[mythtv-users] Repeating audio on SMP machine
Yan-Fa Li
yanfali at best.com
Tue Dec 7 09:24:56 UTC 2004
Take a look at Robert Love's web page. Schedutils might have something
you can try:
http://www.tech9.net/rml/schedutils/
Yan
Kane Tse wrote:
>>
>> Try forcing the capture card and audio capture IRQ processing onto one
>> CPU. It could be that the process is being migrated between CPUs and
>> the driver is not completely SMP safe.
>>
>
>
> I read up on the smp_affinity file and made the modifications, but
> ultimately, it didn't help. I made the following observations:
>
> 1) When I did a 'cat /proc/irq/17/smp_affinity', I got 00000001
> suggesting the IRQ was redirected to CPU0. But when I did:
>
> cd /proc/irq/17
> cat smp_affinity
>
> I got 00000002. Which is really strange, since they should be the same
> file right? Also, this also suggested that input is already handled by
> one CPU, namely CPU0.
>
> 2) When I did an lspci -v | grep 17:, I noticed CPU0 was handling
> almost all the processing for that card, so did an echo 2 >
> /proc/irq/17/smp_affinity to redirect IRQ processing onto CPU1.
> Performing a 'cat /proc/irq/17/smp_affinity' showed 00000002, which
> suggested that IRQ processing was being handled by CPU1. This was
> confirmed by lspci -v | grep 17:. But after a few minutes, 'cat
> /proc/irq/17/smp_affinity' reverted back to 00000001... for no reason,
> and the problems didn't go away (although i couldn't tell if this was
> before or after the reversion back to 00000001.
>
> What's up with that? Why did it revert back?
>
>
>
>
> Anyway, it seems to me that CPU0 was already directed to process input
> from IRQ17, so maybe that wasn't my problem. But tonight when I get
> home I am going to clean-install Fedora Core 3 on that machine and see
> if things improve or not.
>
> Thanks for your help though. And let me know if you have any other ideas.
>
> Is anyone else out there running a bt8x8 and btaudio on an SMP setup?
> Which OS are you using? Any problems?
>
>
More information about the mythtv-users
mailing list