1 '\" te 2 .\" Copyright (c) 2004, Sun Microsystems, Inc. 3 .\" The contents of this file are subject to the terms of the Common Development and Distribution License (the "License"). You may not use this file except in compliance with the License. 4 .\" You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. See the License for the specific language governing permissions and limitations under the License. 5 .\" When distributing Covered Code, include this CDDL HEADER in each file and include the License file at usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your own identifying information: Portions Copyright [yyyy] [name of copyright owner] 6 .TH CPUTRACK 1 "Apr 19, 2004" 7 .SH NAME 8 cputrack \- monitor process and LWP behavior using CPU performance counters 9 .SH SYNOPSIS 10 .LP 11 .nf 12 \fBcputrack\fR \fB-c\fR \fIeventspec\fR [\fB-c\fR \fIeventspec\fR]... [\fB-efntvD\fR] 13 [\fB-N\fR \fIcount\fR] [\fB-o\fR \fIpathname\fR] [\fB-T\fR \fIinterval\fR] \fIcommand\fR [\fIargs\fR] 14 .fi 15 16 .LP 17 .nf 18 \fBcputrack\fR \fB-c\fR \fIeventspec\fR [\fB-c\fR \fIeventspec\fR]... \fB-p\fR \fIpid\fR [\fB-efntvD\fR] 19 [\fB-N\fR \fIcount\fR] [\fB-o\fR \fIpathname\fR] [\fB-T\fR \fIinterval\fR] 20 .fi 21 22 .LP 23 .nf 24 \fBcputrack\fR \fB-h\fR 25 .fi 26 27 .SH DESCRIPTION 28 .sp 29 .LP 30 The \fBcputrack\fR utility allows \fBCPU\fR performance counters to be used to 31 monitor the behavior of a process or family of processes running on the system. 32 If \fIinterval\fR is specified with the \fB-T\fR option, \fBcputrack\fR samples 33 activity every \fIinterval\fR seconds, repeating forever. If a \fIcount\fR is 34 specified with the \fB-N\fR option, the statistics are repeated \fIcount\fR 35 times for each process tracked. If neither are specified, an interval of one 36 second is used. If \fIcommand\fR and optional \fIargs\fR are specified, 37 \fBcputrack\fR runs the command with the arguments given while monitoring the 38 specified \fBCPU\fR performance events. Alternatively, the process \fBID\fR of 39 an existing process can be specified using the \fB-p\fR option. 40 .sp 41 .LP 42 Because \fBcputrack\fR is an unprivileged program, it is subject to the same 43 restrictions that apply to \fBtruss\fR(1). For example, \fBsetuid\fR(2) 44 executables cannot be tracked. 45 .SH OPTIONS 46 .sp 47 .LP 48 The following options are supported: 49 .sp 50 .ne 2 51 .na 52 \fB\fB-c\fR \fIeventspec\fR\fR 53 .ad 54 .RS 16n 55 Specifies a set of events for the \fBCPU\fR performance counters to monitor. 56 The syntax of these event specifications is: 57 .sp 58 .in +2 59 .nf 60 [picn=]\fIeventn\fR[,attr[\fIn\fR][=\fIval\fR]][,[picn=]\fIeventn\fR 61 [,attr[n][=\fIval\fR]],...,] 62 .fi 63 .in -2 64 .sp 65 66 You can use the \fB-h\fR option to obtain a list of available events and 67 attributes. This causes generation of the usage message. You can omit an 68 explicit counter assignment, in which case \fBcpustat\fR attempts to choose a 69 capable counter automatically. 70 .sp 71 Attribute values can be expressed in hexadecimal, octal, or decimal notation, 72 in a format suitable for \fBstrtoll\fR(3C). An attribute present in the event 73 specification without an explicit value receives a default value of \fB1\fR. An 74 attribute without a corresponding counter number is applied to all counters in 75 the specification. 76 .sp 77 The semantics of these event specifications can be determined by reading the 78 \fBCPU\fR manufacturer's documentation for the events. 79 .sp 80 Multiple \fB-c\fR options can be specified, in which case \fBcputrack\fR cycles 81 between the different event settings on each sample. 82 .RE 83 84 .sp 85 .ne 2 86 .na 87 \fB\fB-D\fR\fR 88 .ad 89 .RS 16n 90 Enables debug mode. 91 .RE 92 93 .sp 94 .ne 2 95 .na 96 \fB\fB-e\fR\fR 97 .ad 98 .RS 16n 99 Follows all \fBexec\fR(2), or \fBexecve\fR(2) system calls. 100 .RE 101 102 .sp 103 .ne 2 104 .na 105 \fB\fB-f\fR\fR 106 .ad 107 .RS 16n 108 Follows all children created by \fBfork\fR(2), \fBfork1\fR(2), or 109 \fBvfork\fR(2) system calls. 110 .RE 111 112 .sp 113 .ne 2 114 .na 115 \fB\fB-h\fR\fR 116 .ad 117 .RS 16n 118 Prints an extended help message on how to use the utility, how to program the 119 processor-dependent counters, and where to look for more detailed information. 120 .RE 121 122 .sp 123 .ne 2 124 .na 125 \fB\fB-n\fR\fR 126 .ad 127 .RS 16n 128 Omits all header output (useful if \fBcputrack\fR is the beginning of a 129 pipeline). 130 .RE 131 132 .sp 133 .ne 2 134 .na 135 \fB\fB-N\fR \fIcount\fR\fR 136 .ad 137 .RS 16n 138 Specifies the maximum number of \fBCPU\fR performance counter samples to take 139 before exiting. 140 .RE 141 142 .sp 143 .ne 2 144 .na 145 \fB\fB-o\fR \fIoutfile\fR\fR 146 .ad 147 .RS 16n 148 Specifies file to be used for the \fBcputrack\fR output. 149 .RE 150 151 .sp 152 .ne 2 153 .na 154 \fB\fB-p\fR \fIpid\fR\fR 155 .ad 156 .RS 16n 157 Interprets the argument as the process \fBID\fR of an existing process to which 158 process counter context should be attached and monitored. 159 .RE 160 161 .sp 162 .ne 2 163 .na 164 \fB\fB-t\fR\fR 165 .ad 166 .RS 16n 167 Prints an additional column of processor cycle counts, if available on the 168 current architecture. 169 .RE 170 171 .sp 172 .ne 2 173 .na 174 \fB\fB-T\fR \fIinterval\fR\fR 175 .ad 176 .RS 16n 177 Specifies the interval between \fBCPU\fR performance counter samples in 178 seconds. Very small intervals may cause some samples to be skipped. See 179 WARNINGS. 180 .RE 181 182 .sp 183 .ne 2 184 .na 185 \fB\fB-v\fR\fR 186 .ad 187 .RS 16n 188 Enables more verbose output. 189 .RE 190 191 .SH USAGE 192 .sp 193 .LP 194 The operating system enforces certain restrictions on the tracing of processes. 195 In particular, a command whose object file cannot be read by a user cannot be 196 tracked by that user; set-uid and set-gid commands can only be tracked by a 197 privileged user. Unless it is run by a privileged user, \fBcputrack\fR loses 198 control of any process that performs an \fBexec()\fR of a set-id or unreadable 199 object file. Such processes continue normally, though independently of 200 \fBcputrack\fR, from the point of the \fBexec()\fR. 201 .sp 202 .LP 203 The system may run out of per-user process slots when the \fB-f\fR option is 204 used, since \fBcputrack\fR runs one controlling process for each process being 205 tracked. 206 .sp 207 .LP 208 The times printed by \fBcputrack\fR correspond to the wallclock time when the 209 hardware counters were actually sample. The time is derived from the same 210 timebase as \fBgethrtime\fR(3C). 211 .sp 212 .LP 213 The \fBcputrack\fR utility attaches performance counter context to each process 214 that it examines. The presence of this context allows the performance counters 215 to be multiplexed between different processes on the system, but it cannot be 216 used at the same time as the \fBcpustat\fR(1M) utility. 217 .sp 218 .LP 219 Once an instance of the \fBcpustat\fR utility is running, further attempts to 220 run \fBcputrack\fR will fail until all instances of \fBcpustat\fR terminate. 221 .sp 222 .LP 223 Sometimes \fBcputrack\fR provides sufficient flexibility and prints sufficient 224 statistics to make adding the observation code to an application unnecessary. 225 However, more control is occasionally desired. Because the same performance 226 counter context is used by both the application itself and by the agent LWP 227 injected into the application by \fBcputrack\fR, it is possible for an 228 application to interact with the counter context to achieve some interesting 229 capabilities. See \fBcpc_enable\fR(3CPC). 230 .sp 231 .LP 232 The processor cycle counts enabled by the \fB-t\fR option always apply to both 233 user and system modes, regardless of the settings applied to the performance 234 counter registers. 235 .sp 236 .LP 237 The output of \fBcputrack\fR is designed to be readily parseable by 238 \fBawk\fR(1) and \fBperl\fR(1), thereby allowing performance tools to be 239 composed by embedding \fBcputrack\fR in scripts. Alternatively, tools may be 240 constructed directly using the same \fBAPI\fRs that \fBcputrack\fR is built 241 upon, using the facilities of \fBlibcpc\fR(3LIB) and \fBlibpctx\fR(3LIB). See 242 \fBcpc\fR(3CPC). 243 .sp 244 .LP 245 Although \fBcputrack\fR uses performance counter context to maintain separate 246 performance counter values for each LWP, some of the events that can be counted 247 will inevitably be impacted by other activities occurring on the system, 248 particularly for limited resources that are shared between processes (for 249 example, cache miss rates). For such events, it may also be interesting to 250 observe overall system behavior with \fBcpustat\fR(1M). 251 .sp 252 .LP 253 For the \fB-T\fR \fIinterval\fR option, if \fIinterval\fR is specified as zero, 254 no periodic sampling is performed. The performance counters are only sampled 255 when the process creates or destroys an \fBLWP\fR, or it invokes \fBfork\fR(2), 256 \fBexec\fR(2), or \fBexit\fR(2). 257 .SH EXAMPLES 258 .SS "SPARC" 259 .LP 260 \fBExample 1 \fRUsing Performance Counters to Count Clock Cycles 261 .sp 262 .LP 263 In this example, the utility is being used on a machine containing an 264 UltraSPARC-III+ processor. The counters are set to count processor clock cycles 265 and instructions dispatched in user mode while running the \fBsleep\fR(1) 266 command. 267 268 .sp 269 .in +2 270 .nf 271 example% \fBcputrack -c pic0=Cycle_cnt,pic1=Instr_cnt sleep 10\fR 272 273 274 time lwp event pic0 pic1 275 1.007 1 tick 765308 219233 276 2.007 1 tick 0 0 277 4.017 1 tick 0 0 278 6.007 1 tick 0 0 279 8.007 1 tick 0 0 280 10.007 1 tick 0 0 281 10.017 1 exit 844703 228058 282 283 .fi 284 .in -2 285 .sp 286 287 .LP 288 \fBExample 2 \fRCounting External Cache References and Misses 289 .sp 290 .LP 291 This example shows more verbose output while following the \fBfork()\fR and 292 \fBexec()\fR of a simple shell script on an UltraSPARC machine. The counters 293 are measuring the number of external cache references and external cache 294 misses. Notice that the explicit \fBpic0\fR and \fBpic1\fR names can be omitted 295 where there are no ambiguities. 