[Intel-gfx] [PATCH v4 8/8] hack: align dumb buffer stride to 4k to allow for gtt remapping

[Daniel Vetter]

On Wed, Jan 30, 2019 at 05:38:03PM +0200, Ville Syrjälä wrote:
> On Wed, Jan 30, 2019 at 11:06:07AM +0100, Daniel Vetter wrote:
> > On Wed, Jan 30, 2019 at 10:54:15AM +0100, Daniel Vetter wrote:
> > > On Fri, Jan 18, 2019 at 07:11:06PM +0200, Ville Syrjala wrote:
> > > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > 
> > > > v2: Leave the stride alone for buffers that look to be for the cursor
> > > > ---
> > > >  drivers/gpu/drm/i915/i915_gem.c | 7 ++++++-
> > > >  1 file changed, 6 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> > > > index 1e7c95d0fea1..b4b34519af80 100644
> > > > --- a/drivers/gpu/drm/i915/i915_gem.c
> > > > +++ b/drivers/gpu/drm/i915/i915_gem.c
> > > > @@ -745,7 +745,12 @@ i915_gem_dumb_create(struct drm_file *file,
> > > >  		     struct drm_mode_create_dumb *args)
> > > >  {
> > > >  	/* have to work out size/pitch and return them */
> > > > -	args->pitch = ALIGN(args->width * DIV_ROUND_UP(args->bpp, 8), 64);
> > > > +	if (args->bpp == 32 && (args->width == 64 ||
> > > > +				args->width == 128 ||
> > > > +				args->width == 256))
> > > > +		args->pitch = ALIGN(args->width * DIV_ROUND_UP(args->bpp, 8), 64);
> > > > +	else
> > > > +		args->pitch = ALIGN(args->width * DIV_ROUND_UP(args->bpp, 8), 4096);
> > > 
> > > Shouldn't we convert this into a non-hack and just do it anytime the pitch
> > > is too wide for the display to handle?
> > > 
> > > Or am I missing something somewhere? -modesetting et all will dtrt
> > > because tiling, but this should help with boot splashes and stuff like
> > > that (but those tend to not do side-by-side, so maybe that's why the get
> > > away with it).
> > 
> > Correction: We need this, and before we start bumping the limits. Any dumb
> > buffer you create (within the limits) we must be able to scan out. So we
> > definitely need to have this for super-big buffers on gen7+. Even if it's
> > fairly theoretical.
> 
> Is there some userpsace that actually wants to create a huge dumb
> buffer like that?

Given that the core drm_mode_dumb_create doesn't check anything, I guess
not. Would be good to add.

> Looks like we don't even check against the max fb dimensions in
> dumb_create so you can create any sized dumb buffer you want.
> So even now there is no guarantee that the addfb will actually
> succeed.

Well we do tell userspace up to what limit we expect it to succeed, and
dumb_create is supposed to come up with the correct stride. I think
there's no good reason to give userspace a stride that doesn't work at
all, and it's essentially this patch, but with a slightly different if
condition.
-Daniel

> 
> > 
> > I think we should even have an igt which creates a max size buffer (using
> > dumb create) and then makes sure we can scan that out.
> > -Daniel
> > 
> > > -Daniel
> > > 
> > > >  	args->size = args->pitch * args->height;
> > > >  	return i915_gem_create(file, to_i915(dev),
> > > >  			       args->size, &args->handle);
> > > > -- 
> > > > 2.19.2
> > > > 
> > > > _______________________________________________
> > > > Intel-gfx mailing list
> > > > Intel-gfx at lists.freedesktop.org
> > > > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
> > > 
> > > -- 
> > > Daniel Vetter
> > > Software Engineer, Intel Corporation
> > > http://blog.ffwll.ch
> > 
> > -- 
> > Daniel Vetter
> > Software Engineer, Intel Corporation
> > http://blog.ffwll.ch
> 
> -- 
> Ville Syrjälä
> Intel

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch