Re: Regression in 4.8 - CPU speed set very low

From: Larry Finger
Date: Mon Sep 26 2016 - 20:48:40 EST

On 09/26/2016 07:21 PM, Rafael J. Wysocki wrote:
On Tue, Sep 27, 2016 at 1:53 AM, Larry Finger <Larry.Finger@xxxxxxxxxxxx> wrote:
On 09/26/2016 05:16 PM, Rafael J. Wysocki wrote:

On Tue, Sep 27, 2016 at 12:09 AM, Larry Finger

<Larry.Finger@xxxxxxxxxxxx> wrote:

Maybe it's better to try diagnose the problem instead of spending more
time on bisection.

In my original post, I asked for such help, but nothing until today. I had
no idea what to check, but now I have a better idea.

I'd like to know whether or not 4.7 was definitely good, though.

I never saw this problem with 4.7, but given the difficulty in triggering
the problem, my tests may not have been definitive.

If it is one of them, it may be a while before I dare call this one
In one respect, that is good as I will be traveling tomorrow and

What does "cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver" say?


You probably don't need to worry about all of the cpufreq changes in
4.8-rc, then. Only a few of them affect intel_pstate and I don't see
how any of them may lead to the observed symptoms.

First off, if you have a reproducer, please run it on 4.7 and see if
you can trigger the issue in there.

I'm running 4.8-rc7 at the moment hoping to trigger the problem and get the data requested by Srinivas. Once I get that, I will try 4.7 again.

Second, it would be good to have a look at the output from the
cpu_frequency and pstate_sample tracepoints around when the issue
triggers. The pstate_sample one would be more interesting.

But for both we need a reproducer anyway.

I do not have a reliable reproducer. The condition has always happened when running a high-compute job such as a 'make -j8' on the kernel, or building the RPM for openSUSE's implementation of VirtualBox. The latter is what I'm using for most of my testing.

It also would be good to rule out the thermal throttling (as per the
Srinivas' comments).

For now, please tell me what's in