xref: /openbmc/linux/drivers/gpu/drm/drm_buddy.c (revision ecefa105)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2021 Intel Corporation
4  */
5 
6 #include <linux/kmemleak.h>
7 #include <linux/module.h>
8 #include <linux/sizes.h>
9 
10 #include <drm/drm_buddy.h>
11 
12 static struct kmem_cache *slab_blocks;
13 
14 static struct drm_buddy_block *drm_block_alloc(struct drm_buddy *mm,
15 					       struct drm_buddy_block *parent,
16 					       unsigned int order,
17 					       u64 offset)
18 {
19 	struct drm_buddy_block *block;
20 
21 	BUG_ON(order > DRM_BUDDY_MAX_ORDER);
22 
23 	block = kmem_cache_zalloc(slab_blocks, GFP_KERNEL);
24 	if (!block)
25 		return NULL;
26 
27 	block->header = offset;
28 	block->header |= order;
29 	block->parent = parent;
30 
31 	BUG_ON(block->header & DRM_BUDDY_HEADER_UNUSED);
32 	return block;
33 }
34 
35 static void drm_block_free(struct drm_buddy *mm,
36 			   struct drm_buddy_block *block)
37 {
38 	kmem_cache_free(slab_blocks, block);
39 }
40 
41 static void list_insert_sorted(struct drm_buddy *mm,
42 			       struct drm_buddy_block *block)
43 {
44 	struct drm_buddy_block *node;
45 	struct list_head *head;
46 
47 	head = &mm->free_list[drm_buddy_block_order(block)];
48 	if (list_empty(head)) {
49 		list_add(&block->link, head);
50 		return;
51 	}
52 
53 	list_for_each_entry(node, head, link)
54 		if (drm_buddy_block_offset(block) < drm_buddy_block_offset(node))
55 			break;
56 
57 	__list_add(&block->link, node->link.prev, &node->link);
58 }
59 
60 static void mark_allocated(struct drm_buddy_block *block)
61 {
62 	block->header &= ~DRM_BUDDY_HEADER_STATE;
63 	block->header |= DRM_BUDDY_ALLOCATED;
64 
65 	list_del(&block->link);
66 }
67 
68 static void mark_free(struct drm_buddy *mm,
69 		      struct drm_buddy_block *block)
70 {
71 	block->header &= ~DRM_BUDDY_HEADER_STATE;
72 	block->header |= DRM_BUDDY_FREE;
73 
74 	list_insert_sorted(mm, block);
75 }
76 
77 static void mark_split(struct drm_buddy_block *block)
78 {
79 	block->header &= ~DRM_BUDDY_HEADER_STATE;
80 	block->header |= DRM_BUDDY_SPLIT;
81 
82 	list_del(&block->link);
83 }
84 
85 /**
86  * drm_buddy_init - init memory manager
87  *
88  * @mm: DRM buddy manager to initialize
89  * @size: size in bytes to manage
90  * @chunk_size: minimum page size in bytes for our allocations
91  *
92  * Initializes the memory manager and its resources.
93  *
94  * Returns:
95  * 0 on success, error code on failure.
96  */
97 int drm_buddy_init(struct drm_buddy *mm, u64 size, u64 chunk_size)
98 {
99 	unsigned int i;
100 	u64 offset;
101 
102 	if (size < chunk_size)
103 		return -EINVAL;
104 
105 	if (chunk_size < PAGE_SIZE)
106 		return -EINVAL;
107 
108 	if (!is_power_of_2(chunk_size))
109 		return -EINVAL;
110 
111 	size = round_down(size, chunk_size);
112 
113 	mm->size = size;
114 	mm->avail = size;
115 	mm->chunk_size = chunk_size;
116 	mm->max_order = ilog2(size) - ilog2(chunk_size);
117 
118 	BUG_ON(mm->max_order > DRM_BUDDY_MAX_ORDER);
119 
120 	mm->free_list = kmalloc_array(mm->max_order + 1,
121 				      sizeof(struct list_head),
122 				      GFP_KERNEL);
123 	if (!mm->free_list)
124 		return -ENOMEM;
125 
126 	for (i = 0; i <= mm->max_order; ++i)
127 		INIT_LIST_HEAD(&mm->free_list[i]);
128 
129 	mm->n_roots = hweight64(size);
130 
131 	mm->roots = kmalloc_array(mm->n_roots,
132 				  sizeof(struct drm_buddy_block *),
133 				  GFP_KERNEL);
134 	if (!mm->roots)
135 		goto out_free_list;
136 
137 	offset = 0;
138 	i = 0;
139 
140 	/*
141 	 * Split into power-of-two blocks, in case we are given a size that is
142 	 * not itself a power-of-two.
