tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
perf Documentation: Document intel-hybrid support
Add some words and examples to help understanding of
Intel hybrid perf support.
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20210427070139.25256-27-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
authored by
Jin Yao
and committed by
Arnaldo Carvalho de Melo
2750ce1d
a37f3b88
+217
+214
tools/perf/Documentation/intel-hybrid.txt
+214
tools/perf/Documentation/intel-hybrid.txt
···
1
+
Intel hybrid support
2
+
--------------------
3
+
Support for Intel hybrid events within perf tools.
4
+
5
+
For some Intel platforms, such as AlderLake, which is hybrid platform and
6
+
it consists of atom cpu and core cpu. Each cpu has dedicated event list.
7
+
Part of events are available on core cpu, part of events are available
8
+
on atom cpu and even part of events are available on both.
9
+
10
+
Kernel exports two new cpu pmus via sysfs:
11
+
/sys/devices/cpu_core
12
+
/sys/devices/cpu_atom
13
+
14
+
The 'cpus' files are created under the directories. For example,
15
+
16
+
cat /sys/devices/cpu_core/cpus
17
+
0-15
18
+
19
+
cat /sys/devices/cpu_atom/cpus
20
+
16-23
21
+
22
+
It indicates cpu0-cpu15 are core cpus and cpu16-cpu23 are atom cpus.
23
+
24
+
Quickstart
25
+
26
+
List hybrid event
27
+
-----------------
28
+
29
+
As before, use perf-list to list the symbolic event.
30
+
31
+
perf list
32
+
33
+
inst_retired.any
34
+
[Fixed Counter: Counts the number of instructions retired. Unit: cpu_atom]
35
+
inst_retired.any
36
+
[Number of instructions retired. Fixed Counter - architectural event. Unit: cpu_core]
37
+
38
+
The 'Unit: xxx' is added to brief description to indicate which pmu
39
+
the event is belong to. Same event name but with different pmu can
40
+
be supported.
41
+
42
+
Enable hybrid event with a specific pmu
43
+
---------------------------------------
44
+
45
+
To enable a core only event or atom only event, following syntax is supported:
46
+
47
+
cpu_core/<event name>/
48
+
or
49
+
cpu_atom/<event name>/
50
+
51
+
For example, count the 'cycles' event on core cpus.
52
+
53
+
perf stat -e cpu_core/cycles/
54
+
55
+
Create two events for one hardware event automatically
56
+
------------------------------------------------------
57
+
58
+
When creating one event and the event is available on both atom and core,
59
+
two events are created automatically. One is for atom, the other is for
60
+
core. Most of hardware events and cache events are available on both
61
+
cpu_core and cpu_atom.
62
+
63
+
For hardware events, they have pre-defined configs (e.g. 0 for cycles).
64
+
But on hybrid platform, kernel needs to know where the event comes from
65
+
(from atom or from core). The original perf event type PERF_TYPE_HARDWARE
66
+
can't carry pmu information. So now this type is extended to be PMU aware
67
+
type. The PMU type ID is stored at attr.config[63:32].
68
+
69
+
PMU type ID is retrieved from sysfs.
70
+
/sys/devices/cpu_atom/type
71
+
/sys/devices/cpu_core/type
72
+
73
+
The new attr.config layout for PERF_TYPE_HARDWARE:
74
+
75
+
PERF_TYPE_HARDWARE: 0xEEEEEEEE000000AA
76
+
AA: hardware event ID
77
+
EEEEEEEE: PMU type ID
78
+
79
+
Cache event is similar. The type PERF_TYPE_HW_CACHE is extended to be
80
+
PMU aware type. The PMU type ID is stored at attr.config[63:32].
81
+
82
+
The new attr.config layout for PERF_TYPE_HW_CACHE:
83
+
84
+
PERF_TYPE_HW_CACHE: 0xEEEEEEEE00DDCCBB
85
+
BB: hardware cache ID
86
+
CC: hardware cache op ID
87
+
DD: hardware cache op result ID
88
+
EEEEEEEE: PMU type ID
89
+
90
+
When enabling a hardware event without specified pmu, such as,
91
+
perf stat -e cycles -a (use system-wide in this example), two events
92
+
are created automatically.
