Leo Yan [Wed, 12 Nov 2025 18:24:36 +0000 (18:24 +0000)]
perf arm_spe: Report register access in record
Record register access info for load / store operations.
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 12 Nov 2025 18:24:35 +0000 (18:24 +0000)]
perf arm_spe: Introduce data processing macro for SVE operations
Introduce the ARM_SPE_OP_DP (data processing) macro as associated
information for SVE operations. For SVE register access, only
ARM_SPE_OP_SVE is set; for SVE data processing, both ARM_SPE_OP_SVE and
ARM_SPE_OP_DP are set together.
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 12 Nov 2025 18:24:34 +0000 (18:24 +0000)]
perf arm_spe: Consolidate operation types
Consolidate operation types in a way:
(a) Extract the second-level types into separate enums.
(b) The second-level types for memory and SIMD operations are classified
by modules. E.g., an operation may relate to general register,
SIMD/FP, SVE, etc.
(c) The associated information tells details. E.g., an operation is
load or store, whether it is atomic operation, etc.
Start the enum items for the second-level types from 8 to accommodate
more entries within a 32-bit integer.
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 12 Nov 2025 18:24:33 +0000 (18:24 +0000)]
perf arm_spe: Remove unused operation types
Remove unused SVE operation types. These operations will be reintroduced
in subsequent refactoring, but with a different format.
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 12 Nov 2025 18:24:31 +0000 (18:24 +0000)]
perf arm_spe: Decode ASE and FP fields in other operation
Add a check for other operation, which prevents any incorrectly
classifying. Parse the ASE and FP fields.
After:
. 0000002f: 48 06 OTHER ASE FP INSN-OTHER
. 00000031: b2 08 80 48 01 08 00 ff ff VA 0xffff000801488008
. 0000003a: 9a 00 00 LAT 0 XLAT
. 0000003d: 42 16 EV RETIRED L1D-ACCESS TLB-ACCESS
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Rename the macro to SPE_OP_PKT_OTHER_SUBCLASS_SVE to unify naming.
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 12 Nov 2025 18:24:28 +0000 (18:24 +0000)]
perf arm_spe: Unify operation naming
Rename extended subclass and SVE/SME register access subclass, so that
the naming can be consistent cross all sub classes.
Add an log "SVE-SME-REG" for the SVE/SME register access, this is easier
for parsing.
Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 12 Nov 2025 18:24:27 +0000 (18:24 +0000)]
perf arm_spe: Fix memset subclass in operation
The operation subclass is extracted from bits [7..1] of the payload.
Since bit [0] is not parsed, there is no chance to match the memset type
(0x25). As a result, the memset payload is never parsed successfully.
Instead of extracting a unified bit field, change to extract the
specific bits for each operation subclass.
Fixes: 34fb60400e32 ("perf arm-spe: Add raw decoding for SPEv1.3 MTE and MOPS load/store") Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 13 Nov 2025 18:05:13 +0000 (10:05 -0800)]
perf tool_pmu: More accurately set the cpus for tool events
The user and system time events can record on different CPUs, but for
all other events a single CPU map of just CPU 0 makes sense. In
parse-events detect a tool PMU and then pass the perf_event_attr so
that the tool_pmu can return CPUs specific for the event. This avoids
a CPU map of all online CPUs being used for events like
duration_time. Avoiding this avoids the evlist CPUs containing CPUs
for which duration_time just gives 0. Minimizing the evlist CPUs can
remove unnecessary sched_setaffinity syscalls that delay metric
calculations.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 13 Nov 2025 18:05:11 +0000 (10:05 -0800)]
perf stat: Reduce scope of ru_stats
The ru_stats are used to capture user and system time stats when a
process exits. These are then applied to user and system time tool
events if their reads fail due to the process terminating. Reduce the
scope now the metric code no longer reads these values.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 13 Nov 2025 18:05:10 +0000 (10:05 -0800)]
perf stat-shadow: Read tool events directly
When reading time values for metrics don't use the globals updated in
builtin-stat, just read the events as regular events. The only
exception is for time events where nanoseconds need converting to
seconds as metrics assume time metrics are in seconds.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 13 Nov 2025 18:05:09 +0000 (10:05 -0800)]
perf tool_pmu: Use old_count when computing count values for time events
When running in interval mode every third count of a time event isn't
showing properly:
```
$ perf stat -e duration_time -a -I 1000
1.001082862 1,002,290,425 duration_time
2.004264262 1,003,183,516 duration_time
3.007381401 <not counted> duration_time
4.011160141 1,003,705,631 duration_time
5.014515385 1,003,290,110 duration_time
6.018539680 <not counted> duration_time
7.022065321 1,003,591,720 duration_time
```
The regression came in with a different fix, found through bisection,
commit 68cb1567439f ("perf tool_pmu: Fix aggregation on
duration_time"). The issue is caused by the enabled and running time
of the event matching the old_count's and creating a delta of 0, which
is indicative of an error.
Fixes: 68cb1567439f ("perf tool_pmu: Fix aggregation on duration_time") Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 13 Nov 2025 18:05:07 +0000 (10:05 -0800)]
libperf cpumap: Reduce allocations and sorting in intersect
On hybrid platforms the CPU maps are often disjoint. Rather than copy
CPUs and trim, compute the number of common CPUs, if none early exit,
otherwise copy in an sorted order. This avoids memory allocation in
the disjoint case and avoids a second malloc and useless sort in the
previous trim cases.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Wed, 12 Nov 2025 19:53:11 +0000 (11:53 -0800)]
perf stat: Display metric-only for 0 counters
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 <irogers@google.com> Tested-by: Thomas Falcon <thomas.falcon@intel.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Wed, 12 Nov 2025 19:53:08 +0000 (11:53 -0800)]
perf test: Be tolerant of missing json metric none value
print_metric_only_json and print_metric_end in stat-display.c may
create a metric value of "none" which fails validation as isfloat. Add
a helper to properly validate metric numeric values.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
builtin-script.c:347:36: error: unused function 'evsel_script' [-Werror,-Wunused-function] Fixes: 3622990efaab ("perf script: Change metric format to use json metrics") Signed-off-by: James Clark <james.clark@linaro.org> Reviewed-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The instructions event is now provided in json meaning the has_event
test always succeeds. Switch to using non-legacy event names in the
affected metrics.
