3 * Copyright IBM Corporation, 2012
4 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
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.
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.
16 #include <linux/cgroup.h>
17 #include <linux/slab.h>
18 #include <linux/hugetlb.h>
19 #include <linux/hugetlb_cgroup.h>
21 struct hugetlb_cgroup {
22 struct cgroup_subsys_state css;
24 * the counter to account for hugepages from hugetlb.
26 struct res_counter hugepage[HUGE_MAX_HSTATE];
29 struct cgroup_subsys hugetlb_subsys __read_mostly;
30 static struct hugetlb_cgroup *root_h_cgroup __read_mostly;
33 struct hugetlb_cgroup *hugetlb_cgroup_from_css(struct cgroup_subsys_state *s)
35 return container_of(s, struct hugetlb_cgroup, css);
39 struct hugetlb_cgroup *hugetlb_cgroup_from_cgroup(struct cgroup *cgroup)
41 return hugetlb_cgroup_from_css(cgroup_subsys_state(cgroup,
46 struct hugetlb_cgroup *hugetlb_cgroup_from_task(struct task_struct *task)
48 return hugetlb_cgroup_from_css(task_subsys_state(task,
52 static inline bool hugetlb_cgroup_is_root(struct hugetlb_cgroup *h_cg)
54 return (h_cg == root_h_cgroup);
57 static inline struct hugetlb_cgroup *parent_hugetlb_cgroup(struct cgroup *cg)
61 return hugetlb_cgroup_from_cgroup(cg->parent);
64 static inline bool hugetlb_cgroup_have_usage(struct cgroup *cg)
67 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cg);
69 for (idx = 0; idx < hugetlb_max_hstate; idx++) {
70 if ((res_counter_read_u64(&h_cg->hugepage[idx], RES_USAGE)) > 0)
76 static struct cgroup_subsys_state *hugetlb_cgroup_create(struct cgroup *cgroup)
79 struct cgroup *parent_cgroup;
80 struct hugetlb_cgroup *h_cgroup, *parent_h_cgroup;
82 h_cgroup = kzalloc(sizeof(*h_cgroup), GFP_KERNEL);
84 return ERR_PTR(-ENOMEM);
86 parent_cgroup = cgroup->parent;
88 parent_h_cgroup = hugetlb_cgroup_from_cgroup(parent_cgroup);
89 for (idx = 0; idx < HUGE_MAX_HSTATE; idx++)
90 res_counter_init(&h_cgroup->hugepage[idx],
91 &parent_h_cgroup->hugepage[idx]);
93 root_h_cgroup = h_cgroup;
94 for (idx = 0; idx < HUGE_MAX_HSTATE; idx++)
95 res_counter_init(&h_cgroup->hugepage[idx], NULL);
97 return &h_cgroup->css;
100 static void hugetlb_cgroup_destroy(struct cgroup *cgroup)
102 struct hugetlb_cgroup *h_cgroup;
104 h_cgroup = hugetlb_cgroup_from_cgroup(cgroup);
110 * Should be called with hugetlb_lock held.
111 * Since we are holding hugetlb_lock, pages cannot get moved from
112 * active list or uncharged from the cgroup, So no need to get
113 * page reference and test for page active here. This function
116 static void hugetlb_cgroup_move_parent(int idx, struct cgroup *cgroup,
120 struct res_counter *counter;
121 struct res_counter *fail_res;
122 struct hugetlb_cgroup *page_hcg;
123 struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_cgroup(cgroup);
124 struct hugetlb_cgroup *parent = parent_hugetlb_cgroup(cgroup);
126 page_hcg = hugetlb_cgroup_from_page(page);
128 * We can have pages in active list without any cgroup
129 * ie, hugepage with less than 3 pages. We can safely
130 * ignore those pages.
132 if (!page_hcg || page_hcg != h_cg)
135 csize = PAGE_SIZE << compound_order(page);
137 parent = root_h_cgroup;
138 /* root has no limit */
139 res_counter_charge_nofail(&parent->hugepage[idx],
142 counter = &h_cg->hugepage[idx];
143 res_counter_uncharge_until(counter, counter->parent, csize);
145 set_hugetlb_cgroup(page, parent);
151 * Force the hugetlb cgroup to empty the hugetlb resources by moving them to
154 static int hugetlb_cgroup_pre_destroy(struct cgroup *cgroup)
158 int ret = 0, idx = 0;
161 if (cgroup_task_count(cgroup) ||
162 !list_empty(&cgroup->children)) {
167 spin_lock(&hugetlb_lock);
168 list_for_each_entry(page, &h->hugepage_activelist, lru)
169 hugetlb_cgroup_move_parent(idx, cgroup, page);
171 spin_unlock(&hugetlb_lock);
175 } while (hugetlb_cgroup_have_usage(cgroup));
180 int hugetlb_cgroup_charge_cgroup(int idx, unsigned long nr_pages,
181 struct hugetlb_cgroup **ptr)
184 struct res_counter *fail_res;
185 struct hugetlb_cgroup *h_cg = NULL;
186 unsigned long csize = nr_pages * PAGE_SIZE;
188 if (hugetlb_cgroup_disabled())
191 * We don't charge any cgroup if the compound page have less
194 if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
198 h_cg = hugetlb_cgroup_from_task(current);
199 if (!css_tryget(&h_cg->css)) {
205 ret = res_counter_charge(&h_cg->hugepage[idx], csize, &fail_res);
212 void hugetlb_cgroup_commit_charge(int idx, unsigned long nr_pages,
213 struct hugetlb_cgroup *h_cg,
216 if (hugetlb_cgroup_disabled() || !h_cg)
219 spin_lock(&hugetlb_lock);
220 set_hugetlb_cgroup(page, h_cg);
221 spin_unlock(&hugetlb_lock);
226 * Should be called with hugetlb_lock held
228 void hugetlb_cgroup_uncharge_page(int idx, unsigned long nr_pages,
231 struct hugetlb_cgroup *h_cg;
232 unsigned long csize = nr_pages * PAGE_SIZE;
234 if (hugetlb_cgroup_disabled())
236 VM_BUG_ON(!spin_is_locked(&hugetlb_lock));
237 h_cg = hugetlb_cgroup_from_page(page);
240 set_hugetlb_cgroup(page, NULL);
241 res_counter_uncharge(&h_cg->hugepage[idx], csize);
245 void hugetlb_cgroup_uncharge_cgroup(int idx, unsigned long nr_pages,
246 struct hugetlb_cgroup *h_cg)
248 unsigned long csize = nr_pages * PAGE_SIZE;
250 if (hugetlb_cgroup_disabled() || !h_cg)
253 if (huge_page_order(&hstates[idx]) < HUGETLB_CGROUP_MIN_ORDER)
256 res_counter_uncharge(&h_cg->hugepage[idx], csize);
260 struct cgroup_subsys hugetlb_subsys = {
262 .create = hugetlb_cgroup_create,
263 .pre_destroy = hugetlb_cgroup_pre_destroy,
264 .destroy = hugetlb_cgroup_destroy,
265 .subsys_id = hugetlb_subsys_id,