linux/arch/riscv/net/bpf_jit_comp64.c

1268 lines
31 KiB
C
Raw Normal View History

bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
// SPDX-License-Identifier: GPL-2.0
/* BPF JIT compiler for RV64G
*
* Copyright(c) 2019 Björn Töpel <bjorn.topel@gmail.com>
*
*/
#include <linux/bitfield.h>
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
#include <linux/bpf.h>
#include <linux/filter.h>
#include "bpf_jit.h"
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
#define RV_REG_TCC RV_REG_A6
#define RV_REG_TCC_SAVED RV_REG_S6 /* Store A6 in S6 if program do calls */
static const int regmap[] = {
[BPF_REG_0] = RV_REG_A5,
[BPF_REG_1] = RV_REG_A0,
[BPF_REG_2] = RV_REG_A1,
[BPF_REG_3] = RV_REG_A2,
[BPF_REG_4] = RV_REG_A3,
[BPF_REG_5] = RV_REG_A4,
[BPF_REG_6] = RV_REG_S1,
[BPF_REG_7] = RV_REG_S2,
[BPF_REG_8] = RV_REG_S3,
[BPF_REG_9] = RV_REG_S4,
[BPF_REG_FP] = RV_REG_S5,
[BPF_REG_AX] = RV_REG_T0,
};
static const int pt_regmap[] = {
[RV_REG_A0] = offsetof(struct pt_regs, a0),
[RV_REG_A1] = offsetof(struct pt_regs, a1),
[RV_REG_A2] = offsetof(struct pt_regs, a2),
[RV_REG_A3] = offsetof(struct pt_regs, a3),
[RV_REG_A4] = offsetof(struct pt_regs, a4),
[RV_REG_A5] = offsetof(struct pt_regs, a5),
[RV_REG_S1] = offsetof(struct pt_regs, s1),
[RV_REG_S2] = offsetof(struct pt_regs, s2),
[RV_REG_S3] = offsetof(struct pt_regs, s3),
[RV_REG_S4] = offsetof(struct pt_regs, s4),
[RV_REG_S5] = offsetof(struct pt_regs, s5),
[RV_REG_T0] = offsetof(struct pt_regs, t0),
};
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
enum {
RV_CTX_F_SEEN_TAIL_CALL = 0,
RV_CTX_F_SEEN_CALL = RV_REG_RA,
RV_CTX_F_SEEN_S1 = RV_REG_S1,
RV_CTX_F_SEEN_S2 = RV_REG_S2,
RV_CTX_F_SEEN_S3 = RV_REG_S3,
RV_CTX_F_SEEN_S4 = RV_REG_S4,
RV_CTX_F_SEEN_S5 = RV_REG_S5,
RV_CTX_F_SEEN_S6 = RV_REG_S6,
};
static u8 bpf_to_rv_reg(int bpf_reg, struct rv_jit_context *ctx)
{
u8 reg = regmap[bpf_reg];
switch (reg) {
case RV_CTX_F_SEEN_S1:
case RV_CTX_F_SEEN_S2:
case RV_CTX_F_SEEN_S3:
case RV_CTX_F_SEEN_S4:
case RV_CTX_F_SEEN_S5:
case RV_CTX_F_SEEN_S6:
__set_bit(reg, &ctx->flags);
}
return reg;
};
static bool seen_reg(int reg, struct rv_jit_context *ctx)
{
switch (reg) {
case RV_CTX_F_SEEN_CALL:
case RV_CTX_F_SEEN_S1:
case RV_CTX_F_SEEN_S2:
case RV_CTX_F_SEEN_S3:
case RV_CTX_F_SEEN_S4:
case RV_CTX_F_SEEN_S5:
case RV_CTX_F_SEEN_S6:
return test_bit(reg, &ctx->flags);
}
return false;
}
static void mark_fp(struct rv_jit_context *ctx)
{
__set_bit(RV_CTX_F_SEEN_S5, &ctx->flags);
}
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
static void mark_call(struct rv_jit_context *ctx)
{
__set_bit(RV_CTX_F_SEEN_CALL, &ctx->flags);
}
static bool seen_call(struct rv_jit_context *ctx)
{
return test_bit(RV_CTX_F_SEEN_CALL, &ctx->flags);
}
static void mark_tail_call(struct rv_jit_context *ctx)
{
__set_bit(RV_CTX_F_SEEN_TAIL_CALL, &ctx->flags);
}
static bool seen_tail_call(struct rv_jit_context *ctx)
{
return test_bit(RV_CTX_F_SEEN_TAIL_CALL, &ctx->flags);
}
static u8 rv_tail_call_reg(struct rv_jit_context *ctx)
{
mark_tail_call(ctx);
if (seen_call(ctx)) {
__set_bit(RV_CTX_F_SEEN_S6, &ctx->flags);
return RV_REG_S6;
}
return RV_REG_A6;
}
static bool is_32b_int(s64 val)
{
return -(1L << 31) <= val && val < (1L << 31);
}
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
static bool in_auipc_jalr_range(s64 val)
{
/*
* auipc+jalr can reach any signed PC-relative offset in the range
* [-2^31 - 2^11, 2^31 - 2^11).
*/
return (-(1L << 31) - (1L << 11)) <= val &&
val < ((1L << 31) - (1L << 11));
}
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
static void emit_imm(u8 rd, s64 val, struct rv_jit_context *ctx)
{
/* Note that the immediate from the add is sign-extended,
* which means that we need to compensate this by adding 2^12,
* when the 12th bit is set. A simpler way of doing this, and
* getting rid of the check, is to just add 2**11 before the
* shift. The "Loading a 32-Bit constant" example from the
* "Computer Organization and Design, RISC-V edition" book by
* Patterson/Hennessy highlights this fact.
*
* This also means that we need to process LSB to MSB.
*/
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
s64 upper = (val + (1 << 11)) >> 12;
/* Sign-extend lower 12 bits to 64 bits since immediates for li, addiw,
* and addi are signed and RVC checks will perform signed comparisons.
