Re: [PATCH 1/2] Add scaled time to taskstats based process accounting

[Michael Neuling <mikey@neuling.org>]

In message <46C805B0.1000300@linux.vnet.ibm.com> you wrote:
> Andrew Morton wrote:
> >>  
> >>  	unsigned int rt_priority;
> >> -	cputime_t utime, stime;
> >> +	cputime_t utime, stime, utimescaled, stimescaled;
> > 
> > Adding 8 or 16 bytes to the task_struct for all architectures for something
> > which only powerpc uses?
> > 
> > Is there any prospect that other CPUs can use this?
> > 
> 
> Hi, Andrew,
> 
> There is definitely the prospect for other architectures to use this
> feature
> 
> x86 provides the APERF and MPERF model specific registers.
> The ratio of APERF to MPERF gives the current scaled load on the
> system (acpi-cpufreq, get_measured_perf()) I have been looking at
> exploiting this functionality for x-series, but ran into a problem;
> as per the specification, APERF and MPERF are to be reset to 0
> upon reading them. As a result, I am still figuring out a good
> way to share the data amongst the ondemand governor and utimescaled
> statistics.
> 
> I think for now, we can
> 
> 1. Put utimescaled and stimescaled under an #ifdef for ARCH_POWERPC

... or even #ifdef TASKSTATS

> 2. Add utimescaled and stimescaled and add a big fat comment stating
>    that work for other architectures is on it's way.
> 
> In either case, I think the functionality is useful and can be
> exploited by other architectures. The powerpc port is complete and
> I think the implementation would provide a good reference for
> other implementations to follow.
> 
> -- 
> 	Warm Regards,
> 	Balbir Singh
> 	Linux Technology Center
> 	IBM, ISTL
>