drm_mm fixes, take 3, final?

drm_mm fixes, take 3, final?

[Chris Wilson]

With a lot of polish applied, Joonas has reviewed the series - all but
for [04/38] "lib: Add a simple prime number generator"
[lib/prime_numbers.c]. Anyone feel like poking around at a bit of number
theory?

Other than it would appear to be ready for Daniel to sort out the merge
between drm-misc/i915... Please do take a look and see if you can spot
anything else that needs fixing/improving.
-Chris

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 01/38] drm/i915: Use the MRU stack search after evicting

[Chris Wilson]

When we evict from the GTT to make room for an object, the hole we
create is put onto the MRU stack inside the drm_mm range manager. On the
next search pass, we can speed up a PIN_HIGH allocation by referencing
that stack for the new hole.

v2: Pull together the 3 identical implements (ahem, a couple were
outdated) into a common routine for allocating a node and evicting as
necessary.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/i915/gvt/aperture_gm.c | 33 +++++-----------
 drivers/gpu/drm/i915/i915_gem_gtt.c    | 72 ++++++++++++++++++++++++----------
 drivers/gpu/drm/i915/i915_gem_gtt.h    |  5 +++
 drivers/gpu/drm/i915/i915_vma.c        | 40 ++-----------------
 4 files changed, 70 insertions(+), 80 deletions(-)

diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c b/drivers/gpu/drm/i915/gvt/aperture_gm.c
index 7d33b607bc89..1bb7a5b80d47 100644
--- a/drivers/gpu/drm/i915/gvt/aperture_gm.c
+++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c
@@ -48,47 +48,34 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
 {
 	struct intel_gvt *gvt = vgpu->gvt;
 	struct drm_i915_private *dev_priv = gvt->dev_priv;
-	u32 alloc_flag, search_flag;
+	unsigned int flags;
 	u64 start, end, size;
 	struct drm_mm_node *node;
-	int retried = 0;
 	int ret;
 
 	if (high_gm) {
-		search_flag = DRM_MM_SEARCH_BELOW;
-		alloc_flag = DRM_MM_CREATE_TOP;
 		node = &vgpu->gm.high_gm_node;
 		size = vgpu_hidden_sz(vgpu);
 		start = gvt_hidden_gmadr_base(gvt);
 		end = gvt_hidden_gmadr_end(gvt);
+		flags = PIN_HIGH;
 	} else {
-		search_flag = DRM_MM_SEARCH_DEFAULT;
-		alloc_flag = DRM_MM_CREATE_DEFAULT;
 		node = &vgpu->gm.low_gm_node;
 		size = vgpu_aperture_sz(vgpu);
 		start = gvt_aperture_gmadr_base(gvt);
 		end = gvt_aperture_gmadr_end(gvt);
+		flags = PIN_MAPPABLE;
 	}
 
 	mutex_lock(&dev_priv->drm.struct_mutex);
-search_again:
-	ret = drm_mm_insert_node_in_range_generic(&dev_priv->ggtt.base.mm,
-						  node, size, 4096,
-						  I915_COLOR_UNEVICTABLE,
-						  start, end, search_flag,
-						  alloc_flag);
-	if (ret) {
-		ret = i915_gem_evict_something(&dev_priv->ggtt.base,
-					       size, 4096,
-					       I915_COLOR_UNEVICTABLE,
-					       start, end, 0);
-		if (ret == 0 && ++retried < 3)
-			goto search_again;
-
-		gvt_err("fail to alloc %s gm space from host, retried %d\n",
-				high_gm ? "high" : "low", retried);
-	}
+	ret = i915_gem_gtt_insert(&dev_priv->ggtt.base, node,
+				  size, 4096, I915_COLOR_UNEVICTABLE,
+				  start, end, flags);
 	mutex_unlock(&dev_priv->drm.struct_mutex);
+	if (ret)
+		gvt_err("fail to alloc %s gm space from host\n",
+			high_gm ? "high" : "low");
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 2083c899ab78..97fd66e4e3d0 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2056,7 +2056,6 @@ static int gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt)
 	struct i915_address_space *vm = &ppgtt->base;
 	struct drm_i915_private *dev_priv = ppgtt->base.i915;
 	struct i915_ggtt *ggtt = &dev_priv->ggtt;
-	bool retried = false;
 	int ret;
 
 	/* PPGTT PDEs reside in the GGTT and consists of 512 entries. The
@@ -2069,29 +2068,14 @@ static int gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt)
 	if (ret)
 		return ret;
 
-alloc:
-	ret = drm_mm_insert_node_in_range_generic(&ggtt->base.mm, &ppgtt->node,
-						  GEN6_PD_SIZE, GEN6_PD_ALIGN,
-						  I915_COLOR_UNEVICTABLE,
-						  0, ggtt->base.total,
-						  DRM_MM_TOPDOWN);
-	if (ret == -ENOSPC && !retried) {
-		ret = i915_gem_evict_something(&ggtt->base,
-					       GEN6_PD_SIZE, GEN6_PD_ALIGN,
-					       I915_COLOR_UNEVICTABLE,
-					       0, ggtt->base.total,
-					       0);
-		if (ret)
-			goto err_out;
-
-		retried = true;
-		goto alloc;
-	}
-
+	ret = i915_gem_gtt_insert(&ggtt->base, &ppgtt->node,
+				  GEN6_PD_SIZE, GEN6_PD_ALIGN,
+				  I915_COLOR_UNEVICTABLE,
+				  0, ggtt->base.total,
+				  PIN_HIGH);
 	if (ret)
 		goto err_out;
 
-
 	if (ppgtt->node.start < ggtt->mappable_end)
 		DRM_DEBUG("Forced to use aperture for PDEs\n");
 
@@ -3565,3 +3549,49 @@ i915_get_ggtt_vma_pages(struct i915_vma *vma)
 	return ret;
 }
 
+int i915_gem_gtt_insert(struct i915_address_space *vm,
+			struct drm_mm_node *node,
+			u64 size, u64 alignment, unsigned long colour,
+			u64 start, u64 end, unsigned int flags)
+{
+	u32 search_flag, alloc_flag;
+	int err;
+
+	lockdep_assert_held(&vm->i915->drm.struct_mutex);
+
+	if (flags & PIN_HIGH) {
+		search_flag = DRM_MM_SEARCH_BELOW;
+		alloc_flag = DRM_MM_CREATE_TOP;
+	} else {
+		search_flag = DRM_MM_SEARCH_DEFAULT;
+		alloc_flag = DRM_MM_CREATE_DEFAULT;
+	}
+
+	/* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks,
+	 * so we know that we always have a minimum alignment of 4096.
+	 * The drm_mm range manager is optimised to return results
+	 * with zero alignment, so where possible use the optimal
+	 * path.
+	 */
+	GEM_BUG_ON(size & 4095);
+	if (alignment <= 4096)
+		alignment = 0;
+
+	err = drm_mm_insert_node_in_range_generic(&vm->mm, node,
+						  size, alignment, colour,
+						  start, end,
+						  search_flag, alloc_flag);
+	if (err != -ENOSPC)
+		return err;
+
+	err = i915_gem_evict_something(vm, size, alignment, colour,
+				       start, end, flags);
+	if (err)
+		return err;
+
+	search_flag = DRM_MM_SEARCH_DEFAULT;
+	return drm_mm_insert_node_in_range_generic(&vm->mm, node,
+						   size, alignment, colour,
+						   start, end,
+						   search_flag, alloc_flag);
+}
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 8965bbb13db7..3a178978d3ce 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -520,6 +520,11 @@ int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
 void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
 			       struct sg_table *pages);
 
+int i915_gem_gtt_insert(struct i915_address_space *vm,
+			struct drm_mm_node *node,
+			u64 size, u64 alignment, unsigned long colour,
+			u64 start, u64 end, unsigned int flags);
+
 /* Flags used by pin/bind&friends. */
 #define PIN_NONBLOCK		BIT(0)
 #define PIN_MAPPABLE		BIT(1)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 37c3eebe8316..f709c9b76358 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -410,43 +410,11 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 				goto err_unpin;
 		}
 	} else {
-		u32 search_flag, alloc_flag;
-
-		if (flags & PIN_HIGH) {
-			search_flag = DRM_MM_SEARCH_BELOW;
-			alloc_flag = DRM_MM_CREATE_TOP;
-		} else {
-			search_flag = DRM_MM_SEARCH_DEFAULT;
-			alloc_flag = DRM_MM_CREATE_DEFAULT;
-		}
-
-		/* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks,
-		 * so we know that we always have a minimum alignment of 4096.
-		 * The drm_mm range manager is optimised to return results
-		 * with zero alignment, so where possible use the optimal
-		 * path.
-		 */
-		if (alignment <= 4096)
-			alignment = 0;
-
-search_free:
-		ret = drm_mm_insert_node_in_range_generic(&vma->vm->mm,
-							  &vma->node,
-							  size, alignment,
-							  obj->cache_level,
-							  start, end,
-							  search_flag,
-							  alloc_flag);
-		if (ret) {
-			ret = i915_gem_evict_something(vma->vm, size, alignment,
-						       obj->cache_level,
-						       start, end,
-						       flags);
-			if (ret == 0)
-				goto search_free;
-
+		ret = i915_gem_gtt_insert(vma->vm, &vma->node,
+					  size, alignment, obj->cache_level,
+					  start, end, flags);
+		if (ret)
 			goto err_unpin;
-		}
 
 		GEM_BUG_ON(vma->node.start < start);
 		GEM_BUG_ON(vma->node.start + vma->node.size > end);
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v3 01/38] drm/i915: Use the MRU stack search after evicting

[Chris Wilson]

On Fri, Dec 16, 2016 at 07:25:13PM +0000, Chris Wilson wrote:
> When we evict from the GTT to make room for an object, the hole we
> create is put onto the MRU stack inside the drm_mm range manager. On the
> next search pass, we can speed up a PIN_HIGH allocation by referencing
> that stack for the new hole.
> 
> v2: Pull together the 3 identical implements (ahem, a couple were
> outdated) into a common routine for allocating a node and evicting as
> necessary.

On noes, I realized that I have used the correct spelling of colour.
Needs a s/colour/color/ for consistency.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 02/38] drm: Use drm_mm_nodes() as shorthand for the list of nodes under struct drm_mm

[Chris Wilson]

Fairly commonly we want to inspect the node list on the struct drm_mm,
which is buried within an embedded node. Bring it to the surface with a
bit of syntatic sugar.

Note this was intended to be split from commit ad579002c8ec ("drm: Add
drm_mm_for_each_node_safe()") before being applied, but my timing sucks.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/drm_mm.c |  8 ++++----
 include/drm/drm_mm.h     | 18 +++++++++++++++---
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 7573661302a4..4257c86cc305 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -138,7 +138,7 @@ static void show_leaks(struct drm_mm *mm)
 	if (!buf)
 		return;
 
-	list_for_each_entry(node, __drm_mm_nodes(mm), node_list) {
+	list_for_each_entry(node, drm_mm_nodes(mm), node_list) {
 		struct stack_trace trace = {
 			.entries = entries,
 			.max_entries = STACKDEPTH
@@ -320,7 +320,7 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 		if (hole->start < end)
 			return -ENOSPC;
 	} else {
-		hole = list_entry(__drm_mm_nodes(mm), typeof(*hole), node_list);
+		hole = list_entry(drm_mm_nodes(mm), typeof(*hole), node_list);
 	}
 
 	hole = list_last_entry(&hole->node_list, typeof(*hole), node_list);
@@ -883,7 +883,7 @@ EXPORT_SYMBOL(drm_mm_scan_remove_block);
  */
 bool drm_mm_clean(const struct drm_mm *mm)
 {
-	const struct list_head *head = __drm_mm_nodes(mm);
+	const struct list_head *head = drm_mm_nodes(mm);
 
 	return (head->next->next == head);
 }
@@ -929,7 +929,7 @@ EXPORT_SYMBOL(drm_mm_init);
  */
 void drm_mm_takedown(struct drm_mm *mm)
 {
-	if (WARN(!list_empty(__drm_mm_nodes(mm)),
+	if (WARN(!list_empty(drm_mm_nodes(mm)),
 		 "Memory manager not clean during takedown.\n"))
 		show_leaks(mm);
 
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 5c7f15875b6a..f6a68ed5ecaf 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -180,7 +180,19 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
 	return __drm_mm_hole_node_end(hole_node);
 }
 
-#define __drm_mm_nodes(mm) (&(mm)->head_node.node_list)
+/**
+ * drm_mm_nodes - list of nodes under the drm_mm range manager
+ * @mm: the struct drm_mm range manger
+ *
+ * As the drm_mm range manager hides its node_list deep with its
+ * structure, extracting it looks painful and repetitive. This is
+ * not expected to be used outside of the drm_mm_for_each_node()
+ * macros and similar internal functions.
+ *
+ * Returns:
+ * The node list, may be empty.
+ */
+#define drm_mm_nodes(mm) (&(mm)->head_node.node_list)
 
 /**
  * drm_mm_for_each_node - iterator to walk over all allocated nodes
@@ -191,7 +203,7 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  * with list_for_each, so not save against removal of elements.
  */
 #define drm_mm_for_each_node(entry, mm) \
-	list_for_each_entry(entry, __drm_mm_nodes(mm), node_list)
+	list_for_each_entry(entry, drm_mm_nodes(mm), node_list)
 
 /**
  * drm_mm_for_each_node_safe - iterator to walk over all allocated nodes
@@ -203,7 +215,7 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  * with list_for_each_safe, so save against removal of elements.
  */
 #define drm_mm_for_each_node_safe(entry, next, mm) \
-	list_for_each_entry_safe(entry, next, __drm_mm_nodes(mm), node_list)
+	list_for_each_entry_safe(entry, next, drm_mm_nodes(mm), node_list)
 
 #define __drm_mm_for_each_hole(entry, mm, hole_start, hole_end, backwards) \
 	for (entry = list_entry((backwards) ? (mm)->hole_stack.prev : (mm)->hole_stack.next, struct drm_mm_node, hole_stack); \
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 03/38] drm: Compile time enabling for asserts in drm_mm

[Chris Wilson]

Use CONFIG_DRM_DEBUG_MM to conditionally enable the internal and
validation checking using BUG_ON. Ideally these paths should all be
exercised by CI selftests (with the asserts enabled).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 45 +++++++++++++++++++++++----------------------
 include/drm/drm_mm.h     |  8 +++++++-
 2 files changed, 30 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 4257c86cc305..9756542abe4c 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -237,7 +237,7 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 	u64 adj_start = hole_start;
 	u64 adj_end = hole_end;
 
-	BUG_ON(node->allocated);
+	DRM_MM_BUG_ON(node->allocated);
 
 	if (mm->color_adjust)
 		mm->color_adjust(hole_node, color, &adj_start, &adj_end);
@@ -258,8 +258,8 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 		}
 	}
 
-	BUG_ON(adj_start < hole_start);
-	BUG_ON(adj_end > hole_end);
+	DRM_MM_BUG_ON(adj_start < hole_start);
+	DRM_MM_BUG_ON(adj_end > hole_end);
 
 	if (adj_start == hole_start) {
 		hole_node->hole_follows = 0;
@@ -276,7 +276,7 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 
 	drm_mm_interval_tree_add_node(hole_node, node);
 
-	BUG_ON(node->start + node->size > adj_end);
+	DRM_MM_BUG_ON(node->start + node->size > adj_end);
 
 	node->hole_follows = 0;
 	if (__drm_mm_hole_node_start(node) < hole_end) {
@@ -409,7 +409,7 @@ static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
 	u64 adj_start = hole_start;
 	u64 adj_end = hole_end;
 
-	BUG_ON(!hole_node->hole_follows || node->allocated);
+	DRM_MM_BUG_ON(!hole_node->hole_follows || node->allocated);
 
 	if (adj_start < start)
 		adj_start = start;
@@ -450,10 +450,10 @@ static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
 
 	drm_mm_interval_tree_add_node(hole_node, node);
 
-	BUG_ON(node->start < start);
-	BUG_ON(node->start < adj_start);
-	BUG_ON(node->start + node->size > adj_end);
-	BUG_ON(node->start + node->size > end);
+	DRM_MM_BUG_ON(node->start < start);
+	DRM_MM_BUG_ON(node->start < adj_start);
+	DRM_MM_BUG_ON(node->start + node->size > adj_end);
+	DRM_MM_BUG_ON(node->start + node->size > end);
 
 	node->hole_follows = 0;
 	if (__drm_mm_hole_node_start(node) < hole_end) {
@@ -519,22 +519,21 @@ void drm_mm_remove_node(struct drm_mm_node *node)
 	struct drm_mm *mm = node->mm;
 	struct drm_mm_node *prev_node;
 
-	if (WARN_ON(!node->allocated))
-		return;
-
-	BUG_ON(node->scanned_block || node->scanned_prev_free
-				   || node->scanned_next_free);
+	DRM_MM_BUG_ON(!node->allocated);
+	DRM_MM_BUG_ON(node->scanned_block ||
+		      node->scanned_prev_free ||
+		      node->scanned_next_free);
 
 	prev_node =
 	    list_entry(node->node_list.prev, struct drm_mm_node, node_list);
 
 	if (node->hole_follows) {
-		BUG_ON(__drm_mm_hole_node_start(node) ==
-		       __drm_mm_hole_node_end(node));
+		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) ==
+			      __drm_mm_hole_node_end(node));
 		list_del(&node->hole_stack);
 	} else
-		BUG_ON(__drm_mm_hole_node_start(node) !=
-		       __drm_mm_hole_node_end(node));
+		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) !=
+			      __drm_mm_hole_node_end(node));
 
 
 	if (!prev_node->hole_follows) {
@@ -578,7 +577,7 @@ static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
 	u64 adj_end;
 	u64 best_size;
 
-	BUG_ON(mm->scanned_blocks);
+	DRM_MM_BUG_ON(mm->scanned_blocks);
 
 	best = NULL;
 	best_size = ~0UL;
@@ -622,7 +621,7 @@ static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_
 	u64 adj_end;
 	u64 best_size;
 
-	BUG_ON(mm->scanned_blocks);
+	DRM_MM_BUG_ON(mm->scanned_blocks);
 
 	best = NULL;
 	best_size = ~0UL;
@@ -668,6 +667,8 @@ static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_
  */
 void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new)
 {
+	DRM_MM_BUG_ON(!old->allocated);
+
 	list_replace(&old->node_list, &new->node_list);
 	list_replace(&old->hole_stack, &new->hole_stack);
 	rb_replace_node(&old->rb, &new->rb, &old->mm->interval_tree);
@@ -798,7 +799,7 @@ bool drm_mm_scan_add_block(struct drm_mm_node *node)
 
 	mm->scanned_blocks++;
 
-	BUG_ON(node->scanned_block);
+	DRM_MM_BUG_ON(node->scanned_block);
 	node->scanned_block = 1;
 
 	prev_node = list_entry(node->node_list.prev, struct drm_mm_node,
@@ -859,7 +860,7 @@ bool drm_mm_scan_remove_block(struct drm_mm_node *node)
 
 	mm->scanned_blocks--;
 
-	BUG_ON(!node->scanned_block);
+	DRM_MM_BUG_ON(!node->scanned_block);
 	node->scanned_block = 0;
 
 	prev_node = list_entry(node->node_list.prev, struct drm_mm_node,
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index f6a68ed5ecaf..525543019896 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -48,6 +48,12 @@
 #include <linux/stackdepot.h>
 #endif
 
+#ifdef CONFIG_DRM_DEBUG_MM
+#define DRM_MM_BUG_ON(expr) BUG_ON(expr)
+#else
+#define DRM_MM_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr)
+#endif
+
 enum drm_mm_search_flags {
 	DRM_MM_SEARCH_DEFAULT =		0,
 	DRM_MM_SEARCH_BEST =		1 << 0,
@@ -155,7 +161,7 @@ static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
  */
 static inline u64 drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
 {
-	BUG_ON(!hole_node->hole_follows);
+	DRM_MM_BUG_ON(!hole_node->hole_follows);
 	return __drm_mm_hole_node_start(hole_node);
 }
 
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 04/38] lib: Add a simple prime number generator

[Chris Wilson]

Prime numbers are interesting for testing components that use multiplies
and divides, such as testing DRM's struct drm_mm alignment computations.

v2: Move to lib/, add selftest
v3: Fix initial constants (exclude 0/1 from being primes)
v4: More RCU markup to keep 0day/sparse happy

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Lukas Wunner <lukas@wunner.de>
---
 include/linux/prime_numbers.h |  13 +++
 lib/Kconfig                   |   7 ++
 lib/Makefile                  |   2 +
 lib/prime_numbers.c           | 238 ++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 260 insertions(+)
 create mode 100644 include/linux/prime_numbers.h
 create mode 100644 lib/prime_numbers.c

diff --git a/include/linux/prime_numbers.h b/include/linux/prime_numbers.h
new file mode 100644
index 000000000000..877f6acbd0b6
--- /dev/null
+++ b/include/linux/prime_numbers.h
@@ -0,0 +1,13 @@
+#ifndef __LINUX_PRIME_NUMBERS_H
+#define __LINUX_PRIME_NUMBERS_H
+
+#include <linux/types.h>
+
+bool is_prime_number(unsigned long x);
+unsigned long next_prime_number(unsigned long x);
+
+/* A useful white-lie here is that 1 is prime. */
+#define for_each_prime_number(prime, max) \
+	for (prime = 1;	prime < (max); prime = next_prime_number(prime))
+
+#endif /* !__LINUX_PRIME_NUMBERS_H */
diff --git a/lib/Kconfig b/lib/Kconfig
index 260a80e313b9..1788a1f50d28 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -550,4 +550,11 @@ config STACKDEPOT
 config SBITMAP
 	bool
 
+config PRIME_NUMBERS
+	tristate "Prime number generator"
+	default n
+	help
+	  Provides a helper module to generate prime numbers. Useful for writing
+	  test code, especially when checking multiplication and divison.
+
 endmenu
diff --git a/lib/Makefile b/lib/Makefile
index 50144a3aeebd..c664143fd917 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -197,6 +197,8 @@ obj-$(CONFIG_ASN1) += asn1_decoder.o
 
 obj-$(CONFIG_FONT_SUPPORT) += fonts/
 
+obj-$(CONFIG_PRIME_NUMBERS) += prime_numbers.o
+
 hostprogs-y	:= gen_crc32table
 clean-files	:= crc32table.h
 
diff --git a/lib/prime_numbers.c b/lib/prime_numbers.c
new file mode 100644
index 000000000000..ec2959520653
--- /dev/null
+++ b/lib/prime_numbers.c
@@ -0,0 +1,238 @@
+#define pr_fmt(fmt) "prime numbers: " fmt "\n"
+
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/prime_numbers.h>
+#include <linux/slab.h>
+
+struct primes {
+	struct rcu_head rcu;
+	unsigned long last, sz;
+	unsigned long primes[];
+};
+
+#if BITS_PER_LONG == 64
+static const struct primes small_primes = {
+	.last = 61,
+	.sz = 64,
+	.primes = { 0x28208a20a08a28acUL }
+};
+#elif BITS_PER_LONG == 32
+static const struct primes small_primes = {
+	.last = 31,
+	.sz = 32,
+	.primes = { 0xa08a28acUL }
+};
+#else
+#error "unhandled BITS_PER_LONG"
+#endif
+
+static DEFINE_MUTEX(lock);
+static const struct primes __rcu *primes = RCU_INITIALIZER(&small_primes);
+
+static unsigned long selftest_max;
+
+static bool slow_is_prime_number(unsigned long x)
+{
+	unsigned long y = int_sqrt(x);
+
+	while (y > 1) {
+		if ((x % y) == 0)
+			break;
+		y--;
+	}
+
+	return y == 1;
+}
+
+static unsigned long slow_next_prime_number(unsigned long x)
+{
+	for (;;) {
+		if (slow_is_prime_number(++x))
+			return x;
+	}
+}
+
+static unsigned long mark_multiples(unsigned long x,
+				    unsigned long *p,
+				    unsigned long start,
+				    unsigned long end)
+{
+	unsigned long m;
+
+	m = 2 * x;
+	if (m < start)
+		m = roundup(start, x);
+
+	while (m < end) {
+		__clear_bit(m, p);
+		m += x;
+	}
+
+	return x;
+}
+
+static const struct primes *expand_to_next(unsigned long x)
+{
+	const struct primes *p;
+	struct primes *new;
+	unsigned long sz, y;
+
+	rcu_read_unlock();
+
+	/* Betrand's Theorem states:
+	 * 	For all n > 1, there exists a prime p: n < p <= 2*n.
+	 */
+	sz = 2 * x + 1;
+	if (sz < x)
+		return NULL;
+
+	sz = round_up(sz, BITS_PER_LONG);
+	new = kmalloc(sizeof(*new) + sz / sizeof(long), GFP_KERNEL);
+	if (!new)
+		return NULL;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (x < p->last) {
+		kfree(new);
+		goto relock;
+	}
+
+	/* Where memory permits, track the primes using the
+	 * Sieve of Eratosthenes.
+	 */
+	memcpy(new->primes, p->primes, p->sz / BITS_PER_LONG * sizeof(long));
+	memset(new->primes + p->sz / BITS_PER_LONG,
+	       0xff, (sz - p->sz) / BITS_PER_LONG * sizeof(long));
+	for (y = 2UL; y < sz; y = find_next_bit(new->primes, sz, y + 1))
+		new->last = mark_multiples(y, new->primes, p->sz, sz);
+	new->sz = sz;
+
+	rcu_assign_pointer(primes, new);
+	if (p != &small_primes)
+		kfree_rcu((struct primes *)p, rcu);
+	p = new;
+
+relock:
+	rcu_read_lock();
+	mutex_unlock(&lock);
+	return p;
+}
+
+static const struct primes *get_primes(unsigned long x)
+{
+	const struct primes *p;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	if (!p || x >= p->last)
+		p = expand_to_next(x);
+
+	/* returns under RCU iff p != NULL */
+	return p;
+}
+
+unsigned long next_prime_number(unsigned long x)
+{
+	const struct primes *p;
+
+	p = get_primes(x);
+	if (unlikely(!p))
+		return slow_next_prime_number(x);
+
+	x = find_next_bit(p->primes, p->last, x + 1);
+	rcu_read_unlock();
+
+	return x;
+}
+EXPORT_SYMBOL(next_prime_number);
+
+bool is_prime_number(unsigned long x)
+{
+	const struct primes *p;
+	bool result;
+
+	p = get_primes(x);
+	if (unlikely(!p))
+		return slow_is_prime_number(x);
+
+	result = test_bit(x, p->primes);
+	rcu_read_unlock();
+
+	return result;
+}
+EXPORT_SYMBOL(is_prime_number);
+
+static void dump_primes(void)
+{
+	const struct primes *p;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...x%lx}",
+		p->last, p->sz, p->primes[p->sz / BITS_PER_LONG - 1]);
+	rcu_read_unlock();
+}
+
+static int selftest(unsigned long max)
+{
+	unsigned long x, last;
+
+	if (!max)
+		return 0;
+
+	for (last = 0, x = 2; x < max; x++) {
+		bool slow = slow_is_prime_number(x);
+		bool fast = is_prime_number(x);
+
+		if (slow != fast) {
+			pr_err("inconsistent result for is-prime(%lu): slow=%s, fast=%s!",
+			       x, slow ? "yes" : "no", fast ? "yes" : "no");
+			goto err;
+		}
+
+		if (!slow)
+			continue;
+
+		if (next_prime_number(last) != x) {
+			pr_err("incorrect result for next-prime(%lu): expected %lu, got %lu",
+			       last, x, next_prime_number(last));
+			goto err;
+		}
+		last = x;
+	}
+
+	pr_info("selftest(%lu) passed, last prime was %lu", x, last);
+	return 0;
+
+err:
+	dump_primes();
+	return -EINVAL;
+}
+
+static int __init primes_init(void)
+{
+	return selftest(selftest_max);
+}
+
+static void __exit primes_exit(void)
+{
+	const struct primes *p;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (p != &small_primes) {
+		kfree_rcu((struct primes *)p, rcu);
+		rcu_assign_pointer(p, &small_primes);
+	}
+	mutex_unlock(&lock);
+}
+
+module_init(primes_init);
+module_exit(primes_exit);
+
+module_param_named(selftest, selftest_max, ulong, 0400);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v3 04/38] lib: Add a simple prime number generator

[Chris Wilson]

On Fri, Dec 16, 2016 at 07:25:16PM +0000, Chris Wilson wrote:
> +static void __exit primes_exit(void)
> +{
> +	const struct primes *p;
> +
> +	mutex_lock(&lock);
> +	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
> +	if (p != &small_primes) {
> +		kfree_rcu((struct primes *)p, rcu);
> +		rcu_assign_pointer(p, &small_primes);

Too much sparse appleasing, too little thinking. It's only the module
shutdown path, but we might as well be correct: s/p/primes here.
-Chris
> 

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH] lib: Add a simple prime number generator

[Chris Wilson]

Prime numbers are interesting for testing components that use multiplies
and divides, such as testing DRM's struct drm_mm alignment computations.

v2: Move to lib/, add selftest
v3: Fix initial constants (exclude 0/1 from being primes)
v4: More RCU markup to keep 0day/sparse happy
v5: Fix RCU unwind on module exit, add to kselftests

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Lukas Wunner <lukas@wunner.de>
---
 include/linux/prime_numbers.h                |  13 ++
 lib/Kconfig                                  |   7 +
 lib/Makefile                                 |   2 +
 lib/prime_numbers.c                          | 238 +++++++++++++++++++++++++++
 tools/testing/selftests/lib/prime_numbers.sh |  15 ++
 5 files changed, 275 insertions(+)
 create mode 100644 include/linux/prime_numbers.h
 create mode 100644 lib/prime_numbers.c
 create mode 100755 tools/testing/selftests/lib/prime_numbers.sh

diff --git a/include/linux/prime_numbers.h b/include/linux/prime_numbers.h
new file mode 100644
index 000000000000..877f6acbd0b6
--- /dev/null
+++ b/include/linux/prime_numbers.h
@@ -0,0 +1,13 @@
+#ifndef __LINUX_PRIME_NUMBERS_H
+#define __LINUX_PRIME_NUMBERS_H
+
+#include <linux/types.h>
+
+bool is_prime_number(unsigned long x);
+unsigned long next_prime_number(unsigned long x);
+
+/* A useful white-lie here is that 1 is prime. */
+#define for_each_prime_number(prime, max) \
+	for (prime = 1;	prime < (max); prime = next_prime_number(prime))
+
+#endif /* !__LINUX_PRIME_NUMBERS_H */
diff --git a/lib/Kconfig b/lib/Kconfig
index 260a80e313b9..1788a1f50d28 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -550,4 +550,11 @@ config STACKDEPOT
 config SBITMAP
 	bool
 
+config PRIME_NUMBERS
+	tristate "Prime number generator"
+	default n
+	help
+	  Provides a helper module to generate prime numbers. Useful for writing
+	  test code, especially when checking multiplication and divison.
+
 endmenu
diff --git a/lib/Makefile b/lib/Makefile
index 50144a3aeebd..c664143fd917 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -197,6 +197,8 @@ obj-$(CONFIG_ASN1) += asn1_decoder.o
 
 obj-$(CONFIG_FONT_SUPPORT) += fonts/
 
+obj-$(CONFIG_PRIME_NUMBERS) += prime_numbers.o
+
 hostprogs-y	:= gen_crc32table
 clean-files	:= crc32table.h
 
diff --git a/lib/prime_numbers.c b/lib/prime_numbers.c
new file mode 100644
index 000000000000..f5a704033a3c
--- /dev/null
+++ b/lib/prime_numbers.c
@@ -0,0 +1,238 @@
+#define pr_fmt(fmt) "prime numbers: " fmt "\n"
+
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/prime_numbers.h>
+#include <linux/slab.h>
+
+struct primes {
+	struct rcu_head rcu;
+	unsigned long last, sz;
+	unsigned long primes[];
+};
+
+#if BITS_PER_LONG == 64
+static const struct primes small_primes = {
+	.last = 61,
+	.sz = 64,
+	.primes = { 0x28208a20a08a28acUL }
+};
+#elif BITS_PER_LONG == 32
+static const struct primes small_primes = {
+	.last = 31,
+	.sz = 32,
+	.primes = { 0xa08a28acUL }
+};
+#else
+#error "unhandled BITS_PER_LONG"
+#endif
+
+static DEFINE_MUTEX(lock);
+static const struct primes __rcu *primes = RCU_INITIALIZER(&small_primes);
+
+static unsigned long selftest_max;
+
+static bool slow_is_prime_number(unsigned long x)
+{
+	unsigned long y = int_sqrt(x);
+
+	while (y > 1) {
+		if ((x % y) == 0)
+			break;
+		y--;
+	}
+
+	return y == 1;
+}
+
+static unsigned long slow_next_prime_number(unsigned long x)
+{
+	for (;;) {
+		if (slow_is_prime_number(++x))
+			return x;
+	}
+}
+
+static unsigned long mark_multiples(unsigned long x,
+				    unsigned long *p,
+				    unsigned long start,
+				    unsigned long end)
+{
+	unsigned long m;
+
+	m = 2 * x;
+	if (m < start)
+		m = roundup(start, x);
+
+	while (m < end) {
+		__clear_bit(m, p);
+		m += x;
+	}
+
+	return x;
+}
+
+static const struct primes *expand_to_next(unsigned long x)
+{
+	const struct primes *p;
+	struct primes *new;
+	unsigned long sz, y;
+
+	rcu_read_unlock();
+
+	/* Betrand's Theorem states:
+	 * 	For all n > 1, there exists a prime p: n < p <= 2*n.
+	 */
+	sz = 2 * x + 1;
+	if (sz < x)
+		return NULL;
+
+	sz = round_up(sz, BITS_PER_LONG);
+	new = kmalloc(sizeof(*new) + sz / sizeof(long), GFP_KERNEL);
+	if (!new)
+		return NULL;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (x < p->last) {
+		kfree(new);
+		goto relock;
+	}
+
+	/* Where memory permits, track the primes using the
+	 * Sieve of Eratosthenes.
+	 */
+	memcpy(new->primes, p->primes, p->sz / BITS_PER_LONG * sizeof(long));
+	memset(new->primes + p->sz / BITS_PER_LONG,
+	       0xff, (sz - p->sz) / BITS_PER_LONG * sizeof(long));
+	for (y = 2UL; y < sz; y = find_next_bit(new->primes, sz, y + 1))
+		new->last = mark_multiples(y, new->primes, p->sz, sz);
+	new->sz = sz;
+
+	rcu_assign_pointer(primes, new);
+	if (p != &small_primes)
+		kfree_rcu((struct primes *)p, rcu);
+	p = new;
+
+relock:
+	rcu_read_lock();
+	mutex_unlock(&lock);
+	return p;
+}
+
+static const struct primes *get_primes(unsigned long x)
+{
+	const struct primes *p;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	if (!p || x >= p->last)
+		p = expand_to_next(x);
+
+	/* returns under RCU iff p != NULL */
+	return p;
+}
+
+unsigned long next_prime_number(unsigned long x)
+{
+	const struct primes *p;
+
+	p = get_primes(x);
+	if (unlikely(!p))
+		return slow_next_prime_number(x);
+
+	x = find_next_bit(p->primes, p->last, x + 1);
+	rcu_read_unlock();
+
+	return x;
+}
+EXPORT_SYMBOL(next_prime_number);
+
+bool is_prime_number(unsigned long x)
+{
+	const struct primes *p;
+	bool result;
+
+	p = get_primes(x);
+	if (unlikely(!p))
+		return slow_is_prime_number(x);
+
+	result = test_bit(x, p->primes);
+	rcu_read_unlock();
+
+	return result;
+}
+EXPORT_SYMBOL(is_prime_number);
+
+static void dump_primes(void)
+{
+	const struct primes *p;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...x%lx}",
+		p->last, p->sz, p->primes[p->sz / BITS_PER_LONG - 1]);
+	rcu_read_unlock();
+}
+
+static int selftest(unsigned long max)
+{
+	unsigned long x, last;
+
+	if (!max)
+		return 0;
+
+	for (last = 0, x = 2; x < max; x++) {
+		bool slow = slow_is_prime_number(x);
+		bool fast = is_prime_number(x);
+
+		if (slow != fast) {
+			pr_err("inconsistent result for is-prime(%lu): slow=%s, fast=%s!",
+			       x, slow ? "yes" : "no", fast ? "yes" : "no");
+			goto err;
+		}
+
+		if (!slow)
+			continue;
+
+		if (next_prime_number(last) != x) {
+			pr_err("incorrect result for next-prime(%lu): expected %lu, got %lu",
+			       last, x, next_prime_number(last));
+			goto err;
+		}
+		last = x;
+	}
+
+	pr_info("selftest(%lu) passed, last prime was %lu", x, last);
+	return 0;
+
+err:
+	dump_primes();
+	return -EINVAL;
+}
+
+static int __init primes_init(void)
+{
+	return selftest(selftest_max);
+}
+
+static void __exit primes_exit(void)
+{
+	const struct primes *p;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (p != &small_primes) {
+		rcu_assign_pointer(primes, &small_primes);
+		kfree_rcu((struct primes *)p, rcu);
+	}
+	mutex_unlock(&lock);
+}
+
+module_init(primes_init);
+module_exit(primes_exit);
+
+module_param_named(selftest, selftest_max, ulong, 0400);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
diff --git a/tools/testing/selftests/lib/prime_numbers.sh b/tools/testing/selftests/lib/prime_numbers.sh
new file mode 100755
index 000000000000..da4cbcd766f5
--- /dev/null
+++ b/tools/testing/selftests/lib/prime_numbers.sh
@@ -0,0 +1,15 @@
+#!/bin/sh
+# Checks fast/slow prime_number generation for inconsistencies
+
+if ! /sbin/modprobe -q -r prime_numbers; then
+	echo "prime_numbers: [SKIP]"
+	exit 77
+fi
+
+if /sbin/modprobe -q prime_numbers selftest=65536; then
+	/sbin/modprobe -q -r prime_numbers
+	echo "prime_numbers: ok"
+else
+	echo "prime_numbers: [FAIL]"
+	exit 1
+fi
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v6] lib: Add a simple prime number generator

[Chris Wilson]

Prime numbers are interesting for testing components that use multiplies
and divides, such as testing DRM's struct drm_mm alignment computations.

v2: Move to lib/, add selftest
v3: Fix initial constants (exclude 0/1 from being primes)
v4: More RCU markup to keep 0day/sparse happy
v5: Fix RCU unwind on module exit, add to kselftests
v6: Tidy computation of bitmap size

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Lukas Wunner <lukas@wunner.de>
---
 include/linux/prime_numbers.h                |  23 +++
 lib/Kconfig                                  |   7 +
 lib/Makefile                                 |   2 +
 lib/prime_numbers.c                          | 266 +++++++++++++++++++++++++++
 tools/testing/selftests/lib/prime_numbers.sh |  15 ++
 5 files changed, 313 insertions(+)
 create mode 100644 include/linux/prime_numbers.h
 create mode 100644 lib/prime_numbers.c
 create mode 100755 tools/testing/selftests/lib/prime_numbers.sh

diff --git a/include/linux/prime_numbers.h b/include/linux/prime_numbers.h
new file mode 100644
index 000000000000..6ba642c3f95d
--- /dev/null
+++ b/include/linux/prime_numbers.h
@@ -0,0 +1,23 @@
+#ifndef __LINUX_PRIME_NUMBERS_H
+#define __LINUX_PRIME_NUMBERS_H
+
+#include <linux/types.h>
+
+bool is_prime_number(unsigned long x);
+unsigned long next_prime_number(unsigned long x);
+
+/**
+ * for_each_prime_number - iterate over each prime upto a value
+ * @prime: the current prime number in this iteration
+ * @max: the upper limit
+ *
+ * Starting from 1 (which is only considered prime for convenience
+ * of using for_each_prime_number(), a useful white lie), iterate over each
+ * prime number up to the @max value. On each iteration, @prime is set to the
+ * current prime number. @max should be less than ULONG_MAX to ensure
+ * termination.
+ */
+#define for_each_prime_number(prime, max) \
+	for (prime = 1;	prime <= (max); prime = next_prime_number(prime))
+
+#endif /* !__LINUX_PRIME_NUMBERS_H */
diff --git a/lib/Kconfig b/lib/Kconfig
index 260a80e313b9..1788a1f50d28 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -550,4 +550,11 @@ config STACKDEPOT
 config SBITMAP
 	bool
 
+config PRIME_NUMBERS
+	tristate "Prime number generator"
+	default n
+	help
+	  Provides a helper module to generate prime numbers. Useful for writing
+	  test code, especially when checking multiplication and divison.
+
 endmenu
diff --git a/lib/Makefile b/lib/Makefile
index 50144a3aeebd..c664143fd917 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -197,6 +197,8 @@ obj-$(CONFIG_ASN1) += asn1_decoder.o
 
 obj-$(CONFIG_FONT_SUPPORT) += fonts/
 
+obj-$(CONFIG_PRIME_NUMBERS) += prime_numbers.o
+
 hostprogs-y	:= gen_crc32table
 clean-files	:= crc32table.h
 
diff --git a/lib/prime_numbers.c b/lib/prime_numbers.c
new file mode 100644
index 000000000000..02795aa0e820
--- /dev/null
+++ b/lib/prime_numbers.c
@@ -0,0 +1,266 @@
+#define pr_fmt(fmt) "prime numbers: " fmt "\n"
+
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/prime_numbers.h>
+#include <linux/slab.h>
+
+#define bitmap_size(nbits) (BITS_TO_LONGS(nbits) * sizeof(unsigned long))
+
+struct primes {
+	struct rcu_head rcu;
+	unsigned long last, sz;
+	unsigned long primes[];
+};
+
+#if BITS_PER_LONG == 64
+static const struct primes small_primes = {
+	.last = 61,
+	.sz = 64,
+	.primes = { 0x28208a20a08a28acUL }
+};
+#elif BITS_PER_LONG == 32
+static const struct primes small_primes = {
+	.last = 31,
+	.sz = 32,
+	.primes = { 0xa08a28acUL }
+};
+#else
+#error "unhandled BITS_PER_LONG"
+#endif
+
+static DEFINE_MUTEX(lock);
+static const struct primes __rcu *primes = RCU_INITIALIZER(&small_primes);
+
+static unsigned long selftest_max;
+
+static bool slow_is_prime_number(unsigned long x)
+{
+	unsigned long y = int_sqrt(x);
+
+	while (y > 1) {
+		if ((x % y) == 0)
+			break;
+		y--;
+	}
+
+	return y == 1;
+}
+
+static unsigned long slow_next_prime_number(unsigned long x)
+{
+	while (x < ULONG_MAX && !slow_is_prime_number(++x))
+		;
+
+	return x;
+}
+
+static unsigned long mark_multiples(unsigned long x,
+				    unsigned long *p,
+				    unsigned long start,
+				    unsigned long end)
+{
+	unsigned long m;
+
+	m = 2 * x;
+	if (m < start)
+		m = roundup(start, x);
+
+	while (m < end) {
+		__clear_bit(m, p);
+		m += x;
+	}
+
+	return x;
+}
+
+static const struct primes *expand_to_next_prime(unsigned long x)
+{
+	const struct primes *p;
+	struct primes *new;
+	unsigned long sz, y;
+
+	rcu_read_unlock();
+
+	/* Betrand's Theorem states:
+	 * 	For all n > 1, there exists a prime p: n < p <= 2*n.
+	 */
+	sz = 2 * x + 1;
+	if (sz < x)
+		return NULL;
+
+	sz = round_up(sz, BITS_PER_LONG);
+	new = kmalloc(sizeof(*new) + bitmap_size(sz), GFP_KERNEL);
+	if (!new)
+		return NULL;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (x < p->last) {
+		kfree(new);
+		goto relock;
+	}
+
+	/* Where memory permits, track the primes using the
+	 * Sieve of Eratosthenes.
+	 */
+	bitmap_copy(new->primes, p->primes, p->sz);
+	memset(new->primes + BITS_TO_LONGS(p->sz), -1, bitmap_size(sz - p->sz));
+	for (y = 2UL; y < sz; y = find_next_bit(new->primes, sz, y + 1))
+		new->last = mark_multiples(y, new->primes, p->sz, sz);
+	new->sz = sz;
+
+	BUG_ON(new->last <= x);
+
+	rcu_assign_pointer(primes, new);
+	if (p != &small_primes)
+		kfree_rcu((struct primes *)p, rcu);
+	p = new;
+
+relock:
+	rcu_read_lock();
+	mutex_unlock(&lock);
+	return p;
+}
+
+static const struct primes *get_primes(unsigned long x)
+{
+	const struct primes *p;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	if (!p || x >= p->last)
+		p = expand_to_next_prime(x);
+
+	/* returns under RCU iff p != NULL */
+	return p;
+}
+
+/**
+ * next_prime_number - return the next prime number
+ * @x: the starting point for searching to test
+ *
+ * A prime number is an integer greater than 1 that is only divisible by
+ * itself and 1.  The set of prime numbers is computed using the Sieve of
+ * Eratoshenes (on finding a prime, all multiples of that prime are removed
+ * from the set) enabling a fast lookup of the next prime number larger than
+ * @x. If the seive fails (memory limitation), the search falls back to using
+ * slow trial-divison, up to the value of ULONG_MAX (which is reported as the
+ * final prime as a sentinel).
+ *
+ * Returns: the next prime number larger than @x
+ */
+unsigned long next_prime_number(unsigned long x)
+{
+	const struct primes *p;
+
+	p = get_primes(x);
+	if (unlikely(!p))
+		return slow_next_prime_number(x);
+
+	x = find_next_bit(p->primes, p->last, x + 1);
+	rcu_read_unlock();
+
+	return x;
+}
+EXPORT_SYMBOL(next_prime_number);
+
+/**
+ * is_prime_number - test whether the given number is prime
+ * @x: the number to test
+ *
+ * A prime number is an integer greater than 1 that is only divisible by
+ * itself and 1. Internally a cache of prime numbers is kept (to speed up
+ * searching for sequential primes, see next_prime_number()), but if the number
+ * falls outside of that cache, its primality is tested using trial-divison.
+ *
+ * Returns: true if @x is prime, false for composite numbers.
+ */
+bool is_prime_number(unsigned long x)
+{
+	const struct primes *p;
+	bool result;
+
+	p = get_primes(x);
+	if (unlikely(!p))
+		return slow_is_prime_number(x);
+
+	result = test_bit(x, p->primes);
+	rcu_read_unlock();
+
+	return result;
+}
+EXPORT_SYMBOL(is_prime_number);
+
+static void dump_primes(void)
+{
+	const struct primes *p;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...x%lx}",
+		p->last, p->sz, p->primes[BITS_TO_LONGS(p->sz) - 1]);
+	rcu_read_unlock();
+}
+
+static int selftest(unsigned long max)
+{
+	unsigned long x, last;
+
+	if (!max)
+		return 0;
+
+	for (last = 0, x = 2; x < max; x++) {
+		bool slow = slow_is_prime_number(x);
+		bool fast = is_prime_number(x);
+
+		if (slow != fast) {
+			pr_err("inconsistent result for is-prime(%lu): slow=%s, fast=%s!",
+			       x, slow ? "yes" : "no", fast ? "yes" : "no");
+			goto err;
+		}
+
+		if (!slow)
+			continue;
+
+		if (next_prime_number(last) != x) {
+			pr_err("incorrect result for next-prime(%lu): expected %lu, got %lu",
+			       last, x, next_prime_number(last));
+			goto err;
+		}
+		last = x;
+	}
+
+	pr_info("selftest(%lu) passed, last prime was %lu", x, last);
+	return 0;
+
+err:
+	dump_primes();
+	return -EINVAL;
+}
+
+static int __init primes_init(void)
+{
+	return selftest(selftest_max);
+}
+
+static void __exit primes_exit(void)
+{
+	const struct primes *p;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (p != &small_primes) {
+		rcu_assign_pointer(primes, &small_primes);
+		kfree_rcu((struct primes *)p, rcu);
+	}
+	mutex_unlock(&lock);
+}
+
+module_init(primes_init);
+module_exit(primes_exit);
+
+module_param_named(selftest, selftest_max, ulong, 0400);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
diff --git a/tools/testing/selftests/lib/prime_numbers.sh b/tools/testing/selftests/lib/prime_numbers.sh
new file mode 100755
index 000000000000..da4cbcd766f5
--- /dev/null
+++ b/tools/testing/selftests/lib/prime_numbers.sh
@@ -0,0 +1,15 @@
+#!/bin/sh
+# Checks fast/slow prime_number generation for inconsistencies
+
+if ! /sbin/modprobe -q -r prime_numbers; then
+	echo "prime_numbers: [SKIP]"
+	exit 77
+fi
+
+if /sbin/modprobe -q prime_numbers selftest=65536; then
+	/sbin/modprobe -q -r prime_numbers
+	echo "prime_numbers: ok"
+else
+	echo "prime_numbers: [FAIL]"
+	exit 1
+fi
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 05/38] drm: Add a simple generator of random permutations

[Chris Wilson]

When testing, we want a random but yet reproducible order in which to
process elements. Here we create an array which is a random (using the
Tausworthe PRNG) permutation of the order in which to execute.

