1 /*
2 * Copyright (c) 2006-2021 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/task.h>
31 #include <libkern/libkern.h>
32 #include <machine/atomic.h>
33 #include <mach/coalition.h>
34 #include <os/log.h>
35 #include <sys/coalition.h>
36 #include <sys/proc.h>
37 #include <sys/proc_internal.h>
38 #include <sys/kdebug.h>
39 #include <sys/kern_memorystatus.h>
40 #include <vm/vm_protos.h>
41
42 #include "kern_memorystatus_internal.h"
43
44 /*
45 * All memory pressure policy decisions should live here, and there should be
46 * as little mechanism as possible. This file prioritizes readability.
47 */
48
49 #pragma mark Policy Function Declarations
50
51 #if CONFIG_JETSAM
52 static bool memorystatus_check_aggressive_jetsam_needed(int *jld_idle_kills);
53 #endif /* CONFIG_JETSAM */
54
55 #pragma mark Memorystatus Health Check
56
57 /*
58 * Each subsystem that relies on the memorystatus thread
59 * for resource exhaustion should put a health check in this section.
60 * The memorystatus thread runs all of the health checks
61 * to determine if the system is healthy. If the system is unhealthy
62 * it picks an action based on the system health status. See the
63 * Memorystatus Thread Actions section below.
64 */
65
66 extern bool vm_compressor_needs_to_swap(bool wake_memorystatus_thread);
67 extern boolean_t vm_compressor_low_on_space(void);
68 extern bool vm_compressor_compressed_pages_nearing_limit(void);
69 extern bool vm_compressor_is_thrashing(void);
70 extern bool vm_compressor_swapout_is_ripe(void);
71
72 static void
memorystatus_health_check(memorystatus_system_health_t * status)73 memorystatus_health_check(memorystatus_system_health_t *status)
74 {
75 memset(status, 0, sizeof(memorystatus_system_health_t));
76 #if CONFIG_JETSAM
77 status->msh_available_pages_below_pressure = memorystatus_avail_pages_below_pressure();
78 status->msh_available_pages_below_critical = memorystatus_avail_pages_below_critical();
79 status->msh_compressor_is_low_on_space = (vm_compressor_low_on_space() == TRUE);
80 status->msh_compressed_pages_nearing_limit = vm_compressor_compressed_pages_nearing_limit();
81 status->msh_swapout_is_ripe = vm_compressor_swapout_is_ripe();
82 if (!status->msh_swapout_is_ripe) {
83 status->msh_compressor_is_thrashing = !memorystatus_swap_all_apps && vm_compressor_is_thrashing();
84 #if CONFIG_PHANTOM_CACHE
85 status->msh_phantom_cache_pressure = os_atomic_load(&memorystatus_phantom_cache_pressure, acquire);
86 #else
87 status->msh_phantom_cache_pressure = false;
88 #endif /* CONFIG_PHANTOM_CACHE */
89 } else {
90 status->msh_compressor_is_thrashing = false;
91 status->msh_phantom_cache_pressure = false;
92 }
93 if (!memorystatus_swap_all_apps &&
94 (status->msh_swapout_is_ripe || status->msh_phantom_cache_pressure) &&
95 !(status->msh_compressor_is_thrashing && status->msh_compressor_is_low_on_space)) {
96 status->msh_filecache_is_thrashing = true;
97 }
98 if (os_atomic_load(&memorystatus_compressor_space_shortage, relaxed)) {
99 status->msh_compressor_is_low_on_space = true;
100 }
101 status->msh_swappable_compressor_segments_over_limit = memorystatus_swap_over_trigger(100);
102 status->msh_swapin_queue_over_limit = memorystatus_swapin_over_trigger();
103 status->msh_swap_low_on_space = vm_swap_low_on_space();
104 status->msh_swap_out_of_space = vm_swap_out_of_space();
105 #endif /* CONFIG_JETSAM */
106 status->msh_zone_map_is_exhausted = os_atomic_load(&memorystatus_zone_map_is_exhausted, relaxed);
107 }
108
109 bool
memorystatus_is_system_healthy(const memorystatus_system_health_t * status)110 memorystatus_is_system_healthy(const memorystatus_system_health_t *status)
111 {
112 #if CONFIG_JETSAM
113 return !(status->msh_available_pages_below_critical
114 || status->msh_compressor_is_low_on_space
115 || status->msh_compressor_is_thrashing
116 || status->msh_filecache_is_thrashing
117 || status->msh_zone_map_is_exhausted);
118 #else /* CONFIG_JETSAM */
119 return !status->msh_zone_map_is_exhausted;
120 #endif /* CONFIG_JETSAM */
121 }
122
123
124 #pragma mark Memorystatus Thread Actions
125
126 /*
127 * This section picks the appropriate memorystatus_action & deploys it.
