kernel/linux.git/tools/perf/util/stat-display.c, branch v7.2-rc5

perf stat: Fix false NMI watchdog warning in aggregation modes

2026-06-12T00:21:37+00:00

In aggregation modes (e.g. --per-socket, --per-die, etc.), a counter might not be scheduled or counted on specific aggregate groups if it was not assigned to the CPUs belonging to those groups. However, the printout() check triggers the "print_free_counters_hint" logic unconditionally for any supported counter with a missing count. This results in a false "Some events weren't counted. Try disabling the NMI watchdog" warning. Furthermore, the NMI watchdog only reserves performance counters on core PMUs. Uncore PMU events (e.g. CHA, IMC) are not affected by the NMI watchdog, but their failures also falsely triggered this warning. This warning was originally introduced in commit 02d492e5dcb72c00 ("perf stat: Issue a HW watchdog disable hint") To fix this, restrict setting of print_free_counters_hint to only trigger for core PMU events by checking counter->pmu and counter->pmu->is_core. Example before/after: $ perf stat -M lpm_miss_lat --metric-only --per-socket -a -- sleep 1 Before: Performance counter stats for 'system wide': ns lpm_miss_lat_rem ns lpm_miss_lat_loc S0 126 202.3 207.9 S1 126 231.9 259.3 1.006029831 seconds time elapsed Some events weren't counted. Try disabling the NMI watchdog: echo 0 > /proc/sys/kernel/nmi_watchdog perf stat ... echo 1 > /proc/sys/kernel/nmi_watchdog After: Performance counter stats for 'system wide': ns lpm_miss_lat_rem ns lpm_miss_lat_loc S0 126 202.3 207.9 S1 126 231.9 259.3 1.006029831 seconds time elapsed Assisted-by: Gemini:gemini-next Reviewed-by: Ian Rogers Signed-off-by: Chun-Tse Shao Cc: Namhyung Kim Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Make metric only column line up with header

2026-05-22T20:25:02+00:00

Since some time the metric-only output columns are messed up and do not line up with the header, which makes it hard to read. I haven't bisected it, but presumably it was broken for some time. There were multiple problems: - The dummy pm invocation at the beginning did print a bogus field - The column computation in pm did not agree with the header length - The color escape strings from highlighting confuse printf's field length computation Fix all those. I simplified the column width computation significantly, ignoring EVNAME_LEN, MGROUP_LEN, config->unit_width which don't seem to be useful in the metric only context. It now only uses the actual unit width as well as config->metric_only_len. The result is more code removed than added. Before: % perf stat --topdown -a -I 1000 + time % tma_backend_bound % tma_frontend_bound % tma_bad_speculation % tma_retiring 1.000190386 45.5 40.0 5.3 9.2 2.005185654 45.3 40.1 5.6 9.0 3.009193207 45.4 39.9 5.6 9.1 After: % perf stat --topdown -a -I 1000 + time % tma_backend_bound % tma_frontend_bound % tma_bad_speculation % tma_retiring 1.000810024 46.3 39.7 5.3 8.7 2.004800656 45.8 39.8 5.4 8.9 3.008804783 46.0 39.6 5.4 9.0 Reviewed-by: Ian Rogers Signed-off-by: Andi Kleen Cc: Namhyung Kim Signed-off-by: Arnaldo Carvalho de Melo

libperf cpumap: Make index and nr types unsigned

2026-04-01T21:50:53+00:00

The index into the cpumap array and the number of entries within the array can never be negative, so let's make them unsigned. This is prompted by reports that gcc 13 with -O6 is giving a alloc-size-larger-than errors. The change makes the cpumap changes and then updates the declaration of index variables throughout perf and libperf to be unsigned. The two things are hard to separate as compiler warnings about mixing signed and unsigned types breaks the build. Reported-by: Chingbin Li Closes: https://lore.kernel.org/lkml/20260212025127.841090-1-liqb365@163.com/ Tested-by: Chingbin Li Signed-off-by: Ian Rogers Signed-off-by: Namhyung Kim