Note these are simple helpers intended to be merged upstream in lib/

v2: Tidier code by David Herrmann
v3: Add reminder that this code is intended to be temporary, with at
least the bulk of the prandom changes going to lib/

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: David Herrmann <dh.herrmann@gmail.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/Kconfig          |  4 ++++
 drivers/gpu/drm/Makefile         |  1 +
 drivers/gpu/drm/lib/drm_random.c | 41 ++++++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/lib/drm_random.h | 25 ++++++++++++++++++++++++
 4 files changed, 71 insertions(+)
 create mode 100644 drivers/gpu/drm/lib/drm_random.c
 create mode 100644 drivers/gpu/drm/lib/drm_random.h

diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index ebfe8404c25f..45a1c7468e88 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -321,3 +321,7 @@ config DRM_SAVAGE
 	  chipset. If M is selected the module will be called savage.
 
 endif # DRM_LEGACY
+
+config DRM_LIB_RANDOM
+	bool
+	default n
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index b9ae4280de9d..6bb416360ae4 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -18,6 +18,7 @@ drm-y       :=	drm_auth.o drm_bufs.o drm_cache.o \
 		drm_plane.o drm_color_mgmt.o drm_print.o \
 		drm_dumb_buffers.o drm_mode_config.o
 
+drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o
 drm-$(CONFIG_COMPAT) += drm_ioc32.o
 drm-$(CONFIG_DRM_GEM_CMA_HELPER) += drm_gem_cma_helper.o
 drm-$(CONFIG_PCI) += ati_pcigart.o
diff --git a/drivers/gpu/drm/lib/drm_random.c b/drivers/gpu/drm/lib/drm_random.c
new file mode 100644
index 000000000000..7b12a68c3b54
--- /dev/null
+++ b/drivers/gpu/drm/lib/drm_random.c
@@ -0,0 +1,41 @@
+#include <linux/bitops.h>
+#include <linux/kernel.h>
+#include <linux/random.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+
+#include "drm_random.h"
+
+static inline u32 drm_prandom_u32_max_state(u32 ep_ro, struct rnd_state *state)
+{
+	return upper_32_bits((u64)prandom_u32_state(state) * ep_ro);
+}
+
+void drm_random_reorder(unsigned int *order, unsigned int count,
+			struct rnd_state *state)
+{
+	unsigned int i, j;
+
+	for (i = 0; i < count; ++i) {
+		BUILD_BUG_ON(sizeof(unsigned int) > sizeof(u32));
+		j = drm_prandom_u32_max_state(count, state);
+		swap(order[i], order[j]);
+	}
+}
+EXPORT_SYMBOL(drm_random_reorder);
+
+unsigned int *drm_random_order(unsigned int count, struct rnd_state *state)
+{
+	unsigned int *order, i;
+
+	order = kmalloc_array(count, sizeof(*order), GFP_TEMPORARY);
+	if (!order)
+		return order;
+
+	for (i = 0; i < count; i++)
+		order[i] = i;
+
+	drm_random_reorder(order, count, state);
+	return order;
+}
+EXPORT_SYMBOL(drm_random_order);
diff --git a/drivers/gpu/drm/lib/drm_random.h b/drivers/gpu/drm/lib/drm_random.h
new file mode 100644
index 000000000000..a78644bea7f9
--- /dev/null
+++ b/drivers/gpu/drm/lib/drm_random.h
@@ -0,0 +1,25 @@
+#ifndef __DRM_RANDOM_H__
+#define __DRM_RANDOM_H__
+
+/* This is a temporary home for a couple of utility functions that should
+ * be transposed to lib/ at the earliest convenience.
+ */
+
+#include <linux/random.h>
+
+#define DRM_RND_STATE_INITIALIZER(seed__) ({				\
+	struct rnd_state state__;					\
+	prandom_seed_state(&state__, (seed__));				\
+	state__;							\
+})
+
+#define DRM_RND_STATE(name__, seed__) \
+	struct rnd_state name__ = DRM_RND_STATE_INITIALIZER(seed__)
+
+unsigned int *drm_random_order(unsigned int count,
+			       struct rnd_state *state);
+void drm_random_reorder(unsigned int *order,
+			unsigned int count,
+			struct rnd_state *state);
+
+#endif /* !__DRM_RANDOM_H__ */
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 06/38] drm: Add some kselftests for the DRM range manager (struct drm_mm)

[Chris Wilson]

First we introduce a smattering of infrastructure for writing selftests.
The idea is that we have a test module that exercises a particular
portion of the exported API, and that module provides a set of tests
that can either be run as an ensemble via kselftest or individually via
an igt harness (in this case igt/drm_mm). To accommodate selecting
individual tests, we export a boolean parameter to control selection of
each test - that is hidden inside a bunch of reusable boilerplate macros
to keep writing the tests simple.

v2: Choose a random random_seed unless one is specified by the user.
v3: More parameters to control max_iterations and max_prime of the
tests.

Testcase: igt/drm_mm
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/Kconfig                       |  15 ++++
 drivers/gpu/drm/Makefile                      |   1 +
 drivers/gpu/drm/selftests/drm_mm_selftests.h  |   8 ++
 drivers/gpu/drm/selftests/drm_selftest.c      | 109 ++++++++++++++++++++++++++
 drivers/gpu/drm/selftests/drm_selftest.h      |  41 ++++++++++
 drivers/gpu/drm/selftests/test-drm_mm.c       |  58 ++++++++++++++
 tools/testing/selftests/drivers/gpu/drm_mm.sh |  15 ++++
 7 files changed, 247 insertions(+)
 create mode 100644 drivers/gpu/drm/selftests/drm_mm_selftests.h
 create mode 100644 drivers/gpu/drm/selftests/drm_selftest.c
 create mode 100644 drivers/gpu/drm/selftests/drm_selftest.h
 create mode 100644 drivers/gpu/drm/selftests/test-drm_mm.c
 create mode 100755 tools/testing/selftests/drivers/gpu/drm_mm.sh

diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index 45a1c7468e88..29146fa83001 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -48,6 +48,21 @@ config DRM_DEBUG_MM
 
 	  If in doubt, say "N".
 
+config DRM_DEBUG_MM_SELFTEST
+	tristate "kselftests for DRM range manager (struct drm_mm)"
+	depends on DRM
+	depends on DEBUG_KERNEL
+	select PRIME_NUMBERS
+	select DRM_LIB_RANDOM
+	default n
+	help
+	  This option provides a kernel module that can be used to test
+	  the DRM range manager (drm_mm) and its API. This option is not
+	  useful for distributions or general kernels, but only for kernel
+	  developers working on DRM and associated drivers.
+
+	  If in doubt, say "N".
+
 config DRM_KMS_HELPER
 	tristate
 	depends on DRM
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 6bb416360ae4..c0d1aed8588b 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -38,6 +38,7 @@ drm_kms_helper-$(CONFIG_DRM_KMS_CMA_HELPER) += drm_fb_cma_helper.o
 drm_kms_helper-$(CONFIG_DRM_DP_AUX_CHARDEV) += drm_dp_aux_dev.o
 
 obj-$(CONFIG_DRM_KMS_HELPER) += drm_kms_helper.o
+obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/test-drm_mm.o
 
 CFLAGS_drm_trace_points.o := -I$(src)
 
diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
new file mode 100644
index 000000000000..1610e0a63a5b
--- /dev/null
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -0,0 +1,8 @@
+/* List each unit test as selftest(name, function)
+ *
+ * The name is used as both an enum and expanded as igt__name to create
+ * a module parameter. It must be unique and legal for a C identifier.
+ *
+ * Tests are executed in order by igt/drm_mm
+ */
+selftest(sanitycheck, igt_sanitycheck) /* keep first (selfcheck for igt) */
diff --git a/drivers/gpu/drm/selftests/drm_selftest.c b/drivers/gpu/drm/selftests/drm_selftest.c
new file mode 100644
index 000000000000..e29ed9faef5b
--- /dev/null
+++ b/drivers/gpu/drm/selftests/drm_selftest.c
@@ -0,0 +1,109 @@
+/*
+ * Copyright © 2016 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include <linux/compiler.h>
+
+#define selftest(name, func) __idx_##name,
+enum {
+#include TESTS
+};
+#undef selftest
+
+#define selftest(n, f) [__idx_##n] = { .name = #n, .func = f },
+static struct drm_selftest {
+	bool enabled;
+	const char *name;
+	int (*func)(void *);
+} selftests[] = {
+#include TESTS
+};
+#undef selftest
+
+/* Embed the line number into the parameter name so that we can order tests */
+#define param(n) __PASTE(igt__, __PASTE(__PASTE(__LINE__, __), n))
+#define selftest_0(n, func, id) \
+module_param_named(id, selftests[__idx_##n].enabled, bool, 0400);
+#define selftest(n, func) selftest_0(n, func, param(n))
+#include TESTS
+#undef selftest
+
+static void set_default_test_all(struct drm_selftest *st, unsigned long count)
+{
+	unsigned long i;
+
+	for (i = 0; i < count; i++)
+		if (st[i].enabled)
+			return;
+
+	for (i = 0; i < count; i++)
+		st[i].enabled = true;
+}
+
+static int run_selftests(struct drm_selftest *st,
+			 unsigned long count,
+			 void *data)
+{
+	int err = 0;
+
+	set_default_test_all(st, count);
+
+	/* Tests are listed in natural order in drm_*_selftests.h */
+	for (; count--; st++) {
+		if (!st->enabled)
+			continue;
+
+		pr_debug("drm: Running %s\n", st->name);
+		err = st->func(data);
+		if (err)
+			break;
+	}
+
+	if (WARN(err > 0 || err == -ENOTTY,
+		 "%s returned %d, conflicting with selftest's magic values!\n",
+		 st->name, err))
+		err = -1;
+
+	rcu_barrier();
+	return err;
+}
+
+static int __maybe_unused
+__drm_subtests(const char *caller,
+	       const struct drm_subtest *st,
+	       int count,
+	       void *data)
+{
+	int err;
+
+	for (; count--; st++) {
+		pr_debug("Running %s/%s\n", caller, st->name);
+		err = st->func(data);
+		if (err) {
+			pr_err("%s: %s failed with error %d\n",
+			       caller, st->name, err);
+			return err;
+		}
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/selftests/drm_selftest.h b/drivers/gpu/drm/selftests/drm_selftest.h
new file mode 100644
index 000000000000..c784ec02ff53
--- /dev/null
+++ b/drivers/gpu/drm/selftests/drm_selftest.h
@@ -0,0 +1,41 @@
+/*
+ * Copyright © 2016 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef __DRM_SELFTEST_H__
+#define __DRM_SELFTEST_H__
+
+struct drm_subtest {
+	int (*func)(void *data);
+	const char *name;
+};
+
+static int __drm_subtests(const char *caller,
+			  const struct drm_subtest *st,
+			  int count,
+			  void *data);
+#define drm_subtests(T, data) \
+	__drm_subtests(__func__, T, ARRAY_SIZE(T), data)
+
+#define SUBTEST(x) { x, #x }
+
+#endif /* __DRM_SELFTEST_H__ */
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
new file mode 100644
index 000000000000..682f5f86f465
--- /dev/null
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -0,0 +1,58 @@
+/*
+ * Test cases for the drm_mm range manager
+ */
+
+#define pr_fmt(fmt) "drm_mm: " fmt
+
+#include <linux/module.h>
+#include <linux/prime_numbers.h>
+#include <linux/slab.h>
+#include <linux/random.h>
+#include <linux/vmalloc.h>
+
+#include <drm/drm_mm.h>
+
+#include "../lib/drm_random.h"
+
+#define TESTS "drm_mm_selftests.h"
+#include "drm_selftest.h"
+
+static unsigned int random_seed;
+static unsigned int max_iterations = 8192;
+static unsigned int max_prime = 128;
+
+static int igt_sanitycheck(void *ignored)
+{
+	pr_info("%s - ok!\n", __func__);
+	return 0;
+}
+
+#include "drm_selftest.c"
+
+static int __init test_drm_mm_init(void)
+{
+	int err;
+
+	while (!random_seed)
+		random_seed = get_random_int();
+
+	pr_info("Testing DRM range manger (struct drm_mm), with random_seed=0x%x max_iterations=%u max_prime=%u\n",
+		random_seed, max_iterations, max_prime);
+	err = run_selftests(selftests, ARRAY_SIZE(selftests), NULL);
+
+	return err > 0 ? 0 : err;
+}
+
+static void __exit test_drm_mm_exit(void)
+{
+}
+
+module_init(test_drm_mm_init);
+module_exit(test_drm_mm_exit);
+
+module_param(random_seed, uint, 0400);
+module_param(max_iterations, uint, 0400);
+module_param(max_prime, uint, 0400);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
diff --git a/tools/testing/selftests/drivers/gpu/drm_mm.sh b/tools/testing/selftests/drivers/gpu/drm_mm.sh
new file mode 100755
index 000000000000..96dd55c92799
--- /dev/null
+++ b/tools/testing/selftests/drivers/gpu/drm_mm.sh
@@ -0,0 +1,15 @@
+#!/bin/sh
+# Runs API tests for struct drm_mm (DRM range manager)
+
+if ! /sbin/modprobe -n -q test-drm_mm; then
+       echo "drivers/gpu/drm_mm: [skip]"
+       exit 77
+fi
+
+if /sbin/modprobe -q test-drm_mm; then
+       /sbin/modprobe -q -r test-drm_mm
+       echo "drivers/gpu/drm_mm: ok"
+else
+       echo "drivers/gpu/drm_mm: [FAIL]"
+       exit 1
+fi
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 07/38] drm: kselftest for drm_mm_init()

[Chris Wilson]

Simple first test to just exercise initialisation of struct drm_mm.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 114 +++++++++++++++++++++++++++
 2 files changed, 115 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 1610e0a63a5b..844dd29db540 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -6,3 +6,4 @@
  * Tests are executed in order by igt/drm_mm
  */
 selftest(sanitycheck, igt_sanitycheck) /* keep first (selfcheck for igt) */
+selftest(init, igt_init)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 682f5f86f465..87ad147670da 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -27,6 +27,120 @@ static int igt_sanitycheck(void *ignored)
 	return 0;
 }
 
+static bool assert_no_holes(const struct drm_mm *mm)
+{
+	struct drm_mm_node *hole;
+	u64 hole_start, hole_end;
+	unsigned long count;
+
+	count = 0;
+	drm_mm_for_each_hole(hole, mm, hole_start, hole_end)
+		count++;
+	if (count) {
+		pr_err("Expected to find no holes (after reserve), found %lu instead\n", count);
+		return false;
+	}
+
+	drm_mm_for_each_node(hole, mm) {
+		if (hole->hole_follows) {
+			pr_err("Hole follows node, expected none!\n");
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static bool assert_one_hole(const struct drm_mm *mm, u64 start, u64 end)
+{
+	struct drm_mm_node *hole;
+	u64 hole_start, hole_end;
+	unsigned long count;
+	bool ok = true;
+
+	if (end <= start)
+		return true;
+
+	count = 0;
+	drm_mm_for_each_hole(hole, mm, hole_start, hole_end) {
+		if (start != hole_start || end != hole_end) {
+			if (ok)
+				pr_err("empty mm has incorrect hole, found (%llx, %llx), expect (%llx, %llx)\n",
+				       hole_start, hole_end,
+				       start, end);
+			ok = false;
+		}
+		count++;
+	}
+	if (count != 1) {
+		pr_err("Expected to find one hole, found %lu instead\n", count);
+		ok = false;
+	}
+
+	return ok;
+}
+
+static int igt_init(void *ignored)
+{
+	const unsigned int size = 4096;
+	struct drm_mm mm;
+	struct drm_mm_node tmp;
+	int ret = -EINVAL;
+
+	/* Start with some simple checks on initialising the struct drm_mm */
+	memset(&mm, 0, sizeof(mm));
+	if (drm_mm_initialized(&mm)) {
+		pr_err("zeroed mm claims to be initialized\n");
+		return ret;
+	}
+
+	memset(&mm, 0xff, sizeof(mm));
+	drm_mm_init(&mm, 0, size);
+	if (!drm_mm_initialized(&mm)) {
+		pr_err("mm claims not to be initialized\n");
+		goto out;
+	}
+
+	if (!drm_mm_clean(&mm)) {
+		pr_err("mm not empty on creation\n");
+		goto out;
+	}
+
+	/* After creation, it should all be one massive hole */
+	if (!assert_one_hole(&mm, 0, size)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	memset(&tmp, 0, sizeof(tmp));
+	tmp.start = 0;
+	tmp.size = size;
+	ret = drm_mm_reserve_node(&mm, &tmp);
+	if (ret) {
+		pr_err("failed to reserve whole drm_mm\n");
+		goto out;
+	}
+
+	/* After filling the range entirely, there should be no holes */
+	if (!assert_no_holes(&mm)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	/* And then after emptying it again, the massive hole should be back */
+	drm_mm_remove_node(&tmp);
+	if (!assert_one_hole(&mm, 0, size)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+out:
+	if (ret)
+		drm_mm_debug_table(&mm, __func__);
+	drm_mm_takedown(&mm);
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 08/38] drm: kselftest for drm_mm_debug()

[Chris Wilson]

Simple test to just exercise calling the debug dumper on the drm_mm.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |  1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 35 ++++++++++++++++++++++++++++
 2 files changed, 36 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 844dd29db540..0265f09e92fa 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -7,3 +7,4 @@
  */
 selftest(sanitycheck, igt_sanitycheck) /* keep first (selfcheck for igt) */
 selftest(init, igt_init)
+selftest(debug, igt_debug)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 87ad147670da..434320061d9e 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -141,6 +141,41 @@ static int igt_init(void *ignored)
 	return ret;
 }
 
+static int igt_debug(void *ignored)
+{
+	struct drm_mm mm;
+	struct drm_mm_node nodes[2];
+	int ret;
+
+	/* Create a small drm_mm with a couple of nodes and a few holes, and
+	 * check that the debug iterator doesn't explode over a trivial drm_mm.
+	 */
+
+	drm_mm_init(&mm, 0, 4096);
+
+	memset(nodes, 0, sizeof(nodes));
+	nodes[0].start = 512;
+	nodes[0].size = 1024;
+	ret = drm_mm_reserve_node(&mm, &nodes[0]);
+	if (ret) {
+		pr_err("failed to reserve node[0] {start=%lld, size=%lld)\n",
+		       nodes[0].start, nodes[0].size);
+		return ret;
+	}
+
+	nodes[1].size = 1024;
+	nodes[1].start = 4096 - 512 - nodes[1].size;
+	ret = drm_mm_reserve_node(&mm, &nodes[1]);
+	if (ret) {
+		pr_err("failed to reserve node[1] {start=%lld, size=%lld)\n",
+		       nodes[1].start, nodes[1].size);
+		return ret;
+	}
+
+	drm_mm_debug_table(&mm, __func__);
+	return 0;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 09/38] drm: kselftest for drm_mm_reserve_node()

[Chris Wilson]

Exercise drm_mm_reserve_node(), check that we can't reserve an already
occupied range and that the lists are correct after reserving/removing.

v2: Check for invalid node reservation.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 275 +++++++++++++++++++++++++++
 2 files changed, 276 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 0265f09e92fa..693d85677e7f 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -8,3 +8,4 @@
 selftest(sanitycheck, igt_sanitycheck) /* keep first (selfcheck for igt) */
 selftest(init, igt_init)
 selftest(debug, igt_debug)
+selftest(reserve, igt_reserve)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 434320061d9e..673519cf41e1 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -80,6 +80,57 @@ static bool assert_one_hole(const struct drm_mm *mm, u64 start, u64 end)
 	return ok;
 }
 
+static bool assert_continuous(const struct drm_mm *mm, u64 size)
+{
+	struct drm_mm_node *node, *check, *found;
+	unsigned long n;
+	u64 addr;
+
+	if (!assert_no_holes(mm))
+		return false;
+
+	n = 0;
+	addr = 0;
+	drm_mm_for_each_node(node, mm) {
+		if (node->start != addr) {
+			pr_err("node[%ld] list out of order, expected %llx found %llx\n",
+			       n, addr, node->start);
+			return false;
+		}
+
+		if (node->size != size) {
+			pr_err("node[%ld].size incorrect, expected %llx, found %llx\n",
+			       n, size, node->size);
+			return false;
+		}
+
+		if (node->hole_follows) {
+			pr_err("node[%ld] is followed by a hole!\n", n);
+			return false;
+		}
+
+		found = NULL;
+		drm_mm_for_each_node_in_range(check, mm, addr, addr + size) {
+			if (node != check) {
+				pr_err("lookup return wrong node, expected start %llx, found %llx\n",
+				       node->start, check->start);
+				return false;
+			}
+			found = check;
+		}
+		if (!found) {
+			pr_err("lookup failed for node %llx + %llx\n",
+			       addr, size);
+			return false;
+		}
+
+		addr += size;
+		n++;
+	}
+
+	return true;
+}
+
 static int igt_init(void *ignored)
 {
 	const unsigned int size = 4096;
@@ -176,6 +227,230 @@ static int igt_debug(void *ignored)
 	return 0;
 }
 
+static struct drm_mm_node *set_node(struct drm_mm_node *node,
+				    u64 start, u64 size)
+{
+	node->start = start;
+	node->size = size;
+	return node;
+}
+
+static bool expect_reserve_fail(struct drm_mm *mm, struct drm_mm_node *node)
+{
+	int err;
+
+	err = drm_mm_reserve_node(mm, node);
+	if (likely(err == -ENOSPC))
+		return true;
+
+	if (!err) {
+		pr_err("impossible reserve succeeded, node %llu + %llu\n",
+		       node->start, node->size);
+		drm_mm_remove_node(node);
+	} else {
+		pr_err("impossible reserve failed with wrong error %d [expected %d], node %llu + %llu\n",
+		       err, -ENOSPC, node->start, node->size);
+	}
+	return false;
+}
+
+static bool check_reserve_boundaries(struct drm_mm *mm,
+				     unsigned int count,
+				     u64 size)
+{
+	const struct boundary {
+		u64 start, size;
+		const char *name;
+	} boundaries[] = {
+#define B(st, sz) { (st), (sz), "{ " #st ", " #sz "}" }
+		B(0, 0),
+		B(-size, 0),
+		B(size, 0),
+		B(size * count, 0),
+		B(-size, size),
+		B(-size, -size),
+		B(-size, 2*size),
+		B(0, -size),
+		B(size, -size),
+		B(count*size, size),
+		B(count*size, -size),
+		B(count*size, count*size),
+		B(count*size, -count*size),
+		B(count*size, -(count+1)*size),
+		B((count+1)*size, size),
+		B((count+1)*size, -size),
+		B((count+1)*size, -2*size),
+#undef B
+	};
+	struct drm_mm_node tmp = {};
+	int n;
+
+	for (n = 0; n < ARRAY_SIZE(boundaries); n++) {
+		if (!expect_reserve_fail(mm,
+					 set_node(&tmp,
+						  boundaries[n].start,
+						  boundaries[n].size))) {
+			pr_err("boundary[%d:%s] failed, count=%u, size=%lld\n",
+			       n, boundaries[n].name, count, size);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static int __igt_reserve(unsigned int count, u64 size)
+{
+	DRM_RND_STATE(prng, random_seed);
+	struct drm_mm mm;
+	struct drm_mm_node tmp, *nodes, *node, *next;
+	unsigned int *order, n, m, o = 0;
+	int ret, err;
+
+	/* For exercising drm_mm_reserve_node(), we want to check that
+	 * reservations outside of the drm_mm range are rejected, and to
+	 * overlapping and otherwise already occupied ranges. Afterwards,
+	 * the tree and nodes should be intact.
+	 */
+
+	DRM_MM_BUG_ON(!count);
+	DRM_MM_BUG_ON(!size);
+
+	ret = -ENOMEM;
+	order = drm_random_order(count, &prng);
+	if (!order)
+		goto err;
+
+	nodes = vzalloc(sizeof(*nodes) * count);
+	if (!nodes)
+		goto err_order;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, count * size);
+
+	if (!check_reserve_boundaries(&mm, count, size))
+		goto out;
+
+	for (n = 0; n < count; n++) {
+		nodes[n].start = order[n] * size;
+		nodes[n].size = size;
+
+		err = drm_mm_reserve_node(&mm, &nodes[n]);
+		if (err) {
+			pr_err("reserve failed, step %d, start %llu\n",
+			       n, nodes[n].start);
+			ret = err;
+			goto out;
+		}
+
+		if (!drm_mm_node_allocated(&nodes[n])) {
+			pr_err("reserved node not allocated! step %d, start %llu\n",
+			       n, nodes[n].start);
+			goto out;
+		}
+
+		if (!expect_reserve_fail(&mm, &nodes[n]))
+			goto out;
+	}
+
+	/* After random insertion the nodes should be in order */
+	if (!assert_continuous(&mm, size))
+		goto out;
+
+	/* Repeated use should then fail */
+	drm_random_reorder(order, count, &prng);
+	for (n = 0; n < count; n++) {
+		if (!expect_reserve_fail(&mm,
+					 set_node(&tmp, order[n] * size, 1)))
+			goto out;
+
+		/* Remove and reinsert should work */
+		drm_mm_remove_node(&nodes[order[n]]);
+		err = drm_mm_reserve_node(&mm, &nodes[order[n]]);
+		if (err) {
+			pr_err("reserve failed, step %d, start %llu\n",
+			       n, nodes[n].start);
+			ret = err;
+			goto out;
+		}
+	}
+
+	if (!assert_continuous(&mm, size))
+		goto out;
+
+	/* Overlapping use should then fail */
+	for (n = 0; n < count; n++) {
+		if (!expect_reserve_fail(&mm, set_node(&tmp, 0, size*count)))
+			goto out;
+	}
+	for (n = 0; n < count; n++) {
+		if (!expect_reserve_fail(&mm,
+					 set_node(&tmp,
+						  size * n,
+						  size * (count - n))))
+			goto out;
+	}
+
+	/* Remove several, reinsert, check full */
+	for_each_prime_number(n, min(max_prime, count)) {
+		for (m = 0; m < n; m++) {
+			node = &nodes[order[(o + m) % count]];
+			drm_mm_remove_node(node);
+		}
+
+		for (m = 0; m < n; m++) {
+			node = &nodes[order[(o + m) % count]];
+			err = drm_mm_reserve_node(&mm, node);
+			if (err) {
+				pr_err("reserve failed, step %d/%d, start %llu\n",
+				       m, n, node->start);
+				ret = err;
+				goto out;
+			}
+		}
+
+		o += n;
+
+		if (!assert_continuous(&mm, size))
+			goto out;
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	vfree(nodes);
+err_order:
+	kfree(order);
+err:
+	return ret;
+}
+
+static int igt_reserve(void *ignored)
+{
+	const unsigned int count = min_t(unsigned int, BIT(10), max_iterations);
+	int n, ret;
+
+	for_each_prime_number(n, 54) {
+		u64 size = BIT_ULL(n);
+
+		ret = __igt_reserve(count, size - 1);
+		if (ret)
+			return ret;
+
+		ret = __igt_reserve(count, size);
+		if (ret)
+			return ret;
+
+		ret = __igt_reserve(count, size + 1);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 10/38] drm: kselftest for drm_mm_insert_node()

[Chris Wilson]

Exercise drm_mm_insert_node(), check that we can't overfill a range and
that the lists are correct after reserving/removing.

v2: Extract helpers for the repeated tests
v3: Iterate over all allocation flags

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 234 +++++++++++++++++++++++++++
 2 files changed, 235 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 693d85677e7f..727c6d7255e0 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -9,3 +9,4 @@ selftest(sanitycheck, igt_sanitycheck) /* keep first (selfcheck for igt) */
 selftest(init, igt_init)
 selftest(debug, igt_debug)
 selftest(reserve, igt_reserve)
+selftest(insert, igt_insert)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 673519cf41e1..2d78a5a0f98c 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -21,6 +21,23 @@ static unsigned int random_seed;
 static unsigned int max_iterations = 8192;
 static unsigned int max_prime = 128;
 
+enum {
+	DEFAULT,
+	TOPDOWN,
+	BEST,
+};
+
+static const struct insert_mode {
+	const char *name;
+	unsigned int search_flags;
+	unsigned int create_flags;
+} insert_modes[] = {
+	[DEFAULT] = { "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
+	[TOPDOWN] = { "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
+	[BEST] = { "best", DRM_MM_SEARCH_BEST, DRM_MM_CREATE_DEFAULT },
+	{}
+};
+
 static int igt_sanitycheck(void *ignored)
 {
 	pr_info("%s - ok!\n", __func__);
@@ -131,6 +148,48 @@ static bool assert_continuous(const struct drm_mm *mm, u64 size)
 	return true;
 }
 
+static u64 misalignment(struct drm_mm_node *node, u64 alignment)
+{
+	u64 rem;
+
+	if (!alignment)
+		return 0;
+
+	div64_u64_rem(node->start, alignment, &rem);
+	return rem;
+}
+
+static bool assert_node(struct drm_mm_node *node, struct drm_mm *mm,
+			u64 size, u64 alignment, unsigned long color)
+{
+	bool ok = true;
+
+	if (!drm_mm_node_allocated(node) || node->mm != mm) {
+		pr_err("node not allocated\n");
+		ok = false;
+	}
+
+	if (node->size != size) {
+		pr_err("node has wrong size, found %llu, expected %llu\n",
+		       node->size, size);
+		ok = false;
+	}
+
+	if (misalignment(node, alignment)) {
+		pr_err("node is misalinged, start %llx rem %llu, expected alignment %llu\n",
+		       node->start, misalignment(node, alignment), alignment);
+		ok = false;
+	}
+
+	if (node->color != color) {
+		pr_err("node has wrong color, found %lu, expected %lu\n",
+		       node->color, color);
+		ok = false;
+	}
+
+	return ok;
+}
+
 static int igt_init(void *ignored)
 {
 	const unsigned int size = 4096;
@@ -451,6 +510,181 @@ static int igt_reserve(void *ignored)
 	return 0;
 }
 
+static bool expect_insert(struct drm_mm *mm, struct drm_mm_node *node,
+			  u64 size, u64 alignment, unsigned long color,
+			  const struct insert_mode *mode)
+{
+	int err;
+
+	err = drm_mm_insert_node_generic(mm, node,
+					 size, alignment, color,
+					 mode->search_flags,
+					 mode->create_flags);
+	if (err) {
+		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) failed with err=%d\n",
+		       size, alignment, color, mode->name, err);
+		return false;
+	}
+
+	if (!assert_node(node, mm, size, alignment, color)) {
+		drm_mm_remove_node(node);
+		return false;
+	}
+
+	return true;
+}
+
+static bool expect_insert_fail(struct drm_mm *mm, u64 size)
+{
+	struct drm_mm_node tmp = {};
+	int err;
+
+	err = drm_mm_insert_node(mm, &tmp, size, 0, DRM_MM_SEARCH_DEFAULT);
+	if (likely(err == -ENOSPC))
+		return true;
+
+	if (!err) {
+		pr_err("impossible insert succeeded, node %llu + %llu\n",
+		       tmp.start, tmp.size);
+		drm_mm_remove_node(&tmp);
+	} else {
+		pr_err("impossible insert failed with wrong error %d [expected %d], size %llu\n",
+		       err, -ENOSPC, size);
+	}
+	return false;
+}
+
+static int __igt_insert(unsigned int count, u64 size)
+{
+	DRM_RND_STATE(prng, random_seed);
+	const struct insert_mode *mode;
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int *order, n, m, o = 0;
+	int ret;
+
+	/* Fill a range with lots of nodes, check it doesn't fail too early */
+
+	DRM_MM_BUG_ON(!count);
+	DRM_MM_BUG_ON(!size);
+
+	ret = -ENOMEM;
+	nodes = vzalloc(count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	order = drm_random_order(count, &prng);
+	if (!order)
+		goto err_nodes;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, count * size);
+
+	for (mode = insert_modes; mode->name; mode++) {
+		for (n = 0; n < count; n++) {
+			if (!expect_insert(&mm, &nodes[n], size, 0, n, mode)) {
+				pr_err("%s insert failed, size %llu step %d\n",
+				       mode->name, size, n);
+				goto out;
+			}
+		}
+
+		/* After random insertion the nodes should be in order */
+		if (!assert_continuous(&mm, size))
+			goto out;
+
+		/* Repeated use should then fail */
+		if (!expect_insert_fail(&mm, size))
+			goto out;
+
+		/* Remove one and reinsert, as the only hole it should refill itself */
+		for (n = 0; n < count; n++) {
+			u64 addr = nodes[n].start;
+
+			drm_mm_remove_node(&nodes[n]);
+			if (!expect_insert(&mm, &nodes[n], size, 0, n, mode)) {
+				pr_err("%s reinsert failed, size %llu step %d\n",
+				       mode->name, size, n);
+				goto out;
+			}
+
+			if (nodes[n].start != addr) {
+				pr_err("%s reinsert node moved, step %d, expected %llx, found %llx\n",
+				       mode->name, n, addr, nodes[n].start);
+				goto out;
+			}
+
+			if (!assert_continuous(&mm, size))
+				goto out;
+		}
+
+		/* Remove several, reinsert, check full */
+		for_each_prime_number(n, min(max_prime, count)) {
+			for (m = 0; m < n; m++) {
+				node = &nodes[order[(o + m) % count]];
+				drm_mm_remove_node(node);
+			}
+
+			for (m = 0; m < n; m++) {
+				node = &nodes[order[(o + m) % count]];
+				if (!expect_insert(&mm, node, size, 0, n, mode)) {
+					pr_err("%s multiple reinsert failed, size %llu step %d\n",
+					       mode->name, size, n);
+					goto out;
+				}
+			}
+
+			o += n;
+
+			if (!assert_continuous(&mm, size))
+				goto out;
+
+			if (!expect_insert_fail(&mm, size))
+				goto out;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(!drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
+static int igt_insert(void *ignored)
+{
+	const unsigned int count = min_t(unsigned int, BIT(10), max_iterations);
+	unsigned int n;
+	int ret;
+
+	for_each_prime_number(n, 54) {
+		u64 size = BIT_ULL(n);
+
+		ret = __igt_insert(count, size - 1);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert(count, size);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert(count, size + 1);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 11/38] drm: kselftest for drm_mm_replace_node()

[Chris Wilson]

Reuse drm_mm_insert_node() with a temporary node to exercise
drm_mm_replace_node(). We use the previous test in order to exercise the
various lists following replacement.

v2: Check that we copy across the important (user) details of the node.
The internal details (such as lists and hole tracking) we hope to detect
errors by exercise.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |  1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 65 +++++++++++++++++++++++++---
 2 files changed, 60 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 727c6d7255e0..dca726baa65d 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -10,3 +10,4 @@ selftest(init, igt_init)
 selftest(debug, igt_debug)
 selftest(reserve, igt_reserve)
 selftest(insert, igt_insert)
+selftest(replace, igt_replace)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 2d78a5a0f98c..fa5b12c57161 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -554,7 +554,7 @@ static bool expect_insert_fail(struct drm_mm *mm, u64 size)
 	return false;
 }
 
-static int __igt_insert(unsigned int count, u64 size)
+static int __igt_insert(unsigned int count, u64 size, bool replace)
 {
 	DRM_RND_STATE(prng, random_seed);
 	const struct insert_mode *mode;
@@ -569,7 +569,7 @@ static int __igt_insert(unsigned int count, u64 size)
 	DRM_MM_BUG_ON(!size);
 
 	ret = -ENOMEM;
-	nodes = vzalloc(count * sizeof(*nodes));
+	nodes = vmalloc(count * sizeof(*nodes));
 	if (!nodes)
 		goto err;
 
@@ -582,11 +582,37 @@ static int __igt_insert(unsigned int count, u64 size)
 
 	for (mode = insert_modes; mode->name; mode++) {
 		for (n = 0; n < count; n++) {
-			if (!expect_insert(&mm, &nodes[n], size, 0, n, mode)) {
+			struct drm_mm_node tmp;
+
+			node = replace ? &tmp : &nodes[n];
+			memset(node, 0, sizeof(*node));
+			if (!expect_insert(&mm, node, size, 0, n, mode)) {
 				pr_err("%s insert failed, size %llu step %d\n",
 				       mode->name, size, n);
 				goto out;
 			}
+
+			if (replace) {
+				drm_mm_replace_node(&tmp, &nodes[n]);
+				if (drm_mm_node_allocated(&tmp)) {
+					pr_err("replaced old-node still allocated! step %d\n",
+					       n);
+					goto out;
+				}
+
+				if (!assert_node(&nodes[n], &mm, size, 0, n)) {
+					pr_err("replaced node did not inherit parameters, size %llu step %d\n",
+					       size, n);
+					goto out;
+				}
+
+				if (tmp.start != nodes[n].start) {
+					pr_err("replaced node mismatch location expected [%llx + %llx], found [%llx + %llx]\n",
+					       tmp.start, size,
+					       nodes[n].start, nodes[n].size);
+					goto out;
+				}
+			}
 		}
 
 		/* After random insertion the nodes should be in order */
@@ -669,17 +695,44 @@ static int igt_insert(void *ignored)
 	for_each_prime_number(n, 54) {
 		u64 size = BIT_ULL(n);
 
-		ret = __igt_insert(count, size - 1);
+		ret = __igt_insert(count, size - 1, false);
 		if (ret)
 			return ret;
 
-		ret = __igt_insert(count, size);
+		ret = __igt_insert(count, size, false);
 		if (ret)
 			return ret;
 
-		ret = __igt_insert(count, size + 1);
+		ret = __igt_insert(count, size + 1, false);
+	}
+
+	return 0;
+}
+
+static int igt_replace(void *ignored)
+{
+	const unsigned int count = min_t(unsigned int, BIT(10), max_iterations);
+	unsigned int n;
+	int ret;
+
+	/* Reuse igt_insert to exercise replacement by inserting a dummy node,
+	 * then replacing it with the intended node. We want to check that
+	 * the tree is intact and all the information we need is carried
+	 * across to the target node.
+	 */
+
+	for_each_prime_number(n, 54) {
+		u64 size = BIT_ULL(n);
+
+		ret = __igt_insert(count, size - 1, true);
 		if (ret)
 			return ret;
+
+		ret = __igt_insert(count, size, true);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert(count, size + 1, true);
 	}
 
 	return 0;
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 12/38] drm: kselftest for drm_mm_insert_node_in_range()

[Chris Wilson]

Exercise drm_mm_insert_node_in_range(), check that we only allocate from
the specified range.

v2: Use all allocation flags

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 264 +++++++++++++++++++++++++++
 2 files changed, 265 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index dca726baa65d..92b2c1cb10fa 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -11,3 +11,4 @@ selftest(debug, igt_debug)
 selftest(reserve, igt_reserve)
 selftest(insert, igt_insert)
 selftest(replace, igt_replace)
+selftest(insert_range, igt_insert_range)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index fa5b12c57161..cbd4f09e6b2f 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -738,6 +738,270 @@ static int igt_replace(void *ignored)
 	return 0;
 }
 
+static bool expect_insert_in_range(struct drm_mm *mm, struct drm_mm_node *node,
+				   u64 size, u64 alignment, unsigned long color,
+				   u64 range_start, u64 range_end,
+				   const struct insert_mode *mode)
+{
+	int err;
+
+	err = drm_mm_insert_node_in_range_generic(mm, node,
+						  size, alignment, color,
+						  range_start, range_end,
+						  mode->search_flags,
+						  mode->create_flags);
+	if (err) {
+		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) nto range [%llx, %llx] failed with err=%d\n",
+		       size, alignment, color, mode->name,
+		       range_start, range_end, err);
+		return false;
+	}
+
+	if (!assert_node(node, mm, size, alignment, color)) {
+		drm_mm_remove_node(node);
+		return false;
+	}
+
+	return true;
+}
+
+static bool expect_insert_in_range_fail(struct drm_mm *mm,
+					u64 size,
+					u64 range_start,
+					u64 range_end)
+{
+	struct drm_mm_node tmp = {};
+	int err;
+
+	err = drm_mm_insert_node_in_range_generic(mm, &tmp,
+						  size, 0, 0,
+						  range_start, range_end,
+						  DRM_MM_SEARCH_DEFAULT,
+						  DRM_MM_CREATE_DEFAULT);
+	if (likely(err == -ENOSPC))
+		return true;
+
+	if (!err) {
+		pr_err("impossible insert succeeded, node %llx + %llu, range [%llx, %llx]\n",
+		       tmp.start, tmp.size, range_start, range_end);
+		drm_mm_remove_node(&tmp);
+	} else {
+		pr_err("impossible insert failed with wrong error %d [expected %d], size %llu, range [%llx, %llx]\n",
+		       err, -ENOSPC, size, range_start, range_end);
+	}
+
+	return false;
+}
+
+static bool assert_contiguous_in_range(struct drm_mm *mm,
+				       u64 size,
+				       u64 start,
+				       u64 end)
+{
+	struct drm_mm_node *node;
+	unsigned int n;
+
+	if (!expect_insert_in_range_fail(mm, size, start, end))
+		return false;
+
+	n = div64_u64(start + size - 1, size);
+	drm_mm_for_each_node(node, mm) {
+		if (node->start < start || node->start + node->size > end) {
+			pr_err("node %d out of range, address [%llx + %llu], range [%llx, %llx]\n",
+			       n, node->start, node->start + node->size, start, end);
+			return false;
+		}
+
+		if (node->start != n * size) {
+			pr_err("node %d out of order, expected start %llx, found %llx\n",
+			       n, n * size, node->start);
+			return false;
+		}
+
+		if (node->size != size) {
+			pr_err("node %d has wrong size, expected size %llx, found %llx\n",
+			       n, size, node->size);
+			return false;
+		}
+
+		if (node->hole_follows && drm_mm_hole_node_end(node) < end) {
+			pr_err("node %d is followed by a hole!\n", n);
+			return false;
+		}
+
+		n++;
+	}
+
+	drm_mm_for_each_node_in_range(node, mm, 0, start) {
+		if (node) {
+			pr_err("node before start: node=%llx+%llu, start=%llx\n",
+			       node->start, node->size, start);
+			return false;
+		}
+	}
+
+	drm_mm_for_each_node_in_range(node, mm, end, U64_MAX) {
+		if (node) {
+			pr_err("node after end: node=%llx+%llu, end=%llx\n",
+			       node->start, node->size, end);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static int __igt_insert_range(unsigned int count, u64 size, u64 start, u64 end)
+{
+	const struct insert_mode *mode;
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int n, start_n, end_n;
+	int ret;
+
+	DRM_MM_BUG_ON(!count);
+	DRM_MM_BUG_ON(!size);
+	DRM_MM_BUG_ON(end <= start);
+
+	/* Very similar to __igt_insert(), but now instead of populating the
+	 * full range of the drm_mm, we try to fill a small portion of it.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, count * size);
+
+	start_n = div64_u64(start + size - 1, size);
+	end_n = div64_u64(end - size, size);
+
+	for (mode = insert_modes; mode->name; mode++) {
+		for (n = start_n; n <= end_n; n++) {
+			if (!expect_insert_in_range(&mm, &nodes[n],
+						    size, size, n,
+						    start, end, mode)) {
+				pr_err("%s insert failed, size %llu, step %d [%d, %d], range [%llx, %llx]\n",
+				       mode->name, size, n,
+				       start_n, end_n,
+				       start, end);
+				goto out;
+			}
+		}
+
+		if (!assert_contiguous_in_range(&mm, size, start, end)) {
+			pr_err("%s: range [%llx, %llx] not full after initialisation, size=%llu\n",
+			       mode->name, start, end, size);
+			goto out;
+		}
+
+		/* Remove one and reinsert, it should refill itself */
+		for (n = start_n; n <= end_n; n++) {
+			u64 addr = nodes[n].start;
+
+			drm_mm_remove_node(&nodes[n]);
+			if (!expect_insert_in_range(&mm, &nodes[n],
+						    size, size, n,
+						    start, end, mode)) {
+				pr_err("%s reinsert failed, step %d\n", mode->name, n);
+				goto out;
+			}
+
+			if (nodes[n].start != addr) {
+				pr_err("%s reinsert node moved, step %d, expected %llx, found %llx\n",
+				       mode->name, n, addr, nodes[n].start);
+				goto out;
+			}
+		}
+
+		if (!assert_contiguous_in_range(&mm, size, start, end)) {
+			pr_err("%s: range [%llx, %llx] not full after reinsertion, size=%llu\n",
+			       mode->name, start, end, size);
+			goto out;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(!drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	vfree(nodes);
+err:
+	return ret;
+}
+
+static int insert_outside_range(void)
+{
+	struct drm_mm mm;
+	const unsigned int start = 1024;
+	const unsigned int end = 2048;
+	const unsigned int size = end - start;
+
+	drm_mm_init(&mm, start, size);
+
+	if (!expect_insert_in_range_fail(&mm, 1, 0, start))
+		return -EINVAL;
+
+	if (!expect_insert_in_range_fail(&mm, 1, end, end - size))
+		return -EINVAL;
+
+	if (!expect_insert_in_range_fail(&mm, 1, end, end + size))
+		return -EINVAL;
+
+	drm_mm_takedown(&mm);
+	return 0;
+}
+
+static int igt_insert_range(void *ignored)
+{
+	const unsigned int count = min_t(unsigned int, BIT(13), max_iterations);
+	unsigned int n;
+	int ret;
+
+	/* Check that requests outside the bounds of drm_mm are rejected. */
+	ret = insert_outside_range();
+	if (ret)
+		return ret;
+
+	for_each_prime_number(n, 50) {
+		const u64 size = BIT_ULL(n);
+		const u64 max = count * size;
+
+		ret = __igt_insert_range(count, size, 0, max);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, 1, max);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, 0, max - 1);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, 0, max/2);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, max/2, max);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, max/4+1, 3*max/4-1);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH] drm: kselftest for drm_mm_insert_node_in_range()

[Chris Wilson]

Exercise drm_mm_insert_node_in_range(), check that we only allocate from
the specified range.

v2: Use all allocation flags
v3: Don't pass in invalid ranges - these will be asserted later.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 269 +++++++++++++++++++++++++++
 2 files changed, 270 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index dca726baa65d..92b2c1cb10fa 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -11,3 +11,4 @@ selftest(debug, igt_debug)
 selftest(reserve, igt_reserve)
 selftest(insert, igt_insert)
 selftest(replace, igt_replace)
+selftest(insert_range, igt_insert_range)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index fa5b12c57161..cfbc3f53798e 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -738,6 +738,275 @@ static int igt_replace(void *ignored)
 	return 0;
 }
 
