On Mon, 2007-10-22 at 22:39 -0700, Stephane Eranian wrote: > Yanmin, > > On Tue, Oct 23, 2007 at 08:56:51AM +0800, Zhang, Yanmin wrote: > > I Instrumented pfm_core_probe_pmu and my processor famiy is 6, model is 15. > > Below is the log messages. > > > > perfmon: not an AMD processor > > perfmon: AMD64 PMU detection failed > > perfmon: unsupported family=6 > > perfmon: Intel P4 PMU detection failed > > YMZ---pfm_core_probe_pmu: x86=6 x86_model=15 > > perfmon: nmi_watchdog=0 nmi_active=0 force_nmi=0 > > perfmon: PEBS supported, enabled > > perfmon: Intel Core PMU detected, 4 PMCs, 5 PMDs, 5 counters (31 bits) > > perfmon: Intel Core PMU installed > > perfmon: detected architecural perfmon v2 > > perfmon: num_gen=2 width=40 num_fixed=3 width=40 > > perfmon: register Intel architectural PMU error -16 > > perfmon: added sampling format pebs64_p4 > > perfmon: added sampling format pebs_core > > This looks all normal. Intel Core succeeds, Intel architectural would > have succeeded without Intel Core. There can be at most ONE description > registered. > > Now, I realize that if the link order would have been different you would > not necessarily have had the same sequence. For instance, you could have > seen: Intel Arch succeeds, Intel Core fails. And that would work but with > limited features, e.g., no PEBS. I will add a check in perfmon_arch_intel.c > to fails for 6/15. This way, it will always give a chance to > perfmon_intel_core.c. > For newer models, that use 6/XX (XX != 15), this will work correctly, as > perfmon_intel_core will fail, and perfmon_intel_arch will succeed if it > finds architectural perfmon v1 or v2. That will be great while both CONFIG=y/m.
--yanmin _______________________________________________ perfmon mailing list [email protected] http://www.hpl.hp.com/hosted/linux/mail-archives/perfmon/
