From: David Rientjes <rientjes@google.com>
To: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Chinner <david@fromorbit.com>, Michal Hocko <mhocko@suse.cz>,
Andrew Morton <akpm@linux-foundation.org>,
linux-mm@kvack.org, cgroups@vger.kernel.org,
linux-kernel@vger.kernel.org
Subject: Re: [patch] mm: memcg: do not declare OOM from __GFP_NOFAIL allocations
Date: Tue, 3 Dec 2013 22:10:15 -0800 (PST) [thread overview]
Message-ID: <alpine.DEB.2.02.1312032204270.2597@chino.kir.corp.google.com> (raw)
In-Reply-To: <20131204052542.GY3556@cmpxchg.org>
On Wed, 4 Dec 2013, Johannes Weiner wrote:
> However, the GFP_NOFS | __GFP_NOFAIL task stuck in the page allocator
> may hold filesystem locks that could prevent a third party from
> freeing memory and/or exiting, so we can not guarantee that only the
> __GFP_NOFAIL task is getting stuck, it might well trap other tasks.
> The same applies to open-coded GFP_NOFS allocation loops of course
> unless they cycle the filesystem locks while looping.
>
Yup. I think we should do this:
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2631,6 +2631,11 @@ rebalance:
pages_reclaimed)) {
/* Wait for some write requests to complete then retry */
wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
+
+ /* Allocations that cannot fail must allocate from somewhere */
+ if (gfp_mask & __GFP_NOFAIL)
+ alloc_flags |= ALLOC_HARDER;
+
goto rebalance;
} else {
/*
so that it gets the same behavior as GFP_ATOMIC and is allowed to allocate
from memory reserves (although not enough to totally deplete memory). We
need to leave some memory reserves around in case another process with
__GFP_FS invokes the oom killer and the victim needs memory to exit since
the GFP_NOFS | __GFP_NOFAIL failure wasn't only because reclaim was
limited due to !__GFP_FS.
The only downside of this is that it might become harder in the future to
ever make a case to remove __GFP_NOFAIL entirely since the behavior of the
page allocator is changed with this and it's not equivalent to coding the
retry directly in the caller.
On a tangent, GFP_NOWAIT | __GFP_NOFAIL and GFP_ATOMIC | __GFP_NOFAIL
actually allows allocations to fail. Nothing currently does that, but I
wonder if we should do this for correctness:
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2535,17 +2535,19 @@ rebalance:
}
}
- /* Atomic allocations - we can't balance anything */
- if (!wait)
- goto nopage;
-
- /* Avoid recursion of direct reclaim */
- if (current->flags & PF_MEMALLOC)
- goto nopage;
-
- /* Avoid allocations with no watermarks from looping endlessly */
- if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
- goto nopage;
+ if (likely(!(gfp_mask & __GFP_NOFAIL))) {
+ /* Atomic allocations - we can't balance anything */
+ if (!wait)
+ goto nopage;
+
+ /* Avoid recursion of direct reclaim */
+ if (current->flags & PF_MEMALLOC)
+ goto nopage;
+
+ /* Avoid allocations without watermarks from looping forever */
+ if (test_thread_flag(TIF_MEMDIE))
+ goto nopage;
+ }
/*
* Try direct compaction. The first pass is asynchronous. Subsequent
It can be likely() because the __GFP_NOFAIL restart from the first patch
above will likely now succeed since there's access to memory reserves and
we never actually get here but once for __GFP_NOFAIL. Thoughts on either
patch?
prev parent reply other threads:[~2013-12-04 6:10 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-11-22 17:17 [patch] mm: memcg: do not declare OOM from __GFP_NOFAIL allocations Johannes Weiner
2013-11-27 1:01 ` David Rientjes
2013-11-27 3:33 ` David Rientjes
2013-11-27 16:39 ` Johannes Weiner
2013-11-27 21:38 ` David Rientjes
2013-11-27 22:53 ` Johannes Weiner
2013-11-27 23:34 ` David Rientjes
2013-11-28 10:20 ` Michal Hocko
2013-11-29 23:46 ` David Rientjes
2013-12-02 13:22 ` Michal Hocko
2013-12-02 23:02 ` David Rientjes
2013-12-03 22:25 ` Johannes Weiner
2013-12-03 23:40 ` David Rientjes
2013-12-04 3:01 ` Johannes Weiner
2013-12-04 4:34 ` Dave Chinner
2013-12-04 5:25 ` Johannes Weiner
2013-12-04 6:10 ` David Rientjes [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=alpine.DEB.2.02.1312032204270.2597@chino.kir.corp.google.com \
--to=rientjes@google.com \
--cc=akpm@linux-foundation.org \
--cc=cgroups@vger.kernel.org \
--cc=david@fromorbit.com \
--cc=hannes@cmpxchg.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@suse.cz \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).