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[[EVENT_MODIFIERS]]
43EVENT MODIFIERS
44---------------
45
46Events can optionally have a modifier by appending a colon and one or
47more modifiers. Modifiers allow the user to restrict the events to be
48counted. The following modifiers exist:
49
50 u - user-space counting
51 k - kernel counting
52 h - hypervisor counting
53 I - non idle counting
54 G - guest counting (in KVM guests)
55 H - host counting (not in KVM guests)
56 p - precise level
57 P - use maximum detected precise level
58 S - read sample value (PERF_SAMPLE_READ)
59 D - pin the event to the PMU
60 W - group is weak and will fallback to non-group if not schedulable,
61
62The 'p' modifier can be used for specifying how precise the instruction
63address should be. The 'p' modifier can be specified multiple times:
64
65 0 - SAMPLE_IP can have arbitrary skid
66 1 - SAMPLE_IP must have constant skid
67 2 - SAMPLE_IP requested to have 0 skid
68 3 - SAMPLE_IP must have 0 skid, or uses randomization to avoid
69     sample shadowing effects.
70
71For Intel systems precise event sampling is implemented with PEBS
72which supports up to precise-level 2, and precise level 3 for
73some special cases
74
75On AMD systems it is implemented using IBS (up to precise-level 2).
76The precise modifier works with event types 0x76 (cpu-cycles, CPU
77clocks not halted) and 0xC1 (micro-ops retired). Both events map to
78IBS execution sampling (IBS op) with the IBS Op Counter Control bit
79(IbsOpCntCtl) set respectively (see AMD64 Architecture Programmer’s
80Manual Volume 2: System Programming, 13.3 Instruction-Based
81Sampling). Examples to use IBS:
82
83 perf record -a -e cpu-cycles:p ...    # use ibs op counting cycles
84 perf record -a -e r076:p ...          # same as -e cpu-cycles:p
85 perf record -a -e r0C1:p ...          # use ibs op counting micro-ops
86
87RAW HARDWARE EVENT DESCRIPTOR
88-----------------------------
89Even when an event is not available in a symbolic form within perf right now,
90it can be encoded in a per processor specific way.
91
92For instance For x86 CPUs NNN represents the raw register encoding with the
93layout of IA32_PERFEVTSELx MSRs (see [Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide] Figure 30-1 Layout
94of IA32_PERFEVTSELx MSRs) or AMD's PerfEvtSeln (see [AMD64 Architecture Programmer’s Manual Volume 2: System Programming], Page 344,
95Figure 13-7 Performance Event-Select Register (PerfEvtSeln)).
96
97Note: Only the following bit fields can be set in x86 counter
98registers: event, umask, edge, inv, cmask. Esp. guest/host only and
99OS/user mode flags must be setup using <<EVENT_MODIFIERS, EVENT
100MODIFIERS>>.
101
102Example:
103
104If the Intel docs for a QM720 Core i7 describe an event as:
105
106  Event  Umask  Event Mask
107  Num.   Value  Mnemonic    Description                        Comment
108
109  A8H      01H  LSD.UOPS    Counts the number of micro-ops     Use cmask=1 and
110                            delivered by loop stream detector  invert to count
111                                                               cycles
112
113raw encoding of 0x1A8 can be used:
114
115 perf stat -e r1a8 -a sleep 1
116 perf record -e r1a8 ...
117
118It's also possible to use pmu syntax:
119
120 perf record -e r1a8 -a sleep 1
121 perf record -e cpu/r1a8/ ...
122 perf record -e cpu/r0x1a8/ ...
123
124You should refer to the processor specific documentation for getting these
125details. Some of them are referenced in the SEE ALSO section below.
126
127ARBITRARY PMUS
128--------------
129
130perf also supports an extended syntax for specifying raw parameters
131to PMUs. Using this typically requires looking up the specific event
132in the CPU vendor specific documentation.
133
134The available PMUs and their raw parameters can be listed with
135
136  ls /sys/devices/*/format
137
138For example the raw event "LSD.UOPS" core pmu event above could
139be specified as
140
141  perf stat -e cpu/event=0xa8,umask=0x1,name=LSD.UOPS_CYCLES,cmask=0x1/ ...
142
143  or using extended name syntax
144
145  perf stat -e cpu/event=0xa8,umask=0x1,cmask=0x1,name=\'LSD.UOPS_CYCLES:cmask=0x1\'/ ...
146
147PER SOCKET PMUS
148---------------
149
150Some PMUs are not associated with a core, but with a whole CPU socket.
151Events on these PMUs generally cannot be sampled, but only counted globally
152with perf stat -a. They can be bound to one logical CPU, but will measure
153all the CPUs in the same socket.
154
155This example measures memory bandwidth every second
156on the first memory controller on socket 0 of a Intel Xeon system
157
158  perf stat -C 0 -a uncore_imc_0/cas_count_read/,uncore_imc_0/cas_count_write/ -I 1000 ...
