[deliverable/linux.git] / drivers / gpu / drm / ttm / ttm_memory.c

/**************************************************************************
 *
 * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 **************************************************************************/

#include "ttm/ttm_memory.h"
#include "ttm/ttm_module.h"
#include "ttm/ttm_page_alloc.h"
#include <linux/spinlock.h>
#include <linux/sched.h>
#include <linux/wait.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/slab.h>

#define TTM_MEMORY_ALLOC_RETRIES 4

struct ttm_mem_zone {
	struct kobject kobj;
	struct ttm_mem_global *glob;
	const char *name;
	uint64_t zone_mem;
	uint64_t emer_mem;
	uint64_t max_mem;
	uint64_t swap_limit;
	uint64_t used_mem;
};

static struct attribute ttm_mem_sys = {
	.name = "zone_memory",
	.mode = S_IRUGO
};
static struct attribute ttm_mem_emer = {
	.name = "emergency_memory",
	.mode = S_IRUGO | S_IWUSR
};
static struct attribute ttm_mem_max = {
	.name = "available_memory",
	.mode = S_IRUGO | S_IWUSR
};
static struct attribute ttm_mem_swap = {
	.name = "swap_limit",
	.mode = S_IRUGO | S_IWUSR
};
static struct attribute ttm_mem_used = {
	.name = "used_memory",
	.mode = S_IRUGO
};

static void ttm_mem_zone_kobj_release(struct kobject *kobj)
{
	struct ttm_mem_zone *zone =
		container_of(kobj, struct ttm_mem_zone, kobj);

	printk(KERN_INFO TTM_PFX
	       "Zone %7s: Used memory at exit: %llu kiB.\n",
	       zone->name, (unsigned long long) zone->used_mem >> 10);
	kfree(zone);
}

static ssize_t ttm_mem_zone_show(struct kobject *kobj,
				 struct attribute *attr,
				 char *buffer)
{
	struct ttm_mem_zone *zone =
		container_of(kobj, struct ttm_mem_zone, kobj);
	uint64_t val = 0;

	spin_lock(&zone->glob->lock);
	if (attr == &ttm_mem_sys)
		val = zone->zone_mem;
	else if (attr == &ttm_mem_emer)
		val = zone->emer_mem;
	else if (attr == &ttm_mem_max)
		val = zone->max_mem;
	else if (attr == &ttm_mem_swap)
		val = zone->swap_limit;
	else if (attr == &ttm_mem_used)
		val = zone->used_mem;
	spin_unlock(&zone->glob->lock);

	return snprintf(buffer, PAGE_SIZE, "%llu\n",
			(unsigned long long) val >> 10);
}

static void ttm_check_swapping(struct ttm_mem_global *glob);

static ssize_t ttm_mem_zone_store(struct kobject *kobj,
				  struct attribute *attr,
				  const char *buffer,
				  size_t size)
{
	struct ttm_mem_zone *zone =
		container_of(kobj, struct ttm_mem_zone, kobj);
	int chars;
	unsigned long val;
	uint64_t val64;

	chars = sscanf(buffer, "%lu", &val);
	if (chars == 0)
		return size;

	val64 = val;
	val64 <<= 10;

	spin_lock(&zone->glob->lock);
	if (val64 > zone->zone_mem)
		val64 = zone->zone_mem;
	if (attr == &ttm_mem_emer) {
		zone->emer_mem = val64;
		if (zone->max_mem > val64)
			zone->max_mem = val64;
	} else if (attr == &ttm_mem_max) {
		zone->max_mem = val64;
		if (zone->emer_mem < val64)
			zone->emer_mem = val64;
	} else if (attr == &ttm_mem_swap)
		zone->swap_limit = val64;
	spin_unlock(&zone->glob->lock);

	ttm_check_swapping(zone->glob);

	return size;
}

static struct attribute *ttm_mem_zone_attrs[] = {
	&ttm_mem_sys,
	&ttm_mem_emer,
	&ttm_mem_max,
	&ttm_mem_swap,
	&ttm_mem_used,
	NULL
};

static const struct sysfs_ops ttm_mem_zone_ops = {
	.show = &ttm_mem_zone_show,
	.store = &ttm_mem_zone_store
};

static struct kobj_type ttm_mem_zone_kobj_type = {
	.release = &ttm_mem_zone_kobj_release,
	.sysfs_ops = &ttm_mem_zone_ops,
	.default_attrs = ttm_mem_zone_attrs,
};

static void ttm_mem_global_kobj_release(struct kobject *kobj)
{
	struct ttm_mem_global *glob =
		container_of(kobj, struct ttm_mem_global, kobj);

	kfree(glob);
}

static struct kobj_type ttm_mem_glob_kobj_type = {
	.release = &ttm_mem_global_kobj_release,
};

static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
					bool from_wq, uint64_t extra)
{
	unsigned int i;
	struct ttm_mem_zone *zone;
	uint64_t target;

	for (i = 0; i < glob->num_zones; ++i) {
		zone = glob->zones[i];

		if (from_wq)
			target = zone->swap_limit;
		else if (capable(CAP_SYS_ADMIN))
			target = zone->emer_mem;
		else
			target = zone->max_mem;

		target = (extra > target) ? 0ULL : target;

		if (zone->used_mem > target)
			return true;
	}
	return false;
}

/**
 * At this point we only support a single shrink callback.
 * Extend this if needed, perhaps using a linked list of callbacks.
 * Note that this function is reentrant:
 * many threads may try to swap out at any given time.
 */

static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
		       uint64_t extra)
{
	int ret;
	struct ttm_mem_shrink *shrink;

	spin_lock(&glob->lock);
	if (glob->shrink == NULL)
		goto out;

	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
		shrink = glob->shrink;
		spin_unlock(&glob->lock);
		ret = shrink->do_shrink(shrink);
		spin_lock(&glob->lock);
		if (unlikely(ret != 0))
			goto out;
	}
out:
	spin_unlock(&glob->lock);
}


static void ttm_shrink_work(struct work_struct *work)
{
	struct ttm_mem_global *glob =
	    container_of(work, struct ttm_mem_global, work);

	ttm_shrink(glob, true, 0ULL);
}

static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
				    const struct sysinfo *si)
{
	struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
	uint64_t mem;
	int ret;

	if (unlikely(!zone))
		return -ENOMEM;

	mem = si->totalram - si->totalhigh;
	mem *= si->mem_unit;

	zone->name = "kernel";
	zone->zone_mem = mem;
	zone->max_mem = mem >> 1;
	zone->emer_mem = (mem >> 1) + (mem >> 2);
	zone->swap_limit = zone->max_mem - (mem >> 3);
	zone->used_mem = 0;
	zone->glob = glob;
	glob->zone_kernel = zone;
	ret = kobject_init_and_add(
		&zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
	if (unlikely(ret != 0)) {
		kobject_put(&zone->kobj);
		return ret;
	}
	glob->zones[glob->num_zones++] = zone;
	return 0;
}

#ifdef CONFIG_HIGHMEM
static int ttm_mem_init_highmem_zone(struct ttm_mem_global *glob,
				     const struct sysinfo *si)
{
	struct ttm_mem_zone *zone;
	uint64_t mem;
	int ret;

	if (si->totalhigh == 0)
		return 0;

	zone = kzalloc(sizeof(*zone), GFP_KERNEL);
	if (unlikely(!zone))
		return -ENOMEM;

	mem = si->totalram;
	mem *= si->mem_unit;

	zone->name = "highmem";
	zone->zone_mem = mem;
	zone->max_mem = mem >> 1;
	zone->emer_mem = (mem >> 1) + (mem >> 2);
	zone->swap_limit = zone->max_mem - (mem >> 3);
	zone->used_mem = 0;
	zone->glob = glob;
	glob->zone_highmem = zone;
	ret = kobject_init_and_add(
		&zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
	if (unlikely(ret != 0)) {
		kobject_put(&zone->kobj);
		return ret;
	}
	glob->zones[glob->num_zones++] = zone;
	return 0;
}
#else
static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob,
				   const struct sysinfo *si)
{
	struct ttm_mem_zone *zone = kzalloc(sizeof(*zone), GFP_KERNEL);
	uint64_t mem;
	int ret;

	if (unlikely(!zone))
		return -ENOMEM;

	mem = si->totalram;
	mem *= si->mem_unit;

	/**
	 * No special dma32 zone needed.
	 */

	if (mem <= ((uint64_t) 1ULL << 32)) {
		kfree(zone);
		return 0;
	}

	/*
	 * Limit max dma32 memory to 4GB for now
	 * until we can figure out how big this
	 * zone really is.
	 */

	mem = ((uint64_t) 1ULL << 32);
	zone->name = "dma32";
	zone->zone_mem = mem;
	zone->max_mem = mem >> 1;
	zone->emer_mem = (mem >> 1) + (mem >> 2);
	zone->swap_limit = zone->max_mem - (mem >> 3);
	zone->used_mem = 0;
	zone->glob = glob;
	glob->zone_dma32 = zone;
	ret = kobject_init_and_add(
		&zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
	if (unlikely(ret != 0)) {
		kobject_put(&zone->kobj);
		return ret;
	}
	glob->zones[glob->num_zones++] = zone;
	return 0;
}
#endif

int ttm_mem_global_init(struct ttm_mem_global *glob)
{
	struct sysinfo si;
	int ret;
	int i;
	struct ttm_mem_zone *zone;

	spin_lock_init(&glob->lock);
	glob->swap_queue = create_singlethread_workqueue("ttm_swap");
	INIT_WORK(&glob->work, ttm_shrink_work);
	init_waitqueue_head(&glob->queue);
	ret = kobject_init_and_add(
		&glob->kobj, &ttm_mem_glob_kobj_type, ttm_get_kobj(), "memory_accounting");
	if (unlikely(ret != 0)) {
		kobject_put(&glob->kobj);
		return ret;
	}

	si_meminfo(&si);

	ret = ttm_mem_init_kernel_zone(glob, &si);
	if (unlikely(ret != 0))
		goto out_no_zone;
#ifdef CONFIG_HIGHMEM
	ret = ttm_mem_init_highmem_zone(glob, &si);
	if (unlikely(ret != 0))
		goto out_no_zone;
#else
	ret = ttm_mem_init_dma32_zone(glob, &si);
	if (unlikely(ret != 0))
		goto out_no_zone;
#endif
	for (i = 0; i < glob->num_zones; ++i) {
		zone = glob->zones[i];
		printk(KERN_INFO TTM_PFX
		       "Zone %7s: Available graphics memory: %llu kiB.\n",
		       zone->name, (unsigned long long) zone->max_mem >> 10);
	}
	ttm_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE));
	return 0;
out_no_zone:
	ttm_mem_global_release(glob);
	return ret;
}
EXPORT_SYMBOL(ttm_mem_global_init);

void ttm_mem_global_release(struct ttm_mem_global *glob)
{
	unsigned int i;
	struct ttm_mem_zone *zone;

	/* let the page allocator first stop the shrink work. */
	ttm_page_alloc_fini();

	flush_workqueue(glob->swap_queue);
	destroy_workqueue(glob->swap_queue);
	glob->swap_queue = NULL;
	for (i = 0; i < glob->num_zones; ++i) {
		zone = glob->zones[i];
		kobject_del(&zone->kobj);
		kobject_put(&zone->kobj);
			}
	kobject_del(&glob->kobj);
	kobject_put(&glob->kobj);
}
EXPORT_SYMBOL(ttm_mem_global_release);

static void ttm_check_swapping(struct ttm_mem_global *glob)
{
	bool needs_swapping = false;
	unsigned int i;
	struct ttm_mem_zone *zone;

	spin_lock(&glob->lock);
	for (i = 0; i < glob->num_zones; ++i) {
		zone = glob->zones[i];
		if (zone->used_mem > zone->swap_limit) {
			needs_swapping = true;
			break;
		}
	}

	spin_unlock(&glob->lock);

	if (unlikely(needs_swapping))
		(void)queue_work(glob->swap_queue, &glob->work);

}

static void ttm_mem_global_free_zone(struct ttm_mem_global *glob,
				     struct ttm_mem_zone *single_zone,
				     uint64_t amount)
{
	unsigned int i;
	struct ttm_mem_zone *zone;

	spin_lock(&glob->lock);
	for (i = 0; i < glob->num_zones; ++i) {
		zone = glob->zones[i];
		if (single_zone && zone != single_zone)
			continue;
		zone->used_mem -= amount;
	}
	spin_unlock(&glob->lock);
}

void ttm_mem_global_free(struct ttm_mem_global *glob,
			 uint64_t amount)
{
	return ttm_mem_global_free_zone(glob, NULL, amount);
}
EXPORT_SYMBOL(ttm_mem_global_free);

static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
				  struct ttm_mem_zone *single_zone,
				  uint64_t amount, bool reserve)
{
	uint64_t limit;
	int ret = -ENOMEM;
	unsigned int i;
	struct ttm_mem_zone *zone;

	spin_lock(&glob->lock);
	for (i = 0; i < glob->num_zones; ++i) {
		zone = glob->zones[i];
		if (single_zone && zone != single_zone)
			continue;

		limit = (capable(CAP_SYS_ADMIN)) ?
			zone->emer_mem : zone->max_mem;

		if (zone->used_mem > limit)
			goto out_unlock;
	}

	if (reserve) {
		for (i = 0; i < glob->num_zones; ++i) {
			zone = glob->zones[i];
			if (single_zone && zone != single_zone)
				continue;
			zone->used_mem += amount;
		}
	}

	ret = 0;
out_unlock:
	spin_unlock(&glob->lock);
	ttm_check_swapping(glob);

	return ret;
}


static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
				     struct ttm_mem_zone *single_zone,
				     uint64_t memory,
				     bool no_wait, bool interruptible)
{
	int count = TTM_MEMORY_ALLOC_RETRIES;

	while (unlikely(ttm_mem_global_reserve(glob,
					       single_zone,
					       memory, true)
			!= 0)) {
		if (no_wait)
			return -ENOMEM;
		if (unlikely(count-- == 0))
			return -ENOMEM;
		ttm_shrink(glob, false, memory + (memory >> 2) + 16);
	}

	return 0;
}

int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
			 bool no_wait, bool interruptible)
{
	/**
	 * Normal allocations of kernel memory are registered in
	 * all zones.
	 */

	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
					 interruptible);
}
EXPORT_SYMBOL(ttm_mem_global_alloc);

int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
			      struct page *page,
			      bool no_wait, bool interruptible)
{

	struct ttm_mem_zone *zone = NULL;

	/**
	 * Page allocations may be registed in a single zone
	 * only if highmem or !dma32.
	 */

#ifdef CONFIG_HIGHMEM
	if (PageHighMem(page) && glob->zone_highmem != NULL)
		zone = glob->zone_highmem;
#else
	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
		zone = glob->zone_kernel;
#endif
	return ttm_mem_global_alloc_zone(glob, zone, PAGE_SIZE, no_wait,
					 interruptible);
}

void ttm_mem_global_free_page(struct ttm_mem_global *glob, struct page *page)
{
	struct ttm_mem_zone *zone = NULL;

#ifdef CONFIG_HIGHMEM
	if (PageHighMem(page) && glob->zone_highmem != NULL)
		zone = glob->zone_highmem;
#else
	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
		zone = glob->zone_kernel;
#endif
	ttm_mem_global_free_zone(glob, zone, PAGE_SIZE);
}


size_t ttm_round_pot(size_t size)
{
	if ((size & (size - 1)) == 0)
		return size;
	else if (size > PAGE_SIZE)
		return PAGE_ALIGN(size);
	else {
		size_t tmp_size = 4;

		while (tmp_size < size)
			tmp_size <<= 1;

		return tmp_size;
	}
	return 0;
}
EXPORT_SYMBOL(ttm_round_pot);
Commit	Line	Data
ba4e7d97 TH	1	/**************************************************************************
	2	*
	3	* Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA
	4	* All Rights Reserved.
	5	*
	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:
	13	*
	14	* The above copyright notice and this permission notice (including the
	15	* next paragraph) shall be included in all copies or substantial portions
	16	* of the Software.
	17	*
	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.
	25	*
	26	**************************************************************************/
	27
	28	#include "ttm/ttm_memory.h"
5fd9cbad	29	#include "ttm/ttm_module.h"
1403b1a3	30	#include "ttm/ttm_page_alloc.h"
ba4e7d97 TH	31	#include <linux/spinlock.h>
	32	#include <linux/sched.h>
	33	#include <linux/wait.h>
	34	#include <linux/mm.h>
	35	#include <linux/module.h>
5a0e3ad6	36	#include <linux/slab.h>
ba4e7d97	37
ba4e7d97 TH	38	#define TTM_MEMORY_ALLOC_RETRIES 4
ba4e7d97 TH	39
5fd9cbad TH	40	struct ttm_mem_zone {
	41	struct kobject kobj;
	42	struct ttm_mem_global *glob;
	43	const char *name;
	44	uint64_t zone_mem;
	45	uint64_t emer_mem;
	46	uint64_t max_mem;
	47	uint64_t swap_limit;
	48	uint64_t used_mem;
	49	};
	50
	51	static struct attribute ttm_mem_sys = {
	52	.name = "zone_memory",
	53	.mode = S_IRUGO
	54	};
	55	static struct attribute ttm_mem_emer = {
	56	.name = "emergency_memory",
	57	.mode = S_IRUGO \| S_IWUSR
	58	};
	59	static struct attribute ttm_mem_max = {
	60	.name = "available_memory",
	61	.mode = S_IRUGO \| S_IWUSR
	62	};
	63	static struct attribute ttm_mem_swap = {
	64	.name = "swap_limit",
	65	.mode = S_IRUGO \| S_IWUSR
	66	};
	67	static struct attribute ttm_mem_used = {
	68	.name = "used_memory",
	69	.mode = S_IRUGO
	70	};
	71
	72	static void ttm_mem_zone_kobj_release(struct kobject *kobj)
	73	{
	74	struct ttm_mem_zone *zone =
	75	container_of(kobj, struct ttm_mem_zone, kobj);
	76
	77	printk(KERN_INFO TTM_PFX
	78	"Zone %7s: Used memory at exit: %llu kiB.\n",
	79	zone->name, (unsigned long long) zone->used_mem >> 10);
	80	kfree(zone);
	81	}
	82
	83	static ssize_t ttm_mem_zone_show(struct kobject *kobj,
	84	struct attribute *attr,
	85	char *buffer)
	86	{
	87	struct ttm_mem_zone *zone =
	88	container_of(kobj, struct ttm_mem_zone, kobj);
	89	uint64_t val = 0;
	90
	91	spin_lock(&zone->glob->lock);
	92	if (attr == &ttm_mem_sys)
	93	val = zone->zone_mem;
	94	else if (attr == &ttm_mem_emer)
	95	val = zone->emer_mem;
	96	else if (attr == &ttm_mem_max)
	97	val = zone->max_mem;
	98	else if (attr == &ttm_mem_swap)
	99	val = zone->swap_limit;
	100	else if (attr == &ttm_mem_used)
	101	val = zone->used_mem;
	102	spin_unlock(&zone->glob->lock);
	103
104	return snprintf(buffer, PAGE_SIZE, "%llu\n",
105	(unsigned long long) val >> 10);
106	}
107
108	static void ttm_check_swapping(struct ttm_mem_global *glob);
109
110	static ssize_t ttm_mem_zone_store(struct kobject *kobj,
111	struct attribute *attr,
112	const char *buffer,
113	size_t size)
114	{
115	struct ttm_mem_zone *zone =
116	container_of(kobj, struct ttm_mem_zone, kobj);
117	int chars;
118	unsigned long val;
119	uint64_t val64;
120
121	chars = sscanf(buffer, "%lu", &val);
122	if (chars == 0)
123	return size;
124
125	val64 = val;
126	val64 <<= 10;
127
128	spin_lock(&zone->glob->lock);
129	if (val64 > zone->zone_mem)
130	val64 = zone->zone_mem;
131	if (attr == &ttm_mem_emer) {
132	zone->emer_mem = val64;
133	if (zone->max_mem > val64)
134	zone->max_mem = val64;
135	} else if (attr == &ttm_mem_max) {
136	zone->max_mem = val64;
137	if (zone->emer_mem < val64)
138	zone->emer_mem = val64;
139	} else if (attr == &ttm_mem_swap)
140	zone->swap_limit = val64;
141	spin_unlock(&zone->glob->lock);
142
143	ttm_check_swapping(zone->glob);
144
145	return size;
146	}
147
148	static struct attribute *ttm_mem_zone_attrs[] = {
149	&ttm_mem_sys,
150	&ttm_mem_emer,
151	&ttm_mem_max,
152	&ttm_mem_swap,
153	&ttm_mem_used,
154	NULL
155	};
156
52cf25d0	157	static const struct sysfs_ops ttm_mem_zone_ops = {
5fd9cbad TH	158	.show = &ttm_mem_zone_show,
	159	.store = &ttm_mem_zone_store
	160	};
	161
	162	static struct kobj_type ttm_mem_zone_kobj_type = {
	163	.release = &ttm_mem_zone_kobj_release,
	164	.sysfs_ops = &ttm_mem_zone_ops,
	165	.default_attrs = ttm_mem_zone_attrs,
	166	};
	167
	168	static void ttm_mem_global_kobj_release(struct kobject *kobj)
	169	{
	170	struct ttm_mem_global *glob =
	171	container_of(kobj, struct ttm_mem_global, kobj);
	172
	173	kfree(glob);
	174	}
	175
	176	static struct kobj_type ttm_mem_glob_kobj_type = {
	177	.release = &ttm_mem_global_kobj_release,
	178	};
	179
	180	static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
	181	bool from_wq, uint64_t extra)
	182	{
	183	unsigned int i;
	184	struct ttm_mem_zone *zone;
	185	uint64_t target;
	186
	187	for (i = 0; i < glob->num_zones; ++i) {
	188	zone = glob->zones[i];
	189
	190	if (from_wq)
	191	target = zone->swap_limit;
	192	else if (capable(CAP_SYS_ADMIN))
	193	target = zone->emer_mem;
	194	else
	195	target = zone->max_mem;
	196
	197	target = (extra > target) ? 0ULL : target;
	198
	199	if (zone->used_mem > target)
	200	return true;
	201	}
	202	return false;
	203	}
	204
ba4e7d97 TH	205	/**
	206	* At this point we only support a single shrink callback.
	207	* Extend this if needed, perhaps using a linked list of callbacks.
	208	* Note that this function is reentrant:
	209	* many threads may try to swap out at any given time.
	210	*/
	211
5fd9cbad	212	static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
ba4e7d97 TH	213	uint64_t extra)
	214	{
	215	int ret;
	216	struct ttm_mem_shrink *shrink;
ba4e7d97 TH	217
	218	spin_lock(&glob->lock);
	219	if (glob->shrink == NULL)
	220	goto out;
	221
5fd9cbad	222	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
ba4e7d97 TH	223	shrink = glob->shrink;
	224	spin_unlock(&glob->lock);
	225	ret = shrink->do_shrink(shrink);
	226	spin_lock(&glob->lock);
	227	if (unlikely(ret != 0))
	228	goto out;
	229	}
	230	out:
	231	spin_unlock(&glob->lock);
	232	}
	233
5fd9cbad TH	234
5fd9cbad TH	235
ba4e7d97 TH	236	static void ttm_shrink_work(struct work_struct *work)
	237	{
	238	struct ttm_mem_global *glob =
	239	container_of(work, struct ttm_mem_global, work);
	240
	241	ttm_shrink(glob, true, 0ULL);
	242	}
	243
5fd9cbad TH	244	static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
	245	const struct sysinfo *si)
	246	{
	247	struct ttm_mem_zone zone = kzalloc(sizeof(zone), GFP_KERNEL);
	248	uint64_t mem;
759e4f83	249	int ret;
5fd9cbad TH	250
	251	if (unlikely(!zone))
	252	return -ENOMEM;
	253
	254	mem = si->totalram - si->totalhigh;
	255	mem *= si->mem_unit;
	256
	257	zone->name = "kernel";
	258	zone->zone_mem = mem;
	259	zone->max_mem = mem >> 1;
	260	zone->emer_mem = (mem >> 1) + (mem >> 2);
	261	zone->swap_limit = zone->max_mem - (mem >> 3);
	262	zone->used_mem = 0;
	263	zone->glob = glob;
	264	glob->zone_kernel = zone;
b642ed06 RD	265	ret = kobject_init_and_add(
b642ed06 RD	266	&zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
759e4f83 TH	267	if (unlikely(ret != 0)) {
	268	kobject_put(&zone->kobj);
	269	return ret;
	270	}
	271	glob->zones[glob->num_zones++] = zone;
	272	return 0;
5fd9cbad TH	273	}
	274
	275	#ifdef CONFIG_HIGHMEM
	276	static int ttm_mem_init_highmem_zone(struct ttm_mem_global *glob,
	277	const struct sysinfo *si)
	278	{
46a79fa0	279	struct ttm_mem_zone *zone;
5fd9cbad	280	uint64_t mem;
759e4f83	281	int ret;
5fd9cbad	282
5fd9cbad TH	283	if (si->totalhigh == 0)
	284	return 0;
	285
46a79fa0 DC	286	zone = kzalloc(sizeof(*zone), GFP_KERNEL);
	287	if (unlikely(!zone))
	288	return -ENOMEM;
	289
5fd9cbad TH	290	mem = si->totalram;
	291	mem *= si->mem_unit;
	292
	293	zone->name = "highmem";
	294	zone->zone_mem = mem;
	295	zone->max_mem = mem >> 1;
	296	zone->emer_mem = (mem >> 1) + (mem >> 2);
	297	zone->swap_limit = zone->max_mem - (mem >> 3);
	298	zone->used_mem = 0;
	299	zone->glob = glob;
	300	glob->zone_highmem = zone;
b642ed06 RD	301	ret = kobject_init_and_add(
b642ed06 RD	302	&zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
759e4f83 TH	303	if (unlikely(ret != 0)) {
	304	kobject_put(&zone->kobj);
	305	return ret;
	306	}
	307	glob->zones[glob->num_zones++] = zone;
	308	return 0;
5fd9cbad TH	309	}
	310	#else
	311	static int ttm_mem_init_dma32_zone(struct ttm_mem_global *glob,
	312	const struct sysinfo *si)
	313	{
	314	struct ttm_mem_zone zone = kzalloc(sizeof(zone), GFP_KERNEL);
	315	uint64_t mem;
759e4f83	316	int ret;
5fd9cbad TH	317
	318	if (unlikely(!zone))
	319	return -ENOMEM;
	320
	321	mem = si->totalram;
	322	mem *= si->mem_unit;
	323
	324	/**
	325	* No special dma32 zone needed.
	326	*/
	327
ec42a6e7 DA	328	if (mem <= ((uint64_t) 1ULL << 32)) {
ec42a6e7 DA	329	kfree(zone);
5fd9cbad	330	return 0;
ec42a6e7	331	}
5fd9cbad TH	332
	333	/*
	334	* Limit max dma32 memory to 4GB for now
	335	* until we can figure out how big this
	336	* zone really is.
	337	*/
	338
	339	mem = ((uint64_t) 1ULL << 32);
	340	zone->name = "dma32";
	341	zone->zone_mem = mem;
	342	zone->max_mem = mem >> 1;
	343	zone->emer_mem = (mem >> 1) + (mem >> 2);
	344	zone->swap_limit = zone->max_mem - (mem >> 3);
	345	zone->used_mem = 0;
	346	zone->glob = glob;
	347	glob->zone_dma32 = zone;
b642ed06 RD	348	ret = kobject_init_and_add(
b642ed06 RD	349	&zone->kobj, &ttm_mem_zone_kobj_type, &glob->kobj, zone->name);
759e4f83 TH	350	if (unlikely(ret != 0)) {
	351	kobject_put(&zone->kobj);
	352	return ret;
	353	}
	354	glob->zones[glob->num_zones++] = zone;
	355	return 0;
5fd9cbad TH	356	}
	357	#endif
	358
ba4e7d97 TH	359	int ttm_mem_global_init(struct ttm_mem_global *glob)
	360	{
	361	struct sysinfo si;
5fd9cbad TH	362	int ret;
	363	int i;
	364	struct ttm_mem_zone *zone;
ba4e7d97 TH	365
	366	spin_lock_init(&glob->lock);
	367	glob->swap_queue = create_singlethread_workqueue("ttm_swap");
	368	INIT_WORK(&glob->work, ttm_shrink_work);
	369	init_waitqueue_head(&glob->queue);
b642ed06 RD	370	ret = kobject_init_and_add(
b642ed06 RD	371	&glob->kobj, &ttm_mem_glob_kobj_type, ttm_get_kobj(), "memory_accounting");
759e4f83 TH	372	if (unlikely(ret != 0)) {
	373	kobject_put(&glob->kobj);
	374	return ret;
	375	}
ba4e7d97 TH	376
	377	si_meminfo(&si);
	378
5fd9cbad TH	379	ret = ttm_mem_init_kernel_zone(glob, &si);
	380	if (unlikely(ret != 0))
	381	goto out_no_zone;
	382	#ifdef CONFIG_HIGHMEM
	383	ret = ttm_mem_init_highmem_zone(glob, &si);
	384	if (unlikely(ret != 0))
	385	goto out_no_zone;
	386	#else
	387	ret = ttm_mem_init_dma32_zone(glob, &si);
	388	if (unlikely(ret != 0))
	389	goto out_no_zone;
	390	#endif
	391	for (i = 0; i < glob->num_zones; ++i) {
	392	zone = glob->zones[i];
	393	printk(KERN_INFO TTM_PFX
	394	"Zone %7s: Available graphics memory: %llu kiB.\n",
	395	zone->name, (unsigned long long) zone->max_mem >> 10);
	396	}
c96af79e	397	ttm_page_alloc_init(glob, glob->zone_kernel->max_mem/(2*PAGE_SIZE));
ba4e7d97	398	return 0;
5fd9cbad TH	399	out_no_zone:
	400	ttm_mem_global_release(glob);
	401	return ret;
ba4e7d97 TH	402	}
	403	EXPORT_SYMBOL(ttm_mem_global_init);
	404
	405	void ttm_mem_global_release(struct ttm_mem_global *glob)
	406	{
5fd9cbad TH	407	unsigned int i;
	408	struct ttm_mem_zone *zone;
	409
1403b1a3 PN	410	/* let the page allocator first stop the shrink work. */
	411	ttm_page_alloc_fini();
	412
ba4e7d97 TH	413	flush_workqueue(glob->swap_queue);
	414	destroy_workqueue(glob->swap_queue);
	415	glob->swap_queue = NULL;
5fd9cbad TH	416	for (i = 0; i < glob->num_zones; ++i) {
	417	zone = glob->zones[i];
	418	kobject_del(&zone->kobj);
	419	kobject_put(&zone->kobj);
1403b1a3	420	}
5fd9cbad TH	421	kobject_del(&glob->kobj);
5fd9cbad TH	422	kobject_put(&glob->kobj);
ba4e7d97 TH	423	}
	424	EXPORT_SYMBOL(ttm_mem_global_release);
	425
5fd9cbad	426	static void ttm_check_swapping(struct ttm_mem_global *glob)
ba4e7d97	427	{
5fd9cbad TH	428	bool needs_swapping = false;
	429	unsigned int i;
	430	struct ttm_mem_zone *zone;
ba4e7d97 TH	431
ba4e7d97 TH	432	spin_lock(&glob->lock);
5fd9cbad TH	433	for (i = 0; i < glob->num_zones; ++i) {
	434	zone = glob->zones[i];
	435	if (zone->used_mem > zone->swap_limit) {
	436	needs_swapping = true;
	437	break;
	438	}
	439	}
	440
ba4e7d97 TH	441	spin_unlock(&glob->lock);
	442
	443	if (unlikely(needs_swapping))
	444	(void)queue_work(glob->swap_queue, &glob->work);
	445
	446	}
	447
5fd9cbad TH	448	static void ttm_mem_global_free_zone(struct ttm_mem_global *glob,
	449	struct ttm_mem_zone *single_zone,
	450	uint64_t amount)
ba4e7d97	451	{
5fd9cbad TH	452	unsigned int i;
	453	struct ttm_mem_zone *zone;
	454
ba4e7d97	455	spin_lock(&glob->lock);
5fd9cbad TH	456	for (i = 0; i < glob->num_zones; ++i) {
	457	zone = glob->zones[i];
	458	if (single_zone && zone != single_zone)
	459	continue;
	460	zone->used_mem -= amount;
	461	}
ba4e7d97 TH	462	spin_unlock(&glob->lock);
	463	}
	464
5fd9cbad TH	465	void ttm_mem_global_free(struct ttm_mem_global *glob,
	466	uint64_t amount)
	467	{
	468	return ttm_mem_global_free_zone(glob, NULL, amount);
	469	}
4bfd75cb	470	EXPORT_SYMBOL(ttm_mem_global_free);
5fd9cbad	471
ba4e7d97	472	static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
5fd9cbad TH	473	struct ttm_mem_zone *single_zone,
5fd9cbad TH	474	uint64_t amount, bool reserve)
ba4e7d97 TH	475	{
ba4e7d97 TH	476	uint64_t limit;
ba4e7d97	477	int ret = -ENOMEM;
5fd9cbad TH	478	unsigned int i;
5fd9cbad TH	479	struct ttm_mem_zone *zone;
ba4e7d97 TH	480
ba4e7d97 TH	481	spin_lock(&glob->lock);
5fd9cbad TH	482	for (i = 0; i < glob->num_zones; ++i) {
	483	zone = glob->zones[i];
	484	if (single_zone && zone != single_zone)
	485	continue;
ba4e7d97	486
5fd9cbad TH	487	limit = (capable(CAP_SYS_ADMIN)) ?
5fd9cbad TH	488	zone->emer_mem : zone->max_mem;
ba4e7d97	489
5fd9cbad TH	490	if (zone->used_mem > limit)
	491	goto out_unlock;
	492	}
ba4e7d97 TH	493
ba4e7d97 TH	494	if (reserve) {
5fd9cbad TH	495	for (i = 0; i < glob->num_zones; ++i) {
	496	zone = glob->zones[i];
	497	if (single_zone && zone != single_zone)
	498	continue;
	499	zone->used_mem += amount;
	500	}
ba4e7d97	501	}
5fd9cbad	502
ba4e7d97 TH	503	ret = 0;
	504	out_unlock:
	505	spin_unlock(&glob->lock);
	506	ttm_check_swapping(glob);
	507
	508	return ret;
	509	}
	510
5fd9cbad TH	511
	512	static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
	513	struct ttm_mem_zone *single_zone,
	514	uint64_t memory,
	515	bool no_wait, bool interruptible)
ba4e7d97 TH	516	{
	517	int count = TTM_MEMORY_ALLOC_RETRIES;
	518
5fd9cbad TH	519	while (unlikely(ttm_mem_global_reserve(glob,
	520	single_zone,
	521	memory, true)
ba4e7d97 TH	522	!= 0)) {
	523	if (no_wait)
	524	return -ENOMEM;
	525	if (unlikely(count-- == 0))
	526	return -ENOMEM;
	527	ttm_shrink(glob, false, memory + (memory >> 2) + 16);
	528	}
	529
	530	return 0;
	531	}
	532
5fd9cbad TH	533	int ttm_mem_global_alloc(struct ttm_mem_global *glob, uint64_t memory,
	534	bool no_wait, bool interruptible)
	535	{
	536	/**
	537	* Normal allocations of kernel memory are registered in
	538	* all zones.
	539	*/
	540
	541	return ttm_mem_global_alloc_zone(glob, NULL, memory, no_wait,
	542	interruptible);
	543	}
4bfd75cb	544	EXPORT_SYMBOL(ttm_mem_global_alloc);
5fd9cbad TH	545
	546	int ttm_mem_global_alloc_page(struct ttm_mem_global *glob,
	547	struct page *page,
	548	bool no_wait, bool interruptible)
	549	{
	550
	551	struct ttm_mem_zone *zone = NULL;
	552
	553	/**
	554	* Page allocations may be registed in a single zone
	555	* only if highmem or !dma32.
	556	*/
	557
	558	#ifdef CONFIG_HIGHMEM
	559	if (PageHighMem(page) && glob->zone_highmem != NULL)
	560	zone = glob->zone_highmem;
	561	#else
	562	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
	563	zone = glob->zone_kernel;
	564	#endif
	565	return ttm_mem_global_alloc_zone(glob, zone, PAGE_SIZE, no_wait,
	566	interruptible);
	567	}
	568
	569	void ttm_mem_global_free_page(struct ttm_mem_global glob, struct page page)
	570	{
	571	struct ttm_mem_zone *zone = NULL;
	572
	573	#ifdef CONFIG_HIGHMEM
	574	if (PageHighMem(page) && glob->zone_highmem != NULL)
	575	zone = glob->zone_highmem;
	576	#else
	577	if (glob->zone_dma32 && page_to_pfn(page) > 0x00100000UL)
	578	zone = glob->zone_kernel;
	579	#endif
	580	ttm_mem_global_free_zone(glob, zone, PAGE_SIZE);
	581	}
	582
	583
ba4e7d97 TH	584	size_t ttm_round_pot(size_t size)
	585	{
	586	if ((size & (size - 1)) == 0)
	587	return size;
	588	else if (size > PAGE_SIZE)
	589	return PAGE_ALIGN(size);
	590	else {
	591	size_t tmp_size = 4;
	592
	593	while (tmp_size < size)
	594	tmp_size <<= 1;
	595
	596	return tmp_size;
	597	}
	598	return 0;
	599	}
4bfd75cb	600	EXPORT_SYMBOL(ttm_round_pot);