Using Etnaviv with mxsfb-drm

Using Etnaviv with mxsfb-drm

[Frieder Schrempf]

Hi Lucas, hi Marek,

while doing some tests on i.MX8MM with Etnaviv and mxsfb-drm (using the 
patches for DSIM, GPC, BLK-CTL, etc., see branch at [1]), I noticed that 
I don't seem to be able to run glmark2:

~# glmark2-es2-drm
=======================================================
     glmark2 2017.07
=======================================================
     OpenGL Information
     GL_VENDOR:     etnaviv
     GL_RENDERER:   Vivante GC600 rev 4653
     GL_VERSION:    OpenGL ES 2.0 Mesa 20.2.4
=======================================================
[build] use-vbo=false:Error: Failed to create FB: -22
Segmentation fault

Running a Qt application with the eglfs_kms backend produces a similar 
error, while running kmscube or applications without GPU rendering works 
fine.

What am I missing? When I used imx-drm, I didn't have any such issues. 
How is mxsfb-drm different?

Thanks in advance
Frieder

[1] https://github.com/fschrempf/linux/tree/v5.10-mx8mm-graphics
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: Using Etnaviv with mxsfb-drm

[Lucas Stach]

Hi Frieder,

Am Montag, den 14.12.2020, 13:33 +0100 schrieb Frieder Schrempf:
> Hi Lucas, hi Marek,
> 
> while doing some tests on i.MX8MM with Etnaviv and mxsfb-drm (using the 
> patches for DSIM, GPC, BLK-CTL, etc., see branch at [1]), I noticed that 
> I don't seem to be able to run glmark2:
> 
> ~# glmark2-es2-drm
> =======================================================
>      glmark2 2017.07
> =======================================================
>      OpenGL Information
>      GL_VENDOR:     etnaviv
>      GL_RENDERER:   Vivante GC600 rev 4653
>      GL_VERSION:    OpenGL ES 2.0 Mesa 20.2.4
> =======================================================
> [build] use-vbo=false:Error: Failed to create FB: -22
> Segmentation fault
> 
> Running a Qt application with the eglfs_kms backend produces a similar 
> error, while running kmscube or applications without GPU rendering works 
> fine.
> 
> What am I missing? When I used imx-drm, I didn't have any such issues. 
> How is mxsfb-drm different?

This could have multiple rot causes. First, please make sure that you
have sufficient CMA memory. Also the eLCDIF hardware has some
unfortunate restrictions on buffer stride, which surfaced a bug in the
etnaviv buffer layout code also on some very common display
resolutions. Please make sure you have [1] applied to get rid of this
bug.

Regards,
Lucas

[1] https://cgit.freedesktop.org/mesa/mesa/commit/?id=3862cec314de4a82821d9694ae2eec6f629ec782

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: Using Etnaviv with mxsfb-drm

[Frieder Schrempf]

Hi Lucas,

On 14.12.20 14:06, Lucas Stach wrote:
> Hi Frieder,
> 
> Am Montag, den 14.12.2020, 13:33 +0100 schrieb Frieder Schrempf:
>> Hi Lucas, hi Marek,
>>
>> while doing some tests on i.MX8MM with Etnaviv and mxsfb-drm (using the
>> patches for DSIM, GPC, BLK-CTL, etc., see branch at [1]), I noticed that
>> I don't seem to be able to run glmark2:
>>
>> ~# glmark2-es2-drm
>> =======================================================
>>       glmark2 2017.07
>> =======================================================
>>       OpenGL Information
>>       GL_VENDOR:     etnaviv
>>       GL_RENDERER:   Vivante GC600 rev 4653
>>       GL_VERSION:    OpenGL ES 2.0 Mesa 20.2.4
>> =======================================================
>> [build] use-vbo=false:Error: Failed to create FB: -22
>> Segmentation fault
>>
>> Running a Qt application with the eglfs_kms backend produces a similar
>> error, while running kmscube or applications without GPU rendering works
>> fine.
>>
>> What am I missing? When I used imx-drm, I didn't have any such issues.
>> How is mxsfb-drm different?
> 
> This could have multiple rot causes. First, please make sure that you
> have sufficient CMA memory. Also the eLCDIF hardware has some
> unfortunate restrictions on buffer stride, which surfaced a bug in the
> etnaviv buffer layout code also on some very common display
> resolutions. Please make sure you have [1] applied to get rid of this
> bug.

Great, thanks for pointing that out. Your patch indeed solves the issue 
for me (glmark2 doesn't crash anymore)!

Thanks a lot
Frieder
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel