#18156 closed defect (fixed)
nss-3.90 generates illegal instructions on older intel processors
| Reported by: | pierre | Owned by: | blfs-book |
|---|---|---|---|
| Priority: | normal | Milestone: | 12.0 |
| Component: | BOOK | Version: | git |
| Severity: | normal | Keywords: | |
| Cc: |
Description ¶
See https://bugzilla.mozilla.org/show_bug.cgi?id=1836925
Basically, nss-3.90 uses unconditionally an assembler instruction named adcxq, which is only present on Broadwell and later intel processors.
AFAICT, it is also only present on Ryzen processors for AMD.
There is a patch at https://phabricator.services.mozilla.com/D180068, which removes optimized curve25519 support (I suppose non optimized support is still there).
Change History (7)
comment:1 by , 23 months ago
comment:2 by , 23 months ago
I do have adx on my AMD Ryzen 9 3900X, but do not on my Intel i7-12700K. I'm not sure what that means.
comment:3 by , 23 months ago
It's for arbitrary precision arithmetic. According to https://en.wikichip.org/wiki/intel/core_i7/i7-12700k, it should be present in your CPU
follow-up: 5 comment:4 by , 23 months ago
Replying to pierre:
There is a patch at https://phabricator.services.mozilla.com/D180068, which removes optimized curve25519 support (I suppose non optimized support is still there).
Why don't they add a proper runtime CPU feature detection instead? They are already doing this for AVX anyway.
comment:5 by , 23 months ago
Replying to Xi Ruoyao:
Replying to pierre:
There is a patch at https://phabricator.services.mozilla.com/D180068, which removes optimized curve25519 support (I suppose non optimized support is still there).
Why don't they add a proper runtime CPU feature detection instead? They are already doing this for AVX anyway.
Well, not sure... I guess they made a quick fix now, and they will do a better one later.
Whether the instruction is accepted or not can be seen with
grep adx /proc/cpuinfo(not sure for AMD).