• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *
3  * Copyright IBM Corporation, 2012
4  * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
5  *
6  * Cgroup v2
7  * Copyright (C) 2019 Red Hat, Inc.
8  * Author: Giuseppe Scrivano <gscrivan@redhat.com>
9  *
10  * This program is free software; you can redistribute it and/or modify it
11  * under the terms of version 2.1 of the GNU Lesser General Public License
12  * as published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it would be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
17  *
18  */
19 
20 #include <linux/cgroup.h>
21 #include <linux/page_counter.h>
22 #include <linux/slab.h>
23 #include <linux/hugetlb.h>
24 #include <linux/hugetlb_cgroup.h>
25 
26 #define MEMFILE_PRIVATE(x, val)	(((x) << 16) | (val))
27 #define MEMFILE_IDX(val)	(((val) >> 16) & 0xffff)
28 #define MEMFILE_ATTR(val)	((val) & 0xffff)
29 
30 #define hugetlb_cgroup_from_counter(counter, idx)                   \
31 	container_of(counter, struct hugetlb_cgroup, hugepage[idx])
32 
33 static struct hugetlb_cgroup *root_h_cgroup __read_mostly;
34 
35 static inline struct page_counter *
__hugetlb_cgroup_counter_from_cgroup(struct hugetlb_cgroup * h_cg,int idx,bool rsvd)36 __hugetlb_cgroup_counter_from_cgroup(struct hugetlb_cgroup *h_cg, int idx,
37 				     bool rsvd)
38 {
39 	if (rsvd)
40 		return &h_cg->rsvd_hugepage[idx];
41 	return &h_cg->hugepage[idx];
42 }
43 
44 static inline struct page_counter *
hugetlb_cgroup_counter_from_cgroup(struct hugetlb_cgroup * h_cg,int idx)45 hugetlb_cgroup_counter_from_cgroup(struct hugetlb_cgroup *h_cg, int idx)
46 {
47 	return __hugetlb_cgroup_counter_from_cgroup(h_cg, idx, false);
48 }
49 
50 static inline struct page_counter *
hugetlb_cgroup_counter_from_cgroup_rsvd(struct hugetlb_cgroup * h_cg,int idx)51 hugetlb_cgroup_counter_from_cgroup_rsvd(struct hugetlb_cgroup *h_cg, int idx)
52 {
53 	return __hugetlb_cgroup_counter_from_cgroup(h_cg, idx, true);
54 }
55 
56 static inline
hugetlb_cgroup_from_css(struct cgroup_subsys_state * s)57 struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s)
58 {
59 	return s ? container_of(s, struct hugetlb_cgroup, css) : NULL;
60 }
61 
62 static inline
hugetlb_cgroup_from_task(struct task_struct * task)63 struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task)
64 {
65 	return hugetlb_cgroup_from_css(task_css(task, hugetlb_cgrp_id));
66 }
67 
hugetlb_cgroup_is_root(struct hugetlb_cgroup * h_cg)68 static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg)
69 {
70 	return (h_cg == root_h_cgroup);
71 }
72 
73 static inline struct hugetlb_cgroup *
parent_hugetlb_cgroup(struct hugetlb_cgroup * h_cg)74 parent_hugetlb_cgroup(struct hugetlb_cgroup *h_cg)
75 {
76 	return hugetlb_cgroup_from_css(h_cg->css.parent);
77 }
78 
hugetlb_cgroup_have_usage(struct hugetlb_cgroup * h_cg)79 static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg)
80 {
81 	int idx;
82 
83 	for (idx = 0; idx < hugetlb_max_hstate; idx++) {
84 		if (page_counter_read(
85 				hugetlb_cgroup_counter_from_cgroup(h_cg, idx)))
86 			return true;
87 	}
88 	return false;
89 }
90 
hugetlb_cgroup_init(struct hugetlb_cgroup * h_cgroup,struct hugetlb_cgroup * parent_h_cgroup)91 static void hugetlb_cgroup_init(struct hugetlb_cgroup *h_cgroup,
92 				struct hugetlb_cgroup *parent_h_cgroup)
93 {
94 	int idx;
95 
96 	for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) {
97 		struct page_counter *fault_parent = NULL;
98 		struct page_counter *rsvd_parent = NULL;
99 		unsigned long limit;
100 		int ret;
101 
102 		if (parent_h_cgroup) {
103 			fault_parent = hugetlb_cgroup_counter_from_cgroup(
104 				parent_h_cgroup, idx);
105 			rsvd_parent = hugetlb_cgroup_counter_from_cgroup_rsvd(
106 				parent_h_cgroup, idx);
107 		}
108 		page_counter_init(hugetlb_cgroup_counter_from_cgroup(h_cgroup,
109 								     idx),
110 				  fault_parent);
111 		page_counter_init(
112 			hugetlb_cgroup_counter_from_cgroup_rsvd(h_cgroup, idx),
113 			rsvd_parent);
114 
115 		limit = round_down(PAGE_COUNTER_MAX,
116 				   pages_per_huge_page(&hstates[idx]));
117 
118 		ret = page_counter_set_max(
119 			hugetlb_cgroup_counter_from_cgroup(h_cgroup, idx),
120 			limit);
121 		VM_BUG_ON(ret);
122 		ret = page_counter_set_max(
123 			hugetlb_cgroup_counter_from_cgroup_rsvd(h_cgroup, idx),
124 			limit);
125 		VM_BUG_ON(ret);
126 	}
127 }
128 
129 static struct cgroup_subsys_state *
hugetlb_cgroup_css_alloc(struct cgroup_subsys_state * parent_css)130 hugetlb_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
131 {
132 	struct hugetlb_cgroup *parent_h_cgroup = hugetlb_cgroup_from_css(parent_css);
133 	struct hugetlb_cgroup *h_cgroup;
134 
135 	h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL);
136 	if (!h_cgroup)
137 		return ERR_PTR(-ENOMEM);
138 
139 	if (!parent_h_cgroup)
140 		root_h_cgroup = h_cgroup;
141 
142 	hugetlb_cgroup_init(h_cgroup, parent_h_cgroup);
143 	return &h_cgroup->css;
144 }
145 
hugetlb_cgroup_css_free(struct cgroup_subsys_state * css)146 static void hugetlb_cgroup_css_free(struct cgroup_subsys_state *css)
147 {
148 	struct hugetlb_cgroup *h_cgroup;
149 
150 	h_cgroup = hugetlb_cgroup_from_css(css);
151 	kfree(h_cgroup);
152 }
153 
154 /*
155  * Should be called with hugetlb_lock held.
156  * Since we are holding hugetlb_lock, pages cannot get moved from
157  * active list or uncharged from the cgroup, So no need to get
158  * page reference and test for page active here. This function
159  * cannot fail.
160  */
hugetlb_cgroup_move_parent(int idx,struct hugetlb_cgroup * h_cg,struct page * page)161 static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg,
162 				       struct page *page)
163 {
164 	unsigned int nr_pages;
165 	struct page_counter *counter;
166 	struct hugetlb_cgroup *page_hcg;
167 	struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(h_cg);
168 
169 	page_hcg = hugetlb_cgroup_from_page(page);
170 	/*
171 	 * We can have pages in active list without any cgroup
172 	 * ie, hugepage with less than 3 pages. We can safely
173 	 * ignore those pages.
174 	 */
175 	if (!page_hcg || page_hcg != h_cg)
176 		goto out;
177 
178 	nr_pages = compound_nr(page);
179 	if (!parent) {
180 		parent = root_h_cgroup;
181 		/* root has no limit */
182 		page_counter_charge(&parent->hugepage[idx], nr_pages);
183 	}
184 	counter = &h_cg->hugepage[idx];
185 	/* Take the pages off the local counter */
186 	page_counter_cancel(counter, nr_pages);
187 
188 	set_hugetlb_cgroup(page, parent);
189 out:
190 	return;
191 }
192 
193 /*
194  * Force the hugetlb cgroup to empty the hugetlb resources by moving them to
195  * the parent cgroup.
196  */
hugetlb_cgroup_css_offline(struct cgroup_subsys_state * css)197 static void hugetlb_cgroup_css_offline(struct cgroup_subsys_state *css)
198 {
199 	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
200 	struct hstate *h;
201 	struct page *page;
202 	int idx;
203 
204 	do {
205 		idx = 0;
206 		for_each_hstate(h) {
207 			spin_lock_irq(&hugetlb_lock);
208 			list_for_each_entry(page, &h->hugepage_activelist, lru)
209 				hugetlb_cgroup_move_parent(idx, h_cg, page);
210 
211 			spin_unlock_irq(&hugetlb_lock);
212 			idx++;
213 		}
214 		cond_resched();
215 	} while (hugetlb_cgroup_have_usage(h_cg));
216 }
217 
hugetlb_event(struct hugetlb_cgroup * hugetlb,int idx,enum hugetlb_memory_event event)218 static inline void hugetlb_event(struct hugetlb_cgroup *hugetlb, int idx,
219 				 enum hugetlb_memory_event event)
220 {
221 	atomic_long_inc(&hugetlb->events_local[idx][event]);
222 	cgroup_file_notify(&hugetlb->events_local_file[idx]);
223 
224 	do {
225 		atomic_long_inc(&hugetlb->events[idx][event]);
226 		cgroup_file_notify(&hugetlb->events_file[idx]);
227 	} while ((hugetlb = parent_hugetlb_cgroup(hugetlb)) &&
228 		 !hugetlb_cgroup_is_root(hugetlb));
229 }
230 
__hugetlb_cgroup_charge_cgroup(int idx,unsigned long nr_pages,struct hugetlb_cgroup ** ptr,bool rsvd)231 static int __hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages,
232 					  struct hugetlb_cgroup **ptr,
233 					  bool rsvd)
234 {
235 	int ret = 0;
236 	struct page_counter *counter;
237 	struct hugetlb_cgroup *h_cg = NULL;
238 
239 	if (hugetlb_cgroup_disabled())
240 		goto done;
241 	/*
242 	 * We don't charge any cgroup if the compound page have less
243 	 * than 3 pages.
244 	 */
245 	if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
246 		goto done;
247 again:
248 	rcu_read_lock();
249 	h_cg = hugetlb_cgroup_from_task(current);
250 	if (!css_tryget(&h_cg->css)) {
251 		rcu_read_unlock();
252 		goto again;
253 	}
254 	rcu_read_unlock();
255 
256 	if (!page_counter_try_charge(
257 		    __hugetlb_cgroup_counter_from_cgroup(h_cg, idx, rsvd),
258 		    nr_pages, &counter)) {
259 		ret = -ENOMEM;
260 		hugetlb_event(h_cg, idx, HUGETLB_MAX);
261 		css_put(&h_cg->css);
262 		goto done;
263 	}
264 	/* Reservations take a reference to the css because they do not get
265 	 * reparented.
266 	 */
267 	if (!rsvd)
268 		css_put(&h_cg->css);
269 done:
270 	*ptr = h_cg;
271 	return ret;
272 }
273 
hugetlb_cgroup_charge_cgroup(int idx,unsigned long nr_pages,struct hugetlb_cgroup ** ptr)274 int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages,
275 				 struct hugetlb_cgroup **ptr)
276 {
277 	return __hugetlb_cgroup_charge_cgroup(idx, nr_pages, ptr, false);
278 }
279 
hugetlb_cgroup_charge_cgroup_rsvd(int idx,unsigned long nr_pages,struct hugetlb_cgroup ** ptr)280 int hugetlb_cgroup_charge_cgroup_rsvd(int idx, unsigned long nr_pages,
281 				      struct hugetlb_cgroup **ptr)
282 {
283 	return __hugetlb_cgroup_charge_cgroup(idx, nr_pages, ptr, true);
284 }
285 
286 /* Should be called with hugetlb_lock held */
__hugetlb_cgroup_commit_charge(int idx,unsigned long nr_pages,struct hugetlb_cgroup * h_cg,struct page * page,bool rsvd)287 static void __hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages,
288 					   struct hugetlb_cgroup *h_cg,
289 					   struct page *page, bool rsvd)
290 {
291 	if (hugetlb_cgroup_disabled() || !h_cg)
292 		return;
293 
294 	__set_hugetlb_cgroup(page, h_cg, rsvd);
295 	return;
296 }
297 
hugetlb_cgroup_commit_charge(int idx,unsigned long nr_pages,struct hugetlb_cgroup * h_cg,struct page * page)298 void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages,
299 				  struct hugetlb_cgroup *h_cg,
300 				  struct page *page)
301 {
302 	__hugetlb_cgroup_commit_charge(idx, nr_pages, h_cg, page, false);
303 }
304 
hugetlb_cgroup_commit_charge_rsvd(int idx,unsigned long nr_pages,struct hugetlb_cgroup * h_cg,struct page * page)305 void hugetlb_cgroup_commit_charge_rsvd(int idx, unsigned long nr_pages,
306 				       struct hugetlb_cgroup *h_cg,
307 				       struct page *page)
308 {
309 	__hugetlb_cgroup_commit_charge(idx, nr_pages, h_cg, page, true);
310 }
311 
312 /*
313  * Should be called with hugetlb_lock held
314  */
__hugetlb_cgroup_uncharge_page(int idx,unsigned long nr_pages,struct page * page,bool rsvd)315 static void __hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages,
316 					   struct page *page, bool rsvd)
317 {
318 	struct hugetlb_cgroup *h_cg;
319 
320 	if (hugetlb_cgroup_disabled())
321 		return;
322 	lockdep_assert_held(&hugetlb_lock);
323 	h_cg = __hugetlb_cgroup_from_page(page, rsvd);
324 	if (unlikely(!h_cg))
325 		return;
326 	__set_hugetlb_cgroup(page, NULL, rsvd);
327 
328 	page_counter_uncharge(__hugetlb_cgroup_counter_from_cgroup(h_cg, idx,
329 								   rsvd),
330 			      nr_pages);
331 
332 	if (rsvd)
333 		css_put(&h_cg->css);
334 
335 	return;
336 }
337 
hugetlb_cgroup_uncharge_page(int idx,unsigned long nr_pages,struct page * page)338 void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages,
339 				  struct page *page)
340 {
341 	__hugetlb_cgroup_uncharge_page(idx, nr_pages, page, false);
342 }
343 
hugetlb_cgroup_uncharge_page_rsvd(int idx,unsigned long nr_pages,struct page * page)344 void hugetlb_cgroup_uncharge_page_rsvd(int idx, unsigned long nr_pages,
345 				       struct page *page)
346 {
347 	__hugetlb_cgroup_uncharge_page(idx, nr_pages, page, true);
348 }
349 
__hugetlb_cgroup_uncharge_cgroup(int idx,unsigned long nr_pages,struct hugetlb_cgroup * h_cg,bool rsvd)350 static void __hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages,
351 					     struct hugetlb_cgroup *h_cg,
352 					     bool rsvd)
353 {
354 	if (hugetlb_cgroup_disabled() || !h_cg)
355 		return;
356 
357 	if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
358 		return;
359 
360 	page_counter_uncharge(__hugetlb_cgroup_counter_from_cgroup(h_cg, idx,
361 								   rsvd),
362 			      nr_pages);
363 
364 	if (rsvd)
365 		css_put(&h_cg->css);
366 }
367 
hugetlb_cgroup_uncharge_cgroup(int idx,unsigned long nr_pages,struct hugetlb_cgroup * h_cg)368 void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages,
369 				    struct hugetlb_cgroup *h_cg)
370 {
371 	__hugetlb_cgroup_uncharge_cgroup(idx, nr_pages, h_cg, false);
372 }
373 
hugetlb_cgroup_uncharge_cgroup_rsvd(int idx,unsigned long nr_pages,struct hugetlb_cgroup * h_cg)374 void hugetlb_cgroup_uncharge_cgroup_rsvd(int idx, unsigned long nr_pages,
375 					 struct hugetlb_cgroup *h_cg)
376 {
377 	__hugetlb_cgroup_uncharge_cgroup(idx, nr_pages, h_cg, true);
378 }
379 
hugetlb_cgroup_uncharge_counter(struct resv_map * resv,unsigned long start,unsigned long end)380 void hugetlb_cgroup_uncharge_counter(struct resv_map *resv, unsigned long start,
381 				     unsigned long end)
382 {
383 	if (hugetlb_cgroup_disabled() || !resv || !resv->reservation_counter ||
384 	    !resv->css)
385 		return;
386 
387 	page_counter_uncharge(resv->reservation_counter,
388 			      (end - start) * resv->pages_per_hpage);
389 	css_put(resv->css);
390 }
391 
hugetlb_cgroup_uncharge_file_region(struct resv_map * resv,struct file_region * rg,unsigned long nr_pages,bool region_del)392 void hugetlb_cgroup_uncharge_file_region(struct resv_map *resv,
393 					 struct file_region *rg,
394 					 unsigned long nr_pages,
395 					 bool region_del)
396 {
397 	if (hugetlb_cgroup_disabled() || !resv || !rg || !nr_pages)
398 		return;
399 
400 	if (rg->reservation_counter && resv->pages_per_hpage && nr_pages > 0 &&
401 	    !resv->reservation_counter) {
402 		page_counter_uncharge(rg->reservation_counter,
403 				      nr_pages * resv->pages_per_hpage);
404 		/*
405 		 * Only do css_put(rg->css) when we delete the entire region
406 		 * because one file_region must hold exactly one css reference.
407 		 */
408 		if (region_del)
409 			css_put(rg->css);
410 	}
411 }
412 
413 enum {
414 	RES_USAGE,
415 	RES_RSVD_USAGE,
416 	RES_LIMIT,
417 	RES_RSVD_LIMIT,
418 	RES_MAX_USAGE,
419 	RES_RSVD_MAX_USAGE,
420 	RES_FAILCNT,
421 	RES_RSVD_FAILCNT,
422 };
423 
hugetlb_cgroup_read_u64(struct cgroup_subsys_state * css,struct cftype * cft)424 static u64 hugetlb_cgroup_read_u64(struct cgroup_subsys_state *css,
425 				   struct cftype *cft)
426 {
427 	struct page_counter *counter;
428 	struct page_counter *rsvd_counter;
429 	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
430 
431 	counter = &h_cg->hugepage[MEMFILE_IDX(cft->private)];
432 	rsvd_counter = &h_cg->rsvd_hugepage[MEMFILE_IDX(cft->private)];
433 
434 	switch (MEMFILE_ATTR(cft->private)) {
435 	case RES_USAGE:
436 		return (u64)page_counter_read(counter) * PAGE_SIZE;
437 	case RES_RSVD_USAGE:
438 		return (u64)page_counter_read(rsvd_counter) * PAGE_SIZE;
439 	case RES_LIMIT:
440 		return (u64)counter->max * PAGE_SIZE;
441 	case RES_RSVD_LIMIT:
442 		return (u64)rsvd_counter->max * PAGE_SIZE;
443 	case RES_MAX_USAGE:
444 		return (u64)counter->watermark * PAGE_SIZE;
445 	case RES_RSVD_MAX_USAGE:
446 		return (u64)rsvd_counter->watermark * PAGE_SIZE;
447 	case RES_FAILCNT:
448 		return counter->failcnt;
449 	case RES_RSVD_FAILCNT:
450 		return rsvd_counter->failcnt;
451 	default:
452 		BUG();
453 	}
454 }
455 
hugetlb_cgroup_read_u64_max(struct seq_file * seq,void * v)456 static int hugetlb_cgroup_read_u64_max(struct seq_file *seq, void *v)
457 {
458 	int idx;
459 	u64 val;
460 	struct cftype *cft = seq_cft(seq);
461 	unsigned long limit;
462 	struct page_counter *counter;
463 	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(seq_css(seq));
464 
465 	idx = MEMFILE_IDX(cft->private);
466 	counter = &h_cg->hugepage[idx];
467 
468 	limit = round_down(PAGE_COUNTER_MAX,
469 			   pages_per_huge_page(&hstates[idx]));
470 
471 	switch (MEMFILE_ATTR(cft->private)) {
472 	case RES_RSVD_USAGE:
473 		counter = &h_cg->rsvd_hugepage[idx];
474 		fallthrough;
475 	case RES_USAGE:
476 		val = (u64)page_counter_read(counter);
477 		seq_printf(seq, "%llu\n", val * PAGE_SIZE);
478 		break;
479 	case RES_RSVD_LIMIT:
480 		counter = &h_cg->rsvd_hugepage[idx];
481 		fallthrough;
482 	case RES_LIMIT:
483 		val = (u64)counter->max;
484 		if (val == limit)
485 			seq_puts(seq, "max\n");
486 		else
487 			seq_printf(seq, "%llu\n", val * PAGE_SIZE);
488 		break;
489 	default:
490 		BUG();
491 	}
492 
493 	return 0;
494 }
495 
496 static DEFINE_MUTEX(hugetlb_limit_mutex);
497 
hugetlb_cgroup_write(struct kernfs_open_file * of,char * buf,size_t nbytes,loff_t off,const char * max)498 static ssize_t hugetlb_cgroup_write(struct kernfs_open_file *of,
499 				    char *buf, size_t nbytes, loff_t off,
500 				    const char *max)
501 {
502 	int ret, idx;
503 	unsigned long nr_pages;
504 	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
505 	bool rsvd = false;
506 
507 	if (hugetlb_cgroup_is_root(h_cg)) /* Can't set limit on root */
508 		return -EINVAL;
509 
510 	buf = strstrip(buf);
511 	ret = page_counter_memparse(buf, max, &nr_pages);
512 	if (ret)
513 		return ret;
514 
515 	idx = MEMFILE_IDX(of_cft(of)->private);
516 	nr_pages = round_down(nr_pages, pages_per_huge_page(&hstates[idx]));
517 
518 	switch (MEMFILE_ATTR(of_cft(of)->private)) {
519 	case RES_RSVD_LIMIT:
520 		rsvd = true;
521 		fallthrough;
522 	case RES_LIMIT:
523 		mutex_lock(&hugetlb_limit_mutex);
524 		ret = page_counter_set_max(
525 			__hugetlb_cgroup_counter_from_cgroup(h_cg, idx, rsvd),
526 			nr_pages);
527 		mutex_unlock(&hugetlb_limit_mutex);
528 		break;
529 	default:
530 		ret = -EINVAL;
531 		break;
532 	}
533 	return ret ?: nbytes;
534 }
535 
hugetlb_cgroup_write_legacy(struct kernfs_open_file * of,char * buf,size_t nbytes,loff_t off)536 static ssize_t hugetlb_cgroup_write_legacy(struct kernfs_open_file *of,
537 					   char *buf, size_t nbytes, loff_t off)
538 {
539 	return hugetlb_cgroup_write(of, buf, nbytes, off, "-1");
540 }
541 
hugetlb_cgroup_write_dfl(struct kernfs_open_file * of,char * buf,size_t nbytes,loff_t off)542 static ssize_t hugetlb_cgroup_write_dfl(struct kernfs_open_file *of,
543 					char *buf, size_t nbytes, loff_t off)
544 {
545 	return hugetlb_cgroup_write(of, buf, nbytes, off, "max");
546 }
547 
hugetlb_cgroup_reset(struct kernfs_open_file * of,char * buf,size_t nbytes,loff_t off)548 static ssize_t hugetlb_cgroup_reset(struct kernfs_open_file *of,
549 				    char *buf, size_t nbytes, loff_t off)
550 {
551 	int ret = 0;
552 	struct page_counter *counter, *rsvd_counter;
553 	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
554 
555 	counter = &h_cg->hugepage[MEMFILE_IDX(of_cft(of)->private)];
556 	rsvd_counter = &h_cg->rsvd_hugepage[MEMFILE_IDX(of_cft(of)->private)];
557 
558 	switch (MEMFILE_ATTR(of_cft(of)->private)) {
559 	case RES_MAX_USAGE:
560 		page_counter_reset_watermark(counter);
561 		break;
562 	case RES_RSVD_MAX_USAGE:
563 		page_counter_reset_watermark(rsvd_counter);
564 		break;
565 	case RES_FAILCNT:
566 		counter->failcnt = 0;
567 		break;
568 	case RES_RSVD_FAILCNT:
569 		rsvd_counter->failcnt = 0;
570 		break;
571 	default:
572 		ret = -EINVAL;
573 		break;
574 	}
575 	return ret ?: nbytes;
576 }
577 
mem_fmt(char * buf,int size,unsigned long hsize)578 static char *mem_fmt(char *buf, int size, unsigned long hsize)
579 {
580 	if (hsize >= (1UL << 30))
581 		snprintf(buf, size, "%luGB", hsize >> 30);
582 	else if (hsize >= (1UL << 20))
583 		snprintf(buf, size, "%luMB", hsize >> 20);
584 	else
585 		snprintf(buf, size, "%luKB", hsize >> 10);
586 	return buf;
587 }
588 
__hugetlb_events_show(struct seq_file * seq,bool local)589 static int __hugetlb_events_show(struct seq_file *seq, bool local)
590 {
591 	int idx;
592 	long max;
593 	struct cftype *cft = seq_cft(seq);
594 	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(seq_css(seq));
595 
596 	idx = MEMFILE_IDX(cft->private);
597 
598 	if (local)
599 		max = atomic_long_read(&h_cg->events_local[idx][HUGETLB_MAX]);
600 	else
601 		max = atomic_long_read(&h_cg->events[idx][HUGETLB_MAX]);
602 
603 	seq_printf(seq, "max %lu\n", max);
604 
605 	return 0;
606 }
607 
hugetlb_events_show(struct seq_file * seq,void * v)608 static int hugetlb_events_show(struct seq_file *seq, void *v)
609 {
610 	return __hugetlb_events_show(seq, false);
611 }
612 
hugetlb_events_local_show(struct seq_file * seq,void * v)613 static int hugetlb_events_local_show(struct seq_file *seq, void *v)
614 {
615 	return __hugetlb_events_show(seq, true);
616 }
617 
__hugetlb_cgroup_file_dfl_init(int idx)618 static void __init __hugetlb_cgroup_file_dfl_init(int idx)
619 {
620 	char buf[32];
621 	struct cftype *cft;
622 	struct hstate *h = &hstates[idx];
623 
624 	/* format the size */
625 	mem_fmt(buf, sizeof(buf), huge_page_size(h));
626 
627 	/* Add the limit file */
628 	cft = &h->cgroup_files_dfl[0];
629 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max", buf);
630 	cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT);
631 	cft->seq_show = hugetlb_cgroup_read_u64_max;
632 	cft->write = hugetlb_cgroup_write_dfl;
633 	cft->flags = CFTYPE_NOT_ON_ROOT;
634 
635 	/* Add the reservation limit file */
636 	cft = &h->cgroup_files_dfl[1];
637 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.max", buf);
638 	cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_LIMIT);
639 	cft->seq_show = hugetlb_cgroup_read_u64_max;
640 	cft->write = hugetlb_cgroup_write_dfl;
641 	cft->flags = CFTYPE_NOT_ON_ROOT;
642 
643 	/* Add the current usage file */
644 	cft = &h->cgroup_files_dfl[2];
645 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.current", buf);
646 	cft->private = MEMFILE_PRIVATE(idx, RES_USAGE);
647 	cft->seq_show = hugetlb_cgroup_read_u64_max;
648 	cft->flags = CFTYPE_NOT_ON_ROOT;
649 
650 	/* Add the current reservation usage file */
651 	cft = &h->cgroup_files_dfl[3];
652 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.current", buf);
653 	cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_USAGE);
654 	cft->seq_show = hugetlb_cgroup_read_u64_max;
655 	cft->flags = CFTYPE_NOT_ON_ROOT;
656 
657 	/* Add the events file */
658 	cft = &h->cgroup_files_dfl[4];
659 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.events", buf);
660 	cft->private = MEMFILE_PRIVATE(idx, 0);
661 	cft->seq_show = hugetlb_events_show;
662 	cft->file_offset = offsetof(struct hugetlb_cgroup, events_file[idx]);
663 	cft->flags = CFTYPE_NOT_ON_ROOT;
664 
665 	/* Add the events.local file */
666 	cft = &h->cgroup_files_dfl[5];
667 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.events.local", buf);
668 	cft->private = MEMFILE_PRIVATE(idx, 0);
669 	cft->seq_show = hugetlb_events_local_show;
670 	cft->file_offset = offsetof(struct hugetlb_cgroup,
671 				    events_local_file[idx]);
672 	cft->flags = CFTYPE_NOT_ON_ROOT;
673 
674 	/* NULL terminate the last cft */
675 	cft = &h->cgroup_files_dfl[6];
676 	memset(cft, 0, sizeof(*cft));
677 
678 	WARN_ON(cgroup_add_dfl_cftypes(&hugetlb_cgrp_subsys,
679 				       h->cgroup_files_dfl));
680 }
681 
__hugetlb_cgroup_file_legacy_init(int idx)682 static void __init __hugetlb_cgroup_file_legacy_init(int idx)
683 {
684 	char buf[32];
685 	struct cftype *cft;
686 	struct hstate *h = &hstates[idx];
687 
688 	/* format the size */
689 	mem_fmt(buf, sizeof(buf), huge_page_size(h));
690 
691 	/* Add the limit file */
692 	cft = &h->cgroup_files_legacy[0];
693 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf);
694 	cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT);
695 	cft->read_u64 = hugetlb_cgroup_read_u64;
696 	cft->write = hugetlb_cgroup_write_legacy;
697 
698 	/* Add the reservation limit file */
699 	cft = &h->cgroup_files_legacy[1];
700 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.limit_in_bytes", buf);
701 	cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_LIMIT);
702 	cft->read_u64 = hugetlb_cgroup_read_u64;
703 	cft->write = hugetlb_cgroup_write_legacy;
704 
705 	/* Add the usage file */
706 	cft = &h->cgroup_files_legacy[2];
707 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf);
708 	cft->private = MEMFILE_PRIVATE(idx, RES_USAGE);
709 	cft->read_u64 = hugetlb_cgroup_read_u64;
710 
711 	/* Add the reservation usage file */
712 	cft = &h->cgroup_files_legacy[3];
713 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.usage_in_bytes", buf);
714 	cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_USAGE);
715 	cft->read_u64 = hugetlb_cgroup_read_u64;
716 
717 	/* Add the MAX usage file */
718 	cft = &h->cgroup_files_legacy[4];
719 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf);
720 	cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE);
721 	cft->write = hugetlb_cgroup_reset;
722 	cft->read_u64 = hugetlb_cgroup_read_u64;
723 
724 	/* Add the MAX reservation usage file */
725 	cft = &h->cgroup_files_legacy[5];
726 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.max_usage_in_bytes", buf);
727 	cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_MAX_USAGE);
728 	cft->write = hugetlb_cgroup_reset;
729 	cft->read_u64 = hugetlb_cgroup_read_u64;
730 
731 	/* Add the failcntfile */
732 	cft = &h->cgroup_files_legacy[6];
733 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf);
734 	cft->private = MEMFILE_PRIVATE(idx, RES_FAILCNT);
735 	cft->write = hugetlb_cgroup_reset;
736 	cft->read_u64 = hugetlb_cgroup_read_u64;
737 
738 	/* Add the reservation failcntfile */
739 	cft = &h->cgroup_files_legacy[7];
740 	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.rsvd.failcnt", buf);
741 	cft->private = MEMFILE_PRIVATE(idx, RES_RSVD_FAILCNT);
742 	cft->write = hugetlb_cgroup_reset;
743 	cft->read_u64 = hugetlb_cgroup_read_u64;
744 
745 	/* NULL terminate the last cft */
746 	cft = &h->cgroup_files_legacy[8];
747 	memset(cft, 0, sizeof(*cft));
748 
749 	WARN_ON(cgroup_add_legacy_cftypes(&hugetlb_cgrp_subsys,
750 					  h->cgroup_files_legacy));
751 }
752 
__hugetlb_cgroup_file_init(int idx)753 static void __init __hugetlb_cgroup_file_init(int idx)
754 {
755 	__hugetlb_cgroup_file_dfl_init(idx);
756 	__hugetlb_cgroup_file_legacy_init(idx);
757 }
758 
hugetlb_cgroup_file_init(void)759 void __init hugetlb_cgroup_file_init(void)
760 {
761 	struct hstate *h;
762 
763 	for_each_hstate(h) {
764 		/*
765 		 * Add cgroup control files only if the huge page consists
766 		 * of more than two normal pages. This is because we use
767 		 * page[2].private for storing cgroup details.
768 		 */
769 		if (huge_page_order(h) >= HUGETLB_CGROUP_MIN_ORDER)
770 			__hugetlb_cgroup_file_init(hstate_index(h));
771 	}
772 }
773 
774 /*
775  * hugetlb_lock will make sure a parallel cgroup rmdir won't happen
776  * when we migrate hugepages
777  */
hugetlb_cgroup_migrate(struct page * oldhpage,struct page * newhpage)778 void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage)
779 {
780 	struct hugetlb_cgroup *h_cg;
781 	struct hugetlb_cgroup *h_cg_rsvd;
782 	struct hstate *h = page_hstate(oldhpage);
783 
784 	if (hugetlb_cgroup_disabled())
785 		return;
786 
787 	spin_lock_irq(&hugetlb_lock);
788 	h_cg = hugetlb_cgroup_from_page(oldhpage);
789 	h_cg_rsvd = hugetlb_cgroup_from_page_rsvd(oldhpage);
790 	set_hugetlb_cgroup(oldhpage, NULL);
791 	set_hugetlb_cgroup_rsvd(oldhpage, NULL);
792 
793 	/* move the h_cg details to new cgroup */
794 	set_hugetlb_cgroup(newhpage, h_cg);
795 	set_hugetlb_cgroup_rsvd(newhpage, h_cg_rsvd);
796 	list_move(&newhpage->lru, &h->hugepage_activelist);
797 	spin_unlock_irq(&hugetlb_lock);
798 	return;
799 }
800 
801 static struct cftype hugetlb_files[] = {
802 	{} /* terminate */
803 };
804 
805 struct cgroup_subsys hugetlb_cgrp_subsys = {
806 	.css_alloc	= hugetlb_cgroup_css_alloc,
807 	.css_offline	= hugetlb_cgroup_css_offline,
808 	.css_free	= hugetlb_cgroup_css_free,
809 	.dfl_cftypes	= hugetlb_files,
810 	.legacy_cftypes	= hugetlb_files,
811 };
812