Announcement

Collapse
No announcement yet.

There's Another Linux Kernel Power Problem

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • There's Another Linux Kernel Power Problem

    Phoronix: There's Another Linux Kernel Power Problem

    After last year discovering a major Linux kernel power regression that was widely debated until the Phoronix test automation software bisected the problem to get to the bottom of the situation, there's more active power regressions today on the Linux desktop. As I've mentioned on Twitter and in other articles in weeks prior there's a few regressions, but one of them for at least some notebooks is causing a very significant increase in power consumption. This situation that remains unresolved as of the Linux 3.7 kernel can cause the system to be going through about 20% more power.

    http://www.phoronix.com/vr.php?view=18242

  • #2
    ...due to single-handedly writing hundreds of articles per month...
    Perhaps you should try writing with both hands. I hear typing is much faster that way...

    Comment


    • #3
      see
      https://bbs.archlinux.org/viewtopic.php?id=149901
      https://bbs.archlinux.de/viewtopic.php?id=22089 (German)

      Comment


      • #4
        Originally posted by jwilliams View Post
        Perhaps you should try writing with both hands. I hear typing is much faster that way...
        good one

        Comment


        • #5
          Michael +100 karma points.

          I think its indeed may have sense to outsource some of the editorial work and concentrate on articles about own projects, like this one.
          Thanks, Michael!

          Comment


          • #6
            Originally posted by crazycheese View Post
            Michael +100 karma points.

            I think its indeed may have sense to outsource some of the editorial work and concentrate on articles about own projects, like this one.
            Thanks, Michael!
            Good idea, maybe get help from the knowledgeable posters here to help with bisecting or running further tests?

            Comment


            • #7
              Given that 3.7 lacks drops that the others have, maybe it's a cpuidle regression?

              Or maybe the scheduler is migrating cores too much, not keeping all the cores at a higher frequency than they would normally be? Can easily check that with perf. I did see http://klaig.blogspot.com/2012/12/fo...-can-make.html today, but the versions listed aren't consistent with this, though perhaps there was a minor regression in 3.6 that was amplified by some change in 3.7.

              Comment


              • #8
                Not sure if this is the same thing but,

                I have been testing 3.7 since the rc1, and later in the development I started noticing power issues on my Asus X53E with the Sandy Bridge graphics. I finally noticed that powertop 2.1 would report the GPU running 100 percent of the time, and the exhaust was running hot also. I checked the DMESG, and it did show that the i915 kernel module was enabling the sleep states, but obviously it wasn't working..

                normally a reboot of the laptop would temporarily fix the issue, least until I tried booting the unit cold again..

                only other issue is that 3.6 and 3.7 don't seem to work right with reboots, and power off, I have to manually turn it off with the switch. 3.5 doesn't have any of these issues.

                when running 3.7 without the gpu bugging, I am getting much better power usage, normally idles under 8 watts, where normally I get 8.2 watts according to powertop or the gnome power monitor.

                Comment


                • #9
                  Originally posted by Lederhosen View Post
                  I have been testing 3.7 since the rc1, and later in the development I started noticing power issues on my Asus X53E with the Sandy Bridge graphics. I finally noticed that powertop 2.1 would report the GPU running 100 percent of the time, and the exhaust was running hot also. I checked the DMESG, and it did show that the i915 kernel module was enabling the sleep states, but obviously it wasn't working..

                  normally a reboot of the laptop would temporarily fix the issue, least until I tried booting the unit cold again..

                  only other issue is that 3.6 and 3.7 don't seem to work right with reboots, and power off, I have to manually turn it off with the switch. 3.5 doesn't have any of these issues.

                  when running 3.7 without the gpu bugging, I am getting much better power usage, normally idles under 8 watts, where normally I get 8.2 watts according to powertop or the gnome power monitor.
                  Has this been reported to the kernel developers?

                  ...
                  I guess this doesn't affect me, as I have 1 AMD Neo-based laptop & 1 Atom N270.

                  Comment


                  • #10
                    Originally posted by Ibidem View Post
                    Has this been reported to the kernel developers?

                    ...
                    I guess this doesn't affect me, as I have 1 AMD Neo-based laptop & 1 Atom N270.

                    I believe it's related to this one: oddly enough, no solution yet.

                    But right yesterday I found another issue in the 3.6.x kernels which prevents PCIE ASPM to be enabled when on battery power... at least on my laptop.

                    Comment


                    • #11
                      I too have noticed significantly less battery life on my ArchLinux Intel i5 (Sandy Bridge) based laptop since moving to Kernel 3.6. For a long while, I thought perhaps my battery was getting old; however, moving back to an older Kernel seems to have resolved the issues. Here's hoping that this really bad regression gets fixed~!

                      Comment


                      • #12
                        Originally posted by Ibidem View Post
                        Has this been reported to the kernel developers?
                        "I have also alerted other select parties and stakeholders to the outstanding issues, but once again, there is little external activity until the problem is tossed into the public spotlight."

                        I hope those "select parties" are kernel devs.

                        Comment


                        • #13
                          Originally posted by Ibidem View Post
                          Has this been reported to the kernel developers?
                          ...
                          Yup: https://bugzilla.kernel.org/show_bug.cgi?id=48721

                          Comment


                          • #14
                            "I have also alerted other select parties and stakeholders to the outstanding issues, but once again, there is little external activity until the problem is tossed into the public spotlight."

                            I find it incredible that even someone like Michael is experiencing this very same thing! The developers really should be more thankful for all the free testing they get from both volunteers and the solutions volunteers provide. Phoronix test suite is a real asset Michael is providing.

                            Comment


                            • #15
                              Open source development often happens in the direction that sparks the interest of some developer. Usually developers like doing new features for themselves rather than hunting for some obscure bug that is really hard to trigger.

                              When money is involved even really skilled people will do what ever is asked from them, even useless shit like porting standards compliant web pages to support IE 5-8
                              Last edited by varikonniemi; 12-10-2012, 05:37 AM.

                              Comment

                              Working...
                              X