Hi, I suspect the problem is that the kernel does not recognize the model number for NHM-EX. I am sure it is not in my latest 2.6.29 patch. But if you use the GIT tree, model 46 is there so it should recognize it.
What's in /sys/kernel/perfmon/pmu_desc/model? Did you compile perfmon_intel_arch or perfmon_intel_nhm? Looking at the code in perfmon_intel_arch, I found some missing Nehalem model numbers. But that shouldn't have caused the problem. On Tue, Nov 30, 2010 at 7:05 AM, Vishal Gupta <vis...@cc.gatech.edu> wrote: > Hi, > > Thanks for pointing me to that post. So I downloaded the latest libpfm > and pfmon versions from CVS, but I still get the same error with them. > pfmon is detecting Nehalem PMU on my machine...below is the output > with -I option. I'm also pasting the output with debug and verbose > options enabled if that can help in figuring out the issue. Thanks for > your help. > > r...@host pfmon # pfmon -I > detected host CPUs: 64-way 1064MHz/24.0MB -- Intel(R) Xeon(R) CPU > X7560 @ 2.27GHz (stepping 6) > detected pfmon PMU model: Intel Nehalem > detected libpfm PMU model: Intel Nehalem > max counters/set: 7 > supported pfmon PMU models: [AMD64] [Pentium 4] [Intel Core] [Intel > Atom] [Intel Nehalem] [Intel architectural PMU] > supported sampling modules: [inst-hist] [detailed] [compact] [raw] > pfmlib version: 3.10 > kernel perfmon version: 2.9 > kernel clock resolution: 1ns (1000000000Hz) > host kernel architecture: x86_64 > > > r...@host pfmon # ./pfmon --debug --verbose ls > pfmlib_intel_nhm.c (pfm_nhm_get_num_event_masks.1493): event > UNHALTED_CORE_CYCLES numasks=0 > measuring at user privilege level ONLY > pfmlib_intel_nhm.c (pfm_nhm_get_num_event_masks.1493): event > OFFCORE_RESPONSE_0 numasks=17 > pfmlib_intel_nhm.c (pfm_nhm_get_num_event_masks.1493): event > OFFCORE_RESPONSE_0 numasks=17 > pfmlib_intel_nhm.c (pfm_nhm_get_num_event_masks.1493): event > OFFCORE_RESPONSE_0 numasks=17 > 1 event set(s) defined > long sampling periods(val/mask/seed): 0/0x0/0 > short sampling periods(val/mask/seed): 0/0x0/0 > unable to detect unavailable pmcs > detection of unavailable registers failed, leave it to the kernel to > decide > run_measurements.480: [5296] library dispatch for set0 > pfmlib_intel_nhm.c (pfm_nhm_get_num_event_masks.1493): event > UNHALTED_CORE_CYCLES numasks=0 > [FIXED_CTRL(pmc16)=0xa0 pmi0=1 en0=0x0 any0=0 pmi1=1 en1=0x2 any1=0 > pmi2=1 en2=0x0 any2=0] UNHALTED_CORE_CYCLES > [FIXED_CTR1(pmd17)] > pmd setup for event set0: > [pmd17 set=0 ival=0x0 long_rate=0x0 short_rate=0x0 mask=0x0 seed=0 > randomize=n] > exec-pattern=* > [5299] started task: ls > pfmon_sdesc_new.883: [5296] [5299] SDESC_NEW /bin/ls parent=4273792 > pid=5296 tid=5299 flags=0x7fff000014b3 > follow_exec=n follow_vfork=n follow_fork=n follow_pthread=n > [5299] 64-bit binary > task_pfm_init.1281: [5296] in: [5299] ctxid=-1 monitoring=0 > refcnt=1: /bin/ls > [5299] monitoring /bin/ls... > task_pfm_init.1393: [5296] setup perfmon ctx for [5299] monitoring=1 > refcnt=1: /bin/ls > can't create perfmon context: Function not implemented > > Vishal > > On Nov 29, 2010, at 19:57, Corey Ashford wrote: > >> On 11/29/2010 01:25 PM, Vishal Gupta wrote: >>> I digged into it a little bit more. The machine I'm using is a >>> quad-socket machine with four Intel X7560 processors. X7560 is a >>> Nehalem-EX processor. As pfmon runs fine on another Nehalem machine >>> I've >>> tested, does that mean that pfmon is broken for Nehalem-EX >>> processors? >>> Any clues on how to go about fixing this? >>> >>> Error message: >>> >>> r...@host linux-2.6 # pfmon ls >>> detection of unavailable registers failed, leave it to the kernel >>> to decide >>> can't create perfmon context: Function not implemented >>> >> >> I see a posting from Vince Weaver back in June that says for Nehalem- >> EX you need to check out the latest libpfm3 from CVS and build that, >> rather than using the released version. >> >> I'm not certain of that, though, because I thought there had been a >> final release of libpfm3 since then. >> >> In any case, it wouldn't hurt to give that a shot. >> >> - Corey > > > ------------------------------------------------------------------------------ > Increase Visibility of Your 3D Game App & Earn a Chance To Win $500! > Tap into the largest installed PC base & get more eyes on your game by > optimizing for Intel(R) Graphics Technology. Get started today with the > Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs. > http://p.sf.net/sfu/intelisp-dev2dev > _______________________________________________ > perfmon2-devel mailing list > perfmon2-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/perfmon2-devel > ------------------------------------------------------------------------------ Increase Visibility of Your 3D Game App & Earn a Chance To Win $500! Tap into the largest installed PC base & get more eyes on your game by optimizing for Intel(R) Graphics Technology. Get started today with the Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs. http://p.sf.net/sfu/intelisp-dev2dev _______________________________________________ perfmon2-devel mailing list perfmon2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
