[yocto] [meta-intel] "Crazy" Xorg memory usage after upgrading from Daisy to Fido

Tue Nov 24 15:44:19 PST 2015

Hi Vincent,

I may have made some progress. The undesirable memory usage within Xorg isn’t there if I create an xorg.conf file containing:

Section “Device”
  Identifier “Intel Video”
  Driver “intel”
  Option “TearFree” “true"
EndSection

So it looks as if I need to enable “TearFree”. I didn’t need to add this for the 2.99.910 version of xf86-video-intel included with ‘daisy’.

Chris

> On 23 Nov 2015, at 23:48, Chris Tapp <opensource at keylevel.com> wrote:
> 
> Hi Vincent,
> 
> I’ve finally got back to being able to investigate this further.
> 
> I’ve now moved on to “jethro” and I’m seeing exactly the same behaviour. I’ve tried with kernel versions 3.14.39, 3.19.5 and 4.1.8.
> 
>> On 10 Jun 2015, at 03:50, Cheah, Vincent Beng Keat <vincent.beng.keat.cheah at intel.com> wrote:
>> 
>> Hi Chris,
>> 
>> I don’t have any idea with regard to the issue that you are getting below. All the work that we are doing here so far is on CHV (yocto-kernel-3.19.5 standard/base branch).
>> 
>> From your statement below, it looks to me that you are upgrading meta-intel from Daisy to Fido branch which are using yocto-kernel-3.14 (meta-intel/isg/valleyisland BSP). I'm not sure if you are able to reproduce this with yocto-kernel-3.19.5 (standard/base branch) from the meta-intel common directory. Also, comparing Daisy branch against Fido, it seems like there are lot of changes in the user-space stacks, which I'm not sure could cause the issue below.
>> 
>> 
>> Daisy 1.6.2
>> 	Kernel 3.4, 3.10, 3.14 (Supportable common base)
>> 	Xorg-server 1.15
>> 	Wayland/Weston 1.4.0
>> 	Xf86-video-intel 2.99.910
>> 	Libdrm 2.4.52
>> 	MESA 9.2.5
>> 	Cairo 1.12.16
>> 	libVA 1.3.1 (from meta-intel)
>> 	Intel-VA-driver 1.3.2 (from meta-intel)
>> 	GStreamer 1.2.3
>> 	GStreamer-VAAPI 0.5.8 (from meta-intel)
>> 
>> 
>> Dizzy 1.7.1
>> 	Kernel 3.10, 3.14, 3.17 (Supportable common base)
>> 	Xorg-server 1.15.1
>> 	Wayland/Weston 1.5.0
>> 	Xf86-video-intel 2.99.912
>> 	Libdrm 2.4.54
>> 	MESA 10.1.3
>> 	Cairo 1.12.16
>> 	libVA 1.3.1 (from meta-intel)
>> 	Intel-VA-driver 1.3.2 (from meta-intel)
>> 	GStreamer 1.4.1
>> 	GStreamer-VAAPI 0.5.8 (from meta-intel)
>> 
>> 
>> Fido 1.8
>> 	Kernel 3.14, 3.19 (supportable comon base)
>> 	Xorg-server 1.16.3
>> 	Wayland/weston 1.6.0
>> 	Xf86-video-intel 2.99.917
>> 	Libdrm 2.4.59
>> 	Mesa 10.4.4
>> 	Cairo 1.12.18
>> 	LibVA 1.5.0 (from meta-intel)
>> 	Intel-VA-driver 1.5.0 (from meta-intel)
>> 	Gstreamer 1.4.5
>> 	Gstreamer-vaapi 0.5.10 (from meta-intel)
>> 
>> 
>> ... Vincent
>> 
>> -----Original Message-----
>> From: Chang, Rebecca Swee Fun
>> Sent: Wednesday, June 10, 2015 9:08 AM
>> To: Cheah, Vincent Beng Keat
>> Cc: meta-intel at yoctoproject.org; Chris Tapp; Yocto Project; Wold, Saul; 'Paul Eggleton'
>> Subject: RE: [meta-intel] "Crazy" Xorg memory usage after upgrading from Daisy to Fido
>> 
>> Hi Vincent,
>> 
>> Can you help to comment on this issue mentioned by Chris?
>> Thanks.
>> 
>> Regards,
>> Rebecca
>> 
>>> -----Original Message-----
>>> From: Paul Eggleton [mailto:paul.eggleton at linux.intel.com]
>>> Sent: 09 June, 2015 12:15 AM
>>> To: Chang, Rebecca Swee Fun
>>> Cc: meta-intel at yoctoproject.org; Chris Tapp; Yocto Project; Wold, Saul
>>> Subject: Re: [meta-intel] "Crazy" Xorg memory usage after upgrading
