]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/kern_memorystatus.c
projects / apple / xnu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xnu-7195.101.1.tar.gz
[apple/xnu.git] / bsd / kern / kern_memorystatus.c
1 /*
2 * Copyright (c) 2006-2019 Apple 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
30 #include <kern/sched_prim.h>
31 #include <kern/kalloc.h>
32 #include <kern/assert.h>
33 #include <kern/debug.h>
34 #include <kern/locks.h>
35 #include <kern/task.h>
36 #include <kern/thread.h>
37 #include <kern/host.h>
38 #include <kern/policy_internal.h>
39 #include <kern/thread_group.h>
40
41 #include <corpses/task_corpse.h>
42 #include <libkern/libkern.h>
43 #include <mach/coalition.h>
44 #include <mach/mach_time.h>
45 #include <mach/task.h>
46 #include <mach/host_priv.h>
47 #include <mach/mach_host.h>
48 #include <os/log.h>
49 #include <pexpert/pexpert.h>
50 #include <sys/coalition.h>
51 #include <sys/kern_event.h>
52 #include <sys/proc.h>
53 #include <sys/proc_info.h>
54 #include <sys/reason.h>
55 #include <sys/signal.h>
56 #include <sys/signalvar.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysproto.h>
59 #include <sys/wait.h>
60 #include <sys/tree.h>
61 #include <sys/priv.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_pageout.h>
64 #include <vm/vm_protos.h>
65 #include <mach/machine/sdt.h>
66 #include <libkern/section_keywords.h>
67 #include <stdatomic.h>
68
69 #include <IOKit/IOBSD.h>
70
71 #if CONFIG_FREEZE
72 #include <vm/vm_map.h>
73 #endif /* CONFIG_FREEZE */
74
75 #include <sys/kern_memorystatus.h>
76 #include <sys/kern_memorystatus_freeze.h>
77 #include <sys/kern_memorystatus_notify.h>
78
79 /* For logging clarity */
80 static const char *memorystatus_kill_cause_name[] = {
81 "", /* kMemorystatusInvalid */
82 "jettisoned", /* kMemorystatusKilled */
83 "highwater", /* kMemorystatusKilledHiwat */
84 "vnode-limit", /* kMemorystatusKilledVnodes */
85 "vm-pageshortage", /* kMemorystatusKilledVMPageShortage */
86 "proc-thrashing", /* kMemorystatusKilledProcThrashing */
87 "fc-thrashing", /* kMemorystatusKilledFCThrashing */
88 "per-process-limit", /* kMemorystatusKilledPerProcessLimit */
89 "disk-space-shortage", /* kMemorystatusKilledDiskSpaceShortage */
90 "idle-exit", /* kMemorystatusKilledIdleExit */
91 "zone-map-exhaustion", /* kMemorystatusKilledZoneMapExhaustion */
92 "vm-compressor-thrashing", /* kMemorystatusKilledVMCompressorThrashing */
93 "vm-compressor-space-shortage", /* kMemorystatusKilledVMCompressorSpaceShortage */
94 };
95
96 static const char *
97 memorystatus_priority_band_name(int32_t priority)
98 {
99 switch (priority) {
100 case JETSAM_PRIORITY_FOREGROUND:
101 return "FOREGROUND";
102 case JETSAM_PRIORITY_AUDIO_AND_ACCESSORY:
103 return "AUDIO_AND_ACCESSORY";
104 case JETSAM_PRIORITY_CONDUCTOR:
105 return "CONDUCTOR";
106 case JETSAM_PRIORITY_DRIVER_APPLE:
107 return "DRIVER_APPLE";
108 case JETSAM_PRIORITY_HOME:
109 return "HOME";
110 case JETSAM_PRIORITY_EXECUTIVE:
111 return "EXECUTIVE";
112 case JETSAM_PRIORITY_IMPORTANT:
113 return "IMPORTANT";
114 case JETSAM_PRIORITY_CRITICAL:
115 return "CRITICAL";
116 }
117
118 return "?";
119 }
120
121 /* Does cause indicate vm or fc thrashing? */
122 static boolean_t
123 is_reason_thrashing(unsigned cause)
124 {
125 switch (cause) {
126 case kMemorystatusKilledFCThrashing:
127 case kMemorystatusKilledVMCompressorThrashing:
128 case kMemorystatusKilledVMCompressorSpaceShortage:
129 return TRUE;
130 default:
131 return FALSE;
132 }
133 }
134
135 /* Is the zone map almost full? */
136 static boolean_t
137 is_reason_zone_map_exhaustion(unsigned cause)
138 {
139 if (cause == kMemorystatusKilledZoneMapExhaustion) {
140 return TRUE;
141 }
142 return FALSE;
143 }
144
145 /*
146 * Returns the current zone map size and capacity to include in the jetsam snapshot.
147 * Defined in zalloc.c
148 */
149 extern void get_zone_map_size(uint64_t *current_size, uint64_t *capacity);
150
151 /*
152 * Returns the name of the largest zone and its size to include in the jetsam snapshot.
153 * Defined in zalloc.c
154 */
155 extern void get_largest_zone_info(char *zone_name, size_t zone_name_len, uint64_t *zone_size);
156
157 /*
158 * Active / Inactive limit support
159 * proc list must be locked
160 *
161 * The SET_*** macros are used to initialize a limit
162 * for the first time.
163 *
164 * The CACHE_*** macros are use to cache the limit that will
165 * soon be in effect down in the ledgers.
166 */
167
168 #define SET_ACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
169 MACRO_BEGIN \
170 (p)->p_memstat_memlimit_active = (limit); \
171 if (is_fatal) { \
172 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
173 } else { \
174 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
175 } \
176 MACRO_END
177
178 #define SET_INACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
179 MACRO_BEGIN \
180 (p)->p_memstat_memlimit_inactive = (limit); \
181 if (is_fatal) { \
182 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
183 } else { \
184 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
185 } \
186 MACRO_END
187
188 #define CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal) \
189 MACRO_BEGIN \
190 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_active; \
191 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) { \
192 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
193 is_fatal = TRUE; \
194 } else { \
195 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
196 is_fatal = FALSE; \
197 } \
198 MACRO_END
199
200 #define CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal) \
201 MACRO_BEGIN \
202 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_inactive; \
203 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) { \
204 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
205 is_fatal = TRUE; \
206 } else { \
207 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
208 is_fatal = FALSE; \
209 } \
210 MACRO_END
211
212
213 /* General tunables */
214
215 unsigned long delta_percentage = 5;
216 unsigned long critical_threshold_percentage = 5;
217 // On embedded devices with more than 3GB of memory we lower the critical percentage.
218 uint64_t config_jetsam_large_memory_cutoff = 3UL * (1UL << 30);
219 unsigned long critical_threshold_percentage_larger_devices = 4;
220 unsigned long delta_percentage_larger_devices = 4;
221 unsigned long idle_offset_percentage = 5;
222 unsigned long pressure_threshold_percentage = 15;
223 unsigned long policy_more_free_offset_percentage = 5;
224 unsigned long sysproc_aging_aggr_threshold_percentage = 7;
225
226 /*
227 * default jetsam snapshot support
228 */
229 memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot;
230 memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot_copy;
231
232 #if CONFIG_FREEZE
233 memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot_freezer;
234 /*
235 * The size of the freezer snapshot is given by memorystatus_jetsam_snapshot_max / JETSAM_SNAPSHOT_FREEZER_MAX_FACTOR
236 * The freezer snapshot can be much smaller than the default snapshot
237 * because it only includes apps that have been killed and dasd consumes it every 30 minutes.
238 * Since the snapshots are always wired we don't want to overallocate too much.
239 */
240 #define JETSAM_SNAPSHOT_FREEZER_MAX_FACTOR 20
241 unsigned int memorystatus_jetsam_snapshot_freezer_max;
242 unsigned int memorystatus_jetsam_snapshot_freezer_size;
243 TUNABLE(bool, memorystatus_jetsam_use_freezer_snapshot, "kern.jetsam_user_freezer_snapshot", true);
244 #endif /* CONFIG_FREEZE */
245
246 unsigned int memorystatus_jetsam_snapshot_count = 0;
247 unsigned int memorystatus_jetsam_snapshot_copy_count = 0;
248 unsigned int memorystatus_jetsam_snapshot_max = 0;
249 unsigned int memorystatus_jetsam_snapshot_size = 0;
250 uint64_t memorystatus_jetsam_snapshot_last_timestamp = 0;
251 uint64_t memorystatus_jetsam_snapshot_timeout = 0;
252
253 #if DEVELOPMENT || DEBUG
254 /*
255 * On development and debug kernels, we allow one pid to take ownership
256 * of some memorystatus data structures for testing purposes (via memorystatus_control).
257 * If there's an owner, then only they may consume the jetsam snapshot & set freezer probabilities.
258 * This is used when testing these interface to avoid racing with other
259 * processes on the system that typically use them (namely OSAnalytics & dasd).
260 */
261 static pid_t memorystatus_testing_pid = 0;
262 SYSCTL_INT(_kern, OID_AUTO, memorystatus_testing_pid, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_testing_pid, 0, "");
263 #endif /* DEVELOPMENT || DEBUG */
264 static void memorystatus_init_jetsam_snapshot_header(memorystatus_jetsam_snapshot_t *snapshot);
265
266 /* General memorystatus stuff */
267
268 uint64_t memorystatus_sysprocs_idle_delay_time = 0;
269 uint64_t memorystatus_apps_idle_delay_time = 0;
270 /* Some devices give entitled apps a higher memory limit */
271 #if __arm64__
272 int32_t memorystatus_entitled_max_task_footprint_mb = 0;
273
274 #if DEVELOPMENT || DEBUG
275 SYSCTL_INT(_kern, OID_AUTO, entitled_max_task_pmem, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_entitled_max_task_footprint_mb, 0, "");
276 #endif /* DEVELOPMENT || DEBUG */
277 #endif /* __arm64__ */
278
279 static LCK_GRP_DECLARE(memorystatus_jetsam_fg_band_lock_grp,
280 "memorystatus_jetsam_fg_band");
281 LCK_MTX_DECLARE(memorystatus_jetsam_fg_band_lock,
282 &memorystatus_jetsam_fg_band_lock_grp);
283
284 /* Idle guard handling */
285
286 static int32_t memorystatus_scheduled_idle_demotions_sysprocs = 0;
287 static int32_t memorystatus_scheduled_idle_demotions_apps = 0;
288
289 static void memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2);
290 static void memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state);
291 static void memorystatus_reschedule_idle_demotion_locked(void);
292 int memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap);
293 vm_pressure_level_t convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
294 boolean_t is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
295 void memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear);
296 void memorystatus_send_low_swap_note(void);
297 int memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index);
298 boolean_t memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, unsigned int band, int aggr_count,
299 uint32_t *errors, uint64_t *memory_reclaimed);
300 uint64_t memorystatus_available_memory_internal(proc_t p);
301
302 unsigned int memorystatus_level = 0;
303 static int memorystatus_list_count = 0;
304 memstat_bucket_t memstat_bucket[MEMSTAT_BUCKET_COUNT];
305 static thread_call_t memorystatus_idle_demotion_call;
306 uint64_t memstat_idle_demotion_deadline = 0;
307 int system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
308 int applications_aging_band = JETSAM_PRIORITY_IDLE;
309
310 #define isProcessInAgingBands(p) ((isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) || (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)))
311
312 #define kJetsamAgingPolicyNone (0)
313 #define kJetsamAgingPolicyLegacy (1)
314 #define kJetsamAgingPolicySysProcsReclaimedFirst (2)
315 #define kJetsamAgingPolicyAppsReclaimedFirst (3)
316 #define kJetsamAgingPolicyMax kJetsamAgingPolicyAppsReclaimedFirst
317
318 unsigned int jetsam_aging_policy = kJetsamAgingPolicySysProcsReclaimedFirst;
319
320 extern int corpse_for_fatal_memkill;
321 extern uint64_t vm_purgeable_purge_task_owned(task_t task);
322 boolean_t memorystatus_allowed_vm_map_fork(task_t);
323 #if DEVELOPMENT || DEBUG
324 void memorystatus_abort_vm_map_fork(task_t);
325 #endif
326
327 /*
328 * Idle delay timeout factors for daemons based on relaunch behavior. Only used in
329 * kJetsamAgingPolicySysProcsReclaimedFirst aging policy.
330 */
331 #define kJetsamSysProcsIdleDelayTimeLowRatio (5)
332 #define kJetsamSysProcsIdleDelayTimeMedRatio (2)
333 #define kJetsamSysProcsIdleDelayTimeHighRatio (1)
334 static_assert(kJetsamSysProcsIdleDelayTimeLowRatio <= DEFERRED_IDLE_EXIT_TIME_SECS, "sysproc idle delay time for low relaunch daemons would be 0");
335
336 /*
337 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, treat apps as well
338 * behaved daemons for aging purposes.
339 */
340 #define kJetsamAppsIdleDelayTimeRatio (kJetsamSysProcsIdleDelayTimeLowRatio)
341
342 static uint64_t
343 memorystatus_sysprocs_idle_time(proc_t p)
344 {
345 /*
346 * The kJetsamAgingPolicySysProcsReclaimedFirst aging policy uses the relaunch behavior to
347 * determine the exact idle deferred time provided to the daemons. For all other aging
348 * policies, simply return the default aging idle time.
349 */
350 if (jetsam_aging_policy != kJetsamAgingPolicySysProcsReclaimedFirst) {
351 return memorystatus_sysprocs_idle_delay_time;
352 }
353
354 uint64_t idle_delay_time = 0;
355 /*
356 * For system processes, base the idle delay time on the
357 * jetsam relaunch behavior specified by launchd. The idea
358 * is to provide extra protection to the daemons which would
359 * relaunch immediately after jetsam.
360 */
361 switch (p->p_memstat_relaunch_flags) {
362 case P_MEMSTAT_RELAUNCH_UNKNOWN:
363 case P_MEMSTAT_RELAUNCH_LOW:
364 idle_delay_time = memorystatus_sysprocs_idle_delay_time / kJetsamSysProcsIdleDelayTimeLowRatio;
365 break;
366 case P_MEMSTAT_RELAUNCH_MED:
367 idle_delay_time = memorystatus_sysprocs_idle_delay_time / kJetsamSysProcsIdleDelayTimeMedRatio;
368 break;
369 case P_MEMSTAT_RELAUNCH_HIGH:
370 idle_delay_time = memorystatus_sysprocs_idle_delay_time / kJetsamSysProcsIdleDelayTimeHighRatio;
371 break;
372 default:
373 panic("Unknown relaunch flags on process!");
374 break;
375 }
376 return idle_delay_time;
377 }
378
379 static uint64_t
380 memorystatus_apps_idle_time(__unused proc_t p)
381 {
382 /*
383 * For kJetsamAgingPolicySysProcsReclaimedFirst, the Apps are considered as low
384 * relaunch candidates. So only provide limited protection to them. In the other
385 * aging policies, return the default aging idle time.
386 */
387 if (jetsam_aging_policy != kJetsamAgingPolicySysProcsReclaimedFirst) {
388 return memorystatus_apps_idle_delay_time;
389 }
390
391 return memorystatus_apps_idle_delay_time / kJetsamAppsIdleDelayTimeRatio;
392 }
393
394
395 #if 0
396
397 /* Keeping around for future use if we need a utility that can do this OR an app that needs a dynamic adjustment. */
398
399 static int
400 sysctl_set_jetsam_aging_policy SYSCTL_HANDLER_ARGS
401 {
402 #pragma unused(oidp, arg1, arg2)
403
404 int error = 0, val = 0;
405 memstat_bucket_t *old_bucket = 0;
406 int old_system_procs_aging_band = 0, new_system_procs_aging_band = 0;
407 int old_applications_aging_band = 0, new_applications_aging_band = 0;
408 proc_t p = NULL, next_proc = NULL;
409
410
411 error = sysctl_io_number(req, jetsam_aging_policy, sizeof(int), &val, NULL);
412 if (error || !req->newptr) {
413 return error;
414 }
415
416 if ((val < 0) || (val > kJetsamAgingPolicyMax)) {
417 printf("jetsam: ordering policy sysctl has invalid value - %d\n", val);
418 return EINVAL;
419 }
420
421 /*
422 * We need to synchronize with any potential adding/removal from aging bands
423 * that might be in progress currently. We use the proc_list_lock() just for
424 * consistency with all the routines dealing with 'aging' processes. We need
425 * a lighterweight lock.
426 */
427 proc_list_lock();
428
429 old_system_procs_aging_band = system_procs_aging_band;
430 old_applications_aging_band = applications_aging_band;
431
432 switch (val) {
433 case kJetsamAgingPolicyNone:
434 new_system_procs_aging_band = JETSAM_PRIORITY_IDLE;
435 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
436 break;
437
438 case kJetsamAgingPolicyLegacy:
439 /*
440 * Legacy behavior where some daemons get a 10s protection once and only before the first clean->dirty->clean transition before going into IDLE band.
441 */
442 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
443 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
444 break;
445
446 case kJetsamAgingPolicySysProcsReclaimedFirst:
447 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
448 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
449 break;
450
451 case kJetsamAgingPolicyAppsReclaimedFirst:
452 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
453 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
454 break;
455
456 default:
457 break;
458 }
459
460 if (old_system_procs_aging_band && (old_system_procs_aging_band != new_system_procs_aging_band)) {
461 old_bucket = &memstat_bucket[old_system_procs_aging_band];
462 p = TAILQ_FIRST(&old_bucket->list);
463
464 while (p) {
465 next_proc = TAILQ_NEXT(p, p_memstat_list);
466
467 if (isSysProc(p)) {
468 if (new_system_procs_aging_band == JETSAM_PRIORITY_IDLE) {
469 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
470 }
471
472 memorystatus_update_priority_locked(p, new_system_procs_aging_band, false, true);
473 }
474
475 p = next_proc;
476 continue;
477 }
478 }
479
480 if (old_applications_aging_band && (old_applications_aging_band != new_applications_aging_band)) {
481 old_bucket = &memstat_bucket[old_applications_aging_band];
482 p = TAILQ_FIRST(&old_bucket->list);
483
484 while (p) {
485 next_proc = TAILQ_NEXT(p, p_memstat_list);
486
487 if (isApp(p)) {
488 if (new_applications_aging_band == JETSAM_PRIORITY_IDLE) {
489 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
490 }
491
492 memorystatus_update_priority_locked(p, new_applications_aging_band, false, true);
493 }
494
495 p = next_proc;
496 continue;
497 }
498 }
499
500 jetsam_aging_policy = val;
501 system_procs_aging_band = new_system_procs_aging_band;
502 applications_aging_band = new_applications_aging_band;
503
504 proc_list_unlock();
505
506 return 0;
507 }
508
509 SYSCTL_PROC(_kern, OID_AUTO, set_jetsam_aging_policy, CTLTYPE_INT | CTLFLAG_RW,
510 0, 0, sysctl_set_jetsam_aging_policy, "I", "Jetsam Aging Policy");
511 #endif /*0*/
512
513 static int
514 sysctl_jetsam_set_sysprocs_idle_delay_time SYSCTL_HANDLER_ARGS
515 {
516 #pragma unused(oidp, arg1, arg2)
517
518 int error = 0, val = 0, old_time_in_secs = 0;
519 uint64_t old_time_in_ns = 0;
520
521 absolutetime_to_nanoseconds(memorystatus_sysprocs_idle_delay_time, &old_time_in_ns);
522 old_time_in_secs = (int) (old_time_in_ns / NSEC_PER_SEC);
523
524 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
525 if (error || !req->newptr) {
526 return error;
527 }
528
529 if ((val < 0) || (val > INT32_MAX)) {
530 printf("jetsam: new idle delay interval has invalid value.\n");
531 return EINVAL;
532 }
533
534 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
535
536 return 0;
537 }
538
539 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_sysprocs_idle_delay_time, CTLTYPE_INT | CTLFLAG_RW,
540 0, 0, sysctl_jetsam_set_sysprocs_idle_delay_time, "I", "Aging window for system processes");
541
542
543 static int
544 sysctl_jetsam_set_apps_idle_delay_time SYSCTL_HANDLER_ARGS
545 {
546 #pragma unused(oidp, arg1, arg2)
547
548 int error = 0, val = 0, old_time_in_secs = 0;
549 uint64_t old_time_in_ns = 0;
550
551 absolutetime_to_nanoseconds(memorystatus_apps_idle_delay_time, &old_time_in_ns);
552 old_time_in_secs = (int) (old_time_in_ns / NSEC_PER_SEC);
553
554 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
555 if (error || !req->newptr) {
556 return error;
557 }
558
559 if ((val < 0) || (val > INT32_MAX)) {
560 printf("jetsam: new idle delay interval has invalid value.\n");
561 return EINVAL;
562 }
563
564 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
565
566 return 0;
567 }
568
569 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_apps_idle_delay_time, CTLTYPE_INT | CTLFLAG_RW,
570 0, 0, sysctl_jetsam_set_apps_idle_delay_time, "I", "Aging window for applications");
571
572 SYSCTL_INT(_kern, OID_AUTO, jetsam_aging_policy, CTLTYPE_INT | CTLFLAG_RD, &jetsam_aging_policy, 0, "");
573
574 static unsigned int memorystatus_dirty_count = 0;
575
576 SYSCTL_INT(_kern, OID_AUTO, max_task_pmem, CTLFLAG_RD | CTLFLAG_LOCKED | CTLFLAG_MASKED, &max_task_footprint_mb, 0, "");
577
578 static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. */
579 static boolean_t proc_jetsam_state_is_active_locked(proc_t);
580
581 #if __arm64__
582 int legacy_footprint_bonus_mb = 50; /* This value was chosen after looking at the top 30 apps
583 * that needed the additional room in their footprint when
584 * the 'correct' accounting methods were applied to them.
585 */
586
587 #if DEVELOPMENT || DEBUG
588 SYSCTL_INT(_kern, OID_AUTO, legacy_footprint_bonus_mb, CTLFLAG_RW | CTLFLAG_LOCKED, &legacy_footprint_bonus_mb, 0, "");
589 #endif /* DEVELOPMENT || DEBUG */
590 /*
591 * Raise the inactive and active memory limits to new values.
592 * Will only raise the limits and will do nothing if either of the current
593 * limits are 0.
594 * Caller must hold the proc_list_lock
595 */
596 static void
597 memorystatus_raise_memlimit(proc_t p, int new_memlimit_active, int new_memlimit_inactive)
598 {
599 int memlimit_mb_active = 0, memlimit_mb_inactive = 0;
600 boolean_t memlimit_active_is_fatal = FALSE, memlimit_inactive_is_fatal = FALSE, use_active_limit = FALSE;
601
602 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
603
604 if (p->p_memstat_memlimit_active > 0) {
605 memlimit_mb_active = p->p_memstat_memlimit_active;
606 } else if (p->p_memstat_memlimit_active == -1) {
607 memlimit_mb_active = max_task_footprint_mb;
608 } else {
609 /*
610 * Nothing to do for '0' which is
611 * a special value only used internally
612 * to test 'no limits'.
613 */
614 return;
615 }
616
617 if (p->p_memstat_memlimit_inactive > 0) {
618 memlimit_mb_inactive = p->p_memstat_memlimit_inactive;
619 } else if (p->p_memstat_memlimit_inactive == -1) {
620 memlimit_mb_inactive = max_task_footprint_mb;
621 } else {
622 /*
623 * Nothing to do for '0' which is
624 * a special value only used internally
625 * to test 'no limits'.
626 */
627 return;
628 }
629
630 memlimit_mb_active = MAX(new_memlimit_active, memlimit_mb_active);
631 memlimit_mb_inactive = MAX(new_memlimit_inactive, memlimit_mb_inactive);
632
633 memlimit_active_is_fatal = (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL);
634 memlimit_inactive_is_fatal = (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL);
635
636 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_mb_active, memlimit_active_is_fatal);
637 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_mb_inactive, memlimit_inactive_is_fatal);
638
639 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
640 use_active_limit = TRUE;
641 CACHE_ACTIVE_LIMITS_LOCKED(p, memlimit_active_is_fatal);
642 } else {
643 CACHE_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive_is_fatal);
644 }
645
646 if (memorystatus_highwater_enabled) {
647 task_set_phys_footprint_limit_internal(p->task,
648 (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1,
649 NULL, /*return old value */
650 use_active_limit, /*active limit?*/
651 (use_active_limit ? memlimit_active_is_fatal : memlimit_inactive_is_fatal));
652 }
653 }
654
655 void
656 memorystatus_act_on_legacy_footprint_entitlement(proc_t p, boolean_t footprint_increase)
657 {
658 int memlimit_mb_active = 0, memlimit_mb_inactive = 0;
659
660 if (p == NULL) {
661 return;
662 }
663
664 proc_list_lock();
665
666 if (p->p_memstat_memlimit_active > 0) {
667 memlimit_mb_active = p->p_memstat_memlimit_active;
668 } else if (p->p_memstat_memlimit_active == -1) {
669 memlimit_mb_active = max_task_footprint_mb;
670 } else {
671 /*
672 * Nothing to do for '0' which is
673 * a special value only used internally
674 * to test 'no limits'.
675 */
676 proc_list_unlock();
677 return;
678 }
679
680 if (p->p_memstat_memlimit_inactive > 0) {
681 memlimit_mb_inactive = p->p_memstat_memlimit_inactive;
682 } else if (p->p_memstat_memlimit_inactive == -1) {
683 memlimit_mb_inactive = max_task_footprint_mb;
684 } else {
685 /*
686 * Nothing to do for '0' which is
687 * a special value only used internally
688 * to test 'no limits'.
689 */
690 proc_list_unlock();
691 return;
692 }
693
694 if (footprint_increase) {
695 memlimit_mb_active += legacy_footprint_bonus_mb;
696 memlimit_mb_inactive += legacy_footprint_bonus_mb;
697 } else {
698 memlimit_mb_active -= legacy_footprint_bonus_mb;
699 if (memlimit_mb_active == max_task_footprint_mb) {
700 memlimit_mb_active = -1; /* reverting back to default system limit */
701 }
702
703 memlimit_mb_inactive -= legacy_footprint_bonus_mb;
704 if (memlimit_mb_inactive == max_task_footprint_mb) {
705 memlimit_mb_inactive = -1; /* reverting back to default system limit */
706 }
707 }
708 memorystatus_raise_memlimit(p, memlimit_mb_active, memlimit_mb_inactive);
709
710 proc_list_unlock();
711 }
712
713 void
714 memorystatus_act_on_ios13extended_footprint_entitlement(proc_t p)
715 {
716 if (max_mem < 1500ULL * 1024 * 1024 ||
717 max_mem > 2ULL * 1024 * 1024 * 1024) {
718 /* ios13extended_footprint is only for 2GB devices */
719 return;
720 }
721 /* limit to "almost 2GB" */
722 proc_list_lock();
723 memorystatus_raise_memlimit(p, 1800, 1800);
724 proc_list_unlock();
725 }
726
727 void
728 memorystatus_act_on_entitled_task_limit(proc_t p)
729 {
730 if (memorystatus_entitled_max_task_footprint_mb == 0) {
731 // Entitlement is not supported on this device.
732 return;
733 }
734 proc_list_lock();
735 memorystatus_raise_memlimit(p, memorystatus_entitled_max_task_footprint_mb, memorystatus_entitled_max_task_footprint_mb);
736 proc_list_unlock();
737 }
738 #endif /* __arm64__ */
739
740 SYSCTL_INT(_kern, OID_AUTO, memorystatus_level, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_level, 0, "");
741
742 int
743 memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
744 {
745 user_addr_t level = 0;
746
747 level = args->level;
748
749 if (copyout(&memorystatus_level, level, sizeof(memorystatus_level)) != 0) {
750 return EFAULT;
751 }
752
753 return 0;
754 }
755
756 static void memorystatus_thread(void *param __unused, wait_result_t wr __unused);
757
758 /* Memory Limits */
759
760 static boolean_t memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
761 static boolean_t memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
762
763
764 static int memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
765
766 static int memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry);
767
768 static int memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
769
770 static int memorystatus_cmd_get_memlimit_excess_np(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
771
772 static void memorystatus_get_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t *p_entry);
773 static int memorystatus_set_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t *p_entry);
774
775 int proc_get_memstat_priority(proc_t, boolean_t);
776
777 static boolean_t memorystatus_idle_snapshot = 0;
778
779 unsigned int memorystatus_delta = 0;
780
781 /* Jetsam Loop Detection */
782 static boolean_t memorystatus_jld_enabled = FALSE; /* Enable jetsam loop detection */
783 static uint32_t memorystatus_jld_eval_period_msecs = 0; /* Init pass sets this based on device memory size */
784 static int memorystatus_jld_eval_aggressive_count = 3; /* Raise the priority max after 'n' aggressive loops */
785 static int memorystatus_jld_eval_aggressive_priority_band_max = 15; /* Kill aggressively up through this band */
786
787 /*
788 * A FG app can request that the aggressive jetsam mechanism display some leniency in the FG band. This 'lenient' mode is described as:
789 * --- if aggressive jetsam kills an app in the FG band and gets back >=AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD memory, it will stop the aggressive march further into and up the jetsam bands.
790 *
791 * RESTRICTIONS:
792 * - Such a request is respected/acknowledged only once while that 'requesting' app is in the FG band i.e. if aggressive jetsam was
793 * needed and the 'lenient' mode was deployed then that's it for this special mode while the app is in the FG band.
794 *
795 * - If the app is still in the FG band and aggressive jetsam is needed again, there will be no stop-and-check the next time around.
796 *
797 * - Also, the transition of the 'requesting' app away from the FG band will void this special behavior.
798 */
799
800 #define AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD 25
801 boolean_t memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
802 boolean_t memorystatus_aggressive_jetsam_lenient = FALSE;
803
804 #if DEVELOPMENT || DEBUG
805 /*
806 * Jetsam Loop Detection tunables.
807 */
808
809 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_period_msecs, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_period_msecs, 0, "");
810 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_count, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_count, 0, "");
811 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_priority_band_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_priority_band_max, 0, "");
812 #endif /* DEVELOPMENT || DEBUG */
813
814 static uint32_t kill_under_pressure_cause = 0;
815
816 /*
817 * snapshot support for memstats collected at boot.
818 */
819 static memorystatus_jetsam_snapshot_t memorystatus_at_boot_snapshot;
820
821 static void memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count);
822 static boolean_t memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount);
823 static void memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime);
824
825 static void memorystatus_clear_errors(void);
826 static void memorystatus_get_task_phys_footprint_page_counts(task_t task,
827 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
828 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
829 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
830 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages, uint64_t *frozen_to_swap_pages);
831
832 static void memorystatus_get_task_memory_region_count(task_t task, uint64_t *count);
833
834 static uint32_t memorystatus_build_state(proc_t p);
835 //static boolean_t memorystatus_issue_pressure_kevent(boolean_t pressured);
836
837 static boolean_t memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, os_reason_t jetsam_reason, int32_t *priority,
838 uint32_t *errors, uint64_t *memory_reclaimed);
839 static boolean_t memorystatus_kill_processes_aggressive(uint32_t cause, int aggr_count, int32_t priority_max, uint32_t *errors, uint64_t *memory_reclaimed);
840 static boolean_t memorystatus_kill_hiwat_proc(uint32_t *errors, boolean_t *purged, uint64_t *memory_reclaimed);
841
842 static boolean_t memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause);
843
844 /* Priority Band Sorting Routines */
845 static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order);
846 static int memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order);
847 static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index);
848 static int memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz);
849
850 /* qsort routines */
851 typedef int (*cmpfunc_t)(const void *a, const void *b);
852 extern void qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
853 static int memstat_asc_cmp(const void *a, const void *b);
854
855 /* VM pressure */
856
857 extern unsigned int vm_page_free_count;
858 extern unsigned int vm_page_active_count;
859 extern unsigned int vm_page_inactive_count;
860 extern unsigned int vm_page_throttled_count;
861 extern unsigned int vm_page_purgeable_count;
862 extern unsigned int vm_page_wire_count;
863 extern unsigned int vm_page_speculative_count;
864
865 #if CONFIG_JETSAM
866 #define MEMORYSTATUS_LOG_AVAILABLE_PAGES memorystatus_available_pages
867 #else /* CONFIG_JETSAM */
868 #define MEMORYSTATUS_LOG_AVAILABLE_PAGES (vm_page_active_count + vm_page_inactive_count + vm_page_free_count + vm_page_speculative_count)
869 #endif /* CONFIG_JETSAM */
870 #if CONFIG_SECLUDED_MEMORY
871 extern unsigned int vm_page_secluded_count;
872 extern unsigned int vm_page_secluded_count_over_target;
873 #endif /* CONFIG_SECLUDED_MEMORY */
874
875 /* Aggressive jetsam pages threshold for sysproc aging policy */
876 unsigned int memorystatus_sysproc_aging_aggr_pages = 0;
877
878 #if CONFIG_JETSAM
879 unsigned int memorystatus_available_pages = (unsigned int)-1;
880 unsigned int memorystatus_available_pages_pressure = 0;
881 unsigned int memorystatus_available_pages_critical = 0;
882 unsigned int memorystatus_available_pages_critical_base = 0;
883 unsigned int memorystatus_available_pages_critical_idle_offset = 0;
884
885 #if DEVELOPMENT || DEBUG
886 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
887 #else
888 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_MASKED | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
889 #endif /* DEVELOPMENT || DEBUG */
890
891 static unsigned int memorystatus_jetsam_policy = kPolicyDefault;
892 unsigned int memorystatus_policy_more_free_offset_pages = 0;
893 static void memorystatus_update_levels_locked(boolean_t critical_only);
894 static unsigned int memorystatus_thread_wasted_wakeup = 0;
895
896 /* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
897 extern void vm_thrashing_jetsam_done(void);
898 static int memorystatus_cmd_set_jetsam_memory_limit(pid_t pid, int32_t high_water_mark, __unused int32_t *retval, boolean_t is_fatal_limit);
899 #if DEVELOPMENT || DEBUG
900 static inline uint32_t
901 roundToNearestMB(uint32_t in)
902 {
903 return (in + ((1 << 20) - 1)) >> 20;
904 }
905
906 static int memorystatus_cmd_increase_jetsam_task_limit(pid_t pid, uint32_t byte_increase);
907 #endif
908
909 int32_t max_kill_priority = JETSAM_PRIORITY_MAX;
910
911 #else /* CONFIG_JETSAM */
912
913 uint64_t memorystatus_available_pages = (uint64_t)-1;
914 uint64_t memorystatus_available_pages_pressure = (uint64_t)-1;
915 uint64_t memorystatus_available_pages_critical = (uint64_t)-1;
916
917 int32_t max_kill_priority = JETSAM_PRIORITY_IDLE;
918 #endif /* CONFIG_JETSAM */
919
920 #if DEVELOPMENT || DEBUG
921
922 static LCK_GRP_DECLARE(disconnect_page_mappings_lck_grp, "disconnect_page_mappings");
923 static LCK_MTX_DECLARE(disconnect_page_mappings_mutex, &disconnect_page_mappings_lck_grp);
924
925 extern bool kill_on_no_paging_space;
926 #endif /* DEVELOPMENT || DEBUG */
927
928
929 /* Debug */
930
931 extern struct knote *vm_find_knote_from_pid(pid_t, struct klist *);
932
933 #if DEVELOPMENT || DEBUG
934
935 static unsigned int memorystatus_debug_dump_this_bucket = 0;
936
937 static void
938 memorystatus_debug_dump_bucket_locked(unsigned int bucket_index)
939 {
940 proc_t p = NULL;
941 uint64_t bytes = 0;
942 int ledger_limit = 0;
943 unsigned int b = bucket_index;
944 boolean_t traverse_all_buckets = FALSE;
945
946 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
947 traverse_all_buckets = TRUE;
948 b = 0;
949 } else {
950 traverse_all_buckets = FALSE;
951 b = bucket_index;
952 }
953
954 /*
955 * footprint reported in [pages / MB ]
956 * limits reported as:
957 * L-limit proc's Ledger limit
958 * C-limit proc's Cached limit, should match Ledger
959 * A-limit proc's Active limit
960 * IA-limit proc's Inactive limit
961 * F==Fatal, NF==NonFatal
962 */
963
964 printf("memorystatus_debug_dump ***START*(PAGE_SIZE_64=%llu)**\n", PAGE_SIZE_64);
965 printf("bucket [pid] [pages / MB] [state] [EP / RP / AP] dirty deadline [L-limit / C-limit / A-limit / IA-limit] name\n");
966 p = memorystatus_get_first_proc_locked(&b, traverse_all_buckets);
967 while (p) {
968 bytes = get_task_phys_footprint(p->task);
969 task_get_phys_footprint_limit(p->task, &ledger_limit);
970 printf("%2d [%5d] [%5lld /%3lldMB] 0x%-8x [%2d / %2d / %2d] 0x%-3x %10lld [%3d / %3d%s / %3d%s / %3d%s] %s\n",
971 b, p->p_pid,
972 (bytes / PAGE_SIZE_64), /* task's footprint converted from bytes to pages */
973 (bytes / (1024ULL * 1024ULL)), /* task's footprint converted from bytes to MB */
974 p->p_memstat_state, p->p_memstat_effectivepriority, p->p_memstat_requestedpriority, p->p_memstat_assertionpriority,
975 p->p_memstat_dirty, p->p_memstat_idledeadline,
976 ledger_limit,
977 p->p_memstat_memlimit,
978 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"),
979 p->p_memstat_memlimit_active,
980 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL ? "F " : "NF"),
981 p->p_memstat_memlimit_inactive,
982 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL ? "F " : "NF"),
983 (*p->p_name ? p->p_name : "unknown"));
984 p = memorystatus_get_next_proc_locked(&b, p, traverse_all_buckets);
985 }
986 printf("memorystatus_debug_dump ***END***\n");
987 }
988
989 static int
990 sysctl_memorystatus_debug_dump_bucket SYSCTL_HANDLER_ARGS
991 {
992 #pragma unused(oidp, arg2)
993 int bucket_index = 0;
994 int error;
995 error = SYSCTL_OUT(req, arg1, sizeof(int));
996 if (error || !req->newptr) {
997 return error;
998 }
999 error = SYSCTL_IN(req, &bucket_index, sizeof(int));
1000 if (error || !req->newptr) {
1001 return error;
1002 }
1003 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1004 /*
1005 * All jetsam buckets will be dumped.
1006 */
1007 } else {
1008 /*
1009 * Only a single bucket will be dumped.
1010 */
1011 }
1012
1013 proc_list_lock();
1014 memorystatus_debug_dump_bucket_locked(bucket_index);
1015 proc_list_unlock();
1016 memorystatus_debug_dump_this_bucket = bucket_index;
1017 return error;
1018 }
1019
1020 /*
1021 * Debug aid to look at jetsam buckets and proc jetsam fields.
1022 * Use this sysctl to act on a particular jetsam bucket.
1023 * Writing the sysctl triggers the dump.
