Message ID | 20140807151741.GP19379@twins.programming.kicks-ass.net |
---|---|
State | Changes Requested, archived |
Delegated to: | David Miller |
Headers | show |
On Thu, 7 Aug 2014 17:17:41 +0200 Peter Zijlstra <peterz@infradead.org> wrote: > Subject: atm: Fix blocking in wait loop > > One should not call blocking primitives inside a wait loop, since both > require task_struct::state to sleep, so the inner will destroy the outer > state. > > In this instance sigd_enq() will possible sleep for alloc_skb(), now if > I understand the code right, we do not actually need to call sigd_enq() > after the initial prepare_to_wait(), because we test the termination > condition before schedule() anyhow. > > So we can simply move it up a bit and avoid the entire confusion. > > Signed-off-by: Peter Zijlstra <peterz@infradead.org> > --- > net/atm/svc.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/net/atm/svc.c b/net/atm/svc.c > index d8e5d0c2ebbc..445ac238b69b 100644 > --- a/net/atm/svc.c > +++ b/net/atm/svc.c > @@ -297,8 +297,8 @@ static int svc_listen(struct socket *sock, int backlog) > goto out; > } > set_bit(ATM_VF_WAITING, &vcc->flags); > - prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); > sigd_enq(vcc, as_listen, NULL, NULL, &vcc->local); > + prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); > while (test_bit(ATM_VF_WAITING, &vcc->flags) && sigd) { > schedule(); > prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); This isn't the only place that we queue a message for the signalling daemon after a prepare_to_wait() uninterruptibly so this patch would be incomplete as is. What bothers me is the TASK_UNINTERRUPTIBLE -- I don't have a good reason why any of these should be sleeping uninterruptibly. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Aug 07, 2014 at 12:59:48PM -0400, chas williams - CONTRACTOR wrote: > On Thu, 7 Aug 2014 17:17:41 +0200 > Peter Zijlstra <peterz@infradead.org> wrote: > > > Subject: atm: Fix blocking in wait loop > > > > One should not call blocking primitives inside a wait loop, since both > > require task_struct::state to sleep, so the inner will destroy the outer > > state. > > > > In this instance sigd_enq() will possible sleep for alloc_skb(), now if > > I understand the code right, we do not actually need to call sigd_enq() > > after the initial prepare_to_wait(), because we test the termination > > condition before schedule() anyhow. > > > > So we can simply move it up a bit and avoid the entire confusion. > > > > Signed-off-by: Peter Zijlstra <peterz@infradead.org> > > --- > > net/atm/svc.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/net/atm/svc.c b/net/atm/svc.c > > index d8e5d0c2ebbc..445ac238b69b 100644 > > --- a/net/atm/svc.c > > +++ b/net/atm/svc.c > > @@ -297,8 +297,8 @@ static int svc_listen(struct socket *sock, int backlog) > > goto out; > > } > > set_bit(ATM_VF_WAITING, &vcc->flags); > > - prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); > > sigd_enq(vcc, as_listen, NULL, NULL, &vcc->local); > > + prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); > > while (test_bit(ATM_VF_WAITING, &vcc->flags) && sigd) { > > schedule(); > > prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); > > > This isn't the only place that we queue a message for the signalling > daemon after a prepare_to_wait() uninterruptibly so this patch would > be incomplete as is. I'm not sure I follow, this is the only place you do so while then going to sleep. All other sites don't sleep while they're enqueued on the waitqueue. > What bothers me is the TASK_UNINTERRUPTIBLE -- I don't have a good > reason why any of these should be sleeping uninterruptibly. That's a whole different story, there's tons of ugly in there, but its all ancient code.
On Thu, Aug 07, 2014 at 07:25:14PM +0200, Peter Zijlstra wrote: > > This isn't the only place that we queue a message for the signalling > > daemon after a prepare_to_wait() uninterruptibly so this patch would > > be incomplete as is. > > I'm not sure I follow, this is the only place you do so while then going > to sleep. All other sites don't sleep while they're enqueued on the > waitqueue. Argh, my brain is fried, you're quite right. I'll go have another stab at it tomorrow or so.
diff --git a/net/atm/svc.c b/net/atm/svc.c index d8e5d0c2ebbc..445ac238b69b 100644 --- a/net/atm/svc.c +++ b/net/atm/svc.c @@ -297,8 +297,8 @@ static int svc_listen(struct socket *sock, int backlog) goto out; } set_bit(ATM_VF_WAITING, &vcc->flags); - prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); sigd_enq(vcc, as_listen, NULL, NULL, &vcc->local); + prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE); while (test_bit(ATM_VF_WAITING, &vcc->flags) && sigd) { schedule(); prepare_to_wait(sk_sleep(sk), &wait, TASK_UNINTERRUPTIBLE);