Linux Audio

Check our new training course

Embedded Linux training

Mar 31-Apr 8, 2025
Register
Loading...
v5.4
  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/page_counter.h>
 18#include <linux/slab.h>
 19#include <linux/hugetlb.h>
 20#include <linux/hugetlb_cgroup.h>
 21
 22struct hugetlb_cgroup {
 23	struct cgroup_subsys_state css;
 24	/*
 25	 * the counter to account for hugepages from hugetlb.
 26	 */
 27	struct page_counter hugepage[HUGE_MAX_HSTATE];
 28};
 29
 30#define MEMFILE_PRIVATE(x, val)	(((x) << 16) | (val))
 31#define MEMFILE_IDX(val)	(((val) >> 16) & 0xffff)
 32#define MEMFILE_ATTR(val)	((val) & 0xffff)
 33
 34static struct hugetlb_cgroup *root_h_cgroup __read_mostly;
 35
 36static inline
 37struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s)
 38{
 39	return s ? container_of(s, struct hugetlb_cgroup, css) : NULL;
 40}
 41
 42static inline
 43struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task)
 44{
 45	return hugetlb_cgroup_from_css(task_css(task, hugetlb_cgrp_id));
 46}
 47
 48static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg)
 49{
 50	return (h_cg == root_h_cgroup);
 51}
 52
 53static inline struct hugetlb_cgroup *
 54parent_hugetlb_cgroup(struct hugetlb_cgroup *h_cg)
 55{
 56	return hugetlb_cgroup_from_css(h_cg->css.parent);
 57}
 58
 59static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg)
 60{
 61	int idx;
 62
 63	for (idx = 0; idx < hugetlb_max_hstate; idx++) {
 64		if (page_counter_read(&h_cg->hugepage[idx]))
 65			return true;
 66	}
 67	return false;
 68}
 69
 70static void hugetlb_cgroup_init(struct hugetlb_cgroup *h_cgroup,
 71				struct hugetlb_cgroup *parent_h_cgroup)
 72{
 73	int idx;
 74
 75	for (idx = 0; idx < HUGE_MAX_HSTATE; idx++) {
 76		struct page_counter *counter = &h_cgroup->hugepage[idx];
 77		struct page_counter *parent = NULL;
 78		unsigned long limit;
 79		int ret;
 80
 81		if (parent_h_cgroup)
 82			parent = &parent_h_cgroup->hugepage[idx];
 83		page_counter_init(counter, parent);
 84
 85		limit = round_down(PAGE_COUNTER_MAX,
 86				   1 << huge_page_order(&hstates[idx]));
 87		ret = page_counter_set_max(counter, limit);
 88		VM_BUG_ON(ret);
 89	}
 90}
 91
 92static struct cgroup_subsys_state *
 93hugetlb_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
 94{
 95	struct hugetlb_cgroup *parent_h_cgroup = hugetlb_cgroup_from_css(parent_css);
 96	struct hugetlb_cgroup *h_cgroup;
 
 97
 98	h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL);
 99	if (!h_cgroup)
100		return ERR_PTR(-ENOMEM);
101
102	if (!parent_h_cgroup)
 
 
 
 
103		root_h_cgroup = h_cgroup;
104
105	hugetlb_cgroup_init(h_cgroup, parent_h_cgroup);
 
106	return &h_cgroup->css;
107}
108
109static void hugetlb_cgroup_css_free(struct cgroup_subsys_state *css)
110{
111	struct hugetlb_cgroup *h_cgroup;
112
113	h_cgroup = hugetlb_cgroup_from_css(css);
114	kfree(h_cgroup);
115}
116
117
118/*
119 * Should be called with hugetlb_lock held.
120 * Since we are holding hugetlb_lock, pages cannot get moved from
121 * active list or uncharged from the cgroup, So no need to get
122 * page reference and test for page active here. This function
123 * cannot fail.
124 */
125static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg,
126				       struct page *page)
127{
128	unsigned int nr_pages;
129	struct page_counter *counter;
 