perf stat: Ensure metrics are displayed even with failed events

2026-02-06T21:18:32+00:00

Currently, `perf stat` skips or hides metrics when the underlying hardware events cannot be counted (e.g., due to insufficient permissions or unsupported events). In `--metric-only` mode, this often results in missing columns or blank spaces, making the output difficult to parse. Modify the logic to ensure metrics are consistently displayed by propagating NAN (Not a Number) through the expression evaluator. Specifically: 1. Update `prepare_metric()` in stat-shadow.c to treat uncounted events (where `run == 0`) as NAN. This leverages the existing math in expr.y to propagate NAN through metric expressions. 2. Remove the early return in the display logic's `printout()` function that was previously skipping metrics in `--metric-only` mode for failed events. l 3. Simplify `perf_stat__skip_metric_event()` to no longer depend on event runtime. Tested: 1. `perf all metrics test` did not crash while paranoid is 2. 2. Multiple combinations with `CPUs_utilized` while paranoid is 2. $ ./perf stat -M CPUs_utilized -a -- sleep 1 Performance counter stats for 'system wide': msec cpu-clock:u # nan CPUs CPUs_utilized 1,006,356,120 duration_time 1.004375550 seconds time elapsed $ ./perf stat -M CPUs_utilized -a -j -- sleep 1 {"counter-value" : "", "unit" : "msec", "event" : "cpu-clock:u", "event-runtime" : 0, "pcnt-running" : 100.00, "metric-value" : "nan", "metric-unit" : "CPUs CPUs_utilized"} {"counter-value" : "1006642462.000000", "unit" : "", "event" : "duration_time", "event-runtime" : 1, "pcnt-running" : 100.00} $ ./perf stat -M CPUs_utilized -a --metric-only -- sleep 1 Performance counter stats for 'system wide': CPUs CPUs_utilized nan 1.004424652 seconds time elapsed $ ./perf stat -M CPUs_utilized -a --metric-only -j -- sleep 1 {"CPUs CPUs_utilized" : "none"} Reviewed-by: Ian Rogers Signed-off-by: Chun-Tse Shao Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Ingo Molnar Cc: James Clark Cc: Jiri Olsa Cc: Kan Liang Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Yang Li Signed-off-by: Arnaldo Carvalho de Melo

perf stat display: Make %f precision consistent

2026-01-12T18:55:06+00:00

Commit bc22de9bcdb22491 ("perf stat: Display time in precision based on std deviation") added multirun workload elapsed time. There was an effort to make the precision in the output most useful for the user, however, when gathering over runs it means the formatting varies. This change just makes the output format fixed. Before: ``` $ while :; do perf stat --null --repeat 3 sleep 0.1 2>&1 | grep elapsed; done 0.101140 +- 0.000149 seconds time elapsed ( +- 0.15% ) 0.1011396 +- 0.0000218 seconds time elapsed ( +- 0.02% ) 0.101331 +- 0.000124 seconds time elapsed ( +- 0.12% ) ^C $ while :; do perf stat --null --repeat 3 sleep 1 2>&1 | grep elapsed; done 1.001317 +- 0.000146 seconds time elapsed ( +- 0.01% ) 1.001377 +- 0.000172 seconds time elapsed ( +- 0.02% ) 1.00253 +- 0.00131 seconds time elapsed ( +- 0.13% ) ``` After: ``` $ while :; do perf stat --null --repeat 3 sleep 0.1 2>&1 | grep elapsed; done 0.101406408 +- 0.000064778 seconds time elapsed ( +- 0.06% ) 0.101367315 +- 0.000027253 seconds time elapsed ( +- 0.03% ) 0.101434164 +- 0.000084750 seconds time elapsed ( +- 0.08% ) ^C $ while :; do perf stat --null --repeat 3 sleep 1 2>&1 | grep elapsed; done 1.001525467 +- 0.000051703 seconds time elapsed ( +- 0.01% ) 1.001375093 +- 0.000116200 seconds time elapsed ( +- 0.01% ) 1.001141025 +- 0.000046361 seconds time elapsed ( +- 0.00% ) ``` Closes: https://lore.kernel.org/lkml/aTQRgAOpKyI53TEq@gmail.com/ Suggested-by: Ingo Molnar Signed-off-by: Ian Rogers Tested-by: Ingo Molnar Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Chun-Tse Shao Cc: James Clark Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Display metric-only for 0 counters

2025-11-18T01:31:59+00:00

0 counters may occur in hypervisor settings but metric-only output is always expected. This resolves an issue in the "perf stat STD output linter" test. Signed-off-by: Ian Rogers Tested-by: Thomas Falcon Signed-off-by: Namhyung Kim

perf stat: Align metric output without events

2025-11-12T01:00:31+00:00

One of my concern in the perf stat output was the alignment in the metrics and shadow stats. I think it missed to calculate the basic output length using COUNTS_LEN and EVNAME_LEN but missed to add the unit length like "msec" and surround 2 spaces. I'm not sure why it's not printed below though. But anyway, now it shows correctly aligned metric output. $ perf stat true Performance counter stats for 'true': 859,772 task-clock # 0.395 CPUs utilized 0 context-switches # 0.000 /sec 0 cpu-migrations # 0.000 /sec 56 page-faults # 65.134 K/sec 1,075,022 instructions # 0.86 insn per cycle 1,255,911 cycles # 1.461 GHz 220,573 branches # 256.548 M/sec 7,381 branch-misses # 3.35% of all branches TopdownL1 # 19.2 % tma_retiring # 28.6 % tma_backend_bound # 9.5 % tma_bad_speculation # 42.6 % tma_frontend_bound 0.002174871 seconds time elapsed ^ | 0.002154000 seconds user | 0.000000000 seconds sys here Reviewed-by: Ian Rogers Signed-off-by: Namhyung Kim