>>> from Daisy to Fido
>>> 
>>> Rebecca, is this something you or one of your colleagues would be able
>>> to help with?
>>> 
>>> Thanks,
>>> Paul
>>> 
>>> On Friday 05 June 2015 08:29:00 Chris Tapp wrote:
>>>> I’ve got an application that I’ve had running nicely under Daisy for
>>>> some time. As Daisy is now a bit old, I decided to move the
>>>> application to
>>> Fido.
>>>> I’m using the meta-intel/isg/valleyisland BSP and also switched to
>>>> using its Fido branch.
>>>> 
>>>> The move only required a few minor changes and allowed me to drop a
>>>> Daisy “updates” layer that I had been using for things like gstreamer-1.0.
>>>> 
>>>> However, there is one behaviour which is killing me - I keep getting
>>>> oom-killer events!
>>>> 
>>>> The application is basically an OpenGL-ES 2.0 application that
>>>> renders various bits of text, images and streams captured from a
>>>> gstreamer pipeline at 60 Hz to a 1080 screen.
>>>> 
>>>> Under Daisy this generally took just under 50% CPU and used a modest
>>>> percentage of the 4 GB system memory - i.e. no where near running
>>>> out and usage was just about static.
>>>> 
>>>> Under Fido the CPU usage is about the same and the memory used by
>>>> the application itself looks reasonable when compared to Daisy (and
>>>> usage is static). However, the memory used by XOrg is far from
>>>> constant or stable - it basically has a VSZ value cycling from about
>>>> 630m to 2989m with the cycle period being in the order of 5 to 10
>>>> seconds. Peaks in XOrg memory usage coincide with stutters in video
>>>> playback within my app (audio is unaffected).
>>>> 
>>>> Monitoring /proc/meminfo when this is going on shows that “Shmem”
>>>> usage is following the same pattern as the memory used by XOrg (i.e.
>>>> Shmem usage is high at the same time). If the values are plotted on
>>>> a graph they appear to show that Shmem usage grows linearly and then
>>>> falls rapidly when nearly all the free memory has been exhausted,
>>>> perhaps in response to a delayed garbage collection run.
>>>> 
>>>> Does anyone have any ideas as to what I should be looking at to work
>>>> out what’s going on?
>>>> 
>>>> Are there any significant changes between XOrg under Daisy and Fido
>>>> that could be causing this?
>>>> 
>>>> Could this be related to the meta-intel video drivers?
>>>> 
>>>> Any feedback / comments would be really appreciated.
>>>> 
>>>> Thanks :-)
>>>> 
>>>> --
>>>> 
>>>> Chris Tapp
>>>> opensource at keylevel.com
>>>> www.keylevel.com
>>>> 
>>>> ----
>>>> You can tell you're getting older when your car insurance gets real cheap!
>>> 
>>> --
>>> 
>>> Paul Eggleton
>>> Intel Open Source Technology Centre
> 
> --
> 
> Chris Tapp
> opensource at keylevel.com
> www.keylevel.com
> 
> ----
> You can tell you're getting older when your car insurance gets real cheap!
> 
> --
> _______________________________________________
> yocto mailing list
> yocto at yoctoproject.org
> https://lists.yoctoproject.org/listinfo/yocto

--

Chris Tapp
opensource at keylevel.com
www.keylevel.com

----
You can tell you're getting older when your car insurance gets real cheap!

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 842 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.yoctoproject.org/pipermail/yocto/attachments/20151124/fdafef8d/attachment.pgp>