Re: [Patch 1/4] Delay accounting: Initialization

[Marcelo Tosatti <marcelo.tosatti@cyclades.com>]

On Tue, Nov 15, 2005 at 10:20:35AM -0200, Marcelo Tosatti wrote:
> On Tue, Nov 15, 2005 at 10:19:17AM -0500, Shailabh Nagar wrote:
> > Marcelo Tosatti wrote:
> > > On Mon, Nov 14, 2005 at 08:20:17PM -0800, Andrew Morton wrote:
> > > 
> > >>Shailabh Nagar <nagar@watson.ibm.com> wrote:
> > >>
> > >>>+	*ts = sched_clock();
> > >>
> > >>I'm not sure that it's kosher to use sched_clock() for fine-grained
> > >>timestamping like this.  Ingo had issues with it last time this happened?  
> 
> Maybe Ingo had some other issue other than !use_rtc ? Better check.
> 
> > > If the system boots with use_rtc == 0 you're going to get jiffies based
> > > resolution from sched_clock(). I have a 1GHz Pentium 3 around here which
> > > does that.
> > 
> > Good point, thanks. This reemphasizes the need for better normalization
> > at output time.
> 
> > > Maybe use do_gettimeofday() for such systems?
> > 
> > Perhaps getnstimeofday() so resolution isn't reduced to msec level unnecessarily.
Yep.
> > In these patches, userspace takes responsibility for handling wraparound so
> > delivering a reasonably high-resolution delay data from the kernel is preferable.
> > 
> > > 
> > > Would be nice to have a sort of per-arch overridable "gettime()" function?
> > > 
> > 
> > Provided as part of this patch ?
> 
> Yep, think so. 

Actually I dont think there is any kind of "get_timestamp()" style 
API now. Maybe it would be nice to create one? Its generic functionality 
after all.

blktrace is also using sched_clock(), too.


> My comment meant that its nice to hide away architecture 
> speficic code from generic code, so you don't have to add #ifdef's and 
> such.
> 
> Not sure about the nicer way to do that.
> 
> -
> 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/