Alexandre Ghiti <alexghiti@rivosinc.com> says:
This series optimizes the tlb flushes on riscv which used to simply
flush the whole tlb whatever the size of the range to flush or the size
of the stride.
Patch 3 introduces a threshold that is microarchitecture specific and
will very likely be modified by vendors, not sure though which mechanism
we'll use to do that (dt? alternatives? vendor initialization code?).
* b4-shazam-merge:
riscv: Improve flush_tlb_kernel_range()
riscv: Make __flush_tlb_range() loop over pte instead of flushing the whole tlb
riscv: Improve flush_tlb_range() for hugetlb pages
riscv: Improve tlb_flush()
Link: https://lore.kernel.org/r/20231030133027.19542-1-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Currently, when the range to flush covers more than one page (a 4K page or
a hugepage), __flush_tlb_range() flushes the whole tlb. Flushing the whole
tlb comes with a greater cost than flushing a single entry so we should
flush single entries up to a certain threshold so that:
threshold * cost of flushing a single entry < cost of flushing the whole
tlb.
Co-developed-by: Mayuresh Chitale <mchitale@ventanamicro.com>
Signed-off-by: Mayuresh Chitale <mchitale@ventanamicro.com>
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Tested-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com> # On RZ/Five SMARC
Reviewed-by: Samuel Holland <samuel.holland@sifive.com>
Tested-by: Samuel Holland <samuel.holland@sifive.com>
Link: https://lore.kernel.org/r/20231030133027.19542-4-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Jisheng Zhang <jszhang@kernel.org> says:
This series renews one of my last year RFC patch[1], tries to improve
the vdso layout a bit.
patch1 removes useless symbols
patch2 merges .data section of vdso into .rodata because they are
readonly
patch3 is the real renew patch, it removes hardcoded 0x800 .text start
addr. But I rewrite the commit msg per Andrew's suggestions and move
move .note, .eh_frame_hdr, and .eh_frame between .rodata and .text to
keep the actual code well away from the non-instruction data.
* b4-shazam-merge:
riscv: vdso.lds.S: remove hardcoded 0x800 .text start addr
riscv: vdso.lds.S: merge .data section into .rodata section
riscv: vdso.lds.S: drop __alt_start and __alt_end symbols
Link: https://lore.kernel.org/linux-riscv/20221123161805.1579-1-jszhang@kernel.org/ [1]
Link: https://lore.kernel.org/r/20230912072015.2424-1-jszhang@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
I believe the hardcoded 0x800 and related comments come from the long
history VDSO_TEXT_OFFSET in x86 vdso code, but commit 5b93049337
("x86 vDSO: generate vdso-syms.lds") and commit f6b46ebf90 ("x86
vDSO: new layout") removes the comment and hard coding for x86.
Similar as x86 and other arch, riscv doesn't need the rigid layout
using VDSO_TEXT_OFFSET since it "no longer matters to the kernel".
so we could remove the hard coding now, and removing it brings a
small vdso.so and aligns with other architectures.
Also, having enough separation between data and text is important for
I-cache, so similar as x86, move .note, .eh_frame_hdr, and .eh_frame
between .rodata and .text.
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Tested-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Link: https://lore.kernel.org/r/20230912072015.2424-4-jszhang@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The .data section doesn't need to be separate from .rodata section,
they are both readonly.
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Tested-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Link: https://lore.kernel.org/r/20230912072015.2424-3-jszhang@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
These two symbols are not used, remove them.
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Tested-by: Emil Renner Berthing <emil.renner.berthing@canonical.com>
Link: https://lore.kernel.org/r/20230912072015.2424-2-jszhang@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Instructions can write to x0, so we should simulate these instructions
normally.
Currently, the kernel hangs if an instruction who writes to x0 is
simulated.
Fixes: c22b0bcb1d ("riscv: Add kprobes supported")
Cc: stable@vger.kernel.org
Signed-off-by: Nam Cao <namcaov@gmail.com>
Reviewed-by: Charlie Jenkins <charlie@rivosinc.com>
Acked-by: Guo Ren <guoren@kernel.org>
Link: https://lore.kernel.org/r/20230829182500.61875-1-namcaov@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
uprobes expects is_trap_insn() to return true for any trap instructions,
not just the one used for installing uprobe. The current default
implementation only returns true for 16-bit c.ebreak if C extension is
enabled. This can confuse uprobes if a 32-bit ebreak generates a trap
exception from userspace: uprobes asks is_trap_insn() who says there is no
trap, so uprobes assume a probe was there before but has been removed, and
return to the trap instruction. This causes an infinite loop of entering
and exiting trap handler.
