sched/numa: Fix a possible divide-by-zero
authorXie XiuQi <xiexiuqi@huawei.com>
Sat, 20 Apr 2019 08:34:16 +0000 (16:34 +0800)
committerIngo Molnar <mingo@kernel.org>
Thu, 25 Apr 2019 17:58:54 +0000 (19:58 +0200)
commita860fa7b96e1a1c974556327aa1aee852d434c21
tree242bb6433d229e091cc9f75d775bdc43e1b0b026
parentcd8dead0c39457e58ec1d36db93aedca811d48f1
sched/numa: Fix a possible divide-by-zero

sched_clock_cpu() may not be consistent between CPUs. If a task
migrates to another CPU, then se.exec_start is set to that CPU's
rq_clock_task() by update_stats_curr_start(). Specifically, the new
value might be before the old value due to clock skew.

So then if in numa_get_avg_runtime() the expression:

  'now - p->last_task_numa_placement'

ends up as -1, then the divider '*period + 1' in task_numa_placement()
is 0 and things go bang. Similar to update_curr(), check if time goes
backwards to avoid this.

[ peterz: Wrote new changelog. ]
[ mingo: Tweaked the code comment. ]

Signed-off-by: Xie XiuQi <xiexiuqi@huawei.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: cj.chengjian@huawei.com
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20190425080016.GX11158@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/fair.c