93
+
94
+
------------------------------------------------------------
95
+
perf_event_attr:
96
+
size 120
97
+
config 0x400000000
98
+
sample_type IDENTIFIER
99
+
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
100
+
disabled 1
101
+
inherit 1
102
+
exclude_guest 1
103
+
------------------------------------------------------------
104
+
105
+
and
106
+
107
+
------------------------------------------------------------
108
+
perf_event_attr:
109
+
size 120
110
+
config 0x800000000
111
+
sample_type IDENTIFIER
112
+
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
113
+
disabled 1
114
+
inherit 1
115
+
exclude_guest 1
116
+
------------------------------------------------------------
117
+
118
+
type 0 is PERF_TYPE_HARDWARE.
119
+
0x4 in 0x400000000 indicates it's cpu_core pmu.
120
+
0x8 in 0x800000000 indicates it's cpu_atom pmu (atom pmu type id is random).
121
+
122
+
The kernel creates 'cycles' (0x400000000) on cpu0-cpu15 (core cpus),
123
+
and create 'cycles' (0x800000000) on cpu16-cpu23 (atom cpus).
124
+
125
+
For perf-stat result, it displays two events:
126
+
127
+
Performance counter stats for 'system wide':
128
+
129
+
6,744,979 cpu_core/cycles/
130
+
1,965,552 cpu_atom/cycles/
131
+
132
+
The first 'cycles' is core event, the second 'cycles' is atom event.
133
+
134
+
Thread mode example:
135
+
--------------------
136
+
137
+
perf-stat reports the scaled counts for hybrid event and with a percentage
138
+
displayed. The percentage is the event's running time/enabling time.
139
+
140
+
One example, 'triad_loop' runs on cpu16 (atom core), while we can see the
141
+
scaled value for core cycles is 160,444,092 and the percentage is 0.47%.
142
+
143
+
perf stat -e cycles -- taskset -c 16 ./triad_loop
144
+
145
+
As previous, two events are created.
146
+
147
+
------------------------------------------------------------
148
+
perf_event_attr:
149
+
size 120
150
+
config 0x400000000
151
+
sample_type IDENTIFIER
152
+
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
153
+
disabled 1
154
+
inherit 1
155
+
enable_on_exec 1
156
+
exclude_guest 1
157
+
------------------------------------------------------------
158
+
159
+
and
160
+
161
+
------------------------------------------------------------
162
+
perf_event_attr:
163
+
size 120
164
+
config 0x800000000
165
+
sample_type IDENTIFIER
166
+
read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
167
+
disabled 1
168
+
inherit 1
169
+
enable_on_exec 1
170
+
exclude_guest 1
171
+
------------------------------------------------------------
172
+
173
+
Performance counter stats for 'taskset -c 16 ./triad_loop':
174
+
175
+
233,066,666 cpu_core/cycles/ (0.43%)
176
+
604,097,080 cpu_atom/cycles/ (99.57%)
177
+
178
+
perf-record:
179
+
------------
180
+
181
+
If there is no '-e' specified in perf record, on hybrid platform,
182
+
it creates two default 'cycles' and adds them to event list. One
183
+
is for core, the other is for atom.
184
+
185
+
perf-stat:
186
+
----------
187
+
188
+
If there is no '-e' specified in perf stat, on hybrid platform,
189
+
besides of software events, following events are created and
190
+
added to event list in order.
191
+
192
+
cpu_core/cycles/,
193
+
cpu_atom/cycles/,
194
+
cpu_core/instructions/,
195
+
cpu_atom/instructions/,
196
+
cpu_core/branches/,
197
+
cpu_atom/branches/,
198
+
cpu_core/branch-misses/,
199
+
cpu_atom/branch-misses/
200
+
201
+
Of course, both perf-stat and perf-record support to enable
202
+
hybrid event with a specific pmu.
203
+
204
+
e.g.
205
+
perf stat -e cpu_core/cycles/
206
+
perf stat -e cpu_atom/cycles/
207
+
perf stat -e cpu_core/r1a/
208
+
perf stat -e cpu_atom/L1-icache-loads/
209
+
perf stat -e cpu_core/cycles/,cpu_atom/instructions/
210
+
perf stat -e '{cpu_core/cycles/,cpu_core/instructions/}'
211
+
212
+
But '{cpu_core/cycles/,cpu_atom/instructions/}' will return
213
+
warning and disable grouping, because the pmus in group are
214
+
not matched (cpu_core vs. cpu_atom).
+1
tools/perf/Documentation/perf-record.txt
+1
tools/perf/Documentation/perf-record.txt