Re: can't flush tlb on e500

From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
To: Saito Hideo <hsaito.ppc@gmail.com>
Cc: linuxppc-dev list <linuxppc-dev@ozlabs.org>,
	linux-kernel@vger.kernel.org,
	Kumar Gala <galak@gate.crashing.org>
Subject: Re: can't flush tlb on e500
Date: Fri, 22 May 2009 19:42:16 +1000	[thread overview]
Message-ID: <1242985336.26104.30.camel@pasglop> (raw)
In-Reply-To: <1242984470.26104.29.camel@pasglop>

On Fri, 2009-05-22 at 19:27 +1000, Benjamin Herrenschmidt wrote:
> On Wed, 2009-05-20 at 19:12 +0900, Saito Hideo wrote:
> 
> > I think that the tlb should be cleared before mm->context.id is set
> > MMU_NO_CONTEXT.
> 
> You are right, this definitely looks like a bug on platforms that have
> HW support for the tlbil instruction (and thus care about the PID for
> flushing) which afaik is only the case of recent freescale chips.

In fact, you are doubly right in that it also happens on other platforms
because local_flush_tlb_mm() will check if the PID is MMU_NO_CONTEXT
regardless of what tlbilx supports.. oops

Looks like I only ran my context torture test with CONFIG_SMP enabled.
Shame on me.

> Have you verified that this change fixes your problem ?
> 
> Can you re-submit to linuxppc-dev@ozlabs.org mailing list, along with
> proper changeset comment and signed-off-by: line ?
> 
> Cheers,
> Ben.
> 
> > --- arch/powerpc/mm/mmu_context_nohash.c.orig	2009-03-24
> > 08:12:14.000000000 +0900
> > +++ arch/powerpc/mm/mmu_context_nohash.c	2009-05-20 18:33:53.000000000 +0900
> > @@ -122,22 +122,22 @@ static unsigned int steal_context_up(uns
> >  	struct mm_struct *mm;
> >  	int cpu = smp_processor_id();
> > 
> >  	/* Pick up the victim mm */
> >  	mm = context_mm[id];
> > 
> >  	pr_debug("[%d] steal context %d from mm @%p\n", cpu, id, mm);
> > 
> > -	/* Mark this mm has having no context anymore */
> > -	mm->context.id = MMU_NO_CONTEXT;
> > -
> >  	/* Flush the TLB for that context */
> >  	local_flush_tlb_mm(mm);
> > 
> > +	/* Mark this mm has having no context anymore */
> > +	mm->context.id = MMU_NO_CONTEXT;
> > +
> >  	/* XXX This clear should ultimately be part of local_flush_tlb_mm */
> >  	__clear_bit(id, stale_map[cpu]);
> > 
> >  	return id;
> >  }
> > 
> >  #ifdef DEBUG_MAP_CONSISTENCY
> >  static void context_check_map(void)
> > --
> > 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/
> 
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev