linux.git/tools/perf/util/stat-display.c, branch for-next

perf stat: Add metric-threshold to json output

2024-10-17T19:44:26+00:00

When the threshold isn't unknown add a value to the json like:
"metric-threshold" : "good"

A more complete example:
```
$ perf stat -a -j -I 1000
{"interval" : 1.001089747, "counter-value" : "16045.281449", "unit" : "msec", "event" : "cpu-clock", "event-runtime" : 16045355135, "pcnt-running" : 100.00, "metric-value" : "16.045281", "metric-unit" : "CPUs utilized"}
{"interval" : 1.001089747, "counter-value" : "10003.000000", "unit" : "", "event" : "context-switches", "event-runtime" : 16045314844, "pcnt-running" : 100.00, "metric-value" : "623.423156", "metric-unit" : "/sec"}
{"interval" : 1.001089747, "counter-value" : "328.000000", "unit" : "", "event" : "cpu-migrations", "event-runtime" : 16045321403, "pcnt-running" : 100.00, "metric-value" : "20.442147", "metric-unit" : "/sec"}
{"interval" : 1.001089747, "counter-value" : "20114.000000", "unit" : "", "event" : "page-faults", "event-runtime" : 16045355927, "pcnt-running" : 100.00, "metric-value" : "1.253577", "metric-unit" : "K/sec"}
{"interval" : 1.001089747, "counter-value" : "4066679471.000000", "unit" : "", "event" : "instructions", "event-runtime" : 16045369123, "pcnt-running" : 100.00, "metric-value" : "1.628330", "metric-unit" : "insn per cycle"}
{"interval" : 1.001089747, "counter-value" : "2497454658.000000", "unit" : "", "event" : "cycles", "event-runtime" : 16045374810, "pcnt-running" : 100.00, "metric-value" : "0.155650", "metric-unit" : "GHz"}
{"interval" : 1.001089747, "counter-value" : "914974294.000000", "unit" : "", "event" : "branches", "event-runtime" : 16045379877, "pcnt-running" : 100.00, "metric-value" : "57.024509", "metric-unit" : "M/sec"}
{"interval" : 1.001089747, "counter-value" : "9237201.000000", "unit" : "", "event" : "branch-misses", "event-runtime" : 16045375017, "pcnt-running" : 100.00, "metric-value" : "1.009559", "metric-unit" : "of all branches", "metric-threshold" : "good"}
{"interval" : 1.001089747, "event-runtime" : 16045397172, "pcnt-running" : 100.00, "metricgroup" : "TopdownL1"}
{"interval" : 1.001089747, "metric-value" : "22.036686", "metric-unit" : "%  tma_backend_bound", "metric-threshold" : "bad"}
{"interval" : 1.001089747, "metric-value" : "7.610161", "metric-unit" : "%  tma_bad_speculation", "metric-threshold" : "good"}
{"interval" : 1.001089747, "metric-value" : "36.729687", "metric-unit" : "%  tma_frontend_bound", "metric-threshold" : "bad"}
{"interval" : 1.001089747, "metric-value" : "33.623465", "metric-unit" : "%  tma_retiring"}
...
```

Signed-off-by: Ian Rogers 
Cc: Yicong Yang 
Cc: Weilin Wang 
Cc: Will Deacon 
Cc: James Clark 
Cc: Mike Leach 
Cc: Leo Yan 
Cc: Sumanth Korikkar 
Cc: Thomas Richter 
Cc: Tim Chen 
Cc: John Garry 
Link: https://lore.kernel.org/r/20241017175356.783793-7-irogers@google.com
Signed-off-by: Namhyung Kim

perf stat: Change color to threshold in print_metric

2024-10-17T19:44:26+00:00

Colors don't mean things in CSV and JSON output, switch to a threshold
enum value that the standard output can convert to a color. Updating
the CSV and JSON output will be later changes.

Signed-off-by: Ian Rogers 
Acked-by: Namhyung Kim 
Cc: Yicong Yang 
Cc: Weilin Wang 
Cc: Will Deacon 
Cc: James Clark 
Cc: Mike Leach 
Cc: Leo Yan 
Cc: Sumanth Korikkar 
Cc: Thomas Richter 
Cc: Tim Chen 
Cc: John Garry 
Link: https://lore.kernel.org/r/20241017175356.783793-6-irogers@google.com
Signed-off-by: Namhyung Kim

perf stat: Drop metric-unit if unit is NULL

2024-10-17T19:44:26+00:00

Avoid cases like:
```
$ perf stat -a -M topdownl1 -j -I 1000
...
{"interval" : 11.127757275, "counter-value" : "85715898.000000", "unit" : "", "event" : "IDQ.MITE_UOPS", "event-runtime" : 988376123, "pcnt-running" : 100.00, "metric-value" : "0.000000", "metric-unit" : "(null)"}
...
```

