| Message ID | 1559887659-23121-3-git-send-email-yang.shi@linux.alibaba.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Make deferred split shrinker memcg aware | expand |
Hi, Yang, On 07.06.2019 09:07, Yang Shi wrote: > Currently THP deferred split shrinker is not memcg aware, this may cause > premature OOM with some configuration. For example the below test would > run into premature OOM easily: > > $ cgcreate -g memory:thp > $ echo 4G > /sys/fs/cgroup/memory/thp/memory/limit_in_bytes > $ cgexec -g memory:thp transhuge-stress 4000 > > transhuge-stress comes from kernel selftest. > > It is easy to hit OOM, but there are still a lot THP on the deferred > split queue, memcg direct reclaim can't touch them since the deferred > split shrinker is not memcg aware. > > Convert deferred split shrinker memcg aware by introducing per memcg > deferred split queue. The THP should be on either per node or per memcg > deferred split queue if it belongs to a memcg. When the page is > immigrated to the other memcg, it will be immigrated to the target > memcg's deferred split queue too. > > And, move deleting THP from deferred split queue in page free before > memcg uncharge so that the page's memcg information is available. > > Reuse the second tail page's deferred_list for per memcg list since the > same THP can't be on multiple deferred split queues. > > Cc: Kirill Tkhai <ktkhai@virtuozzo.com> > Cc: Johannes Weiner <hannes@cmpxchg.org> > Cc: Michal Hocko <mhocko@suse.com> > Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com> > Cc: Hugh Dickins <hughd@google.com> > Cc: Shakeel Butt <shakeelb@google.com> > Cc: David Rientjes <rientjes@google.com> > Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> > --- > include/linux/huge_mm.h | 15 ++++++++++ > include/linux/memcontrol.h | 4 +++ > include/linux/mm_types.h | 1 + > mm/huge_memory.c | 71 +++++++++++++++++++++++++++++++++------------- > mm/memcontrol.c | 19 +++++++++++++ > mm/swap.c | 4 +++ > 6 files changed, 94 insertions(+), 20 deletions(-) > > diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h > index 7cd5c15..8137c3a 100644 > --- a/include/linux/huge_mm.h > +++ b/include/linux/huge_mm.h > @@ -250,6 +250,17 @@ static inline bool thp_migration_supported(void) > return IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION); > } > > +static inline struct list_head *page_deferred_list(struct page *page) > +{ > + /* > + * Global or memcg deferred list in the second tail pages is > + * occupied by compound_head. > + */ > + return &page[2].deferred_list; > +} > + > +extern void del_thp_from_deferred_split_queue(struct page *); > + > #else /* CONFIG_TRANSPARENT_HUGEPAGE */ > #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; }) > #define HPAGE_PMD_MASK ({ BUILD_BUG(); 0; }) > @@ -368,6 +379,10 @@ static inline bool thp_migration_supported(void) > { > return false; > } > + > +static inline void del_thp_from_deferred_split_queue(struct page *page) > +{ > +} > #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ > > #endif /* _LINUX_HUGE_MM_H */ > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h > index bc74d6a..5d3c10c 100644 > --- a/include/linux/memcontrol.h > +++ b/include/linux/memcontrol.h > @@ -316,6 +316,10 @@ struct mem_cgroup { > struct list_head event_list; > spinlock_t event_list_lock; > > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE > + struct deferred_split deferred_split_queue; > +#endif > + > struct mem_cgroup_per_node *nodeinfo[0]; > /* WARNING: nodeinfo must be the last member here */ > }; > diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h > index 8ec38b1..4eabf80 100644 > --- a/include/linux/mm_types.h > +++ b/include/linux/mm_types.h > @@ -139,6 +139,7 @@ struct page { > struct { /* Second tail page of compound page */ > unsigned long _compound_pad_1; /* compound_head */ > unsigned long _compound_pad_2; > + /* For both global and memcg */ > struct list_head deferred_list; > }; > struct { /* Page table pages */ > diff --git a/mm/huge_memory.c b/mm/huge_memory.c > index 81cf759..3307697 100644 > --- a/mm/huge_memory.c > +++ b/mm/huge_memory.c > @@ -492,10 +492,15 @@ pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma) > return pmd; > } > > -static inline struct list_head *page_deferred_list(struct page *page) > +static inline struct deferred_split *get_deferred_split_queue(struct page *page) > { > - /* ->lru in the tail pages is occupied by compound_head. */ > - return &page[2].deferred_list; > + struct mem_cgroup *memcg = compound_head(page)->mem_cgroup; > + struct pglist_data *pgdat = NODE_DATA(page_to_nid(page)); > + > + if (memcg) > + return &memcg->deferred_split_queue; > + else > + return &pgdat->deferred_split_queue; memory_cgrp_subsys is not early initialized, so at the beginning of boot root_mem_cgroup is NULL, and pages will use &pgdat->deferred_split_queue list head. But after root_mem_cgroup is initialized, another list head will be used, won't it?! So there will be two different list heads used for same cgroup. This may be a reason of a problem (I won't say you, where the problem will occur). > } > > void prep_transhuge_page(struct page *page) > @@ -2658,7 +2663,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) > { > struct page *head = compound_head(page); > struct pglist_data *pgdata = NODE_DATA(page_to_nid(head)); > - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; > + struct deferred_split *ds_queue = get_deferred_split_queue(page); > struct anon_vma *anon_vma = NULL; > struct address_space *mapping = NULL; > int count, mapcount, extra_pins, ret; > @@ -2792,25 +2797,36 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) > return ret; > } > > -void free_transhuge_page(struct page *page) > +void del_thp_from_deferred_split_queue(struct page *page) > { > - struct pglist_data *pgdata = NODE_DATA(page_to_nid(page)); > - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; > - unsigned long flags; > - > - spin_lock_irqsave(&ds_queue->split_queue_lock, flags); > - if (!list_empty(page_deferred_list(page))) { > - ds_queue->split_queue_len--; > - list_del(page_deferred_list(page)); > + /* > + * The THP may be not on LRU at this point, e.g. the old page of > + * NUMA migration. And PageTransHuge is not enough to distinguish > + * with other compound page, e.g. skb, THP destructor is not used > + * anymore and will be removed, so the compound order sounds like > + * the only choice here. > + */ > + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { > + struct deferred_split *ds_queue = get_deferred_split_queue(page); > + unsigned long flags; > + spin_lock_irqsave(&ds_queue->split_queue_lock, flags); > + if (!list_empty(page_deferred_list(page))) { > + ds_queue->split_queue_len--; > + list_del(page_deferred_list(page)); > + } > + spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); > } > - spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); > +} > + > +void free_transhuge_page(struct page *page) > +{ > free_compound_page(page); > } > > void deferred_split_huge_page(struct page *page) > { > - struct pglist_data *pgdata = NODE_DATA(page_to_nid(page)); > - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; > + struct deferred_split *ds_queue = get_deferred_split_queue(page); > + struct mem_cgroup *memcg = compound_head(page)->mem_cgroup; > unsigned long flags; > > VM_BUG_ON_PAGE(!PageTransHuge(page), page); > @@ -2820,6 +2836,9 @@ void deferred_split_huge_page(struct page *page) > count_vm_event(THP_DEFERRED_SPLIT_PAGE); > list_add_tail(page_deferred_list(page), &ds_queue->split_queue); > ds_queue->split_queue_len++; > + if (memcg) > + memcg_set_shrinker_bit(memcg, page_to_nid(page), > + deferred_split_shrinker.id); > } > spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); > } > @@ -2827,8 +2846,15 @@ void deferred_split_huge_page(struct page *page) > static unsigned long deferred_split_count(struct shrinker *shrink, > struct shrink_control *sc) > { > - struct pglist_data *pgdata = NODE_DATA(sc->nid); > - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; > + struct deferred_split *ds_queue; > + > + if (!sc->memcg) { > + struct pglist_data *pgdata = NODE_DATA(sc->nid); > + ds_queue = &pgdata->deferred_split_queue; > + return READ_ONCE(ds_queue->split_queue_len); > + } > + > + ds_queue = &sc->memcg->deferred_split_queue; > return READ_ONCE(ds_queue->split_queue_len); > } > > @@ -2836,12 +2862,17 @@ static unsigned long deferred_split_scan(struct shrinker *shrink, > struct shrink_control *sc) > { > struct pglist_data *pgdata = NODE_DATA(sc->nid); > - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; > + struct deferred_split *ds_queue; > unsigned long flags; > LIST_HEAD(list), *pos, *next; > struct page *page; > int split = 0; > > + if (sc->memcg) > + ds_queue = &sc->memcg->deferred_split_queue; > + else > + ds_queue = &pgdata->deferred_split_queue; > + > spin_lock_irqsave(&ds_queue->split_queue_lock, flags); > /* Take pin on all head pages to avoid freeing them under us */ > list_for_each_safe(pos, next, &ds_queue->split_queue) { > @@ -2888,7 +2919,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink, > .count_objects = deferred_split_count, > .scan_objects = deferred_split_scan, > .seeks = DEFAULT_SEEKS, > - .flags = SHRINKER_NUMA_AWARE, > + .flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE, > }; > > #ifdef CONFIG_DEBUG_FS > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > index e50a2db..fe7e544 100644 > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -4579,6 +4579,11 @@ static struct mem_cgroup *mem_cgroup_alloc(void) > #ifdef CONFIG_CGROUP_WRITEBACK > INIT_LIST_HEAD(&memcg->cgwb_list); > #endif > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE > + spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); > + INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); > + memcg->deferred_split_queue.split_queue_len = 0; > +#endif > idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); > return memcg; > fail: > @@ -4949,6 +4954,20 @@ static int mem_cgroup_move_account(struct page *page, > __mod_memcg_state(to, NR_WRITEBACK, nr_pages); > } > > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE > + if (compound && !list_empty(page_deferred_list(page))) { > + spin_lock(&from->deferred_split_queue.split_queue_lock); > + list_del(page_deferred_list(page)); > + from->deferred_split_queue.split_queue_len--; > + spin_unlock(&from->deferred_split_queue.split_queue_lock); Won't be better to assign page->mem_cgroup = to; after removing from one list and before linking to another list? There is possible no a problem, but another people writing code on top of this may not expect such the behavior. > + spin_lock(&to->deferred_split_queue.split_queue_lock); > + list_add_tail(page_deferred_list(page), > + &to->deferred_split_queue.split_queue); > + to->deferred_split_queue.split_queue_len++; > + spin_unlock(&to->deferred_split_queue.split_queue_lock); > + } > +#endif > /* > * It is safe to change page->mem_cgroup here because the page > * is referenced, charged, and isolated - we can't race with > diff --git a/mm/swap.c b/mm/swap.c > index 3a75722..3348295 100644 > --- a/mm/swap.c > +++ b/mm/swap.c > @@ -69,6 +69,10 @@ static void __page_cache_release(struct page *page) > del_page_from_lru_list(page, lruvec, page_off_lru(page)); > spin_unlock_irqrestore(&pgdat->lru_lock, flags); > } > + > + /* Delete THP from deferred split queue before memcg uncharge */ > + del_thp_from_deferred_split_queue(page); > + > __ClearPageWaiters(page); > mem_cgroup_uncharge(page); > }
On 6/10/19 1:21 AM, Kirill Tkhai wrote: > Hi, Yang, > > On 07.06.2019 09:07, Yang Shi wrote: >> Currently THP deferred split shrinker is not memcg aware, this may cause >> premature OOM with some configuration. For example the below test would >> run into premature OOM easily: >> >> $ cgcreate -g memory:thp >> $ echo 4G > /sys/fs/cgroup/memory/thp/memory/limit_in_bytes >> $ cgexec -g memory:thp transhuge-stress 4000 >> >> transhuge-stress comes from kernel selftest. >> >> It is easy to hit OOM, but there are still a lot THP on the deferred >> split queue, memcg direct reclaim can't touch them since the deferred >> split shrinker is not memcg aware. >> >> Convert deferred split shrinker memcg aware by introducing per memcg >> deferred split queue. The THP should be on either per node or per memcg >> deferred split queue if it belongs to a memcg. When the page is >> immigrated to the other memcg, it will be immigrated to the target >> memcg's deferred split queue too. >> >> And, move deleting THP from deferred split queue in page free before >> memcg uncharge so that the page's memcg information is available. >> >> Reuse the second tail page's deferred_list for per memcg list since the >> same THP can't be on multiple deferred split queues. >> >> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> >> Cc: Johannes Weiner <hannes@cmpxchg.org> >> Cc: Michal Hocko <mhocko@suse.com> >> Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com> >> Cc: Hugh Dickins <hughd@google.com> >> Cc: Shakeel Butt <shakeelb@google.com> >> Cc: David Rientjes <rientjes@google.com> >> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> >> --- >> include/linux/huge_mm.h | 15 ++++++++++ >> include/linux/memcontrol.h | 4 +++ >> include/linux/mm_types.h | 1 + >> mm/huge_memory.c | 71 +++++++++++++++++++++++++++++++++------------- >> mm/memcontrol.c | 19 +++++++++++++ >> mm/swap.c | 4 +++ >> 6 files changed, 94 insertions(+), 20 deletions(-) >> >> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h >> index 7cd5c15..8137c3a 100644 >> --- a/include/linux/huge_mm.h >> +++ b/include/linux/huge_mm.h >> @@ -250,6 +250,17 @@ static inline bool thp_migration_supported(void) >> return IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION); >> } >> >> +static inline struct list_head *page_deferred_list(struct page *page) >> +{ >> + /* >> + * Global or memcg deferred list in the second tail pages is >> + * occupied by compound_head. >> + */ >> + return &page[2].deferred_list; >> +} >> + >> +extern void del_thp_from_deferred_split_queue(struct page *); >> + >> #else /* CONFIG_TRANSPARENT_HUGEPAGE */ >> #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; }) >> #define HPAGE_PMD_MASK ({ BUILD_BUG(); 0; }) >> @@ -368,6 +379,10 @@ static inline bool thp_migration_supported(void) >> { >> return false; >> } >> + >> +static inline void del_thp_from_deferred_split_queue(struct page *page) >> +{ >> +} >> #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ >> >> #endif /* _LINUX_HUGE_MM_H */ >> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h >> index bc74d6a..5d3c10c 100644 >> --- a/include/linux/memcontrol.h >> +++ b/include/linux/memcontrol.h >> @@ -316,6 +316,10 @@ struct mem_cgroup { >> struct list_head event_list; >> spinlock_t event_list_lock; >> >> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE >> + struct deferred_split deferred_split_queue; >> +#endif >> + >> struct mem_cgroup_per_node *nodeinfo[0]; >> /* WARNING: nodeinfo must be the last member here */ >> }; >> diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h >> index 8ec38b1..4eabf80 100644 >> --- a/include/linux/mm_types.h >> +++ b/include/linux/mm_types.h >> @@ -139,6 +139,7 @@ struct page { >> struct { /* Second tail page of compound page */ >> unsigned long _compound_pad_1; /* compound_head */ >> unsigned long _compound_pad_2; >> + /* For both global and memcg */ >> struct list_head deferred_list; >> }; >> struct { /* Page table pages */ >> diff --git a/mm/huge_memory.c b/mm/huge_memory.c >> index 81cf759..3307697 100644 >> --- a/mm/huge_memory.c >> +++ b/mm/huge_memory.c >> @@ -492,10 +492,15 @@ pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma) >> return pmd; >> } >> >> -static inline struct list_head *page_deferred_list(struct page *page) >> +static inline struct deferred_split *get_deferred_split_queue(struct page *page) >> { >> - /* ->lru in the tail pages is occupied by compound_head. */ >> - return &page[2].deferred_list; >> + struct mem_cgroup *memcg = compound_head(page)->mem_cgroup; >> + struct pglist_data *pgdat = NODE_DATA(page_to_nid(page)); >> + >> + if (memcg) >> + return &memcg->deferred_split_queue; >> + else >> + return &pgdat->deferred_split_queue; > memory_cgrp_subsys is not early initialized, so at the beginning of boot > root_mem_cgroup is NULL, and pages will use &pgdat->deferred_split_queue > list head. But after root_mem_cgroup is initialized, another list head > will be used, won't it?! So there will be two different list heads used > for same cgroup. > > This may be a reason of a problem (I won't say you, where the problem will > occur). Actually I was wondering this too. But, it looks impossible. The memcg is initialized by subsys_initcall, there are still fs, device and late initcalls after it. The init process is run much later, so I can't think of any THP could be put on deferred split queue before memcg is initialized. There may be compound page allocation, i.e. skb, but we don't care them. > >> } >> >> void prep_transhuge_page(struct page *page) >> @@ -2658,7 +2663,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) >> { >> struct page *head = compound_head(page); >> struct pglist_data *pgdata = NODE_DATA(page_to_nid(head)); >> - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; >> + struct deferred_split *ds_queue = get_deferred_split_queue(page); >> struct anon_vma *anon_vma = NULL; >> struct address_space *mapping = NULL; >> int count, mapcount, extra_pins, ret; >> @@ -2792,25 +2797,36 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) >> return ret; >> } >> >> -void free_transhuge_page(struct page *page) >> +void del_thp_from_deferred_split_queue(struct page *page) >> { >> - struct pglist_data *pgdata = NODE_DATA(page_to_nid(page)); >> - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; >> - unsigned long flags; >> - >> - spin_lock_irqsave(&ds_queue->split_queue_lock, flags); >> - if (!list_empty(page_deferred_list(page))) { >> - ds_queue->split_queue_len--; >> - list_del(page_deferred_list(page)); >> + /* >> + * The THP may be not on LRU at this point, e.g. the old page of >> + * NUMA migration. And PageTransHuge is not enough to distinguish >> + * with other compound page, e.g. skb, THP destructor is not used >> + * anymore and will be removed, so the compound order sounds like >> + * the only choice here. >> + */ >> + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { >> + struct deferred_split *ds_queue = get_deferred_split_queue(page); >> + unsigned long flags; >> + spin_lock_irqsave(&ds_queue->split_queue_lock, flags); >> + if (!list_empty(page_deferred_list(page))) { >> + ds_queue->split_queue_len--; >> + list_del(page_deferred_list(page)); >> + } >> + spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); >> } >> - spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); >> +} >> + >> +void free_transhuge_page(struct page *page) >> +{ >> free_compound_page(page); >> } >> >> void deferred_split_huge_page(struct page *page) >> { >> - struct pglist_data *pgdata = NODE_DATA(page_to_nid(page)); >> - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; >> + struct deferred_split *ds_queue = get_deferred_split_queue(page); >> + struct mem_cgroup *memcg = compound_head(page)->mem_cgroup; >> unsigned long flags; >> >> VM_BUG_ON_PAGE(!PageTransHuge(page), page); >> @@ -2820,6 +2836,9 @@ void deferred_split_huge_page(struct page *page) >> count_vm_event(THP_DEFERRED_SPLIT_PAGE); >> list_add_tail(page_deferred_list(page), &ds_queue->split_queue); >> ds_queue->split_queue_len++; >> + if (memcg) >> + memcg_set_shrinker_bit(memcg, page_to_nid(page), >> + deferred_split_shrinker.id); >> } >> spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); >> } >> @@ -2827,8 +2846,15 @@ void deferred_split_huge_page(struct page *page) >> static unsigned long deferred_split_count(struct shrinker *shrink, >> struct shrink_control *sc) >> { >> - struct pglist_data *pgdata = NODE_DATA(sc->nid); >> - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; >> + struct deferred_split *ds_queue; >> + >> + if (!sc->memcg) { >> + struct pglist_data *pgdata = NODE_DATA(sc->nid); >> + ds_queue = &pgdata->deferred_split_queue; >> + return READ_ONCE(ds_queue->split_queue_len); >> + } >> + >> + ds_queue = &sc->memcg->deferred_split_queue; >> return READ_ONCE(ds_queue->split_queue_len); >> } >> >> @@ -2836,12 +2862,17 @@ static unsigned long deferred_split_scan(struct shrinker *shrink, >> struct shrink_control *sc) >> { >> struct pglist_data *pgdata = NODE_DATA(sc->nid); >> - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; >> + struct deferred_split *ds_queue; >> unsigned long flags; >> LIST_HEAD(list), *pos, *next; >> struct page *page; >> int split = 0; >> >> + if (sc->memcg) >> + ds_queue = &sc->memcg->deferred_split_queue; >> + else >> + ds_queue = &pgdata->deferred_split_queue; >> + >> spin_lock_irqsave(&ds_queue->split_queue_lock, flags); >> /* Take pin on all head pages to avoid freeing them under us */ >> list_for_each_safe(pos, next, &ds_queue->split_queue) { >> @@ -2888,7 +2919,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink, >> .count_objects = deferred_split_count, >> .scan_objects = deferred_split_scan, >> .seeks = DEFAULT_SEEKS, >> - .flags = SHRINKER_NUMA_AWARE, >> + .flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE, >> }; >> >> #ifdef CONFIG_DEBUG_FS >> diff --git a/mm/memcontrol.c b/mm/memcontrol.c >> index e50a2db..fe7e544 100644 >> --- a/mm/memcontrol.c >> +++ b/mm/memcontrol.c >> @@ -4579,6 +4579,11 @@ static struct mem_cgroup *mem_cgroup_alloc(void) >> #ifdef CONFIG_CGROUP_WRITEBACK >> INIT_LIST_HEAD(&memcg->cgwb_list); >> #endif >> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE >> + spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); >> + INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); >> + memcg->deferred_split_queue.split_queue_len = 0; >> +#endif >> idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); >> return memcg; >> fail: >> @@ -4949,6 +4954,20 @@ static int mem_cgroup_move_account(struct page *page, >> __mod_memcg_state(to, NR_WRITEBACK, nr_pages); >> } >> >> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE >> + if (compound && !list_empty(page_deferred_list(page))) { >> + spin_lock(&from->deferred_split_queue.split_queue_lock); >> + list_del(page_deferred_list(page)); >> + from->deferred_split_queue.split_queue_len--; >> + spin_unlock(&from->deferred_split_queue.split_queue_lock); > Won't be better to assign > > page->mem_cgroup = to; > > after removing from one list and before linking to another list? > There is possible no a problem, but another people writing code > on top of this may not expect such the behavior. It should be fine too. >> + spin_lock(&to->deferred_split_queue.split_queue_lock); >> + list_add_tail(page_deferred_list(page), >> + &to->deferred_split_queue.split_queue); >> + to->deferred_split_queue.split_queue_len++; >> + spin_unlock(&to->deferred_split_queue.split_queue_lock); >> + } >> +#endif >> /* >> * It is safe to change page->mem_cgroup here because the page >> * is referenced, charged, and isolated - we can't race with >> diff --git a/mm/swap.c b/mm/swap.c >> index 3a75722..3348295 100644 >> --- a/mm/swap.c >> +++ b/mm/swap.c >> @@ -69,6 +69,10 @@ static void __page_cache_release(struct page *page) >> del_page_from_lru_list(page, lruvec, page_off_lru(page)); >> spin_unlock_irqrestore(&pgdat->lru_lock, flags); >> } >> + >> + /* Delete THP from deferred split queue before memcg uncharge */ >> + del_thp_from_deferred_split_queue(page); >> + >> __ClearPageWaiters(page); >> mem_cgroup_uncharge(page); >> }
On Fri, Jun 07, 2019 at 02:07:37PM +0800, Yang Shi wrote: > + /* > + * The THP may be not on LRU at this point, e.g. the old page of > + * NUMA migration. And PageTransHuge is not enough to distinguish > + * with other compound page, e.g. skb, THP destructor is not used > + * anymore and will be removed, so the compound order sounds like > + * the only choice here. > + */ > + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { What happens if the page is the same order as THP is not THP? Why removing of destructor is required?
On 6/11/19 7:47 PM, Kirill A. Shutemov wrote: > On Fri, Jun 07, 2019 at 02:07:37PM +0800, Yang Shi wrote: >> + /* >> + * The THP may be not on LRU at this point, e.g. the old page of >> + * NUMA migration. And PageTransHuge is not enough to distinguish >> + * with other compound page, e.g. skb, THP destructor is not used >> + * anymore and will be removed, so the compound order sounds like >> + * the only choice here. >> + */ >> + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { > What happens if the page is the same order as THP is not THP? Why removing It may corrupt the deferred split queue since it is never added into the list, but deleted here. > of destructor is required? Due to the change to free_transhuge_page() (extracted deferred split queue manipulation and moved before memcg uncharge since page->mem_cgroup is needed), it just calls free_compound_page(). So, it sounds pointless to still keep THP specific destructor. It looks there is not a good way to tell if the compound page is THP in free_page path or not, we may keep the destructor just for this? >
On Tue, Jun 11, 2019 at 10:06:36PM -0700, Yang Shi wrote: > > > On 6/11/19 7:47 PM, Kirill A. Shutemov wrote: > > On Fri, Jun 07, 2019 at 02:07:37PM +0800, Yang Shi wrote: > > > + /* > > > + * The THP may be not on LRU at this point, e.g. the old page of > > > + * NUMA migration. And PageTransHuge is not enough to distinguish > > > + * with other compound page, e.g. skb, THP destructor is not used > > > + * anymore and will be removed, so the compound order sounds like > > > + * the only choice here. > > > + */ > > > + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { > > What happens if the page is the same order as THP is not THP? Why removing > > It may corrupt the deferred split queue since it is never added into the > list, but deleted here. > > > of destructor is required? > > Due to the change to free_transhuge_page() (extracted deferred split queue > manipulation and moved before memcg uncharge since page->mem_cgroup is > needed), it just calls free_compound_page(). So, it sounds pointless to > still keep THP specific destructor. > > It looks there is not a good way to tell if the compound page is THP in > free_page path or not, we may keep the destructor just for this? Other option would be to move mem_cgroup_uncharge(page); from __page_cache_release() to destructors. Destructors will be able to call it as it fits.
On 6/12/19 3:09 AM, Kirill A. Shutemov wrote: > On Tue, Jun 11, 2019 at 10:06:36PM -0700, Yang Shi wrote: >> >> On 6/11/19 7:47 PM, Kirill A. Shutemov wrote: >>> On Fri, Jun 07, 2019 at 02:07:37PM +0800, Yang Shi wrote: >>>> + /* >>>> + * The THP may be not on LRU at this point, e.g. the old page of >>>> + * NUMA migration. And PageTransHuge is not enough to distinguish >>>> + * with other compound page, e.g. skb, THP destructor is not used >>>> + * anymore and will be removed, so the compound order sounds like >>>> + * the only choice here. >>>> + */ >>>> + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { >>> What happens if the page is the same order as THP is not THP? Why removing >> It may corrupt the deferred split queue since it is never added into the >> list, but deleted here. >> >>> of destructor is required? >> Due to the change to free_transhuge_page() (extracted deferred split queue >> manipulation and moved before memcg uncharge since page->mem_cgroup is >> needed), it just calls free_compound_page(). So, it sounds pointless to >> still keep THP specific destructor. >> >> It looks there is not a good way to tell if the compound page is THP in >> free_page path or not, we may keep the destructor just for this? > Other option would be to move mem_cgroup_uncharge(page); from > __page_cache_release() to destructors. Destructors will be able to > call it as it fits. Yes, it is an option. Since __page_cache_release() is called by __put_single_page() too, so mem_cgroup_uncharge() has to be called in both __put_single_page() and the desctructor (free_compound_page() which is called by both THP and other compound page except HugeTLB). But, it sounds acceptable IMHO. >
diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 7cd5c15..8137c3a 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -250,6 +250,17 @@ static inline bool thp_migration_supported(void) return IS_ENABLED(CONFIG_ARCH_ENABLE_THP_MIGRATION); } +static inline struct list_head *page_deferred_list(struct page *page) +{ + /* + * Global or memcg deferred list in the second tail pages is + * occupied by compound_head. + */ + return &page[2].deferred_list; +} + +extern void del_thp_from_deferred_split_queue(struct page *); + #else /* CONFIG_TRANSPARENT_HUGEPAGE */ #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; }) #define HPAGE_PMD_MASK ({ BUILD_BUG(); 0; }) @@ -368,6 +379,10 @@ static inline bool thp_migration_supported(void) { return false; } + +static inline void del_thp_from_deferred_split_queue(struct page *page) +{ +} #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ #endif /* _LINUX_HUGE_MM_H */ diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index bc74d6a..5d3c10c 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h @@ -316,6 +316,10 @@ struct mem_cgroup { struct list_head event_list; spinlock_t event_list_lock; +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + struct deferred_split deferred_split_queue; +#endif + struct mem_cgroup_per_node *nodeinfo[0]; /* WARNING: nodeinfo must be the last member here */ }; diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 8ec38b1..4eabf80 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -139,6 +139,7 @@ struct page { struct { /* Second tail page of compound page */ unsigned long _compound_pad_1; /* compound_head */ unsigned long _compound_pad_2; + /* For both global and memcg */ struct list_head deferred_list; }; struct { /* Page table pages */ diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 81cf759..3307697 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -492,10 +492,15 @@ pmd_t maybe_pmd_mkwrite(pmd_t pmd, struct vm_area_struct *vma) return pmd; } -static inline struct list_head *page_deferred_list(struct page *page) +static inline struct deferred_split *get_deferred_split_queue(struct page *page) { - /* ->lru in the tail pages is occupied by compound_head. */ - return &page[2].deferred_list; + struct mem_cgroup *memcg = compound_head(page)->mem_cgroup; + struct pglist_data *pgdat = NODE_DATA(page_to_nid(page)); + + if (memcg) + return &memcg->deferred_split_queue; + else + return &pgdat->deferred_split_queue; } void prep_transhuge_page(struct page *page) @@ -2658,7 +2663,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) { struct page *head = compound_head(page); struct pglist_data *pgdata = NODE_DATA(page_to_nid(head)); - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; + struct deferred_split *ds_queue = get_deferred_split_queue(page); struct anon_vma *anon_vma = NULL; struct address_space *mapping = NULL; int count, mapcount, extra_pins, ret; @@ -2792,25 +2797,36 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) return ret; } -void free_transhuge_page(struct page *page) +void del_thp_from_deferred_split_queue(struct page *page) { - struct pglist_data *pgdata = NODE_DATA(page_to_nid(page)); - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; - unsigned long flags; - - spin_lock_irqsave(&ds_queue->split_queue_lock, flags); - if (!list_empty(page_deferred_list(page))) { - ds_queue->split_queue_len--; - list_del(page_deferred_list(page)); + /* + * The THP may be not on LRU at this point, e.g. the old page of + * NUMA migration. And PageTransHuge is not enough to distinguish + * with other compound page, e.g. skb, THP destructor is not used + * anymore and will be removed, so the compound order sounds like + * the only choice here. + */ + if (PageTransHuge(page) && compound_order(page) == HPAGE_PMD_ORDER) { + struct deferred_split *ds_queue = get_deferred_split_queue(page); + unsigned long flags; + spin_lock_irqsave(&ds_queue->split_queue_lock, flags); + if (!list_empty(page_deferred_list(page))) { + ds_queue->split_queue_len--; + list_del(page_deferred_list(page)); + } + spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); } - spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); +} + +void free_transhuge_page(struct page *page) +{ free_compound_page(page); } void deferred_split_huge_page(struct page *page) { - struct pglist_data *pgdata = NODE_DATA(page_to_nid(page)); - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; + struct deferred_split *ds_queue = get_deferred_split_queue(page); + struct mem_cgroup *memcg = compound_head(page)->mem_cgroup; unsigned long flags; VM_BUG_ON_PAGE(!PageTransHuge(page), page); @@ -2820,6 +2836,9 @@ void deferred_split_huge_page(struct page *page) count_vm_event(THP_DEFERRED_SPLIT_PAGE); list_add_tail(page_deferred_list(page), &ds_queue->split_queue); ds_queue->split_queue_len++; + if (memcg) + memcg_set_shrinker_bit(memcg, page_to_nid(page), + deferred_split_shrinker.id); } spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags); } @@ -2827,8 +2846,15 @@ void deferred_split_huge_page(struct page *page) static unsigned long deferred_split_count(struct shrinker *shrink, struct shrink_control *sc) { - struct pglist_data *pgdata = NODE_DATA(sc->nid); - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; + struct deferred_split *ds_queue; + + if (!sc->memcg) { + struct pglist_data *pgdata = NODE_DATA(sc->nid); + ds_queue = &pgdata->deferred_split_queue; + return READ_ONCE(ds_queue->split_queue_len); + } + + ds_queue = &sc->memcg->deferred_split_queue; return READ_ONCE(ds_queue->split_queue_len); } @@ -2836,12 +2862,17 @@ static unsigned long deferred_split_scan(struct shrinker *shrink, struct shrink_control *sc) { struct pglist_data *pgdata = NODE_DATA(sc->nid); - struct deferred_split *ds_queue = &pgdata->deferred_split_queue; + struct deferred_split *ds_queue; unsigned long flags; LIST_HEAD(list), *pos, *next; struct page *page; int split = 0; + if (sc->memcg) + ds_queue = &sc->memcg->deferred_split_queue; + else + ds_queue = &pgdata->deferred_split_queue; + spin_lock_irqsave(&ds_queue->split_queue_lock, flags); /* Take pin on all head pages to avoid freeing them under us */ list_for_each_safe(pos, next, &ds_queue->split_queue) { @@ -2888,7 +2919,7 @@ static unsigned long deferred_split_scan(struct shrinker *shrink, .count_objects = deferred_split_count, .scan_objects = deferred_split_scan, .seeks = DEFAULT_SEEKS, - .flags = SHRINKER_NUMA_AWARE, + .flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE, }; #ifdef CONFIG_DEBUG_FS diff --git a/mm/memcontrol.c b/mm/memcontrol.c index e50a2db..fe7e544 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -4579,6 +4579,11 @@ static struct mem_cgroup *mem_cgroup_alloc(void) #ifdef CONFIG_CGROUP_WRITEBACK INIT_LIST_HEAD(&memcg->cgwb_list); #endif +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); + INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); + memcg->deferred_split_queue.split_queue_len = 0; +#endif idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); return memcg; fail: @@ -4949,6 +4954,20 @@ static int mem_cgroup_move_account(struct page *page, __mod_memcg_state(to, NR_WRITEBACK, nr_pages); } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + if (compound && !list_empty(page_deferred_list(page))) { + spin_lock(&from->deferred_split_queue.split_queue_lock); + list_del(page_deferred_list(page)); + from->deferred_split_queue.split_queue_len--; + spin_unlock(&from->deferred_split_queue.split_queue_lock); + + spin_lock(&to->deferred_split_queue.split_queue_lock); + list_add_tail(page_deferred_list(page), + &to->deferred_split_queue.split_queue); + to->deferred_split_queue.split_queue_len++; + spin_unlock(&to->deferred_split_queue.split_queue_lock); + } +#endif /* * It is safe to change page->mem_cgroup here because the page * is referenced, charged, and isolated - we can't race with diff --git a/mm/swap.c b/mm/swap.c index 3a75722..3348295 100644 --- a/mm/swap.c +++ b/mm/swap.c @@ -69,6 +69,10 @@ static void __page_cache_release(struct page *page) del_page_from_lru_list(page, lruvec, page_off_lru(page)); spin_unlock_irqrestore(&pgdat->lru_lock, flags); } + + /* Delete THP from deferred split queue before memcg uncharge */ + del_thp_from_deferred_split_queue(page); + __ClearPageWaiters(page); mem_cgroup_uncharge(page); }
Currently THP deferred split shrinker is not memcg aware, this may cause premature OOM with some configuration. For example the below test would run into premature OOM easily: $ cgcreate -g memory:thp $ echo 4G > /sys/fs/cgroup/memory/thp/memory/limit_in_bytes $ cgexec -g memory:thp transhuge-stress 4000 transhuge-stress comes from kernel selftest. It is easy to hit OOM, but there are still a lot THP on the deferred split queue, memcg direct reclaim can't touch them since the deferred split shrinker is not memcg aware. Convert deferred split shrinker memcg aware by introducing per memcg deferred split queue. The THP should be on either per node or per memcg deferred split queue if it belongs to a memcg. When the page is immigrated to the other memcg, it will be immigrated to the target memcg's deferred split queue too. And, move deleting THP from deferred split queue in page free before memcg uncharge so that the page's memcg information is available. Reuse the second tail page's deferred_list for per memcg list since the same THP can't be on multiple deferred split queues. Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.com> Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> --- include/linux/huge_mm.h | 15 ++++++++++ include/linux/memcontrol.h | 4 +++ include/linux/mm_types.h | 1 + mm/huge_memory.c | 71 +++++++++++++++++++++++++++++++++------------- mm/memcontrol.c | 19 +++++++++++++ mm/swap.c | 4 +++ 6 files changed, 94 insertions(+), 20 deletions(-)