drm_mm range manager fixes, take 2

drm_mm range manager fixes, take 2

[Chris Wilson]

The goal of this series is to fix top-down allocations to be allocated
from the top and aligned correctly, introduce bottom-up allocations, and
speed up searches and tighten evictions.

More polish on the test cases to reduce code duplication and to improve
expectation checking. And a little more polish on the patches leading up
to the fixes.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v2 01/40] 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outdated) into a common routine for allocating a node and evicting as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> #v1
---
 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 ef00d36680c9..4543d7fa7fc2 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");
 
@@ -3567,3 +3551,49 @@ 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

[PATCH v2 02/40] drm/i915: Simplify i915_gtt_color_adjust()

[Chris Wilson]

If we remember that node_list is a circular list containing the fake
head_node, we can use a simple list_next_entry() and skip the NULL check
for the allocated check against the head_node.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 4543d7fa7fc2..4d82f38a000b 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2707,10 +2707,8 @@ static void i915_gtt_color_adjust(struct drm_mm_node *node,
 	if (node->color != color)
 		*start += 4096;
 
-	node = list_first_entry_or_null(&node->node_list,
-					struct drm_mm_node,
-					node_list);
-	if (node && node->allocated && node->color != color)
+	node = list_next_entry(node, node_list);
+	if (node->allocated && node->color != color)
 		*end -= 4096;
 }
 
-- 
2.11.0

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

[PATCH v2 03/40] drm: Add drm_mm_for_each_node_safe()

[Chris Wilson]

A complement to drm_mm_for_each_node(), wraps list_for_each_entry_safe()
for walking the list of nodes safe against removal.

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     | 19 ++++++++++++++++---
 2 files changed, 20 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index ca1e344f318d..6e0735539545 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, &mm->head_node.node_list, node_list) {
+	list_for_each_entry(node, __drm_mm_nodes(mm), node_list) {
 		struct stack_trace trace = {
 			.entries = entries,
 			.max_entries = STACKDEPTH
@@ -320,8 +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(&mm->head_node.node_list,
-				  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);
@@ -884,7 +883,7 @@ EXPORT_SYMBOL(drm_mm_scan_remove_block);
  */
 bool drm_mm_clean(struct drm_mm * mm)
 {
-	struct list_head *head = &mm->head_node.node_list;
+	struct list_head *head = __drm_mm_nodes(mm);
 
 	return (head->next->next == head);
 }
@@ -930,7 +929,7 @@ EXPORT_SYMBOL(drm_mm_init);
  */
 void drm_mm_takedown(struct drm_mm *mm)
 {
-	if (WARN(!list_empty(&mm->head_node.node_list),
+	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 0b8371795aeb..0cc1b78c9ec2 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -179,6 +179,8 @@ static inline u64 drm_mm_hole_node_end(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_for_each_node - iterator to walk over all allocated nodes
  * @entry: drm_mm_node structure to assign to in each iteration step
@@ -187,9 +189,20 @@ static inline u64 drm_mm_hole_node_end(struct drm_mm_node *hole_node)
  * This iterator walks over all nodes in the range allocator. It is implemented
  * with list_for_each, so not save against removal of elements.
  */
-#define drm_mm_for_each_node(entry, mm) list_for_each_entry(entry, \
-						&(mm)->head_node.node_list, \
-						node_list)
+#define drm_mm_for_each_node(entry, mm) \
+	list_for_each_entry(entry, __drm_mm_nodes(mm), node_list)
+
+/**
+ * drm_mm_for_each_node_safe - iterator to walk over all allocated nodes
+ * @entry: drm_mm_node structure to assign to in each iteration step
+ * @next: drm_mm_node structure to store the next step
+ * @mm: drm_mm allocator to walk
+ *
+ * This iterator walks over all nodes in the range allocator. It is implemented
+ * 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)
 
 #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

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

[PATCH v2 04/40] drm: Constify the drm_mm API

[Chris Wilson]

Mark up the pointers as constant through the API where appropriate.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/drm_mm.c            | 24 ++++++++++++------------
 drivers/gpu/drm/i915/i915_gem_gtt.c |  2 +-
 include/drm/drm_mm.h                | 27 +++++++++++++--------------
 3 files changed, 26 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 6e0735539545..7573661302a4 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -174,9 +174,9 @@ INTERVAL_TREE_DEFINE(struct drm_mm_node, rb,
 		     START, LAST, static inline, drm_mm_interval_tree)
 
 struct drm_mm_node *
-__drm_mm_interval_first(struct drm_mm *mm, u64 start, u64 last)
+__drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last)
 {
-	return drm_mm_interval_tree_iter_first(&mm->interval_tree,
+	return drm_mm_interval_tree_iter_first((struct rb_root *)&mm->interval_tree,
 					       start, last);
 }
 EXPORT_SYMBOL(__drm_mm_interval_first);
@@ -881,9 +881,9 @@ EXPORT_SYMBOL(drm_mm_scan_remove_block);
  * True if the allocator is completely free, false if there's still a node
  * allocated in it.
  */
-bool drm_mm_clean(struct drm_mm * mm)
+bool drm_mm_clean(const struct drm_mm *mm)
 {
-	struct list_head *head = __drm_mm_nodes(mm);
+	const struct list_head *head = __drm_mm_nodes(mm);
 
 	return (head->next->next == head);
 }
@@ -897,7 +897,7 @@ EXPORT_SYMBOL(drm_mm_clean);
  *
  * Note that @mm must be cleared to 0 before calling this function.
  */
-void drm_mm_init(struct drm_mm * mm, u64 start, u64 size)
+void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 {
 	INIT_LIST_HEAD(&mm->hole_stack);
 	mm->scanned_blocks = 0;
@@ -936,8 +936,8 @@ void drm_mm_takedown(struct drm_mm *mm)
 }
 EXPORT_SYMBOL(drm_mm_takedown);
 
-static u64 drm_mm_debug_hole(struct drm_mm_node *entry,
-				     const char *prefix)
+static u64 drm_mm_debug_hole(const struct drm_mm_node *entry,
+			     const char *prefix)
 {
 	u64 hole_start, hole_end, hole_size;
 
@@ -958,9 +958,9 @@ static u64 drm_mm_debug_hole(struct drm_mm_node *entry,
  * @mm: drm_mm allocator to dump
  * @prefix: prefix to use for dumping to dmesg
  */
-void drm_mm_debug_table(struct drm_mm *mm, const char *prefix)
+void drm_mm_debug_table(const struct drm_mm *mm, const char *prefix)
 {
-	struct drm_mm_node *entry;
+	const struct drm_mm_node *entry;
 	u64 total_used = 0, total_free = 0, total = 0;
 
 	total_free += drm_mm_debug_hole(&mm->head_node, prefix);
@@ -979,7 +979,7 @@ void drm_mm_debug_table(struct drm_mm *mm, const char *prefix)
 EXPORT_SYMBOL(drm_mm_debug_table);
 
 #if defined(CONFIG_DEBUG_FS)
-static u64 drm_mm_dump_hole(struct seq_file *m, struct drm_mm_node *entry)
+static u64 drm_mm_dump_hole(struct seq_file *m, const struct drm_mm_node *entry)
 {
 	u64 hole_start, hole_end, hole_size;
 
@@ -1000,9 +1000,9 @@ static u64 drm_mm_dump_hole(struct seq_file *m, struct drm_mm_node *entry)
  * @m: seq_file to dump to
  * @mm: drm_mm allocator to dump
  */
-int drm_mm_dump_table(struct seq_file *m, struct drm_mm *mm)
+int drm_mm_dump_table(struct seq_file *m, const struct drm_mm *mm)
 {
-	struct drm_mm_node *entry;
+	const struct drm_mm_node *entry;
 	u64 total_used = 0, total_free = 0, total = 0;
 
 	total_free += drm_mm_dump_hole(m, &mm->head_node);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 4d82f38a000b..97fd66e4e3d0 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -2699,7 +2699,7 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
 	dma_unmap_sg(kdev, pages->sgl, pages->nents, PCI_DMA_BIDIRECTIONAL);
 }
 
-static void i915_gtt_color_adjust(struct drm_mm_node *node,
+static void i915_gtt_color_adjust(const struct drm_mm_node *node,
 				  unsigned long color,
 				  u64 *start,
 				  u64 *end)
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 0cc1b78c9ec2..5c7f15875b6a 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -102,7 +102,8 @@ struct drm_mm {
 	u64 scan_end;
 	struct drm_mm_node *prev_scanned_node;
 
-	void (*color_adjust)(struct drm_mm_node *node, unsigned long color,
+	void (*color_adjust)(const struct drm_mm_node *node,
+			     unsigned long color,
 			     u64 *start, u64 *end);
 };
 
@@ -116,7 +117,7 @@ struct drm_mm {
  * Returns:
  * True if the @node is allocated.
  */
-static inline bool drm_mm_node_allocated(struct drm_mm_node *node)
+static inline bool drm_mm_node_allocated(const struct drm_mm_node *node)
 {
 	return node->allocated;
 }
@@ -131,12 +132,12 @@ static inline bool drm_mm_node_allocated(struct drm_mm_node *node)
  * Returns:
  * True if the @mm is initialized.
  */
-static inline bool drm_mm_initialized(struct drm_mm *mm)
+static inline bool drm_mm_initialized(const struct drm_mm *mm)
 {
 	return mm->hole_stack.next;
 }
 
-static inline u64 __drm_mm_hole_node_start(struct drm_mm_node *hole_node)
+static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
 {
 	return hole_node->start + hole_node->size;
 }
@@ -152,13 +153,13 @@ static inline u64 __drm_mm_hole_node_start(struct drm_mm_node *hole_node)
  * Returns:
  * Start of the subsequent hole.
  */
-static inline u64 drm_mm_hole_node_start(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);
 	return __drm_mm_hole_node_start(hole_node);
 }
 
-static inline u64 __drm_mm_hole_node_end(struct drm_mm_node *hole_node)
+static inline u64 __drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
 {
 	return list_next_entry(hole_node, node_list)->start;
 }
@@ -174,7 +175,7 @@ static inline u64 __drm_mm_hole_node_end(struct drm_mm_node *hole_node)
  * Returns:
  * End of the subsequent hole.
  */
-static inline u64 drm_mm_hole_node_end(struct drm_mm_node *hole_node)
+static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
 {
 	return __drm_mm_hole_node_end(hole_node);
 }
@@ -314,14 +315,12 @@ static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
 
 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_init(struct drm_mm *mm, u64 start, u64 size);
 void drm_mm_takedown(struct drm_mm *mm);
-bool drm_mm_clean(struct drm_mm *mm);
+bool drm_mm_clean(const struct drm_mm *mm);
 
 struct drm_mm_node *
-__drm_mm_interval_first(struct drm_mm *mm, u64 start, u64 last);
+__drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
 
 /**
  * drm_mm_for_each_node_in_range - iterator to walk over a range of
@@ -355,9 +354,9 @@ void drm_mm_init_scan_with_range(struct drm_mm *mm,
 bool drm_mm_scan_add_block(struct drm_mm_node *node);
 bool drm_mm_scan_remove_block(struct drm_mm_node *node);
 
-void drm_mm_debug_table(struct drm_mm *mm, const char *prefix);
+void drm_mm_debug_table(const struct drm_mm *mm, const char *prefix);
 #ifdef CONFIG_DEBUG_FS
-int drm_mm_dump_table(struct seq_file *m, struct drm_mm *mm);
+int drm_mm_dump_table(struct seq_file *m, const struct drm_mm *mm);
 #endif
 
 #endif
-- 
2.11.0

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

[PATCH v2 05/40] 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 7573661302a4..2b76167ef39f 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 5c7f15875b6a..0b0485aff9df 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 v2 06/40] 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                       |  13 +++
 drivers/gpu/drm/Makefile                      |   2 +
 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       |  55 +++++++++++++
 tools/testing/selftests/drivers/gpu/drm_mm.sh |  15 ++++
 7 files changed, 243 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 ebfe8404c25f..d1363d21d3d1 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -48,6 +48,19 @@ 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
+	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 b9ae4280de9d..c8aed3688b20 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -18,6 +18,8 @@ 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
 
+obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/test-drm_mm.o
+
 drm-$(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/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..844d4625931e
--- /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.
+ */
+
+#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 combine_0(x, y) x ## y
+#define combine(x, y) combine_0(x, y)
+#define param(n) combine(igt__, combine(combine(__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..4c061baccf28
--- /dev/null
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -0,0 +1,55 @@
+/*
+ * Test cases for the drm_mm range manager
+ */
+
+#define pr_fmt(fmt) "drm_mm: " fmt
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/random.h>
+#include <linux/vmalloc.h>
+
+#include <drm/drm_mm.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 v2 07/40] 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.

v2: Tidier code by David Herrmann

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>
---
 drivers/gpu/drm/Kconfig          |  6 ++++++
 drivers/gpu/drm/Makefile         |  1 +
 drivers/gpu/drm/lib/drm_random.c | 41 ++++++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/lib/drm_random.h | 21 ++++++++++++++++++++
 4 files changed, 69 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 d1363d21d3d1..2e6ae95459e4 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -52,6 +52,7 @@ config DRM_DEBUG_MM_SELFTEST
 	tristate "kselftests for DRM range manager (struct drm_mm)"
 	depends on DRM
 	depends on DEBUG_KERNEL
+	select DRM_LIB_RANDOM
 	default n
 	help
 	  This option provides a kernel module that can be used to test
@@ -334,3 +335,8 @@ 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 c8aed3688b20..0fa16275fdae 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-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o
 obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/test-drm_mm.o
 
 drm-$(CONFIG_COMPAT) += drm_ioc32.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..507e72721b6f
--- /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 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 = 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..931c8f4a261d
--- /dev/null
+++ b/drivers/gpu/drm/lib/drm_random.h
@@ -0,0 +1,21 @@
+#ifndef __DRM_RANDOM_H__
+#define __DRM_RANDOM_H
+
+#include <linux/random.h>
+
+#define RND_STATE_INITIALIZER(seed__) ({				\
+	struct rnd_state state__;					\
+	prandom_seed_state(&state__, (seed__));				\
+	state__;							\
+})
+
+#define RND_STATE(name__, seed__) \
+	struct rnd_state name__ = 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

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

[PATCH v2 08/40] drm: Add a simple prime number generator

[Chris Wilson]

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

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/Kconfig                 |   4 +
 drivers/gpu/drm/Makefile                |   1 +
 drivers/gpu/drm/lib/drm_prime_numbers.c | 175 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/lib/drm_prime_numbers.h |  10 ++
 4 files changed, 190 insertions(+)
 create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.c
 create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.h

diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index 2e6ae95459e4..93895898d596 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -53,6 +53,7 @@ config DRM_DEBUG_MM_SELFTEST
 	depends on DRM
 	depends on DEBUG_KERNEL
 	select DRM_LIB_RANDOM
+	select DRM_LIB_PRIMES
 	default n
 	help
 	  This option provides a kernel module that can be used to test
@@ -340,3 +341,6 @@ config DRM_LIB_RANDOM
 	bool
 	default n
 
+config DRM_LIB_PRIMES
+	bool
+	default n
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 0fa16275fdae..bbd390fa8914 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -19,6 +19,7 @@ drm-y       :=	drm_auth.o drm_bufs.o drm_cache.o \
 		drm_dumb_buffers.o drm_mode_config.o
 
 drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o
+obj-$(CONFIG_DRM_LIB_PRIMES) += lib/drm_prime_numbers.o
 obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/test-drm_mm.o
 
 drm-$(CONFIG_COMPAT) += drm_ioc32.o
diff --git a/drivers/gpu/drm/lib/drm_prime_numbers.c b/drivers/gpu/drm/lib/drm_prime_numbers.c
new file mode 100644
index 000000000000..839563d9b787
--- /dev/null
+++ b/drivers/gpu/drm/lib/drm_prime_numbers.c
@@ -0,0 +1,175 @@
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+
+#include "drm_prime_numbers.h"
+
+static DEFINE_MUTEX(lock);
+
+static struct primes {
+	struct rcu_head rcu;
+	unsigned long last, sz;
+	unsigned long primes[];
+} __rcu *primes;
+
+static bool slow_is_prime_number(unsigned long x)
+{
+	unsigned long y = int_sqrt(x) + 1;
+
+	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 = (start / x + 1) * x;
+
+	while (m < end) {
+		__clear_bit(m, p);
+		m += x;
+	}
+
+	return x;
+}
+
+static struct primes *expand(unsigned long x)
+{
+	unsigned long sz, y, prev;
+	struct primes *p, *new;
+
+	sz = x * x;
+	if (sz < x)
+		return NULL;
+
+	mutex_lock(&lock);
+	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
+	if (p && x < p->last)
+		goto unlock;
+
+	sz = round_up(sz, BITS_PER_LONG);
+	new = kmalloc(sizeof(*new) + sz / sizeof(long), GFP_KERNEL);
+	if (!new) {
+		p = NULL;
+		goto unlock;
+	}
+
+	/* Where memory permits, track the primes using the
+	 * Sieve of Eratosthenes.
+	 */
+	if (p) {
+		prev = p->sz;
+		memcpy(new->primes, p->primes, prev / BITS_PER_LONG);
+	} else {
+		prev = 0;
+	}
+	memset(new->primes + prev / BITS_PER_LONG,
+	       0xff, (sz - prev) / sizeof(long));
+	for (y = 2UL; y < sz; y = find_next_bit(new->primes, sz, y + 1))
+		new->last = mark_multiples(y, new->primes, prev, sz);
+	new->sz = sz;
+
+	rcu_assign_pointer(primes, new);
+	if (p)
+		kfree_rcu(p, rcu);
+	p = new;
+
+unlock:
+	mutex_unlock(&lock);
+	return p;
+}
+
+unsigned long drm_next_prime_number(unsigned long x)
+{
+	struct primes *p;
+
+	if (x < 2)
+		return 2;
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	if (!p || x >= p->last) {
+		rcu_read_unlock();
+
+		p = expand(x);
+		if (!p)
+			return slow_next_prime_number(x);
+
+		rcu_read_lock();
+	}
+
+	x = find_next_bit(p->primes, p->last, x + 1);
+	rcu_read_unlock();
+
+	return x;
+}
+EXPORT_SYMBOL(drm_next_prime_number);
+
+bool drm_is_prime_number(unsigned long x)
+{
+	struct primes *p;
+	bool result;
+
+	switch (x) {
+	case 0:
+		return false;
+	case 1:
+	case 2:
+	case 3:
+		return true;
+	}
+
+	rcu_read_lock();
+	p = rcu_dereference(primes);
+	if (!p || x >= p->last) {
+		rcu_read_unlock();
+
+		p = expand(x);
+		if (!p)
+			return slow_is_prime_number(x);
+
+		rcu_read_lock();
+	}
+
+	result = test_bit(x, p->primes);
+	rcu_read_unlock();
+
+	return result;
+}
+EXPORT_SYMBOL(drm_is_prime_number);
+
+static int __init drm_primes_init(void)
+{
+	return 0;
+}
+
+static void __exit drm_primes_exit(void)
+{
+	if (primes)
+		kfree_rcu(primes, rcu);
+}
+
+module_init(drm_primes_init);
+module_exit(drm_primes_exit);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
diff --git a/drivers/gpu/drm/lib/drm_prime_numbers.h b/drivers/gpu/drm/lib/drm_prime_numbers.h
new file mode 100644
index 000000000000..7bc58cf9a86c
--- /dev/null
+++ b/drivers/gpu/drm/lib/drm_prime_numbers.h
@@ -0,0 +1,10 @@
+#ifndef __DRM_PRIMES_H__
+#define __DRM_PRIMES_H__
+
+bool drm_is_prime_number(unsigned long x);
+unsigned long drm_next_prime_number(unsigned long x);
+
+#define drm_for_each_prime(prime, max) \
+	for (prime = 1;	prime < (max); prime = drm_next_prime_number(prime))
+
+#endif /* __DRM_PRIMES_H__ */
-- 
2.11.0

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

[PATCH v2 09/40] 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>
---
 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 4c061baccf28..ccef8e249d37 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -24,6 +24,120 @@ 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

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

[PATCH v2 10/40] 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>
---
 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 ccef8e249d37..262e44f8f9fb 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -138,6 +138,41 @@ 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

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

[PATCH v2 11/40] 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      | 279 +++++++++++++++++++++++++++
 2 files changed, 280 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 262e44f8f9fb..2e40f94cb9d3 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -11,6 +11,9 @@
 
 #include <drm/drm_mm.h>
 
+#include "../lib/drm_random.h"
+#include "../lib/drm_prime_numbers.h"
+
 #define TESTS "drm_mm_selftests.h"
 #include "drm_selftest.h"
 
@@ -77,6 +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;
@@ -173,6 +227,231 @@ 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 (err != -ENOSPC) {
+		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;
+	}
+
+	return true;
+}
+
+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)
+{
+	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 */
+	drm_for_each_prime(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;
+
+	drm_for_each_prime(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

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

[PATCH v2 12/40] 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>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 220 +++++++++++++++++++++++++++
 2 files changed, 221 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 2e40f94cb9d3..9621820a1cda 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -131,6 +131,48 @@ static bool assert_continuous(const struct drm_mm *mm, u64 size)
 	return true;
 }
 
+static u64 misaligned(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 (misaligned(node, alignment)) {
+		pr_err("node is misalinged, start %llx rem %llu, expected alignment %llu\n",
+		       node->start, misaligned(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;
@@ -452,6 +494,184 @@ static int igt_reserve(void *ignored)
 	return 0;
 }
 
+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 (err != -ENOSPC)  {
+		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;
+	}
+
+	return true;
+}
+
+static const struct insert_mode {
+	const char *name;
+	unsigned int search_flags;
+	unsigned int create_flags;
+} insert_modes[] = {
+	{ "default", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
+	{ "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
+	{ "best", DRM_MM_SEARCH_BEST, DRM_MM_CREATE_DEFAULT },
+	{}
+};
+
+static int __igt_insert(unsigned int count, u64 size)
+{
+	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, err;
+
+	/* 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++) {
+			node = &nodes[n];
+			err = drm_mm_insert_node_generic(&mm, node, size, 0, n,
+							 mode->search_flags,
+							 mode->create_flags);
+			if (err || !assert_node(node, &mm, size, 0, n)) {
+				pr_err("%s insert failed, size %llu step %d\n",
+				       mode->name, size, n);
+				ret = err ?: -EINVAL;
+				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]);
+			err = drm_mm_insert_node_generic(&mm,
+							 &nodes[n], size, 0, n,
+							 mode->search_flags,
+							 mode->create_flags);
+			if (err || !assert_node(&nodes[n], &mm, size, 0, n)) {
+				pr_err("%s reinsert failed, step %d\n",
+				       mode->name, n);
+				ret = err ?: -EINVAL;
+				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 */
+		drm_for_each_prime(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_insert_node_generic(&mm,
+								 node, size, 0, n,
+								 mode->search_flags,
+								 mode->create_flags);
+				if (err || !assert_node(node, &mm, size, 0, n)) {
+					pr_err("insert failed, step %d, start %llu\n",
+					       n, node->start);
+					ret = err ?: -EINVAL;
+					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;
+
+	drm_for_each_prime(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 v2 13/40] 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      | 64 +++++++++++++++++++++++++---
 2 files changed, 59 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 9621820a1cda..dca146478b06 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -526,7 +526,7 @@ static const struct insert_mode {
 	{}
 };
 
-static int __igt_insert(unsigned int count, u64 size)
+static int __igt_insert(unsigned int count, u64 size, bool replace)
 {
 	RND_STATE(prng, random_seed);
 	const struct insert_mode *mode;
@@ -541,7 +541,7 @@ 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;
 
@@ -554,7 +554,10 @@ static int __igt_insert(unsigned int count, u64 size)
 
 	for (mode = insert_modes; mode->name; mode++) {
 		for (n = 0; n < count; n++) {
-			node = &nodes[n];
+			struct drm_mm_node tmp;
+
+			node = replace ? &tmp : &nodes[n];
+			memset(node, 0, sizeof(*node));
 			err = drm_mm_insert_node_generic(&mm, node, size, 0, n,
 							 mode->search_flags,
 							 mode->create_flags);
@@ -564,6 +567,28 @@ static int __igt_insert(unsigned int count, u64 size)
 				ret = err ?: -EINVAL;
 				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 */
@@ -656,17 +681,44 @@ static int igt_insert(void *ignored)
 	drm_for_each_prime(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, 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.
+	 */
+
+	drm_for_each_prime(n, 54) {
+		u64 size = BIT_ULL(n);
+
+		ret = __igt_insert(count, size - 1, true);
 		if (ret)
 			return ret;
 
-		ret = __igt_insert(count, size + 1);
+		ret = __igt_insert(count, size, true);
 		if (ret)
 			return ret;
+
+		ret = __igt_insert(count, size + 1, true);
 	}
 
 	return 0;
-- 
2.11.0

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

[PATCH v2 14/40] 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>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 246 +++++++++++++++++++++++++++
 2 files changed, 247 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 dca146478b06..94cd0741c5b6 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -724,6 +724,252 @@ static int igt_replace(void *ignored)
 	return 0;
 }
 
+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 (err != -ENOSPC)  {
+		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;
+	}
+
+	return true;
+}
+
+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, err;
+
+	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++) {
+			err = drm_mm_insert_node_in_range_generic(&mm, &nodes[n],
+								  size, size, n,
+								  start, end,
+								  mode->search_flags,
+								  mode->create_flags);
+			if (err || !assert_node(&nodes[n], &mm, size, size, n)) {
+				pr_err("%s insert failed, size %llu, step %d [%d, %d], range [%llx, %llx], err=%d\n",
+				       mode->name, size, n,
+				       start_n, end_n,
+				       start, end,
+				       err);
+				ret = err ?: -EINVAL;
+				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]);
+			err = drm_mm_insert_node_in_range_generic(&mm, &nodes[n],
+								  size, size, n,
+								  start, end,
+								  mode->search_flags,
+								  mode->create_flags);
+			if (err || !assert_node(&nodes[n], &mm, size, size, n)) {
+				pr_err("%s reinsert failed, step %d\n", mode->name, n);
+				ret = err ?: -EINVAL;
+				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;
+
+	drm_for_each_prime(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 v2 15/40] 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      | 110 +++++++++++++++++++++++++++
 2 files changed, 113 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 94cd0741c5b6..dc3aee222158 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -970,6 +970,116 @@ 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, err;
+
+	/* 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;
+
+		drm_for_each_prime(prime, max_count) {
+			u64 size = drm_next_prime_number(prime);
+
+			err = drm_mm_insert_node_generic(&mm,
+							 &nodes[i],
+							 size, prime, i,
+							 mode->search_flags,
+							 mode->create_flags);
+			if (err || !assert_node(&nodes[i], &mm, size, prime, i)) {
+				pr_err("%s insert failed with alignment=%d",
+				       mode->name, prime);
+				ret = err ?: -EINVAL;
+				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;
+		int err;
+
+		node = kzalloc(sizeof(*node), GFP_KERNEL);
+		if (!node) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		align = BIT_ULL(bit);
+		size = BIT_ULL(bit-1) + 1;
+		err = drm_mm_insert_node_generic(&mm, node, size, align, bit,
+						 DRM_MM_SEARCH_DEFAULT,
+						 DRM_MM_CREATE_DEFAULT);
+		if (err || !assert_node(node, &mm, size, align, bit)) {
+			pr_err("insert failed with alignment=%llx [%d]",
+			       align, bit);
+			ret = err ?: -EINVAL;
+			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

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

[PATCH v2 16/40] 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 dc3aee222158..4881752d6424 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -524,6 +524,10 @@ static const struct insert_mode {
 	{ "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
 	{ "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_insert(unsigned int count, u64 size, bool replace)
@@ -1080,6 +1084,339 @@ 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, misaligned(&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)
+{
+	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;
+			}
+		}
+
+		drm_for_each_prime(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

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

[PATCH v2 17/40] 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 4881752d6424..6c4679304358 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1249,6 +1249,7 @@ 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,
@@ -1261,7 +1262,9 @@ 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))
@@ -1279,6 +1282,12 @@ 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,
@@ -1360,7 +1369,7 @@ 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);
@@ -1374,7 +1383,7 @@ 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);
@@ -1392,7 +1401,7 @@ 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);
@@ -1417,6 +1426,102 @@ static int igt_evict(void *ignored)
 	return ret;
 }
 
+static int igt_evict_range(void *ignored)
+{
+	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;
+			}
+		}
+
+		drm_for_each_prime(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/2, 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 v2 18/40] 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      | 123 +++++++++++++++++++++++++++
 2 files changed, 124 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 6c4679304358..fc76d787cfd4 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1522,6 +1522,129 @@ 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)
+{
+	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, err;
+
+	/* 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++) {
+			err = drm_mm_insert_node_generic(&mm,
+							 &nodes[n], size, 0, n,
+							 DRM_MM_SEARCH_BELOW,
+							 DRM_MM_CREATE_TOP);
+			if (err || !assert_node(&nodes[n], &mm, size, 0, n)) {
+				pr_err("insert failed, size %u step %d\n", size, n);
+				ret = err ?: -EINVAL;
+				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);
+		drm_for_each_prime(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]];
+				err = drm_mm_insert_node_generic(&mm, node, size, 0, 0,
+								 DRM_MM_SEARCH_BELOW,
+								 DRM_MM_CREATE_TOP);
+				if (err || !assert_node(node, &mm, size, 0, 0)) {
+					pr_err("insert failed, step %d/%d\n", m, n);
+					ret = err ?: -EINVAL;
+					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 v2 19/40] 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      | 185 +++++++++++++++++++++++++++
 2 files changed, 186 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 fc76d787cfd4..6b3af04eb582 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1645,6 +1645,191 @@ 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;
+		}
+
+		err = drm_mm_insert_node_generic(&mm, node, n, 0, n,
+						 DRM_MM_SEARCH_DEFAULT,
+						 DRM_MM_CREATE_DEFAULT);
+		if (err || !assert_node(node, &mm, n, 0, n)) {
+			pr_err("insert failed, step %d\n", n);
+			kfree(node);
+			ret = err ?: -EINVAL;
+			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");
+			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 = misaligned(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);
+				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;
+			}
+
+			err = drm_mm_insert_node_generic(&mm, node, n, n, n,
+							 mode->search_flags,
+							 mode->create_flags);
+			if (err || !assert_node(node, &mm, n, n, n)) {
+				pr_err("%s insert failed, step %d, err=%d\n",
+				       mode->name, n, err);
+				kfree(node);
+				ret = err ?: -EINVAL;
+				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

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

[PATCH v2 20/40] 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>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 159 +++++++++++++++++++++++++++
 2 files changed, 160 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 6b3af04eb582..0a051abc8f13 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1830,6 +1830,165 @@ 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, misaligned(&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)
+{
+	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++) {
+		err = drm_mm_insert_node_generic(&mm, &nodes[n].node,
+						 1, 0, color++,
+						 DRM_MM_SEARCH_DEFAULT,
+						 DRM_MM_CREATE_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 <= 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;
+			}
+		}
+
+		drm_for_each_prime(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 v2 21/40] 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>
---
 drivers/gpu/drm/selftests/drm_mm_selftests.h |   1 +
 drivers/gpu/drm/selftests/test-drm_mm.c      | 119 ++++++++++++++++++++++++++-
 2 files changed, 116 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 0a051abc8f13..43575074eca5 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1831,6 +1831,7 @@ 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,
@@ -1844,7 +1845,9 @@ 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))
@@ -1862,6 +1865,12 @@ 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;
 
@@ -1933,7 +1942,7 @@ 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);
@@ -1946,7 +1955,7 @@ 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);
@@ -1963,7 +1972,7 @@ 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);
@@ -1989,6 +1998,108 @@ static int igt_color_evict(void *ignored)
 	return ret;
 }
 
+static int igt_color_evict_range(void *ignored)
+{
+	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++) {
+		err = drm_mm_insert_node_generic(&mm, &nodes[n].node,
+						 1, 0, color++,
+						 DRM_MM_SEARCH_DEFAULT,
+						 DRM_MM_CREATE_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, 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;
+			}
+		}
+
+		drm_for_each_prime(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

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

[PATCH v2 22/40] 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 v2 23/40] 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 2b76167ef39f..2d02ab0925a9 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 43575074eca5..f70d92e77ff6 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1086,7 +1086,7 @@ 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);
 }
@@ -1148,7 +1148,7 @@ 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 0b0485aff9df..803f8bd79e62 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;
@@ -249,7 +249,7 @@ 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);
@@ -272,7 +272,7 @@ 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,
@@ -282,7 +282,7 @@ 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,
@@ -309,7 +309,7 @@ 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)
@@ -349,11 +349,11 @@ __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

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

[PATCH v2 24/40] 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>
---
 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 2d02ab0925a9..ead164093ac7 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 803f8bd79e62..47ca6502ef29 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 v2 25/40] 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>
---
 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 ead164093ac7..6450690f5578 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

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

[PATCH v2 26/40] 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 6450690f5578..14a5ef505f1b 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 47ca6502ef29..2f28a806d015 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -330,7 +330,19 @@ 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 v2 27/40] 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>
---
 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 14a5ef505f1b..57267845b7d4 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 == 0);
+
 	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 == 0 || 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

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

[PATCH v2 28/40] 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 57267845b7d4..0f1396b0d39a 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 == 0);
+	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 == 0 || 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 f70d92e77ff6..c8cbfb88b803 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1084,11 +1084,11 @@ 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)
@@ -1128,7 +1128,7 @@ 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,
@@ -1140,18 +1140,16 @@ 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;
 	}
 
@@ -1165,19 +1163,20 @@ 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];
@@ -1211,19 +1210,21 @@ 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);
@@ -1257,15 +1258,16 @@ 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;
@@ -1277,7 +1279,7 @@ 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;
 	}
@@ -1840,15 +1842,16 @@ 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;
@@ -1860,7 +1863,7 @@ 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 2f28a806d015..4db57bef98e0 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;
 };
 
 /**
@@ -366,18 +375,22 @@ __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

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

[PATCH v2 29/40] 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>
---
 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 0f1396b0d39a..6bb61f2212f8 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 v2 30/40] 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 6bb61f2212f8..04b1d36c4ebc 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 == 0);
-	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 4db57bef98e0..8539832fac0f 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;
 };
 
 /**
@@ -375,11 +374,6 @@ __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,
@@ -387,6 +381,33 @@ 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 v2 31/40] 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 04b1d36c4ebc..956782e7b092 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 v2 32/40] 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 956782e7b092..ff1e62c066e8 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 == 0 || 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 c8cbfb88b803..a9fed0637297 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1169,7 +1169,7 @@ 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);
@@ -1216,7 +1216,7 @@ 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);
@@ -1266,7 +1266,8 @@ 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))
@@ -1850,7 +1851,8 @@ 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 8539832fac0f..639d00eb8f28 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;
 };
 
 /**
@@ -376,11 +377,9 @@ __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
@@ -389,10 +388,10 @@ 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
@@ -402,10 +401,13 @@ 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

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

[PATCH v2 33/40] 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 ff1e62c066e8..ffa439b2bd2c 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 639d00eb8f28..6ee87e1455bf 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

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

[PATCH v2 34/40] 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>
---
 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 ffa439b2bd2c..91c89ae09b26 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 6ee87e1455bf..a1532eb0ffbc 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

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

[PATCH v2 35/40] 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>
---
 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 91c89ae09b26..22db356e3ebc 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 a9fed0637297..9899e8364350 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1132,6 +1132,7 @@ 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;
@@ -1156,6 +1157,21 @@ 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;
 }
 
@@ -1269,7 +1285,7 @@ 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;
 
@@ -1854,7 +1870,7 @@ 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 a1532eb0ffbc..6e5911fa9018 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -410,6 +410,7 @@ 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 v2 36/40] 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>
---
 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 22db356e3ebc..da9f98690e97 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 9899e8364350..a9ed018aac12 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -42,7 +42,7 @@ 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;
 		}
@@ -104,7 +104,7 @@ 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;
 		}
@@ -787,7 +787,8 @@ 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;
 		}
@@ -1307,11 +1308,12 @@ 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, misaligned(&tmp, alignment),
-		       tmp.start, tmp.hole_follows);
+		       tmp.start, drm_mm_hole_follows(&tmp));
 		err = -EINVAL;
 	}
 
@@ -1590,7 +1592,7 @@ 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;
@@ -1624,7 +1626,7 @@ 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;
@@ -1679,7 +1681,7 @@ 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 6e5911fa9018..e78c01cf1135 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 v2 37/40] 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>
---
 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 da9f98690e97..062a2a82efd6 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

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

[PATCH v2 38/40] 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>
---
 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 062a2a82efd6..c6fce14178cc 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 e78c01cf1135..782a51d7bbab 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -273,40 +273,6 @@ 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,
@@ -316,6 +282,7 @@ 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
@@ -347,6 +314,61 @@ 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 v2 39/40] 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>
---
 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                     | 507 +++++++++++++++------------
 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      |  97 ++---
 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, 442 insertions(+), 476 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 c6fce14178cc..7eb531092793 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 +305,124 @@ 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);
 
-	if (node->start == hole_start) {
-		hole->hole_follows = 0;
-		list_del(&hole->hole_stack);
+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;
+}
+
+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;
 	}
+	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->hole_follows = 0;
-	if (end != hole_end) {
-		list_add(&node->hole_stack, &mm->hole_stack);
-		node->hole_follows = 1;
+		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 +430,104 @@ 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 +545,20 @@ 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;
-	}
+	node->allocated = false;
 
-	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;
-		}
-	}
-
-	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 +573,23 @@ 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 +632,7 @@ 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 +648,7 @@ 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 == 0 || size > end - start);
@@ -601,7 +663,7 @@ 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 +721,7 @@ 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 +733,7 @@ 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 +827,7 @@ 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 +854,22 @@ 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->head_node.size = -size;
+	add_hole(&mm->head_node);
 
-	mm->interval_tree = RB_ROOT;
-
-	mm->color_adjust = NULL;
+	mm->scan_active = 0;
 }
 EXPORT_SYMBOL(drm_mm_init);
 
@@ -828,18 +891,16 @@ 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 +931,16 @@ 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 a9ed018aac12..97468b8b33a0 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -499,7 +499,7 @@ 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 (err != -ENOSPC)  {
 		if (!err) {
 			pr_err("impossible insert succeeded, node %llu + %llu\n",
@@ -517,16 +517,16 @@ static bool expect_insert_fail(struct drm_mm *mm, u64 size)
 
 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", DRM_MM_SEARCH_DEFAULT, DRM_MM_CREATE_DEFAULT },
-	{ "top-down", DRM_MM_SEARCH_BELOW, DRM_MM_CREATE_TOP },
-	{ "best", DRM_MM_SEARCH_BEST, DRM_MM_CREATE_DEFAULT },
+	{ "best", DRM_MM_INSERT_BEST },
+	{ "bottom-up", DRM_MM_INSERT_LOW },
+	{ "top-down", DRM_MM_INSERT_HIGH },
+	{ "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 },
 	{}
 };
 
@@ -563,8 +563,7 @@ static int __igt_insert(unsigned int count, u64 size, bool replace)
 			node = replace ? &tmp : &nodes[n];
 			memset(node, 0, sizeof(*node));
 			err = drm_mm_insert_node_generic(&mm, node, size, 0, n,
-							 mode->search_flags,
-							 mode->create_flags);
+							 mode->mode);
 			if (err || !assert_node(node, &mm, size, 0, n)) {
 				pr_err("%s insert failed, size %llu step %d\n",
 				       mode->name, size, n);
@@ -610,8 +609,7 @@ static int __igt_insert(unsigned int count, u64 size, bool replace)
 			drm_mm_remove_node(&nodes[n]);
 			err = drm_mm_insert_node_generic(&mm,
 							 &nodes[n], size, 0, n,
-							 mode->search_flags,
-							 mode->create_flags);
+							 mode->mode);
 			if (err || !assert_node(&nodes[n], &mm, size, 0, n)) {
 				pr_err("%s reinsert failed, step %d\n",
 				       mode->name, n);
@@ -640,8 +638,7 @@ static int __igt_insert(unsigned int count, u64 size, bool replace)
 				node = &nodes[order[(o + m) % count]];
 				err = drm_mm_insert_node_generic(&mm,
 								 node, size, 0, n,
-								 mode->search_flags,
-								 mode->create_flags);
+								 mode->mode);
 				if (err || !assert_node(node, &mm, size, 0, n)) {
 					pr_err("insert failed, step %d, start %llu\n",
 					       n, node->start);
@@ -736,11 +733,10 @@ 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 (err != -ENOSPC)  {
 		if (!err) {
 			pr_err("impossible insert succeeded, node %llx + %llu, range [%llx, %llx]\n",
@@ -844,11 +840,10 @@ static int __igt_insert_range(unsigned int count, u64 size, u64 start, u64 end)
 
 	for (mode = insert_modes; mode->name; mode++) {
 		for (n = start_n; n <= end_n; n++) {
-			err = drm_mm_insert_node_in_range_generic(&mm, &nodes[n],
-								  size, size, n,
-								  start, end,
-								  mode->search_flags,
-								  mode->create_flags);
+			err = drm_mm_insert_node_in_range(&mm, &nodes[n],
+							  size, size, n,
+							  start, end,
+							  mode->mode);
 			if (err || !assert_node(&nodes[n], &mm, size, size, n)) {
 				pr_err("%s insert failed, size %llu, step %d [%d, %d], range [%llx, %llx], err=%d\n",
 				       mode->name, size, n,
@@ -871,11 +866,10 @@ static int __igt_insert_range(unsigned int count, u64 size, u64 start, u64 end)
 			u64 addr = nodes[n].start;
 
 			drm_mm_remove_node(&nodes[n]);
-			err = drm_mm_insert_node_in_range_generic(&mm, &nodes[n],
-								  size, size, n,
-								  start, end,
-								  mode->search_flags,
-								  mode->create_flags);
+			err = drm_mm_insert_node_in_range(&mm, &nodes[n],
+							  size, size, n,
+							  start, end,
+							  mode->mode);
 			if (err || !assert_node(&nodes[n], &mm, size, size, n)) {
 				pr_err("%s reinsert failed, step %d\n", mode->name, n);
 				ret = err ?: -EINVAL;
@@ -1004,8 +998,7 @@ static int igt_align(void *ignored)
 			err = drm_mm_insert_node_generic(&mm,
 							 &nodes[i],
 							 size, prime, i,
-							 mode->search_flags,
-							 mode->create_flags);
+							 mode->mode);
 			if (err || !assert_node(&nodes[i], &mm, size, prime, i)) {
 				pr_err("%s insert failed with alignment=%d",
 				       mode->name, prime);
@@ -1054,9 +1047,8 @@ static int igt_align_pot(int max)
 
 		align = BIT_ULL(bit);
 		size = BIT_ULL(bit-1) + 1;
-		err = drm_mm_insert_node_generic(&mm, node, size, align, bit,
-						 DRM_MM_SEARCH_DEFAULT,
-						 DRM_MM_CREATE_DEFAULT);
+		err = drm_mm_insert_node_generic(&mm, node,
+						 size, align, bit, 0);
 		if (err || !assert_node(node, &mm, size, align, bit)) {
 			pr_err("insert failed with alignment=%llx [%d]",
 			       align, bit);
@@ -1284,7 +1276,7 @@ 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))
@@ -1292,8 +1284,7 @@ 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);
@@ -1368,8 +1359,7 @@ 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;
@@ -1477,8 +1467,7 @@ 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;
@@ -1584,8 +1573,7 @@ static int igt_topdown(void *ignored)
 		for (n = 0; n < count; n++) {
 			err = drm_mm_insert_node_generic(&mm,
 							 &nodes[n], size, 0, n,
-							 DRM_MM_SEARCH_BELOW,
-							 DRM_MM_CREATE_TOP);
+							 DRM_MM_INSERT_HIGH);
 			if (err || !assert_node(&nodes[n], &mm, size, 0, n)) {
 				pr_err("insert failed, size %u step %d\n", size, n);
 				ret = err ?: -EINVAL;
@@ -1618,8 +1606,7 @@ static int igt_topdown(void *ignored)
 
 				node = &nodes[order[(o + m) % count]];
 				err = drm_mm_insert_node_generic(&mm, node, size, 0, 0,
-								 DRM_MM_SEARCH_BELOW,
-								 DRM_MM_CREATE_TOP);
+								 DRM_MM_INSERT_HIGH);
 				if (err || !assert_node(node, &mm, size, 0, 0)) {
 					pr_err("insert failed, step %d/%d\n", m, n);
 					ret = err ?: -EINVAL;
@@ -1719,9 +1706,7 @@ static int igt_color(void *ignored)
 			goto out;
 		}
 
-		err = drm_mm_insert_node_generic(&mm, node, n, 0, n,
-						 DRM_MM_SEARCH_DEFAULT,
-						 DRM_MM_CREATE_DEFAULT);
+		err = drm_mm_insert_node_generic(&mm, node, n, 0, n, 0);
 		if (err || !assert_node(node, &mm, n, 0, n)) {
 			pr_err("insert failed, step %d\n", n);
 			kfree(node);
@@ -1805,8 +1790,7 @@ static int igt_color(void *ignored)
 			}
 
 			err = drm_mm_insert_node_generic(&mm, node, n, n, n,
-							 mode->search_flags,
-							 mode->create_flags);
+							 mode->mode);
 			if (err || !assert_node(node, &mm, n, n, n)) {
 				pr_err("%s insert failed, step %d, err=%d\n",
 				       mode->name, n, err);
@@ -1870,7 +1854,7 @@ 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))
@@ -1878,8 +1862,7 @@ 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);
@@ -1952,9 +1935,7 @@ static int igt_color_evict(void *ignored)
 	mm.color_adjust = separate_adjacent_colors;
 	for (n = 0; n < total_size; n++) {
 		err = drm_mm_insert_node_generic(&mm, &nodes[n].node,
-						 1, 0, color++,
-						 DRM_MM_SEARCH_DEFAULT,
-						 DRM_MM_CREATE_DEFAULT);
+						 1, 0, color++, 0);
 		if (err) {
 			pr_err("insert failed, step %d\n", n);
 			ret = err;
@@ -2054,9 +2035,7 @@ static int igt_color_evict_range(void *ignored)
 	mm.color_adjust = separate_adjacent_colors;
 	for (n = 0; n < total_size; n++) {
 		err = drm_mm_insert_node_generic(&mm, &nodes[n].node,
-						 1, 0, color++,
-						 DRM_MM_SEARCH_DEFAULT,
-						 DRM_MM_CREATE_DEFAULT);
+						 1, 0, color++, 0);
 		if (err) {
 			pr_err("insert failed, step %d\n", n);
 			ret = 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 782a51d7bbab..3bc42a31307c 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)
@@ -239,14 +236,6 @@ 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
@@ -262,57 +251,27 @@ 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
@@ -321,8 +280,7 @@ 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.
  *
@@ -333,13 +291,11 @@ 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);
 }
 
 /**
@@ -347,8 +303,6 @@ 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.
@@ -360,13 +314,9 @@ 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);
@@ -413,7 +363,7 @@ 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
@@ -422,7 +372,7 @@ 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.
@@ -436,12 +386,11 @@ 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 v2 40/40] 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      | 102 +++++++++++++++++++++++++++
 2 files changed, 103 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 97468b8b33a0..081b5a1c565f 100644
--- a/drivers/gpu/drm/selftests/test-drm_mm.c
+++ b/drivers/gpu/drm/selftests/test-drm_mm.c
@@ -1653,6 +1653,108 @@ static int igt_topdown(void *ignored)
 	return ret;
 }
 
+static int igt_bottomup(void *ignored)
+{
+	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, err;
+
+	/* 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++) {
+			err = drm_mm_insert_node_generic(&mm, &nodes[n],
+							 size, 0, n,
+							 DRM_MM_INSERT_LOW);
+			if (err || !assert_node(&nodes[n], &mm, size, 0, n)) {
+				pr_err("bottomup insert failed, size %u step %d\n", size, n);
+				ret = err ?: -EINVAL;
+				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);
+		drm_for_each_prime(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]];
+				err = drm_mm_insert_node_generic(&mm, node, size, 0, 0,
+								 DRM_MM_INSERT_LOW);
+				if (err || !assert_node(node, &mm, size, 0, 0)) {
+					pr_err("insert failed, step %d/%d\n", m, n);
+					ret = err ?: -EINVAL;
+					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

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

Re: [PATCH v2 01/40] drm/i915: Use the MRU stack search after evicting

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +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.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> #v1

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 03/40] drm: Add drm_mm_for_each_node_safe()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> A complement to drm_mm_for_each_node(), wraps list_for_each_entry_safe()
> for walking the list of nodes safe against removal.
> 

Most of the diff is about __drm_mm_nodes(mm), which could be split into
own patch and keep the R-b's.

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
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 [v2,01/40] drm/i915: Use the MRU stack search after evicting

[Patchwork]

== Series Details ==

Series: series starting with [v2,01/40] drm/i915: Use the MRU stack search after evicting
URL   : https://patchwork.freedesktop.org/series/16906/
State : success

== Summary ==

Series 16906v1 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/16906/revisions/1/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-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 

428e9f939ad72727abcd02be76fdf41be3672a5b drm-tip: 2016y-12m-16d-08h-25m-46s UTC integration manifest
5bb157c drm: kselftest for drm_mm and bottom-up allocation
4e89e38 drm: Improve drm_mm search (and fix topdown allocation) with rbtrees
25596d7 drm: Use drm_mm_insert_node_in_range_generic() for everyone
419b0874 drm: Apply range restriction after color adjustment when allocation
ddaf9d5 drm: Wrap drm_mm_node.hole_follows
d680764 drm: Apply tight eviction scanning to color_adjust
d8a5adf drm: Simplify drm_mm scan-list manipulation
3abacac drm: Optimise power-of-two alignments in drm_mm_scan_add_block()
48ff21d drm: Compute tight evictions for drm_mm_scan
71bad2c drm: Fix application of color vs range restriction when scanning drm_mm
2e4ed5a drm: Unconditionally do the range check in drm_mm_scan_add_block()
64b0df4 drm: Rename prev_node to hole in drm_mm_scan_add_block()
78fe28e drm: Extract struct drm_mm_scan from struct drm_mm
9d2a85a drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()
1031757 drm: Simplify drm_mm_clean()
87dd432 drm: Detect overflow in drm_mm_reserve_node()
e797ab0 drm: Fix kerneldoc for drm_mm_scan_remove_block()
0d8045e drm: Promote drm_mm alignment to u64
c93021e drm/i915: Build DRM range manager selftests for CI
20763e9 drm: kselftest for drm_mm and restricted color eviction
184bc9575 drm: kselftest for drm_mm and color eviction
cad08ee drm: kselftest for drm_mm and color adjustment
4c17168 drm: kselftest for drm_mm and top-down allocation
c669929 drm: kselftest for drm_mm and range restricted eviction
fed4139 drm: kselftest for drm_mm and eviction
dda6cf3 drm: kselftest for drm_mm and alignment
d735a21 drm: kselftest for drm_mm_insert_node_in_range()
10bbc02 drm: kselftest for drm_mm_replace_node()
04e4898 drm: kselftest for drm_mm_insert_node()
450f1db drm: kselftest for drm_mm_reserve_node()
5ccc47c drm: kselftest for drm_mm_debug()
1d041c9 drm: kselftest for drm_mm_init()
8de9fad drm: Add a simple prime number generator
02bbb5b drm: Add a simple generator of random permutations
75629c2 drm: Add some kselftests for the DRM range manager (struct drm_mm)
ef10a68 drm: Compile time enabling for asserts in drm_mm
0f89700 drm: Constify the drm_mm API
36f3c26 drm: Add drm_mm_for_each_node_safe()
9f0be0f drm/i915: Simplify i915_gtt_color_adjust()
c1d54ce drm/i915: Use the MRU stack search after evicting

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3306/
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 08/40] drm: Add a simple prime number generator

[Lukas Wunner]

On Fri, Dec 16, 2016 at 07:46:46AM +0000, Chris Wilson wrote:
> Prime numbers are interesting for testing components that use multiplies
> and divides, such as testing struct drm_mm alignment computations.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/Kconfig                 |   4 +
>  drivers/gpu/drm/Makefile                |   1 +
>  drivers/gpu/drm/lib/drm_prime_numbers.c | 175 ++++++++++++++++++++++++++++++++
>  drivers/gpu/drm/lib/drm_prime_numbers.h |  10 ++
>  4 files changed, 190 insertions(+)
>  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.c
>  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.h

Hm, why not put this in lib/ ?  Don't see anything DRM-specific here
at first glance and this might be useful to others.  Or others might
come up with improvements and they'll be more likely to discover it
outside of DRM.

Same for the random permutations patch.

Thanks,

Lukas

> 
> diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
> index 2e6ae95459e4..93895898d596 100644
> --- a/drivers/gpu/drm/Kconfig
> +++ b/drivers/gpu/drm/Kconfig
> @@ -53,6 +53,7 @@ config DRM_DEBUG_MM_SELFTEST
>  	depends on DRM
>  	depends on DEBUG_KERNEL
>  	select DRM_LIB_RANDOM
> +	select DRM_LIB_PRIMES
>  	default n
>  	help
>  	  This option provides a kernel module that can be used to test
> @@ -340,3 +341,6 @@ config DRM_LIB_RANDOM
>  	bool
>  	default n
>  
> +config DRM_LIB_PRIMES
> +	bool
> +	default n
> diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
> index 0fa16275fdae..bbd390fa8914 100644
> --- a/drivers/gpu/drm/Makefile
> +++ b/drivers/gpu/drm/Makefile
> @@ -19,6 +19,7 @@ drm-y       :=	drm_auth.o drm_bufs.o drm_cache.o \
>  		drm_dumb_buffers.o drm_mode_config.o
>  
>  drm-$(CONFIG_DRM_LIB_RANDOM) += lib/drm_random.o
> +obj-$(CONFIG_DRM_LIB_PRIMES) += lib/drm_prime_numbers.o
>  obj-$(CONFIG_DRM_DEBUG_MM_SELFTEST) += selftests/test-drm_mm.o
>  
>  drm-$(CONFIG_COMPAT) += drm_ioc32.o
> diff --git a/drivers/gpu/drm/lib/drm_prime_numbers.c b/drivers/gpu/drm/lib/drm_prime_numbers.c
> new file mode 100644
> index 000000000000..839563d9b787
> --- /dev/null
> +++ b/drivers/gpu/drm/lib/drm_prime_numbers.c
> @@ -0,0 +1,175 @@
> +#include <linux/module.h>
> +#include <linux/mutex.h>
> +#include <linux/slab.h>
> +
> +#include "drm_prime_numbers.h"
> +
> +static DEFINE_MUTEX(lock);
> +
> +static struct primes {
> +	struct rcu_head rcu;
> +	unsigned long last, sz;
> +	unsigned long primes[];
> +} __rcu *primes;
> +
> +static bool slow_is_prime_number(unsigned long x)
> +{
> +	unsigned long y = int_sqrt(x) + 1;
> +
> +	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 = (start / x + 1) * x;
> +
> +	while (m < end) {
> +		__clear_bit(m, p);
> +		m += x;
> +	}
> +
> +	return x;
> +}
> +
> +static struct primes *expand(unsigned long x)
> +{
> +	unsigned long sz, y, prev;
> +	struct primes *p, *new;
> +
> +	sz = x * x;
> +	if (sz < x)
> +		return NULL;
> +
> +	mutex_lock(&lock);
> +	p = rcu_dereference_protected(primes, lockdep_is_held(&lock));
> +	if (p && x < p->last)
> +		goto unlock;
> +
> +	sz = round_up(sz, BITS_PER_LONG);
> +	new = kmalloc(sizeof(*new) + sz / sizeof(long), GFP_KERNEL);
> +	if (!new) {
> +		p = NULL;
> +		goto unlock;
> +	}
> +
> +	/* Where memory permits, track the primes using the
> +	 * Sieve of Eratosthenes.
> +	 */
> +	if (p) {
> +		prev = p->sz;
> +		memcpy(new->primes, p->primes, prev / BITS_PER_LONG);
> +	} else {
> +		prev = 0;
> +	}
> +	memset(new->primes + prev / BITS_PER_LONG,
> +	       0xff, (sz - prev) / sizeof(long));
> +	for (y = 2UL; y < sz; y = find_next_bit(new->primes, sz, y + 1))
> +		new->last = mark_multiples(y, new->primes, prev, sz);
> +	new->sz = sz;
> +
> +	rcu_assign_pointer(primes, new);
> +	if (p)
> +		kfree_rcu(p, rcu);
> +	p = new;
> +
> +unlock:
> +	mutex_unlock(&lock);
> +	return p;
> +}
> +
> +unsigned long drm_next_prime_number(unsigned long x)
> +{
> +	struct primes *p;
> +
> +	if (x < 2)
> +		return 2;
> +
> +	rcu_read_lock();
> +	p = rcu_dereference(primes);
> +	if (!p || x >= p->last) {
> +		rcu_read_unlock();
> +
> +		p = expand(x);
> +		if (!p)
> +			return slow_next_prime_number(x);
> +
> +		rcu_read_lock();
> +	}
> +
> +	x = find_next_bit(p->primes, p->last, x + 1);
> +	rcu_read_unlock();
> +
> +	return x;
> +}
> +EXPORT_SYMBOL(drm_next_prime_number);
> +
> +bool drm_is_prime_number(unsigned long x)
> +{
> +	struct primes *p;
> +	bool result;
> +
> +	switch (x) {
> +	case 0:
> +		return false;
> +	case 1:
> +	case 2:
> +	case 3:
> +		return true;
> +	}
> +
> +	rcu_read_lock();
> +	p = rcu_dereference(primes);
> +	if (!p || x >= p->last) {
> +		rcu_read_unlock();
> +
> +		p = expand(x);
> +		if (!p)
> +			return slow_is_prime_number(x);
> +
> +		rcu_read_lock();
> +	}
> +
> +	result = test_bit(x, p->primes);
> +	rcu_read_unlock();
> +
> +	return result;
> +}
> +EXPORT_SYMBOL(drm_is_prime_number);
> +
> +static int __init drm_primes_init(void)
> +{
> +	return 0;
> +}
> +
> +static void __exit drm_primes_exit(void)
> +{
> +	if (primes)
> +		kfree_rcu(primes, rcu);
> +}
> +
> +module_init(drm_primes_init);
> +module_exit(drm_primes_exit);
> +
> +MODULE_AUTHOR("Intel Corporation");
> +MODULE_LICENSE("GPL");
> diff --git a/drivers/gpu/drm/lib/drm_prime_numbers.h b/drivers/gpu/drm/lib/drm_prime_numbers.h
> new file mode 100644
> index 000000000000..7bc58cf9a86c
> --- /dev/null
> +++ b/drivers/gpu/drm/lib/drm_prime_numbers.h
> @@ -0,0 +1,10 @@
> +#ifndef __DRM_PRIMES_H__
> +#define __DRM_PRIMES_H__
> +
> +bool drm_is_prime_number(unsigned long x);
> +unsigned long drm_next_prime_number(unsigned long x);
> +
> +#define drm_for_each_prime(prime, max) \
> +	for (prime = 1;	prime < (max); prime = drm_next_prime_number(prime))
> +
> +#endif /* __DRM_PRIMES_H__ */
> -- 
> 2.11.0
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 07/40] drm: Add a simple generator of random permutations

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> 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.
> 
> v2: Tidier code by David Herrmann
> 
> 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>

<SNIP>

> @@ -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 prandom_u32_max_state(u32 ep_ro, struct rnd_state *state)
> +{
> +	return upper_32_bits((u64)prandom_u32_state(state) * ep_ro);
> +}
> +

To be submitted upstream. If you prefix the function here, there won't
be a conflict when the upstream part gets merged.

> +++ b/drivers/gpu/drm/lib/drm_random.h
> @@ -0,0 +1,21 @@
> +#ifndef __DRM_RANDOM_H__
> +#define __DRM_RANDOM_H
> +
> +#include <linux/random.h>
> +
> +#define RND_STATE_INITIALIZER(seed__) ({				\
> +	struct rnd_state state__;					\
> +	prandom_seed_state(&state__, (seed__));				\
> +	state__;							\
> +})
> +
> +#define RND_STATE(name__, seed__) \
> +	struct rnd_state name__ = RND_STATE_INITIALIZER(seed__)

For upstream submission too. Same comment as above.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 09/40] drm: kselftest for drm_mm_init()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> 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>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 08/40] drm: Add a simple prime number generator

[Chris Wilson]

On Fri, Dec 16, 2016 at 10:31:17AM +0100, Lukas Wunner wrote:
> On Fri, Dec 16, 2016 at 07:46:46AM +0000, Chris Wilson wrote:
> > Prime numbers are interesting for testing components that use multiplies
> > and divides, such as testing struct drm_mm alignment computations.
> > 
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > ---
> >  drivers/gpu/drm/Kconfig                 |   4 +
> >  drivers/gpu/drm/Makefile                |   1 +
> >  drivers/gpu/drm/lib/drm_prime_numbers.c | 175 ++++++++++++++++++++++++++++++++
> >  drivers/gpu/drm/lib/drm_prime_numbers.h |  10 ++
> >  4 files changed, 190 insertions(+)
> >  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.c
> >  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.h
> 
> Hm, why not put this in lib/ ?  Don't see anything DRM-specific here
> at first glance and this might be useful to others.  Or others might
> come up with improvements and they'll be more likely to discover it
> outside of DRM.

Because that is a 3+ month cycle before I can then apply the testcases,
and without the testscases do you want the bugfixes?

If I put in in drm/lib then lift it, I can use it immediately and drop
the local copy once merged.
-Chris

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

Re: [PATCH v2 10/40] drm: kselftest for drm_mm_debug()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> Simple test to just exercise calling the debug dumper on the drm_mm.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

This is rather meta already. Not entirely sure how good of a selftest
this is when we do not validate the generated output, or do you at the
runner side?

Code itself is (but I'm unsure of the usefulness as is);

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 10/40] drm: kselftest for drm_mm_debug()

[Chris Wilson]

On Fri, Dec 16, 2016 at 11:44:39AM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> > Simple test to just exercise calling the debug dumper on the drm_mm.
> > 
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> 
> This is rather meta already. Not entirely sure how good of a selftest
> this is when we do not validate the generated output, or do you at the
> runner side?

No, it is just to ensure we have coverage of that function. All it does is
make sure it doesn't explode, under very tame circumstances.

I thought of doing a few mocks to capture the output and decided that
was asking a little too much for a debug function. Still don't have
coverage for the debugfs dumper...
-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

Re: [PATCH v2 08/40] drm: Add a simple prime number generator

[Lukas Wunner]

On Fri, Dec 16, 2016 at 09:43:54AM +0000, Chris Wilson wrote:
> On Fri, Dec 16, 2016 at 10:31:17AM +0100, Lukas Wunner wrote:
> > On Fri, Dec 16, 2016 at 07:46:46AM +0000, Chris Wilson wrote:
> > > Prime numbers are interesting for testing components that use multiplies
> > > and divides, such as testing struct drm_mm alignment computations.
> > > 
> > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > > ---
> > >  drivers/gpu/drm/Kconfig                 |   4 +
> > >  drivers/gpu/drm/Makefile                |   1 +
> > >  drivers/gpu/drm/lib/drm_prime_numbers.c | 175 ++++++++++++++++++++++++++++++++
> > >  drivers/gpu/drm/lib/drm_prime_numbers.h |  10 ++
> > >  4 files changed, 190 insertions(+)
> > >  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.c
> > >  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.h
> > 
> > Hm, why not put this in lib/ ?  Don't see anything DRM-specific here
> > at first glance and this might be useful to others.  Or others might
> > come up with improvements and they'll be more likely to discover it
> > outside of DRM.
> 
> Because that is a 3+ month cycle before I can then apply the testcases,
> and without the testscases do you want the bugfixes?

Do patches for lib/ have to go through a different tree?
Don't think so, I've seen e.g. changes to lib/ucs2_string.c
go through the EFI tree.  It seems to me lib/ is sort of
free for all.


> If I put in in drm/lib then lift it, I can use it immediately and drop
> the local copy once merged.

That is also workable of course.  Anyway, it was just a suggestion.

Thanks,

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

Re: [PATCH v2 24/40] drm: Fix kerneldoc for drm_mm_scan_remove_block()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

It's like a Christmas gift for Daniel.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 25/40] drm: Detect overflow in drm_mm_reserve_node()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

I was about to suggest an additional check (but didn't). A combined
check is much more clever.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 08/40] drm: Add a simple prime number generator

[Chris Wilson]

On Fri, Dec 16, 2016 at 11:08:10AM +0100, Lukas Wunner wrote:
> On Fri, Dec 16, 2016 at 09:43:54AM +0000, Chris Wilson wrote:
> > On Fri, Dec 16, 2016 at 10:31:17AM +0100, Lukas Wunner wrote:
> > > On Fri, Dec 16, 2016 at 07:46:46AM +0000, Chris Wilson wrote:
> > > > Prime numbers are interesting for testing components that use multiplies
> > > > and divides, such as testing struct drm_mm alignment computations.
> > > > 
> > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > > > ---
> > > >  drivers/gpu/drm/Kconfig                 |   4 +
> > > >  drivers/gpu/drm/Makefile                |   1 +
> > > >  drivers/gpu/drm/lib/drm_prime_numbers.c | 175 ++++++++++++++++++++++++++++++++
> > > >  drivers/gpu/drm/lib/drm_prime_numbers.h |  10 ++
> > > >  4 files changed, 190 insertions(+)
> > > >  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.c
> > > >  create mode 100644 drivers/gpu/drm/lib/drm_prime_numbers.h
> > > 
> > > Hm, why not put this in lib/ ?  Don't see anything DRM-specific here
> > > at first glance and this might be useful to others.  Or others might
> > > come up with improvements and they'll be more likely to discover it
> > > outside of DRM.
> > 
> > Because that is a 3+ month cycle before I can then apply the testcases,
> > and without the testscases do you want the bugfixes?
> 
> Do patches for lib/ have to go through a different tree?
> Don't think so, I've seen e.g. changes to lib/ucs2_string.c
> go through the EFI tree.  It seems to me lib/ is sort of
> free for all.

Hmm, I was expecting to shepherd them through say Andrew Morton.
lib/random32.c is maintained by David Miller, so definitely would like
to present a simple set of pre-reviewed patches.

But it looks like we could create lib/prime_numbers.c without too much
consternation.
-Chris

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

Re: [PATCH v2 37/40] drm: Apply range restriction after color adjustment when allocation

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 38/40] drm: Use drm_mm_insert_node_in_range_generic() for everyone

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 36/40] drm: Wrap drm_mm_node.hole_follows

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> Insulate users from changed to the internal hole tracking within

changes ----------------^

> struct drm_mm_node by using an accessor for hole_follows.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

The function name could be soemthing beginning with drm_mm_node...

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 34/40] drm: Simplify drm_mm scan-list manipulation

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 27/40] drm: Add asserts to catch overflow in drm_mm_init() and drm_mm_init_scan()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

<SNIP>

> @@ -729,6 +729,8 @@ void drm_mm_init_scan(struct drm_mm *mm,
>  		      u64 alignment,
>  		      unsigned long color)
>  {
> +	DRM_MM_BUG_ON(size == 0);

Nitpicking, DIM will complain that this should be !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)
>  {

Ditto, could have simply DRM_MM_BUG_ON(!size);

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas

> +	DRM_MM_BUG_ON(start >= end);
> +	DRM_MM_BUG_ON(size == 0 || size > end - start);
> +
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
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

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           | 230 ++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 252 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..ba08be0ff1ed
--- /dev/null
+++ b/lib/prime_numbers.c
@@ -0,0 +1,230 @@
+#define pr_fmt(fmt) "prime numbers: " #fmt
+
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/prime_numbers.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.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 = { 0x28208a20a08a28ae }
+};
+#elif BITS_PER_LONG == 32
+static const struct primes small_primes = {
+	.last = 31,
+	.sz = 32,
+	.primes = { 0xa08a28ae }
+};
+#else
+#error "unhandled BITS_PER_LONG"
+#endif
+
+static DEFINE_MUTEX(lock);
+static const struct primes __rcu *primes = &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 | __GFP_NORETRY | __GFP_NOWARN);
+	if (!new)
+		new = vmalloc(sizeof(*new) + sz / sizeof(long));
+	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 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 = slow_is_prime_number(x);
+
+		if (slow != fast) {
+			pr_err("inconsistent result for is-prime(%lu): slow=%s, fast=%s!\n",
+			       x, slow ? "yes" : "no", fast ? "yes" : "no");
+			rcu_read_lock();
+			pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...x%lx}",
+				primes->last, primes->sz,
+				primes->primes[primes->sz / BITS_PER_LONG - 1]);
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+
+		if (!slow)
+			continue;
+
+		if (next_prime_number(last) != x) {
+			pr_err("incorrect result for next-prime(%lu): expected %lu, got %lu\n",
+			       last, x, next_prime_number(last));
+			rcu_read_lock();
+			pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...0x%lx}",
+				primes->last, primes->sz,
+				primes->primes[primes->sz / BITS_PER_LONG - 1]);
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+		last = x;
+	}
+
+	pr_info("selftest(%lu) passed, last prime was %lu\n", x, last);
+	return 0;
+}
+
+static int __init primes_init(void)
+{
+	return selftest(selftest_max);
+}
+
+static void __exit primes_exit(void)
+{
+	if (primes != &small_primes)
+		kfree_rcu((struct primes *)primes, rcu);
+}
+
+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 v2 36/40] drm: Wrap drm_mm_node.hole_follows

[Chris Wilson]

On Fri, Dec 16, 2016 at 03:04:31PM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> > Insulate users from changed to the internal hole tracking within
> 
> changes ----------------^
> 
> > struct drm_mm_node by using an accessor for hole_follows.
> > 
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> 
> The function name could be soemthing beginning with drm_mm_node...

Didn't like it as drm_mm_node_hole_follows(), and rationalised it away as
saying this is really drm_mm state that happens to be hidden away inside 
the user's drm_mm_node. I guess we should look at some /* private */ markup.
-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

Re: [PATCH v2 03/40] drm: Add drm_mm_for_each_node_safe()

[Daniel Vetter]

On Fri, Dec 16, 2016 at 11:06:25AM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> > A complement to drm_mm_for_each_node(), wraps list_for_each_entry_safe()
> > for walking the list of nodes safe against removal.
> > 
> 
> Most of the diff is about __drm_mm_nodes(mm), which could be split into
> own patch and keep the R-b's.

Added a note and applied to drm-misc, thanks.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 36/40] drm: Wrap drm_mm_node.hole_follows

[Joonas Lahtinen]

On pe, 2016-12-16 at 13:31 +0000, Chris Wilson wrote:
> On Fri, Dec 16, 2016 at 03:04:31PM +0200, Joonas Lahtinen wrote:
> > 
> > On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> > > 
> > > Insulate users from changed to the internal hole tracking within
> > 
> > changes ----------------^
> > 
> > > 
> > > struct drm_mm_node by using an accessor for hole_follows.
> > > 
> > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > 
> > The function name could be soemthing beginning with drm_mm_node...
> 
> Didn't like it as drm_mm_node_hole_follows(), and rationalised it away as
> saying this is really drm_mm state that happens to be hidden away inside 
> the user's drm_mm_node. I guess we should look at some /* private */ markup.

That makes sense now that you mention it.

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 04/40] drm: Constify the drm_mm API

[Daniel Vetter]

On Fri, Dec 16, 2016 at 07:46:42AM +0000, Chris Wilson wrote:
> Mark up the pointers as constant through the API where appropriate.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

Ok merged this and the patch right before, then Chris told me on irc that
I need to stop because he's mixing a v3 of this series.
-Daniel

> ---
>  drivers/gpu/drm/drm_mm.c            | 24 ++++++++++++------------
>  drivers/gpu/drm/i915/i915_gem_gtt.c |  2 +-
>  include/drm/drm_mm.h                | 27 +++++++++++++--------------
>  3 files changed, 26 insertions(+), 27 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
> index 6e0735539545..7573661302a4 100644
> --- a/drivers/gpu/drm/drm_mm.c
> +++ b/drivers/gpu/drm/drm_mm.c
> @@ -174,9 +174,9 @@ INTERVAL_TREE_DEFINE(struct drm_mm_node, rb,
>  		     START, LAST, static inline, drm_mm_interval_tree)
>  
>  struct drm_mm_node *
> -__drm_mm_interval_first(struct drm_mm *mm, u64 start, u64 last)
> +__drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last)
>  {
> -	return drm_mm_interval_tree_iter_first(&mm->interval_tree,
> +	return drm_mm_interval_tree_iter_first((struct rb_root *)&mm->interval_tree,
>  					       start, last);
>  }
>  EXPORT_SYMBOL(__drm_mm_interval_first);
> @@ -881,9 +881,9 @@ EXPORT_SYMBOL(drm_mm_scan_remove_block);
>   * True if the allocator is completely free, false if there's still a node
>   * allocated in it.
>   */
> -bool drm_mm_clean(struct drm_mm * mm)
> +bool drm_mm_clean(const struct drm_mm *mm)
>  {
> -	struct list_head *head = __drm_mm_nodes(mm);
> +	const struct list_head *head = __drm_mm_nodes(mm);
>  
>  	return (head->next->next == head);
>  }
> @@ -897,7 +897,7 @@ EXPORT_SYMBOL(drm_mm_clean);
>   *
>   * Note that @mm must be cleared to 0 before calling this function.
>   */
> -void drm_mm_init(struct drm_mm * mm, u64 start, u64 size)
> +void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
>  {
>  	INIT_LIST_HEAD(&mm->hole_stack);
>  	mm->scanned_blocks = 0;
> @@ -936,8 +936,8 @@ void drm_mm_takedown(struct drm_mm *mm)
>  }
>  EXPORT_SYMBOL(drm_mm_takedown);
>  
> -static u64 drm_mm_debug_hole(struct drm_mm_node *entry,
> -				     const char *prefix)
> +static u64 drm_mm_debug_hole(const struct drm_mm_node *entry,
> +			     const char *prefix)
>  {
>  	u64 hole_start, hole_end, hole_size;
>  
> @@ -958,9 +958,9 @@ static u64 drm_mm_debug_hole(struct drm_mm_node *entry,
>   * @mm: drm_mm allocator to dump
>   * @prefix: prefix to use for dumping to dmesg
>   */
> -void drm_mm_debug_table(struct drm_mm *mm, const char *prefix)
> +void drm_mm_debug_table(const struct drm_mm *mm, const char *prefix)
>  {
> -	struct drm_mm_node *entry;
> +	const struct drm_mm_node *entry;
>  	u64 total_used = 0, total_free = 0, total = 0;
>  
>  	total_free += drm_mm_debug_hole(&mm->head_node, prefix);
> @@ -979,7 +979,7 @@ void drm_mm_debug_table(struct drm_mm *mm, const char *prefix)
>  EXPORT_SYMBOL(drm_mm_debug_table);
>  
>  #if defined(CONFIG_DEBUG_FS)
> -static u64 drm_mm_dump_hole(struct seq_file *m, struct drm_mm_node *entry)
> +static u64 drm_mm_dump_hole(struct seq_file *m, const struct drm_mm_node *entry)
>  {
>  	u64 hole_start, hole_end, hole_size;
>  
> @@ -1000,9 +1000,9 @@ static u64 drm_mm_dump_hole(struct seq_file *m, struct drm_mm_node *entry)
>   * @m: seq_file to dump to
>   * @mm: drm_mm allocator to dump
>   */
> -int drm_mm_dump_table(struct seq_file *m, struct drm_mm *mm)
> +int drm_mm_dump_table(struct seq_file *m, const struct drm_mm *mm)
>  {
> -	struct drm_mm_node *entry;
> +	const struct drm_mm_node *entry;
>  	u64 total_used = 0, total_free = 0, total = 0;
>  
>  	total_free += drm_mm_dump_hole(m, &mm->head_node);
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> index 4d82f38a000b..97fd66e4e3d0 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> @@ -2699,7 +2699,7 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
>  	dma_unmap_sg(kdev, pages->sgl, pages->nents, PCI_DMA_BIDIRECTIONAL);
>  }
>  
> -static void i915_gtt_color_adjust(struct drm_mm_node *node,
> +static void i915_gtt_color_adjust(const struct drm_mm_node *node,
>  				  unsigned long color,
>  				  u64 *start,
>  				  u64 *end)
> diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
> index 0cc1b78c9ec2..5c7f15875b6a 100644
> --- a/include/drm/drm_mm.h
> +++ b/include/drm/drm_mm.h
> @@ -102,7 +102,8 @@ struct drm_mm {
>  	u64 scan_end;
>  	struct drm_mm_node *prev_scanned_node;
>  
> -	void (*color_adjust)(struct drm_mm_node *node, unsigned long color,
> +	void (*color_adjust)(const struct drm_mm_node *node,
> +			     unsigned long color,
>  			     u64 *start, u64 *end);
>  };
>  
> @@ -116,7 +117,7 @@ struct drm_mm {
>   * Returns:
>   * True if the @node is allocated.
>   */
> -static inline bool drm_mm_node_allocated(struct drm_mm_node *node)
> +static inline bool drm_mm_node_allocated(const struct drm_mm_node *node)
>  {
>  	return node->allocated;
>  }
> @@ -131,12 +132,12 @@ static inline bool drm_mm_node_allocated(struct drm_mm_node *node)
>   * Returns:
>   * True if the @mm is initialized.
>   */
> -static inline bool drm_mm_initialized(struct drm_mm *mm)
> +static inline bool drm_mm_initialized(const struct drm_mm *mm)
>  {
>  	return mm->hole_stack.next;
>  }
>  
> -static inline u64 __drm_mm_hole_node_start(struct drm_mm_node *hole_node)
> +static inline u64 __drm_mm_hole_node_start(const struct drm_mm_node *hole_node)
>  {
>  	return hole_node->start + hole_node->size;
>  }
> @@ -152,13 +153,13 @@ static inline u64 __drm_mm_hole_node_start(struct drm_mm_node *hole_node)
>   * Returns:
>   * Start of the subsequent hole.
>   */
> -static inline u64 drm_mm_hole_node_start(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);
>  	return __drm_mm_hole_node_start(hole_node);
>  }
>  
> -static inline u64 __drm_mm_hole_node_end(struct drm_mm_node *hole_node)
> +static inline u64 __drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
>  {
>  	return list_next_entry(hole_node, node_list)->start;
>  }
> @@ -174,7 +175,7 @@ static inline u64 __drm_mm_hole_node_end(struct drm_mm_node *hole_node)
>   * Returns:
>   * End of the subsequent hole.
>   */
> -static inline u64 drm_mm_hole_node_end(struct drm_mm_node *hole_node)
> +static inline u64 drm_mm_hole_node_end(const struct drm_mm_node *hole_node)
>  {
>  	return __drm_mm_hole_node_end(hole_node);
>  }
> @@ -314,14 +315,12 @@ static inline int drm_mm_insert_node_in_range(struct drm_mm *mm,
>  
>  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_init(struct drm_mm *mm, u64 start, u64 size);
>  void drm_mm_takedown(struct drm_mm *mm);
> -bool drm_mm_clean(struct drm_mm *mm);
> +bool drm_mm_clean(const struct drm_mm *mm);
>  
>  struct drm_mm_node *
> -__drm_mm_interval_first(struct drm_mm *mm, u64 start, u64 last);
> +__drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last);
>  
>  /**
>   * drm_mm_for_each_node_in_range - iterator to walk over a range of
> @@ -355,9 +354,9 @@ void drm_mm_init_scan_with_range(struct drm_mm *mm,
>  bool drm_mm_scan_add_block(struct drm_mm_node *node);
>  bool drm_mm_scan_remove_block(struct drm_mm_node *node);
>  
> -void drm_mm_debug_table(struct drm_mm *mm, const char *prefix);
> +void drm_mm_debug_table(const struct drm_mm *mm, const char *prefix);
>  #ifdef CONFIG_DEBUG_FS
> -int drm_mm_dump_table(struct seq_file *m, struct drm_mm *mm);
> +int drm_mm_dump_table(struct seq_file *m, const struct drm_mm *mm);
>  #endif
>  
>  #endif
> -- 
> 2.11.0
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 39/40] drm: Improve drm_mm search (and fix topdown allocation) with rbtrees

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

<SNIP>

> +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;

Missed this split.

With that addressed, this is;

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

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

[Chris Wilson]

On Fri, Dec 16, 2016 at 01:23:23PM +0000, Chris Wilson wrote:
> 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
> 
> 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           | 230 ++++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 252 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..ba08be0ff1ed
> --- /dev/null
> +++ b/lib/prime_numbers.c
> @@ -0,0 +1,230 @@
> +#define pr_fmt(fmt) "prime numbers: " #fmt
> +
> +#include <linux/module.h>
> +#include <linux/mutex.h>
> +#include <linux/prime_numbers.h>
> +#include <linux/slab.h>
> +#include <linux/vmalloc.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 = { 0x28208a20a08a28ae }

Ugh, extracted this from primes[0] and forgot the mark up skips 0,1

> +	for (last = 0, x = 2; x < max; x++) {
> +		bool slow = slow_is_prime_number(x);
> +		bool fast = slow_is_prime_number(x);

Which didn't help!
-Chris

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

Re: [PATCH v2 39/40] drm: Improve drm_mm search (and fix topdown allocation) with rbtrees

[Chris Wilson]

On Fri, Dec 16, 2016 at 03:46:18PM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> > 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>
> 
> <SNIP>
> 
> > +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;
> 
> Missed this split.

You must own shares in a company selling braces! :-p
-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

Re: [PATCH v2 12/40] drm: kselftest for drm_mm_insert_node()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> 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>

<SNIP>
 
> +static u64 misaligned(struct drm_mm_node *node, u64 alignment)

I'm not sure if 'misalignment' would be better name or not. This makes
me think of bool returning one.

> +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 (err != -ENOSPC)  {

For speed (this function gets called a lot);

if (likely(err == -ENOSPC))
	return true;

> +static int __igt_insert(unsigned int count, u64 size)
> +{

<SNIP>

> 
> +	for (mode = insert_modes; mode->name; mode++) {
> +		for (n = 0; n < count; n++) {
> +			node = &nodes[n];
> +			err = drm_mm_insert_node_generic(&mm, node, size, 0, n,
> +							 mode->search_flags,
> +							 mode->create_flags);
> +			if (err || !assert_node(node, &mm, size, 0, n)) {
> +				pr_err("%s insert failed, size %llu step %d\n",
> +				       mode->name, size, n);
> +				ret = err ?: -EINVAL;
> +				goto out;
> +			}

This construct is three times in this patch; Could be
expect_insert_generic?

Apart from nitpicks, no complaints; was much easier to review now!

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH v3] 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)

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           | 230 ++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 252 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..8c3cdd05d4ae
--- /dev/null
+++ b/lib/prime_numbers.c
@@ -0,0 +1,230 @@
+#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>
+#include <linux/vmalloc.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 = { 0x28208a20a08a28ac }
+};
+#elif BITS_PER_LONG == 32
+static const struct primes small_primes = {
+	.last = 31,
+	.sz = 32,
+	.primes = { 0xa08a28ac }
+};
+#else
+#error "unhandled BITS_PER_LONG"
+#endif
+
+static DEFINE_MUTEX(lock);
+static const struct primes __rcu *primes = &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 | __GFP_NORETRY | __GFP_NOWARN);
+	if (!new)
+		new = vmalloc(sizeof(*new) + sz / sizeof(long));
+	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 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");
+			rcu_read_lock();
+			pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...x%lx}",
+				primes->last, primes->sz,
+				primes->primes[primes->sz / BITS_PER_LONG - 1]);
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+
+		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));
+			rcu_read_lock();
+			pr_info("primes.{last=%lu, .sz=%lu, .primes[]=...0x%lx}",
+				primes->last, primes->sz,
+				primes->primes[primes->sz / BITS_PER_LONG - 1]);
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+		last = x;
+	}
+
+	pr_info("selftest(%lu) passed, last prime was %lu", x, last);
+	return 0;
+}
+
+static int __init primes_init(void)
+{
+	return selftest(selftest_max);
+}
+
+static void __exit primes_exit(void)
+{
+	if (primes != &small_primes)
+		kfree_rcu((struct primes *)primes, rcu);
+}
+
+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

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

Re: [PATCH v2 14/40] drm: kselftest for drm_mm_insert_node_in_range()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> 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>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 35/40] drm: Apply tight eviction scanning to color_adjust

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

It's not optimal as a separate function, but good for now.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 35/40] drm: Apply tight eviction scanning to color_adjust

[Chris Wilson]

On Fri, Dec 16, 2016 at 04:14:30PM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> > 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>
> 
> It's not optimal as a separate function, but good for now.

Yes, it's nasty that it requires a separate step by the caller to catch
for residual overlap. The problem is that we do require the
mm.color_adjust to be performed to tell if the user thinks there will be
overlap - and that requires the final state of the hole and its
neighbours.

Alternatives I considered were just a color_expand parameter and always
evicting the neighbouring nodes (but that is not great for us as most of
our nodes have one colour). Or I considered searching the node list to
find the overlaps and flag them (presuming eviction happened as we
intended). They all seemed to be worse than just asking the user to
check after eviction if they are using color_adjust.
-Chris

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

Re: [PATCH v2 20/40] drm: kselftest for drm_mm and color eviction

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> 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>

<SNIP>

> +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)

expect_evict_color? evict_color gave me an impression of single-way
action (when reading the next patch).

Just a nitpick, so;

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 21/40] drm: kselftest for drm_mm and restricted color eviction

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> 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>

<SNIP> 

> +static int igt_color_evict_range(void *ignored)
> +{

<SNIP>

> +	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;
> +			}

This evict_color + error check again repeats, could maybe be helper.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH v2 29/40] drm: Rename prev_node to hole in drm_mm_scan_add_block()

[Joonas Lahtinen]

On pe, 2016-12-16 at 07:47 +0000, Chris Wilson wrote:
> 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>

I'm not purely against, so if you think it's more intuitive;

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH v2 20/40] drm: kselftest for drm_mm and color eviction

[Chris Wilson]

On Fri, Dec 16, 2016 at 04:38:12PM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> > 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>
> 
> <SNIP>
> 
> > +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)
> 
> expect_evict_color? evict_color gave me an impression of single-way
> action (when reading the next patch).

I thought of this and evict_something() as being the final level test
runner (with the callers iterating over various paramters).
expect_insert_fail/expect_insert_in_range_fail definitely are the single
action check result matches expectations (return bool). Not convinced
these two fall into the same category, so I'll leave as is.
-Chris

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

Re: [PATCH v2 12/40] drm: kselftest for drm_mm_insert_node()

[Chris Wilson]

On Fri, Dec 16, 2016 at 04:02:12PM +0200, Joonas Lahtinen wrote:
> On pe, 2016-12-16 at 07:46 +0000, Chris Wilson wrote:
> > 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>
> 
> <SNIP>
>  
> > +static u64 misaligned(struct drm_mm_node *node, u64 alignment)
> 
> I'm not sure if 'misalignment' would be better name or not. This makes
> me think of bool returning one.
> 
> > +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 (err != -ENOSPC)  {
> 
> For speed (this function gets called a lot);
> 
> if (likely(err == -ENOSPC))
> 	return true;
> 
> > +static int __igt_insert(unsigned int count, u64 size)
> > +{
> 
> <SNIP>
> 
> > 
> > +	for (mode = insert_modes; mode->name; mode++) {
> > +		for (n = 0; n < count; n++) {
> > +			node = &nodes[n];
> > +			err = drm_mm_insert_node_generic(&mm, node, size, 0, n,
> > +							 mode->search_flags,
> > +							 mode->create_flags);
> > +			if (err || !assert_node(node, &mm, size, 0, n)) {
> > +				pr_err("%s insert failed, size %llu step %d\n",
> > +				       mode->name, size, n);
> > +				ret = err ?: -EINVAL;
> > +				goto out;
> > +			}
> 
> This construct is three times in this patch; Could be
> expect_insert_generic?

In this patch alone, indeed it gets used again and again. I resisted
making it an insert and check function simply because I didn't want to
fall into the trap of using it everywhere and so missing out exercising
drm_mm_insert_node and friends.

However, 3 occurences of the same in one patch...
-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

✗ Fi.CI.BAT: failure for series starting with [v2,01/40] drm/i915: Use the MRU stack search after evicting (rev3)

[Patchwork]

== Series Details ==

Series: series starting with [v2,01/40] drm/i915: Use the MRU stack search after evicting (rev3)
URL   : https://patchwork.freedesktop.org/series/16906/
State : failure

== Summary ==

  CC      net/ipv4/tcp_cubic.o
  CC      net/ipv4/xfrm4_policy.o
  LD      drivers/net/phy/built-in.o
  CC      net/ipv4/xfrm4_state.o
  CC      net/ipv4/xfrm4_input.o
  CC      net/ipv4/xfrm4_output.o
  CC      net/ipv4/xfrm4_protocol.o
  CC      net/ipv6/exthdrs_offload.o
  CC      net/ipv6/inet6_hashtables.o
  CC      net/ipv6/mcast_snoop.o
  LD      drivers/rtc/built-in.o
  LD      drivers/pci/pcie/pcieportdrv.o
  LD [M]  drivers/net/ethernet/intel/igbvf/igbvf.o
  LD      drivers/acpi/acpica/acpi.o
  LD      kernel/events/built-in.o
  LD      net/netlink/built-in.o
  LD      net/unix/unix_diag.o
  LD      kernel/sched/built-in.o
  LD      drivers/acpi/acpica/built-in.o
  LD      drivers/pci/pcie/aer/aerdriver.o
  LD      kernel/built-in.o
  LD      drivers/pci/pcie/aer/built-in.o
  LD      drivers/usb/gadget/libcomposite.o
  LD      drivers/usb/storage/usb-storage.o
  LD      drivers/pci/pcie/built-in.o
  LD      drivers/acpi/built-in.o
  LD      drivers/usb/storage/built-in.o
  LD      net/key/built-in.o
  LD      drivers/tty/serial/8250/8250.o
  LD [M]  drivers/usb/serial/usbserial.o
  LD      drivers/thermal/thermal_sys.o
  LD      mm/built-in.o
  LD      drivers/thermal/built-in.o
  LD [M]  drivers/misc/mei/mei-me.o
  LD      lib/raid6/raid6_pq.o
  LD      drivers/misc/built-in.o
  LD      lib/raid6/built-in.o
  LD      drivers/spi/built-in.o
  LD      drivers/iommu/built-in.o
  LD      drivers/usb/gadget/udc/udc-core.o
  LD      drivers/usb/gadget/udc/built-in.o
  LD      drivers/usb/gadget/built-in.o
scripts/Makefile.build:544: recipe for target 'drivers/gpu/drm' failed
make[2]: *** [drivers/gpu/drm] Error 2
scripts/Makefile.build:544: recipe for target 'drivers/gpu' failed
make[1]: *** [drivers/gpu] Error 2
make[1]: *** Waiting for unfinished jobs....
  LD      net/packet/built-in.o
  LD      drivers/video/fbdev/core/fb.o
  LD      drivers/video/fbdev/core/built-in.o
  LD [M]  sound/pci/hda/snd-hda-codec-generic.o
  LD      drivers/video/fbdev/built-in.o
  LD [M]  drivers/net/ethernet/intel/e1000/e1000.o
  LD      sound/pci/built-in.o
  LD      net/unix/unix.o
  LD      net/unix/built-in.o
  LD      drivers/scsi/scsi_mod.o
  LD      drivers/video/console/built-in.o
  LD      drivers/video/built-in.o
  LD      sound/built-in.o
  LD [M]  drivers/net/ethernet/broadcom/genet/genet.o
  LD      drivers/tty/serial/8250/8250_base.o
  LD      drivers/tty/serial/8250/built-in.o
  LD      drivers/tty/serial/built-in.o
  LD      drivers/pci/built-in.o
  LD      drivers/scsi/sd_mod.o
  LD      drivers/scsi/built-in.o
  LD [M]  drivers/net/ethernet/intel/igb/igb.o
  LD      drivers/md/md-mod.o
  AR      lib/lib.a
  LD      drivers/md/built-in.o
  EXPORTS lib/lib-ksyms.o
  CC      arch/x86/kernel/cpu/capflags.o
  LD      arch/x86/kernel/cpu/built-in.o
  LD      drivers/usb/core/usbcore.o
  LD      arch/x86/kernel/built-in.o
  LD      drivers/usb/core/built-in.o
  LD      drivers/usb/host/xhci-hcd.o
  LD      arch/x86/built-in.o
  LD      drivers/tty/vt/built-in.o
  LD      net/xfrm/built-in.o
  LD      drivers/tty/built-in.o
  LD      net/ipv6/ipv6.o
  LD      fs/btrfs/btrfs.o
  LD      net/ipv6/built-in.o
  LD      lib/built-in.o
  LD      fs/btrfs/built-in.o
  LD      fs/ext4/ext4.o
  LD      fs/ext4/built-in.o
  LD      drivers/usb/host/built-in.o
  LD      drivers/usb/built-in.o
  LD      fs/built-in.o
  LD      net/ipv4/built-in.o
  LD [M]  drivers/net/ethernet/intel/e1000e/e1000e.o
  LD      net/core/built-in.o
  LD      net/built-in.o
  LD      drivers/net/ethernet/built-in.o
  LD      drivers/net/built-in.o
Makefile:988: recipe for target 'drivers' failed
make: *** [drivers] Error 2

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

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

[kbuild test robot]

Hi Chris,

[auto build test WARNING on linus/master]
[also build test WARNING on v4.9 next-20161216]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Chris-Wilson/lib-Add-a-simple-prime-number-generator/20161217-013805
reproduce:
        # apt-get install sparse
        make ARCH=x86_64 allmodconfig
        make C=1 CF=-D__CHECK_ENDIAN__


sparse warnings: (new ones prefixed by >>)

   include/linux/compiler.h:253:8: sparse: attribute 'no_sanitize_address': unknown attribute
>> lib/prime_numbers.c:220:20: sparse: incompatible types in comparison expression (different address spaces)

vim +220 lib/prime_numbers.c

   204				return -EINVAL;
   205			}
   206			last = x;
   207		}
   208	
   209		pr_info("selftest(%lu) passed, last prime was %lu", x, last);
   210		return 0;
   211	}
   212	
   213	static int __init primes_init(void)
   214	{
   215		return selftest(selftest_max);
   216	}
   217	
   218	static void __exit primes_exit(void)
   219	{
 > 220		if (primes != &small_primes)
   221			kfree_rcu((struct primes *)primes, rcu);
   222	}
   223	
   224	module_init(primes_init);
   225	module_exit(primes_exit);
   226	
   227	module_param_named(selftest, selftest_max, ulong, 0400);
   228	

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx