Just like other insns with GPR operands, CRC32 with only register
operands should not get a suffix added unless in suffix-always mode.
Do away with CRC32_Fixup() altogether, using other more generic logic
instead.
Unlike for non-zero values passed to USED_REX(), where rex_used gets
updated only when the respective bit was actually set in the encoding,
zero getting passed in is not further guarded, yet such a (potentially
"empty") REX prefix takes effect only when there are registers numbered
4 and up.
There's only a very limited set of modes that this function gets invoked
with - avoid it being more generic than it needs to be. This may, down
the road, allow actually doing away with the function altogether.
This eliminates a first improperly used "USED_REX (0)".
While some insns support both XOP.W based operand swapping and 256-bit
operation (XOP.L=1), many others don't support one or both.
For {L,S}LWPCB also fix the so far not decoded ModRM.mod == 3
restriction.
Take the opportunity and replace the custom OP_LWP_E() and OP_LWPCB_E()
routines by suitable other, non-custom operanbd specifiers.
Just like other VEX-encoded scalar insns do.
Besides a testcase for this behavior also introduce one to verify that
XOP scalar insns don't honor -mavxscalar=256, as they don't ignore
XOP.L.
There's no need for custom operand handling here, except for the VEX.W
controlled operand swapping and the printing of the remaining 4-bit
immediate. VEX.W can be handled just like 4-operand insns.
Also take the opportunity and drop the stray indirection through
vex_w_table[].
There's no need for custom operand handling here, except for the VEX.W
controlled operand swapping. The latter can be easily integrated into
OP_REG_VexI4().
The unnecessary XOP.L decoding had caught my eye, together with the not
really expected operand specifiers. Drop this decode step, and instead
make sure XOP.W and XOP.PP don't get ignored. For the latter, do this in
a form applicable to all XOP insns, rather than adding extra table
layers - there are no encodings with the field non-zero. Besides these
two, for the scalar forms XOP.L actually needs to also be zero.
Since we have these macros, there's no point having unnecessary table
depth.
VFPCLASSP{S,D} are now the first instance of using two %-prefixed
macros, which has pointed out a problem with the implementation. Instead
of using custom code in various case blocks, do the macro accumulation
centralized at the top of the main loop of putop(), and zap the
accumulated macros at the bottom of that loop once it has been
processed.
For all of these only the 512-bit forms are valid, so drop 256-bit ones
from the integer insert/extract variants.
Also replace EXxmmq by the more natural (here) EXymm.
Just like their AVX counterparts do for VEX.L.
At this occasion also make EVEX.W have the same effect as VEX.W on the
printing of VPINSR{B,W}'s operands, bringing them also in sync with
VPEXTR{B,W}.
Unlike for the EVEX-encoded versions, the VEX ones failed to decode
VEX.W. Once the necessary adjustments are done, it becomes obvious that
the EVEX and VEX table entries for VCVTPS2PH are identical and can hence
be folded.
The duplication is not only space inefficient, but also risks entries
going out of sync (some of which that I became aware of while doing this
work will get addressed subsequently). Right here note that for
VGF2P8MULB this also addresses the prior lack of EVEX.W decoding (i.e. a
first example of out of sync entries).
This introduces EXxEVexR to some VEX templates, on the basis that this
operand is benign there and only relevant when EVEX encoding ends up
reaching these entries.
There's only a single user, that that one can do fine with the
alternative, as the "Vex" aspect of the other operand kind is meaningful
only on 3-operand insns.
While doing this I noticed that I didn't need to do the same adjustment
in the EVEX tables, and voilà - there was a bug, which gets fixed at the
same time (see the testsuite changes).
Along the lines of 4102be5cf9 ("x86: replace EXxmm_mdq by
EXVexWdqScalar"), but in the opposite direction, replace EXdScalar/
EXqScalar by EXxmm_md/EXxmm_mq respectively, rendering d_scalar_mode and
q_scalar_mode unused. The change is done this way to improve telling
apart operands affected here from ones using EXbScalar/EXwScalar, which
work sufficiently differently. Additionally this increases similarity
between several VEX-encoded insns and their EVEX-encoded counterparts.
We check register-only source operand to decide if two source operands of
VEX encoded instructions should be swapped. But source operands in AMX
instructions with two source operands swapped are all register-only
operand. Add SwapSources to indicate two source operands should be
swapped.
gas/
* config/tc-i386.c (build_modrm_byte): Check vexswapsources to
swap two source operands.
opcodes/
* i386-gen.c (opcode_modifiers): Add VexSwapSources.
* i386-opc.h (VexSwapSources): New.
(i386_opcode_modifier): Add vexswapsources.
* i386-opc.tbl: Add VexSwapSources to BMI2 and BMI instructions
with two source operands swapped.
