[Intel-gfx] On SNB: Hangcheck timer elapsed... GPU hung

[Andrew Lutomirski]

On Wed, Feb 16, 2011 at 8:20 AM, Andrew Lutomirski <luto at mit.edu> wrote:
> On Wed, Feb 16, 2011 at 8:13 AM, Ivan Bulatovic <combuster at gmx.com> wrote:
>> On Tue, 2011-02-15 at 11:46 +1000, Ted Phelps wrote:
>>> Hi Andy,
>>>
>>> Andi Kleen writes:
>>> > Ted Phelps <phelps at gnusto.com> writes:
>>> >
>>> > > Apologies if this is a known issue, but I haven't been able to convince
>>> > > myself that someone is looking after it.  I've been seeing this issue
>>> > > with Linux kernel 2.6.37, 2.6.38-rc4 and the most recent merge of Linus's
>>> > > git tree and drm-intel-fixes.  I'm happy to provide more information,
>>> > > apply patches, run tools, read code, as requested.
>>> >
>>> > Do you use displayport dual-head? I had this problem with dual head.
>>> > No such issue with only a single monitor.
>>>
>>> I'm using only one head attached to the DVI connector.
>
> I'm curious what userspace you're all running.  I'm using
> xf86-video-intel from git on Feb 4 with 2.6.37 (completely unpatched!)
> and my i7-2600 is quite stable.  The only problem I have is that
> compiz has hung twice since I got the machine.  In both cases, killall
> -9 compiz rescued the system.
>
> Fedora 14's xf86-video-intel didn't work so well (compiz dropped
> enough frames that the effects weren't really visible).
>
> I'm single head on DisplayPort.
>
> (Even firefox 4 webgl works nicely.)

I spoke too soon.  I ran 2.6.38-rc5 (actually 6f576d5 from Linus'
tree) for a few hours, and X hung.  I could still move the cursor, but
the cursor icon didn't change and everything was frozen.  I could
still switch VTs, though.

Xorg.0.log said:

[ 26219.458] (EE) intel(0): failed to set cursor: Input/output error
[ 26219.488] (EE) intel(0): failed to set cursor: Input/output error
[ 26219.518] (EE) intel(0): failed to set cursor: Input/output error
[ 26219.523] (WW) intel(0): intel_uxa_prepare_access: bo map failed:
Input/output error
[ 26219.529] (EE) intel(0): failed to set cursor: Input/output error
[ 26219.587] (WW) intel(0): intel_uxa_prepare_access: bo map failed:
Input/output error
[ 26219.633] (WW) intel(0): intel_uxa_prepare_access: bo map failed:
Input/output error
[ 26219.633] (WW) intel(0): intel_uxa_prepare_access: bo map failed:
Input/output error

and the kernel said:

[25918.930325] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer
elapsed... GPU hung
[25918.931803] [drm:i915_do_wait_request] *ERROR* i915_do_wait_request
returns -11 (awaiting 2652998 at 2652995, next 2653000)
[25919.218729] compiz[1949]: segfault at 0 ip 00007fd5957f5ea6 sp
00007fff9d0394d0 error 6 in i965_dri.so[7fd595770000+372000]
abrt[2932]: saved core dump of pid 1949 (/usr/bin/compiz) to
/var/spool/abrt/ccpp-1298262996-1949.new/coredump (77180928 bytes)

lspci says:
00:02.0 VGA compatible controller [0300]: Intel Corporation Sandy
Bridge Integrated Graphics Controller [8086:0102] (rev 09)
so I don't think the other patch in this thread will do anything.

2.6.37 has been stable and the userspace is identical.

gdb on the core file gives garbage (the faulting address looks legit
but does not correspond to any module).  Curiously, "info shared"
doesn't show libdrm or any of the intel_drv stuff.

--Andy