1024 * Usage: sysctl kern.memorystatus_debug_dump_this_bucket=<bucket_index>
1025 */
1026
1027 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_debug_dump_this_bucket, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_debug_dump_this_bucket, 0, sysctl_memorystatus_debug_dump_bucket, "I", "");
1028
1029
1030 /* Debug aid to aid determination of limit */
1031
1032 static int
1033 sysctl_memorystatus_highwater_enable SYSCTL_HANDLER_ARGS
1034 {
1035 #pragma unused(oidp, arg2)
1036 proc_t p;
1037 unsigned int b = 0;
1038 int error, enable = 0;
1039 boolean_t use_active; /* use the active limit and active limit attributes */
1040 boolean_t is_fatal;
1041
1042 error = SYSCTL_OUT(req, arg1, sizeof(int));
1043 if (error || !req->newptr) {
1044 return error;
1045 }
1046
1047 error = SYSCTL_IN(req, &enable, sizeof(int));
1048 if (error || !req->newptr) {
1049 return error;
1050 }
1051
1052 if (!(enable == 0 || enable == 1)) {
1053 return EINVAL;
1054 }
1055
1056 proc_list_lock();
1057
1058 p = memorystatus_get_first_proc_locked(&b, TRUE);
1059 while (p) {
1060 use_active = proc_jetsam_state_is_active_locked(p);
1061
1062 if (enable) {
1063 if (use_active == TRUE) {
1064 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
1065 } else {
1066 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
1067 }
1068 } else {
1069 /*
1070 * Disabling limits does not touch the stored variants.
1071 * Set the cached limit fields to system_wide defaults.
1072 */
1073 p->p_memstat_memlimit = -1;
1074 p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
1075 is_fatal = TRUE;
1076 }
1077
1078 /*
1079 * Enforce the cached limit by writing to the ledger.
1080 */
1081 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit: -1, NULL, use_active, is_fatal);
1082
1083 p = memorystatus_get_next_proc_locked(&b, p, TRUE);
1084 }
1085
1086 memorystatus_highwater_enabled = enable;
1087
1088 proc_list_unlock();
1089
1090 return 0;
1091 }
1092
1093 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_highwater_enabled, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_highwater_enabled, 0, sysctl_memorystatus_highwater_enable, "I", "");
1094
1095 SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
1096
1097 #if CONFIG_JETSAM
1098 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
1099 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
1100 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
1101 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_policy_more_free_offset_pages, CTLFLAG_RW, &memorystatus_policy_more_free_offset_pages, 0, "");
1102
1103 static unsigned int memorystatus_jetsam_panic_debug = 0;
1104
1105 #if VM_PRESSURE_EVENTS
1106
1107 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_pressure, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_pressure, 0, "");
1108
1109 #endif /* VM_PRESSURE_EVENTS */
1110
1111 #endif /* CONFIG_JETSAM */
1112
1113 #endif /* DEVELOPMENT || DEBUG */
1114
1115 extern kern_return_t kernel_thread_start_priority(thread_continue_t continuation,
1116 void *parameter,
1117 integer_t priority,
1118 thread_t *new_thread);
1119
1120 #if DEVELOPMENT || DEBUG
1121
1122 static int
1123 sysctl_memorystatus_disconnect_page_mappings SYSCTL_HANDLER_ARGS
1124 {
1125 #pragma unused(arg1, arg2)
1126 int error = 0, pid = 0;
1127 proc_t p;
1128
1129 error = sysctl_handle_int(oidp, &pid, 0, req);
1130 if (error || !req->newptr) {
1131 return error;
1132 }
1133
1134 lck_mtx_lock(&disconnect_page_mappings_mutex);
1135
1136 if (pid == -1) {
1137 vm_pageout_disconnect_all_pages();
1138 } else {
1139 p = proc_find(pid);
1140
1141 if (p != NULL) {
1142 error = task_disconnect_page_mappings(p->task);
1143
1144 proc_rele(p);
1145
1146 if (error) {
1147 error = EIO;
1148 }
1149 } else {
1150 error = EINVAL;
1151 }
1152 }
1153 lck_mtx_unlock(&disconnect_page_mappings_mutex);
1154
1155 return error;
1156 }
1157
1158 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_disconnect_page_mappings, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
1159 0, 0, &sysctl_memorystatus_disconnect_page_mappings, "I", "");
1160
1161 #endif /* DEVELOPMENT || DEBUG */
1162
1163 /*
1164 * Sorts the given bucket.
1165 *
1166 * Input:
1167 * bucket_index - jetsam priority band to be sorted.
1168 * sort_order - JETSAM_SORT_xxx from kern_memorystatus.h
1169 * Currently sort_order is only meaningful when handling
1170 * coalitions.
1171 *
1172 * proc_list_lock must be held by the caller.
1173 */
1174 static void
1175 memorystatus_sort_bucket_locked(unsigned int bucket_index, int sort_order)
1176 {
1177 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
1178 if (memstat_bucket[bucket_index].count == 0) {
1179 return;
1180 }
1181
1182 switch (bucket_index) {
1183 case JETSAM_PRIORITY_FOREGROUND:
1184 if (memorystatus_sort_by_largest_coalition_locked(bucket_index, sort_order) == 0) {
1185 /*
1186 * Fall back to per process sorting when zero coalitions are found.
1187 */
1188 memorystatus_sort_by_largest_process_locked(bucket_index);
1189 }
1190 break;
1191 default:
1192 memorystatus_sort_by_largest_process_locked(bucket_index);
1193 break;
1194 }
1195 }
1196
1197 /*
1198 * Picks the sorting routine for a given jetsam priority band.
1199 *
1200 * Input:
1201 * bucket_index - jetsam priority band to be sorted.
1202 * sort_order - JETSAM_SORT_xxx from kern_memorystatus.h
1203 * Currently sort_order is only meaningful when handling
1204 * coalitions.
1205 *
1206 * Return:
1207 * 0 on success
1208 * non-0 on failure
1209 */
1210 static int
1211 memorystatus_sort_bucket(unsigned int bucket_index, int sort_order)
1212 {
1213 int coal_sort_order;
1214
1215 /*
1216 * Verify the jetsam priority
1217 */
1218 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1219 return EINVAL;
1220 }
1221
1222 #if DEVELOPMENT || DEBUG
1223 if (sort_order == JETSAM_SORT_DEFAULT) {
1224 coal_sort_order = COALITION_SORT_DEFAULT;
1225 } else {
1226 coal_sort_order = sort_order; /* only used for testing scenarios */
1227 }
1228 #else
1229 /* Verify default */
1230 if (sort_order == JETSAM_SORT_DEFAULT) {
1231 coal_sort_order = COALITION_SORT_DEFAULT;
1232 } else {
1233 return EINVAL;
1234 }
1235 #endif
1236
1237 proc_list_lock();
1238 memorystatus_sort_bucket_locked(bucket_index, coal_sort_order);
1239 proc_list_unlock();
1240
1241 return 0;
1242 }
1243
1244 /*
1245 * Sort processes by size for a single jetsam bucket.
1246 */
1247
1248 static void
1249 memorystatus_sort_by_largest_process_locked(unsigned int bucket_index)
1250 {
1251 proc_t p = NULL, insert_after_proc = NULL, max_proc = NULL;
1252 proc_t next_p = NULL, prev_max_proc = NULL;
1253 uint32_t pages = 0, max_pages = 0;
1254 memstat_bucket_t *current_bucket;
1255
1256 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1257 return;
1258 }
1259
1260 current_bucket = &memstat_bucket[bucket_index];
1261
1262 p = TAILQ_FIRST(&current_bucket->list);
1263
1264 while (p) {
1265 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1266 max_pages = pages;
1267 max_proc = p;
1268 prev_max_proc = p;
1269
1270 while ((next_p = TAILQ_NEXT(p, p_memstat_list)) != NULL) {
1271 /* traversing list until we find next largest process */
1272 p = next_p;
1273 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1274 if (pages > max_pages) {
1275 max_pages = pages;
1276 max_proc = p;
1277 }
1278 }
1279
1280 if (prev_max_proc != max_proc) {
1281 /* found a larger process, place it in the list */
1282 TAILQ_REMOVE(&current_bucket->list, max_proc, p_memstat_list);
1283 if (insert_after_proc == NULL) {
1284 TAILQ_INSERT_HEAD(&current_bucket->list, max_proc, p_memstat_list);
1285 } else {
1286 TAILQ_INSERT_AFTER(&current_bucket->list, insert_after_proc, max_proc, p_memstat_list);
1287 }
1288 prev_max_proc = max_proc;
1289 }
1290
1291 insert_after_proc = max_proc;
1292
1293 p = TAILQ_NEXT(max_proc, p_memstat_list);
1294 }
1295 }
1296
1297 proc_t
1298 memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search)
1299 {
1300 memstat_bucket_t *current_bucket;
1301 proc_t next_p;
1302
1303 if ((*bucket_index) >= MEMSTAT_BUCKET_COUNT) {
1304 return NULL;
1305 }
1306
1307 current_bucket = &memstat_bucket[*bucket_index];
1308 next_p = TAILQ_FIRST(&current_bucket->list);
1309 if (!next_p && search) {
1310 while (!next_p && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1311 current_bucket = &memstat_bucket[*bucket_index];
1312 next_p = TAILQ_FIRST(&current_bucket->list);
1313 }
1314 }
1315
1316 return next_p;
1317 }
1318
1319 proc_t
1320 memorystatus_get_next_proc_locked(unsigned int *bucket_index, proc_t p, boolean_t search)
1321 {
1322 memstat_bucket_t *current_bucket;
1323 proc_t next_p;
1324
1325 if (!p || ((*bucket_index) >= MEMSTAT_BUCKET_COUNT)) {
1326 return NULL;
1327 }
1328
1329 next_p = TAILQ_NEXT(p, p_memstat_list);
1330 while (!next_p && search && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1331 current_bucket = &memstat_bucket[*bucket_index];
1332 next_p = TAILQ_FIRST(&current_bucket->list);
1333 }
1334
1335 return next_p;
1336 }
1337
1338 /*
1339 * Structure to hold state for a jetsam thread.
1340 * Typically there should be a single jetsam thread
1341 * unless parallel jetsam is enabled.
1342 */
1343 struct jetsam_thread_state {
1344 uint8_t inited; /* boolean - if the thread is initialized */
1345 uint8_t limit_to_low_bands; /* boolean */
1346 int memorystatus_wakeup; /* wake channel */
1347 int index; /* jetsam thread index */
1348 thread_t thread; /* jetsam thread pointer */
1349 } *jetsam_threads;
1350
1351 /* Maximum number of jetsam threads allowed */
1352 #define JETSAM_THREADS_LIMIT 3
1353
1354 /* Number of active jetsam threads */
1355 _Atomic int active_jetsam_threads = 1;
1356
1357 /* Number of maximum jetsam threads configured */
1358 int max_jetsam_threads = JETSAM_THREADS_LIMIT;
1359
1360 /*
1361 * Global switch for enabling fast jetsam. Fast jetsam is
1362 * hooked up via the system_override() system call. It has the
1363 * following effects:
1364 * - Raise the jetsam threshold ("clear-the-deck")
1365 * - Enabled parallel jetsam on eligible devices
1366 */
1367 #if __AMP__
1368 int fast_jetsam_enabled = 1;
1369 #else /* __AMP__ */
1370 int fast_jetsam_enabled = 0;
1371 #endif /* __AMP__ */
1372
1373 #if CONFIG_DIRTYSTATUS_TRACKING
1374 int dirtystatus_tracking_enabled = 0;
1375 SYSCTL_INT(_kern, OID_AUTO, dirtystatus_tracking_enabled, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &dirtystatus_tracking_enabled, 0, "");
1376 #endif
1377
1378 /* Routine to find the jetsam state structure for the current jetsam thread */
1379 static inline struct jetsam_thread_state *
1380 jetsam_current_thread(void)
1381 {
1382 for (int thr_id = 0; thr_id < max_jetsam_threads; thr_id++) {
1383 if (jetsam_threads[thr_id].thread == current_thread()) {
1384 return &(jetsam_threads[thr_id]);
1385 }
1386 }
1387 return NULL;
1388 }
1389
1390
1391 __private_extern__ void
1392 memorystatus_init(void)
1393 {
1394 kern_return_t result;
1395 int i;
1396
1397 #if CONFIG_FREEZE
1398 memorystatus_freeze_jetsam_band = JETSAM_PRIORITY_UI_SUPPORT;
1399 memorystatus_frozen_processes_max = FREEZE_PROCESSES_MAX;
1400 memorystatus_frozen_shared_mb_max = ((MAX_FROZEN_SHARED_MB_PERCENT * max_task_footprint_mb) / 100); /* 10% of the system wide task limit */
1401 memorystatus_freeze_shared_mb_per_process_max = (memorystatus_frozen_shared_mb_max / 4);
1402 memorystatus_freeze_pages_min = FREEZE_PAGES_MIN;
1403 memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
1404 memorystatus_max_frozen_demotions_daily = MAX_FROZEN_PROCESS_DEMOTIONS;
1405 memorystatus_thaw_count_demotion_threshold = MIN_THAW_DEMOTION_THRESHOLD;
1406 #endif
1407
1408 #if DEVELOPMENT || DEBUG
1409 if (kill_on_no_paging_space) {
1410 max_kill_priority = JETSAM_PRIORITY_MAX;
1411 }
1412 #endif
1413
1414 /* Init buckets */
1415 for (i = 0; i < MEMSTAT_BUCKET_COUNT; i++) {
1416 TAILQ_INIT(&memstat_bucket[i].list);
1417 memstat_bucket[i].count = 0;
1418 memstat_bucket[i].relaunch_high_count = 0;
1419 }
1420 memorystatus_idle_demotion_call = thread_call_allocate((thread_call_func_t)memorystatus_perform_idle_demotion, NULL);
1421
1422 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
1423 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
1424
1425 #if CONFIG_JETSAM
1426 /* Apply overrides */
1427 if (!PE_parse_boot_argn("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage))) {
1428 PE_get_default("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage));
1429 }
1430 if (delta_percentage == 0) {
1431 delta_percentage = 5;
1432 }
1433 if (max_mem > config_jetsam_large_memory_cutoff) {
1434 critical_threshold_percentage = critical_threshold_percentage_larger_devices;
1435 delta_percentage = delta_percentage_larger_devices;
1436 }
1437 assert(delta_percentage < 100);
1438 if (!PE_parse_boot_argn("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage))) {
1439 PE_get_default("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage));
1440 }
1441 assert(critical_threshold_percentage < 100);
1442 PE_get_default("kern.jetsam_idle_offset", &idle_offset_percentage, sizeof(idle_offset_percentage));
1443 assert(idle_offset_percentage < 100);
1444 PE_get_default("kern.jetsam_pressure_threshold", &pressure_threshold_percentage, sizeof(pressure_threshold_percentage));
1445 assert(pressure_threshold_percentage < 100);
1446 PE_get_default("kern.jetsam_freeze_threshold", &freeze_threshold_percentage, sizeof(freeze_threshold_percentage));
1447 assert(freeze_threshold_percentage < 100);
1448
1449
1450 if (!PE_parse_boot_argn("jetsam_aging_policy", &jetsam_aging_policy,
1451 sizeof(jetsam_aging_policy))) {
1452 if (!PE_get_default("kern.jetsam_aging_policy", &jetsam_aging_policy,
1453 sizeof(jetsam_aging_policy))) {
1454 jetsam_aging_policy = kJetsamAgingPolicySysProcsReclaimedFirst;
1455 }
1456 }
1457
1458 if (jetsam_aging_policy > kJetsamAgingPolicyMax) {
1459 jetsam_aging_policy = kJetsamAgingPolicySysProcsReclaimedFirst;
1460 }
1461
1462 switch (jetsam_aging_policy) {
1463 case kJetsamAgingPolicyNone:
1464 system_procs_aging_band = JETSAM_PRIORITY_IDLE;
1465 applications_aging_band = JETSAM_PRIORITY_IDLE;
1466 break;
1467
1468 case kJetsamAgingPolicyLegacy:
1469 /*
1470 * Legacy behavior where some daemons get a 10s protection once
1471 * AND only before the first clean->dirty->clean transition before
1472 * going into IDLE band.
1473 */
1474 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1475 applications_aging_band = JETSAM_PRIORITY_IDLE;
1476 break;
1477
1478 case kJetsamAgingPolicySysProcsReclaimedFirst:
1479 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1480 applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1481 break;
1482
1483 case kJetsamAgingPolicyAppsReclaimedFirst:
1484 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1485 applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1486 break;
1487
1488 default:
1489 break;
1490 }
1491
1492 /*
1493 * The aging bands cannot overlap with the JETSAM_PRIORITY_ELEVATED_INACTIVE
1494 * band and must be below it in priority. This is so that we don't have to make
1495 * our 'aging' code worry about a mix of processes, some of which need to age
1496 * and some others that need to stay elevated in the jetsam bands.
1497 */
1498 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > system_procs_aging_band);
1499 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > applications_aging_band);
1500
1501 /* Take snapshots for idle-exit kills by default? First check the boot-arg... */
1502 if (!PE_parse_boot_argn("jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot))) {
1503 /* ...no boot-arg, so check the device tree */
1504 PE_get_default("kern.jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot));
1505 }
1506
1507 memorystatus_delta = (unsigned int) (delta_percentage * atop_64(max_mem) / 100);
1508 memorystatus_available_pages_critical_idle_offset = (unsigned int) (idle_offset_percentage * atop_64(max_mem) / 100);
1509 memorystatus_available_pages_critical_base = (unsigned int) ((critical_threshold_percentage / delta_percentage) * memorystatus_delta);
1510 memorystatus_policy_more_free_offset_pages = (unsigned int) ((policy_more_free_offset_percentage / delta_percentage) * memorystatus_delta);
1511 memorystatus_sysproc_aging_aggr_pages = (unsigned int) (sysproc_aging_aggr_threshold_percentage * atop_64(max_mem) / 100);
1512
1513 /* Jetsam Loop Detection */
1514 if (max_mem <= (512 * 1024 * 1024)) {
1515 /* 512 MB devices */
1516 memorystatus_jld_eval_period_msecs = 8000; /* 8000 msecs == 8 second window */
1517 } else {
1518 /* 1GB and larger devices */
1519 memorystatus_jld_eval_period_msecs = 6000; /* 6000 msecs == 6 second window */
1520 }
1521
1522 memorystatus_jld_enabled = TRUE;
1523
1524 /* No contention at this point */
1525 memorystatus_update_levels_locked(FALSE);
1526
1527 #endif /* CONFIG_JETSAM */
1528
1529 #if __arm64__
1530 if (!PE_parse_boot_argn("entitled_max_task_pmem", &memorystatus_entitled_max_task_footprint_mb,
1531 sizeof(memorystatus_entitled_max_task_footprint_mb))) {
1532 if (!PE_get_default("kern.entitled_max_task_pmem", &memorystatus_entitled_max_task_footprint_mb,
1533 sizeof(memorystatus_entitled_max_task_footprint_mb))) {
1534 // entitled_max_task_pmem is not supported on this system.
1535 memorystatus_entitled_max_task_footprint_mb = 0;
1536 }
1537 }
1538 if (memorystatus_entitled_max_task_footprint_mb > max_mem / (1UL << 20) || memorystatus_entitled_max_task_footprint_mb < 0) {
1539 os_log_with_startup_serial(OS_LOG_DEFAULT, "Invalid value (%d) for entitled_max_task_pmem. Setting to 0",
1540 memorystatus_entitled_max_task_footprint_mb);
1541 }
1542 #endif /* __arm64__ */
1543
1544 memorystatus_jetsam_snapshot_max = maxproc;
1545
1546 memorystatus_jetsam_snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
1547 (sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_max);
1548
1549 memorystatus_jetsam_snapshot = kalloc_flags(memorystatus_jetsam_snapshot_size, Z_WAITOK | Z_ZERO);
1550 if (!memorystatus_jetsam_snapshot) {
1551 panic("Could not allocate memorystatus_jetsam_snapshot");
1552 }
1553
1554 memorystatus_jetsam_snapshot_copy = kalloc_flags(memorystatus_jetsam_snapshot_size, Z_WAITOK | Z_ZERO);
1555 if (!memorystatus_jetsam_snapshot_copy) {
1556 panic("Could not allocate memorystatus_jetsam_snapshot_copy");
1557 }
1558
1559 #if CONFIG_FREEZE
1560 memorystatus_jetsam_snapshot_freezer_max = memorystatus_jetsam_snapshot_max / JETSAM_SNAPSHOT_FREEZER_MAX_FACTOR;
1561 memorystatus_jetsam_snapshot_freezer_size = sizeof(memorystatus_jetsam_snapshot_t) +
1562 (sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_freezer_max);
1563
1564 memorystatus_jetsam_snapshot_freezer = kalloc_flags(memorystatus_jetsam_snapshot_freezer_size, Z_WAITOK | Z_ZERO);
1565 if (!memorystatus_jetsam_snapshot_freezer) {
1566 panic("Could not allocate memorystatus_jetsam_snapshot_freezer");
1567 }
1568 #endif /* CONFIG_FREEZE */
1569
1570 nanoseconds_to_absolutetime((uint64_t)JETSAM_SNAPSHOT_TIMEOUT_SECS * NSEC_PER_SEC, &memorystatus_jetsam_snapshot_timeout);
1571
1572 memset(&memorystatus_at_boot_snapshot, 0, sizeof(memorystatus_jetsam_snapshot_t));
1573
1574 #if CONFIG_FREEZE
1575 memorystatus_freeze_threshold = (unsigned int) ((freeze_threshold_percentage / delta_percentage) * memorystatus_delta);
1576 #endif
1577
1578 /* Check the boot-arg to see if fast jetsam is allowed */
1579 if (!PE_parse_boot_argn("fast_jetsam_enabled", &fast_jetsam_enabled, sizeof(fast_jetsam_enabled))) {
1580 fast_jetsam_enabled = 0;
1581 }
1582
1583 /* Check the boot-arg to configure the maximum number of jetsam threads */
1584 if (!PE_parse_boot_argn("max_jetsam_threads", &max_jetsam_threads, sizeof(max_jetsam_threads))) {
1585 max_jetsam_threads = JETSAM_THREADS_LIMIT;
1586 }
1587
1588 /* Restrict the maximum number of jetsam threads to JETSAM_THREADS_LIMIT */
1589 if (max_jetsam_threads > JETSAM_THREADS_LIMIT) {
1590 max_jetsam_threads = JETSAM_THREADS_LIMIT;
1591 }
1592
1593 /* For low CPU systems disable fast jetsam mechanism */
1594 if (vm_pageout_state.vm_restricted_to_single_processor == TRUE) {
1595 max_jetsam_threads = 1;
1596 fast_jetsam_enabled = 0;
1597 }
1598
1599 /* Initialize the jetsam_threads state array */
1600 jetsam_threads = zalloc_permanent(sizeof(struct jetsam_thread_state) *
1601 max_jetsam_threads, ZALIGN(struct jetsam_thread_state));
1602
1603 /* Initialize all the jetsam threads */
1604 for (i = 0; i < max_jetsam_threads; i++) {
1605 jetsam_threads[i].inited = FALSE;
1606 jetsam_threads[i].index = i;
1607 result = kernel_thread_start_priority(memorystatus_thread, NULL, 95 /* MAXPRI_KERNEL */, &jetsam_threads[i].thread);
1608 if (result != KERN_SUCCESS) {
1609 panic("Could not create memorystatus_thread %d", i);
1610 }
1611 thread_deallocate(jetsam_threads[i].thread);
1612 }
1613 }
1614
1615 /* Centralised for the purposes of allowing panic-on-jetsam */
1616 extern void
1617 vm_run_compactor(void);
1618 extern void
1619 vm_wake_compactor_swapper(void);
1620
1621 /*
1622 * The jetsam no frills kill call
1623 * Return: 0 on success
1624 * error code on failure (EINVAL...)
1625 */
1626 static int
1627 jetsam_do_kill(proc_t p, int jetsam_flags, os_reason_t jetsam_reason)
1628 {
1629 int error = 0;
1630 error = exit_with_reason(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags, jetsam_reason);
1631 return error;
1632 }
1633
1634 /*
1635 * Wrapper for processes exiting with memorystatus details
1636 */
1637 static boolean_t
1638 memorystatus_do_kill(proc_t p, uint32_t cause, os_reason_t jetsam_reason, uint64_t *footprint_of_killed_proc)
1639 {
1640 int error = 0;
1641 __unused pid_t victim_pid = p->p_pid;
1642 uint64_t footprint = get_task_phys_footprint(p->task);
1643 #if (KDEBUG_LEVEL >= KDEBUG_LEVEL_STANDARD)
1644 int32_t memstat_effectivepriority = p->p_memstat_effectivepriority;
1645 #endif /* (KDEBUG_LEVEL >= KDEBUG_LEVEL_STANDARD) */
1646
1647 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_START,
1648 victim_pid, cause, vm_page_free_count, footprint, 0);
1649 DTRACE_MEMORYSTATUS4(memorystatus_do_kill, proc_t, p, os_reason_t, jetsam_reason, uint32_t, cause, uint64_t, footprint);
1650 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
1651 if (memorystatus_jetsam_panic_debug & (1 << cause)) {
1652 panic("memorystatus_do_kill(): jetsam debug panic (cause: %d)", cause);
1653 }
1654 #else
1655 #pragma unused(cause)
1656 #endif
1657
1658 if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
1659 printf("memorystatus: killing process %d [%s] in high band %s (%d) - memorystatus_available_pages: %llu\n", p->p_pid,
1660 (*p->p_name ? p->p_name : "unknown"),
1661 memorystatus_priority_band_name(p->p_memstat_effectivepriority), p->p_memstat_effectivepriority,
1662 (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES);
1663 }
1664
1665 /*
1666 * The jetsam_reason (os_reason_t) has enough information about the kill cause.
1667 * We don't really need jetsam_flags anymore, so it's okay that not all possible kill causes have been mapped.
1668 */
1669 int jetsam_flags = P_LTERM_JETSAM;
1670 switch (cause) {
1671 case kMemorystatusKilledHiwat: jetsam_flags |= P_JETSAM_HIWAT; break;
1672 case kMemorystatusKilledVnodes: jetsam_flags |= P_JETSAM_VNODE; break;
1673 case kMemorystatusKilledVMPageShortage: jetsam_flags |= P_JETSAM_VMPAGESHORTAGE; break;
1674 case kMemorystatusKilledVMCompressorThrashing:
1675 case kMemorystatusKilledVMCompressorSpaceShortage: jetsam_flags |= P_JETSAM_VMTHRASHING; break;
1676 case kMemorystatusKilledFCThrashing: jetsam_flags |= P_JETSAM_FCTHRASHING; break;
1677 case kMemorystatusKilledPerProcessLimit: jetsam_flags |= P_JETSAM_PID; break;
1678 case kMemorystatusKilledIdleExit: jetsam_flags |= P_JETSAM_IDLEEXIT; break;
1679 }
1680 error = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
1681 *footprint_of_killed_proc = ((error == 0) ? footprint : 0);
1682
1683 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_END,
1684 victim_pid, memstat_effectivepriority, vm_page_free_count, error, 0);
1685
1686 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_COMPACTOR_RUN)) | DBG_FUNC_START,
1687 victim_pid, cause, vm_page_free_count, *footprint_of_killed_proc, 0);
1688
1689 if (jetsam_reason->osr_code == JETSAM_REASON_VNODE) {
1690 /*
1691 * vnode jetsams are syncronous and not caused by memory pressure.
1692 * Running the compactor on this thread adds significant latency to the filesystem operation
1693 * that triggered this jetsam.
1694 * Kick of compactor thread asyncronously instead.
1695 */
1696 vm_wake_compactor_swapper();
1697 } else {
1698 vm_run_compactor();
1699 }
1700
1701 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_COMPACTOR_RUN)) | DBG_FUNC_END,
1702 victim_pid, cause, vm_page_free_count, 0, 0);
1703
1704 return error == 0;
1705 }
1706
1707 /*
1708 * Node manipulation
1709 */
1710
1711 static void
1712 memorystatus_check_levels_locked(void)
1713 {
1714 #if CONFIG_JETSAM
1715 /* Update levels */
1716 memorystatus_update_levels_locked(TRUE);
1717 #else /* CONFIG_JETSAM */
1718 /*
1719 * Nothing to do here currently since we update
1720 * memorystatus_available_pages in vm_pressure_response.
1721 */
1722 #endif /* CONFIG_JETSAM */
1723 }
1724
1725 /*
1726 * Pin a process to a particular jetsam band when it is in the background i.e. not doing active work.
1727 * For an application: that means no longer in the FG band
1728 * For a daemon: that means no longer in its 'requested' jetsam priority band
1729 */
1730
1731 int
1732 memorystatus_update_inactive_jetsam_priority_band(pid_t pid, uint32_t op_flags, int jetsam_prio, boolean_t effective_now)
1733 {
1734 int error = 0;
1735 boolean_t enable = FALSE;
1736 proc_t p = NULL;
1737
1738 if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE) {
1739 enable = TRUE;
1740 } else if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE) {
1741 enable = FALSE;
1742 } else {
1743 return EINVAL;
1744 }
1745
1746 p = proc_find(pid);
1747 if (p != NULL) {
1748 if ((enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) ||
1749 (!enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == 0))) {
1750 /*
1751 * No change in state.
1752 */
1753 } else {
1754 proc_list_lock();
1755
1756 if (enable) {
1757 p->p_memstat_state |= P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
1758 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1759
1760 if (effective_now) {
1761 if (p->p_memstat_effectivepriority < jetsam_prio) {
1762 if (memorystatus_highwater_enabled) {
1763 /*
1764 * Process is about to transition from
1765 * inactive --> active
1766 * assign active state
1767 */
1768 boolean_t is_fatal;
1769 boolean_t use_active = TRUE;
1770 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
1771 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
1772 }
1773 memorystatus_update_priority_locked(p, jetsam_prio, FALSE, FALSE);
1774 }
1775 } else {
1776 if (isProcessInAgingBands(p)) {
1777 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1778 }
1779 }
1780 } else {
1781 p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
1782 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1783
1784 if (effective_now) {
1785 if (p->p_memstat_effectivepriority == jetsam_prio) {
1786 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1787 }
1788 } else {
1789 if (isProcessInAgingBands(p)) {
1790 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1791 }
1792 }
1793 }
1794
1795 proc_list_unlock();
1796 }
1797 proc_rele(p);
1798 error = 0;
1799 } else {
1800 error = ESRCH;
1801 }
1802
1803 return error;
1804 }
1805
1806 static void
1807 memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2)
1808 {
1809 proc_t p;
1810 uint64_t current_time = 0, idle_delay_time = 0;
1811 int demote_prio_band = 0;
1812 memstat_bucket_t *demotion_bucket;
1813
1814 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion()\n");
1815
1816 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_START, 0, 0, 0, 0, 0);
1817
1818 current_time = mach_absolute_time();
1819
1820 proc_list_lock();
1821
1822 demote_prio_band = JETSAM_PRIORITY_IDLE + 1;
1823
1824 for (; demote_prio_band < JETSAM_PRIORITY_MAX; demote_prio_band++) {
1825 if (demote_prio_band != system_procs_aging_band && demote_prio_band != applications_aging_band) {
1826 continue;
1827 }
1828
1829 demotion_bucket = &memstat_bucket[demote_prio_band];
1830 p = TAILQ_FIRST(&demotion_bucket->list);
1831
1832 while (p) {
1833 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion() found %d\n", p->p_pid);
1834
1835 assert(p->p_memstat_idledeadline);
1836
1837 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
1838
1839 if (current_time >= p->p_memstat_idledeadline) {
1840 if ((isSysProc(p) &&
1841 ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) != P_DIRTY_IDLE_EXIT_ENABLED)) || /* system proc marked dirty*/
1842 task_has_assertions((struct task *)(p->task))) { /* has outstanding assertions which might indicate outstanding work too */
1843 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_time(p) : memorystatus_apps_idle_time(p);
1844
1845 p->p_memstat_idledeadline += idle_delay_time;
1846 p = TAILQ_NEXT(p, p_memstat_list);
1847 } else {
1848 proc_t next_proc = NULL;
1849
1850 next_proc = TAILQ_NEXT(p, p_memstat_list);
1851 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1852
1853 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false, true);
1854
1855 p = next_proc;
1856 continue;
1857 }
1858 } else {
1859 // No further candidates
1860 break;
1861 }
1862 }
1863 }
1864
1865 memorystatus_reschedule_idle_demotion_locked();
1866
1867 proc_list_unlock();
1868
1869 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_END, 0, 0, 0, 0, 0);
1870 }
1871
1872 static void
1873 memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state)
1874 {
1875 boolean_t present_in_sysprocs_aging_bucket = FALSE;
1876 boolean_t present_in_apps_aging_bucket = FALSE;
1877 uint64_t idle_delay_time = 0;
1878
1879 if (jetsam_aging_policy == kJetsamAgingPolicyNone) {
1880 return;
1881 }
1882
1883 if ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) ||
1884 (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION)) {
1885 /*
1886 * This process isn't going to be making the trip to the lower bands.
1887 */
1888 return;
1889 }
1890
1891 if (isProcessInAgingBands(p)) {
1892 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
1893 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) != P_DIRTY_AGING_IN_PROGRESS);
1894 }
1895
1896 if (isSysProc(p) && system_procs_aging_band) {
1897 present_in_sysprocs_aging_bucket = TRUE;
1898 } else if (isApp(p) && applications_aging_band) {
1899 present_in_apps_aging_bucket = TRUE;
1900 }
1901 }
1902
1903 assert(!present_in_sysprocs_aging_bucket);
1904 assert(!present_in_apps_aging_bucket);
1905
1906 MEMORYSTATUS_DEBUG(1, "memorystatus_schedule_idle_demotion_locked: scheduling demotion to idle band for pid %d (dirty:0x%x, set_state %d, demotions %d).\n",
1907 p->p_pid, p->p_memstat_dirty, set_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
1908
1909 if (isSysProc(p)) {
1910 assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
1911 }
1912
1913 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_time(p) : memorystatus_apps_idle_time(p);
1914 if (set_state) {
1915 p->p_memstat_dirty |= P_DIRTY_AGING_IN_PROGRESS;
1916 p->p_memstat_idledeadline = mach_absolute_time() + idle_delay_time;
1917 }
1918
1919 assert(p->p_memstat_idledeadline);
1920
1921 if (isSysProc(p) && present_in_sysprocs_aging_bucket == FALSE) {
1922 memorystatus_scheduled_idle_demotions_sysprocs++;
1923 } else if (isApp(p) && present_in_apps_aging_bucket == FALSE) {
1924 memorystatus_scheduled_idle_demotions_apps++;
1925 }
1926 }
1927
1928 void
1929 memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clear_state)
1930 {
1931 boolean_t present_in_sysprocs_aging_bucket = FALSE;
1932 boolean_t present_in_apps_aging_bucket = FALSE;
1933
1934 if (!system_procs_aging_band && !applications_aging_band) {
1935 return;
1936 }
1937
1938 if ((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == 0) {
1939 return;
1940 }
1941
1942 if (isProcessInAgingBands(p)) {
1943 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
1944 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == P_DIRTY_AGING_IN_PROGRESS);
1945 }
1946
1947 if (isSysProc(p) && system_procs_aging_band) {
1948 assert(p->p_memstat_effectivepriority == system_procs_aging_band);
1949 assert(p->p_memstat_idledeadline);
1950 present_in_sysprocs_aging_bucket = TRUE;
1951 } else if (isApp(p) && applications_aging_band) {
1952 assert(p->p_memstat_effectivepriority == applications_aging_band);
1953 assert(p->p_memstat_idledeadline);
1954 present_in_apps_aging_bucket = TRUE;
1955 }
1956 }
1957
1958 MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for pid %d (clear_state %d, demotions %d).\n",
1959 p->p_pid, clear_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
1960
1961
1962 if (clear_state) {
1963 p->p_memstat_idledeadline = 0;
1964 p->p_memstat_dirty &= ~P_DIRTY_AGING_IN_PROGRESS;
1965 }
1966
1967 if (isSysProc(p) && present_in_sysprocs_aging_bucket == TRUE) {
1968 memorystatus_scheduled_idle_demotions_sysprocs--;
1969 assert(memorystatus_scheduled_idle_demotions_sysprocs >= 0);
1970 } else if (isApp(p) && present_in_apps_aging_bucket == TRUE) {
1971 memorystatus_scheduled_idle_demotions_apps--;
1972 assert(memorystatus_scheduled_idle_demotions_apps >= 0);
1973 }
1974
1975 assert((memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps) >= 0);
1976 }
1977
1978 static void
1979 memorystatus_reschedule_idle_demotion_locked(void)
1980 {
1981 if (0 == (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps)) {
1982 if (memstat_idle_demotion_deadline) {
1983 /* Transitioned 1->0, so cancel next call */
1984 thread_call_cancel(memorystatus_idle_demotion_call);
1985 memstat_idle_demotion_deadline = 0;
1986 }
1987 } else {
1988 memstat_bucket_t *demotion_bucket;
1989 proc_t p = NULL, p1 = NULL, p2 = NULL;
1990
1991 if (system_procs_aging_band) {
1992 demotion_bucket = &memstat_bucket[system_procs_aging_band];
1993 p1 = TAILQ_FIRST(&demotion_bucket->list);
1994
1995 p = p1;
1996 }
1997
1998 if (applications_aging_band) {
1999 demotion_bucket = &memstat_bucket[applications_aging_band];
2000 p2 = TAILQ_FIRST(&demotion_bucket->list);
2001
2002 if (p1 && p2) {
2003 p = (p1->p_memstat_idledeadline > p2->p_memstat_idledeadline) ? p2 : p1;
2004 } else {
2005 p = (p1 == NULL) ? p2 : p1;
2006 }
2007 }
2008
2009 assert(p);
2010
2011 if (p != NULL) {
2012 assert(p && p->p_memstat_idledeadline);
2013 if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline) {
2014 thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
2015 memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
2016 }
2017 }
2018 }
2019 }
2020
2021 /*
2022 * List manipulation
2023 */
2024
2025 int
2026 memorystatus_add(proc_t p, boolean_t locked)
2027 {
2028 memstat_bucket_t *bucket;
2029
2030 MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding pid %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
2031
2032 if (!locked) {
2033 proc_list_lock();
2034 }
2035
2036 DTRACE_MEMORYSTATUS2(memorystatus_add, proc_t, p, int32_t, p->p_memstat_effectivepriority);
2037
2038 /* Processes marked internal do not have priority tracked */
2039 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2040 goto exit;
2041 }
2042
2043 /*
2044 * Opt out system processes from being frozen by default.
2045 * For coalition-based freezing, we only want to freeze sysprocs that have specifically opted in.
2046 */
2047 if (isSysProc(p)) {
2048 p->p_memstat_state |= P_MEMSTAT_FREEZE_DISABLED;
2049 }
2050 #if CONFIG_FREEZE
2051 memorystatus_freeze_init_proc(p);
2052 #endif
2053
2054 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2055
2056 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
2057 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs - 1);
2058 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
2059 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps - 1);
2060 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2061 /*
2062 * Entering the idle band.
2063 * Record idle start time.
2064 */
2065 p->p_memstat_idle_start = mach_absolute_time();
2066 }
2067
2068 TAILQ_INSERT_TAIL(&bucket->list, p, p_memstat_list);
2069 bucket->count++;
2070 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2071 bucket->relaunch_high_count++;
2072 }
2073
2074 memorystatus_list_count++;
2075
2076 memorystatus_check_levels_locked();
2077
2078 exit:
2079 if (!locked) {
2080 proc_list_unlock();
2081 }
2082
2083 return 0;
2084 }
2085
2086 /*
2087 * Description:
2088 * Moves a process from one jetsam bucket to another.
2089 * which changes the LRU position of the process.
2090 *
2091 * Monitors transition between buckets and if necessary
2092 * will update cached memory limits accordingly.
2093 *
2094 * skip_demotion_check:
2095 * - if the 'jetsam aging policy' is NOT 'legacy':
2096 * When this flag is TRUE, it means we are going
2097 * to age the ripe processes out of the aging bands and into the
2098 * IDLE band and apply their inactive memory limits.
2099 *
2100 * - if the 'jetsam aging policy' is 'legacy':
2101 * When this flag is TRUE, it might mean the above aging mechanism
2102 * OR
2103 * It might be that we have a process that has used up its 'idle deferral'
2104 * stay that is given to it once per lifetime. And in this case, the process
2105 * won't be going through any aging codepaths. But we still need to apply
2106 * the right inactive limits and so we explicitly set this to TRUE if the
2107 * new priority for the process is the IDLE band.
2108 */
2109 void
2110 memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check)
2111 {
2112 memstat_bucket_t *old_bucket, *new_bucket;
2113
2114 assert(priority < MEMSTAT_BUCKET_COUNT);
2115
2116 /* Ensure that exit isn't underway, leaving the proc retained but removed from its bucket */
2117 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2118 return;
2119 }
2120
2121 MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting %s(%d) to priority %d, inserting at %s\n",
2122 (*p->p_name ? p->p_name : "unknown"), p->p_pid, priority, head_insert ? "head" : "tail");
2123
2124 DTRACE_MEMORYSTATUS3(memorystatus_update_priority, proc_t, p, int32_t, p->p_memstat_effectivepriority, int, priority);
2125
2126 old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2127
2128 if (skip_demotion_check == FALSE) {
2129 if (isSysProc(p)) {
2130 /*
2131 * For system processes, the memorystatus_dirty_* routines take care of adding/removing
2132 * the processes from the aging bands and balancing the demotion counts.
2133 * We can, however, override that if the process has an 'elevated inactive jetsam band' attribute.
2134 */
2135
2136 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2137 /*
2138 * 2 types of processes can use the non-standard elevated inactive band:
2139 * - Frozen processes that always land in memorystatus_freeze_jetsam_band
2140 * OR
2141 * - processes that specifically opt-in to the elevated inactive support e.g. docked processes.
2142 */
2143 #if CONFIG_FREEZE
2144 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
2145 if (priority <= memorystatus_freeze_jetsam_band) {
2146 priority = memorystatus_freeze_jetsam_band;
2147 }
2148 } else
2149 #endif /* CONFIG_FREEZE */
2150 {
2151 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE) {
2152 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2153 }
2154 }
2155 assert(!(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2156 }
2157 } else if (isApp(p)) {
2158 /*
2159 * Check to see if the application is being lowered in jetsam priority. If so, and:
2160 * - it has an 'elevated inactive jetsam band' attribute, then put it in the appropriate band.
2161 * - it is a normal application, then let it age in the aging band if that policy is in effect.
2162 */
2163
2164 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2165 #if CONFIG_FREEZE
2166 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
2167 if (priority <= memorystatus_freeze_jetsam_band) {
2168 priority = memorystatus_freeze_jetsam_band;
2169 }
2170 } else
2171 #endif /* CONFIG_FREEZE */
2172 {
2173 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE) {
2174 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2175 }
2176 }
2177 } else {
2178 if (applications_aging_band) {
2179 if (p->p_memstat_effectivepriority == applications_aging_band) {
2180 assert(old_bucket->count == (memorystatus_scheduled_idle_demotions_apps + 1));
2181 }
2182
2183 if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && (priority <= applications_aging_band)) {
2184 assert(!(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2185 priority = applications_aging_band;
2186 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2187 }
2188 }
2189 }
2190 }
2191 }
2192
2193 if ((system_procs_aging_band && (priority == system_procs_aging_band)) || (applications_aging_band && (priority == applications_aging_band))) {
2194 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
2195 }
2196
2197 #if DEVELOPMENT || DEBUG
2198 if (priority == JETSAM_PRIORITY_IDLE && /* if the process is on its way into the IDLE band */
2199 skip_demotion_check == FALSE && /* and it isn't via the path that will set the INACTIVE memlimits */
2200 (p->p_memstat_dirty & P_DIRTY_TRACK) && /* and it has 'DIRTY' tracking enabled */
2201 ((p->p_memstat_memlimit != p->p_memstat_memlimit_inactive) || /* and we notice that the current limit isn't the right value (inactive) */
2202 ((p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) ? (!(p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT)) : (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT)))) { /* OR type (fatal vs non-fatal) */
2203 printf("memorystatus_update_priority_locked: on %s with 0x%x, prio: %d and %d\n", p->p_name, p->p_memstat_state, priority, p->p_memstat_memlimit); /* then we must catch this */
2204 }
2205 #endif /* DEVELOPMENT || DEBUG */
2206
2207 TAILQ_REMOVE(&old_bucket->list, p, p_memstat_list);
2208 old_bucket->count--;
2209 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2210 old_bucket->relaunch_high_count--;
2211 }
2212
2213 new_bucket = &memstat_bucket[priority];
2214 if (head_insert) {
2215 TAILQ_INSERT_HEAD(&new_bucket->list, p, p_memstat_list);
2216 } else {
2217 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
2218 }
2219 new_bucket->count++;
2220 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2221 new_bucket->relaunch_high_count++;
2222 }
2223
2224 if (memorystatus_highwater_enabled) {
2225 boolean_t is_fatal;
2226 boolean_t use_active;
2227
2228 /*
2229 * If cached limit data is updated, then the limits
2230 * will be enforced by writing to the ledgers.
2231 */
2232 boolean_t ledger_update_needed = TRUE;
2233
2234 /*
2235 * Here, we must update the cached memory limit if the task
2236 * is transitioning between:
2237 * active <--> inactive
2238 * FG <--> BG
2239 * but:
2240 * dirty <--> clean is ignored
2241 *
2242 * We bypass non-idle processes that have opted into dirty tracking because
2243 * a move between buckets does not imply a transition between the
2244 * dirty <--> clean state.
2245 */
2246
2247 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
2248 if (skip_demotion_check == TRUE && priority == JETSAM_PRIORITY_IDLE) {
2249 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2250 use_active = FALSE;
2251 } else {
2252 ledger_update_needed = FALSE;
2253 }
2254 } else if ((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) {
2255 /*
2256 * inactive --> active
2257 * BG --> FG
2258 * assign active state
2259 */
2260 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2261 use_active = TRUE;
2262 } else if ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
2263 /*
2264 * active --> inactive
2265 * FG --> BG
2266 * assign inactive state
2267 */
2268 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2269 use_active = FALSE;
2270 } else {
2271 /*
2272 * The transition between jetsam priority buckets apparently did
2273 * not affect active/inactive state.
2274 * This is not unusual... especially during startup when
2275 * processes are getting established in their respective bands.
2276 */
2277 ledger_update_needed = FALSE;
2278 }
2279
2280 /*
2281 * Enforce the new limits by writing to the ledger
2282 */
2283 if (ledger_update_needed) {
2284 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
2285
2286 MEMORYSTATUS_DEBUG(3, "memorystatus_update_priority_locked: new limit on pid %d (%dMB %s) priority old --> new (%d --> %d) dirty?=0x%x %s\n",
2287 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2288 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, priority, p->p_memstat_dirty,
2289 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2290 }
2291 }
2292
2293 /*
2294 * Record idle start or idle delta.
2295 */
2296 if (p->p_memstat_effectivepriority == priority) {
2297 /*
2298 * This process is not transitioning between
2299 * jetsam priority buckets. Do nothing.
2300 */
2301 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2302 uint64_t now;
2303 /*
2304 * Transitioning out of the idle priority bucket.
2305 * Record idle delta.
2306 */
2307 assert(p->p_memstat_idle_start != 0);
2308 now = mach_absolute_time();
2309 if (now > p->p_memstat_idle_start) {
2310 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2311 }
2312
2313 /*
2314 * About to become active and so memory footprint could change.
2315 * So mark it eligible for freeze-considerations next time around.
2316 */
2317 if (p->p_memstat_state & P_MEMSTAT_FREEZE_IGNORE) {
2318 p->p_memstat_state &= ~P_MEMSTAT_FREEZE_IGNORE;
2319 }
2320 } else if (priority == JETSAM_PRIORITY_IDLE) {
2321 /*
2322 * Transitioning into the idle priority bucket.
2323 * Record idle start.
2324 */
2325 p->p_memstat_idle_start = mach_absolute_time();
2326 }
2327
2328 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CHANGE_PRIORITY), p->p_pid, priority, p->p_memstat_effectivepriority, 0, 0);
2329
2330 p->p_memstat_effectivepriority = priority;
2331
2332 #if CONFIG_SECLUDED_MEMORY
2333 if (secluded_for_apps &&
2334 task_could_use_secluded_mem(p->task)) {
2335 task_set_can_use_secluded_mem(
2336 p->task,
2337 (priority >= JETSAM_PRIORITY_FOREGROUND));
2338 }
2339 #endif /* CONFIG_SECLUDED_MEMORY */
2340
2341 memorystatus_check_levels_locked();
2342 }
2343
2344 int
2345 memorystatus_relaunch_flags_update(proc_t p, int relaunch_flags)
2346 {
2347 p->p_memstat_relaunch_flags = relaunch_flags;
2348 KDBG(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_RELAUNCH_FLAGS), p->p_pid, relaunch_flags, 0, 0, 0);
2349 return 0;
2350 }
2351
2352 /*
2353 *
2354 * Description: Update the jetsam priority and memory limit attributes for a given process.
2355 *
2356 * Parameters:
2357 * p init this process's jetsam information.
2358 * priority The jetsam priority band
2359 * user_data user specific data, unused by the kernel
2360 * is_assertion When true, a priority update is driven by an assertion.
2361 * effective guards against race if process's update already occurred
2362 * update_memlimit When true we know this is the init step via the posix_spawn path.
2363 *
2364 * memlimit_active Value in megabytes; The monitored footprint level while the
2365 * process is active. Exceeding it may result in termination
2366 * based on it's associated fatal flag.
2367 *
2368 * memlimit_active_is_fatal When a process is active and exceeds its memory footprint,
2369 * this describes whether or not it should be immediately fatal.
2370 *
2371 * memlimit_inactive Value in megabytes; The monitored footprint level while the
2372 * process is inactive. Exceeding it may result in termination
2373 * based on it's associated fatal flag.
2374 *
2375 * memlimit_inactive_is_fatal When a process is inactive and exceeds its memory footprint,
2376 * this describes whether or not it should be immediatly fatal.
2377 *
2378 * Returns: 0 Success
2379 * non-0 Failure
2380 */
2381
2382 int
2383 memorystatus_update(proc_t p, int priority, uint64_t user_data, boolean_t is_assertion, boolean_t effective, boolean_t update_memlimit,
2384 int32_t memlimit_active, boolean_t memlimit_active_is_fatal,
2385 int32_t memlimit_inactive, boolean_t memlimit_inactive_is_fatal)
2386 {
2387 int ret;
2388 boolean_t head_insert = false;
2389
2390 MEMORYSTATUS_DEBUG(1, "memorystatus_update: changing (%s) pid %d: priority %d, user_data 0x%llx\n", (*p->p_name ? p->p_name : "unknown"), p->p_pid, priority, user_data);
2391
2392 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_START, p->p_pid, priority, user_data, effective, 0);
2393
2394 if (priority == -1) {
2395 /* Use as shorthand for default priority */
2396 priority = JETSAM_PRIORITY_DEFAULT;
2397 } else if ((priority == system_procs_aging_band) || (priority == applications_aging_band)) {
2398 /* Both the aging bands are reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
2399 priority = JETSAM_PRIORITY_IDLE;
2400 } else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
2401 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
2402 priority = JETSAM_PRIORITY_IDLE;
2403 head_insert = TRUE;
2404 } else if ((priority < 0) || (priority >= MEMSTAT_BUCKET_COUNT)) {
2405 /* Sanity check */
2406 ret = EINVAL;
2407 goto out;
2408 }
2409
2410 proc_list_lock();
2411
2412 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2413
2414 if (effective && (p->p_memstat_state & P_MEMSTAT_PRIORITYUPDATED)) {
2415 ret = EALREADY;
2416 proc_list_unlock();
2417 MEMORYSTATUS_DEBUG(1, "memorystatus_update: effective change specified for pid %d, but change already occurred.\n", p->p_pid);
2418 goto out;
2419 }
2420
2421 if ((p->p_memstat_state & P_MEMSTAT_TERMINATED) || ((p->p_listflag & P_LIST_EXITED) != 0)) {
2422 /*
2423 * This could happen when a process calling posix_spawn() is exiting on the jetsam thread.
2424 */
2425 ret = EBUSY;
2426 proc_list_unlock();
2427 goto out;
2428 }
2429
2430 p->p_memstat_state |= P_MEMSTAT_PRIORITYUPDATED;
2431 p->p_memstat_userdata = user_data;
2432
2433 if (is_assertion) {
2434 if (priority == JETSAM_PRIORITY_IDLE) {
2435 /*
2436 * Assertions relinquish control when the process is heading to IDLE.
2437 */
2438 if (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION) {
2439 /*
2440 * Mark the process as no longer being managed by assertions.
2441 */
2442 p->p_memstat_state &= ~P_MEMSTAT_PRIORITY_ASSERTION;
2443 } else {
2444 /*
2445 * Ignore an idle priority transition if the process is not
2446 * already managed by assertions. We won't treat this as
2447 * an error, but we will log the unexpected behavior and bail.
2448 */
2449 os_log(OS_LOG_DEFAULT, "memorystatus: Ignore assertion driven idle priority. Process not previously controlled %s:%d\n",
2450 (*p->p_name ? p->p_name : "unknown"), p->p_pid);
2451
2452 ret = 0;
2453 proc_list_unlock();
2454 goto out;
2455 }
2456 } else {
2457 /*
2458 * Process is now being managed by assertions,
2459 */
2460 p->p_memstat_state |= P_MEMSTAT_PRIORITY_ASSERTION;
2461 }
2462
2463 /* Always update the assertion priority in this path */
2464
2465 p->p_memstat_assertionpriority = priority;
2466
2467 int memstat_dirty_flags = memorystatus_dirty_get(p, TRUE); /* proc_list_lock is held */
2468
2469 if (memstat_dirty_flags != 0) {
2470 /*
2471 * Calculate maximum priority only when dirty tracking processes are involved.
2472 */
2473 int maxpriority;
2474 if (memstat_dirty_flags & PROC_DIRTY_IS_DIRTY) {
2475 maxpriority = MAX(p->p_memstat_assertionpriority, p->p_memstat_requestedpriority);
2476 } else {
2477 /* clean */
2478
2479 if (memstat_dirty_flags & PROC_DIRTY_ALLOWS_IDLE_EXIT) {
2480 /*
2481 * The aging policy must be evaluated and applied here because runnningboardd
2482 * has relinquished its hold on the jetsam priority by attempting to move a
2483 * clean process to the idle band.
2484 */
2485
2486 int newpriority = JETSAM_PRIORITY_IDLE;
2487 if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
2488 newpriority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
2489 }
2490
2491 maxpriority = MAX(p->p_memstat_assertionpriority, newpriority );
2492
2493 if (newpriority == system_procs_aging_band) {
2494 memorystatus_schedule_idle_demotion_locked(p, FALSE);
2495 }
2496 } else {
2497 /*
2498 * Preserves requestedpriority when the process does not support pressured exit.
2499 */
2500 maxpriority = MAX(p->p_memstat_assertionpriority, p->p_memstat_requestedpriority);
2501 }
2502 }
2503 priority = maxpriority;
2504 }
2505 } else {
2506 p->p_memstat_requestedpriority = priority;
2507 }
2508
2509 if (update_memlimit) {
2510 boolean_t is_fatal;
2511 boolean_t use_active;
2512
2513 /*
2514 * Posix_spawn'd processes come through this path to instantiate ledger limits.
2515 * Forked processes do not come through this path, so no ledger limits exist.
2516 * (That's why forked processes can consume unlimited memory.)
2517 */
2518
2519 MEMORYSTATUS_DEBUG(3, "memorystatus_update(enter): pid %d, priority %d, dirty=0x%x, Active(%dMB %s), Inactive(%dMB, %s)\n",
2520 p->p_pid, priority, p->p_memstat_dirty,
2521 memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
2522 memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
2523
2524 if (memlimit_active <= 0) {
2525 /*
2526 * This process will have a system_wide task limit when active.
2527 * System_wide task limit is always fatal.
2528 * It's quite common to see non-fatal flag passed in here.
2529 * It's not an error, we just ignore it.
2530 */
2531
2532 /*
2533 * For backward compatibility with some unexplained launchd behavior,
2534 * we allow a zero sized limit. But we still enforce system_wide limit
2535 * when written to the ledgers.
2536 */
2537
2538 if (memlimit_active < 0) {
2539 memlimit_active = -1; /* enforces system_wide task limit */
2540 }
2541 memlimit_active_is_fatal = TRUE;
2542 }
2543
2544 if (memlimit_inactive <= 0) {
2545 /*
2546 * This process will have a system_wide task limit when inactive.
2547 * System_wide task limit is always fatal.
2548 */
2549
2550 memlimit_inactive = -1;
2551 memlimit_inactive_is_fatal = TRUE;
2552 }
2553
2554 /*
2555 * Initialize the active limit variants for this process.
2556 */
2557 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
2558
2559 /*
2560 * Initialize the inactive limit variants for this process.
2561 */
2562 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
2563
2564 /*
2565 * Initialize the cached limits for target process.
2566 * When the target process is dirty tracked, it's typically
2567 * in a clean state. Non dirty tracked processes are
2568 * typically active (Foreground or above).
2569 * But just in case, we don't make assumptions...
2570 */
2571
2572 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
2573 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2574 use_active = TRUE;
2575 } else {
2576 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2577 use_active = FALSE;
2578 }
2579
2580 /*
2581 * Enforce the cached limit by writing to the ledger.
2582 */
2583 if (memorystatus_highwater_enabled) {
2584 /* apply now */
2585 task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, use_active, is_fatal);
2586
2587 MEMORYSTATUS_DEBUG(3, "memorystatus_update: init: limit on pid %d (%dMB %s) targeting priority(%d) dirty?=0x%x %s\n",
2588 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2589 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), priority, p->p_memstat_dirty,
2590 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2591 }
2592 }
2593
2594 /*
2595 * We can't add to the aging bands buckets here.
2596 * But, we could be removing it from those buckets.
2597 * Check and take appropriate steps if so.
2598 */
2599
2600 if (isProcessInAgingBands(p)) {
2601 if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && isApp(p) && (priority > applications_aging_band)) {
2602 /*
2603 * Runningboardd is pulling up an application that is in the aging band.
2604 * We reset the app's state here so that it'll get a fresh stay in the
2605 * aging band on the way back.
2606 *
2607 * We always handled the app 'aging' in the memorystatus_update_priority_locked()
2608 * function. Daemons used to be handled via the dirty 'set/clear/track' path.
2609 * But with extensions (daemon-app hybrid), runningboardd is now going through
2610 * this routine for daemons too and things have gotten a bit tangled. This should
2611 * be simplified/untangled at some point and might require some assistance from
2612 * runningboardd.
2613 */
2614 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2615 } else {
2616 memorystatus_invalidate_idle_demotion_locked(p, FALSE);
2617 }
2618 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2619 } else {
2620 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy && priority == JETSAM_PRIORITY_IDLE) {
2621 /*
2622 * Daemons with 'inactive' limits will go through the dirty tracking codepath.
2623 * This path deals with apps that may have 'inactive' limits e.g. WebContent processes.
2624 * If this is the legacy aging policy we explicitly need to apply those limits. If it
2625 * is any other aging policy, then we don't need to worry because all processes
2626 * will go through the aging bands and then the demotion thread will take care to
2627 * move them into the IDLE band and apply the required limits.
2628 */
2629 memorystatus_update_priority_locked(p, priority, head_insert, TRUE);
2630 }
2631 }
2632
2633 memorystatus_update_priority_locked(p, priority, head_insert, FALSE);
2634
2635 proc_list_unlock();
2636 ret = 0;
2637
2638 out:
2639 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_END, ret, 0, 0, 0, 0);
2640
2641 return ret;
2642 }
2643
2644 int
2645 memorystatus_remove(proc_t p)
2646 {
2647 int ret;
2648 memstat_bucket_t *bucket;
2649 boolean_t reschedule = FALSE;
2650
2651 MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing pid %d\n", p->p_pid);
2652
2653 /*
2654 * Check if this proc is locked (because we're performing a freeze).
2655 * If so, we fail and instruct the caller to try again later.
2656 */
2657 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
2658 return EAGAIN;
2659 }
2660
2661 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2662
2663 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2664
2665 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
2666 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs);
2667 reschedule = TRUE;
2668 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
2669 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps);
2670 reschedule = TRUE;
2671 }
2672
2673 /*
2674 * Record idle delta
2675 */
2676
2677 if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2678 uint64_t now = mach_absolute_time();
2679 if (now > p->p_memstat_idle_start) {
2680 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2681 }
2682 }
2683
2684 TAILQ_REMOVE(&bucket->list, p, p_memstat_list);
2685 bucket->count--;
2686 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2687 bucket->relaunch_high_count--;
2688 }
2689
2690 memorystatus_list_count--;
2691
2692 /* If awaiting demotion to the idle band, clean up */
2693 if (reschedule) {
2694 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2695 memorystatus_reschedule_idle_demotion_locked();
2696 }
2697
2698 memorystatus_check_levels_locked();
2699
2700 #if CONFIG_FREEZE
2701 if (p->p_memstat_state & (P_MEMSTAT_FROZEN)) {
2702 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
2703 p->p_memstat_state &= ~P_MEMSTAT_REFREEZE_ELIGIBLE;
2704 memorystatus_refreeze_eligible_count--;
2705 }
2706
2707 memorystatus_frozen_count--;
2708 memorystatus_frozen_shared_mb -= p->p_memstat_freeze_sharedanon_pages;
2709 p->p_memstat_freeze_sharedanon_pages = 0;
2710 }
2711
2712 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
2713 memorystatus_suspended_count--;
2714 }
2715 #endif
2716
2717 #if DEVELOPMENT || DEBUG
2718 if (p->p_pid == memorystatus_testing_pid) {
2719 memorystatus_testing_pid = 0;
2720 }
2721 #endif /* DEVELOPMENT || DEBUG */
2722
2723 if (p) {
2724 ret = 0;
2725 } else {
2726 ret = ESRCH;
2727 }
2728
2729 return ret;
2730 }
2731
2732 /*
2733 * Validate dirty tracking flags with process state.
2734 *
2735 * Return:
2736 * 0 on success
2737 * non-0 on failure
2738 *
2739 * The proc_list_lock is held by the caller.
2740 */
2741
2742 static int
2743 memorystatus_validate_track_flags(struct proc *target_p, uint32_t pcontrol)
2744 {
2745 /* See that the process isn't marked for termination */
2746 if (target_p->p_memstat_dirty & P_DIRTY_TERMINATED) {
2747 return EBUSY;
2748 }
2749
2750 /* Idle exit requires that process be tracked */
2751 if ((pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) &&
2752 !(pcontrol & PROC_DIRTY_TRACK)) {
2753 return EINVAL;
2754 }
2755
2756 /* 'Launch in progress' tracking requires that process have enabled dirty tracking too. */
2757 if ((pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) &&
2758 !(pcontrol & PROC_DIRTY_TRACK)) {
2759 return EINVAL;
2760 }
2761
2762 /* Only one type of DEFER behavior is allowed.*/
2763 if ((pcontrol & PROC_DIRTY_DEFER) &&
2764 (pcontrol & PROC_DIRTY_DEFER_ALWAYS)) {
2765 return EINVAL;
2766 }
2767
2768 /* Deferral is only relevant if idle exit is specified */
2769 if (((pcontrol & PROC_DIRTY_DEFER) ||
2770 (pcontrol & PROC_DIRTY_DEFER_ALWAYS)) &&
2771 !(pcontrol & PROC_DIRTY_ALLOWS_IDLE_EXIT)) {
2772 return EINVAL;
2773 }
2774
2775 return 0;
2776 }
2777
2778 static void
2779 memorystatus_update_idle_priority_locked(proc_t p)
2780 {
2781 int32_t priority;
2782
2783 MEMORYSTATUS_DEBUG(1, "memorystatus_update_idle_priority_locked(): pid %d dirty 0x%X\n", p->p_pid, p->p_memstat_dirty);
2784
2785 assert(isSysProc(p));
2786
2787 if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
2788 priority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
2789 } else {
2790 priority = p->p_memstat_requestedpriority;
2791 }
2792
2793 if (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION) {
2794 /*
2795 * This process has a jetsam priority managed by an assertion.
2796 * Policy is to choose the max priority.
2797 */
2798 if (p->p_memstat_assertionpriority > priority) {
2799 os_log(OS_LOG_DEFAULT, "memorystatus: assertion priority %d overrides priority %d for %s:%d\n",
2800 p->p_memstat_assertionpriority, priority,
2801 (*p->p_name ? p->p_name : "unknown"), p->p_pid);
2802 priority = p->p_memstat_assertionpriority;
2803 }
2804 }
2805
2806 if (priority != p->p_memstat_effectivepriority) {
2807 if ((jetsam_aging_policy == kJetsamAgingPolicyLegacy) &&
2808 (priority == JETSAM_PRIORITY_IDLE)) {
2809 /*
2810 * This process is on its way into the IDLE band. The system is
2811 * using 'legacy' jetsam aging policy. That means, this process
2812 * has already used up its idle-deferral aging time that is given
2813 * once per its lifetime. So we need to set the INACTIVE limits
2814 * explicitly because it won't be going through the demotion paths
2815 * that take care to apply the limits appropriately.
2816 */
2817
2818 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2819 /*
2820 * This process has the 'elevated inactive jetsam band' attribute.
2821 * So, there will be no trip to IDLE after all.
2822 * Instead, we pin the process in the elevated band,
2823 * where its ACTIVE limits will apply.
2824 */
2825
2826 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2827 }
2828
2829 memorystatus_update_priority_locked(p, priority, false, true);
2830 } else {
2831 memorystatus_update_priority_locked(p, priority, false, false);
2832 }
2833 }
2834 }
2835
2836 /*
2837 * Processes can opt to have their state tracked by the kernel, indicating when they are busy (dirty) or idle
2838 * (clean). They may also indicate that they support termination when idle, with the result that they are promoted
2839 * to their desired, higher, jetsam priority when dirty (and are therefore killed later), and demoted to the low
2840 * priority idle band when clean (and killed earlier, protecting higher priority procesess).
2841 *
2842 * If the deferral flag is set, then newly tracked processes will be protected for an initial period (as determined by
2843 * memorystatus_sysprocs_idle_delay_time); if they go clean during this time, then they will be moved to a deferred-idle band
2844 * with a slightly higher priority, guarding against immediate termination under memory pressure and being unable to
2845 * make forward progress. Finally, when the guard expires, they will be moved to the standard, lowest-priority, idle
2846 * band. The deferral can be cleared early by clearing the appropriate flag.
2847 *
2848 * The deferral timer is active only for the duration that the process is marked as guarded and clean; if the process
2849 * is marked dirty, the timer will be cancelled. Upon being subsequently marked clean, the deferment will either be
2850 * re-enabled or the guard state cleared, depending on whether the guard deadline has passed.
2851 */
2852
2853 int
2854 memorystatus_dirty_track(proc_t p, uint32_t pcontrol)
2855 {
2856 unsigned int old_dirty;
2857 boolean_t reschedule = FALSE;
2858 boolean_t already_deferred = FALSE;
2859 boolean_t defer_now = FALSE;
2860 int ret = 0;
2861
2862 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_TRACK),
2863 p->p_pid, p->p_memstat_dirty, pcontrol, 0, 0);
2864
2865 proc_list_lock();
2866
2867 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2868 /*
2869 * Process is on its way out.
2870 */
2871 ret = EBUSY;
2872 goto exit;
2873 }
2874
2875 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2876 ret = EPERM;
2877 goto exit;
2878 }
2879
2880 if ((ret = memorystatus_validate_track_flags(p, pcontrol)) != 0) {
2881 /* error */
2882 goto exit;
2883 }
2884
2885 old_dirty = p->p_memstat_dirty;
2886
2887 /* These bits are cumulative, as per <rdar://problem/11159924> */
2888 if (pcontrol & PROC_DIRTY_TRACK) {
2889 p->p_memstat_dirty |= P_DIRTY_TRACK;
2890 }
2891
2892 if (pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) {
2893 p->p_memstat_dirty |= P_DIRTY_ALLOW_IDLE_EXIT;
2894 }
2895
2896 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
2897 p->p_memstat_dirty |= P_DIRTY_LAUNCH_IN_PROGRESS;
2898 }
2899
2900 if (old_dirty & P_DIRTY_AGING_IN_PROGRESS) {
2901 already_deferred = TRUE;
2902 }
2903
2904
2905 /* This can be set and cleared exactly once. */
2906 if (pcontrol & (PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) {
2907 if ((pcontrol & (PROC_DIRTY_DEFER)) &&
2908 !(old_dirty & P_DIRTY_DEFER)) {
2909 p->p_memstat_dirty |= P_DIRTY_DEFER;
2910 }
2911
2912 if ((pcontrol & (PROC_DIRTY_DEFER_ALWAYS)) &&
2913 !(old_dirty & P_DIRTY_DEFER_ALWAYS)) {
2914 p->p_memstat_dirty |= P_DIRTY_DEFER_ALWAYS;
2915 }
2916
2917 defer_now = TRUE;
2918 }
2919
2920 MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for pid %d\n",
2921 ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) ? "Y" : "N",
2922 defer_now ? "Y" : "N",
2923 p->p_memstat_dirty & P_DIRTY ? "Y" : "N",
2924 p->p_pid);
2925
2926 /* Kick off or invalidate the idle exit deferment if there's a state transition. */
2927 if (!(p->p_memstat_dirty & P_DIRTY_IS_DIRTY)) {
2928 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
2929 if (defer_now && !already_deferred) {
2930 /*
2931 * Request to defer a clean process that's idle-exit enabled
2932 * and not already in the jetsam deferred band. Most likely a
2933 * new launch.
2934 */
2935 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2936 reschedule = TRUE;
2937 } else if (!defer_now) {
2938 /*
2939 * The process isn't asking for the 'aging' facility.
2940 * Could be that it is:
2941 */
2942
2943 if (already_deferred) {
2944 /*
2945 * already in the aging bands. Traditionally,
2946 * some processes have tried to use this to
2947 * opt out of the 'aging' facility.
2948 */
2949
2950 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2951 } else {
2952 /*
2953 * agnostic to the 'aging' facility. In that case,
2954 * we'll go ahead and opt it in because this is likely
2955 * a new launch (clean process, dirty tracking enabled)
2956 */
2957
2958 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2959 }
2960
2961 reschedule = TRUE;
2962 }
2963 }
2964 } else {
2965 /*
2966 * We are trying to operate on a dirty process. Dirty processes have to
2967 * be removed from the deferred band. The question is do we reset the
2968 * deferred state or not?
2969 *
2970 * This could be a legal request like:
2971 * - this process had opted into the 'aging' band
2972 * - but it's now dirty and requests to opt out.
2973 * In this case, we remove the process from the band and reset its
2974 * state too. It'll opt back in properly when needed.
2975 *
2976 * OR, this request could be a user-space bug. E.g.:
2977 * - this process had opted into the 'aging' band when clean
2978 * - and, then issues another request to again put it into the band except
2979 * this time the process is dirty.
2980 * The process going dirty, as a transition in memorystatus_dirty_set(), will pull the process out of
2981 * the deferred band with its state intact. So our request below is no-op.
2982 * But we do it here anyways for coverage.
2983 *
2984 * memorystatus_update_idle_priority_locked()
2985 * single-mindedly treats a dirty process as "cannot be in the aging band".
2986 */
2987
2988 if (!defer_now && already_deferred) {
2989 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2990 reschedule = TRUE;
2991 } else {
2992 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
2993
2994 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
2995 reschedule = TRUE;
2996 }
2997 }
2998
2999 memorystatus_update_idle_priority_locked(p);
3000
3001 if (reschedule) {
3002 memorystatus_reschedule_idle_demotion_locked();
3003 }
3004
3005 ret = 0;
3006
3007 exit:
3008 proc_list_unlock();
3009
3010 return ret;
3011 }
3012
3013 int
3014 memorystatus_dirty_set(proc_t p, boolean_t self, uint32_t pcontrol)
3015 {
3016 int ret;
3017 boolean_t kill = false;
3018 boolean_t reschedule = FALSE;
3019 boolean_t was_dirty = FALSE;
3020 boolean_t now_dirty = FALSE;
3021 #if CONFIG_DIRTYSTATUS_TRACKING
3022 boolean_t notify_change = FALSE;
3023 dirty_status_change_event_t change_event;
3024 #endif
3025
3026 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_set(): %d %d 0x%x 0x%x\n", self, p->p_pid, pcontrol, p->p_memstat_dirty);
3027 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_SET), p->p_pid, self, pcontrol, 0, 0);
3028
3029 proc_list_lock();
3030
3031 if ((p->p_listflag & P_LIST_EXITED) != 0) {
3032 /*
3033 * Process is on its way out.
3034 */
3035 ret = EBUSY;
3036 goto exit;
3037 }
3038
3039 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
3040 ret = EPERM;
3041 goto exit;
3042 }
3043
3044 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3045 was_dirty = TRUE;
3046 }
3047
3048 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
3049 /* Dirty tracking not enabled */
3050 ret = EINVAL;
3051 } else if (pcontrol && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
3052 /*
3053 * Process is set to be terminated and we're attempting to mark it dirty.
3054 * Set for termination and marking as clean is OK - see <rdar://problem/10594349>.
3055 */
3056 ret = EBUSY;
3057 } else {
3058 int flag = (self == TRUE) ? P_DIRTY : P_DIRTY_SHUTDOWN;
3059 if (pcontrol && !(p->p_memstat_dirty & flag)) {
3060 /* Mark the process as having been dirtied at some point */
3061 p->p_memstat_dirty |= (flag | P_DIRTY_MARKED);
3062 memorystatus_dirty_count++;
3063 ret = 0;
3064 } else if ((pcontrol == 0) && (p->p_memstat_dirty & flag)) {
3065 if ((flag == P_DIRTY_SHUTDOWN) && (!(p->p_memstat_dirty & P_DIRTY))) {
3066 /* Clearing the dirty shutdown flag, and the process is otherwise clean - kill */
3067 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3068 kill = true;
3069 } else if ((flag == P_DIRTY) && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
3070 /* Kill previously terminated processes if set clean */
3071 kill = true;
3072 }
3073 p->p_memstat_dirty &= ~flag;
3074 memorystatus_dirty_count--;
3075 ret = 0;
3076 } else {
3077 /* Already set */
3078 ret = EALREADY;
3079 }
3080 }
3081
3082 if (ret != 0) {
3083 goto exit;
3084 }
3085
3086 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3087 now_dirty = TRUE;
3088 }
3089
3090 if ((was_dirty == TRUE && now_dirty == FALSE) ||
3091 (was_dirty == FALSE && now_dirty == TRUE)) {
3092 #if CONFIG_DIRTYSTATUS_TRACKING
3093 if (dirtystatus_tracking_enabled) {
3094 uint32_t pages = 0;
3095 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
3096 change_event.dsc_pid = p->p_pid;
3097 change_event.dsc_event_type = (now_dirty == TRUE) ? kDirtyStatusChangedDirty : kDirtyStatusChangedClean;
3098 change_event.dsc_time = mach_absolute_time();
3099 change_event.dsc_pages = pages;
3100 change_event.dsc_priority = p->p_memstat_effectivepriority;
3101 strlcpy(&change_event.dsc_process_name[0], p->p_name, sizeof(change_event.dsc_process_name));
3102 notify_change = TRUE;
3103 }
3104 #endif
3105
3106 /* Manage idle exit deferral, if applied */
3107 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3108 /*
3109 * Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band OR it might be heading back
3110 * there once it's clean again. For the legacy case, this only applies if it has some protection window left.
3111 * P_DIRTY_DEFER: one-time protection window given at launch
3112 * P_DIRTY_DEFER_ALWAYS: protection window given for every dirty->clean transition. Like non-legacy mode.
3113 *
3114 * Non-Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band. It will always stop over
3115 * in that band on it's way to IDLE.
3116 */
3117
3118 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3119 /*
3120 * New dirty process i.e. "was_dirty == FALSE && now_dirty == TRUE"
3121 *
3122 * The process will move from its aging band to its higher requested
3123 * jetsam band.
3124 */
3125 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
3126
3127 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
3128 reschedule = TRUE;
3129 } else {
3130 /*
3131 * Process is back from "dirty" to "clean".
3132 */
3133
3134 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
3135 if (((p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) == FALSE) &&
3136 (mach_absolute_time() >= p->p_memstat_idledeadline)) {
3137 /*
3138 * The process' hasn't enrolled in the "always defer after dirty"
3139 * mode and its deadline has expired. It currently
3140 * does not reside in any of the aging buckets.
3141 *
3142 * It's on its way to the JETSAM_PRIORITY_IDLE
3143 * bucket via memorystatus_update_idle_priority_locked()
3144 * below.
3145 *
3146 * So all we need to do is reset all the state on the
3147 * process that's related to the aging bucket i.e.
3148 * the AGING_IN_PROGRESS flag and the timer deadline.
3149 */
3150
3151 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3152 reschedule = TRUE;
3153 } else {
3154 /*
3155 * Process enrolled in "always stop in deferral band after dirty" OR
3156 * it still has some protection window left and so
3157 * we just re-arm the timer without modifying any
3158 * state on the process iff it still wants into that band.
3159 */
3160
3161 if (p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) {
3162 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3163 reschedule = TRUE;
3164 } else if (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) {
3165 memorystatus_schedule_idle_demotion_locked(p, FALSE);
3166 reschedule = TRUE;
3167 }
3168 }
3169 } else {
3170 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3171 reschedule = TRUE;
3172 }
3173 }
3174 }
3175
3176 memorystatus_update_idle_priority_locked(p);
3177
3178 if (memorystatus_highwater_enabled) {
3179 boolean_t ledger_update_needed = TRUE;
3180 boolean_t use_active;
3181 boolean_t is_fatal;
3182 /*
3183 * We are in this path because this process transitioned between
3184 * dirty <--> clean state. Update the cached memory limits.
