kernel: return from interrupt

[John Baldwin]

On Thursday 11 November 2004 11:49 pm, Stephan Uphoff wrote:
> On Thu, 2004-11-11 at 23:34, Anurekh Saxena wrote:
> > On Thu, 11 Nov 2004 17:55:52 -0500, Stephan Uphoff <ups at tree.com> wrote:
> > > On Thu, 2004-11-11 at 12:58, Anurekh Saxena wrote:
> > > > Hi,
> > > >
> > > > I was under the impression that the 5.3 release had an option for
> > > > full preemption.
> > > > If I am correct, why does the kernel refuse to schedule on a
> > > > return_from_interrupt if its not
> > > > going back to userland?
> > > > I can understand this being a problem if interrupts were nested, or
> > > > return from a page fault in a
> > > > critical section.
> > > > Please correct me if I am wrong, but if a *high* priority interrupt
> > > > thread is ready to run, it
> > > > should be given a chance. Presuming the *interrupted* kernel path is
> > > > going to give up the CPU
> > > > fast enough is probably not a good idea.
> > > >
> > > >
> > > > I hope I have sent this to the right mailing list.
> > > >
> > > > Thanks,
> > > > Anurekh
> > >
> > > This should work if you have "options PREEMPTION" in your config file.
> > > You may also want to try "options FULL_PREEMPTION".
> >
> > I wasnt looking at the  FULL_PREEMPTION option at all. With that
> > enabled, the kernel will
> > call mi_switch when it adds the thread to the runqueue. Thanks for the
> > input.
> >
> > > Can you describe your problems / observations?
> >
> > I was expecting the common return_from_intr path to be used as a
> > preemption point.
> > It was an incorrect observation, and also probably wouldn't work with
> > the ast implementation.
> >
> > > The exception seems to be fast interrupts.
> > > You may want to try the following untested patch to allow preemption
> > > triggered by fast interrupts.
> >
> > That is interesting. I didn't see that the OWEPREEMPT flag is
> > deliberately cleared.
> > Do you why that is done?
>
> I assume that this is just a forgotten temporary fix for scheduler
> problems. I will try to verify this in the next days.

Yes, I've axed it in my jhb_preemption branch, but haven't had time to test it 
recently.

> >  I dont see why a handler will explicitly call
> > maybe_preempt, but it
> > could try to add some thread to the runqueue.
>
> wakeup and scheduling soft interrupt threads would come to mind.
>
> > Thanks for the feedback.
> >
> > -Anurekh
> >
> > > Index: intr_machdep.c
> > > ===================================================================
> > > RCS file: /cvsroot/src/sys/i386/i386/intr_machdep.c,v
> > > retrieving revision 1.11
> > > diff -u -r1.11 intr_machdep.c
> > > --- intr_machdep.c      3 Nov 2004 18:03:06 -0000       1.11
> > > +++ intr_machdep.c      11 Nov 2004 22:31:19 -0000
> > > @@ -205,7 +205,9 @@
> > >                 isrc->is_pic->pic_eoi_source(isrc);
> > >                 error = 0;
> > >                 /* XXX */
> > > +#if 0
> > >                 td->td_pflags &= ~TDP_OWEPREEMPT;
> > > +#endif
> > >                 critical_exit();
> > >         } else {
> > >                 /*

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org