1 perf-c2c(1) 1 perf-c2c(1)
2 =========== 2 ===========
3 3
4 NAME 4 NAME
5 ---- 5 ----
6 perf-c2c - Shared Data C2C/HITM Analyzer. 6 perf-c2c - Shared Data C2C/HITM Analyzer.
7 7
8 SYNOPSIS 8 SYNOPSIS
9 -------- 9 --------
10 [verse] 10 [verse]
11 'perf c2c record' [<options>] <command> 11 'perf c2c record' [<options>] <command>
12 'perf c2c record' [<options>] \-- [<record com !! 12 'perf c2c record' [<options>] -- [<record command options>] <command>
13 'perf c2c report' [<options>] 13 'perf c2c report' [<options>]
14 14
15 DESCRIPTION 15 DESCRIPTION
16 ----------- 16 -----------
17 C2C stands for Cache To Cache. 17 C2C stands for Cache To Cache.
18 18
19 The perf c2c tool provides means for Shared Da 19 The perf c2c tool provides means for Shared Data C2C/HITM analysis. It allows
20 you to track down the cacheline contentions. 20 you to track down the cacheline contentions.
21 21
22 On Intel, the tool is based on load latency an !! 22 On x86, the tool is based on load latency and precise store facility events
23 provided by Intel CPUs. On PowerPC, the tool u 23 provided by Intel CPUs. On PowerPC, the tool uses random instruction sampling
24 with thresholding feature. On AMD, the tool us !! 24 with thresholding feature.
25 limitations, perf c2c is not supported on Zen3 <<
26 sample load and store operations, therefore ha <<
27 required. See linkperf:perf-arm-spe[1] for a s <<
28 statistical nature of Arm SPE sampling, not ev <<
29 sampled. <<
30 25
31 These events provide: 26 These events provide:
32 - memory address of the access 27 - memory address of the access
33 - type of the access (load and store details 28 - type of the access (load and store details)
34 - latency (in cycles) of the load access 29 - latency (in cycles) of the load access
35 30
36 The c2c tool provide means to record this data 31 The c2c tool provide means to record this data and report back access details
37 for cachelines with highest contention - highe 32 for cachelines with highest contention - highest number of HITM accesses.
38 33
39 The basic workflow with this tool follows the 34 The basic workflow with this tool follows the standard record/report phase.
40 User uses the record command to record events 35 User uses the record command to record events data and report command to
41 display it. 36 display it.
42 37
43 38
44 RECORD OPTIONS 39 RECORD OPTIONS
45 -------------- 40 --------------
46 -e:: 41 -e::
47 --event=:: 42 --event=::
48 Select the PMU event. Use 'perf c2c re 43 Select the PMU event. Use 'perf c2c record -e list'
49 to list available events. 44 to list available events.
50 45
51 -v:: 46 -v::
52 --verbose:: 47 --verbose::
53 Be more verbose (show counter open err 48 Be more verbose (show counter open errors, etc).
54 49
55 -l:: 50 -l::
56 --ldlat:: 51 --ldlat::
57 Configure mem-loads latency. Supported !! 52 Configure mem-loads latency. (x86 only)
58 only. Ignored on other archs. <<
59 53
60 -k:: 54 -k::
61 --all-kernel:: 55 --all-kernel::
62 Configure all used events to run in ke 56 Configure all used events to run in kernel space.
63 57
64 -u:: 58 -u::
65 --all-user:: 59 --all-user::
66 Configure all used events to run in us 60 Configure all used events to run in user space.
67 61
68 REPORT OPTIONS 62 REPORT OPTIONS
69 -------------- 63 --------------
70 -k:: 64 -k::
71 --vmlinux=<file>:: 65 --vmlinux=<file>::
72 vmlinux pathname 66 vmlinux pathname
73 67
74 -v:: 68 -v::
75 --verbose:: 69 --verbose::
76 Be more verbose (show counter open err 70 Be more verbose (show counter open errors, etc).
77 71
78 -i:: 72 -i::
79 --input:: 73 --input::
80 Specify the input file to process. 74 Specify the input file to process.
81 75
82 -N:: 76 -N::
83 --node-info:: 77 --node-info::
84 Show extra node info in report (see NO 78 Show extra node info in report (see NODE INFO section)
85 79
86 -c:: 80 -c::
87 --coalesce:: 81 --coalesce::
88 Specify sorting fields for single cach 82 Specify sorting fields for single cacheline display.
89 Following fields are available: tid,pi 83 Following fields are available: tid,pid,iaddr,dso
90 (see COALESCE) 84 (see COALESCE)
91 85
92 -g:: 86 -g::
93 --call-graph:: 87 --call-graph::
94 Setup callchains parameters. 88 Setup callchains parameters.
95 Please refer to perf-report man page f 89 Please refer to perf-report man page for details.
96 90
97 --stdio:: 91 --stdio::
98 Force the stdio output (see STDIO OUTP 92 Force the stdio output (see STDIO OUTPUT)
99 93
100 --stats:: 94 --stats::
101 Display only statistic tables and forc 95 Display only statistic tables and force stdio mode.
102 96
103 --full-symbols:: 97 --full-symbols::
104 Display full length of symbols. 98 Display full length of symbols.
105 99
106 --no-source:: 100 --no-source::
107 Do not display Source:Line column. 101 Do not display Source:Line column.
108 102
109 --show-all:: 103 --show-all::
110 Show all captured HITM lines, with no 104 Show all captured HITM lines, with no regard to HITM % 0.0005 limit.
111 105
112 -f:: 106 -f::
113 --force:: 107 --force::
114 Don't do ownership validation. 108 Don't do ownership validation.
115 109
116 -d:: 110 -d::
117 --display:: 111 --display::
118 Switch to HITM type (rmt, lcl) or peer !! 112 Switch to HITM type (rmt, lcl) to display and sort on. Total HITMs as default.
119 and sort on. Total HITMs (tot) as defa <<
120 as default. <<
121 113
122 --stitch-lbr:: 114 --stitch-lbr::
123 Show callgraph with stitched LBRs, whi 115 Show callgraph with stitched LBRs, which may have more complete
124 callgraph. The perf.data file must hav 116 callgraph. The perf.data file must have been obtained using
125 perf c2c record --call-graph lbr. 117 perf c2c record --call-graph lbr.
126 Disabled by default. In common cases w 118 Disabled by default. In common cases with call stack overflows,
127 it can recreate better call stacks tha 119 it can recreate better call stacks than the default lbr call stack
128 output. But this approach is not foolp !! 120 output. But this approach is not full proof. There can be cases
129 where it creates incorrect call stacks 121 where it creates incorrect call stacks from incorrect matches.
130 The known limitations include exceptio 122 The known limitations include exception handing such as
131 setjmp/longjmp will have calls/returns 123 setjmp/longjmp will have calls/returns not match.
132 124
133 --double-cl:: <<
134 Group the detection of shared cachelin <<
135 granularity. Some architectures have a <<
136 feature, which causes cacheline sharin <<
137 size is doubled. <<
138 <<
139 C2C RECORD 125 C2C RECORD
140 ---------- 126 ----------
141 The perf c2c record command setup options rela 127 The perf c2c record command setup options related to HITM cacheline analysis
142 and calls standard perf record command. 128 and calls standard perf record command.
143 129
144 Following perf record options are configured b 130 Following perf record options are configured by default:
145 (check perf record man page for details) 131 (check perf record man page for details)
146 132
147 -W,-d,--phys-data,--sample-cpu 133 -W,-d,--phys-data,--sample-cpu
148 134
149 Unless specified otherwise with '-e' option, f 135 Unless specified otherwise with '-e' option, following events are monitored by
150 default on Intel: !! 136 default on x86:
151 137
152 cpu/mem-loads,ldlat=30/P 138 cpu/mem-loads,ldlat=30/P
153 cpu/mem-stores/P 139 cpu/mem-stores/P
154 140
155 following on AMD: <<
156 <<
157 ibs_op// <<
158 <<
159 and following on PowerPC: 141 and following on PowerPC:
160 142
161 cpu/mem-loads/ 143 cpu/mem-loads/
162 cpu/mem-stores/ 144 cpu/mem-stores/
163 145
164 User can pass any 'perf record' option behind 146 User can pass any 'perf record' option behind '--' mark, like (to enable
165 callchains and system wide monitoring): 147 callchains and system wide monitoring):
166 148
167 $ perf c2c record -- -g -a 149 $ perf c2c record -- -g -a
168 150
169 Please check RECORD OPTIONS section for specif 151 Please check RECORD OPTIONS section for specific c2c record options.
170 152
171 C2C REPORT 153 C2C REPORT
172 ---------- 154 ----------
173 The perf c2c report command displays shared da 155 The perf c2c report command displays shared data analysis. It comes in two
174 display modes: stdio and tui (default). 156 display modes: stdio and tui (default).
