[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20170123155638.db6036219cb6ab2582be104e@linux-foundation.org>
Date: Mon, 23 Jan 2017 15:56:38 -0800
From: Andrew Morton <akpm@...ux-foundation.org>
To: Arnd Bergmann <arnd@...db.de>
Cc: Vlastimil Babka <vbabka@...e.cz>,
Mel Gorman <mgorman@...hsingularity.net>,
Michal Hocko <mhocko@...e.com>,
Johannes Weiner <hannes@...xchg.org>,
Joonsoo Kim <iamjoonsoo.kim@....com>, linux-mm@...ck.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mm: ensure alloc_flags in slow path are initialized
On Mon, 23 Jan 2017 13:16:12 +0100 Arnd Bergmann <arnd@...db.de> wrote:
> The __alloc_pages_slowpath() has gotten rather complex and gcc
> is no longer able to follow the gotos and prove that the
> alloc_flags variable is initialized at the time it is used:
>
> mm/page_alloc.c: In function '__alloc_pages_slowpath':
> mm/page_alloc.c:3565:15: error: 'alloc_flags' may be used uninitialized in this function [-Werror=maybe-uninitialized]
>
> To be honest, I can't figure that out either, maybe it is or
> maybe not, but moving the existing initialization up a little
> higher looks safe and makes it obvious to both me and gcc that
> the initialization comes before the first use.
>
> ...
>
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3591,6 +3591,13 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
> gfp_mask &= ~__GFP_ATOMIC;
>
> + /*
> + * The fast path uses conservative alloc_flags to succeed only until
> + * kswapd needs to be woken up, and to avoid the cost of setting up
> + * alloc_flags precisely. So we do that now.
> + */
> + alloc_flags = gfp_to_alloc_flags(gfp_mask);
> +
> retry_cpuset:
> compaction_retries = 0;
> no_progress_loops = 0;
> @@ -3607,14 +3614,6 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> if (!ac->preferred_zoneref->zone)
> goto nopage;
>
> -
> - /*
> - * The fast path uses conservative alloc_flags to succeed only until
> - * kswapd needs to be woken up, and to avoid the cost of setting up
> - * alloc_flags precisely. So we do that now.
> - */
> - alloc_flags = gfp_to_alloc_flags(gfp_mask);
> -
> if (gfp_mask & __GFP_KSWAPD_RECLAIM)
> wake_all_kswapds(order, ac);
hm. But we later do
if (gfp_pfmemalloc_allowed(gfp_mask))
alloc_flags = ALLOC_NO_WATERMARKS;
...
if (read_mems_allowed_retry(cpuset_mems_cookie))
goto retry_cpuset;
so with your patch there's a path where we can rerun everything with
alloc_flags == ALLOC_NO_WATERMARKS. That's changed behaviour.
When I saw the test robot warning I did this, which I think preserves
behaviour?
--- a/mm/page_alloc.c~mm-consolidate-gfp_nofail-checks-in-the-allocator-slowpath-fix
+++ a/mm/page_alloc.c
@@ -3577,6 +3577,14 @@ retry_cpuset:
no_progress_loops = 0;
compact_priority = DEF_COMPACT_PRIORITY;
cpuset_mems_cookie = read_mems_allowed_begin();
+
+ /*
+ * The fast path uses conservative alloc_flags to succeed only until
+ * kswapd needs to be woken up, and to avoid the cost of setting up
+ * alloc_flags precisely. So we do that now.
+ */
+ alloc_flags = gfp_to_alloc_flags(gfp_mask);
+
/*
* We need to recalculate the starting point for the zonelist iterator
* because we might have used different nodemask in the fast path, or
@@ -3588,14 +3596,6 @@ retry_cpuset:
if (!ac->preferred_zoneref->zone)
goto nopage;
-
- /*
- * The fast path uses conservative alloc_flags to succeed only until
- * kswapd needs to be woken up, and to avoid the cost of setting up
- * alloc_flags precisely. So we do that now.
- */
- alloc_flags = gfp_to_alloc_flags(gfp_mask);
-
if (gfp_mask & __GFP_KSWAPD_RECLAIM)
wake_all_kswapds(order, ac);
_
Powered by blists - more mailing lists