mm/hugetlb_cgroup.c

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