]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/kern_memorystatus.c
projects / apple / xnu.git / blob
? search:


5c3410624cf99db7f39d92748d3646caab3403f9
[apple/xnu.git] / bsd / kern / kern_memorystatus.c
1 /*
2 * Copyright (c) 2006-2018 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 <libkern/libkern.h>
44 #include <mach/coalition.h>
45 #include <mach/mach_time.h>
46 #include <mach/task.h>
47 #include <mach/host_priv.h>
48 #include <mach/mach_host.h>
49 #include <os/log.h>
50 #include <pexpert/pexpert.h>
51 #include <sys/coalition.h>
52 #include <sys/kern_event.h>
53 #include <sys/proc.h>
54 #include <sys/proc_info.h>
55 #include <sys/reason.h>
56 #include <sys/signal.h>
57 #include <sys/signalvar.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysproto.h>
60 #include <sys/wait.h>
61 #include <sys/tree.h>
62 #include <sys/priv.h>
63 #include <vm/vm_pageout.h>
64 #include <vm/vm_protos.h>
65
66 #if CONFIG_FREEZE
67 #include <vm/vm_map.h>
68 #endif /* CONFIG_FREEZE */
69
70 #include <sys/kern_memorystatus.h>
71
72 #include <mach/machine/sdt.h>
73 #include <libkern/section_keywords.h>
74 #include <stdatomic.h>
75
76 /* For logging clarity */
77 static const char *memorystatus_kill_cause_name[] = {
78 "", /* kMemorystatusInvalid */
79 "jettisoned", /* kMemorystatusKilled */
80 "highwater", /* kMemorystatusKilledHiwat */
81 "vnode-limit", /* kMemorystatusKilledVnodes */
82 "vm-pageshortage", /* kMemorystatusKilledVMPageShortage */
83 "proc-thrashing", /* kMemorystatusKilledProcThrashing */
84 "fc-thrashing", /* kMemorystatusKilledFCThrashing */
85 "per-process-limit", /* kMemorystatusKilledPerProcessLimit */
86 "disk-space-shortage", /* kMemorystatusKilledDiskSpaceShortage */
87 "idle-exit", /* kMemorystatusKilledIdleExit */
88 "zone-map-exhaustion", /* kMemorystatusKilledZoneMapExhaustion */
89 "vm-compressor-thrashing", /* kMemorystatusKilledVMCompressorThrashing */
90 "vm-compressor-space-shortage", /* kMemorystatusKilledVMCompressorSpaceShortage */
91 };
92
93 static const char *
94 memorystatus_priority_band_name(int32_t priority)
95 {
96 switch (priority) {
97 case JETSAM_PRIORITY_FOREGROUND:
98 return "FOREGROUND";
99 case JETSAM_PRIORITY_AUDIO_AND_ACCESSORY:
100 return "AUDIO_AND_ACCESSORY";
101 case JETSAM_PRIORITY_CONDUCTOR:
102 return "CONDUCTOR";
103 case JETSAM_PRIORITY_HOME:
104 return "HOME";
105 case JETSAM_PRIORITY_EXECUTIVE:
106 return "EXECUTIVE";
107 case JETSAM_PRIORITY_IMPORTANT:
108 return "IMPORTANT";
109 case JETSAM_PRIORITY_CRITICAL:
110 return "CRITICAL";
111 }
112
113 return "?";
114 }
115
116 /* Does cause indicate vm or fc thrashing? */
117 static boolean_t
118 is_reason_thrashing(unsigned cause)
119 {
120 switch (cause) {
121 case kMemorystatusKilledFCThrashing:
122 case kMemorystatusKilledVMCompressorThrashing:
123 case kMemorystatusKilledVMCompressorSpaceShortage:
124 return TRUE;
125 default:
126 return FALSE;
127 }
128 }
129
130 /* Is the zone map almost full? */
131 static boolean_t
132 is_reason_zone_map_exhaustion(unsigned cause)
133 {
134 if (cause == kMemorystatusKilledZoneMapExhaustion) {
135 return TRUE;
136 }
137 return FALSE;
138 }
139
140 /*
141 * Returns the current zone map size and capacity to include in the jetsam snapshot.
142 * Defined in zalloc.c
143 */
144 extern void get_zone_map_size(uint64_t *current_size, uint64_t *capacity);
145
146 /*
147 * Returns the name of the largest zone and its size to include in the jetsam snapshot.
148 * Defined in zalloc.c
149 */
150 extern void get_largest_zone_info(char *zone_name, size_t zone_name_len, uint64_t *zone_size);
151
152 /* These are very verbose printfs(), enable with
153 * MEMORYSTATUS_DEBUG_LOG
154 */
155 #if MEMORYSTATUS_DEBUG_LOG
156 #define MEMORYSTATUS_DEBUG(cond, format, ...) \
157 do { \
158 if (cond) { printf(format, ##__VA_ARGS__); } \
159 } while(0)
160 #else
161 #define MEMORYSTATUS_DEBUG(cond, format, ...)
162 #endif
163
164 /*
165 * Active / Inactive limit support
166 * proc list must be locked
167 *
168 * The SET_*** macros are used to initialize a limit
169 * for the first time.
170 *
171 * The CACHE_*** macros are use to cache the limit that will
172 * soon be in effect down in the ledgers.
173 */
174
175 #define SET_ACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
176 MACRO_BEGIN \
177 (p)->p_memstat_memlimit_active = (limit); \
178 if (is_fatal) { \
179 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
180 } else { \
181 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
182 } \
183 MACRO_END
184
185 #define SET_INACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
186 MACRO_BEGIN \
187 (p)->p_memstat_memlimit_inactive = (limit); \
188 if (is_fatal) { \
189 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
190 } else { \
191 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
192 } \
193 MACRO_END
194
195 #define CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal) \
196 MACRO_BEGIN \
197 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_active; \
198 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) { \
199 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
200 is_fatal = TRUE; \
201 } else { \
202 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
203 is_fatal = FALSE; \
204 } \
205 MACRO_END
206
207 #define CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal) \
208 MACRO_BEGIN \
209 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_inactive; \
210 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) { \
211 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
212 is_fatal = TRUE; \
213 } else { \
214 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
215 is_fatal = FALSE; \
216 } \
217 MACRO_END
218
219
220 /* General tunables */
221
222 unsigned long delta_percentage = 5;
223 unsigned long critical_threshold_percentage = 5;
224 unsigned long idle_offset_percentage = 5;
225 unsigned long pressure_threshold_percentage = 15;
226 unsigned long freeze_threshold_percentage = 50;
227 unsigned long policy_more_free_offset_percentage = 5;
228
229 /* General memorystatus stuff */
230
231 struct klist memorystatus_klist;
232 static lck_mtx_t memorystatus_klist_mutex;
233
234 static void memorystatus_klist_lock(void);
235 static void memorystatus_klist_unlock(void);
236
237 static uint64_t memorystatus_sysprocs_idle_delay_time = 0;
238 static uint64_t memorystatus_apps_idle_delay_time = 0;
239
240 /*
241 * Memorystatus kevents
242 */
243
244 static int filt_memorystatusattach(struct knote *kn, struct kevent_internal_s *kev);
245 static void filt_memorystatusdetach(struct knote *kn);
246 static int filt_memorystatus(struct knote *kn, long hint);
247 static int filt_memorystatustouch(struct knote *kn, struct kevent_internal_s *kev);
248 static int filt_memorystatusprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev);
249
250 SECURITY_READ_ONLY_EARLY(struct filterops) memorystatus_filtops = {
251 .f_attach = filt_memorystatusattach,
252 .f_detach = filt_memorystatusdetach,
253 .f_event = filt_memorystatus,
254 .f_touch = filt_memorystatustouch,
255 .f_process = filt_memorystatusprocess,
256 };
257
258 enum {
259 kMemorystatusNoPressure = 0x1,
260 kMemorystatusPressure = 0x2,
261 kMemorystatusLowSwap = 0x4,
262 kMemorystatusProcLimitWarn = 0x8,
263 kMemorystatusProcLimitCritical = 0x10
264 };
265
266 /* Idle guard handling */
267
268 static int32_t memorystatus_scheduled_idle_demotions_sysprocs = 0;
269 static int32_t memorystatus_scheduled_idle_demotions_apps = 0;
270
271 static thread_call_t memorystatus_idle_demotion_call;
272
273 static void memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2);
274 static void memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state);
275 static void memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clean_state);
276 static void memorystatus_reschedule_idle_demotion_locked(void);
277
278 static void memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check);
279
280 int memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap);
281
282 vm_pressure_level_t convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
283
284 boolean_t is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
285 void memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear);
286 void memorystatus_send_low_swap_note(void);
287
288 unsigned int memorystatus_level = 0;
289
290 static int memorystatus_list_count = 0;
291
292
293 #define MEMSTAT_BUCKET_COUNT (JETSAM_PRIORITY_MAX + 1)
294
295 typedef struct memstat_bucket {
296 TAILQ_HEAD(, proc) list;
297 int count;
298 } memstat_bucket_t;
299
300 memstat_bucket_t memstat_bucket[MEMSTAT_BUCKET_COUNT];
301
302 int memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index);
303
304 uint64_t memstat_idle_demotion_deadline = 0;
305
306 int system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
307 int applications_aging_band = JETSAM_PRIORITY_IDLE;
308
309 #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)))
310
311 /*
312 * Checking the p_memstat_state almost always requires the proc_list_lock
313 * because the jetsam thread could be on the other core changing the state.
314 *
315 * App -- almost always managed by a system process. Always have dirty tracking OFF. Can include extensions too.
316 * System Processes -- not managed by anybody. Always have dirty tracking ON. Can include extensions (here) too.
317 */
318 #define isApp(p) ((p->p_memstat_state & P_MEMSTAT_MANAGED) || ! (p->p_memstat_dirty & P_DIRTY_TRACK))
319 #define isSysProc(p) ( ! (p->p_memstat_state & P_MEMSTAT_MANAGED) || (p->p_memstat_dirty & P_DIRTY_TRACK))
320
321 #define kJetsamAgingPolicyNone (0)
322 #define kJetsamAgingPolicyLegacy (1)
323 #define kJetsamAgingPolicySysProcsReclaimedFirst (2)
324 #define kJetsamAgingPolicyAppsReclaimedFirst (3)
325 #define kJetsamAgingPolicyMax kJetsamAgingPolicyAppsReclaimedFirst
326
327 unsigned int jetsam_aging_policy = kJetsamAgingPolicyLegacy;
328
329 extern int corpse_for_fatal_memkill;
330 extern unsigned long total_corpses_count(void) __attribute__((pure));
331 extern void task_purge_all_corpses(void);
332 extern uint64_t vm_purgeable_purge_task_owned(task_t task);
333 boolean_t memorystatus_allowed_vm_map_fork(task_t);
334 #if DEVELOPMENT || DEBUG
335 void memorystatus_abort_vm_map_fork(task_t);
336 #endif
337
338 #if 0
339
340 /* Keeping around for future use if we need a utility that can do this OR an app that needs a dynamic adjustment. */
341
342 static int
343 sysctl_set_jetsam_aging_policy SYSCTL_HANDLER_ARGS
344 {
345 #pragma unused(oidp, arg1, arg2)
346
347 int error = 0, val = 0;
348 memstat_bucket_t *old_bucket = 0;
349 int old_system_procs_aging_band = 0, new_system_procs_aging_band = 0;
350 int old_applications_aging_band = 0, new_applications_aging_band = 0;
351 proc_t p = NULL, next_proc = NULL;
352
353
354 error = sysctl_io_number(req, jetsam_aging_policy, sizeof(int), &val, NULL);
355 if (error || !req->newptr) {
356 return error;
357 }
358
359 if ((val < 0) || (val > kJetsamAgingPolicyMax)) {
360 printf("jetsam: ordering policy sysctl has invalid value - %d\n", val);
361 return EINVAL;
362 }
363
364 /*
365 * We need to synchronize with any potential adding/removal from aging bands
366 * that might be in progress currently. We use the proc_list_lock() just for
367 * consistency with all the routines dealing with 'aging' processes. We need
368 * a lighterweight lock.
369 */
370 proc_list_lock();
371
372 old_system_procs_aging_band = system_procs_aging_band;
373 old_applications_aging_band = applications_aging_band;
374
375 switch (val) {
376 case kJetsamAgingPolicyNone:
377 new_system_procs_aging_band = JETSAM_PRIORITY_IDLE;
378 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
379 break;
380
381 case kJetsamAgingPolicyLegacy:
382 /*
383 * Legacy behavior where some daemons get a 10s protection once and only before the first clean->dirty->clean transition before going into IDLE band.
384 */
385 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
386 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
387 break;
388
389 case kJetsamAgingPolicySysProcsReclaimedFirst:
390 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
391 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
392 break;
393
394 case kJetsamAgingPolicyAppsReclaimedFirst:
395 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
396 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
397 break;
398
399 default:
400 break;
401 }
402
403 if (old_system_procs_aging_band && (old_system_procs_aging_band != new_system_procs_aging_band)) {
404 old_bucket = &memstat_bucket[old_system_procs_aging_band];
405 p = TAILQ_FIRST(&old_bucket->list);
406
407 while (p) {
408 next_proc = TAILQ_NEXT(p, p_memstat_list);
409
410 if (isSysProc(p)) {
411 if (new_system_procs_aging_band == JETSAM_PRIORITY_IDLE) {
412 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
413 }
414
415 memorystatus_update_priority_locked(p, new_system_procs_aging_band, false, true);
416 }
417
418 p = next_proc;
419 continue;
420 }
421 }
422
423 if (old_applications_aging_band && (old_applications_aging_band != new_applications_aging_band)) {
424 old_bucket = &memstat_bucket[old_applications_aging_band];
425 p = TAILQ_FIRST(&old_bucket->list);
426
427 while (p) {
428 next_proc = TAILQ_NEXT(p, p_memstat_list);
429
430 if (isApp(p)) {
431 if (new_applications_aging_band == JETSAM_PRIORITY_IDLE) {
432 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
433 }
434
435 memorystatus_update_priority_locked(p, new_applications_aging_band, false, true);
436 }
437
438 p = next_proc;
439 continue;
440 }
441 }
442
443 jetsam_aging_policy = val;
444 system_procs_aging_band = new_system_procs_aging_band;
445 applications_aging_band = new_applications_aging_band;
446
447 proc_list_unlock();
448
449 return 0;
450 }
451
452 SYSCTL_PROC(_kern, OID_AUTO, set_jetsam_aging_policy, CTLTYPE_INT | CTLFLAG_RW,
453 0, 0, sysctl_set_jetsam_aging_policy, "I", "Jetsam Aging Policy");
454 #endif /*0*/
455
456 static int
457 sysctl_jetsam_set_sysprocs_idle_delay_time SYSCTL_HANDLER_ARGS
458 {
459 #pragma unused(oidp, arg1, arg2)
460
461 int error = 0, val = 0, old_time_in_secs = 0;
462 uint64_t old_time_in_ns = 0;
463
464 absolutetime_to_nanoseconds(memorystatus_sysprocs_idle_delay_time, &old_time_in_ns);
465 old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
466
467 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
468 if (error || !req->newptr) {
469 return error;
470 }
471
472 if ((val < 0) || (val > INT32_MAX)) {
473 printf("jetsam: new idle delay interval has invalid value.\n");
474 return EINVAL;
475 }
476
477 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
478
479 return 0;
480 }
481
482 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_sysprocs_idle_delay_time, CTLTYPE_INT | CTLFLAG_RW,
483 0, 0, sysctl_jetsam_set_sysprocs_idle_delay_time, "I", "Aging window for system processes");
484
485
486 static int
487 sysctl_jetsam_set_apps_idle_delay_time SYSCTL_HANDLER_ARGS
488 {
489 #pragma unused(oidp, arg1, arg2)
490
491 int error = 0, val = 0, old_time_in_secs = 0;
492 uint64_t old_time_in_ns = 0;
493
494 absolutetime_to_nanoseconds(memorystatus_apps_idle_delay_time, &old_time_in_ns);
495 old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
496
497 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
498 if (error || !req->newptr) {
499 return error;
500 }
501
502 if ((val < 0) || (val > INT32_MAX)) {
503 printf("jetsam: new idle delay interval has invalid value.\n");
504 return EINVAL;
505 }
506
507 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
508
509 return 0;
510 }
511
512 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_apps_idle_delay_time, CTLTYPE_INT | CTLFLAG_RW,
513 0, 0, sysctl_jetsam_set_apps_idle_delay_time, "I", "Aging window for applications");
514
515 SYSCTL_INT(_kern, OID_AUTO, jetsam_aging_policy, CTLTYPE_INT | CTLFLAG_RD, &jetsam_aging_policy, 0, "");
516
517 static unsigned int memorystatus_dirty_count = 0;
518
519 SYSCTL_INT(_kern, OID_AUTO, max_task_pmem, CTLFLAG_RD | CTLFLAG_LOCKED | CTLFLAG_MASKED, &max_task_footprint_mb, 0, "");
520
521 #if CONFIG_EMBEDDED
522
523 SYSCTL_INT(_kern, OID_AUTO, memorystatus_level, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_level, 0, "");
524
525 #endif /* CONFIG_EMBEDDED */
526
527 int
528 memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
529 {
530 user_addr_t level = 0;
531
532 level = args->level;
533
534 if (copyout(&memorystatus_level, level, sizeof(memorystatus_level)) != 0) {
535 return EFAULT;
536 }
537
538 return 0;
539 }
540
541 static proc_t memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search);
542 static proc_t memorystatus_get_next_proc_locked(unsigned int *bucket_index, proc_t p, boolean_t search);
543
544 static void memorystatus_thread(void *param __unused, wait_result_t wr __unused);
545
546 /* Memory Limits */
547
548 static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. */
549
550 static boolean_t proc_jetsam_state_is_active_locked(proc_t);
551 static boolean_t memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
552 static boolean_t memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
553
554
555 static int memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
556
557 static int memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry);
558
559 static int memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
560
561 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);
562
563 int proc_get_memstat_priority(proc_t, boolean_t);
564
565 static boolean_t memorystatus_idle_snapshot = 0;
566
567 unsigned int memorystatus_delta = 0;
568
569 /* Jetsam Loop Detection */
570 static boolean_t memorystatus_jld_enabled = FALSE; /* Enable jetsam loop detection */
571 static uint32_t memorystatus_jld_eval_period_msecs = 0; /* Init pass sets this based on device memory size */
572 static int memorystatus_jld_eval_aggressive_count = 3; /* Raise the priority max after 'n' aggressive loops */
573 static int memorystatus_jld_eval_aggressive_priority_band_max = 15; /* Kill aggressively up through this band */
574
575 /*
576 * A FG app can request that the aggressive jetsam mechanism display some leniency in the FG band. This 'lenient' mode is described as:
577 * --- 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.
578 *
579 * RESTRICTIONS:
580 * - Such a request is respected/acknowledged only once while that 'requesting' app is in the FG band i.e. if aggressive jetsam was
581 * needed and the 'lenient' mode was deployed then that's it for this special mode while the app is in the FG band.
582 *
583 * - 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.
584 *
585 * - Also, the transition of the 'requesting' app away from the FG band will void this special behavior.
586 */
587
588 #define AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD 25
589 boolean_t memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
590 boolean_t memorystatus_aggressive_jetsam_lenient = FALSE;
591
592 #if DEVELOPMENT || DEBUG
593 /*
594 * Jetsam Loop Detection tunables.
595 */
596
597 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_period_msecs, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_period_msecs, 0, "");
598 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_count, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_count, 0, "");
599 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_priority_band_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_priority_band_max, 0, "");
600 #endif /* DEVELOPMENT || DEBUG */
601
602 static uint32_t kill_under_pressure_cause = 0;
603
604 /*
605 * default jetsam snapshot support
606 */
607 static memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot;
608 static memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot_copy;
609 #define memorystatus_jetsam_snapshot_list memorystatus_jetsam_snapshot->entries
610 static unsigned int memorystatus_jetsam_snapshot_count = 0;
611 static unsigned int memorystatus_jetsam_snapshot_copy_count = 0;
612 static unsigned int memorystatus_jetsam_snapshot_max = 0;
613 static unsigned int memorystatus_jetsam_snapshot_size = 0;
614 static uint64_t memorystatus_jetsam_snapshot_last_timestamp = 0;
615 static uint64_t memorystatus_jetsam_snapshot_timeout = 0;
616 #define JETSAM_SNAPSHOT_TIMEOUT_SECS 30
617
618 /*
619 * snapshot support for memstats collected at boot.
620 */
621 static memorystatus_jetsam_snapshot_t memorystatus_at_boot_snapshot;
622
623 static void memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count);
624 static boolean_t memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount);
625 static void memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime);
626
627 static void memorystatus_clear_errors(void);
628 static void memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages);
629 static void memorystatus_get_task_phys_footprint_page_counts(task_t task,
630 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
631 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
632 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
633 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages);
634
635 static void memorystatus_get_task_memory_region_count(task_t task, uint64_t *count);
636
637 static uint32_t memorystatus_build_state(proc_t p);
638 //static boolean_t memorystatus_issue_pressure_kevent(boolean_t pressured);
639
640 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, uint32_t *errors);
641 static boolean_t memorystatus_kill_top_process_aggressive(uint32_t cause, int aggr_count, int32_t priority_max, uint32_t *errors);
642 static boolean_t memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, unsigned int band, int aggr_count, uint32_t *errors);
643 static boolean_t memorystatus_kill_hiwat_proc(uint32_t *errors, boolean_t *purged);
644
645 static boolean_t memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause);
646
647 /* Priority Band Sorting Routines */
648 static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order);
649 static int memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order);
650 static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index);
651 static int memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz);
652
653 /* qsort routines */
654 typedef int (*cmpfunc_t)(const void *a, const void *b);
655 extern void qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
656 static int memstat_asc_cmp(const void *a, const void *b);
657
658 /* VM pressure */
659
660 extern unsigned int vm_page_free_count;
661 extern unsigned int vm_page_active_count;
662 extern unsigned int vm_page_inactive_count;
663 extern unsigned int vm_page_throttled_count;
664 extern unsigned int vm_page_purgeable_count;
665 extern unsigned int vm_page_wire_count;
666 #if CONFIG_SECLUDED_MEMORY
667 extern unsigned int vm_page_secluded_count;
668 #endif /* CONFIG_SECLUDED_MEMORY */
669
670 #if CONFIG_JETSAM
671 unsigned int memorystatus_available_pages = (unsigned int)-1;
672 unsigned int memorystatus_available_pages_pressure = 0;
673 unsigned int memorystatus_available_pages_critical = 0;
674 static unsigned int memorystatus_available_pages_critical_base = 0;
675 static unsigned int memorystatus_available_pages_critical_idle_offset = 0;
676
677 #if DEVELOPMENT || DEBUG
678 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
679 #else
680 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_MASKED | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
681 #endif /* DEVELOPMENT || DEBUG */
682
683 static unsigned int memorystatus_jetsam_policy = kPolicyDefault;
684 unsigned int memorystatus_policy_more_free_offset_pages = 0;
685 static void memorystatus_update_levels_locked(boolean_t critical_only);
686 static unsigned int memorystatus_thread_wasted_wakeup = 0;
687
688 /* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
689 extern void vm_thrashing_jetsam_done(void);
690 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);
691
692 int32_t max_kill_priority = JETSAM_PRIORITY_MAX;
693
694 #else /* CONFIG_JETSAM */
695
696 uint64_t memorystatus_available_pages = (uint64_t)-1;
697 uint64_t memorystatus_available_pages_pressure = (uint64_t)-1;
698 uint64_t memorystatus_available_pages_critical = (uint64_t)-1;
699
700 int32_t max_kill_priority = JETSAM_PRIORITY_IDLE;
701 #endif /* CONFIG_JETSAM */
702
703 unsigned int memorystatus_frozen_count = 0;
704 unsigned int memorystatus_frozen_processes_max = 0;
705 unsigned int memorystatus_frozen_shared_mb = 0;
706 unsigned int memorystatus_frozen_shared_mb_max = 0;
707 unsigned int memorystatus_freeze_shared_mb_per_process_max = 0; /* Max. MB allowed per process to be freezer-eligible. */
708 unsigned int memorystatus_freeze_private_shared_pages_ratio = 2; /* Ratio of private:shared pages for a process to be freezer-eligible. */
709 unsigned int memorystatus_suspended_count = 0;
710 unsigned int memorystatus_thaw_count = 0;
711 unsigned int memorystatus_refreeze_eligible_count = 0; /* # of processes currently thawed i.e. have state on disk & in-memory */
712
713 #if VM_PRESSURE_EVENTS
714
715 boolean_t memorystatus_warn_process(pid_t pid, __unused boolean_t is_active, __unused boolean_t is_fatal, boolean_t exceeded);
716
717 vm_pressure_level_t memorystatus_vm_pressure_level = kVMPressureNormal;
718
719 /*
720 * We use this flag to signal if we have any HWM offenders
721 * on the system. This way we can reduce the number of wakeups
722 * of the memorystatus_thread when the system is between the
723 * "pressure" and "critical" threshold.
724 *
725 * The (re-)setting of this variable is done without any locks
726 * or synchronization simply because it is not possible (currently)
727 * to keep track of HWM offenders that drop down below their memory
728 * limit and/or exit. So, we choose to burn a couple of wasted wakeups
729 * by allowing the unguarded modification of this variable.
730 */
731 boolean_t memorystatus_hwm_candidates = 0;
732
733 static int memorystatus_send_note(int event_code, void *data, size_t data_length);
734
735 /*
736 * This value is the threshold that a process must meet to be considered for scavenging.
737 */
738 #if CONFIG_EMBEDDED
739 #define VM_PRESSURE_MINIMUM_RSIZE 6 /* MB */
740 #else /* CONFIG_EMBEDDED */
741 #define VM_PRESSURE_MINIMUM_RSIZE 10 /* MB */
742 #endif /* CONFIG_EMBEDDED */
743
744 uint32_t vm_pressure_task_footprint_min = VM_PRESSURE_MINIMUM_RSIZE;
745
746 #if DEVELOPMENT || DEBUG
747 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_vm_pressure_task_footprint_min, CTLFLAG_RW | CTLFLAG_LOCKED, &vm_pressure_task_footprint_min, 0, "");
748 #endif /* DEVELOPMENT || DEBUG */
749
750 #endif /* VM_PRESSURE_EVENTS */
751
752
753 #if DEVELOPMENT || DEBUG
754
755 lck_grp_attr_t *disconnect_page_mappings_lck_grp_attr;
756 lck_grp_t *disconnect_page_mappings_lck_grp;
757 static lck_mtx_t disconnect_page_mappings_mutex;
758
759 extern boolean_t kill_on_no_paging_space;
760 #endif /* DEVELOPMENT || DEBUG */
761
762
763 /*
764 * Table that expresses the probability of a process
765 * being used in the next hour.
766 */
767 typedef struct memorystatus_internal_probabilities {
768 char proc_name[MAXCOMLEN + 1];
769 int use_probability;
770 } memorystatus_internal_probabilities_t;
771
772 static memorystatus_internal_probabilities_t *memorystatus_global_probabilities_table = NULL;
773 static size_t memorystatus_global_probabilities_size = 0;
774
775 /* Freeze */
776
777 #if CONFIG_FREEZE
778 boolean_t memorystatus_freeze_enabled = FALSE;
779 int memorystatus_freeze_wakeup = 0;
780 int memorystatus_freeze_jetsam_band = 0; /* the jetsam band which will contain P_MEMSTAT_FROZEN processes */
781
782 lck_grp_attr_t *freezer_lck_grp_attr;
783 lck_grp_t *freezer_lck_grp;
784 static lck_mtx_t freezer_mutex;
785
786 static inline boolean_t memorystatus_can_freeze_processes(void);
787 static boolean_t memorystatus_can_freeze(boolean_t *memorystatus_freeze_swap_low);
788 static boolean_t memorystatus_is_process_eligible_for_freeze(proc_t p);
789 static void memorystatus_freeze_thread(void *param __unused, wait_result_t wr __unused);
790 static boolean_t memorystatus_freeze_thread_should_run(void);
791
792 void memorystatus_disable_freeze(void);
793
794 /* Thresholds */
795 static unsigned int memorystatus_freeze_threshold = 0;
796
797 static unsigned int memorystatus_freeze_pages_min = 0;
798 static unsigned int memorystatus_freeze_pages_max = 0;
799
800 static unsigned int memorystatus_freeze_suspended_threshold = FREEZE_SUSPENDED_THRESHOLD_DEFAULT;
801
802 static unsigned int memorystatus_freeze_daily_mb_max = FREEZE_DAILY_MB_MAX_DEFAULT;
803 static uint64_t memorystatus_freeze_budget_pages_remaining = 0; //remaining # of pages that can be frozen to disk
804 static boolean_t memorystatus_freeze_degradation = FALSE; //protected by the freezer mutex. Signals we are in a degraded freeze mode.
805
806 static unsigned int memorystatus_max_frozen_demotions_daily = 0;
807 static unsigned int memorystatus_thaw_count_demotion_threshold = 0;
808
809 /* Stats */
810 static uint64_t memorystatus_freeze_pageouts = 0;
811
812 /* Throttling */
813 #define DEGRADED_WINDOW_MINS (30)
814 #define NORMAL_WINDOW_MINS (24 * 60)
815
816 static throttle_interval_t throttle_intervals[] = {
817 { DEGRADED_WINDOW_MINS, 1, 0, 0, { 0, 0 }},
818 { NORMAL_WINDOW_MINS, 1, 0, 0, { 0, 0 }},
819 };
820 throttle_interval_t *degraded_throttle_window = &throttle_intervals[0];
821 throttle_interval_t *normal_throttle_window = &throttle_intervals[1];
822
823 extern uint64_t vm_swap_get_free_space(void);
824 extern boolean_t vm_swap_max_budget(uint64_t *);
825
826 static void memorystatus_freeze_update_throttle(uint64_t *budget_pages_allowed);
827
828 static uint64_t memorystatus_freezer_thread_next_run_ts = 0;
829
830 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_count, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_frozen_count, 0, "");
831 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_thaw_count, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_thaw_count, 0, "");
832 SYSCTL_QUAD(_kern, OID_AUTO, memorystatus_freeze_pageouts, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_freeze_pageouts, "");
833 SYSCTL_QUAD(_kern, OID_AUTO, memorystatus_freeze_budget_pages_remaining, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_freeze_budget_pages_remaining, "");
834
835 #endif /* CONFIG_FREEZE */
836
837 /* Debug */
838
839 extern struct knote *vm_find_knote_from_pid(pid_t, struct klist *);
840
841 #if DEVELOPMENT || DEBUG
842
843 static unsigned int memorystatus_debug_dump_this_bucket = 0;
844
845 static void
846 memorystatus_debug_dump_bucket_locked(unsigned int bucket_index)
847 {
848 proc_t p = NULL;
849 uint64_t bytes = 0;
850 int ledger_limit = 0;
851 unsigned int b = bucket_index;
852 boolean_t traverse_all_buckets = FALSE;
853
854 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
855 traverse_all_buckets = TRUE;
856 b = 0;
857 } else {
858 traverse_all_buckets = FALSE;
859 b = bucket_index;
860 }
861
862 /*
863 * footprint reported in [pages / MB ]
864 * limits reported as:
865 * L-limit proc's Ledger limit
866 * C-limit proc's Cached limit, should match Ledger
867 * A-limit proc's Active limit
868 * IA-limit proc's Inactive limit
869 * F==Fatal, NF==NonFatal
870 */
871
872 printf("memorystatus_debug_dump ***START*(PAGE_SIZE_64=%llu)**\n", PAGE_SIZE_64);
873 printf("bucket [pid] [pages / MB] [state] [EP / RP] dirty deadline [L-limit / C-limit / A-limit / IA-limit] name\n");
874 p = memorystatus_get_first_proc_locked(&b, traverse_all_buckets);
875 while (p) {
876 bytes = get_task_phys_footprint(p->task);
877 task_get_phys_footprint_limit(p->task, &ledger_limit);
878 printf("%2d [%5d] [%5lld /%3lldMB] 0x%-8x [%2d / %2d] 0x%-3x %10lld [%3d / %3d%s / %3d%s / %3d%s] %s\n",
879 b, p->p_pid,
880 (bytes / PAGE_SIZE_64), /* task's footprint converted from bytes to pages */
881 (bytes / (1024ULL * 1024ULL)), /* task's footprint converted from bytes to MB */
882 p->p_memstat_state, p->p_memstat_effectivepriority, p->p_memstat_requestedpriority, p->p_memstat_dirty, p->p_memstat_idledeadline,
883 ledger_limit,
884 p->p_memstat_memlimit,
885 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"),
886 p->p_memstat_memlimit_active,
887 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL ? "F " : "NF"),
888 p->p_memstat_memlimit_inactive,
889 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL ? "F " : "NF"),
890 (*p->p_name ? p->p_name : "unknown"));
891 p = memorystatus_get_next_proc_locked(&b, p, traverse_all_buckets);
892 }
893 printf("memorystatus_debug_dump ***END***\n");
894 }
895
896 static int
897 sysctl_memorystatus_debug_dump_bucket SYSCTL_HANDLER_ARGS
898 {
899 #pragma unused(oidp, arg2)
900 int bucket_index = 0;
901 int error;
902 error = SYSCTL_OUT(req, arg1, sizeof(int));
903 if (error || !req->newptr) {
904 return error;
905 }
906 error = SYSCTL_IN(req, &bucket_index, sizeof(int));
907 if (error || !req->newptr) {
908 return error;
909 }
910 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
911 /*
912 * All jetsam buckets will be dumped.
913 */
914 } else {
915 /*
916 * Only a single bucket will be dumped.
917 */
918 }
919
920 proc_list_lock();
921 memorystatus_debug_dump_bucket_locked(bucket_index);
922 proc_list_unlock();
923 memorystatus_debug_dump_this_bucket = bucket_index;
924 return error;
925 }
926
927 /*
928 * Debug aid to look at jetsam buckets and proc jetsam fields.
929 * Use this sysctl to act on a particular jetsam bucket.
930 * Writing the sysctl triggers the dump.
931 * Usage: sysctl kern.memorystatus_debug_dump_this_bucket=<bucket_index>
932 */
933
934 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", "");
935
936
937 /* Debug aid to aid determination of limit */
938
939 static int
940 sysctl_memorystatus_highwater_enable SYSCTL_HANDLER_ARGS
941 {
942 #pragma unused(oidp, arg2)
943 proc_t p;
944 unsigned int b = 0;
945 int error, enable = 0;
946 boolean_t use_active; /* use the active limit and active limit attributes */
947 boolean_t is_fatal;
948
949 error = SYSCTL_OUT(req, arg1, sizeof(int));
950 if (error || !req->newptr) {
951 return error;
952 }
953
954 error = SYSCTL_IN(req, &enable, sizeof(int));
955 if (error || !req->newptr) {
956 return error;
957 }
958
959 if (!(enable == 0 || enable == 1)) {
960 return EINVAL;
961 }
962
963 proc_list_lock();
964
965 p = memorystatus_get_first_proc_locked(&b, TRUE);
966 while (p) {
967 use_active = proc_jetsam_state_is_active_locked(p);
968
969 if (enable) {
970 if (use_active == TRUE) {
971 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
972 } else {
973 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
974 }
975 } else {
976 /*
977 * Disabling limits does not touch the stored variants.
978 * Set the cached limit fields to system_wide defaults.
979 */
980 p->p_memstat_memlimit = -1;
981 p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
982 is_fatal = TRUE;
983 }
984
985 /*
986 * Enforce the cached limit by writing to the ledger.
987 */
988 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit: -1, NULL, use_active, is_fatal);
989
990 p = memorystatus_get_next_proc_locked(&b, p, TRUE);
991 }
992
993 memorystatus_highwater_enabled = enable;
994
995 proc_list_unlock();
996
997 return 0;
998 }
999
1000 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_highwater_enabled, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_highwater_enabled, 0, sysctl_memorystatus_highwater_enable, "I", "");
1001
1002 #if VM_PRESSURE_EVENTS
1003
1004 /*
1005 * This routine is used for targeted notifications regardless of system memory pressure
1006 * and regardless of whether or not the process has already been notified.
1007 * It bypasses and has no effect on the only-one-notification per soft-limit policy.
1008 *
1009 * "memnote" is the current user.
1010 */
1011
1012 static int
1013 sysctl_memorystatus_vm_pressure_send SYSCTL_HANDLER_ARGS
1014 {
1015 #pragma unused(arg1, arg2)
1016
1017 int error = 0, pid = 0;
1018 struct knote *kn = NULL;
1019 boolean_t found_knote = FALSE;
1020 int fflags = 0; /* filter flags for EVFILT_MEMORYSTATUS */
1021 uint64_t value = 0;
1022
1023 error = sysctl_handle_quad(oidp, &value, 0, req);
1024 if (error || !req->newptr) {
1025 return error;
1026 }
1027
1028 /*
1029 * Find the pid in the low 32 bits of value passed in.
1030 */
1031 pid = (int)(value & 0xFFFFFFFF);
1032
1033 /*
1034 * Find notification in the high 32 bits of the value passed in.
1035 */
1036 fflags = (int)((value >> 32) & 0xFFFFFFFF);
1037
1038 /*
1039 * For backwards compatibility, when no notification is
1040 * passed in, default to the NOTE_MEMORYSTATUS_PRESSURE_WARN
1041 */
1042 if (fflags == 0) {
1043 fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
1044 // printf("memorystatus_vm_pressure_send: using default notification [0x%x]\n", fflags);
1045 }
1046
1047 /*
1048 * See event.h ... fflags for EVFILT_MEMORYSTATUS
1049 */
1050 if (!((fflags == NOTE_MEMORYSTATUS_PRESSURE_NORMAL) ||
1051 (fflags == NOTE_MEMORYSTATUS_PRESSURE_WARN) ||
1052 (fflags == NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) ||
1053 (fflags == NOTE_MEMORYSTATUS_LOW_SWAP) ||
1054 (fflags == NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) ||
1055 (fflags == NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) ||
1056 (((fflags & NOTE_MEMORYSTATUS_MSL_STATUS) != 0 &&
1057 ((fflags & ~NOTE_MEMORYSTATUS_MSL_STATUS) == 0))))) {
1058 printf("memorystatus_vm_pressure_send: notification [0x%x] not supported \n", fflags);
1059 error = 1;
1060 return error;
1061 }
1062
1063 /*
1064 * Forcibly send pid a memorystatus notification.
1065 */
1066
1067 memorystatus_klist_lock();
1068
1069 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
1070 proc_t knote_proc = knote_get_kq(kn)->kq_p;
1071 pid_t knote_pid = knote_proc->p_pid;
1072
1073 if (knote_pid == pid) {
1074 /*
1075 * Forcibly send this pid a memorystatus notification.
1076 */
1077 kn->kn_fflags = fflags;
1078 found_knote = TRUE;
1079 }
1080 }
1081
1082 if (found_knote) {
1083 KNOTE(&memorystatus_klist, 0);
1084 printf("memorystatus_vm_pressure_send: (value 0x%llx) notification [0x%x] sent to process [%d] \n", value, fflags, pid);
1085 error = 0;
1086 } else {
1087 printf("memorystatus_vm_pressure_send: (value 0x%llx) notification [0x%x] not sent to process [%d] (none registered?)\n", value, fflags, pid);
1088 error = 1;
1089 }
1090
1091 memorystatus_klist_unlock();
1092
1093 return error;
1094 }
1095
1096 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_send, CTLTYPE_QUAD | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
1097 0, 0, &sysctl_memorystatus_vm_pressure_send, "Q", "");
1098
1099 #endif /* VM_PRESSURE_EVENTS */
1100
1101 SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
1102
1103 #if CONFIG_JETSAM
1104 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
1105 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
1106 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
1107 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_policy_more_free_offset_pages, CTLFLAG_RW, &memorystatus_policy_more_free_offset_pages, 0, "");
1108
1109 static unsigned int memorystatus_jetsam_panic_debug = 0;
1110 static unsigned int memorystatus_jetsam_policy_offset_pages_diagnostic = 0;
1111
1112 /* Diagnostic code */
1113
1114 enum {
1115 kJetsamDiagnosticModeNone = 0,
1116 kJetsamDiagnosticModeAll = 1,
1117 kJetsamDiagnosticModeStopAtFirstActive = 2,
1118 kJetsamDiagnosticModeCount
1119 } jetsam_diagnostic_mode = kJetsamDiagnosticModeNone;
1120
1121 static int jetsam_diagnostic_suspended_one_active_proc = 0;
1122
1123 static int
1124 sysctl_jetsam_diagnostic_mode SYSCTL_HANDLER_ARGS
1125 {
1126 #pragma unused(arg1, arg2)
1127
1128 const char *diagnosticStrings[] = {
1129 "jetsam: diagnostic mode: resetting critical level.",
1130 "jetsam: diagnostic mode: will examine all processes",
1131 "jetsam: diagnostic mode: will stop at first active process"
1132 };
1133
1134 int error, val = jetsam_diagnostic_mode;
1135 boolean_t changed = FALSE;
1136
1137 error = sysctl_handle_int(oidp, &val, 0, req);
1138 if (error || !req->newptr) {
1139 return error;
1140 }
1141 if ((val < 0) || (val >= kJetsamDiagnosticModeCount)) {
1142 printf("jetsam: diagnostic mode: invalid value - %d\n", val);
1143 return EINVAL;
1144 }
1145
1146 proc_list_lock();
1147
1148 if ((unsigned int) val != jetsam_diagnostic_mode) {
1149 jetsam_diagnostic_mode = val;
1150
1151 memorystatus_jetsam_policy &= ~kPolicyDiagnoseActive;
1152
1153 switch (jetsam_diagnostic_mode) {
1154 case kJetsamDiagnosticModeNone:
1155 /* Already cleared */
1156 break;
1157 case kJetsamDiagnosticModeAll:
1158 memorystatus_jetsam_policy |= kPolicyDiagnoseAll;
1159 break;
1160 case kJetsamDiagnosticModeStopAtFirstActive:
1161 memorystatus_jetsam_policy |= kPolicyDiagnoseFirst;
1162 break;
1163 default:
1164 /* Already validated */
1165 break;
1166 }
1167
1168 memorystatus_update_levels_locked(FALSE);
1169 changed = TRUE;
1170 }
1171
1172 proc_list_unlock();
1173
1174 if (changed) {
1175 printf("%s\n", diagnosticStrings[val]);
1176 }
1177
1178 return 0;
1179 }
1180
1181 SYSCTL_PROC(_debug, OID_AUTO, jetsam_diagnostic_mode, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY,
1182 &jetsam_diagnostic_mode, 0, sysctl_jetsam_diagnostic_mode, "I", "Jetsam Diagnostic Mode");
1183
1184 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jetsam_policy_offset_pages_diagnostic, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_jetsam_policy_offset_pages_diagnostic, 0, "");
1185
1186 #if VM_PRESSURE_EVENTS
1187
1188 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_pressure, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_available_pages_pressure, 0, "");
1189
1190 #endif /* VM_PRESSURE_EVENTS */
1191
1192 #endif /* CONFIG_JETSAM */
1193
1194 #if CONFIG_FREEZE
1195
1196 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_jetsam_band, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_jetsam_band, 0, "");
1197 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_daily_mb_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_daily_mb_max, 0, "");
1198 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_degraded_mode, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_freeze_degradation, 0, "");
1199
1200 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_threshold, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_threshold, 0, "");
1201
1202 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_pages_min, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_pages_min, 0, "");
1203 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_pages_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_pages_max, 0, "");
1204
1205 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_refreeze_eligible_count, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_refreeze_eligible_count, 0, "");
1206 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_processes_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_frozen_processes_max, 0, "");
1207
1208 /*
1209 * Max. shared-anonymous memory in MB that can be held by frozen processes in the high jetsam band.
1210 * "0" means no limit.
1211 * Default is 10% of system-wide task limit.
1212 */
1213
1214 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_shared_mb_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_frozen_shared_mb_max, 0, "");
1215 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_shared_mb, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_frozen_shared_mb, 0, "");
1216
1217 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_shared_mb_per_process_max, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_shared_mb_per_process_max, 0, "");
1218 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_private_shared_pages_ratio, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_private_shared_pages_ratio, 0, "");
1219
1220 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_min_processes, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_suspended_threshold, 0, "");
1221
1222 /*
1223 * max. # of frozen process demotions we will allow in our daily cycle.
1224 */
1225 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_max_freeze_demotions_daily, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_max_frozen_demotions_daily, 0, "");
1226 /*
1227 * min # of thaws needed by a process to protect it from getting demoted into the IDLE band.
1228 */
1229 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_thaw_count_demotion_threshold, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_thaw_count_demotion_threshold, 0, "");
1230
1231 boolean_t memorystatus_freeze_throttle_enabled = TRUE;
1232 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_throttle_enabled, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_throttle_enabled, 0, "");
1233
1234 /*
1235 * When set to true, this keeps frozen processes in the compressor pool in memory, instead of swapping them out to disk.
1236 * Exposed via the sysctl kern.memorystatus_freeze_to_memory.
1237 */
1238 boolean_t memorystatus_freeze_to_memory = FALSE;
1239 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_to_memory, CTLFLAG_RW | CTLFLAG_LOCKED, &memorystatus_freeze_to_memory, 0, "");
1240
1241 #define VM_PAGES_FOR_ALL_PROCS (2)
1242 /*
1243 * Manual trigger of freeze and thaw for dev / debug kernels only.
1244 */
1245 static int
1246 sysctl_memorystatus_freeze SYSCTL_HANDLER_ARGS
1247 {
1248 #pragma unused(arg1, arg2)
1249 int error, pid = 0;
1250 proc_t p;
1251 int freezer_error_code = 0;
1252
1253 if (memorystatus_freeze_enabled == FALSE) {
1254 printf("sysctl_freeze: Freeze is DISABLED\n");
1255 return ENOTSUP;
1256 }
1257
1258 error = sysctl_handle_int(oidp, &pid, 0, req);
1259 if (error || !req->newptr) {
1260 return error;
1261 }
1262
1263 if (pid == VM_PAGES_FOR_ALL_PROCS) {
1264 vm_pageout_anonymous_pages();
1265
1266 return 0;
1267 }
1268
1269 lck_mtx_lock(&freezer_mutex);
1270
1271 p = proc_find(pid);
1272 if (p != NULL) {
1273 uint32_t purgeable, wired, clean, dirty, shared;
1274 uint32_t max_pages = 0, state = 0;
1275
1276 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
1277 /*
1278 * Freezer backed by the compressor and swap file(s)
1279 * will hold compressed data.
1280 *
1281 * Set the sysctl kern.memorystatus_freeze_to_memory to true to keep compressed data from
1282 * being swapped out to disk. Note that this disables freezer swap support globally,
1283 * not just for the process being frozen.
1284 *
1285 *
1286 * We don't care about the global freezer budget or the process's (min/max) budget here.
1287 * The freeze sysctl is meant to force-freeze a process.
1288 *
1289 * We also don't update any global or process stats on this path, so that the jetsam/ freeze
1290 * logic remains unaffected. The tasks we're performing here are: freeze the process, set the
1291 * P_MEMSTAT_FROZEN bit, and elevate the process to a higher band (if the freezer is active).
1292 */
1293 max_pages = memorystatus_freeze_pages_max;
1294 } else {
1295 /*
1296 * We only have the compressor without any swap.
1297 */
1298 max_pages = UINT32_MAX - 1;
1299 }
1300
1301 proc_list_lock();
1302 state = p->p_memstat_state;
1303 proc_list_unlock();
1304
1305 /*
1306 * The jetsam path also verifies that the process is a suspended App. We don't care about that here.
1307 * We simply ensure that jetsam is not already working on the process and that the process has not
1308 * explicitly disabled freezing.
1309 */
1310 if (state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FREEZE_DISABLED)) {
1311 printf("sysctl_freeze: p_memstat_state check failed, process is%s%s%s\n",
1312 (state & P_MEMSTAT_TERMINATED) ? " terminated" : "",
1313 (state & P_MEMSTAT_LOCKED) ? " locked" : "",
1314 (state & P_MEMSTAT_FREEZE_DISABLED) ? " unfreezable" : "");
1315
1316 proc_rele(p);
1317 lck_mtx_unlock(&freezer_mutex);
1318 return EPERM;
1319 }
1320
1321 error = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, &freezer_error_code, FALSE /* eval only */);
1322
1323 if (error) {
1324 char reason[128];
1325 if (freezer_error_code == FREEZER_ERROR_EXCESS_SHARED_MEMORY) {
1326 strlcpy(reason, "too much shared memory", 128);
1327 }
1328
1329 if (freezer_error_code == FREEZER_ERROR_LOW_PRIVATE_SHARED_RATIO) {
1330 strlcpy(reason, "low private-shared pages ratio", 128);
1331 }
1332
1333 if (freezer_error_code == FREEZER_ERROR_NO_COMPRESSOR_SPACE) {
1334 strlcpy(reason, "no compressor space", 128);
1335 }
1336
1337 if (freezer_error_code == FREEZER_ERROR_NO_SWAP_SPACE) {
1338 strlcpy(reason, "no swap space", 128);
1339 }
1340
1341 printf("sysctl_freeze: task_freeze failed: %s\n", reason);
1342
1343 if (error == KERN_NO_SPACE) {
1344 /* Make it easy to distinguish between failures due to low compressor/ swap space and other failures. */
1345 error = ENOSPC;
1346 } else {
1347 error = EIO;
1348 }
1349 } else {
1350 proc_list_lock();
1351 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) == 0) {
1352 p->p_memstat_state |= P_MEMSTAT_FROZEN;
1353 memorystatus_frozen_count++;
1354 }
1355 p->p_memstat_frozen_count++;
1356
1357
1358 proc_list_unlock();
1359
1360 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
1361 /*
1362 * We elevate only if we are going to swap out the data.
1363 */
1364 error = memorystatus_update_inactive_jetsam_priority_band(pid, MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE,
1365 memorystatus_freeze_jetsam_band, TRUE);
1366
1367 if (error) {
1368 printf("sysctl_freeze: Elevating frozen process to higher jetsam band failed with %d\n", error);
1369 }
1370 }
1371 }
1372
1373 proc_rele(p);
1374
1375 lck_mtx_unlock(&freezer_mutex);
1376 return error;
1377 } else {
1378 printf("sysctl_freeze: Invalid process\n");
1379 }
1380
1381
1382 lck_mtx_unlock(&freezer_mutex);
1383 return EINVAL;
1384 }
1385
1386 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_freeze, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
1387 0, 0, &sysctl_memorystatus_freeze, "I", "");
1388
1389 static int
1390 sysctl_memorystatus_available_pages_thaw SYSCTL_HANDLER_ARGS
1391 {
1392 #pragma unused(arg1, arg2)
1393
1394 int error, pid = 0;
1395 proc_t p;
1396
1397 if (memorystatus_freeze_enabled == FALSE) {
1398 return ENOTSUP;
1399 }
1400
1401 error = sysctl_handle_int(oidp, &pid, 0, req);
1402 if (error || !req->newptr) {
1403 return error;
1404 }
1405
1406 if (pid == VM_PAGES_FOR_ALL_PROCS) {
1407 do_fastwake_warmup_all();
1408 return 0;
1409 } else {
1410 p = proc_find(pid);
1411 if (p != NULL) {
1412 error = task_thaw(p->task);
1413
1414 if (error) {
1415 error = EIO;
1416 } else {
1417 /*
1418 * task_thaw() succeeded.
1419 *
1420 * We increment memorystatus_frozen_count on the sysctl freeze path.
1421 * And so we need the P_MEMSTAT_FROZEN to decrement the frozen count
1422 * when this process exits.
1423 *
1424 * proc_list_lock();
1425 * p->p_memstat_state &= ~P_MEMSTAT_FROZEN;
1426 * proc_list_unlock();
1427 */
1428 }
1429 proc_rele(p);
1430 return error;
1431 }
1432 }
1433
1434 return EINVAL;
1435 }
1436
1437 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_thaw, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
1438 0, 0, &sysctl_memorystatus_available_pages_thaw, "I", "");
1439
1440 typedef struct _global_freezable_status {
1441 boolean_t freeze_pages_threshold_crossed;
1442 boolean_t freeze_eligible_procs_available;
1443 boolean_t freeze_scheduled_in_future;
1444 }global_freezable_status_t;
1445
1446 typedef struct _proc_freezable_status {
1447 boolean_t freeze_has_memstat_state;
1448 boolean_t freeze_has_pages_min;
1449 int freeze_has_probability;
1450 boolean_t freeze_attempted;
1451 uint32_t p_memstat_state;
1452 uint32_t p_pages;
1453 int p_freeze_error_code;
1454 int p_pid;
1455 char p_name[MAXCOMLEN + 1];
1456 }proc_freezable_status_t;
1457
1458 #define MAX_FREEZABLE_PROCESSES 100
1459
1460 static int
1461 memorystatus_freezer_get_status(user_addr_t buffer, size_t buffer_size, int32_t *retval)
1462 {
1463 uint32_t proc_count = 0, i = 0;
1464 global_freezable_status_t *list_head;
1465 proc_freezable_status_t *list_entry;
1466 size_t list_size = 0;
1467 proc_t p;
1468 memstat_bucket_t *bucket;
1469 uint32_t state = 0, pages = 0, entry_count = 0;
1470 boolean_t try_freeze = TRUE;
1471 int error = 0, probability_of_use = 0;
1472
1473
1474 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE == FALSE) {
1475 return ENOTSUP;
1476 }
1477
1478 list_size = sizeof(global_freezable_status_t) + (sizeof(proc_freezable_status_t) * MAX_FREEZABLE_PROCESSES);
1479
1480 if (buffer_size < list_size) {
1481 return EINVAL;
1482 }
1483
1484 list_head = (global_freezable_status_t*)kalloc(list_size);
1485 if (list_head == NULL) {
1486 return ENOMEM;
1487 }
1488
1489 memset(list_head, 0, list_size);
1490
1491 list_size = sizeof(global_freezable_status_t);
1492
1493 proc_list_lock();
1494
1495 uint64_t curr_time = mach_absolute_time();
1496
1497 list_head->freeze_pages_threshold_crossed = (memorystatus_available_pages < memorystatus_freeze_threshold);
1498 list_head->freeze_eligible_procs_available = ((memorystatus_suspended_count - memorystatus_frozen_count) > memorystatus_freeze_suspended_threshold);
1499 list_head->freeze_scheduled_in_future = (curr_time < memorystatus_freezer_thread_next_run_ts);
1500
1501 list_entry = (proc_freezable_status_t*) ((uintptr_t)list_head + sizeof(global_freezable_status_t));
1502
1503 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
1504
1505 entry_count = (memorystatus_global_probabilities_size / sizeof(memorystatus_internal_probabilities_t));
1506
1507 p = memorystatus_get_first_proc_locked(&i, FALSE);
1508 proc_count++;
1509
1510 while ((proc_count <= MAX_FREEZABLE_PROCESSES) &&
1511 (p) &&
1512 (list_size < buffer_size)) {
1513 if (isApp(p) == FALSE) {
1514 p = memorystatus_get_next_proc_locked(&i, p, FALSE);
1515 proc_count++;
1516 continue;
1517 }
1518
1519 strlcpy(list_entry->p_name, p->p_name, MAXCOMLEN + 1);
1520
1521 list_entry->p_pid = p->p_pid;
1522
1523 state = p->p_memstat_state;
1524
1525 if ((state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FREEZE_DISABLED | P_MEMSTAT_FREEZE_IGNORE)) ||
1526 !(state & P_MEMSTAT_SUSPENDED)) {
1527 try_freeze = list_entry->freeze_has_memstat_state = FALSE;
1528 } else {
1529 try_freeze = list_entry->freeze_has_memstat_state = TRUE;
1530 }
1531
1532 list_entry->p_memstat_state = state;
1533
1534 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1535 if (pages < memorystatus_freeze_pages_min) {
1536 try_freeze = list_entry->freeze_has_pages_min = FALSE;
1537 } else {
1538 list_entry->freeze_has_pages_min = TRUE;
1539 if (try_freeze != FALSE) {
1540 try_freeze = TRUE;
1541 }
1542 }
1543
1544 list_entry->p_pages = pages;
1545
1546 if (entry_count) {
1547 uint32_t j = 0;
1548 for (j = 0; j < entry_count; j++) {
1549 if (strncmp(memorystatus_global_probabilities_table[j].proc_name,
1550 p->p_name,
1551 MAXCOMLEN + 1) == 0) {
1552 probability_of_use = memorystatus_global_probabilities_table[j].use_probability;
1553 break;
1554 }
1555 }
1556
1557 list_entry->freeze_has_probability = probability_of_use;
1558
1559 if (probability_of_use && try_freeze != FALSE) {
1560 try_freeze = TRUE;
1561 } else {
1562 try_freeze = FALSE;
1563 }
1564 } else {
1565 if (try_freeze != FALSE) {
1566 try_freeze = TRUE;
1567 }
1568 list_entry->freeze_has_probability = -1;
1569 }
1570
1571 if (try_freeze) {
1572 uint32_t purgeable, wired, clean, dirty, shared;
1573 uint32_t max_pages = 0;
1574 int freezer_error_code = 0;
1575
1576 error = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, &freezer_error_code, TRUE /* eval only */);
1577
1578 if (error) {
1579 list_entry->p_freeze_error_code = freezer_error_code;
1580 }
1581
1582 list_entry->freeze_attempted = TRUE;
1583 }
1584
1585 list_entry++;
1586
1587 list_size += sizeof(proc_freezable_status_t);
1588
1589 p = memorystatus_get_next_proc_locked(&i, p, FALSE);
1590 proc_count++;
1591 }
1592
1593 proc_list_unlock();
1594
1595 buffer_size = list_size;
1596
1597 error = copyout(list_head, buffer, buffer_size);
1598 if (error == 0) {
1599 *retval = buffer_size;
1600 } else {
1601 *retval = 0;
1602 }
1603
1604 list_size = sizeof(global_freezable_status_t) + (sizeof(proc_freezable_status_t) * MAX_FREEZABLE_PROCESSES);
1605 kfree(list_head, list_size);
1606
1607 MEMORYSTATUS_DEBUG(1, "memorystatus_freezer_get_status: returning %d (%lu - size)\n", error, (unsigned long)*list_size);
1608
1609 return error;
1610 }
1611
1612 static int
1613 memorystatus_freezer_control(int32_t flags, user_addr_t buffer, size_t buffer_size, int32_t *retval)
1614 {
1615 int err = ENOTSUP;
1616
1617 if (flags == FREEZER_CONTROL_GET_STATUS) {
1618 err = memorystatus_freezer_get_status(buffer, buffer_size, retval);
1619 }
1620
1621 return err;
1622 }
1623
1624 #endif /* CONFIG_FREEZE */
1625
1626 #endif /* DEVELOPMENT || DEBUG */
1627
1628 extern kern_return_t kernel_thread_start_priority(thread_continue_t continuation,
1629 void *parameter,
1630 integer_t priority,
1631 thread_t *new_thread);
1632
1633 #if DEVELOPMENT || DEBUG
1634
1635 static int
1636 sysctl_memorystatus_disconnect_page_mappings SYSCTL_HANDLER_ARGS
1637 {
1638 #pragma unused(arg1, arg2)
1639 int error = 0, pid = 0;
1640 proc_t p;
1641
1642 error = sysctl_handle_int(oidp, &pid, 0, req);
1643 if (error || !req->newptr) {
1644 return error;
1645 }
1646
1647 lck_mtx_lock(&disconnect_page_mappings_mutex);
1648
1649 if (pid == -1) {
1650 vm_pageout_disconnect_all_pages();
1651 } else {
1652 p = proc_find(pid);
1653
1654 if (p != NULL) {
1655 error = task_disconnect_page_mappings(p->task);
1656
1657 proc_rele(p);
1658
1659 if (error) {
1660 error = EIO;
1661 }
1662 } else {
1663 error = EINVAL;
1664 }
1665 }
1666 lck_mtx_unlock(&disconnect_page_mappings_mutex);
1667
1668 return error;
1669 }
1670
1671 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_disconnect_page_mappings, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
1672 0, 0, &sysctl_memorystatus_disconnect_page_mappings, "I", "");
1673
1674 #endif /* DEVELOPMENT || DEBUG */
1675
1676
1677 /*
1678 * Picks the sorting routine for a given jetsam priority band.
1679 *
1680 * Input:
1681 * bucket_index - jetsam priority band to be sorted.
1682 * sort_order - JETSAM_SORT_xxx from kern_memorystatus.h
1683 * Currently sort_order is only meaningful when handling
1684 * coalitions.
1685 *
1686 * Return:
1687 * 0 on success
1688 * non-0 on failure
1689 */
1690 static int
1691 memorystatus_sort_bucket(unsigned int bucket_index, int sort_order)
1692 {
1693 int coal_sort_order;
1694
1695 /*
1696 * Verify the jetsam priority
1697 */
1698 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1699 return EINVAL;
1700 }
1701
1702 #if DEVELOPMENT || DEBUG
1703 if (sort_order == JETSAM_SORT_DEFAULT) {
1704 coal_sort_order = COALITION_SORT_DEFAULT;
1705 } else {
1706 coal_sort_order = sort_order; /* only used for testing scenarios */
1707 }
1708 #else
1709 /* Verify default */
1710 if (sort_order == JETSAM_SORT_DEFAULT) {
1711 coal_sort_order = COALITION_SORT_DEFAULT;
1712 } else {
1713 return EINVAL;
1714 }
1715 #endif
1716
1717 proc_list_lock();
1718
1719 if (memstat_bucket[bucket_index].count == 0) {
1720 proc_list_unlock();
1721 return 0;
1722 }
1723
1724 switch (bucket_index) {
1725 case JETSAM_PRIORITY_FOREGROUND:
1726 if (memorystatus_sort_by_largest_coalition_locked(bucket_index, coal_sort_order) == 0) {
1727 /*
1728 * Fall back to per process sorting when zero coalitions are found.
1729 */
1730 memorystatus_sort_by_largest_process_locked(bucket_index);
1731 }
1732 break;
1733 default:
1734 memorystatus_sort_by_largest_process_locked(bucket_index);
1735 break;
1736 }
1737 proc_list_unlock();
1738
1739 return 0;
1740 }
1741
1742 /*
1743 * Sort processes by size for a single jetsam bucket.
1744 */
1745
1746 static void
1747 memorystatus_sort_by_largest_process_locked(unsigned int bucket_index)
1748 {
1749 proc_t p = NULL, insert_after_proc = NULL, max_proc = NULL;
1750 proc_t next_p = NULL, prev_max_proc = NULL;
1751 uint32_t pages = 0, max_pages = 0;
1752 memstat_bucket_t *current_bucket;
1753
1754 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1755 return;
1756 }
1757
1758 current_bucket = &memstat_bucket[bucket_index];
1759
1760 p = TAILQ_FIRST(&current_bucket->list);
1761
1762 while (p) {
1763 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1764 max_pages = pages;
1765 max_proc = p;
1766 prev_max_proc = p;
1767
1768 while ((next_p = TAILQ_NEXT(p, p_memstat_list)) != NULL) {
1769 /* traversing list until we find next largest process */
1770 p = next_p;
1771 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
1772 if (pages > max_pages) {
1773 max_pages = pages;
1774 max_proc = p;
1775 }
1776 }
1777
1778 if (prev_max_proc != max_proc) {
1779 /* found a larger process, place it in the list */
1780 TAILQ_REMOVE(&current_bucket->list, max_proc, p_memstat_list);
1781 if (insert_after_proc == NULL) {
1782 TAILQ_INSERT_HEAD(&current_bucket->list, max_proc, p_memstat_list);
1783 } else {
1784 TAILQ_INSERT_AFTER(&current_bucket->list, insert_after_proc, max_proc, p_memstat_list);
1785 }
1786 prev_max_proc = max_proc;
1787 }
1788
1789 insert_after_proc = max_proc;
1790
1791 p = TAILQ_NEXT(max_proc, p_memstat_list);
1792 }
1793 }
1794
1795 static proc_t
1796 memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search)
1797 {
1798 memstat_bucket_t *current_bucket;
1799 proc_t next_p;
1800
1801 if ((*bucket_index) >= MEMSTAT_BUCKET_COUNT) {
1802 return NULL;
1803 }
1804
1805 current_bucket = &memstat_bucket[*bucket_index];
1806 next_p = TAILQ_FIRST(&current_bucket->list);
1807 if (!next_p && search) {
1808 while (!next_p && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1809 current_bucket = &memstat_bucket[*bucket_index];
1810 next_p = TAILQ_FIRST(&current_bucket->list);
1811 }
1812 }
1813
1814 return next_p;
1815 }
1816
1817 static proc_t
1818 memorystatus_get_next_proc_locked(unsigned int *bucket_index, proc_t p, boolean_t search)
1819 {
1820 memstat_bucket_t *current_bucket;
1821 proc_t next_p;
1822
1823 if (!p || ((*bucket_index) >= MEMSTAT_BUCKET_COUNT)) {
1824 return NULL;
1825 }
1826
1827 next_p = TAILQ_NEXT(p, p_memstat_list);
1828 while (!next_p && search && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1829 current_bucket = &memstat_bucket[*bucket_index];
1830 next_p = TAILQ_FIRST(&current_bucket->list);
1831 }
1832
1833 return next_p;
1834 }
1835
1836 /*
1837 * Structure to hold state for a jetsam thread.
1838 * Typically there should be a single jetsam thread
1839 * unless parallel jetsam is enabled.
1840 */
1841 struct jetsam_thread_state {
1842 boolean_t inited; /* if the thread is initialized */
1843 int memorystatus_wakeup; /* wake channel */
1844 int index; /* jetsam thread index */
1845 thread_t thread; /* jetsam thread pointer */
1846 } *jetsam_threads;
1847
1848 /* Maximum number of jetsam threads allowed */
1849 #define JETSAM_THREADS_LIMIT 3
1850
1851 /* Number of active jetsam threads */
1852 _Atomic int active_jetsam_threads = 1;
1853
1854 /* Number of maximum jetsam threads configured */
1855 int max_jetsam_threads = JETSAM_THREADS_LIMIT;
1856
1857 /*
1858 * Global switch for enabling fast jetsam. Fast jetsam is
1859 * hooked up via the system_override() system call. It has the
1860 * following effects:
1861 * - Raise the jetsam threshold ("clear-the-deck")
1862 * - Enabled parallel jetsam on eligible devices
1863 */
1864 int fast_jetsam_enabled = 0;
1865
1866 /* Routine to find the jetsam state structure for the current jetsam thread */
1867 static inline struct jetsam_thread_state *
1868 jetsam_current_thread(void)
1869 {
1870 for (int thr_id = 0; thr_id < max_jetsam_threads; thr_id++) {
1871 if (jetsam_threads[thr_id].thread == current_thread()) {
1872 return &(jetsam_threads[thr_id]);
1873 }
1874 }
1875 panic("jetsam_current_thread() is being called from a non-jetsam thread\n");
1876 /* Contol should not reach here */
1877 return NULL;
1878 }
1879
1880
1881 __private_extern__ void
1882 memorystatus_init(void)
1883 {
1884 kern_return_t result;
1885 int i;
1886
1887 #if CONFIG_FREEZE
1888 memorystatus_freeze_jetsam_band = JETSAM_PRIORITY_UI_SUPPORT;
1889 memorystatus_frozen_processes_max = FREEZE_PROCESSES_MAX;
1890 memorystatus_frozen_shared_mb_max = ((MAX_FROZEN_SHARED_MB_PERCENT * max_task_footprint_mb) / 100); /* 10% of the system wide task limit */
1891 memorystatus_freeze_shared_mb_per_process_max = (memorystatus_frozen_shared_mb_max / 4);
1892 memorystatus_freeze_pages_min = FREEZE_PAGES_MIN;
1893 memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
1894 memorystatus_max_frozen_demotions_daily = MAX_FROZEN_PROCESS_DEMOTIONS;
1895 memorystatus_thaw_count_demotion_threshold = MIN_THAW_DEMOTION_THRESHOLD;
1896 #endif
1897
1898 #if DEVELOPMENT || DEBUG
1899 disconnect_page_mappings_lck_grp_attr = lck_grp_attr_alloc_init();
1900 disconnect_page_mappings_lck_grp = lck_grp_alloc_init("disconnect_page_mappings", disconnect_page_mappings_lck_grp_attr);
1901
1902 lck_mtx_init(&disconnect_page_mappings_mutex, disconnect_page_mappings_lck_grp, NULL);
1903
1904 if (kill_on_no_paging_space == TRUE) {
1905 max_kill_priority = JETSAM_PRIORITY_MAX;
1906 }
1907 #endif
1908
1909
1910 /* Init buckets */
1911 for (i = 0; i < MEMSTAT_BUCKET_COUNT; i++) {
1912 TAILQ_INIT(&memstat_bucket[i].list);
1913 memstat_bucket[i].count = 0;
1914 }
1915 memorystatus_idle_demotion_call = thread_call_allocate((thread_call_func_t)memorystatus_perform_idle_demotion, NULL);
1916
1917 #if CONFIG_JETSAM
1918 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
1919 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
1920
1921 /* Apply overrides */
1922 PE_get_default("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage));
1923 if (delta_percentage == 0) {
1924 delta_percentage = 5;
1925 }
1926 assert(delta_percentage < 100);
1927 PE_get_default("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage));
1928 assert(critical_threshold_percentage < 100);
1929 PE_get_default("kern.jetsam_idle_offset", &idle_offset_percentage, sizeof(idle_offset_percentage));
1930 assert(idle_offset_percentage < 100);
1931 PE_get_default("kern.jetsam_pressure_threshold", &pressure_threshold_percentage, sizeof(pressure_threshold_percentage));
1932 assert(pressure_threshold_percentage < 100);
1933 PE_get_default("kern.jetsam_freeze_threshold", &freeze_threshold_percentage, sizeof(freeze_threshold_percentage));
1934 assert(freeze_threshold_percentage < 100);
1935
1936 if (!PE_parse_boot_argn("jetsam_aging_policy", &jetsam_aging_policy,
1937 sizeof(jetsam_aging_policy))) {
1938 if (!PE_get_default("kern.jetsam_aging_policy", &jetsam_aging_policy,
1939 sizeof(jetsam_aging_policy))) {
1940 jetsam_aging_policy = kJetsamAgingPolicyLegacy;
1941 }
1942 }
1943
1944 if (jetsam_aging_policy > kJetsamAgingPolicyMax) {
1945 jetsam_aging_policy = kJetsamAgingPolicyLegacy;
1946 }
1947
1948 switch (jetsam_aging_policy) {
1949 case kJetsamAgingPolicyNone:
1950 system_procs_aging_band = JETSAM_PRIORITY_IDLE;
1951 applications_aging_band = JETSAM_PRIORITY_IDLE;
1952 break;
1953
1954 case kJetsamAgingPolicyLegacy:
1955 /*
1956 * Legacy behavior where some daemons get a 10s protection once
1957 * AND only before the first clean->dirty->clean transition before
1958 * going into IDLE band.
1959 */
1960 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1961 applications_aging_band = JETSAM_PRIORITY_IDLE;
1962 break;
1963
1964 case kJetsamAgingPolicySysProcsReclaimedFirst:
1965 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1966 applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1967 break;
1968
1969 case kJetsamAgingPolicyAppsReclaimedFirst:
1970 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1971 applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1972 break;
1973
1974 default:
1975 break;
1976 }
1977
1978 /*
1979 * The aging bands cannot overlap with the JETSAM_PRIORITY_ELEVATED_INACTIVE
1980 * band and must be below it in priority. This is so that we don't have to make
1981 * our 'aging' code worry about a mix of processes, some of which need to age
1982 * and some others that need to stay elevated in the jetsam bands.
1983 */
1984 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > system_procs_aging_band);
1985 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > applications_aging_band);
1986
1987 /* Take snapshots for idle-exit kills by default? First check the boot-arg... */
1988 if (!PE_parse_boot_argn("jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot))) {
1989 /* ...no boot-arg, so check the device tree */
1990 PE_get_default("kern.jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot));
1991 }
1992
1993 memorystatus_delta = delta_percentage * atop_64(max_mem) / 100;
1994 memorystatus_available_pages_critical_idle_offset = idle_offset_percentage * atop_64(max_mem) / 100;
1995 memorystatus_available_pages_critical_base = (critical_threshold_percentage / delta_percentage) * memorystatus_delta;
1996 memorystatus_policy_more_free_offset_pages = (policy_more_free_offset_percentage / delta_percentage) * memorystatus_delta;
1997
1998 /* Jetsam Loop Detection */
1999 if (max_mem <= (512 * 1024 * 1024)) {
2000 /* 512 MB devices */
2001 memorystatus_jld_eval_period_msecs = 8000; /* 8000 msecs == 8 second window */
2002 } else {
2003 /* 1GB and larger devices */
2004 memorystatus_jld_eval_period_msecs = 6000; /* 6000 msecs == 6 second window */
2005 }
2006
2007 memorystatus_jld_enabled = TRUE;
2008
2009 /* No contention at this point */
2010 memorystatus_update_levels_locked(FALSE);
2011
2012 #endif /* CONFIG_JETSAM */
2013
2014 memorystatus_jetsam_snapshot_max = maxproc;
2015
2016 memorystatus_jetsam_snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
2017 (sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_max);
2018
2019 memorystatus_jetsam_snapshot =
2020 (memorystatus_jetsam_snapshot_t*)kalloc(memorystatus_jetsam_snapshot_size);
2021 if (!memorystatus_jetsam_snapshot) {
2022 panic("Could not allocate memorystatus_jetsam_snapshot");
2023 }
2024
2025 memorystatus_jetsam_snapshot_copy =
2026 (memorystatus_jetsam_snapshot_t*)kalloc(memorystatus_jetsam_snapshot_size);
2027 if (!memorystatus_jetsam_snapshot_copy) {
2028 panic("Could not allocate memorystatus_jetsam_snapshot_copy");
2029 }
2030
2031 nanoseconds_to_absolutetime((uint64_t)JETSAM_SNAPSHOT_TIMEOUT_SECS * NSEC_PER_SEC, &memorystatus_jetsam_snapshot_timeout);
2032
2033 memset(&memorystatus_at_boot_snapshot, 0, sizeof(memorystatus_jetsam_snapshot_t));
2034
2035 #if CONFIG_FREEZE
2036 memorystatus_freeze_threshold = (freeze_threshold_percentage / delta_percentage) * memorystatus_delta;
2037 #endif
2038
2039 /* Check the boot-arg to see if fast jetsam is allowed */
2040 if (!PE_parse_boot_argn("fast_jetsam_enabled", &fast_jetsam_enabled, sizeof(fast_jetsam_enabled))) {
2041 fast_jetsam_enabled = 0;
2042 }
2043
2044 /* Check the boot-arg to configure the maximum number of jetsam threads */
2045 if (!PE_parse_boot_argn("max_jetsam_threads", &max_jetsam_threads, sizeof(max_jetsam_threads))) {
2046 max_jetsam_threads = JETSAM_THREADS_LIMIT;
2047 }
2048
2049 /* Restrict the maximum number of jetsam threads to JETSAM_THREADS_LIMIT */
2050 if (max_jetsam_threads > JETSAM_THREADS_LIMIT) {
2051 max_jetsam_threads = JETSAM_THREADS_LIMIT;
2052 }
2053
2054 /* For low CPU systems disable fast jetsam mechanism */
2055 if (vm_pageout_state.vm_restricted_to_single_processor == TRUE) {
2056 max_jetsam_threads = 1;
2057 fast_jetsam_enabled = 0;
2058 }
2059
2060 /* Initialize the jetsam_threads state array */
2061 jetsam_threads = kalloc(sizeof(struct jetsam_thread_state) * max_jetsam_threads);
2062
2063 /* Initialize all the jetsam threads */
2064 for (i = 0; i < max_jetsam_threads; i++) {
2065 result = kernel_thread_start_priority(memorystatus_thread, NULL, 95 /* MAXPRI_KERNEL */, &jetsam_threads[i].thread);
2066 if (result == KERN_SUCCESS) {
2067 jetsam_threads[i].inited = FALSE;
2068 jetsam_threads[i].index = i;
2069 thread_deallocate(jetsam_threads[i].thread);
2070 } else {
2071 panic("Could not create memorystatus_thread %d", i);
2072 }
2073 }
2074 }
2075
2076 /* Centralised for the purposes of allowing panic-on-jetsam */
2077 extern void
2078 vm_run_compactor(void);
2079
2080 /*
2081 * The jetsam no frills kill call
2082 * Return: 0 on success
2083 * error code on failure (EINVAL...)
2084 */
2085 static int
2086 jetsam_do_kill(proc_t p, int jetsam_flags, os_reason_t jetsam_reason)
2087 {
2088 int error = 0;
2089 error = exit_with_reason(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags, jetsam_reason);
2090 return error;
2091 }
2092
2093 /*
2094 * Wrapper for processes exiting with memorystatus details
2095 */
2096 static boolean_t
2097 memorystatus_do_kill(proc_t p, uint32_t cause, os_reason_t jetsam_reason)
2098 {
2099 int error = 0;
2100 __unused pid_t victim_pid = p->p_pid;
2101
2102 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_START,
2103 victim_pid, cause, vm_page_free_count, 0, 0);
2104
2105 DTRACE_MEMORYSTATUS3(memorystatus_do_kill, proc_t, p, os_reason_t, jetsam_reason, uint32_t, cause);
2106 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
2107 if (memorystatus_jetsam_panic_debug & (1 << cause)) {
2108 panic("memorystatus_do_kill(): jetsam debug panic (cause: %d)", cause);
2109 }
2110 #else
2111 #pragma unused(cause)
2112 #endif
2113
2114 if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
2115 printf("memorystatus: killing process %d [%s] in high band %s (%d) - memorystatus_available_pages: %llu\n", p->p_pid,
2116 (*p->p_name ? p->p_name : "unknown"),
2117 memorystatus_priority_band_name(p->p_memstat_effectivepriority), p->p_memstat_effectivepriority,
2118 (uint64_t)memorystatus_available_pages);
2119 }
2120
2121 /*
2122 * The jetsam_reason (os_reason_t) has enough information about the kill cause.
2123 * We don't really need jetsam_flags anymore, so it's okay that not all possible kill causes have been mapped.
2124 */
2125 int jetsam_flags = P_LTERM_JETSAM;
2126 switch (cause) {
2127 case kMemorystatusKilledHiwat: jetsam_flags |= P_JETSAM_HIWAT; break;
2128 case kMemorystatusKilledVnodes: jetsam_flags |= P_JETSAM_VNODE; break;
2129 case kMemorystatusKilledVMPageShortage: jetsam_flags |= P_JETSAM_VMPAGESHORTAGE; break;
2130 case kMemorystatusKilledVMCompressorThrashing:
2131 case kMemorystatusKilledVMCompressorSpaceShortage: jetsam_flags |= P_JETSAM_VMTHRASHING; break;
2132 case kMemorystatusKilledFCThrashing: jetsam_flags |= P_JETSAM_FCTHRASHING; break;
2133 case kMemorystatusKilledPerProcessLimit: jetsam_flags |= P_JETSAM_PID; break;
2134 case kMemorystatusKilledIdleExit: jetsam_flags |= P_JETSAM_IDLEEXIT; break;
2135 }
2136 error = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
2137
2138 KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_END,
2139 victim_pid, cause, vm_page_free_count, error, 0);
2140
2141 vm_run_compactor();
2142
2143 return error == 0;
2144 }
2145
2146 /*
2147 * Node manipulation
2148 */
2149
2150 static void
2151 memorystatus_check_levels_locked(void)
2152 {
2153 #if CONFIG_JETSAM
2154 /* Update levels */
2155 memorystatus_update_levels_locked(TRUE);
2156 #else /* CONFIG_JETSAM */
2157 /*
2158 * Nothing to do here currently since we update
2159 * memorystatus_available_pages in vm_pressure_response.
2160 */
2161 #endif /* CONFIG_JETSAM */
2162 }
2163
2164 /*
2165 * Pin a process to a particular jetsam band when it is in the background i.e. not doing active work.
2166 * For an application: that means no longer in the FG band
2167 * For a daemon: that means no longer in its 'requested' jetsam priority band
2168 */
2169
2170 int
2171 memorystatus_update_inactive_jetsam_priority_band(pid_t pid, uint32_t op_flags, int jetsam_prio, boolean_t effective_now)
2172 {
2173 int error = 0;
2174 boolean_t enable = FALSE;
2175 proc_t p = NULL;
2176
2177 if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE) {
2178 enable = TRUE;
2179 } else if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE) {
2180 enable = FALSE;
2181 } else {
2182 return EINVAL;
2183 }
2184
2185 p = proc_find(pid);
2186 if (p != NULL) {
2187 if ((enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) ||
2188 (!enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == 0))) {
2189 /*
2190 * No change in state.
2191 */
2192 } else {
2193 proc_list_lock();
2194
2195 if (enable) {
2196 p->p_memstat_state |= P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
2197 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2198
2199 if (effective_now) {
2200 if (p->p_memstat_effectivepriority < jetsam_prio) {
2201 if (memorystatus_highwater_enabled) {
2202 /*
2203 * Process is about to transition from
2204 * inactive --> active
2205 * assign active state
2206 */
2207 boolean_t is_fatal;
2208 boolean_t use_active = TRUE;
2209 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2210 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
2211 }
2212 memorystatus_update_priority_locked(p, jetsam_prio, FALSE, FALSE);
2213 }
2214 } else {
2215 if (isProcessInAgingBands(p)) {
2216 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2217 }
2218 }
2219 } else {
2220 p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
2221 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2222
2223 if (effective_now) {
2224 if (p->p_memstat_effectivepriority == jetsam_prio) {
2225 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2226 }
2227 } else {
2228 if (isProcessInAgingBands(p)) {
2229 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2230 }
2231 }
2232 }
2233
2234 proc_list_unlock();
2235 }
2236 proc_rele(p);
2237 error = 0;
2238 } else {
2239 error = ESRCH;
2240 }
2241
2242 return error;
2243 }
2244
2245 static void
2246 memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2)
2247 {
2248 proc_t p;
2249 uint64_t current_time = 0, idle_delay_time = 0;
2250 int demote_prio_band = 0;
2251 memstat_bucket_t *demotion_bucket;
2252
2253 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion()\n");
2254
2255 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_START, 0, 0, 0, 0, 0);
2256
2257 current_time = mach_absolute_time();
2258
2259 proc_list_lock();
2260
2261 demote_prio_band = JETSAM_PRIORITY_IDLE + 1;
2262
2263 for (; demote_prio_band < JETSAM_PRIORITY_MAX; demote_prio_band++) {
2264 if (demote_prio_band != system_procs_aging_band && demote_prio_band != applications_aging_band) {
2265 continue;
2266 }
2267
2268 demotion_bucket = &memstat_bucket[demote_prio_band];
2269 p = TAILQ_FIRST(&demotion_bucket->list);
2270
2271 while (p) {
2272 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion() found %d\n", p->p_pid);
2273
2274 assert(p->p_memstat_idledeadline);
2275
2276 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
2277
2278 if (current_time >= p->p_memstat_idledeadline) {
2279 if ((isSysProc(p) &&
2280 ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) != P_DIRTY_IDLE_EXIT_ENABLED)) || /* system proc marked dirty*/
2281 task_has_assertions((struct task *)(p->task))) { /* has outstanding assertions which might indicate outstanding work too */
2282 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_delay_time : memorystatus_apps_idle_delay_time;
2283
2284 p->p_memstat_idledeadline += idle_delay_time;
2285 p = TAILQ_NEXT(p, p_memstat_list);
2286 } else {
2287 proc_t next_proc = NULL;
2288
2289 next_proc = TAILQ_NEXT(p, p_memstat_list);
2290 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2291
2292 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false, true);
2293
2294 p = next_proc;
2295 continue;
2296 }
2297 } else {
2298 // No further candidates
2299 break;
2300 }
2301 }
2302 }
2303
2304 memorystatus_reschedule_idle_demotion_locked();
2305
2306 proc_list_unlock();
2307
2308 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_END, 0, 0, 0, 0, 0);
2309 }
2310
2311 static void
2312 memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state)
2313 {
2314 boolean_t present_in_sysprocs_aging_bucket = FALSE;
2315 boolean_t present_in_apps_aging_bucket = FALSE;
2316 uint64_t idle_delay_time = 0;
2317
2318 if (jetsam_aging_policy == kJetsamAgingPolicyNone) {
2319 return;
2320 }
2321
2322 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2323 /*
2324 * This process isn't going to be making the trip to the lower bands.
2325 */
2326 return;
2327 }
2328
2329 if (isProcessInAgingBands(p)) {
2330 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
2331 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) != P_DIRTY_AGING_IN_PROGRESS);
2332 }
2333
2334 if (isSysProc(p) && system_procs_aging_band) {
2335 present_in_sysprocs_aging_bucket = TRUE;
2336 } else if (isApp(p) && applications_aging_band) {
2337 present_in_apps_aging_bucket = TRUE;
2338 }
2339 }
2340
2341 assert(!present_in_sysprocs_aging_bucket);
2342 assert(!present_in_apps_aging_bucket);
2343
2344 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",
2345 p->p_pid, p->p_memstat_dirty, set_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
2346
2347 if (isSysProc(p)) {
2348 assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
2349 }
2350
2351 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_delay_time : memorystatus_apps_idle_delay_time;
2352
2353 if (set_state) {
2354 p->p_memstat_dirty |= P_DIRTY_AGING_IN_PROGRESS;
2355 p->p_memstat_idledeadline = mach_absolute_time() + idle_delay_time;
2356 }
2357
2358 assert(p->p_memstat_idledeadline);
2359
2360 if (isSysProc(p) && present_in_sysprocs_aging_bucket == FALSE) {
2361 memorystatus_scheduled_idle_demotions_sysprocs++;
2362 } else if (isApp(p) && present_in_apps_aging_bucket == FALSE) {
2363 memorystatus_scheduled_idle_demotions_apps++;
2364 }
2365 }
2366
2367 static void
2368 memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clear_state)
2369 {
2370 boolean_t present_in_sysprocs_aging_bucket = FALSE;
2371 boolean_t present_in_apps_aging_bucket = FALSE;
2372
2373 if (!system_procs_aging_band && !applications_aging_band) {
2374 return;
2375 }
2376
2377 if ((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == 0) {
2378 return;
2379 }
2380
2381 if (isProcessInAgingBands(p)) {
2382 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
2383 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == P_DIRTY_AGING_IN_PROGRESS);
2384 }
2385
2386 if (isSysProc(p) && system_procs_aging_band) {
2387 assert(p->p_memstat_effectivepriority == system_procs_aging_band);
2388 assert(p->p_memstat_idledeadline);
2389 present_in_sysprocs_aging_bucket = TRUE;
2390 } else if (isApp(p) && applications_aging_band) {
2391 assert(p->p_memstat_effectivepriority == applications_aging_band);
2392 assert(p->p_memstat_idledeadline);
2393 present_in_apps_aging_bucket = TRUE;
2394 }
2395 }
2396
2397 MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for pid %d (clear_state %d, demotions %d).\n",
2398 p->p_pid, clear_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
2399
2400
2401 if (clear_state) {
2402 p->p_memstat_idledeadline = 0;
2403 p->p_memstat_dirty &= ~P_DIRTY_AGING_IN_PROGRESS;
2404 }
2405
2406 if (isSysProc(p) && present_in_sysprocs_aging_bucket == TRUE) {
2407 memorystatus_scheduled_idle_demotions_sysprocs--;
2408 assert(memorystatus_scheduled_idle_demotions_sysprocs >= 0);
2409 } else if (isApp(p) && present_in_apps_aging_bucket == TRUE) {
2410 memorystatus_scheduled_idle_demotions_apps--;
2411 assert(memorystatus_scheduled_idle_demotions_apps >= 0);
2412 }
2413
2414 assert((memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps) >= 0);
2415 }
2416
2417 static void
2418 memorystatus_reschedule_idle_demotion_locked(void)
2419 {
2420 if (0 == (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps)) {
2421 if (memstat_idle_demotion_deadline) {
2422 /* Transitioned 1->0, so cancel next call */
2423 thread_call_cancel(memorystatus_idle_demotion_call);
2424 memstat_idle_demotion_deadline = 0;
2425 }
2426 } else {
2427 memstat_bucket_t *demotion_bucket;
2428 proc_t p = NULL, p1 = NULL, p2 = NULL;
2429
2430 if (system_procs_aging_band) {
2431 demotion_bucket = &memstat_bucket[system_procs_aging_band];
2432 p1 = TAILQ_FIRST(&demotion_bucket->list);
2433
2434 p = p1;
2435 }
2436
2437 if (applications_aging_band) {
2438 demotion_bucket = &memstat_bucket[applications_aging_band];
2439 p2 = TAILQ_FIRST(&demotion_bucket->list);
2440
2441 if (p1 && p2) {
2442 p = (p1->p_memstat_idledeadline > p2->p_memstat_idledeadline) ? p2 : p1;
2443 } else {
2444 p = (p1 == NULL) ? p2 : p1;
2445 }
2446 }
2447
2448 assert(p);
2449
2450 if (p != NULL) {
2451 assert(p && p->p_memstat_idledeadline);
2452 if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline) {
2453 thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
2454 memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
2455 }
2456 }
2457 }
2458 }
2459
2460 /*
2461 * List manipulation
2462 */
2463
2464 int
2465 memorystatus_add(proc_t p, boolean_t locked)
2466 {
2467 memstat_bucket_t *bucket;
2468
2469 MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding pid %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
2470
2471 if (!locked) {
2472 proc_list_lock();
2473 }
2474
2475 DTRACE_MEMORYSTATUS2(memorystatus_add, proc_t, p, int32_t, p->p_memstat_effectivepriority);
2476
2477 /* Processes marked internal do not have priority tracked */
2478 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2479 goto exit;
2480 }
2481
2482 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2483
2484 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
2485 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs - 1);
2486 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
2487 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps - 1);
2488 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2489 /*
2490 * Entering the idle band.
2491 * Record idle start time.
2492 */
2493 p->p_memstat_idle_start = mach_absolute_time();
2494 }
2495
2496 TAILQ_INSERT_TAIL(&bucket->list, p, p_memstat_list);
2497 bucket->count++;
2498
2499 memorystatus_list_count++;
2500
2501 memorystatus_check_levels_locked();
2502
2503 exit:
2504 if (!locked) {
2505 proc_list_unlock();
2506 }
2507
2508 return 0;
2509 }
2510
2511 /*
2512 * Description:
2513 * Moves a process from one jetsam bucket to another.
2514 * which changes the LRU position of the process.
2515 *
2516 * Monitors transition between buckets and if necessary
2517 * will update cached memory limits accordingly.
2518 *
2519 * skip_demotion_check:
2520 * - if the 'jetsam aging policy' is NOT 'legacy':
2521 * When this flag is TRUE, it means we are going
2522 * to age the ripe processes out of the aging bands and into the
2523 * IDLE band and apply their inactive memory limits.
2524 *
2525 * - if the 'jetsam aging policy' is 'legacy':
2526 * When this flag is TRUE, it might mean the above aging mechanism
2527 * OR
2528 * It might be that we have a process that has used up its 'idle deferral'
2529 * stay that is given to it once per lifetime. And in this case, the process
2530 * won't be going through any aging codepaths. But we still need to apply
2531 * the right inactive limits and so we explicitly set this to TRUE if the
2532 * new priority for the process is the IDLE band.
2533 */
2534 void
2535 memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check)
2536 {
2537 memstat_bucket_t *old_bucket, *new_bucket;
2538
2539 assert(priority < MEMSTAT_BUCKET_COUNT);
2540
2541 /* Ensure that exit isn't underway, leaving the proc retained but removed from its bucket */
2542 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2543 return;
2544 }
2545
2546 MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting %s(%d) to priority %d, inserting at %s\n",
2547 (*p->p_name ? p->p_name : "unknown"), p->p_pid, priority, head_insert ? "head" : "tail");
2548
2549 DTRACE_MEMORYSTATUS3(memorystatus_update_priority, proc_t, p, int32_t, p->p_memstat_effectivepriority, int, priority);
2550
2551 #if DEVELOPMENT || DEBUG
2552 if (priority == JETSAM_PRIORITY_IDLE && /* if the process is on its way into the IDLE band */
2553 skip_demotion_check == FALSE && /* and it isn't via the path that will set the INACTIVE memlimits */
2554 (p->p_memstat_dirty & P_DIRTY_TRACK) && /* and it has 'DIRTY' tracking enabled */
2555 ((p->p_memstat_memlimit != p->p_memstat_memlimit_inactive) || /* and we notice that the current limit isn't the right value (inactive) */
2556 ((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) */
2557 panic("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 */
2558 }
2559 #endif /* DEVELOPMENT || DEBUG */
2560
2561 old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2562
2563 if (skip_demotion_check == FALSE) {
2564 if (isSysProc(p)) {
2565 /*
2566 * For system processes, the memorystatus_dirty_* routines take care of adding/removing
2567 * the processes from the aging bands and balancing the demotion counts.
2568 * We can, however, override that if the process has an 'elevated inactive jetsam band' attribute.
2569 */
2570
2571 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2572 /*
2573 * 2 types of processes can use the non-standard elevated inactive band:
2574 * - Frozen processes that always land in memorystatus_freeze_jetsam_band
2575 * OR
2576 * - processes that specifically opt-in to the elevated inactive support e.g. docked processes.
2577 */
2578 #if CONFIG_FREEZE
2579 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
2580 if (priority <= memorystatus_freeze_jetsam_band) {
2581 priority = memorystatus_freeze_jetsam_band;
2582 }
2583 } else
2584 #endif /* CONFIG_FREEZE */
2585 {
2586 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE) {
2587 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2588 }
2589 }
2590 assert(!(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2591 }
2592 } else if (isApp(p)) {
2593 /*
2594 * Check to see if the application is being lowered in jetsam priority. If so, and:
2595 * - it has an 'elevated inactive jetsam band' attribute, then put it in the appropriate band.
2596 * - it is a normal application, then let it age in the aging band if that policy is in effect.
2597 */
2598
2599 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
2600 #if CONFIG_FREEZE
2601 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
2602 if (priority <= memorystatus_freeze_jetsam_band) {
2603 priority = memorystatus_freeze_jetsam_band;
2604 }
2605 } else
2606 #endif /* CONFIG_FREEZE */
2607 {
2608 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE) {
2609 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2610 }
2611 }
2612 } else {
2613 if (applications_aging_band) {
2614 if (p->p_memstat_effectivepriority == applications_aging_band) {
2615 assert(old_bucket->count == (memorystatus_scheduled_idle_demotions_apps + 1));
2616 }
2617
2618 if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && (priority <= applications_aging_band)) {
2619 assert(!(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2620 priority = applications_aging_band;
2621 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2622 }
2623 }
2624 }
2625 }
2626 }
2627
2628 if ((system_procs_aging_band && (priority == system_procs_aging_band)) || (applications_aging_band && (priority == applications_aging_band))) {
2629 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
2630 }
2631
2632 TAILQ_REMOVE(&old_bucket->list, p, p_memstat_list);
2633 old_bucket->count--;
2634
2635 new_bucket = &memstat_bucket[priority];
2636 if (head_insert) {
2637 TAILQ_INSERT_HEAD(&new_bucket->list, p, p_memstat_list);
2638 } else {
2639 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
2640 }
2641 new_bucket->count++;
2642
2643 if (memorystatus_highwater_enabled) {
2644 boolean_t is_fatal;
2645 boolean_t use_active;
2646
2647 /*
2648 * If cached limit data is updated, then the limits
2649 * will be enforced by writing to the ledgers.
2650 */
2651 boolean_t ledger_update_needed = TRUE;
2652
2653 /*
2654 * Here, we must update the cached memory limit if the task
2655 * is transitioning between:
2656 * active <--> inactive
2657 * FG <--> BG
2658 * but:
2659 * dirty <--> clean is ignored
2660 *
2661 * We bypass non-idle processes that have opted into dirty tracking because
2662 * a move between buckets does not imply a transition between the
2663 * dirty <--> clean state.
2664 */
2665
2666 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
2667 if (skip_demotion_check == TRUE && priority == JETSAM_PRIORITY_IDLE) {
2668 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2669 use_active = FALSE;
2670 } else {
2671 ledger_update_needed = FALSE;
2672 }
2673 } else if ((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) {
2674 /*
2675 * inactive --> active
2676 * BG --> FG
2677 * assign active state
2678 */
2679 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2680 use_active = TRUE;
2681 } else if ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
2682 /*
2683 * active --> inactive
2684 * FG --> BG
2685 * assign inactive state
2686 */
2687 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2688 use_active = FALSE;
2689 } else {
2690 /*
2691 * The transition between jetsam priority buckets apparently did
2692 * not affect active/inactive state.
2693 * This is not unusual... especially during startup when
2694 * processes are getting established in their respective bands.
2695 */
2696 ledger_update_needed = FALSE;
2697 }
2698
2699 /*
2700 * Enforce the new limits by writing to the ledger
2701 */
2702 if (ledger_update_needed) {
2703 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
2704
2705 MEMORYSTATUS_DEBUG(3, "memorystatus_update_priority_locked: new limit on pid %d (%dMB %s) priority old --> new (%d --> %d) dirty?=0x%x %s\n",
2706 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2707 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, priority, p->p_memstat_dirty,
2708 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2709 }
2710 }
2711
2712 /*
2713 * Record idle start or idle delta.
2714 */
2715 if (p->p_memstat_effectivepriority == priority) {
2716 /*
2717 * This process is not transitioning between
2718 * jetsam priority buckets. Do nothing.
2719 */
2720 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2721 uint64_t now;
2722 /*
2723 * Transitioning out of the idle priority bucket.
2724 * Record idle delta.
2725 */
2726 assert(p->p_memstat_idle_start != 0);
2727 now = mach_absolute_time();
2728 if (now > p->p_memstat_idle_start) {
2729 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2730 }
2731
2732 /*
2733 * About to become active and so memory footprint could change.
2734 * So mark it eligible for freeze-considerations next time around.
2735 */
2736 if (p->p_memstat_state & P_MEMSTAT_FREEZE_IGNORE) {
2737 p->p_memstat_state &= ~P_MEMSTAT_FREEZE_IGNORE;
2738 }
2739 } else if (priority == JETSAM_PRIORITY_IDLE) {
2740 /*
2741 * Transitioning into the idle priority bucket.
2742 * Record idle start.
2743 */
2744 p->p_memstat_idle_start = mach_absolute_time();
2745 }
2746
2747 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CHANGE_PRIORITY), p->p_pid, priority, p->p_memstat_effectivepriority, 0, 0);
2748
2749 p->p_memstat_effectivepriority = priority;
2750
2751 #if CONFIG_SECLUDED_MEMORY
2752 if (secluded_for_apps &&
2753 task_could_use_secluded_mem(p->task)) {
2754 task_set_can_use_secluded_mem(
2755 p->task,
2756 (priority >= JETSAM_PRIORITY_FOREGROUND));
2757 }
2758 #endif /* CONFIG_SECLUDED_MEMORY */
2759
2760 memorystatus_check_levels_locked();
2761 }
2762
2763 /*
2764 *
2765 * Description: Update the jetsam priority and memory limit attributes for a given process.
2766 *
2767 * Parameters:
2768 * p init this process's jetsam information.
2769 * priority The jetsam priority band
2770 * user_data user specific data, unused by the kernel
2771 * effective guards against race if process's update already occurred
2772 * update_memlimit When true we know this is the init step via the posix_spawn path.
2773 *
2774 * memlimit_active Value in megabytes; The monitored footprint level while the
2775 * process is active. Exceeding it may result in termination
2776 * based on it's associated fatal flag.
2777 *
2778 * memlimit_active_is_fatal When a process is active and exceeds its memory footprint,
2779 * this describes whether or not it should be immediately fatal.
2780 *
2781 * memlimit_inactive Value in megabytes; The monitored footprint level while the
2782 * process is inactive. Exceeding it may result in termination
2783 * based on it's associated fatal flag.
2784 *
2785 * memlimit_inactive_is_fatal When a process is inactive and exceeds its memory footprint,
2786 * this describes whether or not it should be immediatly fatal.
2787 *
2788 * Returns: 0 Success
2789 * non-0 Failure
2790 */
2791
2792 int
2793 memorystatus_update(proc_t p, int priority, uint64_t user_data, boolean_t effective, boolean_t update_memlimit,
2794 int32_t memlimit_active, boolean_t memlimit_active_is_fatal,
2795 int32_t memlimit_inactive, boolean_t memlimit_inactive_is_fatal)
2796 {
2797 int ret;
2798 boolean_t head_insert = false;
2799
2800 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);
2801
2802 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_START, p->p_pid, priority, user_data, effective, 0);
2803
2804 if (priority == -1) {
2805 /* Use as shorthand for default priority */
2806 priority = JETSAM_PRIORITY_DEFAULT;
2807 } else if ((priority == system_procs_aging_band) || (priority == applications_aging_band)) {
2808 /* Both the aging bands are reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
2809 priority = JETSAM_PRIORITY_IDLE;
2810 } else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
2811 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
2812 priority = JETSAM_PRIORITY_IDLE;
2813 head_insert = TRUE;
2814 } else if ((priority < 0) || (priority >= MEMSTAT_BUCKET_COUNT)) {
2815 /* Sanity check */
2816 ret = EINVAL;
2817 goto out;
2818 }
2819
2820 proc_list_lock();
2821
2822 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2823
2824 if (effective && (p->p_memstat_state & P_MEMSTAT_PRIORITYUPDATED)) {
2825 ret = EALREADY;
2826 proc_list_unlock();
2827 MEMORYSTATUS_DEBUG(1, "memorystatus_update: effective change specified for pid %d, but change already occurred.\n", p->p_pid);
2828 goto out;
2829 }
2830
2831 if ((p->p_memstat_state & P_MEMSTAT_TERMINATED) || ((p->p_listflag & P_LIST_EXITED) != 0)) {
2832 /*
2833 * This could happen when a process calling posix_spawn() is exiting on the jetsam thread.
2834 */
2835 ret = EBUSY;
2836 proc_list_unlock();
2837 goto out;
2838 }
2839
2840 p->p_memstat_state |= P_MEMSTAT_PRIORITYUPDATED;
2841 p->p_memstat_userdata = user_data;
2842 p->p_memstat_requestedpriority = priority;
2843
2844 if (update_memlimit) {
2845 boolean_t is_fatal;
2846 boolean_t use_active;
2847
2848 /*
2849 * Posix_spawn'd processes come through this path to instantiate ledger limits.
2850 * Forked processes do not come through this path, so no ledger limits exist.
2851 * (That's why forked processes can consume unlimited memory.)
2852 */
2853
2854 MEMORYSTATUS_DEBUG(3, "memorystatus_update(enter): pid %d, priority %d, dirty=0x%x, Active(%dMB %s), Inactive(%dMB, %s)\n",
2855 p->p_pid, priority, p->p_memstat_dirty,
2856 memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
2857 memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
2858
2859 if (memlimit_active <= 0) {
2860 /*
2861 * This process will have a system_wide task limit when active.
2862 * System_wide task limit is always fatal.
2863 * It's quite common to see non-fatal flag passed in here.
2864 * It's not an error, we just ignore it.
2865 */
2866
2867 /*
2868 * For backward compatibility with some unexplained launchd behavior,
2869 * we allow a zero sized limit. But we still enforce system_wide limit
2870 * when written to the ledgers.
2871 */
2872
2873 if (memlimit_active < 0) {
2874 memlimit_active = -1; /* enforces system_wide task limit */
2875 }
2876 memlimit_active_is_fatal = TRUE;
2877 }
2878
2879 if (memlimit_inactive <= 0) {
2880 /*
2881 * This process will have a system_wide task limit when inactive.
2882 * System_wide task limit is always fatal.
2883 */
2884
2885 memlimit_inactive = -1;
2886 memlimit_inactive_is_fatal = TRUE;
2887 }
2888
2889 /*
2890 * Initialize the active limit variants for this process.
2891 */
2892 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
2893
2894 /*
2895 * Initialize the inactive limit variants for this process.
2896 */
2897 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
2898
2899 /*
2900 * Initialize the cached limits for target process.
2901 * When the target process is dirty tracked, it's typically
2902 * in a clean state. Non dirty tracked processes are
2903 * typically active (Foreground or above).
2904 * But just in case, we don't make assumptions...
2905 */
2906
2907 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
2908 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
2909 use_active = TRUE;
2910 } else {
2911 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
2912 use_active = FALSE;
2913 }
2914
2915 /*
2916 * Enforce the cached limit by writing to the ledger.
2917 */
2918 if (memorystatus_highwater_enabled) {
2919 /* apply now */
2920 task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, use_active, is_fatal);
2921
2922 MEMORYSTATUS_DEBUG(3, "memorystatus_update: init: limit on pid %d (%dMB %s) targeting priority(%d) dirty?=0x%x %s\n",
2923 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2924 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), priority, p->p_memstat_dirty,
2925 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2926 }
2927 }
2928
2929 /*
2930 * We can't add to the aging bands buckets here.
2931 * But, we could be removing it from those buckets.
2932 * Check and take appropriate steps if so.
2933 */
2934
2935 if (isProcessInAgingBands(p)) {
2936 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2937 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2938 } else {
2939 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy && priority == JETSAM_PRIORITY_IDLE) {
2940 /*
2941 * Daemons with 'inactive' limits will go through the dirty tracking codepath.
2942 * This path deals with apps that may have 'inactive' limits e.g. WebContent processes.
2943 * If this is the legacy aging policy we explicitly need to apply those limits. If it
2944 * is any other aging policy, then we don't need to worry because all processes
2945 * will go through the aging bands and then the demotion thread will take care to
2946 * move them into the IDLE band and apply the required limits.
2947 */
2948 memorystatus_update_priority_locked(p, priority, head_insert, TRUE);
2949 }
2950 }
2951
2952 memorystatus_update_priority_locked(p, priority, head_insert, FALSE);
2953
2954 proc_list_unlock();
2955 ret = 0;
2956
2957 out:
2958 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_END, ret, 0, 0, 0, 0);
2959
2960 return ret;
2961 }
2962
2963 int
2964 memorystatus_remove(proc_t p, boolean_t locked)
2965 {
2966 int ret;
2967 memstat_bucket_t *bucket;
2968 boolean_t reschedule = FALSE;
2969
2970 MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing pid %d\n", p->p_pid);
2971
2972 if (!locked) {
2973 proc_list_lock();
2974 }
2975
2976 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2977
2978 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2979
2980 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
2981 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs);
2982 reschedule = TRUE;
2983 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
2984 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps);
2985 reschedule = TRUE;
2986 }
2987
2988 /*
2989 * Record idle delta
2990 */
2991
2992 if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2993 uint64_t now = mach_absolute_time();
2994 if (now > p->p_memstat_idle_start) {
2995 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2996 }
2997 }
2998
2999 TAILQ_REMOVE(&bucket->list, p, p_memstat_list);
3000 bucket->count--;
3001
3002 memorystatus_list_count--;
3003
3004 /* If awaiting demotion to the idle band, clean up */
3005 if (reschedule) {
3006 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3007 memorystatus_reschedule_idle_demotion_locked();
3008 }
3009
3010 memorystatus_check_levels_locked();
3011
3012 #if CONFIG_FREEZE
3013 if (p->p_memstat_state & (P_MEMSTAT_FROZEN)) {
3014 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
3015 p->p_memstat_state &= ~P_MEMSTAT_REFREEZE_ELIGIBLE;
3016 memorystatus_refreeze_eligible_count--;
3017 }
3018
3019 memorystatus_frozen_count--;
3020 memorystatus_frozen_shared_mb -= p->p_memstat_freeze_sharedanon_pages;
3021 p->p_memstat_freeze_sharedanon_pages = 0;
3022 }
3023
3024 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
3025 memorystatus_suspended_count--;
3026 }
3027 #endif
3028
3029 if (!locked) {
3030 proc_list_unlock();
3031 }
3032
3033 if (p) {
3034 ret = 0;
3035 } else {
3036 ret = ESRCH;
3037 }
3038
3039 return ret;
3040 }
3041
3042 /*
3043 * Validate dirty tracking flags with process state.
3044 *
3045 * Return:
3046 * 0 on success
3047 * non-0 on failure
3048 *
3049 * The proc_list_lock is held by the caller.
3050 */
3051
3052 static int
3053 memorystatus_validate_track_flags(struct proc *target_p, uint32_t pcontrol)
3054 {
3055 /* See that the process isn't marked for termination */
3056 if (target_p->p_memstat_dirty & P_DIRTY_TERMINATED) {
3057 return EBUSY;
3058 }
3059
3060 /* Idle exit requires that process be tracked */
3061 if ((pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) &&
3062 !(pcontrol & PROC_DIRTY_TRACK)) {
3063 return EINVAL;
3064 }
3065
3066 /* 'Launch in progress' tracking requires that process have enabled dirty tracking too. */
3067 if ((pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) &&
3068 !(pcontrol & PROC_DIRTY_TRACK)) {
3069 return EINVAL;
3070 }
3071
3072 /* Only one type of DEFER behavior is allowed.*/
3073 if ((pcontrol & PROC_DIRTY_DEFER) &&
3074 (pcontrol & PROC_DIRTY_DEFER_ALWAYS)) {
3075 return EINVAL;
3076 }
3077
3078 /* Deferral is only relevant if idle exit is specified */
3079 if (((pcontrol & PROC_DIRTY_DEFER) ||
3080 (pcontrol & PROC_DIRTY_DEFER_ALWAYS)) &&
3081 !(pcontrol & PROC_DIRTY_ALLOWS_IDLE_EXIT)) {
3082 return EINVAL;
3083 }
3084
3085 return 0;
3086 }
3087
3088 static void
3089 memorystatus_update_idle_priority_locked(proc_t p)
3090 {
3091 int32_t priority;
3092
3093 MEMORYSTATUS_DEBUG(1, "memorystatus_update_idle_priority_locked(): pid %d dirty 0x%X\n", p->p_pid, p->p_memstat_dirty);
3094
3095 assert(isSysProc(p));
3096
3097 if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED | P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
3098 priority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
3099 } else {
3100 priority = p->p_memstat_requestedpriority;
3101 }
3102
3103 if (priority != p->p_memstat_effectivepriority) {
3104 if ((jetsam_aging_policy == kJetsamAgingPolicyLegacy) &&
3105 (priority == JETSAM_PRIORITY_IDLE)) {
3106 /*
3107 * This process is on its way into the IDLE band. The system is
3108 * using 'legacy' jetsam aging policy. That means, this process
3109 * has already used up its idle-deferral aging time that is given
3110 * once per its lifetime. So we need to set the INACTIVE limits
3111 * explicitly because it won't be going through the demotion paths
3112 * that take care to apply the limits appropriately.
3113 */
3114
3115 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
3116 /*
3117 * This process has the 'elevated inactive jetsam band' attribute.
3118 * So, there will be no trip to IDLE after all.
3119 * Instead, we pin the process in the elevated band,
3120 * where its ACTIVE limits will apply.
3121 */
3122
3123 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
3124 }
3125
3126 memorystatus_update_priority_locked(p, priority, false, true);
3127 } else {
3128 memorystatus_update_priority_locked(p, priority, false, false);
3129 }
3130 }
3131 }
3132
3133 /*
3134 * Processes can opt to have their state tracked by the kernel, indicating when they are busy (dirty) or idle
3135 * (clean). They may also indicate that they support termination when idle, with the result that they are promoted
3136 * to their desired, higher, jetsam priority when dirty (and are therefore killed later), and demoted to the low
3137 * priority idle band when clean (and killed earlier, protecting higher priority procesess).
3138 *
3139 * If the deferral flag is set, then newly tracked processes will be protected for an initial period (as determined by
3140 * memorystatus_sysprocs_idle_delay_time); if they go clean during this time, then they will be moved to a deferred-idle band
3141 * with a slightly higher priority, guarding against immediate termination under memory pressure and being unable to
3142 * make forward progress. Finally, when the guard expires, they will be moved to the standard, lowest-priority, idle
3143 * band. The deferral can be cleared early by clearing the appropriate flag.
3144 *
3145 * The deferral timer is active only for the duration that the process is marked as guarded and clean; if the process
3146 * is marked dirty, the timer will be cancelled. Upon being subsequently marked clean, the deferment will either be
3147 * re-enabled or the guard state cleared, depending on whether the guard deadline has passed.
3148 */
3149
3150 int
3151 memorystatus_dirty_track(proc_t p, uint32_t pcontrol)
3152 {
3153 unsigned int old_dirty;
3154 boolean_t reschedule = FALSE;
3155 boolean_t already_deferred = FALSE;
3156 boolean_t defer_now = FALSE;
3157 int ret = 0;
3158
3159 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_TRACK),
3160 p->p_pid, p->p_memstat_dirty, pcontrol, 0, 0);
3161
3162 proc_list_lock();
3163
3164 if ((p->p_listflag & P_LIST_EXITED) != 0) {
3165 /*
3166 * Process is on its way out.
3167 */
3168 ret = EBUSY;
3169 goto exit;
3170 }
3171
3172 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
3173 ret = EPERM;
3174 goto exit;
3175 }
3176
3177 if ((ret = memorystatus_validate_track_flags(p, pcontrol)) != 0) {
3178 /* error */
3179 goto exit;
3180 }
3181
3182 old_dirty = p->p_memstat_dirty;
3183
3184 /* These bits are cumulative, as per <rdar://problem/11159924> */
3185 if (pcontrol & PROC_DIRTY_TRACK) {
3186 p->p_memstat_dirty |= P_DIRTY_TRACK;
3187 }
3188
3189 if (pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) {
3190 p->p_memstat_dirty |= P_DIRTY_ALLOW_IDLE_EXIT;
3191 }
3192
3193 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
3194 p->p_memstat_dirty |= P_DIRTY_LAUNCH_IN_PROGRESS;
3195 }
3196
3197 if (old_dirty & P_DIRTY_AGING_IN_PROGRESS) {
3198 already_deferred = TRUE;
3199 }
3200
3201
3202 /* This can be set and cleared exactly once. */
3203 if (pcontrol & (PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) {
3204 if ((pcontrol & (PROC_DIRTY_DEFER)) &&
3205 !(old_dirty & P_DIRTY_DEFER)) {
3206 p->p_memstat_dirty |= P_DIRTY_DEFER;
3207 }
3208
3209 if ((pcontrol & (PROC_DIRTY_DEFER_ALWAYS)) &&
3210 !(old_dirty & P_DIRTY_DEFER_ALWAYS)) {
3211 p->p_memstat_dirty |= P_DIRTY_DEFER_ALWAYS;
3212 }
3213
3214 defer_now = TRUE;
3215 }
3216
3217 MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for pid %d\n",
3218 ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) ? "Y" : "N",
3219 defer_now ? "Y" : "N",
3220 p->p_memstat_dirty & P_DIRTY ? "Y" : "N",
3221 p->p_pid);
3222
3223 /* Kick off or invalidate the idle exit deferment if there's a state transition. */
3224 if (!(p->p_memstat_dirty & P_DIRTY_IS_DIRTY)) {
3225 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3226 if (defer_now && !already_deferred) {
3227 /*
3228 * Request to defer a clean process that's idle-exit enabled
3229 * and not already in the jetsam deferred band. Most likely a
3230 * new launch.
3231 */
3232 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3233 reschedule = TRUE;
3234 } else if (!defer_now) {
3235 /*
3236 * The process isn't asking for the 'aging' facility.
3237 * Could be that it is:
3238 */
3239
3240 if (already_deferred) {
3241 /*
3242 * already in the aging bands. Traditionally,
3243 * some processes have tried to use this to
3244 * opt out of the 'aging' facility.
3245 */
3246
3247 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3248 } else {
3249 /*
3250 * agnostic to the 'aging' facility. In that case,
3251 * we'll go ahead and opt it in because this is likely
3252 * a new launch (clean process, dirty tracking enabled)
3253 */
3254
3255 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3256 }
3257
3258 reschedule = TRUE;
3259 }
3260 }
3261 } else {
3262 /*
3263 * We are trying to operate on a dirty process. Dirty processes have to
3264 * be removed from the deferred band. The question is do we reset the
3265 * deferred state or not?
3266 *
3267 * This could be a legal request like:
3268 * - this process had opted into the 'aging' band
3269 * - but it's now dirty and requests to opt out.
3270 * In this case, we remove the process from the band and reset its
3271 * state too. It'll opt back in properly when needed.
3272 *
3273 * OR, this request could be a user-space bug. E.g.:
3274 * - this process had opted into the 'aging' band when clean
3275 * - and, then issues another request to again put it into the band except
3276 * this time the process is dirty.
3277 * The process going dirty, as a transition in memorystatus_dirty_set(), will pull the process out of
3278 * the deferred band with its state intact. So our request below is no-op.
3279 * But we do it here anyways for coverage.
3280 *
3281 * memorystatus_update_idle_priority_locked()
3282 * single-mindedly treats a dirty process as "cannot be in the aging band".
3283 */
3284
3285 if (!defer_now && already_deferred) {
3286 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3287 reschedule = TRUE;
3288 } else {
3289 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
3290
3291 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
3292 reschedule = TRUE;
3293 }
3294 }
3295
3296 memorystatus_update_idle_priority_locked(p);
3297
3298 if (reschedule) {
3299 memorystatus_reschedule_idle_demotion_locked();
3300 }
3301
3302 ret = 0;
3303
3304 exit:
3305 proc_list_unlock();
3306
3307 return ret;
3308 }
3309
3310 int
3311 memorystatus_dirty_set(proc_t p, boolean_t self, uint32_t pcontrol)
3312 {
3313 int ret;
3314 boolean_t kill = false;
3315 boolean_t reschedule = FALSE;
3316 boolean_t was_dirty = FALSE;
3317 boolean_t now_dirty = FALSE;
3318
3319 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_set(): %d %d 0x%x 0x%x\n", self, p->p_pid, pcontrol, p->p_memstat_dirty);
3320 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_SET), p->p_pid, self, pcontrol, 0, 0);
3321
3322 proc_list_lock();
3323
3324 if ((p->p_listflag & P_LIST_EXITED) != 0) {
3325 /*
3326 * Process is on its way out.
3327 */
3328 ret = EBUSY;
3329 goto exit;
3330 }
3331
3332 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
3333 ret = EPERM;
3334 goto exit;
3335 }
3336
3337 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3338 was_dirty = TRUE;
3339 }
3340
3341 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
3342 /* Dirty tracking not enabled */
3343 ret = EINVAL;
3344 } else if (pcontrol && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
3345 /*
3346 * Process is set to be terminated and we're attempting to mark it dirty.
3347 * Set for termination and marking as clean is OK - see <rdar://problem/10594349>.
3348 */
3349 ret = EBUSY;
3350 } else {
3351 int flag = (self == TRUE) ? P_DIRTY : P_DIRTY_SHUTDOWN;
3352 if (pcontrol && !(p->p_memstat_dirty & flag)) {
3353 /* Mark the process as having been dirtied at some point */
3354 p->p_memstat_dirty |= (flag | P_DIRTY_MARKED);
3355 memorystatus_dirty_count++;
3356 ret = 0;
3357 } else if ((pcontrol == 0) && (p->p_memstat_dirty & flag)) {
3358 if ((flag == P_DIRTY_SHUTDOWN) && (!(p->p_memstat_dirty & P_DIRTY))) {
3359 /* Clearing the dirty shutdown flag, and the process is otherwise clean - kill */
3360 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3361 kill = true;
3362 } else if ((flag == P_DIRTY) && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
3363 /* Kill previously terminated processes if set clean */
3364 kill = true;
3365 }
3366 p->p_memstat_dirty &= ~flag;
3367 memorystatus_dirty_count--;
3368 ret = 0;
3369 } else {
3370 /* Already set */
3371 ret = EALREADY;
3372 }
3373 }
3374
3375 if (ret != 0) {
3376 goto exit;
3377 }
3378
3379 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3380 now_dirty = TRUE;
3381 }
3382
3383 if ((was_dirty == TRUE && now_dirty == FALSE) ||
3384 (was_dirty == FALSE && now_dirty == TRUE)) {
3385 /* Manage idle exit deferral, if applied */
3386 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3387 /*
3388 * Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band OR it might be heading back
3389 * there once it's clean again. For the legacy case, this only applies if it has some protection window left.
3390 * P_DIRTY_DEFER: one-time protection window given at launch
3391 * P_DIRTY_DEFER_ALWAYS: protection window given for every dirty->clean transition. Like non-legacy mode.
3392 *
3393 * Non-Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band. It will always stop over
3394 * in that band on it's way to IDLE.
3395 */
3396
3397 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3398 /*
3399 * New dirty process i.e. "was_dirty == FALSE && now_dirty == TRUE"
3400 *
3401 * The process will move from its aging band to its higher requested
3402 * jetsam band.
3403 */
3404 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
3405
3406 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
3407 reschedule = TRUE;
3408 } else {
3409 /*
3410 * Process is back from "dirty" to "clean".
3411 */
3412
3413 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
3414 if (((p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) == FALSE) &&
3415 (mach_absolute_time() >= p->p_memstat_idledeadline)) {
3416 /*
3417 * The process' hasn't enrolled in the "always defer after dirty"
3418 * mode and its deadline has expired. It currently
3419 * does not reside in any of the aging buckets.
3420 *
3421 * It's on its way to the JETSAM_PRIORITY_IDLE
3422 * bucket via memorystatus_update_idle_priority_locked()
3423 * below.
3424 *
3425 * So all we need to do is reset all the state on the
3426 * process that's related to the aging bucket i.e.
3427 * the AGING_IN_PROGRESS flag and the timer deadline.
3428 */
3429
3430 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3431 reschedule = TRUE;
3432 } else {
3433 /*
3434 * Process enrolled in "always stop in deferral band after dirty" OR
3435 * it still has some protection window left and so
3436 * we just re-arm the timer without modifying any
3437 * state on the process iff it still wants into that band.
3438 */
3439
3440 if (p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) {
3441 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3442 reschedule = TRUE;
3443 } else if (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) {
3444 memorystatus_schedule_idle_demotion_locked(p, FALSE);
3445 reschedule = TRUE;
3446 }
3447 }
3448 } else {
3449 memorystatus_schedule_idle_demotion_locked(p, TRUE);
3450 reschedule = TRUE;
3451 }
3452 }
3453 }
3454
3455 memorystatus_update_idle_priority_locked(p);
3456
3457 if (memorystatus_highwater_enabled) {
3458 boolean_t ledger_update_needed = TRUE;
3459 boolean_t use_active;
3460 boolean_t is_fatal;
3461 /*
3462 * We are in this path because this process transitioned between
3463 * dirty <--> clean state. Update the cached memory limits.
3464 */
3465
3466 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
3467 /*
3468 * process is pinned in elevated band
3469 * or
3470 * process is dirty
3471 */
3472 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
3473 use_active = TRUE;
3474 ledger_update_needed = TRUE;
3475 } else {
3476 /*
3477 * process is clean...but if it has opted into pressured-exit
3478 * we don't apply the INACTIVE limit till the process has aged
3479 * out and is entering the IDLE band.
3480 * See memorystatus_update_priority_locked() for that.
3481 */
3482
3483 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3484 ledger_update_needed = FALSE;
3485 } else {
3486 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
3487 use_active = FALSE;
3488 ledger_update_needed = TRUE;
3489 }
3490 }
3491
3492 /*
3493 * Enforce the new limits by writing to the ledger.
3494 *
3495 * This is a hot path and holding the proc_list_lock while writing to the ledgers,
3496 * (where the task lock is taken) is bad. So, we temporarily drop the proc_list_lock.
3497 * We aren't traversing the jetsam bucket list here, so we should be safe.
3498 * See rdar://21394491.
3499 */
3500
3501 if (ledger_update_needed && proc_ref_locked(p) == p) {
3502 int ledger_limit;
3503 if (p->p_memstat_memlimit > 0) {
3504 ledger_limit = p->p_memstat_memlimit;
3505 } else {
3506 ledger_limit = -1;
3507 }
3508 proc_list_unlock();
3509 task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, use_active, is_fatal);
3510 proc_list_lock();
3511 proc_rele_locked(p);
3512
3513 MEMORYSTATUS_DEBUG(3, "memorystatus_dirty_set: new limit on pid %d (%dMB %s) priority(%d) dirty?=0x%x %s\n",
3514 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
3515 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
3516 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
3517 }
3518 }
3519
3520 /* If the deferral state changed, reschedule the demotion timer */
3521 if (reschedule) {
3522 memorystatus_reschedule_idle_demotion_locked();
3523 }
3524 }
3525
3526 if (kill) {
3527 if (proc_ref_locked(p) == p) {
3528 proc_list_unlock();
3529 psignal(p, SIGKILL);
3530 proc_list_lock();
3531 proc_rele_locked(p);
3532 }
3533 }
3534
3535 exit:
3536 proc_list_unlock();
3537
3538 return ret;
3539 }
3540
3541 int
3542 memorystatus_dirty_clear(proc_t p, uint32_t pcontrol)
3543 {
3544 int ret = 0;
3545
3546 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_clear(): %d 0x%x 0x%x\n", p->p_pid, pcontrol, p->p_memstat_dirty);
3547
3548 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_CLEAR), p->p_pid, pcontrol, 0, 0, 0);
3549
3550 proc_list_lock();
3551
3552 if ((p->p_listflag & P_LIST_EXITED) != 0) {
3553 /*
3554 * Process is on its way out.
3555 */
3556 ret = EBUSY;
3557 goto exit;
3558 }
3559
3560 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
3561 ret = EPERM;
3562 goto exit;
3563 }
3564
3565 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
3566 /* Dirty tracking not enabled */
3567 ret = EINVAL;
3568 goto exit;
3569 }
3570
3571 if (!pcontrol || (pcontrol & (PROC_DIRTY_LAUNCH_IN_PROGRESS | PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) == 0) {
3572 ret = EINVAL;
3573 goto exit;
3574 }
3575
3576 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
3577 p->p_memstat_dirty &= ~P_DIRTY_LAUNCH_IN_PROGRESS;
3578 }
3579
3580 /* This can be set and cleared exactly once. */
3581 if (pcontrol & (PROC_DIRTY_DEFER | PROC_DIRTY_DEFER_ALWAYS)) {
3582 if (p->p_memstat_dirty & P_DIRTY_DEFER) {
3583 p->p_memstat_dirty &= ~(P_DIRTY_DEFER);
3584 }
3585
3586 if (p->p_memstat_dirty & P_DIRTY_DEFER_ALWAYS) {
3587 p->p_memstat_dirty &= ~(P_DIRTY_DEFER_ALWAYS);
3588 }
3589
3590 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3591 memorystatus_update_idle_priority_locked(p);
3592 memorystatus_reschedule_idle_demotion_locked();
3593 }
3594
3595 ret = 0;
3596 exit:
3597 proc_list_unlock();
3598
3599 return ret;
3600 }
3601
3602 int
3603 memorystatus_dirty_get(proc_t p)
3604 {
3605 int ret = 0;
3606
3607 proc_list_lock();
3608
3609 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3610 ret |= PROC_DIRTY_TRACKED;
3611 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3612 ret |= PROC_DIRTY_ALLOWS_IDLE_EXIT;
3613 }
3614 if (p->p_memstat_dirty & P_DIRTY) {
3615 ret |= PROC_DIRTY_IS_DIRTY;
3616 }
3617 if (p->p_memstat_dirty & P_DIRTY_LAUNCH_IN_PROGRESS) {
3618 ret |= PROC_DIRTY_LAUNCH_IS_IN_PROGRESS;
3619 }
3620 }
3621
3622 proc_list_unlock();
3623
3624 return ret;
3625 }
3626
3627 int
3628 memorystatus_on_terminate(proc_t p)
3629 {
3630 int sig;
3631
3632 proc_list_lock();
3633
3634 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3635
3636 if ((p->p_memstat_dirty & (P_DIRTY_TRACK | P_DIRTY_IS_DIRTY)) == P_DIRTY_TRACK) {
3637 /* Clean; mark as terminated and issue SIGKILL */
3638 sig = SIGKILL;
3639 } else {
3640 /* Dirty, terminated, or state tracking is unsupported; issue SIGTERM to allow cleanup */
3641 sig = SIGTERM;
3642 }
3643
3644 proc_list_unlock();
3645
3646 return sig;
3647 }
3648
3649 void
3650 memorystatus_on_suspend(proc_t p)
3651 {
3652 #if CONFIG_FREEZE
3653 uint32_t pages;
3654 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
3655 #endif
3656 proc_list_lock();
3657 #if CONFIG_FREEZE
3658 memorystatus_suspended_count++;
3659 #endif
3660 p->p_memstat_state |= P_MEMSTAT_SUSPENDED;
3661 proc_list_unlock();
3662 }
3663
3664 void
3665 memorystatus_on_resume(proc_t p)
3666 {
3667 #if CONFIG_FREEZE
3668 boolean_t frozen;
3669 pid_t pid;
3670 #endif
3671
3672 proc_list_lock();
3673
3674 #if CONFIG_FREEZE
3675 frozen = (p->p_memstat_state & P_MEMSTAT_FROZEN);
3676 if (frozen) {
3677 /*
3678 * Now that we don't _thaw_ a process completely,
3679 * resuming it (and having some on-demand swapins)
3680 * shouldn't preclude it from being counted as frozen.
3681 *
3682 * memorystatus_frozen_count--;
3683 *
3684 * We preserve the P_MEMSTAT_FROZEN state since the process
3685 * could have state on disk AND so will deserve some protection
3686 * in the jetsam bands.
3687 */
3688 if ((p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) == 0) {
3689 p->p_memstat_state |= P_MEMSTAT_REFREEZE_ELIGIBLE;
3690 memorystatus_refreeze_eligible_count++;
3691 }
3692 p->p_memstat_thaw_count++;
3693
3694 memorystatus_thaw_count++;
3695 }
3696
3697 memorystatus_suspended_count--;
3698
3699 pid = p->p_pid;
3700 #endif
3701
3702 /*
3703 * P_MEMSTAT_FROZEN will remain unchanged. This used to be:
3704 * p->p_memstat_state &= ~(P_MEMSTAT_SUSPENDED | P_MEMSTAT_FROZEN);
3705 */
3706 p->p_memstat_state &= ~P_MEMSTAT_SUSPENDED;
3707
3708 proc_list_unlock();
3709
3710 #if CONFIG_FREEZE
3711 if (frozen) {
3712 memorystatus_freeze_entry_t data = { pid, FALSE, 0 };
3713 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
3714 }
3715 #endif
3716 }
3717
3718 void
3719 memorystatus_on_inactivity(proc_t p)
3720 {
3721 #pragma unused(p)
3722 #if CONFIG_FREEZE
3723 /* Wake the freeze thread */
3724 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
3725 #endif
3726 }
3727
3728 /*
3729 * The proc_list_lock is held by the caller.
3730 */
3731 static uint32_t
3732 memorystatus_build_state(proc_t p)
3733 {
3734 uint32_t snapshot_state = 0;
3735
3736 /* General */
3737 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
3738 snapshot_state |= kMemorystatusSuspended;
3739 }
3740 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
3741 snapshot_state |= kMemorystatusFrozen;
3742 }
3743 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
3744 snapshot_state |= kMemorystatusWasThawed;
3745 }
3746
3747 /* Tracking */
3748 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3749 snapshot_state |= kMemorystatusTracked;
3750 }
3751 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3752 snapshot_state |= kMemorystatusSupportsIdleExit;
3753 }
3754 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3755 snapshot_state |= kMemorystatusDirty;
3756 }
3757
3758 return snapshot_state;
3759 }
3760
3761 static boolean_t
3762 kill_idle_exit_proc(void)
3763 {
3764 proc_t p, victim_p = PROC_NULL;
3765 uint64_t current_time;
3766 boolean_t killed = FALSE;
3767 unsigned int i = 0;
3768 os_reason_t jetsam_reason = OS_REASON_NULL;
3769
3770 /* Pick next idle exit victim. */
3771 current_time = mach_absolute_time();
3772
3773 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_IDLE_EXIT);
3774 if (jetsam_reason == OS_REASON_NULL) {
3775 printf("kill_idle_exit_proc: failed to allocate jetsam reason\n");
3776 }
3777
3778 proc_list_lock();
3779
3780 p = memorystatus_get_first_proc_locked(&i, FALSE);
3781 while (p) {
3782 /* No need to look beyond the idle band */
3783 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
3784 break;
3785 }
3786
3787 if ((p->p_memstat_dirty & (P_DIRTY_ALLOW_IDLE_EXIT | P_DIRTY_IS_DIRTY | P_DIRTY_TERMINATED)) == (P_DIRTY_ALLOW_IDLE_EXIT)) {
3788 if (current_time >= p->p_memstat_idledeadline) {
3789 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3790 victim_p = proc_ref_locked(p);
3791 break;
3792 }
3793 }
3794
3795 p = memorystatus_get_next_proc_locked(&i, p, FALSE);
3796 }
3797
3798 proc_list_unlock();
3799
3800 if (victim_p) {
3801 printf("memorystatus: killing_idle_process pid %d [%s]\n", victim_p->p_pid, (*victim_p->p_name ? victim_p->p_name : "unknown"));
3802 killed = memorystatus_do_kill(victim_p, kMemorystatusKilledIdleExit, jetsam_reason);
3803 proc_rele(victim_p);
3804 } else {
3805 os_reason_free(jetsam_reason);
3806 }
3807
3808 return killed;
3809 }
3810
3811 static void
3812 memorystatus_thread_wake(void)
3813 {
3814 int thr_id = 0;
3815 int active_thr = atomic_load(&active_jetsam_threads);
3816
3817 /* Wakeup all the jetsam threads */
3818 for (thr_id = 0; thr_id < active_thr; thr_id++) {
3819 thread_wakeup((event_t)&jetsam_threads[thr_id].memorystatus_wakeup);
3820 }
3821 }
3822
3823 #if CONFIG_JETSAM
3824
3825 static void
3826 memorystatus_thread_pool_max()
3827 {
3828 /* Increase the jetsam thread pool to max_jetsam_threads */
3829 int max_threads = max_jetsam_threads;
3830 printf("Expanding memorystatus pool to %d!\n", max_threads);
3831 atomic_store(&active_jetsam_threads, max_threads);
3832 }
3833
3834 static void
3835 memorystatus_thread_pool_default()
3836 {
3837 /* Restore the jetsam thread pool to a single thread */
3838 printf("Reverting memorystatus pool back to 1\n");
3839 atomic_store(&active_jetsam_threads, 1);
3840 }
3841
3842 #endif /* CONFIG_JETSAM */
3843
3844 extern void vm_pressure_response(void);
3845
3846 static int
3847 memorystatus_thread_block(uint32_t interval_ms, thread_continue_t continuation)
3848 {
3849 struct jetsam_thread_state *jetsam_thread = jetsam_current_thread();
3850
3851 if (interval_ms) {
3852 assert_wait_timeout(&jetsam_thread->memorystatus_wakeup, THREAD_UNINT, interval_ms, NSEC_PER_MSEC);
3853 } else {
3854 assert_wait(&jetsam_thread->memorystatus_wakeup, THREAD_UNINT);
3855 }
3856
3857 return thread_block(continuation);
3858 }
3859
3860 static boolean_t
3861 memorystatus_avail_pages_below_pressure(void)
3862 {
3863 #if CONFIG_EMBEDDED
3864 /*
3865 * Instead of CONFIG_EMBEDDED for these *avail_pages* routines, we should
3866 * key off of the system having dynamic swap support. With full swap support,
3867 * the system shouldn't really need to worry about various page thresholds.
3868 */
3869 return memorystatus_available_pages <= memorystatus_available_pages_pressure;
3870 #else /* CONFIG_EMBEDDED */
3871 return FALSE;
3872 #endif /* CONFIG_EMBEDDED */
3873 }
3874
3875 static boolean_t
3876 memorystatus_avail_pages_below_critical(void)
3877 {
3878 #if CONFIG_EMBEDDED
3879 return memorystatus_available_pages <= memorystatus_available_pages_critical;
3880 #else /* CONFIG_EMBEDDED */
3881 return FALSE;
3882 #endif /* CONFIG_EMBEDDED */
3883 }
3884
3885 static boolean_t
3886 memorystatus_post_snapshot(int32_t priority, uint32_t cause)
3887 {
3888 #if CONFIG_EMBEDDED
3889 #pragma unused(cause)
3890 /*
3891 * Don't generate logs for steady-state idle-exit kills,
3892 * unless it is overridden for debug or by the device
3893 * tree.
3894 */
3895
3896 return (priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot;
3897
3898 #else /* CONFIG_EMBEDDED */
3899 /*
3900 * Don't generate logs for steady-state idle-exit kills,
3901 * unless
3902 * - it is overridden for debug or by the device
3903 * tree.
3904 * OR
3905 * - the kill causes are important i.e. not kMemorystatusKilledIdleExit
3906 */
3907
3908 boolean_t snapshot_eligible_kill_cause = (is_reason_thrashing(cause) || is_reason_zone_map_exhaustion(cause));
3909 return (priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot || snapshot_eligible_kill_cause;
3910 #endif /* CONFIG_EMBEDDED */
3911 }
3912
3913 static boolean_t
3914 memorystatus_action_needed(void)
3915 {
3916 #if CONFIG_EMBEDDED
3917 return is_reason_thrashing(kill_under_pressure_cause) ||
3918 is_reason_zone_map_exhaustion(kill_under_pressure_cause) ||
3919 memorystatus_available_pages <= memorystatus_available_pages_pressure;
3920 #else /* CONFIG_EMBEDDED */
3921 return is_reason_thrashing(kill_under_pressure_cause) ||
3922 is_reason_zone_map_exhaustion(kill_under_pressure_cause);
3923 #endif /* CONFIG_EMBEDDED */
3924 }
3925
3926 #if CONFIG_FREEZE
3927 extern void vm_swap_consider_defragmenting(int);
3928
3929 /*
3930 * This routine will _jetsam_ all frozen processes
3931 * and reclaim the swap space immediately.
3932 *
3933 * So freeze has to be DISABLED when we call this routine.
3934 */
3935
3936 void
3937 memorystatus_disable_freeze(void)
3938 {
3939 memstat_bucket_t *bucket;
3940 int bucket_count = 0, retries = 0;
3941 boolean_t retval = FALSE, killed = FALSE;
3942 uint32_t errors = 0, errors_over_prev_iteration = 0;
3943 os_reason_t jetsam_reason = 0;
3944 unsigned int band = 0;
3945 proc_t p = PROC_NULL, next_p = PROC_NULL;
3946
3947 assert(memorystatus_freeze_enabled == FALSE);
3948
3949 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_DISK_SPACE_SHORTAGE);
3950 if (jetsam_reason == OS_REASON_NULL) {
3951 printf("memorystatus_disable_freeze: failed to allocate jetsam reason\n");
3952 }
3953
3954 /*
3955 * Let's relocate all frozen processes into band 8. Demoted frozen processes
3956 * are sitting in band 0 currently and it's possible to have a frozen process
3957 * in the FG band being actively used. We don't reset its frozen state when
3958 * it is resumed because it has state on disk.
3959 *
3960 * We choose to do this relocation rather than implement a new 'kill frozen'
3961 * process function for these reasons:
3962 * - duplication of code: too many kill functions exist and we need to rework them better.
3963 * - disk-space-shortage kills are rare
3964 * - not having the 'real' jetsam band at time of the this frozen kill won't preclude us
3965 * from answering any imp. questions re. jetsam policy/effectiveness.
3966 *
3967 * This is essentially what memorystatus_update_inactive_jetsam_priority_band() does while
3968 * avoiding the application of memory limits.
3969 */
3970
3971 again:
3972 proc_list_lock();
3973
3974 band = JETSAM_PRIORITY_IDLE;
3975 p = PROC_NULL;
3976 next_p = PROC_NULL;
3977
3978 next_p = memorystatus_get_first_proc_locked(&band, TRUE);
3979 while (next_p) {
3980 p = next_p;
3981 next_p = memorystatus_get_next_proc_locked(&band, p, TRUE);
3982
3983 if (p->p_memstat_effectivepriority > JETSAM_PRIORITY_FOREGROUND) {
3984 break;
3985 }
3986
3987 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) == FALSE) {
3988 continue;
3989 }
3990
3991 if (p->p_memstat_state & P_MEMSTAT_ERROR) {
3992 p->p_memstat_state &= ~P_MEMSTAT_ERROR;
3993 }
3994
3995 if (p->p_memstat_effectivepriority == memorystatus_freeze_jetsam_band) {
3996 continue;
3997 }
3998
3999 /*
4000 * We explicitly add this flag here so the process looks like a normal
4001 * frozen process i.e. P_MEMSTAT_FROZEN and P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND.
4002 * We don't bother with assigning the 'active' memory
4003 * limits at this point because we are going to be killing it soon below.
4004 */
4005 p->p_memstat_state |= P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
4006 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
4007
4008 memorystatus_update_priority_locked(p, memorystatus_freeze_jetsam_band, FALSE, TRUE);
4009 }
4010
4011 bucket = &memstat_bucket[memorystatus_freeze_jetsam_band];
4012 bucket_count = bucket->count;
4013 proc_list_unlock();
4014
4015 /*
4016 * Bucket count is already stale at this point. But, we don't expect
4017 * freezing to continue since we have already disabled the freeze functionality.
4018 * However, an existing freeze might be in progress. So we might miss that process
4019 * in the first go-around. We hope to catch it in the next.
4020 */
4021
4022 errors_over_prev_iteration = 0;
4023 while (bucket_count) {
4024 bucket_count--;
4025
4026 /*
4027 * memorystatus_kill_elevated_process() drops a reference,
4028 * so take another one so we can continue to use this exit reason
4029 * even after it returns.
4030 */
4031
4032 os_reason_ref(jetsam_reason);
4033 retval = memorystatus_kill_elevated_process(
4034 kMemorystatusKilledDiskSpaceShortage,
4035 jetsam_reason,
4036 memorystatus_freeze_jetsam_band,
4037 0, /* the iteration of aggressive jetsam..ignored here */
4038 &errors);
4039
4040 if (errors > 0) {
4041 printf("memorystatus_disable_freeze: memorystatus_kill_elevated_process returned %d error(s)\n", errors);
4042 errors_over_prev_iteration += errors;
4043 errors = 0;
4044 }
4045
4046 if (retval == 0) {
4047 /*
4048 * No frozen processes left to kill.
4049 */
4050 break;
4051 }
4052
4053 killed = TRUE;
4054 }
4055
4056 proc_list_lock();
4057
4058 if (memorystatus_frozen_count) {
4059 /*
4060 * A frozen process snuck in and so
4061 * go back around to kill it. That
4062 * process may have been resumed and
4063 * put into the FG band too. So we
4064 * have to do the relocation again.
4065 */
4066 assert(memorystatus_freeze_enabled == FALSE);
4067
4068 retries++;
4069 if (retries < 3) {
4070 proc_list_unlock();
4071 goto again;
4072 }
4073 #if DEVELOPMENT || DEBUG
4074 panic("memorystatus_disable_freeze: Failed to kill all frozen processes, memorystatus_frozen_count = %d, errors = %d",
4075 memorystatus_frozen_count, errors_over_prev_iteration);
4076 #endif /* DEVELOPMENT || DEBUG */
4077 }
4078 proc_list_unlock();
4079
4080 os_reason_free(jetsam_reason);
4081
4082 if (killed) {
4083 vm_swap_consider_defragmenting(VM_SWAP_FLAGS_FORCE_DEFRAG | VM_SWAP_FLAGS_FORCE_RECLAIM);
4084
4085 proc_list_lock();
4086 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4087 sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
4088 uint64_t timestamp_now = mach_absolute_time();
4089 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4090 memorystatus_jetsam_snapshot->js_gencount++;
4091 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4092 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4093 proc_list_unlock();
4094 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4095 if (!ret) {
4096 proc_list_lock();
4097 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4098 proc_list_unlock();
4099 }
4100 } else {
4101 proc_list_unlock();
4102 }
4103 }
4104
4105 return;
4106 }
4107 #endif /* CONFIG_FREEZE */
4108
4109 static boolean_t
4110 memorystatus_act_on_hiwat_processes(uint32_t *errors, uint32_t *hwm_kill, boolean_t *post_snapshot, __unused boolean_t *is_critical)
4111 {
4112 boolean_t purged = FALSE;
4113 boolean_t killed = memorystatus_kill_hiwat_proc(errors, &purged);
4114
4115 if (killed) {
4116 *hwm_kill = *hwm_kill + 1;
4117 *post_snapshot = TRUE;
4118 return TRUE;
4119 } else {
4120 if (purged == FALSE) {
4121 /* couldn't purge and couldn't kill */
4122 memorystatus_hwm_candidates = FALSE;
4123 }
4124 }
4125
4126 #if CONFIG_JETSAM
4127 /* No highwater processes to kill. Continue or stop for now? */
4128 if (!is_reason_thrashing(kill_under_pressure_cause) &&
4129 !is_reason_zone_map_exhaustion(kill_under_pressure_cause) &&
4130 (memorystatus_available_pages > memorystatus_available_pages_critical)) {
4131 /*
4132 * We are _not_ out of pressure but we are above the critical threshold and there's:
4133 * - no compressor thrashing
4134 * - enough zone memory
4135 * - no more HWM processes left.
4136 * For now, don't kill any other processes.
4137 */
4138
4139 if (*hwm_kill == 0) {
4140 memorystatus_thread_wasted_wakeup++;
4141 }
4142
4143 *is_critical = FALSE;
4144
4145 return TRUE;
4146 }
4147 #endif /* CONFIG_JETSAM */
4148
4149 return FALSE;
4150 }
4151
4152 static boolean_t
4153 memorystatus_act_aggressive(uint32_t cause, os_reason_t jetsam_reason, int *jld_idle_kills, boolean_t *corpse_list_purged, boolean_t *post_snapshot)
4154 {
4155 if (memorystatus_jld_enabled == TRUE) {
4156 boolean_t killed;
4157 uint32_t errors = 0;
4158
4159 /* Jetsam Loop Detection - locals */
4160 memstat_bucket_t *bucket;
4161 int jld_bucket_count = 0;
4162 struct timeval jld_now_tstamp = {0, 0};
4163 uint64_t jld_now_msecs = 0;
4164 int elevated_bucket_count = 0;
4165
4166 /* Jetsam Loop Detection - statics */
4167 static uint64_t jld_timestamp_msecs = 0;
4168 static int jld_idle_kill_candidates = 0; /* Number of available processes in band 0,1 at start */
4169 static int jld_eval_aggressive_count = 0; /* Bumps the max priority in aggressive loop */
4170 static int32_t jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
4171 /*
4172 * Jetsam Loop Detection: attempt to detect
4173 * rapid daemon relaunches in the lower bands.
4174 */
4175
4176 microuptime(&jld_now_tstamp);
4177
4178 /*
4179 * Ignore usecs in this calculation.
4180 * msecs granularity is close enough.
4181 */
4182 jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
4183
4184 proc_list_lock();
4185 switch (jetsam_aging_policy) {
4186 case kJetsamAgingPolicyLegacy:
4187 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
4188 jld_bucket_count = bucket->count;
4189 bucket = &memstat_bucket[JETSAM_PRIORITY_AGING_BAND1];
4190 jld_bucket_count += bucket->count;
4191 break;
4192 case kJetsamAgingPolicySysProcsReclaimedFirst:
4193 case kJetsamAgingPolicyAppsReclaimedFirst:
4194 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
4195 jld_bucket_count = bucket->count;
4196 bucket = &memstat_bucket[system_procs_aging_band];
4197 jld_bucket_count += bucket->count;
4198 bucket = &memstat_bucket[applications_aging_band];
4199 jld_bucket_count += bucket->count;
4200 break;
4201 case kJetsamAgingPolicyNone:
4202 default:
4203 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
4204 jld_bucket_count = bucket->count;
4205 break;
4206 }
4207
4208 bucket = &memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE];
4209 elevated_bucket_count = bucket->count;
4210
4211 proc_list_unlock();
4212
4213 /*
4214 * memorystatus_jld_eval_period_msecs is a tunable
4215 * memorystatus_jld_eval_aggressive_count is a tunable
4216 * memorystatus_jld_eval_aggressive_priority_band_max is a tunable
4217 */
4218 if ((jld_bucket_count == 0) ||
4219 (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
4220 /*
4221 * Refresh evaluation parameters
4222 */
4223 jld_timestamp_msecs = jld_now_msecs;
4224 jld_idle_kill_candidates = jld_bucket_count;
4225 *jld_idle_kills = 0;
4226 jld_eval_aggressive_count = 0;
4227 jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
4228 }
4229
4230 if (*jld_idle_kills > jld_idle_kill_candidates) {
4231 jld_eval_aggressive_count++;
4232
4233 #if DEVELOPMENT || DEBUG
4234 printf("memorystatus: aggressive%d: beginning of window: %lld ms, : timestamp now: %lld ms\n",
4235 jld_eval_aggressive_count,
4236 jld_timestamp_msecs,
4237 jld_now_msecs);
4238 printf("memorystatus: aggressive%d: idle candidates: %d, idle kills: %d\n",
4239 jld_eval_aggressive_count,
4240 jld_idle_kill_candidates,
4241 *jld_idle_kills);
4242 #endif /* DEVELOPMENT || DEBUG */
4243
4244 if ((jld_eval_aggressive_count == memorystatus_jld_eval_aggressive_count) &&
4245 (total_corpses_count() > 0) && (*corpse_list_purged == FALSE)) {
4246 /*
4247 * If we reach this aggressive cycle, corpses might be causing memory pressure.
4248 * So, in an effort to avoid jetsams in the FG band, we will attempt to purge
4249 * corpse memory prior to this final march through JETSAM_PRIORITY_UI_SUPPORT.
4250 */
4251 task_purge_all_corpses();
4252 *corpse_list_purged = TRUE;
4253 } else if (jld_eval_aggressive_count > memorystatus_jld_eval_aggressive_count) {
4254 /*
4255 * Bump up the jetsam priority limit (eg: the bucket index)
4256 * Enforce bucket index sanity.
4257 */
4258 if ((memorystatus_jld_eval_aggressive_priority_band_max < 0) ||
4259 (memorystatus_jld_eval_aggressive_priority_band_max >= MEMSTAT_BUCKET_COUNT)) {
4260 /*
4261 * Do nothing. Stick with the default level.
4262 */
4263 } else {
4264 jld_priority_band_max = memorystatus_jld_eval_aggressive_priority_band_max;
4265 }
4266 }
4267
4268 /* Visit elevated processes first */
4269 while (elevated_bucket_count) {
4270 elevated_bucket_count--;
4271
4272 /*
4273 * memorystatus_kill_elevated_process() drops a reference,
4274 * so take another one so we can continue to use this exit reason
4275 * even after it returns.
4276 */
4277
4278 os_reason_ref(jetsam_reason);
4279 killed = memorystatus_kill_elevated_process(
4280 cause,
4281 jetsam_reason,
4282 JETSAM_PRIORITY_ELEVATED_INACTIVE,
4283 jld_eval_aggressive_count,
4284 &errors);
4285
4286 if (killed) {
4287 *post_snapshot = TRUE;
4288 if (memorystatus_avail_pages_below_pressure()) {
4289 /*
4290 * Still under pressure.
4291 * Find another pinned processes.
4292 */
4293 continue;
4294 } else {
4295 return TRUE;
4296 }
4297 } else {
4298 /*
4299 * No pinned processes left to kill.
4300 * Abandon elevated band.
4301 */
4302 break;
4303 }
4304 }
4305
4306 /*
4307 * memorystatus_kill_top_process_aggressive() allocates its own
4308 * jetsam_reason so the kMemorystatusKilledProcThrashing cause
4309 * is consistent throughout the aggressive march.
4310 */
4311 killed = memorystatus_kill_top_process_aggressive(
4312 kMemorystatusKilledProcThrashing,
4313 jld_eval_aggressive_count,
4314 jld_priority_band_max,
4315 &errors);
4316
4317 if (killed) {
4318 /* Always generate logs after aggressive kill */
4319 *post_snapshot = TRUE;
4320 *jld_idle_kills = 0;
4321 return TRUE;
4322 }
4323 }
4324
4325 return FALSE;
4326 }
4327
4328 return FALSE;
4329 }
4330
4331
4332 static void
4333 memorystatus_thread(void *param __unused, wait_result_t wr __unused)
4334 {
4335 boolean_t post_snapshot = FALSE;
4336 uint32_t errors = 0;
4337 uint32_t hwm_kill = 0;
4338 boolean_t sort_flag = TRUE;
4339 boolean_t corpse_list_purged = FALSE;
4340 int jld_idle_kills = 0;
4341 struct jetsam_thread_state *jetsam_thread = jetsam_current_thread();
4342
4343 if (jetsam_thread->inited == FALSE) {
4344 /*
4345 * It's the first time the thread has run, so just mark the thread as privileged and block.
4346 * This avoids a spurious pass with unset variables, as set out in <rdar://problem/9609402>.
4347 */
4348
4349 char name[32];
4350 thread_wire(host_priv_self(), current_thread(), TRUE);
4351 snprintf(name, 32, "VM_memorystatus_%d", jetsam_thread->index + 1);
4352
4353 if (jetsam_thread->index == 0) {
4354 if (vm_pageout_state.vm_restricted_to_single_processor == TRUE) {
4355 thread_vm_bind_group_add();
4356 }
4357 }
4358 thread_set_thread_name(current_thread(), name);
4359 jetsam_thread->inited = TRUE;
4360 memorystatus_thread_block(0, memorystatus_thread);
4361 }
4362
4363 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_START,
4364 memorystatus_available_pages, memorystatus_jld_enabled, memorystatus_jld_eval_period_msecs, memorystatus_jld_eval_aggressive_count, 0);
4365
4366 /*
4367 * Jetsam aware version.
4368 *
4369 * The VM pressure notification thread is working it's way through clients in parallel.
4370 *
4371 * So, while the pressure notification thread is targeting processes in order of
4372 * increasing jetsam priority, we can hopefully reduce / stop it's work by killing
4373 * any processes that have exceeded their highwater mark.
4374 *
4375 * If we run out of HWM processes and our available pages drops below the critical threshold, then,
4376 * we target the least recently used process in order of increasing jetsam priority (exception: the FG band).
4377 */
4378 while (memorystatus_action_needed()) {
4379 boolean_t killed;
4380 int32_t priority;
4381 uint32_t cause;
4382 uint64_t jetsam_reason_code = JETSAM_REASON_INVALID;
4383 os_reason_t jetsam_reason = OS_REASON_NULL;
4384
4385 cause = kill_under_pressure_cause;
4386 switch (cause) {
4387 case kMemorystatusKilledFCThrashing:
4388 jetsam_reason_code = JETSAM_REASON_MEMORY_FCTHRASHING;
4389 break;
4390 case kMemorystatusKilledVMCompressorThrashing:
4391 jetsam_reason_code = JETSAM_REASON_MEMORY_VMCOMPRESSOR_THRASHING;
4392 break;
4393 case kMemorystatusKilledVMCompressorSpaceShortage:
4394 jetsam_reason_code = JETSAM_REASON_MEMORY_VMCOMPRESSOR_SPACE_SHORTAGE;
4395 break;
4396 case kMemorystatusKilledZoneMapExhaustion:
4397 jetsam_reason_code = JETSAM_REASON_ZONE_MAP_EXHAUSTION;
4398 break;
4399 case kMemorystatusKilledVMPageShortage:
4400 /* falls through */
4401 default:
4402 jetsam_reason_code = JETSAM_REASON_MEMORY_VMPAGESHORTAGE;
4403 cause = kMemorystatusKilledVMPageShortage;
4404 break;
4405 }
4406
4407 /* Highwater */
4408 boolean_t is_critical = TRUE;
4409 if (memorystatus_act_on_hiwat_processes(&errors, &hwm_kill, &post_snapshot, &is_critical)) {
4410 if (is_critical == FALSE) {
4411 /*
4412 * For now, don't kill any other processes.
4413 */
4414 break;
4415 } else {
4416 goto done;
4417 }
4418 }
4419
4420 jetsam_reason = os_reason_create(OS_REASON_JETSAM, jetsam_reason_code);
4421 if (jetsam_reason == OS_REASON_NULL) {
4422 printf("memorystatus_thread: failed to allocate jetsam reason\n");
4423 }
4424
4425 if (memorystatus_act_aggressive(cause, jetsam_reason, &jld_idle_kills, &corpse_list_purged, &post_snapshot)) {
4426 goto done;
4427 }
4428
4429 /*
4430 * memorystatus_kill_top_process() drops a reference,
4431 * so take another one so we can continue to use this exit reason
4432 * even after it returns
4433 */
4434 os_reason_ref(jetsam_reason);
4435
4436 /* LRU */
4437 killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, jetsam_reason, &priority, &errors);
4438 sort_flag = FALSE;
4439
4440 if (killed) {
4441 if (memorystatus_post_snapshot(priority, cause) == TRUE) {
4442 post_snapshot = TRUE;
4443 }
4444
4445 /* Jetsam Loop Detection */
4446 if (memorystatus_jld_enabled == TRUE) {
4447 if ((priority == JETSAM_PRIORITY_IDLE) || (priority == system_procs_aging_band) || (priority == applications_aging_band)) {
4448 jld_idle_kills++;
4449 } else {
4450 /*
4451 * We've reached into bands beyond idle deferred.
4452 * We make no attempt to monitor them
4453 */
4454 }
4455 }
4456
4457 if ((priority >= JETSAM_PRIORITY_UI_SUPPORT) && (total_corpses_count() > 0) && (corpse_list_purged == FALSE)) {
4458 /*
4459 * If we have jetsammed a process in or above JETSAM_PRIORITY_UI_SUPPORT
4460 * then we attempt to relieve pressure by purging corpse memory.
4461 */
4462 task_purge_all_corpses();
4463 corpse_list_purged = TRUE;
4464 }
4465 goto done;
4466 }
4467
4468 if (memorystatus_avail_pages_below_critical()) {
4469 /*
4470 * Still under pressure and unable to kill a process - purge corpse memory
4471 */
4472 if (total_corpses_count() > 0) {
4473 task_purge_all_corpses();
4474 corpse_list_purged = TRUE;
4475 }
4476
4477 if (memorystatus_avail_pages_below_critical()) {
4478 /*
4479 * Still under pressure and unable to kill a process - panic
4480 */
4481 panic("memorystatus_jetsam_thread: no victim! available pages:%llu\n", (uint64_t)memorystatus_available_pages);
4482 }
4483 }
4484
4485 done:
4486
4487 /*
4488 * We do not want to over-kill when thrashing has been detected.
4489 * To avoid that, we reset the flag here and notify the
4490 * compressor.
4491 */
4492 if (is_reason_thrashing(kill_under_pressure_cause)) {
4493 kill_under_pressure_cause = 0;
4494 #if CONFIG_JETSAM
4495 vm_thrashing_jetsam_done();
4496 #endif /* CONFIG_JETSAM */
4497 } else if (is_reason_zone_map_exhaustion(kill_under_pressure_cause)) {
4498 kill_under_pressure_cause = 0;
4499 }
4500
4501 os_reason_free(jetsam_reason);
4502 }
4503
4504 kill_under_pressure_cause = 0;
4505
4506 if (errors) {
4507 memorystatus_clear_errors();
4508 }
4509
4510 if (post_snapshot) {
4511 proc_list_lock();
4512 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4513 sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
4514 uint64_t timestamp_now = mach_absolute_time();
4515 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4516 memorystatus_jetsam_snapshot->js_gencount++;
4517 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4518 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4519 proc_list_unlock();
4520 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4521 if (!ret) {
4522 proc_list_lock();
4523 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4524 proc_list_unlock();
4525 }
4526 } else {
4527 proc_list_unlock();
4528 }
4529 }
4530
4531 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_END,
4532 memorystatus_available_pages, 0, 0, 0, 0);
4533
4534 memorystatus_thread_block(0, memorystatus_thread);
4535 }
4536
4537 /*
4538 * Returns TRUE:
4539 * when an idle-exitable proc was killed
4540 * Returns FALSE:
4541 * when there are no more idle-exitable procs found
4542 * when the attempt to kill an idle-exitable proc failed
4543 */
4544 boolean_t
4545 memorystatus_idle_exit_from_VM(void)
4546 {
4547 /*
4548 * This routine should no longer be needed since we are
4549 * now using jetsam bands on all platforms and so will deal
4550 * with IDLE processes within the memorystatus thread itself.
