[Mesa-dev] [PATCH v2 07/13] st/va: add headless support, i.e. VA_DISPLAY_DRM
Emil Velikov
emil.l.velikov at gmail.com
Thu Oct 29 15:21:00 PDT 2015
On 29 October 2015 at 12:05, Christian König <deathsimple at vodafone.de> wrote:
>
>> + drm_info = (struct drm_state *) ctx->drm_state;
>> + if (!drm_info) {
>> + FREE(drv);
>> + return VA_STATUS_ERROR_INVALID_PARAMETER;
>> + }
>> +
>> +#if GALLIUM_STATIC_TARGETS
>> + drm_fd = drm_info->fd;
>> +#else
>> + drm_fd = dup(drm_info->fd);
>> +#endif
>> +
>> + if (drm_fd < 0) {
>> + FREE(drv);
>> + return VA_STATUS_ERROR_INVALID_PARAMETER;
>> + }
>> +
>> + drv->vscreen = CALLOC_STRUCT(vl_screen);
>> + if (!drv->vscreen)
>> + goto error_screen;
>> +
>> +#if GALLIUM_STATIC_TARGETS
>> + drv->vscreen->pscreen = dd_create_screen(drm_fd);
>> +#else
>> + if (pipe_loader_drm_probe_fd(&drv->dev, drm_fd))
>> + drv->vscreen->pscreen = pipe_loader_create_screen(drv->dev,
>> PIPE_SEARCH_DIR);
>> +#endif
>> +
>> + if (!drv->vscreen->pscreen)
>> + goto error_pipe;
>> +
>> + }
>
>
> It would be nice to have this snip-set in something like a vl_winsys_drm.c
> if possible, but that's not a hard requirement and can happen when we
> actually need this elsewhere as well.
>
I doubt we'll need it elsewhere any time soon, so whenever one wants
to hack on something different it'll be fine :-P Although we should
consider splitting out the non platform specifics out of
vl_winsys_dri.c and perhaps fetching the xlib/xcb symbols via
dlsym/dlopen.
> Emil any further comments?
With the VA_RT_FORMAT_YUV400 handling dropped, and the
VA_RT_FORMAT_RGB32 + vlVaDestroySurfaces comments in "st/va:
implement VaCreateSurfaces2 and VaQuerySurfaceAttributes" the whole 9
patches look great. Fwiw
Reviewed-by: Emil Velikov <emil.l.velikov at gmail.com>
Thank you for sticking around and addressing with all the comments Julien !
Christian, you are definitely more authoritative person on the topic,
so I'd leave the final decision up-to you.
> Did you static target cleanup already land? I
> would like to get at least quite a bunch of the patches upstream sooner than
> later.
>
Most people value their sanity and haven't looked at the static target
patches :( That said, none of that work is not a blocker for landing
these.
Cheers,
Emil
More information about the mesa-dev
mailing list