Bug#628080: xserver-xorg-video-radeon: Low performance after a period of inactivity

[Stephen Kitt <steve@sk2.org>]

Hi,

On Wed, Sep 14, 2011 at 07:14:12PM +0200, dayer wrote:
> El mié, 14-09-2011 a las 11:23 +0200, Michel Dänzer escribió:
> > Since several IRQs are affected, I suspect this is a fundamental problem
> > in the kernel/BIOS rather than in any driver.
> 
> I think the problem is my graphic card ATI Gigabyte HD4650 AGP 1GB. I'm
> using it one or two years without problem until April 2011, when IRQs
> problems started. In Windows XP there's no problem. And if I try
> connecting an old ATI 9600 AGP 256MB in Linux there's no problem with
> IRQs.

I've been having a similar problem for a while now, with a similar
graphics card (an XFX HD4650 AGP with 1GB of RAM): at every boot,
regardless of irqpoll, the card's IRQ would end up being disabled. I
hadn't bothered filing a bug because like Michel, I thought it was
probably due to a BIOS bug.

Today though I noticed a pattern: in every single instance, just after
the IRQ got disabled, the kernel logged
[   55.404002] Disabling IRQ #16
[   56.912023] hda-intel: azx_get_response timeout, switching to polling mode: last cmd=0x000f0001

So I wondered whether this might be somehow related to the
snd-hda-intel driver attempting to initialize the HDMI audio
output... I blacklisted the snd-hda-intel driver, and hey presto, no
more errors.

In the original submitter's case, both IRQ #16 and IRQ #17 got
disabled; but on my system IRQ #16 is the VGA controller, and #17 the
audio device. I noted too that on the first report mentioning IRQ #17,
the handlers include azx_interrupt which belongs to snd-hda-intel...

So it might be worth trying to add
	blacklist snd-hda-intel
to /etc/modprobe.d/local.conf for example, and running
	update-initramfs -u
to make sure it's not loaded from the initramfs.

Regards,

Stephen