]> 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-6153.81.5.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 <IOKit/IOBSD.h>
42
43 #include <corpses/task_corpse.h>
44 #include <libkern/libkern.h>
45 #include <mach/coalition.h>
46 #include <mach/mach_time.h>
47 #include <mach/task.h>
48 #include <mach/host_priv.h>
49 #include <mach/mach_host.h>
50 #include <os/log.h>
51 #include <pexpert/pexpert.h>
52 #include <sys/coalition.h>
53 #include <sys/kern_event.h>
54 #include <sys/proc.h>
55 #include <sys/proc_info.h>
56 #include <sys/reason.h>
57 #include <sys/signal.h>
58 #include <sys/signalvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysproto.h>
61 #include <sys/wait.h>
62 #include <sys/tree.h>
63 #include <sys/priv.h>
64 #include <vm/vm_pageout.h>
65 #include <vm/vm_protos.h>
66 #include <mach/machine/sdt.h>
67 #include <libkern/section_keywords.h>
68 #include <stdatomic.h>
69
70 #if CONFIG_FREEZE
71 #include <vm/vm_map.h>
72 #endif /* CONFIG_FREEZE */
73
74 #include <sys/kern_memorystatus.h>
75 #include <sys/kern_memorystatus_freeze.h>
76 #include <sys/kern_memorystatus_notify.h>
77
78 /* For logging clarity */
79 static const char *memorystatus_kill_cause_name[] = {
80 "", /* kMemorystatusInvalid */
81 "jettisoned", /* kMemorystatusKilled */
82 "highwater", /* kMemorystatusKilledHiwat */
83 "vnode-limit", /* kMemorystatusKilledVnodes */
84 "vm-pageshortage", /* kMemorystatusKilledVMPageShortage */
85 "proc-thrashing", /* kMemorystatusKilledProcThrashing */
86 "fc-thrashing", /* kMemorystatusKilledFCThrashing */
87 "per-process-limit", /* kMemorystatusKilledPerProcessLimit */
88 "disk-space-shortage", /* kMemorystatusKilledDiskSpaceShortage */
89 "idle-exit", /* kMemorystatusKilledIdleExit */
90 "zone-map-exhaustion", /* kMemorystatusKilledZoneMapExhaustion */
91 "vm-compressor-thrashing", /* kMemorystatusKilledVMCompressorThrashing */
92 "vm-compressor-space-shortage", /* kMemorystatusKilledVMCompressorSpaceShortage */
93 };
94
95 static const char *
96 memorystatus_priority_band_name(int32_t priority)
97 {
98 switch (priority) {
99 case JETSAM_PRIORITY_FOREGROUND:
100 return "FOREGROUND";
101 case JETSAM_PRIORITY_AUDIO_AND_ACCESSORY:
102 return "AUDIO_AND_ACCESSORY";
103 case JETSAM_PRIORITY_CONDUCTOR:
104 return "CONDUCTOR";
105 case JETSAM_PRIORITY_DRIVER_APPLE:
106 return "DRIVER_APPLE";
107 case JETSAM_PRIORITY_HOME:
108 return "HOME";
109 case JETSAM_PRIORITY_EXECUTIVE:
110 return "EXECUTIVE";
111 case JETSAM_PRIORITY_IMPORTANT:
112 return "IMPORTANT";
113 case JETSAM_PRIORITY_CRITICAL:
114 return "CRITICAL";
115 }
116
117 return "?";
118 }
119
120 /* Does cause indicate vm or fc thrashing? */
121 static boolean_t
122 is_reason_thrashing(unsigned cause)
123 {
124 switch (cause) {
125 case kMemorystatusKilledFCThrashing:
126 case kMemorystatusKilledVMCompressorThrashing:
127 case kMemorystatusKilledVMCompressorSpaceShortage:
128 return TRUE;
129 default:
130 return FALSE;
131 }
132 }
133
134 /* Is the zone map almost full? */
135 static boolean_t
136 is_reason_zone_map_exhaustion(unsigned cause)
137 {
138 if (cause == kMemorystatusKilledZoneMapExhaustion) {
139 return TRUE;
140 }
141 return FALSE;
142 }
143
144 /*
145 * Returns the current zone map size and capacity to include in the jetsam snapshot.
146 * Defined in zalloc.c
147 */
148 extern void get_zone_map_size(uint64_t *current_size, uint64_t *capacity);
149
150 /*
151 * Returns the name of the largest zone and its size to include in the jetsam snapshot.
152 * Defined in zalloc.c
153 */
154 extern void get_largest_zone_info(char *zone_name, size_t zone_name_len, uint64_t *zone_size);
155
156 /*
157 * Active / Inactive limit support
158 * proc list must be locked
159 *
160 * The SET_*** macros are used to initialize a limit
161 * for the first time.
162 *
163 * The CACHE_*** macros are use to cache the limit that will
164 * soon be in effect down in the ledgers.
165 */
166
167 #define SET_ACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
168 MACRO_BEGIN \
169 (p)->p_memstat_memlimit_active = (limit); \
170 if (is_fatal) { \
171 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
172 } else { \
173 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
174 } \
175 MACRO_END
176
177 #define SET_INACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
178 MACRO_BEGIN \
179 (p)->p_memstat_memlimit_inactive = (limit); \
180 if (is_fatal) { \
181 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
182 } else { \
183 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
184 } \
185 MACRO_END
186
187 #define CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal) \
188 MACRO_BEGIN \
189 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_active; \
190 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) { \
191 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
192 is_fatal = TRUE; \
193 } else { \
194 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
195 is_fatal = FALSE; \
196 } \
197 MACRO_END
198
199 #define CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal) \
200 MACRO_BEGIN \
201 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_inactive; \
202 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) { \
203 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
204 is_fatal = TRUE; \
205 } else { \
206 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
207 is_fatal = FALSE; \
208 } \
209 MACRO_END
210
211
212 /* General tunables */
213
214 unsigned long delta_percentage = 5;
215 unsigned long critical_threshold_percentage = 5;
216 // On embedded devices with more than 3GB of memory we lower the critical percentage.
217 uint64_t config_jetsam_large_memory_cutoff = 3UL * (1UL << 30);
218 unsigned long critical_threshold_percentage_larger_devices = 4;
219 unsigned long delta_percentage_larger_devices = 4;
220 unsigned long idle_offset_percentage = 5;
221 unsigned long pressure_threshold_percentage = 15;
222 unsigned long policy_more_free_offset_percentage = 5;
223 unsigned long sysproc_aging_aggr_threshold_percentage = 7;
224
225 /*
226 * default jetsam snapshot support
227 */
228 memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot;
229 memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot_copy;
230 unsigned int memorystatus_jetsam_snapshot_count = 0;
231 unsigned int memorystatus_jetsam_snapshot_copy_count = 0;
232 unsigned int memorystatus_jetsam_snapshot_max = 0;
233 unsigned int memorystatus_jetsam_snapshot_size = 0;
234 uint64_t memorystatus_jetsam_snapshot_last_timestamp = 0;
235 uint64_t memorystatus_jetsam_snapshot_timeout = 0;
236
237 /* General memorystatus stuff */
238
239 uint64_t memorystatus_sysprocs_idle_delay_time = 0;
240 uint64_t memorystatus_apps_idle_delay_time = 0;
241
242 static lck_grp_attr_t *memorystatus_jetsam_fg_band_lock_grp_attr;
243 static lck_grp_t *memorystatus_jetsam_fg_band_lock_grp;
244 lck_mtx_t memorystatus_jetsam_fg_band_lock;
245
246 /* Idle guard handling */
247
248 static int32_t memorystatus_scheduled_idle_demotions_sysprocs = 0;
249 static int32_t memorystatus_scheduled_idle_demotions_apps = 0;
250
251 static void memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2);
252 static void memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state);
253 static void memorystatus_reschedule_idle_demotion_locked(void);
254 int memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap);
255 vm_pressure_level_t convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
256 boolean_t is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
257 void memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear);
258 void memorystatus_send_low_swap_note(void);
259 int memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index);
260 boolean_t memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, unsigned int band, int aggr_count,
261 uint32_t *errors, uint64_t *memory_reclaimed);
262 uint64_t memorystatus_available_memory_internal(proc_t p);
263
264 unsigned int memorystatus_level = 0;
265 static int memorystatus_list_count = 0;
266 memstat_bucket_t memstat_bucket[MEMSTAT_BUCKET_COUNT];
267 static thread_call_t memorystatus_idle_demotion_call;
268 uint64_t memstat_idle_demotion_deadline = 0;
269 int system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
270 int applications_aging_band = JETSAM_PRIORITY_IDLE;
271
272 #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)))
273
274 #define kJetsamAgingPolicyNone (0)
275 #define kJetsamAgingPolicyLegacy (1)
276 #define kJetsamAgingPolicySysProcsReclaimedFirst (2)
277 #define kJetsamAgingPolicyAppsReclaimedFirst (3)
278 #define kJetsamAgingPolicyMax kJetsamAgingPolicyAppsReclaimedFirst
279
280 unsigned int jetsam_aging_policy = kJetsamAgingPolicySysProcsReclaimedFirst;
281
282 extern int corpse_for_fatal_memkill;
283 extern uint64_t vm_purgeable_purge_task_owned(task_t task);
284 boolean_t memorystatus_allowed_vm_map_fork(task_t);
285 #if DEVELOPMENT || DEBUG
286 void memorystatus_abort_vm_map_fork(task_t);
287 #endif
288
289 /*
290 * Idle delay timeout factors for daemons based on relaunch behavior. Only used in
291 * kJetsamAgingPolicySysProcsReclaimedFirst aging policy.
292 */
293 #define kJetsamSysProcsIdleDelayTimeLowRatio (5)
294 #define kJetsamSysProcsIdleDelayTimeMedRatio (2)
295 #define kJetsamSysProcsIdleDelayTimeHighRatio (1)
296 static_assert(kJetsamSysProcsIdleDelayTimeLowRatio <= DEFERRED_IDLE_EXIT_TIME_SECS, "sysproc idle delay time for low relaunch daemons would be 0");
297
298 /*
299 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, treat apps as well
300 * behaved daemons for aging purposes.
301 */
302 #define kJetsamAppsIdleDelayTimeRatio (kJetsamSysProcsIdleDelayTimeLowRatio)
303
304 static uint64_t
305 memorystatus_sysprocs_idle_time(proc_t p)
306 {
307 /*
308 * The kJetsamAgingPolicySysProcsReclaimedFirst aging policy uses the relaunch behavior to
309 * determine the exact idle deferred time provided to the daemons. For all other aging
310 * policies, simply return the default aging idle time.
311 */
312 if (jetsam_aging_policy != kJetsamAgingPolicySysProcsReclaimedFirst) {
313 return memorystatus_sysprocs_idle_delay_time;
314 }
315
316 uint64_t idle_delay_time = 0;
317 /*
318 * For system processes, base the idle delay time on the
319 * jetsam relaunch behavior specified by launchd. The idea
320 * is to provide extra protection to the daemons which would
321 * relaunch immediately after jetsam.
322 */
323 switch (p->p_memstat_relaunch_flags) {
324 case P_MEMSTAT_RELAUNCH_UNKNOWN:
325 case P_MEMSTAT_RELAUNCH_LOW:
326 idle_delay_time = memorystatus_sysprocs_idle_delay_time / kJetsamSysProcsIdleDelayTimeLowRatio;
327 break;
328 case P_MEMSTAT_RELAUNCH_MED:
329 idle_delay_time = memorystatus_sysprocs_idle_delay_time / kJetsamSysProcsIdleDelayTimeMedRatio;
330 break;
331 case P_MEMSTAT_RELAUNCH_HIGH:
332 idle_delay_time = memorystatus_sysprocs_idle_delay_time / kJetsamSysProcsIdleDelayTimeHighRatio;
333 break;
334 default:
335 panic("Unknown relaunch flags on process!");
336 break;
337 }
338 return idle_delay_time;
339 }
340
341 static uint64_t
342 memorystatus_apps_idle_time(__unused proc_t p)
343 {
344 /*
345 * For kJetsamAgingPolicySysProcsReclaimedFirst, the Apps are considered as low
346 * relaunch candidates. So only provide limited protection to them. In the other
347 * aging policies, return the default aging idle time.
348 */
349 if (jetsam_aging_policy != kJetsamAgingPolicySysProcsReclaimedFirst) {
350 return memorystatus_apps_idle_delay_time;
351 }
352
353 return memorystatus_apps_idle_delay_time / kJetsamAppsIdleDelayTimeRatio;
354 }
355
356
357 #if 0
358
359 /* Keeping around for future use if we need a utility that can do this OR an app that needs a dynamic adjustment. */
360
361 static int
362 sysctl_set_jetsam_aging_policy SYSCTL_HANDLER_ARGS
363 {
364 #pragma unused(oidp, arg1, arg2)
365
366 int error = 0, val = 0;
367 memstat_bucket_t *old_bucket = 0;
368 int old_system_procs_aging_band = 0, new_system_procs_aging_band = 0;
369 int old_applications_aging_band = 0, new_applications_aging_band = 0;
370 proc_t p = NULL, next_proc = NULL;
371
372
373 error = sysctl_io_number(req, jetsam_aging_policy, sizeof(int), &val, NULL);
374 if (error || !req->newptr) {
375 return error;
376 }
377
378 if ((val < 0) || (val > kJetsamAgingPolicyMax)) {
379 printf("jetsam: ordering policy sysctl has invalid value - %d\n", val);
380 return EINVAL;
381 }
382
383 /*
384 * We need to synchronize with any potential adding/removal from aging bands
385 * that might be in progress currently. We use the proc_list_lock() just for
386 * consistency with all the routines dealing with 'aging' processes. We need
387 * a lighterweight lock.
388 */
389 proc_list_lock();
390
391 old_system_procs_aging_band = system_procs_aging_band;
392 old_applications_aging_band = applications_aging_band;
393
394 switch (val) {
395 case kJetsamAgingPolicyNone:
396 new_system_procs_aging_band = JETSAM_PRIORITY_IDLE;
397 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
398 break;
399
400 case kJetsamAgingPolicyLegacy:
401 /*
402 * Legacy behavior where some daemons get a 10s protection once and only before the first clean->dirty->clean transition before going into IDLE band.
403 */
404 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
405 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
406 break;
407
408 case kJetsamAgingPolicySysProcsReclaimedFirst:
409 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
410 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
411 break;
412
413 case kJetsamAgingPolicyAppsReclaimedFirst:
414 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
415 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
416 break;
417
418 default:
419 break;
420 }
421
422 if (old_system_procs_aging_band && (old_system_procs_aging_band != new_system_procs_aging_band)) {
423 old_bucket = &memstat_bucket[old_system_procs_aging_band];
424 p = TAILQ_FIRST(&old_bucket->list);
425
426 while (p) {
427 next_proc = TAILQ_NEXT(p, p_memstat_list);
428
429 if (isSysProc(p)) {
430 if (new_system_procs_aging_band == JETSAM_PRIORITY_IDLE) {
431 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
432 }
433
434 memorystatus_update_priority_locked(p, new_system_procs_aging_band, false, true);
435 }
436
437 p = next_proc;
438 continue;
439 }
440 }
441
442 if (old_applications_aging_band && (old_applications_aging_band != new_applications_aging_band)) {
443 old_bucket = &memstat_bucket[old_applications_aging_band];
444 p = TAILQ_FIRST(&old_bucket->list);
445
446 while (p) {
447 next_proc = TAILQ_NEXT(p, p_memstat_list);
448
449 if (isApp(p)) {
450 if (new_applications_aging_band == JETSAM_PRIORITY_IDLE) {
451 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
452 }
453
454 memorystatus_update_priority_locked(p, new_applications_aging_band, false, true);
455 }
456
457 p = next_proc;
458 continue;
459 }
460 }
461
462 jetsam_aging_policy = val;
463 system_procs_aging_band = new_system_procs_aging_band;
464 applications_aging_band = new_applications_aging_band;
465
466 proc_list_unlock();
467
468 return 0;
469 }
470
471 SYSCTL_PROC(_kern, OID_AUTO, set_jetsam_aging_policy, CTLTYPE_INT | CTLFLAG_RW,
472 0, 0, sysctl_set_jetsam_aging_policy, "I", "Jetsam Aging Policy");
473 #endif /*0*/
474
475 static int
476 sysctl_jetsam_set_sysprocs_idle_delay_time SYSCTL_HANDLER_ARGS
477 {
478 #pragma unused(oidp, arg1, arg2)
479
480 int error = 0, val = 0, old_time_in_secs = 0;
481 uint64_t old_time_in_ns = 0;
482
483 absolutetime_to_nanoseconds(memorystatus_sysprocs_idle_delay_time, &old_time_in_ns);
484 old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
485
486 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
487 if (error || !req->newptr) {
488 return error;
489 }
490
491 if ((val < 0) || (val > INT32_MAX)) {
492 printf("jetsam: new idle delay interval has invalid value.\n");
493 return EINVAL;
494 }
495
496 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
497
498 return 0;
499 }
500
501 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_sysprocs_idle_delay_time, CTLTYPE_INT | CTLFLAG_RW,
502 0, 0, sysctl_jetsam_set_sysprocs_idle_delay_time, "I", "Aging window for system processes");
503
504
505 static int
506 sysctl_jetsam_set_apps_idle_delay_time SYSCTL_HANDLER_ARGS
507 {
508 #pragma unused(oidp, arg1, arg2)
509
510 int error = 0, val = 0, old_time_in_secs = 0;
511 uint64_t old_time_in_ns = 0;
512
513 absolutetime_to_nanoseconds(memorystatus_apps_idle_delay_time, &old_time_in_ns);
514 old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
515
516 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
517 if (error || !req->newptr) {
518 return error;
519 }
520
521 if ((val < 0) || (val > INT32_MAX)) {
522 printf("jetsam: new idle delay interval has invalid value.\n");
523 return EINVAL;
524 }
525
526 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
527
528 return 0;
529 }
530
531 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_apps_idle_delay_time, CTLTYPE_INT | CTLFLAG_RW,
532 0, 0, sysctl_jetsam_set_apps_idle_delay_time, "I", "Aging window for applications");
533
534 SYSCTL_INT(_kern, OID_AUTO, jetsam_aging_policy, CTLTYPE_INT | CTLFLAG_RD, &jetsam_aging_policy, 0, "");
535
536 static unsigned int memorystatus_dirty_count = 0;
537
538 SYSCTL_INT(_kern, OID_AUTO, max_task_pmem, CTLFLAG_RD | CTLFLAG_LOCKED | CTLFLAG_MASKED, &max_task_footprint_mb, 0, "");
539
540 static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. */
541 static boolean_t proc_jetsam_state_is_active_locked(proc_t);
542
543 #if __arm64__
544 #if CONFIG_MEMORYSTATUS
545 int legacy_footprint_bonus_mb = 50; /* This value was chosen after looking at the top 30 apps
546 * that needed the additional room in their footprint when
547 * the 'correct' accounting methods were applied to them.
548 */
549
550 #if DEVELOPMENT || DEBUG
551 SYSCTL_INT(_kern, OID_AUTO, legacy_footprint_bonus_mb, CTLFLAG_RW | CTLFLAG_LOCKED, &legacy_footprint_bonus_mb, 0, "");
552 #endif /* DEVELOPMENT || DEBUG */
553
554 void
555 memorystatus_act_on_legacy_footprint_entitlement(proc_t p, boolean_t footprint_increase)
556 {
557 int memlimit_mb_active = 0, memlimit_mb_inactive = 0;
558 boolean_t memlimit_active_is_fatal = FALSE, memlimit_inactive_is_fatal = 0, use_active_limit = FALSE;
559
560 if (p == NULL) {
561 return;
562 }
563
564 proc_list_lock();
565
566 if (p->p_memstat_memlimit_active > 0) {
567 memlimit_mb_active = p->p_memstat_memlimit_active;
568 } else if (p->p_memstat_memlimit_active == -1) {
569 memlimit_mb_active = max_task_footprint_mb;
570 } else {
571 /*
572 * Nothing to do for '0' which is
573 * a special value only used internally
574 * to test 'no limits'.
575 */
576 proc_list_unlock();
577 return;
578 }
579
580 if (p->p_memstat_memlimit_inactive > 0) {
581 memlimit_mb_inactive = p->p_memstat_memlimit_inactive;
582 } else if (p->p_memstat_memlimit_inactive == -1) {
583 memlimit_mb_inactive = max_task_footprint_mb;
584 } else {
585 /*
586 * Nothing to do for '0' which is
587 * a special value only used internally
588 * to test 'no limits'.
589 */
590 proc_list_unlock();
591 return;
592 }
593
594 if (footprint_increase) {
595 memlimit_mb_active += legacy_footprint_bonus_mb;
596 memlimit_mb_inactive += legacy_footprint_bonus_mb;
597 } else {
598 memlimit_mb_active -= legacy_footprint_bonus_mb;
599 if (memlimit_mb_active == max_task_footprint_mb) {
600 memlimit_mb_active = -1; /* reverting back to default system limit */
601 }
602
603 memlimit_mb_inactive -= legacy_footprint_bonus_mb;
604 if (memlimit_mb_inactive == max_task_footprint_mb) {
605 memlimit_mb_inactive = -1; /* reverting back to default system limit */
606 }
607 }
608
609 memlimit_active_is_fatal = (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL);
610 memlimit_inactive_is_fatal = (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL);
611
612 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_mb_active, memlimit_active_is_fatal);
613 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_mb_inactive, memlimit_inactive_is_fatal);
614
615 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
616 use_active_limit = TRUE;
617 CACHE_ACTIVE_LIMITS_LOCKED(p, memlimit_active_is_fatal);
618 } else {
619 CACHE_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive_is_fatal);
620 }
621
622
623 if (memorystatus_highwater_enabled) {
624 task_set_phys_footprint_limit_internal(p->task,
625 (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1,
626 NULL, /*return old value */
627 use_active_limit, /*active limit?*/
628 (use_active_limit ? memlimit_active_is_fatal : memlimit_inactive_is_fatal));
629 }
630
631 proc_list_unlock();
632 }
633
634 void
635 memorystatus_act_on_ios13extended_footprint_entitlement(proc_t p)
636 {
637 int memlimit_mb_active = 0, memlimit_mb_inactive = 0;
638 boolean_t memlimit_active_is_fatal = FALSE, memlimit_inactive_is_fatal = 0, use_active_limit = FALSE;
639
640 if (max_mem < 1500ULL * 1024 * 1024 ||
641 max_mem > 2ULL * 1024 * 1024 * 1024) {
642 /* ios13extended_footprint is only for 2GB devices */
643 return;
644 }
645
646 proc_list_lock();
647
648 if (p->p_memstat_memlimit_active > 0) {
649 memlimit_mb_active = p->p_memstat_memlimit_active;
650 } else if (p->p_memstat_memlimit_active == -1) {
651 memlimit_mb_active = max_task_footprint_mb;
652 } else {
653 /*
654 * Nothing to do for '0' which is
655 * a special value only used internally
656 * to test 'no limits'.
657 */
658 proc_list_unlock();
659 return;
660 }
661
662 if (p->p_memstat_memlimit_inactive > 0) {
663 memlimit_mb_inactive = p->p_memstat_memlimit_inactive;
664 } else if (p->p_memstat_memlimit_inactive == -1) {
665 memlimit_mb_inactive = max_task_footprint_mb;
666 } else {
667 /*
668 * Nothing to do for '0' which is
669 * a special value only used internally
670 * to test 'no limits'.
671 */
672 proc_list_unlock();
673 return;
674 }
675
676 /* limit to "almost 2GB" */
677 int ios13extended_footprint_mb = 1800;
678 if (memlimit_mb_active > ios13extended_footprint_mb) {
679 /* do not lower the current limit */
680 proc_list_unlock();
681 return;
682 }
683 memlimit_mb_active = ios13extended_footprint_mb;
684 memlimit_mb_inactive = ios13extended_footprint_mb;
685
686 memlimit_active_is_fatal = (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL);
687 memlimit_inactive_is_fatal = (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL);
688
689 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_mb_active, memlimit_active_is_fatal);
690 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_mb_inactive, memlimit_inactive_is_fatal);
691
692 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
693 use_active_limit = TRUE;
694 CACHE_ACTIVE_LIMITS_LOCKED(p, memlimit_active_is_fatal);
695 } else {
696 CACHE_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive_is_fatal);
697 }
698
699
700 if (memorystatus_highwater_enabled) {
701 task_set_phys_footprint_limit_internal(p->task,
702 (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1,
703 NULL, /*return old value */
704 use_active_limit, /*active limit?*/
705 (use_active_limit ? memlimit_active_is_fatal : memlimit_inactive_is_fatal));
706 }
707
708 proc_list_unlock();
709 }
710
711 #endif /* CONFIG_MEMORYSTATUS */
712 #endif /* __arm64__ */
713
714 #if CONFIG_EMBEDDED
715
716 SYSCTL_INT(_kern, OID_AUTO, memorystatus_level, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_level, 0, "");
717
718 #endif /* CONFIG_EMBEDDED */
719
720 int
721 memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
722 {
723 user_addr_t level = 0;
724
725 level = args->level;
726
727 if (copyout(&memorystatus_level, level, sizeof(memorystatus_level)) != 0) {
728 return EFAULT;
729 }
730
731 return 0;
732 }
733
734 static void memorystatus_thread(void *param __unused, wait_result_t wr __unused);
735
736 /* Memory Limits */
737
738 static boolean_t memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
739 static boolean_t memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
740
741
742 static int memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
743
744 static int memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry);
745
746 static int memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
747
748 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);
749
750 static void memorystatus_get_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t *p_entry);
751 static int memorystatus_set_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t *p_entry);
752
753 int proc_get_memstat_priority(proc_t, boolean_t);
754
755 static boolean_t memorystatus_idle_snapshot = 0;
756
757 unsigned int memorystatus_delta = 0;
758
759 /* Jetsam Loop Detection */
760 static boolean_t memorystatus_jld_enabled = FALSE; /* Enable jetsam loop detection */
761 static uint32_t memorystatus_jld_eval_period_msecs = 0; /* Init pass sets this based on device memory size */
762 static int memorystatus_jld_eval_aggressive_count = 3; /* Raise the priority max after 'n' aggressive loops */
763 static int memorystatus_jld_eval_aggressive_priority_band_max = 15; /* Kill aggressively up through this band */
764
765 /*
766 * A FG app can request that the aggressive jetsam mechanism display some leniency in the FG band. This 'lenient' mode is described as:
767 * --- 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.
768 *
769 * RESTRICTIONS:
770 * - Such a request is respected/acknowledged only once while that 'requesting' app is in the FG band i.e. if aggressive jetsam was
771 * needed and the 'lenient' mode was deployed then that's it for this special mode while the app is in the FG band.
772 *
773 * - 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.
774 *
775 * - Also, the transition of the 'requesting' app away from the FG band will void this special behavior.
776 */
777
778 #define AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD 25
779 boolean_t memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
780 boolean_t memorystatus_aggressive_jetsam_lenient = FALSE;
781
782 #if DEVELOPMENT || DEBUG
783 /*
784 * Jetsam Loop Detection tunables.
785 */
786
787 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_period_msecs, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_period_msecs, 0, "");
788 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_count, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_count, 0, "");
789 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_priority_band_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_priority_band_max, 0, "");
790 #endif /* DEVELOPMENT || DEBUG */
791
792 static uint32_t kill_under_pressure_cause = 0;
793
794 /*
795 * snapshot support for memstats collected at boot.
796 */
797 static memorystatus_jetsam_snapshot_t memorystatus_at_boot_snapshot;
798
799 static void memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count);
800 static boolean_t memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount);
801 static void memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime);
802
803 static void memorystatus_clear_errors(void);
804 static void memorystatus_get_task_phys_footprint_page_counts(task_t task,
805 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
806 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
807 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
808 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages);
809
810 static void memorystatus_get_task_memory_region_count(task_t task, uint64_t *count);
811
812 static uint32_t memorystatus_build_state(proc_t p);
813 //static boolean_t memorystatus_issue_pressure_kevent(boolean_t pressured);
814
815 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,
816 uint32_t *errors, uint64_t *memory_reclaimed);
817 static boolean_t memorystatus_kill_processes_aggressive(uint32_t cause, int aggr_count, int32_t priority_max, uint32_t *errors, uint64_t *memory_reclaimed);
818 static boolean_t memorystatus_kill_hiwat_proc(uint32_t *errors, boolean_t *purged, uint64_t *memory_reclaimed);
819
820 static boolean_t memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause);
821
822 /* Priority Band Sorting Routines */
823 static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order);
824 static int memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order);
825 static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index);
826 static int memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz);
827
828 /* qsort routines */
829 typedef int (*cmpfunc_t)(const void *a, const void *b);
830 extern void qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
831 static int memstat_asc_cmp(const void *a, const void *b);
832
833 /* VM pressure */
834
835 extern unsigned int vm_page_free_count;
836 extern unsigned int vm_page_active_count;
837 extern unsigned int vm_page_inactive_count;
838 extern unsigned int vm_page_throttled_count;
839 extern unsigned int vm_page_purgeable_count;
840 extern unsigned int vm_page_wire_count;
841 #if CONFIG_SECLUDED_MEMORY
842 extern unsigned int vm_page_secluded_count;
843 extern unsigned int vm_page_secluded_count_over_target;
844 #endif /* CONFIG_SECLUDED_MEMORY */
845
846 /* Aggressive jetsam pages threshold for sysproc aging policy */
847 unsigned int memorystatus_sysproc_aging_aggr_pages = 0;
848
849 #if CONFIG_JETSAM
850 unsigned int memorystatus_available_pages = (unsigned int)-1;
851 unsigned int memorystatus_available_pages_pressure = 0;
852 unsigned int memorystatus_available_pages_critical = 0;
853 unsigned int memorystatus_available_pages_critical_base = 0;
854 unsigned int memorystatus_available_pages_critical_idle_offset = 0;
855
856 #if DEVELOPMENT || DEBUG
857 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
858 #else
859 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_MASKED | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
860 #endif /* DEVELOPMENT || DEBUG */
861
862 static unsigned int memorystatus_jetsam_policy = kPolicyDefault;
863 unsigned int memorystatus_policy_more_free_offset_pages = 0;
864 static void memorystatus_update_levels_locked(boolean_t critical_only);
865 static unsigned int memorystatus_thread_wasted_wakeup = 0;
866
867 /* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
868 extern void vm_thrashing_jetsam_done(void);
869 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);
870 #if DEVELOPMENT || DEBUG
871 static inline uint32_t
872 roundToNearestMB(uint32_t in)
873 {
874 return (in + ((1 << 20) - 1)) >> 20;
875 }
876
877 static int memorystatus_cmd_increase_jetsam_task_limit(pid_t pid, uint32_t byte_increase);
878 #endif
879
880 int32_t max_kill_priority = JETSAM_PRIORITY_MAX;
881
882 #else /* CONFIG_JETSAM */
883
884 uint64_t memorystatus_available_pages = (uint64_t)-1;
885 uint64_t memorystatus_available_pages_pressure = (uint64_t)-1;
886 uint64_t memorystatus_available_pages_critical = (uint64_t)-1;
887
888 int32_t max_kill_priority = JETSAM_PRIORITY_IDLE;
889 #endif /* CONFIG_JETSAM */
890
891 #if DEVELOPMENT || DEBUG
892
893 lck_grp_attr_t *disconnect_page_mappings_lck_grp_attr;
894 lck_grp_t *disconnect_page_mappings_lck_grp;
895 static lck_mtx_t disconnect_page_mappings_mutex;
896
897 extern boolean_t kill_on_no_paging_space;
898 #endif /* DEVELOPMENT || DEBUG */
899
900
901 /* Debug */
902
903 extern struct knote *vm_find_knote_from_pid(pid_t, struct klist *);
904
905 #if DEVELOPMENT || DEBUG
906
907 static unsigned int memorystatus_debug_dump_this_bucket = 0;
908
909 static void
910 memorystatus_debug_dump_bucket_locked(unsigned int bucket_index)
911 {
912 proc_t p = NULL;
913 uint64_t bytes = 0;
914 int ledger_limit = 0;
915 unsigned int b = bucket_index;
916 boolean_t traverse_all_buckets = FALSE;
917
918 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
919 traverse_all_buckets = TRUE;
920 b = 0;
921 } else {
922 traverse_all_buckets = FALSE;
923 b = bucket_index;
924 }
925
926 /*
927 * footprint reported in [pages / MB ]
928 * limits reported as:
929 * L-limit proc's Ledger limit
930 * C-limit proc's Cached limit, should match Ledger
931 * A-limit proc's Active limit
932 * IA-limit proc's Inactive limit
933 * F==Fatal, NF==NonFatal
934 */
935
936 printf("memorystatus_debug_dump ***START*(PAGE_SIZE_64=%llu)**\n", PAGE_SIZE_64);
937 printf("bucket [pid] [pages / MB] [state] [EP / RP / AP] dirty deadline [L-limit / C-limit / A-limit / IA-limit] name\n");
938 p = memorystatus_get_first_proc_locked(&b, traverse_all_buckets);
939 while (p) {
940 bytes = get_task_phys_footprint(p->task);
941 task_get_phys_footprint_limit(p->task, &ledger_limit);
942 printf("%2d [%5d] [%5lld /%3lldMB] 0x%-8x [%2d / %2d / %2d] 0x%-3x %10lld [%3d / %3d%s / %3d%s / %3d%s] %s\n",
943 b, p->p_pid,
944 (bytes / PAGE_SIZE_64), /* task's footprint converted from bytes to pages */
945 (bytes / (1024ULL * 1024ULL)), /* task's footprint converted from bytes to MB */
946 p->p_memstat_state, p->p_memstat_effectivepriority, p->p_memstat_requestedpriority, p->p_memstat_assertionpriority,
947 p->p_memstat_dirty, p->p_memstat_idledeadline,
948 ledger_limit,
949 p->p_memstat_memlimit,
950 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"),
951 p->p_memstat_memlimit_active,
952 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL ? "F " : "NF"),
953 p->p_memstat_memlimit_inactive,
954 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL ? "F " : "NF"),
955 (*p->p_name ? p->p_name : "unknown"));
956 p = memorystatus_get_next_proc_locked(&b, p, traverse_all_buckets);
957 }
958 printf("memorystatus_debug_dump ***END***\n");
959 }
960
961 static int
962 sysctl_memorystatus_debug_dump_bucket SYSCTL_HANDLER_ARGS
963 {
964 #pragma unused(oidp, arg2)
965 int bucket_index = 0;
966 int error;
967 error = SYSCTL_OUT(req, arg1, sizeof(int));
968 if (error || !req->newptr) {
969 return error;
970 }
971 error = SYSCTL_IN(req, &bucket_index, sizeof(int));
972 if (error || !req->newptr) {
973 return error;
974 }
975 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
976 /*
977 * All jetsam buckets will be dumped.
978 */
979 } else {
980 /*
981 * Only a single bucket will be dumped.
982 */
983 }
984
985 proc_list_lock();
986 memorystatus_debug_dump_bucket_locked(bucket_index);
987 proc_list_unlock();
988 memorystatus_debug_dump_this_bucket = bucket_index;
989 return error;
990 }
991
992 /*
993 * Debug aid to look at jetsam buckets and proc jetsam fields.
994 * Use this sysctl to act on a particular jetsam bucket.
995 * Writing the sysctl triggers the dump.