* i386-tbl.h: Regenerated.
The unprivileged CSR should be controlled by other specific specs rather
than the privileged spec. For example, the debug CSR should be controlled
by the debug spec, and the float CSR should be controlled by the float
spec. User may use assembler options to choose what the debug and other
specs they want, or may encode the versions of specs into the architecture
string directly. Since we haven't decided which one is better, we set the
defined and aborted versions of unprivileged CSR to PRIV_SPEC_CLASS_NONE
in the include/opcode/riscv-opc.h, to tell assembler don't check priv spec
versions for them. However, these PRIV_SPEC_CLASS_NONE will be changed
to FLOAT_SPEC_CLASS_* and DEBUG_SPEC_CLASS_* in the future.
gas/
* config/tc-riscv.c (riscv_csr_class_check): Removed. Move the
checking into riscv_csr_address.
(riscv_csr_version_check): Likewise.
(riscv_csr_address): New function. Return the suitable CSR address
after checking the ISA dependency and versions. Issue warnings if
we find any conflict and -mcsr-check is set. CSR_CLASS_F and
CSR_CLASS_DEBUG are unprivileged CSR for now, so don't check the
priv spec versions for them.
(reg_csr_lookup_internal): Call riscv_csr_address to find the
suitable CSR address.
* testsuite/gas/riscv/priv-reg-fail-fext.d: Remove -mpriv-spec=1.11.
* testsuite/gas/riscv/priv-reg-fail-read-only-01.d: Likewise.
* testsuite/gas/riscv/priv-reg-fail-rv32-only.d: Likewise.
* testsuite/gas/riscv/priv-reg-fail-fext.l: We don't care the
priv spec warnings here. These warnings are added by accident.
Remove them and only focus on the ISA dependency warnings.
* testsuite/gas/riscv/priv-reg-fail-rv32-only.l: Likewise.
* testsuite/gas/riscv/priv-reg-fail-read-only-01.l: Likewise.
* testsuite/gas/riscv/priv-reg-fail-version-1p9.l: Updated since
dscratch0 and dscratch1 are regarded as the unprivileged CSR rather
than the privileged ones.
* testsuite/gas/riscv/priv-reg-fail-version-1p9p1.l: Likewise.
* testsuite/gas/riscv/priv-reg-fail-version-1p10.l: Likewise.
* testsuite/gas/riscv/priv-reg-fail-version-1p11.l: Likewise.
* testsuite/gas/riscv/priv-reg.s: Likewise. Add missing debug CSR.
* testsuite/gas/riscv/priv-reg-version-1p9.d: Likewise.
* testsuite/gas/riscv/priv-reg-version-1p9p1.d: Likewise.
* testsuite/gas/riscv/priv-reg-version-1p10.d: Likewise.
* testsuite/gas/riscv/priv-reg-version-1p11.d: Likewise.
* testsuite/gas/riscv/csr-dw-regnums.d: Likewise.
* testsuite/gas/riscv/csr-dw-regnums.s: Likewise.
include/
* opcode/riscv-opc.h: Support the unprivileged CSR. The versions
of the unprivileged CSR should be PRIV_SPEC_CLASS_NONE for now.
* opcode/riscv.h (enum riscv_csr_class): Add CSR_CLASS_DEBUG.
opcodes/
* riscv-dis.c (print_insn_args, case 'E'): Updated. Let the
unprivileged CSR can also be initialized.
Rename VecSIB to SIB to support Intel Advanced Matrix Extensions which
introduces instructions with a mandatory SIB byte which isn't a vector
SIB (VSIB).
gas/
* config/tc-i386.c (check_VecOperands): Replace vecsib with sib.
Replace VecSIB128, VecSIB256 and VecSIB512 with VECSIB128,
VECSIB256 and VECSIB512, respectively.
(build_modrm_byte): Replace vecsib with sib.
opcodes/
* i386-gen.c (opcode_modifiers): Replace VecSIB with SIB.
(VecSIB128): Renamed to ...
(VECSIB128): This.
(VecSIB256): Renamed to ...
(VECSIB256): This.
(VecSIB512): Renamed to ...
(VECSIB512): This.
(VecSIB): Renamed to ...
(SIB): This.
(i386_opcode_modifier): Replace vecsib with sib.
* i386-opc.tbl (VexSIB128): New.
(VecSIB256): Likewise.
(VecSIB512): Likewise.
Replace VecSIB=1, VecSIB=2 and VecSIB=3 with VexSIB128, VecSIB256
and VecSIB512, respectively.
The {Q|Q} construct on lgdt/lidt and the slightly different {Q|IQ}
struck me as odd when encountering. Realizing they both have the same
effect, let's free up the I macro by using the former form of construct
where needed (there aren't that many uses overall).
