Most recent kernel where this bug did not occur:
Linux kernel should have support for software coordination of P-states and
support for _PSD ACPI 3.0 method.
The original patch for this was sent on the mailing list here:
This patch will support software coordination of P-states across logical CPUs
in cases where CPUs share a P-state and BIOS supports the _PSD method and lets
OS know of this P-states getting shared across CPUs.
There has been some other changes in the code being touched by the above patch
in last two months. Below is the rebased patchset to current base kernel
Created attachment 6826 [details]
patch 1/4 : P-state software coordination changes for acpi core
Created attachment 6827 [details]
patch 2/4 : P-state software coordination changes for acpi-cpufreq driver
Created attachment 6828 [details]
patch 3/4 : P-state software coordination changes for speedstep-centrino driver
Created attachment 6829 [details]
patch 4/4 : P-state software coordination final changes to turn on in _PDC
Created attachment 7219 [details]
patch 2/4 : P-state software coordination changes for acpi-cpufreq driver (rebased with 2.6.16-rc1-git6)
Updated patchset against 2.6.16-rc1-git6. Patch 1, 3 and 4 above applies
cleanly. But, patch 2had a conflict and Updating that one here.
patches 1-4 applied to acpi-test tree
Created attachment 7878 [details]
Patch to treat hardware coordinated CPUs as independent in cpufreq
Addon patch that applies over all the original patches. This changes the
behavior with hardware coordination. Hardware coordination will be treated as
independent CPUs by cpufreq. That way we can do per cpu monitoring of P-states.
patches 1-4 shipped in 2.6.17-git9
patch in comment #8 added to acpi-test tree
patch in comment #8 shipped after 2.6.17-git17
in time for 2.6.18-rc1. closed.