Re: [PATCH][next] leds: Avoid -Wflex-array-member-not-at-end warning

From: Thomas Weißschuh
Date: Sat Mar 29 2025 - 05:33:01 EST

Hi!

On 2025-03-28 12:51:02-0600, Gustavo A. R. Silva wrote:
> On 28/03/25 12:31, Thomas Weißschuh wrote:
> > On 2025-03-28 08:33:22-0600, Gustavo A. R. Silva wrote:
> > > -Wflex-array-member-not-at-end was introduced in GCC-14, and we are
> > > getting ready to enable it, globally.
> > >
> > > Use the `DEFINE_RAW_FLEX()` helper for an on-stack definition of
> > > a flexible structure where the size of the flexible-array member
> > > is known at compile-time, and refactor the rest of the code,
> > > accordingly.
> > >
> > > So, with these changes, fix the following warning:
> > >
> > > drivers/leds/leds-cros_ec.c:70:40: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]
> > >
> > > Signed-off-by: Gustavo A. R. Silva <gustavoars@xxxxxxxxxx>
> > > ---
> > > drivers/leds/leds-cros_ec.c | 26 +++++++++++---------------
> > > 1 file changed, 11 insertions(+), 15 deletions(-)
> > >
> > > diff --git a/drivers/leds/leds-cros_ec.c b/drivers/leds/leds-cros_ec.c
> > > index 275522b81ea5..6eab0474f52d 100644
> > > --- a/drivers/leds/leds-cros_ec.c
> > > +++ b/drivers/leds/leds-cros_ec.c
> > > @@ -66,24 +66,20 @@ static int cros_ec_led_send_cmd(struct cros_ec_device *cros_ec,
> > > union cros_ec_led_cmd_data *arg)
> > > {
> > > int ret;
> > > - struct {
> > > - struct cros_ec_command msg;
> > > - union cros_ec_led_cmd_data data;
> > > - } __packed buf = {
> > > - .msg = {
> > > - .version = 1,
> > > - .command = EC_CMD_LED_CONTROL,
> > > - .insize = sizeof(arg->resp),
> > > - .outsize = sizeof(arg->req),
> > > - },
> > > - .data.req = arg->req
> > > - };
> > > -
> > > - ret = cros_ec_cmd_xfer_status(cros_ec, &buf.msg);
> > > + DEFINE_RAW_FLEX(struct cros_ec_command, msg, data,
> > > + sizeof(union cros_ec_led_cmd_data));
> > > +
> > > + msg->version = 1;
> > > + msg->command = EC_CMD_LED_CONTROL;
> > > + msg->insize = sizeof(arg->resp);
> > > + msg->outsize = sizeof(arg->req);
> > > + *(struct ec_params_led_control *)msg->data = arg->req;
> >
> > To be honest this looks really ugly and it's not at all obvious what is
>
> We can do something like this, instead:
>
> diff --git a/drivers/leds/leds-cros_ec.c b/drivers/leds/leds-cros_ec.c
> index 275522b81ea5..c7235f4e577b 100644
> --- a/drivers/leds/leds-cros_ec.c
> +++ b/drivers/leds/leds-cros_ec.c
> @@ -66,24 +66,24 @@ static int cros_ec_led_send_cmd(struct cros_ec_device *cros_ec,
> union cros_ec_led_cmd_data *arg)
> {
> int ret;
> - struct {
> - struct cros_ec_command msg;
> - union cros_ec_led_cmd_data data;
> - } __packed buf = {
> - .msg = {
> - .version = 1,
> - .command = EC_CMD_LED_CONTROL,
> - .insize = sizeof(arg->resp),
> - .outsize = sizeof(arg->req),
> - },
> - .data.req = arg->req
> - };
> -
> - ret = cros_ec_cmd_xfer_status(cros_ec, &buf.msg);
> + DEFINE_RAW_FLEX(struct cros_ec_command, msg, data,
> + sizeof(union cros_ec_led_cmd_data));
> + struct ec_params_led_control *req =
> + (struct ec_params_led_control *)msg->data;
> + struct ec_response_led_control *resp =
> + (struct ec_response_led_control *)msg->data;
> +
> + msg->version = 1;
> + msg->command = EC_CMD_LED_CONTROL;
> + msg->insize = sizeof(arg->resp);
> + msg->outsize = sizeof(arg->req);
> + *req = arg->req;
> +
> + ret = cros_ec_cmd_xfer_status(cros_ec, msg);
> if (ret < 0)
> return ret;
>
> - arg->resp = buf.data.resp;
> + arg->resp = *resp;
>
> return 0;
> }

My issue was with the general usage of DEFINE_RAW_FLEX(), the reply was
placed badly. For somebody not knowing this construct it's not clear
what is happening under the hood.
It's probably fine in a regular header file with some explanation,
but in a random driver it looks off.

The following is what I had in mind. Now actually tested.

--- a/drivers/leds/leds-cros_ec.c
+++ b/drivers/leds/leds-cros_ec.c
@@ -60,31 +60,18 @@ static inline struct cros_ec_led_priv *cros_ec_led_cdev_to_priv(struct led_class
union cros_ec_led_cmd_data {
struct ec_params_led_control req;
struct ec_response_led_control resp;
-} __packed;
+};

static int cros_ec_led_send_cmd(struct cros_ec_device *cros_ec,
union cros_ec_led_cmd_data *arg)
{
int ret;
- struct {
- struct cros_ec_command msg;
- union cros_ec_led_cmd_data data;
- } __packed buf = {
- .msg = {
- .version = 1,
- .command = EC_CMD_LED_CONTROL,
- .insize = sizeof(arg->resp),
- .outsize = sizeof(arg->req),
- },
- .data.req = arg->req
- };
-
- ret = cros_ec_cmd_xfer_status(cros_ec, &buf.msg);
+
+ ret = cros_ec_cmd(cros_ec, 1, EC_CMD_LED_CONTROL,
+ &arg->req, sizeof(arg->req), &arg->resp, sizeof(arg->resp));
if (ret < 0)
return ret;

- arg->resp = buf.data.resp;
-
return 0;
}


>
> as in other cases:
>
> https://lore.kernel.org/linux-hardening/Z-a4meHAy-t58bcE@kspp/
>
> -Gustavo
>
> > going on. We have the utility function cros_ec_cmd() which would be the
> > nicer alternative. (Without having verified that it avoids the warning).
> > While it is slightly more expensive, I don't think it matters.
> > And if it does, the helper can be optimized.
> >
> > (The same goes for my other cros_ec drivers)
> >
> > > +
> > > + ret = cros_ec_cmd_xfer_status(cros_ec, msg);
> > > if (ret < 0)
> > > return ret;
> > > - arg->resp = buf.data.resp;
> > > + arg->resp = *(struct ec_response_led_control *)msg->data;
> > > return 0;
> > > }


Thomas