*/
s64 lower = ((val & 0xfff) << 52) >> 52;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
int shift;
if (is_32b_int(val)) {
if (upper)
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_lui(rd, upper, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (!upper) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_li(rd, lower, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
return;
}
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addiw(rd, rd, lower, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
return;
}
shift = __ffs(upper);
upper >>= shift;
shift += 12;
emit_imm(rd, upper, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_slli(rd, rd, shift, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (lower)
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(rd, rd, lower, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
static void __build_epilogue(bool is_tail_call, struct rv_jit_context *ctx)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
{
int stack_adjust = ctx->stack_size, store_offset = stack_adjust - 8;
if (seen_reg(RV_REG_RA, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_RA, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_FP, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
if (seen_reg(RV_REG_S1, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_S1, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S2, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_S2, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S3, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_S3, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S4, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_S4, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S5, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_S5, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S6, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_S6, store_offset, RV_REG_SP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(RV_REG_SP, RV_REG_SP, stack_adjust, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
/* Set return value. */
if (!is_tail_call)
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(RV_REG_A0, RV_REG_A5, ctx);
emit_jalr(RV_REG_ZERO, is_tail_call ? RV_REG_T3 : RV_REG_RA,
is_tail_call ? 4 : 0, /* skip TCC init */
ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
static void emit_bcc(u8 cond, u8 rd, u8 rs, int rvoff,
struct rv_jit_context *ctx)
{
switch (cond) {
case BPF_JEQ:
emit(rv_beq(rd, rs, rvoff >> 1), ctx);
return;
case BPF_JGT:
emit(rv_bltu(rs, rd, rvoff >> 1), ctx);
return;
case BPF_JLT:
emit(rv_bltu(rd, rs, rvoff >> 1), ctx);
return;
case BPF_JGE:
emit(rv_bgeu(rd, rs, rvoff >> 1), ctx);
return;
case BPF_JLE:
emit(rv_bgeu(rs, rd, rvoff >> 1), ctx);
return;
case BPF_JNE:
emit(rv_bne(rd, rs, rvoff >> 1), ctx);
return;
case BPF_JSGT:
emit(rv_blt(rs, rd, rvoff >> 1), ctx);
return;
case BPF_JSLT:
emit(rv_blt(rd, rs, rvoff >> 1), ctx);
return;
case BPF_JSGE:
emit(rv_bge(rd, rs, rvoff >> 1), ctx);
return;
case BPF_JSLE:
emit(rv_bge(rs, rd, rvoff >> 1), ctx);
}
}
static void emit_branch(u8 cond, u8 rd, u8 rs, int rvoff,
struct rv_jit_context *ctx)
{
s64 upper, lower;
if (is_13b_int(rvoff)) {
emit_bcc(cond, rd, rs, rvoff, ctx);
return;
}
/* Adjust for jal */
rvoff -= 4;
/* Transform, e.g.:
* bne rd,rs,foo
* to
* beq rd,rs,<.L1>
* (auipc foo)
* jal(r) foo
* .L1
*/
cond = invert_bpf_cond(cond);
if (is_21b_int(rvoff)) {
emit_bcc(cond, rd, rs, 8, ctx);
emit(rv_jal(RV_REG_ZERO, rvoff >> 1), ctx);
return;
}
/* 32b No need for an additional rvoff adjustment, since we
* get that from the auipc at PC', where PC = PC' + 4.
*/
upper = (rvoff + (1 << 11)) >> 12;
lower = rvoff & 0xfff;
emit_bcc(cond, rd, rs, 12, ctx);
emit(rv_auipc(RV_REG_T1, upper), ctx);
emit(rv_jalr(RV_REG_ZERO, RV_REG_T1, lower), ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
static void emit_zext_32(u8 reg, struct rv_jit_context *ctx)
{
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_slli(reg, reg, 32, ctx);
emit_srli(reg, reg, 32, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
static int emit_bpf_tail_call(int insn, struct rv_jit_context *ctx)
{
int tc_ninsn, off, start_insn = ctx->ninsns;
u8 tcc = rv_tail_call_reg(ctx);
/* a0: &ctx
* a1: &array
* a2: index
*
* if (index >= array->map.max_entries)
* goto out;
*/
tc_ninsn = insn ? ctx->offset[insn] - ctx->offset[insn - 1] :
ctx->offset[0];
emit_zext_32(RV_REG_A2, ctx);
off = offsetof(struct bpf_array, map.max_entries);
if (is_12b_check(off, insn))
return -1;
emit(rv_lwu(RV_REG_T1, off, RV_REG_A1), ctx);
off = ninsns_rvoff(tc_ninsn - (ctx->ninsns - start_insn));
emit_branch(BPF_JGE, RV_REG_A2, RV_REG_T1, off, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
bpf: Change value of MAX_TAIL_CALL_CNT from 32 to 33 In the current code, the actual max tail call count is 33 which is greater than MAX_TAIL_CALL_CNT (defined as 32). The actual limit is not consistent with the meaning of MAX_TAIL_CALL_CNT and thus confusing at first glance. We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow bpf programs to tail-call other bpf programs") and commit f9dabe016b63 ("bpf: Undo off-by-one in interpreter tail call count limit"). In order to avoid changing existing behavior, the actual limit is 33 now, this is reasonable. After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can see there exists failed testcase. On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set: # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:0 ret 34 != 33 FAIL On some archs: # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:1 ret 34 != 33 FAIL Although the above failed testcase has been fixed in commit 18935a72eb25 ("bpf/tests: Fix error in tail call limit tests"), it would still be good to change the value of MAX_TAIL_CALL_CNT from 32 to 33 to make the code more readable. The 32-bit x86 JIT was using a limit of 32, just fix the wrong comments and limit to 33 tail calls as the constant MAX_TAIL_CALL_CNT updated. For the mips64 JIT, use "ori" instead of "addiu" as suggested by Johan Almbladh. For the riscv JIT, use RV_REG_TCC directly to save one register move as suggested by Björn Töpel. For the other implementations, no function changes, it does not change the current limit 33, the new value of MAX_TAIL_CALL_CNT can reflect the actual max tail call count, the related tail call testcases in test_bpf module and selftests can work well for the interpreter and the JIT. Here are the test results on x86_64: # uname -m x86_64 # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf test_suite=test_tail_calls # dmesg | tail -1 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [0/8 JIT'ed] # rmmod test_bpf # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf test_suite=test_tail_calls # dmesg | tail -1 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [8/8 JIT'ed] # rmmod test_bpf # ./test_progs -t tailcalls #142 tailcalls:OK Summary: 1/11 PASSED, 0 SKIPPED, 0 FAILED Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Tested-by: Johan Almbladh <johan.almbladh@anyfinetworks.com> Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> Acked-by: Björn Töpel <bjorn@kernel.org> Acked-by: Johan Almbladh <johan.almbladh@anyfinetworks.com> Acked-by: Ilya Leoshkevich <iii@linux.ibm.com> Link: https://lore.kernel.org/bpf/1636075800-3264-1-git-send-email-yangtiezhu@loongson.cn
2021-11-05 09:30:00 +08:00
/* if (--TCC < 0)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
* goto out;
*/
bpf: Change value of MAX_TAIL_CALL_CNT from 32 to 33 In the current code, the actual max tail call count is 33 which is greater than MAX_TAIL_CALL_CNT (defined as 32). The actual limit is not consistent with the meaning of MAX_TAIL_CALL_CNT and thus confusing at first glance. We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow bpf programs to tail-call other bpf programs") and commit f9dabe016b63 ("bpf: Undo off-by-one in interpreter tail call count limit"). In order to avoid changing existing behavior, the actual limit is 33 now, this is reasonable. After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can see there exists failed testcase. On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set: # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:0 ret 34 != 33 FAIL On some archs: # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:1 ret 34 != 33 FAIL Although the above failed testcase has been fixed in commit 18935a72eb25 ("bpf/tests: Fix error in tail call limit tests"), it would still be good to change the value of MAX_TAIL_CALL_CNT from 32 to 33 to make the code more readable. The 32-bit x86 JIT was using a limit of 32, just fix the wrong comments and limit to 33 tail calls as the constant MAX_TAIL_CALL_CNT updated. For the mips64 JIT, use "ori" instead of "addiu" as suggested by Johan Almbladh. For the riscv JIT, use RV_REG_TCC directly to save one register move as suggested by Björn Töpel. For the other implementations, no function changes, it does not change the current limit 33, the new value of MAX_TAIL_CALL_CNT can reflect the actual max tail call count, the related tail call testcases in test_bpf module and selftests can work well for the interpreter and the JIT. Here are the test results on x86_64: # uname -m x86_64 # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf test_suite=test_tail_calls # dmesg | tail -1 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [0/8 JIT'ed] # rmmod test_bpf # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf test_suite=test_tail_calls # dmesg | tail -1 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [8/8 JIT'ed] # rmmod test_bpf # ./test_progs -t tailcalls #142 tailcalls:OK Summary: 1/11 PASSED, 0 SKIPPED, 0 FAILED Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Tested-by: Johan Almbladh <johan.almbladh@anyfinetworks.com> Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> Acked-by: Björn Töpel <bjorn@kernel.org> Acked-by: Johan Almbladh <johan.almbladh@anyfinetworks.com> Acked-by: Ilya Leoshkevich <iii@linux.ibm.com> Link: https://lore.kernel.org/bpf/1636075800-3264-1-git-send-email-yangtiezhu@loongson.cn
2021-11-05 09:30:00 +08:00
emit_addi(RV_REG_TCC, tcc, -1, ctx);
off = ninsns_rvoff(tc_ninsn - (ctx->ninsns - start_insn));
bpf: Change value of MAX_TAIL_CALL_CNT from 32 to 33 In the current code, the actual max tail call count is 33 which is greater than MAX_TAIL_CALL_CNT (defined as 32). The actual limit is not consistent with the meaning of MAX_TAIL_CALL_CNT and thus confusing at first glance. We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow bpf programs to tail-call other bpf programs") and commit f9dabe016b63 ("bpf: Undo off-by-one in interpreter tail call count limit"). In order to avoid changing existing behavior, the actual limit is 33 now, this is reasonable. After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can see there exists failed testcase. On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set: # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:0 ret 34 != 33 FAIL On some archs: # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:1 ret 34 != 33 FAIL Although the above failed testcase has been fixed in commit 18935a72eb25 ("bpf/tests: Fix error in tail call limit tests"), it would still be good to change the value of MAX_TAIL_CALL_CNT from 32 to 33 to make the code more readable. The 32-bit x86 JIT was using a limit of 32, just fix the wrong comments and limit to 33 tail calls as the constant MAX_TAIL_CALL_CNT updated. For the mips64 JIT, use "ori" instead of "addiu" as suggested by Johan Almbladh. For the riscv JIT, use RV_REG_TCC directly to save one register move as suggested by Björn Töpel. For the other implementations, no function changes, it does not change the current limit 33, the new value of MAX_TAIL_CALL_CNT can reflect the actual max tail call count, the related tail call testcases in test_bpf module and selftests can work well for the interpreter and the JIT. Here are the test results on x86_64: # uname -m x86_64 # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf test_suite=test_tail_calls # dmesg | tail -1 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [0/8 JIT'ed] # rmmod test_bpf # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf test_suite=test_tail_calls # dmesg | tail -1 test_bpf: test_tail_calls: Summary: 8 PASSED, 0 FAILED, [8/8 JIT'ed] # rmmod test_bpf # ./test_progs -t tailcalls #142 tailcalls:OK Summary: 1/11 PASSED, 0 SKIPPED, 0 FAILED Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Tested-by: Johan Almbladh <johan.almbladh@anyfinetworks.com> Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> Acked-by: Björn Töpel <bjorn@kernel.org> Acked-by: Johan Almbladh <johan.almbladh@anyfinetworks.com> Acked-by: Ilya Leoshkevich <iii@linux.ibm.com> Link: https://lore.kernel.org/bpf/1636075800-3264-1-git-send-email-yangtiezhu@loongson.cn
2021-11-05 09:30:00 +08:00
emit_branch(BPF_JSLT, RV_REG_TCC, RV_REG_ZERO, off, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
/* prog = array->ptrs[index];
* if (!prog)
* goto out;
*/
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_slli(RV_REG_T2, RV_REG_A2, 3, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_A1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
off = offsetof(struct bpf_array, ptrs);
if (is_12b_check(off, insn))
return -1;
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_T2, off, RV_REG_T2, ctx);
off = ninsns_rvoff(tc_ninsn - (ctx->ninsns - start_insn));
emit_branch(BPF_JEQ, RV_REG_T2, RV_REG_ZERO, off, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
/* goto *(prog->bpf_func + 4); */
off = offsetof(struct bpf_prog, bpf_func);
if (is_12b_check(off, insn))
return -1;
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_ld(RV_REG_T3, off, RV_REG_T2, ctx);
__build_epilogue(true, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
return 0;
}
static void init_regs(u8 *rd, u8 *rs, const struct bpf_insn *insn,
struct rv_jit_context *ctx)
{
u8 code = insn->code;
switch (code) {
case BPF_JMP | BPF_JA:
case BPF_JMP | BPF_CALL:
case BPF_JMP | BPF_EXIT:
case BPF_JMP | BPF_TAIL_CALL:
break;
default:
*rd = bpf_to_rv_reg(insn->dst_reg, ctx);
}
if (code & (BPF_ALU | BPF_X) || code & (BPF_ALU64 | BPF_X) ||
code & (BPF_JMP | BPF_X) || code & (BPF_JMP32 | BPF_X) ||
code & BPF_LDX || code & BPF_STX)
*rs = bpf_to_rv_reg(insn->src_reg, ctx);
}
static void emit_zext_32_rd_rs(u8 *rd, u8 *rs, struct rv_jit_context *ctx)
