]>
Commit | Line | Data |
---|---|---|
3a1bd924 TH |
1 | /************************************************************************** |
2 | * | |
3 | * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA. | |
ba004e39 | 4 | * Copyright 2016 Intel Corporation |
3a1bd924 TH |
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 | |
ba004e39 CW |
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. | |
3a1bd924 TH |
38 | * |
39 | * Aligned allocations can also see improvement. | |
40 | * | |
41 | * Authors: | |
96de0e25 | 42 | * Thomas Hellström <thomas-at-tungstengraphics-dot-com> |
3a1bd924 TH |
43 | */ |
44 | ||
760285e7 DH |
45 | #include <drm/drmP.h> |
46 | #include <drm/drm_mm.h> | |
1d58420b | 47 | #include <linux/slab.h> |
fa8a1238 | 48 | #include <linux/seq_file.h> |
2d1a8a48 | 49 | #include <linux/export.h> |
202b52b7 | 50 | #include <linux/interval_tree_generic.h> |
1d58420b | 51 | |
93110be6 DV |
52 | /** |
53 | * DOC: Overview | |
54 | * | |
55 | * drm_mm provides a simple range allocator. The drivers are free to use the | |
56 | * resource allocator from the linux core if it suits them, the upside of drm_mm | |
57 | * is that it's in the DRM core. Which means that it's easier to extend for | |
58 | * some of the crazier special purpose needs of gpus. | |
59 | * | |
60 | * The main data struct is &drm_mm, allocations are tracked in &drm_mm_node. | |
61 | * Drivers are free to embed either of them into their own suitable | |
62 | * datastructures. drm_mm itself will not do any allocations of its own, so if | |
63 | * drivers choose not to embed nodes they need to still allocate them | |
64 | * themselves. | |
65 | * | |
66 | * The range allocator also supports reservation of preallocated blocks. This is | |
67 | * useful for taking over initial mode setting configurations from the firmware, | |
68 | * where an object needs to be created which exactly matches the firmware's | |
69 | * scanout target. As long as the range is still free it can be inserted anytime | |
70 | * after the allocator is initialized, which helps with avoiding looped | |
ba004e39 | 71 | * dependencies in the driver load sequence. |
93110be6 DV |
72 | * |
73 | * drm_mm maintains a stack of most recently freed holes, which of all | |
74 | * simplistic datastructures seems to be a fairly decent approach to clustering | |
75 | * allocations and avoiding too much fragmentation. This means free space | |
76 | * searches are O(num_holes). Given that all the fancy features drm_mm supports | |
77 | * something better would be fairly complex and since gfx thrashing is a fairly | |
78 | * steep cliff not a real concern. Removing a node again is O(1). | |
79 | * | |
80 | * drm_mm supports a few features: Alignment and range restrictions can be | |
81 | * supplied. Further more every &drm_mm_node has a color value (which is just an | |
ba004e39 | 82 | * opaque unsigned long) which in conjunction with a driver callback can be used |
93110be6 DV |
83 | * to implement sophisticated placement restrictions. The i915 DRM driver uses |
84 | * this to implement guard pages between incompatible caching domains in the | |
85 | * graphics TT. | |
86 | * | |
ba004e39 CW |
87 | * Two behaviors are supported for searching and allocating: bottom-up and |
88 | * top-down. The default is bottom-up. Top-down allocation can be used if the | |
89 | * memory area has different restrictions, or just to reduce fragmentation. | |
62347f9e | 90 | * |
93110be6 DV |
91 | * Finally iteration helpers to walk all nodes and all holes are provided as are |
92 | * some basic allocator dumpers for debugging. | |
1c9bd1ed DV |
93 | * |
94 | * Note that this range allocator is not thread-safe, drivers need to protect | |
95 | * modifications with their on locking. The idea behind this is that for a full | |
96 | * memory manager additional data needs to be protected anyway, hence internal | |
97 | * locking would be fully redundant. | |
93110be6 DV |
98 | */ |
99 | ||
c700c67b | 100 | static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm, |
440fd528 | 101 | u64 size, |
71733207 | 102 | u64 alignment, |
c700c67b | 103 | unsigned long color, |
440fd528 TR |
104 | u64 start, |
105 | u64 end, | |
c700c67b | 106 | enum drm_mm_search_flags flags); |
1d58420b | 107 | |
5705670d | 108 | #ifdef CONFIG_DRM_DEBUG_MM |
93ce75fa CW |
109 | #include <linux/stackdepot.h> |
110 | ||
5705670d CW |
111 | #define STACKDEPTH 32 |
112 | #define BUFSZ 4096 | |
113 | ||
114 | static noinline void save_stack(struct drm_mm_node *node) | |
115 | { | |
116 | unsigned long entries[STACKDEPTH]; | |
117 | struct stack_trace trace = { | |
118 | .entries = entries, | |
119 | .max_entries = STACKDEPTH, | |
120 | .skip = 1 | |
121 | }; | |
122 | ||
123 | save_stack_trace(&trace); | |
124 | if (trace.nr_entries != 0 && | |
125 | trace.entries[trace.nr_entries-1] == ULONG_MAX) | |
126 | trace.nr_entries--; | |
127 | ||
128 | /* May be called under spinlock, so avoid sleeping */ | |
129 | node->stack = depot_save_stack(&trace, GFP_NOWAIT); | |
130 | } | |
131 | ||
132 | static void show_leaks(struct drm_mm *mm) | |
133 | { | |
134 | struct drm_mm_node *node; | |
135 | unsigned long entries[STACKDEPTH]; | |
136 | char *buf; | |
137 | ||
138 | buf = kmalloc(BUFSZ, GFP_KERNEL); | |
139 | if (!buf) | |
140 | return; | |
141 | ||
2bc98c86 | 142 | list_for_each_entry(node, drm_mm_nodes(mm), node_list) { |
5705670d CW |
143 | struct stack_trace trace = { |
144 | .entries = entries, | |
145 | .max_entries = STACKDEPTH | |
146 | }; | |
147 | ||
148 | if (!node->stack) { | |
149 | DRM_ERROR("node [%08llx + %08llx]: unknown owner\n", | |
150 | node->start, node->size); | |
151 | continue; | |
152 | } | |
153 | ||
154 | depot_fetch_stack(node->stack, &trace); | |
155 | snprint_stack_trace(buf, BUFSZ, &trace, 0); | |
156 | DRM_ERROR("node [%08llx + %08llx]: inserted at\n%s", | |
157 | node->start, node->size, buf); | |
158 | } | |
159 | ||
160 | kfree(buf); | |
161 | } | |
162 | ||
163 | #undef STACKDEPTH | |
164 | #undef BUFSZ | |
165 | #else | |
166 | static void save_stack(struct drm_mm_node *node) { } | |
167 | static void show_leaks(struct drm_mm *mm) { } | |
168 | #endif | |
169 | ||
202b52b7 CW |
170 | #define START(node) ((node)->start) |
171 | #define LAST(node) ((node)->start + (node)->size - 1) | |
172 | ||
173 | INTERVAL_TREE_DEFINE(struct drm_mm_node, rb, | |
174 | u64, __subtree_last, | |
175 | START, LAST, static inline, drm_mm_interval_tree) | |
176 | ||
177 | struct drm_mm_node * | |
45b186f1 | 178 | __drm_mm_interval_first(const struct drm_mm *mm, u64 start, u64 last) |
202b52b7 | 179 | { |
45b186f1 | 180 | return drm_mm_interval_tree_iter_first((struct rb_root *)&mm->interval_tree, |
202b52b7 CW |
181 | start, last); |
182 | } | |
522e85dd | 183 | EXPORT_SYMBOL(__drm_mm_interval_first); |
202b52b7 CW |
184 | |
185 | static void drm_mm_interval_tree_add_node(struct drm_mm_node *hole_node, | |
186 | struct drm_mm_node *node) | |
187 | { | |
188 | struct drm_mm *mm = hole_node->mm; | |
189 | struct rb_node **link, *rb; | |
190 | struct drm_mm_node *parent; | |
191 | ||
192 | node->__subtree_last = LAST(node); | |
193 | ||
194 | if (hole_node->allocated) { | |
195 | rb = &hole_node->rb; | |
196 | while (rb) { | |
197 | parent = rb_entry(rb, struct drm_mm_node, rb); | |
198 | if (parent->__subtree_last >= node->__subtree_last) | |
199 | break; | |
200 | ||
201 | parent->__subtree_last = node->__subtree_last; | |
202 | rb = rb_parent(rb); | |
203 | } | |
204 | ||
205 | rb = &hole_node->rb; | |
206 | link = &hole_node->rb.rb_right; | |
207 | } else { | |
208 | rb = NULL; | |
209 | link = &mm->interval_tree.rb_node; | |
210 | } | |
211 | ||
212 | while (*link) { | |
213 | rb = *link; | |
214 | parent = rb_entry(rb, struct drm_mm_node, rb); | |
215 | if (parent->__subtree_last < node->__subtree_last) | |
216 | parent->__subtree_last = node->__subtree_last; | |
217 | if (node->start < parent->start) | |
218 | link = &parent->rb.rb_left; | |
219 | else | |
220 | link = &parent->rb.rb_right; | |
221 | } | |
222 | ||
223 | rb_link_node(&node->rb, rb, link); | |
224 | rb_insert_augmented(&node->rb, | |
225 | &mm->interval_tree, | |
226 | &drm_mm_interval_tree_augment); | |
227 | } | |
228 | ||
9fc935de DV |
229 | static void drm_mm_insert_helper(struct drm_mm_node *hole_node, |
230 | struct drm_mm_node *node, | |
71733207 | 231 | u64 size, u64 alignment, |
62347f9e | 232 | unsigned long color, |
adb040b8 | 233 | u64 range_start, u64 range_end, |
62347f9e | 234 | enum drm_mm_allocator_flags flags) |
3a1bd924 | 235 | { |
ea7b1dd4 | 236 | struct drm_mm *mm = hole_node->mm; |
440fd528 TR |
237 | u64 hole_start = drm_mm_hole_node_start(hole_node); |
238 | u64 hole_end = drm_mm_hole_node_end(hole_node); | |
239 | u64 adj_start = hole_start; | |
240 | u64 adj_end = hole_end; | |
ea7b1dd4 | 241 | |
adb040b8 | 242 | DRM_MM_BUG_ON(!drm_mm_hole_follows(hole_node) || node->allocated); |
