| Message ID | 1495103306-7041-1-git-send-email-tomi.valkeinen@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Tomi, Thank you for the patch. On Thursday 18 May 2017 13:28:26 Tomi Valkeinen wrote: > omap_gem uses page alignment for buffer stride. The related calculations > are a bit off, though, as byte stride of 4096 gets aligned to 8192, > instead of 4096. This patch fixes those calculations. > > Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> > --- > drivers/gpu/drm/omapdrm/omap_gem.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/omapdrm/omap_gem.c > b/drivers/gpu/drm/omapdrm/omap_gem.c index 13abf221d153..4c41000ff4c4 > 100644 > --- a/drivers/gpu/drm/omapdrm/omap_gem.c > +++ b/drivers/gpu/drm/omapdrm/omap_gem.c > @@ -182,7 +182,7 @@ static void evict_entry(struct drm_gem_object *obj, > size_t size = PAGE_SIZE * n; > loff_t off = mmap_offset(obj) + > (entry->obj_pgoff << PAGE_SHIFT); > - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); > + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); How about int m = round_up(omap_obj->width << fmt, PAGE_SIZE); instead of open-coding it ? I find that a bit easier to understand. By the way, shifting left by fmt should be fine for TILFMT_8BIT, TILFMT_16BIT and TILFMT_32BIT that evaluate to 0, 1 and 2 respectively, but how does it work with TILFMT_PAGE ? fmt is computed by gem2fmt() in call cases, which returns TILFMT_PAGE in the default case (no tiled flag set). Can this happen in practice ? > > if (m > 1) { > int i; > @@ -424,7 +424,7 @@ static int fault_2d(struct drm_gem_object *obj, > * into account in some of the math, so figure out virtual stride > * in pages > */ > - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); > + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); > > /* We don't use vmf->pgoff since that has the fake offset: */ > pgoff = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
On 18/05/17 13:59, Laurent Pinchart wrote: > Hi Tomi, > > Thank you for the patch. > > On Thursday 18 May 2017 13:28:26 Tomi Valkeinen wrote: >> omap_gem uses page alignment for buffer stride. The related calculations >> are a bit off, though, as byte stride of 4096 gets aligned to 8192, >> instead of 4096. This patch fixes those calculations. >> >> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> >> --- >> drivers/gpu/drm/omapdrm/omap_gem.c | 4 ++-- >> 1 file changed, 2 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/gpu/drm/omapdrm/omap_gem.c >> b/drivers/gpu/drm/omapdrm/omap_gem.c index 13abf221d153..4c41000ff4c4 >> 100644 >> --- a/drivers/gpu/drm/omapdrm/omap_gem.c >> +++ b/drivers/gpu/drm/omapdrm/omap_gem.c >> @@ -182,7 +182,7 @@ static void evict_entry(struct drm_gem_object *obj, >> size_t size = PAGE_SIZE * n; >> loff_t off = mmap_offset(obj) + >> (entry->obj_pgoff << PAGE_SHIFT); >> - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); >> + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); > > How about > > int m = round_up(omap_obj->width << fmt, PAGE_SIZE); > > instead of open-coding it ? I find that a bit easier to understand. That should be: round_up(omap_obj->width << fmt, PAGE_SIZE) / PAGE_SIZE; Yes, I think that's more understandable. > By the way, shifting left by fmt should be fine for TILFMT_8BIT, TILFMT_16BIT > and TILFMT_32BIT that evaluate to 0, 1 and 2 respectively, but how does it > work with TILFMT_PAGE ? fmt is computed by gem2fmt() in call cases, which > returns TILFMT_PAGE in the default case (no tiled flag set). Can this happen > in practice ? These functions are only called for 2D buffers. I do find shifting by 'enum tiler_fmt' quite ugly, though... Tomi
Hi Tomi, On Thursday 18 May 2017 14:14:35 Tomi Valkeinen wrote: > On 18/05/17 13:59, Laurent Pinchart wrote: > > On Thursday 18 May 2017 13:28:26 Tomi Valkeinen wrote: > >> omap_gem uses page alignment for buffer stride. The related calculations > >> are a bit off, though, as byte stride of 4096 gets aligned to 8192, > >> instead of 4096. This patch fixes those calculations. > >> > >> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> > >> --- > >> > >> drivers/gpu/drm/omapdrm/omap_gem.c | 4 ++-- > >> 1 file changed, 2 insertions(+), 2 deletions(-) > >> > >> diff --git a/drivers/gpu/drm/omapdrm/omap_gem.c > >> b/drivers/gpu/drm/omapdrm/omap_gem.c index 13abf221d153..4c41000ff4c4 > >> 100644 > >> --- a/drivers/gpu/drm/omapdrm/omap_gem.c > >> +++ b/drivers/gpu/drm/omapdrm/omap_gem.c > >> @@ -182,7 +182,7 @@ static void evict_entry(struct drm_gem_object *obj, > >> size_t size = PAGE_SIZE * n; > >> loff_t off = mmap_offset(obj) + > >> (entry->obj_pgoff << PAGE_SHIFT); > >> - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); > >> + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); > > > > How about > > > > int m = round_up(omap_obj->width << fmt, PAGE_SIZE); > > > > instead of open-coding it ? I find that a bit easier to understand. > > That should be: > > round_up(omap_obj->width << fmt, PAGE_SIZE) / PAGE_SIZE; Sorry, I meant DIV_ROUND_UP, not round_up. > Yes, I think that's more understandable. > > > By the way, shifting left by fmt should be fine for TILFMT_8BIT, > > TILFMT_16BIT and TILFMT_32BIT that evaluate to 0, 1 and 2 respectively, > > but how does it work with TILFMT_PAGE ? fmt is computed by gem2fmt() in > > call cases, which returns TILFMT_PAGE in the default case (no tiled flag > > set). Can this happen in practice ? > > These functions are only called for 2D buffers. I do find shifting by > 'enum tiler_fmt' quite ugly, though... So do I. This should be added to the infinite todo list :-)
On Thu, May 18, 2017 at 02:14:35PM +0300, Tomi Valkeinen wrote: > On 18/05/17 13:59, Laurent Pinchart wrote: > > Hi Tomi, > > > > Thank you for the patch. > > > > On Thursday 18 May 2017 13:28:26 Tomi Valkeinen wrote: > >> omap_gem uses page alignment for buffer stride. The related calculations > >> are a bit off, though, as byte stride of 4096 gets aligned to 8192, > >> instead of 4096. This patch fixes those calculations. > >> > >> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> > >> --- > >> drivers/gpu/drm/omapdrm/omap_gem.c | 4 ++-- > >> 1 file changed, 2 insertions(+), 2 deletions(-) > >> > >> diff --git a/drivers/gpu/drm/omapdrm/omap_gem.c > >> b/drivers/gpu/drm/omapdrm/omap_gem.c index 13abf221d153..4c41000ff4c4 > >> 100644 > >> --- a/drivers/gpu/drm/omapdrm/omap_gem.c > >> +++ b/drivers/gpu/drm/omapdrm/omap_gem.c > >> @@ -182,7 +182,7 @@ static void evict_entry(struct drm_gem_object *obj, > >> size_t size = PAGE_SIZE * n; > >> loff_t off = mmap_offset(obj) + > >> (entry->obj_pgoff << PAGE_SHIFT); > >> - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); > >> + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); > > > > How about > > > > int m = round_up(omap_obj->width << fmt, PAGE_SIZE); > > > > instead of open-coding it ? I find that a bit easier to understand. > > That should be: > > round_up(omap_obj->width << fmt, PAGE_SIZE) / PAGE_SIZE; Or a bit more optimally just DIV_ROUND_UP()
diff --git a/drivers/gpu/drm/omapdrm/omap_gem.c b/drivers/gpu/drm/omapdrm/omap_gem.c index 13abf221d153..4c41000ff4c4 100644 --- a/drivers/gpu/drm/omapdrm/omap_gem.c +++ b/drivers/gpu/drm/omapdrm/omap_gem.c @@ -182,7 +182,7 @@ static void evict_entry(struct drm_gem_object *obj, size_t size = PAGE_SIZE * n; loff_t off = mmap_offset(obj) + (entry->obj_pgoff << PAGE_SHIFT); - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); if (m > 1) { int i; @@ -424,7 +424,7 @@ static int fault_2d(struct drm_gem_object *obj, * into account in some of the math, so figure out virtual stride * in pages */ - const int m = 1 + ((omap_obj->width << fmt) / PAGE_SIZE); + const int m = 1 + (((omap_obj->width - 1) << fmt) / PAGE_SIZE); /* We don't use vmf->pgoff since that has the fake offset: */ pgoff = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
omap_gem uses page alignment for buffer stride. The related calculations are a bit off, though, as byte stride of 4096 gets aligned to 8192, instead of 4096. This patch fixes those calculations. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> --- drivers/gpu/drm/omapdrm/omap_gem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)