linux-stable.git/tools/perf/util/stat-shadow.c, branch linux-5.7.y

perf stat: Fix no metric header if --per-socket and --metric-only set

2020-04-14T11:49:26+00:00

We received a report that was no metric header displayed if --per-socket
and --metric-only were both set.

It's hard for script to parse the perf-stat output. This patch fixes this
issue.

Before:

  root@kbl-ppc:~# perf stat -a -M CPI --metric-only --per-socket
  ^C
   Performance counter stats for 'system wide':

  S0        8                  2.6

         2.215270071 seconds time elapsed

  root@kbl-ppc:~# perf stat -a -M CPI --metric-only --per-socket -I1000
  #           time socket cpus
       1.000411692 S0        8                  2.2
       2.001547952 S0        8                  3.4
       3.002446511 S0        8                  3.4
       4.003346157 S0        8                  4.0
       5.004245736 S0        8                  0.3

After:

  root@kbl-ppc:~# perf stat -a -M CPI --metric-only --per-socket
  ^C
   Performance counter stats for 'system wide':

                               CPI
  S0        8                  2.1

         1.813579830 seconds time elapsed

  root@kbl-ppc:~# perf stat -a -M CPI --metric-only --per-socket -I1000
  #           time socket cpus                  CPI
       1.000415122 S0        8                  3.2
       2.001630051 S0        8                  2.9
       3.002612278 S0        8                  4.3
       4.003523594 S0        8                  3.0
       5.004504256 S0        8                  3.7

Signed-off-by: Jin Yao 
Acked-by: Jiri Olsa 
Cc: Alexander Shishkin 
Cc: Andi Kleen 
Cc: Kan Liang 
Cc: Peter Zijlstra 
Link: http://lore.kernel.org/lkml/20200331180226.25915-1-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf expr: Straighten exprparse()/exprfind_other() interface

2020-03-10T00:43:24+00:00

Now that we have a flex parser we don't need to update the parsed string
pointer, so the interface can just be passed the pointer to the
expression instead of a pointer to pointer.

Signed-off-by: Jiri Olsa 
Reviewed-by: Andi Kleen 
Cc: Alexander Shishkin 
Cc: John Garry 
Cc: Kajol Jain 
Cc: Michael Petlan 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Ravi Bangoria 
Link: http://lore.kernel.org/lkml/20200228093616.67125-5-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Don't report a null stalled cycles per insn metric

2020-02-10T19:30:09+00:00

For data collected on machines with front end stalled cycles supported,
such as found on modern AMD CPU families, commit 146540fb545b ("perf
stat: Always separate stalled cycles per insn") introduces a new line in
CSV output with a leading comma that upsets some automated scripts.
Scripts have to use "-e ex_ret_instr" to work around this issue, after
upgrading to a version of perf with that commit.

We could add "if (have_frontend_stalled && !config->csv_sep)" to the not
(total && avg) else clause, to emphasize that CSV users are usually
scripts, and are written to do only what is needed, i.e., they wouldn't
typically invoke "perf stat" without specifying an explicit event list.

But - let alone CSV output - why should users now tolerate a constant
0-reporting extra line in regular terminal output?:

BEFORE:

$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1

 Performance counter stats for 'system wide':

       181,110,981      instructions              #    0.58  insn per cycle
                                                  #    0.00  stalled cycles per insn
       309,876,469      cycles

       1.002202582 seconds time elapsed

The user would not like to see the now permanent:

  "0.00  stalled cycles per insn"

line fixture, as it gives no useful information.

So this patch removes the printing of the zeroed stalled cycles line
altogether, almost reverting the very original commit fb4605ba47e7
("perf stat: Check for frontend stalled for metrics"), which seems like
it was written to normalize --metric-only column output of common Intel
machines at the time: modern Intel machines have ceased to support the
genericised frontend stalled metrics AFAICT.

AFTER:

$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1

 Performance counter stats for 'system wide':

       244,071,432      instructions              #    0.69  insn per cycle
       355,353,490      cycles

       1.001862516 seconds time elapsed

Output behaviour when stalled cycles is indeed measured is not affected
(BEFORE == AFTER):

$ sudo perf stat --all-cpus -einstructions,cycles,stalled-cycles-frontend -- sleep 1

 Performance counter stats for 'system wide':

       247,227,799      instructions              #    0.63  insn per cycle
                                                  #    0.26  stalled cycles per insn
       394,745,636      cycles
        63,194,485      stalled-cycles-frontend   #   16.01% frontend cycles idle

       1.002079770 seconds time elapsed

Fixes: 146540fb545b ("perf stat: Always separate stalled cycles per insn")
Signed-off-by: Kim Phillips 
Acked-by: Andi Kleen 
Acked-by: Jiri Olsa 
Acked-by: Song Liu 
Cc: Alexander Shishkin 
Cc: Cong Wang 
Cc: Davidlohr Bueso 
Cc: Jin Yao 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: http://lore.kernel.org/lkml/20200207230613.26709-1-kim.phillips@amd.com
Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Fix free memory access / memory leaks in metrics

2019-09-25T19:26:40+00:00

Make sure to not free the name passed in by the caller, but free all the
allocated ids when parsing expressions.

The loop at the end knows that the first entry shouldn't be freed, so
make sure the caller name is the first entry.

Fixes

  % perf stat -M IpB,IpCall,IpTB,IPC,Retiring_SMT,Frontend_Bound_SMT,Kernel_Utilization,CPU_Utilization --metric-only -a -I 1000 sleep 2

  valgrind:
       1.009943231 ==21527== Invalid read of size 1
  ==21527==    at 0x483CB74: strcmp (vg_replace_strmem.c:849)
  ==21527==    by 0x582CF8: collect_all_aliases (stat-display.c:554)
  ==21527==    by 0x582EB3: collect_data (stat-display.c:577)
  ==21527==    by 0x583A32: print_counter_aggr (stat-display.c:806)
  ==21527==    by 0x584FAD: perf_evlist__print_counters (stat-display.c:1200)
  ==21527==    by 0x45133A: print_counters (builtin-stat.c:655)
  ==21527==    by 0x450629: process_interval (builtin-stat.c:353)
  ==21527==    by 0x450FBD: __run_perf_stat (builtin-stat.c:564)
  ==21527==    by 0x451285: run_perf_stat (builtin-stat.c:636)
  ==21527==    by 0x454619: cmd_stat (builtin-stat.c:1966)
  ==21527==    by 0x4D557D: run_builtin (perf.c:310)
  ==21527==    by 0x4D57EA: handle_internal_command (perf.c:362)
  ==21527==  Address 0x12826cd0 is 0 bytes inside a block of size 25 free'd
  ==21527==    at 0x4839A0C: free (vg_replace_malloc.c:540)
  ==21527==    by 0x627041: __zfree (zalloc.c:13)
  ==21527==    by 0x57F66A: generic_metric (stat-shadow.c:814)
  ==21527==    by 0x580B21: perf_stat__print_shadow_stats (stat-shadow.c:1057)
  ==21527==    by 0x58418E: print_metric_headers (stat-display.c:943)
  ==21527==    by 0x5844BC: print_interval (stat-display.c:1004)
  ==21527==    by 0x584DEB: perf_evlist__print_counters (stat-display.c:1172)
  ==21527==    by 0x45133A: print_counters (builtin-stat.c:655)
  ==21527==    by 0x450629: process_interval (builtin-stat.c:353)
  ==21527==    by 0x450FBD: __run_perf_stat (builtin-stat.c:564)
  ==21527==    by 0x451285: run_perf_stat (builtin-stat.c:636)
  ==21527==    by 0x454619: cmd_stat (builtin-stat.c:1966)
  ==21527==  Block was alloc'd at
  ==21527==    at 0x483880B: malloc (vg_replace_malloc.c:309)
  ==21527==    by 0x51677DE: strdup (in /usr/lib64/libc-2.29.so)
  ==21527==    by 0x506457: parse_events_name (parse-events.c:1754)
  ==21527==    by 0x5550BB: parse_events_parse (parse-events.y:214)
  ==21527==    by 0x50694D: parse_events__scanner (parse-events.c:1887)
  ==21527==    by 0x506AEF: parse_events (parse-events.c:1927)
  ==21527==    by 0x521D8B: metricgroup__parse_groups (metricgroup.c:527)
  ==21527==    by 0x45156F: parse_metric_groups (builtin-stat.c:721)
  ==21527==    by 0x6228A9: get_value (parse-options.c:243)
  ==21527==    by 0x62363F: parse_short_opt (parse-options.c:348)
  ==21527==    by 0x62363F: parse_options_step (parse-options.c:536)
  ==21527==    by 0x62363F: parse_options_subcommand (parse-options.c:651)
  ==21527==    by 0x453C1D: cmd_stat (builtin-stat.c:1718)
  ==21527==    by 0x4D557D: run_builtin (perf.c:310)

and also a leak report.

