projects / thirdparty / kernel / stable-queue.git / blame
| Commit | Line | Data |
|---|---|---|
| cc6b8f45 GKH |
1 | From 0f397f2c90ce68821ee864c2c53baafe78de765d Mon Sep 17 00:00:00 2001 |
| 2 | From: Kirill Tkhai <tkhai@yandex.ru> | |
| 3 | Date: Tue, 20 May 2014 13:33:42 +0400 | |
| 4 | Subject: sched/dl: Fix race in dl_task_timer() | |
| 5 | ||
| 6 | From: Kirill Tkhai <tkhai@yandex.ru> | |
| 7 | ||
| 8 | commit 0f397f2c90ce68821ee864c2c53baafe78de765d upstream. | |
| 9 | ||
| 10 | Throttled task is still on rq, and it may be moved to other cpu | |
| 11 | if user is playing with sched_setaffinity(). Therefore, unlocked | |
| 12 | task_rq() access makes the race. | |
| 13 | ||
| 14 | Juri Lelli reports he got this race when dl_bandwidth_enabled() | |
| 15 | was not set. | |
| 16 | ||
| 17 | Other thing, pointed by Peter Zijlstra: | |
| 18 | ||
| 19 | "Now I suppose the problem can still actually happen when | |
| 20 | you change the root domain and trigger a effective affinity | |
| 21 | change that way". | |
| 22 | ||
| 23 | To fix that we do the same as made in __task_rq_lock(). We do not | |
| 24 | use __task_rq_lock() itself, because it has a useful lockdep check, | |
| 25 | which is not correct in case of dl_task_timer(). We do not need | |
| 26 | pi_lock locked here. This case is an exception (PeterZ): | |
| 27 | ||
| 28 | "The only reason we don't strictly need ->pi_lock now is because | |
| 29 | we're guaranteed to have p->state == TASK_RUNNING here and are | |
| 30 | thus free of ttwu races". | |
| 31 | ||
| 32 | Signed-off-by: Kirill Tkhai <tkhai@yandex.ru> | |
| 33 | Signed-off-by: Peter Zijlstra <peterz@infradead.org> | |
| 34 | Cc: Linus Torvalds <torvalds@linux-foundation.org> | |
| 35 | Link: http://lkml.kernel.org/r/3056991400578422@web14g.yandex.ru | |
| 36 | Signed-off-by: Ingo Molnar <mingo@kernel.org> | |
| 37 | Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> | |
| 38 | ||
| 39 | --- | |
| 40 | kernel/sched/deadline.c | 10 +++++++++- | |
| 41 | 1 file changed, 9 insertions(+), 1 deletion(-) | |
| 42 | ||
| 43 | --- a/kernel/sched/deadline.c | |
| 44 | +++ b/kernel/sched/deadline.c | |
| 45 | @@ -490,9 +490,17 @@ static enum hrtimer_restart dl_task_time | |
| 46 | struct sched_dl_entity, | |
| 47 | dl_timer); | |
| 48 | struct task_struct *p = dl_task_of(dl_se); | |
| 49 | - struct rq *rq = task_rq(p); | |
| 50 | + struct rq *rq; | |
| 51 | +again: | |
| 52 | + rq = task_rq(p); | |
| 53 | raw_spin_lock(&rq->lock); | |
| 54 | ||
| 55 | + if (rq != task_rq(p)) { | |
| 56 | + /* Task was moved, retrying. */ | |
| 57 | + raw_spin_unlock(&rq->lock); | |
| 58 | + goto again; | |
| 59 | + } | |
| 60 | + | |
| 61 | /* | |
| 62 | * We need to take care of a possible races here. In fact, the | |
| 63 | * task might have changed its scheduling policy to something |