Android / Etnaviv / i.MX6 update

Lucas Stach l.stach at pengutronix.de
Wed Aug 1 15:55:53 UTC 2018


Am Mittwoch, den 01.08.2018, 17:53 +0200 schrieb Martin Fuzzey:
> On 01/08/18 17:26, Lucas Stach wrote:
> > Am Mittwoch, den 01.08.2018, 17:03 +0200 schrieb Martin Fuzzey:
> > 
> > The DRM card nodes can change their numbering for many reasons,
> > like DT
> > node ordering, device probe ordering, module vs. built-in configs,
> > etc.
> > So please don't base any assumptions on the device number.
> 
> There is still a hardcoded path in
> mesa3d/src/gallium/winsys/imx/drm/imx_drm_winsys.c
>        .gpu_fd = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC)

Yes, this needs fixing. This is one of the unclean corners of etnaviv
that I didn't get around to fix yet.

Regards,
Lucas

> So that definitely won't work if the GPU comes second.
> The GPU used to be the only render node so it was "ok" before but now
> that imx-drm has a render node too there is a problem.
> 
> The other paths are defined by android properties definied in a .rc
> file in /system/vendor/etc/init
> 
> 	setprop drm.gpu.vendor_name etnaviv
> 	setprop gralloc.gbm.device /dev/dri/renderD129
>   	setprop hwc.drm.device /dev/dri/card1
> 
> Those are little less problematic assuming we know the configuration
> the init file applies to.
> 
> It looks like we really need a more general version of Rob's patch to
> let us use names everywhere?
> 
> But it could be messy to implement because there are several
> different components involved.
> 
> Maybe we could add a couple of helper functions to libdrm to find the
> render / master node based on a vendor name?
> 
> 
> Regards,
> 
> Martin
> 
> 


More information about the etnaviv mailing list