[PATCH] of/irq: add missing of_node_put() for interrupt parent node

[PATCH] of/irq: add missing of_node_put() for interrupt parent node

[Clément Léger]

After calling of_irq_parse_one(), the node provided in the of_phandle_args
has a refcount increment by one. Add missing of_node_put in of_irq_get()
to decrement the refcount once used.

Signed-off-by: Clément Léger <clement.leger@bootlin.com>
---

While debugging a refcount problem with OF_DYNAMIC enabled (which is
actually the only case were node refcount are really used), I noticed that
platform_get_irq() was actually incrementing the refcount of an interrupt
controller node. Digging into that function shows that it calls
of_irq_get() which calls of_irq_parse_one() and finally of_irq_parse_raw().
Since it seems sane that the node returned in the of_phandle_args has a
refcount incremented, I thought it is better to put the of_node_put() in
the user even though it was hard to find any user doing so.

 drivers/of/irq.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/drivers/of/irq.c b/drivers/of/irq.c
index e9bf5236ed89..174900072c18 100644
--- a/drivers/of/irq.c
+++ b/drivers/of/irq.c
@@ -438,10 +438,16 @@ int of_irq_get(struct device_node *dev, int index)
 		return rc;
 
 	domain = irq_find_host(oirq.np);
-	if (!domain)
-		return -EPROBE_DEFER;
+	if (!domain) {
+		rc = -EPROBE_DEFER;
+		goto out;
+	}
 
-	return irq_create_of_mapping(&oirq);
+	rc = irq_create_of_mapping(&oirq);
+out:
+	of_node_put(oirq.np);
+
+	return rc;
 }
 EXPORT_SYMBOL_GPL(of_irq_get);
 
-- 
2.39.0

Re: [PATCH] of/irq: add missing of_node_put() for interrupt parent node

[Rob Herring]

On Tue, 17 Jan 2023 15:49:29 +0100, Clément Léger wrote:
> After calling of_irq_parse_one(), the node provided in the of_phandle_args
> has a refcount increment by one. Add missing of_node_put in of_irq_get()
> to decrement the refcount once used.
> 
> Signed-off-by: Clément Léger <clement.leger@bootlin.com>
> ---
> 
> While debugging a refcount problem with OF_DYNAMIC enabled (which is
> actually the only case were node refcount are really used), I noticed that
> platform_get_irq() was actually incrementing the refcount of an interrupt
> controller node. Digging into that function shows that it calls
> of_irq_get() which calls of_irq_parse_one() and finally of_irq_parse_raw().
> Since it seems sane that the node returned in the of_phandle_args has a
> refcount incremented, I thought it is better to put the of_node_put() in
> the user even though it was hard to find any user doing so.
> 
>  drivers/of/irq.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
> 

Applied, thanks!

Re: [PATCH] of/irq: add missing of_node_put() for interrupt parent node

[Rob Herring]

+Saravana, Jean

On Tue, Jan 17, 2023 at 8:47 AM Clément Léger <clement.leger@bootlin.com> wrote:
>
> After calling of_irq_parse_one(), the node provided in the of_phandle_args
> has a refcount increment by one. Add missing of_node_put in of_irq_get()
> to decrement the refcount once used.
>
> Signed-off-by: Clément Léger <clement.leger@bootlin.com>
> ---
>
> While debugging a refcount problem with OF_DYNAMIC enabled (which is
> actually the only case were node refcount are really used), I noticed that
> platform_get_irq() was actually incrementing the refcount of an interrupt
> controller node. Digging into that function shows that it calls
> of_irq_get() which calls of_irq_parse_one() and finally of_irq_parse_raw().
> Since it seems sane that the node returned in the of_phandle_args has a
> refcount incremented, I thought it is better to put the of_node_put() in
> the user even though it was hard to find any user doing so.

