Re: [PATCH 5/10] Input: adp5588-keys - added a check key_val

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

Hi Denis,

On Thu, Sep 19, 2024 at 05:29:14PM +0300, Denis Arefev wrote:
> No upstream commit exists for this commit.

Sorry, what does this mean?

> 
> If the adp5588_read function returns 0, then there will be an
> overflow of the kpad->keycode[key_val - 1] buffer.
> 
> If the adp5588_read function returns a negative value, then the
> logic is broken - the wrong value is used as an index of
> the kpad->keycode array.
> 
> Found by Linux Verification Center (linuxtesting.org) with SVACE.
> 
> Fixes: 69a4af606ed4 ("Input: adp5588-keys - support GPI events for ADP5588 devices")
> Cc: stable@vger.kernel.org
> Signed-off-by: Denis Arefev <arefev@swemel.ru>
> ---
>  drivers/input/keyboard/adp5588-keys.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/input/keyboard/adp5588-keys.c b/drivers/input/keyboard/adp5588-keys.c
> index 90a59b973d00..19be8054eb5f 100644
> --- a/drivers/input/keyboard/adp5588-keys.c
> +++ b/drivers/input/keyboard/adp5588-keys.c
> @@ -272,6 +272,8 @@ static void adp5588_report_events(struct adp5588_kpad *kpad, int ev_cnt)
>  		int key = adp5588_read(kpad->client, Key_EVENTA + i);
>  		int key_val = key & KEY_EV_MASK;
>  
> +		if (key_val <= 0)
> +			continue;

We should be checking the original value (key) and not masked value.
Masked value will never be negative.

		int key, key_vali, key_press;

		key = adp5588_read(kpad->client, Key_EVENTA + i);
		if (key < 0)
			continue;

		key_val = key & key & KEY_EV_MASK;
		key_press = key & KEY_EV_PRESSED;
		if (key_val >= GPI_PIN_BASE && key_val <= GPI_PIN_END) {
			...
		} else if (key_val > 0) {
			...
		}

Thanks.

-- 
Dmitry