Re: [PATCH] don't use flush_tlb_all in suspend time

[Shaohua Li <shaohua.li@intel.com>]

On Sun, 2006-04-30 at 14:04 +0200, Pavel Machek wrote:
> On So 29-04-06 23:57:21, Andrew Morton wrote:
> > Shaohua Li <shaohua.li@intel.com> wrote:
> > >
> > > On Sun, 2006-04-30 at 06:45 +0000, Pavel Machek wrote:
> > > > Hi!
> > > > 
> > > > > flush_tlb_all uses on_each_cpu, which will disable/enable interrupt.
> > > > > In suspend/resume time, this will make interrupt wrongly enabled.
> > > > 
> > > > > diff -puN arch/i386/mm/init.c~flush_tlb_all_check arch/i386/mm/init.c
> > > > > --- linux-2.6.17-rc3/arch/i386/mm/init.c~flush_tlb_all_check	2006-04-29 08:47:05.000000000 +0800
> > > > > +++ linux-2.6.17-rc3-root/arch/i386/mm/init.c	2006-04-29 08:48:15.000000000 +0800
> > > > > @@ -420,7 +420,10 @@ void zap_low_mappings (void)
> > > > >  #else
> > > > >  		set_pgd(swapper_pg_dir+i, __pgd(0));
> > > > >  #endif
> > > > > -	flush_tlb_all();
> > > > > +	if (cpus_weight(cpu_online_map) == 1)
> > > > > +		local_flush_tlb();
> > > > > +	else
> > > > > +		flush_tlb_all();
> > > > >  }
> > > > >
> > > > 
> > > > Either it is okay to enable interrupts here -> unneccessary and ugly
> > > > test, or it is not, and then we are broken in SMP case.
> > > It's not broken in SMP case, APs are offlined here in suspend/resume.
> > > 
> > 
> > In which case, how's about this?
> 
> Certainly better, I'd say.
> 
> > @@ -420,7 +421,14 @@ void zap_low_mappings (void)
> >  #else
> >  		set_pgd(swapper_pg_dir+i, __pgd(0));
> >  #endif
> > -	if (cpus_weight(cpu_online_map) == 1)
> > +	/*
> > +	 * We can be called at suspend/resume time, with local interrupts
> > +	 * disabled.  But flush_tlb_all() requires that local interrupts be
> > +	 * enabled.
> > +	 *
> > +	 * Happily, the APs are not yet started, so we can use local_flush_tlb()	 * in that case
> > +	 */
> > +	if (num_online_cpus() == 1)
> >  		local_flush_tlb();
> >  	else
> >  		flush_tlb_all();
Sorry for the delay. Last week is holiday here.

> But this still scares. It means calling convention is "may enable
> interrupts with >1 cpu, may not with == 1 cpu". 
Then we need port x86_64's implementation. I'll try if I can work it
out.

Thanks,
Shaohua
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/