Re: [perfmon2] [PATCH] 1/1 - add kernel support for POWER4 and POWER6 to perfmon2
Status: Beta
Brought to you by:
seranian
From: Stephane E. <er...@hp...> - 2007-11-29 12:25:45
|
Phil, On Thu, Nov 29, 2007 at 12:07:52PM +0100, Philip Mucci wrote: > Hi folks, > > The Sparc64 port has registers like this as well. While I've long > argued that perfmon2 should have a general mechanism to support > counters without an interrupt bit (indeed even architectures that > don't give you a PMU interrupt), I'd be just as happy in this case if: > > a) perfmon complained if the user requested notification/sampling on > a non-interrupting counter > b) PAPI had knowledge of counters that could not interrupt > I looked at the code again. I had forgotten about something which was added for Barcelona. I was wrong about perfmon2 not knowing about interrupt-generating data register. There is a flag (PFM_REG_INTR) which is used to build a bitmask of interrupt-generating pmds. The idea is to limit scanning of PMD on interrupt to those known to generate some. Today, pfm_write_pmds() ignores PFM_REGFL_OVFL_NOTIFY on non counter. I think we can easily change that to return an error when set on non-interrupting data registers. I'll make that change if that helps. |