weston rdp black screen

Fri Sep 18 08:15:51 UTC 2020

On 18.09.2020 09:53, Pekka Paalanen wrote:
> On Thu, 17 Sep 2020 19:10:53 +0200
> Maciej Pijanowski <maciej.pijanowski at 3mdeb.com> wrote:
>
>> On 10.09.2020 16:32, Maciej Pijanowski wrote:
>>> Hello,
>>>
>>> I'm having exactly the same issue as described in the [1].
>>> I have a custom Yocto build running on the embedded board
>>> with Mali GPU. Mali GPU was also used in the mentioned thread.
>>> So maybe there is something in common.
>>>
>>> When RDP backend is used as a main backend
>>> (in the core section) I can see the screen content in the RDP client.
>>> When the core backend is DRM and the RDP is started via the screen-share
>>> plugin, I can only see the black screen in the RDP client (although
>>> the input works
>>> correctly).
>>>
>>> I tried running the same combination on Fedora VM or on the RPi4
>>> and in those cases the screen-share works correctly with RDP.
>>>
>>> I've been looking for some differences in the logs between those setups.
>>> I have noticed the EGL config in case of Mali to be different than in case
>>> of the other ones.
>>>
>>> Mali:
>>>
>>> [07:24:23.919] Bad/unknown DRM format code 0x00000000.
>>> [07:24:23.919] Unable to use first choice EGL config with XRGB8888,
>>> succeeded with alternate ARGB8888.
>>> ...
>>> [07:24:23.923] GL version: OpenGL ES 2.0 "fa33ed7"
>>> [07:24:23.923] GLSL version: OpenGL ES GLSL ES 1.00
>>> [07:24:23.923] GL vendor: ARM
>>> [07:24:23.923] GL renderer: Mali-400 MP
>>> ...
>>> [07:24:24.006] Chosen EGL config details: id: 9 rgba: 8 8 8 8 buf: 32
>>> dep: 24 stcl: 8 int: 0-10 type: win|pix|pbf|swap_preserved vis_id:
>>> ARGB8888
>>> (0x34325241)
>>>
>>> RPi:
>>>
>>> [07:04:41.917] GL version: OpenGL ES 3.1 Mesa 20.0.2
>>> [07:04:41.917] GLSL version: OpenGL ES GLSL ES 3.10
>>> [07:04:41.917] GL vendor: Broadcom
>>> [07:04:41.917] GL renderer: V3D 4.2
>>> ...
>>> [07:04:42.032] Chosen EGL config details: id: 17 rgba: 8 8 8 0 buf: 24
>>> dep: 0 stcl: 0 int: 1-1 type: win vis_id: XRGB8888 (0x34325258)
>>>
>>>
>>> I've changed the format on the RPi ARGB8888 and it still works fine.
>>>   
>> So I've been investigating this issue further. I've checked for the
>> pixeldata
>> here [2]
>> and in the cases with the black screen there was nothing dumped.
>>
>> I have traced it further back to the glReadPixels call [3]. Once I switched
>> the pixel format to the PIXMAN_a8b8g8r8 (gl_format to GL_RGBA) it started
>> to work correctly. Both VNC [4] and RDP backends now work fine with the
>> screen-share on platform with Mali GPU and libMali blob.
>>
>> As I got this working, I could just patch the code to use the given
>> pixel format.
>> But I would prefer to fix things upstream. Any hints on what would be
>> "the proper"
>> way of fixing this issue are welcome.
> Hi,
>
> that is some good debugging you did. See this:
>
> $ git ngrep -p -A 3 read_pixels\(
> compositor/screen-share.c=817=shared_output_repainted(struct wl_listener *listener, void *data)
> --
> compositor/screen-share.c:882:		so->output->compositor->renderer->read_pixels(
> compositor/screen-share.c-883-			so->output, PIXMAN_a8r8g8b8, so->tmp_data,
> compositor/screen-share.c-884-			x, y_orig, width, height);
> compositor/screen-share.c-885-
> libweston/screenshooter.c=110=screenshooter_frame_notify(struct wl_listener *listener, void *data)
> --
> libweston/screenshooter.c:131:	compositor->renderer->read_pixels(output,
> libweston/screenshooter.c-132-			     compositor->read_format, pixels,
> libweston/screenshooter.c-133-			     0, 0, output->current_mode->width,
> libweston/screenshooter.c-134-			     output->current_mode->height);
> --
> libweston/screenshooter.c=262=weston_recorder_frame_notify(struct wl_listener *listener, void *data)
> --
> libweston/screenshooter.c:321:		compositor->renderer->read_pixels(output,
> libweston/screenshooter.c-322-				compositor->read_format, recorder->rect,
> libweston/screenshooter.c-323-				r[i].x1, y_orig, width, height);
> libweston/screenshooter.c-324-
> --
> tests/weston-test.c=431=test_screenshot_frame_notify(struct wl_listener *listener, void *data)
> --
> tests/weston-test.c:454:	compositor->renderer->read_pixels(output,
> tests/weston-test.c-455-					  compositor->read_format,
> tests/weston-test.c-456-					  pixels,
> tests/weston-test.c-457-					  0, 0,
>
> Notice how screen-share is using a hard-coded pixel format, but every
> other caller is using compositor->read_format.
>
> compositor->read_format is determined in gl_renderer_setup() based on
> the availability of GL_EXT_read_format_bgra. Also
> pixman_renderer_output_set_buffer() sets it if you happen to be using
> Pixman-renderer instead.
>
> So, screen-share needs to start using compositor->read_format instead
> of hard-coding.

We will try to fix that according to suggestion and submit a patch.

Thanks,

>
>
> Thanks,
> pq
>
>
>> I also added Stefan Agner to CC as he was changing this part of the code
>> lately in screen-share and we were discussing this matter in [5]. I hope
>> this is fine.
>>
>> Thanks,
>>
>> [2]
>> https://gitlab.freedesktop.org/wayland/weston/-/blob/master/compositor/screen-share.c#L888
>> [3]
>> https://github.com/3mdeb/weston/blob/use_GL_RGBA/libweston/renderer-gl/gl-renderer.c#L1551
>> [4] https://gitlab.freedesktop.org/wayland/weston/-/merge_requests/36
>> [5]
>> https://gitlab.freedesktop.org/wayland/weston/-/merge_requests/362#note_623870
>>
>>> One more observation - if I forced the DRM backend to use the pixman
>>> renderer
>>> (by adding the use-pixman=true in the weston.ini), the cloned content
>>> appeared
>>> correctly in the RDP client. Of course the performance and usability
>>> dropped,
>>> as the GPU was no longer used for rendering.
>>> It clearly suggests that the problem is somewhere around the Mali library.
>>>
>>> Any suggestions how to approach further debugging of these kind of issues
>>> are highly appreciated.
>>>
>>> Thank you
>>>
>>> [1]
>>> https://lists.freedesktop.org/archives/wayland-devel/2019-April/040382.html
>>>
>>> -- 
>>> Maciej Pijanowski
>>> Embedded Systems Engineer
>>> GPG: 9963C36AAC3B2B46
>>> https://3mdeb.com | @3mdeb_com
>>>
>>> _______________________________________________
>>> wayland-devel mailing list
>>> wayland-devel at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/wayland-devel
>
> _______________________________________________
> wayland-devel mailing list
> wayland-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/wayland-devel

-- 
Maciej Pijanowski
Embedded Systems Engineer
GPG: 9963C36AAC3B2B46
https://3mdeb.com | @3mdeb_com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/wayland-devel/attachments/20200918/ff6d1c94/attachment.htm>