Re: [PATCH 2/3] x86/vdso: Rearrange do_hres() to improve code generation

[Thomas Gleixner]

On Thu, 4 Oct 2018, Andy Lutomirski wrote:
> index 18c8a78d1ec9..419de7552c2f 100644
> --- a/arch/x86/entry/vdso/vclock_gettime.c
> +++ b/arch/x86/entry/vdso/vclock_gettime.c
> @@ -147,10 +147,9 @@ notrace static int do_hres(clockid_t clk, struct timespec *ts)
>  
>  	do {
>  		seq = gtod_read_begin(gtod);
> -		ts->tv_sec = base->sec;
> +		cycles = vgetcyc(gtod->vclock_mode);
>  		ns = base->nsec;
>  		last = gtod->cycle_last;
> -		cycles = vgetcyc(gtod->vclock_mode);
>  		if (unlikely((s64)cycles < 0))
>  			return vdso_fallback_gettime(clk, ts);
>  		if (cycles > last)
> @@ -158,7 +157,7 @@ notrace static int do_hres(clockid_t clk, struct timespec *ts)
>  		ns >>= gtod->shift;
>  	} while (unlikely(gtod_read_retry(gtod, seq)));
>  
> -	ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
> +	ts->tv_sec = base->sec + __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);

You cannot access base->sec outside of the seqcount protected region. It
might have been incremented by now and you'll get a time jump by a full
second.

Thanks,

	tglx