There is a compiler bug in latest MSVC, which at the time of writing is
19.36.32535: given `switch (x)`, where `x` is 64 bits wide, the compiler
generates code that computes an incorrect jump table index. E.g. if
`x` is zero, it ends up reading the table entry at index -1.
Disassembling single floating points with immediate values currently
gives wrong results on big endian hosts (like s390x), e.g.:
./cstool/cstool m68k40 'f2 3c 44 22 40 49 0e 56'
0 f2 3c 44 22 40 49 0e 56 fadd.s #0.000000, fp0
While it should be (like on x86):
./cstool/cstool m68k40 'f2 3c 44 22 40 49 0e 56'
0 f2 3c 44 22 40 49 0e 56 fadd.s #3.141500, fp0
The problem is that these single float values are supposed to be stored
in the 32-bit "simm" field of struct cs_m68k_op (see e.g. the printing
of M68K_FPU_SIZE_SINGLE in printAddressingMode() in M68KInstPrinter.c),
but currently the immediate is only written to the 64-bit "imm" field
of the union in cs_m68k_op. This works on little endian systems, since
the least significant bytes overlap in the union there. For example,
let's assume that the value 0x01020304 gets written to "imm":
04 03 02 01 00 00 00 00 uint64_t imm
xx xx xx xx xx xx xx xx double dimm;
xx xx xx xx .. .. .. .. float simm;
But on big endian hosts, the important bytes do not overlap, so "simm"
is always zero there:
00 00 00 00 01 02 03 04 uint64_t imm
xx xx xx xx xx xx xx xx double dimm;
xx xx xx xx .. .. .. .. float simm;
To fix the problem, let's always set "simm" explicitly, this works on
both, big endian and little endian hosts.
Thanks to Michal Schulz for his initial analysis of the problem
(in #1710) and to Travis Finkenauer for providing an easy example
to reproduce the issue (in #1931).
Closes: https://github.com/capstone-engine/capstone/issues/1710
* Replace one- and sign-extend with MathExtra.h functions.
This fixes incorrect sign extensions and reachable
aborts().
* Wrap around PC relative values instead of abort in case of overflow.
---------
Co-authored-by: billow <billow.fun@gmail.com>
* Update sysop inc file
* Fix missing braces warning
* Handle new system operands
* Fix build errors by renaming.
* Fix segfault
* Fix segfault
* Add custom MCOperand valiadtors
* Add AArch64 case for getFeatureBits
* Fix infinite loop
* Fix braces warning.
* Implement loopuo by name for sys operands
* Fix incorrect translation which remove else if statements.
* Fix several segfaults
* Rename GetRegFromClass patch
* Fix segfaults and asserts
* Fix segfault
* Move MRI setting to Mapping
* Remove unused code
* Add add_op_X functinos for AArch64.
* Add fill detail functins
* Handle RegWithShiftExtend operands
* Handle TypedVectorList operands.
* Handle ComplexRoatation operands
* Handle MemExtend operands
* Handle ImmRangeScale operands
* Handle ExactFPImm operands
* Handle GPRSeqPairsClass operands
* Handle Imm8OptLsl operands
* Handle ImmScale operands
* Handle LogicalImm operands
* Handle Matrix operands
* Handle SME Matrix tiles and vectors.
* Handle normal operands.
* Fix segfault.
* Handle PostInc operands.
* Reorder VecLayout enum to have no duplicate enum value.
* Handle PredicateAsCounter operands
* Handle ZPRasFPR operands
* Handle VectorIndex operands
* Handle UImm12Offset operands.
* Move reg suffix to enum val to single function.
* Handle SVERegOp operands
* Handle SVELogicalImm operands
* Handle SImm operand
* Handle PrefetchOp operands
* Handle Imm and ImmHex operands
* Handle GPR64as32 and GPR64x8 operands
* Add missing break
* Handle FPImm operand
* Handle ExtendedRegister opreand
* Handle CondCode operands
* Handle BTIHintOp operands
* Handle BarrierOption operands
* Handle BarrierXSOption
* Add not implemeted case again
* Handle ArithExtend operands
* Handle AdrpLabel and AlignedLabel operands
* Handle AMNoIndex operands
* Handle AddSubImm operands
* Handle MSRSystemRegisters and MRSSystemRegister operands
* Handle PSBHntOp and RPRFMOperand operands
* Remove unused variables
* Handle InverseCondCode operands
* Handle ImplicityTypedVectorList operands
* Handle ShiftedRegister operands
* Handle Shifter operands
* Handle SIMDType10Operand operands
* Handle SVCROp operands
* Handle SVEPattern operands
* Handle SVEVecLenSpecifier operands
* Handle SysCROperands
* Handle SysXzrPair operands
* Handle PState operands
* Handle VRegOperands
* Primt SME oeprands.