{
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(RV_REG_T2, *rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(RV_REG_T2, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(RV_REG_T1, *rs, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(RV_REG_T1, ctx);
*rd = RV_REG_T2;
*rs = RV_REG_T1;
}
static void emit_sext_32_rd_rs(u8 *rd, u8 *rs, struct rv_jit_context *ctx)
{
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addiw(RV_REG_T2, *rd, 0, ctx);
emit_addiw(RV_REG_T1, *rs, 0, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
*rd = RV_REG_T2;
*rs = RV_REG_T1;
}
static void emit_zext_32_rd_t1(u8 *rd, struct rv_jit_context *ctx)
{
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(RV_REG_T2, *rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(RV_REG_T2, ctx);
emit_zext_32(RV_REG_T1, ctx);
*rd = RV_REG_T2;
}
static void emit_sext_32_rd(u8 *rd, struct rv_jit_context *ctx)
{
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addiw(RV_REG_T2, *rd, 0, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
*rd = RV_REG_T2;
}
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
static int emit_jump_and_link(u8 rd, s64 rvoff, bool force_jalr,
struct rv_jit_context *ctx)
{
s64 upper, lower;
if (rvoff && is_21b_int(rvoff) && !force_jalr) {
emit(rv_jal(rd, rvoff >> 1), ctx);
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
return 0;
} else if (in_auipc_jalr_range(rvoff)) {
upper = (rvoff + (1 << 11)) >> 12;
lower = rvoff & 0xfff;
emit(rv_auipc(RV_REG_T1, upper), ctx);
emit(rv_jalr(rd, RV_REG_T1, lower), ctx);
return 0;
}
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
pr_err("bpf-jit: target offset 0x%llx is out of range\n", rvoff);
return -ERANGE;
}
static bool is_signed_bpf_cond(u8 cond)
{
return cond == BPF_JSGT || cond == BPF_JSLT ||
cond == BPF_JSGE || cond == BPF_JSLE;
}
static int emit_call(bool fixed, u64 addr, struct rv_jit_context *ctx)
{
s64 off = 0;
u64 ip;
u8 rd;
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
int ret;
if (addr && ctx->insns) {
ip = (u64)(long)(ctx->insns + ctx->ninsns);
off = addr - ip;
}
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
ret = emit_jump_and_link(RV_REG_RA, off, !fixed, ctx);
if (ret)
return ret;
rd = bpf_to_rv_reg(BPF_REG_0, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(rd, RV_REG_A0, ctx);
return 0;
}
#define BPF_FIXUP_OFFSET_MASK GENMASK(26, 0)
#define BPF_FIXUP_REG_MASK GENMASK(31, 27)
bool ex_handler_bpf(const struct exception_table_entry *ex,
struct pt_regs *regs)
{
off_t offset = FIELD_GET(BPF_FIXUP_OFFSET_MASK, ex->fixup);
int regs_offset = FIELD_GET(BPF_FIXUP_REG_MASK, ex->fixup);
*(unsigned long *)((void *)regs + pt_regmap[regs_offset]) = 0;
regs->epc = (unsigned long)&ex->fixup - offset;
return true;
}
/* For accesses to BTF pointers, add an entry to the exception table */
static int add_exception_handler(const struct bpf_insn *insn,
struct rv_jit_context *ctx,
int dst_reg, int insn_len)
{
struct exception_table_entry *ex;
unsigned long pc;
off_t offset;
if (!ctx->insns || !ctx->prog->aux->extable || BPF_MODE(insn->code) != BPF_PROBE_MEM)
return 0;
if (WARN_ON_ONCE(ctx->nexentries >= ctx->prog->aux->num_exentries))
return -EINVAL;
if (WARN_ON_ONCE(insn_len > ctx->ninsns))
return -EINVAL;
if (WARN_ON_ONCE(!rvc_enabled() && insn_len == 1))
return -EINVAL;
ex = &ctx->prog->aux->extable[ctx->nexentries];
pc = (unsigned long)&ctx->insns[ctx->ninsns - insn_len];
offset = pc - (long)&ex->insn;
if (WARN_ON_ONCE(offset >= 0 || offset < INT_MIN))
return -ERANGE;
ex->insn = pc;
/*
* Since the extable follows the program, the fixup offset is always
* negative and limited to BPF_JIT_REGION_SIZE. Store a positive value
* to keep things simple, and put the destination register in the upper
* bits. We don't need to worry about buildtime or runtime sort
* modifying the upper bits because the table is already sorted, and
* isn't part of the main exception table.
*/
offset = (long)&ex->fixup - (pc + insn_len * sizeof(u16));
if (!FIELD_FIT(BPF_FIXUP_OFFSET_MASK, offset))
return -ERANGE;
ex->fixup = FIELD_PREP(BPF_FIXUP_OFFSET_MASK, offset) |
FIELD_PREP(BPF_FIXUP_REG_MASK, dst_reg);
ex->type = EX_TYPE_BPF;
ctx->nexentries++;
return 0;
}
int bpf_jit_emit_insn(const struct bpf_insn *insn, struct rv_jit_context *ctx,
bool extra_pass)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
{
bool is64 = BPF_CLASS(insn->code) == BPF_ALU64 ||
BPF_CLASS(insn->code) == BPF_JMP;
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
int s, e, rvoff, ret, i = insn - ctx->prog->insnsi;
struct bpf_prog_aux *aux = ctx->prog->aux;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
u8 rd = -1, rs = -1, code = insn->code;
s16 off = insn->off;
s32 imm = insn->imm;
init_regs(&rd, &rs, insn, ctx);
switch (code) {
/* dst = src */
case BPF_ALU | BPF_MOV | BPF_X:
case BPF_ALU64 | BPF_MOV | BPF_X:
if (imm == 1) {
/* Special mov32 for zext */
emit_zext_32(rd, ctx);
break;
}
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(rd, rs, ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
/* dst = dst OP src */
case BPF_ALU | BPF_ADD | BPF_X:
case BPF_ALU64 | BPF_ADD | BPF_X:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(rd, rd, rs, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_SUB | BPF_X:
case BPF_ALU64 | BPF_SUB | BPF_X:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
if (is64)
emit_sub(rd, rd, rs, ctx);
else
emit_subw(rd, rd, rs, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_AND | BPF_X:
case BPF_ALU64 | BPF_AND | BPF_X:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_and(rd, rd, rs, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_OR | BPF_X:
case BPF_ALU64 | BPF_OR | BPF_X:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_or(rd, rd, rs, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_XOR | BPF_X:
case BPF_ALU64 | BPF_XOR | BPF_X:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_xor(rd, rd, rs, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_MUL | BPF_X:
case BPF_ALU64 | BPF_MUL | BPF_X:
emit(is64 ? rv_mul(rd, rd, rs) : rv_mulw(rd, rd, rs), ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_DIV | BPF_X:
case BPF_ALU64 | BPF_DIV | BPF_X:
emit(is64 ? rv_divu(rd, rd, rs) : rv_divuw(rd, rd, rs), ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_MOD | BPF_X:
case BPF_ALU64 | BPF_MOD | BPF_X:
emit(is64 ? rv_remu(rd, rd, rs) : rv_remuw(rd, rd, rs), ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_LSH | BPF_X:
case BPF_ALU64 | BPF_LSH | BPF_X:
emit(is64 ? rv_sll(rd, rd, rs) : rv_sllw(rd, rd, rs), ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_RSH | BPF_X:
case BPF_ALU64 | BPF_RSH | BPF_X:
emit(is64 ? rv_srl(rd, rd, rs) : rv_srlw(rd, rd, rs), ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_ARSH | BPF_X:
case BPF_ALU64 | BPF_ARSH | BPF_X:
emit(is64 ? rv_sra(rd, rd, rs) : rv_sraw(rd, rd, rs), ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* dst = -dst */
case BPF_ALU | BPF_NEG:
case BPF_ALU64 | BPF_NEG:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sub(rd, RV_REG_ZERO, rd, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* dst = BSWAP##imm(dst) */
case BPF_ALU | BPF_END | BPF_FROM_LE:
switch (imm) {
case 16:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_slli(rd, rd, 48, ctx);
