Commit | Line | Data |
---|---|---|
2bc64a20 AK |
1 | /* |
2 | * | |
3 | * Copyright IBM Corporation, 2012 | |
4 | * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> | |
5 | * | |
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. | |
9 | * | |
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. | |
13 | * | |
14 | */ | |
15 | ||
16 | #include <linux/cgroup.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/hugetlb.h> | |
19 | #include <linux/hugetlb_cgroup.h> | |
20 | ||
21 | struct hugetlb_cgroup { | |
22 | struct cgroup_subsys_state css; | |
23 | /* | |
24 | * the counter to account for hugepages from hugetlb. | |
25 | */ | |
26 | struct res_counter hugepage[HUGE_MAX_HSTATE]; | |
27 | }; | |
28 | ||
abb8206c AK |
29 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
30 | #define MEMFILE_IDX(val) (((val) >> 16) & 0xffff) | |
31 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
32 | ||
2bc64a20 AK |
33 | struct cgroup_subsys hugetlb_subsys __read_mostly; |
34 | static struct hugetlb_cgroup *root_h_cgroup __read_mostly; | |
35 | ||
36 | static inline | |
37 | struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s) | |
38 | { | |
a7c6d554 | 39 | return s ? container_of(s, struct hugetlb_cgroup, css) : NULL; |
2bc64a20 AK |
40 | } |
41 | ||
42 | static inline | |
43 | struct hugetlb_cgroup *hugetlb_cgroup_from_cgroup(struct cgroup *cgroup) | |
44 | { | |
8af01f56 | 45 | return hugetlb_cgroup_from_css(cgroup_css(cgroup, hugetlb_subsys_id)); |
2bc64a20 AK |
46 | } |
47 | ||
48 | static inline | |
49 | struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task) | |
50 | { | |
8af01f56 | 51 | return hugetlb_cgroup_from_css(task_css(task, hugetlb_subsys_id)); |
2bc64a20 AK |
52 | } |
53 | ||
54 | static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg) | |
55 | { | |
56 | return (h_cg == root_h_cgroup); | |
57 | } | |
58 | ||
3f798518 TH |
59 | static inline struct hugetlb_cgroup * |
60 | parent_hugetlb_cgroup(struct hugetlb_cgroup *h_cg) | |
2bc64a20 | 61 | { |
63876986 | 62 | return hugetlb_cgroup_from_css(css_parent(&h_cg->css)); |
2bc64a20 AK |
63 | } |
64 | ||
3f798518 | 65 | static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg) |
2bc64a20 AK |
66 | { |
67 | int idx; | |
2bc64a20 AK |
68 | |
69 | for (idx = 0; idx < hugetlb_max_hstate; idx++) { | |
70 | if ((res_counter_read_u64(&h_cg->hugepage[idx], RES_USAGE)) > 0) | |
71 | return true; | |
72 | } | |
73 | return false; | |
74 | } | |
75 | ||
eb95419b TH |
76 | static struct cgroup_subsys_state * |
77 | hugetlb_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
2bc64a20 | 78 | { |
eb95419b TH |
79 | struct hugetlb_cgroup *parent_h_cgroup = hugetlb_cgroup_from_css(parent_css); |
80 | struct hugetlb_cgroup *h_cgroup; | |
2bc64a20 | 81 | int idx; |
2bc64a20 AK |
82 | |
83 | h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL); | |
84 | if (!h_cgroup) | |
85 | return ERR_PTR(-ENOMEM); | |
86 | ||
eb95419b | 87 | if (parent_h_cgroup) { |
2bc64a20 AK |
88 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) |
89 | res_counter_init(&h_cgroup->hugepage[idx], | |
90 | &parent_h_cgroup->hugepage[idx]); | |
91 | } else { | |
92 | root_h_cgroup = h_cgroup; | |
93 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) | |
94 | res_counter_init(&h_cgroup->hugepage[idx], NULL); | |
95 | } | |
96 | return &h_cgroup->css; | |
97 | } | |
98 | ||
eb95419b | 99 | static void hugetlb_cgroup_css_free(struct cgroup_subsys_state *css) |
2bc64a20 AK |
100 | { |
101 | struct hugetlb_cgroup *h_cgroup; | |
102 | ||
eb95419b | 103 | h_cgroup = hugetlb_cgroup_from_css(css); |
2bc64a20 AK |
104 | kfree(h_cgroup); |
105 | } | |
106 | ||
da1def55 AK |
107 | |
108 | /* | |
109 | * Should be called with hugetlb_lock held. | |
110 | * Since we are holding hugetlb_lock, pages cannot get moved from | |
111 | * active list or uncharged from the cgroup, So no need to get | |
112 | * page reference and test for page active here. This function | |
113 | * cannot fail. | |
114 | */ | |
3f798518 | 115 | static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg, |
da1def55 AK |
116 | struct page *page) |
117 | { | |
118 | int csize; | |
119 | struct res_counter *counter; | |
120 | struct res_counter *fail_res; | |
121 | struct hugetlb_cgroup *page_hcg; | |
3f798518 | 122 | struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(h_cg); |
da1def55 AK |
123 | |
124 | page_hcg = hugetlb_cgroup_from_page(page); | |
125 | /* | |
126 | * We can have pages in active list without any cgroup | |
127 | * ie, hugepage with less than 3 pages. We can safely | |
128 | * ignore those pages. | |
129 | */ | |
130 | if (!page_hcg || page_hcg != h_cg) | |
131 | goto out; | |
132 | ||
133 | csize = PAGE_SIZE << compound_order(page); | |
134 | if (!parent) { | |
135 | parent = root_h_cgroup; | |
136 | /* root has no limit */ | |
137 | res_counter_charge_nofail(&parent->hugepage[idx], | |
138 | csize, &fail_res); | |
139 | } | |
140 | counter = &h_cg->hugepage[idx]; | |
141 | res_counter_uncharge_until(counter, counter->parent, csize); | |
142 | ||
143 | set_hugetlb_cgroup(page, parent); | |
144 | out: | |
145 | return; | |
146 | } | |
147 | ||
148 | /* | |
149 | * Force the hugetlb cgroup to empty the hugetlb resources by moving them to | |
150 | * the parent cgroup. | |
151 | */ | |
eb95419b | 152 | static void hugetlb_cgroup_css_offline(struct cgroup_subsys_state *css) |
2bc64a20 | 153 | { |
eb95419b | 154 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css); |
da1def55 AK |
155 | struct hstate *h; |
156 | struct page *page; | |
9d093cb1 | 157 | int idx = 0; |
da1def55 AK |
158 | |
159 | do { | |
da1def55 AK |
160 | for_each_hstate(h) { |
161 | spin_lock(&hugetlb_lock); | |
162 | list_for_each_entry(page, &h->hugepage_activelist, lru) | |
3f798518 | 163 | hugetlb_cgroup_move_parent(idx, h_cg, page); |
da1def55 AK |
164 | |
165 | spin_unlock(&hugetlb_lock); | |
166 | idx++; | |
167 | } | |
168 | cond_resched(); | |
3f798518 | 169 | } while (hugetlb_cgroup_have_usage(h_cg)); |
2bc64a20 AK |
170 | } |
171 | ||
6d76dcf4 AK |
172 | int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, |
173 | struct hugetlb_cgroup **ptr) | |
174 | { | |
175 | int ret = 0; | |
176 | struct res_counter *fail_res; | |
177 | struct hugetlb_cgroup *h_cg = NULL; | |
178 | unsigned long csize = nr_pages * PAGE_SIZE; | |
179 | ||
180 | if (hugetlb_cgroup_disabled()) | |
181 | goto done; | |
182 | /* | |
183 | * We don't charge any cgroup if the compound page have less | |
184 | * than 3 pages. | |
185 | */ | |
186 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
187 | goto done; | |
188 | again: | |
189 | rcu_read_lock(); | |
190 | h_cg = hugetlb_cgroup_from_task(current); | |
191 | if (!css_tryget(&h_cg->css)) { | |
192 | rcu_read_unlock(); | |
193 | goto again; | |
194 | } | |
195 | rcu_read_unlock(); | |
196 | ||
197 | ret = res_counter_charge(&h_cg->hugepage[idx], csize, &fail_res); | |
198 | css_put(&h_cg->css); | |
199 | done: | |
200 | *ptr = h_cg; | |
201 | return ret; | |
202 | } | |
203 | ||
94ae8ba7 | 204 | /* Should be called with hugetlb_lock held */ |
6d76dcf4 AK |
205 | void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, |
206 | struct hugetlb_cgroup *h_cg, | |
207 | struct page *page) | |
208 | { | |
209 | if (hugetlb_cgroup_disabled() || !h_cg) | |
210 | return; | |
211 | ||
6d76dcf4 | 212 | set_hugetlb_cgroup(page, h_cg); |
6d76dcf4 AK |
213 | return; |
214 | } | |
215 | ||
216 | /* | |
217 | * Should be called with hugetlb_lock held | |
218 | */ | |
219 | void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, | |
220 | struct page *page) | |
221 | { | |
222 | struct hugetlb_cgroup *h_cg; | |
223 | unsigned long csize = nr_pages * PAGE_SIZE; | |
224 | ||
225 | if (hugetlb_cgroup_disabled()) | |
226 | return; | |
227 | VM_BUG_ON(!spin_is_locked(&hugetlb_lock)); | |
228 | h_cg = hugetlb_cgroup_from_page(page); | |
229 | if (unlikely(!h_cg)) | |
230 | return; | |
231 | set_hugetlb_cgroup(page, NULL); | |
232 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
233 | return; | |
234 | } | |
235 | ||
236 | void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, | |
237 | struct hugetlb_cgroup *h_cg) | |
238 | { | |
239 | unsigned long csize = nr_pages * PAGE_SIZE; | |
240 | ||
241 | if (hugetlb_cgroup_disabled() || !h_cg) | |
242 | return; | |
243 | ||
244 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
245 | return; | |
246 | ||
247 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
248 | return; | |
249 | } | |
250 | ||
abb8206c AK |
251 | static ssize_t hugetlb_cgroup_read(struct cgroup *cgroup, struct cftype *cft, |
252 | struct file *file, char __user *buf, | |
253 | size_t nbytes, loff_t *ppos) | |
254 | { | |
255 | u64 val; | |
256 | char str[64]; | |
257 | int idx, name, len; | |
258 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
259 | ||
260 | idx = MEMFILE_IDX(cft->private); | |
261 | name = MEMFILE_ATTR(cft->private); | |
262 | ||
263 | val = res_counter_read_u64(&h_cg->hugepage[idx], name); | |
264 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
265 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
266 | } | |
267 | ||
268 | static int hugetlb_cgroup_write(struct cgroup *cgroup, struct cftype *cft, | |
269 | const char *buffer) | |
270 | { | |
271 | int idx, name, ret; | |
272 | unsigned long long val; | |
273 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
274 | ||
275 | idx = MEMFILE_IDX(cft->private); | |
276 | name = MEMFILE_ATTR(cft->private); | |
277 | ||
278 | switch (name) { | |
279 | case RES_LIMIT: | |
280 | if (hugetlb_cgroup_is_root(h_cg)) { | |
281 | /* Can't set limit on root */ | |
282 | ret = -EINVAL; | |
283 | break; | |
284 | } | |
285 | /* This function does all necessary parse...reuse it */ | |
286 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
287 | if (ret) | |
288 | break; | |
289 | ret = res_counter_set_limit(&h_cg->hugepage[idx], val); | |
290 | break; | |
291 | default: | |
292 | ret = -EINVAL; | |
293 | break; | |
294 | } | |
295 | return ret; | |
296 | } | |
297 | ||
298 | static int hugetlb_cgroup_reset(struct cgroup *cgroup, unsigned int event) | |
299 | { | |
300 | int idx, name, ret = 0; | |
301 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
302 | ||
303 | idx = MEMFILE_IDX(event); | |
304 | name = MEMFILE_ATTR(event); | |
305 | ||
306 | switch (name) { | |
307 | case RES_MAX_USAGE: | |
308 | res_counter_reset_max(&h_cg->hugepage[idx]); | |
309 | break; | |
310 | case RES_FAILCNT: | |
311 | res_counter_reset_failcnt(&h_cg->hugepage[idx]); | |
312 | break; | |
313 | default: | |
314 | ret = -EINVAL; | |
315 | break; | |
316 | } | |
317 | return ret; | |
318 | } | |
319 | ||
320 | static char *mem_fmt(char *buf, int size, unsigned long hsize) | |
321 | { | |
322 | if (hsize >= (1UL << 30)) | |
323 | snprintf(buf, size, "%luGB", hsize >> 30); | |
324 | else if (hsize >= (1UL << 20)) | |
325 | snprintf(buf, size, "%luMB", hsize >> 20); | |
326 | else | |
327 | snprintf(buf, size, "%luKB", hsize >> 10); | |
328 | return buf; | |
329 | } | |
330 | ||
7179e7bf | 331 | static void __init __hugetlb_cgroup_file_init(int idx) |
abb8206c AK |
332 | { |
333 | char buf[32]; | |
334 | struct cftype *cft; | |
335 | struct hstate *h = &hstates[idx]; | |
336 | ||
337 | /* format the size */ | |
338 | mem_fmt(buf, 32, huge_page_size(h)); | |
339 | ||
340 | /* Add the limit file */ | |
341 | cft = &h->cgroup_files[0]; | |
342 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf); | |
343 | cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT); | |
344 | cft->read = hugetlb_cgroup_read; | |
345 | cft->write_string = hugetlb_cgroup_write; | |
346 | ||
347 | /* Add the usage file */ | |
348 | cft = &h->cgroup_files[1]; | |
349 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf); | |
350 | cft->private = MEMFILE_PRIVATE(idx, RES_USAGE); | |
351 | cft->read = hugetlb_cgroup_read; | |
352 | ||
353 | /* Add the MAX usage file */ | |
354 | cft = &h->cgroup_files[2]; | |
355 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf); | |
356 | cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE); | |
357 | cft->trigger = hugetlb_cgroup_reset; | |
358 | cft->read = hugetlb_cgroup_read; | |
359 | ||
360 | /* Add the failcntfile */ | |
361 | cft = &h->cgroup_files[3]; | |
362 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf); | |
363 | cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT); | |
364 | cft->trigger = hugetlb_cgroup_reset; | |
365 | cft->read = hugetlb_cgroup_read; | |
366 | ||
367 | /* NULL terminate the last cft */ | |
368 | cft = &h->cgroup_files[4]; | |
369 | memset(cft, 0, sizeof(*cft)); | |
370 | ||
371 | WARN_ON(cgroup_add_cftypes(&hugetlb_subsys, h->cgroup_files)); | |
372 | ||
7179e7bf JW |
373 | return; |
374 | } | |
375 | ||
376 | void __init hugetlb_cgroup_file_init(void) | |
377 | { | |
378 | struct hstate *h; | |
379 | ||
380 | for_each_hstate(h) { | |
381 | /* | |
382 | * Add cgroup control files only if the huge page consists | |
383 | * of more than two normal pages. This is because we use | |
384 | * page[2].lru.next for storing cgroup details. | |
385 | */ | |
386 | if (huge_page_order(h) >= HUGETLB_CGROUP_MIN_ORDER) | |
387 | __hugetlb_cgroup_file_init(hstate_index(h)); | |
388 | } | |
abb8206c AK |
389 | } |
390 | ||
75754681 AK |
391 | /* |
392 | * hugetlb_lock will make sure a parallel cgroup rmdir won't happen | |
393 | * when we migrate hugepages | |
394 | */ | |
8e6ac7fa AK |
395 | void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage) |
396 | { | |
397 | struct hugetlb_cgroup *h_cg; | |
94ae8ba7 | 398 | struct hstate *h = page_hstate(oldhpage); |
8e6ac7fa AK |
399 | |
400 | if (hugetlb_cgroup_disabled()) | |
401 | return; | |
402 | ||
403 | VM_BUG_ON(!PageHuge(oldhpage)); | |
404 | spin_lock(&hugetlb_lock); | |
405 | h_cg = hugetlb_cgroup_from_page(oldhpage); | |
406 | set_hugetlb_cgroup(oldhpage, NULL); | |
8e6ac7fa AK |
407 | |
408 | /* move the h_cg details to new cgroup */ | |
409 | set_hugetlb_cgroup(newhpage, h_cg); | |
94ae8ba7 | 410 | list_move(&newhpage->lru, &h->hugepage_activelist); |
8e6ac7fa | 411 | spin_unlock(&hugetlb_lock); |
8e6ac7fa AK |
412 | return; |
413 | } | |
414 | ||
2bc64a20 AK |
415 | struct cgroup_subsys hugetlb_subsys = { |
416 | .name = "hugetlb", | |
92fb9748 TH |
417 | .css_alloc = hugetlb_cgroup_css_alloc, |
418 | .css_offline = hugetlb_cgroup_css_offline, | |
419 | .css_free = hugetlb_cgroup_css_free, | |
420 | .subsys_id = hugetlb_subsys_id, | |
2bc64a20 | 421 | }; |