143 	 */
144 	do {
145 		struct drm_buddy_block *root;
146 		unsigned int order;
147 		u64 root_size;
148 
149 		root_size = rounddown_pow_of_two(size);
150 		order = ilog2(root_size) - ilog2(chunk_size);
151 
152 		root = drm_block_alloc(mm, NULL, order, offset);
153 		if (!root)
154 			goto out_free_roots;
155 
156 		mark_free(mm, root);
157 
158 		BUG_ON(i > mm->max_order);
159 		BUG_ON(drm_buddy_block_size(mm, root) < chunk_size);
160 
161 		mm->roots[i] = root;
162 
163 		offset += root_size;
164 		size -= root_size;
165 		i++;
166 	} while (size);
167 
168 	return 0;
169 
170 out_free_roots:
171 	while (i--)
172 		drm_block_free(mm, mm->roots[i]);
173 	kfree(mm->roots);
174 out_free_list:
175 	kfree(mm->free_list);
176 	return -ENOMEM;
177 }
178 EXPORT_SYMBOL(drm_buddy_init);
179 
180 /**
181  * drm_buddy_fini - tear down the memory manager
182  *
183  * @mm: DRM buddy manager to free
184  *
185  * Cleanup memory manager resources and the freelist
186  */
187 void drm_buddy_fini(struct drm_buddy *mm)
188 {
189 	int i;
190 
191 	for (i = 0; i < mm->n_roots; ++i) {
192 		WARN_ON(!drm_buddy_block_is_free(mm->roots[i]));
193 		drm_block_free(mm, mm->roots[i]);
194 	}
195 
196 	WARN_ON(mm->avail != mm->size);
197 
198 	kfree(mm->roots);
199 	kfree(mm->free_list);
200 }
201 EXPORT_SYMBOL(drm_buddy_fini);
202 
203 static int split_block(struct drm_buddy *mm,
204 		       struct drm_buddy_block *block)
205 {
206 	unsigned int block_order = drm_buddy_block_order(block) - 1;
207 	u64 offset = drm_buddy_block_offset(block);
208 
209 	BUG_ON(!drm_buddy_block_is_free(block));
210 	BUG_ON(!drm_buddy_block_order(block));
211 
212 	block->left = drm_block_alloc(mm, block, block_order, offset);
213 	if (!block->left)
214 		return -ENOMEM;
215 
216 	block->right = drm_block_alloc(mm, block, block_order,
217 				       offset + (mm->chunk_size << block_order));
218 	if (!block->right) {
219 		drm_block_free(mm, block->left);
220 		return -ENOMEM;
221 	}
222 
223 	mark_free(mm, block->left);
224 	mark_free(mm, block->right);
225 
226 	mark_split(block);
227 
228 	return 0;
229 }
230 
231 static struct drm_buddy_block *
232 __get_buddy(struct drm_buddy_block *block)
233 {
234 	struct drm_buddy_block *parent;
235 
236 	parent = block->parent;
237 	if (!parent)
238 		return NULL;
239 
240 	if (parent->left == block)
241 		return parent->right;
242 
243 	return parent->left;
244 }
245 
246 /**
247  * drm_get_buddy - get buddy address
248  *
249  * @block: DRM buddy block
250  *
251  * Returns the corresponding buddy block for @block, or NULL
252  * if this is a root block and can't be merged further.
253  * Requires some kind of locking to protect against
254  * any concurrent allocate and free operations.