b0b7af18 | 243 | |
6b9d89b4 CW |
244 | if (mm->color_adjust) |
245 | mm->color_adjust(hole_node, color, &adj_start, &adj_end); | |
1d58420b | 246 | |
adb040b8 CW |
247 | adj_start = max(adj_start, range_start); |
248 | adj_end = min(adj_end, range_end); | |
249 | ||
62347f9e LK |
250 | if (flags & DRM_MM_CREATE_TOP) |
251 | adj_start = adj_end - size; | |
252 | ||
6b9d89b4 | 253 | if (alignment) { |
71733207 | 254 | u64 rem; |
440fd528 | 255 | |
71733207 | 256 | div64_u64_rem(adj_start, alignment, &rem); |
440fd528 | 257 | if (rem) { |
62347f9e | 258 | if (flags & DRM_MM_CREATE_TOP) |
440fd528 | 259 | adj_start -= rem; |
62347f9e | 260 | else |
440fd528 | 261 | adj_start += alignment - rem; |
62347f9e | 262 | } |
6b9d89b4 CW |
263 | } |
264 | ||
265 | if (adj_start == hole_start) { | |
ea7b1dd4 | 266 | hole_node->hole_follows = 0; |
6b9d89b4 CW |
267 | list_del(&hole_node->hole_stack); |
268 | } | |
ea7b1dd4 | 269 | |
6b9d89b4 | 270 | node->start = adj_start; |
ea7b1dd4 DV |
271 | node->size = size; |
272 | node->mm = mm; | |
6b9d89b4 | 273 | node->color = color; |
b0b7af18 | 274 | node->allocated = 1; |
3a1bd924 | 275 | |
ea7b1dd4 DV |
276 | list_add(&node->node_list, &hole_node->node_list); |
277 | ||
202b52b7 CW |
278 | drm_mm_interval_tree_add_node(hole_node, node); |
279 | ||
adb040b8 CW |
280 | DRM_MM_BUG_ON(node->start < range_start); |
281 | DRM_MM_BUG_ON(node->start < adj_start); | |
b3ee963f | 282 | DRM_MM_BUG_ON(node->start + node->size > adj_end); |
adb040b8 | 283 | DRM_MM_BUG_ON(node->start + node->size > range_end); |
ea7b1dd4 | 284 | |
6b9d89b4 | 285 | node->hole_follows = 0; |
9e8944ab | 286 | if (__drm_mm_hole_node_start(node) < hole_end) { |
ea7b1dd4 DV |
287 | list_add(&node->hole_stack, &mm->hole_stack); |
288 | node->hole_follows = 1; | |
1d58420b | 289 | } |
5705670d CW |
290 | |
291 | save_stack(node); | |
9fc935de DV |
292 | } |
293 | ||
e18c0412 DV |
294 | /** |
295 | * drm_mm_reserve_node - insert an pre-initialized node | |
296 | * @mm: drm_mm allocator to insert @node into | |
297 | * @node: drm_mm_node to insert | |
298 | * | |
299 | * This functions inserts an already set-up drm_mm_node into the allocator, | |
300 | * meaning that start, size and color must be set by the caller. This is useful | |
301 | * to initialize the allocator with preallocated objects which must be set-up | |
302 | * before the range allocator can be set-up, e.g. when taking over a firmware | |
303 | * framebuffer. | |
304 | * | |
305 | * Returns: | |
306 | * 0 on success, -ENOSPC if there's no hole where @node is. | |
307 | */ | |
338710e7 | 308 | int drm_mm_reserve_node(struct drm_mm *mm, struct drm_mm_node *node) |
5973c7ee | 309 | { |
202b52b7 | 310 | u64 end = node->start + node->size; |
b3a070cc | 311 | struct drm_mm_node *hole; |
202b52b7 | 312 | u64 hole_start, hole_end; |
2db86dfc | 313 | u64 adj_start, adj_end; |
338710e7 | 314 | |
b80d3942 | 315 | end = node->start + node->size; |
c820186d CW |
316 | if (unlikely(end <= node->start)) |
317 | return -ENOSPC; | |
b80d3942 | 318 | |
338710e7 | 319 | /* Find the relevant hole to add our node to */ |
202b52b7 CW |
320 | hole = drm_mm_interval_tree_iter_first(&mm->interval_tree, |
321 | node->start, ~(u64)0); | |
322 | if (hole) { | |
323 | if (hole->start < end) | |
324 | return -ENOSPC; | |
325 | } else { | |
2bc98c86 | 326 | hole = list_entry(drm_mm_nodes(mm), typeof(*hole), node_list); |
202b52b7 | 327 | } |
5973c7ee | 328 | |
202b52b7 | 329 | hole = list_last_entry(&hole->node_list, typeof(*hole), node_list); |
3f85fb34 | 330 | if (!drm_mm_hole_follows(hole)) |
202b52b7 | 331 | return -ENOSPC; |
5973c7ee | 332 | |
2db86dfc CW |
333 | adj_start = hole_start = __drm_mm_hole_node_start(hole); |
334 | adj_end = hole_end = __drm_mm_hole_node_end(hole); | |
335 | ||
336 | if (mm->color_adjust) | |
337 | mm->color_adjust(hole, node->color, &adj_start, &adj_end); | |
338 | ||
339 | if (adj_start > node->start || adj_end < end) | |
202b52b7 | 340 | return -ENOSPC; |
5973c7ee | 341 | |
202b52b7 CW |
342 | node->mm = mm; |
343 | node->allocated = 1; | |
5973c7ee | 344 | |
202b52b7 | 345 | list_add(&node->node_list, &hole->node_list); |
5973c7ee | 346 | |
202b52b7 CW |
347 | drm_mm_interval_tree_add_node(hole, node); |
348 | ||
349 | if (node->start == hole_start) { | |
350 | hole->hole_follows = 0; | |
a7879005 | 351 | list_del(&hole->hole_stack); |
202b52b7 CW |
352 | } |
353 | ||
354 | node->hole_follows = 0; | |
355 | if (end != hole_end) { | |
356 | list_add(&node->hole_stack, &mm->hole_stack); | |
357 | node->hole_follows = 1; | |
5973c7ee CW |
358 | } |
359 | ||
5705670d CW |
360 | save_stack(node); |
361 | ||
202b52b7 | 362 | return 0; |
5973c7ee | 363 | } |
338710e7 | 364 | EXPORT_SYMBOL(drm_mm_reserve_node); |
5973c7ee | 365 | |
b0b7af18 | 366 | /** |
e18c0412 DV |
367 | * drm_mm_insert_node_in_range_generic - ranged search for space and insert @node |
368 | * @mm: drm_mm to allocate from | |
369 | * @node: preallocate node to insert | |
370 | * @size: size of the allocation | |
371 | * @alignment: alignment of the allocation | |
372 | * @color: opaque tag value to use for this node | |
373 | * @start: start of the allowed range for this node | |
374 | * @end: end of the allowed range for this node | |
62347f9e LK |
375 | * @sflags: flags to fine-tune the allocation search |
376 | * @aflags: flags to fine-tune the allocation behavior | |
e18c0412 DV |
377 | * |
378 | * The preallocated node must be cleared to 0. | |
379 | * | |
380 | * Returns: | |
381 | * 0 on success, -ENOSPC if there's no suitable hole. | |
3a1bd924 | 382 | */ |
b8103450 | 383 | int drm_mm_insert_node_in_range_generic(struct drm_mm *mm, struct drm_mm_node *node, |
71733207 | 384 | u64 size, u64 alignment, |
62347f9e | 385 | unsigned long color, |
440fd528 | 386 | u64 start, u64 end, |
62347f9e LK |
387 | enum drm_mm_search_flags sflags, |
388 | enum drm_mm_allocator_flags aflags) | |
3a1bd924 | 389 | { |
b0b7af18 DV |
390 | struct drm_mm_node *hole_node; |
391 | ||
aafdcfd3 CW |
392 | if (WARN_ON(size == 0)) |
393 | return -EINVAL; | |
394 | ||
b8103450 CW |
395 | hole_node = drm_mm_search_free_in_range_generic(mm, |
396 | size, alignment, color, | |
62347f9e | 397 | start, end, sflags); |
b0b7af18 DV |
398 | if (!hole_node) |
399 | return -ENOSPC; | |
400 | ||
adb040b8 CW |
401 | drm_mm_insert_helper(hole_node, node, |
402 | size, alignment, color, | |
403 | start, end, aflags); | |
b0b7af18 DV |
404 | return 0; |
405 | } | |
b8103450 CW |
406 | EXPORT_SYMBOL(drm_mm_insert_node_in_range_generic); |
407 | ||
b0b7af18 | 408 | /** |
e18c0412 DV |
409 | * drm_mm_remove_node - Remove a memory node from the allocator. |
410 | * @node: drm_mm_node to remove | |
411 | * | |
412 | * This just removes a node from its drm_mm allocator. The node does not need to | |
413 | * be cleared again before it can be re-inserted into this or any other drm_mm | |
ba004e39 | 414 | * allocator. It is a bug to call this function on a unallocated node. |
b0b7af18 DV |
415 | */ |
416 | void drm_mm_remove_node(struct drm_mm_node *node) | |
417 | { | |
ea7b1dd4 DV |
418 | struct drm_mm *mm = node->mm; |
419 | struct drm_mm_node *prev_node; | |
3a1bd924 | 420 | |
b3ee963f | 421 | DRM_MM_BUG_ON(!node->allocated); |
f29051f1 | 422 | DRM_MM_BUG_ON(node->scanned_block); |
3a1bd924 | 423 | |
ea7b1dd4 DV |
424 | prev_node = |
425 | list_entry(node->node_list.prev, struct drm_mm_node, node_list); | |
709ea971 | 426 | |
3f85fb34 | 427 | if (drm_mm_hole_follows(node)) { |
b3ee963f CW |
428 | DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) == |
429 | __drm_mm_hole_node_end(node)); | |
ea7b1dd4 | 430 | list_del(&node->hole_stack); |
3f85fb34 | 431 | } else { |
b3ee963f CW |
432 | DRM_MM_BUG_ON(__drm_mm_hole_node_start(node) != |
433 | __drm_mm_hole_node_end(node)); | |
3f85fb34 | 434 | } |
9e8944ab | 435 | |
3f85fb34 | 436 | if (!drm_mm_hole_follows(prev_node)) { |
ea7b1dd4 DV |
437 | prev_node->hole_follows = 1; |
438 | list_add(&prev_node->hole_stack, &mm->hole_stack); | |
439 | } else | |
440 | list_move(&prev_node->hole_stack, &mm->hole_stack); | |
441 | ||
202b52b7 | 442 | drm_mm_interval_tree_remove(node, &mm->interval_tree); |
ea7b1dd4 | 443 | list_del(&node->node_list); |
b0b7af18 DV |
444 | node->allocated = 0; |
445 | } | |
446 | EXPORT_SYMBOL(drm_mm_remove_node); | |
447 | ||
71733207 | 448 | static int check_free_hole(u64 start, u64 end, u64 size, u64 alignment) |
7a6b2896 | 449 | { |
75214733 | 450 | if (end - start < size) |
7a6b2896 DV |
451 | return 0; |
452 | ||
453 | if (alignment) { | |
71733207 | 454 | u64 rem; |
440fd528 | 455 | |
71733207 | 456 | div64_u64_rem(start, alignment, &rem); |
046d669c | 457 | if (rem) |
440fd528 | 458 | start += alignment - rem; |
7a6b2896 DV |
459 | } |
460 | ||
6b9d89b4 | 461 | return end >= start + size; |
7a6b2896 DV |
462 | } |
463 | ||
c700c67b | 464 | static struct drm_mm_node *drm_mm_search_free_in_range_generic(const struct drm_mm *mm, |
440fd528 | 465 | u64 size, |
71733207 | 466 | u64 alignment, |
6b9d89b4 | 467 | unsigned long color, |
440fd528 TR |
468 | u64 start, |
469 | u64 end, | |
31e5d7c6 | 470 | enum drm_mm_search_flags flags) |
a2e68e92 | 471 | { |
a2e68e92 JG |
472 | struct drm_mm_node *entry; |
473 | struct drm_mm_node *best; | |
440fd528 TR |
474 | u64 adj_start; |
475 | u64 adj_end; | |
476 | u64 best_size; | |
a2e68e92 | 477 | |
9a71e277 | 478 | DRM_MM_BUG_ON(mm->scan_active); |
709ea971 | 479 | |
a2e68e92 JG |
480 | best = NULL; |
481 | best_size = ~0UL; | |
482 | ||
62347f9e LK |
483 | __drm_mm_for_each_hole(entry, mm, adj_start, adj_end, |
484 | flags & DRM_MM_SEARCH_BELOW) { | |
440fd528 | 485 | u64 hole_size = adj_end - adj_start; |
145bccd2 | 486 | |
6b9d89b4 CW |
487 | if (mm->color_adjust) { |
488 | mm->color_adjust(entry, color, &adj_start, &adj_end); | |
489 | if (adj_end <= adj_start) | |
490 | continue; | |
491 | } | |
492 | ||
3db93756 CW |
493 | adj_start = max(adj_start, start); |
494 | adj_end = min(adj_end, end); | |
495 | ||
75214733 | 496 | if (!check_free_hole(adj_start, adj_end, size, alignment)) |
a2e68e92 JG |
497 | continue; |
498 | ||
31e5d7c6 | 499 | if (!(flags & DRM_MM_SEARCH_BEST)) |
7a6b2896 | 500 | return entry; |
a2e68e92 | 501 | |
145bccd2 | 502 | if (hole_size < best_size) { |
7a6b2896 | 503 | best = entry; |
145bccd2 | 504 | best_size = hole_size; |
a2e68e92 JG |
505 | } |
506 | } | |
507 | ||
508 | return best; | |
509 | } | |
a2e68e92 | 510 | |
b0b7af18 | 511 | /** |
e18c0412 DV |
512 | * drm_mm_replace_node - move an allocation from @old to @new |
513 | * @old: drm_mm_node to remove from the allocator | |
514 | * @new: drm_mm_node which should inherit @old's allocation | |
515 | * | |
516 | * This is useful for when drivers embed the drm_mm_node structure and hence | |
517 | * can't move allocations by reassigning pointers. It's a combination of remove | |
518 | * and insert with the guarantee that the allocation start will match. | |
b0b7af18 DV |
519 | */ |
520 | void drm_mm_replace_node(struct drm_mm_node *old, struct drm_mm_node *new) | |
521 | { | |
b3ee963f CW |
522 | DRM_MM_BUG_ON(!old->allocated); |
523 | ||
b0b7af18 | 524 | list_replace(&old->node_list, &new->node_list); |
2bbd4492 | 525 | list_replace(&old->hole_stack, &new->hole_stack); |
202b52b7 | 526 | rb_replace_node(&old->rb, &new->rb, &old->mm->interval_tree); |
b0b7af18 DV |
527 | new->hole_follows = old->hole_follows; |
528 | new->mm = old->mm; | |
529 | new->start = old->start; | |
530 | new->size = old->size; | |
6b9d89b4 | 531 | new->color = old->color; |
202b52b7 | 532 | new->__subtree_last = old->__subtree_last; |
b0b7af18 DV |
533 | |
534 | old->allocated = 0; | |
535 | new->allocated = 1; | |
536 | } | |
537 | EXPORT_SYMBOL(drm_mm_replace_node); | |
538 | ||
93110be6 DV |
539 | /** |
540 | * DOC: lru scan roaster | |
541 | * | |
542 | * Very often GPUs need to have continuous allocations for a given object. When | |
543 | * evicting objects to make space for a new one it is therefore not most | |
544 | * efficient when we simply start to select all objects from the tail of an LRU | |
545 | * until there's a suitable hole: Especially for big objects or nodes that | |
546 | * otherwise have special allocation constraints there's a good chance we evict | |
ba004e39 | 547 | * lots of (smaller) objects unnecessarily. |
93110be6 DV |
548 | * |
549 | * The DRM range allocator supports this use-case through the scanning | |
550 | * interfaces. First a scan operation needs to be initialized with | |
9a71e277 | 551 | * drm_mm_scan_init() or drm_mm_scan_init_with_range(). The driver adds |
3fa489da CW |
552 | * objects to the roster (probably by walking an LRU list, but this can be |
553 | * freely implemented) (using drm_mm_scan_add_block()) until a suitable hole | |
554 | * is found or there are no further evictable objects. | |
93110be6 | 555 | * |
ba004e39 | 556 | * The driver must walk through all objects again in exactly the reverse |
93110be6 DV |
557 | * order to restore the allocator state. Note that while the allocator is used |
558 | * in the scan mode no other operation is allowed. | |
559 | * | |
3fa489da CW |
560 | * Finally the driver evicts all objects selected (drm_mm_scan_remove_block() |
561 | * reported true) in the scan, and any overlapping nodes after color adjustment | |
562 | * (drm_mm_scan_evict_color()). Adding and removing an object is O(1), and | |
563 | * since freeing a node is also O(1) the overall complexity is | |
564 | * O(scanned_objects). So like the free stack which needs to be walked before a | |
565 | * scan operation even begins this is linear in the number of objects. It | |
566 | * doesn't seem to hurt too badly. | |
93110be6 DV |
567 | */ |
568 | ||
d935cc61 | 569 | /** |
9a71e277 CW |
570 | * drm_mm_scan_init_with_range - initialize range-restricted lru scanning |
571 | * @scan: scan state | |
e18c0412 DV |
572 | * @mm: drm_mm to scan |
573 | * @size: size of the allocation | |
574 | * @alignment: alignment of the allocation | |
575 | * @color: opaque tag value to use for the allocation | |
576 | * @start: start of the allowed range for the allocation | |
577 | * @end: end of the allowed range for the allocation | |
0b04d474 | 578 | * @flags: flags to specify how the allocation will be performed afterwards |
d935cc61 DV |
579 | * |
580 | * This simply sets up the scanning routines with the parameters for the desired | |
0b04d474 | 581 | * hole. |
d935cc61 | 582 | * |
e18c0412 DV |
583 | * Warning: |
584 | * As long as the scan list is non-empty, no other operations than | |
d935cc61 DV |
585 | * adding/removing nodes to/from the scan list are allowed. |
586 | */ | |
9a71e277 CW |
587 | void drm_mm_scan_init_with_range(struct drm_mm_scan *scan, |
588 | struct drm_mm *mm, | |
440fd528 | 589 | u64 size, |
71733207 | 590 | u64 alignment, |
6b9d89b4 | 591 | unsigned long color, |
440fd528 | 592 | u64 start, |
0b04d474 CW |
593 | u64 end, |
594 | unsigned int flags) | |
d935cc61 | 595 | { |
6259a56b CW |
596 | DRM_MM_BUG_ON(start >= end); |
597 | DRM_MM_BUG_ON(!size || size > end - start); | |
9a71e277 CW |
598 | DRM_MM_BUG_ON(mm->scan_active); |
599 | ||
600 | scan->mm = mm; | |
601 | ||
9a956b15 CW |
602 | if (alignment <= 1) |
603 | alignment = 0; | |
604 | ||
9a71e277 CW |
605 | scan->color = color; |
606 | scan->alignment = alignment; | |
9a956b15 | 607 | scan->remainder_mask = is_power_of_2(alignment) ? alignment - 1 : 0; |
9a71e277 | 608 | scan->size = size; |
0b04d474 | 609 | scan->flags = flags; |
9a71e277 CW |
610 | |
611 | DRM_MM_BUG_ON(end <= start); | |
612 | scan->range_start = start; | |
613 | scan->range_end = end; | |
6259a56b | 614 | |
9a71e277 CW |
615 | scan->hit_start = U64_MAX; |
616 | scan->hit_end = 0; | |
d935cc61 | 617 | } |
9a71e277 | 618 | EXPORT_SYMBOL(drm_mm_scan_init_with_range); |
d935cc61 | 619 | |
709ea971 | 620 | /** |
e18c0412 | 621 | * drm_mm_scan_add_block - add a node to the scan list |
9b8b75de | 622 | * @scan: the active drm_mm scanner |
e18c0412 DV |
623 | * @node: drm_mm_node to add |
624 | * | |
709ea971 DV |
625 | * Add a node to the scan list that might be freed to make space for the desired |
626 | * hole. | |
627 | * | |
e18c0412 DV |
628 | * Returns: |
629 | * True if a hole has been found, false otherwise. | |
709ea971 | 630 | */ |
9a71e277 CW |
631 | bool drm_mm_scan_add_block(struct drm_mm_scan *scan, |
632 | struct drm_mm_node *node) | |
709ea971 | 633 | { |
9a71e277 | 634 | struct drm_mm *mm = scan->mm; |
4a6c156f | 635 | struct drm_mm_node *hole; |
440fd528 | 636 | u64 hole_start, hole_end; |
268c6498 | 637 | u64 col_start, col_end; |
440fd528 | 638 | u64 adj_start, adj_end; |
709ea971 | 639 | |
9a71e277 CW |
640 | DRM_MM_BUG_ON(node->mm != mm); |
641 | DRM_MM_BUG_ON(!node->allocated); | |
b3ee963f | 642 | DRM_MM_BUG_ON(node->scanned_block); |
0b04d474 | 643 | node->scanned_block = true; |
9a71e277 | 644 | mm->scan_active++; |
709ea971 | 645 | |
f29051f1 CW |
646 | /* Remove this block from the node_list so that we enlarge the hole |
647 | * (distance between the end of our previous node and the start of | |
648 | * or next), without poisoning the link so that we can restore it | |
649 | * later in drm_mm_scan_remove_block(). | |
650 | */ | |
4a6c156f | 651 | hole = list_prev_entry(node, node_list); |
f29051f1 CW |
652 | DRM_MM_BUG_ON(list_next_entry(hole, node_list) != node); |
653 | __list_del_entry(&node->node_list); | |
709ea971 | 654 | |
268c6498 CW |
655 | hole_start = __drm_mm_hole_node_start(hole); |
656 | hole_end = __drm_mm_hole_node_end(hole); | |
d935cc61 | 657 | |
268c6498 CW |
658 | col_start = hole_start; |
659 | col_end = hole_end; | |
901593f2 | 660 | if (mm->color_adjust) |
268c6498 CW |
661 | mm->color_adjust(hole, scan->color, &col_start, &col_end); |
662 | ||
663 | adj_start = max(col_start, scan->range_start); | |
664 | adj_end = min(col_end, scan->range_end); | |
0b04d474 CW |
665 | if (adj_end <= adj_start || adj_end - adj_start < scan->size) |
666 | return false; | |
667 | ||
668 | if (scan->flags == DRM_MM_CREATE_TOP) | |
669 | adj_start = adj_end - scan->size; | |
670 | ||
671 | if (scan->alignment) { | |
672 | u64 rem; | |
673 | ||
9a956b15 CW |
674 | if (likely(scan->remainder_mask)) |
675 | rem = adj_start & scan->remainder_mask; | |
676 | else | |
677 | div64_u64_rem(adj_start, scan->alignment, &rem); | |
0b04d474 CW |
678 | if (rem) { |
679 | adj_start -= rem; | |
680 | if (scan->flags != DRM_MM_CREATE_TOP) | |
681 | adj_start += scan->alignment; | |
682 | if (adj_start < max(col_start, scan->range_start) || | |
683 | min(col_end, scan->range_end) - adj_start < scan->size) | |
684 | return false; | |
685 | ||
686 | if (adj_end <= adj_start || | |
687 | adj_end - adj_start < scan->size) | |
688 | return false; | |
689 | } | |
690 | } | |
901593f2 | 691 | |
3fa489da CW |
692 | scan->hit_start = adj_start; |
693 | scan->hit_end = adj_start + scan->size; | |
709ea971 | 694 | |
0b04d474 CW |
695 | DRM_MM_BUG_ON(scan->hit_start >= scan->hit_end); |
696 | DRM_MM_BUG_ON(scan->hit_start < hole_start); | |
697 | DRM_MM_BUG_ON(scan->hit_end > hole_end); | |
698 | ||
699 | return true; | |
709ea971 DV |
700 | } |
701 | EXPORT_SYMBOL(drm_mm_scan_add_block); | |
702 | ||
703 | /** | |
e18c0412 | 704 | * drm_mm_scan_remove_block - remove a node from the scan list |
9b8b75de | 705 | * @scan: the active drm_mm scanner |
e18c0412 | 706 | * @node: drm_mm_node to remove |
709ea971 | 707 | * |
ba004e39 CW |
708 | * Nodes _must_ be removed in exactly the reverse order from the scan list as |
709 | * they have been added (e.g. using list_add as they are added and then | |
710 | * list_for_each over that eviction list to remove), otherwise the internal | |
711 | * state of the memory manager will be corrupted. | |
709ea971 DV |
712 | * |
713 | * When the scan list is empty, the selected memory nodes can be freed. An | |
31e5d7c6 DH |
714 | * immediately following drm_mm_search_free with !DRM_MM_SEARCH_BEST will then |
715 | * return the just freed block (because its at the top of the free_stack list). | |
709ea971 | 716 | * |
e18c0412 DV |
717 | * Returns: |
718 | * True if this block should be evicted, false otherwise. Will always | |
719 | * return false when no hole has been found. | |
709ea971 | 720 | */ |
9a71e277 CW |
721 | bool drm_mm_scan_remove_block(struct drm_mm_scan *scan, |
722 | struct drm_mm_node *node) | |
709ea971 | 723 | { |
ea7b1dd4 | 724 | struct drm_mm_node *prev_node; |
709ea971 | 725 | |
9a71e277 | 726 | DRM_MM_BUG_ON(node->mm != scan->mm); |
b3ee963f | 727 | DRM_MM_BUG_ON(!node->scanned_block); |
0b04d474 | 728 | node->scanned_block = false; |
709ea971 | 729 | |
9a71e277 CW |
730 | DRM_MM_BUG_ON(!node->mm->scan_active); |
731 | node->mm->scan_active--; | |
732 | ||
f29051f1 CW |
733 | /* During drm_mm_scan_add_block() we decoupled this node leaving |
734 | * its pointers intact. Now that the caller is walking back along | |
735 | * the eviction list we can restore this block into its rightful | |
736 | * place on the full node_list. To confirm that the caller is walking | |
737 | * backwards correctly we check that prev_node->next == node->next, | |
738 | * i.e. both believe the same node should be on the other side of the | |
739 | * hole. | |
740 | */ | |
9a71e277 | 741 | prev_node = list_prev_entry(node, node_list); |
f29051f1 CW |
742 | DRM_MM_BUG_ON(list_next_entry(prev_node, node_list) != |
743 | list_next_entry(node, node_list)); | |
ea7b1dd4 | 744 | list_add(&node->node_list, &prev_node->node_list); |
709ea971 | 745 | |
0b04d474 | 746 | return (node->start + node->size > scan->hit_start && |
9a71e277 | 747 | node->start < scan->hit_end); |
709ea971 DV |
748 | } |
749 | EXPORT_SYMBOL(drm_mm_scan_remove_block); | |
750 | ||
3fa489da CW |
751 | /** |
752 | * drm_mm_scan_color_evict - evict overlapping nodes on either side of hole | |
753 | * @scan: drm_mm scan with target hole | |
754 | * | |
755 | * After completing an eviction scan and removing the selected nodes, we may | |
756 | * need to remove a few more nodes from either side of the target hole if | |
757 | * mm.color_adjust is being used. | |
758 | * | |
759 | * Returns: | |
760 | * A node to evict, or NULL if there are no overlapping nodes. | |
761 | */ | |
762 | struct drm_mm_node *drm_mm_scan_color_evict(struct drm_mm_scan *scan) | |
763 | { | |
764 | struct drm_mm *mm = scan->mm; | |
765 | struct drm_mm_node *hole; | |
766 | u64 hole_start, hole_end; | |
767 | ||
768 | DRM_MM_BUG_ON(list_empty(&mm->hole_stack)); | |
769 | ||
770 | if (!mm->color_adjust) | |
771 | return NULL; | |
772 | ||
773 | hole = list_first_entry(&mm->hole_stack, typeof(*hole), hole_stack); | |
774 | hole_start = __drm_mm_hole_node_start(hole); | |
775 | hole_end = __drm_mm_hole_node_end(hole); | |
776 | ||
777 | DRM_MM_BUG_ON(hole_start > scan->hit_start); | |
778 | DRM_MM_BUG_ON(hole_end < scan->hit_end); | |
779 | ||
780 | mm->color_adjust(hole, scan->color, &hole_start, &hole_end); | |
781 | if (hole_start > scan->hit_start) | |
782 | return hole; | |
783 | if (hole_end < scan->hit_end) | |
784 | return list_next_entry(hole, node_list); | |
785 | ||
786 | return NULL; | |
787 | } | |
788 | EXPORT_SYMBOL(drm_mm_scan_color_evict); | |
789 | ||
e18c0412 DV |
790 | /** |
791 | * drm_mm_init - initialize a drm-mm allocator | |
792 | * @mm: the drm_mm structure to initialize | |
793 | * @start: start of the range managed by @mm | |
794 | * @size: end of the range managed by @mm | |
795 | * | |
796 | * Note that @mm must be cleared to 0 before calling this function. | |
797 | */ | |
45b186f1 | 798 | void drm_mm_init(struct drm_mm *mm, u64 start, u64 size) |
1d58420b | 799 | { |
6259a56b CW |
800 | DRM_MM_BUG_ON(start + size <= start); |
801 | ||
ea7b1dd4 | 802 | INIT_LIST_HEAD(&mm->hole_stack); |
9a71e277 | 803 | mm->scan_active = 0; |
3a1bd924 | 804 | |
ea7b1dd4 DV |
805 | /* Clever trick to avoid a special case in the free hole tracking. */ |
806 | INIT_LIST_HEAD(&mm->head_node.node_list); | |
cc98e6ce | 807 | mm->head_node.allocated = 0; |
ea7b1dd4 | 808 | mm->head_node.hole_follows = 1; |
ea7b1dd4 DV |
809 | mm->head_node.mm = mm; |
810 | mm->head_node.start = start + size; | |
811 | mm->head_node.size = start - mm->head_node.start; | |
812 | list_add_tail(&mm->head_node.hole_stack, &mm->hole_stack); | |
813 | ||
202b52b7 CW |
814 | mm->interval_tree = RB_ROOT; |
815 | ||
6b9d89b4 | 816 | mm->color_adjust = NULL; |
3a1bd924 | 817 | } |
673a394b | 818 | EXPORT_SYMBOL(drm_mm_init); |
3a1bd924 | 819 | |
e18c0412 DV |
820 | /** |
821 | * drm_mm_takedown - clean up a drm_mm allocator | |
822 | * @mm: drm_mm allocator to clean up | |
823 | * | |
824 | * Note that it is a bug to call this function on an allocator which is not | |
825 | * clean. | |
826 | */ | |
5705670d | 827 | void drm_mm_takedown(struct drm_mm *mm) |
3a1bd924 | 828 | { |
ac9bb7b7 | 829 | if (WARN(!drm_mm_clean(mm), |
5705670d CW |
830 | "Memory manager not clean during takedown.\n")) |
831 | show_leaks(mm); | |
3a1bd924 | 832 | } |
f453ba04 | 833 | EXPORT_SYMBOL(drm_mm_takedown); |
fa8a1238 | 834 | |
b5c3714f | 835 | static u64 drm_mm_dump_hole(struct drm_printer *p, const struct drm_mm_node *entry) |
99d7e48e | 836 | { |
440fd528 | 837 | u64 hole_start, hole_end, hole_size; |
ea7b1dd4 | 838 | |
2c54b133 DV |
839 | if (entry->hole_follows) { |
840 | hole_start = drm_mm_hole_node_start(entry); | |
841 | hole_end = drm_mm_hole_node_end(entry); | |
842 | hole_size = hole_end - hole_start; | |
b5c3714f | 843 | drm_printf(p, "%#018llx-%#018llx: %llu: free\n", hole_start, |
440fd528 | 844 | hole_end, hole_size); |
3a359f0b DV |
845 | return hole_size; |
846 | } | |
847 | ||
848 | return 0; | |
849 | } | |
850 | ||
e18c0412 | 851 | /** |
b5c3714f DV |
852 | * drm_mm_print - print allocator state |
853 | * @mm: drm_mm allocator to print | |
854 | * @p: DRM printer to use | |
e18c0412 | 855 | */ |
b5c3714f | 856 | void drm_mm_print(const struct drm_mm *mm, struct drm_printer *p) |
3a359f0b | 857 | { |
45b186f1 | 858 | const struct drm_mm_node *entry; |
440fd528 | 859 | u64 total_used = 0, total_free = 0, total = 0; |
3a359f0b | 860 | |
b5c3714f | 861 | total_free += drm_mm_dump_hole(p, &mm->head_node); |
ea7b1dd4 DV |
862 | |
863 | drm_mm_for_each_node(entry, mm) { | |
b5c3714f | 864 | drm_printf(p, "%#018llx-%#018llx: %llu: used\n", entry->start, |
440fd528 | 865 | entry->start + entry->size, entry->size); |
ea7b1dd4 | 866 | total_used += entry->size; |
b5c3714f | 867 | total_free += drm_mm_dump_hole(p, entry); |
fa8a1238 | 868 | } |
ea7b1dd4 DV |
869 | total = total_free + total_used; |
870 | ||
b5c3714f | 871 | drm_printf(p, "total: %llu, used %llu free %llu\n", total, |
440fd528 | 872 | total_used, total_free); |
fa8a1238 | 873 | } |
b5c3714f | 874 | EXPORT_SYMBOL(drm_mm_print); |