drm/nv50: implement custom vram mm

This is required on nv50 as we need to be able to have more precise control
over physical VRAM allocations to avoid buffer corruption when using
buffers of mixed memory types.

This removes some nasty overallocation/alignment that we were previously
using to "control" this problem.

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

1 files changed, 271 insertions, 0 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_mm.c b/drivers/gpu/drm/nouveau/nouveau_mm.c
new file mode 100644
index 000000000000..cdbb11eb701b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_mm.c
@@ -0,0 +1,271 @@
+/*
+ * 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
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_mm.h"
+
+static inline void
+region_put(struct nouveau_mm *rmm, struct nouveau_mm_node *a)
+{
+	list_del(&a->nl_entry);
+	list_del(&a->fl_entry);
+	kfree(a);
+}
+
+static struct nouveau_mm_node *
+region_split(struct nouveau_mm *rmm, 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;
+
+	b->offset = a->offset;
+	b->length = size;
+	b->free   = a->free;
+	b->type   = a->type;
+	a->offset += size;
+	a->length -= size;
+	list_add_tail(&b->nl_entry, &a->nl_entry);
+	if (b->free)
+		list_add_tail(&b->fl_entry, &a->fl_entry);
+	return b;
+}
+
+static struct nouveau_mm_node *
+nouveau_mm_merge(struct nouveau_mm *rmm, struct nouveau_mm_node *this)
+{
+	struct nouveau_mm_node *prev, *next;
+
+	/* try to merge with free adjacent entries of same type */
+	prev = list_entry(this->nl_entry.prev, struct nouveau_mm_node, nl_entry);
+	if (this->nl_entry.prev != &rmm->nodes) {
+		if (prev->free && prev->type == this->type) {
+			prev->length += this->length;
+			region_put(rmm, this);
+			this = prev;
+		}
+	}
+
+	next = list_entry(this->nl_entry.next, struct nouveau_mm_node, nl_entry);
+	if (this->nl_entry.next != &rmm->nodes) {
+		if (next->free && next->type == this->type) {
+			next->offset  = this->offset;
+			next->length += this->length;
+			region_put(rmm, this);
+			this = next;
+		}
+	}
+
+	return this;
+}
+
+void
+nouveau_mm_put(struct nouveau_mm *rmm, struct nouveau_mm_node *this)
+{
+	u32 block_s, block_l;
+
+	this->free = true;
+	list_add(&this->fl_entry, &rmm->free);
+	this = nouveau_mm_merge(rmm, this);
+
+	/* any entirely free blocks now?  we'll want to remove typing
+	 * on them now so they can be use for any memory allocation
+	 */
+	block_s = roundup(this->offset, rmm->block_size);
+	if (block_s + rmm->block_size > this->offset + this->length)
+		return;
+
+	/* split off any still-typed region at the start */
+	if (block_s != this->offset) {
+		if (!region_split(rmm, this, block_s - this->offset))
+			return;
+	}
+
+	/* split off the soon-to-be-untyped block(s) */
+	block_l = rounddown(this->length, rmm->block_size);
+	if (block_l != this->length) {
+		this = region_split(rmm, this, block_l);
+		if (!this)
+			return;
+	}
+
+	/* mark as having no type, and retry merge with any adjacent
+	 * untyped blocks
+	 */
+	this->type = 0;
+	nouveau_mm_merge(rmm, this);
+}
+
+int
+nouveau_mm_get(struct nouveau_mm *rmm, int type, u32 size, u32 size_nc,
+	       u32 align, struct nouveau_mm_node **pnode)
+{
+	struct nouveau_mm_node *this, *tmp, *next;
+	u32 splitoff, avail, alloc;
+
+	list_for_each_entry_safe(this, tmp, &rmm->free, fl_entry) {
+		next = list_entry(this->nl_entry.next, struct nouveau_mm_node, nl_entry);
+		if (this->nl_entry.next == &rmm->nodes)
+			next = NULL;
+
+		/* skip wrongly typed blocks */
+		if (this->type && this->type != type)
+			continue;
+
+		/* account for alignment */
+		splitoff = this->offset & (align - 1);
+		if (splitoff)
+			splitoff = align - splitoff;
+
+		if (this->length <= splitoff)
+			continue;
+
+		/* determine total memory available from this, and
+		 * the next block (if appropriate)
+		 */
+		avail = this->length;
+		if (next && next->free && (!next->type || next->type == type))
+			avail += next->length;
+
+		avail -= splitoff;
+
+		/* determine allocation size */
+		if (size_nc) {
+			alloc = min(avail, size);
+			alloc = rounddown(alloc, size_nc);
+			if (alloc == 0)
+				continue;
+		} else {
+			alloc = size;
+			if (avail < alloc)
+				continue;
+		}
+
+		/* untyped block, split off a chunk that's a multiple
+		 * of block_size and type it
+		 */
+		if (!this->type) {
+			u32 block = roundup(alloc + splitoff, rmm->block_size);
+			if (this->length < block)
+				continue;
+
+			this = region_split(rmm, this, block);
+			if (!this)
+				return -ENOMEM;
+
+			this->type = type;
+		}
+
+		/* stealing memory from adjacent block */
+		if (alloc > this->length) {
+			u32 amount = alloc - (this->length - splitoff);
+
+			if (!next->type) {
+				amount = roundup(amount, rmm->block_size);
+
+				next = region_split(rmm, next, amount);
+				if (!next)
+					return -ENOMEM;
+
+				next->type = type;
+			}
+
+			this->length += amount;
+			next->offset += amount;
+			next->length -= amount;
+			if (!next->length) {
+				list_del(&next->nl_entry);
+				list_del(&next->fl_entry);
+				kfree(next);
+			}
+		}
+
+		if (splitoff) {
+			if (!region_split(rmm, this, splitoff))
+				return -ENOMEM;
+		}
+
+		this = region_split(rmm, this, alloc);
+		if (this == NULL)
+			return -ENOMEM;
+
+		this->free = false;
+		list_del(&this->fl_entry);
+		*pnode = this;
+		return 0;
+	}
+
+	return -ENOMEM;
+}
+
+int
+nouveau_mm_init(struct nouveau_mm **prmm, u32 offset, u32 length, u32 block)
+{
+	struct nouveau_mm *rmm;
+	struct nouveau_mm_node *heap;
+
+	heap = kzalloc(sizeof(*heap), GFP_KERNEL);
+	if (!heap)
+		return -ENOMEM;
+	heap->free = true;
+	heap->offset = roundup(offset, block);
+	heap->length = rounddown(offset + length, block) - heap->offset;
+
+	rmm = kzalloc(sizeof(*rmm), GFP_KERNEL);
+	if (!rmm) {
+		kfree(heap);
+		return -ENOMEM;
+	}
+	rmm->block_size = block;
+	mutex_init(&rmm->mutex);
+	INIT_LIST_HEAD(&rmm->nodes);
+	INIT_LIST_HEAD(&rmm->free);
+	list_add(&heap->nl_entry, &rmm->nodes);
+	list_add(&heap->fl_entry, &rmm->free);
+
+	*prmm = rmm;
+	return 0;
+}
+
+int
+nouveau_mm_fini(struct nouveau_mm **prmm)
+{
+	struct nouveau_mm *rmm = *prmm;
+	struct nouveau_mm_node *heap =
+		list_first_entry(&rmm->nodes, struct nouveau_mm_node, nl_entry);
+
+	if (!list_is_singular(&rmm->nodes))
+		return -EBUSY;
+
+	kfree(heap);
+	kfree(rmm);
+	*prmm = NULL;
+	return 0;
+}