1 #include <linux/swap_cgroup.h>
2 #include <linux/vmalloc.h>
5 #include <linux/swapops.h> /* depends on mm.h include */
7 static DEFINE_MUTEX(swap_cgroup_mutex);
8 struct swap_cgroup_ctrl {
14 static struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES];
19 #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup))
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.
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".
32 * TODO: we can push these buffers out to HIGHMEM.
36 * allocate buffer for swap_cgroup.
38 static int swap_cgroup_prepare(int type)
41 struct swap_cgroup_ctrl *ctrl;
42 unsigned long idx, max;
44 ctrl = &swap_cgroup_ctrl[type];
46 for (idx = 0; idx < ctrl->length; idx++) {
47 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
50 ctrl->map[idx] = page;
55 for (idx = 0; idx < max; idx++)
56 __free_page(ctrl->map[idx]);
61 static struct swap_cgroup *lookup_swap_cgroup(swp_entry_t ent,
62 struct swap_cgroup_ctrl **ctrlp)
64 pgoff_t offset = swp_offset(ent);
65 struct swap_cgroup_ctrl *ctrl;
67 struct swap_cgroup *sc;
69 ctrl = &swap_cgroup_ctrl[swp_type(ent)];
73 mappage = ctrl->map[offset / SC_PER_PAGE];
74 sc = page_address(mappage);
75 return sc + offset % SC_PER_PAGE;
79 * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry.
80 * @ent: swap entry to be cmpxchged
84 * Returns old id at success, 0 at failure.
85 * (There is no mem_cgroup using 0 as its id)
87 unsigned short swap_cgroup_cmpxchg(swp_entry_t ent,
88 unsigned short old, unsigned short new)
90 struct swap_cgroup_ctrl *ctrl;
91 struct swap_cgroup *sc;
93 unsigned short retval;
95 sc = lookup_swap_cgroup(ent, &ctrl);
97 spin_lock_irqsave(&ctrl->lock, flags);
103 spin_unlock_irqrestore(&ctrl->lock, flags);
108 * swap_cgroup_record - record mem_cgroup for this swp_entry.
109 * @ent: swap entry to be recorded into
110 * @id: mem_cgroup to be recorded
112 * Returns old value at success, 0 at failure.
113 * (Of course, old value can be 0.)
115 unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id)
117 struct swap_cgroup_ctrl *ctrl;
118 struct swap_cgroup *sc;
122 sc = lookup_swap_cgroup(ent, &ctrl);
124 spin_lock_irqsave(&ctrl->lock, flags);
127 spin_unlock_irqrestore(&ctrl->lock, flags);
133 * lookup_swap_cgroup_id - lookup mem_cgroup id tied to swap entry
134 * @ent: swap entry to be looked up.
136 * Returns ID of mem_cgroup at success. 0 at failure. (0 is invalid ID)
138 unsigned short lookup_swap_cgroup_id(swp_entry_t ent)
140 return lookup_swap_cgroup(ent, NULL)->id;
143 int swap_cgroup_swapon(int type, unsigned long max_pages)
146 unsigned long array_size;
147 unsigned long length;
148 struct swap_cgroup_ctrl *ctrl;
150 if (!do_swap_account)
153 length = DIV_ROUND_UP(max_pages, SC_PER_PAGE);
154 array_size = length * sizeof(void *);
156 array = vzalloc(array_size);
160 ctrl = &swap_cgroup_ctrl[type];
161 mutex_lock(&swap_cgroup_mutex);
162 ctrl->length = length;
164 spin_lock_init(&ctrl->lock);
165 if (swap_cgroup_prepare(type)) {
166 /* memory shortage */
169 mutex_unlock(&swap_cgroup_mutex);
173 mutex_unlock(&swap_cgroup_mutex);
177 pr_info("couldn't allocate enough memory for swap_cgroup\n");
178 pr_info("swap_cgroup can be disabled by swapaccount=0 boot option\n");
182 void swap_cgroup_swapoff(int type)
185 unsigned long i, length;
186 struct swap_cgroup_ctrl *ctrl;
188 if (!do_swap_account)
191 mutex_lock(&swap_cgroup_mutex);
192 ctrl = &swap_cgroup_ctrl[type];
194 length = ctrl->length;
197 mutex_unlock(&swap_cgroup_mutex);
200 for (i = 0; i < length; i++) {
201 struct page *page = map[i];