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 <drm/drmP.h>
46 #include <drm/drm_mm.h>
47 #include <linux/slab.h>
48 #include <linux/seq_file.h>
49 #include <linux/export.h>
50
51 extern int drm_vma_debug;
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 on 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 static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
102 u64 size,
103 u64 alignment,
104 unsigned long color,
105 u64 start,
106 u64 end,
107 enum drm_mm_search_flags flags);
108
109 #ifdef CONFIG_DRM_DEBUG_MM
110 #include <linux/stackdepot.h>
111
112 #define STACKDEPTH 32
113 #define BUFSZ 4096
114
save_stack(struct drm_mm_node * node)115 static noinline void save_stack(struct drm_mm_node *node)
116 {
117 unsigned long entries[STACKDEPTH];
118 struct stack_trace trace = {
119 .entries = entries,
120 .max_entries = STACKDEPTH,
121 .skip = 1
122 };
123
124 save_stack_trace(&trace);
125 if (trace.nr_entries != 0 &&
126 trace.entries[trace.nr_entries-1] == ULONG_MAX)
127 trace.nr_entries--;
128
129 /* May be called under spinlock, so avoid sleeping */
130 node->stack = depot_save_stack(&trace, GFP_NOWAIT);
131 }
132
show_leaks(struct drm_mm * mm)133 static void show_leaks(struct drm_mm *mm)
134 {
135 struct drm_mm_node *node;
136 unsigned long entries[STACKDEPTH];
137 char *buf;
138
139 buf = kmalloc(BUFSZ, M_DRM, GFP_KERNEL);
140 if (!buf)
141 return;
142
143 list_for_each_entry(node, drm_mm_nodes(mm), node_list) {
144 struct stack_trace trace = {
145 .entries = entries,
146 .max_entries = STACKDEPTH
147 };
148
149 if (!node->stack) {
150 DRM_ERROR("node [%08llx + %08llx]: unknown owner\n",
151 node->start, node->size);
152 continue;
153 }
154
155 depot_fetch_stack(node->stack, &trace);
156 snprint_stack_trace(buf, BUFSZ, &trace, 0);
157 DRM_ERROR("node [%08llx + %08llx]: inserted at\n%s",
158 node->start, node->size, buf);
159 }
160
161 kfree(buf);
162 }
163
164 #undef STACKDEPTH
165 #undef BUFSZ
166 #else
save_stack(struct drm_mm_node * node)167 static void save_stack(struct drm_mm_node *node) { }
show_leaks(struct drm_mm * mm)168 static void show_leaks(struct drm_mm *mm) { }
169 #endif
170
171 #define START(node) ((node)->start)
172 #define LAST(node) ((node)->start + (node)->size - 1)
173
174 #ifndef __DragonFly__
INTERVAL_TREE_DEFINE(struct drm_mm_node,rb,u64,__subtree_last,START,LAST,static inline,drm_mm_interval_tree)175 INTERVAL_TREE_DEFINE(struct drm_mm_node, rb,
176 u64, __subtree_last,
177 START, LAST, static inline, drm_mm_interval_tree)
178 #else
179 static struct drm_mm_node *
180 drm_mm_interval_tree_iter_first(struct rb_root *rb, u64 start, u64 last)
181 {
182 struct drm_mm *mm = container_of(rb, typeof(*mm), interval_tree);
183 struct drm_mm_node *node;
184
185 drm_mm_for_each_node(node, mm) {
186 if (LAST(node) >= start && START(node) <= last)
187 return node;
188 }
189 return NULL;
190 }
191 #endif
192
193 struct drm_mm_node *
194 __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last)
195 {
196 return drm_mm_interval_tree_iter_first((struct rb_root *)&mm->interval_tree,
197 start, last) ?: (struct drm_mm_node *)&mm->head_node;
198 }
199 EXPORT_SYMBOL(__drm_mm_interval_first);
200
drm_mm_insert_helper(struct drm_mm_node * hole_node,struct drm_mm_node * node,u64 size,u64 alignment,unsigned long color,u64 range_start,u64 range_end,enum drm_mm_allocator_flags flags)201 static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
202 struct drm_mm_node *node,
203 u64 size, u64 alignment,
204 unsigned long color,
205 u64 range_start, u64 range_end,
206 enum drm_mm_allocator_flags flags)
207 {
208 struct drm_mm *mm = hole_node->mm;
209 u64 hole_start = drm_mm_hole_node_start(hole_node);
210 u64 hole_end = drm_mm_hole_node_end(hole_node);
211 u64 adj_start = hole_start;
212 u64 adj_end = hole_end;
213
214 DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated);
215
216 if (mm->color_adjust)