996 * Usage: sysctl kern.memorystatus_debug_dump_this_bucket=<bucket_index>
997 */
998
999 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", "");
1000
1001
1002 /* Debug aid to aid determination of limit */
1003
1004 static int
1005 sysctl_memorystatus_highwater_enable SYSCTL_HANDLER_ARGS
1006 {
1007 #pragma unused(oidp, arg2)
1008 proc_t p;
1009 unsigned int b = 0;
1010 int error, enable = 0;
1011 boolean_t use_active; /* use the active limit and active limit attributes */
1012 boolean_t is_fatal;
1013
1014 error = SYSCTL_OUT(req, arg1, sizeof(int));
1015 if (error || !req->newptr) {
1016 return error;
1017 }
1018
1019 error = SYSCTL_IN(req, &enable, sizeof(int));
1020 if (error || !req->newptr) {
1021 return error;
1022 }
1023
1024 if (!(enable == 0 || enable == 1)) {
1025 return EINVAL;
1026 }
1027
1028 proc_list_lock();
1029
1030 p = memorystatus_get_first_proc_locked(&b, TRUE);
1031 while (p) {
1032 use_active = proc_jetsam_state_is_active_locked(p);
1033
1034 if (enable) {
1035 if (use_active == TRUE) {
1036 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
1037 } else {
1038 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
1039 }
1040 } else {
1041 /*
1042 * Disabling limits does not touch the stored variants.
1043 * Set the cached limit fields to system_wide defaults.
1044 */
1045 p->p_memstat_memlimit = -1;
1046 p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
1047 is_fatal = TRUE;
1048 }
1049
1050 /*
1051 * Enforce the cached limit by writing to the ledger.
1052 */
1053 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit: -1, NULL, use_active, is_fatal);
1054
1055 p = memorystatus_get_next_proc_locked(&b, p, TRUE);
1056 }
1057
1058 memorystatus_highwater_enabled = enable;
1059
1060 proc_list_unlock();
1061
1062 return 0;
1063 }
1064
1065 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_highwater_enabled, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_highwater_enabled, 0, sysctl_memorystatus_highwater_enable, "I", "");
1066
1067 SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
1068
1069 #if CONFIG_JETSAM
1070 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
1071 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
1072 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
1073 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_policy_more_free_offset_pages, CTLFLAG_RW, &memorystatus_policy_more_free_offset_pages, 0, "");
1074
1075 static unsigned int memorystatus_jetsam_panic_debug = 0;
1076
1077 #if VM_PRESSURE_EVENTS
1078
1079 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_pressure, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_pressure, 0, "");
1080
1081 #endif /* VM_PRESSURE_EVENTS */
1082
1083 #endif /* CONFIG_JETSAM */
1084
1085 #endif /* DEVELOPMENT || DEBUG */
1086
1087 extern kern_return_t kernel_thread_start_priority(thread_continue_t continuation,
1088 void *parameter,
1089 integer_t priority,
1090 thread_t *new_thread);
1091
1092 #if DEVELOPMENT || DEBUG
1093
1094 static int
1095 sysctl_memorystatus_disconnect_page_mappings SYSCTL_HANDLER_ARGS
1096 {
1097 #pragma unused(arg1, arg2)
1098 int error = 0, pid = 0;
1099 proc_t p;
1100
1101 error = sysctl_handle_int(oidp, &pid, 0, req);
1102 if (error || !req->newptr) {
1103 return error;
1104 }
1105
1106 lck_mtx_lock(&disconnect_page_mappings_mutex);
1107
1108 if (pid == -1) {
1109 vm_pageout_disconnect_all_pages();
1110 } else {
1111 p = proc_find(pid);
1112
1113 if (p != NULL) {
1114 error = task_disconnect_page_mappings(p->task);
1115
1116 proc_rele(p);
1117
1118 if (error) {
1119 error = EIO;
1120 }
1121 } else {
1122 error = EINVAL;
1123 }
1124 }
1125 lck_mtx_unlock(&disconnect_page_mappings_mutex);
1126
1127 return error;
1128 }
1129
1130 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_disconnect_page_mappings, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
1131 0, 0, &sysctl_memorystatus_disconnect_page_mappings, "I", "");
1132
1133 #endif /* DEVELOPMENT || DEBUG */
1134
1135
1136 /*
1137 * Picks the sorting routine for a given jetsam priority band.
1138 *
1139 * Input:
1140 * bucket_index - jetsam priority band to be sorted.
1141 * sort_order - JETSAM_SORT_xxx from kern_memorystatus.h
1142 * Currently sort_order is only meaningful when handling
1143 * coalitions.
1144 *
1145 * Return:
1146 * 0 on success
1147 * non-0 on failure
1148 */
1149 static int
1150 memorystatus_sort_bucket(unsigned int bucket_index, int sort_order)
1151 {
1152 int coal_sort_order;
1153
1154 /*
1155 * Verify the jetsam priority
1156 */
1157 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1158 return EINVAL;
1159 }
1160
1161 #if DEVELOPMENT || DEBUG
1162 if (sort_order == JETSAM_SORT_DEFAULT) {
1163 coal_sort_order = COALITION_SORT_DEFAULT;
1164 } else {
1165 coal_sort_order = sort_order; /* only used for testing scenarios */
1166 }
1167 #else
1168 /* Verify default */
1169 if (sort_order == JETSAM_SORT_DEFAULT) {
1170 coal_sort_order = COALITION_SORT_DEFAULT;
1171 } else {
1172 return EINVAL;
1173 }
1174 #endif
1175
1176 proc_list_lock();
1177
1178 if (memstat_bucket[bucket_index].count == 0) {
1179 proc_list_unlock();
1180 return 0;
1181 }
1182
1183 switch (bucket_index) {
1184 case JETSAM_PRIORITY_FOREGROUND:
1185 if (memorystatus_sort_by_largest_coalition_locked(bucket_index, coal_sort_order) == 0) {
1186 /*
1187 * Fall back to per process sorting when zero coalitions are found.
1188 */
1189 memorystatus_sort_by_largest_process_locked(bucket_index);
1190 }
1191 break;
1192 default:
1193 memorystatus_sort_by_largest_process_locked(bucket_index);
1194 break;
1195 }
1196 proc_list_unlock();
1197
1198 return 0;
1199 }
1200
1201 /*
1202 * Sort processes by size for a single jetsam bucket.
1203 */
1204
1205 static void
1206 memorystatus_sort_by_largest_process_locked(unsigned int bucket_index)
1207 {
1208 proc_t p = NULL, insert_after_proc = NULL, max_proc = NULL;
1209 proc_t next_p = NULL, prev_max_proc = NULL;
1210 uint32_t pages = 0, max_pages = 0;
1211 memstat_bucket_t *current_bucket;
1212
1213 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1214 return;
1215 }
1216
1217 current_bucket = &memstat_bucket[bucket_index];
1218
1219 p = TAILQ_FIRST(&current_bucket->list);
1220
1221 while (p) {
1222 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1223 max_pages = pages;
1224 max_proc = p;
1225 prev_max_proc = p;
1226
1227 while ((next_p = TAILQ_NEXT(p, p_memstat_list)) != NULL) {
1228 /* traversing list until we find next largest process */
1229 p = next_p;
1230 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1231 if (pages > max_pages) {
1232 max_pages = pages;
1233 max_proc = p;
1234 }
1235 }
1236
1237 if (prev_max_proc != max_proc) {
1238 /* found a larger process, place it in the list */
1239 TAILQ_REMOVE(&current_bucket->list, max_proc, p_memstat_list);
1240 if (insert_after_proc == NULL) {
1241 TAILQ_INSERT_HEAD(&current_bucket->list, max_proc, p_memstat_list);
1242 } else {
1243 TAILQ_INSERT_AFTER(&current_bucket->list, insert_after_proc, max_proc, p_memstat_list);
1244 }
1245 prev_max_proc = max_proc;
1246 }
1247
1248 insert_after_proc = max_proc;
1249
1250 p = TAILQ_NEXT(max_proc, p_memstat_list);
1251 }
1252 }
1253
1254 proc_t
1255 memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search)
1256 {
1257 memstat_bucket_t *current_bucket;
1258 proc_t next_p;
1259
1260 if ((*bucket_index) >= MEMSTAT_BUCKET_COUNT) {
1261 return NULL;
1262 }
1263
1264 current_bucket = &memstat_bucket[*bucket_index];
1265 next_p = TAILQ_FIRST(&current_bucket->list);
1266 if (!next_p && search) {
1267 while (!next_p && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1268 current_bucket = &memstat_bucket[*bucket_index];
1269 next_p = TAILQ_FIRST(&current_bucket->list);
1270 }
1271 }
1272
1273 return next_p;
1274 }
1275
1276 proc_t
1277 memorystatus_get_next_proc_locked(unsigned int *bucket_index, proc_t p, boolean_t search)
1278 {
1279 memstat_bucket_t *current_bucket;
1280 proc_t next_p;
1281
1282 if (!p || ((*bucket_index) >= MEMSTAT_BUCKET_COUNT)) {
1283 return NULL;
1284 }
1285
1286 next_p = TAILQ_NEXT(p, p_memstat_list);
1287 while (!next_p && search && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1288 current_bucket = &memstat_bucket[*bucket_index];
1289 next_p = TAILQ_FIRST(&current_bucket->list);
1290 }
1291
1292 return next_p;
1293 }
1294
1295 /*
1296 * Structure to hold state for a jetsam thread.
1297 * Typically there should be a single jetsam thread
1298 * unless parallel jetsam is enabled.
1299 */
1300 struct jetsam_thread_state {
1301 uint8_t inited; /* boolean - if the thread is initialized */
1302 uint8_t limit_to_low_bands; /* boolean */
1303 int memorystatus_wakeup; /* wake channel */
1304 int index; /* jetsam thread index */
1305 thread_t thread; /* jetsam thread pointer */
1306 } *jetsam_threads;
1307
1308 /* Maximum number of jetsam threads allowed */
1309 #define JETSAM_THREADS_LIMIT 3
1310
1311 /* Number of active jetsam threads */
1312 _Atomic int active_jetsam_threads = 1;
1313
1314 /* Number of maximum jetsam threads configured */
1315 int max_jetsam_threads = JETSAM_THREADS_LIMIT;
1316
1317 /*
1318 * Global switch for enabling fast jetsam. Fast jetsam is
1319 * hooked up via the system_override() system call. It has the
1320 * following effects:
1321 * - Raise the jetsam threshold ("clear-the-deck")
1322 * - Enabled parallel jetsam on eligible devices
1323 */
1324 #if __AMP__
1325 int fast_jetsam_enabled = 1;
1326 #else /* __AMP__ */
1327 int fast_jetsam_enabled = 0;
1328 #endif /* __AMP__ */
1329
1330 /* Routine to find the jetsam state structure for the current jetsam thread */
1331 static inline struct jetsam_thread_state *
1332 jetsam_current_thread(void)
1333 {
1334 for (int thr_id = 0; thr_id < max_jetsam_threads; thr_id++) {
1335 if (jetsam_threads[thr_id].thread == current_thread()) {
1336 return &(jetsam_threads[thr_id]);
1337 }
1338 }
1339 return NULL;
1340 }
1341
1342
1343 __private_extern__ void
1344 memorystatus_init(void)
1345 {
1346 kern_return_t result;
1347 int i;
1348
1349 #if CONFIG_FREEZE
1350 memorystatus_freeze_jetsam_band = JETSAM_PRIORITY_UI_SUPPORT;
1351 memorystatus_frozen_processes_max = FREEZE_PROCESSES_MAX;
1352 memorystatus_frozen_shared_mb_max = ((MAX_FROZEN_SHARED_MB_PERCENT * max_task_footprint_mb) / 100); /* 10% of the system wide task limit */
1353 memorystatus_freeze_shared_mb_per_process_max = (memorystatus_frozen_shared_mb_max / 4);
1354 memorystatus_freeze_pages_min = FREEZE_PAGES_MIN;
1355 memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
1356 memorystatus_max_frozen_demotions_daily = MAX_FROZEN_PROCESS_DEMOTIONS;
1357 memorystatus_thaw_count_demotion_threshold = MIN_THAW_DEMOTION_THRESHOLD;
1358 #endif
1359
1360 #if DEVELOPMENT || DEBUG
1361 disconnect_page_mappings_lck_grp_attr = lck_grp_attr_alloc_init();
1362 disconnect_page_mappings_lck_grp = lck_grp_alloc_init("disconnect_page_mappings", disconnect_page_mappings_lck_grp_attr);
1363
1364 lck_mtx_init(&disconnect_page_mappings_mutex, disconnect_page_mappings_lck_grp, NULL);
1365
1366 if (kill_on_no_paging_space == TRUE) {
1367 max_kill_priority = JETSAM_PRIORITY_MAX;
1368 }
1369 #endif
1370
1371 memorystatus_jetsam_fg_band_lock_grp_attr = lck_grp_attr_alloc_init();
1372 memorystatus_jetsam_fg_band_lock_grp =
1373 lck_grp_alloc_init("memorystatus_jetsam_fg_band", memorystatus_jetsam_fg_band_lock_grp_attr);
1374 lck_mtx_init(&memorystatus_jetsam_fg_band_lock, memorystatus_jetsam_fg_band_lock_grp, NULL);
1375
1376 /* Init buckets */
1377 for (i = 0; i < MEMSTAT_BUCKET_COUNT; i++) {
1378 TAILQ_INIT(&memstat_bucket[i].list);
1379 memstat_bucket[i].count = 0;
1380 memstat_bucket[i].relaunch_high_count = 0;
1381 }
1382 memorystatus_idle_demotion_call = thread_call_allocate((thread_call_func_t)memorystatus_perform_idle_demotion, NULL);
1383
1384 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
1385 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
1386
1387 #if CONFIG_JETSAM
1388 /* Apply overrides */
1389 if (!PE_parse_boot_argn("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage))) {
1390 PE_get_default("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage));
1391 }
1392 if (delta_percentage == 0) {
1393 delta_percentage = 5;
1394 }
1395 if (max_mem > config_jetsam_large_memory_cutoff) {
1396 critical_threshold_percentage = critical_threshold_percentage_larger_devices;
1397 delta_percentage = delta_percentage_larger_devices;
1398 }
1399 assert(delta_percentage < 100);
1400 if (!PE_parse_boot_argn("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage))) {
1401 PE_get_default("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage));
1402 }
1403 assert(critical_threshold_percentage < 100);
1404 PE_get_default("kern.jetsam_idle_offset", &idle_offset_percentage, sizeof(idle_offset_percentage));
1405 assert(idle_offset_percentage < 100);
1406 PE_get_default("kern.jetsam_pressure_threshold", &pressure_threshold_percentage, sizeof(pressure_threshold_percentage));
1407 assert(pressure_threshold_percentage < 100);
1408 PE_get_default("kern.jetsam_freeze_threshold", &freeze_threshold_percentage, sizeof(freeze_threshold_percentage));
1409 assert(freeze_threshold_percentage < 100);
1410
1411
1412 if (!PE_parse_boot_argn("jetsam_aging_policy", &jetsam_aging_policy,
1413 sizeof(jetsam_aging_policy))) {
1414 if (!PE_get_default("kern.jetsam_aging_policy", &jetsam_aging_policy,
1415 sizeof(jetsam_aging_policy))) {
1416 jetsam_aging_policy = kJetsamAgingPolicySysProcsReclaimedFirst;
1417 }
1418 }
1419
1420 if (jetsam_aging_policy > kJetsamAgingPolicyMax) {
1421 jetsam_aging_policy = kJetsamAgingPolicySysProcsReclaimedFirst;
1422 }
1423
1424 switch (jetsam_aging_policy) {
1425 case kJetsamAgingPolicyNone:
1426 system_procs_aging_band = JETSAM_PRIORITY_IDLE;
1427 applications_aging_band = JETSAM_PRIORITY_IDLE;
1428 break;
1429
1430 case kJetsamAgingPolicyLegacy:
1431 /*
1432 * Legacy behavior where some daemons get a 10s protection once
1433 * AND only before the first clean->dirty->clean transition before
1434 * going into IDLE band.
1435 */
1436 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1437 applications_aging_band = JETSAM_PRIORITY_IDLE;
1438 break;
1439
1440 case kJetsamAgingPolicySysProcsReclaimedFirst:
1441 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1442 applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1443 break;
1444
1445 case kJetsamAgingPolicyAppsReclaimedFirst:
1446 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1447 applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1448 break;
1449
1450 default:
1451 break;
1452 }
1453
1454 /*
1455 * The aging bands cannot overlap with the JETSAM_PRIORITY_ELEVATED_INACTIVE
1456 * band and must be below it in priority. This is so that we don't have to make
1457 * our 'aging' code worry about a mix of processes, some of which need to age
1458 * and some others that need to stay elevated in the jetsam bands.
1459 */
1460 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > system_procs_aging_band);
1461 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > applications_aging_band);
1462
1463 /* Take snapshots for idle-exit kills by default? First check the boot-arg... */
1464 if (!PE_parse_boot_argn("jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot))) {
1465 /* ...no boot-arg, so check the device tree */
1466 PE_get_default("kern.jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot));
1467 }
1468
1469 memorystatus_delta = delta_percentage * atop_64(max_mem) / 100;
1470 memorystatus_available_pages_critical_idle_offset = idle_offset_percentage * atop_64(max_mem) / 100;
1471 memorystatus_available_pages_critical_base = (critical_threshold_percentage / delta_percentage) * memorystatus_delta;
1472 memorystatus_policy_more_free_offset_pages = (policy_more_free_offset_percentage / delta_percentage) * memorystatus_delta;
1473 memorystatus_sysproc_aging_aggr_pages = sysproc_aging_aggr_threshold_percentage * atop_64(max_mem) / 100;
1474
1475 /* Jetsam Loop Detection */
1476 if (max_mem <= (512 * 1024 * 1024)) {
1477 /* 512 MB devices */
1478 memorystatus_jld_eval_period_msecs = 8000; /* 8000 msecs == 8 second window */
1479 } else {
1480 /* 1GB and larger devices */
1481 memorystatus_jld_eval_period_msecs = 6000; /* 6000 msecs == 6 second window */
1482 }
1483
1484 memorystatus_jld_enabled = TRUE;
1485
1486 /* No contention at this point */
1487 memorystatus_update_levels_locked(FALSE);
1488
1489 #endif /* CONFIG_JETSAM */
1490
1491 memorystatus_jetsam_snapshot_max = maxproc;
1492
1493 memorystatus_jetsam_snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
1494 (sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_max);
1495
1496 memorystatus_jetsam_snapshot =
1497 (memorystatus_jetsam_snapshot_t*)kalloc(memorystatus_jetsam_snapshot_size);
1498 if (!memorystatus_jetsam_snapshot) {
1499 panic("Could not allocate memorystatus_jetsam_snapshot");
1500 }
1501
1502 memorystatus_jetsam_snapshot_copy =
1503 (memorystatus_jetsam_snapshot_t*)kalloc(memorystatus_jetsam_snapshot_size);
1504 if (!memorystatus_jetsam_snapshot_copy) {
1505 panic("Could not allocate memorystatus_jetsam_snapshot_copy");
1506 }
1507
1508 nanoseconds_to_absolutetime((uint64_t)JETSAM_SNAPSHOT_TIMEOUT_SECS * NSEC_PER_SEC, &memorystatus_jetsam_snapshot_timeout);
1509
1510 memset(&memorystatus_at_boot_snapshot, 0, sizeof(memorystatus_jetsam_snapshot_t));
1511
1512 #if CONFIG_FREEZE
1513 memorystatus_freeze_threshold = (freeze_threshold_percentage / delta_percentage) * memorystatus_delta;
1514 #endif
1515
1516 /* Check the boot-arg to see if fast jetsam is allowed */
1517 if (!PE_parse_boot_argn("fast_jetsam_enabled", &fast_jetsam_enabled, sizeof(fast_jetsam_enabled))) {
1518 fast_jetsam_enabled = 0;
1519 }
1520
1521 /* Check the boot-arg to configure the maximum number of jetsam threads */
1522 if (!PE_parse_boot_argn("max_jetsam_threads", &max_jetsam_threads, sizeof(max_jetsam_threads))) {
1523 max_jetsam_threads = JETSAM_THREADS_LIMIT;
1524 }
1525
1526 /* Restrict the maximum number of jetsam threads to JETSAM_THREADS_LIMIT */
1527 if (max_jetsam_threads > JETSAM_THREADS_LIMIT) {
1528 max_jetsam_threads = JETSAM_THREADS_LIMIT;
1529 }
1530
1531 /* For low CPU systems disable fast jetsam mechanism */
1532 if (vm_pageout_state.vm_restricted_to_single_processor == TRUE) {
1533 max_jetsam_threads = 1;
1534 fast_jetsam_enabled = 0;
1535 }
1536
1537 /* Initialize the jetsam_threads state array */
1538 jetsam_threads = kalloc(sizeof(struct jetsam_thread_state) * max_jetsam_threads);
1539
1540 /* Initialize all the jetsam threads */
1541 for (i = 0; i < max_jetsam_threads; i++) {
1542 jetsam_threads[i].inited = FALSE;
1543 jetsam_threads[i].index = i;
1544 result = kernel_thread_start_priority(memorystatus_thread, NULL, 95 /* MAXPRI_KERNEL */, &jetsam_threads[i].thread);
1545 if (result != KERN_SUCCESS) {
1546 panic("Could not create memorystatus_thread %d", i);
1547 }
1548 thread_deallocate(jetsam_threads[i].thread);
1549 }
1550 }
1551
1552 /* Centralised for the purposes of allowing panic-on-jetsam */
1553 extern void
1554 vm_run_compactor(void);
1555
1556 /*
1557 * The jetsam no frills kill call
1558 * Return: 0 on success
1559 * error code on failure (EINVAL...)
1560 */
1561 static int
1562 jetsam_do_kill(proc_t p, int jetsam_flags, os_reason_t jetsam_reason)
1563 {
1564 int error = 0;
1565 error = exit_with_reason(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags, jetsam_reason);
1566 return error;
1567 }
1568
1569 /*
1570 * Wrapper for processes exiting with memorystatus details
1571 */
1572 static boolean_t
1573 memorystatus_do_kill(proc_t p, uint32_t cause, os_reason_t jetsam_reason, uint64_t *footprint_of_killed_proc)
1574 {
1575 int error = 0;
1576 __unused pid_t victim_pid = p->p_pid;
1577 uint64_t footprint = get_task_phys_footprint(p->task);
1578 #if (KDEBUG_LEVEL >= KDEBUG_LEVEL_STANDARD)
1579 int32_t memstat_effectivepriority = p->p_memstat_effectivepriority;
1580 #endif /* (KDEBUG_LEVEL >= KDEBUG_LEVEL_STANDARD) */
1581
1582 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_START,
1583 victim_pid, cause, vm_page_free_count, footprint, 0);
1584 DTRACE_MEMORYSTATUS4(memorystatus_do_kill, proc_t, p, os_reason_t, jetsam_reason, uint32_t, cause, uint64_t, footprint);
1585 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
1586 if (memorystatus_jetsam_panic_debug & (1 << cause)) {
1587 panic("memorystatus_do_kill(): jetsam debug panic (cause: %d)", cause);
1588 }
1589 #else
1590 #pragma unused(cause)
1591 #endif
1592
1593 if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
1594 printf("memorystatus: killing process %d [%s] in high band %s (%d) - memorystatus_available_pages: %llu\n", p->p_pid,
1595 (*p->p_name ? p->p_name : "unknown"),
1596 memorystatus_priority_band_name(p->p_memstat_effectivepriority), p->p_memstat_effectivepriority,
1597 (uint64_t)memorystatus_available_pages);
1598 }
1599
1600 /*
1601 * The jetsam_reason (os_reason_t) has enough information about the kill cause.
1602 * We don't really need jetsam_flags anymore, so it's okay that not all possible kill causes have been mapped.
1603 */
1604 int jetsam_flags = P_LTERM_JETSAM;
1605 switch (cause) {
1606 case kMemorystatusKilledHiwat: jetsam_flags |= P_JETSAM_HIWAT; break;
1607 case kMemorystatusKilledVnodes: jetsam_flags |= P_JETSAM_VNODE; break;
1608 case kMemorystatusKilledVMPageShortage: jetsam_flags |= P_JETSAM_VMPAGESHORTAGE; break;
1609 case kMemorystatusKilledVMCompressorThrashing:
1610 case kMemorystatusKilledVMCompressorSpaceShortage: jetsam_flags |= P_JETSAM_VMTHRASHING; break;
1611 case kMemorystatusKilledFCThrashing: jetsam_flags |= P_JETSAM_FCTHRASHING; break;
1612 case kMemorystatusKilledPerProcessLimit: jetsam_flags |= P_JETSAM_PID; break;
1613 case kMemorystatusKilledIdleExit: jetsam_flags |= P_JETSAM_IDLEEXIT; break;
1614 }
1615 error = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
1616 *footprint_of_killed_proc = ((error == 0) ? footprint : 0);
1617
1618 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_END,
1619 victim_pid, memstat_effectivepriority, vm_page_free_count, error, 0);
1620
1621 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_COMPACTOR_RUN)) | DBG_FUNC_START,
1622 victim_pid, cause, vm_page_free_count, *footprint_of_killed_proc, 0);
1623
1624 vm_run_compactor();
1625
1626 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_COMPACTOR_RUN)) | DBG_FUNC_END,
1627 victim_pid, cause, vm_page_free_count, 0, 0);
1628
1629 return error == 0;
1630 }
1631
1632 /*
1633 * Node manipulation
1634 */
1635
1636 static void
1637 memorystatus_check_levels_locked(void)
1638 {
1639 #if CONFIG_JETSAM
1640 /* Update levels */
1641 memorystatus_update_levels_locked(TRUE);
1642 #else /* CONFIG_JETSAM */
1643 /*
1644 * Nothing to do here currently since we update
1645 * memorystatus_available_pages in vm_pressure_response.
1646 */
1647 #endif /* CONFIG_JETSAM */
1648 }
1649
1650 /*
1651 * Pin a process to a particular jetsam band when it is in the background i.e. not doing active work.
1652 * For an application: that means no longer in the FG band
1653 * For a daemon: that means no longer in its 'requested' jetsam priority band
1654 */
1655
1656 int
1657 memorystatus_update_inactive_jetsam_priority_band(pid_t pid, uint32_t op_flags, int jetsam_prio, boolean_t effective_now)
1658 {
1659 int error = 0;
1660 boolean_t enable = FALSE;
1661 proc_t p = NULL;
1662
1663 if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE) {
1664 enable = TRUE;
1665 } else if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE) {
1666 enable = FALSE;
1667 } else {
1668 return EINVAL;
1669 }
1670
1671 p = proc_find(pid);
1672 if (p != NULL) {
1673 if ((enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) ||
1674 (!enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == 0))) {
1675 /*
1676 * No change in state.
1677 */
1678 } else {
1679 proc_list_lock();
1680
1681 if (enable) {
1682 p->p_memstat_state |= P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
1683 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1684
1685 if (effective_now) {
1686 if (p->p_memstat_effectivepriority < jetsam_prio) {
1687 if (memorystatus_highwater_enabled) {
1688 /*
1689 * Process is about to transition from
1690 * inactive --> active
1691 * assign active state
1692 */
1693 boolean_t is_fatal;
1694 boolean_t use_active = TRUE;
1695 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
1696 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
1697 }
1698 memorystatus_update_priority_locked(p, jetsam_prio, FALSE, FALSE);
1699 }
1700 } else {
1701 if (isProcessInAgingBands(p)) {
1702 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1703 }
1704 }
1705 } else {
1706 p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
1707 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1708
1709 if (effective_now) {
1710 if (p->p_memstat_effectivepriority == jetsam_prio) {
1711 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1712 }
1713 } else {
1714 if (isProcessInAgingBands(p)) {
1715 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1716 }
1717 }
1718 }
1719
1720 proc_list_unlock();
1721 }
1722 proc_rele(p);
1723 error = 0;
1724 } else {
1725 error = ESRCH;
1726 }
1727
1728 return error;
1729 }
1730
1731 static void
1732 memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2)
1733 {
1734 proc_t p;
1735 uint64_t current_time = 0, idle_delay_time = 0;
1736 int demote_prio_band = 0;
1737 memstat_bucket_t *demotion_bucket;
1738
1739 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion()\n");
1740
1741 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_START, 0, 0, 0, 0, 0);
1742
1743 current_time = mach_absolute_time();
1744
1745 proc_list_lock();
1746
1747 demote_prio_band = JETSAM_PRIORITY_IDLE + 1;
1748
1749 for (; demote_prio_band < JETSAM_PRIORITY_MAX; demote_prio_band++) {
1750 if (demote_prio_band != system_procs_aging_band && demote_prio_band != applications_aging_band) {
1751 continue;
1752 }
1753
1754 demotion_bucket = &memstat_bucket[demote_prio_band];
1755 p = TAILQ_FIRST(&demotion_bucket->list);
1756
1757 while (p) {
1758 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion() found %d\n", p->p_pid);
1759
1760 assert(p->p_memstat_idledeadline);
1761
1762 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
1763
1764 if (current_time >= p->p_memstat_idledeadline) {
1765 if ((isSysProc(p) &&
1766 ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) != P_DIRTY_IDLE_EXIT_ENABLED)) || /* system proc marked dirty*/
1767 task_has_assertions((struct task *)(p->task))) { /* has outstanding assertions which might indicate outstanding work too */
1768 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_time(p) : memorystatus_apps_idle_time(p);
1769
1770 p->p_memstat_idledeadline += idle_delay_time;
1771 p = TAILQ_NEXT(p, p_memstat_list);
1772 } else {
1773 proc_t next_proc = NULL;
1774
1775 next_proc = TAILQ_NEXT(p, p_memstat_list);
1776 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1777
1778 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false, true);
1779
1780 p = next_proc;
1781 continue;
1782 }
1783 } else {
1784 // No further candidates
1785 break;
1786 }
1787 }
1788 }
1789
1790 memorystatus_reschedule_idle_demotion_locked();
1791
1792 proc_list_unlock();
1793
1794 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_END, 0, 0, 0, 0, 0);
1795 }
1796
1797 static void
1798 memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state)
1799 {
1800 boolean_t present_in_sysprocs_aging_bucket = FALSE;
1801 boolean_t present_in_apps_aging_bucket = FALSE;
1802 uint64_t idle_delay_time = 0;
1803
1804 if (jetsam_aging_policy == kJetsamAgingPolicyNone) {
1805 return;
1806 }
1807
1808 if ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) ||
1809 (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION)) {
1810 /*
1811 * This process isn't going to be making the trip to the lower bands.
1812 */
1813 return;
1814 }
1815
1816 if (isProcessInAgingBands(p)) {
1817 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
1818 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) != P_DIRTY_AGING_IN_PROGRESS);
1819 }
1820
1821 if (isSysProc(p) && system_procs_aging_band) {
1822 present_in_sysprocs_aging_bucket = TRUE;
1823 } else if (isApp(p) && applications_aging_band) {
1824 present_in_apps_aging_bucket = TRUE;
1825 }
1826 }
1827
1828 assert(!present_in_sysprocs_aging_bucket);
1829 assert(!present_in_apps_aging_bucket);
1830
1831 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",
1832 p->p_pid, p->p_memstat_dirty, set_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
1833
1834 if (isSysProc(p)) {
1835 assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
1836 }
1837
1838 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_time(p) : memorystatus_apps_idle_time(p);
1839 if (set_state) {
1840 p->p_memstat_dirty |= P_DIRTY_AGING_IN_PROGRESS;
1841 p->p_memstat_idledeadline = mach_absolute_time() + idle_delay_time;
1842 }
1843
1844 assert(p->p_memstat_idledeadline);
1845
1846 if (isSysProc(p) && present_in_sysprocs_aging_bucket == FALSE) {
1847 memorystatus_scheduled_idle_demotions_sysprocs++;
1848 } else if (isApp(p) && present_in_apps_aging_bucket == FALSE) {
1849 memorystatus_scheduled_idle_demotions_apps++;
1850 }
1851 }
1852
1853 void
1854 memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clear_state)
1855 {
1856 boolean_t present_in_sysprocs_aging_bucket = FALSE;
1857 boolean_t present_in_apps_aging_bucket = FALSE;
1858
1859 if (!system_procs_aging_band && !applications_aging_band) {
1860 return;
1861 }
1862
1863 if ((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == 0) {
1864 return;
1865 }
1866
1867 if (isProcessInAgingBands(p)) {
1868 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
1869 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == P_DIRTY_AGING_IN_PROGRESS);
1870 }
1871
1872 if (isSysProc(p) && system_procs_aging_band) {
1873 assert(p->p_memstat_effectivepriority == system_procs_aging_band);
1874 assert(p->p_memstat_idledeadline);
1875 present_in_sysprocs_aging_bucket = TRUE;
1876 } else if (isApp(p) && applications_aging_band) {
1877 assert(p->p_memstat_effectivepriority == applications_aging_band);
1878 assert(p->p_memstat_idledeadline);
1879 present_in_apps_aging_bucket = TRUE;
1880 }
1881 }
1882
1883 MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for pid %d (clear_state %d, demotions %d).\n",
1884 p->p_pid, clear_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
1885
1886
1887 if (clear_state) {
1888 p->p_memstat_idledeadline = 0;
1889 p->p_memstat_dirty &= ~P_DIRTY_AGING_IN_PROGRESS;
1890 }
1891
1892 if (isSysProc(p) && present_in_sysprocs_aging_bucket == TRUE) {
1893 memorystatus_scheduled_idle_demotions_sysprocs--;
1894 assert(memorystatus_scheduled_idle_demotions_sysprocs >= 0);
1895 } else if (isApp(p) && present_in_apps_aging_bucket == TRUE) {
1896 memorystatus_scheduled_idle_demotions_apps--;
1897 assert(memorystatus_scheduled_idle_demotions_apps >= 0);
1898 }
1899
1900 assert((memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps) >= 0);
1901 }
1902
1903 static void
1904 memorystatus_reschedule_idle_demotion_locked(void)
1905 {
1906 if (0 == (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps)) {
1907 if (memstat_idle_demotion_deadline) {
1908 /* Transitioned 1->0, so cancel next call */
1909 thread_call_cancel(memorystatus_idle_demotion_call);
1910 memstat_idle_demotion_deadline = 0;
1911 }
1912 } else {
1913 memstat_bucket_t *demotion_bucket;
1914 proc_t p = NULL, p1 = NULL, p2 = NULL;
1915
1916 if (system_procs_aging_band) {
1917 demotion_bucket = &memstat_bucket[system_procs_aging_band];
1918 p1 = TAILQ_FIRST(&demotion_bucket->list);
1919
1920 p = p1;
1921 }
1922
1923 if (applications_aging_band) {
1924 demotion_bucket = &memstat_bucket[applications_aging_band];
1925 p2 = TAILQ_FIRST(&demotion_bucket->list);
1926
1927 if (p1 && p2) {
1928 p = (p1->p_memstat_idledeadline > p2->p_memstat_idledeadline) ? p2 : p1;
1929 } else {
1930 p = (p1 == NULL) ? p2 : p1;
1931 }
1932 }
1933
1934 assert(p);
1935
1936 if (p != NULL) {
1937 assert(p && p->p_memstat_idledeadline);
1938 if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline) {
1939 thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
1940 memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
1941 }
1942 }
1943 }
1944 }
1945
1946 /*
1947 * List manipulation
1948 */
1949
1950 int
1951 memorystatus_add(proc_t p, boolean_t locked)
1952 {
1953 memstat_bucket_t *bucket;
1954
1955 MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding pid %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
1956
1957 if (!locked) {
1958 proc_list_lock();
1959 }
1960
1961 DTRACE_MEMORYSTATUS2(memorystatus_add, proc_t, p, int32_t, p->p_memstat_effectivepriority);
1962
1963 /* Processes marked internal do not have priority tracked */
1964 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
1965 goto exit;
1966 }
1967
1968 /*
1969 * Opt out system processes from being frozen by default.
1970 * For coalition-based freezing, we only want to freeze sysprocs that have specifically opted in.
