| 1 | /* |
| 2 | * Copyright (c) 2000-2016 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 | #include <kern/policy_internal.h> |
| 30 | #include <mach/task_policy.h> |
| 31 | |
| 32 | #include <mach/mach_types.h> |
| 33 | #include <mach/task_server.h> |
| 34 | |
| 35 | #include <kern/host.h> /* host_priv_self() */ |
| 36 | #include <mach/host_priv.h> /* host_get_special_port() */ |
| 37 | #include <mach/host_special_ports.h> /* RESOURCE_NOTIFY_PORT */ |
| 38 | #include <kern/sched.h> |
| 39 | #include <kern/task.h> |
| 40 | #include <mach/thread_policy.h> |
| 41 | #include <sys/errno.h> |
| 42 | #include <sys/resource.h> |
| 43 | #include <machine/limits.h> |
| 44 | #include <kern/ledger.h> |
| 45 | #include <kern/thread_call.h> |
| 46 | #include <kern/sfi.h> |
| 47 | #include <kern/coalition.h> |
| 48 | #if CONFIG_TELEMETRY |
| 49 | #include <kern/telemetry.h> |
| 50 | #endif |
| 51 | #if CONFIG_EMBEDDED |
| 52 | #include <kern/kalloc.h> |
| 53 | #include <sys/errno.h> |
| 54 | #endif /* CONFIG_EMBEDDED */ |
| 55 | |
| 56 | #if IMPORTANCE_INHERITANCE |
| 57 | #include <ipc/ipc_importance.h> |
| 58 | #if IMPORTANCE_TRACE |
| 59 | #include <mach/machine/sdt.h> |
| 60 | #endif /* IMPORTANCE_TRACE */ |
| 61 | #endif /* IMPORTANCE_INHERITACE */ |
| 62 | |
| 63 | #include <sys/kdebug.h> |
| 64 | |
| 65 | /* |
| 66 | * Task Policy |
| 67 | * |
| 68 | * This subsystem manages task and thread IO priority and backgrounding, |
| 69 | * as well as importance inheritance, process suppression, task QoS, and apptype. |
| 70 | * These properties have a suprising number of complex interactions, so they are |
| 71 | * centralized here in one state machine to simplify the implementation of those interactions. |
| 72 | * |
| 73 | * Architecture: |
| 74 | * Threads and tasks have two policy fields: requested, effective. |
| 75 | * Requested represents the wishes of each interface that influences task policy. |
| 76 | * Effective represents the distillation of that policy into a set of behaviors. |
| 77 | * |
| 78 | * Each thread making a modification in the policy system passes a 'pending' struct, |
| 79 | * which tracks updates that will be applied after dropping the policy engine lock. |
| 80 | * |
| 81 | * Each interface that has an input into the task policy state machine controls a field in requested. |
| 82 | * If the interface has a getter, it returns what is in the field in requested, but that is |
| 83 | * not necessarily what is actually in effect. |
| 84 | * |
| 85 | * All kernel subsystems that behave differently based on task policy call into |
| 86 | * the proc_get_effective_(task|thread)_policy functions, which return the decision of the task policy state machine |
| 87 | * for that subsystem by querying only the 'effective' field. |
| 88 | * |
| 89 | * Policy change operations: |
| 90 | * Here are the steps to change a policy on a task or thread: |
| 91 | * 1) Lock task |
| 92 | * 2) Change requested field for the relevant policy |
| 93 | * 3) Run a task policy update, which recalculates effective based on requested, |
| 94 | * then takes a diff between the old and new versions of requested and calls the relevant |
| 95 | * other subsystems to apply these changes, and updates the pending field. |
| 96 | * 4) Unlock task |
| 97 | * 5) Run task policy update complete, which looks at the pending field to update |
| 98 | * subsystems which cannot be touched while holding the task lock. |
| 99 | * |
| 100 | * To add a new requested policy, add the field in the requested struct, the flavor in task.h, |
| 101 | * the setter and getter in proc_(set|get)_task_policy*, |
| 102 | * then set up the effects of that behavior in task_policy_update*. If the policy manifests |
| 103 | * itself as a distinct effective policy, add it to the effective struct and add it to the |
| 104 | * proc_get_effective_task_policy accessor. |
| 105 | * |
| 106 | * Most policies are set via proc_set_task_policy, but policies that don't fit that interface |
| 107 | * roll their own lock/set/update/unlock/complete code inside this file. |
| 108 | * |
| 109 | * |
| 110 | * Suppression policy |
| 111 | * |
| 112 | * These are a set of behaviors that can be requested for a task. They currently have specific |
| 113 | * implied actions when they're enabled, but they may be made customizable in the future. |
| 114 | * |
| 115 | * When the affected task is boosted, we temporarily disable the suppression behaviors |
| 116 | * so that the affected process has a chance to run so it can call the API to permanently |
| 117 | * disable the suppression behaviors. |
| 118 | * |
| 119 | * Locking |
| 120 | * |
| 121 | * Changing task policy on a task takes the task lock. |
| 122 | * Changing task policy on a thread takes the thread mutex. |
| 123 | * Task policy changes that affect threads will take each thread's mutex to update it if necessary. |
| 124 | * |
| 125 | * Querying the effective policy does not take a lock, because callers |
| 126 | * may run in interrupt context or other place where locks are not OK. |
| 127 | * |
| 128 | * This means that any notification of state change needs to be externally synchronized. |
| 129 | * We do this by idempotent callouts after the state has changed to ask |
| 130 | * other subsystems to update their view of the world. |
| 131 | * |
| 132 | * TODO: Move all cpu/wakes/io monitor code into a separate file |
| 133 | * TODO: Move all importance code over to importance subsystem |
| 134 | * TODO: Move all taskwatch code into a separate file |
| 135 | * TODO: Move all VM importance code into a separate file |
| 136 | */ |
| 137 | |
| 138 | /* Task policy related helper functions */ |
| 139 | static void proc_set_task_policy_locked(task_t task, int category, int flavor, int value, int value2); |
| 140 | |
| 141 | static void task_policy_update_locked(task_t task, task_pend_token_t pend_token); |
| 142 | static void task_policy_update_internal_locked(task_t task, boolean_t in_create, task_pend_token_t pend_token); |
| 143 | |
| 144 | /* For attributes that have two scalars as input/output */ |
| 145 | static void proc_set_task_policy2(task_t task, int category, int flavor, int value1, int value2); |
| 146 | static void proc_get_task_policy2(task_t task, int category, int flavor, int *value1, int *value2); |
| 147 | |
| 148 | static boolean_t task_policy_update_coalition_focal_tasks(task_t task, int prev_role, int next_role, task_pend_token_t pend_token); |
| 149 | |
| 150 | static uint64_t task_requested_bitfield(task_t task); |
| 151 | static uint64_t task_effective_bitfield(task_t task); |
| 152 | |
| 153 | /* Convenience functions for munging a policy bitfield into a tracepoint */ |
| 154 | static uintptr_t trequested_0(task_t task); |
| 155 | static uintptr_t trequested_1(task_t task); |
| 156 | static uintptr_t teffective_0(task_t task); |
| 157 | static uintptr_t teffective_1(task_t task); |
| 158 | |
| 159 | /* CPU limits helper functions */ |
| 160 | static int task_set_cpuusage(task_t task, uint8_t percentage, uint64_t interval, uint64_t deadline, int scope, int entitled); |
| 161 | static int task_get_cpuusage(task_t task, uint8_t *percentagep, uint64_t *intervalp, uint64_t *deadlinep, int *scope); |
| 162 | static int task_enable_cpumon_locked(task_t task); |
| 163 | static int task_disable_cpumon(task_t task); |
| 164 | static int task_clear_cpuusage_locked(task_t task, int cpumon_entitled); |
| 165 | static int task_apply_resource_actions(task_t task, int type); |
| 166 | static void task_action_cpuusage(thread_call_param_t param0, thread_call_param_t param1); |
| 167 | |
| 168 | #ifdef MACH_BSD |
| 169 | typedef struct proc * proc_t; |
| 170 | int proc_pid(void *proc); |
| 171 | extern int proc_selfpid(void); |
| 172 | extern char * proc_name_address(void *p); |
| 173 | extern char * proc_best_name(proc_t proc); |
| 174 | |
| 175 | extern int proc_pidpathinfo_internal(proc_t p, uint64_t arg, |
| 176 | char *buffer, uint32_t buffersize, |
| 177 | int32_t *retval); |
| 178 | #endif /* MACH_BSD */ |
| 179 | |
| 180 | |
| 181 | #if CONFIG_EMBEDDED |
| 182 | /* TODO: make CONFIG_TASKWATCH */ |
| 183 | /* Taskwatch related helper functions */ |
| 184 | static void set_thread_appbg(thread_t thread, int setbg,int importance); |
| 185 | static void add_taskwatch_locked(task_t task, task_watch_t * twp); |
| 186 | static void remove_taskwatch_locked(task_t task, task_watch_t * twp); |
| 187 | static void task_watch_lock(void); |
| 188 | static void task_watch_unlock(void); |
| 189 | static void apply_appstate_watchers(task_t task); |
| 190 | |
| 191 | typedef struct task_watcher { |
| 192 | queue_chain_t tw_links; /* queueing of threads */ |
| 193 | task_t tw_task; /* task that is being watched */ |
| 194 | thread_t tw_thread; /* thread that is watching the watch_task */ |
| 195 | int tw_state; /* the current app state of the thread */ |
| 196 | int tw_importance; /* importance prior to backgrounding */ |
| 197 | } task_watch_t; |
| 198 | |
| 199 | typedef struct thread_watchlist { |
| 200 | thread_t thread; /* thread being worked on for taskwatch action */ |
| 201 | int importance; /* importance to be restored if thread is being made active */ |
| 202 | } thread_watchlist_t; |
| 203 | |
| 204 | #endif /* CONFIG_EMBEDDED */ |
| 205 | |
| 206 | extern int memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap); |
| 207 | |
| 208 | /* Importance Inheritance related helper functions */ |
| 209 | |
| 210 | #if IMPORTANCE_INHERITANCE |
| 211 | |
| 212 | static void task_importance_mark_live_donor(task_t task, boolean_t donating); |
| 213 | static void task_importance_mark_receiver(task_t task, boolean_t receiving); |
| 214 | static void task_importance_mark_denap_receiver(task_t task, boolean_t denap); |
| 215 | |
| 216 | static boolean_t task_is_marked_live_importance_donor(task_t task); |
| 217 | static boolean_t task_is_importance_receiver(task_t task); |
| 218 | static boolean_t task_is_importance_denap_receiver(task_t task); |
| 219 | |
| 220 | static int task_importance_hold_internal_assertion(task_t target_task, uint32_t count); |
| 221 | |
| 222 | static void task_add_importance_watchport(task_t task, mach_port_t port, int *boostp); |
| 223 | static void task_importance_update_live_donor(task_t target_task); |
| 224 | |
| 225 | static void task_set_boost_locked(task_t task, boolean_t boost_active); |
| 226 | |
| 227 | #endif /* IMPORTANCE_INHERITANCE */ |
| 228 | |
| 229 | #if IMPORTANCE_TRACE |
| 230 | #define __imptrace_only |
| 231 | #else /* IMPORTANCE_TRACE */ |
| 232 | #define __imptrace_only __unused |
| 233 | #endif /* !IMPORTANCE_TRACE */ |
| 234 | |
| 235 | #if IMPORTANCE_INHERITANCE |
| 236 | #define __imp_only |
| 237 | #else |
| 238 | #define __imp_only __unused |
| 239 | #endif |
| 240 | |
| 241 | /* |
| 242 | * Default parameters for certain policies |
| 243 | */ |
| 244 | |
| 245 | int proc_standard_daemon_tier = THROTTLE_LEVEL_TIER1; |
| 246 | int proc_suppressed_disk_tier = THROTTLE_LEVEL_TIER1; |
| 247 | int proc_tal_disk_tier = THROTTLE_LEVEL_TIER1; |
| 248 | |
| 249 | int proc_graphics_timer_qos = (LATENCY_QOS_TIER_0 & 0xFF); |
| 250 | |
| 251 | #if CONFIG_EMBEDDED |
| 252 | const int proc_default_bg_iotier = THROTTLE_LEVEL_TIER3; |
| 253 | #else |
| 254 | const int proc_default_bg_iotier = THROTTLE_LEVEL_TIER2; |
| 255 | #endif |
| 256 | |
| 257 | /* Latency/throughput QoS fields remain zeroed, i.e. TIER_UNSPECIFIED at creation */ |
| 258 | const struct task_requested_policy default_task_requested_policy = { |
| 259 | .trp_bg_iotier = proc_default_bg_iotier |
| 260 | }; |
| 261 | const struct task_effective_policy default_task_effective_policy = {}; |
| 262 | |
| 263 | /* |
| 264 | * Default parameters for CPU usage monitor. |
| 265 | * |
| 266 | * Default setting is 50% over 3 minutes. |
| 267 | */ |
| 268 | #define DEFAULT_CPUMON_PERCENTAGE 50 |
| 269 | #define DEFAULT_CPUMON_INTERVAL (3 * 60) |
| 270 | |
| 271 | uint8_t proc_max_cpumon_percentage; |
| 272 | uint64_t proc_max_cpumon_interval; |
| 273 | |
| 274 | |
| 275 | kern_return_t |
| 276 | qos_latency_policy_validate(task_latency_qos_t ltier) { |
| 277 | if ((ltier != LATENCY_QOS_TIER_UNSPECIFIED) && |
| 278 | ((ltier > LATENCY_QOS_TIER_5) || (ltier < LATENCY_QOS_TIER_0))) |
| 279 | return KERN_INVALID_ARGUMENT; |
| 280 | |
| 281 | return KERN_SUCCESS; |
| 282 | } |
| 283 | |
| 284 | kern_return_t |
| 285 | qos_throughput_policy_validate(task_throughput_qos_t ttier) { |
| 286 | if ((ttier != THROUGHPUT_QOS_TIER_UNSPECIFIED) && |
| 287 | ((ttier > THROUGHPUT_QOS_TIER_5) || (ttier < THROUGHPUT_QOS_TIER_0))) |
| 288 | return KERN_INVALID_ARGUMENT; |
| 289 | |
| 290 | return KERN_SUCCESS; |
| 291 | } |
| 292 | |
| 293 | static kern_return_t |
| 294 | task_qos_policy_validate(task_qos_policy_t qosinfo, mach_msg_type_number_t count) { |
| 295 | if (count < TASK_QOS_POLICY_COUNT) |
| 296 | return KERN_INVALID_ARGUMENT; |
| 297 | |
| 298 | task_latency_qos_t ltier = qosinfo->task_latency_qos_tier; |
| 299 | task_throughput_qos_t ttier = qosinfo->task_throughput_qos_tier; |
| 300 | |
| 301 | kern_return_t kr = qos_latency_policy_validate(ltier); |
| 302 | |
| 303 | if (kr != KERN_SUCCESS) |
| 304 | return kr; |
| 305 | |
| 306 | kr = qos_throughput_policy_validate(ttier); |
| 307 | |
| 308 | return kr; |
| 309 | } |
| 310 | |
| 311 | uint32_t |
| 312 | qos_extract(uint32_t qv) { |
| 313 | return (qv & 0xFF); |
| 314 | } |
| 315 | |
| 316 | uint32_t |
| 317 | qos_latency_policy_package(uint32_t qv) { |
| 318 | return (qv == LATENCY_QOS_TIER_UNSPECIFIED) ? LATENCY_QOS_TIER_UNSPECIFIED : ((0xFF << 16) | qv); |
| 319 | } |
| 320 | |
| 321 | uint32_t |
| 322 | qos_throughput_policy_package(uint32_t qv) { |
| 323 | return (qv == THROUGHPUT_QOS_TIER_UNSPECIFIED) ? THROUGHPUT_QOS_TIER_UNSPECIFIED : ((0xFE << 16) | qv); |
| 324 | } |
| 325 | |
| 326 | /* TEMPORARY boot-arg controlling task_policy suppression (App Nap) */ |
| 327 | static boolean_t task_policy_suppression_disable = FALSE; |
| 328 | |
| 329 | kern_return_t |
| 330 | task_policy_set( |
| 331 | task_t task, |
| 332 | task_policy_flavor_t flavor, |
| 333 | task_policy_t policy_info, |
| 334 | mach_msg_type_number_t count) |
| 335 | { |
| 336 | kern_return_t result = KERN_SUCCESS; |
| 337 | |
| 338 | if (task == TASK_NULL || task == kernel_task) |
| 339 | return (KERN_INVALID_ARGUMENT); |
| 340 | |
| 341 | switch (flavor) { |
| 342 | |
| 343 | case TASK_CATEGORY_POLICY: { |
| 344 | task_category_policy_t info = (task_category_policy_t)policy_info; |
| 345 | |
| 346 | if (count < TASK_CATEGORY_POLICY_COUNT) |
| 347 | return (KERN_INVALID_ARGUMENT); |
| 348 | |
| 349 | #if CONFIG_EMBEDDED |
| 350 | /* On embedded, you can't modify your own role. */ |
| 351 | if (current_task() == task) |
| 352 | return (KERN_INVALID_ARGUMENT); |
| 353 | #endif |
| 354 | |
| 355 | switch(info->role) { |
| 356 | case TASK_FOREGROUND_APPLICATION: |
| 357 | case TASK_BACKGROUND_APPLICATION: |
| 358 | case TASK_DEFAULT_APPLICATION: |
| 359 | proc_set_task_policy(task, |
| 360 | TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE, |
| 361 | info->role); |
| 362 | break; |
| 363 | |
| 364 | case TASK_CONTROL_APPLICATION: |
| 365 | if (task != current_task() || task->sec_token.val[0] != 0) |
| 366 | result = KERN_INVALID_ARGUMENT; |
| 367 | else |
| 368 | proc_set_task_policy(task, |
| 369 | TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE, |
| 370 | info->role); |
| 371 | break; |
| 372 | |
| 373 | case TASK_GRAPHICS_SERVER: |
| 374 | /* TODO: Restrict this role to FCFS <rdar://problem/12552788> */ |
| 375 | if (task != current_task() || task->sec_token.val[0] != 0) |
| 376 | result = KERN_INVALID_ARGUMENT; |
| 377 | else |
| 378 | proc_set_task_policy(task, |
| 379 | TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE, |
| 380 | info->role); |
| 381 | break; |
| 382 | default: |
| 383 | result = KERN_INVALID_ARGUMENT; |
| 384 | break; |
| 385 | } /* switch (info->role) */ |
| 386 | |
| 387 | break; |
| 388 | } |
| 389 | |
| 390 | /* Desired energy-efficiency/performance "quality-of-service" */ |
| 391 | case TASK_BASE_QOS_POLICY: |
| 392 | case TASK_OVERRIDE_QOS_POLICY: |
| 393 | { |
| 394 | task_qos_policy_t qosinfo = (task_qos_policy_t)policy_info; |
| 395 | kern_return_t kr = task_qos_policy_validate(qosinfo, count); |
| 396 | |
| 397 | if (kr != KERN_SUCCESS) |
| 398 | return kr; |
| 399 | |
| 400 | |
| 401 | uint32_t lqos = qos_extract(qosinfo->task_latency_qos_tier); |
| 402 | uint32_t tqos = qos_extract(qosinfo->task_throughput_qos_tier); |
| 403 | |
| 404 | proc_set_task_policy2(task, TASK_POLICY_ATTRIBUTE, |
| 405 | flavor == TASK_BASE_QOS_POLICY ? TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS : TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS, |
| 406 | lqos, tqos); |
| 407 | } |
| 408 | break; |
| 409 | |
| 410 | case TASK_BASE_LATENCY_QOS_POLICY: |
| 411 | { |
| 412 | task_qos_policy_t qosinfo = (task_qos_policy_t)policy_info; |
| 413 | kern_return_t kr = task_qos_policy_validate(qosinfo, count); |
| 414 | |
| 415 | if (kr != KERN_SUCCESS) |
| 416 | return kr; |
| 417 | |
| 418 | uint32_t lqos = qos_extract(qosinfo->task_latency_qos_tier); |
| 419 | |
| 420 | proc_set_task_policy(task, TASK_POLICY_ATTRIBUTE, TASK_BASE_LATENCY_QOS_POLICY, lqos); |
| 421 | } |
| 422 | break; |
| 423 | |
| 424 | case TASK_BASE_THROUGHPUT_QOS_POLICY: |
| 425 | { |
| 426 | task_qos_policy_t qosinfo = (task_qos_policy_t)policy_info; |
| 427 | kern_return_t kr = task_qos_policy_validate(qosinfo, count); |
| 428 | |
| 429 | if (kr != KERN_SUCCESS) |
| 430 | return kr; |
| 431 | |
| 432 | uint32_t tqos = qos_extract(qosinfo->task_throughput_qos_tier); |
| 433 | |
| 434 | proc_set_task_policy(task, TASK_POLICY_ATTRIBUTE, TASK_BASE_THROUGHPUT_QOS_POLICY, tqos); |
| 435 | } |
| 436 | break; |
| 437 | |
| 438 | case TASK_SUPPRESSION_POLICY: |
| 439 | { |
| 440 | #if CONFIG_EMBEDDED |
| 441 | /* |
| 442 | * Suppression policy is not enabled for embedded |
| 443 | * because apps aren't marked as denap receivers |
| 444 | */ |
| 445 | result = KERN_INVALID_ARGUMENT; |
| 446 | break; |
| 447 | #else /* CONFIG_EMBEDDED */ |
| 448 | |
| 449 | task_suppression_policy_t info = (task_suppression_policy_t)policy_info; |
| 450 | |
| 451 | if (count < TASK_SUPPRESSION_POLICY_COUNT) |
| 452 | return (KERN_INVALID_ARGUMENT); |
| 453 | |
| 454 | struct task_qos_policy qosinfo; |
| 455 | |
| 456 | qosinfo.task_latency_qos_tier = info->timer_throttle; |
| 457 | qosinfo.task_throughput_qos_tier = info->throughput_qos; |
| 458 | |
| 459 | kern_return_t kr = task_qos_policy_validate(&qosinfo, TASK_QOS_POLICY_COUNT); |
| 460 | |
| 461 | if (kr != KERN_SUCCESS) |
| 462 | return kr; |
| 463 | |
| 464 | /* TEMPORARY disablement of task suppression */ |
| 465 | if (task_policy_suppression_disable && info->active) |
| 466 | return KERN_SUCCESS; |
| 467 | |
| 468 | struct task_pend_token pend_token = {}; |
| 469 | |
| 470 | task_lock(task); |
| 471 | |
| 472 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 473 | (IMPORTANCE_CODE(IMP_TASK_SUPPRESSION, info->active)) | DBG_FUNC_START, |
| 474 | proc_selfpid(), task_pid(task), trequested_0(task), |
| 475 | trequested_1(task), 0); |
| 476 | |
| 477 | task->requested_policy.trp_sup_active = (info->active) ? 1 : 0; |
| 478 | task->requested_policy.trp_sup_lowpri_cpu = (info->lowpri_cpu) ? 1 : 0; |
| 479 | task->requested_policy.trp_sup_timer = qos_extract(info->timer_throttle); |
| 480 | task->requested_policy.trp_sup_disk = (info->disk_throttle) ? 1 : 0; |
| 481 | task->requested_policy.trp_sup_throughput = qos_extract(info->throughput_qos); |
| 482 | task->requested_policy.trp_sup_cpu = (info->suppressed_cpu) ? 1 : 0; |
| 483 | task->requested_policy.trp_sup_bg_sockets = (info->background_sockets) ? 1 : 0; |
| 484 | |
| 485 | task_policy_update_locked(task, &pend_token); |
| 486 | |
| 487 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 488 | (IMPORTANCE_CODE(IMP_TASK_SUPPRESSION, info->active)) | DBG_FUNC_END, |
| 489 | proc_selfpid(), task_pid(task), trequested_0(task), |
| 490 | trequested_1(task), 0); |
| 491 | |
| 492 | task_unlock(task); |
| 493 | |
| 494 | task_policy_update_complete_unlocked(task, &pend_token); |
| 495 | |
| 496 | break; |
| 497 | |
| 498 | #endif /* CONFIG_EMBEDDED */ |
| 499 | } |
| 500 | |
| 501 | default: |
| 502 | result = KERN_INVALID_ARGUMENT; |
| 503 | break; |
| 504 | } |
| 505 | |
| 506 | return (result); |
| 507 | } |
| 508 | |
| 509 | /* Sets BSD 'nice' value on the task */ |
| 510 | kern_return_t |
| 511 | task_importance( |
| 512 | task_t task, |
| 513 | integer_t importance) |
| 514 | { |
| 515 | if (task == TASK_NULL || task == kernel_task) |
| 516 | return (KERN_INVALID_ARGUMENT); |
| 517 | |
| 518 | task_lock(task); |
| 519 | |
| 520 | if (!task->active) { |
| 521 | task_unlock(task); |
| 522 | |
| 523 | return (KERN_TERMINATED); |
| 524 | } |
| 525 | |
| 526 | if (proc_get_effective_task_policy(task, TASK_POLICY_ROLE) >= TASK_CONTROL_APPLICATION) { |
| 527 | task_unlock(task); |
| 528 | |
| 529 | return (KERN_INVALID_ARGUMENT); |
| 530 | } |
| 531 | |
| 532 | task->importance = importance; |
| 533 | |
| 534 | struct task_pend_token pend_token = {}; |
| 535 | |
| 536 | task_policy_update_locked(task, &pend_token); |
| 537 | |
| 538 | task_unlock(task); |
| 539 | |
| 540 | task_policy_update_complete_unlocked(task, &pend_token); |
| 541 | |
| 542 | return (KERN_SUCCESS); |
| 543 | } |
| 544 | |
| 545 | kern_return_t |
| 546 | task_policy_get( |
| 547 | task_t task, |
| 548 | task_policy_flavor_t flavor, |
| 549 | task_policy_t policy_info, |
| 550 | mach_msg_type_number_t *count, |
| 551 | boolean_t *get_default) |
| 552 | { |
| 553 | if (task == TASK_NULL || task == kernel_task) |
| 554 | return (KERN_INVALID_ARGUMENT); |
| 555 | |
| 556 | switch (flavor) { |
| 557 | |
| 558 | case TASK_CATEGORY_POLICY: |
| 559 | { |
| 560 | task_category_policy_t info = (task_category_policy_t)policy_info; |
| 561 | |
| 562 | if (*count < TASK_CATEGORY_POLICY_COUNT) |
| 563 | return (KERN_INVALID_ARGUMENT); |
| 564 | |
| 565 | if (*get_default) |
| 566 | info->role = TASK_UNSPECIFIED; |
| 567 | else |
| 568 | info->role = proc_get_task_policy(task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE); |
| 569 | break; |
| 570 | } |
| 571 | |
| 572 | case TASK_BASE_QOS_POLICY: /* FALLTHRU */ |
| 573 | case TASK_OVERRIDE_QOS_POLICY: |
| 574 | { |
| 575 | task_qos_policy_t info = (task_qos_policy_t)policy_info; |
| 576 | |
| 577 | if (*count < TASK_QOS_POLICY_COUNT) |
| 578 | return (KERN_INVALID_ARGUMENT); |
| 579 | |
| 580 | if (*get_default) { |
| 581 | info->task_latency_qos_tier = LATENCY_QOS_TIER_UNSPECIFIED; |
| 582 | info->task_throughput_qos_tier = THROUGHPUT_QOS_TIER_UNSPECIFIED; |
| 583 | } else if (flavor == TASK_BASE_QOS_POLICY) { |
| 584 | int value1, value2; |
| 585 | |
| 586 | proc_get_task_policy2(task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS, &value1, &value2); |
| 587 | |
| 588 | info->task_latency_qos_tier = qos_latency_policy_package(value1); |
| 589 | info->task_throughput_qos_tier = qos_throughput_policy_package(value2); |
| 590 | |
| 591 | } else if (flavor == TASK_OVERRIDE_QOS_POLICY) { |
| 592 | int value1, value2; |
| 593 | |
| 594 | proc_get_task_policy2(task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS, &value1, &value2); |
| 595 | |
| 596 | info->task_latency_qos_tier = qos_latency_policy_package(value1); |
| 597 | info->task_throughput_qos_tier = qos_throughput_policy_package(value2); |
| 598 | } |
| 599 | |
| 600 | break; |
| 601 | } |
| 602 | |
| 603 | case TASK_POLICY_STATE: |
| 604 | { |
| 605 | task_policy_state_t info = (task_policy_state_t)policy_info; |
| 606 | |
| 607 | if (*count < TASK_POLICY_STATE_COUNT) |
| 608 | return (KERN_INVALID_ARGUMENT); |
| 609 | |
| 610 | /* Only root can get this info */ |
| 611 | if (current_task()->sec_token.val[0] != 0) |
| 612 | return KERN_PROTECTION_FAILURE; |
| 613 | |
| 614 | if (*get_default) { |
| 615 | info->requested = 0; |
| 616 | info->effective = 0; |
| 617 | info->pending = 0; |
| 618 | info->imp_assertcnt = 0; |
| 619 | info->imp_externcnt = 0; |
| 620 | info->flags = 0; |
| 621 | info->imp_transitions = 0; |
| 622 | } else { |
| 623 | task_lock(task); |
| 624 | |
| 625 | info->requested = task_requested_bitfield(task); |
| 626 | info->effective = task_effective_bitfield(task); |
| 627 | info->pending = 0; |
| 628 | |
| 629 | info->tps_requested_policy = *(uint64_t*)(&task->requested_policy); |
| 630 | info->tps_effective_policy = *(uint64_t*)(&task->effective_policy); |
| 631 | |
| 632 | info->flags = 0; |
| 633 | if (task->task_imp_base != NULL) { |
| 634 | info->imp_assertcnt = task->task_imp_base->iit_assertcnt; |
| 635 | info->imp_externcnt = IIT_EXTERN(task->task_imp_base); |
| 636 | info->flags |= (task_is_marked_importance_receiver(task) ? TASK_IMP_RECEIVER : 0); |
| 637 | info->flags |= (task_is_marked_importance_denap_receiver(task) ? TASK_DENAP_RECEIVER : 0); |
| 638 | info->flags |= (task_is_marked_importance_donor(task) ? TASK_IMP_DONOR : 0); |
| 639 | info->flags |= (task_is_marked_live_importance_donor(task) ? TASK_IMP_LIVE_DONOR : 0); |
| 640 | info->imp_transitions = task->task_imp_base->iit_transitions; |
| 641 | } else { |
| 642 | info->imp_assertcnt = 0; |
| 643 | info->imp_externcnt = 0; |
| 644 | info->imp_transitions = 0; |
| 645 | } |
| 646 | task_unlock(task); |
| 647 | } |
| 648 | |
| 649 | break; |
| 650 | } |
| 651 | |
| 652 | case TASK_SUPPRESSION_POLICY: |
| 653 | { |
| 654 | task_suppression_policy_t info = (task_suppression_policy_t)policy_info; |
| 655 | |
| 656 | if (*count < TASK_SUPPRESSION_POLICY_COUNT) |
| 657 | return (KERN_INVALID_ARGUMENT); |
| 658 | |
| 659 | task_lock(task); |
| 660 | |
| 661 | if (*get_default) { |
| 662 | info->active = 0; |
| 663 | info->lowpri_cpu = 0; |
| 664 | info->timer_throttle = LATENCY_QOS_TIER_UNSPECIFIED; |
| 665 | info->disk_throttle = 0; |
| 666 | info->cpu_limit = 0; |
| 667 | info->suspend = 0; |
| 668 | info->throughput_qos = 0; |
| 669 | info->suppressed_cpu = 0; |
| 670 | } else { |
| 671 | info->active = task->requested_policy.trp_sup_active; |
| 672 | info->lowpri_cpu = task->requested_policy.trp_sup_lowpri_cpu; |
| 673 | info->timer_throttle = qos_latency_policy_package(task->requested_policy.trp_sup_timer); |
| 674 | info->disk_throttle = task->requested_policy.trp_sup_disk; |
| 675 | info->cpu_limit = 0; |
| 676 | info->suspend = 0; |
| 677 | info->throughput_qos = qos_throughput_policy_package(task->requested_policy.trp_sup_throughput); |
| 678 | info->suppressed_cpu = task->requested_policy.trp_sup_cpu; |
| 679 | info->background_sockets = task->requested_policy.trp_sup_bg_sockets; |
| 680 | } |
| 681 | |
| 682 | task_unlock(task); |
| 683 | break; |
| 684 | } |
| 685 | |
| 686 | default: |
| 687 | return (KERN_INVALID_ARGUMENT); |
| 688 | } |
| 689 | |
| 690 | return (KERN_SUCCESS); |
| 691 | } |
| 692 | |
| 693 | /* |
| 694 | * Called at task creation |
| 695 | * We calculate the correct effective but don't apply it to anything yet. |
| 696 | * The threads, etc will inherit from the task as they get created. |
| 697 | */ |
| 698 | void |
| 699 | task_policy_create(task_t task, task_t parent_task) |
| 700 | { |
| 701 | task->requested_policy.trp_apptype = parent_task->requested_policy.trp_apptype; |
| 702 | |
| 703 | task->requested_policy.trp_int_darwinbg = parent_task->requested_policy.trp_int_darwinbg; |
| 704 | task->requested_policy.trp_ext_darwinbg = parent_task->requested_policy.trp_ext_darwinbg; |
| 705 | task->requested_policy.trp_int_iotier = parent_task->requested_policy.trp_int_iotier; |
| 706 | task->requested_policy.trp_ext_iotier = parent_task->requested_policy.trp_ext_iotier; |
| 707 | task->requested_policy.trp_int_iopassive = parent_task->requested_policy.trp_int_iopassive; |
| 708 | task->requested_policy.trp_ext_iopassive = parent_task->requested_policy.trp_ext_iopassive; |
| 709 | task->requested_policy.trp_bg_iotier = parent_task->requested_policy.trp_bg_iotier; |
| 710 | task->requested_policy.trp_terminated = parent_task->requested_policy.trp_terminated; |
| 711 | task->requested_policy.trp_qos_clamp = parent_task->requested_policy.trp_qos_clamp; |
| 712 | |
| 713 | if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE && !task_is_exec_copy(task)) { |
| 714 | /* Do not update the apptype for exec copy task */ |
| 715 | if (parent_task->requested_policy.trp_boosted) { |
| 716 | task->requested_policy.trp_apptype = TASK_APPTYPE_DAEMON_INTERACTIVE; |
| 717 | task_importance_mark_donor(task, TRUE); |
| 718 | } else { |
| 719 | task->requested_policy.trp_apptype = TASK_APPTYPE_DAEMON_BACKGROUND; |
| 720 | task_importance_mark_receiver(task, FALSE); |
| 721 | } |
| 722 | } |
| 723 | |
| 724 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 725 | (IMPORTANCE_CODE(IMP_UPDATE, (IMP_UPDATE_TASK_CREATE | TASK_POLICY_TASK))) | DBG_FUNC_START, |
| 726 | task_pid(task), teffective_0(task), |
| 727 | teffective_1(task), task->priority, 0); |
| 728 | |
| 729 | task_policy_update_internal_locked(task, TRUE, NULL); |
| 730 | |
| 731 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 732 | (IMPORTANCE_CODE(IMP_UPDATE, (IMP_UPDATE_TASK_CREATE | TASK_POLICY_TASK))) | DBG_FUNC_END, |
| 733 | task_pid(task), teffective_0(task), |
| 734 | teffective_1(task), task->priority, 0); |
| 735 | |
| 736 | task_importance_update_live_donor(task); |
| 737 | } |
| 738 | |
| 739 | |
| 740 | static void |
| 741 | task_policy_update_locked(task_t task, task_pend_token_t pend_token) |
| 742 | { |
| 743 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 744 | (IMPORTANCE_CODE(IMP_UPDATE, TASK_POLICY_TASK) | DBG_FUNC_START), |
| 745 | task_pid(task), teffective_0(task), |
| 746 | teffective_1(task), task->priority, 0); |
| 747 | |
| 748 | task_policy_update_internal_locked(task, FALSE, pend_token); |
| 749 | |
| 750 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 751 | (IMPORTANCE_CODE(IMP_UPDATE, TASK_POLICY_TASK)) | DBG_FUNC_END, |
| 752 | task_pid(task), teffective_0(task), |
| 753 | teffective_1(task), task->priority, 0); |
| 754 | } |
| 755 | |
| 756 | /* |
| 757 | * One state update function TO RULE THEM ALL |
| 758 | * |
| 759 | * This function updates the task or thread effective policy fields |
| 760 | * and pushes the results to the relevant subsystems. |
| 761 | * |
| 762 | * Must call update_complete after unlocking the task, |
| 763 | * as some subsystems cannot be updated while holding the task lock. |
| 764 | * |
| 765 | * Called with task locked, not thread |
| 766 | */ |
| 767 | |
| 768 | static void |
| 769 | task_policy_update_internal_locked(task_t task, boolean_t in_create, task_pend_token_t pend_token) |
| 770 | { |
| 771 | /* |
| 772 | * Step 1: |
| 773 | * Gather requested policy |
| 774 | */ |
| 775 | |
| 776 | struct task_requested_policy requested = task->requested_policy; |
| 777 | |
| 778 | /* |
| 779 | * Step 2: |
| 780 | * Calculate new effective policies from requested policy and task state |
| 781 | * Rules: |
| 782 | * Don't change requested, it won't take effect |
| 783 | */ |
| 784 | |
| 785 | struct task_effective_policy next = {}; |
| 786 | |
| 787 | /* Update task role */ |
| 788 | next.tep_role = requested.trp_role; |
| 789 | |
| 790 | /* Set task qos clamp and ceiling */ |
| 791 | next.tep_qos_clamp = requested.trp_qos_clamp; |
| 792 | |
| 793 | if (requested.trp_apptype == TASK_APPTYPE_APP_DEFAULT || |
| 794 | requested.trp_apptype == TASK_APPTYPE_APP_TAL) { |
| 795 | |
| 796 | switch (next.tep_role) { |
| 797 | case TASK_FOREGROUND_APPLICATION: |
| 798 | /* Foreground apps get urgent scheduler priority */ |
| 799 | next.tep_qos_ui_is_urgent = 1; |
| 800 | next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED; |
| 801 | break; |
| 802 | |
| 803 | case TASK_BACKGROUND_APPLICATION: |
| 804 | /* This is really 'non-focal but on-screen' */ |
| 805 | next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED; |
| 806 | break; |
| 807 | |
| 808 | case TASK_DEFAULT_APPLICATION: |
| 809 | /* This is 'may render UI but we don't know if it's focal/nonfocal' */ |
| 810 | next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED; |
| 811 | break; |
| 812 | |
| 813 | case TASK_NONUI_APPLICATION: |
| 814 | /* i.e. 'off-screen' */ |
| 815 | next.tep_qos_ceiling = THREAD_QOS_LEGACY; |
| 816 | break; |
| 817 | |
| 818 | case TASK_CONTROL_APPLICATION: |
| 819 | case TASK_GRAPHICS_SERVER: |
| 820 | next.tep_qos_ui_is_urgent = 1; |
| 821 | next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED; |
| 822 | break; |
| 823 | |
| 824 | case TASK_THROTTLE_APPLICATION: |
| 825 | /* i.e. 'TAL launch' */ |
| 826 | next.tep_qos_ceiling = THREAD_QOS_UTILITY; |
| 827 | break; |
| 828 | |
| 829 | case TASK_UNSPECIFIED: |
| 830 | default: |
| 831 | /* Apps that don't have an application role get |
| 832 | * USER_INTERACTIVE and USER_INITIATED squashed to LEGACY */ |
| 833 | next.tep_qos_ceiling = THREAD_QOS_LEGACY; |
| 834 | break; |
| 835 | } |
| 836 | } else { |
| 837 | /* Daemons get USER_INTERACTIVE squashed to USER_INITIATED */ |
| 838 | next.tep_qos_ceiling = THREAD_QOS_USER_INITIATED; |
| 839 | } |
| 840 | |
| 841 | /* Calculate DARWIN_BG */ |
| 842 | boolean_t wants_darwinbg = FALSE; |
| 843 | boolean_t wants_all_sockets_bg = FALSE; /* Do I want my existing sockets to be bg */ |
| 844 | boolean_t wants_watchersbg = FALSE; /* Do I want my pidbound threads to be bg */ |
| 845 | |
| 846 | /* |
| 847 | * If DARWIN_BG has been requested at either level, it's engaged. |
| 848 | * Only true DARWIN_BG changes cause watchers to transition. |
| 849 | * |
| 850 | * Backgrounding due to apptype does. |
| 851 | */ |
| 852 | if (requested.trp_int_darwinbg || requested.trp_ext_darwinbg) |
| 853 | wants_watchersbg = wants_all_sockets_bg = wants_darwinbg = TRUE; |
| 854 | |
| 855 | /* Background TAL apps are throttled when TAL is enabled */ |
| 856 | if (requested.trp_apptype == TASK_APPTYPE_APP_TAL && |
| 857 | requested.trp_role == TASK_BACKGROUND_APPLICATION && |
| 858 | requested.trp_tal_enabled == 1) { |
| 859 | next.tep_tal_engaged = 1; |
| 860 | } |
| 861 | |
| 862 | if ((requested.trp_apptype == TASK_APPTYPE_APP_DEFAULT || |
| 863 | requested.trp_apptype == TASK_APPTYPE_APP_TAL) && |
| 864 | requested.trp_role == TASK_THROTTLE_APPLICATION) { |
| 865 | next.tep_tal_engaged = 1; |
| 866 | } |
| 867 | |
| 868 | /* Adaptive daemons are DARWIN_BG unless boosted, and don't get network throttled. */ |
| 869 | if (requested.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE && |
| 870 | requested.trp_boosted == 0) |
| 871 | wants_darwinbg = TRUE; |
| 872 | |
| 873 | /* Background daemons are always DARWIN_BG, no exceptions, and don't get network throttled. */ |
| 874 | if (requested.trp_apptype == TASK_APPTYPE_DAEMON_BACKGROUND) |
| 875 | wants_darwinbg = TRUE; |
| 876 | |
| 877 | if (next.tep_qos_clamp == THREAD_QOS_BACKGROUND || next.tep_qos_clamp == THREAD_QOS_MAINTENANCE) |
| 878 | wants_darwinbg = TRUE; |
| 879 | |
| 880 | /* Calculate side effects of DARWIN_BG */ |
| 881 | |
| 882 | if (wants_darwinbg) { |
| 883 | next.tep_darwinbg = 1; |
| 884 | /* darwinbg tasks always create bg sockets, but we don't always loop over all sockets */ |
| 885 | next.tep_new_sockets_bg = 1; |
| 886 | next.tep_lowpri_cpu = 1; |
| 887 | } |
| 888 | |
| 889 | if (wants_all_sockets_bg) |
| 890 | next.tep_all_sockets_bg = 1; |
| 891 | |
| 892 | if (wants_watchersbg) |
| 893 | next.tep_watchers_bg = 1; |
| 894 | |
| 895 | /* Calculate low CPU priority */ |
| 896 | |
| 897 | boolean_t wants_lowpri_cpu = FALSE; |
| 898 | |
| 899 | if (wants_darwinbg) |
| 900 | wants_lowpri_cpu = TRUE; |
| 901 | |
| 902 | if (next.tep_tal_engaged) |
| 903 | wants_lowpri_cpu = TRUE; |
| 904 | |
| 905 | if (requested.trp_sup_lowpri_cpu && requested.trp_boosted == 0) |
| 906 | wants_lowpri_cpu = TRUE; |
| 907 | |
| 908 | if (wants_lowpri_cpu) |
| 909 | next.tep_lowpri_cpu = 1; |
| 910 | |
| 911 | /* Calculate IO policy */ |
| 912 | |
| 913 | /* Update BG IO policy (so we can see if it has changed) */ |
| 914 | next.tep_bg_iotier = requested.trp_bg_iotier; |
| 915 | |
| 916 | int iopol = THROTTLE_LEVEL_TIER0; |
| 917 | |
| 918 | if (wants_darwinbg) |
| 919 | iopol = MAX(iopol, requested.trp_bg_iotier); |
| 920 | |
| 921 | if (requested.trp_apptype == TASK_APPTYPE_DAEMON_STANDARD) |
| 922 | iopol = MAX(iopol, proc_standard_daemon_tier); |
| 923 | |
| 924 | if (requested.trp_sup_disk && requested.trp_boosted == 0) |
| 925 | iopol = MAX(iopol, proc_suppressed_disk_tier); |
| 926 | |
| 927 | if (next.tep_tal_engaged) |
| 928 | iopol = MAX(iopol, proc_tal_disk_tier); |
| 929 | |
| 930 | if (next.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) |
| 931 | iopol = MAX(iopol, thread_qos_policy_params.qos_iotier[next.tep_qos_clamp]); |
| 932 | |
| 933 | iopol = MAX(iopol, requested.trp_int_iotier); |
| 934 | iopol = MAX(iopol, requested.trp_ext_iotier); |
| 935 | |
| 936 | next.tep_io_tier = iopol; |
| 937 | |
| 938 | /* Calculate Passive IO policy */ |
| 939 | |
| 940 | if (requested.trp_ext_iopassive || requested.trp_int_iopassive) |
| 941 | next.tep_io_passive = 1; |
| 942 | |
| 943 | /* Calculate suppression-active flag */ |
| 944 | boolean_t memorystatus_appnap_transition = FALSE; |
| 945 | |
| 946 | if (requested.trp_sup_active && requested.trp_boosted == 0) |
| 947 | next.tep_sup_active = 1; |
| 948 | |
| 949 | if (task->effective_policy.tep_sup_active != next.tep_sup_active) |
| 950 | memorystatus_appnap_transition = TRUE; |
| 951 | |
| 952 | /* Calculate timer QOS */ |
| 953 | int latency_qos = requested.trp_base_latency_qos; |
| 954 | |
| 955 | if (requested.trp_sup_timer && requested.trp_boosted == 0) |
| 956 | latency_qos = requested.trp_sup_timer; |
| 957 | |
| 958 | if (next.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) |
| 959 | latency_qos = MAX(latency_qos, (int)thread_qos_policy_params.qos_latency_qos[next.tep_qos_clamp]); |
| 960 | |
| 961 | if (requested.trp_over_latency_qos != 0) |
| 962 | latency_qos = requested.trp_over_latency_qos; |
| 963 | |
| 964 | /* Treat the windowserver special */ |
| 965 | if (requested.trp_role == TASK_GRAPHICS_SERVER) |
| 966 | latency_qos = proc_graphics_timer_qos; |
| 967 | |
| 968 | next.tep_latency_qos = latency_qos; |
| 969 | |
| 970 | /* Calculate throughput QOS */ |
| 971 | int through_qos = requested.trp_base_through_qos; |
| 972 | |
| 973 | if (requested.trp_sup_throughput && requested.trp_boosted == 0) |
| 974 | through_qos = requested.trp_sup_throughput; |
| 975 | |
| 976 | if (next.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) |
| 977 | through_qos = MAX(through_qos, (int)thread_qos_policy_params.qos_through_qos[next.tep_qos_clamp]); |
| 978 | |
| 979 | if (requested.trp_over_through_qos != 0) |
| 980 | through_qos = requested.trp_over_through_qos; |
| 981 | |
| 982 | next.tep_through_qos = through_qos; |
| 983 | |
| 984 | /* Calculate suppressed CPU priority */ |
| 985 | if (requested.trp_sup_cpu && requested.trp_boosted == 0) |
| 986 | next.tep_suppressed_cpu = 1; |
| 987 | |
| 988 | /* |
| 989 | * Calculate background sockets |
| 990 | * Don't take into account boosting to limit transition frequency. |
| 991 | */ |
| 992 | if (requested.trp_sup_bg_sockets){ |
| 993 | next.tep_all_sockets_bg = 1; |
| 994 | next.tep_new_sockets_bg = 1; |
| 995 | } |
| 996 | |
| 997 | /* Apply SFI Managed class bit */ |
| 998 | next.tep_sfi_managed = requested.trp_sfi_managed; |
| 999 | |
| 1000 | /* Calculate 'live donor' status for live importance */ |
| 1001 | switch (requested.trp_apptype) { |
| 1002 | case TASK_APPTYPE_APP_TAL: |
| 1003 | case TASK_APPTYPE_APP_DEFAULT: |
| 1004 | if (requested.trp_ext_darwinbg == 0) |
| 1005 | next.tep_live_donor = 1; |
| 1006 | else |
| 1007 | next.tep_live_donor = 0; |
| 1008 | break; |
| 1009 | |
| 1010 | case TASK_APPTYPE_DAEMON_INTERACTIVE: |
| 1011 | case TASK_APPTYPE_DAEMON_STANDARD: |
| 1012 | case TASK_APPTYPE_DAEMON_ADAPTIVE: |
| 1013 | case TASK_APPTYPE_DAEMON_BACKGROUND: |
| 1014 | default: |
| 1015 | next.tep_live_donor = 0; |
| 1016 | break; |
| 1017 | } |
| 1018 | |
| 1019 | if (requested.trp_terminated) { |
| 1020 | /* |
| 1021 | * Shoot down the throttles that slow down exit or response to SIGTERM |
| 1022 | * We don't need to shoot down: |
| 1023 | * passive (don't want to cause others to throttle) |
| 1024 | * all_sockets_bg (don't need to iterate FDs on every exit) |
| 1025 | * new_sockets_bg (doesn't matter for exiting process) |
| 1026 | * pidsuspend (jetsam-ed BG process shouldn't run again) |
| 1027 | * watchers_bg (watcher threads don't need to be unthrottled) |
| 1028 | * latency_qos (affects userspace timers only) |
| 1029 | */ |
| 1030 | |
| 1031 | next.tep_terminated = 1; |
| 1032 | next.tep_darwinbg = 0; |
| 1033 | next.tep_lowpri_cpu = 0; |
| 1034 | next.tep_io_tier = THROTTLE_LEVEL_TIER0; |
| 1035 | next.tep_tal_engaged = 0; |
| 1036 | next.tep_role = TASK_UNSPECIFIED; |
| 1037 | next.tep_suppressed_cpu = 0; |
| 1038 | } |
| 1039 | |
| 1040 | /* |
| 1041 | * Step 3: |
| 1042 | * Swap out old policy for new policy |
| 1043 | */ |
| 1044 | |
| 1045 | struct task_effective_policy prev = task->effective_policy; |
| 1046 | |
| 1047 | /* This is the point where the new values become visible to other threads */ |
| 1048 | task->effective_policy = next; |
| 1049 | |
| 1050 | /* Don't do anything further to a half-formed task */ |
| 1051 | if (in_create) |
| 1052 | return; |
| 1053 | |
| 1054 | if (task == kernel_task) |
| 1055 | panic("Attempting to set task policy on kernel_task"); |
| 1056 | |
| 1057 | /* |
| 1058 | * Step 4: |
| 1059 | * Pend updates that can't be done while holding the task lock |
| 1060 | */ |
| 1061 | |
| 1062 | if (prev.tep_all_sockets_bg != next.tep_all_sockets_bg) |
| 1063 | pend_token->tpt_update_sockets = 1; |
| 1064 | |
| 1065 | /* Only re-scan the timer list if the qos level is getting less strong */ |
| 1066 | if (prev.tep_latency_qos > next.tep_latency_qos) |
| 1067 | pend_token->tpt_update_timers = 1; |
| 1068 | |
| 1069 | #if CONFIG_EMBEDDED |
| 1070 | if (prev.tep_watchers_bg != next.tep_watchers_bg) |
| 1071 | pend_token->tpt_update_watchers = 1; |
| 1072 | #endif /* CONFIG_EMBEDDED */ |
| 1073 | |
| 1074 | if (prev.tep_live_donor != next.tep_live_donor) |
| 1075 | pend_token->tpt_update_live_donor = 1; |
| 1076 | |
| 1077 | /* |
| 1078 | * Step 5: |
| 1079 | * Update other subsystems as necessary if something has changed |
| 1080 | */ |
| 1081 | |
| 1082 | boolean_t update_threads = FALSE, update_sfi = FALSE; |
| 1083 | |
| 1084 | /* |
| 1085 | * Check for the attributes that thread_policy_update_internal_locked() consults, |
| 1086 | * and trigger thread policy re-evaluation. |
| 1087 | */ |
| 1088 | if (prev.tep_io_tier != next.tep_io_tier || |
| 1089 | prev.tep_bg_iotier != next.tep_bg_iotier || |
| 1090 | prev.tep_io_passive != next.tep_io_passive || |
| 1091 | prev.tep_darwinbg != next.tep_darwinbg || |
| 1092 | prev.tep_qos_clamp != next.tep_qos_clamp || |
| 1093 | prev.tep_qos_ceiling != next.tep_qos_ceiling || |
| 1094 | prev.tep_qos_ui_is_urgent != next.tep_qos_ui_is_urgent || |
| 1095 | prev.tep_latency_qos != next.tep_latency_qos || |
| 1096 | prev.tep_through_qos != next.tep_through_qos || |
| 1097 | prev.tep_lowpri_cpu != next.tep_lowpri_cpu || |
| 1098 | prev.tep_new_sockets_bg != next.tep_new_sockets_bg || |
| 1099 | prev.tep_terminated != next.tep_terminated ) |
| 1100 | update_threads = TRUE; |
| 1101 | |
| 1102 | /* |
| 1103 | * Check for the attributes that sfi_thread_classify() consults, |
| 1104 | * and trigger SFI re-evaluation. |
| 1105 | */ |
| 1106 | if (prev.tep_latency_qos != next.tep_latency_qos || |
| 1107 | prev.tep_role != next.tep_role || |
| 1108 | prev.tep_sfi_managed != next.tep_sfi_managed ) |
| 1109 | update_sfi = TRUE; |
| 1110 | |
| 1111 | /* Reflect task role transitions into the coalition role counters */ |
| 1112 | if (prev.tep_role != next.tep_role) { |
| 1113 | if (task_policy_update_coalition_focal_tasks(task, prev.tep_role, next.tep_role, pend_token)) |
| 1114 | update_sfi = TRUE; |
| 1115 | } |
| 1116 | |
| 1117 | boolean_t update_priority = FALSE; |
| 1118 | |
| 1119 | int priority = BASEPRI_DEFAULT; |
| 1120 | int max_priority = MAXPRI_USER; |
| 1121 | |
| 1122 | if (next.tep_lowpri_cpu) { |
| 1123 | priority = MAXPRI_THROTTLE; |
| 1124 | max_priority = MAXPRI_THROTTLE; |
| 1125 | } else if (next.tep_suppressed_cpu) { |
| 1126 | priority = MAXPRI_SUPPRESSED; |
| 1127 | max_priority = MAXPRI_SUPPRESSED; |
| 1128 | } else { |
| 1129 | switch (next.tep_role) { |
| 1130 | case TASK_CONTROL_APPLICATION: |
| 1131 | priority = BASEPRI_CONTROL; |
| 1132 | break; |
| 1133 | case TASK_GRAPHICS_SERVER: |
| 1134 | priority = BASEPRI_GRAPHICS; |
| 1135 | max_priority = MAXPRI_RESERVED; |
| 1136 | break; |
| 1137 | default: |
| 1138 | break; |
| 1139 | } |
| 1140 | |
| 1141 | /* factor in 'nice' value */ |
| 1142 | priority += task->importance; |
| 1143 | |
| 1144 | if (task->effective_policy.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) { |
| 1145 | int qos_clamp_priority = thread_qos_policy_params.qos_pri[task->effective_policy.tep_qos_clamp]; |
| 1146 | |
| 1147 | priority = MIN(priority, qos_clamp_priority); |
| 1148 | max_priority = MIN(max_priority, qos_clamp_priority); |
| 1149 | } |
| 1150 | |
| 1151 | if (priority > max_priority) |
| 1152 | priority = max_priority; |
| 1153 | else if (priority < MINPRI) |
| 1154 | priority = MINPRI; |
| 1155 | } |
| 1156 | |
| 1157 | assert(priority <= max_priority); |
| 1158 | |
| 1159 | /* avoid extra work if priority isn't changing */ |
| 1160 | if (priority != task->priority || |
| 1161 | max_priority != task->max_priority ) { |
| 1162 | /* update the scheduling priority for the task */ |
| 1163 | task->max_priority = max_priority; |
| 1164 | task->priority = priority; |
| 1165 | update_priority = TRUE; |
| 1166 | } |
| 1167 | |
| 1168 | /* Loop over the threads in the task: |
| 1169 | * only once |
| 1170 | * only if necessary |
| 1171 | * with one thread mutex hold per thread |
| 1172 | */ |
| 1173 | if (update_threads || update_priority || update_sfi) { |
| 1174 | thread_t thread; |
| 1175 | |
| 1176 | queue_iterate(&task->threads, thread, thread_t, task_threads) { |
| 1177 | struct task_pend_token thread_pend_token = {}; |
| 1178 | |
| 1179 | if (update_sfi) |
| 1180 | thread_pend_token.tpt_update_thread_sfi = 1; |
| 1181 | |
| 1182 | if (update_priority || update_threads) |
| 1183 | thread_policy_update_tasklocked(thread, |
| 1184 | task->priority, task->max_priority, |
| 1185 | &thread_pend_token); |
| 1186 | |
| 1187 | assert(!thread_pend_token.tpt_update_sockets); |
| 1188 | |
| 1189 | // Slightly risky, as we still hold the task lock... |
| 1190 | thread_policy_update_complete_unlocked(thread, &thread_pend_token); |
| 1191 | } |
| 1192 | } |
| 1193 | |
| 1194 | /* |
| 1195 | * Use the app-nap transitions to influence the |
| 1196 | * transition of the process within the jetsam band. |
| 1197 | * On macOS only. |
| 1198 | */ |
| 1199 | if (memorystatus_appnap_transition == TRUE) { |
| 1200 | if (task->effective_policy.tep_sup_active == 1) { |
| 1201 | memorystatus_update_priority_for_appnap(((proc_t) task->bsd_info), TRUE); |
| 1202 | } else { |
| 1203 | memorystatus_update_priority_for_appnap(((proc_t) task->bsd_info), FALSE); |
| 1204 | } |
| 1205 | } |
| 1206 | } |
| 1207 | |
| 1208 | |
| 1209 | /* |
| 1210 | * Yet another layering violation. We reach out and bang on the coalition directly. |
| 1211 | */ |
| 1212 | static boolean_t |
| 1213 | task_policy_update_coalition_focal_tasks(task_t task, |
| 1214 | int prev_role, |
| 1215 | int next_role, |
| 1216 | task_pend_token_t pend_token) |
| 1217 | { |
| 1218 | boolean_t sfi_transition = FALSE; |
| 1219 | uint32_t new_count = 0; |
| 1220 | |
| 1221 | /* task moving into/out-of the foreground */ |
| 1222 | if (prev_role != TASK_FOREGROUND_APPLICATION && next_role == TASK_FOREGROUND_APPLICATION) { |
| 1223 | if (task_coalition_adjust_focal_count(task, 1, &new_count) && (new_count == 1)) { |
| 1224 | sfi_transition = TRUE; |
| 1225 | pend_token->tpt_update_tg_ui_flag = TRUE; |
| 1226 | } |
| 1227 | } else if (prev_role == TASK_FOREGROUND_APPLICATION && next_role != TASK_FOREGROUND_APPLICATION) { |
| 1228 | if (task_coalition_adjust_focal_count(task, -1, &new_count) && (new_count == 0)) { |
| 1229 | sfi_transition = TRUE; |
| 1230 | pend_token->tpt_update_tg_ui_flag = TRUE; |
| 1231 | } |
| 1232 | } |
| 1233 | |
| 1234 | /* task moving into/out-of background */ |
| 1235 | if (prev_role != TASK_BACKGROUND_APPLICATION && next_role == TASK_BACKGROUND_APPLICATION) { |
| 1236 | if (task_coalition_adjust_nonfocal_count(task, 1, &new_count) && (new_count == 1)) |
| 1237 | sfi_transition = TRUE; |
| 1238 | } else if (prev_role == TASK_BACKGROUND_APPLICATION && next_role != TASK_BACKGROUND_APPLICATION) { |
| 1239 | if (task_coalition_adjust_nonfocal_count(task, -1, &new_count) && (new_count == 0)) |
| 1240 | sfi_transition = TRUE; |
| 1241 | } |
| 1242 | |
| 1243 | if (sfi_transition) |
| 1244 | pend_token->tpt_update_coal_sfi = 1; |
| 1245 | return sfi_transition; |
| 1246 | } |
| 1247 | |
| 1248 | #if CONFIG_SCHED_SFI |
| 1249 | |
| 1250 | /* coalition object is locked */ |
| 1251 | static void |
| 1252 | task_sfi_reevaluate_cb(coalition_t coal, void *ctx, task_t task) |
| 1253 | { |
| 1254 | thread_t thread; |
| 1255 | |
| 1256 | /* unused for now */ |
| 1257 | (void)coal; |
| 1258 | |
| 1259 | /* skip the task we're re-evaluating on behalf of: it's already updated */ |
| 1260 | if (task == (task_t)ctx) |
| 1261 | return; |
| 1262 | |
| 1263 | task_lock(task); |
| 1264 | |
| 1265 | queue_iterate(&task->threads, thread, thread_t, task_threads) { |
| 1266 | sfi_reevaluate(thread); |
| 1267 | } |
| 1268 | |
| 1269 | task_unlock(task); |
| 1270 | } |
| 1271 | #endif /* CONFIG_SCHED_SFI */ |
| 1272 | |
| 1273 | /* |
| 1274 | * Called with task unlocked to do things that can't be done while holding the task lock |
| 1275 | */ |
| 1276 | void |
| 1277 | task_policy_update_complete_unlocked(task_t task, task_pend_token_t pend_token) |
| 1278 | { |
| 1279 | #ifdef MACH_BSD |
| 1280 | if (pend_token->tpt_update_sockets) |
| 1281 | proc_apply_task_networkbg(task->bsd_info, THREAD_NULL); |
| 1282 | #endif /* MACH_BSD */ |
| 1283 | |
| 1284 | /* The timer throttle has been removed or reduced, we need to look for expired timers and fire them */ |
| 1285 | if (pend_token->tpt_update_timers) |
| 1286 | ml_timer_evaluate(); |
| 1287 | |
| 1288 | #if CONFIG_EMBEDDED |
| 1289 | if (pend_token->tpt_update_watchers) |
| 1290 | apply_appstate_watchers(task); |
| 1291 | #endif /* CONFIG_EMBEDDED */ |
| 1292 | |
| 1293 | if (pend_token->tpt_update_live_donor) |
| 1294 | task_importance_update_live_donor(task); |
| 1295 | |
| 1296 | #if CONFIG_SCHED_SFI |
| 1297 | /* use the resource coalition for SFI re-evaluation */ |
| 1298 | if (pend_token->tpt_update_coal_sfi) |
| 1299 | coalition_for_each_task(task->coalition[COALITION_TYPE_RESOURCE], |
| 1300 | (void *)task, task_sfi_reevaluate_cb); |
| 1301 | #endif /* CONFIG_SCHED_SFI */ |
| 1302 | |
| 1303 | } |
| 1304 | |
| 1305 | /* |
| 1306 | * Initiate a task policy state transition |
| 1307 | * |
| 1308 | * Everything that modifies requested except functions that need to hold the task lock |
| 1309 | * should use this function |
| 1310 | * |
| 1311 | * Argument validation should be performed before reaching this point. |
| 1312 | * |
| 1313 | * TODO: Do we need to check task->active? |
| 1314 | */ |
| 1315 | void |
| 1316 | proc_set_task_policy(task_t task, |
| 1317 | int category, |
| 1318 | int flavor, |
| 1319 | int value) |
| 1320 | { |
| 1321 | struct task_pend_token pend_token = {}; |
| 1322 | |
| 1323 | task_lock(task); |
| 1324 | |
| 1325 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 1326 | (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_START, |
| 1327 | task_pid(task), trequested_0(task), |
| 1328 | trequested_1(task), value, 0); |
| 1329 | |
| 1330 | proc_set_task_policy_locked(task, category, flavor, value, 0); |
| 1331 | |
| 1332 | task_policy_update_locked(task, &pend_token); |
| 1333 | |
| 1334 | |
| 1335 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 1336 | (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_END, |
| 1337 | task_pid(task), trequested_0(task), |
| 1338 | trequested_1(task), tpending(&pend_token), 0); |
| 1339 | |
| 1340 | task_unlock(task); |
| 1341 | |
| 1342 | task_policy_update_complete_unlocked(task, &pend_token); |
| 1343 | } |
| 1344 | |
| 1345 | /* |
| 1346 | * Variant of proc_set_task_policy() that sets two scalars in the requested policy structure. |
| 1347 | * Same locking rules apply. |
| 1348 | */ |
| 1349 | void |
| 1350 | proc_set_task_policy2(task_t task, |
| 1351 | int category, |
| 1352 | int flavor, |
| 1353 | int value, |
| 1354 | int value2) |
| 1355 | { |
| 1356 | struct task_pend_token pend_token = {}; |
| 1357 | |
| 1358 | task_lock(task); |
| 1359 | |
| 1360 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 1361 | (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_START, |
| 1362 | task_pid(task), trequested_0(task), |
| 1363 | trequested_1(task), value, 0); |
| 1364 | |
| 1365 | proc_set_task_policy_locked(task, category, flavor, value, value2); |
| 1366 | |
| 1367 | task_policy_update_locked(task, &pend_token); |
| 1368 | |
| 1369 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 1370 | (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_END, |
| 1371 | task_pid(task), trequested_0(task), |
| 1372 | trequested_1(task), tpending(&pend_token), 0); |
| 1373 | |
| 1374 | task_unlock(task); |
| 1375 | |
| 1376 | task_policy_update_complete_unlocked(task, &pend_token); |
| 1377 | } |
| 1378 | |
| 1379 | /* |
| 1380 | * Set the requested state for a specific flavor to a specific value. |
| 1381 | * |
| 1382 | * TODO: |
| 1383 | * Verify that arguments to non iopol things are 1 or 0 |
| 1384 | */ |
| 1385 | static void |
| 1386 | proc_set_task_policy_locked(task_t task, |
| 1387 | int category, |
| 1388 | int flavor, |
| 1389 | int value, |
| 1390 | int value2) |
| 1391 | { |
| 1392 | int tier, passive; |
| 1393 | |
| 1394 | struct task_requested_policy requested = task->requested_policy; |
| 1395 | |
| 1396 | switch (flavor) { |
| 1397 | |
| 1398 | /* Category: EXTERNAL and INTERNAL */ |
| 1399 | |
| 1400 | case TASK_POLICY_DARWIN_BG: |
| 1401 | if (category == TASK_POLICY_EXTERNAL) |
| 1402 | requested.trp_ext_darwinbg = value; |
| 1403 | else |
| 1404 | requested.trp_int_darwinbg = value; |
| 1405 | break; |
| 1406 | |
| 1407 | case TASK_POLICY_IOPOL: |
| 1408 | proc_iopol_to_tier(value, &tier, &passive); |
| 1409 | if (category == TASK_POLICY_EXTERNAL) { |
| 1410 | requested.trp_ext_iotier = tier; |
| 1411 | requested.trp_ext_iopassive = passive; |
| 1412 | } else { |
| 1413 | requested.trp_int_iotier = tier; |
| 1414 | requested.trp_int_iopassive = passive; |
| 1415 | } |
| 1416 | break; |
| 1417 | |
| 1418 | case TASK_POLICY_IO: |
| 1419 | if (category == TASK_POLICY_EXTERNAL) |
| 1420 | requested.trp_ext_iotier = value; |
| 1421 | else |
| 1422 | requested.trp_int_iotier = value; |
| 1423 | break; |
| 1424 | |
| 1425 | case TASK_POLICY_PASSIVE_IO: |
| 1426 | if (category == TASK_POLICY_EXTERNAL) |
| 1427 | requested.trp_ext_iopassive = value; |
| 1428 | else |
| 1429 | requested.trp_int_iopassive = value; |
| 1430 | break; |
| 1431 | |
| 1432 | /* Category: INTERNAL */ |
| 1433 | |
| 1434 | case TASK_POLICY_DARWIN_BG_IOPOL: |
| 1435 | assert(category == TASK_POLICY_INTERNAL); |
| 1436 | proc_iopol_to_tier(value, &tier, &passive); |
| 1437 | requested.trp_bg_iotier = tier; |
| 1438 | break; |
| 1439 | |
| 1440 | /* Category: ATTRIBUTE */ |
| 1441 | |
| 1442 | case TASK_POLICY_TAL: |
| 1443 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1444 | requested.trp_tal_enabled = value; |
| 1445 | break; |
| 1446 | |
| 1447 | case TASK_POLICY_BOOST: |
| 1448 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1449 | requested.trp_boosted = value; |
| 1450 | break; |
| 1451 | |
| 1452 | case TASK_POLICY_ROLE: |
| 1453 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1454 | requested.trp_role = value; |
| 1455 | break; |
| 1456 | |
| 1457 | case TASK_POLICY_TERMINATED: |
| 1458 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1459 | requested.trp_terminated = value; |
| 1460 | break; |
| 1461 | |
| 1462 | case TASK_BASE_LATENCY_QOS_POLICY: |
| 1463 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1464 | requested.trp_base_latency_qos = value; |
| 1465 | break; |
| 1466 | |
| 1467 | case TASK_BASE_THROUGHPUT_QOS_POLICY: |
| 1468 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1469 | requested.trp_base_through_qos = value; |
| 1470 | break; |
| 1471 | |
| 1472 | case TASK_POLICY_SFI_MANAGED: |
| 1473 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1474 | requested.trp_sfi_managed = value; |
| 1475 | break; |
| 1476 | |
| 1477 | case TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS: |
| 1478 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1479 | requested.trp_base_latency_qos = value; |
| 1480 | requested.trp_base_through_qos = value2; |
| 1481 | break; |
| 1482 | |
| 1483 | case TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS: |
| 1484 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1485 | requested.trp_over_latency_qos = value; |
| 1486 | requested.trp_over_through_qos = value2; |
| 1487 | break; |
| 1488 | |
| 1489 | default: |
| 1490 | panic("unknown task policy: %d %d %d %d", category, flavor, value, value2); |
| 1491 | break; |
| 1492 | } |
| 1493 | |
| 1494 | task->requested_policy = requested; |
| 1495 | } |
| 1496 | |
| 1497 | /* |
| 1498 | * Gets what you set. Effective values may be different. |
| 1499 | */ |
| 1500 | int |
| 1501 | proc_get_task_policy(task_t task, |
| 1502 | int category, |
| 1503 | int flavor) |
| 1504 | { |
| 1505 | int value = 0; |
| 1506 | |
| 1507 | task_lock(task); |
| 1508 | |
| 1509 | struct task_requested_policy requested = task->requested_policy; |
| 1510 | |
| 1511 | switch (flavor) { |
| 1512 | case TASK_POLICY_DARWIN_BG: |
| 1513 | if (category == TASK_POLICY_EXTERNAL) |
| 1514 | value = requested.trp_ext_darwinbg; |
| 1515 | else |
| 1516 | value = requested.trp_int_darwinbg; |
| 1517 | break; |
| 1518 | case TASK_POLICY_IOPOL: |
| 1519 | if (category == TASK_POLICY_EXTERNAL) |
| 1520 | value = proc_tier_to_iopol(requested.trp_ext_iotier, |
| 1521 | requested.trp_ext_iopassive); |
| 1522 | else |
| 1523 | value = proc_tier_to_iopol(requested.trp_int_iotier, |
| 1524 | requested.trp_int_iopassive); |
| 1525 | break; |
| 1526 | case TASK_POLICY_IO: |
| 1527 | if (category == TASK_POLICY_EXTERNAL) |
| 1528 | value = requested.trp_ext_iotier; |
| 1529 | else |
| 1530 | value = requested.trp_int_iotier; |
| 1531 | break; |
| 1532 | case TASK_POLICY_PASSIVE_IO: |
| 1533 | if (category == TASK_POLICY_EXTERNAL) |
| 1534 | value = requested.trp_ext_iopassive; |
| 1535 | else |
| 1536 | value = requested.trp_int_iopassive; |
| 1537 | break; |
| 1538 | case TASK_POLICY_DARWIN_BG_IOPOL: |
| 1539 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1540 | value = proc_tier_to_iopol(requested.trp_bg_iotier, 0); |
| 1541 | break; |
| 1542 | case TASK_POLICY_ROLE: |
| 1543 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1544 | value = requested.trp_role; |
| 1545 | break; |
| 1546 | case TASK_POLICY_SFI_MANAGED: |
| 1547 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1548 | value = requested.trp_sfi_managed; |
| 1549 | break; |
| 1550 | default: |
| 1551 | panic("unknown policy_flavor %d", flavor); |
| 1552 | break; |
| 1553 | } |
| 1554 | |
| 1555 | task_unlock(task); |
| 1556 | |
| 1557 | return value; |
| 1558 | } |
| 1559 | |
| 1560 | /* |
| 1561 | * Variant of proc_get_task_policy() that returns two scalar outputs. |
| 1562 | */ |
| 1563 | void |
| 1564 | proc_get_task_policy2(task_t task, |
| 1565 | __assert_only int category, |
| 1566 | int flavor, |
| 1567 | int *value1, |
| 1568 | int *value2) |
| 1569 | { |
| 1570 | task_lock(task); |
| 1571 | |
| 1572 | struct task_requested_policy requested = task->requested_policy; |
| 1573 | |
| 1574 | switch (flavor) { |
| 1575 | case TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS: |
| 1576 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1577 | *value1 = requested.trp_base_latency_qos; |
| 1578 | *value2 = requested.trp_base_through_qos; |
| 1579 | break; |
| 1580 | |
| 1581 | case TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS: |
| 1582 | assert(category == TASK_POLICY_ATTRIBUTE); |
| 1583 | *value1 = requested.trp_over_latency_qos; |
| 1584 | *value2 = requested.trp_over_through_qos; |
| 1585 | break; |
| 1586 | |
| 1587 | default: |
| 1588 | panic("unknown policy_flavor %d", flavor); |
| 1589 | break; |
| 1590 | } |
| 1591 | |
| 1592 | task_unlock(task); |
| 1593 | } |
| 1594 | |
| 1595 | /* |
| 1596 | * Function for querying effective state for relevant subsystems |
| 1597 | * Gets what is actually in effect, for subsystems which pull policy instead of receive updates. |
| 1598 | * |
| 1599 | * ONLY the relevant subsystem should query this. |
| 1600 | * NEVER take a value from the 'effective' function and stuff it into a setter. |
| 1601 | * |
| 1602 | * NOTE: This accessor does not take the task lock. |
| 1603 | * Notifications of state updates need to be externally synchronized with state queries. |
| 1604 | * This routine *MUST* remain interrupt safe, as it is potentially invoked |
| 1605 | * within the context of a timer interrupt. It is also called in KDP context for stackshot. |
| 1606 | */ |
| 1607 | int |
| 1608 | proc_get_effective_task_policy(task_t task, |
| 1609 | int flavor) |
| 1610 | { |
| 1611 | int value = 0; |
| 1612 | |
| 1613 | switch (flavor) { |
| 1614 | case TASK_POLICY_DARWIN_BG: |
| 1615 | /* |
| 1616 | * This backs the KPI call proc_pidbackgrounded to find |
| 1617 | * out if a pid is backgrounded. |
| 1618 | * It is used to communicate state to the VM system, as well as |
| 1619 | * prioritizing requests to the graphics system. |
| 1620 | * Returns 1 for background mode, 0 for normal mode |
| 1621 | */ |
| 1622 | value = task->effective_policy.tep_darwinbg; |
| 1623 | break; |
| 1624 | case TASK_POLICY_ALL_SOCKETS_BG: |
| 1625 | /* |
| 1626 | * do_background_socket() calls this to determine what it should do to the proc's sockets |
| 1627 | * Returns 1 for background mode, 0 for normal mode |
| 1628 | * |
| 1629 | * This consults both thread and task so un-DBGing a thread while the task is BG |
| 1630 | * doesn't get you out of the network throttle. |
| 1631 | */ |
| 1632 | value = task->effective_policy.tep_all_sockets_bg; |
| 1633 | break; |
| 1634 | case TASK_POLICY_LATENCY_QOS: |
| 1635 | /* |
| 1636 | * timer arming calls into here to find out the timer coalescing level |
| 1637 | * Returns a QoS tier (0-6) |
| 1638 | */ |
| 1639 | value = task->effective_policy.tep_latency_qos; |
| 1640 | break; |
| 1641 | case TASK_POLICY_THROUGH_QOS: |
| 1642 | /* |
| 1643 | * This value is passed into the urgency callout from the scheduler |
| 1644 | * to the performance management subsystem. |
| 1645 | * Returns a QoS tier (0-6) |
| 1646 | */ |
| 1647 | value = task->effective_policy.tep_through_qos; |
| 1648 | break; |
| 1649 | case TASK_POLICY_ROLE: |
| 1650 | /* |
| 1651 | * This controls various things that ask whether a process is foreground, |
| 1652 | * like SFI, VM, access to GPU, etc |
| 1653 | */ |
| 1654 | value = task->effective_policy.tep_role; |
| 1655 | break; |
| 1656 | case TASK_POLICY_WATCHERS_BG: |
| 1657 | /* |
| 1658 | * This controls whether or not a thread watching this process should be BG. |
| 1659 | */ |
| 1660 | value = task->effective_policy.tep_watchers_bg; |
| 1661 | break; |
| 1662 | case TASK_POLICY_SFI_MANAGED: |
| 1663 | /* |
| 1664 | * This controls whether or not a process is targeted for specific control by thermald. |
| 1665 | */ |
| 1666 | value = task->effective_policy.tep_sfi_managed; |
| 1667 | break; |
| 1668 | default: |
| 1669 | panic("unknown policy_flavor %d", flavor); |
| 1670 | break; |
| 1671 | } |
| 1672 | |
| 1673 | return value; |
| 1674 | } |
| 1675 | |
| 1676 | /* |
| 1677 | * Convert from IOPOL_* values to throttle tiers. |
| 1678 | * |
| 1679 | * TODO: Can this be made more compact, like an array lookup |
| 1680 | * Note that it is possible to support e.g. IOPOL_PASSIVE_STANDARD in the future |
| 1681 | */ |
| 1682 | |
| 1683 | void |
| 1684 | proc_iopol_to_tier(int iopolicy, int *tier, int *passive) |
| 1685 | { |
| 1686 | *passive = 0; |
| 1687 | *tier = 0; |
| 1688 | switch (iopolicy) { |
| 1689 | case IOPOL_IMPORTANT: |
| 1690 | *tier = THROTTLE_LEVEL_TIER0; |
| 1691 | break; |
| 1692 | case IOPOL_PASSIVE: |
| 1693 | *tier = THROTTLE_LEVEL_TIER0; |
| 1694 | *passive = 1; |
| 1695 | break; |
| 1696 | case IOPOL_STANDARD: |
| 1697 | *tier = THROTTLE_LEVEL_TIER1; |
| 1698 | break; |
| 1699 | case IOPOL_UTILITY: |
| 1700 | *tier = THROTTLE_LEVEL_TIER2; |
| 1701 | break; |
| 1702 | case IOPOL_THROTTLE: |
| 1703 | *tier = THROTTLE_LEVEL_TIER3; |
| 1704 | break; |
| 1705 | default: |
| 1706 | panic("unknown I/O policy %d", iopolicy); |
| 1707 | break; |
| 1708 | } |
| 1709 | } |
| 1710 | |
| 1711 | int |
| 1712 | proc_tier_to_iopol(int tier, int passive) |
| 1713 | { |
| 1714 | if (passive == 1) { |
| 1715 | switch (tier) { |
| 1716 | case THROTTLE_LEVEL_TIER0: |
| 1717 | return IOPOL_PASSIVE; |
| 1718 | default: |
| 1719 | panic("unknown passive tier %d", tier); |
| 1720 | return IOPOL_DEFAULT; |
| 1721 | } |
| 1722 | } else { |
| 1723 | switch (tier) { |
| 1724 | case THROTTLE_LEVEL_NONE: |
| 1725 | case THROTTLE_LEVEL_TIER0: |
| 1726 | return IOPOL_DEFAULT; |
| 1727 | case THROTTLE_LEVEL_TIER1: |
| 1728 | return IOPOL_STANDARD; |
| 1729 | case THROTTLE_LEVEL_TIER2: |
| 1730 | return IOPOL_UTILITY; |
| 1731 | case THROTTLE_LEVEL_TIER3: |
| 1732 | return IOPOL_THROTTLE; |
| 1733 | default: |
| 1734 | panic("unknown tier %d", tier); |
| 1735 | return IOPOL_DEFAULT; |
| 1736 | } |
| 1737 | } |
| 1738 | } |
| 1739 | |
| 1740 | int |
| 1741 | proc_darwin_role_to_task_role(int darwin_role, int* task_role) |
| 1742 | { |
| 1743 | integer_t role = TASK_UNSPECIFIED; |
| 1744 | |
| 1745 | switch (darwin_role) { |
| 1746 | case PRIO_DARWIN_ROLE_DEFAULT: |
| 1747 | role = TASK_UNSPECIFIED; |
| 1748 | break; |
| 1749 | case PRIO_DARWIN_ROLE_UI_FOCAL: |
| 1750 | role = TASK_FOREGROUND_APPLICATION; |
| 1751 | break; |
| 1752 | case PRIO_DARWIN_ROLE_UI: |
| 1753 | role = TASK_DEFAULT_APPLICATION; |
| 1754 | break; |
| 1755 | case PRIO_DARWIN_ROLE_NON_UI: |
| 1756 | role = TASK_NONUI_APPLICATION; |
| 1757 | break; |
| 1758 | case PRIO_DARWIN_ROLE_UI_NON_FOCAL: |
| 1759 | role = TASK_BACKGROUND_APPLICATION; |
| 1760 | break; |
| 1761 | case PRIO_DARWIN_ROLE_TAL_LAUNCH: |
| 1762 | role = TASK_THROTTLE_APPLICATION; |
| 1763 | break; |
| 1764 | default: |
| 1765 | return EINVAL; |
| 1766 | } |
| 1767 | |
| 1768 | *task_role = role; |
| 1769 | |
| 1770 | return 0; |
| 1771 | } |
| 1772 | |
| 1773 | int |
| 1774 | proc_task_role_to_darwin_role(int task_role) |
| 1775 | { |
| 1776 | switch (task_role) { |
| 1777 | case TASK_FOREGROUND_APPLICATION: |
| 1778 | return PRIO_DARWIN_ROLE_UI_FOCAL; |
| 1779 | case TASK_BACKGROUND_APPLICATION: |
| 1780 | return PRIO_DARWIN_ROLE_UI_NON_FOCAL; |
| 1781 | case TASK_NONUI_APPLICATION: |
| 1782 | return PRIO_DARWIN_ROLE_NON_UI; |
| 1783 | case TASK_DEFAULT_APPLICATION: |
| 1784 | return PRIO_DARWIN_ROLE_UI; |
| 1785 | case TASK_THROTTLE_APPLICATION: |
| 1786 | return PRIO_DARWIN_ROLE_TAL_LAUNCH; |
| 1787 | case TASK_UNSPECIFIED: |
| 1788 | default: |
| 1789 | return PRIO_DARWIN_ROLE_DEFAULT; |
| 1790 | } |
| 1791 | } |
| 1792 | |
| 1793 | |
| 1794 | /* TODO: remove this variable when interactive daemon audit period is over */ |
| 1795 | extern boolean_t ipc_importance_interactive_receiver; |
| 1796 | |
| 1797 | /* |
| 1798 | * Called at process exec to initialize the apptype, qos clamp, and qos seed of a process |
| 1799 | * |
| 1800 | * TODO: Make this function more table-driven instead of ad-hoc |
| 1801 | */ |
| 1802 | void |
| 1803 | proc_set_task_spawnpolicy(task_t task, int apptype, int qos_clamp, int role, |
| 1804 | ipc_port_t * portwatch_ports, int portwatch_count) |
| 1805 | { |
| 1806 | struct task_pend_token pend_token = {}; |
| 1807 | |
| 1808 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 1809 | (IMPORTANCE_CODE(IMP_TASK_APPTYPE, apptype)) | DBG_FUNC_START, |
| 1810 | task_pid(task), trequested_0(task), trequested_1(task), |
| 1811 | apptype, 0); |
| 1812 | |
| 1813 | switch (apptype) { |
| 1814 | case TASK_APPTYPE_APP_TAL: |
| 1815 | case TASK_APPTYPE_APP_DEFAULT: |
| 1816 | /* Apps become donors via the 'live-donor' flag instead of the static donor flag */ |
| 1817 | task_importance_mark_donor(task, FALSE); |
| 1818 | task_importance_mark_live_donor(task, TRUE); |
| 1819 | task_importance_mark_receiver(task, FALSE); |
| 1820 | #if CONFIG_EMBEDDED |
| 1821 | task_importance_mark_denap_receiver(task, FALSE); |
| 1822 | #else |
| 1823 | /* Apps are de-nap recievers on desktop for suppression behaviors */ |
| 1824 | task_importance_mark_denap_receiver(task, TRUE); |
| 1825 | #endif /* CONFIG_EMBEDDED */ |
| 1826 | break; |
| 1827 | |
| 1828 | case TASK_APPTYPE_DAEMON_INTERACTIVE: |
| 1829 | task_importance_mark_donor(task, TRUE); |
| 1830 | task_importance_mark_live_donor(task, FALSE); |
| 1831 | |
| 1832 | /* |
| 1833 | * A boot arg controls whether interactive daemons are importance receivers. |
| 1834 | * Normally, they are not. But for testing their behavior as an adaptive |
| 1835 | * daemon, the boot-arg can be set. |
| 1836 | * |
| 1837 | * TODO: remove this when the interactive daemon audit period is over. |
| 1838 | */ |
| 1839 | task_importance_mark_receiver(task, /* FALSE */ ipc_importance_interactive_receiver); |
| 1840 | task_importance_mark_denap_receiver(task, FALSE); |
| 1841 | break; |
| 1842 | |
| 1843 | case TASK_APPTYPE_DAEMON_STANDARD: |
| 1844 | task_importance_mark_donor(task, TRUE); |
| 1845 | task_importance_mark_live_donor(task, FALSE); |
| 1846 | task_importance_mark_receiver(task, FALSE); |
| 1847 | task_importance_mark_denap_receiver(task, FALSE); |
| 1848 | break; |
| 1849 | |
| 1850 | case TASK_APPTYPE_DAEMON_ADAPTIVE: |
| 1851 | task_importance_mark_donor(task, FALSE); |
| 1852 | task_importance_mark_live_donor(task, FALSE); |
| 1853 | task_importance_mark_receiver(task, TRUE); |
| 1854 | task_importance_mark_denap_receiver(task, FALSE); |
| 1855 | break; |
| 1856 | |
| 1857 | case TASK_APPTYPE_DAEMON_BACKGROUND: |
| 1858 | task_importance_mark_donor(task, FALSE); |
