mm: memcg: relayout structure mem_cgroup to avoid cache interference

0day reported one -22.7% regression for will-it-scale page_fault2
case [1] on a 4 sockets 144 CPU platform, and bisected to it to be
caused by Waiman's optimization (commit bd0b230fe1) of saving one
'struct page_counter' space for 'struct mem_cgroup'.

Initially we thought it was due to the cache alignment change introduced
by the patch, but further debug shows that it is due to some hot data
members ('vmstats_local', 'vmstats_percpu', 'vmstats') sit in 2 adjacent
cacheline (2N and 2N+1 cacheline), and when adjacent cache line prefetch
is enabled, it triggers an "extended level" of cache false sharing for
2 adjacent cache lines.

So exchange the 2 member blocks, while keeping mostly the original
cache alignment, which can restore and even enhance the performance,
and save 64 bytes of space for 'struct mem_cgroup' (from 2880 to 2816,
with 0day's default RHEL-8.3 kernel config)

[1]. https://lore.kernel.org/lkml/20201102091543.GM31092@shao2-debian/

Fixes: bd0b230fe145 ("mm/memcg: unify swap and memsw page counters")
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Feng Tang <feng.tang@intel.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index a80c59a..922a7f6 100644 (file)
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -282,20 +282,6 @@ struct mem_cgroup {
 
        MEMCG_PADDING(_pad1_);
 
-       /*
-        * set > 0 if pages under this cgroup are moving to other cgroup.
-        */
-       atomic_t                moving_account;
-       struct task_struct      *move_lock_task;
-
-       /* Legacy local VM stats and events */
-       struct memcg_vmstats_percpu __percpu *vmstats_local;
-
-       /* Subtree VM stats and events (batched updates) */
-       struct memcg_vmstats_percpu __percpu *vmstats_percpu;
-
-       MEMCG_PADDING(_pad2_);
-
        atomic_long_t           vmstats[MEMCG_NR_STAT];
        atomic_long_t           vmevents[NR_VM_EVENT_ITEMS];
 
@@ -317,6 +303,20 @@ struct mem_cgroup {
        struct list_head objcg_list; /* list of inherited objcgs */
 #endif
 
+       MEMCG_PADDING(_pad2_);
+
+       /*
+        * set > 0 if pages under this cgroup are moving to other cgroup.
+        */
+       atomic_t                moving_account;
+       struct task_struct      *move_lock_task;
+
+       /* Legacy local VM stats and events */
+       struct memcg_vmstats_percpu __percpu *vmstats_local;
+
+       /* Subtree VM stats and events (batched updates) */
+       struct memcg_vmstats_percpu __percpu *vmstats_percpu;
+
 #ifdef CONFIG_CGROUP_WRITEBACK
        struct list_head cgwb_list;
        struct wb_domain cgwb_domain;