In continuation of Mobile Users Beware: Linux Has Major Power Regression
, what's causing this regression in the mainline Linux kernel between Linux 2.6.37 and 2.6.38 (and still outstanding in 2.6.39) is still a mystery. To some surprise, even when disabling power management on one of the ThinkPads, the problem is still present.
In the article published earlier, it was shown that this major performance regression that's affected multiple notebooks of several CPU generations and GPUs of different vendors isn't attributed to any back-light differences, EIST (Enhanced Intel SpeedStep Technology) was still functioning for dynamically changing the CPU clock/voltage states, and there are no other obvious differences.
The latest test to be carried out was to see what would happen when disabling "Intel SpeedStep" and "CPU Power Management" from the BIOS of the Lenovo ThinkPad T60 with the Core Duo T2400. When doing so and testing both the 2.6.37 and Linux 2.6.38 kernels, the problem still persisted.
This power issue is even to the extent that a Linux 2.6.37 kernel when EIST/CPU-PM (Power Management) is disabled from the BIOS that under that condition it even consumes less power than a Linux 2.6.38 kernel when the power management and SpeedStep Technology are enabled.
So disabling EIST / CPU power management from the BIOS of the ThinkPad seems to have no impact on this dramatic regression.
The only common denominator so far is that it's occurring on Intel hardware. However, that's all I've been able to test on as far as mobile devices are concerned. All of the netbooks / notebooks here are Intel powered and for desktops / workstations / servers it's also skewed towards Intel by a factor of 6~9 to 1 AMD. That's simply because AMD hasn't been sending out any processors or other hardware to Phoronix (the last time was Opterons a few generations ago), so the AMD hardware around is usually limited to what I end up purchasing. Intel hardware is much easier to come by. When buying any notebooks, they tend to be ThinkPads or Mac Book Pros, which also trend towards Intel. Unfortunately for the AMD desktops and servers, there isn't the power monitoring support as mentioned in the earlier article.
There's also now at least one Phoronix reader
who has also noticed issues with Linux 2.6.38+ and as a result is still using the Linux 2.6.37 kernel.