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 #include "ttm/ttm_memory.h"
29 #include "ttm/ttm_module.h"
30 #include <linux/spinlock.h>
31 #include <linux/sched.h>
32 #include <linux/wait.h>
34 #include <linux/module.h>
36 #define TTM_MEMORY_ALLOC_RETRIES 4
40 struct ttm_mem_global *glob;
49 static struct attribute ttm_mem_sys = {
50 .name = "zone_memory",
53 static struct attribute ttm_mem_emer = {
54 .name = "emergency_memory",
55 .mode = S_IRUGO | S_IWUSR
57 static struct attribute ttm_mem_max = {
58 .name = "available_memory",
59 .mode = S_IRUGO | S_IWUSR
61 static struct attribute ttm_mem_swap = {
63 .mode = S_IRUGO | S_IWUSR
65 static struct attribute ttm_mem_used = {
66 .name = "used_memory",
70 static void ttm_mem_zone_kobj_release(struct kobject *kobj)
72 struct ttm_mem_zone *zone =
73 container_of(kobj, struct ttm_mem_zone, kobj);
75 printk(KERN_INFO TTM_PFX
76 "Zone %7s: Used memory at exit: %llu kiB.\n",
77 zone->name, (unsigned long long) zone->used_mem >> 10);
81 static ssize_t ttm_mem_zone_show(struct kobject *kobj,
82 struct attribute *attr,
85 struct ttm_mem_zone *zone =
86 container_of(kobj, struct ttm_mem_zone, kobj);
89 spin_lock(&zone->glob->lock);
90 if (attr == &ttm_mem_sys)
92 else if (attr == &ttm_mem_emer)
94 else if (attr == &ttm_mem_max)
96 else if (attr == &ttm_mem_swap)
97 val = zone->swap_limit;
98 else if (attr == &ttm_mem_used)
100 spin_unlock(&zone->glob->lock);
102 return snprintf(buffer, PAGE_SIZE, "%llu\n",
103 (unsigned long long) val >> 10);
106 static void ttm_check_swapping(struct ttm_mem_global *glob);
108 static ssize_t ttm_mem_zone_store(struct kobject *kobj,
109 struct attribute *attr,
113 struct ttm_mem_zone *zone =
114 container_of(kobj, struct ttm_mem_zone, kobj);
119 chars = sscanf(buffer, "%lu", &val);
126 spin_lock(&zone->glob->lock);
127 if (val64 > zone->zone_mem)
128 val64 = zone->zone_mem;
129 if (attr == &ttm_mem_emer) {
130 zone->emer_mem = val64;
131 if (zone->max_mem > val64)
132 zone->max_mem = val64;
133 } else if (attr == &ttm_mem_max) {
134 zone->max_mem = val64;
135 if (zone->emer_mem < val64)
136 zone->emer_mem = val64;
137 } else if (attr == &ttm_mem_swap)
138 zone->swap_limit = val64;
139 spin_unlock(&zone->glob->lock);
141 ttm_check_swapping(zone->glob);
146 static struct attribute *ttm_mem_zone_attrs[] = {
155 static const struct sysfs_ops ttm_mem_zone_ops = {
156 .show = &ttm_mem_zone_show,
157 .store = &ttm_mem_zone_store
160 static struct kobj_type ttm_mem_zone_kobj_type = {
161 .release = &ttm_mem_zone_kobj_release,
162 .sysfs_ops = &ttm_mem_zone_ops,
163 .default_attrs = ttm_mem_zone_attrs,
166 static void ttm_mem_global_kobj_release(struct kobject *kobj)
168 struct ttm_mem_global *glob =
169 container_of(kobj, struct ttm_mem_global, kobj);
174 static struct kobj_type ttm_mem_glob_kobj_type = {
175 .release = &ttm_mem_global_kobj_release,
178 static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
179 bool from_wq, uint64_t extra)
182 struct ttm_mem_zone *zone;
185 for (i = 0; i < glob->num_zones; ++i) {
186 zone = glob->zones[i];
189 target = zone->swap_limit;
190 else if (capable(CAP_SYS_ADMIN))
191 target = zone->emer_mem;
193 target = zone->max_mem;
195 target = (extra > target) ? 0ULL : target;
197 if (zone->used_mem > target)
204 * At this point we only support a single shrink callback.
205 * Extend this if needed, perhaps using a linked list of callbacks.
206 * Note that this function is reentrant:
207 * many threads may try to swap out at any given time.
210 static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
214 struct ttm_mem_shrink *shrink;
216 spin_lock(&glob->lock);
217 if (glob->shrink == NULL)
220 while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
221 shrink = glob->shrink;
222 spin_unlock(&glob->lock);
223 ret = shrink->do_shrink(shrink);
224 spin_lock(&glob->lock);
225 if (unlikely(ret != 0))
229 spin_unlock(&glob->lock);
234 static void ttm_shrink_work(struct work_struct *work)
236 struct ttm_mem_global *glob =
237 container_of(work, struct ttm_mem_global, work);
239 ttm_shrink(glob, true, 0ULL);
242 static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
243 const struct sysinfo *si)
245 struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
252 mem = si->totalram - si->totalhigh;
255 zone->name = "kernel";
256 zone->zone_mem = mem;
257 zone->max_mem = mem >> 1;
258 zone->emer_mem = (mem >> 1) + (mem >> 2);
259 zone->swap_limit = zone->max_mem - (mem >> 3);
262 glob->zone_kernel = zone;
263 ret = kobject_init_and_add(
264 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
265 if (unlikely(ret != 0)) {
266 kobject_put(&zone->kobj);
269 glob->zones[glob->num_zones++] = zone;
273 #ifdef CONFIG_HIGHMEM
274 static int ttm_mem_init_highmem_zone(struct ttm_mem_global *glob,
275 const struct sysinfo *si)
277 struct ttm_mem_zone *zone;
281 if (si->totalhigh == 0)
284 zone = kzalloc(sizeof(*zone), GFP_KERNEL);
291 zone->name = "highmem";
292 zone->zone_mem = mem;
293 zone->max_mem = mem >> 1;
294 zone->emer_mem = (mem >> 1) + (mem >> 2);
295 zone->swap_limit = zone->max_mem - (mem >> 3);
298 glob->zone_highmem = zone;
299 ret = kobject_init_and_add(
300 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
301 if (unlikely(ret != 0)) {
302 kobject_put(&zone->kobj);
305 glob->zones[glob->num_zones++] = zone;
309 static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob,
310 const struct sysinfo *si)
312 struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
323 * No special dma32 zone needed.
326 if (mem <= ((uint64_t) 1ULL << 32)) {
332 * Limit max dma32 memory to 4GB for now
333 * until we can figure out how big this
337 mem = ((uint64_t) 1ULL << 32);
338 zone->name = "dma32";
339 zone->zone_mem = mem;
340 zone->max_mem = mem >> 1;
341 zone->emer_mem = (mem >> 1) + (mem >> 2);
342 zone->swap_limit = zone->max_mem - (mem >> 3);
345 glob->zone_dma32 = zone;
346 ret = kobject_init_and_add(
347 &zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
348 if (unlikely(ret != 0)) {
349 kobject_put(&zone->kobj);
352 glob->zones[glob->num_zones++] = zone;
357 int ttm_mem_global_init(struct ttm_mem_global *glob)
362 struct ttm_mem_zone *zone;
364 spin_lock_init(&glob->lock);
365 glob->swap_queue = create_singlethread_workqueue("ttm_swap");
366 INIT_WORK(&glob->work, ttm_shrink_work);
367 init_waitqueue_head(&glob->queue);
368 ret = kobject_init_and_add(
369 &glob->kobj, &ttm_mem_glob_kobj_type, ttm_get_kobj(), "memory_accounting");
370 if (unlikely(ret != 0)) {
371 kobject_put(&glob->kobj);
377 ret = ttm_mem_init_kernel_zone(glob, &si);
378 if (unlikely(ret != 0))
380 #ifdef CONFIG_HIGHMEM
381 ret = ttm_mem_init_highmem_zone(glob, &si);
382 if (unlikely(ret != 0))
385 ret = ttm_mem_init_dma32_zone(glob, &si);
386 if (unlikely(ret != 0))
389 for (i = 0; i < glob->num_zones; ++i) {
390 zone = glob->zones[i];
391 printk(KERN_INFO TTM_PFX
392 "Zone %7s: Available graphics memory: %llu kiB.\n",
393 zone->name, (unsigned long long) zone->max_mem >> 10);
397 ttm_mem_global_release(glob);
400 EXPORT_SYMBOL(ttm_mem_global_init);
402 void ttm_mem_global_release(struct ttm_mem_global *glob)
405 struct ttm_mem_zone *zone;
407 flush_workqueue(glob->swap_queue);
408 destroy_workqueue(glob->swap_queue);
409 glob->swap_queue = NULL;
410 for (i = 0; i < glob->num_zones; ++i) {
411 zone = glob->zones[i];
412 kobject_del(&zone->kobj);
413 kobject_put(&zone->kobj);
415 kobject_del(&glob->kobj);
416 kobject_put(&glob->kobj);
418 EXPORT_SYMBOL(ttm_mem_global_release);
420 static void ttm_check_swapping(struct ttm_mem_global *glob)
422 bool needs_swapping = false;
424 struct ttm_mem_zone *zone;
426 spin_lock(&glob->lock);
427 for (i = 0; i < glob->num_zones; ++i) {
428 zone = glob->zones[i];
429 if (zone->used_mem > zone->swap_limit) {
430 needs_swapping = true;
435 spin_unlock(&glob->lock);
437 if (unlikely(needs_swapping))
438 (void)queue_work(glob->swap_queue, &glob->work);
442 static void ttm_mem_global_free_zone(struct ttm_mem_global *glob,
443 struct ttm_mem_zone *single_zone,
447 struct ttm_mem_zone *zone;
449 spin_lock(&glob->lock);
450 for (i = 0; i < glob->num_zones; ++i) {
451 zone = glob->zones[i];
452 if (single_zone && zone != single_zone)
454 zone->used_mem -= amount;
456 spin_unlock(&glob->lock);
459 void ttm_mem_global_free(struct ttm_mem_global *glob,
462 return ttm_mem_global_free_zone(glob, NULL, amount);
464 EXPORT_SYMBOL(ttm_mem_global_free);
466 static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
467 struct ttm_mem_zone *single_zone,
468 uint64_t amount, bool reserve)
473 struct ttm_mem_zone *zone;
475 spin_lock(&glob->lock);
476 for (i = 0; i < glob->num_zones; ++i) {
477 zone = glob->zones[i];
478 if (single_zone && zone != single_zone)
481 limit = (capable(CAP_SYS_ADMIN)) ?
482 zone->emer_mem : zone->max_mem;
484 if (zone->used_mem > limit)
489 for (i = 0; i < glob->num_zones; ++i) {
490 zone = glob->zones[i];
491 if (single_zone && zone != single_zone)
493 zone->used_mem += amount;
499 spin_unlock(&glob->lock);
500 ttm_check_swapping(glob);
506 static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
507 struct ttm_mem_zone *single_zone,
509 bool no_wait, bool interruptible)
511 int count = TTM_MEMORY_ALLOC_RETRIES;
513 while (unlikely(ttm_mem_global_reserve(glob,
519 if (unlikely(count-- == 0))
521 ttm_shrink(glob, false, memory + (memory >> 2) + 16);
527 int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
528 bool no_wait, bool interruptible)
531 * Normal allocations of kernel memory are registered in
535 return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
538 EXPORT_SYMBOL(ttm_mem_global_alloc);
540 int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
542 bool no_wait, bool interruptible)
545 struct ttm_mem_zone *zone = NULL;
548 * Page allocations may be registed in a single zone
549 * only if highmem or !dma32.
552 #ifdef CONFIG_HIGHMEM
553 if (PageHighMem(page) && glob->zone_highmem != NULL)
554 zone = glob->zone_highmem;
556 if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
557 zone = glob->zone_kernel;
559 return ttm_mem_global_alloc_zone(glob, zone, PAGE_SIZE, no_wait,
563 void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page)
565 struct ttm_mem_zone *zone = NULL;
567 #ifdef CONFIG_HIGHMEM
568 if (PageHighMem(page) && glob->zone_highmem != NULL)
569 zone = glob->zone_highmem;
571 if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
572 zone = glob->zone_kernel;
574 ttm_mem_global_free_zone(glob, zone, PAGE_SIZE);
578 size_t ttm_round_pot(size_t size)
580 if ((size & (size - 1)) == 0)
582 else if (size > PAGE_SIZE)
583 return PAGE_ALIGN(size);
587 while (tmp_size < size)
594 EXPORT_SYMBOL(ttm_round_pot);