3185 */
3186
3187 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
3188 /*
3189 * process is pinned in elevated band
3190 * or
3191 * process is dirty
3192 */
3193 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
3194 use_active = TRUE;
3195 ledger_update_needed = TRUE;
3196 } else {
3197 /*
3198 * process is clean...but if it has opted into pressured-exit
3199 * we don't apply the INACTIVE limit till the process has aged
3200 * out and is entering the IDLE band.
3201 * See memorystatus_update_priority_locked() for that.
3202 */
3203
3204 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3205 ledger_update_needed = FALSE;
3206 } else {
3207 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
3208 use_active = FALSE;
3209 ledger_update_needed = TRUE;
3210 }
3211 }
3212
3213 /*
3214 * Enforce the new limits by writing to the ledger.
3215 *
3216 * This is a hot path and holding the proc_list_lock while writing to the ledgers,
3217 * (where the task lock is taken) is bad. So, we temporarily drop the proc_list_lock.
3218 * We aren't traversing the jetsam bucket list here, so we should be safe.
3219 * See rdar://21394491.
3220 */
3221
3222 if (ledger_update_needed && proc_ref_locked(p) == p) {
3223 int ledger_limit;
3224 if (p->p_memstat_memlimit > 0) {
3225 ledger_limit = p->p_memstat_memlimit;
3226 } else {
3227 ledger_limit = -1;
3228 }
3229 proc_list_unlock();
3230 task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, use_active, is_fatal);
3231 proc_list_lock();
3232 proc_rele_locked(p);
3233
3234 MEMORYSTATUS_DEBUG(3, "memorystatus_dirty_set: new limit on pid %d (%dMB %s) priority(%d) dirty?=0x%x %s\n",
3235 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
3236 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
3237 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
3238 }
3239 }
3240
3241 /* If the deferral state changed, reschedule the demotion timer */
3242 if (reschedule) {
3243 memorystatus_reschedule_idle_demotion_locked();
3244 }
3245 }
3246
3247 if (kill) {
3248 if (proc_ref_locked(p) == p) {
3249 proc_list_unlock();
3250 psignal(p, SIGKILL);
3251 proc_list_lock();
3252 proc_rele_locked(p);
3253 }
3254 }
3255
3256 exit:
3257 proc_list_unlock();
3258
3259 #if CONFIG_DIRTYSTATUS_TRACKING
3260 // Before returning, let's notify the dirtiness status if we have to
3261 if (notify_change) {
3262 memorystatus_send_dirty_status_change_note(&change_event, sizeof(change_event));
3263 }
3264 #endif
3265
3266 return ret;
3267 }
3268
3269 int
3270 memorystatus_dirty_clear(proc_t p, uint32_t pcontrol)
3271 {
3272 int ret = 0;
3273
3274 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_clear(): %d 0x%x 0x%x\n", p->p_pid, pcontrol, p->p_memstat_dirty);
3275
3276 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_CLEAR), p->p_pid, pcontrol, 0, 0, 0);
3277
3278 proc_list_lock();
3279
3280 if ((p->p_listflag & P_LIST_EXITED) != 0) {
3281 /*
3282 * Process is on its way out.
3283 */
3284 ret = EBUSY;
3285 goto exit;
3286 }
3287
3288 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
3289 ret = EPERM;
3290 goto exit;
3291 }
3292
3293 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
3294 /* Dirty tracking not enabled */
3295 ret = EINVAL;
3296 goto exit;
3297 }
3298
3299 if (!pcontrol || (pcontrol & (PROC_DIRTY_LAUNCH_IN_PROGRESS | PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) == 0) {
3300 ret = EINVAL;
3301 goto exit;
3302 }
3303
3304 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
3305 p->p_memstat_dirty &= ~P_DIRTY_LAUNCH_IN_PROGRESS;
3306 }
3307
3308 /* This can be set and cleared exactly once. */
3309 if (pcontrol & (PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) {
3310 if (p->p_memstat_dirty & P_DIRTY_DEFER) {
3311 p->p_memstat_dirty &= ~(P_DIRTY_DEFER);
3312 }
3313
3314 if (p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) {
3315 p->p_memstat_dirty &= ~(P_DIRTY_DEFER_ALWAYS);
3316 }
3317
3318 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3319 memorystatus_update_idle_priority_locked(p);
3320 memorystatus_reschedule_idle_demotion_locked();
3321 }
3322
3323 ret = 0;
3324 exit:
3325 proc_list_unlock();
3326
3327 return ret;
3328 }
3329
3330 int
3331 memorystatus_dirty_get(proc_t p, boolean_t locked)
3332 {
3333 int ret = 0;
3334
3335 if (!locked) {
3336 proc_list_lock();
3337 }
3338
3339 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3340 ret |= PROC_DIRTY_TRACKED;
3341 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3342 ret |= PROC_DIRTY_ALLOWS_IDLE_EXIT;
3343 }
3344 if (p->p_memstat_dirty & P_DIRTY) {
3345 ret |= PROC_DIRTY_IS_DIRTY;
3346 }
3347 if (p->p_memstat_dirty & P_DIRTY_LAUNCH_IN_PROGRESS) {
3348 ret |= PROC_DIRTY_LAUNCH_IS_IN_PROGRESS;
3349 }
3350 }
3351
3352 if (!locked) {
3353 proc_list_unlock();
3354 }
3355
3356 return ret;
3357 }
3358
3359 int
3360 memorystatus_on_terminate(proc_t p)
3361 {
3362 int sig;
3363
3364 proc_list_lock();
3365
3366 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3367
3368 if (((p->p_memstat_dirty & (P_DIRTY_TRACK | P_DIRTY_IS_DIRTY)) == P_DIRTY_TRACK) ||
3369 (p->p_memstat_state & P_MEMSTAT_SUSPENDED)) {
3370 /*
3371 * Mark as terminated and issue SIGKILL if:-
3372 * - process is clean, or,
3373 * - if process is dirty but suspended. This case is likely
3374 * an extension because apps don't opt into dirty-tracking
3375 * and daemons aren't suspended.
3376 */
3377 #if DEVELOPMENT || DEBUG
3378 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
3379 os_log(OS_LOG_DEFAULT, "memorystatus: sending suspended process %s (pid %d) SIGKILL",
3380 (*p->p_name ? p->p_name : "unknown"), p->p_pid);
3381 }
3382 #endif /* DEVELOPMENT || DEBUG */
3383 sig = SIGKILL;
3384 } else {
3385 /* Dirty, terminated, or state tracking is unsupported; issue SIGTERM to allow cleanup */
3386 sig = SIGTERM;
3387 }
3388
3389 proc_list_unlock();
3390
3391 return sig;
3392 }
3393
3394 void
3395 memorystatus_on_suspend(proc_t p)
3396 {
3397 #if CONFIG_FREEZE
3398 uint32_t pages;
3399 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
3400 #endif
3401 proc_list_lock();
3402 #if CONFIG_FREEZE
3403 memorystatus_suspended_count++;
3404 #endif
3405 p->p_memstat_state |= P_MEMSTAT_SUSPENDED;
3406 proc_list_unlock();
3407 }
3408
3409 extern uint64_t memorystatus_thaw_count_since_boot;
3410
3411 void
3412 memorystatus_on_resume(proc_t p)
3413 {
3414 #if CONFIG_FREEZE
3415 boolean_t frozen;
3416 pid_t pid;
3417 #endif
3418
3419 proc_list_lock();
3420
3421 #if CONFIG_FREEZE
3422 frozen = (p->p_memstat_state & P_MEMSTAT_FROZEN);
3423 if (frozen) {
3424 /*
3425 * Now that we don't _thaw_ a process completely,
3426 * resuming it (and having some on-demand swapins)
3427 * shouldn't preclude it from being counted as frozen.
3428 *
3429 * memorystatus_frozen_count--;
3430 *
3431 * We preserve the P_MEMSTAT_FROZEN state since the process
3432 * could have state on disk AND so will deserve some protection
3433 * in the jetsam bands.
3434 */
3435 if ((p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) == 0) {
3436 p->p_memstat_state |= P_MEMSTAT_REFREEZE_ELIGIBLE;
3437 memorystatus_refreeze_eligible_count++;
3438 }
3439 if (p->p_memstat_thaw_count == 0 || p->p_memstat_last_thaw_interval < memorystatus_freeze_current_interval) {
3440 os_atomic_inc(&(memorystatus_freezer_stats.mfs_processes_thawed), relaxed);
3441 }
3442 p->p_memstat_last_thaw_interval = memorystatus_freeze_current_interval;
3443 p->p_memstat_thaw_count++;
3444
3445 memorystatus_thaw_count++;
3446 memorystatus_thaw_count_since_boot++;
3447 }
3448
3449 memorystatus_suspended_count--;
3450
3451 pid = p->p_pid;
3452 #endif
3453
3454 /*
3455 * P_MEMSTAT_FROZEN will remain unchanged. This used to be:
3456 * p->p_memstat_state &= ~(P_MEMSTAT_SUSPENDED | P_MEMSTAT_FROZEN);
3457 */
3458 p->p_memstat_state &= ~P_MEMSTAT_SUSPENDED;
3459
3460 proc_list_unlock();
3461
3462 #if CONFIG_FREEZE
3463 if (frozen) {
3464 memorystatus_freeze_entry_t data = { pid, FALSE, 0 };
3465 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
3466 }
3467 #endif
3468 }
3469
3470 void
3471 memorystatus_on_inactivity(proc_t p)
3472 {
3473 #pragma unused(p)
3474 #if CONFIG_FREEZE
3475 /* Wake the freeze thread */
3476 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
3477 #endif
3478 }
3479
3480 /*
3481 * The proc_list_lock is held by the caller.
3482 */
3483 static uint32_t
3484 memorystatus_build_state(proc_t p)
3485 {
3486 uint32_t snapshot_state = 0;
3487
3488 /* General */
3489 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
3490 snapshot_state |= kMemorystatusSuspended;
3491 }
3492 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
3493 snapshot_state |= kMemorystatusFrozen;
3494 }
3495 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
3496 snapshot_state |= kMemorystatusWasThawed;
3497 }
3498 if (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION) {
3499 snapshot_state |= kMemorystatusAssertion;
3500 }
3501
3502 /* Tracking */
3503 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3504 snapshot_state |= kMemorystatusTracked;
3505 }
3506 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3507 snapshot_state |= kMemorystatusSupportsIdleExit;
3508 }
3509 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3510 snapshot_state |= kMemorystatusDirty;
3511 }
3512
3513 return snapshot_state;
3514 }
3515
3516 static boolean_t
3517 kill_idle_exit_proc(void)
3518 {
3519 proc_t p, victim_p = PROC_NULL;
3520 uint64_t current_time, footprint_of_killed_proc;
3521 boolean_t killed = FALSE;
3522 unsigned int i = 0;
3523 os_reason_t jetsam_reason = OS_REASON_NULL;
3524
3525 /* Pick next idle exit victim. */
3526 current_time = mach_absolute_time();
3527
3528 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_IDLE_EXIT);
3529 if (jetsam_reason == OS_REASON_NULL) {
3530 printf("kill_idle_exit_proc: failed to allocate jetsam reason\n");
3531 }
3532
3533 proc_list_lock();
3534
3535 p = memorystatus_get_first_proc_locked(&i, FALSE);
3536 while (p) {
3537 /* No need to look beyond the idle band */
3538 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
3539 break;
3540 }
3541
3542 if ((p->p_memstat_dirty & (P_DIRTY_ALLOW_IDLE_EXIT | P_DIRTY_IS_DIRTY | P_DIRTY_TERMINATED)) == (P_DIRTY_ALLOW_IDLE_EXIT)) {
3543 if (current_time >= p->p_memstat_idledeadline) {
3544 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3545 victim_p = proc_ref_locked(p);
3546 break;
3547 }
3548 }
3549
3550 p = memorystatus_get_next_proc_locked(&i, p, FALSE);
3551 }
3552
3553 proc_list_unlock();
3554
3555 if (victim_p) {
3556 printf("memorystatus: killing_idle_process pid %d [%s] jetsam_reason->osr_code: %llu\n", victim_p->p_pid, (*victim_p->p_name ? victim_p->p_name : "unknown"), jetsam_reason->osr_code);
3557 killed = memorystatus_do_kill(victim_p, kMemorystatusKilledIdleExit, jetsam_reason, &footprint_of_killed_proc);
3558 proc_rele(victim_p);
3559 } else {
3560 os_reason_free(jetsam_reason);
3561 }
3562
3563 return killed;
3564 }
3565
3566 static void
3567 memorystatus_thread_wake(void)
3568 {
3569 int thr_id = 0;
3570 int active_thr = atomic_load(&active_jetsam_threads);
3571
3572 /* Wakeup all the jetsam threads */
3573 for (thr_id = 0; thr_id < active_thr; thr_id++) {
3574 thread_wakeup((event_t)&jetsam_threads[thr_id].memorystatus_wakeup);
3575 }
3576 }
3577
3578 #if CONFIG_JETSAM
3579
3580 static void
3581 memorystatus_thread_pool_max()
3582 {
3583 /* Increase the jetsam thread pool to max_jetsam_threads */
3584 int max_threads = max_jetsam_threads;
3585 printf("Expanding memorystatus pool to %d!\n", max_threads);
3586 atomic_store(&active_jetsam_threads, max_threads);
3587 }
3588
3589 static void
3590 memorystatus_thread_pool_default()
3591 {
3592 /* Restore the jetsam thread pool to a single thread */
3593 printf("Reverting memorystatus pool back to 1\n");
3594 atomic_store(&active_jetsam_threads, 1);
3595 }
3596
3597 #endif /* CONFIG_JETSAM */
3598
3599 extern void vm_pressure_response(void);
3600
3601 static int
3602 memorystatus_thread_block(uint32_t interval_ms, thread_continue_t continuation)
3603 {
3604 struct jetsam_thread_state *jetsam_thread = jetsam_current_thread();
3605
3606 assert(jetsam_thread != NULL);
3607 if (interval_ms) {
3608 assert_wait_timeout(&jetsam_thread->memorystatus_wakeup, THREAD_UNINT, interval_ms, NSEC_PER_MSEC);
3609 } else {
3610 assert_wait(&jetsam_thread->memorystatus_wakeup, THREAD_UNINT);
3611 }
3612
3613 return thread_block(continuation);
3614 }
3615
3616 static boolean_t
3617 memorystatus_avail_pages_below_pressure(void)
3618 {
3619 #if CONFIG_JETSAM
3620 return memorystatus_available_pages <= memorystatus_available_pages_pressure;
3621 #else /* CONFIG_JETSAM */
3622 return FALSE;
3623 #endif /* CONFIG_JETSAM */
3624 }
3625
3626 static boolean_t
3627 memorystatus_avail_pages_below_critical(void)
3628 {
3629 #if CONFIG_JETSAM
3630 return memorystatus_available_pages <= memorystatus_available_pages_critical;
3631 #else /* CONFIG_JETSAM */
3632 return FALSE;
3633 #endif /* CONFIG_JETSAM */
3634 }
3635
3636 static boolean_t
3637 memorystatus_post_snapshot(int32_t priority, uint32_t cause)
3638 {
3639 boolean_t is_idle_priority;
3640
3641 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
3642 is_idle_priority = (priority == JETSAM_PRIORITY_IDLE);
3643 } else {
3644 is_idle_priority = (priority == JETSAM_PRIORITY_IDLE || priority == JETSAM_PRIORITY_IDLE_DEFERRED);
3645 }
3646 #if CONFIG_JETSAM
3647 #pragma unused(cause)
3648 /*
3649 * Don't generate logs for steady-state idle-exit kills,
3650 * unless it is overridden for debug or by the device
3651 * tree.
3652 */
3653
3654 return !is_idle_priority || memorystatus_idle_snapshot;
3655
3656 #else /* CONFIG_JETSAM */
3657 /*
3658 * Don't generate logs for steady-state idle-exit kills,
3659 * unless
3660 * - it is overridden for debug or by the device
3661 * tree.
3662 * OR
3663 * - the kill causes are important i.e. not kMemorystatusKilledIdleExit
3664 */
3665
3666 boolean_t snapshot_eligible_kill_cause = (is_reason_thrashing(cause) || is_reason_zone_map_exhaustion(cause));
3667 return !is_idle_priority || memorystatus_idle_snapshot || snapshot_eligible_kill_cause;
3668 #endif /* CONFIG_JETSAM */
3669 }
3670
3671 static boolean_t
3672 memorystatus_action_needed(void)
3673 {
3674 #if CONFIG_JETSAM
3675 return is_reason_thrashing(kill_under_pressure_cause) ||
3676 is_reason_zone_map_exhaustion(kill_under_pressure_cause) ||
3677 memorystatus_available_pages <= memorystatus_available_pages_pressure;
3678 #else /* CONFIG_JETSAM */
3679 return is_reason_thrashing(kill_under_pressure_cause) ||
3680 is_reason_zone_map_exhaustion(kill_under_pressure_cause);
3681 #endif /* CONFIG_JETSAM */
3682 }
3683
3684 static boolean_t
3685 memorystatus_act_on_hiwat_processes(uint32_t *errors, uint32_t *hwm_kill, boolean_t *post_snapshot, __unused boolean_t *is_critical, uint64_t *memory_reclaimed)
3686 {
3687 boolean_t purged = FALSE, killed = FALSE;
3688
3689 *memory_reclaimed = 0;
3690 killed = memorystatus_kill_hiwat_proc(errors, &purged, memory_reclaimed);
3691
3692 if (killed) {
3693 *hwm_kill = *hwm_kill + 1;
3694 *post_snapshot = TRUE;
3695 return TRUE;
3696 } else {
3697 if (purged == FALSE) {
3698 /* couldn't purge and couldn't kill */
3699 memorystatus_hwm_candidates = FALSE;
3700 }
3701 }
3702
3703 #if CONFIG_JETSAM
3704 /* No highwater processes to kill. Continue or stop for now? */
3705 if (!is_reason_thrashing(kill_under_pressure_cause) &&
3706 !is_reason_zone_map_exhaustion(kill_under_pressure_cause) &&
3707 (memorystatus_available_pages > memorystatus_available_pages_critical)) {
3708 /*
3709 * We are _not_ out of pressure but we are above the critical threshold and there's:
3710 * - no compressor thrashing
3711 * - enough zone memory
3712 * - no more HWM processes left.
3713 * For now, don't kill any other processes.
3714 */
3715
3716 if (*hwm_kill == 0) {
3717 memorystatus_thread_wasted_wakeup++;
3718 }
3719
3720 *is_critical = FALSE;
3721
3722 return TRUE;
3723 }
3724 #endif /* CONFIG_JETSAM */
3725
3726 return FALSE;
3727 }
3728
3729 /*
3730 * kJetsamHighRelaunchCandidatesThreshold defines the percentage of candidates
3731 * in the idle & deferred bands that need to be bad candidates in order to trigger
3732 * aggressive jetsam.
3733 */
3734 #define kJetsamHighRelaunchCandidatesThreshold (100)
3735
3736 /* kJetsamMinCandidatesThreshold defines the minimum number of candidates in the
3737 * idle/deferred bands to trigger aggressive jetsam. This value basically decides
3738 * how much memory the system is ready to hold in the lower bands without triggering
3739 * aggressive jetsam. This number should ideally be tuned based on the memory config
3740 * of the device.
3741 */
3742 #define kJetsamMinCandidatesThreshold (5)
3743
3744 static boolean_t
3745 memorystatus_aggressive_jetsam_needed_sysproc_aging(__unused int jld_eval_aggressive_count, __unused int *jld_idle_kills, __unused int jld_idle_kill_candidates, int *total_candidates, int *elevated_bucket_count)
3746 {
3747 boolean_t aggressive_jetsam_needed = false;
3748
3749 /*
3750 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, we maintain the jetsam
3751 * relaunch behavior for all daemons. Also, daemons and apps are aged in deferred bands on
3752 * every dirty->clean transition. For this aging policy, the best way to determine if
3753 * aggressive jetsam is needed, is to see if the kill candidates are mostly bad candidates.
3754 * If yes, then we need to go to higher bands to reclaim memory.
3755 */
3756 proc_list_lock();
3757 /* Get total candidate counts for idle and idle deferred bands */
3758 *total_candidates = memstat_bucket[JETSAM_PRIORITY_IDLE].count + memstat_bucket[system_procs_aging_band].count;
3759 /* Get counts of bad kill candidates in idle and idle deferred bands */
3760 int bad_candidates = memstat_bucket[JETSAM_PRIORITY_IDLE].relaunch_high_count + memstat_bucket[system_procs_aging_band].relaunch_high_count;
3761
3762 *elevated_bucket_count = memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE].count;
3763
3764 proc_list_unlock();
3765
3766 /* Check if the number of bad candidates is greater than kJetsamHighRelaunchCandidatesThreshold % */
3767 aggressive_jetsam_needed = (((bad_candidates * 100) / *total_candidates) >= kJetsamHighRelaunchCandidatesThreshold);
3768
3769 /*
3770 * Since the new aging policy bases the aggressive jetsam trigger on percentage of
3771 * bad candidates, it is prone to being overly aggressive. In order to mitigate that,
3772 * make sure the system is really under memory pressure before triggering aggressive
3773 * jetsam.
3774 */
3775 if (memorystatus_available_pages > memorystatus_sysproc_aging_aggr_pages) {
3776 aggressive_jetsam_needed = false;
3777 }
3778
3779 #if DEVELOPMENT || DEBUG
3780 printf("memorystatus: aggressive%d: [%s] Bad Candidate Threshold Check (total: %d, bad: %d, threshold: %d %%); Memory Pressure Check (available_pgs: %llu, threshold_pgs: %llu)\n",
3781 jld_eval_aggressive_count, aggressive_jetsam_needed ? "PASSED" : "FAILED", *total_candidates, bad_candidates,
3782 kJetsamHighRelaunchCandidatesThreshold, (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES, (uint64_t)memorystatus_sysproc_aging_aggr_pages);
3783 #endif /* DEVELOPMENT || DEBUG */
3784 return aggressive_jetsam_needed;
3785 }
3786
3787 /*
3788 * Gets memory back from various system caches.
3789 * Called before jetsamming in the foreground band in the hope that we'll
3790 * avoid a jetsam.
3791 */
3792 static void
3793 memorystatus_approaching_fg_band(boolean_t *corpse_list_purged)
3794 {
3795 assert(corpse_list_purged != NULL);
3796 pmap_release_pages_fast();
3797 memorystatus_issue_fg_band_notify();
3798 if (total_corpses_count() > 0 && !*corpse_list_purged) {
3799 task_purge_all_corpses();
3800 *corpse_list_purged = TRUE;
3801 }
3802 }
3803
3804 static boolean_t
3805 memorystatus_aggressive_jetsam_needed_default(__unused int jld_eval_aggressive_count, int *jld_idle_kills, int jld_idle_kill_candidates, int *total_candidates, int *elevated_bucket_count)
3806 {
3807 boolean_t aggressive_jetsam_needed = false;
3808 /* Jetsam Loop Detection - locals */
3809 memstat_bucket_t *bucket;
3810 int jld_bucket_count = 0;
3811
3812 proc_list_lock();
3813 switch (jetsam_aging_policy) {
3814 case kJetsamAgingPolicyLegacy:
3815 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3816 jld_bucket_count = bucket->count;
3817 bucket = &memstat_bucket[JETSAM_PRIORITY_AGING_BAND1];
3818 jld_bucket_count += bucket->count;
3819 break;
3820 case kJetsamAgingPolicyAppsReclaimedFirst:
3821 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3822 jld_bucket_count = bucket->count;
3823 bucket = &memstat_bucket[system_procs_aging_band];
3824 jld_bucket_count += bucket->count;
3825 bucket = &memstat_bucket[applications_aging_band];
3826 jld_bucket_count += bucket->count;
3827 break;
3828 case kJetsamAgingPolicyNone:
3829 default:
3830 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3831 jld_bucket_count = bucket->count;
3832 break;
3833 }
3834
3835 bucket = &memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE];
3836 *elevated_bucket_count = bucket->count;
3837 *total_candidates = jld_bucket_count;
3838 proc_list_unlock();
3839
3840 aggressive_jetsam_needed = (*jld_idle_kills > jld_idle_kill_candidates);
3841
3842 #if DEVELOPMENT || DEBUG
3843 if (aggressive_jetsam_needed) {
3844 printf("memorystatus: aggressive%d: idle candidates: %d, idle kills: %d\n",
3845 jld_eval_aggressive_count,
3846 jld_idle_kill_candidates,
3847 *jld_idle_kills);
3848 }
3849 #endif /* DEVELOPMENT || DEBUG */
3850 return aggressive_jetsam_needed;
3851 }
3852
3853 static boolean_t
3854 memorystatus_act_aggressive(uint32_t cause, os_reason_t jetsam_reason, int *jld_idle_kills, boolean_t *corpse_list_purged, boolean_t *post_snapshot, uint64_t *memory_reclaimed)
3855 {
3856 boolean_t aggressive_jetsam_needed = false;
3857 boolean_t killed;
3858 uint32_t errors = 0;
3859 uint64_t footprint_of_killed_proc = 0;
3860 int elevated_bucket_count = 0;
3861 int total_candidates = 0;
3862 *memory_reclaimed = 0;
3863
3864 /*
3865 * The aggressive jetsam logic looks at the number of times it has been in the
3866 * aggressive loop to determine the max priority band it should kill upto. The
3867 * static variables below are used to track that property.
3868 *
3869 * To reset those values, the implementation checks if it has been
3870 * memorystatus_jld_eval_period_msecs since the parameters were reset.
3871 */
3872 static int jld_eval_aggressive_count = 0;
3873 static int32_t jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
3874 static uint64_t jld_timestamp_msecs = 0;
3875 static int jld_idle_kill_candidates = 0;
3876
3877 if (memorystatus_jld_enabled == FALSE) {
3878 /* If aggressive jetsam is disabled, nothing to do here */
3879 return FALSE;
3880 }
3881
3882 /* Get current timestamp (msecs only) */
3883 struct timeval jld_now_tstamp = {0, 0};
3884 uint64_t jld_now_msecs = 0;
3885 microuptime(&jld_now_tstamp);
3886 jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
3887
3888 /*
3889 * The aggressive jetsam logic looks at the number of candidates and their
3890 * properties to decide if aggressive jetsam should be engaged.
3891 */
3892 if (jetsam_aging_policy == kJetsamAgingPolicySysProcsReclaimedFirst) {
3893 /*
3894 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, the logic looks at the number of
3895 * candidates in the idle and deferred band and how many out of them are marked as high relaunch
3896 * probability.
3897 */
3898 aggressive_jetsam_needed = memorystatus_aggressive_jetsam_needed_sysproc_aging(jld_eval_aggressive_count,
3899 jld_idle_kills, jld_idle_kill_candidates, &total_candidates, &elevated_bucket_count);
3900 } else {
3901 /*
3902 * The other aging policies look at number of candidate processes over a specific time window and
3903 * evaluate if the system is in a jetsam loop. If yes, aggressive jetsam is triggered.
3904 */
3905 aggressive_jetsam_needed = memorystatus_aggressive_jetsam_needed_default(jld_eval_aggressive_count,
3906 jld_idle_kills, jld_idle_kill_candidates, &total_candidates, &elevated_bucket_count);
3907 }
3908
3909 /*
3910 * Check if its been really long since the aggressive jetsam evaluation
3911 * parameters have been refreshed. This logic also resets the jld_eval_aggressive_count
3912 * counter to make sure we reset the aggressive jetsam severity.
3913 */
3914 boolean_t param_reval = false;
3915
3916 if ((total_candidates == 0) ||
3917 (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
3918 jld_timestamp_msecs = jld_now_msecs;
3919 jld_idle_kill_candidates = total_candidates;
3920 *jld_idle_kills = 0;
3921 jld_eval_aggressive_count = 0;
3922 jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
3923 param_reval = true;
3924 }
3925
3926 /*
3927 * If the parameters have been updated, re-evaluate the aggressive_jetsam_needed condition for
3928 * the non kJetsamAgingPolicySysProcsReclaimedFirst policy since its based on jld_idle_kill_candidates etc.
3929 */
3930 if ((param_reval == true) && (jetsam_aging_policy != kJetsamAgingPolicySysProcsReclaimedFirst)) {
3931 aggressive_jetsam_needed = (*jld_idle_kills > jld_idle_kill_candidates);
3932 }
3933
3934 /*
3935 * It is also possible that the system is down to a very small number of processes in the candidate
3936 * bands. In that case, the decisions made by the memorystatus_aggressive_jetsam_needed_* routines
3937 * would not be useful. In that case, do not trigger aggressive jetsam.
3938 */
3939 if (total_candidates < kJetsamMinCandidatesThreshold) {
3940 #if DEVELOPMENT || DEBUG
3941 printf("memorystatus: aggressive: [FAILED] Low Candidate Count (current: %d, threshold: %d)\n", total_candidates, kJetsamMinCandidatesThreshold);
3942 #endif /* DEVELOPMENT || DEBUG */
3943 aggressive_jetsam_needed = false;
3944 }
3945
3946 if (aggressive_jetsam_needed == false) {
3947 /* Either the aging policy or the candidate count decided that aggressive jetsam is not needed. Nothing more to do here. */
3948 return FALSE;
3949 }
3950
3951 /* Looks like aggressive jetsam is needed */
3952 jld_eval_aggressive_count++;
3953
3954 if (jld_eval_aggressive_count == memorystatus_jld_eval_aggressive_count) {
3955 memorystatus_approaching_fg_band(corpse_list_purged);
3956 } else if (jld_eval_aggressive_count > memorystatus_jld_eval_aggressive_count) {
3957 /*
3958 * Bump up the jetsam priority limit (eg: the bucket index)
3959 * Enforce bucket index sanity.
3960 */
3961 if ((memorystatus_jld_eval_aggressive_priority_band_max < 0) ||
3962 (memorystatus_jld_eval_aggressive_priority_band_max >= MEMSTAT_BUCKET_COUNT)) {
3963 /*
3964 * Do nothing. Stick with the default level.
3965 */
3966 } else {
3967 jld_priority_band_max = memorystatus_jld_eval_aggressive_priority_band_max;
3968 }
3969 }
3970
3971 /* Visit elevated processes first */
3972 while (elevated_bucket_count) {
3973 elevated_bucket_count--;
3974
3975 /*
3976 * memorystatus_kill_elevated_process() drops a reference,
3977 * so take another one so we can continue to use this exit reason
3978 * even after it returns.
3979 */
3980
3981 os_reason_ref(jetsam_reason);
3982 killed = memorystatus_kill_elevated_process(
3983 cause,
3984 jetsam_reason,
3985 JETSAM_PRIORITY_ELEVATED_INACTIVE,
3986 jld_eval_aggressive_count,
3987 &errors, &footprint_of_killed_proc);
3988 if (killed) {
3989 *post_snapshot = TRUE;
3990 *memory_reclaimed += footprint_of_killed_proc;
3991 if (memorystatus_avail_pages_below_pressure()) {
3992 /*
3993 * Still under pressure.
3994 * Find another pinned processes.
3995 */
3996 continue;
3997 } else {
3998 return TRUE;
3999 }
4000 } else {
4001 /*
4002 * No pinned processes left to kill.
4003 * Abandon elevated band.
4004 */
4005 break;
4006 }
4007 }
4008
4009 /*
4010 * memorystatus_kill_processes_aggressive() allocates its own
4011 * jetsam_reason so the kMemorystatusKilledProcThrashing cause
4012 * is consistent throughout the aggressive march.
4013 */
4014 killed = memorystatus_kill_processes_aggressive(
4015 kMemorystatusKilledProcThrashing,
4016 jld_eval_aggressive_count,
4017 jld_priority_band_max,
4018 &errors, &footprint_of_killed_proc);
4019
4020 if (killed) {
4021 /* Always generate logs after aggressive kill */
4022 *post_snapshot = TRUE;
4023 *memory_reclaimed += footprint_of_killed_proc;
4024 *jld_idle_kills = 0;
4025 return TRUE;
4026 }
4027
4028 return FALSE;
4029 }
4030
4031
4032 static void
4033 memorystatus_thread(void *param __unused, wait_result_t wr __unused)
4034 {
4035 boolean_t post_snapshot = FALSE;
4036 uint32_t errors = 0;
4037 uint32_t hwm_kill = 0;
4038 boolean_t sort_flag = TRUE;
4039 boolean_t corpse_list_purged = FALSE;
4040 int jld_idle_kills = 0;
4041 struct jetsam_thread_state *jetsam_thread = jetsam_current_thread();
4042 uint64_t total_memory_reclaimed = 0;
4043
4044 assert(jetsam_thread != NULL);
4045 if (jetsam_thread->inited == FALSE) {
4046 /*
4047 * It's the first time the thread has run, so just mark the thread as privileged and block.
4048 * This avoids a spurious pass with unset variables, as set out in <rdar://problem/9609402>.
4049 */
4050
4051 char name[32];
4052 thread_wire(host_priv_self(), current_thread(), TRUE);
4053 snprintf(name, 32, "VM_memorystatus_%d", jetsam_thread->index + 1);
4054
4055 /* Limit all but one thread to the lower jetsam bands, as that's where most of the victims are. */
4056 if (jetsam_thread->index == 0) {
4057 if (vm_pageout_state.vm_restricted_to_single_processor == TRUE) {
4058 thread_vm_bind_group_add();
4059 }
4060 jetsam_thread->limit_to_low_bands = FALSE;
4061 } else {
4062 jetsam_thread->limit_to_low_bands = TRUE;
4063 }
4064 #if CONFIG_THREAD_GROUPS
4065 thread_group_vm_add();
4066 #endif
4067 thread_set_thread_name(current_thread(), name);
4068 jetsam_thread->inited = TRUE;
4069 memorystatus_thread_block(0, memorystatus_thread);
4070 }
4071
4072 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_START,
4073 MEMORYSTATUS_LOG_AVAILABLE_PAGES, memorystatus_jld_enabled, memorystatus_jld_eval_period_msecs, memorystatus_jld_eval_aggressive_count, 0);
4074
4075 /*
4076 * Jetsam aware version.
4077 *
4078 * The VM pressure notification thread is working it's way through clients in parallel.
4079 *
4080 * So, while the pressure notification thread is targeting processes in order of
4081 * increasing jetsam priority, we can hopefully reduce / stop it's work by killing
4082 * any processes that have exceeded their highwater mark.
4083 *
4084 * If we run out of HWM processes and our available pages drops below the critical threshold, then,
4085 * we target the least recently used process in order of increasing jetsam priority (exception: the FG band).
4086 */
4087 while (memorystatus_action_needed()) {
4088 boolean_t killed;
4089 int32_t priority;
4090 uint32_t cause;
4091 uint64_t memory_reclaimed = 0;
4092 uint64_t jetsam_reason_code = JETSAM_REASON_INVALID;
4093 os_reason_t jetsam_reason = OS_REASON_NULL;
4094
4095 cause = kill_under_pressure_cause;
4096 switch (cause) {
4097 case kMemorystatusKilledFCThrashing:
4098 jetsam_reason_code = JETSAM_REASON_MEMORY_FCTHRASHING;
4099 break;
4100 case kMemorystatusKilledVMCompressorThrashing:
4101 jetsam_reason_code = JETSAM_REASON_MEMORY_VMCOMPRESSOR_THRASHING;
4102 break;
4103 case kMemorystatusKilledVMCompressorSpaceShortage:
4104 jetsam_reason_code = JETSAM_REASON_MEMORY_VMCOMPRESSOR_SPACE_SHORTAGE;
4105 break;
4106 case kMemorystatusKilledZoneMapExhaustion:
4107 jetsam_reason_code = JETSAM_REASON_ZONE_MAP_EXHAUSTION;
4108 break;
4109 case kMemorystatusKilledVMPageShortage:
4110 /* falls through */
4111 default:
4112 jetsam_reason_code = JETSAM_REASON_MEMORY_VMPAGESHORTAGE;
4113 cause = kMemorystatusKilledVMPageShortage;
4114 break;
4115 }
4116
4117 /* Highwater */
4118 boolean_t is_critical = TRUE;
4119 if (memorystatus_act_on_hiwat_processes(&errors, &hwm_kill, &post_snapshot, &is_critical, &memory_reclaimed)) {
4120 total_memory_reclaimed += memory_reclaimed;
4121 if (is_critical == FALSE) {
4122 /*
4123 * For now, don't kill any other processes.
4124 */
4125 break;
4126 } else {
4127 goto done;
4128 }
4129 }
4130
4131 jetsam_reason = os_reason_create(OS_REASON_JETSAM, jetsam_reason_code);
4132 if (jetsam_reason == OS_REASON_NULL) {
4133 printf("memorystatus_thread: failed to allocate jetsam reason\n");
4134 }
4135
4136 /* Only unlimited jetsam threads should act aggressive */
4137 if (!jetsam_thread->limit_to_low_bands &&
4138 memorystatus_act_aggressive(cause, jetsam_reason, &jld_idle_kills, &corpse_list_purged, &post_snapshot, &memory_reclaimed)) {
4139 total_memory_reclaimed += memory_reclaimed;
4140 goto done;
4141 }
4142
4143 /*
4144 * memorystatus_kill_top_process() drops a reference,
4145 * so take another one so we can continue to use this exit reason
4146 * even after it returns
4147 */
4148 os_reason_ref(jetsam_reason);
4149
4150 /* LRU */
4151 killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, jetsam_reason, &priority, &errors, &memory_reclaimed);
4152 sort_flag = FALSE;
4153
4154 if (killed) {
4155 total_memory_reclaimed += memory_reclaimed;
4156 if (memorystatus_post_snapshot(priority, cause) == TRUE) {
4157 post_snapshot = TRUE;
4158 }
4159
4160 /* Jetsam Loop Detection */
4161 if (memorystatus_jld_enabled == TRUE) {
4162 if ((priority == JETSAM_PRIORITY_IDLE) || (priority == system_procs_aging_band) || (priority == applications_aging_band)) {
4163 jld_idle_kills++;
4164 } else {
4165 /*
4166 * We've reached into bands beyond idle deferred.
4167 * We make no attempt to monitor them
4168 */
4169 }
4170 }
4171
4172 /*
4173 * If we have jetsammed a process in or above JETSAM_PRIORITY_UI_SUPPORT
4174 * then we attempt to relieve pressure by purging corpse memory and notifying
4175 * anybody wanting to know this.
4176 */
4177 if (priority >= JETSAM_PRIORITY_UI_SUPPORT) {
4178 memorystatus_approaching_fg_band(&corpse_list_purged);
4179 }
4180 goto done;
4181 }
4182
4183 if (memorystatus_avail_pages_below_critical()) {
4184 /*
4185 * Still under pressure and unable to kill a process - purge corpse memory
4186 * and get everything back from the pmap.