4551 *
4552 * But we still use it because we observed that macos systems
4553 * started heavy compression/swapping with a bunch of
4554 * idle-exitable processes alive and doing nothing. We decided
4555 * to rather kill those processes than start swapping earlier.
4556 */
4557
4558 return kill_idle_exit_proc();
4559 }
4560
4561 /*
4562 * Callback invoked when allowable physical memory footprint exceeded
4563 * (dirty pages + IOKit mappings)
4564 *
4565 * This is invoked for both advisory, non-fatal per-task high watermarks,
4566 * as well as the fatal task memory limits.
4567 */
4568 void
4569 memorystatus_on_ledger_footprint_exceeded(boolean_t warning, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
4570 {
4571 os_reason_t jetsam_reason = OS_REASON_NULL;
4572
4573 proc_t p = current_proc();
4574
4575 #if VM_PRESSURE_EVENTS
4576 if (warning == TRUE) {
4577 /*
4578 * This is a warning path which implies that the current process is close, but has
4579 * not yet exceeded its per-process memory limit.
4580 */
4581 if (memorystatus_warn_process(p->p_pid, memlimit_is_active, memlimit_is_fatal, FALSE /* not exceeded */) != TRUE) {
4582 /* Print warning, since it's possible that task has not registered for pressure notifications */
4583 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);
4584 }
4585 return;
4586 }
4587 #endif /* VM_PRESSURE_EVENTS */
4588
4589 if (memlimit_is_fatal) {
4590 /*
4591 * If this process has no high watermark or has a fatal task limit, then we have been invoked because the task
4592 * has violated either the system-wide per-task memory limit OR its own task limit.
4593 */
4594 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_PERPROCESSLIMIT);
4595 if (jetsam_reason == NULL) {
4596 printf("task_exceeded footprint: failed to allocate jetsam reason\n");
4597 } else if (corpse_for_fatal_memkill != 0 && proc_send_synchronous_EXC_RESOURCE(p) == FALSE) {
4598 /* Set OS_REASON_FLAG_GENERATE_CRASH_REPORT to generate corpse */
4599 jetsam_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
4600 }
4601
4602 if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit, jetsam_reason) != TRUE) {
4603 printf("task_exceeded_footprint: failed to kill the current task (exiting?).\n");
4604 }
4605 } else {
4606 /*
4607 * HWM offender exists. Done without locks or synchronization.
4608 * See comment near its declaration for more details.
4609 */
4610 memorystatus_hwm_candidates = TRUE;
4611
4612 #if VM_PRESSURE_EVENTS
4613 /*
4614 * The current process is not in the warning path.
4615 * This path implies the current process has exceeded a non-fatal (soft) memory limit.
4616 * Failure to send note is ignored here.
4617 */
4618 (void)memorystatus_warn_process(p->p_pid, memlimit_is_active, memlimit_is_fatal, TRUE /* exceeded */);
4619
4620 #endif /* VM_PRESSURE_EVENTS */
4621 }
4622 }
4623
4624 void
4625 memorystatus_log_exception(const int max_footprint_mb, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
4626 {
4627 proc_t p = current_proc();
4628
4629 /*
4630 * The limit violation is logged here, but only once per process per limit.
4631 * Soft memory limit is a non-fatal high-water-mark
4632 * Hard memory limit is a fatal custom-task-limit or system-wide per-task memory limit.
4633 */
4634
4635 os_log_with_startup_serial(OS_LOG_DEFAULT, "EXC_RESOURCE -> %s[%d] exceeded mem limit: %s%s %d MB (%s)\n",
4636 (*p->p_name ? p->p_name : "unknown"), p->p_pid, (memlimit_is_active ? "Active" : "Inactive"),
4637 (memlimit_is_fatal ? "Hard" : "Soft"), max_footprint_mb,
4638 (memlimit_is_fatal ? "fatal" : "non-fatal"));
4639
4640 return;
4641 }
4642
4643
4644 /*
4645 * Description:
4646 * Evaluates process state to determine which limit
4647 * should be applied (active vs. inactive limit).
4648 *
4649 * Processes that have the 'elevated inactive jetsam band' attribute
4650 * are first evaluated based on their current priority band.
4651 * presently elevated ==> active
4652 *
4653 * Processes that opt into dirty tracking are evaluated
4654 * based on clean vs dirty state.
4655 * dirty ==> active
4656 * clean ==> inactive
4657 *
4658 * Process that do not opt into dirty tracking are
4659 * evalulated based on priority level.
4660 * Foreground or above ==> active
4661 * Below Foreground ==> inactive
4662 *
4663 * Return: TRUE if active
4664 * False if inactive
4665 */
4666
4667 static boolean_t
4668 proc_jetsam_state_is_active_locked(proc_t p)
4669 {
4670 if ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) &&
4671 (p->p_memstat_effectivepriority == JETSAM_PRIORITY_ELEVATED_INACTIVE)) {
4672 /*
4673 * process has the 'elevated inactive jetsam band' attribute
4674 * and process is present in the elevated band
4675 * implies active state
4676 */
4677 return TRUE;
4678 } else if (p->p_memstat_dirty & P_DIRTY_TRACK) {
4679 /*
4680 * process has opted into dirty tracking
4681 * active state is based on dirty vs. clean
4682 */
4683 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
4684 /*
4685 * process is dirty
4686 * implies active state
4687 */
4688 return TRUE;
4689 } else {
4690 /*
4691 * process is clean
4692 * implies inactive state
4693 */
4694 return FALSE;
4695 }
4696 } else if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
4697 /*
4698 * process is Foreground or higher
4699 * implies active state
4700 */
4701 return TRUE;
4702 } else {
4703 /*
4704 * process found below Foreground
4705 * implies inactive state
4706 */
4707 return FALSE;
4708 }
4709 }
4710
4711 static boolean_t
4712 memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason)
4713 {
4714 boolean_t res;
4715
4716 uint32_t errors = 0;
4717
4718 if (victim_pid == -1) {
4719 /* No pid, so kill first process */
4720 res = memorystatus_kill_top_process(TRUE, TRUE, cause, jetsam_reason, NULL, &errors);
4721 } else {
4722 res = memorystatus_kill_specific_process(victim_pid, cause, jetsam_reason);
4723 }
4724
4725 if (errors) {
4726 memorystatus_clear_errors();
4727 }
4728
4729 if (res == TRUE) {
4730 /* Fire off snapshot notification */
4731 proc_list_lock();
4732 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
4733 sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
4734 uint64_t timestamp_now = mach_absolute_time();
4735 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
4736 if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
4737 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
4738 proc_list_unlock();
4739 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
4740 if (!ret) {
4741 proc_list_lock();
4742 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
4743 proc_list_unlock();
4744 }
4745 } else {
4746 proc_list_unlock();
4747 }
4748 }
4749
4750 return res;
4751 }
4752
4753 /*
4754 * Jetsam a specific process.
4755 */
4756 static boolean_t
4757 memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason)
4758 {
4759 boolean_t killed;
4760 proc_t p;
4761 uint64_t killtime = 0;
4762 clock_sec_t tv_sec;
4763 clock_usec_t tv_usec;
4764 uint32_t tv_msec;
4765
4766 /* TODO - add a victim queue and push this into the main jetsam thread */
4767
4768 p = proc_find(victim_pid);
4769 if (!p) {
4770 os_reason_free(jetsam_reason);
4771 return FALSE;
4772 }
4773
4774 proc_list_lock();
4775
4776 if (memorystatus_jetsam_snapshot_count == 0) {
4777 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
4778 }
4779
4780 killtime = mach_absolute_time();
4781 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
4782 tv_msec = tv_usec / 1000;
4783
4784 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
4785
4786 proc_list_unlock();
4787
4788 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_specific_process pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
4789 (unsigned long)tv_sec, tv_msec, victim_pid, (*p->p_name ? p->p_name : "unknown"),
4790 memorystatus_kill_cause_name[cause], p->p_memstat_effectivepriority, (uint64_t)memorystatus_available_pages);
4791
4792 killed = memorystatus_do_kill(p, cause, jetsam_reason);
4793 proc_rele(p);
4794
4795 return killed;
4796 }
4797
4798
4799 /*
4800 * Toggle the P_MEMSTAT_TERMINATED state.
4801 * Takes the proc_list_lock.
4802 */
4803 void
4804 proc_memstat_terminated(proc_t p, boolean_t set)
4805 {
4806 #if DEVELOPMENT || DEBUG
4807 if (p) {
4808 proc_list_lock();
4809 if (set == TRUE) {
4810 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
4811 } else {
4812 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
4813 }
4814 proc_list_unlock();
4815 }
4816 #else
4817 #pragma unused(p, set)
4818 /*
4819 * do nothing
4820 */
4821 #endif /* DEVELOPMENT || DEBUG */
4822 return;
4823 }
4824
4825
4826 #if CONFIG_JETSAM
4827 /*
4828 * This is invoked when cpulimits have been exceeded while in fatal mode.
4829 * The jetsam_flags do not apply as those are for memory related kills.
4830 * We call this routine so that the offending process is killed with
4831 * a non-zero exit status.
4832 */
4833 void
4834 jetsam_on_ledger_cpulimit_exceeded(void)
4835 {
4836 int retval = 0;
4837 int jetsam_flags = 0; /* make it obvious */
4838 proc_t p = current_proc();
4839 os_reason_t jetsam_reason = OS_REASON_NULL;
4840
4841 printf("task_exceeded_cpulimit: killing pid %d [%s]\n",
4842 p->p_pid, (*p->p_name ? p->p_name : "(unknown)"));
4843
4844 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_CPULIMIT);
4845 if (jetsam_reason == OS_REASON_NULL) {
4846 printf("task_exceeded_cpulimit: unable to allocate memory for jetsam reason\n");
4847 }
4848
4849 retval = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
4850
4851 if (retval) {
4852 printf("task_exceeded_cpulimit: failed to kill current task (exiting?).\n");
4853 }
4854 }
4855
4856 #endif /* CONFIG_JETSAM */
4857
4858 static void
4859 memorystatus_get_task_memory_region_count(task_t task, uint64_t *count)
4860 {
4861 assert(task);
4862 assert(count);
4863
4864 *count = get_task_memory_region_count(task);
4865 }
4866
4867
4868 #define MEMORYSTATUS_VM_MAP_FORK_ALLOWED 0x100000000
4869 #define MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED 0x200000000
4870
4871 #if DEVELOPMENT || DEBUG
4872
4873 /*
4874 * Sysctl only used to test memorystatus_allowed_vm_map_fork() path.
4875 * set a new pidwatch value
4876 * or
4877 * get the current pidwatch value
4878 *
4879 * The pidwatch_val starts out with a PID to watch for in the map_fork path.
4880 * Its value is:
4881 * - OR'd with MEMORYSTATUS_VM_MAP_FORK_ALLOWED if we allow the map_fork.
4882 * - OR'd with MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED if we disallow the map_fork.
4883 * - set to -1ull if the map_fork() is aborted for other reasons.
4884 */
4885
4886 uint64_t memorystatus_vm_map_fork_pidwatch_val = 0;
4887
4888 static int sysctl_memorystatus_vm_map_fork_pidwatch SYSCTL_HANDLER_ARGS {
4889 #pragma unused(oidp, arg1, arg2)
4890
4891 uint64_t new_value = 0;
4892 uint64_t old_value = 0;
4893 int error = 0;
4894
4895 /*
4896 * The pid is held in the low 32 bits.
4897 * The 'allowed' flags are in the upper 32 bits.
4898 */
4899 old_value = memorystatus_vm_map_fork_pidwatch_val;
4900
4901 error = sysctl_io_number(req, old_value, sizeof(old_value), &new_value, NULL);
4902
4903 if (error || !req->newptr) {
4904 /*
4905 * No new value passed in.
4906 */
4907 return error;
4908 }
4909
4910 /*
4911 * A new pid was passed in via req->newptr.
4912 * Ignore any attempt to set the higher order bits.
4913 */
4914 memorystatus_vm_map_fork_pidwatch_val = new_value & 0xFFFFFFFF;
4915 printf("memorystatus: pidwatch old_value = 0x%llx, new_value = 0x%llx \n", old_value, new_value);
4916
4917 return error;
4918 }
4919
4920 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_map_fork_pidwatch, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_MASKED,
4921 0, 0, sysctl_memorystatus_vm_map_fork_pidwatch, "Q", "get/set pid watched for in vm_map_fork");
4922
4923
4924 /*
4925 * Record if a watched process fails to qualify for a vm_map_fork().
4926 */
4927 void
4928 memorystatus_abort_vm_map_fork(task_t task)
4929 {
4930 if (memorystatus_vm_map_fork_pidwatch_val != 0) {
4931 proc_t p = get_bsdtask_info(task);
4932 if (p != NULL && memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid) {
4933 memorystatus_vm_map_fork_pidwatch_val = -1ull;
4934 }
4935 }
4936 }
4937
4938 static void
4939 set_vm_map_fork_pidwatch(task_t task, uint64_t x)
4940 {
4941 if (memorystatus_vm_map_fork_pidwatch_val != 0) {
4942 proc_t p = get_bsdtask_info(task);
4943 if (p && (memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid)) {
4944 memorystatus_vm_map_fork_pidwatch_val |= x;
4945 }
4946 }
4947 }
4948
4949 #else /* DEVELOPMENT || DEBUG */
4950
4951
4952 static void
4953 set_vm_map_fork_pidwatch(task_t task, uint64_t x)
4954 {
4955 #pragma unused(task)
4956 #pragma unused(x)
4957 }
4958
4959 #endif /* DEVELOPMENT || DEBUG */
4960
4961 /*
4962 * Called during EXC_RESOURCE handling when a process exceeds a soft
4963 * memory limit. This is the corpse fork path and here we decide if
4964 * vm_map_fork will be allowed when creating the corpse.
4965 * The task being considered is suspended.
4966 *
4967 * By default, a vm_map_fork is allowed to proceed.
4968 *
4969 * A few simple policy assumptions:
4970 * Desktop platform is not considered in this path.
4971 * The vm_map_fork is always allowed.
4972 *
4973 * If the device has a zero system-wide task limit,
4974 * then the vm_map_fork is allowed.
4975 *
4976 * And if a process's memory footprint calculates less
4977 * than or equal to half of the system-wide task limit,
4978 * then the vm_map_fork is allowed. This calculation
4979 * is based on the assumption that a process can
4980 * munch memory up to the system-wide task limit.
4981 */
4982 boolean_t
4983 memorystatus_allowed_vm_map_fork(task_t task)
4984 {
4985 boolean_t is_allowed = TRUE; /* default */
4986
4987 #if CONFIG_EMBEDDED
4988
4989 uint64_t footprint_in_bytes;
4990 uint64_t max_allowed_bytes;
4991
4992 if (max_task_footprint_mb == 0) {
4993 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_ALLOWED);
4994 return is_allowed;
4995 }
4996
4997 footprint_in_bytes = get_task_phys_footprint(task);
4998
4999 /*
5000 * Maximum is 1/4 of the system-wide task limit.
5001 */
5002 max_allowed_bytes = ((uint64_t)max_task_footprint_mb * 1024 * 1024) >> 2;
5003
5004 if (footprint_in_bytes > max_allowed_bytes) {
5005 printf("memorystatus disallowed vm_map_fork %lld %lld\n", footprint_in_bytes, max_allowed_bytes);
5006 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED);
5007 return !is_allowed;
5008 }
5009 #endif /* CONFIG_EMBEDDED */
5010
5011 set_vm_map_fork_pidwatch(task, MEMORYSTATUS_VM_MAP_FORK_ALLOWED);
5012 return is_allowed;
5013 }
5014
5015 static void
5016 memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages)
5017 {
5018 assert(task);
5019 assert(footprint);
5020
5021 uint64_t pages;
5022
5023 pages = (get_task_phys_footprint(task) / PAGE_SIZE_64);
5024 assert(((uint32_t)pages) == pages);
5025 *footprint = (uint32_t)pages;
5026
5027 if (max_footprint_lifetime) {
5028 pages = (get_task_phys_footprint_lifetime_max(task) / PAGE_SIZE_64);
5029 assert(((uint32_t)pages) == pages);
5030 *max_footprint_lifetime = (uint32_t)pages;
5031 }
5032 if (purgeable_pages) {
5033 pages = (get_task_purgeable_size(task) / PAGE_SIZE_64);
5034 assert(((uint32_t)pages) == pages);
5035 *purgeable_pages = (uint32_t)pages;
5036 }
5037 }
5038
5039 static void
5040 memorystatus_get_task_phys_footprint_page_counts(task_t task,
5041 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
5042 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
5043 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
5044 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages)
5045 {
5046 assert(task);
5047
5048 if (internal_pages) {
5049 *internal_pages = (get_task_internal(task) / PAGE_SIZE_64);
5050 }
5051
5052 if (internal_compressed_pages) {
5053 *internal_compressed_pages = (get_task_internal_compressed(task) / PAGE_SIZE_64);
5054 }
5055
5056 if (purgeable_nonvolatile_pages) {
5057 *purgeable_nonvolatile_pages = (get_task_purgeable_nonvolatile(task) / PAGE_SIZE_64);
5058 }
5059
5060 if (purgeable_nonvolatile_compressed_pages) {
5061 *purgeable_nonvolatile_compressed_pages = (get_task_purgeable_nonvolatile_compressed(task) / PAGE_SIZE_64);
5062 }
5063
5064 if (alternate_accounting_pages) {
5065 *alternate_accounting_pages = (get_task_alternate_accounting(task) / PAGE_SIZE_64);
5066 }
5067
5068 if (alternate_accounting_compressed_pages) {
5069 *alternate_accounting_compressed_pages = (get_task_alternate_accounting_compressed(task) / PAGE_SIZE_64);
5070 }
5071
5072 if (iokit_mapped_pages) {
5073 *iokit_mapped_pages = (get_task_iokit_mapped(task) / PAGE_SIZE_64);
5074 }
5075
5076 if (page_table_pages) {
5077 *page_table_pages = (get_task_page_table(task) / PAGE_SIZE_64);
5078 }
5079 }
5080
5081 /*
5082 * This routine only acts on the global jetsam event snapshot.
5083 * Updating the process's entry can race when the memorystatus_thread
5084 * has chosen to kill a process that is racing to exit on another core.
5085 */
5086 static void
5087 memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime)
5088 {
5089 memorystatus_jetsam_snapshot_entry_t *entry = NULL;
5090 memorystatus_jetsam_snapshot_t *snapshot = NULL;
5091 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
5092
5093 unsigned int i;
5094
5095 LCK_MTX_ASSERT(proc_list_mlock, LCK_MTX_ASSERT_OWNED);
5096
5097 if (memorystatus_jetsam_snapshot_count == 0) {
5098 /*
5099 * No active snapshot.
5100 * Nothing to do.
5101 */
5102 return;
5103 }
5104
5105 /*
5106 * Sanity check as this routine should only be called
5107 * from a jetsam kill path.
5108 */
5109 assert(kill_cause != 0 && killtime != 0);
5110
5111 snapshot = memorystatus_jetsam_snapshot;
5112 snapshot_list = memorystatus_jetsam_snapshot->entries;
5113
5114 for (i = 0; i < memorystatus_jetsam_snapshot_count; i++) {
5115 if (snapshot_list[i].pid == p->p_pid) {
5116 entry = &snapshot_list[i];
5117
5118 if (entry->killed || entry->jse_killtime) {
5119 /*
5120 * We apparently raced on the exit path
5121 * for this process, as it's snapshot entry
5122 * has already recorded a kill.
5123 */
5124 assert(entry->killed && entry->jse_killtime);
5125 break;
5126 }
5127
5128 /*
5129 * Update the entry we just found in the snapshot.
5130 */
5131
5132 entry->killed = kill_cause;
5133 entry->jse_killtime = killtime;
5134 entry->jse_gencount = snapshot->js_gencount;
5135 entry->jse_idle_delta = p->p_memstat_idle_delta;
5136 #if CONFIG_FREEZE
5137 entry->jse_thaw_count = p->p_memstat_thaw_count;
5138 #else /* CONFIG_FREEZE */
5139 entry->jse_thaw_count = 0;
5140 #endif /* CONFIG_FREEZE */
5141
5142 /*
5143 * If a process has moved between bands since snapshot was
5144 * initialized, then likely these fields changed too.
5145 */
5146 if (entry->priority != p->p_memstat_effectivepriority) {
5147 strlcpy(entry->name, p->p_name, sizeof(entry->name));
5148 entry->priority = p->p_memstat_effectivepriority;
5149 entry->state = memorystatus_build_state(p);
5150 entry->user_data = p->p_memstat_userdata;
5151 entry->fds = p->p_fd->fd_nfiles;
5152 }
5153
5154 /*
5155 * Always update the page counts on a kill.
5156 */
5157
5158 uint32_t pages = 0;
5159 uint32_t max_pages_lifetime = 0;
5160 uint32_t purgeable_pages = 0;
5161
5162 memorystatus_get_task_page_counts(p->task, &pages, &max_pages_lifetime, &purgeable_pages);
5163 entry->pages = (uint64_t)pages;
5164 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
5165 entry->purgeable_pages = (uint64_t)purgeable_pages;
5166
5167 uint64_t internal_pages = 0;
5168 uint64_t internal_compressed_pages = 0;
5169 uint64_t purgeable_nonvolatile_pages = 0;
5170 uint64_t purgeable_nonvolatile_compressed_pages = 0;
5171 uint64_t alternate_accounting_pages = 0;
5172 uint64_t alternate_accounting_compressed_pages = 0;
5173 uint64_t iokit_mapped_pages = 0;
5174 uint64_t page_table_pages = 0;
5175
5176 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
5177 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
5178 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
5179 &iokit_mapped_pages, &page_table_pages);
5180
5181 entry->jse_internal_pages = internal_pages;
5182 entry->jse_internal_compressed_pages = internal_compressed_pages;
5183 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
5184 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
5185 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
5186 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
5187 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
5188 entry->jse_page_table_pages = page_table_pages;
5189
5190 uint64_t region_count = 0;
5191 memorystatus_get_task_memory_region_count(p->task, &region_count);
5192 entry->jse_memory_region_count = region_count;
5193
5194 goto exit;
5195 }
5196 }
5197
5198 if (entry == NULL) {
5199 /*
5200 * The entry was not found in the snapshot, so the process must have
5201 * launched after the snapshot was initialized.
5202 * Let's try to append the new entry.
5203 */
5204 if (memorystatus_jetsam_snapshot_count < memorystatus_jetsam_snapshot_max) {
5205 /*
5206 * A populated snapshot buffer exists
5207 * and there is room to init a new entry.
5208 */
5209 assert(memorystatus_jetsam_snapshot_count == snapshot->entry_count);
5210
5211 unsigned int next = memorystatus_jetsam_snapshot_count;
5212
5213 if (memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[next], (snapshot->js_gencount)) == TRUE) {
5214 entry = &snapshot_list[next];
5215 entry->killed = kill_cause;
5216 entry->jse_killtime = killtime;
5217
5218 snapshot->entry_count = ++next;
5219 memorystatus_jetsam_snapshot_count = next;
5220
5221 if (memorystatus_jetsam_snapshot_count >= memorystatus_jetsam_snapshot_max) {
5222 /*
5223 * We just used the last slot in the snapshot buffer.
5224 * We only want to log it once... so we do it here
5225 * when we notice we've hit the max.
5226 */
5227 printf("memorystatus: WARNING snapshot buffer is full, count %d\n",
5228 memorystatus_jetsam_snapshot_count);
5229 }
5230 }
5231 }
5232 }
5233
5234 exit:
5235 if (entry == NULL) {
5236 /*
5237 * If we reach here, the snapshot buffer could not be updated.
5238 * Most likely, the buffer is full, in which case we would have
5239 * logged a warning in the previous call.
5240 *
5241 * For now, we will stop appending snapshot entries.
5242 * When the buffer is consumed, the snapshot state will reset.
5243 */
5244
5245 MEMORYSTATUS_DEBUG(4, "memorystatus_update_jetsam_snapshot_entry_locked: failed to update pid %d, priority %d, count %d\n",
5246 p->p_pid, p->p_memstat_effectivepriority, memorystatus_jetsam_snapshot_count);
5247 }
5248
5249 return;
5250 }
5251
5252 #if CONFIG_JETSAM
5253 void
5254 memorystatus_pages_update(unsigned int pages_avail)
5255 {
5256 memorystatus_available_pages = pages_avail;
5257
5258 #if VM_PRESSURE_EVENTS
5259 /*
5260 * Since memorystatus_available_pages changes, we should
5261 * re-evaluate the pressure levels on the system and
5262 * check if we need to wake the pressure thread.
5263 * We also update memorystatus_level in that routine.
5264 */
5265 vm_pressure_response();
5266
5267 if (memorystatus_available_pages <= memorystatus_available_pages_pressure) {
5268 if (memorystatus_hwm_candidates || (memorystatus_available_pages <= memorystatus_available_pages_critical)) {
5269 memorystatus_thread_wake();
5270 }
5271 }
5272 #if CONFIG_FREEZE
5273 /*
5274 * We can't grab the freezer_mutex here even though that synchronization would be correct to inspect
5275 * the # of frozen processes and wakeup the freezer thread. Reason being that we come here into this
5276 * code with (possibly) the page-queue locks held and preemption disabled. So trying to grab a mutex here
5277 * will result in the "mutex with preemption disabled" panic.
5278 */
5279
5280 if (memorystatus_freeze_thread_should_run() == TRUE) {
5281 /*
5282 * The freezer thread is usually woken up by some user-space call i.e. pid_hibernate(any process).
5283 * That trigger isn't invoked often enough and so we are enabling this explicit wakeup here.
5284 */
5285 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5286 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
5287 }
5288 }
5289 #endif /* CONFIG_FREEZE */
5290
5291 #else /* VM_PRESSURE_EVENTS */
5292
5293 boolean_t critical, delta;
5294
5295 if (!memorystatus_delta) {
5296 return;
5297 }
5298
5299 critical = (pages_avail < memorystatus_available_pages_critical) ? TRUE : FALSE;
5300 delta = ((pages_avail >= (memorystatus_available_pages + memorystatus_delta))
5301 || (memorystatus_available_pages >= (pages_avail + memorystatus_delta))) ? TRUE : FALSE;
5302
5303 if (critical || delta) {
5304 unsigned int total_pages;
5305
5306 total_pages = (unsigned int) atop_64(max_mem);
5307 #if CONFIG_SECLUDED_MEMORY
5308 total_pages -= vm_page_secluded_count;
5309 #endif /* CONFIG_SECLUDED_MEMORY */
5310 memorystatus_level = memorystatus_available_pages * 100 / total_pages;
5311 memorystatus_thread_wake();
5312 }
5313 #endif /* VM_PRESSURE_EVENTS */
5314 }
5315 #endif /* CONFIG_JETSAM */
5316
5317 static boolean_t
5318 memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount)
5319 {
5320 clock_sec_t tv_sec;
5321 clock_usec_t tv_usec;
5322 uint32_t pages = 0;
5323 uint32_t max_pages_lifetime = 0;
5324 uint32_t purgeable_pages = 0;
5325 uint64_t internal_pages = 0;
5326 uint64_t internal_compressed_pages = 0;
5327 uint64_t purgeable_nonvolatile_pages = 0;
5328 uint64_t purgeable_nonvolatile_compressed_pages = 0;
5329 uint64_t alternate_accounting_pages = 0;
5330 uint64_t alternate_accounting_compressed_pages = 0;
5331 uint64_t iokit_mapped_pages = 0;
5332 uint64_t page_table_pages = 0;
5333 uint64_t region_count = 0;
5334 uint64_t cids[COALITION_NUM_TYPES];
5335
5336 memset(entry, 0, sizeof(memorystatus_jetsam_snapshot_entry_t));
5337
5338 entry->pid = p->p_pid;
5339 strlcpy(&entry->name[0], p->p_name, sizeof(entry->name));
5340 entry->priority = p->p_memstat_effectivepriority;
5341
5342 memorystatus_get_task_page_counts(p->task, &pages, &max_pages_lifetime, &purgeable_pages);
5343 entry->pages = (uint64_t)pages;
5344 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
5345 entry->purgeable_pages = (uint64_t)purgeable_pages;
5346
5347 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
5348 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
5349 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
5350 &iokit_mapped_pages, &page_table_pages);
5351
5352 entry->jse_internal_pages = internal_pages;
5353 entry->jse_internal_compressed_pages = internal_compressed_pages;
5354 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
5355 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
5356 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
5357 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
5358 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
5359 entry->jse_page_table_pages = page_table_pages;
5360
5361 memorystatus_get_task_memory_region_count(p->task, &region_count);
5362 entry->jse_memory_region_count = region_count;
5363
5364 entry->state = memorystatus_build_state(p);
5365 entry->user_data = p->p_memstat_userdata;
5366 memcpy(&entry->uuid[0], &p->p_uuid[0], sizeof(p->p_uuid));
5367 entry->fds = p->p_fd->fd_nfiles;
5368
5369 absolutetime_to_microtime(get_task_cpu_time(p->task), &tv_sec, &tv_usec);
5370 entry->cpu_time.tv_sec = (int64_t)tv_sec;
5371 entry->cpu_time.tv_usec = (int64_t)tv_usec;
5372
5373 assert(p->p_stats != NULL);
5374 entry->jse_starttime = p->p_stats->ps_start; /* abstime process started */
5375 entry->jse_killtime = 0; /* abstime jetsam chose to kill process */
5376 entry->killed = 0; /* the jetsam kill cause */
5377 entry->jse_gencount = gencount; /* indicates a pass through jetsam thread, when process was targeted to be killed */
5378
5379 entry->jse_idle_delta = p->p_memstat_idle_delta; /* Most recent timespan spent in idle-band */
5380
5381 #if CONFIG_FREEZE
5382 entry->jse_thaw_count = p->p_memstat_thaw_count;
5383 #else /* CONFIG_FREEZE */
5384 entry->jse_thaw_count = 0;
5385 #endif /* CONFIG_FREEZE */
5386
5387 proc_coalitionids(p, cids);
5388 entry->jse_coalition_jetsam_id = cids[COALITION_TYPE_JETSAM];
5389
5390 return TRUE;
5391 }
5392
5393 static void
5394 memorystatus_init_snapshot_vmstats(memorystatus_jetsam_snapshot_t *snapshot)
5395 {
5396 kern_return_t kr = KERN_SUCCESS;
5397 mach_msg_type_number_t count = HOST_VM_INFO64_COUNT;
5398 vm_statistics64_data_t vm_stat;
5399
5400 if ((kr = host_statistics64(host_self(), HOST_VM_INFO64, (host_info64_t)&vm_stat, &count)) != KERN_SUCCESS) {
5401 printf("memorystatus_init_jetsam_snapshot_stats: host_statistics64 failed with %d\n", kr);
5402 memset(&snapshot->stats, 0, sizeof(snapshot->stats));
5403 } else {
5404 snapshot->stats.free_pages = vm_stat.free_count;
5405 snapshot->stats.active_pages = vm_stat.active_count;
5406 snapshot->stats.inactive_pages = vm_stat.inactive_count;
5407 snapshot->stats.throttled_pages = vm_stat.throttled_count;
5408 snapshot->stats.purgeable_pages = vm_stat.purgeable_count;
5409 snapshot->stats.wired_pages = vm_stat.wire_count;
5410
5411 snapshot->stats.speculative_pages = vm_stat.speculative_count;
5412 snapshot->stats.filebacked_pages = vm_stat.external_page_count;
5413 snapshot->stats.anonymous_pages = vm_stat.internal_page_count;
5414 snapshot->stats.compressions = vm_stat.compressions;
5415 snapshot->stats.decompressions = vm_stat.decompressions;
5416 snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
5417 snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
5418 }
5419
5420 get_zone_map_size(&snapshot->stats.zone_map_size, &snapshot->stats.zone_map_capacity);
5421 get_largest_zone_info(snapshot->stats.largest_zone_name, sizeof(snapshot->stats.largest_zone_name),
5422 &snapshot->stats.largest_zone_size);
5423 }
5424
5425 /*
5426 * Collect vm statistics at boot.
5427 * Called only once (see kern_exec.c)
5428 * Data can be consumed at any time.
5429 */
5430 void
5431 memorystatus_init_at_boot_snapshot()
5432 {
5433 memorystatus_init_snapshot_vmstats(&memorystatus_at_boot_snapshot);
5434 memorystatus_at_boot_snapshot.entry_count = 0;
5435 memorystatus_at_boot_snapshot.notification_time = 0; /* updated when consumed */
5436 memorystatus_at_boot_snapshot.snapshot_time = mach_absolute_time();
5437 }
5438
5439 static void
5440 memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count )
5441 {
5442 proc_t p, next_p;
5443 unsigned int b = 0, i = 0;
5444
5445 memorystatus_jetsam_snapshot_t *snapshot = NULL;
5446 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
5447 unsigned int snapshot_max = 0;
5448
5449 LCK_MTX_ASSERT(proc_list_mlock, LCK_MTX_ASSERT_OWNED);
5450
5451 if (od_snapshot) {
5452 /*
5453 * This is an on_demand snapshot
5454 */
5455 snapshot = od_snapshot;
5456 snapshot_list = od_snapshot->entries;
5457 snapshot_max = ods_list_count;
5458 } else {
5459 /*
5460 * This is a jetsam event snapshot
5461 */
5462 snapshot = memorystatus_jetsam_snapshot;
5463 snapshot_list = memorystatus_jetsam_snapshot->entries;
5464 snapshot_max = memorystatus_jetsam_snapshot_max;
5465 }
5466
5467 /*
5468 * Init the snapshot header information
5469 */
5470 memorystatus_init_snapshot_vmstats(snapshot);
5471 snapshot->snapshot_time = mach_absolute_time();
5472 snapshot->notification_time = 0;
5473 snapshot->js_gencount = 0;
5474
5475 next_p = memorystatus_get_first_proc_locked(&b, TRUE);
5476 while (next_p) {
5477 p = next_p;
5478 next_p = memorystatus_get_next_proc_locked(&b, p, TRUE);
5479
5480 if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i], snapshot->js_gencount)) {
5481 continue;
5482 }
5483
5484 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",
5485 p->p_pid,
5486 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],
5487 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]);
5488
5489 if (++i == snapshot_max) {
5490 break;
5491 }
5492 }
5493
5494 snapshot->entry_count = i;
5495
5496 if (!od_snapshot) {
5497 /* update the system buffer count */
5498 memorystatus_jetsam_snapshot_count = i;
5499 }
5500 }
5501
5502 #if DEVELOPMENT || DEBUG
5503
5504 #if CONFIG_JETSAM
5505 static int
5506 memorystatus_cmd_set_panic_bits(user_addr_t buffer, uint32_t buffer_size)
5507 {
5508 int ret;
5509 memorystatus_jetsam_panic_options_t debug;
5510
5511 if (buffer_size != sizeof(memorystatus_jetsam_panic_options_t)) {
5512 return EINVAL;
5513 }
5514
5515 ret = copyin(buffer, &debug, buffer_size);
5516 if (ret) {
5517 return ret;
5518 }
5519
5520 /* Panic bits match kMemorystatusKilled* enum */
5521 memorystatus_jetsam_panic_debug = (memorystatus_jetsam_panic_debug & ~debug.mask) | (debug.data & debug.mask);
5522
5523 /* Copyout new value */
5524 debug.data = memorystatus_jetsam_panic_debug;
5525 ret = copyout(&debug, buffer, sizeof(memorystatus_jetsam_panic_options_t));
5526
5527 return ret;
5528 }
5529 #endif /* CONFIG_JETSAM */
5530
5531 /*
5532 * Triggers a sort_order on a specified jetsam priority band.
5533 * This is for testing only, used to force a path through the sort
5534 * function.
5535 */
5536 static int
5537 memorystatus_cmd_test_jetsam_sort(int priority, int sort_order)
5538 {
5539 int error = 0;
5540
5541 unsigned int bucket_index = 0;
5542
5543 if (priority == -1) {
5544 /* Use as shorthand for default priority */
5545 bucket_index = JETSAM_PRIORITY_DEFAULT;
5546 } else {
5547 bucket_index = (unsigned int)priority;
5548 }
5549
5550 error = memorystatus_sort_bucket(bucket_index, sort_order);
5551
5552 return error;
5553 }
5554
5555 #endif /* DEVELOPMENT || DEBUG */
5556
5557 /*
5558 * Prepare the process to be killed (set state, update snapshot) and kill it.
5559 */
5560 static uint64_t memorystatus_purge_before_jetsam_success = 0;
5561
5562 static boolean_t
5563 memorystatus_kill_proc(proc_t p, uint32_t cause, os_reason_t jetsam_reason, boolean_t *killed)
5564 {
5565 pid_t aPid = 0;
5566 uint32_t aPid_ep = 0;
5567
5568 uint64_t killtime = 0;
5569 clock_sec_t tv_sec;
5570 clock_usec_t tv_usec;
5571 uint32_t tv_msec;
5572 boolean_t retval = FALSE;
5573 uint64_t num_pages_purged = 0;
5574
5575 aPid = p->p_pid;
5576 aPid_ep = p->p_memstat_effectivepriority;
5577
5578 if (cause != kMemorystatusKilledVnodes && cause != kMemorystatusKilledZoneMapExhaustion) {
5579 /*
5580 * Genuine memory pressure and not other (vnode/zone) resource exhaustion.
5581 */
5582 boolean_t success = FALSE;
5583
5584 networking_memstatus_callout(p, cause);
5585 num_pages_purged = vm_purgeable_purge_task_owned(p->task);
5586
5587 if (num_pages_purged) {
5588 /*
5589 * We actually purged something and so let's
5590 * check if we need to continue with the kill.