If there is no unit then drop the metric-value too as:
Suggested-by: Tim Chen 

Signed-off-by: Ian Rogers 
Cc: Yicong Yang 
Cc: Weilin Wang 
Cc: Will Deacon 
Cc: James Clark 
Cc: Mike Leach 
Cc: Leo Yan 
Cc: Sumanth Korikkar 
Cc: Thomas Richter 
Cc: Tim Chen 
Cc: John Garry 
Link: https://lore.kernel.org/r/20241017175356.783793-5-irogers@google.com
Signed-off-by: Namhyung Kim

perf stat: Display "none" for NaN with metric only json

2024-10-17T19:44:26+00:00

Return earlier for an empty unit case. If snprintf of the fmt doesn't
produce digits between vals and ends, as happens with NaN, make the
value "none" as happens in print_metric_end.

Signed-off-by: Ian Rogers 
Acked-by: Namhyung Kim 
Cc: Yicong Yang 
Cc: Weilin Wang 
Cc: Will Deacon 
Cc: James Clark 
Cc: Mike Leach 
Cc: Leo Yan 
Cc: Sumanth Korikkar 
Cc: Thomas Richter 
Cc: Tim Chen 
Cc: John Garry 
Link: https://lore.kernel.org/r/20241017175356.783793-4-irogers@google.com
Signed-off-by: Namhyung Kim

perf tool_pmu: Move expr literals to tool_pmu

2024-10-11T06:40:32+00:00

Add the expr literals like "#smt_on" as tool events, this allows stat
events to give the values. On my laptop with hyperthreading enabled:

```
$ perf stat -e "has_pmem,num_cores,num_cpus,num_cpus_online,num_dies,num_packages,smt_on,system_tsc_freq" true

 Performance counter stats for 'true':

                 0      has_pmem
                 8      num_cores
                16      num_cpus
                16      num_cpus_online
                 1      num_dies
                 1      num_packages
                 1      smt_on
     2,496,000,000      system_tsc_freq

       0.001113637 seconds time elapsed

       0.001218000 seconds user
       0.000000000 seconds sys
```

And with hyperthreading disabled:
```
$ perf stat -e "has_pmem,num_cores,num_cpus,num_cpus_online,num_dies,num_packages,smt_on,system_tsc_freq" true

 Performance counter stats for 'true':

                 0      has_pmem
                 8      num_cores
                16      num_cpus
                 8      num_cpus_online
                 1      num_dies
                 1      num_packages
                 0      smt_on
     2,496,000,000      system_tsc_freq

       0.000802115 seconds time elapsed

       0.000000000 seconds user
       0.000806000 seconds sys
```

As zero matters for these values, in stat-display
should_skip_zero_counter only skip the zero value if it is not the
first aggregation index.

The tool event implementations are used in expr but not evaluated as
events for simplicity. Also core_wide isn't made a tool event as it
requires command line parameters.

Signed-off-by: Ian Rogers 
Acked-by: Namhyung Kim 
Link: https://lore.kernel.org/r/20241002032016.333748-8-irogers@google.com
Signed-off-by: Namhyung Kim

perf tool_pmu: Factor tool events into their own PMU

2024-10-11T06:40:32+00:00

Rather than treat tool events as a special kind of event, create a
tool only PMU where the events/aliases match the existing
duration_time, user_time and system_time events. Remove special
parsing and printing support for the tool events, but add function
calls for when PMU functions are called on a tool_pmu.

Move the tool PMU code in evsel into tool_pmu.c to better encapsulate
the tool event behavior in that file.

Signed-off-by: Ian Rogers 
Acked-by: Namhyung Kim 
Link: https://lore.kernel.org/r/20241002032016.333748-5-irogers@google.com
Signed-off-by: Namhyung Kim

perf stat: Remove evlist__add_default_attrs use strings

2024-09-26T20:26:11+00:00

add_default_atttributes would add evsels by having pre-created
perf_event_attr, however, this needed fixing for hybrid as the
extended PMU type was necessary for each core PMU. The logic for this
was in an arch specific x86 function and wasn't present for ARM,
meaning that default events weren't being opened on all PMUs on
ARM. Change the creation of the default events to use parse_events and
strings as that will open the events on all PMUs.

Rather than try to detect events on PMUs before parsing, parse the
event but skip its output in stat-display.

The previous order of hardware events was: cycles,
stalled-cycles-frontend, stalled-cycles-backend, instructions. As
instructions is a more fundamental concept the order is changed to:
instructions, cycles, stalled-cycles-frontend, stalled-cycles-backend.