Instead of using the default implementation, implement this function
speficially for riscv with checks for both ebreak and c.ebreak.
Fixes: 74784081aa ("riscv: Add uprobes supported")
Signed-off-by: Nam Cao <namcaov@gmail.com>
Tested-by: Björn Töpel <bjorn@rivosinc.com>
Reviewed-by: Guo Ren <guoren@kernel.org>
Link: https://lore.kernel.org/r/20230829083614.117748-1-namcaov@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
If misaligned_access_speed percpu var isn't so called "HWPROBE
MISALIGNED UNKNOWN", it means the probe has happened(this is possible
for example, hotplug off then hotplug on one cpu), and the percpu var
has been set, don't probe again in this case.
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Fixes: 584ea6564b ("RISC-V: Probe for unaligned access speed")
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230912154040.3306-1-jszhang@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
In the current riscv implementation, blocking syscalls like read() may
not correctly restart after being interrupted by ptrace. This problem
arises when the syscall restart process in arch_do_signal_or_restart()
is bypassed due to changes to the regs->cause register, such as an
ebreak instruction.
Steps to reproduce:
1. Interrupt the tracee process with PTRACE_SEIZE & PTRACE_INTERRUPT.
2. Backup original registers and instruction at new_pc.
3. Change pc to new_pc, and inject an instruction (like ebreak) to this
address.
4. Resume with PTRACE_CONT and wait for the process to stop again after
executing ebreak.
5. Restore original registers and instructions, and detach from the
tracee process.
6. Now the read() syscall in tracee will return -1 with errno set to
ERESTARTSYS.
Specifically, during an interrupt, the regs->cause changes from
EXC_SYSCALL to EXC_BREAKPOINT due to the injected ebreak, which is
inaccessible via ptrace so we cannot restore it. This alteration breaks
the syscall restart condition and ends the read() syscall with an
ERESTARTSYS error. According to include/linux/errno.h, it should never
be seen by user programs. X86 can avoid this issue as it checks the
syscall condition using a register (orig_ax) exposed to user space.
Arm64 handles syscall restart before calling get_signal, where it could
be paused and inspected by ptrace/debugger.
This patch adjusts the riscv implementation to arm64 style, which also
checks syscall using a kernel register (syscallno). It ensures the
syscall restart process is not bypassed when changes to the cause
register occur, providing more consistent behavior across various
architectures.
For a simplified reproduction program, feel free to visit:
https://github.com/ancientmodern/riscv-ptrace-bug-demo.
Signed-off-by: Haorong Lu <ancientmodern4@gmail.com>
Link: https://lore.kernel.org/r/20230803224458.4156006-1-ancientmodern4@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Clément Léger <cleger@rivosinc.com> says:
Since commit 61cadb9 ("Provide new description of misaligned load/store
behavior compatible with privileged architecture.") in the RISC-V ISA
manual, it is stated that misaligned load/store might not be supported.
However, the RISC-V kernel uABI describes that misaligned accesses are
supported. In order to support that, this series adds support for S-mode
handling of misaligned accesses as well support for prctl(PR_UNALIGN).
Handling misaligned access in kernel allows for a finer grain control
of the misaligned accesses behavior, and thanks to the prctl() call,
can allow disabling misaligned access emulation to generate SIGBUS. User
space can then optimize its software by removing such access based on
SIGBUS generation.
This series is useful when using a SBI implementation that does not
handle misaligned traps as well as detecting misaligned accesses
generated by userspace application using the prctrl(PR_SET_UNALIGN)
feature.
This series can be tested using the spike simulator[1] and a modified
openSBI version[2] which allows to always delegate misaligned load/store to
S-mode. A test[3] that exercise various instructions/registers can be
executed to verify the unaligned access support.
[1] https://github.com/riscv-software-src/riscv-isa-sim
[2] https://github.com/rivosinc/opensbi/tree/dev/cleger/no_misaligned
[3] https://github.com/clementleger/unaligned_test
* b4-shazam-merge:
riscv: add support for PR_SET_UNALIGN and PR_GET_UNALIGN
riscv: report misaligned accesses emulation to hwprobe
riscv: annotate check_unaligned_access_boot_cpu() with __init
riscv: add support for sysctl unaligned_enabled control
riscv: add floating point insn support to misaligned access emulation
riscv: report perf event for misaligned fault
riscv: add support for misaligned trap handling in S-mode
riscv: remove unused functions in traps_misaligned.c
Link: https://lore.kernel.org/r/20231004151405.521596-1-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
A hwprobe pair key is signed, but the hwprobe vDSO function was
only checking that the upper bound was valid. In order to help
avoid this type of problem in the future, and in anticipation of
this check becoming more complicated with sparse keys, introduce
and use a "key is valid" predicate function for the check.