Reported-by: Thomas Richter <tmricht@linux.ibm.com> Closes: https://lore.kernel.org/linux-perf-users/3e80f453-f015-4f4f-93d3-8df6bb6b3c95@linux.ibm.com/ Fixes: 0012e0fa221b ("perf jevents: Add legacy-hardware and legacy-cache json") Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: Thomas Richter <tmricht@linux.ibm.com> Tested-by: Thomas Richter <tmricht@linux.ibm.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Mon, 10 Nov 2025 01:31:51 +0000 (17:31 -0800)]
perf build: Remove NO_AUXTRACE build option
The NO_AUXTRACE build option was used when the __get_cpuid feature
test failed or if it was provided on the command line. The option no
longer avoids a dependency on a library and so having the option is
just adding complexity to the code base. Remove the option
CONFIG_AUXTRACE from Build files and HAVE_AUXTRACE_SUPPORT by assuming
it is always defined.
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Mon, 10 Nov 2025 01:31:50 +0000 (17:31 -0800)]
tool build: Remove __get_cpuid feature test
This feature test is no longer used so remove.
The function tested by the feature test is used in:
tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c
however, the Makefile just assumes the presence of the function and
doesn't perform a build feature test for it.
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Mon, 10 Nov 2025 01:31:48 +0000 (17:31 -0800)]
perf intel-pt: Use the perf provided "cpuid.h"
Rather than having a feature test and include of <cpuid.h> for the
__get_cpuid function, use the cpuid function provided by
tools/perf/arch/x86/util/cpuid.h.
Signed-off-by: Ian Rogers <irogers@google.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Zide Chen [Wed, 12 Nov 2025 16:48:23 +0000 (08:48 -0800)]
perf test: Add a perf event fallback test
This adds test cases to verify the precise ip fallback logic:
- If the system supports precise ip, for an event given with the maximum
precision level, it should be able to decrease precise_ip to find a
supported level.
- The same fallback behavior should also work in more complex scenarios,
such as event groups or when PEBS is involved
Additional fallback tests, such as those covering missing feature cases,
can be added in the future.
Suggested-by: Ian Rogers <irogers@google.com> Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com> Signed-off-by: Zide Chen <zide.chen@intel.com> Reviewed-by: Ian Rogers <irogers!@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Namhyung Kim [Thu, 6 Nov 2025 07:28:34 +0000 (23:28 -0800)]
perf stat: Align metric output without events
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.
Ian Rogers [Tue, 11 Nov 2025 21:22:06 +0000 (13:22 -0800)]
perf tool_pmu: Make core_wide and target_cpu json events
For the sake of better documentation, add core_wide and target_cpu to
the tool.json. When the values of system_wide and
user_requested_cpu_list are unknown, use the values from the global
stat_config.
Example output showing how '-a' modifies the values in `perf stat`:
```
$ perf stat -e core_wide,target_cpu true
$ perf list
...
tool:
core_wide
[1 if not SMT,if SMT are events being gathered on all SMT threads 1 otherwise 0. Unit: tool]
...
target_cpu
[1 if CPUs being analyzed,0 if threads/processes. Unit: tool]
...
```
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 11 Nov 2025 21:22:05 +0000 (13:22 -0800)]
perf test stat csv: Update test expectations and events
Explicitly use a metric rather than implicitly expecting '-e
instructions,cycles' to produce a metric. Use a metric with software
events to make it more compatible.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 11 Nov 2025 21:22:04 +0000 (13:22 -0800)]
perf test stat: Update test expectations and events
test_stat_record_report and test_stat_record_script used default
output which triggers a bug when sending metrics. As this isn't
relevant to the test switch to using named software events.
Update the match in test_hybrid as the cycles event is now cpu-cycles
to workaround potential ARM issues.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 11 Nov 2025 21:21:59 +0000 (13:21 -0800)]
perf test stat+json: Improve metric-only testing
When testing metric-only, pass a metric to perf rather than expecting
a hard coded metric value to be generated.
Remove keys that were really metric-only units and instead don't
expect metric only to have a matching json key as it encodes metrics
as {"metric_name", "metric_value"}.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 11 Nov 2025 21:21:56 +0000 (13:21 -0800)]
perf stat: Fix default metricgroup display on hybrid
The logic to skip output of a default metric line was firing on
Alderlake and not displaying 'TopdownL1 (cpu_atom)'. Remove the
need_full_name check as it is equivalent to the different PMU test in
the cases we care about, merge the 'if's and flip the evsel of the PMU
test. The 'if' is now basically saying, if the output matches the last
printed output then skip the output.
Ian Rogers [Tue, 11 Nov 2025 21:21:55 +0000 (13:21 -0800)]
perf stat: Remove hard coded shadow metrics
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
Ian Rogers [Tue, 11 Nov 2025 21:21:54 +0000 (13:21 -0800)]
perf script: Change metric format to use json metrics
The metric format option isn't properly supported. This change
improves that by making the sample events update the counts of an
evsel, where the shadow metric code expects to read the values. To
support printing metrics, metrics need to be found. This is done on
the first attempt to print a metric. Every metric is parsed and then
the evsels in the metric's evlist compared to those in perf script
using the perf_event_attr type and config. If the metric matches then
it is added for printing. As an event in the perf script's evlist may
have >1 metric id, or different leader for aggregation, the first
metric matched will be displayed in those cases.
An example use is:
```
$ perf record -a -e '{instructions,cpu-cycles}:S' -a -- sleep 1
$ perf script -F period,metric
...
867817
metric: 0.30 insn per cycle
125394
metric: 0.04 insn per cycle
313516
metric: 0.11 insn per cycle
metric: 1.00 insn per cycle
```
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 11 Nov 2025 21:21:52 +0000 (13:21 -0800)]
perf jevents: Add metric DefaultShowEvents
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
Ian Rogers [Tue, 11 Nov 2025 21:21:51 +0000 (13:21 -0800)]
perf jevents: Add set of common metrics based on default ones
Add support to getting a common set of metrics from a default
table. It simplifies the generation to add json metrics at the same
time. The metrics added are CPUs_utilized, cs_per_second,
migrations_per_second, page_faults_per_second, insn_per_cycle,
stalled_cycles_per_instruction, frontend_cycles_idle,
backend_cycles_idle, cycles_frequency, branch_frequency and
branch_miss_rate based on the shadow metric definitions.
