On (20/01/28 16:22), Tomasz Figa wrote:
[..]
> > > > Then, if q->allow_cache_hint is set, then default to a cache sync (cache clean)
> > > > in the prepare for OUTPUT buffers and a cache sync (cache invalidate) in the
> > > > finish for CAPTURE buffers.
> > >
> > > We alter default cache sync behaviour based both on queue ->memory
> > > type and queue ->dma_dir. Shall both of those cases depend on
> > > ->allow_cache_hints, or q->memory can be independent?
> > >
> > > static void set_buffer_cache_hints(struct vb2_queue *q,
> > > struct vb2_buffer *vb,
> > > struct v4l2_buffer *b)
> > > {
> > > if (!q->allow_cache_hints)
> > > return;
> > >
> > > /*
> > > * DMA exporter should take care of cache syncs, so we can avoid
> > > * explicit ->prepare()/->finish() syncs. For other ->memory types
> > > * we always need ->prepare() or/and ->finish() cache sync.
> > > */
> > > if (q->memory == VB2_MEMORY_DMABUF) {
> > > vb->need_cache_sync_on_finish = 0;
> > > vb->need_cache_sync_on_prepare = 0;
> > > return;
> > > }
> >
> > I personally would prefer this to be above ->allow_cache_hints check,
> > IOW to be independent. Because we need to skip flush/invalidation for
> > DMABUF anyway, that's what allocators do in ->prepare() and ->finish()
> > currently.
>
> I think it's even more than personal preference. We shouldn't even
> attempt to sync imported DMA-bufs, so the code above may result in
> incorrect behavior.
Sure. Allocators test buf->db_attach in prepare()/finish(). This patchset
removes that logic, because we move it to the core code. But if the
conclusion will be to clear ->need_cache_sync_on_finish/prepare only
when ->allow_cache_hints was set, then buf->db_attach bail out must stay
in allocators. And this is where I have preferences :)
-ss
