Re: [KJ] [patch] fix common mistake in polling loops

[Om N.]

On 8/7/06, Darren Jenkins <darrenrjenkins@xxxxxxxxx> wrote:
> G'day
>
> On 8/8/06, Pavel Machek <pavel@xxxxxxx> wrote:
> > Hi!
> >
> > > >> Well, whoever wrote thi has some serious problems (in attitude
> > > >> department). *Any* loop you design may take half a minute under
> > > >> streange circumstances.
> > >
> > > 6.
> > > common mistake in polling loops [from Linus]:
> >
> > Yes, Linus was wrong here. Or more precisely, he's right original code
> > is broken, but his suggested "fix" is worse than the original.
> >
> >         unsigned long timeout = jiffies + HZ/2;
> >         for (;;) {
> >                 if (ready())
> >                         return 0;
> > [IMAGINE HALF A SECOND DELAY HERE]
> >                 if (time_after(timeout, jiffies))
> >                         break;
> >                 msleep(10);
> >         }
> >
> > Oops.
> >
> > > >Actually it may be broken, depending on use. In some cases this loop
> > > >may want to poll the hardware 50 times, 10msec appart... and your loop
> > > >can poll it only once in extreme conditions.
> > > >
> > > >Actually your loop is totally broken, and may poll only once (without
> > > >any delay) and then directly timeout :-P -- that will break _any_
> > > >user.
> > >
> > > The Idea is that we are checking some event in external hardware that
> > > we know will complete in a given time (This time is not dependant on
> > > system activity but is fixed). After that time if the event has not
> > > happened we know something has borked.
> >
> > But you have to make sure YOU CHECK READY AFTER THE TIMEOUT. Linus'
> > code does not do that.
> >                                                                 Pavel
>
>
> Sorry I did not realise that was your problem with the code.
> Would it help if we just explicitly added a
     unsigned long timeout = jiffies + HZ/2;
      for (;;) {
              if (ready())
                      return 0;
[IMAGINE HALF A SECOND DELAY HERE]
              if (time_after(timeout, jiffies)) {
                      if (ready())
                             return 0;
                      break;
               }
              msleep(10);
      }
Wouldn't this be better than adding a check after the break of loop?

> if (ready())
>         return 0;
>
> after the loop, in the example code? so people wont miss adding
> something like that in?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/