lkml.org 
[lkml]   [2004]   [Jul]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: gettimeofday nanoseconds patch (makes it possible for the posix-timer functions to return higher accuracy)
From
Date
On Wed, 2004-07-14 at 09:41, Christoph Lameter wrote:
> I am working on some timer issues for the IA64 in order to make the timers
> more efficient and return results with a higher accuracy.
>
> However, in various locations do_gettimeofday is used and then the
> resulting usecs are multiplied by 1000 to obtain nanoseconds. This is in
> particular problematic for the posix timers and especially clock_gettime.
>
> The following patch introduces a new gettimeofday function using
> struct timespec instead of struct timeval. If a platforms supports time
> interpolation then the new gettimeofday will use that to provide a
> gettimeofday function with higher accuracy and then also clock_gettime
> will return with nanosecond accuracy.

Honestly, I'm not a fan of the patch. It realistically only helps ia64
and and adds more confusing code to the generic time code. If there
isn't an real/immediate need for this, I'd wait to 2.7 for a better
cleanup.

None the less, I do understand the desire for the change (and am working
to address it in 2.7), so could you at least use a better name then
gettimeofday()? Maybe get_ns_time() or something? Its just too similar
to do_gettimeofday and the syscall gettimeofday().

Really, I feel the cleaner method is to fix do_gettimeofday() so it
returns a timespec and then convert it to a timeval in
sys_gettimeofday(). However this would add overhead to the syscall, so I
doubt folks would go for it.

> I would be interested in feedback on this approach. In this context
> the time interpolator patches that are being discussed on linux-ia64
> would also be of interest. Those provide a generic way to utilize
> any memory mapped or CPU counter to do the time interpolations for any
> platforms and would allow an easy way to realize a nanosecond resolution
> for all platforms.

I think the ia64 time interpolation code is a step in the right
direction (def better then the i386 bits), but it still isn't the
cleanest and clearest way. My plan is to select a reliable timesource
for the system, then use a periodic interrupt to accumulate time from
the timesource (in order to avoid overflows). This avoids lost tick
issues and cleanly separates the timer subsystem from the time of day
subsystem.

> The patch is against 2.6.8-rc1
>
> Index: linux-2.6.7/kernel/time.c
> ===================================================================
> --- linux-2.6.7.orig/kernel/time.c
> +++ linux-2.6.7/kernel/time.c
> @@ -421,6 +421,40 @@
>
> EXPORT_SYMBOL(current_kernel_time);
>
> +#ifdef TIME_INTERPOLATION
> +void gettimeofday (struct timespec *tv)
> +{
> + unsigned long seq;
> +
> + do {
> + seq = read_seqbegin(&xtime_lock);
> + tv->tv_sec = xtime.tv_sec;
> + tv->tv_nsec = xtime.tv_nsec+time_interpolator_get_offset();
> + } while (unlikely(read_seqretry(&xtime_lock, seq)));
> +
> + while (unlikely(tv->tv_nsec >= NSEC_PER_SEC)) {
> + tv->tv_nsec -= NSEC_PER_SEC;
> + ++tv->tv_sec;
> + }
> +}

You'll need to cap time_interpolator_get_offset() at the maximum NTP
tick size, or else you may have time go backwards right after a timer
interrupt.

thanks
-john




-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 14:04    [W:0.104 / U:0.076 seconds]
©2003-2014 Jasper Spaans. Advertise on this site