175 157
176 The report command workflow is following: 158 The report command workflow is following:
177 - sort all the data based on the cacheline a 159 - sort all the data based on the cacheline address
178 - store access details for each cacheline 160 - store access details for each cacheline
179 - sort all cachelines based on user settings 161 - sort all cachelines based on user settings
180 - display data 162 - display data
181 163
182 In general perf report output consist of 2 bas 164 In general perf report output consist of 2 basic views:
183 1) most expensive cachelines list 165 1) most expensive cachelines list
184 2) offsets details for each cacheline 166 2) offsets details for each cacheline
185 167
186 For each cacheline in the 1) list we display f 168 For each cacheline in the 1) list we display following data:
187 (Both stdio and TUI modes follow the same fiel 169 (Both stdio and TUI modes follow the same fields output)
188 170
189 Index 171 Index
190 - zero based index to identify the cacheline 172 - zero based index to identify the cacheline
191 173
192 Cacheline 174 Cacheline
193 - cacheline address (hex number) 175 - cacheline address (hex number)
194 176
195 Rmt/Lcl Hitm (Display with HITM types) !! 177 Rmt/Lcl Hitm
196 - cacheline percentage of all Remote/Local H 178 - cacheline percentage of all Remote/Local HITM accesses
197 179
198 Peer Snoop (Display with peer type) !! 180 LLC Load Hitm - Total, LclHitm, RmtHitm
199 - cacheline percentage of all peer accesses <<
200 <<
201 LLC Load Hitm - Total, LclHitm, RmtHitm (For <<
202 - count of Total/Local/Remote load HITMs 181 - count of Total/Local/Remote load HITMs
203 182
204 Load Peer - Total, Local, Remote (For displa <<
205 - count of Total/Local/Remote load from peer <<
206 <<
207 Total records 183 Total records
208 - sum of all cachelines accesses 184 - sum of all cachelines accesses
209 185
210 Total loads 186 Total loads
211 - sum of all load accesses 187 - sum of all load accesses
212 188
213 Total stores 189 Total stores
214 - sum of all store accesses 190 - sum of all store accesses
215 191
216 Store Reference - L1Hit, L1Miss, N/A !! 192 Store Reference - L1Hit, L1Miss
217 L1Hit - store accesses that hit L1 193 L1Hit - store accesses that hit L1
218 L1Miss - store accesses that missed L1 194 L1Miss - store accesses that missed L1
219 N/A - store accesses with memory level is <<
220 195
221 Core Load Hit - FB, L1, L2 196 Core Load Hit - FB, L1, L2
222 - count of load hits in FB (Fill Buffer), L1 197 - count of load hits in FB (Fill Buffer), L1 and L2 cache
223 198
224 LLC Load Hit - LlcHit, LclHitm 199 LLC Load Hit - LlcHit, LclHitm
225 - count of LLC load accesses, includes LLC h 200 - count of LLC load accesses, includes LLC hits and LLC HITMs
226 201
227 RMT Load Hit - RmtHit, RmtHitm 202 RMT Load Hit - RmtHit, RmtHitm
228 - count of remote load accesses, includes re !! 203 - count of remote load accesses, includes remote hits and remote HITMs
229 on Arm neoverse cores, RmtHit is used to a <<
230 includes remote DRAM or any upward cache l <<
231 204
232 Load Dram - Lcl, Rmt 205 Load Dram - Lcl, Rmt
233 - count of local and remote DRAM accesses 206 - count of local and remote DRAM accesses
234 207
235 For each offset in the 2) list we display foll 208 For each offset in the 2) list we display following data:
236 209
237 HITM - Rmt, Lcl (Display with HITM types) !! 210 HITM - Rmt, Lcl
238 - % of Remote/Local HITM accesses for given 211 - % of Remote/Local HITM accesses for given offset within cacheline
239 212
240 Peer Snoop - Rmt, Lcl (Display with peer typ !! 213 Store Refs - L1 Hit, L1 Miss
241 - % of Remote/Local peer accesses for given !! 214 - % of store accesses that hit/missed L1 for given offset within cacheline
242 <<
243 Store Refs - L1 Hit, L1 Miss, N/A <<
244 - % of store accesses that hit L1, missed L1 <<
245 level for given offset within cacheline <<
246 215
247 Data address - Offset 216 Data address - Offset
248 - offset address 217 - offset address
249 218
250 Pid 219 Pid
251 - pid of the process responsible for the acc 220 - pid of the process responsible for the accesses
252 221
253 Tid 222 Tid
254 - tid of the process responsible for the acc 223 - tid of the process responsible for the accesses
255 224
256 Code address 225 Code address
257 - code address responsible for the accesses 226 - code address responsible for the accesses
258 227
259 cycles - rmt hitm, lcl hitm, load (Display w !! 228 cycles - rmt hitm, lcl hitm, load
260 - sum of cycles for given accesses - Remot 229 - sum of cycles for given accesses - Remote/Local HITM and generic load
261 230
262 cycles - rmt peer, lcl peer, load (Display w <<
263 - sum of cycles for given accesses - Remot <<
264 <<
265 cpu cnt 231 cpu cnt
