Re: [PATCH 1/2] scalable rw_mutex

Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]
From: Andrew Morton
Date: Wednesday, May 16, 2007 - 4:28 pm

On Sat, 12 May 2007 11:06:24 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:


Like this:


From: Andrew Morton <akpm@linux-foundation.org>

per-cpu counters presently must iterate over all possible CPUs in the
exhaustive percpu_counter_sum().

But it can be much better to only iterate over the presently-online CPUs.  To
do this, we must arrange for an offlined CPU's count to be spilled into the
counter's central count.

We can do this for all percpu_counters in the machine by linking them into a
single global list and walking that list at CPU_DEAD time.

(I hope.  Might have race windows in which the percpu_counter_sum() count is
inaccurate?)


Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 include/linux/percpu_counter.h |   18 ++------
 lib/percpu_counter.c           |   66 +++++++++++++++++++++++++++++++
 2 files changed, 72 insertions(+), 12 deletions(-)

diff -puN lib/percpu_counter.c~percpu_counters-use-cpu-notifiers lib/percpu_counter.c
--- a/lib/percpu_counter.c~percpu_counters-use-cpu-notifiers
+++ a/lib/percpu_counter.c
@@ -3,8 +3,17 @@
  */
 
 #include <linux/percpu_counter.h>
+#include <linux/notifier.h>
+#include <linux/mutex.h>
+#include <linux/init.h>
+#include <linux/cpu.h>
 #include <linux/module.h>
 
+#ifdef CONFIG_HOTPLUG_CPU
+static LIST_HEAD(percpu_counters);
+static DEFINE_MUTEX(percpu_counters_lock);
+#endif
+
 void percpu_counter_mod(struct percpu_counter *fbc, s32 amount)
 {
 	long count;
@@ -44,3 +53,60 @@ s64 percpu_counter_sum(struct percpu_cou
 	return ret < 0 ? 0 : ret;
 }
 EXPORT_SYMBOL(percpu_counter_sum);
+
+void percpu_counter_init(struct percpu_counter *fbc, s64 amount)
+{
+	spin_lock_init(&fbc->lock);
+	fbc->count = amount;
+	fbc->counters = alloc_percpu(s32);
+#ifdef CONFIG_HOTPLUG_CPU
+	mutex_lock(&percpu_counters_lock);
+	list_add(&fbc->list, &percpu_counters);
+	mutex_unlock(&percpu_counters_lock);
+#endif
+}
+EXPORT_SYMBOL(percpu_counter_init);
+
+void percpu_counter_destroy(struct percpu_counter *fbc)
+{
+	free_percpu(fbc->counters);
+#ifdef CONFIG_HOTPLUG_CPU
+	mutex_lock(&percpu_counters_lock);
+	list_del(&fbc->list);
+	mutex_unlock(&percpu_counters_lock);
+#endif
+}
+EXPORT_SYMBOL(percpu_counter_destroy);
+
+#ifdef CONFIG_HOTPLUG_CPU
+static int __cpuinit percpu_counter_hotcpu_callback(struct notifier_block *nb,
+					unsigned long action, void *hcpu)
+{
+	unsigned int cpu;
+	struct percpu_counter *fbc;
+
+	if (action != CPU_DEAD)
+		return NOTIFY_OK;
+
+	cpu = (unsigned long)hcpu;
+	mutex_lock(&percpu_counters_lock);
+	list_for_each_entry(fbc, &percpu_counters, list) {
+		s32 *pcount;
+
+		spin_lock(&fbc->lock);
+		pcount = per_cpu_ptr(fbc->counters, cpu);
+		fbc->count += *pcount;
+		*pcount = 0;
+		spin_unlock(&fbc->lock);
+	}
+	mutex_unlock(&percpu_counters_lock);
+	return NOTIFY_OK;
+}
+
+static int __init percpu_counter_startup(void)
+{
+	hotcpu_notifier(percpu_counter_hotcpu_callback, 0);
+	return 0;
+}
+module_init(percpu_counter_startup);
+#endif
diff -puN include/linux/percpu.h~percpu_counters-use-cpu-notifiers include/linux/percpu.h
diff -puN include/linux/percpu_counter.h~percpu_counters-use-cpu-notifiers include/linux/percpu_counter.h
--- a/include/linux/percpu_counter.h~percpu_counters-use-cpu-notifiers
+++ a/include/linux/percpu_counter.h
@@ -8,6 +8,7 @@
 
 #include <linux/spinlock.h>
 #include <linux/smp.h>
+#include <linux/list.h>
 #include <linux/threads.h>
 #include <linux/percpu.h>
 #include <linux/types.h>
@@ -17,6 +18,9 @@
 struct percpu_counter {
 	spinlock_t lock;
 	s64 count;
+#ifdef CONFIG_HOTPLUG_CPU
+	struct list_head list;	/* All percpu_counters are on a list */
+#endif
 	s32 *counters;
 };
 
