diff mbox series

[X,1/4] stop_machine, sched: Fix migrate_swap() vs. active_balance() deadlock

Message ID 20190321234412.11113-2-mfo@canonical.com
State New
Headers show
Series LP#1821259 Fix for deadlock in cpu_stopper | expand

Commit Message

Mauricio Faria de Oliveira March 21, 2019, 11:44 p.m. UTC 
From: Peter Zijlstra <peterz@infradead.org>

BugLink: https://bugs.launchpad.net/bugs/1821259

Matt reported the following deadlock:

CPU0					CPU1

schedule(.prev=migrate/0)		<fault>
  pick_next_task()			  ...
    idle_balance()			    migrate_swap()
      active_balance()			      stop_two_cpus()
						spin_lock(stopper0->lock)
						spin_lock(stopper1->lock)
						ttwu(migrate/0)
						  smp_cond_load_acquire() -- waits for schedule()
        stop_one_cpu(1)
	  spin_lock(stopper1->lock) -- waits for stopper lock

Fix this deadlock by taking the wakeups out from under stopper->lock.
This allows the active_balance() to queue the stop work and finish the
context switch, which in turn allows the wakeup from migrate_swap() to
observe the context and complete the wakeup.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reported-by: Matt Fleming <matt@codeblueprint.co.uk>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20180420095005.GH4064@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
(backported from commit 0b26351b910fb8fe6a056f8a1bbccabe50c0e19f)
[mfo: backport:
 - hunk 1:
   - refresh context lines
   - include 'linux/sched.h' instead of 'linux/sched/wake_q.h' which is
     code moved by upstream commit eb61baf69871 ("sched/headers: Move the
     wake-queue types and interfaces from sched.h into <linux/sched/wake_q.h>")
 - hunk 2:
   - refresh context lines
   - s/bool/void/ return type of cpu_stop_queue_work() and other 2 changes
     which are not relevant to this fix, due to lack of upstream commits:
     - 'enabled' variable:
       commit 1b034bd989aa ("stop_machine: Make cpu_stop_queue_work() and
       stop_one_cpu_nowait() return bool")
     - else 'if (work->done)' condition
       commit dd2e3121e3cb ("stop_machine: Shift the 'done != NULL' check
       from cpu_stop_signal_done() to callers")
   - s/DEFINE_WAKE_Q/WAKE_Q/ due to lack of upstream commit 194a6b5b9cb6
     ("sched/wake_q: Rename WAKE_Q to DEFINE_WAKE_Q")
 - hunk 3:
   - refresh context lines
   - s/DEFINE_WAKE_Q/WAKE_Q/ due to lack of upstream commit 194a6b5b9cb6
     ("sched/wake_q: Rename WAKE_Q to DEFINE_WAKE_Q")
 - hunk 4:
   - refresh context lines
 - hunk 5:
   - merge with hunk 4
   - refresh context lines]
Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com>
---
 kernel/stop_machine.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)
diff mbox series

Patch

diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c
index a3bbaee77c58..71435be8bd25 100644
--- a/kernel/stop_machine.c
+++ b/kernel/stop_machine.c
@@ -21,6 +21,7 @@ 
 #include <linux/smpboot.h>
 #include <linux/atomic.h>
 #include <linux/lglock.h>
+#include <linux/sched.h>
 
 /*
  * Structure to determine completion condition and record errors.  May
@@ -74,24 +75,28 @@  static void cpu_stop_signal_done(struct cpu_stop_done *done, bool executed)
 }
 
 static void __cpu_stop_queue_work(struct cpu_stopper *stopper,
-					struct cpu_stop_work *work)
+					struct cpu_stop_work *work,
+					struct wake_q_head *wakeq)
 {
 	list_add_tail(&work->list, &stopper->works);
-	wake_up_process(stopper->thread);
+	wake_q_add(wakeq, stopper->thread);
 }
 
 /* queue @work to @stopper.  if offline, @work is completed immediately */
 static void cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work)
 {
 	struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
+	WAKE_Q(wakeq);
 	unsigned long flags;
 
 	spin_lock_irqsave(&stopper->lock, flags);
 	if (stopper->enabled)
-		__cpu_stop_queue_work(stopper, work);
+		__cpu_stop_queue_work(stopper, work, &wakeq);
 	else
 		cpu_stop_signal_done(work->done, false);
 	spin_unlock_irqrestore(&stopper->lock, flags);
+
+	wake_up_q(&wakeq);
 }
 
 /**
@@ -221,6 +226,7 @@  static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1,
 {
 	struct cpu_stopper *stopper1 = per_cpu_ptr(&cpu_stopper, cpu1);
 	struct cpu_stopper *stopper2 = per_cpu_ptr(&cpu_stopper, cpu2);
+	WAKE_Q(wakeq);
 	int err;
 
 	lg_double_lock(&stop_cpus_lock, cpu1, cpu2);
@@ -232,13 +238,15 @@  static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1,
 		goto unlock;
 
 	err = 0;
-	__cpu_stop_queue_work(stopper1, work1);
-	__cpu_stop_queue_work(stopper2, work2);
+	__cpu_stop_queue_work(stopper1, work1, &wakeq);
+	__cpu_stop_queue_work(stopper2, work2, &wakeq);
 unlock:
 	spin_unlock(&stopper2->lock);
 	spin_unlock_irq(&stopper1->lock);
 	lg_double_unlock(&stop_cpus_lock, cpu1, cpu2);
 
+	wake_up_q(&wakeq);
+
 	return err;
 }
 /**