1971 */
1972 if (isSysProc(p)) {
1973 p->p_memstat_state |= P_MEMSTAT_FREEZE_DISABLED;
1974 }
1975
1976 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
1977
1978 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
1979 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs - 1);
1980 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
1981 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps - 1);
1982 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
1983 /*
1984 * Entering the idle band.
1985 * Record idle start time.
1986 */
1987 p->p_memstat_idle_start = mach_absolute_time();
1988 }
1989
1990 TAILQ_INSERT_TAIL(&bucket->list, p, p_memstat_list);
1991 bucket->count++;
1992 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
1993 bucket->relaunch_high_count++;
1994 }
1995
1996 memorystatus_list_count++;
1997
1998 memorystatus_check_levels_locked();
1999
2000 exit:
2001 if (!locked) {
2002 proc_list_unlock();
2003 }
2004
2005 return 0;
2006 }
2007
2008 /*
2009 * Description:
2010 * Moves a process from one jetsam bucket to another.
2011 * which changes the LRU position of the process.
2012 *
2013 * Monitors transition between buckets and if necessary
2014 * will update cached memory limits accordingly.
2015 *
2016 * skip_demotion_check:
2017 * - if the 'jetsam aging policy' is NOT 'legacy':
2018 * When this flag is TRUE, it means we are going
2019 * to age the ripe processes out of the aging bands and into the
2020 * IDLE band and apply their inactive memory limits.
2021 *
2022 * - if the 'jetsam aging policy' is 'legacy':
2023 * When this flag is TRUE, it might mean the above aging mechanism
2024 * OR
2025 * It might be that we have a process that has used up its 'idle deferral'
2026 * stay that is given to it once per lifetime. And in this case, the process
2027 * won't be going through any aging codepaths. But we still need to apply
2028 * the right inactive limits and so we explicitly set this to TRUE if the
2029 * new priority for the process is the IDLE band.
2030 */
2031 void
2032 memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check)
2033 {
2034 memstat_bucket_t *old_bucket, *new_bucket;
2035
2036 assert(priority < MEMSTAT_BUCKET_COUNT);
2037
2038 /* Ensure that exit isn't underway, leaving the proc retained but removed from its bucket */
2039 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2040 return;
2041 }
2042
2043 MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting %s(%d) to priority %d, inserting at %s\n",
2044 (*p->p_name ? p->p_name : "unknown"), p->p_pid, priority, head_insert ? "head" : "tail");
2045
2046 DTRACE_MEMORYSTATUS3(memorystatus_update_priority, proc_t, p, int32_t, p->p_memstat_effectivepriority, int, priority);
2047
2048 old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2049
2050 if (skip_demotion_check == FALSE) {
2051 if (isSysProc(p)) {
2052 /*
2053 * For system processes, the memorystatus_dirty_* routines take care of adding/removing
2054 * the processes from the aging bands and balancing the demotion counts.
2055 * We can, however, override that if the process has an 'elevated inactive jetsam band' attribute.
2056 */
2057
2058 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2059 /*
2060 * 2 types of processes can use the non-standard elevated inactive band:
2061 * - Frozen processes that always land in memorystatus_freeze_jetsam_band
2062 * OR
2063 * - processes that specifically opt-in to the elevated inactive support e.g. docked processes.
2064 */
2065 #if CONFIG_FREEZE
2066 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
2067 if (priority <= memorystatus_freeze_jetsam_band) {
2068 priority = memorystatus_freeze_jetsam_band;
2069 }
2070 } else
2071 #endif /* CONFIG_FREEZE */
2072 {
2073 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE) {
2074 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2075 }
2076 }
2077 assert(!(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2078 }
2079 } else if (isApp(p)) {
2080 /*
2081 * Check to see if the application is being lowered in jetsam priority. If so, and:
2082 * - it has an 'elevated inactive jetsam band' attribute, then put it in the appropriate band.
2083 * - it is a normal application, then let it age in the aging band if that policy is in effect.
2084 */
2085
2086 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2087 #if CONFIG_FREEZE
2088 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
2089 if (priority <= memorystatus_freeze_jetsam_band) {
2090 priority = memorystatus_freeze_jetsam_band;
2091 }
2092 } else
2093 #endif /* CONFIG_FREEZE */
2094 {
2095 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE) {
2096 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2097 }
2098 }
2099 } else {
2100 if (applications_aging_band) {
2101 if (p->p_memstat_effectivepriority == applications_aging_band) {
2102 assert(old_bucket->count == (memorystatus_scheduled_idle_demotions_apps + 1));
2103 }
2104
2105 if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && (priority <= applications_aging_band)) {
2106 assert(!(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2107 priority = applications_aging_band;
2108 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2109 }
2110 }
2111 }
2112 }
2113 }
2114
2115 if ((system_procs_aging_band && (priority == system_procs_aging_band)) || (applications_aging_band && (priority == applications_aging_band))) {
2116 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
2117 }
2118
2119 #if DEVELOPMENT || DEBUG
2120 if (priority == JETSAM_PRIORITY_IDLE && /* if the process is on its way into the IDLE band */
2121 skip_demotion_check == FALSE && /* and it isn't via the path that will set the INACTIVE memlimits */
2122 (p->p_memstat_dirty & P_DIRTY_TRACK) && /* and it has 'DIRTY' tracking enabled */
2123 ((p->p_memstat_memlimit != p->p_memstat_memlimit_inactive) || /* and we notice that the current limit isn't the right value (inactive) */
2124 ((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) */
2125 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 */
2126 }
2127 #endif /* DEVELOPMENT || DEBUG */
2128
2129 TAILQ_REMOVE(&old_bucket->list, p, p_memstat_list);
2130 old_bucket->count--;
2131 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2132 old_bucket->relaunch_high_count--;
2133 }
2134
2135 new_bucket = &memstat_bucket[priority];
2136 if (head_insert) {
2137 TAILQ_INSERT_HEAD(&new_bucket->list, p, p_memstat_list);
2138 } else {
2139 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
2140 }
2141 new_bucket->count++;
2142 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2143 new_bucket->relaunch_high_count++;
2144 }
2145
2146 if (memorystatus_highwater_enabled) {
2147 boolean_t is_fatal;
2148 boolean_t use_active;
2149
2150 /*
2151 * If cached limit data is updated, then the limits
2152 * will be enforced by writing to the ledgers.
2153 */
2154 boolean_t ledger_update_needed = TRUE;
2155
2156 /*
2157 * Here, we must update the cached memory limit if the task
2158 * is transitioning between:
2159 * active <--> inactive
2160 * FG <--> BG
2161 * but:
2162 * dirty <--> clean is ignored
2163 *
2164 * We bypass non-idle processes that have opted into dirty tracking because
2165 * a move between buckets does not imply a transition between the
2166 * dirty <--> clean state.
2167 */
2168
2169 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
2170 if (skip_demotion_check == TRUE && priority == JETSAM_PRIORITY_IDLE) {
2171 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2172 use_active = FALSE;
2173 } else {
2174 ledger_update_needed = FALSE;
2175 }
2176 } else if ((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) {
2177 /*
2178 * inactive --> active
2179 * BG --> FG
2180 * assign active state
2181 */
2182 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2183 use_active = TRUE;
2184 } else if ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
2185 /*
2186 * active --> inactive
2187 * FG --> BG
2188 * assign inactive state
2189 */
2190 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2191 use_active = FALSE;
2192 } else {
2193 /*
2194 * The transition between jetsam priority buckets apparently did
2195 * not affect active/inactive state.
2196 * This is not unusual... especially during startup when
2197 * processes are getting established in their respective bands.
2198 */
2199 ledger_update_needed = FALSE;
2200 }
2201
2202 /*
2203 * Enforce the new limits by writing to the ledger
2204 */
2205 if (ledger_update_needed) {
2206 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
2207
2208 MEMORYSTATUS_DEBUG(3, "memorystatus_update_priority_locked: new limit on pid %d (%dMB %s) priority old --> new (%d --> %d) dirty?=0x%x %s\n",
2209 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2210 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, priority, p->p_memstat_dirty,
2211 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2212 }
2213 }
2214
2215 /*
2216 * Record idle start or idle delta.
2217 */
2218 if (p->p_memstat_effectivepriority == priority) {
2219 /*
2220 * This process is not transitioning between
2221 * jetsam priority buckets. Do nothing.
2222 */
2223 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2224 uint64_t now;
2225 /*
2226 * Transitioning out of the idle priority bucket.
2227 * Record idle delta.
2228 */
2229 assert(p->p_memstat_idle_start != 0);
2230 now = mach_absolute_time();
2231 if (now > p->p_memstat_idle_start) {
2232 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2233 }
2234
2235 /*
2236 * About to become active and so memory footprint could change.
2237 * So mark it eligible for freeze-considerations next time around.
2238 */
2239 if (p->p_memstat_state & P_MEMSTAT_FREEZE_IGNORE) {
2240 p->p_memstat_state &= ~P_MEMSTAT_FREEZE_IGNORE;
2241 }
2242 } else if (priority == JETSAM_PRIORITY_IDLE) {
2243 /*
2244 * Transitioning into the idle priority bucket.
2245 * Record idle start.
2246 */
2247 p->p_memstat_idle_start = mach_absolute_time();
2248 }
2249
2250 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CHANGE_PRIORITY), p->p_pid, priority, p->p_memstat_effectivepriority, 0, 0);
2251
2252 p->p_memstat_effectivepriority = priority;
2253
2254 #if CONFIG_SECLUDED_MEMORY
2255 if (secluded_for_apps &&
2256 task_could_use_secluded_mem(p->task)) {
2257 task_set_can_use_secluded_mem(
2258 p->task,
2259 (priority >= JETSAM_PRIORITY_FOREGROUND));
2260 }
2261 #endif /* CONFIG_SECLUDED_MEMORY */
2262
2263 memorystatus_check_levels_locked();
2264 }
2265
2266 int
2267 memorystatus_relaunch_flags_update(proc_t p, int relaunch_flags)
2268 {
2269 p->p_memstat_relaunch_flags = relaunch_flags;
2270 KDBG(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_RELAUNCH_FLAGS), p->p_pid, relaunch_flags, 0, 0, 0);
2271 return 0;
2272 }
2273
2274 /*
2275 *
2276 * Description: Update the jetsam priority and memory limit attributes for a given process.
2277 *
2278 * Parameters:
2279 * p init this process's jetsam information.
2280 * priority The jetsam priority band
2281 * user_data user specific data, unused by the kernel
2282 * is_assertion When true, a priority update is driven by an assertion.
2283 * effective guards against race if process's update already occurred
2284 * update_memlimit When true we know this is the init step via the posix_spawn path.
2285 *
2286 * memlimit_active Value in megabytes; The monitored footprint level while the
2287 * process is active. Exceeding it may result in termination
2288 * based on it's associated fatal flag.
2289 *
2290 * memlimit_active_is_fatal When a process is active and exceeds its memory footprint,
2291 * this describes whether or not it should be immediately fatal.
2292 *
2293 * memlimit_inactive Value in megabytes; The monitored footprint level while the
2294 * process is inactive. Exceeding it may result in termination
2295 * based on it's associated fatal flag.
2296 *
2297 * memlimit_inactive_is_fatal When a process is inactive and exceeds its memory footprint,
2298 * this describes whether or not it should be immediatly fatal.
2299 *
2300 * Returns: 0 Success
2301 * non-0 Failure
2302 */
2303
2304 int
2305 memorystatus_update(proc_t p, int priority, uint64_t user_data, boolean_t is_assertion, boolean_t effective, boolean_t update_memlimit,
2306 int32_t memlimit_active, boolean_t memlimit_active_is_fatal,
2307 int32_t memlimit_inactive, boolean_t memlimit_inactive_is_fatal)
2308 {
2309 int ret;
2310 boolean_t head_insert = false;
2311
2312 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);
2313
2314 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_START, p->p_pid, priority, user_data, effective, 0);
2315
2316 if (priority == -1) {
2317 /* Use as shorthand for default priority */
2318 priority = JETSAM_PRIORITY_DEFAULT;
2319 } else if ((priority == system_procs_aging_band) || (priority == applications_aging_band)) {
2320 /* Both the aging bands are reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
2321 priority = JETSAM_PRIORITY_IDLE;
2322 } else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
2323 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
2324 priority = JETSAM_PRIORITY_IDLE;
2325 head_insert = TRUE;
2326 } else if ((priority < 0) || (priority >= MEMSTAT_BUCKET_COUNT)) {
2327 /* Sanity check */
2328 ret = EINVAL;
2329 goto out;
2330 }
2331
2332 proc_list_lock();
2333
2334 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2335
2336 if (effective && (p->p_memstat_state & P_MEMSTAT_PRIORITYUPDATED)) {
2337 ret = EALREADY;
2338 proc_list_unlock();
2339 MEMORYSTATUS_DEBUG(1, "memorystatus_update: effective change specified for pid %d, but change already occurred.\n", p->p_pid);
2340 goto out;
2341 }
2342
2343 if ((p->p_memstat_state & P_MEMSTAT_TERMINATED) || ((p->p_listflag & P_LIST_EXITED) != 0)) {
2344 /*
2345 * This could happen when a process calling posix_spawn() is exiting on the jetsam thread.
2346 */
2347 ret = EBUSY;
2348 proc_list_unlock();
2349 goto out;
2350 }
2351
2352 p->p_memstat_state |= P_MEMSTAT_PRIORITYUPDATED;
2353 p->p_memstat_userdata = user_data;
2354
2355 if (is_assertion) {
2356 if (priority == JETSAM_PRIORITY_IDLE) {
2357 /*
2358 * Assertions relinquish control when the process is heading to IDLE.
2359 */
2360 if (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION) {
2361 /*
2362 * Mark the process as no longer being managed by assertions.
2363 */
2364 p->p_memstat_state &= ~P_MEMSTAT_PRIORITY_ASSERTION;
2365 } else {
2366 /*
2367 * Ignore an idle priority transition if the process is not
2368 * already managed by assertions. We won't treat this as
2369 * an error, but we will log the unexpected behavior and bail.
2370 */
2371 os_log(OS_LOG_DEFAULT, "memorystatus: Ignore assertion driven idle priority. Process not previously controlled %s:%d\n",
2372 (*p->p_name ? p->p_name : "unknown"), p->p_pid);
2373
2374 ret = 0;
2375 proc_list_unlock();
2376 goto out;
2377 }
2378 } else {
2379 /*
2380 * Process is now being managed by assertions,
2381 */
2382 p->p_memstat_state |= P_MEMSTAT_PRIORITY_ASSERTION;
2383 }
2384
2385 /* Always update the assertion priority in this path */
2386
2387 p->p_memstat_assertionpriority = priority;
2388
2389 int memstat_dirty_flags = memorystatus_dirty_get(p, TRUE); /* proc_list_lock is held */
2390
2391 if (memstat_dirty_flags != 0) {
2392 /*
2393 * Calculate maximum priority only when dirty tracking processes are involved.
2394 */
2395 int maxpriority;
2396 if (memstat_dirty_flags & PROC_DIRTY_IS_DIRTY) {
2397 maxpriority = MAX(p->p_memstat_assertionpriority, p->p_memstat_requestedpriority);
2398 } else {
2399 /* clean */
2400
2401 if (memstat_dirty_flags & PROC_DIRTY_ALLOWS_IDLE_EXIT) {
2402 /*
2403 * The aging policy must be evaluated and applied here because runnningboardd
2404 * has relinquished its hold on the jetsam priority by attempting to move a
2405 * clean process to the idle band.
2406 */
2407
2408 int newpriority = JETSAM_PRIORITY_IDLE;
2409 if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
2410 newpriority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
2411 }
2412
2413 maxpriority = MAX(p->p_memstat_assertionpriority, newpriority );
2414
2415 if (newpriority == system_procs_aging_band) {
2416 memorystatus_schedule_idle_demotion_locked(p, FALSE);
2417 }
2418 } else {
2419 /*
2420 * Preserves requestedpriority when the process does not support pressured exit.
2421 */
2422 maxpriority = MAX(p->p_memstat_assertionpriority, p->p_memstat_requestedpriority);
2423 }
2424 }
2425 priority = maxpriority;
2426 }
2427 } else {
2428 p->p_memstat_requestedpriority = priority;
2429 }
2430
2431 if (update_memlimit) {
2432 boolean_t is_fatal;
2433 boolean_t use_active;
2434
2435 /*
2436 * Posix_spawn'd processes come through this path to instantiate ledger limits.
2437 * Forked processes do not come through this path, so no ledger limits exist.
2438 * (That's why forked processes can consume unlimited memory.)
2439 */
2440
2441 MEMORYSTATUS_DEBUG(3, "memorystatus_update(enter): pid %d, priority %d, dirty=0x%x, Active(%dMB %s), Inactive(%dMB, %s)\n",
2442 p->p_pid, priority, p->p_memstat_dirty,
2443 memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
2444 memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
2445
2446 if (memlimit_active <= 0) {
2447 /*
2448 * This process will have a system_wide task limit when active.
2449 * System_wide task limit is always fatal.
2450 * It's quite common to see non-fatal flag passed in here.
2451 * It's not an error, we just ignore it.
2452 */
2453
2454 /*
2455 * For backward compatibility with some unexplained launchd behavior,
2456 * we allow a zero sized limit. But we still enforce system_wide limit
2457 * when written to the ledgers.
2458 */
2459
2460 if (memlimit_active < 0) {
2461 memlimit_active = -1; /* enforces system_wide task limit */
2462 }
2463 memlimit_active_is_fatal = TRUE;
2464 }
2465
2466 if (memlimit_inactive <= 0) {
2467 /*
2468 * This process will have a system_wide task limit when inactive.
2469 * System_wide task limit is always fatal.
2470 */
2471
2472 memlimit_inactive = -1;
2473 memlimit_inactive_is_fatal = TRUE;
2474 }
2475
2476 /*
2477 * Initialize the active limit variants for this process.
2478 */
2479 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
2480
2481 /*
2482 * Initialize the inactive limit variants for this process.
2483 */
2484 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
2485
2486 /*
2487 * Initialize the cached limits for target process.
2488 * When the target process is dirty tracked, it's typically
2489 * in a clean state. Non dirty tracked processes are
2490 * typically active (Foreground or above).
2491 * But just in case, we don't make assumptions...
2492 */
2493
2494 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
2495 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2496 use_active = TRUE;
2497 } else {
2498 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2499 use_active = FALSE;
2500 }
2501
2502 /*
2503 * Enforce the cached limit by writing to the ledger.
2504 */
2505 if (memorystatus_highwater_enabled) {
2506 /* apply now */
2507 task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, use_active, is_fatal);
2508
2509 MEMORYSTATUS_DEBUG(3, "memorystatus_update: init: limit on pid %d (%dMB %s) targeting priority(%d) dirty?=0x%x %s\n",
2510 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2511 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), priority, p->p_memstat_dirty,
2512 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2513 }
2514 }
2515
2516 /*
2517 * We can't add to the aging bands buckets here.
2518 * But, we could be removing it from those buckets.
2519 * Check and take appropriate steps if so.
2520 */
2521
2522 if (isProcessInAgingBands(p)) {
2523 if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && isApp(p) && (priority > applications_aging_band)) {
2524 /*
2525 * Runningboardd is pulling up an application that is in the aging band.
2526 * We reset the app's state here so that it'll get a fresh stay in the
2527 * aging band on the way back.
2528 *
2529 * We always handled the app 'aging' in the memorystatus_update_priority_locked()
2530 * function. Daemons used to be handled via the dirty 'set/clear/track' path.
2531 * But with extensions (daemon-app hybrid), runningboardd is now going through
2532 * this routine for daemons too and things have gotten a bit tangled. This should
2533 * be simplified/untangled at some point and might require some assistance from
2534 * runningboardd.
2535 */
2536 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2537 } else {
2538 memorystatus_invalidate_idle_demotion_locked(p, FALSE);
2539 }
2540 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2541 } else {
2542 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy && priority == JETSAM_PRIORITY_IDLE) {
2543 /*
2544 * Daemons with 'inactive' limits will go through the dirty tracking codepath.
2545 * This path deals with apps that may have 'inactive' limits e.g. WebContent processes.
2546 * If this is the legacy aging policy we explicitly need to apply those limits. If it
2547 * is any other aging policy, then we don't need to worry because all processes
2548 * will go through the aging bands and then the demotion thread will take care to
2549 * move them into the IDLE band and apply the required limits.
2550 */
2551 memorystatus_update_priority_locked(p, priority, head_insert, TRUE);
2552 }
2553 }
2554
2555 memorystatus_update_priority_locked(p, priority, head_insert, FALSE);
2556
2557 proc_list_unlock();
2558 ret = 0;
2559
2560 out:
2561 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_END, ret, 0, 0, 0, 0);
2562
2563 return ret;
2564 }
2565
2566 int
2567 memorystatus_remove(proc_t p)
2568 {
2569 int ret;
2570 memstat_bucket_t *bucket;
2571 boolean_t reschedule = FALSE;
2572
2573 MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing pid %d\n", p->p_pid);
2574
2575 /*
2576 * Check if this proc is locked (because we're performing a freeze).
2577 * If so, we fail and instruct the caller to try again later.
2578 */
2579 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
2580 return EAGAIN;
2581 }
2582
2583 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2584
2585 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2586
2587 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
2588 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs);
2589 reschedule = TRUE;
2590 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
2591 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps);
2592 reschedule = TRUE;
2593 }
2594
2595 /*
2596 * Record idle delta
2597 */
2598
2599 if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2600 uint64_t now = mach_absolute_time();
2601 if (now > p->p_memstat_idle_start) {
2602 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2603 }
2604 }
2605
2606 TAILQ_REMOVE(&bucket->list, p, p_memstat_list);
2607 bucket->count--;
2608 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
2609 bucket->relaunch_high_count--;
2610 }
2611
2612 memorystatus_list_count--;
2613
2614 /* If awaiting demotion to the idle band, clean up */
2615 if (reschedule) {
2616 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2617 memorystatus_reschedule_idle_demotion_locked();
2618 }
2619
2620 memorystatus_check_levels_locked();
2621
2622 #if CONFIG_FREEZE
2623 if (p->p_memstat_state & (P_MEMSTAT_FROZEN)) {
2624 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
2625 p->p_memstat_state &= ~P_MEMSTAT_REFREEZE_ELIGIBLE;
2626 memorystatus_refreeze_eligible_count--;
2627 }
2628
2629 memorystatus_frozen_count--;
2630 memorystatus_frozen_shared_mb -= p->p_memstat_freeze_sharedanon_pages;
2631 p->p_memstat_freeze_sharedanon_pages = 0;
2632 }
2633
2634 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
2635 memorystatus_suspended_count--;
2636 }
2637 #endif
2638
2639 if (p) {
2640 ret = 0;
2641 } else {
2642 ret = ESRCH;
2643 }
2644
2645 return ret;
2646 }
2647
2648 /*
2649 * Validate dirty tracking flags with process state.
2650 *
2651 * Return:
2652 * 0 on success
2653 * non-0 on failure
2654 *
2655 * The proc_list_lock is held by the caller.
2656 */
2657
2658 static int
2659 memorystatus_validate_track_flags(struct proc *target_p, uint32_t pcontrol)
2660 {
2661 /* See that the process isn't marked for termination */
2662 if (target_p->p_memstat_dirty & P_DIRTY_TERMINATED) {
2663 return EBUSY;
2664 }
2665
2666 /* Idle exit requires that process be tracked */
2667 if ((pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) &&
2668 !(pcontrol & PROC_DIRTY_TRACK)) {
2669 return EINVAL;
2670 }
2671
2672 /* 'Launch in progress' tracking requires that process have enabled dirty tracking too. */
2673 if ((pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) &&
2674 !(pcontrol & PROC_DIRTY_TRACK)) {
2675 return EINVAL;
2676 }
2677
2678 /* Only one type of DEFER behavior is allowed.*/
2679 if ((pcontrol & PROC_DIRTY_DEFER) &&
2680 (pcontrol & PROC_DIRTY_DEFER_ALWAYS)) {
2681 return EINVAL;
2682 }
2683
2684 /* Deferral is only relevant if idle exit is specified */
2685 if (((pcontrol & PROC_DIRTY_DEFER) ||
2686 (pcontrol & PROC_DIRTY_DEFER_ALWAYS)) &&
2687 !(pcontrol & PROC_DIRTY_ALLOWS_IDLE_EXIT)) {
2688 return EINVAL;
2689 }
2690
2691 return 0;
2692 }
2693
2694 static void
2695 memorystatus_update_idle_priority_locked(proc_t p)
2696 {
2697 int32_t priority;
2698
2699 MEMORYSTATUS_DEBUG(1, "memorystatus_update_idle_priority_locked(): pid %d dirty 0x%X\n", p->p_pid, p->p_memstat_dirty);
2700
2701 assert(isSysProc(p));
2702
2703 if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
2704 priority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
2705 } else {
2706 priority = p->p_memstat_requestedpriority;
2707 }
2708
2709 if (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION) {
2710 /*
2711 * This process has a jetsam priority managed by an assertion.
2712 * Policy is to choose the max priority.
2713 */
2714 if (p->p_memstat_assertionpriority > priority) {
2715 os_log(OS_LOG_DEFAULT, "memorystatus: assertion priority %d overrides priority %d for %s:%d\n",
2716 p->p_memstat_assertionpriority, priority,
2717 (*p->p_name ? p->p_name : "unknown"), p->p_pid);
2718 priority = p->p_memstat_assertionpriority;
2719 }
2720 }
2721
2722 if (priority != p->p_memstat_effectivepriority) {
2723 if ((jetsam_aging_policy == kJetsamAgingPolicyLegacy) &&
2724 (priority == JETSAM_PRIORITY_IDLE)) {
2725 /*
2726 * This process is on its way into the IDLE band. The system is
2727 * using 'legacy' jetsam aging policy. That means, this process
2728 * has already used up its idle-deferral aging time that is given
2729 * once per its lifetime. So we need to set the INACTIVE limits
2730 * explicitly because it won't be going through the demotion paths
2731 * that take care to apply the limits appropriately.
2732 */
2733
2734 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2735 /*
2736 * This process has the 'elevated inactive jetsam band' attribute.
2737 * So, there will be no trip to IDLE after all.
2738 * Instead, we pin the process in the elevated band,
2739 * where its ACTIVE limits will apply.
2740 */
2741
2742 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2743 }
2744
2745 memorystatus_update_priority_locked(p, priority, false, true);
2746 } else {
2747 memorystatus_update_priority_locked(p, priority, false, false);
2748 }
2749 }
2750 }
2751
2752 /*
2753 * Processes can opt to have their state tracked by the kernel, indicating when they are busy (dirty) or idle
2754 * (clean). They may also indicate that they support termination when idle, with the result that they are promoted
2755 * to their desired, higher, jetsam priority when dirty (and are therefore killed later), and demoted to the low
2756 * priority idle band when clean (and killed earlier, protecting higher priority procesess).
2757 *
2758 * If the deferral flag is set, then newly tracked processes will be protected for an initial period (as determined by
2759 * memorystatus_sysprocs_idle_delay_time); if they go clean during this time, then they will be moved to a deferred-idle band
2760 * with a slightly higher priority, guarding against immediate termination under memory pressure and being unable to
2761 * make forward progress. Finally, when the guard expires, they will be moved to the standard, lowest-priority, idle
2762 * band. The deferral can be cleared early by clearing the appropriate flag.
2763 *
2764 * The deferral timer is active only for the duration that the process is marked as guarded and clean; if the process
2765 * is marked dirty, the timer will be cancelled. Upon being subsequently marked clean, the deferment will either be
2766 * re-enabled or the guard state cleared, depending on whether the guard deadline has passed.
2767 */
2768
2769 int
2770 memorystatus_dirty_track(proc_t p, uint32_t pcontrol)
2771 {
2772 unsigned int old_dirty;
2773 boolean_t reschedule = FALSE;
2774 boolean_t already_deferred = FALSE;
2775 boolean_t defer_now = FALSE;
2776 int ret = 0;
2777
2778 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_TRACK),
2779 p->p_pid, p->p_memstat_dirty, pcontrol, 0, 0);
2780
2781 proc_list_lock();
2782
2783 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2784 /*
2785 * Process is on its way out.
2786 */
2787 ret = EBUSY;
2788 goto exit;
2789 }
2790
2791 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2792 ret = EPERM;
2793 goto exit;
2794 }
2795
2796 if ((ret = memorystatus_validate_track_flags(p, pcontrol)) != 0) {
2797 /* error */
2798 goto exit;
2799 }
2800
2801 old_dirty = p->p_memstat_dirty;
2802
2803 /* These bits are cumulative, as per <rdar://problem/11159924> */
2804 if (pcontrol & PROC_DIRTY_TRACK) {
2805 p->p_memstat_dirty |= P_DIRTY_TRACK;
2806 }
2807
2808 if (pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) {
2809 p->p_memstat_dirty |= P_DIRTY_ALLOW_IDLE_EXIT;
2810 }
2811
2812 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
2813 p->p_memstat_dirty |= P_DIRTY_LAUNCH_IN_PROGRESS;
2814 }
2815
2816 if (old_dirty & P_DIRTY_AGING_IN_PROGRESS) {
2817 already_deferred = TRUE;
2818 }
2819
2820
2821 /* This can be set and cleared exactly once. */
2822 if (pcontrol & (PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) {
2823 if ((pcontrol & (PROC_DIRTY_DEFER)) &&
2824 !(old_dirty & P_DIRTY_DEFER)) {
2825 p->p_memstat_dirty |= P_DIRTY_DEFER;
2826 }
2827
2828 if ((pcontrol & (PROC_DIRTY_DEFER_ALWAYS)) &&
2829 !(old_dirty & P_DIRTY_DEFER_ALWAYS)) {
2830 p->p_memstat_dirty |= P_DIRTY_DEFER_ALWAYS;
2831 }
2832
2833 defer_now = TRUE;
2834 }
2835
2836 MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for pid %d\n",
2837 ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) ? "Y" : "N",
2838 defer_now ? "Y" : "N",
2839 p->p_memstat_dirty & P_DIRTY ? "Y" : "N",
2840 p->p_pid);
2841
2842 /* Kick off or invalidate the idle exit deferment if there's a state transition. */
2843 if (!(p->p_memstat_dirty & P_DIRTY_IS_DIRTY)) {
2844 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
2845 if (defer_now && !already_deferred) {
2846 /*
2847 * Request to defer a clean process that's idle-exit enabled
2848 * and not already in the jetsam deferred band. Most likely a
2849 * new launch.
2850 */
2851 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2852 reschedule = TRUE;
2853 } else if (!defer_now) {
2854 /*
2855 * The process isn't asking for the 'aging' facility.
2856 * Could be that it is:
2857 */
2858
2859 if (already_deferred) {
2860 /*
2861 * already in the aging bands. Traditionally,
2862 * some processes have tried to use this to
2863 * opt out of the 'aging' facility.
2864 */
2865
2866 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2867 } else {
2868 /*
2869 * agnostic to the 'aging' facility. In that case,
2870 * we'll go ahead and opt it in because this is likely
2871 * a new launch (clean process, dirty tracking enabled)
2872 */
2873
2874 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2875 }
2876
2877 reschedule = TRUE;
2878 }
2879 }
2880 } else {
2881 /*
2882 * We are trying to operate on a dirty process. Dirty processes have to
2883 * be removed from the deferred band. The question is do we reset the
2884 * deferred state or not?
2885 *
2886 * This could be a legal request like:
2887 * - this process had opted into the 'aging' band
2888 * - but it's now dirty and requests to opt out.
2889 * In this case, we remove the process from the band and reset its
2890 * state too. It'll opt back in properly when needed.
2891 *
2892 * OR, this request could be a user-space bug. E.g.:
2893 * - this process had opted into the 'aging' band when clean
2894 * - and, then issues another request to again put it into the band except
2895 * this time the process is dirty.
2896 * The process going dirty, as a transition in memorystatus_dirty_set(), will pull the process out of
2897 * the deferred band with its state intact. So our request below is no-op.
2898 * But we do it here anyways for coverage.
2899 *
2900 * memorystatus_update_idle_priority_locked()
2901 * single-mindedly treats a dirty process as "cannot be in the aging band".
2902 */
2903
2904 if (!defer_now && already_deferred) {
2905 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2906 reschedule = TRUE;
2907 } else {
2908 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
2909
2910 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
2911 reschedule = TRUE;
2912 }
2913 }
2914
2915 memorystatus_update_idle_priority_locked(p);
2916
2917 if (reschedule) {
2918 memorystatus_reschedule_idle_demotion_locked();
2919 }
2920
2921 ret = 0;
2922
2923 exit:
2924 proc_list_unlock();
2925
2926 return ret;
2927 }
2928
2929 int
2930 memorystatus_dirty_set(proc_t p, boolean_t self, uint32_t pcontrol)
2931 {
2932 int ret;
2933 boolean_t kill = false;
2934 boolean_t reschedule = FALSE;
2935 boolean_t was_dirty = FALSE;
2936 boolean_t now_dirty = FALSE;
2937
2938 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_set(): %d %d 0x%x 0x%x\n", self, p->p_pid, pcontrol, p->p_memstat_dirty);
2939 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_SET), p->p_pid, self, pcontrol, 0, 0);
2940
2941 proc_list_lock();
2942
2943 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2944 /*
2945 * Process is on its way out.
2946 */
2947 ret = EBUSY;
2948 goto exit;
2949 }
2950
2951 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2952 ret = EPERM;
2953 goto exit;
2954 }
2955
2956 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
2957 was_dirty = TRUE;
2958 }
2959
2960 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
2961 /* Dirty tracking not enabled */
2962 ret = EINVAL;
2963 } else if (pcontrol && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
2964 /*
2965 * Process is set to be terminated and we're attempting to mark it dirty.
2966 * Set for termination and marking as clean is OK - see <rdar://problem/10594349>.
2967 */
2968 ret = EBUSY;
2969 } else {
2970 int flag = (self == TRUE) ? P_DIRTY : P_DIRTY_SHUTDOWN;
2971 if (pcontrol && !(p->p_memstat_dirty & flag)) {
2972 /* Mark the process as having been dirtied at some point */
2973 p->p_memstat_dirty |= (flag | P_DIRTY_MARKED);
2974 memorystatus_dirty_count++;
2975 ret = 0;
2976 } else if ((pcontrol == 0) && (p->p_memstat_dirty & flag)) {
2977 if ((flag == P_DIRTY_SHUTDOWN) && (!(p->p_memstat_dirty & P_DIRTY))) {
2978 /* Clearing the dirty shutdown flag, and the process is otherwise clean - kill */
2979 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
2980 kill = true;
2981 } else if ((flag == P_DIRTY) && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
2982 /* Kill previously terminated processes if set clean */
2983 kill = true;
2984 }
2985 p->p_memstat_dirty &= ~flag;
2986 memorystatus_dirty_count--;
2987 ret = 0;
2988 } else {
2989 /* Already set */
2990 ret = EALREADY;
2991 }
2992 }
2993
2994 if (ret != 0) {
2995 goto exit;
2996 }
2997
2998 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
2999 now_dirty = TRUE;
3000 }
3001
3002 if ((was_dirty == TRUE && now_dirty == FALSE) ||
3003 (was_dirty == FALSE && now_dirty == TRUE)) {
3004 /* Manage idle exit deferral, if applied */
3005 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3006 /*
3007 * Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band OR it might be heading back
3008 * there once it's clean again. For the legacy case, this only applies if it has some protection window left.
3009 * P_DIRTY_DEFER: one-time protection window given at launch
3010 * P_DIRTY_DEFER_ALWAYS: protection window given for every dirty->clean transition. Like non-legacy mode.
3011 *
3012 * Non-Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band. It will always stop over
3013 * in that band on it's way to IDLE.
