Xen devel - Jun 2007 - [Q] about clock_gettime accuracy on i386(&x86_64).

Hi,

I am now compareing clock_gettime() accuracy 
between Native Linux and Dom0.(i386 and x86_64)

Dom0 accuracy is microsecond.
(nanosecond 3-digit is not changed keep some value)
But Linux accuracy is nanosecond.
(nanosecond 3-digit is changed each clock_gettime call)

Is there any plan to fix this?
Or would you tell me the reason?

Thanks
Atsushi SAKAI




_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi,

I solve previous question, but other problem occured.
My question is CLOCK_MONOTONIC is available on PV-guest/x86.

CONFIG_GENERIC_TIME is off for current PV-kernel(i386).
In this case, directly calls do_gettimeofday.
This causes the problem for Xen.
do_gettimeofday is not MONOTONIC.
this causes clock_gettime(CLOCK_MONOTONIC,&tp) is not monotomic.

Is there any plan to solve this problem?

I attached the test program for 2CPU case (like Core2Duo).


Anyway, the reason for tick changes is that
getnstimeofday()  changes between Native and PV-guest.

In PV kernel, just calls do_gettimeofday.(it causes tickness microsecond.)
But In Native Linux Kernel, CONFIG_GENERIC_TIME is on.
The function(getnstimeofday()) has an accuracy of nanoseconds at most.

Thanks
Atsushi SAKAI


Atsushi SAKAI <sakaia@jp.fujitsu.com> wrote:
> Hi,
> 
> I am now compareing clock_gettime() accuracy 
> between Native Linux and Dom0.(i386 and x86_64)
> 
> Dom0 accuracy is microsecond.
> (nanosecond 3-digit is not changed keep some value)
> But Linux accuracy is nanosecond.
> (nanosecond 3-digit is changed each clock_gettime call)
> 
> Is there any plan to fix this?
> Or would you tell me the reason?
> 
> Thanks
> Atsushi SAKAI
> 
> 
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 13/6/07 11:35, "Atsushi SAKAI" <sakaia@jp.fujitsu.com> wrote:
> Is there any plan to solve this problem?
gettimeofday() could track last returned timestamp and clamp further
timestamps to be no smaller. We probably do need this in our gettimeofday()
implementation.

 -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel