Skip to contentSkip to conversation

cpuset: Make cpuset hotplug synchronous

PATCH Applied Prateek Sood 3 participants 6 messages Jan 24, 2020 – Feb 12, 2020

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cpuset: Make cpuset hotplug synchronous

Hi Tejun & Peter,

It seems that after following patch we can make cpuset_hotplug_workfn synchronous:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/kernel/cgroup/cpuset.c?h=v5.5-rc7&id=d74b27d63a8bebe2fe634944e4ebdc7b10db7a39

Could you please share your opinion on the same and below patch.

Thanks,
Prateek

Quoted 1 line

8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8

Convert cpuset_hotplug_workfn() into synchronous call for cpu hotplug path. For memory hotplug path it still gets queued as a work item.

Since cpuset_hotplug_workfn() can be made synchronous for cpu hotplug path, it is not required to wait for cpuset hotplug while thawing processes.

---
 include/linux/cpuset.h |  3 ---
 kernel/cgroup/cpuset.c | 31 +++++++++++++++++++------------
 kernel/power/process.c |  2 --
 3 files changed, 19 insertions(+), 17 deletions(-)
include/linux/cpuset.h +0 -3
index 04c20de66..cede4cb 100644
@@ -54,7 +54,6 @@ static inline void cpuset_dec(void)
 extern void cpuset_init_smp(void);
 extern void cpuset_force_rebuild(void);
 extern void cpuset_update_active_cpus(void);
-extern void cpuset_wait_for_hotplug(void);
 extern void cpuset_read_lock(void);
 extern void cpuset_read_unlock(void);
 extern void cpuset_cpus_allowed(struct task_struct *p, struct cpumask *mask);
@@ -176,8 +175,6 @@ static inline void cpuset_update_active_cpus(void)
 	partition_sched_domains(1, NULL, NULL);
 }
 
-static inline void cpuset_wait_for_hotplug(void) { }
-
 static inline void cpuset_read_lock(void) { }
 static inline void cpuset_read_unlock(void) { }
 
kernel/cgroup/cpuset.c +19 -12
index 58f5073..cafd4d2 100644
@@ -3101,7 +3101,7 @@ static void cpuset_hotplug_update_tasks(struct cpuset *cs, struct tmpmasks *tmp)
 }
 
 /**
- * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset
+ * cpuset_hotplug - 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
@@ -3116,7 +3116,7 @@ static void cpuset_hotplug_update_tasks(struct cpuset *cs, struct tmpmasks *tmp)
  * Note that CPU offlining during suspend is ignored.  We don't modify
  * cpusets across suspend/resume cycles at all.
  */
-static void cpuset_hotplug_workfn(struct work_struct *work)
+static void cpuset_hotplug(bool use_cpu_hp_lock)
 {
 	static cpumask_t new_cpus;
 	static nodemask_t new_mems;
@@ -3201,25 +3201,32 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
 	/* rebuild sched domains if cpus_allowed has changed */
 	if (cpus_updated || force_rebuild) {
 		force_rebuild = false;
-		rebuild_sched_domains();
+		if (use_cpu_hp_lock)
+			rebuild_sched_domains();
+		else {
+			/* Acquiring cpu_hotplug_lock is not required.
+			 * When cpuset_hotplug() is called in hotplug path,
+			 * cpu_hotplug_lock is held by the hotplug context
+			 * which is waiting for cpuhp_thread_fun to indicate
+			 * completion of callback.
+			*/
+			percpu_down_write(&cpuset_rwsem);
+			rebuild_sched_domains_locked();
+			percpu_up_write(&cpuset_rwsem);
+		}
 	}
 
 	free_cpumasks(NULL, ptmp);
 }
 
-void cpuset_update_active_cpus(void)
+static void cpuset_hotplug_workfn(struct work_struct *work)
 {
-	/*
-	 * 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.
-	 */
-	schedule_work(&cpuset_hotplug_work);
+	cpuset_hotplug(true);
 }
 
-void cpuset_wait_for_hotplug(void)
+void cpuset_update_active_cpus(void)
 {
-	flush_work(&cpuset_hotplug_work);
+	cpuset_hotplug(false);
 }
 
 /*
kernel/power/process.c +0 -2
index 4b6a54d..08f7019 100644
@@ -204,8 +204,6 @@ void thaw_processes(void)
 	__usermodehelper_set_disable_depth(UMH_FREEZING);
 	thaw_workqueues();
 
-	cpuset_wait_for_hotplug();
-
 	read_lock(&tasklist_lock);
 	for_each_process_thread(g, p) {
 		/* No other threads should have PF_SUSPEND_TASK set */
Signature · 2 lines
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc., 
is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.
#1 lore raw

Hi Tejun & Peter,

Could you please share your feedback on this patch for making

cpuset_hotplug_workfn synchronous.

Thanks,

Prateek

Quoted 126 lines · Prateek Sood

Hi Tejun & Peter,

It seems that after following patch we can make cpuset_hotplug_workfn synchronous:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/kernel/cgroup/cpuset.c?h=v5.5-rc7&id=d74b27d63a8bebe2fe634944e4ebdc7b10db7a39

Could you please share your opinion on the same and below patch.

Thanks,
Prateek

Quoted 1 line

8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8>8

Convert cpuset_hotplug_workfn() into synchronous call for cpu hotplug path. For memory hotplug path it still gets queued as a work item.

Since cpuset_hotplug_workfn() can be made synchronous for cpu hotplug path, it is not required to wait for cpuset hotplug while thawing processes.

