6 perf-list - List all symbolic event types
11 'perf list' [--no-desc] [--long-desc]
12 [hw|sw|cache|tracepoint|pmu|sdt|metric|metricgroup|event_glob]
16 This command displays the symbolic event types which can be selected in the
17 various perf commands with the -e option.
23 Print extra event descriptions. (default)
26 Don't print descriptions.
30 Print longer event descriptions.
33 Enable debugging output.
36 Print how named events are resolved internally into perf events, and also
37 any extra expressions computed by perf stat.
40 Print deprecated events. By default the deprecated events are hidden.
46 Events can optionally have a modifier by appending a colon and one or
47 more modifiers. Modifiers allow the user to restrict the events to be
48 counted. The following modifiers exist:
50 u - user-space counting
52 h - hypervisor counting
54 G - guest counting (in KVM guests)
55 H - host counting (not in KVM guests)
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 e - group or event are exclusive and do not share the PMU
63 The 'p' modifier can be used for specifying how precise the instruction
64 address should be. The 'p' modifier can be specified multiple times:
66 0 - SAMPLE_IP can have arbitrary skid
67 1 - SAMPLE_IP must have constant skid
68 2 - SAMPLE_IP requested to have 0 skid
69 3 - SAMPLE_IP must have 0 skid, or uses randomization to avoid
70 sample shadowing effects.
72 For Intel systems precise event sampling is implemented with PEBS
73 which supports up to precise-level 2, and precise level 3 for
76 On AMD systems it is implemented using IBS (up to precise-level 2).
77 The precise modifier works with event types 0x76 (cpu-cycles, CPU
78 clocks not halted) and 0xC1 (micro-ops retired). Both events map to
79 IBS execution sampling (IBS op) with the IBS Op Counter Control bit
80 (IbsOpCntCtl) set respectively (see AMD64 Architecture Programmer’s
81 Manual Volume 2: System Programming, 13.3 Instruction-Based
82 Sampling). Examples to use IBS:
84 perf record -a -e cpu-cycles:p ... # use ibs op counting cycles
85 perf record -a -e r076:p ... # same as -e cpu-cycles:p
86 perf record -a -e r0C1:p ... # use ibs op counting micro-ops
88 RAW HARDWARE EVENT DESCRIPTOR
89 -----------------------------
90 Even when an event is not available in a symbolic form within perf right now,
91 it can be encoded in a per processor specific way.
93 For instance For x86 CPUs NNN represents the raw register encoding with the
94 layout of IA32_PERFEVTSELx MSRs (see [Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide] Figure 30-1 Layout
95 of IA32_PERFEVTSELx MSRs) or AMD's PerfEvtSeln (see [AMD64 Architecture Programmer’s Manual Volume 2: System Programming], Page 344,
96 Figure 13-7 Performance Event-Select Register (PerfEvtSeln)).
98 Note: Only the following bit fields can be set in x86 counter
99 registers: event, umask, edge, inv, cmask. Esp. guest/host only and
100 OS/user mode flags must be setup using <<EVENT_MODIFIERS, EVENT
105 If the Intel docs for a QM720 Core i7 describe an event as:
107 Event Umask Event Mask
108 Num. Value Mnemonic Description Comment
110 A8H 01H LSD.UOPS Counts the number of micro-ops Use cmask=1 and
111 delivered by loop stream detector invert to count
114 raw encoding of 0x1A8 can be used:
116 perf stat -e r1a8 -a sleep 1
117 perf record -e r1a8 ...
119 It's also possible to use pmu syntax:
121 perf record -e r1a8 -a sleep 1
122 perf record -e cpu/r1a8/ ...
123 perf record -e cpu/r0x1a8/ ...
125 You should refer to the processor specific documentation for getting these
126 details. Some of them are referenced in the SEE ALSO section below.
131 perf also supports an extended syntax for specifying raw parameters
132 to PMUs. Using this typically requires looking up the specific event
133 in the CPU vendor specific documentation.
135 The available PMUs and their raw parameters can be listed with
137 ls /sys/devices/*/format
139 For example the raw event "LSD.UOPS" core pmu event above could
142 perf stat -e cpu/event=0xa8,umask=0x1,name=LSD.UOPS_CYCLES,cmask=0x1/ ...
144 or using extended name syntax
146 perf stat -e cpu/event=0xa8,umask=0x1,cmask=0x1,name=\'LSD.UOPS_CYCLES:cmask=0x1\'/ ...
151 Some PMUs are not associated with a core, but with a whole CPU socket.
152 Events on these PMUs generally cannot be sampled, but only counted globally
153 with perf stat -a. They can be bound to one logical CPU, but will measure
154 all the CPUs in the same socket.
156 This example measures memory bandwidth every second
157 on the first memory controller on socket 0 of a Intel Xeon system
159 perf stat -C 0 -a uncore_imc_0/cas_count_read/,uncore_imc_0/cas_count_write/ -I 1000 ...
