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 | { | |
39 | return container_of(s, struct hugetlb_cgroup, css); | |
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 | ||
59 | static inline struct hugetlb_cgroup *parent_hugetlb_cgroup(struct cgroup *cg) | |
60 | { | |
61 | if (!cg->parent) | |
62 | return NULL; | |
63 | return hugetlb_cgroup_from_cgroup(cg->parent); | |
64 | } | |
65 | ||
66 | static inline bool hugetlb_cgroup_have_usage(struct cgroup *cg) | |
67 | { | |
68 | int idx; | |
69 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cg); | |
70 | ||
71 | for (idx = 0; idx < hugetlb_max_hstate; idx++) { | |
72 | if ((res_counter_read_u64(&h_cg->hugepage[idx], RES_USAGE)) > 0) | |
73 | return true; | |
74 | } | |
75 | return false; | |
76 | } | |
77 | ||
92fb9748 | 78 | static struct cgroup_subsys_state *hugetlb_cgroup_css_alloc(struct cgroup *cgroup) |
2bc64a20 AK |
79 | { |
80 | int idx; | |
81 | struct cgroup *parent_cgroup; | |
82 | struct hugetlb_cgroup *h_cgroup, *parent_h_cgroup; | |
83 | ||
84 | h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL); | |
85 | if (!h_cgroup) | |
86 | return ERR_PTR(-ENOMEM); | |
87 | ||
88 | parent_cgroup = cgroup->parent; | |
89 | if (parent_cgroup) { | |
90 | parent_h_cgroup = hugetlb_cgroup_from_cgroup(parent_cgroup); | |
91 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) | |
92 | res_counter_init(&h_cgroup->hugepage[idx], | |
93 | &parent_h_cgroup->hugepage[idx]); | |
94 | } else { | |
95 | root_h_cgroup = h_cgroup; | |
96 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) | |
97 | res_counter_init(&h_cgroup->hugepage[idx], NULL); | |
98 | } | |
99 | return &h_cgroup->css; | |
100 | } | |
101 | ||
92fb9748 | 102 | static void hugetlb_cgroup_css_free(struct cgroup *cgroup) |
2bc64a20 AK |
103 | { |
104 | struct hugetlb_cgroup *h_cgroup; | |
105 | ||
106 | h_cgroup = hugetlb_cgroup_from_cgroup(cgroup); | |
107 | kfree(h_cgroup); | |
108 | } | |
109 | ||
da1def55 AK |
110 | |
111 | /* | |
112 | * Should be called with hugetlb_lock held. | |
113 | * Since we are holding hugetlb_lock, pages cannot get moved from | |
114 | * active list or uncharged from the cgroup, So no need to get | |
115 | * page reference and test for page active here. This function | |
116 | * cannot fail. | |
117 | */ | |
118 | static void hugetlb_cgroup_move_parent(int idx, struct cgroup *cgroup, | |
119 | struct page *page) | |
120 | { | |
121 | int csize; | |
122 | struct res_counter *counter; | |
123 | struct res_counter *fail_res; | |
124 | struct hugetlb_cgroup *page_hcg; | |
125 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
126 | struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(cgroup); | |
127 | ||
128 | page_hcg = hugetlb_cgroup_from_page(page); | |
129 | /* | |
130 | * We can have pages in active list without any cgroup | |
131 | * ie, hugepage with less than 3 pages. We can safely | |
132 | * ignore those pages. | |
133 | */ | |
134 | if (!page_hcg || page_hcg != h_cg) | |
135 | goto out; | |
136 | ||
137 | csize = PAGE_SIZE << compound_order(page); | |
138 | if (!parent) { | |
139 | parent = root_h_cgroup; | |
140 | /* root has no limit */ | |
141 | res_counter_charge_nofail(&parent->hugepage[idx], | |
142 | csize, &fail_res); | |
143 | } | |
144 | counter = &h_cg->hugepage[idx]; | |
145 | res_counter_uncharge_until(counter, counter->parent, csize); | |
146 | ||
147 | set_hugetlb_cgroup(page, parent); | |
148 | out: | |
149 | return; | |
150 | } | |
151 | ||
152 | /* | |
153 | * Force the hugetlb cgroup to empty the hugetlb resources by moving them to | |
154 | * the parent cgroup. | |
155 | */ | |
92fb9748 | 156 | static void hugetlb_cgroup_css_offline(struct cgroup *cgroup) |
2bc64a20 | 157 | { |
da1def55 AK |
158 | struct hstate *h; |
159 | struct page *page; | |
9d093cb1 | 160 | int idx = 0; |
da1def55 AK |
161 | |
162 | do { | |
da1def55 AK |
163 | for_each_hstate(h) { |
164 | spin_lock(&hugetlb_lock); | |
165 | list_for_each_entry(page, &h->hugepage_activelist, lru) | |
166 | hugetlb_cgroup_move_parent(idx, cgroup, page); | |
167 | ||
168 | spin_unlock(&hugetlb_lock); | |
169 | idx++; | |
170 | } | |
171 | cond_resched(); | |
172 | } while (hugetlb_cgroup_have_usage(cgroup)); | |
2bc64a20 AK |
173 | } |
174 | ||
6d76dcf4 AK |
175 | int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, |
176 | struct hugetlb_cgroup **ptr) | |
177 | { | |
178 | int ret = 0; | |
179 | struct res_counter *fail_res; | |
180 | struct hugetlb_cgroup *h_cg = NULL; | |
181 | unsigned long csize = nr_pages * PAGE_SIZE; | |
182 | ||
183 | if (hugetlb_cgroup_disabled()) | |
184 | goto done; | |
185 | /* | |
186 | * We don't charge any cgroup if the compound page have less | |
187 | * than 3 pages. | |
188 | */ | |
189 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
190 | goto done; | |
191 | again: | |
192 | rcu_read_lock(); | |
193 | h_cg = hugetlb_cgroup_from_task(current); | |
194 | if (!css_tryget(&h_cg->css)) { | |
195 | rcu_read_unlock(); | |
196 | goto again; | |
197 | } | |
198 | rcu_read_unlock(); | |
199 | ||
200 | ret = res_counter_charge(&h_cg->hugepage[idx], csize, &fail_res); | |
201 | css_put(&h_cg->css); | |
202 | done: | |
203 | *ptr = h_cg; | |
204 | return ret; | |
205 | } | |
206 | ||
94ae8ba7 | 207 | /* Should be called with hugetlb_lock held */ |
6d76dcf4 AK |
208 | void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, |
209 | struct hugetlb_cgroup *h_cg, | |
210 | struct page *page) | |
211 | { | |
212 | if (hugetlb_cgroup_disabled() || !h_cg) | |
213 | return; | |
214 | ||
6d76dcf4 | 215 | set_hugetlb_cgroup(page, h_cg); |
6d76dcf4 AK |
216 | return; |
217 | } | |
218 | ||
219 | /* | |
220 | * Should be called with hugetlb_lock held | |
221 | */ | |
222 | void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, | |
223 | struct page *page) | |
224 | { | |
225 | struct hugetlb_cgroup *h_cg; | |
226 | unsigned long csize = nr_pages * PAGE_SIZE; | |
227 | ||
228 | if (hugetlb_cgroup_disabled()) | |
229 | return; | |
230 | VM_BUG_ON(!spin_is_locked(&hugetlb_lock)); | |
231 | h_cg = hugetlb_cgroup_from_page(page); | |
232 | if (unlikely(!h_cg)) | |
233 | return; | |
234 | set_hugetlb_cgroup(page, NULL); | |
235 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
236 | return; | |
237 | } | |
238 | ||
239 | void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, | |
240 | struct hugetlb_cgroup *h_cg) | |
241 | { | |
242 | unsigned long csize = nr_pages * PAGE_SIZE; | |
243 | ||
244 | if (hugetlb_cgroup_disabled() || !h_cg) | |
245 | return; | |
246 | ||
247 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
248 | return; | |
249 | ||
250 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
251 | return; | |
252 | } | |
253 | ||
abb8206c AK |
254 | static ssize_t hugetlb_cgroup_read(struct cgroup *cgroup, struct cftype *cft, |
255 | struct file *file, char __user *buf, | |
256 | size_t nbytes, loff_t *ppos) | |
257 | { | |
258 | u64 val; | |
259 | char str[64]; | |
260 | int idx, name, len; | |
261 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
262 | ||
263 | idx = MEMFILE_IDX(cft->private); | |
264 | name = MEMFILE_ATTR(cft->private); | |
265 | ||
266 | val = res_counter_read_u64(&h_cg->hugepage[idx], name); | |
267 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
268 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
269 | } | |
270 | ||
271 | static int hugetlb_cgroup_write(struct cgroup *cgroup, struct cftype *cft, | |
272 | const char *buffer) | |
273 | { | |
274 | int idx, name, ret; | |
275 | unsigned long long val; | |
276 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
277 | ||
278 | idx = MEMFILE_IDX(cft->private); | |
279 | name = MEMFILE_ATTR(cft->private); | |
280 | ||
281 | switch (name) { | |
282 | case RES_LIMIT: | |
283 | if (hugetlb_cgroup_is_root(h_cg)) { | |
284 | /* Can't set limit on root */ | |
285 | ret = -EINVAL; | |
286 | break; | |
287 | } | |
288 | /* This function does all necessary parse...reuse it */ | |
289 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
290 | if (ret) | |
291 | break; | |
292 | ret = res_counter_set_limit(&h_cg->hugepage[idx], val); | |
293 | break; | |
294 | default: | |
295 | ret = -EINVAL; | |
296 | break; | |
297 | } | |
298 | return ret; | |
299 | } | |
300 | ||
301 | static int hugetlb_cgroup_reset(struct cgroup *cgroup, unsigned int event) | |
302 | { | |
303 | int idx, name, ret = 0; | |
304 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
305 | ||
306 | idx = MEMFILE_IDX(event); | |
307 | name = MEMFILE_ATTR(event); | |
308 | ||
309 | switch (name) { | |
310 | case RES_MAX_USAGE: | |
311 | res_counter_reset_max(&h_cg->hugepage[idx]); | |
312 | break; | |
313 | case RES_FAILCNT: | |
314 | res_counter_reset_failcnt(&h_cg->hugepage[idx]); | |
315 | break; | |
316 | default: | |
317 | ret = -EINVAL; | |
318 | break; | |
319 | } | |
320 | return ret; | |
321 | } | |
322 | ||
323 | static char *mem_fmt(char *buf, int size, unsigned long hsize) | |
324 | { | |
325 | if (hsize >= (1UL << 30)) | |
326 | snprintf(buf, size, "%luGB", hsize >> 30); | |
327 | else if (hsize >= (1UL << 20)) | |
328 | snprintf(buf, size, "%luMB", hsize >> 20); | |
329 | else | |
330 | snprintf(buf, size, "%luKB", hsize >> 10); | |
331 | return buf; | |
332 | } | |
333 | ||
7179e7bf | 334 | static void __init __hugetlb_cgroup_file_init(int idx) |
abb8206c AK |
335 | { |
336 | char buf[32]; | |
337 | struct cftype *cft; | |
338 | struct hstate *h = &hstates[idx]; | |
339 | ||
340 | /* format the size */ | |
341 | mem_fmt(buf, 32, huge_page_size(h)); | |
342 | ||
343 | /* Add the limit file */ | |
344 | cft = &h->cgroup_files[0]; | |
345 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf); | |
346 | cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT); | |
347 | cft->read = hugetlb_cgroup_read; | |
348 | cft->write_string = hugetlb_cgroup_write; | |
349 | ||
350 | /* Add the usage file */ | |
351 | cft = &h->cgroup_files[1]; | |
352 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf); | |
353 | cft->private = MEMFILE_PRIVATE(idx, RES_USAGE); | |
354 | cft->read = hugetlb_cgroup_read; | |
355 | ||
356 | /* Add the MAX usage file */ | |
357 | cft = &h->cgroup_files[2]; | |
358 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf); | |
359 | cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE); | |
360 | cft->trigger = hugetlb_cgroup_reset; | |
361 | cft->read = hugetlb_cgroup_read; | |
362 | ||
363 | /* Add the failcntfile */ | |
364 | cft = &h->cgroup_files[3]; | |
365 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf); | |
366 | cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT); | |
367 | cft->trigger = hugetlb_cgroup_reset; | |
368 | cft->read = hugetlb_cgroup_read; | |
369 | ||
370 | /* NULL terminate the last cft */ | |
371 | cft = &h->cgroup_files[4]; | |
372 | memset(cft, 0, sizeof(*cft)); | |
373 | ||
374 | WARN_ON(cgroup_add_cftypes(&hugetlb_subsys, h->cgroup_files)); | |
375 | ||
7179e7bf JW |
376 | return; |
377 | } | |
378 | ||
379 | void __init hugetlb_cgroup_file_init(void) | |
380 | { | |
381 | struct hstate *h; | |
382 | ||
383 | for_each_hstate(h) { | |
384 | /* | |
385 | * Add cgroup control files only if the huge page consists | |
386 | * of more than two normal pages. This is because we use | |
387 | * page[2].lru.next for storing cgroup details. | |
388 | */ | |
389 | if (huge_page_order(h) >= HUGETLB_CGROUP_MIN_ORDER) | |
390 | __hugetlb_cgroup_file_init(hstate_index(h)); | |
391 | } | |
abb8206c AK |
392 | } |
393 | ||
75754681 AK |
394 | /* |
395 | * hugetlb_lock will make sure a parallel cgroup rmdir won't happen | |
396 | * when we migrate hugepages | |
397 | */ | |
8e6ac7fa AK |
398 | void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage) |
399 | { | |
400 | struct hugetlb_cgroup *h_cg; | |
94ae8ba7 | 401 | struct hstate *h = page_hstate(oldhpage); |
8e6ac7fa AK |
402 | |
403 | if (hugetlb_cgroup_disabled()) | |
404 | return; | |
405 | ||
406 | VM_BUG_ON(!PageHuge(oldhpage)); | |
407 | spin_lock(&hugetlb_lock); | |
408 | h_cg = hugetlb_cgroup_from_page(oldhpage); | |
409 | set_hugetlb_cgroup(oldhpage, NULL); | |
8e6ac7fa AK |
410 | |
411 | /* move the h_cg details to new cgroup */ | |
412 | set_hugetlb_cgroup(newhpage, h_cg); | |
94ae8ba7 | 413 | list_move(&newhpage->lru, &h->hugepage_activelist); |
8e6ac7fa | 414 | spin_unlock(&hugetlb_lock); |
8e6ac7fa AK |
415 | return; |
416 | } | |
417 | ||
2bc64a20 AK |
418 | struct cgroup_subsys hugetlb_subsys = { |
419 | .name = "hugetlb", | |
92fb9748 TH |
420 | .css_alloc = hugetlb_cgroup_css_alloc, |
421 | .css_offline = hugetlb_cgroup_css_offline, | |
422 | .css_free = hugetlb_cgroup_css_free, | |
423 | .subsys_id = hugetlb_subsys_id, | |
2bc64a20 | 424 | }; |