xref: /linux/tools/perf/Documentation/perf-list.txt (revision c4bbe83d27c2446a033cc0381c3fb6be5e8c41c7)
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[[EVENT_MODIFIERS]]
51EVENT MODIFIERS
52---------------
53
54Events can optionally have a modifier by appending a colon and one or
55more modifiers. Modifiers allow the user to restrict the events to be
56counted. The following modifiers exist:
57
58 u - user-space counting
59 k - kernel counting
60 h - hypervisor counting
61 I - non idle counting
62 G - guest counting (in KVM guests)
63 H - host counting (not in KVM guests)
64 p - precise level
65 P - use maximum detected precise level
66 S - read sample value (PERF_SAMPLE_READ)
67 D - pin the event to the PMU
68 W - group is weak and will fallback to non-group if not schedulable,
69 e - group or event are exclusive and do not share the PMU
70
71The 'p' modifier can be used for specifying how precise the instruction
72address should be. The 'p' modifier can be specified multiple times:
73
74 0 - SAMPLE_IP can have arbitrary skid
75 1 - SAMPLE_IP must have constant skid
76 2 - SAMPLE_IP requested to have 0 skid
77 3 - SAMPLE_IP must have 0 skid, or uses randomization to avoid
78     sample shadowing effects.
79
80For Intel systems precise event sampling is implemented with PEBS
81which supports up to precise-level 2, and precise level 3 for
82some special cases
83
84On AMD systems it is implemented using IBS OP (up to precise-level 2).
85Unlike Intel PEBS which provides levels of precision, AMD core pmu is
86inherently non-precise and IBS is inherently precise. (i.e. ibs_op//,
87ibs_op//p, ibs_op//pp and ibs_op//ppp are all same). The precise modifier
88works with event types 0x76 (cpu-cycles, CPU clocks not halted) and 0xC1
89(micro-ops retired). Both events map to IBS execution sampling (IBS op)
90with the IBS Op Counter Control bit (IbsOpCntCtl) set respectively (see the
91Core Complex (CCX) -> Processor x86 Core -> Instruction Based Sampling (IBS)
92section of the [AMD Processor Programming Reference (PPR)] relevant to the
93family, model and stepping of the processor being used).
94
95Manual Volume 2: System Programming, 13.3 Instruction-Based
96Sampling). Examples to use IBS:
97
98 perf record -a -e cpu-cycles:p ...    # use ibs op counting cycles
99 perf record -a -e r076:p ...          # same as -e cpu-cycles:p
100 perf record -a -e r0C1:p ...          # use ibs op counting micro-ops
101
102RAW HARDWARE EVENT DESCRIPTOR
103-----------------------------
104Even when an event is not available in a symbolic form within perf right now,
105it can be encoded in a per processor specific way.
106
107For instance on x86 CPUs, N is a hexadecimal value that represents the raw register encoding with the
108layout of IA32_PERFEVTSELx MSRs (see [Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide] Figure 30-1 Layout
109of IA32_PERFEVTSELx MSRs) or AMD's PERF_CTL MSRs (see the
110Core Complex (CCX) -> Processor x86 Core -> MSR Registers section of the
111[AMD Processor Programming Reference (PPR)] relevant to the family, model
112and stepping of the processor being used).
113
114Note: Only the following bit fields can be set in x86 counter
115registers: event, umask, edge, inv, cmask. Esp. guest/host only and
116OS/user mode flags must be setup using <<EVENT_MODIFIERS, EVENT
117MODIFIERS>>.
118
119Example:
120
121If the Intel docs for a QM720 Core i7 describe an event as:
122
123  Event  Umask  Event Mask
124  Num.   Value  Mnemonic    Description                        Comment
125
126  A8H      01H  LSD.UOPS    Counts the number of micro-ops     Use cmask=1 and
127                            delivered by loop stream detector  invert to count
128                                                               cycles
129
130raw encoding of 0x1A8 can be used:
131
132 perf stat -e r1a8 -a sleep 1
133 perf record -e r1a8 ...
134
135It's also possible to use pmu syntax:
136
137 perf record -e r1a8 -a sleep 1
138 perf record -e cpu/r1a8/ ...
139 perf record -e cpu/r0x1a8/ ...
140
141Some processors, like those from AMD, support event codes and unit masks
142larger than a byte. In such cases, the bits corresponding to the event
143configuration parameters can be seen with:
144
145  cat /sys/bus/event_source/devices/<pmu>/format/<config>
146
147Example:
148
149If the AMD docs for an EPYC 7713 processor describe an event as:
150
151  Event  Umask  Event Mask
152  Num.   Value  Mnemonic                        Description
153
154  28FH     03H  op_cache_hit_miss.op_cache_hit  Counts Op Cache micro-tag
155                                                hit events.
156
157raw encoding of 0x0328F cannot be used since the upper nibble of the
158EventSelect bits have to be specified via bits 32-35 as can be seen with:
159
160  cat /sys/bus/event_source/devices/cpu/format/event
161
162raw encoding of 0x20000038F should be used instead:
163
164 perf stat -e r20000038f -a sleep 1
165 perf record -e r20000038f ...
166
167It's also possible to use pmu syntax:
168
169 perf record -e r20000038f -a sleep 1
170 perf record -e cpu/r20000038f/ ...
171 perf record -e cpu/r0x20000038f/ ...
172
173You should refer to the processor specific documentation for getting these
174details. Some of them are referenced in the SEE ALSO section below.
175
176ARBITRARY PMUS
177--------------
178
179perf also supports an extended syntax for specifying raw parameters
180to PMUs. Using this typically requires looking up the specific event
181in the CPU vendor specific documentation.
182
183The available PMUs and their raw parameters can be listed with
184
185  ls /sys/devices/*/format
186
187For example the raw event "LSD.UOPS" core pmu event above could
188be specified as
189
190  perf stat -e cpu/event=0xa8,umask=0x1,name=LSD.UOPS_CYCLES,cmask=0x1/ ...
191
192  or using extended name syntax
193
194  perf stat -e cpu/event=0xa8,umask=0x1,cmask=0x1,name=\'LSD.UOPS_CYCLES:cmask=0x1\'/ ...
195
196PER SOCKET PMUS
197---------------
198
199Some PMUs are not associated with a core, but with a whole CPU socket.
200Events on these PMUs generally cannot be sampled, but only counted globally
201with perf stat -a. They can be bound to one logical CPU, but will measure
202all the CPUs in the same socket.
203
204This example measures memory bandwidth every second
205on the first memory controller on socket 0 of a Intel Xeon system
206
207  perf stat -C 0 -a uncore_imc_0/cas_count_read/,uncore_imc_0/cas_count_write/ -I 1000 ...
208
209Each memory controller has its own PMU.  Measuring the complete system
210bandwidth would require specifying all imc PMUs (see perf list output),
211and adding the values together. To simplify creation of multiple events,
212prefix and glob matching is supported in the PMU name, and the prefix
213'uncore_' is also ignored when performing the match. So the command above
214can be expanded to all memory controllers by using the syntaxes:
215
216  perf stat -C 0 -a imc/cas_count_read/,imc/cas_count_write/ -I 1000 ...
217  perf stat -C 0 -a *imc*/cas_count_read/,*imc*/cas_count_write/ -I 1000 ...
218
219This example measures the combined core power every second
220
221  perf stat -I 1000 -e power/energy-cores/  -a
222
223ACCESS RESTRICTIONS
224-------------------
225
226For non root users generally only context switched PMU events are available.
227This is normally only the events in the cpu PMU, the predefined events
228like cycles and instructions and some software events.
229
230Other PMUs and global measurements are normally root only.
231Some event qualifiers, such as "any", are also root only.
232
233This can be overridden by setting the kernel.perf_event_paranoid
234sysctl to -1, which allows non root to use these events.
235
236For accessing trace point events perf needs to have read access to
237/sys/kernel/tracing, even when perf_event_paranoid is in a relaxed
238setting.
239
240TRACING
241-------
242
243Some PMUs control advanced hardware tracing capabilities, such as Intel PT,
244that allows low overhead execution tracing.  These are described in a separate
245intel-pt.txt document.
246
247PARAMETERIZED EVENTS
248--------------------
249
250Some pmu events listed by 'perf-list' will be displayed with '?' in them. For
251example:
252
253  hv_gpci/dtbp_ptitc,phys_processor_idx=?/
254
255This means that when provided as an event, a value for '?' must
256also be supplied. For example:
257
258  perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
259
260EVENT QUALIFIERS:
261
262It is also possible to add extra qualifiers to an event:
263
264percore:
265
266Sums up the event counts for all hardware threads in a core, e.g.:
267
268
269  perf stat -e cpu/event=0,umask=0x3,percore=1/
270
271
272EVENT GROUPS
273------------
274
275Perf supports time based multiplexing of events, when the number of events
276active exceeds the number of hardware performance counters. Multiplexing
277can cause measurement errors when the workload changes its execution
278profile.
279
280When metrics are computed using formulas from event counts, it is useful to
281ensure some events are always measured together as a group to minimize multiplexing
282errors. Event groups can be specified using { }.
283
284  perf stat -e '{instructions,cycles}' ...
285
286The number of available performance counters depend on the CPU. A group
287cannot contain more events than available counters.
288For example Intel Core CPUs typically have four generic performance counters
289for the core, plus three fixed counters for instructions, cycles and
290ref-cycles. Some special events have restrictions on which counter they
291can schedule, and may not support multiple instances in a single group.
292When too many events are specified in the group some of them will not
293be measured.
294
295Globally pinned events can limit the number of counters available for
296other groups. On x86 systems, the NMI watchdog pins a counter by default.
297The nmi watchdog can be disabled as root with
298
299	echo 0 > /proc/sys/kernel/nmi_watchdog
300
301Events from multiple different PMUs cannot be mixed in a group, with
302some exceptions for software events.
303
304LEADER SAMPLING
305---------------
306
307perf also supports group leader sampling using the :S specifier.
308
309  perf record -e '{cycles,instructions}:S' ...
310  perf report --group
311
312Normally all events in an event group sample, but with :S only
313the first event (the leader) samples, and it only reads the values of the
314other events in the group.
315
316However, in the case AUX area events (e.g. Intel PT or CoreSight), the AUX
317area event must be the leader, so then the second event samples, not the first.
318
319OPTIONS
320-------
321
322Without options all known events will be listed.
323
324To limit the list use:
325
326. 'hw' or 'hardware' to list hardware events such as cache-misses, etc.
327
328. 'sw' or 'software' to list software events such as context switches, etc.
329
330. 'cache' or 'hwcache' to list hardware cache events such as L1-dcache-loads, etc.
331
332. 'tracepoint' to list all tracepoint events, alternatively use
333  'subsys_glob:event_glob' to filter by tracepoint subsystems such as sched,
334  block, etc.
335
336. 'pmu' to print the kernel supplied PMU events.
337
338. 'sdt' to list all Statically Defined Tracepoint events.
339
340. 'metric' to list metrics
341
342. 'metricgroup' to list metricgroups with metrics.
343
344. If none of the above is matched, it will apply the supplied glob to all
345  events, printing the ones that match.
346
347. As a last resort, it will do a substring search in all event names.
348
349One or more types can be used at the same time, listing the events for the
350types specified.
351
352Support raw format:
353
354. '--raw-dump', shows the raw-dump of all the events.
355. '--raw-dump [hw|sw|cache|tracepoint|pmu|event_glob]', shows the raw-dump of
356  a certain kind of events.
357
358SEE ALSO
359--------
360linkperf:perf-stat[1], linkperf:perf-top[1],
361linkperf:perf-record[1],
362http://www.intel.com/sdm/[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide],
363https://bugzilla.kernel.org/show_bug.cgi?id=206537[AMD Processor Programming Reference (PPR)]
364