RE: [PATCH] cpufreq: intel_pstate: Implement passive mode with HWP enabled

["Doug Smythies" <dsmythies@telus.net>]

Hi Rafael,

Thank you for your reply.

On 2020.07.16 05:08 Rafael J. Wysocki wrote:
> On Wed, Jul 15, 2020 at 10:39 PM Doug Smythies <dsmythies@telus.net> wrote:
>> On 2020.07.14 11:16 Rafael J. Wysocki wrote:
>> >
>> > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>> ...
>> > Since the passive mode hasn't worked with HWP at all, and it is not going to
>> > the default for HWP systems anyway, I don't see any drawbacks related to making
>> > this change, so I would consider this as 5.9 material unless there are any
>> > serious objections.
>>
>> Good point.

Actually, for those users that default to passive mode upon boot,
this would mean they would find themselves using this.
Also, it isn't obvious, from the typical "what driver and what governor"
inquiry.

>> Some of the tests I do involve labour intensive post processing of data.
>> I want to automate some of that work, and it will take time.
>> We might be into the 5.9-rc series before I have detailed feedback.
>>
>> However, so far:
>>
>> Inverse impulse response test [1]:
>>
>> High level test, i5-9600K, HWP-passive (this patch), ondemand:
>> 3101 tests. 0 failures. (GOOD)
>>
>> From [1], re-stated:
>> > . High level: i5-9600K: 2453 tests, 60 failures, 2.45% fail rate. (HWP-active - powersave)
>> > . Verify acpi-cpufreq/ondemand works fine: i5-9600K: 8975 tests. 0 failures.
>>
>> My version of that cool Alexander named pipe test [2] serialized workflow:
>>
>> HWP-passive (this patch), performance: PASS.
>>
>> From [2], re-stated, and also re-tested.
>> HWP-disabled passive - performance: FAIL.
> 
> But I'm not quite sure how this is related to this patch?

It isn't. The point being that it is different.
But yes, that failure is because of our other discussion [3].

> 
> This test would still fail without the patch if the kernel was started
> with intel_pstate=passive in the kernel command line, wouldn't it.

Yes.

> 
> > Although, I believe the issue to be EPB management, [3].
> 
> Well, that's kind of unexpected.
> 
> If this really is the case, then it looks like the effect of the EPB
> setting on the processor is different depending on whether or not HWP
> is enabled.
> 
>> And yes, I did see the reply to [3] that came earlier,
>> And have now re-done the test, with the referenced patch added.
>> It still is FAIL. I reply to the [3] thread, eventually.
>>
>> [1] https://marc.info/?l=linux-pm&m=159354421400342&w=2
>> [2] https://marc.info/?l=linux-pm&m=159155067328641&w=2
>> [3] https://marc.info/?l=linux-pm&m=159438804230744&w=2
>>
>> Kernel:
>>
>> b08284a541ad (HEAD -> k58rc5-doug) cpufreq: intel_pstate: Avoid enabling HWP if EPP is not supported
>> 063fd7ccabfe cpufreq: intel_pstate: Implement passive mode with HWP enabled
>> 730ccf5054e9 cpufreq: intel_pstate: Allow raw energy performance preference value
>> bee36df01c68 cpufreq: intel_pstate: Allow enable/disable energy efficiency
>> 199629d8200e cpufreq: intel_pstate: Fix active mode setting from command line
>> 11ba468877bb (tag: v5.8-rc5, origin/master, origin/HEAD, master) Linux 5.8-rc5
>>
>> Rules for this work:
>>
>> . never use x86_energy_perf_policy.
>> . For HWP disabled: never change from active to passive or via versa, but rather do it via boot.
>> . after boot always check and reset the various power limit log bits that are set.
>> . never compile the kernel (well, until after any tests), which will set those bits again.
>> . never run prime95 high heat torture test, which will set those bits again.
>> . try to never do anything else that will set those bits again.
>>
>> To be clear, I do allow changing governors within the context of the above rules.
> 
> Thanks for the feedback!