Forgot your password?
typodupeerror
Intel Ubuntu Linux

Kernel Bug Means Linux Power Usage Remains High 179

Posted by timothy
from the wrong-kind-of-power-user dept.
An anonymous reader writes "The significant Linux kernel power regression reported back in April, which ended up being attributed to PCI-E Active State Power Management, is still not resolved even as Ubuntu 11.10 and Fedora 16 approach. Until Linux is able to handle ASPM in a manner more like Windows or the device drivers explicitly set the ASPM flag, users of many modern laptops need to use the "pcie_aspm=force" option to regain much of their battery life. At least a power bug affecting newer Intel hardware with the "energy performance bias" feature has been fixed. There's more information in this LaunchPad bug report and in the latest power consumption testing."
This discussion has been archived. No new comments can be posted.

Kernel Bug Means Linux Power Usage Remains High

Comments Filter:
  • by Aighearach (97333)

    I wonder how much wasted power a beowulf cluster of those would have?

  • by jaymzter (452402) on Friday October 07, 2011 @06:57PM (#37645114) Homepage

    Seems like its been so long since relevant technical submissions were made to /. I remember being able to learn so many interesting OS tricks from poster's comments to articles and hearing about new software.

    Now it's mostly just crap about who pissed on who's patents...

    • by Aighearach (97333)

      Did you think the Grand Taco left for no reason?

    • Re:Finally! (Score:5, Insightful)

      by cadeon (977561) on Friday October 07, 2011 @07:27PM (#37645234)

      Unfortunately, that's what the majority of the news is these days.

      Years ago, a kernel regression that didn't result in a lockup or massive data corruption would have been borderline slow-news-day material. Today, software quality as a whole has increased, and there's not as much of that (or as many groundbreaking new features) going on. There's still some interesting stuff going on in the mobile world, but PCs and Servers have largely been figured out for the time being. At least compared to what it was a while back.

      As much as I'd like to jump on this "Blame slashdot, slashdot sucks now" bandwagon, they're just reporting what's happening, IMHO.

      And if they aren't reporting what you think is newsworthy, blame yourself for not submitting 'real' stories and/or not drinking from the firehose.

    • Re:Finally! (Score:4, Insightful)

      by djlowe (41723) * on Friday October 07, 2011 @07:40PM (#37645282)

      Now it's mostly just crap about who pissed on who's patents

      Well, that's marginally better than the copyright wars that reigned here not so long ago... or the global warming debate... or... what was before that? I forget.

      I imagine, however, that those generated more revenue. Patent battles among corporations are pretty much a battle among giants, and most of us here are just nerdly peons, fairly removed from such. They're gonna do whatever they want, work it out in the end, and the rest of us will get shat upon, one way or another.

      From here in the "cheap seats"? Shit is shit, regardless of who is dumping it on you, or so it seems to me.

      Cynically,

      dj

      • by ultranova (717540)

        Well, that's marginally better than the copyright wars that reigned here not so long ago... or the global warming debate... or... what was before that? I forget.

        Religion, specifically Intelligent Design. Which typically degenerated into a poo-flinging contest even faster than those two since neither side knew neither theology nor science (or at least the people who made the most noise didn't).

        My theory is that these topics get posted cyclically to keep people from getting bored. It's not unlike crop rotati

      • by gardyloo (512791)

        ... or... what was before that? I forget.
         

        Ponies!!!11111!1 and hot grits.

  • ... that a good unexplained fire and a stabbing wouldn't fix.

  • by kvvbassboy (2010962) on Friday October 07, 2011 @06:59PM (#37645124)

    It's a problem with the BIOS manufacturers, and the BIOS incorrectly reporting its ASPM capability. When an OEM installs Windows on a laptop, it can correctly tune these settings. But for a fresh install of Linux that YOU performed, a database of every motherboard + BIOS combination needs to be maintained in the open to set the force PCIE ASPM flag. If set wrongly, when the BIOS doesn't support it, it could lead to locking which is far more serious.

    There are other solutions to effectively manage power in Linux, like Jupiter [jupiterapplet.org].

    For more (and better) information, see the following links: About the Kernel 3.0 "Power Regression" Myth [fewt.com] and PCIe, power management, and problematic BIOSes [lwn.net]

    • If the BIOS reports ASPM it'll be used. If it doesn't, then Linux can't assume that it works.

      • Yes, but apparently a lot of the time the BIOS doesn't report ASPM even if it does support. You are right in that Linux can't assume anything, but OEMs are aware of it.
      • Re: (Score:3, Insightful)

        by hedwards (940851)

        And this sort of thing really ought to be used to slap MS upside the head for behaving irresponsibly. Years back when ACPI was first coming out and a significant number of motherboard models were shipped with a broken DSDT that would only function with Windows. The company creating the firmware didn't care and MS had the money to work around the problem leaving Windows the only platform that would work correctly.

        MS could have solved the problem by refusing to implement work arounds, but opted to go out of

        • Re: (Score:2, Funny)

          And this sort of thing really ought to be used to slap MS upside the head for behaving irresponsibly.

          Wow, I knew someone would find a way to blame Microsoft for this... but so early in the thread - well done! Hey, there's a little girl missing in the Los Angeles area - could you find a way to blame Microsoft for her disappearance as well?

          It's like the "Six Degrees of Kevin Bacon" game, but for Linux fanboys!

          • Um.. The "Halloween Documents"?

            That was a series of leaked emails from 1999-ish where Microsoft had discussed that hardware was "too standard" so actually encouraged this as a way for OEMs that sell finished systems to look better... And to spike the budding Open Source as well.

            Intel happily chipped in because the pushed specs like USB where every device can be super cheap... And controlled by the CPU... all those $39 printers, winmodems, GMA900, etc all sucked up CPU so Intel could sell more... And all tho

          • by hedwards (940851) on Friday October 07, 2011 @10:47PM (#37645854)

            It's pretty well established that the ACPI implementation that MS was using with Win XP was non-standard. The one that folks had access to wouldn't compile the DSDTs that were coming with a lot of the computers because they were buggy and non-standard. MS had the advantage of controlling the only validation program that mattered and could hard code into their OS the bits necessary to work with the most common bugs.

            Unfortunately for Linux, *BSD and everybody else, those coders didn't have access to that information and had to go to a huge amount of work to rewrite the DSDT and load that so that it would work as the standard specify.

            Also, nice ad hominem you've got there, I am not a Linux fanboy,

            • by Lehk228 (705449)
              seems to me that such features could be set up using a function probing tool that tested the function of each command and logged it so they system could know what does work, what does not work, and what crashes the system, submit results back to t he repository and linux will have a more complete and accurate database than windows has (include information on lower value probabilistic errors rather than just does/does not work, warn the user when yea this model is supposed to work but there is a 1/2000 chan
    • by NoSig (1919688)
      How about on first boot on a new motherboard, Linux tries to set PCIE on, then runs tests that are going to result in the locking you mentioned if PCIE isn't actually available. Is that possible?
      • Re: (Score:1, Flamebait)

        by billcopc (196330)

        Just because it's possible doesn't mean the zealots actually care about even trying. It's so much cooler to play the blame game, rather than focus on what's actually important: making things work!

        The net result is that someone else has to implement the hackish-yet-perfectly-acceptable fix. Kernel devs could tackle it, but they won't, so someone else will. That someone else is often Redhat or Ubuntu, which means the fixes don't travel back upstream.

        Even though it's the BIOS makers' fault, as an end user,

        • Re: (Score:3, Interesting)

          by Microlith (54737)

          Just because it's possible doesn't mean the zealots actually care about even trying. It's so much cooler to play the blame game, rather than focus on what's actually important: making things work!

          I like how you come out of the gate with an insult, and expect people to fix things for you.

          PROTIP: Kernel devs would rather play it safe than risk causing data loss.

          That someone else is often Redhat or Ubuntu, which means the fixes don't travel back upstream.

          No, it's unlikely they will either. I doubt they want to

          • That would be like if the company that made your engine created a bug where piston 3 would miss-fire ever 8th stroke, and then YOU demanding that the ECU programmers test for the bug when you start the car and not fire the 3rd piston each 8th revolution.
      • Sure... Then Linux takes 20 minutes to boot.

        The problem is that things like the ACPI are by chipset, and manufacturer, model and firmware rev. PCI vendor and ID strings aren't enough to just make a table to check at start... Let alone all the hardware interactions in an end user machine.

        Even trying to look up the settings on websites that track this stuff for some random PC you didn't build yourself is an hours long daunting task... And Big Box stuff often never makes those lists.

    • by mcgrew (92797) *

      Thank you for that, it'll be useful when I get Linux on this driveless subnotebook. Power isn't a problem with the box that runs on AC.

  • Why use Intel icon when it's a Linux issue?
    • Re:Intel? (Score:5, Funny)

      by Anonymous Coward on Friday October 07, 2011 @07:57PM (#37645344)

      It's a motherboard issue. BIOS supports ASPM but doesn't advertise it => Linux doesn't try using hardware that isn't there => power usage goes up => Linux gets blamed. Fucking Microsoft.

    • ASPM is implemented incorrectly in many motherboards, either fails to advertise it supports it when it does, or advertises it supports it when it doesn't

      MS have worked around the issue rather than getting the manufacturers to fix it .. so now everyone else has to follow suit ...

      I suspect MS do it wrong on a few systems, but no-one notices because everyone else does it wrong as well

      Blame the MB, Bios manufacturers for not caring about quality, and blame MS for not using the clout they have to force people t

  • - replace BIOS (perhaps demand that the manufacturer update it)
    - create a large (online) database of MBs (best identified by a list of hardware) that support ASPM and check it upon installation

    • One of the HUGE advantages of linux (and other *nix derivatives) is the ability to seamlessly move the OS from one machine to another without having to make ANY changes. The only exception is moving from nvidia to ati video cards (or vise-versa), but the OS will still boot. These "check at install time" fixes are very dangerous to that huge feature. You guys are actually demanding a software fix to a firmware bug. Absolutely unbelievable...
      • by Urkki (668283)

        I think affected users are demanding a fix for a bug. They won't much care what the fix is or where the bug is, they just want their po... email on the go.

        Solution to your "move from one machine to another" thing would be to tie certain settings to certain hardware signature. If signature changes, revert settings to safe values at boot. It would be a very useful thing to have overall, until the happy day comes when there are no hardware or firmware or software bugs in the world.

        Also, apparently there is a s

      • One of the HUGE advantages of linux (and other *nix derivatives) is the ability to seamlessly move the OS from one machine to another without having to make ANY changes. The only exception is moving from nvidia to ati video cards (or vise-versa), but the OS will still boot. These "check at install time" fixes are very dangerous to that huge feature.

        You could check it at the beginning of every boot. If the setting was correct, continue. If it wasn't, automatically change it and reboot if necessary.

  • They have been playing the blame game. Seriously though, I don't care whether it's an upstream issue or a downstream issue, 30% increase in power consumption is pretty big issue no matter what the reason. That's the reason why I am still on Ubuntu 10.10 and haven't upgraded. If they don't fix this before the support window runs out for 10.10 I am switching to something else (maybe even Windows). I use laptop on my battery all the time (in the bathroom, at the library, coffee shop etc.). A 30% reduction in b
    • Try using Jupiter (http://www.jupiterapplet.org/). It's not there in the repos, but there is a ppa available for it: https://launchpad.net/~webupd8team/+archive/jupiter [launchpad.net] .
    • How long do you SPEND in the bathroom...?
  • I'm not so sure about this. I enabled the fix as proposed by Phoronix and saw a 15% battery life improvement; I'm now getting almost 5 hours, which is pretty good for this system.But 15% was not "significant" really. So "to regain much of their battery life" seems like an exaggeration in par with the alarmist tone of the articles in Phoronix. Sure, there's a problem, and I certainly appreciate Phoronix's efforts to pinpoint the cause and offer a workaround, but it's certainly not as bad as they've been maki

"How to make a million dollars: First, get a million dollars." -- Steve Martin

Working...