Re: [PATCH] mm: memcg/slab: properly handle kmem_caches reparented to root_mem_cgroup
On Thu, Jun 20, 2019 at 08:48:00AM -0700, Shakeel Butt wrote: > On Wed, Jun 19, 2019 at 6:57 PM Roman Gushchin wrote: > > > > As a result of reparenting a kmem_cache might belong to the root > > memory cgroup. It happens when a top-level memory cgroup is removed, > > and all associated kmem_caches are reparented to the root memory > > cgroup. > > > > The root memory cgroup is special, and requires a special handling. > > Let's make sure that we don't try to charge or uncharge it, > > and we handle system-wide vmstats exactly as for root kmem_caches. > > > > Note, that we still need to alter the kmem_cache reference counter, > > so that the kmem_cache can be released properly. > > > > The issue was discovered by running CRIU tests; the following warning > > did appear: > > > > [ 381.345960] WARNING: CPU: 0 PID: 11655 at mm/page_counter.c:62 > > page_counter_cancel+0x26/0x30 > > [ 381.345992] Modules linked in: > > [ 381.345998] CPU: 0 PID: 11655 Comm: kworker/0:8 Not tainted > > 5.2.0-rc5-next-20190618+ #1 > > [ 381.346001] Hardware name: Google Google Compute Engine/Google > > Compute Engine, BIOS Google 01/01/2011 > > [ 381.346010] Workqueue: memcg_kmem_cache kmemcg_workfn > > [ 381.346013] RIP: 0010:page_counter_cancel+0x26/0x30 > > [ 381.346017] Code: 1f 44 00 00 0f 1f 44 00 00 48 89 f0 53 48 f7 d8 > > f0 48 0f c1 07 48 29 f0 48 89 c3 48 89 c6 e8 61 ff ff ff 48 85 db 78 > > 02 5b c3 <0f> 0b 5b c3 66 0f 1f 44 00 00 0f 1f 44 00 00 48 85 ff 74 41 > > 41 55 > > [ 381.346019] RSP: 0018:b3b34319f990 EFLAGS: 00010086 > > [ 381.346022] RAX: fffc RBX: fffc RCX: > > 0004 > > [ 381.346024] RDX: RSI: fffc RDI: > > 9c2cd7165270 > > [ 381.346026] RBP: 0004 R08: R09: > > 0001 > > [ 381.346028] R10: 00c8 R11: 9c2cd684e660 R12: > > fffc > > [ 381.346030] R13: 0002 R14: 0006 R15: > > 9c2c8ce1f200 > > [ 381.346033] FS: () GS:9c2cd820() > > knlGS: > > [ 381.346039] CS: 0010 DS: ES: CR0: 80050033 > > [ 381.346041] CR2: 007be000 CR3: 0001cdbfc005 CR4: > > 001606f0 > > [ 381.346043] DR0: DR1: DR2: > > > > [ 381.346045] DR3: DR6: fffe0ff0 DR7: > > 0400 > > [ 381.346047] Call Trace: > > [ 381.346054] page_counter_uncharge+0x1d/0x30 > > [ 381.346065] __memcg_kmem_uncharge_memcg+0x39/0x60 > > [ 381.346071] __free_slab+0x34c/0x460 > > [ 381.346079] deactivate_slab.isra.80+0x57d/0x6d0 > > [ 381.346088] ? add_lock_to_list.isra.36+0x9c/0xf0 > > [ 381.346095] ? __lock_acquire+0x252/0x1410 > > [ 381.346106] ? cpumask_next_and+0x19/0x20 > > [ 381.346110] ? slub_cpu_dead+0xd0/0xd0 > > [ 381.346113] flush_cpu_slab+0x36/0x50 > > [ 381.346117] ? slub_cpu_dead+0xd0/0xd0 > > [ 381.346125] on_each_cpu_mask+0x51/0x70 > > [ 381.346131] ? ksm_migrate_page+0x60/0x60 > > [ 381.346134] on_each_cpu_cond_mask+0xab/0x100 > > [ 381.346143] __kmem_cache_shrink+0x56/0x320 > > [ 381.346150] ? ret_from_fork+0x3a/0x50 > > [ 381.346157] ? unwind_next_frame+0x73/0x480 > > [ 381.346176] ? __lock_acquire+0x252/0x1410 > > [ 381.346188] ? kmemcg_workfn+0x21/0x50 > > [ 