Hi, On Thu, Nov 30, 2006 at 12:21:41AM +0100, Andi Kleen wrote: > > > An alternative approach might be to change perfmon so that it works out > > whether it is being called by an idle thread > > > > if ((current->flags & PF_IDLE) && (other stuff to do with irqs?)) > > return; > > The problem is that the performance counters just keep running in the CPU. > Perfmon needs to do something actively to disable them. > Exactly.
> Actually on x86 they usually stop in true idle state in hardware, but > they don't do in polling mode and it sometimes seems to depend on > the firmware. So it mostly would be for idle=poll > Most likely. But we also have the deal with other architectures such as MIPS,IA64, and PPC. > But if you do walk clock time profiling exactly because they stop > a profiler should account for this somehow. Otherwise the profiling time > doesn't add up to 100% > Yes. Note that perfmon does maintain the duration when monitornig was active. So it is possible to determine active time and use it to scale counts. -- -Stephane - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
