xref: /openbmc/linux/drivers/gpu/drm/drm_mm.c (revision 3a83e4e6)
1 /**************************************************************************
2  *
3  * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA.
4  * Copyright 2016 Intel Corporation
5  * All Rights Reserved.
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a
8  * copy of this software and associated documentation files (the
9  * "Software"), to deal in the Software without restriction, including
10  * without limitation the rights to use, copy, modify, merge, publish,
11  * distribute, sub license, and/or sell copies of the Software, and to
12  * permit persons to whom the Software is furnished to do so, subject to
13  * the following conditions:
14  *
15  * The above copyright notice and this permission notice (including the
16  * next paragraph) shall be included in all copies or substantial portions
17  * of the Software.
18  *
19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
22  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
23  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
24  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
25  * USE OR OTHER DEALINGS IN THE SOFTWARE.
26  *
27  *
28  **************************************************************************/
29 
30 /*
31  * Generic simple memory manager implementation. Intended to be used as a base
32  * class implementation for more advanced memory managers.
33  *
34  * Note that the algorithm used is quite simple and there might be substantial
35  * performance gains if a smarter free list is implemented. Currently it is
36  * just an unordered stack of free regions. This could easily be improved if
37  * an RB-tree is used instead. At least if we expect heavy fragmentation.
38  *
39  * Aligned allocations can also see improvement.
40  *
41  * Authors:
42  * Thomas Hellström <thomas-at-tungstengraphics-dot-com>
43  */
44 
45 #include <linux/export.h>
46 #include <linux/interval_tree_generic.h>
47 #include <linux/seq_file.h>
48 #include <linux/slab.h>
49 #include <linux/stacktrace.h>
50 
51 #include <drm/drm_mm.h>
52 
53 /**
54  * DOC: Overview
55  *
56  * drm_mm provides a simple range allocator. The drivers are free to use the
57  * resource allocator from the linux core if it suits them, the upside of drm_mm
58  * is that it's in the DRM core. Which means that it's easier to extend for
59  * some of the crazier special purpose needs of gpus.
60  *
61  * The main data struct is &drm_mm, allocations are tracked in &drm_mm_node.
62  * Drivers are free to embed either of them into their own suitable
63  * datastructures. drm_mm itself will not do any memory allocations of its own,
64  * so if drivers choose not to embed nodes they need to still allocate them
65  * themselves.
66  *
67  * The range allocator also supports reservation of preallocated blocks. This is
68  * useful for taking over initial mode setting configurations from the firmware,
69  * where an object needs to be created which exactly matches the firmware's
70  * scanout target. As long as the range is still free it can be inserted anytime
71  * after the allocator is initialized, which helps with avoiding looped
72  * dependencies in the driver load sequence.
73  *
74  * drm_mm maintains a stack of most recently freed holes, which of all
75  * simplistic datastructures seems to be a fairly decent approach to clustering
76  * allocations and avoiding too much fragmentation. This means free space
77  * searches are O(num_holes). Given that all the fancy features drm_mm supports
78  * something better would be fairly complex and since gfx thrashing is a fairly
79  * steep cliff not a real concern. Removing a node again is O(1).
80  *
81  * drm_mm supports a few features: Alignment and range restrictions can be
82  * supplied. Furthermore every &drm_mm_node has a color value (which is just an
83  * opaque unsigned long) which in conjunction with a driver callback can be used
84  * to implement sophisticated placement restrictions. The i915 DRM driver uses
85  * this to implement guard pages between incompatible caching domains in the
86  * graphics TT.
87  *
88  * Two behaviors are supported for searching and allocating: bottom-up and
89  * top-down. The default is bottom-up. Top-down allocation can be used if the
90  * memory area has different restrictions, or just to reduce fragmentation.
91  *
92  * Finally iteration helpers to walk all nodes and all holes are provided as are
93  * some basic allocator dumpers for debugging.
94  *
95  * Note that this range allocator is not thread-safe, drivers need to protect
96  * modifications with their own locking. The idea behind this is that for a full
97  * memory manager additional data needs to be protected anyway, hence internal
98  * locking would be fully redundant.