Committer testing:

Before:

  # perf stat -M IpB,IpCall,IpTB,IPC,Retiring_SMT,Frontend_Bound_SMT,Kernel_Utilization,CPU_Utilization --metric-only -a -I 1000 sleep 2
  #           time      CPU_Utilization
       1.000470810                      free(): double free detected in tcache 2
  Aborted (core dumped)
  #

After:

  # perf stat -M IpB,IpCall,IpTB,IPC,Retiring_SMT,Frontend_Bound_SMT,Kernel_Utilization,CPU_Utilization --metric-only -a -I 1000 sleep 2
  #           time      CPU_Utilization
       1.000494752                  0.1
       2.001105112                  0.1
  #

Signed-off-by: Andi Kleen 
Acked-by: Jiri Olsa 
Tested-by: Arnaldo Carvalho de Melo 
Link: http://lore.kernel.org/lkml/20190923233339.25326-3-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo

perf metricgroup: Support multiple events for metricgroup

2019-09-01T01:27:52+00:00

Some uncore metrics don't work as expected. For example, on
cascadelakex:

  root@lkp-csl-2sp2:~# perf stat -M UNC_M_PMM_BANDWIDTH.TOTAL -a -- sleep 1

   Performance counter stats for 'system wide':

           1841092      unc_m_pmm_rpq_inserts
           3680816      unc_m_pmm_wpq_inserts

       1.001775055 seconds time elapsed

  root@lkp-csl-2sp2:~# perf stat -M UNC_M_PMM_READ_LATENCY -a -- sleep 1

   Performance counter stats for 'system wide':

         860649746      unc_m_pmm_rpq_occupancy.all
           1840557      unc_m_pmm_rpq_inserts
       12790627455      unc_m_clockticks

       1.001773348 seconds time elapsed

No metrics 'UNC_M_PMM_BANDWIDTH.TOTAL' or 'UNC_M_PMM_READ_LATENCY' are
reported.

The issue is, the case of an alias expanding to mulitple events is not
supported, typically the uncore events.  (see comments in
find_evsel_group()).

For UNC_M_PMM_BANDWIDTH.TOTAL in above example, the expanded event group
is '{unc_m_pmm_rpq_inserts,unc_m_pmm_wpq_inserts}:W', but the actual
events passed to find_evsel_group are:

  unc_m_pmm_rpq_inserts
  unc_m_pmm_rpq_inserts
  unc_m_pmm_rpq_inserts
  unc_m_pmm_rpq_inserts
  unc_m_pmm_rpq_inserts
  unc_m_pmm_rpq_inserts
  unc_m_pmm_wpq_inserts
  unc_m_pmm_wpq_inserts
  unc_m_pmm_wpq_inserts
  unc_m_pmm_wpq_inserts
  unc_m_pmm_wpq_inserts
  unc_m_pmm_wpq_inserts

For this multiple events case, it's not supported well.

This patch introduces a new field 'metric_leader' in struct evsel. The
first event is considered as a metric leader. For the rest of same
events, they point to the first event via it's metric_leader field in
struct evsel.

This design is for adding the counting results of all same events to the
first event in group (the metric_leader).

With this patch,

  root@lkp-csl-2sp2:~# perf stat -M UNC_M_PMM_BANDWIDTH.TOTAL -a -- sleep 1

   Performance counter stats for 'system wide':

           1842108      unc_m_pmm_rpq_inserts     #    337.2 MB/sec  UNC_M_PMM_BANDWIDTH.TOTAL
           3682209      unc_m_pmm_wpq_inserts

       1.001819706 seconds time elapsed

  root@lkp-csl-2sp2:~# perf stat -M UNC_M_PMM_READ_LATENCY -a -- sleep 1

   Performance counter stats for 'system wide':

         861970685      unc_m_pmm_rpq_occupancy.all #    219.4 ns  UNC_M_PMM_READ_LATENCY
           1842772      unc_m_pmm_rpq_inserts
       12790196356      unc_m_clockticks

       1.001749103 seconds time elapsed

Now we can see the correct metrics 'UNC_M_PMM_BANDWIDTH.TOTAL' and
'UNC_M_PMM_READ_LATENCY'.

