Re: [v6 PATCH 07/11] mm: vmscan: add per memcg shrinker nr_deferred

From: Yang Shi
Date: Fri Feb 05 2021 - 15:14:14 EST


On Fri, Feb 5, 2021 at 6:38 AM Kirill Tkhai <ktkhai@xxxxxxxxxxxxx> wrote:
>
> On 04.02.2021 20:17, Yang Shi wrote:
> > On Thu, Feb 4, 2021 at 12:31 AM Kirill Tkhai <ktkhai@xxxxxxxxxxxxx> wrote:
> >>
> >> On 03.02.2021 20:20, Yang Shi wrote:
> >>> Currently the number of deferred objects are per shrinker, but some slabs, for example,
> >>> vfs inode/dentry cache are per memcg, this would result in poor isolation among memcgs.
> >>>
> >>> The deferred objects typically are generated by __GFP_NOFS allocations, one memcg with
> >>> excessive __GFP_NOFS allocations may blow up deferred objects, then other innocent memcgs
> >>> may suffer from over shrink, excessive reclaim latency, etc.
> >>>
> >>> For example, two workloads run in memcgA and memcgB respectively, workload in B is vfs
> >>> heavy workload. Workload in A generates excessive deferred objects, then B's vfs cache
> >>> might be hit heavily (drop half of caches) by B's limit reclaim or global reclaim.
> >>>
> >>> We observed this hit in our production environment which was running vfs heavy workload
> >>> shown as the below tracing log:
> >>>
> >>> <...>-409454 [016] .... 28286961.747146: mm_shrink_slab_start: super_cache_scan+0x0/0x1a0 ffff9a83046f3458:
> >>> nid: 1 objects to shrink 3641681686040 gfp_flags GFP_HIGHUSER_MOVABLE|__GFP_ZERO pgs_scanned 1 lru_pgs 15721
> >>> cache items 246404277 delta 31345 total_scan 123202138
> >>> <...>-409454 [022] .... 28287105.928018: mm_shrink_slab_end: super_cache_scan+0x0/0x1a0 ffff9a83046f3458:
> >>> nid: 1 unused scan count 3641681686040 new scan count 3641798379189 total_scan 602
> >>> last shrinker return val 123186855
> >>>
> >>> The vfs cache and page cache ration was 10:1 on this machine, and half of caches were dropped.
> >>> This also resulted in significant amount of page caches were dropped due to inodes eviction.
> >>>
> >>> Make nr_deferred per memcg for memcg aware shrinkers would solve the unfairness and bring
> >>> better isolation.
> >>>
> >>> When memcg is not enabled (!CONFIG_MEMCG or memcg disabled), the shrinker's nr_deferred
> >>> would be used. And non memcg aware shrinkers use shrinker's nr_deferred all the time.
> >>>
> >>> Signed-off-by: Yang Shi <shy828301@xxxxxxxxx>
> >>> ---
> >>> include/linux/memcontrol.h | 7 +++---
> >>> mm/vmscan.c | 45 ++++++++++++++++++++++++--------------
> >>> 2 files changed, 33 insertions(+), 19 deletions(-)
> >>>
> >>> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> >>> index 4c9253896e25..c457fc7bc631 100644
> >>> --- a/include/linux/memcontrol.h
> >>> +++ b/include/linux/memcontrol.h
> >>> @@ -93,12 +93,13 @@ struct lruvec_stat {
> >>> };
> >>>
> >>> /*
> >>> - * Bitmap of shrinker::id corresponding to memcg-aware shrinkers,
> >>> - * which have elements charged to this memcg.
> >>> + * Bitmap and deferred work of shrinker::id corresponding to memcg-aware
> >>> + * shrinkers, which have elements charged to this memcg.
> >>> */
> >>> struct shrinker_info {
> >>> struct rcu_head rcu;
> >>> - unsigned long map[];
> >>> + atomic_long_t *nr_deferred;
> >>> + unsigned long *map;
> >>> };
> >>>
> >>> /*
> >>> diff --git a/mm/vmscan.c b/mm/vmscan.c
> >>> index dc0d69e081b0..d9126f12890f 100644
> >>> --- a/mm/vmscan.c
> >>> +++ b/mm/vmscan.c
> >>> @@ -196,10 +196,12 @@ static void free_shrinker_info_rcu(struct rcu_head *head)
> >>> }
> >>>
> >>> static int expand_one_shrinker_info(struct mem_cgroup *memcg,
> >>> - int size, int old_size)
> >>> + int m_size, int d_size,
> >>> + int old_m_size, int old_d_size)
> >>> {
> >>> struct shrinker_info *new, *old;
> >>> int nid;
> >>> + int size = m_size + d_size;
> >>>
> >>> for_each_node(nid) {
> >>> old = rcu_dereference_protected(
> >>> @@ -212,9 +214,15 @@ static int expand_one_shrinker_info(struct mem_cgroup *memcg,
> >>> if (!new)
> >>> return -ENOMEM;
> >>>
> >>> - /* Set all old bits, clear all new bits */
> >>> - memset(new->map, (int)0xff, old_size);
> >>> - memset((void *)new->map + old_size, 0, size - old_size);
> >>> + new->nr_deferred = (atomic_long_t *)(new + 1);
> >>> + new->map = (void *)new->nr_deferred + d_size;
> >>> +
> >>> + /* map: set all old bits, clear all new bits */
> >>> + memset(new->map, (int)0xff, old_m_size);
> >>> + memset((void *)new->map + old_m_size, 0, m_size - old_m_size);
> >>> + /* nr_deferred: copy old values, clear all new values */
> >>> + memcpy(new->nr_deferred, old->nr_deferred, old_d_size);
> >>> + memset((void *)new->nr_deferred + old_d_size, 0, d_size - old_d_size);
> >>>
> >>> rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_info, new);
> >>> call_rcu(&old->rcu, free_shrinker_info_rcu);
> >>> @@ -229,9 +237,6 @@ void free_shrinker_info(struct mem_cgroup *memcg)
> >>> struct shrinker_info *info;
> >>> int nid;
> >>>
> >>> - if (mem_cgroup_is_root(memcg))
> >>> - return;
> >>> -
> >>> for_each_node(nid) {
> >>> pn = mem_cgroup_nodeinfo(memcg, nid);
> >>> info = rcu_dereference_protected(pn->shrinker_info, true);
> >>> @@ -244,12 +249,13 @@ int alloc_shrinker_info(struct mem_cgroup *memcg)
> >>> {
> >>> struct shrinker_info *info;
> >>> int nid, size, ret = 0;
> >>> -
> >>> - if (mem_cgroup_is_root(memcg))
> >>> - return 0;
> >>> + int m_size, d_size = 0;
> >>>
> >>> down_write(&shrinker_rwsem);
> >>> - size = NR_MAX_TO_SHR_MAP_SIZE(shrinker_nr_max);
> >>> + m_size = NR_MAX_TO_SHR_MAP_SIZE(shrinker_nr_max);
> >>> + d_size = shrinker_nr_max * sizeof(atomic_long_t);
> >>> + size = m_size + d_size;
> >>> +
> >>> for_each_node(nid) {
> >>> info = kvzalloc_node(sizeof(*info) + size, GFP_KERNEL, nid);
> >>> if (!info) {
> >>> @@ -257,6 +263,8 @@ int alloc_shrinker_info(struct mem_cgroup *memcg)
> >>> ret = -ENOMEM;
> >>> break;
> >>> }
> >>> + info->nr_deferred = (atomic_long_t *)(info + 1);
> >>> + info->map = (void *)info->nr_deferred + d_size;
> >>> rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_info, info);
> >>> }
> >>> up_write(&shrinker_rwsem);
> >>> @@ -268,10 +276,16 @@ static int expand_shrinker_info(int new_id)
> >>> {
> >>> int size, old_size, ret = 0;
> >>> int new_nr_max = new_id + 1;
> >>> + int m_size, d_size = 0;
> >>> + int old_m_size, old_d_size = 0;
> >>> struct mem_cgroup *memcg;
> >>>
> >>> - size = NR_MAX_TO_SHR_MAP_SIZE(new_nr_max);
> >>> - old_size = NR_MAX_TO_SHR_MAP_SIZE(shrinker_nr_max);
> >>> + m_size = NR_MAX_TO_SHR_MAP_SIZE(new_nr_max);
> >>> + d_size = new_nr_max * sizeof(atomic_long_t);
> >>> + size = m_size + d_size;
> >>> + old_m_size = NR_MAX_TO_SHR_MAP_SIZE(shrinker_nr_max);
> >>> + old_d_size = shrinker_nr_max * sizeof(atomic_long_t);
> >>> + old_size = old_m_size + old_d_size;
> >>> if (size <= old_size)
> >>> goto out;
> >>
> >> Before this patch we used to allocate shrinker_info with BITS_PER_LONG batching.
> >> So, first registered shrinker used to allocate a map of unsigned long size, and
> >> we could to allocate 63 more shrinkers without maps expanding.
> >>
> >> After this patch we will expand maps on every shrinker registration, won't we?
> >
> > Yes, I'm supposed "maps" means "info".I'm supposed the most shrinkers
> > should be registered at boot time, and typically very few memcgs are
> > created at boot time so I didn't treat it as a hot path.
>
> Not so. Every mount adds at least one shrinker, so they can actively be added
> during normal system work.
>
> E.g., on our production system (containers) several thousand shrinkers
> is not a rare situation.

