drm/i915: buddy allocator

Simple buddy allocator. We want to allocate properly aligned
power-of-two blocks to promote usage of huge-pages for the GTT, so 64K,
2M and possibly even 1G. While we do support allocating stuff at a
specific offset, it is more intended for preallocating portions of the
address space, say for an initial framebuffer, for other uses drm_mm is
probably a much better fit. Anyway, hopefully this can all be thrown
away if we eventually move to having the core MM manage device memory.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190809202926.14545-2-matthew.auld@intel.com

drivers/gpu/drm/i915/Makefile

@@ -127,6 +127,7 @@ gem-y += \
 i915-y += \
 	  $(gem-y) \
 	  i915_active.o \
+	  i915_buddy.o \
 	  i915_cmd_parser.o \
 	  i915_gem_evict.o \
 	  i915_gem_fence_reg.o \

drivers/gpu/drm/i915/i915_buddy.c

+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2019 Intel Corporation
+ */
+
+#include <linux/slab.h>
+
+#include "i915_buddy.h"
+
+#include "i915_gem.h"
+#include "i915_globals.h"
+#include "i915_utils.h"
+
+static struct i915_global_block {
+	struct i915_global base;
+	struct kmem_cache *slab_blocks;
+} global;
+
+static void i915_global_buddy_shrink(void)
+{
+	kmem_cache_shrink(global.slab_blocks);
+}
+
+static void i915_global_buddy_exit(void)
+{
+	kmem_cache_destroy(global.slab_blocks);
+}
+
+static struct i915_global_block global = { {
+	.shrink = i915_global_buddy_shrink,
+	.exit = i915_global_buddy_exit,
+} };
+
+int __init i915_global_buddy_init(void)
+{
+	global.slab_blocks = KMEM_CACHE(i915_buddy_block, SLAB_HWCACHE_ALIGN);
+	if (!global.slab_blocks)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static struct i915_buddy_block *i915_block_alloc(struct i915_buddy_block *parent,
+						 unsigned int order,
+						 u64 offset)
+{
+	struct i915_buddy_block *block;
+
+	block = kmem_cache_zalloc(global.slab_blocks, GFP_KERNEL);
+	if (!block)
+		return NULL;
+
+	block->header = offset;
+	block->header |= order;
+	block->parent = parent;
+
+	return block;
+}
+
+static void i915_block_free(struct i915_buddy_block *block)
+{
+	kmem_cache_free(global.slab_blocks, block);
+}
+
+static void mark_allocated(struct i915_buddy_block *block)
+{
+	block->header &= ~I915_BUDDY_HEADER_STATE;
+	block->header |= I915_BUDDY_ALLOCATED;
+
+	list_del(&block->link);
+}
+
+static void mark_free(struct i915_buddy_mm *mm,
+		      struct i915_buddy_block *block)
+{
+	block->header &= ~I915_BUDDY_HEADER_STATE;
+	block->header |= I915_BUDDY_FREE;
+
+	list_add(&block->link,
+		 &mm->free_list[i915_buddy_block_order(block)]);
+}
+
+static void mark_split(struct i915_buddy_block *block)
+{
+	block->header &= ~I915_BUDDY_HEADER_STATE;
+	block->header |= I915_BUDDY_SPLIT;
+
+	list_del(&block->link);
+}
+
+int i915_buddy_init(struct i915_buddy_mm *mm, u64 size, u64 chunk_size)
+{
+	unsigned int i;
+	u64 offset;
+
+	if (size < chunk_size)
+		return -EINVAL;
+
+	if (chunk_size < PAGE_SIZE)
+		return -EINVAL;
+
+	if (!is_power_of_2(chunk_size))
+		return -EINVAL;
+
+	size = round_down(size, chunk_size);
+
+	mm->size = size;
+	mm->chunk_size = chunk_size;
+	mm->max_order = ilog2(size) - ilog2(chunk_size);
+
+	GEM_BUG_ON(mm->max_order > I915_BUDDY_MAX_ORDER);
+
+	mm->free_list = kmalloc_array(mm->max_order + 1,
+				      sizeof(struct list_head),
+				      GFP_KERNEL);
+	if (!mm->free_list)
+		return -ENOMEM;
+
+	for (i = 0; i <= mm->max_order; ++i)
+		INIT_LIST_HEAD(&mm->free_list[i]);
+
+	mm->n_roots = hweight64(size);
+
+	mm->roots = kmalloc_array(mm->n_roots,
+				  sizeof(struct i915_buddy_block *),
+				  GFP_KERNEL);
+	if (!mm->roots)
+		goto out_free_list;
+
+	offset = 0;
+	i = 0;
+
+	/*
+	 * Split into power-of-two blocks, in case we are given a size that is
+	 * not itself a power-of-two.
+	 */
+	do {
+		struct i915_buddy_block *root;
+		unsigned int order;
+		u64 root_size;
+
+		root_size = rounddown_pow_of_two(size);
+		order = ilog2(root_size) - ilog2(chunk_size);
+
+		root = i915_block_alloc(NULL, order, offset);
+		if (!root)
+			goto out_free_roots;
+
+		mark_free(mm, root);
+
+		GEM_BUG_ON(i > mm->max_order);
+		GEM_BUG_ON(i915_buddy_block_size(mm, root) < chunk_size);
+
+		mm->roots[i] = root;
+
+		offset += root_size;
+		size -= root_size;
+		i++;
+	} while (size);
+
+	return 0;
+
+out_free_roots:
+	while (i--)
+		i915_block_free(mm->roots[i]);
+	kfree(mm->roots);
+out_free_list:
+	kfree(mm->free_list);
+	return -ENOMEM;
+}
+
+void i915_buddy_fini(struct i915_buddy_mm *mm)
+{
+	int err = 0;
+	int i;
+
+	for (i = 0; i < mm->n_roots; ++i) {
+		if (!i915_buddy_block_is_free(mm->roots[i])) {
+			err = -EBUSY;
+			continue;
+		}
+
+		i915_block_free(mm->roots[i]);
+	}
+
+	kfree(mm->roots);
+	kfree(mm->free_list);
+}
+
+static int split_block(struct i915_buddy_mm *mm,
+		       struct i915_buddy_block *block)
+{
+	unsigned int block_order = i915_buddy_block_order(block) - 1;
+	u64 offset = i915_buddy_block_offset(block);
+
+	GEM_BUG_ON(!i915_buddy_block_is_free(block));
+	GEM_BUG_ON(!i915_buddy_block_order(block));
+
+	block->left = i915_block_alloc(block, block_order, offset);
+	if (!block->left)
+		return -ENOMEM;
+
+	block->right = i915_block_alloc(block, block_order,
+					offset + (mm->chunk_size << block_order));
+	if (!block->right) {
+		i915_block_free(block->left);
+		return -ENOMEM;
+	}
+
+	mark_free(mm, block->left);
+	mark_free(mm, block->right);
+
+	mark_split(block);
+
+	return 0;
+}
+
+static struct i915_buddy_block *
+get_buddy(struct i915_buddy_block *block)
+{
+	struct i915_buddy_block *parent;
+
+	parent = block->parent;
+	if (!parent)
+		return NULL;
+
+	if (parent->left == block)
+		return parent->right;
+
+	return parent->left;
+}
+
+static void __i915_buddy_free(struct i915_buddy_mm *mm,
+			      struct i915_buddy_block *block)
+{
+	struct i915_buddy_block *parent;
+
+	while ((parent = block->parent)) {
+		struct i915_buddy_block *buddy;
+
+		buddy = get_buddy(block);
+
+		if (!i915_buddy_block_is_free(buddy))
+			break;
+
+		list_del(&buddy->link);
+
+		i915_block_free(block);
+		i915_block_free(buddy);
+
+		block = parent;
+	}
+
+	mark_free(mm, block);
+}
+
+void i915_buddy_free(struct i915_buddy_mm *mm,
+		     struct i915_buddy_block *block)
+{
+	GEM_BUG_ON(!i915_buddy_block_is_allocated(block));
+	__i915_buddy_free(mm, block);
+}
+
+void i915_buddy_free_list(struct i915_buddy_mm *mm, struct list_head *objects)
+{
+	struct i915_buddy_block *block, *on;
+
+	list_for_each_entry_safe(block, on, objects, link)
+		i915_buddy_free(mm, block);
+	INIT_LIST_HEAD(objects);
+}
+
+/*
+ * Allocate power-of-two block. The order value here translates to:
+ *
+ *   0 = 2^0 * mm->chunk_size
+ *   1 = 2^1 * mm->chunk_size
+ *   2 = 2^2 * mm->chunk_size
+ *   ...
+ */
+struct i915_buddy_block *
+i915_buddy_alloc(struct i915_buddy_mm *mm, unsigned int order)
+{
+	struct i915_buddy_block *block = NULL;
+	unsigned int i;
+	int err;
+
+	for (i = order; i <= mm->max_order; ++i) {
+		block = list_first_entry_or_null(&mm->free_list[i],
+						 struct i915_buddy_block,
+						 link);
+		if (block)
+			break;
+	}
+
+	if (!block)
+		return ERR_PTR(-ENOSPC);
+
+	GEM_BUG_ON(!i915_buddy_block_is_free(block));
+
+	while (i != order) {
+		err = split_block(mm, block);
+		if (unlikely(err))
+			goto out_free;
+
+		/* Go low */
+		block = block->left;
+		i--;
+	}
+
+	mark_allocated(block);
+	return block;
+
+out_free:
+	__i915_buddy_free(mm, block);
+	return ERR_PTR(err);
+}
+
+static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
+{
+	return s1 <= e2 && e1 >= s2;
+}
+
+static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
+{
+	return s1 <= s2 && e1 >= e2;
+}
+
+/*
+ * Allocate range. Note that it's safe to chain together multiple alloc_ranges
+ * with the same blocks list.
+ *
+ * Intended for pre-allocating portions of the address space, for example to
+ * reserve a block for the initial framebuffer or similar, hence the expectation
+ * here is that i915_buddy_alloc() is still the main vehicle for
+ * allocations, so if that's not the case then the drm_mm range allocator is
+ * probably a much better fit, and so you should probably go use that instead.
+ */
+int i915_buddy_alloc_range(struct i915_buddy_mm *mm,
+			   struct list_head *blocks,
+			   u64 start, u64 size)
+{
+	struct i915_buddy_block *block;
+	struct i915_buddy_block *buddy;
+	LIST_HEAD(allocated);
+	LIST_HEAD(dfs);
+	u64 end;
+	int err;
+	int i;
+
+	if (size < mm->chunk_size)
+		return -EINVAL;
+
+	if (!IS_ALIGNED(size | start, mm->chunk_size))
+		return -EINVAL;
+
+	if (range_overflows(start, size, mm->size))
+		return -EINVAL;
+
+	for (i = 0; i < mm->n_roots; ++i)
+		list_add_tail(&mm->roots[i]->tmp_link, &dfs);
+
+	end = start + size - 1;
+
+	do {
+		u64 block_start;
+		u64 block_end;
+
+		block = list_first_entry_or_null(&dfs,
+						 struct i915_buddy_block,
+						 tmp_link);
+		if (!block)
+			break;
+
+		list_del(&block->tmp_link);
+
+		block_start = i915_buddy_block_offset(block);
+		block_end = block_start + i915_buddy_block_size(mm, block) - 1;
+
+		if (!overlaps(start, end, block_start, block_end))
+			continue;
+
+		if (i915_buddy_block_is_allocated(block)) {
+			err = -ENOSPC;
+			goto err_free;
+		}
+
+		if (contains(start, end, block_start, block_end)) {
+			if (!i915_buddy_block_is_free(block)) {
+				err = -ENOSPC;
+				goto err_free;
+			}
+
+			mark_allocated(block);
+			list_add_tail(&block->link, &allocated);
+			continue;
+		}
+
+		if (!i915_buddy_block_is_split(block)) {
+			err = split_block(mm, block);
+			if (unlikely(err))
+				goto err_undo;
+		}
+
+		list_add(&block->right->tmp_link, &dfs);
+		list_add(&block->left->tmp_link, &dfs);
+	} while (1);
+
+	list_splice_tail(&allocated, blocks);
+	return 0;
+
+err_undo:
+	/*
+	 * We really don't want to leave around a bunch of split blocks, since
+	 * bigger is better, so make sure we merge everything back before we
+	 * free the allocated blocks.
+	 */
+	buddy = get_buddy(block);
+	if (buddy &&
+	    (i915_buddy_block_is_free(block) &&
+	     i915_buddy_block_is_free(buddy)))
+		__i915_buddy_free(mm, block);
+
+err_free:
+	i915_buddy_free_list(mm, &allocated);
+	return err;
+}
+
+#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
+#include "selftests/i915_buddy.c"
+#endif

