Re: Short sleeps.

[Steve Baker <sjbaker1@airmail.net>]

Mark Collins wrote:

> Well, that could be a problem with such high-resolution sleeps, and could be
> a wasted effort (although the loopback idea sounds good...).

The infamous paperclip hack *does* work - so it's true that in theory, if you
could find a clean way to cause a kernel-level interrupt in 1ms, you could
block for that long.
 
> When you give control to another process, that process could block for <x>
> milliseconds, then the schedular passes controll to another process, which
> blocks for <y> milliseconds and so on.

I'm assuming that my sleeping process has higher priority than the other
game threads - and that there are no other *significant* processes running.
 
> Amd, looking at my FIrst Edition of Linux Device Drivers, on page 137, there
> is mention of the udelay() function, which is pretty high resolution, BUT, it
> blocks. The books says that there are no functions (or fails to mention
> functions) for such high resolution delays.

I don't see a 'udelay' in any modern Linux docs.

> Of course, this is quite an old book, and things might have changed since
> then...

Yes.  Early Linux was not very standardized.  Also, if it was a book on device
drivers, there may well be some functions inside the kernel that drivers can
use that are not available to user-level processes.
 
> I just tried a very simple program, which uses sleep(), which looked like it
> had no delay. I'm not sure if this blocks, but if you pass 0 as an argument,
> it MAY continue performing tasks in the schedule queue for one cycle. Stick
> that in a loop like the one I mentioned eariler, and you may be in luck...

Yes - come to think of it, people do that a lot on IRIX machines using
an IRIX-specific function called 'sginap'.  The problem with sginap (and
probably this suggestion) is that if you have a higher process priority
than the other threads you'd like to run, then it's just like a busy-wait
and if your process priority is lower then you get kicked off the CPU for
10ms come-what-may.


----------------------------- Steve Baker -------------------------------
Mail : <sjbaker1@airmail.net>   WorkMail: <sjbaker@link.com>
URLs : http://www.sjbaker.org
       http://plib.sf.net http://tuxaqfh.sf.net http://tuxkart.sf.net
       http://prettypoly.sf.net http://freeglut.sf.net
       http://toobular.sf.net   http://lodestone.sf.net