Bug 16850

Created attachment 17897 [details]
Xorg.0.log

Created attachment 17898 [details]
xorg.conf

It would probably be good to confirm from the kernel side first what exactly this message means/indicates. Can you do that Nikos?

(In reply to comment #3)
> It would probably be good to confirm from the kernel side first what exactly
> this message means/indicates. Can you do that Nikos?

Just posted: http://thread.gmane.org/gmane.linux.kernel/712176

I just went on and completely disabled DRM in the kernel and loaded the proprietary fglrx driver instead to see what happens.  The warnings still appear.  It seems it's entirely a kernel issue?

Note that this does not happen with 2.6.25 kernels.  I will try to build some 2.6.26 ones with various config options and see if a specific option is responsible for this.

I've made a bit of progress.  I changed the kernel config:

  - CONFIG_PCI_MSI=y
  + # CONFIG_PCI_MSI is not set

In other words I disabled Message Signaled Interrupt support.  Now instead of "do_IRQ: 0.83 No irq handler for vector" I get:

  +------ PCI-Express Device Error ------+
  Error Severity          : Uncorrected (Non-Fatal)
  PCIE Bus Error type     : Transaction Layer
  Flow Control Protocol   : First
  Receiver ID             : 0010
  VendorID=1106h, DeviceID=a208h, Bus=00h, Device=02h, Function=00h
  Broadcast error_detected message
  Broadcast mmio_enabled message
  Broadcast resume message
  AER driver successfully recovered

(Repeated as many times as the "do_IRQ" message previously.)

I hope this helps :P

Just in case someone needs it, here's the complete dmesg: http://realnc.pastebin.com/d6534029

And here my kernel configuration: http://realnc.pastebin.com/dafc97a2

(I am also attaching those two in case the links disappear.)

Created attachment 17911 [details]
dmesg

Created attachment 17912 [details]
.config

(In reply to comment #5)
> I just went on and completely disabled DRM in the kernel and loaded the
> proprietary fglrx driver instead to see what happens.  The warnings still
> appear.  

And sync to vblank doesn't work correctly either?

> It seems it's entirely a kernel issue?

Quite possibly, so I still think it's a good idea to continue tracking it down there first.

> Note that this does not happen with 2.6.25 kernels.  I will try to build some
> 2.6.26 ones with various config options and see if a specific option is
> responsible for this.

What if you disable CONFIG_PCIEAER? In the worst case, you could try git bisect between 2.6.25 and 2.6.26 to find the change(s) that introduced the problem.

(In reply to comment #9)
> (In reply to comment #5)
> > I just went on and completely disabled DRM in the kernel and loaded the
> > proprietary fglrx driver instead to see what happens.  The warnings still
> > appear.  
> 
> And sync to vblank doesn't work correctly either?

I works with fglrx.  But not through DRI configuration (vblank_mode).  fglrx doesn't allow for that.  Rather through Catalyst Control Center.


> > Note that this does not happen with 2.6.25 kernels.  I will try to build some
> > 2.6.26 ones with various config options and see if a specific option is
> > responsible for this.
> 
> What if you disable CONFIG_PCIEAER? In the worst case, you could try git bisect
> between 2.6.25 and 2.6.26 to find the change(s) that introduced the problem.

I'll try CONFIG_PCIEAER when I get back to my machine.  As far as I understand it, the problem is that the hardware is doing something the kernel doesn't expect it to do.

This bug has been open long enough.  In the end, I was never able to track the cause down, but today I updated to kernel 2.6.26.2, git master of X.Org and git master of xf86-video-ati and it's fixed. So I'm closing this.