Re: bug in select() (was Re: {sys_,/dev/}epoll waiting timeout)

[Jamie Lokier]

Mark Mielke wrote:
> On Tue, Jan 28, 2003 at 09:45:00AM +0000, Jamie Lokier wrote:
> > Davide Libenzi wrote:
> > > ( if Tms > 0 )
> > Which is unfortunate, because that doesn't allow for a value of Tms ==
> > 0 which is needed when you want to sleep and wake up on every jiffie
> > on systems where HZ >= 1000.  Tms == 0 is taken already, to mean do
> > not wait at all.
> 
> To some degree, isn't this the equivalent of yield()?

No.  If you want a process to wake every HZ tick, do a little work and
then sleep again, yield() won't do that.

If HZ >= 1000, you simply can't use Linux poll() to do that; you have
to use select().  (Or epoll_wait()).

Even if select() is changed to do double-rounding-up like poll(), it
will still do this because select() times have microsecond
granularity.

-- Jamie