296 297 .sp 298 .in +2 299 .nf 300 example% \fBcputrack -fev -c EC_ref,EC_hit /bin/ulimit -c\fR 301 302 303 time pid lwp event pic0 pic1 304 0.007 101142 1 init_lwp 805286 20023 305 0.023 101142 1 fork # 101143 306 0.026 101143 1 init_lwp 1015382 24461 307 0.029 101143 1 fini_lwp 1025546 25074 308 0.029 101143 1 exec 1025546 25074 309 0.000 101143 1 exec \e 310 # '/usr/bin/sh /usr/bin/basename\e 311 /bin/ulimit' 312 0.039 101143 1 init_lwp 1025546 25074 313 0.050 101143 1 fini_lwp 1140482 27806 314 0.050 101143 1 exec 1140482 27806 315 0.000 101143 1 exec # '/usr/bin/expr \e 316 //bin/ulimit : \(.*[^/]\)/*$ : .*/\(..*\) : \(.*\)$ | //bin/ulimi' 317 0.059 101143 1 init_lwp 1140482 27806 318 0.075 101143 1 fini_lwp 1237647 30207 319 0.075 101143 1 exit 1237647 30207 320 unlimited 321 0.081 101142 1 fini_lwp 953383 23814 322 0.081 101142 1 exit 953383 23814 323 .fi 324 .in -2 325 .sp 326 327 .SS "x86" 328 .LP 329 \fBExample 3 \fRCounting Instructions 330 .sp 331 .LP 332 This example shows how many instructions were executed in the application and 333 in the kernel to print the date on a Pentium III machine: 334 335 .sp 336 .in +2 337 .nf 338 example% \fBcputrack -c inst_retired,inst_retired,nouser1,sys1 date\fR 339 340 341 time lwp event pic0 pic1 342 Fri Aug 20 20:03:08 PDT 1999 343 0.072 1 exit 246725 339666 344 .fi 345 .in -2 346 .sp 347 348 .LP 349 \fBExample 4 \fRCounting TLB Hits 350 .sp 351 .LP 352 This example shows how to use processor-specific attributes to count TLB hits 353 on a Pentium 4 machine: 354 355 .sp 356 .in +2 357 .nf 358 example% \fBcputrack -c ITLB_reference,emask=1 date\fR 359 360 361 time lwp event pic0 362 Fri Aug 20 20:03:08 PDT 1999 363 0.072 1 exit 246725 364 .fi 365 .in -2 366 .sp 367 368 .SH WARNINGS 369 .sp 370 .LP 371 By running any instance of the \fBcpustat\fR(1M) utility, all existing 372 performance counter context is forcibly invalidated across the machine. This 373 may in turn cause all invocations of the \fBcputrack\fR command to exit 374 prematurely with unspecified errors. 375 .sp 376 .LP 377 If \fBcpustat\fR is invoked on a system that has \fBCPU\fR performance counters 378 which are not supported by Solaris, the following message appears: 379 .sp 380 .in +2 381 .nf 382 cputrack: cannot access performance counters - Operation not applicable 383 .fi 384 .in -2 385 .sp 386 387 .sp 388 .LP 389 This error message implies that \fBcpc_open()\fR has failed and is documented 390 in \fBcpc_open\fR(3CPC). Review this documentation for more information about 391 the problem and possible solutions. 392 .sp 393 .LP 394 If a short interval is requested, \fBcputrack\fR may not be able to keep up 395 with the desired sample rate. In this case, some samples may be dropped. 396 .SH ATTRIBUTES 397 .sp 398 .LP 399 See \fBattributes\fR(5) for descriptions of the following attributes: 400 .sp 401 402 .sp 403 .TS 404 box; 405 c | c 406 l | l . 407 ATTRIBUTE TYPE ATTRIBUTE VALUE 408 Interface Stability Evolving 409 .TE 410 411 .SH SEE ALSO 412 .sp 413 .LP 414 \fBawk\fR(1), \fBperl\fR(1), \fBproc\fR(1), \fBtruss\fR(1), \fBprstat\fR(1M), 415 \fBcpustat\fR(1M), \fBexec\fR(2), \fBexit\fR(2), \fBfork\fR(2), 416 \fBsetuid\fR(2), \fBvfork\fR(2), \fBgethrtime\fR(3C), \fBstrtoll\fR(3C), 417 \fBcpc\fR(3CPC), \fBcpc_bind_pctx\fR(3CPC), \fBcpc_enable\fR(3CPC), 418 \fBcpc_open\fR(3CPC), \fBlibcpc\fR(3LIB), \fBlibpctx\fR(3LIB), \fBproc\fR(4), 419 \fBattributes\fR(5)