5591 */
5592 if (cause == kMemorystatusKilledHiwat) {
5593 uint64_t footprint_in_bytes = get_task_phys_footprint(p->task);
5594 uint64_t memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
5595 success = (footprint_in_bytes <= memlimit_in_bytes);
5596 } else {
5597 success = (memorystatus_avail_pages_below_pressure() == FALSE);
5598 }
5599
5600 if (success) {
5601 memorystatus_purge_before_jetsam_success++;
5602
5603 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: purged %llu pages from pid %d [%s] and avoided %s\n",
5604 num_pages_purged, aPid, (*p->p_name ? p->p_name : "unknown"), memorystatus_kill_cause_name[cause]);
5605
5606 *killed = FALSE;
5607
5608 return TRUE;
5609 }
5610 }
5611 }
5612
5613 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
5614 MEMORYSTATUS_DEBUG(1, "jetsam: %s pid %d [%s] - %lld Mb > 1 (%d Mb)\n",
5615 (memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing",
5616 aPid, (*p->p_name ? p->p_name : "unknown"),
5617 (footprint_in_bytes / (1024ULL * 1024ULL)), /* converted bytes to MB */
5618 p->p_memstat_memlimit);
5619 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
5620
5621 killtime = mach_absolute_time();
5622 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5623 tv_msec = tv_usec / 1000;
5624
5625 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
5626 if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
5627 if (cause == kMemorystatusKilledHiwat) {
5628 MEMORYSTATUS_DEBUG(1, "jetsam: suspending pid %d [%s] for diagnosis - memorystatus_available_pages: %d\n",
5629 aPid, (*p->p_name ? p->p_name: "(unknown)"), memorystatus_available_pages);
5630 } else {
5631 int activeProcess = p->p_memstat_state & P_MEMSTAT_FOREGROUND;
5632 if (activeProcess) {
5633 MEMORYSTATUS_DEBUG(1, "jetsam: suspending pid %d [%s] (active) for diagnosis - memorystatus_available_pages: %d\n",
5634 aPid, (*p->p_name ? p->p_name: "(unknown)"), memorystatus_available_pages);
5635
5636 if (memorystatus_jetsam_policy & kPolicyDiagnoseFirst) {
5637 jetsam_diagnostic_suspended_one_active_proc = 1;
5638 printf("jetsam: returning after suspending first active proc - %d\n", aPid);
5639 }
5640 }
5641 }
5642
5643 proc_list_lock();
5644 /* This diagnostic code is going away soon. Ignore the kMemorystatusInvalid cause here. */
5645 memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusInvalid, killtime);
5646 proc_list_unlock();
5647
5648 p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
5649
5650 if (p) {
5651 task_suspend(p->task);
5652 *killed = TRUE;
5653 }
5654 } else
5655 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
5656 {
5657 proc_list_lock();
5658 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5659 proc_list_unlock();
5660
5661 char kill_reason_string[128];
5662
5663 if (cause == kMemorystatusKilledHiwat) {
5664 strlcpy(kill_reason_string, "killing_highwater_process", 128);
5665 } else {
5666 if (aPid_ep == JETSAM_PRIORITY_IDLE) {
5667 strlcpy(kill_reason_string, "killing_idle_process", 128);
5668 } else {
5669 strlcpy(kill_reason_string, "killing_top_process", 128);
5670 }
5671 }
5672
5673 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: %s pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
5674 (unsigned long)tv_sec, tv_msec, kill_reason_string,
5675 aPid, (*p->p_name ? p->p_name : "unknown"),
5676 memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
5677
5678 /*
5679 * memorystatus_do_kill drops a reference, so take another one so we can
5680 * continue to use this exit reason even after memorystatus_do_kill()
5681 * returns
5682 */
5683 os_reason_ref(jetsam_reason);
5684
5685 retval = memorystatus_do_kill(p, cause, jetsam_reason);
5686
5687 *killed = retval;
5688 }
5689
5690 return retval;
5691 }
5692
5693 /*
5694 * Jetsam the first process in the queue.
5695 */
5696 static boolean_t
5697 memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, os_reason_t jetsam_reason,
5698 int32_t *priority, uint32_t *errors)
5699 {
5700 pid_t aPid;
5701 proc_t p = PROC_NULL, next_p = PROC_NULL;
5702 boolean_t new_snapshot = FALSE, force_new_snapshot = FALSE, killed = FALSE, freed_mem = FALSE;
5703 unsigned int i = 0;
5704 uint32_t aPid_ep;
5705 int32_t local_max_kill_prio = JETSAM_PRIORITY_IDLE;
5706
5707 #ifndef CONFIG_FREEZE
5708 #pragma unused(any)
5709 #endif
5710
5711 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5712 memorystatus_available_pages, 0, 0, 0, 0);
5713
5714
5715 #if CONFIG_JETSAM
5716 if (sort_flag == TRUE) {
5717 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
5718 }
5719
5720 local_max_kill_prio = max_kill_priority;
5721
5722 force_new_snapshot = FALSE;
5723
5724 #else /* CONFIG_JETSAM */
5725
5726 if (sort_flag == TRUE) {
5727 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_IDLE, JETSAM_SORT_DEFAULT);
5728 }
5729
5730 /*
5731 * On macos, we currently only have 2 reasons to be here:
5732 *
5733 * kMemorystatusKilledZoneMapExhaustion
5734 * AND
5735 * kMemorystatusKilledVMCompressorSpaceShortage
5736 *
5737 * If we are here because of kMemorystatusKilledZoneMapExhaustion, we will consider
5738 * any and all processes as eligible kill candidates since we need to avoid a panic.
5739 *
5740 * Since this function can be called async. it is harder to toggle the max_kill_priority
5741 * value before and after a call. And so we use this local variable to set the upper band
5742 * on the eligible kill bands.
5743 */
5744 if (cause == kMemorystatusKilledZoneMapExhaustion) {
5745 local_max_kill_prio = JETSAM_PRIORITY_MAX;
5746 } else {
5747 local_max_kill_prio = max_kill_priority;
5748 }
5749
5750 /*
5751 * And, because we are here under extreme circumstances, we force a snapshot even for
5752 * IDLE kills.
5753 */
5754 force_new_snapshot = TRUE;
5755
5756 #endif /* CONFIG_JETSAM */
5757
5758 proc_list_lock();
5759
5760 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5761 while (next_p && (next_p->p_memstat_effectivepriority <= local_max_kill_prio)) {
5762 #if DEVELOPMENT || DEBUG
5763 int procSuspendedForDiagnosis;
5764 #endif /* DEVELOPMENT || DEBUG */
5765
5766 p = next_p;
5767 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5768
5769 #if DEVELOPMENT || DEBUG
5770 procSuspendedForDiagnosis = p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED;
5771 #endif /* DEVELOPMENT || DEBUG */
5772
5773 aPid = p->p_pid;
5774 aPid_ep = p->p_memstat_effectivepriority;
5775
5776 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5777 continue; /* with lock held */
5778 }
5779
5780 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
5781 if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && procSuspendedForDiagnosis) {
5782 printf("jetsam: continuing after ignoring proc suspended already for diagnosis - %d\n", aPid);
5783 continue;
5784 }
5785 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
5786
5787 if (cause == kMemorystatusKilledVnodes) {
5788 /*
5789 * If the system runs out of vnodes, we systematically jetsam
5790 * processes in hopes of stumbling onto a vnode gain that helps
5791 * the system recover. The process that happens to trigger
5792 * this path has no known relationship to the vnode shortage.
5793 * Deadlock avoidance: attempt to safeguard the caller.
5794 */
5795
5796 if (p == current_proc()) {
5797 /* do not jetsam the current process */
5798 continue;
5799 }
5800 }
5801
5802 #if CONFIG_FREEZE
5803 boolean_t skip;
5804 boolean_t reclaim_proc = !(p->p_memstat_state & P_MEMSTAT_LOCKED);
5805 if (any || reclaim_proc) {
5806 skip = FALSE;
5807 } else {
5808 skip = TRUE;
5809 }
5810
5811 if (skip) {
5812 continue;
5813 } else
5814 #endif
5815 {
5816 if (proc_ref_locked(p) == p) {
5817 /*
5818 * Mark as terminated so that if exit1() indicates success, but the process (for example)
5819 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
5820 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
5821 * acquisition of the proc lock.
5822 */
5823 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5824 } else {
5825 /*
5826 * We need to restart the search again because
5827 * proc_ref_locked _can_ drop the proc_list lock
5828 * and we could have lost our stored next_p via
5829 * an exit() on another core.
5830 */
5831 i = 0;
5832 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5833 continue;
5834 }
5835
5836 /*
5837 * Capture a snapshot if none exists and:
5838 * - we are forcing a new snapshot creation, either because:
5839 * - on a particular platform we need these snapshots every time, OR
5840 * - a boot-arg/embedded device tree property has been set.
5841 * - priority was not requested (this is something other than an ambient kill)
5842 * - the priority was requested *and* the targeted process is not at idle priority
5843 */
5844 if ((memorystatus_jetsam_snapshot_count == 0) &&
5845 (force_new_snapshot || memorystatus_idle_snapshot || ((!priority) || (priority && (aPid_ep != JETSAM_PRIORITY_IDLE))))) {
5846 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5847 new_snapshot = TRUE;
5848 }
5849
5850 proc_list_unlock();
5851
5852 freed_mem = memorystatus_kill_proc(p, cause, jetsam_reason, &killed); /* purged and/or killed 'p' */
5853 /* Success? */
5854 if (freed_mem) {
5855 if (killed) {
5856 if (priority) {
5857 *priority = aPid_ep;
5858 }
5859 } else {
5860 /* purged */
5861 proc_list_lock();
5862 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5863 proc_list_unlock();
5864 }
5865 proc_rele(p);
5866 goto exit;
5867 }
5868
5869 /*
5870 * Failure - first unwind the state,
5871 * then fall through to restart the search.
5872 */
5873 proc_list_lock();
5874 proc_rele_locked(p);
5875 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5876 p->p_memstat_state |= P_MEMSTAT_ERROR;
5877 *errors += 1;
5878
5879 i = 0;
5880 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5881 }
5882 }
5883
5884 proc_list_unlock();
5885
5886 exit:
5887 os_reason_free(jetsam_reason);
5888
5889 /* Clear snapshot if freshly captured and no target was found */
5890 if (new_snapshot && !killed) {
5891 proc_list_lock();
5892 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5893 proc_list_unlock();
5894 }
5895
5896 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
5897 memorystatus_available_pages, killed ? aPid : 0, 0, 0, 0);
5898
5899 return killed;
5900 }
5901
5902 /*
5903 * Jetsam aggressively
5904 */
5905 static boolean_t
5906 memorystatus_kill_top_process_aggressive(uint32_t cause, int aggr_count,
5907 int32_t priority_max, uint32_t *errors)
5908 {
5909 pid_t aPid;
5910 proc_t p = PROC_NULL, next_p = PROC_NULL;
5911 boolean_t new_snapshot = FALSE, killed = FALSE;
5912 int kill_count = 0;
5913 unsigned int i = 0;
5914 int32_t aPid_ep = 0;
5915 unsigned int memorystatus_level_snapshot = 0;
5916 uint64_t killtime = 0;
5917 clock_sec_t tv_sec;
5918 clock_usec_t tv_usec;
5919 uint32_t tv_msec;
5920 os_reason_t jetsam_reason = OS_REASON_NULL;
5921
5922 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5923 memorystatus_available_pages, priority_max, 0, 0, 0);
5924
5925 memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
5926
5927 jetsam_reason = os_reason_create(OS_REASON_JETSAM, cause);
5928 if (jetsam_reason == OS_REASON_NULL) {
5929 printf("memorystatus_kill_top_process_aggressive: failed to allocate exit reason\n");
5930 }
5931
5932 proc_list_lock();
5933
5934 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5935 while (next_p) {
5936 #if DEVELOPMENT || DEBUG
5937 int activeProcess;
5938 int procSuspendedForDiagnosis;
5939 #endif /* DEVELOPMENT || DEBUG */
5940
5941 if (((next_p->p_listflag & P_LIST_EXITED) != 0) ||
5942 ((unsigned int)(next_p->p_memstat_effectivepriority) != i)) {
5943 /*
5944 * We have raced with next_p running on another core.
5945 * It may be exiting or it may have moved to a different
5946 * jetsam priority band. This means we have lost our
5947 * place in line while traversing the jetsam list. We
5948 * attempt to recover by rewinding to the beginning of the band
5949 * we were already traversing. By doing this, we do not guarantee
5950 * that no process escapes this aggressive march, but we can make
5951 * skipping an entire range of processes less likely. (PR-21069019)
5952 */
5953
5954 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: rewinding band %d, %s(%d) moved or exiting.\n",
5955 aggr_count, i, (*next_p->p_name ? next_p->p_name : "unknown"), next_p->p_pid);
5956
5957 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5958 continue;
5959 }
5960
5961 p = next_p;
5962 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5963
5964 if (p->p_memstat_effectivepriority > priority_max) {
5965 /*
5966 * Bail out of this killing spree if we have
5967 * reached beyond the priority_max jetsam band.
5968 * That is, we kill up to and through the
5969 * priority_max jetsam band.
5970 */
5971 proc_list_unlock();
5972 goto exit;
5973 }
5974
5975 #if DEVELOPMENT || DEBUG
5976 activeProcess = p->p_memstat_state & P_MEMSTAT_FOREGROUND;
5977 procSuspendedForDiagnosis = p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED;
5978 #endif /* DEVELOPMENT || DEBUG */
5979
5980 aPid = p->p_pid;
5981 aPid_ep = p->p_memstat_effectivepriority;
5982
5983 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5984 continue;
5985 }
5986
5987 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
5988 if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && procSuspendedForDiagnosis) {
5989 printf("jetsam: continuing after ignoring proc suspended already for diagnosis - %d\n", aPid);
5990 continue;
5991 }
5992 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
5993
5994 /*
5995 * Capture a snapshot if none exists.
5996 */
5997 if (memorystatus_jetsam_snapshot_count == 0) {
5998 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
5999 new_snapshot = TRUE;
6000 }
6001
6002 /*
6003 * Mark as terminated so that if exit1() indicates success, but the process (for example)
6004 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
6005 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
6006 * acquisition of the proc lock.
6007 */
6008 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
6009
6010 killtime = mach_absolute_time();
6011 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
6012 tv_msec = tv_usec / 1000;
6013
6014 /* Shift queue, update stats */
6015 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
6016
6017 /*
6018 * In order to kill the target process, we will drop the proc_list_lock.
6019 * To guaranteee that p and next_p don't disappear out from under the lock,
6020 * we must take a ref on both.
6021 * If we cannot get a reference, then it's likely we've raced with
6022 * that process exiting on another core.
6023 */
6024 if (proc_ref_locked(p) == p) {
6025 if (next_p) {
6026 while (next_p && (proc_ref_locked(next_p) != next_p)) {
6027 proc_t temp_p;
6028
6029 /*
6030 * We must have raced with next_p exiting on another core.
6031 * Recover by getting the next eligible process in the band.
6032 */
6033
6034 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: skipping %d [%s] (exiting?)\n",
6035 aggr_count, next_p->p_pid, (*next_p->p_name ? next_p->p_name : "(unknown)"));
6036
6037 temp_p = next_p;
6038 next_p = memorystatus_get_next_proc_locked(&i, temp_p, TRUE);
6039 }
6040 }
6041 proc_list_unlock();
6042
6043 printf("%lu.%03d memorystatus: %s%d pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
6044 (unsigned long)tv_sec, tv_msec,
6045 ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "killing_idle_process_aggressive" : "killing_top_process_aggressive"),
6046 aggr_count, aPid, (*p->p_name ? p->p_name : "unknown"),
6047 memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
6048
6049 memorystatus_level_snapshot = memorystatus_level;
6050
6051 /*
6052 * memorystatus_do_kill() drops a reference, so take another one so we can
6053 * continue to use this exit reason even after memorystatus_do_kill()
6054 * returns.
6055 */
6056 os_reason_ref(jetsam_reason);
6057 killed = memorystatus_do_kill(p, cause, jetsam_reason);
6058
6059 /* Success? */
6060 if (killed) {
6061 proc_rele(p);
6062 kill_count++;
6063 p = NULL;
6064 killed = FALSE;
6065
6066 /*
6067 * Continue the killing spree.
6068 */
6069 proc_list_lock();
6070 if (next_p) {
6071 proc_rele_locked(next_p);
6072 }
6073
6074 if (aPid_ep == JETSAM_PRIORITY_FOREGROUND && memorystatus_aggressive_jetsam_lenient == TRUE) {
6075 if (memorystatus_level > memorystatus_level_snapshot && ((memorystatus_level - memorystatus_level_snapshot) >= AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD)) {
6076 #if DEVELOPMENT || DEBUG
6077 printf("Disabling Lenient mode after one-time deployment.\n");
6078 #endif /* DEVELOPMENT || DEBUG */
6079 memorystatus_aggressive_jetsam_lenient = FALSE;
6080 break;
6081 }
6082 }
6083
6084 continue;
6085 }
6086
6087 /*
6088 * Failure - first unwind the state,
6089 * then fall through to restart the search.
6090 */
6091 proc_list_lock();
6092 proc_rele_locked(p);
6093 if (next_p) {
6094 proc_rele_locked(next_p);
6095 }
6096 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6097 p->p_memstat_state |= P_MEMSTAT_ERROR;
6098 *errors += 1;
6099 p = NULL;
6100 }
6101
6102 /*
6103 * Failure - restart the search at the beginning of
6104 * the band we were already traversing.
6105 *
6106 * We might have raced with "p" exiting on another core, resulting in no
6107 * ref on "p". Or, we may have failed to kill "p".
6108 *
6109 * Either way, we fall thru to here, leaving the proc in the
6110 * P_MEMSTAT_TERMINATED or P_MEMSTAT_ERROR state.
6111 *
6112 * And, we hold the the proc_list_lock at this point.
6113 */
6114
6115 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6116 }
6117
6118 proc_list_unlock();
6119
6120 exit:
6121 os_reason_free(jetsam_reason);
6122
6123 /* Clear snapshot if freshly captured and no target was found */
6124 if (new_snapshot && (kill_count == 0)) {
6125 proc_list_lock();
6126 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6127 proc_list_unlock();
6128 }
6129
6130 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
6131 memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
6132
6133 if (kill_count > 0) {
6134 return TRUE;
6135 } else {
6136 return FALSE;
6137 }
6138 }
6139
6140 static boolean_t
6141 memorystatus_kill_hiwat_proc(uint32_t *errors, boolean_t *purged)
6142 {
6143 pid_t aPid = 0;
6144 proc_t p = PROC_NULL, next_p = PROC_NULL;
6145 boolean_t new_snapshot = FALSE, killed = FALSE, freed_mem = FALSE;
6146 unsigned int i = 0;
6147 uint32_t aPid_ep;
6148 os_reason_t jetsam_reason = OS_REASON_NULL;
6149 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
6150 memorystatus_available_pages, 0, 0, 0, 0);
6151
6152 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_HIGHWATER);
6153 if (jetsam_reason == OS_REASON_NULL) {
6154 printf("memorystatus_kill_hiwat_proc: failed to allocate exit reason\n");
6155 }
6156
6157 proc_list_lock();
6158
6159 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6160 while (next_p) {
6161 uint64_t footprint_in_bytes = 0;
6162 uint64_t memlimit_in_bytes = 0;
6163 boolean_t skip = 0;
6164
6165 p = next_p;
6166 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6167
6168 aPid = p->p_pid;
6169 aPid_ep = p->p_memstat_effectivepriority;
6170
6171 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
6172 continue;
6173 }
6174
6175 /* skip if no limit set */
6176 if (p->p_memstat_memlimit <= 0) {
6177 continue;
6178 }
6179
6180 footprint_in_bytes = get_task_phys_footprint(p->task);
6181 memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
6182 skip = (footprint_in_bytes <= memlimit_in_bytes);
6183
6184 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
6185 if (!skip && (memorystatus_jetsam_policy & kPolicyDiagnoseActive)) {
6186 if (p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED) {
6187 continue;
6188 }
6189 }
6190 #endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
6191
6192 #if CONFIG_FREEZE
6193 if (!skip) {
6194 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
6195 skip = TRUE;
6196 } else {
6197 skip = FALSE;
6198 }
6199 }
6200 #endif
6201
6202 if (skip) {
6203 continue;
6204 } else {
6205 if (memorystatus_jetsam_snapshot_count == 0) {
6206 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
6207 new_snapshot = TRUE;
6208 }
6209
6210 if (proc_ref_locked(p) == p) {
6211 /*
6212 * Mark as terminated so that if exit1() indicates success, but the process (for example)
6213 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
6214 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
6215 * acquisition of the proc lock.
6216 */
6217 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
6218
6219 proc_list_unlock();
6220 } else {
6221 /*
6222 * We need to restart the search again because
6223 * proc_ref_locked _can_ drop the proc_list lock
6224 * and we could have lost our stored next_p via
6225 * an exit() on another core.
6226 */
6227 i = 0;
6228 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6229 continue;
6230 }
6231
6232 freed_mem = memorystatus_kill_proc(p, kMemorystatusKilledHiwat, jetsam_reason, &killed); /* purged and/or killed 'p' */
6233
6234 /* Success? */
6235 if (freed_mem) {
6236 if (killed == FALSE) {
6237 /* purged 'p'..don't reset HWM candidate count */
6238 *purged = TRUE;
6239
6240 proc_list_lock();
6241 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6242 proc_list_unlock();
6243 }
6244 proc_rele(p);
6245 goto exit;
6246 }
6247 /*
6248 * Failure - first unwind the state,
6249 * then fall through to restart the search.
6250 */
6251 proc_list_lock();
6252 proc_rele_locked(p);
6253 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6254 p->p_memstat_state |= P_MEMSTAT_ERROR;
6255 *errors += 1;
6256
6257 i = 0;
6258 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
6259 }
6260 }
6261
6262 proc_list_unlock();
6263
6264 exit:
6265 os_reason_free(jetsam_reason);
6266
6267 /* Clear snapshot if freshly captured and no target was found */
6268 if (new_snapshot && !killed) {
6269 proc_list_lock();
6270 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6271 proc_list_unlock();
6272 }
6273
6274 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_END,
6275 memorystatus_available_pages, killed ? aPid : 0, 0, 0, 0);
6276
6277 return killed;
6278 }
6279
6280 /*
6281 * Jetsam a process pinned in the elevated band.
6282 *
6283 * Return: true -- at least one pinned process was jetsammed
6284 * false -- no pinned process was jetsammed
6285 */
6286 static boolean_t
6287 memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, unsigned int band, int aggr_count, uint32_t *errors)
6288 {
6289 pid_t aPid = 0;
6290 proc_t p = PROC_NULL, next_p = PROC_NULL;
6291 boolean_t new_snapshot = FALSE, killed = FALSE;
6292 int kill_count = 0;
6293 uint32_t aPid_ep;
6294 uint64_t killtime = 0;
6295 clock_sec_t tv_sec;
6296 clock_usec_t tv_usec;
6297 uint32_t tv_msec;
6298
6299
6300 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
6301 memorystatus_available_pages, 0, 0, 0, 0);
6302
6303 #if CONFIG_FREEZE
6304 boolean_t consider_frozen_only = FALSE;
6305
6306 if (band == (unsigned int) memorystatus_freeze_jetsam_band) {
6307 consider_frozen_only = TRUE;
6308 }
6309 #endif /* CONFIG_FREEZE */
6310
6311 proc_list_lock();
6312
6313 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
6314 while (next_p) {
6315 p = next_p;
6316 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
6317
6318 aPid = p->p_pid;
6319 aPid_ep = p->p_memstat_effectivepriority;
6320
6321 /*
6322 * Only pick a process pinned in this elevated band
6323 */
6324 if (!(p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
6325 continue;
6326 }
6327
6328 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
6329 continue;
6330 }
6331
6332 #if CONFIG_FREEZE
6333 if (consider_frozen_only && !(p->p_memstat_state & P_MEMSTAT_FROZEN)) {
6334 continue;
6335 }
6336
6337 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
6338 continue;
6339 }
6340 #endif /* CONFIG_FREEZE */
6341
6342 #if DEVELOPMENT || DEBUG
6343 MEMORYSTATUS_DEBUG(1, "jetsam: elevated%d process pid %d [%s] - memorystatus_available_pages: %d\n",
6344 aggr_count,
6345 aPid, (*p->p_name ? p->p_name : "unknown"),
6346 memorystatus_available_pages);
6347 #endif /* DEVELOPMENT || DEBUG */
6348
6349 if (memorystatus_jetsam_snapshot_count == 0) {
6350 memorystatus_init_jetsam_snapshot_locked(NULL, 0);
6351 new_snapshot = TRUE;
6352 }
6353
6354 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
6355
6356 killtime = mach_absolute_time();
6357 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
6358 tv_msec = tv_usec / 1000;
6359
6360 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
6361
6362 if (proc_ref_locked(p) == p) {
6363 proc_list_unlock();
6364
6365 os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_top_process_elevated%d pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
6366 (unsigned long)tv_sec, tv_msec,
6367 aggr_count,
6368 aPid, (*p->p_name ? p->p_name : "unknown"),
6369 memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
6370
6371 /*
6372 * memorystatus_do_kill drops a reference, so take another one so we can
6373 * continue to use this exit reason even after memorystatus_do_kill()
6374 * returns
6375 */
6376 os_reason_ref(jetsam_reason);
6377 killed = memorystatus_do_kill(p, cause, jetsam_reason);
6378
6379 /* Success? */
6380 if (killed) {
6381 proc_rele(p);
6382 kill_count++;
6383 goto exit;
6384 }
6385
6386 /*
6387 * Failure - first unwind the state,
6388 * then fall through to restart the search.
6389 */
6390 proc_list_lock();
6391 proc_rele_locked(p);
6392 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
6393 p->p_memstat_state |= P_MEMSTAT_ERROR;
6394 *errors += 1;
6395 }
6396
6397 /*
6398 * Failure - restart the search.
6399 *
6400 * We might have raced with "p" exiting on another core, resulting in no
6401 * ref on "p". Or, we may have failed to kill "p".
6402 *
6403 * Either way, we fall thru to here, leaving the proc in the
6404 * P_MEMSTAT_TERMINATED state or P_MEMSTAT_ERROR state.
6405 *
6406 * And, we hold the the proc_list_lock at this point.
6407 */
6408
6409 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
6410 }
6411
6412 proc_list_unlock();
6413
6414 exit:
6415 os_reason_free(jetsam_reason);
6416
6417 /* Clear snapshot if freshly captured and no target was found */
6418 if (new_snapshot && (kill_count == 0)) {
6419 proc_list_lock();
6420 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
6421 proc_list_unlock();
6422 }
6423
6424 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
6425 memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
6426
6427 return killed;
6428 }
6429
6430 static boolean_t
6431 memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause)
6432 {
6433 /*
6434 * TODO: allow a general async path
6435 *
6436 * NOTE: If a new async kill cause is added, make sure to update memorystatus_thread() to
6437 * add the appropriate exit reason code mapping.
6438 */
6439 if ((victim_pid != -1) ||
6440 (cause != kMemorystatusKilledVMPageShortage &&
6441 cause != kMemorystatusKilledVMCompressorThrashing &&
6442 cause != kMemorystatusKilledVMCompressorSpaceShortage &&
6443 cause != kMemorystatusKilledFCThrashing &&
6444 cause != kMemorystatusKilledZoneMapExhaustion)) {
6445 return FALSE;
6446 }
6447
6448 kill_under_pressure_cause = cause;
6449 memorystatus_thread_wake();
6450 return TRUE;
6451 }
6452
6453 boolean_t
6454 memorystatus_kill_on_VM_compressor_space_shortage(boolean_t async)
6455 {
6456 if (async) {
6457 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMCompressorSpaceShortage);
6458 } else {
6459 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMCOMPRESSOR_SPACE_SHORTAGE);
6460 if (jetsam_reason == OS_REASON_NULL) {
6461 printf("memorystatus_kill_on_VM_compressor_space_shortage -- sync: failed to allocate jetsam reason\n");
6462 }
6463
6464 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMCompressorSpaceShortage, jetsam_reason);
6465 }
6466 }
6467
6468 #if CONFIG_JETSAM
6469 boolean_t
6470 memorystatus_kill_on_VM_compressor_thrashing(boolean_t async)
6471 {
6472 if (async) {
6473 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMCompressorThrashing);
6474 } else {
6475 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMCOMPRESSOR_THRASHING);
6476 if (jetsam_reason == OS_REASON_NULL) {
6477 printf("memorystatus_kill_on_VM_compressor_thrashing -- sync: failed to allocate jetsam reason\n");
6478 }
6479
6480 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMCompressorThrashing, jetsam_reason);
6481 }
6482 }
6483
6484 boolean_t
6485 memorystatus_kill_on_VM_page_shortage(boolean_t async)
6486 {
6487 if (async) {
6488 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMPageShortage);
6489 } else {
6490 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMPAGESHORTAGE);
6491 if (jetsam_reason == OS_REASON_NULL) {
6492 printf("memorystatus_kill_on_VM_page_shortage -- sync: failed to allocate jetsam reason\n");
6493 }
6494
6495 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMPageShortage, jetsam_reason);
6496 }
6497 }
6498
6499 boolean_t
6500 memorystatus_kill_on_FC_thrashing(boolean_t async)
6501 {
6502 if (async) {
6503 return memorystatus_kill_process_async(-1, kMemorystatusKilledFCThrashing);
6504 } else {
6505 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_FCTHRASHING);
6506 if (jetsam_reason == OS_REASON_NULL) {
6507 printf("memorystatus_kill_on_FC_thrashing -- sync: failed to allocate jetsam reason\n");
6508 }
6509
6510 return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing, jetsam_reason);
6511 }
6512 }
6513
6514 boolean_t
6515 memorystatus_kill_on_vnode_limit(void)
6516 {
6517 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_VNODE);
6518 if (jetsam_reason == OS_REASON_NULL) {
6519 printf("memorystatus_kill_on_vnode_limit: failed to allocate jetsam reason\n");
6520 }
6521
6522 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes, jetsam_reason);
6523 }
6524
6525 #endif /* CONFIG_JETSAM */
6526
6527 boolean_t
6528 memorystatus_kill_on_zone_map_exhaustion(pid_t pid)
6529 {
6530 boolean_t res = FALSE;
6531 if (pid == -1) {
6532 res = memorystatus_kill_process_async(-1, kMemorystatusKilledZoneMapExhaustion);
6533 } else {
6534 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_ZONE_MAP_EXHAUSTION);
6535 if (jetsam_reason == OS_REASON_NULL) {
6536 printf("memorystatus_kill_on_zone_map_exhaustion: failed to allocate jetsam reason\n");
6537 }
6538
6539 res = memorystatus_kill_process_sync(pid, kMemorystatusKilledZoneMapExhaustion, jetsam_reason);
6540 }
6541 return res;
6542 }
6543
6544 #if CONFIG_FREEZE
6545
6546 __private_extern__ void
6547 memorystatus_freeze_init(void)
6548 {
6549 kern_return_t result;
6550 thread_t thread;
6551
6552 freezer_lck_grp_attr = lck_grp_attr_alloc_init();
6553 freezer_lck_grp = lck_grp_alloc_init("freezer", freezer_lck_grp_attr);
6554
6555 lck_mtx_init(&freezer_mutex, freezer_lck_grp, NULL);
6556
6557 /*
6558 * This is just the default value if the underlying
6559 * storage device doesn't have any specific budget.
6560 * We check with the storage layer in memorystatus_freeze_update_throttle()
6561 * before we start our freezing the first time.
6562 */
6563 memorystatus_freeze_budget_pages_remaining = (memorystatus_freeze_daily_mb_max * 1024 * 1024) / PAGE_SIZE;
6564
6565 result = kernel_thread_start(memorystatus_freeze_thread, NULL, &thread);
6566 if (result == KERN_SUCCESS) {
6567 proc_set_thread_policy(thread, TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER2);
6568 proc_set_thread_policy(thread, TASK_POLICY_INTERNAL, TASK_POLICY_PASSIVE_IO, TASK_POLICY_ENABLE);
6569 thread_set_thread_name(thread, "VM_freezer");
6570
6571 thread_deallocate(thread);
6572 } else {
6573 panic("Could not create memorystatus_freeze_thread");
6574 }
6575 }
6576
6577 static boolean_t
6578 memorystatus_is_process_eligible_for_freeze(proc_t p)
6579 {
6580 /*
6581 * Called with proc_list_lock held.
6582 */
6583
6584 LCK_MTX_ASSERT(proc_list_mlock, LCK_MTX_ASSERT_OWNED);
6585
6586 boolean_t should_freeze = FALSE;
6587 uint32_t state = 0, entry_count = 0, pages = 0, i = 0;
6588 int probability_of_use = 0;
6589
6590 if (isApp(p) == FALSE) {
6591 goto out;
6592 }
6593
6594 state = p->p_memstat_state;
6595
6596 if ((state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FREEZE_DISABLED | P_MEMSTAT_FREEZE_IGNORE)) ||
6597 !(state & P_MEMSTAT_SUSPENDED)) {
6598 goto out;
6599 }
6600
6601 /* Only freeze processes meeting our minimum resident page criteria */
6602 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL);
6603 if (pages < memorystatus_freeze_pages_min) {
6604 goto out;
6605 }
6606
6607 entry_count = (memorystatus_global_probabilities_size / sizeof(memorystatus_internal_probabilities_t));
6608
6609 if (entry_count) {
6610 for (i = 0; i < entry_count; i++) {
6611 if (strncmp(memorystatus_global_probabilities_table[i].proc_name,
6612 p->p_name,
6613 MAXCOMLEN + 1) == 0) {
6614 probability_of_use = memorystatus_global_probabilities_table[i].use_probability;
6615 break;
6616 }
6617 }
6618
6619 if (probability_of_use == 0) {
6620 goto out;
6621 }
6622 }
6623
6624 should_freeze = TRUE;
6625 out:
6626 return should_freeze;
6627 }
6628
6629 /*
6630 * Synchronously freeze the passed proc. Called with a reference to the proc held.
6631 *
6632 * Doesn't deal with re-freezing because this is called on a specific process and
6633 * not by the freezer thread. If that changes, we'll have to teach it about
6634 * refreezing a frozen process.
6635 *
6636 * Returns EINVAL or the value returned by task_freeze().
6637 */
6638 int
6639 memorystatus_freeze_process_sync(proc_t p)
6640 {
6641 int ret = EINVAL;
6642 pid_t aPid = 0;
6643 boolean_t memorystatus_freeze_swap_low = FALSE;
6644 int freezer_error_code = 0;
6645
6646 lck_mtx_lock(&freezer_mutex);
6647
6648 if (p == NULL) {
6649 printf("memorystatus_freeze_process_sync: Invalid process\n");
6650 goto exit;
6651 }
6652
6653 if (memorystatus_freeze_enabled == FALSE) {
6654 printf("memorystatus_freeze_process_sync: Freezing is DISABLED\n");
6655 goto exit;
6656 }
6657
6658 if (!memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
6659 printf("memorystatus_freeze_process_sync: Low compressor and/or low swap space...skipping freeze\n");
6660 goto exit;
6661 }
6662
6663 memorystatus_freeze_update_throttle(&memorystatus_freeze_budget_pages_remaining);
6664 if (!memorystatus_freeze_budget_pages_remaining) {
6665 printf("memorystatus_freeze_process_sync: exit with NO available budget\n");
6666 goto exit;
6667 }
6668
6669 proc_list_lock();
6670
6671 if (p != NULL) {
6672 uint32_t purgeable, wired, clean, dirty, shared;
6673 uint32_t max_pages, i;
6674
6675 aPid = p->p_pid;
6676
6677 /* Ensure the process is eligible for freezing */
6678 if (memorystatus_is_process_eligible_for_freeze(p) == FALSE) {
6679 proc_list_unlock();
6680 goto exit;
6681 }
6682
6683 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
6684 max_pages = MIN(memorystatus_freeze_pages_max, memorystatus_freeze_budget_pages_remaining);
6685 } else {
6686 /*
6687 * We only have the compressor without any swap.
6688 */
6689 max_pages = UINT32_MAX - 1;
6690 }
6691
6692 /* Mark as locked temporarily to avoid kill */
6693 p->p_memstat_state |= P_MEMSTAT_LOCKED;
6694 proc_list_unlock();
6695
6696 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_START,
6697 memorystatus_available_pages, 0, 0, 0, 0);
6698
6699 ret = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, &freezer_error_code, FALSE /* eval only */);
6700
6701 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_END,
6702 memorystatus_available_pages, aPid, 0, 0, 0);
6703
6704 DTRACE_MEMORYSTATUS6(memorystatus_freeze, proc_t, p, unsigned int, memorystatus_available_pages, boolean_t, purgeable, unsigned int, wired, uint32_t, clean, uint32_t, dirty);
6705
6706 MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_process_sync: task_freeze %s for pid %d [%s] - "
6707 "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, max_pages %d, shared %d\n",
6708 (ret == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (*p->p_name ? p->p_name : "(unknown)"),
6709 memorystatus_available_pages, purgeable, wired, clean, dirty, max_pages, shared);
6710
6711 proc_list_lock();
6712
6713 if (ret == KERN_SUCCESS) {
6714 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: freezing (specific) pid %d [%s]...done",
6715 aPid, (*p->p_name ? p->p_name : "unknown"));
6716
6717 memorystatus_freeze_entry_t data = { aPid, TRUE, dirty };
6718
6719 p->p_memstat_freeze_sharedanon_pages += shared;
6720
6721 memorystatus_frozen_shared_mb += shared;
6722
6723 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) == 0) {
6724 p->p_memstat_state |= P_MEMSTAT_FROZEN;
6725 memorystatus_frozen_count++;
6726 }
6727
6728 p->p_memstat_frozen_count++;
6729
6730 /*
6731 * Still keeping the P_MEMSTAT_LOCKED bit till we are actually done elevating this frozen process
6732 * to its higher jetsam band.
6733 */
6734 proc_list_unlock();
6735
6736 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
6737
6738 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
6739 ret = memorystatus_update_inactive_jetsam_priority_band(p->p_pid, MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE,
6740 memorystatus_freeze_jetsam_band, TRUE);
6741
6742 if (ret) {
6743 printf("Elevating the frozen process failed with %d\n", ret);
6744 /* not fatal */
6745 ret = 0;
6746 }
6747
6748 proc_list_lock();
6749
6750 /* Update stats */
6751 for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
6752 throttle_intervals[i].pageouts += dirty;
6753 }
6754 } else {
6755 proc_list_lock();
6756 }
6757
6758 memorystatus_freeze_pageouts += dirty;
6759
6760 if (memorystatus_frozen_count == (memorystatus_frozen_processes_max - 1)) {
6761 /*
6762 * Add some eviction logic here? At some point should we
6763 * jetsam a process to get back its swap space so that we
6764 * can freeze a more eligible process at this moment in time?
6765 */
6766 }
6767 } else {
6768 char reason[128];
6769 if (freezer_error_code == FREEZER_ERROR_EXCESS_SHARED_MEMORY) {
6770 strlcpy(reason, "too much shared memory", 128);
6771 }
6772
6773 if (freezer_error_code == FREEZER_ERROR_LOW_PRIVATE_SHARED_RATIO) {
6774 strlcpy(reason, "low private-shared pages ratio", 128);
6775 }
6776
6777 if (freezer_error_code == FREEZER_ERROR_NO_COMPRESSOR_SPACE) {
6778 strlcpy(reason, "no compressor space", 128);
6779 }
6780
6781 if (freezer_error_code == FREEZER_ERROR_NO_SWAP_SPACE) {
6782 strlcpy(reason, "no swap space", 128);
6783 }
6784
6785 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: freezing (specific) pid %d [%s]...skipped (%s)",
6786 aPid, (*p->p_name ? p->p_name : "unknown"), reason);
6787 p->p_memstat_state |= P_MEMSTAT_FREEZE_IGNORE;
6788 }
6789
6790 p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
6791 proc_list_unlock();
6792 }
6793
6794 exit:
6795 lck_mtx_unlock(&freezer_mutex);
6796
6797 return ret;
6798 }
6799
6800 static int
6801 memorystatus_freeze_top_process(void)
6802 {
6803 pid_t aPid = 0;
6804 int ret = -1;
6805 proc_t p = PROC_NULL, next_p = PROC_NULL;
6806 unsigned int i = 0;
6807 unsigned int band = JETSAM_PRIORITY_IDLE;
6808 boolean_t refreeze_processes = FALSE;
6809
6810 proc_list_lock();
6811
6812 if (memorystatus_frozen_count >= memorystatus_frozen_processes_max) {
6813 /*
6814 * Freezer is already full but we are here and so let's
6815 * try to refreeze any processes we might have thawed
6816 * in the past and push out their compressed state out.
6817 */
6818 refreeze_processes = TRUE;
6819 band = (unsigned int) memorystatus_freeze_jetsam_band;
6820 }
6821
6822 freeze_process:
6823
6824 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
6825 while (next_p) {
6826 kern_return_t kr;
6827 uint32_t purgeable, wired, clean, dirty, shared;
6828 uint32_t max_pages = 0;
6829 int freezer_error_code = 0;
6830
6831 p = next_p;
6832 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
6833
6834 aPid = p->p_pid;
6835
6836 if (p->p_memstat_effectivepriority != (int32_t) band) {
6837 /*
6838 * We shouldn't be freezing processes outside the
6839 * prescribed band.
6840 */
6841 break;
6842 }
6843
6844 /* Ensure the process is eligible for (re-)freezing */
6845 if (refreeze_processes) {
6846 /*
6847 * Has to have been frozen once before.
6848 */
6849 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) == FALSE) {
6850 continue;
6851 }
6852
6853 /*
6854 * Has to have been resumed once before.
6855 */
6856 if ((p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) == FALSE) {
6857 continue;
6858 }
6859
6860 /*
6861 * Not currently being looked at for something.
6862 */
6863 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
6864 continue;
6865 }
6866
6867 /*
6868 * We are going to try and refreeze and so re-evaluate
6869 * the process. We don't want to double count the shared
6870 * memory. So deduct the old snapshot here.
