Re: [PATCH] kfifo: fix warn_unused_result

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 11 Dec 2009 10:18:28 +0100 Stefani Seibold <stefani@seibold.net> wrote:

> As requested by Andrew Morton:
> 
> This patch fix the "ignoring return value of '...', declared with
> attribute warn_unused_result" compiler warning in several users of the
> new kfifo API.
> 
> The patch-set is against current mm tree from 11-Dec-2009
> 
> ...
>
> --- mmotm/drivers/char/nozomi.c	2009-12-11 08:31:46.670736197 +0100
> +++ linux-2.6.32/drivers/char/nozomi.c	2009-12-11 09:25:46.941436203 +0100
> @@ -685,8 +685,9 @@ static int nozomi_read_config_table(stru
>  		dump_table(dc);
>  
>  		for (i = PORT_MDM; i < MAX_PORT; i++) {
> -			kfifo_alloc(&dc->port[i].fifo_ul,
> -				FIFO_BUFFER_SIZE_UL, GFP_ATOMIC);
> +			if (kfifo_alloc(&dc->port[i].fifo_ul,
> +				FIFO_BUFFER_SIZE_UL, GFP_ATOMIC))
> +					BUG();

No, we can't do this.  GFP_ATOMIC allocations are unreliable and can
fail.  The calling code *has* to detect the failure and then take some
recovery action.

It would be better to leave the warning in place, rather than to add
this runtime landmine.

>  		input_sync(kp.dev);
> -		kfifo_in_locked(&sonypi_device.input_fifo,
> +		if (kfifo_in_locked(&sonypi_device.input_fifo,
>  			(unsigned char *)&kp, sizeof(kp),
> -			&sonypi_device.input_fifo_lock);
> +			&sonypi_device.input_fifo_lock) != sizeof(kp))
> +				BUG();

The rest of the patch seems to be adding BUG()s if kfifo_in() fails. 
All over the place.

If that's the appropriate way to handle failure for these callsites
then it would be neater to do this in the callee.  ie, add a new

unsigned int kfifo_in_nonpartial(struct kfifo *fifo,
                        const unsigned char *from, unsigned int len)
{
	unsigned int ret = kfifo_in(fifo, from, len);

	BUG_ON(ret != len);
	return ret;
}