With there now being several multi-character macros also re-do "alt"
handling slightly: Terminate it when finding the closing brace, rather
than after the next single character. Also set the flag only when
actually in Intel syntax mode.
Multiple -M options can be specified in any order. Therefore stright
assignment to fields affected needs to be avoided, such that earlier
options' effects won't be discarded. This was in particular a problem
for -Msuffix followed by certain of the other sub-options.
While updating documentation, take the liberty and also drop the
redundant mentioning of being able to comma-separate multiple options.
There's clearly a shortage of available macro characters, as can be seen
from the various two-character macros that had to be introduced. Don't
waste characters for things that can be expressed differently. In the
case of J this alternative is {l|}.
SYSRET can't use the same macro as IRET, since there's no 16-bit operand
size form of it. Re-use LQ for it instead.
Doing so made obvious that outside of 64-bit mode {,V}CVTSI2S{S,D} and
PTWRITE should have an 'l' suffix printed only in suffix-always mode.
We do know some conflicts among different privileged specs. For linker,
the safest approach is that don't allow the object linked with others which
may cause conflicts. But this may cause inconvenience since not all objects
with conflicting priv specs are linked will cause problems. But it is hard
to know the detailed conflict cases for linker, so we probably need a option
to tell linker that we do know there are no conflicts, or we are willing to
take risks to link the objects with conflicted priv specs. But the option
is still under discussion.
Therefore, we can report warnings rather than errors when linking the objects
with conflicted priv specs. This not only makes the linker more flexible,
but also warns people that the conflicts may happen. We also need to update
the output priv spec version once the input priv spec is newer.
bfd/
* elfxx-riscv.c (struct priv_spec_t priv_specs[]): Move them from
opcodes/riscv-opc.c to bfd/elfxx-riscv.c, since we need it in linker.
(riscv_get_priv_spec_class): Likewise.
(riscv_get_priv_spec_name): Likewise.
(riscv_get_priv_spec_class_from_numbers): New function, convert
the version numbers into string, then call riscv_get_priv_spec_class
to get the priv spec class.
* elfxx-riscv.h (riscv_get_priv_spec_class): Move forward declaration
from include/opcode/riscv.h to bfd/elfxx-riscv.h.
(riscv_get_priv_spec_name): Likewise.
(riscv_get_priv_spec_class_from_numbers): New forward declaration.
(opcode/riscv.h): Include it in the header rather than elfxx-riscv.c.
* elfnn-riscv.c (riscv_merge_attributes): Get the priv spec classes
of input and output objects form their priv spec attributes by
riscv_get_priv_spec_class_from_numbers. Report warning rather than
errors when linking objects with differnet priv spec versions. We do
know v1.9.1 may have conflicts to other versions, so report the
warning, too. After that, update the output priv spec version to the
newest one so far.
gas/
* config/tc-riscv.c (buf_size, buf): Remove the unused variables.
(riscv_set_default_priv_spec): Get the priv spec version from the
priv spec attributes by riscv_get_priv_spec_class_from_numbers.
include/
* opcode/riscv.h (riscv_get_priv_spec_class): Move the function
forward declarations to bfd/elfxx-riscv.h.
(riscv_get_priv_spec_name): Likewise.
opcodes/
* riscv-opc.c: Move the structures and functions to bfd/elfxx-riscv.c.
* riscv-dis.c: Include elfxx-riscv.h.
ld/
* testsuite/ld-riscv-elf/attr-merge-priv-spec-failed-01.d: Updated.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-failed-02.d: Updated.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-failed-03.d: Updated.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-failed-04.d: Updated.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-failed-05.d: Updated.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-failed-06.d: Updated.
There is a conflict between v1.9 and v1.9.1 - CSR MISA address. MISA is
0xf10 in v1.9, but change to 0x301 in v1.9.1. The change made MISA writable,
but may also cause risk of compatibility. Binutils already support the
-mpriv-spec options and ELF priv attributes, which can used to choose what
privileged spec you want, and then give a correponding CSR name and address
to use. But Gdb and other tools don't have the simialr mechanism for now.
However, there are two things can be confirmed,
1. If we don't have a way to control the priv specs, then the changes, like
MISA, will cause risk and hard to maintain.
2. We get the guarantee that the CSR address won't be reused in the future
specs, even if it is dropped.
I'm not sure if Gdb needs to care about the priv spec versions, it is still
discussing. But drop the priv spec v1.9, and make sure that we won't reuse
the CSR address is a useful solution for now. Also, we might drop the v1.9.1
in a year or two. After that, specs above v1.10 should be compatible anyway.
gas/
* testsuite/gas/riscv/priv-reg-fail-version-1p9.d: Removed.
* testsuite/gas/riscv/priv-reg-fail-version-1p9.l: Likewise.
