monitor/mwait support for idle
Jeff Roberson
jroberson at jroberson.net
Mon Apr 21 22:43:13 UTC 2008
On Mon, 21 Apr 2008, Andrew Gallatin wrote:
>
> Jeff Roberson writes:
> > http://people.freebsd.org/~jeff/mwait.diff
> >
> > This patch implements support for the x86/amd64 monitor and mwait
> > instructions in the idle loop. This also implements idle loop selection
> > via a sysctl string. The following loops are supported, in
> > decreasing order of performance and power consumption:
> >
> > spin - Simply returns
> > mwait - Always use mwait to sleep. CPU enters C0 or C1 depending on
> > how busy it is.
> > mwait_hlt - Use mwait when busy but fall back to hlt/acpi when not.
> > hlt - pure hlt loop
> > acpi - uses acpi_cpu_idle if available and hlt if not. This is the
> > default.
> >
>
> Something which may be a bit confusing is that machines like recent
> Core2 Xeons will go into C1E when hlt is executed, depending on an MSR
> setting that most BIOSes enable (bit 25, MSR 0x1a0). I think C1E
> might be a deeper sleep than what is reached by mwait. I confess that
> I don't know much about C1E, other than it kills network io intensive
> performance, and I turn off this (mis)feature whenever I can.
>
> It will be interesting to see how much mwait helps. I'm sure it will
> save power over my current workaround of disabling hlt altogether :)
I don't have any useful measurements for power. For performance I see as
much as 10-15% in some workloads. This is due to both latency and cpu
time reduction in wakeup heavy workloads that also idle a lot.
Jeff
>
> Drew
>
More information about the freebsd-arch
mailing list