Re: [PATCH] irq: fix possible null-pointer deref irq_domain_to_irq

Re: [PATCH] irq: fix possible null-pointer deref irq_domain_to_irq

[Rob Herring]

Jamie,

On 11/25/2011 09:35 AM, Jamie Iles wrote:
> Hi Rob,
> 
> On Wed, Nov 23, 2011 at 12:23:08PM +0000, Jamie Iles wrote:
>> On Thu, Nov 10, 2011 at 07:37:32AM -0600, Rob Herring wrote:
>>> On 11/10/2011 05:14 AM, Jamie Iles wrote:
>>>> On Thu, Nov 10, 2011 at 10:53:03AM +0000, Jamie Iles wrote:
>>>>> It is optional for an irqdomain to have a to_irq() method, and for
>>>>> simple domains they often don't require any operations at all - just
>>>>> hwirq to Linux irq translation.  Check we have valid ops before
>>>>> dereferencing them.
>>>>>
>>>>> Patch originally by Rob Herring.
>>>>>
>>>>> Suggested-by: Rob Herring <robherring2-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>
>>>> Naturally git send-email doesn't know how to convert this into a CC, so 
>>>> Rob is now CC'd!
>>>>
>>>> Jamie
>>>>
>>>>> Cc: Thomas Gleixner <tglx-hfZtesqFncYOwBW4kG4KsQ@public.gmane.org>
>>>>> Cc: Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>
>>>>> Signed-off-by: Jamie Iles <jamie-wmLquQDDieKakBO8gow8eQ@public.gmane.org>
>>>>> ---
>>>>>
>>>>> Rob, I can't see that you've already posted this but I didn't want to
>>>>> hold Marc's GIC patches up.
>>>
>>> Thanks. I didn't get around to it yesterday.
>>>
>>> Acked-by: Rob Herring <rob.herring-bsGFqQB8/DxBDgjK7y7TUQ@public.gmane.org>
>>
>> I've just noticed in kernel/irq/irqdomain.c, that the kerneldoc for 
>> irq_domain_add() says that a valid ops structure is required, so perhaps 
>> this fix isn't right and we should just have an empty ops structure for 
>> simple controllers...
> 
> I wonder if this might make more sense instead.  I'll post a proper 
> patch if you think so.
> 
> Regards,
> 
> Jamie
> 
> diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
> index 200ce83..92c3484 100644
> --- a/kernel/irq/irqdomain.c
> +++ b/kernel/irq/irqdomain.c
> @@ -143,11 +143,6 @@ int irq_domain_simple_dt_translate(struct irq_domain *d,
>  	return 0;
>  }
>  
> -struct irq_domain_ops irq_domain_simple_ops = {
> -	.dt_translate = irq_domain_simple_dt_translate,
> -};
> -EXPORT_SYMBOL_GPL(irq_domain_simple_ops);
> -
>  /**
>   * irq_domain_create_simple() - Set up a 'simple' translation range
>   */
> @@ -181,4 +176,11 @@ void irq_domain_generate_simple(const struct of_device_id *match,
>  		pr_info("no node found\n");
>  }
>  EXPORT_SYMBOL_GPL(irq_domain_generate_simple);
> +#else /* CONFIG_OF_IRQ */
> +#define irq_domain_simple_dt_translate NULL
>  #endif /* CONFIG_OF_IRQ */
> +
> +struct irq_domain_ops irq_domain_simple_ops = {
> +	.dt_translate = irq_domain_simple_dt_translate,

.dt_translate is surrounded by CONFIG_OF, so it will break on !CONFIG_OF.

> +};
> +EXPORT_SYMBOL_GPL(irq_domain_simple_ops);

The extern declaration in irqdomain.h is surrounded by CONFIG_OF_IRQ, so
it needs to be moved out of that as well.

On Sparc, CONFIG_OF is true, but CONFIG_OF_IRQ is false. So that
combination needs to be handled as well.

Rob

Re: [PATCH] irq: fix possible null-pointer deref irq_domain_to_irq

[Jamie Iles]

Hi Rob,

On Sun, Nov 27, 2011 at 09:11:01AM -0600, Rob Herring wrote:
> On 11/25/2011 09:35 AM, Jamie Iles wrote:
> > I wonder if this might make more sense instead.  I'll post a proper 
> > patch if you think so.
> > 
> > Regards,
> > 
> > Jamie
> > 
> > diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
> > index 200ce83..92c3484 100644
> > --- a/kernel/irq/irqdomain.c
> > +++ b/kernel/irq/irqdomain.c
> > @@ -143,11 +143,6 @@ int irq_domain_simple_dt_translate(struct irq_domain *d,
> >  	return 0;
> >  }
> >  
> > -struct irq_domain_ops irq_domain_simple_ops = {
> > -	.dt_translate = irq_domain_simple_dt_translate,
> > -};
> > -EXPORT_SYMBOL_GPL(irq_domain_simple_ops);
> > -
> >  /**
> >   * irq_domain_create_simple() - Set up a 'simple' translation range
> >   */
> > @@ -181,4 +176,11 @@ void irq_domain_generate_simple(const struct of_device_id *match,
> >  		pr_info("no node found\n");
> >  }
> >  EXPORT_SYMBOL_GPL(irq_domain_generate_simple);
> > +#else /* CONFIG_OF_IRQ */
> > +#define irq_domain_simple_dt_translate NULL
> >  #endif /* CONFIG_OF_IRQ */
> > +
> > +struct irq_domain_ops irq_domain_simple_ops = {
> > +	.dt_translate = irq_domain_simple_dt_translate,
> 
> .dt_translate is surrounded by CONFIG_OF, so it will break on !CONFIG_OF.

Ahh, good spot!

> > +};
> > +EXPORT_SYMBOL_GPL(irq_domain_simple_ops);
> 
> The extern declaration in irqdomain.h is surrounded by CONFIG_OF_IRQ, so
> it needs to be moved out of that as well.
> 
> On Sparc, CONFIG_OF is true, but CONFIG_OF_IRQ is false. So that
> combination needs to be handled as well.

OK, I'm travelling for a couple of days now, but I'll respin something 
when I'm back.  I guess it's a case of always exporting the simple ops 
when CONFIG_IRQDOMAIN=y, and only setting the .dt_translate when 
CONFIG_OF_IRQ=y though.

Jamie