xref: /linux/tools/perf/Documentation/perf-list.txt (revision e96fddb32931d007db12b1fce9b5e8e4c080401b)
1perf-list(1)
2============
3
4NAME
5----
6perf-list - List all symbolic event types
7
8SYNOPSIS
9--------
10[verse]
11'perf list' [--no-desc] [--long-desc]
12            [hw|sw|cache|tracepoint|pmu|sdt|metric|metricgroup|event_glob]
13
14DESCRIPTION
15-----------
16This command displays the symbolic event types which can be selected in the
17various perf commands with the -e option.
18
19OPTIONS
20-------
21-d::
22--desc::
23Print extra event descriptions. (default)
24
25--no-desc::
26Don't print descriptions.
27
28-v::
29--long-desc::
30Print longer event descriptions.
31
32--debug::
33Enable debugging output.
34
35--details::
36Print how named events are resolved internally into perf events, and also
37any extra expressions computed by perf stat.
38
39--deprecated::
40Print deprecated events. By default the deprecated events are hidden.
41
42--unit::
43Print PMU events and metrics limited to the specific PMU name.
44(e.g. --unit cpu, --unit msr, --unit cpu_core, --unit cpu_atom)
45
46-j::
47--json::
48Output in JSON format.
49
50-o::
51--output=::
52	Output file name. By default output is written to stdout.
53
54[[EVENT_MODIFIERS]]
55EVENT MODIFIERS
56---------------
57
58Events can optionally have a modifier by appending a colon and one or
59more modifiers. Modifiers allow the user to restrict the events to be
60counted. The following modifiers exist:
61
62 u - user-space counting
63 k - kernel counting
64 h - hypervisor counting
65 I - non idle counting
66 G - guest counting (in KVM guests)
67 H - host counting (not in KVM guests)
68 p - precise level
69 P - use maximum detected precise level
70 S - read sample value (PERF_SAMPLE_READ)
71 D - pin the event to the PMU
72 W - group is weak and will fallback to non-group if not schedulable,
73 e - group or event are exclusive and do not share the PMU
74
75The 'p' modifier can be used for specifying how precise the instruction
76address should be. The 'p' modifier can be specified multiple times:
77
78 0 - SAMPLE_IP can have arbitrary skid
79 1 - SAMPLE_IP must have constant skid
80 2 - SAMPLE_IP requested to have 0 skid
81 3 - SAMPLE_IP must have 0 skid, or uses randomization to avoid
82     sample shadowing effects.
83
84For Intel systems precise event sampling is implemented with PEBS
85which supports up to precise-level 2, and precise level 3 for
86some special cases
87
88On AMD systems it is implemented using IBS OP (up to precise-level 2).
89Unlike Intel PEBS which provides levels of precision, AMD core pmu is
90inherently non-precise and IBS is inherently precise. (i.e. ibs_op//,
91ibs_op//p, ibs_op//pp and ibs_op//ppp are all same). The precise modifier
92works with event types 0x76 (cpu-cycles, CPU clocks not halted) and 0xC1
93(micro-ops retired). Both events map to IBS execution sampling (IBS op)
94with the IBS Op Counter Control bit (IbsOpCntCtl) set respectively (see the
95Core Complex (CCX) -> Processor x86 Core -> Instruction Based Sampling (IBS)
96section of the [AMD Processor Programming Reference (PPR)] relevant to the
97family, model and stepping of the processor being used).
98
99Manual Volume 2: System Programming, 13.3 Instruction-Based
100Sampling). Examples to use IBS:
101
102 perf record -a -e cpu-cycles:p ...    # use ibs op counting cycles
103 perf record -a -e r076:p ...          # same as -e cpu-cycles:p
104 perf record -a -e r0C1:p ...          # use ibs op counting micro-ops
105
106RAW HARDWARE EVENT DESCRIPTOR
107-----------------------------
108Even when an event is not available in a symbolic form within perf right now,
109it can be encoded in a per processor specific way.
110
111For instance on x86 CPUs, N is a hexadecimal value that represents the raw register encoding with the
112layout of IA32_PERFEVTSELx MSRs (see [Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide] Figure 30-1 Layout
113of IA32_PERFEVTSELx MSRs) or AMD's PERF_CTL MSRs (see the
114Core Complex (CCX) -> Processor x86 Core -> MSR Registers section of the
115[AMD Processor Programming Reference (PPR)] relevant to the family, model
116and stepping of the processor being used).
