Re: [PATCH v3 2/3] autofs-5.1.1 - use monotonic clock for pthread cond timed wait.

[Ian Kent <raven@themaw.net>]

On Thu, 2015-09-17 at 17:12 +0800, Ning Yu wrote:
> On Thu, Sep 17, 2015 at 4:39 PM, Ian Kent <raven@themaw.net> wrote:
> > On Thu, 2015-09-17 at 15:19 +0800, Ning Yu wrote:
> >> On Thu, Sep 17, 2015 at 2:49 PM, Ian Kent <raven@themaw.net> wrote:
> >> > On Thu, 2015-09-17 at 11:48 +0800, Yu Ning wrote:
> >> >> The default PTHREAD_COND_INITIALIZER initializer uses realtime clock,
> >> >> we need to switch to use the monotic clock.
> >> >>
> >> >> Signed-off-by: Yu Ning <ning.yu@ubuntu.com>
> >> >> ---
> >> >>  lib/alarm.c | 13 ++++++++++++-
> >> >>  1 file changed, 12 insertions(+), 1 deletion(-)
> >> >>
> >> >> diff --git a/lib/alarm.c b/lib/alarm.c
> >> >> index 65a80ae..5b98b2d 100755
> >> >> --- a/lib/alarm.c
> >> >> +++ b/lib/alarm.c
> >> >> @@ -23,7 +23,7 @@ struct alarm {
> >> >>  };
> >> >>
> >> >>  static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
> >> >> -static pthread_cond_t cond = PTHREAD_COND_INITIALIZER;
> >> >> +static pthread_cond_t cond;
> >> >>  static LIST_HEAD(alarms);
> >> >>
> >> >>  #define alarm_lock() \
> >> >> @@ -212,6 +212,7 @@ int alarm_start_handler(void)
> >> >>       pthread_t thid;
> >> >>       pthread_attr_t attrs;
> >> >>       pthread_attr_t *pattrs = &attrs;
> >> >> +     pthread_condattr_t condattrs;
> >> >>       int status;
> >> >>
> >> >>       status = pthread_attr_init(pattrs);
> >> >> @@ -224,8 +225,18 @@ int alarm_start_handler(void)
> >> >>  #endif
> >> >>       }
> >> >>
> >> >> +     status = pthread_condattr_init(&condattrs);
> >> >> +     if (status)
> >> >> +             fatal(status);
> >> >> +
> >> >> +     pthread_condattr_setclock(&condattrs, CLOCK_MONOTONIC);
> >> >> +     pthread_cond_init(&cond, &condattrs);
> >> >> +
> >> >>       status = pthread_create(&thid, pattrs, alarm_handler, NULL);
> >> >>
> >> >> +     pthread_condattr_destroy(&condattrs);
> >> >> +     pthread_condattr_destroy(&cond);
> >> >
> >> > I don't think we can do this.
> >> >
> >> > The condition is used by the alarm_handler() for the duration of it's
> >> > life so I don't think we can't destroy its attributes after thread
> >> > creation.
> >> >
> >> > In fact I'm not sure we even need the pthread_condattr_destroy(&cond)
> >> > but I'm not sure.
> >> >
> >> > Destroying the condition attributes (condattrs) OTOH must not affect any
> >> > condition which it has been used to initialize so destroying that is
> >> > fine.
> >> >
> >> > Let me have a closer look at this, maybe I've misunderstood what's going
> >> > on, and report back later.
> >> >
> >>
> >> I think you are right, we shouldn't destroy the cond attrs, nor the cond
> >> (and I used wrong function to destroy cond...) if the thread is successfully
> >> created.
> >>
> >> In fact I think we should do the cleanup at the end of the thread, however
> >> since the thread is expected to live __forever__, maybe we can simply ignore
> >> the cleanup.
> >
> > That's what I'm thinking too.
> >
> > I don't think there is any problem assuming the monotonic clock is
> > available (I think it's been around a very long time) so there's no
> > point in complicating things to cater for that case.
> >
> > Although I am thinking of checking the return and calling the fatal()
> > function on failure as I do for other pthreads functions that should
> > never fail, so if it does fail we'll get a simple message and a core so
> > we know where to look.
> >
> 
> Yes, those messages will definitely be very useful for debugging, do
> you mind that I update my patch to add those checking and messages? Or
> will it be better to submit a new patch for that job?

Sure, send an updated patch, I'm not able to start work on it quite yet.

> 
> >>
> >> >> +
> >> >>       if (pattrs)
> >> >>               pthread_attr_destroy(pattrs);
> >> >>
> >> >
> >> >
> >
> >


--
To unsubscribe from this list: send the line "unsubscribe autofs" in