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 | { | |
45 | return hugetlb_cgroup_from_css(cgroup_subsys_state(cgroup, | |
46 | hugetlb_subsys_id)); | |
47 | } | |
48 | ||
49 | static inline | |
50 | struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task) | |
51 | { | |
52 | return hugetlb_cgroup_from_css(task_subsys_state(task, | |
53 | hugetlb_subsys_id)); | |
54 | } | |
55 | ||
56 | static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg) | |
57 | { | |
58 | return (h_cg == root_h_cgroup); | |
59 | } | |
60 | ||
61 | static inline struct hugetlb_cgroup *parent_hugetlb_cgroup(struct cgroup *cg) | |
62 | { | |
63 | if (!cg->parent) | |
64 | return NULL; | |
65 | return hugetlb_cgroup_from_cgroup(cg->parent); | |
66 | } | |
67 | ||
68 | static inline bool hugetlb_cgroup_have_usage(struct cgroup *cg) | |
69 | { | |
70 | int idx; | |
71 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cg); | |
72 | ||
73 | for (idx = 0; idx < hugetlb_max_hstate; idx++) { | |
74 | if ((res_counter_read_u64(&h_cg->hugepage[idx], RES_USAGE)) > 0) | |
75 | return true; | |
76 | } | |
77 | return false; | |
78 | } | |
79 | ||
92fb9748 | 80 | static struct cgroup_subsys_state *hugetlb_cgroup_css_alloc(struct cgroup *cgroup) |
2bc64a20 AK |
81 | { |
82 | int idx; | |
83 | struct cgroup *parent_cgroup; | |
84 | struct hugetlb_cgroup *h_cgroup, *parent_h_cgroup; | |
85 | ||
86 | h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL); | |
87 | if (!h_cgroup) | |
88 | return ERR_PTR(-ENOMEM); | |
89 | ||
90 | parent_cgroup = cgroup->parent; | |
91 | if (parent_cgroup) { | |
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]); | |
96 | } else { | |
97 | root_h_cgroup = h_cgroup; | |
98 | for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) | |
99 | res_counter_init(&h_cgroup->hugepage[idx], NULL); | |
100 | } | |
101 | return &h_cgroup->css; | |
102 | } | |
103 | ||
92fb9748 | 104 | static void hugetlb_cgroup_css_free(struct cgroup *cgroup) |
2bc64a20 AK |
105 | { |
106 | struct hugetlb_cgroup *h_cgroup; | |
107 | ||
108 | h_cgroup = hugetlb_cgroup_from_cgroup(cgroup); | |
109 | kfree(h_cgroup); | |
110 | } | |
111 | ||
da1def55 AK |
112 | |
113 | /* | |
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 | |
118 | * cannot fail. | |
119 | */ | |
120 | static void hugetlb_cgroup_move_parent(int idx, struct cgroup *cgroup, | |
121 | struct page *page) | |
122 | { | |
123 | int csize; | |
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); | |
129 | ||
130 | page_hcg = hugetlb_cgroup_from_page(page); | |
131 | /* | |
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. | |
135 | */ | |
136 | if (!page_hcg || page_hcg != h_cg) | |
137 | goto out; | |
138 | ||
139 | csize = PAGE_SIZE << compound_order(page); | |
140 | if (!parent) { | |
141 | parent = root_h_cgroup; | |
142 | /* root has no limit */ | |
143 | res_counter_charge_nofail(&parent->hugepage[idx], | |
144 | csize, &fail_res); | |
145 | } | |
146 | counter = &h_cg->hugepage[idx]; | |
147 | res_counter_uncharge_until(counter, counter->parent, csize); | |
148 | ||
149 | set_hugetlb_cgroup(page, parent); | |
150 | out: | |
151 | return; | |
152 | } | |
153 | ||
154 | /* | |
155 | * Force the hugetlb cgroup to empty the hugetlb resources by moving them to | |
156 | * the parent cgroup. | |
157 | */ | |
92fb9748 | 158 | static void hugetlb_cgroup_css_offline(struct cgroup *cgroup) |
2bc64a20 | 159 | { |
da1def55 AK |
160 | struct hstate *h; |
161 | struct page *page; | |
9d093cb1 | 162 | int idx = 0; |
da1def55 AK |
163 | |
164 | do { | |
da1def55 AK |
165 | for_each_hstate(h) { |
166 | spin_lock(&hugetlb_lock); | |
167 | list_for_each_entry(page, &h->hugepage_activelist, lru) | |
168 | hugetlb_cgroup_move_parent(idx, cgroup, page); | |
169 | ||
170 | spin_unlock(&hugetlb_lock); | |
171 | idx++; | |
172 | } | |
173 | cond_resched(); | |
174 | } while (hugetlb_cgroup_have_usage(cgroup)); | |
2bc64a20 AK |
175 | } |
176 | ||
6d76dcf4 AK |
177 | int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages, |
178 | struct hugetlb_cgroup **ptr) | |
179 | { | |
180 | int ret = 0; | |
181 | struct res_counter *fail_res; | |
182 | struct hugetlb_cgroup *h_cg = NULL; | |
183 | unsigned long csize = nr_pages * PAGE_SIZE; | |
184 | ||
185 | if (hugetlb_cgroup_disabled()) | |
186 | goto done; | |
187 | /* | |
188 | * We don't charge any cgroup if the compound page have less | |
189 | * than 3 pages. | |
190 | */ | |
191 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
192 | goto done; | |
193 | again: | |
194 | rcu_read_lock(); | |
195 | h_cg = hugetlb_cgroup_from_task(current); | |
196 | if (!css_tryget(&h_cg->css)) { | |
197 | rcu_read_unlock(); | |
198 | goto again; | |
199 | } | |
200 | rcu_read_unlock(); | |
201 | ||
202 | ret = res_counter_charge(&h_cg->hugepage[idx], csize, &fail_res); | |
203 | css_put(&h_cg->css); | |
204 | done: | |
205 | *ptr = h_cg; | |
206 | return ret; | |
207 | } | |
208 | ||
94ae8ba7 | 209 | /* Should be called with hugetlb_lock held */ |
6d76dcf4 AK |
210 | void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages, |
211 | struct hugetlb_cgroup *h_cg, | |
212 | struct page *page) | |
213 | { | |
214 | if (hugetlb_cgroup_disabled() || !h_cg) | |
215 | return; | |
216 | ||
6d76dcf4 | 217 | set_hugetlb_cgroup(page, h_cg); |
6d76dcf4 AK |
218 | return; |
219 | } | |
220 | ||
221 | /* | |
222 | * Should be called with hugetlb_lock held | |
223 | */ | |
224 | void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages, | |
225 | struct page *page) | |
226 | { | |
227 | struct hugetlb_cgroup *h_cg; | |
228 | unsigned long csize = nr_pages * PAGE_SIZE; | |
229 | ||
230 | if (hugetlb_cgroup_disabled()) | |
231 | return; | |
232 | VM_BUG_ON(!spin_is_locked(&hugetlb_lock)); | |
233 | h_cg = hugetlb_cgroup_from_page(page); | |
234 | if (unlikely(!h_cg)) | |
235 | return; | |
236 | set_hugetlb_cgroup(page, NULL); | |
237 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
238 | return; | |
239 | } | |
240 | ||
241 | void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages, | |
242 | struct hugetlb_cgroup *h_cg) | |
243 | { | |
244 | unsigned long csize = nr_pages * PAGE_SIZE; | |
245 | ||
246 | if (hugetlb_cgroup_disabled() || !h_cg) | |
247 | return; | |
248 | ||
249 | if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER) | |
250 | return; | |
251 | ||
252 | res_counter_uncharge(&h_cg->hugepage[idx], csize); | |
253 | return; | |
254 | } | |
255 | ||
abb8206c AK |
256 | static ssize_t hugetlb_cgroup_read(struct cgroup *cgroup, struct cftype *cft, |
257 | struct file *file, char __user *buf, | |
258 | size_t nbytes, loff_t *ppos) | |
259 | { | |
260 | u64 val; | |
261 | char str[64]; | |
262 | int idx, name, len; | |
263 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
264 | ||
265 | idx = MEMFILE_IDX(cft->private); | |
266 | name = MEMFILE_ATTR(cft->private); | |
267 | ||
268 | val = res_counter_read_u64(&h_cg->hugepage[idx], name); | |
269 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
270 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
271 | } | |
272 | ||
273 | static int hugetlb_cgroup_write(struct cgroup *cgroup, struct cftype *cft, | |
274 | const char *buffer) | |
275 | { | |
276 | int idx, name, ret; | |
277 | unsigned long long val; | |
278 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
279 | ||
280 | idx = MEMFILE_IDX(cft->private); | |
281 | name = MEMFILE_ATTR(cft->private); | |
282 | ||
283 | switch (name) { | |
284 | case RES_LIMIT: | |
285 | if (hugetlb_cgroup_is_root(h_cg)) { | |
286 | /* Can't set limit on root */ | |
287 | ret = -EINVAL; | |
288 | break; | |
289 | } | |
290 | /* This function does all necessary parse...reuse it */ | |
291 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
292 | if (ret) | |
293 | break; | |
294 | ret = res_counter_set_limit(&h_cg->hugepage[idx], val); | |
295 | break; | |
296 | default: | |
297 | ret = -EINVAL; | |
298 | break; | |
299 | } | |
300 | return ret; | |
301 | } | |
302 | ||
303 | static int hugetlb_cgroup_reset(struct cgroup *cgroup, unsigned int event) | |
304 | { | |
305 | int idx, name, ret = 0; | |
306 | struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup); | |
307 | ||
308 | idx = MEMFILE_IDX(event); | |
309 | name = MEMFILE_ATTR(event); | |
310 | ||
311 | switch (name) { | |
312 | case RES_MAX_USAGE: | |
313 | res_counter_reset_max(&h_cg->hugepage[idx]); | |
314 | break; | |
315 | case RES_FAILCNT: | |
316 | res_counter_reset_failcnt(&h_cg->hugepage[idx]); | |
317 | break; | |
318 | default: | |
319 | ret = -EINVAL; | |
320 | break; | |
321 | } | |
322 | return ret; | |
323 | } | |
324 | ||
325 | static char *mem_fmt(char *buf, int size, unsigned long hsize) | |
326 | { | |
327 | if (hsize >= (1UL << 30)) | |
328 | snprintf(buf, size, "%luGB", hsize >> 30); | |
329 | else if (hsize >= (1UL << 20)) | |
330 | snprintf(buf, size, "%luMB", hsize >> 20); | |
331 | else | |
332 | snprintf(buf, size, "%luKB", hsize >> 10); | |
333 | return buf; | |
334 | } | |
335 | ||
7179e7bf | 336 | static void __init __hugetlb_cgroup_file_init(int idx) |
abb8206c AK |
337 | { |
338 | char buf[32]; | |
339 | struct cftype *cft; | |
340 | struct hstate *h = &hstates[idx]; | |
341 | ||
342 | /* format the size */ | |
343 | mem_fmt(buf, 32, huge_page_size(h)); | |
344 | ||
345 | /* Add the limit file */ | |
346 | cft = &h->cgroup_files[0]; | |
347 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf); | |
348 | cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT); | |
349 | cft->read = hugetlb_cgroup_read; | |
350 | cft->write_string = hugetlb_cgroup_write; | |
351 | ||
352 | /* Add the usage file */ | |
353 | cft = &h->cgroup_files[1]; | |
354 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf); | |
355 | cft->private = MEMFILE_PRIVATE(idx, RES_USAGE); | |
356 | cft->read = hugetlb_cgroup_read; | |
357 | ||
358 | /* Add the MAX usage file */ | |
359 | cft = &h->cgroup_files[2]; | |
360 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf); | |
361 | cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE); | |
362 | cft->trigger = hugetlb_cgroup_reset; | |
363 | cft->read = hugetlb_cgroup_read; | |
364 | ||
365 | /* Add the failcntfile */ | |
366 | cft = &h->cgroup_files[3]; | |
367 | snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf); | |
368 | cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT); | |
369 | cft->trigger = hugetlb_cgroup_reset; | |
370 | cft->read = hugetlb_cgroup_read; | |
371 | ||
372 | /* NULL terminate the last cft */ | |
373 | cft = &h->cgroup_files[4]; | |
374 | memset(cft, 0, sizeof(*cft)); | |
375 | ||
376 | WARN_ON(cgroup_add_cftypes(&hugetlb_subsys, h->cgroup_files)); | |
377 | ||
7179e7bf JW |
378 | return; |
379 | } | |
380 | ||
381 | void __init hugetlb_cgroup_file_init(void) | |
382 | { | |
383 | struct hstate *h; | |
384 | ||
385 | for_each_hstate(h) { | |
386 | /* | |
387 | * Add cgroup control files only if the huge page consists | |
388 | * of more than two normal pages. This is because we use | |
389 | * page[2].lru.next for storing cgroup details. | |
390 | */ | |
391 | if (huge_page_order(h) >= HUGETLB_CGROUP_MIN_ORDER) | |
392 | __hugetlb_cgroup_file_init(hstate_index(h)); | |
393 | } | |
abb8206c AK |
394 | } |
395 | ||
75754681 AK |
396 | /* |
397 | * hugetlb_lock will make sure a parallel cgroup rmdir won't happen | |
398 | * when we migrate hugepages | |
399 | */ | |
8e6ac7fa AK |
400 | void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage) |
401 | { | |
402 | struct hugetlb_cgroup *h_cg; | |
94ae8ba7 | 403 | struct hstate *h = page_hstate(oldhpage); |
8e6ac7fa AK |
404 | |
405 | if (hugetlb_cgroup_disabled()) | |
406 | return; | |
407 | ||
408 | VM_BUG_ON(!PageHuge(oldhpage)); | |
409 | spin_lock(&hugetlb_lock); | |
410 | h_cg = hugetlb_cgroup_from_page(oldhpage); | |
411 | set_hugetlb_cgroup(oldhpage, NULL); | |
8e6ac7fa AK |
412 | |
413 | /* move the h_cg details to new cgroup */ | |
414 | set_hugetlb_cgroup(newhpage, h_cg); | |
94ae8ba7 | 415 | list_move(&newhpage->lru, &h->hugepage_activelist); |
8e6ac7fa | 416 | spin_unlock(&hugetlb_lock); |
8e6ac7fa AK |
417 | return; |
418 | } | |
419 | ||
2bc64a20 AK |
420 | struct cgroup_subsys hugetlb_subsys = { |
421 | .name = "hugetlb", | |
92fb9748 TH |
422 | .css_alloc = hugetlb_cgroup_css_alloc, |
423 | .css_offline = hugetlb_cgroup_css_offline, | |
424 | .css_free = hugetlb_cgroup_css_free, | |
425 | .subsys_id = hugetlb_subsys_id, | |
2bc64a20 | 426 | }; |