117
118Note: Only the following bit fields can be set in x86 counter
119registers: event, umask, edge, inv, cmask. Esp. guest/host only and
120OS/user mode flags must be setup using <<EVENT_MODIFIERS, EVENT
121MODIFIERS>>.
122
123Example:
124
125If the Intel docs for a QM720 Core i7 describe an event as:
126
127  Event  Umask  Event Mask
128  Num.   Value  Mnemonic    Description                        Comment
129
130  A8H      01H  LSD.UOPS    Counts the number of micro-ops     Use cmask=1 and
131                            delivered by loop stream detector  invert to count
132                                                               cycles
133
134raw encoding of 0x1A8 can be used:
135
136 perf stat -e r1a8 -a sleep 1
137 perf record -e r1a8 ...
138
139It's also possible to use pmu syntax:
140
141 perf record -e r1a8 -a sleep 1
142 perf record -e cpu/r1a8/ ...
143 perf record -e cpu/r0x1a8/ ...
144
145Some processors, like those from AMD, support event codes and unit masks
146larger than a byte. In such cases, the bits corresponding to the event
147configuration parameters can be seen with:
148
149  cat /sys/bus/event_source/devices/<pmu>/format/<config>
150
151Example:
152
153If the AMD docs for an EPYC 7713 processor describe an event as:
154
155  Event  Umask  Event Mask
156  Num.   Value  Mnemonic                        Description
157
158  28FH     03H  op_cache_hit_miss.op_cache_hit  Counts Op Cache micro-tag
159                                                hit events.
160
161raw encoding of 0x0328F cannot be used since the upper nibble of the
162EventSelect bits have to be specified via bits 32-35 as can be seen with:
163
164  cat /sys/bus/event_source/devices/cpu/format/event
165
166raw encoding of 0x20000038F should be used instead:
167
168 perf stat -e r20000038f -a sleep 1
169 perf record -e r20000038f ...
170
171It's also possible to use pmu syntax:
172
173 perf record -e r20000038f -a sleep 1
174 perf record -e cpu/r20000038f/ ...
175 perf record -e cpu/r0x20000038f/ ...
176
177You should refer to the processor specific documentation for getting these
178details. Some of them are referenced in the SEE ALSO section below.
179
180ARBITRARY PMUS
181--------------
182
183perf also supports an extended syntax for specifying raw parameters
184to PMUs. Using this typically requires looking up the specific event
185in the CPU vendor specific documentation.
186
187The available PMUs and their raw parameters can be listed with
188
189  ls /sys/devices/*/format
190
191For example the raw event "LSD.UOPS" core pmu event above could
192be specified as
193
194  perf stat -e cpu/event=0xa8,umask=0x1,name=LSD.UOPS_CYCLES,cmask=0x1/ ...
195
196  or using extended name syntax
197
198  perf stat -e cpu/event=0xa8,umask=0x1,cmask=0x1,name=\'LSD.UOPS_CYCLES:cmask=0x1\'/ ...
199
200PER SOCKET PMUS
201---------------
202
203Some PMUs are not associated with a core, but with a whole CPU socket.
204Events on these PMUs generally cannot be sampled, but only counted globally
205with perf stat -a. They can be bound to one logical CPU, but will measure
206all the CPUs in the same socket.
207
208This example measures memory bandwidth every second
209on the first memory controller on socket 0 of a Intel Xeon system
210
211  perf stat -C 0 -a uncore_imc_0/cas_count_read/,uncore_imc_0/cas_count_write/ -I 1000 ...
212
213Each memory controller has its own PMU.  Measuring the complete system
214bandwidth would require specifying all imc PMUs (see perf list output),
215and adding the values together. To simplify creation of multiple events,
216prefix and glob matching is supported in the PMU name, and the prefix
217'uncore_' is also ignored when performing the match. So the command above
218can be expanded to all memory controllers by using the syntaxes:
219
220  perf stat -C 0 -a imc/cas_count_read/,imc/cas_count_write/ -I 1000 ...
221  perf stat -C 0 -a *imc*/cas_count_read/,*imc*/cas_count_write/ -I 1000 ...
222
223This example measures the combined core power every second
224
225  perf stat -I 1000 -e power/energy-cores/  -a
226
227ACCESS RESTRICTIONS
228-------------------
229
230For non root users generally only context switched PMU events are available.
