On Fri 24-04-20 12:51:03, Johannes Weiner wrote:
> On Fri, Apr 24, 2020 at 06:21:03PM +0200, Michal Hocko wrote:
> > On Fri 24-04-20 11:10:13, Johannes Weiner wrote:
> > > On Fri, Apr 24, 2020 at 04:29:58PM +0200, Michal Hocko wrote:
> > > > On Fri 24-04-20 09:14:50, Johannes Weiner wrote:
> > > > > On Thu, Apr 23, 2020 at 02:16:29AM -0400, Yafang Shao wrote:
> > > > > > This patch is an improvement of a previous version[1], as the previous
> > > > > > version is not easy to understand.
> > > > > > This issue persists in the newest kernel, I have to resend the fix. As
> > > > > > the implementation is changed, I drop Roman's ack from the previous
> > > > > > version.
> > > > >
> > > > > Now that I understand the problem, I much prefer the previous version.
> > > > >
> > > > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > > > > index 745697906ce3..2bf91ae1e640 100644
> > > > > --- a/mm/memcontrol.c
> > > > > +++ b/mm/memcontrol.c
> > > > > @@ -6332,8 +6332,19 @@ enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root,
> > > > >
> > > > > if (!root)
> > > > > root = root_mem_cgroup;
> > > > > - if (memcg == root)
> > > > > + if (memcg == root) {
> > > > > + /*
> > > > > + * The cgroup is the reclaim root in this reclaim
> > > > > + * cycle, and therefore not protected. But it may have
> > > > > + * stale effective protection values from previous
> > > > > + * cycles in which it was not the reclaim root - for
> > > > > + * example, global reclaim followed by limit reclaim.
> > > > > + * Reset these values for mem_cgroup_protection().
> > > > > + */
> > > > > + memcg->memory.emin = 0;
> > > > > + memcg->memory.elow = 0;
> > > > > return MEMCG_PROT_NONE;
> > > > > + }
> > > >
> > > > Could you be more specific why you prefer this over the
> > > > mem_cgroup_protection which doesn't change the effective value?
> > > > Isn't it easier to simply ignore effective value for the reclaim roots?
> > >
> > > Because now both mem_cgroup_protection() and mem_cgroup_protected()
> > > have to know about the reclaim root semantics, instead of just the one
> > > central place.
> >
> > Yes this is true but it is also potentially overwriting the state with
> > a parallel reclaim which can lead to surprising results
>
> Checking in mem_cgroup_protection() doesn't avoid the fundamental race:
>
> root
> `- A (low=2G, elow=2G, max=3G)
> `- A1 (low=2G, elow=2G)
>
> If A does limit reclaim while global reclaim races, the memcg == root
> check in mem_cgroup_protection() will reliably calculate the "right"
> scan value for A, which has no pages, and the wrong scan value for A1
> where the memory actually is.
I am sorry but I do not see how A1 would get wrong scan value.
- Global reclaim
- A.elow = 2G
- A1.elow = min(A1.low, A1.usage) ; if (A.children_low_usage < A.elow)
- A reclaim.
- A.elow = stale/undefined
- A1.elow = A1.low
if mem_cgroup_protection returns 0 for A's reclaim targeting A (assuming
the check is there) then not a big deal as there are no pages there as
you say.
Let's compare the GR (global reclaim), AR (A reclaim).
GR(A1.elow) <= AR(A1.elow) by definition, right? For A1.low
overcommitted we have
min(A1.low, A1.usage) * A.elow / A.children_low_usage <= min(A1.low, A1.usage)
because A.elow <= A.children_low_usage
so in both cases we have GR(A1.elow) <= AR(A1.elow) which means that
racing reclaims will behave sanely because the protection for the
external pressure pressure is not violated. A is going to reclaim A1
less than the global reclaim but that should be OK.
Or what do I miss?
> I'm okay with fixing the case where a really old left-over value is
> used by target reclaim.
>
> I don't see a point in special casing this one instance of a
> fundamental race condition at the expense of less robust code.
I am definitely not calling to fragment the code. I do agree that having
a special case in mem_cgroup_protection is quite non-intuitive.
The existing code is quite hard to reason about in its current form
as we can see. If we can fix all that in mem_cgroup_protected then no
objections from me at all.
--
Michal Hocko
SUSE Labs