Following this change the default perf stat output on an alderlake
looks like:
```
$ perf stat -a -- sleep 2
Ian Rogers [Tue, 11 Nov 2025 21:21:50 +0000 (13:21 -0800)]
perf expr: Add #target_cpu literal
For CPU nanoseconds a lot of the stat-shadow metrics use either
task-clock or cpu-clock, the latter being used when
target__has_cpu. Add a #target_cpu literal so that json metrics can
perform the same test.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 11 Nov 2025 21:21:49 +0000 (13:21 -0800)]
perf metricgroup: Add care to picking the evsel for displaying a metric
Rather than using the first evsel in the matched events, try to find
the least shared non-tool evsel. The aim is to pick the first evsel
that typifies the metric within the list of metrics.
This addresses an issue where Default metric group metrics may lose
their counter value due to how the stat displaying hides counters for
default event/metric output.
For a metricgroup like TopdownL1 on an Intel Alderlake the change is,
before there are 4 events with metrics:
```
$ perf stat -M topdownL1 -a sleep 1
Namhyung Kim [Tue, 11 Nov 2025 07:59:44 +0000 (23:59 -0800)]
perf tools: Fix missing feature check for inherit + SAMPLE_READ
It should also have PERF_SAMPLE_TID to enable inherit and PERF_SAMPLE_READ
on recent kernels. Not having _TID makes the feature check wrongly detect
the inherit and _READ support.
It was reported that the following command failed due to the error in
the missing feature check on Intel SPR machines.
$ perf record -e '{cpu/mem-loads-aux/S,cpu/mem-loads,ldlat=3/PS}' -- ls
Error:
Failure to open event 'cpu/mem-loads,ldlat=3/PS' on PMU 'cpu' which will be removed.
Invalid event (cpu/mem-loads,ldlat=3/PS) in per-thread mode, enable system wide with '-a'.
Reviewed-by: Ian Rogers <irogers@google.com> Fixes: 3b193a57baf15c468 ("perf tools: Detect missing kernel features properly") Reported-and-tested-by: Chen, Zide <zide.chen@intel.com> Closes: https://lore.kernel.org/lkml/20251022220802.1335131-1-zide.chen@intel.com/ Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Sun, 9 Nov 2025 00:59:59 +0000 (16:59 -0800)]
perf test: Add test that command line period overrides sysfs/json values
The behavior of weak terms is subtle, add a test that they aren't
accidentally broken. The test finds an event with a weak 'period' and
then overrides it. In no such event is present then the test skips.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Fri, 7 Nov 2025 17:07:12 +0000 (09:07 -0800)]
perf tool: Add a delegate_tool that just delegates actions to another tool
Add an ability to be able to compose perf_tools, by having one perform
an action and then calling a delegate. Currently the perf_tools have
if-then-elses setting the callback and then if-then-elses within the
callback. Understanding the behavior is complex as it is in two places
and logic for numerous operations, within things like perf inject, is
interwoven. By chaining perf_tools together based on command line
options this kind of code can be avoided.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Fri, 7 Nov 2025 17:07:11 +0000 (09:07 -0800)]
perf tool: Add the perf_tool argument to all callbacks
Getting context for what a tool is doing, such as the perf_inject
instance, using container_of the tool is a common pattern in the
code. This isn't possible event_op2, event_op3 and event_op4 callbacks
as the tool isn't passed. Add the argument and then fix function
signatures to match. As tools maybe reading a tool from somewhere
else, change that code to use the passed in tool.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Add JSON metrics for i.MX94 DDR Performance Monitor.
Reviewed-by: Peng Fan <peng.fan@nxp.com> Reviewed-by: Frank Li <Frank.Li@nxp.com> Signed-off-by: Xu Yang <xu.yang_2@nxp.com> Acked-by: Ian Rogers <irogers@google.com> Acked-by: Conor Dooley <conor.dooley@microchip.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Namhyung Kim [Thu, 6 Nov 2025 19:00:23 +0000 (11:00 -0800)]
perf record: Make sure to update build-ID cache
Recent change on enabling --buildid-mmap by default brought an issue
with build-id handling. With build-ID in MMAP2 records, we don't need
to save the build-ID table in the header of a perf data file.
But the actual file contents still need to be cached in the debug
directory for annotation etc. Split the build-ID header processing and
caching and make sure perf record to save hit DSOs in the build-ID cache
by moving perf_session__cache_build_ids() to the end of the record__
finish_output().
Reviewed-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Fri, 24 Oct 2025 17:58:41 +0000 (10:58 -0700)]
perf jevents: Make all tables static
The tables created by jevents.py are only used within the pmu-events.c
file. Change the declarations of those global variables to be static
to encapsulate this.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Fri, 24 Oct 2025 17:58:37 +0000 (10:58 -0700)]
perf metricgroup: Update comment on location of metric_event list
Update comment as the stat_config no longer holds all metrics.
Signed-off-by: Ian Rogers <irogers@google.com> Fixes: faebee18d720 ("perf stat: Move metric list from config to evlist") Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Fri, 31 Oct 2025 19:42:16 +0000 (12:42 -0700)]
perf tools: Cache counter names for raw samples on s390
Searching all event names is slower now that legacy names are
included. Add a cache to avoid long iterative searches. Note, the
cache isn't cleaned up and is as such a memory leak, however, globally
reachable leaks like this aren't treated as leaks by leak sanitizer.
Reported-by: Thomas Richter <tmricht@linux.ibm.com> Closes: https://lore.kernel.org/linux-perf-users/09943f4f-516c-4b93-877c-e4a64ed61d38@linux.ibm.com/ Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: Thomas Richter <tmricht@linux.ibm.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Namhyung Kim [Mon, 19 May 2025 23:25:39 +0000 (16:25 -0700)]
perf trace: Increase syscall handler map size to 1024
The syscalls_sys_{enter,exit} map in augmented_raw_syscalls.bpf.c has
max entries of 512. Usually syscall numbers are smaller than this but
x86 has x32 ABI where syscalls start from 512.
That makes trace__init_syscalls_bpf_prog_array_maps() fail in the middle
of the loop when it accesses those keys. As the loop iteration is not
ordered by syscall numbers anymore, the failure can affect non-x32
syscalls.
Let's increase the map size to 1024 so that it can handle those ABIs
too. While most systems won't need this, increasing the size will be
safer for potential future changes.
Reviewed-by: Howard Chu <howardchu95@gmail.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Chu Guangqing [Fri, 31 Oct 2025 02:58:10 +0000 (10:58 +0800)]
perf vendor events AmpereOneX: Fix spelling typo in the metrics file
The json file incorrectly used "acceses" instead of "accesses".
Signed-off-by: Chu Guangqing <chuguangqing@inspur.com> Reviewed-by: James Clark <james.clark@linaro.org> Reviewed-by: Ilkka Koskinen <ilkka@os.amperecomputing.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Chu Guangqing <chuguangqing@inspur.com> Reviewed-by: James Clark <james.clark@linaro.org> Reviewed-by: Ilkka Koskinen <ilkka@os.amperecomputing.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Shuai Xue [Thu, 23 Oct 2025 01:50:43 +0000 (09:50 +0800)]
perf record: skip synthesize event when open evsel failed
When using perf record with the `--overwrite` option, a segmentation fault
occurs if an event fails to open. For example:
perf record -e cycles-ct -F 1000 -a --overwrite
Error:
cycles-ct:H: PMU Hardware doesn't support sampling/overflow-interrupts. Try 'perf stat'
perf: Segmentation fault
#0 0x6466b6 in dump_stack debug.c:366
#1 0x646729 in sighandler_dump_stack debug.c:378
#2 0x453fd1 in sigsegv_handler builtin-record.c:722
#3 0x7f8454e65090 in __restore_rt libc-2.32.so[54090]
#4 0x6c5671 in __perf_event__synthesize_id_index synthetic-events.c:1862
#5 0x6c5ac0 in perf_event__synthesize_id_index synthetic-events.c:1943
#6 0x458090 in record__synthesize builtin-record.c:2075
#7 0x45a85a in __cmd_record builtin-record.c:2888
#8 0x45deb6 in cmd_record builtin-record.c:4374
#9 0x4e5e33 in run_builtin perf.c:349
#10 0x4e60bf in handle_internal_command perf.c:401
#11 0x4e6215 in run_argv perf.c:448
#12 0x4e653a in main perf.c:555
#13 0x7f8454e4fa72 in __libc_start_main libc-2.32.so[3ea72]
#14 0x43a3ee in _start ??:0
The --overwrite option implies --tail-synthesize, which collects non-sample
events reflecting the system status when recording finishes. However, when
evsel opening fails (e.g., unsupported event 'cycles-ct'), session->evlist
is not initialized and remains NULL. The code unconditionally calls
record__synthesize() in the error path, which iterates through the NULL
evlist pointer and causes a segfault.
To fix it, move the record__synthesize() call inside the error check block, so
it's only called when there was no error during recording, ensuring that evlist
is properly initialized.
Namhyung Kim [Thu, 30 Oct 2025 04:01:39 +0000 (21:01 -0700)]
perf lock contention: Load kernel map before lookup
On some machines, it caused troubles when it tried to find kernel
symbols. I think it's because kernel modules and kallsyms are messed
up during load and split.
Basically we want to make sure the kernel map is loaded and the code has
it in the lock_contention_read(). But recently we added more lookups in
the lock_contention_prepare() which is called before _read().
Also the kernel map (kallsyms) may not be the first one in the group
like on ARM. Let's use machine__kernel_map() rather than just loading
the first map.
Reviewed-by: Ian Rogers <irogers@google.com> Fixes: 688d2e8de231c54e ("perf lock contention: Add -l/--lock-addr option") Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Tue, 28 Oct 2025 15:38:20 +0000 (08:38 -0700)]
perf test workload: Add thread count argument to thloop
Allow the number of threads for the thloop workload to be increased
beyond the normal 2. Add error checking to the parsed time and thread
count values.
Signed-off-by: Ian Rogers <irogers@google.com> Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
The line_len is only set on success. Check the return value instead.
util/hwmon_pmu.c: In function ‘perf_pmus__read_hwmon_pmus’:
util/hwmon_pmu.c:742:20: warning: ‘line_len’ may be used uninitialized [-Wmaybe-uninitialized]
742 | if (line_len > 0 && line[line_len - 1] == '\n')
| ^
util/hwmon_pmu.c:719:24: note: ‘line_len’ was declared here
719 | size_t line_len;
Fixes: 53cc0b351ec9 ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs") Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
To avoid hardcoding the offset value for synthetic event IDs
in multiple auxtrace modules (arm-spe, cs-etm, intel-pt, etc.),
and to improve code reusability, this patch unifies
the handling of the ID offset via a dedicated helper function.
Signed-off-by: tanze <tanze@kylinos.cn> Reviewed-by: Adrian Hunter <adrian.hunter@intel.com> Tested-by: Leo Yan <leo.yan@arm.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Sat, 25 Oct 2025 20:28:34 +0000 (13:28 -0700)]
perf stat: Add/fix bperf cgroup max events workarounds
Commit b8308511f6e0 bumped the max events to 1024 but this results in
BPF verifier issues if the number of command line events is too
large. Workaround this by:
1) moving the constants to a header file to share between BPF and perf
C code,
2) testing that the maximum number of events doesn't cause BPF
verifier issues in debug builds,
3) lower the max events from 1024 to 128,
4) in perf stat, if there are more events than the BPF counters can
support then disable BPF counter usage.
The rodata setup is factored into its own function to avoid
duplicating it in the testing code.
