Lines Matching +full:charge +full:- +full:current +full:- +full:limit +full:- +full:mapping
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* memcontrol.c - Memory Controller
19 * Charge lifetime sanitation
35 #include <linux/page-flags.h>
36 #include <linux/backing-dev.h>
107 * Cgroups above their limits are maintained in a RB-Tree, independent of
199 * limit reclaim to prevent infinite loops, if they ever occur.
204 /* for encoding cft->private value on file */
236 return tsk_is_oom_victim(current) || fatal_signal_pending(current) || in task_is_dying()
237 (current->flags & PF_EXITING); in task_is_dying()
245 return &memcg->vmpressure; in memcg_to_vmpressure()
250 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; in vmpressure_to_css()
266 * objcg->nr_charged_bytes can't have an arbitrary byte value. in obj_cgroup_release()
270 * 1) CPU0: objcg == stock->cached_objcg in obj_cgroup_release()
275 * objcg->nr_charged_bytes = PAGE_SIZE - 92 in obj_cgroup_release()
277 * 92 bytes are added to stock->nr_bytes in obj_cgroup_release()
279 * 92 bytes are added to objcg->nr_charged_bytes in obj_cgroup_release()
284 nr_bytes = atomic_read(&objcg->nr_charged_bytes); in obj_cgroup_release()
285 WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); in obj_cgroup_release()
292 list_del(&objcg->list); in obj_cgroup_release()
309 ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, in obj_cgroup_alloc()
315 INIT_LIST_HEAD(&objcg->list); in obj_cgroup_alloc()
324 objcg = rcu_replace_pointer(memcg->objcg, NULL, true); in memcg_reparent_objcgs()
329 xchg(&objcg->memcg, parent); in memcg_reparent_objcgs()
330 css_get(&parent->css); in memcg_reparent_objcgs()
331 list_add(&objcg->list, &parent->objcg_list); in memcg_reparent_objcgs()
334 list_for_each_entry(iter, &memcg->objcg_list, list) { in memcg_reparent_objcgs()
335 css_get(&parent->css); in memcg_reparent_objcgs()
336 xchg(&iter->memcg, parent); in memcg_reparent_objcgs()
337 css_put(&memcg->css); in memcg_reparent_objcgs()
339 list_splice(&memcg->objcg_list, &parent->objcg_list); in memcg_reparent_objcgs()
343 percpu_ref_kill(&objcg->refcnt); in memcg_reparent_objcgs()
350 * but only a few kmem-limited. Or also, if we have, for instance, 200
351 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
354 * The current size of the caches array is stored in memcg_nr_cache_ids. It
375 * the alloc/free process all the time. In a small machine, 4 kmem-limited
416 mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); in memcg_expand_one_shrinker_map()
423 return -ENOMEM; in memcg_expand_one_shrinker_map()
426 memset(new->map, (int)0xff, old_size); in memcg_expand_one_shrinker_map()
427 memset((void *)new->map + old_size, 0, size - old_size); in memcg_expand_one_shrinker_map()
429 rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); in memcg_expand_one_shrinker_map()
430 call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); in memcg_expand_one_shrinker_map()
447 map = rcu_dereference_protected(pn->shrinker_map, true); in memcg_free_shrinker_maps()
450 rcu_assign_pointer(pn->shrinker_map, NULL); in memcg_free_shrinker_maps()
468 ret = -ENOMEM; in memcg_alloc_shrinker_maps()
471 rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); in memcg_alloc_shrinker_maps()
514 map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); in memcg_set_shrinker_bit()
517 set_bit(shrinker_id, map->map); in memcg_set_shrinker_bit()
523 * mem_cgroup_css_from_page - css of the memcg associated with a page
537 memcg = page->mem_cgroup; in mem_cgroup_css_from_page()
542 return &memcg->css; in mem_cgroup_css_from_page()
546 * page_cgroup_ino - return inode number of the memcg a page is charged to
564 memcg = page->mem_cgroup; in page_cgroup_ino()
575 while (memcg && !(memcg->css.flags & CSS_ONLINE)) in page_cgroup_ino()
578 ino = cgroup_ino(memcg->css.cgroup); in page_cgroup_ino()
588 return memcg->nodeinfo[nid]; in mem_cgroup_page_nodeinfo()
609 struct rb_node **p = &mctz->rb_root.rb_node; in __mem_cgroup_insert_exceeded()
614 if (mz->on_tree) in __mem_cgroup_insert_exceeded()
617 mz->usage_in_excess = new_usage_in_excess; in __mem_cgroup_insert_exceeded()
618 if (!mz->usage_in_excess) in __mem_cgroup_insert_exceeded()
624 if (mz->usage_in_excess < mz_node->usage_in_excess) { in __mem_cgroup_insert_exceeded()
625 p = &(*p)->rb_left; in __mem_cgroup_insert_exceeded()
631 * limit by the same amount in __mem_cgroup_insert_exceeded()
633 else if (mz->usage_in_excess >= mz_node->usage_in_excess) in __mem_cgroup_insert_exceeded()
634 p = &(*p)->rb_right; in __mem_cgroup_insert_exceeded()
638 mctz->rb_rightmost = &mz->tree_node; in __mem_cgroup_insert_exceeded()
640 rb_link_node(&mz->tree_node, parent, p); in __mem_cgroup_insert_exceeded()
641 rb_insert_color(&mz->tree_node, &mctz->rb_root); in __mem_cgroup_insert_exceeded()
642 mz->on_tree = true; in __mem_cgroup_insert_exceeded()
648 if (!mz->on_tree) in __mem_cgroup_remove_exceeded()
651 if (&mz->tree_node == mctz->rb_rightmost) in __mem_cgroup_remove_exceeded()
652 mctz->rb_rightmost = rb_prev(&mz->tree_node); in __mem_cgroup_remove_exceeded()
654 rb_erase(&mz->tree_node, &mctz->rb_root); in __mem_cgroup_remove_exceeded()
655 mz->on_tree = false; in __mem_cgroup_remove_exceeded()
663 spin_lock_irqsave(&mctz->lock, flags); in mem_cgroup_remove_exceeded()
665 spin_unlock_irqrestore(&mctz->lock, flags); in mem_cgroup_remove_exceeded()
672 struct lruvec *lruvec = &mz->lruvec; in soft_limit_excess()
677 unsigned long nr_pages = page_counter_read(&memcg->memory); in soft_limit_excess()
679 unsigned long soft_limit = READ_ONCE(memcg->soft_limit); in soft_limit_excess()
683 excess = nr_pages - soft_limit; in soft_limit_excess()
705 * We have to update the tree if mz is on RB-tree or in mem_cgroup_update_tree()
708 if (excess || mz->on_tree) { in mem_cgroup_update_tree()
711 spin_lock_irqsave(&mctz->lock, flags); in mem_cgroup_update_tree()
712 /* if on-tree, remove it */ in mem_cgroup_update_tree()
713 if (mz->on_tree) in mem_cgroup_update_tree()
716 * Insert again. mz->usage_in_excess will be updated. in mem_cgroup_update_tree()
720 spin_unlock_irqrestore(&mctz->lock, flags); in mem_cgroup_update_tree()
746 if (!mctz->rb_rightmost) in __mem_cgroup_largest_soft_limit_node()
749 mz = rb_entry(mctz->rb_rightmost, in __mem_cgroup_largest_soft_limit_node()
757 if (!soft_limit_excess(mz->memcg) || in __mem_cgroup_largest_soft_limit_node()
758 !css_tryget(&mz->memcg->css)) in __mem_cgroup_largest_soft_limit_node()
769 spin_lock_irq(&mctz->lock); in mem_cgroup_largest_soft_limit_node()
771 spin_unlock_irq(&mctz->lock); in mem_cgroup_largest_soft_limit_node()
776 * __mod_memcg_state - update cgroup memory statistics
778 * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item
791 x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); in __mod_memcg_state()
799 __this_cpu_add(memcg->vmstats_local->stat[idx], x); in __mod_memcg_state()
801 atomic_long_add(x, &mi->vmstats[idx]); in __mod_memcg_state()
804 __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); in __mod_memcg_state()
812 parent = parent_mem_cgroup(pn->memcg); in parent_nodeinfo()
826 memcg = pn->memcg; in __mod_memcg_lruvec_state()
832 __this_cpu_add(pn->lruvec_stat_local->count[idx], val); in __mod_memcg_lruvec_state()
837 x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); in __mod_memcg_lruvec_state()
842 for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) in __mod_memcg_lruvec_state()
843 atomic_long_add(x, &pi->lruvec_stat[idx]); in __mod_memcg_lruvec_state()
846 __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); in __mod_memcg_lruvec_state()
850 * __mod_lruvec_state - update lruvec memory statistics
857 * change of state at this level: per-node, per-cgroup, per-lruvec.
887 * when we free the slab object, we need to update the per-memcg in __mod_lruvec_slab_state()
911 * __count_memcg_events - account VM events in a cgroup
928 x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); in __count_memcg_events()
936 __this_cpu_add(memcg->vmstats_local->events[idx], x); in __count_memcg_events()
938 atomic_long_add(x, &mi->vmevents[idx]); in __count_memcg_events()
941 __this_cpu_write(memcg->vmstats_percpu->events[idx], x); in __count_memcg_events()
946 return atomic_long_read(&memcg->vmevents[event]); in memcg_events()
955 x += per_cpu(memcg->vmstats_local->events[event], cpu); in memcg_events_local()
968 nr_pages = -nr_pages; /* for event */ in mem_cgroup_charge_statistics()
971 __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); in mem_cgroup_charge_statistics()
979 val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); in mem_cgroup_event_ratelimit()
980 next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); in mem_cgroup_event_ratelimit()
982 if ((long)(next - val) < 0) { in mem_cgroup_event_ratelimit()
993 __this_cpu_write(memcg->vmstats_percpu->targets[target], next); in mem_cgroup_event_ratelimit()
1005 /* threshold event is triggered in finer grain than soft limit */ in memcg_check_events()
1021 * mm_update_next_owner() may clear mm->owner to NULL in mem_cgroup_from_task()
1036 * Obtain a reference on mm->memcg and returns it if successful. Otherwise
1057 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); in get_mem_cgroup_from_mm()
1061 } while (!css_tryget(&memcg->css)); in get_mem_cgroup_from_mm()
1071 * Obtain a reference on page->memcg and returns it if successful. Otherwise
1076 struct mem_cgroup *memcg = page->mem_cgroup; in get_mem_cgroup_from_page()
1083 if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) in get_mem_cgroup_from_page()
1095 return current->active_memcg; in active_memcg()
1105 if (memcg && WARN_ON_ONCE(!css_tryget(&memcg->css))) in get_active_memcg()
1118 /* Memcg to charge can't be determined. */ in memcg_kmem_bypass()
1119 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) in memcg_kmem_bypass()
1126 * If active memcg is set, do not fallback to current->mm->memcg.
