Re: tasklets in 2.4.6

[Andrea Arcangeli]

On Mon, Jul 09, 2001 at 02:39:46PM +0300, Ville Nummela wrote:
> Ville Nummela <[e-mail]> writes:
> 
> > Jeff Garzik <[e-mail]> writes:
> > > As I hacked around to fix this, I noticed Andrea's ksoftirq patch
> > > already fixed this.  So, I decided to look over his patch instead.
> > I tried that patch too, but with not so good results: The code LOOKS good to
> > mee too, but for some reason it still seems to stuck in looping the tasklet
> > code only. btw. I'm trying this on PPC, it might have something to do with
> > that.. :)
> 
> Stupid is stupid does.. I had used Adreas patch for a wrong kernel version,
> and therefore the patch hadn't quite succeeded. The right patch works well
> on the PPC too :-)

btw, I tell to you too, if you have an SMP you also need to fix this plain
scheduler cpu affinity bug or you can deadlock at boot when ksoftirqd
kicks in:

	ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.7pre3aa1/00_cpus_allowed-1

attached here too:

--- 2.4.4pre3aa/kernel/sched.c.~1~	Sat Apr 14 15:49:11 2001
+++ 2.4.4pre3aa/kernel/sched.c	Sun Apr 15 18:31:14 2001
@@ -765,6 +765,8 @@
 	goto repeat_schedule;
 
 still_running:
+	if (!(prev->cpus_allowed & (1UL << this_cpu)))
+		goto still_running_back;
 	c = goodness(prev, this_cpu, prev->active_mm);
 	next = prev;
 	goto still_running_back;

Andrea