drm/nouveau/core: add support for reverse mm allocations

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/core/core/mm.c

@@ -1,5 +1,5 @@
 /*
- * Copyright 2010 Red Hat Inc.
+ * Copyright 2012 Red Hat Inc.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -22,19 +22,52 @@
  * Authors: Ben Skeggs
  */
 
-#include <core/os.h>
+#include "core/os.h"
-#include <core/mm.h>
+#include "core/mm.h"
 
-static inline void
+#define node(root, dir) ((root)->nl_entry.dir == &mm->nodes) ? NULL : \
-region_put(struct nouveau_mm *mm, struct nouveau_mm_node *a)
+	list_entry((root)->nl_entry.dir, struct nouveau_mm_node, nl_entry)
+
+void
+nouveau_mm_free(struct nouveau_mm *mm, struct nouveau_mm_node **pthis)
 {
-	list_del(&a->nl_entry);
+	struct nouveau_mm_node *this = *pthis;
-	list_del(&a->fl_entry);
+
-	kfree(a);
+	if (this) {
+		struct nouveau_mm_node *prev = node(this, prev);
+		struct nouveau_mm_node *next = node(this, next);
+
+		if (prev && prev->type == 0) {
+			prev->length += this->length;
+			list_del(&this->nl_entry);
+			kfree(this); this = prev;
+		}
+
+		if (next && next->type == 0) {
+			next->offset  = this->offset;
+			next->length += this->length;
+			if (this->type == 0)
+				list_del(&this->fl_entry);
+			list_del(&this->nl_entry);
+			kfree(this); this = NULL;
+		}
+
+		if (this && this->type != 0) {
+			list_for_each_entry(prev, &mm->free, fl_entry) {
+				if (this->offset < prev->offset)
+					break;
+			}
+
+			list_add_tail(&this->fl_entry, &prev->fl_entry);
+			this->type = 0;
+		}
+	}
+
+	*pthis = NULL;
 }
 
 static struct nouveau_mm_node *
-region_split(struct nouveau_mm *mm, struct nouveau_mm_node *a, u32 size)
+region_head(struct nouveau_mm *mm, struct nouveau_mm_node *a, u32 size)
 {
 	struct nouveau_mm_node *b;
 
@@ -56,38 +89,12 @@ region_split(struct nouveau_mm *mm, struct nouveau_mm_node *a, u32 size)
 	return b;
 }
 
-#define node(root, dir) ((root)->nl_entry.dir == &mm->nodes) ? NULL : \
-	list_entry((root)->nl_entry.dir, struct nouveau_mm_node, nl_entry)
-
-void
-nouveau_mm_put(struct nouveau_mm *mm, struct nouveau_mm_node *this)
-{
-	struct nouveau_mm_node *prev = node(this, prev);
-	struct nouveau_mm_node *next = node(this, next);
-
-	list_add(&this->fl_entry, &mm->free);
-	this->type = 0;
-
-	if (prev && prev->type == 0) {
-		prev->length += this->length;
-		region_put(mm, this);
-		this = prev;
-	}
-
-	if (next && next->type == 0) {
-		next->offset  = this->offset;
-		next->length += this->length;
-		region_put(mm, this);
-	}
-}
-
 int
-nouveau_mm_get(struct nouveau_mm *mm, int type, u32 size, u32 size_nc,
+nouveau_mm_head(struct nouveau_mm *mm, u8 type, u32 size_max, u32 size_min,
-	       u32 align, struct nouveau_mm_node **pnode)
+		u32 align, struct nouveau_mm_node **pnode)
 {
 	struct nouveau_mm_node *prev, *this, *next;
-	u32 min = size_nc ? size_nc : size;
+	u32 mask = align - 1;
-	u32 align_mask = align - 1;
 	u32 splitoff;
 	u32 s, e;
 
@@ -103,16 +110,86 @@ nouveau_mm_get(struct nouveau_mm *mm, int type, u32 size, u32 size_nc,
 		if (next && next->type != type)
 			e = rounddown(e, mm->block_size);
 
-		s  = (s + align_mask) & ~align_mask;
+		s  = (s + mask) & ~mask;
-		e &= ~align_mask;
+		e &= ~mask;
-		if (s > e || e - s < min)
+		if (s > e || e - s < size_min)
 			continue;
 
 		splitoff = s - this->offset;
-		if (splitoff && !region_split(mm, this, splitoff))
+		if (splitoff && !region_head(mm, this, splitoff))
 			return -ENOMEM;
 
-		this = region_split(mm, this, min(size, e - s));
+		this = region_head(mm, this, min(size_max, e - s));
+		if (!this)
+			return -ENOMEM;
+
+		this->type = type;
+		list_del(&this->fl_entry);
+		*pnode = this;
+		return 0;
+	}
+
+	return -ENOSPC;
+}
+
+static struct nouveau_mm_node *
+region_tail(struct nouveau_mm *mm, struct nouveau_mm_node *a, u32 size)
+{
+	struct nouveau_mm_node *b;
+
+	if (a->length == size)
+		return a;
+
+	b = kmalloc(sizeof(*b), GFP_KERNEL);
+	if (unlikely(b == NULL))
+		return NULL;
+
+	a->length -= size;
+	b->offset  = a->offset + a->length;
+	b->length  = size;
+	b->type    = a->type;
+
+	list_add(&b->nl_entry, &a->nl_entry);
+	if (b->type == 0)
+		list_add(&b->fl_entry, &a->fl_entry);
+	return b;
+}
+
+int
+nouveau_mm_tail(struct nouveau_mm *mm, u8 type, u32 size_max, u32 size_min,
+		u32 align, struct nouveau_mm_node **pnode)
+{
+	struct nouveau_mm_node *prev, *this, *next;
+	u32 mask = align - 1;
+
+	list_for_each_entry_reverse(this, &mm->free, fl_entry) {
+		u32 e = this->offset + this->length;
+		u32 s = this->offset;
+		u32 c = 0, a;
+
+		prev = node(this, prev);
+		if (prev && prev->type != type)
+			s = roundup(s, mm->block_size);
+
+		next = node(this, next);
+		if (next && next->type != type) {
+			e = rounddown(e, mm->block_size);
+			c = next->offset - e;
+		}
+
+		s = (s + mask) & ~mask;
+		a = e - s;
+		if (s > e || a < size_min)
+			continue;
+
+		a  = min(a, size_max);
+		s  = (e - a) & ~mask;
+		c += (e - s) - a;
+
+		if (c && !region_tail(mm, this, c))
+			return -ENOMEM;
+
+		this = region_tail(mm, this, a);
 		if (!this)
 			return -ENOMEM;
 
@@ -147,6 +224,7 @@ nouveau_mm_init(struct nouveau_mm *mm, u32 offset, u32 length, u32 block)
 	list_add_tail(&node->nl_entry, &mm->nodes);
 	list_add_tail(&node->fl_entry, &mm->free);
 	mm->heap_nodes++;
+	mm->heap_size += length;
 	return 0;
 }
 
@@ -158,15 +236,8 @@ nouveau_mm_fini(struct nouveau_mm *mm)
 	int nodes = 0;
 
 	list_for_each_entry(node, &mm->nodes, nl_entry) {
-		if (nodes++ == mm->heap_nodes) {
+		if (nodes++ == mm->heap_nodes)
-			printk(KERN_ERR "nouveau_mm in use at destroy time!\n");
-			list_for_each_entry(node, &mm->nodes, nl_entry) {
-				printk(KERN_ERR "0x%02x: 0x%08x 0x%08x\n",
-				       node->type, node->offset, node->length);
-			}
-			WARN_ON(1);
 			return -EBUSY;
-		}
 	}
 
 	kfree(heap);

drivers/gpu/drm/nouveau/core/include/core/mm.h

@@ -1,27 +1,3 @@
-/*
- * Copyright 2010 Red Hat Inc.
- *
- * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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.
- *
- * Authors: Ben Skeggs
- */
-
 #ifndef __NOUVEAU_MM_H__
 #define __NOUVEAU_MM_H__
 
@@ -43,13 +19,15 @@ struct nouveau_mm {
 
 	u32 block_size;
 	int heap_nodes;
+	u32 heap_size;
 };
 
 int  nouveau_mm_init(struct nouveau_mm *, u32 offset, u32 length, u32 block);
 int  nouveau_mm_fini(struct nouveau_mm *);
-int  nouveau_mm_pre(struct nouveau_mm *);
+int  nouveau_mm_head(struct nouveau_mm *, u8 type, u32 size_max, u32 size_min,
-int  nouveau_mm_get(struct nouveau_mm *, int type, u32 size, u32 size_nc,
+		     u32 align, struct nouveau_mm_node **);
-		    u32 align, struct nouveau_mm_node **);
+int  nouveau_mm_tail(struct nouveau_mm *, u8 type, u32 size_max, u32 size_min,
-void nouveau_mm_put(struct nouveau_mm *, struct nouveau_mm_node *);
+		     u32 align, struct nouveau_mm_node **);
+void nouveau_mm_free(struct nouveau_mm *, struct nouveau_mm_node **);
 
 #endif

drivers/gpu/drm/nouveau/core/subdev/fb/nv50_vram.c

@@ -65,7 +65,7 @@ nv50_vram_del(struct drm_device *dev, struct nouveau_mem **pmem)
 		this = list_first_entry(&mem->regions, struct nouveau_mm_node, rl_entry);
 
 		list_del(&this->rl_entry);
-		nouveau_mm_put(mm, this);
+		nouveau_mm_free(mm, &this);
 	}
 
 	if (mem->tag) {
@@ -78,7 +78,7 @@ nv50_vram_del(struct drm_device *dev, struct nouveau_mem **pmem)
 }
 
 int
-nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc,
+nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 ncmin,
 	      u32 memtype, struct nouveau_mem **pmem)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -87,13 +87,14 @@ nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc,
 	struct nouveau_mem *mem;
 	int comp = (memtype & 0x300) >> 8;
 	int type = (memtype & 0x07f);
+	int back = (memtype & 0x800);
 	int ret;
 
-	if (!types[type])
-		return -EINVAL;
 	size >>= 12;
 	align >>= 12;
-	size_nc >>= 12;
+	ncmin >>= 12;
+	if (!ncmin)
+		ncmin = size;
 
 	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 	if (!mem)
@@ -119,8 +120,13 @@ nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc,
 	mem->memtype = (comp << 7) | type;
 	mem->size = size;
 
+	type = types[type];
 	do {
-		ret = nouveau_mm_get(mm, types[type], size, size_nc, align, &r);
+		if (back)
+			ret = nouveau_mm_tail(mm, type, size, ncmin, align, &r);
+		else
+			ret = nouveau_mm_head(mm, type, size, ncmin, align, &r);
+
 		if (ret) {
 			mutex_unlock(&mm->mutex);
 			nv50_vram_del(dev, &mem);

drivers/gpu/drm/nouveau/core/subdev/fb/nvc0_vram.c

@@ -58,17 +58,21 @@ nvc0_vram_flags_valid(struct drm_device *dev, u32 tile_flags)
 
 int
 nvc0_vram_new(struct drm_device *dev, u64 size, u32 align, u32 ncmin,
-	      u32 type, struct nouveau_mem **pmem)
+	      u32 memtype, struct nouveau_mem **pmem)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_mm *mm = &dev_priv->engine.vram.mm;
 	struct nouveau_mm_node *r;
 	struct nouveau_mem *mem;
+	int type = (memtype & 0x0ff);
+	int back = (memtype & 0x800);
 	int ret;
 
 	size  >>= 12;
 	align >>= 12;
 	ncmin >>= 12;
+	if (!ncmin)
+		ncmin = size;
 
 	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 	if (!mem)
@@ -76,12 +80,15 @@ nvc0_vram_new(struct drm_device *dev, u64 size, u32 align, u32 ncmin,
 
 	INIT_LIST_HEAD(&mem->regions);
 	mem->dev = dev_priv->dev;
-	mem->memtype = (type & 0xff);
+	mem->memtype = type;
 	mem->size = size;
 
 	mutex_lock(&mm->mutex);
 	do {
-		ret = nouveau_mm_get(mm, 1, size, ncmin, align, &r);
+		if (back)
+			ret = nouveau_mm_tail(mm, 1, size, ncmin, align, &r);
+		else
+			ret = nouveau_mm_head(mm, 1, size, ncmin, align, &r);
 		if (ret) {
 			mutex_unlock(&mm->mutex);
 			nv50_vram_del(dev, &mem);

drivers/gpu/drm/nouveau/core/subdev/vm/base.c

@@ -282,7 +282,8 @@ nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
 	int ret;
 
 	mutex_lock(&vm->mm.mutex);
-	ret = nouveau_mm_get(&vm->mm, page_shift, msize, 0, align, &vma->node);
+	ret = nouveau_mm_head(&vm->mm, page_shift, msize, msize, align,
+			     &vma->node);
 	if (unlikely(ret != 0)) {
 		mutex_unlock(&vm->mm.mutex);
 		return ret;
@@ -303,9 +304,8 @@ nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
 		if (ret) {
 			if (pde != fpde)
 				nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
-			nouveau_mm_put(&vm->mm, vma->node);
+			nouveau_mm_free(&vm->mm, &vma->node);
 			mutex_unlock(&vm->mm.mutex);
-			vma->node = NULL;
 			return ret;
 		}
 	}
@@ -330,8 +330,7 @@ nouveau_vm_put(struct nouveau_vma *vma)
 
 	mutex_lock(&vm->mm.mutex);
 	nouveau_vm_unmap_pgt(vm, vma->node->type != vm->spg_shift, fpde, lpde);
-	nouveau_mm_put(&vm->mm, vma->node);
+	nouveau_mm_free(&vm->mm, &vma->node);
-	vma->node = NULL;
 	mutex_unlock(&vm->mm.mutex);
 }