3014 */
3015
3016 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3017 /*
3018 * New dirty process i.e. "was_dirty == FALSE && now_dirty == TRUE"
3019 *
3020 * The process will move from its aging band to its higher requested
3021 * jetsam band.
3022 */
3023 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
3024
3025 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
3026 reschedule = TRUE;
3027 } else {
3028 /*
3029 * Process is back from "dirty" to "clean".
3030 */
3031
3032 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
3033 if (((p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) == FALSE) &&
3034 (mach_absolute_time() >= p->p_memstat_idledeadline)) {
3035 /*
3036 * The process' hasn't enrolled in the "always defer after dirty"
3037 * mode and its deadline has expired. It currently
3038 * does not reside in any of the aging buckets.
3039 *
3040 * It's on its way to the JETSAM_PRIORITY_IDLE
3041 * bucket via memorystatus_update_idle_priority_locked()
3042 * below.
3043 *
3044 * So all we need to do is reset all the state on the
3045 * process that's related to the aging bucket i.e.
3046 * the AGING_IN_PROGRESS flag and the timer deadline.
3047 */
3048
3049 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3050 reschedule = TRUE;
3051 } else {
3052 /*
3053 * Process enrolled in "always stop in deferral band after dirty" OR
3054 * it still has some protection window left and so
3055 * we just re-arm the timer without modifying any
3056 * state on the process iff it still wants into that band.
3057 */
3058
3059 if (p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) {
3060 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3061 reschedule = TRUE;
3062 } else if (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) {
3063 memorystatus_schedule_idle_demotion_locked(p, FALSE);
3064 reschedule = TRUE;
3065 }
3066 }
3067 } else {
3068 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3069 reschedule = TRUE;
3070 }
3071 }
3072 }
3073
3074 memorystatus_update_idle_priority_locked(p);
3075
3076 if (memorystatus_highwater_enabled) {
3077 boolean_t ledger_update_needed = TRUE;
3078 boolean_t use_active;
3079 boolean_t is_fatal;
3080 /*
3081 * We are in this path because this process transitioned between
3082 * dirty <--> clean state. Update the cached memory limits.
3083 */
3084
3085 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
3086 /*
3087 * process is pinned in elevated band
3088 * or
3089 * process is dirty
3090 */
3091 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
3092 use_active = TRUE;
3093 ledger_update_needed = TRUE;
3094 } else {
3095 /*
3096 * process is clean...but if it has opted into pressured-exit
3097 * we don't apply the INACTIVE limit till the process has aged
3098 * out and is entering the IDLE band.
3099 * See memorystatus_update_priority_locked() for that.
3100 */
3101
3102 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3103 ledger_update_needed = FALSE;
3104 } else {
3105 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
3106 use_active = FALSE;
3107 ledger_update_needed = TRUE;
3108 }
3109 }
3110
3111 /*
3112 * Enforce the new limits by writing to the ledger.
3113 *
3114 * This is a hot path and holding the proc_list_lock while writing to the ledgers,
3115 * (where the task lock is taken) is bad. So, we temporarily drop the proc_list_lock.
3116 * We aren't traversing the jetsam bucket list here, so we should be safe.
3117 * See rdar://21394491.
3118 */
3119
3120 if (ledger_update_needed && proc_ref_locked(p) == p) {
3121 int ledger_limit;
3122 if (p->p_memstat_memlimit > 0) {
3123 ledger_limit = p->p_memstat_memlimit;
3124 } else {
3125 ledger_limit = -1;
3126 }
3127 proc_list_unlock();
3128 task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, use_active, is_fatal);
3129 proc_list_lock();
3130 proc_rele_locked(p);
3131
3132 MEMORYSTATUS_DEBUG(3, "memorystatus_dirty_set: new limit on pid %d (%dMB %s) priority(%d) dirty?=0x%x %s\n",
3133 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
3134 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
3135 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
3136 }
3137 }
3138
3139 /* If the deferral state changed, reschedule the demotion timer */
3140 if (reschedule) {
3141 memorystatus_reschedule_idle_demotion_locked();
3142 }
3143 }
3144
3145 if (kill) {
3146 if (proc_ref_locked(p) == p) {
3147 proc_list_unlock();
3148 psignal(p, SIGKILL);
3149 proc_list_lock();
3150 proc_rele_locked(p);
3151 }
3152 }
3153
3154 exit:
3155 proc_list_unlock();
3156
3157 return ret;
3158 }
3159
3160 int
3161 memorystatus_dirty_clear(proc_t p, uint32_t pcontrol)
3162 {
3163 int ret = 0;
3164
3165 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_clear(): %d 0x%x 0x%x\n", p->p_pid, pcontrol, p->p_memstat_dirty);
3166
3167 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_CLEAR), p->p_pid, pcontrol, 0, 0, 0);
3168
3169 proc_list_lock();
3170
3171 if ((p->p_listflag & P_LIST_EXITED) != 0) {
3172 /*
3173 * Process is on its way out.
3174 */
3175 ret = EBUSY;
3176 goto exit;
3177 }
3178
3179 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
3180 ret = EPERM;
3181 goto exit;
3182 }
3183
3184 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
3185 /* Dirty tracking not enabled */
3186 ret = EINVAL;
3187 goto exit;
3188 }
3189
3190 if (!pcontrol || (pcontrol & (PROC_DIRTY_LAUNCH_IN_PROGRESS | PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) == 0) {
3191 ret = EINVAL;
3192 goto exit;
3193 }
3194
3195 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
3196 p->p_memstat_dirty &= ~P_DIRTY_LAUNCH_IN_PROGRESS;
3197 }
3198
3199 /* This can be set and cleared exactly once. */
3200 if (pcontrol & (PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) {
3201 if (p->p_memstat_dirty & P_DIRTY_DEFER) {
3202 p->p_memstat_dirty &= ~(P_DIRTY_DEFER);
3203 }
3204
3205 if (p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) {
3206 p->p_memstat_dirty &= ~(P_DIRTY_DEFER_ALWAYS);
3207 }
3208
3209 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3210 memorystatus_update_idle_priority_locked(p);
3211 memorystatus_reschedule_idle_demotion_locked();
3212 }
3213
3214 ret = 0;
3215 exit:
3216 proc_list_unlock();
3217
3218 return ret;
3219 }
3220
3221 int
3222 memorystatus_dirty_get(proc_t p, boolean_t locked)
3223 {
3224 int ret = 0;
3225
3226 if (!locked) {
3227 proc_list_lock();
3228 }
3229
3230 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3231 ret |= PROC_DIRTY_TRACKED;
3232 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3233 ret |= PROC_DIRTY_ALLOWS_IDLE_EXIT;
3234 }
3235 if (p->p_memstat_dirty & P_DIRTY) {
3236 ret |= PROC_DIRTY_IS_DIRTY;
3237 }
3238 if (p->p_memstat_dirty & P_DIRTY_LAUNCH_IN_PROGRESS) {
3239 ret |= PROC_DIRTY_LAUNCH_IS_IN_PROGRESS;
3240 }
3241 }
3242
3243 if (!locked) {
3244 proc_list_unlock();
3245 }
3246
3247 return ret;
3248 }
3249
3250 int
3251 memorystatus_on_terminate(proc_t p)
3252 {
3253 int sig;
3254
3255 proc_list_lock();
3256
3257 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3258
3259 if ((p->p_memstat_dirty & (P_DIRTY_TRACK | P_DIRTY_IS_DIRTY)) == P_DIRTY_TRACK) {
3260 /* Clean; mark as terminated and issue SIGKILL */
3261 sig = SIGKILL;
3262 } else {
3263 /* Dirty, terminated, or state tracking is unsupported; issue SIGTERM to allow cleanup */
3264 sig = SIGTERM;
3265 }
3266
3267 proc_list_unlock();
3268
3269 return sig;
3270 }
3271
3272 void
3273 memorystatus_on_suspend(proc_t p)
3274 {
3275 #if CONFIG_FREEZE
3276 uint32_t pages;
3277 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
3278 #endif
3279 proc_list_lock();
3280 #if CONFIG_FREEZE
3281 memorystatus_suspended_count++;
3282 #endif
3283 p->p_memstat_state |= P_MEMSTAT_SUSPENDED;
3284 proc_list_unlock();
3285 }
3286
3287 void
3288 memorystatus_on_resume(proc_t p)
3289 {
3290 #if CONFIG_FREEZE
3291 boolean_t frozen;
3292 pid_t pid;
3293 #endif
3294
3295 proc_list_lock();
3296
3297 #if CONFIG_FREEZE
3298 frozen = (p->p_memstat_state & P_MEMSTAT_FROZEN);
3299 if (frozen) {
3300 /*
3301 * Now that we don't _thaw_ a process completely,
3302 * resuming it (and having some on-demand swapins)
3303 * shouldn't preclude it from being counted as frozen.
3304 *
3305 * memorystatus_frozen_count--;
3306 *
3307 * We preserve the P_MEMSTAT_FROZEN state since the process
3308 * could have state on disk AND so will deserve some protection
3309 * in the jetsam bands.
3310 */
3311 if ((p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) == 0) {
3312 p->p_memstat_state |= P_MEMSTAT_REFREEZE_ELIGIBLE;
3313 memorystatus_refreeze_eligible_count++;
3314 }
3315 p->p_memstat_thaw_count++;
3316
3317 memorystatus_thaw_count++;
3318 }
3319
3320 memorystatus_suspended_count--;
3321
3322 pid = p->p_pid;
3323 #endif
3324
3325 /*
3326 * P_MEMSTAT_FROZEN will remain unchanged. This used to be:
3327 * p->p_memstat_state &= ~(P_MEMSTAT_SUSPENDED | P_MEMSTAT_FROZEN);
3328 */
3329 p->p_memstat_state &= ~P_MEMSTAT_SUSPENDED;
3330
3331 proc_list_unlock();
3332
3333 #if CONFIG_FREEZE
3334 if (frozen) {
3335 memorystatus_freeze_entry_t data = { pid, FALSE, 0 };
3336 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
3337 }
3338 #endif
3339 }
3340
3341 void
3342 memorystatus_on_inactivity(proc_t p)
3343 {
3344 #pragma unused(p)
3345 #if CONFIG_FREEZE
3346 /* Wake the freeze thread */
3347 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
3348 #endif
3349 }
3350
3351 /*
3352 * The proc_list_lock is held by the caller.
3353 */
3354 static uint32_t
3355 memorystatus_build_state(proc_t p)
3356 {
3357 uint32_t snapshot_state = 0;
3358
3359 /* General */
3360 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
3361 snapshot_state |= kMemorystatusSuspended;
3362 }
3363 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
3364 snapshot_state |= kMemorystatusFrozen;
3365 }
3366 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
3367 snapshot_state |= kMemorystatusWasThawed;
3368 }
3369 if (p->p_memstat_state & P_MEMSTAT_PRIORITY_ASSERTION) {
3370 snapshot_state |= kMemorystatusAssertion;
3371 }
3372
3373 /* Tracking */
3374 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3375 snapshot_state |= kMemorystatusTracked;
3376 }
3377 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3378 snapshot_state |= kMemorystatusSupportsIdleExit;
3379 }
3380 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3381 snapshot_state |= kMemorystatusDirty;
3382 }
3383
3384 return snapshot_state;
3385 }
3386
3387 static boolean_t
3388 kill_idle_exit_proc(void)
3389 {
3390 proc_t p, victim_p = PROC_NULL;
3391 uint64_t current_time, footprint_of_killed_proc;
3392 boolean_t killed = FALSE;
3393 unsigned int i = 0;
3394 os_reason_t jetsam_reason = OS_REASON_NULL;
3395
3396 /* Pick next idle exit victim. */
3397 current_time = mach_absolute_time();
3398
3399 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_IDLE_EXIT);
3400 if (jetsam_reason == OS_REASON_NULL) {
3401 printf("kill_idle_exit_proc: failed to allocate jetsam reason\n");
3402 }
3403
3404 proc_list_lock();
3405
3406 p = memorystatus_get_first_proc_locked(&i, FALSE);
3407 while (p) {
3408 /* No need to look beyond the idle band */
3409 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
3410 break;
3411 }
3412
3413 if ((p->p_memstat_dirty & (P_DIRTY_ALLOW_IDLE_EXIT | P_DIRTY_IS_DIRTY | P_DIRTY_TERMINATED)) == (P_DIRTY_ALLOW_IDLE_EXIT)) {
3414 if (current_time >= p->p_memstat_idledeadline) {
3415 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3416 victim_p = proc_ref_locked(p);
3417 break;
3418 }
3419 }
3420
3421 p = memorystatus_get_next_proc_locked(&i, p, FALSE);
3422 }
3423
3424 proc_list_unlock();
3425
3426 if (victim_p) {
3427 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);
3428 killed = memorystatus_do_kill(victim_p, kMemorystatusKilledIdleExit, jetsam_reason, &footprint_of_killed_proc);
3429 proc_rele(victim_p);
3430 } else {
3431 os_reason_free(jetsam_reason);
3432 }
3433
3434 return killed;
3435 }
3436
3437 static void
3438 memorystatus_thread_wake(void)
3439 {
3440 int thr_id = 0;
3441 int active_thr = atomic_load(&active_jetsam_threads);
3442
3443 /* Wakeup all the jetsam threads */
3444 for (thr_id = 0; thr_id < active_thr; thr_id++) {
3445 thread_wakeup((event_t)&jetsam_threads[thr_id].memorystatus_wakeup);
3446 }
3447 }
3448
3449 #if CONFIG_JETSAM
3450
3451 static void
3452 memorystatus_thread_pool_max()
3453 {
3454 /* Increase the jetsam thread pool to max_jetsam_threads */
3455 int max_threads = max_jetsam_threads;
3456 printf("Expanding memorystatus pool to %d!\n", max_threads);
3457 atomic_store(&active_jetsam_threads, max_threads);
3458 }
3459
3460 static void
3461 memorystatus_thread_pool_default()
3462 {
3463 /* Restore the jetsam thread pool to a single thread */
3464 printf("Reverting memorystatus pool back to 1\n");
3465 atomic_store(&active_jetsam_threads, 1);
3466 }
3467
3468 #endif /* CONFIG_JETSAM */
3469
3470 extern void vm_pressure_response(void);
3471
3472 static int
3473 memorystatus_thread_block(uint32_t interval_ms, thread_continue_t continuation)
3474 {
3475 struct jetsam_thread_state *jetsam_thread = jetsam_current_thread();
3476
3477 assert(jetsam_thread != NULL);
3478 if (interval_ms) {
3479 assert_wait_timeout(&jetsam_thread->memorystatus_wakeup, THREAD_UNINT, interval_ms, NSEC_PER_MSEC);
3480 } else {
3481 assert_wait(&jetsam_thread->memorystatus_wakeup, THREAD_UNINT);
3482 }
3483
3484 return thread_block(continuation);
3485 }
3486
3487 static boolean_t
3488 memorystatus_avail_pages_below_pressure(void)
3489 {
3490 #if CONFIG_EMBEDDED
3491 /*
3492 * Instead of CONFIG_EMBEDDED for these *avail_pages* routines, we should
3493 * key off of the system having dynamic swap support. With full swap support,
3494 * the system shouldn't really need to worry about various page thresholds.
3495 */
3496 return memorystatus_available_pages <= memorystatus_available_pages_pressure;
3497 #else /* CONFIG_EMBEDDED */
3498 return FALSE;
3499 #endif /* CONFIG_EMBEDDED */
3500 }
3501
3502 static boolean_t
3503 memorystatus_avail_pages_below_critical(void)
3504 {
3505 #if CONFIG_EMBEDDED
3506 return memorystatus_available_pages <= memorystatus_available_pages_critical;
3507 #else /* CONFIG_EMBEDDED */
3508 return FALSE;
3509 #endif /* CONFIG_EMBEDDED */
3510 }
3511
3512 static boolean_t
3513 memorystatus_post_snapshot(int32_t priority, uint32_t cause)
3514 {
3515 boolean_t is_idle_priority;
3516
3517 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
3518 is_idle_priority = (priority == JETSAM_PRIORITY_IDLE);
3519 } else {
3520 is_idle_priority = (priority == JETSAM_PRIORITY_IDLE || priority == JETSAM_PRIORITY_IDLE_DEFERRED);
3521 }
3522 #if CONFIG_EMBEDDED
3523 #pragma unused(cause)
3524 /*
3525 * Don't generate logs for steady-state idle-exit kills,
3526 * unless it is overridden for debug or by the device
3527 * tree.
3528 */
3529
3530 return !is_idle_priority || memorystatus_idle_snapshot;
3531
3532 #else /* CONFIG_EMBEDDED */
3533 /*
3534 * Don't generate logs for steady-state idle-exit kills,
3535 * unless
3536 * - it is overridden for debug or by the device
3537 * tree.
3538 * OR
3539 * - the kill causes are important i.e. not kMemorystatusKilledIdleExit
3540 */
3541
3542 boolean_t snapshot_eligible_kill_cause = (is_reason_thrashing(cause) || is_reason_zone_map_exhaustion(cause));
3543 return !is_idle_priority || memorystatus_idle_snapshot || snapshot_eligible_kill_cause;
3544 #endif /* CONFIG_EMBEDDED */
3545 }
3546
3547 static boolean_t
3548 memorystatus_action_needed(void)
3549 {
3550 #if CONFIG_EMBEDDED
3551 return is_reason_thrashing(kill_under_pressure_cause) ||
3552 is_reason_zone_map_exhaustion(kill_under_pressure_cause) ||
3553 memorystatus_available_pages <= memorystatus_available_pages_pressure;
3554 #else /* CONFIG_EMBEDDED */
3555 return is_reason_thrashing(kill_under_pressure_cause) ||
3556 is_reason_zone_map_exhaustion(kill_under_pressure_cause);
3557 #endif /* CONFIG_EMBEDDED */
3558 }
3559
3560 static boolean_t
3561 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)
3562 {
3563 boolean_t purged = FALSE, killed = FALSE;
3564
3565 *memory_reclaimed = 0;
3566 killed = memorystatus_kill_hiwat_proc(errors, &purged, memory_reclaimed);
3567
3568 if (killed) {
3569 *hwm_kill = *hwm_kill + 1;
3570 *post_snapshot = TRUE;
3571 return TRUE;
3572 } else {
3573 if (purged == FALSE) {
3574 /* couldn't purge and couldn't kill */
3575 memorystatus_hwm_candidates = FALSE;
3576 }
3577 }
3578
3579 #if CONFIG_JETSAM
3580 /* No highwater processes to kill. Continue or stop for now? */
3581 if (!is_reason_thrashing(kill_under_pressure_cause) &&
3582 !is_reason_zone_map_exhaustion(kill_under_pressure_cause) &&
3583 (memorystatus_available_pages > memorystatus_available_pages_critical)) {
3584 /*
3585 * We are _not_ out of pressure but we are above the critical threshold and there's:
3586 * - no compressor thrashing
3587 * - enough zone memory
3588 * - no more HWM processes left.
3589 * For now, don't kill any other processes.
3590 */
3591
3592 if (*hwm_kill == 0) {
3593 memorystatus_thread_wasted_wakeup++;
3594 }
3595
3596 *is_critical = FALSE;
3597
3598 return TRUE;
3599 }
3600 #endif /* CONFIG_JETSAM */
3601
3602 return FALSE;
3603 }
3604
3605 /*
3606 * kJetsamHighRelaunchCandidatesThreshold defines the percentage of candidates
3607 * in the idle & deferred bands that need to be bad candidates in order to trigger
3608 * aggressive jetsam.
3609 */
3610 #define kJetsamHighRelaunchCandidatesThreshold (100)
3611
3612 /* kJetsamMinCandidatesThreshold defines the minimum number of candidates in the
3613 * idle/deferred bands to trigger aggressive jetsam. This value basically decides
3614 * how much memory the system is ready to hold in the lower bands without triggering
3615 * aggressive jetsam. This number should ideally be tuned based on the memory config
3616 * of the device.
3617 */
3618 #define kJetsamMinCandidatesThreshold (5)
3619
3620 static boolean_t
3621 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)
3622 {
3623 boolean_t aggressive_jetsam_needed = false;
3624
3625 /*
3626 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, we maintain the jetsam
3627 * relaunch behavior for all daemons. Also, daemons and apps are aged in deferred bands on
3628 * every dirty->clean transition. For this aging policy, the best way to determine if
3629 * aggressive jetsam is needed, is to see if the kill candidates are mostly bad candidates.
3630 * If yes, then we need to go to higher bands to reclaim memory.
3631 */
3632 proc_list_lock();
3633 /* Get total candidate counts for idle and idle deferred bands */
3634 *total_candidates = memstat_bucket[JETSAM_PRIORITY_IDLE].count + memstat_bucket[system_procs_aging_band].count;
3635 /* Get counts of bad kill candidates in idle and idle deferred bands */
3636 int bad_candidates = memstat_bucket[JETSAM_PRIORITY_IDLE].relaunch_high_count + memstat_bucket[system_procs_aging_band].relaunch_high_count;
3637
3638 *elevated_bucket_count = memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE].count;
3639
3640 proc_list_unlock();
3641
3642 /* Check if the number of bad candidates is greater than kJetsamHighRelaunchCandidatesThreshold % */
3643 aggressive_jetsam_needed = (((bad_candidates * 100) / *total_candidates) >= kJetsamHighRelaunchCandidatesThreshold);
3644
3645 /*
3646 * Since the new aging policy bases the aggressive jetsam trigger on percentage of
3647 * bad candidates, it is prone to being overly aggressive. In order to mitigate that,
3648 * make sure the system is really under memory pressure before triggering aggressive
3649 * jetsam.
3650 */
3651 if (memorystatus_available_pages > memorystatus_sysproc_aging_aggr_pages) {
3652 aggressive_jetsam_needed = false;
3653 }
3654
3655 #if DEVELOPMENT || DEBUG
3656 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",
3657 jld_eval_aggressive_count, aggressive_jetsam_needed ? "PASSED" : "FAILED", *total_candidates, bad_candidates,
3658 kJetsamHighRelaunchCandidatesThreshold, (uint64_t)memorystatus_available_pages, (uint64_t)memorystatus_sysproc_aging_aggr_pages);
3659 #endif /* DEVELOPMENT || DEBUG */
3660 return aggressive_jetsam_needed;
3661 }
3662
3663 static boolean_t
3664 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)
3665 {
3666 boolean_t aggressive_jetsam_needed = false;
3667 /* Jetsam Loop Detection - locals */
3668 memstat_bucket_t *bucket;
3669 int jld_bucket_count = 0;
3670
3671 proc_list_lock();
3672 switch (jetsam_aging_policy) {
3673 case kJetsamAgingPolicyLegacy:
3674 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3675 jld_bucket_count = bucket->count;
3676 bucket = &memstat_bucket[JETSAM_PRIORITY_AGING_BAND1];
3677 jld_bucket_count += bucket->count;
3678 break;
3679 case kJetsamAgingPolicyAppsReclaimedFirst:
3680 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3681 jld_bucket_count = bucket->count;
3682 bucket = &memstat_bucket[system_procs_aging_band];
3683 jld_bucket_count += bucket->count;
3684 bucket = &memstat_bucket[applications_aging_band];
3685 jld_bucket_count += bucket->count;
3686 break;
3687 case kJetsamAgingPolicyNone:
3688 default:
3689 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3690 jld_bucket_count = bucket->count;
3691 break;
3692 }
3693
3694 bucket = &memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE];
3695 *elevated_bucket_count = bucket->count;
3696 *total_candidates = jld_bucket_count;
3697 proc_list_unlock();
3698
3699 aggressive_jetsam_needed = (*jld_idle_kills > jld_idle_kill_candidates);
3700
3701 #if DEVELOPMENT || DEBUG
3702 if (aggressive_jetsam_needed) {
3703 printf("memorystatus: aggressive%d: idle candidates: %d, idle kills: %d\n",
3704 jld_eval_aggressive_count,
3705 jld_idle_kill_candidates,
3706 *jld_idle_kills);
3707 }
3708 #endif /* DEVELOPMENT || DEBUG */
3709 return aggressive_jetsam_needed;
3710 }
3711
3712 static boolean_t
3713 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)
3714 {
3715 boolean_t aggressive_jetsam_needed = false;
3716 boolean_t killed;
3717 uint32_t errors = 0;
3718 uint64_t footprint_of_killed_proc = 0;
3719 int elevated_bucket_count = 0;
3720 int total_candidates = 0;
3721 *memory_reclaimed = 0;
3722
3723 /*
3724 * The aggressive jetsam logic looks at the number of times it has been in the
3725 * aggressive loop to determine the max priority band it should kill upto. The
3726 * static variables below are used to track that property.
3727 *
3728 * To reset those values, the implementation checks if it has been
3729 * memorystatus_jld_eval_period_msecs since the parameters were reset.
3730 */
3731 static int jld_eval_aggressive_count = 0;
3732 static int32_t jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
3733 static uint64_t jld_timestamp_msecs = 0;
3734 static int jld_idle_kill_candidates = 0;
3735
3736 if (memorystatus_jld_enabled == FALSE) {
3737 /* If aggressive jetsam is disabled, nothing to do here */
3738 return FALSE;
3739 }
3740
3741 /* Get current timestamp (msecs only) */
3742 struct timeval jld_now_tstamp = {0, 0};
3743 uint64_t jld_now_msecs = 0;
3744 microuptime(&jld_now_tstamp);
3745 jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
3746
3747 /*
3748 * The aggressive jetsam logic looks at the number of candidates and their
3749 * properties to decide if aggressive jetsam should be engaged.
3750 */
3751 if (jetsam_aging_policy == kJetsamAgingPolicySysProcsReclaimedFirst) {
3752 /*
3753 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, the logic looks at the number of
3754 * candidates in the idle and deferred band and how many out of them are marked as high relaunch
3755 * probability.
3756 */
3757 aggressive_jetsam_needed = memorystatus_aggressive_jetsam_needed_sysproc_aging(jld_eval_aggressive_count,
3758 jld_idle_kills, jld_idle_kill_candidates, &total_candidates, &elevated_bucket_count);
3759 } else {
3760 /*
3761 * The other aging policies look at number of candidate processes over a specific time window and
3762 * evaluate if the system is in a jetsam loop. If yes, aggressive jetsam is triggered.
3763 */
3764 aggressive_jetsam_needed = memorystatus_aggressive_jetsam_needed_default(jld_eval_aggressive_count,
3765 jld_idle_kills, jld_idle_kill_candidates, &total_candidates, &elevated_bucket_count);
3766 }
3767
3768 /*
3769 * Check if its been really long since the aggressive jetsam evaluation
3770 * parameters have been refreshed. This logic also resets the jld_eval_aggressive_count
3771 * counter to make sure we reset the aggressive jetsam severity.
3772 */
3773 boolean_t param_reval = false;
3774
3775 if ((total_candidates == 0) ||
3776 (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
3777 jld_timestamp_msecs = jld_now_msecs;
3778 jld_idle_kill_candidates = total_candidates;
3779 *jld_idle_kills = 0;
3780 jld_eval_aggressive_count = 0;
3781 jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
3782 param_reval = true;
3783 }
3784
3785 /*
3786 * If the parameters have been updated, re-evaluate the aggressive_jetsam_needed condition for
3787 * the non kJetsamAgingPolicySysProcsReclaimedFirst policy since its based on jld_idle_kill_candidates etc.
3788 */
3789 if ((param_reval == true) && (jetsam_aging_policy != kJetsamAgingPolicySysProcsReclaimedFirst)) {
3790 aggressive_jetsam_needed = (*jld_idle_kills > jld_idle_kill_candidates);
3791 }
3792
3793 /*
3794 * It is also possible that the system is down to a very small number of processes in the candidate
3795 * bands. In that case, the decisions made by the memorystatus_aggressive_jetsam_needed_* routines
3796 * would not be useful. In that case, do not trigger aggressive jetsam.
3797 */
3798 if (total_candidates < kJetsamMinCandidatesThreshold) {
3799 #if DEVELOPMENT || DEBUG
3800 printf("memorystatus: aggressive: [FAILED] Low Candidate Count (current: %d, threshold: %d)\n", total_candidates, kJetsamMinCandidatesThreshold);
3801 #endif /* DEVELOPMENT || DEBUG */
3802 aggressive_jetsam_needed = false;
3803 }
3804
3805 if (aggressive_jetsam_needed == false) {
3806 /* Either the aging policy or the candidate count decided that aggressive jetsam is not needed. Nothing more to do here. */
3807 return FALSE;
3808 }
3809
3810 /* Looks like aggressive jetsam is needed */
3811 jld_eval_aggressive_count++;
3812
3813 if (jld_eval_aggressive_count == memorystatus_jld_eval_aggressive_count) {
3814 memorystatus_issue_fg_band_notify();
3815
3816 /*
3817 * If we reach this aggressive cycle, corpses might be causing memory pressure.
3818 * So, in an effort to avoid jetsams in the FG band, we will attempt to purge
3819 * corpse memory prior to this final march through JETSAM_PRIORITY_UI_SUPPORT.
3820 */
3821 if (total_corpses_count() > 0 && !*corpse_list_purged) {
3822 task_purge_all_corpses();
3823 *corpse_list_purged = TRUE;
3824 }
3825 } else if (jld_eval_aggressive_count > memorystatus_jld_eval_aggressive_count) {
3826 /*
3827 * Bump up the jetsam priority limit (eg: the bucket index)
3828 * Enforce bucket index sanity.
3829 */
3830 if ((memorystatus_jld_eval_aggressive_priority_band_max < 0) ||
3831 (memorystatus_jld_eval_aggressive_priority_band_max >= MEMSTAT_BUCKET_COUNT)) {
3832 /*
3833 * Do nothing. Stick with the default level.
3834 */
3835 } else {
3836 jld_priority_band_max = memorystatus_jld_eval_aggressive_priority_band_max;
3837 }
3838 }
3839
3840 /* Visit elevated processes first */
3841 while (elevated_bucket_count) {
3842 elevated_bucket_count--;
3843
3844 /*
3845 * memorystatus_kill_elevated_process() drops a reference,
3846 * so take another one so we can continue to use this exit reason
3847 * even after it returns.
3848 */
3849
3850 os_reason_ref(jetsam_reason);
3851 killed = memorystatus_kill_elevated_process(
3852 cause,
3853 jetsam_reason,
3854 JETSAM_PRIORITY_ELEVATED_INACTIVE,
3855 jld_eval_aggressive_count,
3856 &errors, &footprint_of_killed_proc);
3857 if (killed) {
3858 *post_snapshot = TRUE;
3859 *memory_reclaimed += footprint_of_killed_proc;
3860 if (memorystatus_avail_pages_below_pressure()) {
3861 /*
3862 * Still under pressure.
3863 * Find another pinned processes.
3864 */
3865 continue;
3866 } else {
3867 return TRUE;
3868 }
3869 } else {
3870 /*
3871 * No pinned processes left to kill.
3872 * Abandon elevated band.
3873 */
3874 break;
3875 }
3876 }
3877
3878 /*
3879 * memorystatus_kill_processes_aggressive() allocates its own
3880 * jetsam_reason so the kMemorystatusKilledProcThrashing cause
3881 * is consistent throughout the aggressive march.
3882 */
3883 killed = memorystatus_kill_processes_aggressive(
3884 kMemorystatusKilledProcThrashing,
3885 jld_eval_aggressive_count,
3886 jld_priority_band_max,
3887 &errors, &footprint_of_killed_proc);
3888
3889 if (killed) {
3890 /* Always generate logs after aggressive kill */
3891 *post_snapshot = TRUE;
3892 *memory_reclaimed += footprint_of_killed_proc;
3893 *jld_idle_kills = 0;
3894 return TRUE;
3895 }
3896
3897 return FALSE;
3898 }
3899
3900
3901 static void
3902 memorystatus_thread(void *param __unused, wait_result_t wr __unused)
3903 {
3904 boolean_t post_snapshot = FALSE;
3905 uint32_t errors = 0;
3906 uint32_t hwm_kill = 0;
3907 boolean_t sort_flag = TRUE;
3908 boolean_t corpse_list_purged = FALSE;
3909 int jld_idle_kills = 0;
3910 struct jetsam_thread_state *jetsam_thread = jetsam_current_thread();
3911 uint64_t total_memory_reclaimed = 0;
3912
3913 assert(jetsam_thread != NULL);
3914 if (jetsam_thread->inited == FALSE) {
3915 /*
3916 * It's the first time the thread has run, so just mark the thread as privileged and block.
3917 * This avoids a spurious pass with unset variables, as set out in <rdar://problem/9609402>.
3918 */
3919
3920 char name[32];
3921 thread_wire(host_priv_self(), current_thread(), TRUE);
3922 snprintf(name, 32, "VM_memorystatus_%d", jetsam_thread->index + 1);
3923
3924 /* Limit all but one thread to the lower jetsam bands, as that's where most of the victims are. */
3925 if (jetsam_thread->index == 0) {
3926 if (vm_pageout_state.vm_restricted_to_single_processor == TRUE) {
3927 thread_vm_bind_group_add();
3928 }
3929 jetsam_thread->limit_to_low_bands = FALSE;
3930 } else {
3931 jetsam_thread->limit_to_low_bands = TRUE;
3932 }
3933 thread_set_thread_name(current_thread(), name);
3934 jetsam_thread->inited = TRUE;
3935 memorystatus_thread_block(0, memorystatus_thread);
3936 }
3937
3938 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_START,
3939 memorystatus_available_pages, memorystatus_jld_enabled, memorystatus_jld_eval_period_msecs, memorystatus_jld_eval_aggressive_count, 0);
3940
3941 /*
3942 * Jetsam aware version.
3943 *
3944 * The VM pressure notification thread is working it's way through clients in parallel.
3945 *
3946 * So, while the pressure notification thread is targeting processes in order of
3947 * increasing jetsam priority, we can hopefully reduce / stop it's work by killing
3948 * any processes that have exceeded their highwater mark.
3949 *
3950 * If we run out of HWM processes and our available pages drops below the critical threshold, then,
3951 * we target the least recently used process in order of increasing jetsam priority (exception: the FG band).
3952 */
3953 while (memorystatus_action_needed()) {
3954 boolean_t killed;
3955 int32_t priority;
3956 uint32_t cause;
3957 uint64_t memory_reclaimed = 0;
3958 uint64_t jetsam_reason_code = JETSAM_REASON_INVALID;
3959 os_reason_t jetsam_reason = OS_REASON_NULL;
3960
3961 cause = kill_under_pressure_cause;
3962 switch (cause) {
3963 case kMemorystatusKilledFCThrashing:
3964 jetsam_reason_code = JETSAM_REASON_MEMORY_FCTHRASHING;
3965 break;
3966 case kMemorystatusKilledVMCompressorThrashing:
3967 jetsam_reason_code = JETSAM_REASON_MEMORY_VMCOMPRESSOR_THRASHING;
3968 break;
3969 case kMemorystatusKilledVMCompressorSpaceShortage:
3970 jetsam_reason_code = JETSAM_REASON_MEMORY_VMCOMPRESSOR_SPACE_SHORTAGE;
3971 break;
3972 case kMemorystatusKilledZoneMapExhaustion:
3973 jetsam_reason_code = JETSAM_REASON_ZONE_MAP_EXHAUSTION;
3974 break;
3975 case kMemorystatusKilledVMPageShortage:
3976 /* falls through */
3977 default:
3978 jetsam_reason_code = JETSAM_REASON_MEMORY_VMPAGESHORTAGE;
3979 cause = kMemorystatusKilledVMPageShortage;
3980 break;
3981 }
3982
3983 /* Highwater */
3984 boolean_t is_critical = TRUE;
3985 if (memorystatus_act_on_hiwat_processes(&errors, &hwm_kill, &post_snapshot, &is_critical, &memory_reclaimed)) {
3986 total_memory_reclaimed += memory_reclaimed;
3987 if (is_critical == FALSE) {
3988 /*
3989 * For now, don't kill any other processes.
3990 */
3991 break;
3992 } else {
3993 goto done;
3994 }
3995 }
3996
3997 jetsam_reason = os_reason_create(OS_REASON_JETSAM, jetsam_reason_code);
3998 if (jetsam_reason == OS_REASON_NULL) {
3999 printf("memorystatus_thread: failed to allocate jetsam reason\n");
4000 }
4001
4002 /* Only unlimited jetsam threads should act aggressive */
4003 if (!jetsam_thread->limit_to_low_bands &&
4004 memorystatus_act_aggressive(cause, jetsam_reason, &jld_idle_kills, &corpse_list_purged, &post_snapshot, &memory_reclaimed)) {
4005 total_memory_reclaimed += memory_reclaimed;
4006 goto done;
4007 }
4008
4009 /*
4010 * memorystatus_kill_top_process() drops a reference,
4011 * so take another one so we can continue to use this exit reason
4012 * even after it returns
4013 */
4014 os_reason_ref(jetsam_reason);
4015
4016 /* LRU */
4017 killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, jetsam_reason, &priority, &errors, &memory_reclaimed);
4018 sort_flag = FALSE;
4019
4020 if (killed) {
4021 total_memory_reclaimed += memory_reclaimed;
4022 if (memorystatus_post_snapshot(priority, cause) == TRUE) {
4023 post_snapshot = TRUE;
4024 }
4025
4026 /* Jetsam Loop Detection */
4027 if (memorystatus_jld_enabled == TRUE) {
4028 if ((priority == JETSAM_PRIORITY_IDLE) || (priority == system_procs_aging_band) || (priority == applications_aging_band)) {
4029 jld_idle_kills++;
4030 } else {
4031 /*
4032 * We've reached into bands beyond idle deferred.
4033 * We make no attempt to monitor them
4034 */
4035 }
4036 }
4037
4038 /*
4039 * If we have jetsammed a process in or above JETSAM_PRIORITY_UI_SUPPORT
4040 * then we attempt to relieve pressure by purging corpse memory and notifying
4041 * anybody wanting to know this.
4042 */
4043 if (priority >= JETSAM_PRIORITY_UI_SUPPORT) {
4044 memorystatus_issue_fg_band_notify();
4045 if (total_corpses_count() > 0 && !corpse_list_purged) {
4046 task_purge_all_corpses();
4047 corpse_list_purged = TRUE;
4048 }
4049 }
4050 goto done;
4051 }
4052
4053 if (memorystatus_avail_pages_below_critical()) {
4054 /*
4055 * Still under pressure and unable to kill a process - purge corpse memory
4056 */
4057 if (total_corpses_count() > 0) {
4058 task_purge_all_corpses();
4059 corpse_list_purged = TRUE;
4060 }
4061
4062 if (!jetsam_thread->limit_to_low_bands && memorystatus_avail_pages_below_critical()) {
4063 /*
4064 * Still under pressure and unable to kill a process - panic
4065 */
4066 panic("memorystatus_jetsam_thread: no victim! available pages:%llu\n", (uint64_t)memorystatus_available_pages);
4067 }
4068 }
4069
4070 done:
4071
4072 /*
4073 * We do not want to over-kill when thrashing has been detected.
4074 * To avoid that, we reset the flag here and notify the
4075 * compressor.
4076 */
4077 if (is_reason_thrashing(kill_under_pressure_cause)) {
4078 kill_under_pressure_cause = 0;
4079 #if CONFIG_JETSAM
4080 vm_thrashing_jetsam_done();
4081 #endif /* CONFIG_JETSAM */
4082 } else if (is_reason_zone_map_exhaustion(kill_under_pressure_cause)) {
4083 kill_under_pressure_cause = 0;
4084 }
4085
4086 os_reason_free(jetsam_reason);
4087 }
4088
4089 kill_under_pressure_cause = 0;
4090
4091 if (errors) {
4092 memorystatus_clear_errors();
4093 }
4094
4095 if (post_snapshot) {
4096 proc_list_lock();
4097 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4098 sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
4099 uint64_t timestamp_now = mach_absolute_time();
4100 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4101 memorystatus_jetsam_snapshot->js_gencount++;
4102 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4103 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4104 proc_list_unlock();
4105 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4106 if (!ret) {
4107 proc_list_lock();
4108 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4109 proc_list_unlock();
4110 }
4111 } else {
4112 proc_list_unlock();
4113 }
4114 }
4115
4116 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_END,
4117 memorystatus_available_pages, total_memory_reclaimed, 0, 0, 0);
4118
4119 memorystatus_thread_block(0, memorystatus_thread);
4120 }
4121
4122 /*
4123 * Returns TRUE:
4124 * when an idle-exitable proc was killed
4125 * Returns FALSE:
4126 * when there are no more idle-exitable procs found
4127 * when the attempt to kill an idle-exitable proc failed
4128 */
4129 boolean_t
4130 memorystatus_idle_exit_from_VM(void)
4131 {
4132 /*
4133 * This routine should no longer be needed since we are
4134 * now using jetsam bands on all platforms and so will deal
4135 * with IDLE processes within the memorystatus thread itself.
4136 *
4137 * But we still use it because we observed that macos systems
4138 * started heavy compression/swapping with a bunch of
4139 * idle-exitable processes alive and doing nothing. We decided
4140 * to rather kill those processes than start swapping earlier.
4141 */
4142
4143 return kill_idle_exit_proc();
4144 }
4145
4146 /*
4147 * Callback invoked when allowable physical memory footprint exceeded
4148 * (dirty pages + IOKit mappings)
4149 *
4150 * This is invoked for both advisory, non-fatal per-task high watermarks,
4151 * as well as the fatal task memory limits.
4152 */
4153 void
4154 memorystatus_on_ledger_footprint_exceeded(boolean_t warning, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
4155 {
4156 os_reason_t jetsam_reason = OS_REASON_NULL;
4157
4158 proc_t p = current_proc();
4159
4160 #if VM_PRESSURE_EVENTS
4161 if (warning == TRUE) {
4162 /*
4163 * This is a warning path which implies that the current process is close, but has
4164 * not yet exceeded its per-process memory limit.
4165 */
4166 if (memorystatus_warn_process(p->p_pid, memlimit_is_active, memlimit_is_fatal, FALSE /* not exceeded */) != TRUE) {
4167 /* Print warning, since it's possible that task has not registered for pressure notifications */
4168 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);
4169 }
4170 return;
4171 }
4172 #endif /* VM_PRESSURE_EVENTS */
4173
4174 if (memlimit_is_fatal) {
4175 /*
4176 * If this process has no high watermark or has a fatal task limit, then we have been invoked because the task
4177 * has violated either the system-wide per-task memory limit OR its own task limit.
4178 */
4179 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_PERPROCESSLIMIT);
4180 if (jetsam_reason == NULL) {
4181 printf("task_exceeded footprint: failed to allocate jetsam reason\n");
4182 } else if (corpse_for_fatal_memkill != 0 && proc_send_synchronous_EXC_RESOURCE(p) == FALSE) {
4183 /* Set OS_REASON_FLAG_GENERATE_CRASH_REPORT to generate corpse */
4184 jetsam_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
4185 }
4186
4187 if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit, jetsam_reason) != TRUE) {
4188 printf("task_exceeded_footprint: failed to kill the current task (exiting?).\n");
4189 }
4190 } else {
4191 /*
4192 * HWM offender exists. Done without locks or synchronization.
4193 * See comment near its declaration for more details.
4194 */
4195 memorystatus_hwm_candidates = TRUE;
4196
4197 #if VM_PRESSURE_EVENTS
4198 /*
4199 * The current process is not in the warning path.
4200 * This path implies the current process has exceeded a non-fatal (soft) memory limit.
4201 * Failure to send note is ignored here.
4202 */
4203 (void)memorystatus_warn_process(p->p_pid, memlimit_is_active, memlimit_is_fatal, TRUE /* exceeded */);
4204
4205 #endif /* VM_PRESSURE_EVENTS */
4206 }
4207 }
4208
4209 void
4210 memorystatus_log_exception(const int max_footprint_mb, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
4211 {
4212 proc_t p = current_proc();
4213
4214 /*
4215 * The limit violation is logged here, but only once per process per limit.
4216 * Soft memory limit is a non-fatal high-water-mark
4217 * Hard memory limit is a fatal custom-task-limit or system-wide per-task memory limit.
4218 */
4219
4220 os_log_with_startup_serial(OS_LOG_DEFAULT, "EXC_RESOURCE -> %s[%d] exceeded mem limit: %s%s %d MB (%s)\n",
4221 ((p && *p->p_name) ? p->p_name : "unknown"), (p ? p->p_pid : -1), (memlimit_is_active ? "Active" : "Inactive"),
4222 (memlimit_is_fatal ? "Hard" : "Soft"), max_footprint_mb,
4223 (memlimit_is_fatal ? "fatal" : "non-fatal"));
4224
4225 return;
4226 }
4227
4228
4229 /*
4230 * Description:
4231 * Evaluates process state to determine which limit
4232 * should be applied (active vs. inactive limit).
4233 *
4234 * Processes that have the 'elevated inactive jetsam band' attribute
4235 * are first evaluated based on their current priority band.
4236 * presently elevated ==> active
4237 *
4238 * Processes that opt into dirty tracking are evaluated
4239 * based on clean vs dirty state.
4240 * dirty ==> active
4241 * clean ==> inactive
4242 *
4243 * Process that do not opt into dirty tracking are
4244 * evalulated based on priority level.
4245 * Foreground or above ==> active
4246 * Below Foreground ==> inactive
4247 *
4248 * Return: TRUE if active
4249 * False if inactive
4250 */
4251
4252 static boolean_t
4253 proc_jetsam_state_is_active_locked(proc_t p)
4254 {
4255 if ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) &&
4256 (p->p_memstat_effectivepriority == JETSAM_PRIORITY_ELEVATED_INACTIVE)) {
4257 /*
4258 * process has the 'elevated inactive jetsam band' attribute
4259 * and process is present in the elevated band
4260 * implies active state
4261 */
4262 return TRUE;
4263 } else if (p->p_memstat_dirty & P_DIRTY_TRACK) {
4264 /*
4265 * process has opted into dirty tracking
4266 * active state is based on dirty vs. clean
4267 */
4268 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
4269 /*
4270 * process is dirty
4271 * implies active state
4272 */
4273 return TRUE;
4274 } else {
4275 /*
4276 * process is clean
4277 * implies inactive state
4278 */
4279 return FALSE;
4280 }
4281 } else if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
4282 /*
4283 * process is Foreground or higher
4284 * implies active state
4285 */
4286 return TRUE;
4287 } else {
4288 /*
4289 * process found below Foreground
4290 * implies inactive state
4291 */
4292 return FALSE;
4293 }
4294 }
4295
4296 static boolean_t
4297 memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason)
4298 {
4299 boolean_t res;
4300
4301 uint32_t errors = 0;
4302 uint64_t memory_reclaimed = 0;
4303
4304 if (victim_pid == -1) {
4305 /* No pid, so kill first process */
4306 res = memorystatus_kill_top_process(TRUE, TRUE, cause, jetsam_reason, NULL, &errors, &memory_reclaimed);
4307 } else {
4308 res = memorystatus_kill_specific_process(victim_pid, cause, jetsam_reason);
4309 }
4310
4311 if (errors) {
4312 memorystatus_clear_errors();
4313 }
4314
4315 if (res == TRUE) {
4316 /* Fire off snapshot notification */
4317 proc_list_lock();
4318 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4319 sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
4320 uint64_t timestamp_now = mach_absolute_time();
4321 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4322 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4323 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4324 proc_list_unlock();
4325 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4326 if (!ret) {
4327 proc_list_lock();
4328 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4329 proc_list_unlock();
4330 }
4331 } else {
4332 proc_list_unlock();
4333 }
4334 }
4335
4336 return res;
4337 }
4338
4339 /*
4340 * Jetsam a specific process.
4341 */
4342 static boolean_t
4343 memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason)
4344 {
4345 boolean_t killed;
4346 proc_t p;
4347 uint64_t killtime = 0;
4348 uint64_t footprint_of_killed_proc;
4349 clock_sec_t tv_sec;
4350 clock_usec_t tv_usec;
4351 uint32_t tv_msec;
4352
4353 /* TODO - add a victim queue and push this into the main jetsam thread */
4354
4355 p = proc_find(victim_pid);
4356 if (!p) {
4357 os_reason_free(jetsam_reason);
4358 return FALSE;
4359 }
4360
4361 proc_list_lock();
4362
4363 if (memorystatus_jetsam_snapshot_count == 0) {
4364 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
4365 }
4366
4367 killtime = mach_absolute_time();
4368 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
4369 tv_msec = tv_usec / 1000;
4370
4371 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
4372
4373 proc_list_unlock();
4374
4375 killed = memorystatus_do_kill(p, cause, jetsam_reason, &footprint_of_killed_proc);
4376
4377 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",
4378 (unsigned long)tv_sec, tv_msec, victim_pid, ((p && *p->p_name) ? p->p_name : "unknown"),
4379 memorystatus_kill_cause_name[cause], (p ? p->p_memstat_effectivepriority: -1),
4380 footprint_of_killed_proc >> 10, (uint64_t)memorystatus_available_pages);
4381
4382 proc_rele(p);
4383
4384 return killed;
4385 }
4386
4387
4388 /*
4389 * Toggle the P_MEMSTAT_TERMINATED state.
4390 * Takes the proc_list_lock.
4391 */
4392 void
4393 proc_memstat_terminated(proc_t p, boolean_t set)
4394 {
4395 #if DEVELOPMENT || DEBUG
4396 if (p) {
4397 proc_list_lock();
4398 if (set == TRUE) {
4399 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
4400 } else {
4401 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
4402 }
4403 proc_list_unlock();
4404 }
4405 #else
4406 #pragma unused(p, set)
4407 /*
4408 * do nothing
4409 */
4410 #endif /* DEVELOPMENT || DEBUG */
4411 return;
4412 }
4413
4414
4415 #if CONFIG_JETSAM
4416 /*
4417 * This is invoked when cpulimits have been exceeded while in fatal mode.
4418 * The jetsam_flags do not apply as those are for memory related kills.
4419 * We call this routine so that the offending process is killed with
4420 * a non-zero exit status.
4421 */
4422 void
4423 jetsam_on_ledger_cpulimit_exceeded(void)
4424 {
4425 int retval = 0;
4426 int jetsam_flags = 0; /* make it obvious */
4427 proc_t p = current_proc();
4428 os_reason_t jetsam_reason = OS_REASON_NULL;
4429
4430 printf("task_exceeded_cpulimit: killing pid %d [%s]\n",
4431 p->p_pid, (*p->p_name ? p->p_name : "(unknown)"));
4432
4433 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_CPULIMIT);
4434 if (jetsam_reason == OS_REASON_NULL) {
4435 printf("task_exceeded_cpulimit: unable to allocate memory for jetsam reason\n");
4436 }
4437
4438 retval = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
4439
4440 if (retval) {
4441 printf("task_exceeded_cpulimit: failed to kill current task (exiting?).\n");
4442 }
4443 }
4444
4445 #endif /* CONFIG_JETSAM */
4446
4447 static void
4448 memorystatus_get_task_memory_region_count(task_t task, uint64_t *count)
4449 {
4450 assert(task);
4451 assert(count);
4452
4453 *count = get_task_memory_region_count(task);
4454 }
4455
4456
4457 #define MEMORYSTATUS_VM_MAP_FORK_ALLOWED 0x100000000
4458 #define MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED 0x200000000
4459
4460 #if DEVELOPMENT || DEBUG
4461
4462 /*
4463 * Sysctl only used to test memorystatus_allowed_vm_map_fork() path.
4464 * set a new pidwatch value
4465 * or
4466 * get the current pidwatch value
4467 *
4468 * The pidwatch_val starts out with a PID to watch for in the map_fork path.
4469 * Its value is:
4470 * - OR'd with MEMORYSTATUS_VM_MAP_FORK_ALLOWED if we allow the map_fork.
4471 * - OR'd with MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED if we disallow the map_fork.
4472 * - set to -1ull if the map_fork() is aborted for other reasons.
4473 */
4474
4475 uint64_t memorystatus_vm_map_fork_pidwatch_val = 0;
4476
4477 static int sysctl_memorystatus_vm_map_fork_pidwatch SYSCTL_HANDLER_ARGS {
4478 #pragma unused(oidp, arg1, arg2)
4479
4480 uint64_t new_value = 0;
4481 uint64_t old_value = 0;
4482 int error = 0;
4483
4484 /*
4485 * The pid is held in the low 32 bits.
4486 * The 'allowed' flags are in the upper 32 bits.
4487 */
4488 old_value = memorystatus_vm_map_fork_pidwatch_val;
4489
4490 error = sysctl_io_number(req, old_value, sizeof(old_value), &new_value, NULL);
4491
4492 if (error || !req->newptr) {
4493 /*
4494 * No new value passed in.
4495 */
4496 return error;
4497 }
4498
4499 /*
4500 * A new pid was passed in via req->newptr.
4501 * Ignore any attempt to set the higher order bits.
4502 */
4503 memorystatus_vm_map_fork_pidwatch_val = new_value & 0xFFFFFFFF;
4504 printf("memorystatus: pidwatch old_value = 0x%llx, new_value = 0x%llx \n", old_value, new_value);
4505
4506 return error;
4507 }
4508
4509 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_map_fork_pidwatch, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_MASKED,
4510 0, 0, sysctl_memorystatus_vm_map_fork_pidwatch, "Q", "get/set pid watched for in vm_map_fork");
4511
4512
4513 /*
4514 * Record if a watched process fails to qualify for a vm_map_fork().
4515 */
4516 void
4517 memorystatus_abort_vm_map_fork(task_t task)
4518 {
4519 if (memorystatus_vm_map_fork_pidwatch_val != 0) {
4520 proc_t p = get_bsdtask_info(task);
4521 if (p != NULL && memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid) {
4522 memorystatus_vm_map_fork_pidwatch_val = -1ull;
4523 }
4524 }
4525 }
4526
4527 static void
4528 set_vm_map_fork_pidwatch(task_t task, uint64_t x)
4529 {
4530 if (memorystatus_vm_map_fork_pidwatch_val != 0) {
4531 proc_t p = get_bsdtask_info(task);
4532 if (p && (memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid)) {
4533 memorystatus_vm_map_fork_pidwatch_val |= x;
4534 }
4535 }
4536 }
4537
4538 #else /* DEVELOPMENT || DEBUG */
4539
4540
4541 static void
4542 set_vm_map_fork_pidwatch(task_t task, uint64_t x)
4543 {
4544 #pragma unused(task)
4545 #pragma unused(x)
4546 }
4547
4548 #endif /* DEVELOPMENT || DEBUG */
4549
4550 /*
4551 * Called during EXC_RESOURCE handling when a process exceeds a soft
4552 * memory limit. This is the corpse fork path and here we decide if
4553 * vm_map_fork will be allowed when creating the corpse.
4554 * The task being considered is suspended.
4555 *
4556 * By default, a vm_map_fork is allowed to proceed.
4557 *
4558 * A few simple policy assumptions:
4559 * Desktop platform is not considered in this path.
4560 * The vm_map_fork is always allowed.
4561 *
4562 * If the device has a zero system-wide task limit,
4563 * then the vm_map_fork is allowed.
4564 *
4565 * And if a process's memory footprint calculates less
4566 * than or equal to quarter of the system-wide task limit,
4567 * then the vm_map_fork is allowed. This calculation
4568 * is based on the assumption that a process can
4569 * munch memory up to the system-wide task limit.
4570 */
4571 extern boolean_t corpse_threshold_system_limit;
4572 boolean_t
4573 memorystatus_allowed_vm_map_fork(task_t task)
4574 {
4575 boolean_t is_allowed = TRUE; /* default */
4576
4577 #if CONFIG_EMBEDDED
4578
4579 uint64_t footprint_in_bytes;
4580 uint64_t max_allowed_bytes;
4581
4582 if (max_task_footprint_mb == 0) {
4583 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_ALLOWED);
4584 return is_allowed;
4585 }
4586
4587 footprint_in_bytes = get_task_phys_footprint(task);
4588
4589 /*
4590 * Maximum is 1/4 of the system-wide task limit by default.
4591 */
4592 max_allowed_bytes = ((uint64_t)max_task_footprint_mb * 1024 * 1024) >> 2;
4593
4594 #if DEBUG || DEVELOPMENT
4595 if (corpse_threshold_system_limit) {
4596 max_allowed_bytes = (uint64_t)max_task_footprint_mb * (1UL << 20);
4597 }
4598 #endif /* DEBUG || DEVELOPMENT */
4599
4600 if (footprint_in_bytes > max_allowed_bytes) {
4601 printf("memorystatus disallowed vm_map_fork %lld %lld\n", footprint_in_bytes, max_allowed_bytes);
4602 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED);
4603 return !is_allowed;
4604 }
4605 #endif /* CONFIG_EMBEDDED */
4606
4607 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_ALLOWED);
4608 return is_allowed;
4609 }
4610
4611 void
4612 memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages)
4613 {
4614 assert(task);
4615 assert(footprint);
4616
4617 uint64_t pages;
4618
4619 pages = (get_task_phys_footprint(task) / PAGE_SIZE_64);
4620 assert(((uint32_t)pages) == pages);
4621 *footprint = (uint32_t)pages;
4622
4623 if (max_footprint_lifetime) {
4624 pages = (get_task_phys_footprint_lifetime_max(task) / PAGE_SIZE_64);
4625 assert(((uint32_t)pages) == pages);
4626 *max_footprint_lifetime = (uint32_t)pages;
4627 }
4628 if (purgeable_pages) {
4629 pages = (get_task_purgeable_size(task) / PAGE_SIZE_64);
4630 assert(((uint32_t)pages) == pages);
4631 *purgeable_pages = (uint32_t)pages;
4632 }
4633 }
4634
4635 static void
4636 memorystatus_get_task_phys_footprint_page_counts(task_t task,
4637 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
4638 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
4639 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
4640 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages)
4641 {
4642 assert(task);
4643
4644 if (internal_pages) {
4645 *internal_pages = (get_task_internal(task) / PAGE_SIZE_64);
4646 }
4647
4648 if (internal_compressed_pages) {
4649 *internal_compressed_pages = (get_task_internal_compressed(task) / PAGE_SIZE_64);
4650 }
4651
4652 if (purgeable_nonvolatile_pages) {
4653 *purgeable_nonvolatile_pages = (get_task_purgeable_nonvolatile(task) / PAGE_SIZE_64);
4654 }
4655
4656 if (purgeable_nonvolatile_compressed_pages) {
4657 *purgeable_nonvolatile_compressed_pages = (get_task_purgeable_nonvolatile_compressed(task) / PAGE_SIZE_64);
4658 }
4659
4660 if (alternate_accounting_pages) {
4661 *alternate_accounting_pages = (get_task_alternate_accounting(task) / PAGE_SIZE_64);
4662 }
4663
4664 if (alternate_accounting_compressed_pages) {
4665 *alternate_accounting_compressed_pages = (get_task_alternate_accounting_compressed(task) / PAGE_SIZE_64);
4666 }
4667
4668 if (iokit_mapped_pages) {
4669 *iokit_mapped_pages = (get_task_iokit_mapped(task) / PAGE_SIZE_64);
4670 }
4671
4672 if (page_table_pages) {
4673 *page_table_pages = (get_task_page_table(task) / PAGE_SIZE_64);
4674 }
4675 }
4676
4677 /*
4678 * This routine only acts on the global jetsam event snapshot.
4679 * Updating the process's entry can race when the memorystatus_thread
4680 * has chosen to kill a process that is racing to exit on another core.
4681 */
4682 static void
4683 memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime)
4684 {
4685 memorystatus_jetsam_snapshot_entry_t *entry = NULL;
4686 memorystatus_jetsam_snapshot_t *snapshot = NULL;
4687 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
4688
4689 unsigned int i;
4690
4691 LCK_MTX_ASSERT(proc_list_mlock, LCK_MTX_ASSERT_OWNED);
4692
4693 if (memorystatus_jetsam_snapshot_count == 0) {
4694 /*
4695 * No active snapshot.
4696 * Nothing to do.
4697 */
4698 return;
4699 }
4700
4701 /*
4702 * Sanity check as this routine should only be called
4703 * from a jetsam kill path.
4704 */
4705 assert(kill_cause != 0 && killtime != 0);
4706
4707 snapshot = memorystatus_jetsam_snapshot;
4708 snapshot_list = memorystatus_jetsam_snapshot->entries;
4709
4710 for (i = 0; i < memorystatus_jetsam_snapshot_count; i++) {
4711 if (snapshot_list[i].pid == p->p_pid) {
4712 entry = &snapshot_list[i];
4713
4714 if (entry->killed || entry->jse_killtime) {
4715 /*
4716 * We apparently raced on the exit path
4717 * for this process, as it's snapshot entry
4718 * has already recorded a kill.
4719 */
4720 assert(entry->killed && entry->jse_killtime);
4721 break;
4722 }
4723
4724 /*
4725 * Update the entry we just found in the snapshot.
4726 */
4727
4728 entry->killed = kill_cause;
4729 entry->jse_killtime = killtime;
4730 entry->jse_gencount = snapshot->js_gencount;
4731 entry->jse_idle_delta = p->p_memstat_idle_delta;
4732 #if CONFIG_FREEZE
4733 entry->jse_thaw_count = p->p_memstat_thaw_count;
4734 #else /* CONFIG_FREEZE */
4735 entry->jse_thaw_count = 0;
4736 #endif /* CONFIG_FREEZE */
4737
4738 /*
4739 * If a process has moved between bands since snapshot was
4740 * initialized, then likely these fields changed too.
4741 */
4742 if (entry->priority != p->p_memstat_effectivepriority) {
4743 strlcpy(entry->name, p->p_name, sizeof(entry->name));
4744 entry->priority = p->p_memstat_effectivepriority;
4745 entry->state = memorystatus_build_state(p);
4746 entry->user_data = p->p_memstat_userdata;
4747 entry->fds = p->p_fd->fd_nfiles;
4748 }
4749
4750 /*
4751 * Always update the page counts on a kill.
4752 */
4753
4754 uint32_t pages = 0;
4755 uint32_t max_pages_lifetime = 0;
4756 uint32_t purgeable_pages = 0;
4757
4758 memorystatus_get_task_page_counts(p->task, &pages, &max_pages_lifetime, &purgeable_pages);
4759 entry->pages = (uint64_t)pages;
4760 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
4761 entry->purgeable_pages = (uint64_t)purgeable_pages;
4762
4763 uint64_t internal_pages = 0;
4764 uint64_t internal_compressed_pages = 0;
4765 uint64_t purgeable_nonvolatile_pages = 0;
4766 uint64_t purgeable_nonvolatile_compressed_pages = 0;
4767 uint64_t alternate_accounting_pages = 0;
4768 uint64_t alternate_accounting_compressed_pages = 0;
4769 uint64_t iokit_mapped_pages = 0;
4770 uint64_t page_table_pages = 0;
4771
4772 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
4773 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
4774 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
4775 &iokit_mapped_pages, &page_table_pages);
4776
4777 entry->jse_internal_pages = internal_pages;
4778 entry->jse_internal_compressed_pages = internal_compressed_pages;
4779 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
4780 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
4781 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
4782 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
4783 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
4784 entry->jse_page_table_pages = page_table_pages;
4785
4786 uint64_t region_count = 0;
4787 memorystatus_get_task_memory_region_count(p->task, &region_count);
4788 entry->jse_memory_region_count = region_count;
4789
4790 goto exit;
4791 }
4792 }
4793
4794 if (entry == NULL) {
4795 /*
4796 * The entry was not found in the snapshot, so the process must have
4797 * launched after the snapshot was initialized.
4798 * Let's try to append the new entry.
4799 */
4800 if (memorystatus_jetsam_snapshot_count < memorystatus_jetsam_snapshot_max) {
4801 /*
4802 * A populated snapshot buffer exists
4803 * and there is room to init a new entry.
4804 */
4805 assert(memorystatus_jetsam_snapshot_count == snapshot->entry_count);
4806
4807 unsigned int next = memorystatus_jetsam_snapshot_count;
4808
4809 if (memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[next], (snapshot->js_gencount)) == TRUE) {
4810 entry = &snapshot_list[next];
4811 entry->killed = kill_cause;
4812 entry->jse_killtime = killtime;
4813
4814 snapshot->entry_count = ++next;
4815 memorystatus_jetsam_snapshot_count = next;
4816
4817 if (memorystatus_jetsam_snapshot_count >= memorystatus_jetsam_snapshot_max) {
4818 /*
4819 * We just used the last slot in the snapshot buffer.
4820 * We only want to log it once... so we do it here
4821 * when we notice we've hit the max.
4822 */
4823 printf("memorystatus: WARNING snapshot buffer is full, count %d\n",
4824 memorystatus_jetsam_snapshot_count);
4825 }
4826 }
4827 }
4828 }
4829
4830 exit:
4831 if (entry == NULL) {
4832 /*
4833 * If we reach here, the snapshot buffer could not be updated.
4834 * Most likely, the buffer is full, in which case we would have
4835 * logged a warning in the previous call.
4836 *
4837 * For now, we will stop appending snapshot entries.
4838 * When the buffer is consumed, the snapshot state will reset.
4839 */
4840
4841 MEMORYSTATUS_DEBUG(4, "memorystatus_update_jetsam_snapshot_entry_locked: failed to update pid %d, priority %d, count %d\n",
4842 p->p_pid, p->p_memstat_effectivepriority, memorystatus_jetsam_snapshot_count);
4843 }
4844
4845 return;
4846 }
4847
4848 #if CONFIG_JETSAM
4849 void
4850 memorystatus_pages_update(unsigned int pages_avail)
4851 {
4852 memorystatus_available_pages = pages_avail;
4853
4854 #if VM_PRESSURE_EVENTS
4855 /*
4856 * Since memorystatus_available_pages changes, we should
4857 * re-evaluate the pressure levels on the system and
4858 * check if we need to wake the pressure thread.
4859 * We also update memorystatus_level in that routine.
4860 */
4861 vm_pressure_response();
4862
4863 if (memorystatus_available_pages <= memorystatus_available_pages_pressure) {
4864 if (memorystatus_hwm_candidates || (memorystatus_available_pages <= memorystatus_available_pages_critical)) {
4865 memorystatus_thread_wake();
4866 }
4867 }
4868 #if CONFIG_FREEZE
4869 /*
4870 * We can't grab the freezer_mutex here even though that synchronization would be correct to inspect
4871 * the # of frozen processes and wakeup the freezer thread. Reason being that we come here into this
4872 * code with (possibly) the page-queue locks held and preemption disabled. So trying to grab a mutex here
4873 * will result in the "mutex with preemption disabled" panic.
4874 */
4875
4876 if (memorystatus_freeze_thread_should_run() == TRUE) {
4877 /*
4878 * The freezer thread is usually woken up by some user-space call i.e. pid_hibernate(any process).
4879 * That trigger isn't invoked often enough and so we are enabling this explicit wakeup here.
4880 */
4881 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
4882 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
4883 }
4884 }
4885 #endif /* CONFIG_FREEZE */
4886
4887 #else /* VM_PRESSURE_EVENTS */
4888
4889 boolean_t critical, delta;
4890
4891 if (!memorystatus_delta) {
4892 return;
4893 }
4894
4895 critical = (pages_avail < memorystatus_available_pages_critical) ? TRUE : FALSE;
4896 delta = ((pages_avail >= (memorystatus_available_pages + memorystatus_delta))
4897 || (memorystatus_available_pages >= (pages_avail + memorystatus_delta))) ? TRUE : FALSE;
4898
4899 if (critical || delta) {
4900 unsigned int total_pages;
4901
4902 total_pages = (unsigned int) atop_64(max_mem);
4903 #if CONFIG_SECLUDED_MEMORY
4904 total_pages -= vm_page_secluded_count;
4905 #endif /* CONFIG_SECLUDED_MEMORY */
4906 memorystatus_level = memorystatus_available_pages * 100 / total_pages;
4907 memorystatus_thread_wake();
4908 }
4909 #endif /* VM_PRESSURE_EVENTS */
4910 }
4911 #endif /* CONFIG_JETSAM */
4912
4913 static boolean_t
4914 memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount)
4915 {
4916 clock_sec_t tv_sec;
4917 clock_usec_t tv_usec;
4918 uint32_t pages = 0;
4919 uint32_t max_pages_lifetime = 0;
4920 uint32_t purgeable_pages = 0;
4921 uint64_t internal_pages = 0;
4922 uint64_t internal_compressed_pages = 0;
4923 uint64_t purgeable_nonvolatile_pages = 0;
4924 uint64_t purgeable_nonvolatile_compressed_pages = 0;
4925 uint64_t alternate_accounting_pages = 0;
4926 uint64_t alternate_accounting_compressed_pages = 0;
4927 uint64_t iokit_mapped_pages = 0;
4928 uint64_t page_table_pages = 0;
4929 uint64_t region_count = 0;
4930 uint64_t cids[COALITION_NUM_TYPES];
4931
4932 memset(entry, 0, sizeof(memorystatus_jetsam_snapshot_entry_t));
4933
4934 entry->pid = p->p_pid;
4935 strlcpy(&entry->name[0], p->p_name, sizeof(entry->name));
4936 entry->priority = p->p_memstat_effectivepriority;
4937
4938 memorystatus_get_task_page_counts(p->task, &pages, &max_pages_lifetime, &purgeable_pages);
4939 entry->pages = (uint64_t)pages;
4940 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
4941 entry->purgeable_pages = (uint64_t)purgeable_pages;
4942
4943 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
4944 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
4945 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
4946 &iokit_mapped_pages, &page_table_pages);
4947
4948 entry->jse_internal_pages = internal_pages;
4949 entry->jse_internal_compressed_pages = internal_compressed_pages;
4950 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
4951 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
4952 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
4953 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
4954 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
4955 entry->jse_page_table_pages = page_table_pages;
4956
4957 memorystatus_get_task_memory_region_count(p->task, &region_count);
4958 entry->jse_memory_region_count = region_count;
4959
4960 entry->state = memorystatus_build_state(p);
4961 entry->user_data = p->p_memstat_userdata;
4962 memcpy(&entry->uuid[0], &p->p_uuid[0], sizeof(p->p_uuid));
4963 entry->fds = p->p_fd->fd_nfiles;
4964
4965 absolutetime_to_microtime(get_task_cpu_time(p->task), &tv_sec, &tv_usec);
4966 entry->cpu_time.tv_sec = (int64_t)tv_sec;
4967 entry->cpu_time.tv_usec = (int64_t)tv_usec;
4968
4969 assert(p->p_stats != NULL);
4970 entry->jse_starttime = p->p_stats->ps_start; /* abstime process started */
4971 entry->jse_killtime = 0; /* abstime jetsam chose to kill process */
4972 entry->killed = 0; /* the jetsam kill cause */
4973 entry->jse_gencount = gencount; /* indicates a pass through jetsam thread, when process was targeted to be killed */
4974
4975 entry->jse_idle_delta = p->p_memstat_idle_delta; /* Most recent timespan spent in idle-band */
4976
4977 #if CONFIG_FREEZE
4978 entry->jse_thaw_count = p->p_memstat_thaw_count;
4979 #else /* CONFIG_FREEZE */
4980 entry->jse_thaw_count = 0;
4981 #endif /* CONFIG_FREEZE */
4982
4983 proc_coalitionids(p, cids);
4984 entry->jse_coalition_jetsam_id = cids[COALITION_TYPE_JETSAM];
4985
4986 return TRUE;
4987 }
4988
4989 static void
4990 memorystatus_init_snapshot_vmstats(memorystatus_jetsam_snapshot_t *snapshot)
4991 {
4992 kern_return_t kr = KERN_SUCCESS;
4993 mach_msg_type_number_t count = HOST_VM_INFO64_COUNT;
4994 vm_statistics64_data_t vm_stat;
4995
4996 if ((kr = host_statistics64(host_self(), HOST_VM_INFO64, (host_info64_t)&vm_stat, &count)) != KERN_SUCCESS) {
4997 printf("memorystatus_init_jetsam_snapshot_stats: host_statistics64 failed with %d\n", kr);
4998 memset(&snapshot->stats, 0, sizeof(snapshot->stats));
4999 } else {
5000 snapshot->stats.free_pages = vm_stat.free_count;
5001 snapshot->stats.active_pages = vm_stat.active_count;
5002 snapshot->stats.inactive_pages = vm_stat.inactive_count;
5003 snapshot->stats.throttled_pages = vm_stat.throttled_count;
5004 snapshot->stats.purgeable_pages = vm_stat.purgeable_count;
5005 snapshot->stats.wired_pages = vm_stat.wire_count;
5006
5007 snapshot->stats.speculative_pages = vm_stat.speculative_count;
5008 snapshot->stats.filebacked_pages = vm_stat.external_page_count;
5009 snapshot->stats.anonymous_pages = vm_stat.internal_page_count;
5010 snapshot->stats.compressions = vm_stat.compressions;
5011 snapshot->stats.decompressions = vm_stat.decompressions;
5012 snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
5013 snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
5014 }
5015
5016 get_zone_map_size(&snapshot->stats.zone_map_size, &snapshot->stats.zone_map_capacity);
5017
5018 bzero(snapshot->stats.largest_zone_name, sizeof(snapshot->stats.largest_zone_name));
5019 get_largest_zone_info(snapshot->stats.largest_zone_name, sizeof(snapshot->stats.largest_zone_name),
5020 &snapshot->stats.largest_zone_size);
5021 }
5022
5023 /*
5024 * Collect vm statistics at boot.
