drm/i915: Store required fence size/alignment for GGTT vma

The fence size/alignment is a combination of the vma size plus object
tiling parameters. Those parameters are rarely changed, making the fence
size/alignemnt roughly constant for the lifetime of the VMA. We can
simplify subsequent calculations by precalculating the size/alignment
required for GGTT vma taking fencing into account (with an update if we
do change the tiling or stride).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170109161613.11881-4-chris@chris-wilson.co.uk

6 files changed, 73 insertions, 81 deletions

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 8185229f370f..5ee76628be98 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3360,11 +3360,10 @@ int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj,
 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,
+u32 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv, u32 size,
 			   int tiling_mode, unsigned int stride);
-u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size,
-				int tiling_mode, unsigned int stride,
-				bool fenced);
+u32 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u32 size,
+				int tiling_mode, unsigned int stride);
 
 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 07cc0d01915f..1f9496e587dc 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2026,10 +2026,10 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv)
  * Return the required global GTT size for an object, taking into account
  * potential fence register mapping.
  */
-u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv,
-			   u64 size, int tiling_mode, unsigned int stride)
+u32 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv,
+			   u32 size, int tiling_mode, unsigned int stride)
 {
-	u64 ggtt_size;
+	u32 ggtt_size;
 
 	GEM_BUG_ON(!size);
 
@@ -2062,14 +2062,12 @@ u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv,
  * @size: object size
  * @tiling_mode: tiling mode
  * @stride: tiling stride
- * @fenced: is fenced alignment required or not
  *
  * Return the required global GTT alignment for an object, taking into account
  * potential fence register mapping.
  */
-u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size,
-				int tiling_mode, unsigned int stride,
-				bool fenced)
+u32 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u32 size,
+				int tiling_mode, unsigned int stride)
 {
 	GEM_BUG_ON(!size);
 
@@ -2077,9 +2075,7 @@ u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size,
 	 * Minimum alignment is 4k (GTT page size), but might be greater
 	 * if a fence register is needed for the object.
 	 */
-	if (INTEL_GEN(dev_priv) >= 4 ||
-	    (!fenced && (IS_G33(dev_priv) || IS_PINEVIEW(dev_priv))) ||
-	    tiling_mode == I915_TILING_NONE)
+	if (INTEL_GEN(dev_priv) >= 4 || tiling_mode == I915_TILING_NONE)
 		return 4096;
 
 	/*
@@ -3558,7 +3554,7 @@ i915_gem_object_unpin_from_display_plane(struct i915_vma *vma)
 		return;
 
 	if (--vma->obj->pin_display == 0)
-		vma->display_alignment = 0;
+		vma->display_alignment = 4096;
 
 	/* Bump the LRU to try and avoid premature eviction whilst flipping  */
 	if (!i915_vma_is_active(vma))
@@ -3703,11 +3699,6 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
 			return ERR_PTR(-ENOSPC);
 
 		if (flags & PIN_MAPPABLE) {
-			u32 fence_size;
-
-			fence_size = i915_gem_get_ggtt_size(dev_priv, vma->size,
-							    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
@@ -3715,7 +3706,7 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
 			 * the object in and out of the Global GTT and
 			 * waste a lot of cycles under the mutex.
 			 */
-			if (fence_size > dev_priv->ggtt.mappable_end)
+			if (vma->fence_size > dev_priv->ggtt.mappable_end)
 				return ERR_PTR(-E2BIG);
 
 			/* If NONBLOCK is set the caller is optimistically
@@ -3734,7 +3725,7 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
 			 * we could try to minimise harm to others.
 			 */
 			if (flags & PIN_NONBLOCK &&
-			    fence_size > dev_priv->ggtt.mappable_end / 2)
+			    vma->fence_size > dev_priv->ggtt.mappable_end / 2)
 				return ERR_PTR(-ENOSPC);
 		}
 
diff --git a/drivers/gpu/drm/i915/i915_gem_fence_reg.c b/drivers/gpu/drm/i915/i915_gem_fence_reg.c
index 26f242359fbd..8b37c4cb311a 100644
--- a/drivers/gpu/drm/i915/i915_gem_fence_reg.c
+++ b/drivers/gpu/drm/i915/i915_gem_fence_reg.c
@@ -79,11 +79,11 @@ static void i965_write_fence_reg(struct drm_i915_fence_reg *fence,
 	if (vma) {
 		unsigned int stride = i915_gem_object_get_stride(vma->obj);
 		u32 row_size = i915_gem_object_get_tile_row_size(vma->obj);
-		u32 size = rounddown((u32)vma->node.size, row_size);
+		u32 size = rounddown((u32)vma->fence_size, row_size);
 
 		GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
 		GEM_BUG_ON(vma->node.start & 4095);
-		GEM_BUG_ON(vma->node.size & 4095);
+		GEM_BUG_ON(vma->fence_size & 4095);
 		GEM_BUG_ON(stride & 127);
 
 		val = (vma->node.start + size - 4096) << 32;
@@ -128,8 +128,8 @@ static void i915_write_fence_reg(struct drm_i915_fence_reg *fence,
 
 		GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
 		GEM_BUG_ON(vma->node.start & ~I915_FENCE_START_MASK);
-		GEM_BUG_ON(!is_power_of_2(vma->node.size));
-		GEM_BUG_ON(vma->node.start & (vma->node.size - 1));
+		GEM_BUG_ON(!is_power_of_2(vma->fence_size));
+		GEM_BUG_ON(vma->node.start & (vma->fence_size - 1));
 
 		if (is_y_tiled && HAS_128_BYTE_Y_TILING(fence->i915))
 			stride /= 128;
@@ -140,7 +140,7 @@ static void i915_write_fence_reg(struct drm_i915_fence_reg *fence,
 		val = vma->node.start;
 		if (is_y_tiled)
 			val |= BIT(I830_FENCE_TILING_Y_SHIFT);
-		val |= I915_FENCE_SIZE_BITS(vma->node.size);
+		val |= I915_FENCE_SIZE_BITS(vma->fence_size);
 		val |= ilog2(stride) << I830_FENCE_PITCH_SHIFT;
 
 		val |= I830_FENCE_REG_VALID;
@@ -162,20 +162,18 @@ static void i830_write_fence_reg(struct drm_i915_fence_reg *fence,
 
 	val = 0;
 	if (vma) {
-		unsigned int tiling = i915_gem_object_get_tiling(vma->obj);
-		bool is_y_tiled = tiling == I915_TILING_Y;
 		unsigned int stride = i915_gem_object_get_stride(vma->obj);
 
 		GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
 		GEM_BUG_ON(vma->node.start & ~I830_FENCE_START_MASK);
-		GEM_BUG_ON(!is_power_of_2(vma->node.size));
+		GEM_BUG_ON(!is_power_of_2(vma->fence_size));
 		GEM_BUG_ON(!is_power_of_2(stride / 128));
-		GEM_BUG_ON(vma->node.start & (vma->node.size - 1));
+		GEM_BUG_ON(vma->node.start & (vma->fence_size - 1));
 
 		val = vma->node.start;
-		if (is_y_tiled)
+		if (i915_gem_object_get_tiling(vma->obj) == I915_TILING_Y)
 			val |= BIT(I830_FENCE_TILING_Y_SHIFT);
-		val |= I830_FENCE_SIZE_BITS(vma->node.size);
+		val |= I830_FENCE_SIZE_BITS(vma->fence_size);
 		val |= ilog2(stride / 128) << I830_FENCE_PITCH_SHIFT;
 		val |= I830_FENCE_REG_VALID;
 	}
diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c
index 51b8d71876b7..23a896cd934f 100644
--- a/drivers/gpu/drm/i915/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/i915_gem_tiling.c
@@ -120,25 +120,18 @@ i915_tiling_ok(struct drm_i915_private *dev_priv,
 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;
+	struct drm_i915_private *i915 = vma->vm->i915;
+	u32 size, alignment;
 
 	if (!i915_vma_is_map_and_fenceable(vma))
 		return true;
 
-	if (INTEL_GEN(dev_priv) == 3) {
-		if (vma->node.start & ~I915_FENCE_START_MASK)
-			return false;
-	} else {
-		if (vma->node.start & ~I830_FENCE_START_MASK)
-			return false;
-	}
-
-	size = i915_gem_get_ggtt_size(dev_priv, vma->size, tiling_mode, stride);
+	size = i915_gem_get_ggtt_size(i915, vma->size, tiling_mode, stride);
 	if (vma->node.size < size)
 		return false;
 
-	if (vma->node.start & (size - 1))
+	alignment = i915_gem_get_ggtt_alignment(i915, vma->size, tiling_mode, stride);
+	if (vma->node.start & (alignment - 1))
 		return false;
 
 	return true;
@@ -156,6 +149,9 @@ i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj,
 		return 0;
 
 	list_for_each_entry(vma, &obj->vma_list, obj_link) {
+		if (!i915_vma_is_ggtt(vma))
+			break;
+
 		if (i915_vma_fence_prepare(vma, tiling_mode, stride))
 			continue;
 
@@ -277,10 +273,18 @@ i915_gem_set_tiling(struct drm_device *dev, void *data,
 			mutex_unlock(&obj->mm.lock);
 
 			list_for_each_entry(vma, &obj->vma_list, obj_link) {
-				if (!vma->fence)
-					continue;
-
-				vma->fence->dirty = true;
+				if (!i915_vma_is_ggtt(vma))
+					break;
+
+				vma->fence_size = i915_gem_get_ggtt_size(dev_priv, vma->size,
+									 args->tiling_mode,
+									 args->stride);
+				vma->fence_alignment = i915_gem_get_ggtt_alignment(dev_priv, vma->size,
+										   args->tiling_mode,
+										   args->stride);
+
+				if (vma->fence)
+					vma->fence->dirty = true;
 			}
 			obj->tiling_and_stride =
 				args->stride | args->tiling_mode;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 734f77b7697f..a605d735662c 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -91,6 +91,7 @@ __i915_vma_create(struct drm_i915_gem_object *obj,
 	vma->vm = vm;
 	vma->obj = obj;
 	vma->size = obj->base.size;
+	vma->display_alignment = 4096;
 
 	if (view) {
 		vma->ggtt_view = *view;
@@ -110,6 +111,17 @@ __i915_vma_create(struct drm_i915_gem_object *obj,
 	}
 
 	if (i915_is_ggtt(vm)) {
+		GEM_BUG_ON(overflows_type(vma->size, u32));
+		vma->fence_size = i915_gem_get_ggtt_size(vm->i915, vma->size,
+							 i915_gem_object_get_tiling(obj),
+							 i915_gem_object_get_stride(obj));
+		GEM_BUG_ON(vma->fence_size & 4095);
+
+		vma->fence_alignment = i915_gem_get_ggtt_alignment(vm->i915, vma->size,
+								   i915_gem_object_get_tiling(obj),
+								   i915_gem_object_get_stride(obj));
+		GEM_BUG_ON(!is_power_of_2(vma->fence_alignment));
+
 		vma->flags |= I915_VMA_GGTT;
 		list_add(&vma->obj_link, &obj->vma_list);
 	} else {
@@ -277,34 +289,24 @@ i915_vma_misplaced(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 
 void __i915_vma_set_map_and_fenceable(struct i915_vma *vma)
 {
-	struct drm_i915_gem_object *obj = vma->obj;
-	struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
 	bool mappable, fenceable;
-	u32 fence_size, fence_alignment;
 
-	fence_size = i915_gem_get_ggtt_size(dev_priv,
-					    vma->size,
-					    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);
-	GEM_BUG_ON(!is_power_of_2(fence_alignment));
-
-	fenceable = (vma->node.size == fence_size &&
-		     (vma->node.start & (fence_alignment - 1)) == 0);
-
-	mappable = (vma->node.start + fence_size <=
-		    dev_priv->ggtt.mappable_end);
+	GEM_BUG_ON(!i915_vma_is_ggtt(vma));
+	GEM_BUG_ON(!vma->fence_size);
 
 	/*
 	 * Explicitly disable for rotated VMA since the display does not
 	 * need the fence and the VMA is not accessible to other users.
 	 */
-	if (mappable && fenceable &&
-	    vma->ggtt_view.type != I915_GGTT_VIEW_ROTATED)
+	if (vma->ggtt_view.type == I915_GGTT_VIEW_ROTATED)
+		return;
+
+	fenceable = (vma->node.size >= vma->fence_size &&
+		     (vma->node.start & (vma->fence_alignment - 1)) == 0);
+
+	mappable = vma->node.start + vma->fence_size <= i915_vm_to_ggtt(vma->vm)->mappable_end;
+
+	if (mappable && fenceable)
 		vma->flags |= I915_VMA_CAN_FENCE;
 	else
 		vma->flags &= ~I915_VMA_CAN_FENCE;
@@ -371,17 +373,12 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 	GEM_BUG_ON(drm_mm_node_allocated(&vma->node));
 
 	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_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));
-	GEM_BUG_ON(!is_power_of_2(alignment));
+	alignment = max(alignment, vma->display_alignment);
+	if (flags & PIN_MAPPABLE) {
+		size = max_t(typeof(size), size, vma->fence_size);
+		alignment = max_t(typeof(alignment),
+				  alignment, vma->fence_alignment);
+	}
 
 	start = flags & PIN_OFFSET_BIAS ? flags & PIN_OFFSET_MASK : 0;
 
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index e3b2b3b1e056..a969bbb65871 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -55,6 +55,9 @@ struct i915_vma {
 	u64 size;
 	u64 display_alignment;
 
+	u32 fence_size;
+	u32 fence_alignment;
+
 	unsigned int flags;
 	/**
 	 * How many users have pinned this object in GTT space. The following