Signed-off-by: Jin Yao 
Cc: Alexander Shishkin 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Peter Zijlstra 
Link: http://lore.kernel.org/lkml/20190828055932.8269-5-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf metricgroup: Scale the metric result

2019-09-01T01:27:52+00:00

Some metrics define the scale unit, such as

    {
        "BriefDescription": "Intel Optane DC persistent memory read latency (ns). Derived from unc_m_pmm_rpq_occupancy.all",
        "Counter": "0,1,2,3",
        "EventCode": "0xE0",
        "EventName": "UNC_M_PMM_READ_LATENCY",
        "MetricExpr": "UNC_M_PMM_RPQ_OCCUPANCY.ALL / UNC_M_PMM_RPQ_INSERTS / UNC_M_CLOCKTICKS",
        "MetricName": "UNC_M_PMM_READ_LATENCY",
        "PerPkg": "1",
        "ScaleUnit": "6000000000ns",
        "UMask": "0x1",
        "Unit": "iMC"
    },

For above example, the ratio should be,

ratio = (UNC_M_PMM_RPQ_OCCUPANCY.ALL / UNC_M_PMM_RPQ_INSERTS / UNC_M_CLOCKTICKS) * 6000000000

But in current code, the ratio is not scaled ( * 6000000000)

With this patch, the ratio is scaled and the unit (ns) is printed.

For example,
  #    219.4 ns  UNC_M_PMM_READ_LATENCY

Signed-off-by: Jin Yao 
Cc: Alexander Shishkin 
Cc: Andi Kleen 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Peter Zijlstra 
Link: http://lore.kernel.org/lkml/20190828055932.8269-4-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

libperf: Move perf_event_attr field from perf's evsel to libperf's perf_evsel

2019-07-29T21:34:45+00:00

Move the perf_event_attr struct fron 'struct evsel' to 'struct perf_evsel'.

Committer notes:

Fixed up these:

 tools/perf/arch/arm/util/auxtrace.c
 tools/perf/arch/arm/util/cs-etm.c
 tools/perf/arch/arm64/util/arm-spe.c
 tools/perf/arch/s390/util/auxtrace.c
 tools/perf/util/cs-etm.c

Also

  cc1: warnings being treated as errors
  tests/sample-parsing.c: In function 'do_test':
  tests/sample-parsing.c:162: error: missing initializer
  tests/sample-parsing.c:162: error: (near initialization for 'evsel.core.cpus')

   	struct evsel evsel = {
   		.needs_swap = false,
  -		.core.attr = {
  -			.sample_type = sample_type,
  -			.read_format = read_format,
  +		.core = {
  +			. attr = {
  +				.sample_type = sample_type,
  +				.read_format = read_format,
  +			},

  [perfbuilder@a70e4eeb5549 /]$ gcc --version |& head -1
  gcc (GCC) 4.4.7

Also we don't need to include perf_event.h in
tools/perf/lib/include/perf/evsel.h, forward declaring 'struct
perf_event_attr' is enough. And this even fixes the build in some
systems where things are used somewhere down the include path from
perf_event.h without defining __always_inline.

Signed-off-by: Jiri Olsa 
Cc: Alexander Shishkin 
Cc: Alexey Budankov 
Cc: Andi Kleen 
Cc: Michael Petlan 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/20190721112506.12306-43-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf evlist: Rename struct perf_evlist to struct evlist

2019-07-29T21:34:42+00:00

Rename struct perf_evlist to struct evlist, so we don't have a name
clash when we add struct perf_evlist in libperf.

Committer notes:

Added fixes to build on arm64, from Jiri and from me
(tools/perf/util/cs-etm.c)

Signed-off-by: Jiri Olsa 
Cc: Alexander Shishkin 
Cc: Alexey Budankov 
Cc: Andi Kleen 
Cc: Michael Petlan 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/20190721112506.12306-6-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf evsel: Rename struct perf_evsel to struct evsel