128 */
129
130 /*
131 * Inspects the state of various resources in the system to see if
132 * the system is healthy. If the system is not healthy, picks a
133 * memorystatus_action_t to recover the system.
134 *
135 * Every time the memorystatus thread wakes up it calls into here
136 * to pick an action. It will continue performing memorystatus actions until this
137 * function returns MEMORYSTATUS_KILL_NONE. At that point the thread will block.
138 */
139 memorystatus_action_t
memorystatus_pick_action(struct jetsam_thread_state * jetsam_thread,uint32_t * kill_cause,bool highwater_remaining,bool suspended_swappable_apps_remaining,bool swappable_apps_remaining,int * jld_idle_kills)140 memorystatus_pick_action(struct jetsam_thread_state *jetsam_thread,
141 uint32_t *kill_cause,
142 bool highwater_remaining,
143 bool suspended_swappable_apps_remaining,
144 bool swappable_apps_remaining,
145 int *jld_idle_kills)
146 {
147 memorystatus_system_health_t status;
148 memorystatus_health_check(&status);
149 memorystatus_log_system_health(&status);
150 bool is_system_healthy = memorystatus_is_system_healthy(&status);
151
152 #if CONFIG_JETSAM
153 if (status.msh_available_pages_below_pressure || !is_system_healthy) {
154 /*
155 * If swap is enabled, first check if we're running low or are out of swap space.
156 */
157 if (memorystatus_swap_all_apps && jetsam_kill_on_low_swap) {
158 if (swappable_apps_remaining && status.msh_swap_out_of_space) {
159 *kill_cause = kMemorystatusKilledLowSwap;
160 return MEMORYSTATUS_KILL_SWAPPABLE;
161 } else if (suspended_swappable_apps_remaining && status.msh_swap_low_on_space) {
162 *kill_cause = kMemorystatusKilledLowSwap;
163 return MEMORYSTATUS_KILL_SUSPENDED_SWAPPABLE;
164 }
165 }
166
167 /*
168 * We're below the pressure level or the system is unhealthy,
169 * regardless of the system health let's check if we should be swapping
170 * and if there are high watermark kills left to do.
171 */
172 if (memorystatus_swap_all_apps) {
173 if (status.msh_swappable_compressor_segments_over_limit && !vm_swapout_thread_running && !os_atomic_load(&vm_swapout_wake_pending, relaxed)) {
174 /*
175 * TODO: The swapper will keep running until it has drained the entire early swapout queue.
176 * That might be overly aggressive & we should look into tuning it.
177 * See rdar://84102304.
178 */
179 return MEMORYSTATUS_WAKE_SWAPPER;
180 } else if (status.msh_swapin_queue_over_limit) {
181 return MEMORYSTATUS_PROCESS_SWAPIN_QUEUE;
182 } else if (status.msh_swappable_compressor_segments_over_limit) {
183 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: Skipping swap wakeup because the swap thread is already running. vm_swapout_thread_running=%d, vm_swapout_wake_pending=%d\n", vm_swapout_thread_running, os_atomic_load(&vm_swapout_wake_pending, relaxed));
184 }
185 }
186
187 if (highwater_remaining) {
188 *kill_cause = kMemorystatusKilledHiwat;
189 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: Looking for highwatermark kills.\n");
190 return MEMORYSTATUS_KILL_HIWATER;
191 }
192 }
193
194 if (is_system_healthy) {
195 *kill_cause = 0;
196 return MEMORYSTATUS_KILL_NONE;
197 }
198
199 /*
200 * At this point the system is unhealthy and there are no
201 * more highwatermark processes to kill.
202 */
203
204 if (!jetsam_thread->limit_to_low_bands) {
205 if (memorystatus_check_aggressive_jetsam_needed(jld_idle_kills)) {
206 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: Starting aggressive jetsam.\n");
207 *kill_cause = kMemorystatusKilledProcThrashing;
208 return MEMORYSTATUS_KILL_AGGRESSIVE;
209 }
210 }
211 /*
212 * The system is unhealthy and we either don't need aggressive jetsam
213 * or are not allowed to deploy it.
214 * Kill in priority order. We'll use LRU within every band except the
215 * FG (which will be sorted by coalition role).
216 */
217 *kill_cause = memorystatus_pick_kill_cause(&status);
218 return MEMORYSTATUS_KILL_TOP_PROCESS;
219 #else /* CONFIG_JETSAM */
220 (void) jetsam_thread;
221 (void) jld_idle_kills;
222 (void) suspended_swappable_apps_remaining;
223 (void) swappable_apps_remaining;
224 /*
225 * Without CONFIG_JETSAM, we only kill if the system is unhealthy.
226 * There is no aggressive jetsam and no
227 * early highwatermark killing.
228 */
229 if (is_system_healthy) {
230 *kill_cause = 0;
231 return MEMORYSTATUS_KILL_NONE;
232 }
233 if (highwater_remaining) {
234 *kill_cause = kMemorystatusKilledHiwat;
235 return MEMORYSTATUS_KILL_HIWATER;
236 } else {
237 *kill_cause = memorystatus_pick_kill_cause(&status);
238 return MEMORYSTATUS_KILL_TOP_PROCESS;
239 }
240 #endif /* CONFIG_JETSAM */
241 }
242
243 #pragma mark Aggressive Jetsam
244 /*
245 * This section defines when we deploy aggressive jetsam.
246 * Aggressive jetsam kills everything up to the jld_priority_band_max band.
247 */
248
249 #if CONFIG_JETSAM
250
251 static bool
252 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);
253
254 /*
255 * kJetsamHighRelaunchCandidatesThreshold defines the percentage of candidates
256 * in the idle & deferred bands that need to be bad candidates in order to trigger
257 * aggressive jetsam.
258 */
259 #define kJetsamHighRelaunchCandidatesThreshold (100)
260
261 /* kJetsamMinCandidatesThreshold defines the minimum number of candidates in the
262 * idle/deferred bands to trigger aggressive jetsam. This value basically decides
263 * how much memory the system is ready to hold in the lower bands without triggering
264 * aggressive jetsam. This number should ideally be tuned based on the memory config
265 * of the device.
266 */
267 #define kJetsamMinCandidatesThreshold (5)
268
269 static bool
memorystatus_check_aggressive_jetsam_needed(int * jld_idle_kills)270 memorystatus_check_aggressive_jetsam_needed(int *jld_idle_kills)
271 {
272 bool aggressive_jetsam_needed = false;
273 int total_candidates = 0;
274 /*
275 * The aggressive jetsam logic looks at the number of times it has been in the
276 * aggressive loop to determine the max priority band it should kill upto. The
277 * static variables below are used to track that property.
278 *
279 * To reset those values, the implementation checks if it has been
280 * memorystatus_jld_eval_period_msecs since the parameters were reset.
281 */
282
283 if (memorystatus_jld_enabled == FALSE) {
284 /* If aggressive jetsam is disabled, nothing to do here */
285 return FALSE;
286 }
287
288 /* Get current timestamp (msecs only) */
289 struct timeval jld_now_tstamp = {0, 0};
290 uint64_t jld_now_msecs = 0;
291 microuptime(&jld_now_tstamp);
292 jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
293
294 /*
295 * Look at the number of candidates in the idle and deferred band and
296 * how many out of them are marked as high relaunch probability.
297 */
298 aggressive_jetsam_needed = memorystatus_aggressive_jetsam_needed_sysproc_aging(jld_eval_aggressive_count,
299 jld_idle_kills, jld_idle_kill_candidates, &total_candidates);
300
301 /*
302 * Check if its been really long since the aggressive jetsam evaluation
303 * parameters have been refreshed. This logic also resets the jld_eval_aggressive_count
304 * counter to make sure we reset the aggressive jetsam severity.
305 */
306 boolean_t param_reval = false;
307
308 if ((total_candidates == 0) ||
309 (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
310 jld_timestamp_msecs = jld_now_msecs;
311 jld_idle_kill_candidates = total_candidates;
312 *jld_idle_kills = 0;
313 jld_eval_aggressive_count = 0;
314 jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
315 param_reval = true;
316 }
317
318 /*
319 * It is also possible that the system is down to a very small number of processes in the candidate
320 * bands. In that case, the decisions made by the memorystatus_aggressive_jetsam_needed_* routines
321 * would not be useful. In that case, do not trigger aggressive jetsam.
322 */
323 if (total_candidates < kJetsamMinCandidatesThreshold) {
324 #if DEVELOPMENT || DEBUG
325 printf("memorystatus: aggressive: [FAILED] Low Candidate Count (current: %d, threshold: %d)\n", total_candidates, kJetsamMinCandidatesThreshold);
326 #endif /* DEVELOPMENT || DEBUG */
327 aggressive_jetsam_needed = false;
328 }
329 return aggressive_jetsam_needed;
330 }
331
332 static bool
memorystatus_aggressive_jetsam_needed_sysproc_aging(__unused int eval_aggressive_count,__unused int * idle_kills,__unused int idle_kill_candidates,int * total_candidates)333 memorystatus_aggressive_jetsam_needed_sysproc_aging(__unused int eval_aggressive_count, __unused int *idle_kills, __unused int idle_kill_candidates, int *total_candidates)
334 {
335 bool aggressive_jetsam_needed = false;
336
337 /*
338 * For the kJetsamAgingPolicySysProcsReclaimedFirst aging policy, we maintain the jetsam
339 * relaunch behavior for all daemons. Also, daemons and apps are aged in deferred bands on
340 * every dirty->clean transition. For this aging policy, the best way to determine if
341 * aggressive jetsam is needed, is to see if the kill candidates are mostly bad candidates.
342 * If yes, then we need to go to higher bands to reclaim memory.
343 */
344 proc_list_lock();
345 /* Get total candidate counts for idle and idle deferred bands */
346 *total_candidates = memstat_bucket[JETSAM_PRIORITY_IDLE].count + memstat_bucket[system_procs_aging_band].count;
347 /* Get counts of bad kill candidates in idle and idle deferred bands */
348 int bad_candidates = memstat_bucket[JETSAM_PRIORITY_IDLE].relaunch_high_count + memstat_bucket[system_procs_aging_band].relaunch_high_count;
349
350 proc_list_unlock();
351
352 /* Check if the number of bad candidates is greater than kJetsamHighRelaunchCandidatesThreshold % */
353 aggressive_jetsam_needed = (((bad_candidates * 100) / *total_candidates) >= kJetsamHighRelaunchCandidatesThreshold);
354
355 /*
356 * Since the new aging policy bases the aggressive jetsam trigger on percentage of
357 * bad candidates, it is prone to being overly aggressive. In order to mitigate that,
358 * make sure the system is really under memory pressure before triggering aggressive
359 * jetsam.
360 */
361 if (memorystatus_available_pages > memorystatus_sysproc_aging_aggr_pages) {
362 aggressive_jetsam_needed = false;
363 }
364
365 #if DEVELOPMENT || DEBUG
366 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",
367 eval_aggressive_count, aggressive_jetsam_needed ? "PASSED" : "FAILED", *total_candidates, bad_candidates,
368 kJetsamHighRelaunchCandidatesThreshold, (uint64_t)MEMORYSTATUS_LOG_AVAILABLE_PAGES, (uint64_t)memorystatus_sysproc_aging_aggr_pages);
369 #endif /* DEVELOPMENT || DEBUG */
370 return aggressive_jetsam_needed;
371 }
372
373 #endif /* CONFIG_JETSAM */
374
375 #pragma mark Freezer
376 #if CONFIG_FREEZE
377 /*
378 * Freezer policies
379 */
380
381 /*
382 * These functions determine what is eligible for the freezer
383 * and the order that we consider freezing them
384 */
385
386 /*
387 * Checks if the given process is eligible for the freezer.