Closes: https://lore.kernel.org/lkml/CAP-5=fVABSBZnsmtRn1uF-k-G1GWM-L5SgiinhPTfHbQsKXb_g@mail.gmail.com/
Acked-by: Namhyung Kim 
Signed-off-by: Ian Rogers 
[Don't display unsupported default events except 'cycles']
Acked-by: Kan Liang 
Signed-off-by: James Clark 
Cc: Yang Jihong 
Cc: Dominique Martinet 
Cc: Colin Ian King 
Cc: Howard Chu 
Cc: Ze Gao 
Cc: Yicong Yang 
Cc: Weilin Wang 
Cc: Will Deacon 
Cc: Mike Leach 
Cc: Jing Zhang 
Cc: Yang Li 
Cc: Leo Yan 
Cc: ak@linux.intel.com
Cc: Athira Rajeev 
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong 
Cc: John Garry 
Link: https://lore.kernel.org/r/20240926144851.245903-4-james.clark@linaro.org
Signed-off-by: Namhyung Kim

perf stat: Uniquify event name improvements

2024-09-26T20:26:11+00:00

Without aggregation on Intel:
```
$ perf stat -e instructions,cycles ...
```
Will use "cycles" for the name of the legacy cycles event but as
"instructions" has a sysfs name it will and a "[cpu]" PMU suffix. This
often breaks things as the space between the event and the PMU name
look like an extra column. The existing uniquify logic was also
uniquifying in cases when all events are core and not with uncore
events, it was not correctly handling modifiers, etc.

Change the logic so that an initial pass that can disable
uniquification is run. For individual counters, disable uniquification
in more cases such as for consistency with legacy events or for
libpfm4 events. Don't use the "[pmu]" style suffix in uniquification,
always use "pmu/.../". Change how modifiers/terms are handled in the
uniquification so that they look like parse-able events.

This fixes "102: perf stat metrics (shadow stat) test:" that has been
failing due to "instructions [cpu]" breaking its column/awk logic when
values aren't aggregated. This started happening when instructions
could match a sysfs rather than a legacy event, so the fixes tag
reflects this.

Fixes: 617824a7f0f7 ("perf parse-events: Prefer sysfs/JSON hardware events over legacy")
Acked-by: Namhyung Kim 
Signed-off-by: Ian Rogers 
[ Fix Intel TPEBS counting mode test ]
Acked-by: Kan Liang 
Signed-off-by: James Clark 
Cc: Yang Jihong 
Cc: Dominique Martinet 
Cc: Colin Ian King 
Cc: Howard Chu 
Cc: Ze Gao 
Cc: Yicong Yang 
Cc: Weilin Wang 
Cc: Will Deacon 
Cc: Mike Leach 
Cc: Jing Zhang 
Cc: Yang Li 
Cc: Leo Yan 
Cc: ak@linux.intel.com
Cc: Athira Rajeev 
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong 
Cc: John Garry 
Link: https://lore.kernel.org/r/20240926144851.245903-3-james.clark@linaro.org
Signed-off-by: Namhyung Kim

perf stat: Display iostat headers correctly

2024-08-16T22:35:18+00:00

Currently we'll only print metric headers for metric leader in
aggregration mode. This will make `perf iostat` header not shown
since it'll aggregrated globally but don't have metric events:

  root@ubuntu204:/home/yang/linux/tools/perf# ./perf stat --iostat --timeout 1000
   Performance counter stats for 'system wide':
      port
  0000:00                    0                    0                    0                    0
  0000:80                    0                    0                    0                    0
  [...]

Fix this by excluding the iostat in the check of printing metric
headers. Then we can see the headers:

  root@ubuntu204:/home/yang/linux/tools/perf# ./perf stat --iostat --timeout 1000
   Performance counter stats for 'system wide':
      port             Inbound Read(MB)    Inbound Write(MB)    Outbound Read(MB)   Outbound Write(MB)
  0000:00                    0                    0                    0                    0
  0000:80                    0                    0                    0                    0
  [...]

Fixes: 193a9e30207f5477 ("perf stat: Don't display metric header for non-leader uncore events")
Signed-off-by: Yicong Yang 
Acked-by: Namhyung Kim 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jonathan Cameron 
Cc: Junhao He 
Cc: linuxarm@huawei.com
Cc: Mark Rutland 
Cc: Peter Zijlstra 
Cc: Shameerali Kolothum Thodi 
Cc: Zeng Tao 
Link: https://lore.kernel.org/r/20240802065800.48774-1-yangyicong@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Use field separator in the metric header

2024-06-28T17:58:09+00:00

It didn't use the passed field separator (using -x option) when it
prints the metric headers and always put "," between the fields.

Before:
  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:
  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:

Tested-by: Arnaldo Carvalho de Melo 
Link: https://lore.kernel.org/r/20240628000604.1296808-2-namhyung@kernel.org
Signed-off-by: Namhyung Kim