[yocto] [meta-raspberrypi][PATCH 5/5] rpi-default-providers: Switch providers according to used gfx stack

Thu Aug 13 08:43:43 PDT 2015

Hello,

On 08/13/2015 09:22 AM, Javier Martinez Canillas wrote:
> Hello Andreas,
> 
> On 08/12/2015 10:22 PM, Andreas Müller wrote:
>> On Wed, Aug 12, 2015 at 7:15 PM, Andreas Müller
>> <schnitzeltony at googlemail.com> wrote:
>>> FYI: I managed to get the vc4 driver loaded (should be in my repo
>>> branch vc4-2). With this I get some repeating kernel error messages
>>> (don't have them here). I am sure that I read something about these
>>> messages when preparing vc4 (yes I started similar before you sent
>>> patches).
>>>
> 
> Awesome, I tried to get it working yesterday but couldn't. Good work!
> 
>>> Hope I have some energy left tonight to check further and let you know...
>>>
>> >From xorg perspective all looks fine
>>
>> [595923.730] (II) modeset(0): [DRI2] Setup complete
>> [595923.730] (II) modeset(0): [DRI2]   DRI driver: vc4
>> [595923.730] (II) modeset(0): [DRI2]   VDPAU driver: vc4
>> [595923.740] (--) RandR disabled
>> [595923.745] (II) AIGLX: enabled GLX_MESA_copy_sub_buffer
>> [595923.745] (II) AIGLX: enabled GLX_ARB_create_context
>> [595923.745] (II) AIGLX: enabled GLX_ARB_create_context_profile
>> [595923.745] (II) AIGLX: enabled GLX_EXT_create_context_es2_profile
>> [595923.745] (II) AIGLX: enabled GLX_INTEL_swap_event
>> [595923.745] (II) AIGLX: enabled GLX_SGI_swap_control and GLX_MESA_swap_control
>> [595923.745] (II) AIGLX: enabled GLX_EXT_framebuffer_sRGB
>> [595923.745] (II) AIGLX: enabled GLX_ARB_fbconfig_float
>> [595923.745] (II) AIGLX: GLX_EXT_texture_from_pixmap backed by buffer objects
>> [595923.747] (II) AIGLX: Loaded and initialized vc4
>> [595923.747] (II) GLX: Initialized DRI2 GL provider for screen 0
>> [595923.782] (II) modeset(0): Setting screen physical size to 338 x 270
>>
>> but kernel complains periodically ~6s with
>>
>> [   36.814922] [drm:vc4_submit_cl_ioctl] *ERROR* Rendering requires
>> BOs to validate
>> [   43.060516] [drm:vc4_submit_cl_ioctl] *ERROR* Rendering requires
>> BOs to validate
>> [   49.325115] [drm:vc4_submit_cl_ioctl] *ERROR* Rendering requires
>> BOs to validate
>> [   55.558433] [drm:vc4_submit_cl_ioctl] *ERROR* Rendering requires
>> BOs to validate
>>
> 
> Yes, I was able to reproduce the issue. My X -verbose output:
> 
> http://hastebin.com/onovosojuw.md
> 
>> Will check what this message want me to say - anybody out there with
>> helping hints?
>>
> 
> No clue. I was looking and the error is in the VC4_SUBMIT_CL ioctl cmd
> handler (vc4_submit_cl_ioctl) in drivers/gpu/drm/vc4/vc4_gem.c.
> 
> AFAIU bo_handle_count is supposed to always be > 0 but somehow mesa is
> passing 0 on it.
> 
> The ioctl call is in vc4_flush (src/gallium/drivers/vc4/vc4_context.c)
> in mesa. So it seems this is a mesa issue. I've asked Eric Anholt in
> #dri-devel on IRC if his kernel is supposed to work with mesa 10.5.8
> or if there is a minimum version / sha1 that is needed.
>

So I spent a lot of time trying to figure out what is wrong with mesa
10.5.8 and then gave up and tried updating mesa to the version we are
using for our Tizen port that works. I did the change in the meta layer
just for testing purposes. This is the diff [0] on top of the kernel
patches and Andreas' latest changes.

I did this because it was a known mesa version that works and indeed
Xorg starts and the "Rendering requires BOs to validate" error is gone.
The complete Xorg log is at [1].

Also, glmark2 says:

=======================================================
    glmark2 2014.03
=======================================================
    OpenGL Information
    GL_VENDOR:     Broadcom
    GL_RENDERER:   Gallium 0.4 on VC4
    GL_VERSION:    2.1 Mesa 10.7.0-devel (git-1762568fd39b)
=======================================================

And I see the horse and the box spinning. It gives me a 33-35 FPS but
the last texture-filter=linear test fails with:

[drm:vc4_use_bo] *ERROR* BO index -980025344 greater than BO count 9

So there are still some issues. The glkmar2 log is at [2].

I'll post the v2 of the vc4 kernel recipe patches since the problem
seems to be with mesa 10.5.8 and also is working for us with Weston.

At least so we all are on the same page and testing with the same
set of patches.

Best regards,
-- 
Javier Martinez Canillas
Open Source Group
Samsung Research America

[0]: http://hastebin.com/karakapeqa.diff
[1]: http://hastebin.com/foyanotema.vhdl
[2]: http://hastebin.com/ikotadavop.vbs