1PERF-MEM(1)                       perf Manual                      PERF-MEM(1)
2
3
4

NAME

6       perf-mem - Profile memory accesses
7

SYNOPSIS

9       perf mem [<options>] (record [<command>] | report)
10

DESCRIPTION

12       "perf mem record" runs a command and gathers memory operation data from
13       it, into perf.data. Perf record options are accepted and are passed
14       through.
15
16       "perf mem report" displays the result. It invokes perf report with the
17       right set of options to display a memory access profile. By default,
18       loads and stores are sampled. Use the -t option to limit to loads or
19       stores.
20
21       Note that on Intel systems the memory latency reported is the
22       use-latency, not the pure load (or store latency). Use latency includes
23       any pipeline queueing delays in addition to the memory subsystem
24       latency.
25
26       On Arm64 this uses SPE to sample load and store operations, therefore
27       hardware and kernel support is required. See perf-arm-spe(1) for a
28       setup guide. Due to the statistical nature of SPE sampling, not every
29       memory operation will be sampled.
30

OPTIONS

32       <command>...
33           Any command you can specify in a shell.
34
35       -i, --input=<file>
36           Input file name.
37
38       -f, --force
39           Don’t do ownership validation
40
41       -t, --type=<type>
42           Select the memory operation type: load or store (default:
43           load,store)
44
45       -D, --dump-raw-samples
46           Dump the raw decoded samples on the screen in a format that is easy
47           to parse with one sample per line.
48
49       -x, --field-separator=<separator>
50           Specify the field separator used when dump raw samples (-D option).
51           By default, The separator is the space character.
52
53       -C, --cpu=<cpu>
54           Monitor only on the list of CPUs provided. Multiple CPUs can be
55           provided as a comma-separated list with no space: 0,1. Ranges of
56           CPUs are specified with -: 0-2. Default is to monitor all CPUS.
57
58       -U, --hide-unresolved
59           Only display entries resolved to a symbol.
60
61       -p, --phys-data
62           Record/Report sample physical addresses
63
64       --data-page-size
65           Record/Report sample data address page size
66

RECORD OPTIONS

68       -e, --event <event>
69           Event selector. Use perf mem record -e list to list available
70           events.
71
72       -K, --all-kernel
73           Configure all used events to run in kernel space.
74
75       -U, --all-user
76           Configure all used events to run in user space.
77
78       -v, --verbose
79           Be more verbose (show counter open errors, etc)
80
81       --ldlat <n>
82           Specify desired latency for loads event. Supported on Intel and
83           Arm64 processors only. Ignored on other archs.
84
85       In addition, for report all perf report options are valid, and for
86       record all perf record options.
87

SEE ALSO

89       perf-record(1), perf-report(1), perf-arm-spe(1)
90
91
92
93perf                              11/28/2023                       PERF-MEM(1)
Impressum