Re: 2.6.19-rc1: known regressions (v2)

[Prakash Punnoor]

Am Samstag 07 Oktober 2006 23:46 schrieb Adrian Bunk:
> Subject    : snd-hda-intel <-> forcedeth MSI problem
> References : http://lkml.org/lkml/2006/10/5/40
> Submitter  : Prakash Punnoor <prakash@xxxxxxxxxx>
> Status     : unknown

AFAIK, strictly speaking this is not a regression, but rather a new feature 
breaking: MSI capability of snd-hda-intel. Alsa dev  
Takashi Iwai doesn't think it is snd-hda-intel doing something wrong, but 
rather an IRQ routing probem. As such I am not sure who to cc....

While I don't really understand the details, I guess it is due to snd and net 
sharing the interrupt in "pure" APIC more:

 23:     232796          0   IO-APIC-fasteoi  HDA Intel, eth0

So both are physically on the same INT line? But MSI of snd want to assign 
another IRQ to itself:

 23:       7486          0   IO-APIC-fasteoi  eth0
316:          0          0   PCI-MSI-edge     HDA Intel

So it seems that irq23 still receives signals from snd and thus the nobody 
cared message gets printed.

Thus my work-around is simply to disable MSI of snd by 
snd-hda-intel.disable_msi=1 and all is well again.

So if this cannot be fixed for 2.6.19, I humbly suggest deactivating MSI of 
snd-intel-hda by default.

Cheers,
-- 
(°=                 =°)
//\ Prakash Punnoor /\\
V_/                 \_V

Attachment: pgp00000.pgp
Description: PGP signature