perf stat: Remove "unit" workarounds for metric-only

2025-11-12T00:48:35+00:00

Remove code that tested the "unit" as in KB/sec for certain hard coded metric values and did workarounds. Signed-off-by: Ian Rogers Signed-off-by: Namhyung Kim

perf stat: Remove hard coded shadow metrics

2025-11-12T00:48:35+00:00

Now that the metrics are encoded in common json the hard coded printing means the metrics are shown twice. Remove the hard coded version. This means that when specifying events, and those events correspond to a hard coded metric, the metric will no longer be displayed. The metric will be displayed if the metric is requested. Due to the adhoc printing in the previous approach it was often found frustrating, the new approach avoids this. The default perf stat output on an alderlake now looks like: ``` $ perf stat -a -- sleep 1 Performance counter stats for 'system wide': 19,697 context-switches # nan cs/sec cs_per_second TopdownL1 (cpu_core) # 10.7 % tma_bad_speculation # 24.9 % tma_frontend_bound TopdownL1 (cpu_core) # 34.3 % tma_backend_bound # 30.1 % tma_retiring 6,593 page-faults # nan faults/sec page_faults_per_second 729,065,658 cpu_atom/cpu-cycles/ # nan GHz cycles_frequency (49.79%) 1,605,131,101 cpu_core/cpu-cycles/ # nan GHz cycles_frequency # 19.7 % tma_bad_speculation # 14.2 % tma_retiring (50.14%) # 37.3 % tma_frontend_bound (50.31%) 87,302,268 cpu_atom/branches/ # nan M/sec branch_frequency (60.27%) 512,046,956 cpu_core/branches/ # nan M/sec branch_frequency 1,111 cpu-migrations # nan migrations/sec migrations_per_second # 28.8 % tma_backend_bound (60.26%) 0.00 msec cpu-clock # 0.0 CPUs CPUs_utilized 392,509,323 cpu_atom/instructions/ # 0.6 instructions insn_per_cycle (60.19%) 2,990,369,310 cpu_core/instructions/ # 1.9 instructions insn_per_cycle 3,493,478 cpu_atom/branch-misses/ # 5.9 % branch_miss_rate (49.69%) 7,297,531 cpu_core/branch-misses/ # 1.4 % branch_miss_rate 1.006621701 seconds time elapsed ``` Signed-off-by: Ian Rogers Signed-off-by: Namhyung Kim

perf jevents: Add metric DefaultShowEvents

2025-11-12T00:48:35+00:00

Some Default group metrics require their events showing for consistency with perf's previous behavior. Add a flag to indicate when this is the case and use it in stat-display. As events are coming from Default metrics remove that default hardware and software events from perf stat. Following this change the default perf stat output on an alderlake looks like: ``` $ perf stat -a -- sleep 1 Performance counter stats for 'system wide': 20,550 context-switches # nan cs/sec cs_per_second TopdownL1 (cpu_core) # 9.0 % tma_bad_speculation # 28.1 % tma_frontend_bound TopdownL1 (cpu_core) # 29.2 % tma_backend_bound # 33.7 % tma_retiring 6,685 page-faults # nan faults/sec page_faults_per_second 790,091,064 cpu_atom/cpu-cycles/ # nan GHz cycles_frequency (49.83%) 2,563,918,366 cpu_core/cpu-cycles/ # nan GHz cycles_frequency # 12.3 % tma_bad_speculation # 14.5 % tma_retiring (50.20%) # 33.8 % tma_frontend_bound (50.24%) 76,390,322 cpu_atom/branches/ # nan M/sec branch_frequency (60.20%) 1,015,173,047 cpu_core/branches/ # nan M/sec branch_frequency 1,325 cpu-migrations # nan migrations/sec migrations_per_second # 39.3 % tma_backend_bound (60.17%) 0.00 msec cpu-clock # 0.000 CPUs utilized # 0.0 CPUs CPUs_utilized 554,347,072 cpu_atom/instructions/ # 0.64 insn per cycle # 0.6 instructions insn_per_cycle (60.14%) 5,228,931,991 cpu_core/instructions/ # 2.04 insn per cycle # 2.0 instructions insn_per_cycle 4,308,874 cpu_atom/branch-misses/ # 5.65% of all branches # 5.6 % branch_miss_rate (49.76%) 9,890,606 cpu_core/branch-misses/ # 0.97% of all branches # 1.0 % branch_miss_rate 1.005477803 seconds time elapsed ``` Signed-off-by: Ian Rogers Signed-off-by: Namhyung Kim