omapdrm + etnaviv memory leak

[Lucas Stach]

Am Donnerstag, den 24.05.2018, 10:43 +0200 schrieb Lucas Stach:
> Hi Tomi,
> 
> Am Donnerstag, den 24.05.2018, 11:39 +0300 schrieb Tomi Valkeinen:
> > Hi Lucas,
> > 
> > On 23/05/18 11:53, Lucas Stach wrote:
> > 
> > > > With each run, I can see buffers being left lying around,
> > > > visible
> > > > in
> > > > both omapdrm's and etnaviv's 'gem' debugfs file. And they're
> > > > there
> > > > even after killing X.
> > > > 
> > > > If I try to rmmod etnaviv, I get the warnings below. Unloading
> > > > omapdrm is not possible, as it's being referenced by something
> > > > (presumably by etnaviv having imported omapdrm's dmabufs).
> > > > 
> > > > I haven't debugged this much yet, but we do use dmabuf import &
> > > > export successfully with omapdrm and v4l2. Has etnaviv dmabuf
> > > > import/export been tested?
> > > 
> > > Yes, dma-buf import/export with etnaviv is extensively being
> > > used,
> > > as
> > > we need to work with imx-drm on the scanout side and a V4L2
> > > driven
> > > VPU
> > > for video-decode.
> > 
> > I managed to create a simple test case for this, and I can see the
> > leak
> > happen without omapdrm too, with etna + vgem combination.
> > 
> > https://github.com/tomba/kmsxx/blob/etna-debug/utils/buftest.cpp
> > 
> > With etna_bo_map(), buffers leak.
> 
> Thanks, this is really helpful! I'll have a look.

Digging in it seems that etnaviv mmap is pretty broken for anything
which isn't a etnaviv shm buffer. I'll get back to you once I have a
patchset ready for test.

Regards,
Lucas