Messages in this thread Patch in this message | | | Subject | [PATCH v2 2/6] sched: Do not pick a task which is switching on other cpu | From | Kirill Tkhai <> | Date | Mon, 22 Sep 2014 22:32:10 +0400 |
| |
From: Kirill Tkhai <ktkhai@parallels.com>
Architectures, which define __ARCH_WANT_UNLOCKED_CTXSW, may pull a task when it's in the middle of schedule().
CPU1(task1 calls schedule) CPU2 ... schedule() ... idle_balance() ... load_balance() ... ... schedule() ... prepare_lock_switch() ... raw_spin_unlock(&rq1->lock) ... ... raw_spin_lock(&rq1->lock) ... detach_tasks(); ... can_migrate_task(task1) ... attach_tasks(); <--- move task1 to rq2 ... raw_spin_unlock(&rq1->lock) ... context_switch() <--- switch to task1's stack ... ... (using task1's stack) (using task1's stack) ... ... context_switch() ...
Parallel use of a single stack is not a good idea.
Signed-off-by: Kirill Tkhai <ktkhai@parallels.com> Cc: <stable@vger.kernel.org> # Should this go to stable? --- kernel/sched/core.c | 11 +++-------- kernel/sched/deadline.c | 7 ++++++- kernel/sched/fair.c | 3 +++ kernel/sched/rt.c | 7 ++++++- kernel/sched/sched.h | 16 ++++++++++++++++ 5 files changed, 34 insertions(+), 10 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 2a93b87..5b864e9 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -1700,15 +1700,10 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) #ifdef CONFIG_SMP /* - * If the owning (remote) cpu is still in the middle of schedule() with - * this task as prev, wait until its done referencing the task. + * Note, that p is dequeued at the moment. But it still + * may be "prev" in the middle of schedule() on other cpu. */ - while (p->on_cpu) - cpu_relax(); - /* - * Pairs with the smp_wmb() in finish_lock_switch(). - */ - smp_rmb(); + cpu_relax__while_on_cpu(p); p->sched_contributes_to_load = !!task_contributes_to_load(p); p->state = TASK_WAKING; diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index aaa5abb..ea0ba33 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -1364,7 +1364,9 @@ static int push_dl_task(struct rq *rq) next_task = task; goto retry; } - +#ifdef __ARCH_WANT_UNLOCKED_CTXSW + cpu_relax__while_on_cpu(next_task); +#endif deactivate_task(rq, next_task, 0); set_task_cpu(next_task, later_rq->cpu); activate_task(later_rq, next_task, 0); @@ -1451,6 +1453,9 @@ static int pull_dl_task(struct rq *this_rq) ret = 1; +#ifdef __ARCH_WANT_UNLOCKED_CTXSW + cpu_relax__while_on_cpu(p); +#endif deactivate_task(src_rq, p, 0); set_task_cpu(p, this_cpu); activate_task(this_rq, p, 0); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index 420bc98..80c5064 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -5298,6 +5298,9 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env) schedstat_inc(env->sd, lb_hot_gained[env->idle]); schedstat_inc(p, se.statistics.nr_forced_migrations); } +#ifdef __ARCH_WANT_UNLOCKED_CTXSW + cpu_relax__while_on_cpu(p); +#endif return 1; } diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 2e6a774..de356b0 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -1734,7 +1734,9 @@ static int push_rt_task(struct rq *rq) next_task = task; goto retry; } - +#ifdef __ARCH_WANT_UNLOCKED_CTXSW + cpu_relax__while_on_cpu(next_task); +#endif deactivate_task(rq, next_task, 0); set_task_cpu(next_task, lowest_rq->cpu); activate_task(lowest_rq, next_task, 0); @@ -1823,6 +1825,9 @@ static int pull_rt_task(struct rq *this_rq) ret = 1; +#ifdef __ARCH_WANT_UNLOCKED_CTXSW + cpu_relax__while_on_cpu(p); +#endif deactivate_task(src_rq, p, 0); set_task_cpu(p, this_cpu); activate_task(this_rq, p, 0); diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 1bc6aad..9c07d72 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -1034,6 +1034,22 @@ static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ /* + * If the owning (remote) cpu is still in the middle of schedule() with + * this task as prev, wait until its done referencing the task. + */ +static inline void cpu_relax__while_on_cpu(struct task_struct *p) +{ +#ifdef CONFIG_SMP + while (p->on_cpu) + cpu_relax(); + /* + * Pairs with the smp_wmb() in finish_lock_switch(). + */ + smp_rmb(); +#endif +} + +/* * wake flags */ #define WF_SYNC 0x01 /* waker goes to sleep after wakeup */
| |