Re: [PATCH 10/13] perf report: Do not blindly use env->cpu[al.cpu].socket_id

["Wangnan (F)" <wangnan0@huawei.com>]

On 2015/9/10 3:50, Arnaldo Carvalho de Melo wrote:
> From: Arnaldo Carvalho de Melo <acme@redhat.com>
>
> As al.cpu may be -1, i.e. no PERF_SAMPLE_CPU, and env->cpu may be NULL.
>
> Rely instead on the work now done in perf_event__preprocess_sample(),
> that does all those checks.
>
> Reported-by: Wang Nan <wangnan0@huawei.com>
> Based-on-a-patch-by: Jiri Olsa <jolsa@kernel.org>
> Cc: Adrian Hunter <adrian.hunter@intel.com>
> Cc: Borislav Petkov <bp@suse.de>
> Cc: David Ahern <dsahern@gmail.com>
> Cc: Frederic Weisbecker <fweisbec@gmail.com>
> Cc: Kan Liang <kan.liang@intel.com>
> Cc: Namhyung Kim <namhyung@kernel.org>
> Cc: Stephane Eranian <eranian@google.com>
> Cc: Wang Nan <wangnan0@huawei.com>
> Link: http://lkml.kernel.org/n/tip-2lw80g5ehsrec7tozhmnjgxw@git.kernel.org
> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
> ---
>   tools/perf/builtin-report.c | 4 ----
>   1 file changed, 4 deletions(-)
>
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index 4b432453922f..9b5083630a56 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -150,7 +150,6 @@ static int process_sample_event(struct perf_tool *tool,
>   		.add_entry_cb 		= hist_iter__report_callback,
>   	};
>   	int ret = 0;
> -	struct perf_env *env = evsel->evlist->env;
>   
>   	if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
>   		pr_debug("problem processing %d event, skipping it.\n",
> @@ -158,9 +157,6 @@ static int process_sample_event(struct perf_tool *tool,
>   		return -1;
>   	}
>   
> -	/* read socket id from perf.data for perf report */
> -	al.socket = env->cpu[al.cpu].socket_id;
> -
>   	if (rep->hide_unresolved && al.sym == NULL)
>   		goto out_put;
>   

I tested this patch on an 8 cores SOC. No segfault found now.

Normal case:

  # ./perf record -a ls
  # ...
  # ./perf_arm64 report -v  --stdio -s socket,cpu
  build id event received for [kernel.kallsyms]: ...
  # To display the perf.data header info, please use 
--header/--header-only options.
  #
  symsrc__init: cannot get elf header.
  Looking at the vmlinux_path (7 entries long)
  Failed to open /proc/kcore. Note /proc/kcore requires CAP_SYS_RAWIO 
capability to access.
  Using /proc/kallsyms for symbols
  Failed to open /sbin/adbd, continuing without symbols
  #
  # Total Lost Samples: 0
  #
  # Samples: 291  of event 'cycles'
  # Event count (approx.): 40968659
  #
  # Overhead  Socket  CPU
  # ........  ......  ...
  #
      58.01%     000  004
      35.61%     000  007
       3.65%     000  005
       1.23%     001  000
       1.18%     001  001
       0.19%     000  006
       0.08%     001  002
       0.07%     001  003

If one of CPUs is offlined before 'perf record':

  # ./perf record -a ls
  # ...
  # ./perf_arm64 report -v  --stdio -s socket,cpu
  build id event received for [kernel.kallsyms]: 
d287ff3393fb1a01d3a785c3a1dac6e63d973bce
  # To display the perf.data header info, please use 
--header/--header-only options.
  #
  symsrc__init: cannot get elf header.
  Looking at the vmlinux_path (7 entries long)
  Failed to open /proc/kcore. Note /proc/kcore requires CAP_SYS_RAWIO 
capability to access.
  Using /proc/kallsyms for symbols
  Failed to open /sbin/adbd, continuing without symbols
  #
  # Total Lost Samples: 0
  #
  # Samples: 304  of event 'cycles'
  # Event count (approx.): 42027298
  #
  # Overhead  Socket  CPU
  # ........  ......  ...
  #
      60.25%    -001  004
      32.69%    -001  007
       5.00%    -001  005
       1.34%    -001  003
       0.37%    -001  000
       0.20%    -001  006
       0.08%    -001  002
       0.07%    -001  001

And if CPU is offlined during perf record (between build_cpu_topology() 
and perf_env__read_cpu_topology_map()):

  # ./perf report -v  --stdio -s socket,cpu
  build id event received for [kernel.kallsyms]: ...
  core_id number is too big.You may need to upgrade the perf tool. <--- 
*please see this line*
  # To display the perf.data header info, please use 
--header/--header-only options.
  #
  symsrc__init: cannot get elf header.
  Looking at the vmlinux_path (7 entries long)
  Failed to open /proc/kcore. Note /proc/kcore requires CAP_SYS_RAWIO 
capability to access.
  Using /proc/kallsyms for symbols
  #
  # Total Lost Samples: 0
  #
  # Samples: 278  of event 'cycles'
  # Event count (approx.): 37729429
  #
  # Overhead  Socket  CPU
  # ........  ......  ...
  #
      59.06%    -001  004
      36.93%    -001  007
       1.33%    -001  000
       1.18%    -001  005
       0.82%    -001  006
       0.50%    -001  001
       0.08%    -001  002
       0.08%    -001  003