Phoronix: ASPM Kernel Power Fix Won't Land Until Linux 3.3
For those still wondering about the patches that appeared last month to mimic the power management behavior of Windows within the Linux kernel as a proper fix to the well known ASPM Linux kernel power regression, here's an update on the matter...
Well, I am rather happy here with my system and I use pcie_aspm=force on all machines w. PCIE without hassles. Still it would be nice to have it fixed in Kernel but maybe devs want to check out faulty hardware first to blacklist it. Otherwise people might end up with strange results with the next official kernel.
It's always astonishing to me how much hardware bugs/flaws become visible and public during FOSS driver development.
You're making a mistake of even trying to discuss with his... assessments. The bug has been confirmed on launchpad.net by ubuntu devs plus there's a dedicated ubu wiki page for this particular issue and there some sorry #%@ specialist pop up like the issue itself wasn't PITA enough. It's not worth discussing here.
So, I'm the guy that wrote the article and I have more experience with the issue than you can ever pretend to comprehend. You can discount it all you want, but I have proven the regression false on multiple occasions.
The bug has been confirmed by people that don't comprehend the basics of configuration and testing, it is not a kernel problem it is a configuration problem. Is there room for power optimization in the kernel? Sure. Is there a regression? No.
If you look at the following data set with the claimed fix - you will find that the "fix" really didn't amount to much except in the case of a small number of machines - as I implied.
Check out the T420 thinkpads with the "fix", 14.89W - 9.14W - 19.54W, see the pattern? They are all using too much power because they are all configured wrong. If I can make them use 7 watts with a simple configuration change. Hmm..
The test procedure isn't valid either, they should reboot the computer before testing for the baseline, and test with no apps running. Then do the same with the new kernel. They don't do that though, so there is no telling what was running before and what is running after.
There is also no collection of data used to gather the metrics so it can't be analyzed, meaning unfortunately that it is all invalid. Amateur test cases at best.
My ASUS (linked above) with the regression is currently sitting at 38 degrees Celsius, with Chromium, Thunderbird, Pidgin and a bunch of other apps running while connected to AC power.
I can see now how it is a kernel problem though, thanks for pointing it out, genius.