emit_srli(rd, rd, 48, ctx);
break;
case 32:
if (!aux->verifier_zext)
emit_zext_32(rd, ctx);
break;
case 64:
/* Do nothing */
break;
}
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
case BPF_ALU | BPF_END | BPF_FROM_BE:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_li(RV_REG_T2, 0, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (imm == 16)
goto out_be;
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (imm == 32)
goto out_be;
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
emit_slli(RV_REG_T2, RV_REG_T2, 8, ctx);
emit_srli(rd, rd, 8, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
out_be:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(RV_REG_T1, rd, 0xff, ctx);
emit_add(RV_REG_T2, RV_REG_T2, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(rd, RV_REG_T2, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* dst = imm */
case BPF_ALU | BPF_MOV | BPF_K:
case BPF_ALU64 | BPF_MOV | BPF_K:
emit_imm(rd, imm, ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
/* dst = dst OP imm */
case BPF_ALU | BPF_ADD | BPF_K:
case BPF_ALU64 | BPF_ADD | BPF_K:
if (is_12b_int(imm)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(rd, rd, imm, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
} else {
emit_imm(RV_REG_T1, imm, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(rd, rd, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_SUB | BPF_K:
case BPF_ALU64 | BPF_SUB | BPF_K:
if (is_12b_int(-imm)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(rd, rd, -imm, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
} else {
emit_imm(RV_REG_T1, imm, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sub(rd, rd, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_AND | BPF_K:
case BPF_ALU64 | BPF_AND | BPF_K:
if (is_12b_int(imm)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(rd, rd, imm, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
} else {
emit_imm(RV_REG_T1, imm, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_and(rd, rd, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_OR | BPF_K:
case BPF_ALU64 | BPF_OR | BPF_K:
if (is_12b_int(imm)) {
emit(rv_ori(rd, rd, imm), ctx);
} else {
emit_imm(RV_REG_T1, imm, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_or(rd, rd, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_XOR | BPF_K:
case BPF_ALU64 | BPF_XOR | BPF_K:
if (is_12b_int(imm)) {
emit(rv_xori(rd, rd, imm), ctx);
} else {
emit_imm(RV_REG_T1, imm, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_xor(rd, rd, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_MUL | BPF_K:
case BPF_ALU64 | BPF_MUL | BPF_K:
emit_imm(RV_REG_T1, imm, ctx);
emit(is64 ? rv_mul(rd, rd, RV_REG_T1) :
rv_mulw(rd, rd, RV_REG_T1), ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_DIV | BPF_K:
case BPF_ALU64 | BPF_DIV | BPF_K:
emit_imm(RV_REG_T1, imm, ctx);
emit(is64 ? rv_divu(rd, rd, RV_REG_T1) :
rv_divuw(rd, rd, RV_REG_T1), ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_MOD | BPF_K:
case BPF_ALU64 | BPF_MOD | BPF_K:
emit_imm(RV_REG_T1, imm, ctx);
emit(is64 ? rv_remu(rd, rd, RV_REG_T1) :
rv_remuw(rd, rd, RV_REG_T1), ctx);
if (!is64 && !aux->verifier_zext)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit_zext_32(rd, ctx);
break;
case BPF_ALU | BPF_LSH | BPF_K:
case BPF_ALU64 | BPF_LSH | BPF_K:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_slli(rd, rd, imm, ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_RSH | BPF_K:
case BPF_ALU64 | BPF_RSH | BPF_K:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
if (is64)
emit_srli(rd, rd, imm, ctx);
else
emit(rv_srliw(rd, rd, imm), ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ALU | BPF_ARSH | BPF_K:
case BPF_ALU64 | BPF_ARSH | BPF_K:
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
if (is64)
emit_srai(rd, rd, imm, ctx);
else
emit(rv_sraiw(rd, rd, imm), ctx);
if (!is64 && !aux->verifier_zext)
emit_zext_32(rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* JUMP off */
case BPF_JMP | BPF_JA:
rvoff = rv_offset(i, off, ctx);
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
ret = emit_jump_and_link(RV_REG_ZERO, rvoff, false, ctx);
if (ret)
return ret;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* IF (dst COND src) JUMP off */
case BPF_JMP | BPF_JEQ | BPF_X:
case BPF_JMP32 | BPF_JEQ | BPF_X:
case BPF_JMP | BPF_JGT | BPF_X:
case BPF_JMP32 | BPF_JGT | BPF_X:
case BPF_JMP | BPF_JLT | BPF_X:
case BPF_JMP32 | BPF_JLT | BPF_X:
case BPF_JMP | BPF_JGE | BPF_X:
case BPF_JMP32 | BPF_JGE | BPF_X:
case BPF_JMP | BPF_JLE | BPF_X:
case BPF_JMP32 | BPF_JLE | BPF_X:
case BPF_JMP | BPF_JNE | BPF_X:
case BPF_JMP32 | BPF_JNE | BPF_X:
case BPF_JMP | BPF_JSGT | BPF_X:
case BPF_JMP32 | BPF_JSGT | BPF_X:
case BPF_JMP | BPF_JSLT | BPF_X:
case BPF_JMP32 | BPF_JSLT | BPF_X:
case BPF_JMP | BPF_JSGE | BPF_X:
case BPF_JMP32 | BPF_JSGE | BPF_X:
case BPF_JMP | BPF_JSLE | BPF_X:
case BPF_JMP32 | BPF_JSLE | BPF_X:
case BPF_JMP | BPF_JSET | BPF_X:
case BPF_JMP32 | BPF_JSET | BPF_X:
rvoff = rv_offset(i, off, ctx);
if (!is64) {
s = ctx->ninsns;
if (is_signed_bpf_cond(BPF_OP(code)))
emit_sext_32_rd_rs(&rd, &rs, ctx);
else
emit_zext_32_rd_rs(&rd, &rs, ctx);
e = ctx->ninsns;
/* Adjust for extra insns */
rvoff -= ninsns_rvoff(e - s);
}
if (BPF_OP(code) == BPF_JSET) {
/* Adjust for and */
rvoff -= 4;
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_and(RV_REG_T1, rd, rs, ctx);
emit_branch(BPF_JNE, RV_REG_T1, RV_REG_ZERO, rvoff,
ctx);
} else {
emit_branch(BPF_OP(code), rd, rs, rvoff, ctx);
}
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* IF (dst COND imm) JUMP off */
case BPF_JMP | BPF_JEQ | BPF_K:
case BPF_JMP32 | BPF_JEQ | BPF_K:
case BPF_JMP | BPF_JGT | BPF_K:
case BPF_JMP32 | BPF_JGT | BPF_K:
case BPF_JMP | BPF_JLT | BPF_K:
case BPF_JMP32 | BPF_JLT | BPF_K:
case BPF_JMP | BPF_JGE | BPF_K:
case BPF_JMP32 | BPF_JGE | BPF_K:
case BPF_JMP | BPF_JLE | BPF_K:
case BPF_JMP32 | BPF_JLE | BPF_K:
case BPF_JMP | BPF_JNE | BPF_K:
case BPF_JMP32 | BPF_JNE | BPF_K:
case BPF_JMP | BPF_JSGT | BPF_K:
case BPF_JMP32 | BPF_JSGT | BPF_K:
case BPF_JMP | BPF_JSLT | BPF_K:
case BPF_JMP32 | BPF_JSLT | BPF_K:
case BPF_JMP | BPF_JSGE | BPF_K:
case BPF_JMP32 | BPF_JSGE | BPF_K:
case BPF_JMP | BPF_JSLE | BPF_K:
case BPF_JMP32 | BPF_JSLE | BPF_K:
rvoff = rv_offset(i, off, ctx);
s = ctx->ninsns;
if (imm) {
emit_imm(RV_REG_T1, imm, ctx);
rs = RV_REG_T1;
} else {
/* If imm is 0, simply use zero register. */
rs = RV_REG_ZERO;
}
if (!is64) {
if (is_signed_bpf_cond(BPF_OP(code)))
emit_sext_32_rd(&rd, ctx);
else
emit_zext_32_rd_t1(&rd, ctx);
}
e = ctx->ninsns;
/* Adjust for extra insns */
rvoff -= ninsns_rvoff(e - s);
emit_branch(BPF_OP(code), rd, rs, rvoff, ctx);
break;
case BPF_JMP | BPF_JSET | BPF_K:
case BPF_JMP32 | BPF_JSET | BPF_K:
rvoff = rv_offset(i, off, ctx);
s = ctx->ninsns;
if (is_12b_int(imm)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_andi(RV_REG_T1, rd, imm, ctx);
} else {
emit_imm(RV_REG_T1, imm, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_and(RV_REG_T1, rd, RV_REG_T1, ctx);
}
/* For jset32, we should clear the upper 32 bits of t1, but
* sign-extension is sufficient here and saves one instruction,
* as t1 is used only in comparison against zero.
