Re: [RFCv4] timerfd: add TFD_NOTIFY_CLOCK_SET to watch for clock changes

[Alexander Shishkin <virtuoso@slind.org>]

On Thu, Mar 10, 2011 at 09:48:03AM +0100, Arnd Bergmann wrote:
> On Wednesday 09 March 2011 15:36:51 Alexander Shishkin wrote:
> > This is another attempt to approach notifying userspace about system
> > clock changes. The other one is using an eventfd and a syscall [1]. In
> > the course of discussing the necessity of a syscall for this kind of
> > notifications, it was suggested that this functionality can be achieved
> > via timers [2] (and timerfd in particular [3]). This idea got quite
> > some support [4], [5], [6] and some vague criticism [7], so I decided
> > to try and go a bit further with it.
> 
> I don't understand from your description or from the patch how user
> space gets notified. From your description, I would have expected
> to see a change to the timerfd_poll() function to check if the
> clock has changed, possibly returning POLLPRI, but the only such
> change I can see is in the timerfd_read() function. Don't you need
> to change poll() so that a task knows when to call read()?

Luckily, no changes for the _poll function were required, because the
notification code reuses the ctx->ticks counter of timerfds. IOW, poll
wakes up in the same way as before.

> > +/* TFD_NOTIFY_CLOCK_SET timers go here */
> > +static DEFINE_SPINLOCK(notifiers_lock);
> > +static LIST_HEAD(notifiers_list);
> 
> Maybe I was the only one to be confused by this, but I think t
> he naming is slightly misleading, because it's easy to confuse
> with a struct notifier_block. You could of course use the
> notifier API instead of building your own, but if you don't,
> I'd recommend coming up with a different name.

Point taken.

> I also think that a mutex would be better here than a spinlock.
> It's unlikely to be under heavy contention, but if you have
> a lot of threads, it could be held for a significant amount
> of time.

Indeed, thanks!

Regards,
--
Alex