Skip to contentSkip to conversation

New perf ilist app

PATCH Superseded Ian Rogers 1 participant 17 messages Jun 11, 2025 perf events

v2 of 10 Superseded Latest is v10 →

lore
Share & tools
lore

Ian Rogers

author reviewer
#0 lore raw

v2 cover New perf ilist app

This patch series builds up to the addition of a new ilist app written in python using textual [1] for the UI. The app presents perf PMUs and events, displays the event information as in perf list while at the bottom of the console showing recent activity of the event in total and across all CPUs.

The first part of the patches are a few perf and perf python C API fixes, most importantly the counter reading in python supports tool PMUs.

The second part of the patches adds event json for the software PMU and makes the tracepoint PMU support iteration of events and the like. Without these improvements the tracepoint and software PMUs will appear to have no events in the ilist app. As the software PMU moves parsing to json, the legacy hard coded parsing is removed. This has proven controversial for hardware events and so that cleanup isn't done here.

The final patch adds the ilist command. To run it you need the updated perf.cpython.so in your PYTHONPATH and then execute the script. Expanding PMUs and then selecting events will cause event informatin to be displayed in the top-right and the counters values to be displayed as sparklines and counts in the bottom half of the screen.

Some thoughts on the series:

- The PMU changes will conflict with the addition of the DRM PMU:
https://lore.kernel.org/lkml/[email protected]/
when these two are merged together ilist will show yet more counters. It'd be nice if the DRM stuff could land and then I can rebase these patches.

- The parse-events clean up of the software and tracepoint PMU. The software PMU hard coding to be legacy first has similar issues and will conflict with the clean up in:
https://lore.kernel.org/lkml/[email protected]/
Moving the software work to json means we don't need special parse events terms for software events, etc. We can just treat things like regular PMUs with json, etc. I'd much rather we had less special case logic so that series is best rebased on top of this work and it should drop the changes for software terms, etc. which this series removes. Maybe one day the whole event parsing can be much more regular in how PMUs are treated but there's always "cycles".

- Should python libraries have feature tests? How does this get packaged outside the kernel tree? I think these are open questions. Clearly textual is kind of a big dependency and we've largely been moving in the direction of fewer dependencies recently. Hopefully the app makes it clear why I think this one is worth carrying. We carry libslang as a dependency and I think textual clearly far surpasses it.

- How to launch? Currently I run tools/perf/python/ilist.py but it would be much nicer if we could do perf ilist much as we do for perf-archive.sh. There are probably other scripts that should be perf commands like flamegraph and gecko. It'd be nice to follow up the series with something to make using these commands easy.

[1] https://textual.textualize.io/

v2: In the jevents event description duplication, some minor changes
accidentally missed from v1 meaning that in v1 the descriptions were still duplicated. Expand the cover letter with some thoughts on the series.

Ian Rogers (15):
perf hwmon_pmu: Avoid shortening hwmon PMU name
perf parse-events: Minor tidy up of event_type helper
perf python: In str(evsel) use the evsel__pmu_name helper perf python: Fix thread check in pyrf_evsel__read
perf python: Correct pyrf_evsel__read for tool PMUs
perf python: Add basic PMU abstraction and pmus sequence perf python: Add function returning dictionary of all events on a PMU perf jevents: If the long_desc and desc are identical then drop the
long_desc
perf jevents: Add common software event json
perf pmu: Tolerate failure to read the type for wellknown PMUs perf parse-events: Remove non-json software events
perf tp_pmu: Factor existing tracepoint logic to new file perf tp_pmu: Add event APIs
perf list: Remove tracepoint printing code
perf ilist: Add new python ilist command

 tools/perf/builtin-list.c                     |  47 ++--
 .../arch/common/common/software.json          |  92 ++++++
 tools/perf/pmu-events/empty-pmu-events.c      | 266 +++++++++++-------
 tools/perf/pmu-events/jevents.py              |  18 +-
 tools/perf/python/ilist.py                    | 238 ++++++++++++++++
 tools/perf/util/Build                         |   1 +
 tools/perf/util/evsel.c                       |  21 +-
 tools/perf/util/hwmon_pmu.c                   |   2 +-
 tools/perf/util/parse-events.c                | 225 +++++----------
 tools/perf/util/parse-events.h                |   3 +-
 tools/perf/util/parse-events.l                |  38 +--
 tools/perf/util/parse-events.y                |  29 +-
 tools/perf/util/pmu.c                         |  44 ++-
 tools/perf/util/print-events.c                |  95 -------
 tools/perf/util/print-events.h                |   1 -
 tools/perf/util/python.c                      | 248 +++++++++++++++-
 tools/perf/util/tp_pmu.c                      | 209 ++++++++++++++
 tools/perf/util/tp_pmu.h                      |  19 ++
 18 files changed, 1144 insertions(+), 452 deletions(-)
 create mode 100644 tools/perf/pmu-events/arch/common/common/software.json
 create mode 100755 tools/perf/python/ilist.py
 create mode 100644 tools/perf/util/tp_pmu.c
 create mode 100644 tools/perf/util/tp_pmu.h
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#1 lore raw

PATCH v2 1/15 perf hwmon_pmu: Avoid shortening hwmon PMU name

Long names like ucsi_source_psy_USBC000:001 when prefixed with hwmon_ exceed the buffer size and the last digit is lost. This causes confusion with similar names like ucsi_source_psy_USBC000:002. Extend the buffer size to avoid this.

Fixes 53cc0b351ec9 ("perf hwmon_pmu: Add a tool PMU exposing events from hwmon in sysfs")
---
 tools/perf/util/hwmon_pmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/hwmon_pmu.c +1 -1
index c25e7296f1c1..75683c543994 100644
@@ -344,7 +344,7 @@ static int hwmon_pmu__read_events(struct hwmon_pmu *pmu)
 
 struct perf_pmu *hwmon_pmu__new(struct list_head *pmus, int hwmon_dir, const char *sysfs_name, const char *name)
 {
-	char buf[32];
+	char buf[64];
 	struct hwmon_pmu *hwm;
 	__u32 type = PERF_PMU_TYPE_HWMON_START + strtoul(sysfs_name + 5, NULL, 10);
 
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#2 lore raw

PATCH v2 2/15 perf parse-events: Minor tidy up of event_type helper

In reply to Ian Rogers

Add missing breakpoint and raw types. Avoid a switch, just use a lookup array. Switch the type to unsigned to avoid checking negative values.

---
 tools/perf/util/parse-events.c | 31 +++++++++++++------------------
 tools/perf/util/parse-events.h |  2 +-
 2 files changed, 14 insertions(+), 19 deletions(-)
tools/perf/util/parse-events.c +13 -18
index 7f34e602fc08..9dd0216cfae4 100644
@@ -135,26 +135,21 @@ const struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = {
 	},
 };
 
-const char *event_type(int type)
-{
-	switch (type) {
-	case PERF_TYPE_HARDWARE:
-		return "hardware";
-
-	case PERF_TYPE_SOFTWARE:
-		return "software";
-
-	case PERF_TYPE_TRACEPOINT:
-		return "tracepoint";
-
-	case PERF_TYPE_HW_CACHE:
-		return "hardware-cache";
+static const char *const event_types[] = {
+	[PERF_TYPE_HARDWARE]	= "hardware",
+	[PERF_TYPE_SOFTWARE]	= "software",
+	[PERF_TYPE_TRACEPOINT]	= "tracepoint",
+	[PERF_TYPE_HW_CACHE]	= "hardware-cache",
+	[PERF_TYPE_RAW]		= "raw",
+	[PERF_TYPE_BREAKPOINT]	= "breakpoint",
+};
 
-	default:
-		break;
-	}
+const char *event_type(size_t type)
+{
+	if (type >= PERF_TYPE_MAX)
+		return "unknown";
 
-	return "unknown";
+	return event_types[type];
 }
 
 static char *get_config_str(const struct parse_events_terms *head_terms,
tools/perf/util/parse-events.h +1 -1
index 1c20ed0879aa..b47bf2810112 100644
@@ -21,7 +21,7 @@ struct option;
 struct perf_pmu;
 struct strbuf;
 
-const char *event_type(int type);
+const char *event_type(size_t type);
 
 /* Arguments encoded in opt->value. */
 struct parse_events_option_args {
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#3 lore raw

PATCH v2 3/15 perf python: In str(evsel) use the evsel__pmu_name helper

In reply to Ian Rogers

The evsel__pmu_name helper will internally use evsel__find_pmu that handles legacy events, extended types, etc. in determining a PMU and will provide a better value than just trying to access the PMU's name directly as the PMU may not have been computed.

---
 tools/perf/util/python.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
tools/perf/util/python.c +1 -4
index 82666bcd2eda..49e1db3feb87 100644
@@ -924,10 +924,7 @@ static PyObject *pyrf_evsel__str(PyObject *self)
 	struct pyrf_evsel *pevsel = (void *)self;
 	struct evsel *evsel = &pevsel->evsel;
 
-	if (!evsel->pmu)
-		return PyUnicode_FromFormat("evsel(%s)", evsel__name(evsel));
-
-	return PyUnicode_FromFormat("evsel(%s/%s/)", evsel->pmu->name, evsel__name(evsel));
+	return PyUnicode_FromFormat("evsel(%s/%s/)", evsel__pmu_name(evsel), evsel__name(evsel));
 }
 
 static PyMethodDef pyrf_evsel__methods[] = {
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#4 lore raw

PATCH v2 4/15 perf python: Fix thread check in pyrf_evsel__read

In reply to Ian Rogers

The CPU index is incorrectly checked rather than the thread index.

Fixes 739621f65702 ("perf python: Add evsel read method")
---
 tools/perf/util/python.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/python.c +1 -1
index 49e1db3feb87..b724478fe34a 100644
@@ -909,7 +909,7 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 		return NULL;
 	}
 	thread_idx = perf_thread_map__idx(evsel->core.threads, thread);
-	if (cpu_idx < 0) {
+	if (thread_idx < 0) {
 		PyErr_Format(PyExc_TypeError, "Thread %d is not part of evsel's threads",
 			     thread);
 		return NULL;
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#5 lore raw

PATCH v2 5/15 perf python: Correct pyrf_evsel__read for tool PMUs

In reply to Ian Rogers

Tool PMUs assume that stat's process_counter_values is being used to read the counters. Specifically they hold onto old values in evsel->prev_raw_counts and give the cumulative count based off of this value. Update pyrf_evsel__read to allocate counts and prev_raw_counts, use evsel__read_counter rather than perf_evsel__read so tool PMUs are read from not just perf_event_open events, make the returned pyrf_counts_values contain the delta value rather than the cumulative value.

Fixes 739621f65702 ("perf python: Add evsel read method")
---
 tools/perf/util/python.c | 34 +++++++++++++++++++++++++++++++---
 1 file changed, 31 insertions(+), 3 deletions(-)
tools/perf/util/python.c +31 -3
index b724478fe34a..201fe04fb6dd 100644
@@ -10,6 +10,7 @@
 #endif
 #include <perf/mmap.h>
 #include "callchain.h"
+#include "counts.h"
 #include "evlist.h"
 #include "evsel.h"
 #include "event.h"
@@ -893,7 +894,7 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 {
 	struct evsel *evsel = &pevsel->evsel;
 	int cpu = 0, cpu_idx, thread = 0, thread_idx;
-	struct perf_counts_values counts;
+	struct perf_counts_values *old_count, *new_count;
 	struct pyrf_counts_values *count_values = PyObject_New(struct pyrf_counts_values,
 							       &pyrf_counts_values__type);
 
@@ -914,8 +915,35 @@ static PyObject *pyrf_evsel__read(struct pyrf_evsel *pevsel,
 			     thread);
 		return NULL;
 	}
-	perf_evsel__read(&(evsel->core), cpu_idx, thread_idx, &counts);
-	count_values->values = counts;
+
+	/*
+	 * Ensure evsel's counts and prev_raw_counts are allocated, the latter
+	 * used by tool PMUs to compute the cumulative count as expected by
+	 * stat's process_counter_values.
+	 */
+	if (evsel->counts == NULL) {
+		int nthreads = perf_thread_map__nr(evsel->core.threads);
+		int ncpus = perf_cpu_map__nr(evsel->core.cpus);
+
+		evsel->counts = perf_counts__new(ncpus, nthreads);
+		if (evsel->counts == NULL)
+			return NULL;
+		evsel->prev_raw_counts = perf_counts__new(ncpus, nthreads);
+		if (evsel->prev_raw_counts == NULL)
+			return NULL;
+	}
+	/* Set up pointers to the old and newly read counter values. */
+	old_count = perf_counts(evsel->prev_raw_counts, cpu_idx, thread_idx);
+	new_count = perf_counts(evsel->counts, cpu_idx, thread_idx);
+	/* Update the value in evsel->counts. */
+	evsel__read_counter(evsel, cpu_idx, thread_idx);
+	/* Copy the value and turn it into the delta from old_count. */
+	count_values->values = *new_count;
+	count_values->values.val -= old_count->val;
+	count_values->values.ena -= old_count->ena;
+	count_values->values.run -= old_count->run;
+	/* Save the new count over the old_count for the next read. */
+	*old_count = *new_count;
 	return (PyObject *)count_values;
 }
 
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#6 lore raw

PATCH v2 6/15 perf python: Add basic PMU abstraction and pmus sequence

In reply to Ian Rogers

Add an ability to iterate over PMUs and a basic PMU type then can just show the PMU's name.

An example usage:

```
$ python
Python 3.12.9 (main, Feb  5 2025, 01:31:18) [GCC 14.2.0] on linux
>>> import perf
>>> list(perf.pmus())
[pmu(cpu), pmu(breakpoint), pmu(cstate_core), pmu(cstate_pkg),
pmu(hwmon_acpitz), pmu(hwmon_ac), pmu(hwmon_bat0),
pmu(hwmon_coretemp), pmu(hwmon_iwlwifi_1), pmu(hwmon_nvme),
pmu(hwmon_thinkpad), pmu(hwmon_ucsi_source_psy_usbc000_0),
pmu(hwmon_ucsi_source_psy_usbc000_0), pmu(i915), pmu(intel_bts),
pmu(intel_pt), pmu(kprobe), pmu(msr), pmu(power), pmu(software),
pmu(tool), pmu(tracepoint), pmu(uncore_arb), pmu(uncore_cbox_0),
pmu(uncore_cbox_1), pmu(uncore_cbox_2), pmu(uncore_cbox_3),
pmu(uncore_cbox_4), pmu(uncore_cbox_5), pmu(uncore_cbox_6),
pmu(uncore_cbox_7), pmu(uncore_clock), pmu(uncore_imc_free_running_0),
pmu(uncore_imc_free_running_1), pmu(uprobe)]
```
---
 tools/perf/util/python.c | 140 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 140 insertions(+)
tools/perf/util/python.c +140 -0
index 201fe04fb6dd..33b23d56dfb1 100644
@@ -627,6 +627,138 @@ static int pyrf_thread_map__setup_types(void)
 	return PyType_Ready(&pyrf_thread_map__type);
 }
 
+/**
+ * A python wrapper for perf_pmus that are globally owned by the pmus.c code.
+ */
+struct pyrf_pmu {
+	PyObject_HEAD
+
+	struct perf_pmu *pmu;
+};
+
+static void pyrf_pmu__delete(struct pyrf_pmu *ppmu)
+{
+	Py_TYPE(ppmu)->tp_free((PyObject *)ppmu);
+}
+
+static PyObject *pyrf_pmu__name(PyObject *self)
+{
+	struct pyrf_pmu *ppmu = (void *)self;
+
+	return PyUnicode_FromString(ppmu->pmu->name);
+}
+
+static PyObject *pyrf_pmu__repr(PyObject *self)
+{
+	struct pyrf_pmu *ppmu = (void *)self;
+
+	return PyUnicode_FromFormat("pmu(%s)", ppmu->pmu->name);
+}
+
+static const char pyrf_pmu__doc[] = PyDoc_STR("perf Performance Monitoring Unit (PMU) object.");
+
+static PyMethodDef pyrf_pmu__methods[] = {
+	{
+		.ml_name  = "name",
+		.ml_meth  = (PyCFunction)pyrf_pmu__name,
+		.ml_flags = METH_NOARGS,
+		.ml_doc	  = PyDoc_STR("Name of the PMU including suffixes.")
+	},
+	{ .ml_name = NULL, }
+};
+
+/** The python type for a perf.pmu. */
+static PyTypeObject pyrf_pmu__type = {
+	PyVarObject_HEAD_INIT(NULL, 0)
+	.tp_name	= "perf.pmu",
+	.tp_basicsize	= sizeof(struct pyrf_pmu),
+	.tp_dealloc	= (destructor)pyrf_pmu__delete,
+	.tp_flags	= Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE,
+	.tp_doc		= pyrf_pmu__doc,
+	.tp_methods	= pyrf_pmu__methods,
+	.tp_str         = pyrf_pmu__name,
+	.tp_repr        = pyrf_pmu__repr,
+};
+
+static int pyrf_pmu__setup_types(void)
+{
+	pyrf_pmu__type.tp_new = PyType_GenericNew;
+	return PyType_Ready(&pyrf_pmu__type);
+}
+
+
+/** A python iterator for pmus that has no equivalent in the C code. */
+struct pyrf_pmu_iterator {
+	PyObject_HEAD
+	struct perf_pmu *pmu;
+};
+
+static void pyrf_pmu_iterator__dealloc(struct pyrf_pmu_iterator *self)
+{
+	Py_TYPE(self)->tp_free((PyObject *) self);
+}
+
+static PyObject *pyrf_pmu_iterator__new(PyTypeObject *type, PyObject *args __maybe_unused,
+					PyObject *kwds __maybe_unused)
+{
+	struct pyrf_pmu_iterator *itr = (void *)type->tp_alloc(type, 0);
+
+	if (itr != NULL)
+		itr->pmu = perf_pmus__scan(/*pmu=*/NULL);
+
+	return (PyObject *) itr;
+}
+
+static PyObject *pyrf_pmu_iterator__iter(PyObject *self)
+{
+	Py_INCREF(self);
+	return self;
+}
+
+static PyObject *pyrf_pmu_iterator__iternext(PyObject *self)
+{
+	struct pyrf_pmu_iterator *itr = (void *)self;
+	struct pyrf_pmu *ppmu;
+
+	if (itr->pmu == NULL) {
+		PyErr_SetNone(PyExc_StopIteration);
+		return NULL;
+	}
+	// Create object to return.
+	ppmu = PyObject_New(struct pyrf_pmu, &pyrf_pmu__type);
+	if (ppmu) {
+		ppmu->pmu = itr->pmu;
+		// Advance iterator.
+		itr->pmu = perf_pmus__scan(itr->pmu);
+	}
+	return (PyObject *)ppmu;
+}
+
+/** The python type for the PMU iterator. */
+static PyTypeObject pyrf_pmu_iterator__type = {
+	PyVarObject_HEAD_INIT(NULL, 0)
+	.tp_name = "pmus.iterator",
+	.tp_doc = "Iterator for the pmus string sequence.",
+	.tp_basicsize = sizeof(struct pyrf_pmu_iterator),
+	.tp_itemsize = 0,
+	.tp_flags = Py_TPFLAGS_DEFAULT,
+	.tp_new = pyrf_pmu_iterator__new,
+	.tp_dealloc = (destructor) pyrf_pmu_iterator__dealloc,
+	.tp_iter = pyrf_pmu_iterator__iter,
+	.tp_iternext = pyrf_pmu_iterator__iternext,
+};
+
+static int pyrf_pmu_iterator__setup_types(void)
+{
+	return PyType_Ready(&pyrf_pmu_iterator__type);
+}
+
+static PyObject *pyrf__pmus(PyObject *self, PyObject *args)
+{
+	// Calling the class creates an instance of the iterator.
+	return PyObject_CallObject((PyObject *) &pyrf_pmu_iterator__type, /*args=*/NULL);
+}
+
 struct pyrf_counts_values {
 	PyObject_HEAD
 
@@ -1613,6 +1745,12 @@ static PyMethodDef perf__methods[] = {
 		.ml_flags = METH_VARARGS,
 		.ml_doc	  = PyDoc_STR("Parse a string of events and return an evlist.")
 	},
+	{
+		.ml_name  = "pmus",
+		.ml_meth  = (PyCFunction) pyrf__pmus,
+		.ml_flags = METH_NOARGS,
+		.ml_doc	  = PyDoc_STR("Returns a sequence of pmus.")
+	},
 	{ .ml_name = NULL, }
 };
 
@@ -1640,6 +1778,8 @@ PyMODINIT_FUNC PyInit_perf(void)
 	    pyrf_evsel__setup_types() < 0 ||
 	    pyrf_thread_map__setup_types() < 0 ||
 	    pyrf_cpu_map__setup_types() < 0 ||
+	    pyrf_pmu_iterator__setup_types() < 0 ||
+	    pyrf_pmu__setup_types() < 0 ||
 	    pyrf_counts_values__setup_types() < 0)
 		return module;
 
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#7 lore raw

PATCH v2 7/15 perf python: Add function returning dictionary of all events on a PMU

In reply to Ian Rogers

Allow all events on a PMU to be gathered, similar to how perf list gathers event information.

An example usage:

```
$ python
Python 3.12.9 (main, Feb  5 2025, 01:31:18) [GCC 14.2.0] on linux
>>> import perf
>>> for pmu in perf.pmus():
...   print(pmu.events())
...
[{'name': 'mem_load_retired.l3_hit', 'desc': 'Retired load instructions...
```
---
 tools/perf/util/python.c | 67 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 67 insertions(+)
tools/perf/util/python.c +67 -0
index 33b23d56dfb1..f86a82548636 100644
@@ -648,6 +648,67 @@ static PyObject *pyrf_pmu__name(PyObject *self)
 	return PyUnicode_FromString(ppmu->pmu->name);
 }
 
+static bool add_to_dict(PyObject *dict, const char *key, const char *value)
+{
+	PyObject *pkey, *pvalue;
+	bool ret;
+
+	if (value == NULL)
+		return true;
+
+	pkey = PyUnicode_FromString(key);
+	pvalue = PyUnicode_FromString(value);
+
+	ret = pkey && pvalue && PyDict_SetItem(dict, pkey, pvalue) == 0;
+	Py_XDECREF(pkey);
+	Py_XDECREF(pvalue);
+	return ret;
+}
+
+static int pyrf_pmu__events_cb(void *state, struct pmu_event_info *info)
+{
+	PyObject *py_list = state;
+	PyObject *dict = PyDict_New();
+
+	if (!dict)
+		return -ENOMEM;
+
+	if (!add_to_dict(dict, "name", info->name) ||
+	    !add_to_dict(dict, "alias", info->alias) ||
+	    !add_to_dict(dict, "scale_unit", info->scale_unit) ||
+	    !add_to_dict(dict, "desc", info->desc) ||
+	    !add_to_dict(dict, "long_desc", info->long_desc) ||
+	    !add_to_dict(dict, "encoding_desc", info->encoding_desc) ||
+	    !add_to_dict(dict, "topic", info->topic) ||
+	    !add_to_dict(dict, "event_type_desc", info->event_type_desc) ||
+	    !add_to_dict(dict, "str", info->str) ||
+	    !add_to_dict(dict, "deprecated", info->deprecated ? "deprecated" : NULL) ||
+	    PyList_Append(py_list, dict) != 0) {
+		Py_DECREF(dict);
+		return -ENOMEM;
+	}
+	Py_DECREF(dict);
+	return 0;
+}
+
+static PyObject *pyrf_pmu__events(PyObject *self)
+{
+	struct pyrf_pmu *ppmu = (void *)self;
+	PyObject *py_list = PyList_New(0);
+
+	if (!py_list)
+		return NULL;
+
+	if (perf_pmu__for_each_event(ppmu->pmu,
+				     /*skip_duplicate_pmus=*/false,
+				     py_list,
+				     pyrf_pmu__events_cb) != 0) {
+		Py_DECREF(py_list);
+		return NULL;
+	}
+	return py_list;
+}
+
 static PyObject *pyrf_pmu__repr(PyObject *self)
 {
 	struct pyrf_pmu *ppmu = (void *)self;
@@ -658,6 +719,12 @@ static PyObject *pyrf_pmu__repr(PyObject *self)
 static const char pyrf_pmu__doc[] = PyDoc_STR("perf Performance Monitoring Unit (PMU) object.");
 
 static PyMethodDef pyrf_pmu__methods[] = {
+	{
+		.ml_name  = "events",
+		.ml_meth  = (PyCFunction)pyrf_pmu__events,
+		.ml_flags = METH_NOARGS,
+		.ml_doc	  = PyDoc_STR("Name of the PMU including suffixes.")
+	},
 	{
 		.ml_name  = "name",
 		.ml_meth  = (PyCFunction)pyrf_pmu__name,
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#8 lore raw

PATCH v2 8/15 perf jevents: If the long_desc and desc are identical then drop the long_desc

In reply to Ian Rogers

If the short and long descriptions are the same then save space and don't store both of them. When storing the desc in the perf_pmu_alias, don't duplicate the desc into the long_desc.

By avoiding storing the duplicate the size of the events string in the binary on x86 is reduced by 29,840 bytes.

---
 tools/perf/pmu-events/empty-pmu-events.c | 128 +++++++++++------------
 tools/perf/pmu-events/jevents.py         |   3 +
 tools/perf/util/pmu.c                    |   3 +-
 3 files changed, 68 insertions(+), 66 deletions(-)
tools/perf/pmu-events/empty-pmu-events.c +64 -64
index d4017007a991..a4569a74db07 100644
@@ -40,38 +40,38 @@ static const char *const big_c_string =
 /* offset=1475 */ "dispatch_blocked.any\000other\000Memory cluster signals to block micro-op dispatch for any reason\000event=9,period=200000,umask=0x20\000\00000\000\000\000\000\000"
 /* offset=1608 */ "eist_trans\000other\000Number of Enhanced Intel SpeedStep(R) Technology (EIST) transitions\000event=0x3a,period=200000\000\00000\000\000\000\000\000"
 /* offset=1726 */ "hisi_sccl,ddrc\000"
-/* offset=1741 */ "uncore_hisi_ddrc.flux_wcmd\000uncore\000DDRC write commands\000event=2\000\00000\000\000\000\000DDRC write commands\000"
-/* offset=1830 */ "uncore_cbox\000"
-/* offset=1842 */ "unc_cbo_xsnp_response.miss_eviction\000uncore\000A cross-core snoop resulted from L3 Eviction which misses in some processor core\000event=0x22,umask=0x81\000\00000\000\000\000\000A cross-core snoop resulted from L3 Eviction which misses in some processor core\000"
-/* offset=2076 */ "event-hyphen\000uncore\000UNC_CBO_HYPHEN\000event=0xe0\000\00000\000\000\000\000UNC_CBO_HYPHEN\000"
-/* offset=2144 */ "event-two-hyph\000uncore\000UNC_CBO_TWO_HYPH\000event=0xc0\000\00000\000\000\000\000UNC_CBO_TWO_HYPH\000"
-/* offset=2218 */ "hisi_sccl,l3c\000"
-/* offset=2232 */ "uncore_hisi_l3c.rd_hit_cpipe\000uncore\000Total read hits\000event=7\000\00000\000\000\000\000Total read hits\000"
-/* offset=2315 */ "uncore_imc_free_running\000"
-/* offset=2339 */ "uncore_imc_free_running.cache_miss\000uncore\000Total cache misses\000event=0x12\000\00000\000\000\000\000Total cache misses\000"
-/* offset=2437 */ "uncore_imc\000"
-/* offset=2448 */ "uncore_imc.cache_hits\000uncore\000Total cache hits\000event=0x34\000\00000\000\000\000\000Total cache hits\000"
-/* offset=2529 */ "uncore_sys_ddr_pmu\000"
-/* offset=2548 */ "sys_ddr_pmu.write_cycles\000uncore\000ddr write-cycles event\000event=0x2b\000v8\00000\000\000\000\000\000"
-/* offset=2624 */ "uncore_sys_ccn_pmu\000"
-/* offset=2643 */ "sys_ccn_pmu.read_cycles\000uncore\000ccn read-cycles event\000config=0x2c\0000x01\00000\000\000\000\000\000"
-/* offset=2720 */ "uncore_sys_cmn_pmu\000"
-/* offset=2739 */ "sys_cmn_pmu.hnf_cache_miss\000uncore\000Counts total cache misses in first lookup result (high priority)\000eventid=1,type=5\000(434|436|43c|43a).*\00000\000\000\000\000\000"
-/* offset=2882 */ "CPI\000\0001 / IPC\000\000\000\000\000\000\000\00000"
-/* offset=2904 */ "IPC\000group1\000inst_retired.any / cpu_clk_unhalted.thread\000\000\000\000\000\000\000\00000"
-/* offset=2967 */ "Frontend_Bound_SMT\000\000idq_uops_not_delivered.core / (4 * (cpu_clk_unhalted.thread / 2 * (1 + cpu_clk_unhalted.one_thread_active / cpu_clk_unhalted.ref_xclk)))\000\000\000\000\000\000\000\00000"
-/* offset=3133 */ "dcache_miss_cpi\000\000l1d\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000"
-/* offset=3197 */ "icache_miss_cycles\000\000l1i\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000"
-/* offset=3264 */ "cache_miss_cycles\000group1\000dcache_miss_cpi + icache_miss_cycles\000\000\000\000\000\000\000\00000"
-/* offset=3335 */ "DCache_L2_All_Hits\000\000l2_rqsts.demand_data_rd_hit + l2_rqsts.pf_hit + l2_rqsts.rfo_hit\000\000\000\000\000\000\000\00000"
-/* offset=3429 */ "DCache_L2_All_Miss\000\000max(l2_rqsts.all_demand_data_rd - l2_rqsts.demand_data_rd_hit, 0) + l2_rqsts.pf_miss + l2_rqsts.rfo_miss\000\000\000\000\000\000\000\00000"
-/* offset=3563 */ "DCache_L2_All\000\000DCache_L2_All_Hits + DCache_L2_All_Miss\000\000\000\000\000\000\000\00000"
-/* offset=3627 */ "DCache_L2_Hits\000\000d_ratio(DCache_L2_All_Hits, DCache_L2_All)\000\000\000\000\000\000\000\00000"
-/* offset=3695 */ "DCache_L2_Misses\000\000d_ratio(DCache_L2_All_Miss, DCache_L2_All)\000\000\000\000\000\000\000\00000"
-/* offset=3765 */ "M1\000\000ipc + M2\000\000\000\000\000\000\000\00000"
-/* offset=3787 */ "M2\000\000ipc + M1\000\000\000\000\000\000\000\00000"
-/* offset=3809 */ "M3\000\0001 / M3\000\000\000\000\000\000\000\00000"
-/* offset=3829 */ "L1D_Cache_Fill_BW\000\00064 * l1d.replacement / 1e9 / duration_time\000\000\000\000\000\000\000\00000"
+/* offset=1741 */ "uncore_hisi_ddrc.flux_wcmd\000uncore\000DDRC write commands\000event=2\000\00000\000\000\000\000\000"
+/* offset=1811 */ "uncore_cbox\000"
+/* offset=1823 */ "unc_cbo_xsnp_response.miss_eviction\000uncore\000A cross-core snoop resulted from L3 Eviction which misses in some processor core\000event=0x22,umask=0x81\000\00000\000\000\000\000\000"
+/* offset=1977 */ "event-hyphen\000uncore\000UNC_CBO_HYPHEN\000event=0xe0\000\00000\000\000\000\000\000"
+/* offset=2031 */ "event-two-hyph\000uncore\000UNC_CBO_TWO_HYPH\000event=0xc0\000\00000\000\000\000\000\000"
+/* offset=2089 */ "hisi_sccl,l3c\000"
+/* offset=2103 */ "uncore_hisi_l3c.rd_hit_cpipe\000uncore\000Total read hits\000event=7\000\00000\000\000\000\000\000"
+/* offset=2171 */ "uncore_imc_free_running\000"
+/* offset=2195 */ "uncore_imc_free_running.cache_miss\000uncore\000Total cache misses\000event=0x12\000\00000\000\000\000\000\000"
+/* offset=2275 */ "uncore_imc\000"
+/* offset=2286 */ "uncore_imc.cache_hits\000uncore\000Total cache hits\000event=0x34\000\00000\000\000\000\000\000"
+/* offset=2351 */ "uncore_sys_ddr_pmu\000"
+/* offset=2370 */ "sys_ddr_pmu.write_cycles\000uncore\000ddr write-cycles event\000event=0x2b\000v8\00000\000\000\000\000\000"
+/* offset=2446 */ "uncore_sys_ccn_pmu\000"
+/* offset=2465 */ "sys_ccn_pmu.read_cycles\000uncore\000ccn read-cycles event\000config=0x2c\0000x01\00000\000\000\000\000\000"
+/* offset=2542 */ "uncore_sys_cmn_pmu\000"
+/* offset=2561 */ "sys_cmn_pmu.hnf_cache_miss\000uncore\000Counts total cache misses in first lookup result (high priority)\000eventid=1,type=5\000(434|436|43c|43a).*\00000\000\000\000\000\000"
+/* offset=2704 */ "CPI\000\0001 / IPC\000\000\000\000\000\000\000\00000"
+/* offset=2726 */ "IPC\000group1\000inst_retired.any / cpu_clk_unhalted.thread\000\000\000\000\000\000\000\00000"
+/* offset=2789 */ "Frontend_Bound_SMT\000\000idq_uops_not_delivered.core / (4 * (cpu_clk_unhalted.thread / 2 * (1 + cpu_clk_unhalted.one_thread_active / cpu_clk_unhalted.ref_xclk)))\000\000\000\000\000\000\000\00000"
+/* offset=2955 */ "dcache_miss_cpi\000\000l1d\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000"
+/* offset=3019 */ "icache_miss_cycles\000\000l1i\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000"
+/* offset=3086 */ "cache_miss_cycles\000group1\000dcache_miss_cpi + icache_miss_cycles\000\000\000\000\000\000\000\00000"
+/* offset=3157 */ "DCache_L2_All_Hits\000\000l2_rqsts.demand_data_rd_hit + l2_rqsts.pf_hit + l2_rqsts.rfo_hit\000\000\000\000\000\000\000\00000"
+/* offset=3251 */ "DCache_L2_All_Miss\000\000max(l2_rqsts.all_demand_data_rd - l2_rqsts.demand_data_rd_hit, 0) + l2_rqsts.pf_miss + l2_rqsts.rfo_miss\000\000\000\000\000\000\000\00000"
+/* offset=3385 */ "DCache_L2_All\000\000DCache_L2_All_Hits + DCache_L2_All_Miss\000\000\000\000\000\000\000\00000"
+/* offset=3449 */ "DCache_L2_Hits\000\000d_ratio(DCache_L2_All_Hits, DCache_L2_All)\000\000\000\000\000\000\000\00000"
+/* offset=3517 */ "DCache_L2_Misses\000\000d_ratio(DCache_L2_All_Miss, DCache_L2_All)\000\000\000\000\000\000\000\00000"
+/* offset=3587 */ "M1\000\000ipc + M2\000\000\000\000\000\000\000\00000"
+/* offset=3609 */ "M2\000\000ipc + M1\000\000\000\000\000\000\000\00000"
+/* offset=3631 */ "M3\000\0001 / M3\000\000\000\000\000\000\000\00000"
+/* offset=3651 */ "L1D_Cache_Fill_BW\000\00064 * l1d.replacement / 1e9 / duration_time\000\000\000\000\000\000\000\00000"
 ;
 
