On 12/14/23 08:36, Vincent Guittot wrote:

On Thu, 14 Dec 2023 at 09:30, Lukasz Luba [off-list ref] wrote:

quoted

On 12/12/23 14:27, Vincent Guittot wrote:

[snip]

quoted

quoted

       update_rq_clock(rq);
-     thermal_pressure = arch_scale_thermal_pressure(cpu_of(rq));
-     update_thermal_load_avg(rq_clock_thermal(rq), rq, thermal_pressure);
+     hw_pressure = arch_scale_hw_pressure(cpu_of(rq));
+     update_hw_load_avg(rq_clock_task(rq), rq, hw_pressure);

We switch to task clock here, could you tell me why?
Don't we have to maintain the boot command parameter for the shift?

This should have been part of the patch5 that I finally removed. IMO,
the additional time shift with rq_clock_thermal is no more needed now
that we have 2 separates signals

I didn't like the left-shift which causes the signal to converge slowly.
I rather wanted right-shift to converge (react faster), so you have my
vote for this change. Also, I agree that with the two-signal approach
this shift trick can go away now. I just worry about the dropped boot
parameter.

So, are going to send that patach5 which removes the
'sched_thermal_decay_shift' and documentation bit?