* testsuite/gas/riscv/priv-reg-version-1p9.d: Likewise.
include/
* opcode/riscv-opc.h: Update the defined versions of CSR from
PRIV_SPEC_CLASS_1P9 to PRIV_SPEC_CLASS_1P9P1. Also, drop the
MISA DECLARE_CSR_ALIAS since it's aborted version is v1.9.
* opcode/riscv.h (enum riscv_priv_spec_class): Remove
PRIV_SPEC_CLASS_1P9.
opcodes/
* riscv-opc.c (priv_specs): Remove v1.9 and PRIV_SPEC_CLASS_1P9.
Prior to this patch, the information describing the AArch64 system
registers was separate from the information describing which system
registers are available depending on the CPU feature set. Indeed, the
latter was implemented as a separate function from the main table with
the system register information.
This patch remedies this situation and puts the feature information into
the system register table itself.
This has several advantages:
* Having all the information described in one place is easier to
maintain.
* The logic to check whether a system register is supported now becomes
trivial (and much more efficient).
Since this patch ended up touching every line of the system register
table, I took the opportunity to make the formatting more consistent and
remove some redundant comments.
Note that there is still more refactoring that could be done along the
same lines here (e.g. with the TLB instructions) but this seemed like a
reasonable first pass.
Testing:
* Regression tested an x64 -> aarch64-none-elf cross binutils.
* Built aarch64-none-elf cross toolchain, checked newlib startup
code still works.
* Bootstrapped binutils on aarch64-linux-gnu, regression tested.
* Built aarch64 kernel using new binutils with allyesconfig.
OK for master? If so, I'll need a maintainer to commit on my behalf
since I don't have write access.
Thanks,
Alex
---
include/ChangeLog:
2020-06-11 Alex Coplan <alex.coplan@arm.com>
* opcode/aarch64.h (aarch64_sys_reg): Add required features to struct
describing system registers.
opcodes/ChangeLog:
2020-06-11 Alex Coplan <alex.coplan@arm.com>
* aarch64-opc.c (SYSREG): New macro for describing system registers.
(SR_CORE): Likewise.
(SR_FEAT): Likewise.
(SR_RNG): Likewise.
(SR_V8_1): Likewise.
(SR_V8_2): Likewise.
(SR_V8_3): Likewise.
(SR_V8_4): Likewise.
(SR_PAN): Likewise.
(SR_RAS): Likewise.
(SR_SSBS): Likewise.
(SR_SVE): Likewise.
(SR_ID_PFR2): Likewise.
(SR_PROFILE): Likewise.
(SR_MEMTAG): Likewise.
(SR_SCXTNUM): Likewise.
(aarch64_sys_regs): Refactor to store feature information in the table.
(aarch64_sys_reg_supported_p): Collapse logic for system registers
that now describe their own features.
(aarch64_pstatefield_supported_p): Likewise.
All of data size, rep, lock, and rex prefixes are invalid with VEX- and
alike encoded insns. Make sure they get printed explicitly in all cases,
to signal the anomaly. With this, do away with "rex_ignored" - if there
is a rex prefix, we want to print it anyway for VEX etc (and there's
nothing "ignored" about it in the first place - such an instruction will
raise #UD).
Neither the legacy nor the VEX-encoded forms are permitted with register
operands, just like is already the case for their store forms as well as
{,V}MOV{L,H}PS.
At the same time, besides folding respective vex_len_table[] entries,
adjust adjacent related legacy mod_table[] entries:
- when the prefix was already decoded, PREFIX_OPCODE is pointless,
- limit the amount of string literals by using X consistently on all
{,V}MOV{L,H}P{S,D} forms.
For major opcodes allowing only packed FP kinds of operands, i.e. the
ones where legacy and AVX decoding uses the X macro, we can do so for
AVX512 as well, by attaching to the checking logic the "EVEX.W must
match presence of embedded 66 prefix" rule. (Encodings not following
this general pattern simply may not gain the PREFIX_OPCODE attribute.)
Note that testing of the thus altered decoding has already been put in
place by "x86: correct decoding of packed-FP-only AVX encodings".
This can also be at least partly applied to scalar-FP-only insns (i.e.
V{,U}COMIS{S,D}) as well as the vector-FP forms of insns also allowing
scalar encodings (e.g. VADDP{S,D}).
Take the opportunity and also fix EVEX-encoded VMOVNTP{S,D} as well as
to-memory forms of VMOV{L,H}PS and both forms of VMOV{L,H}PD to wrongly
disassemble with only register operands.
Various AVX insns utilizing the X macro fail to reject F3/F2 embedded
prefix encodings. As the PREFIX_OPCODE attribute wasn't used by any
non-legacy-encoded insns so far, re-use it to achieve the intended
effect.