drivers/gpu/drm/i915/i915_buddy.h

+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2019 Intel Corporation
+ */
+
+#ifndef __I915_BUDDY_H__
+#define __I915_BUDDY_H__
+
+#include <linux/bitops.h>
+#include <linux/list.h>
+
+struct i915_buddy_block {
+#define I915_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
+#define I915_BUDDY_HEADER_STATE  GENMASK_ULL(11, 10)
+#define   I915_BUDDY_ALLOCATED	   (1 << 10)
+#define   I915_BUDDY_FREE	   (2 << 10)
+#define   I915_BUDDY_SPLIT	   (3 << 10)
+#define I915_BUDDY_HEADER_ORDER  GENMASK_ULL(9, 0)
+	u64 header;
+
+	struct i915_buddy_block *left;
+	struct i915_buddy_block *right;
+	struct i915_buddy_block *parent;
+
+	void *private; /* owned by creator */
+
+	/*
+	 * While the block is allocated by the user through i915_buddy_alloc*,
+	 * the user has ownership of the link, for example to maintain within
+	 * a list, if so desired. As soon as the block is freed with
+	 * i915_buddy_free* ownership is given back to the mm.
+	 */
+	struct list_head link;
+	struct list_head tmp_link;
+};
+
+#define I915_BUDDY_MAX_ORDER  I915_BUDDY_HEADER_ORDER
+
+/*
+ * Binary Buddy System.
+ *
+ * Locking should be handled by the user, a simple mutex around
+ * i915_buddy_alloc* and i915_buddy_free* should suffice.
+ */
+struct i915_buddy_mm {
+	/* Maintain a free list for each order. */
+	struct list_head *free_list;
+
+	/*
+	 * Maintain explicit binary tree(s) to track the allocation of the
+	 * address space. This gives us a simple way of finding a buddy block
+	 * and performing the potentially recursive merge step when freeing a
+	 * block.  Nodes are either allocated or free, in which case they will
+	 * also exist on the respective free list.
+	 */
+	struct i915_buddy_block **roots;
+
+	/*
+	 * Anything from here is public, and remains static for the lifetime of
+	 * the mm. Everything above is considered do-not-touch.
+	 */
+	unsigned int n_roots;
+	unsigned int max_order;
+
+	/* Must be at least PAGE_SIZE */
+	u64 chunk_size;
+	u64 size;
+};
+
+static inline u64
+i915_buddy_block_offset(struct i915_buddy_block *block)
+{
+	return block->header & I915_BUDDY_HEADER_OFFSET;
+}
+
+static inline unsigned int
+i915_buddy_block_order(struct i915_buddy_block *block)
+{
+	return block->header & I915_BUDDY_HEADER_ORDER;
+}
+
+static inline unsigned int
+i915_buddy_block_state(struct i915_buddy_block *block)
+{
+	return block->header & I915_BUDDY_HEADER_STATE;
+}
+
+static inline bool
+i915_buddy_block_is_allocated(struct i915_buddy_block *block)
+{
+	return i915_buddy_block_state(block) == I915_BUDDY_ALLOCATED;
+}
+
+static inline bool
+i915_buddy_block_is_free(struct i915_buddy_block *block)
+{
+	return i915_buddy_block_state(block) == I915_BUDDY_FREE;
+}
+
+static inline bool
+i915_buddy_block_is_split(struct i915_buddy_block *block)
+{
+	return i915_buddy_block_state(block) == I915_BUDDY_SPLIT;
+}
+
+static inline u64
+i915_buddy_block_size(struct i915_buddy_mm *mm,
+		      struct i915_buddy_block *block)
+{
+	return mm->chunk_size << i915_buddy_block_order(block);
+}
+
+int i915_buddy_init(struct i915_buddy_mm *mm, u64 size, u64 chunk_size);
+
+void i915_buddy_fini(struct i915_buddy_mm *mm);
+
+struct i915_buddy_block *
+i915_buddy_alloc(struct i915_buddy_mm *mm, unsigned int order);
+
+int i915_buddy_alloc_range(struct i915_buddy_mm *mm,
+			   struct list_head *blocks,
+			   u64 start, u64 size);
+
+void i915_buddy_free(struct i915_buddy_mm *mm, struct i915_buddy_block *block);
+
+void i915_buddy_free_list(struct i915_buddy_mm *mm, struct list_head *objects);
+
+#endif

drivers/gpu/drm/i915/i915_globals.c

@@ -62,6 +62,7 @@ static void __i915_globals_cleanup(void)
 
 static __initconst int (* const initfn[])(void) = {
 	i915_global_active_init,
+	i915_global_buddy_init,
 	i915_global_context_init,
 	i915_global_gem_context_init,
 	i915_global_objects_init,

drivers/gpu/drm/i915/i915_globals.h

@@ -27,6 +27,7 @@ void i915_globals_exit(void);
 
 /* constructors */
 int i915_global_active_init(void);
+int i915_global_buddy_init(void);
 int i915_global_context_init(void);
 int i915_global_gem_context_init(void);
 int i915_global_objects_init(void);

drivers/gpu/drm/i915/selftests/i915_mock_selftests.h

@@ -25,3 +25,4 @@ selftest(evict, i915_gem_evict_mock_selftests)
 selftest(gtt, i915_gem_gtt_mock_selftests)
 selftest(hugepages, i915_gem_huge_page_mock_selftests)
 selftest(contexts, i915_gem_context_mock_selftests)
+selftest(buddy, i915_buddy_mock_selftests)