255  */
256 struct drm_buddy_block *
257 drm_get_buddy(struct drm_buddy_block *block)
258 {
259 	return __get_buddy(block);
260 }
261 EXPORT_SYMBOL(drm_get_buddy);
262 
263 static void __drm_buddy_free(struct drm_buddy *mm,
264 			     struct drm_buddy_block *block)
265 {
266 	struct drm_buddy_block *parent;
267 
268 	while ((parent = block->parent)) {
269 		struct drm_buddy_block *buddy;
270 
271 		buddy = __get_buddy(block);
272 
273 		if (!drm_buddy_block_is_free(buddy))
274 			break;
275 
276 		list_del(&buddy->link);
277 
278 		drm_block_free(mm, block);
279 		drm_block_free(mm, buddy);
280 
281 		block = parent;
282 	}
283 
284 	mark_free(mm, block);
285 }
286 
287 /**
288  * drm_buddy_free_block - free a block
289  *
290  * @mm: DRM buddy manager
291  * @block: block to be freed
292  */
293 void drm_buddy_free_block(struct drm_buddy *mm,
294 			  struct drm_buddy_block *block)
295 {
296 	BUG_ON(!drm_buddy_block_is_allocated(block));
297 	mm->avail += drm_buddy_block_size(mm, block);
298 	__drm_buddy_free(mm, block);
299 }
300 EXPORT_SYMBOL(drm_buddy_free_block);
301 
302 /**
303  * drm_buddy_free_list - free blocks
304  *
305  * @mm: DRM buddy manager
306  * @objects: input list head to free blocks
307  */
308 void drm_buddy_free_list(struct drm_buddy *mm, struct list_head *objects)
309 {
310 	struct drm_buddy_block *block, *on;
311 
312 	list_for_each_entry_safe(block, on, objects, link) {
313 		drm_buddy_free_block(mm, block);
314 		cond_resched();
315 	}
316 	INIT_LIST_HEAD(objects);
317 }
318 EXPORT_SYMBOL(drm_buddy_free_list);
319 
320 static inline bool overlaps(u64 s1, u64 e1, u64 s2, u64 e2)
321 {
322 	return s1 <= e2 && e1 >= s2;
323 }
324 
325 static inline bool contains(u64 s1, u64 e1, u64 s2, u64 e2)
326 {
327 	return s1 <= s2 && e1 >= e2;
328 }
329 
330 static struct drm_buddy_block *
331 alloc_range_bias(struct drm_buddy *mm,
332 		 u64 start, u64 end,
333 		 unsigned int order)
334 {
335 	struct drm_buddy_block *block;
336 	struct drm_buddy_block *buddy;
337 	LIST_HEAD(dfs);
338 	int err;
339 	int i;
340 
341 	end = end - 1;
342 
343 	for (i = 0; i < mm->n_roots; ++i)
344 		list_add_tail(&mm->roots[i]->tmp_link, &dfs);
345 
346 	do {
347 		u64 block_start;
348 		u64 block_end;
349 
350 		block = list_first_entry_or_null(&dfs,
351 						 struct drm_buddy_block,
352 						 tmp_link);
353 		if (!block)
354 			break;
355 
356 		list_del(&block->tmp_link);
357 
358 		if (drm_buddy_block_order(block) < order)
359 			continue;
360 
361 		block_start = drm_buddy_block_offset(block);
362 		block_end = block_start + drm_buddy_block_size(mm, block) - 1;
363 
364 		if (!overlaps(start, end, block_start, block_end))
365 			continue;
366 
367 		if (drm_buddy_block_is_allocated(block))
368 			continue;
369 
370 		if (contains(start, end, block_start, block_end) &&
371 		    order == drm_buddy_block_order(block)) {
372 			/*
373 			 * Find the free block within the range.
374 			 */
375 			if (drm_buddy_block_is_free(block))
376 				return block;
377 
378 			continue;
379 		}
380 
381 		if (!drm_buddy_block_is_split(block)) {
382 			err = split_block(mm, block);
383 			if (unlikely(err))
384 				goto err_undo;
385 		}
386 
387 		list_add(&block->right->tmp_link, &dfs);
388 		list_add(&block->left->tmp_link, &dfs);
389 	} while (1);
390 
391 	return ERR_PTR(-ENOSPC);
392 
393 err_undo:
394 	/*
395 	 * We really don't want to leave around a bunch of split blocks, since
396 	 * bigger is better, so make sure we merge everything back before we
397 	 * free the allocated blocks.
