Re: [PATCH] kernel/sysctl_binary.c: improve the usage of return value 'result'

[Eric W. Biederman]

Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> writes:

> On Tue, 06 Aug 2013 15:29:42 +0800 Chen Gang <gang.chen@xxxxxxxxxxx> wrote:
>
>> Improve the usage of return value 'result', so not only can make code
>> clearer to readers, but also can improve the performance.
>
> It used to be pervasive kernel style do to
>
> 	ret = -ENOMEM;
> 	foo = alloc(...);
> 	if (!foo)
> 		goto out;
>
> whereas nowadays people usually do the more straightforward
>
> 	foo = alloc(...);
> 	if (!foo) {
> 		ret = -ENOMEM;
> 		goto out;
> 	}
>
> The thinking was that the old style generated better code, but for the
> life of me I can't remember why :(

Because doing the assignment outside of the if() goto .  Allows the
compiler to emit the if() goto as a single branch.

While a smart compiler may perform the code motion across the branch,
it is much easier for the compiler to branch to somewhere else perform
the assignment and then branch out.

Eric


> Your patch switches from old-style to new-style.  And it appears to
> have increased the text size.  I did this, to switch three sites back
> to old-style:
>
> --- a/kernel/sysctl_binary.c~kernel-sysctl_binaryc-improve-the-usage-of-return-value-result-fix
> +++ a/kernel/sysctl_binary.c
> @@ -941,17 +941,15 @@ static ssize_t bin_string(struct file *f
>  		copied = result;
>  		lastp = oldval + copied - 1;
>  
> -		if (get_user(ch, lastp)) {
> -			result = -EFAULT;
> +		result = -EFAULT;
> +		if (get_user(ch, lastp))
>  			goto out;
> -		}
>  
>  		/* Trim off the trailing newline */
>  		if (ch == '\n') {
> -			if (put_user('\0', lastp)) {
> -				result = -EFAULT;
> +			result = -EFAULT;
> +			if (put_user('\0', lastp))
>  				goto out;
> -			}
>  			copied -= 1;
>  		}
>  	}
> @@ -976,11 +974,10 @@ static ssize_t bin_intvec(struct file *f
>  	char *buffer;
>  	ssize_t result;
>  
> +	result = -ENOMEM;
>  	buffer = kmalloc(BUFSZ, GFP_KERNEL);
> -	if (!buffer) {
> -		result = -ENOMEM;
> +	if (!buffer)
>  		goto out;
> -	}
>  
>  	if (oldval && oldlen) {
>  		unsigned __user *vec = oldval;
> _
>
> and kernel/sysctl_binary.o's .text got six bytes smaller.
>
> Now, smaller text doesn't mean faster code.  But it probably means
> larger cache footprint, which can mean slower code.
>
> IOW, it isn't obvious that this was an improvement.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/