Messages in this thread | | | Subject | Re: [PATCH] posix-timers: RCU conversion | From | john stultz <> | Date | Mon, 04 Apr 2011 11:08:21 -0700 |
| |
On Sun, 2011-04-03 at 18:54 +0200, Eric Dumazet wrote: > Andrew, I submitted this patch some time ago and got no feedback from > Thomas, John, Richard or Paul.
Sorry about that. Its been on my list.
It looks ok to me, but I not very familiar with idr locking rules (though looking at the code the comments state that idr_find can be safely called the way you do here).
CC'ing Oleg as I think he knows the code and is always good at catching things.
thanks -john
> [PATCH] posix-timers: RCU conversion > > Ben Nagy reported a scalability problem with KVM/QEMU that hit very hard > a single spinlock (idr_lock) in posix-timers code, on its 48 core > machine. > > Even on a 16 cpu machine (2x4x2), a single test can show 98% of cpu time > used in ticket_spin_lock, from lock_timer > > Ref: http://www.spinics.net/lists/kvm/msg51526.html > > Switching to RCU is quite easy, IDR being already RCU ready. > > idr_lock should be locked only for an insert/delete, not a lookup. > > Benchmark on a 2x4x2 machine, 16 processes calling timer_gettime(). > > Before : > > real 1m18.669s > user 0m1.346s > sys 1m17.180s > > After : > > real 0m3.296s > user 0m1.366s > sys 0m1.926s > > Reported-by: Ben Nagy <ben@iagu.net> > Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com> > Tested-by: Ben Nagy <ben@iagu.net> > Cc: Avi Kivity <avi@redhat.com> > Cc: Thomas Gleixner <tglx@linutronix.de> > Cc: John Stultz <johnstul@us.ibm.com> > Cc: Richard Cochran <richard.cochran@omicron.at> > Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> > --- > include/linux/posix-timers.h | 1 + > kernel/posix-timers.c | 25 ++++++++++++++----------- > 2 files changed, 15 insertions(+), 11 deletions(-) > > diff --git a/include/linux/posix-timers.h b/include/linux/posix-timers.h > index d51243a..5dc27ca 100644 > --- a/include/linux/posix-timers.h > +++ b/include/linux/posix-timers.h > @@ -81,6 +81,7 @@ struct k_itimer { > unsigned long expires; > } mmtimer; > } it; > + struct rcu_head rcu; > }; > > struct k_clock { > diff --git a/kernel/posix-timers.c b/kernel/posix-timers.c > index 4c01249..acb9be9 100644 > --- a/kernel/posix-timers.c > +++ b/kernel/posix-timers.c > @@ -491,6 +491,13 @@ static struct k_itimer * alloc_posix_timer(void) > return tmr; > } > > +static void k_itimer_rcu_free(struct rcu_head *head) > +{ > + struct k_itimer *tmr = container_of(head, struct k_itimer, rcu); > + > + kmem_cache_free(posix_timers_cache, tmr); > +} > + > #define IT_ID_SET 1 > #define IT_ID_NOT_SET 0 > static void release_posix_timer(struct k_itimer *tmr, int it_id_set) > @@ -503,7 +510,7 @@ static void release_posix_timer(struct k_itimer *tmr, int it_id_set) > } > put_pid(tmr->it_pid); > sigqueue_free(tmr->sigq); > - kmem_cache_free(posix_timers_cache, tmr); > + call_rcu(&tmr->rcu, k_itimer_rcu_free); > } > > static struct k_clock *clockid_to_kclock(const clockid_t id) > @@ -631,22 +638,18 @@ out: > static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags) > { > struct k_itimer *timr; > - /* > - * Watch out here. We do a irqsave on the idr_lock and pass the > - * flags part over to the timer lock. Must not let interrupts in > - * while we are moving the lock. > - */ > - spin_lock_irqsave(&idr_lock, *flags); > + > + rcu_read_lock(); > timr = idr_find(&posix_timers_id, (int)timer_id); > if (timr) { > - spin_lock(&timr->it_lock); > + spin_lock_irqsave(&timr->it_lock, *flags); > if (timr->it_signal == current->signal) { > - spin_unlock(&idr_lock); > + rcu_read_unlock(); > return timr; > } > - spin_unlock(&timr->it_lock); > + spin_unlock_irqrestore(&timr->it_lock, *flags); > } > - spin_unlock_irqrestore(&idr_lock, *flags); > + rcu_read_unlock(); > > return NULL; > } > >
| |