 static const struct compact_pmu_event pmu_events__common_tool[] = {
@@ -107,21 +107,21 @@ static const struct compact_pmu_event pmu_events__test_soc_cpu_default_core[] =
 { 1373 }, /* segment_reg_loads.any\000other\000Number of segment register loads\000event=6,period=200000,umask=0x80\000\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_cpu_hisi_sccl_ddrc[] = {
-{ 1741 }, /* uncore_hisi_ddrc.flux_wcmd\000uncore\000DDRC write commands\000event=2\000\00000\000\000\000\000DDRC write commands\000 */
+{ 1741 }, /* uncore_hisi_ddrc.flux_wcmd\000uncore\000DDRC write commands\000event=2\000\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_cpu_hisi_sccl_l3c[] = {
-{ 2232 }, /* uncore_hisi_l3c.rd_hit_cpipe\000uncore\000Total read hits\000event=7\000\00000\000\000\000\000Total read hits\000 */
+{ 2103 }, /* uncore_hisi_l3c.rd_hit_cpipe\000uncore\000Total read hits\000event=7\000\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_cpu_uncore_cbox[] = {
-{ 2076 }, /* event-hyphen\000uncore\000UNC_CBO_HYPHEN\000event=0xe0\000\00000\000\000\000\000UNC_CBO_HYPHEN\000 */
-{ 2144 }, /* event-two-hyph\000uncore\000UNC_CBO_TWO_HYPH\000event=0xc0\000\00000\000\000\000\000UNC_CBO_TWO_HYPH\000 */
-{ 1842 }, /* unc_cbo_xsnp_response.miss_eviction\000uncore\000A cross-core snoop resulted from L3 Eviction which misses in some processor core\000event=0x22,umask=0x81\000\00000\000\000\000\000A cross-core snoop resulted from L3 Eviction which misses in some processor core\000 */
+{ 1977 }, /* event-hyphen\000uncore\000UNC_CBO_HYPHEN\000event=0xe0\000\00000\000\000\000\000\000 */
+{ 2031 }, /* event-two-hyph\000uncore\000UNC_CBO_TWO_HYPH\000event=0xc0\000\00000\000\000\000\000\000 */
+{ 1823 }, /* unc_cbo_xsnp_response.miss_eviction\000uncore\000A cross-core snoop resulted from L3 Eviction which misses in some processor core\000event=0x22,umask=0x81\000\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_cpu_uncore_imc[] = {
-{ 2448 }, /* uncore_imc.cache_hits\000uncore\000Total cache hits\000event=0x34\000\00000\000\000\000\000Total cache hits\000 */
+{ 2286 }, /* uncore_imc.cache_hits\000uncore\000Total cache hits\000event=0x34\000\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_cpu_uncore_imc_free_running[] = {
-{ 2339 }, /* uncore_imc_free_running.cache_miss\000uncore\000Total cache misses\000event=0x12\000\00000\000\000\000\000Total cache misses\000 */
+{ 2195 }, /* uncore_imc_free_running.cache_miss\000uncore\000Total cache misses\000event=0x12\000\00000\000\000\000\000\000 */
 
 };
 
@@ -139,41 +139,41 @@ const struct pmu_table_entry pmu_events__test_soc_cpu[] = {
 {
      .entries = pmu_events__test_soc_cpu_hisi_sccl_l3c,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_cpu_hisi_sccl_l3c),
-     .pmu_name = { 2218 /* hisi_sccl,l3c\000 */ },
+     .pmu_name = { 2089 /* hisi_sccl,l3c\000 */ },
 },
 {
      .entries = pmu_events__test_soc_cpu_uncore_cbox,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_cpu_uncore_cbox),
-     .pmu_name = { 1830 /* uncore_cbox\000 */ },
+     .pmu_name = { 1811 /* uncore_cbox\000 */ },
 },
 {
      .entries = pmu_events__test_soc_cpu_uncore_imc,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_cpu_uncore_imc),
-     .pmu_name = { 2437 /* uncore_imc\000 */ },
+     .pmu_name = { 2275 /* uncore_imc\000 */ },
 },
 {
      .entries = pmu_events__test_soc_cpu_uncore_imc_free_running,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_cpu_uncore_imc_free_running),
-     .pmu_name = { 2315 /* uncore_imc_free_running\000 */ },
+     .pmu_name = { 2171 /* uncore_imc_free_running\000 */ },
 },
 };
 
 static const struct compact_pmu_event pmu_metrics__test_soc_cpu_default_core[] = {
-{ 2882 }, /* CPI\000\0001 / IPC\000\000\000\000\000\000\000\00000 */
-{ 3563 }, /* DCache_L2_All\000\000DCache_L2_All_Hits + DCache_L2_All_Miss\000\000\000\000\000\000\000\00000 */
-{ 3335 }, /* DCache_L2_All_Hits\000\000l2_rqsts.demand_data_rd_hit + l2_rqsts.pf_hit + l2_rqsts.rfo_hit\000\000\000\000\000\000\000\00000 */
-{ 3429 }, /* DCache_L2_All_Miss\000\000max(l2_rqsts.all_demand_data_rd - l2_rqsts.demand_data_rd_hit, 0) + l2_rqsts.pf_miss + l2_rqsts.rfo_miss\000\000\000\000\000\000\000\00000 */
-{ 3627 }, /* DCache_L2_Hits\000\000d_ratio(DCache_L2_All_Hits, DCache_L2_All)\000\000\000\000\000\000\000\00000 */
-{ 3695 }, /* DCache_L2_Misses\000\000d_ratio(DCache_L2_All_Miss, DCache_L2_All)\000\000\000\000\000\000\000\00000 */
-{ 2967 }, /* Frontend_Bound_SMT\000\000idq_uops_not_delivered.core / (4 * (cpu_clk_unhalted.thread / 2 * (1 + cpu_clk_unhalted.one_thread_active / cpu_clk_unhalted.ref_xclk)))\000\000\000\000\000\000\000\00000 */
-{ 2904 }, /* IPC\000group1\000inst_retired.any / cpu_clk_unhalted.thread\000\000\000\000\000\000\000\00000 */
-{ 3829 }, /* L1D_Cache_Fill_BW\000\00064 * l1d.replacement / 1e9 / duration_time\000\000\000\000\000\000\000\00000 */
-{ 3765 }, /* M1\000\000ipc + M2\000\000\000\000\000\000\000\00000 */
-{ 3787 }, /* M2\000\000ipc + M1\000\000\000\000\000\000\000\00000 */
-{ 3809 }, /* M3\000\0001 / M3\000\000\000\000\000\000\000\00000 */
-{ 3264 }, /* cache_miss_cycles\000group1\000dcache_miss_cpi + icache_miss_cycles\000\000\000\000\000\000\000\00000 */
-{ 3133 }, /* dcache_miss_cpi\000\000l1d\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000 */
-{ 3197 }, /* icache_miss_cycles\000\000l1i\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000 */
+{ 2704 }, /* CPI\000\0001 / IPC\000\000\000\000\000\000\000\00000 */
+{ 3385 }, /* DCache_L2_All\000\000DCache_L2_All_Hits + DCache_L2_All_Miss\000\000\000\000\000\000\000\00000 */
+{ 3157 }, /* DCache_L2_All_Hits\000\000l2_rqsts.demand_data_rd_hit + l2_rqsts.pf_hit + l2_rqsts.rfo_hit\000\000\000\000\000\000\000\00000 */
+{ 3251 }, /* DCache_L2_All_Miss\000\000max(l2_rqsts.all_demand_data_rd - l2_rqsts.demand_data_rd_hit, 0) + l2_rqsts.pf_miss + l2_rqsts.rfo_miss\000\000\000\000\000\000\000\00000 */
+{ 3449 }, /* DCache_L2_Hits\000\000d_ratio(DCache_L2_All_Hits, DCache_L2_All)\000\000\000\000\000\000\000\00000 */
+{ 3517 }, /* DCache_L2_Misses\000\000d_ratio(DCache_L2_All_Miss, DCache_L2_All)\000\000\000\000\000\000\000\00000 */
+{ 2789 }, /* Frontend_Bound_SMT\000\000idq_uops_not_delivered.core / (4 * (cpu_clk_unhalted.thread / 2 * (1 + cpu_clk_unhalted.one_thread_active / cpu_clk_unhalted.ref_xclk)))\000\000\000\000\000\000\000\00000 */
+{ 2726 }, /* IPC\000group1\000inst_retired.any / cpu_clk_unhalted.thread\000\000\000\000\000\000\000\00000 */
+{ 3651 }, /* L1D_Cache_Fill_BW\000\00064 * l1d.replacement / 1e9 / duration_time\000\000\000\000\000\000\000\00000 */
+{ 3587 }, /* M1\000\000ipc + M2\000\000\000\000\000\000\000\00000 */
+{ 3609 }, /* M2\000\000ipc + M1\000\000\000\000\000\000\000\00000 */
+{ 3631 }, /* M3\000\0001 / M3\000\000\000\000\000\000\000\00000 */
+{ 3086 }, /* cache_miss_cycles\000group1\000dcache_miss_cpi + icache_miss_cycles\000\000\000\000\000\000\000\00000 */
+{ 2955 }, /* dcache_miss_cpi\000\000l1d\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000 */
+{ 3019 }, /* icache_miss_cycles\000\000l1i\\-loads\\-misses / inst_retired.any\000\000\000\000\000\000\000\00000 */
 
 };
 
@@ -186,13 +186,13 @@ const struct pmu_table_entry pmu_metrics__test_soc_cpu[] = {
 };
 
 static const struct compact_pmu_event pmu_events__test_soc_sys_uncore_sys_ccn_pmu[] = {
-{ 2643 }, /* sys_ccn_pmu.read_cycles\000uncore\000ccn read-cycles event\000config=0x2c\0000x01\00000\000\000\000\000\000 */
+{ 2465 }, /* sys_ccn_pmu.read_cycles\000uncore\000ccn read-cycles event\000config=0x2c\0000x01\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_sys_uncore_sys_cmn_pmu[] = {
-{ 2739 }, /* sys_cmn_pmu.hnf_cache_miss\000uncore\000Counts total cache misses in first lookup result (high priority)\000eventid=1,type=5\000(434|436|43c|43a).*\00000\000\000\000\000\000 */
+{ 2561 }, /* sys_cmn_pmu.hnf_cache_miss\000uncore\000Counts total cache misses in first lookup result (high priority)\000eventid=1,type=5\000(434|436|43c|43a).*\00000\000\000\000\000\000 */
 };
 static const struct compact_pmu_event pmu_events__test_soc_sys_uncore_sys_ddr_pmu[] = {
-{ 2548 }, /* sys_ddr_pmu.write_cycles\000uncore\000ddr write-cycles event\000event=0x2b\000v8\00000\000\000\000\000\000 */
+{ 2370 }, /* sys_ddr_pmu.write_cycles\000uncore\000ddr write-cycles event\000event=0x2b\000v8\00000\000\000\000\000\000 */
 
