[PATCHv2 1/1] omap: Ptr "isr_reg" tracked as NULL was dereferenced

[Evgeny Kuznetsov]

On Wed, 2010-10-13 at 17:20 +0530, ext Varadarajan, Charulatha wrote:
> 
> <<snip>>
> 
> > > >
> > > > From: Evgeny Kuznetsov <ext-eugeny.kuznetsov at nokia.com>
> > > >
> > > > Value of "isr_reg" pointer is depend on configuration and GPIO method.
> > > > Potentially it may have NULL value and it is dereferenced later
> > > > in code. If pointer is NULL there is some kernel issue.
> > >
> > > Can you elaborate?
> > "isr_reg" should not be NULL. But if it is NULL then there is kernel
> > bug. And WARN_ON() used to show it.
> > I did not see this bug, this is potentially may happen.
> > >
> > > > Warning and exit from function are added in this case.
> > > > Also compilation check is added for correct architecture
> > > > configuration.
> > > >
> > > > Signed-off-by: Evgeny Kuznetsov <EXT-Eugeny.Kuznetsov at nokia.com>
> > > > ---
> > > >  arch/arm/plat-omap/gpio.c |   18 ++++++++++++++++++
> > > >  1 files changed, 18 insertions(+), 0 deletions(-)
> > > >
> > > > diff --git a/arch/arm/plat-omap/gpio.c b/arch/arm/plat-omap/gpio.c
> > > > index c05c653..d04913c 100644
> > > > --- a/arch/arm/plat-omap/gpio.c
> > > > +++ b/arch/arm/plat-omap/gpio.c
> > > > @@ -1318,6 +1318,23 @@ static void gpio_irq_handler(unsigned int irq,
> > > > struct irq_desc *desc)
> > > >  	if (bank->method == METHOD_GPIO_44XX)
> > > >  		isr_reg = bank->base + OMAP4_GPIO_IRQSTATUS0;
> > > >  #endif
> > > > +
> > > > +#if !defined(CONFIG_ARCH_OMAP1) &&		\
> > > > +		!defined(CONFIG_ARCH_OMAP15XX) &&	\
> > > > +		!defined(CONFIG_ARCH_OMAP16XX) &&	\
> > > > +		!defined(CONFIG_ARCH_OMAP730) &&	\
> > > > +		!defined(CONFIG_ARCH_OMAP850) &&	\
> > > > +		!defined(CONFIG_ARCH_OMAP2) &&	\
> > > > +		!defined(CONFIG_ARCH_OMAP3) &&	\
> > > > +		!defined(CONFIG_ARCH_OMAP4)
> > > > +
> > > > +#error "Incorrect arch configuration"
> > >
> > > This is not required. If the architecture is not one of the above
> > > mentioned, gpio_irq_handler() will not be used/called at all.
> > This could be removed.
> > 
> > > Also all the possible gpio methods for a given OMAP architecture are
> > > already considered with "#ifdef"s and (bank->method) checks in
> > > gpio_irq_handler().
> > It is not cover all cased, e.g. for OMAP4 arch:
> > ....
> > 	#if defined(CONFIG_ARCH_OMAP4)
> > 		if (bank->method == METHOD_GPIO_44XX)
> > 			isr_reg = bank->base + OMAP4_GPIO_IRQSTATUS0;
> > 	#endif
> > .....
> > 
> > If (bank->method != METHOD_GPIO_44XX) then isr_reg will be NULL.
> > This should not happen, but potentially may have place.
> 
> When would it fail? If it is CONFIG_ARCH_OMAP4, the gpio method can
> only be METHOD_GPIO_44XX. Else if it is for some other OMAP architecture,
> the gpio_method is similarly taken care. So this cannot happen.
This is similar check as e.g. in _enable_gpio_irqbank() there default
case for bank->methos switch is WARN_ON(1).
Just to warn in case.
> 
> > 
> > >
> > > > +
> > > > +#endif
> > > > +
> > > > +	if (WARN_ON(!isr_reg))
> > > > +		goto exit;
> > >
> > > For the above mentioned reason, this isr_reg would be non-NULL. Have
> > > you observed this error anytime?
> > I did not see this bug, this is potentially may happen.
> > >
> > > Also, the omap-gpio code has similar code spread all over and has to be
> > > anyway cleaned-up. Is there any reason why gpio_irq_handler() alone is
> > > addressed in this patch?
> > Here "isr_reg" is used later in code and may cause oops if it is NULL.
> > >
> > > > +
> > > >  	while(1) {
> > > >  		u32 isr_saved, level_mask = 0;
> > > >  		u32 enabled;
> > > > @@ -1377,6 +1394,7 @@ static void gpio_irq_handler(unsigned int irq,
> > > > struct irq_desc *desc)
> > > >  	configured, we must unmask the bank interrupt only after
> > > >  	handler(s) are executed in order to avoid spurious bank
> > > >  	interrupt */
> > > > +exit:
> > > >  	if (!unmasked)
> > > >  		desc->chip->unmask(irq);
> > > >
> > > > --
> > > > 1.6.3.3
> > >
> > 
>