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.
43 Events can optionally have a modifier by appending a colon and one or
44 more modifiers. Modifiers allow the user to restrict the events to be
45 counted. The following modifiers exist:
47 u - user-space counting
49 h - hypervisor counting
51 G - guest counting (in KVM guests)
52 H - host counting (not in KVM guests)
54 P - use maximum detected precise level
55 S - read sample value (PERF_SAMPLE_READ)
56 D - pin the event to the PMU
57 W - group is weak and will fallback to non-group if not schedulable,
58 only supported in 'perf stat' for now.
60 The 'p' modifier can be used for specifying how precise the instruction
61 address should be. The 'p' modifier can be specified multiple times:
63 0 - SAMPLE_IP can have arbitrary skid
64 1 - SAMPLE_IP must have constant skid
65 2 - SAMPLE_IP requested to have 0 skid
66 3 - SAMPLE_IP must have 0 skid, or uses randomization to avoid
67 sample shadowing effects.
69 For Intel systems precise event sampling is implemented with PEBS
70 which supports up to precise-level 2, and precise level 3 for
73 On AMD systems it is implemented using IBS (up to precise-level 2).
74 The precise modifier works with event types 0x76 (cpu-cycles, CPU
75 clocks not halted) and 0xC1 (micro-ops retired). Both events map to
76 IBS execution sampling (IBS op) with the IBS Op Counter Control bit
77 (IbsOpCntCtl) set respectively (see AMD64 Architecture Programmer’s
78 Manual Volume 2: System Programming, 13.3 Instruction-Based
79 Sampling). Examples to use IBS:
81 perf record -a -e cpu-cycles:p ... # use ibs op counting cycles
82 perf record -a -e r076:p ... # same as -e cpu-cycles:p
83 perf record -a -e r0C1:p ... # use ibs op counting micro-ops
85 RAW HARDWARE EVENT DESCRIPTOR
86 -----------------------------
87 Even when an event is not available in a symbolic form within perf right now,
88 it can be encoded in a per processor specific way.
90 For instance For x86 CPUs NNN represents the raw register encoding with the
91 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
92 of IA32_PERFEVTSELx MSRs) or AMD's PerfEvtSeln (see [AMD64 Architecture Programmer’s Manual Volume 2: System Programming], Page 344,
93 Figure 13-7 Performance Event-Select Register (PerfEvtSeln)).
95 Note: Only the following bit fields can be set in x86 counter
96 registers: event, umask, edge, inv, cmask. Esp. guest/host only and
97 OS/user mode flags must be setup using <<EVENT_MODIFIERS, EVENT
102 If the Intel docs for a QM720 Core i7 describe an event as:
104 Event Umask Event Mask
105 Num. Value Mnemonic Description Comment
107 A8H 01H LSD.UOPS Counts the number of micro-ops Use cmask=1 and
108 delivered by loop stream detector invert to count
111 raw encoding of 0x1A8 can be used:
113 perf stat -e r1a8 -a sleep 1
114 perf record -e r1a8 ...
116 You should refer to the processor specific documentation for getting these
117 details. Some of them are referenced in the SEE ALSO section below.
122 perf also supports an extended syntax for specifying raw parameters
123 to PMUs. Using this typically requires looking up the specific event
124 in the CPU vendor specific documentation.
126 The available PMUs and their raw parameters can be listed with
128 ls /sys/devices/*/format
130 For example the raw event "LSD.UOPS" core pmu event above could
133 perf stat -e cpu/event=0xa8,umask=0x1,name=LSD.UOPS_CYCLES,cmask=0x1/ ...
135 or using extended name syntax
137 perf stat -e cpu/event=0xa8,umask=0x1,cmask=0x1,name=\'LSD.UOPS_CYCLES:cmask=0x1\'/ ...
142 Some PMUs are not associated with a core, but with a whole CPU socket.
143 Events on these PMUs generally cannot be sampled, but only counted globally
144 with perf stat -a. They can be bound to one logical CPU, but will measure
145 all the CPUs in the same socket.
147 This example measures memory bandwidth every second
148 on the first memory controller on socket 0 of a Intel Xeon system
150 perf stat -C 0 -a uncore_imc_0/cas_count_read/,uncore_imc_0/cas_count_write/ -I 1000 ...
152 Each memory controller has its own PMU. Measuring the complete system
153 bandwidth would require specifying all imc PMUs (see perf list output),
154 and adding the values together. To simplify creation of multiple events,
155 prefix and glob matching is supported in the PMU name, and the prefix
156 'uncore_' is also ignored when performing the match. So the command above
157 can be expanded to all memory controllers by using the syntaxes:
159 perf stat -C 0 -a imc/cas_count_read/,imc/cas_count_write/ -I 1000 ...
160 perf stat -C 0 -a *imc*/cas_count_read/,*imc*/cas_count_write/ -I 1000 ...
162 This example measures the combined core power every second
164 perf stat -I 1000 -e power/energy-cores/ -a
169 For non root users generally only context switched PMU events are available.
170 This is normally only the events in the cpu PMU, the predefined events
171 like cycles and instructions and some software events.
173 Other PMUs and global measurements are normally root only.
174 Some event qualifiers, such as "any", are also root only.
176 This can be overriden by setting the kernel.perf_event_paranoid
177 sysctl to -1, which allows non root to use these events.
179 For accessing trace point events perf needs to have read access to
180 /sys/kernel/debug/tracing, even when perf_event_paranoid is in a relaxed
186 Some PMUs control advanced hardware tracing capabilities, such as Intel PT,
187 that allows low overhead execution tracing. These are described in a separate
188 intel-pt.txt document.
193 Some pmu events listed by 'perf-list' will be displayed with '?' in them. For
196 hv_gpci/dtbp_ptitc,phys_processor_idx=?/
198 This means that when provided as an event, a value for '?' must
199 also be supplied. For example:
201 perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
206 Perf supports time based multiplexing of events, when the number of events
207 active exceeds the number of hardware performance counters. Multiplexing
208 can cause measurement errors when the workload changes its execution
211 When metrics are computed using formulas from event counts, it is useful to
212 ensure some events are always measured together as a group to minimize multiplexing
213 errors. Event groups can be specified using { }.
215 perf stat -e '{instructions,cycles}' ...
217 The number of available performance counters depend on the CPU. A group
218 cannot contain more events than available counters.
219 For example Intel Core CPUs typically have four generic performance counters
220 for the core, plus three fixed counters for instructions, cycles and
221 ref-cycles. Some special events have restrictions on which counter they
222 can schedule, and may not support multiple instances in a single group.
223 When too many events are specified in the group some of them will not
226 Globally pinned events can limit the number of counters available for
227 other groups. On x86 systems, the NMI watchdog pins a counter by default.
228 The nmi watchdog can be disabled as root with
230 echo 0 > /proc/sys/kernel/nmi_watchdog
232 Events from multiple different PMUs cannot be mixed in a group, with
233 some exceptions for software events.
238 perf also supports group leader sampling using the :S specifier.
240 perf record -e '{cycles,instructions}:S' ...
243 Normally all events in an event group sample, but with :S only
244 the first event (the leader) samples, and it only reads the values of the
245 other events in the group.
250 Without options all known events will be listed.
252 To limit the list use:
254 . 'hw' or 'hardware' to list hardware events such as cache-misses, etc.
256 . 'sw' or 'software' to list software events such as context switches, etc.
258 . 'cache' or 'hwcache' to list hardware cache events such as L1-dcache-loads, etc.
260 . 'tracepoint' to list all tracepoint events, alternatively use
261 'subsys_glob:event_glob' to filter by tracepoint subsystems such as sched,
264 . 'pmu' to print the kernel supplied PMU events.
266 . 'sdt' to list all Statically Defined Tracepoint events.
268 . 'metric' to list metrics
270 . 'metricgroup' to list metricgroups with metrics.
272 . If none of the above is matched, it will apply the supplied glob to all
273 events, printing the ones that match.
275 . As a last resort, it will do a substring search in all event names.
277 One or more types can be used at the same time, listing the events for the
282 . '--raw-dump', shows the raw-dump of all the events.
283 . '--raw-dump [hw|sw|cache|tracepoint|pmu|event_glob]', shows the raw-dump of
284 a certain kind of events.
288 linkperf:perf-stat[1], linkperf:perf-top[1],
289 linkperf:perf-record[1],
290 http://www.intel.com/sdm/[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide],
291 http://support.amd.com/us/Processor_TechDocs/24593_APM_v2.pdf[AMD64 Architecture Programmer’s Manual Volume 2: System Programming]