Arch Linux

MikeGooB

Member

Registered: 2026-07-19

Posts: 2

BIOS settings when building packages [Non Issue]

Hello,

I had (no longer have) issues to install some packages from the AUR repository, the building process failed with SIGKILL and SIGSEGV, definitely memory issues.
The packages I tried were paru that uses rust and nomacs that uses C++.
paru could be built without any issue using cargo with the debug profile (without --release) but failed similarly with cargo --release.
Talking with DeepSeek, it hypothesized several issues and fixes, neither of them were useful, but it gave me the idea to check my BIOS for multithreading issues (as it was one of the hypothesis for the issue).

I have Intel(R) Core(TM) i9-14900K and my options could be intel specific, but I disabled HyperThreading and changed the profile from Extreme to Performance.
Apparently this fixed my issue, but I have no clue of why (my current theory is that hyperthreading or the extreme profile starved the threads, but I'm not sure).

Does anyone know why this seemingly fixed the issue? Maybe it was a coincidence (despite me having this issue from a couple of days, across multiple rebootings, not just a one time issue).

I have no problem sharing logs or specs if needed.

Last edited by MikeGooB (2026-07-19 16:12:27)

dimich

Member

From: Kharkiv, Ukraine

Registered: 2009-11-03

Posts: 718

Re: BIOS settings when building packages [Non Issue]

I disabled HyperThreading and changed the profile from Extreme to Performance.

Speculation: number of CPU cores changed, performance changed => race between parallel threads stopped reproducing.

MikeGooB

Member

Registered: 2026-07-19

Posts: 2

Re: BIOS settings when building packages [Non Issue]

I disabled HyperThreading and changed the profile from Extreme to Performance.

Speculation: number of CPU cores changed, performance changed => race between parallel threads stopped reproducing.

Talking from my ignorance on the topic, but isn't the issue of multithreading race condition an issue that should be handled by the developer of the code using sempahores and other techniques? Or am I misunderstandign something (this is uncharted territory for me)?

cryptearth

Member

Registered: 2024-02-03

Posts: 2,269

Re: BIOS settings when building packages [Non Issue]

the issue is how intel (and amd for that matter, as they do it differently) actually physical implement hyper-threading
i'm not an expert on that topic but suffice it to say: hyper-threading does not magically double what your cpu has physically - but it only changes how the hardware is used logically
as for building software: it highly depends on the language and the developers implementation if and how well the compilation can be parallelized which also depends on how smart the compiler is to detect that
if there's a difference between enabled and disabled hyper-threading this sure can be a hint towards hardware issues - but it can also be as simple as the software doesn't play nice with how the hardware expects and requires
SIGKILL should only happen when a deadlock is detected and a thread/process needs to be hard killed external
SIGSEGV usually happens when ppointer arithmetik goes wrong - often buffer-overflows or use-after-free or such ... these are the more classic race conditions because the hardware ends up in a state it should not / is not expected
disable hyper-threading might help to eliminate these race conditions resulting in successful build
as sanity check you could try reenable hyper-threading but limit the parallel build threads or even disable multithread build alltogether

unless you're able to reproduce the very same issue in a constant manner i doubt it's faulty hardware but rather just coincidental race conditions

dimich

Member

From: Kharkiv, Ukraine

Registered: 2009-11-03

Posts: 718

Re: BIOS settings when building packages [Non Issue]

isn't the issue of multithreading race condition an issue that should be handled by the developer of the code using sempahores and other techniques?

That's completely true. However, often race condition bugs are hard to reproduce, they may pop up only in specific circumstances.
Also some build systems may automatically evaluate number of parallel jobs to run by number of CPU cores. Afaik they don't distinguish hyperthreading virtual cores from real ones.

seth

Member

From: Won't reply 2 private help req

Registered: 2012-09-03

Posts: 76,958

Re: BIOS settings when building packages [Non Issue]

Speculation: number of CPU cores changed, number of parallel jobs changed => no longer running OOM?

But you'd have to replicate the failing setup to analyze what's going wrong.

---

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc