Fedora 7 Test Update: kernel-2.6.22.2-52.fc7

[Michal Jaegermann]

On Mon, Aug 13, 2007 at 03:36:22PM -0400, Dave Jones wrote:
> On Mon, Aug 13, 2007 at 03:34:23PM -0400, Dave Jones wrote:
>  > On Mon, Aug 13, 2007 at 12:11:08PM -0600, Michal Jaegermann wrote:
>  >  > On Mon, Aug 13, 2007 at 10:01:02AM -0700, updates at fedoraproject.org wrote:
>  >  > > --------------------------------------------------------------------------------
>  >  > > Fedora Test Update Notification
>  >  > > FEDORA-2007-1549
>  >  > > 2007-08-13 09:58:54.176915
>  >  > > --------------------------------------------------------------------------------
>  >  > > 
>  >  > > Name        : kernel
>  >  > > Product     : Fedora 7
>  >  > > Version     : 2.6.22.2
>  >  > > Release     : 52.fc7
>  >  > ....
>  >  > > 
>  >  > > Detect broken timers on some AMD dual-core machines: fixes
>  >  > > hangs and failure to boot.
>  >  > 
>  >  > My x86_64 test machine is actually a single core.  This kernel,
>  >  > grabbed from koji, prints for me "agpgart: Detected AGP bridge 0"
>  >  > line and immediately reboots. Something like 'boot_delay=200' is
>  >  > required to see that or otherwise you are looking at rebooting
>  >  > machine right away.  On some occasions a machine may hang in a
>  >  > reboot and requires a powerdown.
>  > 
>  > Does booting with agp=off get past that?
> 
> Also initcall_debug would be useful to add, just to be sure that
> we're not dying in something immediately after AGP.

Yes, booting with 'agp=off' succeeds but running a graphics desktop
in this state is painful. :-)  See bz #249174.

With initcall_debug I see
....
calling initcall 0xffffffff814291d6: pci_iommu_init+0x0/0x17()
agpgart: Detected AGP bridge 0

and a machine locks up instead of rebooting.

I believe that this is another instance of
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=249174
Yesterday I posted there in comments 21 and 22 pictures of
backtraces from 2.6.22.1-32.fc6 on the same hardware.  That kernel,
for a change, gets "invalid opcode" and produces a trace instead of
just going away in some fashion.  Timings on PCI reads?

   Michal