217 mm->color_adjust(hole_node, color, &adj_start, &adj_end);
218
219 adj_start = max(adj_start, range_start);
220 adj_end = min(adj_end, range_end);
221
222 if (flags & DRM_MM_CREATE_TOP)
223 adj_start = adj_end - size;
224
225 if (alignment) {
226 u64 rem;
227
228 div64_u64_rem(adj_start, alignment, &rem);
229 if (rem) {
230 if (flags & DRM_MM_CREATE_TOP)
231 adj_start -= rem;
232 else
233 adj_start += alignment - rem;
234 }
235 }
236
237 if (adj_start == hole_start) {
238 hole_node->hole_follows = 0;
239 list_del(&hole_node->hole_stack);
240 }
241
242 node->start = adj_start;
243 node->size = size;
244 node->mm = mm;
245 node->color = color;
246 node->allocated = 1;
247
248 list_add(&node->node_list, &hole_node->node_list);
249
250 DRM_MM_BUG_ON(node->start < range_start);
251 DRM_MM_BUG_ON(node->start < adj_start);
252 DRM_MM_BUG_ON(node->start + node->size > adj_end);
253 DRM_MM_BUG_ON(node->start + node->size > range_end);
254
255 node->hole_follows = 0;
256 if (__drm_mm_hole_node_start(node) < hole_end) {
257 list_add(&node->hole_stack, &mm->hole_stack);
258 node->hole_follows = 1;
259 }
260
261 save_stack(node);
262 }
263
264 /**
265 * drm_mm_reserve_node - insert an pre-initialized node
266 * @mm: drm_mm allocator to insert @node into
267 * @node: drm_mm_node to insert
268 *
269 * This functions inserts an already set-up &drm_mm_node into the allocator,
270 * meaning that start, size and color must be set by the caller. All other
271 * fields must be cleared to 0. This is useful to initialize the allocator with
272 * preallocated objects which must be set-up before the range allocator can be
273 * set-up, e.g. when taking over a firmware framebuffer.
274 *
275 * Returns:
276 * 0 on success, -ENOSPC if there's no hole where @node is.
277 */
drm_mm_reserve_node(struct drm_mm * mm,struct drm_mm_node * node)278 int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node)
279 {
280 u64 end = node->start + node->size;
281 struct drm_mm_node *hole;
282 u64 hole_start, hole_end;
283 u64 adj_start, adj_end;
284
285 end = node->start + node->size;
286 if (unlikely(end <= node->start))
287 return -ENOSPC;
288
289 /* Find the relevant hole to add our node to */
290 hole = drm_mm_interval_tree_iter_first(&mm->interval_tree,
291 node->start, ~(u64)0);
292 if (hole) {
293 if (hole->start < end)
294 return -ENOSPC;
295 } else {
296 hole = list_entry(drm_mm_nodes(mm), typeof(*hole), node_list);
297 }
298
299 hole = list_last_entry(&hole->node_list, typeof(*hole), node_list);
300 if (!drm_mm_hole_follows(hole))
301 return -ENOSPC;
302
303 adj_start = hole_start = __drm_mm_hole_node_start(hole);
304 adj_end = hole_end = __drm_mm_hole_node_end(hole);
305
306 if (mm->color_adjust)
307 mm->color_adjust(hole, node->color, &adj_start, &adj_end);
308
309 if (adj_start > node->start || adj_end < end)
310 return -ENOSPC;
311
312 node->mm = mm;
313 node->allocated = 1;
314
315 list_add(&node->node_list, &hole->node_list);
316
317 #if 0
318 drm_mm_interval_tree_add_node(hole, node);
319 #endif
320
321 if (node->start == hole_start) {
322 hole->hole_follows = 0;
323 list_del(&hole->hole_stack);
324 }
325
326 node->hole_follows = 0;
327 if (end != hole_end) {
328 list_add(&node->hole_stack, &mm->hole_stack);
329 node->hole_follows = 1;
330 }
331
332 save_stack(node);
333
334 return 0;
335 }
336 EXPORT_SYMBOL(drm_mm_reserve_node);
337
338 /**
339 * drm_mm_insert_node_in_range_generic - ranged search for space and insert @node
340 * @mm: drm_mm to allocate from
341 * @node: preallocate node to insert
342 * @size: size of the allocation
343 * @alignment: alignment of the allocation
344 * @color: opaque tag value to use for this node
345 * @start: start of the allowed range for this node
346 * @end: end of the allowed range for this node
347 * @sflags: flags to fine-tune the allocation search
348 * @aflags: flags to fine-tune the allocation behavior
349 *
350 * The preallocated @node must be cleared to 0.
351 *
352 * Returns:
353 * 0 on success, -ENOSPC if there's no suitable hole.
354 */
drm_mm_insert_node_in_range_generic(struct drm_mm * mm,struct drm_mm_node * node,u64 size,u64 alignment,unsigned long color,u64 start,u64 end,enum drm_mm_search_flags sflags,enum drm_mm_allocator_flags aflags)355 int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *node,
356 u64 size, u64 alignment,
357 unsigned long color,
358 u64 start, u64 end,
359 enum drm_mm_search_flags sflags,
360 enum drm_mm_allocator_flags aflags)
361 {
362 struct drm_mm_node *hole_node;
363
364 if (WARN_ON(size == 0))
365 return -EINVAL;
366
367 hole_node = drm_mm_search_free_in_range_generic(mm,
368 size, alignment, color,
369 start, end, sflags);
370 if (!hole_node)
371 return -ENOSPC;
372
373 drm_mm_insert_helper(hole_node, node,
374 size, alignment, color,
375 start, end, aflags);
376 return 0;
377 }
378 EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic);
379
380 /**
381 * drm_mm_remove_node - Remove a memory node from the allocator.
382 * @node: drm_mm_node to remove
383 *
384 * This just removes a node from its drm_mm allocator. The node does not need to
385 * be cleared again before it can be re-inserted into this or any other drm_mm
386 * allocator. It is a bug to call this function on a unallocated node.
387 */
drm_mm_remove_node(struct drm_mm_node * node)388 void drm_mm_remove_node(struct drm_mm_node *node)
389 {
390 struct drm_mm *mm = node->mm;
391 struct drm_mm_node *prev_node;
392
393 if (drm_vma_debug & 2) {
394 drm_vma_debug &= ~2;
395 }
396
397 DRM_MM_BUG_ON(!node->allocated);
398 DRM_MM_BUG_ON(node->scanned_block);
399
400 prev_node =
401 list_entry(node->node_list.prev, struct drm_mm_node, node_list);
402
403 if (drm_mm_hole_follows(node)) {
404 DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) ==
405 __drm_mm_hole_node_end(node));
406 list_del(&node->hole_stack);
407 } else {
408 DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) !=
409 __drm_mm_hole_node_end(node));
410 }
411
412 if (!drm_mm_hole_follows(prev_node)) {
413 prev_node->hole_follows = 1;
414 list_add(&prev_node->hole_stack, &mm->hole_stack);
415 } else
416 list_move(&prev_node->hole_stack, &mm->hole_stack);
417
418 list_del(&node->node_list);
419 node->allocated = 0;
420 }
421 EXPORT_SYMBOL(drm_mm_remove_node);
422
check_free_hole(u64 start,u64 end,u64 size,u64 alignment)423 static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment)
424 {
425 if (end - start < size)
426 return 0;
427
428 if (alignment) {
429 u64 rem;
430
431 div64_u64_rem(start, alignment, &rem);
432 if (rem)
433 start += alignment - rem;
434 }
435
436 return end >= start + size;
437 }
438
drm_mm_search_free_in_range_generic(const struct drm_mm * mm,u64 size,u64 alignment,unsigned long color,u64 start,u64 end,enum drm_mm_search_flags flags)439 static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm,
440 u64 size,
441 u64 alignment,
442 unsigned long color,
443 u64 start,
444 u64 end,
445 enum drm_mm_search_flags flags)
446 {
447 struct drm_mm_node *entry;
448 struct drm_mm_node *best;
449 u64 adj_start;
450 u64 adj_end;
451 u64 best_size;
452
453 DRM_MM_BUG_ON(mm->scan_active);
454
455 best = NULL;
456 best_size = ~0UL;
457
458 __drm_mm_for_each_hole(entry, mm, adj_start, adj_end,
459 flags & DRM_MM_SEARCH_BELOW) {
460 u64 hole_size = adj_end - adj_start;
461
462 if (mm->color_adjust) {
463 mm->color_adjust(entry, color, &adj_start, &adj_end);
464 if (adj_end <= adj_start)
465 continue;
466 }
467
468 adj_start = max(adj_start, start);
469 adj_end = min(adj_end, end);
470
471 if (!check_free_hole(adj_start, adj_end, size, alignment))
472 continue;
473
474 if (!(flags & DRM_MM_SEARCH_BEST))
475 return entry;
476
477 if (hole_size < best_size) {
478 best = entry;
479 best_size = hole_size;
480 }
481 }
482
483 return best;
484 }
485
486 /**
487 * drm_mm_replace_node - move an allocation from @old to @new
488 * @old: drm_mm_node to remove from the allocator
489 * @new: drm_mm_node which should inherit @old's allocation
490 *
491 * This is useful for when drivers embed the drm_mm_node structure and hence
492 * can't move allocations by reassigning pointers. It's a combination of remove
493 * and insert with the guarantee that the allocation start will match.
494 */
drm_mm_replace_node(struct drm_mm_node * old,struct drm_mm_node * new)495 void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new)
496 {
497 DRM_MM_BUG_ON(!old->allocated);
498
499 list_replace(&old->node_list, &new->node_list);
500 list_replace(&old->hole_stack, &new->hole_stack);
501 new->hole_follows = old->hole_follows;
502 new->mm = old->mm;
503 new->start = old->start;
504 new->size = old->size;
505 new->color = old->color;
506
507 old->allocated = 0;
508 new->allocated = 1;
509 }
510 EXPORT_SYMBOL(drm_mm_replace_node);
511
512 /**
513 * DOC: lru scan roster
514 *
515 * Very often GPUs need to have continuous allocations for a given object. When
516 * evicting objects to make space for a new one it is therefore not most
517 * efficient when we simply start to select all objects from the tail of an LRU
518 * until there's a suitable hole: Especially for big objects or nodes that
519 * otherwise have special allocation constraints there's a good chance we evict
520 * lots of (smaller) objects unnecessarily.
521 *
522 * The DRM range allocator supports this use-case through the scanning
523 * interfaces. First a scan operation needs to be initialized with
524 * drm_mm_scan_init() or drm_mm_scan_init_with_range(). The driver adds
525 * objects to the roster, probably by walking an LRU list, but this can be
526 * freely implemented. Eviction candiates are added using
527 * drm_mm_scan_add_block() until a suitable hole is found or there are no
528 * further evictable objects. Eviction roster metadata is tracked in struct
529 * &drm_mm_scan.
530 *
531 * The driver must walk through all objects again in exactly the reverse
532 * order to restore the allocator state. Note that while the allocator is used
533 * in the scan mode no other operation is allowed.
534 *
535 * Finally the driver evicts all objects selected (drm_mm_scan_remove_block()
536 * reported true) in the scan, and any overlapping nodes after color adjustment
537 * (drm_mm_scan_color_evict()). Adding and removing an object is O(1), and
538 * since freeing a node is also O(1) the overall complexity is
539 * O(scanned_objects). So like the free stack which needs to be walked before a
540 * scan operation even begins this is linear in the number of objects. It
541 * doesn't seem to hurt too badly.
