| 1 | /* |
| 2 | * Copyright (c) 2011 Apple Computer, Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | |
| 29 | /* sysctl interface for parameters from user-land */ |
| 30 | |
| 31 | #include <kern/debug.h> |
| 32 | #include <libkern/libkern.h> |
| 33 | #include <pexpert/pexpert.h> |
| 34 | #include <sys/param.h> |
| 35 | #include <sys/mman.h> |
| 36 | #include <sys/stat.h> |
| 37 | #include <sys/sysctl.h> |
| 38 | #include <sys/kauth.h> |
| 39 | |
| 40 | #include <kperf/action.h> |
| 41 | #include <kperf/context.h> |
| 42 | #include <kperf/kdebug_trigger.h> |
| 43 | #include <kperf/kperf.h> |
| 44 | #include <kperf/kperfbsd.h> |
| 45 | #include <kperf/kptimer.h> |
| 46 | #include <kperf/pet.h> |
| 47 | #include <kperf/lazy.h> |
| 48 | |
| 49 | #include <sys/ktrace.h> |
| 50 | |
| 51 | /* Requests from kperf sysctls. */ |
| 52 | enum kperf_request { |
| 53 | REQ_SAMPLING, |
| 54 | REQ_RESET, |
| 55 | |
| 56 | REQ_ACTION_COUNT, |
| 57 | REQ_ACTION_SAMPLERS, |
| 58 | REQ_ACTION_USERDATA, |
| 59 | REQ_ACTION_FILTER_BY_TASK, |
| 60 | REQ_ACTION_FILTER_BY_PID, |
| 61 | REQ_ACTION_UCALLSTACK_DEPTH, |
| 62 | REQ_ACTION_KCALLSTACK_DEPTH, |
| 63 | |
| 64 | REQ_TIMER_COUNT, |
| 65 | REQ_TIMER_PERIOD, |
| 66 | REQ_TIMER_PET, |
| 67 | REQ_TIMER_ACTION, |
| 68 | |
| 69 | REQ_KDBG_CSWITCH, |
| 70 | |
| 71 | REQ_BLESS, |
| 72 | REQ_BLESS_PREEMPT, |
| 73 | |
| 74 | REQ_PET_IDLE_RATE, |
| 75 | REQ_LIGHTWEIGHT_PET, |
| 76 | |
| 77 | REQ_KDEBUG_FILTER, |
| 78 | REQ_KDEBUG_ACTION, |
| 79 | |
| 80 | REQ_LAZY_WAIT_TIME_THRESHOLD, |
| 81 | REQ_LAZY_WAIT_ACTION, |
| 82 | REQ_LAZY_CPU_TIME_THRESHOLD, |
| 83 | REQ_LAZY_CPU_ACTION, |
| 84 | }; |
| 85 | |
| 86 | int kperf_debug_level = 0; |
| 87 | |
| 88 | #if DEVELOPMENT || DEBUG |
| 89 | _Atomic long long kperf_pending_ipis = 0; |
| 90 | #endif /* DEVELOPMENT || DEBUG */ |
| 91 | |
| 92 | /* |
| 93 | * kperf has unique requirements from sysctl. |
| 94 | * |
| 95 | * For simple queries like the number of actions, the normal sysctl style |
| 96 | * of get/set works well. |
| 97 | * |
| 98 | * However, when requesting information about something specific, like an |
| 99 | * action, user space needs to provide some contextual information. This |
| 100 | * information is stored in a uint64_t array that includes the context, like |
| 101 | * the action ID it is interested in. If user space is getting the value from |
| 102 | * the kernel, then the get side of the sysctl is valid. If it is setting the |
| 103 | * value, then the get pointers are left NULL. |
| 104 | * |
| 105 | * These functions handle marshalling and unmarshalling data from sysctls. |
| 106 | */ |
| 107 | |
| 108 | static int |
| 109 | kperf_sysctl_get_set_uint32(struct sysctl_req *req, |
| 110 | uint32_t (*get)(void), int (*set)(uint32_t)) |
| 111 | { |
| 112 | assert(req != NULL); |
| 113 | assert(get != NULL); |
| 114 | assert(set != NULL); |
| 115 | |
| 116 | uint32_t value = 0; |
| 117 | if (req->oldptr) { |
| 118 | value = get(); |
| 119 | } |
| 120 | |
| 121 | int error = sysctl_io_number(req, bigValue: value, valueSize: sizeof(value), pValue: &value, NULL); |
| 122 | |
| 123 | if (error || !req->newptr) { |
| 124 | return error; |
| 125 | } |
| 126 | |
| 127 | return set(value); |
| 128 | } |
| 129 | |
| 130 | static int |
| 131 | kperf_sysctl_get_set_int(struct sysctl_req *req, |
| 132 | int (*get)(void), int (*set)(int)) |
| 133 | { |
| 134 | assert(req != NULL); |
| 135 | assert(get != NULL); |
| 136 | assert(set != NULL); |
| 137 | |
| 138 | int value = 0; |
| 139 | if (req->oldptr) { |
| 140 | value = get(); |
