1 perf-arm-spe(1)
2 ================
3
4 NAME
5 ----
6 perf-arm-spe - Support for Arm Statistical Pro
7
8 SYNOPSIS
9 --------
10 [verse]
11 'perf record' -e arm_spe//
12
13 DESCRIPTION
14 -----------
15
16 The SPE (Statistical Profiling Extension) feat
17 events down to individual instructions. Rathe
18 instruction to sample and then captures data f
19 in cycles. For loads and stores it also includ
20
21 The sampling has 5 stages:
22
23 1. Choose an operation
24 2. Collect data about the operation
25 3. Optionally discard the record based on a
26 4. Write the record to memory
27 5. Interrupt when the buffer is full
28
29 Choose an operation
30 ~~~~~~~~~~~~~~~~~~~
31
32 This is chosen from a sample population, for S
33 architectural instructions or all micro-ops. S
34 architecture provides a mechanism for the SPE
35 sample. This minimum interval is used by the d
36 perturbation is also added to the sampling int
37
38 Collect data about the operation
39 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
40
41 Program counter, PMU events, timings and data
42 Sampling ensures there is only one sampled ope
43
44 Optionally discard the record based on a filte
45 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
46
47 Based on programmable criteria, choose whether
48 discarded then the flow stops here for this sa
49
50 Write the record to memory
51 ~~~~~~~~~~~~~~~~~~~~~~~~~~
52
53 The record is appended to a memory buffer
54
55 Interrupt when the buffer is full
56 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
57
58 When the buffer fills, an interrupt is sent an
59 Perf saves the raw data in the perf.data file.
60
61 Opening the file
62 ----------------
63
64 Up until this point no decoding of the SPE dat
65 recorded file is opened with 'perf report' or
66 the data, Perf generates "synthetic samples" a
67 recording. These samples are the same as if no
68 although they may have more attributes associa
69 just the instruction pointer, but an SPE sampl
70
71 Why Sampling?
72 -------------
73
74 - Sampling, rather than tracing, cuts down th
75 hardware. Only one sampled operation is in fl
76
77 - Allows precise attribution data, including:
78 addresses.
79
80 - Allows correlation between an instruction a
81 indicates which particular cache was hit, but
82 different implementations can have different
83
84 However, SPE does not provide any call-graph i
85
86 Collisions
87 ----------
88
89 When an operation is sampled while a previous
90 occurs. The new sample is dropped. Collisions
91 should be set to avoid collisions.
92
93 The 'sample_collision' PMU event can be used t
94 count is based on collisions _before_ filterin
95 number for samples dropped that would have mad
96 guide.
97
98 The effect of microarchitectural sampling
99 -----------------------------------------
100
101 If an implementation samples micro-operations
102 be weighted accordingly.
103
104 For example, if a given instruction A is alway
105 becomes twice as likely to appear in the sampl
106
107 The coarse effect of conversions, and, if appl
108 estimated from the 'sample_pop' and 'inst_reti
109
110 Kernel Requirements
111 -------------------
112
113 The ARM_SPE_PMU config must be set to build as
114
115 Depending on CPU model, the kernel may need to
116 (kpti=off). If KPTI needs to be disabled, this
117 inaccessible. Try passing 'kpti=off' on the ke
118
119 For the full criteria that determine whether K
120 unmap_kernel_at_el0() in the kernel sources. C
121 are on the CPUs in kpti_safe_list, or on Arm v
122
123 The SPE interrupt must also be described by th
124 disabled (or isn't required to be disabled) bu
125 /sys/bus/event_source/devices/, then it's poss
126 ACPI or DT. In this case no warning will be pr
127
128 Capturing SPE with perf command-line tools
129 ------------------------------------------
130
131 You can record a session with SPE samples:
132
133 perf record -e arm_spe// -- ./mybench
134
135 The sample period is set from the -c option, a
136 it's recommended to set this to a higher value
137
138 Config parameters
139 ~~~~~~~~~~~~~~~~~
140
141 These are placed between the // in the event a
142 arm_spe/load_filter=1,min_latency=10/'
143
144 branch_filter=1 - collect branches only
145 event_filter=<mask> - filter on specific eve
146 jitter=1 - use jitter to avoid re
147 load_filter=1 - collect loads only (PM
148 min_latency=<n> - collect only samples w
149 pa_enable=1 - collect physical addre
150 pct_enable=1 - collect physical times
151 store_filter=1 - collect stores only (P
152 ts_enable=1 - enable timestamping wi
153
154 +++*+++ Latency is the total latency from the
155 than only the execution latency.
156
157 Only some events can be filtered on; these inc
158
159 bit 1 - instruction retired (i.e. omit s
160 bit 3 - L1D refill
161 bit 5 - TLB refill
162 bit 7 - mispredict
163 bit 11 - misaligned access
164
165 So to sample just retired instructions:
166
167 perf record -e arm_spe/event_filter=2/ -- ./
168
169 or just mispredicted branches:
170
171 perf record -e arm_spe/event_filter=0x80/ --
172
173 Viewing the data
174 ~~~~~~~~~~~~~~~~~
175
176 By default perf report and perf script will as
177 attributes/events of the SPE record. Because i
178 them, the samples in these groups are not nece
179 groups:
180
181 Available samples
182 0 arm_spe//
183 0 dummy:u
184 21 l1d-miss
185 897 l1d-access
186 5 llc-miss
187 7 llc-access
188 2 tlb-miss
189 1K tlb-access
190 36 branch-miss
191 0 remote-access
192 900 memory
193
194 The arm_spe// and dummy:u events are implement
195
196 To get a full list of unique samples that are
197 generate 'instruction' samples. The period opt
198 instruction unless you want to further downsam
199
200 perf report --itrace=i1i
201
202 Memory access details are also stored on the s
203
204 perf report --mem-mode
205
206 Common errors
207 ~~~~~~~~~~~~~
208
209 - "Cannot find PMU `arm_spe'. Missing kernel
210
211 Module not built or loaded, KPTI not disabl
212 or running on a VM. See 'Kernel Requirement
213
214 - "Arm SPE CONTEXT packets not found in the t
215
216 Root privilege is required to collect conte
217 assigning PIDs to kernel samples. For users
218
219 - Excessively large perf.data file size
220
221 Increase sampling interval (see above)
222
223
224 SEE ALSO
225 --------
226
227 linkperf:perf-record[1], linkperf:perf-script[
228 linkperf:perf-inject[1]
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.