Signed-off-by: Ian Rogers <irogers@google.com> Fixes: b8308511f6e0 ("perf stat bperf cgroup: Increase MAX_EVENTS from 32 to 1024") Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Leo Yan [Wed, 22 Oct 2025 14:28:49 +0000 (15:28 +0100)]
perf cs-etm: Mute enumeration value warning
When the OpenCSD library introduces a new enumeration value (for example,
in the v1.7.1 release), the perf build fails with an error:
util/cs-etm-decoder/cs-etm-decoder.c:600:10: error: enumeration value 'OCSD_GEN_TRC_ELEM_ITMTRACE' not explicitly handled in switch [-Werror, -Wswitch-enum]
600 | switch (elem->elem_type) {
| ^~~~~~~~~~~~~~~
1 error generated.
Convert to if-else sentences to mute the enumeration value warning,
which can avoid build failures whenever the lib is updated.
No functional change.
Suggested-by: Suzuki K Poulose <suzuki.poulose@arm.com> Signed-off-by: Leo Yan <leo.yan@arm.com> Reviewed-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com> Reviewed-by: Leo Yan <leo.yan@arm.com> Acked-by: Will Deacon <will@kernel.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
James Clark [Wed, 22 Oct 2025 11:02:40 +0000 (12:02 +0100)]
perf annotate: Fix Clang build by adding block in switch case
Clang and GCC disagree with what constitutes a "declaration after
statement". GCC allows declarations in switch cases without an extra
block, as long as it's immediately after the label. Clang does not.
Unfortunately this is the case even in the latest versions of both
compilers. The only option that makes them behave in the same way is
-Wpedantic, which can't be enabled in Perf because of the number of
warnings it generates.
Add a block to fix the Clang build, which is the only thing we can do.
ui/browsers/annotate.c:1008:4: error: use of undeclared identifier 'al'
al = annotated_source__get_line(notes->src, offset);
ui/browsers/annotate.c:1009:24: error: use of undeclared identifier 'al'
browser->curr_hot = al ? &al->rb_node : NULL;
ui/browsers/annotate.c:1009:30: error: use of undeclared identifier 'al'
browser->curr_hot = al ? &al->rb_node : NULL;
ui/browsers/annotate.c:1000:8: error: mixing declarations and code is incompatible with standards before C99 [-Werror,-Wdeclaration-after-statement]
s64 offset = annotate_browser__curr_hot_offset(browser);
Fixes: ad83f3b7155d ("perf c2c annotate: Start from the contention line") Signed-off-by: James Clark <james.clark@linaro.org> Reviewed-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Zecheng Li [Mon, 13 Oct 2025 18:16:02 +0000 (18:16 +0000)]
perf annotate: Invalidate register states for untracked instructions
When tracking variable types, instructions that modify a pointer value
in an untracked way can lead to incorrect type propagation. To prevent
this, invalidate the register state when encountering such instructions.
This change invalidates pointer types for various arithmetic and bitwise
operations that current pointer offset tracking doesn't support, like
imul, shl, and, inc, etc.
A special case is added for 'xor reg, reg', which is a common idiom for
zeroing a register. For this, the register state is updated to be a
constant with a value of 0.
This could introduce slight regressions if a variable is zeroed and then
reused. This can be addressed in the future by using all DWARF locations
for instruction tracking instead of only the first one.
Signed-off-by: Zecheng Li <zecheng@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Zecheng Li [Mon, 13 Oct 2025 18:16:01 +0000 (18:16 +0000)]
perf annotate: Save pointer offset in stack state
The tracked pointer offset was not being preserved in the stack state,
which could lead to incorrect type analysis. This change adds a
ptr_offset field to the type_state_stack struct and passes it to
set_stack_state and findnew_stack_state to ensure the offset is
preserved after the pointer is loaded from a stack location. It improves
the type annotation coverage and quality.
Signed-off-by: Zecheng Li <zecheng@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Zecheng Li [Mon, 13 Oct 2025 18:16:00 +0000 (18:16 +0000)]
perf annotate: Track arithmetic instructions on pointers
Track the arithmetic operations on registers with pointer types. We
handle only add, sub and lea instructions. The original pointer
information needs to be preserved for getting outermost struct types.
For example, reg0 points to a struct cfs_rq, when we add 0x10 to reg0,
it should preserve the information of struct cfs_rq + 0x10 in the
register instead of a pointer type to the child field at 0x10.
Details:
1. struct type_state_reg now includes an offset, indicating if the
register points to the start or an internal part of its associated
type. This offset is used in mem to reg and reg to stack mem
transfers, and also applied to the final type offset.
2. lea offset(%sp/%fp), reg is now treated as taking the address of a
stack variable. It worked fine in most cases, but an issue with this
approach is the pointer type may not exist.
3. lea offset(%base), reg is handled by moving the type from %base and
adding an offset, similar to an add operation followed by a mov reg
to reg.
4. Non-stack variables from DWARF with non-zero offsets in their
location expressions are now accepted with register offset tracking.
Multi-register addressing modes in LEA are not supported.
Signed-off-by: Zecheng Li <zecheng@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Zecheng Li [Mon, 13 Oct 2025 18:15:59 +0000 (18:15 +0000)]
perf annotate: Track address registers via TSR_KIND_POINTER
Introduce TSR_KIND_POINTER to improve the data type profiler's ability
to track pointer-based memory accesses and address register variables.
TSR_KIND_POINTER represents that the location holds a pointer type to
the type in the type state. The semantics match the `breg` registers
that describe a memory location.
This change implements handling for this new kind in mov instructions
and in the check_matching_type() function. When a TSR_KIND_POINTER is
moved to the stack, the stack state size is set to the architecture's
pointer size.
Signed-off-by: Zecheng Li <zecheng@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Zecheng Li [Mon, 13 Oct 2025 18:15:58 +0000 (18:15 +0000)]
perf annotate: Skip annotating data types to lea instructions
Introduce a helper function is_address_gen_insn() to check
arch-dependent address generation instructions like lea in x86. Remove
type annotation on these instructions since they are not accessing
memory. It should be counted as `no_mem_ops`.
Signed-off-by: Zecheng Li <zecheng@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tianyou Li [Mon, 20 Oct 2025 07:30:05 +0000 (15:30 +0800)]
perf annotate: Check return value of evsel__get_arch() properly
Check the error code of evsel__get_arch() in the symbol__annotate().
Previously it checked non-zero value but after the refactoring it does
only for negative values.