266 - number of cpus that participated on the 232 - number of cpus that participated on the access
267 233
268 Symbol 234 Symbol
269 - code symbol related to the 'Code address 235 - code symbol related to the 'Code address' value
270 236
271 Shared Object 237 Shared Object
272 - shared object name related to the 'Code 238 - shared object name related to the 'Code address' value
273 239
274 Source:Line 240 Source:Line
275 - source information related to the 'Code 241 - source information related to the 'Code address' value
276 242
277 Node 243 Node
278 - nodes participating on the access (see N 244 - nodes participating on the access (see NODE INFO section)
279 245
280 NODE INFO 246 NODE INFO
281 --------- 247 ---------
282 The 'Node' field displays nodes that accesses 248 The 'Node' field displays nodes that accesses given cacheline
283 offset. Its output comes in 3 flavors: 249 offset. Its output comes in 3 flavors:
284 - node IDs separated by ',' 250 - node IDs separated by ','
285 - node IDs with stats for each ID, in follow 251 - node IDs with stats for each ID, in following format:
286 Node{cpus %hitms %stores} (Display with !! 252 Node{cpus %hitms %stores}
287 Node{cpus %peers %stores} (Display with <<
288 - node IDs with list of affected CPUs in fol 253 - node IDs with list of affected CPUs in following format:
289 Node{cpu list} 254 Node{cpu list}
290 255
291 User can switch between above flavors with -N 256 User can switch between above flavors with -N option or
292 use 'n' key to interactively switch in TUI mod 257 use 'n' key to interactively switch in TUI mode.
293 258
294 COALESCE 259 COALESCE
295 -------- 260 --------
296 User can specify how to sort offsets for cache 261 User can specify how to sort offsets for cacheline.
297 262
298 Following fields are available and governs the 263 Following fields are available and governs the final
299 output fields set for cacheline offsets output !! 264 output fields set for caheline offsets output:
300 265
301 tid - coalesced by process TIDs 266 tid - coalesced by process TIDs
302 pid - coalesced by process PIDs 267 pid - coalesced by process PIDs
303 iaddr - coalesced by code address, following 268 iaddr - coalesced by code address, following fields are displayed:
304 Code address, Code symbol, Shared 269 Code address, Code symbol, Shared Object, Source line
305 dso - coalesced by shared object 270 dso - coalesced by shared object
306 271
307 By default the coalescing is setup with 'pid,i 272 By default the coalescing is setup with 'pid,iaddr'.
308 273
309 STDIO OUTPUT 274 STDIO OUTPUT
310 ------------ 275 ------------
311 The stdio output displays data on standard out 276 The stdio output displays data on standard output.
312 277
313 Following tables are displayed: 278 Following tables are displayed:
314 Trace Event Information 279 Trace Event Information
315 - overall statistics of memory accesses 280 - overall statistics of memory accesses
316 281
317 Global Shared Cache Line Event Information 282 Global Shared Cache Line Event Information
318 - overall statistics on shared cachelines 283 - overall statistics on shared cachelines
319 284
320 Shared Data Cache Line Table 285 Shared Data Cache Line Table
321 - list of most expensive cachelines 286 - list of most expensive cachelines
322 287
323 Shared Cache Line Distribution Pareto 288 Shared Cache Line Distribution Pareto
324 - list of all accessed offsets for each cach 289 - list of all accessed offsets for each cacheline
325 290
326 TUI OUTPUT 291 TUI OUTPUT
327 ---------- 292 ----------
328 The TUI output provides interactive interface 293 The TUI output provides interactive interface to navigate
329 through cachelines list and to display offset 294 through cachelines list and to display offset details.
330 295
331 For details please refer to the help window by 296 For details please refer to the help window by pressing '?' key.
332 297
333 CREDITS 298 CREDITS
334 ------- 299 -------
335 Although Don Zickus, Dick Fowles and Joe Mario 300 Although Don Zickus, Dick Fowles and Joe Mario worked together
336 to get this implemented, we got lots of early 301 to get this implemented, we got lots of early help from Arnaldo
337 Carvalho de Melo, Stephane Eranian, Jiri Olsa 302 Carvalho de Melo, Stephane Eranian, Jiri Olsa and Andi Kleen.
338 303
339 C2C BLOG 304 C2C BLOG
340 -------- 305 --------
341 Check Joe's blog on c2c tool for detailed use 306 Check Joe's blog on c2c tool for detailed use case explanation:
342 https://joemario.github.io/blog/2016/09/01/c 307 https://joemario.github.io/blog/2016/09/01/c2c-blog/
343 308
344 SEE ALSO 309 SEE ALSO
345 -------- 310 --------
346 linkperf:perf-record[1], linkperf:perf-mem[1], !! 311 linkperf:perf-record[1], linkperf:perf-mem[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.