This commit refactors the AArch64 and partially the ARM
plugin to make it Capstone v6 compatible.
Due to the big API changes in Capstone v6 several changes
had to be made.
Because we need to be compatible to Capstone v4 and v5
many include guards are added as well.
Overview of changes done:
**ARM**
- Instruction alias were introduced. This leads to different
decoding and analysis paths taken for certain instructions.
Some alias have their IL code generated like the real
instruction now (no special handling needed anymore).
This change is responsible for many changes you'll encounter.
- The operand details of each instruction are now always the
one of the real instruction. Also for alias.
For example, if "MOV <Wd>, #<imm>" is an alias for `ORR <Wd>, WZR, #<imm>`,
the details by CS hold all three operands of "ORR".
Before, they held only the two of "MOV".
- Several bugs in variable and argument generation
were fixed. Especially the default variable width
for ARM Thumb was changed to 32bit instead of the 16bit.
**AArch64/ARM64**
The changes listed above for ARM, also apply to AArch64.
Additionally:
- Capstone v6 changed the name ARM64 now everywhere to AArch64.
To be compatible with Capstone v4/v5 AArch64 names must be wrapped
into macros which resolve the name, depending on the CS version used.
- Capstone v6 is now more consistent with
register real and alias names.
From now on we use the register alias by default.
**List squashed commit messages:**
[AArch64 CS v6 BEGIN] Change subproject config to use cs-auto-sync-aarch64 branch
Replace ARM64 with version sensitive macros.
Exclude alias if CS version >= 6
Update access to writeback member
Exclude instr alias from inclusion
Update memory operand printing to json.
Enable real instr. detail only for AArch64
Set correct arch name in meson.build for CS
Fix U/SBFM instructions and their alias.
Mark parameters with RZ_OUt/BORROW
Optimize register extension to skip some, if the width already matches.
Adapt width and lsb of U/SBFM alias instructions (ImmR and ImmS are from U/SBFM).
Fix tests correct semantic buy bad syntax
Pass alias MOV instructions to mov()
Handle CSET and CSETM alias
Fix lsl, lsr and asr by handling them as alias.
Fix mov alias.
Handle TST alias
Fix CNEG, CINV alias
Fix bfi and bfxil alias.
Fix sign extensions.
Fix compare instructions.
Fix NEG, NGC, NGCS, NEGS, MVN
Fix CINC
Fix multiply instructions.
Fix ROR
Run clang-format
Handle CMP for ESIL
Handle new position of memory disponents of post index operands.
Fix post-index operations.
Add missing writeback checks for Post and preindex
Handle UBFM and SBFM alias
Handl BFM alias
Handle CMP, CSET and CINC alias
Update meson file of for cs-aarch64 branch
Fix asm tests. Use reg alias now.
Fix condition confusion and incorrect operand usage.
Fix plf test.
Run clang-format
Use register alias in tests
Add support for fp and lr reg alias assembly.
Use reg alias in test
Rename cond tranlate functions r2 -> rz
Fix condition check which assume 0 == invalid.
Fix issues intruduced by rebase
Set CS commit to current next branch.
Rename ARM64 -> AArch64
Add missing source file to meson.build
Remove DisassemblerExtension.c file for CS v5
Update to newest capstone next branch
Bump up CS version
REVERT ME: Get Capstone v4/v5 via git clone until new tars are released.
Wrap setting of CS_DETAIL_REAL into CS version check
Add maybe-unitialized to Capstone C args.
Fix CS pre v6 build by adding guards.
Use reg alias now printed by default by CS.
Bump CS version to most recent next.
Fix build errors due to stircter alias handling in ARM.
Fix RzIL tests introduced by alias introduction to ARM.
Fix ESIL bugs introduced with ARM alias introduction.
- stackptr hasn't been set for POP and PUSH
Add support again for Thumb1 pop/push
Handle PUSHW and POPW alias
Update test case
Add more POP and PUSH alias and enrich detail for other versions of them.
Fix incorrect mem access width guesses for ARM thumb.
Set POP return info if it writes to PC
Fix tests about default var size and POP mem write direction.
Bump CS version to newest next.
Fix incorrect tests.
- TriCore: Functions were in ro section.
- Default arg width in ARM thumb is 32bit.
Revert check for a set stackptr.
stackptr is used in different ways:
1. Safes the offset from the stack frame base.
2. Is interpreted as somthing else for x86 and I cannot find out what, in a reasonably time.
Hence we cannot use it here consistently.
Remove check for non existing ARM_GRP_RET in CSv5
Fix incorrect stack offsets of variables.
'push <reg-list>' instructions for which the second register was the FP,
reset the stackptr variable to 0. This led to wrong bp offsets in the variable names.
In this case it was +0xc.
Bump CS version.
Add copy of meta-programming macros for capstone-sys build.
Update capstone-next.wrap
Use bracket-less met-programming macro to fix Windows build warnings.
Update wrap files for Capstone with branch names
Add new meta-programming macro
Add workaround for MSVC pre-processor bug.
* Remove macros checking for CS_API_VERSION 3
* Remove CS 3 from meson_options and subprojects.
* Remove Capstone v3 from CI
* Remove CS_API_VERSION >= 4 checks since always true from now on.
* Add CSv5 workflow to CI.
* Handle ARM pre v6 build errors.