5025 * Called only once (see kern_exec.c)
5026 * Data can be consumed at any time.
5027 */
5028 void
5029 memorystatus_init_at_boot_snapshot()
5030 {
5031 memorystatus_init_snapshot_vmstats(&memorystatus_at_boot_snapshot);
5032 memorystatus_at_boot_snapshot.entry_count = 0;
5033 memorystatus_at_boot_snapshot.notification_time = 0; /* updated when consumed */
5034 memorystatus_at_boot_snapshot.snapshot_time = mach_absolute_time();
5035 }
5036
5037 static void
5038 memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count )
5039 {
5040 proc_t p, next_p;
5041 unsigned int b = 0, i = 0;
5042
5043 memorystatus_jetsam_snapshot_t *snapshot = NULL;
5044 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
5045 unsigned int snapshot_max = 0;
5046
5047 LCK_MTX_ASSERT(proc_list_mlock, LCK_MTX_ASSERT_OWNED);
5048
5049 if (od_snapshot) {
5050 /*
5051 * This is an on_demand snapshot
5052 */
5053 snapshot = od_snapshot;
5054 snapshot_list = od_snapshot->entries;
5055 snapshot_max = ods_list_count;
5056 } else {
5057 /*
5058 * This is a jetsam event snapshot
5059 */
5060 snapshot = memorystatus_jetsam_snapshot;
5061 snapshot_list = memorystatus_jetsam_snapshot->entries;
5062 snapshot_max = memorystatus_jetsam_snapshot_max;
5063 }
5064
5065 /*
5066 * Init the snapshot header information
5067 */
5068 memorystatus_init_snapshot_vmstats(snapshot);
5069 snapshot->snapshot_time = mach_absolute_time();
5070 snapshot->notification_time = 0;
5071 snapshot->js_gencount = 0;
5072
5073 next_p = memorystatus_get_first_proc_locked(&b, TRUE);
5074 while (next_p) {
5075 p = next_p;
5076 next_p = memorystatus_get_next_proc_locked(&b, p, TRUE);
5077
5078 if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i], snapshot->js_gencount)) {
5079 continue;
5080 }
5081
5082 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",
5083 p->p_pid,
5084 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],
5085 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]);
5086
5087 if (++i == snapshot_max) {
5088 break;
5089 }
5090 }
5091
5092 snapshot->entry_count = i;
5093
5094 if (!od_snapshot) {
5095 /* update the system buffer count */
5096 memorystatus_jetsam_snapshot_count = i;
5097 }
5098 }
5099
5100 #if DEVELOPMENT || DEBUG
5101
5102 #if CONFIG_JETSAM
5103 static int
5104 memorystatus_cmd_set_panic_bits(user_addr_t buffer, uint32_t buffer_size)
5105 {
5106 int ret;
5107 memorystatus_jetsam_panic_options_t debug;
5108
5109 if (buffer_size != sizeof(memorystatus_jetsam_panic_options_t)) {
5110 return EINVAL;
5111 }
5112
5113 ret = copyin(buffer, &debug, buffer_size);
5114 if (ret) {
5115 return ret;
5116 }
5117
5118 /* Panic bits match kMemorystatusKilled* enum */
5119 memorystatus_jetsam_panic_debug = (memorystatus_jetsam_panic_debug & ~debug.mask) | (debug.data & debug.mask);
5120
5121 /* Copyout new value */
5122 debug.data = memorystatus_jetsam_panic_debug;
5123 ret = copyout(&debug, buffer, sizeof(memorystatus_jetsam_panic_options_t));
5124
5125 return ret;
5126 }
5127 #endif /* CONFIG_JETSAM */
5128
5129 /*
5130 * Triggers a sort_order on a specified jetsam priority band.
5131 * This is for testing only, used to force a path through the sort
5132 * function.
5133 */
5134 static int
5135 memorystatus_cmd_test_jetsam_sort(int priority, int sort_order)
5136 {
5137 int error = 0;
5138
5139 unsigned int bucket_index = 0;
5140
5141 if (priority == -1) {
5142 /* Use as shorthand for default priority */
5143 bucket_index = JETSAM_PRIORITY_DEFAULT;
5144 } else {
5145 bucket_index = (unsigned int)priority;
5146 }
5147
5148 error = memorystatus_sort_bucket(bucket_index, sort_order);
5149
5150 return error;
5151 }
5152
5153 #endif /* DEVELOPMENT || DEBUG */
5154
5155 /*
5156 * Prepare the process to be killed (set state, update snapshot) and kill it.
5157 */
5158 static uint64_t memorystatus_purge_before_jetsam_success = 0;
5159
5160 static boolean_t
5161 memorystatus_kill_proc(proc_t p, uint32_t cause, os_reason_t jetsam_reason, boolean_t *killed, uint64_t *footprint_of_killed_proc)
5162 {
5163 pid_t aPid = 0;
5164 uint32_t aPid_ep = 0;
5165
5166 uint64_t killtime = 0;
5167 clock_sec_t tv_sec;
5168 clock_usec_t tv_usec;
5169 uint32_t tv_msec;
5170 boolean_t retval = FALSE;
5171
5172 aPid = p->p_pid;
5173 aPid_ep = p->p_memstat_effectivepriority;
5174
5175 if (cause != kMemorystatusKilledVnodes && cause != kMemorystatusKilledZoneMapExhaustion) {
5176 /*
5177 * Genuine memory pressure and not other (vnode/zone) resource exhaustion.
5178 */
5179 boolean_t success = FALSE;
5180 uint64_t num_pages_purged;
5181 uint64_t num_pages_reclaimed = 0;
5182 uint64_t num_pages_unsecluded = 0;
5183
5184 networking_memstatus_callout(p, cause);
5185 num_pages_purged = vm_purgeable_purge_task_owned(p->task);
5186 num_pages_reclaimed += num_pages_purged;
5187 #if CONFIG_SECLUDED_MEMORY
5188 if (cause == kMemorystatusKilledVMPageShortage &&
5189 vm_page_secluded_count > 0 &&
5190 task_can_use_secluded_mem(p->task, FALSE)) {
5191 /*
5192 * We're about to kill a process that has access
5193 * to the secluded pool. Drain that pool into the
5194 * free or active queues to make these pages re-appear
5195 * as "available", which might make us no longer need
5196 * to kill that process.
5197 * Since the secluded pool does not get refilled while
5198 * a process has access to it, it should remain
5199 * drained.
5200 */
5201 num_pages_unsecluded = vm_page_secluded_drain();
5202 num_pages_reclaimed += num_pages_unsecluded;
5203 }
5204 #endif /* CONFIG_SECLUDED_MEMORY */
5205
5206 if (num_pages_reclaimed) {
5207 /*
5208 * We actually reclaimed something and so let's
5209 * check if we need to continue with the kill.
5210 */
5211 if (cause == kMemorystatusKilledHiwat) {
5212 uint64_t footprint_in_bytes = get_task_phys_footprint(p->task);
5213 uint64_t memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
5214 success = (footprint_in_bytes <= memlimit_in_bytes);
5215 } else {
5216 success = (memorystatus_avail_pages_below_pressure() == FALSE);
5217 #if CONFIG_SECLUDED_MEMORY
5218 if (!success && num_pages_unsecluded) {
5219 /*
5220 * We just drained the secluded pool
5221 * because we're about to kill a
5222 * process that has access to it.
5223 * This is an important process and
5224 * we'd rather not kill it unless
5225 * absolutely necessary, so declare
5226 * success even if draining the pool
5227 * did not quite get us out of the
5228 * "pressure" level but still got
5229 * us out of the "critical" level.
5230 */
5231 success = (memorystatus_avail_pages_below_critical() == FALSE);
5232 }
5233 #endif /* CONFIG_SECLUDED_MEMORY */
5234 }
5235
5236 if (success) {
5237 memorystatus_purge_before_jetsam_success++;
5238
5239 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: reclaimed %llu pages (%llu purged, %llu unsecluded) from pid %d [%s] and avoided %s\n",
5240 num_pages_reclaimed, num_pages_purged, num_pages_unsecluded, aPid, ((p && *p->p_name) ? p->p_name : "unknown"), memorystatus_kill_cause_name[cause]);
5241
5242 *killed = FALSE;
5243
5244 return TRUE;
5245 }
5246 }
5247 }
5248
5249 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
5250 MEMORYSTATUS_DEBUG(1, "jetsam: killing pid %d [%s] - %lld Mb > 1 (%d Mb)\n",
5251 aPid, (*p->p_name ? p->p_name : "unknown"),
5252 (footprint_in_bytes / (1024ULL * 1024ULL)), /* converted bytes to MB */
5253 p->p_memstat_memlimit);
5254 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
5255
5256 killtime = mach_absolute_time();
5257 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5258 tv_msec = tv_usec / 1000;
5259
5260 proc_list_lock();
5261 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5262 proc_list_unlock();
5263
5264 char kill_reason_string[128];
5265
5266 if (cause == kMemorystatusKilledHiwat) {
5267 strlcpy(kill_reason_string, "killing_highwater_process", 128);
5268 } else {
5269 if (aPid_ep == JETSAM_PRIORITY_IDLE) {
5270 strlcpy(kill_reason_string, "killing_idle_process", 128);
5271 } else {
5272 strlcpy(kill_reason_string, "killing_top_process", 128);
5273 }
5274 }
5275
5276 /*
5277 * memorystatus_do_kill drops a reference, so take another one so we can
5278 * continue to use this exit reason even after memorystatus_do_kill()
5279 * returns
5280 */
5281 os_reason_ref(jetsam_reason);
5282
5283 retval = memorystatus_do_kill(p, cause, jetsam_reason, footprint_of_killed_proc);
5284 *killed = retval;
5285
5286 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: %s pid %d [%s] (%s %d) %lluKB - memorystatus_available_pages: %llu",
5287 (unsigned long)tv_sec, tv_msec, kill_reason_string,
5288 aPid, ((p && *p->p_name) ? p->p_name : "unknown"),
5289 memorystatus_kill_cause_name[cause], aPid_ep,
5290 (*footprint_of_killed_proc) >> 10, (uint64_t)memorystatus_available_pages);
5291
5292 return retval;
5293 }
5294
5295 /*
5296 * Jetsam the first process in the queue.
5297 */
5298 static boolean_t
5299 memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, os_reason_t jetsam_reason,
5300 int32_t *priority, uint32_t *errors, uint64_t *memory_reclaimed)
5301 {
5302 pid_t aPid;
5303 proc_t p = PROC_NULL, next_p = PROC_NULL;
5304 boolean_t new_snapshot = FALSE, force_new_snapshot = FALSE, killed = FALSE, freed_mem = FALSE;
5305 unsigned int i = 0;
5306 uint32_t aPid_ep;
5307 int32_t local_max_kill_prio = JETSAM_PRIORITY_IDLE;
5308 uint64_t footprint_of_killed_proc = 0;
5309
5310 #ifndef CONFIG_FREEZE
5311 #pragma unused(any)
5312 #endif
5313
5314 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5315 memorystatus_available_pages, 0, 0, 0, 0);
5316
5317
5318 #if CONFIG_JETSAM
5319 if (sort_flag == TRUE) {
5320 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
5321 }
5322
5323 local_max_kill_prio = max_kill_priority;
5324
5325 force_new_snapshot = FALSE;
5326
5327 #else /* CONFIG_JETSAM */
5328
5329 if (sort_flag == TRUE) {
5330 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_IDLE, JETSAM_SORT_DEFAULT);
5331 }
5332
5333 /*
5334 * On macos, we currently only have 2 reasons to be here:
5335 *
5336 * kMemorystatusKilledZoneMapExhaustion
5337 * AND
5338 * kMemorystatusKilledVMCompressorSpaceShortage
5339 *
5340 * If we are here because of kMemorystatusKilledZoneMapExhaustion, we will consider
5341 * any and all processes as eligible kill candidates since we need to avoid a panic.
5342 *
5343 * Since this function can be called async. it is harder to toggle the max_kill_priority
5344 * value before and after a call. And so we use this local variable to set the upper band
5345 * on the eligible kill bands.
5346 */
5347 if (cause == kMemorystatusKilledZoneMapExhaustion) {
5348 local_max_kill_prio = JETSAM_PRIORITY_MAX;
5349 } else {
5350 local_max_kill_prio = max_kill_priority;
5351 }
5352
5353 /*
5354 * And, because we are here under extreme circumstances, we force a snapshot even for
5355 * IDLE kills.
5356 */
5357 force_new_snapshot = TRUE;
5358
5359 #endif /* CONFIG_JETSAM */
5360
5361 if (cause != kMemorystatusKilledZoneMapExhaustion &&
5362 jetsam_current_thread() != NULL &&
5363 jetsam_current_thread()->limit_to_low_bands &&
5364 local_max_kill_prio > JETSAM_PRIORITY_BACKGROUND) {
5365 local_max_kill_prio = JETSAM_PRIORITY_BACKGROUND;
5366 }
5367
5368 proc_list_lock();
5369
5370 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5371 while (next_p && (next_p->p_memstat_effectivepriority <= local_max_kill_prio)) {
5372 p = next_p;
5373 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5374
5375
5376 aPid = p->p_pid;
5377 aPid_ep = p->p_memstat_effectivepriority;
5378
5379 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5380 continue; /* with lock held */
5381 }
5382
5383 if (cause == kMemorystatusKilledVnodes) {
5384 /*
5385 * If the system runs out of vnodes, we systematically jetsam
5386 * processes in hopes of stumbling onto a vnode gain that helps
5387 * the system recover. The process that happens to trigger
5388 * this path has no known relationship to the vnode shortage.
5389 * Deadlock avoidance: attempt to safeguard the caller.
5390 */
5391
5392 if (p == current_proc()) {
5393 /* do not jetsam the current process */
5394 continue;
5395 }
5396 }
5397
5398 #if CONFIG_FREEZE
5399 boolean_t skip;
5400 boolean_t reclaim_proc = !(p->p_memstat_state & P_MEMSTAT_LOCKED);
5401 if (any || reclaim_proc) {
5402 skip = FALSE;
5403 } else {
5404 skip = TRUE;
5405 }
5406
5407 if (skip) {
5408 continue;
5409 } else
5410 #endif
5411 {
5412 if (proc_ref_locked(p) == p) {
5413 /*
5414 * Mark as terminated so that if exit1() indicates success, but the process (for example)
5415 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
5416 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
5417 * acquisition of the proc lock.
5418 */
5419 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5420 } else {
5421 /*
5422 * We need to restart the search again because
5423 * proc_ref_locked _can_ drop the proc_list lock
5424 * and we could have lost our stored next_p via
5425 * an exit() on another core.
5426 */
5427 i = 0;
5428 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5429 continue;
5430 }
5431
5432 /*
5433 * Capture a snapshot if none exists and:
5434 * - we are forcing a new snapshot creation, either because:
5435 * - on a particular platform we need these snapshots every time, OR
5436 * - a boot-arg/embedded device tree property has been set.
5437 * - priority was not requested (this is something other than an ambient kill)
5438 * - the priority was requested *and* the targeted process is not at idle priority
5439 */
5440 if ((memorystatus_jetsam_snapshot_count == 0) &&
5441 (force_new_snapshot || memorystatus_idle_snapshot || ((!priority) || (priority && (aPid_ep != JETSAM_PRIORITY_IDLE))))) {
5442 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5443 new_snapshot = TRUE;
5444 }
5445
5446 proc_list_unlock();
5447
5448 freed_mem = memorystatus_kill_proc(p, cause, jetsam_reason, &killed, &footprint_of_killed_proc); /* purged and/or killed 'p' */
5449 /* Success? */
5450 if (freed_mem) {
5451 if (killed) {
5452 *memory_reclaimed = footprint_of_killed_proc;
5453 if (priority) {
5454 *priority = aPid_ep;
5455 }
5456 } else {
5457 /* purged */
5458 proc_list_lock();
5459 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5460 proc_list_unlock();
5461 }
5462 proc_rele(p);
5463 goto exit;
5464 }
5465
5466 /*
5467 * Failure - first unwind the state,
5468 * then fall through to restart the search.
5469 */
5470 proc_list_lock();
5471 proc_rele_locked(p);
5472 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5473 p->p_memstat_state |= P_MEMSTAT_ERROR;
5474 *errors += 1;
5475
5476 i = 0;
5477 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5478 }
5479 }
5480
5481 proc_list_unlock();
5482
5483 exit:
5484 os_reason_free(jetsam_reason);
5485
5486 if (!killed) {
5487 *memory_reclaimed = 0;
5488
5489 /* Clear snapshot if freshly captured and no target was found */
5490 if (new_snapshot) {
5491 proc_list_lock();
5492 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5493 proc_list_unlock();
5494 }
5495 }
5496
5497 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
5498 memorystatus_available_pages, killed ? aPid : 0, killed, *memory_reclaimed, 0);
5499
5500 return killed;
5501 }
5502
5503 /*
5504 * Jetsam aggressively
5505 */
5506 static boolean_t
5507 memorystatus_kill_processes_aggressive(uint32_t cause, int aggr_count,
5508 int32_t priority_max, uint32_t *errors, uint64_t *memory_reclaimed)
5509 {
5510 pid_t aPid;
5511 proc_t p = PROC_NULL, next_p = PROC_NULL;
5512 boolean_t new_snapshot = FALSE, killed = FALSE;
5513 int kill_count = 0;
5514 unsigned int i = 0;
5515 int32_t aPid_ep = 0;
5516 unsigned int memorystatus_level_snapshot = 0;
5517 uint64_t killtime = 0;
5518 clock_sec_t tv_sec;
5519 clock_usec_t tv_usec;
5520 uint32_t tv_msec;
5521 os_reason_t jetsam_reason = OS_REASON_NULL;
5522 uint64_t footprint_of_killed_proc = 0;
5523
5524 *memory_reclaimed = 0;
5525
5526 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5527 memorystatus_available_pages, priority_max, 0, 0, 0);
5528
5529 if (priority_max >= JETSAM_PRIORITY_FOREGROUND) {
5530 /*
5531 * Check if aggressive jetsam has been asked to kill upto or beyond the
5532 * JETSAM_PRIORITY_FOREGROUND bucket. If yes, sort the FG band based on
5533 * coalition footprint.
5534 */
5535 memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
5536 }
5537
5538 jetsam_reason = os_reason_create(OS_REASON_JETSAM, cause);
5539 if (jetsam_reason == OS_REASON_NULL) {
5540 printf("memorystatus_kill_processes_aggressive: failed to allocate exit reason\n");
5541 }
5542
5543 proc_list_lock();
5544
5545 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5546 while (next_p) {
5547 if (((next_p->p_listflag & P_LIST_EXITED) != 0) ||
5548 ((unsigned int)(next_p->p_memstat_effectivepriority) != i)) {
5549 /*
5550 * We have raced with next_p running on another core.
5551 * It may be exiting or it may have moved to a different
5552 * jetsam priority band. This means we have lost our
5553 * place in line while traversing the jetsam list. We
5554 * attempt to recover by rewinding to the beginning of the band
5555 * we were already traversing. By doing this, we do not guarantee
5556 * that no process escapes this aggressive march, but we can make
5557 * skipping an entire range of processes less likely. (PR-21069019)
5558 */
5559
5560 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: rewinding band %d, %s(%d) moved or exiting.\n",
5561 aggr_count, i, (*next_p->p_name ? next_p->p_name : "unknown"), next_p->p_pid);
5562
5563 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5564 continue;
5565 }
5566
5567 p = next_p;
5568 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5569
5570 if (p->p_memstat_effectivepriority > priority_max) {
5571 /*
5572 * Bail out of this killing spree if we have
5573 * reached beyond the priority_max jetsam band.
5574 * That is, we kill up to and through the
5575 * priority_max jetsam band.
5576 */
5577 proc_list_unlock();
5578 goto exit;
5579 }
5580
5581 aPid = p->p_pid;
5582 aPid_ep = p->p_memstat_effectivepriority;
5583
5584 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5585 continue;
5586 }
5587
5588 /*
5589 * Capture a snapshot if none exists.
5590 */
5591 if (memorystatus_jetsam_snapshot_count == 0) {
5592 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5593 new_snapshot = TRUE;
5594 }
5595
5596 /*
5597 * Mark as terminated so that if exit1() indicates success, but the process (for example)
5598 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
5599 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
5600 * acquisition of the proc lock.
5601 */
5602 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5603
5604 killtime = mach_absolute_time();
5605 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5606 tv_msec = tv_usec / 1000;
5607
5608 /* Shift queue, update stats */
5609 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5610
5611 /*
5612 * In order to kill the target process, we will drop the proc_list_lock.
5613 * To guaranteee that p and next_p don't disappear out from under the lock,
5614 * we must take a ref on both.
5615 * If we cannot get a reference, then it's likely we've raced with
5616 * that process exiting on another core.
5617 */
5618 if (proc_ref_locked(p) == p) {
5619 if (next_p) {
5620 while (next_p && (proc_ref_locked(next_p) != next_p)) {
5621 proc_t temp_p;
5622
5623 /*
5624 * We must have raced with next_p exiting on another core.
5625 * Recover by getting the next eligible process in the band.
5626 */
5627
5628 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: skipping %d [%s] (exiting?)\n",
5629 aggr_count, next_p->p_pid, (*next_p->p_name ? next_p->p_name : "(unknown)"));
5630
5631 temp_p = next_p;
5632 next_p = memorystatus_get_next_proc_locked(&i, temp_p, TRUE);
5633 }
5634 }
5635 proc_list_unlock();
5636
5637 printf("%lu.%03d memorystatus: %s%d pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
5638 (unsigned long)tv_sec, tv_msec,
5639 ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "killing_idle_process_aggressive" : "killing_top_process_aggressive"),
5640 aggr_count, aPid, (*p->p_name ? p->p_name : "unknown"),
5641 memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
5642
5643 memorystatus_level_snapshot = memorystatus_level;
5644
5645 /*
5646 * memorystatus_do_kill() drops a reference, so take another one so we can
5647 * continue to use this exit reason even after memorystatus_do_kill()
5648 * returns.
5649 */
5650 os_reason_ref(jetsam_reason);
5651 killed = memorystatus_do_kill(p, cause, jetsam_reason, &footprint_of_killed_proc);
5652
5653 /* Success? */
5654 if (killed) {
5655 *memory_reclaimed += footprint_of_killed_proc;
5656 proc_rele(p);
5657 kill_count++;
5658 p = NULL;
5659 killed = FALSE;
5660
5661 /*
5662 * Continue the killing spree.
5663 */
5664 proc_list_lock();
5665 if (next_p) {
5666 proc_rele_locked(next_p);
5667 }
5668
5669 if (aPid_ep == JETSAM_PRIORITY_FOREGROUND && memorystatus_aggressive_jetsam_lenient == TRUE) {
5670 if (memorystatus_level > memorystatus_level_snapshot && ((memorystatus_level - memorystatus_level_snapshot) >= AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD)) {
5671 #if DEVELOPMENT || DEBUG
5672 printf("Disabling Lenient mode after one-time deployment.\n");
5673 #endif /* DEVELOPMENT || DEBUG */
5674 memorystatus_aggressive_jetsam_lenient = FALSE;
5675 break;
5676 }
5677 }
5678
5679 continue;
5680 }
5681
5682 /*
5683 * Failure - first unwind the state,
5684 * then fall through to restart the search.
5685 */
5686 proc_list_lock();
5687 proc_rele_locked(p);
5688 if (next_p) {
5689 proc_rele_locked(next_p);
5690 }
5691 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5692 p->p_memstat_state |= P_MEMSTAT_ERROR;
5693 *errors += 1;
5694 p = NULL;
5695 }
5696
5697 /*
5698 * Failure - restart the search at the beginning of
5699 * the band we were already traversing.
5700 *
5701 * We might have raced with "p" exiting on another core, resulting in no
5702 * ref on "p". Or, we may have failed to kill "p".
5703 *
5704 * Either way, we fall thru to here, leaving the proc in the
5705 * P_MEMSTAT_TERMINATED or P_MEMSTAT_ERROR state.
5706 *
5707 * And, we hold the the proc_list_lock at this point.
5708 */
5709
5710 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5711 }
5712
5713 proc_list_unlock();
5714
5715 exit:
5716 os_reason_free(jetsam_reason);
5717
5718 /* Clear snapshot if freshly captured and no target was found */
5719 if (new_snapshot && (kill_count == 0)) {
5720 proc_list_lock();
5721 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5722 proc_list_unlock();
5723 }
5724
5725 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
5726 memorystatus_available_pages, 0, kill_count, *memory_reclaimed, 0);
5727
5728 if (kill_count > 0) {
5729 return TRUE;
5730 } else {
5731 return FALSE;
5732 }
5733 }
5734
5735 static boolean_t
5736 memorystatus_kill_hiwat_proc(uint32_t *errors, boolean_t *purged, uint64_t *memory_reclaimed)
5737 {
5738 pid_t aPid = 0;
5739 proc_t p = PROC_NULL, next_p = PROC_NULL;
5740 boolean_t new_snapshot = FALSE, killed = FALSE, freed_mem = FALSE;
5741 unsigned int i = 0;
5742 uint32_t aPid_ep;
5743 os_reason_t jetsam_reason = OS_REASON_NULL;
5744 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
5745 memorystatus_available_pages, 0, 0, 0, 0);
5746
5747 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_HIGHWATER);
5748 if (jetsam_reason == OS_REASON_NULL) {
5749 printf("memorystatus_kill_hiwat_proc: failed to allocate exit reason\n");
5750 }
5751
5752 proc_list_lock();
5753
5754 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5755 while (next_p) {
5756 uint64_t footprint_in_bytes = 0;
5757 uint64_t memlimit_in_bytes = 0;
5758 boolean_t skip = 0;
5759
5760 p = next_p;
5761 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5762
5763 aPid = p->p_pid;
5764 aPid_ep = p->p_memstat_effectivepriority;
5765
5766 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5767 continue;
5768 }
5769
5770 /* skip if no limit set */
5771 if (p->p_memstat_memlimit <= 0) {
5772 continue;
5773 }
5774
5775 footprint_in_bytes = get_task_phys_footprint(p->task);
5776 memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
5777 skip = (footprint_in_bytes <= memlimit_in_bytes);
5778
5779 #if CONFIG_FREEZE
5780 if (!skip) {
5781 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
5782 skip = TRUE;
5783 } else {
5784 skip = FALSE;
5785 }
5786 }
5787 #endif
5788
5789 if (skip) {
5790 continue;
5791 } else {
5792 if (memorystatus_jetsam_snapshot_count == 0) {
5793 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5794 new_snapshot = TRUE;
5795 }
5796
5797 if (proc_ref_locked(p) == p) {
5798 /*
5799 * Mark as terminated so that if exit1() indicates success, but the process (for example)
5800 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
5801 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
5802 * acquisition of the proc lock.
5803 */
5804 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5805
5806 proc_list_unlock();
5807 } else {
5808 /*
5809 * We need to restart the search again because
5810 * proc_ref_locked _can_ drop the proc_list lock
5811 * and we could have lost our stored next_p via
5812 * an exit() on another core.
5813 */
5814 i = 0;
5815 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5816 continue;
5817 }
5818
5819 footprint_in_bytes = 0;
5820 freed_mem = memorystatus_kill_proc(p, kMemorystatusKilledHiwat, jetsam_reason, &killed, &footprint_in_bytes); /* purged and/or killed 'p' */
5821
5822 /* Success? */
5823 if (freed_mem) {
5824 if (killed == FALSE) {
5825 /* purged 'p'..don't reset HWM candidate count */
5826 *purged = TRUE;
5827
5828 proc_list_lock();
5829 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5830 proc_list_unlock();
5831 } else {
5832 *memory_reclaimed = footprint_in_bytes;
5833 }
5834 proc_rele(p);
5835 goto exit;
5836 }
5837 /*
5838 * Failure - first unwind the state,
5839 * then fall through to restart the search.
5840 */
5841 proc_list_lock();
5842 proc_rele_locked(p);
5843 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5844 p->p_memstat_state |= P_MEMSTAT_ERROR;
5845 *errors += 1;
5846
5847 i = 0;
5848 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5849 }
5850 }
5851
5852 proc_list_unlock();
5853
5854 exit:
5855 os_reason_free(jetsam_reason);
5856
5857 if (!killed) {
5858 *memory_reclaimed = 0;
5859
5860 /* Clear snapshot if freshly captured and no target was found */
5861 if (new_snapshot) {
5862 proc_list_lock();
5863 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5864 proc_list_unlock();
5865 }
5866 }
5867
5868 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_END,
5869 memorystatus_available_pages, killed ? aPid : 0, killed, *memory_reclaimed, 0);
5870
5871 return killed;
5872 }
5873
5874 /*
5875 * Jetsam a process pinned in the elevated band.
5876 *
5877 * Return: true -- a pinned process was jetsammed
5878 * false -- no pinned process was jetsammed
5879 */
5880 boolean_t
5881 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)
5882 {
5883 pid_t aPid = 0;
5884 proc_t p = PROC_NULL, next_p = PROC_NULL;
5885 boolean_t new_snapshot = FALSE, killed = FALSE;
5886 int kill_count = 0;
5887 uint32_t aPid_ep;
5888 uint64_t killtime = 0;
5889 clock_sec_t tv_sec;
5890 clock_usec_t tv_usec;
5891 uint32_t tv_msec;
5892 uint64_t footprint_of_killed_proc = 0;
5893
5894
5895 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5896 memorystatus_available_pages, 0, 0, 0, 0);
5897
5898 #if CONFIG_FREEZE
5899 boolean_t consider_frozen_only = FALSE;
5900
5901 if (band == (unsigned int) memorystatus_freeze_jetsam_band) {
5902 consider_frozen_only = TRUE;
5903 }
5904 #endif /* CONFIG_FREEZE */
5905
5906 proc_list_lock();
5907
5908 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
5909 while (next_p) {
5910 p = next_p;
5911 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
5912
5913 aPid = p->p_pid;
5914 aPid_ep = p->p_memstat_effectivepriority;
5915
5916 /*
5917 * Only pick a process pinned in this elevated band
5918 */
5919 if (!(p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
5920 continue;
5921 }
5922
5923 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5924 continue;
5925 }
5926
5927 #if CONFIG_FREEZE
5928 if (consider_frozen_only && !(p->p_memstat_state & P_MEMSTAT_FROZEN)) {
5929 continue;
5930 }
5931
5932 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
5933 continue;
5934 }
5935 #endif /* CONFIG_FREEZE */
5936
5937 #if DEVELOPMENT || DEBUG
5938 MEMORYSTATUS_DEBUG(1, "jetsam: elevated%d process pid %d [%s] - memorystatus_available_pages: %d\n",
5939 aggr_count,
5940 aPid, (*p->p_name ? p->p_name : "unknown"),
5941 memorystatus_available_pages);
5942 #endif /* DEVELOPMENT || DEBUG */
5943
5944 if (memorystatus_jetsam_snapshot_count == 0) {
5945 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5946 new_snapshot = TRUE;
5947 }
5948
5949 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5950
5951 killtime = mach_absolute_time();
5952 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5953 tv_msec = tv_usec / 1000;
5954
5955 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5956
5957 if (proc_ref_locked(p) == p) {
5958 proc_list_unlock();
5959
5960 /*
5961 * memorystatus_do_kill drops a reference, so take another one so we can
5962 * continue to use this exit reason even after memorystatus_do_kill()
5963 * returns
5964 */
5965 os_reason_ref(jetsam_reason);
5966 killed = memorystatus_do_kill(p, cause, jetsam_reason, &footprint_of_killed_proc);
5967
5968 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",
5969 (unsigned long)tv_sec, tv_msec,
5970 aggr_count,
5971 aPid, ((p && *p->p_name) ? p->p_name : "unknown"),
5972 memorystatus_kill_cause_name[cause], aPid_ep,
5973 footprint_of_killed_proc >> 10, (uint64_t)memorystatus_available_pages);
5974
5975 /* Success? */
5976 if (killed) {
5977 *memory_reclaimed = footprint_of_killed_proc;
5978 proc_rele(p);
5979 kill_count++;
5980 goto exit;
5981 }
5982
5983 /*
5984 * Failure - first unwind the state,
5985 * then fall through to restart the search.
5986 */
5987 proc_list_lock();
5988 proc_rele_locked(p);
5989 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5990 p->p_memstat_state |= P_MEMSTAT_ERROR;
5991 *errors += 1;
5992 }
5993
5994 /*
5995 * Failure - restart the search.
5996 *
5997 * We might have raced with "p" exiting on another core, resulting in no
5998 * ref on "p". Or, we may have failed to kill "p".
5999 *
6000 * Either way, we fall thru to here, leaving the proc in the
6001 * P_MEMSTAT_TERMINATED state or P_MEMSTAT_ERROR state.
6002 *
6003 * And, we hold the the proc_list_lock at this point.
6004 */
6005
6006 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
6007 }
6008
6009 proc_list_unlock();
6010
6011 exit:
6012 os_reason_free(jetsam_reason);
6013
6014 if (kill_count == 0) {
6015 *memory_reclaimed = 0;
6016
6017 /* Clear snapshot if freshly captured and no target was found */
6018 if (new_snapshot) {
6019 proc_list_lock();
6020 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6021 proc_list_unlock();
6022 }
6023 }
6024
6025 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
6026 memorystatus_available_pages, killed ? aPid : 0, kill_count, *memory_reclaimed, 0);
6027
6028 return killed;
6029 }
6030
6031 static boolean_t
6032 memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause)
6033 {
6034 /*
6035 * TODO: allow a general async path
6036 *
6037 * NOTE: If a new async kill cause is added, make sure to update memorystatus_thread() to
6038 * add the appropriate exit reason code mapping.
6039 */
6040 if ((victim_pid != -1) ||
6041 (cause != kMemorystatusKilledVMPageShortage &&
6042 cause != kMemorystatusKilledVMCompressorThrashing &&
6043 cause != kMemorystatusKilledVMCompressorSpaceShortage &&
6044 cause != kMemorystatusKilledFCThrashing &&
6045 cause != kMemorystatusKilledZoneMapExhaustion)) {
6046 return FALSE;
6047 }
6048
6049 kill_under_pressure_cause = cause;
6050 memorystatus_thread_wake();
6051 return TRUE;
6052 }
6053
6054 boolean_t
6055 memorystatus_kill_on_VM_compressor_space_shortage(boolean_t async)
6056 {
6057 if (async) {
6058 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMCompressorSpaceShortage);
6059 } else {
6060 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMCOMPRESSOR_SPACE_SHORTAGE);
6061 if (jetsam_reason == OS_REASON_NULL) {
6062 printf("memorystatus_kill_on_VM_compressor_space_shortage -- sync: failed to allocate jetsam reason\n");
6063 }
6064
6065 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMCompressorSpaceShortage, jetsam_reason);
6066 }
6067 }
6068
6069 #if CONFIG_JETSAM
6070 boolean_t
6071 memorystatus_kill_on_VM_compressor_thrashing(boolean_t async)
6072 {
6073 if (async) {
6074 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMCompressorThrashing);
6075 } else {
6076 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMCOMPRESSOR_THRASHING);
6077 if (jetsam_reason == OS_REASON_NULL) {
6078 printf("memorystatus_kill_on_VM_compressor_thrashing -- sync: failed to allocate jetsam reason\n");
6079 }
6080
6081 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMCompressorThrashing, jetsam_reason);
6082 }
6083 }
6084
6085 boolean_t
6086 memorystatus_kill_on_VM_page_shortage(boolean_t async)
6087 {
6088 if (async) {
6089 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMPageShortage);
6090 } else {
6091 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMPAGESHORTAGE);
6092 if (jetsam_reason == OS_REASON_NULL) {
6093 printf("memorystatus_kill_on_VM_page_shortage -- sync: failed to allocate jetsam reason\n");
6094 }
6095
6096 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMPageShortage, jetsam_reason);
6097 }
6098 }
6099
6100 boolean_t
6101 memorystatus_kill_on_FC_thrashing(boolean_t async)
6102 {
6103 if (async) {
6104 return memorystatus_kill_process_async(-1, kMemorystatusKilledFCThrashing);
6105 } else {
6106 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_FCTHRASHING);
6107 if (jetsam_reason == OS_REASON_NULL) {
6108 printf("memorystatus_kill_on_FC_thrashing -- sync: failed to allocate jetsam reason\n");
6109 }
6110
6111 return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing, jetsam_reason);
6112 }
6113 }
6114
6115 boolean_t
6116 memorystatus_kill_on_vnode_limit(void)
6117 {
6118 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_VNODE);
6119 if (jetsam_reason == OS_REASON_NULL) {
6120 printf("memorystatus_kill_on_vnode_limit: failed to allocate jetsam reason\n");
6121 }
6122
6123 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes, jetsam_reason);
6124 }
6125
6126 #endif /* CONFIG_JETSAM */
6127
6128 boolean_t
6129 memorystatus_kill_on_zone_map_exhaustion(pid_t pid)
6130 {
6131 boolean_t res = FALSE;
6132 if (pid == -1) {
6133 res = memorystatus_kill_process_async(-1, kMemorystatusKilledZoneMapExhaustion);
6134 } else {
6135 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_ZONE_MAP_EXHAUSTION);
6136 if (jetsam_reason == OS_REASON_NULL) {
6137 printf("memorystatus_kill_on_zone_map_exhaustion: failed to allocate jetsam reason\n");
6138 }
6139
6140 res = memorystatus_kill_process_sync(pid, kMemorystatusKilledZoneMapExhaustion, jetsam_reason);
6141 }
6142 return res;
6143 }
6144
6145 void
6146 memorystatus_on_pageout_scan_end(void)
6147 {
6148 /* No-op */
6149 }
6150
6151 /* Return both allocated and actual size, since there's a race between allocation and list compilation */
6152 static int
6153 memorystatus_get_priority_list(memorystatus_priority_entry_t **list_ptr, size_t *buffer_size, size_t *list_size, boolean_t size_only)
6154 {
6155 uint32_t list_count, i = 0;
6156 memorystatus_priority_entry_t *list_entry;
6157 proc_t p;
6158
6159 list_count = memorystatus_list_count;
6160 *list_size = sizeof(memorystatus_priority_entry_t) * list_count;
6161
6162 /* Just a size check? */
6163 if (size_only) {
6164 return 0;
6165 }
6166
6167 /* Otherwise, validate the size of the buffer */
6168 if (*buffer_size < *list_size) {
6169 return EINVAL;
6170 }
6171
6172 *list_ptr = (memorystatus_priority_entry_t*)kalloc(*list_size);
6173 if (!*list_ptr) {
6174 return ENOMEM;
6175 }
6176
6177 memset(*list_ptr, 0, *list_size);
6178
6179 *buffer_size = *list_size;
6180 *list_size = 0;
6181
6182 list_entry = *list_ptr;
6183
6184 proc_list_lock();
6185
6186 p = memorystatus_get_first_proc_locked(&i, TRUE);
6187 while (p && (*list_size < *buffer_size)) {
6188 list_entry->pid = p->p_pid;
6189 list_entry->priority = p->p_memstat_effectivepriority;
6190 list_entry->user_data = p->p_memstat_userdata;
6191
6192 if (p->p_memstat_memlimit <= 0) {
6193 task_get_phys_footprint_limit(p->task, &list_entry->limit);
6194 } else {
6195 list_entry->limit = p->p_memstat_memlimit;
6196 }
6197
6198 list_entry->state = memorystatus_build_state(p);
6199 list_entry++;
6200
6201 *list_size += sizeof(memorystatus_priority_entry_t);
6202
6203 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6204 }
6205
6206 proc_list_unlock();
6207
6208 MEMORYSTATUS_DEBUG(1, "memorystatus_get_priority_list: returning %lu for size\n", (unsigned long)*list_size);
6209
6210 return 0;
6211 }
6212
6213 static int
6214 memorystatus_get_priority_pid(pid_t pid, user_addr_t buffer, size_t buffer_size)
6215 {
6216 int error = 0;
6217 memorystatus_priority_entry_t mp_entry;
6218 kern_return_t ret;
6219
6220 /* Validate inputs */
6221 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_entry_t))) {
6222 return EINVAL;
6223 }
6224
6225 proc_t p = proc_find(pid);
6226 if (!p) {
6227 return ESRCH;
6228 }
6229
6230 memset(&mp_entry, 0, sizeof(memorystatus_priority_entry_t));
6231
6232 mp_entry.pid = p->p_pid;
6233 mp_entry.priority = p->p_memstat_effectivepriority;
6234 mp_entry.user_data = p->p_memstat_userdata;
6235 if (p->p_memstat_memlimit <= 0) {
6236 ret = task_get_phys_footprint_limit(p->task, &mp_entry.limit);
6237 if (ret != KERN_SUCCESS) {
6238 proc_rele(p);
6239 return EINVAL;
6240 }
6241 } else {
6242 mp_entry.limit = p->p_memstat_memlimit;
6243 }
6244 mp_entry.state = memorystatus_build_state(p);
6245
6246 proc_rele(p);
6247
6248 error = copyout(&mp_entry, buffer, buffer_size);
6249
6250 return error;
6251 }
6252
6253 static int
6254 memorystatus_cmd_get_priority_list(pid_t pid, user_addr_t buffer, size_t buffer_size, int32_t *retval)
6255 {
6256 int error = 0;
6257 boolean_t size_only;
6258 size_t list_size;
6259
6260 /*
6261 * When a non-zero pid is provided, the 'list' has only one entry.
6262 */
6263
6264 size_only = ((buffer == USER_ADDR_NULL) ? TRUE: FALSE);
6265
6266 if (pid != 0) {
6267 list_size = sizeof(memorystatus_priority_entry_t) * 1;
6268 if (!size_only) {
6269 error = memorystatus_get_priority_pid(pid, buffer, buffer_size);
6270 }
6271 } else {
6272 memorystatus_priority_entry_t *list = NULL;
6273 error = memorystatus_get_priority_list(&list, &buffer_size, &list_size, size_only);
6274
6275 if (error == 0) {
6276 if (!size_only) {
6277 error = copyout(list, buffer, list_size);
6278 }
6279 }
6280
6281 if (list) {
6282 kfree(list, buffer_size);
6283 }
6284 }
6285
6286 if (error == 0) {
6287 *retval = list_size;
6288 }
6289
6290 return error;
6291 }
6292
6293 static void
6294 memorystatus_clear_errors(void)
6295 {
6296 proc_t p;
6297 unsigned int i = 0;
6298
6299 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_START, 0, 0, 0, 0, 0);
6300
6301 proc_list_lock();
6302
6303 p = memorystatus_get_first_proc_locked(&i, TRUE);
6304 while (p) {
6305 if (p->p_memstat_state & P_MEMSTAT_ERROR) {
6306 p->p_memstat_state &= ~P_MEMSTAT_ERROR;
6307 }
6308 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6309 }
6310
6311 proc_list_unlock();
6312
6313 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_END, 0, 0, 0, 0, 0);
6314 }
6315
6316 #if CONFIG_JETSAM
6317 static void
6318 memorystatus_update_levels_locked(boolean_t critical_only)
6319 {
6320 memorystatus_available_pages_critical = memorystatus_available_pages_critical_base;
6321
6322 /*
6323 * If there's an entry in the first bucket, we have idle processes.
6324 */
6325
6326 memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
6327 if (first_bucket->count) {
6328 memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
6329
6330 if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure) {
6331 /*
6332 * The critical threshold must never exceed the pressure threshold
6333 */
6334 memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
6335 }
6336 }
6337
6338 if (memorystatus_jetsam_policy & kPolicyMoreFree) {
6339 memorystatus_available_pages_critical += memorystatus_policy_more_free_offset_pages;
6340 }
6341
6342 if (critical_only) {
6343 return;
6344 }
6345
6346 #if VM_PRESSURE_EVENTS
6347 memorystatus_available_pages_pressure = pressure_threshold_percentage * (atop_64(max_mem) / 100);
6348 #endif
6349 }
6350
6351 void
6352 memorystatus_fast_jetsam_override(boolean_t enable_override)
6353 {
6354 /* If fast jetsam is not enabled, simply return */
6355 if (!fast_jetsam_enabled) {
6356 return;
6357 }
6358
6359 if (enable_override) {
6360 if ((memorystatus_jetsam_policy & kPolicyMoreFree) == kPolicyMoreFree) {
6361 return;
6362 }
6363 proc_list_lock();
6364 memorystatus_jetsam_policy |= kPolicyMoreFree;
6365 memorystatus_thread_pool_max();
6366 memorystatus_update_levels_locked(TRUE);
6367 proc_list_unlock();
6368 } else {
6369 if ((memorystatus_jetsam_policy & kPolicyMoreFree) == 0) {
6370 return;
6371 }
6372 proc_list_lock();
6373 memorystatus_jetsam_policy &= ~kPolicyMoreFree;
6374 memorystatus_thread_pool_default();
6375 memorystatus_update_levels_locked(TRUE);
6376 proc_list_unlock();
6377 }
6378 }
6379
6380
6381 static int
6382 sysctl_kern_memorystatus_policy_more_free SYSCTL_HANDLER_ARGS
6383 {
6384 #pragma unused(arg1, arg2, oidp)
6385 int error = 0, more_free = 0;
6386
6387 /*
6388 * TODO: Enable this privilege check?
6389 *
6390 * error = priv_check_cred(kauth_cred_get(), PRIV_VM_JETSAM, 0);
6391 * if (error)
6392 * return (error);
6393 */
6394
6395 error = sysctl_handle_int(oidp, &more_free, 0, req);
6396 if (error || !req->newptr) {
6397 return error;
6398 }
6399
6400 if (more_free) {
6401 memorystatus_fast_jetsam_override(true);
6402 } else {
6403 memorystatus_fast_jetsam_override(false);
6404 }
6405
6406 return 0;
6407 }
6408 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_policy_more_free, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
6409 0, 0, &sysctl_kern_memorystatus_policy_more_free, "I", "");
6410
6411 #endif /* CONFIG_JETSAM */
6412
6413 /*
6414 * Get the at_boot snapshot
6415 */
6416 static int
6417 memorystatus_get_at_boot_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6418 {
6419 size_t input_size = *snapshot_size;
6420
6421 /*
6422 * The at_boot snapshot has no entry list.
6423 */
6424 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t);
6425
6426 if (size_only) {
6427 return 0;
6428 }
6429
6430 /*
6431 * Validate the size of the snapshot buffer
6432 */
6433 if (input_size < *snapshot_size) {
6434 return EINVAL;
6435 }
6436
6437 /*
6438 * Update the notification_time only
6439 */
6440 memorystatus_at_boot_snapshot.notification_time = mach_absolute_time();
6441 *snapshot = &memorystatus_at_boot_snapshot;
6442
6443 MEMORYSTATUS_DEBUG(7, "memorystatus_get_at_boot_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%d)\n",
6444 (long)input_size, (long)*snapshot_size, 0);
6445 return 0;
6446 }
6447
6448 /*
6449 * Get the previous fully populated snapshot
6450 */
6451 static int
6452 memorystatus_get_jetsam_snapshot_copy(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6453 {
6454 size_t input_size = *snapshot_size;
6455
6456 if (memorystatus_jetsam_snapshot_copy_count > 0) {
6457 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_copy_count));
6458 } else {
6459 *snapshot_size = 0;
6460 }
6461
6462 if (size_only) {
6463 return 0;
6464 }
6465
6466 if (input_size < *snapshot_size) {
6467 return EINVAL;
6468 }
6469
6470 *snapshot = memorystatus_jetsam_snapshot_copy;
6471
6472 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot_copy: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6473 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_copy_count);
6474
6475 return 0;
6476 }
6477
6478 static int
6479 memorystatus_get_on_demand_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6480 {
6481 size_t input_size = *snapshot_size;
6482 uint32_t ods_list_count = memorystatus_list_count;
6483 memorystatus_jetsam_snapshot_t *ods = NULL; /* The on_demand snapshot buffer */
6484
6485 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (ods_list_count));
6486
6487 if (size_only) {
6488 return 0;
6489 }
6490
6491 /*
6492 * Validate the size of the snapshot buffer.
6493 * This is inherently racey. May want to revisit
6494 * this error condition and trim the output when
6495 * it doesn't fit.
6496 */
6497 if (input_size < *snapshot_size) {
6498 return EINVAL;
6499 }
6500
6501 /*
6502 * Allocate and initialize a snapshot buffer.
6503 */
6504 ods = (memorystatus_jetsam_snapshot_t *)kalloc(*snapshot_size);
6505 if (!ods) {
6506 return ENOMEM;
6507 }
6508
6509 memset(ods, 0, *snapshot_size);
6510
6511 proc_list_lock();
6512 memorystatus_init_jetsam_snapshot_locked(ods, ods_list_count);
6513 proc_list_unlock();
6514
6515 /*
6516 * Return the kernel allocated, on_demand buffer.
6517 * The caller of this routine will copy the data out
6518 * to user space and then free the kernel allocated
6519 * buffer.
6520 */
6521 *snapshot = ods;
6522
6523 MEMORYSTATUS_DEBUG(7, "memorystatus_get_on_demand_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6524 (long)input_size, (long)*snapshot_size, (long)ods_list_count);
6525
6526 return 0;
6527 }
6528
6529 static int
6530 memorystatus_get_jetsam_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
6531 {
6532 size_t input_size = *snapshot_size;
6533
6534 if (memorystatus_jetsam_snapshot_count > 0) {
6535 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count));
6536 } else {
6537 *snapshot_size = 0;
6538 }
6539
6540 if (size_only) {
6541 return 0;
6542 }
6543
6544 if (input_size < *snapshot_size) {
6545 return EINVAL;
6546 }
6547
6548 *snapshot = memorystatus_jetsam_snapshot;
6549
6550 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
6551 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_count);
6552
6553 return 0;
6554 }
6555
6556
6557 static int
6558 memorystatus_cmd_get_jetsam_snapshot(int32_t flags, user_addr_t buffer, size_t buffer_size, int32_t *retval)
6559 {
6560 int error = EINVAL;
6561 boolean_t size_only;
6562 boolean_t is_default_snapshot = FALSE;
6563 boolean_t is_on_demand_snapshot = FALSE;
6564 boolean_t is_at_boot_snapshot = FALSE;
6565 memorystatus_jetsam_snapshot_t *snapshot;
6566
6567 size_only = ((buffer == USER_ADDR_NULL) ? TRUE : FALSE);
6568
6569 if (flags == 0) {
6570 /* Default */
6571 is_default_snapshot = TRUE;
6572 error = memorystatus_get_jetsam_snapshot(&snapshot, &buffer_size, size_only);
6573 } else {
6574 if (flags & ~(MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT | MEMORYSTATUS_SNAPSHOT_COPY)) {
6575 /*
6576 * Unsupported bit set in flag.
6577 */
6578 return EINVAL;
6579 }
6580
6581 if (flags & (flags - 0x1)) {
6582 /*
6583 * Can't have multiple flags set at the same time.
6584 */
6585 return EINVAL;
6586 }
6587
6588 if (flags & MEMORYSTATUS_SNAPSHOT_ON_DEMAND) {
6589 is_on_demand_snapshot = TRUE;
6590 /*
6591 * When not requesting the size only, the following call will allocate
6592 * an on_demand snapshot buffer, which is freed below.
6593 */
6594 error = memorystatus_get_on_demand_snapshot(&snapshot, &buffer_size, size_only);
6595 } else if (flags & MEMORYSTATUS_SNAPSHOT_AT_BOOT) {
6596 is_at_boot_snapshot = TRUE;
6597 error = memorystatus_get_at_boot_snapshot(&snapshot, &buffer_size, size_only);
6598 } else if (flags & MEMORYSTATUS_SNAPSHOT_COPY) {
6599 error = memorystatus_get_jetsam_snapshot_copy(&snapshot, &buffer_size, size_only);
6600 } else {
6601 /*
6602 * Invalid flag setting.
6603 */
6604 return EINVAL;
6605 }
6606 }
6607
6608 if (error) {
6609 goto out;
6610 }
6611
6612 /*
6613 * Copy the data out to user space and clear the snapshot buffer.
6614 * If working with the jetsam snapshot,
6615 * clearing the buffer means, reset the count.
6616 * If working with an on_demand snapshot
6617 * clearing the buffer means, free it.
6618 * If working with the at_boot snapshot
6619 * there is nothing to clear or update.
6620 * If working with a copy of the snapshot
6621 * there is nothing to clear or update.
6622 */
6623 if (!size_only) {
6624 if ((error = copyout(snapshot, buffer, buffer_size)) == 0) {
6625 if (is_default_snapshot) {
6626 /*
6627 * The jetsam snapshot is never freed, its count is simply reset.
6628 * However, we make a copy for any parties that might be interested
6629 * in the previous fully populated snapshot.
6630 */
6631 proc_list_lock();
6632 memcpy(memorystatus_jetsam_snapshot_copy, memorystatus_jetsam_snapshot, memorystatus_jetsam_snapshot_size);
6633 memorystatus_jetsam_snapshot_copy_count = memorystatus_jetsam_snapshot_count;
6634 snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6635 memorystatus_jetsam_snapshot_last_timestamp = 0;
6636 proc_list_unlock();
6637 }
6638 }
6639
6640 if (is_on_demand_snapshot) {
6641 /*
6642 * The on_demand snapshot is always freed,
6643 * even if the copyout failed.
6644 */
6645 if (snapshot) {
6646 kfree(snapshot, buffer_size);
6647 }
6648 }
6649 }
6650
6651 if (error == 0) {
6652 *retval = buffer_size;
6653 }
6654 out:
6655 return error;
6656 }
6657
6658 /*
6659 * Routine: memorystatus_cmd_grp_set_priorities
6660 * Purpose: Update priorities for a group of processes.
6661 *
6662 * [priority]
6663 * Move each process out of its effective priority
6664 * band and into a new priority band.
6665 * Maintains relative order from lowest to highest priority.
6666 * In single band, maintains relative order from head to tail.
6667 *
6668 * eg: before [effectivepriority | pid]
6669 * [18 | p101 ]
6670 * [17 | p55, p67, p19 ]
6671 * [12 | p103 p10 ]
6672 * [ 7 | p25 ]
6673 * [ 0 | p71, p82, ]
6674 *
6675 * after [ new band | pid]
6676 * [ xxx | p71, p82, p25, p103, p10, p55, p67, p19, p101]
6677 *
6678 * Returns: 0 on success, else non-zero.
6679 *
6680 * Caveat: We know there is a race window regarding recycled pids.
6681 * A process could be killed before the kernel can act on it here.
6682 * If a pid cannot be found in any of the jetsam priority bands,
6683 * then we simply ignore it. No harm.
6684 * But, if the pid has been recycled then it could be an issue.
6685 * In that scenario, we might move an unsuspecting process to the new
6686 * priority band. It's not clear how the kernel can safeguard
6687 * against this, but it would be an extremely rare case anyway.
6688 * The caller of this api might avoid such race conditions by
6689 * ensuring that the processes passed in the pid list are suspended.
6690 */
6691
6692
6693 static int
6694 memorystatus_cmd_grp_set_priorities(user_addr_t buffer, size_t buffer_size)
6695 {
6696 /*
6697 * We only handle setting priority
6698 * per process
6699 */
6700
6701 int error = 0;
6702 memorystatus_properties_entry_v1_t *entries = NULL;
6703 uint32_t entry_count = 0;
6704
6705 /* This will be the ordered proc list */
6706 typedef struct memorystatus_internal_properties {
6707 proc_t proc;
6708 int32_t priority;
6709 } memorystatus_internal_properties_t;
6710
6711 memorystatus_internal_properties_t *table = NULL;
6712 size_t table_size = 0;
6713 uint32_t table_count = 0;
6714
6715 uint32_t i = 0;
6716 uint32_t bucket_index = 0;
6717 boolean_t head_insert;
6718 int32_t new_priority;
6719
6720 proc_t p;
6721
6722 /* Verify inputs */
6723 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
6724 error = EINVAL;
6725 goto out;
6726 }
6727
6728 entry_count = (buffer_size / sizeof(memorystatus_properties_entry_v1_t));
6729 if ((entries = (memorystatus_properties_entry_v1_t *)kalloc(buffer_size)) == NULL) {
6730 error = ENOMEM;
6731 goto out;
6732 }
6733
6734 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);
6735
6736 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
6737 goto out;
6738 }
6739
6740 /* Verify sanity of input priorities */
6741 if (entries[0].version == MEMORYSTATUS_MPE_VERSION_1) {
6742 if ((buffer_size % MEMORYSTATUS_MPE_VERSION_1_SIZE) != 0) {
6743 error = EINVAL;
6744 goto out;
6745 }
6746 } else {
6747 error = EINVAL;
6748 goto out;
6749 }
6750
6751 for (i = 0; i < entry_count; i++) {
6752 if (entries[i].priority == -1) {
6753 /* Use as shorthand for default priority */
6754 entries[i].priority = JETSAM_PRIORITY_DEFAULT;
6755 } else if ((entries[i].priority == system_procs_aging_band) || (entries[i].priority == applications_aging_band)) {
6756 /* Both the aging bands are reserved for internal use;
6757 * if requested, adjust to JETSAM_PRIORITY_IDLE. */
6758 entries[i].priority = JETSAM_PRIORITY_IDLE;
6759 } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
6760 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle
6761 * queue */
6762 /* Deal with this later */
6763 } else if ((entries[i].priority < 0) || (entries[i].priority >= MEMSTAT_BUCKET_COUNT)) {
6764 /* Sanity check */
6765 error = EINVAL;
6766 goto out;
6767 }
6768 }
6769
6770 table_size = sizeof(memorystatus_internal_properties_t) * entry_count;
6771 if ((table = (memorystatus_internal_properties_t *)kalloc(table_size)) == NULL) {
6772 error = ENOMEM;
6773 goto out;
6774 }
6775 memset(table, 0, table_size);
6776
6777
6778 /*
6779 * For each jetsam bucket entry, spin through the input property list.
6780 * When a matching pid is found, populate an adjacent table with the
6781 * appropriate proc pointer and new property values.
6782 * This traversal automatically preserves order from lowest
6783 * to highest priority.
6784 */
6785
6786 bucket_index = 0;
6787
6788 proc_list_lock();
6789
6790 /* Create the ordered table */
6791 p = memorystatus_get_first_proc_locked(&bucket_index, TRUE);
6792 while (p && (table_count < entry_count)) {
6793 for (i = 0; i < entry_count; i++) {
6794 if (p->p_pid == entries[i].pid) {
6795 /* Build the table data */
6796 table[table_count].proc = p;
6797 table[table_count].priority = entries[i].priority;
6798 table_count++;
6799 break;
6800 }
6801 }
6802 p = memorystatus_get_next_proc_locked(&bucket_index, p, TRUE);
6803 }
6804
6805 /* We now have ordered list of procs ready to move */
6806 for (i = 0; i < table_count; i++) {
6807 p = table[i].proc;
6808 assert(p != NULL);
6809
6810 /* Allow head inserts -- but relative order is now */
6811 if (table[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
6812 new_priority = JETSAM_PRIORITY_IDLE;
6813 head_insert = true;
6814 } else {
6815 new_priority = table[i].priority;
6816 head_insert = false;
6817 }
6818
6819 /* Not allowed */
6820 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
6821 continue;
6822 }
6823
6824 /*
6825 * Take appropriate steps if moving proc out of
6826 * either of the aging bands.
6827 */
6828 if ((p->p_memstat_effectivepriority == system_procs_aging_band) || (p->p_memstat_effectivepriority == applications_aging_band)) {
6829 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
6830 }
6831
6832 memorystatus_update_priority_locked(p, new_priority, head_insert, false);
6833 }
6834
6835 proc_list_unlock();
6836
6837 /*
6838 * if (table_count != entry_count)
6839 * then some pids were not found in a jetsam band.
6840 * harmless but interesting...
6841 */
6842 out:
6843 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);
6844
6845 if (entries) {
6846 kfree(entries, buffer_size);
6847 }
6848 if (table) {
6849 kfree(table, table_size);
6850 }
6851
6852 return error;
6853 }
6854
6855 memorystatus_internal_probabilities_t *memorystatus_global_probabilities_table = NULL;
6856 size_t memorystatus_global_probabilities_size = 0;
6857
6858 static int
6859 memorystatus_cmd_grp_set_probabilities(user_addr_t buffer, size_t buffer_size)
6860 {
6861 int error = 0;
6862 memorystatus_properties_entry_v1_t *entries = NULL;
6863 uint32_t entry_count = 0, i = 0;
6864 memorystatus_internal_probabilities_t *tmp_table_new = NULL, *tmp_table_old = NULL;
6865 size_t tmp_table_new_size = 0, tmp_table_old_size = 0;
6866
6867 /* Verify inputs */
6868 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
6869 error = EINVAL;
6870 goto out;
6871 }
6872
6873 entry_count = (buffer_size / sizeof(memorystatus_properties_entry_v1_t));
6874
6875 if ((entries = (memorystatus_properties_entry_v1_t *) kalloc(buffer_size)) == NULL) {
6876 error = ENOMEM;
6877 goto out;
6878 }
6879
6880 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);
6881
6882 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
6883 goto out;
6884 }
6885
6886 if (entries[0].version == MEMORYSTATUS_MPE_VERSION_1) {
6887 if ((buffer_size % MEMORYSTATUS_MPE_VERSION_1_SIZE) != 0) {
6888 error = EINVAL;
6889 goto out;
6890 }
6891 } else {
6892 error = EINVAL;
6893 goto out;
6894 }
6895
6896 /* Verify sanity of input priorities */
6897 for (i = 0; i < entry_count; i++) {
6898 /*
6899 * 0 - low probability of use.
6900 * 1 - high probability of use.
6901 *
6902 * Keeping this field an int (& not a bool) to allow
6903 * us to experiment with different values/approaches
6904 * later on.
6905 */
6906 if (entries[i].use_probability > 1) {
6907 error = EINVAL;
6908 goto out;
6909 }
6910 }
6911
6912 tmp_table_new_size = sizeof(memorystatus_internal_probabilities_t) * entry_count;
6913
6914 if ((tmp_table_new = (memorystatus_internal_probabilities_t *) kalloc(tmp_table_new_size)) == NULL) {
6915 error = ENOMEM;
6916 goto out;
6917 }
6918 memset(tmp_table_new, 0, tmp_table_new_size);
6919
6920 proc_list_lock();
6921
6922 if (memorystatus_global_probabilities_table) {
6923 tmp_table_old = memorystatus_global_probabilities_table;
6924 tmp_table_old_size = memorystatus_global_probabilities_size;
6925 }
6926
6927 memorystatus_global_probabilities_table = tmp_table_new;
6928 memorystatus_global_probabilities_size = tmp_table_new_size;
6929 tmp_table_new = NULL;
6930
6931 for (i = 0; i < entry_count; i++) {
6932 /* Build the table data */
6933 strlcpy(memorystatus_global_probabilities_table[i].proc_name, entries[i].proc_name, MAXCOMLEN + 1);
6934 memorystatus_global_probabilities_table[i].use_probability = entries[i].use_probability;
6935 }
6936
6937 proc_list_unlock();
6938
6939 out:
6940 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);
6941
6942 if (entries) {
6943 kfree(entries, buffer_size);
6944 entries = NULL;
6945 }
6946
6947 if (tmp_table_old) {
6948 kfree(tmp_table_old, tmp_table_old_size);
6949 tmp_table_old = NULL;
6950 }
6951
6952 return error;
6953 }
6954
6955 static int
6956 memorystatus_cmd_grp_set_properties(int32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
6957 {
6958 int error = 0;
6959
6960 if ((flags & MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY) == MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY) {
6961 error = memorystatus_cmd_grp_set_priorities(buffer, buffer_size);
6962 } else if ((flags & MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY) == MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY) {
6963 error = memorystatus_cmd_grp_set_probabilities(buffer, buffer_size);
6964 } else {
6965 error = EINVAL;
6966 }
6967
6968 return error;
6969 }
6970
6971 /*
6972 * This routine is used to update a process's jetsam priority position and stored user_data.
6973 * It is not used for the setting of memory limits, which is why the last 6 args to the
6974 * memorystatus_update() call are 0 or FALSE.
6975 *
6976 * Flags passed into this call are used to distinguish the motivation behind a jetsam priority
6977 * transition. By default, the kernel updates the process's original requested priority when
6978 * no flag is passed. But when the MEMORYSTATUS_SET_PRIORITY_ASSERTION flag is used, the kernel
6979 * updates the process's assertion driven priority.
6980 *
6981 * The assertion flag was introduced for use by the device's assertion mediator (eg: runningboardd).
6982 * When an assertion is controlling a process's jetsam priority, it may conflict with that process's
6983 * dirty/clean (active/inactive) jetsam state. The kernel attempts to resolve a priority transition
6984 * conflict by reviewing the process state and then choosing the maximum jetsam band at play,
6985 * eg: requested priority versus assertion priority.
6986 */
6987
6988 static int
6989 memorystatus_cmd_set_priority_properties(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
6990 {
6991 int error = 0;
6992 boolean_t is_assertion = FALSE; /* priority is driven by an assertion */
6993 memorystatus_priority_properties_t mpp_entry;
6994
6995 /* Validate inputs */
6996 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_properties_t))) {
6997 return EINVAL;
6998 }
6999
7000 /* Validate flags */
7001 if (flags == 0) {
7002 /*
7003 * Default. This path updates requestedpriority.
7004 */
7005 } else {
7006 if (flags & ~(MEMORYSTATUS_SET_PRIORITY_ASSERTION)) {
7007 /*
7008 * Unsupported bit set in flag.
7009 */
7010 return EINVAL;
7011 } else if (flags & MEMORYSTATUS_SET_PRIORITY_ASSERTION) {
7012 is_assertion = TRUE;
7013 }
7014 }
7015
7016 error = copyin(buffer, &mpp_entry, buffer_size);
7017
7018 if (error == 0) {
7019 proc_t p;
7020
7021 p = proc_find(pid);
7022 if (!p) {
7023 return ESRCH;
7024 }
7025
7026 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
7027 proc_rele(p);
7028 return EPERM;
7029 }
7030
7031 if (is_assertion) {
7032 os_log(OS_LOG_DEFAULT, "memorystatus: set assertion priority(%d) target %s:%d\n",
7033 mpp_entry.priority, (*p->p_name ? p->p_name : "unknown"), p->p_pid);
7034 }
7035
7036 error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, is_assertion, FALSE, FALSE, 0, 0, FALSE, FALSE);
7037 proc_rele(p);
7038 }
7039
7040 return error;
7041 }
7042
7043 static int
7044 memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7045 {
7046 int error = 0;
7047 memorystatus_memlimit_properties_t mmp_entry;
7048
7049 /* Validate inputs */
7050 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
7051 return EINVAL;
7052 }
7053
7054 error = copyin(buffer, &mmp_entry, buffer_size);
7055
7056 if (error == 0) {
7057 error = memorystatus_set_memlimit_properties(pid, &mmp_entry);
7058 }
7059
7060 return error;
7061 }
7062
7063 static void
7064 memorystatus_get_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t* p_entry)
7065 {
7066 memset(p_entry, 0, sizeof(memorystatus_memlimit_properties_t));
7067
7068 if (p->p_memstat_memlimit_active > 0) {
7069 p_entry->memlimit_active = p->p_memstat_memlimit_active;
7070 } else {
7071 task_convert_phys_footprint_limit(-1, &p_entry->memlimit_active);
7072 }
7073
7074 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) {
7075 p_entry->memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7076 }
7077
7078 /*
7079 * Get the inactive limit and attributes
7080 */
7081 if (p->p_memstat_memlimit_inactive <= 0) {
7082 task_convert_phys_footprint_limit(-1, &p_entry->memlimit_inactive);
7083 } else {
7084 p_entry->memlimit_inactive = p->p_memstat_memlimit_inactive;
7085 }
7086 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) {
7087 p_entry->memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7088 }
7089 }
7090
7091 /*
7092 * When getting the memlimit settings, we can't simply call task_get_phys_footprint_limit().
7093 * That gets the proc's cached memlimit and there is no guarantee that the active/inactive
7094 * limits will be the same in the no-limit case. Instead we convert limits <= 0 using
7095 * task_convert_phys_footprint_limit(). It computes the same limit value that would be written
7096 * to the task's ledgers via task_set_phys_footprint_limit().
7097 */
7098 static int
7099 memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7100 {
7101 memorystatus_memlimit_properties2_t mmp_entry;
7102
7103 /* Validate inputs */
7104 if ((pid == 0) || (buffer == USER_ADDR_NULL) ||
7105 ((buffer_size != sizeof(memorystatus_memlimit_properties_t)) &&
7106 (buffer_size != sizeof(memorystatus_memlimit_properties2_t)))) {
7107 return EINVAL;
7108 }
7109
7110 memset(&mmp_entry, 0, sizeof(memorystatus_memlimit_properties2_t));
7111
7112 proc_t p = proc_find(pid);
7113 if (!p) {
7114 return ESRCH;
7115 }
7116
7117 /*
7118 * Get the active limit and attributes.
7119 * No locks taken since we hold a reference to the proc.
7120 */
7121
7122 memorystatus_get_memlimit_properties_internal(p, &mmp_entry.v1);
7123
7124 #if CONFIG_JETSAM
7125 #if DEVELOPMENT || DEBUG
7126 /*
7127 * Get the limit increased via SPI
7128 */
7129 mmp_entry.memlimit_increase = roundToNearestMB(p->p_memlimit_increase);
7130 mmp_entry.memlimit_increase_bytes = p->p_memlimit_increase;
7131 #endif /* DEVELOPMENT || DEBUG */
7132 #endif /* CONFIG_JETSAM */
7133
7134 proc_rele(p);
7135
7136 int error = copyout(&mmp_entry, buffer, buffer_size);
7137
7138 return error;
7139 }
7140
7141
7142 /*
7143 * SPI for kbd - pr24956468
7144 * This is a very simple snapshot that calculates how much a
7145 * process's phys_footprint exceeds a specific memory limit.
7146 * Only the inactive memory limit is supported for now.
7147 * The delta is returned as bytes in excess or zero.
7148 */
7149 static int
7150 memorystatus_cmd_get_memlimit_excess_np(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
7151 {
7152 int error = 0;
7153 uint64_t footprint_in_bytes = 0;
7154 uint64_t delta_in_bytes = 0;
7155 int32_t memlimit_mb = 0;
7156 uint64_t memlimit_bytes = 0;
7157
7158 /* Validate inputs */
7159 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(uint64_t)) || (flags != 0)) {
7160 return EINVAL;
7161 }
7162
7163 proc_t p = proc_find(pid);
7164 if (!p) {
7165 return ESRCH;
7166 }
7167
7168 /*
7169 * Get the inactive limit.
7170 * No locks taken since we hold a reference to the proc.
7171 */
7172
7173 if (p->p_memstat_memlimit_inactive <= 0) {
7174 task_convert_phys_footprint_limit(-1, &memlimit_mb);
7175 } else {
7176 memlimit_mb = p->p_memstat_memlimit_inactive;
7177 }
7178
7179 footprint_in_bytes = get_task_phys_footprint(p->task);
7180
7181 proc_rele(p);
7182
7183 memlimit_bytes = memlimit_mb * 1024 * 1024; /* MB to bytes */
7184
7185 /*
7186 * Computed delta always returns >= 0 bytes
7187 */
7188 if (footprint_in_bytes > memlimit_bytes) {
7189 delta_in_bytes = footprint_in_bytes - memlimit_bytes;
7190 }
7191
7192 error = copyout(&delta_in_bytes, buffer, sizeof(delta_in_bytes));
7193
7194 return error;
7195 }
7196
7197
7198 static int
7199 memorystatus_cmd_get_pressure_status(int32_t *retval)
7200 {
7201 int error;
7202
7203 /* Need privilege for check */
7204 error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
7205 if (error) {
7206 return error;
7207 }
7208
7209 /* Inherently racy, so it's not worth taking a lock here */
7210 *retval = (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
7211
7212 return error;
7213 }
7214
7215 int
7216 memorystatus_get_pressure_status_kdp()
7217 {
7218 return (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
7219 }
7220
7221 /*
7222 * Every process, including a P_MEMSTAT_INTERNAL process (currently only pid 1), is allowed to set a HWM.
7223 *
7224 * This call is inflexible -- it does not distinguish between active/inactive, fatal/non-fatal
7225 * So, with 2-level HWM preserving previous behavior will map as follows.
7226 * - treat the limit passed in as both an active and inactive limit.
7227 * - treat the is_fatal_limit flag as though it applies to both active and inactive limits.
7228 *
7229 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK
7230 * - the is_fatal_limit is FALSE, meaning the active and inactive limits are non-fatal/soft
7231 * - so mapping is (active/non-fatal, inactive/non-fatal)
7232 *
7233 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT
7234 * - the is_fatal_limit is TRUE, meaning the process's active and inactive limits are fatal/hard
7235 * - so mapping is (active/fatal, inactive/fatal)
7236 */
7237
7238 #if CONFIG_JETSAM
7239 static int
7240 memorystatus_cmd_set_jetsam_memory_limit(pid_t pid, int32_t high_water_mark, __unused int32_t *retval, boolean_t is_fatal_limit)
7241 {
7242 int error = 0;
7243 memorystatus_memlimit_properties_t entry;
7244
7245 entry.memlimit_active = high_water_mark;
7246 entry.memlimit_active_attr = 0;
7247 entry.memlimit_inactive = high_water_mark;
7248 entry.memlimit_inactive_attr = 0;
7249
7250 if (is_fatal_limit == TRUE) {
7251 entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7252 entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7253 }
7254
7255 error = memorystatus_set_memlimit_properties(pid, &entry);
7256 return error;
7257 }
7258 #endif /* CONFIG_JETSAM */
7259
7260 static int
7261 memorystatus_set_memlimit_properties_internal(proc_t p, memorystatus_memlimit_properties_t *p_entry)
7262 {
7263 int error = 0;
7264
7265 LCK_MTX_ASSERT(proc_list_mlock, LCK_MTX_ASSERT_OWNED);
7266
7267 /*
7268 * Store the active limit variants in the proc.
7269 */
7270 SET_ACTIVE_LIMITS_LOCKED(p, p_entry->memlimit_active, p_entry->memlimit_active_attr);
7271
7272 /*
7273 * Store the inactive limit variants in the proc.
7274 */
7275 SET_INACTIVE_LIMITS_LOCKED(p, p_entry->memlimit_inactive, p_entry->memlimit_inactive_attr);
7276
7277 /*
7278 * Enforce appropriate limit variant by updating the cached values
7279 * and writing the ledger.
7280 * Limit choice is based on process active/inactive state.
7281 */
7282
7283 if (memorystatus_highwater_enabled) {
7284 boolean_t is_fatal;
7285 boolean_t use_active;
7286
7287 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
7288 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
7289 use_active = TRUE;
7290 } else {
7291 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
7292 use_active = FALSE;
7293 }
7294
7295 /* Enforce the limit by writing to the ledgers */
7296 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;
7297
7298 MEMORYSTATUS_DEBUG(3, "memorystatus_set_memlimit_properties: new limit on pid %d (%dMB %s) current priority (%d) dirty_state?=0x%x %s\n",
7299 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
7300 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
7301 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
7302 DTRACE_MEMORYSTATUS2(memorystatus_set_memlimit, proc_t, p, int32_t, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1));
7303 }
7304
7305 return error;
7306 }
7307
7308 static int
7309 memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry)
7310 {
7311 memorystatus_memlimit_properties_t set_entry;
7312
7313 proc_t p = proc_find(pid);
7314 if (!p) {
7315 return ESRCH;
7316 }
7317
7318 /*
7319 * Check for valid attribute flags.
7320 */
7321 const uint32_t valid_attrs = MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7322 if ((entry->memlimit_active_attr & (~valid_attrs)) != 0) {
7323 proc_rele(p);
7324 return EINVAL;
7325 }
7326 if ((entry->memlimit_inactive_attr & (~valid_attrs)) != 0) {
7327 proc_rele(p);
7328 return EINVAL;
7329 }
7330
7331 /*
7332 * Setup the active memlimit properties
7333 */
7334 set_entry.memlimit_active = entry->memlimit_active;
7335 set_entry.memlimit_active_attr = entry->memlimit_active_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7336
7337 /*
7338 * Setup the inactive memlimit properties
7339 */
7340 set_entry.memlimit_inactive = entry->memlimit_inactive;
7341 set_entry.memlimit_inactive_attr = entry->memlimit_inactive_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7342
7343 /*
7344 * Setting a limit of <= 0 implies that the process has no
7345 * high-water-mark and has no per-task-limit. That means
7346 * the system_wide task limit is in place, which by the way,
7347 * is always fatal.
7348 */
7349
7350 if (set_entry.memlimit_active <= 0) {
7351 /*
7352 * Enforce the fatal system_wide task limit while process is active.
7353 */
7354 set_entry.memlimit_active = -1;
7355 set_entry.memlimit_active_attr = MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7356 }
7357 #if CONFIG_JETSAM
7358 #if DEVELOPMENT || DEBUG
7359 else {
7360 /* add the current increase to it, for roots */
7361 set_entry.memlimit_active += roundToNearestMB(p->p_memlimit_increase);
7362 }
7363 #endif /* DEVELOPMENT || DEBUG */
7364 #endif /* CONFIG_JETSAM */
7365
7366 if (set_entry.memlimit_inactive <= 0) {
7367 /*
7368 * Enforce the fatal system_wide task limit while process is inactive.
7369 */
7370 set_entry.memlimit_inactive = -1;
7371 set_entry.memlimit_inactive_attr = MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7372 }
7373 #if CONFIG_JETSAM
7374 #if DEVELOPMENT || DEBUG
7375 else {
7376 /* add the current increase to it, for roots */
7377 set_entry.memlimit_inactive += roundToNearestMB(p->p_memlimit_increase);
7378 }
7379 #endif /* DEVELOPMENT || DEBUG */
7380 #endif /* CONFIG_JETSAM */
7381
7382 proc_list_lock();
7383
7384 int error = memorystatus_set_memlimit_properties_internal(p, &set_entry);
7385
7386 proc_list_unlock();
7387 proc_rele(p);
7388
7389 return error;
7390 }
7391
7392 /*
7393 * Returns the jetsam priority (effective or requested) of the process
7394 * associated with this task.
7395 */
7396 int
7397 proc_get_memstat_priority(proc_t p, boolean_t effective_priority)
7398 {
7399 if (p) {
7400 if (effective_priority) {
7401 return p->p_memstat_effectivepriority;
7402 } else {
7403 return p->p_memstat_requestedpriority;
7404 }
7405 }
7406 return 0;
7407 }
7408
7409 static int
7410 memorystatus_get_process_is_managed(pid_t pid, int *is_managed)
7411 {
7412 proc_t p = NULL;
7413
7414 /* Validate inputs */
7415 if (pid == 0) {
7416 return EINVAL;
7417 }
7418
7419 p = proc_find(pid);
7420 if (!p) {
7421 return ESRCH;
7422 }
7423
7424 proc_list_lock();
7425 *is_managed = ((p->p_memstat_state & P_MEMSTAT_MANAGED) ? 1 : 0);
7426 proc_rele_locked(p);
7427 proc_list_unlock();
7428
7429 return 0;
7430 }
7431
7432 static int
7433 memorystatus_set_process_is_managed(pid_t pid, boolean_t set_managed)
7434 {
7435 proc_t p = NULL;
7436
7437 /* Validate inputs */
7438 if (pid == 0) {
7439 return EINVAL;
7440 }
7441
7442 p = proc_find(pid);
7443 if (!p) {
7444 return ESRCH;
7445 }
7446
7447 proc_list_lock();
7448 if (set_managed == TRUE) {
7449 p->p_memstat_state |= P_MEMSTAT_MANAGED;
7450 /*
7451 * The P_MEMSTAT_MANAGED bit is set by assertiond for Apps.
7452 * Also opt them in to being frozen (they might have started
7453 * off with the P_MEMSTAT_FREEZE_DISABLED bit set.)
7454 */
7455 p->p_memstat_state &= ~P_MEMSTAT_FREEZE_DISABLED;
7456 } else {
7457 p->p_memstat_state &= ~P_MEMSTAT_MANAGED;
7458 }
7459 proc_rele_locked(p);
7460 proc_list_unlock();
7461
7462 return 0;
7463 }
7464
7465 int
7466 memorystatus_control(struct proc *p __unused, struct memorystatus_control_args *args, int *ret)
7467 {
7468 int error = EINVAL;
7469 boolean_t skip_auth_check = FALSE;
7470 os_reason_t jetsam_reason = OS_REASON_NULL;
7471
7472 #if !CONFIG_JETSAM
7473 #pragma unused(ret)
7474 #pragma unused(jetsam_reason)
7475 #endif
7476
7477 /* We don't need entitlements if we're setting/ querying the freeze preference for a process. Skip the check below. */
7478 if (args->command == MEMORYSTATUS_CMD_SET_PROCESS_IS_FREEZABLE || args->command == MEMORYSTATUS_CMD_GET_PROCESS_IS_FREEZABLE) {
7479 skip_auth_check = TRUE;
7480 }
7481
7482 /* Need to be root or have entitlement. */
7483 if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT) && !skip_auth_check) {
7484 error = EPERM;
7485 goto out;
7486 }
7487
7488 /*
7489 * Sanity check.
7490 * Do not enforce it for snapshots.
7491 */
7492 if (args->command != MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT) {
7493 if (args->buffersize > MEMORYSTATUS_BUFFERSIZE_MAX) {
7494 error = EINVAL;
7495 goto out;
7496 }
7497 }
7498
7499 switch (args->command) {
7500 case MEMORYSTATUS_CMD_GET_PRIORITY_LIST:
7501 error = memorystatus_cmd_get_priority_list(args->pid, args->buffer, args->buffersize, ret);
7502 break;
7503 case MEMORYSTATUS_CMD_SET_PRIORITY_PROPERTIES:
7504 error = memorystatus_cmd_set_priority_properties(args->pid, args->flags, args->buffer, args->buffersize, ret);
7505 break;
7506 case MEMORYSTATUS_CMD_SET_MEMLIMIT_PROPERTIES:
7507 error = memorystatus_cmd_set_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
7508 break;
7509 case MEMORYSTATUS_CMD_GET_MEMLIMIT_PROPERTIES:
7510 error = memorystatus_cmd_get_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
7511 break;
7512 case MEMORYSTATUS_CMD_GET_MEMLIMIT_EXCESS:
7513 error = memorystatus_cmd_get_memlimit_excess_np(args->pid, args->flags, args->buffer, args->buffersize, ret);
7514 break;
7515 case MEMORYSTATUS_CMD_GRP_SET_PROPERTIES:
7516 error = memorystatus_cmd_grp_set_properties((int32_t)args->flags, args->buffer, args->buffersize, ret);
7517 break;
7518 case MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT:
7519 error = memorystatus_cmd_get_jetsam_snapshot((int32_t)args->flags, args->buffer, args->buffersize, ret);
7520 break;
7521 case MEMORYSTATUS_CMD_GET_PRESSURE_STATUS:
7522 error = memorystatus_cmd_get_pressure_status(ret);
7523 break;
7524 #if CONFIG_JETSAM
7525 case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
7526 /*
7527 * This call does not distinguish between active and inactive limits.
7528 * Default behavior in 2-level HWM world is to set both.
7529 * Non-fatal limit is also assumed for both.
7530 */
7531 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, FALSE);
7532 break;
7533 case MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT:
7534 /*
7535 * This call does not distinguish between active and inactive limits.
7536 * Default behavior in 2-level HWM world is to set both.
7537 * Fatal limit is also assumed for both.
7538 */
7539 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
7540 break;
7541 #endif /* CONFIG_JETSAM */
7542 /* Test commands */
7543 #if DEVELOPMENT || DEBUG
7544 case MEMORYSTATUS_CMD_TEST_JETSAM:
7545 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_GENERIC);
7546 if (jetsam_reason == OS_REASON_NULL) {
7547 printf("memorystatus_control: failed to allocate jetsam reason\n");
7548 }
7549
7550 error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled, jetsam_reason) ? 0 : EINVAL;
7551 break;
7552 case MEMORYSTATUS_CMD_TEST_JETSAM_SORT:
7553 error = memorystatus_cmd_test_jetsam_sort(args->pid, (int32_t)args->flags);
7554 break;
7555 #if CONFIG_JETSAM
7556 case MEMORYSTATUS_CMD_SET_JETSAM_PANIC_BITS:
7557 error = memorystatus_cmd_set_panic_bits(args->buffer, args->buffersize);
7558 break;
7559 #endif /* CONFIG_JETSAM */
7560 #else /* DEVELOPMENT || DEBUG */
7561 #pragma unused(jetsam_reason)
7562 #endif /* DEVELOPMENT || DEBUG */
7563 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_ENABLE:
7564 if (memorystatus_aggressive_jetsam_lenient_allowed == FALSE) {
7565 #if DEVELOPMENT || DEBUG
7566 printf("Enabling Lenient Mode\n");
7567 #endif /* DEVELOPMENT || DEBUG */
7568
7569 memorystatus_aggressive_jetsam_lenient_allowed = TRUE;
7570 memorystatus_aggressive_jetsam_lenient = TRUE;
7571 error = 0;
7572 }
7573 break;
7574 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_DISABLE:
7575 #if DEVELOPMENT || DEBUG
7576 printf("Disabling Lenient mode\n");
7577 #endif /* DEVELOPMENT || DEBUG */
7578 memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
7579 memorystatus_aggressive_jetsam_lenient = FALSE;
7580 error = 0;
7581 break;
7582 case MEMORYSTATUS_CMD_GET_AGGRESSIVE_JETSAM_LENIENT_MODE:
7583 *ret = (memorystatus_aggressive_jetsam_lenient ? 1 : 0);
7584 error = 0;
7585 break;
7586 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE:
7587 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE:
7588 error = memorystatus_low_mem_privileged_listener(args->command);
7589 break;
7590
7591 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE:
7592 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE:
7593 error = memorystatus_update_inactive_jetsam_priority_band(args->pid, args->command, JETSAM_PRIORITY_ELEVATED_INACTIVE, args->flags ? TRUE : FALSE);
7594 break;
7595 case MEMORYSTATUS_CMD_SET_PROCESS_IS_MANAGED:
7596 error = memorystatus_set_process_is_managed(args->pid, args->flags);
7597 break;
7598
7599 case MEMORYSTATUS_CMD_GET_PROCESS_IS_MANAGED:
7600 error = memorystatus_get_process_is_managed(args->pid, ret);
7601 break;
7602
7603 #if CONFIG_FREEZE
7604 case MEMORYSTATUS_CMD_SET_PROCESS_IS_FREEZABLE:
7605 error = memorystatus_set_process_is_freezable(args->pid, args->flags ? TRUE : FALSE);
7606 break;
7607
7608 case MEMORYSTATUS_CMD_GET_PROCESS_IS_FREEZABLE:
7609 error = memorystatus_get_process_is_freezable(args->pid, ret);
7610 break;
7611
7612 #if DEVELOPMENT || DEBUG
7613 case MEMORYSTATUS_CMD_FREEZER_CONTROL:
7614 error = memorystatus_freezer_control(args->flags, args->buffer, args->buffersize, ret);
7615 break;
7616 #endif /* DEVELOPMENT || DEBUG */
7617 #endif /* CONFIG_FREEZE */
7618
7619 #if CONFIG_JETSAM
7620 #if DEVELOPMENT || DEBUG
7621 case MEMORYSTATUS_CMD_INCREASE_JETSAM_TASK_LIMIT:
7622 error = memorystatus_cmd_increase_jetsam_task_limit(args->pid, args->flags);
7623 break;
7624 #endif /* DEVELOPMENT */
7625 #endif /* CONFIG_JETSAM */
7626
7627 default:
7628 break;
7629 }
7630
7631 out:
7632 return error;
7633 }
7634
7635 /* Coalition support */
7636
7637 /* sorting info for a particular priority bucket */
7638 typedef struct memstat_sort_info {
7639 coalition_t msi_coal;
7640 uint64_t msi_page_count;
7641 pid_t msi_pid;
7642 int msi_ntasks;
7643 } memstat_sort_info_t;
7644
7645 /*
7646 * qsort from smallest page count to largest page count
7647 *
7648 * return < 0 for a < b
7649 * 0 for a == b
7650 * > 0 for a > b
7651 */
7652 static int
7653 memstat_asc_cmp(const void *a, const void *b)
7654 {
7655 const memstat_sort_info_t *msA = (const memstat_sort_info_t *)a;
7656 const memstat_sort_info_t *msB = (const memstat_sort_info_t *)b;
7657
7658 return (int)((uint64_t)msA->msi_page_count - (uint64_t)msB->msi_page_count);
7659 }
7660
7661 /*
7662 * Return the number of pids rearranged during this sort.
7663 */
7664 static int
7665 memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order)
7666 {
7667 #define MAX_SORT_PIDS 80
7668 #define MAX_COAL_LEADERS 10
7669
7670 unsigned int b = bucket_index;
7671 int nleaders = 0;
7672 int ntasks = 0;
7673 proc_t p = NULL;
7674 coalition_t coal = COALITION_NULL;
7675 int pids_moved = 0;
7676 int total_pids_moved = 0;
7677 int i;
7678
7679 /*
7680 * The system is typically under memory pressure when in this
7681 * path, hence, we want to avoid dynamic memory allocation.
7682 */
7683 memstat_sort_info_t leaders[MAX_COAL_LEADERS];
7684 pid_t pid_list[MAX_SORT_PIDS];
7685
7686 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
7687 return 0;
7688 }
7689
7690 /*
7691 * Clear the array that holds coalition leader information
7692 */
7693 for (i = 0; i < MAX_COAL_LEADERS; i++) {
7694 leaders[i].msi_coal = COALITION_NULL;
7695 leaders[i].msi_page_count = 0; /* will hold total coalition page count */
7696 leaders[i].msi_pid = 0; /* will hold coalition leader pid */
7697 leaders[i].msi_ntasks = 0; /* will hold the number of tasks in a coalition */
7698 }
7699
7700 p = memorystatus_get_first_proc_locked(&b, FALSE);
7701 while (p) {
7702 coal = task_get_coalition(p->task, COALITION_TYPE_JETSAM);
7703 if (coalition_is_leader(p->task, coal)) {
7704 if (nleaders < MAX_COAL_LEADERS) {
7705 int coal_ntasks = 0;
7706 uint64_t coal_page_count = coalition_get_page_count(coal, &coal_ntasks);
7707 leaders[nleaders].msi_coal = coal;
7708 leaders[nleaders].msi_page_count = coal_page_count;
7709 leaders[nleaders].msi_pid = p->p_pid; /* the coalition leader */
7710 leaders[nleaders].msi_ntasks = coal_ntasks;
7711 nleaders++;
7712 } else {
7713 /*
7714 * We've hit MAX_COAL_LEADERS meaning we can handle no more coalitions.
7715 * Abandoned coalitions will linger at the tail of the priority band
7716 * when this sort session ends.
7717 * TODO: should this be an assert?
7718 */
7719 printf("%s: WARNING: more than %d leaders in priority band [%d]\n",
7720 __FUNCTION__, MAX_COAL_LEADERS, bucket_index);
7721 break;
7722 }
7723 }
7724 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
7725 }
7726
7727 if (nleaders == 0) {
7728 /* Nothing to sort */
7729 return 0;
7730 }
7731
7732 /*
7733 * Sort the coalition leader array, from smallest coalition page count
7734 * to largest coalition page count. When inserted in the priority bucket,
7735 * smallest coalition is handled first, resulting in the last to be jetsammed.
7736 */
7737 if (nleaders > 1) {
7738 qsort(leaders, nleaders, sizeof(memstat_sort_info_t), memstat_asc_cmp);
7739 }
7740
7741 #if 0
7742 for (i = 0; i < nleaders; i++) {
7743 printf("%s: coal_leader[%d of %d] pid[%d] pages[%llu] ntasks[%d]\n",
7744 __FUNCTION__, i, nleaders, leaders[i].msi_pid, leaders[i].msi_page_count,
7745 leaders[i].msi_ntasks);
7746 }
7747 #endif
7748
7749 /*
7750 * During coalition sorting, processes in a priority band are rearranged
7751 * by being re-inserted at the head of the queue. So, when handling a
7752 * list, the first process that gets moved to the head of the queue,
7753 * ultimately gets pushed toward the queue tail, and hence, jetsams last.
7754 *
7755 * So, for example, the coalition leader is expected to jetsam last,
7756 * after its coalition members. Therefore, the coalition leader is
7757 * inserted at the head of the queue first.
7758 *
7759 * After processing a coalition, the jetsam order is as follows:
7760 * undefs(jetsam first), extensions, xpc services, leader(jetsam last)
7761 */
7762
7763 /*
7764 * Coalition members are rearranged in the priority bucket here,
7765 * based on their coalition role.
7766 */
7767 total_pids_moved = 0;
7768 for (i = 0; i < nleaders; i++) {
7769 /* a bit of bookkeeping */
7770 pids_moved = 0;
7771
7772 /* Coalition leaders are jetsammed last, so move into place first */
7773 pid_list[0] = leaders[i].msi_pid;
7774 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list, 1);
7775
7776 /* xpc services should jetsam after extensions */
7777 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_XPC,
7778 coal_sort_order, pid_list, MAX_SORT_PIDS);
7779
7780 if (ntasks > 0) {
7781 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
7782 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
7783 }
7784
7785 /* extensions should jetsam after unmarked processes */
7786 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_EXT,
7787 coal_sort_order, pid_list, MAX_SORT_PIDS);
7788
7789 if (ntasks > 0) {
7790 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
7791 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
7792 }
7793
7794 /* undefined coalition members should be the first to jetsam */
7795 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_UNDEF,
7796 coal_sort_order, pid_list, MAX_SORT_PIDS);
7797
7798 if (ntasks > 0) {
7799 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
7800 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
7801 }
7802
7803 #if 0
7804 if (pids_moved == leaders[i].msi_ntasks) {
7805 /*
7806 * All the pids in the coalition were found in this band.
7807 */
7808 printf("%s: pids_moved[%d] equal total coalition ntasks[%d] \n", __FUNCTION__,
7809 pids_moved, leaders[i].msi_ntasks);
7810 } else if (pids_moved > leaders[i].msi_ntasks) {
7811 /*
7812 * Apparently new coalition members showed up during the sort?
7813 */
7814 printf("%s: pids_moved[%d] were greater than expected coalition ntasks[%d] \n", __FUNCTION__,
7815 pids_moved, leaders[i].msi_ntasks);
7816 } else {
7817 /*
7818 * Apparently not all the pids in the coalition were found in this band?
7819 */
7820 printf("%s: pids_moved[%d] were less than expected coalition ntasks[%d] \n", __FUNCTION__,
7821 pids_moved, leaders[i].msi_ntasks);
7822 }
7823 #endif
7824
7825 total_pids_moved += pids_moved;
7826 } /* end for */
7827
7828 return total_pids_moved;
7829 }
7830
7831
7832 /*
7833 * Traverse a list of pids, searching for each within the priority band provided.
7834 * If pid is found, move it to the front of the priority band.
7835 * Never searches outside the priority band provided.
7836 *
7837 * Input:
7838 * bucket_index - jetsam priority band.
7839 * pid_list - pointer to a list of pids.
7840 * list_sz - number of pids in the list.
7841 *
7842 * Pid list ordering is important in that,
7843 * pid_list[n] is expected to jetsam ahead of pid_list[n+1].
7844 * The sort_order is set by the coalition default.
7845 *
7846 * Return:
7847 * the number of pids found and hence moved within the priority band.
7848 */
7849 static int
7850 memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz)
7851 {
7852 memstat_bucket_t *current_bucket;
7853 int i;
7854 int found_pids = 0;
7855
7856 if ((pid_list == NULL) || (list_sz <= 0)) {
7857 return 0;
7858 }
7859
7860 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
7861 return 0;
7862 }
7863
7864 current_bucket = &memstat_bucket[bucket_index];
7865 for (i = 0; i < list_sz; i++) {
7866 unsigned int b = bucket_index;
7867 proc_t p = NULL;
7868 proc_t aProc = NULL;
7869 pid_t aPid;
7870 int list_index;
7871
7872 list_index = ((list_sz - 1) - i);
7873 aPid = pid_list[list_index];
7874
7875 /* never search beyond bucket_index provided */
7876 p = memorystatus_get_first_proc_locked(&b, FALSE);
7877 while (p) {
7878 if (p->p_pid == aPid) {
7879 aProc = p;
7880 break;
7881 }
7882 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
7883 }
7884
7885 if (aProc == NULL) {
7886 /* pid not found in this band, just skip it */
7887 continue;
7888 } else {
7889 TAILQ_REMOVE(&current_bucket->list, aProc, p_memstat_list);
7890 TAILQ_INSERT_HEAD(&current_bucket->list, aProc, p_memstat_list);
7891 found_pids++;
7892 }
7893 }
7894 return found_pids;
7895 }
7896
7897 int
7898 memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index)
7899 {
7900 int32_t i = JETSAM_PRIORITY_IDLE;
7901 int count = 0;
7902
7903 if (max_bucket_index >= MEMSTAT_BUCKET_COUNT) {
7904 return -1;
7905 }
7906
7907 while (i <= max_bucket_index) {
7908 count += memstat_bucket[i++].count;
7909 }
7910
7911 return count;
7912 }
7913
7914 int
7915 memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap)
7916 {
7917 #if !CONFIG_JETSAM
7918 if (!p || (!isApp(p)) || (p->p_memstat_state & (P_MEMSTAT_INTERNAL | P_MEMSTAT_MANAGED))) {
7919 /*
7920 * Ineligible processes OR system processes e.g. launchd.
7921 *
7922 * We also skip processes that have the P_MEMSTAT_MANAGED bit set, i.e.
7923 * they're managed by assertiond. These are iOS apps that have been ported
7924 * to macOS. assertiond might be in the process of modifying the app's
7925 * priority / memory limit - so it might have the proc_list lock, and then try
7926 * to take the task lock. Meanwhile we've entered this function with the task lock
7927 * held, and we need the proc_list lock below. So we'll deadlock with assertiond.
7928 *
7929 * It should be fine to read the P_MEMSTAT_MANAGED bit without the proc_list
7930 * lock here, since assertiond only sets this bit on process launch.
7931 */
7932 return -1;
7933 }
7934
7935 /*
7936 * For macOS only:
7937 * We would like to use memorystatus_update() here to move the processes
7938 * within the bands. Unfortunately memorystatus_update() calls
7939 * memorystatus_update_priority_locked() which uses any band transitions
7940 * as an indication to modify ledgers. For that it needs the task lock
7941 * and since we came into this function with the task lock held, we'll deadlock.
7942 *
7943 * Unfortunately we can't completely disable ledger updates because we still
7944 * need the ledger updates for a subset of processes i.e. daemons.
7945 * When all processes on all platforms support memory limits, we can simply call
7946 * memorystatus_update().
7947 *
7948 * It also has some logic to deal with 'aging' which, currently, is only applicable
7949 * on CONFIG_JETSAM configs. So, till every platform has CONFIG_JETSAM we'll need
7950 * to do this explicit band transition.
7951 */
7952
7953 memstat_bucket_t *current_bucket, *new_bucket;
7954 int32_t priority = 0;
7955
7956 proc_list_lock();
7957
7958 if (((p->p_listflag & P_LIST_EXITED) != 0) ||
7959 (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED))) {
7960 /*
7961 * If the process is on its way out OR
7962 * jetsam has alread tried and failed to kill this process,
7963 * let's skip the whole jetsam band transition.
7964 */
7965 proc_list_unlock();
7966 return 0;
7967 }
7968
7969 if (is_appnap) {
7970 current_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
7971 new_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
7972 priority = JETSAM_PRIORITY_IDLE;
7973 } else {
7974 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
7975 /*
7976 * It is possible that someone pulled this process
7977 * out of the IDLE band without updating its app-nap
7978 * parameters.
7979 */
7980 proc_list_unlock();
7981 return 0;
7982 }
7983
7984 current_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
7985 new_bucket = &memstat_bucket[p->p_memstat_requestedpriority];
7986 priority = p->p_memstat_requestedpriority;
7987 }
7988
7989 TAILQ_REMOVE(&current_bucket->list, p, p_memstat_list);
7990 current_bucket->count--;
7991 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
7992 current_bucket->relaunch_high_count--;
7993 }
7994 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
7995 new_bucket->count++;
7996 if (p->p_memstat_relaunch_flags & (P_MEMSTAT_RELAUNCH_HIGH)) {
7997 new_bucket->relaunch_high_count++;
7998 }
7999 /*
8000 * Record idle start or idle delta.
8001 */
8002 if (p->p_memstat_effectivepriority == priority) {
8003 /*
8004 * This process is not transitioning between
8005 * jetsam priority buckets. Do nothing.
8006 */
8007 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
8008 uint64_t now;
8009 /*
8010 * Transitioning out of the idle priority bucket.
8011 * Record idle delta.
8012 */
8013 assert(p->p_memstat_idle_start != 0);
8014 now = mach_absolute_time();
8015 if (now > p->p_memstat_idle_start) {
8016 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
8017 }
8018 } else if (priority == JETSAM_PRIORITY_IDLE) {
8019 /*
8020 * Transitioning into the idle priority bucket.
8021 * Record idle start.
8022 */
8023 p->p_memstat_idle_start = mach_absolute_time();
8024 }
8025
8026 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CHANGE_PRIORITY), p->p_pid, priority, p->p_memstat_effectivepriority, 0, 0);
8027
8028 p->p_memstat_effectivepriority = priority;
8029
8030 proc_list_unlock();
8031
8032 return 0;
8033
8034 #else /* !CONFIG_JETSAM */
8035 #pragma unused(p)
8036 #pragma unused(is_appnap)
8037 return -1;
8038 #endif /* !CONFIG_JETSAM */
8039 }
8040
8041 uint64_t
8042 memorystatus_available_memory_internal(proc_t p)
8043 {
8044 #ifdef XNU_TARGET_OS_OSX
8045 #pragma unused(p)
8046 return 0;
8047 #else
8048 const uint64_t footprint_in_bytes = get_task_phys_footprint(p->task);
8049 int32_t memlimit_mb;
8050 int64_t memlimit_bytes;
8051 int64_t rc;
8052
8053 if (isApp(p) == FALSE) {
8054 return 0;
8055 }
8056
8057 if (p->p_memstat_memlimit > 0) {
8058 memlimit_mb = p->p_memstat_memlimit;
8059 } else if (task_convert_phys_footprint_limit(-1, &memlimit_mb) != KERN_SUCCESS) {
8060 return 0;
8061 }
8062
8063 if (memlimit_mb <= 0) {
8064 memlimit_bytes = INT_MAX & ~((1 << 20) - 1);
8065 } else {
8066 memlimit_bytes = ((int64_t) memlimit_mb) << 20;
8067 }
8068
8069 rc = memlimit_bytes - footprint_in_bytes;
8070
8071 return (rc >= 0) ? rc : 0;
8072 #endif
8073 }
8074
8075 int
8076 memorystatus_available_memory(struct proc *p, __unused struct memorystatus_available_memory_args *args, uint64_t *ret)
8077 {
8078 *ret = memorystatus_available_memory_internal(p);
8079
8080 return 0;
8081 }
8082
8083 #if CONFIG_JETSAM
8084 #if DEVELOPMENT || DEBUG
8085 static int
8086 memorystatus_cmd_increase_jetsam_task_limit(pid_t pid, uint32_t byte_increase)
8087 {
8088 memorystatus_memlimit_properties_t mmp_entry;
8089
8090 /* Validate inputs */
8091 if ((pid == 0) || (byte_increase == 0)) {
8092 return EINVAL;
8093 }
8094
8095 proc_t p = proc_find(pid);
8096
8097 if (!p) {
8098 return ESRCH;
8099 }
8100
8101 const uint32_t current_memlimit_increase = roundToNearestMB(p->p_memlimit_increase);
8102 const uint32_t page_aligned_increase = round_page(p->p_memlimit_increase + byte_increase); /* round to page */
8103
8104 proc_list_lock();
8105
8106 memorystatus_get_memlimit_properties_internal(p, &mmp_entry);
8107
8108 if (mmp_entry.memlimit_active > 0) {
8109 mmp_entry.memlimit_active -= current_memlimit_increase;
8110 mmp_entry.memlimit_active += roundToNearestMB(page_aligned_increase);
8111 }
8112
8113 if (mmp_entry.memlimit_inactive > 0) {
8114 mmp_entry.memlimit_inactive -= current_memlimit_increase;
8115 mmp_entry.memlimit_inactive += roundToNearestMB(page_aligned_increase);
8116 }
8117
8118 /*
8119 * Store the updated delta limit in the proc.
8120 */
8121 p->p_memlimit_increase = page_aligned_increase;
8122
8123 int error = memorystatus_set_memlimit_properties_internal(p, &mmp_entry);
8124
8125 proc_list_unlock();
8126 proc_rele(p);
8127
8128 return error;
8129 }
8130 #endif /* DEVELOPMENT */
8131 #endif /* CONFIG_JETSAM */
mirror of XNU