Re: [PATCH v2 14/16] perf tools: add new mem command for memory access profiling

[Arnaldo Carvalho de Melo <acme@ghostprotocols.net>]

Em Tue, Nov 06, 2012 at 04:57:59PM +0100, Stephane Eranian escreveu:
> On Tue, Nov 6, 2012 at 4:50 PM, Arnaldo Carvalho de Melo
> <acme@ghostprotocols.net> wrote:
> > Em Tue, Nov 06, 2012 at 12:44:46PM -0300, Arnaldo Carvalho de Melo escreveu:
> >> [root@sandy ~]# perf record -g -a -e cpu/mem-stores/
> >> ^C[ perf record: Woken up 25 times to write data ]
> >> [ perf record: Captured and wrote 7.419 MB perf.data (~324160 samples) ]
> >>
> >> Yay, got some numbers.
> >
> > But then the results out of:
> >
> > $ perf mem -t load rep --stdio
> >
> > Are bogus at least on the callchains:
> >
> I think it's because we are not using the period in the hist_entry but
> the cost and there
> is no cost for stores.
> 
> I think you should start with loads and no callchain. I am interested
> in your toughts on how to
> fix the data symbol resolution problem in perf. In V2, I modified the
> kernel to at least
> help perf distinguish MAP__FUNCTION from MAP__VARIABLE. But there is
> still something
> wrong.

Yeah, I saw that, now that I'm playing with this patchset I'll work on
that, after lunch.
 
> 
> > # ========
> > # captured on: Tue Nov  6 12:46:21 2012
> > # hostname : sandy.ghostprotocols.net
> > # os release : 3.7.0-rc2+
> > # perf version : 3.7.rc4.gfaa41f
> > # arch : x86_64
> > # nrcpus online : 8
> > # nrcpus avail : 8
> > # cpudesc : Intel(R) Core(TM) i7-2920XM CPU @ 2.50GHz
> > # cpuid : GenuineIntel,6,42,7
> > # total memory : 16220228 kB
> > # cmdline : /home/acme/bin/perf record -g -a -e cpu/mem-stores/
> > # event : name = cpu/mem-stores/, type = 4, config = 0x2cd, config1 = 0x0, config2 = 0x0, excl_usr = 0, excl_kern = 0, excl_ho
> > # HEADER_CPU_TOPOLOGY info available, use -I to display
> > # HEADER_NUMA_TOPOLOGY info available, use -I to display
> > # pmu mappings: cpu = 4, software = 1, tracepoint = 2, uncore_cbox_0 = 6, uncore_cbox_1 = 7, uncore_cbox_2 = 8, uncore_cbox_3
> > # ========
> > #
> > # Samples: 98  of event 'cpu/mem-stores/'
> > # Total cost : 98
> > # Sort order : cost,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked
> > #
> > # Overhead      Samples     Cost             Memory access                                          Symbol               Share
> > # ........  ...........  .......  ........................  ..............................................  ..................
> > #
> >     19.39%           19      N/A                            [k] csd_unlock                                  [kernel.kallsyms]
> >             |
> >             --- csd_unlock
> >                |
> >                |--6242.11%-- generic_smp_call_function_single_interrupt
> >                |          smp_call_function_single_interrupt
> >                |          call_function_single_interrupt
> >                |          cpuidle_enter
> >                |          cpuidle_enter_state
> >                |          cpuidle_idle_call
> >                |          cpu_idle
> >                |          |
> >                |          |--85.08%-- start_secondary
> >                |          |
> >                |           --14.92%-- rest_init
> >                |                     start_kernel
> >                |                     x86_64_start_reservations
> >                |                     x86_64_start_kernel
> >                |
> >                |--100.00%-- smp_call_function_single_interrupt
> >                |          call_function_single_interrupt
> >                |          cpuidle_enter
> >                |          cpuidle_enter_state
> >                |          cpuidle_idle_call
> >                |          cpu_idle
> >                |          start_secondary
> >                 --97088126703734472704.00%-- [...]
> >
> >      5.10%            5      N/A                            [k] _raw_spin_lock_irqsave                      [kernel.kallsyms]
> >             |
> >
> >
> >
> > Ideas?
> >
> > - Arnaldo