| 141 | } |
| 142 | |
| 143 | int error = sysctl_io_number(req, bigValue: value, valueSize: sizeof(value), pValue: &value, NULL); |
| 144 | |
| 145 | if (error || !req->newptr) { |
| 146 | return error; |
| 147 | } |
| 148 | |
| 149 | return set(value); |
| 150 | } |
| 151 | |
| 152 | static int |
| 153 | kperf_sysctl_get_set_uint64(struct sysctl_req *req, |
| 154 | uint64_t (*get)(void), int (*set)(uint64_t)) |
| 155 | { |
| 156 | assert(req != NULL); |
| 157 | assert(get != NULL); |
| 158 | assert(set != NULL); |
| 159 | |
| 160 | uint64_t value = 0; |
| 161 | if (req->oldptr) { |
| 162 | value = get(); |
| 163 | } |
| 164 | |
| 165 | int error = sysctl_io_number(req, bigValue: (long long)value, valueSize: sizeof(value), pValue: &value, NULL); |
| 166 | |
| 167 | if (error || !req->newptr) { |
| 168 | return error; |
| 169 | } |
| 170 | |
| 171 | return set(value); |
| 172 | } |
| 173 | |
| 174 | static int |
| 175 | kperf_sysctl_get_set_unsigned_uint32(struct sysctl_req *req, |
| 176 | int (*get)(unsigned int, uint32_t *), int (*set)(unsigned int, uint32_t)) |
| 177 | { |
| 178 | assert(req != NULL); |
| 179 | assert(get != NULL); |
| 180 | assert(set != NULL); |
| 181 | |
| 182 | int error = 0; |
| 183 | uint64_t inputs[2] = {}; |
| 184 | |
| 185 | if (req->newptr == USER_ADDR_NULL) { |
| 186 | return EFAULT; |
| 187 | } |
| 188 | |
| 189 | if ((error = copyin(req->newptr, inputs, sizeof(inputs)))) { |
| 190 | return error; |
| 191 | } |
| 192 | |
| 193 | unsigned int action_id = (unsigned int)inputs[0]; |
| 194 | uint32_t new_value = (uint32_t)inputs[1]; |
| 195 | |
| 196 | if (req->oldptr != USER_ADDR_NULL) { |
| 197 | uint32_t value_out = 0; |
| 198 | if ((error = get(action_id, &value_out))) { |
| 199 | return error; |
| 200 | } |
| 201 | |
| 202 | inputs[1] = value_out; |
| 203 | |
| 204 | return copyout(inputs, req->oldptr, sizeof(inputs)); |
| 205 | } else { |
| 206 | return set(action_id, new_value); |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | /* |
| 211 | * These functions are essentially the same as the generic |
| 212 | * kperf_sysctl_get_set_unsigned_uint32, except they have unique input sizes. |
| 213 | */ |
| 214 | |
| 215 | static int |
| 216 | sysctl_timer_period(struct sysctl_req *req) |
| 217 | { |
| 218 | uint64_t inputs[2] = {}; |
| 219 | |
| 220 | if (req->newptr == USER_ADDR_NULL) { |
| 221 | return EFAULT; |
| 222 | } |
| 223 | |
| 224 | int error = 0; |
| 225 | if ((error = copyin(req->newptr, inputs, sizeof(inputs)))) { |
| 226 | return error; |
| 227 | } |
| 228 | unsigned int timer = (unsigned int)inputs[0]; |
| 229 | uint64_t new_period = inputs[1]; |
| 230 | |
| 231 | if (req->oldptr != USER_ADDR_NULL) { |
| 232 | uint64_t period_out = 0; |
| 233 | if ((error = kptimer_get_period(timerid: timer, period_out: &period_out))) { |
| 234 | return error; |
| 235 | } |
| 236 | |
| 237 | inputs[1] = period_out; |
| 238 | return copyout(inputs, req->oldptr, sizeof(inputs)); |
| 239 | } else { |
| 240 | return kptimer_set_period(timerid: timer, period: new_period); |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | static int |
| 245 | sysctl_action_filter(struct sysctl_req *req, bool is_task_t) |
| 246 | { |
| 247 | int error = 0; |
| 248 | uint64_t inputs[2] = {}; |
| 249 | |
| 250 | assert(req != NULL); |
| 251 | |
| 252 | if (req->newptr == USER_ADDR_NULL) { |
| 253 | return EFAULT; |
| 254 | } |
| 255 | |
| 256 | if ((error = copyin(req->newptr, inputs, sizeof(inputs)))) { |
| 257 | return error; |
| 258 | } |
| 259 | |
| 260 | unsigned int actionid = (unsigned int)inputs[0]; |
| 261 | int new_filter = (int)inputs[1]; |
| 262 | |
| 263 | if (req->oldptr != USER_ADDR_NULL) { |
| 264 | int filter_out; |
| 265 | if ((error = kperf_action_get_filter(actionid, pid_out: &filter_out))) { |
| 266 | return error; |
| 267 | } |
| 268 | |
| 269 | inputs[1] = (uint64_t)filter_out; |
| 270 | return copyout(inputs, req->oldptr, sizeof(inputs)); |
| 271 | } else { |
| 272 | int pid = is_task_t ? kperf_port_to_pid(portname: (mach_port_name_t)new_filter) |
| 273 | : new_filter; |
| 274 | |
| 275 | return kperf_action_set_filter(actionid, pid); |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | static int |
| 280 | sysctl_bless(struct sysctl_req *req) |
| 281 | { |
| 282 | int value = ktrace_get_owning_pid(); |
| 283 | int error = sysctl_io_number(req, bigValue: value, valueSize: sizeof(value), pValue: &value, NULL); |
| 284 | |
| 285 | if (error || !req->newptr) { |
| 286 | return error; |
| 287 | } |
| 288 | |
| 289 | return ktrace_set_owning_pid(pid: value); |
| 290 | } |
| 291 | |
| 292 | /* sysctl handlers that use the generic functions */ |
| 293 | |
| 294 | static int |
| 295 | sysctl_action_samplers(struct sysctl_req *req) |
| 296 | { |
| 297 | return kperf_sysctl_get_set_unsigned_uint32(req, |
| 298 | get: kperf_action_get_samplers, set: kperf_action_set_samplers); |
| 299 | } |
| 300 | |
| 301 | static int |
| 302 | sysctl_action_userdata(struct sysctl_req *req) |
| 303 | { |
| 304 | return kperf_sysctl_get_set_unsigned_uint32(req, |
| 305 | get: kperf_action_get_userdata, set: kperf_action_set_userdata); |
| 306 | } |
| 307 | |
| 308 | static int |
| 309 | sysctl_action_ucallstack_depth(struct sysctl_req *req) |
| 310 | { |
| 311 | return kperf_sysctl_get_set_unsigned_uint32(req, |
| 312 | get: kperf_action_get_ucallstack_depth, set: kperf_action_set_ucallstack_depth); |
| 313 | } |
| 314 | |
| 315 | static int |
| 316 | sysctl_action_kcallstack_depth(struct sysctl_req *req) |
| 317 | { |
| 318 | return kperf_sysctl_get_set_unsigned_uint32(req, |
| 319 | get: kperf_action_get_kcallstack_depth, set: kperf_action_set_kcallstack_depth); |
| 320 | } |
| 321 | |
| 322 | static int |
| 323 | sysctl_kdebug_action(struct sysctl_req *req) |
| 324 | { |
| 325 | return kperf_sysctl_get_set_int(req, get: kperf_kdebug_get_action, |
| 326 | set: kperf_kdebug_set_action); |
| 327 | } |
| 328 | |
| 329 | static int |
| 330 | sysctl_kdebug_filter(struct sysctl_req *req) |
| 331 | { |
| 332 | assert(req != NULL); |
| 333 | |
| 334 | if (req->oldptr != USER_ADDR_NULL) { |
| 335 | struct kperf_kdebug_filter *filter = NULL; |
| 336 | uint32_t n_debugids = kperf_kdebug_get_filter(filter: &filter); |
| 337 | size_t filter_size = KPERF_KDEBUG_FILTER_SIZE(n_debugids); |
| 338 | |
| 339 | if (n_debugids == 0) { |
| 340 | return EINVAL; |
| 341 | } |
| 342 | |
| 343 | return SYSCTL_OUT(req, filter, filter_size); |
| 344 | } else if (req->newptr != USER_ADDR_NULL) { |
| 345 | return kperf_kdebug_set_filter(user_filter: req->newptr, user_size: (uint32_t)req->newlen); |
| 346 | } else { |
| 347 | return EINVAL; |
| 348 | } |
| 349 | } |
| 350 | |
| 351 | static uint32_t |
| 352 | kperf_sampling_get(void) |
| 353 | { |
| 354 | return kperf_is_sampling(); |
| 355 | } |
| 356 | |
| 357 | static int |
| 358 | kperf_sampling_set(uint32_t sample_start) |
| 359 | { |
| 360 | if (sample_start) { |
| 361 | return kperf_enable_sampling(); |
| 362 | } else { |
| 363 | return kperf_disable_sampling(); |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | static int |
| 368 | sysctl_sampling(struct sysctl_req *req) |
| 369 | { |
| 370 | return kperf_sysctl_get_set_uint32(req, get: kperf_sampling_get, |
| 371 | set: kperf_sampling_set); |
| 372 | } |
| 373 | |
| 374 | static int |
| 375 | sysctl_action_count(struct sysctl_req *req) |
| 376 | { |
| 377 | return kperf_sysctl_get_set_uint32(req, get: kperf_action_get_count, |
| 378 | set: kperf_action_set_count); |
| 379 | } |
| 380 | |
| 381 | static int |
| 382 | sysctl_timer_count(struct sysctl_req *req) |
| 383 | { |
| 384 | return kperf_sysctl_get_set_uint32(req, get: kptimer_get_count, |
| 385 | set: kptimer_set_count); |
| 386 | } |
| 387 | |
| 388 | static int |
| 389 | sysctl_timer_action(struct sysctl_req *req) |
| 390 | { |
| 391 | return kperf_sysctl_get_set_unsigned_uint32(req, get: kptimer_get_action, |
| 392 | set: kptimer_set_action); |
| 393 | } |
| 394 | |
| 395 | static int |
| 396 | sysctl_timer_pet(struct sysctl_req *req) |
| 397 | { |
| 398 | return kperf_sysctl_get_set_uint32(req, get: kptimer_get_pet_timerid, |
| 399 | set: kptimer_set_pet_timerid); |
| 400 | } |
| 401 | |
| 402 | static int |
| 403 | sysctl_bless_preempt(struct sysctl_req *req) |
| 404 | { |
| 405 | return sysctl_io_number(req, bigValue: ktrace_root_set_owner_allowed, |
| 406 | valueSize: sizeof(ktrace_root_set_owner_allowed), |
| 407 | pValue: &ktrace_root_set_owner_allowed, NULL); |
| 408 | } |
| 409 | |
| 410 | static int |
| 411 | sysctl_kperf_reset(struct sysctl_req *req) |
| 412 | { |
| 413 | int should_reset = 0; |
| 414 | |
| 415 | int error = sysctl_io_number(req, bigValue: should_reset, valueSize: sizeof(should_reset), |
| 416 | pValue: &should_reset, NULL); |
| 417 | if (error) { |
| 418 | return error; |
| 419 | } |
| 420 | |
| 421 | if (should_reset) { |
| 422 | ktrace_reset(KTRACE_KPERF); |
| 423 | } |
| 424 | return 0; |
| 425 | } |
| 426 | |
| 427 | static int |
| 428 | sysctl_pet_idle_rate(struct sysctl_req *req) |
| 429 | { |
| 430 | return kperf_sysctl_get_set_int(req, get: kppet_get_idle_rate, |
| 431 | set: kppet_set_idle_rate); |
| 432 | } |
| 433 | |
| 434 | static int |
| 435 | sysctl_lightweight_pet(struct sysctl_req *req) |
| 436 | { |
| 437 | return kperf_sysctl_get_set_int(req, get: kppet_get_lightweight_pet, |
| 438 | set: kppet_set_lightweight_pet); |
| 439 | } |
| 440 | |
| 441 | static int |
| 442 | sysctl_kdbg_cswitch(struct sysctl_req *req) |
| 443 | { |
| 444 | return kperf_sysctl_get_set_int(req, get: kperf_kdbg_cswitch_get, |
| 445 | set: kperf_kdbg_cswitch_set); |
| 446 | } |
| 447 | |
| 448 | static int |
| 449 | sysctl_lazy_wait_time_threshold(struct sysctl_req *req) |
| 450 | { |
| 451 | return kperf_sysctl_get_set_uint64(req, get: kperf_lazy_get_wait_time_threshold, |
| 452 | set: kperf_lazy_set_wait_time_threshold); |
| 453 | } |
| 454 | |
| 455 | static int |
| 456 | sysctl_lazy_wait_action(struct sysctl_req *req) |
| 457 | { |
| 458 | return kperf_sysctl_get_set_int(req, get: kperf_lazy_get_wait_action, |
| 459 | set: kperf_lazy_set_wait_action); |
| 460 | } |
| 461 | |
| 462 | static int |
| 463 | sysctl_lazy_cpu_time_threshold(struct sysctl_req *req) |
| 464 | { |
| 465 | return kperf_sysctl_get_set_uint64(req, get: kperf_lazy_get_cpu_time_threshold, |
| 466 | set: kperf_lazy_set_cpu_time_threshold); |
| 467 | } |
| 468 | |
| 469 | static int |
| 470 | sysctl_lazy_cpu_action(struct sysctl_req *req) |
| 471 | { |
| 472 | return kperf_sysctl_get_set_int(req, get: kperf_lazy_get_cpu_action, |
| 473 | set: kperf_lazy_set_cpu_action); |
| 474 | } |
| 475 | |
| 476 | static int |
| 477 | kperf_sysctl SYSCTL_HANDLER_ARGS |
| 478 | { |
| 479 | #pragma unused(oidp, arg2) |
| 480 | int ret; |
| 481 | enum kperf_request type = (enum kperf_request)arg1; |
| 482 | |
| 483 | ktrace_lock(); |
| 484 | |
| 485 | if (req->oldptr == USER_ADDR_NULL && req->newptr != USER_ADDR_NULL) { |
| 486 | if ((ret = ktrace_configure(KTRACE_KPERF))) { |
| 487 | ktrace_unlock(); |
| 488 | return ret; |
| 489 | } |
| 490 | } else { |
| 491 | if ((ret = ktrace_read_check())) { |
| 492 | ktrace_unlock(); |
| 493 | return ret; |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | /* which request */ |
| 498 | switch (type) { |
| 499 | case REQ_ACTION_COUNT: |
| 500 | ret = sysctl_action_count(req); |
| 501 | break; |
| 502 | case REQ_ACTION_SAMPLERS: |
| 503 | ret = sysctl_action_samplers(req); |
| 504 | break; |
| 505 | case REQ_ACTION_USERDATA: |
| 506 | ret = sysctl_action_userdata(req); |
| 507 | break; |
| 508 | case REQ_TIMER_COUNT: |
| 509 | ret = sysctl_timer_count(req); |
| 510 | break; |
| 511 | case REQ_TIMER_PERIOD: |
| 512 | ret = sysctl_timer_period(req); |
| 513 | break; |
| 514 | case REQ_TIMER_PET: |
| 515 | ret = sysctl_timer_pet(req); |
| 516 | break; |
| 517 | case REQ_TIMER_ACTION: |
| 518 | ret = sysctl_timer_action(req); |
| 519 | break; |
| 520 | case REQ_SAMPLING: |
| 521 | ret = sysctl_sampling(req); |
| 522 | break; |
| 523 | case REQ_KDBG_CSWITCH: |
| 524 | ret = sysctl_kdbg_cswitch(req); |
| 525 | break; |
| 526 | case REQ_ACTION_FILTER_BY_TASK: |
| 527 | ret = sysctl_action_filter(req, true); |
| 528 | break; |
| 529 | case REQ_ACTION_FILTER_BY_PID: |
| 530 | ret = sysctl_action_filter(req, false); |
| 531 | break; |
| 532 | case REQ_KDEBUG_ACTION: |
| 533 | ret = sysctl_kdebug_action(req); |
| 534 | break; |
| 535 | case REQ_KDEBUG_FILTER: |
| 536 | ret = sysctl_kdebug_filter(req); |
| 537 | break; |
| 538 | case REQ_PET_IDLE_RATE: |
| 539 | ret = sysctl_pet_idle_rate(req); |
| 540 | break; |
| 541 | case REQ_BLESS_PREEMPT: |
| 542 | ret = sysctl_bless_preempt(req); |
| 543 | break; |
| 544 | case REQ_RESET: |
| 545 | ret = sysctl_kperf_reset(req); |
| 546 | break; |
| 547 | case REQ_ACTION_UCALLSTACK_DEPTH: |
| 548 | ret = sysctl_action_ucallstack_depth(req); |
| 549 | break; |
| 550 | case REQ_ACTION_KCALLSTACK_DEPTH: |
| 551 | ret = sysctl_action_kcallstack_depth(req); |
| 552 | break; |
| 553 | case REQ_LIGHTWEIGHT_PET: |
| 554 | ret = sysctl_lightweight_pet(req); |
| 555 | break; |
| 556 | case REQ_LAZY_WAIT_TIME_THRESHOLD: |
| 557 | ret = sysctl_lazy_wait_time_threshold(req); |
| 558 | break; |
| 559 | case REQ_LAZY_WAIT_ACTION: |
| 560 | ret = sysctl_lazy_wait_action(req); |
| 561 | break; |
| 562 | case REQ_LAZY_CPU_TIME_THRESHOLD: |
| 563 | ret = sysctl_lazy_cpu_time_threshold(req); |
| 564 | break; |
| 565 | case REQ_LAZY_CPU_ACTION: |
| 566 | ret = sysctl_lazy_cpu_action(req); |
| 567 | break; |
| 568 | default: |
| 569 | ret = ENOENT; |
| 570 | break; |
| 571 | } |
| 572 | |
| 573 | ktrace_unlock(); |
| 574 | |
| 575 | return ret; |
| 576 | } |
| 577 | |
| 578 | static int |
| 579 | kperf_sysctl_bless_handler SYSCTL_HANDLER_ARGS |
| 580 | { |
| 581 | #pragma unused(oidp, arg2) |
| 582 | int ret; |
| 583 | |
| 584 | ktrace_lock(); |
| 585 | |
| 586 | /* if setting a new "blessed pid" (ktrace owning pid) */ |
| 587 | if (req->newptr != USER_ADDR_NULL) { |
| 588 | /* |
| 589 | * root can bypass the ktrace check when a flag is set (for |
| 590 | * backwards compatibility) or when ownership is maintained over |
| 591 | * subsystems resets (to allow the user space process that set |
| 592 | * ownership to unset it). |