Fixes: 0669729eb0afb0cf ("perf annotate: Factor out evsel__get_arch()") Suggested-by: James Clark <james.clark@linaro.org> Acked-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Tianyou Li <tianyou.li@intel.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tianyou Li [Mon, 20 Oct 2025 07:30:04 +0000 (15:30 +0800)]
perf annotate: fix a crash when annotate the same symbol with 's' and 'T'
When perf report with annotation for a symbol, press 's' and 'T', then exit
the annotate browser. Once annotate the same symbol, the annotate browser
will crash.
The browser.arch was required to be correctly updated when data type
feature was enabled by 'T'. Usually it was initialized by symbol__annotate2
function. If a symbol has already been correctly annotated at the first
time, it should not call the symbol__annotate2 function again, thus the
browser.arch will not get initialized. Then at the second time to show the
annotate browser, the data type needs to be displayed but the browser.arch
is empty.
Stack trace as below:
Perf: Segmentation fault
-------- backtrace --------
#0 0x55d365 in ui__signal_backtrace setup.c:0
#1 0x7f5ff1a3e930 in __restore_rt libc.so.6[3e930]
#2 0x570f08 in arch__is perf[570f08]
#3 0x562186 in annotate_get_insn_location perf[562186]
#4 0x562626 in __hist_entry__get_data_type annotate.c:0
#5 0x56476d in annotation_line__write perf[56476d]
#6 0x54e2db in annotate_browser__write annotate.c:0
#7 0x54d061 in ui_browser__list_head_refresh perf[54d061]
#8 0x54dc9e in annotate_browser__refresh annotate.c:0
#9 0x54c03d in __ui_browser__refresh browser.c:0
#10 0x54ccf8 in ui_browser__run perf[54ccf8]
#11 0x54eb92 in __hist_entry__tui_annotate perf[54eb92]
#12 0x552293 in do_annotate hists.c:0
#13 0x55941c in evsel__hists_browse hists.c:0
#14 0x55b00f in evlist__tui_browse_hists perf[55b00f]
#15 0x42ff02 in cmd_report perf[42ff02]
#16 0x494008 in run_builtin perf.c:0
#17 0x494305 in handle_internal_command perf.c:0
#18 0x410547 in main perf[410547]
#19 0x7f5ff1a295d0 in __libc_start_call_main libc.so.6[295d0]
#20 0x7f5ff1a29680 in __libc_start_main@@GLIBC_2.34 libc.so.6[29680]
#21 0x410b75 in _start perf[410b75]
Fixes: 1d4374afd000 ("perf annotate: Add 'T' hot key to toggle data type display") Reviewed-by: James Clark <james.clark@linaro.org> Tested-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Tianyou Li <tianyou.li@intel.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Namhyung Kim [Tue, 21 Oct 2025 03:07:50 +0000 (12:07 +0900)]
perf annotate: Fix build with NO_SLANG=1
The recent change for perf c2c annotate broke build without slang
support like below.
builtin-annotate.c: In function 'hists__find_annotations':
builtin-annotate.c:522:73: error: 'NO_ADDR' undeclared (first use in this function); did you mean 'NR_ADDR'?
522 | key = hist_entry__tui_annotate(he, evsel, NULL, NO_ADDR);
| ^~~~~~~
| NR_ADDR
builtin-annotate.c:522:73: note: each undeclared identifier is reported only once for each function it appears in
builtin-annotate.c:522:31: error: too many arguments to function 'hist_entry__tui_annotate'
522 | key = hist_entry__tui_annotate(he, evsel, NULL, NO_ADDR);
| ^~~~~~~~~~~~~~~~~~~~~~~~
In file included from util/sort.h:6,
from builtin-annotate.c:28:
util/hist.h:756:19: note: declared here
756 | static inline int hist_entry__tui_annotate(struct hist_entry *he __maybe_unused,
| ^~~~~~~~~~~~~~~~~~~~~~~~
And I noticed that it missed to update the other side of #ifdef
HAVE_SLANG_SUPPORT. Let's fix it.
Cc: Tianyou Li <tianyou.li@intel.com> Fixes: cd3466cd2639783d ("perf c2c: Add annotation support to perf c2c report") Signed-off-by: Namhyung Kim <namhyung@kernel.org>
When doing an in source build, $(OUTPUT) is empty so the rule has the
same input and output file. Suppress the warning by only adding the rule
when doing an out of source build. The same condition already exists for
the clean rule for json files.
Signed-off-by: James Clark <james.clark@linaro.org> Tested-by: Namhyung Kim <namhyung@kernel.org> Tested-by: Leo Yan <leo.yan@arm.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
James Clark [Mon, 20 Oct 2025 16:08:26 +0000 (17:08 +0100)]
perf jevents: Fix build when there are other json files in the tree
The unquoted glob *.json will expand to a real file if, for example,
there is any file in the Perf source ending in .json. This can happen
when using tools like Bear and clangd which generate a
compile_commands.json file. With the glob already expanded by the shell,
the find command will fail to wildcard any real json events files.
Fix it by wrapping the star in quotes so it's passed to find rather than
the shell.
This fixes the following build error (most of the diff output omitted):
$ make V=1 -C tools/perf O=/tmp/perf_build_with_json
Fixes: 4bb55de4ff03 ("perf jevents: Support copying the source json files to OUTPUT") Signed-off-by: James Clark <james.clark@linaro.org> Reviewed-by: Leo Yan <leo.yan@arm.com> Tested-by: Leo Yan <leo.yan@arm.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Fri, 17 Oct 2025 23:03:57 +0000 (16:03 -0700)]
perf parse-events: Make X modifier more respectful of groups
Events with an X modifier were reordered within a group, for example
slots was made the leader in:
```
$ perf record -e '{cpu/mem-stores/ppu,cpu/slots/uX}' -- sleep 1
```
Fix by making `dont_regroup` evsels always use their index for
sorting. Make the cur_leader, when fixing the groups, be that of
`dont_regroup` evsel so that the `dont_regroup` evsel doesn't become a
leader.