161 Each memory controller has its own PMU. Measuring the complete system
162 bandwidth would require specifying all imc PMUs (see perf list output),
163 and adding the values together. To simplify creation of multiple events,
164 prefix and glob matching is supported in the PMU name, and the prefix
165 'uncore_' is also ignored when performing the match. So the command above
166 can be expanded to all memory controllers by using the syntaxes:
168 perf stat -C 0 -a imc/cas_count_read/,imc/cas_count_write/ -I 1000 ...
169 perf stat -C 0 -a *imc*/cas_count_read/,*imc*/cas_count_write/ -I 1000 ...
171 This example measures the combined core power every second
173 perf stat -I 1000 -e power/energy-cores/ -a
178 For non root users generally only context switched PMU events are available.
179 This is normally only the events in the cpu PMU, the predefined events
180 like cycles and instructions and some software events.
182 Other PMUs and global measurements are normally root only.
183 Some event qualifiers, such as "any", are also root only.
185 This can be overridden by setting the kernel.perf_event_paranoid
186 sysctl to -1, which allows non root to use these events.
188 For accessing trace point events perf needs to have read access to
189 /sys/kernel/debug/tracing, even when perf_event_paranoid is in a relaxed
195 Some PMUs control advanced hardware tracing capabilities, such as Intel PT,
196 that allows low overhead execution tracing. These are described in a separate
197 intel-pt.txt document.
202 Some pmu events listed by 'perf-list' will be displayed with '?' in them. For
205 hv_gpci/dtbp_ptitc,phys_processor_idx=?/
207 This means that when provided as an event, a value for '?' must
208 also be supplied. For example:
210 perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
214 It is also possible to add extra qualifiers to an event:
218 Sums up the event counts for all hardware threads in a core, e.g.:
221 perf stat -e cpu/event=0,umask=0x3,percore=1/
227 Perf supports time based multiplexing of events, when the number of events
228 active exceeds the number of hardware performance counters. Multiplexing
229 can cause measurement errors when the workload changes its execution
232 When metrics are computed using formulas from event counts, it is useful to
233 ensure some events are always measured together as a group to minimize multiplexing
234 errors. Event groups can be specified using { }.
236 perf stat -e '{instructions,cycles}' ...
238 The number of available performance counters depend on the CPU. A group
239 cannot contain more events than available counters.
240 For example Intel Core CPUs typically have four generic performance counters
241 for the core, plus three fixed counters for instructions, cycles and
242 ref-cycles. Some special events have restrictions on which counter they
243 can schedule, and may not support multiple instances in a single group.
244 When too many events are specified in the group some of them will not
247 Globally pinned events can limit the number of counters available for
248 other groups. On x86 systems, the NMI watchdog pins a counter by default.
249 The nmi watchdog can be disabled as root with
251 echo 0 > /proc/sys/kernel/nmi_watchdog
253 Events from multiple different PMUs cannot be mixed in a group, with
254 some exceptions for software events.
259 perf also supports group leader sampling using the :S specifier.
261 perf record -e '{cycles,instructions}:S' ...
264 Normally all events in an event group sample, but with :S only
265 the first event (the leader) samples, and it only reads the values of the
266 other events in the group.
268 However, in the case AUX area events (e.g. Intel PT or CoreSight), the AUX
269 area event must be the leader, so then the second event samples, not the first.
274 Without options all known events will be listed.
276 To limit the list use:
278 . 'hw' or 'hardware' to list hardware events such as cache-misses, etc.
280 . 'sw' or 'software' to list software events such as context switches, etc.
282 . 'cache' or 'hwcache' to list hardware cache events such as L1-dcache-loads, etc.
284 . 'tracepoint' to list all tracepoint events, alternatively use
285 'subsys_glob:event_glob' to filter by tracepoint subsystems such as sched,
288 . 'pmu' to print the kernel supplied PMU events.
290 . 'sdt' to list all Statically Defined Tracepoint events.
292 . 'metric' to list metrics
294 . 'metricgroup' to list metricgroups with metrics.
296 . If none of the above is matched, it will apply the supplied glob to all
297 events, printing the ones that match.
299 . As a last resort, it will do a substring search in all event names.
301 One or more types can be used at the same time, listing the events for the
306 . '--raw-dump', shows the raw-dump of all the events.
307 . '--raw-dump [hw|sw|cache|tracepoint|pmu|event_glob]', shows the raw-dump of
308 a certain kind of events.
312 linkperf:perf-stat[1], linkperf:perf-top[1],
313 linkperf:perf-record[1],
314 http://www.intel.com/sdm/[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide],
315 http://support.amd.com/us/Processor_TechDocs/24593_APM_v2.pdf[AMD64 Architecture Programmer’s Manual Volume 2: System Programming]