 };
 
@@ -200,17 +200,17 @@ const struct pmu_table_entry pmu_events__test_soc_sys[] = {
 {
      .entries = pmu_events__test_soc_sys_uncore_sys_ccn_pmu,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_sys_uncore_sys_ccn_pmu),
-     .pmu_name = { 2624 /* uncore_sys_ccn_pmu\000 */ },
+     .pmu_name = { 2446 /* uncore_sys_ccn_pmu\000 */ },
 },
 {
      .entries = pmu_events__test_soc_sys_uncore_sys_cmn_pmu,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_sys_uncore_sys_cmn_pmu),
-     .pmu_name = { 2720 /* uncore_sys_cmn_pmu\000 */ },
+     .pmu_name = { 2542 /* uncore_sys_cmn_pmu\000 */ },
 },
 {
      .entries = pmu_events__test_soc_sys_uncore_sys_ddr_pmu,
      .num_entries = ARRAY_SIZE(pmu_events__test_soc_sys_uncore_sys_ddr_pmu),
-     .pmu_name = { 2529 /* uncore_sys_ddr_pmu\000 */ },
+     .pmu_name = { 2351 /* uncore_sys_ddr_pmu\000 */ },
 },
 };
 
tools/perf/pmu-events/jevents.py +3 -0
index a1899f35ec74..e821155151ec 100755
@@ -397,6 +397,9 @@ class JsonEvent:
       self.desc += extra_desc
     if self.long_desc and extra_desc:
       self.long_desc += extra_desc
+    if self.desc and self.long_desc and self.desc == self.long_desc:
+        # Avoid duplicated descriptions.
+        self.long_desc = None
     if arch_std:
       if arch_std.lower() in _arch_std_events:
         event = _arch_std_events[arch_std.lower()].event
tools/perf/util/pmu.c +1 -2
index 609828513f6c..76b7ded7fbf4 100644
@@ -622,8 +622,7 @@ static int perf_pmu__new_alias(struct perf_pmu *pmu, const char *name,
 
 	alias->name = strdup(name);
 	alias->desc = desc ? strdup(desc) : NULL;
-	alias->long_desc = long_desc ? strdup(long_desc) :
-				desc ? strdup(desc) : NULL;
+	alias->long_desc = long_desc ? strdup(long_desc) : NULL;
 	alias->topic = topic ? strdup(topic) : NULL;
 	alias->pmu_name = pmu_name ? strdup(pmu_name) : NULL;
 	if (unit) {
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#9 lore raw

PATCH v2 9/15 perf jevents: Add common software event json

In reply to Ian Rogers

Add json for software events so that in perf list the events can have a description. Common json exists for the tool PMU but it has no sysfs equivalent. Modify the map_for_pmu code to return the common map (rather than an architecture specific one) when a PMU with a common name is being looked for, this allows the events to be found.

---
 .../arch/common/common/software.json          |  92 ++++++
 tools/perf/pmu-events/empty-pmu-events.c      | 266 +++++++++++-------
 tools/perf/pmu-events/jevents.py              |  15 +-
 3 files changed, 264 insertions(+), 109 deletions(-)
 create mode 100644 tools/perf/pmu-events/arch/common/common/software.json
tools/perf/pmu-events/arch/common/common/software.json new file +92 -0
new file mode 100644
index 000000000000..3af2f565a601
@@ -0,0 +1,92 @@
+[
+  {
+    "Unit": "software",
+    "EventName": "cpu-clock",
+    "BriefDescription": "Per-CPU high-resolution timer based event",
+    "ConfigCode": "0"
+  },
+  {
+    "Unit": "software",
+    "EventName": "task-clock",
+    "BriefDescription": "Task based high-resolution timer based event",
+    "ConfigCode": "1"
+  },
+  {
+    "Unit": "software",
+    "EventName": "faults",
+    "BriefDescription": "Number of page faults [This event is an alias of page-faults]",
+    "ConfigCode": "2"
+  },
+  {
+    "Unit": "software",
+    "EventName": "page-faults",
+    "BriefDescription": "Number of page faults [This event is an alias of faults]",
+    "ConfigCode": "2"
+  },
+  {
+    "Unit": "software",
+    "EventName": "context-switches",
+    "BriefDescription": "Number of context switches [This event is an alias of cs]",
+    "ConfigCode": "3"
+  },
+  {
+    "Unit": "software",
+    "EventName": "cs",
+    "BriefDescription": "Number of context switches [This event is an alias of context-switches]",
+    "ConfigCode": "3"
+  },
+  {
+    "Unit": "software",
+    "EventName": "cpu-migrations",
+    "BriefDescription": "Number of times a process has migrated to a new CPU [This event is an alias of migrations]",
+    "ConfigCode": "4"
+  },
+  {
+    "Unit": "software",
+    "EventName": "migrations",
+    "BriefDescription": "Number of times a process has migrated to a new CPU [This event is an alias of cpu-migrations]",
+    "ConfigCode": "4"
+  },
+  {
+    "Unit": "software",
+    "EventName": "minor-faults",
+    "BriefDescription": "Number of minor page faults. Minor faults don't require I/O to handle",
+    "ConfigCode": "5"
+  },
+  {
+    "Unit": "software",
+    "EventName": "major-faults",
+    "BriefDescription": "Number of major page faults. Major faults require I/O to handle",
+    "ConfigCode": "6"
+  },
+  {
+    "Unit": "software",
+    "EventName": "alignment-faults",
+    "BriefDescription": "Number of kernel handled memory alignment faults",
+    "ConfigCode": "7"
+  },
+  {
+    "Unit": "software",
+    "EventName": "emulation-faults",
+    "BriefDescription": "Number of kernel handled unimplemented instruction faults handled through emulation",
+    "ConfigCode": "8"
+  },
+  {
+    "Unit": "software",
+    "EventName": "dummy",
+    "BriefDescription": "A placeholder event that doesn't count anything",
+    "ConfigCode": "9"
+  },
+  {
+    "Unit": "software",
+    "EventName": "bpf-output",
+    "BriefDescription": "An event used by BPF programs to write to the perf ring buffer",
+    "ConfigCode": "10"
+  },
+  {
+    "Unit": "software",
+    "EventName": "cgroup-switches",
+    "BriefDescription": "Number of context switches to a task in a different cgroup",
+    "ConfigCode": "11"
+  }
+]
tools/perf/pmu-events/empty-pmu-events.c +158 -108

361 lines not on this page · raw message

tools/perf/pmu-events/jevents.py +14 -1
index e821155151ec..76c1e7b0bc22 100755
@@ -295,6 +295,7 @@ class JsonEvent:
           'cpu_atom': 'cpu_atom',
           'ali_drw': 'ali_drw',
           'arm_cmn': 'arm_cmn',
+          'software': 'software',
           'tool': 'tool',
       }
       return table[unit] if unit in table else f'uncore_{unit.lower()}'
@@ -1158,8 +1159,20 @@ static const struct pmu_events_map *map_for_pmu(struct perf_pmu *pmu)
 {
         struct perf_cpu cpu = {-1};
 
-        if (pmu)
+        if (pmu) {
+                for (size_t i=0; i < ARRAY_SIZE(pmu_events__common); i++) {
+                        const char *pmu_name = &big_c_string[pmu_events__common[i].pmu_name.offset];
+
+                        if (!strcmp(pmu_name, pmu->name)) {
+                                const struct pmu_events_map *map = &pmu_events_map[0];
+
+                                while (strcmp("common", map->arch))
+                                        map++;
+                                return map;
+                        }
+                }
                 cpu = perf_cpu_map__min(pmu->cpus);
+        }
         return map_for_cpu(cpu);
 }
 
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#10 lore raw

PATCH v2 10/15 perf pmu: Tolerate failure to read the type for wellknown PMUs

In reply to Ian Rogers

If sysfs isn't mounted then we may fail to read a PMU's type. In this situation resort to lookup of wellknown types. Only applies to software, tracepoint and breakpoint PMUs.

---
 tools/perf/util/pmu.c | 34 ++++++++++++++++++++++++++++++++--
 1 file changed, 32 insertions(+), 2 deletions(-)
tools/perf/util/pmu.c +32 -2
index 76b7ded7fbf4..c0c12880146a 100644
@@ -1180,6 +1180,32 @@ int perf_pmu__init(struct perf_pmu *pmu, __u32 type, const char *name)
 	return 0;
 }
 
+static __u32 wellknown_pmu_type(const char *pmu_name)
+{
+	struct {
+		const char *pmu_name;
+		__u32 type;
+	} wellknown_pmus[] = {
+		{
+			"software",
+			PERF_TYPE_SOFTWARE
+		},
+		{
+			"tracepoint",
+			PERF_TYPE_TRACEPOINT
+		},
+		{
+			"breakpoint",
+			PERF_TYPE_BREAKPOINT
+		},
+	};
+	for (size_t i = 0; i < ARRAY_SIZE(wellknown_pmus); i++) {
+		if (!strcmp(wellknown_pmus[i].pmu_name, pmu_name))
+			return wellknown_pmus[i].type;
+	}
+	return PERF_TYPE_MAX;
+}
+
 struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char *name,
 				  bool eager_load)
 {
@@ -1199,8 +1225,12 @@ struct perf_pmu *perf_pmu__lookup(struct list_head *pmus, int dirfd, const char
 	 * that type value is successfully assigned (return 1).
 	 */
 	if (perf_pmu__scan_file_at(pmu, dirfd, "type", "%u", &pmu->type) != 1) {
-		perf_pmu__delete(pmu);
-		return NULL;
+		/* Double check the PMU's name isn't wellknown. */
+		pmu->type = wellknown_pmu_type(name);
+		if (pmu->type == PERF_TYPE_MAX) {
+			perf_pmu__delete(pmu);
+			return NULL;
+		}
 	}
 
 	/*
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#11 lore raw

PATCH v2 11/15 perf parse-events: Remove non-json software events

In reply to Ian Rogers

Remove the hard coded encodings from parse-events. This has the consequence that software events are matched using the sysfs/json priority, will be case insensitive and will be wildcarded across PMUs. As there were software and hardware types in the parsing code, the removal means software vs hardware logic can be removed and hardware assumed.

Now the perf json provides detailed descriptions of software events, remove the previous listing support that didn't contain event descriptions. When globbing is required for the "sw" option in perf list, use string PMU globbing as was done previously for the tool PMU.

