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>
40 #define TTM_MEMORY_ALLOC_RETRIES 4
44 struct ttm_mem_global *glob;
53 static struct attribute ttm_mem_sys = {
54 .name = "zone_memory",
57 static struct attribute ttm_mem_emer = {
58 .name = "emergency_memory",
59 .mode = S_IRUGO | S_IWUSR
61 static struct attribute ttm_mem_max = {
62 .name = "available_memory",
63 .mode = S_IRUGO | S_IWUSR
65 static struct attribute ttm_mem_swap = {
67 .mode = S_IRUGO | S_IWUSR
69 static struct attribute ttm_mem_used = {
70 .name = "used_memory",
74 static void ttm_mem_zone_kobj_release(struct kobject *kobj)
76 struct ttm_mem_zone *zone =
77 container_of(kobj, struct ttm_mem_zone, kobj);
79 pr_info("Zone %7s: Used memory at exit: %llu kiB\n",
80 zone->name, (unsigned long long)zone->used_mem >> 10);
84 static ssize_t ttm_mem_zone_show(struct kobject *kobj,
85 struct attribute *attr,
88 struct ttm_mem_zone *zone =
89 container_of(kobj, struct ttm_mem_zone, kobj);
92 lockmgr(&zone->glob->lock, LK_EXCLUSIVE);
93 if (attr == &ttm_mem_sys)
95 else if (attr == &ttm_mem_emer)
97 else if (attr == &ttm_mem_max)
99 else if (attr == &ttm_mem_swap)
100 val = zone->swap_limit;
101 else if (attr == &ttm_mem_used)
102 val = zone->used_mem;
103 lockmgr(&zone->glob->lock, LK_RELEASE);
105 return ksnprintf(buffer, PAGE_SIZE, "%llu\n",
106 (unsigned long long) val >> 10);
109 static void ttm_check_swapping(struct ttm_mem_global *glob);
111 static ssize_t ttm_mem_zone_store(struct kobject *kobj,
112 struct attribute *attr,
116 struct ttm_mem_zone *zone =
117 container_of(kobj, struct ttm_mem_zone, kobj);
122 chars = ksscanf(buffer, "%lu", &val);
129 lockmgr(&zone->glob->lock, LK_EXCLUSIVE);
130 if (val64 > zone->zone_mem)
131 val64 = zone->zone_mem;
132 if (attr == &ttm_mem_emer) {
133 zone->emer_mem = val64;
134 if (zone->max_mem > val64)
135 zone->max_mem = val64;
136 } else if (attr == &ttm_mem_max) {
137 zone->max_mem = val64;
138 if (zone->emer_mem < val64)
139 zone->emer_mem = val64;
140 } else if (attr == &ttm_mem_swap)
141 zone->swap_limit = val64;
142 lockmgr(&zone->glob->lock, LK_RELEASE);
144 ttm_check_swapping(zone->glob);
149 static struct attribute *ttm_mem_zone_attrs[] = {
158 static const struct sysfs_ops ttm_mem_zone_ops = {
159 .show = &ttm_mem_zone_show,
160 .store = &ttm_mem_zone_store
163 static struct kobj_type ttm_mem_zone_kobj_type = {
164 .release = &ttm_mem_zone_kobj_release,
165 .sysfs_ops = &ttm_mem_zone_ops,
166 .default_attrs = ttm_mem_zone_attrs,
169 static void ttm_mem_global_kobj_release(struct kobject *kobj)
171 struct ttm_mem_global *glob =
172 container_of(kobj, struct ttm_mem_global, kobj);
177 static struct kobj_type ttm_mem_glob_kobj_type = {
178 .release = &ttm_mem_global_kobj_release,
181 static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
182 bool from_wq, uint64_t extra)
185 struct ttm_mem_zone *zone;
188 for (i = 0; i < glob->num_zones; ++i) {
189 zone = glob->zones[i];
192 target = zone->swap_limit;
193 else if (caps_priv_check_self(SYSCAP_NOVM_MLOCK) == 0)
194 target = zone->emer_mem;
196 target = zone->max_mem;
198 target = (extra > target) ? 0ULL : target;
200 if (zone->used_mem > target)
207 * At this point we only support a single shrink callback.
208 * Extend this if needed, perhaps using a linked list of callbacks.
209 * Note that this function is reentrant:
210 * many threads may try to swap out at any given time.
213 static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
214 uint64_t extra, struct ttm_operation_ctx *ctx)
218 lockmgr(&glob->lock, LK_EXCLUSIVE);
220 while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
221 lockmgr(&glob->lock, LK_RELEASE);
222 ret = ttm_bo_swapout(glob->bo_glob, ctx);
223 lockmgr(&glob->lock, LK_EXCLUSIVE);
224 if (unlikely(ret != 0))
228 lockmgr(&glob->lock, LK_RELEASE);
231 static void ttm_shrink_work(struct work_struct *work)
233 struct ttm_operation_ctx ctx = {
234 .interruptible = false,
237 struct ttm_mem_global *glob =
238 container_of(work, struct ttm_mem_global, work);
240 ttm_shrink(glob, true, 0ULL, &ctx);
243 static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
246 struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
249 zone->name = "kernel";
250 zone->zone_mem = mem;
251 zone->max_mem = mem >> 1;
252 zone->emer_mem = (mem >> 1) + (mem >> 2);
253 zone->swap_limit = zone->max_mem - (mem >> 3);
256 glob->zone_kernel = zone;
257 ret = kobject_init_and_add(
258 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
259 if (unlikely(ret != 0)) {
260 kobject_put(&zone->kobj);
263 glob->zones[glob->num_zones++] = zone;
267 #ifdef CONFIG_HIGHMEM
269 static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob,
272 struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
276 * No special dma32 zone needed.
279 if ((physmem * PAGE_SIZE) <= ((uint64_t) 1ULL << 32)) {
285 * Limit max dma32 memory to 4GB for now
286 * until we can figure out how big this
289 if (mem > ((uint64_t) 1ULL << 32))
290 mem = ((uint64_t) 1ULL << 32);
292 zone->name = "dma32";
293 zone->zone_mem = mem;
294 zone->max_mem = mem >> 1;
295 zone->emer_mem = (mem >> 1) + (mem >> 2);
296 zone->swap_limit = zone->max_mem - (mem >> 3);
299 glob->zone_dma32 = zone;
300 ret = kobject_init_and_add(
301 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
302 if (unlikely(ret != 0)) {
303 kobject_put(&zone->kobj);
306 glob->zones[glob->num_zones++] = zone;
311 int ttm_mem_global_init(struct ttm_mem_global *glob)
316 struct ttm_mem_zone *zone;
318 lockinit(&glob->lock, "ttmemglob", 0, 0);
319 glob->swap_queue = create_singlethread_workqueue("ttm_swap");
320 INIT_WORK(&glob->work, ttm_shrink_work);
321 ret = kobject_init_and_add(
322 &glob->kobj, &ttm_mem_glob_kobj_type, ttm_get_kobj(), "memory_accounting");
323 if (unlikely(ret != 0)) {
324 kobject_put(&glob->kobj);
329 * Managed contiguous memory for TTM. Only use kernel-reserved
330 * dma memory for TTM, which can be controlled via /boot/loader.conf
331 * (e.g. vm.dma_reserved=256m). This is the only truly dependable
334 mem = (uint64_t)vm_contig_avail_pages() * PAGE_SIZE;
336 ret = ttm_mem_init_kernel_zone(glob, mem);
337 if (unlikely(ret != 0))
339 #ifdef CONFIG_HIGHMEM
340 ret = ttm_mem_init_highmem_zone(glob, &si);
341 if (unlikely(ret != 0))
344 ret = ttm_mem_init_dma32_zone(glob, mem);
345 if (unlikely(ret != 0))
348 for (i = 0; i < glob->num_zones; ++i) {
349 zone = glob->zones[i];
350 pr_info("Zone %7s: Available graphics memory: %llu kiB\n",
351 zone->name, (unsigned long long)zone->max_mem >> 10);
353 ttm_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE));
354 ttm_dma_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE));
357 ttm_mem_global_release(glob);
360 EXPORT_SYMBOL(ttm_mem_global_init);
362 void ttm_mem_global_release(struct ttm_mem_global *glob)
365 struct ttm_mem_zone *zone;
367 /* let the page allocator first stop the shrink work. */
368 ttm_page_alloc_fini();
369 ttm_dma_page_alloc_fini();
371 flush_workqueue(glob->swap_queue);
372 destroy_workqueue(glob->swap_queue);
373 glob->swap_queue = NULL;
374 for (i = 0; i < glob->num_zones; ++i) {
375 zone = glob->zones[i];
376 kobject_del(&zone->kobj);
377 kobject_put(&zone->kobj);
379 kobject_del(&glob->kobj);
380 kobject_put(&glob->kobj);
382 EXPORT_SYMBOL(ttm_mem_global_release);
384 static void ttm_check_swapping(struct ttm_mem_global *glob)
386 bool needs_swapping = false;
388 struct ttm_mem_zone *zone;
390 lockmgr(&glob->lock, LK_EXCLUSIVE);
391 for (i = 0; i < glob->num_zones; ++i) {
392 zone = glob->zones[i];
393 if (zone->used_mem > zone->swap_limit) {
394 needs_swapping = true;
399 lockmgr(&glob->lock, LK_RELEASE);
401 if (unlikely(needs_swapping))
402 (void)queue_work(glob->swap_queue, &glob->work);
406 static void ttm_mem_global_free_zone(struct ttm_mem_global *glob,
407 struct ttm_mem_zone *single_zone,
411 struct ttm_mem_zone *zone;
413 lockmgr(&glob->lock, LK_EXCLUSIVE);
414 for (i = 0; i < glob->num_zones; ++i) {
415 zone = glob->zones[i];
416 if (single_zone && zone != single_zone)
418 zone->used_mem -= amount;
420 lockmgr(&glob->lock, LK_RELEASE);
423 void ttm_mem_global_free(struct ttm_mem_global *glob,
426 return ttm_mem_global_free_zone(glob, NULL, amount);
428 EXPORT_SYMBOL(ttm_mem_global_free);
431 * check if the available mem is under lower memory limit
433 * a. if no swap disk at all or free swap space is under swap_mem_limit
434 * but available system mem is bigger than sys_mem_limit, allow TTM
437 * b. if the available system mem is less than sys_mem_limit but free
438 * swap disk is bigger than swap_mem_limit, allow TTM allocation.
441 ttm_check_under_lowerlimit(struct ttm_mem_global *glob,
443 struct ttm_operation_ctx *ctx)
450 if (ctx->flags & TTM_OPT_FLAG_FORCE_ALLOC)
453 available = get_nr_swap_pages() + si_mem_available();
454 available -= num_pages;
455 if (available < glob->lower_mem_limit)
461 EXPORT_SYMBOL(ttm_check_under_lowerlimit);
463 static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
464 struct ttm_mem_zone *single_zone,
465 uint64_t amount, bool reserve)
470 struct ttm_mem_zone *zone;
472 lockmgr(&glob->lock, LK_EXCLUSIVE);
473 for (i = 0; i < glob->num_zones; ++i) {
474 zone = glob->zones[i];
475 if (single_zone && zone != single_zone)
478 limit = (caps_priv_check_self(SYSCAP_NOVM_MLOCK) == 0) ?
479 zone->emer_mem : zone->max_mem;
481 if (zone->used_mem > limit)
486 for (i = 0; i < glob->num_zones; ++i) {
487 zone = glob->zones[i];
488 if (single_zone && zone != single_zone)
490 zone->used_mem += amount;
496 lockmgr(&glob->lock, LK_RELEASE);
497 ttm_check_swapping(glob);
503 static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
504 struct ttm_mem_zone *single_zone,
506 struct ttm_operation_ctx *ctx)
508 int count = TTM_MEMORY_ALLOC_RETRIES;
510 while (unlikely(ttm_mem_global_reserve(glob,
514 if (ctx->no_wait_gpu)
516 if (unlikely(count-- == 0))
518 ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
524 int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
525 struct ttm_operation_ctx *ctx)
528 * Normal allocations of kernel memory are registered in
532 return ttm_mem_global_alloc_zone(glob, NULL, memory, ctx);
534 EXPORT_SYMBOL(ttm_mem_global_alloc);
536 int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
537 struct page *page, uint64_t size,
538 struct ttm_operation_ctx *ctx)
540 struct ttm_mem_zone *zone = NULL;
543 * Page allocations may be registed in a single zone
544 * only if highmem or !dma32.
547 #ifdef CONFIG_HIGHMEM
548 if (PageHighMem(page) && glob->zone_highmem != NULL)
549 zone = glob->zone_highmem;
551 if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
552 zone = glob->zone_kernel;
554 return ttm_mem_global_alloc_zone(glob, zone, size, ctx);
557 void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page,
560 struct ttm_mem_zone *zone = NULL;
562 #ifdef CONFIG_HIGHMEM
563 if (PageHighMem(page) && glob->zone_highmem != NULL)
564 zone = glob->zone_highmem;
566 if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
567 zone = glob->zone_kernel;
569 ttm_mem_global_free_zone(glob, zone, size);
572 size_t ttm_round_pot(size_t size)
574 if ((size & (size - 1)) == 0)
576 else if (size > PAGE_SIZE)
577 return PAGE_ALIGN(size);
581 while (tmp_size < size)
588 EXPORT_SYMBOL(ttm_round_pot);
590 uint64_t ttm_get_kernel_zone_memory_size(struct ttm_mem_global *glob)
592 return glob->zone_kernel->max_mem;
594 EXPORT_SYMBOL(ttm_get_kernel_zone_memory_size);