Fixes: aa5af0aa90 ("RISC-V: Add hwprobe vDSO function and data")
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Evan Green <evan@rivosinc.com>
Link: https://lore.kernel.org/r/20231010165101.14942-2-ajones@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Sami Tolvanen <samitolvanen@google.com> says:
This series adds Shadow Call Stack (SCS) support for RISC-V. SCS
uses compiler instrumentation to store return addresses in a
separate shadow stack to protect them against accidental or
malicious overwrites. More information about SCS can be found
here:
https://clang.llvm.org/docs/ShadowCallStack.html
Patch 1 is from Deepak, and it simplifies VMAP_STACK overflow
handling by adding support for accessing per-CPU variables
directly in assembly. The patch is included in this series to
make IRQ stack switching cleaner with SCS, and I've simply
rebased it and fixed a couple of minor issues. Patch 2 uses this
functionality to clean up the stack switching by moving duplicate
code into a single function. On RISC-V, the compiler uses the
gp register for storing the current shadow call stack pointer,
which is incompatible with global pointer relaxation. Patch 3
moves global pointer loading into a macro that can be easily
disabled with SCS. Patch 4 implements SCS register loading and
switching, and allows the feature to be enabled, and patch 5 adds
separate per-CPU IRQ shadow call stacks when CONFIG_IRQ_STACKS is
enabled. Patch 6 fixes the backward-edge CFI test in lkdtm for
RISC-V.
Note that this series requires Clang 17. Earlier Clang versions
support SCS on RISC-V, but use the x18 register instead of gp,
which isn't ideal. gcc has SCS support for arm64, but I'm not
aware of plans to support RISC-V. Once the Zicfiss extension is
ratified, it's probably preferable to use hardware-backed shadow
stacks instead of SCS on hardware that supports the extension,
and we may want to consider implementing CONFIG_DYNAMIC_SCS to
patch between the implementation at runtime (similarly to the
arm64 implementation, which switches to SCS when hardware PAC
support isn't available).
* b4-shazam-merge:
lkdtm: Fix CFI_BACKWARD on RISC-V
riscv: Use separate IRQ shadow call stacks
riscv: Implement Shadow Call Stack
riscv: Move global pointer loading to a macro
riscv: Deduplicate IRQ stack switching
riscv: VMAP_STACK overflow detection thread-safe
Link: https://lore.kernel.org/r/20230927224757.1154247-8-samitolvanen@google.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Now that trap support is ready to handle misalignment errors in S-mode,
allow the user to control the behavior of misaligned accesses using
prctl(PR_SET_UNALIGN). Add an align_ctl flag in thread_struct which
will be used to determine if we should SIGBUS the process or not on
such fault.
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-9-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
hwprobe provides a way to report if misaligned access are emulated. In
order to correctly populate that feature, we can check if it actually
traps when doing a misaligned access. This can be checked using an
exception table entry which will actually be used when a misaligned
access is done from kernel mode.
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-8-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This function is solely called as an initcall, thus annotate it with
__init.
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Reviewed-by: Evan Green <evan@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-7-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This sysctl tuning option allows the user to disable misaligned access
handling globally on the system. This will also be used by misaligned
detection code to temporarily disable misaligned access handling.
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-6-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This support is partially based of openSBI misaligned emulation floating
point instruction support. It provides support for the existing
floating point instructions (both for 32/64 bits as well as compressed
ones). Since floating point registers are not part of the pt_regs
struct, we need to modify them directly using some assembly. We also
dirty the pt_regs status in case we modify them to be sure context
switch will save FP state. With this support, Linux is on par with
openSBI support.
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-5-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Misalignment trap handling is only supported for M-mode and uses direct
accesses to user memory. In S-mode, when handling usermode fault, this
requires to use the get_user()/put_user() accessors. Implement
load_u8(), store_u8() and get_insn() using these accessors for
userspace and direct text access for kernel.
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-3-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Replace macros by the only two function calls that are done from this
file, store_u8() and load_u8().
Signed-off-by: Clément Léger <cleger@rivosinc.com>
Link: https://lore.kernel.org/r/20231004151405.521596-2-cleger@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Sunil V L <sunilvl@ventanamicro.com> says:
This series is a set of patches which were originally part of RFC v1 series
[1] to add ACPI support in RISC-V interrupt controllers. Since these
patches are independent of the interrupt controllers, creating this new
series which helps to merge instead of waiting for big series.
This set of patches primarily adds support below ECR [2] which is approved
by the ASWG and adds below features.
- Get CBO block sizes from RHCT on ACPI based systems.