6871 */
6872 memorystatus_frozen_shared_mb -= p->p_memstat_freeze_sharedanon_pages;
6873 p->p_memstat_freeze_sharedanon_pages = 0;
6874
6875 p->p_memstat_state &= ~P_MEMSTAT_REFREEZE_ELIGIBLE;
6876 memorystatus_refreeze_eligible_count--;
6877 } else {
6878 if (memorystatus_is_process_eligible_for_freeze(p) == FALSE) {
6879 continue; // with lock held
6880 }
6881 }
6882
6883 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
6884 /*
6885 * Freezer backed by the compressor and swap file(s)
6886 * will hold compressed data.
6887 */
6888
6889 max_pages = MIN(memorystatus_freeze_pages_max, memorystatus_freeze_budget_pages_remaining);
6890 } else {
6891 /*
6892 * We only have the compressor pool.
6893 */
6894 max_pages = UINT32_MAX - 1;
6895 }
6896
6897 /* Mark as locked temporarily to avoid kill */
6898 p->p_memstat_state |= P_MEMSTAT_LOCKED;
6899
6900 p = proc_ref_locked(p);
6901 if (!p) {
6902 break;
6903 }
6904
6905 proc_list_unlock();
6906
6907 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_START,
6908 memorystatus_available_pages, 0, 0, 0, 0);
6909
6910 kr = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, &freezer_error_code, FALSE /* eval only */);
6911
6912 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_END,
6913 memorystatus_available_pages, aPid, 0, 0, 0);
6914
6915 MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_top_process: task_freeze %s for pid %d [%s] - "
6916 "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, max_pages %d, shared %d\n",
6917 (kr == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (*p->p_name ? p->p_name : "(unknown)"),
6918 memorystatus_available_pages, purgeable, wired, clean, dirty, max_pages, shared);
6919
6920 proc_list_lock();
6921
6922 /* Success? */
6923 if (KERN_SUCCESS == kr) {
6924 if (refreeze_processes) {
6925 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: Refreezing (general) pid %d [%s]...done",
6926 aPid, (*p->p_name ? p->p_name : "unknown"));
6927 } else {
6928 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: freezing (general) pid %d [%s]...done",
6929 aPid, (*p->p_name ? p->p_name : "unknown"));
6930 }
6931
6932 memorystatus_freeze_entry_t data = { aPid, TRUE, dirty };
6933
6934 p->p_memstat_freeze_sharedanon_pages += shared;
6935
6936 memorystatus_frozen_shared_mb += shared;
6937
6938 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) == 0) {
6939 p->p_memstat_state |= P_MEMSTAT_FROZEN;
6940 memorystatus_frozen_count++;
6941 }
6942
6943 p->p_memstat_frozen_count++;
6944
6945 /*
6946 * Still keeping the P_MEMSTAT_LOCKED bit till we are actually done elevating this frozen process
6947 * to its higher jetsam band.
6948 */
6949 proc_list_unlock();
6950
6951 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
6952
6953 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
6954 ret = memorystatus_update_inactive_jetsam_priority_band(p->p_pid, MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE, memorystatus_freeze_jetsam_band, TRUE);
6955
6956 if (ret) {
6957 printf("Elevating the frozen process failed with %d\n", ret);
6958 /* not fatal */
6959 ret = 0;
6960 }
6961
6962 proc_list_lock();
6963
6964 /* Update stats */
6965 for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
6966 throttle_intervals[i].pageouts += dirty;
6967 }
6968 } else {
6969 proc_list_lock();
6970 }
6971
6972 memorystatus_freeze_pageouts += dirty;
6973
6974 if (memorystatus_frozen_count == (memorystatus_frozen_processes_max - 1)) {
6975 /*
6976 * Add some eviction logic here? At some point should we
6977 * jetsam a process to get back its swap space so that we
6978 * can freeze a more eligible process at this moment in time?
6979 */
6980 }
6981
6982 /* Return KERN_SUCCESS */
6983 ret = kr;
6984
6985 p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
6986 proc_rele_locked(p);
6987
6988 /*
6989 * We froze a process successfully. We can stop now
6990 * and see if that helped.
6991 */
6992
6993 break;
6994 } else {
6995 p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
6996
6997 if (refreeze_processes == TRUE) {
6998 if ((freezer_error_code == FREEZER_ERROR_EXCESS_SHARED_MEMORY) ||
6999 (freezer_error_code == FREEZER_ERROR_LOW_PRIVATE_SHARED_RATIO)) {
7000 /*
7001 * Keeping this prior-frozen process in this high band when
7002 * we failed to re-freeze it due to bad shared memory usage
7003 * could cause excessive pressure on the lower bands.
7004 * We need to demote it for now. It'll get re-evaluated next
7005 * time because we don't set the P_MEMSTAT_FREEZE_IGNORE
7006 * bit.
7007 */
7008
7009 p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
7010 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
7011 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, TRUE, TRUE);
7012 }
7013 } else {
7014 p->p_memstat_state |= P_MEMSTAT_FREEZE_IGNORE;
7015 }
7016
7017 proc_rele_locked(p);
7018
7019 char reason[128];
7020 if (freezer_error_code == FREEZER_ERROR_EXCESS_SHARED_MEMORY) {
7021 strlcpy(reason, "too much shared memory", 128);
7022 }
7023
7024 if (freezer_error_code == FREEZER_ERROR_LOW_PRIVATE_SHARED_RATIO) {
7025 strlcpy(reason, "low private-shared pages ratio", 128);
7026 }
7027
7028 if (freezer_error_code == FREEZER_ERROR_NO_COMPRESSOR_SPACE) {
7029 strlcpy(reason, "no compressor space", 128);
7030 }
7031
7032 if (freezer_error_code == FREEZER_ERROR_NO_SWAP_SPACE) {
7033 strlcpy(reason, "no swap space", 128);
7034 }
7035
7036 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: freezing (general) pid %d [%s]...skipped (%s)",
7037 aPid, (*p->p_name ? p->p_name : "unknown"), reason);
7038
7039 if (vm_compressor_low_on_space() || vm_swap_low_on_space()) {
7040 break;
7041 }
7042 }
7043 }
7044
7045 if ((ret == -1) &&
7046 (memorystatus_refreeze_eligible_count >= MIN_THAW_REFREEZE_THRESHOLD) &&
7047 (refreeze_processes == FALSE)) {
7048 /*
7049 * We failed to freeze a process from the IDLE
7050 * band AND we have some thawed processes
7051 * AND haven't tried refreezing as yet.
7052 * Let's try and re-freeze processes in the
7053 * frozen band that have been resumed in the past
7054 * and so have brought in state from disk.
7055 */
7056
7057 band = (unsigned int) memorystatus_freeze_jetsam_band;
7058
7059 refreeze_processes = TRUE;
7060
7061 goto freeze_process;
7062 }
7063
7064 proc_list_unlock();
7065
7066 return ret;
7067 }
7068
7069 static inline boolean_t
7070 memorystatus_can_freeze_processes(void)
7071 {
7072 boolean_t ret;
7073
7074 proc_list_lock();
7075
7076 if (memorystatus_suspended_count) {
7077 memorystatus_freeze_suspended_threshold = MIN(memorystatus_freeze_suspended_threshold, FREEZE_SUSPENDED_THRESHOLD_DEFAULT);
7078
7079 if ((memorystatus_suspended_count - memorystatus_frozen_count) > memorystatus_freeze_suspended_threshold) {
7080 ret = TRUE;
7081 } else {
7082 ret = FALSE;
7083 }
7084 } else {
7085 ret = FALSE;
7086 }
7087
7088 proc_list_unlock();
7089
7090 return ret;
7091 }
7092
7093 static boolean_t
7094 memorystatus_can_freeze(boolean_t *memorystatus_freeze_swap_low)
7095 {
7096 boolean_t can_freeze = TRUE;
7097
7098 /* Only freeze if we're sufficiently low on memory; this holds off freeze right
7099 * after boot, and is generally is a no-op once we've reached steady state. */
7100 if (memorystatus_available_pages > memorystatus_freeze_threshold) {
7101 return FALSE;
7102 }
7103
7104 /* Check minimum suspended process threshold. */
7105 if (!memorystatus_can_freeze_processes()) {
7106 return FALSE;
7107 }
7108 assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
7109
7110 if (!VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
7111 /*
7112 * In-core compressor used for freezing WITHOUT on-disk swap support.
7113 */
7114 if (vm_compressor_low_on_space()) {
7115 if (*memorystatus_freeze_swap_low) {
7116 *memorystatus_freeze_swap_low = TRUE;
7117 }
7118
7119 can_freeze = FALSE;
7120 } else {
7121 if (*memorystatus_freeze_swap_low) {
7122 *memorystatus_freeze_swap_low = FALSE;
7123 }
7124
7125 can_freeze = TRUE;
7126 }
7127 } else {
7128 /*
7129 * Freezing WITH on-disk swap support.
7130 *
7131 * In-core compressor fronts the swap.
7132 */
7133 if (vm_swap_low_on_space()) {
7134 if (*memorystatus_freeze_swap_low) {
7135 *memorystatus_freeze_swap_low = TRUE;
7136 }
7137
7138 can_freeze = FALSE;
7139 }
7140 }
7141
7142 return can_freeze;
7143 }
7144
7145 /*
7146 * This function evaluates if the currently frozen processes deserve
7147 * to stay in the higher jetsam band. If the # of thaws of a process
7148 * is below our threshold, then we will demote that process into the IDLE
7149 * band and put it at the head. We don't immediately kill the process here
7150 * because it already has state on disk and so it might be worth giving
7151 * it another shot at getting thawed/resumed and used.
7152 */
7153 static void
7154 memorystatus_demote_frozen_processes(void)
7155 {
7156 unsigned int band = (unsigned int) memorystatus_freeze_jetsam_band;
7157 unsigned int demoted_proc_count = 0;
7158 proc_t p = PROC_NULL, next_p = PROC_NULL;
7159
7160 proc_list_lock();
7161
7162 if (memorystatus_freeze_enabled == FALSE) {
7163 /*
7164 * Freeze has been disabled likely to
7165 * reclaim swap space. So don't change
7166 * any state on the frozen processes.
7167 */
7168 proc_list_unlock();
7169 return;
7170 }
7171
7172 next_p = memorystatus_get_first_proc_locked(&band, FALSE);
7173 while (next_p) {
7174 p = next_p;
7175 next_p = memorystatus_get_next_proc_locked(&band, p, FALSE);
7176
7177 if ((p->p_memstat_state & P_MEMSTAT_FROZEN) == FALSE) {
7178 continue;
7179 }
7180
7181 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
7182 continue;
7183 }
7184
7185 if (p->p_memstat_thaw_count < memorystatus_thaw_count_demotion_threshold) {
7186 p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
7187 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
7188
7189 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, TRUE, TRUE);
7190 #if DEVELOPMENT || DEBUG
7191 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus_demote_frozen_process pid %d [%s]",
7192 p->p_pid, (*p->p_name ? p->p_name : "unknown"));
7193 #endif /* DEVELOPMENT || DEBUG */
7194
7195 /*
7196 * The freezer thread will consider this a normal app to be frozen
7197 * because it is in the IDLE band. So we don't need the
7198 * P_MEMSTAT_REFREEZE_ELIGIBLE state here. Also, if it gets resumed
7199 * we'll correctly count it as eligible for re-freeze again.
7200 *
7201 * We don't drop the frozen count because this process still has
7202 * state on disk. So there's a chance it gets resumed and then it
7203 * should land in the higher jetsam band. For that it needs to
7204 * remain marked frozen.
7205 */
7206 if (p->p_memstat_state & P_MEMSTAT_REFREEZE_ELIGIBLE) {
7207 p->p_memstat_state &= ~P_MEMSTAT_REFREEZE_ELIGIBLE;
7208 memorystatus_refreeze_eligible_count--;
7209 }
7210
7211 demoted_proc_count++;
7212 }
7213
7214 if (demoted_proc_count == memorystatus_max_frozen_demotions_daily) {
7215 break;
7216 }
7217 }
7218
7219 memorystatus_thaw_count = 0;
7220 proc_list_unlock();
7221 }
7222
7223
7224 /*
7225 * This function will do 4 things:
7226 *
7227 * 1) check to see if we are currently in a degraded freezer mode, and if so:
7228 * - check to see if our window has expired and we should exit this mode, OR,
7229 * - return a budget based on the degraded throttle window's max. pageouts vs current pageouts.
7230 *
7231 * 2) check to see if we are in a NEW normal window and update the normal throttle window's params.
7232 *
7233 * 3) check what the current normal window allows for a budget.
7234 *
7235 * 4) calculate the current rate of pageouts for DEGRADED_WINDOW_MINS duration. If that rate is below
7236 * what we would normally expect, then we are running low on our daily budget and need to enter
7237 * degraded perf. mode.
7238 */
7239
7240 static void
7241 memorystatus_freeze_update_throttle(uint64_t *budget_pages_allowed)
7242 {
7243 clock_sec_t sec;
7244 clock_nsec_t nsec;
7245 mach_timespec_t ts;
7246
7247 unsigned int freeze_daily_pageouts_max = 0;
7248
7249 #if DEVELOPMENT || DEBUG
7250 if (!memorystatus_freeze_throttle_enabled) {
7251 /*
7252 * No throttling...we can use the full budget everytime.
7253 */
7254 *budget_pages_allowed = UINT64_MAX;
7255 return;
7256 }
7257 #endif
7258
7259 clock_get_system_nanotime(&sec, &nsec);
7260 ts.tv_sec = sec;
7261 ts.tv_nsec = nsec;
7262
7263 struct throttle_interval_t *interval = NULL;
7264
7265 if (memorystatus_freeze_degradation == TRUE) {
7266 interval = degraded_throttle_window;
7267
7268 if (CMP_MACH_TIMESPEC(&ts, &interval->ts) >= 0) {
7269 memorystatus_freeze_degradation = FALSE;
7270 interval->pageouts = 0;
7271 interval->max_pageouts = 0;
7272 } else {
7273 *budget_pages_allowed = interval->max_pageouts - interval->pageouts;
7274 }
7275 }
7276
7277 interval = normal_throttle_window;
7278
7279 if (CMP_MACH_TIMESPEC(&ts, &interval->ts) >= 0) {
7280 /*
7281 * New throttle window.
7282 * Rollover any unused budget.
7283 * Also ask the storage layer what the new budget needs to be.
7284 */
7285 uint64_t freeze_daily_budget = 0;
7286 unsigned int daily_budget_pageouts = 0;
7287
7288 if (vm_swap_max_budget(&freeze_daily_budget)) {
7289 memorystatus_freeze_daily_mb_max = (freeze_daily_budget / (1024 * 1024));
7290 os_log_with_startup_serial(OS_LOG_DEFAULT, "memorystatus: memorystatus_freeze_daily_mb_max set to %dMB\n", memorystatus_freeze_daily_mb_max);
7291 }
7292
7293 freeze_daily_pageouts_max = memorystatus_freeze_daily_mb_max * (1024 * 1024 / PAGE_SIZE);
7294
7295 daily_budget_pageouts = (interval->burst_multiple * (((uint64_t)interval->mins * freeze_daily_pageouts_max) / NORMAL_WINDOW_MINS));
7296 interval->max_pageouts = (interval->max_pageouts - interval->pageouts) + daily_budget_pageouts;
7297
7298 interval->ts.tv_sec = interval->mins * 60;
7299 interval->ts.tv_nsec = 0;
7300 ADD_MACH_TIMESPEC(&interval->ts, &ts);
7301 /* Since we update the throttle stats pre-freeze, adjust for overshoot here */
7302 if (interval->pageouts > interval->max_pageouts) {
7303 interval->pageouts -= interval->max_pageouts;
7304 } else {
7305 interval->pageouts = 0;
7306 }
7307 *budget_pages_allowed = interval->max_pageouts;
7308
7309 memorystatus_demote_frozen_processes();
7310 } else {
7311 /*
7312 * Current throttle window.
7313 * Deny freezing if we have no budget left.
7314 * Try graceful degradation if we are within 25% of:
7315 * - the daily budget, and
7316 * - the current budget left is below our normal budget expectations.
7317 */
7318
7319 #if DEVELOPMENT || DEBUG
7320 /*
7321 * This can only happen in the INTERNAL configs because we allow modifying the daily budget for testing.
7322 */
7323
7324 if (freeze_daily_pageouts_max > interval->max_pageouts) {
7325 /*
7326 * We just bumped the daily budget. Re-evaluate our normal window params.
7327 */
7328 interval->max_pageouts = (interval->burst_multiple * (((uint64_t)interval->mins * freeze_daily_pageouts_max) / NORMAL_WINDOW_MINS));
7329 memorystatus_freeze_degradation = FALSE; //we'll re-evaluate this below...
7330 }
7331 #endif /* DEVELOPMENT || DEBUG */
7332
7333 if (memorystatus_freeze_degradation == FALSE) {
7334 if (interval->pageouts >= interval->max_pageouts) {
7335 *budget_pages_allowed = 0;
7336 } else {
7337 int budget_left = interval->max_pageouts - interval->pageouts;
7338 int budget_threshold = (freeze_daily_pageouts_max * FREEZE_DEGRADATION_BUDGET_THRESHOLD) / 100;
7339
7340 mach_timespec_t time_left = {0, 0};
7341
7342 time_left.tv_sec = interval->ts.tv_sec;
7343 time_left.tv_nsec = 0;
7344
7345 SUB_MACH_TIMESPEC(&time_left, &ts);
7346
7347 if (budget_left <= budget_threshold) {
7348 /*
7349 * For the current normal window, calculate how much we would pageout in a DEGRADED_WINDOW_MINS duration.
7350 * And also calculate what we would pageout for the same DEGRADED_WINDOW_MINS duration if we had the full
7351 * daily pageout budget.
7352 */
7353
7354 unsigned int current_budget_rate_allowed = ((budget_left / time_left.tv_sec) / 60) * DEGRADED_WINDOW_MINS;
7355 unsigned int normal_budget_rate_allowed = (freeze_daily_pageouts_max / NORMAL_WINDOW_MINS) * DEGRADED_WINDOW_MINS;
7356
7357 /*
7358 * The current rate of pageouts is below what we would expect for
7359 * the normal rate i.e. we have below normal budget left and so...
7360 */
7361
7362 if (current_budget_rate_allowed < normal_budget_rate_allowed) {
7363 memorystatus_freeze_degradation = TRUE;
7364 degraded_throttle_window->max_pageouts = current_budget_rate_allowed;
7365 degraded_throttle_window->pageouts = 0;
7366
7367 /*
7368 * Switch over to the degraded throttle window so the budget
7369 * doled out is based on that window.
7370 */
7371 interval = degraded_throttle_window;
7372 }
7373 }
7374
7375 *budget_pages_allowed = interval->max_pageouts - interval->pageouts;
7376 }
7377 }
7378 }
7379
7380 MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_update_throttle_interval: throttle updated - %d frozen (%d max) within %dm; %dm remaining; throttle %s\n",
7381 interval->pageouts, interval->max_pageouts, interval->mins, (interval->ts.tv_sec - ts->tv_sec) / 60,
7382 interval->throttle ? "on" : "off");
7383 }
7384
7385 static void
7386 memorystatus_freeze_thread(void *param __unused, wait_result_t wr __unused)
7387 {
7388 static boolean_t memorystatus_freeze_swap_low = FALSE;
7389
7390 lck_mtx_lock(&freezer_mutex);
7391
7392 if (memorystatus_freeze_enabled) {
7393 if ((memorystatus_frozen_count < memorystatus_frozen_processes_max) ||
7394 (memorystatus_refreeze_eligible_count >= MIN_THAW_REFREEZE_THRESHOLD)) {
7395 if (memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
7396 /* Only freeze if we've not exceeded our pageout budgets.*/
7397 memorystatus_freeze_update_throttle(&memorystatus_freeze_budget_pages_remaining);
7398
7399 if (memorystatus_freeze_budget_pages_remaining) {
7400 memorystatus_freeze_top_process();
7401 }
7402 }
7403 }
7404 }
7405
7406 /*
7407 * We use memorystatus_apps_idle_delay_time because if/when we adopt aging for applications,
7408 * it'll tie neatly into running the freezer once we age an application.
7409 *
7410 * Till then, it serves as a good interval that can be tuned via a sysctl too.
7411 */
7412 memorystatus_freezer_thread_next_run_ts = mach_absolute_time() + memorystatus_apps_idle_delay_time;
7413
7414 assert_wait((event_t) &memorystatus_freeze_wakeup, THREAD_UNINT);
7415 lck_mtx_unlock(&freezer_mutex);
7416
7417 thread_block((thread_continue_t) memorystatus_freeze_thread);
7418 }
7419
7420 static boolean_t
7421 memorystatus_freeze_thread_should_run(void)
7422 {
7423 /*
7424 * No freezer_mutex held here...see why near call-site
7425 * within memorystatus_pages_update().
7426 */
7427
7428 boolean_t should_run = FALSE;
7429
7430 if (memorystatus_freeze_enabled == FALSE) {
7431 goto out;
7432 }
7433
7434 if (memorystatus_available_pages > memorystatus_freeze_threshold) {
7435 goto out;
7436 }
7437
7438 if ((memorystatus_frozen_count >= memorystatus_frozen_processes_max) &&
7439 (memorystatus_refreeze_eligible_count < MIN_THAW_REFREEZE_THRESHOLD)) {
7440 goto out;
7441 }
7442
7443 if (memorystatus_frozen_shared_mb_max && (memorystatus_frozen_shared_mb >= memorystatus_frozen_shared_mb_max)) {
7444 goto out;
7445 }
7446
7447 uint64_t curr_time = mach_absolute_time();
7448
7449 if (curr_time < memorystatus_freezer_thread_next_run_ts) {
7450 goto out;
7451 }
7452
7453 should_run = TRUE;
7454
7455 out:
7456 return should_run;
7457 }
7458
7459 static int
7460 sysctl_memorystatus_do_fastwake_warmup_all SYSCTL_HANDLER_ARGS
7461 {
7462 #pragma unused(oidp, req, arg1, arg2)
7463
7464 /* Need to be root or have entitlement */
7465 if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT)) {
7466 return EPERM;
7467 }
7468
7469 if (memorystatus_freeze_enabled == FALSE) {
7470 return ENOTSUP;
7471 }
7472
7473 do_fastwake_warmup_all();
7474
7475 return 0;
7476 }
7477
7478 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_do_fastwake_warmup_all, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
7479 0, 0, &sysctl_memorystatus_do_fastwake_warmup_all, "I", "");
7480
7481 #endif /* CONFIG_FREEZE */
7482
7483 #if VM_PRESSURE_EVENTS
7484
7485 #if CONFIG_MEMORYSTATUS
7486
7487 static int
7488 memorystatus_send_note(int event_code, void *data, size_t data_length)
7489 {
7490 int ret;
7491 struct kev_msg ev_msg;
7492
7493 ev_msg.vendor_code = KEV_VENDOR_APPLE;
7494 ev_msg.kev_class = KEV_SYSTEM_CLASS;
7495 ev_msg.kev_subclass = KEV_MEMORYSTATUS_SUBCLASS;
7496
7497 ev_msg.event_code = event_code;
7498
7499 ev_msg.dv[0].data_length = data_length;
7500 ev_msg.dv[0].data_ptr = data;
7501 ev_msg.dv[1].data_length = 0;
7502
7503 ret = kev_post_msg(&ev_msg);
7504 if (ret) {
7505 printf("%s: kev_post_msg() failed, err %d\n", __func__, ret);
7506 }
7507
7508 return ret;
7509 }
7510
7511 boolean_t
7512 memorystatus_warn_process(pid_t pid, __unused boolean_t is_active, __unused boolean_t is_fatal, boolean_t limit_exceeded)
7513 {
7514 boolean_t ret = FALSE;
7515 boolean_t found_knote = FALSE;
7516 struct knote *kn = NULL;
7517 int send_knote_count = 0;
7518
7519 /*
7520 * See comment in sysctl_memorystatus_vm_pressure_send.
7521 */
7522
7523 memorystatus_klist_lock();
7524
7525 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
7526 proc_t knote_proc = knote_get_kq(kn)->kq_p;
7527 pid_t knote_pid = knote_proc->p_pid;
7528
7529 if (knote_pid == pid) {
7530 /*
7531 * By setting the "fflags" here, we are forcing
7532 * a process to deal with the case where it's
7533 * bumping up into its memory limits. If we don't
7534 * do this here, we will end up depending on the
7535 * system pressure snapshot evaluation in
7536 * filt_memorystatus().
7537 */
7538
7539 #if CONFIG_EMBEDDED
7540 if (!limit_exceeded) {
7541 /*
7542 * Intentionally set either the unambiguous limit warning,
7543 * the system-wide critical or the system-wide warning
7544 * notification bit.
7545 */
7546
7547 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
7548 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
7549 found_knote = TRUE;
7550 send_knote_count++;
7551 } else if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
7552 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
7553 found_knote = TRUE;
7554 send_knote_count++;
7555 } else if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
7556 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
7557 found_knote = TRUE;
7558 send_knote_count++;
7559 }
7560 } else {
7561 /*
7562 * Send this notification when a process has exceeded a soft limit.
7563 */
7564 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
7565 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
7566 found_knote = TRUE;
7567 send_knote_count++;
7568 }
7569 }
7570 #else /* CONFIG_EMBEDDED */
7571 if (!limit_exceeded) {
7572 /*
7573 * Processes on desktop are not expecting to handle a system-wide
7574 * critical or system-wide warning notification from this path.
7575 * Intentionally set only the unambiguous limit warning here.
7576 *
7577 * If the limit is soft, however, limit this to one notification per
7578 * active/inactive limit (per each registered listener).
7579 */
7580
7581 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
7582 found_knote = TRUE;
7583 if (!is_fatal) {
7584 /*
7585 * Restrict proc_limit_warn notifications when
7586 * non-fatal (soft) limit is at play.
7587 */
7588 if (is_active) {
7589 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE) {
7590 /*
7591 * Mark this knote for delivery.
7592 */
7593 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
7594 /*
7595 * And suppress it from future notifications.
7596 */
7597 kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
7598 send_knote_count++;
7599 }
7600 } else {
7601 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE) {
7602 /*
7603 * Mark this knote for delivery.
7604 */
7605 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
7606 /*
7607 * And suppress it from future notifications.
7608 */
7609 kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
7610 send_knote_count++;
7611 }
7612 }
7613 } else {
7614 /*
7615 * No restriction on proc_limit_warn notifications when
7616 * fatal (hard) limit is at play.
7617 */
7618 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
7619 send_knote_count++;
7620 }
7621 }
7622 } else {
7623 /*
7624 * Send this notification when a process has exceeded a soft limit,
7625 */
7626
7627 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
7628 found_knote = TRUE;
7629 if (!is_fatal) {
7630 /*
7631 * Restrict critical notifications for soft limits.
7632 */
7633
7634 if (is_active) {
7635 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE) {
7636 /*
7637 * Suppress future proc_limit_critical notifications
7638 * for the active soft limit.
7639 */
7640 kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
7641 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
7642 send_knote_count++;
7643 }
7644 } else {
7645 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE) {
7646 /*
7647 * Suppress future proc_limit_critical_notifications
7648 * for the inactive soft limit.
7649 */
7650 kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
7651 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
7652 send_knote_count++;
7653 }
7654 }
7655 } else {
7656 /*
7657 * We should never be trying to send a critical notification for
7658 * a hard limit... the process would be killed before it could be
7659 * received.
7660 */
7661 panic("Caught sending pid %d a critical warning for a fatal limit.\n", pid);
7662 }
7663 }
7664 }
7665 #endif /* CONFIG_EMBEDDED */
7666 }
7667 }
7668
7669 if (found_knote) {
7670 if (send_knote_count > 0) {
7671 KNOTE(&memorystatus_klist, 0);
7672 }
7673 ret = TRUE;
7674 }
7675
7676 memorystatus_klist_unlock();
7677
7678 return ret;
7679 }
7680
7681 /*
7682 * Can only be set by the current task on itself.
7683 */
7684 int
7685 memorystatus_low_mem_privileged_listener(uint32_t op_flags)
7686 {
7687 boolean_t set_privilege = FALSE;
7688 /*
7689 * Need an entitlement check here?
7690 */
7691 if (op_flags == MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE) {
7692 set_privilege = TRUE;
7693 } else if (op_flags == MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE) {
7694 set_privilege = FALSE;
7695 } else {
7696 return EINVAL;
7697 }
7698
7699 return task_low_mem_privileged_listener(current_task(), set_privilege, NULL);
7700 }
7701
7702 int
7703 memorystatus_send_pressure_note(pid_t pid)
7704 {
7705 MEMORYSTATUS_DEBUG(1, "memorystatus_send_pressure_note(): pid %d\n", pid);
7706 return memorystatus_send_note(kMemorystatusPressureNote, &pid, sizeof(pid));
7707 }
7708
7709 void
7710 memorystatus_send_low_swap_note(void)
7711 {
7712 struct knote *kn = NULL;
7713
7714 memorystatus_klist_lock();
7715 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
7716 /* We call is_knote_registered_modify_task_pressure_bits to check if the sfflags for the
7717 * current note contain NOTE_MEMORYSTATUS_LOW_SWAP. Once we find one note in the memorystatus_klist
7718 * that has the NOTE_MEMORYSTATUS_LOW_SWAP flags in its sfflags set, we call KNOTE with
7719 * kMemoryStatusLowSwap as the hint to process and update all knotes on the memorystatus_klist accordingly. */
7720 if (is_knote_registered_modify_task_pressure_bits(kn, NOTE_MEMORYSTATUS_LOW_SWAP, NULL, 0, 0) == TRUE) {
7721 KNOTE(&memorystatus_klist, kMemorystatusLowSwap);
7722 break;
7723 }
7724 }
7725
7726 memorystatus_klist_unlock();
7727 }
7728
7729 boolean_t
7730 memorystatus_bg_pressure_eligible(proc_t p)
7731 {
7732 boolean_t eligible = FALSE;
7733
7734 proc_list_lock();
7735
7736 MEMORYSTATUS_DEBUG(1, "memorystatus_bg_pressure_eligible: pid %d, state 0x%x\n", p->p_pid, p->p_memstat_state);
7737
7738 /* Foreground processes have already been dealt with at this point, so just test for eligibility */
7739 if (!(p->p_memstat_state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_SUSPENDED | P_MEMSTAT_FROZEN))) {
7740 eligible = TRUE;
7741 }
7742
7743 if (p->p_memstat_effectivepriority < JETSAM_PRIORITY_BACKGROUND_OPPORTUNISTIC) {
7744 /*
7745 * IDLE and IDLE_DEFERRED bands contain processes
7746 * that have dropped memory to be under their inactive
7747 * memory limits. And so they can't really give back
7748 * anything.
7749 */
7750 eligible = FALSE;
7751 }
7752
7753 proc_list_unlock();
7754
7755 return eligible;
7756 }
7757
7758 boolean_t
7759 memorystatus_is_foreground_locked(proc_t p)
7760 {
7761 return (p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND) ||
7762 (p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND_SUPPORT);
7763 }
7764
7765 /*
7766 * This is meant for stackshot and kperf -- it does not take the proc_list_lock
7767 * to access the p_memstat_dirty field.
7768 */
7769 void
7770 memorystatus_proc_flags_unsafe(void * v, boolean_t *is_dirty, boolean_t *is_dirty_tracked, boolean_t *allow_idle_exit)
7771 {
7772 if (!v) {
7773 *is_dirty = FALSE;
7774 *is_dirty_tracked = FALSE;
7775 *allow_idle_exit = FALSE;
7776 } else {
7777 proc_t p = (proc_t)v;
7778 *is_dirty = (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) != 0;
7779 *is_dirty_tracked = (p->p_memstat_dirty & P_DIRTY_TRACK) != 0;
7780 *allow_idle_exit = (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) != 0;
7781 }
7782 }
7783
7784 #endif /* CONFIG_MEMORYSTATUS */
7785
7786 /*
7787 * Trigger levels to test the mechanism.
7788 * Can be used via a sysctl.
7789 */
7790 #define TEST_LOW_MEMORY_TRIGGER_ONE 1
7791 #define TEST_LOW_MEMORY_TRIGGER_ALL 2
7792 #define TEST_PURGEABLE_TRIGGER_ONE 3
7793 #define TEST_PURGEABLE_TRIGGER_ALL 4
7794 #define TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ONE 5
7795 #define TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL 6
7796
7797 boolean_t memorystatus_manual_testing_on = FALSE;
7798 vm_pressure_level_t memorystatus_manual_testing_level = kVMPressureNormal;
7799
7800 extern struct knote *
7801 vm_pressure_select_optimal_candidate_to_notify(struct klist *, int, boolean_t);
7802
7803
7804 #define VM_PRESSURE_NOTIFY_WAIT_PERIOD 10000 /* milliseconds */
7805
7806 #if DEBUG
7807 #define VM_PRESSURE_DEBUG(cond, format, ...) \
7808 do { \
7809 if (cond) { printf(format, ##__VA_ARGS__); } \
7810 } while(0)
7811 #else
7812 #define VM_PRESSURE_DEBUG(cond, format, ...)
7813 #endif
7814
7815 #define INTER_NOTIFICATION_DELAY (250000) /* .25 second */
7816
7817 void
7818 memorystatus_on_pageout_scan_end(void)
7819 {
7820 /* No-op */
7821 }
7822
7823 /*
7824 * kn_max - knote
7825 *
7826 * knote_pressure_level - to check if the knote is registered for this notification level.
7827 *
7828 * task - task whose bits we'll be modifying
7829 *
7830 * pressure_level_to_clear - if the task has been notified of this past level, clear that notification bit so that if/when we revert to that level, the task will be notified again.
7831 *
7832 * pressure_level_to_set - the task is about to be notified of this new level. Update the task's bit notification information appropriately.
7833 *
7834 */
7835
7836 boolean_t
7837 is_knote_registered_modify_task_pressure_bits(struct knote *kn_max, int knote_pressure_level, task_t task, vm_pressure_level_t pressure_level_to_clear, vm_pressure_level_t pressure_level_to_set)
7838 {
7839 if (kn_max->kn_sfflags & knote_pressure_level) {
7840 if (pressure_level_to_clear && task_has_been_notified(task, pressure_level_to_clear) == TRUE) {
7841 task_clear_has_been_notified(task, pressure_level_to_clear);
7842 }
7843
7844 task_mark_has_been_notified(task, pressure_level_to_set);
7845 return TRUE;
7846 }
7847
7848 return FALSE;
7849 }
7850
7851 void
7852 memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear)
7853 {
7854 struct knote *kn = NULL;
7855
7856 memorystatus_klist_lock();
7857 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
7858 proc_t p = PROC_NULL;
7859 struct task* t = TASK_NULL;
7860
7861 p = knote_get_kq(kn)->kq_p;
7862 proc_list_lock();
7863 if (p != proc_ref_locked(p)) {
7864 p = PROC_NULL;
7865 proc_list_unlock();
7866 continue;
7867 }
7868 proc_list_unlock();
7869
7870 t = (struct task *)(p->task);
7871
7872 task_clear_has_been_notified(t, pressure_level_to_clear);
7873
7874 proc_rele(p);
7875 }
7876
7877 memorystatus_klist_unlock();
7878 }
7879
7880 extern kern_return_t vm_pressure_notify_dispatch_vm_clients(boolean_t target_foreground_process);
7881
7882 struct knote *
7883 vm_pressure_select_optimal_candidate_to_notify(struct klist *candidate_list, int level, boolean_t target_foreground_process);
7884
7885 /*
7886 * Used by the vm_pressure_thread which is
7887 * signalled from within vm_pageout_scan().
7888 */
7889 static void vm_dispatch_memory_pressure(void);
7890 void consider_vm_pressure_events(void);
7891
7892 void
7893 consider_vm_pressure_events(void)
7894 {
7895 vm_dispatch_memory_pressure();
7896 }
7897 static void
7898 vm_dispatch_memory_pressure(void)
7899 {
7900 memorystatus_update_vm_pressure(FALSE);
7901 }
7902
7903 extern vm_pressure_level_t
7904 convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
7905
7906 struct knote *
7907 vm_pressure_select_optimal_candidate_to_notify(struct klist *candidate_list, int level, boolean_t target_foreground_process)
7908 {
7909 struct knote *kn = NULL, *kn_max = NULL;
7910 uint64_t resident_max = 0; /* MB */
7911 struct timeval curr_tstamp = {0, 0};
7912 int elapsed_msecs = 0;
7913 int selected_task_importance = 0;
7914 static int pressure_snapshot = -1;
7915 boolean_t pressure_increase = FALSE;
7916
7917 if (pressure_snapshot == -1) {
7918 /*
7919 * Initial snapshot.
7920 */
7921 pressure_snapshot = level;
7922 pressure_increase = TRUE;
7923 } else {
7924 if (level && (level >= pressure_snapshot)) {
7925 pressure_increase = TRUE;
7926 } else {
7927 pressure_increase = FALSE;
7928 }
7929
7930 pressure_snapshot = level;
7931 }
7932
7933 if (pressure_increase == TRUE) {
7934 /*
7935 * We'll start by considering the largest
7936 * unimportant task in our list.
7937 */
7938 selected_task_importance = INT_MAX;
7939 } else {
7940 /*
7941 * We'll start by considering the largest
7942 * important task in our list.
7943 */
7944 selected_task_importance = 0;
7945 }
7946
7947 microuptime(&curr_tstamp);
7948
7949 SLIST_FOREACH(kn, candidate_list, kn_selnext) {
7950 uint64_t resident_size = 0; /* MB */
7951 proc_t p = PROC_NULL;
7952 struct task* t = TASK_NULL;
7953 int curr_task_importance = 0;
7954 boolean_t consider_knote = FALSE;
7955 boolean_t privileged_listener = FALSE;
7956
7957 p = knote_get_kq(kn)->kq_p;
7958 proc_list_lock();
7959 if (p != proc_ref_locked(p)) {
7960 p = PROC_NULL;
7961 proc_list_unlock();
7962 continue;
7963 }
7964 proc_list_unlock();
7965
7966 #if CONFIG_MEMORYSTATUS
7967 if (target_foreground_process == TRUE && !memorystatus_is_foreground_locked(p)) {
7968 /*
7969 * Skip process not marked foreground.
7970 */
7971 proc_rele(p);
7972 continue;
7973 }
7974 #endif /* CONFIG_MEMORYSTATUS */
7975
7976 t = (struct task *)(p->task);
7977
7978 timevalsub(&curr_tstamp, &p->vm_pressure_last_notify_tstamp);
7979 elapsed_msecs = curr_tstamp.tv_sec * 1000 + curr_tstamp.tv_usec / 1000;
7980
7981 vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(level);
7982
7983 if ((kn->kn_sfflags & dispatch_level) == 0) {
7984 proc_rele(p);
7985 continue;
7986 }
7987
7988 #if CONFIG_MEMORYSTATUS
7989 if (target_foreground_process == FALSE && !memorystatus_bg_pressure_eligible(p)) {
7990 VM_PRESSURE_DEBUG(1, "[vm_pressure] skipping process %d\n", p->p_pid);
7991 proc_rele(p);
7992 continue;
7993 }
7994 #endif /* CONFIG_MEMORYSTATUS */
7995
7996 #if CONFIG_EMBEDDED
7997 curr_task_importance = p->p_memstat_effectivepriority;
7998 #else /* CONFIG_EMBEDDED */
7999 curr_task_importance = task_importance_estimate(t);
8000 #endif /* CONFIG_EMBEDDED */
8001
8002 /*
8003 * Privileged listeners are only considered in the multi-level pressure scheme
8004 * AND only if the pressure is increasing.
8005 */
8006 if (level > 0) {
8007 if (task_has_been_notified(t, level) == FALSE) {
8008 /*
8009 * Is this a privileged listener?
8010 */
8011 if (task_low_mem_privileged_listener(t, FALSE, &privileged_listener) == 0) {
8012 if (privileged_listener) {
8013 kn_max = kn;
8014 proc_rele(p);
8015 goto done_scanning;
8016 }
8017 }
8018 } else {
8019 proc_rele(p);
8020 continue;
8021 }
8022 } else if (level == 0) {
8023 /*
8024 * Task wasn't notified when the pressure was increasing and so
8025 * no need to notify it that the pressure is decreasing.