* Fix PPC build errors for CS < v6
* Fix correct macro name
* Remove wrong include.
* Fix fallthrough warning
* Fix getter macros for WB and post-fix flags.
* Fix faulty macro name
* Check for all float extensions for vabs.
* Add return to assert cases.
* Update cs to 46154e8605aaefdcca5fecf4ea88b92db5a40ad3
* Remove cs3 wrap file.
---------
Co-authored-by: wargio <wargio@libero.it>
ARM
- Updated ARM identifiers and API changes.
- Fixed shift amounts to be always decimal.
- Fixed incorrect usages of memory disponents.
- Fixed restoring of condition codes.
- Updated IDs for new ones.
- Fixed 8-byte stores.
- Added flag checking for MOV with shifts.
Capstone Integration
- Removed capstone-auto-sync subproject and replaced with next.
- Checked for NEON features.
- Used https://github.com/capstone-engine/capstone/pull/2122 for better system operand support.
- Distinguished between 32 and 64-bit cc checks.
- Checked for CS API version >5.
- Checked for CS_NEXT_VERSION instead of CS_API_MAJOR.
- Fixed VSTn and VLDn instructions to use corrected memory operands.
- Excluded multiple instruction aliases that are no longer valid IDs.
- Used CS_NEXT_VERSION as include guard.
- Added CS v6 include guards.
- Added CS v6 support to branch conditions for Rzil.
- Added more CSv6 guards.
PPC
- Fixed: Compare instructions do not use the branch predicate.
- Fixed conditional and CTR checks.
- Fixed MTSPR and MFSPR instructions.
- Excluded more branch aliases no longer present in v6 from switch cases.
- Fixed incorrectly assigned variables.
- Fixed condition checks for branches.
- Used ITE for condition checks, to prevent unnecessary reads of CTR or CR.
- Fixed rzil tests with new semantics using register 0.
- Updated include guards for Capstone versions to use CS_NEXT_VERSION.
- Initialized spr_name to prevent uninitialized use.
- Used mem.offset register for CSv6.
- Always used real operand details.
- Used mem operand for DCBZ.
- Fixed tests that are semantically identical.
- Handled LI LIS alias.
- Fixed rzil tests (with simplified semantics).
- Fixed branch alias with new cond test method.
- Handled clrl. alias.
- Handled SL/SR alias.
- Fixed conditional branches in ESIL.
- Fixed possible multiplication result overflow.
- Added RZ_NONNULL.
- Added 0 register.
- Handled LIS alias.
- Fixed: Print crX reg name in CS v6.
- Handled all general branch instructions into a single case statement.
- Fixed no semantic issues in tests.
- Added newly discovered calls.
- Moved direction check to inline function.
- Added link of root cause for broken test.
- Added QPX support.
- Set Capstone next branch to latest commit.
- Fixed uninitialized warning.
- Set CS next branch to newest commit.
* core/cmd: Adjust math commands
* shell: make `?x` commands a parsing failure
We have moved the `?x` commands to `%x`, thus now no command should
start with `?`. This patch makes the parser fail to parse `?x`
strings.
* core/tui: use APIs in panels
* There's no `%q` anymore, use `%=`
Functions like MEM_read32 could still perform unaligned accesses on
OpenBSD/sparc64 with gcc, possibly related to gcc bug 51628. The
solution is to switch to a safe memcpy-based solution and let the
compiler do any optimizations.
Apple Clang 4.2 on Mac OS X 10.7 Lion does support avx2, but it has
issues with some of the asm syntax used, in particular jc and some mov
syntax.
So we add an additional check for a short assembly snippet.
The test for avx512 could result in false-positives because the
_mm256_setr_epi64x intrinsic requires only AVX, not AVX512. Specifically what
is needed here is AVX512VL.
In addition, not passing the -mavx* args to the asm compilation failed the
build on macOS High Sierra and below (Apple LLVM version 10.0.0,
clang-1000.11.45.5)
This commit rewrites completely how RzBinPlugin are used to load info
into the RzBinObject structure.
Main changes are:
- Avoid looping multiple times to set the same data:
now the data is cycled only once to rebase addresses, demangle strings
and to make each loaded structure similar to each other.
- rz_bin_demangle is now demangling only strings:
the old implementation was a hack over a hack to add classes & methods
into the RzBinObject while demangling a string.
This allowed methods and classes to be added to RzBinObject when for
example invoking `is` or other printing commands.
This was meant to be an "optimization" to avoid looping and doing
things on-the-fly.
- Now the demangling is done after guessing the language used in the bin.
This is done to avoid looping to various demanglers hoping that a
string is going to be correctly demangled.
After each string is demangled, the string is given to the language
related function which adds classes, methods and fields into the
RzBinObject fields.
- Swift handling is a special case, since the demangler can be disabled
via -Duse_swift_demangler.
The issue with the swift demangler is that is not correctly demangling
strings and outputting swift code, but instead is creating something
that sometimes works, sometimes doesn't.
Also the old code only added fields and never methods for "reasons",
and the new code just uses what the old code was doing as is.
- The new code also takes advantage of the language detection, which now
happens before the demangling of symbols, imports and relocs.
* Improve cmd_help.yaml and fix a small bug
* Add support for 64bit random numbers
* Make a distinction between 32- and 64-bit PRNG
* Add documentation for new functions
* Change the RBTree test to use 32-bit random numbers
