1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 * 21 * Portions Copyright 2006-2008 John Birrell jb@freebsd.org 22 * 23 * $FreeBSD$ 24 * 25 */ 26 27 /* 28 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 29 * Use is subject to license terms. 30 */ 31 32 #include <sys/cdefs.h> 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/conf.h> 36 #include <sys/cpuvar.h> 37 #include <sys/fcntl.h> 38 #include <sys/filio.h> 39 #include <sys/kdb.h> 40 #include <sys/kernel.h> 41 #include <sys/kmem.h> 42 #include <sys/kthread.h> 43 #include <sys/limits.h> 44 #include <sys/linker.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/module.h> 48 #include <sys/mutex.h> 49 #include <sys/poll.h> 50 #include <sys/proc.h> 51 #include <sys/selinfo.h> 52 #include <sys/smp.h> 53 #include <sys/uio.h> 54 #include <sys/unistd.h> 55 #include <machine/stdarg.h> 56 57 #include <sys/cyclic.h> 58 #include <sys/dtrace.h> 59 #include <sys/dtrace_bsd.h> 60 61 #define PROF_NAMELEN 15 62 63 #define PROF_PROFILE 0 64 #define PROF_TICK 1 65 #define PROF_PREFIX_PROFILE "profile-" 66 #define PROF_PREFIX_TICK "tick-" 67 68 /* 69 * Regardless of platform, there are five artificial frames in the case of the 70 * profile provider: 71 * 72 * profile_fire 73 * cyclic_expire 74 * cyclic_fire 75 * [ cbe ] 76 * [ locore ] 77 * 78 * On amd64, there are two frames associated with locore: one in locore, and 79 * another in common interrupt dispatch code. (i386 has not been modified to 80 * use this common layer.) Further, on i386, the interrupted instruction 81 * appears as its own stack frame. All of this means that we need to add one 82 * frame for amd64, and then take one away for both amd64 and i386. 83 * 84 * On SPARC, the picture is further complicated because the compiler 85 * optimizes away tail-calls -- so the following frames are optimized away: 86 * 87 * profile_fire 88 * cyclic_expire 89 * 90 * This gives three frames. However, on DEBUG kernels, the cyclic_expire 91 * frame cannot be tail-call eliminated, yielding four frames in this case. 92 * 93 * All of the above constraints lead to the mess below. Yes, the profile 94 * provider should ideally figure this out on-the-fly by hiting one of its own 95 * probes and then walking its own stack trace. This is complicated, however, 96 * and the static definition doesn't seem to be overly brittle. Still, we 97 * allow for a manual override in case we get it completely wrong. 98 */ 99 #ifdef __amd64 100 #define PROF_ARTIFICIAL_FRAMES 7 101 #else 102 #ifdef __i386 103 #define PROF_ARTIFICIAL_FRAMES 6 104 #else 105 #ifdef __sparc 106 #ifdef DEBUG 107 #define PROF_ARTIFICIAL_FRAMES 4 108 #else 109 #define PROF_ARTIFICIAL_FRAMES 3 110 #endif 111 #endif 112 #endif 113 #endif 114 115 #ifdef __mips 116 /* 117 * This value is bogus just to make module compilable on mips 118 */ 119 #define PROF_ARTIFICIAL_FRAMES 3 120 #endif 121 122 typedef struct profile_probe { 123 char prof_name[PROF_NAMELEN]; 124 dtrace_id_t prof_id; 125 int prof_kind; 126 hrtime_t prof_interval; 127 cyclic_id_t prof_cyclic; 128 } profile_probe_t; 129 130 typedef struct profile_probe_percpu { 131 hrtime_t profc_expected; 132 hrtime_t profc_interval; 133 profile_probe_t *profc_probe; 134 } profile_probe_percpu_t; 135 136 static d_open_t profile_open; 137 static int profile_unload(void); 138 static void profile_create(hrtime_t, char *, int); 139 static void profile_destroy(void *, dtrace_id_t, void *); 140 static void profile_enable(void *, dtrace_id_t, void *); 141 static void profile_disable(void *, dtrace_id_t, void *); 142 static void profile_load(void *); 143 static void profile_provide(void *, dtrace_probedesc_t *); 144 145 static int profile_rates[] = { 146 97, 199, 499, 997, 1999, 147 4001, 4999, 0, 0, 0, 148 0, 0, 0, 0, 0, 149 0, 0, 0, 0, 0 150 }; 151 152 static int profile_ticks[] = { 153 1, 10, 100, 500, 1000, 154 5000, 0, 0, 0, 0, 155 0, 0, 0, 0, 0 156 }; 157 158 /* 159 * profile_max defines the upper bound on the number of profile probes that 160 * can exist (this is to prevent malicious or clumsy users from exhausing 161 * system resources by creating a slew of profile probes). At mod load time, 162 * this gets its value from PROFILE_MAX_DEFAULT or profile-max-probes if it's 163 * present in the profile.conf file. 164 */ 165 #define PROFILE_MAX_DEFAULT 1000 /* default max. number of probes */ 166 static uint32_t profile_max = PROFILE_MAX_DEFAULT; 167 /* maximum number of profile probes */ 168 static uint32_t profile_total; /* current number of profile probes */ 169 170 static struct cdevsw profile_cdevsw = { 171 .d_version = D_VERSION, 172 .d_open = profile_open, 173 .d_name = "profile", 174 }; 175 176 static dtrace_pattr_t profile_attr = { 177 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 178 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, 179 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 180 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, 181 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, 182 }; 183 184 static dtrace_pops_t profile_pops = { 185 profile_provide, 186 NULL, 187 profile_enable, 188 profile_disable, 189 NULL, 190 NULL, 191 NULL, 192 NULL, 193 NULL, 194 profile_destroy 195 }; 196 197 static struct cdev *profile_cdev; 198 static dtrace_provider_id_t profile_id; 199 static hrtime_t profile_interval_min = NANOSEC / 5000; /* 5000 hz */ 200 static int profile_aframes = 0; /* override */ 201 202 static void 203 profile_fire(void *arg) 204 { 205 profile_probe_percpu_t *pcpu = arg; 206 profile_probe_t *prof = pcpu->profc_probe; 207 hrtime_t late; 208 solaris_cpu_t *c = &solaris_cpu[curcpu]; 209 210 late = gethrtime() - pcpu->profc_expected; 211 pcpu->profc_expected += pcpu->profc_interval; 212 213 dtrace_probe(prof->prof_id, c->cpu_profile_pc, 214 c->cpu_profile_upc, late, 0, 0); 215 } 216 217 static void 218 profile_tick(void *arg) 219 { 220 profile_probe_t *prof = arg; 221 solaris_cpu_t *c = &solaris_cpu[curcpu]; 222 223 dtrace_probe(prof->prof_id, c->cpu_profile_pc, 224 c->cpu_profile_upc, 0, 0, 0); 225 } 226 227 static void 228 profile_create(hrtime_t interval, char *name, int kind) 229 { 230 profile_probe_t *prof; 231 232 if (interval < profile_interval_min) 233 return; 234 235 if (dtrace_probe_lookup(profile_id, NULL, NULL, name) != 0) 236 return; 237 238 atomic_add_32(&profile_total, 1); 239 if (profile_total > profile_max) { 240 atomic_add_32(&profile_total, -1); 241 return; 242 } 243 244 prof = kmem_zalloc(sizeof (profile_probe_t), KM_SLEEP); 245 (void) strcpy(prof->prof_name, name); 246 prof->prof_interval = interval; 247 prof->prof_cyclic = CYCLIC_NONE; 248 prof->prof_kind = kind; 249 prof->prof_id = dtrace_probe_create(profile_id, 250 NULL, NULL, name, 251 profile_aframes ? profile_aframes : PROF_ARTIFICIAL_FRAMES, prof); 252 } 253 254 /*ARGSUSED*/ 255 static void 256 profile_provide(void *arg, dtrace_probedesc_t *desc) 257 { 258 int i, j, rate, kind; 259 hrtime_t val = 0, mult = 1, len = 0; 260 char *name, *suffix = NULL; 261 262 const struct { 263 char *prefix; 264 int kind; 265 } types[] = { 266 { PROF_PREFIX_PROFILE, PROF_PROFILE }, 267 { PROF_PREFIX_TICK, PROF_TICK }, 268 { 0, 0 } 269 }; 270 271 const struct { 272 char *name; 273 hrtime_t mult; 274 } suffixes[] = { 275 { "ns", NANOSEC / NANOSEC }, 276 { "nsec", NANOSEC / NANOSEC }, 277 { "us", NANOSEC / MICROSEC }, 278 { "usec", NANOSEC / MICROSEC }, 279 { "ms", NANOSEC / MILLISEC }, 280 { "msec", NANOSEC / MILLISEC }, 281 { "s", NANOSEC / SEC }, 282 { "sec", NANOSEC / SEC }, 283 { "m", NANOSEC * (hrtime_t)60 }, 284 { "min", NANOSEC * (hrtime_t)60 }, 285 { "h", NANOSEC * (hrtime_t)(60 * 60) }, 286 { "hour", NANOSEC * (hrtime_t)(60 * 60) }, 287 { "d", NANOSEC * (hrtime_t)(24 * 60 * 60) }, 288 { "day", NANOSEC * (hrtime_t)(24 * 60 * 60) }, 289 { "hz", 0 }, 290 { NULL } 291 }; 292 293 if (desc == NULL) { 294 char n[PROF_NAMELEN]; 295 296 /* 297 * If no description was provided, provide all of our probes. 298 */ 299 for (i = 0; i < sizeof (profile_rates) / sizeof (int); i++) { 300 if ((rate = profile_rates[i]) == 0) 301 continue; 302 303 (void) snprintf(n, PROF_NAMELEN, "%s%d", 304 PROF_PREFIX_PROFILE, rate); 305 profile_create(NANOSEC / rate, n, PROF_PROFILE); 306 } 307 308 for (i = 0; i < sizeof (profile_ticks) / sizeof (int); i++) { 309 if ((rate = profile_ticks[i]) == 0) 310 continue; 311 312 (void) snprintf(n, PROF_NAMELEN, "%s%d", 313 PROF_PREFIX_TICK, rate); 314 profile_create(NANOSEC / rate, n, PROF_TICK); 315 } 316 317 return; 318 } 319 320 name = desc->dtpd_name; 321 322 for (i = 0; types[i].prefix != NULL; i++) { 323 len = strlen(types[i].prefix); 324 325 if (strncmp(name, types[i].prefix, len) != 0) 326 continue; 327 break; 328 } 329 330 if (types[i].prefix == NULL) 331 return; 332 333 kind = types[i].kind; 334 j = strlen(name) - len; 335 336 /* 337 * We need to start before any time suffix. 338 */ 339 for (j = strlen(name); j >= len; j--) { 340 if (name[j] >= '0' && name[j] <= '9') 341 break; 342 suffix = &name[j]; 343 } 344 345 ASSERT(suffix != NULL); 346 347 /* 348 * Now determine the numerical value present in the probe name. 349 */ 350 for (; j >= len; j--) { 351 if (name[j] < '0' || name[j] > '9') 352 return; 353 354 val += (name[j] - '0') * mult; 355 mult *= (hrtime_t)10; 356 } 357 358 if (val == 0) 359 return; 360 361 /* 362 * Look-up the suffix to determine the multiplier. 363 */ 364 for (i = 0, mult = 0; suffixes[i].name != NULL; i++) { 365 if (strcasecmp(suffixes[i].name, suffix) == 0) { 366 mult = suffixes[i].mult; 367 break; 368 } 369 } 370 371 if (suffixes[i].name == NULL && *suffix != '\0') 372 return; 373 374 if (mult == 0) { 375 /* 376 * The default is frequency-per-second. 377 */ 378 val = NANOSEC / val; 379 } else { 380 val *= mult; 381 } 382 383 profile_create(val, name, kind); 384 } 385 386 /* ARGSUSED */ 387 static void 388 profile_destroy(void *arg, dtrace_id_t id, void *parg) 389 { 390 profile_probe_t *prof = parg; 391 392 ASSERT(prof->prof_cyclic == CYCLIC_NONE); 393 kmem_free(prof, sizeof (profile_probe_t)); 394 395 ASSERT(profile_total >= 1); 396 atomic_add_32(&profile_total, -1); 397 } 398 399 /*ARGSUSED*/ 400 static void 401 profile_online(void *arg, cpu_t *cpu, cyc_handler_t *hdlr, cyc_time_t *when) 402 { 403 profile_probe_t *prof = arg; 404 profile_probe_percpu_t *pcpu; 405 406 pcpu = kmem_zalloc(sizeof (profile_probe_percpu_t), KM_SLEEP); 407 pcpu->profc_probe = prof; 408 409 hdlr->cyh_func = profile_fire; 410 hdlr->cyh_arg = pcpu; 411 412 when->cyt_interval = prof->prof_interval; 413 when->cyt_when = gethrtime() + when->cyt_interval; 414 415 pcpu->profc_expected = when->cyt_when; 416 pcpu->profc_interval = when->cyt_interval; 417 } 418 419 /*ARGSUSED*/ 420 static void 421 profile_offline(void *arg, cpu_t *cpu, void *oarg) 422 { 423 profile_probe_percpu_t *pcpu = oarg; 424 425 ASSERT(pcpu->profc_probe == arg); 426 kmem_free(pcpu, sizeof (profile_probe_percpu_t)); 427 } 428 429 /* ARGSUSED */ 430 static void 431 profile_enable(void *arg, dtrace_id_t id, void *parg) 432 { 433 profile_probe_t *prof = parg; 434 cyc_omni_handler_t omni; 435 cyc_handler_t hdlr; 436 cyc_time_t when; 437 438 ASSERT(prof->prof_interval != 0); 439 ASSERT(MUTEX_HELD(&cpu_lock)); 440 441 if (prof->prof_kind == PROF_TICK) { 442 hdlr.cyh_func = profile_tick; 443 hdlr.cyh_arg = prof; 444 445 when.cyt_interval = prof->prof_interval; 446 when.cyt_when = gethrtime() + when.cyt_interval; 447 } else { 448 ASSERT(prof->prof_kind == PROF_PROFILE); 449 omni.cyo_online = profile_online; 450 omni.cyo_offline = profile_offline; 451 omni.cyo_arg = prof; 452 } 453 454 if (prof->prof_kind == PROF_TICK) { 455 prof->prof_cyclic = cyclic_add(&hdlr, &when); 456 } else { 457 prof->prof_cyclic = cyclic_add_omni(&omni); 458 } 459 } 460 461 /* ARGSUSED */ 462 static void 463 profile_disable(void *arg, dtrace_id_t id, void *parg) 464 { 465 profile_probe_t *prof = parg; 466 467 ASSERT(prof->prof_cyclic != CYCLIC_NONE); 468 ASSERT(MUTEX_HELD(&cpu_lock)); 469 470 cyclic_remove(prof->prof_cyclic); 471 prof->prof_cyclic = CYCLIC_NONE; 472 } 473 474 static void 475 profile_load(void *dummy) 476 { 477 /* Create the /dev/dtrace/profile entry. */ 478 profile_cdev = make_dev(&profile_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, 479 "dtrace/profile"); 480 481 if (dtrace_register("profile", &profile_attr, DTRACE_PRIV_USER, 482 NULL, &profile_pops, NULL, &profile_id) != 0) 483 return; 484 } 485 486 487 static int 488 profile_unload() 489 { 490 int error = 0; 491 492 if ((error = dtrace_unregister(profile_id)) != 0) 493 return (error); 494 495 destroy_dev(profile_cdev); 496 497 return (error); 498 } 499 500 /* ARGSUSED */ 501 static int 502 profile_modevent(module_t mod __unused, int type, void *data __unused) 503 { 504 int error = 0; 505 506 switch (type) { 507 case MOD_LOAD: 508 break; 509 510 case MOD_UNLOAD: 511 break; 512 513 case MOD_SHUTDOWN: 514 break; 515 516 default: 517 error = EOPNOTSUPP; 518 break; 519 520 } 521 return (error); 522 } 523 524 /* ARGSUSED */ 525 static int 526 profile_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) 527 { 528 return (0); 529 } 530 531 SYSINIT(profile_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, profile_load, NULL); 532 SYSUNINIT(profile_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, profile_unload, NULL); 533 534 DEV_MODULE(profile, profile_modevent, NULL); 535 MODULE_VERSION(profile, 1); 536 MODULE_DEPEND(profile, dtrace, 1, 1, 1); 537 MODULE_DEPEND(profile, cyclic, 1, 1, 1); 538 MODULE_DEPEND(profile, opensolaris, 1, 1, 1); 539