Additionally, the series contains a patch to improve acpi_os_ioremap().
[1] - https://lore.kernel.org/lkml/20230803175202.3173957-1-sunilvl@ventanamicro.com/
[2] - https://drive.google.com/file/d/1sKbOa8m1UZw1JkquZYe3F1zQBN1xXsaf/view?usp=sharing
* b4-shazam-merge:
RISC-V: cacheflush: Initialize CBO variables on ACPI systems
RISC-V: ACPI: RHCT: Add function to get CBO block sizes
RISC-V: ACPI: Update the return value of acpi_get_rhct()
RISC-V: ACPI: Enhance acpi_os_ioremap with MMIO remapping
Link: https://lore.kernel.org/r/20231018124007.1306159-1-sunilvl@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The interrupt entries are expected to be in the .irqentry.text section.
For example, for kprobes to work properly, exception code cannot be
probed; this is ensured by blacklisting addresses in the .irqentry.text
section.
Fixes: 7db91e57a0 ("RISC-V: Task implementation")
Signed-off-by: Nam Cao <namcaov@gmail.com>
Link: https://lore.kernel.org/r/20230821145708.21270-1-namcaov@gmail.com
Cc: stable@vger.kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Extensions prefixed with "Su" won't corrupt the workaround in many
cases. The only exception is when the first multi-letter extension in the
ISA string begins with "Su" and is not prefixed with an underscore.
For instance, following ISA string can confuse this QEMU workaround.
* "rv64imacsuclic" (RV64I + M + A + C + "Suclic")
However, this case is very unlikely because extensions prefixed by either
"Z", "Sm" or "Ss" will most likely precede first.
For instance, the "Suclic" extension (draft as of now) will be placed after
related "Smclic" and "Ssclic" extensions. It's also highly likely that
other unprivileged extensions like "Zba" will precede.
It's also possible to suppress the issue in the QEMU workaround with an
underscore. Following ISA string won't confuse the QEMU workaround.
* "rv64imac_suclic" (RV64I + M + A + C + delimited "Suclic")
This fix is to tell kernel developers the nature of this workaround
precisely. There are some "Su*" extensions to be ratified but don't worry
about this workaround too much.
This commit comes with other minor editorial fixes (for minor wording and
spacing issues, without changing the meaning).
Signed-off-by: Tsukasa OI <research_trasio@irq.a4lg.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/8a127608cf6194a6d288289f2520bd1744b81437.1690350252.git.research_trasio@irq.a4lg.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Andrew Jones <ajones@ventanamicro.com> says:
In order for usermode to issue cbo.zero, it needs privilege granted to
issue the extension instruction (patch 2) and to know that the extension
is available and its block size (patch 3). Patch 1 could be separate from
this series (it just fixes up some error messages), patches 4-5 convert
the hwprobe selftest to a statically-linked, TAP test and patch 6 adds a
new hwprobe test for the new information as well as testing CBO
instructions can or cannot be issued as appropriate.
* b4-shazam-merge:
RISC-V: selftests: Add CBO tests
RISC-V: selftests: Convert hwprobe test to kselftest API
RISC-V: selftests: Statically link hwprobe test
RISC-V: hwprobe: Expose Zicboz extension and its block size
RISC-V: Enable cbo.zero in usermode
RISC-V: Make zicbom/zicboz errors consistent
Link: https://lore.kernel.org/r/20230918131518.56803-8-ajones@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Song Shuai <songshuaishuai@tinylab.org> says:
This series contains a cleanup for riscv_kexec_relocate() and two fixups
for KEXEC_FILE and had passed the basic kexec test in my 64bit Qemu-virt.
You can use this kexec-tools[3] to test the kexec-file-syscall and these patches.
riscv: kexec: Cleanup riscv_kexec_relocate (patch1)
==================================================
For readability and simplicity, cleanup the riscv_kexec_relocate code:
- Re-sort the first 4 `mv` instructions against `riscv_kexec_method()`
- Eliminate registers for debugging (s9,s10,s11) and storing const-value (s5,s6)
- Replace `jalr` with `jr` for no-link jump
riscv: kexec: Align the kexeced kernel entry (patch2)
==================================================
The current riscv boot protocol requires 2MB alignment for RV64
and 4MB alignment for RV32.
In KEXEC_FILE path, the elf_find_pbase() function should align
the kexeced kernel entry according to the requirement, otherwise
the kexeced kernel would silently BUG at the setup_vm().
riscv: kexec: Remove -fPIE for PURGATORY_CFLAGS (patch3)
==================================================
With CONFIG_RELOCATABLE enabled, KBUILD_CFLAGS had a -fPIE option
and then the purgatory/string.o was built to reference _ctype symbol
via R_RISCV_GOT_HI20 relocations which can't be handled by purgatory.
