Re: How is Code in do_sys_settimeofday() safe in case of SMP andNest Kernel Path?

[Christoph Lameter]

On Sun, 1 Oct 2006, Dong Feng wrote:

> --- kernel/time.c.orig	2006-09-30 23:21:29.000000000 +0800
> +++ kernel/time.c	2006-09-30 23:38:18.000000000 +0800
> @@ -107,7 +107,16 @@ asmlinkage long sys_gettimeofday(struct
> 			return -EFAULT;
> 	}
> 	if (unlikely(tz != NULL)) {
> -		if (copy_to_user(tz, &sys_tz, sizeof(sys_tz)))
> +		struct timezone ktz;
> +		unsigned long seq;
> +
> +		do {
> +                	seq = read_seqbegin(&xtime_lock);
> +			ktz.tz_minuteswest = sys_tz.tz_minuteswest;
> +			ktz.tz_dsttime = sys_tz.tz_dsttime;
> +        	} while (unlikely(read_seqretry(&xtime_lock, seq)));
> +
> +		if (copy_to_user(tz, &ktz, sizeof(ktz)))
> 			return -EFAULT;

I really hate adding overhead to gettimeofday() and we would have to take 
the seqlock in all places when we reference tz. Maybe we can tolerate the 
resulting race?

If we assume word size transfers then we only have an issue on 32 bit 
platforms. The result of the race would be that tz_minuteswest and 
tz_dsttime disagree. So we may get daylight savings time wrong.
But then we are already changing the timezone and are potentially warping time.
gettimofday may be unstable anyways. So it may be okay to leave the race 
in. Just add some comments explaining the situation.
-
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/