+static bool expect_insert_in_range(struct drm_mm *mm, struct drm_mm_node *node,
+				   u64 size, u64 alignment, unsigned long color,
+				   u64 range_start, u64 range_end,
+				   const struct insert_mode *mode)
+{
+	int err;
+
+	err = drm_mm_insert_node_in_range_generic(mm, node,
+						  size, alignment, color,
+						  range_start, range_end,
+						  mode->search_flags,
+						  mode->create_flags);
+	if (err) {
+		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) nto range [%llx, %llx] failed with err=%d\n",
+		       size, alignment, color, mode->name,
+		       range_start, range_end, err);
+		return false;
+	}
+
+	if (!assert_node(node, mm, size, alignment, color)) {
+		drm_mm_remove_node(node);
+		return false;
+	}
+
+	return true;
+}
+
+static bool expect_insert_in_range_fail(struct drm_mm *mm,
+					u64 size,
+					u64 range_start,
+					u64 range_end)
+{
+	struct drm_mm_node tmp = {};
+	int err;
+
+	err = drm_mm_insert_node_in_range_generic(mm, &tmp,
+						  size, 0, 0,
+						  range_start, range_end,
+						  DRM_MM_SEARCH_DEFAULT,
+						  DRM_MM_CREATE_DEFAULT);
+	if (likely(err == -ENOSPC))
+		return true;
+
+	if (!err) {
+		pr_err("impossible insert succeeded, node %llx + %llu, range [%llx, %llx]\n",
+		       tmp.start, tmp.size, range_start, range_end);
+		drm_mm_remove_node(&tmp);
+	} else {
+		pr_err("impossible insert failed with wrong error %d [expected %d], size %llu, range [%llx, %llx]\n",
+		       err, -ENOSPC, size, range_start, range_end);
+	}
+
+	return false;
+}
+
+static bool assert_contiguous_in_range(struct drm_mm *mm,
+				       u64 size,
+				       u64 start,
+				       u64 end)
+{
+	struct drm_mm_node *node;
+	unsigned int n;
+
+	if (!expect_insert_in_range_fail(mm, size, start, end))
+		return false;
+
+	n = div64_u64(start + size - 1, size);
+	drm_mm_for_each_node(node, mm) {
+		if (node->start < start || node->start + node->size > end) {
+			pr_err("node %d out of range, address [%llx + %llu], range [%llx, %llx]\n",
+			       n, node->start, node->start + node->size, start, end);
+			return false;
+		}
+
+		if (node->start != n * size) {
+			pr_err("node %d out of order, expected start %llx, found %llx\n",
+			       n, n * size, node->start);
+			return false;
+		}
+
+		if (node->size != size) {
+			pr_err("node %d has wrong size, expected size %llx, found %llx\n",
+			       n, size, node->size);
+			return false;
+		}
+
+		if (node->hole_follows && drm_mm_hole_node_end(node) < end) {
+			pr_err("node %d is followed by a hole!\n", n);
+			return false;
+		}
+
+		n++;
+	}
+
+	drm_mm_for_each_node_in_range(node, mm, 0, start) {
+		if (node) {
+			pr_err("node before start: node=%llx+%llu, start=%llx\n",
+			       node->start, node->size, start);
+			return false;
+		}
+	}
+
+	drm_mm_for_each_node_in_range(node, mm, end, U64_MAX) {
+		if (node) {
+			pr_err("node after end: node=%llx+%llu, end=%llx\n",
+			       node->start, node->size, end);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static int __igt_insert_range(unsigned int count, u64 size, u64 start, u64 end)
+{
+	const struct insert_mode *mode;
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int n, start_n, end_n;
+	int ret;
+
+	DRM_MM_BUG_ON(!count);
+	DRM_MM_BUG_ON(!size);
+	DRM_MM_BUG_ON(end <= start);
+
+	/* Very similar to __igt_insert(), but now instead of populating the
+	 * full range of the drm_mm, we try to fill a small portion of it.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, count * size);
+
+	start_n = div64_u64(start + size - 1, size);
+	end_n = div64_u64(end - size, size);
+
+	for (mode = insert_modes; mode->name; mode++) {
+		for (n = start_n; n <= end_n; n++) {
+			if (!expect_insert_in_range(&mm, &nodes[n],
+						    size, size, n,
+						    start, end, mode)) {
+				pr_err("%s insert failed, size %llu, step %d [%d, %d], range [%llx, %llx]\n",
+				       mode->name, size, n,
+				       start_n, end_n,
+				       start, end);
+				goto out;
+			}
+		}
+
+		if (!assert_contiguous_in_range(&mm, size, start, end)) {
+			pr_err("%s: range [%llx, %llx] not full after initialisation, size=%llu\n",
+			       mode->name, start, end, size);
+			goto out;
+		}
+
+		/* Remove one and reinsert, it should refill itself */
+		for (n = start_n; n <= end_n; n++) {
+			u64 addr = nodes[n].start;
+
+			drm_mm_remove_node(&nodes[n]);
+			if (!expect_insert_in_range(&mm, &nodes[n],
+						    size, size, n,
+						    start, end, mode)) {
+				pr_err("%s reinsert failed, step %d\n", mode->name, n);
+				goto out;
+			}
+
+			if (nodes[n].start != addr) {
+				pr_err("%s reinsert node moved, step %d, expected %llx, found %llx\n",
+				       mode->name, n, addr, nodes[n].start);
+				goto out;
+			}
+		}
+
+		if (!assert_contiguous_in_range(&mm, size, start, end)) {
+			pr_err("%s: range [%llx, %llx] not full after reinsertion, size=%llu\n",
+			       mode->name, start, end, size);
+			goto out;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(!drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	vfree(nodes);
+err:
+	return ret;
+}
+
+static int insert_outside_range(void)
+{
+	struct drm_mm mm;
+	const unsigned int start = 1024;
+	const unsigned int end = 2048;
+	const unsigned int size = end - start;
+
+	drm_mm_init(&mm, start, size);
+
+	if (!expect_insert_in_range_fail(&mm, 1, 0, start))
+		return -EINVAL;
+
+	if (!expect_insert_in_range_fail(&mm, size,
+					 start - size/2, start + (size+1)/2))
+		return -EINVAL;
+
+	if (!expect_insert_in_range_fail(&mm, size,
+					 end - (size+1)/2, end + size/2))
+		return -EINVAL;
+
+	if (!expect_insert_in_range_fail(&mm, 1, end, end + size))
+		return -EINVAL;
+
+	drm_mm_takedown(&mm);
+	return 0;
+}
+
+static int igt_insert_range(void *ignored)
+{
+	const unsigned int count = min_t(unsigned int, BIT(13), max_iterations);
+	unsigned int n;
+	int ret;
+
+	/* Check that requests outside the bounds of drm_mm are rejected. */
+	ret = insert_outside_range();
+	if (ret)
+		return ret;
+
+	for_each_prime_number(n, 50) {
+		const u64 size = BIT_ULL(n);
+		const u64 max = count * size;
+
+		ret = __igt_insert_range(count, size, 0, max);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, 1, max);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, 0, max - 1);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, 0, max/2);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, max/2, max);
+		if (ret)
+			return ret;
+
+		ret = __igt_insert_range(count, size, max/4+1, 3*max/4-1);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 13/38] drm: kselftest for drm_mm and alignment

[Chris Wilson]

Check that we can request alignment to any power-of-two or prime using a
plain drm_mm_node_insert(), and also handle a reasonable selection of
primes.

v2: Exercise all allocation flags

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   3 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 103 +++++++++++++++++++++++++++
 2 files changed, 106 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 92b2c1cb10fa..a7a3763f8b20 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -12,3 +12,6 @@ selftest(reserve, igt_reserve)
 selftest(insert, igt_insert)
 selftest(replace, igt_replace)
 selftest(insert_range, igt_insert_range)
+selftest(align, igt_align)
+selftest(align32, igt_align32)
+selftest(align64, igt_align64)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index cbd4f09e6b2f..e1c034ea7e73 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1002,6 +1002,109 @@ static int igt_insert_range(void *ignored)
 	return 0;
 }
 
+static int igt_align(void *ignored)
+{
+	const struct insert_mode *mode;
+	const unsigned int max_count = min(8192u, max_prime);
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int prime;
+	int ret = -EINVAL;
+
+	/* For each of the possible insertion modes, we pick a few
+	 * arbitrary alignments and check that the inserted node
+	 * meets our requirements.
+	 */
+
+	nodes = vzalloc(max_count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	drm_mm_init(&mm, 1, U64_MAX - 2);
+
+	for (mode = insert_modes; mode->name; mode++) {
+		unsigned int i = 0;
+
+		for_each_prime_number(prime, max_count) {
+			u64 size = next_prime_number(prime);
+
+			if (!expect_insert(&mm, &nodes[i],
+					   size, prime, i,
+					   mode)) {
+				pr_err("%s insert failed with alignment=%d",
+				       mode->name, prime);
+				goto out;
+			}
+
+			i++;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(!drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	vfree(nodes);
+err:
+	return ret;
+}
+
+static int igt_align_pot(int max)
+{
+	struct drm_mm mm;
+	struct drm_mm_node *node, *next;
+	int bit;
+	int ret = -EINVAL;
+
+	/* Check that we can align to the full u64 address space */
+
+	drm_mm_init(&mm, 1, U64_MAX - 1);
+
+	for (bit = max - 1; bit; bit--) {
+		u64 align, size;
+
+		node = kzalloc(sizeof(*node), GFP_KERNEL);
+		if (!node) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		align = BIT_ULL(bit);
+		size = BIT_ULL(bit-1) + 1;
+		if (!expect_insert(&mm, node,
+				   size, align, bit,
+				   &insert_modes[0])) {
+			pr_err("insert failed with alignment=%llx [%d]",
+			       align, bit);
+			goto out;
+		}
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm) {
+		drm_mm_remove_node(node);
+		kfree(node);
+	}
+	drm_mm_takedown(&mm);
+	return ret;
+}
+
+static int igt_align32(void *ignored)
+{
+	return igt_align_pot(32);
+}
+
+static int igt_align64(void *ignored)
+{
+	return igt_align_pot(64);
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 14/38] drm: kselftest for drm_mm and eviction

[Chris Wilson]

Check that we add arbitrary blocks to the eviction scanner in order to
find the first minimal hole that matches our request.

v2: Refactor out some common eviction code for later

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 337 +++++++++++++++++++++++++++
 2 files changed, 338 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index a7a3763f8b20..a31b4458c7eb 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -15,3 +15,4 @@ selftest(insert_range, igt_insert_range)
 selftest(align, igt_align)
 selftest(align32, igt_align32)
 selftest(align64, igt_align64)
+selftest(evict, igt_evict)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index e1c034ea7e73..10142bc4e07e 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -36,6 +36,10 @@ static const struct insert_mode {
 	[TOPDOWN] = { "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
 	[BEST] = { "best", DRM_MM_SEARCH_BEST, DRM_MM_CREATE_DEFAULT },
 	{}
+}, evict_modes[] = {
+	{ "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
+	{ "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
+	{}
 };
 
 static int igt_sanitycheck(void *ignored)
@@ -1105,6 +1109,339 @@ static int igt_align64(void *ignored)
 	return igt_align_pot(64);
 }
 
+static void show_scan(const struct drm_mm *scan)
+{
+	pr_info("scan: hit [%llx, %llx], size=%lld, align=%d, color=%ld\n",
+		scan->scan_hit_start, scan->scan_hit_end,
+		scan->scan_size, scan->scan_alignment, scan->scan_color);
+}
+
+static void show_holes(const struct drm_mm *mm, int count)
+{
+	u64 hole_start, hole_end;
+	struct drm_mm_node *hole;
+
+	drm_mm_for_each_hole(hole, mm, hole_start, hole_end) {
+		struct drm_mm_node *next = list_next_entry(hole, node_list);
+		const char *node1 = NULL, *node2 = NULL;
+
+		if (hole->allocated)
+			node1 = kasprintf(GFP_KERNEL,
+					  "[%llx + %lld, color=%ld], ",
+					  hole->start, hole->size, hole->color);
+
+		if (next->allocated)
+			node2 = kasprintf(GFP_KERNEL,
+					  ", [%llx + %lld, color=%ld]",
+					  next->start, next->size, next->color);
+
+		pr_info("%sHole [%llx - %llx, size %lld]%s\n",
+			node1,
+			hole_start, hole_end, hole_end - hole_start,
+			node2);
+
+		kfree(node2);
+		kfree(node1);
+
+		if (!--count)
+			break;
+	}
+}
+
+struct evict_node {
+	struct drm_mm_node node;
+	struct list_head link;
+};
+
+static bool evict_nodes(struct drm_mm *mm,
+			struct evict_node *nodes,
+			unsigned int *order,
+			unsigned int count,
+			struct list_head *evict_list)
+{
+	struct evict_node *e, *en;
+	unsigned int i;
+
+	for (i = 0; i < count; i++) {
+		e = &nodes[order ? order[i] : i];
+		list_add(&e->link, evict_list);
+		if (drm_mm_scan_add_block(&e->node))
+			break;
+	}
+	list_for_each_entry_safe(e, en, evict_list, link) {
+		if (!drm_mm_scan_remove_block(&e->node))
+			list_del(&e->link);
+	}
+	if (list_empty(evict_list)) {
+		pr_err("Failed to find eviction: size=%lld [avail=%d], align=%d (color=%lu)\n",
+		       mm->scan_size, count,
+		       mm->scan_alignment,
+		       mm->scan_color);
+		return false;
+	}
+
+	list_for_each_entry(e, evict_list, link)
+		drm_mm_remove_node(&e->node);
+
+	return true;
+}
+
+static bool evict_nothing(struct drm_mm *mm,
+			  unsigned int total_size,
+			  struct evict_node *nodes)
+{
+	LIST_HEAD(evict_list);
+	struct evict_node *e;
+	struct drm_mm_node *node;
+	unsigned int n;
+
+	drm_mm_init_scan(mm, 1, 0, 0);
+	for (n = 0; n < total_size; n++) {
+		e = &nodes[n];
+		list_add(&e->link, &evict_list);
+		drm_mm_scan_add_block(&e->node);
+	}
+	list_for_each_entry(e, &evict_list, link)
+		drm_mm_scan_remove_block(&e->node);
+
+	for (n = 0; n < total_size; n++) {
+		e = &nodes[n];
+
+		if (!drm_mm_node_allocated(&e->node)) {
+			pr_err("node[%d] no longer allocated!\n", n);
+			return false;
+		}
+
+		e->link.next = NULL;
+	}
+
+	drm_mm_for_each_node(node, mm) {
+		e = container_of(node, typeof(*e), node);
+		e->link.next = &e->link;
+	}
+
+	for (n = 0; n < total_size; n++) {
+		e = &nodes[n];
+
+		if (!e->link.next) {
+			pr_err("node[%d] no longer connected!\n", n);
+			return false;
+		}
+	}
+
+	return assert_continuous(mm, nodes[0].node.size);
+}
+
+static bool evict_everything(struct drm_mm *mm,
+			     unsigned int total_size,
+			     struct evict_node *nodes)
+{
+	LIST_HEAD(evict_list);
+	struct evict_node *e;
+	unsigned int n;
+	int err;
+
+	drm_mm_init_scan(mm, total_size, 0, 0);
+	for (n = 0; n < total_size; n++) {
+		e = &nodes[n];
+		list_add(&e->link, &evict_list);
+		drm_mm_scan_add_block(&e->node);
+	}
+	list_for_each_entry(e, &evict_list, link) {
+		if (!drm_mm_scan_remove_block(&e->node)) {
+			pr_err("Node %lld not marked for eviction!\n",
+			       e->node.start);
+			list_del(&e->link);
+		}
+	}
+
+	list_for_each_entry(e, &evict_list, link)
+		drm_mm_remove_node(&e->node);
+
+	if (!assert_one_hole(mm, 0, total_size))
+		return false;
+
+	list_for_each_entry(e, &evict_list, link) {
+		err = drm_mm_reserve_node(mm, &e->node);
+		if (err) {
+			pr_err("Failed to reinsert node after eviction: start=%llx\n",
+			       e->node.start);
+			return false;
+		}
+	}
+
+	return assert_continuous(mm, nodes[0].node.size);
+}
+
+static int evict_something(struct drm_mm *mm,
+			   struct evict_node *nodes,
+			   unsigned int *order,
+			   unsigned int count,
+			   unsigned int size,
+			   unsigned int alignment,
+			   const struct insert_mode *mode)
+{
+	LIST_HEAD(evict_list);
+	struct evict_node *e;
+	struct drm_mm_node tmp;
+	int err;
+
+	drm_mm_init_scan(mm, size, alignment, 0);
+	if (!evict_nodes(mm,
+			 nodes, order, count,
+			 &evict_list))
+		return -EINVAL;
+
+	memset(&tmp, 0, sizeof(tmp));
+	err = drm_mm_insert_node_generic(mm, &tmp, size, alignment, 0,
+					 mode->search_flags,
+					 mode->create_flags);
+	if (err) {
+		pr_err("Failed to insert into eviction hole: size=%d, align=%d\n",
+		       size, alignment);
+		show_scan(mm);
+		show_holes(mm, 3);
+		return err;
+	}
+
+	if (!assert_node(&tmp, mm, size, alignment, 0) || tmp.hole_follows) {
+		pr_err("Inserted did not fill the eviction hole: size=%lld [%d], align=%d [rem=%lld], start=%llx, hole-follows?=%d\n",
+		       tmp.size, size,
+		       alignment, misalignment(&tmp, alignment),
+		       tmp.start, tmp.hole_follows);
+		err = -EINVAL;
+	}
+
+	drm_mm_remove_node(&tmp);
+	if (err)
+		return err;
+
+	list_for_each_entry(e, &evict_list, link) {
+		err = drm_mm_reserve_node(mm, &e->node);
+		if (err) {
+			pr_err("Failed to reinsert node after eviction: start=%llx\n",
+			       e->node.start);
+			return err;
+		}
+	}
+
+	if (!assert_continuous(mm, nodes[0].node.size)) {
+		pr_err("range is no longer continuous\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int igt_evict(void *ignored)
+{
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int size = 8192;
+	const struct insert_mode *mode;
+	struct drm_mm mm;
+	struct evict_node *nodes;
+	struct drm_mm_node *node, *next;
+	unsigned int *order, n;
+	int ret, err;
+
+	/* Here we populate a full drm_mm and then try and insert a new node
+	 * by evicting other nodes in a random order. The drm_mm_scan should
+	 * pick the first matching hole it finds from the random list. We
+	 * repeat that for different allocation strategies, alignments and
+	 * sizes to try and stress the hole finder.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(size * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	order = drm_random_order(size, &prng);
+	if (!order)
+		goto err_nodes;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, size);
+	for (n = 0; n < size; n++) {
+		err = drm_mm_insert_node(&mm, &nodes[n].node, 1, 0,
+					 DRM_MM_SEARCH_DEFAULT);
+		if (err) {
+			pr_err("insert failed, step %d\n", n);
+			ret = err;
+			goto out;
+		}
+	}
+
+	/* First check that using the scanner doesn't break the mm */
+	if (!evict_nothing(&mm, size, nodes)) {
+		pr_err("evict_nothing() failed\n");
+		goto out;
+	}
+	if (!evict_everything(&mm, size, nodes)) {
+		pr_err("evict_everything() failed\n");
+		goto out;
+	}
+
+	for (mode = evict_modes; mode->name; mode++) {
+		for (n = 1; n <= size; n <<= 1) {
+			drm_random_reorder(order, size, &prng);
+			err = evict_something(&mm,
+					      nodes, order, size,
+					      n, 1,
+					      mode);
+			if (err) {
+				pr_err("%s evict_something(size=%u) failed\n",
+				       mode->name, n);
+				ret = err;
+				goto out;
+			}
+		}
+
+		for (n = 1; n < size; n <<= 1) {
+			drm_random_reorder(order, size, &prng);
+			err = evict_something(&mm,
+					      nodes, order, size,
+					      size/2, n,
+					      mode);
+			if (err) {
+				pr_err("%s evict_something(size=%u, alignment=%u) failed\n",
+				       mode->name, size/2, n);
+				ret = err;
+				goto out;
+			}
+		}
+
+		for_each_prime_number(n, min(size, max_prime)) {
+			unsigned int nsize = (size - n + 1) / 2;
+
+			DRM_MM_BUG_ON(!nsize);
+
+			drm_random_reorder(order, size, &prng);
+			err = evict_something(&mm,
+					      nodes, order, size,
+					      nsize, n,
+					      mode);
+			if (err) {
+				pr_err("%s evict_something(size=%u, alignment=%u) failed\n",
+				       mode->name, nsize, n);
+				ret = err;
+				goto out;
+			}
+		}
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 15/38] drm: kselftest for drm_mm and range restricted eviction

[Chris Wilson]

Check that we add arbitrary blocks to a restrited eviction scanner in
order to find the first minimal hole that matches our request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 113 ++++++++++++++++++++++++++-
 2 files changed, 110 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index a31b4458c7eb..965aca65c160 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -16,3 +16,4 @@ selftest(align, igt_align)
 selftest(align32, igt_align32)
 selftest(align64, igt_align64)
 selftest(evict, igt_evict)
+selftest(evict_range, igt_evict_range)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 10142bc4e07e..f662d98177ee 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1274,6 +1274,7 @@ static bool evict_everything(struct drm_mm *mm,
 }
 
 static int evict_something(struct drm_mm *mm,
+			   u64 range_start, u64 range_end,
 			   struct evict_node *nodes,
 			   unsigned int *order,
 			   unsigned int count,
@@ -1286,7 +1287,9 @@ static int evict_something(struct drm_mm *mm,
 	struct drm_mm_node tmp;
 	int err;
 
-	drm_mm_init_scan(mm, size, alignment, 0);
+	drm_mm_init_scan_with_range(mm,
+				    size, alignment, 0,
+				    range_start, range_end);
 	if (!evict_nodes(mm,
 			 nodes, order, count,
 			 &evict_list))
@@ -1304,6 +1307,12 @@ static int evict_something(struct drm_mm *mm,
 		return err;
 	}
 
+	if (tmp.start < range_start || tmp.start + tmp.size > range_end) {
+		pr_err("Inserted [address=%llu + %llu] did not fit into the request range [%llu, %llu]\n",
+		       tmp.start, tmp.size, range_start, range_end);
+		err = -EINVAL;
+	}
+
 	if (!assert_node(&tmp, mm, size, alignment, 0) || tmp.hole_follows) {
 		pr_err("Inserted did not fill the eviction hole: size=%lld [%d], align=%d [rem=%lld], start=%llx, hole-follows?=%d\n",
 		       tmp.size, size,
@@ -1385,7 +1394,7 @@ static int igt_evict(void *ignored)
 	for (mode = evict_modes; mode->name; mode++) {
 		for (n = 1; n <= size; n <<= 1) {
 			drm_random_reorder(order, size, &prng);
-			err = evict_something(&mm,
+			err = evict_something(&mm, 0, U64_MAX,
 					      nodes, order, size,
 					      n, 1,
 					      mode);
@@ -1399,7 +1408,7 @@ static int igt_evict(void *ignored)
 
 		for (n = 1; n < size; n <<= 1) {
 			drm_random_reorder(order, size, &prng);
-			err = evict_something(&mm,
+			err = evict_something(&mm, 0, U64_MAX,
 					      nodes, order, size,
 					      size/2, n,
 					      mode);
@@ -1417,7 +1426,7 @@ static int igt_evict(void *ignored)
 			DRM_MM_BUG_ON(!nsize);
 
 			drm_random_reorder(order, size, &prng);
-			err = evict_something(&mm,
+			err = evict_something(&mm, 0, U64_MAX,
 					      nodes, order, size,
 					      nsize, n,
 					      mode);
@@ -1442,6 +1451,102 @@ static int igt_evict(void *ignored)
 	return ret;
 }
 
+static int igt_evict_range(void *ignored)
+{
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int size = 8192;
+	const unsigned int range_size = size / 2;
+	const unsigned int range_start = size / 4;
+	const unsigned int range_end = range_start + range_size;
+	const struct insert_mode *mode;
+	struct drm_mm mm;
+	struct evict_node *nodes;
+	struct drm_mm_node *node, *next;
+	unsigned int *order, n;
+	int ret, err;
+
+	/* Like igt_evict() but now we are limiting the search to a
+	 * small portion of the full drm_mm.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(size * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	order = drm_random_order(size, &prng);
+	if (!order)
+		goto err_nodes;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, size);
+	for (n = 0; n < size; n++) {
+		err = drm_mm_insert_node(&mm, &nodes[n].node, 1, 0,
+					 DRM_MM_SEARCH_DEFAULT);
+		if (err) {
+			pr_err("insert failed, step %d\n", n);
+			ret = err;
+			goto out;
+		}
+	}
+
+	for (mode = evict_modes; mode->name; mode++) {
+		for (n = 1; n <= range_size; n <<= 1) {
+			drm_random_reorder(order, size, &prng);
+			err = evict_something(&mm, range_start, range_end,
+					      nodes, order, size,
+					      n, 1,
+					      mode);
+			if (err) {
+				pr_err("%s evict_something(size=%u) failed with range [%u, %u]\n",
+				       mode->name, n, range_start, range_end);
+				goto out;
+			}
+		}
+
+		for (n = 1; n <= range_size; n <<= 1) {
+			drm_random_reorder(order, size, &prng);
+			err = evict_something(&mm, range_start, range_end,
+					      nodes, order, size,
+					      range_size/2, n,
+					      mode);
+			if (err) {
+				pr_err("%s evict_something(size=%u, alignment=%u) failed with range [%u, %u]\n",
+				       mode->name, range_size/2, n, range_start, range_end);
+				goto out;
+			}
+		}
+
+		for_each_prime_number(n, min(range_size, max_prime)) {
+			unsigned int nsize = (range_size - n + 1) / 2;
+
+			DRM_MM_BUG_ON(!nsize);
+
+			drm_random_reorder(order, size, &prng);
+			err = evict_something(&mm, range_start, range_end,
+					      nodes, order, size,
+					      nsize, n,
+					      mode);
+			if (err) {
+				pr_err("%s evict_something(size=%u, alignment=%u) failed with range [%u, %u]\n",
+				       mode->name, nsize, n, range_start, range_end);
+				goto out;
+			}
+		}
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 16/38] drm: kselftest for drm_mm and top-down allocation

[Chris Wilson]

Check that if we request top-down allocation from drm_mm_insert_node()
we receive the next available hole from the top.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 119 +++++++++++++++++++++++++++
 2 files changed, 120 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 965aca65c160..cd508e3d6538 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -17,3 +17,4 @@ selftest(align32, igt_align32)
 selftest(align64, igt_align64)
 selftest(evict, igt_evict)
 selftest(evict_range, igt_evict_range)
+selftest(topdown, igt_topdown)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index f662d98177ee..a499895a76c4 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1547,6 +1547,125 @@ static int igt_evict_range(void *ignored)
 	return ret;
 }
 
+static unsigned int node_index(const struct drm_mm_node *node)
+{
+	return div64_u64(node->start, node->size);
+}
+
+static int igt_topdown(void *ignored)
+{
+	const struct insert_mode *topdown = &insert_modes[TOPDOWN];
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int count = 8192;
+	unsigned int size;
+	unsigned long *bitmap = NULL;
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int *order, n, m, o = 0;
+	int ret;
+
+	/* When allocating top-down, we expect to be returned a node
+	 * from a suitable hole at the top of the drm_mm. We check that
+	 * the returned node does match the highest available slot.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	bitmap = kzalloc(count / BITS_PER_LONG * sizeof(unsigned long),
+			 GFP_TEMPORARY);
+	if (!bitmap)
+		goto err_nodes;
+
+	order = drm_random_order(count, &prng);
+	if (!order)
+		goto err_bitmap;
+
+	ret = -EINVAL;
+	for (size = 1; size <= 64; size <<= 1) {
+		drm_mm_init(&mm, 0, size*count);
+		for (n = 0; n < count; n++) {
+			if (!expect_insert(&mm, &nodes[n],
+					   size, 0, n,
+					   topdown)) {
+				pr_err("insert failed, size %u step %d\n", size, n);
+				goto out;
+			}
+
+			if (nodes[n].hole_follows) {
+				pr_err("hole after topdown insert %d, start=%llx\n, size=%u",
+				       n, nodes[n].start, size);
+				goto out;
+			}
+
+			if (!assert_one_hole(&mm, 0, size*(count - n - 1)))
+				goto out;
+		}
+
+		if (!assert_continuous(&mm, size))
+			goto out;
+
+		drm_random_reorder(order, count, &prng);
+		for_each_prime_number(n, min(count, max_prime)) {
+			for (m = 0; m < n; m++) {
+				node = &nodes[order[(o + m) % count]];
+				drm_mm_remove_node(node);
+				__set_bit(node_index(node), bitmap);
+			}
+
+			for (m = 0; m < n; m++) {
+				unsigned int last;
+
+				node = &nodes[order[(o + m) % count]];
+				if (!expect_insert(&mm, node,
+						   size, 0, 0,
+						   topdown)) {
+					pr_err("insert failed, step %d/%d\n", m, n);
+					goto out;
+				}
+
+				if (node->hole_follows) {
+					pr_err("hole after topdown insert %d/%d, start=%llx\n",
+					       m, n, node->start);
+					goto out;
+				}
+
+				last = find_last_bit(bitmap, count);
+				if (node_index(node) != last) {
+					pr_err("node %d/%d, size %d, not inserted into upmost hole, expected %d, found %d\n",
+					       m, n, size, last, node_index(node));
+					goto out;
+				}
+
+				__clear_bit(last, bitmap);
+			}
+
+			DRM_MM_BUG_ON(find_first_bit(bitmap, count) != count);
+
+			o += n;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_bitmap:
+	kfree(bitmap);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH] drm: kselftest for drm_mm and top-down allocation

[Chris Wilson]

Check that if we request top-down allocation from drm_mm_insert_node()
we receive the next available hole from the top.

v2: Flip sign on conditional assert.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 119 +++++++++++++++++++++++++++
 2 files changed, 120 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 965aca65c160..cd508e3d6538 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -17,3 +17,4 @@ selftest(align32, igt_align32)
 selftest(align64, igt_align64)
 selftest(evict, igt_evict)
 selftest(evict_range, igt_evict_range)
+selftest(topdown, igt_topdown)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 403543c67e52..4530c514016e 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1552,6 +1552,125 @@ static int igt_evict_range(void *ignored)
 	return ret;
 }
 
+static unsigned int node_index(const struct drm_mm_node *node)
+{
+	return div64_u64(node->start, node->size);
+}
+
+static int igt_topdown(void *ignored)
+{
+	const struct insert_mode *topdown = &insert_modes[TOPDOWN];
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int count = 8192;
+	unsigned int size;
+	unsigned long *bitmap = NULL;
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int *order, n, m, o = 0;
+	int ret;
+
+	/* When allocating top-down, we expect to be returned a node
+	 * from a suitable hole at the top of the drm_mm. We check that
+	 * the returned node does match the highest available slot.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	bitmap = kzalloc(count / BITS_PER_LONG * sizeof(unsigned long),
+			 GFP_TEMPORARY);
+	if (!bitmap)
+		goto err_nodes;
+
+	order = drm_random_order(count, &prng);
+	if (!order)
+		goto err_bitmap;
+
+	ret = -EINVAL;
+	for (size = 1; size <= 64; size <<= 1) {
+		drm_mm_init(&mm, 0, size*count);
+		for (n = 0; n < count; n++) {
+			if (!expect_insert(&mm, &nodes[n],
+					   size, 0, n,
+					   topdown)) {
+				pr_err("insert failed, size %u step %d\n", size, n);
+				goto out;
+			}
+
+			if (nodes[n].hole_follows) {
+				pr_err("hole after topdown insert %d, start=%llx\n, size=%u",
+				       n, nodes[n].start, size);
+				goto out;
+			}
+
+			if (!assert_one_hole(&mm, 0, size*(count - n - 1)))
+				goto out;
+		}
+
+		if (!assert_continuous(&mm, size))
+			goto out;
+
+		drm_random_reorder(order, count, &prng);
+		for_each_prime_number(n, min(count, max_prime)) {
+			for (m = 0; m < n; m++) {
+				node = &nodes[order[(o + m) % count]];
+				drm_mm_remove_node(node);
+				__set_bit(node_index(node), bitmap);
+			}
+
+			for (m = 0; m < n; m++) {
+				unsigned int last;
+
+				node = &nodes[order[(o + m) % count]];
+				if (!expect_insert(&mm, node,
+						   size, 0, 0,
+						   topdown)) {
+					pr_err("insert failed, step %d/%d\n", m, n);
+					goto out;
+				}
+
+				if (node->hole_follows) {
+					pr_err("hole after topdown insert %d/%d, start=%llx\n",
+					       m, n, node->start);
+					goto out;
+				}
+
+				last = find_last_bit(bitmap, count);
+				if (node_index(node) != last) {
+					pr_err("node %d/%d, size %d, not inserted into upmost hole, expected %d, found %d\n",
+					       m, n, size, last, node_index(node));
+					goto out;
+				}
+
+				__clear_bit(last, bitmap);
+			}
+
+			DRM_MM_BUG_ON(find_first_bit(bitmap, count) != count);
+
+			o += n;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(!drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_bitmap:
+	kfree(bitmap);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 17/38] drm: kselftest for drm_mm and color adjustment

[Chris Wilson]

Check that after applying the driver's color adjustment, fitting of the
node and its alignment are still correct.

v2: s/no_color_touching/separate_adjacent_colors/

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 183 +++++++++++++++++++++++++++
 2 files changed, 184 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index cd508e3d6538..ff44f39a1826 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -18,3 +18,4 @@ selftest(align64, igt_align64)
 selftest(evict, igt_evict)
 selftest(evict_range, igt_evict_range)
 selftest(topdown, igt_topdown)
+selftest(color, igt_color)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index a499895a76c4..71520a21de8d 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1666,6 +1666,189 @@ static int igt_topdown(void *ignored)
 	return ret;
 }
 
+static void separate_adjacent_colors(const struct drm_mm_node *node,
+				     unsigned long color,
+				     u64 *start,
+				     u64 *end)
+{
+	if (node->allocated && node->color != color)
+		++*start;
+
+	node = list_next_entry(node, node_list);
+	if (node->allocated && node->color != color)
+		--*end;
+}
+
+static bool colors_abutt(const struct drm_mm_node *node)
+{
+	if (!node->hole_follows &&
+	    list_next_entry(node, node_list)->allocated) {
+		pr_err("colors abutt; %ld [%llx + %llx] is next to %ld [%llx + %llx]!\n",
+		       node->color, node->start, node->size,
+		       list_next_entry(node, node_list)->color,
+		       list_next_entry(node, node_list)->start,
+		       list_next_entry(node, node_list)->size);
+		return true;
+	}
+
+	return false;
+}
+
+static int igt_color(void *ignored)
+{
+	const unsigned int count = min(4096u, max_iterations);
+	const struct insert_mode *mode;
+	struct drm_mm mm;
+	struct drm_mm_node *node, *nn;
+	unsigned int n;
+	int ret = -EINVAL, err;
+
+	/* Color adjustment complicates everything. First we just check
+	 * that when we insert a node we apply any color_adjustment callback.
+	 * The callback we use should ensure that there is a gap between
+	 * any two nodes, and so after each insertion we check that those
+	 * holes are inserted and that they are preserved.
+	 */
+
+	drm_mm_init(&mm, 0, U64_MAX);
+
+	for (n = 1; n <= count; n++) {
+		node = kzalloc(sizeof(*node), GFP_KERNEL);
+		if (!node) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		if (!expect_insert(&mm, node,
+				   n, 0, n,
+				   &insert_modes[0])) {
+			pr_err("insert failed, step %d\n", n);
+			kfree(node);
+			goto out;
+		}
+	}
+
+	drm_mm_for_each_node_safe(node, nn, &mm) {
+		if (node->color != node->size) {
+			pr_err("invalid color stored: expected %lld, found %ld\n",
+			       node->size, node->color);
+
+			goto out;
+		}
+
+		drm_mm_remove_node(node);
+		kfree(node);
+	}
+
+	/* Now, let's start experimenting with applying a color callback */
+	mm.color_adjust = separate_adjacent_colors;
+	for (mode = insert_modes; mode->name; mode++) {
+		u64 last;
+
+		node = kzalloc(sizeof(*node), GFP_KERNEL);
+		if (!node) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		node->size = 1 + 2*count;
+		node->color = node->size;
+
+		err = drm_mm_reserve_node(&mm, node);
+		if (err) {
+			pr_err("initial reserve failed!\n");
+			ret = err;
+			goto out;
+		}
+
+		last = node->start + node->size;
+
+		for (n = 1; n <= count; n++) {
+			int rem;
+
+			node = kzalloc(sizeof(*node), GFP_KERNEL);
+			if (!node) {
+				ret = -ENOMEM;
+				goto out;
+			}
+
+			node->start = last;
+			node->size = n + count;
+			node->color = node->size;
+
+			err = drm_mm_reserve_node(&mm, node);
+			if (err != -ENOSPC) {
+				pr_err("reserve %d did not report color overlap! err=%d\n",
+				       n, err);
+				goto out;
+			}
+
+			node->start += n + 1;
+			rem = misalignment(node, n + count);
+			node->start += n + count - rem;
+
+			err = drm_mm_reserve_node(&mm, node);
+			if (err) {
+				pr_err("reserve %d failed, err=%d\n", n, err);
+				ret = err;
+				goto out;
+			}
+
+			last = node->start + node->size;
+		}
+
+		for (n = 1; n <= count; n++) {
+			node = kzalloc(sizeof(*node), GFP_KERNEL);
+			if (!node) {
+				ret = -ENOMEM;
+				goto out;
+			}
+
+			if (!expect_insert(&mm, node,
+					   n, n, n,
+					   mode)) {
+				pr_err("%s insert failed, step %d\n",
+				       mode->name, n);
+				kfree(node);
+				goto out;
+			}
+		}
+
+		drm_mm_for_each_node_safe(node, nn, &mm) {
+			u64 rem;
+
+			if (node->color != node->size) {
+				pr_err("%s invalid color stored: expected %lld, found %ld\n",
+				       mode->name, node->size, node->color);
+
+				goto out;
+			}
+
+			if (colors_abutt(node))
+				goto out;
+
+			div64_u64_rem(node->start, node->size, &rem);
+			if (rem) {
+				pr_err("%s colored node misaligned, start=%llx expected alignment=%lld [rem=%lld]\n",
+				       mode->name, node->start, node->size, rem);
+				goto out;
+			}
+
+			drm_mm_remove_node(node);
+			kfree(node);
+		}
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, nn, &mm) {
+		drm_mm_remove_node(node);
+		kfree(node);
+	}
+	drm_mm_takedown(&mm);
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 18/38] drm: kselftest for drm_mm and color eviction

[Chris Wilson]

Check that after applying the driver's color adjustment, eviction
scanning find a suitable hole.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 156 +++++++++++++++++++++++++++
 2 files changed, 157 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index ff44f39a1826..0a3a7e32e5f7 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -19,3 +19,4 @@ selftest(evict, igt_evict)
 selftest(evict_range, igt_evict_range)
 selftest(topdown, igt_topdown)
 selftest(color, igt_color)
+selftest(color_evict, igt_color_evict)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 71520a21de8d..9dd94da15bd5 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1849,6 +1849,162 @@ static int igt_color(void *ignored)
 	return ret;
 }
 
+static int evict_color(struct drm_mm *mm,
+		       struct evict_node *nodes,
+		       unsigned int *order,
+		       unsigned int count,
+		       unsigned int size,
+		       unsigned int alignment,
+		       unsigned long color,
+		       const struct insert_mode *mode)
+{
+	LIST_HEAD(evict_list);
+	struct evict_node *e;
+	struct drm_mm_node tmp;
+	int err;
+
+	drm_mm_init_scan(mm, size, alignment, color);
+	if (!evict_nodes(mm,
+			 nodes, order, count,
+			 &evict_list))
+		return -EINVAL;
+
+	memset(&tmp, 0, sizeof(tmp));
+	err = drm_mm_insert_node_generic(mm, &tmp, size, alignment, color,
+					 mode->search_flags,
+					 mode->create_flags);
+	if (err) {
+		pr_err("Failed to insert into eviction hole: size=%d, align=%d, color=%lu, err=%d\n",
+		       size, alignment, color, err);
+		show_scan(mm);
+		show_holes(mm, 3);
+		return err;
+	}
+
+	if (colors_abutt(&tmp))
+		err = -EINVAL;
+
+	if (!assert_node(&tmp, mm, size, alignment, color)) {
+		pr_err("Inserted did not fit the eviction hole: size=%lld [%d], align=%d [rem=%lld], start=%llx\n",
+		       tmp.size, size,
+		       alignment, misalignment(&tmp, alignment), tmp.start);
+		err = -EINVAL;
+	}
+
+	drm_mm_remove_node(&tmp);
+	if (err)
+		return err;
+
+	list_for_each_entry(e, &evict_list, link) {
+		err = drm_mm_reserve_node(mm, &e->node);
+		if (err) {
+			pr_err("Failed to reinsert node after eviction: start=%llx\n",
+			       e->node.start);
+			return err;
+		}
+	}
+
+	return 0;
+}
+
+static int igt_color_evict(void *ignored)
+{
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int total_size = min(8192u, max_iterations);
+	const struct insert_mode *mode;
+	unsigned long color = 0;
+	struct drm_mm mm;
+	struct evict_node *nodes;
+	struct drm_mm_node *node, *next;
+	unsigned int *order, n;
+	int ret, err;
+
+	/* Check that the drm_mm_scan also honours color adjustment when
+	 * choosing its victims to create a hole. Our color_adjust does not
+	 * allow two nodes to be placed together without an intervening hole
+	 * enlarging the set of victims that must be evicted.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(total_size * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	order = drm_random_order(total_size, &prng);
+	if (!order)
+		goto err_nodes;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, 2*total_size - 1);
+	mm.color_adjust = separate_adjacent_colors;
+	for (n = 0; n < total_size; n++) {
+		if (!expect_insert(&mm, &nodes[n].node,
+				   1, 0, color++,
+				   &insert_modes[0])) {
+			pr_err("insert failed, step %d\n", n);
+			goto out;
+		}
+	}
+
+	for (mode = evict_modes; mode->name; mode++) {
+		for (n = 1; n <= total_size; n <<= 1) {
+			drm_random_reorder(order, total_size, &prng);
+			err = evict_color(&mm,
+					  nodes, order, total_size,
+					  n, 1, color++,
+					  mode);
+			if (err) {
+				pr_err("%s evict_color(size=%u) failed\n",
+				       mode->name, n);
+				goto out;
+			}
+		}
+
+		for (n = 1; n < total_size; n <<= 1) {
+			drm_random_reorder(order, total_size, &prng);
+			err = evict_color(&mm,
+					  nodes, order, total_size,
+					  total_size/2, n, color++,
+					  mode);
+			if (err) {
+				pr_err("%s evict_color(size=%u, alignment=%u) failed\n",
+				       mode->name, total_size/2, n);
+				goto out;
+			}
+		}
+
+		for_each_prime_number(n, min(total_size, max_prime)) {
+			unsigned int nsize = (total_size - n + 1) / 2;
+
+			DRM_MM_BUG_ON(!nsize);
+
+			drm_random_reorder(order, total_size, &prng);
+			err = evict_color(&mm,
+					  nodes, order, total_size,
+					  nsize, n, color++,
+					  mode);
+			if (err) {
+				pr_err("%s evict_color(size=%u, alignment=%u) failed\n",
+				       mode->name, nsize, n);
+				goto out;
+			}
+		}
+	}
+
+	ret = 0;
+out:
+	if (ret)
+		drm_mm_debug_table(&mm, __func__);
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 19/38] drm: kselftest for drm_mm and restricted color eviction

[Chris Wilson]

Check that after applying the driver's color adjustment, restricted
eviction scanning find a suitable hole.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 116 ++++++++++++++++++++++++++-
 2 files changed, 113 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 0a3a7e32e5f7..6a4575fdc1c0 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -20,3 +20,4 @@ selftest(evict_range, igt_evict_range)
 selftest(topdown, igt_topdown)
 selftest(color, igt_color)
 selftest(color_evict, igt_color_evict)
+selftest(color_evict_range, igt_color_evict_range)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 9dd94da15bd5..7e90886f5558 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1850,6 +1850,7 @@ static int igt_color(void *ignored)
 }
 
 static int evict_color(struct drm_mm *mm,
+		       u64 range_start, u64 range_end,
 		       struct evict_node *nodes,
 		       unsigned int *order,
 		       unsigned int count,
@@ -1863,7 +1864,9 @@ static int evict_color(struct drm_mm *mm,
 	struct drm_mm_node tmp;
 	int err;
 
-	drm_mm_init_scan(mm, size, alignment, color);
+	drm_mm_init_scan_with_range(mm,
+				    size, alignment, color,
+				    range_start, range_end);
 	if (!evict_nodes(mm,
 			 nodes, order, count,
 			 &evict_list))
@@ -1881,6 +1884,12 @@ static int evict_color(struct drm_mm *mm,
 		return err;
 	}
 
+	if (tmp.start < range_start || tmp.start + tmp.size > range_end) {
+		pr_err("Inserted [address=%llu + %llu] did not fit into the request range [%llu, %llu]\n",
+		       tmp.start, tmp.size, range_start, range_end);
+		err = -EINVAL;
+	}
+
 	if (colors_abutt(&tmp))
 		err = -EINVAL;
 
@@ -1949,7 +1958,7 @@ static int igt_color_evict(void *ignored)
 	for (mode = evict_modes; mode->name; mode++) {
 		for (n = 1; n <= total_size; n <<= 1) {
 			drm_random_reorder(order, total_size, &prng);
-			err = evict_color(&mm,
+			err = evict_color(&mm, 0, U64_MAX,
 					  nodes, order, total_size,
 					  n, 1, color++,
 					  mode);
@@ -1962,7 +1971,7 @@ static int igt_color_evict(void *ignored)
 
 		for (n = 1; n < total_size; n <<= 1) {
 			drm_random_reorder(order, total_size, &prng);
-			err = evict_color(&mm,
+			err = evict_color(&mm, 0, U64_MAX,
 					  nodes, order, total_size,
 					  total_size/2, n, color++,
 					  mode);
@@ -1979,7 +1988,7 @@ static int igt_color_evict(void *ignored)
 			DRM_MM_BUG_ON(!nsize);
 
 			drm_random_reorder(order, total_size, &prng);
-			err = evict_color(&mm,
+			err = evict_color(&mm, 0, U64_MAX,
 					  nodes, order, total_size,
 					  nsize, n, color++,
 					  mode);
@@ -2005,6 +2014,105 @@ static int igt_color_evict(void *ignored)
 	return ret;
 }
 
+static int igt_color_evict_range(void *ignored)
+{
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int total_size = 8192;
+	const unsigned int range_size = total_size / 2;
+	const unsigned int range_start = total_size / 4;
+	const unsigned int range_end = range_start + range_size;
+	const struct insert_mode *mode;
+	unsigned long color = 0;
+	struct drm_mm mm;
+	struct evict_node *nodes;
+	struct drm_mm_node *node, *next;
+	unsigned int *order, n;
+	int ret, err;
+
+	/* Like igt_color_evict(), but limited to small portion of the full
+	 * drm_mm range.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(total_size * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	order = drm_random_order(total_size, &prng);
+	if (!order)
+		goto err_nodes;
+
+	ret = -EINVAL;
+	drm_mm_init(&mm, 0, 2*total_size - 1);
+	mm.color_adjust = separate_adjacent_colors;
+	for (n = 0; n < total_size; n++) {
+		if (!expect_insert(&mm, &nodes[n].node,
+				   1, 0, color++,
+				   &insert_modes[0])) {
+			pr_err("insert failed, step %d\n", n);
+			goto out;
+		}
+	}
+
+	for (mode = evict_modes; mode->name; mode++) {
+		for (n = 1; n <= range_size; n <<= 1) {
+			drm_random_reorder(order, range_size, &prng);
+			err = evict_color(&mm, range_start, range_end,
+					  nodes, order, total_size,
+					  n, 1, color++,
+					  mode);
+			if (err) {
+				pr_err("%s evict_color(size=%u) failed for range [%x, %x]\n",
+				       mode->name, n, range_start, range_end);
+				goto out;
+			}
+		}
+
+		for (n = 1; n < range_size; n <<= 1) {
+			drm_random_reorder(order, total_size, &prng);
+			err = evict_color(&mm, range_start, range_end,
+					  nodes, order, total_size,
+					  range_size/2, n, color++,
+					  mode);
+			if (err) {
+				pr_err("%s evict_color(size=%u, alignment=%u) failed for range [%x, %x]\n",
+				       mode->name, total_size/2, n, range_start, range_end);
+				goto out;
+			}
+		}
+
+		for_each_prime_number(n, min(range_size, max_prime)) {
+			unsigned int nsize = (range_size - n + 1) / 2;
+
+			DRM_MM_BUG_ON(!nsize);
+
+			drm_random_reorder(order, total_size, &prng);
+			err = evict_color(&mm, range_start, range_end,
+					  nodes, order, total_size,
+					  nsize, n, color++,
+					  mode);
+			if (err) {
+				pr_err("%s evict_color(size=%u, alignment=%u) failed for range [%x, %x]\n",
+				       mode->name, nsize, n, range_start, range_end);
+				goto out;
+			}
+		}
+	}
+
+	ret = 0;
+out:
+	if (ret)
+		drm_mm_debug_table(&mm, __func__);
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 #include "drm_selftest.c"
 
 static int __init test_drm_mm_init(void)
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 20/38] drm/i915: Build DRM range manager selftests for CI

[Chris Wilson]

Build the struct drm_mm selftests so that we can trivially run them
within our CI.

"Enable debug, become developer." - Joonas Lahtinen

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/i915/Kconfig.debug | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/i915/Kconfig.debug b/drivers/gpu/drm/i915/Kconfig.debug
index 597648c7a645..598551dbf62c 100644
--- a/drivers/gpu/drm/i915/Kconfig.debug
+++ b/drivers/gpu/drm/i915/Kconfig.debug
@@ -24,6 +24,7 @@ config DRM_I915_DEBUG
         select X86_MSR # used by igt/pm_rpm
         select DRM_VGEM # used by igt/prime_vgem (dmabuf interop checks)
         select DRM_DEBUG_MM if DRM=y
+	select DRM_DEBUG_MM_SELFTEST
 	select DRM_I915_SW_FENCE_DEBUG_OBJECTS
         default n
         help
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 21/38] drm: Promote drm_mm alignment to u64

[Chris Wilson]

In places (e.g. i915.ko), the alignment is exported to userspace as u64
and there now exists hardware for which we can indeed utilize a u64
alignment. As such, we need to keep 64bit integers throughout when
handling alignment.

Testcase: igt/drm_mm/align64
Testcase: igt/gem_exec_alignment
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/drm_mm.c                | 37 +++++++++++++++------------------
 drivers/gpu/drm/selftests/test-drm_mm.c |  4 ++--
 include/drm/drm_mm.h                    | 16 +++++++-------
 3 files changed, 27 insertions(+), 30 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 9756542abe4c..fd2667052c73 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -93,12 +93,12 @@
 
 static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
 						u64 size,
-						unsigned alignment,
+						u64 alignment,
 						unsigned long color,
 						enum drm_mm_search_flags flags);
 static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
 						u64 size,
-						unsigned alignment,
+						u64 alignment,
 						unsigned long color,
 						u64 start,
 						u64 end,
@@ -227,7 +227,7 @@ static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node,
 
 static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 				 struct drm_mm_node *node,
-				 u64 size, unsigned alignment,
+				 u64 size, u64 alignment,
 				 unsigned long color,
 				 enum drm_mm_allocator_flags flags)
 {
@@ -246,10 +246,9 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 		adj_start = adj_end - size;
 
 	if (alignment) {
-		u64 tmp = adj_start;
-		unsigned rem;
+		u64 rem;
 
-		rem = do_div(tmp, alignment);
+		div64_u64_rem(adj_start, alignment, &rem);
 		if (rem) {
 			if (flags & DRM_MM_CREATE_TOP)
 				adj_start -= rem;
@@ -376,7 +375,7 @@ EXPORT_SYMBOL(drm_mm_reserve_node);
  * 0 on success, -ENOSPC if there's no suitable hole.
  */
 int drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
-			       u64 size, unsigned alignment,
+			       u64 size, u64 alignment,
 			       unsigned long color,
 			       enum drm_mm_search_flags sflags,
 			       enum drm_mm_allocator_flags aflags)
@@ -398,7 +397,7 @@ EXPORT_SYMBOL(drm_mm_insert_node_generic);
 
 static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
 				       struct drm_mm_node *node,
-				       u64 size, unsigned alignment,
+				       u64 size, u64 alignment,
 				       unsigned long color,
 				       u64 start, u64 end,
 				       enum drm_mm_allocator_flags flags)
@@ -423,10 +422,9 @@ static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
 		adj_start = adj_end - size;
 
 	if (alignment) {
-		u64 tmp = adj_start;
-		unsigned rem;
+		u64 rem;
 
-		rem = do_div(tmp, alignment);
+		div64_u64_rem(adj_start, alignment, &rem);
 		if (rem) {
 			if (flags & DRM_MM_CREATE_TOP)
 				adj_start -= rem;
@@ -482,7 +480,7 @@ static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
  * 0 on success, -ENOSPC if there's no suitable hole.
  */
 int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *node,
-					u64 size, unsigned alignment,
+					u64 size, u64 alignment,
 					unsigned long color,
 					u64 start, u64 end,
 					enum drm_mm_search_flags sflags,
@@ -548,16 +546,15 @@ void drm_mm_remove_node(struct drm_mm_node *node)
 }
 EXPORT_SYMBOL(drm_mm_remove_node);
 
-static int check_free_hole(u64 start, u64 end, u64 size, unsigned alignment)
+static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment)
 {
 	if (end - start < size)
 		return 0;
 
 	if (alignment) {
-		u64 tmp = start;
-		unsigned rem;
+		u64 rem;
 
-		rem = do_div(tmp, alignment);
+		div64_u64_rem(start, alignment, &rem);
 		if (rem)
 			start += alignment - rem;
 	}
@@ -567,7 +564,7 @@ static int check_free_hole(u64 start, u64 end, u64 size, unsigned alignment)
 
 static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
 						      u64 size,
-						      unsigned alignment,
+						      u64 alignment,
 						      unsigned long color,
 						      enum drm_mm_search_flags flags)
 {
@@ -609,7 +606,7 @@ static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
 
 static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
 							u64 size,
-							unsigned alignment,
+							u64 alignment,
 							unsigned long color,
 							u64 start,
 							u64 end,
@@ -729,7 +726,7 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  */
 void drm_mm_init_scan(struct drm_mm *mm,
 		      u64 size,
-		      unsigned alignment,
+		      u64 alignment,
 		      unsigned long color)
 {
 	mm->scan_color = color;
@@ -762,7 +759,7 @@ EXPORT_SYMBOL(drm_mm_init_scan);
  */
 void drm_mm_init_scan_with_range(struct drm_mm *mm,
 				 u64 size,
-				 unsigned alignment,
+				 u64 alignment,
 				 unsigned long color,
 				 u64 start,
 				 u64 end)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 7e90886f5558..ea21c477837c 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1111,7 +1111,7 @@ static int igt_align64(void *ignored)
 
 static void show_scan(const struct drm_mm *scan)
 {
-	pr_info("scan: hit [%llx, %llx], size=%lld, align=%d, color=%ld\n",
+	pr_info("scan: hit [%llx, %llx], size=%lld, align=%lld, color=%ld\n",
 		scan->scan_hit_start, scan->scan_hit_end,
 		scan->scan_size, scan->scan_alignment, scan->scan_color);
 }
@@ -1173,7 +1173,7 @@ static bool evict_nodes(struct drm_mm *mm,
 			list_del(&e->link);
 	}
 	if (list_empty(evict_list)) {
-		pr_err("Failed to find eviction: size=%lld [avail=%d], align=%d (color=%lu)\n",
+		pr_err("Failed to find eviction: size=%lld [avail=%d], align=%lld (color=%lu)\n",
 		       mm->scan_size, count,
 		       mm->scan_alignment,
 		       mm->scan_color);
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 525543019896..0ce8c3678c11 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -98,12 +98,12 @@ struct drm_mm {
 	struct rb_root interval_tree;
 
 	unsigned int scan_check_range : 1;
-	unsigned scan_alignment;
+	unsigned int scanned_blocks;
 	unsigned long scan_color;
+	u64 scan_alignment;
 	u64 scan_size;
 	u64 scan_hit_start;
 	u64 scan_hit_end;
-	unsigned scanned_blocks;
 	u64 scan_start;
 	u64 scan_end;
 	struct drm_mm_node *prev_scanned_node;
@@ -261,7 +261,7 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node);
 int drm_mm_insert_node_generic(struct drm_mm *mm,
 			       struct drm_mm_node *node,
 			       u64 size,
-			       unsigned alignment,
+			       u64 alignment,
 			       unsigned long color,
 			       enum drm_mm_search_flags sflags,
 			       enum drm_mm_allocator_flags aflags);
@@ -284,7 +284,7 @@ int drm_mm_insert_node_generic(struct drm_mm *mm,
 static inline int drm_mm_insert_node(struct drm_mm *mm,
 				     struct drm_mm_node *node,
 				     u64 size,
-				     unsigned alignment,
+				     u64 alignment,
 				     enum drm_mm_search_flags flags)
 {
 	return drm_mm_insert_node_generic(mm, node, size, alignment, 0, flags,
@@ -294,7 +294,7 @@ static inline int drm_mm_insert_node(struct drm_mm *mm,
 int drm_mm_insert_node_in_range_generic(struct drm_mm *mm,
 					struct drm_mm_node *node,
 					u64 size,
-					unsigned alignment,
+					u64 alignment,
 					unsigned long color,
 					u64 start,
 					u64 end,
@@ -321,7 +321,7 @@ int drm_mm_insert_node_in_range_generic(struct drm_mm *mm,
 static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
 					      struct drm_mm_node *node,
 					      u64 size,
-					      unsigned alignment,
+					      u64 alignment,
 					      u64 start,
 					      u64 end,
 					      enum drm_mm_search_flags flags)
@@ -361,11 +361,11 @@ __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
 
 void drm_mm_init_scan(struct drm_mm *mm,
 		      u64 size,
-		      unsigned alignment,
+		      u64 alignment,
 		      unsigned long color);
 void drm_mm_init_scan_with_range(struct drm_mm *mm,
 				 u64 size,
-				 unsigned alignment,
+				 u64 alignment,
 				 unsigned long color,
 				 u64 start,
 				 u64 end);
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 22/38] drm: Fix kerneldoc for drm_mm_scan_remove_block()

[Chris Wilson]

The nodes must be removed in the *reverse* order. This is correct in the
overview, but backwards in the function description. Whilst here add
Intel's copyright statement and tweak some formatting.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 34 ++++++++++++++++++----------------
 include/drm/drm_mm.h     | 19 +++++++++++++------
 2 files changed, 31 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index fd2667052c73..767cfd05c628 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -1,6 +1,7 @@
 /**************************************************************************
  *
  * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA.
+ * Copyright 2016 Intel Corporation
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -31,9 +32,9 @@
  * class implementation for more advanced memory managers.
  *
  * Note that the algorithm used is quite simple and there might be substantial
- * performance gains if a smarter free list is implemented. Currently it is just an
- * unordered stack of free regions. This could easily be improved if an RB-tree
- * is used instead. At least if we expect heavy fragmentation.
+ * performance gains if a smarter free list is implemented. Currently it is
+ * just an unordered stack of free regions. This could easily be improved if
+ * an RB-tree is used instead. At least if we expect heavy fragmentation.
  *
  * Aligned allocations can also see improvement.
  *
@@ -67,7 +68,7 @@
  * where an object needs to be created which exactly matches the firmware's
  * scanout target. As long as the range is still free it can be inserted anytime
  * after the allocator is initialized, which helps with avoiding looped
- * depencies in the driver load sequence.
+ * dependencies in the driver load sequence.
  *
  * drm_mm maintains a stack of most recently freed holes, which of all
  * simplistic datastructures seems to be a fairly decent approach to clustering
@@ -78,14 +79,14 @@
  *
  * drm_mm supports a few features: Alignment and range restrictions can be
  * supplied. Further more every &drm_mm_node has a color value (which is just an
- * opaqua unsigned long) which in conjunction with a driver callback can be used
+ * opaque unsigned long) which in conjunction with a driver callback can be used
  * to implement sophisticated placement restrictions. The i915 DRM driver uses
  * this to implement guard pages between incompatible caching domains in the
  * graphics TT.
  *
- * Two behaviors are supported for searching and allocating: bottom-up and top-down.
- * The default is bottom-up. Top-down allocation can be used if the memory area
- * has different restrictions, or just to reduce fragmentation.
+ * Two behaviors are supported for searching and allocating: bottom-up and
+ * top-down. The default is bottom-up. Top-down allocation can be used if the
+ * memory area has different restrictions, or just to reduce fragmentation.
  *
  * Finally iteration helpers to walk all nodes and all holes are provided as are
  * some basic allocator dumpers for debugging.
@@ -510,7 +511,7 @@ EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic);
  *
  * This just removes a node from its drm_mm allocator. The node does not need to
  * be cleared again before it can be re-inserted into this or any other drm_mm
- * allocator. It is a bug to call this function on a un-allocated node.
+ * allocator. It is a bug to call this function on a unallocated node.
  */
 void drm_mm_remove_node(struct drm_mm_node *node)
 {
@@ -689,16 +690,16 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  * efficient when we simply start to select all objects from the tail of an LRU
  * until there's a suitable hole: Especially for big objects or nodes that
  * otherwise have special allocation constraints there's a good chance we evict
- * lots of (smaller) objects unecessarily.
+ * lots of (smaller) objects unnecessarily.
  *
  * The DRM range allocator supports this use-case through the scanning
  * interfaces. First a scan operation needs to be initialized with
- * drm_mm_init_scan() or drm_mm_init_scan_with_range(). The the driver adds
+ * drm_mm_init_scan() or drm_mm_init_scan_with_range(). The driver adds
  * objects to the roaster (probably by walking an LRU list, but this can be
  * freely implemented) until a suitable hole is found or there's no further
- * evitable object.
+ * evictable object.
  *
- * The the driver must walk through all objects again in exactly the reverse
+ * The driver must walk through all objects again in exactly the reverse
  * order to restore the allocator state. Note that while the allocator is used
  * in the scan mode no other operation is allowed.
  *
@@ -838,9 +839,10 @@ EXPORT_SYMBOL(drm_mm_scan_add_block);
  * drm_mm_scan_remove_block - remove a node from the scan list
  * @node: drm_mm_node to remove
  *
- * Nodes _must_ be removed in the exact same order from the scan list as they
- * have been added, otherwise the internal state of the memory manager will be
- * corrupted.
+ * Nodes _must_ be removed in exactly the reverse order from the scan list as
+ * they have been added (e.g. using list_add as they are added and then
+ * list_for_each over that eviction list to remove), otherwise the internal
+ * state of the memory manager will be corrupted.
  *
  * When the scan list is empty, the selected memory nodes can be freed. An
  * immediately following drm_mm_search_free with !DRM_MM_SEARCH_BEST will then
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 0ce8c3678c11..7eeb98b5bf70 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -1,6 +1,7 @@
 /**************************************************************************
  *
  * Copyright 2006-2008 Tungsten Graphics, Inc., Cedar Park, TX. USA.
+ * Copyright 2016 Intel Corporation
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -117,7 +118,10 @@ struct drm_mm {
  * drm_mm_node_allocated - checks whether a node is allocated
  * @node: drm_mm_node to check
  *
- * Drivers should use this helpers for proper encapusulation of drm_mm
+ * Drivers are required to clear a node prior to using it with the
+ * drm_mm range manager.
+ *
+ * Drivers should use this helper for proper encapsulation of drm_mm
  * internals.
  *
  * Returns:
@@ -132,7 +136,10 @@ static inline bool drm_mm_node_allocated(const struct drm_mm_node *node)
  * drm_mm_initialized - checks whether an allocator is initialized
  * @mm: drm_mm to check
  *
- * Drivers should use this helpers for proper encapusulation of drm_mm
+ * Drivers should clear the struct drm_mm prior to initialisation if they
+ * want to use this function.
+ *
+ * Drivers should use this helper for proper encapsulation of drm_mm
  * internals.
  *
  * Returns:
@@ -152,8 +159,8 @@ static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
  * drm_mm_hole_node_start - computes the start of the hole following @node
  * @hole_node: drm_mm_node which implicitly tracks the following hole
  *
- * This is useful for driver-sepific debug dumpers. Otherwise drivers should not
- * inspect holes themselves. Drivers must check first whether a hole indeed
+ * This is useful for driver-specific debug dumpers. Otherwise drivers should
+ * not inspect holes themselves. Drivers must check first whether a hole indeed
  * follows by looking at node->hole_follows.
  *
  * Returns:
@@ -174,8 +181,8 @@ static inline u64 __drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  * drm_mm_hole_node_end - computes the end of the hole following @node
  * @hole_node: drm_mm_node which implicitly tracks the following hole
  *
- * This is useful for driver-sepific debug dumpers. Otherwise drivers should not
- * inspect holes themselves. Drivers must check first whether a hole indeed
+ * This is useful for driver-specific debug dumpers. Otherwise drivers should
+ * not inspect holes themselves. Drivers must check first whether a hole indeed
  * follows by looking at node->hole_follows.
  *
  * Returns:
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 23/38] drm: Detect overflow in drm_mm_reserve_node()

[Chris Wilson]

Protect ourselves from a caller passing in node.start + node.size that
will overflow and trick us into reserving that node.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 767cfd05c628..370cb8ee91c9 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -308,10 +308,9 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 	u64 hole_start, hole_end;
 	u64 adj_start, adj_end;
 
-	if (WARN_ON(node->size == 0))
-		return -EINVAL;
-
 	end = node->start + node->size;
+	if (unlikely(end <= node->start))
+		return -ENOSPC;
 
 	/* Find the relevant hole to add our node to */
 	hole = drm_mm_interval_tree_iter_first(&mm->interval_tree,
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 24/38] drm: Simplify drm_mm_clean()

[Chris Wilson]

Since commit ea7b1dd44867 ("drm: mm: track free areas implicitly"),
to test whether there are any nodes allocated within the range manager,
we merely have to ask whether the node_list is empty.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 19 +------------------
 include/drm/drm_mm.h     | 14 +++++++++++++-
 2 files changed, 14 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 370cb8ee91c9..e0419cf09bbb 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -873,22 +873,6 @@ bool drm_mm_scan_remove_block(struct drm_mm_node *node)
 EXPORT_SYMBOL(drm_mm_scan_remove_block);
 
 /**
- * drm_mm_clean - checks whether an allocator is clean
- * @mm: drm_mm allocator to check
- *
- * Returns:
- * True if the allocator is completely free, false if there's still a node
- * allocated in it.
- */
-bool drm_mm_clean(const struct drm_mm *mm)
-{
-	const struct list_head *head = drm_mm_nodes(mm);
-
-	return (head->next->next == head);
-}
-EXPORT_SYMBOL(drm_mm_clean);
-
-/**
  * drm_mm_init - initialize a drm-mm allocator
  * @mm: the drm_mm structure to initialize
  * @start: start of the range managed by @mm
@@ -928,10 +912,9 @@ EXPORT_SYMBOL(drm_mm_init);
  */
 void drm_mm_takedown(struct drm_mm *mm)
 {
-	if (WARN(!list_empty(drm_mm_nodes(mm)),
+	if (WARN(!drm_mm_clean(mm),
 		 "Memory manager not clean during takedown.\n"))
 		show_leaks(mm);
-
 }
 EXPORT_SYMBOL(drm_mm_takedown);
 
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 7eeb98b5bf70..72e0c0ddf8d0 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -342,7 +342,19 @@ void drm_mm_remove_node(struct drm_mm_node *node);
 void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new);
 void drm_mm_init(struct drm_mm *mm, u64 start, u64 size);
 void drm_mm_takedown(struct drm_mm *mm);
-bool drm_mm_clean(const struct drm_mm *mm);
+
+/**
+ * drm_mm_clean - checks whether an allocator is clean
+ * @mm: drm_mm allocator to check
+ *
+ * Returns:
+ * True if the allocator is completely free, false if there's still a node
+ * allocated in it.
+ */
+static inline bool drm_mm_clean(const struct drm_mm *mm)
+{
+	return list_empty(drm_mm_nodes(mm));
+}
 
 struct drm_mm_node *
 __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 25/38] drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()

[Chris Wilson]

A simple assert to ensure that we don't overflow start + size when
initialising the drm_mm, or its scanner.

In future, we may want to switch to tracking the value of ranges (rather
than size) so that we can cover the full u64, for example like resource
tracking.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index e0419cf09bbb..b80305484124 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -729,6 +729,8 @@ void drm_mm_init_scan(struct drm_mm *mm,
 		      u64 alignment,
 		      unsigned long color)
 {
+	DRM_MM_BUG_ON(!size);
+
 	mm->scan_color = color;
 	mm->scan_alignment = alignment;
 	mm->scan_size = size;
@@ -764,6 +766,9 @@ void drm_mm_init_scan_with_range(struct drm_mm *mm,
 				 u64 start,
 				 u64 end)
 {
+	DRM_MM_BUG_ON(start >= end);
+	DRM_MM_BUG_ON(!size || size > end - start);
+
 	mm->scan_color = color;
 	mm->scan_alignment = alignment;
 	mm->scan_size = size;
@@ -882,6 +887,8 @@ EXPORT_SYMBOL(drm_mm_scan_remove_block);
  */
 void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 {
+	DRM_MM_BUG_ON(start + size <= start);
+
 	INIT_LIST_HEAD(&mm->hole_stack);
 	mm->scanned_blocks = 0;
 
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 26/38] drm: Extract struct drm_mm_scan from struct drm_mm

[Chris Wilson]

The scan state occupies a large proportion of the struct drm_mm and is
rarely used and only contains temporary state. That makes it suitable to
moving to its struct and onto the stack of the callers.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c                | 124 ++++++++++++++++++--------------
 drivers/gpu/drm/etnaviv/etnaviv_mmu.c   |   7 +-
 drivers/gpu/drm/i915/i915_gem_evict.c   |  19 +++--
 drivers/gpu/drm/selftests/test-drm_mm.c |  45 ++++++------
 include/drm/drm_mm.h                    |  43 +++++++----
 5 files changed, 138 insertions(+), 100 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index b80305484124..21bd2e13738b 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -574,7 +574,7 @@ static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
 	u64 adj_end;
 	u64 best_size;
 
-	DRM_MM_BUG_ON(mm->scanned_blocks);
+	DRM_MM_BUG_ON(mm->scan_active);
 
 	best = NULL;
 	best_size = ~0UL;
@@ -618,7 +618,7 @@ static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_
 	u64 adj_end;
 	u64 best_size;
 
-	DRM_MM_BUG_ON(mm->scanned_blocks);
+	DRM_MM_BUG_ON(mm->scan_active);
 
 	best = NULL;
 	best_size = ~0UL;
@@ -693,7 +693,7 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  *
  * The DRM range allocator supports this use-case through the scanning
  * interfaces. First a scan operation needs to be initialized with
- * drm_mm_init_scan() or drm_mm_init_scan_with_range(). The driver adds
+ * drm_mm_scan_init() or drm_mm_scan_init_with_range(). The driver adds
  * objects to the roaster (probably by walking an LRU list, but this can be
  * freely implemented) until a suitable hole is found or there's no further
  * evictable object.
@@ -710,7 +710,8 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  */
 
 /**
- * drm_mm_init_scan - initialize lru scanning
+ * drm_mm_scan_init - initialize lru scanning
+ * @scan: scan state
  * @mm: drm_mm to scan
  * @size: size of the allocation
  * @alignment: alignment of the allocation
@@ -724,26 +725,33 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  * As long as the scan list is non-empty, no other operations than
  * adding/removing nodes to/from the scan list are allowed.
  */
-void drm_mm_init_scan(struct drm_mm *mm,
+void drm_mm_scan_init(struct drm_mm_scan *scan,
+		      struct drm_mm *mm,
 		      u64 size,
 		      u64 alignment,
 		      unsigned long color)
 {
 	DRM_MM_BUG_ON(!size);
+	DRM_MM_BUG_ON(mm->scan_active);
 
-	mm->scan_color = color;
-	mm->scan_alignment = alignment;
-	mm->scan_size = size;
-	mm->scanned_blocks = 0;
-	mm->scan_hit_start = 0;
-	mm->scan_hit_end = 0;
-	mm->scan_check_range = 0;
-	mm->prev_scanned_node = NULL;
+	scan->mm = mm;
+
+	scan->color = color;
+	scan->alignment = alignment;
+	scan->size = size;
+
+	scan->check_range = 0;
+
+	scan->hit_start = U64_MAX;
+	scan->hit_end = 0;
+
+	scan->prev_scanned_node = NULL;
 }
-EXPORT_SYMBOL(drm_mm_init_scan);
+EXPORT_SYMBOL(drm_mm_scan_init);
 
 /**
- * drm_mm_init_scan - initialize range-restricted lru scanning
+ * drm_mm_scan_init_with_range - initialize range-restricted lru scanning
+ * @scan: scan state
  * @mm: drm_mm to scan
  * @size: size of the allocation
  * @alignment: alignment of the allocation
@@ -759,7 +767,8 @@ EXPORT_SYMBOL(drm_mm_init_scan);
  * As long as the scan list is non-empty, no other operations than
  * adding/removing nodes to/from the scan list are allowed.
  */
-void drm_mm_init_scan_with_range(struct drm_mm *mm,
+void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
+				 struct drm_mm *mm,
 				 u64 size,
 				 u64 alignment,
 				 unsigned long color,
@@ -768,19 +777,25 @@ void drm_mm_init_scan_with_range(struct drm_mm *mm,
 {
 	DRM_MM_BUG_ON(start >= end);
 	DRM_MM_BUG_ON(!size || size > end - start);
+	DRM_MM_BUG_ON(mm->scan_active);
+
+	scan->mm = mm;
+
+	scan->color = color;
+	scan->alignment = alignment;
+	scan->size = size;
+
+	DRM_MM_BUG_ON(end <= start);
+	scan->range_start = start;
+	scan->range_end = end;
+	scan->check_range = 1;
 
-	mm->scan_color = color;
-	mm->scan_alignment = alignment;
-	mm->scan_size = size;
-	mm->scanned_blocks = 0;
-	mm->scan_hit_start = 0;
-	mm->scan_hit_end = 0;
-	mm->scan_start = start;
-	mm->scan_end = end;
-	mm->scan_check_range = 1;
-	mm->prev_scanned_node = NULL;
+	scan->hit_start = U64_MAX;
+	scan->hit_end = 0;
+
+	scan->prev_scanned_node = NULL;
 }
-EXPORT_SYMBOL(drm_mm_init_scan_with_range);
+EXPORT_SYMBOL(drm_mm_scan_init_with_range);
 
 /**
  * drm_mm_scan_add_block - add a node to the scan list
@@ -792,46 +807,46 @@ EXPORT_SYMBOL(drm_mm_init_scan_with_range);
  * Returns:
  * True if a hole has been found, false otherwise.
  */
-bool drm_mm_scan_add_block(struct drm_mm_node *node)
+bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
+			   struct drm_mm_node *node)
 {
-	struct drm_mm *mm = node->mm;
+	struct drm_mm *mm = scan->mm;
 	struct drm_mm_node *prev_node;
 	u64 hole_start, hole_end;
 	u64 adj_start, adj_end;
 
-	mm->scanned_blocks++;
-
+	DRM_MM_BUG_ON(node->mm != mm);
+	DRM_MM_BUG_ON(!node->allocated);
 	DRM_MM_BUG_ON(node->scanned_block);
 	node->scanned_block = 1;
+	mm->scan_active++;
 
-	prev_node = list_entry(node->node_list.prev, struct drm_mm_node,
-			       node_list);
+	prev_node = list_prev_entry(node, node_list);
 
 	node->scanned_preceeds_hole = prev_node->hole_follows;
 	prev_node->hole_follows = 1;
 	list_del(&node->node_list);
 	node->node_list.prev = &prev_node->node_list;
-	node->node_list.next = &mm->prev_scanned_node->node_list;
-	mm->prev_scanned_node = node;
+	node->node_list.next = &scan->prev_scanned_node->node_list;
+	scan->prev_scanned_node = node;
 
 	adj_start = hole_start = drm_mm_hole_node_start(prev_node);
 	adj_end = hole_end = drm_mm_hole_node_end(prev_node);
 
-	if (mm->scan_check_range) {
-		if (adj_start < mm->scan_start)
-			adj_start = mm->scan_start;
-		if (adj_end > mm->scan_end)
-			adj_end = mm->scan_end;
+	if (scan->check_range) {
+		if (adj_start < scan->range_start)
+			adj_start = scan->range_start;
+		if (adj_end > scan->range_end)
+			adj_end = scan->range_end;
 	}
 
 	if (mm->color_adjust)
-		mm->color_adjust(prev_node, mm->scan_color,
-				 &adj_start, &adj_end);
+		mm->color_adjust(prev_node, scan->color, &adj_start, &adj_end);
 
 	if (check_free_hole(adj_start, adj_end,
-			    mm->scan_size, mm->scan_alignment)) {
-		mm->scan_hit_start = hole_start;
-		mm->scan_hit_end = hole_end;
+			    scan->size, scan->alignment)) {
+		scan->hit_start = hole_start;
+		scan->hit_end = hole_end;
 		return true;
 	}
 
@@ -856,24 +871,25 @@ EXPORT_SYMBOL(drm_mm_scan_add_block);
  * True if this block should be evicted, false otherwise. Will always
  * return false when no hole has been found.
  */
-bool drm_mm_scan_remove_block(struct drm_mm_node *node)
+bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
+			      struct drm_mm_node *node)
 {
-	struct drm_mm *mm = node->mm;
 	struct drm_mm_node *prev_node;
 
-	mm->scanned_blocks--;
-
+	DRM_MM_BUG_ON(node->mm != scan->mm);
 	DRM_MM_BUG_ON(!node->scanned_block);
 	node->scanned_block = 0;
 
-	prev_node = list_entry(node->node_list.prev, struct drm_mm_node,
-			       node_list);
+	DRM_MM_BUG_ON(!node->mm->scan_active);
+	node->mm->scan_active--;
+
+	prev_node = list_prev_entry(node, node_list);
 
 	prev_node->hole_follows = node->scanned_preceeds_hole;
 	list_add(&node->node_list, &prev_node->node_list);
 
-	 return (drm_mm_hole_node_end(node) > mm->scan_hit_start &&
-		 node->start < mm->scan_hit_end);
+	return (drm_mm_hole_node_end(node) > scan->hit_start &&
+		node->start < scan->hit_end);
 }
 EXPORT_SYMBOL(drm_mm_scan_remove_block);
 
@@ -890,7 +906,7 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 	DRM_MM_BUG_ON(start + size <= start);
 
 	INIT_LIST_HEAD(&mm->hole_stack);
-	mm->scanned_blocks = 0;
+	mm->scan_active = 0;
 
 	/* Clever trick to avoid a special case in the free hole tracking. */
 	INIT_LIST_HEAD(&mm->head_node.node_list);
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
index 169ac96e8f08..fe1e886dcabb 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
@@ -113,6 +113,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 
 	while (1) {
 		struct etnaviv_vram_mapping *m, *n;
+		struct drm_mm_scan scan;
 		struct list_head list;
 		bool found;
 
@@ -134,7 +135,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 		}
 
 		/* Try to retire some entries */
-		drm_mm_init_scan(&mmu->mm, size, 0, 0);
+		drm_mm_scan_init(&scan, &mmu->mm, size, 0, 0);
 
 		found = 0;
 		INIT_LIST_HEAD(&list);
@@ -151,7 +152,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 				continue;
 
 			list_add(&free->scan_node, &list);
-			if (drm_mm_scan_add_block(&free->vram_node)) {
+			if (drm_mm_scan_add_block(&scan, &free->vram_node)) {
 				found = true;
 				break;
 			}
@@ -171,7 +172,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 		 * can leave the block pinned.
 		 */
 		list_for_each_entry_safe(m, n, &list, scan_node)
-			if (!drm_mm_scan_remove_block(&m->vram_node))
+			if (!drm_mm_scan_remove_block(&scan, &m->vram_node))
 				list_del_init(&m->scan_node);
 
 		/*
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 6457fd0c33a8..6db0d73c0aa7 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -51,7 +51,10 @@ static bool ggtt_is_idle(struct drm_i915_private *dev_priv)
 }
 
 static bool
-mark_free(struct i915_vma *vma, unsigned int flags, struct list_head *unwind)
+mark_free(struct drm_mm_scan *scan,
+	  struct i915_vma *vma,
+	  unsigned int flags,
+	  struct list_head *unwind)
 {
 	if (i915_vma_is_pinned(vma))
 		return false;
@@ -63,7 +66,7 @@ mark_free(struct i915_vma *vma, unsigned int flags, struct list_head *unwind)
 		return false;
 
 	list_add(&vma->exec_list, unwind);
-	return drm_mm_scan_add_block(&vma->node);
+	return drm_mm_scan_add_block(scan, &vma->node);
 }
 
 /**
@@ -97,6 +100,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
 			 unsigned flags)
 {
 	struct drm_i915_private *dev_priv = vm->i915;
+	struct drm_mm_scan scan;
 	struct list_head eviction_list;
 	struct list_head *phases[] = {
 		&vm->inactive_list,
@@ -123,11 +127,12 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 * object on the TAIL.
 	 */
 	if (start != 0 || end != vm->total) {
-		drm_mm_init_scan_with_range(&vm->mm, min_size,
+		drm_mm_scan_init_with_range(&scan, &vm->mm, min_size,
 					    alignment, cache_level,
 					    start, end);
 	} else
-		drm_mm_init_scan(&vm->mm, min_size, alignment, cache_level);
+		drm_mm_scan_init(&scan, &vm->mm, min_size,
+				 alignment, cache_level);
 
 	/* Retire before we search the active list. Although we have
 	 * reasonable accuracy in our retirement lists, we may have
@@ -144,13 +149,13 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	phase = phases;
 	do {
 		list_for_each_entry(vma, *phase, vm_link)
-			if (mark_free(vma, flags, &eviction_list))
+			if (mark_free(&scan, vma, flags, &eviction_list))
 				goto found;
 	} while (*++phase);
 
 	/* Nothing found, clean up and bail out! */
 	list_for_each_entry_safe(vma, next, &eviction_list, exec_list) {
-		ret = drm_mm_scan_remove_block(&vma->node);
+		ret = drm_mm_scan_remove_block(&scan, &vma->node);
 		BUG_ON(ret);
 
 		INIT_LIST_HEAD(&vma->exec_list);
@@ -199,7 +204,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 * of any of our objects, thus corrupting the list).
 	 */
 	list_for_each_entry_safe(vma, next, &eviction_list, exec_list) {
-		if (drm_mm_scan_remove_block(&vma->node))
+		if (drm_mm_scan_remove_block(&scan, &vma->node))
 			__i915_vma_pin(vma);
 		else
 			list_del_init(&vma->exec_list);
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index ea21c477837c..b96b2964aa01 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1109,11 +1109,11 @@ static int igt_align64(void *ignored)
 	return igt_align_pot(64);
 }
 
-static void show_scan(const struct drm_mm *scan)
+static void show_scan(const struct drm_mm_scan *scan)
 {
 	pr_info("scan: hit [%llx, %llx], size=%lld, align=%lld, color=%ld\n",
-		scan->scan_hit_start, scan->scan_hit_end,
-		scan->scan_size, scan->scan_alignment, scan->scan_color);
+		scan->hit_start, scan->hit_end,
+		scan->size, scan->alignment, scan->color);
 }
 
 static void show_holes(const struct drm_mm *mm, int count)
@@ -1153,7 +1153,7 @@ struct evict_node {
 	struct list_head link;
 };
 
-static bool evict_nodes(struct drm_mm *mm,
+static bool evict_nodes(struct drm_mm_scan *scan,
 			struct evict_node *nodes,
 			unsigned int *order,
 			unsigned int count,
@@ -1165,18 +1165,16 @@ static bool evict_nodes(struct drm_mm *mm,
 	for (i = 0; i < count; i++) {
 		e = &nodes[order ? order[i] : i];
 		list_add(&e->link, evict_list);
-		if (drm_mm_scan_add_block(&e->node))
+		if (drm_mm_scan_add_block(scan, &e->node))
 			break;
 	}
 	list_for_each_entry_safe(e, en, evict_list, link) {
-		if (!drm_mm_scan_remove_block(&e->node))
+		if (!drm_mm_scan_remove_block(scan, &e->node))
 			list_del(&e->link);
 	}
 	if (list_empty(evict_list)) {
 		pr_err("Failed to find eviction: size=%lld [avail=%d], align=%lld (color=%lu)\n",
-		       mm->scan_size, count,
-		       mm->scan_alignment,
-		       mm->scan_color);
+		       scan->size, count, scan->alignment, scan->color);
 		return false;
 	}
 
@@ -1190,19 +1188,20 @@ static bool evict_nothing(struct drm_mm *mm,
 			  unsigned int total_size,
 			  struct evict_node *nodes)
 {
+	struct drm_mm_scan scan;
 	LIST_HEAD(evict_list);
 	struct evict_node *e;
 	struct drm_mm_node *node;
 	unsigned int n;
 
-	drm_mm_init_scan(mm, 1, 0, 0);
+	drm_mm_scan_init(&scan, mm, 1, 0, 0);
 	for (n = 0; n < total_size; n++) {
 		e = &nodes[n];
 		list_add(&e->link, &evict_list);
-		drm_mm_scan_add_block(&e->node);
+		drm_mm_scan_add_block(&scan, &e->node);
 	}
 	list_for_each_entry(e, &evict_list, link)
-		drm_mm_scan_remove_block(&e->node);
+		drm_mm_scan_remove_block(&scan, &e->node);
 
 	for (n = 0; n < total_size; n++) {
 		e = &nodes[n];
@@ -1236,19 +1235,21 @@ static bool evict_everything(struct drm_mm *mm,
 			     unsigned int total_size,
 			     struct evict_node *nodes)
 {
+	struct drm_mm_scan scan;
 	LIST_HEAD(evict_list);
 	struct evict_node *e;
 	unsigned int n;
 	int err;
 
-	drm_mm_init_scan(mm, total_size, 0, 0);
+	drm_mm_scan_init(&scan, mm, total_size, 0, 0);
 	for (n = 0; n < total_size; n++) {
 		e = &nodes[n];
 		list_add(&e->link, &evict_list);
-		drm_mm_scan_add_block(&e->node);
+		if (drm_mm_scan_add_block(&scan, &e->node))
+			break;
 	}
 	list_for_each_entry(e, &evict_list, link) {
-		if (!drm_mm_scan_remove_block(&e->node)) {
+		if (!drm_mm_scan_remove_block(&scan, &e->node)) {
 			pr_err("Node %lld not marked for eviction!\n",
 			       e->node.start);
 			list_del(&e->link);
@@ -1282,15 +1283,16 @@ static int evict_something(struct drm_mm *mm,
 			   unsigned int alignment,
 			   const struct insert_mode *mode)
 {
+	struct drm_mm_scan scan;
 	LIST_HEAD(evict_list);
 	struct evict_node *e;
 	struct drm_mm_node tmp;
 	int err;
 
-	drm_mm_init_scan_with_range(mm,
+	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, 0,
 				    range_start, range_end);
-	if (!evict_nodes(mm,
+	if (!evict_nodes(&scan,
 			 nodes, order, count,
 			 &evict_list))
 		return -EINVAL;
@@ -1302,7 +1304,7 @@ static int evict_something(struct drm_mm *mm,
 	if (err) {
 		pr_err("Failed to insert into eviction hole: size=%d, align=%d\n",
 		       size, alignment);
-		show_scan(mm);
+		show_scan(&scan);
 		show_holes(mm, 3);
 		return err;
 	}
@@ -1859,15 +1861,16 @@ static int evict_color(struct drm_mm *mm,
 		       unsigned long color,
 		       const struct insert_mode *mode)
 {
+	struct drm_mm_scan scan;
 	LIST_HEAD(evict_list);
 	struct evict_node *e;
 	struct drm_mm_node tmp;
 	int err;
 
-	drm_mm_init_scan_with_range(mm,
+	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, color,
 				    range_start, range_end);
-	if (!evict_nodes(mm,
+	if (!evict_nodes(&scan,
 			 nodes, order, count,
 			 &evict_list))
 		return -EINVAL;
@@ -1879,7 +1882,7 @@ static int evict_color(struct drm_mm *mm,
 	if (err) {
 		pr_err("Failed to insert into eviction hole: size=%d, align=%d, color=%lu, err=%d\n",
 		       size, alignment, color, err);
-		show_scan(mm);
+		show_scan(&scan);
 		show_holes(mm, 3);
 		return err;
 	}
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 72e0c0ddf8d0..fcad718c5fb4 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -98,20 +98,29 @@ struct drm_mm {
 	/* Keep an interval_tree for fast lookup of drm_mm_nodes by address. */
 	struct rb_root interval_tree;
 
-	unsigned int scan_check_range : 1;
-	unsigned int scanned_blocks;
-	unsigned long scan_color;
-	u64 scan_alignment;
-	u64 scan_size;
-	u64 scan_hit_start;
-	u64 scan_hit_end;
-	u64 scan_start;
-	u64 scan_end;
-	struct drm_mm_node *prev_scanned_node;
-
 	void (*color_adjust)(const struct drm_mm_node *node,
 			     unsigned long color,
 			     u64 *start, u64 *end);
+
+	unsigned long scan_active;
+};
+
+struct drm_mm_scan {
+	struct drm_mm *mm;
+
+	u64 size;
+	u64 alignment;
+
+	u64 range_start;
+	u64 range_end;
+
+	u64 hit_start;
+	u64 hit_end;
+
+	struct drm_mm_node *prev_scanned_node;
+
+	unsigned long color;
+	bool check_range : 1;
 };
 
 /**
@@ -378,18 +387,22 @@ __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
 	     node__ && node__->start < (end__);				\
 	     node__ = list_next_entry(node__, node_list))
 
-void drm_mm_init_scan(struct drm_mm *mm,
+void drm_mm_scan_init(struct drm_mm_scan *scan,
+		      struct drm_mm *mm,
 		      u64 size,
 		      u64 alignment,
 		      unsigned long color);
-void drm_mm_init_scan_with_range(struct drm_mm *mm,
+void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
+				 struct drm_mm *mm,
 				 u64 size,
 				 u64 alignment,
 				 unsigned long color,
 				 u64 start,
 				 u64 end);
-bool drm_mm_scan_add_block(struct drm_mm_node *node);
-bool drm_mm_scan_remove_block(struct drm_mm_node *node);
+bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
+			   struct drm_mm_node *node);
+bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
+			      struct drm_mm_node *node);
 
 void drm_mm_debug_table(const struct drm_mm *mm, const char *prefix);
 #ifdef CONFIG_DEBUG_FS
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 27/38] drm: Rename prev_node to hole in drm_mm_scan_add_block()

[Chris Wilson]

Acknowledging that we were building up the hole was more useful to me
when reading the code, than knowing the relationship between this node
and the previous node.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 21bd2e13738b..459f10ca5714 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -811,7 +811,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 			   struct drm_mm_node *node)
 {
 	struct drm_mm *mm = scan->mm;
-	struct drm_mm_node *prev_node;
+	struct drm_mm_node *hole;
 	u64 hole_start, hole_end;
 	u64 adj_start, adj_end;
 
@@ -821,17 +821,17 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	node->scanned_block = 1;
 	mm->scan_active++;
 
-	prev_node = list_prev_entry(node, node_list);
+	hole = list_prev_entry(node, node_list);
 
-	node->scanned_preceeds_hole = prev_node->hole_follows;
-	prev_node->hole_follows = 1;
+	node->scanned_preceeds_hole = hole->hole_follows;
+	hole->hole_follows = 1;
 	list_del(&node->node_list);
-	node->node_list.prev = &prev_node->node_list;
+	node->node_list.prev = &hole->node_list;
 	node->node_list.next = &scan->prev_scanned_node->node_list;
 	scan->prev_scanned_node = node;
 
-	adj_start = hole_start = drm_mm_hole_node_start(prev_node);
-	adj_end = hole_end = drm_mm_hole_node_end(prev_node);
+	adj_start = hole_start = drm_mm_hole_node_start(hole);
+	adj_end = hole_end = drm_mm_hole_node_end(hole);
 
 	if (scan->check_range) {
 		if (adj_start < scan->range_start)
@@ -841,7 +841,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	}
 
 	if (mm->color_adjust)
-		mm->color_adjust(prev_node, scan->color, &adj_start, &adj_end);
+		mm->color_adjust(hole, scan->color, &adj_start, &adj_end);
 
 	if (check_free_hole(adj_start, adj_end,
 			    scan->size, scan->alignment)) {
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 28/38] drm: Unconditionally do the range check in drm_mm_scan_add_block()

[Chris Wilson]

Doing the check is trivial (low cost in comparison to overall eviction)
and helps simplify the code.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c              | 53 +++--------------------------------
 drivers/gpu/drm/i915/i915_gem_evict.c | 10 ++-----
 include/drm/drm_mm.h                  | 33 ++++++++++++++++++----
 3 files changed, 34 insertions(+), 62 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 459f10ca5714..c68f79149b9a 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -710,46 +710,6 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  */
 
 /**
- * drm_mm_scan_init - initialize lru scanning
- * @scan: scan state
- * @mm: drm_mm to scan
- * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @color: opaque tag value to use for the allocation
- *
- * This simply sets up the scanning routines with the parameters for the desired
- * hole. Note that there's no need to specify allocation flags, since they only
- * change the place a node is allocated from within a suitable hole.
- *
- * Warning:
- * As long as the scan list is non-empty, no other operations than
- * adding/removing nodes to/from the scan list are allowed.
- */
-void drm_mm_scan_init(struct drm_mm_scan *scan,
-		      struct drm_mm *mm,
-		      u64 size,
-		      u64 alignment,
-		      unsigned long color)
-{
-	DRM_MM_BUG_ON(!size);
-	DRM_MM_BUG_ON(mm->scan_active);
-
-	scan->mm = mm;
-
-	scan->color = color;
-	scan->alignment = alignment;
-	scan->size = size;
-
-	scan->check_range = 0;
-
-	scan->hit_start = U64_MAX;
-	scan->hit_end = 0;
-
-	scan->prev_scanned_node = NULL;
-}
-EXPORT_SYMBOL(drm_mm_scan_init);
-
-/**
  * drm_mm_scan_init_with_range - initialize range-restricted lru scanning
  * @scan: scan state
  * @mm: drm_mm to scan
@@ -788,7 +748,6 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 	DRM_MM_BUG_ON(end <= start);
 	scan->range_start = start;
 	scan->range_end = end;
-	scan->check_range = 1;
 
 	scan->hit_start = U64_MAX;
 	scan->hit_end = 0;
@@ -830,15 +789,11 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	node->node_list.next = &scan->prev_scanned_node->node_list;
 	scan->prev_scanned_node = node;
 
-	adj_start = hole_start = drm_mm_hole_node_start(hole);
-	adj_end = hole_end = drm_mm_hole_node_end(hole);
+	hole_start = drm_mm_hole_node_start(hole);
+	hole_end = drm_mm_hole_node_end(hole);
 
-	if (scan->check_range) {
-		if (adj_start < scan->range_start)
-			adj_start = scan->range_start;
-		if (adj_end > scan->range_end)
-			adj_end = scan->range_end;
-	}
+	adj_start = max(hole_start, scan->range_start);
+	adj_end = min(hole_end, scan->range_end);
 
 	if (mm->color_adjust)
 		mm->color_adjust(hole, scan->color, &adj_start, &adj_end);
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 6db0d73c0aa7..77ded288534b 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -126,13 +126,9 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 * On each list, the oldest objects lie at the HEAD with the freshest
 	 * object on the TAIL.
 	 */
-	if (start != 0 || end != vm->total) {
-		drm_mm_scan_init_with_range(&scan, &vm->mm, min_size,
-					    alignment, cache_level,
-					    start, end);
-	} else
-		drm_mm_scan_init(&scan, &vm->mm, min_size,
-				 alignment, cache_level);
+	drm_mm_scan_init_with_range(&scan, &vm->mm,
+				    min_size, alignment, cache_level,
+				    start, end);
 
 	/* Retire before we search the active list. Although we have
 	 * reasonable accuracy in our retirement lists, we may have
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index fcad718c5fb4..bae0f10da8e3 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -120,7 +120,6 @@ struct drm_mm_scan {
 	struct drm_mm_node *prev_scanned_node;
 
 	unsigned long color;
-	bool check_range : 1;
 };
 
 /**
@@ -387,11 +386,6 @@ __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
 	     node__ && node__->start < (end__);				\
 	     node__ = list_next_entry(node__, node_list))
 
-void drm_mm_scan_init(struct drm_mm_scan *scan,
-		      struct drm_mm *mm,
-		      u64 size,
-		      u64 alignment,
-		      unsigned long color);
 void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 struct drm_mm *mm,
 				 u64 size,
@@ -399,6 +393,33 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 unsigned long color,
 				 u64 start,
 				 u64 end);
+
+/**
+ * drm_mm_scan_init - initialize lru scanning
+ * @scan: scan state
+ * @mm: drm_mm to scan
+ * @size: size of the allocation
+ * @alignment: alignment of the allocation
+ * @color: opaque tag value to use for the allocation
+ *
+ * This simply sets up the scanning routines with the parameters for the desired
+ * hole. Note that there's no need to specify allocation flags, since they only
+ * change the place a node is allocated from within a suitable hole.
+ *
+ * Warning:
+ * As long as the scan list is non-empty, no other operations than
+ * adding/removing nodes to/from the scan list are allowed.
+ */
+static inline void drm_mm_scan_init(struct drm_mm_scan *scan,
+				    struct drm_mm *mm,
+				    u64 size,
+				    u64 alignment,
+				    unsigned long color)
+{
+	drm_mm_scan_init_with_range(scan, mm, size, alignment, color,
+				    0, U64_MAX);
+}
+
 bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 			   struct drm_mm_node *node);
 bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 29/38] drm: Fix application of color vs range restriction when scanning drm_mm

[Chris Wilson]

The range restriction should be applied after the color adjustment, or
else we may inadvertently apply the color adjustment to the restricted
hole (and not against its neighbours).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index c68f79149b9a..1b5613bcb35e 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -772,6 +772,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	struct drm_mm *mm = scan->mm;
 	struct drm_mm_node *hole;
 	u64 hole_start, hole_end;
+	u64 col_start, col_end;
 	u64 adj_start, adj_end;
 
 	DRM_MM_BUG_ON(node->mm != mm);
@@ -789,14 +790,16 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	node->node_list.next = &scan->prev_scanned_node->node_list;
 	scan->prev_scanned_node = node;
 
-	hole_start = drm_mm_hole_node_start(hole);
-	hole_end = drm_mm_hole_node_end(hole);
-
-	adj_start = max(hole_start, scan->range_start);
-	adj_end = min(hole_end, scan->range_end);
+	hole_start = __drm_mm_hole_node_start(hole);
+	hole_end = __drm_mm_hole_node_end(hole);
 
+	col_start = hole_start;
+	col_end = hole_end;
 	if (mm->color_adjust)
-		mm->color_adjust(hole, scan->color, &adj_start, &adj_end);
+		mm->color_adjust(hole, scan->color, &col_start, &col_end);
+
+	adj_start = max(col_start, scan->range_start);
+	adj_end = min(col_end, scan->range_end);
 
 	if (check_free_hole(adj_start, adj_end,
 			    scan->size, scan->alignment)) {
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 30/38] drm: Compute tight evictions for drm_mm_scan

[Chris Wilson]

Compute the minimal required hole during scan and only evict those nodes
that overlap. This enables us to reduce the number of nodes we need to
evict to the bare minimum.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c                | 60 +++++++++++++++++++++++++++------
 drivers/gpu/drm/etnaviv/etnaviv_mmu.c   |  2 +-
 drivers/gpu/drm/i915/i915_gem_evict.c   |  3 +-
 drivers/gpu/drm/selftests/test-drm_mm.c | 10 +++---
 include/drm/drm_mm.h                    | 22 ++++++------
 5 files changed, 71 insertions(+), 26 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 1b5613bcb35e..189ab84c5a59 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -718,10 +718,10 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  * @color: opaque tag value to use for the allocation
  * @start: start of the allowed range for the allocation
  * @end: end of the allowed range for the allocation
+ * @flags: flags to specify how the allocation will be performed afterwards
  *
  * This simply sets up the scanning routines with the parameters for the desired
- * hole. Note that there's no need to specify allocation flags, since they only
- * change the place a node is allocated from within a suitable hole.
+ * hole.
  *
  * Warning:
  * As long as the scan list is non-empty, no other operations than
@@ -733,7 +733,8 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 u64 alignment,
 				 unsigned long color,
 				 u64 start,
-				 u64 end)
+				 u64 end,
+				 unsigned int flags)
 {
 	DRM_MM_BUG_ON(start >= end);
 	DRM_MM_BUG_ON(!size || size > end - start);
@@ -744,6 +745,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 	scan->color = color;
 	scan->alignment = alignment;
 	scan->size = size;
+	scan->flags = flags;
 
 	DRM_MM_BUG_ON(end <= start);
 	scan->range_start = start;
@@ -778,7 +780,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	DRM_MM_BUG_ON(node->mm != mm);
 	DRM_MM_BUG_ON(!node->allocated);
 	DRM_MM_BUG_ON(node->scanned_block);
-	node->scanned_block = 1;
+	node->scanned_block = true;
 	mm->scan_active++;
 
 	hole = list_prev_entry(node, node_list);
@@ -800,15 +802,53 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 
 	adj_start = max(col_start, scan->range_start);
 	adj_end = min(col_end, scan->range_end);
+	if (adj_end <= adj_start || adj_end - adj_start < scan->size)
+		return false;
+
+	if (scan->flags == DRM_MM_CREATE_TOP)
+		adj_start = adj_end - scan->size;
+
+	if (scan->alignment) {
+		u64 rem;
+
+		div64_u64_rem(adj_start, scan->alignment, &rem);
+		if (rem) {
+			adj_start -= rem;
+			if (scan->flags != DRM_MM_CREATE_TOP)
+				adj_start += scan->alignment;
+			if (adj_start < max(col_start, scan->range_start) ||
+			    min(col_end, scan->range_end) - adj_start < scan->size)
+				return false;
+
+			if (adj_end <= adj_start ||
+			    adj_end - adj_start < scan->size)
+				return false;
+		}
+	}
 
-	if (check_free_hole(adj_start, adj_end,
-			    scan->size, scan->alignment)) {
+	if (mm->color_adjust) {
+		/* If allocations need adjusting due to neighbouring colours,
+		 * we do not have enough information to decide if we need
+		 * to evict nodes on either side of [adj_start, adj_end].
+		 * What almost works is
+		 * hit_start = adj_start + (hole_start - col_start);
+		 * hit_end = adj_start + scan->size + (hole_end - col_end);
+		 * but because the decision is only made on the final hole,
+		 * we may underestimate the required adjustments for an
+		 * interior allocation.
+		 */
 		scan->hit_start = hole_start;
 		scan->hit_end = hole_end;
-		return true;
+	} else {
+		scan->hit_start = adj_start;
+		scan->hit_end = adj_start + scan->size;
 	}
 
-	return false;
+	DRM_MM_BUG_ON(scan->hit_start >= scan->hit_end);
+	DRM_MM_BUG_ON(scan->hit_start < hole_start);
+	DRM_MM_BUG_ON(scan->hit_end > hole_end);
+
+	return true;
 }
 EXPORT_SYMBOL(drm_mm_scan_add_block);
 
@@ -836,7 +876,7 @@ bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
 
 	DRM_MM_BUG_ON(node->mm != scan->mm);
 	DRM_MM_BUG_ON(!node->scanned_block);
-	node->scanned_block = 0;
+	node->scanned_block = false;
 
 	DRM_MM_BUG_ON(!node->mm->scan_active);
 	node->mm->scan_active--;
@@ -846,7 +886,7 @@ bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
 	prev_node->hole_follows = node->scanned_preceeds_hole;
 	list_add(&node->node_list, &prev_node->node_list);
 
-	return (drm_mm_hole_node_end(node) > scan->hit_start &&
+	return (node->start + node->size > scan->hit_start &&
 		node->start < scan->hit_end);
 }
 EXPORT_SYMBOL(drm_mm_scan_remove_block);
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
index fe1e886dcabb..2dae3169ce48 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
@@ -135,7 +135,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 		}
 
 		/* Try to retire some entries */
-		drm_mm_scan_init(&scan, &mmu->mm, size, 0, 0);
+		drm_mm_scan_init(&scan, &mmu->mm, size, 0, 0, 0);
 
 		found = 0;
 		INIT_LIST_HEAD(&list);
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 77ded288534b..2741498cdf2b 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -128,7 +128,8 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 */
 	drm_mm_scan_init_with_range(&scan, &vm->mm,
 				    min_size, alignment, cache_level,
-				    start, end);
+				    start, end,
+				    flags & PIN_HIGH ? DRM_MM_CREATE_TOP : 0);
 
 	/* Retire before we search the active list. Although we have
 	 * reasonable accuracy in our retirement lists, we may have
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index b96b2964aa01..71a75043e8b5 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1194,7 +1194,7 @@ static bool evict_nothing(struct drm_mm *mm,
 	struct drm_mm_node *node;
 	unsigned int n;
 
-	drm_mm_scan_init(&scan, mm, 1, 0, 0);
+	drm_mm_scan_init(&scan, mm, 1, 0, 0, 0);
 	for (n = 0; n < total_size; n++) {
 		e = &nodes[n];
 		list_add(&e->link, &evict_list);
@@ -1241,7 +1241,7 @@ static bool evict_everything(struct drm_mm *mm,
 	unsigned int n;
 	int err;
 
-	drm_mm_scan_init(&scan, mm, total_size, 0, 0);
+	drm_mm_scan_init(&scan, mm, total_size, 0, 0, 0);
 	for (n = 0; n < total_size; n++) {
 		e = &nodes[n];
 		list_add(&e->link, &evict_list);
@@ -1291,7 +1291,8 @@ static int evict_something(struct drm_mm *mm,
 
 	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, 0,
-				    range_start, range_end);
+				    range_start, range_end,
+				    mode->create_flags);
 	if (!evict_nodes(&scan,
 			 nodes, order, count,
 			 &evict_list))
@@ -1869,7 +1870,8 @@ static int evict_color(struct drm_mm *mm,
 
 	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, color,
-				    range_start, range_end);
+				    range_start, range_end,
+				    mode->create_flags);
 	if (!evict_nodes(&scan,
 			 nodes, order, count,
 			 &evict_list))
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index bae0f10da8e3..606336fc229a 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -120,6 +120,7 @@ struct drm_mm_scan {
 	struct drm_mm_node *prev_scanned_node;
 
 	unsigned long color;
+	unsigned int flags;
 };
 
 /**
@@ -388,11 +389,9 @@ __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
 
 void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 struct drm_mm *mm,
-				 u64 size,
-				 u64 alignment,
-				 unsigned long color,
-				 u64 start,
-				 u64 end);
+				 u64 size, u64 alignment, unsigned long color,
+				 u64 start, u64 end,
+				 unsigned int flags);
 
 /**
  * drm_mm_scan_init - initialize lru scanning
@@ -401,10 +400,10 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
  * @size: size of the allocation
  * @alignment: alignment of the allocation
  * @color: opaque tag value to use for the allocation
+ * @flags: flags to specify how the allocation will be performed afterwards
  *
  * This simply sets up the scanning routines with the parameters for the desired
- * hole. Note that there's no need to specify allocation flags, since they only
- * change the place a node is allocated from within a suitable hole.
+ * hole.
  *
  * Warning:
  * As long as the scan list is non-empty, no other operations than
@@ -414,10 +413,13 @@ static inline void drm_mm_scan_init(struct drm_mm_scan *scan,
 				    struct drm_mm *mm,
 				    u64 size,
 				    u64 alignment,
-				    unsigned long color)
+				    unsigned long color,
+				    unsigned int flags)
 {
-	drm_mm_scan_init_with_range(scan, mm, size, alignment, color,
-				    0, U64_MAX);
+	drm_mm_scan_init_with_range(scan, mm,
+				    size, alignment, color,
+				    0, U64_MAX,
+				    flags);
 }
 
 bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 31/38] drm: Optimise power-of-two alignments in drm_mm_scan_add_block()

[Chris Wilson]

For power-of-two alignments, we can avoid the 64bit divide and do a
simple bitwise add instead.

v2: s/alignment_mask/remainder_mask/

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 9 ++++++++-
 include/drm/drm_mm.h     | 1 +
 2 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 189ab84c5a59..0441d84fba74 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -742,8 +742,12 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 
 	scan->mm = mm;
 
+	if (alignment <= 1)
+		alignment = 0;
+
 	scan->color = color;
 	scan->alignment = alignment;
+	scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
 	scan->size = size;
 	scan->flags = flags;
 
@@ -811,7 +815,10 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	if (scan->alignment) {
 		u64 rem;
 
-		div64_u64_rem(adj_start, scan->alignment, &rem);
+		if (likely(scan->remainder_mask))
+			rem = adj_start & scan->remainder_mask;
+		else
+			div64_u64_rem(adj_start, scan->alignment, &rem);
 		if (rem) {
 			adj_start -= rem;
 			if (scan->flags != DRM_MM_CREATE_TOP)
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 606336fc229a..d6701d56ea74 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -110,6 +110,7 @@ struct drm_mm_scan {
 
 	u64 size;
 	u64 alignment;
+	u64 remainder_mask;
 
 	u64 range_start;
 	u64 range_end;
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 32/38] drm: Simplify drm_mm scan-list manipulation

[Chris Wilson]

Since we mandate a strict reverse-order of drm_mm_scan_remove_block()
after drm_mm_scan_add_block() we can further simplify the list
manipulations when generating the temporary scan-hole.

v2: Highlight the games being played with the lists to track the scan
holes without allocation.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 35 ++++++++++++++++++-----------------
 include/drm/drm_mm.h     |  7 +------
 2 files changed, 19 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 0441d84fba74..ccca8dafb7fc 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -518,9 +518,7 @@ void drm_mm_remove_node(struct drm_mm_node *node)
 	struct drm_mm_node *prev_node;
 
 	DRM_MM_BUG_ON(!node->allocated);
-	DRM_MM_BUG_ON(node->scanned_block ||
-		      node->scanned_prev_free ||
-		      node->scanned_next_free);
+	DRM_MM_BUG_ON(node->scanned_block);
 
 	prev_node =
 	    list_entry(node->node_list.prev, struct drm_mm_node, node_list);
@@ -757,8 +755,6 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 
 	scan->hit_start = U64_MAX;
 	scan->hit_end = 0;
-
-	scan->prev_scanned_node = NULL;
 }
 EXPORT_SYMBOL(drm_mm_scan_init_with_range);
 
@@ -787,14 +783,14 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	node->scanned_block = true;
 	mm->scan_active++;
 
+	/* Remove this block from the node_list so that we enlarge the hole
+	 * (distance between the end of our previous node and the start of
+	 * or next), without poisoning the link so that we can restore it
+	 * later in drm_mm_scan_remove_block().
+	 */
 	hole = list_prev_entry(node, node_list);
-
-	node->scanned_preceeds_hole = hole->hole_follows;
-	hole->hole_follows = 1;
-	list_del(&node->node_list);
-	node->node_list.prev = &hole->node_list;
-	node->node_list.next = &scan->prev_scanned_node->node_list;
-	scan->prev_scanned_node = node;
+	DRM_MM_BUG_ON(list_next_entry(hole, node_list) != node);
+	__list_del_entry(&node->node_list);
 
 	hole_start = __drm_mm_hole_node_start(hole);
 	hole_end = __drm_mm_hole_node_end(hole);
@@ -888,9 +884,17 @@ bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
 	DRM_MM_BUG_ON(!node->mm->scan_active);
 	node->mm->scan_active--;
 
+	/* During drm_mm_scan_add_block() we decoupled this node leaving
+	 * its pointers intact. Now that the caller is walking back along
+	 * the eviction list we can restore this block into its rightful
+	 * place on the full node_list. To confirm that the caller is walking
+	 * backwards correctly we check that prev_node->next == node->next,
+	 * i.e. both believe the same node should be on the other side of the
+	 * hole.
+	 */
 	prev_node = list_prev_entry(node, node_list);
-
-	prev_node->hole_follows = node->scanned_preceeds_hole;
+	DRM_MM_BUG_ON(list_next_entry(prev_node, node_list) !=
+		      list_next_entry(node, node_list));
 	list_add(&node->node_list, &prev_node->node_list);
 
 	return (node->start + node->size > scan->hit_start &&
@@ -917,9 +921,6 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 	INIT_LIST_HEAD(&mm->head_node.node_list);
 	mm->head_node.allocated = 0;
 	mm->head_node.hole_follows = 1;
-	mm->head_node.scanned_block = 0;
-	mm->head_node.scanned_prev_free = 0;
-	mm->head_node.scanned_next_free = 0;
 	mm->head_node.mm = mm;
 	mm->head_node.start = start + size;
 	mm->head_node.size = start - mm->head_node.start;
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index d6701d56ea74..ff120b7d0f85 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -74,11 +74,8 @@ struct drm_mm_node {
 	struct list_head hole_stack;
 	struct rb_node rb;
 	unsigned hole_follows : 1;
-	unsigned scanned_block : 1;
-	unsigned scanned_prev_free : 1;
-	unsigned scanned_next_free : 1;
-	unsigned scanned_preceeds_hole : 1;
 	unsigned allocated : 1;
+	bool scanned_block : 1;
 	unsigned long color;
 	u64 start;
 	u64 size;
@@ -118,8 +115,6 @@ struct drm_mm_scan {
 	u64 hit_start;
 	u64 hit_end;
 
-	struct drm_mm_node *prev_scanned_node;
-
 	unsigned long color;
 	unsigned int flags;
 };
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 33/38] drm: Apply tight eviction scanning to color_adjust

[Chris Wilson]

Using mm->color_adjust makes the eviction scanner much tricker since we
don't know the actual neighbours of the target hole until after it is
created (after scanning is complete). To work out whether we need to
evict the neighbours because they impact upon the hole, we have to then
check the hole afterwards - requiring an extra step in the user of the
eviction scanner when they apply color_adjust.

v2: Massage kerneldoc.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c                | 76 ++++++++++++++++++++++-----------
 drivers/gpu/drm/i915/i915_gem_evict.c   |  7 +++
 drivers/gpu/drm/selftests/test-drm_mm.c | 20 ++++++++-
 include/drm/drm_mm.h                    |  1 +
 4 files changed, 77 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index ccca8dafb7fc..b59978fe4c6e 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -692,19 +692,21 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  * The DRM range allocator supports this use-case through the scanning
  * interfaces. First a scan operation needs to be initialized with
  * drm_mm_scan_init() or drm_mm_scan_init_with_range(). The driver adds
- * objects to the roaster (probably by walking an LRU list, but this can be
- * freely implemented) until a suitable hole is found or there's no further
- * evictable object.
+ * objects to the roster (probably by walking an LRU list, but this can be
+ * freely implemented) (using drm_mm_scan_add_block()) until a suitable hole
+ * is found or there are no further evictable objects.
  *
  * The driver must walk through all objects again in exactly the reverse
  * order to restore the allocator state. Note that while the allocator is used
  * in the scan mode no other operation is allowed.
  *
- * Finally the driver evicts all objects selected in the scan. Adding and
- * removing an object is O(1), and since freeing a node is also O(1) the overall
- * complexity is O(scanned_objects). So like the free stack which needs to be
- * walked before a scan operation even begins this is linear in the number of
- * objects. It doesn't seem to hurt badly.
+ * Finally the driver evicts all objects selected (drm_mm_scan_remove_block()
+ * reported true) in the scan, and any overlapping nodes after color adjustment
+ * (drm_mm_scan_evict_color()). Adding and removing an object is O(1), and
+ * since freeing a node is also O(1) the overall complexity is
+ * O(scanned_objects). So like the free stack which needs to be walked before a
+ * scan operation even begins this is linear in the number of objects. It
+ * doesn't seem to hurt too badly.
  */
 
 /**
@@ -829,23 +831,8 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 		}
 	}
 
-	if (mm->color_adjust) {
-		/* If allocations need adjusting due to neighbouring colours,
-		 * we do not have enough information to decide if we need
-		 * to evict nodes on either side of [adj_start, adj_end].
-		 * What almost works is
-		 * hit_start = adj_start + (hole_start - col_start);
-		 * hit_end = adj_start + scan->size + (hole_end - col_end);
-		 * but because the decision is only made on the final hole,
-		 * we may underestimate the required adjustments for an
-		 * interior allocation.
-		 */
-		scan->hit_start = hole_start;
-		scan->hit_end = hole_end;
-	} else {
-		scan->hit_start = adj_start;
-		scan->hit_end = adj_start + scan->size;
-	}
+	scan->hit_start = adj_start;
+	scan->hit_end = adj_start + scan->size;
 
 	DRM_MM_BUG_ON(scan->hit_start >= scan->hit_end);
 	DRM_MM_BUG_ON(scan->hit_start < hole_start);
@@ -903,6 +890,45 @@ bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
 EXPORT_SYMBOL(drm_mm_scan_remove_block);
 
 /**
+ * drm_mm_scan_color_evict - evict overlapping nodes on either side of hole
+ * @scan: drm_mm scan with target hole
+ *
+ * After completing an eviction scan and removing the selected nodes, we may
+ * need to remove a few more nodes from either side of the target hole if
+ * mm.color_adjust is being used.
+ *
+ * Returns:
+ * A node to evict, or NULL if there are no overlapping nodes.
+ */
+struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan)
+{
+	struct drm_mm *mm = scan->mm;
+	struct drm_mm_node *hole;
+	u64 hole_start, hole_end;
+
+	DRM_MM_BUG_ON(list_empty(&mm->hole_stack));
+
+	if (!mm->color_adjust)
+		return NULL;
+
+	hole = list_first_entry(&mm->hole_stack, typeof(*hole), hole_stack);
+	hole_start = __drm_mm_hole_node_start(hole);
+	hole_end = __drm_mm_hole_node_end(hole);
+
+	DRM_MM_BUG_ON(hole_start > scan->hit_start);
+	DRM_MM_BUG_ON(hole_end < scan->hit_end);
+
+	mm->color_adjust(hole, scan->color, &hole_start, &hole_end);
+	if (hole_start > scan->hit_start)
+		return hole;
+	if (hole_end < scan->hit_end)
+		return list_next_entry(hole, node_list);
+
+	return NULL;
+}
+EXPORT_SYMBOL(drm_mm_scan_color_evict);
+
+/**
  * drm_mm_init - initialize a drm-mm allocator
  * @mm: the drm_mm structure to initialize
  * @start: start of the range managed by @mm
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 2741498cdf2b..50129ec1caab 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -108,6 +108,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
 		NULL,
 	}, **phase;
 	struct i915_vma *vma, *next;
+	struct drm_mm_node *node;
 	int ret;
 
 	lockdep_assert_held(&vm->i915->drm.struct_mutex);
@@ -218,6 +219,12 @@ i915_gem_evict_something(struct i915_address_space *vm,
 		if (ret == 0)
 			ret = i915_vma_unbind(vma);
 	}
+
+	while (ret == 0 && (node = drm_mm_scan_color_evict(&scan))) {
+		vma = container_of(node, struct i915_vma, node);
+		ret = i915_vma_unbind(vma);
+	}
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 71a75043e8b5..8b9d8956fb51 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1157,6 +1157,7 @@ static bool evict_nodes(struct drm_mm_scan *scan,
 			struct evict_node *nodes,
 			unsigned int *order,
 			unsigned int count,
+			bool use_color,
 			struct list_head *evict_list)
 {
 	struct evict_node *e, *en;
@@ -1181,6 +1182,21 @@ static bool evict_nodes(struct drm_mm_scan *scan,
 	list_for_each_entry(e, evict_list, link)
 		drm_mm_remove_node(&e->node);
 
+	if (use_color) {
+		struct drm_mm_node *node;
+
+		while ((node = drm_mm_scan_color_evict(scan))) {
+			e = container_of(node, typeof(*e), node);
+			drm_mm_remove_node(&e->node);
+			list_add(&e->link, evict_list);
+		}
+	} else {
+		if (drm_mm_scan_color_evict(scan)) {
+			pr_err("drm_mm_scan_color_evict unexpectedly reported overlapping nodes!\n");
+			return false;
+		}
+	}
+
 	return true;
 }
 
@@ -1294,7 +1310,7 @@ static int evict_something(struct drm_mm *mm,
 				    range_start, range_end,
 				    mode->create_flags);
 	if (!evict_nodes(&scan,
-			 nodes, order, count,
+			 nodes, order, count, false,
 			 &evict_list))
 		return -EINVAL;
 
@@ -1873,7 +1889,7 @@ static int evict_color(struct drm_mm *mm,
 				    range_start, range_end,
 				    mode->create_flags);
 	if (!evict_nodes(&scan,
-			 nodes, order, count,
+			 nodes, order, count, true,
 			 &evict_list))
 		return -EINVAL;
 
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index ff120b7d0f85..aed93cbc4bde 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -422,6 +422,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 			   struct drm_mm_node *node);
 bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
 			      struct drm_mm_node *node);
+struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan);
 
 void drm_mm_debug_table(const struct drm_mm *mm, const char *prefix);
 #ifdef CONFIG_DEBUG_FS
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 34/38] drm: Wrap drm_mm_node.hole_follows

[Chris Wilson]

Insulate users from changed to the internal hole tracking within
struct drm_mm_node by using an accessor for hole_follows.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c                | 12 ++++++------
 drivers/gpu/drm/i915/i915_vma.c         |  4 ++--
 drivers/gpu/drm/selftests/test-drm_mm.c | 18 ++++++++++--------
 include/drm/drm_mm.h                    | 22 +++++++++++++++++++---
 4 files changed, 37 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index b59978fe4c6e..c0024719f32b 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -323,7 +323,7 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 	}
 
 	hole = list_last_entry(&hole->node_list, typeof(*hole), node_list);
-	if (!hole->hole_follows)
+	if (!drm_mm_hole_follows(hole))
 		return -ENOSPC;
 
 	adj_start = hole_start = __drm_mm_hole_node_start(hole);
@@ -408,7 +408,7 @@ static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
 	u64 adj_start = hole_start;
 	u64 adj_end = hole_end;
 
-	DRM_MM_BUG_ON(!hole_node->hole_follows || node->allocated);
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
 
 	if (adj_start < start)
 		adj_start = start;
@@ -523,16 +523,16 @@ void drm_mm_remove_node(struct drm_mm_node *node)
 	prev_node =
 	    list_entry(node->node_list.prev, struct drm_mm_node, node_list);
 
-	if (node->hole_follows) {
+	if (drm_mm_hole_follows(node)) {
 		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) ==
 			      __drm_mm_hole_node_end(node));
 		list_del(&node->hole_stack);
-	} else
+	} else {
 		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) !=
 			      __drm_mm_hole_node_end(node));
+	}
 
-
-	if (!prev_node->hole_follows) {
+	if (!drm_mm_hole_follows(prev_node)) {
 		prev_node->hole_follows = 1;
 		list_add(&prev_node->hole_stack, &mm->hole_stack);
 	} else
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index f709c9b76358..34374c4133b5 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -322,11 +322,11 @@ bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long cache_level)
 	GEM_BUG_ON(list_empty(&node->node_list));
 
 	other = list_prev_entry(node, node_list);
-	if (color_differs(other, cache_level) && !other->hole_follows)
+	if (color_differs(other, cache_level) && !drm_mm_hole_follows(other))
 		return false;
 
 	other = list_next_entry(node, node_list);
-	if (color_differs(other, cache_level) && !node->hole_follows)
+	if (color_differs(other, cache_level) && !drm_mm_hole_follows(node))
 		return false;
 
 	return true;
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 8b9d8956fb51..989e49e49f1c 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -63,7 +63,7 @@ static bool assert_no_holes(const struct drm_mm *mm)
 	}
 
 	drm_mm_for_each_node(hole, mm) {
-		if (hole->hole_follows) {
+		if (drm_mm_hole_follows(hole)) {
 			pr_err("Hole follows node, expected none!\n");
 			return false;
 		}
@@ -125,7 +125,7 @@ static bool assert_continuous(const struct drm_mm *mm, u64 size)
 			return false;
 		}
 
-		if (node->hole_follows) {
+		if (drm_mm_hole_follows(node)) {
 			pr_err("node[%ld] is followed by a hole!\n", n);
 			return false;
 		}
@@ -828,7 +828,8 @@ static bool assert_contiguous_in_range(struct drm_mm *mm,
 			return false;
 		}
 
-		if (node->hole_follows && drm_mm_hole_node_end(node) < end) {
+		if (drm_mm_hole_follows(node) &&
+		    drm_mm_hole_node_end(node) < end) {
 			pr_err("node %d is followed by a hole!\n", n);
 			return false;
 		}
@@ -1332,11 +1333,12 @@ static int evict_something(struct drm_mm *mm,
 		err = -EINVAL;
 	}
 
-	if (!assert_node(&tmp, mm, size, alignment, 0) || tmp.hole_follows) {
+	if (!assert_node(&tmp, mm, size, alignment, 0) ||
+	    drm_mm_hole_follows(&tmp)) {
 		pr_err("Inserted did not fill the eviction hole: size=%lld [%d], align=%d [rem=%lld], start=%llx, hole-follows?=%d\n",
 		       tmp.size, size,
 		       alignment, misalignment(&tmp, alignment),
-		       tmp.start, tmp.hole_follows);
+		       tmp.start, drm_mm_hole_follows(&tmp));
 		err = -EINVAL;
 	}
 
@@ -1613,7 +1615,7 @@ static int igt_topdown(void *ignored)
 				goto out;
 			}
 
-			if (nodes[n].hole_follows) {
+			if (drm_mm_hole_follows(&nodes[n])) {
 				pr_err("hole after topdown insert %d, start=%llx\n, size=%u",
 				       n, nodes[n].start, size);
 				goto out;
@@ -1645,7 +1647,7 @@ static int igt_topdown(void *ignored)
 					goto out;
 				}
 
-				if (node->hole_follows) {
+				if (drm_mm_hole_follows(node)) {
 					pr_err("hole after topdown insert %d/%d, start=%llx\n",
 					       m, n, node->start);
 					goto out;
@@ -1700,7 +1702,7 @@ static void separate_adjacent_colors(const struct drm_mm_node *node,
 
 static bool colors_abutt(const struct drm_mm_node *node)
 {
-	if (!node->hole_follows &&
+	if (!drm_mm_hole_follows(node) &&
 	    list_next_entry(node, node_list)->allocated) {
 		pr_err("colors abutt; %ld [%llx + %llx] is next to %ld [%llx + %llx]!\n",
 		       node->color, node->start, node->size,
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index aed93cbc4bde..7da7a171d6d5 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -155,6 +155,22 @@ static inline bool drm_mm_initialized(const struct drm_mm *mm)
 	return mm->hole_stack.next;
 }
 
+/**
+ * drm_mm_hole_follows - checks whether a hole follows this node
+ * @node: drm_mm_node to check
+ *
+ * Holes are embedded into the drm_mm using the tail of a drm_mm_node.
+ * If you wish to know whether a hole follows this particular node,
+ * query this function.
+ *
+ * Returns:
+ * True if a hole follows the @node.
+ */
+static inline bool drm_mm_hole_follows(const struct drm_mm_node *node)
+{
+	return node->hole_follows;
+}
+
 static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
 {
 	return hole_node->start + hole_node->size;
@@ -166,14 +182,14 @@ static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
  *
  * This is useful for driver-specific debug dumpers. Otherwise drivers should
  * not inspect holes themselves. Drivers must check first whether a hole indeed
- * follows by looking at node->hole_follows.
+ * follows by looking at drm_mm_hole_follows()
  *
  * Returns:
  * Start of the subsequent hole.
  */
 static inline u64 drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
 {
-	DRM_MM_BUG_ON(!hole_node->hole_follows);
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node));
 	return __drm_mm_hole_node_start(hole_node);
 }
 
@@ -188,7 +204,7 @@ static inline u64 __drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  *
  * This is useful for driver-specific debug dumpers. Otherwise drivers should
  * not inspect holes themselves. Drivers must check first whether a hole indeed
- * follows by looking at node->hole_follows.
+ * follows by looking at drm_mm_hole_follows().
  *
  * Returns:
  * End of the subsequent hole.
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 35/38] drm: Apply range restriction after color adjustment when allocation

[Chris Wilson]

mm->color_adjust() compares the hole with its neighbouring nodes. They
only abutt before we restrict the hole, so we have to apply color_adjust
before we apply the range restriction.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index c0024719f32b..e279ac7c7aaf 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -410,14 +410,12 @@ static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
 
 	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
 
-	if (adj_start < start)
-		adj_start = start;
-	if (adj_end > end)
-		adj_end = end;
-
 	if (mm->color_adjust)
 		mm->color_adjust(hole_node, color, &adj_start, &adj_end);
 
+	adj_start = max(adj_start, start);
+	adj_end = min(adj_end, end);
+
 	if (flags & DRM_MM_CREATE_TOP)
 		adj_start = adj_end - size;
 
@@ -625,17 +623,15 @@ static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_
 			       flags & DRM_MM_SEARCH_BELOW) {
 		u64 hole_size = adj_end - adj_start;
 
-		if (adj_start < start)
-			adj_start = start;
-		if (adj_end > end)
-			adj_end = end;
-
 		if (mm->color_adjust) {
 			mm->color_adjust(entry, color, &adj_start, &adj_end);
 			if (adj_end <= adj_start)
 				continue;
 		}
 
+		adj_start = max(adj_start, start);
+		adj_end = min(adj_end, end);
+
 		if (!check_free_hole(adj_start, adj_end, size, alignment))
 			continue;
 
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 36/38] drm: Use drm_mm_insert_node_in_range_generic() for everyone

[Chris Wilson]

Remove a superfluous helper as drm_mm_insert_node is equivalent to
insert_node_in_range with a range of (0, U64_MAX).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/drm_mm.c | 166 ++++-------------------------------------------
 include/drm/drm_mm.h     |  90 +++++++++++++++----------
 2 files changed, 67 insertions(+), 189 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index e279ac7c7aaf..58a7e3bbe130 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -92,11 +92,6 @@
  * some basic allocator dumpers for debugging.
  */
 
-static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
-						u64 size,
-						u64 alignment,
-						unsigned long color,
-						enum drm_mm_search_flags flags);
 static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
 						u64 size,
 						u64 alignment,
@@ -230,6 +225,7 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 				 struct drm_mm_node *node,
 				 u64 size, u64 alignment,
 				 unsigned long color,
+				 u64 range_start, u64 range_end,
 				 enum drm_mm_allocator_flags flags)
 {
 	struct drm_mm *mm = hole_node->mm;
@@ -238,11 +234,14 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 	u64 adj_start = hole_start;
 	u64 adj_end = hole_end;
 
-	DRM_MM_BUG_ON(node->allocated);
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
 
 	if (mm->color_adjust)
 		mm->color_adjust(hole_node, color, &adj_start, &adj_end);
 
+	adj_start = max(adj_start, range_start);
+	adj_end = min(adj_end, range_end);
+
 	if (flags & DRM_MM_CREATE_TOP)
 		adj_start = adj_end - size;
 
@@ -258,9 +257,6 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 		}
 	}
 
-	DRM_MM_BUG_ON(adj_start < hole_start);
-	DRM_MM_BUG_ON(adj_end > hole_end);
-
 	if (adj_start == hole_start) {
 		hole_node->hole_follows = 0;
 		list_del(&hole_node->hole_stack);
@@ -276,7 +272,10 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 
 	drm_mm_interval_tree_add_node(hole_node, node);
 
+	DRM_MM_BUG_ON(node->start < range_start);
+	DRM_MM_BUG_ON(node->start < adj_start);
 	DRM_MM_BUG_ON(node->start + node->size > adj_end);
+	DRM_MM_BUG_ON(node->start + node->size > range_end);
 
 	node->hole_follows = 0;
 	if (__drm_mm_hole_node_start(node) < hole_end) {
@@ -360,107 +359,6 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 EXPORT_SYMBOL(drm_mm_reserve_node);
 
 /**
- * drm_mm_insert_node_generic - search for space and insert @node
- * @mm: drm_mm to allocate from
- * @node: preallocate node to insert
- * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @color: opaque tag value to use for this node
- * @sflags: flags to fine-tune the allocation search
- * @aflags: flags to fine-tune the allocation behavior
- *
- * The preallocated node must be cleared to 0.
- *
- * Returns:
- * 0 on success, -ENOSPC if there's no suitable hole.
- */
-int drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
-			       u64 size, u64 alignment,
-			       unsigned long color,
-			       enum drm_mm_search_flags sflags,
-			       enum drm_mm_allocator_flags aflags)
-{
-	struct drm_mm_node *hole_node;
-
-	if (WARN_ON(size == 0))
-		return -EINVAL;
-
-	hole_node = drm_mm_search_free_generic(mm, size, alignment,
-					       color, sflags);
-	if (!hole_node)
-		return -ENOSPC;
-
-	drm_mm_insert_helper(hole_node, node, size, alignment, color, aflags);
-	return 0;
-}
-EXPORT_SYMBOL(drm_mm_insert_node_generic);
-
-static void drm_mm_insert_helper_range(struct drm_mm_node *hole_node,
-				       struct drm_mm_node *node,
-				       u64 size, u64 alignment,
-				       unsigned long color,
-				       u64 start, u64 end,
-				       enum drm_mm_allocator_flags flags)
-{
-	struct drm_mm *mm = hole_node->mm;
-	u64 hole_start = drm_mm_hole_node_start(hole_node);
-	u64 hole_end = drm_mm_hole_node_end(hole_node);
-	u64 adj_start = hole_start;
-	u64 adj_end = hole_end;
-
-	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
-
-	if (mm->color_adjust)
-		mm->color_adjust(hole_node, color, &adj_start, &adj_end);
-
-	adj_start = max(adj_start, start);
-	adj_end = min(adj_end, end);
-
-	if (flags & DRM_MM_CREATE_TOP)
-		adj_start = adj_end - size;
-
-	if (alignment) {
-		u64 rem;
-
-		div64_u64_rem(adj_start, alignment, &rem);
-		if (rem) {
-			if (flags & DRM_MM_CREATE_TOP)
-				adj_start -= rem;
-			else
-				adj_start += alignment - rem;
-		}
-	}
-
-	if (adj_start == hole_start) {
-		hole_node->hole_follows = 0;
-		list_del(&hole_node->hole_stack);
-	}
-
-	node->start = adj_start;
-	node->size = size;
-	node->mm = mm;
-	node->color = color;
-	node->allocated = 1;
-
-	list_add(&node->node_list, &hole_node->node_list);
-
-	drm_mm_interval_tree_add_node(hole_node, node);
-
-	DRM_MM_BUG_ON(node->start < start);
-	DRM_MM_BUG_ON(node->start < adj_start);
-	DRM_MM_BUG_ON(node->start + node->size > adj_end);
-	DRM_MM_BUG_ON(node->start + node->size > end);
-
-	node->hole_follows = 0;
-	if (__drm_mm_hole_node_start(node) < hole_end) {
-		list_add(&node->hole_stack, &mm->hole_stack);
-		node->hole_follows = 1;
-	}
-
-	save_stack(node);
-}
-
-/**
  * drm_mm_insert_node_in_range_generic - ranged search for space and insert @node
  * @mm: drm_mm to allocate from
  * @node: preallocate node to insert
@@ -495,9 +393,9 @@ int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *n
 	if (!hole_node)
 		return -ENOSPC;
 
-	drm_mm_insert_helper_range(hole_node, node,
-				   size, alignment, color,
-				   start, end, aflags);
+	drm_mm_insert_helper(hole_node, node,
+			     size, alignment, color,
+			     start, end, aflags);
 	return 0;
 }
 EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic);
@@ -558,48 +456,6 @@ static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment)
 	return end >= start + size;
 }
 
-static struct drm_mm_node *drm_mm_search_free_generic(const struct drm_mm *mm,
-						      u64 size,
-						      u64 alignment,
-						      unsigned long color,
-						      enum drm_mm_search_flags flags)
-{
-	struct drm_mm_node *entry;
-	struct drm_mm_node *best;
-	u64 adj_start;
-	u64 adj_end;
-	u64 best_size;
-
-	DRM_MM_BUG_ON(mm->scan_active);
-
-	best = NULL;
-	best_size = ~0UL;
-
-	__drm_mm_for_each_hole(entry, mm, adj_start, adj_end,
-			       flags & DRM_MM_SEARCH_BELOW) {
-		u64 hole_size = adj_end - adj_start;
-
-		if (mm->color_adjust) {
-			mm->color_adjust(entry, color, &adj_start, &adj_end);
-			if (adj_end <= adj_start)
-				continue;
-		}
-
-		if (!check_free_hole(adj_start, adj_end, size, alignment))
-			continue;
-
-		if (!(flags & DRM_MM_SEARCH_BEST))
-			return entry;
-
-		if (hole_size < best_size) {
-			best = entry;
-			best_size = hole_size;
-		}
-	}
-
-	return best;
-}
-
 static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
 							u64 size,
 							u64 alignment,
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 7da7a171d6d5..92ec5759caae 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -285,40 +285,6 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  * Basic range manager support (drm_mm.c)
  */
 int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node);
-
-int drm_mm_insert_node_generic(struct drm_mm *mm,
-			       struct drm_mm_node *node,
-			       u64 size,
-			       u64 alignment,
-			       unsigned long color,
-			       enum drm_mm_search_flags sflags,
-			       enum drm_mm_allocator_flags aflags);
-/**
- * drm_mm_insert_node - search for space and insert @node
- * @mm: drm_mm to allocate from
- * @node: preallocate node to insert
- * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @flags: flags to fine-tune the allocation
- *
- * This is a simplified version of drm_mm_insert_node_generic() with @color set
- * to 0.
- *
- * The preallocated node must be cleared to 0.
- *
- * Returns:
- * 0 on success, -ENOSPC if there's no suitable hole.
- */
-static inline int drm_mm_insert_node(struct drm_mm *mm,
-				     struct drm_mm_node *node,
-				     u64 size,
-				     u64 alignment,
-				     enum drm_mm_search_flags flags)
-{
-	return drm_mm_insert_node_generic(mm, node, size, alignment, 0, flags,
-					  DRM_MM_CREATE_DEFAULT);
-}
-
 int drm_mm_insert_node_in_range_generic(struct drm_mm *mm,
 					struct drm_mm_node *node,
 					u64 size,
@@ -328,6 +294,7 @@ int drm_mm_insert_node_in_range_generic(struct drm_mm *mm,
 					u64 end,
 					enum drm_mm_search_flags sflags,
 					enum drm_mm_allocator_flags aflags);
+
 /**
  * drm_mm_insert_node_in_range - ranged search for space and insert @node
  * @mm: drm_mm to allocate from
@@ -359,6 +326,61 @@ static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
 						   DRM_MM_CREATE_DEFAULT);
 }
 
+/**
+ * drm_mm_insert_node_generic - search for space and insert @node
+ * @mm: drm_mm to allocate from
+ * @node: preallocate node to insert
+ * @size: size of the allocation
+ * @alignment: alignment of the allocation
+ * @color: opaque tag value to use for this node
+ * @sflags: flags to fine-tune the allocation search
+ * @aflags: flags to fine-tune the allocation behavior
+ *
+ * The preallocated node must be cleared to 0.
+ *
+ * Returns:
+ * 0 on success, -ENOSPC if there's no suitable hole.
+ */
+static inline int
+drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
+			   u64 size, u64 alignment,
+			   unsigned long color,
+			   enum drm_mm_search_flags sflags,
+			   enum drm_mm_allocator_flags aflags)
+{
+	return drm_mm_insert_node_in_range_generic(mm, node,
+						   size, alignment, 0,
+						   0, U64_MAX,
+						   sflags, aflags);
+}
+
+/**
+ * drm_mm_insert_node - search for space and insert @node
+ * @mm: drm_mm to allocate from
+ * @node: preallocate node to insert
+ * @size: size of the allocation
+ * @alignment: alignment of the allocation
+ * @flags: flags to fine-tune the allocation
+ *
+ * This is a simplified version of drm_mm_insert_node_generic() with @color set
+ * to 0.
+ *
+ * The preallocated node must be cleared to 0.
+ *
+ * Returns:
+ * 0 on success, -ENOSPC if there's no suitable hole.
+ */
+static inline int drm_mm_insert_node(struct drm_mm *mm,
+				     struct drm_mm_node *node,
+				     u64 size,
+				     u64 alignment,
+				     enum drm_mm_search_flags flags)
+{
+	return drm_mm_insert_node_generic(mm, node,
+					  size, alignment, 0,
+					  flags, DRM_MM_CREATE_DEFAULT);
+}
+
 void drm_mm_remove_node(struct drm_mm_node *node);
 void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new);
 void drm_mm_init(struct drm_mm *mm, u64 start, u64 size);
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3 37/38] drm: Improve drm_mm search (and fix topdown allocation) with rbtrees

[Chris Wilson]

The drm_mm range manager claimed to support top-down insertion, but it
was neither searching for the top-most hole that could fit the
allocation request nor fitting the request to the hole correctly.

In order to search the range efficiently, we create a secondary index
for the holes using either their size or their address. This index
allows us to find the smallest hole or the hole at the bottom or top of
the range efficiently, whilst keeping the hole stack to rapidly service
evictions.

v2: Search for holes both high and low. Rename flags to mode.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c  |  16 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c |  20 +-
 drivers/gpu/drm/armada/armada_gem.c          |   4 +-
 drivers/gpu/drm/drm_mm.c                     | 511 +++++++++++++++------------
 drivers/gpu/drm/drm_vma_manager.c            |   3 +-
 drivers/gpu/drm/etnaviv/etnaviv_mmu.c        |   8 +-
 drivers/gpu/drm/i915/i915_gem.c              |  10 +-
 drivers/gpu/drm/i915/i915_gem_evict.c        |   9 +-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c   |   5 +-
 drivers/gpu/drm/i915/i915_gem_gtt.c          |  39 +-
 drivers/gpu/drm/i915/i915_gem_stolen.c       |   6 +-
 drivers/gpu/drm/msm/msm_gem.c                |   3 +-
 drivers/gpu/drm/msm/msm_gem_vma.c            |   3 +-
 drivers/gpu/drm/selftests/test-drm_mm.c      |  58 ++-
 drivers/gpu/drm/sis/sis_mm.c                 |   6 +-
 drivers/gpu/drm/tegra/gem.c                  |   4 +-
 drivers/gpu/drm/ttm/ttm_bo_manager.c         |  18 +-
 drivers/gpu/drm/vc4/vc4_crtc.c               |   2 +-
 drivers/gpu/drm/vc4/vc4_hvs.c                |   3 +-
 drivers/gpu/drm/vc4/vc4_plane.c              |   6 +-
 drivers/gpu/drm/via/via_mm.c                 |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c       |  10 +-
 include/drm/drm_mm.h                         | 135 +++----
 23 files changed, 434 insertions(+), 449 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 00f46b0e076d..d841fcb2e709 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -97,8 +97,7 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 {
 	struct amdgpu_gtt_mgr *mgr = man->priv;
 	struct drm_mm_node *node = mem->mm_node;
-	enum drm_mm_search_flags sflags = DRM_MM_SEARCH_BEST;
-	enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+	enum drm_mm_insert_mode mode;
 	unsigned long fpfn, lpfn;
 	int r;
 
@@ -115,15 +114,14 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 	else
 		lpfn = man->size;
 
-	if (place && place->flags & TTM_PL_FLAG_TOPDOWN) {
-		sflags = DRM_MM_SEARCH_BELOW;
-		aflags = DRM_MM_CREATE_TOP;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (place && place->mode & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
 
 	spin_lock(&mgr->lock);
-	r = drm_mm_insert_node_in_range_generic(&mgr->mm, node, mem->num_pages,
-						mem->page_alignment, 0,
-						fpfn, lpfn, sflags, aflags);
+	r = drm_mm_insert_node_in_range(&mgr->mm, node,
+					mem->num_pages, mem->page_alignment, 0,
+					fpfn, lpfn, mode);
 	spin_unlock(&mgr->lock);
 
 	if (!r) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
index d710226a0fff..5f106ad815ce 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
@@ -97,8 +97,7 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 	struct amdgpu_vram_mgr *mgr = man->priv;
 	struct drm_mm *mm = &mgr->mm;
 	struct drm_mm_node *nodes;
-	enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
-	enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+	enum drm_mm_insert_mode mode;
 	unsigned long lpfn, num_nodes, pages_per_node, pages_left;
 	unsigned i;
 	int r;
@@ -121,10 +120,9 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 	if (!nodes)
 		return -ENOMEM;
 
-	if (place->flags & TTM_PL_FLAG_TOPDOWN) {
-		sflags = DRM_MM_SEARCH_BELOW;
-		aflags = DRM_MM_CREATE_TOP;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (place->flags & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
 
 	pages_left = mem->num_pages;
 
@@ -135,13 +133,11 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 
 		if (pages == pages_per_node)
 			alignment = pages_per_node;
-		else
-			sflags |= DRM_MM_SEARCH_BEST;
 
-		r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
-							alignment, 0,
-							place->fpfn, lpfn,
-							sflags, aflags);
+		r = drm_mm_insert_node_in_range(mm, &nodes[i],
+						pages, alignment, 0,
+						place->fpfn, lpfn,
+						mode);
 		if (unlikely(r))
 			goto error;
 
diff --git a/drivers/gpu/drm/armada/armada_gem.c b/drivers/gpu/drm/armada/armada_gem.c
index 768087ddb046..65b029bc84b5 100644
--- a/drivers/gpu/drm/armada/armada_gem.c
+++ b/drivers/gpu/drm/armada/armada_gem.c
@@ -149,8 +149,8 @@ armada_gem_linear_back(struct drm_device *dev, struct armada_gem_object *obj)
 			return -ENOSPC;
 
 		mutex_lock(&priv->linear_lock);
-		ret = drm_mm_insert_node(&priv->linear, node, size, align,
-					 DRM_MM_SEARCH_DEFAULT);
+		ret = drm_mm_insert_node_generic(&priv->linear, node,
+						 size, align, 0, 0);
 		mutex_unlock(&priv->linear_lock);
 		if (ret) {
 			kfree(node);
diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 58a7e3bbe130..39b964fcdd6f 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -92,14 +92,6 @@
  * some basic allocator dumpers for debugging.
  */
 
-static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
-						u64 size,
-						u64 alignment,
-						unsigned long color,
-						u64 start,
-						u64 end,
-						enum drm_mm_search_flags flags);
-
 #ifdef CONFIG_DRM_DEBUG_MM
 #include <linux/stackdepot.h>
 
@@ -221,69 +213,47 @@ static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node,
 			    &drm_mm_interval_tree_augment);
 }
 
-static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
-				 struct drm_mm_node *node,
-				 u64 size, u64 alignment,
-				 unsigned long color,
-				 u64 range_start, u64 range_end,
-				 enum drm_mm_allocator_flags flags)
+#define RB_INSERT(root, member, expr) do { \
+	struct rb_node **link = &root.rb_node, *rb = NULL; \
+	u64 x = expr(node); \
+	while (*link) { \
+		rb = *link; \
+		if (x < expr(rb_entry(rb, struct drm_mm_node, member))) \
+			link = &rb->rb_left; \
+		else \
+			link = &rb->rb_right; \
+	} \
+	rb_link_node(&node->member, rb, link); \
+	rb_insert_color(&node->member, &root); \
+} while (0)
+
+#define HOLE_SIZE(NODE) ((NODE)->hole_size)
+#define HOLE_ADDR(NODE) (__drm_mm_hole_node_start(NODE))
+
+static void add_hole(struct drm_mm_node *node)
 {
-	struct drm_mm *mm = hole_node->mm;
-	u64 hole_start = drm_mm_hole_node_start(hole_node);
-	u64 hole_end = drm_mm_hole_node_end(hole_node);
-	u64 adj_start = hole_start;
-	u64 adj_end = hole_end;
-
-	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
-
-	if (mm->color_adjust)
-		mm->color_adjust(hole_node, color, &adj_start, &adj_end);
-
-	adj_start = max(adj_start, range_start);
-	adj_end = min(adj_end, range_end);
-
-	if (flags & DRM_MM_CREATE_TOP)
-		adj_start = adj_end - size;
-
-	if (alignment) {
-		u64 rem;
-
-		div64_u64_rem(adj_start, alignment, &rem);
-		if (rem) {
-			if (flags & DRM_MM_CREATE_TOP)
-				adj_start -= rem;
-			else
-				adj_start += alignment - rem;
-		}
-	}
-
-	if (adj_start == hole_start) {
-		hole_node->hole_follows = 0;
-		list_del(&hole_node->hole_stack);
-	}
+	struct drm_mm *mm = node->mm;
 
-	node->start = adj_start;
-	node->size = size;
-	node->mm = mm;
-	node->color = color;
-	node->allocated = 1;
+	node->hole_size =
+		__drm_mm_hole_node_end(node) - __drm_mm_hole_node_start(node);
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(node));
 
-	list_add(&node->node_list, &hole_node->node_list);
+	RB_INSERT(mm->holes_size, rb_hole_size, HOLE_SIZE);
+	RB_INSERT(mm->holes_addr, rb_hole_addr, HOLE_ADDR);
 
-	drm_mm_interval_tree_add_node(hole_node, node);
+	list_add(&node->hole_stack, &mm->hole_stack);
+}
 
-	DRM_MM_BUG_ON(node->start < range_start);
-	DRM_MM_BUG_ON(node->start < adj_start);
-	DRM_MM_BUG_ON(node->start + node->size > adj_end);
-	DRM_MM_BUG_ON(node->start + node->size > range_end);
+static void rm_hole(struct drm_mm_node *node)
+{
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(node));
 
-	node->hole_follows = 0;
-	if (__drm_mm_hole_node_start(node) < hole_end) {
-		list_add(&node->hole_stack, &mm->hole_stack);
-		node->hole_follows = 1;
-	}
+	list_del(&node->hole_stack);
+	rb_erase(&node->rb_hole_size, &node->mm->holes_size);
+	rb_erase(&node->rb_hole_addr, &node->mm->holes_addr);
+	node->hole_size = 0;
 
-	save_stack(node);
+	DRM_MM_BUG_ON(drm_mm_hole_follows(node));
 }
 
 /**
@@ -313,7 +283,7 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 
 	/* Find the relevant hole to add our node to */
 	hole = drm_mm_interval_tree_iter_first(&mm->interval_tree,
-					       node->start, ~(u64)0);
+					       node->start, U64_MAX);
 	if (hole) {
 		if (hole->start < end)
 			return -ENOSPC;
@@ -321,12 +291,12 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 		hole = list_entry(drm_mm_nodes(mm), typeof(*hole), node_list);
 	}
 
-	hole = list_last_entry(&hole->node_list, typeof(*hole), node_list);
+	hole = list_prev_entry(hole, node_list);
 	if (!drm_mm_hole_follows(hole))
 		return -ENOSPC;
 
 	adj_start = hole_start = __drm_mm_hole_node_start(hole);
-	adj_end = hole_end = __drm_mm_hole_node_end(hole);
+	adj_end = hole_end = hole_start + hole->hole_size;
 
 	if (mm->color_adjust)
 		mm->color_adjust(hole, node->color, &adj_start, &adj_end);
@@ -335,31 +305,128 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 		return -ENOSPC;
 
 	node->mm = mm;
-	node->allocated = 1;
 
 	list_add(&node->node_list, &hole->node_list);
-
 	drm_mm_interval_tree_add_node(hole, node);
+	node->allocated = true;
+	node->hole_size = 0;
+
+	rm_hole(hole);
+	if (node->start > hole_start)
+		add_hole(hole);
+	if (end < hole_end)
+		add_hole(node);
+
+	save_stack(node);
+	return 0;
+}
+EXPORT_SYMBOL(drm_mm_reserve_node);
+
+static inline struct drm_mm_node *rb_hole_size_to_node(struct rb_node *rb)
+{
+	return rb ? rb_entry(rb, struct drm_mm_node, rb_hole_size) : NULL;
+}
+
+static inline struct drm_mm_node *rb_hole_addr_to_node(struct rb_node *rb)
+{
+	return rb ? rb_entry(rb, struct drm_mm_node, rb_hole_addr) : NULL;
+}
 
-	if (node->start == hole_start) {
-		hole->hole_follows = 0;
-		list_del(&hole->hole_stack);
+static inline u64 rb_hole_size(struct rb_node *rb)
+{
+	return rb_entry(rb, struct drm_mm_node, rb_hole_size)->hole_size;
+}
+
+static struct drm_mm_node *best_hole(struct drm_mm *mm, u64 size)
+{
+	struct rb_node *best = NULL;
+	struct rb_node **link = &mm->holes_size.rb_node;
+
+	while (*link) {
+		struct rb_node *rb = *link;
+		if (size <= rb_hole_size(rb)) {
+			link = &rb->rb_left;
+			best = rb;
+		} else {
+			link = &rb->rb_right;
+		}
 	}
 
-	node->hole_follows = 0;
-	if (end != hole_end) {
-		list_add(&node->hole_stack, &mm->hole_stack);
-		node->hole_follows = 1;
+	return rb_hole_size_to_node(best);
+}
+
+static struct drm_mm_node *find_hole(struct drm_mm *mm, u64 addr)
+{
+	struct drm_mm_node *node = NULL;
+	struct rb_node **link = &mm->holes_addr.rb_node;
+
+	while (*link) {
+		u64 hole_start;
+
+		node = rb_hole_addr_to_node(*link);
+		hole_start = __drm_mm_hole_node_start(node);
+
+		if (addr < hole_start)
+			link = &node->rb_hole_addr.rb_left;
+		else if (addr > hole_start + node->hole_size)
+			link = &node->rb_hole_addr.rb_right;
+		else
+			break;
 	}
 
-	save_stack(node);
+	return node;
+}
 
-	return 0;
+static struct drm_mm_node *
+first_hole(struct drm_mm *mm,
+	   u64 start, u64 end, u64 size,
+	   enum drm_mm_insert_mode mode)
+{
+	if (RB_EMPTY_ROOT(&mm->holes_size))
+		return NULL;
+
+	switch (mode) {
+	default:
+	case DRM_MM_INSERT_BEST:
+		return best_hole(mm, size);
+
+	case DRM_MM_INSERT_LOW:
+		return find_hole(mm, start);
+
+	case DRM_MM_INSERT_HIGH:
+		return find_hole(mm, end);
+
+	case DRM_MM_INSERT_EVICT:
+		return list_first_entry_or_null(&mm->hole_stack,
+						struct drm_mm_node,
+						hole_stack);
+	}
+}
+
+static struct drm_mm_node *
+next_hole(struct drm_mm *mm,
+	  struct drm_mm_node *node,
+	  enum drm_mm_insert_mode mode)
+{
+	switch (mode) {
+	default:
+	case DRM_MM_INSERT_BEST:
+		return rb_hole_size_to_node(rb_next(&node->rb_hole_size));
+
+	case DRM_MM_INSERT_LOW:
+		return rb_hole_addr_to_node(rb_next(&node->rb_hole_addr));
+
+	case DRM_MM_INSERT_HIGH:
+		return rb_hole_addr_to_node(rb_prev(&node->rb_hole_addr));
+
+	case DRM_MM_INSERT_EVICT:
+		node = list_next_entry(node, hole_stack);
+		return &node->hole_stack == &mm->hole_stack ? NULL : node;
+	}
 }
-EXPORT_SYMBOL(drm_mm_reserve_node);
 
 /**
- * drm_mm_insert_node_in_range_generic - ranged search for space and insert @node
+ * drm_mm_insert_node_in_range - ranged search for space and insert @node
  * @mm: drm_mm to allocate from
  * @node: preallocate node to insert
  * @size: size of the allocation
@@ -367,38 +434,104 @@ EXPORT_SYMBOL(drm_mm_reserve_node);
  * @color: opaque tag value to use for this node
  * @start: start of the allowed range for this node
  * @end: end of the allowed range for this node
- * @sflags: flags to fine-tune the allocation search
- * @aflags: flags to fine-tune the allocation behavior
+ * @mode: fine-tune the allocation search and placement
  *
  * The preallocated node must be cleared to 0.
  *
  * Returns:
  * 0 on success, -ENOSPC if there's no suitable hole.
  */
-int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *node,
-					u64 size, u64 alignment,
-					unsigned long color,
-					u64 start, u64 end,
-					enum drm_mm_search_flags sflags,
-					enum drm_mm_allocator_flags aflags)
+int drm_mm_insert_node_in_range(struct drm_mm * const mm,
+				struct drm_mm_node * const node,
+				u64 size, u64 alignment,
+				unsigned long color,
+				u64 range_start, u64 range_end,
+				enum drm_mm_insert_mode mode)
 {
-	struct drm_mm_node *hole_node;
+	struct drm_mm_node *hole;
+	u64 remainder_mask;
 
-	if (WARN_ON(size == 0))
-		return -EINVAL;
+	DRM_MM_BUG_ON(range_start >= range_end);
 
-	hole_node = drm_mm_search_free_in_range_generic(mm,
-							size, alignment, color,
-							start, end, sflags);
-	if (!hole_node)
+	if (unlikely(size == 0 || range_end - range_start < size))
 		return -ENOSPC;
 
-	drm_mm_insert_helper(hole_node, node,
-			     size, alignment, color,
-			     start, end, aflags);
-	return 0;
+	if (alignment <= 1)
+		alignment = 0;
+
+	remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
+	for (hole = first_hole(mm, range_start, range_end, size, mode); hole;
+	     hole = next_hole(mm, hole, mode)) {
+		u64 hole_start = __drm_mm_hole_node_start(hole);
+		u64 hole_end = hole_start + hole->hole_size;
+		u64 adj_start, adj_end;
+		u64 col_start, col_end;
+
+		if (mode == DRM_MM_INSERT_LOW && hole_start >= range_end)
+			break;
+
+		if (mode == DRM_MM_INSERT_HIGH && hole_end <= range_start)
+			break;
+
+		col_start = hole_start;
+		col_end = hole_end;
+		if (mm->color_adjust)
+			mm->color_adjust(hole, color, &col_start, &col_end);
+
+		adj_start = max(col_start, range_start);
+		adj_end = min(col_end, range_end);
+
+		if (adj_end <= adj_start || adj_end - adj_start < size)
+			continue;
+
+		if (mode == DRM_MM_INSERT_HIGH)
+			adj_start = adj_end - size;
+
+		if (alignment) {
+			u64 rem;
+
+			if (likely(remainder_mask))
+				rem = adj_start & remainder_mask;
+			else
+				div64_u64_rem(adj_start, alignment, &rem);
+			if (rem) {
+				adj_start -= rem;
+				if (mode != DRM_MM_INSERT_HIGH)
+					adj_start += alignment;
+
+				if (adj_start < max(col_start, range_start) ||
+				    min(col_end, range_end) - adj_start < size)
+					continue;
+
+				if (adj_end <= adj_start ||
+				    adj_end - adj_start < size)
+					continue;
+			}
+		}
+
+		node->mm = mm;
+		node->size = size;
+		node->start = adj_start;
+		node->color = color;
+		node->hole_size = 0;
+
+		list_add(&node->node_list, &hole->node_list);
+		drm_mm_interval_tree_add_node(hole, node);
+		node->allocated = true;
+
+		rm_hole(hole);
+		if (adj_start > hole_start)
+			add_hole(hole);
+		if (adj_start + size < hole_end)
+			add_hole(node);
+
+		save_stack(node);
+		return 0;
+	}
+
+	return -ENOSPC;
 }
-EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic);
+EXPORT_SYMBOL(drm_mm_insert_node_in_range);
 
 /**
  * drm_mm_remove_node - Remove a memory node from the allocator.
@@ -416,92 +549,20 @@ void drm_mm_remove_node(struct drm_mm_node *node)
 	DRM_MM_BUG_ON(!node->allocated);
 	DRM_MM_BUG_ON(node->scanned_block);
 
-	prev_node =
-	    list_entry(node->node_list.prev, struct drm_mm_node, node_list);
-
-	if (drm_mm_hole_follows(node)) {
-		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) ==
-			      __drm_mm_hole_node_end(node));
-		list_del(&node->hole_stack);
-	} else {
-		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) !=
-			      __drm_mm_hole_node_end(node));
-	}
+	prev_node = list_prev_entry(node, node_list);
 
-	if (!drm_mm_hole_follows(prev_node)) {
-		prev_node->hole_follows = 1;
-		list_add(&prev_node->hole_stack, &mm->hole_stack);
-	} else
-		list_move(&prev_node->hole_stack, &mm->hole_stack);
+	if (drm_mm_hole_follows(node))
+		rm_hole(node);
 
 	drm_mm_interval_tree_remove(node, &mm->interval_tree);
 	list_del(&node->node_list);
-	node->allocated = 0;
-}
-EXPORT_SYMBOL(drm_mm_remove_node);
-
-static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment)
-{
-	if (end - start < size)
-		return 0;
-
-	if (alignment) {
-		u64 rem;
-
-		div64_u64_rem(start, alignment, &rem);
-		if (rem)
-			start += alignment - rem;
-	}
-
-	return end >= start + size;
-}
-
-static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
-							u64 size,
-							u64 alignment,
-							unsigned long color,
-							u64 start,
-							u64 end,
-							enum drm_mm_search_flags flags)
-{
-	struct drm_mm_node *entry;
-	struct drm_mm_node *best;
-	u64 adj_start;
-	u64 adj_end;
-	u64 best_size;
-
-	DRM_MM_BUG_ON(mm->scan_active);
-
-	best = NULL;
-	best_size = ~0UL;
-
-	__drm_mm_for_each_hole(entry, mm, adj_start, adj_end,
-			       flags & DRM_MM_SEARCH_BELOW) {
-		u64 hole_size = adj_end - adj_start;
-
-		if (mm->color_adjust) {
-			mm->color_adjust(entry, color, &adj_start, &adj_end);
-			if (adj_end <= adj_start)
-				continue;
-		}
-
-		adj_start = max(adj_start, start);
-		adj_end = min(adj_end, end);
-
-		if (!check_free_hole(adj_start, adj_end, size, alignment))
-			continue;
-
-		if (!(flags & DRM_MM_SEARCH_BEST))
-			return entry;
-
-		if (hole_size < best_size) {
-			best = entry;
-			best_size = hole_size;
-		}
-	}
+	node->allocated = false;
 
-	return best;
+	if (drm_mm_hole_follows(prev_node))
+		rm_hole(prev_node);
+	add_hole(prev_node);
 }
+EXPORT_SYMBOL(drm_mm_remove_node);
 
 /**
  * drm_mm_replace_node - move an allocation from @old to @new
@@ -516,18 +577,23 @@ void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new)
 {
 	DRM_MM_BUG_ON(!old->allocated);
 
+	*new = *old;
+
 	list_replace(&old->node_list, &new->node_list);
-	list_replace(&old->hole_stack, &new->hole_stack);
 	rb_replace_node(&old->rb, &new->rb, &old->mm->interval_tree);
-	new->hole_follows = old->hole_follows;
-	new->mm = old->mm;
-	new->start = old->start;
-	new->size = old->size;
-	new->color = old->color;
-	new->__subtree_last = old->__subtree_last;
-
-	old->allocated = 0;
-	new->allocated = 1;
+
+	if (drm_mm_hole_follows(old)) {
+		list_replace(&old->hole_stack, &new->hole_stack);
+		rb_replace_node(&old->rb_hole_size,
+				&new->rb_hole_size,
+				&old->mm->holes_size);
+		rb_replace_node(&old->rb_hole_addr,
+				&new->rb_hole_addr,
+				&old->mm->holes_addr);
+	}
+
+	old->allocated = false;
+	new->allocated = true;
 }
 EXPORT_SYMBOL(drm_mm_replace_node);
 
@@ -570,7 +636,7 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  * @color: opaque tag value to use for the allocation
  * @start: start of the allowed range for the allocation
  * @end: end of the allowed range for the allocation
- * @flags: flags to specify how the allocation will be performed afterwards
+ * @mode: fine-tune the allocation search and placement
  *
  * This simply sets up the scanning routines with the parameters for the desired
  * hole.
@@ -586,7 +652,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 unsigned long color,
 				 u64 start,
 				 u64 end,
-				 unsigned int flags)
+				 enum drm_mm_insert_mode mode)
 {
 	DRM_MM_BUG_ON(start >= end);
 	DRM_MM_BUG_ON(!size || size > end - start);
@@ -601,7 +667,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 	scan->alignment = alignment;
 	scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
 	scan->size = size;
-	scan->flags = flags;
+	scan->mode = mode;
 
 	DRM_MM_BUG_ON(end <= start);
 	scan->range_start = start;
@@ -659,7 +725,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	if (adj_end <= adj_start || adj_end - adj_start < scan->size)
 		return false;
 
-	if (scan->flags == DRM_MM_CREATE_TOP)
+	if (scan->mode == DRM_MM_INSERT_HIGH)
 		adj_start = adj_end - scan->size;
 
 	if (scan->alignment) {
@@ -671,7 +737,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 			div64_u64_rem(adj_start, scan->alignment, &rem);
 		if (rem) {
 			adj_start -= rem;
-			if (scan->flags != DRM_MM_CREATE_TOP)
+			if (scan->mode != DRM_MM_INSERT_HIGH)
 				adj_start += scan->alignment;
 			if (adj_start < max(col_start, scan->range_start) ||
 			    min(col_end, scan->range_end) - adj_start < scan->size)
@@ -765,7 +831,7 @@ struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan)
 
 	hole = list_first_entry(&mm->hole_stack, typeof(*hole), hole_stack);
 	hole_start = __drm_mm_hole_node_start(hole);
-	hole_end = __drm_mm_hole_node_end(hole);
+	hole_end = hole_start + hole->hole_size;
 
 	DRM_MM_BUG_ON(hole_start > scan->hit_start);
 	DRM_MM_BUG_ON(hole_end < scan->hit_end);
@@ -792,21 +858,22 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 {
 	DRM_MM_BUG_ON(start + size <= start);
 
+	mm->color_adjust = NULL;
+
 	INIT_LIST_HEAD(&mm->hole_stack);
-	mm->scan_active = 0;
+	mm->interval_tree = RB_ROOT;
+	mm->holes_size = RB_ROOT;
+	mm->holes_addr = RB_ROOT;
 
 	/* Clever trick to avoid a special case in the free hole tracking. */
 	INIT_LIST_HEAD(&mm->head_node.node_list);
-	mm->head_node.allocated = 0;
-	mm->head_node.hole_follows = 1;
+	mm->head_node.allocated = false;
 	mm->head_node.mm = mm;
 	mm->head_node.start = start + size;
-	mm->head_node.size = start - mm->head_node.start;
-	list_add_tail(&mm->head_node.hole_stack, &mm->hole_stack);
-
-	mm->interval_tree = RB_ROOT;
+	mm->head_node.size = -size;
+	add_hole(&mm->head_node);
 
-	mm->color_adjust = NULL;
+	mm->scan_active = 0;
 }
 EXPORT_SYMBOL(drm_mm_init);
 
@@ -828,18 +895,16 @@ EXPORT_SYMBOL(drm_mm_takedown);
 static u64 drm_mm_debug_hole(const struct drm_mm_node *entry,
 			     const char *prefix)
 {
-	u64 hole_start, hole_end, hole_size;
-
-	if (entry->hole_follows) {
-		hole_start = drm_mm_hole_node_start(entry);
-		hole_end = drm_mm_hole_node_end(entry);
-		hole_size = hole_end - hole_start;
-		pr_debug("%s %#llx-%#llx: %llu: free\n", prefix, hole_start,
-			 hole_end, hole_size);
-		return hole_size;
+	u64 start, size;
+
+	size = entry->hole_size;
+	if (size) {
+		start = drm_mm_hole_node_start(entry);
+		pr_debug("%s %#llx-%#llx: %llu: free\n",
+			 prefix, start, start + size, size);
 	}
 
-	return 0;
+	return size;
 }
 
 /**
@@ -870,18 +935,16 @@ EXPORT_SYMBOL(drm_mm_debug_table);
 #if defined(CONFIG_DEBUG_FS)
 static u64 drm_mm_dump_hole(struct seq_file *m, const struct drm_mm_node *entry)
 {
-	u64 hole_start, hole_end, hole_size;
-
-	if (entry->hole_follows) {
-		hole_start = drm_mm_hole_node_start(entry);
-		hole_end = drm_mm_hole_node_end(entry);
-		hole_size = hole_end - hole_start;
-		seq_printf(m, "%#018llx-%#018llx: %llu: free\n", hole_start,
-			   hole_end, hole_size);
-		return hole_size;
+	u64 start, size;
+
+	size = entry->hole_size;
+	if (size) {
+		start = drm_mm_hole_node_start(entry);
+		seq_printf(m, "%#018llx-%#018llx: %llu: free\n",
+			   start, start + size, size);
 	}
 
-	return 0;
+	return size;
 }
 
 /**
diff --git a/drivers/gpu/drm/drm_vma_manager.c b/drivers/gpu/drm/drm_vma_manager.c
index 20cc33d1bfc1..d9100b565198 100644
--- a/drivers/gpu/drm/drm_vma_manager.c
+++ b/drivers/gpu/drm/drm_vma_manager.c
@@ -212,8 +212,7 @@ int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
 		goto out_unlock;
 	}
 
-	ret = drm_mm_insert_node(&mgr->vm_addr_space_mm, &node->vm_node,
-				 pages, 0, DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node(&mgr->vm_addr_space_mm, &node->vm_node, pages);
 	if (ret)
 		goto out_unlock;
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
index 2dae3169ce48..69ab98c143dc 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
@@ -107,6 +107,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 				   struct drm_mm_node *node, size_t size)
 {
 	struct etnaviv_vram_mapping *free = NULL;
+	enum drm_mm_insert_mode mode = DRM_MM_INSERT_LOW;
 	int ret;
 
 	lockdep_assert_held(&mmu->lock);
@@ -118,8 +119,9 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 		bool found;
 
 		ret = drm_mm_insert_node_in_range(&mmu->mm, node,
-			size, 0, mmu->last_iova, ~0UL,
-			DRM_MM_SEARCH_DEFAULT);
+						  size, 0, 0,
+						  mmu->last_iova, U64_MAX,
+						  mode);
 
 		if (ret != -ENOSPC)
 			break;
@@ -187,6 +189,8 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 			list_del_init(&m->scan_node);
 		}
 
+		mode = DRM_MM_INSERT_EVICT;
+
 		/*
 		 * We removed enough mappings so that the new allocation will
 		 * succeed.  Ensure that the MMU will be flushed before the
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index f86a71d9fe37..d3be3d320d0d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -69,12 +69,10 @@ insert_mappable_node(struct i915_ggtt *ggtt,
                      struct drm_mm_node *node, u32 size)
 {
 	memset(node, 0, sizeof(*node));
-	return drm_mm_insert_node_in_range_generic(&ggtt->base.mm, node,
-						   size, 0,
-						   I915_COLOR_UNEVICTABLE,
-						   0, ggtt->mappable_end,
-						   DRM_MM_SEARCH_DEFAULT,
-						   DRM_MM_CREATE_DEFAULT);
+	return drm_mm_insert_node_in_range(&ggtt->base.mm, node,
+					   size, 0, I915_COLOR_UNEVICTABLE,
+					   0, ggtt->mappable_end,
+					   DRM_MM_INSERT_LOW);
 }
 
 static void
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 50129ec1caab..f92f63e8749a 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -109,6 +109,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	}, **phase;
 	struct i915_vma *vma, *next;
 	struct drm_mm_node *node;
+	enum drm_mm_insert_mode mode;
 	int ret;
 
 	lockdep_assert_held(&vm->i915->drm.struct_mutex);
@@ -127,10 +128,14 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 * On each list, the oldest objects lie at the HEAD with the freshest
 	 * object on the TAIL.
 	 */
+	mode = DRM_MM_INSERT_BEST;
+	if (flags & PIN_HIGH)
+		mode = DRM_MM_INSERT_HIGH;
+	if (flags & PIN_MAPPABLE)
+		mode = DRM_MM_INSERT_LOW;
 	drm_mm_scan_init_with_range(&scan, &vm->mm,
 				    min_size, alignment, cache_level,
-				    start, end,
-				    flags & PIN_HIGH ? DRM_MM_CREATE_TOP : 0);
+				    start, end, mode);
 
 	/* Retire before we search the active list. Although we have
 	 * reasonable accuracy in our retirement lists, we may have
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index d665a33229bd..98b0963a1aaf 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -436,12 +436,11 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 					       PIN_MAPPABLE | PIN_NONBLOCK);
 		if (IS_ERR(vma)) {
 			memset(&cache->node, 0, sizeof(cache->node));
-			ret = drm_mm_insert_node_in_range_generic
+			ret = drm_mm_insert_node_in_range
 				(&ggtt->base.mm, &cache->node,
 				 4096, 0, I915_COLOR_UNEVICTABLE,
 				 0, ggtt->mappable_end,
-				 DRM_MM_SEARCH_DEFAULT,
-				 DRM_MM_CREATE_DEFAULT);
+				 DRM_MM_INSERT_LOW);
 			if (ret) /* no inactive aperture space, use cpu reloc */
 				return NULL;
 		} else {
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 97fd66e4e3d0..f1b685896833 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2734,12 +2734,10 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv)
 		return ret;
 
 	/* Reserve a mappable slot for our lockless error capture */
-	ret = drm_mm_insert_node_in_range_generic(&ggtt->base.mm,
-						  &ggtt->error_capture,
-						  4096, 0,
-						  I915_COLOR_UNEVICTABLE,
-						  0, ggtt->mappable_end,
-						  0, 0);
+	ret = drm_mm_insert_node_in_range(&ggtt->base.mm, &ggtt->error_capture,
+					  4096, 0, I915_COLOR_UNEVICTABLE,
+					  0, ggtt->mappable_end,
+					  DRM_MM_INSERT_LOW);
 	if (ret)
 		return ret;
 
@@ -3554,18 +3552,16 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 			u64 size, u64 alignment, unsigned long colour,
 			u64 start, u64 end, unsigned int flags)
 {
-	u32 search_flag, alloc_flag;
+	enum drm_mm_insert_mode mode;
 	int err;
 
 	lockdep_assert_held(&vm->i915->drm.struct_mutex);
 
-	if (flags & PIN_HIGH) {
-		search_flag = DRM_MM_SEARCH_BELOW;
-		alloc_flag = DRM_MM_CREATE_TOP;
-	} else {
-		search_flag = DRM_MM_SEARCH_DEFAULT;
-		alloc_flag = DRM_MM_CREATE_DEFAULT;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (flags & PIN_HIGH)
+		mode = DRM_MM_INSERT_HIGH;
+	if (flags & PIN_MAPPABLE)
+		mode = DRM_MM_INSERT_LOW;
 
 	/* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks,
 	 * so we know that we always have a minimum alignment of 4096.
@@ -3577,10 +3573,9 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 	if (alignment <= 4096)
 		alignment = 0;
 
-	err = drm_mm_insert_node_in_range_generic(&vm->mm, node,
-						  size, alignment, colour,
-						  start, end,
-						  search_flag, alloc_flag);
+	err = drm_mm_insert_node_in_range(&vm->mm, node,
+					  size, alignment, colour,
+					  start, end, mode);
 	if (err != -ENOSPC)
 		return err;
 
@@ -3589,9 +3584,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 	if (err)
 		return err;
 
-	search_flag = DRM_MM_SEARCH_DEFAULT;
-	return drm_mm_insert_node_in_range_generic(&vm->mm, node,
-						   size, alignment, colour,
-						   start, end,
-						   search_flag, alloc_flag);
+	return drm_mm_insert_node_in_range(&vm->mm, node,
+					   size, alignment, colour,
+					   start, end, DRM_MM_INSERT_EVICT);
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index efc0e748ef89..00341e0fae6b 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -62,9 +62,9 @@ int i915_gem_stolen_insert_node_in_range(struct drm_i915_private *dev_priv,
 		start = 4096;
 
 	mutex_lock(&dev_priv->mm.stolen_lock);
-	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node, size,
-					  alignment, start, end,
-					  DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node,
+					  size, alignment, 0,
+					  start, end, DRM_MM_INSERT_BEST);
 	mutex_unlock(&dev_priv->mm.stolen_lock);
 
 	return ret;
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index cd06cfd94687..412669062cb7 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -54,8 +54,7 @@ static struct page **get_pages_vram(struct drm_gem_object *obj,
 	if (!p)
 		return ERR_PTR(-ENOMEM);
 
-	ret = drm_mm_insert_node(&priv->vram.mm, msm_obj->vram_node,
-			npages, 0, DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node(&priv->vram.mm, msm_obj->vram_node, npages);
 	if (ret) {
 		drm_free_large(p);
 		return ERR_PTR(ret);
diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c
index a311d26ccb21..b654eca7636a 100644
--- a/drivers/gpu/drm/msm/msm_gem_vma.c
+++ b/drivers/gpu/drm/msm/msm_gem_vma.c
@@ -45,8 +45,7 @@ msm_gem_map_vma(struct msm_gem_address_space *aspace,
 	if (WARN_ON(drm_mm_node_allocated(&vma->node)))
 		return 0;
 
-	ret = drm_mm_insert_node(&aspace->mm, &vma->node, npages,
-			0, DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node(&aspace->mm, &vma->node, npages);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 989e49e49f1c..26e7001c846d 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -22,23 +22,24 @@ static unsigned int max_iterations = 8192;
 static unsigned int max_prime = 128;
 
 enum {
-	DEFAULT,
-	TOPDOWN,
 	BEST,
+	BOTTOMUP,
+	TOPDOWN,
+	EVICT,
 };
 
 static const struct insert_mode {
 	const char *name;
-	unsigned int search_flags;
-	unsigned int create_flags;
+	enum drm_mm_insert_mode mode;
 } insert_modes[] = {
-	[DEFAULT] = { "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
-	[TOPDOWN] = { "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
-	[BEST] = { "best", DRM_MM_SEARCH_BEST, DRM_MM_CREATE_DEFAULT },
+	[BEST] = { "best", DRM_MM_INSERT_BEST },
+	[BOTTOMUP] = { "bottom-up", DRM_MM_INSERT_LOW },
+	[TOPDOWN] = { "top-down", DRM_MM_INSERT_HIGH },
+	[EVICT] = { "evict", DRM_MM_INSERT_EVICT },
 	{}
 }, evict_modes[] = {
-	{ "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
-	{ "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
+	{ "bottom-up", DRM_MM_INSERT_LOW },
+	{ "top-down", DRM_MM_INSERT_HIGH },
 	{}
 };
 
@@ -522,8 +523,7 @@ static bool expect_insert(struct drm_mm *mm, struct drm_mm_node *node,
 
 	err = drm_mm_insert_node_generic(mm, node,
 					 size, alignment, color,
-					 mode->search_flags,
-					 mode->create_flags);
+					 mode->mode);
 	if (err) {
 		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) failed with err=%d\n",
 		       size, alignment, color, mode->name, err);
@@ -543,7 +543,7 @@ static bool expect_insert_fail(struct drm_mm *mm, u64 size)
 	struct drm_mm_node tmp = {};
 	int err;
 
-	err = drm_mm_insert_node(mm, &tmp, size, 0, DRM_MM_SEARCH_DEFAULT);
+	err = drm_mm_insert_node(mm, &tmp, size);
 	if (likely(err == -ENOSPC))
 		return true;
 
@@ -749,11 +749,10 @@ static bool expect_insert_in_range(struct drm_mm *mm, struct drm_mm_node *node,
 {
 	int err;
 
-	err = drm_mm_insert_node_in_range_generic(mm, node,
-						  size, alignment, color,
-						  range_start, range_end,
-						  mode->search_flags,
-						  mode->create_flags);
+	err = drm_mm_insert_node_in_range(mm, node,
+					  size, alignment, color,
+					  range_start, range_end,
+					  mode->mode);
 	if (err) {
 		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) nto range [%llx, %llx] failed with err=%d\n",
 		       size, alignment, color, mode->name,
@@ -777,11 +776,10 @@ static bool expect_insert_in_range_fail(struct drm_mm *mm,
 	struct drm_mm_node tmp = {};
 	int err;
 
-	err = drm_mm_insert_node_in_range_generic(mm, &tmp,
-						  size, 0, 0,
-						  range_start, range_end,
-						  DRM_MM_SEARCH_DEFAULT,
-						  DRM_MM_CREATE_DEFAULT);
+	err = drm_mm_insert_node_in_range(mm, &tmp,
+					  size, 0, 0,
+					  range_start, range_end,
+					  0);
 	if (likely(err == -ENOSPC))
 		return true;
 
@@ -1309,7 +1307,7 @@ static int evict_something(struct drm_mm *mm,
 	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, 0,
 				    range_start, range_end,
-				    mode->create_flags);
+				    mode->mode);
 	if (!evict_nodes(&scan,
 			 nodes, order, count, false,
 			 &evict_list))
@@ -1317,8 +1315,7 @@ static int evict_something(struct drm_mm *mm,
 
 	memset(&tmp, 0, sizeof(tmp));
 	err = drm_mm_insert_node_generic(mm, &tmp, size, alignment, 0,
-					 mode->search_flags,
-					 mode->create_flags);
+					 DRM_MM_INSERT_EVICT);
 	if (err) {
 		pr_err("Failed to insert into eviction hole: size=%d, align=%d\n",
 		       size, alignment);
@@ -1393,8 +1390,7 @@ static int igt_evict(void *ignored)
 	ret = -EINVAL;
 	drm_mm_init(&mm, 0, size);
 	for (n = 0; n < size; n++) {
-		err = drm_mm_insert_node(&mm, &nodes[n].node, 1, 0,
-					 DRM_MM_SEARCH_DEFAULT);
+		err = drm_mm_insert_node(&mm, &nodes[n].node, 1);
 		if (err) {
 			pr_err("insert failed, step %d\n", n);
 			ret = err;
@@ -1502,8 +1498,7 @@ static int igt_evict_range(void *ignored)
 	ret = -EINVAL;
 	drm_mm_init(&mm, 0, size);
 	for (n = 0; n < size; n++) {
-		err = drm_mm_insert_node(&mm, &nodes[n].node, 1, 0,
-					 DRM_MM_SEARCH_DEFAULT);
+		err = drm_mm_insert_node(&mm, &nodes[n].node, 1);
 		if (err) {
 			pr_err("insert failed, step %d\n", n);
 			ret = err;
@@ -1889,7 +1884,7 @@ static int evict_color(struct drm_mm *mm,
 	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, color,
 				    range_start, range_end,
-				    mode->create_flags);
+				    mode->mode);
 	if (!evict_nodes(&scan,
 			 nodes, order, count, true,
 			 &evict_list))
@@ -1897,8 +1892,7 @@ static int evict_color(struct drm_mm *mm,
 
 	memset(&tmp, 0, sizeof(tmp));
 	err = drm_mm_insert_node_generic(mm, &tmp, size, alignment, color,
-					 mode->search_flags,
-					 mode->create_flags);
+					 DRM_MM_INSERT_EVICT);
 	if (err) {
 		pr_err("Failed to insert into eviction hole: size=%d, align=%d, color=%lu, err=%d\n",
 		       size, alignment, color, err);
diff --git a/drivers/gpu/drm/sis/sis_mm.c b/drivers/gpu/drm/sis/sis_mm.c
index 03defda77766..1622db24cd39 100644
--- a/drivers/gpu/drm/sis/sis_mm.c
+++ b/drivers/gpu/drm/sis/sis_mm.c
@@ -109,8 +109,7 @@ static int sis_drm_alloc(struct drm_device *dev, struct drm_file *file,
 	if (pool == AGP_TYPE) {
 		retval = drm_mm_insert_node(&dev_priv->agp_mm,
 					    &item->mm_node,
-					    mem->size, 0,
-					    DRM_MM_SEARCH_DEFAULT);
+					    mem->size);
 		offset = item->mm_node.start;
 	} else {
 #if defined(CONFIG_FB_SIS) || defined(CONFIG_FB_SIS_MODULE)
@@ -122,8 +121,7 @@ static int sis_drm_alloc(struct drm_device *dev, struct drm_file *file,
 #else
 		retval = drm_mm_insert_node(&dev_priv->vram_mm,
 					    &item->mm_node,
-					    mem->size, 0,
-					    DRM_MM_SEARCH_DEFAULT);
+					    mem->size);
 		offset = item->mm_node.start;
 #endif
 	}
diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c
index c08e5279eeac..a6051d6c6de6 100644
--- a/drivers/gpu/drm/tegra/gem.c
+++ b/drivers/gpu/drm/tegra/gem.c
@@ -128,8 +128,8 @@ static int tegra_bo_iommu_map(struct tegra_drm *tegra, struct tegra_bo *bo)
 	if (!bo->mm)
 		return -ENOMEM;
 
-	err = drm_mm_insert_node_generic(&tegra->mm, bo->mm, bo->gem.size,
-					 PAGE_SIZE, 0, 0, 0);
+	err = drm_mm_insert_node_generic(&tegra->mm,
+					 bo->mm, bo->gem.size, PAGE_SIZE, 0, 0);
 	if (err < 0) {
 		dev_err(tegra->drm->dev, "out of I/O virtual memory: %zd\n",
 			err);
diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c
index aa0bd054d3e9..66be8c50ffdb 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_manager.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c
@@ -54,9 +54,8 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man,
 {
 	struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
 	struct drm_mm *mm = &rman->mm;
-	struct drm_mm_node *node = NULL;
-	enum drm_mm_search_flags sflags = DRM_MM_SEARCH_BEST;
-	enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+	struct drm_mm_node *node;
+	enum drm_mm_insert_mode mode;
 	unsigned long lpfn;
 	int ret;
 
@@ -68,16 +67,15 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man,
 	if (!node)
 		return -ENOMEM;
 
-	if (place->flags & TTM_PL_FLAG_TOPDOWN) {
-		sflags = DRM_MM_SEARCH_BELOW;
-		aflags = DRM_MM_CREATE_TOP;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (place->flags & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
 
 	spin_lock(&rman->lock);
-	ret = drm_mm_insert_node_in_range_generic(mm, node, mem->num_pages,
+	ret = drm_mm_insert_node_in_range(mm, node,
+					  mem->num_pages,
 					  mem->page_alignment, 0,
-					  place->fpfn, lpfn,
-					  sflags, aflags);
+					  place->fpfn, lpfn, mode);
 	spin_unlock(&rman->lock);
 
 	if (unlikely(ret)) {
diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c
index a0fd3e66bc4b..baacdf62c83e 100644
--- a/drivers/gpu/drm/vc4/vc4_crtc.c
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -589,7 +589,7 @@ static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
 
 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
-				 dlist_count, 1, 0);
+				 dlist_count);
 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
 	if (ret)
 		return ret;
diff --git a/drivers/gpu/drm/vc4/vc4_hvs.c b/drivers/gpu/drm/vc4/vc4_hvs.c
index 6fbab1c82cb1..4aba0fa56289 100644
--- a/drivers/gpu/drm/vc4/vc4_hvs.c
+++ b/drivers/gpu/drm/vc4/vc4_hvs.c
@@ -141,8 +141,7 @@ static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
 	int ret, i;
 	u32 __iomem *dst_kernel;
 
-	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS, 1,
-				 0);
+	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
 	if (ret) {
 		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
 			  ret);
diff --git a/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c
index 110d1518f5d5..c1f06897136b 100644
--- a/drivers/gpu/drm/vc4/vc4_plane.c
+++ b/drivers/gpu/drm/vc4/vc4_plane.c
@@ -514,9 +514,9 @@ static int vc4_plane_mode_set(struct drm_plane *plane,
 	if (lbm_size) {
 		if (!vc4_state->lbm.allocated) {
 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
-			ret = drm_mm_insert_node(&vc4->hvs->lbm_mm,
-						 &vc4_state->lbm,
-						 lbm_size, 32, 0);
+			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
+							 &vc4_state->lbm,
+							 lbm_size, 32, 0, 0);
 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
 		} else {
 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
diff --git a/drivers/gpu/drm/via/via_mm.c b/drivers/gpu/drm/via/via_mm.c
index a04ef1c992d9..4217d66a5cc6 100644
--- a/drivers/gpu/drm/via/via_mm.c
+++ b/drivers/gpu/drm/via/via_mm.c
@@ -140,11 +140,11 @@ int via_mem_alloc(struct drm_device *dev, void *data,
 	if (mem->type == VIA_MEM_AGP)
 		retval = drm_mm_insert_node(&dev_priv->agp_mm,
 					    &item->mm_node,
-					    tmpSize, 0, DRM_MM_SEARCH_DEFAULT);
+					    tmpSize);
 	else
 		retval = drm_mm_insert_node(&dev_priv->vram_mm,
 					    &item->mm_node,
-					    tmpSize, 0, DRM_MM_SEARCH_DEFAULT);
+					    tmpSize);
 	if (retval)
 		goto fail_alloc;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index aa04fb0159a7..77cb7c627e09 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -673,16 +673,10 @@ static bool vmw_cmdbuf_try_alloc(struct vmw_cmdbuf_man *man,
  
 	memset(info->node, 0, sizeof(*info->node));
 	spin_lock_bh(&man->lock);
-	ret = drm_mm_insert_node_generic(&man->mm, info->node, info->page_size,
-					 0, 0,
-					 DRM_MM_SEARCH_DEFAULT,
-					 DRM_MM_CREATE_DEFAULT);
+	ret = drm_mm_insert_node(&man->mm, info->node, info->page_size);
 	if (ret) {
 		vmw_cmdbuf_man_process(man);
-		ret = drm_mm_insert_node_generic(&man->mm, info->node,
-						 info->page_size, 0, 0,
-						 DRM_MM_SEARCH_DEFAULT,
-						 DRM_MM_CREATE_DEFAULT);
+		ret = drm_mm_insert_node(&man->mm, info->node, info->page_size);
 	}
 
 	spin_unlock_bh(&man->lock);
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 92ec5759caae..6558cd08d7c9 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -55,32 +55,27 @@
 #define DRM_MM_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr)
 #endif
 
-enum drm_mm_search_flags {
-	DRM_MM_SEARCH_DEFAULT =		0,
-	DRM_MM_SEARCH_BEST =		1 << 0,
-	DRM_MM_SEARCH_BELOW =		1 << 1,
+enum drm_mm_insert_mode {
+	DRM_MM_INSERT_BEST = 0,
+	DRM_MM_INSERT_LOW,
+	DRM_MM_INSERT_HIGH,
+	DRM_MM_INSERT_EVICT,
 };
 
-enum drm_mm_allocator_flags {
-	DRM_MM_CREATE_DEFAULT =		0,
-	DRM_MM_CREATE_TOP =		1 << 0,
-};
-
-#define DRM_MM_BOTTOMUP DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT
-#define DRM_MM_TOPDOWN DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP
-
 struct drm_mm_node {
+	struct drm_mm *mm;
 	struct list_head node_list;
 	struct list_head hole_stack;
 	struct rb_node rb;
-	unsigned hole_follows : 1;
-	unsigned allocated : 1;
-	bool scanned_block : 1;
-	unsigned long color;
+	struct rb_node rb_hole_size;
+	struct rb_node rb_hole_addr;
 	u64 start;
 	u64 size;
 	u64 __subtree_last;
-	struct drm_mm *mm;
+	u64 hole_size;
+	unsigned long color;
+	bool allocated : 1;
+	bool scanned_block : 1;
 #ifdef CONFIG_DRM_DEBUG_MM
 	depot_stack_handle_t stack;
 #endif
@@ -94,6 +89,8 @@ struct drm_mm {
 	struct drm_mm_node head_node;
 	/* Keep an interval_tree for fast lookup of drm_mm_nodes by address. */
 	struct rb_root interval_tree;
+	struct rb_root holes_size;
+	struct rb_root holes_addr;
 
 	void (*color_adjust)(const struct drm_mm_node *node,
 			     unsigned long color,
@@ -116,7 +113,7 @@ struct drm_mm_scan {
 	u64 hit_end;
 
 	unsigned long color;
-	unsigned int flags;
+	enum drm_mm_insert_mode mode;
 };
 
 /**
@@ -168,7 +165,7 @@ static inline bool drm_mm_initialized(const struct drm_mm *mm)
  */
 static inline bool drm_mm_hole_follows(const struct drm_mm_node *node)
 {
-	return node->hole_follows;
+	return node->hole_size;
 }
 
 static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
@@ -251,14 +248,6 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
 #define drm_mm_for_each_node_safe(entry, next, mm) \
 	list_for_each_entry_safe(entry, next, drm_mm_nodes(mm), node_list)
 
-#define __drm_mm_for_each_hole(entry, mm, hole_start, hole_end, backwards) \
-	for (entry = list_entry((backwards) ? (mm)->hole_stack.prev : (mm)->hole_stack.next, struct drm_mm_node, hole_stack); \
-	     &entry->hole_stack != &(mm)->hole_stack ? \
-	     hole_start = drm_mm_hole_node_start(entry), \
-	     hole_end = drm_mm_hole_node_end(entry), \
-	     1 : 0; \
-	     entry = list_entry((backwards) ? entry->hole_stack.prev : entry->hole_stack.next, struct drm_mm_node, hole_stack))
-
 /**
  * drm_mm_for_each_hole - iterator to walk over all holes
  * @entry: drm_mm_node used internally to track progress
@@ -274,57 +263,27 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  * Implementation Note:
  * We need to inline list_for_each_entry in order to be able to set hole_start
  * and hole_end on each iteration while keeping the macro sane.
- *
- * The __drm_mm_for_each_hole version is similar, but with added support for
- * going backwards.
  */
-#define drm_mm_for_each_hole(entry, mm, hole_start, hole_end) \
-	__drm_mm_for_each_hole(entry, mm, hole_start, hole_end, 0)
+#define drm_mm_for_each_hole(pos, mm, hole_start, hole_end) \
+	for (pos = list_first_entry(&(mm)->hole_stack, \
+				    typeof(*pos), hole_stack); \
+	     &pos->hole_stack != &(mm)->hole_stack ? \
+	     hole_start = drm_mm_hole_node_start(pos), \
+	     hole_end = hole_start + pos->hole_size : 0; \
+	     pos = list_next_entry(pos, hole_stack))
 
 /*
  * Basic range manager support (drm_mm.c)
  */
 int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node);
-int drm_mm_insert_node_in_range_generic(struct drm_mm *mm,
-					struct drm_mm_node *node,
-					u64 size,
-					u64 alignment,
-					unsigned long color,
-					u64 start,
-					u64 end,
-					enum drm_mm_search_flags sflags,
-					enum drm_mm_allocator_flags aflags);
-
-/**
- * drm_mm_insert_node_in_range - ranged search for space and insert @node
- * @mm: drm_mm to allocate from
- * @node: preallocate node to insert
- * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @start: start of the allowed range for this node
- * @end: end of the allowed range for this node
- * @flags: flags to fine-tune the allocation
- *
- * This is a simplified version of drm_mm_insert_node_in_range_generic() with
- * @color set to 0.
- *
- * The preallocated node must be cleared to 0.
- *
- * Returns:
- * 0 on success, -ENOSPC if there's no suitable hole.
- */
-static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
-					      struct drm_mm_node *node,
-					      u64 size,
-					      u64 alignment,
-					      u64 start,
-					      u64 end,
-					      enum drm_mm_search_flags flags)
-{
-	return drm_mm_insert_node_in_range_generic(mm, node, size, alignment,
-						   0, start, end, flags,
-						   DRM_MM_CREATE_DEFAULT);
-}
+int drm_mm_insert_node_in_range(struct drm_mm *mm,
+				struct drm_mm_node *node,
+				u64 size,
+				u64 alignment,
+				unsigned long color,
+				u64 start,
+				u64 end,
+				enum drm_mm_insert_mode mode);
 
 /**
  * drm_mm_insert_node_generic - search for space and insert @node
@@ -333,8 +292,7 @@ static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
  * @size: size of the allocation
  * @alignment: alignment of the allocation
  * @color: opaque tag value to use for this node
- * @sflags: flags to fine-tune the allocation search
- * @aflags: flags to fine-tune the allocation behavior
+ * @mode: fine-tune the allocation search and placement
  *
  * The preallocated node must be cleared to 0.
  *
@@ -345,13 +303,11 @@ static inline int
 drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
 			   u64 size, u64 alignment,
 			   unsigned long color,
-			   enum drm_mm_search_flags sflags,
-			   enum drm_mm_allocator_flags aflags)
+			   enum drm_mm_insert_mode mode)
 {
-	return drm_mm_insert_node_in_range_generic(mm, node,
-						   size, alignment, 0,
-						   0, U64_MAX,
-						   sflags, aflags);
+	return drm_mm_insert_node_in_range(mm, node,
+					   size, alignment, color,
+					   0, U64_MAX, mode);
 }
 
 /**
@@ -359,8 +315,6 @@ drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
  * @mm: drm_mm to allocate from
  * @node: preallocate node to insert
  * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @flags: flags to fine-tune the allocation
  *
  * This is a simplified version of drm_mm_insert_node_generic() with @color set
  * to 0.
@@ -372,13 +326,9 @@ drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
  */
 static inline int drm_mm_insert_node(struct drm_mm *mm,
 				     struct drm_mm_node *node,
-				     u64 size,
-				     u64 alignment,
-				     enum drm_mm_search_flags flags)
+				     u64 size)
 {
-	return drm_mm_insert_node_generic(mm, node,
-					  size, alignment, 0,
-					  flags, DRM_MM_CREATE_DEFAULT);
+	return drm_mm_insert_node_generic(mm, node, size, 0, 0, 0);
 }
 
 void drm_mm_remove_node(struct drm_mm_node *node);
@@ -425,7 +375,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 struct drm_mm *mm,
 				 u64 size, u64 alignment, unsigned long color,
 				 u64 start, u64 end,
-				 unsigned int flags);
+				 enum drm_mm_insert_mode mode);
 
 /**
  * drm_mm_scan_init - initialize lru scanning
@@ -434,7 +384,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
  * @size: size of the allocation
  * @alignment: alignment of the allocation
  * @color: opaque tag value to use for the allocation
- * @flags: flags to specify how the allocation will be performed afterwards
+ * @mode: fine-tune the allocation search and placement
  *
  * This simply sets up the scanning routines with the parameters for the desired
  * hole.
@@ -448,12 +398,11 @@ static inline void drm_mm_scan_init(struct drm_mm_scan *scan,
 				    u64 size,
 				    u64 alignment,
 				    unsigned long color,
-				    unsigned int flags)
+				    enum drm_mm_insert_mode mode)
 {
 	drm_mm_scan_init_with_range(scan, mm,
 				    size, alignment, color,
-				    0, U64_MAX,
-				    flags);
+				    0, U64_MAX, mode);
 }
 
 bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3] drm: Improve drm_mm search (and fix topdown allocation) with rbtrees

[Chris Wilson]

The drm_mm range manager claimed to support top-down insertion, but it
was neither searching for the top-most hole that could fit the
allocation request nor fitting the request to the hole correctly.

In order to search the range efficiently, we create a secondary index
for the holes using either their size or their address. This index
allows us to find the smallest hole or the hole at the bottom or top of
the range efficiently, whilst keeping the hole stack to rapidly service
evictions.

v2: Search for holes both high and low. Rename flags to mode.
v3: Discover rb_entry_safe() and use it!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c  |  16 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c |  20 +-
 drivers/gpu/drm/armada/armada_gem.c          |   4 +-
 drivers/gpu/drm/drm_mm.c                     | 511 +++++++++++++++------------
 drivers/gpu/drm/drm_vma_manager.c            |   3 +-
 drivers/gpu/drm/etnaviv/etnaviv_mmu.c        |   8 +-
 drivers/gpu/drm/i915/i915_gem.c              |  10 +-
 drivers/gpu/drm/i915/i915_gem_evict.c        |   9 +-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c   |   5 +-
 drivers/gpu/drm/i915/i915_gem_gtt.c          |  39 +-
 drivers/gpu/drm/i915/i915_gem_stolen.c       |   6 +-
 drivers/gpu/drm/msm/msm_gem.c                |   3 +-
 drivers/gpu/drm/msm/msm_gem_vma.c            |   3 +-
 drivers/gpu/drm/selftests/test-drm_mm.c      |  58 ++-
 drivers/gpu/drm/sis/sis_mm.c                 |   6 +-
 drivers/gpu/drm/tegra/gem.c                  |   4 +-
 drivers/gpu/drm/ttm/ttm_bo_manager.c         |  18 +-
 drivers/gpu/drm/vc4/vc4_crtc.c               |   2 +-
 drivers/gpu/drm/vc4/vc4_hvs.c                |   3 +-
 drivers/gpu/drm/vc4/vc4_plane.c              |   6 +-
 drivers/gpu/drm/via/via_mm.c                 |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c       |  10 +-
 include/drm/drm_mm.h                         | 135 +++----
 23 files changed, 434 insertions(+), 449 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 00f46b0e076d..d841fcb2e709 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -97,8 +97,7 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 {
 	struct amdgpu_gtt_mgr *mgr = man->priv;
 	struct drm_mm_node *node = mem->mm_node;
-	enum drm_mm_search_flags sflags = DRM_MM_SEARCH_BEST;
-	enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+	enum drm_mm_insert_mode mode;
 	unsigned long fpfn, lpfn;
 	int r;
 
@@ -115,15 +114,14 @@ int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 	else
 		lpfn = man->size;
 
-	if (place && place->flags & TTM_PL_FLAG_TOPDOWN) {
-		sflags = DRM_MM_SEARCH_BELOW;
-		aflags = DRM_MM_CREATE_TOP;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (place && place->mode & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
 
 	spin_lock(&mgr->lock);
-	r = drm_mm_insert_node_in_range_generic(&mgr->mm, node, mem->num_pages,
-						mem->page_alignment, 0,
-						fpfn, lpfn, sflags, aflags);
+	r = drm_mm_insert_node_in_range(&mgr->mm, node,
+					mem->num_pages, mem->page_alignment, 0,
+					fpfn, lpfn, mode);
 	spin_unlock(&mgr->lock);
 
 	if (!r) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
index d710226a0fff..5f106ad815ce 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
@@ -97,8 +97,7 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 	struct amdgpu_vram_mgr *mgr = man->priv;
 	struct drm_mm *mm = &mgr->mm;
 	struct drm_mm_node *nodes;
-	enum drm_mm_search_flags sflags = DRM_MM_SEARCH_DEFAULT;
-	enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+	enum drm_mm_insert_mode mode;
 	unsigned long lpfn, num_nodes, pages_per_node, pages_left;
 	unsigned i;
 	int r;
@@ -121,10 +120,9 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 	if (!nodes)
 		return -ENOMEM;
 
-	if (place->flags & TTM_PL_FLAG_TOPDOWN) {
-		sflags = DRM_MM_SEARCH_BELOW;
-		aflags = DRM_MM_CREATE_TOP;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (place->flags & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
 
 	pages_left = mem->num_pages;
 
@@ -135,13 +133,11 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
 
 		if (pages == pages_per_node)
 			alignment = pages_per_node;
-		else
-			sflags |= DRM_MM_SEARCH_BEST;
 
-		r = drm_mm_insert_node_in_range_generic(mm, &nodes[i], pages,
-							alignment, 0,
-							place->fpfn, lpfn,
-							sflags, aflags);
+		r = drm_mm_insert_node_in_range(mm, &nodes[i],
+						pages, alignment, 0,
+						place->fpfn, lpfn,
+						mode);
 		if (unlikely(r))
 			goto error;
 
diff --git a/drivers/gpu/drm/armada/armada_gem.c b/drivers/gpu/drm/armada/armada_gem.c
index 768087ddb046..65b029bc84b5 100644
--- a/drivers/gpu/drm/armada/armada_gem.c
+++ b/drivers/gpu/drm/armada/armada_gem.c
@@ -149,8 +149,8 @@ armada_gem_linear_back(struct drm_device *dev, struct armada_gem_object *obj)
 			return -ENOSPC;
 
 		mutex_lock(&priv->linear_lock);
-		ret = drm_mm_insert_node(&priv->linear, node, size, align,
-					 DRM_MM_SEARCH_DEFAULT);
+		ret = drm_mm_insert_node_generic(&priv->linear, node,
+						 size, align, 0, 0);
 		mutex_unlock(&priv->linear_lock);
 		if (ret) {
 			kfree(node);
diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 58a7e3bbe130..960c9cd7dc23 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -92,14 +92,6 @@
  * some basic allocator dumpers for debugging.
  */
 
-static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
-						u64 size,
-						u64 alignment,
-						unsigned long color,
-						u64 start,
-						u64 end,
-						enum drm_mm_search_flags flags);
-
 #ifdef CONFIG_DRM_DEBUG_MM
 #include <linux/stackdepot.h>
 
@@ -221,69 +213,47 @@ static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node,
 			    &drm_mm_interval_tree_augment);
 }
 
-static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
-				 struct drm_mm_node *node,
-				 u64 size, u64 alignment,
-				 unsigned long color,
-				 u64 range_start, u64 range_end,
-				 enum drm_mm_allocator_flags flags)
+#define RB_INSERT(root, member, expr) do { \
+	struct rb_node **link = &root.rb_node, *rb = NULL; \
+	u64 x = expr(node); \
+	while (*link) { \
+		rb = *link; \
+		if (x < expr(rb_entry(rb, struct drm_mm_node, member))) \
+			link = &rb->rb_left; \
+		else \
+			link = &rb->rb_right; \
+	} \
+	rb_link_node(&node->member, rb, link); \
+	rb_insert_color(&node->member, &root); \
+} while (0)
+
+#define HOLE_SIZE(NODE) ((NODE)->hole_size)
+#define HOLE_ADDR(NODE) (__drm_mm_hole_node_start(NODE))
+
+static void add_hole(struct drm_mm_node *node)
 {
-	struct drm_mm *mm = hole_node->mm;
-	u64 hole_start = drm_mm_hole_node_start(hole_node);
-	u64 hole_end = drm_mm_hole_node_end(hole_node);
-	u64 adj_start = hole_start;
-	u64 adj_end = hole_end;
-
-	DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
-
-	if (mm->color_adjust)
-		mm->color_adjust(hole_node, color, &adj_start, &adj_end);
-
-	adj_start = max(adj_start, range_start);
-	adj_end = min(adj_end, range_end);
-
-	if (flags & DRM_MM_CREATE_TOP)
-		adj_start = adj_end - size;
-
-	if (alignment) {
-		u64 rem;
-
-		div64_u64_rem(adj_start, alignment, &rem);
-		if (rem) {
-			if (flags & DRM_MM_CREATE_TOP)
-				adj_start -= rem;
-			else
-				adj_start += alignment - rem;
-		}
-	}
-
-	if (adj_start == hole_start) {
-		hole_node->hole_follows = 0;
-		list_del(&hole_node->hole_stack);
-	}
+	struct drm_mm *mm = node->mm;
 
-	node->start = adj_start;
-	node->size = size;
-	node->mm = mm;
-	node->color = color;
-	node->allocated = 1;
+	node->hole_size =
+		__drm_mm_hole_node_end(node) - __drm_mm_hole_node_start(node);
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(node));
 
-	list_add(&node->node_list, &hole_node->node_list);
+	RB_INSERT(mm->holes_size, rb_hole_size, HOLE_SIZE);
+	RB_INSERT(mm->holes_addr, rb_hole_addr, HOLE_ADDR);
 
-	drm_mm_interval_tree_add_node(hole_node, node);
+	list_add(&node->hole_stack, &mm->hole_stack);
+}
 
-	DRM_MM_BUG_ON(node->start < range_start);
-	DRM_MM_BUG_ON(node->start < adj_start);
-	DRM_MM_BUG_ON(node->start + node->size > adj_end);
-	DRM_MM_BUG_ON(node->start + node->size > range_end);
+static void rm_hole(struct drm_mm_node *node)
+{
+	DRM_MM_BUG_ON(!drm_mm_hole_follows(node));
 
-	node->hole_follows = 0;
-	if (__drm_mm_hole_node_start(node) < hole_end) {
-		list_add(&node->hole_stack, &mm->hole_stack);
-		node->hole_follows = 1;
-	}
+	list_del(&node->hole_stack);
+	rb_erase(&node->rb_hole_size, &node->mm->holes_size);
+	rb_erase(&node->rb_hole_addr, &node->mm->holes_addr);
+	node->hole_size = 0;
 
-	save_stack(node);
+	DRM_MM_BUG_ON(drm_mm_hole_follows(node));
 }
 
 /**
@@ -313,7 +283,7 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 
 	/* Find the relevant hole to add our node to */
 	hole = drm_mm_interval_tree_iter_first(&mm->interval_tree,
-					       node->start, ~(u64)0);
+					       node->start, U64_MAX);
 	if (hole) {
 		if (hole->start < end)
 			return -ENOSPC;
@@ -321,12 +291,12 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 		hole = list_entry(drm_mm_nodes(mm), typeof(*hole), node_list);
 	}
 
-	hole = list_last_entry(&hole->node_list, typeof(*hole), node_list);
+	hole = list_prev_entry(hole, node_list);
 	if (!drm_mm_hole_follows(hole))
 		return -ENOSPC;
 
 	adj_start = hole_start = __drm_mm_hole_node_start(hole);
-	adj_end = hole_end = __drm_mm_hole_node_end(hole);
+	adj_end = hole_end = hole_start + hole->hole_size;
 
 	if (mm->color_adjust)
 		mm->color_adjust(hole, node->color, &adj_start, &adj_end);
@@ -335,31 +305,128 @@ int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
 		return -ENOSPC;
 
 	node->mm = mm;
-	node->allocated = 1;
 
 	list_add(&node->node_list, &hole->node_list);
-
 	drm_mm_interval_tree_add_node(hole, node);
+	node->allocated = true;
+	node->hole_size = 0;
+
+	rm_hole(hole);
+	if (node->start > hole_start)
+		add_hole(hole);
+	if (end < hole_end)
+		add_hole(node);
+
+	save_stack(node);
+	return 0;
+}
+EXPORT_SYMBOL(drm_mm_reserve_node);
+
+static inline struct drm_mm_node *rb_hole_size_to_node(struct rb_node *rb)
+{
+	return rb_entry_safe(rb, struct drm_mm_node, rb_hole_size);
+}
+
+static inline struct drm_mm_node *rb_hole_addr_to_node(struct rb_node *rb)
+{
+	return rb_entry_safe(rb, struct drm_mm_node, rb_hole_addr);
+}
 
-	if (node->start == hole_start) {
-		hole->hole_follows = 0;
-		list_del(&hole->hole_stack);
+static inline u64 rb_hole_size(struct rb_node *rb)
+{
+	return rb_entry(rb, struct drm_mm_node, rb_hole_size)->hole_size;
+}
+
+static struct drm_mm_node *best_hole(struct drm_mm *mm, u64 size)
+{
+	struct rb_node *best = NULL;
+	struct rb_node **link = &mm->holes_size.rb_node;
+
+	while (*link) {
+		struct rb_node *rb = *link;
+		if (size <= rb_hole_size(rb)) {
+			link = &rb->rb_left;
+			best = rb;
+		} else {
+			link = &rb->rb_right;
+		}
 	}
 
-	node->hole_follows = 0;
-	if (end != hole_end) {
-		list_add(&node->hole_stack, &mm->hole_stack);
-		node->hole_follows = 1;
+	return rb_hole_size_to_node(best);
+}
+
+static struct drm_mm_node *find_hole(struct drm_mm *mm, u64 addr)
+{
+	struct drm_mm_node *node = NULL;
+	struct rb_node **link = &mm->holes_addr.rb_node;
+
+	while (*link) {
+		u64 hole_start;
+
+		node = rb_hole_addr_to_node(*link);
+		hole_start = __drm_mm_hole_node_start(node);
+
+		if (addr < hole_start)
+			link = &node->rb_hole_addr.rb_left;
+		else if (addr > hole_start + node->hole_size)
+			link = &node->rb_hole_addr.rb_right;
+		else
+			break;
 	}
 
-	save_stack(node);
+	return node;
+}
 
-	return 0;
+static struct drm_mm_node *
+first_hole(struct drm_mm *mm,
+	   u64 start, u64 end, u64 size,
+	   enum drm_mm_insert_mode mode)
+{
+	if (RB_EMPTY_ROOT(&mm->holes_size))
+		return NULL;
+
+	switch (mode) {
+	default:
+	case DRM_MM_INSERT_BEST:
+		return best_hole(mm, size);
+
+	case DRM_MM_INSERT_LOW:
+		return find_hole(mm, start);
+
+	case DRM_MM_INSERT_HIGH:
+		return find_hole(mm, end);
+
+	case DRM_MM_INSERT_EVICT:
+		return list_first_entry_or_null(&mm->hole_stack,
+						struct drm_mm_node,
+						hole_stack);
+	}
+}
+
+static struct drm_mm_node *
+next_hole(struct drm_mm *mm,
+	  struct drm_mm_node *node,
+	  enum drm_mm_insert_mode mode)
+{
+	switch (mode) {
+	default:
+	case DRM_MM_INSERT_BEST:
+		return rb_hole_size_to_node(rb_next(&node->rb_hole_size));
+
+	case DRM_MM_INSERT_LOW:
+		return rb_hole_addr_to_node(rb_next(&node->rb_hole_addr));
+
+	case DRM_MM_INSERT_HIGH:
+		return rb_hole_addr_to_node(rb_prev(&node->rb_hole_addr));
+
+	case DRM_MM_INSERT_EVICT:
+		node = list_next_entry(node, hole_stack);
+		return &node->hole_stack == &mm->hole_stack ? NULL : node;
+	}
 }
-EXPORT_SYMBOL(drm_mm_reserve_node);
 
 /**
- * drm_mm_insert_node_in_range_generic - ranged search for space and insert @node
+ * drm_mm_insert_node_in_range - ranged search for space and insert @node
  * @mm: drm_mm to allocate from
  * @node: preallocate node to insert
  * @size: size of the allocation
@@ -367,38 +434,104 @@ EXPORT_SYMBOL(drm_mm_reserve_node);
  * @color: opaque tag value to use for this node
  * @start: start of the allowed range for this node
  * @end: end of the allowed range for this node
- * @sflags: flags to fine-tune the allocation search
- * @aflags: flags to fine-tune the allocation behavior
+ * @mode: fine-tune the allocation search and placement
  *
  * The preallocated node must be cleared to 0.
  *
  * Returns:
  * 0 on success, -ENOSPC if there's no suitable hole.
  */
-int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *node,
-					u64 size, u64 alignment,
-					unsigned long color,
-					u64 start, u64 end,
-					enum drm_mm_search_flags sflags,
-					enum drm_mm_allocator_flags aflags)
+int drm_mm_insert_node_in_range(struct drm_mm * const mm,
+				struct drm_mm_node * const node,
+				u64 size, u64 alignment,
+				unsigned long color,
+				u64 range_start, u64 range_end,
+				enum drm_mm_insert_mode mode)
 {
-	struct drm_mm_node *hole_node;
+	struct drm_mm_node *hole;
+	u64 remainder_mask;
 
-	if (WARN_ON(size == 0))
-		return -EINVAL;
+	DRM_MM_BUG_ON(range_start >= range_end);
 
-	hole_node = drm_mm_search_free_in_range_generic(mm,
-							size, alignment, color,
-							start, end, sflags);
-	if (!hole_node)
+	if (unlikely(size == 0 || range_end - range_start < size))
 		return -ENOSPC;
 
-	drm_mm_insert_helper(hole_node, node,
-			     size, alignment, color,
-			     start, end, aflags);
-	return 0;
+	if (alignment <= 1)
+		alignment = 0;
+
+	remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
+	for (hole = first_hole(mm, range_start, range_end, size, mode); hole;
+	     hole = next_hole(mm, hole, mode)) {
+		u64 hole_start = __drm_mm_hole_node_start(hole);
+		u64 hole_end = hole_start + hole->hole_size;
+		u64 adj_start, adj_end;
+		u64 col_start, col_end;
+
+		if (mode == DRM_MM_INSERT_LOW && hole_start >= range_end)
+			break;
+
+		if (mode == DRM_MM_INSERT_HIGH && hole_end <= range_start)
+			break;
+
+		col_start = hole_start;
+		col_end = hole_end;
+		if (mm->color_adjust)
+			mm->color_adjust(hole, color, &col_start, &col_end);
+
+		adj_start = max(col_start, range_start);
+		adj_end = min(col_end, range_end);
+
+		if (adj_end <= adj_start || adj_end - adj_start < size)
+			continue;
+
+		if (mode == DRM_MM_INSERT_HIGH)
+			adj_start = adj_end - size;
+
+		if (alignment) {
+			u64 rem;
+
+			if (likely(remainder_mask))
+				rem = adj_start & remainder_mask;
+			else
+				div64_u64_rem(adj_start, alignment, &rem);
+			if (rem) {
+				adj_start -= rem;
+				if (mode != DRM_MM_INSERT_HIGH)
+					adj_start += alignment;
+
+				if (adj_start < max(col_start, range_start) ||
+				    min(col_end, range_end) - adj_start < size)
+					continue;
+
+				if (adj_end <= adj_start ||
+				    adj_end - adj_start < size)
+					continue;
+			}
+		}
+
+		node->mm = mm;
+		node->size = size;
+		node->start = adj_start;
+		node->color = color;
+		node->hole_size = 0;
+
+		list_add(&node->node_list, &hole->node_list);
+		drm_mm_interval_tree_add_node(hole, node);
+		node->allocated = true;
+
+		rm_hole(hole);
+		if (adj_start > hole_start)
+			add_hole(hole);
+		if (adj_start + size < hole_end)
+			add_hole(node);
+
+		save_stack(node);
+		return 0;
+	}
+
+	return -ENOSPC;
 }
-EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic);
+EXPORT_SYMBOL(drm_mm_insert_node_in_range);
 
 /**
  * drm_mm_remove_node - Remove a memory node from the allocator.
@@ -416,92 +549,20 @@ void drm_mm_remove_node(struct drm_mm_node *node)
 	DRM_MM_BUG_ON(!node->allocated);
 	DRM_MM_BUG_ON(node->scanned_block);
 
-	prev_node =
-	    list_entry(node->node_list.prev, struct drm_mm_node, node_list);
-
-	if (drm_mm_hole_follows(node)) {
-		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) ==
-			      __drm_mm_hole_node_end(node));
-		list_del(&node->hole_stack);
-	} else {
-		DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) !=
-			      __drm_mm_hole_node_end(node));
-	}
+	prev_node = list_prev_entry(node, node_list);
 
-	if (!drm_mm_hole_follows(prev_node)) {
-		prev_node->hole_follows = 1;
-		list_add(&prev_node->hole_stack, &mm->hole_stack);
-	} else
-		list_move(&prev_node->hole_stack, &mm->hole_stack);
+	if (drm_mm_hole_follows(node))
+		rm_hole(node);
 
 	drm_mm_interval_tree_remove(node, &mm->interval_tree);
 	list_del(&node->node_list);
-	node->allocated = 0;
-}
-EXPORT_SYMBOL(drm_mm_remove_node);
-
-static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment)
-{
-	if (end - start < size)
-		return 0;
-
-	if (alignment) {
-		u64 rem;
-
-		div64_u64_rem(start, alignment, &rem);
-		if (rem)
-			start += alignment - rem;
-	}
-
-	return end >= start + size;
-}
-
-static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
-							u64 size,
-							u64 alignment,
-							unsigned long color,
-							u64 start,
-							u64 end,
-							enum drm_mm_search_flags flags)
-{
-	struct drm_mm_node *entry;
-	struct drm_mm_node *best;
-	u64 adj_start;
-	u64 adj_end;
-	u64 best_size;
-
-	DRM_MM_BUG_ON(mm->scan_active);
-
-	best = NULL;
-	best_size = ~0UL;
-
-	__drm_mm_for_each_hole(entry, mm, adj_start, adj_end,
-			       flags & DRM_MM_SEARCH_BELOW) {
-		u64 hole_size = adj_end - adj_start;
-
-		if (mm->color_adjust) {
-			mm->color_adjust(entry, color, &adj_start, &adj_end);
-			if (adj_end <= adj_start)
-				continue;
-		}
-
-		adj_start = max(adj_start, start);
-		adj_end = min(adj_end, end);
-
-		if (!check_free_hole(adj_start, adj_end, size, alignment))
-			continue;
-
-		if (!(flags & DRM_MM_SEARCH_BEST))
-			return entry;
-
-		if (hole_size < best_size) {
-			best = entry;
-			best_size = hole_size;
-		}
-	}
+	node->allocated = false;
 
-	return best;
+	if (drm_mm_hole_follows(prev_node))
+		rm_hole(prev_node);
+	add_hole(prev_node);
 }
+EXPORT_SYMBOL(drm_mm_remove_node);
 
 /**
  * drm_mm_replace_node - move an allocation from @old to @new
@@ -516,18 +577,23 @@ void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new)
 {
 	DRM_MM_BUG_ON(!old->allocated);
 
+	*new = *old;
+
 	list_replace(&old->node_list, &new->node_list);
-	list_replace(&old->hole_stack, &new->hole_stack);
 	rb_replace_node(&old->rb, &new->rb, &old->mm->interval_tree);
-	new->hole_follows = old->hole_follows;
-	new->mm = old->mm;
-	new->start = old->start;
-	new->size = old->size;
-	new->color = old->color;
-	new->__subtree_last = old->__subtree_last;
-
-	old->allocated = 0;
-	new->allocated = 1;
+
+	if (drm_mm_hole_follows(old)) {
+		list_replace(&old->hole_stack, &new->hole_stack);
+		rb_replace_node(&old->rb_hole_size,
+				&new->rb_hole_size,
+				&old->mm->holes_size);
+		rb_replace_node(&old->rb_hole_addr,
+				&new->rb_hole_addr,
+				&old->mm->holes_addr);
+	}
+
+	old->allocated = false;
+	new->allocated = true;
 }
 EXPORT_SYMBOL(drm_mm_replace_node);
 
@@ -570,7 +636,7 @@ EXPORT_SYMBOL(drm_mm_replace_node);
  * @color: opaque tag value to use for the allocation
  * @start: start of the allowed range for the allocation
  * @end: end of the allowed range for the allocation
- * @flags: flags to specify how the allocation will be performed afterwards
+ * @mode: fine-tune the allocation search and placement
  *
  * This simply sets up the scanning routines with the parameters for the desired
  * hole.
@@ -586,7 +652,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 unsigned long color,
 				 u64 start,
 				 u64 end,
-				 unsigned int flags)
+				 enum drm_mm_insert_mode mode)
 {
 	DRM_MM_BUG_ON(start >= end);
 	DRM_MM_BUG_ON(!size || size > end - start);
@@ -601,7 +667,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 	scan->alignment = alignment;
 	scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
 	scan->size = size;
-	scan->flags = flags;
+	scan->mode = mode;
 
 	DRM_MM_BUG_ON(end <= start);
 	scan->range_start = start;
@@ -659,7 +725,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 	if (adj_end <= adj_start || adj_end - adj_start < scan->size)
 		return false;
 
-	if (scan->flags == DRM_MM_CREATE_TOP)
+	if (scan->mode == DRM_MM_INSERT_HIGH)
 		adj_start = adj_end - scan->size;
 
 	if (scan->alignment) {
@@ -671,7 +737,7 @@ bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
 			div64_u64_rem(adj_start, scan->alignment, &rem);
 		if (rem) {
 			adj_start -= rem;
-			if (scan->flags != DRM_MM_CREATE_TOP)
+			if (scan->mode != DRM_MM_INSERT_HIGH)
 				adj_start += scan->alignment;
 			if (adj_start < max(col_start, scan->range_start) ||
 			    min(col_end, scan->range_end) - adj_start < scan->size)
@@ -765,7 +831,7 @@ struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan)
 
 	hole = list_first_entry(&mm->hole_stack, typeof(*hole), hole_stack);
 	hole_start = __drm_mm_hole_node_start(hole);
-	hole_end = __drm_mm_hole_node_end(hole);
+	hole_end = hole_start + hole->hole_size;
 
 	DRM_MM_BUG_ON(hole_start > scan->hit_start);
 	DRM_MM_BUG_ON(hole_end < scan->hit_end);
@@ -792,21 +858,22 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 {
 	DRM_MM_BUG_ON(start + size <= start);
 
+	mm->color_adjust = NULL;
+
 	INIT_LIST_HEAD(&mm->hole_stack);
-	mm->scan_active = 0;
+	mm->interval_tree = RB_ROOT;
+	mm->holes_size = RB_ROOT;
+	mm->holes_addr = RB_ROOT;
 
 	/* Clever trick to avoid a special case in the free hole tracking. */
 	INIT_LIST_HEAD(&mm->head_node.node_list);
-	mm->head_node.allocated = 0;
-	mm->head_node.hole_follows = 1;
+	mm->head_node.allocated = false;
 	mm->head_node.mm = mm;
 	mm->head_node.start = start + size;
-	mm->head_node.size = start - mm->head_node.start;
-	list_add_tail(&mm->head_node.hole_stack, &mm->hole_stack);
-
-	mm->interval_tree = RB_ROOT;
+	mm->head_node.size = -size;
+	add_hole(&mm->head_node);
 
-	mm->color_adjust = NULL;
+	mm->scan_active = 0;
 }
 EXPORT_SYMBOL(drm_mm_init);
 
@@ -828,18 +895,16 @@ EXPORT_SYMBOL(drm_mm_takedown);
 static u64 drm_mm_debug_hole(const struct drm_mm_node *entry,
 			     const char *prefix)
 {
-	u64 hole_start, hole_end, hole_size;
-
-	if (entry->hole_follows) {
-		hole_start = drm_mm_hole_node_start(entry);
-		hole_end = drm_mm_hole_node_end(entry);
-		hole_size = hole_end - hole_start;
-		pr_debug("%s %#llx-%#llx: %llu: free\n", prefix, hole_start,
-			 hole_end, hole_size);
-		return hole_size;
+	u64 start, size;
+
+	size = entry->hole_size;
+	if (size) {
+		start = drm_mm_hole_node_start(entry);
+		pr_debug("%s %#llx-%#llx: %llu: free\n",
+			 prefix, start, start + size, size);
 	}
 
-	return 0;
+	return size;
 }
 
 /**
@@ -870,18 +935,16 @@ EXPORT_SYMBOL(drm_mm_debug_table);
 #if defined(CONFIG_DEBUG_FS)
 static u64 drm_mm_dump_hole(struct seq_file *m, const struct drm_mm_node *entry)
 {
-	u64 hole_start, hole_end, hole_size;
-
-	if (entry->hole_follows) {
-		hole_start = drm_mm_hole_node_start(entry);
-		hole_end = drm_mm_hole_node_end(entry);
-		hole_size = hole_end - hole_start;
-		seq_printf(m, "%#018llx-%#018llx: %llu: free\n", hole_start,
-			   hole_end, hole_size);
-		return hole_size;
+	u64 start, size;
+
+	size = entry->hole_size;
+	if (size) {
+		start = drm_mm_hole_node_start(entry);
+		seq_printf(m, "%#018llx-%#018llx: %llu: free\n",
+			   start, start + size, size);
 	}
 
-	return 0;
+	return size;
 }
 
 /**
diff --git a/drivers/gpu/drm/drm_vma_manager.c b/drivers/gpu/drm/drm_vma_manager.c
index 20cc33d1bfc1..d9100b565198 100644
--- a/drivers/gpu/drm/drm_vma_manager.c
+++ b/drivers/gpu/drm/drm_vma_manager.c
@@ -212,8 +212,7 @@ int drm_vma_offset_add(struct drm_vma_offset_manager *mgr,
 		goto out_unlock;
 	}
 
-	ret = drm_mm_insert_node(&mgr->vm_addr_space_mm, &node->vm_node,
-				 pages, 0, DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node(&mgr->vm_addr_space_mm, &node->vm_node, pages);
 	if (ret)
 		goto out_unlock;
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
index 2dae3169ce48..69ab98c143dc 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_mmu.c
@@ -107,6 +107,7 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 				   struct drm_mm_node *node, size_t size)
 {
 	struct etnaviv_vram_mapping *free = NULL;
+	enum drm_mm_insert_mode mode = DRM_MM_INSERT_LOW;
 	int ret;
 
 	lockdep_assert_held(&mmu->lock);
@@ -118,8 +119,9 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 		bool found;
 
 		ret = drm_mm_insert_node_in_range(&mmu->mm, node,
-			size, 0, mmu->last_iova, ~0UL,
-			DRM_MM_SEARCH_DEFAULT);
+						  size, 0, 0,
+						  mmu->last_iova, U64_MAX,
+						  mode);
 
 		if (ret != -ENOSPC)
 			break;
@@ -187,6 +189,8 @@ static int etnaviv_iommu_find_iova(struct etnaviv_iommu *mmu,
 			list_del_init(&m->scan_node);
 		}
 
+		mode = DRM_MM_INSERT_EVICT;
+
 		/*
 		 * We removed enough mappings so that the new allocation will
 		 * succeed.  Ensure that the MMU will be flushed before the
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 1490db6a22a2..691f2a87ca1b 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -69,12 +69,10 @@ insert_mappable_node(struct i915_ggtt *ggtt,
                      struct drm_mm_node *node, u32 size)
 {
 	memset(node, 0, sizeof(*node));
-	return drm_mm_insert_node_in_range_generic(&ggtt->base.mm, node,
-						   size, 0,
-						   I915_COLOR_UNEVICTABLE,
-						   0, ggtt->mappable_end,
-						   DRM_MM_SEARCH_DEFAULT,
-						   DRM_MM_CREATE_DEFAULT);
+	return drm_mm_insert_node_in_range(&ggtt->base.mm, node,
+					   size, 0, I915_COLOR_UNEVICTABLE,
+					   0, ggtt->mappable_end,
+					   DRM_MM_INSERT_LOW);
 }
 
 static void
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 50129ec1caab..f92f63e8749a 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -109,6 +109,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	}, **phase;
 	struct i915_vma *vma, *next;
 	struct drm_mm_node *node;
+	enum drm_mm_insert_mode mode;
 	int ret;
 
 	lockdep_assert_held(&vm->i915->drm.struct_mutex);
@@ -127,10 +128,14 @@ i915_gem_evict_something(struct i915_address_space *vm,
 	 * On each list, the oldest objects lie at the HEAD with the freshest
 	 * object on the TAIL.
 	 */
+	mode = DRM_MM_INSERT_BEST;
+	if (flags & PIN_HIGH)
+		mode = DRM_MM_INSERT_HIGH;
+	if (flags & PIN_MAPPABLE)
+		mode = DRM_MM_INSERT_LOW;
 	drm_mm_scan_init_with_range(&scan, &vm->mm,
 				    min_size, alignment, cache_level,
-				    start, end,
-				    flags & PIN_HIGH ? DRM_MM_CREATE_TOP : 0);
+				    start, end, mode);
 
 	/* Retire before we search the active list. Although we have
 	 * reasonable accuracy in our retirement lists, we may have
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index c64438f8171c..3e435caf7221 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -436,12 +436,11 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
 					       PIN_MAPPABLE | PIN_NONBLOCK);
 		if (IS_ERR(vma)) {
 			memset(&cache->node, 0, sizeof(cache->node));
-			ret = drm_mm_insert_node_in_range_generic
+			ret = drm_mm_insert_node_in_range
 				(&ggtt->base.mm, &cache->node,
 				 4096, 0, I915_COLOR_UNEVICTABLE,
 				 0, ggtt->mappable_end,
-				 DRM_MM_SEARCH_DEFAULT,
-				 DRM_MM_CREATE_DEFAULT);
+				 DRM_MM_INSERT_LOW);
 			if (ret) /* no inactive aperture space, use cpu reloc */
 				return NULL;
 		} else {
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index ad8c5dec0370..d35c46b477a6 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2722,12 +2722,10 @@ int i915_gem_init_ggtt(struct drm_i915_private *dev_priv)
 		return ret;
 
 	/* Reserve a mappable slot for our lockless error capture */
-	ret = drm_mm_insert_node_in_range_generic(&ggtt->base.mm,
-						  &ggtt->error_capture,
-						  4096, 0,
-						  I915_COLOR_UNEVICTABLE,
-						  0, ggtt->mappable_end,
-						  0, 0);
+	ret = drm_mm_insert_node_in_range(&ggtt->base.mm, &ggtt->error_capture,
+					  4096, 0, I915_COLOR_UNEVICTABLE,
+					  0, ggtt->mappable_end,
+					  DRM_MM_INSERT_LOW);
 	if (ret)
 		return ret;
 
@@ -3542,18 +3540,16 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 			u64 size, u64 alignment, unsigned long colour,
 			u64 start, u64 end, unsigned int flags)
 {
-	u32 search_flag, alloc_flag;
+	enum drm_mm_insert_mode mode;
 	int err;
 
 	lockdep_assert_held(&vm->i915->drm.struct_mutex);
 
-	if (flags & PIN_HIGH) {
-		search_flag = DRM_MM_SEARCH_BELOW;
-		alloc_flag = DRM_MM_CREATE_TOP;
-	} else {
-		search_flag = DRM_MM_SEARCH_DEFAULT;
-		alloc_flag = DRM_MM_CREATE_DEFAULT;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (flags & PIN_HIGH)
+		mode = DRM_MM_INSERT_HIGH;
+	if (flags & PIN_MAPPABLE)
+		mode = DRM_MM_INSERT_LOW;
 
 	/* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks,
 	 * so we know that we always have a minimum alignment of 4096.
@@ -3565,10 +3561,9 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 	if (alignment <= 4096)
 		alignment = 0;
 
-	err = drm_mm_insert_node_in_range_generic(&vm->mm, node,
-						  size, alignment, colour,
-						  start, end,
-						  search_flag, alloc_flag);
+	err = drm_mm_insert_node_in_range(&vm->mm, node,
+					  size, alignment, colour,
+					  start, end, mode);
 	if (err != -ENOSPC)
 		return err;
 
@@ -3577,9 +3572,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 	if (err)
 		return err;
 
-	search_flag = DRM_MM_SEARCH_DEFAULT;
-	return drm_mm_insert_node_in_range_generic(&vm->mm, node,
-						   size, alignment, colour,
-						   start, end,
-						   search_flag, alloc_flag);
+	return drm_mm_insert_node_in_range(&vm->mm, node,
+					   size, alignment, colour,
+					   start, end, DRM_MM_INSERT_EVICT);
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index f1a80bfa9919..f6d507faf83b 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -55,9 +55,9 @@ int i915_gem_stolen_insert_node_in_range(struct drm_i915_private *dev_priv,
 		return -ENODEV;
 
 	mutex_lock(&dev_priv->mm.stolen_lock);
-	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node, size,
-					  alignment, start, end,
-					  DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node,
+					  size, alignment, 0,
+					  start, end, DRM_MM_INSERT_BEST);
 	mutex_unlock(&dev_priv->mm.stolen_lock);
 
 	return ret;
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index cd06cfd94687..412669062cb7 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -54,8 +54,7 @@ static struct page **get_pages_vram(struct drm_gem_object *obj,
 	if (!p)
 		return ERR_PTR(-ENOMEM);
 
-	ret = drm_mm_insert_node(&priv->vram.mm, msm_obj->vram_node,
-			npages, 0, DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node(&priv->vram.mm, msm_obj->vram_node, npages);
 	if (ret) {
 		drm_free_large(p);
 		return ERR_PTR(ret);
diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c
index a311d26ccb21..b654eca7636a 100644
--- a/drivers/gpu/drm/msm/msm_gem_vma.c
+++ b/drivers/gpu/drm/msm/msm_gem_vma.c
@@ -45,8 +45,7 @@ msm_gem_map_vma(struct msm_gem_address_space *aspace,
 	if (WARN_ON(drm_mm_node_allocated(&vma->node)))
 		return 0;
 
-	ret = drm_mm_insert_node(&aspace->mm, &vma->node, npages,
-			0, DRM_MM_SEARCH_DEFAULT);
+	ret = drm_mm_insert_node(&aspace->mm, &vma->node, npages);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index d852b009f646..4ae76534ea97 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -22,23 +22,24 @@ static unsigned int max_iterations = 8192;
 static unsigned int max_prime = 128;
 
 enum {
-	DEFAULT,
-	TOPDOWN,
 	BEST,
+	BOTTOMUP,
+	TOPDOWN,
+	EVICT,
 };
 
 static const struct insert_mode {
 	const char *name;
-	unsigned int search_flags;
-	unsigned int create_flags;
+	enum drm_mm_insert_mode mode;
 } insert_modes[] = {
-	[DEFAULT] = { "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
-	[TOPDOWN] = { "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
-	[BEST] = { "best", DRM_MM_SEARCH_BEST, DRM_MM_CREATE_DEFAULT },
+	[BEST] = { "best", DRM_MM_INSERT_BEST },
+	[BOTTOMUP] = { "bottom-up", DRM_MM_INSERT_LOW },
+	[TOPDOWN] = { "top-down", DRM_MM_INSERT_HIGH },
+	[EVICT] = { "evict", DRM_MM_INSERT_EVICT },
 	{}
 }, evict_modes[] = {
-	{ "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
-	{ "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
+	{ "bottom-up", DRM_MM_INSERT_LOW },
+	{ "top-down", DRM_MM_INSERT_HIGH },
 	{}
 };
 
@@ -522,8 +523,7 @@ static bool expect_insert(struct drm_mm *mm, struct drm_mm_node *node,
 
 	err = drm_mm_insert_node_generic(mm, node,
 					 size, alignment, color,
-					 mode->search_flags,
-					 mode->create_flags);
+					 mode->mode);
 	if (err) {
 		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) failed with err=%d\n",
 		       size, alignment, color, mode->name, err);
@@ -543,7 +543,7 @@ static bool expect_insert_fail(struct drm_mm *mm, u64 size)
 	struct drm_mm_node tmp = {};
 	int err;
 
-	err = drm_mm_insert_node(mm, &tmp, size, 0, DRM_MM_SEARCH_DEFAULT);
+	err = drm_mm_insert_node(mm, &tmp, size);
 	if (likely(err == -ENOSPC))
 		return true;
 
@@ -749,11 +749,10 @@ static bool expect_insert_in_range(struct drm_mm *mm, struct drm_mm_node *node,
 {
 	int err;
 
-	err = drm_mm_insert_node_in_range_generic(mm, node,
-						  size, alignment, color,
-						  range_start, range_end,
-						  mode->search_flags,
-						  mode->create_flags);
+	err = drm_mm_insert_node_in_range(mm, node,
+					  size, alignment, color,
+					  range_start, range_end,
+					  mode->mode);
 	if (err) {
 		pr_err("insert (size=%llu, alignment=%llu, color=%lu, mode=%s) nto range [%llx, %llx] failed with err=%d\n",
 		       size, alignment, color, mode->name,
@@ -777,11 +776,10 @@ static bool expect_insert_in_range_fail(struct drm_mm *mm,
 	struct drm_mm_node tmp = {};
 	int err;
 
-	err = drm_mm_insert_node_in_range_generic(mm, &tmp,
-						  size, 0, 0,
-						  range_start, range_end,
-						  DRM_MM_SEARCH_DEFAULT,
-						  DRM_MM_CREATE_DEFAULT);
+	err = drm_mm_insert_node_in_range(mm, &tmp,
+					  size, 0, 0,
+					  range_start, range_end,
+					  0);
 	if (likely(err == -ENOSPC))
 		return true;
 
@@ -1314,7 +1312,7 @@ static int evict_something(struct drm_mm *mm,
 	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, 0,
 				    range_start, range_end,
-				    mode->create_flags);
+				    mode->mode);
 	if (!evict_nodes(&scan,
 			 nodes, order, count, false,
 			 &evict_list))
@@ -1322,8 +1320,7 @@ static int evict_something(struct drm_mm *mm,
 
 	memset(&tmp, 0, sizeof(tmp));
 	err = drm_mm_insert_node_generic(mm, &tmp, size, alignment, 0,
-					 mode->search_flags,
-					 mode->create_flags);
+					 DRM_MM_INSERT_EVICT);
 	if (err) {
 		pr_err("Failed to insert into eviction hole: size=%d, align=%d\n",
 		       size, alignment);
@@ -1398,8 +1395,7 @@ static int igt_evict(void *ignored)
 	ret = -EINVAL;
 	drm_mm_init(&mm, 0, size);
 	for (n = 0; n < size; n++) {
-		err = drm_mm_insert_node(&mm, &nodes[n].node, 1, 0,
-					 DRM_MM_SEARCH_DEFAULT);
+		err = drm_mm_insert_node(&mm, &nodes[n].node, 1);
 		if (err) {
 			pr_err("insert failed, step %d\n", n);
 			ret = err;
@@ -1507,8 +1503,7 @@ static int igt_evict_range(void *ignored)
 	ret = -EINVAL;
 	drm_mm_init(&mm, 0, size);
 	for (n = 0; n < size; n++) {
-		err = drm_mm_insert_node(&mm, &nodes[n].node, 1, 0,
-					 DRM_MM_SEARCH_DEFAULT);
+		err = drm_mm_insert_node(&mm, &nodes[n].node, 1);
 		if (err) {
 			pr_err("insert failed, step %d\n", n);
 			ret = err;
@@ -1894,7 +1889,7 @@ static int evict_color(struct drm_mm *mm,
 	drm_mm_scan_init_with_range(&scan, mm,
 				    size, alignment, color,
 				    range_start, range_end,
-				    mode->create_flags);
+				    mode->mode);
 	if (!evict_nodes(&scan,
 			 nodes, order, count, true,
 			 &evict_list))
@@ -1902,8 +1897,7 @@ static int evict_color(struct drm_mm *mm,
 
 	memset(&tmp, 0, sizeof(tmp));
 	err = drm_mm_insert_node_generic(mm, &tmp, size, alignment, color,
-					 mode->search_flags,
-					 mode->create_flags);
+					 DRM_MM_INSERT_EVICT);
 	if (err) {
 		pr_err("Failed to insert into eviction hole: size=%d, align=%d, color=%lu, err=%d\n",
 		       size, alignment, color, err);
diff --git a/drivers/gpu/drm/sis/sis_mm.c b/drivers/gpu/drm/sis/sis_mm.c
index 03defda77766..1622db24cd39 100644
--- a/drivers/gpu/drm/sis/sis_mm.c
+++ b/drivers/gpu/drm/sis/sis_mm.c
@@ -109,8 +109,7 @@ static int sis_drm_alloc(struct drm_device *dev, struct drm_file *file,
 	if (pool == AGP_TYPE) {
 		retval = drm_mm_insert_node(&dev_priv->agp_mm,
 					    &item->mm_node,
-					    mem->size, 0,
-					    DRM_MM_SEARCH_DEFAULT);
+					    mem->size);
 		offset = item->mm_node.start;
 	} else {
 #if defined(CONFIG_FB_SIS) || defined(CONFIG_FB_SIS_MODULE)
@@ -122,8 +121,7 @@ static int sis_drm_alloc(struct drm_device *dev, struct drm_file *file,
 #else
 		retval = drm_mm_insert_node(&dev_priv->vram_mm,
 					    &item->mm_node,
-					    mem->size, 0,
-					    DRM_MM_SEARCH_DEFAULT);
+					    mem->size);
 		offset = item->mm_node.start;
 #endif
 	}
diff --git a/drivers/gpu/drm/tegra/gem.c b/drivers/gpu/drm/tegra/gem.c
index c08e5279eeac..a6051d6c6de6 100644
--- a/drivers/gpu/drm/tegra/gem.c
+++ b/drivers/gpu/drm/tegra/gem.c
@@ -128,8 +128,8 @@ static int tegra_bo_iommu_map(struct tegra_drm *tegra, struct tegra_bo *bo)
 	if (!bo->mm)
 		return -ENOMEM;
 
