[PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[Michael Walle]

Up until commit 6a45b0e2589f ("gpiolib: Introduce
gpiochip_irqchip_add_domain()") all irq_domains were allocated
by gpiolib itself and thus gpiolib also takes care of freeing it.

With gpiochip_irqchip_add_domain() a user of gpiolib can associate an
irq_domain with the gpio_chip. This irq_domain is not managed by
gpiolib and therefore must not be freed by gpiolib.

Fixes: 6a45b0e2589f ("gpiolib: Introduce gpiochip_irqchip_add_domain()")
Reported-by: Jiawen Wu <jiawenwu@trustnetic.com>
Signed-off-by: Michael Walle <mwalle@kernel.org>
---
 drivers/gpio/gpiolib.c      | 3 ++-
 include/linux/gpio/driver.h | 8 ++++++++
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index 7c6fefcbebd1..b0e3a3de2bf2 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -1740,7 +1740,7 @@ static void gpiochip_irqchip_remove(struct gpio_chip *gc)
 	}
 
 	/* Remove all IRQ mappings and delete the domain */
-	if (gc->irq.domain) {
+	if (!gc->irq.domain_is_ext && gc->irq.domain) {
 		unsigned int irq;
 
 		for (offset = 0; offset < gc->ngpio; offset++) {
@@ -1786,6 +1786,7 @@ int gpiochip_irqchip_add_domain(struct gpio_chip *gc,
 
 	gc->to_irq = gpiochip_to_irq;
 	gc->irq.domain = domain;
+	gc->irq.domain_is_ext = true;
 
 	/*
 	 * Using barrier() here to prevent compiler from reordering
diff --git a/include/linux/gpio/driver.h b/include/linux/gpio/driver.h
index 5c6db5533be6..3dbfed83823b 100644
--- a/include/linux/gpio/driver.h
+++ b/include/linux/gpio/driver.h
@@ -251,6 +251,14 @@ struct gpio_irq_chip {
 	 */
 	bool initialized;
 
+	/**
+	 * @domain_is_ext:
+	 *
+	 * True it the irq_domain was allocated outside of gpiolib, in which
+	 * case gpiolib won't free the irq_domain itself.
+	 */
+	bool domain_is_ext;
+
 	/**
 	 * @init_hw: optional routine to initialize hardware before
 	 * an IRQ chip will be added. This is quite useful when

---
base-commit: 659140df086c67f5eb77bbffcab34a7e8accd2a8
change-id: 20230616-fixes-gpiolib-irq-domain-2b69ecae0fdd

Re: [PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[Linus Walleij]

On Fri, Jun 16, 2023 at 9:30 AM Michael Walle <mwalle@kernel.org> wrote:

> Up until commit 6a45b0e2589f ("gpiolib: Introduce
> gpiochip_irqchip_add_domain()") all irq_domains were allocated
> by gpiolib itself and thus gpiolib also takes care of freeing it.
>
> With gpiochip_irqchip_add_domain() a user of gpiolib can associate an
> irq_domain with the gpio_chip. This irq_domain is not managed by
> gpiolib and therefore must not be freed by gpiolib.
>
> Fixes: 6a45b0e2589f ("gpiolib: Introduce gpiochip_irqchip_add_domain()")
> Reported-by: Jiawen Wu <jiawenwu@trustnetic.com>
> Signed-off-by: Michael Walle <mwalle@kernel.org>

Oh right.

I guess we can't devres it some way...
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

Yours,
Linus Walleij

Re: [PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[Andy Shevchenko]

On Fri, Jun 16, 2023 at 09:30:06AM +0200, Michael Walle wrote:
> Up until commit 6a45b0e2589f ("gpiolib: Introduce
> gpiochip_irqchip_add_domain()") all irq_domains were allocated
> by gpiolib itself and thus gpiolib also takes care of freeing it.
> 
> With gpiochip_irqchip_add_domain() a user of gpiolib can associate an
> irq_domain with the gpio_chip. This irq_domain is not managed by
> gpiolib and therefore must not be freed by gpiolib.

With or without below nit-pick
Reviewed-by: Andy Shevchenko <andy@kernel.org>

...

>  	/* Remove all IRQ mappings and delete the domain */
> -	if (gc->irq.domain) {
> +	if (!gc->irq.domain_is_ext && gc->irq.domain) {

Looking at this, perhaps positive about allocation?

	domain_is_allocated?

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[Michael Walle]

Am 2023-06-16 16:13, schrieb Andy Shevchenko:
> On Fri, Jun 16, 2023 at 09:30:06AM +0200, Michael Walle wrote:
>> Up until commit 6a45b0e2589f ("gpiolib: Introduce
>> gpiochip_irqchip_add_domain()") all irq_domains were allocated
>> by gpiolib itself and thus gpiolib also takes care of freeing it.
>> 
>> With gpiochip_irqchip_add_domain() a user of gpiolib can associate an
>> irq_domain with the gpio_chip. This irq_domain is not managed by
>> gpiolib and therefore must not be freed by gpiolib.
> 
> With or without below nit-pick
> Reviewed-by: Andy Shevchenko <andy@kernel.org>
> 
> ...
> 
>>  	/* Remove all IRQ mappings and delete the domain */
>> -	if (gc->irq.domain) {
>> +	if (!gc->irq.domain_is_ext && gc->irq.domain) {
> 
> Looking at this, perhaps positive about allocation?
> 
> 	domain_is_allocated?

I mean all domains are allocated ;)

domain_is_allocated_externally? I'm bad with short and concise
names..

-michael

Re: [PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[Andy Shevchenko]

On Fri, Jun 16, 2023 at 9:02 PM Michael Walle <michael@walle.cc> wrote:
> Am 2023-06-16 16:13, schrieb Andy Shevchenko:
> > On Fri, Jun 16, 2023 at 09:30:06AM +0200, Michael Walle wrote:

...

> >>      /* Remove all IRQ mappings and delete the domain */
> >> -    if (gc->irq.domain) {
> >> +    if (!gc->irq.domain_is_ext && gc->irq.domain) {
> >
> > Looking at this, perhaps positive about allocation?
> >
> >       domain_is_allocated?
>
> I mean all domains are allocated ;)
>
> domain_is_allocated_externally? I'm bad with short and concise
> names..

Naming is one of the hardest problems in software...
Your variant is long, but conscious. I dunno. Bart, Linus, do you have
any ideas?

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] gpiolib: Fix irq_domain resource tracking for gpiochip_irqchip_add_domain()

[Linus Walleij]

On Sat, Jun 17, 2023 at 4:25 PM Andy Shevchenko
<andy.shevchenko@gmail.com> wrote:
> On Fri, Jun 16, 2023 at 9:02 PM Michael Walle <michael@walle.cc> wrote:
> > Am 2023-06-16 16:13, schrieb Andy Shevchenko:
> > > On Fri, Jun 16, 2023 at 09:30:06AM +0200, Michael Walle wrote:
>
> ...
>
> > >>      /* Remove all IRQ mappings and delete the domain */
> > >> -    if (gc->irq.domain) {
> > >> +    if (!gc->irq.domain_is_ext && gc->irq.domain) {
> > >
> > > Looking at this, perhaps positive about allocation?
> > >
> > >       domain_is_allocated?
> >
> > I mean all domains are allocated ;)
> >
> > domain_is_allocated_externally? I'm bad with short and concise
> > names..
>
> Naming is one of the hardest problems in software...
> Your variant is long, but conscious. I dunno. Bart, Linus, do you have
> any ideas?

No, this name is fine. It is meant for human readers and they
will immediately understand.

Yours,
Linus Walleij