@@ -26,18 +30,8 @@ struct percpu_counter {
 #define FBC_BATCH	(NR_CPUS*4)
 #endif
 
-static inline void percpu_counter_init(struct percpu_counter *fbc, s64 amount)
-{
-	spin_lock_init(&fbc->lock);
-	fbc->count = amount;
-	fbc->counters = alloc_percpu(s32);
-}
-
-static inline void percpu_counter_destroy(struct percpu_counter *fbc)
-{
-	free_percpu(fbc->counters);
-}
-
+void percpu_counter_init(struct percpu_counter *fbc, s64 amount);
+void percpu_counter_destroy(struct percpu_counter *fbc);
 void percpu_counter_mod(struct percpu_counter *fbc, s32 amount);
 s64 percpu_counter_sum(struct percpu_counter *fbc);
 
_

and then this:


From: Andrew Morton <akpm@linux-foundation.org>

Now that we have implemented hotunplug-time counter spilling,
percpu_counter_sum() only needs to look at online CPus.


Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 lib/percpu_counter.c |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)

diff -puN lib/percpu_counter.c~percpu_counters-use-for_each_online_cpu lib/percpu_counter.c
--- a/lib/percpu_counter.c~percpu_counters-use-for_each_online_cpu
+++ a/lib/percpu_counter.c
@@ -45,7 +45,7 @@ s64 percpu_counter_sum(struct percpu_cou
 
 	spin_lock(&fbc->lock);
 	ret = fbc->count;
-	for_each_possible_cpu(cpu) {
+	for_each_online_cpu(cpu) {
 		s32 *pcount = per_cpu_ptr(fbc->counters, cpu);
 		ret += *pcount;
 	}
_

-
Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]

Messages in current thread:
[PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Fri May 11, 6:15 am)
Re: [PATCH 1/2] scalable rw_mutex, Christoph Hellwig, (Fri May 11, 7:03 am)
Re: [PATCH 1/2] scalable rw_mutex, Andrew Morton, (Fri May 11, 9:31 am)
Re: [PATCH 1/2] scalable rw_mutex, Christoph Lameter, (Fri May 11, 10:07 am)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Fri May 11, 10:57 am)
Re: [PATCH 1/2] scalable rw_mutex, Andrew Morton, (Fri May 11, 11:05 am)
Re: [PATCH 1/2] scalable rw_mutex, Oleg Nesterov, (Fri May 11, 4:00 pm)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Sat May 12, 12:39 am)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Sat May 12, 6:41 am)
Re: [PATCH 1/2] scalable rw_mutex, Oleg Nesterov, (Sat May 12, 9:04 am)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Sat May 12, 9:57 am)
Re: [PATCH 1/2] scalable rw_mutex, Oleg Nesterov, (Sat May 12, 11:03 am)
Re: [PATCH 1/2] scalable rw_mutex, Andrew Morton, (Sat May 12, 11:06 am)
Re: [PATCH 1/2] scalable rw_mutex, Andrew Morton, (Sat May 12, 11:11 am)
Re: [PATCH 1/2] scalable rw_mutex, Oleg Nesterov, (Sat May 12, 11:12 am)
Re: [PATCH 1/2] scalable rw_mutex, Andi Kleen, (Sat May 12, 11:55 am)
Re: [PATCH 1/2] scalable rw_mutex, Andi Kleen, (Sat May 12, 12:21 pm)
Re: [PATCH 1/2] scalable rw_mutex, Oleg Nesterov, (Sat May 12, 2:42 pm)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Mon May 14, 3:59 am)
Re: [PATCH 1/2] scalable rw_mutex, Nick Piggin, (Mon May 14, 4:36 am)
Re: [PATCH 1/2] scalable rw_mutex, Paul E. McKenney, (Mon May 14, 5:36 pm)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Tue May 15, 12:43 am)
Re: [PATCH 1/2] scalable rw_mutex, Paul E. McKenney, (Tue May 15, 8:29 am)
Re: [PATCH 1/2] scalable rw_mutex, Peter Zijlstra, (Tue May 15, 9:17 am)
Re: [PATCH 1/2] scalable rw_mutex, Paul E. McKenney, (Tue May 15, 11:52 am)
Re: [PATCH 1/2] scalable rw_mutex, Andrew Morton, (Wed May 16, 4:28 pm)
Re: [PATCH 1/2] scalable rw_mutex, Christoph Lameter, (Wed May 16, 4:40 pm)
Re: [PATCH 1/2] scalable rw_mutex, Andrew Morton, (Wed May 16, 5:24 pm)