On a tigerlake this patch corrects this and meets expectations in:
```
$ perf stat -e '{cpu/mem-stores/,cpu/slots/uX}' -a -- sleep 0.1
Closes: https://lore.kernel.org/lkml/18f20d38-070c-4e17-bc90-cf7102e1e53d@linux.intel.com/ Fixes: 035c17893082 ("perf parse-events: Add 'X' modifier to exclude an event from being regrouped") Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tianyou Li [Mon, 13 Oct 2025 14:48:11 +0000 (22:48 +0800)]
perf c2c annotate: Start from the contention line
Add support to highlight the contention line in the annotate browser,
use 'TAB'/'UNTAB' to refocus to the contention line.
Signed-off-by: Tianyou Li <tianyou.li@intel.com> Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com> Reviewed-by: Thomas Falcon <thomas.falcon@intel.com> Reviewed-by: Jiebin Sun <jiebin.sun@intel.com> Reviewed-by: Pan Deng <pan.deng@intel.com> Reviewed-by: Zhiguo Zhou <zhiguo.zhou@intel.com> Reviewed-by: Wangyang Guo <wangyang.guo@intel.com> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tianyou Li [Mon, 13 Oct 2025 14:48:10 +0000 (22:48 +0800)]
perf c2c: Add annotation support to perf c2c report
Perf c2c report currently specified the code address and source:line
information in the cacheline browser, while it is lack of annotation
support like perf report to directly show the disassembly code for
the particular symbol shared that same cacheline. This patches add
a key 'a' binding to the cacheline browser which reuse the annotation
browser to show the disassembly view for easier analysis of cacheline
contentions.
Signed-off-by: Tianyou Li <tianyou.li@intel.com> Reviewed-by: Dapeng Mi <dapeng1.mi@linux.intel.com> Reviewed-by: Thomas Falcon <thomas.falcon@intel.com> Reviewed-by: Jiebin Sun <jiebin.sun@intel.com> Reviewed-by: Pan Deng <pan.deng@intel.com> Reviewed-by: Zhiguo Zhou <zhiguo.zhou@intel.com> Reviewed-by: Wangyang Guo <wangyang.guo@intel.com> Tested-by: Ravi Bangoria <ravi.bangoria@amd.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 16 Oct 2025 15:07:18 +0000 (08:07 -0700)]
perf stat bperf cgroup: Increase MAX_EVENTS from 32 to 1024
The MAX_EVENTS value ensured a counted loop presumably to satisfy the
BPF verifier. It is possible to go past 32 events when gathering
uncore events. Increase the amount to 1024 as that should provide some
amount of headroom.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 16 Oct 2025 22:22:28 +0000 (15:22 -0700)]
perf ilist: Add PMU information to metrics
Duplicate metrics may exist on hybrid platforms, with the metric's PMU
being used to select the metric to use. Incorporate the metric PMU
into the ilist display and support opening it just for a given PMU.
Before:
```
⭘ Interactive Perf List
├── ▼ TopdownL1 tma_backend_bound
│ ├── tma_backend_bound Counts the total number of issue slots that were
│ ├── ▶ tma_backend_bound_group not consumed by the backend due to backend stalls
│ ├── tma_backend_bound Counts the total number of issue slots that were
│ ├── ▶ tma_backend_bound_group not consumed by the backend due to backend stalls.
│ ├── tma_bad_speculation Note that uops must be available for consumption
│ ├── ▶ tma_bad_speculation_group in order for this event to count. If a uop is not
│ ├── tma_bad_speculation available (IQ is empty), this event will not count
│ ├── ▶ tma_bad_speculation_group cpu_atom@TOPDOWN_BE_BOUND.ALL@ / (5 *
│ ├── tma_frontend_bound cpu_atom@CPU_CLK_UNHALTED.CORE@)
│ ├── ▶ tma_frontend_bound_group tma_backend_bound > 0.1
│ ├── tma_frontend_bound ▆▆
│ ├── ▶ tma_frontend_bound_group
│ ├── tma_retiring
│ ├── ▶ tma_retiring_group
│ ├── tma_retiring
│ └── ▶ tma_retiring_group
├── ▶ TopdownL2
total▄▄▅▅▆▅▅▂▁▁▁▁▂▃▂▂▃▄▄▇▇█▆▆▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▄▅▅▅▄▆▆▆▅▅▅▅▅▅▇▇▇▇▆▅▆▆▆▆▅▅▅▄▃▃▃▃▃▃▃▃▃▃▄▄▄▅▅▅▅▅▆▆▆▆▆▆
cpu0▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu1▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▃▃▃▃▄▄▄▄▄▄▄▄▄▄▅▅▅▅▅▅▅▅▅▅▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇█████▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
cpu3▁▁▁▁▁▁▁▁▁▄▄▄▄▄▄▄▄▄▄█████▆▆▆▆▆▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅
cpu4████▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂
cpu5▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▃▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▇▇▇▇▇▇▇▇▇▇▆▆
cpu6▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇█████▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
cpu7▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu8▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▃▃▃▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▅▅▅▅▅▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu9▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂█████▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅
cpu10▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu11▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇█████▁▁▁▁▁▁▁▁▁
cpu12▁▁▁▁▁▁▁▁▁▁▁▁▁▁▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu13▁▁▁▁▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu14▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇█████▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
cpu15▁▁▁▁▁▁▁▁▁▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu16▃▃▃▃▃▃▃▃▃▄▄▃▃▃▃▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄▄▃▃▃▃▃▃▃▃▄▄▄▄▄▄▁▁▃▃▃▃▃▃▃▃▃▃▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
cpu17▁▁▁▁▁▄▄▅▅▅▅▅▅▅▅▄▄▄▄▄▄▄▄▃▃▃▃▂▂▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▄▄▄▄▃▃▃▃▂▂▂▂▄▄▄▄▄▄▆▆▆▆▆▆▆▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
s Search n Next p Previous c Collapse ^q Quit ▏^p palette
```
After:
```
⭘ Interactive Perf List
├── ▼ TopdownL1 tma_backend_bound
│ ├── tma_backend_bound (cpu_atom) Counts the total number of issue slots that were
│ ├── ▼ tma_backend_bound_group (cpu_atom) not consumed by the backend due to backend stalls
│ │ ├── tma_core_bound (cpu_atom) Counts the total number of issue slots that were
│ │ ├── ▶ tma_core_bound_group (cpu_atom not consumed by the backend due to backend stalls.