| 593 | */ |
| 594 | if (!((ktrace_root_set_owner_allowed || |
| 595 | ktrace_keep_ownership_on_reset) && |
| 596 | kauth_cred_issuser(cred: kauth_cred_get()))) { |
| 597 | if ((ret = ktrace_configure(KTRACE_KPERF))) { |
| 598 | ktrace_unlock(); |
| 599 | return ret; |
| 600 | } |
| 601 | } |
| 602 | } else { |
| 603 | if ((ret = ktrace_read_check())) { |
| 604 | ktrace_unlock(); |
| 605 | return ret; |
| 606 | } |
| 607 | } |
| 608 | |
| 609 | /* which request */ |
| 610 | if ((uintptr_t)arg1 == REQ_BLESS) { |
| 611 | ret = sysctl_bless(req); |
| 612 | } else { |
| 613 | ret = ENOENT; |
| 614 | } |
| 615 | |
| 616 | ktrace_unlock(); |
| 617 | |
| 618 | return ret; |
| 619 | } |
| 620 | |
| 621 | /* root kperf node */ |
| 622 | |
| 623 | SYSCTL_NODE(, OID_AUTO, kperf, CTLFLAG_RW | CTLFLAG_LOCKED, 0, |
| 624 | "kperf" ); |
| 625 | |
| 626 | /* actions */ |
| 627 | |
| 628 | SYSCTL_NODE(_kperf, OID_AUTO, action, CTLFLAG_RW | CTLFLAG_LOCKED, 0, |
| 629 | "action" ); |
| 630 | |
| 631 | SYSCTL_PROC(_kperf_action, OID_AUTO, count, |
| 632 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED | |
| 633 | CTLFLAG_MASKED, |
| 634 | (void *)REQ_ACTION_COUNT, |
| 635 | sizeof(int), kperf_sysctl, "I" , "Number of actions" ); |
| 636 | |
| 637 | SYSCTL_PROC(_kperf_action, OID_AUTO, samplers, |
| 638 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 639 | (void *)REQ_ACTION_SAMPLERS, |
| 640 | 3 * sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 641 | "What to sample when a trigger fires an action" ); |
| 642 | |
| 643 | SYSCTL_PROC(_kperf_action, OID_AUTO, userdata, |
| 644 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 645 | (void *)REQ_ACTION_USERDATA, |
| 646 | 3 * sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 647 | "User data to attribute to action" ); |
| 648 | |
| 649 | SYSCTL_PROC(_kperf_action, OID_AUTO, filter_by_task, |
| 650 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 651 | (void *)REQ_ACTION_FILTER_BY_TASK, |
| 652 | 3 * sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 653 | "Apply a task filter to the action" ); |
| 654 | |
| 655 | SYSCTL_PROC(_kperf_action, OID_AUTO, filter_by_pid, |
| 656 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 657 | (void *)REQ_ACTION_FILTER_BY_PID, |
| 658 | 3 * sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 659 | "Apply a pid filter to the action" ); |
| 660 | |
| 661 | SYSCTL_PROC(_kperf_action, OID_AUTO, ucallstack_depth, |
| 662 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 663 | (void *)REQ_ACTION_UCALLSTACK_DEPTH, |
| 664 | sizeof(int), kperf_sysctl, "I" , |
| 665 | "Maximum number of frames to include in user callstacks" ); |
| 666 | |
| 667 | SYSCTL_PROC(_kperf_action, OID_AUTO, kcallstack_depth, |
| 668 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 669 | (void *)REQ_ACTION_KCALLSTACK_DEPTH, |
| 670 | sizeof(int), kperf_sysctl, "I" , |
| 671 | "Maximum number of frames to include in kernel callstacks" ); |
| 672 | |
| 673 | /* timers */ |
| 674 | |
| 675 | SYSCTL_NODE(_kperf, OID_AUTO, timer, CTLFLAG_RW | CTLFLAG_LOCKED, 0, |
| 676 | "timer" ); |
| 677 | |
| 678 | SYSCTL_PROC(_kperf_timer, OID_AUTO, count, |
| 679 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 680 | | CTLFLAG_MASKED, |
| 681 | (void *)REQ_TIMER_COUNT, |
| 682 | sizeof(int), kperf_sysctl, "I" , "Number of time triggers" ); |
| 683 | |
| 684 | SYSCTL_PROC(_kperf_timer, OID_AUTO, period, |
| 685 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 686 | (void *)REQ_TIMER_PERIOD, |