---
 tools/perf/builtin-list.c      | 19 ++++++-------
 tools/perf/util/parse-events.c | 51 ----------------------------------
 tools/perf/util/parse-events.h |  1 -
 tools/perf/util/parse-events.l | 38 +++++++++----------------
 tools/perf/util/parse-events.y | 29 ++++++++-----------
 tools/perf/util/print-events.c |  2 --
 6 files changed, 33 insertions(+), 107 deletions(-)
tools/perf/builtin-list.c +9 -10
index e9b595d75df2..674bb0afbf93 100644
@@ -623,16 +623,17 @@ int cmd_list(int argc, const char **argv)
 		else if (strcmp(argv[i], "sw") == 0 ||
 			 strcmp(argv[i], "software") == 0) {
 			char *old_pmu_glob = default_ps.pmu_glob;
+			static const char * const sw_globs[] = { "software", "tool" };
 
-			print_symbol_events(&print_cb, ps, PERF_TYPE_SOFTWARE,
-					event_symbols_sw, PERF_COUNT_SW_MAX);
-			default_ps.pmu_glob = strdup("tool");
-			if (!default_ps.pmu_glob) {
-				ret = -1;
-				goto out;
+			for (size_t j = 0; j < ARRAY_SIZE(sw_globs); j++) {
+				default_ps.pmu_glob = strdup(sw_globs[j]);
+				if (!default_ps.pmu_glob) {
+					ret = -1;
+					goto out;
+				}
+				perf_pmus__print_pmu_events(&print_cb, ps);
+				zfree(&default_ps.pmu_glob);
 			}
-			perf_pmus__print_pmu_events(&print_cb, ps);
-			zfree(&default_ps.pmu_glob);
 			default_ps.pmu_glob = old_pmu_glob;
 		} else if (strcmp(argv[i], "cache") == 0 ||
 			 strcmp(argv[i], "hwcache") == 0)
@@ -679,8 +680,6 @@ int cmd_list(int argc, const char **argv)
 			default_ps.event_glob = s;
 			print_symbol_events(&print_cb, ps, PERF_TYPE_HARDWARE,
 					event_symbols_hw, PERF_COUNT_HW_MAX);
-			print_symbol_events(&print_cb, ps, PERF_TYPE_SOFTWARE,
-					event_symbols_sw, PERF_COUNT_SW_MAX);
 			print_hwcache_events(&print_cb, ps);
 			perf_pmus__print_pmu_events(&print_cb, ps);
 			print_tracepoint_events(&print_cb, ps);
tools/perf/util/parse-events.c +0 -51

57 lines not on this page · raw message

tools/perf/util/parse-events.h +0 -1
index b47bf2810112..62dc7202e3ba 100644
@@ -264,7 +264,6 @@ struct event_symbol {
 	const char	*alias;
 };
 extern const struct event_symbol event_symbols_hw[];
-extern const struct event_symbol event_symbols_sw[];
 
 char *parse_events_formats_error_string(char *additional_terms);
 
tools/perf/util/parse-events.l +13 -25
index 4af7b9c1f44d..2034590eb789 100644
@@ -117,12 +117,12 @@ do {								\
 	yyless(0);						\
 } while (0)
 
-static int sym(yyscan_t scanner, int type, int config)
+static int sym(yyscan_t scanner, int config)
 {
 	YYSTYPE *yylval = parse_events_get_lval(scanner);
 
-	yylval->num = (type << 16) + config;
-	return type == PERF_TYPE_HARDWARE ? PE_VALUE_SYM_HW : PE_VALUE_SYM_SW;
+	yylval->num = config;
+	return PE_VALUE_SYM_HW;
 }
 
 static int term(yyscan_t scanner, enum parse_events__term_type type)
@@ -391,28 +391,16 @@ r0x{num_raw_hex}	{ return str(yyscanner, PE_RAW); }
 <<EOF>>			{ BEGIN(INITIAL); }
 }
 
-cpu-cycles|cycles				{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES); }
-stalled-cycles-frontend|idle-cycles-frontend	{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_STALLED_CYCLES_FRONTEND); }
-stalled-cycles-backend|idle-cycles-backend	{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_STALLED_CYCLES_BACKEND); }
-instructions					{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS); }
-cache-references				{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_CACHE_REFERENCES); }
-cache-misses					{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_CACHE_MISSES); }
-branch-instructions|branches			{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_BRANCH_INSTRUCTIONS); }
-branch-misses					{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_BRANCH_MISSES); }
-bus-cycles					{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_BUS_CYCLES); }
-ref-cycles					{ return sym(yyscanner, PERF_TYPE_HARDWARE, PERF_COUNT_HW_REF_CPU_CYCLES); }
-cpu-clock					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_CPU_CLOCK); }
-task-clock					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_TASK_CLOCK); }
-page-faults|faults				{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_PAGE_FAULTS); }
-minor-faults					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_PAGE_FAULTS_MIN); }
-major-faults					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_PAGE_FAULTS_MAJ); }
-context-switches|cs				{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_CONTEXT_SWITCHES); }
-cpu-migrations|migrations			{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_CPU_MIGRATIONS); }
-alignment-faults				{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_ALIGNMENT_FAULTS); }
-emulation-faults				{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_EMULATION_FAULTS); }
-dummy						{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); }
-bpf-output					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_BPF_OUTPUT); }
-cgroup-switches					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_CGROUP_SWITCHES); }
+cpu-cycles|cycles				{ return sym(yyscanner, PERF_COUNT_HW_CPU_CYCLES); }
+stalled-cycles-frontend|idle-cycles-frontend	{ return sym(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_FRONTEND); }
+stalled-cycles-backend|idle-cycles-backend	{ return sym(yyscanner, PERF_COUNT_HW_STALLED_CYCLES_BACKEND); }
+instructions					{ return sym(yyscanner, PERF_COUNT_HW_INSTRUCTIONS); }
+cache-references				{ return sym(yyscanner, PERF_COUNT_HW_CACHE_REFERENCES); }
+cache-misses					{ return sym(yyscanner, PERF_COUNT_HW_CACHE_MISSES); }
+branch-instructions|branches			{ return sym(yyscanner, PERF_COUNT_HW_BRANCH_INSTRUCTIONS); }
+branch-misses					{ return sym(yyscanner, PERF_COUNT_HW_BRANCH_MISSES); }
+bus-cycles					{ return sym(yyscanner, PERF_COUNT_HW_BUS_CYCLES); }
+ref-cycles					{ return sym(yyscanner, PERF_COUNT_HW_REF_CPU_CYCLES); }
 
 {lc_type}			{ return str(yyscanner, PE_LEGACY_CACHE); }
 {lc_type}-{lc_op_result}	{ return str(yyscanner, PE_LEGACY_CACHE); }
tools/perf/util/parse-events.y +11 -18
index f888cbb076d6..a2361c0040d7 100644
@@ -55,7 +55,7 @@ static void free_list_evsel(struct list_head* list_evsel)
 %}
 
 %token PE_START_EVENTS PE_START_TERMS
-%token PE_VALUE PE_VALUE_SYM_HW PE_VALUE_SYM_SW PE_TERM
+%token PE_VALUE PE_VALUE_SYM_HW PE_TERM
 %token PE_EVENT_NAME
 %token PE_RAW PE_NAME
 %token PE_MODIFIER_EVENT PE_MODIFIER_BP PE_BP_COLON PE_BP_SLASH
@@ -66,10 +66,8 @@ static void free_list_evsel(struct list_head* list_evsel)
 %token PE_TERM_HW
 %type <num> PE_VALUE
 %type <num> PE_VALUE_SYM_HW
-%type <num> PE_VALUE_SYM_SW
 %type <mod> PE_MODIFIER_EVENT
 %type <term_type> PE_TERM
-%type <num> value_sym
 %type <str> PE_RAW
 %type <str> PE_NAME
 %type <str> PE_LEGACY_CACHE
@@ -306,24 +304,19 @@ PE_NAME sep_dc
 	$$ = list;
 }
 
-value_sym:
-PE_VALUE_SYM_HW
-|
-PE_VALUE_SYM_SW
-
 event_legacy_symbol:
-value_sym '/' event_config '/'
+PE_VALUE_SYM_HW '/' event_config '/'
 {
 	struct list_head *list;
-	int type = $1 >> 16;
-	int config = $1 & 255;
 	int err;
-	bool wildcard = (type == PERF_TYPE_HARDWARE || type == PERF_TYPE_HW_CACHE);
 
 	list = alloc_list();
 	if (!list)
 		YYNOMEM;
-	err = parse_events_add_numeric(_parse_state, list, type, config, $3, wildcard);
+	err = parse_events_add_numeric(_parse_state, list,
+				       PERF_TYPE_HARDWARE, $1,
+				       $3,
+				       /*wildcard=*/true);
 	parse_events_terms__delete($3);
 	if (err) {
 		free_list_evsel(list);
@@ -332,18 +325,18 @@ value_sym '/' event_config '/'
 	$$ = list;
 }
 |
-value_sym sep_slash_slash_dc
+PE_VALUE_SYM_HW sep_slash_slash_dc
 {
 	struct list_head *list;
-	int type = $1 >> 16;
-	int config = $1 & 255;
-	bool wildcard = (type == PERF_TYPE_HARDWARE || type == PERF_TYPE_HW_CACHE);
 	int err;
 
 	list = alloc_list();
 	if (!list)
 		YYNOMEM;
-	err = parse_events_add_numeric(_parse_state, list, type, config, /*head_config=*/NULL, wildcard);
+	err = parse_events_add_numeric(_parse_state, list,
+				       PERF_TYPE_HARDWARE, $1,
+				       /*head_config=*/NULL,
+				       /*wildcard=*/true);
 	if (err)
 		PE_ABORT(err);
 	$$ = list;
tools/perf/util/print-events.c +0 -2
index a786cbfb0ff5..362e93fc33d2 100644
@@ -387,8 +387,6 @@ void print_events(const struct print_callbacks *print_cb, void *print_state)
 {
 	print_symbol_events(print_cb, print_state, PERF_TYPE_HARDWARE,
 			event_symbols_hw, PERF_COUNT_HW_MAX);
-	print_symbol_events(print_cb, print_state, PERF_TYPE_SOFTWARE,
-			event_symbols_sw, PERF_COUNT_SW_MAX);
 
 	print_hwcache_events(print_cb, print_state);
 
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#12 lore raw

PATCH v2 12/15 perf tp_pmu: Factor existing tracepoint logic to new file

In reply to Ian Rogers

Start the creation of a tracepoint PMU abstraction. Tracepoint events don't follow the regular sysfs perf conventions. Eventually the new PMU abstraction will bridge the gap so tracepoint events look more like regular perf ones.

---
 tools/perf/util/Build          |   1 +
 tools/perf/util/evsel.c        |  21 +----
 tools/perf/util/parse-events.c | 147 ++++++++++++++-------------------
 tools/perf/util/tp_pmu.c       |  95 +++++++++++++++++++++
 tools/perf/util/tp_pmu.h       |  12 +++
 5 files changed, 170 insertions(+), 106 deletions(-)
 create mode 100644 tools/perf/util/tp_pmu.c
 create mode 100644 tools/perf/util/tp_pmu.h
tools/perf/util/Build +1 -0
index 7910d908c814..a06ff82f9dd7 100644
@@ -86,6 +86,7 @@ perf-util-y += pmu-flex.o
 perf-util-y += pmu-bison.o
 perf-util-y += hwmon_pmu.o
 perf-util-y += tool_pmu.o
+perf-util-y += tp_pmu.o
 perf-util-y += svghelper.o
 perf-util-y += trace-event-info.o
 perf-util-y += trace-event-scripting.o
tools/perf/util/evsel.c +2 -19
index d55482f094bf..beab9710fbf5 100644
@@ -58,6 +58,7 @@
 #include "pmus.h"
 #include "hwmon_pmu.h"
 #include "tool_pmu.h"
+#include "tp_pmu.h"
 #include "rlimit.h"
 #include "../perf-sys.h"
 #include "util/parse-branch-options.h"
@@ -570,24 +571,6 @@ struct evsel *evsel__clone(struct evsel *dest, struct evsel *orig)
 	return NULL;
 }
 
-static int trace_event__id(const char *sys, const char *name)
-{
-	char *tp_dir = get_events_file(sys);
-	char path[PATH_MAX];
-	int id, err;
-
-	if (!tp_dir)
-		return -1;
-
-	scnprintf(path, PATH_MAX, "%s/%s/id", tp_dir, name);
-	put_events_file(tp_dir);
-	err = filename__read_int(path, &id);
-	if (err)
-		return err;
-
-	return id;
-}
-
 /*
  * Returns pointer with encoded error via <linux/err.h> interface.
  */
@@ -621,7 +604,7 @@ struct evsel *evsel__newtp_idx(const char *sys, const char *name, int idx, bool
 	event_attr_init(&attr);
 
 	if (format) {
-		id = trace_event__id(sys, name);
+		id = tp_pmu__id(sys, name);
 		if (id < 0) {
 			err = id;
 			goto out_free;
tools/perf/util/parse-events.c +60 -87

196 lines not on this page · raw message

tools/perf/util/tp_pmu.c new file +95 -0

95 lines not on this page · raw message

tools/perf/util/tp_pmu.h new file +12 -0
new file mode 100644
index 000000000000..49537303bd73
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
+#ifndef __TP_PMU_H
+#define __TP_PMU_H
+
+typedef int (*tp_sys_callback)(void *state, const char *sys_name);
+typedef int (*tp_event_callback)(void *state, const char *sys_name, const char *evt_name);
+
+int tp_pmu__id(const char *sys, const char *name);
+int tp_pmu__for_each_tp_event(const char *sys, void *state, tp_event_callback cb);
+int tp_pmu__for_each_tp_sys(void *state, tp_sys_callback cb);
+
+#endif /* __TP_PMU_H */
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#13 lore raw

PATCH v2 13/15 perf tp_pmu: Add event APIs

In reply to Ian Rogers

Add event APIs for the tracepoint PMU allowing things like perf list to function using it. For perf list add the tracepoint format in the long description (shown with -v).

---
 tools/perf/util/pmu.c    |   7 +++
 tools/perf/util/tp_pmu.c | 114 +++++++++++++++++++++++++++++++++++++++
 tools/perf/util/tp_pmu.h |   7 +++
 3 files changed, 128 insertions(+)
tools/perf/util/pmu.c +7 -0
index c0c12880146a..2b7671d5bbb4 100644
@@ -23,6 +23,7 @@
 #include "hwmon_pmu.h"
 #include "pmus.h"
 #include "tool_pmu.h"
+#include "tp_pmu.h"
 #include <util/pmu-bison.h>
 #include <util/pmu-flex.h>
 #include "parse-events.h"
@@ -1976,6 +1977,8 @@ bool perf_pmu__have_event(struct perf_pmu *pmu, const char *name)
 		return false;
 	if (perf_pmu__is_tool(pmu) && tool_pmu__skip_event(name))
 		return false;
+	if (perf_pmu__is_tracepoint(pmu))
+		return tp_pmu__have_event(pmu, name);
 	if (perf_pmu__is_hwmon(pmu))
 		return hwmon_pmu__have_event(pmu, name);
 	if (perf_pmu__find_alias(pmu, name, /*load=*/ true) != NULL)
@@ -1989,6 +1992,8 @@ size_t perf_pmu__num_events(struct perf_pmu *pmu)
 {
 	size_t nr;
 
+	if (perf_pmu__is_tracepoint(pmu))
+		return tp_pmu__num_events(pmu);
 	if (perf_pmu__is_hwmon(pmu))
 		return hwmon_pmu__num_events(pmu);
 
@@ -2057,6 +2062,8 @@ int perf_pmu__for_each_event(struct perf_pmu *pmu, bool skip_duplicate_pmus,
 	struct hashmap_entry *entry;
 	size_t bkt;
 
+	if (perf_pmu__is_tracepoint(pmu))
+		return tp_pmu__for_each_event(pmu, state, cb);
 	if (perf_pmu__is_hwmon(pmu))
 		return hwmon_pmu__for_each_event(pmu, state, cb);
 
tools/perf/util/tp_pmu.c +114 -0

122 lines not on this page · raw message

tools/perf/util/tp_pmu.h +7 -0
index 49537303bd73..30456bd6943d 100644
@@ -2,6 +2,8 @@
 #ifndef __TP_PMU_H
 #define __TP_PMU_H
 
+#include "pmu.h"
+
 typedef int (*tp_sys_callback)(void *state, const char *sys_name);
 typedef int (*tp_event_callback)(void *state, const char *sys_name, const char *evt_name);
 
@@ -9,4 +11,9 @@ int tp_pmu__id(const char *sys, const char *name);
 int tp_pmu__for_each_tp_event(const char *sys, void *state, tp_event_callback cb);
 int tp_pmu__for_each_tp_sys(void *state, tp_sys_callback cb);
 
+bool perf_pmu__is_tracepoint(const struct perf_pmu *pmu);
+int tp_pmu__for_each_event(struct perf_pmu *pmu, void *state, pmu_event_callback cb);
+size_t tp_pmu__num_events(struct perf_pmu *pmu);
+bool tp_pmu__have_event(struct perf_pmu *pmu, const char *name);
+
 #endif /* __TP_PMU_H */
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#14 lore raw

PATCH v2 14/15 perf list: Remove tracepoint printing code

In reply to Ian Rogers

Now that the tp_pmu can iterate and describe events remove the custom tracepoint printing logic, this avoids perf list showing the tracepoint events twice.

---
 tools/perf/builtin-list.c      | 28 +++++++---
 tools/perf/util/print-events.c | 93 ----------------------------------
 tools/perf/util/print-events.h |  1 -
 3 files changed, 22 insertions(+), 100 deletions(-)
tools/perf/builtin-list.c +22 -6
index 674bb0afbf93..3216be404f27 100644
@@ -614,9 +614,18 @@ int cmd_list(int argc, const char **argv)
 	for (i = 0; i < argc; ++i) {
 		char *sep, *s;
 
-		if (strcmp(argv[i], "tracepoint") == 0)
-			print_tracepoint_events(&print_cb, ps);
-		else if (strcmp(argv[i], "hw") == 0 ||
+		if (strcmp(argv[i], "tracepoint") == 0) {
+			char *old_pmu_glob = default_ps.pmu_glob;
+
+			default_ps.pmu_glob = strdup("tracepoint");
+			if (!default_ps.pmu_glob) {
+				ret = -1;
+				goto out;
+			}
+			perf_pmus__print_pmu_events(&print_cb, ps);
+			zfree(&default_ps.pmu_glob);
+			default_ps.pmu_glob = old_pmu_glob;
+		} else if (strcmp(argv[i], "hw") == 0 ||
 			 strcmp(argv[i], "hardware") == 0)
 			print_symbol_events(&print_cb, ps, PERF_TYPE_HARDWARE,
 					event_symbols_hw, PERF_COUNT_HW_MAX);
@@ -658,6 +667,7 @@ int cmd_list(int argc, const char **argv)
 #endif
 		else if ((sep = strchr(argv[i], ':')) != NULL) {
 			char *old_pmu_glob = default_ps.pmu_glob;
+			char *old_event_glob = default_ps.event_glob;
 
 			default_ps.event_glob = strdup(argv[i]);
 			if (!default_ps.event_glob) {
@@ -665,13 +675,20 @@ int cmd_list(int argc, const char **argv)
 				goto out;
 			}
 
-			print_tracepoint_events(&print_cb, ps);
+			default_ps.pmu_glob = strdup("tracepoint");
+			if (!default_ps.pmu_glob) {
+				zfree(&default_ps.event_glob);
+				ret = -1;
+				goto out;
+			}
+			perf_pmus__print_pmu_events(&print_cb, ps);
+			default_ps.pmu_glob = old_pmu_glob;
 			print_sdt_events(&print_cb, ps);
 			default_ps.metrics = true;
 			default_ps.metricgroups = true;
 			metricgroup__print(&print_cb, ps);
 			zfree(&default_ps.event_glob);
-			default_ps.pmu_glob = old_pmu_glob;
+			default_ps.event_glob = old_event_glob;
 		} else {
 			if (asprintf(&s, "*%s*", argv[i]) < 0) {
 				printf("Critical: Not enough memory! Trying to continue...\n");
@@ -682,7 +699,6 @@ int cmd_list(int argc, const char **argv)
 					event_symbols_hw, PERF_COUNT_HW_MAX);
 			print_hwcache_events(&print_cb, ps);
 			perf_pmus__print_pmu_events(&print_cb, ps);
-			print_tracepoint_events(&print_cb, ps);
 			print_sdt_events(&print_cb, ps);
 			default_ps.metrics = true;
 			default_ps.metricgroups = true;
tools/perf/util/print-events.c +0 -93

105 lines not on this page · raw message

tools/perf/util/print-events.h +0 -1
index 8f19c2bea64a..02511ed6ecad 100644
@@ -37,7 +37,6 @@ void print_sdt_events(const struct print_callbacks *print_cb, void *print_state)
 void print_symbol_events(const struct print_callbacks *print_cb, void *print_state,
 			 unsigned int type, const struct event_symbol *syms,
 			 unsigned int max);
-void print_tracepoint_events(const struct print_callbacks *print_cb, void *print_state);
 bool is_event_supported(u8 type, u64 config);
 
 #endif /* __PERF_PRINT_EVENTS_H */
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#15 lore raw

PATCH v2 15/15 perf ilist: Add new python ilist command

In reply to Ian Rogers

The perf ilist command is a textual app [1] similar to perf list. In the top-left pane a tree of PMUs is displayed. Selecting a PMU expands the events within it. Selecting an event displays the perf list style event information in the top-right pane.

When an event is selected it is opened and the counters on each CPU the event is for are periodically read. The bottom of the screen contains a scrollable set of sparklines showing the events in total and on each CPU. Scrolling below the sparklines shows the same data as raw counts. The sparklines are small graphs where the height of the bar is in relation to maximum of the other counts in the graph.

By default the counts are read with an interval of 0.1 seconds (10 times per second). A -I/--interval command line option allows the interval to be changed. The oldest read counts are dropped when the counts fill the line causing the sparkline to move from right to left.

