Message ID | 20200720052348.428564-1-alex.shi@linux.alibaba.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | mm/vmstat: don't do count if no needs | expand |
From 4cb977d34227e4bafa95b8da5e47dbd8b6141d26 Mon Sep 17 00:00:00 2001 From: Alex Shi <alex.shi@linux.alibaba.com> Date: Mon, 20 Jul 2020 11:50:12 +0800 Subject: [PATCH v2] mm/vmstat: don't do count if no needs For couple of vmstat account funcs, the caller usually doesn't check the delta value, if delta == 0, irq or atomic operator is a waste. That's better to be skipped, also add unlikey() since !delta is less happened. Signed-off-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: linux-mm@kvack.org Cc: linux-kernel@vger.kernel.org --- mm/vmstat.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/mm/vmstat.c b/mm/vmstat.c index 3fb23a21f6dd..b02d99dff443 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -321,6 +321,9 @@ void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, long x; long t; + if (unlikely(!delta)) + return; + x = delta + __this_cpu_read(*p); t = __this_cpu_read(pcp->stat_threshold); @@ -341,6 +344,9 @@ void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, long x; long t; + if (unlikely(!delta)) + return; + x = delta + __this_cpu_read(*p); t = __this_cpu_read(pcp->stat_threshold); @@ -484,6 +490,9 @@ static inline void mod_zone_state(struct zone *zone, s8 __percpu *p = pcp->vm_stat_diff + item; long o, n, t, z; + if (unlikely(!delta)) + return; + do { z = 0; /* overflow to zone counters */ @@ -518,6 +527,9 @@ static inline void mod_zone_state(struct zone *zone, void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, long delta) { + if (unlikely(!delta)) + return; + mod_zone_state(zone, item, delta, 0); } EXPORT_SYMBOL(mod_zone_page_state); @@ -541,6 +553,9 @@ static inline void mod_node_state(struct pglist_data *pgdat, s8 __percpu *p = pcp->vm_node_stat_diff + item; long o, n, t, z; + if (unlikely(!delta)) + return; + do { z = 0; /* overflow to node counters */ @@ -604,6 +619,9 @@ void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, { unsigned long flags; + if (unlikely(!delta)) + return; + local_irq_save(flags); __mod_zone_page_state(zone, item, delta); local_irq_restore(flags); @@ -647,6 +665,9 @@ void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, { unsigned long flags; + if (unlikely(!delta)) + return; + local_irq_save(flags); __mod_node_page_state(pgdat, item, delta); local_irq_restore(flags);
On Mon, 20 Jul 2020 13:42:45 +0800 Alex Shi <alex.shi@linux.alibaba.com> wrote: > > For couple of vmstat account funcs, the caller usually doesn't check the > delta value, if delta == 0, irq or atomic operator is a waste. That's > better to be skipped, also add unlikey() since !delta is less happened. > > ... > > --- a/mm/vmstat.c > +++ b/mm/vmstat.c > @@ -321,6 +321,9 @@ void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, > long x; > long t; > > + if (unlikely(!delta)) > + return; > + Do we know that delta==0 is frequent enough to make the change a net benefit?
在 2020/7/21 上午7:14, Andrew Morton 写道: >> --- a/mm/vmstat.c >> +++ b/mm/vmstat.c >> @@ -321,6 +321,9 @@ void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, >> long x; >> long t; >> >> + if (unlikely(!delta)) >> + return; >> + > Do we know that delta==0 is frequent enough to make the change a net > benefit? Uh,I check again. it seems no much place delta==0, forget this patch. Thanks Alex
diff --git a/mm/vmstat.c b/mm/vmstat.c index 3fb23a21f6dd..91f28146daa7 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -321,6 +321,9 @@ void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item, long x; long t; + if (!delta) + return; + x = delta + __this_cpu_read(*p); t = __this_cpu_read(pcp->stat_threshold); @@ -341,6 +344,9 @@ void __mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, long x; long t; + if (!delta) + return; + x = delta + __this_cpu_read(*p); t = __this_cpu_read(pcp->stat_threshold); @@ -484,6 +490,9 @@ static inline void mod_zone_state(struct zone *zone, s8 __percpu *p = pcp->vm_stat_diff + item; long o, n, t, z; + if (!delta) + return; + do { z = 0; /* overflow to zone counters */ @@ -518,6 +527,9 @@ static inline void mod_zone_state(struct zone *zone, void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, long delta) { + if (!delta) + return; + mod_zone_state(zone, item, delta, 0); } EXPORT_SYMBOL(mod_zone_page_state); @@ -541,6 +553,9 @@ static inline void mod_node_state(struct pglist_data *pgdat, s8 __percpu *p = pcp->vm_node_stat_diff + item; long o, n, t, z; + if (!delta) + return; + do { z = 0; /* overflow to node counters */ @@ -581,6 +596,9 @@ EXPORT_SYMBOL(mod_node_page_state); void inc_node_state(struct pglist_data *pgdat, enum node_stat_item item) { + if (!delta) + return; + mod_node_state(pgdat, item, 1, 1); } @@ -604,6 +622,9 @@ void mod_zone_page_state(struct zone *zone, enum zone_stat_item item, { unsigned long flags; + if (!delta) + return; + local_irq_save(flags); __mod_zone_page_state(zone, item, delta); local_irq_restore(flags); @@ -647,6 +668,9 @@ void mod_node_page_state(struct pglist_data *pgdat, enum node_stat_item item, { unsigned long flags; + if (!delta) + return; + local_irq_save(flags); __mod_node_page_state(pgdat, item, delta); local_irq_restore(flags);
For couple of vmstat account funcs, the caller usually doesn't check the delta value, if delta == 0, irq or atomic operator is a waste. That's better to be skipped. Signed-off-by: Alex Shi <alex.shi@linux.alibaba.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: linux-mm@kvack.org Cc: linux-kernel@vger.kernel.org --- mm/vmstat.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)