As a consequence, the kernel failed kexec_load_file() with:
[ 880.386562] kexec_image: The entry point of kernel at 0x80200000
[ 880.388650] kexec_image: Unknown rela relocation: 20
[ 880.389173] kexec_image: Error loading purgatory ret=-8
So remove the -fPIE option for PURGATORY_CFLAGS to generate
R_RISCV_PCREL_HI20 relocations type making puragtory work as it was.
arch/riscv/kernel/elf_kexec.c | 8 ++++-
arch/riscv/kernel/kexec_relocate.S | 52 +++++++++++++-----------------
arch/riscv/purgatory/Makefile | 4 +++
3 files changed, 34 insertions(+), 30 deletions(-)
* b4-shazam-merge:
riscv: kexec: Remove -fPIE for PURGATORY_CFLAGS
riscv: kexec: Align the kexeced kernel entry
riscv: kexec: Cleanup riscv_kexec_relocate
Link: https://lore.kernel.org/r/20230907103304.590739-1-songshuaishuai@tinylab.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
When both CONFIG_IRQ_STACKS and SCS are enabled, also use a separate
per-CPU shadow call stack.
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/r/20230927224757.1154247-13-samitolvanen@google.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Implement CONFIG_SHADOW_CALL_STACK for RISC-V. When enabled, the
compiler injects instructions to all non-leaf C functions to
store the return address to the shadow stack and unconditionally
load it again before returning, which makes it harder to corrupt
the return address through a stack overflow, for example.
The active shadow call stack pointer is stored in the gp
register, which makes SCS incompatible with gp relaxation. Use
--no-relax-gp to ensure gp relaxation is disabled and disable
global pointer loading. Add SCS pointers to struct thread_info,
implement SCS initialization, and task switching
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/r/20230927224757.1154247-12-samitolvanen@google.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
In Clang 17, -fsanitize=shadow-call-stack uses the newly declared
platform register gp for storing shadow call stack pointers. As
this is obviously incompatible with gp relaxation, in preparation
for CONFIG_SHADOW_CALL_STACK support, move global pointer loading
to a single macro, which we can cleanly disable when SCS is used
instead.
Link: https://reviews.llvm.org/rGaa1d2693c256
Link: a484e843e6
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/r/20230927224757.1154247-11-samitolvanen@google.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
With CONFIG_IRQ_STACKS, we switch to a separate per-CPU IRQ stack
before calling handle_riscv_irq or __do_softirq. We currently
have duplicate inline assembly snippets for stack switching in
both code paths. Now that we can access per-CPU variables in
assembly, implement call_on_irq_stack in assembly, and use that
instead of redundant inline assembly.
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Guo Ren <guoren@kernel.org>
Link: https://lore.kernel.org/r/20230927224757.1154247-10-samitolvanen@google.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Enhance the acpi_os_ioremap() to support opregions in MMIO space. Also,
have strict checks using EFI memory map to allow remapping the RAM similar
to arm64.
Signed-off-by: Sunil V L <sunilvl@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20231018124007.1306159-2-sunilvl@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Expose Zicboz through hwprobe and also provide a key to extract its
respective block size. Opportunistically add a macro and apply it to
current extensions in order to avoid duplicating code.
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Evan Green <evan@rivosinc.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230918131518.56803-11-ajones@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
When Zicboz is present, enable its instruction (cbo.zero) in
usermode by setting its respective senvcfg bit. We don't bother
trying to set this bit per-task, which would also require an
interface for tasks to request enabling and/or disabling. Instead,
permanently set the bit for each hart which has the extension when
bringing it online.
This patch also introduces riscv_cpu_has_extension_[un]likely()
functions to check a specific hart's ISA bitmap for extensions.
Prior to checking the specific hart's bitmap in these functions
we try the bitmap which represents the LCD of extensions, but only
when we know it will use its optimized, alternatives path by gating
its call on CONFIG_RISCV_ALTERNATIVE. When alternatives are used, the
compiler ensures that the invocation of the LCD search becomes a
constant true or false. When it's true, even the new functions will
completely vanish from their callsites. OTOH, when the LCD check is
false, we need to do a search of the hart's ISA bitmap. Had we also
checked the LCD bitmap without the use of alternatives, then we would
have ended up with two bitmap searches instead of one.
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230918131518.56803-10-ajones@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
commit c818fea83d ("riscv: say disabling zicbom if no or bad
riscv,cbom-block-size found") improved the error messages for
zicbom but zicboz was missed since its patches were in flight
at the same time. Get 'em now.