231This is normally only the events in the cpu PMU, the predefined events
232like cycles and instructions and some software events.
233
234Other PMUs and global measurements are normally root only.
235Some event qualifiers, such as "any", are also root only.
236
237This can be overridden by setting the kernel.perf_event_paranoid
238sysctl to -1, which allows non root to use these events.
239
240For accessing trace point events perf needs to have read access to
241/sys/kernel/tracing, even when perf_event_paranoid is in a relaxed
242setting.
243
244TRACING
245-------
246
247Some PMUs control advanced hardware tracing capabilities, such as Intel PT,
248that allows low overhead execution tracing.  These are described in a separate
249intel-pt.txt document.
250
251PARAMETERIZED EVENTS
252--------------------
253
254Some pmu events listed by 'perf-list' will be displayed with '?' in them. For
255example:
256
257  hv_gpci/dtbp_ptitc,phys_processor_idx=?/
258
259This means that when provided as an event, a value for '?' must
260also be supplied. For example:
261
262  perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
263
264EVENT QUALIFIERS:
265
266It is also possible to add extra qualifiers to an event:
267
268percore:
269
270Sums up the event counts for all hardware threads in a core, e.g.:
271
272
273  perf stat -e cpu/event=0,umask=0x3,percore=1/
274
275
276EVENT GROUPS
277------------
278
279Perf supports time based multiplexing of events, when the number of events
280active exceeds the number of hardware performance counters. Multiplexing
281can cause measurement errors when the workload changes its execution
282profile.
283
284When metrics are computed using formulas from event counts, it is useful to
285ensure some events are always measured together as a group to minimize multiplexing
286errors. Event groups can be specified using { }.
287
288  perf stat -e '{instructions,cycles}' ...
289
290The number of available performance counters depend on the CPU. A group
291cannot contain more events than available counters.
292For example Intel Core CPUs typically have four generic performance counters
293for the core, plus three fixed counters for instructions, cycles and
294ref-cycles. Some special events have restrictions on which counter they
295can schedule, and may not support multiple instances in a single group.
296When too many events are specified in the group some of them will not
297be measured.
298
299Globally pinned events can limit the number of counters available for
300other groups. On x86 systems, the NMI watchdog pins a counter by default.
301The nmi watchdog can be disabled as root with
302
303	echo 0 > /proc/sys/kernel/nmi_watchdog
304
305Events from multiple different PMUs cannot be mixed in a group, with
306some exceptions for software events.
307
308LEADER SAMPLING
309---------------
310
311perf also supports group leader sampling using the :S specifier.
312
313  perf record -e '{cycles,instructions}:S' ...
314  perf report --group
315
316Normally all events in an event group sample, but with :S only
317the first event (the leader) samples, and it only reads the values of the
318other events in the group.
319
320However, in the case AUX area events (e.g. Intel PT or CoreSight), the AUX
321area event must be the leader, so then the second event samples, not the first.
322
323OPTIONS
324-------
325
326Without options all known events will be listed.
327
328To limit the list use:
329
330. 'hw' or 'hardware' to list hardware events such as cache-misses, etc.
331
332. 'sw' or 'software' to list software events such as context switches, etc.
333
334. 'cache' or 'hwcache' to list hardware cache events such as L1-dcache-loads, etc.
335
336. 'tracepoint' to list all tracepoint events, alternatively use
337  'subsys_glob:event_glob' to filter by tracepoint subsystems such as sched,
338  block, etc.
339
340. 'pmu' to print the kernel supplied PMU events.
341
342. 'sdt' to list all Statically Defined Tracepoint events.
343
344. 'metric' to list metrics
345
346. 'metricgroup' to list metricgroups with metrics.
347
348. If none of the above is matched, it will apply the supplied glob to all
349  events, printing the ones that match.
350
351. As a last resort, it will do a substring search in all event names.
352
353One or more types can be used at the same time, listing the events for the
354types specified.
355
356Support raw format:
357
358. '--raw-dump', shows the raw-dump of all the events.
359. '--raw-dump [hw|sw|cache|tracepoint|pmu|event_glob]', shows the raw-dump of
360  a certain kind of events.
361
362SEE ALSO
363--------
364linkperf:perf-stat[1], linkperf:perf-top[1],
365linkperf:perf-record[1],
366http://www.intel.com/sdm/[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide],
367https://bugzilla.kernel.org/show_bug.cgi?id=206537[AMD Processor Programming Reference (PPR)]
368