[Intel-gfx] [PATCH] drm/i915/selftests: Adjust y-tiling height for older machines

[Rodrigo Vivi]

On Fri, Jul 06, 2018 at 09:39:14PM +0100, Chris Wilson wrote:
> Quoting Rodrigo Vivi (2018-07-06 21:27:52)
> > On Fri, Jul 06, 2018 at 06:15:37PM +0100, Chris Wilson wrote:
> > > Older machines do not have the 128-byte tile width format for
> > > I915_TILING_Y and so we must adapt our reference swizzle.
> > > 
> > > Testcase: igt/drv_selftest/live_objects #gdg
> > 
> > The change below itself makes sense to me, but I'm trying to understand
> > where this came from....
> 
> The result doesn't look right, so scrap it.
>  
> > Looking to https://intel-gfx-ci.01.org/tree/drm-tip/igt@drv_selftest@live_objects.html
> > is this related to issues on fi-gdg-551?
> 
> Would only apply to gdg in the farm.
>  
> > Or is this related to that APL bugzilla entry?

that flash of a moment when I wondered you could be actually targeting this:
https://bugs.freedesktop.org/show_bug.cgi?id=107113


> 
> Which?
>  
> > > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > > ---
> > >  drivers/gpu/drm/i915/selftests/i915_gem_object.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_object.c b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
> > > index 6fe71865b710..8a35d2f70671 100644
> > > --- a/drivers/gpu/drm/i915/selftests/i915_gem_object.c
> > > +++ b/drivers/gpu/drm/i915/selftests/i915_gem_object.c
> > > @@ -171,7 +171,7 @@ static u64 tiled_offset(const struct tile *tile, u64 v)
> > >               v += x;
> > >       } else {
> > >               const unsigned int ytile_span = 16;
> > 
> > could we also figure this value from somewhere else instead of
> > leaving it hardcoded for all platforms here?
> 
> The only place where manual detiling is used inside the kernel. And if
> we were, it would be a lot of specialised code, where obfuscation of
> magic macros is unlikely to help (careful handling of cachelines being
> at the forefront). Interesting question as to whether we do provide a
> bounce buffer mmap to replace GTT mmap? Just say no.
> -Chris