On 07/23/2015 03:07 AM, Stephane Eranian wrote: > On Wed, Jul 22, 2015 at 4:32 PM, William Cohen <wco...@redhat.com> wrote: >> Hi All, >> >> A while back I added event support to libpfm and papi for the ARM >> Cortex A53 based on the publicly available documentation. However, >> I didn't have a means of testing it until recently when I got a >> 96board dragonboard 410c >> (https://www.96boards.org/products/ce/dragonboard410c/) running Ubuntu >> 15.04 with a custom kernel: > > Nice little board and very affordable! > >> >> $ uname -a >> Linux linaro-alip 4.0.0-linaro-lt-qcom #1 SMP PREEMPT Thu Jun 18 09:43:10 >> UTC 2015 aarch64 aarch64 aarch64 GNU/Linux >> >> I took a little time to build and run libpfm from libpfm-4.5.0.tar.gz >> and the local validate test things loook good: >> >> $ ./validate >> Libpfm structure tests: >> libpfm ABI version : 0 >> pfm_pmu_info_t : Passed >> pfm_event_info_t : Passed >> pfm_event_attr_info_t : Passed >> pfm_pmu_encode_arg_t : Passed >> pfm_perf_encode_arg_t : Passed >> Libpfm internal table tests: >> checking perf (80 events): Passed >> checking perf_raw (1 events): Passed >> Architecture specific tests: >> All tests passed >> >> >> Similarly built papi from the upstream git repository and ran the tests >> >> $ sudo ./run_tests.sh >& papi_git-ac53g.log >> >> Attached are the results of the tests and below are some comments >> about the tests. >> >> The cortex a53 doesn't have support for monitoring floating point events so >> a number of tests are "SKIPPED" or "PASSED with WARNING": >> >> api.c PASSED with WARNING >> Running ctests/calibrate:calibrate.c SKIPPED >> >> >> Only supported on specific architectures: >> >> Running ctests/data_range:data_range.c SKIPPED >> Running ctests/earprofile:earprofile.c SKIPPED >> Running ctests/p4_lst_ins:p4_lst_ins.c SKIPPED >> Running ctests/zero_smp:zero_smp.c SKIPPED >> >> >> Missing PAPI_REF_CYC event: >> Running ctests/cycle_ratio:cycle_ratio.c SKIPPED >> >> Missing PAPI_L1_TCM event: >> Running ctests/tenth:tenth.c SKIPPED >> >> arm using 20MHz for REAL_CYCLES >> Running ctests/zero:zero.c FAILED >> >> Don't have openshmem >> Running ctests/zero_shmem:zero_shmem.c SKIPPED >> >> >> Seem to be related to running the run_test.sh with sudo: >> Running ctests/exeinfo:exeinfo.c FAILED >> Running ctests/profile_pthreads:profile_pthreads.c FAILED >> Running ctests/profile_twoevents:profile_twoevents.c FAILED >> >> Looks like need to take a closer look at the domain (kernel/user) control: >> Running >> components/perf_event/tests/perf_event_system_wide:perf_event_system_wide.c >> FAILED >> > Are you saying, there is something wrong withe the priv level > filtering in libpfm4 > for ARM64?
Right now the output of the test is looking wrong on the cortex ac53 board and I need to do some additional investigation. The ac53 events are very similar to other armv8 hardware. I should get an xgene or seattle machine and do some comparisons between them. The problem could be in places other than libpfm4 such as the kernel. -Will ------------------------------------------------------------------------------ _______________________________________________ perfmon2-devel mailing list perfmon2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
