On Mon, Jan 20, 2014 at 08:26:25PM +0100, David Herrmann wrote: > Instead of rounding down to the next lower page-boundary, round up. > dma-buf guarantees that we can map buffers in multiples of a page, so if > an exporter does not page-align, do it ourselves. > > This avoids issues if the exported buffer contains an unaligned size and > we crop it. In this case, the buffer is too small for the UDL CRTC. So we > round up to page-size now and avoid black borders. Worst case is we end up > reading out some random kernel memory, but we can never fault as the whole > page has the same access-rights. And in this case it's an issue of the > buggy exporting driver, not the importing one. > > Signed-off-by: David Herrmann <dh.herrmann@xxxxxxxxx> > --- > drivers/gpu/drm/udl/udl_gem.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/udl/udl_gem.c b/drivers/gpu/drm/udl/udl_gem.c > index df963a1..1069e57 100644 > --- a/drivers/gpu/drm/udl/udl_gem.c > +++ b/drivers/gpu/drm/udl/udl_gem.c > @@ -227,7 +227,7 @@ static int udl_prime_create(struct drm_device *dev, > struct udl_gem_object *obj; > int npages; > > - npages = size / PAGE_SIZE; > + npages = PAGE_ALIGN(size) >> PAGE_SHIFT; Imo the proper patch would be to reject exporting anything which isn't page-aligned as a dma-buf in the driver core (together with a quick review of the docs to make sure this requirement is clear). This is just a bug in the exporter. -Daniel > > *obj_p = NULL; > obj = udl_gem_alloc_object(dev, npages * PAGE_SIZE); > -- > 1.8.5.3 > -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
