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 | static struct hugetlb_cgroup *root_h_cgroup __read_mostly; |
34 | ||
35 | static inline | |
36 | struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s) | |
37 | { | |
a7c6d554 | 38 | return s ? container_of(s, struct hugetlb_cgroup, css) : NULL; |
2bc64a20 AK |
39 | } |
40 | ||
2bc64a20 AK |
41 | static inline |
42 | struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task) | |
43 | { | |
073219e9 | 44 | return hugetlb_cgroup_from_css(task_css(task, hugetlb_cgrp_id)); |
2bc64a20 AK |
45 | } |
46 | ||
47 | static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg) | |
48 | { | |
49 | return (h_cg == root_h_cgroup); | |
50 | } | |
51 | ||
3f798518 TH |
52 | static inline struct hugetlb_cgroup * |
53 | parent_hugetlb_cgroup(struct hugetlb_cgroup *h_cg) | |
2bc64a20 | 54 | { |
63876986 | 55 | return hugetlb_cgroup_from_css(css_parent(&h_cg->css)); |
2bc64a20 AK |
56 | } |
57 | ||
3f798518 | 58 | static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg) |
2bc64a20 AK |
59 | { |
60 | int idx; | |
2bc64a20 AK |
61 | |
62 | for (idx = 0; idx < hugetlb_max_hstate; idx++) { | |
63 | if ((res_counter_read_u64(&h_cg->hugepage[idx], RES_USAGE)) > 0) | |
64 | return true; | |
65 | } | |
66 | return false; | |
67 | } | |
68 | ||
eb95419b TH |
69 | static struct cgroup_subsys_state * |
70 | hugetlb_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
2bc64a20 | 71 | { |
eb95419b TH |
72 | struct hugetlb_cgroup *parent_h_cgroup = hugetlb_cgroup_from_css(parent_css); |
73 | struct hugetlb_cgroup *h_cgroup; | |
2bc64a20 | 74 | int idx; |
2bc64a20 AK |
75 | |
76 | h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL); | |
77 | if (!h_cgroup) | |
78 | return ERR_PTR(-ENOMEM); | |
79 | ||
eb95419b | 80 | if (parent_h_cgroup) { |
2bc64a20 AK |
81 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) |
82 | res_counter_init(&h_cgroup->hugepage[idx], | |
83 | &parent_h_cgroup->hugepage[idx]); | |
84 | } else { | |
85 | root_h_cgroup = h_cgroup; | |
86 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) | |
87 | res_counter_init(&h_cgroup->hugepage[idx], NULL); | |
88 | } | |
89 | return &h_cgroup->css; | |
90 | } | |
91 | ||
eb95419b | 92 | static void hugetlb_cgroup_css_free(struct cgroup_subsys_state *css) |
2bc64a20 AK |
93 | { |
94 | struct hugetlb_cgroup *h_cgroup; | |
95 | ||
eb95419b | 96 | h_cgroup = hugetlb_cgroup_from_css(css); |
2bc64a20 AK |
97 | kfree(h_cgroup); |
98 | } | |
99 | ||
da1def55 AK |
100 | |
101 | /* | |
102 | * Should be called with hugetlb_lock held. | |
103 | * Since we are holding hugetlb_lock, pages cannot get moved from | |
104 | * active list or uncharged from the cgroup, So no need to get | |
105 | * page reference and test for page active here. This function | |
106 | * cannot fail. | |
107 | */ | |
3f798518 | 108 | static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg, |
da1def55 AK |
109 | struct page *page) |
110 | { | |
111 | int csize; | |
112 | struct res_counter *counter; | |
113 | struct res_counter *fail_res; | |
114 | struct hugetlb_cgroup *page_hcg; | |
3f798518 | 115 | struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(h_cg); |
da1def55 AK |
116 | |
117 | page_hcg = hugetlb_cgroup_from_page(page); | |
118 | /* | |
119 | * We can have pages in active list without any cgroup | |
120 | * ie, hugepage with less than 3 pages. We can safely | |
121 | * ignore those pages. | |
122 | */ | |
123 | if (!page_hcg || page_hcg != h_cg) | |
124 | goto out; | |
125 | ||
126 | csize = PAGE_SIZE << compound_order(page); | |
127 | if (!parent) { | |
128 | parent = root_h_cgroup; | |
129 | /* root has no limit */ | |
130 | res_counter_charge_nofail(&parent->hugepage[idx], | |
131 | csize, &fail_res); | |
132 | } | |
133 | counter = &h_cg->hugepage[idx]; | |
134 | res_counter_uncharge_until(counter, counter->parent, csize); | |
135 | ||
136 | set_hugetlb_cgroup(page, parent); | |
137 | out: | |
138 | return; | |
139 | } | |
140 | ||
141 | /* | |
142 | * Force the hugetlb cgroup to empty the hugetlb resources by moving them to | |
143 | * the parent cgroup. | |
144 | */ | |
eb95419b | 145 | static void hugetlb_cgroup_css_offline(struct cgroup_subsys_state *css) |
2bc64a20 | 146 | { |
eb95419b | 147 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css); |
da1def55 AK |
148 | struct hstate *h; |
149 | struct page *page; | |
9d093cb1 | 150 | int idx = 0; |
da1def55 AK |
151 | |
152 | do { | |
da1def55 AK |
153 | for_each_hstate(h) { |
154 | spin_lock(&hugetlb_lock); | |
155 | list_for_each_entry(page, &h->hugepage_activelist, lru) | |
3f798518 | 156 | hugetlb_cgroup_move_parent(idx, h_cg, page); |
da1def55 AK |
157 | |
158 | spin_unlock(&hugetlb_lock); | |
159 | idx++; | |
160 | } | |
161 | cond_resched(); | |
3f798518 | 162 | } while (hugetlb_cgroup_have_usage(h_cg)); |
2bc64a20 AK |
163 | } |
164 | ||
6d76dcf4 AK |
165 | int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, |
166 | struct hugetlb_cgroup **ptr) | |
167 | { | |
168 | int ret = 0; | |
169 | struct res_counter *fail_res; | |
170 | struct hugetlb_cgroup *h_cg = NULL; | |
171 | unsigned long csize = nr_pages * PAGE_SIZE; | |
172 | ||
173 | if (hugetlb_cgroup_disabled()) | |
174 | goto done; | |
175 | /* | |
176 | * We don't charge any cgroup if the compound page have less | |
177 | * than 3 pages. | |
178 | */ | |
179 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
180 | goto done; | |
181 | again: | |
182 | rcu_read_lock(); | |
183 | h_cg = hugetlb_cgroup_from_task(current); | |
184 | if (!css_tryget(&h_cg->css)) { | |
185 | rcu_read_unlock(); | |
186 | goto again; | |
187 | } | |
188 | rcu_read_unlock(); | |
189 | ||
190 | ret = res_counter_charge(&h_cg->hugepage[idx], csize, &fail_res); | |
191 | css_put(&h_cg->css); | |
192 | done: | |
193 | *ptr = h_cg; | |
194 | return ret; | |
195 | } | |
196 | ||
94ae8ba7 | 197 | /* Should be called with hugetlb_lock held */ |
6d76dcf4 AK |
198 | void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, |
199 | struct hugetlb_cgroup *h_cg, | |
200 | struct page *page) | |
201 | { | |
202 | if (hugetlb_cgroup_disabled() || !h_cg) | |
203 | return; | |
204 | ||
6d76dcf4 | 205 | set_hugetlb_cgroup(page, h_cg); |
6d76dcf4 AK |
206 | return; |
207 | } | |
208 | ||
209 | /* | |
210 | * Should be called with hugetlb_lock held | |
211 | */ | |
212 | void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, | |
213 | struct page *page) | |
214 | { | |
215 | struct hugetlb_cgroup *h_cg; | |
216 | unsigned long csize = nr_pages * PAGE_SIZE; | |
217 | ||
218 | if (hugetlb_cgroup_disabled()) | |
219 | return; | |
220 | VM_BUG_ON(!spin_is_locked(&hugetlb_lock)); | |
221 | h_cg = hugetlb_cgroup_from_page(page); | |
222 | if (unlikely(!h_cg)) | |
223 | return; | |
224 | set_hugetlb_cgroup(page, NULL); | |
225 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
226 | return; | |
227 | } | |
228 | ||
229 | void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, | |
230 | struct hugetlb_cgroup *h_cg) | |
231 | { | |
232 | unsigned long csize = nr_pages * PAGE_SIZE; | |
233 | ||
234 | if (hugetlb_cgroup_disabled() || !h_cg) | |
235 | return; | |
236 | ||
237 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
238 | return; | |
239 | ||
240 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
241 | return; | |
242 | } | |
243 | ||
716f479d TH |
244 | static u64 hugetlb_cgroup_read_u64(struct cgroup_subsys_state *css, |
245 | struct cftype *cft) | |
abb8206c | 246 | { |
716f479d | 247 | int idx, name; |
182446d0 | 248 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css); |
abb8206c AK |
249 | |
250 | idx = MEMFILE_IDX(cft->private); | |
251 | name = MEMFILE_ATTR(cft->private); | |
252 | ||
716f479d | 253 | return res_counter_read_u64(&h_cg->hugepage[idx], name); |
abb8206c AK |
254 | } |
255 | ||
182446d0 | 256 | static int hugetlb_cgroup_write(struct cgroup_subsys_state *css, |
4d3bb511 | 257 | struct cftype *cft, char *buffer) |
abb8206c AK |
258 | { |
259 | int idx, name, ret; | |
260 | unsigned long long val; | |
182446d0 | 261 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css); |
abb8206c AK |
262 | |
263 | idx = MEMFILE_IDX(cft->private); | |
264 | name = MEMFILE_ATTR(cft->private); | |
265 | ||
266 | switch (name) { | |
267 | case RES_LIMIT: | |
268 | if (hugetlb_cgroup_is_root(h_cg)) { | |
269 | /* Can't set limit on root */ | |
270 | ret = -EINVAL; | |
271 | break; | |
272 | } | |
273 | /* This function does all necessary parse...reuse it */ | |
274 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
275 | if (ret) | |
276 | break; | |
277 | ret = res_counter_set_limit(&h_cg->hugepage[idx], val); | |
278 | break; | |
279 | default: | |
280 | ret = -EINVAL; | |
281 | break; | |
282 | } | |
283 | return ret; | |
284 | } | |
285 | ||
182446d0 TH |
286 | static int hugetlb_cgroup_reset(struct cgroup_subsys_state *css, |
287 | unsigned int event) | |
abb8206c AK |
288 | { |
289 | int idx, name, ret = 0; | |
182446d0 | 290 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css); |
abb8206c AK |
291 | |
292 | idx = MEMFILE_IDX(event); | |
293 | name = MEMFILE_ATTR(event); | |
294 | ||
295 | switch (name) { | |
296 | case RES_MAX_USAGE: | |
297 | res_counter_reset_max(&h_cg->hugepage[idx]); | |
298 | break; | |
299 | case RES_FAILCNT: | |
300 | res_counter_reset_failcnt(&h_cg->hugepage[idx]); | |
301 | break; | |
302 | default: | |
303 | ret = -EINVAL; | |
304 | break; | |
305 | } | |
306 | return ret; | |
307 | } | |
308 | ||
309 | static char *mem_fmt(char *buf, int size, unsigned long hsize) | |
310 | { | |
311 | if (hsize >= (1UL << 30)) | |
312 | snprintf(buf, size, "%luGB", hsize >> 30); | |
313 | else if (hsize >= (1UL << 20)) | |
314 | snprintf(buf, size, "%luMB", hsize >> 20); | |
315 | else | |
316 | snprintf(buf, size, "%luKB", hsize >> 10); | |
317 | return buf; | |
318 | } | |
319 | ||
7179e7bf | 320 | static void __init __hugetlb_cgroup_file_init(int idx) |
abb8206c AK |
321 | { |
322 | char buf[32]; | |
323 | struct cftype *cft; | |
324 | struct hstate *h = &hstates[idx]; | |
325 | ||
326 | /* format the size */ | |
327 | mem_fmt(buf, 32, huge_page_size(h)); | |
328 | ||
329 | /* Add the limit file */ | |
330 | cft = &h->cgroup_files[0]; | |
331 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf); | |
332 | cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT); | |
716f479d | 333 | cft->read_u64 = hugetlb_cgroup_read_u64; |
abb8206c AK |
334 | cft->write_string = hugetlb_cgroup_write; |
335 | ||
336 | /* Add the usage file */ | |
337 | cft = &h->cgroup_files[1]; | |
338 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf); | |
339 | cft->private = MEMFILE_PRIVATE(idx, RES_USAGE); | |
716f479d | 340 | cft->read_u64 = hugetlb_cgroup_read_u64; |
abb8206c AK |
341 | |
342 | /* Add the MAX usage file */ | |
343 | cft = &h->cgroup_files[2]; | |
344 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf); | |
345 | cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE); | |
346 | cft->trigger = hugetlb_cgroup_reset; | |
716f479d | 347 | cft->read_u64 = hugetlb_cgroup_read_u64; |
abb8206c AK |
348 | |
349 | /* Add the failcntfile */ | |
350 | cft = &h->cgroup_files[3]; | |
351 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf); | |
352 | cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT); | |
353 | cft->trigger = hugetlb_cgroup_reset; | |
716f479d | 354 | cft->read_u64 = hugetlb_cgroup_read_u64; |
abb8206c AK |
355 | |
356 | /* NULL terminate the last cft */ | |
357 | cft = &h->cgroup_files[4]; | |
358 | memset(cft, 0, sizeof(*cft)); | |
359 | ||
073219e9 | 360 | WARN_ON(cgroup_add_cftypes(&hugetlb_cgrp_subsys, h->cgroup_files)); |
abb8206c | 361 | |
7179e7bf JW |
362 | return; |
363 | } | |
364 | ||
365 | void __init hugetlb_cgroup_file_init(void) | |
366 | { | |
367 | struct hstate *h; | |
368 | ||
369 | for_each_hstate(h) { | |
370 | /* | |
371 | * Add cgroup control files only if the huge page consists | |
372 | * of more than two normal pages. This is because we use | |
373 | * page[2].lru.next for storing cgroup details. | |
374 | */ | |
375 | if (huge_page_order(h) >= HUGETLB_CGROUP_MIN_ORDER) | |
376 | __hugetlb_cgroup_file_init(hstate_index(h)); | |
377 | } | |
abb8206c AK |
378 | } |
379 | ||
75754681 AK |
380 | /* |
381 | * hugetlb_lock will make sure a parallel cgroup rmdir won't happen | |
382 | * when we migrate hugepages | |
383 | */ | |
8e6ac7fa AK |
384 | void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage) |
385 | { | |
386 | struct hugetlb_cgroup *h_cg; | |
94ae8ba7 | 387 | struct hstate *h = page_hstate(oldhpage); |
8e6ac7fa AK |
388 | |
389 | if (hugetlb_cgroup_disabled()) | |
390 | return; | |
391 | ||
309381fe | 392 | VM_BUG_ON_PAGE(!PageHuge(oldhpage), oldhpage); |
8e6ac7fa AK |
393 | spin_lock(&hugetlb_lock); |
394 | h_cg = hugetlb_cgroup_from_page(oldhpage); | |
395 | set_hugetlb_cgroup(oldhpage, NULL); | |
8e6ac7fa AK |
396 | |
397 | /* move the h_cg details to new cgroup */ | |
398 | set_hugetlb_cgroup(newhpage, h_cg); | |
94ae8ba7 | 399 | list_move(&newhpage->lru, &h->hugepage_activelist); |
8e6ac7fa | 400 | spin_unlock(&hugetlb_lock); |
8e6ac7fa AK |
401 | return; |
402 | } | |
403 | ||
073219e9 | 404 | struct cgroup_subsys hugetlb_cgrp_subsys = { |
92fb9748 TH |
405 | .css_alloc = hugetlb_cgroup_css_alloc, |
406 | .css_offline = hugetlb_cgroup_css_offline, | |
407 | .css_free = hugetlb_cgroup_css_free, | |
2bc64a20 | 408 | }; |