While investigating [1], I stumbled back on this. Was the failing case
you had using interrupts-extended? It looks to me like that path has a
get, but the 'interrupts' path does not. If so, this change is wrong.

Rob

[1] https://lore.kernel.org/all/20230228174019.4004581-1-jjhiblot@traphandler.com/


>
>  drivers/of/irq.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/of/irq.c b/drivers/of/irq.c
> index e9bf5236ed89..174900072c18 100644
> --- a/drivers/of/irq.c
> +++ b/drivers/of/irq.c
> @@ -438,10 +438,16 @@ int of_irq_get(struct device_node *dev, int index)
>                 return rc;
>
>         domain = irq_find_host(oirq.np);
> -       if (!domain)
> -               return -EPROBE_DEFER;
> +       if (!domain) {
> +               rc = -EPROBE_DEFER;
> +               goto out;
> +       }
>
> -       return irq_create_of_mapping(&oirq);
> +       rc = irq_create_of_mapping(&oirq);
> +out:
> +       of_node_put(oirq.np);
> +
> +       return rc;
>  }
>  EXPORT_SYMBOL_GPL(of_irq_get);
>
> --
> 2.39.0
>

Re: [PATCH] of/irq: add missing of_node_put() for interrupt parent node

[Clément Léger]

Le Tue, 28 Feb 2023 17:14:18 -0600,
Rob Herring <robh+dt@kernel.org> a écrit :

> +Saravana, Jean
> 
> On Tue, Jan 17, 2023 at 8:47 AM Clément Léger <clement.leger@bootlin.com> wrote:
> >
> > After calling of_irq_parse_one(), the node provided in the of_phandle_args
> > has a refcount increment by one. Add missing of_node_put in of_irq_get()
> > to decrement the refcount once used.
> >
> > Signed-off-by: Clément Léger <clement.leger@bootlin.com>
> > ---
> >
> > While debugging a refcount problem with OF_DYNAMIC enabled (which is
> > actually the only case were node refcount are really used), I noticed that
> > platform_get_irq() was actually incrementing the refcount of an interrupt
> > controller node. Digging into that function shows that it calls
> > of_irq_get() which calls of_irq_parse_one() and finally of_irq_parse_raw().
> > Since it seems sane that the node returned in the of_phandle_args has a
> > refcount incremented, I thought it is better to put the of_node_put() in
> > the user even though it was hard to find any user doing so.  
> 
> While investigating [1], I stumbled back on this. Was the failing case
> you had using interrupts-extended? It looks to me like that path has a
> get, but the 'interrupts' path does not. If so, this change is wrong.

In my case, it was with a classic "interrupts" property. I can take
another look at the internal code to be sure this fix is correct.

Clément

> 
> Rob
> 
> [1] https://lore.kernel.org/all/20230228174019.4004581-1-jjhiblot@traphandler.com/
> 
> 
> >
> >  drivers/of/irq.c | 12 +++++++++---
> >  1 file changed, 9 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/of/irq.c b/drivers/of/irq.c
> > index e9bf5236ed89..174900072c18 100644
> > --- a/drivers/of/irq.c
> > +++ b/drivers/of/irq.c
> > @@ -438,10 +438,16 @@ int of_irq_get(struct device_node *dev, int index)
> >                 return rc;
> >
> >         domain = irq_find_host(oirq.np);
> > -       if (!domain)
> > -               return -EPROBE_DEFER;
> > +       if (!domain) {
> > +               rc = -EPROBE_DEFER;
> > +               goto out;
> > +       }
> >
> > -       return irq_create_of_mapping(&oirq);
> > +       rc = irq_create_of_mapping(&oirq);
> > +out:
> > +       of_node_put(oirq.np);
> > +
> > +       return rc;
> >  }
> >  EXPORT_SYMBOL_GPL(of_irq_get);
> >
> > --
> > 2.39.0
> >  



-- 
Clément Léger,
Embedded Linux and Kernel engineer at Bootlin
https://bootlin.com