388 * Processes can only be frozen if this returns true.
389 */
390 bool
memorystatus_is_process_eligible_for_freeze(proc_t p)391 memorystatus_is_process_eligible_for_freeze(proc_t p)
392 {
393 /*
394 * Called with proc_list_lock held.
395 */
396
397 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
398
399 bool should_freeze = false;
400 uint32_t state = 0, pages = 0;
401 bool first_consideration = true;
402 task_t task;
403
404 state = p->p_memstat_state;
405
406 if (state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FREEZE_DISABLED | P_MEMSTAT_FREEZE_IGNORE)) {
407 if (state & P_MEMSTAT_FREEZE_DISABLED) {
408 p->p_memstat_freeze_skip_reason = kMemorystatusFreezeSkipReasonDisabled;
409 }
410 goto out;
411 }
412
413 task = proc_task(p);
414
415 if (isSysProc(p)) {
416 /*
417 * Daemon:- We consider freezing it if:
418 * - it belongs to a coalition and the leader is frozen, and,
419 * - its role in the coalition is XPC service.
420 *
421 * We skip memory size requirements in this case.
422 */
423 int task_role_in_coalition = 0;
424 proc_t leader_proc = memorystatus_get_coalition_leader_and_role(p, &task_role_in_coalition);
425 if (leader_proc == PROC_NULL || leader_proc == p) {
426 /*
427 * Jetsam coalition is leaderless or the leader is not an app.
428 * Either way, don't freeze this proc.
429 */
430 goto out;
431 }
432
433 /* Leader must be frozen */
434 if (!(leader_proc->p_memstat_state & P_MEMSTAT_FROZEN)) {
435 goto out;
436 }
437 /* Only freeze XPC services */
438 if (task_role_in_coalition == COALITION_TASKROLE_XPC) {
439 should_freeze = true;
440 }
441
442 goto out;
443 } else {
444 /*
445 * Application. Only freeze if it's suspended.
446 */
447 if (!(state & P_MEMSTAT_SUSPENDED)) {
448 goto out;
449 }
450 }
451
452 /*
453 * We're interested in tracking what percentage of
454 * eligible apps actually get frozen.
455 * To avoid skewing the metrics towards processes which
456 * are considered more frequently, we only track failures once
457 * per process.
458 */
459 first_consideration = !(state & P_MEMSTAT_FREEZE_CONSIDERED);
460
461 if (first_consideration) {
462 memorystatus_freezer_stats.mfs_process_considered_count++;
463 p->p_memstat_state |= P_MEMSTAT_FREEZE_CONSIDERED;
464 }
465
466 /* Only freeze applications meeting our minimum resident page criteria */
467 memorystatus_get_task_page_counts(proc_task(p), &pages, NULL, NULL);
468 if (pages < memorystatus_freeze_pages_min) {
469 if (first_consideration) {
470 memorystatus_freezer_stats.mfs_error_below_min_pages_count++;
471 }
472 p->p_memstat_freeze_skip_reason = kMemorystatusFreezeSkipReasonBelowMinPages;
473 goto out;
474 }
475
476 /* Don't freeze processes that are already exiting on core. It may have started exiting
477 * after we chose it for freeze, but before we obtained the proc_list_lock.
478 * NB: This is only possible if we're coming in from memorystatus_freeze_process_sync.
479 * memorystatus_freeze_top_process holds the proc_list_lock while it traverses the bands.
480 */
481 if (proc_list_exited(p)) {
482 if (first_consideration) {
483 memorystatus_freezer_stats.mfs_error_other_count++;
484 }
485 p->p_memstat_freeze_skip_reason = kMemorystatusFreezeSkipReasonOther;
486 goto out;
487 }
488
489 if (!memorystatus_freezer_use_ordered_list) {
490 /*
491 * We're not using the ordered list so we need to check
492 * that dasd recommended the process. Note that the ordered list
493 * algorithm only considers processes on the list in the first place
494 * so there's no need to double check here.
495 */
496 if (!memorystatus_freeze_process_is_recommended(p)) {
497 if (first_consideration) {
498 memorystatus_freezer_stats.mfs_error_low_probability_of_use_count++;
499 }
500 p->p_memstat_freeze_skip_reason = kMemorystatusFreezeSkipReasonLowProbOfUse;
501 goto out;
502 }
503 }
504
505 if (!(state & P_MEMSTAT_FROZEN) && p->p_memstat_effectivepriority > memorystatus_freeze_max_candidate_band) {
506 /*
507 * Proc has been elevated by something else.
508 * Don't freeze it.
509 */
510 if (first_consideration) {
511 memorystatus_freezer_stats.mfs_error_elevated_count++;
512 }
513 p->p_memstat_freeze_skip_reason = kMemorystatusFreezeSkipReasonElevated;
514 goto out;
515 }
516
517 should_freeze = true;
518 out:
519 if (should_freeze && !(state & P_MEMSTAT_FROZEN)) {
520 /*
521 * Reset the skip reason. If it's killed before we manage to actually freeze it
522 * we failed to consider it early enough.
523 */
524 p->p_memstat_freeze_skip_reason = kMemorystatusFreezeSkipReasonNone;
525 if (!first_consideration) {
526 /*
527 * We're freezing this for the first time and we previously considered it ineligible.
528 * Bump the considered count so that we track this as 1 failure
529 * and 1 success.
530 */
531 memorystatus_freezer_stats.mfs_process_considered_count++;
532 }
533 }
534 return should_freeze;
535 }
536
537 bool
memorystatus_freeze_proc_is_refreeze_eligible(proc_t p)538 memorystatus_freeze_proc_is_refreeze_eligible(proc_t p)
539 {
540 return (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) != 0;
541 }
542
543
544 static proc_t
memorystatus_freeze_pick_refreeze_process(proc_t last_p)545 memorystatus_freeze_pick_refreeze_process(proc_t last_p)
546 {
547 proc_t p = PROC_NULL, next_p = PROC_NULL;
548 unsigned int band = (unsigned int) memorystatus_freeze_jetsam_band;
549 if (last_p == PROC_NULL) {
550 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
551 } else {
552 next_p = memorystatus_get_next_proc_locked(&band, last_p, FALSE);
553 }
554 while (next_p) {
555 p = next_p;
556 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
557 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) && !memorystatus_freeze_proc_is_refreeze_eligible(p)) {
558 /* Process is already frozen & hasn't been thawed. */
559 continue;
560 }
561 /*
562 * Has to have been frozen once before.
563 */
564 if (!(p->p_memstat_state & P_MEMSTAT_FROZEN)) {
565 continue;
566 }
567
568 /*
569 * Not currently being looked at for something.
570 */
571 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
572 continue;
573 }
574 /*
575 * Found it
576 */
577 break;
578 }
579 return p;
580 }
581
582 proc_t
memorystatus_freeze_pick_process(struct memorystatus_freeze_list_iterator * iterator)583 memorystatus_freeze_pick_process(struct memorystatus_freeze_list_iterator *iterator)
584 {
585 proc_t p = PROC_NULL, next_p = PROC_NULL;
586 unsigned int band = JETSAM_PRIORITY_IDLE;
587
588 LCK_MTX_ASSERT(&proc_list_mlock, LCK_MTX_ASSERT_OWNED);
589 /*
590 * If the freezer is full, only consider refreezes.
591 */
592 if (iterator->refreeze_only || memorystatus_frozen_count >= memorystatus_frozen_processes_max) {
593 if (!iterator->refreeze_only) {
594 /*
595 * The first time the iterator starts to return refreeze
596 * candidates, we need to reset the last pointer b/c it's pointing into the wrong band.
597 */
598 iterator->last_p = PROC_NULL;
599 iterator->refreeze_only = true;
600 }
601 iterator->last_p = memorystatus_freeze_pick_refreeze_process(iterator->last_p);
602 return iterator->last_p;
603 }
604
605 /*
606 * Search for the next freezer candidate.
607 */
608 if (memorystatus_freezer_use_ordered_list) {
609 next_p = memorystatus_freezer_candidate_list_get_proc(
610 &memorystatus_global_freeze_list,
611 (iterator->global_freeze_list_index)++,
612 &memorystatus_freezer_stats.mfs_freeze_pid_mismatches);
613 } else if (iterator->last_p == PROC_NULL) {
614 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
615 } else {
616 next_p = memorystatus_get_next_proc_locked(&band, iterator->last_p, FALSE);
617 }
618 while (next_p) {
619 p = next_p;
620 if (memorystatus_is_process_eligible_for_freeze(p)) {
621 iterator->last_p = p;
622 return iterator->last_p;
623 } else {
624 if (memorystatus_freezer_use_ordered_list) {
625 next_p = memorystatus_freezer_candidate_list_get_proc(
626 &memorystatus_global_freeze_list,
627 (iterator->global_freeze_list_index)++,
628 &memorystatus_freezer_stats.mfs_freeze_pid_mismatches);
629 } else {
630 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
631 }
632 }
633 }
634
635 /*
636 * Failed to find a new freezer candidate.
637 * Try to re-freeze.
638 */
639 if (memorystatus_refreeze_eligible_count >= MIN_THAW_REFREEZE_THRESHOLD) {
640 assert(!iterator->refreeze_only);
641 iterator->refreeze_only = true;
642 iterator->last_p = memorystatus_freeze_pick_refreeze_process(PROC_NULL);
643 return iterator->last_p;
644 }
645 return PROC_NULL;
646 }
647
648 /*
649 * memorystatus_pages_update calls this function whenever the number
650 * of available pages changes. It wakes the freezer thread iff the function returns
651 * true. The freezer thread will try to freeze (or refreeze) up to 1 process
652 * before blocking again.
653 *
654 * Note the freezer thread is also woken up by memorystatus_on_inactivity.
655 */
656
657 bool
memorystatus_freeze_thread_should_run()658 memorystatus_freeze_thread_should_run()
659 {
660 /*
661 * No freezer_mutex held here...see why near call-site
662 * within memorystatus_pages_update().
663 */
664
665 if (memorystatus_freeze_enabled == FALSE) {
666 return false;
667 }
668
669 if (memorystatus_available_pages > memorystatus_freeze_threshold) {
670 return false;
671 }
672
673 memorystatus_freezer_stats.mfs_below_threshold_count++;
674
675 if ((memorystatus_frozen_count >= memorystatus_frozen_processes_max)) {
676 /*
677 * Consider this as a skip even if we wake up to refreeze because
678 * we won't freeze any new procs.
679 */
680 memorystatus_freezer_stats.mfs_skipped_full_count++;
681 if (memorystatus_refreeze_eligible_count < MIN_THAW_REFREEZE_THRESHOLD) {
682 return false;
683 }
684 }
685
686 if (memorystatus_frozen_shared_mb_max && (memorystatus_frozen_shared_mb >= memorystatus_frozen_shared_mb_max)) {
687 memorystatus_freezer_stats.mfs_skipped_shared_mb_high_count++;
688 return false;
689 }
690
691 uint64_t curr_time = mach_absolute_time();
692
693 if (curr_time < memorystatus_freezer_thread_next_run_ts) {
694 return false;
695 }
696
697 return true;
698 }
699
700 size_t
memorystatus_pick_freeze_count_for_wakeup()701 memorystatus_pick_freeze_count_for_wakeup()
702 {
703 size_t num_to_freeze = 0;
704 if (!memorystatus_swap_all_apps) {
705 num_to_freeze = 1;
706 } else {
707 /*
708 * When app swap is enabled, we want the freezer thread to aggressively freeze
709 * all candidates so we clear out space for the fg working set.
710 * But we still cap it to the current size of the candidate bands to avoid
711 * consuming excessive CPU if there's a lot of churn in the candidate band.
712 */
713 proc_list_lock();
714 for (unsigned int band = JETSAM_PRIORITY_IDLE; band <= memorystatus_freeze_max_candidate_band; band++) {
715 num_to_freeze += memstat_bucket[band].count;
716 }
717 proc_list_unlock();
718 }
719
720 return num_to_freeze;
721 }
722
723 #endif /* CONFIG_FREEZE */
724