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230918131518.56803-9-ajones@ventanamicro.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The current riscv boot protocol requires 2MB alignment for RV64
and 4MB alignment for RV32.
In KEXEC_FILE path, the elf_find_pbase() function should align
the kexeced kernel entry according to the requirement, otherwise
the kexeced kernel would silently BUG at the setup_vm().
Fixes: 8acea455fa ("RISC-V: Support for kexec_file on panic")
Signed-off-by: Song Shuai <songshuaishuai@tinylab.org>
Link: https://lore.kernel.org/r/20230907103304.590739-3-songshuaishuai@tinylab.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
For readability and simplicity, cleanup the riscv_kexec_relocate code:
- Re-sort the first 4 `mv` instructions against `riscv_kexec_method()`
- Eliminate registers for debugging (s9,s10,s11) and storing const-value (s5,s6)
- Replace `jalr` with `jr` for no-link jump
I tested this on Qemu virt machine and works as it was.
Signed-off-by: Song Shuai <songshuaishuai@tinylab.org>
Link: https://lore.kernel.org/r/20230907103304.590739-2-songshuaishuai@tinylab.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Puranjay Mohan <puranjay12@gmail.com> says:
Here is some data to prove the V2 fixes the problem:
Without this series:
root@rv-selftester:~/src/kselftest/bpf# time ./test_tag
test_tag: OK (40945 tests)
real 7m47.562s
user 0m24.145s
sys 6m37.064s
With this series applied:
root@rv-selftester:~/src/selftest/bpf# time ./test_tag
test_tag: OK (40945 tests)
real 7m29.472s
user 0m25.865s
sys 6m18.401s
BPF programs currently consume a page each on RISCV. For systems with many BPF
programs, this adds significant pressure to instruction TLB. High iTLB pressure
usually causes slow down for the whole system.
Song Liu introduced the BPF prog pack allocator[1] to mitigate the above issue.
It packs multiple BPF programs into a single huge page. It is currently only
enabled for the x86_64 BPF JIT.
I enabled this allocator on the ARM64 BPF JIT[2]. It is being reviewed now.
This patch series enables the BPF prog pack allocator for the RISCV BPF JIT.
======================================================
Performance Analysis of prog pack allocator on RISCV64
======================================================
Test setup:
===========
Host machine: Debian GNU/Linux 11 (bullseye)
Qemu Version: QEMU emulator version 8.0.3 (Debian 1:8.0.3+dfsg-1)
u-boot-qemu Version: 2023.07+dfsg-1
opensbi Version: 1.3-1
To test the performance of the BPF prog pack allocator on RV, a stresser
tool[4] linked below was built. This tool loads 8 BPF programs on the system and
triggers 5 of them in an infinite loop by doing system calls.
The runner script starts 20 instances of the above which loads 8*20=160 BPF
programs on the system, 5*20=100 of which are being constantly triggered.
The script is passed a command which would be run in the above environment.
The script was run with following perf command:
./run.sh "perf stat -a \
-e iTLB-load-misses \
-e dTLB-load-misses \
-e dTLB-store-misses \
-e instructions \
--timeout 60000"
The output of the above command is discussed below before and after enabling the
BPF prog pack allocator.
The tests were run on qemu-system-riscv64 with 8 cpus, 16G memory. The rootfs
was created using Bjorn's riscv-cross-builder[5] docker container linked below.
Results
=======
Before enabling prog pack allocator:
------------------------------------
Performance counter stats for 'system wide':
4939048 iTLB-load-misses
5468689 dTLB-load-misses
465234 dTLB-store-misses
1441082097998 instructions
60.045791200 seconds time elapsed
After enabling prog pack allocator:
-----------------------------------
Performance counter stats for 'system wide':
3430035 iTLB-load-misses
5008745 dTLB-load-misses
409944 dTLB-store-misses
1441535637988 instructions
60.046296600 seconds time elapsed
Improvements in metrics
=======================
It was expected that the iTLB-load-misses would decrease as now a single huge
page is used to keep all the BPF programs compared to a single page for each
program earlier.
--------------------------------------------
The improvement in iTLB-load-misses: -30.5 %
--------------------------------------------
I repeated this expriment more than 100 times in different setups and the
improvement was always greater than 30%.
This patch series is boot tested on the Starfive VisionFive 2 board[6].
The performance analysis was not done on the board because it doesn't
expose iTLB-load-misses, etc. The stresser program was run on the board to test
the loading and unloading of BPF programs
[1] https://lore.kernel.org/bpf/20220204185742.271030-1-song@kernel.org/
[2] https://lore.kernel.org/all/20230626085811.3192402-1-puranjay12@gmail.com/
[3] https://lore.kernel.org/all/20230626085811.3192402-2-puranjay12@gmail.com/
[4] https://github.com/puranjaymohan/BPF-Allocator-Bench
[5] https://github.com/bjoto/riscv-cross-builder
[6] https://www.starfivetech.com/en/site/boards
* b4-shazam-merge:
bpf, riscv: use prog pack allocator in the BPF JIT
riscv: implement a memset like function for text
riscv: extend patch_text_nosync() for multiple pages
bpf: make bpf_prog_pack allocator portable
Link: https://lore.kernel.org/r/20230831131229.497941-1-puranjay12@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Alexandre Ghiti <alexghiti@rivosinc.com> says:
The following KASLR implementation allows to randomize the kernel mapping:
- virtually: we expect the bootloader to provide a seed in the device-tree
- physically: only implemented in the EFI stub, it relies on the firmware to
provide a seed using EFI_RNG_PROTOCOL. arm64 has a similar implementation
hence the patch 3 factorizes KASLR related functions for riscv to take
advantage.
The new virtual kernel location is limited by the early page table that only
has one PUD and with the PMD alignment constraint, the kernel can only take
< 512 positions.
* b4-shazam-merge:
riscv: libstub: Implement KASLR by using generic functions
libstub: Fix compilation warning for rv32
arm64: libstub: Move KASLR handling functions to kaslr.c
riscv: Dump out kernel offset information on panic
riscv: Introduce virtual kernel mapping KASLR
Link: https://lore.kernel.org/r/20230722123850.634544-1-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This resurrects the vector ptrace() support that was removed for 6.5 due
to some bugs cropping up as part of the GDB review process.
* b4-shazam-merge:
RISC-V: Add ptrace support for vectors
Link: https://lore.kernel.org/r/20230825050248.32681-1-andy.chiu@sifive.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Prabhakar <prabhakar.csengg@gmail.com> says:
From: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
non-coherent DMA support for AX45MP
====================================
On the Andes AX45MP core, cache coherency is a specification option so it
may not be supported. In this case DMA will fail. To get around with this
issue this patch series does the below:
1] Andes alternative ports is implemented as errata which checks if the
IOCP is missing and only then applies to CMO errata. One vendor specific
SBI EXT (ANDES_SBI_EXT_IOCP_SW_WORKAROUND) is implemented as part of
errata.
Below are the configs which Andes port provides (and are selected by
RZ/Five):
- ERRATA_ANDES
- ERRATA_ANDES_CMO
OpenSBI patch supporting ANDES_SBI_EXT_IOCP_SW_WORKAROUND SBI is now
part v1.3 release.
2] Andes AX45MP core has a Programmable Physical Memory Attributes (PMA)
block that allows dynamic adjustment of memory attributes in the runtime.
It contains a configurable amount of PMA entries implemented as CSR
registers to control the attributes of memory locations in interest.
OpenSBI configures the PMA regions as required and creates a reserve memory
node and propagates it to the higher boot stack.
Currently OpenSBI (upstream) configures the required PMA region and passes
this a shared DMA pool to Linux.
reserved-memory {
#address-cells = <2>;
#size-cells = <2>;
ranges;
pma_resv0@58000000 {
compatible = "shared-dma-pool";
reg = <0x0 0x58000000 0x0 0x08000000>;
no-map;
linux,dma-default;
};
};
The above shared DMA pool gets appended to Linux DTB so the DMA memory
requests go through this region.
3] We provide callbacks to synchronize specific content between memory and
cache.
4] RZ/Five SoC selects the below configs
- AX45MP_L2_CACHE
- DMA_GLOBAL_POOL
- ERRATA_ANDES
- ERRATA_ANDES_CMO
----------x---------------------x--------------------x---------------x----
* b4-shazam-merge:
soc: renesas: Kconfig: Select the required configs for RZ/Five SoC
cache: Add L2 cache management for Andes AX45MP RISC-V core
dt-bindings: cache: andestech,ax45mp-cache: Add DT binding documentation for L2 cache controller
riscv: mm: dma-noncoherent: nonstandard cache operations support
riscv: errata: Add Andes alternative ports
riscv: asm: vendorid_list: Add Andes Technology to the vendors list
Link: https://lore.kernel.org/r/20230818135723.80612-1-prabhakar.mahadev-lad.rj@bp.renesas.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Evan Green <evan@rivosinc.com> says:
The current setting for the hwprobe bit indicating misaligned access
speed is controlled by a vendor-specific feature probe function. This is
essentially a per-SoC table we have to maintain on behalf of each vendor
going forward. Let's convert that instead to something we detect at
runtime.
We have two assembly routines at the heart of our probe: one that
does a bunch of word-sized accesses (without aligning its input buffer),
and the other that does byte accesses. If we can move a larger number of
bytes using misaligned word accesses than we can with the same amount of
time doing byte accesses, then we can declare misaligned accesses as
"fast".
The tradeoff of reducing this maintenance burden is boot time. We spend
4-6 jiffies per core doing this measurement (0-2 on jiffie edge
alignment, and 4 on measurement). The timing loop was based on
raid6_choose_gen(), which uses (16+1)*N jiffies (where N is the number
of algorithms). By taking only the fastest iteration out of all
attempts for use in the comparison, variance between runs is very low.
On my THead C906, it looks like this:
[ 0.047563] cpu0: Ratio of byte access time to unaligned word access is 4.34, unaligned accesses are fast
Several others have chimed in with results on slow machines with the
older algorithm, which took all runs into account, including noise like
interrupts. Even with this variation, results indicate that in all cases
(fast, slow, and emulated) the measured numbers are nowhere near each
other (always multiple factors away).
* b4-shazam-merge:
RISC-V: alternative: Remove feature_probe_func
RISC-V: Probe for unaligned access speed
Link: https://lore.kernel.org/r/20230818194136.4084400-1-evan@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The BPF JIT needs to write invalid instructions to RX regions of memory to
invalidate removed BPF programs. This needs a function like memset() that
can work with RX memory.
Implement patch_text_set_nosync() which is similar to text_poke_set() of
x86.
Signed-off-by: Puranjay Mohan <puranjay12@gmail.com>
Reviewed-by: Pu Lehui <pulehui@huawei.com>
Acked-by: Björn Töpel <bjorn@kernel.org>
Tested-by: Björn Töpel <bjorn@rivosinc.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20230831131229.497941-4-puranjay12@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The patch_insn_write() function currently doesn't work for multiple pages
of instructions, therefore patch_text_nosync() will fail with a page fault
if called with lengths spanning multiple pages.
This commit extends the patch_insn_write() function to support multiple
pages by copying at max 2 pages at a time in a loop. This implementation
is similar to text_poke_copy() function of x86.
Signed-off-by: Puranjay Mohan <puranjay12@gmail.com>
Reviewed-by: Pu Lehui <pulehui@huawei.com>
Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
Tested-by: Björn Töpel <bjorn@rivosinc.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20230831131229.497941-3-puranjay12@gmail.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
We can now use arm64 functions to handle the move of the kernel physical
mapping: if KASLR is enabled, we will try to get a random seed from the
firmware, if not possible, the kernel will be moved to a location that
suits its alignment constraints.
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com>
Tested-by: Song Shuai <songshuaishuai@tinylab.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/r/20230722123850.634544-6-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Dump out the KASLR virtual kernel offset when panic to help debug kernel.
Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com>
Tested-by: Song Shuai <songshuaishuai@tinylab.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/r/20230722123850.634544-3-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
KASLR implementation relies on a relocatable kernel so that we can move
the kernel mapping.
The seed needed to virtually move the kernel is taken from the device tree,
so we rely on the bootloader to provide a correct seed. Zkr could be used
unconditionnally instead if implemented, but that's for another patch.
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com>
Tested-by: Song Shuai <songshuaishuai@tinylab.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/r/20230722123850.634544-2-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This patch add back the ptrace support with the following fix:
- Define NT_RISCV_CSR and re-number NT_RISCV_VECTOR to prevent
conflicting with gdb's NT_RISCV_CSR.
- Use struct __riscv_v_regset_state to handle ptrace requests
Since gdb does not directly include the note description header in
Linux and has already defined NT_RISCV_CSR as 0x900, we decide to
sync with gdb and renumber NT_RISCV_VECTOR to solve and prevent future
conflicts.
Fixes: 0c59922c76 ("riscv: Add ptrace vector support")
Signed-off-by: Andy Chiu <andy.chiu@sifive.com>
Link: https://lore.kernel.org/r/20230825050248.32681-1-andy.chiu@sifive.com
[Palmer: Drop the unused "size" variable in riscv_vr_set().]
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Add required ports of the Alternative scheme for Andes CPU cores.
I/O Coherence Port (IOCP) provides an AXI interface for connecting external
non-caching masters, such as DMA controllers. IOCP is a specification
option and is disabled on the Renesas RZ/Five SoC due to this reason cache
management needs a software workaround.
Signed-off-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com> # tyre-kicking on a d1
Link: https://lore.kernel.org/r/20230818135723.80612-3-prabhakar.mahadev-lad.rj@bp.renesas.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>