│ │ ├── tma_resource_bound (cpu_atom) Note that uops must be available for consumption
│ │ └── ▶ tma_resource_bound_group (cpu_ in order for this event to count. If a uop is not
│ ├── tma_backend_bound (cpu_core) available (IQ is empty), this event will not count
│ ├── ▶ tma_backend_bound_group (cpu_core) cpu_atom@TOPDOWN_BE_BOUND.ALL@ / (5 *
│ ├── tma_bad_speculation (cpu_atom) cpu_atom@CPU_CLK_UNHALTED.CORE@)
│ ├── ▶ tma_bad_speculation_group (cpu_ato▆▆tma_backend_bound > 0.1
│ ├── tma_bad_speculation (cpu_core)
│ ├── ▶ tma_bad_speculation_group (cpu_cor▃▃
│ ├── tma_frontend_bound (cpu_atom)
│ ├── ▶ tma_frontend_bound_group (cpu_atom
│ ├── tma_frontend_bound (cpu_core)
│ ├── ▶ tma_frontend_bound_group (cpu_core
▌
total▁▁▁▁▂▂▂▂▂▂▂▂▂▃▃▃▃▃▃▃▃▃▃▂▂▂▂▃▃▃▄▄▄▄▄▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▄▄▄▄▄▄▅▅▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▆▇▆▇▇
cpu16▇▇▇▇▇▇▇▇▇▇▇▆▆▁▁▁▁▁▁▁▁▁▁▁▁▂▂▄▄▅▅▅▅▅▅▆▆▆▆▆▆▆▆▇▇▇▇▆▆▆▆▆▆▆▆▅▅▅▅▅▅▅▅▅▅▅▅▅▅▆▆▆▆▄▄▄▄▃▃▄▄▄▄▇▇▇▇▇▇▇▇▇▇
cpu17█▇▇▇▇▇▇▇▇▆▆▆▆▆▆▅▅▅▅▃▃▃▃▂▂▁▁▅▅▅▅▅▅▅▅▄▄▄▄▄▄▄▄▄▄▄▄▃▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▁▁▁▁▅▅▄▄▂▂▇▇▇▇▆▆▅▅▆▆
cpu18▇▇▇▇▇██▇▇▃▃▃▃▃▃▃▃▃▃▂▂▂▂▂▂▂▂▃▃▃▃▃▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄▄▄▄▄▄▅▅▅▅▅▅▅▅
cpu19▇▃▃▄▄▄▄▄▄▄▄▄▄▅▅▅▅▅▅▅▅▁▁▂▂▃▃▃▃▅▅▆▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇██▇▇▇▇▇▇▆▆▅▅▅▅▆▆▄▄▄▄▅▅
cpu20▃▃▃▃▃▃▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▂▂▂▂▂▂▂▂▂▄▄▄▄▅▅▅▅▅▅▆▆▇▇
cpu21▇▇▇▇▇▇▇██▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▆▆▆▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▆▆▆▆▆▆▆▆▆▆▆▆▆▆▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▅▅▄▄▂▂▂▂▂▂▁▁▁▁
cpu22█▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▆▆▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▂▂▁▁▁▁▂▂▂▂▂▂▂▂
cpu23▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▃▃▄▄▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▅▇▇▇▇▆▆██▇▇▇▇▇▇
cpu24▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▂▃▃▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▅▅▇▇▆▆▆▆▆▆▇▇▇▇
cpu25▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▆▆▆▆▇▇▇▇▇▇██
cpu26▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▃▃▄▄▇▇▇▇▇▇▇▇▇▇██▇▇▇▇▇▇▇▇▇▇▆▆▆▆▆▆▆▆▆▆▆▆▇▇▇▇▇▇▇▇▆▆▆▆▆▆▆▆▆▆▂▂▁▁▁▁▂▂▂▂▂▂▂▂
cpu27▂▂▂▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▃▃▄▄▄▄▅▅▅▅▅▅▇▇
total 7.4923074548462605
cpu16 0.2961618003253457
cpu17 0.3065719718925585
cpu18 0.27800656881051855
cpu19 0.28564742078353406
cpu20 0.2764790653117084
s Search n Next p Previous c Collapse ^q Quit ▏^p palette
```
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 16 Oct 2025 22:22:27 +0000 (15:22 -0700)]
perf python: Add PMU argument to parse_metrics
Add an optional PMU argument to parse_metrics to allow restriction of
the particular metrics to be opened. If no argument is provided then
all metrics with the given name/group are opened
Signed-off-by: Ian Rogers <irogers@google.com> Acked-by: Gautam Menghani <gautam@linux.ibm.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Thu, 16 Oct 2025 22:22:26 +0000 (15:22 -0700)]
perf ilist: Don't display deprecated events
Unsupported legacy events are flagged as deprecated. Don't display
these events in ilist as they won't open and there are over 1,000
legacy cache events.
Signed-off-by: Ian Rogers <irogers@google.com> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Sun, 5 Oct 2025 18:24:27 +0000 (11:24 -0700)]
perf test: Switch cycles event to cpu-cycles
Without a PMU perf matches an event against any PMU with the
event. Unfortunately some PMU drivers advertise a "cycles" event which
is typically just a core event. As tests assume a core event, switch
to use "cpu-cycles" that avoids the overloaded "cycles" event on
troublesome PMUs and is so far not overloaded. Note, on x86 this
changes a legacy event into a sysfs one.
Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Ian Rogers [Sun, 5 Oct 2025 18:24:26 +0000 (11:24 -0700)]
perf test parse-events: Remove cpu PMU requirement
In the event parse string, switch "cpu" to "default_core" and then
rewrite this to the first core PMU name prior to parsing. This enables
testing with a PMU on hybrid x86 and other systems that don't use
"cpu" for the core PMU name. The name "default_core" is already used
by jevents. Update test expectations to match.
Signed-off-by: Ian Rogers <irogers@google.com> Tested-by: James Clark <james.clark@linaro.org> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
projects / thirdparty / kernel / linux.git / log
