On Mon, Apr 22, 2013 at 02:56:08PM +0200, Thomas Gleixner wrote:
> Boris, can you please provide the irq16 line of /proc/interrupts
> before you invoke suspend?
>
> If it's shared we know which driver is shutdown before hda_intel and
> perhaps leaves its device in a weird state.
>
> If it's not shared, it still might be shared behind the scene. The
> other device uses MSI(X) and when shutdown MSI is disabled so the
> default IRQ route which points to IRQ16 would be active again.
>
> I assume, that the RCU stuff just causes enough timing difference to
> uncover the underlying issue.
Hmm, as Takashi reminded me, this box has two audio controllers. So
could it be that when we shutdown and disable MSI, the interrupts get
routed to to IRQ16 and they overflow there? Just a wild guess, I guess.
I very well can imagine that those two audio controllers to be one where
some f*cked up BIOS shim switches between the two but can't keep them
apart properly.
Something like the HPET fiascos...
* This is irq counts with my patch to expedite RCU grace periods:
- before suspend:
# grep snd /proc/interrupts
16: 1304 1340 1335 1329 1293 1348
1357 1363 IO-APIC-fasteoi snd_hda_intel
90: 3 3 4 4 4 4
3 3 PCI-MSI-edge snd_hda_intel
- after resume:
# grep snd /proc/interrupts
16: 1354 1389 1386 1380 1344 1400
1409 1495 IO-APIC-fasteoi snd_hda_intel
90: 0 0 0 0 0 0
0 5 PCI-MSI-edge snd_hda_intel
* "normal" RCU grace periods:
- before suspend:
# grep snd /proc/interrupts
16: 109 108 107 108 108 108
109 111 IO-APIC-fasteoi snd_hda_intel
90: 3 3 4 4 4 4
3 3 PCI-MSI-edge snd_hda_intel
- after resume:
# grep snd /proc/interrupts
16: 154 154 153 155 157 157
235 164 IO-APIC-fasteoi snd_hda_intel
90: 0 0 0 0 0 0
5 0 PCI-MSI-edge snd_hda_intel
More "hmm.."
--
Regards/Gruss,
Boris.
Sent from a fat crate under my desk. Formatting is fine.
--
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