398 	 */
399 	buddy = __get_buddy(block);
400 	if (buddy &&
401 	    (drm_buddy_block_is_free(block) &&
402 	     drm_buddy_block_is_free(buddy)))
403 		__drm_buddy_free(mm, block);
404 	return ERR_PTR(err);
405 }
406 
407 static struct drm_buddy_block *
408 get_maxblock(struct drm_buddy *mm, unsigned int order)
409 {
410 	struct drm_buddy_block *max_block = NULL, *node;
411 	unsigned int i;
412 
413 	for (i = order; i <= mm->max_order; ++i) {
414 		if (!list_empty(&mm->free_list[i])) {
415 			node = list_last_entry(&mm->free_list[i],
416 					       struct drm_buddy_block,
417 					       link);
418 			if (!max_block) {
419 				max_block = node;
420 				continue;
421 			}
422 
423 			if (drm_buddy_block_offset(node) >
424 			    drm_buddy_block_offset(max_block)) {
425 				max_block = node;
426 			}
427 		}
428 	}
429 
430 	return max_block;
431 }
432 
433 static struct drm_buddy_block *
434 alloc_from_freelist(struct drm_buddy *mm,
435 		    unsigned int order,
436 		    unsigned long flags)
437 {
438 	struct drm_buddy_block *block = NULL;
439 	unsigned int tmp;
440 	int err;
441 
442 	if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
443 		block = get_maxblock(mm, order);
444 		if (block)
445 			/* Store the obtained block order */
446 			tmp = drm_buddy_block_order(block);
447 	} else {
448 		for (tmp = order; tmp <= mm->max_order; ++tmp) {
449 			if (!list_empty(&mm->free_list[tmp])) {
450 				block = list_last_entry(&mm->free_list[tmp],
451 							struct drm_buddy_block,
452 							link);
453 				if (block)
454 					break;
455 			}
456 		}
457 	}
458 
459 	if (!block)
460 		return ERR_PTR(-ENOSPC);
461 
462 	BUG_ON(!drm_buddy_block_is_free(block));
463 
464 	while (tmp != order) {
465 		err = split_block(mm, block);
466 		if (unlikely(err))
467 			goto err_undo;
468 
469 		block = block->right;
470 		tmp--;
471 	}
472 	return block;
473 
474 err_undo:
475 	if (tmp != order)
476 		__drm_buddy_free(mm, block);
477 	return ERR_PTR(err);
478 }
479 
480 static int __alloc_range(struct drm_buddy *mm,
481 			 struct list_head *dfs,
482 			 u64 start, u64 size,
483 			 struct list_head *blocks)
484 {
485 	struct drm_buddy_block *block;
486 	struct drm_buddy_block *buddy;
487 	LIST_HEAD(allocated);
488 	u64 end;
489 	int err;
490 
491 	end = start + size - 1;
492 
493 	do {
494 		u64 block_start;
495 		u64 block_end;
496 
497 		block = list_first_entry_or_null(dfs,
498 						 struct drm_buddy_block,
499 						 tmp_link);
500 		if (!block)
501 			break;
502 
503 		list_del(&block->tmp_link);
504 
505 		block_start = drm_buddy_block_offset(block);
506 		block_end = block_start + drm_buddy_block_size(mm, block) - 1;
507 
508 		if (!overlaps(start, end, block_start, block_end))
509 			continue;
510 
511 		if (drm_buddy_block_is_allocated(block)) {
512 			err = -ENOSPC;
513 			goto err_free;
514 		}
515 
516 		if (contains(start, end, block_start, block_end)) {
517 			if (!drm_buddy_block_is_free(block)) {
518 				err = -ENOSPC;
519 				goto err_free;
520 			}
521 
522 			mark_allocated(block);
523 			mm->avail -= drm_buddy_block_size(mm, block);
524 			list_add_tail(&block->link, &allocated);
525 			continue;
526 		}
527 
528 		if (!drm_buddy_block_is_split(block)) {
529 			err = split_block(mm, block);
530 			if (unlikely(err))
531 				goto err_undo;
532 		}
533 
534 		list_add(&block->right->tmp_link, dfs);
535 		list_add(&block->left->tmp_link, dfs);
536 	} while (1);
537 
538 	list_splice_tail(&allocated, blocks);
539 	return 0;
540 
541 err_undo:
542 	/*
543 	 * We really don't want to leave around a bunch of split blocks, since
544 	 * bigger is better, so make sure we merge everything back before we
545 	 * free the allocated blocks.
546 	 */
547 	buddy = __get_buddy(block);
548 	if (buddy &&
549 	    (drm_buddy_block_is_free(block) &&
550 	     drm_buddy_block_is_free(buddy)))
551 		__drm_buddy_free(mm, block);
552 
553 err_free:
554 	drm_buddy_free_list(mm, &allocated);
555 	return err;
556 }
557 
558 static int __drm_buddy_alloc_range(struct drm_buddy *mm,
559 				   u64 start,
560 				   u64 size,
561 				   struct list_head *blocks)
562 {
563 	LIST_HEAD(dfs);
564 	int i;
565 
566 	for (i = 0; i < mm->n_roots; ++i)
567 		list_add_tail(&mm->roots[i]->tmp_link, &dfs);
568 
569 	return __alloc_range(mm, &dfs, start, size, blocks);
570 }
571 
572 /**
573  * drm_buddy_block_trim - free unused pages
574  *
575  * @mm: DRM buddy manager
576  * @new_size: original size requested
577  * @blocks: Input and output list of allocated blocks.
578  * MUST contain single block as input to be trimmed.
579  * On success will contain the newly allocated blocks
580  * making up the @new_size. Blocks always appear in
581  * ascending order
582  *
583  * For contiguous allocation, we round up the size to the nearest
584  * power of two value, drivers consume *actual* size, so remaining
585  * portions are unused and can be optionally freed with this function
586  *
587  * Returns:
588  * 0 on success, error code on failure.
589  */
590 int drm_buddy_block_trim(struct drm_buddy *mm,
591 			 u64 new_size,
592 			 struct list_head *blocks)
593 {
594 	struct drm_buddy_block *parent;
595 	struct drm_buddy_block *block;
596 	LIST_HEAD(dfs);
597 	u64 new_start;
598 	int err;
599 
600 	if (!list_is_singular(blocks))
601 		return -EINVAL;
602 
603 	block = list_first_entry(blocks,
604 				 struct drm_buddy_block,
605 				 link);
606 
607 	if (WARN_ON(!drm_buddy_block_is_allocated(block)))
608 		return -EINVAL;
609 
610 	if (new_size > drm_buddy_block_size(mm, block))
611 		return -EINVAL;
612 
613 	if (!new_size || !IS_ALIGNED(new_size, mm->chunk_size))
614 		return -EINVAL;
615 
616 	if (new_size == drm_buddy_block_size(mm, block))
617 		return 0;
618 
619 	list_del(&block->link);
620 	mark_free(mm, block);
621 	mm->avail += drm_buddy_block_size(mm, block);
622 
623 	/* Prevent recursively freeing this node */
624 	parent = block->parent;
625 	block->parent = NULL;
626 
627 	new_start = drm_buddy_block_offset(block);
628 	list_add(&block->tmp_link, &dfs);
629 	err =  __alloc_range(mm, &dfs, new_start, new_size, blocks);
630 	if (err) {
631 		mark_allocated(block);
632 		mm->avail -= drm_buddy_block_size(mm, block);
633 		list_add(&block->link, blocks);
634 	}
635 
636 	block->parent = parent;
637 	return err;
638 }
639 EXPORT_SYMBOL(drm_buddy_block_trim);
640 
641 /**
642  * drm_buddy_alloc_blocks - allocate power-of-two blocks
643  *
644  * @mm: DRM buddy manager to allocate from
645  * @start: start of the allowed range for this block
646  * @end: end of the allowed range for this block
647  * @size: size of the allocation
648  * @min_page_size: alignment of the allocation
649  * @blocks: output list head to add allocated blocks
650  * @flags: DRM_BUDDY_*_ALLOCATION flags
651  *
652  * alloc_range_bias() called on range limitations, which traverses
653  * the tree and returns the desired block.
654  *
655  * alloc_from_freelist() called when *no* range restrictions
656  * are enforced, which picks the block from the freelist.
657  *
658  * Returns:
659  * 0 on success, error code on failure.
660  */
661 int drm_buddy_alloc_blocks(struct drm_buddy *mm,
662 			   u64 start, u64 end, u64 size,
663 			   u64 min_page_size,
664 			   struct list_head *blocks,
665 			   unsigned long flags)
666 {
667 	struct drm_buddy_block *block = NULL;
668 	unsigned int min_order, order;
669 	unsigned long pages;
670 	LIST_HEAD(allocated);
671 	int err;
672 
673 	if (size < mm->chunk_size)
674 		return -EINVAL;
675 
676 	if (min_page_size < mm->chunk_size)
677 		return -EINVAL;
678 
679 	if (!is_power_of_2(min_page_size))
680 		return -EINVAL;
681 
682 	if (!IS_ALIGNED(start | end | size, mm->chunk_size))
683 		return -EINVAL;
684 
685 	if (end > mm->size)
686 		return -EINVAL;
687 
688 	if (range_overflows(start, size, mm->size))
689 		return -EINVAL;
690 
691 	/* Actual range allocation */
692 	if (start + size == end)
693 		return __drm_buddy_alloc_range(mm, start, size, blocks);
694 
695 	if (!IS_ALIGNED(size, min_page_size))
696 		return -EINVAL;
697 
698 	pages = size >> ilog2(mm->chunk_size);
699 	order = fls(pages) - 1;
700 	min_order = ilog2(min_page_size) - ilog2(mm->chunk_size);
701 
702 	do {
703 		order = min(order, (unsigned int)fls(pages) - 1);
704 		BUG_ON(order > mm->max_order);
705 		BUG_ON(order < min_order);
706 
707 		do {
708 			if (flags & DRM_BUDDY_RANGE_ALLOCATION)
709 				/* Allocate traversing within the range */
710 				block = alloc_range_bias(mm, start, end, order);
711 			else
712 				/* Allocate from freelist */
713 				block = alloc_from_freelist(mm, order, flags);
714 
715 			if (!IS_ERR(block))
716 				break;
717 
718 			if (order-- == min_order) {
719 				err = -ENOSPC;
720 				goto err_free;
721 			}
722 		} while (1);
723 
724 		mark_allocated(block);
725 		mm->avail -= drm_buddy_block_size(mm, block);
726 		kmemleak_update_trace(block);
727 		list_add_tail(&block->link, &allocated);
728 
729 		pages -= BIT(order);
730 
731 		if (!pages)
732 			break;
733 	} while (1);
734 
735 	list_splice_tail(&allocated, blocks);
736 	return 0;
737 
738 err_free:
739 	drm_buddy_free_list(mm, &allocated);
740 	return err;
741 }
742 EXPORT_SYMBOL(drm_buddy_alloc_blocks);
743 
744 /**
745  * drm_buddy_block_print - print block information
746  *
747  * @mm: DRM buddy manager
748  * @block: DRM buddy block
749  * @p: DRM printer to use
750  */
751 void drm_buddy_block_print(struct drm_buddy *mm,
752 			   struct drm_buddy_block *block,
753 			   struct drm_printer *p)
754 {
755 	u64 start = drm_buddy_block_offset(block);
756 	u64 size = drm_buddy_block_size(mm, block);
757 
758 	drm_printf(p, "%#018llx-%#018llx: %llu\n", start, start + size, size);
759 }
760 EXPORT_SYMBOL(drm_buddy_block_print);
761 
762 /**
763  * drm_buddy_print - print allocator state
764  *
765  * @mm: DRM buddy manager
766  * @p: DRM printer to use
767  */
768 void drm_buddy_print(struct drm_buddy *mm, struct drm_printer *p)
769 {
770 	int order;
771 
772 	drm_printf(p, "chunk_size: %lluKiB, total: %lluMiB, free: %lluMiB\n",
773 		   mm->chunk_size >> 10, mm->size >> 20, mm->avail >> 20);
774 
775 	for (order = mm->max_order; order >= 0; order--) {
776 		struct drm_buddy_block *block;
777 		u64 count = 0, free;
778 
779 		list_for_each_entry(block, &mm->free_list[order], link) {
780 			BUG_ON(!drm_buddy_block_is_free(block));
781 			count++;
782 		}
783 
784 		drm_printf(p, "order-%d ", order);
785 
786 		free = count * (mm->chunk_size << order);
787 		if (free < SZ_1M)
788 			drm_printf(p, "free: %lluKiB", free >> 10);
789 		else
790 			drm_printf(p, "free: %lluMiB", free >> 20);
791 
792 		drm_printf(p, ", pages: %llu\n", count);
793 	}
794 }
795 EXPORT_SYMBOL(drm_buddy_print);
796 
797 static void drm_buddy_module_exit(void)
798 {
799 	kmem_cache_destroy(slab_blocks);
800 }
801 
802 static int __init drm_buddy_module_init(void)
803 {
804 	slab_blocks = KMEM_CACHE(drm_buddy_block, 0);
805 	if (!slab_blocks)
806 		return -ENOMEM;
807 
808 	return 0;
809 }
810 
811 module_init(drm_buddy_module_init);
812 module_exit(drm_buddy_module_exit);
813 
814 MODULE_DESCRIPTION("DRM Buddy Allocator");
815 MODULE_LICENSE("Dual MIT/GPL");
816