[Intel-gfx] [PATCH 4/5] Align YS tile base address to 64KB

[Damien Lespiau]

On Tue, Mar 31, 2015 at 10:49:22AM -0700, Anuj Phogat wrote:
> On Tue, Mar 31, 2015 at 7:26 AM, Damien Lespiau
> <damien.lespiau at intel.com> wrote:
> > On Mon, Mar 30, 2015 at 02:00:07PM -0700, Anuj Phogat wrote:
> >> Signed-off-by: Anuj Phogat <anuj.phogat at gmail.com>
> >> ---
> >>  intel/intel_bufmgr_gem.c | 7 ++++++-
> >>  1 file changed, 6 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/intel/intel_bufmgr_gem.c b/intel/intel_bufmgr_gem.c
> >> index 7c50e26..775a9f9 100644
> >> --- a/intel/intel_bufmgr_gem.c
> >> +++ b/intel/intel_bufmgr_gem.c
> >> @@ -289,8 +289,13 @@ drm_intel_gem_bo_tile_size(drm_intel_bufmgr_gem *bufmgr_gem, unsigned long size,
> >>       if (*tiling_mode == I915_TILING_NONE)
> >>               return size;
> >>
> >> +     /* Tiled surface base addresses must be tile aligned (64KB aligned
> >> +      * for TileYS, 4KB aligned for all other tile modes).
> >> +      */
> >> +     if (*tiling_mode == I915_TILING_YS)
> >> +             return ROUND_UP_TO(size, 64 * 1024);
> >>       /* 965+ just need multiples of page size for tiling */
> >> -     if (bufmgr_gem->gen >= 4)
> >> +     else if (bufmgr_gem->gen >= 4)
> >>               return ROUND_UP_TO(size, 4096);
> >
> > I'm confused. You're saying you want to align the address of those
> > buffers to 64k, but here we're talking about the object size. At the
> > moment, the kernel places buffers in the address space and it was chosen
> > that the kernel didn't need to know about those tiling formats. So we
> > need something else if that constraint is indeed true (could you tell
> > us the source for this assertion? privately if needed).
> >
> This comment is invalid here. It was meant for surface state in Mesa.
> I'll remove it.

Don't you have the exact same problem? how does the kernel know about
this alignment constraint when resolving the relocation?

-- 
Damien