| 687 | 2 * sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 688 | "Timer number and period" ); |
| 689 | |
| 690 | SYSCTL_PROC(_kperf_timer, OID_AUTO, action, |
| 691 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 692 | (void *)REQ_TIMER_ACTION, |
| 693 | 2 * sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 694 | "Timer number and actionid" ); |
| 695 | |
| 696 | SYSCTL_PROC(_kperf_timer, OID_AUTO, pet_timer, |
| 697 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 698 | | CTLFLAG_MASKED, |
| 699 | (void *)REQ_TIMER_PET, |
| 700 | sizeof(int), kperf_sysctl, "I" , "Which timer ID does PET" ); |
| 701 | |
| 702 | /* kdebug trigger */ |
| 703 | |
| 704 | SYSCTL_NODE(_kperf, OID_AUTO, kdebug, CTLFLAG_RW | CTLFLAG_LOCKED, 0, |
| 705 | "kdebug" ); |
| 706 | |
| 707 | SYSCTL_PROC(_kperf_kdebug, OID_AUTO, action, |
| 708 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 709 | | CTLFLAG_MASKED, |
| 710 | (void*)REQ_KDEBUG_ACTION, |
| 711 | sizeof(int), kperf_sysctl, "I" , "ID of action to trigger on kdebug events" ); |
| 712 | |
| 713 | SYSCTL_PROC(_kperf_kdebug, OID_AUTO, filter, |
| 714 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 715 | (void*)REQ_KDEBUG_FILTER, |
| 716 | sizeof(int), kperf_sysctl, "P" , "The filter that determines which kdebug events trigger a sample" ); |
| 717 | |
| 718 | /* lazy sampling */ |
| 719 | |
| 720 | SYSCTL_NODE(_kperf, OID_AUTO, lazy, CTLFLAG_RW | CTLFLAG_LOCKED, 0, |
| 721 | "lazy" ); |
| 722 | |
| 723 | SYSCTL_PROC(_kperf_lazy, OID_AUTO, wait_time_threshold, |
| 724 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 725 | (void *)REQ_LAZY_WAIT_TIME_THRESHOLD, |
| 726 | sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 727 | "How many ticks a thread must wait to take a sample" ); |
| 728 | |
| 729 | SYSCTL_PROC(_kperf_lazy, OID_AUTO, wait_action, |
| 730 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 731 | (void *)REQ_LAZY_WAIT_ACTION, |
| 732 | sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 733 | "Which action to fire when a thread waits longer than threshold" ); |
| 734 | |
| 735 | SYSCTL_PROC(_kperf_lazy, OID_AUTO, cpu_time_threshold, |
| 736 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 737 | (void *)REQ_LAZY_CPU_TIME_THRESHOLD, |
| 738 | sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 739 | "Minimum number of ticks a CPU must run between samples" ); |
| 740 | |
| 741 | SYSCTL_PROC(_kperf_lazy, OID_AUTO, cpu_action, |
| 742 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 743 | (void *)REQ_LAZY_CPU_ACTION, |
| 744 | sizeof(uint64_t), kperf_sysctl, "UQ" , |
| 745 | "Which action to fire for lazy CPU samples" ); |
| 746 | |
| 747 | /* misc */ |
| 748 | |
| 749 | SYSCTL_PROC(_kperf, OID_AUTO, sampling, |
| 750 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 751 | | CTLFLAG_MASKED, |
| 752 | (void *)REQ_SAMPLING, |
| 753 | sizeof(int), kperf_sysctl, "I" , "Sampling running" ); |
| 754 | |
| 755 | SYSCTL_PROC(_kperf, OID_AUTO, reset, |
| 756 | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MASKED | CTLFLAG_LOCKED, |
| 757 | (void *)REQ_RESET, |
| 758 | 0, kperf_sysctl, "-" , "Reset kperf" ); |
| 759 | |
| 760 | SYSCTL_PROC(_kperf, OID_AUTO, blessed_pid, |
| 761 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED /* must be root */ |
| 762 | | CTLFLAG_MASKED, |
| 763 | (void *)REQ_BLESS, |
| 764 | sizeof(int), kperf_sysctl_bless_handler, "I" , "Blessed pid" ); |
| 765 | |
| 766 | SYSCTL_PROC(_kperf, OID_AUTO, blessed_preempt, |
| 767 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED | |
| 768 | CTLFLAG_MASKED, |
| 769 | (void *)REQ_BLESS_PREEMPT, |
| 770 | sizeof(int), kperf_sysctl, "I" , "Blessed preemption" ); |
| 771 | |
| 772 | SYSCTL_PROC(_kperf, OID_AUTO, kdbg_cswitch, |
| 773 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 774 | | CTLFLAG_MASKED, |
| 775 | (void *)REQ_KDBG_CSWITCH, |
| 776 | sizeof(int), kperf_sysctl, "I" , "Generate context switch info" ); |
| 777 | |
| 778 | SYSCTL_PROC(_kperf, OID_AUTO, pet_idle_rate, |
| 779 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 780 | | CTLFLAG_MASKED, |
| 781 | (void *)REQ_PET_IDLE_RATE, |
| 782 | sizeof(int), kperf_sysctl, "I" , |
| 783 | "Rate at which unscheduled threads are forced to be sampled in " |
| 784 | "PET mode" ); |
| 785 | |
| 786 | SYSCTL_PROC(_kperf, OID_AUTO, lightweight_pet, |
| 787 | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED |
| 788 | | CTLFLAG_MASKED, |
| 789 | (void *)REQ_LIGHTWEIGHT_PET, |
| 790 | sizeof(int), kperf_sysctl, "I" , |
| 791 | "Status of lightweight PET mode" ); |
| 792 | |
| 793 | /* limits */ |
| 794 | |
| 795 | SYSCTL_NODE(_kperf, OID_AUTO, limits, CTLFLAG_RW | CTLFLAG_LOCKED, 0, |
| 796 | "limits" ); |
| 797 | |
| 798 | enum kperf_limit_request { |
| 799 | REQ_LIM_PERIOD_NS, |
| 800 | REQ_LIM_BG_PERIOD_NS, |
| 801 | REQ_LIM_PET_PERIOD_NS, |
| 802 | REQ_LIM_BG_PET_PERIOD_NS, |
| 803 | }; |
| 804 | |
| 805 | static int |
| 806 | kperf_sysctl_limits SYSCTL_HANDLER_ARGS |
| 807 | { |
| 808 | #pragma unused(oidp, arg2) |
| 809 | enum kptimer_period_limit limit = (enum kptimer_period_limit)arg1; |
| 810 | if (limit >= KTPL_MAX) { |
| 811 | return ENOENT; |
| 812 | } |
| 813 | uint64_t period = kptimer_minperiods_ns[limit]; |
| 814 | return sysctl_io_number(req, bigValue: (long long)period, valueSize: sizeof(period), pValue: &period, |
| 815 | NULL); |
| 816 | } |
| 817 | |
| 818 | SYSCTL_PROC(_kperf_limits, OID_AUTO, timer_min_period_ns, |
| 819 | CTLTYPE_QUAD | CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_LOCKED, |
| 820 | (void *)REQ_LIM_PERIOD_NS, sizeof(uint64_t), kperf_sysctl_limits, |
| 821 | "Q" , "Minimum timer period in nanoseconds" ); |
| 822 | SYSCTL_PROC(_kperf_limits, OID_AUTO, timer_min_bg_period_ns, |
| 823 | CTLTYPE_QUAD | CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_LOCKED, |
| 824 | (void *)REQ_LIM_BG_PERIOD_NS, sizeof(uint64_t), kperf_sysctl_limits, |
| 825 | "Q" , "Minimum background timer period in nanoseconds" ); |
| 826 | SYSCTL_PROC(_kperf_limits, OID_AUTO, timer_min_pet_period_ns, |
| 827 | CTLTYPE_QUAD | CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_LOCKED, |
| 828 | (void *)REQ_LIM_PET_PERIOD_NS, sizeof(uint64_t), kperf_sysctl_limits, |
| 829 | "Q" , "Minimum PET timer period in nanoseconds" ); |
| 830 | SYSCTL_PROC(_kperf_limits, OID_AUTO, timer_min_bg_pet_period_ns, |
| 831 | CTLTYPE_QUAD | CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_LOCKED, |
| 832 | (void *)REQ_LIM_BG_PET_PERIOD_NS, sizeof(uint64_t), kperf_sysctl_limits, |
| 833 | "Q" , "Minimum background PET timer period in nanoseconds" ); |
| 834 | SYSCTL_INT(_kperf_limits, OID_AUTO, max_action_count, CTLFLAG_RD | CTLFLAG_LOCKED, |
| 835 | &kperf_max_actions, 0, "Maximum number of kperf actions" ); |
| 836 | |
| 837 | /* debug */ |
| 838 | SYSCTL_INT(_kperf, OID_AUTO, debug_level, CTLFLAG_RW | CTLFLAG_LOCKED, |
| 839 | &kperf_debug_level, 0, "debug level" ); |
| 840 | |
| 841 | #if DEVELOPMENT || DEBUG |
| 842 | SYSCTL_QUAD(_kperf, OID_AUTO, already_pending_ipis, |
| 843 | CTLFLAG_RD | CTLFLAG_LOCKED, |
| 844 | &kperf_pending_ipis, "" ); |
| 845 | #endif /* DEVELOPMENT || DEBUG */ |
| 846 | |