lockdep: re-annotate scheduler runqueues
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Mon, 11 Aug 2008 07:30:22 +0000 (09:30 +0200)
committerIngo Molnar <mingo@elte.hu>
Mon, 11 Aug 2008 07:30:22 +0000 (09:30 +0200)
commit1b12bbc747560ea68bcc132c3d05699e52271da0
tree0e0fe5b7fe07d411251eebdd053e9e7793820248
parent64aa348edc617dea17bbd01ddee4e47886d5ec8c
lockdep: re-annotate scheduler runqueues

Instead of using a per-rq lock class, use the regular nesting operations.

However, take extra care with double_lock_balance() as it can release the
already held rq->lock (and therefore change its nesting class).

So what can happen is:

 spin_lock(rq->lock); // this rq subclass 0

 double_lock_balance(rq, other_rq);
   // release rq
   // acquire other_rq->lock subclass 0
   // acquire rq->lock subclass 1

 spin_unlock(other_rq->lock);

leaving you with rq->lock in subclass 1

So a subsequent double_lock_balance() call can try to nest a subclass 1
lock while already holding a subclass 1 lock.

Fix this by introducing double_unlock_balance() which releases the other
rq's lock, but also re-sets the subclass for this rq's lock to 0.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched.c
kernel/sched_rt.c