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.
22 Don't print descriptions.
26 Print longer event descriptions.
29 Print how named events are resolved internally into perf events, and also
30 any extra expressions computed by perf stat.
37 Events can optionally have a modifier by appending a colon and one or
38 more modifiers. Modifiers allow the user to restrict the events to be
39 counted. The following modifiers exist:
41 u - user-space counting
43 h - hypervisor counting
45 G - guest counting (in KVM guests)
46 H - host counting (not in KVM guests)
48 P - use maximum detected precise level
49 S - read sample value (PERF_SAMPLE_READ)
50 D - pin the event to the PMU
51 W - group is weak and will fallback to non-group if not schedulable,
52 only supported in 'perf stat' for now.
54 The 'p' modifier can be used for specifying how precise the instruction
55 address should be. The 'p' modifier can be specified multiple times:
57 0 - SAMPLE_IP can have arbitrary skid
58 1 - SAMPLE_IP must have constant skid
59 2 - SAMPLE_IP requested to have 0 skid
60 3 - SAMPLE_IP must have 0 skid, or uses randomization to avoid
61 sample shadowing effects.
63 For Intel systems precise event sampling is implemented with PEBS
64 which supports up to precise-level 2, and precise level 3 for
67 On AMD systems it is implemented using IBS (up to precise-level 2).
68 The precise modifier works with event types 0x76 (cpu-cycles, CPU
69 clocks not halted) and 0xC1 (micro-ops retired). Both events map to
70 IBS execution sampling (IBS op) with the IBS Op Counter Control bit
71 (IbsOpCntCtl) set respectively (see AMD64 Architecture Programmer’s
72 Manual Volume 2: System Programming, 13.3 Instruction-Based
73 Sampling). Examples to use IBS:
75 perf record -a -e cpu-cycles:p ... # use ibs op counting cycles
76 perf record -a -e r076:p ... # same as -e cpu-cycles:p
77 perf record -a -e r0C1:p ... # use ibs op counting micro-ops
79 RAW HARDWARE EVENT DESCRIPTOR
80 -----------------------------
81 Even when an event is not available in a symbolic form within perf right now,
82 it can be encoded in a per processor specific way.
84 For instance For x86 CPUs NNN represents the raw register encoding with the
85 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
86 of IA32_PERFEVTSELx MSRs) or AMD's PerfEvtSeln (see [AMD64 Architecture Programmer’s Manual Volume 2: System Programming], Page 344,
87 Figure 13-7 Performance Event-Select Register (PerfEvtSeln)).
89 Note: Only the following bit fields can be set in x86 counter
90 registers: event, umask, edge, inv, cmask. Esp. guest/host only and
91 OS/user mode flags must be setup using <<EVENT_MODIFIERS, EVENT
96 If the Intel docs for a QM720 Core i7 describe an event as:
98 Event Umask Event Mask
99 Num. Value Mnemonic Description Comment
101 A8H 01H LSD.UOPS Counts the number of micro-ops Use cmask=1 and
102 delivered by loop stream detector invert to count
105 raw encoding of 0x1A8 can be used:
107 perf stat -e r1a8 -a sleep 1
108 perf record -e r1a8 ...
110 You should refer to the processor specific documentation for getting these
111 details. Some of them are referenced in the SEE ALSO section below.
116 perf also supports an extended syntax for specifying raw parameters
117 to PMUs. Using this typically requires looking up the specific event
118 in the CPU vendor specific documentation.
120 The available PMUs and their raw parameters can be listed with
122 ls /sys/devices/*/format
124 For example the raw event "LSD.UOPS" core pmu event above could
127 perf stat -e cpu/event=0xa8,umask=0x1,name=LSD.UOPS_CYCLES,cmask=1/ ...
132 Some PMUs are not associated with a core, but with a whole CPU socket.
133 Events on these PMUs generally cannot be sampled, but only counted globally
134 with perf stat -a. They can be bound to one logical CPU, but will measure
135 all the CPUs in the same socket.
137 This example measures memory bandwidth every second
138 on the first memory controller on socket 0 of a Intel Xeon system
140 perf stat -C 0 -a uncore_imc_0/cas_count_read/,uncore_imc_0/cas_count_write/ -I 1000 ...