159
160Each memory controller has its own PMU.  Measuring the complete system
161bandwidth would require specifying all imc PMUs (see perf list output),
162and adding the values together. To simplify creation of multiple events,
163prefix and glob matching is supported in the PMU name, and the prefix
164'uncore_' is also ignored when performing the match. So the command above
165can be expanded to all memory controllers by using the syntaxes:
166
167  perf stat -C 0 -a imc/cas_count_read/,imc/cas_count_write/ -I 1000 ...
168  perf stat -C 0 -a *imc*/cas_count_read/,*imc*/cas_count_write/ -I 1000 ...
169
170This example measures the combined core power every second
171
172  perf stat -I 1000 -e power/energy-cores/  -a
173
174ACCESS RESTRICTIONS
175-------------------
176
177For non root users generally only context switched PMU events are available.
178This is normally only the events in the cpu PMU, the predefined events
179like cycles and instructions and some software events.
180
181Other PMUs and global measurements are normally root only.
182Some event qualifiers, such as "any", are also root only.
183
184This can be overridden by setting the kernel.perf_event_paranoid
185sysctl to -1, which allows non root to use these events.
186
187For accessing trace point events perf needs to have read access to
188/sys/kernel/debug/tracing, even when perf_event_paranoid is in a relaxed
189setting.
190
191TRACING
192-------
193
194Some PMUs control advanced hardware tracing capabilities, such as Intel PT,
195that allows low overhead execution tracing.  These are described in a separate
196intel-pt.txt document.
197
198PARAMETERIZED EVENTS
199--------------------
200
201Some pmu events listed by 'perf-list' will be displayed with '?' in them. For
202example:
203
204  hv_gpci/dtbp_ptitc,phys_processor_idx=?/
205
206This means that when provided as an event, a value for '?' must
207also be supplied. For example:
208
209  perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
210
211EVENT QUALIFIERS:
212
213It is also possible to add extra qualifiers to an event:
214
215percore:
216
217Sums up the event counts for all hardware threads in a core, e.g.:
218
219
220  perf stat -e cpu/event=0,umask=0x3,percore=1/
221
222
223EVENT GROUPS
224------------
225
226Perf supports time based multiplexing of events, when the number of events
227active exceeds the number of hardware performance counters. Multiplexing
228can cause measurement errors when the workload changes its execution
229profile.
230
231When metrics are computed using formulas from event counts, it is useful to
232ensure some events are always measured together as a group to minimize multiplexing
233errors. Event groups can be specified using { }.
234
235  perf stat -e '{instructions,cycles}' ...
236
237The number of available performance counters depend on the CPU. A group
238cannot contain more events than available counters.
239For example Intel Core CPUs typically have four generic performance counters
240for the core, plus three fixed counters for instructions, cycles and
241ref-cycles. Some special events have restrictions on which counter they
242can schedule, and may not support multiple instances in a single group.
243When too many events are specified in the group some of them will not
244be measured.
245
246Globally pinned events can limit the number of counters available for
247other groups. On x86 systems, the NMI watchdog pins a counter by default.
248The nmi watchdog can be disabled as root with
249
250	echo 0 > /proc/sys/kernel/nmi_watchdog
251
252Events from multiple different PMUs cannot be mixed in a group, with
253some exceptions for software events.
254
255LEADER SAMPLING
256---------------
257
258perf also supports group leader sampling using the :S specifier.
259
260  perf record -e '{cycles,instructions}:S' ...
261  perf report --group
262
263Normally all events in an event group sample, but with :S only
264the first event (the leader) samples, and it only reads the values of the
265other events in the group.
266
267However, in the case AUX area events (e.g. Intel PT or CoreSight), the AUX
268area event must be the leader, so then the second event samples, not the first.
269
270OPTIONS
271-------
272
273Without options all known events will be listed.
274
275To limit the list use:
276
277. 'hw' or 'hardware' to list hardware events such as cache-misses, etc.
278
279. 'sw' or 'software' to list software events such as context switches, etc.
280
281. 'cache' or 'hwcache' to list hardware cache events such as L1-dcache-loads, etc.
282
283. 'tracepoint' to list all tracepoint events, alternatively use
284  'subsys_glob:event_glob' to filter by tracepoint subsystems such as sched,
285  block, etc.
286
287. 'pmu' to print the kernel supplied PMU events.
288
289. 'sdt' to list all Statically Defined Tracepoint events.
290
291. 'metric' to list metrics
292
293. 'metricgroup' to list metricgroups with metrics.
294
295. If none of the above is matched, it will apply the supplied glob to all
296  events, printing the ones that match.
297
298. As a last resort, it will do a substring search in all event names.
299
300One or more types can be used at the same time, listing the events for the
301types specified.
302
303Support raw format:
304
305. '--raw-dump', shows the raw-dump of all the events.
306. '--raw-dump [hw|sw|cache|tracepoint|pmu|event_glob]', shows the raw-dump of
307  a certain kind of events.
308
309SEE ALSO
310--------
311linkperf:perf-stat[1], linkperf:perf-top[1],
312linkperf:perf-record[1],
313http://www.intel.com/sdm/[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide],
314http://support.amd.com/us/Processor_TechDocs/24593_APM_v2.pdf[AMD64 Architecture Programmer’s Manual Volume 2: System Programming]
315