Aha, yes, I missed this point.

>
> >> What do you think about batching here?
> >
> > Just off the top of my head, we could allocate, for example, 64
> > nr_deferred (64 * sizeof(atomic_long_t)) so that we just need to
> > expand info for every 64 shrinker registrations. Maybe define it
> > depends on the machine (64 bit - 64, 32 bit - 32).
> >
> > Why 64? Basically a magic number. And when I was investigating that
> > list_lru reparent race issue
> > (https://lore.kernel.org/linux-mm/20201202171749.264354-1-shy828301@xxxxxxxxx/)
> > I happened to notice that there are at most 64 shrinkers registered in
> > our production environment (a typical data center configuration).
> >
> > How do you think about it?
>
> I think 64 is OK for now. We may use some #define to set this value, so we will
> be able to change it easily in the future.

BITS_PER_LONG might be better. We could reuse all the existing logic
without adding too much new code.

>
> >>>
> >>> @@ -280,9 +294,8 @@ static int expand_shrinker_info(int new_id)
> >>>
> >>> memcg = mem_cgroup_iter(NULL, NULL, NULL);
> >>> do {
> >>> - if (mem_cgroup_is_root(memcg))
> >>> - continue;
> >>> - ret = expand_one_shrinker_info(memcg, size, old_size);
> >>> + ret = expand_one_shrinker_info(memcg, m_size, d_size,
> >>> + old_m_size, old_d_size);
> >>> if (ret) {
> >>> mem_cgroup_iter_break(NULL, memcg);
> >>> goto out;
> >>>
> >>
> >>
>