[DynInst_API:] [dyninst/dyninst] 37a827: Fix 64-bit shift count masking in ROSE x86-64 sema...

[bbiiggppiigg <noreply@xxxxxxxxxx>]

  Branch: refs/heads/bbiiggppiigg/fix-rose-shift-constant
  Home:   https://github.com/dyninst/dyninst
  Commit: 37a827c4e51cc5dd9c6530d7944e2acf0eb779f4
      https://github.com/dyninst/dyninst/commit/37a827c4e51cc5dd9c6530d7944e2acf0eb779f4
  Author: wuxx1279 <bbiiggppiigg@xxxxxxxxx>
  Date:   2026-05-31 (Sun, 31 May 2026)

  Changed paths:
    M dataflowAPI/rose/x86_64InstructionSemantics.h
    M parseAPI/src/BoundFactData.C

  Log Message:
  -----------
  Fix 64-bit shift count masking in ROSE x86-64 semantics

x86 masks shift/rotate counts modulo the operand size: mod 32 for
8/16/32-bit operands, but mod 64 for 64-bit operands. The ROSE
semantics extracted the count as Word(5) (extract<0,5>, i.e. mod 32)
for all sizes, so a 64-bit shift like `shrq $0x28` (count 40) was
truncated to 40 & 0x1f = 8.

In jump-table index analysis this produced a bogus bound: an index
computed as `(imm >> 40)` was evaluated as `imm >> 8`, scaling the
StridedInterval by 2^32 and yielding strides/ranges like
2^32[0,10*2^32]. Combined with the 32-bit loop counter in
ReadTable that became an infinite loop.

Fix the 64-bit (case 8) branches of shl/shr/sar/ror to mask the count
mod 64 using Word(6)/extract<0,6>. The <=32-bit branches keep their
correct mod-32 masking.

Also harden StridedInterval::ShiftLeft/ShiftRight: `1 << rhs.low` used
a 32-bit int literal, which is undefined for counts >= 31. Use
(int64_t)1 << rhs.low and bail to top for counts >= 64. Without this,
the now-correct large shift counts would overflow.

Co-Authored-By: Claude Opus 4.8 <noreply@xxxxxxxxxxxxx>



To unsubscribe from these emails, change your notification settings at https://github.com/dyninst/dyninst/settings/notifications