CPU frequency management fails to re-engage on resume from suspend

[Adam Colligan]

Behavior:

CPU frequency management fails to re-engage on resume from suspend. On resume, the CPUs are stuck around a single frequency (usually a low idle frequency).

I use indicator-cpufreq for manual governor control. On resume, indicator-cpufreq indicates the low frequency state in its live cpu performance display in the icon. However, in its governor menu, it still shows as in use whichever governor was selected before suspend, even though that governor is not actually or properly controlling the CPU state.

Upon manually switching to the unused governor in indicator-cpufreq, this behavior resolves, and the newly-chosen governor begins performing its duties. I can then immediately switch back to the desired (pre-suspend) governor with no issues.

PM logs appear to show a successful resume of cpu frequency management, even though the modules did not actually gain effective control of the processor frequencies.

Expected behavior:

On resume, the CPU frequencies should continue being managed as they were before suspend without the need for manual user input. Also, in the event that they do not, an error/warning should appear -- if not to the user, then in the pm suspend/resume logs.

Other information that may or may not be relevant:

On this system, the "powersave" governor is always automatically chosen at startup, although not until a minute or so after session login. The choice of an alternate governor "performance" from indicator-cpufreq is not persistent across reboots. Further, if I manually toggle the governors at the beginning of my first session after boot, and I settle on "performance", some automated script will force it back to "powersave" at the usual time about a minute after login. To use the "performance" governor, I must select it after waiting an interval following login, and I must always switch to powersave and back to performance after each resume from suspend.

ProblemType: Bug
DistroRelease: Ubuntu 15.04
Package: linux-image-3.19.0-25-generic 3.19.0-25.26
ProcVersionSignature: Ubuntu 3.19.0-25.26-generic 3.19.8-ckt2
Uname: Linux 3.19.0-25-generic x86_64
ApportVersion: 2.17.2-0ubuntu1.2
Architecture: amd64
AudioDevicesInUse:
 USER PID ACCESS COMMAND
 /dev/snd/controlC0: adam 3887 F.... pulseaudio
 /dev/snd/controlC1: adam 3887 F.... pulseaudio
CurrentDesktop: Unity
Date: Thu Aug 13 11:41:27 2015
InstallationDate: Installed on 2014-08-10 (368 days ago)
InstallationMedia: Ubuntu 14.04.1 LTS "Trusty Tahr" - Release amd64 (20140722.2)
MachineType: Hewlett-Packard HP ENVY 15 x360 PC
ProcFB: 0 inteldrmfb
ProcKernelCmdLine: BOOT_IMAGE=/vmlinuz-3.19.0-25-generic.efi.signed root=UUID=5ad8035a-f2a4-4334-84d8-e2a8a7396bf8 ro quiet splash vt.handoff=7
RelatedPackageVersions:
 linux-restricted-modules-3.19.0-25-generic N/A
 linux-backports-modules-3.19.0-25-generic N/A
 linux-firmware 1.143.3
SourcePackage: linux
UpgradeStatus: No upgrade log present (probably fresh install)
dmi.bios.date: 01/19/2015
dmi.bios.vendor: Insyde
dmi.bios.version: F.26
dmi.board.asset.tag: Type2 - Board Asset Tag
dmi.board.name: 22D6
dmi.board.vendor: Hewlett-Packard
dmi.board.version: 89.23
dmi.chassis.type: 10
dmi.chassis.vendor: Hewlett-Packard
dmi.chassis.version: Chassis Version
dmi.modalias: dmi:bvnInsyde:bvrF.26:bd01/19/2015:svnHewlett-Packard:pnHPENVY15x360PC:pvr0974100022405F00000420180:rvnHewlett-Packard:rn22D6:rvr89.23:cvnHewlett-Packard:ct10:cvrChassisVersion:
dmi.product.name: HP ENVY 15 x360 PC
dmi.product.version: 0974100022405F00000420180
dmi.sys.vendor: Hewlett-Packard

[Brad Figg (brad-figg)]

This change was made by a bot.

[Adam Colligan (adam-p)]

I have tried to recreate this with the following two mainline kernels:

v4.2-rc6-unstable
v4.2.0-999 (201508130157)

However, it is somewhat difficult to compare the behavior apples-to-apples the way I normally do, as cpupower is not available for those kernels (no linux-tools packages installed for them). I would appreciate any advice on the best way to show the problem's presence/absence in an objective or definitive way across kernels.

[Doug Smythies (dsmythies)]

When a user is using the intel_pstate driver in performance mode, then yes, prior to Kernel 4.2RC1 the CPU frequency will be typically be stuck after resume from suspend.

This is not a pm-ultils issue (which is how I stumbled across this bug report).

To test that the issue is fixed on kernel 4.2, just use primitive commands and not these higher level utilities (that I neither use or know anything about). For example use:

grep MHz /proc/cpuinfo

to observe CPU frequencies.

The commit that, I think, fixes this issue is:

commit 6c1e45917dec5e7c99ba8125fd8cc50f6e482a21
Author: Doug Smythies <email address hidden>
Date: Mon Jun 1 21:12:34 2015 -0700

    intel_pstate: Force setting target pstate when required

[Adam Colligan (adam-p)]

I have not noticed this problem on later kernels; I'm going to mark it as "fix released" for Linux and "invalid" for pm-utils. I hope that's the right course of action.