[PATCH -mm] cpuset,mm: make the write-side sleep if the read-side is not running

[Miao Xie <miaox@cn.fujitsu.com>]

on 2010-5-12 12:32, Andrew Morton wrote:
> On Wed, 12 May 2010 15:20:51 +0800 Miao Xie <miaox@cn.fujitsu.com> wrote:
> 
>> @@ -985,6 +984,7 @@ repeat:
>>  	 * for the read-side.
>>  	 */
>>  	while (ACCESS_ONCE(tsk->mems_allowed_change_disable)) {
>> +		task_unlock(tsk);
>>  		if (!task_curr(tsk))
>>  			yield();
>>  		goto repeat;
> 
> Oh, I meant to mention that.  No yield()s, please.  Their duration is
> highly unpredictable.  Can we do something more deterministic here?

根据Andrew的指摘，我做了下面的patch，如果读端的进程处于内存分配过程中但不在运行，
则让负责修改tsk->mems_allowed的进程睡眠，等待读端结束内存分配操作。

diff --git a/include/linux/cpuset.h b/include/linux/cpuset.h
index 457ed76..d348c47 100644
--- a/include/linux/cpuset.h
+++ b/include/linux/cpuset.h
@@ -117,7 +117,11 @@ static inline void put_mems_allowed(void)
 	 * nodemask.
 	 */
 	smp_mb();
-	--ACCESS_ONCE(current->mems_allowed_change_disable);
+	if (!--ACCESS_ONCE(current->mems_allowed_change_disable)
+	    && unlikely(current->mems_read_done)) {
+		complete(current->mems_read_done);
+		current->mems_read_done = NULL;
+	}
 }
 
 static inline void set_mems_allowed(nodemask_t nodemask)
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 66620fa..8699900 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1423,6 +1423,8 @@ struct task_struct {
 #ifdef CONFIG_CPUSETS
 	nodemask_t mems_allowed;	/* Protected by alloc_lock */
 	int mems_allowed_change_disable;
+	/* for changing mems_allowed and mempolicy */
+	struct completion *mems_read_done;
 	int cpuset_mem_spread_rotor;
 	int cpuset_slab_spread_rotor;
 #endif
@@ -2525,6 +2527,12 @@ static inline void inc_syscw(struct task_struct *tsk)
 extern void task_oncpu_function_call(struct task_struct *p,
 				     void (*func) (void *info), void *info);
 
+/*
+ * Call the function if the target task is not executing right now
+ */
+extern void task_notcurr_function_call(struct task_struct *p,
+				     void (*func) (void *info), void *info);
+
 
 #ifdef CONFIG_MM_OWNER
 extern void mm_update_next_owner(struct mm_struct *mm);
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index d243a22..a471ab2 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -938,6 +938,20 @@ static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
 	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
 }
 
+struct cpuset_task_info {
+	struct task_struct *tsk;
+	struct completion done;
+	int ret;
+};
+
+void set_mems_read_done_for_task(void *_info)
+{
+	struct cpuset_task_info *info = _info;
+
+	info->tsk->mems_read_done = &info->done;
+	info->ret = 1;
+}
+
 /*
  * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy
  * @tsk: the task to change
@@ -950,6 +964,8 @@ static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
 static void cpuset_change_task_nodemask(struct task_struct *tsk,
 					nodemask_t *newmems)
 {
+	struct cpuset_task_info info;
+
 repeat:
 	/*
 	 * Allow tasks that have access to memory reserves because they have
@@ -980,13 +996,23 @@ repeat:
 	smp_mb();
 
 	/*
-	 * Allocating of memory is very fast, we needn't sleep when waitting
-	 * for the read-side.
+	 * If the read-side is running, we needn't sleep when waiting for the
+	 * read-side because allocating page is very fast.
 	 */
 	while (ACCESS_ONCE(tsk->mems_allowed_change_disable)) {
 		task_unlock(tsk);
-		if (!task_curr(tsk))
-			yield();
+		if (!task_curr(tsk)) {
+			info.tsk = tsk;
+			init_completion(&info.done);
+			info.ret = 0;
+
+			task_notcurr_function_call(tsk,
+						   set_mems_read_done_for_task,
+						   &info);
+			if (info.ret)
+				wait_for_completion(&info.done);
+		} else
+			cpu_relax();
 		goto repeat;
 	}
 
diff --git a/kernel/fork.c b/kernel/fork.c
index f4f0951..76a6ec8 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1090,6 +1090,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 #ifdef CONFIG_CPUSETS
 	p->cpuset_mem_spread_rotor = node_random(p->mems_allowed);
 	p->cpuset_slab_spread_rotor = node_random(p->mems_allowed);
+	p->mems_read_done = NULL;
 #endif
 #ifdef CONFIG_TRACE_IRQFLAGS
 	p->irq_events = 0;
diff --git a/kernel/sched.c b/kernel/sched.c
index e298c71..f839f8f 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2217,6 +2217,35 @@ void task_oncpu_function_call(struct task_struct *p,
 	preempt_enable();
 }
 
+/**
+ * task_notcurr_function_call - call a function when a task isn't running
+ * @p:		the task to evaluate
+ * @func:	the function to be called
+ * @info:	the function call argument
+ *
+ * Calls the function @func when the task is not currently running.
+ */
+void task_notcurr_function_call(struct task_struct *p,
+			      void (*func) (void *info), void *info)
+{
+	struct rq *rq;
+	unsigned long flags;
+
+	if (p == current)
+		return;
+
+#ifdef CONFIG_SMP
+	rq = task_rq_lock(p, &flags);
+	if (!task_curr(p))
+		func(info);
+	task_rq_unlock(rq, &flags);
+#else
+	preempt_disable();
+	func(info);
+	preempt_enable();
+#endif
+}
+
 #ifdef CONFIG_SMP
 /*
  * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held.