*/
if (!is64 && imm < 0)
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addiw(RV_REG_T1, RV_REG_T1, 0, ctx);
e = ctx->ninsns;
rvoff -= ninsns_rvoff(e - s);
emit_branch(BPF_JNE, RV_REG_T1, RV_REG_ZERO, rvoff, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* function call */
case BPF_JMP | BPF_CALL:
{
bool fixed;
u64 addr;
mark_call(ctx);
ret = bpf_jit_get_func_addr(ctx->prog, insn, extra_pass, &addr,
&fixed);
if (ret < 0)
return ret;
ret = emit_call(fixed, addr, ctx);
if (ret)
return ret;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
/* tail call */
case BPF_JMP | BPF_TAIL_CALL:
if (emit_bpf_tail_call(i, ctx))
return -1;
break;
/* function return */
case BPF_JMP | BPF_EXIT:
if (i == ctx->prog->len - 1)
break;
rvoff = epilogue_offset(ctx);
riscv, bpf: Fix offset range checking for auipc+jalr on RV64 The existing code in emit_call on RV64 checks that the PC-relative offset to the function fits in 32 bits before calling emit_jump_and_link to emit an auipc+jalr pair. However, this check is incorrect because offsets in the range [2^31 - 2^11, 2^31 - 1] cannot be encoded using auipc+jalr on RV64 (see discussion [1]). The RISC-V spec has recently been updated to reflect this fact [2, 3]. This patch fixes the problem by moving the check on the offset into emit_jump_and_link and modifying it to the correct range of encodable offsets, which is [-2^31 - 2^11, 2^31 - 2^11). This also enforces the check on the offset to other uses of emit_jump_and_link (e.g., BPF_JA) as well. Currently, this bug is unlikely to be triggered, because the memory region from which JITed images are allocated is close enough to kernel text for the offsets to not become too large; and because the bounds on BPF program size are small enough. This patch prevents this problem from becoming an issue if either of these change. [1]: https://groups.google.com/a/groups.riscv.org/forum/#!topic/isa-dev/bwWFhBnnZFQ [2]: https://github.com/riscv/riscv-isa-manual/commit/b1e42e09ac55116dbf9de5e4fb326a5a90e4a993 [3]: https://github.com/riscv/riscv-isa-manual/commit/4c1b2066ebd2965a422e41eb262d0a208a7fea07 Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Link: https://lore.kernel.org/bpf/20200406221604.18547-1-luke.r.nels@gmail.com
2020-04-07 06:16:04 +08:00
ret = emit_jump_and_link(RV_REG_ZERO, rvoff, false, ctx);
if (ret)
return ret;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* dst = imm64 */
case BPF_LD | BPF_IMM | BPF_DW:
{
struct bpf_insn insn1 = insn[1];
u64 imm64;
imm64 = (u64)insn1.imm << 32 | (u32)imm;
emit_imm(rd, imm64, ctx);
return 1;
}
/* LDX: dst = *(size *)(src + off) */
case BPF_LDX | BPF_MEM | BPF_B:
case BPF_LDX | BPF_MEM | BPF_H:
case BPF_LDX | BPF_MEM | BPF_W:
case BPF_LDX | BPF_MEM | BPF_DW:
case BPF_LDX | BPF_PROBE_MEM | BPF_B:
case BPF_LDX | BPF_PROBE_MEM | BPF_H:
case BPF_LDX | BPF_PROBE_MEM | BPF_W:
case BPF_LDX | BPF_PROBE_MEM | BPF_DW:
{
int insn_len, insns_start;
switch (BPF_SIZE(code)) {
case BPF_B:
if (is_12b_int(off)) {
insns_start = ctx->ninsns;
emit(rv_lbu(rd, off, rs), ctx);
insn_len = ctx->ninsns - insns_start;
break;
}
emit_imm(RV_REG_T1, off, ctx);
emit_add(RV_REG_T1, RV_REG_T1, rs, ctx);
insns_start = ctx->ninsns;
emit(rv_lbu(rd, 0, RV_REG_T1), ctx);
insn_len = ctx->ninsns - insns_start;
if (insn_is_zext(&insn[1]))
return 1;
break;
case BPF_H:
if (is_12b_int(off)) {
insns_start = ctx->ninsns;
emit(rv_lhu(rd, off, rs), ctx);
insn_len = ctx->ninsns - insns_start;
break;
}
emit_imm(RV_REG_T1, off, ctx);
emit_add(RV_REG_T1, RV_REG_T1, rs, ctx);
insns_start = ctx->ninsns;
emit(rv_lhu(rd, 0, RV_REG_T1), ctx);
insn_len = ctx->ninsns - insns_start;
if (insn_is_zext(&insn[1]))
return 1;
break;
case BPF_W:
if (is_12b_int(off)) {
insns_start = ctx->ninsns;
emit(rv_lwu(rd, off, rs), ctx);
insn_len = ctx->ninsns - insns_start;
break;
}
emit_imm(RV_REG_T1, off, ctx);
emit_add(RV_REG_T1, RV_REG_T1, rs, ctx);
insns_start = ctx->ninsns;
emit(rv_lwu(rd, 0, RV_REG_T1), ctx);
insn_len = ctx->ninsns - insns_start;
if (insn_is_zext(&insn[1]))
return 1;
break;
case BPF_DW:
if (is_12b_int(off)) {
insns_start = ctx->ninsns;
emit_ld(rd, off, rs, ctx);
insn_len = ctx->ninsns - insns_start;
break;
}
emit_imm(RV_REG_T1, off, ctx);
emit_add(RV_REG_T1, RV_REG_T1, rs, ctx);
insns_start = ctx->ninsns;
emit_ld(rd, 0, RV_REG_T1, ctx);
insn_len = ctx->ninsns - insns_start;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
ret = add_exception_handler(insn, ctx, rd, insn_len);
if (ret)
return ret;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
/* speculation barrier */
case BPF_ST | BPF_NOSPEC:
break;
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
/* ST: *(size *)(dst + off) = imm */
case BPF_ST | BPF_MEM | BPF_B:
emit_imm(RV_REG_T1, imm, ctx);
if (is_12b_int(off)) {
emit(rv_sb(rd, off, RV_REG_T1), ctx);
break;
}
emit_imm(RV_REG_T2, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T2, RV_REG_T2, rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit(rv_sb(RV_REG_T2, 0, RV_REG_T1), ctx);
break;
case BPF_ST | BPF_MEM | BPF_H:
emit_imm(RV_REG_T1, imm, ctx);
if (is_12b_int(off)) {
emit(rv_sh(rd, off, RV_REG_T1), ctx);
break;
}
emit_imm(RV_REG_T2, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T2, RV_REG_T2, rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit(rv_sh(RV_REG_T2, 0, RV_REG_T1), ctx);
break;
case BPF_ST | BPF_MEM | BPF_W:
emit_imm(RV_REG_T1, imm, ctx);
if (is_12b_int(off)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sw(rd, off, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