| 1859 | task_importance_mark_live_donor(task, FALSE); |
| 1860 | task_importance_mark_receiver(task, FALSE); |
| 1861 | task_importance_mark_denap_receiver(task, FALSE); |
| 1862 | break; |
| 1863 | |
| 1864 | case TASK_APPTYPE_NONE: |
| 1865 | break; |
| 1866 | } |
| 1867 | |
| 1868 | if (portwatch_ports != NULL && apptype == TASK_APPTYPE_DAEMON_ADAPTIVE) { |
| 1869 | int portwatch_boosts = 0; |
| 1870 | |
| 1871 | for (int i = 0; i < portwatch_count; i++) { |
| 1872 | ipc_port_t port = NULL; |
| 1873 | |
| 1874 | if ((port = portwatch_ports[i]) != NULL) { |
| 1875 | int boost = 0; |
| 1876 | task_add_importance_watchport(task, port, &boost); |
| 1877 | portwatch_boosts += boost; |
| 1878 | } |
| 1879 | } |
| 1880 | |
| 1881 | if (portwatch_boosts > 0) { |
| 1882 | task_importance_hold_internal_assertion(task, portwatch_boosts); |
| 1883 | } |
| 1884 | } |
| 1885 | |
| 1886 | task_lock(task); |
| 1887 | |
| 1888 | if (apptype == TASK_APPTYPE_APP_TAL) { |
| 1889 | /* TAL starts off enabled by default */ |
| 1890 | task->requested_policy.trp_tal_enabled = 1; |
| 1891 | } |
| 1892 | |
| 1893 | if (apptype != TASK_APPTYPE_NONE) { |
| 1894 | task->requested_policy.trp_apptype = apptype; |
| 1895 | } |
| 1896 | |
| 1897 | #if CONFIG_EMBEDDED |
| 1898 | /* Remove this after launchd starts setting it properly */ |
| 1899 | if (apptype == TASK_APPTYPE_APP_DEFAULT && role == TASK_UNSPECIFIED) { |
| 1900 | task->requested_policy.trp_role = TASK_FOREGROUND_APPLICATION; |
| 1901 | } else |
| 1902 | #endif |
| 1903 | if (role != TASK_UNSPECIFIED) { |
| 1904 | task->requested_policy.trp_role = role; |
| 1905 | } |
| 1906 | |
| 1907 | if (qos_clamp != THREAD_QOS_UNSPECIFIED) { |
| 1908 | task->requested_policy.trp_qos_clamp = qos_clamp; |
| 1909 | } |
| 1910 | |
| 1911 | task_policy_update_locked(task, &pend_token); |
| 1912 | |
| 1913 | task_unlock(task); |
| 1914 | |
| 1915 | /* Ensure the donor bit is updated to be in sync with the new live donor status */ |
| 1916 | pend_token.tpt_update_live_donor = 1; |
| 1917 | |
| 1918 | task_policy_update_complete_unlocked(task, &pend_token); |
| 1919 | |
| 1920 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, |
| 1921 | (IMPORTANCE_CODE(IMP_TASK_APPTYPE, apptype)) | DBG_FUNC_END, |
| 1922 | task_pid(task), trequested_0(task), trequested_1(task), |
| 1923 | task_is_importance_receiver(task), 0); |
| 1924 | } |
| 1925 | |
| 1926 | /* |
| 1927 | * Inherit task role across exec |
| 1928 | */ |
| 1929 | void |
| 1930 | proc_inherit_task_role(task_t new_task, |
| 1931 | task_t old_task) |
| 1932 | { |
| 1933 | int role; |
| 1934 | |
| 1935 | /* inherit the role from old task to new task */ |
| 1936 | role = proc_get_task_policy(old_task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE); |
| 1937 | proc_set_task_policy(new_task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE, role); |
| 1938 | } |
| 1939 | |
| 1940 | extern void *initproc; |
| 1941 | |
| 1942 | /* |
| 1943 | * Compute the default main thread qos for a task |
| 1944 | */ |
| 1945 | int |
| 1946 | task_compute_main_thread_qos(task_t task) |
| 1947 | { |
| 1948 | int primordial_qos = THREAD_QOS_UNSPECIFIED; |
| 1949 | |
| 1950 | int qos_clamp = task->requested_policy.trp_qos_clamp; |
| 1951 | |
| 1952 | switch (task->requested_policy.trp_apptype) { |
| 1953 | case TASK_APPTYPE_APP_TAL: |
| 1954 | case TASK_APPTYPE_APP_DEFAULT: |
| 1955 | primordial_qos = THREAD_QOS_USER_INTERACTIVE; |
| 1956 | break; |
| 1957 | |
| 1958 | case TASK_APPTYPE_DAEMON_INTERACTIVE: |
| 1959 | case TASK_APPTYPE_DAEMON_STANDARD: |
| 1960 | case TASK_APPTYPE_DAEMON_ADAPTIVE: |
| 1961 | primordial_qos = THREAD_QOS_LEGACY; |
| 1962 | break; |
| 1963 | |
| 1964 | case TASK_APPTYPE_DAEMON_BACKGROUND: |
| 1965 | primordial_qos = THREAD_QOS_BACKGROUND; |
| 1966 | break; |
| 1967 | } |
| 1968 | |
| 1969 | if (task->bsd_info == initproc) { |
| 1970 | /* PID 1 gets a special case */ |
| 1971 | primordial_qos = MAX(primordial_qos, THREAD_QOS_USER_INITIATED); |
| 1972 | } |
| 1973 | |
| 1974 | if (qos_clamp != THREAD_QOS_UNSPECIFIED) { |
| 1975 | if (primordial_qos != THREAD_QOS_UNSPECIFIED) { |
| 1976 | primordial_qos = MIN(qos_clamp, primordial_qos); |
| 1977 | } else { |
| 1978 | primordial_qos = qos_clamp; |
| 1979 | } |
| 1980 | } |
| 1981 | |
| 1982 | return primordial_qos; |
| 1983 | } |
| 1984 | |
| 1985 | |
| 1986 | /* for process_policy to check before attempting to set */ |
| 1987 | boolean_t |
| 1988 | proc_task_is_tal(task_t task) |
| 1989 | { |
| 1990 | return (task->requested_policy.trp_apptype == TASK_APPTYPE_APP_TAL) ? TRUE : FALSE; |
| 1991 | } |
| 1992 | |
| 1993 | int |
| 1994 | task_get_apptype(task_t task) |
| 1995 | { |
| 1996 | return task->requested_policy.trp_apptype; |
| 1997 | } |
| 1998 | |
| 1999 | boolean_t |
| 2000 | task_is_daemon(task_t task) |
| 2001 | { |
| 2002 | switch (task->requested_policy.trp_apptype) { |
| 2003 | case TASK_APPTYPE_DAEMON_INTERACTIVE: |
| 2004 | case TASK_APPTYPE_DAEMON_STANDARD: |
| 2005 | case TASK_APPTYPE_DAEMON_ADAPTIVE: |
| 2006 | case TASK_APPTYPE_DAEMON_BACKGROUND: |
| 2007 | return TRUE; |
| 2008 | default: |
| 2009 | return FALSE; |
| 2010 | } |
| 2011 | } |
| 2012 | |
| 2013 | boolean_t |
| 2014 | task_is_app(task_t task) |
| 2015 | { |
| 2016 | switch (task->requested_policy.trp_apptype) { |
| 2017 | case TASK_APPTYPE_APP_DEFAULT: |
| 2018 | case TASK_APPTYPE_APP_TAL: |
| 2019 | return TRUE; |
| 2020 | default: |
| 2021 | return FALSE; |
| 2022 | } |
| 2023 | } |
| 2024 | |
| 2025 | /* for telemetry */ |
| 2026 | integer_t |
| 2027 | task_grab_latency_qos(task_t task) |
| 2028 | { |
| 2029 | return qos_latency_policy_package(proc_get_effective_task_policy(task, TASK_POLICY_LATENCY_QOS)); |
| 2030 | } |
| 2031 | |
| 2032 | /* update the darwin background action state in the flags field for libproc */ |
| 2033 | int |
| 2034 | proc_get_darwinbgstate(task_t task, uint32_t * flagsp) |
| 2035 | { |
| 2036 | if (task->requested_policy.trp_ext_darwinbg) |
| 2037 | *flagsp |= PROC_FLAG_EXT_DARWINBG; |
| 2038 | |
| 2039 | if (task->requested_policy.trp_int_darwinbg) |
| 2040 | *flagsp |= PROC_FLAG_DARWINBG; |
| 2041 | |
| 2042 | #if CONFIG_EMBEDDED |
| 2043 | if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_BACKGROUND) |
| 2044 | *flagsp |= PROC_FLAG_IOS_APPLEDAEMON; |
| 2045 | |
| 2046 | if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE) |
| 2047 | *flagsp |= PROC_FLAG_IOS_IMPPROMOTION; |
| 2048 | #endif /* CONFIG_EMBEDDED */ |
| 2049 | |
| 2050 | if (task->requested_policy.trp_apptype == TASK_APPTYPE_APP_DEFAULT || |
| 2051 | task->requested_policy.trp_apptype == TASK_APPTYPE_APP_TAL) |
| 2052 | *flagsp |= PROC_FLAG_APPLICATION; |
| 2053 | |
| 2054 | if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE) |
| 2055 | *flagsp |= PROC_FLAG_ADAPTIVE; |
| 2056 | |
| 2057 | if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE && |
| 2058 | task->requested_policy.trp_boosted == 1) |
| 2059 | *flagsp |= PROC_FLAG_ADAPTIVE_IMPORTANT; |
| 2060 | |
| 2061 | if (task_is_importance_donor(task)) |
| 2062 | *flagsp |= PROC_FLAG_IMPORTANCE_DONOR; |
| 2063 | |
| 2064 | if (task->effective_policy.tep_sup_active) |
| 2065 | *flagsp |= PROC_FLAG_SUPPRESSED; |
| 2066 | |
| 2067 | return(0); |
| 2068 | } |
| 2069 | |
| 2070 | /* |
| 2071 | * Tracepoint data... Reading the tracepoint data can be somewhat complicated. |
| 2072 | * The current scheme packs as much data into a single tracepoint as it can. |
| 2073 | * |
| 2074 | * Each task/thread requested/effective structure is 64 bits in size. Any |
| 2075 | * given tracepoint will emit either requested or effective data, but not both. |
| 2076 | * |
| 2077 | * A tracepoint may emit any of task, thread, or task & thread data. |
| 2078 | * |
| 2079 | * The type of data emitted varies with pointer size. Where possible, both |
| 2080 | * task and thread data are emitted. In LP32 systems, the first and second |
| 2081 | * halves of either the task or thread data is emitted. |
| 2082 | * |
| 2083 | * The code uses uintptr_t array indexes instead of high/low to avoid |
| 2084 | * confusion WRT big vs little endian. |
| 2085 | * |
| 2086 | * The truth table for the tracepoint data functions is below, and has the |
| 2087 | * following invariants: |
| 2088 | * |
| 2089 | * 1) task and thread are uintptr_t* |
| 2090 | * 2) task may never be NULL |
| 2091 | * |
| 2092 | * |
| 2093 | * LP32 LP64 |
| 2094 | * trequested_0(task, NULL) task[0] task[0] |
| 2095 | * trequested_1(task, NULL) task[1] NULL |
| 2096 | * trequested_0(task, thread) thread[0] task[0] |
| 2097 | * trequested_1(task, thread) thread[1] thread[0] |
| 2098 | * |
| 2099 | * Basically, you get a full task or thread on LP32, and both on LP64. |
| 2100 | * |
| 2101 | * The uintptr_t munging here is squicky enough to deserve a comment. |
| 2102 | * |
| 2103 | * The variables we are accessing are laid out in memory like this: |
| 2104 | * |
| 2105 | * [ LP64 uintptr_t 0 ] |
| 2106 | * [ LP32 uintptr_t 0 ] [ LP32 uintptr_t 1 ] |
| 2107 | * |
| 2108 | * 1 2 3 4 5 6 7 8 |
| 2109 | * |
| 2110 | */ |
| 2111 | |
| 2112 | static uintptr_t |
| 2113 | trequested_0(task_t task) |
| 2114 | { |
| 2115 | static_assert(sizeof(struct task_requested_policy) == sizeof(uint64_t), "size invariant violated"); |
| 2116 | |
| 2117 | uintptr_t* raw = (uintptr_t*)&task->requested_policy; |
| 2118 | |
| 2119 | return raw[0]; |
| 2120 | } |
| 2121 | |
| 2122 | static uintptr_t |
| 2123 | trequested_1(task_t task) |
| 2124 | { |
| 2125 | #if defined __LP64__ |
| 2126 | (void)task; |
| 2127 | return 0; |
| 2128 | #else |
| 2129 | uintptr_t* raw = (uintptr_t*)(&task->requested_policy); |
| 2130 | return raw[1]; |
| 2131 | #endif |
| 2132 | } |
| 2133 | |
| 2134 | static uintptr_t |
| 2135 | teffective_0(task_t task) |
| 2136 | { |
| 2137 | uintptr_t* raw = (uintptr_t*)&task->effective_policy; |
| 2138 | |
| 2139 | return raw[0]; |
| 2140 | } |
| 2141 | |
| 2142 | static uintptr_t |
| 2143 | teffective_1(task_t task) |
| 2144 | { |
| 2145 | #if defined __LP64__ |
| 2146 | (void)task; |
| 2147 | return 0; |
| 2148 | #else |
| 2149 | uintptr_t* raw = (uintptr_t*)(&task->effective_policy); |
| 2150 | return raw[1]; |
| 2151 | #endif |
| 2152 | } |
| 2153 | |
| 2154 | /* dump pending for tracepoint */ |
| 2155 | uint32_t tpending(task_pend_token_t pend_token) { return *(uint32_t*)(void*)(pend_token); } |
| 2156 | |
| 2157 | uint64_t |
| 2158 | task_requested_bitfield(task_t task) |
| 2159 | { |
| 2160 | uint64_t bits = 0; |
| 2161 | struct task_requested_policy requested = task->requested_policy; |
| 2162 | |
| 2163 | bits |= (requested.trp_int_darwinbg ? POLICY_REQ_INT_DARWIN_BG : 0); |
| 2164 | bits |= (requested.trp_ext_darwinbg ? POLICY_REQ_EXT_DARWIN_BG : 0); |
| 2165 | bits |= (requested.trp_int_iotier ? (((uint64_t)requested.trp_int_iotier) << POLICY_REQ_INT_IO_TIER_SHIFT) : 0); |
| 2166 | bits |= (requested.trp_ext_iotier ? (((uint64_t)requested.trp_ext_iotier) << POLICY_REQ_EXT_IO_TIER_SHIFT) : 0); |
| 2167 | bits |= (requested.trp_int_iopassive ? POLICY_REQ_INT_PASSIVE_IO : 0); |
| 2168 | bits |= (requested.trp_ext_iopassive ? POLICY_REQ_EXT_PASSIVE_IO : 0); |
| 2169 | bits |= (requested.trp_bg_iotier ? (((uint64_t)requested.trp_bg_iotier) << POLICY_REQ_BG_IOTIER_SHIFT) : 0); |
| 2170 | bits |= (requested.trp_terminated ? POLICY_REQ_TERMINATED : 0); |
| 2171 | |
| 2172 | bits |= (requested.trp_boosted ? POLICY_REQ_BOOSTED : 0); |
| 2173 | bits |= (requested.trp_tal_enabled ? POLICY_REQ_TAL_ENABLED : 0); |
| 2174 | bits |= (requested.trp_apptype ? (((uint64_t)requested.trp_apptype) << POLICY_REQ_APPTYPE_SHIFT) : 0); |
| 2175 | bits |= (requested.trp_role ? (((uint64_t)requested.trp_role) << POLICY_REQ_ROLE_SHIFT) : 0); |
| 2176 | |
| 2177 | bits |= (requested.trp_sup_active ? POLICY_REQ_SUP_ACTIVE : 0); |
| 2178 | bits |= (requested.trp_sup_lowpri_cpu ? POLICY_REQ_SUP_LOWPRI_CPU : 0); |
| 2179 | bits |= (requested.trp_sup_cpu ? POLICY_REQ_SUP_CPU : 0); |
| 2180 | bits |= (requested.trp_sup_timer ? (((uint64_t)requested.trp_sup_timer) << POLICY_REQ_SUP_TIMER_THROTTLE_SHIFT) : 0); |
| 2181 | bits |= (requested.trp_sup_throughput ? (((uint64_t)requested.trp_sup_throughput) << POLICY_REQ_SUP_THROUGHPUT_SHIFT) : 0); |
| 2182 | bits |= (requested.trp_sup_disk ? POLICY_REQ_SUP_DISK_THROTTLE : 0); |
| 2183 | bits |= (requested.trp_sup_bg_sockets ? POLICY_REQ_SUP_BG_SOCKETS : 0); |
| 2184 | |
| 2185 | bits |= (requested.trp_base_latency_qos ? (((uint64_t)requested.trp_base_latency_qos) << POLICY_REQ_BASE_LATENCY_QOS_SHIFT) : 0); |
| 2186 | bits |= (requested.trp_over_latency_qos ? (((uint64_t)requested.trp_over_latency_qos) << POLICY_REQ_OVER_LATENCY_QOS_SHIFT) : 0); |
| 2187 | bits |= (requested.trp_base_through_qos ? (((uint64_t)requested.trp_base_through_qos) << POLICY_REQ_BASE_THROUGH_QOS_SHIFT) : 0); |
| 2188 | bits |= (requested.trp_over_through_qos ? (((uint64_t)requested.trp_over_through_qos) << POLICY_REQ_OVER_THROUGH_QOS_SHIFT) : 0); |
| 2189 | bits |= (requested.trp_sfi_managed ? POLICY_REQ_SFI_MANAGED : 0); |
| 2190 | bits |= (requested.trp_qos_clamp ? (((uint64_t)requested.trp_qos_clamp) << POLICY_REQ_QOS_CLAMP_SHIFT) : 0); |
| 2191 | |
| 2192 | return bits; |
| 2193 | } |
| 2194 | |
| 2195 | uint64_t |
| 2196 | task_effective_bitfield(task_t task) |
| 2197 | { |
| 2198 | uint64_t bits = 0; |
| 2199 | struct task_effective_policy effective = task->effective_policy; |
| 2200 | |
| 2201 | bits |= (effective.tep_io_tier ? (((uint64_t)effective.tep_io_tier) << POLICY_EFF_IO_TIER_SHIFT) : 0); |
| 2202 | bits |= (effective.tep_io_passive ? POLICY_EFF_IO_PASSIVE : 0); |
| 2203 | bits |= (effective.tep_darwinbg ? POLICY_EFF_DARWIN_BG : 0); |
| 2204 | bits |= (effective.tep_lowpri_cpu ? POLICY_EFF_LOWPRI_CPU : 0); |
| 2205 | bits |= (effective.tep_terminated ? POLICY_EFF_TERMINATED : 0); |
| 2206 | bits |= (effective.tep_all_sockets_bg ? POLICY_EFF_ALL_SOCKETS_BG : 0); |
| 2207 | bits |= (effective.tep_new_sockets_bg ? POLICY_EFF_NEW_SOCKETS_BG : 0); |
| 2208 | bits |= (effective.tep_bg_iotier ? (((uint64_t)effective.tep_bg_iotier) << POLICY_EFF_BG_IOTIER_SHIFT) : 0); |
| 2209 | bits |= (effective.tep_qos_ui_is_urgent ? POLICY_EFF_QOS_UI_IS_URGENT : 0); |
| 2210 | |
| 2211 | bits |= (effective.tep_tal_engaged ? POLICY_EFF_TAL_ENGAGED : 0); |
| 2212 | bits |= (effective.tep_watchers_bg ? POLICY_EFF_WATCHERS_BG : 0); |
| 2213 | bits |= (effective.tep_sup_active ? POLICY_EFF_SUP_ACTIVE : 0); |
| 2214 | bits |= (effective.tep_suppressed_cpu ? POLICY_EFF_SUP_CPU : 0); |
| 2215 | bits |= (effective.tep_role ? (((uint64_t)effective.tep_role) << POLICY_EFF_ROLE_SHIFT) : 0); |
| 2216 | bits |= (effective.tep_latency_qos ? (((uint64_t)effective.tep_latency_qos) << POLICY_EFF_LATENCY_QOS_SHIFT) : 0); |
| 2217 | bits |= (effective.tep_through_qos ? (((uint64_t)effective.tep_through_qos) << POLICY_EFF_THROUGH_QOS_SHIFT) : 0); |
| 2218 | bits |= (effective.tep_sfi_managed ? POLICY_EFF_SFI_MANAGED : 0); |
| 2219 | bits |= (effective.tep_qos_ceiling ? (((uint64_t)effective.tep_qos_ceiling) << POLICY_EFF_QOS_CEILING_SHIFT) : 0); |
| 2220 | |
| 2221 | return bits; |
| 2222 | } |
| 2223 | |
| 2224 | |
| 2225 | /* |
| 2226 | * Resource usage and CPU related routines |
| 2227 | */ |
| 2228 | |
| 2229 | int |
| 2230 | proc_get_task_ruse_cpu(task_t task, uint32_t *policyp, uint8_t *percentagep, uint64_t *intervalp, uint64_t *deadlinep) |
| 2231 | { |
| 2232 | |
| 2233 | int error = 0; |
| 2234 | int scope; |
| 2235 | |
| 2236 | task_lock(task); |
| 2237 | |
| 2238 | |
| 2239 | error = task_get_cpuusage(task, percentagep, intervalp, deadlinep, &scope); |
| 2240 | task_unlock(task); |
| 2241 | |
| 2242 | /* |
| 2243 | * Reverse-map from CPU resource limit scopes back to policies (see comment below). |
| 2244 | */ |
| 2245 | if (scope == TASK_RUSECPU_FLAGS_PERTHR_LIMIT) { |
| 2246 | *policyp = TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC; |
| 2247 | } else if (scope == TASK_RUSECPU_FLAGS_PROC_LIMIT) { |
| 2248 | *policyp = TASK_POLICY_RESOURCE_ATTRIBUTE_THROTTLE; |
| 2249 | } else if (scope == TASK_RUSECPU_FLAGS_DEADLINE) { |
| 2250 | *policyp = TASK_POLICY_RESOURCE_ATTRIBUTE_NONE; |
| 2251 | } |
| 2252 | |
| 2253 | return(error); |
| 2254 | } |
| 2255 | |
| 2256 | /* |
| 2257 | * Configure the default CPU usage monitor parameters. |
| 2258 | * |
| 2259 | * For tasks which have this mechanism activated: if any thread in the |
| 2260 | * process consumes more CPU than this, an EXC_RESOURCE exception will be generated. |
| 2261 | */ |
| 2262 | void |
| 2263 | proc_init_cpumon_params(void) |
| 2264 | { |
| 2265 | /* |
| 2266 | * The max CPU percentage can be configured via the boot-args and |
| 2267 | * a key in the device tree. The boot-args are honored first, then the |
| 2268 | * device tree. |
| 2269 | */ |
| 2270 | if (!PE_parse_boot_argn("max_cpumon_percentage", &proc_max_cpumon_percentage, |
| 2271 | sizeof (proc_max_cpumon_percentage))) |
| 2272 | { |
| 2273 | uint64_t max_percentage = 0ULL; |
| 2274 | |
| 2275 | if (!PE_get_default("kern.max_cpumon_percentage", &max_percentage, |
| 2276 | sizeof(max_percentage))) |
| 2277 | { |
| 2278 | max_percentage = DEFAULT_CPUMON_PERCENTAGE; |
| 2279 | } |
| 2280 | |
| 2281 | assert(max_percentage <= UINT8_MAX); |
| 2282 | proc_max_cpumon_percentage = (uint8_t) max_percentage; |
| 2283 | } |
| 2284 | |
| 2285 | if (proc_max_cpumon_percentage > 100) { |
| 2286 | proc_max_cpumon_percentage = 100; |
| 2287 | } |
| 2288 | |
| 2289 | /* |
| 2290 | * The interval should be specified in seconds. |
| 2291 | * |
| 2292 | * Like the max CPU percentage, the max CPU interval can be configured |
| 2293 | * via boot-args and the device tree. |
| 2294 | */ |
| 2295 | if (!PE_parse_boot_argn("max_cpumon_interval", &proc_max_cpumon_interval, |
| 2296 | sizeof (proc_max_cpumon_interval))) |
| 2297 | { |
| 2298 | if (!PE_get_default("kern.max_cpumon_interval", &proc_max_cpumon_interval, |
| 2299 | sizeof(proc_max_cpumon_interval))) |
| 2300 | { |
| 2301 | proc_max_cpumon_interval = DEFAULT_CPUMON_INTERVAL; |
| 2302 | } |
| 2303 | } |
| 2304 | |
| 2305 | proc_max_cpumon_interval *= NSEC_PER_SEC; |
| 2306 | |
| 2307 | /* TEMPORARY boot arg to control App suppression */ |
| 2308 | PE_parse_boot_argn("task_policy_suppression_disable", |
| 2309 | &task_policy_suppression_disable, |
| 2310 | sizeof(task_policy_suppression_disable)); |
| 2311 | } |
| 2312 | |
| 2313 | /* |
| 2314 | * Currently supported configurations for CPU limits. |
| 2315 | * |
| 2316 | * Policy | Deadline-based CPU limit | Percentage-based CPU limit |
| 2317 | * -------------------------------------+--------------------------+------------------------------ |
| 2318 | * PROC_POLICY_RSRCACT_THROTTLE | ENOTSUP | Task-wide scope only |
| 2319 | * PROC_POLICY_RSRCACT_SUSPEND | Task-wide scope only | ENOTSUP |
| 2320 | * PROC_POLICY_RSRCACT_TERMINATE | Task-wide scope only | ENOTSUP |
| 2321 | * PROC_POLICY_RSRCACT_NOTIFY_KQ | Task-wide scope only | ENOTSUP |
| 2322 | * PROC_POLICY_RSRCACT_NOTIFY_EXC | ENOTSUP | Per-thread scope only |
| 2323 | * |
| 2324 | * A deadline-based CPU limit is actually a simple wallclock timer - the requested action is performed |
| 2325 | * after the specified amount of wallclock time has elapsed. |
| 2326 | * |
| 2327 | * A percentage-based CPU limit performs the requested action after the specified amount of actual CPU time |
| 2328 | * has been consumed -- regardless of how much wallclock time has elapsed -- by either the task as an |
| 2329 | * aggregate entity (so-called "Task-wide" or "Proc-wide" scope, whereby the CPU time consumed by all threads |
| 2330 | * in the task are added together), or by any one thread in the task (so-called "per-thread" scope). |
| 2331 | * |
| 2332 | * We support either deadline != 0 OR percentage != 0, but not both. The original intention in having them |
| 2333 | * share an API was to use actual CPU time as the basis of the deadline-based limit (as in: perform an action |
| 2334 | * after I have used some amount of CPU time; this is different than the recurring percentage/interval model) |
| 2335 | * but the potential consumer of the API at the time was insisting on wallclock time instead. |
| 2336 | * |
| 2337 | * Currently, requesting notification via an exception is the only way to get per-thread scope for a |
| 2338 | * CPU limit. All other types of notifications force task-wide scope for the limit. |
| 2339 | */ |
| 2340 | int |
| 2341 | proc_set_task_ruse_cpu(task_t task, uint32_t policy, uint8_t percentage, uint64_t interval, uint64_t deadline, |
| 2342 | int cpumon_entitled) |
| 2343 | { |
| 2344 | int error = 0; |
| 2345 | int scope; |
| 2346 | |
| 2347 | /* |
| 2348 | * Enforce the matrix of supported configurations for policy, percentage, and deadline. |
| 2349 | */ |
| 2350 | switch (policy) { |
| 2351 | // If no policy is explicitly given, the default is to throttle. |
| 2352 | case TASK_POLICY_RESOURCE_ATTRIBUTE_NONE: |
| 2353 | case TASK_POLICY_RESOURCE_ATTRIBUTE_THROTTLE: |
| 2354 | if (deadline != 0) |
| 2355 | return (ENOTSUP); |
| 2356 | scope = TASK_RUSECPU_FLAGS_PROC_LIMIT; |
| 2357 | break; |
| 2358 | case TASK_POLICY_RESOURCE_ATTRIBUTE_SUSPEND: |
| 2359 | case TASK_POLICY_RESOURCE_ATTRIBUTE_TERMINATE: |
| 2360 | case TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_KQ: |
| 2361 | if (percentage != 0) |
| 2362 | return (ENOTSUP); |
| 2363 | scope = TASK_RUSECPU_FLAGS_DEADLINE; |
| 2364 | break; |
| 2365 | case TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC: |
| 2366 | if (deadline != 0) |
| 2367 | return (ENOTSUP); |
| 2368 | scope = TASK_RUSECPU_FLAGS_PERTHR_LIMIT; |
| 2369 | #ifdef CONFIG_NOMONITORS |
| 2370 | return (error); |
| 2371 | #endif /* CONFIG_NOMONITORS */ |
| 2372 | break; |
| 2373 | default: |
| 2374 | return (EINVAL); |
| 2375 | } |
| 2376 | |
| 2377 | task_lock(task); |
| 2378 | if (task != current_task()) { |
| 2379 | task->policy_ru_cpu_ext = policy; |
| 2380 | } else { |
| 2381 | task->policy_ru_cpu = policy; |
| 2382 | } |
| 2383 | error = task_set_cpuusage(task, percentage, interval, deadline, scope, cpumon_entitled); |
| 2384 | task_unlock(task); |
| 2385 | return(error); |
| 2386 | } |
| 2387 | |
| 2388 | /* TODO: get rid of these */ |
| 2389 | #define TASK_POLICY_CPU_RESOURCE_USAGE 0 |
| 2390 | #define TASK_POLICY_WIREDMEM_RESOURCE_USAGE 1 |
| 2391 | #define TASK_POLICY_VIRTUALMEM_RESOURCE_USAGE 2 |
| 2392 | #define TASK_POLICY_DISK_RESOURCE_USAGE 3 |
| 2393 | #define TASK_POLICY_NETWORK_RESOURCE_USAGE 4 |
| 2394 | #define TASK_POLICY_POWER_RESOURCE_USAGE 5 |
| 2395 | |
| 2396 | #define TASK_POLICY_RESOURCE_USAGE_COUNT 6 |
| 2397 | |
| 2398 | int |
| 2399 | proc_clear_task_ruse_cpu(task_t task, int cpumon_entitled) |
| 2400 | { |
| 2401 | int error = 0; |
| 2402 | int action; |
| 2403 | void * bsdinfo = NULL; |
| 2404 | |
| 2405 | task_lock(task); |
| 2406 | if (task != current_task()) { |
| 2407 | task->policy_ru_cpu_ext = TASK_POLICY_RESOURCE_ATTRIBUTE_DEFAULT; |
| 2408 | } else { |
| 2409 | task->policy_ru_cpu = TASK_POLICY_RESOURCE_ATTRIBUTE_DEFAULT; |
| 2410 | } |
| 2411 | |
| 2412 | error = task_clear_cpuusage_locked(task, cpumon_entitled); |
| 2413 | if (error != 0) |
| 2414 | goto out; |
| 2415 | |
| 2416 | action = task->applied_ru_cpu; |
| 2417 | if (task->applied_ru_cpu_ext != TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) { |
| 2418 | /* reset action */ |
| 2419 | task->applied_ru_cpu_ext = TASK_POLICY_RESOURCE_ATTRIBUTE_NONE; |
| 2420 | } |
| 2421 | if (action != TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) { |
| 2422 | bsdinfo = task->bsd_info; |
| 2423 | task_unlock(task); |
| 2424 | proc_restore_resource_actions(bsdinfo, TASK_POLICY_CPU_RESOURCE_USAGE, action); |
| 2425 | goto out1; |
| 2426 | } |
| 2427 | |
| 2428 | out: |
| 2429 | task_unlock(task); |
| 2430 | out1: |
| 2431 | return(error); |
| 2432 | |
| 2433 | } |
| 2434 | |
| 2435 | /* used to apply resource limit related actions */ |
| 2436 | static int |
| 2437 | task_apply_resource_actions(task_t task, int type) |
| 2438 | { |
| 2439 | int action = TASK_POLICY_RESOURCE_ATTRIBUTE_NONE; |
| 2440 | void * bsdinfo = NULL; |
| 2441 | |
| 2442 | switch (type) { |
| 2443 | case TASK_POLICY_CPU_RESOURCE_USAGE: |
| 2444 | break; |
| 2445 | case TASK_POLICY_WIREDMEM_RESOURCE_USAGE: |
| 2446 | case TASK_POLICY_VIRTUALMEM_RESOURCE_USAGE: |
| 2447 | case TASK_POLICY_DISK_RESOURCE_USAGE: |
| 2448 | case TASK_POLICY_NETWORK_RESOURCE_USAGE: |
| 2449 | case TASK_POLICY_POWER_RESOURCE_USAGE: |
| 2450 | return(0); |
| 2451 | |
| 2452 | default: |
| 2453 | return(1); |
| 2454 | }; |
| 2455 | |
| 2456 | /* only cpu actions for now */ |
| 2457 | task_lock(task); |
| 2458 | |
| 2459 | if (task->applied_ru_cpu_ext == TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) { |
| 2460 | /* apply action */ |
| 2461 | task->applied_ru_cpu_ext = task->policy_ru_cpu_ext; |
| 2462 | action = task->applied_ru_cpu_ext; |
| 2463 | } else { |
| 2464 | action = task->applied_ru_cpu_ext; |
| 2465 | } |
| 2466 | |
| 2467 | if (action != TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) { |
| 2468 | bsdinfo = task->bsd_info; |
| 2469 | task_unlock(task); |
| 2470 | proc_apply_resource_actions(bsdinfo, TASK_POLICY_CPU_RESOURCE_USAGE, action); |
| 2471 | } else |
| 2472 | task_unlock(task); |
| 2473 | |
| 2474 | return(0); |
| 2475 | } |
| 2476 | |
| 2477 | /* |
| 2478 | * XXX This API is somewhat broken; we support multiple simultaneous CPU limits, but the get/set API |
| 2479 | * only allows for one at a time. This means that if there is a per-thread limit active, the other |
| 2480 | * "scopes" will not be accessible via this API. We could change it to pass in the scope of interest |
| 2481 | * to the caller, and prefer that, but there's no need for that at the moment. |
| 2482 | */ |
| 2483 | static int |
| 2484 | task_get_cpuusage(task_t task, uint8_t *percentagep, uint64_t *intervalp, uint64_t *deadlinep, int *scope) |
| 2485 | { |
| 2486 | *percentagep = 0; |
| 2487 | *intervalp = 0; |
| 2488 | *deadlinep = 0; |
| 2489 | |
| 2490 | if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PERTHR_LIMIT) != 0) { |
| 2491 | *scope = TASK_RUSECPU_FLAGS_PERTHR_LIMIT; |
| 2492 | *percentagep = task->rusage_cpu_perthr_percentage; |
| 2493 | *intervalp = task->rusage_cpu_perthr_interval; |
| 2494 | } else if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PROC_LIMIT) != 0) { |
| 2495 | *scope = TASK_RUSECPU_FLAGS_PROC_LIMIT; |
| 2496 | *percentagep = task->rusage_cpu_percentage; |
| 2497 | *intervalp = task->rusage_cpu_interval; |
| 2498 | } else if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_DEADLINE) != 0) { |
| 2499 | *scope = TASK_RUSECPU_FLAGS_DEADLINE; |
| 2500 | *deadlinep = task->rusage_cpu_deadline; |
| 2501 | } else { |
| 2502 | *scope = 0; |
| 2503 | } |
| 2504 | |
| 2505 | return(0); |
| 2506 | } |
| 2507 | |
| 2508 | /* |
| 2509 | * Suspend the CPU usage monitor for the task. Return value indicates |
| 2510 | * if the mechanism was actually enabled. |
| 2511 | */ |
| 2512 | int |
| 2513 | task_suspend_cpumon(task_t task) |
| 2514 | { |
| 2515 | thread_t thread; |
| 2516 | |
| 2517 | task_lock_assert_owned(task); |
| 2518 | |
| 2519 | if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PERTHR_LIMIT) == 0) { |
| 2520 | return KERN_INVALID_ARGUMENT; |
| 2521 | } |
| 2522 | |
| 2523 | #if CONFIG_TELEMETRY |
| 2524 | /* |
| 2525 | * Disable task-wide telemetry if it was ever enabled by the CPU usage |
| 2526 | * monitor's warning zone. |
| 2527 | */ |
| 2528 | telemetry_task_ctl_locked(task, TF_CPUMON_WARNING, 0); |
| 2529 | #endif |
| 2530 | |
| 2531 | /* |
| 2532 | * Suspend monitoring for the task, and propagate that change to each thread. |
| 2533 | */ |
| 2534 | task->rusage_cpu_flags &= ~(TASK_RUSECPU_FLAGS_PERTHR_LIMIT | TASK_RUSECPU_FLAGS_FATAL_CPUMON); |
| 2535 | queue_iterate(&task->threads, thread, thread_t, task_threads) { |
| 2536 | act_set_astledger(thread); |
| 2537 | } |
| 2538 | |
| 2539 | return KERN_SUCCESS; |
| 2540 | } |
| 2541 | |
| 2542 | /* |
| 2543 | * Remove all traces of the CPU monitor. |
| 2544 | */ |
| 2545 | int |
| 2546 | task_disable_cpumon(task_t task) |
| 2547 | { |
| 2548 | int kret; |
| 2549 | |
| 2550 | task_lock_assert_owned(task); |
| 2551 | |
| 2552 | kret = task_suspend_cpumon(task); |
| 2553 | if (kret) return kret; |
| 2554 | |
| 2555 | /* Once we clear these values, the monitor can't be resumed */ |
| 2556 | task->rusage_cpu_perthr_percentage = 0; |
| 2557 | task->rusage_cpu_perthr_interval = 0; |
| 2558 | |
| 2559 | return (KERN_SUCCESS); |
| 2560 | } |
| 2561 | |
| 2562 | |
| 2563 | static int |
| 2564 | task_enable_cpumon_locked(task_t task) |
| 2565 | { |
| 2566 | thread_t thread; |
| 2567 | task_lock_assert_owned(task); |
| 2568 | |
| 2569 | if (task->rusage_cpu_perthr_percentage == 0 || |
| 2570 | task->rusage_cpu_perthr_interval == 0) { |
| 2571 | return KERN_INVALID_ARGUMENT; |
| 2572 | } |
| 2573 | |
| 2574 | task->rusage_cpu_flags |= TASK_RUSECPU_FLAGS_PERTHR_LIMIT; |
| 2575 | queue_iterate(&task->threads, thread, thread_t, task_threads) { |
| 2576 | act_set_astledger(thread); |
| 2577 | } |
| 2578 | |
| 2579 | return KERN_SUCCESS; |
| 2580 | } |
| 2581 | |
| 2582 | int |
| 2583 | task_resume_cpumon(task_t task) |
| 2584 | { |
| 2585 | kern_return_t kret; |
| 2586 | |
| 2587 | if (!task) { |
| 2588 | return EINVAL; |
| 2589 | } |
| 2590 | |
| 2591 | task_lock(task); |
| 2592 | kret = task_enable_cpumon_locked(task); |
| 2593 | task_unlock(task); |
| 2594 | |
| 2595 | return kret; |
| 2596 | } |
| 2597 | |
| 2598 | |
| 2599 | /* duplicate values from bsd/sys/process_policy.h */ |
| 2600 | #define PROC_POLICY_CPUMON_DISABLE 0xFF |
| 2601 | #define PROC_POLICY_CPUMON_DEFAULTS 0xFE |
| 2602 | |
| 2603 | static int |
| 2604 | task_set_cpuusage(task_t task, uint8_t percentage, uint64_t interval, uint64_t deadline, int scope, int cpumon_entitled) |
| 2605 | { |
| 2606 | uint64_t abstime = 0; |
| 2607 | uint64_t limittime = 0; |
| 2608 | |
| 2609 | lck_mtx_assert(&task->lock, LCK_MTX_ASSERT_OWNED); |
| 2610 | |
| 2611 | /* By default, refill once per second */ |
| 2612 | if (interval == 0) |
| 2613 | interval = NSEC_PER_SEC; |
| 2614 | |
| 2615 | if (percentage != 0) { |
| 2616 | if (scope == TASK_RUSECPU_FLAGS_PERTHR_LIMIT) { |
| 2617 | boolean_t warn = FALSE; |
| 2618 | |
| 2619 | /* |
| 2620 | * A per-thread CPU limit on a task generates an exception |
| 2621 | * (LEDGER_ACTION_EXCEPTION) if any one thread in the task |
| 2622 | * exceeds the limit. |
| 2623 | */ |
| 2624 | |
| 2625 | if (percentage == PROC_POLICY_CPUMON_DISABLE) { |
| 2626 | if (cpumon_entitled) { |
| 2627 | /* 25095698 - task_disable_cpumon() should be reliable */ |
| 2628 | task_disable_cpumon(task); |
| 2629 | return 0; |
| 2630 | } |
| 2631 | |
| 2632 | /* |
| 2633 | * This task wishes to disable the CPU usage monitor, but it's |
| 2634 | * missing the required entitlement: |
| 2635 | * com.apple.private.kernel.override-cpumon |
| 2636 | * |
| 2637 | * Instead, treat this as a request to reset its params |
| 2638 | * back to the defaults. |
| 2639 | */ |
| 2640 | warn = TRUE; |
| 2641 | percentage = PROC_POLICY_CPUMON_DEFAULTS; |
| 2642 | } |
| 2643 | |
| 2644 | if (percentage == PROC_POLICY_CPUMON_DEFAULTS) { |
| 2645 | percentage = proc_max_cpumon_percentage; |
| 2646 | interval = proc_max_cpumon_interval; |
| 2647 | } |
| 2648 | |
| 2649 | if (percentage > 100) { |
| 2650 | percentage = 100; |
| 2651 | } |
| 2652 | |
| 2653 | /* |
| 2654 | * Passing in an interval of -1 means either: |
| 2655 | * - Leave the interval as-is, if there's already a per-thread |
| 2656 | * limit configured |
| 2657 | * - Use the system default. |
| 2658 | */ |
| 2659 | if (interval == -1ULL) { |
| 2660 | if (task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PERTHR_LIMIT) { |
| 2661 | interval = task->rusage_cpu_perthr_interval; |
| 2662 | } else { |
| 2663 | interval = proc_max_cpumon_interval; |
| 2664 | } |
| 2665 | } |
| 2666 | |
| 2667 | /* |
| 2668 | * Enforce global caps on CPU usage monitor here if the process is not |
| 2669 | * entitled to escape the global caps. |
| 2670 | */ |
| 2671 | if ((percentage > proc_max_cpumon_percentage) && (cpumon_entitled == 0)) { |
| 2672 | warn = TRUE; |
| 2673 | percentage = proc_max_cpumon_percentage; |
| 2674 | } |
| 2675 | |
| 2676 | if ((interval > proc_max_cpumon_interval) && (cpumon_entitled == 0)) { |
| 2677 | warn = TRUE; |
| 2678 | interval = proc_max_cpumon_interval; |
| 2679 | } |
| 2680 | |
| 2681 | if (warn) { |
| 2682 | int pid = 0; |
| 2683 | const char *procname = "unknown"; |
| 2684 | |
| 2685 | #ifdef MACH_BSD |
| 2686 | pid = proc_selfpid(); |
| 2687 | if (current_task()->bsd_info != NULL) { |
| 2688 | procname = proc_name_address(current_task()->bsd_info); |
| 2689 | } |
| 2690 | #endif |
| 2691 | |
| 2692 | printf("process %s[%d] denied attempt to escape CPU monitor" |
| 2693 | " (missing required entitlement).\n", procname, pid); |
| 2694 | } |
| 2695 | |
| 2696 | /* configure the limit values */ |
| 2697 | task->rusage_cpu_perthr_percentage = percentage; |
| 2698 | task->rusage_cpu_perthr_interval = interval; |
| 2699 | |
| 2700 | /* and enable the CPU monitor */ |
| 2701 | (void)task_enable_cpumon_locked(task); |
| 2702 | } else if (scope == TASK_RUSECPU_FLAGS_PROC_LIMIT) { |
| 2703 | /* |
| 2704 | * Currently, a proc-wide CPU limit always blocks if the limit is |
| 2705 | * exceeded (LEDGER_ACTION_BLOCK). |
| 2706 | */ |
| 2707 | task->rusage_cpu_flags |= TASK_RUSECPU_FLAGS_PROC_LIMIT; |
| 2708 | task->rusage_cpu_percentage = percentage; |
| 2709 | task->rusage_cpu_interval = interval; |
| 2710 | |
| 2711 | limittime = (interval * percentage) / 100; |
| 2712 | nanoseconds_to_absolutetime(limittime, &abstime); |
| 2713 | |
| 2714 | ledger_set_limit(task->ledger, task_ledgers.cpu_time, abstime, 0); |
| 2715 | ledger_set_period(task->ledger, task_ledgers.cpu_time, interval); |
| 2716 | ledger_set_action(task->ledger, task_ledgers.cpu_time, LEDGER_ACTION_BLOCK); |
| 2717 | } |
| 2718 | } |
| 2719 | |
| 2720 | if (deadline != 0) { |
| 2721 | assert(scope == TASK_RUSECPU_FLAGS_DEADLINE); |
| 2722 | |
| 2723 | /* if already in use, cancel and wait for it to cleanout */ |
| 2724 | if (task->rusage_cpu_callt != NULL) { |
| 2725 | task_unlock(task); |
| 2726 | thread_call_cancel_wait(task->rusage_cpu_callt); |
| 2727 | task_lock(task); |
| 2728 | } |
| 2729 | if (task->rusage_cpu_callt == NULL) { |
| 2730 | task->rusage_cpu_callt = thread_call_allocate_with_priority(task_action_cpuusage, (thread_call_param_t)task, THREAD_CALL_PRIORITY_KERNEL); |
| 2731 | } |
| 2732 | /* setup callout */ |
| 2733 | if (task->rusage_cpu_callt != 0) { |
| 2734 | uint64_t save_abstime = 0; |
| 2735 | |
| 2736 | task->rusage_cpu_flags |= TASK_RUSECPU_FLAGS_DEADLINE; |
| 2737 | task->rusage_cpu_deadline = deadline; |
| 2738 | |
| 2739 | nanoseconds_to_absolutetime(deadline, &abstime); |
| 2740 | save_abstime = abstime; |
| 2741 | clock_absolutetime_interval_to_deadline(save_abstime, &abstime); |
| 2742 | thread_call_enter_delayed(task->rusage_cpu_callt, abstime); |
| 2743 | } |
| 2744 | } |
| 2745 | |
| 2746 | return(0); |
| 2747 | } |
| 2748 | |
| 2749 | int |
| 2750 | task_clear_cpuusage(task_t task, int cpumon_entitled) |
| 2751 | { |
| 2752 | int retval = 0; |
| 2753 | |
| 2754 | task_lock(task); |
| 2755 | retval = task_clear_cpuusage_locked(task, cpumon_entitled); |
| 2756 | task_unlock(task); |
| 2757 | |
| 2758 | return(retval); |
| 2759 | } |
| 2760 | |
| 2761 | static int |
| 2762 | task_clear_cpuusage_locked(task_t task, int cpumon_entitled) |
| 2763 | { |
| 2764 | thread_call_t savecallt; |
| 2765 | |
| 2766 | /* cancel percentage handling if set */ |
| 2767 | if (task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PROC_LIMIT) { |
| 2768 | task->rusage_cpu_flags &= ~TASK_RUSECPU_FLAGS_PROC_LIMIT; |
| 2769 | ledger_set_limit(task->ledger, task_ledgers.cpu_time, LEDGER_LIMIT_INFINITY, 0); |
| 2770 | task->rusage_cpu_percentage = 0; |
| 2771 | task->rusage_cpu_interval = 0; |
| 2772 | } |
| 2773 | |
| 2774 | /* |
| 2775 | * Disable the CPU usage monitor. |
| 2776 | */ |
| 2777 | if (cpumon_entitled) { |
| 2778 | task_disable_cpumon(task); |
| 2779 | } |
| 2780 | |
| 2781 | /* cancel deadline handling if set */ |
| 2782 | if (task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_DEADLINE) { |
| 2783 | task->rusage_cpu_flags &= ~TASK_RUSECPU_FLAGS_DEADLINE; |
| 2784 | if (task->rusage_cpu_callt != 0) { |
| 2785 | savecallt = task->rusage_cpu_callt; |
| 2786 | task->rusage_cpu_callt = NULL; |
| 2787 | task->rusage_cpu_deadline = 0; |
| 2788 | task_unlock(task); |
| 2789 | thread_call_cancel_wait(savecallt); |
| 2790 | thread_call_free(savecallt); |
| 2791 | task_lock(task); |
| 2792 | } |
| 2793 | } |
| 2794 | return(0); |
| 2795 | } |
| 2796 | |
| 2797 | /* called by ledger unit to enforce action due to resource usage criteria being met */ |
| 2798 | static void |
| 2799 | task_action_cpuusage(thread_call_param_t param0, __unused thread_call_param_t param1) |
| 2800 | { |
| 2801 | task_t task = (task_t)param0; |
| 2802 | (void)task_apply_resource_actions(task, TASK_POLICY_CPU_RESOURCE_USAGE); |
| 2803 | return; |
| 2804 | } |
| 2805 | |
| 2806 | |
| 2807 | /* |
| 2808 | * Routines for taskwatch and pidbind |
| 2809 | */ |
| 2810 | |
| 2811 | #if CONFIG_EMBEDDED |
| 2812 | |
| 2813 | lck_mtx_t task_watch_mtx; |
| 2814 | |
| 2815 | void |
| 2816 | task_watch_init(void) |
| 2817 | { |
| 2818 | lck_mtx_init(&task_watch_mtx, &task_lck_grp, &task_lck_attr); |
| 2819 | } |
| 2820 | |
| 2821 | static void |
| 2822 | task_watch_lock(void) |
| 2823 | { |
| 2824 | lck_mtx_lock(&task_watch_mtx); |
| 2825 | } |
| 2826 | |
| 2827 | static void |
| 2828 | task_watch_unlock(void) |
| 2829 | { |
| 2830 | lck_mtx_unlock(&task_watch_mtx); |
| 2831 | } |
| 2832 | |
| 2833 | static void |
| 2834 | add_taskwatch_locked(task_t task, task_watch_t * twp) |
| 2835 | { |
| 2836 | queue_enter(&task->task_watchers, twp, task_watch_t *, tw_links); |
| 2837 | task->num_taskwatchers++; |
| 2838 | |
| 2839 | } |
| 2840 | |
| 2841 | static void |
| 2842 | remove_taskwatch_locked(task_t task, task_watch_t * twp) |
| 2843 | { |
| 2844 | queue_remove(&task->task_watchers, twp, task_watch_t *, tw_links); |
| 2845 | task->num_taskwatchers--; |
| 2846 | } |
| 2847 | |
| 2848 | |
| 2849 | int |
| 2850 | proc_lf_pidbind(task_t curtask, uint64_t tid, task_t target_task, int bind) |
| 2851 | { |
| 2852 | thread_t target_thread = NULL; |
| 2853 | int ret = 0, setbg = 0; |
| 2854 | task_watch_t *twp = NULL; |
| 2855 | task_t task = TASK_NULL; |
| 2856 | |
| 2857 | target_thread = task_findtid(curtask, tid); |
| 2858 | if (target_thread == NULL) |
| 2859 | return ESRCH; |
| 2860 | /* holds thread reference */ |
| 2861 | |
| 2862 | if (bind != 0) { |
| 2863 | /* task is still active ? */ |
| 2864 | task_lock(target_task); |
| 2865 | if (target_task->active == 0) { |
| 2866 | task_unlock(target_task); |
| 2867 | ret = ESRCH; |
| 2868 | goto out; |
| 2869 | } |
| 2870 | task_unlock(target_task); |
| 2871 | |
| 2872 | twp = (task_watch_t *)kalloc(sizeof(task_watch_t)); |
| 2873 | if (twp == NULL) { |
| 2874 | task_watch_unlock(); |
| 2875 | ret = ENOMEM; |
| 2876 | goto out; |
| 2877 | } |
| 2878 | |
| 2879 | bzero(twp, sizeof(task_watch_t)); |
| 2880 | |
| 2881 | task_watch_lock(); |
| 2882 | |
| 2883 | if (target_thread->taskwatch != NULL){ |
| 2884 | /* already bound to another task */ |
| 2885 | task_watch_unlock(); |
| 2886 | |
| 2887 | kfree(twp, sizeof(task_watch_t)); |
| 2888 | ret = EBUSY; |
| 2889 | goto out; |
| 2890 | } |
| 2891 | |
| 2892 | task_reference(target_task); |
| 2893 | |
| 2894 | setbg = proc_get_effective_task_policy(target_task, TASK_POLICY_WATCHERS_BG); |
| 2895 | |
| 2896 | twp->tw_task = target_task; /* holds the task reference */ |
| 2897 | twp->tw_thread = target_thread; /* holds the thread reference */ |
| 2898 | twp->tw_state = setbg; |
| 2899 | twp->tw_importance = target_thread->importance; |
| 2900 | |
| 2901 | add_taskwatch_locked(target_task, twp); |
| 2902 | |
| 2903 | target_thread->taskwatch = twp; |
| 2904 | |
| 2905 | task_watch_unlock(); |
| 2906 | |
| 2907 | if (setbg) |
| 2908 | set_thread_appbg(target_thread, setbg, INT_MIN); |
| 2909 | |
| 2910 | /* retain the thread reference as it is in twp */ |
| 2911 | target_thread = NULL; |
| 2912 | } else { |
| 2913 | /* unbind */ |
| 2914 | task_watch_lock(); |
| 2915 | if ((twp = target_thread->taskwatch) != NULL) { |
| 2916 | task = twp->tw_task; |
| 2917 | target_thread->taskwatch = NULL; |
| 2918 | remove_taskwatch_locked(task, twp); |
| 2919 | |
| 2920 | task_watch_unlock(); |
| 2921 | |
| 2922 | task_deallocate(task); /* drop task ref in twp */ |
| 2923 | set_thread_appbg(target_thread, 0, twp->tw_importance); |
| 2924 | thread_deallocate(target_thread); /* drop thread ref in twp */ |
| 2925 | kfree(twp, sizeof(task_watch_t)); |
| 2926 | } else { |
| 2927 | task_watch_unlock(); |
| 2928 | ret = 0; /* return success if it not alredy bound */ |
| 2929 | goto out; |
| 2930 | } |
| 2931 | } |
| 2932 | out: |
| 2933 | thread_deallocate(target_thread); /* drop thread ref acquired in this routine */ |
| 2934 | return(ret); |
| 2935 | } |
| 2936 | |
| 2937 | static void |
| 2938 | set_thread_appbg(thread_t thread, int setbg, __unused int importance) |
| 2939 | { |
| 2940 | int enable = (setbg ? TASK_POLICY_ENABLE : TASK_POLICY_DISABLE); |
| 2941 | |
| 2942 | proc_set_thread_policy(thread, TASK_POLICY_ATTRIBUTE, TASK_POLICY_PIDBIND_BG, enable); |
| 2943 | } |
| 2944 | |
| 2945 | static void |
| 2946 | apply_appstate_watchers(task_t task) |
| 2947 | { |
| 2948 | int numwatchers = 0, i, j, setbg; |
| 2949 | thread_watchlist_t * threadlist; |
| 2950 | task_watch_t * twp; |
| 2951 | |
| 2952 | retry: |
| 2953 | /* if no watchers on the list return */ |
| 2954 | if ((numwatchers = task->num_taskwatchers) == 0) |
| 2955 | return; |
| 2956 | |
| 2957 | threadlist = (thread_watchlist_t *)kalloc(numwatchers*sizeof(thread_watchlist_t)); |
| 2958 | if (threadlist == NULL) |
| 2959 | return; |
| 2960 | |
| 2961 | bzero(threadlist, numwatchers*sizeof(thread_watchlist_t)); |
| 2962 | |
| 2963 | task_watch_lock(); |
| 2964 | /*serialize application of app state changes */ |
| 2965 | |
| 2966 | if (task->watchapplying != 0) { |
| 2967 | lck_mtx_sleep(&task_watch_mtx, LCK_SLEEP_DEFAULT, &task->watchapplying, THREAD_UNINT); |
| 2968 | task_watch_unlock(); |
| 2969 | kfree(threadlist, numwatchers*sizeof(thread_watchlist_t)); |
| 2970 | goto retry; |
| 2971 | } |
| 2972 | |
| 2973 | if (numwatchers != task->num_taskwatchers) { |
| 2974 | task_watch_unlock(); |
| 2975 | kfree(threadlist, numwatchers*sizeof(thread_watchlist_t)); |
| 2976 | goto retry; |
| 2977 | } |
| 2978 | |
| 2979 | setbg = proc_get_effective_task_policy(task, TASK_POLICY_WATCHERS_BG); |
| 2980 | |
| 2981 | task->watchapplying = 1; |
| 2982 | i = 0; |
| 2983 | queue_iterate(&task->task_watchers, twp, task_watch_t *, tw_links) { |
| 2984 | |
| 2985 | threadlist[i].thread = twp->tw_thread; |
| 2986 | thread_reference(threadlist[i].thread); |
| 2987 | if (setbg != 0) { |
| 2988 | twp->tw_importance = twp->tw_thread->importance; |
| 2989 | threadlist[i].importance = INT_MIN; |
| 2990 | } else |
| 2991 | threadlist[i].importance = twp->tw_importance; |
| 2992 | i++; |
| 2993 | if (i > numwatchers) |
| 2994 | break; |
| 2995 | } |
| 2996 | |
| 2997 | task_watch_unlock(); |
| 2998 | |
| 2999 | for (j = 0; j< i; j++) { |
| 3000 | set_thread_appbg(threadlist[j].thread, setbg, threadlist[j].importance); |
| 3001 | thread_deallocate(threadlist[j].thread); |
| 3002 | } |
| 3003 | kfree(threadlist, numwatchers*sizeof(thread_watchlist_t)); |
| 3004 | |
| 3005 | |
| 3006 | task_watch_lock(); |
| 3007 | task->watchapplying = 0; |
| 3008 | thread_wakeup_one(&task->watchapplying); |
| 3009 | task_watch_unlock(); |
| 3010 | } |
| 3011 | |
| 3012 | void |
| 3013 | thead_remove_taskwatch(thread_t thread) |
| 3014 | { |
| 3015 | task_watch_t * twp; |
| 3016 | int importance = 0; |
| 3017 | |
| 3018 | task_watch_lock(); |
| 3019 | if ((twp = thread->taskwatch) != NULL) { |
| 3020 | thread->taskwatch = NULL; |
| 3021 | remove_taskwatch_locked(twp->tw_task, twp); |
| 3022 | } |
| 3023 | task_watch_unlock(); |
| 3024 | if (twp != NULL) { |
| 3025 | thread_deallocate(twp->tw_thread); |
| 3026 | task_deallocate(twp->tw_task); |
| 3027 | importance = twp->tw_importance; |
| 3028 | kfree(twp, sizeof(task_watch_t)); |
| 3029 | /* remove the thread and networkbg */ |
| 3030 | set_thread_appbg(thread, 0, importance); |
| 3031 | } |
| 3032 | } |
| 3033 | |
| 3034 | void |
| 3035 | task_removewatchers(task_t task) |
| 3036 | { |
| 3037 | int numwatchers = 0, i, j; |
| 3038 | task_watch_t ** twplist = NULL; |
| 3039 | task_watch_t * twp = NULL; |
| 3040 | |
| 3041 | retry: |
| 3042 | if ((numwatchers = task->num_taskwatchers) == 0) |
| 3043 | return; |
| 3044 | |
| 3045 | twplist = (task_watch_t **)kalloc(numwatchers*sizeof(task_watch_t *)); |
| 3046 | if (twplist == NULL) |
| 3047 | return; |
| 3048 | |
| 3049 | bzero(twplist, numwatchers*sizeof(task_watch_t *)); |
| 3050 | |
| 3051 | task_watch_lock(); |
| 3052 | if (task->num_taskwatchers == 0) { |
| 3053 | task_watch_unlock(); |
| 3054 | goto out; |
| 3055 | } |
| 3056 | |
| 3057 | if (numwatchers != task->num_taskwatchers) { |
| 3058 | task_watch_unlock(); |
| 3059 | kfree(twplist, numwatchers*sizeof(task_watch_t *)); |
| 3060 | numwatchers = 0; |
| 3061 | goto retry; |
| 3062 | } |
| 3063 | |
| 3064 | i = 0; |
| 3065 | while((twp = (task_watch_t *)dequeue_head(&task->task_watchers)) != NULL) |
| 3066 | { |
| 3067 | twplist[i] = twp; |
| 3068 | task->num_taskwatchers--; |
| 3069 | |
| 3070 | /* |
| 3071 | * Since the linkage is removed and thead state cleanup is already set up, |
| 3072 | * remove the refernce from the thread. |
| 3073 | */ |
| 3074 | twp->tw_thread->taskwatch = NULL; /* removed linkage, clear thread holding ref */ |
| 3075 | i++; |
| 3076 | if ((task->num_taskwatchers == 0) || (i > numwatchers)) |
| 3077 | break; |
| 3078 | } |
| 3079 | |
| 3080 | task_watch_unlock(); |
| 3081 | |
| 3082 | for (j = 0; j< i; j++) { |
| 3083 | |
| 3084 | twp = twplist[j]; |
| 3085 | /* remove thread and network bg */ |
| 3086 | set_thread_appbg(twp->tw_thread, 0, twp->tw_importance); |
| 3087 | thread_deallocate(twp->tw_thread); |
| 3088 | task_deallocate(twp->tw_task); |
| 3089 | kfree(twp, sizeof(task_watch_t)); |
| 3090 | } |
| 3091 | |
| 3092 | out: |
| 3093 | kfree(twplist, numwatchers*sizeof(task_watch_t *)); |
| 3094 | |
| 3095 | } |
| 3096 | #endif /* CONFIG_EMBEDDED */ |
| 3097 | |
| 3098 | /* |
| 3099 | * Routines for importance donation/inheritance/boosting |
| 3100 | */ |
| 3101 | |
| 3102 | static void |
| 3103 | task_importance_update_live_donor(task_t target_task) |
| 3104 | { |
| 3105 | #if IMPORTANCE_INHERITANCE |
| 3106 | |
| 3107 | ipc_importance_task_t task_imp; |
| 3108 | |
| 3109 | task_imp = ipc_importance_for_task(target_task, FALSE); |
| 3110 | if (IIT_NULL != task_imp) { |
| 3111 | ipc_importance_task_update_live_donor(task_imp); |
| 3112 | ipc_importance_task_release(task_imp); |
| 3113 | } |
| 3114 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3115 | } |
| 3116 | |
| 3117 | void |
| 3118 | task_importance_mark_donor(task_t task, boolean_t donating) |
| 3119 | { |
| 3120 | #if IMPORTANCE_INHERITANCE |
| 3121 | ipc_importance_task_t task_imp; |
| 3122 | |
| 3123 | task_imp = ipc_importance_for_task(task, FALSE); |
| 3124 | if (IIT_NULL != task_imp) { |
| 3125 | ipc_importance_task_mark_donor(task_imp, donating); |
| 3126 | ipc_importance_task_release(task_imp); |
| 3127 | } |
| 3128 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3129 | } |
| 3130 | |
| 3131 | void |
| 3132 | task_importance_mark_live_donor(task_t task, boolean_t live_donating) |
| 3133 | { |
| 3134 | #if IMPORTANCE_INHERITANCE |
| 3135 | ipc_importance_task_t task_imp; |
| 3136 | |
| 3137 | task_imp = ipc_importance_for_task(task, FALSE); |
| 3138 | if (IIT_NULL != task_imp) { |
| 3139 | ipc_importance_task_mark_live_donor(task_imp, live_donating); |
| 3140 | ipc_importance_task_release(task_imp); |
| 3141 | } |
| 3142 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3143 | } |
| 3144 | |
| 3145 | void |
| 3146 | task_importance_mark_receiver(task_t task, boolean_t receiving) |
| 3147 | { |
| 3148 | #if IMPORTANCE_INHERITANCE |
| 3149 | ipc_importance_task_t task_imp; |
| 3150 | |
| 3151 | task_imp = ipc_importance_for_task(task, FALSE); |
| 3152 | if (IIT_NULL != task_imp) { |
| 3153 | ipc_importance_task_mark_receiver(task_imp, receiving); |
| 3154 | ipc_importance_task_release(task_imp); |
| 3155 | } |
| 3156 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3157 | } |
| 3158 | |
| 3159 | void |
| 3160 | task_importance_mark_denap_receiver(task_t task, boolean_t denap) |
| 3161 | { |
| 3162 | #if IMPORTANCE_INHERITANCE |
| 3163 | ipc_importance_task_t task_imp; |
| 3164 | |
| 3165 | task_imp = ipc_importance_for_task(task, FALSE); |
| 3166 | if (IIT_NULL != task_imp) { |
| 3167 | ipc_importance_task_mark_denap_receiver(task_imp, denap); |
| 3168 | ipc_importance_task_release(task_imp); |
| 3169 | } |
| 3170 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3171 | } |
| 3172 | |
| 3173 | void |
| 3174 | task_importance_reset(__imp_only task_t task) |
| 3175 | { |
| 3176 | #if IMPORTANCE_INHERITANCE |
| 3177 | ipc_importance_task_t task_imp; |
| 3178 | |
| 3179 | /* TODO: Lower importance downstream before disconnect */ |
| 3180 | task_imp = task->task_imp_base; |
| 3181 | ipc_importance_reset(task_imp, FALSE); |
| 3182 | task_importance_update_live_donor(task); |
| 3183 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3184 | } |
| 3185 | |
| 3186 | #if IMPORTANCE_INHERITANCE |
| 3187 | |
| 3188 | /* |
| 3189 | * Sets the task boost bit to the provided value. Does NOT run the update function. |
| 3190 | * |
| 3191 | * Task lock must be held. |
| 3192 | */ |
| 3193 | static void |
| 3194 | task_set_boost_locked(task_t task, boolean_t boost_active) |
| 3195 | { |
| 3196 | #if IMPORTANCE_TRACE |
| 3197 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (IMPORTANCE_CODE(IMP_BOOST, (boost_active ? IMP_BOOSTED : IMP_UNBOOSTED)) | DBG_FUNC_START), |
| 3198 | proc_selfpid(), task_pid(task), trequested_0(task), trequested_1(task), 0); |
| 3199 | #endif /* IMPORTANCE_TRACE */ |
| 3200 | |
| 3201 | task->requested_policy.trp_boosted = boost_active; |
| 3202 | |
| 3203 | #if IMPORTANCE_TRACE |
| 3204 | if (boost_active == TRUE){ |
| 3205 | DTRACE_BOOST2(boost, task_t, task, int, task_pid(task)); |
| 3206 | } else { |
| 3207 | DTRACE_BOOST2(unboost, task_t, task, int, task_pid(task)); |
| 3208 | } |
| 3209 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (IMPORTANCE_CODE(IMP_BOOST, (boost_active ? IMP_BOOSTED : IMP_UNBOOSTED)) | DBG_FUNC_END), |
| 3210 | proc_selfpid(), task_pid(task), |
| 3211 | trequested_0(task), trequested_1(task), 0); |
| 3212 | #endif /* IMPORTANCE_TRACE */ |
| 3213 | } |
| 3214 | |
| 3215 | /* |
| 3216 | * Sets the task boost bit to the provided value and applies the update. |
| 3217 | * |
| 3218 | * Task lock must be held. Must call update complete after unlocking the task. |
| 3219 | */ |
| 3220 | void |
| 3221 | task_update_boost_locked(task_t task, boolean_t boost_active, task_pend_token_t pend_token) |
| 3222 | { |
| 3223 | task_set_boost_locked(task, boost_active); |
| 3224 | |
| 3225 | task_policy_update_locked(task, pend_token); |
| 3226 | } |
| 3227 | |
| 3228 | /* |
| 3229 | * Check if this task should donate importance. |
| 3230 | * |
| 3231 | * May be called without taking the task lock. In that case, donor status can change |
| 3232 | * so you must check only once for each donation event. |
| 3233 | */ |
| 3234 | boolean_t |
| 3235 | task_is_importance_donor(task_t task) |
| 3236 | { |
| 3237 | if (task->task_imp_base == IIT_NULL) |
| 3238 | return FALSE; |
| 3239 | return ipc_importance_task_is_donor(task->task_imp_base); |
| 3240 | } |
| 3241 | |
| 3242 | /* |
| 3243 | * Query the status of the task's donor mark. |
| 3244 | */ |
| 3245 | boolean_t |
| 3246 | task_is_marked_importance_donor(task_t task) |
| 3247 | { |
| 3248 | if (task->task_imp_base == IIT_NULL) |
| 3249 | return FALSE; |
| 3250 | return ipc_importance_task_is_marked_donor(task->task_imp_base); |
| 3251 | } |
| 3252 | |
| 3253 | /* |
| 3254 | * Query the status of the task's live donor and donor mark. |
| 3255 | */ |
| 3256 | boolean_t |
| 3257 | task_is_marked_live_importance_donor(task_t task) |
| 3258 | { |
| 3259 | if (task->task_imp_base == IIT_NULL) |
| 3260 | return FALSE; |
| 3261 | return ipc_importance_task_is_marked_live_donor(task->task_imp_base); |
| 3262 | } |
| 3263 | |
| 3264 | |
| 3265 | /* |
| 3266 | * This routine may be called without holding task lock |
| 3267 | * since the value of imp_receiver can never be unset. |
| 3268 | */ |
| 3269 | boolean_t |
| 3270 | task_is_importance_receiver(task_t task) |
| 3271 | { |
| 3272 | if (task->task_imp_base == IIT_NULL) |
| 3273 | return FALSE; |
| 3274 | return ipc_importance_task_is_marked_receiver(task->task_imp_base); |
| 3275 | } |
| 3276 | |
| 3277 | /* |
| 3278 | * Query the task's receiver mark. |
| 3279 | */ |
| 3280 | boolean_t |
| 3281 | task_is_marked_importance_receiver(task_t task) |
| 3282 | { |
| 3283 | if (task->task_imp_base == IIT_NULL) |
| 3284 | return FALSE; |
| 3285 | return ipc_importance_task_is_marked_receiver(task->task_imp_base); |
| 3286 | } |
| 3287 | |
| 3288 | /* |
| 3289 | * This routine may be called without holding task lock |
| 3290 | * since the value of de-nap receiver can never be unset. |
| 3291 | */ |
| 3292 | boolean_t |
| 3293 | task_is_importance_denap_receiver(task_t task) |
| 3294 | { |
| 3295 | if (task->task_imp_base == IIT_NULL) |
| 3296 | return FALSE; |
| 3297 | return ipc_importance_task_is_denap_receiver(task->task_imp_base); |
| 3298 | } |
| 3299 | |
| 3300 | /* |
| 3301 | * Query the task's de-nap receiver mark. |
| 3302 | */ |
| 3303 | boolean_t |
| 3304 | task_is_marked_importance_denap_receiver(task_t task) |
| 3305 | { |
| 3306 | if (task->task_imp_base == IIT_NULL) |
| 3307 | return FALSE; |
| 3308 | return ipc_importance_task_is_marked_denap_receiver(task->task_imp_base); |
| 3309 | } |
| 3310 | |
| 3311 | /* |
| 3312 | * This routine may be called without holding task lock |
| 3313 | * since the value of imp_receiver can never be unset. |
| 3314 | */ |
| 3315 | boolean_t |
| 3316 | task_is_importance_receiver_type(task_t task) |
| 3317 | { |
| 3318 | if (task->task_imp_base == IIT_NULL) |
| 3319 | return FALSE; |
| 3320 | return (task_is_importance_receiver(task) || |
| 3321 | task_is_importance_denap_receiver(task)); |
| 3322 | } |
| 3323 | |
| 3324 | /* |
| 3325 | * External importance assertions are managed by the process in userspace |
| 3326 | * Internal importance assertions are the responsibility of the kernel |
| 3327 | * Assertions are changed from internal to external via task_importance_externalize_assertion |
| 3328 | */ |
| 3329 | |
| 3330 | int |
| 3331 | task_importance_hold_internal_assertion(task_t target_task, uint32_t count) |
| 3332 | { |
| 3333 | ipc_importance_task_t task_imp; |
| 3334 | kern_return_t ret; |
| 3335 | |
| 3336 | /* may be first time, so allow for possible importance setup */ |
| 3337 | task_imp = ipc_importance_for_task(target_task, FALSE); |
| 3338 | if (IIT_NULL == task_imp) { |
| 3339 | return EOVERFLOW; |
| 3340 | } |
| 3341 | ret = ipc_importance_task_hold_internal_assertion(task_imp, count); |
| 3342 | ipc_importance_task_release(task_imp); |
| 3343 | |
| 3344 | return (KERN_SUCCESS != ret) ? ENOTSUP : 0; |
| 3345 | } |
| 3346 | |
| 3347 | int |
| 3348 | task_importance_hold_file_lock_assertion(task_t target_task, uint32_t count) |
| 3349 | { |
| 3350 | ipc_importance_task_t task_imp; |
| 3351 | kern_return_t ret; |
| 3352 | |
| 3353 | /* may be first time, so allow for possible importance setup */ |
| 3354 | task_imp = ipc_importance_for_task(target_task, FALSE); |
| 3355 | if (IIT_NULL == task_imp) { |
| 3356 | return EOVERFLOW; |
| 3357 | } |
| 3358 | ret = ipc_importance_task_hold_file_lock_assertion(task_imp, count); |
| 3359 | ipc_importance_task_release(task_imp); |
| 3360 | |
| 3361 | return (KERN_SUCCESS != ret) ? ENOTSUP : 0; |
| 3362 | } |
| 3363 | |
| 3364 | int |
| 3365 | task_importance_hold_legacy_external_assertion(task_t target_task, uint32_t count) |
| 3366 | { |
| 3367 | ipc_importance_task_t task_imp; |
| 3368 | kern_return_t ret; |
| 3369 | |
| 3370 | /* must already have set up an importance */ |
| 3371 | task_imp = target_task->task_imp_base; |
| 3372 | if (IIT_NULL == task_imp) { |
| 3373 | return EOVERFLOW; |
| 3374 | } |
| 3375 | ret = ipc_importance_task_hold_legacy_external_assertion(task_imp, count); |
| 3376 | return (KERN_SUCCESS != ret) ? ENOTSUP : 0; |
| 3377 | } |
| 3378 | |
| 3379 | int |
| 3380 | task_importance_drop_file_lock_assertion(task_t target_task, uint32_t count) |
| 3381 | { |
| 3382 | ipc_importance_task_t task_imp; |
| 3383 | kern_return_t ret; |
| 3384 | |
| 3385 | /* must already have set up an importance */ |
| 3386 | task_imp = target_task->task_imp_base; |
| 3387 | if (IIT_NULL == task_imp) { |
| 3388 | return EOVERFLOW; |
| 3389 | } |
| 3390 | ret = ipc_importance_task_drop_file_lock_assertion(target_task->task_imp_base, count); |
| 3391 | return (KERN_SUCCESS != ret) ? EOVERFLOW : 0; |
| 3392 | } |
| 3393 | |
| 3394 | int |
| 3395 | task_importance_drop_legacy_external_assertion(task_t target_task, uint32_t count) |
| 3396 | { |
| 3397 | ipc_importance_task_t task_imp; |
| 3398 | kern_return_t ret; |
| 3399 | |
| 3400 | /* must already have set up an importance */ |
| 3401 | task_imp = target_task->task_imp_base; |
| 3402 | if (IIT_NULL == task_imp) { |
| 3403 | return EOVERFLOW; |
| 3404 | } |
| 3405 | ret = ipc_importance_task_drop_legacy_external_assertion(task_imp, count); |
| 3406 | return (KERN_SUCCESS != ret) ? EOVERFLOW : 0; |
| 3407 | } |
| 3408 | |
| 3409 | static void |
| 3410 | task_add_importance_watchport(task_t task, mach_port_t port, int *boostp) |
| 3411 | { |
| 3412 | int boost = 0; |
| 3413 | |
| 3414 | __imptrace_only int released_pid = 0; |
| 3415 | __imptrace_only int pid = task_pid(task); |
| 3416 | |
| 3417 | ipc_importance_task_t release_imp_task = IIT_NULL; |
| 3418 | |
| 3419 | if (IP_VALID(port) != 0) { |
| 3420 | ipc_importance_task_t new_imp_task = ipc_importance_for_task(task, FALSE); |
| 3421 | |
| 3422 | ip_lock(port); |
| 3423 | |
| 3424 | /* |
| 3425 | * The port must have been marked tempowner already. |
| 3426 | * This also filters out ports whose receive rights |
| 3427 | * are already enqueued in a message, as you can't |
| 3428 | * change the right's destination once it's already |
| 3429 | * on its way. |
| 3430 | */ |
| 3431 | if (port->ip_tempowner != 0) { |
| 3432 | assert(port->ip_impdonation != 0); |
| 3433 | |
| 3434 | boost = port->ip_impcount; |
| 3435 | if (IIT_NULL != port->ip_imp_task) { |
| 3436 | /* |
| 3437 | * if this port is already bound to a task, |
| 3438 | * release the task reference and drop any |
| 3439 | * watchport-forwarded boosts |
| 3440 | */ |
| 3441 | release_imp_task = port->ip_imp_task; |
| 3442 | port->ip_imp_task = IIT_NULL; |
| 3443 | } |
| 3444 | |
| 3445 | /* mark the port is watching another task (reference held in port->ip_imp_task) */ |
| 3446 | if (ipc_importance_task_is_marked_receiver(new_imp_task)) { |
| 3447 | port->ip_imp_task = new_imp_task; |
| 3448 | new_imp_task = IIT_NULL; |
| 3449 | } |
| 3450 | } |
| 3451 | ip_unlock(port); |
| 3452 | |
| 3453 | if (IIT_NULL != new_imp_task) { |
| 3454 | ipc_importance_task_release(new_imp_task); |
| 3455 | } |
| 3456 | |
| 3457 | if (IIT_NULL != release_imp_task) { |
| 3458 | if (boost > 0) |
| 3459 | ipc_importance_task_drop_internal_assertion(release_imp_task, boost); |
| 3460 | |
| 3461 | // released_pid = task_pid(release_imp_task); /* TODO: Need ref-safe way to get pid */ |
| 3462 | ipc_importance_task_release(release_imp_task); |
| 3463 | } |
| 3464 | #if IMPORTANCE_TRACE |
| 3465 | KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (IMPORTANCE_CODE(IMP_WATCHPORT, 0)) | DBG_FUNC_NONE, |
| 3466 | proc_selfpid(), pid, boost, released_pid, 0); |
| 3467 | #endif /* IMPORTANCE_TRACE */ |
| 3468 | } |
| 3469 | |
| 3470 | *boostp = boost; |
| 3471 | return; |
| 3472 | } |
| 3473 | |
| 3474 | #endif /* IMPORTANCE_INHERITANCE */ |
| 3475 | |
| 3476 | /* |
| 3477 | * Routines for VM to query task importance |
| 3478 | */ |
| 3479 | |
| 3480 | |
| 3481 | /* |
| 3482 | * Order to be considered while estimating importance |
| 3483 | * for low memory notification and purging purgeable memory. |
| 3484 | */ |
| 3485 | #define TASK_IMPORTANCE_FOREGROUND 4 |
| 3486 | #define TASK_IMPORTANCE_NOTDARWINBG 1 |
| 3487 | |
| 3488 | |
| 3489 | /* |
| 3490 | * (Un)Mark the task as a privileged listener for memory notifications. |
| 3491 | * if marked, this task will be among the first to be notified amongst |
| 3492 | * the bulk of all other tasks when the system enters a pressure level |
| 3493 | * of interest to this task. |
| 3494 | */ |
| 3495 | int |
| 3496 | task_low_mem_privileged_listener(task_t task, boolean_t new_value, boolean_t *old_value) |
| 3497 | { |
| 3498 | if (old_value != NULL) { |
| 3499 | *old_value = (boolean_t)task->low_mem_privileged_listener; |
| 3500 | } else { |
| 3501 | task_lock(task); |
| 3502 | task->low_mem_privileged_listener = (uint32_t)new_value; |
| 3503 | task_unlock(task); |
| 3504 | } |
| 3505 | |
| 3506 | return 0; |
| 3507 | } |
| 3508 | |
| 3509 | /* |
| 3510 | * Checks if the task is already notified. |
| 3511 | * |
| 3512 | * Condition: task lock should be held while calling this function. |
| 3513 | */ |
| 3514 | boolean_t |
| 3515 | task_has_been_notified(task_t task, int pressurelevel) |
| 3516 | { |
| 3517 | if (task == NULL) { |
| 3518 | return FALSE; |
| 3519 | } |
| 3520 | |
| 3521 | if (pressurelevel == kVMPressureWarning) |
| 3522 | return (task->low_mem_notified_warn ? TRUE : FALSE); |
| 3523 | else if (pressurelevel == kVMPressureCritical) |
| 3524 | return (task->low_mem_notified_critical ? TRUE : FALSE); |
| 3525 | else |
| 3526 | return TRUE; |
| 3527 | } |
| 3528 | |
| 3529 | |
| 3530 | /* |
| 3531 | * Checks if the task is used for purging. |
| 3532 | * |
| 3533 | * Condition: task lock should be held while calling this function. |
| 3534 | */ |
| 3535 | boolean_t |
| 3536 | task_used_for_purging(task_t task, int pressurelevel) |
| 3537 | { |
| 3538 | if (task == NULL) { |
| 3539 | return FALSE; |
| 3540 | } |
| 3541 | |
| 3542 | if (pressurelevel == kVMPressureWarning) |
| 3543 | return (task->purged_memory_warn ? TRUE : FALSE); |
| 3544 | else if (pressurelevel == kVMPressureCritical) |
| 3545 | return (task->purged_memory_critical ? TRUE : FALSE); |
| 3546 | else |
| 3547 | return TRUE; |
| 3548 | } |
| 3549 | |
| 3550 | |
| 3551 | /* |
| 3552 | * Mark the task as notified with memory notification. |
| 3553 | * |
| 3554 | * Condition: task lock should be held while calling this function. |
| 3555 | */ |
| 3556 | void |
| 3557 | task_mark_has_been_notified(task_t task, int pressurelevel) |
| 3558 | { |
| 3559 | if (task == NULL) { |
| 3560 | return; |
| 3561 | } |
| 3562 | |
| 3563 | if (pressurelevel == kVMPressureWarning) |
| 3564 | task->low_mem_notified_warn = 1; |
| 3565 | else if (pressurelevel == kVMPressureCritical) |
| 3566 | task->low_mem_notified_critical = 1; |
| 3567 | } |
| 3568 | |
| 3569 | |
| 3570 | /* |
| 3571 | * Mark the task as purged. |
| 3572 | * |
| 3573 | * Condition: task lock should be held while calling this function. |
| 3574 | */ |
| 3575 | void |
| 3576 | task_mark_used_for_purging(task_t task, int pressurelevel) |
| 3577 | { |
| 3578 | if (task == NULL) { |
| 3579 | return; |
| 3580 | } |
| 3581 | |
| 3582 | if (pressurelevel == kVMPressureWarning) |
| 3583 | task->purged_memory_warn = 1; |
| 3584 | else if (pressurelevel == kVMPressureCritical) |
| 3585 | task->purged_memory_critical = 1; |
| 3586 | } |
| 3587 | |
| 3588 | |
| 3589 | /* |
| 3590 | * Mark the task eligible for low memory notification. |
| 3591 | * |
| 3592 | * Condition: task lock should be held while calling this function. |
| 3593 | */ |
| 3594 | void |
| 3595 | task_clear_has_been_notified(task_t task, int pressurelevel) |
| 3596 | { |
| 3597 | if (task == NULL) { |
| 3598 | return; |
| 3599 | } |
| 3600 | |
| 3601 | if (pressurelevel == kVMPressureWarning) |
| 3602 | task->low_mem_notified_warn = 0; |
| 3603 | else if (pressurelevel == kVMPressureCritical) |
| 3604 | task->low_mem_notified_critical = 0; |
| 3605 | } |
| 3606 | |
| 3607 | |
| 3608 | /* |
| 3609 | * Mark the task eligible for purging its purgeable memory. |
| 3610 | * |
| 3611 | * Condition: task lock should be held while calling this function. |
| 3612 | */ |
| 3613 | void |
| 3614 | task_clear_used_for_purging(task_t task) |
| 3615 | { |
| 3616 | if (task == NULL) { |
| 3617 | return; |
| 3618 | } |
| 3619 | |
| 3620 | task->purged_memory_warn = 0; |
| 3621 | task->purged_memory_critical = 0; |
| 3622 | } |
| 3623 | |
| 3624 | |
| 3625 | /* |
| 3626 | * Estimate task importance for purging its purgeable memory |
| 3627 | * and low memory notification. |
| 3628 | * |
| 3629 | * Importance is calculated in the following order of criteria: |
| 3630 | * -Task role : Background vs Foreground |
| 3631 | * -Boost status: Not boosted vs Boosted |
| 3632 | * -Darwin BG status. |
| 3633 | * |
| 3634 | * Returns: Estimated task importance. Less important task will have lower |
| 3635 | * estimated importance. |
| 3636 | */ |
| 3637 | int |
| 3638 | task_importance_estimate(task_t task) |
| 3639 | { |
| 3640 | int task_importance = 0; |
| 3641 | |
| 3642 | if (task == NULL) { |
| 3643 | return 0; |
| 3644 | } |
| 3645 | |
| 3646 | if (proc_get_effective_task_policy(task, TASK_POLICY_ROLE) == TASK_FOREGROUND_APPLICATION) |
| 3647 | task_importance += TASK_IMPORTANCE_FOREGROUND; |
| 3648 | |
| 3649 | if (proc_get_effective_task_policy(task, TASK_POLICY_DARWIN_BG) == 0) |
| 3650 | task_importance += TASK_IMPORTANCE_NOTDARWINBG; |
| 3651 | |
| 3652 | return task_importance; |
| 3653 | } |
| 3654 | |
| 3655 | boolean_t |
| 3656 | task_has_assertions(task_t task) |
| 3657 | { |
| 3658 | return (task->task_imp_base->iit_assertcnt? TRUE : FALSE); |
| 3659 | } |
| 3660 | |
| 3661 | |
| 3662 | kern_return_t |
| 3663 | send_resource_violation(typeof(send_cpu_usage_violation) sendfunc, |
| 3664 | task_t violator, |
| 3665 | struct ledger_entry_info *linfo, |
| 3666 | resource_notify_flags_t flags) |
| 3667 | { |
| 3668 | #ifndef MACH_BSD |
| 3669 | return KERN_NOT_SUPPORTED; |
| 3670 | #else |
| 3671 | kern_return_t kr = KERN_SUCCESS; |
| 3672 | proc_t proc = NULL; |
| 3673 | posix_path_t proc_path = ""; |
| 3674 | proc_name_t procname = "<unknown>"; |
| 3675 | int pid = -1; |
| 3676 | clock_sec_t secs; |
| 3677 | clock_nsec_t nsecs; |
| 3678 | mach_timespec_t timestamp; |
| 3679 | thread_t curthread = current_thread(); |
| 3680 | ipc_port_t dstport = MACH_PORT_NULL; |
| 3681 | |
| 3682 | if (!violator) { |
| 3683 | kr = KERN_INVALID_ARGUMENT; goto finish; |
| 3684 | } |
| 3685 | |
| 3686 | /* extract violator information */ |
| 3687 | task_lock(violator); |
| 3688 | if (!(proc = get_bsdtask_info(violator))) { |
| 3689 | task_unlock(violator); |
| 3690 | kr = KERN_INVALID_ARGUMENT; goto finish; |
| 3691 | } |
| 3692 | (void)mig_strncpy(procname, proc_best_name(proc), sizeof(procname)); |
| 3693 | pid = task_pid(violator); |
| 3694 | if (flags & kRNFatalLimitFlag) { |
| 3695 | kr = proc_pidpathinfo_internal(proc, 0, proc_path, |
| 3696 | sizeof(proc_path), NULL); |
| 3697 | } |
| 3698 | task_unlock(violator); |
| 3699 | if (kr) goto finish; |
| 3700 | |
| 3701 | /* violation time ~ now */ |
| 3702 | clock_get_calendar_nanotime(&secs, &nsecs); |
| 3703 | timestamp.tv_sec = (int32_t)secs; |
| 3704 | timestamp.tv_nsec = (int32_t)nsecs; |
| 3705 | /* 25567702 tracks widening mach_timespec_t */ |
| 3706 | |
| 3707 | /* send message */ |
| 3708 | kr = host_get_special_port(host_priv_self(), HOST_LOCAL_NODE, |
| 3709 | HOST_RESOURCE_NOTIFY_PORT, &dstport); |
| 3710 | if (kr) goto finish; |
| 3711 | |
| 3712 | thread_set_honor_qlimit(curthread); |
| 3713 | kr = sendfunc(dstport, |
| 3714 | procname, pid, proc_path, timestamp, |
| 3715 | linfo->lei_balance, linfo->lei_last_refill, |
| 3716 | linfo->lei_limit, linfo->lei_refill_period, |
| 3717 | flags); |
| 3718 | thread_clear_honor_qlimit(curthread); |
| 3719 | |
| 3720 | ipc_port_release_send(dstport); |
| 3721 | |
| 3722 | finish: |
| 3723 | return kr; |
| 3724 | #endif /* MACH_BSD */ |
| 3725 | } |
| 3726 | |
| 3727 | |
| 3728 | /* |
| 3729 | * Resource violations trace four 64-bit integers. For K32, two additional |
| 3730 | * codes are allocated, the first with the low nibble doubled. So if the K64 |
| 3731 | * code is 0x042, the K32 codes would be 0x044 and 0x45. |
| 3732 | */ |
| 3733 | #ifdef __LP64__ |
| 3734 | void |
| 3735 | trace_resource_violation(uint16_t code, |
| 3736 | struct ledger_entry_info *linfo) |
| 3737 | { |
| 3738 | KERNEL_DBG_IST_SANE(KDBG_CODE(DBG_MACH, DBG_MACH_RESOURCE, code), |
| 3739 | linfo->lei_balance, linfo->lei_last_refill, |
| 3740 | linfo->lei_limit, linfo->lei_refill_period); |
| 3741 | } |
| 3742 | #else /* K32 */ |
| 3743 | /* TODO: create/find a trace_two_LLs() for K32 systems */ |
| 3744 | #define MASK32 0xffffffff |
| 3745 | void |
| 3746 | trace_resource_violation(uint16_t code, |
| 3747 | struct ledger_entry_info *linfo) |
| 3748 | { |
| 3749 | int8_t lownibble = (code & 0x3) * 2; |
| 3750 | int16_t codeA = (code & 0xffc) | lownibble; |
| 3751 | int16_t codeB = codeA + 1; |
| 3752 | |
| 3753 | int32_t balance_high = (linfo->lei_balance >> 32) & MASK32; |
| 3754 | int32_t balance_low = linfo->lei_balance & MASK32; |
| 3755 | int32_t last_refill_high = (linfo->lei_last_refill >> 32) & MASK32; |
| 3756 | int32_t last_refill_low = linfo->lei_last_refill & MASK32; |
| 3757 | |
| 3758 | int32_t limit_high = (linfo->lei_limit >> 32) & MASK32; |
| 3759 | int32_t limit_low = linfo->lei_limit & MASK32; |
| 3760 | int32_t refill_period_high = (linfo->lei_refill_period >> 32) & MASK32; |
| 3761 | int32_t refill_period_low = linfo->lei_refill_period & MASK32; |
| 3762 | |
| 3763 | KERNEL_DBG_IST_SANE(KDBG_CODE(DBG_MACH, DBG_MACH_RESOURCE, codeA), |
| 3764 | balance_high, balance_low, |
| 3765 | last_refill_high, last_refill_low); |
| 3766 | KERNEL_DBG_IST_SANE(KDBG_CODE(DBG_MACH, DBG_MACH_RESOURCE, codeB), |
| 3767 | limit_high, limit_low, |
| 3768 | refill_period_high, refill_period_low); |
| 3769 | } |
| 3770 | #endif /* K64/K32 */ |