sched: fix __set_task_cpu() SMP race

Grant Wilson has reported rare SCHED_FAIR_USER crashes on his quad-core
system, which crashes can only be explained via runqueue corruption.

there is a narrow SMP race in __set_task_cpu(): after ->cpu is set up to
a new value, task_rq_lock(p, ...) can be successfuly executed on another
CPU. We must ensure that updates of per-task data have been completed by
this moment.

this bug has been hiding in the Linux scheduler for an eternity (we never
had any explicit barrier for task->cpu in set_task_cpu() - so the bug was
introduced in 2.5.1), but only became visible via set_task_cfs_rq() being
accidentally put after the task->cpu update. It also probably needs a
sufficiently out-of-order CPU to trigger.

Reported-by: Grant Wilson <grant.wilson@zen.co.uk>
Signed-off-by: Dmitry Adamushko <dmitry.adamushko@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

1 files changed, 13 insertions, 7 deletions

diff --git a/kernel/sched.c b/kernel/sched.c
index d62b759753f3..db1f71e31310 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -216,15 +216,15 @@ static inline struct task_group *task_group(struct task_struct *p)
 }
 
 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
-static inline void set_task_cfs_rq(struct task_struct *p)
+static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu)
 {
-        p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)];
-        p->se.parent = task_group(p)->se[task_cpu(p)];
+        p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
+        p->se.parent = task_group(p)->se[cpu];
 }
 
 #else
 
-static inline void set_task_cfs_rq(struct task_struct *p) { }
+static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) { }
 
 #endif  /* CONFIG_FAIR_GROUP_SCHED */
 
@@ -1022,10 +1022,16 @@ unsigned long weighted_cpuload(const int cpu)
 
 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
 {
+        set_task_cfs_rq(p, cpu);
 #ifdef CONFIG_SMP
+        /*
+         * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
+         * successfuly executed on another CPU. We must ensure that updates of
+         * per-task data have been completed by this moment.
+         */
+        smp_wmb();
         task_thread_info(p)->cpu = cpu;
 #endif
-        set_task_cfs_rq(p);
 }
 
 #ifdef CONFIG_SMP
@@ -7088,7 +7094,7 @@ void sched_move_task(struct task_struct *tsk)
         rq = task_rq_lock(tsk, &flags);
 
         if (tsk->sched_class != &fair_sched_class) {
-                set_task_cfs_rq(tsk);
+                set_task_cfs_rq(tsk, task_cpu(tsk));
                 goto done;
         }
 
@@ -7103,7 +7109,7 @@ void sched_move_task(struct task_struct *tsk)
                         tsk->sched_class->put_prev_task(rq, tsk);
         }
 
-        set_task_cfs_rq(tsk);
+        set_task_cfs_rq(tsk, task_cpu(tsk));
 
         if (on_rq) {
                 if (unlikely(running))