4187 */
4188 pmap_release_pages_fast();
4189 if (total_corpses_count() > 0) {
4190 task_purge_all_corpses();
4191 corpse_list_purged = TRUE;
4192 }
4193
4194 if (!jetsam_thread->limit_to_low_bands && memorystatus_avail_pages_below_critical()) {
4195 /*
4196 * Still under pressure and unable to kill a process - panic
4197 */
4198 panic("memorystatus_jetsam_thread: no victim! available pages:%llu\n", (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES);
4199 }
4200 }
4201
4202 done:
4203
4204 /*
4205 * We do not want to over-kill when thrashing has been detected.
4206 * To avoid that, we reset the flag here and notify the
4207 * compressor.
4208 */
4209 if (is_reason_thrashing(kill_under_pressure_cause)) {
4210 kill_under_pressure_cause = 0;
4211 #if CONFIG_JETSAM
4212 vm_thrashing_jetsam_done();
4213 #endif /* CONFIG_JETSAM */
4214 } else if (is_reason_zone_map_exhaustion(kill_under_pressure_cause)) {
4215 kill_under_pressure_cause = 0;
4216 }
4217
4218 os_reason_free(jetsam_reason);
4219 }
4220
4221 kill_under_pressure_cause = 0;
4222
4223 if (errors) {
4224 memorystatus_clear_errors();
4225 }
4226
4227 if (post_snapshot) {
4228 proc_list_lock();
4229 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4230 sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
4231 uint64_t timestamp_now = mach_absolute_time();
4232 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4233 memorystatus_jetsam_snapshot->js_gencount++;
4234 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4235 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4236 proc_list_unlock();
4237 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4238 if (!ret) {
4239 proc_list_lock();
4240 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4241 proc_list_unlock();
4242 }
4243 } else {
4244 proc_list_unlock();
4245 }
4246 }
4247
4248 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_END,
4249 MEMORYSTATUS_LOG_AVAILABLE_PAGES, total_memory_reclaimed, 0, 0, 0);
4250
4251 memorystatus_thread_block(0, memorystatus_thread);
4252 }
4253
4254 /*
4255 * Returns TRUE:
4256 * when an idle-exitable proc was killed
4257 * Returns FALSE:
4258 * when there are no more idle-exitable procs found
4259 * when the attempt to kill an idle-exitable proc failed
4260 */
4261 boolean_t
4262 memorystatus_idle_exit_from_VM(void)
4263 {
4264 /*
4265 * This routine should no longer be needed since we are
4266 * now using jetsam bands on all platforms and so will deal
4267 * with IDLE processes within the memorystatus thread itself.
4268 *
4269 * But we still use it because we observed that macos systems
4270 * started heavy compression/swapping with a bunch of
4271 * idle-exitable processes alive and doing nothing. We decided
4272 * to rather kill those processes than start swapping earlier.
4273 */
4274
4275 return kill_idle_exit_proc();
4276 }
4277
4278 /*
4279 * Callback invoked when allowable physical memory footprint exceeded
4280 * (dirty pages + IOKit mappings)
4281 *
4282 * This is invoked for both advisory, non-fatal per-task high watermarks,
4283 * as well as the fatal task memory limits.
4284 */
4285 void
4286 memorystatus_on_ledger_footprint_exceeded(boolean_t warning, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
4287 {
4288 os_reason_t jetsam_reason = OS_REASON_NULL;
4289
4290 proc_t p = current_proc();
4291
4292 #if VM_PRESSURE_EVENTS
4293 if (warning == TRUE) {
4294 /*
4295 * This is a warning path which implies that the current process is close, but has
4296 * not yet exceeded its per-process memory limit.
4297 */
4298 if (memorystatus_warn_process(p, memlimit_is_active, memlimit_is_fatal, FALSE /* not exceeded */) != TRUE) {
4299 /* Print warning, since it's possible that task has not registered for pressure notifications */
4300 os_log(OS_LOG_DEFAULT, "memorystatus_on_ledger_footprint_exceeded: failed to warn the current task (%d exiting, or no handler registered?).\n", p->p_pid);
4301 }
4302 return;
4303 }
4304 #endif /* VM_PRESSURE_EVENTS */
4305
4306 if (memlimit_is_fatal) {
4307 /*
4308 * If this process has no high watermark or has a fatal task limit, then we have been invoked because the task
4309 * has violated either the system-wide per-task memory limit OR its own task limit.
4310 */
4311 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_PERPROCESSLIMIT);
4312 if (jetsam_reason == NULL) {
4313 printf("task_exceeded footprint: failed to allocate jetsam reason\n");
4314 } else if (corpse_for_fatal_memkill != 0 && proc_send_synchronous_EXC_RESOURCE(p) == FALSE) {
4315 /* Set OS_REASON_FLAG_GENERATE_CRASH_REPORT to generate corpse */
4316 jetsam_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
4317 }
4318
4319 if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit, jetsam_reason) != TRUE) {
4320 printf("task_exceeded_footprint: failed to kill the current task (exiting?).\n");
4321 }
4322 } else {
4323 /*
4324 * HWM offender exists. Done without locks or synchronization.
4325 * See comment near its declaration for more details.
4326 */
4327 memorystatus_hwm_candidates = TRUE;
4328
4329 #if VM_PRESSURE_EVENTS
4330 /*
4331 * The current process is not in the warning path.
4332 * This path implies the current process has exceeded a non-fatal (soft) memory limit.
4333 * Failure to send note is ignored here.
4334 */
4335 (void)memorystatus_warn_process(p, memlimit_is_active, memlimit_is_fatal, TRUE /* exceeded */);
4336
4337 #endif /* VM_PRESSURE_EVENTS */
4338 }
4339 }
4340
4341 void
4342 memorystatus_log_exception(const int max_footprint_mb, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
4343 {
4344 proc_t p = current_proc();
4345
4346 /*
4347 * The limit violation is logged here, but only once per process per limit.
4348 * Soft memory limit is a non-fatal high-water-mark
4349 * Hard memory limit is a fatal custom-task-limit or system-wide per-task memory limit.
4350 */
4351
4352 os_log_with_startup_serial(OS_LOG_DEFAULT, "EXC_RESOURCE -> %s[%d] exceeded mem limit: %s%s %d MB (%s)\n",
4353 ((p && *p->p_name) ? p->p_name : "unknown"), (p ? p->p_pid : -1), (memlimit_is_active ? "Active" : "Inactive"),
4354 (memlimit_is_fatal ? "Hard" : "Soft"), max_footprint_mb,
4355 (memlimit_is_fatal ? "fatal" : "non-fatal"));
4356
4357 return;
4358 }
4359
4360
4361 /*
4362 * Description:
4363 * Evaluates process state to determine which limit
4364 * should be applied (active vs. inactive limit).
4365 *
4366 * Processes that have the 'elevated inactive jetsam band' attribute
4367 * are first evaluated based on their current priority band.
4368 * presently elevated ==> active
4369 *
4370 * Processes that opt into dirty tracking are evaluated
4371 * based on clean vs dirty state.
4372 * dirty ==> active
4373 * clean ==> inactive
4374 *
4375 * Process that do not opt into dirty tracking are
4376 * evalulated based on priority level.
4377 * Foreground or above ==> active
4378 * Below Foreground ==> inactive
4379 *
4380 * Return: TRUE if active
4381 * False if inactive
4382 */
4383
4384 static boolean_t
4385 proc_jetsam_state_is_active_locked(proc_t p)
4386 {
4387 if ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) &&
4388 (p->p_memstat_effectivepriority == JETSAM_PRIORITY_ELEVATED_INACTIVE)) {
4389 /*
4390 * process has the 'elevated inactive jetsam band' attribute
4391 * and process is present in the elevated band
4392 * implies active state
4393 */
4394 return TRUE;
4395 } else if (p->p_memstat_dirty & P_DIRTY_TRACK) {
4396 /*
4397 * process has opted into dirty tracking
4398 * active state is based on dirty vs. clean
4399 */
4400 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
4401 /*
4402 * process is dirty
4403 * implies active state
4404 */
4405 return TRUE;
4406 } else {
4407 /*
4408 * process is clean
4409 * implies inactive state
4410 */
4411 return FALSE;
4412 }
4413 } else if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
4414 /*
4415 * process is Foreground or higher
4416 * implies active state
4417 */
4418 return TRUE;
4419 } else {
4420 /*
4421 * process found below Foreground
4422 * implies inactive state
4423 */
4424 return FALSE;
4425 }
4426 }
4427
4428 static boolean_t
4429 memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason)
4430 {
4431 boolean_t res;
4432
4433 uint32_t errors = 0;
4434 uint64_t memory_reclaimed = 0;
4435
4436 if (victim_pid == -1) {
4437 /* No pid, so kill first process */
4438 res = memorystatus_kill_top_process(TRUE, TRUE, cause, jetsam_reason, NULL, &errors, &memory_reclaimed);
4439 } else {
4440 res = memorystatus_kill_specific_process(victim_pid, cause, jetsam_reason);
4441 }
4442
4443 if (errors) {
4444 memorystatus_clear_errors();
4445 }
4446
4447 if (res == TRUE) {
4448 /* Fire off snapshot notification */
4449 proc_list_lock();
4450 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4451 sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
4452 uint64_t timestamp_now = mach_absolute_time();
4453 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4454 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4455 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4456 proc_list_unlock();
4457 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4458 if (!ret) {
4459 proc_list_lock();
4460 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4461 proc_list_unlock();
4462 }
4463 } else {
4464 proc_list_unlock();
4465 }
4466 }
4467
4468 return res;
4469 }
4470
4471 /*
4472 * Jetsam a specific process.
4473 */
4474 static boolean_t
4475 memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason)
4476 {
4477 boolean_t killed;
4478 proc_t p;
4479 uint64_t killtime = 0;
4480 uint64_t footprint_of_killed_proc;
4481 clock_sec_t tv_sec;
4482 clock_usec_t tv_usec;
4483 uint32_t tv_msec;
4484
4485 /* TODO - add a victim queue and push this into the main jetsam thread */
4486
4487 p = proc_find(victim_pid);
4488 if (!p) {
4489 os_reason_free(jetsam_reason);
4490 return FALSE;
4491 }
4492
4493 proc_list_lock();
4494
4495 if (memorystatus_jetsam_snapshot_count == 0) {
4496 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
4497 }
4498
4499 killtime = mach_absolute_time();
4500 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
4501 tv_msec = tv_usec / 1000;
4502
4503 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
4504
4505 proc_list_unlock();
4506
4507 killed = memorystatus_do_kill(p, cause, jetsam_reason, &footprint_of_killed_proc);
4508
4509 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_specific_process pid %d [%s] (%s %d) %lluKB - memorystatus_available_pages: %llu\n",
4510 (unsigned long)tv_sec, tv_msec, victim_pid, ((p && *p->p_name) ? p->p_name : "unknown"),
4511 memorystatus_kill_cause_name[cause], (p ? p->p_memstat_effectivepriority: -1),
4512 footprint_of_killed_proc >> 10, (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES);
4513
4514 proc_rele(p);
4515
4516 return killed;
4517 }
4518
4519
4520 /*
4521 * Toggle the P_MEMSTAT_TERMINATED state.
4522 * Takes the proc_list_lock.
4523 */
4524 void
4525 proc_memstat_terminated(proc_t p, boolean_t set)
4526 {
4527 #if DEVELOPMENT || DEBUG
4528 if (p) {
4529 proc_list_lock();
4530 if (set == TRUE) {
4531 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
4532 } else {
4533 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
4534 }
4535 proc_list_unlock();
4536 }
4537 #else
4538 #pragma unused(p, set)
4539 /*
4540 * do nothing
4541 */
4542 #endif /* DEVELOPMENT || DEBUG */
4543 return;
4544 }
4545
4546
4547 #if CONFIG_JETSAM
4548 /*
4549 * This is invoked when cpulimits have been exceeded while in fatal mode.
4550 * The jetsam_flags do not apply as those are for memory related kills.
4551 * We call this routine so that the offending process is killed with
4552 * a non-zero exit status.
4553 */
4554 void
4555 jetsam_on_ledger_cpulimit_exceeded(void)
4556 {
4557 int retval = 0;
4558 int jetsam_flags = 0; /* make it obvious */
4559 proc_t p = current_proc();
4560 os_reason_t jetsam_reason = OS_REASON_NULL;
4561
4562 printf("task_exceeded_cpulimit: killing pid %d [%s]\n",
4563 p->p_pid, (*p->p_name ? p->p_name : "(unknown)"));
4564
4565 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_CPULIMIT);
4566 if (jetsam_reason == OS_REASON_NULL) {
4567 printf("task_exceeded_cpulimit: unable to allocate memory for jetsam reason\n");
4568 }
4569
4570 retval = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
4571
4572 if (retval) {
4573 printf("task_exceeded_cpulimit: failed to kill current task (exiting?).\n");
4574 }
4575 }
4576
4577 #endif /* CONFIG_JETSAM */
4578
4579 static void
4580 memorystatus_get_task_memory_region_count(task_t task, uint64_t *count)
4581 {
4582 assert(task);
4583 assert(count);
4584
4585 *count = get_task_memory_region_count(task);
4586 }
4587
4588
4589 #define MEMORYSTATUS_VM_MAP_FORK_ALLOWED 0x100000000
4590 #define MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED 0x200000000
4591
4592 #if DEVELOPMENT || DEBUG
4593
4594 /*
4595 * Sysctl only used to test memorystatus_allowed_vm_map_fork() path.
4596 * set a new pidwatch value
4597 * or
4598 * get the current pidwatch value
4599 *
4600 * The pidwatch_val starts out with a PID to watch for in the map_fork path.
4601 * Its value is:
4602 * - OR'd with MEMORYSTATUS_VM_MAP_FORK_ALLOWED if we allow the map_fork.
4603 * - OR'd with MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED if we disallow the map_fork.
4604 * - set to -1ull if the map_fork() is aborted for other reasons.
4605 */
4606
4607 uint64_t memorystatus_vm_map_fork_pidwatch_val = 0;
4608
4609 static int sysctl_memorystatus_vm_map_fork_pidwatch SYSCTL_HANDLER_ARGS {
4610 #pragma unused(oidp, arg1, arg2)
4611
4612 uint64_t new_value = 0;
4613 uint64_t old_value = 0;
4614 int error = 0;
4615
4616 /*
4617 * The pid is held in the low 32 bits.
4618 * The 'allowed' flags are in the upper 32 bits.
4619 */
4620 old_value = memorystatus_vm_map_fork_pidwatch_val;
4621
4622 error = sysctl_io_number(req, old_value, sizeof(old_value), &new_value, NULL);
4623
4624 if (error || !req->newptr) {
4625 /*
4626 * No new value passed in.
4627 */
4628 return error;
4629 }
4630
4631 /*
4632 * A new pid was passed in via req->newptr.
4633 * Ignore any attempt to set the higher order bits.
4634 */
4635 memorystatus_vm_map_fork_pidwatch_val = new_value & 0xFFFFFFFF;
4636 printf("memorystatus: pidwatch old_value = 0x%llx, new_value = 0x%llx \n", old_value, new_value);
4637
4638 return error;
4639 }
4640
4641 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_map_fork_pidwatch, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_MASKED,
4642 0, 0, sysctl_memorystatus_vm_map_fork_pidwatch, "Q", "get/set pid watched for in vm_map_fork");
4643
4644
4645 /*
4646 * Record if a watched process fails to qualify for a vm_map_fork().
4647 */
4648 void
4649 memorystatus_abort_vm_map_fork(task_t task)
4650 {
4651 if (memorystatus_vm_map_fork_pidwatch_val != 0) {
4652 proc_t p = get_bsdtask_info(task);
4653 if (p != NULL && memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid) {
4654 memorystatus_vm_map_fork_pidwatch_val = -1ull;
4655 }
4656 }
4657 }
4658
4659 static void
4660 set_vm_map_fork_pidwatch(task_t task, uint64_t x)
4661 {
4662 if (memorystatus_vm_map_fork_pidwatch_val != 0) {
4663 proc_t p = get_bsdtask_info(task);
4664 if (p && (memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid)) {
4665 memorystatus_vm_map_fork_pidwatch_val |= x;
4666 }
4667 }
4668 }
4669
4670 #else /* DEVELOPMENT || DEBUG */
4671
4672
4673 static void
4674 set_vm_map_fork_pidwatch(task_t task, uint64_t x)
4675 {
4676 #pragma unused(task)
4677 #pragma unused(x)
4678 }
4679
4680 #endif /* DEVELOPMENT || DEBUG */
4681
4682 /*
4683 * Called during EXC_RESOURCE handling when a process exceeds a soft
4684 * memory limit. This is the corpse fork path and here we decide if
4685 * vm_map_fork will be allowed when creating the corpse.
4686 * The task being considered is suspended.
4687 *
4688 * By default, a vm_map_fork is allowed to proceed.
4689 *
4690 * A few simple policy assumptions:
4691 * If the device has a zero system-wide task limit,
4692 * then the vm_map_fork is allowed. macOS always has a zero
4693 * system wide task limit (unless overriden by a boot-arg).
4694 *
4695 * And if a process's memory footprint calculates less
4696 * than or equal to quarter of the system-wide task limit,
4697 * then the vm_map_fork is allowed. This calculation
4698 * is based on the assumption that a process can
4699 * munch memory up to the system-wide task limit.
4700 */
4701 extern boolean_t corpse_threshold_system_limit;
4702 boolean_t
4703 memorystatus_allowed_vm_map_fork(task_t task)
4704 {
4705 boolean_t is_allowed = TRUE; /* default */
4706
4707 uint64_t footprint_in_bytes;
4708 uint64_t max_allowed_bytes;
4709
4710 if (max_task_footprint_mb == 0) {
4711 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_ALLOWED);
4712 return is_allowed;
4713 }
4714
4715 footprint_in_bytes = get_task_phys_footprint(task);
4716
4717 /*
4718 * Maximum is 1/4 of the system-wide task limit by default.
4719 */
4720 max_allowed_bytes = ((uint64_t)max_task_footprint_mb * 1024 * 1024) >> 2;
4721
4722 #if DEBUG || DEVELOPMENT
4723 if (corpse_threshold_system_limit) {
4724 max_allowed_bytes = (uint64_t)max_task_footprint_mb * (1UL << 20);
4725 }
4726 #endif /* DEBUG || DEVELOPMENT */
4727
4728 if (footprint_in_bytes > max_allowed_bytes) {
4729 printf("memorystatus disallowed vm_map_fork %lld %lld\n", footprint_in_bytes, max_allowed_bytes);
4730 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED);
4731 return !is_allowed;
4732 }
4733
4734 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_ALLOWED);
4735 return is_allowed;
4736 }
4737
4738 void
4739 memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages)
4740 {
4741 assert(task);
4742 assert(footprint);
4743
4744 uint64_t pages;
4745
4746 pages = (get_task_phys_footprint(task) / PAGE_SIZE_64);
4747 assert(((uint32_t)pages) == pages);
4748 *footprint = (uint32_t)pages;
4749
4750 if (max_footprint_lifetime) {
4751 pages = (get_task_phys_footprint_lifetime_max(task) / PAGE_SIZE_64);
4752 assert(((uint32_t)pages) == pages);
4753 *max_footprint_lifetime = (uint32_t)pages;
4754 }
4755 if (purgeable_pages) {
4756 pages = (get_task_purgeable_size(task) / PAGE_SIZE_64);
4757 assert(((uint32_t)pages) == pages);
4758 *purgeable_pages = (uint32_t)pages;
4759 }
4760 }
4761
4762 static void
4763 memorystatus_get_task_phys_footprint_page_counts(task_t task,
4764 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
4765 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
4766 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
4767 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages, uint64_t *frozen_to_swap_pages)
4768 {
4769 assert(task);
4770
4771 if (internal_pages) {
4772 *internal_pages = (get_task_internal(task) / PAGE_SIZE_64);
4773 }
4774
4775 if (internal_compressed_pages) {
4776 *internal_compressed_pages = (get_task_internal_compressed(task) / PAGE_SIZE_64);
4777 }
4778
4779 if (purgeable_nonvolatile_pages) {
4780 *purgeable_nonvolatile_pages = (get_task_purgeable_nonvolatile(task) / PAGE_SIZE_64);
4781 }
4782
4783 if (purgeable_nonvolatile_compressed_pages) {
4784 *purgeable_nonvolatile_compressed_pages = (get_task_purgeable_nonvolatile_compressed(task) / PAGE_SIZE_64);
4785 }
4786
4787 if (alternate_accounting_pages) {
4788 *alternate_accounting_pages = (get_task_alternate_accounting(task) / PAGE_SIZE_64);
4789 }
4790
4791 if (alternate_accounting_compressed_pages) {
4792 *alternate_accounting_compressed_pages = (get_task_alternate_accounting_compressed(task) / PAGE_SIZE_64);
4793 }
4794
4795 if (iokit_mapped_pages) {
4796 *iokit_mapped_pages = (get_task_iokit_mapped(task) / PAGE_SIZE_64);
4797 }
4798
4799 if (page_table_pages) {
4800 *page_table_pages = (get_task_page_table(task) / PAGE_SIZE_64);
4801 }
4802
4803 #if CONFIG_FREEZE
4804 if (frozen_to_swap_pages) {
4805 *frozen_to_swap_pages = (get_task_frozen_to_swap(task) / PAGE_SIZE_64);
4806 }
4807 #else /* CONFIG_FREEZE */
4808 #pragma unused(frozen_to_swap_pages)
4809 #endif /* CONFIG_FREEZE */
4810 }
4811
4812 #if CONFIG_FREEZE
4813 /*
4814 * Copies the source entry into the destination snapshot.
4815 * Returns true on success. Fails if the destination snapshot is full.
4816 * Caller must hold the proc list lock.
4817 */
4818 static bool
4819 memorystatus_jetsam_snapshot_copy_entry_locked(memorystatus_jetsam_snapshot_t *dst_snapshot, unsigned int dst_snapshot_size, const memorystatus_jetsam_snapshot_entry_t *src_entry)
4820 {
4821 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
4822 assert(dst_snapshot);
4823
4824 if (dst_snapshot->entry_count == dst_snapshot_size) {
4825 /* Destination snapshot is full. Can not be updated until it is consumed. */
4826 return false;
4827 }
4828 if (dst_snapshot->entry_count == 0) {
4829 memorystatus_init_jetsam_snapshot_header(dst_snapshot);
4830 }
4831 memorystatus_jetsam_snapshot_entry_t *dst_entry = &dst_snapshot->entries[dst_snapshot->entry_count++];
4832 memcpy(dst_entry, src_entry, sizeof(memorystatus_jetsam_snapshot_entry_t));
4833 return true;
4834 }
4835 #endif /* CONFIG_FREEZE */
4836
4837 static bool
4838 memorystatus_init_jetsam_snapshot_entry_with_kill_locked(memorystatus_jetsam_snapshot_t *snapshot, proc_t p, uint32_t kill_cause, uint64_t killtime, memorystatus_jetsam_snapshot_entry_t **entry)
4839 {
4840 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
4841 memorystatus_jetsam_snapshot_entry_t *snapshot_list = snapshot->entries;
4842 size_t i = snapshot->entry_count;
4843
4844 if (memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i], (snapshot->js_gencount)) == TRUE) {
4845 *entry = &snapshot_list[i];
4846 (*entry)->killed = kill_cause;
4847 (*entry)->jse_killtime = killtime;
4848
4849 snapshot->entry_count = i + 1;
4850 return true;
4851 }
4852 return false;
4853 }
4854
4855 /*
4856 * This routine only acts on the global jetsam event snapshot.
4857 * Updating the process's entry can race when the memorystatus_thread
4858 * has chosen to kill a process that is racing to exit on another core.
4859 */
4860 static void
4861 memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime)
4862 {
4863 memorystatus_jetsam_snapshot_entry_t *entry = NULL;
4864 memorystatus_jetsam_snapshot_t *snapshot = NULL;
4865 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
4866
4867 unsigned int i;
4868 #if CONFIG_FREEZE
4869 bool copied_to_freezer_snapshot = false;
4870 #endif /* CONFIG_FREEZE */
4871
4872 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
4873
4874 if (memorystatus_jetsam_snapshot_count == 0) {
4875 /*
4876 * No active snapshot.
4877 * Nothing to do.
4878 */
4879 goto exit;
4880 }
4881
4882 /*
4883 * Sanity check as this routine should only be called
4884 * from a jetsam kill path.
4885 */
4886 assert(kill_cause != 0 && killtime != 0);
4887
4888 snapshot = memorystatus_jetsam_snapshot;
4889 snapshot_list = memorystatus_jetsam_snapshot->entries;
4890
4891 for (i = 0; i < memorystatus_jetsam_snapshot_count; i++) {
4892 if (snapshot_list[i].pid == p->p_pid) {
4893 entry = &snapshot_list[i];
4894
4895 if (entry->killed || entry->jse_killtime) {
4896 /*
4897 * We apparently raced on the exit path
4898 * for this process, as it's snapshot entry
4899 * has already recorded a kill.
4900 */
4901 assert(entry->killed && entry->jse_killtime);
4902 break;
4903 }
4904
4905 /*
4906 * Update the entry we just found in the snapshot.
4907 */
4908
4909 entry->killed = kill_cause;
4910 entry->jse_killtime = killtime;
4911 entry->jse_gencount = snapshot->js_gencount;
4912 entry->jse_idle_delta = p->p_memstat_idle_delta;
4913 #if CONFIG_FREEZE
4914 entry->jse_thaw_count = p->p_memstat_thaw_count;
4915 entry->jse_freeze_skip_reason = p->p_memstat_freeze_skip_reason;
4916 #else /* CONFIG_FREEZE */
4917 entry->jse_thaw_count = 0;
4918 entry->jse_freeze_skip_reason = kMemorystatusFreezeSkipReasonNone;
4919 #endif /* CONFIG_FREEZE */
4920
4921 /*
4922 * If a process has moved between bands since snapshot was
4923 * initialized, then likely these fields changed too.
4924 */
4925 if (entry->priority != p->p_memstat_effectivepriority) {
4926 strlcpy(entry->name, p->p_name, sizeof(entry->name));
4927 entry->priority = p->p_memstat_effectivepriority;
4928 entry->state = memorystatus_build_state(p);
4929 entry->user_data = p->p_memstat_userdata;
4930 entry->fds = p->p_fd->fd_nfiles;
4931 }
4932
4933 /*
4934 * Always update the page counts on a kill.
4935 */
4936
4937 uint32_t pages = 0;
4938 uint32_t max_pages_lifetime = 0;
4939 uint32_t purgeable_pages = 0;
4940
4941 memorystatus_get_task_page_counts(p->task, &pages, &max_pages_lifetime, &purgeable_pages);
4942 entry->pages = (uint64_t)pages;
4943 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
4944 entry->purgeable_pages = (uint64_t)purgeable_pages;
4945
4946 uint64_t internal_pages = 0;
4947 uint64_t internal_compressed_pages = 0;
4948 uint64_t purgeable_nonvolatile_pages = 0;
4949 uint64_t purgeable_nonvolatile_compressed_pages = 0;
4950 uint64_t alternate_accounting_pages = 0;
4951 uint64_t alternate_accounting_compressed_pages = 0;
4952 uint64_t iokit_mapped_pages = 0;
4953 uint64_t page_table_pages = 0;
4954 uint64_t frozen_to_swap_pages = 0;
4955
4956 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
4957 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
4958 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
4959 &iokit_mapped_pages, &page_table_pages, &frozen_to_swap_pages);
4960
4961 entry->jse_internal_pages = internal_pages;
4962 entry->jse_internal_compressed_pages = internal_compressed_pages;
4963 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
4964 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
4965 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
4966 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
4967 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
4968 entry->jse_page_table_pages = page_table_pages;
4969 entry->jse_frozen_to_swap_pages = frozen_to_swap_pages;
4970
4971 uint64_t region_count = 0;
4972 memorystatus_get_task_memory_region_count(p->task, &region_count);
4973 entry->jse_memory_region_count = region_count;
4974
4975 goto exit;
4976 }
4977 }
4978
4979 if (entry == NULL) {
4980 /*
4981 * The entry was not found in the snapshot, so the process must have
4982 * launched after the snapshot was initialized.
4983 * Let's try to append the new entry.
4984 */
4985 if (memorystatus_jetsam_snapshot_count < memorystatus_jetsam_snapshot_max) {
4986 /*
4987 * A populated snapshot buffer exists
4988 * and there is room to init a new entry.
4989 */
4990 assert(memorystatus_jetsam_snapshot_count == snapshot->entry_count);
4991
4992 if (memorystatus_init_jetsam_snapshot_entry_with_kill_locked(snapshot, p, kill_cause, killtime, &entry)) {
4993 memorystatus_jetsam_snapshot_count++;
4994
4995 if (memorystatus_jetsam_snapshot_count >= memorystatus_jetsam_snapshot_max) {
4996 /*
4997 * We just used the last slot in the snapshot buffer.
4998 * We only want to log it once... so we do it here
4999 * when we notice we've hit the max.
5000 */
5001 printf("memorystatus: WARNING snapshot buffer is full, count %d\n",
5002 memorystatus_jetsam_snapshot_count);
5003 }
5004 }
5005 }
5006 }
5007
5008 exit:
5009 if (entry) {
5010 #if CONFIG_FREEZE
5011 if (memorystatus_jetsam_use_freezer_snapshot && isApp(p)) {
5012 /* This is an app kill. Record it in the freezer snapshot so dasd can incorporate this in its recommendations. */
5013 copied_to_freezer_snapshot = memorystatus_jetsam_snapshot_copy_entry_locked(memorystatus_jetsam_snapshot_freezer, memorystatus_jetsam_snapshot_freezer_max, entry);
5014 if (copied_to_freezer_snapshot && memorystatus_jetsam_snapshot_freezer->entry_count == memorystatus_jetsam_snapshot_freezer_max) {
5015 /*
5016 * We just used the last slot in the freezer snapshot buffer.
5017 * We only want to log it once... so we do it here
5018 * when we notice we've hit the max.
5019 */
5020 os_log_error(OS_LOG_DEFAULT, "memorystatus: WARNING freezer snapshot buffer is full, count %zu",
5021 memorystatus_jetsam_snapshot_freezer->entry_count);
5022 }
5023 }
5024 #endif /* CONFIG_FREEZE */
5025 } else {
5026 /*
5027 * If we reach here, the snapshot buffer could not be updated.
5028 * Most likely, the buffer is full, in which case we would have
5029 * logged a warning in the previous call.
5030 *
5031 * For now, we will stop appending snapshot entries.
5032 * When the buffer is consumed, the snapshot state will reset.
5033 */
5034
5035 MEMORYSTATUS_DEBUG(4, "memorystatus_update_jetsam_snapshot_entry_locked: failed to update pid %d, priority %d, count %d\n",
5036 p->p_pid, p->p_memstat_effectivepriority, memorystatus_jetsam_snapshot_count);
5037
5038 #if CONFIG_FREEZE
5039 /* We still attempt to record this in the freezer snapshot */
5040 if (memorystatus_jetsam_use_freezer_snapshot && isApp(p)) {
5041 snapshot = memorystatus_jetsam_snapshot_freezer;
5042 if (snapshot->entry_count < memorystatus_jetsam_snapshot_freezer_max) {
5043 copied_to_freezer_snapshot = memorystatus_init_jetsam_snapshot_entry_with_kill_locked(snapshot, p, kill_cause, killtime, &entry);
5044 if (copied_to_freezer_snapshot && memorystatus_jetsam_snapshot_freezer->entry_count == memorystatus_jetsam_snapshot_freezer_max) {
5045 /*
5046 * We just used the last slot in the freezer snapshot buffer.
5047 * We only want to log it once... so we do it here
5048 * when we notice we've hit the max.
5049 */
5050 os_log_error(OS_LOG_DEFAULT, "memorystatus: WARNING freezer snapshot buffer is full, count %zu",
5051 memorystatus_jetsam_snapshot_freezer->entry_count);
5052 }
5053 }
5054 }
5055 #endif /* CONFIG_FREEZE */
5056 }
5057
5058 return;
5059 }
5060
5061 #if CONFIG_JETSAM
5062 void
5063 memorystatus_pages_update(unsigned int pages_avail)
5064 {
5065 memorystatus_available_pages = pages_avail;
5066
5067 #if VM_PRESSURE_EVENTS
5068 /*
5069 * Since memorystatus_available_pages changes, we should
5070 * re-evaluate the pressure levels on the system and
5071 * check if we need to wake the pressure thread.
5072 * We also update memorystatus_level in that routine.
5073 */
5074 vm_pressure_response();
5075
5076 if (memorystatus_available_pages <= memorystatus_available_pages_pressure) {
5077 if (memorystatus_hwm_candidates || (memorystatus_available_pages <= memorystatus_available_pages_critical)) {
5078 memorystatus_thread_wake();
5079 }
5080 }
5081 #if CONFIG_FREEZE
5082 /*
5083 * We can't grab the freezer_mutex here even though that synchronization would be correct to inspect
5084 * the # of frozen processes and wakeup the freezer thread. Reason being that we come here into this
5085 * code with (possibly) the page-queue locks held and preemption disabled. So trying to grab a mutex here
5086 * will result in the "mutex with preemption disabled" panic.
5087 */
5088
5089 if (memorystatus_freeze_thread_should_run() == TRUE) {
5090 /*
5091 * The freezer thread is usually woken up by some user-space call i.e. pid_hibernate(any process).
5092 * That trigger isn't invoked often enough and so we are enabling this explicit wakeup here.
5093 */
5094 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5095 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
5096 }
5097 }
5098 #endif /* CONFIG_FREEZE */
5099
5100 #else /* VM_PRESSURE_EVENTS */
5101
5102 boolean_t critical, delta;
5103
5104 if (!memorystatus_delta) {
5105 return;
5106 }
5107
5108 critical = (pages_avail < memorystatus_available_pages_critical) ? TRUE : FALSE;
5109 delta = ((pages_avail >= (memorystatus_available_pages + memorystatus_delta))
5110 || (memorystatus_available_pages >= (pages_avail + memorystatus_delta))) ? TRUE : FALSE;
5111
5112 if (critical || delta) {
5113 unsigned int total_pages;
5114
5115 total_pages = (unsigned int) atop_64(max_mem);
5116 #if CONFIG_SECLUDED_MEMORY
5117 total_pages -= vm_page_secluded_count;
5118 #endif /* CONFIG_SECLUDED_MEMORY */
5119 memorystatus_level = memorystatus_available_pages * 100 / total_pages;
5120 memorystatus_thread_wake();
5121 }
5122 #endif /* VM_PRESSURE_EVENTS */
5123 }
5124 #endif /* CONFIG_JETSAM */
5125
5126 static boolean_t
5127 memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount)
5128 {
5129 clock_sec_t tv_sec;
5130 clock_usec_t tv_usec;
5131 uint32_t pages = 0;
5132 uint32_t max_pages_lifetime = 0;
5133 uint32_t purgeable_pages = 0;
5134 uint64_t internal_pages = 0;
5135 uint64_t internal_compressed_pages = 0;
5136 uint64_t purgeable_nonvolatile_pages = 0;
5137 uint64_t purgeable_nonvolatile_compressed_pages = 0;
5138 uint64_t alternate_accounting_pages = 0;
5139 uint64_t alternate_accounting_compressed_pages = 0;
5140 uint64_t iokit_mapped_pages = 0;
5141 uint64_t page_table_pages = 0;
5142 uint64_t frozen_to_swap_pages = 0;
5143 uint64_t region_count = 0;
5144 uint64_t cids[COALITION_NUM_TYPES];
5145
5146 memset(entry, 0, sizeof(memorystatus_jetsam_snapshot_entry_t));
5147
5148 entry->pid = p->p_pid;
5149 strlcpy(&entry->name[0], p->p_name, sizeof(entry->name));
5150 entry->priority = p->p_memstat_effectivepriority;
5151
5152 memorystatus_get_task_page_counts(p->task, &pages, &max_pages_lifetime, &purgeable_pages);
5153 entry->pages = (uint64_t)pages;
5154 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
5155 entry->purgeable_pages = (uint64_t)purgeable_pages;
5156
5157 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
5158 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
5159 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
5160 &iokit_mapped_pages, &page_table_pages, &frozen_to_swap_pages);
5161
5162 entry->jse_internal_pages = internal_pages;
5163 entry->jse_internal_compressed_pages = internal_compressed_pages;
5164 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
5165 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
5166 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
5167 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
5168 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
5169 entry->jse_page_table_pages = page_table_pages;
5170 entry->jse_frozen_to_swap_pages = frozen_to_swap_pages;
5171
5172 memorystatus_get_task_memory_region_count(p->task, &region_count);
5173 entry->jse_memory_region_count = region_count;
5174
5175 entry->state = memorystatus_build_state(p);
5176 entry->user_data = p->p_memstat_userdata;
5177 memcpy(&entry->uuid[0], &p->p_uuid[0], sizeof(p->p_uuid));
5178 entry->fds = p->p_fd->fd_nfiles;
5179
5180 absolutetime_to_microtime(get_task_cpu_time(p->task), &tv_sec, &tv_usec);
5181 entry->cpu_time.tv_sec = (int64_t)tv_sec;
5182 entry->cpu_time.tv_usec = (int64_t)tv_usec;
5183
5184 assert(p->p_stats != NULL);
5185 entry->jse_starttime = p->p_stats->ps_start; /* abstime process started */
5186 entry->jse_killtime = 0; /* abstime jetsam chose to kill process */
5187 entry->killed = 0; /* the jetsam kill cause */
5188 entry->jse_gencount = gencount; /* indicates a pass through jetsam thread, when process was targeted to be killed */
5189
5190 entry->jse_idle_delta = p->p_memstat_idle_delta; /* Most recent timespan spent in idle-band */
5191
5192 #if CONFIG_FREEZE
5193 entry->jse_freeze_skip_reason = p->p_memstat_freeze_skip_reason;
5194 entry->jse_thaw_count = p->p_memstat_thaw_count;
5195 #else /* CONFIG_FREEZE */
5196 entry->jse_thaw_count = 0;
5197 entry->jse_freeze_skip_reason = kMemorystatusFreezeSkipReasonNone;
5198 #endif /* CONFIG_FREEZE */
5199
5200 proc_coalitionids(p, cids);
5201 entry->jse_coalition_jetsam_id = cids[COALITION_TYPE_JETSAM];
5202
5203 return TRUE;
5204 }
5205
5206 static void
5207 memorystatus_init_snapshot_vmstats(memorystatus_jetsam_snapshot_t *snapshot)
5208 {
5209 kern_return_t kr = KERN_SUCCESS;
5210 mach_msg_type_number_t count = HOST_VM_INFO64_COUNT;
5211 vm_statistics64_data_t vm_stat;
5212
5213 if ((kr = host_statistics64(host_self(), HOST_VM_INFO64, (host_info64_t)&vm_stat, &count)) != KERN_SUCCESS) {
5214 printf("memorystatus_init_jetsam_snapshot_stats: host_statistics64 failed with %d\n", kr);
5215 memset(&snapshot->stats, 0, sizeof(snapshot->stats));
5216 } else {
5217 snapshot->stats.free_pages = vm_stat.free_count;
5218 snapshot->stats.active_pages = vm_stat.active_count;
5219 snapshot->stats.inactive_pages = vm_stat.inactive_count;
5220 snapshot->stats.throttled_pages = vm_stat.throttled_count;
5221 snapshot->stats.purgeable_pages = vm_stat.purgeable_count;
5222 snapshot->stats.wired_pages = vm_stat.wire_count;
5223
5224 snapshot->stats.speculative_pages = vm_stat.speculative_count;
5225 snapshot->stats.filebacked_pages = vm_stat.external_page_count;
5226 snapshot->stats.anonymous_pages = vm_stat.internal_page_count;
5227 snapshot->stats.compressions = vm_stat.compressions;
5228 snapshot->stats.decompressions = vm_stat.decompressions;
5229 snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
5230 snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
5231 }
5232
5233 get_zone_map_size(&snapshot->stats.zone_map_size, &snapshot->stats.zone_map_capacity);
5234
5235 bzero(snapshot->stats.largest_zone_name, sizeof(snapshot->stats.largest_zone_name));
5236 get_largest_zone_info(snapshot->stats.largest_zone_name, sizeof(snapshot->stats.largest_zone_name),
5237 &snapshot->stats.largest_zone_size);
5238 }
5239
5240 /*
5241 * Collect vm statistics at boot.
5242 * Called only once (see kern_exec.c)
5243 * Data can be consumed at any time.
5244 */
5245 void
5246 memorystatus_init_at_boot_snapshot()
5247 {
5248 memorystatus_init_snapshot_vmstats(&memorystatus_at_boot_snapshot);
5249 memorystatus_at_boot_snapshot.entry_count = 0;
5250 memorystatus_at_boot_snapshot.notification_time = 0; /* updated when consumed */
5251 memorystatus_at_boot_snapshot.snapshot_time = mach_absolute_time();
5252 }
5253
5254 static void
5255 memorystatus_init_jetsam_snapshot_header(memorystatus_jetsam_snapshot_t *snapshot)
5256 {
5257 memorystatus_init_snapshot_vmstats(snapshot);
5258 snapshot->snapshot_time = mach_absolute_time();
5259 snapshot->notification_time = 0;
5260 snapshot->js_gencount = 0;
5261 }
5262
5263 static void
5264 memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count )
5265 {
5266 proc_t p, next_p;
5267 unsigned int b = 0, i = 0;
5268
5269 memorystatus_jetsam_snapshot_t *snapshot = NULL;
5270 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
5271 unsigned int snapshot_max = 0;
5272
5273 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
5274
5275 if (od_snapshot) {
5276 /*
5277 * This is an on_demand snapshot
5278 */
5279 snapshot = od_snapshot;
5280 snapshot_list = od_snapshot->entries;
5281 snapshot_max = ods_list_count;
5282 } else {
5283 /*
5284 * This is a jetsam event snapshot
5285 */
5286 snapshot = memorystatus_jetsam_snapshot;
5287 snapshot_list = memorystatus_jetsam_snapshot->entries;
5288 snapshot_max = memorystatus_jetsam_snapshot_max;
5289 }
5290
5291 memorystatus_init_jetsam_snapshot_header(snapshot);
5292
5293 next_p = memorystatus_get_first_proc_locked(&b, TRUE);
5294 while (next_p) {
5295 p = next_p;
5296 next_p = memorystatus_get_next_proc_locked(&b, p, TRUE);
5297
5298 if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i], snapshot->js_gencount)) {
5299 continue;
5300 }
5301
5302 MEMORYSTATUS_DEBUG(0, "jetsam snapshot pid %d, uuid = %02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
5303 p->p_pid,
5304 p->p_uuid[0], p->p_uuid[1], p->p_uuid[2], p->p_uuid[3], p->p_uuid[4], p->p_uuid[5], p->p_uuid[6], p->p_uuid[7],
5305 p->p_uuid[8], p->p_uuid[9], p->p_uuid[10], p->p_uuid[11], p->p_uuid[12], p->p_uuid[13], p->p_uuid[14], p->p_uuid[15]);
5306
5307 if (++i == snapshot_max) {
5308 break;
5309 }
5310 }
5311
5312 snapshot->entry_count = i;
5313
5314 if (!od_snapshot) {
5315 /* update the system buffer count */
5316 memorystatus_jetsam_snapshot_count = i;
5317 }
5318 }
5319
5320 #if DEVELOPMENT || DEBUG
5321
5322 #if CONFIG_JETSAM
5323 static int
5324 memorystatus_cmd_set_panic_bits(user_addr_t buffer, size_t buffer_size)
5325 {
5326 int ret;
5327 memorystatus_jetsam_panic_options_t debug;
5328
5329 if (buffer_size != sizeof(memorystatus_jetsam_panic_options_t)) {
5330 return EINVAL;
5331 }
5332
5333 ret = copyin(buffer, &debug, buffer_size);
5334 if (ret) {
5335 return ret;
5336 }
5337
5338 /* Panic bits match kMemorystatusKilled* enum */
5339 memorystatus_jetsam_panic_debug = (memorystatus_jetsam_panic_debug & ~debug.mask) | (debug.data & debug.mask);
5340
5341 /* Copyout new value */
5342 debug.data = memorystatus_jetsam_panic_debug;
5343 ret = copyout(&debug, buffer, sizeof(memorystatus_jetsam_panic_options_t));
5344
5345 return ret;
5346 }
5347 #endif /* CONFIG_JETSAM */
5348
5349 /*
5350 * Verify that the given bucket has been sorted correctly.
5351 *
5352 * Walks through the bucket and verifies that all pids in the
5353 * expected_order buffer are in that bucket and in the same
5354 * relative order.
5355 *
5356 * The proc_list_lock must be held by the caller.
5357 */
5358 static int
5359 memorystatus_verify_sort_order(unsigned int bucket_index, pid_t *expected_order, size_t num_pids)
5360 {
5361 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
5362
5363 int error = 0;
5364 proc_t p = NULL;
5365 size_t i = 0;
5366
5367 /*
5368 * NB: We allow other procs to be mixed in within the expected ones.
5369 * We just need the expected procs to be in the right order relative to each other.
5370 */
5371 p = memorystatus_get_first_proc_locked(&bucket_index, FALSE);
5372 while (p) {
5373 if (p->p_pid == expected_order[i]) {
5374 i++;
5375 }
5376 if (i == num_pids) {
5377 break;
5378 }
5379 p = memorystatus_get_next_proc_locked(&bucket_index, p, FALSE);
5380 }
5381 if (i != num_pids) {
5382 char buffer[128];
5383 size_t len = sizeof(buffer);
5384 size_t buffer_idx = 0;
5385 os_log_error(OS_LOG_DEFAULT, "memorystatus_verify_sort_order: Processes in bucket %d were not sorted properly\n", bucket_index);
5386 for (i = 0; i < num_pids; i++) {
5387 int num_written = snprintf(buffer + buffer_idx, len - buffer_idx, "%d,", expected_order[i]);
5388 if (num_written <= 0) {
5389 break;
5390 }
5391 if (buffer_idx + (unsigned int) num_written >= len) {
5392 break;
5393 }
5394 buffer_idx += num_written;
5395 }
5396 os_log_error(OS_LOG_DEFAULT, "memorystatus_verify_sort_order: Expected order [%s]", buffer);
5397 memset(buffer, 0, len);
5398 buffer_idx = 0;
5399 p = memorystatus_get_first_proc_locked(&bucket_index, FALSE);
5400 i = 0;
5401 os_log_error(OS_LOG_DEFAULT, "memorystatus_verify_sort_order: Actual order:");
5402 while (p) {
5403 int num_written;
5404 if (buffer_idx == 0) {
5405 num_written = snprintf(buffer + buffer_idx, len - buffer_idx, "%zu: %d,", i, p->p_pid);
5406 } else {
5407 num_written = snprintf(buffer + buffer_idx, len - buffer_idx, "%d,", p->p_pid);
5408 }
5409 if (num_written <= 0) {
5410 break;
5411 }
5412 buffer_idx += (unsigned int) num_written;
5413 assert(buffer_idx <= len);
5414 if (i % 10 == 0) {
5415 os_log_error(OS_LOG_DEFAULT, "memorystatus_verify_sort_order: %s", buffer);
5416 buffer_idx = 0;
5417 }
5418 p = memorystatus_get_next_proc_locked(&bucket_index, p, FALSE);
5419 i++;
5420 }
5421 if (buffer_idx != 0) {
5422 os_log_error(OS_LOG_DEFAULT, "memorystatus_verify_sort_order: %s", buffer);
5423 }
5424 error = EINVAL;
5425 }
5426 return error;
5427 }
5428
5429 /*
5430 * Triggers a sort_order on a specified jetsam priority band.
5431 * This is for testing only, used to force a path through the sort
5432 * function.
5433 */
5434 static int
5435 memorystatus_cmd_test_jetsam_sort(int priority,
5436 int sort_order,
5437 user_addr_t expected_order_user,
5438 size_t expected_order_user_len)
5439 {
5440 int error = 0;
5441 unsigned int bucket_index = 0;
5442 static size_t kMaxPids = 8;
5443 pid_t expected_order[kMaxPids];
5444 size_t copy_size = sizeof(expected_order);
5445 size_t num_pids;
5446
5447 if (expected_order_user_len < copy_size) {
5448 copy_size = expected_order_user_len;
5449 }
5450 num_pids = copy_size / sizeof(pid_t);
5451
5452 error = copyin(expected_order_user, expected_order, copy_size);
5453 if (error != 0) {
5454 return error;
5455 }
5456
5457 if (priority == -1) {
5458 /* Use as shorthand for default priority */
5459 bucket_index = JETSAM_PRIORITY_DEFAULT;
5460 } else {
5461 bucket_index = (unsigned int)priority;
5462 }
5463
5464 /*
5465 * Acquire lock before sorting so we can check the sort order
5466 * while still holding the lock.
5467 */
5468 proc_list_lock();
5469
5470 memorystatus_sort_bucket_locked(bucket_index, sort_order);
5471
5472 if (expected_order_user != CAST_USER_ADDR_T(NULL) && expected_order_user_len > 0) {
5473 error = memorystatus_verify_sort_order(bucket_index, expected_order, num_pids);
5474 }
5475
5476 proc_list_unlock();
5477
5478 return error;
5479 }
5480
5481 #endif /* DEVELOPMENT || DEBUG */
5482
5483 /*
5484 * Prepare the process to be killed (set state, update snapshot) and kill it.
5485 */
5486 static uint64_t memorystatus_purge_before_jetsam_success = 0;
5487
5488 static boolean_t
5489 memorystatus_kill_proc(proc_t p, uint32_t cause, os_reason_t jetsam_reason, boolean_t *killed, uint64_t *footprint_of_killed_proc)
5490 {
5491 pid_t aPid = 0;
5492 uint32_t aPid_ep = 0;
5493
5494 uint64_t killtime = 0;
5495 clock_sec_t tv_sec;
5496 clock_usec_t tv_usec;
5497 uint32_t tv_msec;
5498 boolean_t retval = FALSE;
5499
5500 aPid = p->p_pid;
5501 aPid_ep = p->p_memstat_effectivepriority;
5502
5503 if (cause != kMemorystatusKilledVnodes && cause != kMemorystatusKilledZoneMapExhaustion) {
5504 /*
5505 * Genuine memory pressure and not other (vnode/zone) resource exhaustion.
5506 */
5507 boolean_t success = FALSE;
5508 uint64_t num_pages_purged;
5509 uint64_t num_pages_reclaimed = 0;
5510 uint64_t num_pages_unsecluded = 0;
5511
5512 networking_memstatus_callout(p, cause);
5513 num_pages_purged = vm_purgeable_purge_task_owned(p->task);
5514 num_pages_reclaimed += num_pages_purged;
5515 #if CONFIG_SECLUDED_MEMORY
5516 if (cause == kMemorystatusKilledVMPageShortage &&
5517 vm_page_secluded_count > 0 &&
5518 task_can_use_secluded_mem(p->task, FALSE)) {
5519 /*
5520 * We're about to kill a process that has access
5521 * to the secluded pool. Drain that pool into the
5522 * free or active queues to make these pages re-appear
5523 * as "available", which might make us no longer need
5524 * to kill that process.
5525 * Since the secluded pool does not get refilled while
5526 * a process has access to it, it should remain
5527 * drained.
5528 */
5529 num_pages_unsecluded = vm_page_secluded_drain();
5530 num_pages_reclaimed += num_pages_unsecluded;
5531 }
5532 #endif /* CONFIG_SECLUDED_MEMORY */
5533
5534 if (num_pages_reclaimed) {
5535 /*
5536 * We actually reclaimed something and so let's
5537 * check if we need to continue with the kill.
5538 */
5539 if (cause == kMemorystatusKilledHiwat) {
5540 uint64_t footprint_in_bytes = get_task_phys_footprint(p->task);
5541 uint64_t memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
5542 success = (footprint_in_bytes <= memlimit_in_bytes);
5543 } else {
5544 success = (memorystatus_avail_pages_below_pressure() == FALSE);
5545 #if CONFIG_SECLUDED_MEMORY
5546 if (!success && num_pages_unsecluded) {
5547 /*
5548 * We just drained the secluded pool
5549 * because we're about to kill a
5550 * process that has access to it.
5551 * This is an important process and
5552 * we'd rather not kill it unless
5553 * absolutely necessary, so declare
5554 * success even if draining the pool
5555 * did not quite get us out of the
5556 * "pressure" level but still got
5557 * us out of the "critical" level.
5558 */
5559 success = (memorystatus_avail_pages_below_critical() == FALSE);
5560 }
5561 #endif /* CONFIG_SECLUDED_MEMORY */
5562 }
5563
5564 if (success) {
5565 memorystatus_purge_before_jetsam_success++;
5566
5567 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: reclaimed %llu pages (%llu purged, %llu unsecluded) from pid %d [%s] and avoided %s\n",
5568 num_pages_reclaimed, num_pages_purged, num_pages_unsecluded, aPid, ((p && *p->p_name) ? p->p_name : "unknown"), memorystatus_kill_cause_name[cause]);
5569
5570 *killed = FALSE;
5571
5572 return TRUE;
5573 }
5574 }
5575 }
5576
5577 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
5578 MEMORYSTATUS_DEBUG(1, "jetsam: killing pid %d [%s] - %lld Mb > 1 (%d Mb)\n",
5579 aPid, (*p->p_name ? p->p_name : "unknown"),
5580 (footprint_in_bytes / (1024ULL * 1024ULL)), /* converted bytes to MB */
5581 p->p_memstat_memlimit);
5582 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
5583
5584 killtime = mach_absolute_time();
5585 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5586 tv_msec = tv_usec / 1000;
5587
5588 proc_list_lock();
5589 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5590 proc_list_unlock();
5591
5592 char kill_reason_string[128];
5593
5594 if (cause == kMemorystatusKilledHiwat) {
5595 strlcpy(kill_reason_string, "killing_highwater_process", 128);
5596 } else {
5597 if (aPid_ep == JETSAM_PRIORITY_IDLE) {
5598 strlcpy(kill_reason_string, "killing_idle_process", 128);
5599 } else {
5600 strlcpy(kill_reason_string, "killing_top_process", 128);
5601 }
5602 }
5603
5604 /*
5605 * memorystatus_do_kill drops a reference, so take another one so we can
5606 * continue to use this exit reason even after memorystatus_do_kill()
5607 * returns
5608 */
5609 os_reason_ref(jetsam_reason);
5610
5611 retval = memorystatus_do_kill(p, cause, jetsam_reason, footprint_of_killed_proc);
5612 *killed = retval;
5613
5614 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: %s pid %d [%s] (%s %d) %lluKB - memorystatus_available_pages: %llu",
5615 (unsigned long)tv_sec, tv_msec, kill_reason_string,
5616 aPid, ((p && *p->p_name) ? p->p_name : "unknown"),
5617 memorystatus_kill_cause_name[cause], aPid_ep,
5618 (*footprint_of_killed_proc) >> 10, (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES);
5619
5620 return retval;
5621 }
5622
5623 /*
5624 * Jetsam the first process in the queue.
5625 */
5626 static boolean_t
5627 memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, os_reason_t jetsam_reason,
5628 int32_t *priority, uint32_t *errors, uint64_t *memory_reclaimed)
5629 {
5630 pid_t aPid;
5631 proc_t p = PROC_NULL, next_p = PROC_NULL;
5632 boolean_t new_snapshot = FALSE, force_new_snapshot = FALSE, killed = FALSE, freed_mem = FALSE;
5633 unsigned int i = 0;
5634 uint32_t aPid_ep;
5635 int32_t local_max_kill_prio = JETSAM_PRIORITY_IDLE;
5636 uint64_t footprint_of_killed_proc = 0;
5637
5638 #ifndef CONFIG_FREEZE
5639 #pragma unused(any)
5640 #endif
5641
5642 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5643 MEMORYSTATUS_LOG_AVAILABLE_PAGES, 0, 0, 0, 0);
5644
5645
5646 #if CONFIG_JETSAM
5647 if (sort_flag == TRUE) {
5648 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
5649 }
5650
5651 local_max_kill_prio = max_kill_priority;
5652
5653 force_new_snapshot = FALSE;
5654
5655 #else /* CONFIG_JETSAM */
5656
5657 if (sort_flag == TRUE) {
5658 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_IDLE, JETSAM_SORT_DEFAULT);
5659 }
5660
5661 /*
5662 * On macos, we currently only have 2 reasons to be here:
5663 *
5664 * kMemorystatusKilledZoneMapExhaustion
5665 * AND
5666 * kMemorystatusKilledVMCompressorSpaceShortage
5667 *
5668 * If we are here because of kMemorystatusKilledZoneMapExhaustion, we will consider
5669 * any and all processes as eligible kill candidates since we need to avoid a panic.
5670 *
5671 * Since this function can be called async. it is harder to toggle the max_kill_priority
5672 * value before and after a call. And so we use this local variable to set the upper band
5673 * on the eligible kill bands.
5674 */
5675 if (cause == kMemorystatusKilledZoneMapExhaustion) {
5676 local_max_kill_prio = JETSAM_PRIORITY_MAX;
5677 } else {
5678 local_max_kill_prio = max_kill_priority;
5679 }
5680
5681 /*
5682 * And, because we are here under extreme circumstances, we force a snapshot even for
5683 * IDLE kills.
5684 */
5685 force_new_snapshot = TRUE;
5686
5687 #endif /* CONFIG_JETSAM */
5688
5689 if (cause != kMemorystatusKilledZoneMapExhaustion &&
5690 jetsam_current_thread() != NULL &&
5691 jetsam_current_thread()->limit_to_low_bands &&
5692 local_max_kill_prio > JETSAM_PRIORITY_BACKGROUND) {
5693 local_max_kill_prio = JETSAM_PRIORITY_BACKGROUND;
5694 }
5695
5696 proc_list_lock();
5697
5698 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5699 while (next_p && (next_p->p_memstat_effectivepriority <= local_max_kill_prio)) {
5700 p = next_p;
5701 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5702
5703
5704 aPid = p->p_pid;
5705 aPid_ep = p->p_memstat_effectivepriority;
5706
5707 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5708 continue; /* with lock held */
5709 }
5710
5711 if (cause == kMemorystatusKilledVnodes) {
5712 /*
5713 * If the system runs out of vnodes, we systematically jetsam
5714 * processes in hopes of stumbling onto a vnode gain that helps
5715 * the system recover. The process that happens to trigger
5716 * this path has no known relationship to the vnode shortage.
5717 * Deadlock avoidance: attempt to safeguard the caller.
5718 */
5719
5720 if (p == current_proc()) {
5721 /* do not jetsam the current process */
5722 continue;
5723 }
5724 }
5725
5726 #if CONFIG_FREEZE
5727 boolean_t skip;
5728 boolean_t reclaim_proc = !(p->p_memstat_state & P_MEMSTAT_LOCKED);
5729 if (any || reclaim_proc) {
5730 skip = FALSE;
5731 } else {
5732 skip = TRUE;
5733 }
5734
5735 if (skip) {
5736 continue;
5737 } else
5738 #endif
5739 {
5740 if (proc_ref_locked(p) == p) {
5741 /*
5742 * Mark as terminated so that if exit1() indicates success, but the process (for example)
5743 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
5744 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
5745 * acquisition of the proc lock.
5746 */
5747 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5748 } else {
5749 /*
5750 * We need to restart the search again because
5751 * proc_ref_locked _can_ drop the proc_list lock
5752 * and we could have lost our stored next_p via
5753 * an exit() on another core.
5754 */
5755 i = 0;
5756 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5757 continue;
5758 }
5759
5760 /*
5761 * Capture a snapshot if none exists and:
5762 * - we are forcing a new snapshot creation, either because:
5763 * - on a particular platform we need these snapshots every time, OR
5764 * - a boot-arg/embedded device tree property has been set.
5765 * - priority was not requested (this is something other than an ambient kill)
5766 * - the priority was requested *and* the targeted process is not at idle priority
5767 */
5768 if ((memorystatus_jetsam_snapshot_count == 0) &&
5769 (force_new_snapshot || memorystatus_idle_snapshot || ((!priority) || (priority && (aPid_ep != JETSAM_PRIORITY_IDLE))))) {
5770 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5771 new_snapshot = TRUE;
5772 }
5773
5774 proc_list_unlock();
5775
5776 freed_mem = memorystatus_kill_proc(p, cause, jetsam_reason, &killed, &footprint_of_killed_proc); /* purged and/or killed 'p' */
5777 /* Success? */
5778 if (freed_mem) {
5779 if (killed) {
5780 *memory_reclaimed = footprint_of_killed_proc;
5781 if (priority) {
5782 *priority = aPid_ep;
5783 }
5784 } else {
5785 /* purged */
5786 proc_list_lock();
5787 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5788 proc_list_unlock();
5789 }
5790 proc_rele(p);
5791 goto exit;
5792 }
5793
5794 /*
5795 * Failure - first unwind the state,
5796 * then fall through to restart the search.
5797 */
5798 proc_list_lock();
5799 proc_rele_locked(p);
5800 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5801 p->p_memstat_state |= P_MEMSTAT_ERROR;
5802 *errors += 1;
5803
5804 i = 0;
5805 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5806 }
5807 }
5808
5809 proc_list_unlock();
5810
5811 exit:
5812 os_reason_free(jetsam_reason);
5813
5814 if (!killed) {
5815 *memory_reclaimed = 0;
5816
5817 /* Clear snapshot if freshly captured and no target was found */
5818 if (new_snapshot) {
5819 proc_list_lock();
5820 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5821 proc_list_unlock();
5822 }
5823 }
5824
5825 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
5826 MEMORYSTATUS_LOG_AVAILABLE_PAGES, killed ? aPid : 0, killed, *memory_reclaimed, 0);
5827
5828 return killed;
5829 }
5830
5831 /*
5832 * Jetsam aggressively
5833 */
5834 static boolean_t
5835 memorystatus_kill_processes_aggressive(uint32_t cause, int aggr_count,
5836 int32_t priority_max, uint32_t *errors, uint64_t *memory_reclaimed)
5837 {
5838 pid_t aPid;
5839 proc_t p = PROC_NULL, next_p = PROC_NULL;
5840 boolean_t new_snapshot = FALSE, killed = FALSE;
5841 int kill_count = 0;
5842 unsigned int i = 0;
5843 int32_t aPid_ep = 0;
5844 unsigned int memorystatus_level_snapshot = 0;
5845 uint64_t killtime = 0;
5846 clock_sec_t tv_sec;
5847 clock_usec_t tv_usec;
5848 uint32_t tv_msec;
5849 os_reason_t jetsam_reason = OS_REASON_NULL;
5850 uint64_t footprint_of_killed_proc = 0;
5851
5852 *memory_reclaimed = 0;
5853
5854 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5855 MEMORYSTATUS_LOG_AVAILABLE_PAGES, priority_max, 0, 0, 0);
5856
5857 if (priority_max >= JETSAM_PRIORITY_FOREGROUND) {
5858 /*
5859 * Check if aggressive jetsam has been asked to kill upto or beyond the
5860 * JETSAM_PRIORITY_FOREGROUND bucket. If yes, sort the FG band based on
5861 * coalition footprint.
5862 */
5863 memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
5864 }
5865
5866 jetsam_reason = os_reason_create(OS_REASON_JETSAM, cause);
5867 if (jetsam_reason == OS_REASON_NULL) {
5868 printf("memorystatus_kill_processes_aggressive: failed to allocate exit reason\n");
5869 }
5870
5871 proc_list_lock();
5872
5873 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5874 while (next_p) {
5875 if (((next_p->p_listflag & P_LIST_EXITED) != 0) ||
5876 ((unsigned int)(next_p->p_memstat_effectivepriority) != i)) {
5877 /*
5878 * We have raced with next_p running on another core.
5879 * It may be exiting or it may have moved to a different
5880 * jetsam priority band. This means we have lost our
5881 * place in line while traversing the jetsam list. We
5882 * attempt to recover by rewinding to the beginning of the band
5883 * we were already traversing. By doing this, we do not guarantee
5884 * that no process escapes this aggressive march, but we can make
5885 * skipping an entire range of processes less likely. (PR-21069019)
5886 */
5887
5888 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: rewinding band %d, %s(%d) moved or exiting.\n",
5889 aggr_count, i, (*next_p->p_name ? next_p->p_name : "unknown"), next_p->p_pid);
5890
5891 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5892 continue;
5893 }
5894
5895 p = next_p;
5896 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5897
5898 if (p->p_memstat_effectivepriority > priority_max) {
5899 /*
5900 * Bail out of this killing spree if we have
5901 * reached beyond the priority_max jetsam band.
5902 * That is, we kill up to and through the
5903 * priority_max jetsam band.
5904 */
5905 proc_list_unlock();
5906 goto exit;
5907 }
5908
5909 aPid = p->p_pid;
5910 aPid_ep = p->p_memstat_effectivepriority;
5911
5912 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5913 continue;
5914 }
5915
5916 /*
5917 * Capture a snapshot if none exists.
5918 */
5919 if (memorystatus_jetsam_snapshot_count == 0) {
5920 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5921 new_snapshot = TRUE;
5922 }
5923
5924 /*
5925 * Mark as terminated so that if exit1() indicates success, but the process (for example)
5926 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
5927 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
5928 * acquisition of the proc lock.
5929 */
5930 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5931
5932 killtime = mach_absolute_time();
5933 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5934 tv_msec = tv_usec / 1000;
5935
5936 /* Shift queue, update stats */
5937 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5938
5939 /*
5940 * In order to kill the target process, we will drop the proc_list_lock.
5941 * To guaranteee that p and next_p don't disappear out from under the lock,
5942 * we must take a ref on both.
5943 * If we cannot get a reference, then it's likely we've raced with
5944 * that process exiting on another core.
5945 */
5946 if (proc_ref_locked(p) == p) {
5947 if (next_p) {
5948 while (next_p && (proc_ref_locked(next_p) != next_p)) {
5949 proc_t temp_p;
5950
5951 /*
5952 * We must have raced with next_p exiting on another core.
5953 * Recover by getting the next eligible process in the band.
5954 */
5955
5956 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: skipping %d [%s] (exiting?)\n",
5957 aggr_count, next_p->p_pid, (*next_p->p_name ? next_p->p_name : "(unknown)"));
5958
5959 temp_p = next_p;
5960 next_p = memorystatus_get_next_proc_locked(&i, temp_p, TRUE);
5961 }
5962 }
5963 proc_list_unlock();
5964
5965 printf("%lu.%03d memorystatus: %s%d pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
5966 (unsigned long)tv_sec, tv_msec,
5967 ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "killing_idle_process_aggressive" : "killing_top_process_aggressive"),
5968 aggr_count, aPid, (*p->p_name ? p->p_name : "unknown"),
5969 memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES);
5970
5971 memorystatus_level_snapshot = memorystatus_level;
5972
5973 /*
5974 * memorystatus_do_kill() drops a reference, so take another one so we can
5975 * continue to use this exit reason even after memorystatus_do_kill()
5976 * returns.
5977 */
5978 os_reason_ref(jetsam_reason);
5979 killed = memorystatus_do_kill(p, cause, jetsam_reason, &footprint_of_killed_proc);
5980
5981 /* Success? */
5982 if (killed) {
5983 *memory_reclaimed += footprint_of_killed_proc;
5984 proc_rele(p);
5985 kill_count++;
5986 p = NULL;
5987 killed = FALSE;
5988
5989 /*
5990 * Continue the killing spree.
5991 */
5992 proc_list_lock();
5993 if (next_p) {
5994 proc_rele_locked(next_p);
5995 }
5996
5997 if (aPid_ep == JETSAM_PRIORITY_FOREGROUND && memorystatus_aggressive_jetsam_lenient == TRUE) {
5998 if (memorystatus_level > memorystatus_level_snapshot && ((memorystatus_level - memorystatus_level_snapshot) >= AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD)) {
5999 #if DEVELOPMENT || DEBUG
6000 printf("Disabling Lenient mode after one-time deployment.\n");
6001 #endif /* DEVELOPMENT || DEBUG */
6002 memorystatus_aggressive_jetsam_lenient = FALSE;
6003 break;
6004 }
6005 }
6006
6007 continue;
6008 }
6009
6010 /*
6011 * Failure - first unwind the state,
6012 * then fall through to restart the search.
6013 */
6014 proc_list_lock();
6015 proc_rele_locked(p);
6016 if (next_p) {
6017 proc_rele_locked(next_p);
6018 }
6019 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6020 p->p_memstat_state |= P_MEMSTAT_ERROR;
6021 *errors += 1;
6022 p = NULL;
6023 }
6024
6025 /*
6026 * Failure - restart the search at the beginning of
6027 * the band we were already traversing.
6028 *
6029 * We might have raced with "p" exiting on another core, resulting in no
6030 * ref on "p". Or, we may have failed to kill "p".
6031 *
6032 * Either way, we fall thru to here, leaving the proc in the
6033 * P_MEMSTAT_TERMINATED or P_MEMSTAT_ERROR state.
6034 *
6035 * And, we hold the the proc_list_lock at this point.
6036 */
6037
6038 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6039 }
6040
6041 proc_list_unlock();
6042
6043 exit:
6044 os_reason_free(jetsam_reason);
6045
6046 /* Clear snapshot if freshly captured and no target was found */
6047 if (new_snapshot && (kill_count == 0)) {
6048 proc_list_lock();
6049 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6050 proc_list_unlock();
6051 }
6052
6053 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
6054 MEMORYSTATUS_LOG_AVAILABLE_PAGES, 0, kill_count, *memory_reclaimed, 0);
6055
6056 if (kill_count > 0) {
6057 return TRUE;
6058 } else {
6059 return FALSE;
6060 }
6061 }
6062
6063 static boolean_t
6064 memorystatus_kill_hiwat_proc(uint32_t *errors, boolean_t *purged, uint64_t *memory_reclaimed)
6065 {
6066 pid_t aPid = 0;
6067 proc_t p = PROC_NULL, next_p = PROC_NULL;
6068 boolean_t new_snapshot = FALSE, killed = FALSE, freed_mem = FALSE;
6069 unsigned int i = 0;
6070 uint32_t aPid_ep;
6071 os_reason_t jetsam_reason = OS_REASON_NULL;
6072 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
6073 MEMORYSTATUS_LOG_AVAILABLE_PAGES, 0, 0, 0, 0);
6074
6075 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_HIGHWATER);
6076 if (jetsam_reason == OS_REASON_NULL) {
6077 printf("memorystatus_kill_hiwat_proc: failed to allocate exit reason\n");
6078 }
6079
6080 proc_list_lock();
6081
6082 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6083 while (next_p) {
6084 uint64_t footprint_in_bytes = 0;
6085 uint64_t memlimit_in_bytes = 0;
6086 boolean_t skip = 0;
6087
6088 p = next_p;
6089 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6090
6091 aPid = p->p_pid;
6092 aPid_ep = p->p_memstat_effectivepriority;
6093
6094 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
6095 continue;
6096 }
6097
6098 /* skip if no limit set */
6099 if (p->p_memstat_memlimit <= 0) {
6100 continue;
6101 }
6102
6103 footprint_in_bytes = get_task_phys_footprint(p->task);
6104 memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
6105 skip = (footprint_in_bytes <= memlimit_in_bytes);
6106
6107 #if CONFIG_FREEZE
6108 if (!skip) {
6109 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
6110 skip = TRUE;
6111 } else {
6112 skip = FALSE;
6113 }
6114 }
6115 #endif
6116
6117 if (skip) {
6118 continue;
6119 } else {
6120 if (memorystatus_jetsam_snapshot_count == 0) {
6121 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
6122 new_snapshot = TRUE;
6123 }
6124
6125 if (proc_ref_locked(p) == p) {
6126 /*
6127 * Mark as terminated so that if exit1() indicates success, but the process (for example)
6128 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
6129 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
6130 * acquisition of the proc lock.
6131 */
6132 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
6133
6134 proc_list_unlock();
6135 } else {
6136 /*
6137 * We need to restart the search again because
6138 * proc_ref_locked _can_ drop the proc_list lock
6139 * and we could have lost our stored next_p via
6140 * an exit() on another core.
6141 */
6142 i = 0;
6143 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6144 continue;
6145 }
6146
6147 footprint_in_bytes = 0;
6148 freed_mem = memorystatus_kill_proc(p, kMemorystatusKilledHiwat, jetsam_reason, &killed, &footprint_in_bytes); /* purged and/or killed 'p' */
6149
6150 /* Success? */
6151 if (freed_mem) {
6152 if (killed == FALSE) {
6153 /* purged 'p'..don't reset HWM candidate count */
6154 *purged = TRUE;
6155
6156 proc_list_lock();
6157 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6158 proc_list_unlock();
6159 } else {
6160 *memory_reclaimed = footprint_in_bytes;
6161 }
6162 proc_rele(p);
6163 goto exit;
6164 }
6165 /*
6166 * Failure - first unwind the state,
6167 * then fall through to restart the search.
6168 */
6169 proc_list_lock();
6170 proc_rele_locked(p);
6171 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6172 p->p_memstat_state |= P_MEMSTAT_ERROR;
6173 *errors += 1;
6174
6175 i = 0;
6176 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6177 }
6178 }
6179
6180 proc_list_unlock();
6181
6182 exit:
6183 os_reason_free(jetsam_reason);
6184
6185 if (!killed) {
6186 *memory_reclaimed = 0;
6187
6188 /* Clear snapshot if freshly captured and no target was found */
6189 if (new_snapshot) {
6190 proc_list_lock();
6191 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6192 proc_list_unlock();
6193 }
6194 }
6195
6196 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_END,
6197 MEMORYSTATUS_LOG_AVAILABLE_PAGES, killed ? aPid : 0, killed, *memory_reclaimed, 0);
6198
6199 return killed;
6200 }
6201
6202 /*
6203 * Jetsam a process pinned in the elevated band.
6204 *
6205 * Return: true -- a pinned process was jetsammed
6206 * false -- no pinned process was jetsammed
6207 */
6208 boolean_t
6209 memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, unsigned int band, int aggr_count, uint32_t *errors, uint64_t *memory_reclaimed)
6210 {
6211 pid_t aPid = 0;
6212 proc_t p = PROC_NULL, next_p = PROC_NULL;
6213 boolean_t new_snapshot = FALSE, killed = FALSE;
6214 int kill_count = 0;
6215 uint32_t aPid_ep;
6216 uint64_t killtime = 0;
6217 clock_sec_t tv_sec;
6218 clock_usec_t tv_usec;
6219 uint32_t tv_msec;
6220 uint64_t footprint_of_killed_proc = 0;
6221
6222
6223 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
6224 MEMORYSTATUS_LOG_AVAILABLE_PAGES, 0, 0, 0, 0);
6225
6226 #if CONFIG_FREEZE
6227 boolean_t consider_frozen_only = FALSE;
6228
6229 if (band == (unsigned int) memorystatus_freeze_jetsam_band) {
6230 consider_frozen_only = TRUE;
6231 }
6232 #endif /* CONFIG_FREEZE */
6233
6234 proc_list_lock();
6235
6236 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
6237 while (next_p) {
6238 p = next_p;
6239 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
6240
6241 aPid = p->p_pid;
6242 aPid_ep = p->p_memstat_effectivepriority;
6243
6244 /*
6245 * Only pick a process pinned in this elevated band
6246 */
6247 if (!(p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
6248 continue;
6249 }
6250
6251 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
6252 continue;
6253 }
6254
6255 #if CONFIG_FREEZE
6256 if (consider_frozen_only && !(p->p_memstat_state & P_MEMSTAT_FROZEN)) {
6257 continue;
6258 }
6259
6260 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
6261 continue;
6262 }
6263 #endif /* CONFIG_FREEZE */
6264
6265 #if DEVELOPMENT || DEBUG
6266 MEMORYSTATUS_DEBUG(1, "jetsam: elevated%d process pid %d [%s] - memorystatus_available_pages: %d\n",
6267 aggr_count,
6268 aPid, (*p->p_name ? p->p_name : "unknown"),
6269 MEMORYSTATUS_LOG_AVAILABLE_PAGES);
6270 #endif /* DEVELOPMENT || DEBUG */
6271
6272 if (memorystatus_jetsam_snapshot_count == 0) {
6273 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
6274 new_snapshot = TRUE;
6275 }
6276
6277 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
6278
6279 killtime = mach_absolute_time();
6280 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
6281 tv_msec = tv_usec / 1000;
6282
6283 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
6284
6285 if (proc_ref_locked(p) == p) {
6286 proc_list_unlock();
6287
6288 /*
6289 * memorystatus_do_kill drops a reference, so take another one so we can
6290 * continue to use this exit reason even after memorystatus_do_kill()
6291 * returns
6292 */
6293 os_reason_ref(jetsam_reason);
6294 killed = memorystatus_do_kill(p, cause, jetsam_reason, &footprint_of_killed_proc);
6295
6296 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_top_process_elevated%d pid %d [%s] (%s %d) %lluKB - memorystatus_available_pages: %llu\n",
6297 (unsigned long)tv_sec, tv_msec,
6298 aggr_count,
6299 aPid, ((p && *p->p_name) ? p->p_name : "unknown"),
6300 memorystatus_kill_cause_name[cause], aPid_ep,
6301 footprint_of_killed_proc >> 10, (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES);
6302
6303 /* Success? */
6304 if (killed) {
6305 *memory_reclaimed = footprint_of_killed_proc;
6306 proc_rele(p);
6307 kill_count++;
6308 goto exit;
6309 }
6310
6311 /*
6312 * Failure - first unwind the state,
6313 * then fall through to restart the search.
6314 */
6315 proc_list_lock();
6316 proc_rele_locked(p);
6317 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6318 p->p_memstat_state |= P_MEMSTAT_ERROR;
6319 *errors += 1;
6320 }
6321
6322 /*
6323 * Failure - restart the search.
6324 *
6325 * We might have raced with "p" exiting on another core, resulting in no
6326 * ref on "p". Or, we may have failed to kill "p".
6327 *
6328 * Either way, we fall thru to here, leaving the proc in the
6329 * P_MEMSTAT_TERMINATED state or P_MEMSTAT_ERROR state.
6330 *
6331 * And, we hold the the proc_list_lock at this point.
6332 */
6333
6334 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
6335 }
6336
6337 proc_list_unlock();
6338
6339 exit:
6340 os_reason_free(jetsam_reason);
6341
6342 if (kill_count == 0) {
6343 *memory_reclaimed = 0;
6344
6345 /* Clear snapshot if freshly captured and no target was found */
6346 if (new_snapshot) {
6347 proc_list_lock();
6348 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6349 proc_list_unlock();
6350 }
6351 }
6352
6353 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
6354 MEMORYSTATUS_LOG_AVAILABLE_PAGES, killed ? aPid : 0, kill_count, *memory_reclaimed, 0);
6355
6356 return killed;
6357 }
6358
6359 static boolean_t
6360 memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause)
6361 {
6362 /*
6363 * TODO: allow a general async path
6364 *
6365 * NOTE: If a new async kill cause is added, make sure to update memorystatus_thread() to
6366 * add the appropriate exit reason code mapping.
6367 */
6368 if ((victim_pid != -1) ||
6369 (cause != kMemorystatusKilledVMPageShortage &&
6370 cause != kMemorystatusKilledVMCompressorThrashing &&
6371 cause != kMemorystatusKilledVMCompressorSpaceShortage &&
6372 cause != kMemorystatusKilledFCThrashing &&
6373 cause != kMemorystatusKilledZoneMapExhaustion)) {
6374 return FALSE;
6375 }
6376
6377 kill_under_pressure_cause = cause;
6378 memorystatus_thread_wake();
6379 return TRUE;
6380 }
6381
6382 boolean_t
6383 memorystatus_kill_on_VM_compressor_space_shortage(boolean_t async)
6384 {
6385 if (async) {
6386 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMCompressorSpaceShortage);
6387 } else {
6388 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMCOMPRESSOR_SPACE_SHORTAGE);
6389 if (jetsam_reason == OS_REASON_NULL) {
6390 printf("memorystatus_kill_on_VM_compressor_space_shortage -- sync: failed to allocate jetsam reason\n");
6391 }
6392
6393 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMCompressorSpaceShortage, jetsam_reason);
6394 }
6395 }
6396
6397 #if CONFIG_JETSAM
6398 boolean_t
6399 memorystatus_kill_on_VM_compressor_thrashing(boolean_t async)
6400 {
6401 if (async) {
6402 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMCompressorThrashing);
6403 } else {
6404 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMCOMPRESSOR_THRASHING);
6405 if (jetsam_reason == OS_REASON_NULL) {
6406 printf("memorystatus_kill_on_VM_compressor_thrashing -- sync: failed to allocate jetsam reason\n");
6407 }
6408
6409 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMCompressorThrashing, jetsam_reason);
6410 }
6411 }
6412
6413 boolean_t
6414 memorystatus_kill_on_VM_page_shortage(boolean_t async)
6415 {
6416 if (async) {
6417 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMPageShortage);
6418 } else {
6419 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMPAGESHORTAGE);
6420 if (jetsam_reason == OS_REASON_NULL) {
6421 printf("memorystatus_kill_on_VM_page_shortage -- sync: failed to allocate jetsam reason\n");
6422 }
6423
6424 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMPageShortage, jetsam_reason);
6425 }
6426 }
6427
6428 boolean_t
6429 memorystatus_kill_on_FC_thrashing(boolean_t async)
6430 {
6431 if (async) {
6432 return memorystatus_kill_process_async(-1, kMemorystatusKilledFCThrashing);
6433 } else {
6434 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_FCTHRASHING);
6435 if (jetsam_reason == OS_REASON_NULL) {
6436 printf("memorystatus_kill_on_FC_thrashing -- sync: failed to allocate jetsam reason\n");
6437 }
6438
6439 return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing, jetsam_reason);
6440 }
6441 }
6442
6443 boolean_t
6444 memorystatus_kill_on_vnode_limit(void)
6445 {
6446 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_VNODE);
6447 if (jetsam_reason == OS_REASON_NULL) {
6448 printf("memorystatus_kill_on_vnode_limit: failed to allocate jetsam reason\n");
6449 }
6450
6451 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes, jetsam_reason);
6452 }
6453
6454 #endif /* CONFIG_JETSAM */
6455
6456 boolean_t
6457 memorystatus_kill_on_zone_map_exhaustion(pid_t pid)
6458 {
6459 boolean_t res = FALSE;
6460 if (pid == -1) {
6461 res = memorystatus_kill_process_async(-1, kMemorystatusKilledZoneMapExhaustion);
6462 } else {
6463 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_ZONE_MAP_EXHAUSTION);
6464 if (jetsam_reason == OS_REASON_NULL) {
6465 printf("memorystatus_kill_on_zone_map_exhaustion: failed to allocate jetsam reason\n");
6466 }
6467
6468 res = memorystatus_kill_process_sync(pid, kMemorystatusKilledZoneMapExhaustion, jetsam_reason);
6469 }
6470 return res;
6471 }
6472
6473 void
6474 memorystatus_on_pageout_scan_end(void)
6475 {
6476 /* No-op */
6477 }
6478
6479 /* Return both allocated and actual size, since there's a race between allocation and list compilation */
6480 static int
6481 memorystatus_get_priority_list(memorystatus_priority_entry_t **list_ptr, size_t *buffer_size, size_t *list_size, boolean_t size_only)
6482 {
6483 uint32_t list_count, i = 0;
6484 memorystatus_priority_entry_t *list_entry;
6485 proc_t p;
6486
6487 list_count = memorystatus_list_count;
6488 *list_size = sizeof(memorystatus_priority_entry_t) * list_count;
6489
6490 /* Just a size check? */
6491 if (size_only) {
6492 return 0;
6493 }
6494
6495 /* Otherwise, validate the size of the buffer */
6496 if (*buffer_size < *list_size) {
6497 return EINVAL;
6498 }
6499
6500 *list_ptr = kheap_alloc(KHEAP_TEMP, *list_size, Z_WAITOK | Z_ZERO);
6501 if (!*list_ptr) {
6502 return ENOMEM;
6503 }
6504
6505 *buffer_size = *list_size;
6506 *list_size = 0;
6507
6508 list_entry = *list_ptr;
6509
6510 proc_list_lock();
6511
6512 p = memorystatus_get_first_proc_locked(&i, TRUE);
6513 while (p && (*list_size < *buffer_size)) {
6514 list_entry->pid = p->p_pid;
6515 list_entry->priority = p->p_memstat_effectivepriority;
6516 list_entry->user_data = p->p_memstat_userdata;
6517
6518 if (p->p_memstat_memlimit <= 0) {
6519 task_get_phys_footprint_limit(p->task, &list_entry->limit);
6520 } else {
6521 list_entry->limit = p->p_memstat_memlimit;
6522 }
6523
6524 list_entry->state = memorystatus_build_state(p);
6525 list_entry++;
6526
6527 *list_size += sizeof(memorystatus_priority_entry_t);
6528
6529 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6530 }
6531
6532 proc_list_unlock();
6533
6534 MEMORYSTATUS_DEBUG(1, "memorystatus_get_priority_list: returning %lu for size\n", (unsigned long)*list_size);
6535
6536 return 0;
6537 }
6538
6539 static int
6540 memorystatus_get_priority_pid(pid_t pid, user_addr_t buffer, size_t buffer_size)
6541 {
6542 int error = 0;
6543 memorystatus_priority_entry_t mp_entry;
6544 kern_return_t ret;
6545
6546 /* Validate inputs */
6547 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_entry_t))) {
6548 return EINVAL;
6549 }
6550
6551 proc_t p = proc_find(pid);
6552 if (!p) {
6553 return ESRCH;
6554 }
6555
6556 memset(&mp_entry, 0, sizeof(memorystatus_priority_entry_t));
6557
6558 mp_entry.pid = p->p_pid;
6559 mp_entry.priority = p->p_memstat_effectivepriority;
6560 mp_entry.user_data = p->p_memstat_userdata;
6561 if (p->p_memstat_memlimit <= 0) {
6562 ret = task_get_phys_footprint_limit(p->task, &mp_entry.limit);
6563 if (ret != KERN_SUCCESS) {
6564 proc_rele(p);
6565 return EINVAL;
6566 }
6567 } else {
6568 mp_entry.limit = p->p_memstat_memlimit;
6569 }
6570 mp_entry.state = memorystatus_build_state(p);
6571
6572 proc_rele(p);
6573
6574 error = copyout(&mp_entry, buffer, buffer_size);
6575
6576 return error;
6577 }
6578
6579 static int
6580 memorystatus_cmd_get_priority_list(pid_t pid, user_addr_t buffer, size_t buffer_size, int32_t *retval)
6581 {
6582 int error = 0;
6583 boolean_t size_only;
6584 size_t list_size;
6585
6586 /*
6587 * When a non-zero pid is provided, the 'list' has only one entry.
6588 */
6589
6590 size_only = ((buffer == USER_ADDR_NULL) ? TRUE: FALSE);
6591
6592 if (pid != 0) {
6593 list_size = sizeof(memorystatus_priority_entry_t) * 1;
6594 if (!size_only) {
6595 error = memorystatus_get_priority_pid(pid, buffer, buffer_size);
6596 }
6597 } else {
6598 memorystatus_priority_entry_t *list = NULL;
6599 error = memorystatus_get_priority_list(&list, &buffer_size, &list_size, size_only);
6600
6601 if (error == 0) {
6602 if (!size_only) {
6603 error = copyout(list, buffer, list_size);
6604 }
6605 }
6606
6607 if (list) {
6608 kheap_free(KHEAP_TEMP, list, buffer_size);
6609 }
6610 }
6611
6612 if (error == 0) {
6613 assert(list_size <= INT32_MAX);
6614 *retval = (int32_t) list_size;
6615 }
6616
6617 return error;
6618 }
6619
6620 static void
6621 memorystatus_clear_errors(void)
6622 {
6623 proc_t p;
6624 unsigned int i = 0;
6625
6626 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_START, 0, 0, 0, 0, 0);
6627
6628 proc_list_lock();
6629
6630 p = memorystatus_get_first_proc_locked(&i, TRUE);
6631 while (p) {
6632 if (p->p_memstat_state & P_MEMSTAT_ERROR) {
6633 p->p_memstat_state &= ~P_MEMSTAT_ERROR;
6634 }
6635 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6636 }
6637
6638 proc_list_unlock();
6639
6640 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_END, 0, 0, 0, 0, 0);
6641 }
6642
6643 #if CONFIG_JETSAM
6644 static void
6645 memorystatus_update_levels_locked(boolean_t critical_only)
6646 {
6647 memorystatus_available_pages_critical = memorystatus_available_pages_critical_base;
6648
6649 /*
6650 * If there's an entry in the first bucket, we have idle processes.
6651 */
6652
6653 memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
6654 if (first_bucket->count) {
6655 memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
6656
6657 if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure) {
6658 /*
6659 * The critical threshold must never exceed the pressure threshold
6660 */
6661 memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
6662 }
6663 }
6664
6665 if (memorystatus_jetsam_policy & kPolicyMoreFree) {
6666 memorystatus_available_pages_critical += memorystatus_policy_more_free_offset_pages;
6667 }
6668
6669 if (critical_only) {
6670 return;
6671 }
6672
6673 #if VM_PRESSURE_EVENTS
6674 memorystatus_available_pages_pressure = (int32_t)(pressure_threshold_percentage * (atop_64(max_mem) / 100));
6675 #endif
6676 }
6677
6678 void
6679 memorystatus_fast_jetsam_override(boolean_t enable_override)
6680 {
6681 /* If fast jetsam is not enabled, simply return */
6682 if (!fast_jetsam_enabled) {
6683 return;
6684 }
6685
6686 if (enable_override) {
6687 if ((memorystatus_jetsam_policy & kPolicyMoreFree) == kPolicyMoreFree) {
6688 return;
6689 }
6690 proc_list_lock();
6691 memorystatus_jetsam_policy |= kPolicyMoreFree;
6692 memorystatus_thread_pool_max();
6693 memorystatus_update_levels_locked(TRUE);
6694 proc_list_unlock();
6695 } else {
6696 if ((memorystatus_jetsam_policy & kPolicyMoreFree) == 0) {
6697 return;
6698 }
6699 proc_list_lock();
6700 memorystatus_jetsam_policy &= ~kPolicyMoreFree;
6701 memorystatus_thread_pool_default();
6702 memorystatus_update_levels_locked(TRUE);
6703 proc_list_unlock();
6704 }
6705 }
6706
6707
6708 static int
6709 sysctl_kern_memorystatus_policy_more_free SYSCTL_HANDLER_ARGS
6710 {
6711 #pragma unused(arg1, arg2, oidp)
6712 int error = 0, more_free = 0;
6713
6714 /*
6715 * TODO: Enable this privilege check?
6716 *
6717 * error = priv_check_cred(kauth_cred_get(), PRIV_VM_JETSAM, 0);
6718 * if (error)
6719 * return (error);
6720 */
6721
6722 error = sysctl_handle_int(oidp, &more_free, 0, req);
6723 if (error || !req->newptr) {
6724 return error;
6725 }
6726
6727 if (more_free) {
6728 memorystatus_fast_jetsam_override(true);
6729 } else {
6730 memorystatus_fast_jetsam_override(false);
6731 }
6732
6733 return 0;
6734 }
6735 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_policy_more_free, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
6736 0, 0, &sysctl_kern_memorystatus_policy_more_free, "I", "");
6737
6738 #endif /* CONFIG_JETSAM */
6739
6740 /*
6741 * Get the at_boot snapshot
6742 */
6743 static int
6744 memorystatus_get_at_boot_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6745 {
6746 size_t input_size = *snapshot_size;
6747
6748 /*
6749 * The at_boot snapshot has no entry list.
6750 */
6751 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t);
6752
6753 if (size_only) {
6754 return 0;
6755 }
6756
6757 /*
6758 * Validate the size of the snapshot buffer
6759 */
6760 if (input_size < *snapshot_size) {
6761 return EINVAL;
6762 }
6763
6764 /*
6765 * Update the notification_time only
6766 */
6767 memorystatus_at_boot_snapshot.notification_time = mach_absolute_time();
6768 *snapshot = &memorystatus_at_boot_snapshot;
6769
6770 MEMORYSTATUS_DEBUG(7, "memorystatus_get_at_boot_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%d)\n",
6771 (long)input_size, (long)*snapshot_size, 0);
6772 return 0;
6773 }
6774
6775 /*
6776 * Get the previous fully populated snapshot
6777 */
6778 static int
6779 memorystatus_get_jetsam_snapshot_copy(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6780 {
6781 size_t input_size = *snapshot_size;
6782
6783 if (memorystatus_jetsam_snapshot_copy_count > 0) {
6784 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_copy_count));
6785 } else {
6786 *snapshot_size = 0;
6787 }
6788
6789 if (size_only) {
6790 return 0;
6791 }
6792
6793 if (input_size < *snapshot_size) {
6794 return EINVAL;
6795 }
6796
6797 *snapshot = memorystatus_jetsam_snapshot_copy;
6798
6799 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot_copy: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6800 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_copy_count);
6801
6802 return 0;
6803 }
6804
6805 #if CONFIG_FREEZE
6806 static int
6807 memorystatus_get_jetsam_snapshot_freezer(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6808 {
6809 size_t input_size = *snapshot_size;
6810
6811 if (memorystatus_jetsam_snapshot_freezer->entry_count > 0) {
6812 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_freezer->entry_count));
6813 } else {
6814 *snapshot_size = 0;
6815 }
6816 assert(*snapshot_size <= memorystatus_jetsam_snapshot_freezer_size);
6817
6818 if (size_only) {
6819 return 0;
6820 }
6821
6822 if (input_size < *snapshot_size) {
6823 return EINVAL;
6824 }
6825
6826 *snapshot = memorystatus_jetsam_snapshot_freezer;
6827
6828 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot_freezer: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6829 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_freezer->entry_count);
6830
6831 return 0;
6832 }
6833 #endif /* CONFIG_FREEZE */
6834
6835 static int
6836 memorystatus_get_on_demand_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6837 {
6838 size_t input_size = *snapshot_size;
6839 uint32_t ods_list_count = memorystatus_list_count;
6840 memorystatus_jetsam_snapshot_t *ods = NULL; /* The on_demand snapshot buffer */
6841
6842 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (ods_list_count));
6843
6844 if (size_only) {
6845 return 0;
6846 }
6847
6848 /*
6849 * Validate the size of the snapshot buffer.
6850 * This is inherently racey. May want to revisit
6851 * this error condition and trim the output when
6852 * it doesn't fit.
6853 */
6854 if (input_size < *snapshot_size) {
6855 return EINVAL;
6856 }
6857
6858 /*
6859 * Allocate and initialize a snapshot buffer.
6860 */
6861 ods = kalloc(*snapshot_size);
6862 if (!ods) {
6863 return ENOMEM;
6864 }
6865
6866 memset(ods, 0, *snapshot_size);
6867
6868 proc_list_lock();
6869 memorystatus_init_jetsam_snapshot_locked(ods, ods_list_count);
6870 proc_list_unlock();
6871
6872 /*
6873 * Return the kernel allocated, on_demand buffer.
6874 * The caller of this routine will copy the data out
6875 * to user space and then free the kernel allocated
6876 * buffer.
6877 */
6878 *snapshot = ods;
6879
6880 MEMORYSTATUS_DEBUG(7, "memorystatus_get_on_demand_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6881 (long)input_size, (long)*snapshot_size, (long)ods_list_count);
6882
6883 return 0;
6884 }
6885
6886 static int
6887 memorystatus_get_jetsam_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6888 {
6889 size_t input_size = *snapshot_size;
6890
6891 if (memorystatus_jetsam_snapshot_count > 0) {
6892 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count));
6893 } else {
6894 *snapshot_size = 0;
6895 }
6896
6897 if (size_only) {
6898 return 0;
6899 }
6900
6901 if (input_size < *snapshot_size) {
6902 return EINVAL;
6903 }
6904
6905 *snapshot = memorystatus_jetsam_snapshot;
6906
6907 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6908 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_count);
6909
6910 return 0;
6911 }
6912
6913
6914 static int
6915 memorystatus_cmd_get_jetsam_snapshot(int32_t flags, user_addr_t buffer, size_t buffer_size, int32_t *retval)
6916 {
6917 int error = EINVAL;
6918 boolean_t size_only;
6919 boolean_t is_default_snapshot = FALSE;
6920 boolean_t is_on_demand_snapshot = FALSE;
6921 boolean_t is_at_boot_snapshot = FALSE;
6922 #if CONFIG_FREEZE
6923 bool is_freezer_snapshot = false;
6924 #endif /* CONFIG_FREEZE */
6925 memorystatus_jetsam_snapshot_t *snapshot;
6926
6927 size_only = ((buffer == USER_ADDR_NULL) ? TRUE : FALSE);
6928
6929 if (flags == 0) {
6930 /* Default */
6931 is_default_snapshot = TRUE;
6932 error = memorystatus_get_jetsam_snapshot(&snapshot, &buffer_size, size_only);
6933 } else {
6934 if (flags & ~(MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT | MEMORYSTATUS_SNAPSHOT_COPY | MEMORYSTATUS_FLAGS_SNAPSHOT_FREEZER)) {
6935 /*
6936 * Unsupported bit set in flag.
6937 */
6938 return EINVAL;
6939 }
6940
6941 if (flags & (flags - 0x1)) {
6942 /*
6943 * Can't have multiple flags set at the same time.
6944 */
6945 return EINVAL;
6946 }
6947
6948 if (flags & MEMORYSTATUS_SNAPSHOT_ON_DEMAND) {
6949 is_on_demand_snapshot = TRUE;
6950 /*
6951 * When not requesting the size only, the following call will allocate
6952 * an on_demand snapshot buffer, which is freed below.
6953 */
6954 error = memorystatus_get_on_demand_snapshot(&snapshot, &buffer_size, size_only);
6955 } else if (flags & MEMORYSTATUS_SNAPSHOT_AT_BOOT) {
6956 is_at_boot_snapshot = TRUE;
6957 error = memorystatus_get_at_boot_snapshot(&snapshot, &buffer_size, size_only);
6958 } else if (flags & MEMORYSTATUS_SNAPSHOT_COPY) {
6959 error = memorystatus_get_jetsam_snapshot_copy(&snapshot, &buffer_size, size_only);
6960 #if CONFIG_FREEZE
6961 } else if (flags & MEMORYSTATUS_FLAGS_SNAPSHOT_FREEZER) {
6962 is_freezer_snapshot = true;
6963 error = memorystatus_get_jetsam_snapshot_freezer(&snapshot, &buffer_size, size_only);
6964 #endif /* CONFIG_FREEZE */
6965 } else {
6966 /*
6967 * Invalid flag setting.
6968 */
6969 return EINVAL;
6970 }
6971 }
6972
6973 if (error) {
6974 goto out;
6975 }
6976
6977 /*
6978 * Copy the data out to user space and clear the snapshot buffer.
6979 * If working with the jetsam snapshot,
6980 * clearing the buffer means, reset the count.
6981 * If working with an on_demand snapshot
6982 * clearing the buffer means, free it.
6983 * If working with the at_boot snapshot
6984 * there is nothing to clear or update.
6985 * If working with a copy of the snapshot
6986 * there is nothing to clear or update.
6987 * If working with the freezer snapshot
6988 * clearing the buffer means, reset the count.
6989 */
6990 if (!size_only) {
6991 if ((error = copyout(snapshot, buffer, buffer_size)) == 0) {
6992 #if CONFIG_FREEZE
6993 if (is_default_snapshot || is_freezer_snapshot) {
6994 #else
6995 if (is_default_snapshot) {
6996 #endif /* CONFIG_FREEZE */
6997 /*
6998 * The jetsam snapshot is never freed, its count is simply reset.
6999 * However, we make a copy for any parties that might be interested
7000 * in the previous fully populated snapshot.
7001 */
7002 proc_list_lock();
7003 #if DEVELOPMENT || DEBUG
7004 if (memorystatus_testing_pid != 0 && memorystatus_testing_pid != current_proc()->p_pid) {
7005 /* Snapshot is currently owned by someone else. Don't consume it. */
7006 proc_list_unlock();
7007 goto out;
7008 }
7009 #endif /* (DEVELOPMENT || DEBUG)*/
7010 if (is_default_snapshot) {
7011 memcpy(memorystatus_jetsam_snapshot_copy, memorystatus_jetsam_snapshot, memorystatus_jetsam_snapshot_size);
7012 memorystatus_jetsam_snapshot_copy_count = memorystatus_jetsam_snapshot_count;
7013 snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
7014 memorystatus_jetsam_snapshot_last_timestamp = 0;
7015 }
7016 #if CONFIG_FREEZE
7017 else if (is_freezer_snapshot) {
7018 memorystatus_jetsam_snapshot_freezer->entry_count = 0;
7019 }
7020 #endif /* CONFIG_FREEZE */
7021 proc_list_unlock();
7022 }
7023 }
7024
7025 if (is_on_demand_snapshot) {
7026 /*
7027 * The on_demand snapshot is always freed,
7028 * even if the copyout failed.
7029 */
7030 if (snapshot) {
7031 kfree(snapshot, buffer_size);
7032 }
7033 }
7034 }
7035
7036 out:
7037 if (error == 0) {
7038 assert(buffer_size <= INT32_MAX);
7039 *retval = (int32_t) buffer_size;
7040 }
7041 return error;
7042 }
7043
7044 #if DEVELOPMENT || DEBUG
7045 static int
7046 memorystatus_cmd_set_testing_pid(int32_t flags)
7047 {
7048 int error = EINVAL;
7049 proc_t caller = current_proc();
7050 assert(caller != kernproc);
7051 proc_list_lock();
7052 if (flags & MEMORYSTATUS_FLAGS_SET_TESTING_PID) {
7053 if (memorystatus_testing_pid == 0) {
7054 memorystatus_testing_pid = caller->p_pid;
7055 error = 0;
7056 } else if (memorystatus_testing_pid == caller->p_pid) {
7057 error = 0;
7058 } else {
7059 /* We don't allow ownership to be taken from another proc. */
7060 error = EBUSY;
7061 }
7062 } else if (flags & MEMORYSTATUS_FLAGS_UNSET_TESTING_PID) {
7063 if (memorystatus_testing_pid == caller->p_pid) {
7064 memorystatus_testing_pid = 0;
7065 error = 0;
7066 } else if (memorystatus_testing_pid != 0) {
7067 /* We don't allow ownership to be taken from another proc. */
7068 error = EPERM;
7069 }
7070 }
7071 proc_list_unlock();
7072
7073 return error;
7074 }
7075 #endif /* DEVELOPMENT || DEBUG */
7076
7077 /*
7078 * Routine: memorystatus_cmd_grp_set_priorities
7079 * Purpose: Update priorities for a group of processes.
7080 *
7081 * [priority]
7082 * Move each process out of its effective priority
7083 * band and into a new priority band.
7084 * Maintains relative order from lowest to highest priority.
7085 * In single band, maintains relative order from head to tail.
7086 *
7087 * eg: before [effectivepriority | pid]
7088 * [18 | p101 ]
7089 * [17 | p55, p67, p19 ]
7090 * [12 | p103 p10 ]
7091 * [ 7 | p25 ]
7092 * [ 0 | p71, p82, ]
7093 *
7094 * after [ new band | pid]
7095 * [ xxx | p71, p82, p25, p103, p10, p55, p67, p19, p101]
7096 *
7097 * Returns: 0 on success, else non-zero.
7098 *
7099 * Caveat: We know there is a race window regarding recycled pids.
7100 * A process could be killed before the kernel can act on it here.
7101 * If a pid cannot be found in any of the jetsam priority bands,
7102 * then we simply ignore it. No harm.
7103 * But, if the pid has been recycled then it could be an issue.
7104 * In that scenario, we might move an unsuspecting process to the new
7105 * priority band. It's not clear how the kernel can safeguard
7106 * against this, but it would be an extremely rare case anyway.
7107 * The caller of this api might avoid such race conditions by
7108 * ensuring that the processes passed in the pid list are suspended.
7109 */
7110
7111
7112 static int
7113 memorystatus_cmd_grp_set_priorities(user_addr_t buffer, size_t buffer_size)
7114 {
7115 /*
7116 * We only handle setting priority
7117 * per process
7118 */
7119
7120 int error = 0;
7121 memorystatus_properties_entry_v1_t *entries = NULL;
7122 size_t entry_count = 0;
7123
7124 /* This will be the ordered proc list */
7125 typedef struct memorystatus_internal_properties {
7126 proc_t proc;
7127 int32_t priority;
7128 } memorystatus_internal_properties_t;
7129
7130 memorystatus_internal_properties_t *table = NULL;
7131 size_t table_size = 0;
7132 uint32_t table_count = 0;
7133
7134 size_t i = 0;
7135 uint32_t bucket_index = 0;
7136 boolean_t head_insert;
7137 int32_t new_priority;
7138
7139 proc_t p;
7140
7141 /* Verify inputs */
7142 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
7143 error = EINVAL;
7144 goto out;
7145 }
7146
7147 entry_count = (buffer_size / sizeof(memorystatus_properties_entry_v1_t));
7148 if (entry_count == 0) {
7149 /* buffer size was not large enough for a single entry */
7150 error = EINVAL;
7151 goto out;
7152 }
7153
7154 if ((entries = kheap_alloc(KHEAP_TEMP, buffer_size, Z_WAITOK)) == NULL) {
7155 error = ENOMEM;
7156 goto out;
7157 }
7158
7159 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_START, MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY, entry_count, 0, 0, 0);
7160
7161 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
7162 goto out;
7163 }
7164
7165 /* Verify sanity of input priorities */
7166 if (entries[0].version == MEMORYSTATUS_MPE_VERSION_1) {
7167 if ((buffer_size % MEMORYSTATUS_MPE_VERSION_1_SIZE) != 0) {
7168 error = EINVAL;
7169 goto out;
7170 }
7171 } else {
7172 error = EINVAL;
7173 goto out;
7174 }
7175
7176 for (i = 0; i < entry_count; i++) {
7177 if (entries[i].priority == -1) {
7178 /* Use as shorthand for default priority */
7179 entries[i].priority = JETSAM_PRIORITY_DEFAULT;
7180 } else if ((entries[i].priority == system_procs_aging_band) || (entries[i].priority == applications_aging_band)) {
7181 /* Both the aging bands are reserved for internal use;
7182 * if requested, adjust to JETSAM_PRIORITY_IDLE. */
7183 entries[i].priority = JETSAM_PRIORITY_IDLE;
7184 } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
7185 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle
7186 * queue */
7187 /* Deal with this later */
7188 } else if ((entries[i].priority < 0) || (entries[i].priority >= MEMSTAT_BUCKET_COUNT)) {
7189 /* Sanity check */
7190 error = EINVAL;
7191 goto out;
7192 }
7193 }
7194
7195 table_size = sizeof(memorystatus_internal_properties_t) * entry_count;
7196 if ((table = kheap_alloc(KHEAP_TEMP, table_size, Z_WAITOK | Z_ZERO)) == NULL) {
7197 error = ENOMEM;
7198 goto out;
7199 }
7200
7201
7202 /*
7203 * For each jetsam bucket entry, spin through the input property list.
7204 * When a matching pid is found, populate an adjacent table with the
7205 * appropriate proc pointer and new property values.
7206 * This traversal automatically preserves order from lowest
7207 * to highest priority.
7208 */
7209
7210 bucket_index = 0;
7211
7212 proc_list_lock();
7213
7214 /* Create the ordered table */
7215 p = memorystatus_get_first_proc_locked(&bucket_index, TRUE);
7216 while (p && (table_count < entry_count)) {
7217 for (i = 0; i < entry_count; i++) {
7218 if (p->p_pid == entries[i].pid) {
7219 /* Build the table data */
7220 table[table_count].proc = p;
7221 table[table_count].priority = entries[i].priority;
7222 table_count++;
7223 break;
7224 }
7225 }
7226 p = memorystatus_get_next_proc_locked(&bucket_index, p, TRUE);
7227 }
7228
7229 /* We now have ordered list of procs ready to move */
7230 for (i = 0; i < table_count; i++) {
7231 p = table[i].proc;
7232 assert(p != NULL);
7233
7234 /* Allow head inserts -- but relative order is now */
7235 if (table[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
7236 new_priority = JETSAM_PRIORITY_IDLE;
7237 head_insert = true;
7238 } else {
7239 new_priority = table[i].priority;
7240 head_insert = false;
7241 }
7242
7243 /* Not allowed */
7244 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
7245 continue;
7246 }
7247
7248 /*
7249 * Take appropriate steps if moving proc out of
7250 * either of the aging bands.
7251 */
7252 if ((p->p_memstat_effectivepriority == system_procs_aging_band) || (p->p_memstat_effectivepriority == applications_aging_band)) {
7253 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
7254 }
7255
7256 memorystatus_update_priority_locked(p, new_priority, head_insert, false);
7257 }
7258
7259 proc_list_unlock();
7260
7261 /*
7262 * if (table_count != entry_count)
7263 * then some pids were not found in a jetsam band.
7264 * harmless but interesting...
7265 */
7266 out:
7267 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_END, MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY, entry_count, table_count, 0, 0);
7268
7269 if (entries) {
7270 kheap_free(KHEAP_TEMP, entries, buffer_size);
7271 }
7272 if (table) {
7273 kheap_free(KHEAP_TEMP, table, table_size);
7274 }
7275
7276 return error;
7277 }
7278
7279 memorystatus_internal_probabilities_t *memorystatus_global_probabilities_table = NULL;
7280 size_t memorystatus_global_probabilities_size = 0;
7281
7282 static int
7283 memorystatus_cmd_grp_set_probabilities(user_addr_t buffer, size_t buffer_size)
7284 {
7285 int error = 0;
7286 memorystatus_properties_entry_v1_t *entries = NULL;
7287 size_t entry_count = 0, i = 0;
7288 memorystatus_internal_probabilities_t *tmp_table_new = NULL, *tmp_table_old = NULL;
7289 size_t tmp_table_new_size = 0, tmp_table_old_size = 0;
7290 #if DEVELOPMENT || DEBUG
7291 if (memorystatus_testing_pid != 0 && memorystatus_testing_pid != current_proc()->p_pid) {
7292 /* probabilites are currently owned by someone else. Don't change them. */
7293 error = EPERM;
7294 goto out;
7295 }
7296 #endif /* (DEVELOPMENT || DEBUG)*/
7297
7298 /* Verify inputs */
7299 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
7300 error = EINVAL;
7301 goto out;
7302 }
7303
7304 entry_count = (buffer_size / sizeof(memorystatus_properties_entry_v1_t));
7305
7306 if ((entries = kheap_alloc(KHEAP_TEMP, buffer_size, Z_WAITOK)) == NULL) {
7307 error = ENOMEM;
7308 goto out;
7309 }
7310
7311 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_START, MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY, entry_count, 0, 0, 0);
7312
7313 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
7314 goto out;
7315 }
7316
7317 if (entries[0].version == MEMORYSTATUS_MPE_VERSION_1) {
7318 if ((buffer_size % MEMORYSTATUS_MPE_VERSION_1_SIZE) != 0) {
7319 error = EINVAL;
7320 goto out;
7321 }
7322 } else {
7323 error = EINVAL;
7324 goto out;
7325 }
7326
7327 /* Verify sanity of input priorities */
7328 for (i = 0; i < entry_count; i++) {
7329 /*
7330 * 0 - low probability of use.
7331 * 1 - high probability of use.
7332 *
7333 * Keeping this field an int (& not a bool) to allow
7334 * us to experiment with different values/approaches
7335 * later on.
7336 */
7337 if (entries[i].use_probability > 1) {
7338 error = EINVAL;
7339 goto out;
7340 }
7341 }
7342
7343 tmp_table_new_size = sizeof(memorystatus_internal_probabilities_t) * entry_count;
7344
7345 if ((tmp_table_new = kalloc_flags(tmp_table_new_size, Z_WAITOK | Z_ZERO)) == NULL) {
7346 error = ENOMEM;
7347 goto out;
7348 }
7349
7350 proc_list_lock();
7351
7352 if (memorystatus_global_probabilities_table) {
7353 tmp_table_old = memorystatus_global_probabilities_table;
7354 tmp_table_old_size = memorystatus_global_probabilities_size;
7355 }
7356
7357 memorystatus_global_probabilities_table = tmp_table_new;
7358 memorystatus_global_probabilities_size = tmp_table_new_size;
7359 tmp_table_new = NULL;
7360
7361 for (i = 0; i < entry_count; i++) {
7362 /* Build the table data */
7363 strlcpy(memorystatus_global_probabilities_table[i].proc_name, entries[i].proc_name, MAXCOMLEN + 1);
7364 memorystatus_global_probabilities_table[i].use_probability = entries[i].use_probability;
7365 }
7366
7367 proc_list_unlock();
7368
7369 out:
7370 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_END, MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY, entry_count, tmp_table_new_size, 0, 0);
7371
7372 if (entries) {
7373 kheap_free(KHEAP_TEMP, entries, buffer_size);
7374 entries = NULL;
7375 }
7376
7377 if (tmp_table_old) {
7378 kfree(tmp_table_old, tmp_table_old_size);
7379 tmp_table_old = NULL;
7380 }
7381
7382 return error;
7383 }
7384
7385 static int
7386 memorystatus_cmd_grp_set_properties(int32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7387 {
7388 int error = 0;
7389
7390 if ((flags & MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY) == MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY) {
7391 error = memorystatus_cmd_grp_set_priorities(buffer, buffer_size);
7392 } else if ((flags & MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY) == MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY) {
7393 error = memorystatus_cmd_grp_set_probabilities(buffer, buffer_size);
7394 } else {
7395 error = EINVAL;
7396 }
7397
7398 return error;
7399 }
7400
7401 /*
7402 * This routine is used to update a process's jetsam priority position and stored user_data.
7403 * It is not used for the setting of memory limits, which is why the last 6 args to the
7404 * memorystatus_update() call are 0 or FALSE.
7405 *
7406 * Flags passed into this call are used to distinguish the motivation behind a jetsam priority
7407 * transition. By default, the kernel updates the process's original requested priority when
7408 * no flag is passed. But when the MEMORYSTATUS_SET_PRIORITY_ASSERTION flag is used, the kernel
7409 * updates the process's assertion driven priority.
7410 *
7411 * The assertion flag was introduced for use by the device's assertion mediator (eg: runningboardd).
7412 * When an assertion is controlling a process's jetsam priority, it may conflict with that process's
7413 * dirty/clean (active/inactive) jetsam state. The kernel attempts to resolve a priority transition
7414 * conflict by reviewing the process state and then choosing the maximum jetsam band at play,
7415 * eg: requested priority versus assertion priority.
7416 */
7417
7418 static int
7419 memorystatus_cmd_set_priority_properties(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7420 {
7421 int error = 0;
7422 boolean_t is_assertion = FALSE; /* priority is driven by an assertion */
7423 memorystatus_priority_properties_t mpp_entry;
7424
7425 /* Validate inputs */
7426 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_properties_t))) {
7427 return EINVAL;
7428 }
7429
7430 /* Validate flags */
7431 if (flags == 0) {
7432 /*
7433 * Default. This path updates requestedpriority.
7434 */
7435 } else {
7436 if (flags & ~(MEMORYSTATUS_SET_PRIORITY_ASSERTION)) {
7437 /*
7438 * Unsupported bit set in flag.
7439 */
7440 return EINVAL;
7441 } else if (flags & MEMORYSTATUS_SET_PRIORITY_ASSERTION) {
7442 is_assertion = TRUE;
7443 }
7444 }
7445
7446 error = copyin(buffer, &mpp_entry, buffer_size);
7447
7448 if (error == 0) {
7449 proc_t p;
7450
7451 p = proc_find(pid);
7452 if (!p) {
7453 return ESRCH;
7454 }
7455
7456 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
7457 proc_rele(p);
7458 return EPERM;
7459 }
7460
7461 if (is_assertion) {
7462 os_log(OS_LOG_DEFAULT, "memorystatus: set assertion priority(%d) target %s:%d\n",
7463 mpp_entry.priority, (*p->p_name ? p->p_name : "unknown"), p->p_pid);
7464 }
7465
7466 error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, is_assertion, FALSE, FALSE, 0, 0, FALSE, FALSE);
7467 proc_rele(p);
7468 }
7469
7470 return error;
7471 }
7472
7473 static int
7474 memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7475 {
7476 int error = 0;
7477 memorystatus_memlimit_properties_t mmp_entry;
7478
7479 /* Validate inputs */
7480 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
7481 return EINVAL;
7482 }
7483
7484 error = copyin(buffer, &mmp_entry, buffer_size);
7485
7486 if (error == 0) {
7487 error = memorystatus_set_memlimit_properties(pid, &mmp_entry);
7488 }
7489
7490 return error;
7491 }
7492
7493 static void
7494 memorystatus_get_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t* p_entry)
7495 {
7496 memset(p_entry, 0, sizeof(memorystatus_memlimit_properties_t));
7497
7498 if (p->p_memstat_memlimit_active > 0) {
7499 p_entry->memlimit_active = p->p_memstat_memlimit_active;
7500 } else {
7501 task_convert_phys_footprint_limit(-1, &p_entry->memlimit_active);
7502 }
7503
7504 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) {
7505 p_entry->memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7506 }
7507
7508 /*
7509 * Get the inactive limit and attributes
7510 */
7511 if (p->p_memstat_memlimit_inactive <= 0) {
7512 task_convert_phys_footprint_limit(-1, &p_entry->memlimit_inactive);
7513 } else {
7514 p_entry->memlimit_inactive = p->p_memstat_memlimit_inactive;
7515 }
7516 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) {
7517 p_entry->memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7518 }
7519 }
7520
7521 /*
7522 * When getting the memlimit settings, we can't simply call task_get_phys_footprint_limit().
7523 * That gets the proc's cached memlimit and there is no guarantee that the active/inactive
7524 * limits will be the same in the no-limit case. Instead we convert limits <= 0 using
7525 * task_convert_phys_footprint_limit(). It computes the same limit value that would be written
7526 * to the task's ledgers via task_set_phys_footprint_limit().
7527 */
7528 static int
7529 memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7530 {
7531 memorystatus_memlimit_properties2_t mmp_entry;
7532
7533 /* Validate inputs */
7534 if ((pid == 0) || (buffer == USER_ADDR_NULL) ||
7535 ((buffer_size != sizeof(memorystatus_memlimit_properties_t)) &&
7536 (buffer_size != sizeof(memorystatus_memlimit_properties2_t)))) {
7537 return EINVAL;
7538 }
7539
7540 memset(&mmp_entry, 0, sizeof(memorystatus_memlimit_properties2_t));
7541
7542 proc_t p = proc_find(pid);
7543 if (!p) {
7544 return ESRCH;
7545 }
7546
7547 /*
7548 * Get the active limit and attributes.
7549 * No locks taken since we hold a reference to the proc.
7550 */
7551
7552 memorystatus_get_memlimit_properties_internal(p, &mmp_entry.v1);
7553
7554 #if CONFIG_JETSAM
7555 #if DEVELOPMENT || DEBUG
7556 /*
7557 * Get the limit increased via SPI
7558 */
7559 mmp_entry.memlimit_increase = roundToNearestMB(p->p_memlimit_increase);
7560 mmp_entry.memlimit_increase_bytes = p->p_memlimit_increase;
7561 #endif /* DEVELOPMENT || DEBUG */
7562 #endif /* CONFIG_JETSAM */
7563
7564 proc_rele(p);
7565
7566 int error = copyout(&mmp_entry, buffer, buffer_size);
7567
7568 return error;
7569 }
7570
7571
7572 /*
7573 * SPI for kbd - pr24956468
7574 * This is a very simple snapshot that calculates how much a
7575 * process's phys_footprint exceeds a specific memory limit.
7576 * Only the inactive memory limit is supported for now.
7577 * The delta is returned as bytes in excess or zero.
7578 */
7579 static int
7580 memorystatus_cmd_get_memlimit_excess_np(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7581 {
7582 int error = 0;
7583 uint64_t footprint_in_bytes = 0;
7584 uint64_t delta_in_bytes = 0;
7585 int32_t memlimit_mb = 0;
7586 uint64_t memlimit_bytes = 0;
7587
7588 /* Validate inputs */
7589 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(uint64_t)) || (flags != 0)) {
7590 return EINVAL;
7591 }
7592
7593 proc_t p = proc_find(pid);
7594 if (!p) {
7595 return ESRCH;
7596 }
7597
7598 /*
7599 * Get the inactive limit.
7600 * No locks taken since we hold a reference to the proc.
7601 */
7602
7603 if (p->p_memstat_memlimit_inactive <= 0) {
7604 task_convert_phys_footprint_limit(-1, &memlimit_mb);
7605 } else {
7606 memlimit_mb = p->p_memstat_memlimit_inactive;
7607 }
7608
7609 footprint_in_bytes = get_task_phys_footprint(p->task);
7610
7611 proc_rele(p);
7612
7613 memlimit_bytes = memlimit_mb * 1024 * 1024; /* MB to bytes */
7614
7615 /*
7616 * Computed delta always returns >= 0 bytes
7617 */
7618 if (footprint_in_bytes > memlimit_bytes) {
7619 delta_in_bytes = footprint_in_bytes - memlimit_bytes;
7620 }
7621
7622 error = copyout(&delta_in_bytes, buffer, sizeof(delta_in_bytes));
7623
7624 return error;
7625 }
7626
7627
7628 static int
7629 memorystatus_cmd_get_pressure_status(int32_t *retval)
7630 {
7631 int error;
7632
7633 /* Need privilege for check */
7634 error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
7635 if (error) {
7636 return error;
7637 }
7638
7639 /* Inherently racy, so it's not worth taking a lock here */
7640 *retval = (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
7641
7642 return error;
7643 }
7644
7645 int
7646 memorystatus_get_pressure_status_kdp()
7647 {
7648 return (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
7649 }
7650
7651 /*
7652 * Every process, including a P_MEMSTAT_INTERNAL process (currently only pid 1), is allowed to set a HWM.
7653 *
7654 * This call is inflexible -- it does not distinguish between active/inactive, fatal/non-fatal
7655 * So, with 2-level HWM preserving previous behavior will map as follows.
7656 * - treat the limit passed in as both an active and inactive limit.
7657 * - treat the is_fatal_limit flag as though it applies to both active and inactive limits.
7658 *
7659 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK
7660 * - the is_fatal_limit is FALSE, meaning the active and inactive limits are non-fatal/soft
7661 * - so mapping is (active/non-fatal, inactive/non-fatal)
7662 *
7663 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT
7664 * - the is_fatal_limit is TRUE, meaning the process's active and inactive limits are fatal/hard
7665 * - so mapping is (active/fatal, inactive/fatal)
7666 */
7667
7668 #if CONFIG_JETSAM
7669 static int
7670 memorystatus_cmd_set_jetsam_memory_limit(pid_t pid, int32_t high_water_mark, __unused int32_t *retval, boolean_t is_fatal_limit)
7671 {
7672 int error = 0;
7673 memorystatus_memlimit_properties_t entry;
7674
7675 entry.memlimit_active = high_water_mark;
7676 entry.memlimit_active_attr = 0;
7677 entry.memlimit_inactive = high_water_mark;
7678 entry.memlimit_inactive_attr = 0;
7679
7680 if (is_fatal_limit == TRUE) {
7681 entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7682 entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7683 }
7684
7685 error = memorystatus_set_memlimit_properties(pid, &entry);
7686 return error;
7687 }
7688 #endif /* CONFIG_JETSAM */
7689
7690 static int
7691 memorystatus_set_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t *p_entry)
7692 {
7693 int error = 0;
7694
7695 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
7696
7697 /*
7698 * Store the active limit variants in the proc.
7699 */
7700 SET_ACTIVE_LIMITS_LOCKED(p, p_entry->memlimit_active, p_entry->memlimit_active_attr);
7701
7702 /*
7703 * Store the inactive limit variants in the proc.
7704 */
7705 SET_INACTIVE_LIMITS_LOCKED(p, p_entry->memlimit_inactive, p_entry->memlimit_inactive_attr);
7706
7707 /*
7708 * Enforce appropriate limit variant by updating the cached values
7709 * and writing the ledger.
7710 * Limit choice is based on process active/inactive state.
7711 */
7712
7713 if (memorystatus_highwater_enabled) {
7714 boolean_t is_fatal;
7715 boolean_t use_active;
7716
7717 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
7718 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
7719 use_active = TRUE;
7720 } else {
7721 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
7722 use_active = FALSE;
7723 }
7724
7725 /* Enforce the limit by writing to the ledgers */
7726 error = (task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, use_active, is_fatal) == 0) ? 0 : EINVAL;
7727
7728 MEMORYSTATUS_DEBUG(3, "memorystatus_set_memlimit_properties: new limit on pid %d (%dMB %s) current priority (%d) dirty_state?=0x%x %s\n",
7729 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
7730 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
7731 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
7732 DTRACE_MEMORYSTATUS2(memorystatus_set_memlimit, proc_t, p, int32_t, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1));
7733 }
7734
7735 return error;
7736 }
7737
7738 static int
7739 memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry)
7740 {
7741 memorystatus_memlimit_properties_t set_entry;
7742
7743 proc_t p = proc_find(pid);
7744 if (!p) {
7745 return ESRCH;
7746 }
7747
7748 /*
7749 * Check for valid attribute flags.
7750 */
7751 const uint32_t valid_attrs = MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7752 if ((entry->memlimit_active_attr & (~valid_attrs)) != 0) {
7753 proc_rele(p);
7754 return EINVAL;
7755 }
7756 if ((entry->memlimit_inactive_attr & (~valid_attrs)) != 0) {
7757 proc_rele(p);
7758 return EINVAL;
7759 }
7760
7761 /*
7762 * Setup the active memlimit properties
7763 */
7764 set_entry.memlimit_active = entry->memlimit_active;
7765 set_entry.memlimit_active_attr = entry->memlimit_active_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7766
7767 /*
7768 * Setup the inactive memlimit properties
7769 */
7770 set_entry.memlimit_inactive = entry->memlimit_inactive;
7771 set_entry.memlimit_inactive_attr = entry->memlimit_inactive_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7772
7773 /*
7774 * Setting a limit of <= 0 implies that the process has no
7775 * high-water-mark and has no per-task-limit. That means
7776 * the system_wide task limit is in place, which by the way,
7777 * is always fatal.
7778 */
7779
7780 if (set_entry.memlimit_active <= 0) {
7781 /*
7782 * Enforce the fatal system_wide task limit while process is active.
7783 */
7784 set_entry.memlimit_active = -1;
7785 set_entry.memlimit_active_attr = MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7786 }
7787 #if CONFIG_JETSAM
7788 #if DEVELOPMENT || DEBUG
7789 else {
7790 /* add the current increase to it, for roots */
7791 set_entry.memlimit_active += roundToNearestMB(p->p_memlimit_increase);
7792 }
7793 #endif /* DEVELOPMENT || DEBUG */
7794 #endif /* CONFIG_JETSAM */
7795
7796 if (set_entry.memlimit_inactive <= 0) {
7797 /*
7798 * Enforce the fatal system_wide task limit while process is inactive.
7799 */
7800 set_entry.memlimit_inactive = -1;
7801 set_entry.memlimit_inactive_attr = MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7802 }
7803 #if CONFIG_JETSAM
7804 #if DEVELOPMENT || DEBUG
7805 else {
7806 /* add the current increase to it, for roots */
7807 set_entry.memlimit_inactive += roundToNearestMB(p->p_memlimit_increase);
7808 }
7809 #endif /* DEVELOPMENT || DEBUG */
7810 #endif /* CONFIG_JETSAM */
7811
7812 proc_list_lock();
7813
7814 int error = memorystatus_set_memlimit_properties_internal(p, &set_entry);
7815
7816 proc_list_unlock();
7817 proc_rele(p);
7818
7819 return error;
7820 }
7821
7822 /*
7823 * Returns the jetsam priority (effective or requested) of the process
7824 * associated with this task.
7825 */
7826 int
7827 proc_get_memstat_priority(proc_t p, boolean_t effective_priority)
7828 {
7829 if (p) {
7830 if (effective_priority) {
7831 return p->p_memstat_effectivepriority;
7832 } else {
7833 return p->p_memstat_requestedpriority;
7834 }
7835 }
7836 return 0;
7837 }
7838
7839 static int
7840 memorystatus_get_process_is_managed(pid_t pid, int *is_managed)
7841 {
7842 proc_t p = NULL;
7843
7844 /* Validate inputs */
7845 if (pid == 0) {
7846 return EINVAL;
7847 }
7848
7849 p = proc_find(pid);
7850 if (!p) {
7851 return ESRCH;
7852 }
7853
7854 proc_list_lock();
7855 *is_managed = ((p->p_memstat_state & P_MEMSTAT_MANAGED) ? 1 : 0);
7856 proc_rele_locked(p);
7857 proc_list_unlock();
7858
7859 return 0;
7860 }
7861
7862 static int
7863 memorystatus_set_process_is_managed(pid_t pid, boolean_t set_managed)
7864 {
7865 proc_t p = NULL;
7866
7867 /* Validate inputs */
7868 if (pid == 0) {
7869 return EINVAL;
7870 }
7871
7872 p = proc_find(pid);
7873 if (!p) {
7874 return ESRCH;
7875 }
7876
7877 proc_list_lock();
7878 if (set_managed == TRUE) {
7879 p->p_memstat_state |= P_MEMSTAT_MANAGED;
7880 /*
7881 * The P_MEMSTAT_MANAGED bit is set by assertiond for Apps.
7882 * Also opt them in to being frozen (they might have started
7883 * off with the P_MEMSTAT_FREEZE_DISABLED bit set.)
7884 */
7885 p->p_memstat_state &= ~P_MEMSTAT_FREEZE_DISABLED;
7886 } else {
7887 p->p_memstat_state &= ~P_MEMSTAT_MANAGED;
7888 }
7889 proc_rele_locked(p);
7890 proc_list_unlock();
7891
7892 return 0;
7893 }
7894
7895 int
7896 memorystatus_control(struct proc *p __unused, struct memorystatus_control_args *args, int *ret)
7897 {
7898 int error = EINVAL;
7899 boolean_t skip_auth_check = FALSE;
7900 os_reason_t jetsam_reason = OS_REASON_NULL;
7901
7902 #if !CONFIG_JETSAM
7903 #pragma unused(ret)
7904 #pragma unused(jetsam_reason)
7905 #endif
7906
7907 /* We don't need entitlements if we're setting / querying the freeze preference or frozen status for a process. */
7908 if (args->command == MEMORYSTATUS_CMD_SET_PROCESS_IS_FREEZABLE ||
7909 args->command == MEMORYSTATUS_CMD_GET_PROCESS_IS_FREEZABLE ||
7910 args->command == MEMORYSTATUS_CMD_GET_PROCESS_IS_FROZEN) {
7911 skip_auth_check = TRUE;
7912 }
7913
7914 /* Need to be root or have entitlement. */
7915 if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT) && !skip_auth_check) {
7916 error = EPERM;
7917 goto out;
7918 }
7919
7920 /*
7921 * Sanity check.
7922 * Do not enforce it for snapshots.
7923 */
7924 if (args->command != MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT) {
7925 if (args->buffersize > MEMORYSTATUS_BUFFERSIZE_MAX) {
7926 error = EINVAL;
7927 goto out;
7928 }
7929 }
7930
7931 switch (args->command) {
7932 case MEMORYSTATUS_CMD_GET_PRIORITY_LIST:
7933 error = memorystatus_cmd_get_priority_list(args->pid, args->buffer, args->buffersize, ret);
7934 break;
7935 case MEMORYSTATUS_CMD_SET_PRIORITY_PROPERTIES:
7936 error = memorystatus_cmd_set_priority_properties(args->pid, args->flags, args->buffer, args->buffersize, ret);
7937 break;
7938 case MEMORYSTATUS_CMD_SET_MEMLIMIT_PROPERTIES:
7939 error = memorystatus_cmd_set_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
7940 break;
7941 case MEMORYSTATUS_CMD_GET_MEMLIMIT_PROPERTIES:
7942 error = memorystatus_cmd_get_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
7943 break;
7944 case MEMORYSTATUS_CMD_GET_MEMLIMIT_EXCESS:
7945 error = memorystatus_cmd_get_memlimit_excess_np(args->pid, args->flags, args->buffer, args->buffersize, ret);
7946 break;
7947 case MEMORYSTATUS_CMD_GRP_SET_PROPERTIES:
7948 error = memorystatus_cmd_grp_set_properties((int32_t)args->flags, args->buffer, args->buffersize, ret);
7949 break;
7950 case MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT:
7951 error = memorystatus_cmd_get_jetsam_snapshot((int32_t)args->flags, args->buffer, args->buffersize, ret);
7952 break;
7953 #if DEVELOPMENT || DEBUG
7954 case MEMORYSTATUS_CMD_SET_TESTING_PID:
7955 error = memorystatus_cmd_set_testing_pid((int32_t) args->flags);
7956 break;
7957 #endif
7958 case MEMORYSTATUS_CMD_GET_PRESSURE_STATUS:
7959 error = memorystatus_cmd_get_pressure_status(ret);
7960 break;
7961 #if CONFIG_JETSAM
7962 case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
7963 /*
7964 * This call does not distinguish between active and inactive limits.
7965 * Default behavior in 2-level HWM world is to set both.
7966 * Non-fatal limit is also assumed for both.
7967 */
7968 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, FALSE);
7969 break;
7970 case MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT:
7971 /*
7972 * This call does not distinguish between active and inactive limits.
7973 * Default behavior in 2-level HWM world is to set both.
7974 * Fatal limit is also assumed for both.
7975 */
7976 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
7977 break;
7978 #endif /* CONFIG_JETSAM */
7979 /* Test commands */
7980 #if DEVELOPMENT || DEBUG
7981 case MEMORYSTATUS_CMD_TEST_JETSAM:
7982 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_GENERIC);
7983 if (jetsam_reason == OS_REASON_NULL) {
7984 printf("memorystatus_control: failed to allocate jetsam reason\n");
7985 }
7986
7987 error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled, jetsam_reason) ? 0 : EINVAL;
7988 break;
7989 case MEMORYSTATUS_CMD_TEST_JETSAM_SORT:
7990 error = memorystatus_cmd_test_jetsam_sort(args->pid, (int32_t)args->flags, args->buffer, args->buffersize);
7991 break;
7992 #if CONFIG_JETSAM
7993 case MEMORYSTATUS_CMD_SET_JETSAM_PANIC_BITS:
7994 error = memorystatus_cmd_set_panic_bits(args->buffer, args->buffersize);
7995 break;
7996 #endif /* CONFIG_JETSAM */
7997 #else /* DEVELOPMENT || DEBUG */
7998 #pragma unused(jetsam_reason)
7999 #endif /* DEVELOPMENT || DEBUG */
8000 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_ENABLE:
8001 if (memorystatus_aggressive_jetsam_lenient_allowed == FALSE) {
8002 #if DEVELOPMENT || DEBUG
8003 printf("Enabling Lenient Mode\n");
8004 #endif /* DEVELOPMENT || DEBUG */
8005
8006 memorystatus_aggressive_jetsam_lenient_allowed = TRUE;
8007 memorystatus_aggressive_jetsam_lenient = TRUE;
8008 error = 0;
8009 }
8010 break;
8011 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_DISABLE:
8012 #if DEVELOPMENT || DEBUG
8013 printf("Disabling Lenient mode\n");
8014 #endif /* DEVELOPMENT || DEBUG */
8015 memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
8016 memorystatus_aggressive_jetsam_lenient = FALSE;
8017 error = 0;
8018 break;
8019 case MEMORYSTATUS_CMD_GET_AGGRESSIVE_JETSAM_LENIENT_MODE:
8020 *ret = (memorystatus_aggressive_jetsam_lenient ? 1 : 0);
8021 error = 0;
8022 break;
8023 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE:
8024 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE:
8025 error = memorystatus_low_mem_privileged_listener(args->command);
8026 break;
8027
8028 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE:
8029 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE:
8030 error = memorystatus_update_inactive_jetsam_priority_band(args->pid, args->command, JETSAM_PRIORITY_ELEVATED_INACTIVE, args->flags ? TRUE : FALSE);
8031 break;
8032 case MEMORYSTATUS_CMD_SET_PROCESS_IS_MANAGED:
8033 error = memorystatus_set_process_is_managed(args->pid, args->flags);
8034 break;
8035
8036 case MEMORYSTATUS_CMD_GET_PROCESS_IS_MANAGED:
8037 error = memorystatus_get_process_is_managed(args->pid, ret);
8038 break;
8039
8040 #if CONFIG_FREEZE
8041 case MEMORYSTATUS_CMD_SET_PROCESS_IS_FREEZABLE:
8042 error = memorystatus_set_process_is_freezable(args->pid, args->flags ? TRUE : FALSE);
8043 break;
8044
8045 case MEMORYSTATUS_CMD_GET_PROCESS_IS_FREEZABLE:
8046 error = memorystatus_get_process_is_freezable(args->pid, ret);
8047 break;
8048 case MEMORYSTATUS_CMD_GET_PROCESS_IS_FROZEN:
8049 error = memorystatus_get_process_is_frozen(args->pid, ret);
8050 break;
8051
8052 case MEMORYSTATUS_CMD_FREEZER_CONTROL:
8053 error = memorystatus_freezer_control(args->flags, args->buffer, args->buffersize, ret);
8054 break;
8055 #endif /* CONFIG_FREEZE */
8056
8057 #if CONFIG_JETSAM
8058 #if DEVELOPMENT || DEBUG
8059 case MEMORYSTATUS_CMD_INCREASE_JETSAM_TASK_LIMIT:
8060 error = memorystatus_cmd_increase_jetsam_task_limit(args->pid, args->flags);
8061 break;
8062 #endif /* DEVELOPMENT || DEBUG */
8063 #endif /* CONFIG_JETSAM */
8064
8065 default:
8066 break;
8067 }
8068
8069 out:
8070 return error;
8071 }
8072
8073 /* Coalition support */
8074
8075 /* sorting info for a particular priority bucket */
8076 typedef struct memstat_sort_info {
8077 coalition_t msi_coal;
8078 uint64_t msi_page_count;
8079 pid_t msi_pid;
8080 int msi_ntasks;
8081 } memstat_sort_info_t;
8082
8083 /*
8084 * qsort from smallest page count to largest page count
8085 *
8086 * return < 0 for a < b
8087 * 0 for a == b
8088 * > 0 for a > b
8089 */
8090 static int
8091 memstat_asc_cmp(const void *a, const void *b)
8092 {
8093 const memstat_sort_info_t *msA = (const memstat_sort_info_t *)a;
8094 const memstat_sort_info_t *msB = (const memstat_sort_info_t *)b;
8095
8096 return (int)((uint64_t)msA->msi_page_count - (uint64_t)msB->msi_page_count);
8097 }
8098
8099 /*
8100 * Return the number of pids rearranged during this sort.
8101 */
8102 static int
8103 memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order)
8104 {
8105 #define MAX_SORT_PIDS 80
8106 #define MAX_COAL_LEADERS 10
8107
8108 unsigned int b = bucket_index;
8109 int nleaders = 0;
8110 int ntasks = 0;
8111 proc_t p = NULL;
8112 coalition_t coal = COALITION_NULL;
8113 int pids_moved = 0;
8114 int total_pids_moved = 0;
8115 int i;
8116
8117 /*
8118 * The system is typically under memory pressure when in this
8119 * path, hence, we want to avoid dynamic memory allocation.
8120 */
8121 memstat_sort_info_t leaders[MAX_COAL_LEADERS];
8122 pid_t pid_list[MAX_SORT_PIDS];
8123
8124 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
8125 return 0;
8126 }
8127
8128 /*
8129 * Clear the array that holds coalition leader information
8130 */
8131 for (i = 0; i < MAX_COAL_LEADERS; i++) {
8132 leaders[i].msi_coal = COALITION_NULL;
8133 leaders[i].msi_page_count = 0; /* will hold total coalition page count */
8134 leaders[i].msi_pid = 0; /* will hold coalition leader pid */
8135 leaders[i].msi_ntasks = 0; /* will hold the number of tasks in a coalition */
8136 }
8137
8138 p = memorystatus_get_first_proc_locked(&b, FALSE);
8139 while (p) {
8140 coal = task_get_coalition(p->task, COALITION_TYPE_JETSAM);
8141 if (coalition_is_leader(p->task, coal)) {
8142 if (nleaders < MAX_COAL_LEADERS) {
8143 int coal_ntasks = 0;
8144 uint64_t coal_page_count = coalition_get_page_count(coal, &coal_ntasks);
8145 leaders[nleaders].msi_coal = coal;
8146 leaders[nleaders].msi_page_count = coal_page_count;
8147 leaders[nleaders].msi_pid = p->p_pid; /* the coalition leader */
8148 leaders[nleaders].msi_ntasks = coal_ntasks;
8149 nleaders++;
8150 } else {
8151 /*
8152 * We've hit MAX_COAL_LEADERS meaning we can handle no more coalitions.
8153 * Abandoned coalitions will linger at the tail of the priority band
8154 * when this sort session ends.
8155 * TODO: should this be an assert?
8156 */
8157 printf("%s: WARNING: more than %d leaders in priority band [%d]\n",
8158 __FUNCTION__, MAX_COAL_LEADERS, bucket_index);
8159 break;
8160 }
8161 }
8162 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
8163 }
8164
8165 if (nleaders == 0) {
8166 /* Nothing to sort */
8167 return 0;
8168 }
8169
8170 /*
8171 * Sort the coalition leader array, from smallest coalition page count
8172 * to largest coalition page count. When inserted in the priority bucket,
8173 * smallest coalition is handled first, resulting in the last to be jetsammed.
8174 */
8175 if (nleaders > 1) {
8176 qsort(leaders, nleaders, sizeof(memstat_sort_info_t), memstat_asc_cmp);
8177 }
8178
8179 #if 0
8180 for (i = 0; i < nleaders; i++) {
8181 printf("%s: coal_leader[%d of %d] pid[%d] pages[%llu] ntasks[%d]\n",
8182 __FUNCTION__, i, nleaders, leaders[i].msi_pid, leaders[i].msi_page_count,
8183 leaders[i].msi_ntasks);
8184 }
8185 #endif
8186
8187 /*
8188 * During coalition sorting, processes in a priority band are rearranged
8189 * by being re-inserted at the head of the queue. So, when handling a
8190 * list, the first process that gets moved to the head of the queue,
8191 * ultimately gets pushed toward the queue tail, and hence, jetsams last.
8192 *
8193 * So, for example, the coalition leader is expected to jetsam last,
8194 * after its coalition members. Therefore, the coalition leader is
8195 * inserted at the head of the queue first.
8196 *
8197 * After processing a coalition, the jetsam order is as follows:
8198 * undefs(jetsam first), extensions, xpc services, leader(jetsam last)
8199 */
8200
8201 /*
8202 * Coalition members are rearranged in the priority bucket here,
8203 * based on their coalition role.
8204 */
8205 total_pids_moved = 0;
8206 for (i = 0; i < nleaders; i++) {
8207 /* a bit of bookkeeping */
8208 pids_moved = 0;
8209
8210 /* Coalition leaders are jetsammed last, so move into place first */
8211 pid_list[0] = leaders[i].msi_pid;
8212 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list, 1);
8213
8214 /* xpc services should jetsam after extensions */
8215 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_XPC,
8216 coal_sort_order, pid_list, MAX_SORT_PIDS);
8217
8218 if (ntasks > 0) {
8219 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
8220 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
8221 }
8222
8223 /* extensions should jetsam after unmarked processes */
8224 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_EXT,
8225 coal_sort_order, pid_list, MAX_SORT_PIDS);
8226
8227 if (ntasks > 0) {
8228 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
8229 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
8230 }
8231
8232 /* undefined coalition members should be the first to jetsam */
8233 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_UNDEF,
8234 coal_sort_order, pid_list, MAX_SORT_PIDS);
8235
8236 if (ntasks > 0) {
8237 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
8238 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
8239 }
8240
8241 #if 0
8242 if (pids_moved == leaders[i].msi_ntasks) {
8243 /*
8244 * All the pids in the coalition were found in this band.
8245 */
8246 printf("%s: pids_moved[%d] equal total coalition ntasks[%d] \n", __FUNCTION__,
8247 pids_moved, leaders[i].msi_ntasks);
8248 } else if (pids_moved > leaders[i].msi_ntasks) {
8249 /*
8250 * Apparently new coalition members showed up during the sort?
8251 */
8252 printf("%s: pids_moved[%d] were greater than expected coalition ntasks[%d] \n", __FUNCTION__,
8253 pids_moved, leaders[i].msi_ntasks);
8254 } else {
8255 /*
8256 * Apparently not all the pids in the coalition were found in this band?
8257 */
8258 printf("%s: pids_moved[%d] were less than expected coalition ntasks[%d] \n", __FUNCTION__,
8259 pids_moved, leaders[i].msi_ntasks);
8260 }
8261 #endif
8262
8263 total_pids_moved += pids_moved;
8264 } /* end for */
8265
8266 return total_pids_moved;
8267 }
8268
8269
8270 /*
8271 * Traverse a list of pids, searching for each within the priority band provided.
8272 * If pid is found, move it to the front of the priority band.
8273 * Never searches outside the priority band provided.
8274 *
8275 * Input:
8276 * bucket_index - jetsam priority band.
8277 * pid_list - pointer to a list of pids.
8278 * list_sz - number of pids in the list.
8279 *
8280 * Pid list ordering is important in that,
8281 * pid_list[n] is expected to jetsam ahead of pid_list[n+1].
8282 * The sort_order is set by the coalition default.
8283 *
8284 * Return:
8285 * the number of pids found and hence moved within the priority band.
8286 */
8287 static int
8288 memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz)
8289 {
8290 memstat_bucket_t *current_bucket;
8291 int i;
8292 int found_pids = 0;
8293
8294 if ((pid_list == NULL) || (list_sz <= 0)) {
8295 return 0;
8296 }
8297
8298 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
8299 return 0;
8300 }
8301
8302 current_bucket = &memstat_bucket[bucket_index];
8303 for (i = 0; i < list_sz; i++) {
8304 unsigned int b = bucket_index;
8305 proc_t p = NULL;
8306 proc_t aProc = NULL;
8307 pid_t aPid;
8308 int list_index;
8309
8310 list_index = ((list_sz - 1) - i);
8311 aPid = pid_list[list_index];
8312
8313 /* never search beyond bucket_index provided */
8314 p = memorystatus_get_first_proc_locked(&b, FALSE);
8315 while (p) {
8316 if (p->p_pid == aPid) {
8317 aProc = p;
8318 break;
8319 }
8320 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
8321 }
8322
8323 if (aProc == NULL) {
8324 /* pid not found in this band, just skip it */
8325 continue;
8326 } else {
8327 TAILQ_REMOVE(&current_bucket->list, aProc, p_memstat_list);
8328 TAILQ_INSERT_HEAD(&current_bucket->list, aProc, p_memstat_list);
8329 found_pids++;
8330 }
8331 }
8332 return found_pids;
8333 }
8334
8335 int
8336 memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index)
8337 {
8338 int32_t i = JETSAM_PRIORITY_IDLE;
8339 int count = 0;
8340
8341 if (max_bucket_index >= MEMSTAT_BUCKET_COUNT) {
8342 return -1;
8343 }
8344
8345 while (i <= max_bucket_index) {
8346 count += memstat_bucket[i++].count;
8347 }
8348
8349 return count;
8350 }
8351
8352 int
8353 memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap)
8354 {
8355 #if !CONFIG_JETSAM
8356 if (!p || (!isApp(p)) || (p->p_memstat_state & (P_MEMSTAT_INTERNAL | P_MEMSTAT_MANAGED))) {
8357 /*
8358 * Ineligible processes OR system processes e.g. launchd.
8359 *
8360 * We also skip processes that have the P_MEMSTAT_MANAGED bit set, i.e.
8361 * they're managed by assertiond. These are iOS apps that have been ported
8362 * to macOS. assertiond might be in the process of modifying the app's
8363 * priority / memory limit - so it might have the proc_list lock, and then try
8364 * to take the task lock. Meanwhile we've entered this function with the task lock
8365 * held, and we need the proc_list lock below. So we'll deadlock with assertiond.
8366 *
8367 * It should be fine to read the P_MEMSTAT_MANAGED bit without the proc_list
8368 * lock here, since assertiond only sets this bit on process launch.
8369 */
8370 return -1;
8371 }
8372
8373 /*
8374 * For macOS only:
8375 * We would like to use memorystatus_update() here to move the processes
8376 * within the bands. Unfortunately memorystatus_update() calls
8377 * memorystatus_update_priority_locked() which uses any band transitions
8378 * as an indication to modify ledgers. For that it needs the task lock
8379 * and since we came into this function with the task lock held, we'll deadlock.
8380 *
8381 * Unfortunately we can't completely disable ledger updates because we still
8382 * need the ledger updates for a subset of processes i.e. daemons.
8383 * When all processes on all platforms support memory limits, we can simply call
8384 * memorystatus_update().
8385 *
8386 * It also has some logic to deal with 'aging' which, currently, is only applicable
8387 * on CONFIG_JETSAM configs. So, till every platform has CONFIG_JETSAM we'll need
8388 * to do this explicit band transition.
8389 */
8390
8391 memstat_bucket_t *current_bucket, *new_bucket;
8392 int32_t priority = 0;
8393
8394 proc_list_lock();
8395
8396 if (((p->p_listflag & P_LIST_EXITED) != 0) ||
8397 (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED))) {
8398 /*
8399 * If the process is on its way out OR
8400 * jetsam has alread tried and failed to kill this process,
8401 * let's skip the whole jetsam band transition.
8402 */
8403 proc_list_unlock();
8404 return 0;
8405 }
8406
8407 if (is_appnap) {
8408 current_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
8409 new_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
8410 priority = JETSAM_PRIORITY_IDLE;
8411 } else {
8412 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
8413 /*
8414 * It is possible that someone pulled this process
8415 * out of the IDLE band without updating its app-nap
8416 * parameters.
8417 */
8418 proc_list_unlock();
8419 return 0;
8420 }
8421
8422 current_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
8423 new_bucket = &memstat_bucket[p->p_memstat_requestedpriority];
8424 priority = p->p_memstat_requestedpriority;
8425 }
8426
8427 TAILQ_REMOVE(&current_bucket->list, p, p_memstat_list);
8428 current_bucket->count--;
8429 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
8430 current_bucket->relaunch_high_count--;
8431 }
8432 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
8433 new_bucket->count++;
8434 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
8435 new_bucket->relaunch_high_count++;
8436 }
8437 /*
8438 * Record idle start or idle delta.
8439 */
8440 if (p->p_memstat_effectivepriority == priority) {
8441 /*
8442 * This process is not transitioning between
8443 * jetsam priority buckets. Do nothing.
8444 */
8445 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
8446 uint64_t now;
8447 /*
8448 * Transitioning out of the idle priority bucket.
8449 * Record idle delta.
8450 */
8451 assert(p->p_memstat_idle_start != 0);
8452 now = mach_absolute_time();
8453 if (now > p->p_memstat_idle_start) {
8454 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
8455 }
8456 } else if (priority == JETSAM_PRIORITY_IDLE) {
8457 /*
8458 * Transitioning into the idle priority bucket.
8459 * Record idle start.
8460 */
8461 p->p_memstat_idle_start = mach_absolute_time();
8462 }
8463
8464 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CHANGE_PRIORITY), p->p_pid, priority, p->p_memstat_effectivepriority, 0, 0);
8465
8466 p->p_memstat_effectivepriority = priority;
8467
8468 proc_list_unlock();
8469
8470 return 0;
8471
8472 #else /* !CONFIG_JETSAM */
8473 #pragma unused(p)
8474 #pragma unused(is_appnap)
8475 return -1;
8476 #endif /* !CONFIG_JETSAM */
8477 }
8478
8479 uint64_t
8480 memorystatus_available_memory_internal(struct proc *p)
8481 {
8482 #ifdef XNU_TARGET_OS_OSX
8483 if (p->p_memstat_memlimit <= 0) {
8484 return 0;
8485 }
8486 #endif /* XNU_TARGET_OS_OSX */
8487 const uint64_t footprint_in_bytes = get_task_phys_footprint(p->task);
8488 int32_t memlimit_mb;
8489 int64_t memlimit_bytes;
8490 int64_t rc;
8491
8492 if (isApp(p) == FALSE) {
8493 return 0;
8494 }
8495
8496 if (p->p_memstat_memlimit > 0) {
8497 memlimit_mb = p->p_memstat_memlimit;
8498 } else if (task_convert_phys_footprint_limit(-1, &memlimit_mb) != KERN_SUCCESS) {
8499 return 0;
8500 }
8501
8502 if (memlimit_mb <= 0) {
8503 memlimit_bytes = INT_MAX & ~((1 << 20) - 1);
8504 } else {
8505 memlimit_bytes = ((int64_t) memlimit_mb) << 20;
8506 }
8507
8508 rc = memlimit_bytes - footprint_in_bytes;
8509
8510 return (rc >= 0) ? rc : 0;
8511 }
8512
8513 int
8514 memorystatus_available_memory(struct proc *p, __unused struct memorystatus_available_memory_args *args, uint64_t *ret)
8515 {
8516 *ret = memorystatus_available_memory_internal(p);
8517
8518 return 0;
8519 }
8520
8521 #if CONFIG_JETSAM
8522 #if DEVELOPMENT || DEBUG
8523 static int
8524 memorystatus_cmd_increase_jetsam_task_limit(pid_t pid, uint32_t byte_increase)
8525 {
8526 memorystatus_memlimit_properties_t mmp_entry;
8527
8528 /* Validate inputs */
8529 if ((pid == 0) || (byte_increase == 0)) {
8530 return EINVAL;
8531 }
8532
8533 proc_t p = proc_find(pid);
8534
8535 if (!p) {
8536 return ESRCH;
8537 }
8538
8539 const uint32_t current_memlimit_increase = roundToNearestMB(p->p_memlimit_increase);
8540 /* round to page */
8541 const int32_t page_aligned_increase = (int32_t) MIN(round_page(p->p_memlimit_increase + byte_increase), INT32_MAX);
8542
8543 proc_list_lock();
8544
8545 memorystatus_get_memlimit_properties_internal(p, &mmp_entry);
8546
8547 if (mmp_entry.memlimit_active > 0) {
8548 mmp_entry.memlimit_active -= current_memlimit_increase;
8549 mmp_entry.memlimit_active += roundToNearestMB(page_aligned_increase);
8550 }
8551
8552 if (mmp_entry.memlimit_inactive > 0) {
8553 mmp_entry.memlimit_inactive -= current_memlimit_increase;
8554 mmp_entry.memlimit_inactive += roundToNearestMB(page_aligned_increase);
8555 }
8556
8557 /*
8558 * Store the updated delta limit in the proc.
8559 */
8560 p->p_memlimit_increase = page_aligned_increase;
8561
8562 int error = memorystatus_set_memlimit_properties_internal(p, &mmp_entry);
8563
8564 proc_list_unlock();
8565 proc_rele(p);
8566
8567 return error;
8568 }
8569 #endif /* DEVELOPMENT */
8570 #endif /* CONFIG_JETSAM */
mirror of XNU