Re: [PATCH 1/7] sched: Introduce scale-invariant load tracking

From: Dietmar Eggemann
Date: Wed Oct 08 2014 - 08:54:49 EST


Hi Yuyang,

On 08/10/14 01:50, Yuyang Du wrote:
> Hi Morten,
>
> Sorry for late jumping in.
>
> The problem seems to be self-evident. But for the implementation to be
> equally attractive it needs to account for every freq change for every task,
> or anything less than that makes it less attractive.
>
> But this should be very hard. Intel Architecture has limitation to capture all
> the freq changes in software and also the intel_pstate should have no
> notification.

We encountered this missing notification for current frequency with
Intel systems (e.g. i5-3320M) using the intel_pstate driver while
testing this patch-set. The arch_scale_set_curr_freq call in
__cpufreq_notify_transition [[PATCH 2/7] cpufreq: Architecture specific
callback for frequency changes] will not work on such a system.

In our internal testing, we placed arch_scale_set_curr_freq(cpu->cpu,
sample->freq) into intel_pstate_timer_func [intel_pstate.c] to get the
current frequency for a cpu.

The arch_scale_set_max_freq call in cpufreq_set_policy
[drivers/cpufreq/cpufreq.c] still works although the driver exposes the
max turbo pstate and not the max pstate. That's an additional problem
because we don't want to use turbo states for frequency scaling.

>
> For every task, this makes the updating of the entire queue in load tracking
> more needed, so once again, ping maintainers for the rewrite patches, :)
>
> Thanks,
> Yuyang
>

[...]

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