3 * Copyright IBM Corporation, 2012
4 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2.1 of the GNU Lesser General Public License
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
16 #include <linux/cgroup.h>
17 #include <linux/slab.h>
18 #include <linux/hugetlb.h>
19 #include <linux/hugetlb_cgroup.h>
21 struct hugetlb_cgroup {
22 struct cgroup_subsys_state css;
24 * the counter to account for hugepages from hugetlb.
26 struct res_counter hugepage[HUGE_MAX_HSTATE];
29 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
30 #define MEMFILE_IDX(val) (((val) >> 16) & 0xffff)
31 #define MEMFILE_ATTR(val) ((val) & 0xffff)
33 struct cgroup_subsys hugetlb_subsys __read_mostly;
34 static struct hugetlb_cgroup *root_h_cgroup __read_mostly;
37 struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s)
39 return container_of(s, struct hugetlb_cgroup, css);
43 struct hugetlb_cgroup *hugetlb_cgroup_from_cgroup(struct cgroup *cgroup)
45 return hugetlb_cgroup_from_css(cgroup_subsys_state(cgroup,
50 struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task)
52 return hugetlb_cgroup_from_css(task_subsys_state(task,
56 static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg)
58 return (h_cg == root_h_cgroup);
61 static inline struct hugetlb_cgroup *parent_hugetlb_cgroup(struct cgroup *cg)
65 return hugetlb_cgroup_from_cgroup(cg->parent);
68 static inline bool hugetlb_cgroup_have_usage(struct cgroup *cg)
71 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cg);
73 for (idx = 0; idx < hugetlb_max_hstate; idx++) {
74 if ((res_counter_read_u64(&h_cg->hugepage[idx], RES_USAGE)) > 0)
80 static struct cgroup_subsys_state *hugetlb_cgroup_create(struct cgroup *cgroup)
83 struct cgroup *parent_cgroup;
84 struct hugetlb_cgroup *h_cgroup, *parent_h_cgroup;
86 h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL);
88 return ERR_PTR(-ENOMEM);
90 parent_cgroup = cgroup->parent;
92 parent_h_cgroup = hugetlb_cgroup_from_cgroup(parent_cgroup);
93 for (idx = 0; idx < HUGE_MAX_HSTATE; idx++)
94 res_counter_init(&h_cgroup->hugepage[idx],
95 &parent_h_cgroup->hugepage[idx]);
97 root_h_cgroup = h_cgroup;
98 for (idx = 0; idx < HUGE_MAX_HSTATE; idx++)
99 res_counter_init(&h_cgroup->hugepage[idx], NULL);
101 return &h_cgroup->css;
104 static void hugetlb_cgroup_destroy(struct cgroup *cgroup)
106 struct hugetlb_cgroup *h_cgroup;
108 h_cgroup = hugetlb_cgroup_from_cgroup(cgroup);
114 * Should be called with hugetlb_lock held.
115 * Since we are holding hugetlb_lock, pages cannot get moved from
116 * active list or uncharged from the cgroup, So no need to get
117 * page reference and test for page active here. This function
120 static void hugetlb_cgroup_move_parent(int idx, struct cgroup *cgroup,
124 struct res_counter *counter;
125 struct res_counter *fail_res;
126 struct hugetlb_cgroup *page_hcg;
127 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup);
128 struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(cgroup);
130 page_hcg = hugetlb_cgroup_from_page(page);
132 * We can have pages in active list without any cgroup
133 * ie, hugepage with less than 3 pages. We can safely
134 * ignore those pages.
136 if (!page_hcg || page_hcg != h_cg)
139 csize = PAGE_SIZE << compound_order(page);
141 parent = root_h_cgroup;
142 /* root has no limit */
143 res_counter_charge_nofail(&parent->hugepage[idx],
146 counter = &h_cg->hugepage[idx];
147 res_counter_uncharge_until(counter, counter->parent, csize);
149 set_hugetlb_cgroup(page, parent);
155 * Force the hugetlb cgroup to empty the hugetlb resources by moving them to
158 static int hugetlb_cgroup_pre_destroy(struct cgroup *cgroup)
166 spin_lock(&hugetlb_lock);
167 list_for_each_entry(page, &h->hugepage_activelist, lru)
168 hugetlb_cgroup_move_parent(idx, cgroup, page);
170 spin_unlock(&hugetlb_lock);
174 } while (hugetlb_cgroup_have_usage(cgroup));
179 int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages,
180 struct hugetlb_cgroup **ptr)
183 struct res_counter *fail_res;
184 struct hugetlb_cgroup *h_cg = NULL;
185 unsigned long csize = nr_pages * PAGE_SIZE;
187 if (hugetlb_cgroup_disabled())
190 * We don't charge any cgroup if the compound page have less
193 if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
197 h_cg = hugetlb_cgroup_from_task(current);
198 if (!css_tryget(&h_cg->css)) {
204 ret = res_counter_charge(&h_cg->hugepage[idx], csize, &fail_res);
211 /* Should be called with hugetlb_lock held */
212 void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages,
213 struct hugetlb_cgroup *h_cg,
216 if (hugetlb_cgroup_disabled() || !h_cg)
219 set_hugetlb_cgroup(page, h_cg);
224 * Should be called with hugetlb_lock held
226 void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages,
229 struct hugetlb_cgroup *h_cg;
230 unsigned long csize = nr_pages * PAGE_SIZE;
232 if (hugetlb_cgroup_disabled())
234 VM_BUG_ON(!spin_is_locked(&hugetlb_lock));
235 h_cg = hugetlb_cgroup_from_page(page);
238 set_hugetlb_cgroup(page, NULL);
239 res_counter_uncharge(&h_cg->hugepage[idx], csize);
243 void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages,
244 struct hugetlb_cgroup *h_cg)
246 unsigned long csize = nr_pages * PAGE_SIZE;
248 if (hugetlb_cgroup_disabled() || !h_cg)
251 if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
254 res_counter_uncharge(&h_cg->hugepage[idx], csize);
258 static ssize_t hugetlb_cgroup_read(struct cgroup *cgroup, struct cftype *cft,
259 struct file *file, char __user *buf,
260 size_t nbytes, loff_t *ppos)
265 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup);
267 idx = MEMFILE_IDX(cft->private);
268 name = MEMFILE_ATTR(cft->private);
270 val = res_counter_read_u64(&h_cg->hugepage[idx], name);
271 len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
272 return simple_read_from_buffer(buf, nbytes, ppos, str, len);
275 static int hugetlb_cgroup_write(struct cgroup *cgroup, struct cftype *cft,
279 unsigned long long val;
280 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup);
282 idx = MEMFILE_IDX(cft->private);
283 name = MEMFILE_ATTR(cft->private);
287 if (hugetlb_cgroup_is_root(h_cg)) {
288 /* Can't set limit on root */
292 /* This function does all necessary parse...reuse it */
293 ret = res_counter_memparse_write_strategy(buffer, &val);
296 ret = res_counter_set_limit(&h_cg->hugepage[idx], val);
305 static int hugetlb_cgroup_reset(struct cgroup *cgroup, unsigned int event)
307 int idx, name, ret = 0;
308 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup);
310 idx = MEMFILE_IDX(event);
311 name = MEMFILE_ATTR(event);
315 res_counter_reset_max(&h_cg->hugepage[idx]);
318 res_counter_reset_failcnt(&h_cg->hugepage[idx]);
327 static char *mem_fmt(char *buf, int size, unsigned long hsize)
329 if (hsize >= (1UL << 30))
330 snprintf(buf, size, "%luGB", hsize >> 30);
331 else if (hsize >= (1UL << 20))
332 snprintf(buf, size, "%luMB", hsize >> 20);
334 snprintf(buf, size, "%luKB", hsize >> 10);
338 int __init hugetlb_cgroup_file_init(int idx)
342 struct hstate *h = &hstates[idx];
344 /* format the size */
345 mem_fmt(buf, 32, huge_page_size(h));
347 /* Add the limit file */
348 cft = &h->cgroup_files[0];
349 snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf);
350 cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT);
351 cft->read = hugetlb_cgroup_read;
352 cft->write_string = hugetlb_cgroup_write;
354 /* Add the usage file */
355 cft = &h->cgroup_files[1];
356 snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf);
357 cft->private = MEMFILE_PRIVATE(idx, RES_USAGE);
358 cft->read = hugetlb_cgroup_read;
360 /* Add the MAX usage file */
361 cft = &h->cgroup_files[2];
362 snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf);
363 cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE);
364 cft->trigger = hugetlb_cgroup_reset;
365 cft->read = hugetlb_cgroup_read;
367 /* Add the failcntfile */
368 cft = &h->cgroup_files[3];
369 snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf);
370 cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT);
371 cft->trigger = hugetlb_cgroup_reset;
372 cft->read = hugetlb_cgroup_read;
374 /* NULL terminate the last cft */
375 cft = &h->cgroup_files[4];
376 memset(cft, 0, sizeof(*cft));
378 WARN_ON(cgroup_add_cftypes(&hugetlb_subsys, h->cgroup_files));
384 * hugetlb_lock will make sure a parallel cgroup rmdir won't happen
385 * when we migrate hugepages
387 void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage)
389 struct hugetlb_cgroup *h_cg;
390 struct hstate *h = page_hstate(oldhpage);
392 if (hugetlb_cgroup_disabled())
395 VM_BUG_ON(!PageHuge(oldhpage));
396 spin_lock(&hugetlb_lock);
397 h_cg = hugetlb_cgroup_from_page(oldhpage);
398 set_hugetlb_cgroup(oldhpage, NULL);
400 /* move the h_cg details to new cgroup */
401 set_hugetlb_cgroup(newhpage, h_cg);
402 list_move(&newhpage->lru, &h->hugepage_activelist);
403 spin_unlock(&hugetlb_lock);
407 struct cgroup_subsys hugetlb_subsys = {
409 .create = hugetlb_cgroup_create,
410 .pre_destroy = hugetlb_cgroup_pre_destroy,
411 .destroy = hugetlb_cgroup_destroy,
412 .subsys_id = hugetlb_subsys_id,