Greetings and Happy New Year Stephane,
Ok, thanks to your tips, and digging through Vince’s perf_event_tests codes, I
tracked down this bug - it’s the exclude_guest = 1 flag. The systems I’m
running on are paranoid level 2. Here’s the GDB log… What’s the right way to
fix this? I have a patched library working for me, but it’s not generalized.
Let me know and I’ll send you a patch.
I’m preparing an answer for the other email about Intel event names as well...
Regards,
> Reading symbols from /home/pjmucci/bob/perf_examples/self...done.
> (gdb) break perf_event_open
> Breakpoint 1 at 0x4015ee: file
> /home/pjmucci/bob/perf_examples/../include/perfmon/perf_event.h, line 506.
> (gdb) run
> Starting program: /home/pjmucci/bob/perf_examples/self
> [Thread debugging using libthread_db enabled]
>
> Breakpoint 1, perf_event_open (hw_event_uptr=0x73e1b0, pid=0, cpu=-1,
> group_fd=-1, flags=0) at
> /home/pjmucci/bob/perf_examples/../include/perfmon/perf_event.h:506
> 506 return syscall(
> Missing separate debuginfos, use: debuginfo-install
> glibc-2.12-1.132.el6_5.4.x86_64
> (gdb) p *hw_event_uptr
> $1 = {type = 0, size = 112, config = 0, {sample_period = 0, sample_freq = 0},
> sample_type = 0, read_format = 3, disabled = 1, inherit = 0, pinned = 0,
> exclusive = 0, exclude_user = 0, exclude_kernel = 0, exclude_hv = 0,
> exclude_idle = 0, mmap = 0, comm = 0, freq = 0, inherit_stat = 0,
> enable_on_exec = 0, task = 0, watermark = 0, precise_ip = 0, mmap_data = 0,
> sample_id_all = 0, exclude_host = 0, exclude_guest = 1,
> exclude_callchain_kernel = 0,
> exclude_callchain_user = 0, mmap2 = 0, comm_exec = 0, use_clockid = 0,
> __reserved_1 = 0, {wakeup_events = 0, wakeup_watermark = 0}, bp_type = 0,
> {bp_addr = 0, config1 = 0}, {bp_len = 0, config2 = 0}, branch_sample_type =
> 0,
> sample_regs_user = 0, sample_stack_user = 0, clockid = 0, sample_regs_intr
> = 0, aux_watermark = 0, __reserved_2 = 0}
> (gdb) c
> Continuing.
> self: cannot open event 0: Invalid argument
>
> Program exited with code 01.
> (gdb) run
> Starting program: /home/pjmucci/bob/perf_examples/self
> [Thread debugging using libthread_db enabled]
>
> Breakpoint 1, perf_event_open (hw_event_uptr=0x73e1b0, pid=0, cpu=-1,
> group_fd=-1, flags=0) at
> /home/pjmucci/bob/perf_examples/../include/perfmon/perf_event.h:506
> 506 return syscall(
> (gdb) set hw_event_uptr->exclude_guest=0
> (gdb) c
> Continuing.
> self: cannot open event 0: Permission denied
>
> Program exited with code 01.
> (gdb) run cycles:u
> Starting program: /home/pjmucci/bob/perf_examples/self cycles:u
> [Thread debugging using libthread_db enabled]
>
> Breakpoint 1, perf_event_open (hw_event_uptr=0x73e1b0, pid=0, cpu=-1,
> group_fd=-1, flags=0) at
> /home/pjmucci/bob/perf_examples/../include/perfmon/perf_event.h:506
> 506 return syscall(
> (gdb) set hw_event_uptr->exclude_guest=0
> (gdb) c
> Continuing.
> INITIAL: 2,788 cycles:u (0.00% scaling, raw=2,788, ena=6,001,
> run=6,001)
> Final counts:
> FINAL: 25,899,440,206 cycles:u (0.00% scaling, raw=25,899,440,206,
> ena=9,979,548,313, run=9,979,548,313)
>
> On Dec 27, 2015, at 4:24 AM, Stephane Eranian <eran...@googlemail.com> wrote:
>
> Hi Phil,
>
> On Thu, Dec 24, 2015 at 1:55 AM, Philip Mucci <mu...@icl.utk.edu
> <mailto:mu...@icl.utk.edu>> wrote:
> Hi folks,
>
> Does the nmi_watchdog prevent libpfm from working at all? I’m on an older
> Sandia system (compton)
>
> No, this only affects the number of events you can measure simultaneously.
