Re: [PATCH 1 of 6] x86, UV: smp_processor_id in a preemptable region

[Cliff Wickman <cpw@sgi.com>]

On Wed, Jun 15, 2011 at 06:15:18PM +0200, Ingo Molnar wrote:
> 
> * Cliff Wickman <cpw@sgi.com> wrote:
> 
> > Hi Ingo,
> > 
> > On Wed, Jun 15, 2011 at 05:54:45PM +0200, Ingo Molnar wrote:
> > > 
> > > * Cliff Wickman <cpw@sgi.com> wrote:
> > > 
> > > > On Wed, Jun 15, 2011 at 12:05:17PM +0600, Rakib Mullick wrote:
> > > > > On Wed, Jun 15, 2011 at 5:06 AM, Cliff Wickman <cpw@sgi.com> wrote:
> > > > > > From: Cliff Wickman <cpw@sgi.com>
> > > > > >
> > > > > > Calling smp_processor_id() from within a preemptable region will issue
> > > > > > a warning if DEBUG_PREEMPT is set.
> > > > > >
> > > > > > Diffed against 3.0.0-rc3
> > > > > >
> > > > > > Signed-off-by: Cliff Wickman <cpw@sgi.com>
> > > > > > ---
> > > > > > ?arch/x86/platform/uv/tlb_uv.c | ? ?2 ++
> > > > > > ?1 file changed, 2 insertions(+)
> > > > > >
> > > > > > Index: linux/arch/x86/platform/uv/tlb_uv.c
> > > > > > ===================================================================
> > > > > > --- linux.orig/arch/x86/platform/uv/tlb_uv.c
> > > > > > +++ linux/arch/x86/platform/uv/tlb_uv.c
> > > > > > @@ -1334,7 +1334,9 @@ static ssize_t tunables_write(struct fil
> > > > > >
> > > > > > ? ? ? ?instr[count] = '\0';
> > > > > >
> > > > > > + ? ? ? preempt_disable(); /* avoid DEBUG_PREEMPT warning */
> > > > > 
> > > > > I think above code comment, "avoid DEBUG_PREEMPT warning" should be to
> > > > > something more meaningful. It's a BUG, if smp_processor_id() is called
> > > > > within preemptible context. So, we don't want to hit that BUG.
> > > > 
> > > > I agree that calling smp_processor_id() within a preemptible context is
> > > > going to produce unpredictable results.  In this particular case we just
> > > > need a valid cpu number so that we can find a per-cpu structure.
> > > > That structure contains a reasonable (sanity-checking) limit to the value
> > > > of the tunable that is being written.
> > > 
> > > So what happens if the code gets preempted away and this CPU is 
> > > hotplugged away? You'll reference a CPU ID that does not exist 
> > > anymore.
> > 
> > You're right of course. But we don't support CPU hotplug on the UV 
> > hardware. There are enhancements needed in both the BIOS and Linux 
> > (BAU and GRU among them). They are on our work queue.
> 
> But here you put in yet another roadblock.

So would you say I should really widen the scope of the non-preemptible
region to include everything done with the results of that call to
smp_processor_id()?  
Which in this case is the call to parse_tunables_write().
Like this:

        preempt_disable();
        bcp = &per_cpu(bau_control, smp_processor_id());

        ret = parse_tunables_write(bcp, instr, count);
        preempt_enable_no_resched();

-Cliff
-- 
Cliff Wickman
SGI
cpw@sgi.com
(651) 683-3824