130	struct hugetlb_cgroup *page_hcg;
131	struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(h_cg);
132
133	page_hcg = hugetlb_cgroup_from_page(page);
134	/*
135	 * We can have pages in active list without any cgroup
136	 * ie, hugepage with less than 3 pages. We can safely
137	 * ignore those pages.
138	 */
139	if (!page_hcg || page_hcg != h_cg)
140		goto out;
141
142	nr_pages = compound_nr(page);
143	if (!parent) {
144		parent = root_h_cgroup;
145		/* root has no limit */
146		page_counter_charge(&parent->hugepage[idx], nr_pages);
 
147	}
148	counter = &h_cg->hugepage[idx];
149	/* Take the pages off the local counter */
150	page_counter_cancel(counter, nr_pages);
151
152	set_hugetlb_cgroup(page, parent);
153out:
154	return;
155}
156
157/*
158 * Force the hugetlb cgroup to empty the hugetlb resources by moving them to
159 * the parent cgroup.
160 */
161static void hugetlb_cgroup_css_offline(struct cgroup_subsys_state *css)
162{
163	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
164	struct hstate *h;
165	struct page *page;
166	int idx = 0;
167
168	do {
169		for_each_hstate(h) {
170			spin_lock(&hugetlb_lock);
171			list_for_each_entry(page, &h->hugepage_activelist, lru)
172				hugetlb_cgroup_move_parent(idx, h_cg, page);
173
174			spin_unlock(&hugetlb_lock);
175			idx++;
176		}
177		cond_resched();
178	} while (hugetlb_cgroup_have_usage(h_cg));
179}
180
181int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages,
182				 struct hugetlb_cgroup **ptr)
183{
184	int ret = 0;
185	struct page_counter *counter;
186	struct hugetlb_cgroup *h_cg = NULL;
 
187
188	if (hugetlb_cgroup_disabled())
189		goto done;
190	/*
191	 * We don't charge any cgroup if the compound page have less
192	 * than 3 pages.
193	 */
194	if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
195		goto done;
196again:
197	rcu_read_lock();
198	h_cg = hugetlb_cgroup_from_task(current);
199	if (!css_tryget(&h_cg->css)) {
200		rcu_read_unlock();
201		goto again;
202	}
203	rcu_read_unlock();
204
205	if (!page_counter_try_charge(&h_cg->hugepage[idx], nr_pages, &counter))
206		ret = -ENOMEM;
207	css_put(&h_cg->css);
208done:
209	*ptr = h_cg;
210	return ret;
211}
212
213/* Should be called with hugetlb_lock held */
214void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages,
215				  struct hugetlb_cgroup *h_cg,
216				  struct page *page)
217{
218	if (hugetlb_cgroup_disabled() || !h_cg)
219		return;
220
221	set_hugetlb_cgroup(page, h_cg);
222	return;
223}
224
225/*
226 * Should be called with hugetlb_lock held
227 */
228void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages,
229				  struct page *page)
230{
231	struct hugetlb_cgroup *h_cg;
 
232
233	if (hugetlb_cgroup_disabled())
234		return;
235	lockdep_assert_held(&hugetlb_lock);
236	h_cg = hugetlb_cgroup_from_page(page);
237	if (unlikely(!h_cg))
238		return;
239	set_hugetlb_cgroup(page, NULL);
240	page_counter_uncharge(&h_cg->hugepage[idx], nr_pages);
241	return;
242}
243
244void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages,
245				    struct hugetlb_cgroup *h_cg)
246{
 
 
247	if (hugetlb_cgroup_disabled() || !h_cg)
248		return;
249
250	if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
251		return;
252
253	page_counter_uncharge(&h_cg->hugepage[idx], nr_pages);
254	return;
255}
256
257enum {
258	RES_USAGE,
259	RES_LIMIT,
260	RES_MAX_USAGE,
261	RES_FAILCNT,
262};
263
264static u64 hugetlb_cgroup_read_u64(struct cgroup_subsys_state *css,
265				   struct cftype *cft)
266{
267	struct page_counter *counter;
268	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
269
270	counter = &h_cg->hugepage[MEMFILE_IDX(cft->private)];
 
271
272	switch (MEMFILE_ATTR(cft->private)) {
273	case RES_USAGE:
274		return (u64)page_counter_read(counter) * PAGE_SIZE;
275	case RES_LIMIT:
276		return (u64)counter->max * PAGE_SIZE;
277	case RES_MAX_USAGE:
278		return (u64)counter->watermark * PAGE_SIZE;
279	case RES_FAILCNT:
280		return counter->failcnt;
281	default:
282		BUG();
283	}
284}
285
286static DEFINE_MUTEX(hugetlb_limit_mutex);
287
288static ssize_t hugetlb_cgroup_write(struct kernfs_open_file *of,
289				    char *buf, size_t nbytes, loff_t off)
290{
291	int ret, idx;
292	unsigned long nr_pages;
293	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
294
295	if (hugetlb_cgroup_is_root(h_cg)) /* Can't set limit on root */
296		return -EINVAL;
297
298	buf = strstrip(buf);
299	ret = page_counter_memparse(buf, "-1", &nr_pages);
300	if (ret)
301		return ret;
302
303	idx = MEMFILE_IDX(of_cft(of)->private);
304	nr_pages = round_down(nr_pages, 1 << huge_page_order(&hstates[idx]));
305
306	switch (MEMFILE_ATTR(of_cft(of)->private)) {
307	case RES_LIMIT:
308		mutex_lock(&hugetlb_limit_mutex);
309		ret = page_counter_set_max(&h_cg->hugepage[idx], nr_pages);
310		mutex_unlock(&hugetlb_limit_mutex);
 
 
 
 
 
 
 
311		break;
312	default:
313		ret = -EINVAL;
314		break;
315	}
316	return ret ?: nbytes;
317}
318
319static ssize_t hugetlb_cgroup_reset(struct kernfs_open_file *of,
320				    char *buf, size_t nbytes, loff_t off)
321{
322	int ret = 0;
323	struct page_counter *counter;
324	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(of_css(of));
325
326	counter = &h_cg->hugepage[MEMFILE_IDX(of_cft(of)->private)];
 
327
328	switch (MEMFILE_ATTR(of_cft(of)->private)) {
329	case RES_MAX_USAGE:
330		page_counter_reset_watermark(counter);
331		break;
332	case RES_FAILCNT:
333		counter->failcnt = 0;
334		break;
335	default:
336		ret = -EINVAL;
337		break;
338	}
339	return ret ?: nbytes;
340}
341
342static char *mem_fmt(char *buf, int size, unsigned long hsize)
343{
344	if (hsize >= (1UL << 30))
345		snprintf(buf, size, "%luGB", hsize >> 30);
346	else if (hsize >= (1UL << 20))
347		snprintf(buf, size, "%luMB", hsize >> 20);
348	else
349		snprintf(buf, size, "%luKB", hsize >> 10);
350	return buf;
351}
352
353static void __init __hugetlb_cgroup_file_init(int idx)
354{
355	char buf[32];
356	struct cftype *cft;
357	struct hstate *h = &hstates[idx];
358
359	/* format the size */
360	mem_fmt(buf, 32, huge_page_size(h));
361
362	/* Add the limit file */
363	cft = &h->cgroup_files[0];
364	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.limit_in_bytes", buf);
365	cft->private = MEMFILE_PRIVATE(idx, RES_LIMIT);
366	cft->read_u64 = hugetlb_cgroup_read_u64;
367	cft->write = hugetlb_cgroup_write;
368
369	/* Add the usage file */
370	cft = &h->cgroup_files[1];
371	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.usage_in_bytes", buf);
372	cft->private = MEMFILE_PRIVATE(idx, RES_USAGE);
373	cft->read_u64 = hugetlb_cgroup_read_u64;
374
375	/* Add the MAX usage file */
376	cft = &h->cgroup_files[2];
377	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.max_usage_in_bytes", buf);
378	cft->private = MEMFILE_PRIVATE(idx, RES_MAX_USAGE);
379	cft->write = hugetlb_cgroup_reset;
380	cft->read_u64 = hugetlb_cgroup_read_u64;
381
382	/* Add the failcntfile */
383	cft = &h->cgroup_files[3];
384	snprintf(cft->name, MAX_CFTYPE_NAME, "%s.failcnt", buf);
385	cft->private  = MEMFILE_PRIVATE(idx, RES_FAILCNT);
386	cft->write = hugetlb_cgroup_reset;
387	cft->read_u64 = hugetlb_cgroup_read_u64;
388
389	/* NULL terminate the last cft */
390	cft = &h->cgroup_files[4];
391	memset(cft, 0, sizeof(*cft));
392
393	WARN_ON(cgroup_add_legacy_cftypes(&hugetlb_cgrp_subsys,
394					  h->cgroup_files));
 
395}
396
397void __init hugetlb_cgroup_file_init(void)
398{
399	struct hstate *h;
400
401	for_each_hstate(h) {
402		/*
403		 * Add cgroup control files only if the huge page consists
404		 * of more than two normal pages. This is because we use
405		 * page[2].private for storing cgroup details.
406		 */
407		if (huge_page_order(h) >= HUGETLB_CGROUP_MIN_ORDER)
408			__hugetlb_cgroup_file_init(hstate_index(h));
409	}
410}
411
412/*
413 * hugetlb_lock will make sure a parallel cgroup rmdir won't happen
414 * when we migrate hugepages
415 */
416void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage)
417{
418	struct hugetlb_cgroup *h_cg;
419	struct hstate *h = page_hstate(oldhpage);
420
421	if (hugetlb_cgroup_disabled())
422		return;
423
424	VM_BUG_ON_PAGE(!PageHuge(oldhpage), oldhpage);
425	spin_lock(&hugetlb_lock);
426	h_cg = hugetlb_cgroup_from_page(oldhpage);
427	set_hugetlb_cgroup(oldhpage, NULL);
428
429	/* move the h_cg details to new cgroup */
430	set_hugetlb_cgroup(newhpage, h_cg);
431	list_move(&newhpage->lru, &h->hugepage_activelist);
432	spin_unlock(&hugetlb_lock);
433	return;
434}
435
436struct cgroup_subsys hugetlb_cgrp_subsys = {
437	.css_alloc	= hugetlb_cgroup_css_alloc,
438	.css_offline	= hugetlb_cgroup_css_offline,
439	.css_free	= hugetlb_cgroup_css_free,
440};
v3.15
  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
 21struct 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
 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
 33static struct hugetlb_cgroup *root_h_cgroup __read_mostly;
 34
 35static inline
 36struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s)
 37{
 38	return s ? container_of(s, struct hugetlb_cgroup, css) : NULL;
 39}
 40
 41static inline
 42struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task)
 43{
 44	return hugetlb_cgroup_from_css(task_css(task, hugetlb_cgrp_id));
 45}
 46
 47static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg)
 48{
 49	return (h_cg == root_h_cgroup);
 50}
 51
 52static inline struct hugetlb_cgroup *
 53parent_hugetlb_cgroup(struct hugetlb_cgroup *h_cg)
 54{
 55	return hugetlb_cgroup_from_css(css_parent(&h_cg->css));
 56}
 57
 58static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg)
 59{
 60	int idx;
 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
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 69static struct cgroup_subsys_state *
 70hugetlb_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
 71{
 72	struct hugetlb_cgroup *parent_h_cgroup = hugetlb_cgroup_from_css(parent_css);
 73	struct hugetlb_cgroup *h_cgroup;
 74	int idx;
 75
 76	h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL);
 77	if (!h_cgroup)
 78		return ERR_PTR(-ENOMEM);
 79
 80	if (parent_h_cgroup) {
 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
 92static void hugetlb_cgroup_css_free(struct cgroup_subsys_state *css)
 93{
 94	struct hugetlb_cgroup *h_cgroup;
 95
 96	h_cgroup = hugetlb_cgroup_from_css(css);
 97	kfree(h_cgroup);
 98}
 99
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 */
108static void hugetlb_cgroup_move_parent(int idx, struct hugetlb_cgroup *h_cg,
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;
115	struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(h_cg);
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);
137out:
138	return;
139}
140
141/*
142 * Force the hugetlb cgroup to empty the hugetlb resources by moving them to
143 * the parent cgroup.
144 */
145static void hugetlb_cgroup_css_offline(struct cgroup_subsys_state *css)
146{
147	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
148	struct hstate *h;
149	struct page *page;
150	int idx = 0;
151
152	do {
153		for_each_hstate(h) {
154			spin_lock(&hugetlb_lock);
155			list_for_each_entry(page, &h->hugepage_activelist, lru)
156				hugetlb_cgroup_move_parent(idx, h_cg, page);
157
158			spin_unlock(&hugetlb_lock);
159			idx++;
160		}
161		cond_resched();
162	} while (hugetlb_cgroup_have_usage(h_cg));
163}
164
165int 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;
181again:
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);
192done:
193	*ptr = h_cg;
194	return ret;
195}
196
197/* Should be called with hugetlb_lock held */
198void 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
205	set_hugetlb_cgroup(page, h_cg);
206	return;
207}
208
209/*
210 * Should be called with hugetlb_lock held
211 */
212void 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
229void 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
 
 
 
 
 
 
 
244static u64 hugetlb_cgroup_read_u64(struct cgroup_subsys_state *css,
245				   struct cftype *cft)
246{
247	int idx, name;
248	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
249
250	idx = MEMFILE_IDX(cft->private);
251	name = MEMFILE_ATTR(cft->private);
252
253	return res_counter_read_u64(&h_cg->hugepage[idx], name);
 
 
 
 
 
 
 
 
 
 
 
254}
255
256static int hugetlb_cgroup_write(struct cgroup_subsys_state *css,
257				struct cftype *cft, char *buffer)
 
 
258{
259	int idx, name, ret;
260	unsigned long long val;
261	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
 
 
 
 
 
 
 
 
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
286static int hugetlb_cgroup_reset(struct cgroup_subsys_state *css,
287				unsigned int event)
288{
289	int idx, name, ret = 0;
290	struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
 
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
309static 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
320static void __init __hugetlb_cgroup_file_init(int idx)
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);
333	cft->read_u64 = hugetlb_cgroup_read_u64;
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);
340	cft->read_u64 = hugetlb_cgroup_read_u64;
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;
347	cft->read_u64 = hugetlb_cgroup_read_u64;
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;
354	cft->read_u64 = hugetlb_cgroup_read_u64;
355
356	/* NULL terminate the last cft */
357	cft = &h->cgroup_files[4];
358	memset(cft, 0, sizeof(*cft));
359
360	WARN_ON(cgroup_add_cftypes(&hugetlb_cgrp_subsys, h->cgroup_files));
361
362	return;
363}
364
365void __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	}
378}
379
380/*
381 * hugetlb_lock will make sure a parallel cgroup rmdir won't happen
382 * when we migrate hugepages
383 */
384void hugetlb_cgroup_migrate(struct page *oldhpage, struct page *newhpage)
385{
386	struct hugetlb_cgroup *h_cg;
387	struct hstate *h = page_hstate(oldhpage);
388
389	if (hugetlb_cgroup_disabled())
390		return;
391
392	VM_BUG_ON_PAGE(!PageHuge(oldhpage), oldhpage);
393	spin_lock(&hugetlb_lock);
394	h_cg = hugetlb_cgroup_from_page(oldhpage);
395	set_hugetlb_cgroup(oldhpage, NULL);
396
397	/* move the h_cg details to new cgroup */
398	set_hugetlb_cgroup(newhpage, h_cg);
399	list_move(&newhpage->lru, &h->hugepage_activelist);
400	spin_unlock(&hugetlb_lock);
401	return;
402}
403
404struct cgroup_subsys hugetlb_cgrp_subsys = {
405	.css_alloc	= hugetlb_cgroup_css_alloc,
406	.css_offline	= hugetlb_cgroup_css_offline,
407	.css_free	= hugetlb_cgroup_css_free,
408};