Skip to contentSkip to conversation

Package C states over Haswell platform

Powertop Alexandra Yates 2 participants 2 messages Oct 29, 2014

lore
Share & tools
lore

Alexandra Yates

author
#0 lore raw
Quoted 81 lines
Quoted 81 lines

Hi,

I'm working on DELL E7440 platform (Haswell based) with dual OS: Windows 8.1
and Ubuntu 14.04.1 (based on 3.13).
When I use Win 8.1, using BLA I get ~90% PC7 when idle. However, when I use
the Ubuntu, using powertop I get only as high as PC3.
Since it's the same platform, it seems to rule out a HW issue. Any idea what
can be the problem? Is Haswell fully supported by powertop?

          Package   |             Core    |            CPU 0       CPU 2
                    |                     | C0 active   0.2%        0.1%
                    |                     | POLL        0.0%    0.0 ms

0.0%
0.0 ms
| | C1E-HSW 0.0% 0.1 ms
0.0%
0.0 ms

C2 (pc2)    0.1%    |                     |
C3 (pc3)    3.7%    | C3 (cc3)    0.0%    | C3-HSW      0.0%    0.4 ms

0.0%
0.4 ms
C6 (pc6) 0.0% | C6 (cc6) 0.0% | C6-HSW 0.0% 0.0 ms
0.0%
0.2 ms
C7 (pc7) 0.0% | C7 (cc7) 99.0% | C7s-HSW 0.1% 0.9 ms
0.1%
1.8 ms
C8 (pc8) 0.0% | | C8-HSW 0.1% 1.4 ms
0.2%
2.5 ms
C9 (pc9) 0.0% | | C9-HSW 0.4% 4.0 ms
2.3%
8.0 ms
C10 (pc10) 0.0% | | C10-HSW 98.9% 44.5 ms
97.0%
36.5 ms

                    |             Core    |            CPU 1       CPU 3
                    |                     | C0 active   0.2%        0.1%
                    |                     | POLL        0.0%    0.0 ms

0.0%
0.0 ms
| | C1E-HSW 0.0% 0.0 ms
1.0%
50.1 ms

                    |                     |
                    | C3 (cc3)    0.0%    | C3-HSW      0.0%    0.7 ms
0.0%
0.6 ms
                    | C6 (cc6)    0.0%    | C6-HSW      0.0%    0.5 ms
0.0%
0.0 ms
                    | C7 (cc7)   98.0%    | C7s-HSW     0.1%    0.7 ms

0.4%
25.7 ms
| | C8-HSW 0.0% 4.9 ms
0.0%
1.5 ms
| | C9-HSW 0.6% 4.7 ms
0.7%
19.5 ms

                    |                     | C10-HSW    98.5%   93.5 ms

97.8%
78.0 ms

                    |             GPU     |
                    |                     |
                    | Powered On 12.1%    |
                    | RC6        87.9%    |
                    | RC6p        0.0%    |
                    | RC6pp       0.0%    |

Thanks

Quoted 7 lines
Quoted 7 lines

Rony

_______________________________________________
PowerTop mailing list
PowerTop(a)lists.01.org
https://lists.01.org/mailman/listinfo/powertop

Hi Rony,

Interesting you ask this since I presented at the Open source bridge conference a case study on how to optimize your computer in terms of power using PowerTOP, as an example I used a Haswell box with Ubuntu. I also presented on the Debian conference a similar talk using Debian distro. Both talks happened this year.

There are few things that may be happening.

1- Update to the latest kernel 3.17. From 3.13 to 3.17
there have been few bug fixes that help linux in terms of power
management.

2-At the boot command line of your kernel add the following settings:

       i915.enable_psr=1
       i915.enable_fbc=1
       pcie_aspm=force

The first command enables the graphics card use panel self refresh, the second enables frame buffer compression. The third command helps the network card enter D3 state.
http://wireless.kernel.org/en/users/Documentation/ASPM#Enabling_ASPM_with_setpci

3- Use PowerTOP/tunables to tune power management for all your devices.
If none of the settings disrupt the proper usability of the computer add then to a script file and run this each time you restart your computer. Alternatively you can use powertop --auto_tune

Run PowerTOP at this point you should see your system reaching a deep power state.

Thank you,
Alexandra.

#1 lore raw

Hi,

Alexandra and Daniel, thank you for your suggestions.

1. I forgot to mention that the powertop results that I posted were after performing --autotune
2. The platform's lid is off and I connect using SSH, so I don't think that the graphic card/frame buffer are to blame
3. I don't think that not entering D3 has any affect on entering PC7. The package should enter PC7 even when the device is not in D3
4. Currently I'm trying to avoid updating the kernel. However, if nothing else helps I might have to...

I can think of one thing that might prevent the package from entering high C states. It might be related to the LTR mechanism, which would prevent high PC states when devices report low LTR values (or if there is some other LTR related issue). Is there a way the read the LTR values that devices set during runtime?

BTW, reading LTR values could be an important feature for powertop.

Thanks,

Rony

Quoted 127 lines · Alexandra Yates
Quoted 81 lines
Quoted 81 lines

Hi,

I'm working on DELL E7440 platform (Haswell based) with dual OS: Windows 8.1
and Ubuntu 14.04.1 (based on 3.13).
When I use Win 8.1, using BLA I get ~90% PC7 when idle. However, when I use
the Ubuntu, using powertop I get only as high as PC3.
Since it's the same platform, it seems to rule out a HW issue. Any idea what
can be the problem? Is Haswell fully supported by powertop?

          Package   |             Core    |            CPU 0       CPU 2
                    |                     | C0 active   0.2%        0.1%
                    |                     | POLL        0.0%    0.0 ms

0.0%
0.0 ms
| | C1E-HSW 0.0% 0.1 ms
0.0%
0.0 ms

C2 (pc2)    0.1%    |                     |
C3 (pc3)    3.7%    | C3 (cc3)    0.0%    | C3-HSW      0.0%    0.4 ms

0.0%
0.4 ms
C6 (pc6) 0.0% | C6 (cc6) 0.0% | C6-HSW 0.0% 0.0 ms
0.0%
0.2 ms
C7 (pc7) 0.0% | C7 (cc7) 99.0% | C7s-HSW 0.1% 0.9 ms
0.1%
1.8 ms
C8 (pc8) 0.0% | | C8-HSW 0.1% 1.4 ms
0.2%
2.5 ms
C9 (pc9) 0.0% | | C9-HSW 0.4% 4.0 ms
2.3%
8.0 ms
C10 (pc10) 0.0% | | C10-HSW 98.9% 44.5 ms
97.0%
36.5 ms

                    |             Core    |            CPU 1       CPU 3
                    |                     | C0 active   0.2%        0.1%
                    |                     | POLL        0.0%    0.0 ms

0.0%
0.0 ms
| | C1E-HSW 0.0% 0.0 ms
1.0%
50.1 ms

                    |                     |
                    | C3 (cc3)    0.0%    | C3-HSW      0.0%    0.7 ms
0.0%
0.6 ms
                    | C6 (cc6)    0.0%    | C6-HSW      0.0%    0.5 ms
0.0%
0.0 ms
                    | C7 (cc7)   98.0%    | C7s-HSW     0.1%    0.7 ms

0.4%
25.7 ms
| | C8-HSW 0.0% 4.9 ms
0.0%
1.5 ms
| | C9-HSW 0.6% 4.7 ms
0.7%
19.5 ms

                    |                     | C10-HSW    98.5%   93.5 ms

97.8%
78.0 ms

                    |             GPU     |
                    |                     |
                    | Powered On 12.1%    |
                    | RC6        87.9%    |
                    | RC6p        0.0%    |
                    | RC6pp       0.0%    |

Thanks

Quoted 7 lines
Quoted 7 lines

Rony

_______________________________________________
PowerTop mailing list
PowerTop(a)lists.01.org
https://lists.01.org/mailman/listinfo/powertop

Hi Rony,

Interesting you ask this since I presented at the Open source bridge conference a case study on how to optimize your computer in terms of power using PowerTOP, as an example I used a Haswell box with Ubuntu. I also presented on the Debian conference a similar talk using Debian distro. Both talks happened this year.

There are few things that may be happening.

1- Update to the latest kernel 3.17. From 3.13 to 3.17
there have been few bug fixes that help linux in terms of power
management.

2-At the boot command line of your kernel add the following settings:

       i915.enable_psr=1
       i915.enable_fbc=1
       pcie_aspm=force

The first command enables the graphics card use panel self refresh, the second enables frame buffer compression. The third command helps the network card enter D3 state.

http://wireless.kernel.org/en/users/Documentation/ASPM#Enabling_ASPM_with_setpci

3- Use PowerTOP/tunables to tune power management for all your devices.
If none of the settings disrupt the proper usability of the computer add then to a script file and run this each time you restart your computer. Alternatively you can use powertop --auto_tune

Run PowerTOP at this point you should see your system reaching a deep power state.

Thank you,
Alexandra.