Re: regression caused by cgroups optimization in 3.17-rc2

[Dave Hansen]

On 09/02/2014 12:05 PM, Dave Hansen wrote:
> It does not revert cleanly because of the hunks below.  The code in
> those hunks was removed, so I tried running without properly merging
> them and it spews warnings because counter->usage is seen going negative.
> 
> So, it doesn't appear we can quickly revert this.

I'm fairly confident that I missed some of the cases (especially in the
charge-moving code), but the attached patch does at least work around
the regression for me.  It restores the original performance, or at
least gets _close_ to it.




---

 b/mm/memcontrol.c |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff -puN mm/memcontrol.c~try-partial-revert-of-root-charge-regression-patch mm/memcontrol.c
--- a/mm/memcontrol.c~try-partial-revert-of-root-charge-regression-patch	2014-09-02 12:20:11.209527453 -0700
+++ b/mm/memcontrol.c	2014-09-02 13:10:28.756736862 -0700
@@ -2534,6 +2534,8 @@ static int try_charge(struct mem_cgroup
 	unsigned long long size;
 	int ret = 0;
 
+	if (mem_cgroup_is_root(memcg))
+		goto done;
 retry:
 	if (consume_stock(memcg, nr_pages))
 		goto done;
@@ -2640,6 +2642,9 @@ static void __mem_cgroup_cancel_local_ch
 {
 	unsigned long bytes = nr_pages * PAGE_SIZE;
 
+	if (mem_cgroup_is_root(memcg))
+		return;
+
 	res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
 	if (do_swap_account)
 		res_counter_uncharge_until(&memcg->memsw,
@@ -6440,6 +6445,9 @@ void mem_cgroup_commit_charge(struct pag
 	VM_BUG_ON_PAGE(!page->mapping, page);
 	VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
 
+	if (mem_cgroup_is_root(memcg))
+		return;
+
 	if (mem_cgroup_disabled())
 		return;
 	/*
@@ -6484,6 +6492,9 @@ void mem_cgroup_cancel_charge(struct pag
 {
 	unsigned int nr_pages = 1;
 
+	if (mem_cgroup_is_root(memcg))
+		return;
+
 	if (mem_cgroup_disabled())
 		return;
 	/*
@@ -6509,6 +6520,9 @@ static void uncharge_batch(struct mem_cg
 {
 	unsigned long flags;
 
+	if (mem_cgroup_is_root(memcg))
+		return;
+
 	if (nr_mem)
 		res_counter_uncharge(&memcg->res, nr_mem * PAGE_SIZE);
 	if (nr_memsw)
_