emit_imm(RV_REG_T2, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T2, RV_REG_T2, rd, ctx);
emit_sw(RV_REG_T2, 0, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_ST | BPF_MEM | BPF_DW:
emit_imm(RV_REG_T1, imm, ctx);
if (is_12b_int(off)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(rd, off, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
emit_imm(RV_REG_T2, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T2, RV_REG_T2, rd, ctx);
emit_sd(RV_REG_T2, 0, RV_REG_T1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
/* STX: *(size *)(dst + off) = src */
case BPF_STX | BPF_MEM | BPF_B:
if (is_12b_int(off)) {
emit(rv_sb(rd, off, rs), ctx);
break;
}
emit_imm(RV_REG_T1, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T1, RV_REG_T1, rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit(rv_sb(RV_REG_T1, 0, rs), ctx);
break;
case BPF_STX | BPF_MEM | BPF_H:
if (is_12b_int(off)) {
emit(rv_sh(rd, off, rs), ctx);
break;
}
emit_imm(RV_REG_T1, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T1, RV_REG_T1, rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
emit(rv_sh(RV_REG_T1, 0, rs), ctx);
break;
case BPF_STX | BPF_MEM | BPF_W:
if (is_12b_int(off)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sw(rd, off, rs, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
emit_imm(RV_REG_T1, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T1, RV_REG_T1, rd, ctx);
emit_sw(RV_REG_T1, 0, rs, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_STX | BPF_MEM | BPF_DW:
if (is_12b_int(off)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(rd, off, rs, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
}
emit_imm(RV_REG_T1, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T1, RV_REG_T1, rd, ctx);
emit_sd(RV_REG_T1, 0, rs, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
break;
case BPF_STX | BPF_ATOMIC | BPF_W:
case BPF_STX | BPF_ATOMIC | BPF_DW:
if (insn->imm != BPF_ADD) {
pr_err("bpf-jit: not supported: atomic operation %02x ***\n",
insn->imm);
return -EINVAL;
}
/* atomic_add: lock *(u32 *)(dst + off) += src
* atomic_add: lock *(u64 *)(dst + off) += src
*/
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (off) {
if (is_12b_int(off)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(RV_REG_T1, rd, off, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
} else {
emit_imm(RV_REG_T1, off, ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_add(RV_REG_T1, RV_REG_T1, rd, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}
rd = RV_REG_T1;
}
emit(BPF_SIZE(code) == BPF_W ?
rv_amoadd_w(RV_REG_ZERO, rs, rd, 0, 0) :
rv_amoadd_d(RV_REG_ZERO, rs, rd, 0, 0), ctx);
break;
default:
pr_err("bpf-jit: unknown opcode %02x\n", code);
return -EINVAL;
}
return 0;
}
void bpf_jit_build_prologue(struct rv_jit_context *ctx)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
{
int stack_adjust = 0, store_offset, bpf_stack_adjust;
bpf_stack_adjust = round_up(ctx->prog->aux->stack_depth, 16);
if (bpf_stack_adjust)
mark_fp(ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (seen_reg(RV_REG_RA, ctx))
stack_adjust += 8;
stack_adjust += 8; /* RV_REG_FP */
if (seen_reg(RV_REG_S1, ctx))
stack_adjust += 8;
if (seen_reg(RV_REG_S2, ctx))
stack_adjust += 8;
if (seen_reg(RV_REG_S3, ctx))
stack_adjust += 8;
if (seen_reg(RV_REG_S4, ctx))
stack_adjust += 8;
if (seen_reg(RV_REG_S5, ctx))
stack_adjust += 8;
if (seen_reg(RV_REG_S6, ctx))
stack_adjust += 8;
stack_adjust = round_up(stack_adjust, 16);
stack_adjust += bpf_stack_adjust;
store_offset = stack_adjust - 8;
/* First instruction is always setting the tail-call-counter
* (TCC) register. This instruction is skipped for tail calls.
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
* Force using a 4-byte (non-compressed) instruction.
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
*/
emit(rv_addi(RV_REG_TCC, RV_REG_ZERO, MAX_TAIL_CALL_CNT), ctx);
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(RV_REG_SP, RV_REG_SP, -stack_adjust, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (seen_reg(RV_REG_RA, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_RA, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_FP, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
if (seen_reg(RV_REG_S1, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_S1, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S2, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_S2, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S3, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_S3, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S4, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_S4, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S5, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_S5, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
if (seen_reg(RV_REG_S6, ctx)) {
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_sd(RV_REG_SP, store_offset, RV_REG_S6, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
store_offset -= 8;
}
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(RV_REG_FP, RV_REG_SP, stack_adjust, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
if (bpf_stack_adjust)
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_addi(RV_REG_S5, RV_REG_SP, bpf_stack_adjust, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
/* Program contains calls and tail calls, so RV_REG_TCC need
* to be saved across calls.
