From f44dfc440863562141d4c14e214472a9a5abafa2 Mon Sep 17 00:00:00 2001
From: Tejun Heo <tj@kernel.org>
Date: Mon, 7 Jan 2013 08:51:07 -0800
Subject: [PATCH] --- yaml --- r: 350343 b: refs/heads/master c:
 3a5a6d0c2b0391e159fa5bf1dddb9bf1f35178a0 h: refs/heads/master i:   350341:
 3239239548d6061ef600761bdb2054d9bb662876   350339:
 c428a0d4ec8e96e716f637137e56ee6240c560af   350335:
 1f00c4633bfeb95197737053cf2539a061ad5d8e v: v3

---
 [refs]                |  2 +-
 trunk/kernel/cpuset.c | 39 +++++++++++++++++++++++++++++++++++----
 2 files changed, 36 insertions(+), 5 deletions(-)

diff --git a/[refs] b/[refs]
index 09d49c7d8d51..2d75cda7da3e 100644
--- a/[refs]
+++ b/[refs]
@@ -1,2 +1,2 @@
 ---
-refs/heads/master: deb7aa308ea264b374d1db970174f5728a2faa27
+refs/heads/master: 3a5a6d0c2b0391e159fa5bf1dddb9bf1f35178a0
diff --git a/trunk/kernel/cpuset.c b/trunk/kernel/cpuset.c
index 3d448e646a4a..658eb1a32084 100644
--- a/trunk/kernel/cpuset.c
+++ b/trunk/kernel/cpuset.c
@@ -259,6 +259,13 @@ static char cpuset_name[CPUSET_NAME_LEN];
 static char cpuset_nodelist[CPUSET_NODELIST_LEN];
 static DEFINE_SPINLOCK(cpuset_buffer_lock);
 
+/*
+ * CPU / memory hotplug is handled asynchronously.
+ */
+static void cpuset_hotplug_workfn(struct work_struct *work);
+
+static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);
+
 /*
  * This is ugly, but preserves the userspace API for existing cpuset
  * users. If someone tries to mount the "cpuset" filesystem, we
@@ -1565,6 +1572,19 @@ static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft,
 	struct cpuset *cs = cgroup_cs(cgrp);
 	struct cpuset *trialcs;
 
+	/*
+	 * CPU or memory hotunplug may leave @cs w/o any execution
+	 * resources, in which case the hotplug code asynchronously updates
+	 * configuration and transfers all tasks to the nearest ancestor
+	 * which can execute.
+	 *
+	 * As writes to "cpus" or "mems" may restore @cs's execution
+	 * resources, wait for the previously scheduled operations before
+	 * proceeding, so that we don't end up keep removing tasks added
+	 * after execution capability is restored.
+	 */
+	flush_work(&cpuset_hotplug_work);
+
 	if (!cgroup_lock_live_group(cgrp))
 		return -ENODEV;
 
@@ -2095,7 +2115,7 @@ static void cpuset_propagate_hotplug(struct cpuset *cs)
 }
 
 /**
- * cpuset_handle_hotplug - handle CPU/memory hot[un]plug
+ * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset
  *
  * This function is called after either CPU or memory configuration has
  * changed and updates cpuset accordingly.  The top_cpuset is always
@@ -2110,7 +2130,7 @@ static void cpuset_propagate_hotplug(struct cpuset *cs)
  * Note that CPU offlining during suspend is ignored.  We don't modify
  * cpusets across suspend/resume cycles at all.
  */
-static void cpuset_handle_hotplug(void)
+static void cpuset_hotplug_workfn(struct work_struct *work)
 {
 	static cpumask_t new_cpus, tmp_cpus;
 	static nodemask_t new_mems, tmp_mems;
@@ -2177,7 +2197,18 @@ static void cpuset_handle_hotplug(void)
 
 void cpuset_update_active_cpus(bool cpu_online)
 {
-	cpuset_handle_hotplug();
+	/*
+	 * We're inside cpu hotplug critical region which usually nests
+	 * inside cgroup synchronization.  Bounce actual hotplug processing
+	 * to a work item to avoid reverse locking order.
+	 *
+	 * We still need to do partition_sched_domains() synchronously;
+	 * otherwise, the scheduler will get confused and put tasks to the
+	 * dead CPU.  Fall back to the default single domain.
+	 * cpuset_hotplug_workfn() will rebuild it as necessary.
+	 */
+	partition_sched_domains(1, NULL, NULL);
+	schedule_work(&cpuset_hotplug_work);
 }
 
 #ifdef CONFIG_MEMORY_HOTPLUG
@@ -2189,7 +2220,7 @@ void cpuset_update_active_cpus(bool cpu_online)
 static int cpuset_track_online_nodes(struct notifier_block *self,
 				unsigned long action, void *arg)
 {
-	cpuset_handle_hotplug();
+	schedule_work(&cpuset_hotplug_work);
 	return NOTIFY_OK;
 }
 #endif