vulgraph opened issue #13220:
cranelift/codegen/src/isa/x64/lower.isle(sha538aa5ba) onrelease-25.0.0still has the pre-fixfcopysignlowering — the rules don'tXmm-coercea/bbefore theandnps/andpschain, so an upstreamf64.loadcan sink in and get widened to a 128-bit read.(rule (lower (has_type $F64 (fcopysign a @ (value_type $F64) b))) (let ((sign_bit Xmm (imm $F64 0x8000000000000000))) (x64_orpd (x64_andnpd sign_bit a) (x64_andpd sign_bit b))))Upstream
728fa071(CVE-2026-24116, 40.0.3) adds(a Xmm a)/(b Xmm b)bindings with the comment "force into reg so we don't sink a 128-bit load.". Neither binding nor the comment shows up onrelease-25.0.0.Is this branch still in the security-fix track? Happy to send a focused backport — it's a 4-line change plus the
tests/disas/f64-copysign.watregression test.vulgraph
tschneidereit closed issue #13220:
cranelift/codegen/src/isa/x64/lower.isle(sha538aa5ba) onrelease-25.0.0still has the pre-fixfcopysignlowering — the rules don'tXmm-coercea/bbefore theandnps/andpschain, so an upstreamf64.loadcan sink in and get widened to a 128-bit read.(rule (lower (has_type $F64 (fcopysign a @ (value_type $F64) b))) (let ((sign_bit Xmm (imm $F64 0x8000000000000000))) (x64_orpd (x64_andnpd sign_bit a) (x64_andpd sign_bit b))))Upstream
728fa071(CVE-2026-24116, 40.0.3) adds(a Xmm a)/(b Xmm b)bindings with the comment "force into reg so we don't sink a 128-bit load.". Neither binding nor the comment shows up onrelease-25.0.0.Is this branch still in the security-fix track? Happy to send a focused backport — it's a 4-line change plus the
tests/disas/f64-copysign.watregression test.vulgraph
tschneidereit commented on issue #13220:
See #13221
vulgraph commented on issue #13220:
Thanks for the cross-link to the policy explanation in the sister thread — got it, no further action needed here. Closing.
Last updated: May 03 2026 at 22:13 UTC