*/
if (seen_tail_call(ctx) && seen_call(ctx))
bpf, riscv: Use compressed instructions in the rv64 JIT This patch uses the RVC support and encodings from bpf_jit.h to optimize the rv64 jit. The optimizations work by replacing emit(rv_X(...)) with a call to a helper function emit_X, which will emit a compressed version of the instruction when possible, and when RVC is enabled. The JIT continues to pass all tests in lib/test_bpf.c, and introduces no new failures to test_verifier; both with and without RVC being enabled. Most changes are straightforward replacements of emit(rv_X(...), ctx) with emit_X(..., ctx), with the following exceptions bearing mention; * Change emit_imm to sign-extend the value in "lower", since the checks for RVC (and the instructions themselves) treat the value as signed. Otherwise, small negative immediates will not be recognized as encodable using an RVC instruction. For example, without this change, emit_imm(rd, -1, ctx) would cause lower to become 4095, which is not a 6b int even though a "c.li rd, -1" instruction suffices. * For {BPF_MOV,BPF_ADD} BPF_X, drop using addiw,addw in the 32-bit cases since the values are zero-extended into the upper 32 bits in the following instructions anyways, and the addition commutes with zero-extension. (BPF_SUB BPF_X must still use subw since subtraction does not commute with zero-extension.) This patch avoids optimizing branches and jumps to use RVC instructions since surrounding code often makes assumptions about the sizes of emitted instructions. Optimizing these will require changing these functions (e.g., emit_branch) to dynamically compute jump offsets. The following are examples of the JITed code for the verifier selftest "direct packet read test#3 for CGROUP_SKB OK", without and with RVC enabled, respectively. The former uses 178 bytes, and the latter uses 112, for a ~37% reduction in code size for this example. Without RVC: 0: 02000813 addi a6,zero,32 4: fd010113 addi sp,sp,-48 8: 02813423 sd s0,40(sp) c: 02913023 sd s1,32(sp) 10: 01213c23 sd s2,24(sp) 14: 01313823 sd s3,16(sp) 18: 01413423 sd s4,8(sp) 1c: 03010413 addi s0,sp,48 20: 03056683 lwu a3,48(a0) 24: 02069693 slli a3,a3,0x20 28: 0206d693 srli a3,a3,0x20 2c: 03456703 lwu a4,52(a0) 30: 02071713 slli a4,a4,0x20 34: 02075713 srli a4,a4,0x20 38: 03856483 lwu s1,56(a0) 3c: 02049493 slli s1,s1,0x20 40: 0204d493 srli s1,s1,0x20 44: 03c56903 lwu s2,60(a0) 48: 02091913 slli s2,s2,0x20 4c: 02095913 srli s2,s2,0x20 50: 04056983 lwu s3,64(a0) 54: 02099993 slli s3,s3,0x20 58: 0209d993 srli s3,s3,0x20 5c: 09056a03 lwu s4,144(a0) 60: 020a1a13 slli s4,s4,0x20 64: 020a5a13 srli s4,s4,0x20 68: 00900313 addi t1,zero,9 6c: 006a7463 bgeu s4,t1,0x74 70: 00000a13 addi s4,zero,0 74: 02d52823 sw a3,48(a0) 78: 02e52a23 sw a4,52(a0) 7c: 02952c23 sw s1,56(a0) 80: 03252e23 sw s2,60(a0) 84: 05352023 sw s3,64(a0) 88: 00000793 addi a5,zero,0 8c: 02813403 ld s0,40(sp) 90: 02013483 ld s1,32(sp) 94: 01813903 ld s2,24(sp) 98: 01013983 ld s3,16(sp) 9c: 00813a03 ld s4,8(sp) a0: 03010113 addi sp,sp,48 a4: 00078513 addi a0,a5,0 a8: 00008067 jalr zero,0(ra) With RVC: 0: 02000813 addi a6,zero,32 4: 7179 c.addi16sp sp,-48 6: f422 c.sdsp s0,40(sp) 8: f026 c.sdsp s1,32(sp) a: ec4a c.sdsp s2,24(sp) c: e84e c.sdsp s3,16(sp) e: e452 c.sdsp s4,8(sp) 10: 1800 c.addi4spn s0,sp,48 12: 03056683 lwu a3,48(a0) 16: 1682 c.slli a3,0x20 18: 9281 c.srli a3,0x20 1a: 03456703 lwu a4,52(a0) 1e: 1702 c.slli a4,0x20 20: 9301 c.srli a4,0x20 22: 03856483 lwu s1,56(a0) 26: 1482 c.slli s1,0x20 28: 9081 c.srli s1,0x20 2a: 03c56903 lwu s2,60(a0) 2e: 1902 c.slli s2,0x20 30: 02095913 srli s2,s2,0x20 34: 04056983 lwu s3,64(a0) 38: 1982 c.slli s3,0x20 3a: 0209d993 srli s3,s3,0x20 3e: 09056a03 lwu s4,144(a0) 42: 1a02 c.slli s4,0x20 44: 020a5a13 srli s4,s4,0x20 48: 4325 c.li t1,9 4a: 006a7363 bgeu s4,t1,0x50 4e: 4a01 c.li s4,0 50: d914 c.sw a3,48(a0) 52: d958 c.sw a4,52(a0) 54: dd04 c.sw s1,56(a0) 56: 03252e23 sw s2,60(a0) 5a: 05352023 sw s3,64(a0) 5e: 4781 c.li a5,0 60: 7422 c.ldsp s0,40(sp) 62: 7482 c.ldsp s1,32(sp) 64: 6962 c.ldsp s2,24(sp) 66: 69c2 c.ldsp s3,16(sp) 68: 6a22 c.ldsp s4,8(sp) 6a: 6145 c.addi16sp sp,48 6c: 853e c.mv a0,a5 6e: 8082 c.jr ra Signed-off-by: Luke Nelson <luke.r.nels@gmail.com> Signed-off-by: Alexei Starovoitov <ast@kernel.org> Cc: Björn Töpel <bjorn.topel@gmail.com> Link: https://lore.kernel.org/bpf/20200721025241.8077-4-luke.r.nels@gmail.com
2020-07-21 10:52:40 +08:00
emit_mv(RV_REG_TCC_SAVED, RV_REG_TCC, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
ctx->stack_size = stack_adjust;
}
void bpf_jit_build_epilogue(struct rv_jit_context *ctx)
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
{
__build_epilogue(false, ctx);
bpf, riscv: add BPF JIT for RV64G This commit adds a BPF JIT for RV64G. The JIT is a two-pass JIT, and has a dynamic prolog/epilogue (similar to the MIPS64 BPF JIT) instead of static ones (e.g. x86_64). At the moment the RISC-V Linux port does not support CONFIG_HAVE_KPROBES, which means that CONFIG_BPF_EVENTS is not supported. Thus, no tests involving BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_PERF_EVENT, BPF_PROG_TYPE_KPROBE and BPF_PROG_TYPE_RAW_TRACEPOINT passes. The implementation does not support "far branching" (>4KiB). Test results: # modprobe test_bpf test_bpf: Summary: 378 PASSED, 0 FAILED, [366/366 JIT'ed] # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier ... Summary: 761 PASSED, 507 SKIPPED, 2 FAILED Note that "test_verifier" was run with one build with CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y and one without, otherwise many of the the tests that require unaligned access were skipped. CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 0 No CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS: # echo 1 > /proc/sys/kernel/unprivileged_bpf_disabled # ./test_verifier | grep -c 'NOTE.*unknown align' 59 The two failing test_verifier tests are: "ld_abs: vlan + abs, test 1" "ld_abs: jump around ld_abs" This is due to that "far branching" involved in those tests. All tests where done on QEMU (QEMU emulator version 3.1.50 (v3.1.0-688-g8ae951fbc106)). Signed-off-by: Björn Töpel <bjorn.topel@gmail.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-02-05 20:41:22 +08:00
}