381.346196] ? __mutex_lock+0x99/0x920 > > [ 381.346199] ? kmemcg_workfn+0x21/0x50 > > [ 381.346205] ? kmemcg_workfn+0x21/0x50 > > [ 381.346216] __kmemcg_cache_deactivate_after_rcu+0xe/0x40 > > [ 381.346220] kmemcg_cache_deactivate_after_rcu+0xe/0x20 > > [ 381.346223] kmemcg_workfn+0x31/0x50 > > [ 381.346230] process_one_work+0x23c/0x5e0 > > [ 381.346241] worker_thread+0x3c/0x390 > > [ 381.346248] ? process_one_work+0x5e0/0x5e0 > > [ 381.346252] kthread+0x11d/0x140 > > [ 381.346255] ? kthread_create_on_node+0x60/0x60 > > [ 381.346261] ret_from_fork+0x3a/0x50 > > [ 381.346275] irq event stamp: 10302 > > [ 381.346278] hardirqs last enabled at (10301): [] > > _raw_spin_unlock_irq+0x29/0x40 > > [ 381.346282] hardirqs last disabled at (10302): [] > > on_each_cpu_mask+0x49/0x70 > > [ 381.346287] softirqs last enabled at (10262): [] > > cgroup_idr_replace+0x3a/0x50 > > [ 381.346290] softirqs last disabled at (10260): [] > > cgroup_idr_replace+0x1d/0x50 > > [ 381.346293] ---[ end trace b324ba73eb3659f0 ]--- > > > > Reported-by: Andrei Vagin > > Signed-off-by: Roman Gushchin > > Cc: Christoph Lameter > > Cc: Johannes Weiner > > Cc: Michal Hocko > > Cc: Shakeel Butt > > Cc: Vladimir Davydov > > Cc: Waiman Long > > Cc: David Rientjes > > Cc: Joonsoo Kim > > Cc: Pekka Enberg > > --- > > mm/slab.h | 17 + > > 1 file changed, 13 insertions(+), 4 deletions(-) > > > > diff --git a/mm/slab.h b/mm/slab.h > > index a4c9b9d042de..c02e7f44268b 100644 > > --- a/mm/slab.h > > +++ b/mm/slab.h > > @@ -294,8 +294,12 @@ static __always_inline int
Re: [PATCH] mm: memcg/slab: properly handle kmem_caches reparented to root_mem_cgroup
On Wed, Jun 19, 2019 at 6:57 PM Roman Gushchin wrote:
>
> As a result of reparenting a kmem_cache might belong to the root
> memory cgroup. It happens when a top-level memory cgroup is removed,
> and all associated kmem_caches are reparented to the root memory
> cgroup.
>
> The root memory cgroup is special, and requires a special handling.
> Let's make sure that we don't try to charge or uncharge it,
> and we handle system-wide vmstats exactly as for root kmem_caches.
>
> Note, that we still need to alter the kmem_cache reference counter,
> so that the kmem_cache can be released properly.
>
> The issue was discovered by running CRIU tests; the following warning
> did appear:
>
> [ 381.345960] WARNING: CPU: 0 PID: 11655 at mm/page_counter.c:62
> page_counter_cancel+0x26/0x30
> [ 381.345992] Modules linked in:
> [ 381.345998] CPU: 0 PID: 11655 Comm: kworker/0:8 Not tainted
> 5.2.0-rc5-next-20190618+ #1
> [ 381.346001] Hardware name: Google Google Compute Engine/Google
> Compute Engine, BIOS Google 01/01/2011
> [ 381.346010] Workqueue: memcg_kmem_cache kmemcg_workfn
> [ 381.346013] RIP: 0010:page_counter_cancel+0x26/0x30
> [ 381.346017] Code: 1f 44 00 00 0f 1f 44 00 00 48 89 f0 53 48 f7 d8
> f0 48 0f c1 07 48 29 f0 48 89 c3 48 89 c6 e8 61 ff ff ff 48 85 db 78
> 02 5b c3 <0f> 0b 5b c3 66 0f 1f 44 00 00 0f 1f 44 00 00 48 85 ff 74 41
> 41 55
> [ 381.346019] RSP: 0018:b3b34319f990 EFLAGS: 00010086
> [ 381.346022] RAX: fffc RBX: fffc RCX:
> 0004
> [ 381.346024] RDX: RSI: fffc RDI:
> 9c2cd7165270
> [ 381.346026] RBP: 0004 R08: R09:
> 0001
> [ 381.346028] R10: 00c8 R11: 9c2cd684e660 R12:
> fffc
> [ 381.346030] R13: 0002 R14: 0006 R15:
> 9c2c8ce1f200
> [ 381.346033] FS: () GS:9c2cd820()
> knlGS:
> [ 381.346039] CS: 0010 DS: ES: CR0: 80050033
> [ 381.346041] CR2: 007be000 CR3: 0001cdbfc005 CR4:
> 001606f0
> [ 381.346043] DR0: DR1: DR2:
>
> [ 381.346045] DR3: DR6: fffe0ff0 DR7:
> 0400
> [ 381.346047] Call Trace:
> [ 381.346054] page_counter_uncharge+0x1d/0x30
> [ 381.346065] __memcg_kmem_uncharge_memcg+0x39/0x60
> [ 381.346071] __free_slab+0x34c/0x460
> [ 381.346079] deactivate_slab.isra.80+0x57d/0x6d0
> [ 381.346088] ? add_lock_to_list.isra.36+0x9c/0xf0
> [ 381.346095] ? __lock_acquire+0x252/0x1410
> [ 381.346106] ? cpumask_next_and+0x19/0x20
> [ 381.346110] ? slub_cpu_dead+0xd0/0xd0
> [ 381.346113] flush_cpu_slab+0x36/0x50
> [ 381.346117] ? slub_cpu_dead+0xd0/0xd0
> [ 381.346125] on_each_cpu_mask+0x51/0x70
> [ 381.346131] ? ksm_migrate_page+0x60/0x60
> [ 381.346134] on_each_cpu_cond_mask+0xab/0x100
> [ 381.346143] __kmem_cache_shrink+0x56/0x320
> [ 381.346150] ? ret_from_fork+0x3a/0x50
> [ 381.346157] ? unwind_next_frame+0x73/0x480
> [ 381.346176] ? __lock_acquire+0x252/0x1410
> [ 381.346188] ? kmemcg_workfn+0x21/0x50
> [ 381.346196] ? __mutex_lock+0x99/0x920
> [ 381.346199] ? kmemcg_workfn+0x21/0x50
> [ 381.346205] ? kmemcg_workfn+0x21/0x50
> [ 381.346216] __kmemcg_cache_deactivate_after_rcu+0xe/0x40
> [ 381.346220] kmemcg_cache_deactivate_after_rcu+0xe/0x20
> [ 381.346223] kmemcg_workfn+0x31/0x50
> [ 381.346230] process_one_work+0x23c/0x5e0
> [ 381.346241] worker_thread+0x3c/0x390
> [ 381.346248] ? process_one_work+0x5e0/0x5e0
> [ 381.346252] kthread+0x11d/0x140
> [ 381.346255] ? kthread_create_on_node+0x60/0x60
> [ 381.346261] ret_from_fork+0x3a/0x50
> [ 381.346275] irq event stamp: 10302
> [ 381.346278] hardirqs last enabled at (10301): []
> _raw_spin_unlock_irq+0x29/0x40
> [ 381.346282] hardirqs last disabled at (10302): []
> on_each_cpu_mask+0x49/0x70
> [ 381.346287] softirqs last enabled at (10262): []
> cgroup_idr_replace+0x3a/0x50
> [ 381.346290] softirqs last disabled at (10260): []
> cgroup_idr_replace+0x1d/0x50
> [ 381.346293] ---[ end trace b324ba73eb3659f0 ]---
>
> Reported-by: Andrei Vagin
> Signed-off-by: Roman Gushchin
> Cc: Christoph Lameter
> Cc: Johannes Weiner
> Cc: Michal Hocko
> Cc: Shakeel Butt
> Cc: Vladimir Davydov
> Cc: Waiman Long
> Cc: David Rientjes
> Cc: Joonsoo Kim
> Cc: Pekka Enberg
> ---
> mm/slab.h | 17 +
> 1 file changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/mm/slab.h b/mm/slab.h
> index a4c9b9d042de..c02e7f44268b 100644
> --- a/mm/slab.h
> +++ b/mm/slab.h
> @@ -294,8 +294,12 @@ static __always_inline int memcg_charge_slab(struct page
> *page,
> memcg = parent_mem_cgroup(memcg);
> rcu_read_unlock();
>
> - if (unlikely(!memcg))
> + if (unlikely(!memcg || mem_cgroup_is_root(memcg))) {
> + mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
> +
