Re: [PATCH] mm: page_alloc: Fix setting of ZONE_FAIR_DEPLETED on UP v2

[Leon Romanovsky <leon@xxxxxxx>]

Hi Mel,

On Mon, Sep 8, 2014 at 2:57 PM, Mel Gorman <mgorman@xxxxxxx> wrote:

Commit 4ffeaf35 (mm: page_alloc: reduce cost of the fair zone allocation
policy) arguably broke the fair zone allocation policy on UP with these
hunks.

a/mm/page_alloc.c
b/mm/page_alloc.c
@@ -1612,6 +1612,9 @@ again:
        }

        __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
+       if (zone_page_state(zone, NR_ALLOC_BATCH) == 0 &&
+           !zone_is_fair_depleted(zone))
+               zone_set_flag(zone, ZONE_FAIR_DEPLETED);

        __count_zone_vm_events(PGALLOC, zone, 1 << order);
        zone_statistics(preferred_zone, zone, gfp_flags);
@@ -1966,8 +1985,10 @@ zonelist_scan:
                if (alloc_flags & ALLOC_FAIR) {
                        if (!zone_local(preferred_zone, zone))
                                break;
-                       if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0)
+                       if (zone_is_fair_depleted(zone)) {
+                               nr_fair_skipped++;
                                continue;
+                       }
                }

A <= check was replaced with a ==. On SMP it doesn't matter because
negative values are returned as zero due to per-CPU drift which is not
possible in the UP case. Vlastimil Babka correctly pointed out that this
can wrap negative due to high-order allocations.

However, Leon Romanovsky pointed out that a <= check on zone_page_state
was never correct as zone_page_state returns unsigned long so the root
cause of the breakage was the <= check in the first place.

zone_page_state is an API hazard because of the difference in behaviour
between SMP and UP is very surprising. There is a good reason to allow
NR_ALLOC_BATCH to go negative -- when the counter is reset the negative
value takes recent activity into account. This patch makes zone_page_state
behave the same on SMP and UP as saving one branch on UP is not likely to
make a measurable performance difference.

Reported-by: Vlastimil Babka <vbabka@xxxxxxx>
Reported-by: Leon Romanovsky <leon@xxxxxxx>
Signed-off-by: Mel Gorman <mgorman@xxxxxxx>
---
 include/linux/vmstat.h | 2 --
 1 file changed, 2 deletions(-)

diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
index 82e7db7..cece0f0 100644
--- a/include/linux/vmstat.h
+++ b/include/linux/vmstat.h
@@ -131,10 +131,8 @@ static inline unsigned long zone_page_state(struct zone *zone,
                                        enum zone_stat_item item)
 {
        long x = atomic_long_read(&zone->vm_stat[item]);
-#ifdef CONFIG_SMP
        if (x < 0)
                x = 0;
-#endif
        return x;
 }

Since you are changing vmstat.h, what do you think about change in all similiar places?

diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
index 82e7db7..88d3d3e 100644
--- a/include/linux/vmstat.h
+++ b/include/linux/vmstat.h
@@ -120,10 +120,8 @@ static inline void zone_page_state_add(long x, struct zone *zone,
 static inline unsigned long global_page_state(enum zone_stat_item item)
 {
        long x = atomic_long_read(&vm_stat[item]);
-#ifdef CONFIG_SMP
        if (x < 0)
                x = 0;
-#endif
        return x;
 }
 
@@ -131,10 +129,8 @@ static inline unsigned long zone_page_state(struct zone *zone,
                                        enum zone_stat_item item)
 {
        long x = atomic_long_read(&zone->vm_stat[item]);
-#ifdef CONFIG_SMP
        if (x < 0)
                x = 0;
-#endif
        return x;
 }
 
@@ -153,10 +149,9 @@ static inline unsigned long zone_page_state_snapshot(struct zone *zone,
        int cpu;
        for_each_online_cpu(cpu)
                x += per_cpu_ptr(zone->pageset, cpu)->vm_stat_diff[item];
-
+#endif
        if (x < 0)
                x = 0;
-#endif
        return x;
 }

--

Leon Romanovsky | Independent Linux Consultant

        www.leon.nu | leon@xxxxxxx