[PATCH v1] ARM: Feroceon: Fix OF node and mapping leaks in L2 init

Yuho Choi posted 1 patch 1 month, 1 week ago
arch/arm/mm/cache-feroceon-l2.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
[PATCH v1] ARM: Feroceon: Fix OF node and mapping leaks in L2 init
Posted by Yuho Choi 1 month, 1 week ago
of_find_matching_node() returns a referenced device node, and of_iomap()
creates a temporary mapping. feroceon_of_init() currently leaves both
resources around after init and also returns from the of_iomap() failure
path without dropping the node reference.

Unmap the register mapping after the register update and put the node on
both the failure and success paths.

Fixes: 4b8f7a11c9fb ("ARM: MM: Add DT binding for Feroceon L2 cache")
Signed-off-by: Yuho Choi <dbgh9129@gmail.com>
---
 arch/arm/mm/cache-feroceon-l2.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/arch/arm/mm/cache-feroceon-l2.c b/arch/arm/mm/cache-feroceon-l2.c
index 2bfefb252ffd..1fe931a09d79 100644
--- a/arch/arm/mm/cache-feroceon-l2.c
+++ b/arch/arm/mm/cache-feroceon-l2.c
@@ -373,16 +373,20 @@ int __init feroceon_of_init(void)
 	node = of_find_matching_node(NULL, feroceon_ids);
 	if (node && of_device_is_compatible(node, "marvell,kirkwood-cache")) {
 		base = of_iomap(node, 0);
-		if (!base)
+		if (!base) {
+			of_node_put(node);
 			return -ENOMEM;
+		}
 
 		if (l2_wt_override)
 			writel(readl(base) | L2_WRITETHROUGH_KIRKWOOD, base);
 		else
 			writel(readl(base) & ~L2_WRITETHROUGH_KIRKWOOD, base);
+		iounmap(base);
 	}
 
 	feroceon_l2_init(l2_wt_override);
+	of_node_put(node);
 
 	return 0;
 }
-- 
2.43.0
Re: [PATCH v1] ARM: Feroceon: Fix OF node and mapping leaks in L2 init
Posted by 최유호 1 week, 3 days ago
Hi,

Just a gentle reminder regarding this patch.

I would appreciate any feedback when you have a chance to review it.

Thanks for your time.

Best regards,
Yuho Choi

On Mon, 4 May 2026 at 20:32, Yuho Choi <dbgh9129@gmail.com> wrote:
>
> of_find_matching_node() returns a referenced device node, and of_iomap()
> creates a temporary mapping. feroceon_of_init() currently leaves both
> resources around after init and also returns from the of_iomap() failure
> path without dropping the node reference.
>
> Unmap the register mapping after the register update and put the node on
> both the failure and success paths.
>
> Fixes: 4b8f7a11c9fb ("ARM: MM: Add DT binding for Feroceon L2 cache")
> Signed-off-by: Yuho Choi <dbgh9129@gmail.com>
> ---
>  arch/arm/mm/cache-feroceon-l2.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm/mm/cache-feroceon-l2.c b/arch/arm/mm/cache-feroceon-l2.c
> index 2bfefb252ffd..1fe931a09d79 100644
> --- a/arch/arm/mm/cache-feroceon-l2.c
> +++ b/arch/arm/mm/cache-feroceon-l2.c
> @@ -373,16 +373,20 @@ int __init feroceon_of_init(void)
>         node = of_find_matching_node(NULL, feroceon_ids);
>         if (node && of_device_is_compatible(node, "marvell,kirkwood-cache")) {
>                 base = of_iomap(node, 0);
> -               if (!base)
> +               if (!base) {
> +                       of_node_put(node);
>                         return -ENOMEM;
> +               }
>
>                 if (l2_wt_override)
>                         writel(readl(base) | L2_WRITETHROUGH_KIRKWOOD, base);
>                 else
>                         writel(readl(base) & ~L2_WRITETHROUGH_KIRKWOOD, base);
> +               iounmap(base);
>         }
>
>         feroceon_l2_init(l2_wt_override);
> +       of_node_put(node);
>
>         return 0;
>  }
> --
> 2.43.0
>
Re: [PATCH v1] ARM: Feroceon: Fix OF node and mapping leaks in L2 init
Posted by Andrew Lunn 1 week, 3 days ago
On Tue, Jun 02, 2026 at 10:52:32PM -0400, 최유호 wrote:
> Hi,
> 
> Just a gentle reminder regarding this patch.
> 
> I would appreciate any feedback when you have a chance to review it.

What do you think happens to the machine if it cannot enable L2?  I
think you have much bigger problems than some minor resource leaks.

Please read

https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html

Does this patch fulfil those rules. Please could you point to bug
reports of the leaked resources, so we know it actually bothers
people.

	Andrew