On Fri, Jul 9, 2021 at 6:52 PM Miaohe Lin [off-list ref] wrote:

On 2021/7/10 8:36, Suren Baghdasaryan wrote:

quoted

Add mem_cgroup_disabled check in vmpressure, mem_cgroup_uncharge_swap and
cgroup_throttle_swaprate functions. This minimizes the memcg overhead in
the pagefault and exit_mmap paths when memcgs are disabled using
cgroup_disable=memory command-line option.
This change results in ~2.1% overhead reduction when running PFT test
comparing {CONFIG_MEMCG=n, CONFIG_MEMCG_SWAP=n} against {CONFIG_MEMCG=y,
CONFIG_MEMCG_SWAP=y, cgroup_disable=memory} configuration on an 8-core
ARM64 Android device.

Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Reviewed-by: Shakeel Butt <redacted>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
---
 mm/memcontrol.c | 3 +++
 mm/swapfile.c   | 3 +++
 mm/vmpressure.c | 7 ++++++-
 3 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index ae1f5d0cb581..a228cd51c4bd 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c

@@ -7305,6 +7305,9 @@ void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
      struct mem_cgroup *memcg;
      unsigned short id;

+     if (mem_cgroup_disabled())
+             return;
+
      id = swap_cgroup_record(entry, 0, nr_pages);
      rcu_read_lock();
      memcg = mem_cgroup_from_id(id);

diff --git a/mm/swapfile.c b/mm/swapfile.c
index 1e07d1c776f2..707fa0481bb4 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c

@@ -3778,6 +3778,9 @@ void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
      struct swap_info_struct *si, *next;
      int nid = page_to_nid(page);

+     if (mem_cgroup_disabled())
+             return;
+

Many thanks for your patch. But I'am somewhat confused about this change.
IMO, cgroup_throttle_swaprate() is only related to blk_cgroup and it seems
it's irrelevant to mem_cgroup. Could you please have a explanation for me?

cgroup_throttle_swaprate() is a NoOp when CONFIG_MEMCG=n (see:
https://elixir.bootlin.com/linux/latest/source/include/linux/swap.h#L699),
therefore I assume we can safely skip it when memcgs are disabled via
"cgroup_disable=memory". From perf results I also see no hits on this
function when CONFIG_MEMCG=n.
However, looking into the code, I'm not sure why it should depend on
CONFIG_MEMCG. But it's Friday night and I might be missing some
details here...

Thanks!

quoted

      if (!(gfp_mask & __GFP_IO))
              return;

diff --git a/mm/vmpressure.c b/mm/vmpressure.c
index d69019fc3789..9b172561fded 100644
--- a/mm/vmpressure.c
+++ b/mm/vmpressure.c

@@ -240,7 +240,12 @@ static void vmpressure_work_fn(struct work_struct *work)
 void vmpressure(gfp_t gfp, struct mem_cgroup *memcg, bool tree,
              unsigned long scanned, unsigned long reclaimed)
 {
-     struct vmpressure *vmpr = memcg_to_vmpressure(memcg);
+     struct vmpressure *vmpr;
+
+     if (mem_cgroup_disabled())
+             return;
+
+     vmpr = memcg_to_vmpressure(memcg);

      /*
       * Here we only want to account pressure that userland is able to