Page: 1, 2  Next

ASPM fix for PCIe power regression
DeepDayze
Status: Contributor
Joined: 21 May 2009
Posts: 128
Reply Quote
An interesting article on Phoronix regarding a patch to fix the ASPM regression issues on PCIe systems:

www.phoronix.com/scan.php?page=article&item=linux_aspm_solution&num=1

Will it be possible to integrate this patch into your kernels, Damentz?

Patch is here (for 3.2) but could possibly work with the other kernel versions.
Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1117
Reply Quote
This is pretty interesting. I read this news on phoronix this morning only 15 minutes ago. I'll look to see if there's been any more progress, discussion, or newer patches and put those into the 3.1 kernel I'm working on now.
Back to top
DeepDayze
Status: Contributor
Joined: 21 May 2009
Posts: 128
Reply Quote
As noted there the patch works well and the power issues have been resolved once the patch is applied. Not sure about the power issues on non-PCIe systems but there are other power management bugs still being worked on that might be applicable to such systems
Back to top
ej64
Status: Interested
Joined: 13 Nov 2011
Posts: 18
Location: Germany
Reply Quote
:: damentz wrote ::
... and put those into the 3.1 kernel I'm working on now.

Are they in your latest linux-image-3.1.0-1.dmz.2 of 13-Nov-2011?

I'ld really appreciate this ... using the same Thinkpad model as the patch developer.
Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1117
Reply Quote
:: ej64 wrote ::
:: damentz wrote ::
... and put those into the 3.1 kernel I'm working on now.

Are they in your latest linux-image-3.1.0-1.dmz.2 of 13-Nov-2011?

I'ld really appreciate this ... using the same Thinkpad model as the patch developer.


Yes, check it out and see if it helps.
Back to top
ej64
Status: Interested
Joined: 13 Nov 2011
Posts: 18
Location: Germany
Reply Quote
:: Quote ::
Yes, check it out and see if it helps.

Negative ... I'm slightly confused

It's a Lenovo X220, i5-2520, 4 GB RAM, 128 GB SSD (Samsung 1. Gen), 6-cell 63 Wh battery, with Crunchbang Linux XFCE on Debian testing.
Anything wireless disabled, lowest brightness, no load, I get these power readings from tp_smapi (/sys/devices/platform/smapi/BAT0/power_now) and xfce-power-manager 1.0.10 (unreliable?):
:: Code ::
3.1.0-1.dmz.2-liquorix-amd64                         ~ 12.4 W, ~ 6h 10m remaining
3.1.0-1.dmz.2-liquorix-amd64 + pcie_aspm=force       ~ 11.7 W, ~ 5h 20m (??) remaining
2.6.39-bpo.2-amd64           + pcie_aspm=force       ~  6.6 W, ~ 9h remaining

It seems the patch kind of works, but it's still worse than with pcie_aspm=force. However there's still a major power regression compared to the Debian stock 2.6.39 kernel (from packages.debian.org/squeeze-backports/).
The last config is roughly in line with win7 power readings (X220 at notebookcheck), my one is getting down to ~ 6 W with mics and sound disabled. So, at the moment I'm settling with the 2.6.39 kernel.

Damentz, do you prohibit CPU C4 state? Any ideas?
Back to top
GoinEasy9
Status: Contributor
Joined: 28 Jun 2008
Posts: 84
Location: Manorville, New York
Reply Quote
Check out dmesg and see if ASPM is mentioned. There is a similar thread on the siduction.org forum where they are trying the patch with towo's kernel, also showing that the patch isn't helping. This is the line from dmesg that they are getting:

ACPI _OSC control for PCIe not granted, disabling ASPM

It may be that the patch isn't all it's made out to be.
Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1117
Reply Quote
Read the new benchmarks Phoronix makes, documenting the new and old power issues we're getting in every new kernel release. There's probably something else broken between 2.6.39 and 3.1 that's giving you power consumption problems.

Your best course of action is to identify what subsystem is causing it. It could be anything from the CPU, GPU, motherboard, sound chip, network card, etc.
Back to top
ej64
Status: Interested
Joined: 13 Nov 2011
Posts: 18
Location: Germany
Reply Quote
Thanks everybody for your input -- the topic is solved to me! :)

After GoinEasy9's post I digged deeper ...
1. pcie_aspm=force seems to work differently or even not at all with the patched 3.0+ kernel (at least on my Thinkpad X220)
2. with kernel 3.0 (3.1, too) the low power mode of the i915 chipset has been disabled by default for some bug reports

My solution in the end:

kernel 3.1.0-1.dmz.2-liquorix-amd64 with grub command i915.i915_enable_rc6=1 vga=844

the i915... setting enables the low power modes of the i915 chipset with it's HD3000 internal graphics.
vga=844 was necessary (on my system) because with the pure i915... setting my display was recognized as 1024x768, not as 1366x768.

A hint, if you like to dig by yourself: additional boot parameters that are sometimes recommended aren't helpful to me.

An additional pcie_aspm=force now is counterproductive (at least with an patched kernel)! With this setting the lowest power readings are ~ 6.2 W, without ~ 5.4 W !!

So, with this new low power record on my machine I'm really happy. Thanks again!
Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1117
Reply Quote
Ok, I have to admit, that's some pretty damn good power management you got there. I'll try turning off the pcie_aspm=force parameter myself, didn't realize the force would be even worse.
Back to top
Display posts from previous:   
Page: 1, 2  Next
All times are GMT - 8 Hours