-	err = drm_mm_insert_node_generic(&tegra->mm, bo->mm, bo->gem.size,
-					 PAGE_SIZE, 0, 0, 0);
+	err = drm_mm_insert_node_generic(&tegra->mm,
+					 bo->mm, bo->gem.size, PAGE_SIZE, 0, 0);
 	if (err < 0) {
 		dev_err(tegra->drm->dev, "out of I/O virtual memory: %zd\n",
 			err);
diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c
index aea6a01500e1..493fe6b33353 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_manager.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c
@@ -54,9 +54,8 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man,
 {
 	struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
 	struct drm_mm *mm = &rman->mm;
-	struct drm_mm_node *node = NULL;
-	enum drm_mm_search_flags sflags = DRM_MM_SEARCH_BEST;
-	enum drm_mm_allocator_flags aflags = DRM_MM_CREATE_DEFAULT;
+	struct drm_mm_node *node;
+	enum drm_mm_insert_mode mode;
 	unsigned long lpfn;
 	int ret;
 
@@ -68,16 +67,15 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man,
 	if (!node)
 		return -ENOMEM;
 
-	if (place->flags & TTM_PL_FLAG_TOPDOWN) {
-		sflags = DRM_MM_SEARCH_BELOW;
-		aflags = DRM_MM_CREATE_TOP;
-	}
+	mode = DRM_MM_INSERT_BEST;
+	if (place->flags & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
 
 	spin_lock(&rman->lock);
-	ret = drm_mm_insert_node_in_range_generic(mm, node, mem->num_pages,
+	ret = drm_mm_insert_node_in_range(mm, node,
+					  mem->num_pages,
 					  mem->page_alignment, 0,
-					  place->fpfn, lpfn,
-					  sflags, aflags);
+					  place->fpfn, lpfn, mode);
 	spin_unlock(&rman->lock);
 
 	if (unlikely(ret)) {
diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c
index a0fd3e66bc4b..baacdf62c83e 100644
--- a/drivers/gpu/drm/vc4/vc4_crtc.c
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -589,7 +589,7 @@ static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
 
 	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
 	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
-				 dlist_count, 1, 0);
+				 dlist_count);
 	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
 	if (ret)
 		return ret;
diff --git a/drivers/gpu/drm/vc4/vc4_hvs.c b/drivers/gpu/drm/vc4/vc4_hvs.c
index 6fbab1c82cb1..4aba0fa56289 100644
--- a/drivers/gpu/drm/vc4/vc4_hvs.c
+++ b/drivers/gpu/drm/vc4/vc4_hvs.c
@@ -141,8 +141,7 @@ static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
 	int ret, i;
 	u32 __iomem *dst_kernel;
 
-	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS, 1,
-				 0);
+	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
 	if (ret) {
 		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
 			  ret);
diff --git a/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c
index 110d1518f5d5..c1f06897136b 100644
--- a/drivers/gpu/drm/vc4/vc4_plane.c
+++ b/drivers/gpu/drm/vc4/vc4_plane.c
@@ -514,9 +514,9 @@ static int vc4_plane_mode_set(struct drm_plane *plane,
 	if (lbm_size) {
 		if (!vc4_state->lbm.allocated) {
 			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
-			ret = drm_mm_insert_node(&vc4->hvs->lbm_mm,
-						 &vc4_state->lbm,
-						 lbm_size, 32, 0);
+			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
+							 &vc4_state->lbm,
+							 lbm_size, 32, 0, 0);
 			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
 		} else {
 			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
diff --git a/drivers/gpu/drm/via/via_mm.c b/drivers/gpu/drm/via/via_mm.c
index a04ef1c992d9..4217d66a5cc6 100644
--- a/drivers/gpu/drm/via/via_mm.c
+++ b/drivers/gpu/drm/via/via_mm.c
@@ -140,11 +140,11 @@ int via_mem_alloc(struct drm_device *dev, void *data,
 	if (mem->type == VIA_MEM_AGP)
 		retval = drm_mm_insert_node(&dev_priv->agp_mm,
 					    &item->mm_node,
-					    tmpSize, 0, DRM_MM_SEARCH_DEFAULT);
+					    tmpSize);
 	else
 		retval = drm_mm_insert_node(&dev_priv->vram_mm,
 					    &item->mm_node,
-					    tmpSize, 0, DRM_MM_SEARCH_DEFAULT);
+					    tmpSize);
 	if (retval)
 		goto fail_alloc;
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
index aa04fb0159a7..77cb7c627e09 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cmdbuf.c
@@ -673,16 +673,10 @@ static bool vmw_cmdbuf_try_alloc(struct vmw_cmdbuf_man *man,
  
 	memset(info->node, 0, sizeof(*info->node));
 	spin_lock_bh(&man->lock);
-	ret = drm_mm_insert_node_generic(&man->mm, info->node, info->page_size,
-					 0, 0,
-					 DRM_MM_SEARCH_DEFAULT,
-					 DRM_MM_CREATE_DEFAULT);
+	ret = drm_mm_insert_node(&man->mm, info->node, info->page_size);
 	if (ret) {
 		vmw_cmdbuf_man_process(man);
-		ret = drm_mm_insert_node_generic(&man->mm, info->node,
-						 info->page_size, 0, 0,
-						 DRM_MM_SEARCH_DEFAULT,
-						 DRM_MM_CREATE_DEFAULT);
+		ret = drm_mm_insert_node(&man->mm, info->node, info->page_size);
 	}
 
 	spin_unlock_bh(&man->lock);
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 92ec5759caae..6558cd08d7c9 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -55,32 +55,27 @@
 #define DRM_MM_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr)
 #endif
 
-enum drm_mm_search_flags {
-	DRM_MM_SEARCH_DEFAULT =		0,
-	DRM_MM_SEARCH_BEST =		1 << 0,
-	DRM_MM_SEARCH_BELOW =		1 << 1,
+enum drm_mm_insert_mode {
+	DRM_MM_INSERT_BEST = 0,
+	DRM_MM_INSERT_LOW,
+	DRM_MM_INSERT_HIGH,
+	DRM_MM_INSERT_EVICT,
 };
 
-enum drm_mm_allocator_flags {
-	DRM_MM_CREATE_DEFAULT =		0,
-	DRM_MM_CREATE_TOP =		1 << 0,
-};
-
-#define DRM_MM_BOTTOMUP DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT
-#define DRM_MM_TOPDOWN DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP
-
 struct drm_mm_node {
+	struct drm_mm *mm;
 	struct list_head node_list;
 	struct list_head hole_stack;
 	struct rb_node rb;
-	unsigned hole_follows : 1;
-	unsigned allocated : 1;
-	bool scanned_block : 1;
-	unsigned long color;
+	struct rb_node rb_hole_size;
+	struct rb_node rb_hole_addr;
 	u64 start;
 	u64 size;
 	u64 __subtree_last;
-	struct drm_mm *mm;
+	u64 hole_size;
+	unsigned long color;
+	bool allocated : 1;
+	bool scanned_block : 1;
 #ifdef CONFIG_DRM_DEBUG_MM
 	depot_stack_handle_t stack;
 #endif
@@ -94,6 +89,8 @@ struct drm_mm {
 	struct drm_mm_node head_node;
 	/* Keep an interval_tree for fast lookup of drm_mm_nodes by address. */
 	struct rb_root interval_tree;
+	struct rb_root holes_size;
+	struct rb_root holes_addr;
 
 	void (*color_adjust)(const struct drm_mm_node *node,
 			     unsigned long color,
@@ -116,7 +113,7 @@ struct drm_mm_scan {
 	u64 hit_end;
 
 	unsigned long color;
-	unsigned int flags;
+	enum drm_mm_insert_mode mode;
 };
 
 /**
@@ -168,7 +165,7 @@ static inline bool drm_mm_initialized(const struct drm_mm *mm)
  */
 static inline bool drm_mm_hole_follows(const struct drm_mm_node *node)
 {
-	return node->hole_follows;
+	return node->hole_size;
 }
 
 static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
@@ -251,14 +248,6 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
 #define drm_mm_for_each_node_safe(entry, next, mm) \
 	list_for_each_entry_safe(entry, next, drm_mm_nodes(mm), node_list)
 
-#define __drm_mm_for_each_hole(entry, mm, hole_start, hole_end, backwards) \
-	for (entry = list_entry((backwards) ? (mm)->hole_stack.prev : (mm)->hole_stack.next, struct drm_mm_node, hole_stack); \
-	     &entry->hole_stack != &(mm)->hole_stack ? \
-	     hole_start = drm_mm_hole_node_start(entry), \
-	     hole_end = drm_mm_hole_node_end(entry), \
-	     1 : 0; \
-	     entry = list_entry((backwards) ? entry->hole_stack.prev : entry->hole_stack.next, struct drm_mm_node, hole_stack))
-
 /**
  * drm_mm_for_each_hole - iterator to walk over all holes
  * @entry: drm_mm_node used internally to track progress
@@ -274,57 +263,27 @@ static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
  * Implementation Note:
  * We need to inline list_for_each_entry in order to be able to set hole_start
  * and hole_end on each iteration while keeping the macro sane.
- *
- * The __drm_mm_for_each_hole version is similar, but with added support for
- * going backwards.
  */
-#define drm_mm_for_each_hole(entry, mm, hole_start, hole_end) \
-	__drm_mm_for_each_hole(entry, mm, hole_start, hole_end, 0)
+#define drm_mm_for_each_hole(pos, mm, hole_start, hole_end) \
+	for (pos = list_first_entry(&(mm)->hole_stack, \
+				    typeof(*pos), hole_stack); \
+	     &pos->hole_stack != &(mm)->hole_stack ? \
+	     hole_start = drm_mm_hole_node_start(pos), \
+	     hole_end = hole_start + pos->hole_size : 0; \
+	     pos = list_next_entry(pos, hole_stack))
 
 /*
  * Basic range manager support (drm_mm.c)
  */
 int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node);
-int drm_mm_insert_node_in_range_generic(struct drm_mm *mm,
-					struct drm_mm_node *node,
-					u64 size,
-					u64 alignment,
-					unsigned long color,
-					u64 start,
-					u64 end,
-					enum drm_mm_search_flags sflags,
-					enum drm_mm_allocator_flags aflags);
-
-/**
- * drm_mm_insert_node_in_range - ranged search for space and insert @node
- * @mm: drm_mm to allocate from
- * @node: preallocate node to insert
- * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @start: start of the allowed range for this node
- * @end: end of the allowed range for this node
- * @flags: flags to fine-tune the allocation
- *
- * This is a simplified version of drm_mm_insert_node_in_range_generic() with
- * @color set to 0.
- *
- * The preallocated node must be cleared to 0.
- *
- * Returns:
- * 0 on success, -ENOSPC if there's no suitable hole.
- */
-static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
-					      struct drm_mm_node *node,
-					      u64 size,
-					      u64 alignment,
-					      u64 start,
-					      u64 end,
-					      enum drm_mm_search_flags flags)
-{
-	return drm_mm_insert_node_in_range_generic(mm, node, size, alignment,
-						   0, start, end, flags,
-						   DRM_MM_CREATE_DEFAULT);
-}
+int drm_mm_insert_node_in_range(struct drm_mm *mm,
+				struct drm_mm_node *node,
+				u64 size,
+				u64 alignment,
+				unsigned long color,
+				u64 start,
+				u64 end,
+				enum drm_mm_insert_mode mode);
 
 /**
  * drm_mm_insert_node_generic - search for space and insert @node
@@ -333,8 +292,7 @@ static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
  * @size: size of the allocation
  * @alignment: alignment of the allocation
  * @color: opaque tag value to use for this node
- * @sflags: flags to fine-tune the allocation search
- * @aflags: flags to fine-tune the allocation behavior
+ * @mode: fine-tune the allocation search and placement
  *
  * The preallocated node must be cleared to 0.
  *
@@ -345,13 +303,11 @@ static inline int
 drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
 			   u64 size, u64 alignment,
 			   unsigned long color,
-			   enum drm_mm_search_flags sflags,
-			   enum drm_mm_allocator_flags aflags)
+			   enum drm_mm_insert_mode mode)
 {
-	return drm_mm_insert_node_in_range_generic(mm, node,
-						   size, alignment, 0,
-						   0, U64_MAX,
-						   sflags, aflags);
+	return drm_mm_insert_node_in_range(mm, node,
+					   size, alignment, color,
+					   0, U64_MAX, mode);
 }
 
 /**
@@ -359,8 +315,6 @@ drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
  * @mm: drm_mm to allocate from
  * @node: preallocate node to insert
  * @size: size of the allocation
- * @alignment: alignment of the allocation
- * @flags: flags to fine-tune the allocation
  *
  * This is a simplified version of drm_mm_insert_node_generic() with @color set
  * to 0.
@@ -372,13 +326,9 @@ drm_mm_insert_node_generic(struct drm_mm *mm, struct drm_mm_node *node,
  */
 static inline int drm_mm_insert_node(struct drm_mm *mm,
 				     struct drm_mm_node *node,
-				     u64 size,
-				     u64 alignment,
-				     enum drm_mm_search_flags flags)
+				     u64 size)
 {
-	return drm_mm_insert_node_generic(mm, node,
-					  size, alignment, 0,
-					  flags, DRM_MM_CREATE_DEFAULT);
+	return drm_mm_insert_node_generic(mm, node, size, 0, 0, 0);
 }
 
 void drm_mm_remove_node(struct drm_mm_node *node);
@@ -425,7 +375,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
 				 struct drm_mm *mm,
 				 u64 size, u64 alignment, unsigned long color,
 				 u64 start, u64 end,
-				 unsigned int flags);
+				 enum drm_mm_insert_mode mode);
 
 /**
  * drm_mm_scan_init - initialize lru scanning
@@ -434,7 +384,7 @@ void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
  * @size: size of the allocation
  * @alignment: alignment of the allocation
  * @color: opaque tag value to use for the allocation
- * @flags: flags to specify how the allocation will be performed afterwards
+ * @mode: fine-tune the allocation search and placement
  *
  * This simply sets up the scanning routines with the parameters for the desired
  * hole.
@@ -448,12 +398,11 @@ static inline void drm_mm_scan_init(struct drm_mm_scan *scan,
 				    u64 size,
 				    u64 alignment,
 				    unsigned long color,
-				    unsigned int flags)
+				    enum drm_mm_insert_mode mode)
 {
 	drm_mm_scan_init_with_range(scan, mm,
 				    size, alignment, color,
-				    0, U64_MAX,
-				    flags);
+				    0, U64_MAX, mode);
 }
 
 bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
-- 
2.11.0

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH v3 38/38] drm: kselftest for drm_mm and bottom-up allocation

[Chris Wilson]

Check that if we request bottom-up allocation from drm_mm_insert_node()
we receive the next available hole from the bottom.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 100 +++++++++++++++++++++++++++
 2 files changed, 101 insertions(+)

diff --git a/drivers/gpu/drm/selftests/drm_mm_selftests.h b/drivers/gpu/drm/selftests/drm_mm_selftests.h
index 6a4575fdc1c0..37bbdac52896 100644
--- a/drivers/gpu/drm/selftests/drm_mm_selftests.h
+++ b/drivers/gpu/drm/selftests/drm_mm_selftests.h
@@ -17,6 +17,7 @@ selftest(align32, igt_align32)
 selftest(align64, igt_align64)
 selftest(evict, igt_evict)
 selftest(evict_range, igt_evict_range)
+selftest(bottomup, igt_bottomup)
 selftest(topdown, igt_topdown)
 selftest(color, igt_color)
 selftest(color_evict, igt_color_evict)
diff --git a/drivers/gpu/drm/selftests/test-drm_mm.c b/drivers/gpu/drm/selftests/test-drm_mm.c
index 26e7001c846d..52a1a1d0ba7c 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1682,6 +1682,106 @@ static int igt_topdown(void *ignored)
 	return ret;
 }
 
+static int igt_bottomup(void *ignored)
+{
+	const struct insert_mode *bottomup = &insert_modes[BOTTOMUP];
+	DRM_RND_STATE(prng, random_seed);
+	const unsigned int count = 8192;
+	unsigned int size;
+	unsigned long *bitmap;
+	struct drm_mm mm;
+	struct drm_mm_node *nodes, *node, *next;
+	unsigned int *order, n, m, o = 0;
+	int ret;
+
+	/* Like igt_topdown, but instead of searching for the last hole,
+	 * we search for the first.
+	 */
+
+	ret = -ENOMEM;
+	nodes = vzalloc(count * sizeof(*nodes));
+	if (!nodes)
+		goto err;
+
+	bitmap = kzalloc(count / BITS_PER_LONG * sizeof(unsigned long),
+			 GFP_TEMPORARY);
+	if (!bitmap)
+		goto err_nodes;
+
+	order = drm_random_order(count, &prng);
+	if (!order)
+		goto err_bitmap;
+
+	ret = -EINVAL;
+	for (size = 1; size <= 64; size <<= 1) {
+		drm_mm_init(&mm, 0, size*count);
+		for (n = 0; n < count; n++) {
+			if (!expect_insert(&mm, &nodes[n],
+					   size, 0, n,
+					   bottomup)) {
+				pr_err("bottomup insert failed, size %u step %d\n", size, n);
+				goto out;
+			}
+
+			if (!assert_one_hole(&mm, size*(n + 1), size*count))
+				goto out;
+		}
+
+		if (!assert_continuous(&mm, size))
+			goto out;
+
+		drm_random_reorder(order, count, &prng);
+		for_each_prime_number(n, min(count, max_prime)) {
+			for (m = 0; m < n; m++) {
+				node = &nodes[order[(o + m) % count]];
+				drm_mm_remove_node(node);
+				__set_bit(node_index(node), bitmap);
+			}
+
+			for (m = 0; m < n; m++) {
+				unsigned int first;
+
+				node = &nodes[order[(o + m) % count]];
+				if (!expect_insert(&mm, node,
+						   size, 0, 0,
+						   bottomup)) {
+					pr_err("insert failed, step %d/%d\n", m, n);
+					goto out;
+				}
+
+				first = find_first_bit(bitmap, count);
+				if (node_index(node) != first) {
+					pr_err("node %d/%d not inserted into bottom hole, expected %d, found %d\n",
+					       m, n, first, node_index(node));
+					goto out;
+				}
+				__clear_bit(first, bitmap);
+			}
+
+			DRM_MM_BUG_ON(find_first_bit(bitmap, count) != count);
+
+			o += n;
+		}
+
+		drm_mm_for_each_node_safe(node, next, &mm)
+			drm_mm_remove_node(node);
+		DRM_MM_BUG_ON(!drm_mm_clean(&mm));
+	}
+
+	ret = 0;
+out:
+	drm_mm_for_each_node_safe(node, next, &mm)
+		drm_mm_remove_node(node);
+	drm_mm_takedown(&mm);
+	kfree(order);
+err_bitmap:
+	kfree(bitmap);
+err_nodes:
+	vfree(nodes);
+err:
+	return ret;
+}
+
 static void separate_adjacent_colors(const struct drm_mm_node *node,
 				     unsigned long color,
 				     u64 *start,
-- 
2.11.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✗ Fi.CI.BAT: warning for series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting

[Patchwork]

== Series Details ==

Series: series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting
URL   : https://patchwork.freedesktop.org/series/16934/
State : warning

== Summary ==

Series 16934v1 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/16934/revisions/1/mbox/

Test kms_pipe_crc_basic:
        Subgroup suspend-read-crc-pipe-b:
                pass       -> SKIP       (fi-bxt-j4205)

fi-bdw-5557u     total:247  pass:233  dwarn:0   dfail:0   fail:0   skip:14 
fi-bsw-n3050     total:247  pass:208  dwarn:0   dfail:0   fail:0   skip:39 
fi-bxt-j4205     total:247  pass:222  dwarn:0   dfail:0   fail:0   skip:25 
fi-bxt-t5700     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-j1900     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-n2820     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-hsw-4770      total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-hsw-4770r     total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-ilk-650       total:247  pass:195  dwarn:0   dfail:0   fail:0   skip:52 
fi-ivb-3520m     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-ivb-3770      total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-kbl-7500u     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6260u     total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-skl-6700hq    total:247  pass:227  dwarn:0   dfail:0   fail:0   skip:20 
fi-skl-6700k     total:247  pass:224  dwarn:3   dfail:0   fail:0   skip:20 
fi-skl-6770hq    total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-snb-2520m     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-snb-2600      total:247  pass:215  dwarn:0   dfail:0   fail:0   skip:32 

705f1e8fef81d504f0032df8e21bdc2e74850b3a drm-tip: 2016y-12m-16d-15h-40m-02s UTC integration manifest
7c45ab9 drm: kselftest for drm_mm and bottom-up allocation
f0dc10e drm: Improve drm_mm search (and fix topdown allocation) with rbtrees
9d67b2a drm: Use drm_mm_insert_node_in_range_generic() for everyone
7c31969 drm: Apply range restriction after color adjustment when allocation
a389ffb drm: Wrap drm_mm_node.hole_follows
8f3d65a drm: Apply tight eviction scanning to color_adjust
713e51e drm: Simplify drm_mm scan-list manipulation
3f23d58 drm: Optimise power-of-two alignments in drm_mm_scan_add_block()
4b7fe06 drm: Compute tight evictions for drm_mm_scan
7b43f39 drm: Fix application of color vs range restriction when scanning drm_mm
46ca14a drm: Unconditionally do the range check in drm_mm_scan_add_block()
52d3c07 drm: Rename prev_node to hole in drm_mm_scan_add_block()
843ed0b drm: Extract struct drm_mm_scan from struct drm_mm
bd683ec drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()
54b09d7 drm: Simplify drm_mm_clean()
5d0b310 drm: Detect overflow in drm_mm_reserve_node()
68a3fad drm: Fix kerneldoc for drm_mm_scan_remove_block()
58b86fb drm: Promote drm_mm alignment to u64
200a726 drm/i915: Build DRM range manager selftests for CI
8b85ea3 drm: kselftest for drm_mm and restricted color eviction
bc18224 drm: kselftest for drm_mm and color eviction
aecad93 drm: kselftest for drm_mm and color adjustment
72de4a2 drm: kselftest for drm_mm and top-down allocation
79acff5 drm: kselftest for drm_mm and range restricted eviction
de7a869 drm: kselftest for drm_mm and eviction
7c2f995 drm: kselftest for drm_mm and alignment
a45196d drm: kselftest for drm_mm_insert_node_in_range()
63e7304 drm: kselftest for drm_mm_replace_node()
f859694 drm: kselftest for drm_mm_insert_node()
2f810a4 drm: kselftest for drm_mm_reserve_node()
203008d drm: kselftest for drm_mm_debug()
fd28f93 drm: kselftest for drm_mm_init()
a23ba8f drm: Add some kselftests for the DRM range manager (struct drm_mm)
d742b04 drm: Add a simple generator of random permutations
95765bd lib: Add a simple prime number generator
80ec238 drm: Compile time enabling for asserts in drm_mm
a451708 drm: Use drm_mm_nodes() as shorthand for the list of nodes under struct drm_mm
a4063bcc drm/i915: Use the MRU stack search after evicting

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3312/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✓ Fi.CI.BAT: success for series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev3)

[Patchwork]

== Series Details ==

Series: series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev3)
URL   : https://patchwork.freedesktop.org/series/16934/
State : success

== Summary ==

Series 16934v3 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/16934/revisions/3/mbox/


fi-bdw-5557u     total:247  pass:233  dwarn:0   dfail:0   fail:0   skip:14 
fi-bsw-n3050     total:247  pass:208  dwarn:0   dfail:0   fail:0   skip:39 
fi-bxt-j4205     total:247  pass:222  dwarn:0   dfail:0   fail:0   skip:25 
fi-bxt-t5700     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-j1900     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-n2820     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-hsw-4770      total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-hsw-4770r     total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-ilk-650       total:247  pass:195  dwarn:0   dfail:0   fail:0   skip:52 
fi-ivb-3520m     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-ivb-3770      total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-kbl-7500u     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6260u     total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-skl-6700hq    total:247  pass:227  dwarn:0   dfail:0   fail:0   skip:20 
fi-skl-6700k     total:247  pass:224  dwarn:3   dfail:0   fail:0   skip:20 
fi-skl-6770hq    total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-snb-2520m     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-snb-2600      total:247  pass:215  dwarn:0   dfail:0   fail:0   skip:32 

5415795ad73044788900bdc350076e17ac7f2fa1 drm-tip: 2016y-12m-16d-21h-26m-45s UTC integration manifest
4a05059 drm: kselftest for drm_mm and bottom-up allocation
bc5aece drm: Improve drm_mm search (and fix topdown allocation) with rbtrees
ca2762b drm: Use drm_mm_insert_node_in_range_generic() for everyone
942de50 drm: Apply range restriction after color adjustment when allocation
3147fb9 drm: Wrap drm_mm_node.hole_follows
2d206ac drm: Apply tight eviction scanning to color_adjust
d810876 drm: Simplify drm_mm scan-list manipulation
e96fdad drm: Optimise power-of-two alignments in drm_mm_scan_add_block()
0f46044 drm: Compute tight evictions for drm_mm_scan
e59ad38 drm: Fix application of color vs range restriction when scanning drm_mm
03dd696 drm: Unconditionally do the range check in drm_mm_scan_add_block()
af1bc2d drm: Rename prev_node to hole in drm_mm_scan_add_block()
2b5a909 drm: Extract struct drm_mm_scan from struct drm_mm
8e10673 drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()
61c2b53 drm: Simplify drm_mm_clean()
15bc0f1 drm: Detect overflow in drm_mm_reserve_node()
45f7011 drm: Fix kerneldoc for drm_mm_scan_remove_block()
ac09eaf drm: Promote drm_mm alignment to u64
de66129 drm/i915: Build DRM range manager selftests for CI
2dd8fd3 drm: kselftest for drm_mm and restricted color eviction
8ed6e72 drm: kselftest for drm_mm and color eviction
3cc2340 drm: kselftest for drm_mm and color adjustment
76fe015 drm: kselftest for drm_mm and top-down allocation
fe6cff2 drm: kselftest for drm_mm and range restricted eviction
d98df8c drm: kselftest for drm_mm and eviction
4b7f07f drm: kselftest for drm_mm and alignment
df9a5e8 drm: kselftest for drm_mm_insert_node_in_range()
0f79417 drm: kselftest for drm_mm_replace_node()
1d1356d drm: kselftest for drm_mm_insert_node()
9f836ff drm: kselftest for drm_mm_reserve_node()
23c5177 drm: kselftest for drm_mm_debug()
bf387b8 drm: kselftest for drm_mm_init()
762efae drm: Add some kselftests for the DRM range manager (struct drm_mm)
4632b13 drm: Add a simple generator of random permutations
21cb20f lib: Add a simple prime number generator
ab6e80c drm: Compile time enabling for asserts in drm_mm
e3f4a70 drm: Use drm_mm_nodes() as shorthand for the list of nodes under struct drm_mm
7f6b373 drm/i915: Use the MRU stack search after evicting

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3314/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✗ Fi.CI.BAT: warning for series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev4)

[Patchwork]

== Series Details ==

Series: series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev4)
URL   : https://patchwork.freedesktop.org/series/16934/
State : warning

== Summary ==

Series 16934v4 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/16934/revisions/4/mbox/

Test drv_module_reload:
        Subgroup basic-reload-inject:
                pass       -> DMESG-WARN (fi-ilk-650)

fi-bdw-5557u     total:247  pass:233  dwarn:0   dfail:0   fail:0   skip:14 
fi-bsw-n3050     total:247  pass:208  dwarn:0   dfail:0   fail:0   skip:39 
fi-bxt-j4205     total:247  pass:222  dwarn:0   dfail:0   fail:0   skip:25 
fi-byt-j1900     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-n2820     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-hsw-4770      total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-hsw-4770r     total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-ilk-650       total:247  pass:194  dwarn:1   dfail:0   fail:0   skip:52 
fi-ivb-3520m     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-ivb-3770      total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-kbl-7500u     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6260u     total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-skl-6700hq    total:247  pass:227  dwarn:0   dfail:0   fail:0   skip:20 
fi-skl-6700k     total:247  pass:224  dwarn:3   dfail:0   fail:0   skip:20 
fi-skl-6770hq    total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-snb-2520m     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-snb-2600      total:247  pass:215  dwarn:0   dfail:0   fail:0   skip:32 

5415795ad73044788900bdc350076e17ac7f2fa1 drm-tip: 2016y-12m-16d-21h-26m-45s UTC integration manifest
bed1a81 drm: kselftest for drm_mm and bottom-up allocation
215abbc drm: Improve drm_mm search (and fix topdown allocation) with rbtrees
6508e21 drm: Use drm_mm_insert_node_in_range_generic() for everyone
5337365 drm: Apply range restriction after color adjustment when allocation
63f5948 drm: Wrap drm_mm_node.hole_follows
11cdd09 drm: Apply tight eviction scanning to color_adjust
33d7c8e drm: Simplify drm_mm scan-list manipulation
59f7bd6 drm: Optimise power-of-two alignments in drm_mm_scan_add_block()
91bccb1 drm: Compute tight evictions for drm_mm_scan
263ce3e drm: Fix application of color vs range restriction when scanning drm_mm
9a18627 drm: Unconditionally do the range check in drm_mm_scan_add_block()
caf5cda drm: Rename prev_node to hole in drm_mm_scan_add_block()
209e185 drm: Extract struct drm_mm_scan from struct drm_mm
dcc23a0 drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()
0e4a541 drm: Simplify drm_mm_clean()
3a0d8ee drm: Detect overflow in drm_mm_reserve_node()
14fde39 drm: Fix kerneldoc for drm_mm_scan_remove_block()
806d7ea drm: Promote drm_mm alignment to u64
78e0633 drm/i915: Build DRM range manager selftests for CI
c487357 drm: kselftest for drm_mm and restricted color eviction
134b35b drm: kselftest for drm_mm and color eviction
6f756a7 drm: kselftest for drm_mm and color adjustment
18b416a drm: kselftest for drm_mm and top-down allocation
90c1063 drm: kselftest for drm_mm and range restricted eviction
dbf9b2b drm: kselftest for drm_mm and eviction
417d822 drm: kselftest for drm_mm and alignment
748b844 drm: kselftest for drm_mm_insert_node_in_range()
1d42270 drm: kselftest for drm_mm_replace_node()
bdf62ff drm: kselftest for drm_mm_insert_node()
9505659 drm: kselftest for drm_mm_reserve_node()
cc81ab2 drm: kselftest for drm_mm_debug()
3482524 drm: kselftest for drm_mm_init()
9e7080a drm: Add some kselftests for the DRM range manager (struct drm_mm)
8fe82c3 drm: Add a simple generator of random permutations
e318ceb lib: Add a simple prime number generator
94f3b40 drm: Compile time enabling for asserts in drm_mm
7d05a29 drm: Use drm_mm_nodes() as shorthand for the list of nodes under struct drm_mm
c6a98ad drm/i915: Use the MRU stack search after evicting

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3315/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✗ Fi.CI.BAT: failure for series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev5)

[Patchwork]

== Series Details ==

Series: series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev5)
URL   : https://patchwork.freedesktop.org/series/16934/
State : failure

== Summary ==

Series 16934v5 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/16934/revisions/5/mbox/

Test gem_sync:
        Subgroup basic-store-all:
                pass       -> FAIL       (fi-ivb-3520m)
Test kms_frontbuffer_tracking:
        Subgroup basic:
                pass       -> DMESG-WARN (fi-bxt-j4205)
Test kms_pipe_crc_basic:
        Subgroup suspend-read-crc-pipe-a:
                skip       -> PASS       (fi-bxt-j4205)
        Subgroup suspend-read-crc-pipe-b:
                skip       -> PASS       (fi-bxt-j4205)
        Subgroup suspend-read-crc-pipe-c:
                skip       -> PASS       (fi-bxt-j4205)
Test pm_backlight:
        Subgroup basic-brightness:
                skip       -> PASS       (fi-bxt-j4205)

fi-bdw-5557u     total:247  pass:233  dwarn:0   dfail:0   fail:0   skip:14 
fi-bsw-n3050     total:247  pass:208  dwarn:0   dfail:0   fail:0   skip:39 
fi-bxt-j4205     total:247  pass:225  dwarn:1   dfail:0   fail:0   skip:21 
fi-bxt-t5700     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-j1900     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-n2820     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-hsw-4770      total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-hsw-4770r     total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-ilk-650       total:247  pass:194  dwarn:1   dfail:0   fail:0   skip:52 
fi-ivb-3520m     total:247  pass:225  dwarn:0   dfail:0   fail:1   skip:21 
fi-ivb-3770      total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-kbl-7500u     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6260u     total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-skl-6700hq    total:247  pass:227  dwarn:0   dfail:0   fail:0   skip:20 
fi-skl-6700k     total:247  pass:224  dwarn:3   dfail:0   fail:0   skip:20 
fi-skl-6770hq    total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-snb-2520m     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-snb-2600      total:247  pass:215  dwarn:0   dfail:0   fail:0   skip:32 

2a932d085375c80a1bbb332799db3df9738e8eba drm-tip: 2016y-12m-18d-16h-31m-05s UTC integration manifest
068db82 drm: kselftest for drm_mm and bottom-up allocation
e9786359 drm: Improve drm_mm search (and fix topdown allocation) with rbtrees
281b584 drm: Use drm_mm_insert_node_in_range_generic() for everyone
1a2f2cd drm: Apply range restriction after color adjustment when allocation
bdca8c2 drm: Wrap drm_mm_node.hole_follows
b677fb7 drm: Apply tight eviction scanning to color_adjust
dc19ba8 drm: Simplify drm_mm scan-list manipulation
93942ee drm: Optimise power-of-two alignments in drm_mm_scan_add_block()
fbc42fe drm: Compute tight evictions for drm_mm_scan
d6fc99f drm: Fix application of color vs range restriction when scanning drm_mm
3afaf4c drm: Unconditionally do the range check in drm_mm_scan_add_block()
8c671b2 drm: Rename prev_node to hole in drm_mm_scan_add_block()
4b534c3 drm: Extract struct drm_mm_scan from struct drm_mm
cd896f2 drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()
02be89f drm: Simplify drm_mm_clean()
52f04d9 drm: Detect overflow in drm_mm_reserve_node()
d3299af drm: Fix kerneldoc for drm_mm_scan_remove_block()
77b0dfa drm: Promote drm_mm alignment to u64
248b2b0 drm/i915: Build DRM range manager selftests for CI
35614bf drm: kselftest for drm_mm and restricted color eviction
f877a1d drm: kselftest for drm_mm and color eviction
9d2af48 drm: kselftest for drm_mm and color adjustment
0f2b1bf drm: kselftest for drm_mm and top-down allocation
0b6084f drm: kselftest for drm_mm and range restricted eviction
e0bc8d6 drm: kselftest for drm_mm and eviction
3c6a4d0 drm: kselftest for drm_mm and alignment
ad50c72 drm: kselftest for drm_mm_insert_node_in_range()
5c9d9e4 drm: kselftest for drm_mm_replace_node()
f87b541 drm: kselftest for drm_mm_insert_node()
dd7199b drm: kselftest for drm_mm_reserve_node()
00ff33b drm: kselftest for drm_mm_debug()
d1e9919 drm: kselftest for drm_mm_init()
8cc6695 drm: Add some kselftests for the DRM range manager (struct drm_mm)
fdb84ec drm: Add a simple generator of random permutations
218822b lib: Add a simple prime number generator
a3777fc drm: Compile time enabling for asserts in drm_mm
0947fe8 drm: Use drm_mm_nodes() as shorthand for the list of nodes under struct drm_mm
cf9f618 drm/i915: Use the MRU stack search after evicting

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3323/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✓ Fi.CI.BAT: success for series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev6)

[Patchwork]

== Series Details ==

Series: series starting with [v3,01/38] drm/i915: Use the MRU stack search after evicting (rev6)
URL   : https://patchwork.freedesktop.org/series/16934/
State : success

== Summary ==

Series 16934v6 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/16934/revisions/6/mbox/

Test drv_module_reload:
        Subgroup basic-reload-inject:
                dmesg-warn -> PASS       (fi-skl-6260u)
Test pm_rpm:
        Subgroup basic-pci-d3-state:
                fail       -> PASS       (fi-skl-6260u)
        Subgroup basic-rte:
                fail       -> PASS       (fi-skl-6260u)

fi-bdw-5557u     total:247  pass:233  dwarn:0   dfail:0   fail:0   skip:14 
fi-bsw-n3050     total:247  pass:208  dwarn:0   dfail:0   fail:0   skip:39 
fi-bxt-j4205     total:247  pass:225  dwarn:1   dfail:0   fail:0   skip:21 
fi-byt-j1900     total:247  pass:220  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-n2820     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-hsw-4770      total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-hsw-4770r     total:247  pass:228  dwarn:0   dfail:0   fail:0   skip:19 
fi-ilk-650       total:247  pass:195  dwarn:0   dfail:0   fail:0   skip:52 
fi-ivb-3520m     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-ivb-3770      total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-kbl-7500u     total:247  pass:226  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6260u     total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-skl-6700hq    total:247  pass:227  dwarn:0   dfail:0   fail:0   skip:20 
fi-skl-6700k     total:247  pass:224  dwarn:3   dfail:0   fail:0   skip:20 
fi-skl-6770hq    total:247  pass:234  dwarn:0   dfail:0   fail:0   skip:13 
fi-snb-2520m     total:247  pass:216  dwarn:0   dfail:0   fail:0   skip:31 
fi-snb-2600      total:247  pass:215  dwarn:0   dfail:0   fail:0   skip:32 

7285204135ffcf095bea64196db6ef900d47e048 drm-tip: 2016y-12m-20d-14h-50m-18s UTC integration manifest
f647118 drm: kselftest for drm_mm and bottom-up allocation
457bcc3 drm: Improve drm_mm search (and fix topdown allocation) with rbtrees
98992f9 drm: Use drm_mm_insert_node_in_range_generic() for everyone
37836f0 drm: Apply range restriction after color adjustment when allocation
f9da855 drm: Wrap drm_mm_node.hole_follows
328c23a drm: Apply tight eviction scanning to color_adjust
9ea0bd0 drm: Simplify drm_mm scan-list manipulation
017761c drm: Optimise power-of-two alignments in drm_mm_scan_add_block()
6d874e1 drm: Compute tight evictions for drm_mm_scan
1d16722 drm: Fix application of color vs range restriction when scanning drm_mm
ff0ea90 drm: Unconditionally do the range check in drm_mm_scan_add_block()
abe5592 drm: Rename prev_node to hole in drm_mm_scan_add_block()
fe3ad32 drm: Extract struct drm_mm_scan from struct drm_mm
7c2a427 drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()
5f413fd drm: Simplify drm_mm_clean()
a6125ad drm: Detect overflow in drm_mm_reserve_node()
7121791 drm: Fix kerneldoc for drm_mm_scan_remove_block()
3071d25 drm: Promote drm_mm alignment to u64
373c154 drm/i915: Build DRM range manager selftests for CI
762f6bf drm: kselftest for drm_mm and restricted color eviction
52a9e87 drm: kselftest for drm_mm and color eviction
5398b2b drm: kselftest for drm_mm and color adjustment
19f9d04 drm: kselftest for drm_mm and top-down allocation
b61ae82 drm: kselftest for drm_mm and range restricted eviction
fc01f52 drm: kselftest for drm_mm and eviction
1bc5dc1 drm: kselftest for drm_mm and alignment
7d7daa4 drm: kselftest for drm_mm_insert_node_in_range()
1fef227 drm: kselftest for drm_mm_replace_node()
962f7eb drm: kselftest for drm_mm_insert_node()
ca0de06 drm: kselftest for drm_mm_reserve_node()
4bf0b3a drm: kselftest for drm_mm_debug()
f9a6476f drm: kselftest for drm_mm_init()
7a934bc drm: Add some kselftests for the DRM range manager (struct drm_mm)
2563cec drm: Add a simple generator of random permutations
f5d7bf1 lib: Add a simple prime number generator
50c78e1 drm: Compile time enabling for asserts in drm_mm
cc78309 drm: Use drm_mm_nodes() as shorthand for the list of nodes under struct drm_mm
87a1493 drm/i915: Use the MRU stack search after evicting

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3344/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

drm_mm crash with multi threads

[Mark yao]

[-- Attachment #1: Type: text/html, Size: 4127 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [Intel-gfx] drm_mm crash with multi threads

[Rob Clark]

On Thu, Dec 22, 2016 at 11:07 PM, Mark yao <mark.yao@rock-chips.com> wrote:
> Hi Chris Wilson
>
> We port drm_mm to my internal kernel, with high load test, found following
> crash:
>
> [49451.856244]
> ==================================================================
> [49451.856350] BUG: KASAN: wild-memory-access on address dead000000000108
> [49451.856379] Write of size 8 by task Binder:218_4/683 [49451.856417] CPU:
> 2 PID: 683 Comm: Binder:218_4 Not tainted 4.4.36 #62 [49451.856443] Hardware
> name: Rockchip RK3399 Excavator Board edp (Android) (DT) [49451.856469] Call
> trace: [49451.856519] [<ffffff900808a9d0>] dump_backtrace+0x0/0x230
> [49451.856556] [<ffffff900808ac14>] show_stack+0x14/0x1c [49451.856592]
> [<ffffff90084a4de0>] dump_stack+0xa0/0xc8 [49451.856633]
> [<ffffff900821b700>] kasan_report+0x110/0x4dc [49451.856670]
> [<ffffff900821aa84>] __asan_store8+0x24/0x7c [49451.856715]
> [<ffffff90086158c4>] drm_mm_insert_node_generic+0x2dc/0x464 [49451.856760]
> [<ffffff90086406a8>] rockchip_gem_iommu_map+0x60/0x158 [49451.856794]
> [<ffffff9008640bb4>] rockchip_gem_create_object+0x278/0x488 [49451.856827]
> [<ffffff9008641020>] rockchip_gem_create_with_handle+0x24/0x10c
> [49451.856862] [<ffffff9008641364>] rockchip_gem_create_ioctl+0x3c/0x50
> [49451.856896] [<ffffff900860aee4>] drm_ioctl+0x354/0x52c [49451.856939]
> [<ffffff900823d948>] do_vfs_ioctl+0x670/0x78c [49451.856976]
> [<ffffff900823dac4>] SyS_ioctl+0x60/0x88 [49451.857009] [<ffffff9008082ef0>]
> el0_svc_naked+0x24/0x28
> We only use drm_mm_insert_node_generic to alloc memory, and use
> drm_mm_remove_node to release memory
>
> alloc/release maybe on difference threads.
>
> Seem the problem is threads problem, drm_mm seems is not threads safe, we
> found drm_mm_insert_node_generic and drm_mm_remove_node
> may access same resource with list ops, such as some mm->hole_stack.
>
> After use mutex lock protect drm_mm_remove_node and
> drm_mm_insert_node_generic, the crash disappear.
>
> I'm not familiar with drm mm, Do you know how to fix it?

I don't think drm_mm is intended to be thread safe, but instead the
driver should provide whatever locking is needed before calling in to
drm_mm.  (And presumably you already need some lock to protect driver
level data structures when creating/destroying gem objects.)

BR,
-R

> Thanks.
>
> On 2016年12月17日 03:25, Chris Wilson wrote:
>
> With a lot of polish applied, Joonas has reviewed the series - all but
> for [04/38] "lib: Add a simple prime number generator"
> [lib/prime_numbers.c]. Anyone feel like poking around at a bit of number
> theory?
>
> Other than it would appear to be ready for Daniel to sort out the merge
> between drm-misc/i915... Please do take a look and see if you can spot
> anything else that needs fixing/improving.
> -Chris
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
>
>
> --
> Ｍark Yao
>
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [Intel-gfx] drm_mm crash with multi threads

[Mark yao]

On 2016年12月23日 13:57, Rob Clark wrote:
> On Thu, Dec 22, 2016 at 11:07 PM, Mark yao <mark.yao@rock-chips.com> wrote:
>> Hi Chris Wilson
>>
>> We port drm_mm to my internal kernel, with high load test, found following
>> crash:
>>
>> [49451.856244]
>> ==================================================================
>> [49451.856350] BUG: KASAN: wild-memory-access on address dead000000000108
>> [49451.856379] Write of size 8 by task Binder:218_4/683 [49451.856417] CPU:
>> 2 PID: 683 Comm: Binder:218_4 Not tainted 4.4.36 #62 [49451.856443] Hardware
>> name: Rockchip RK3399 Excavator Board edp (Android) (DT) [49451.856469] Call
>> trace: [49451.856519] [<ffffff900808a9d0>] dump_backtrace+0x0/0x230
>> [49451.856556] [<ffffff900808ac14>] show_stack+0x14/0x1c [49451.856592]
>> [<ffffff90084a4de0>] dump_stack+0xa0/0xc8 [49451.856633]
>> [<ffffff900821b700>] kasan_report+0x110/0x4dc [49451.856670]
>> [<ffffff900821aa84>] __asan_store8+0x24/0x7c [49451.856715]
>> [<ffffff90086158c4>] drm_mm_insert_node_generic+0x2dc/0x464 [49451.856760]
>> [<ffffff90086406a8>] rockchip_gem_iommu_map+0x60/0x158 [49451.856794]
>> [<ffffff9008640bb4>] rockchip_gem_create_object+0x278/0x488 [49451.856827]
>> [<ffffff9008641020>] rockchip_gem_create_with_handle+0x24/0x10c
>> [49451.856862] [<ffffff9008641364>] rockchip_gem_create_ioctl+0x3c/0x50
>> [49451.856896] [<ffffff900860aee4>] drm_ioctl+0x354/0x52c [49451.856939]
>> [<ffffff900823d948>] do_vfs_ioctl+0x670/0x78c [49451.856976]
>> [<ffffff900823dac4>] SyS_ioctl+0x60/0x88 [49451.857009] [<ffffff9008082ef0>]
>> el0_svc_naked+0x24/0x28
>> We only use drm_mm_insert_node_generic to alloc memory, and use
>> drm_mm_remove_node to release memory
>>
>> alloc/release maybe on difference threads.
>>
>> Seem the problem is threads problem, drm_mm seems is not threads safe, we
>> found drm_mm_insert_node_generic and drm_mm_remove_node
>> may access same resource with list ops, such as some mm->hole_stack.
>>
>> After use mutex lock protect drm_mm_remove_node and
>> drm_mm_insert_node_generic, the crash disappear.
>>
>> I'm not familiar with drm mm, Do you know how to fix it?
> I don't think drm_mm is intended to be thread safe, but instead the
> driver should provide whatever locking is needed before calling in to
> drm_mm.  (And presumably you already need some lock to protect driver
> level data structures when creating/destroying gem objects.)
>
> BR,
> -R

Got it,

I also see other's driver, all of them  have a lock to protect memory 
creating/destorying.

Thanks.

>
>> Thanks.
>>
>> On 2016年12月17日 03:25, Chris Wilson wrote:
>>
>> With a lot of polish applied, Joonas has reviewed the series - all but
>> for [04/38] "lib: Add a simple prime number generator"
>> [lib/prime_numbers.c]. Anyone feel like poking around at a bit of number
>> theory?
>>
>> Other than it would appear to be ready for Daniel to sort out the merge
>> between drm-misc/i915... Please do take a look and see if you can spot
>> anything else that needs fixing/improving.
>> -Chris
>>
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>>
>>
>>
>> --
>> Ｍark Yao
>>
>>
>> _______________________________________________
>> Intel-gfx mailing list
>> Intel-gfx@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
>>
>
>


-- 
Ｍark Yao


_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel