1 /**************************************************************************
3 * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 * USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
28 #define pr_fmt(fmt) "[TTM] " fmt
30 #include <drm/ttm/ttm_memory.h>
31 #include <drm/ttm/ttm_module.h>
32 #include <drm/ttm/ttm_page_alloc.h>
33 #include <linux/spinlock.h>
34 #include <linux/sched.h>
35 #include <linux/wait.h>
37 #include <linux/module.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
41 #define TTM_MEMORY_ALLOC_RETRIES 4
45 struct ttm_mem_global *glob;
54 static struct attribute ttm_mem_sys = {
55 .name = "zone_memory",
58 static struct attribute ttm_mem_emer = {
59 .name = "emergency_memory",
60 .mode = S_IRUGO | S_IWUSR
62 static struct attribute ttm_mem_max = {
63 .name = "available_memory",
64 .mode = S_IRUGO | S_IWUSR
66 static struct attribute ttm_mem_swap = {
68 .mode = S_IRUGO | S_IWUSR
70 static struct attribute ttm_mem_used = {
71 .name = "used_memory",
75 static void ttm_mem_zone_kobj_release(struct kobject *kobj)
77 struct ttm_mem_zone *zone =
78 container_of(kobj, struct ttm_mem_zone, kobj);
80 pr_info("Zone %7s: Used memory at exit: %llu kiB\n",
81 zone->name, (unsigned long long)zone->used_mem >> 10);
85 static ssize_t ttm_mem_zone_show(struct kobject *kobj,
86 struct attribute *attr,
89 struct ttm_mem_zone *zone =
90 container_of(kobj, struct ttm_mem_zone, kobj);
93 spin_lock(&zone->glob->lock);
94 if (attr == &ttm_mem_sys)
96 else if (attr == &ttm_mem_emer)
98 else if (attr == &ttm_mem_max)
100 else if (attr == &ttm_mem_swap)
101 val = zone->swap_limit;
102 else if (attr == &ttm_mem_used)
103 val = zone->used_mem;
104 spin_unlock(&zone->glob->lock);
106 return snprintf(buffer, PAGE_SIZE, "%llu\n",
107 (unsigned long long) val >> 10);
110 static void ttm_check_swapping(struct ttm_mem_global *glob);
112 static ssize_t ttm_mem_zone_store(struct kobject *kobj,
113 struct attribute *attr,
117 struct ttm_mem_zone *zone =
118 container_of(kobj, struct ttm_mem_zone, kobj);
123 chars = sscanf(buffer, "%lu", &val);
130 spin_lock(&zone->glob->lock);
131 if (val64 > zone->zone_mem)
132 val64 = zone->zone_mem;
133 if (attr == &ttm_mem_emer) {
134 zone->emer_mem = val64;
135 if (zone->max_mem > val64)
136 zone->max_mem = val64;
137 } else if (attr == &ttm_mem_max) {
138 zone->max_mem = val64;
139 if (zone->emer_mem < val64)
140 zone->emer_mem = val64;
141 } else if (attr == &ttm_mem_swap)
142 zone->swap_limit = val64;
143 spin_unlock(&zone->glob->lock);
145 ttm_check_swapping(zone->glob);
150 static struct attribute *ttm_mem_zone_attrs[] = {
159 static const struct sysfs_ops ttm_mem_zone_ops = {
160 .show = &ttm_mem_zone_show,
161 .store = &ttm_mem_zone_store
164 static struct kobj_type ttm_mem_zone_kobj_type = {
165 .release = &ttm_mem_zone_kobj_release,
166 .sysfs_ops = &ttm_mem_zone_ops,
167 .default_attrs = ttm_mem_zone_attrs,
170 static struct attribute ttm_mem_global_lower_mem_limit = {
171 .name = "lower_mem_limit",
172 .mode = S_IRUGO | S_IWUSR
175 static ssize_t ttm_mem_global_show(struct kobject *kobj,
176 struct attribute *attr,
179 struct ttm_mem_global *glob =
180 container_of(kobj, struct ttm_mem_global, kobj);
183 spin_lock(&glob->lock);
184 val = glob->lower_mem_limit;
185 spin_unlock(&glob->lock);
186 /* convert from number of pages to KB */
187 val <<= (PAGE_SHIFT - 10);
188 return snprintf(buffer, PAGE_SIZE, "%llu\n",
189 (unsigned long long) val);
192 static ssize_t ttm_mem_global_store(struct kobject *kobj,
193 struct attribute *attr,
200 struct ttm_mem_global *glob =
201 container_of(kobj, struct ttm_mem_global, kobj);
203 chars = sscanf(buffer, "%lu", &val);
208 /* convert from KB to number of pages */
209 val64 >>= (PAGE_SHIFT - 10);
211 spin_lock(&glob->lock);
212 glob->lower_mem_limit = val64;
213 spin_unlock(&glob->lock);
218 static void ttm_mem_global_kobj_release(struct kobject *kobj)
220 struct ttm_mem_global *glob =
221 container_of(kobj, struct ttm_mem_global, kobj);
226 static struct attribute *ttm_mem_global_attrs[] = {
227 &ttm_mem_global_lower_mem_limit,
231 static const struct sysfs_ops ttm_mem_global_ops = {
232 .show = &ttm_mem_global_show,
233 .store = &ttm_mem_global_store,
236 static struct kobj_type ttm_mem_glob_kobj_type = {
237 .release = &ttm_mem_global_kobj_release,
238 .sysfs_ops = &ttm_mem_global_ops,
239 .default_attrs = ttm_mem_global_attrs,
242 static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
243 bool from_wq, uint64_t extra)
246 struct ttm_mem_zone *zone;
249 for (i = 0; i < glob->num_zones; ++i) {
250 zone = glob->zones[i];
253 target = zone->swap_limit;
254 else if (capable(CAP_SYS_ADMIN))
255 target = zone->emer_mem;
257 target = zone->max_mem;
259 target = (extra > target) ? 0ULL : target;
261 if (zone->used_mem > target)
268 * At this point we only support a single shrink callback.
269 * Extend this if needed, perhaps using a linked list of callbacks.
270 * Note that this function is reentrant:
271 * many threads may try to swap out at any given time.
274 static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
275 uint64_t extra, struct ttm_operation_ctx *ctx)
279 spin_lock(&glob->lock);
281 while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
282 spin_unlock(&glob->lock);
283 ret = ttm_bo_swapout(glob->bo_glob, ctx);
284 spin_lock(&glob->lock);
285 if (unlikely(ret != 0))
289 spin_unlock(&glob->lock);
292 static void ttm_shrink_work(struct work_struct *work)
294 struct ttm_operation_ctx ctx = {
295 .interruptible = false,
298 struct ttm_mem_global *glob =
299 container_of(work, struct ttm_mem_global, work);
301 ttm_shrink(glob, true, 0ULL, &ctx);
304 static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
305 const struct sysinfo *si)
307 struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
314 mem = si->totalram - si->totalhigh;
317 zone->name = "kernel";
318 zone->zone_mem = mem;
319 zone->max_mem = mem >> 1;
320 zone->emer_mem = (mem >> 1) + (mem >> 2);
321 zone->swap_limit = zone->max_mem - (mem >> 3);
324 glob->zone_kernel = zone;
325 ret = kobject_init_and_add(
326 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
327 if (unlikely(ret != 0)) {
328 kobject_put(&zone->kobj);
331 glob->zones[glob->num_zones++] = zone;
335 #ifdef CONFIG_HIGHMEM
336 static int ttm_mem_init_highmem_zone(struct ttm_mem_global *glob,
337 const struct sysinfo *si)
339 struct ttm_mem_zone *zone;
343 if (si->totalhigh == 0)
346 zone = kzalloc(sizeof(*zone), GFP_KERNEL);
353 zone->name = "highmem";
354 zone->zone_mem = mem;
355 zone->max_mem = mem >> 1;
356 zone->emer_mem = (mem >> 1) + (mem >> 2);
357 zone->swap_limit = zone->max_mem - (mem >> 3);
360 glob->zone_highmem = zone;
361 ret = kobject_init_and_add(
362 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, "%s",
364 if (unlikely(ret != 0)) {
365 kobject_put(&zone->kobj);
368 glob->zones[glob->num_zones++] = zone;
372 static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob,
373 const struct sysinfo *si)
375 struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
386 * No special dma32 zone needed.
389 if (mem <= ((uint64_t) 1ULL << 32)) {
395 * Limit max dma32 memory to 4GB for now
396 * until we can figure out how big this
400 mem = ((uint64_t) 1ULL << 32);
401 zone->name = "dma32";
402 zone->zone_mem = mem;
403 zone->max_mem = mem >> 1;
404 zone->emer_mem = (mem >> 1) + (mem >> 2);
405 zone->swap_limit = zone->max_mem - (mem >> 3);
408 glob->zone_dma32 = zone;
409 ret = kobject_init_and_add(
410 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
411 if (unlikely(ret != 0)) {
412 kobject_put(&zone->kobj);
415 glob->zones[glob->num_zones++] = zone;
420 int ttm_mem_global_init(struct ttm_mem_global *glob)
425 struct ttm_mem_zone *zone;
427 spin_lock_init(&glob->lock);
428 glob->swap_queue = create_singlethread_workqueue("ttm_swap");
429 INIT_WORK(&glob->work, ttm_shrink_work);
430 ret = kobject_init_and_add(
431 &glob->kobj, &ttm_mem_glob_kobj_type, ttm_get_kobj(), "memory_accounting");
432 if (unlikely(ret != 0)) {
433 kobject_put(&glob->kobj);
439 /* set it as 0 by default to keep original behavior of OOM */
440 glob->lower_mem_limit = 0;
442 ret = ttm_mem_init_kernel_zone(glob, &si);
443 if (unlikely(ret != 0))
445 #ifdef CONFIG_HIGHMEM
446 ret = ttm_mem_init_highmem_zone(glob, &si);
447 if (unlikely(ret != 0))
450 ret = ttm_mem_init_dma32_zone(glob, &si);
451 if (unlikely(ret != 0))
454 for (i = 0; i < glob->num_zones; ++i) {
455 zone = glob->zones[i];
456 pr_info("Zone %7s: Available graphics memory: %llu kiB\n",
457 zone->name, (unsigned long long)zone->max_mem >> 10);
459 ttm_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE));
460 ttm_dma_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE));
463 ttm_mem_global_release(glob);
466 EXPORT_SYMBOL(ttm_mem_global_init);
468 void ttm_mem_global_release(struct ttm_mem_global *glob)
471 struct ttm_mem_zone *zone;
473 /* let the page allocator first stop the shrink work. */
474 ttm_page_alloc_fini();
475 ttm_dma_page_alloc_fini();
477 flush_workqueue(glob->swap_queue);
478 destroy_workqueue(glob->swap_queue);
479 glob->swap_queue = NULL;
480 for (i = 0; i < glob->num_zones; ++i) {
481 zone = glob->zones[i];
482 kobject_del(&zone->kobj);
483 kobject_put(&zone->kobj);
485 kobject_del(&glob->kobj);
486 kobject_put(&glob->kobj);
488 EXPORT_SYMBOL(ttm_mem_global_release);
490 static void ttm_check_swapping(struct ttm_mem_global *glob)
492 bool needs_swapping = false;
494 struct ttm_mem_zone *zone;
496 spin_lock(&glob->lock);
497 for (i = 0; i < glob->num_zones; ++i) {
498 zone = glob->zones[i];
499 if (zone->used_mem > zone->swap_limit) {
500 needs_swapping = true;
505 spin_unlock(&glob->lock);
507 if (unlikely(needs_swapping))
508 (void)queue_work(glob->swap_queue, &glob->work);
512 static void ttm_mem_global_free_zone(struct ttm_mem_global *glob,
513 struct ttm_mem_zone *single_zone,
517 struct ttm_mem_zone *zone;
519 spin_lock(&glob->lock);
520 for (i = 0; i < glob->num_zones; ++i) {
521 zone = glob->zones[i];
522 if (single_zone && zone != single_zone)
524 zone->used_mem -= amount;
526 spin_unlock(&glob->lock);
529 void ttm_mem_global_free(struct ttm_mem_global *glob,
532 return ttm_mem_global_free_zone(glob, NULL, amount);
534 EXPORT_SYMBOL(ttm_mem_global_free);
537 * check if the available mem is under lower memory limit
539 * a. if no swap disk at all or free swap space is under swap_mem_limit
540 * but available system mem is bigger than sys_mem_limit, allow TTM
543 * b. if the available system mem is less than sys_mem_limit but free
544 * swap disk is bigger than swap_mem_limit, allow TTM allocation.
547 ttm_check_under_lowerlimit(struct ttm_mem_global *glob,
549 struct ttm_operation_ctx *ctx)
553 if (ctx->flags & TTM_OPT_FLAG_FORCE_ALLOC)
556 available = get_nr_swap_pages() + si_mem_available();
557 available -= num_pages;
558 if (available < glob->lower_mem_limit)
563 EXPORT_SYMBOL(ttm_check_under_lowerlimit);
565 static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
566 struct ttm_mem_zone *single_zone,
567 uint64_t amount, bool reserve)
572 struct ttm_mem_zone *zone;
574 spin_lock(&glob->lock);
575 for (i = 0; i < glob->num_zones; ++i) {
576 zone = glob->zones[i];
577 if (single_zone && zone != single_zone)
580 limit = (capable(CAP_SYS_ADMIN)) ?
581 zone->emer_mem : zone->max_mem;
583 if (zone->used_mem > limit)
588 for (i = 0; i < glob->num_zones; ++i) {
589 zone = glob->zones[i];
590 if (single_zone && zone != single_zone)
592 zone->used_mem += amount;
598 spin_unlock(&glob->lock);
599 ttm_check_swapping(glob);
605 static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
606 struct ttm_mem_zone *single_zone,
608 struct ttm_operation_ctx *ctx)
610 int count = TTM_MEMORY_ALLOC_RETRIES;
612 while (unlikely(ttm_mem_global_reserve(glob,
616 if (ctx->no_wait_gpu)
618 if (unlikely(count-- == 0))
620 ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
626 int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
627 struct ttm_operation_ctx *ctx)
630 * Normal allocations of kernel memory are registered in
634 return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
636 EXPORT_SYMBOL(ttm_mem_global_alloc);
638 int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
639 struct page *page, uint64_t size,
640 struct ttm_operation_ctx *ctx)
642 struct ttm_mem_zone *zone = NULL;
645 * Page allocations may be registed in a single zone
646 * only if highmem or !dma32.
649 #ifdef CONFIG_HIGHMEM
650 if (PageHighMem(page) && glob->zone_highmem != NULL)
651 zone = glob->zone_highmem;
653 if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
654 zone = glob->zone_kernel;
656 return ttm_mem_global_alloc_zone(glob, zone, size, ctx);
659 void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
662 struct ttm_mem_zone *zone = NULL;
664 #ifdef CONFIG_HIGHMEM
665 if (PageHighMem(page) && glob->zone_highmem != NULL)
666 zone = glob->zone_highmem;
668 if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
669 zone = glob->zone_kernel;
671 ttm_mem_global_free_zone(glob, zone, size);
674 size_t ttm_round_pot(size_t size)
676 if ((size & (size - 1)) == 0)
678 else if (size > PAGE_SIZE)
679 return PAGE_ALIGN(size);
683 while (tmp_size < size)
690 EXPORT_SYMBOL(ttm_round_pot);
692 uint64_t ttm_get_kernel_zone_memory_size(struct ttm_mem_global *glob)
694 return glob->zone_kernel->max_mem;
696 EXPORT_SYMBOL(ttm_get_kernel_zone_memory_size);