[1] https://textual.textualize.io/

---
 tools/perf/python/ilist.py | 238 +++++++++++++++++++++++++++++++++++++
 1 file changed, 238 insertions(+)
 create mode 100755 tools/perf/python/ilist.py
tools/perf/python/ilist.py new file +238 -0
new file mode 100755
index 000000000000..30cc70435f7e
@@ -0,0 +1,238 @@
+#!/usr/bin/env python3
+# SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
+"""Interactive perf list."""
+
+import argparse
+from typing import Dict
+import perf
+from textual import on
+from textual.app import App, ComposeResult
+from textual.binding import Binding
+from textual.containers import Horizontal, HorizontalGroup, Vertical, VerticalScroll
+from textual.screen import ModalScreen
+from textual.widgets import Button, Footer, Header, Label, Sparkline, Static, Tree
+
+class ErrorScreen(ModalScreen[bool]):
+    """Pop up dialog for errors."""
+
+    CSS="""
+    ErrorScreen {
+        align: center middle;
+    }
+    """
+    def __init__(self, error: str):
+        self.error = error
+        super().__init__()
+
+    def compose(self) -> ComposeResult:
+        yield Button(f"Error: {self.error}", variant="primary", id="error")
+
+    def on_button_pressed(self, event: Button.Pressed) -> None:
+        self.dismiss(True)
+
+
+class Counter(HorizontalGroup):
+    """Two labels for a CPU and its counter value."""
+
+    CSS="""
+    Label {
+        gutter: 1;
+    }
+    """
+
+    def __init__(self, cpu: int) -> None:
+        self.cpu = cpu
+        super().__init__()
+
+    def compose(self) -> ComposeResult:
+        label = f"cpu{self.cpu}" if self.cpu >= 0 else "total"
+        yield Label(label + " ")
+        yield Label("0", id=f"counter_{label}")
+
+
+class CounterSparkline(HorizontalGroup):
+    """A Sparkline for a performance counter."""
+
+    def __init__(self, cpu: int) -> None:
+        self.cpu = cpu
+        super().__init__()
+
+    def compose(self) -> ComposeResult:
+        label = f"cpu{self.cpu}" if self.cpu >= 0 else "total"
+        yield Label(label)
+        yield Sparkline([], summary_function=max, id=f"sparkline_{label}")
+
+
+class IListApp(App):
+    TITLE = "Interactive Perf List"
+
+    BINDINGS = [
+        Binding(key="q", action="quit", description="Quit the app")
+    ]
+
+    # Make the 'total' sparkline a different color.
+    CSS = """
+        #sparkline_total > .sparkline--min-color {
+            color: $accent;
+        }
+        #sparkline_total > .sparkline--max-color {
+            color: $accent 30%;
+        }
+    """
+
+    def __init__(self, interval: float) -> None:
+        self.interval = interval
+        self.evlist = None
+        super().__init__()
+
+
+    def update_counts(self) -> None:
+        if not self.evlist:
+            return
+
+        def update_count(cpu: int, count: int):
+            # Update the raw count display.
+            counter: Label = self.query(f"#counter_cpu{cpu}" if cpu >= 0 else "#counter_total")
+            if not counter:
+                return
+            counter = counter.first(Label)
+            counter.update(str(count))
+
+            # Update the sparkline.
+            line: Sparkline = self.query(f"#sparkline_cpu{cpu}" if cpu >= 0 else "#sparkline_total")
+            if not line:
+                return
+            line = line.first(Sparkline)
+            # If there are more events than the width, remove the front event.
+            if len(line.data) > line.size.width:
+                line.data.pop(0)
+            line.data.append(count)
+            line.mutate_reactive(Sparkline.data)
+
+        # Update the total and each CPU counts, assume there's just 1 evsel.
+        total = 0
+        self.evlist.disable()
+        for evsel in self.evlist:
+            for cpu in evsel.cpus():
+                aggr = 0
+                for thread in evsel.threads():
+                    counts = evsel.read(cpu, thread)
+                    aggr += counts.val
+                update_count(cpu, aggr)
+                total += aggr
+        update_count(-1, total)
+        self.evlist.enable()
+
+
+    def on_mount(self) -> None:
+        """When App starts set up periodic event updating."""
+        self.update_counts()
+        self.set_interval(self.interval, self.update_counts)
+
+
+    def set_pmu_and_event(self, pmu: str, event: str) -> None:
+        # Remove previous event information.
+        if self.evlist:
+            self.evlist.disable()
+            self.evlist.close()
+            lines = self.query(CounterSparkline)
+            for line in lines:
+                line.remove()
+            lines = self.query(Counter)
+            for line in lines:
+                line.remove()
+
+        def pmu_event_description(pmu: str, event: str) -> str:
+            """Find and format event description for {pmu}/{event}/."""
+            def get_info(info: Dict[str, str], key: str):
+                return (info[key] + "\n") if key in info else ""
+
+            for p in perf.pmus():
+                if p.name() != pmu:
+                    continue
+                for info in p.events():
+                    if "name" not in info or info["name"] != event:
+                        continue
+
+                    desc = get_info(info, "topic")
+                    desc += get_info(info, "event_type_desc")
+                    desc += get_info(info, "desc")
+                    desc += get_info(info, "long_desc")
+                    desc += get_info(info, "encoding_desc")
+                    return desc
+            return "description"
+
+        # Parse event, update event text and description.
+        full_name = event if event.startswith(pmu) or ':' in event else f"{pmu}/{event}/"
+        self.query_one("#event_name", Label).update(full_name)
+        self.query_one("#event_description", Static).update(pmu_event_description(pmu, event))
+
+        # Open the event.
+        try:
+            self.evlist = perf.parse_events(full_name)
+            if self.evlist:
+                self.evlist.open()
+                self.evlist.enable()
+        except:
+            self.evlist = None
+
+        if not self.evlist:
+            self.push_screen(ErrorScreen(f"Failed to open {full_name}"))
+            return
+
+        # Add spark lines for all the CPUs. Note, must be done after
+        # open so that the evlist CPUs have been computed by propagate
+        # maps.
+        lines = self.query_one("#lines")
+        line = CounterSparkline(cpu=-1)
+        lines.mount(line)
+        for cpu in self.evlist.all_cpus():
+            line = CounterSparkline(cpu)
+            lines.mount(line)
+        line = Counter(cpu=-1)
+        lines.mount(line)
+        for cpu in self.evlist.all_cpus():
+            line = Counter(cpu)
+            lines.mount(line)
+
+
+    def compose(self) -> ComposeResult:
+        def pmu_event_tree() -> Tree:
+            """Create tree of PMUs with events under."""
+            tree: Tree[str] = Tree("PMUs")
+            tree.root.expand()
+            for pmu in perf.pmus():
+                pmu_name = pmu.name()
+                pmu_node = tree.root.add(pmu_name, data=pmu_name)
+                for event in sorted(pmu.events(), key=lambda x: x["name"]):
+                    if "name" in event:
+                        e = event["name"]
+                        if "alias" in event:
+                            pmu_node.add_leaf(f'{e} ({event["alias"]})', data=e)
+                        else:
+                            pmu_node.add_leaf(e, data=e)
+            return tree
+
+        yield Header()
+        yield Horizontal(Vertical(pmu_event_tree(), id="events"),
+                         Vertical(Label("event name", id="event_name"),
+                                  Static("description", markup=False, id="event_description")
+                                  ))
+        yield VerticalScroll(id="lines")
+        yield Footer()
+
+
+    @on(Tree.NodeSelected)
+    def on_tree_node_selected(self, event: Tree.NodeSelected[None]) -> None:
+        if event.node.parent and event.node.parent.parent:
+            assert event.node.parent.data is not None
+            assert event.node.data is not None
+            self.set_pmu_and_event(event.node.parent.data, event.node.data)
+
+
+if __name__ == "__main__":
+    ap = argparse.ArgumentParser()
+    ap.add_argument('-I', '--interval', help="Counter update interval in seconds", default=0.1)
+    args = ap.parse_args()
+    app = IListApp(float(args.interval))
+    app.run()
Signature · 1 line
2.50.0.rc0.642.g800a2b2222-goog
author reviewer
#16 lore raw
In reply to Ian Rogers
Quoted 55 lines · Ian Rogers

This patch series builds up to the addition of a new ilist app written in python using textual [1] for the UI. The app presents perf PMUs and events, displays the event information as in perf list while at the bottom of the console showing recent activity of the event in total and across all CPUs.

The first part of the patches are a few perf and perf python C API fixes, most importantly the counter reading in python supports tool PMUs.

The second part of the patches adds event json for the software PMU and makes the tracepoint PMU support iteration of events and the like. Without these improvements the tracepoint and software PMUs will appear to have no events in the ilist app. As the software PMU moves parsing to json, the legacy hard coded parsing is removed. This has proven controversial for hardware events and so that cleanup isn't done here.

The final patch adds the ilist command. To run it you need the updated perf.cpython.so in your PYTHONPATH and then execute the script. Expanding PMUs and then selecting events will cause event informatin to be displayed in the top-right and the counters values to be displayed as sparklines and counts in the bottom half of the screen.

Some thoughts on the series:

- The PMU changes will conflict with the addition of the DRM PMU:
https://lore.kernel.org/lkml/[email protected]/
when these two are merged together ilist will show yet more counters. It'd be nice if the DRM stuff could land and then I can rebase these patches.

- The parse-events clean up of the software and tracepoint PMU. The software PMU hard coding to be legacy first has similar issues and will conflict with the clean up in:
https://lore.kernel.org/lkml/[email protected]/
Moving the software work to json means we don't need special parse events terms for software events, etc. We can just treat things like regular PMUs with json, etc. I'd much rather we had less special case logic so that series is best rebased on top of this work and it should drop the changes for software terms, etc. which this series removes. Maybe one day the whole event parsing can be much more regular in how PMUs are treated but there's always "cycles".

- Should python libraries have feature tests? How does this get packaged outside the kernel tree? I think these are open questions. Clearly textual is kind of a big dependency and we've largely been moving in the direction of fewer dependencies recently. Hopefully the app makes it clear why I think this one is worth carrying. We carry libslang as a dependency and I think textual clearly far surpasses it.

Quoted 5 lines

- How to launch? Currently I run tools/perf/python/ilist.py but it would be much nicer if we could do perf ilist much as we do for perf-archive.sh. There are probably other scripts that should be perf commands like flamegraph and gecko. It'd be nice to follow up the series with something to make using these commands easy.

Some more thoughts:
- The number of events can be quite extensive and so a search UI
would be a good addition. I'd rather it be follow up work.
- The legacy events for hardware and hardware cache lack json, PMU
alias data, .. and so only partially show up. Ugh, legacy events.
- A dialog is shown when an event fails to open, it should probably
mention more things related to perf permissions, run as root, etc. More follow up work imo.
- Showing metrics would be great. Follow up work.
- Textual web (https://github.com/Textualize/textual-web) integration
so you can see the output in a browser would be great. Follow up work.
- The sparklines are against the max in the counts recently read,
maybe this should be something smarter like the maximum possible, or over time, or over all CPUs.. Follow up work.
- People likely have thoughts on the way the UI is, like should the
counts be after the spark lines, etc. I think with the python being only 238 lines (stat-display.c and stat-shadow.c are over >2000 lines) things should be pretty hackable and we can work on things over time. The patch is something of a proof-of-concept for that reason, but it is also something useful, hence posting it.

Thanks,
Ian

Quoted 52 lines

[1] https://textual.textualize.io/

v2: In the jevents event description duplication, some minor changes
accidentally missed from v1 meaning that in v1 the descriptions were still duplicated. Expand the cover letter with some thoughts on the series.

Ian Rogers (15):
perf hwmon_pmu: Avoid shortening hwmon PMU name
perf parse-events: Minor tidy up of event_type helper
perf python: In str(evsel) use the evsel__pmu_name helper perf python: Fix thread check in pyrf_evsel__read
perf python: Correct pyrf_evsel__read for tool PMUs
perf python: Add basic PMU abstraction and pmus sequence perf python: Add function returning dictionary of all events on a PMU perf jevents: If the long_desc and desc are identical then drop the
long_desc
perf jevents: Add common software event json
perf pmu: Tolerate failure to read the type for wellknown PMUs perf parse-events: Remove non-json software events
perf tp_pmu: Factor existing tracepoint logic to new file perf tp_pmu: Add event APIs
perf list: Remove tracepoint printing code
perf ilist: Add new python ilist command

 tools/perf/builtin-list.c                     |  47 ++--
 .../arch/common/common/software.json          |  92 ++++++
 tools/perf/pmu-events/empty-pmu-events.c      | 266 +++++++++++-------
 tools/perf/pmu-events/jevents.py              |  18 +-
 tools/perf/python/ilist.py                    | 238 ++++++++++++++++
 tools/perf/util/Build                         |   1 +
 tools/perf/util/evsel.c                       |  21 +-
 tools/perf/util/hwmon_pmu.c                   |   2 +-
 tools/perf/util/parse-events.c                | 225 +++++----------
 tools/perf/util/parse-events.h                |   3 +-
 tools/perf/util/parse-events.l                |  38 +--
 tools/perf/util/parse-events.y                |  29 +-
 tools/perf/util/pmu.c                         |  44 ++-
 tools/perf/util/print-events.c                |  95 -------
 tools/perf/util/print-events.h                |   1 -
 tools/perf/util/python.c                      | 248 +++++++++++++++-
 tools/perf/util/tp_pmu.c                      | 209 ++++++++++++++
 tools/perf/util/tp_pmu.h                      |  19 ++
 18 files changed, 1144 insertions(+), 452 deletions(-)
 create mode 100644 tools/perf/pmu-events/arch/common/common/software.json
 create mode 100755 tools/perf/python/ilist.py
 create mode 100644 tools/perf/util/tp_pmu.c
 create mode 100644 tools/perf/util/tp_pmu.h

--
2.50.0.rc0.642.g800a2b2222-goog