99  */
100 
101 #ifdef CONFIG_DRM_DEBUG_MM
102 #include <linux/stackdepot.h>
103 
104 #define STACKDEPTH 32
105 #define BUFSZ 4096
106 
107 static noinline void save_stack(struct drm_mm_node *node)
108 {
109 	unsigned long entries[STACKDEPTH];
110 	unsigned int n;
111 
112 	n = stack_trace_save(entries, ARRAY_SIZE(entries), 1);
113 
114 	/* May be called under spinlock, so avoid sleeping */
115 	node->stack = stack_depot_save(entries, n, GFP_NOWAIT);
116 }
117 
118 static void show_leaks(struct drm_mm *mm)
119 {
120 	struct drm_mm_node *node;
121 	unsigned long *entries;
122 	unsigned int nr_entries;
123 	char *buf;
124 
125 	buf = kmalloc(BUFSZ, GFP_KERNEL);
126 	if (!buf)
127 		return;
128 
129 	list_for_each_entry(node, drm_mm_nodes(mm), node_list) {
130 		if (!node->stack) {
131 			DRM_ERROR("node [%08llx + %08llx]: unknown owner\n",
132 				  node->start, node->size);
133 			continue;
134 		}
135 
136 		nr_entries = stack_depot_fetch(node->stack, &entries);
137 		stack_trace_snprint(buf, BUFSZ, entries, nr_entries, 0);
138 		DRM_ERROR("node [%08llx + %08llx]: inserted at\n%s",
139 			  node->start, node->size, buf);
140 	}
141 
142 	kfree(buf);
143 }
144 
145 #undef STACKDEPTH
146 #undef BUFSZ
147 #else
148 static void save_stack(struct drm_mm_node *node) { }
149 static void show_leaks(struct drm_mm *mm) { }
150 #endif
151 
152 #define START(node) ((node)->start)
153 #define LAST(node)  ((node)->start + (node)->size - 1)
154 
155 INTERVAL_TREE_DEFINE(struct drm_mm_node, rb,
156 		     u64, __subtree_last,
157 		     START, LAST, static inline, drm_mm_interval_tree)
158 
159 struct drm_mm_node *
160 __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last)
161 {
162 	return drm_mm_interval_tree_iter_first((struct rb_root_cached *)&mm->interval_tree,
163 					       start, last) ?: (struct drm_mm_node *)&mm->head_node;
164 }
165 EXPORT_SYMBOL(__drm_mm_interval_first);
166 
167 static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node,
168 					  struct drm_mm_node *node)
169 {
170 	struct drm_mm *mm = hole_node->mm;
171 	struct rb_node **link, *rb;
172 	struct drm_mm_node *parent;
173 	bool leftmost;
174 
175 	node->__subtree_last = LAST(node);
176 
177 	if (drm_mm_node_allocated(hole_node)) {
178 		rb = &hole_node->rb;
179 		while (rb) {
180 			parent = rb_entry(rb, struct drm_mm_node, rb);
181 			if (parent->__subtree_last >= node->__subtree_last)
182 				break;
183 
184 			parent->__subtree_last = node->__subtree_last;
185 			rb = rb_parent(rb);
186 		}
187 
188 		rb = &hole_node->rb;
189 		link = &hole_node->rb.rb_right;
190 		leftmost = false;
191 	} else {
192 		rb = NULL;
193 		link = &mm->interval_tree.rb_root.rb_node;
194 		leftmost = true;
195 	}
196 
197 	while (*link) {
198 		rb = *link;
199 		parent = rb_entry(rb, struct drm_mm_node, rb);
200 		if (parent->__subtree_last < node->__subtree_last)
201 			parent->__subtree_last = node->__subtree_last;
202 		if (node->start < parent->start) {
203 			link = &parent->rb.rb_left;
204 		} else {
205 			link = &parent->rb.rb_right;
206 			leftmost = false;
207 		}
208 	}
209 
210 	rb_link_node(&node->rb, rb, link);
211 	rb_insert_augmented_cached(&node->rb, &mm->interval_tree, leftmost,
212 				   &drm_mm_interval_tree_augment);
213 }
214 
215 #define HOLE_SIZE(NODE) ((NODE)->hole_size)
216 #define HOLE_ADDR(NODE) (__drm_mm_hole_node_start(NODE))
217 
218 static u64 rb_to_hole_size(struct rb_node *rb)
219 {
220 	return rb_entry(rb, struct drm_mm_node, rb_hole_size)->hole_size;
221 }
222 
223 static void insert_hole_size(struct rb_root_cached *root,
224 			     struct drm_mm_node *node)
225 {
226 	struct rb_node **link = &root->rb_root.rb_node, *rb = NULL;
227 	u64 x = node->hole_size;
228 	bool first = true;
229 
230 	while (*link) {
231 		rb = *link;
232 		if (x > rb_to_hole_size(rb)) {
233 			link = &rb->rb_left;
234 		} else {
235 			link = &rb->rb_right;
236 			first = false;
237 		}
238 	}
239 
240 	rb_link_node(&node->rb_hole_size, rb, link);
241 	rb_insert_color_cached(&node->rb_hole_size, root, first);
242 }
243 
244 RB_DECLARE_CALLBACKS_MAX(static, augment_callbacks,
245 			 struct drm_mm_node, rb_hole_addr,
246 			 u64, subtree_max_hole, HOLE_SIZE)
247 
248 static void insert_hole_addr(struct rb_root *root, struct drm_mm_node *node)
249 {
250 	struct rb_node **link = &root->rb_node, *rb_parent = NULL;
251 	u64 start = HOLE_ADDR(node), subtree_max_hole = node->subtree_max_hole;
252 	struct drm_mm_node *parent;
253 
254 	while (*link) {
255 		rb_parent = *link;
256 		parent = rb_entry(rb_parent, struct drm_mm_node, rb_hole_addr);
257 		if (parent->subtree_max_hole < subtree_max_hole)
258 			parent->subtree_max_hole = subtree_max_hole;
259 		if (start < HOLE_ADDR(parent))
260 			link = &parent->rb_hole_addr.rb_left;
261 		else
262 			link = &parent->rb_hole_addr.rb_right;
263 	}
264 
265 	rb_link_node(&node->rb_hole_addr, rb_parent, link);
266 	rb_insert_augmented(&node->rb_hole_addr, root, &augment_callbacks);
267 }
268 
269 static void add_hole(struct drm_mm_node *node)
270 {
271 	struct drm_mm *mm = node->mm;
272 
273 	node->hole_size =
274 		__drm_mm_hole_node_end(node) - __drm_mm_hole_node_start(node);
275 	node->subtree_max_hole = node->hole_size;
276 	DRM_MM_BUG_ON(!drm_mm_hole_follows(node));
277 
278 	insert_hole_size(&mm->holes_size, node);
279 	insert_hole_addr(&mm->holes_addr, node);
280 
281 	list_add(&node->hole_stack, &mm->hole_stack);
282 }
283 
284 static void rm_hole(struct drm_mm_node *node)
285 {
286 	DRM_MM_BUG_ON(!drm_mm_hole_follows(node));
287 
288 	list_del(&node->hole_stack);
289 	rb_erase_cached(&node->rb_hole_size, &node->mm->holes_size);
290 	rb_erase_augmented(&node->rb_hole_addr, &node->mm->holes_addr,
291 			   &augment_callbacks);
292 	node->hole_size = 0;
293 	node->subtree_max_hole = 0;
294 
295 	DRM_MM_BUG_ON(drm_mm_hole_follows(node));
296 }
297 
298 static inline struct drm_mm_node *rb_hole_size_to_node(struct rb_node *rb)
299 {
300 	return rb_entry_safe(rb, struct drm_mm_node, rb_hole_size);
301 }
302 
303 static inline struct drm_mm_node *rb_hole_addr_to_node(struct rb_node *rb)
304 {
305 	return rb_entry_safe(rb, struct drm_mm_node, rb_hole_addr);
306 }
307 
308 static struct drm_mm_node *best_hole(struct drm_mm *mm, u64 size)
309 {
310 	struct rb_node *rb = mm->holes_size.rb_root.rb_node;
311 	struct drm_mm_node *best = NULL;
312 
313 	do {
314 		struct drm_mm_node *node =
315 			rb_entry(rb, struct drm_mm_node, rb_hole_size);
316 
317 		if (size <= node->hole_size) {
318 			best = node;
319 			rb = rb->rb_right;
320 		} else {
321 			rb = rb->rb_left;
322 		}
323 	} while (rb);
324 
325 	return best;
326 }
327 
328 static bool usable_hole_addr(struct rb_node *rb, u64 size)
329 {
330 	return rb && rb_hole_addr_to_node(rb)->subtree_max_hole >= size;
331 }
332 
333 static struct drm_mm_node *find_hole_addr(struct drm_mm *mm, u64 addr, u64 size)
334 {
335 	struct rb_node *rb = mm->holes_addr.rb_node;
336 	struct drm_mm_node *node = NULL;
337 
338 	while (rb) {
339 		u64 hole_start;
340 
341 		if (!usable_hole_addr(rb, size))
342 			break;
343 
344 		node = rb_hole_addr_to_node(rb);
345 		hole_start = __drm_mm_hole_node_start(node);
346 
347 		if (addr < hole_start)
348 			rb = node->rb_hole_addr.rb_left;
349 		else if (addr > hole_start + node->hole_size)
350 			rb = node->rb_hole_addr.rb_right;
351 		else
352 			break;
353 	}
354 
355 	return node;
356 }
357 
358 static struct drm_mm_node *
359 first_hole(struct drm_mm *mm,
360 	   u64 start, u64 end, u64 size,
361 	   enum drm_mm_insert_mode mode)
362 {
363 	switch (mode) {
364 	default:
365 	case DRM_MM_INSERT_BEST:
366 		return best_hole(mm, size);
367 
368 	case DRM_MM_INSERT_LOW:
369 		return find_hole_addr(mm, start, size);
370 
371 	case DRM_MM_INSERT_HIGH:
372 		return find_hole_addr(mm, end, size);
373 
374 	case DRM_MM_INSERT_EVICT:
375 		return list_first_entry_or_null(&mm->hole_stack,
376 						struct drm_mm_node,
377 						hole_stack);
378 	}
379 }
380 
381 /**
382  * DECLARE_NEXT_HOLE_ADDR - macro to declare next hole functions
383  * @name: name of function to declare
384  * @first: first rb member to traverse (either rb_left or rb_right).
385  * @last: last rb member to traverse (either rb_right or rb_left).
386  *
387  * This macro declares a function to return the next hole of the addr rb tree.
388  * While traversing the tree we take the searched size into account and only
389  * visit branches with potential big enough holes.
390  */
391 
392 #define DECLARE_NEXT_HOLE_ADDR(name, first, last)			\
393 static struct drm_mm_node *name(struct drm_mm_node *entry, u64 size)	\
394 {									\
395 	struct rb_node *parent, *node = &entry->rb_hole_addr;		\
396 									\
397 	if (!entry || RB_EMPTY_NODE(node))				\
398 		return NULL;						\
399 									\
400 	if (usable_hole_addr(node->first, size)) {			\
401 		node = node->first;					\
402 		while (usable_hole_addr(node->last, size))		\
403 			node = node->last;				\
404 		return rb_hole_addr_to_node(node);			\
405 	}								\
406 									\
407 	while ((parent = rb_parent(node)) && node == parent->first)	\
408 		node = parent;						\
409 									\
410 	return rb_hole_addr_to_node(parent);				\
411 }
412 
413 DECLARE_NEXT_HOLE_ADDR(next_hole_high_addr, rb_left, rb_right)
414 DECLARE_NEXT_HOLE_ADDR(next_hole_low_addr, rb_right, rb_left)
415 
416 static struct drm_mm_node *
417 next_hole(struct drm_mm *mm,
418 	  struct drm_mm_node *node,
419 	  u64 size,
420 	  enum drm_mm_insert_mode mode)
421 {
422 	switch (mode) {
423 	default:
424 	case DRM_MM_INSERT_BEST:
425 		return rb_hole_size_to_node(rb_prev(&node->rb_hole_size));
426 
427 	case DRM_MM_INSERT_LOW:
428 		return next_hole_low_addr(node, size);
429 
430 	case DRM_MM_INSERT_HIGH:
431 		return next_hole_high_addr(node, size);
432 
433 	case DRM_MM_INSERT_EVICT:
434 		node = list_next_entry(node, hole_stack);
435 		return &node->hole_stack == &mm->hole_stack ? NULL : node;
436 	}
437 }
438 
439 /**
440  * drm_mm_reserve_node - insert an pre-initialized node
441  * @mm: drm_mm allocator to insert @node into
442  * @node: drm_mm_node to insert
443  *
444  * This functions inserts an already set-up &drm_mm_node into the allocator,
445  * meaning that start, size and color must be set by the caller. All other
446  * fields must be cleared to 0. This is useful to initialize the allocator with
447  * preallocated objects which must be set-up before the range allocator can be
448  * set-up, e.g. when taking over a firmware framebuffer.
449  *
450  * Returns:
451  * 0 on success, -ENOSPC if there's no hole where @node is.
452  */
453 int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
454 {
455 	struct drm_mm_node *hole;
456 	u64 hole_start, hole_end;
457 	u64 adj_start, adj_end;
458 	u64 end;
459 
460 	end = node->start + node->size;
461 	if (unlikely(end <= node->start))
462 		return -ENOSPC;
463 
464 	/* Find the relevant hole to add our node to */
465 	hole = find_hole_addr(mm, node->start, 0);
466 	if (!hole)
467 		return -ENOSPC;
468 
469 	adj_start = hole_start = __drm_mm_hole_node_start(hole);
470 	adj_end = hole_end = hole_start + hole->hole_size;
471 
472 	if (mm->color_adjust)
473 		mm->color_adjust(hole, node->color, &adj_start, &adj_end);
474 
475 	if (adj_start > node->start || adj_end < end)
476 		return -ENOSPC;
477 
478 	node->mm = mm;
479 
480 	__set_bit(DRM_MM_NODE_ALLOCATED_BIT, &node->flags);
481 	list_add(&node->node_list, &hole->node_list);
482 	drm_mm_interval_tree_add_node(hole, node);
483 	node->hole_size = 0;
484 
485 	rm_hole(hole);
486 	if (node->start > hole_start)
487 		add_hole(hole);
488 	if (end < hole_end)
489 		add_hole(node);
490 
491 	save_stack(node);
492 	return 0;
493 }
494 EXPORT_SYMBOL(drm_mm_reserve_node);
495 
496 static u64 rb_to_hole_size_or_zero(struct rb_node *rb)
497 {
498 	return rb ? rb_to_hole_size(rb) : 0;
499 }
500 
501 /**
502  * drm_mm_insert_node_in_range - ranged search for space and insert @node
503  * @mm: drm_mm to allocate from
504  * @node: preallocate node to insert
505  * @size: size of the allocation
506  * @alignment: alignment of the allocation
507  * @color: opaque tag value to use for this node
508  * @range_start: start of the allowed range for this node
509  * @range_end: end of the allowed range for this node
510  * @mode: fine-tune the allocation search and placement
511  *
512  * The preallocated @node must be cleared to 0.
513  *
514  * Returns:
515  * 0 on success, -ENOSPC if there's no suitable hole.
516  */
517 int drm_mm_insert_node_in_range(struct drm_mm * const mm,
518 				struct drm_mm_node * const node,
519 				u64 size, u64 alignment,
520 				unsigned long color,
521 				u64 range_start, u64 range_end,
522 				enum drm_mm_insert_mode mode)
523 {
524 	struct drm_mm_node *hole;
525 	u64 remainder_mask;
526 	bool once;
527 
528 	DRM_MM_BUG_ON(range_start > range_end);
529 
530 	if (unlikely(size == 0 || range_end - range_start < size))
531 		return -ENOSPC;
532 
533 	if (rb_to_hole_size_or_zero(rb_first_cached(&mm->holes_size)) < size)
534 		return -ENOSPC;
535 
536 	if (alignment <= 1)
537 		alignment = 0;
538 
539 	once = mode & DRM_MM_INSERT_ONCE;
540 	mode &= ~DRM_MM_INSERT_ONCE;
541 
542 	remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
543 	for (hole = first_hole(mm, range_start, range_end, size, mode);
544 	     hole;
545 	     hole = once ? NULL : next_hole(mm, hole, size, mode)) {
546 		u64 hole_start = __drm_mm_hole_node_start(hole);
547 		u64 hole_end = hole_start + hole->hole_size;
548 		u64 adj_start, adj_end;
549 		u64 col_start, col_end;
550 
551 		if (mode == DRM_MM_INSERT_LOW && hole_start >= range_end)
552 			break;
553 
554 		if (mode == DRM_MM_INSERT_HIGH && hole_end <= range_start)
555 			break;
556 
557 		col_start = hole_start;
558 		col_end = hole_end;
559 		if (mm->color_adjust)
560 			mm->color_adjust(hole, color, &col_start, &col_end);
561 
562 		adj_start = max(col_start, range_start);
563 		adj_end = min(col_end, range_end);
564 
565 		if (adj_end <= adj_start || adj_end - adj_start < size)
566 			continue;
567 
568 		if (mode == DRM_MM_INSERT_HIGH)
569 			adj_start = adj_end - size;
570 
571 		if (alignment) {
572 			u64 rem;
573 
574 			if (likely(remainder_mask))
575 				rem = adj_start & remainder_mask;
576 			else
577 				div64_u64_rem(adj_start, alignment, &rem);
578 			if (rem) {
579 				adj_start -= rem;
580 				if (mode != DRM_MM_INSERT_HIGH)
581 					adj_start += alignment;
582 
583 				if (adj_start < max(col_start, range_start) ||
584 				    min(col_end, range_end) - adj_start < size)
585 					continue;
586 
587 				if (adj_end <= adj_start ||
588 				    adj_end - adj_start < size)
589 					continue;
590 			}
591 		}
592 
593 		node->mm = mm;
594 		node->size = size;
595 		node->start = adj_start;
596 		node->color = color;
597 		node->hole_size = 0;
598 
599 		__set_bit(DRM_MM_NODE_ALLOCATED_BIT, &node->flags);
600 		list_add(&node->node_list, &hole->node_list);
601 		drm_mm_interval_tree_add_node(hole, node);
602 
603 		rm_hole(hole);
604 		if (adj_start > hole_start)
605 			add_hole(hole);
606 		if (adj_start + size < hole_end)
607 			add_hole(node);
608 
609 		save_stack(node);
610 		return 0;
611 	}
612 
613 	return -ENOSPC;
614 }
615 EXPORT_SYMBOL(drm_mm_insert_node_in_range);
616 
617 static inline bool drm_mm_node_scanned_block(const struct drm_mm_node *node)
618 {
619 	return test_bit(DRM_MM_NODE_SCANNED_BIT, &node->flags);
620 }
621 
622 /**
623  * drm_mm_remove_node - Remove a memory node from the allocator.
624  * @node: drm_mm_node to remove
625  *
626  * This just removes a node from its drm_mm allocator. The node does not need to
627  * be cleared again before it can be re-inserted into this or any other drm_mm
628  * allocator. It is a bug to call this function on a unallocated node.
629  */
630 void drm_mm_remove_node(struct drm_mm_node *node)
631 {
632 	struct drm_mm *mm = node->mm;
633 	struct drm_mm_node *prev_node;
634 
635 	DRM_MM_BUG_ON(!drm_mm_node_allocated(node));
636 	DRM_MM_BUG_ON(drm_mm_node_scanned_block(node));
637 
638 	prev_node = list_prev_entry(node, node_list);
639 
640 	if (drm_mm_hole_follows(node))
641 		rm_hole(node);
642 
643 	drm_mm_interval_tree_remove(node, &mm->interval_tree);
644 	list_del(&node->node_list);
645 
646 	if (drm_mm_hole_follows(prev_node))
647 		rm_hole(prev_node);
648 	add_hole(prev_node);
649 
650 	clear_bit_unlock(DRM_MM_NODE_ALLOCATED_BIT, &node->flags);
651 }
652 EXPORT_SYMBOL(drm_mm_remove_node);
653 
654 /**
655  * drm_mm_replace_node - move an allocation from @old to @new
656  * @old: drm_mm_node to remove from the allocator
657  * @new: drm_mm_node which should inherit @old's allocation
658  *
659  * This is useful for when drivers embed the drm_mm_node structure and hence
660  * can't move allocations by reassigning pointers. It's a combination of remove
661  * and insert with the guarantee that the allocation start will match.
662  */
663 void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new)
664 {
665 	struct drm_mm *mm = old->mm;
666 
667 	DRM_MM_BUG_ON(!drm_mm_node_allocated(old));
668 
669 	*new = *old;
670 
671 	__set_bit(DRM_MM_NODE_ALLOCATED_BIT, &new->flags);
672 	list_replace(&old->node_list, &new->node_list);
673 	rb_replace_node_cached(&old->rb, &new->rb, &mm->interval_tree);
674 
675 	if (drm_mm_hole_follows(old)) {
676 		list_replace(&old->hole_stack, &new->hole_stack);
677 		rb_replace_node_cached(&old->rb_hole_size,
678 				       &new->rb_hole_size,
679 				       &mm->holes_size);
680 		rb_replace_node(&old->rb_hole_addr,
681 				&new->rb_hole_addr,
682 				&mm->holes_addr);
683 	}
684 
685 	clear_bit_unlock(DRM_MM_NODE_ALLOCATED_BIT, &old->flags);
686 }
687 EXPORT_SYMBOL(drm_mm_replace_node);
688 
689 /**
690  * DOC: lru scan roster
691  *
692  * Very often GPUs need to have continuous allocations for a given object. When
693  * evicting objects to make space for a new one it is therefore not most
694  * efficient when we simply start to select all objects from the tail of an LRU
695  * until there's a suitable hole: Especially for big objects or nodes that
696  * otherwise have special allocation constraints there's a good chance we evict
697  * lots of (smaller) objects unnecessarily.
698  *
699  * The DRM range allocator supports this use-case through the scanning
700  * interfaces. First a scan operation needs to be initialized with
701  * drm_mm_scan_init() or drm_mm_scan_init_with_range(). The driver adds
702  * objects to the roster, probably by walking an LRU list, but this can be
703  * freely implemented. Eviction candiates are added using
704  * drm_mm_scan_add_block() until a suitable hole is found or there are no
705  * further evictable objects. Eviction roster metadata is tracked in &struct
706  * drm_mm_scan.
707  *
708  * The driver must walk through all objects again in exactly the reverse
709  * order to restore the allocator state. Note that while the allocator is used
710  * in the scan mode no other operation is allowed.
711  *
712  * Finally the driver evicts all objects selected (drm_mm_scan_remove_block()
713  * reported true) in the scan, and any overlapping nodes after color adjustment
714  * (drm_mm_scan_color_evict()). Adding and removing an object is O(1), and
715  * since freeing a node is also O(1) the overall complexity is
716  * O(scanned_objects). So like the free stack which needs to be walked before a
717  * scan operation even begins this is linear in the number of objects. It
718  * doesn't seem to hurt too badly.
719  */
720 
721 /**
722  * drm_mm_scan_init_with_range - initialize range-restricted lru scanning
723  * @scan: scan state
724  * @mm: drm_mm to scan
725  * @size: size of the allocation
726  * @alignment: alignment of the allocation
727  * @color: opaque tag value to use for the allocation
728  * @start: start of the allowed range for the allocation
729  * @end: end of the allowed range for the allocation
730  * @mode: fine-tune the allocation search and placement
731  *
732  * This simply sets up the scanning routines with the parameters for the desired
733  * hole.
734  *
735  * Warning:
736  * As long as the scan list is non-empty, no other operations than
737  * adding/removing nodes to/from the scan list are allowed.
738  */
739 void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
740 				 struct drm_mm *mm,
741 				 u64 size,
742 				 u64 alignment,
743 				 unsigned long color,
744 				 u64 start,
745 				 u64 end,
746 				 enum drm_mm_insert_mode mode)
747 {
748 	DRM_MM_BUG_ON(start >= end);
749 	DRM_MM_BUG_ON(!size || size > end - start);
750 	DRM_MM_BUG_ON(mm->scan_active);
751 
752 	scan->mm = mm;
753 
754 	if (alignment <= 1)
755 		alignment = 0;
756 
757 	scan->color = color;
758 	scan->alignment = alignment;
759 	scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
760 	scan->size = size;
761 	scan->mode = mode;
762 
763 	DRM_MM_BUG_ON(end <= start);
764 	scan->range_start = start;
765 	scan->range_end = end;
766 
767 	scan->hit_start = U64_MAX;
768 	scan->hit_end = 0;
769 }
770 EXPORT_SYMBOL(drm_mm_scan_init_with_range);
771 
772 /**
773  * drm_mm_scan_add_block - add a node to the scan list
774  * @scan: the active drm_mm scanner
775  * @node: drm_mm_node to add
776  *
777  * Add a node to the scan list that might be freed to make space for the desired
778  * hole.
779  *
780  * Returns:
781  * True if a hole has been found, false otherwise.
782  */
783 bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
784 			   struct drm_mm_node *node)
785 {
786 	struct drm_mm *mm = scan->mm;
787 	struct drm_mm_node *hole;
788 	u64 hole_start, hole_end;
789 	u64 col_start, col_end;
790 	u64 adj_start, adj_end;
791 
792 	DRM_MM_BUG_ON(node->mm != mm);
793 	DRM_MM_BUG_ON(!drm_mm_node_allocated(node));
794 	DRM_MM_BUG_ON(drm_mm_node_scanned_block(node));
795 	__set_bit(DRM_MM_NODE_SCANNED_BIT, &node->flags);
796 	mm->scan_active++;
797 
798 	/* Remove this block from the node_list so that we enlarge the hole
799 	 * (distance between the end of our previous node and the start of
800 	 * or next), without poisoning the link so that we can restore it
801 	 * later in drm_mm_scan_remove_block().
802 	 */
803 	hole = list_prev_entry(node, node_list);
804 	DRM_MM_BUG_ON(list_next_entry(hole, node_list) != node);
805 	__list_del_entry(&node->node_list);
806 
807 	hole_start = __drm_mm_hole_node_start(hole);
808 	hole_end = __drm_mm_hole_node_end(hole);
809 
810 	col_start = hole_start;
811 	col_end = hole_end;
812 	if (mm->color_adjust)
813 		mm->color_adjust(hole, scan->color, &col_start, &col_end);
814 
815 	adj_start = max(col_start, scan->range_start);
816 	adj_end = min(col_end, scan->range_end);
817 	if (adj_end <= adj_start || adj_end - adj_start < scan->size)
818 		return false;
819 
820 	if (scan->mode == DRM_MM_INSERT_HIGH)
821 		adj_start = adj_end - scan->size;
822 
823 	if (scan->alignment) {
824 		u64 rem;
825 
826 		if (likely(scan->remainder_mask))
827 			rem = adj_start & scan->remainder_mask;
828 		else
829 			div64_u64_rem(adj_start, scan->alignment, &rem);
830 		if (rem) {
831 			adj_start -= rem;
832 			if (scan->mode != DRM_MM_INSERT_HIGH)
833 				adj_start += scan->alignment;
834 			if (adj_start < max(col_start, scan->range_start) ||
835 			    min(col_end, scan->range_end) - adj_start < scan->size)
836 				return false;
837 
838 			if (adj_end <= adj_start ||
839 			    adj_end - adj_start < scan->size)
840 				return false;
841 		}
842 	}
843 
844 	scan->hit_start = adj_start;
845 	scan->hit_end = adj_start + scan->size;
846 
847 	DRM_MM_BUG_ON(scan->hit_start >= scan->hit_end);
848 	DRM_MM_BUG_ON(scan->hit_start < hole_start);
849 	DRM_MM_BUG_ON(scan->hit_end > hole_end);
850 
851 	return true;
852 }
853 EXPORT_SYMBOL(drm_mm_scan_add_block);
854 
855 /**
856  * drm_mm_scan_remove_block - remove a node from the scan list
857  * @scan: the active drm_mm scanner
858  * @node: drm_mm_node to remove
859  *
860  * Nodes **must** be removed in exactly the reverse order from the scan list as
861  * they have been added (e.g. using list_add() as they are added and then
862  * list_for_each() over that eviction list to remove), otherwise the internal
863  * state of the memory manager will be corrupted.
864  *
865  * When the scan list is empty, the selected memory nodes can be freed. An
866  * immediately following drm_mm_insert_node_in_range_generic() or one of the
867  * simpler versions of that function with !DRM_MM_SEARCH_BEST will then return
868  * the just freed block (because it's at the top of the free_stack list).
869  *
870  * Returns:
871  * True if this block should be evicted, false otherwise. Will always
872  * return false when no hole has been found.
873  */
874 bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
875 			      struct drm_mm_node *node)
876 {
877 	struct drm_mm_node *prev_node;
878 
879 	DRM_MM_BUG_ON(node->mm != scan->mm);
880 	DRM_MM_BUG_ON(!drm_mm_node_scanned_block(node));
881 	__clear_bit(DRM_MM_NODE_SCANNED_BIT, &node->flags);
882 
883 	DRM_MM_BUG_ON(!node->mm->scan_active);
884 	node->mm->scan_active--;
885 
886 	/* During drm_mm_scan_add_block() we decoupled this node leaving
887 	 * its pointers intact. Now that the caller is walking back along
888 	 * the eviction list we can restore this block into its rightful
889 	 * place on the full node_list. To confirm that the caller is walking
890 	 * backwards correctly we check that prev_node->next == node->next,
891 	 * i.e. both believe the same node should be on the other side of the
892 	 * hole.
893 	 */
894 	prev_node = list_prev_entry(node, node_list);
895 	DRM_MM_BUG_ON(list_next_entry(prev_node, node_list) !=
896 		      list_next_entry(node, node_list));
897 	list_add(&node->node_list, &prev_node->node_list);
898 
899 	return (node->start + node->size > scan->hit_start &&
900 		node->start < scan->hit_end);
901 }
902 EXPORT_SYMBOL(drm_mm_scan_remove_block);
903 
904 /**
905  * drm_mm_scan_color_evict - evict overlapping nodes on either side of hole
906  * @scan: drm_mm scan with target hole
907  *
908  * After completing an eviction scan and removing the selected nodes, we may
909  * need to remove a few more nodes from either side of the target hole if
910  * mm.color_adjust is being used.
911  *
912  * Returns:
913  * A node to evict, or NULL if there are no overlapping nodes.
914  */
915 struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan)
916 {
917 	struct drm_mm *mm = scan->mm;
918 	struct drm_mm_node *hole;
919 	u64 hole_start, hole_end;
920 
921 	DRM_MM_BUG_ON(list_empty(&mm->hole_stack));
922 
923 	if (!mm->color_adjust)
924 		return NULL;
925 
926 	/*
927 	 * The hole found during scanning should ideally be the first element
928 	 * in the hole_stack list, but due to side-effects in the driver it
929 	 * may not be.
930 	 */
931 	list_for_each_entry(hole, &mm->hole_stack, hole_stack) {
932 		hole_start = __drm_mm_hole_node_start(hole);
933 		hole_end = hole_start + hole->hole_size;
934 
935 		if (hole_start <= scan->hit_start &&
936 		    hole_end >= scan->hit_end)
937 			break;
938 	}
939 
940 	/* We should only be called after we found the hole previously */
941 	DRM_MM_BUG_ON(&hole->hole_stack == &mm->hole_stack);
942 	if (unlikely(&hole->hole_stack == &mm->hole_stack))
943 		return NULL;
944 
945 	DRM_MM_BUG_ON(hole_start > scan->hit_start);
946 	DRM_MM_BUG_ON(hole_end < scan->hit_end);
947 
948 	mm->color_adjust(hole, scan->color, &hole_start, &hole_end);
949 	if (hole_start > scan->hit_start)
950 		return hole;
951 	if (hole_end < scan->hit_end)
952 		return list_next_entry(hole, node_list);
953 
954 	return NULL;
955 }
956 EXPORT_SYMBOL(drm_mm_scan_color_evict);
957 
958 /**
959  * drm_mm_init - initialize a drm-mm allocator
960  * @mm: the drm_mm structure to initialize
961  * @start: start of the range managed by @mm
962  * @size: end of the range managed by @mm
963  *
964  * Note that @mm must be cleared to 0 before calling this function.
965  */
966 void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
967 {
968 	DRM_MM_BUG_ON(start + size <= start);
969 
970 	mm->color_adjust = NULL;
971 
972 	INIT_LIST_HEAD(&mm->hole_stack);
973 	mm->interval_tree = RB_ROOT_CACHED;
974 	mm->holes_size = RB_ROOT_CACHED;
975 	mm->holes_addr = RB_ROOT;
976 
977 	/* Clever trick to avoid a special case in the free hole tracking. */
978 	INIT_LIST_HEAD(&mm->head_node.node_list);
979 	mm->head_node.flags = 0;
980 	mm->head_node.mm = mm;
981 	mm->head_node.start = start + size;
982 	mm->head_node.size = -size;
983 	add_hole(&mm->head_node);
984 
985 	mm->scan_active = 0;
986 }
987 EXPORT_SYMBOL(drm_mm_init);
988 
989 /**
990  * drm_mm_takedown - clean up a drm_mm allocator
991  * @mm: drm_mm allocator to clean up
992  *
993  * Note that it is a bug to call this function on an allocator which is not
994  * clean.
995  */
996 void drm_mm_takedown(struct drm_mm *mm)
997 {
998 	if (WARN(!drm_mm_clean(mm),
999 		 "Memory manager not clean during takedown.\n"))
1000 		show_leaks(mm);
1001 }
1002 EXPORT_SYMBOL(drm_mm_takedown);
1003 
1004 static u64 drm_mm_dump_hole(struct drm_printer *p, const struct drm_mm_node *entry)
1005 {
1006 	u64 start, size;
1007 
1008 	size = entry->hole_size;
1009 	if (size) {
1010 		start = drm_mm_hole_node_start(entry);
1011 		drm_printf(p, "%#018llx-%#018llx: %llu: free\n",
1012 			   start, start + size, size);
1013 	}
1014 
1015 	return size;
1016 }
1017 /**
1018  * drm_mm_print - print allocator state
1019  * @mm: drm_mm allocator to print
1020  * @p: DRM printer to use
1021  */
1022 void drm_mm_print(const struct drm_mm *mm, struct drm_printer *p)
1023 {
1024 	const struct drm_mm_node *entry;
1025 	u64 total_used = 0, total_free = 0, total = 0;
1026 
1027 	total_free += drm_mm_dump_hole(p, &mm->head_node);
1028 
1029 	drm_mm_for_each_node(entry, mm) {
1030 		drm_printf(p, "%#018llx-%#018llx: %llu: used\n", entry->start,
1031 			   entry->start + entry->size, entry->size);
1032 		total_used += entry->size;
1033 		total_free += drm_mm_dump_hole(p, entry);
1034 	}
1035 	total = total_free + total_used;
1036 
1037 	drm_printf(p, "total: %llu, used %llu free %llu\n", total,
1038 		   total_used, total_free);
1039 }
1040 EXPORT_SYMBOL(drm_mm_print);
1041