142 Each memory controller has its own PMU. Measuring the complete system
143 bandwidth would require specifying all imc PMUs (see perf list output),
144 and adding the values together.
146 This example measures the combined core power every second
148 perf stat -I 1000 -e power/energy-cores/ -a
153 For non root users generally only context switched PMU events are available.
154 This is normally only the events in the cpu PMU, the predefined events
155 like cycles and instructions and some software events.
157 Other PMUs and global measurements are normally root only.
158 Some event qualifiers, such as "any", are also root only.
160 This can be overriden by setting the kernel.perf_event_paranoid
161 sysctl to -1, which allows non root to use these events.
163 For accessing trace point events perf needs to have read access to
164 /sys/kernel/debug/tracing, even when perf_event_paranoid is in a relaxed
170 Some PMUs control advanced hardware tracing capabilities, such as Intel PT,
171 that allows low overhead execution tracing. These are described in a separate
172 intel-pt.txt document.
177 Some pmu events listed by 'perf-list' will be displayed with '?' in them. For
180 hv_gpci/dtbp_ptitc,phys_processor_idx=?/
182 This means that when provided as an event, a value for '?' must
183 also be supplied. For example:
185 perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
190 Perf supports time based multiplexing of events, when the number of events
191 active exceeds the number of hardware performance counters. Multiplexing
192 can cause measurement errors when the workload changes its execution
195 When metrics are computed using formulas from event counts, it is useful to
196 ensure some events are always measured together as a group to minimize multiplexing
197 errors. Event groups can be specified using { }.
199 perf stat -e '{instructions,cycles}' ...
201 The number of available performance counters depend on the CPU. A group
202 cannot contain more events than available counters.
203 For example Intel Core CPUs typically have four generic performance counters
204 for the core, plus three fixed counters for instructions, cycles and
205 ref-cycles. Some special events have restrictions on which counter they
206 can schedule, and may not support multiple instances in a single group.
207 When too many events are specified in the group some of them will not
210 Globally pinned events can limit the number of counters available for
211 other groups. On x86 systems, the NMI watchdog pins a counter by default.
212 The nmi watchdog can be disabled as root with
214 echo 0 > /proc/sys/kernel/nmi_watchdog
216 Events from multiple different PMUs cannot be mixed in a group, with
217 some exceptions for software events.
222 perf also supports group leader sampling using the :S specifier.
224 perf record -e '{cycles,instructions}:S' ...
227 Normally all events in a event group sample, but with :S only
228 the first event (the leader) samples, and it only reads the values of the
229 other events in the group.
234 Without options all known events will be listed.
236 To limit the list use:
238 . 'hw' or 'hardware' to list hardware events such as cache-misses, etc.
240 . 'sw' or 'software' to list software events such as context switches, etc.
242 . 'cache' or 'hwcache' to list hardware cache events such as L1-dcache-loads, etc.
244 . 'tracepoint' to list all tracepoint events, alternatively use
245 'subsys_glob:event_glob' to filter by tracepoint subsystems such as sched,
248 . 'pmu' to print the kernel supplied PMU events.
250 . 'sdt' to list all Statically Defined Tracepoint events.
252 . 'metric' to list metrics
254 . 'metricgroup' to list metricgroups with metrics.
256 . If none of the above is matched, it will apply the supplied glob to all
257 events, printing the ones that match.
259 . As a last resort, it will do a substring search in all event names.
261 One or more types can be used at the same time, listing the events for the
266 . '--raw-dump', shows the raw-dump of all the events.
267 . '--raw-dump [hw|sw|cache|tracepoint|pmu|event_glob]', shows the raw-dump of
268 a certain kind of events.
272 linkperf:perf-stat[1], linkperf:perf-top[1],
273 linkperf:perf-record[1],
274 http://www.intel.com/sdm/[Intel® 64 and IA-32 Architectures Software Developer's Manual Volume 3B: System Programming Guide],
275 http://support.amd.com/us/Processor_TechDocs/24593_APM_v2.pdf[AMD64 Architecture Programmer’s Manual Volume 2: System Programming]