[PATCH 08/10] drm/i915: Align GGTT sizes to a fence tile row
Chris Wilson
chris at chris-wilson.co.uk
Fri Dec 23 21:07:00 UTC 2016
Ensure the view occupies the full tile row so that reads/writes into the
VMA do not escape (via fenced detiling) into neighbouring objects - we
will pad the object with scratch pages to satisfy the fence. This
applies the lazy-tiling we employed on gen2/3 to gen4+.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/i915_drv.h | 5 +++--
drivers/gpu/drm/i915/i915_gem.c | 19 +++++++++++++------
drivers/gpu/drm/i915/i915_gem_tiling.c | 14 +++++++++-----
drivers/gpu/drm/i915/i915_vma.c | 8 ++++++--
4 files changed, 31 insertions(+), 15 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 3794b0933531..5ce233e1ebe5 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3439,9 +3439,10 @@ int i915_gem_open(struct drm_device *dev, struct drm_file *file);
void i915_gem_release(struct drm_device *dev, struct drm_file *file);
u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv, u64 size,
- int tiling_mode);
+ int tiling_mode, unsigned int stride);
u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size,
- int tiling_mode, bool fenced);
+ int tiling_mode, unsigned int stride,
+ bool fenced);
int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj,
enum i915_cache_level cache_level);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index b7c3af7292a2..f95bfe5b936a 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2069,16 +2069,21 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv)
* potential fence register mapping.
*/
u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv,
- u64 size, int tiling_mode)
+ u64 size, int tiling_mode, unsigned int stride)
{
u64 ggtt_size;
GEM_BUG_ON(size == 0);
- if (INTEL_GEN(dev_priv) >= 4 ||
- tiling_mode == I915_TILING_NONE)
+ if (tiling_mode == I915_TILING_NONE)
return size;
+ if (INTEL_GEN(dev_priv) >= 4) {
+ ggtt_size = stride;
+ ggtt_size *= tiling_mode == I915_TILING_Y ? 32 : 8;
+ return roundup(size, ggtt_size);
+ }
+
/* Previous chips need a power-of-two fence region when tiling */
if (IS_GEN3(dev_priv))
ggtt_size = 1024*1024;
@@ -2102,7 +2107,8 @@ u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv,
* potential fence register mapping.
*/
u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size,
- int tiling_mode, bool fenced)
+ int tiling_mode, unsigned int stride,
+ bool fenced)
{
GEM_BUG_ON(size == 0);
@@ -2119,7 +2125,7 @@ u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size,
* Previous chips need to be aligned to the size of the smallest
* fence register that can contain the object.
*/
- return i915_gem_get_ggtt_size(dev_priv, size, tiling_mode);
+ return i915_gem_get_ggtt_size(dev_priv, size, tiling_mode, stride);
}
static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj)
@@ -3741,7 +3747,8 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
u32 fence_size;
fence_size = i915_gem_get_ggtt_size(dev_priv, vma->size,
- i915_gem_object_get_tiling(obj));
+ i915_gem_object_get_tiling(obj),
+ i915_gem_object_get_stride(obj));
/* If the required space is larger than the available
* aperture, we will not able to find a slot for the
* object and unbinding the object now will be in
diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c
index 62ad375de6ca..bc9eb7e1da0e 100644
--- a/drivers/gpu/drm/i915/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/i915_gem_tiling.c
@@ -117,7 +117,8 @@ i915_tiling_ok(struct drm_i915_private *dev_priv,
return true;
}
-static bool i915_vma_fence_prepare(struct i915_vma *vma, int tiling_mode)
+static bool i915_vma_fence_prepare(struct i915_vma *vma,
+ int tiling_mode, unsigned int stride)
{
struct drm_i915_private *dev_priv = vma->vm->i915;
u32 size;
@@ -133,7 +134,7 @@ static bool i915_vma_fence_prepare(struct i915_vma *vma, int tiling_mode)
return false;
}
- size = i915_gem_get_ggtt_size(dev_priv, vma->size, tiling_mode);
+ size = i915_gem_get_ggtt_size(dev_priv, vma->size, tiling_mode, stride);
if (vma->node.size < size)
return false;
@@ -145,7 +146,8 @@ static bool i915_vma_fence_prepare(struct i915_vma *vma, int tiling_mode)
/* Make the current GTT allocation valid for the change in tiling. */
static int
-i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, int tiling_mode)
+i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj,
+ int tiling_mode, unsigned int stride)
{
struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
struct i915_vma *vma;
@@ -158,7 +160,7 @@ i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, int tiling_mode)
return 0;
list_for_each_entry(vma, &obj->vma_list, obj_link) {
- if (i915_vma_fence_prepare(vma, tiling_mode))
+ if (i915_vma_fence_prepare(vma, tiling_mode, stride))
continue;
ret = i915_vma_unbind(vma);
@@ -255,7 +257,9 @@ i915_gem_set_tiling(struct drm_device *dev, void *data,
* whilst executing a fenced command for an untiled object.
*/
- err = i915_gem_object_fence_prepare(obj, args->tiling_mode);
+ err = i915_gem_object_fence_prepare(obj,
+ args->tiling_mode,
+ args->stride);
if (!err) {
struct i915_vma *vma;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 3d083e86c4c5..bff6ea567bd5 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -282,10 +282,12 @@ void __i915_vma_set_map_and_fenceable(struct i915_vma *vma)
fence_size = i915_gem_get_ggtt_size(dev_priv,
vma->size,
- i915_gem_object_get_tiling(obj));
+ i915_gem_object_get_tiling(obj),
+ i915_gem_object_get_stride(obj));
fence_alignment = i915_gem_get_ggtt_alignment(dev_priv,
vma->size,
i915_gem_object_get_tiling(obj),
+ i915_gem_object_get_stride(obj),
true);
fenceable = (vma->node.size == fence_size &&
@@ -368,11 +370,13 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
size = max(size, vma->size);
if (flags & PIN_MAPPABLE)
size = i915_gem_get_ggtt_size(dev_priv, size,
- i915_gem_object_get_tiling(obj));
+ i915_gem_object_get_tiling(obj),
+ i915_gem_object_get_stride(obj));
alignment = max(max(alignment, vma->display_alignment),
i915_gem_get_ggtt_alignment(dev_priv, size,
i915_gem_object_get_tiling(obj),
+ i915_gem_object_get_stride(obj),
flags & PIN_MAPPABLE));
start = flags & PIN_OFFSET_BIAS ? flags & PIN_OFFSET_MASK : 0;
--
2.11.0
More information about the Intel-gfx-trybot
mailing list