1136 return get_mem_cgroup_from_mm(current->mm); in get_mem_cgroup_from_current()
1140 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1146 * @root itself, or %NULL after a full round-trip.
1150 * to cancel a hierarchy walk before the round-trip is complete.
1174 if (!root->use_hierarchy && root != root_mem_cgroup) { in mem_cgroup_iter()
1185 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); in mem_cgroup_iter()
1186 iter = &mz->iter; in mem_cgroup_iter()
1188 if (prev && reclaim->generation != iter->generation) in mem_cgroup_iter()
1192 pos = READ_ONCE(iter->position); in mem_cgroup_iter()
1193 if (!pos || css_tryget(&pos->css)) in mem_cgroup_iter()
1196 * css reference reached zero, so iter->position will in mem_cgroup_iter()
1197 * be cleared by ->css_released. However, we should not in mem_cgroup_iter()
1198 * rely on this happening soon, because ->css_released in mem_cgroup_iter()
1199 * is called from a work queue, and by busy-waiting we in mem_cgroup_iter()
1200 * might block it. So we clear iter->position right in mem_cgroup_iter()
1203 (void)cmpxchg(&iter->position, pos, NULL); in mem_cgroup_iter()
1208 css = &pos->css; in mem_cgroup_iter()
1211 css = css_next_descendant_pre(css, &root->css); in mem_cgroup_iter()
1216 * the hierarchy - make sure they see at least in mem_cgroup_iter()
1231 if (css == &root->css) in mem_cgroup_iter()
1246 (void)cmpxchg(&iter->position, pos, memcg); in mem_cgroup_iter()
1249 css_put(&pos->css); in mem_cgroup_iter()
1252 iter->generation++; in mem_cgroup_iter()
1254 reclaim->generation = iter->generation; in mem_cgroup_iter()
1261 css_put(&prev->css); in mem_cgroup_iter()
1267 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1277 css_put(&prev->css); in mem_cgroup_iter_break()
1289 iter = &mz->iter; in __invalidate_reclaim_iterators()
1290 cmpxchg(&iter->position, dead_memcg, NULL); in __invalidate_reclaim_iterators()
1305 * When cgruop1 non-hierarchy mode is used, in invalidate_reclaim_iterators()
1316 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
1322 * descendants and calls @fn for each task. If @fn returns a non-zero
1340 css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); in mem_cgroup_scan_tasks()
1353 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
1357 * This function relies on page->mem_cgroup being stable - see the
1367 lruvec = &pgdat->__lruvec; in mem_cgroup_page_lruvec()
1378 memcg = page->mem_cgroup; in mem_cgroup_page_lruvec()
1380 * Swapcache readahead pages are added to the LRU - and in mem_cgroup_page_lruvec()
1381 * possibly migrated - before they are charged. in mem_cgroup_page_lruvec()
1387 lruvec = &mz->lruvec; in mem_cgroup_page_lruvec()
1391 * we have to be prepared to initialize lruvec->zone here; in mem_cgroup_page_lruvec()
1394 if (unlikely(lruvec->pgdat != pgdat)) in mem_cgroup_page_lruvec()
1395 lruvec->pgdat = pgdat; in mem_cgroup_page_lruvec()
1400 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1425 lru_size = &mz->lru_zone_size[zid][lru]; in mem_cgroup_update_lru_size()
1443 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1453 unsigned long limit; in mem_cgroup_margin() local
1455 count = page_counter_read(&memcg->memory); in mem_cgroup_margin()
1456 limit = READ_ONCE(memcg->memory.max); in mem_cgroup_margin()
1457 if (count < limit) in mem_cgroup_margin()
1458 margin = limit - count; in mem_cgroup_margin()
1461 count = page_counter_read(&memcg->memsw); in mem_cgroup_margin()
1462 limit = READ_ONCE(memcg->memsw.max); in mem_cgroup_margin()
1463 if (count < limit) in mem_cgroup_margin()
1464 margin = min(margin, limit - count); in mem_cgroup_margin()
1476 * moving cgroups. This is for waiting at high-memory pressure
1503 if (mc.moving_task && current != mc.moving_task) { in mem_cgroup_wait_acct_move()
1507 /* moving charge context might have finished. */ in mem_cgroup_wait_acct_move()
1595 * 1) generic big picture -> specifics and details in memory_stat_format()
1596 * 2) reflecting userspace activity -> reflecting kernel heuristics in memory_stat_format()
1598 * Current memory state: in memory_stat_format()
1651 #define K(x) ((x) << (PAGE_SHIFT-10))
1655 * @memcg: The memory cgroup that went over limit
1667 pr_cont_cgroup_path(memcg->css.cgroup); in mem_cgroup_print_oom_context()
1680 * @memcg: The memory cgroup that went over limit
1686 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", in mem_cgroup_print_oom_meminfo()
1687 K((u64)page_counter_read(&memcg->memory)), in mem_cgroup_print_oom_meminfo()
1688 K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); in mem_cgroup_print_oom_meminfo()
1690 pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", in mem_cgroup_print_oom_meminfo()
1691 K((u64)page_counter_read(&memcg->swap)), in mem_cgroup_print_oom_meminfo()
1692 K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); in mem_cgroup_print_oom_meminfo()
1694 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", in mem_cgroup_print_oom_meminfo()
1695 K((u64)page_counter_read(&memcg->memsw)), in mem_cgroup_print_oom_meminfo()
1696 K((u64)memcg->memsw.max), memcg->memsw.failcnt); in mem_cgroup_print_oom_meminfo()
1697 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", in mem_cgroup_print_oom_meminfo()
1698 K((u64)page_counter_read(&memcg->kmem)), in mem_cgroup_print_oom_meminfo()
1699 K((u64)memcg->kmem.max), memcg->kmem.failcnt); in mem_cgroup_print_oom_meminfo()
1703 pr_cont_cgroup_path(memcg->css.cgroup); in mem_cgroup_print_oom_meminfo()
1713 * Return the memory (and swap, if configured) limit for a memcg.
1717 unsigned long max = READ_ONCE(memcg->memory.max); in mem_cgroup_get_max()
1721 max += min(READ_ONCE(memcg->swap.max), in mem_cgroup_get_max()
1725 /* Calculate swap excess capacity from memsw limit */ in mem_cgroup_get_max()
1726 unsigned long swap = READ_ONCE(memcg->memsw.max) - max; in mem_cgroup_get_max()
1736 return page_counter_read(&memcg->memory); in mem_cgroup_size()
1827 * Check OOM-Killer is already running under our hierarchy.
1837 if (iter->oom_lock) { in mem_cgroup_oom_trylock()
1846 iter->oom_lock = true; in mem_cgroup_oom_trylock()
1859 iter->oom_lock = false; in mem_cgroup_oom_trylock()
1876 iter->oom_lock = false; in mem_cgroup_oom_unlock()
1886 iter->under_oom++; in mem_cgroup_mark_under_oom()
1900 if (iter->under_oom > 0) in mem_cgroup_unmark_under_oom()
1901 iter->under_oom--; in mem_cgroup_unmark_under_oom()
1920 oom_wait_memcg = oom_wait_info->memcg; in memcg_oom_wake_function()
1931 * For the following lockless ->under_oom test, the only required in memcg_oom_recover()
1938 if (memcg && memcg->under_oom) in memcg_oom_recover()
1960 * We are in the middle of the charge context here, so we in mem_cgroup_oom()
1965 * handling until the charge can succeed; remember the context and put in mem_cgroup_oom()
1969 * On the other hand, in-kernel OOM killer allows for an async victim in mem_cgroup_oom()
1975 * victim and then we have to bail out from the charge path. in mem_cgroup_oom()
1977 if (memcg->oom_kill_disable) { in mem_cgroup_oom()
1978 if (!current->in_user_fault) in mem_cgroup_oom()
1980 css_get(&memcg->css); in mem_cgroup_oom()
1981 current->memcg_in_oom = memcg; in mem_cgroup_oom()
1982 current->memcg_oom_gfp_mask = mask; in mem_cgroup_oom()
1983 current->memcg_oom_order = order; in mem_cgroup_oom()
2008 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2016 * situation. Sleeping directly in the charge context with all kinds
2026 struct mem_cgroup *memcg = current->memcg_in_oom; in mem_cgroup_oom_synchronize()
2040 owait.wait.private = current; in mem_cgroup_oom_synchronize()
2051 if (locked && !memcg->oom_kill_disable) { in mem_cgroup_oom_synchronize()
2054 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, in mem_cgroup_oom_synchronize()
2055 current->memcg_oom_order); in mem_cgroup_oom_synchronize()
2065 * There is no guarantee that an OOM-lock contender in mem_cgroup_oom_synchronize()
2072 current->memcg_in_oom = NULL; in mem_cgroup_oom_synchronize()
2073 css_put(&memcg->css); in mem_cgroup_oom_synchronize()
2078 * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM
2080 * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM
2083 * by killing all belonging OOM-killable tasks.
2085 * Caller has to call mem_cgroup_put() on the returned non-NULL memcg.
