On Thu, Jan 18, 2018 at 4:30 PM, Kenneth Graunke <kenn...@whitecape.org>
wrote:
> On Friday, January 12, 2018 1:28:00 AM PST Chris Wilson wrote:
> > Quoting Jason Ekstrand (2018-01-12 01:40:52)
> > > This helper should be used carefully as setting tiling is a racy
> > > operation since it potentially interacts with other processes. Still,
> > > it is a useful thing to be able to do.
> > >
> > > Cc: mesa-sta...@lists.freedesktop.org
> > > ---
> > > src/mesa/drivers/dri/i965/brw_bufmgr.c | 27
> +++++++++++++++++++++++++++
> > > src/mesa/drivers/dri/i965/brw_bufmgr.h | 10 ++++++++++
> > > 2 files changed, 37 insertions(+)
> > >
> > > diff --git a/src/mesa/drivers/dri/i965/brw_bufmgr.c
> b/src/mesa/drivers/dri/i965/brw_bufmgr.c
> > > index 469895e..dbd13dd 100644
> > > --- a/src/mesa/drivers/dri/i965/brw_bufmgr.c
> > > +++ b/src/mesa/drivers/dri/i965/brw_bufmgr.c
> > > @@ -1133,6 +1133,33 @@ brw_bo_get_tiling(struct brw_bo *bo, uint32_t
> *tiling_mode,
> > > return 0;
> > > }
> > >
> > > +int
> > > +brw_bo_set_tiling(struct brw_bo *bo, uint32_t tiling_mode,
> > > + uint32_t stride)
> > > +{
> > > + struct brw_bufmgr *bufmgr = bo->bufmgr;
> > > +
> > > + mtx_lock(&bufmgr->lock);
> >
> > This mutex protects the buffer cache, which should not be containing
> > this bo. Changing the tiling of a shared active bo also should not
> > happen because the other parties will have already derived state from
> > the older tiling. So I don't see the purpose of this mutex here.
> >
> > If we will need exclusive access to a bo, let's have a bo->lock.
> > -Chris
>
> I agree with Chris, I don't like this locking. Looking more closely,
> I don't think patches 3-4 are what you want at all.
>
> intel_create_image_from_fds_common calls brw_bo_gem_create_from_prime.
> It's the only caller of that function.
>
> brw_bo_gem_create_from_prime allocates a new BO, does GET_TILING, and
> inserts it into the cache. This is pointless after your patch 4,
> because you immediately whack it.
>
> So...instead...just make brw_bo_gem_create_from_prime take a tiling,
> and SET_TILING it...don't bother with GET_TILING at all. Then, your
> BO isn't in the cache yet (or you have the cache locked), and it's
> safe without a race and redundant work.
>
I can rework things to do it that way. I thought about it briefly but I
got hung up on how to handle the "use the tiling from the BO" case. I
think I can use an int and pass -1 or something. Or maybe add a
create_from_prime_tiled helper.
--Jason
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
