xref: /openbmc/linux/mm/swap_cgroup.c (revision 57ee11ea)
1 #include <linux/swap_cgroup.h>
2 #include <linux/vmalloc.h>
3 #include <linux/mm.h>
4 
5 #include <linux/swapops.h> /* depends on mm.h include */
6 
7 static DEFINE_MUTEX(swap_cgroup_mutex);
8 struct swap_cgroup_ctrl {
9 	struct page **map;
10 	unsigned long length;
11 	spinlock_t	lock;
12 };
13 
14 static struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES];
15 
16 struct swap_cgroup {
17 	unsigned short		id;
18 };
19 #define SC_PER_PAGE	(PAGE_SIZE/sizeof(struct swap_cgroup))
20 
21 /*
22  * SwapCgroup implements "lookup" and "exchange" operations.
23  * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge
24  * against SwapCache. At swap_free(), this is accessed directly from swap.
25  *
26  * This means,
27  *  - we have no race in "exchange" when we're accessed via SwapCache because
28  *    SwapCache(and its swp_entry) is under lock.
29  *  - When called via swap_free(), there is no user of this entry and no race.
30  * Then, we don't need lock around "exchange".
31  *
32  * TODO: we can push these buffers out to HIGHMEM.
33  */
34 
35 /*
36  * allocate buffer for swap_cgroup.
37  */
38 static int swap_cgroup_prepare(int type)
39 {
40 	struct page *page;
41 	struct swap_cgroup_ctrl *ctrl;
42 	unsigned long idx, max;
43 
44 	ctrl = &swap_cgroup_ctrl[type];
45 
46 	for (idx = 0; idx < ctrl->length; idx++) {
47 		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
48 		if (!page)
49 			goto not_enough_page;
50 		ctrl->map[idx] = page;
51 
52 		if (!(idx % SWAP_CLUSTER_MAX))
53 			cond_resched();
54 	}
55 	return 0;
56 not_enough_page:
57 	max = idx;
58 	for (idx = 0; idx < max; idx++)
59 		__free_page(ctrl->map[idx]);
60 
61 	return -ENOMEM;
62 }
63 
64 static struct swap_cgroup *__lookup_swap_cgroup(struct swap_cgroup_ctrl *ctrl,
65 						pgoff_t offset)
66 {
67 	struct page *mappage;
68 	struct swap_cgroup *sc;
69 
70 	mappage = ctrl->map[offset / SC_PER_PAGE];
71 	sc = page_address(mappage);
72 	return sc + offset % SC_PER_PAGE;
73 }
74 
75 static struct swap_cgroup *lookup_swap_cgroup(swp_entry_t ent,
76 					struct swap_cgroup_ctrl **ctrlp)
77 {
78 	pgoff_t offset = swp_offset(ent);
79 	struct swap_cgroup_ctrl *ctrl;
80 
81 	ctrl = &swap_cgroup_ctrl[swp_type(ent)];
82 	if (ctrlp)
83 		*ctrlp = ctrl;
84 	return __lookup_swap_cgroup(ctrl, offset);
85 }
86 
87 /**
88  * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry.
89  * @ent: swap entry to be cmpxchged
90  * @old: old id
91  * @new: new id
92  *
93  * Returns old id at success, 0 at failure.
94  * (There is no mem_cgroup using 0 as its id)
95  */
96 unsigned short swap_cgroup_cmpxchg(swp_entry_t ent,
97 					unsigned short old, unsigned short new)
98 {
99 	struct swap_cgroup_ctrl *ctrl;
100 	struct swap_cgroup *sc;
101 	unsigned long flags;
102 	unsigned short retval;
103 
104 	sc = lookup_swap_cgroup(ent, &ctrl);
105 
106 	spin_lock_irqsave(&ctrl->lock, flags);
107 	retval = sc->id;
108 	if (retval == old)
109 		sc->id = new;
110 	else
111 		retval = 0;
112 	spin_unlock_irqrestore(&ctrl->lock, flags);
113 	return retval;
114 }
115 
116 /**
117  * swap_cgroup_record - record mem_cgroup for a set of swap entries
118  * @ent: the first swap entry to be recorded into
119  * @id: mem_cgroup to be recorded
120  * @nr_ents: number of swap entries to be recorded
121  *
122  * Returns old value at success, 0 at failure.
123  * (Of course, old value can be 0.)
124  */
125 unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id,
126 				  unsigned int nr_ents)
127 {
128 	struct swap_cgroup_ctrl *ctrl;
129 	struct swap_cgroup *sc;
130 	unsigned short old;
131 	unsigned long flags;
132 	pgoff_t offset = swp_offset(ent);
133 	pgoff_t end = offset + nr_ents;
134 
135 	sc = lookup_swap_cgroup(ent, &ctrl);
136 
137 	spin_lock_irqsave(&ctrl->lock, flags);
138 	old = sc->id;
139 	for (;;) {
140 		VM_BUG_ON(sc->id != old);
141 		sc->id = id;
142 		offset++;
143 		if (offset == end)
144 			break;
145 		if (offset % SC_PER_PAGE)
146 			sc++;
147 		else
148 			sc = __lookup_swap_cgroup(ctrl, offset);
149 	}
150 	spin_unlock_irqrestore(&ctrl->lock, flags);
151 
152 	return old;
153 }
154 
155 /**
156  * lookup_swap_cgroup_id - lookup mem_cgroup id tied to swap entry
157  * @ent: swap entry to be looked up.
158  *
159  * Returns ID of mem_cgroup at success. 0 at failure. (0 is invalid ID)
160  */
161 unsigned short lookup_swap_cgroup_id(swp_entry_t ent)
162 {
163 	return lookup_swap_cgroup(ent, NULL)->id;
164 }
165 
166 int swap_cgroup_swapon(int type, unsigned long max_pages)
167 {
168 	void *array;
169 	unsigned long array_size;
170 	unsigned long length;
171 	struct swap_cgroup_ctrl *ctrl;
172 
173 	if (!do_swap_account)
174 		return 0;
175 
176 	length = DIV_ROUND_UP(max_pages, SC_PER_PAGE);
177 	array_size = length * sizeof(void *);
178 
179 	array = vzalloc(array_size);
180 	if (!array)
181 		goto nomem;
182 
183 	ctrl = &swap_cgroup_ctrl[type];
184 	mutex_lock(&swap_cgroup_mutex);
185 	ctrl->length = length;
186 	ctrl->map = array;
187 	spin_lock_init(&ctrl->lock);
188 	if (swap_cgroup_prepare(type)) {
189 		/* memory shortage */
190 		ctrl->map = NULL;
191 		ctrl->length = 0;
192 		mutex_unlock(&swap_cgroup_mutex);
193 		vfree(array);
194 		goto nomem;
195 	}
196 	mutex_unlock(&swap_cgroup_mutex);
197 
198 	return 0;
199 nomem:
200 	pr_info("couldn't allocate enough memory for swap_cgroup\n");
201 	pr_info("swap_cgroup can be disabled by swapaccount=0 boot option\n");
202 	return -ENOMEM;
203 }
204 
205 void swap_cgroup_swapoff(int type)
206 {
207 	struct page **map;
208 	unsigned long i, length;
209 	struct swap_cgroup_ctrl *ctrl;
210 
211 	if (!do_swap_account)
212 		return;
213 
214 	mutex_lock(&swap_cgroup_mutex);
215 	ctrl = &swap_cgroup_ctrl[type];
216 	map = ctrl->map;
217 	length = ctrl->length;
218 	ctrl->map = NULL;
219 	ctrl->length = 0;
220 	mutex_unlock(&swap_cgroup_mutex);
221 
222 	if (map) {
223 		for (i = 0; i < length; i++) {
224 			struct page *page = map[i];
225 			if (page)
226 				__free_page(page);
227 			if (!(i % SWAP_CLUSTER_MAX))
228 				cond_resched();
229 		}
230 		vfree(map);
231 	}
232 }
233