rtmutex: Deboost before waking up the top waiter
authorXunlei Pang <xlpang@redhat.com>
Thu, 23 Mar 2017 14:56:07 +0000 (15:56 +0100)
committerThomas Gleixner <tglx@linutronix.de>
Tue, 4 Apr 2017 09:44:05 +0000 (11:44 +0200)
commit2a1c6029940675abb2217b590512dbf691867ec4
treea975e4e9f643d86a904b6314a3bab21b023fd6cc
parent38bffdac071b720db627bfd2b125a2802a04d419
rtmutex: Deboost before waking up the top waiter

We should deboost before waking the high-priority task, such that we
don't run two tasks with the same "state" (priority, deadline,
sched_class, etc).

In order to make sure the boosting task doesn't start running between
unlock and deboost (due to 'spurious' wakeup), we move the deboost
under the wait_lock, that way its serialized against the wait loop in
__rt_mutex_slowlock().

Doing the deboost early can however lead to priority-inversion if
current would get preempted after the deboost but before waking our
high-prio task, hence we disable preemption before doing deboost, and
enabling it after the wake up is over.

This gets us the right semantic order, but most importantly however;
this change ensures pointer stability for the next patch, where we
have rt_mutex_setprio() cache a pointer to the top-most waiter task.
If we, as before this change, do the wakeup first and then deboost,
this pointer might point into thin air.

[peterz: Changelog + patch munging]
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Xunlei Pang <xlpang@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: juri.lelli@arm.com
Cc: bigeasy@linutronix.de
Cc: mathieu.desnoyers@efficios.com
Cc: jdesfossez@efficios.com
Cc: bristot@redhat.com
Link: http://lkml.kernel.org/r/20170323150216.110065320@infradead.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
kernel/futex.c
kernel/locking/rtmutex.c
kernel/locking/rtmutex_common.h