On Wed, 3 Sep 2014, Gary Mohr wrote: > When I added a modified version of check_scheduability() to the uncore > component it now detects that the two events with constraint conflicts > cannot be used together. This allows PAPI to return an error at add > event time which is better than reporting it when trying to read the > results after the test is completed.
It's odd this is failing though. Which events are causing the issue? It sounds like a bug in the uncore constraint code. If you can reproduce the issue using the "perf" tool then you can probably get the kernel developer's attention. > One of the changes was to remove the test if the NMI watchdog was active > (if (nmi_watchdog_active) return 1) from the function > bug_check_scheduability(). I am hoping that NMI only affects core > events and is not important when using uncore events. Does this sound > like a safe assumption to you ?? Yes, the NMI watchdog just grabs a cycle counter event typically, it wouldn't be grabbing any uncore events. > So I guess since we do not have much influence with the kernel folks, we > will have to go with this PAPI kludge to get better error reporting. I wouldn't give up on getting a kernel response, it just takes some patience and a good reproducible testcase. Though if the bug is in the wild already then we'll probably need the kludge forever. Vince ------------------------------------------------------------------------------ Slashdot TV. Video for Nerds. Stuff that matters. http://tv.slashdot.org/ _______________________________________________ perfmon2-devel mailing list perfmon2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
