* Constify registerinfo.py output
Remove two conditionals separating identical bits of code.
Add "const" markup to MCRegisterDesc and MCRegisterClass.
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify instrinfo-arch.py output
In this case, do not actively strip const.
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the AArch64 backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the EVM backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify M680X backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify M68K backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the Mips backend
The Mips backend has not been regenerated from LLVM recently,
and there are more fixups required than I'd like. Just apply
the fixes to the tables by hand for now.
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the Sparc backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the TMS320C64x backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the X86 backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the XCore backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify systemregister.py output
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the ARM backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the PowerPC backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the MOS65XX backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the SystemZ backend
The mapping of system register to indexes is easy to
generate read-only. Since we know the indexes are
between 0 and 31, use uint8_t instead of unsigned.
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the WASM backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify cs.c
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Constify the BPF backend
Signed-off-by: Richard Henderson <rth@twiddle.net>
* Fixes#1317 - arm thb operand.shift.value
For a more detailed description, see issue #1317.
Release 4.0.0 introduced a new field for ARM operands:
`operand.mem.lshift`. This field was supposed to be a bug fix for #246.
The #246 issue has been fixed in the meantime and the proper shift value
was stored in `operand.shift.value`.
The 4.0.0 changes created a regression in which `operand.shift.value`
was not set for a `tbh [r0, r1, lsl #1]` instruction on ARM and the
value was set in a `operand.mem.lshift` field instead.
As the regression broke some of users codebase (e.g. in
[manticore](https://github.com/trailofbits/manticore/pull/1312) project), we fix it by setting
`operand.shift.value` back again.
As a result, the shift value is set in two fields: `operand.shift.value`
and `operand.mem.lshift`. As the `operand.shift` also stores a `.type`
field, we might want to deprecate `operand.mem.lshift` in the future.
* Add changelog stub
* Update init of cs_detail for AArch64
as @aquynh requested in #1125
* Update init of cs_detail for ARM
as @aquynh requested in #1125
* Update init of cs_detail for EVM
as @aquynh requested in #1125
* Update init of cs_detail for M680X
as @aquynh requested in #1125
* Update init of cs_detail for M68K
as @aquynh requested in #1125
* Update init of cs_detail for Mips
as @aquynh requested in #1125
* Update init of cs_detail for PowerPC
as @aquynh requested in #1125
* Update init of cs_detail for Sparc
as @aquynh requested in #1125
* Update init of cs_detail for SystemZ
as @aquynh requested in #1125
* Update init of cs_detail for TMS320C64x
as @aquynh requested in #1125
* Update init of cs_detail for XCore
as @aquynh requested in #1125
* Comment on init of cs_detail
* wrap long lines
* Declare global arch arrays with contents (#1171)
This eliminates the need for archs_enable() and eliminates the racey
initialization.
This makes the architecture-specific init and option functions
non-static so that they may be called from a different file.
Cherry-picked 853a2870
* Add cs_arch_disallowed_mode_mask global
Cherry-pick 94bce437:
mips: CS_MODE_MIPS32R6 implies CS_MODE_32
Cherry-pick 8998a3a1:
ppc: fix endian check (#1029)
Fixes bug where endianness could not be set for ppc.
Remove `big_endian` field of `cs_struct`.
Added a helper macro `MODE_IS_BIG_ENDIAN()` to check if
`CS_MODE_BIG_ENDIAN` is set.
Refactored `cs_open()` check for valid mode out of arch-specific code
into arch-independent code. Also added a valid mode check to
`cs_option()`. The checks use a new global array
`cs_arch_disallowed_mode_mask[]`.
* Make global arrays static
Make all_arch uint32_t to guarantee a certain number of bits (with
adequate room for growth).
This eliminates the need for archs_enable() and eliminates the racey
initialization.
This makes the architecture-specific init, option, and destroy functions
non-static so that they may be called from a different file.
* Fix undefined shifts
Found by oss-fuzz
uint8_t gets promoted to integer
and integers shift cannot overflow on sign bit
* Fix undefined shifts
shifting 31 bits the sign bit
* Fix undefined shifts
uint8 gets promoted to signed integer
in ARM, MIPS, Sparc
in AArch64, PPC and Xcore
* fix undefined shift in powerpc
* Fix undefined shift in Mips
use mulitply instead
