Hi,
On 23 February 2017 at 20:55, Chris Taylor <ctaylor@xxxxxxxxxx> wrote:
> There is an issue in FIO when using the detailed bandwidth and iops logging with averaging over a period of time > 1 second.
> It seems that usecs overflows which later causes negative time diff values resulting in skewed toward 0 results.
> I have attached a potential fix that should prevent usecs from going beyond 1000000.
Is this problem shown by the following job:
./fio --ioengine=null --write_bw_log=/dev/null --log_avg_msec=2000
--size=100M --runtime=10s --time_based --name=overflow
Among Clang's undefined behaviour sanitizer output was this for the above:
stat.c:2265:43: runtime error: unsigned integer overflow: 1999 - 2000
cannot be represented in type 'unsigned long'
stat.c:2265:28: runtime error: unsigned integer overflow: 3999 -
4294967295 cannot be represented in type 'unsigned long'
> diff --git a/time.c b/time.c
> index f5dc049..c748bee 100644
> --- a/time.c
> +++ b/time.c
> @@ -8,11 +8,18 @@ static unsigned long ns_granularity;
>
> void timeval_add_msec(struct timeval *tv, unsigned int msec)
> {
> - tv->tv_usec += 1000 * msec;
> - if (tv->tv_usec >= 1000000) {
> - tv->tv_usec -= 1000000;
> - tv->tv_sec++;
> - }
> + int adj_usec = 1000 * msec;
> + int adj_sec = 0;
> + tv->tv_usec += adj_usec;
> + if (adj_usec >= 1000000) {
> + adj_sec = adj_usec / 1000000;
> + tv->tv_usec -= adj_sec * 1000000;
> + tv->tv_sec += adj_sec;
> + }
> + if (tv->tv_usec >= 1000000){
> + tv->tv_usec -= 1000000;
> + tv->tv_sec++;
> + }
> }
Is it still safe to use int if you're targeting a 32 bit platform?
--
Sitsofe | http://sucs.org/~sits/
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
