On Mon, 27 Jul 2026 at 03:04, Tomoaki AOKI <[email protected]> wrote:
>
> Another datapoint.
>
> When I've experienced "buzz" was when playing flac audio
> using audacious AND under quite heavy loads (maybe especially
> high I/O loads on staging / packaging) by poudriere builds
> and `make buildworld` in case base llvm is under rebuilding.
>
> Experienced both on pulseaudio and OSS backends, and today
> just experienced on pipewire backend, too.
>
> If this is caused by convergences on buses, I think
> glmark2 (or any other GPU-intensive benchmarks for the computer)
> could eat up bus bandwidth and causing the issue.
>
> Just my 2 JPY. ;-)
>

If you're experiencing "buzz" like I think you are then it's the same
frame being queued
over and over again to the DAC. Typically I've only seen that happen here when I
have the kernel panic (we're not stopping audio DMA at this point, and
I don't know
why it's looping over.)

If it's happening without a panic then it could be worthwhile digging
into whether
there's some driver bug around DMA and interrupt handling latency in high
load scenarios.

I'm unsurprised /i/ have not reproduced it as I don't do crazy stuff
like run poudriere
locally whilst also watching youtube.

If you're interested in reproducing and chasing down /this/ but then I
think this would
be a fine thing to reproduce it locally on a given machine, file a
bug, and then we
can dig into profiling what's going on. eg if we can reproduce it with something
VERY STUPID - eg mpg123 playing an mp3 direct to /dev/dsp* with no other virtual
audio stack running - we can then focus on the kernel side to see if there's
weird driver issues under high IO/interrupt load.

Even in the year of our lord "LLMs are available" there's only so much
one can do
in long email threads; at some point bugs need reproducing, PRs need filing and
developer / user time needs to happen. Let's be methodical about it to help out
the sound folks in figuring it out.



-adrian

Reply via email to