* Fix cs_operand.h include
* Rename arm64 -> aarch64 in python bindings.
* Add Python bindings for SH
* Fix ARM Python bindings (#2127)
* Restructure auto-sync update scripts.
* Move Helper functions to Updater dir
* Move requirements.txt
* Add basic ASUpdater.py
* Run black.
* Add inc file generater to updater
* Add option to select certain inc files fore generation.
* Enable clean build and implement patcher for inc files.
* Format config
* Patch main header files after inc generation.
* Implement clang-format function (unused yet, because it takes forever.)
* Copy generated inc files to arch dir
* Invert clean option (noramlly we need to clean the build dir.)
* Clearify arg doc
* Rename SystemRegister file for AArch64
* Centralize handling of path variables.
* Check if SystemOperands had to be generated before renaming on of its files.
* Replace class parameters by calling get_path
* Remove updater config which only contained paths.
* Add refactor option.
* Remove more path handling in the Configurator.
* Add translation step to updater.
* Fix includes after CppTranslator was moved into the Updater
* Remove updater config
* Fix several issue in the Configurator
* Fix file operations
* Remove addition argument from translator.
* Add Differ step to updater.
* Add path variable for arch_config
* Add diff step.
* Fix typo
* Introduce .clang-format path variable.
* Remove duplicate functions
* Add option to select update steps to execute.
* Check in write functions for write flag.
* Rename PatchMainHeader -> HeaderPatcher
* Move .gitignore
* Add README to vendor dir.
* Add all system operands to cstool output
* Update cstest with aarch64 changes
* Remove wb flag of aarch64 detail struct
* Set updates_flag after decoding
* Set writeback after decoding.
* Rename ARM64 -> AArch64
* Update printer and op mapping
* Exit normally
* Add AArch64 alias
* Fix some tmeplate function calls
* Fix flag check after rebase.
* Fix build by commentig unnused code.
* Add memory operand flag
* Handle memory operands printed via generic printOperand function.
* Handle UImm memory offsets
* Introduce MEM_REG and MEM_IMM op types
* Handle scaled memory immediates
* Check for op_count before checking for mem op at -1 index.
* Update memory operand flags.
* Pass imm/reg memory ops in set_imm/reg to set_mem.
* Add missing set_sme_operand call and fix assert.
* Remove CS_OP_MEM flag before entering switch.
* Preidcates are registers.
* Add shift info always to the previous operand
* Check for generic system regs
* Handle NumLanes = 0 LaneKind = q case
* Replace printImm call with normal print logic. Otherwise ops get added twice to detail.
* Handle FP operands in printOperand.
* Add access information to float operands.
* Rewrite SME matrix handling.
* Set correct SME layouts and allow for immediate range sme offsets.
* Handle cases of unknown system alias by setting their raw values
* Update cstool and header file with new SME offset handling
* Handle SME Tile lists.
* Fix build error in cstest
* Update MC tests for AArch64
* Handle TLBI operands and fix printing bug.
* Fix: Print signed value as signed.
* Add more system alias to detail.
* Remove duplicate hex prefix
* Set correct values for the register info
* Replace tabs with white spaces
* Move string append logic to own function.
* Set DecodeComplete = true before decoding (as originally in the LLVM code).
* Change type of feature argument, since only LLVM features are passed, not CS groups.
* Imitate lower_bound for the index table binary search.
* Remove trailing comments from test files.
* Print shift amount in decimal
* Save detail of shift alias instructions.
* Add extension details fot ext instruction alias
* Print LSB and width in decimal
* Fix LLVM bug. The feature check for V8_2a doesn't check if all features are enabled.
* Fix lower_bounds check.
For m == 0 we wrap around 0 of cause.
* Fix feature check. Add check for FeatureAll since it includes XS
* Operate on temporary MCInst when trying decoding.
* Add lower_bound behavior to IndexTypeStr binsearch.
* Fix MC tests which were incorrect because of missing FeatureAll check
* Add Alias handling for AArch64
* Update system operands with SYSIMM types and add additional sysop category.
* Add macros for meta programming (ARM64 <-> AArch64 selection).
* Fix union/struct confusion and add raw_value member to uninions.
* Allow to set Syntax and mode options for AArch64
* Fix build warning by using correct type
* Print shift value in decimal
* Add missing call to add_cs_detail.
* Update name map files with normalized names.
* Remove unused function
* Add check if detail should be filled.
* Fill detail for real instructions if only real detail is requested.
* Add always the extension.
* Make dir creation log message debug level
* Implement ADR immediate operand printer.
See: c3484b1fdc
* Check for flag registers beeing written and update flag.
* Move multiple CondCode helpers to aarch64.h because they are so freaking useful.
+ Print CC if it is EQ
* Fix incorrectly initialized CC and VectorLayout.
* Add LSL shift type for extensions.
* Fix case when shift amount is 0
* Fix post-index memory instructions.
* Pass raw immediate through getShiftValue to extract actual shift amount
* Setup AArch64 detail ops.
* Add flag for operands part of a list.
* Set vector indices for all relevant registers.
* Add missing call to add_cs_detail for postIncOperands
* Add ugly yet reliable way to determine post-index addressing mode
* Add support for old Capstone register alias.
* Remove leading space before some alias mnemonics.
* add AARCH64 to `cmake.sh`
* add HAS_AARCH64 to `cs.c`
* should probably just reference `cs_operand.h` in `aarch64.h`
* hint compiler at `AArch64_SYSREG` enum type for casting purposes
* update `Makefile` for AARCH64
leaves `CAPSTONE_HAS_ARM64` supported
* `testFeatureBits` platform function check
`testFeatureBits` should check if the platform function is visible first
* update tests to use AARCH64 convention
* hack: avoid enum casts for `MCInst` Values
Apple compiler really hates typecasting a enum, even if bounded from a unsigned. Lets set the raw_value directly
is a hack and needs proper review
* Check for present detail before accessing it.
* Add CS only groups
* Use general map ins_op type
* Fix build warning about str size computation.
* Disable warning about unitialized value for GCC 11.
Imm is initialized and the warning does not appear
in later versions.
* Use correct include guard for PPC
* Add missing requirements
* Update SystemOperand enums.
* Fix overlapping comparison warning
* Fix reachable assert where OpNum is not of type IMM
* Handle 0.0 operand for fcmp
* Fix incorrect variable passed.
* Fix for MacOS which doesn't know the warning and throws another one.
* Make getExtendEncoding static to fix build warning on MSVC.
* Fix build error: 'missing binary operator before token' by checking __GNUC__
* Add string search to add vector layout info.
* Add missing mem disponents of several ldr and str instructions.
* Add 0 immediates to several instructions.
* Rename v regs to q and d variant.
The cs_regname API can not pass the variant name of the register requested.
So we simply emit the default variant name.
* Fix incorrect enum value.
* Fix tests for system operands.
* Fix syntax issues in tests.
* Rename Arm64 -> AArch64 Python bindings.
* Fix Python bindings C structs.
* Fix generation of constants (ARMCC skipped because it starts with ARM)
* Update const files
* Remove -Wmaybe-uninitialized warning since it fails fuzz build
* Add missing comma
* Fix case
* Fix AArch64 Python bindings:
- Do not generate constants automatically (dscript is way too buggy).
- Update printing of details.
* Rename ARM64 -> AArch64 in test_corpus.py
* Rename test_arm64 -> test_aarch64
* Rename ARM-64 -> AArch64
* Fix diff CI test by disassembling AArch64 at former ARM64 place
* Fix several wrong types and remove unnecessary memebers from Python binding
* Fix: Same printing format of detail for cstool, test_ and test_*.py
* Fix: pass correct op index for mov alias with op[1] == reg wzr.
* Set prfm op manuall in case of unnown sysop. set_imm would add it to an memory operand wihtout base.
* Fix: If barrier ops are not set an assert is reached.
We fix it here by simply getting the immediate as the printing code does.
---------
Co-authored-by: Peace-Maker <peace-maker@wcfan.de>
Co-authored-by: Dayton <5340801+watbulb@users.noreply.github.com>
This has several consequences:
- Branch immediates are memory operands from now. Hence they are added manually as immediate.
- Some operands, handled over printOprerand(), are added to a mem operand, but the mem operand
is never closed. There is simply no indication when a memory operand ends.
So we close the mem operand now always, after an offset or disp awas added and the base exists.
* Fail CI if a command fails
* Apply cs_options in MC tests
* Emit a space if the alias has one at position I.
* Fix pop instruction tests.
* Update reglist patch and fix it again in ARMGenCSMappingInsnOps.inc
* Fix VSCCLRM* patch and write attributes.
* Emit new system operand formatting in tests
* Set new syntax in issues.cs
* Fix correct setting of neon lane
* Fix cstest arm SYSM operand type output
* Add implicit reads and writes of SP to POP and PUSH.
* Fix issue tests with new syntax and group names.
---------
Co-authored-by: Rot127 <unisono@quyllur.org>
- Differentiates between banked and non-banked registers.
- Saves the MSR mask.
- PSR field bits are saved separately.
- SYSm value is saved if possible.