>
> Is this a kernel version and perf attr data structure problem? It can’t be
> NMI since perf with raw registers seems to work fine…
>
> Looks like it because you have perf stat working. Your kernel is set to
> paranoid 2, which means you can only monitor
> user level execution.
>
> Normally, the key field is attr.size. It should always be set to attr.size =
> sizeof(attr). But that is in the example.
> Furthermore, the kernel handles older attr struct (with smaller size). Bigger
> are not a problem as long as the excess
> fields are zero. What could cause a problem is:
> - attr.reserved1 != 0
> - attr.sample_type has unknown bits set for kernel
> - attr.read_format has unknown bits set for kernel
> - attr.branch_sample_type has unknown bits set for kernel
>
> I would check that first.
>
>
>
> [pjmucci@compton1 perfmon2-libpfm4.git]$ uname -a
> Linux compton1 2.6.32-131.0.15.el6.x86_64 #1 SMP Tue May 10 15:42:40 EDT 2011
> x86_64 x86_64 x86_64 GNU/Linux
> [pjmucci@compton1 perfmon2-libpfm4.git]$ cat /etc/*ease
> Red Hat Enterprise Linux Workstation release 6.1 (Santiago)
> Red Hat Enterprise Linux Workstation release 6.1 (Santiago)
>
> [pjmucci@compton32 perfmon2-libpfm4.git]$ cat /etc/*ease
> Red Hat Enterprise Linux Server release 6.2 (Santiago)
> Red Hat Enterprise Linux Server release 6.2 (Santiago)
> [pjmucci@compton32 perfmon2-libpfm4.git]$ uname -a
> Linux compton32 2.6.32-220.el6.x86_64 #1 SMP Wed Nov 9 08:03:13 EST 2011
> x86_64 x86_64 x86_64 GNU/Linux
>
> [pjmucci@compton32 perfmon2-libpfm4.git]$ cat
> /proc/sys/kernel/perf_event_paranoid
> 2
> [pjmucci@compton32 perfmon2-libpfm4.git]$ cat /proc/sys/kernel/nmi_watchdog
> 1
> [pjmucci@compton32 perfmon2-libpfm4.git]$ ./examples/check_events
> snb_ep::UOPS_RETIRED:TOTAL_CYCLES:u
> Requested Event: snb_ep::UOPS_RETIRED:TOTAL_CYCLES:u
> Actual Event: snb_ep::UOPS_RETIRED:ALL:k=0:u=1:e=0:i=0:c=16:t=0
> PMU : Intel Sandy Bridge EP
> IDX : 148897869
> Codes : 0x105101c2
> [pjmucci@compton32 perfmon2-libpfm4.git]$ perf stat -e r105301c2:uh sleep 1
>
> Performance counter stats for 'sleep 1':
>
> 0 r105301c2:uh
>
>
> 1.001672975 seconds time elapsed
> [pjmucci@compton32 perfmon2-libpfm4.git]$ ./perf_examples/self_count
> snb_ep::UOPS_RETIRED:TOTAL_CYCLES:u
> self_count: cannot open event 0: Invalid argument
>
>
> [pjmucci@compton32 perfmon2-libpfm4.git]$ ./perf_examples/evt2raw
> snb_ep::UNHALTED_CORE_CYCLES:u
> r53003c:uh
> [pjmucci@compton32 perfmon2-libpfm4.git]$ ./perf_examples/evt2raw
> snb_ep::UNHALTED_REFERENCE_CYCLES:u
> r530300:uh
> [pjmucci@compton32 perfmon2-libpfm4.git]$ perf stat -e r530300:uh sleep 1
>
> Performance counter stats for 'sleep 1':
>
> 3 r530300:uh
>
>
> 1.002638998 seconds time elapsed
>
> [pjmucci@compton32 perfmon2-libpfm4.git]$ perf stat -e r53003c:uh sleep 1
>
> Performance counter stats for 'sleep 1':
>
> 300,065 r53003c:uh
>
>
> 1.001350589 seconds time elapsed
>
> [pjmucci@compton32 perfmon2-libpfm4.git]$ ./perf_examples/self_count
> snb_ep::UNHALTED_CORE_CYCLES:u
> self_count: cannot open event 0: Invalid argument
>
>
>
>
> ------------------------------------------------------------------------------
>
> _______________________________________________
> perfmon2-devel mailing list
> perfmon2-devel@lists.sourceforge.net
> <mailto:perfmon2-devel@lists.sourceforge.net>
> https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
> <https://lists.sourceforge.net/lists/listinfo/perfmon2-devel>
>
>
------------------------------------------------------------------------------
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
