On Fri, Mar 18, 2022 at 01:24:24PM +0000, Hyeonggon Yoo wrote:
> On Fri, Mar 18, 2022 at 08:55:32PM +0800, Boqun Feng wrote:
> > On Wed, Mar 16, 2022 at 03:45:48PM -0700, Namhyung Kim wrote:
> > [...]
> > > @@ -209,6 +210,7 @@ static inline int __sched __down_common(struct semaphore *sem, long state,
> > > long timeout)
> > > {
> > > struct semaphore_waiter waiter;
> > > + bool tracing = false;
> > >
> > > list_add_tail(&waiter.list, &sem->wait_list);
> > > waiter.task = current;
> > > @@ -220,18 +222,28 @@ static inline int __sched __down_common(struct semaphore *sem, long state,
> > > if (unlikely(timeout <= 0))
> > > goto timed_out;
> > > __set_current_state(state);
> > > + if (!tracing) {
> > > + trace_contention_begin(sem, 0);
> >
> > This looks a littl ugly ;-/
>
> I agree this can be simplified a bit.
>
> > Maybe we can rename __down_common() to
> > ___down_common() and implement __down_common() as:
> >
> > static inline int __sched __down_common(...)
> > {
> > int ret;
> > trace_contention_begin(sem, 0);
> > ret = ___down_common(...);
> > trace_contention_end(sem, ret);
> > return ret;
> > }
> >
> > Thoughts?
> >
>
> But IMO inlining tracepoints is generally not a good idea.
> Will increase kernel size a lot.
>
Ah, it's already inlined. Sorry.
> > Regards,
> > Boqun
> >
> > > + tracing = true;
> > > + }
> > > raw_spin_unlock_irq(&sem->lock);
> > > timeout = schedule_timeout(timeout);
> > > raw_spin_lock_irq(&sem->lock);
> > > - if (waiter.up)
> > > + if (waiter.up) {
> > > + trace_contention_end(sem, 0);
> > > return 0;
> > > + }
> > > }
> > >
> > > timed_out:
> > > + if (tracing)
> > > + trace_contention_end(sem, -ETIME);
> > > list_del(&waiter.list);
> > > return -ETIME;
> > >
> > > interrupted:
> > > + if (tracing)
> > > + trace_contention_end(sem, -EINTR);
> > > list_del(&waiter.list);
> > > return -EINTR;
> > > }
> > > --
> > > 2.35.1.894.gb6a874cedc-goog
> > >
