Linux Patched For A New AMD Zen 4 CPU Bug - Erratum #1485

[phoronix]

Phoronix: Linux Patched For A New AMD Zen 4 CPU Bug - Erratum #1485

Merged today to Linux 6.6 Git is a new patch for fixing an AMD erratum CPU bug affecting Zen 4 based processors like the Ryzen 7000 series and EPYC 8004/9004 series...

https://www.phoronix.com/news/Linux-Fixes-AMD-Erratum-1485

[JEBjames]

Michael

Typo/grammar

"Merged to the Linux kernel is a adding around a dozen lines of code for affecting a recently uncovered Zen 4 bug." maybe "Merged to the Linux kernel is around a dozen lines of code for fixing a recently uncovered Zen 4 bug.""

"erroneously appear in such configuration." maybe "erroneously appear in such a configuration.

Other missing commas see something like https://quillbot.com/grammar-check

[avis]

Michael

The most recent mitigations for Zen 4's SRSO are not free at all as your own benchmarking has shown: https://www.phoronix.com/review/amd-...ion-benchmarks

I'm running with spec_rstack_overflow=off because of that.

[fitzie]

i'm rocking mitigations=off and I'm using yesterday's version of curl. #livingdangerously

[Paradigm Shifter]

Nice work, rene!

[NotMine999]

"Living Dangerously" is like "Dropping your pants and sliding on the ice"?

[mlau]

Hm, I tried this on a 7950x3d overnight: boot 6.5.7 with mitigations=off, rebuild gcc/llvm/clang/rust/libreoffice and then a gentoo mips32 userland, all without any compiler failures.
Maybe Rene's setup is unique, or the bios already enables this workaround for me.

[rene]

Hm, I tried this on a 7950x3d overnight: boot 6.5.7 with mitigations=off, rebuild gcc/llvm/clang/rust/libreoffice and then a gentoo mips32 userland, all without any compiler failures.
Maybe Rene's setup is unique, or the bios already enables this workaround for me.

I wrote the specific instruction sequence must occur, as for example in my compiled gcc sparc64 cross compiler, and thus I had published a test initrd (and kernel) https://dl.t2sde.org/amd-zen4-smt-c0fefe/ But given it occurred in this combination for me it is likely many other real world applications can cause such spurlos #UD exceptions, too. All it takes is a similar instruction sequence to sooner or later be executed.

[mlau]

Oh let me read the LKML thread first, my apologies.

EDIT: yeah, the reproducer is awesome. setting bit 5 of this msr really gets rid of the faults. Super Arbeit Rene!

[mlau]

I wrote the specific instruction sequence must occur, as for example in my compiled gcc sparc64 cross compiler, and thus I had published a test initrd (and kernel) https://dl.t2sde.org/amd-zen4-smt-c0fefe/ But given it occurred in this combination for me it is likely many other real world applications can cause such spurlos #UD exceptions, too. All it takes is a similar instruction sequence to sooner or later be executed.

Did you actually notice an increase in speed with mitigations disabled? I remember based on other benchmarks by Michael that zen4 is actually faster with spectre-v2 mitigations enabled, as are latest intel cpus.