2116 * highest-level memory cgroup with oom.group set. in mem_cgroup_get_oom_group()
2119 if (memcg->oom_group) in mem_cgroup_get_oom_group()
2127 css_get(&oom_group->css); in mem_cgroup_get_oom_group()
2137 pr_cont_cgroup_path(memcg->css.cgroup); in mem_cgroup_print_oom_group()
2142 * lock_page_memcg - lock a page->mem_cgroup binding
2160 * path can get away without acquiring the memcg->move_lock in lock_page_memcg()
2174 memcg = head->mem_cgroup; in lock_page_memcg()
2178 if (atomic_read(&memcg->moving_account) <= 0) in lock_page_memcg()
2181 spin_lock_irqsave(&memcg->move_lock, flags); in lock_page_memcg()
2182 if (memcg != head->mem_cgroup) { in lock_page_memcg()
2183 spin_unlock_irqrestore(&memcg->move_lock, flags); in lock_page_memcg()
2188 * When charge migration first begins, we can have locked and in lock_page_memcg()
2192 memcg->move_lock_task = current; in lock_page_memcg()
2193 memcg->move_lock_flags = flags; in lock_page_memcg()
2200 * __unlock_page_memcg - unlock and unpin a memcg
2207 if (memcg && memcg->move_lock_task == current) { in __unlock_page_memcg()
2208 unsigned long flags = memcg->move_lock_flags; in __unlock_page_memcg()
2210 memcg->move_lock_task = NULL; in __unlock_page_memcg()
2211 memcg->move_lock_flags = 0; in __unlock_page_memcg()
2213 spin_unlock_irqrestore(&memcg->move_lock, flags); in __unlock_page_memcg()
2220 * unlock_page_memcg - unlock a page->mem_cgroup binding
2227 __unlock_page_memcg(head->mem_cgroup); in unlock_page_memcg()
2264 * consume_stock: Try to consume stocked charge on this cpu.
2266 * @nr_pages: how many pages to charge.
2268 * The charges will only happen if @memcg matches the current cpu's memcg
2286 if (memcg == stock->cached && stock->nr_pages >= nr_pages) { in consume_stock()
2287 stock->nr_pages -= nr_pages; in consume_stock()
2301 struct mem_cgroup *old = stock->cached; in drain_stock()
2306 if (stock->nr_pages) { in drain_stock()
2307 page_counter_uncharge(&old->memory, stock->nr_pages); in drain_stock()
2309 page_counter_uncharge(&old->memsw, stock->nr_pages); in drain_stock()
2310 stock->nr_pages = 0; in drain_stock()
2313 css_put(&old->css); in drain_stock()
2314 stock->cached = NULL; in drain_stock()
2331 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); in drain_local_stock()
2348 if (stock->cached != memcg) { /* reset if necessary */ in refill_stock()
2350 css_get(&memcg->css); in refill_stock()
2351 stock->cached = memcg; in refill_stock()
2353 stock->nr_pages += nr_pages; in refill_stock()
2355 if (stock->nr_pages > MEMCG_CHARGE_BATCH) in refill_stock()
2362 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2373 * Notify other cpus that system-wide "drain" is running in drain_all_stock()
2376 * per-cpu data. CPU up doesn't touch memcg_stock at all. in drain_all_stock()
2385 memcg = stock->cached; in drain_all_stock()
2386 if (memcg && stock->nr_pages && in drain_all_stock()
2394 !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { in drain_all_stock()
2396 drain_local_stock(&stock->work); in drain_all_stock()
2398 schedule_work_on(cpu, &stock->work); in drain_all_stock()
2420 x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); in memcg_hotplug_cpu_dead()
2423 atomic_long_add(x, &memcg->vmstats[i]); in memcg_hotplug_cpu_dead()
2432 x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); in memcg_hotplug_cpu_dead()
2435 atomic_long_add(x, &pn->lruvec_stat[i]); in memcg_hotplug_cpu_dead()
2443 x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); in memcg_hotplug_cpu_dead()
2446 atomic_long_add(x, &memcg->vmevents[i]); in memcg_hotplug_cpu_dead()
2462 if (page_counter_read(&memcg->memory) <= in reclaim_high()
2463 READ_ONCE(memcg->memory.high)) in reclaim_high()
2498 * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the
2500 * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the
2505 * reasonable delay curve compared to precision-adjusted overage, not
2507 * limit penalises misbehaviour cgroups by slowing them down exponentially. For
2510 * +-------+------------------------+
2512 * +-------+------------------------+
2534 * +-------+------------------------+
2552 overage = usage - high; in calculate_overage()
2562 overage = calculate_overage(page_counter_read(&memcg->memory), in mem_find_max_overage()
2563 READ_ONCE(memcg->memory.high)); in mem_find_max_overage()
2576 overage = calculate_overage(page_counter_read(&memcg->swap), in swap_find_max_overage()
2577 READ_ONCE(memcg->swap.high)); in swap_find_max_overage()
2614 * N-sized allocations are throttled approximately the same as one in calculate_high_delay()
2615 * 4N-sized allocation. in calculate_high_delay()
2618 * larger the current charge patch is than that. in calculate_high_delay()
2625 * and reclaims memory over the high limit.
2632 unsigned int nr_pages = current->memcg_nr_pages_over_high; in mem_cgroup_handle_over_high()
2640 memcg = get_mem_cgroup_from_mm(current->mm); in mem_cgroup_handle_over_high()
2641 current->memcg_nr_pages_over_high = 0; in mem_cgroup_handle_over_high()
2688 if (nr_reclaimed || nr_retries--) { in mem_cgroup_handle_over_high()
2696 * need to account for any ill-begotten jiffies to pay them off later. in mem_cgroup_handle_over_high()
2703 css_put(&memcg->css); in mem_cgroup_handle_over_high()
2727 page_counter_try_charge(&memcg->memsw, batch, &counter)) { in try_charge()
2728 if (page_counter_try_charge(&memcg->memory, batch, &counter)) in try_charge()
2731 page_counter_uncharge(&memcg->memsw, batch); in try_charge()
2756 * under the limit over triggering OOM kills in these cases. in try_charge()
2758 if (unlikely(current->flags & PF_MEMALLOC)) in try_charge()
2761 if (unlikely(task_in_memcg_oom(current))) in try_charge()
2786 * Even though the limit is exceeded at this point, reclaim in try_charge()
2787 * may have been able to free some pages. Retry the charge in try_charge()
2791 * unlikely to succeed so close to the limit, and we fall back in try_charge()
2797 * At task move, charge accounts can be doubly counted. So, it's in try_charge()
2803 if (nr_retries--) in try_charge()
2818 * a forward progress or bypass the charge if the oom killer in try_charge()
2830 return -ENOMEM; in try_charge()
2834 * being freed very soon. Allow memory usage go over the limit in try_charge()
2837 page_counter_charge(&memcg->memory, nr_pages); in try_charge()
2839 page_counter_charge(&memcg->memsw, nr_pages); in try_charge()
2845 refill_stock(memcg, batch - nr_pages); in try_charge()
2852 * not recorded as it most likely matches current's and won't in try_charge()
2853 * change in the meantime. As high limit is checked again before in try_charge()
2859 mem_high = page_counter_read(&memcg->memory) > in try_charge()
2860 READ_ONCE(memcg->memory.high); in try_charge()
2861 swap_high = page_counter_read(&memcg->swap) > in try_charge()
2862 READ_ONCE(memcg->swap.high); in try_charge()
2867 schedule_work(&memcg->high_work); in try_charge()
2879 * Target some best-effort fairness between the tasks, in try_charge()
2883 current->memcg_nr_pages_over_high += batch; in try_charge()
2884 set_notify_resume(current); in try_charge()
2898 page_counter_uncharge(&memcg->memory, nr_pages); in cancel_charge()
2900 page_counter_uncharge(&memcg->memsw, nr_pages); in cancel_charge()
2906 VM_BUG_ON_PAGE(page->mem_cgroup, page); in commit_charge()
2908 * Any of the following ensures page->mem_cgroup stability: in commit_charge()
2910 * - the page lock in commit_charge()
2911 * - LRU isolation in commit_charge()
2912 * - lock_page_memcg() in commit_charge()
2913 * - exclusive reference in commit_charge()
2915 page->mem_cgroup = memcg; in commit_charge()
2936 return -ENOMEM; in memcg_alloc_page_obj_cgroups()
2938 if (cmpxchg(&page->obj_cgroups, NULL, in memcg_alloc_page_obj_cgroups()
2963 * If page->mem_cgroup is set, it's either a simple mem_cgroup pointer in mem_cgroup_from_obj()
2966 * The page->mem_cgroup pointer can be asynchronously changed in mem_cgroup_from_obj()
2970 if (!page->mem_cgroup) in mem_cgroup_from_obj()
2974 * Slab objects are accounted individually, not per-page. in mem_cgroup_from_obj()
2976 * the page->obj_cgroups. in mem_cgroup_from_obj()
2982 off = obj_to_index(page->slab_cache, page, p); in mem_cgroup_from_obj()
2990 /* All other pages use page->mem_cgroup */ in mem_cgroup_from_obj()
2991 return page->mem_cgroup; in mem_cgroup_from_obj()
3006 memcg = mem_cgroup_from_task(current); in get_obj_cgroup_from_current()
3009 objcg = rcu_dereference(memcg->objcg); in get_obj_cgroup_from_current()
3063 * __memcg_kmem_charge: charge a number of kernel pages to a memcg
3064 * @memcg: memory cgroup to charge
3066 * @nr_pages: number of pages to charge
3081 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { in __memcg_kmem_charge()
3089 page_counter_charge(&memcg->kmem, nr_pages); in __memcg_kmem_charge()
3093 return -ENOMEM; in __memcg_kmem_charge()
3106 page_counter_uncharge(&memcg->kmem, nr_pages); in __memcg_kmem_uncharge()
3112 * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup
3113 * @page: page to charge
3128 page->mem_cgroup = memcg; in __memcg_kmem_charge_page()
3132 css_put(&memcg->css); in __memcg_kmem_charge_page()
3144 struct mem_cgroup *memcg = page->mem_cgroup; in __memcg_kmem_uncharge_page()
3152 page->mem_cgroup = NULL; in __memcg_kmem_uncharge_page()
3153 css_put(&memcg->css); in __memcg_kmem_uncharge_page()
3169 if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { in consume_obj_stock()
3170 stock->nr_bytes -= nr_bytes; in consume_obj_stock()
3181 struct obj_cgroup *old = stock->cached_objcg; in drain_obj_stock()
3186 if (stock->nr_bytes) { in drain_obj_stock()
3187 unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; in drain_obj_stock()
3188 unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); in drain_obj_stock()
3196 if (unlikely(!css_tryget(&memcg->css))) in drain_obj_stock()
3201 css_put(&memcg->css); in drain_obj_stock()
3205 * The leftover is flushed to the centralized per-memcg value. in drain_obj_stock()
3207 * to a per-cpu stock (probably, on an other CPU), see in drain_obj_stock()
3210 * How often it's flushed is a trade-off between the memory in drain_obj_stock()
3211 * limit enforcement accuracy and potential CPU contention, in drain_obj_stock()
3214 atomic_add(nr_bytes, &old->nr_charged_bytes); in drain_obj_stock()
3215 stock->nr_bytes = 0; in drain_obj_stock()
3219 stock->cached_objcg = NULL; in drain_obj_stock()
3227 if (stock->cached_objcg) { in obj_stock_flush_required()
3228 memcg = obj_cgroup_memcg(stock->cached_objcg); in obj_stock_flush_required()
3244 if (stock->cached_objcg != objcg) { /* reset if necessary */ in refill_obj_stock()
3247 stock->cached_objcg = objcg; in refill_obj_stock()
3248 stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); in refill_obj_stock()
3250 stock->nr_bytes += nr_bytes; in refill_obj_stock()
3252 if (stock->nr_bytes > PAGE_SIZE) in refill_obj_stock()
3268 * In theory, memcg->nr_charged_bytes can have enough in obj_cgroup_charge()
3269 * pre-charged bytes to satisfy the allocation. However, in obj_cgroup_charge()
3270 * flushing memcg->nr_charged_bytes requires two atomic in obj_cgroup_charge()
3271 * operations, and memcg->nr_charged_bytes can't be big, in obj_cgroup_charge()
3273 * memcg->nr_charged_bytes will be flushed in in obj_cgroup_charge()
3280 if (unlikely(!css_tryget(&memcg->css))) in obj_cgroup_charge()
3285 nr_bytes = size & (PAGE_SIZE - 1); in obj_cgroup_charge()
3292 refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); in obj_cgroup_charge()
3294 css_put(&memcg->css); in obj_cgroup_charge()
3306 * Because head->mem_cgroup is not set on tails, set it now.
3310 struct mem_cgroup *memcg = head->mem_cgroup; in split_page_memcg()
3322 css_get_many(&memcg->css, nr - 1); in split_page_memcg()
3327 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3335 * Returns 0 on success, -EINVAL on failure.
3349 mod_memcg_state(from, MEMCG_SWAP, -1); in mem_cgroup_move_swap_account()
3353 return -EINVAL; in mem_cgroup_move_swap_account()
3359 return -EINVAL; in mem_cgroup_move_swap_account()
3372 struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; in mem_cgroup_resize_max()
3375 if (signal_pending(current)) { in mem_cgroup_resize_max()
3376 ret = -EINTR; in mem_cgroup_resize_max()
3382 * Make sure that the new limit (memsw or memory limit) doesn't in mem_cgroup_resize_max()
3385 limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : in mem_cgroup_resize_max()
3386 max <= memcg->memsw.max; in mem_cgroup_resize_max()
3389 ret = -EINVAL; in mem_cgroup_resize_max()
3392 if (max > counter->max) in mem_cgroup_resize_max()
3408 ret = -EBUSY; in mem_cgroup_resize_max()
3434 mctz = soft_limit_tree_node(pgdat->node_id); in mem_cgroup_soft_limit_reclaim()
3439 * are acceptable as soft limit is best effort anyway. in mem_cgroup_soft_limit_reclaim()
3441 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) in mem_cgroup_soft_limit_reclaim()
3446 * keep exceeding their soft limit and putting the system under in mem_cgroup_soft_limit_reclaim()
3458 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, in mem_cgroup_soft_limit_reclaim()
3462 spin_lock_irq(&mctz->lock); in mem_cgroup_soft_limit_reclaim()
3473 excess = soft_limit_excess(mz->memcg); in mem_cgroup_soft_limit_reclaim()
3484 spin_unlock_irq(&mctz->lock); in mem_cgroup_soft_limit_reclaim()
3485 css_put(&mz->memcg->css); in mem_cgroup_soft_limit_reclaim()
3498 css_put(&next_mz->memcg->css); in mem_cgroup_soft_limit_reclaim()
3513 ret = css_next_child(NULL, &memcg->css); in memcg_has_children()
3527 /* we call try-to-free pages for make this cgroup empty */ in mem_cgroup_force_empty()
3533 while (nr_retries && page_counter_read(&memcg->memory)) { in mem_cgroup_force_empty()
3536 if (signal_pending(current)) in mem_cgroup_force_empty()
3537 return -EINTR; in mem_cgroup_force_empty()
3542 nr_retries--; in mem_cgroup_force_empty()
3559 return -EINVAL; in mem_cgroup_force_empty_write()
3566 return mem_cgroup_from_css(css)->use_hierarchy; in mem_cgroup_hierarchy_read()
3574 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); in mem_cgroup_hierarchy_write()
3576 if (memcg->use_hierarchy == val) in mem_cgroup_hierarchy_write()
3582 * occur, provided the current cgroup has no children. in mem_cgroup_hierarchy_write()
3587 if ((!parent_memcg || !parent_memcg->use_hierarchy) && in mem_cgroup_hierarchy_write()
3590 memcg->use_hierarchy = val; in mem_cgroup_hierarchy_write()
3592 retval = -EBUSY; in mem_cgroup_hierarchy_write()
3594 retval = -EINVAL; in mem_cgroup_hierarchy_write()
3610 val = page_counter_read(&memcg->memory); in mem_cgroup_usage()
3612 val = page_counter_read(&memcg->memsw); in mem_cgroup_usage()
3631 switch (MEMFILE_TYPE(cft->private)) { in mem_cgroup_read_u64()
3633 counter = &memcg->memory; in mem_cgroup_read_u64()
3636 counter = &memcg->memsw; in mem_cgroup_read_u64()
3639 counter = &memcg->kmem; in mem_cgroup_read_u64()
3642 counter = &memcg->tcpmem; in mem_cgroup_read_u64()
3648 switch (MEMFILE_ATTR(cft->private)) { in mem_cgroup_read_u64()
3650 if (counter == &memcg->memory) in mem_cgroup_read_u64()
3652 if (counter == &memcg->memsw) in mem_cgroup_read_u64()
3656 return (u64)counter->max * PAGE_SIZE; in mem_cgroup_read_u64()
3658 return (u64)counter->watermark * PAGE_SIZE; in mem_cgroup_read_u64()
3660 return counter->failcnt; in mem_cgroup_read_u64()
3662 return (u64)memcg->soft_limit * PAGE_SIZE; in mem_cgroup_read_u64()
3676 stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); in memcg_flush_percpu_vmstats()
3680 atomic_long_add(stat[i], &mi->vmstats[i]); in memcg_flush_percpu_vmstats()
3683 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; in memcg_flush_percpu_vmstats()
3692 pn->lruvec_stat_cpu->count[i], cpu); in memcg_flush_percpu_vmstats()
3696 atomic_long_add(stat[i], &pi->lruvec_stat[i]); in memcg_flush_percpu_vmstats()
3711 events[i] += per_cpu(memcg->vmstats_percpu->events[i], in memcg_flush_percpu_vmevents()
3716 atomic_long_add(events[i], &mi->vmevents[i]); in memcg_flush_percpu_vmevents()
3728 BUG_ON(memcg->kmemcg_id >= 0); in memcg_online_kmem()
3729 BUG_ON(memcg->kmem_state); in memcg_online_kmem()
3738 return -ENOMEM; in memcg_online_kmem()
3740 objcg->memcg = memcg; in memcg_online_kmem()
3741 rcu_assign_pointer(memcg->objcg, objcg); in memcg_online_kmem()
3746 * A memory cgroup is considered kmem-online as soon as it gets in memcg_online_kmem()
3751 memcg->kmemcg_id = memcg_id; in memcg_online_kmem()
3752 memcg->kmem_state = KMEM_ONLINE; in memcg_online_kmem()
3763 if (memcg->kmem_state != KMEM_ONLINE) in memcg_offline_kmem()
3766 memcg->kmem_state = KMEM_ALLOCATED; in memcg_offline_kmem()
3774 kmemcg_id = memcg->kmemcg_id; in memcg_offline_kmem()
3782 * ordering is imposed by list_lru_node->lock taken by in memcg_offline_kmem()
3786 css_for_each_descendant_pre(css, &memcg->css) { in memcg_offline_kmem()
3788 BUG_ON(child->kmemcg_id != kmemcg_id); in memcg_offline_kmem()
3789 child->kmemcg_id = parent->kmemcg_id; in memcg_offline_kmem()
3790 if (!memcg->use_hierarchy) in memcg_offline_kmem()
3803 if (unlikely(memcg->kmem_state == KMEM_ONLINE)) in memcg_free_kmem()
3825 ret = page_counter_set_max(&memcg->kmem, max); in memcg_update_kmem_max()
3836 ret = page_counter_set_max(&memcg->tcpmem, max); in memcg_update_tcp_max()
3840 if (!memcg->tcpmem_active) { in memcg_update_tcp_max()
3858 memcg->tcpmem_active = true; in memcg_update_tcp_max()
3877 ret = page_counter_memparse(buf, "-1", &nr_pages); in mem_cgroup_write()
3881 switch (MEMFILE_ATTR(of_cft(of)->private)) { in mem_cgroup_write()
3883 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ in mem_cgroup_write()
3884 ret = -EINVAL; in mem_cgroup_write()
3887 switch (MEMFILE_TYPE(of_cft(of)->private)) { in mem_cgroup_write()
3896 "Please report your usecase to linux-mm@kvack.org if you " in mem_cgroup_write()
3906 memcg->soft_limit = nr_pages; in mem_cgroup_write()
3919 switch (MEMFILE_TYPE(of_cft(of)->private)) { in mem_cgroup_reset()
3921 counter = &memcg->memory; in mem_cgroup_reset()
3924 counter = &memcg->memsw; in mem_cgroup_reset()
3927 counter = &memcg->kmem; in mem_cgroup_reset()
3930 counter = &memcg->tcpmem; in mem_cgroup_reset()
3936 switch (MEMFILE_ATTR(of_cft(of)->private)) { in mem_cgroup_reset()
3941 counter->failcnt = 0; in mem_cgroup_reset()
3953 return mem_cgroup_from_css(css)->move_charge_at_immigrate; in mem_cgroup_move_charge_read()
3963 return -EINVAL; in mem_cgroup_move_charge_write()
3966 * No kind of locking is needed in here, because ->can_attach() will in mem_cgroup_move_charge_write()
3971 memcg->move_charge_at_immigrate = val; in mem_cgroup_move_charge_write()
3978 return -ENOSYS; in mem_cgroup_move_charge_write()
3986 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
4044 seq_printf(m, "%s=%lu", stat->name, in memcg_numa_stat_show()
4045 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, in memcg_numa_stat_show()
4050 stat->lru_mask, false)); in memcg_numa_stat_show()
4056 seq_printf(m, "hierarchical_%s=%lu", stat->name, in memcg_numa_stat_show()
4057 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, in memcg_numa_stat_show()
4062 stat->lru_mask, true)); in memcg_numa_stat_show()
4138 memory = min(memory, READ_ONCE(mi->memory.max)); in memcg_stat_show()
4139 memsw = min(memsw, READ_ONCE(mi->memsw.max)); in memcg_stat_show()
4179 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); in memcg_stat_show()
4181 anon_cost += mz->lruvec.anon_cost; in memcg_stat_show()
4182 file_cost += mz->lruvec.file_cost; in memcg_stat_show()
4209 return -EINVAL; in mem_cgroup_swappiness_write()
4211 if (css->parent) in mem_cgroup_swappiness_write()
4212 memcg->swappiness = val; in mem_cgroup_swappiness_write()
4227 t = rcu_dereference(memcg->thresholds.primary); in __mem_cgroup_threshold()
4229 t = rcu_dereference(memcg->memsw_thresholds.primary); in __mem_cgroup_threshold()
4241 i = t->current_threshold; in __mem_cgroup_threshold()
4249 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) in __mem_cgroup_threshold()
4250 eventfd_signal(t->entries[i].eventfd, 1); in __mem_cgroup_threshold()
4261 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) in __mem_cgroup_threshold()
4262 eventfd_signal(t->entries[i].eventfd, 1); in __mem_cgroup_threshold()
4265 t->current_threshold = i - 1; in __mem_cgroup_threshold()
4286 if (_a->threshold > _b->threshold) in compare_thresholds()
4289 if (_a->threshold < _b->threshold) in compare_thresholds()
4290 return -1; in compare_thresholds()
4301 list_for_each_entry(ev, &memcg->oom_notify, list) in mem_cgroup_oom_notify_cb()
4302 eventfd_signal(ev->eventfd, 1); in mem_cgroup_oom_notify_cb()
4325 ret = page_counter_memparse(args, "-1", &threshold); in __mem_cgroup_usage_register_event()
4329 mutex_lock(&memcg->thresholds_lock); in __mem_cgroup_usage_register_event()
4332 thresholds = &memcg->thresholds; in __mem_cgroup_usage_register_event()
4335 thresholds = &memcg->memsw_thresholds; in __mem_cgroup_usage_register_event()
4341 if (thresholds->primary) in __mem_cgroup_usage_register_event()
4344 size = thresholds->primary ? thresholds->primary->size + 1 : 1; in __mem_cgroup_usage_register_event()
4349 ret = -ENOMEM; in __mem_cgroup_usage_register_event()
4352 new->size = size; in __mem_cgroup_usage_register_event()
4355 if (thresholds->primary) in __mem_cgroup_usage_register_event()
4356 memcpy(new->entries, thresholds->primary->entries, in __mem_cgroup_usage_register_event()
4357 flex_array_size(new, entries, size - 1)); in __mem_cgroup_usage_register_event()
4360 new->entries[size - 1].eventfd = eventfd; in __mem_cgroup_usage_register_event()
4361 new->entries[size - 1].threshold = threshold; in __mem_cgroup_usage_register_event()
4363 /* Sort thresholds. Registering of new threshold isn't time-critical */ in __mem_cgroup_usage_register_event()
4364 sort(new->entries, size, sizeof(*new->entries), in __mem_cgroup_usage_register_event()
4367 /* Find current threshold */ in __mem_cgroup_usage_register_event()
4368 new->current_threshold = -1; in __mem_cgroup_usage_register_event()
4370 if (new->entries[i].threshold <= usage) { in __mem_cgroup_usage_register_event()
4372 * new->current_threshold will not be used until in __mem_cgroup_usage_register_event()
4376 ++new->current_threshold; in __mem_cgroup_usage_register_event()
4382 kfree(thresholds->spare); in __mem_cgroup_usage_register_event()
4383 thresholds->spare = thresholds->primary; in __mem_cgroup_usage_register_event()
4385 rcu_assign_pointer(thresholds->primary, new); in __mem_cgroup_usage_register_event()
4391 mutex_unlock(&memcg->thresholds_lock); in __mem_cgroup_usage_register_event()
4416 mutex_lock(&memcg->thresholds_lock); in __mem_cgroup_usage_unregister_event()
4419 thresholds = &memcg->thresholds; in __mem_cgroup_usage_unregister_event()
4422 thresholds = &memcg->memsw_thresholds; in __mem_cgroup_usage_unregister_event()
4427 if (!thresholds->primary) in __mem_cgroup_usage_unregister_event()
4435 for (i = 0; i < thresholds->primary->size; i++) { in __mem_cgroup_usage_unregister_event()
4436 if (thresholds->primary->entries[i].eventfd != eventfd) in __mem_cgroup_usage_unregister_event()
4442 new = thresholds->spare; in __mem_cgroup_usage_unregister_event()
4455 new->size = size; in __mem_cgroup_usage_unregister_event()
4457 /* Copy thresholds and find current threshold */ in __mem_cgroup_usage_unregister_event()
4458 new->current_threshold = -1; in __mem_cgroup_usage_unregister_event()
4459 for (i = 0, j = 0; i < thresholds->primary->size; i++) { in __mem_cgroup_usage_unregister_event()
4460 if (thresholds->primary->entries[i].eventfd == eventfd) in __mem_cgroup_usage_unregister_event()
4463 new->entries[j] = thresholds->primary->entries[i]; in __mem_cgroup_usage_unregister_event()
4464 if (new->entries[j].threshold <= usage) { in __mem_cgroup_usage_unregister_event()
4466 * new->current_threshold will not be used in __mem_cgroup_usage_unregister_event()
4470 ++new->current_threshold; in __mem_cgroup_usage_unregister_event()
4477 thresholds->spare = thresholds->primary; in __mem_cgroup_usage_unregister_event()
4479 rcu_assign_pointer(thresholds->primary, new); in __mem_cgroup_usage_unregister_event()
4486 kfree(thresholds->spare); in __mem_cgroup_usage_unregister_event()
4487 thresholds->spare = NULL; in __mem_cgroup_usage_unregister_event()
4490 mutex_unlock(&memcg->thresholds_lock); in __mem_cgroup_usage_unregister_event()
4512 return -ENOMEM; in mem_cgroup_oom_register_event()
4516 event->eventfd = eventfd; in mem_cgroup_oom_register_event()
4517 list_add(&event->list, &memcg->oom_notify); in mem_cgroup_oom_register_event()
4520 if (memcg->under_oom) in mem_cgroup_oom_register_event()
4534 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { in mem_cgroup_oom_unregister_event()
4535 if (ev->eventfd == eventfd) { in mem_cgroup_oom_unregister_event()
4536 list_del(&ev->list); in mem_cgroup_oom_unregister_event()
4548 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); in mem_cgroup_oom_control_read()
4549 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); in mem_cgroup_oom_control_read()
4551 atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); in mem_cgroup_oom_control_read()
4561 if (!css->parent || !((val == 0) || (val == 1))) in mem_cgroup_oom_control_write()
4562 return -EINVAL; in mem_cgroup_oom_control_write()
4564 memcg->oom_kill_disable = val; in mem_cgroup_oom_control_write()
4577 return wb_domain_init(&memcg->cgwb_domain, gfp); in memcg_wb_domain_init()
4582 wb_domain_exit(&memcg->cgwb_domain); in memcg_wb_domain_exit()
4587 wb_domain_size_changed(&memcg->cgwb_domain); in memcg_wb_domain_size_changed()
4592 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); in mem_cgroup_wb_domain()
4594 if (!memcg->css.parent) in mem_cgroup_wb_domain()
4597 return &memcg->cgwb_domain; in mem_cgroup_wb_domain()
4606 long x = atomic_long_read(&memcg->vmstats[idx]); in memcg_exact_page_state()
4610 x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; in memcg_exact_page_state()
4617 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
4625 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
4628 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
4638 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); in mem_cgroup_wb_stats()
4649 unsigned long ceiling = min(READ_ONCE(memcg->memory.max), in mem_cgroup_wb_stats()
4650 READ_ONCE(memcg->memory.high)); in mem_cgroup_wb_stats()
4651 unsigned long used = page_counter_read(&memcg->memory); in mem_cgroup_wb_stats()
4653 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); in mem_cgroup_wb_stats()
4662 * trackes ownership per-page while the latter per-inode. This was a
4663 * deliberate design decision because honoring per-page ownership in the
4665 * and deemed unnecessary given that write-sharing an inode across
4666 * different cgroups isn't a common use-case.
4668 * Combined with inode majority-writer ownership switching, this works well
4689 * page - a page whose memcg and writeback ownerships don't match - is
4695 * recorded bdi_writebacks and concurrent in-flight foreign writebacks are
4705 struct mem_cgroup *memcg = page->mem_cgroup; in mem_cgroup_track_foreign_dirty_slowpath()
4709 int oldest = -1; in mem_cgroup_track_foreign_dirty_slowpath()
4720 frn = &memcg->cgwb_frn[i]; in mem_cgroup_track_foreign_dirty_slowpath()
4721 if (frn->bdi_id == wb->bdi->id && in mem_cgroup_track_foreign_dirty_slowpath()
4722 frn->memcg_id == wb->memcg_css->id) in mem_cgroup_track_foreign_dirty_slowpath()
4724 if (time_before64(frn->at, oldest_at) && in mem_cgroup_track_foreign_dirty_slowpath()
4725 atomic_read(&frn->done.cnt) == 1) { in mem_cgroup_track_foreign_dirty_slowpath()
4727 oldest_at = frn->at; in mem_cgroup_track_foreign_dirty_slowpath()
4733 * Re-using an existing one. Update timestamp lazily to in mem_cgroup_track_foreign_dirty_slowpath()
4735 * reasonably up-to-date and significantly shorter than in mem_cgroup_track_foreign_dirty_slowpath()
4743 if (time_before64(frn->at, now - update_intv)) in mem_cgroup_track_foreign_dirty_slowpath()
4744 frn->at = now; in mem_cgroup_track_foreign_dirty_slowpath()
4747 frn = &memcg->cgwb_frn[oldest]; in mem_cgroup_track_foreign_dirty_slowpath()
4748 frn->bdi_id = wb->bdi->id; in mem_cgroup_track_foreign_dirty_slowpath()
4749 frn->memcg_id = wb->memcg_css->id; in mem_cgroup_track_foreign_dirty_slowpath()
4750 frn->at = now; in mem_cgroup_track_foreign_dirty_slowpath()
4757 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); in mem_cgroup_flush_foreign()
4763 struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; in mem_cgroup_flush_foreign()
4771 if (time_after64(frn->at, now - intv) && in mem_cgroup_flush_foreign()
4772 atomic_read(&frn->done.cnt) == 1) { in mem_cgroup_flush_foreign()
4773 frn->at = 0; in mem_cgroup_flush_foreign()
4774 trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); in mem_cgroup_flush_foreign()
4775 cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, in mem_cgroup_flush_foreign()
4777 &frn->done); in mem_cgroup_flush_foreign()
4804 * This is way over-engineered. It tries to support fully configurable
4821 struct mem_cgroup *memcg = event->memcg; in memcg_event_remove()
4823 remove_wait_queue(event->wqh, &event->wait); in memcg_event_remove()
4825 event->unregister_event(memcg, event->eventfd); in memcg_event_remove()
4828 eventfd_signal(event->eventfd, 1); in memcg_event_remove()
4830 eventfd_ctx_put(event->eventfd); in memcg_event_remove()
4832 css_put(&memcg->css); in memcg_event_remove()
4838 * Called with wqh->lock held and interrupts disabled.
4845 struct mem_cgroup *memcg = event->memcg; in memcg_event_wake()
4855 * side will require wqh->lock via remove_wait_queue(), in memcg_event_wake()
4858 spin_lock(&memcg->event_list_lock); in memcg_event_wake()
4859 if (!list_empty(&event->list)) { in memcg_event_wake()
4860 list_del_init(&event->list); in memcg_event_wake()
4865 schedule_work(&event->remove); in memcg_event_wake()
4867 spin_unlock(&memcg->event_list_lock); in memcg_event_wake()
4879 event->wqh = wqh; in memcg_event_ptable_queue_proc()
4880 add_wait_queue(wqh, &event->wait); in memcg_event_ptable_queue_proc()
4909 return -EINVAL; in memcg_write_event_control()
4914 return -EINVAL; in memcg_write_event_control()
4919 return -ENOMEM; in memcg_write_event_control()
4921 event->memcg = memcg; in memcg_write_event_control()
4922 INIT_LIST_HEAD(&event->list); in memcg_write_event_control()
4923 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); in memcg_write_event_control()
4924 init_waitqueue_func_entry(&event->wait, memcg_event_wake); in memcg_write_event_control()
4925 INIT_WORK(&event->remove, memcg_event_remove); in memcg_write_event_control()
4929 ret = -EBADF; in memcg_write_event_control()
4933 event->eventfd = eventfd_ctx_fileget(efile.file); in memcg_write_event_control()
4934 if (IS_ERR(event->eventfd)) { in memcg_write_event_control()
4935 ret = PTR_ERR(event->eventfd); in memcg_write_event_control()
4941 ret = -EBADF; in memcg_write_event_control()
4959 name = cfile.file->f_path.dentry->d_name.name; in memcg_write_event_control()
4962 event->register_event = mem_cgroup_usage_register_event; in memcg_write_event_control()
4963 event->unregister_event = mem_cgroup_usage_unregister_event; in memcg_write_event_control()
4965 event->register_event = mem_cgroup_oom_register_event; in memcg_write_event_control()
4966 event->unregister_event = mem_cgroup_oom_unregister_event; in memcg_write_event_control()
4968 event->register_event = vmpressure_register_event; in memcg_write_event_control()
4969 event->unregister_event = vmpressure_unregister_event; in memcg_write_event_control()
4971 event->register_event = memsw_cgroup_usage_register_event; in memcg_write_event_control()
4972 event->unregister_event = memsw_cgroup_usage_unregister_event; in memcg_write_event_control()
4974 ret = -EINVAL; in memcg_write_event_control()
4983 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, in memcg_write_event_control()
4985 ret = -EINVAL; in memcg_write_event_control()
4993 ret = event->register_event(memcg, event->eventfd, buf); in memcg_write_event_control()
4997 vfs_poll(efile.file, &event->pt); in memcg_write_event_control()
4999 spin_lock(&memcg->event_list_lock); in memcg_write_event_control()
5000 list_add(&event->list, &memcg->event_list); in memcg_write_event_control()
5001 spin_unlock(&memcg->event_list_lock); in memcg_write_event_control()
5013 eventfd_ctx_put(event->eventfd); in memcg_write_event_control()
5154 * Swap-out records and page cache shadow entries need to store memcg
5157 * memory-controlled cgroups to 64k.
5164 * even when there are much fewer than 64k cgroups - possibly none.
5166 * Maintain a private 16-bit ID space for memcg, and allow the ID to
5179 if (memcg->id.id > 0) { in mem_cgroup_id_remove()
5180 idr_remove(&mem_cgroup_idr, memcg->id.id); in mem_cgroup_id_remove()
5181 memcg->id.id = 0; in mem_cgroup_id_remove()
5188 refcount_add(n, &memcg->id.ref); in mem_cgroup_id_get_many()
5193 if (refcount_sub_and_test(n, &memcg->id.ref)) { in mem_cgroup_id_put_many()
5197 css_put(&memcg->css); in mem_cgroup_id_put_many()
5207 * mem_cgroup_from_id - look up a memcg from a memcg id
5216 if (id == -1) in mem_cgroup_from_id()
5235 tmp = -1; in alloc_mem_cgroup_per_node_info()
5240 pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, in alloc_mem_cgroup_per_node_info()
5242 if (!pn->lruvec_stat_local) { in alloc_mem_cgroup_per_node_info()
5247 pn->lruvec_stat_cpu = alloc_percpu_gfp(struct lruvec_stat, in alloc_mem_cgroup_per_node_info()
5249 if (!pn->lruvec_stat_cpu) { in alloc_mem_cgroup_per_node_info()
5250 free_percpu(pn->lruvec_stat_local); in alloc_mem_cgroup_per_node_info()
5255 lruvec_init(&pn->lruvec); in alloc_mem_cgroup_per_node_info()
5256 pn->usage_in_excess = 0; in alloc_mem_cgroup_per_node_info()
5257 pn->lruvec.pgdat = NODE_DATA(node); in alloc_mem_cgroup_per_node_info()
5258 pn->on_tree = false; in alloc_mem_cgroup_per_node_info()
5259 pn->memcg = memcg; in alloc_mem_cgroup_per_node_info()
5261 memcg->nodeinfo[node] = pn; in alloc_mem_cgroup_per_node_info()
5267 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; in free_mem_cgroup_per_node_info()
5272 free_percpu(pn->lruvec_stat_cpu); in free_mem_cgroup_per_node_info()
5273 free_percpu(pn->lruvec_stat_local); in free_mem_cgroup_per_node_info()
5283 free_percpu(memcg->vmstats_percpu); in __mem_cgroup_free()
5284 free_percpu(memcg->vmstats_local); in __mem_cgroup_free()
5306 long error = -ENOMEM; in mem_cgroup_alloc()
5315 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, in mem_cgroup_alloc()
5318 if (memcg->id.id < 0) { in mem_cgroup_alloc()
5319 error = memcg->id.id; in mem_cgroup_alloc()
5323 memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, in mem_cgroup_alloc()
5325 if (!memcg->vmstats_local) in mem_cgroup_alloc()
5328 memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, in mem_cgroup_alloc()
5330 if (!memcg->vmstats_percpu) in mem_cgroup_alloc()
5340 INIT_WORK(&memcg->high_work, high_work_func); in mem_cgroup_alloc()
5341 INIT_LIST_HEAD(&memcg->oom_notify); in mem_cgroup_alloc()
5342 mutex_init(&memcg->thresholds_lock); in mem_cgroup_alloc()
5343 spin_lock_init(&memcg->move_lock); in mem_cgroup_alloc()
5344 vmpressure_init(&memcg->vmpressure); in mem_cgroup_alloc()
5345 INIT_LIST_HEAD(&memcg->event_list); in mem_cgroup_alloc()
5346 spin_lock_init(&memcg->event_list_lock); in mem_cgroup_alloc()
5347 memcg->socket_pressure = jiffies; in mem_cgroup_alloc()
5349 memcg->kmemcg_id = -1; in mem_cgroup_alloc()
5350 INIT_LIST_HEAD(&memcg->objcg_list); in mem_cgroup_alloc()
5353 INIT_LIST_HEAD(&memcg->cgwb_list); in mem_cgroup_alloc()
5355 memcg->cgwb_frn[i].done = in mem_cgroup_alloc()
5359 spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); in mem_cgroup_alloc()
5360 INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); in mem_cgroup_alloc()
5361 memcg->deferred_split_queue.split_queue_len = 0; in mem_cgroup_alloc()
5372 INIT_LIST_HEAD(&memcg->score_node); in mem_cgroup_alloc()
5374 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); in mem_cgroup_alloc()
5387 long error = -ENOMEM; in mem_cgroup_css_alloc()
5396 atomic64_set(&memcg->memcg_reclaimed.app_score, 300); in mem_cgroup_css_alloc()
5399 atomic_set(&memcg->memcg_reclaimed.ub_zram2ufs_ratio, 10); in mem_cgroup_css_alloc()
5400 atomic_set(&memcg->memcg_reclaimed.ub_mem2zram_ratio, 60); in mem_cgroup_css_alloc()
5401 atomic_set(&memcg->memcg_reclaimed.refault_threshold, 50); in mem_cgroup_css_alloc()
5403 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); in mem_cgroup_css_alloc()
5404 memcg->soft_limit = PAGE_COUNTER_MAX; in mem_cgroup_css_alloc()
5405 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); in mem_cgroup_css_alloc()
5407 memcg->swappiness = mem_cgroup_swappiness(parent); in mem_cgroup_css_alloc()
5408 memcg->oom_kill_disable = parent->oom_kill_disable; in mem_cgroup_css_alloc()
5411 page_counter_init(&memcg->memory, NULL); in mem_cgroup_css_alloc()
5412 page_counter_init(&memcg->swap, NULL); in mem_cgroup_css_alloc()
5413 page_counter_init(&memcg->kmem, NULL); in mem_cgroup_css_alloc()
5414 page_counter_init(&memcg->tcpmem, NULL); in mem_cgroup_css_alloc()
5415 } else if (parent->use_hierarchy) { in mem_cgroup_css_alloc()
5416 memcg->use_hierarchy = true; in mem_cgroup_css_alloc()
5417 page_counter_init(&memcg->memory, &parent->memory); in mem_cgroup_css_alloc()
5418 page_counter_init(&memcg->swap, &parent->swap); in mem_cgroup_css_alloc()
5419 page_counter_init(&memcg->kmem, &parent->kmem); in mem_cgroup_css_alloc()
5420 page_counter_init(&memcg->tcpmem, &parent->tcpmem); in mem_cgroup_css_alloc()
5422 page_counter_init(&memcg->memory, &root_mem_cgroup->memory); in mem_cgroup_css_alloc()
5423 page_counter_init(&memcg->swap, &root_mem_cgroup->swap); in mem_cgroup_css_alloc()
5424 page_counter_init(&memcg->kmem, &root_mem_cgroup->kmem); in mem_cgroup_css_alloc()
5425 page_counter_init(&memcg->tcpmem, &root_mem_cgroup->tcpmem); in mem_cgroup_css_alloc()
5438 return &memcg->css; in mem_cgroup_css_alloc()
5448 return &memcg->css; in mem_cgroup_css_alloc()
5466 return -ENOMEM; in mem_cgroup_css_online()
5475 refcount_set(&memcg->id.ref, 1); in mem_cgroup_css_online()
5489 list_del_init(&memcg->score_node); in mem_cgroup_css_offline()
5499 spin_lock(&memcg->event_list_lock); in mem_cgroup_css_offline()
5500 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { in mem_cgroup_css_offline()
5501 list_del_init(&event->list); in mem_cgroup_css_offline()
5502 schedule_work(&event->remove); in mem_cgroup_css_offline()
5504 spin_unlock(&memcg->event_list_lock); in mem_cgroup_css_offline()
5506 page_counter_set_min(&memcg->memory, 0); in mem_cgroup_css_offline()
5507 page_counter_set_low(&memcg->memory, 0); in mem_cgroup_css_offline()
5531 wb_wait_for_completion(&memcg->cgwb_frn[i].done); in mem_cgroup_css_free()
5536 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) in mem_cgroup_css_free()
5539 vmpressure_cleanup(&memcg->vmpressure); in mem_cgroup_css_free()
5540 cancel_work_sync(&memcg->high_work); in mem_cgroup_css_free()
5548 * mem_cgroup_css_reset - reset the states of a mem_cgroup
5557 * The current implementation only resets the essential configurations.
5564 page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); in mem_cgroup_css_reset()
5565 page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); in mem_cgroup_css_reset()
5566 page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); in mem_cgroup_css_reset()
5567 page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); in mem_cgroup_css_reset()
5568 page_counter_set_min(&memcg->memory, 0); in mem_cgroup_css_reset()
5569 page_counter_set_low(&memcg->memory, 0); in mem_cgroup_css_reset()
5570 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); in mem_cgroup_css_reset()
5571 memcg->soft_limit = PAGE_COUNTER_MAX; in mem_cgroup_css_reset()
5572 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); in mem_cgroup_css_reset()
5577 /* Handlers for move charge at task migration. */
5582 /* Try a single bulk charge without reclaim first, kswapd may wake */ in mem_cgroup_do_precharge()
5590 while (count--) { in mem_cgroup_do_precharge()
5666 entry->val = ent.val; in mc_handle_swap_pte()
5681 if (!vma->vm_file) /* anonymous vma */ in mc_handle_file_pte()
5686 /* page is moved even if it's not RSS of this task(page-faulted). */ in mc_handle_file_pte()
5688 return find_get_incore_page(vma->vm_file->f_mapping, in mc_handle_file_pte()
5693 * mem_cgroup_move_account - move account of the page
5695 * @compound: charge the page as compound or small page
5701 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
5720 * page->mem_cgroup of its source page while we change it. in mem_cgroup_move_account()
5722 ret = -EBUSY; in mem_cgroup_move_account()
5726 ret = -EINVAL; in mem_cgroup_move_account()
5727 if (page->mem_cgroup != from) in mem_cgroup_move_account()
5738 __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); in mem_cgroup_move_account()
5747 __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); in mem_cgroup_move_account()
5751 __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); in mem_cgroup_move_account()
5756 __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); in mem_cgroup_move_account()
5761 struct address_space *mapping = page_mapping(page); in mem_cgroup_move_account() local
5763 if (mapping_can_writeback(mapping)) { in mem_cgroup_move_account()
5765 -nr_pages); in mem_cgroup_move_account()
5773 __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); in mem_cgroup_move_account()
5780 * It is safe to change page->mem_cgroup here because the page in mem_cgroup_move_account()
5783 * that would rely on a stable page->mem_cgroup. in mem_cgroup_move_account()
5786 * to save space. As soon as we switch page->mem_cgroup to a in mem_cgroup_move_account()
5792 css_get(&to->css); in mem_cgroup_move_account()
5793 css_put(&from->css); in mem_cgroup_move_account()
5795 page->mem_cgroup = to; in mem_cgroup_move_account()
5804 mem_cgroup_charge_statistics(from, page, -nr_pages); in mem_cgroup_move_account()
5814 * get_mctgt_type - get target type of moving charge
5821 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5823 * move charge. if @target is not NULL, the page is stored in target->page
5826 * target for charge migration. if @target is not NULL, the entry is stored
5827 * in target->ent.
5830 * For now we such page is charge like a regular page would be as for all
5861 if (page->mem_cgroup == mc.from) { in get_mctgt_type()
5866 target->page = page; in get_mctgt_type()
5873 * But we cannot move a tail-page in a THP. in get_mctgt_type()
5879 target->ent = ent; in get_mctgt_type()
5905 if (page->mem_cgroup == mc.from) { in get_mctgt_type_thp()
5909 target->page = page; in get_mctgt_type_thp()
5926 struct vm_area_struct *vma = walk->vma; in mem_cgroup_count_precharge_pte_range()
5945 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); in mem_cgroup_count_precharge_pte_range()
5949 pte_unmap_unlock(pte - 1, ptl); in mem_cgroup_count_precharge_pte_range()
5964 walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); in mem_cgroup_count_precharge()
5978 mc.moving_task = current; in mem_cgroup_precharge_mc()
6005 page_counter_uncharge(&mc.from->memsw, mc.moved_swap); in __mem_cgroup_clear_mc()
6010 * we charged both to->memory and to->memsw, so we in __mem_cgroup_clear_mc()
6011 * should uncharge to->memory. in __mem_cgroup_clear_mc()
6014 page_counter_uncharge(&mc.to->memory, mc.moved_swap); in __mem_cgroup_clear_mc()
6052 /* charge immigration isn't supported on the default hierarchy */ in mem_cgroup_can_attach()
6057 * Multi-process migrations only happen on the default hierarchy in mem_cgroup_can_attach()
6058 * where charge immigration is not used. Perform charge in mem_cgroup_can_attach()
6073 * tunable will only affect upcoming migrations, not the current one. in mem_cgroup_can_attach()
6076 move_flags = READ_ONCE(memcg->move_charge_at_immigrate); in mem_cgroup_can_attach()
6088 if (mm->owner == p) { in mem_cgroup_can_attach()
6123 struct vm_area_struct *vma = walk->vma; in mem_cgroup_move_charge_pte_range()
6142 mc.precharge -= HPAGE_PMD_NR; in mem_cgroup_move_charge_pte_range()
6152 mc.precharge -= HPAGE_PMD_NR; in mem_cgroup_move_charge_pte_range()
6164 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); in mem_cgroup_move_charge_pte_range()
6183 * memcg. There should be somebody mapping the head. in mem_cgroup_move_charge_pte_range()
6191 mc.precharge--; in mem_cgroup_move_charge_pte_range()
6203 mc.precharge--; in mem_cgroup_move_charge_pte_range()
6213 pte_unmap_unlock(pte - 1, ptl); in mem_cgroup_move_charge_pte_range()
6219 * We try charge one by one, but don't do any additional in mem_cgroup_move_charge_pte_range()
6220 * charges to mc.to if we have failed in charge once in attach() in mem_cgroup_move_charge_pte_range()
6241 * for already started RCU-only updates to finish. in mem_cgroup_move_charge()
6243 atomic_inc(&mc.from->moving_account); in mem_cgroup_move_charge()
6251 * to move enough charges, but moving charge is a best-effort in mem_cgroup_move_charge()
6260 * additional charge, the page walk just aborts. in mem_cgroup_move_charge()
6262 walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, in mem_cgroup_move_charge()
6266 atomic_dec(&mc.from->moving_account); in mem_cgroup_move_charge()
6302 root_mem_cgroup->use_hierarchy = true; in mem_cgroup_bind()
6304 root_mem_cgroup->use_hierarchy = false; in mem_cgroup_bind()
6322 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; in memory_current_read()
6328 READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); in memory_min_show()
6343 page_counter_set_min(&memcg->memory, min); in memory_min_write()
6351 READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); in memory_low_show()
6366 page_counter_set_low(&memcg->memory, low); in memory_low_write()
6374 READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); in memory_high_show()
6391 page_counter_set_high(&memcg->memory, high); in memory_high_write()
6394 unsigned long nr_pages = page_counter_read(&memcg->memory); in memory_high_write()
6400 if (signal_pending(current)) in memory_high_write()
6409 reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, in memory_high_write()
6412 if (!reclaimed && !nr_retries--) in memory_high_write()
6423 READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); in memory_max_show()
6440 xchg(&memcg->memory.max, max); in memory_max_write()
6443 unsigned long nr_pages = page_counter_read(&memcg->memory); in memory_max_write()
6448 if (signal_pending(current)) in memory_max_write()
6458 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, in memory_max_write()
6460 nr_reclaims--; in memory_max_write()
6487 __memory_events_show(m, memcg->memory_events); in memory_events_show()
6495 __memory_events_show(m, memcg->memory_events_local); in memory_events_local_show()
6506 return -ENOMEM; in memory_stat_show()
6545 seq_printf(m, "%d\n", memcg->oom_group); in memory_oom_group_show()
6558 return -EINVAL; in memory_oom_group_write()
6565 return -EINVAL; in memory_oom_group_write()
6567 memcg->oom_group = oom_group; in memory_oom_group_write()
6574 .name = "current",
6670 * This makes distribution proportional, but also work-conserving:
6681 * of the ancestor's claim to protection, any unutilized -
6682 * "floating" - protection from up the tree is distributed in
6708 * claimed protection in order to be work-conserving: claimed in effective_protection()
6746 * aren't read atomically - make sure the division is sane. in effective_protection()
6755 unclaimed = parent_effective - siblings_protected; in effective_protection()
6756 unclaimed *= usage - protected; in effective_protection()
6757 unclaimed /= parent_usage - siblings_protected; in effective_protection()
6766 * mem_cgroup_protected - check if memory consumption is in the normal range
6767 * @root: the top ancestor of the sub-tree being checked
6771 * of a top-down tree iteration, not for isolated queries.
6795 usage = page_counter_read(&memcg->memory); in mem_cgroup_calculate_protection()
6800 /* No parent means a non-hierarchical mode on v1 memcg */ in mem_cgroup_calculate_protection()
6805 memcg->memory.emin = READ_ONCE(memcg->memory.min); in mem_cgroup_calculate_protection()
6806 memcg->memory.elow = READ_ONCE(memcg->memory.low); in mem_cgroup_calculate_protection()
6810 parent_usage = page_counter_read(&parent->memory); in mem_cgroup_calculate_protection()
6812 WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, in mem_cgroup_calculate_protection()
6813 READ_ONCE(memcg->memory.min), in mem_cgroup_calculate_protection()
6814 READ_ONCE(parent->memory.emin), in mem_cgroup_calculate_protection()
6815 atomic_long_read(&parent->memory.children_min_usage))); in mem_cgroup_calculate_protection()
6817 WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, in mem_cgroup_calculate_protection()
6818 READ_ONCE(memcg->memory.low), in mem_cgroup_calculate_protection()
6819 READ_ONCE(parent->memory.elow), in mem_cgroup_calculate_protection()
6820 atomic_long_read(&parent->memory.children_low_usage))); in mem_cgroup_calculate_protection()
6824 * mem_cgroup_charge - charge a newly allocated page to a cgroup
6825 * @page: page to charge
6829 * Try to charge @page to the memcg that @mm belongs to, reclaiming
6848 * Every swap fault against a single page tries to charge the in mem_cgroup_charge()
6850 * already charged pages, too. page->mem_cgroup is protected in mem_cgroup_charge()
6855 if (compound_head(page)->mem_cgroup) in mem_cgroup_charge()
6861 if (memcg && !css_tryget_online(&memcg->css)) in mem_cgroup_charge()
6873 css_get(&memcg->css); in mem_cgroup_charge()
6889 * so this is a non-issue here. Memory and swap charge lifetimes in mem_cgroup_charge()
6890 * correspond 1:1 to page and swap slot lifetimes: we charge the in mem_cgroup_charge()
6898 * memory+swap charge, drop the swap entry duplicate. in mem_cgroup_charge()
6904 css_put(&memcg->css); in mem_cgroup_charge()
6926 if (!mem_cgroup_is_root(ug->memcg)) { in uncharge_batch()
6927 page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); in uncharge_batch()
6929 page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); in uncharge_batch()
6930 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) in uncharge_batch()
6931 page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); in uncharge_batch()
6932 memcg_oom_recover(ug->memcg); in uncharge_batch()
6936 __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); in uncharge_batch()
6937 __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); in uncharge_batch()
6938 memcg_check_events(ug->memcg, ug->dummy_page); in uncharge_batch()
6942 css_put(&ug->memcg->css); in uncharge_batch()
6951 if (!page->mem_cgroup) in uncharge_page()
6956 * page->mem_cgroup at this point, we have fully in uncharge_page()
6960 if (ug->memcg != page->mem_cgroup) { in uncharge_page()
6961 if (ug->memcg) { in uncharge_page()
6965 ug->memcg = page->mem_cgroup; in uncharge_page()
6968 css_get(&ug->memcg->css); in uncharge_page()
6972 ug->nr_pages += nr_pages; in uncharge_page()
6975 ug->pgpgout++; in uncharge_page()
6977 ug->nr_kmem += nr_pages; in uncharge_page()
6981 ug->dummy_page = page; in uncharge_page()
6982 page->mem_cgroup = NULL; in uncharge_page()
6983 css_put(&ug->memcg->css); in uncharge_page()
6994 * Note that the list can be a single page->lru; hence the in uncharge_list()
6995 * do-while loop instead of a simple list_for_each_entry(). in uncharge_list()
6997 next = page_list->next; in uncharge_list()
7002 next = page->lru.next; in uncharge_list()
7012 * mem_cgroup_uncharge - uncharge a page
7024 /* Don't touch page->lru of any random page, pre-check: */ in mem_cgroup_uncharge()
7025 if (!page->mem_cgroup) in mem_cgroup_uncharge()
7034 * mem_cgroup_uncharge_list - uncharge a list of page
7050 * mem_cgroup_migrate - charge a page's replacement
7054 * Charge @newpage as a replacement page for @oldpage. @oldpage will
7057 * Both pages must be locked, @newpage->mapping must be set up.
7075 if (newpage->mem_cgroup) in mem_cgroup_migrate()
7079 memcg = oldpage->mem_cgroup; in mem_cgroup_migrate()
7083 /* Force-charge the new page. The old one will be freed soon */ in mem_cgroup_migrate()
7086 page_counter_charge(&memcg->memory, nr_pages); in mem_cgroup_migrate()
7088 page_counter_charge(&memcg->memsw, nr_pages); in mem_cgroup_migrate()
7090 css_get(&memcg->css); in mem_cgroup_migrate()
7114 memcg = mem_cgroup_from_task(current); in mem_cgroup_sk_alloc()
7117 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) in mem_cgroup_sk_alloc()
7119 if (css_tryget(&memcg->css)) in mem_cgroup_sk_alloc()
7120 sk->sk_memcg = memcg; in mem_cgroup_sk_alloc()
7127 if (sk->sk_memcg) in mem_cgroup_sk_free()
7128 css_put(&sk->sk_memcg->css); in mem_cgroup_sk_free()
7132 * mem_cgroup_charge_skmem - charge socket memory
7133 * @memcg: memcg to charge
7134 * @nr_pages: number of pages to charge
7136 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
7137 * @memcg's configured limit, %false if the charge had to be forced.
7146 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { in mem_cgroup_charge_skmem()
7147 memcg->tcpmem_pressure = 0; in mem_cgroup_charge_skmem()
7150 page_counter_charge(&memcg->tcpmem, nr_pages); in mem_cgroup_charge_skmem()
7151 memcg->tcpmem_pressure = 1; in mem_cgroup_charge_skmem()
7169 * mem_cgroup_uncharge_skmem - uncharge socket memory
7176 page_counter_uncharge(&memcg->tcpmem, nr_pages); in mem_cgroup_uncharge_skmem()
7180 mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); in mem_cgroup_uncharge_skmem()
7207 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
7219 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, in mem_cgroup_init()
7228 rtpn->rb_root = RB_ROOT; in mem_cgroup_init()
7229 rtpn->rb_rightmost = NULL; in mem_cgroup_init()
7230 spin_lock_init(&rtpn->lock); in mem_cgroup_init()
7241 while (!refcount_inc_not_zero(&memcg->id.ref)) { in mem_cgroup_id_get_online()
7258 * mem_cgroup_swapout - transfer a memsw charge to swap
7259 * @page: page whose memsw charge to transfer
7260 * @entry: swap entry to move the charge to
7262 * Transfer the memsw charge of @page to @entry.
7276 memcg = page->mem_cgroup; in mem_cgroup_swapout()
7284 * have an ID allocated to it anymore, charge the closest online in mem_cgroup_swapout()
7285 * ancestor for the swap instead and transfer the memory+swap charge. in mem_cgroup_swapout()
7291 mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); in mem_cgroup_swapout()
7297 page->mem_cgroup = NULL; in mem_cgroup_swapout()
7300 page_counter_uncharge(&memcg->memory, nr_entries); in mem_cgroup_swapout()
7304 page_counter_charge(&swap_memcg->memsw, nr_entries); in mem_cgroup_swapout()
7305 page_counter_uncharge(&memcg->memsw, nr_entries); in mem_cgroup_swapout()
7310 * i_pages lock which is taken with interrupts-off. It is in mem_cgroup_swapout()
7312 * only synchronisation we have for updating the per-CPU variables. in mem_cgroup_swapout()
7315 mem_cgroup_charge_statistics(memcg, page, -nr_entries); in mem_cgroup_swapout()
7318 css_put(&memcg->css); in mem_cgroup_swapout()
7322 * mem_cgroup_try_charge_swap - try charging swap space for a page
7324 * @entry: swap entry to charge
7326 * Try to charge @page's memcg for the swap space at @entry.
7328 * Returns 0 on success, -ENOMEM on failure.
7340 memcg = page->mem_cgroup; in mem_cgroup_try_charge_swap()
7354 !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { in mem_cgroup_try_charge_swap()
7358 return -ENOMEM; in mem_cgroup_try_charge_swap()
7363 mem_cgroup_id_get_many(memcg, nr_pages - 1); in mem_cgroup_try_charge_swap()
7372 * mem_cgroup_uncharge_swap - uncharge swap space
7387 page_counter_uncharge(&memcg->swap, nr_pages); in mem_cgroup_uncharge_swap()
7389 page_counter_uncharge(&memcg->memsw, nr_pages); in mem_cgroup_uncharge_swap()
7391 mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); in mem_cgroup_uncharge_swap()
7405 READ_ONCE(memcg->swap.max) - in mem_cgroup_get_nr_swap_pages()
7406 page_counter_read(&memcg->swap)); in mem_cgroup_get_nr_swap_pages()
7421 memcg = page->mem_cgroup; in mem_cgroup_swap_full()
7426 unsigned long usage = page_counter_read(&memcg->swap); in mem_cgroup_swap_full()
7428 if (usage * 2 >= READ_ONCE(memcg->swap.high) || in mem_cgroup_swap_full()
7429 usage * 2 >= READ_ONCE(memcg->swap.max)) in mem_cgroup_swap_full()
7451 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; in swap_current_read()
7457 READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); in swap_high_show()
7472 page_counter_set_high(&memcg->swap, high); in swap_high_write()
7480 READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); in swap_max_show()
7495 xchg(&memcg->swap.max, max); in swap_max_write()
7505 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); in swap_events_show()
7507 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); in swap_events_show()
7509 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); in swap_events_show()
7516 .name = "swap.current",
7577 /* No memory control -> no swap control */ in mem_cgroup_swap_init()