[PATCH] Input: amijoy - fix memory region leak on registration failure

[PATCH] Input: amijoy - fix memory region leak on registration failure

[Thorsten Blum]

If input_register_device() fails in amijoy_init(), release the memory
region for port i before cleaning up the previously registered ports.

Fixes: 127278ce2254 ("Input: joysticks - handle errors when registering input devices")
Cc: stable@vger.kernel.org
Signed-off-by: Thorsten Blum <blum@kernel.org>
---
 drivers/input/joystick/amijoy.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/input/joystick/amijoy.c b/drivers/input/joystick/amijoy.c
index cdb2b8f15348..3c2357ba7dc3 100644
--- a/drivers/input/joystick/amijoy.c
+++ b/drivers/input/joystick/amijoy.c
@@ -132,6 +132,7 @@ static int __init amijoy_init(void)
 		err = input_register_device(amijoy_dev[i]);
 		if (err) {
 			input_free_device(amijoy_dev[i]);
+			release_mem_region(CUSTOM_PHYSADDR + 10 + i * 2, 2);
 			goto fail;
 		}
 	}

Re: [PATCH] Input: amijoy - fix memory region leak on registration failure

[Dmitry Torokhov]

Hi Thorsten,

On Sat, Nov 07, 2026 at 05:01:57PM +0100, Thorsten Blum wrote:
> If input_register_device() fails in amijoy_init(), release the memory
> region for port i before cleaning up the previously registered ports.
>
> Fixes: 127278ce2254 ("Input: joysticks - handle errors when registering input devices")
> Cc: stable@vger.kernel.org
...
>  		err = input_register_device(amijoy_dev[i]);
>  		if (err) {
>  			input_free_device(amijoy_dev[i]);
> +			release_mem_region(CUSTOM_PHYSADDR + 10 + i * 2, 2);
>  			goto fail;
>  		}

Do we really need this in stable? It has been like this for 20 years,
and only affects an unlikely input_register_device() failure (basically
-ENOMEM during init) on Amiga hardware, so please drop Cc: stable.

Without the constraint of keeping the diff minimal for stable, could we
also clean up the error handling here instead of unwinding inline (and
out of order relative to input_allocate_device() / request_mem_region())?
Either add cascading err_release_mem / err_free_dev labels before the
unregister loop so unwinding happens in reverse order without repeating
input_free_device(), or factor per-port initialization out into a small
amijoy_setup_port() helper.

Thanks.

-- 
Dmitry