542 */
543
544 /**
545 * drm_mm_scan_init_with_range - initialize range-restricted lru scanning
546 * @scan: scan state
547 * @mm: drm_mm to scan
548 * @size: size of the allocation
549 * @alignment: alignment of the allocation
550 * @color: opaque tag value to use for the allocation
551 * @start: start of the allowed range for the allocation
552 * @end: end of the allowed range for the allocation
553 * @flags: flags to specify how the allocation will be performed afterwards
554 *
555 * This simply sets up the scanning routines with the parameters for the desired
556 * hole.
557 *
558 * Warning:
559 * As long as the scan list is non-empty, no other operations than
560 * adding/removing nodes to/from the scan list are allowed.
561 */
drm_mm_scan_init_with_range(struct drm_mm_scan * scan,struct drm_mm * mm,u64 size,u64 alignment,unsigned long color,u64 start,u64 end,unsigned int flags)562 void drm_mm_scan_init_with_range(struct drm_mm_scan *scan,
563 struct drm_mm *mm,
564 u64 size,
565 u64 alignment,
566 unsigned long color,
567 u64 start,
568 u64 end,
569 unsigned int flags)
570 {
571 DRM_MM_BUG_ON(start >= end);
572 DRM_MM_BUG_ON(!size || size > end - start);
573 DRM_MM_BUG_ON(mm->scan_active);
574
575 scan->mm = mm;
576
577 if (alignment <= 1)
578 alignment = 0;
579
580 scan->color = color;
581 scan->alignment = alignment;
582 scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0;
583 scan->size = size;
584 scan->flags = flags;
585
586 DRM_MM_BUG_ON(end <= start);
587 scan->range_start = start;
588 scan->range_end = end;
589
590 scan->hit_start = U64_MAX;
591 scan->hit_end = 0;
592 }
593 EXPORT_SYMBOL(drm_mm_scan_init_with_range);
594
595 /**
596 * drm_mm_scan_add_block - add a node to the scan list
597 * @scan: the active drm_mm scanner
598 * @node: drm_mm_node to add
599 *
600 * Add a node to the scan list that might be freed to make space for the desired
601 * hole.
602 *
603 * Returns:
604 * True if a hole has been found, false otherwise.
605 */
drm_mm_scan_add_block(struct drm_mm_scan * scan,struct drm_mm_node * node)606 bool drm_mm_scan_add_block(struct drm_mm_scan *scan,
607 struct drm_mm_node *node)
608 {
609 struct drm_mm *mm = scan->mm;
610 struct drm_mm_node *hole;
611 u64 hole_start, hole_end;
612 u64 col_start, col_end;
613 u64 adj_start, adj_end;
614
615 DRM_MM_BUG_ON(node->mm != mm);
616 DRM_MM_BUG_ON(!node->allocated);
617 DRM_MM_BUG_ON(node->scanned_block);
618 node->scanned_block = true;
619 mm->scan_active++;
620
621 /* Remove this block from the node_list so that we enlarge the hole
622 * (distance between the end of our previous node and the start of
623 * or next), without poisoning the link so that we can restore it
624 * later in drm_mm_scan_remove_block().
625 */
626 hole = list_prev_entry(node, node_list);
627 DRM_MM_BUG_ON(list_next_entry(hole, node_list) != node);
628 __list_del_entry(&node->node_list);
629
630 hole_start = __drm_mm_hole_node_start(hole);
631 hole_end = __drm_mm_hole_node_end(hole);
632
633 col_start = hole_start;
634 col_end = hole_end;
635 if (mm->color_adjust)
636 mm->color_adjust(hole, scan->color, &col_start, &col_end);
637
638 adj_start = max(col_start, scan->range_start);
639 adj_end = min(col_end, scan->range_end);
640 if (adj_end <= adj_start || adj_end - adj_start < scan->size)
641 return false;
642
643 if (scan->flags == DRM_MM_CREATE_TOP)
644 adj_start = adj_end - scan->size;
645
646 if (scan->alignment) {
647 u64 rem;
648
649 if (likely(scan->remainder_mask))
650 rem = adj_start & scan->remainder_mask;
651 else
652 div64_u64_rem(adj_start, scan->alignment, &rem);
653 if (rem) {
654 adj_start -= rem;
655 if (scan->flags != DRM_MM_CREATE_TOP)
656 adj_start += scan->alignment;
657 if (adj_start < max(col_start, scan->range_start) ||
658 min(col_end, scan->range_end) - adj_start < scan->size)
659 return false;
660
661 if (adj_end <= adj_start ||
662 adj_end - adj_start < scan->size)
663 return false;
664 }
665 }
666
667 scan->hit_start = adj_start;
668 scan->hit_end = adj_start + scan->size;
669
670 DRM_MM_BUG_ON(scan->hit_start >= scan->hit_end);
671 DRM_MM_BUG_ON(scan->hit_start < hole_start);
672 DRM_MM_BUG_ON(scan->hit_end > hole_end);
673
674 return true;
675 }
676 EXPORT_SYMBOL(drm_mm_scan_add_block);
677
678 /**
679 * drm_mm_scan_remove_block - remove a node from the scan list
680 * @scan: the active drm_mm scanner
681 * @node: drm_mm_node to remove
682 *
683 * Nodes **must** be removed in exactly the reverse order from the scan list as
684 * they have been added (e.g. using list_add() as they are added and then
685 * list_for_each() over that eviction list to remove), otherwise the internal
686 * state of the memory manager will be corrupted.
687 *
688 * When the scan list is empty, the selected memory nodes can be freed. An
689 * immediately following drm_mm_insert_node_in_range_generic() or one of the
690 * simpler versions of that function with !DRM_MM_SEARCH_BEST will then return
691 * the just freed block (because its at the top of the free_stack list).
692 *
693 * Returns:
694 * True if this block should be evicted, false otherwise. Will always
695 * return false when no hole has been found.
696 */
drm_mm_scan_remove_block(struct drm_mm_scan * scan,struct drm_mm_node * node)697 bool drm_mm_scan_remove_block(struct drm_mm_scan *scan,
698 struct drm_mm_node *node)
699 {
700 struct drm_mm_node *prev_node;
701
702 DRM_MM_BUG_ON(node->mm != scan->mm);
703 DRM_MM_BUG_ON(!node->scanned_block);
704 node->scanned_block = false;
705
706 DRM_MM_BUG_ON(!node->mm->scan_active);
707 node->mm->scan_active--;
708
709 /* During drm_mm_scan_add_block() we decoupled this node leaving
710 * its pointers intact. Now that the caller is walking back along
711 * the eviction list we can restore this block into its rightful
712 * place on the full node_list. To confirm that the caller is walking
713 * backwards correctly we check that prev_node->next == node->next,
714 * i.e. both believe the same node should be on the other side of the
715 * hole.
716 */
717 prev_node = list_prev_entry(node, node_list);
718 DRM_MM_BUG_ON(list_next_entry(prev_node, node_list) !=
719 list_next_entry(node, node_list));
720 list_add(&node->node_list, &prev_node->node_list);
721
722 return (node->start + node->size > scan->hit_start &&
723 node->start < scan->hit_end);
724 }
725 EXPORT_SYMBOL(drm_mm_scan_remove_block);
726
727 /**
728 * drm_mm_scan_color_evict - evict overlapping nodes on either side of hole
729 * @scan: drm_mm scan with target hole
730 *
731 * After completing an eviction scan and removing the selected nodes, we may
732 * need to remove a few more nodes from either side of the target hole if
733 * mm.color_adjust is being used.
734 *
735 * Returns:
736 * A node to evict, or NULL if there are no overlapping nodes.
737 */
drm_mm_scan_color_evict(struct drm_mm_scan * scan)738 struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan)
739 {
740 struct drm_mm *mm = scan->mm;
741 struct drm_mm_node *hole;
742 u64 hole_start, hole_end;
743
744 DRM_MM_BUG_ON(list_empty(&mm->hole_stack));
745
746 if (!mm->color_adjust)
747 return NULL;
748
749 hole = list_first_entry(&mm->hole_stack, typeof(*hole), hole_stack);
750 hole_start = __drm_mm_hole_node_start(hole);
751 hole_end = __drm_mm_hole_node_end(hole);
752
753 DRM_MM_BUG_ON(hole_start > scan->hit_start);
754 DRM_MM_BUG_ON(hole_end < scan->hit_end);
755
756 mm->color_adjust(hole, scan->color, &hole_start, &hole_end);
757 if (hole_start > scan->hit_start)
758 return hole;
759 if (hole_end < scan->hit_end)
760 return list_next_entry(hole, node_list);
761
762 return NULL;
763 }
764 EXPORT_SYMBOL(drm_mm_scan_color_evict);
765
766 /**
767 * drm_mm_init - initialize a drm-mm allocator
768 * @mm: the drm_mm structure to initialize
769 * @start: start of the range managed by @mm
770 * @size: end of the range managed by @mm
771 *
772 * Note that @mm must be cleared to 0 before calling this function.
773 */
drm_mm_init(struct drm_mm * mm,u64 start,u64 size)774 void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
775 {
776 DRM_MM_BUG_ON(start + size <= start);
777
778 INIT_LIST_HEAD(&mm->hole_stack);
779 mm->scan_active = 0;
780
781 /* Clever trick to avoid a special case in the free hole tracking. */
782 INIT_LIST_HEAD(&mm->head_node.node_list);
783 INIT_LIST_HEAD(&mm->head_node.hole_stack);
784 mm->head_node.allocated = 0;
785 mm->head_node.hole_follows = 1;
786 mm->head_node.mm = mm;
787 mm->head_node.start = start + size;
788 mm->head_node.size = start - mm->head_node.start;
789 list_add_tail(&mm->head_node.hole_stack, &mm->hole_stack);
790
791 mm->color_adjust = NULL;
792 }
793 EXPORT_SYMBOL(drm_mm_init);
794
795 /**
796 * drm_mm_takedown - clean up a drm_mm allocator
797 * @mm: drm_mm allocator to clean up
798 *
799 * Note that it is a bug to call this function on an allocator which is not
800 * clean.
801 */
drm_mm_takedown(struct drm_mm * mm)802 void drm_mm_takedown(struct drm_mm *mm)
803 {
804 if (WARN(!drm_mm_clean(mm),
805 "Memory manager not clean during takedown.\n"))
806 show_leaks(mm);
807 }
808 EXPORT_SYMBOL(drm_mm_takedown);
809
drm_mm_dump_hole(struct drm_printer * p,struct drm_mm_node * entry)810 static u64 drm_mm_dump_hole(struct drm_printer *p, struct drm_mm_node *entry)
811 {
812 u64 hole_start, hole_end, hole_size;
813
814 if (entry->hole_follows) {
815 hole_start = drm_mm_hole_node_start(entry);
816 hole_end = drm_mm_hole_node_end(entry);
817 hole_size = hole_end - hole_start;
818 drm_printf(p, "%#018llx-%#018llx: %llu: free\n", hole_start,
819 hole_end, hole_size);
820 return hole_size;
821 }
822
823 return 0;
824 }
825
826 /**
827 * drm_mm_print - print allocator state
828 * @mm: drm_mm allocator to print
829 * @p: DRM printer to use
830 */
drm_mm_print(struct drm_mm * mm,struct drm_printer * p)831 void drm_mm_print(struct drm_mm *mm, struct drm_printer *p)
832 {
833 struct drm_mm_node *entry;
834 u64 total_used = 0, total_free = 0, total = 0;
835
836 total_free += drm_mm_dump_hole(p, &mm->head_node);
837
838 drm_mm_for_each_node(entry, mm) {
839 drm_printf(p, "%#018llx-%#018llx: %llu: used\n", entry->start,
840 entry->start + entry->size, entry->size);
841 total_used += entry->size;
842 total_free += drm_mm_dump_hole(p, entry);
843 }
844 total = total_free + total_used;
845
846 drm_printf(p, "total: %llu, used %llu free %llu\n", total,
847 total_used, total_free);
848 }
849 EXPORT_SYMBOL(drm_mm_print);
850