On Fri, Feb 05, 2010 at 12:45:21PM -0500, Len Brown wrote:
> Jeff,
> What do you see if you apply just the patch below?
> 
> Also, in addition to "powertop -d" to show what the kernel requests,
> please run turbostat to show what the hardware actually did:
> 
> http://userweb.kernel.org/~lenb/acpi/utils/pmtools-latest/turbostat/turbostat.c
> 
> eg.
> # turbostat -d -v sleep 5
> 
> thanks,
> -Len Brown, Intel Open Source Technology Center

With the patch, powertop reports good C3 residency and wakeups
remain very low.  Seems to work.  :)

I attached the powertop & turbostat output with this patch.

However, this confuses me.  In a previous experiment in the
acpi_processor_setup_cpuidle() function, I replaced the pointer
to acpi_idle_enter_bm() with a pointer to
acpi_idle_enter_simple() even when bm_check is nonzero.  With
that, I was able to get into C3, but the wakeups ballooned.  But
the difference between what I did, and what you did, is the
difference between acpi_idle_enter_bm() with acpi_idle_bm_check()
returning zero and acpi_idle_enter_simple().  Those code paths
look almost identical.  The bm path calls acpi_unlazy_tlb(), and
doesn't appear to call the ACPI_FLUSH_CPU_CACHE(), and they call
sched_clock_idle_sleep_event() in different places.  I don't
understand why any of these differences would have had any
significant effect on wakeups.

I'm left wondering if it's a problem on my part.  I should repeat
that previous  experiment and see if there really is something
significantly different there.

BTW, getting a bit off topic, but since the two code paths are
almost identical, is there any reason not to unite them?
Something like the attached patch might work?

Thanks,
Jeff