[home]

Linux

amd76x_pm: power management for AMD K7 SMP

This module has been written by Tony Lindgren in 2002. It enables C2/C3 powersaving mode on AMD K7 SMP systems. This is a very useful thing, as AMD K7 consumes a lot of energy. With this module you may save up to 70 – 90 W in idle mode compared to the default idle mode. See Tony's page http://www.muru.com/linux/amd-smp-idle/ for details.

Several people have been involved in developing, improving and testing this driver. Some of them are:

Tony Lindgren, Johnathan Hicks, Alan Cox, Pasi Savolainen

After some changes in the 2.6 kernel Tony's and Pasi's patches weren't usable any longer and I started to maintain a more recent version just for myself. By doing this I got more and more familiar with the code and started to play around a bit. My main motivation was the fact that I realized a bad influence from this module on the system clock. Some small changes showed an improvement and I asked at LKML if this modified module might be included in some standard kernel tree. I got some suggestions for further improvements from and started to rewrite the idle-loop. And here is the result.

Some people reported that this module works for them. For known issues see below. Just give it a try! All you need to do is apply the recent patch, enable “AMD76x Native Power Management support” in “Power management options” / “ACPI” and read Documentation/amd76x_pm.txt.

Known issues

• System going into sleep instead of idling once after loading the module. Observed on Tyan S2460. (1 report quite some time ago)
• Occasional hard locks possibly caused by amd76x_pm. Observed on Chaintech 7KDD, Tyan S2460 and Tyan S2466. (several reports)
• Weird keyboard repeating after some uptime (1-2 days). Observed on Tyan S2466 and Chaintech 7KDD. (No more reports for recent kernels)
• Disturbing noise with audio playback.
• Failing SATA RAID-1 disks (r/w errors, possibly losing one mirror) on Promise controllers. (No more reports for recent kernels)

Patch usage

• A "standard" patch that provides the base functionality for C2 idling. If you want to just use amd76x_pm, this patch is yours.
• An "extra" patch that provides support for NTH, POS and C3. These extra features are disabled by default. If you are willing to take the risk of using highly experimental code and you feel able to cope with the sources, this might be your patch.
  NB: The "extra" patch is completely outdated.

Release Notes

2.6.28-2
added support for cpuidle

2.6.28-1
adapted to changes in the kernel's idle loop
calculation of sleep_rate removed. (It was erroneous anyway.)
removed the lazy_idle loop
renamed spin_idle to max_poll. The behaviour has completely changed!
extended mode to C0-C3
introduced busy_mode
borrowed some code from processor_idle.c

2.6.24-1
trivial 2.6.24 port

2.6.23-1
trivial 2.6.23 port

2.6.22-1
2.6.22 port

2.6.21-2
removed "synchronized sleep", i.e. one CPU might enter Cx state if all other CPUs are idle but not nessecarily entering Cx state.
run "inner loop" with interrupts enabled
adopted some system clock related functionalities from processor_idle.c

2.6.21-1
sleep_rate calculation using 64bit arithmetics

2.6.20-1
workqueue API changes in 2.6.20

2.6.19-1
Trivial 2.6.19 port.

2.6.18-2
Trivial 2.6.18 port.

2.6.17-2
Removed some double-locking.
Code lifting.

2.6.17-1
Some changes concerning memory barriers.

2.6.16-1
Removed sleep_time in favor of sleep_rate, which gives a better idling measure.
C2/C3 mode now changeable at runtime. Note: syntax has changed.

2.6.15-6
Redesigned the idle function again. Now there shouldn't be any situations left where not all CPUs try to enter C2/C3.
Removed the (ugly) parsing of watch_irqs. Using two separate arrays watch_irqs and watch_limits now.
sleep_time calculation and sysfs entries added.
This patch isn't well tested yet.
Note: module parameter syntax has changed.

2.6.15-4
Major redesign of the idle loop again. Now there is a two-phase process to make sure all or no CPUs go into C3. In phase one each CPU waits for the other to become idle. In phase two all CPUs come to an agreement about going C3 or not. Now using spinlocks. Simplicity has gone :-(

2.6.15-3
Switched from C2 to C3 idling, inspired by processor_idle.c. C2 idling has just been replaced by C3 idling, there is no C2/C3 transition.
Restoring all touched northbridge and southbridge register bits to their original values on module unload.
Note: Please check lazy_idle and spin_idle values, the behaviour might have changed again.

2.6.15-2
Removed a race condition in the idle loop. Clock stability has improved now, maybe other issues also have gone.

2.6.15-1
Using percpu variables.
Note: Please check lazy_idle and spin_idle values, the behaviour might have changed.

2.6.14-2
Minor changes in memory layout.

2.6.14-1
No major changes from 2.6.13-1.

2.6.13-1
Mainly 2.6.13 API changes. No major changes from 2.6.12-jo4-1.

2.6.12-jo4-1
Introduced an "irq rate watcher". This task checks the interrupt rate of some interrupts and prevents entering C2 mode if any of them exceeds a certain limit. This improves the throughput of disk and network devices and gives a better system clock stability too. There are a lot of changes, please take a look at Documentation/amd76x_pm.txt for details.

2.6.12-jo3
The idle loop has been rewritten completely. The impact on the system clock has (hopefully) been reduced. Please check your settings for lazy_idle as the impact on cooling and clock stability may have changed. NTH and POS from the -extra patch are still untestet. C3 idling seems to work.

2.6.12-jo2
A first attempt to improve amd76x_pm.

Download

Here are some patches:

Have fun!

-jo