[parisc-linux] sched-O1-2.4.17-I3.patch, need help

[joel.soete@freebel.net]

Hello James,

Thanks a lot for inforamtion.

Quoting James Bottomley <James.Bottomley@HansenPartnership.com>:

> > >         while (1) {
> > >                 void (*idle)(void) = pm_idle;
> > >                 if (!idle)
> > >                         idle = default_idle;
> > > -               while (!current->need_resched)
> > > +               if (!current->need_resched)
> > >                         idle();
> > >                 schedule();
> > >                 check_pgt_cache();
> >
> > Dunno.
> 
> This is sending the idle thread back through the scheduler at least once
> every 
> clock tick.  On the x86, idle() halts the processor until it receives an
> 
> interrupt.
> 
> I *think* the reason is the idle tick rebalances the run queue but
> doesn't 
> schedule the first task for execution (if there is one), so the call to
> 
> schedule() does that.
> 
> However, there are more changes than just this (I just did the changes
> for a 
> different arch).  You need to change p->processor to p->cpu in the SMP
> code.  
> There may be other struct task changes that bite, but that looks like
> the only 
> one in parisc.

I think that also p->nice would be changed in p->__nice?

> 
> You also get bitten (in I3) by the physical to logical CPU mapping
> problems.  
> These are gone in J0, but now you have to introduce and extra
> migrate_process 
> cross processor interrupt.

I also have a look in J2 and notice that task_struct was deepely changed and so
I  ignore what will I have to do with offset.c.

> 
> Probably the best way to work on this is to start a 2.5 parisc tree...
> 

I do also believe that the limits of my knowledge will yet oblige me to wait
this release.

Thanks anyway for help and precious advises,

Joel

> Good luck.
> 
> James Bottomley
> 
> 
> 
> 


-------------------------------------------------
This mail sent through Tiscalinet Webmail (http://webmail.tiscali.be)