drm_gem_create_mmap_offset / intel_uxa_prepare_access bo problems

drm_gem_create_mmap_offset / intel_uxa_prepare_access bo problems

[Knut Petersen]

Hi Chris!

Problem:
=======
Slowdown of system, missing icons after 16 days kernel uptime and 12 days Xserver uptime.
Xorg log: flooded with "(WW) intel(0): intel_uxa_prepare_access: bo map (use gtt? 1, access 1) failed: No space left on device" lines
dmesg: flooded with "[drm:drm_gem_create_mmap_offset] *ERROR* failed to allocate offset for bo 0" entries


Hardware
=======
AOpen i915GMm-hfs, Pentium-M Dothan

Software:
=======
linux kernel 3.6.6
full Xorg git master, fetch/build 10 November 2012,
video-intel: git f5d79b202dd448e61ab6ffce26fe9cbf9051d770
xserver: git 011f8458805e443ac9130865d2840a929a00cabf
rest of system openSuSE 12.1

Any idea?

cu,
  Knut

Re: drm_gem_create_mmap_offset / intel_uxa_prepare_access bo problems

[Chris Wilson]

On Thu, 22 Nov 2012 16:29:08 +0100, Knut Petersen <Knut_Petersen@t-online.de> wrote:
> Hi Chris!
> 
> Problem:
> =======
> Slowdown of system, missing icons after 16 days kernel uptime and 12 days Xserver uptime.
> Xorg log: flooded with "(WW) intel(0): intel_uxa_prepare_access: bo map (use gtt? 1, access 1) failed: No space left on device" lines
> dmesg: flooded with "[drm:drm_gem_create_mmap_offset] *ERROR* failed to allocate offset for bo 0" entries
> 
> 
> Hardware
> =======
> AOpen i915GMm-hfs, Pentium-M Dothan
> 
> Software:
> =======
> linux kernel 3.6.6
> full Xorg git master, fetch/build 10 November 2012,
> video-intel: git f5d79b202dd448e61ab6ffce26fe9cbf9051d770
> xserver: git 011f8458805e443ac9130865d2840a929a00cabf
> rest of system openSuSE 12.1
> 
> Any idea?

Well you kernel and drm has all the latest protections, which is good
because it's usually a bo leak of some sort. First stop is to check
xrestop, /sys/kernel/debug/dri/0/i915_gem_objects and
intel-gpu-tools/scripts/who.sh

That will undoubtably reveal a large number of objects being held by the
X server...
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: drm_gem_create_mmap_offset / intel_uxa_prepare_access bo problems

[Knut Petersen]

On 22.11.2012 16:37, Chris Wilson wrote:

> Well you kernel and drm has all the latest protections, which is good because it's usually a bo leak of some sort. First stop is to check xrestop, /sys/kernel/debug/dri/0/i915_gem_objects and intel-gpu-tools/scripts/who.sh That will undoubtably reveal a large number of objects being held by the X 
> server... -Chris 

As the server had to be rebooted it will probably take some days to gather more information.

cu,
  Knut

Re: drm_gem_create_mmap_offset / intel_uxa_prepare_access bo problems

[Knut Petersen]

Hi everybody!

xrestop is functional again in xorg master, so I remembered to have a look at
the problem reported two weeks ago. It is caused by qinternet.

qinternet is a small utility  that is frequently used in opensuse systems to control
network connections. It attaches itself to the system tray of the kde panel, but it
also might be used with other desktop environments.

xrestop, proc/sys/fs/file-nr and /sys/kernel/debug/dri/0/i915_gem_objects
show that qinternet allocates one small pixmap every second. So it steadily
increases the number of inactive gem objects, the memory usage and the
number of used file descriptors. It is clear that this causes problems after
some time.

If someone at opensuse is reading this: This is an opensuse problem.
opensuse is a nice distribution, but as long as you do require people to
create a Novell acount with full address, phone number etc. to be able to
access the opensuse bugzilla I will not report bugs there.

cu,
  Knut



On 22.11.2012 16:37, Chris Wilson wrote:
> On Thu, 22 Nov 2012 16:29:08 +0100, Knut Petersen <Knut_Petersen@t-online.de> wrote:
>> Hi Chris!
>>
>> Problem:
>> =======
>> Slowdown of system, missing icons after 16 days kernel uptime and 12 days Xserver uptime.
>> Xorg log: flooded with "(WW) intel(0): intel_uxa_prepare_access: bo map (use gtt? 1, access 1) failed: No space left on device" lines
>> dmesg: flooded with "[drm:drm_gem_create_mmap_offset] *ERROR* failed to allocate offset for bo 0" entries
>>
>>
> Well you kernel and drm has all the latest protections, which is good
> because it's usually a bo leak of some sort. First stop is to check
> xrestop, /sys/kernel/debug/dri/0/i915_gem_objects and
> intel-gpu-tools/scripts/who.sh
>
> That will undoubtably reveal a large number of objects being held by the
> X server...
> -Chris
>