---
  include/linux/cpuset.h |  3 ---
  kernel/cgroup/cpuset.c | 31 +++++++++++++++++++------------
  kernel/power/process.c |  2 --
  3 files changed, 19 insertions(+), 17 deletions(-)
include/linux/cpuset.h +0 -3
index 04c20de66..cede4cb 100644
@@ -54,7 +54,6 @@ static inline void cpuset_dec(void)
  extern void cpuset_init_smp(void);
  extern void cpuset_force_rebuild(void);
  extern void cpuset_update_active_cpus(void);
-extern void cpuset_wait_for_hotplug(void);
  extern void cpuset_read_lock(void);
  extern void cpuset_read_unlock(void);
  extern void cpuset_cpus_allowed(struct task_struct *p, struct cpumask *mask);
@@ -176,8 +175,6 @@ static inline void cpuset_update_active_cpus(void)
  	partition_sched_domains(1, NULL, NULL);
  }
 
-static inline void cpuset_wait_for_hotplug(void) { }
-
  static inline void cpuset_read_lock(void) { }
  static inline void cpuset_read_unlock(void) { }
 
kernel/cgroup/cpuset.c +19 -12
index 58f5073..cafd4d2 100644
@@ -3101,7 +3101,7 @@ static void cpuset_hotplug_update_tasks(struct cpuset *cs, struct tmpmasks *tmp)
  }
 
  /**
- * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset
+ * cpuset_hotplug - 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
@@ -3116,7 +3116,7 @@ static void cpuset_hotplug_update_tasks(struct cpuset *cs, struct tmpmasks *tmp)
   * Note that CPU offlining during suspend is ignored.  We don't modify
   * cpusets across suspend/resume cycles at all.
   */
-static void cpuset_hotplug_workfn(struct work_struct *work)
+static void cpuset_hotplug(bool use_cpu_hp_lock)
  {
  	static cpumask_t new_cpus;
  	static nodemask_t new_mems;
@@ -3201,25 +3201,32 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
  	/* rebuild sched domains if cpus_allowed has changed */
  	if (cpus_updated || force_rebuild) {
  		force_rebuild = false;
-		rebuild_sched_domains();
+		if (use_cpu_hp_lock)
+			rebuild_sched_domains();
+		else {
+			/* Acquiring cpu_hotplug_lock is not required.
+			 * When cpuset_hotplug() is called in hotplug path,
+			 * cpu_hotplug_lock is held by the hotplug context
+			 * which is waiting for cpuhp_thread_fun to indicate
+			 * completion of callback.
+			*/
+			percpu_down_write(&cpuset_rwsem);
+			rebuild_sched_domains_locked();
+			percpu_up_write(&cpuset_rwsem);
+		}
  	}
 
  	free_cpumasks(NULL, ptmp);
  }
 
-void cpuset_update_active_cpus(void)
+static void cpuset_hotplug_workfn(struct work_struct *work)
  {
-	/*
-	 * 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.
-	 */
-	schedule_work(&cpuset_hotplug_work);
+	cpuset_hotplug(true);
  }
 
-void cpuset_wait_for_hotplug(void)
+void cpuset_update_active_cpus(void)
  {
-	flush_work(&cpuset_hotplug_work);
+	cpuset_hotplug(false);
  }
 
  /*
kernel/power/process.c +0 -2
index 4b6a54d..08f7019 100644
@@ -204,8 +204,6 @@ void thaw_processes(void)
  	__usermodehelper_set_disable_depth(UMH_FREEZING);
  	thaw_workqueues();
 
-	cpuset_wait_for_hotplug();
-
  	read_lock(&tasklist_lock);
  	for_each_process_thread(g, p) {
  		/* No other threads should have PF_SUSPEND_TASK set */
Signature · 3 lines
Qualcomm India Private Limited, on behalf of Qualcomm Innovation
Center, Inc., is a member of Code Aurora Forum, a Linux Foundation
Collaborative Project
maintainer
#2 lore raw
Quoted 5 lines · Prateek Sood

Hi Tejun & Peter,

Could you please share your feedback on this patch for making

cpuset_hotplug_workfn synchronous.

Hey, IIRC, we went back forth with this several times. Unless there's a pressing reason to change, I'd rather leave it alone.

Thanks.

Signature · 1 line
tejun
#3 lore raw
Quoted 9 lines · Tejun Heo
Quoted 5 lines · Prateek Sood

Hi Tejun & Peter,

Could you please share your feedback on this patch for making

cpuset_hotplug_workfn synchronous.

Hey, IIRC, we went back forth with this several times. Unless there's a pressing reason to change, I'd rather leave it alone.

I think I have a similar patch in one of the many in-progress series I keep around; IIRC it is the one where I add per-cpu support to SCHED_DEADLINE. The hotplug accounting becomes easier if we do this.

Sadly this series is fairly low on the todo list atm though :/ I have no problems keeping it around until such a time though.

maintainer
#4 lore raw

Hey, Peter.

Quoted 6 lines · Peter Zijlstra

I think I have a similar patch in one of the many in-progress series I keep around; IIRC it is the one where I add per-cpu support to SCHED_DEADLINE. The hotplug accounting becomes easier if we do this.

Sadly this series is fairly low on the todo list atm though :/ I have no problems keeping it around until such a time though.

I see. I don't have anything against the change except that there doesn't seem to be a good reason right now. If it's something which can be useful for planned changes, might as well apply it and let it cook upstream. Will apply to for-5.7.

Thanks.

Signature · 1 line
tejun
maintainer Applied
#5 lore raw
In reply to Prateek Sood
Quoted 8 lines · Prateek Sood

Convert cpuset_hotplug_workfn() into synchronous call for cpu hotplug path. For memory hotplug path it still gets queued as a work item.

Since cpuset_hotplug_workfn() can be made synchronous for cpu hotplug path, it is not required to wait for cpuset hotplug while thawing processes.

Applied to cgroup/for-5.7.

Thanks.

Signature · 1 line
tejun