On Nov 21, 2007 9:08 PM, Gavin Maltby <[EMAIL PROTECTED]> wrote:
> Hi,
>
> Mladen Nikitovic wrote:
> > Hi,
> >
> > I've been informed that there is a "power-aware" version of the original
> > idle-thread in OpenSolaris. My understanding is that this version should
> > execute the halt-instruction, which in turn disables the clock-tree of the
> > processor. In the original version, the idle-thread continues checking
> > different queues etc.
> >
> > As the power-aware idle-thread is not used by default, I'm wondering how I
> > can go about replacing the original idle-thread? Is there a smooth
> > approach? Would be happy for any pointers...At this point I'm not exactly
> > aware of where the code for the power-aware idle thread is located in the
> > source-tree.
>
> The common idle loop is idle() in uts/common/disp/disp.c. Have a look
> at its source and you'll see it calls the function pointed to by
> idle_cpu when it can find no work to do. The uniprocessor loop
> is easier to follows:
>
> while (ncpus == 1) {
> if (cp->cpu_disp->disp_nrunnable == 0) {
> (*idle_cpu)();
> continue;
> }
> idle_exit();
> swtch();
>
> idle_enter(); /* returned from swtch */
> }
>
> Different architectures can assign custom idle function to
> idle_cpu. The default is generic_idle_cpu (see definition
> of idle_cpu pointer in disp.c). For i86pc it can be
> set to cpu_idle or to cpu_idle_mwait.
>
> cpu_idle calls mach_cpu_idle which for i86pc calls the
> halt instruction you refer to.
>
> cpu_idle_mwait is a variant on systems that support monitoring
> a line of memory for their wakeup condition (halt support
> above uses an interrupt).
depends on the macro "__xpv", if it's defined, cpu_idle_mwait is used.
>
> These are basic power awareness aspects - no frequency scaling, voltage
> change etc.
depends on if C1E is supported. If it is, frequency and voltage are turned down.
>There is now support for Intel Speedstep. I've lost
> track of the story for AMD - I know we decided it can't be
> done on multiprocessor systems before family 0x10, but I vaguely
> recall an intention to try it for family 0xf for single-socket
> systems.
>
-Aubrey
Intel OpenSolaris Team.
_______________________________________________
opensolaris-code mailing list
opensolaris-code@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
