Re: [PATCH] cpuidle: Check cpuidle driver before add refcount

[Rafael J. Wysocki]

On Friday, August 30, 2013 09:39:27 PM Daniel Fu wrote:
> Yes, I saw it. 
> Thanks. found one case when  Find PM domain and connect cpuidle to it , in when Kernel booting ,opening early prink testing
> pm_genpd_name_attach_cpuidle()->cpuidle_driver_ref().

I see. OK, I'll queue it up for the second part of the 3.12 merge window.

Thanks,
Rafael


> -----Original Message-----
> From: Rafael J. Wysocki [mailto:rjw@xxxxxxx] 
> Sent: 2013å8æ30æ 21:26
> To: Daniel Fu
> Cc: daniel.lezcano@xxxxxxxxxx; linux-pm@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Johnny Qiu
> Subject: Re: [PATCH] cpuidle: Check cpuidle driver before add refcount
> 
> On Friday, August 30, 2013 08:26:34 PM Daniel Fu wrote:
> > X-NVConfidentiality: public
> > 
> > In SMP, When doing the cpuidle driver registration. There maybe only 1 
> > cpu have register cpuidle(CPU0), but In the process of registration 
> > cpuidle driver, the scheduler maybe schedule the process to the other 
> > CPU (eg. CPU0 have heay load, migrate to CPU1)
> > cpuidle_get_driver() --> get_cpu() , will get CPU1, if CPU1 didn't register cpuidle driver, Will get NULL.
> > 
> > I know we should prevent the registration migrate to the other CPU, but we should check drv before add refcount to prevent kernel panic at least, Right?
> 
> Well, cpuidle_driver_ref() has only one caller, which is not the registration.
> 
> However, the caller apparently assumes that that function may return NULL and
> cpuidle_driver_unref() check drv against NULL too, so cpuidle_driver_ref() should do that either.
> 
> Did you see that problem in testing or just through code inspection?
> 
> Rafael
> 
> 
> > -----Original Message-----
> > From: Rafael J. Wysocki [mailto:rjw@xxxxxxx]
> > Sent: 2013å8æ30æ 20:01
> > To: Daniel Fu
> > Cc: daniel.lezcano@xxxxxxxxxx; linux-pm@xxxxxxxxxxxxxxx; 
> > linux-kernel@xxxxxxxxxxxxxxx; Johnny Qiu
> > Subject: Re: [PATCH] cpuidle: Check cpuidle driver before add refcount
> > 
> > On Friday, August 30, 2013 07:48:22 PM Daniel Fu wrote:
> > > If the current CPU has no cpuidle driver, drv will be NULL.
> > > Check if we get drv successfully before add refount to prevent 
> > > Kernel panic.
> > 
> > What is the actual scenario that may lead to this panic?
> > 
> > Rafael
> > 
> > 
> > > Signed-off-by: Daniel Fu <danifu@xxxxxxxxxx>
> > > ---
> > >  drivers/cpuidle/driver.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/cpuidle/driver.c b/drivers/cpuidle/driver.c 
> > > index
> > > 3ac499d..6e11701 100644
> > > --- a/drivers/cpuidle/driver.c
> > > +++ b/drivers/cpuidle/driver.c
> > > @@ -331,7 +331,8 @@ struct cpuidle_driver *cpuidle_driver_ref(void)
> > >  	spin_lock(&cpuidle_driver_lock);
> > >  
> > >  	drv = cpuidle_get_driver();
> > > -	drv->refcnt++;
> > > +	if (drv)
> > > +		drv->refcnt++;
> > >  
> > >  	spin_unlock(&cpuidle_driver_lock);
> > >  	return drv;
> > > 
> > --
> > I speak only for myself.
> > Rafael J. Wysocki, Intel Open Source Technology Center.
> > 
> --
> I speak only for myself.
> Rafael J. Wysocki, Intel Open Source Technology Center.
> 
-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/