8026 */
8027 if ((task_has_been_notified(t, kVMPressureWarning) == FALSE) && (task_has_been_notified(t, kVMPressureCritical) == FALSE)) {
8028 proc_rele(p);
8029 continue;
8030 }
8031 }
8032
8033 /*
8034 * We don't want a small process to block large processes from
8035 * being notified again. <rdar://problem/7955532>
8036 */
8037 resident_size = (get_task_phys_footprint(t)) / (1024 * 1024ULL); /* MB */
8038
8039 if (resident_size >= vm_pressure_task_footprint_min) {
8040 if (level > 0) {
8041 /*
8042 * Warning or Critical Pressure.
8043 */
8044 if (pressure_increase) {
8045 if ((curr_task_importance < selected_task_importance) ||
8046 ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
8047 /*
8048 * We have found a candidate process which is:
8049 * a) at a lower importance than the current selected process
8050 * OR
8051 * b) has importance equal to that of the current selected process but is larger
8052 */
8053
8054 consider_knote = TRUE;
8055 }
8056 } else {
8057 if ((curr_task_importance > selected_task_importance) ||
8058 ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
8059 /*
8060 * We have found a candidate process which is:
8061 * a) at a higher importance than the current selected process
8062 * OR
8063 * b) has importance equal to that of the current selected process but is larger
8064 */
8065
8066 consider_knote = TRUE;
8067 }
8068 }
8069 } else if (level == 0) {
8070 /*
8071 * Pressure back to normal.
8072 */
8073 if ((curr_task_importance > selected_task_importance) ||
8074 ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
8075 consider_knote = TRUE;
8076 }
8077 }
8078
8079 if (consider_knote) {
8080 resident_max = resident_size;
8081 kn_max = kn;
8082 selected_task_importance = curr_task_importance;
8083 consider_knote = FALSE; /* reset for the next candidate */
8084 }
8085 } else {
8086 /* There was no candidate with enough resident memory to scavenge */
8087 VM_PRESSURE_DEBUG(0, "[vm_pressure] threshold failed for pid %d with %llu resident...\n", p->p_pid, resident_size);
8088 }
8089 proc_rele(p);
8090 }
8091
8092 done_scanning:
8093 if (kn_max) {
8094 VM_DEBUG_CONSTANT_EVENT(vm_pressure_event, VM_PRESSURE_EVENT, DBG_FUNC_NONE, knote_get_kq(kn_max)->kq_p->p_pid, resident_max, 0, 0);
8095 VM_PRESSURE_DEBUG(1, "[vm_pressure] sending event to pid %d with %llu resident\n", knote_get_kq(kn_max)->kq_p->p_pid, resident_max);
8096 }
8097
8098 return kn_max;
8099 }
8100
8101 #define VM_PRESSURE_DECREASED_SMOOTHING_PERIOD 5000 /* milliseconds */
8102 #define WARNING_NOTIFICATION_RESTING_PERIOD 25 /* seconds */
8103 #define CRITICAL_NOTIFICATION_RESTING_PERIOD 25 /* seconds */
8104
8105 uint64_t next_warning_notification_sent_at_ts = 0;
8106 uint64_t next_critical_notification_sent_at_ts = 0;
8107
8108 kern_return_t
8109 memorystatus_update_vm_pressure(boolean_t target_foreground_process)
8110 {
8111 struct knote *kn_max = NULL;
8112 struct knote *kn_cur = NULL, *kn_temp = NULL; /* for safe list traversal */
8113 pid_t target_pid = -1;
8114 struct klist dispatch_klist = { NULL };
8115 proc_t target_proc = PROC_NULL;
8116 struct task *task = NULL;
8117 boolean_t found_candidate = FALSE;
8118
8119 static vm_pressure_level_t level_snapshot = kVMPressureNormal;
8120 static vm_pressure_level_t prev_level_snapshot = kVMPressureNormal;
8121 boolean_t smoothing_window_started = FALSE;
8122 struct timeval smoothing_window_start_tstamp = {0, 0};
8123 struct timeval curr_tstamp = {0, 0};
8124 int elapsed_msecs = 0;
8125 uint64_t curr_ts = mach_absolute_time();
8126
8127 #if !CONFIG_JETSAM
8128 #define MAX_IDLE_KILLS 100 /* limit the number of idle kills allowed */
8129
8130 int idle_kill_counter = 0;
8131
8132 /*
8133 * On desktop we take this opportunity to free up memory pressure
8134 * by immediately killing idle exitable processes. We use a delay
8135 * to avoid overkill. And we impose a max counter as a fail safe
8136 * in case daemons re-launch too fast.
8137 */
8138 while ((memorystatus_vm_pressure_level != kVMPressureNormal) && (idle_kill_counter < MAX_IDLE_KILLS)) {
8139 if (memorystatus_idle_exit_from_VM() == FALSE) {
8140 /* No idle exitable processes left to kill */
8141 break;
8142 }
8143 idle_kill_counter++;
8144
8145 if (memorystatus_manual_testing_on == TRUE) {
8146 /*
8147 * Skip the delay when testing
8148 * the pressure notification scheme.
8149 */
8150 } else {
8151 delay(1000000); /* 1 second */
8152 }
8153 }
8154 #endif /* !CONFIG_JETSAM */
8155
8156 if (level_snapshot != kVMPressureNormal) {
8157 /*
8158 * Check to see if we are still in the 'resting' period
8159 * after having notified all clients interested in
8160 * a particular pressure level.
8161 */
8162
8163 level_snapshot = memorystatus_vm_pressure_level;
8164
8165 if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
8166 if (next_warning_notification_sent_at_ts) {
8167 if (curr_ts < next_warning_notification_sent_at_ts) {
8168 delay(INTER_NOTIFICATION_DELAY * 4 /* 1 sec */);
8169 return KERN_SUCCESS;
8170 }
8171
8172 next_warning_notification_sent_at_ts = 0;
8173 memorystatus_klist_reset_all_for_level(kVMPressureWarning);
8174 }
8175 } else if (level_snapshot == kVMPressureCritical) {
8176 if (next_critical_notification_sent_at_ts) {
8177 if (curr_ts < next_critical_notification_sent_at_ts) {
8178 delay(INTER_NOTIFICATION_DELAY * 4 /* 1 sec */);
8179 return KERN_SUCCESS;
8180 }
8181 next_critical_notification_sent_at_ts = 0;
8182 memorystatus_klist_reset_all_for_level(kVMPressureCritical);
8183 }
8184 }
8185 }
8186
8187 while (1) {
8188 /*
8189 * There is a race window here. But it's not clear
8190 * how much we benefit from having extra synchronization.
8191 */
8192 level_snapshot = memorystatus_vm_pressure_level;
8193
8194 if (prev_level_snapshot > level_snapshot) {
8195 /*
8196 * Pressure decreased? Let's take a little breather
8197 * and see if this condition stays.
8198 */
8199 if (smoothing_window_started == FALSE) {
8200 smoothing_window_started = TRUE;
8201 microuptime(&smoothing_window_start_tstamp);
8202 }
8203
8204 microuptime(&curr_tstamp);
8205 timevalsub(&curr_tstamp, &smoothing_window_start_tstamp);
8206 elapsed_msecs = curr_tstamp.tv_sec * 1000 + curr_tstamp.tv_usec / 1000;
8207
8208 if (elapsed_msecs < VM_PRESSURE_DECREASED_SMOOTHING_PERIOD) {
8209 delay(INTER_NOTIFICATION_DELAY);
8210 continue;
8211 }
8212 }
8213
8214 prev_level_snapshot = level_snapshot;
8215 smoothing_window_started = FALSE;
8216
8217 memorystatus_klist_lock();
8218 kn_max = vm_pressure_select_optimal_candidate_to_notify(&memorystatus_klist, level_snapshot, target_foreground_process);
8219
8220 if (kn_max == NULL) {
8221 memorystatus_klist_unlock();
8222
8223 /*
8224 * No more level-based clients to notify.
8225 *
8226 * Start the 'resting' window within which clients will not be re-notified.
8227 */
8228
8229 if (level_snapshot != kVMPressureNormal) {
8230 if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
8231 nanoseconds_to_absolutetime(WARNING_NOTIFICATION_RESTING_PERIOD * NSEC_PER_SEC, &curr_ts);
8232
8233 /* Next warning notification (if nothing changes) won't be sent before...*/
8234 next_warning_notification_sent_at_ts = mach_absolute_time() + curr_ts;
8235 }
8236
8237 if (level_snapshot == kVMPressureCritical) {
8238 nanoseconds_to_absolutetime(CRITICAL_NOTIFICATION_RESTING_PERIOD * NSEC_PER_SEC, &curr_ts);
8239
8240 /* Next critical notification (if nothing changes) won't be sent before...*/
8241 next_critical_notification_sent_at_ts = mach_absolute_time() + curr_ts;
8242 }
8243 }
8244 return KERN_FAILURE;
8245 }
8246
8247 target_proc = knote_get_kq(kn_max)->kq_p;
8248
8249 proc_list_lock();
8250 if (target_proc != proc_ref_locked(target_proc)) {
8251 target_proc = PROC_NULL;
8252 proc_list_unlock();
8253 memorystatus_klist_unlock();
8254 continue;
8255 }
8256 proc_list_unlock();
8257
8258 target_pid = target_proc->p_pid;
8259
8260 task = (struct task *)(target_proc->task);
8261
8262 if (level_snapshot != kVMPressureNormal) {
8263 if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
8264 if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_WARN, task, 0, kVMPressureWarning) == TRUE) {
8265 found_candidate = TRUE;
8266 }
8267 } else {
8268 if (level_snapshot == kVMPressureCritical) {
8269 if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_CRITICAL, task, 0, kVMPressureCritical) == TRUE) {
8270 found_candidate = TRUE;
8271 }
8272 }
8273 }
8274 } else {
8275 if (kn_max->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
8276 task_clear_has_been_notified(task, kVMPressureWarning);
8277 task_clear_has_been_notified(task, kVMPressureCritical);
8278
8279 found_candidate = TRUE;
8280 }
8281 }
8282
8283 if (found_candidate == FALSE) {
8284 proc_rele(target_proc);
8285 memorystatus_klist_unlock();
8286 continue;
8287 }
8288
8289 SLIST_FOREACH_SAFE(kn_cur, &memorystatus_klist, kn_selnext, kn_temp) {
8290 int knote_pressure_level = convert_internal_pressure_level_to_dispatch_level(level_snapshot);
8291
8292 if (is_knote_registered_modify_task_pressure_bits(kn_cur, knote_pressure_level, task, 0, level_snapshot) == TRUE) {
8293 proc_t knote_proc = knote_get_kq(kn_cur)->kq_p;
8294 pid_t knote_pid = knote_proc->p_pid;
8295 if (knote_pid == target_pid) {
8296 KNOTE_DETACH(&memorystatus_klist, kn_cur);
8297 KNOTE_ATTACH(&dispatch_klist, kn_cur);
8298 }
8299 }
8300 }
8301
8302 KNOTE(&dispatch_klist, (level_snapshot != kVMPressureNormal) ? kMemorystatusPressure : kMemorystatusNoPressure);
8303
8304 SLIST_FOREACH_SAFE(kn_cur, &dispatch_klist, kn_selnext, kn_temp) {
8305 KNOTE_DETACH(&dispatch_klist, kn_cur);
8306 KNOTE_ATTACH(&memorystatus_klist, kn_cur);
8307 }
8308
8309 memorystatus_klist_unlock();
8310
8311 microuptime(&target_proc->vm_pressure_last_notify_tstamp);
8312 proc_rele(target_proc);
8313
8314 if (memorystatus_manual_testing_on == TRUE && target_foreground_process == TRUE) {
8315 break;
8316 }
8317
8318 if (memorystatus_manual_testing_on == TRUE) {
8319 /*
8320 * Testing out the pressure notification scheme.
8321 * No need for delays etc.
8322 */
8323 } else {
8324 uint32_t sleep_interval = INTER_NOTIFICATION_DELAY;
8325 #if CONFIG_JETSAM
8326 unsigned int page_delta = 0;
8327 unsigned int skip_delay_page_threshold = 0;
8328
8329 assert(memorystatus_available_pages_pressure >= memorystatus_available_pages_critical_base);
8330
8331 page_delta = (memorystatus_available_pages_pressure - memorystatus_available_pages_critical_base) / 2;
8332 skip_delay_page_threshold = memorystatus_available_pages_pressure - page_delta;
8333
8334 if (memorystatus_available_pages <= skip_delay_page_threshold) {
8335 /*
8336 * We are nearing the critcal mark fast and can't afford to wait between
8337 * notifications.
8338 */
8339 sleep_interval = 0;
8340 }
8341 #endif /* CONFIG_JETSAM */
8342
8343 if (sleep_interval) {
8344 delay(sleep_interval);
8345 }
8346 }
8347 }
8348
8349 return KERN_SUCCESS;
8350 }
8351
8352 vm_pressure_level_t
8353 convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t internal_pressure_level)
8354 {
8355 vm_pressure_level_t dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
8356
8357 switch (internal_pressure_level) {
8358 case kVMPressureNormal:
8359 {
8360 dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
8361 break;
8362 }
8363
8364 case kVMPressureWarning:
8365 case kVMPressureUrgent:
8366 {
8367 dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_WARN;
8368 break;
8369 }
8370
8371 case kVMPressureCritical:
8372 {
8373 dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
8374 break;
8375 }
8376
8377 default:
8378 break;
8379 }
8380
8381 return dispatch_level;
8382 }
8383
8384 static int
8385 sysctl_memorystatus_vm_pressure_level SYSCTL_HANDLER_ARGS
8386 {
8387 #pragma unused(arg1, arg2, oidp)
8388 #if CONFIG_EMBEDDED
8389 int error = 0;
8390
8391 error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
8392 if (error) {
8393 return error;
8394 }
8395
8396 #endif /* CONFIG_EMBEDDED */
8397 vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(memorystatus_vm_pressure_level);
8398
8399 return SYSCTL_OUT(req, &dispatch_level, sizeof(dispatch_level));
8400 }
8401
8402 #if DEBUG || DEVELOPMENT
8403
8404 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_LOCKED,
8405 0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
8406
8407 #else /* DEBUG || DEVELOPMENT */
8408
8409 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_LOCKED | CTLFLAG_MASKED,
8410 0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
8411
8412 #endif /* DEBUG || DEVELOPMENT */
8413
8414
8415 static int
8416 sysctl_memorypressure_manual_trigger SYSCTL_HANDLER_ARGS
8417 {
8418 #pragma unused(arg1, arg2)
8419
8420 int level = 0;
8421 int error = 0;
8422 int pressure_level = 0;
8423 int trigger_request = 0;
8424 int force_purge;
8425
8426 error = sysctl_handle_int(oidp, &level, 0, req);
8427 if (error || !req->newptr) {
8428 return error;
8429 }
8430
8431 memorystatus_manual_testing_on = TRUE;
8432
8433 trigger_request = (level >> 16) & 0xFFFF;
8434 pressure_level = (level & 0xFFFF);
8435
8436 if (trigger_request < TEST_LOW_MEMORY_TRIGGER_ONE ||
8437 trigger_request > TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL) {
8438 return EINVAL;
8439 }
8440 switch (pressure_level) {
8441 case NOTE_MEMORYSTATUS_PRESSURE_NORMAL:
8442 case NOTE_MEMORYSTATUS_PRESSURE_WARN:
8443 case NOTE_MEMORYSTATUS_PRESSURE_CRITICAL:
8444 break;
8445 default:
8446 return EINVAL;
8447 }
8448
8449 /*
8450 * The pressure level is being set from user-space.
8451 * And user-space uses the constants in sys/event.h
8452 * So we translate those events to our internal levels here.
8453 */
8454 if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
8455 memorystatus_manual_testing_level = kVMPressureNormal;
8456 force_purge = 0;
8457 } else if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_WARN) {
8458 memorystatus_manual_testing_level = kVMPressureWarning;
8459 force_purge = vm_pageout_state.memorystatus_purge_on_warning;
8460 } else if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
8461 memorystatus_manual_testing_level = kVMPressureCritical;
8462 force_purge = vm_pageout_state.memorystatus_purge_on_critical;
8463 }
8464
8465 memorystatus_vm_pressure_level = memorystatus_manual_testing_level;
8466
8467 /* purge according to the new pressure level */
8468 switch (trigger_request) {
8469 case TEST_PURGEABLE_TRIGGER_ONE:
8470 case TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ONE:
8471 if (force_purge == 0) {
8472 /* no purging requested */
8473 break;
8474 }
8475 vm_purgeable_object_purge_one_unlocked(force_purge);
8476 break;
8477 case TEST_PURGEABLE_TRIGGER_ALL:
8478 case TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL:
8479 if (force_purge == 0) {
8480 /* no purging requested */
8481 break;
8482 }
8483 while (vm_purgeable_object_purge_one_unlocked(force_purge)) {
8484 ;
8485 }
8486 break;
8487 }
8488
8489 if ((trigger_request == TEST_LOW_MEMORY_TRIGGER_ONE) ||
8490 (trigger_request == TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ONE)) {
8491 memorystatus_update_vm_pressure(TRUE);
8492 }
8493
8494 if ((trigger_request == TEST_LOW_MEMORY_TRIGGER_ALL) ||
8495 (trigger_request == TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL)) {
8496 while (memorystatus_update_vm_pressure(FALSE) == KERN_SUCCESS) {
8497 continue;
8498 }
8499 }
8500
8501 if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
8502 memorystatus_manual_testing_on = FALSE;
8503 }
8504
8505 return 0;
8506 }
8507
8508 SYSCTL_PROC(_kern, OID_AUTO, memorypressure_manual_trigger, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
8509 0, 0, &sysctl_memorypressure_manual_trigger, "I", "");
8510
8511
8512 SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_warning, CTLFLAG_RW | CTLFLAG_LOCKED, &vm_pageout_state.memorystatus_purge_on_warning, 0, "");
8513 SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_urgent, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &vm_pageout_state.memorystatus_purge_on_urgent, 0, "");
8514 SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_critical, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, &vm_pageout_state.memorystatus_purge_on_critical, 0, "");
8515
8516 #if DEBUG || DEVELOPMENT
8517 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_vm_pressure_events_enabled, CTLFLAG_RW | CTLFLAG_LOCKED, &vm_pressure_events_enabled, 0, "");
8518 #endif
8519
8520 #endif /* VM_PRESSURE_EVENTS */
8521
8522 /* Return both allocated and actual size, since there's a race between allocation and list compilation */
8523 static int
8524 memorystatus_get_priority_list(memorystatus_priority_entry_t **list_ptr, size_t *buffer_size, size_t *list_size, boolean_t size_only)
8525 {
8526 uint32_t list_count, i = 0;
8527 memorystatus_priority_entry_t *list_entry;
8528 proc_t p;
8529
8530 list_count = memorystatus_list_count;
8531 *list_size = sizeof(memorystatus_priority_entry_t) * list_count;
8532
8533 /* Just a size check? */
8534 if (size_only) {
8535 return 0;
8536 }
8537
8538 /* Otherwise, validate the size of the buffer */
8539 if (*buffer_size < *list_size) {
8540 return EINVAL;
8541 }
8542
8543 *list_ptr = (memorystatus_priority_entry_t*)kalloc(*list_size);
8544 if (!*list_ptr) {
8545 return ENOMEM;
8546 }
8547
8548 memset(*list_ptr, 0, *list_size);
8549
8550 *buffer_size = *list_size;
8551 *list_size = 0;
8552
8553 list_entry = *list_ptr;
8554
8555 proc_list_lock();
8556
8557 p = memorystatus_get_first_proc_locked(&i, TRUE);
8558 while (p && (*list_size < *buffer_size)) {
8559 list_entry->pid = p->p_pid;
8560 list_entry->priority = p->p_memstat_effectivepriority;
8561 list_entry->user_data = p->p_memstat_userdata;
8562
8563 if (p->p_memstat_memlimit <= 0) {
8564 task_get_phys_footprint_limit(p->task, &list_entry->limit);
8565 } else {
8566 list_entry->limit = p->p_memstat_memlimit;
8567 }
8568
8569 list_entry->state = memorystatus_build_state(p);
8570 list_entry++;
8571
8572 *list_size += sizeof(memorystatus_priority_entry_t);
8573
8574 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
8575 }
8576
8577 proc_list_unlock();
8578
8579 MEMORYSTATUS_DEBUG(1, "memorystatus_get_priority_list: returning %lu for size\n", (unsigned long)*list_size);
8580
8581 return 0;
8582 }
8583
8584 static int
8585 memorystatus_get_priority_pid(pid_t pid, user_addr_t buffer, size_t buffer_size)
8586 {
8587 int error = 0;
8588 memorystatus_priority_entry_t mp_entry;
8589
8590 /* Validate inputs */
8591 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_entry_t))) {
8592 return EINVAL;
8593 }
8594
8595 proc_t p = proc_find(pid);
8596 if (!p) {
8597 return ESRCH;
8598 }
8599
8600 memset(&mp_entry, 0, sizeof(memorystatus_priority_entry_t));
8601
8602 mp_entry.pid = p->p_pid;
8603 mp_entry.priority = p->p_memstat_effectivepriority;
8604 mp_entry.user_data = p->p_memstat_userdata;
8605 if (p->p_memstat_memlimit <= 0) {
8606 task_get_phys_footprint_limit(p->task, &mp_entry.limit);
8607 } else {
8608 mp_entry.limit = p->p_memstat_memlimit;
8609 }
8610 mp_entry.state = memorystatus_build_state(p);
8611
8612 proc_rele(p);
8613
8614 error = copyout(&mp_entry, buffer, buffer_size);
8615
8616 return error;
8617 }
8618
8619 static int
8620 memorystatus_cmd_get_priority_list(pid_t pid, user_addr_t buffer, size_t buffer_size, int32_t *retval)
8621 {
8622 int error = 0;
8623 boolean_t size_only;
8624 size_t list_size;
8625
8626 /*
8627 * When a non-zero pid is provided, the 'list' has only one entry.
8628 */
8629
8630 size_only = ((buffer == USER_ADDR_NULL) ? TRUE: FALSE);
8631
8632 if (pid != 0) {
8633 list_size = sizeof(memorystatus_priority_entry_t) * 1;
8634 if (!size_only) {
8635 error = memorystatus_get_priority_pid(pid, buffer, buffer_size);
8636 }
8637 } else {
8638 memorystatus_priority_entry_t *list = NULL;
8639 error = memorystatus_get_priority_list(&list, &buffer_size, &list_size, size_only);
8640
8641 if (error == 0) {
8642 if (!size_only) {
8643 error = copyout(list, buffer, list_size);
8644 }
8645 }
8646
8647 if (list) {
8648 kfree(list, buffer_size);
8649 }
8650 }
8651
8652 if (error == 0) {
8653 *retval = list_size;
8654 }
8655
8656 return error;
8657 }
8658
8659 static void
8660 memorystatus_clear_errors(void)
8661 {
8662 proc_t p;
8663 unsigned int i = 0;
8664
8665 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_START, 0, 0, 0, 0, 0);
8666
8667 proc_list_lock();
8668
8669 p = memorystatus_get_first_proc_locked(&i, TRUE);
8670 while (p) {
8671 if (p->p_memstat_state & P_MEMSTAT_ERROR) {
8672 p->p_memstat_state &= ~P_MEMSTAT_ERROR;
8673 }
8674 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
8675 }
8676
8677 proc_list_unlock();
8678
8679 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_END, 0, 0, 0, 0, 0);
8680 }
8681
8682 #if CONFIG_JETSAM
8683 static void
8684 memorystatus_update_levels_locked(boolean_t critical_only)
8685 {
8686 memorystatus_available_pages_critical = memorystatus_available_pages_critical_base;
8687
8688 /*
8689 * If there's an entry in the first bucket, we have idle processes.
8690 */
8691
8692 memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
8693 if (first_bucket->count) {
8694 memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
8695
8696 if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure) {
8697 /*
8698 * The critical threshold must never exceed the pressure threshold
8699 */
8700 memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
8701 }
8702 }
8703
8704 #if DEBUG || DEVELOPMENT
8705 if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
8706 memorystatus_available_pages_critical += memorystatus_jetsam_policy_offset_pages_diagnostic;
8707
8708 if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure) {
8709 /*
8710 * The critical threshold must never exceed the pressure threshold
8711 */
8712 memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
8713 }
8714 }
8715 #endif /* DEBUG || DEVELOPMENT */
8716
8717 if (memorystatus_jetsam_policy & kPolicyMoreFree) {
8718 memorystatus_available_pages_critical += memorystatus_policy_more_free_offset_pages;
8719 }
8720
8721 if (critical_only) {
8722 return;
8723 }
8724
8725 #if VM_PRESSURE_EVENTS
8726 memorystatus_available_pages_pressure = (pressure_threshold_percentage / delta_percentage) * memorystatus_delta;
8727 #if DEBUG || DEVELOPMENT
8728 if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
8729 memorystatus_available_pages_pressure += memorystatus_jetsam_policy_offset_pages_diagnostic;
8730 }
8731 #endif
8732 #endif
8733 }
8734
8735 void
8736 memorystatus_fast_jetsam_override(boolean_t enable_override)
8737 {
8738 /* If fast jetsam is not enabled, simply return */
8739 if (!fast_jetsam_enabled) {
8740 return;
8741 }
8742
8743 if (enable_override) {
8744 if ((memorystatus_jetsam_policy & kPolicyMoreFree) == kPolicyMoreFree) {
8745 return;
8746 }
8747 proc_list_lock();
8748 memorystatus_jetsam_policy |= kPolicyMoreFree;
8749 memorystatus_thread_pool_max();
8750 memorystatus_update_levels_locked(TRUE);
8751 proc_list_unlock();
8752 } else {
8753 if ((memorystatus_jetsam_policy & kPolicyMoreFree) == 0) {
8754 return;
8755 }
8756 proc_list_lock();
8757 memorystatus_jetsam_policy &= ~kPolicyMoreFree;
8758 memorystatus_thread_pool_default();
8759 memorystatus_update_levels_locked(TRUE);
8760 proc_list_unlock();
8761 }
8762 }
8763
8764
8765 static int
8766 sysctl_kern_memorystatus_policy_more_free SYSCTL_HANDLER_ARGS
8767 {
8768 #pragma unused(arg1, arg2, oidp)
8769 int error = 0, more_free = 0;
8770
8771 /*
8772 * TODO: Enable this privilege check?
8773 *
8774 * error = priv_check_cred(kauth_cred_get(), PRIV_VM_JETSAM, 0);
8775 * if (error)
8776 * return (error);
8777 */
8778
8779 error = sysctl_handle_int(oidp, &more_free, 0, req);
8780 if (error || !req->newptr) {
8781 return error;
8782 }
8783
8784 if (more_free) {
8785 memorystatus_fast_jetsam_override(true);
8786 } else {
8787 memorystatus_fast_jetsam_override(false);
8788 }
8789
8790 return 0;
8791 }
8792 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_policy_more_free, CTLTYPE_INT | CTLFLAG_WR | CTLFLAG_LOCKED | CTLFLAG_MASKED,
8793 0, 0, &sysctl_kern_memorystatus_policy_more_free, "I", "");
8794
8795 #endif /* CONFIG_JETSAM */
8796
8797 /*
8798 * Get the at_boot snapshot
8799 */
8800 static int
8801 memorystatus_get_at_boot_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
8802 {
8803 size_t input_size = *snapshot_size;
8804
8805 /*
8806 * The at_boot snapshot has no entry list.
8807 */
8808 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t);
8809
8810 if (size_only) {
8811 return 0;
8812 }
8813
8814 /*
8815 * Validate the size of the snapshot buffer
8816 */
8817 if (input_size < *snapshot_size) {
8818 return EINVAL;
8819 }
8820
8821 /*
8822 * Update the notification_time only
8823 */
8824 memorystatus_at_boot_snapshot.notification_time = mach_absolute_time();
8825 *snapshot = &memorystatus_at_boot_snapshot;
8826
8827 MEMORYSTATUS_DEBUG(7, "memorystatus_get_at_boot_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%d)\n",
8828 (long)input_size, (long)*snapshot_size, 0);
8829 return 0;
8830 }
8831
8832 /*
8833 * Get the previous fully populated snapshot
8834 */
8835 static int
8836 memorystatus_get_jetsam_snapshot_copy(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
8837 {
8838 size_t input_size = *snapshot_size;
8839
8840 if (memorystatus_jetsam_snapshot_copy_count > 0) {
8841 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_copy_count));
8842 } else {
8843 *snapshot_size = 0;
8844 }
8845
8846 if (size_only) {
8847 return 0;
8848 }
8849
8850 if (input_size < *snapshot_size) {
8851 return EINVAL;
8852 }
8853
8854 *snapshot = memorystatus_jetsam_snapshot_copy;
8855
8856 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot_copy: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
8857 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_copy_count);
8858
8859 return 0;
8860 }
8861
8862 static int
8863 memorystatus_get_on_demand_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
8864 {
8865 size_t input_size = *snapshot_size;
8866 uint32_t ods_list_count = memorystatus_list_count;
8867 memorystatus_jetsam_snapshot_t *ods = NULL; /* The on_demand snapshot buffer */
8868
8869 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (ods_list_count));
8870
8871 if (size_only) {
8872 return 0;
8873 }
8874
8875 /*
8876 * Validate the size of the snapshot buffer.
8877 * This is inherently racey. May want to revisit
8878 * this error condition and trim the output when
8879 * it doesn't fit.
8880 */
8881 if (input_size < *snapshot_size) {
8882 return EINVAL;
8883 }
8884
8885 /*
8886 * Allocate and initialize a snapshot buffer.
8887 */
8888 ods = (memorystatus_jetsam_snapshot_t *)kalloc(*snapshot_size);
8889 if (!ods) {
8890 return ENOMEM;
8891 }
8892
8893 memset(ods, 0, *snapshot_size);
8894
8895 proc_list_lock();
8896 memorystatus_init_jetsam_snapshot_locked(ods, ods_list_count);
8897 proc_list_unlock();
8898
8899 /*
8900 * Return the kernel allocated, on_demand buffer.
8901 * The caller of this routine will copy the data out
8902 * to user space and then free the kernel allocated
8903 * buffer.
8904 */
8905 *snapshot = ods;
8906
8907 MEMORYSTATUS_DEBUG(7, "memorystatus_get_on_demand_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
8908 (long)input_size, (long)*snapshot_size, (long)ods_list_count);
8909
8910 return 0;
8911 }
8912
8913 static int
8914 memorystatus_get_jetsam_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only)
8915 {
8916 size_t input_size = *snapshot_size;
8917
8918 if (memorystatus_jetsam_snapshot_count > 0) {
8919 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count));
8920 } else {
8921 *snapshot_size = 0;
8922 }
8923
8924 if (size_only) {
8925 return 0;
8926 }
8927
8928 if (input_size < *snapshot_size) {
8929 return EINVAL;
8930 }
8931
8932 *snapshot = memorystatus_jetsam_snapshot;
8933
8934 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
8935 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_count);
8936
8937 return 0;
8938 }
8939
8940
8941 static int
8942 memorystatus_cmd_get_jetsam_snapshot(int32_t flags, user_addr_t buffer, size_t buffer_size, int32_t *retval)
8943 {
8944 int error = EINVAL;
8945 boolean_t size_only;
8946 boolean_t is_default_snapshot = FALSE;
8947 boolean_t is_on_demand_snapshot = FALSE;
8948 boolean_t is_at_boot_snapshot = FALSE;
8949 memorystatus_jetsam_snapshot_t *snapshot;
8950
8951 size_only = ((buffer == USER_ADDR_NULL) ? TRUE : FALSE);
8952
8953 if (flags == 0) {
8954 /* Default */
8955 is_default_snapshot = TRUE;
8956 error = memorystatus_get_jetsam_snapshot(&snapshot, &buffer_size, size_only);
8957 } else {
8958 if (flags & ~(MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT | MEMORYSTATUS_SNAPSHOT_COPY)) {
8959 /*
8960 * Unsupported bit set in flag.
8961 */
8962 return EINVAL;
8963 }
8964
8965 if (flags & (flags - 0x1)) {
8966 /*
8967 * Can't have multiple flags set at the same time.
8968 */
8969 return EINVAL;
8970 }
8971
8972 if (flags & MEMORYSTATUS_SNAPSHOT_ON_DEMAND) {
8973 is_on_demand_snapshot = TRUE;
8974 /*
8975 * When not requesting the size only, the following call will allocate
8976 * an on_demand snapshot buffer, which is freed below.
8977 */
8978 error = memorystatus_get_on_demand_snapshot(&snapshot, &buffer_size, size_only);
8979 } else if (flags & MEMORYSTATUS_SNAPSHOT_AT_BOOT) {
8980 is_at_boot_snapshot = TRUE;
8981 error = memorystatus_get_at_boot_snapshot(&snapshot, &buffer_size, size_only);
8982 } else if (flags & MEMORYSTATUS_SNAPSHOT_COPY) {
8983 error = memorystatus_get_jetsam_snapshot_copy(&snapshot, &buffer_size, size_only);
8984 } else {
8985 /*
8986 * Invalid flag setting.
8987 */
8988 return EINVAL;
8989 }
8990 }
8991
8992 if (error) {
8993 goto out;
8994 }
8995
8996 /*
8997 * Copy the data out to user space and clear the snapshot buffer.
8998 * If working with the jetsam snapshot,
8999 * clearing the buffer means, reset the count.
9000 * If working with an on_demand snapshot
9001 * clearing the buffer means, free it.
9002 * If working with the at_boot snapshot
9003 * there is nothing to clear or update.
9004 * If working with a copy of the snapshot
9005 * there is nothing to clear or update.
9006 */
9007 if (!size_only) {
9008 if ((error = copyout(snapshot, buffer, buffer_size)) == 0) {
9009 if (is_default_snapshot) {
9010 /*
9011 * The jetsam snapshot is never freed, its count is simply reset.
9012 * However, we make a copy for any parties that might be interested
9013 * in the previous fully populated snapshot.
9014 */
9015 proc_list_lock();
9016 memcpy(memorystatus_jetsam_snapshot_copy, memorystatus_jetsam_snapshot, memorystatus_jetsam_snapshot_size);
9017 memorystatus_jetsam_snapshot_copy_count = memorystatus_jetsam_snapshot_count;
9018 snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
9019 memorystatus_jetsam_snapshot_last_timestamp = 0;
9020 proc_list_unlock();
9021 }
9022 }
9023
9024 if (is_on_demand_snapshot) {
9025 /*
9026 * The on_demand snapshot is always freed,
9027 * even if the copyout failed.
9028 */
9029 if (snapshot) {
9030 kfree(snapshot, buffer_size);
9031 }
9032 }
9033 }
9034
9035 if (error == 0) {
9036 *retval = buffer_size;
9037 }
9038 out:
9039 return error;
9040 }
9041
9042 /*
9043 * Routine: memorystatus_cmd_grp_set_priorities
9044 * Purpose: Update priorities for a group of processes.
9045 *
9046 * [priority]
9047 * Move each process out of its effective priority
9048 * band and into a new priority band.
9049 * Maintains relative order from lowest to highest priority.
9050 * In single band, maintains relative order from head to tail.
9051 *
9052 * eg: before [effectivepriority | pid]
9053 * [18 | p101 ]
9054 * [17 | p55, p67, p19 ]
9055 * [12 | p103 p10 ]
9056 * [ 7 | p25 ]
9057 * [ 0 | p71, p82, ]
9058 *
9059 * after [ new band | pid]
9060 * [ xxx | p71, p82, p25, p103, p10, p55, p67, p19, p101]
9061 *
9062 * Returns: 0 on success, else non-zero.
9063 *
9064 * Caveat: We know there is a race window regarding recycled pids.
9065 * A process could be killed before the kernel can act on it here.
9066 * If a pid cannot be found in any of the jetsam priority bands,
9067 * then we simply ignore it. No harm.
9068 * But, if the pid has been recycled then it could be an issue.
9069 * In that scenario, we might move an unsuspecting process to the new
9070 * priority band. It's not clear how the kernel can safeguard
9071 * against this, but it would be an extremely rare case anyway.
9072 * The caller of this api might avoid such race conditions by
9073 * ensuring that the processes passed in the pid list are suspended.
9074 */
9075
9076
9077 static int
9078 memorystatus_cmd_grp_set_priorities(user_addr_t buffer, size_t buffer_size)
9079 {
9080 /*
9081 * We only handle setting priority
9082 * per process
9083 */
9084
9085 int error = 0;
9086 memorystatus_properties_entry_v1_t *entries = NULL;
9087 uint32_t entry_count = 0;
9088
9089 /* This will be the ordered proc list */
9090 typedef struct memorystatus_internal_properties {
9091 proc_t proc;
9092 int32_t priority;
9093 } memorystatus_internal_properties_t;
9094
9095 memorystatus_internal_properties_t *table = NULL;
9096 size_t table_size = 0;
9097 uint32_t table_count = 0;
9098
9099 uint32_t i = 0;
9100 uint32_t bucket_index = 0;
9101 boolean_t head_insert;
9102 int32_t new_priority;
9103
9104 proc_t p;
9105
9106 /* Verify inputs */
9107 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
9108 error = EINVAL;
9109 goto out;
9110 }
9111
9112 entry_count = (buffer_size / sizeof(memorystatus_properties_entry_v1_t));
9113 if ((entries = (memorystatus_properties_entry_v1_t *)kalloc(buffer_size)) == NULL) {
9114 error = ENOMEM;
9115 goto out;
9116 }
9117
9118 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);
9119
9120 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
9121 goto out;
9122 }
9123
9124 /* Verify sanity of input priorities */
9125 if (entries[0].version == MEMORYSTATUS_MPE_VERSION_1) {
9126 if ((buffer_size % MEMORYSTATUS_MPE_VERSION_1_SIZE) != 0) {
9127 error = EINVAL;
9128 goto out;
9129 }
9130 } else {
9131 error = EINVAL;
9132 goto out;
9133 }
9134
9135 for (i = 0; i < entry_count; i++) {
9136 if (entries[i].priority == -1) {
9137 /* Use as shorthand for default priority */
9138 entries[i].priority = JETSAM_PRIORITY_DEFAULT;
9139 } else if ((entries[i].priority == system_procs_aging_band) || (entries[i].priority == applications_aging_band)) {
9140 /* Both the aging bands are reserved for internal use;
9141 * if requested, adjust to JETSAM_PRIORITY_IDLE. */
9142 entries[i].priority = JETSAM_PRIORITY_IDLE;
9143 } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
9144 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle
9145 * queue */
9146 /* Deal with this later */
9147 } else if ((entries[i].priority < 0) || (entries[i].priority >= MEMSTAT_BUCKET_COUNT)) {
9148 /* Sanity check */
9149 error = EINVAL;
9150 goto out;
9151 }
9152 }
9153
9154 table_size = sizeof(memorystatus_internal_properties_t) * entry_count;
9155 if ((table = (memorystatus_internal_properties_t *)kalloc(table_size)) == NULL) {
9156 error = ENOMEM;
9157 goto out;
9158 }
9159 memset(table, 0, table_size);
9160
9161
9162 /*
9163 * For each jetsam bucket entry, spin through the input property list.
9164 * When a matching pid is found, populate an adjacent table with the
9165 * appropriate proc pointer and new property values.
9166 * This traversal automatically preserves order from lowest
9167 * to highest priority.
9168 */
9169
9170 bucket_index = 0;
9171
9172 proc_list_lock();
9173
9174 /* Create the ordered table */
9175 p = memorystatus_get_first_proc_locked(&bucket_index, TRUE);
9176 while (p && (table_count < entry_count)) {
9177 for (i = 0; i < entry_count; i++) {
9178 if (p->p_pid == entries[i].pid) {
9179 /* Build the table data */
9180 table[table_count].proc = p;
9181 table[table_count].priority = entries[i].priority;
9182 table_count++;
9183 break;
9184 }
9185 }
9186 p = memorystatus_get_next_proc_locked(&bucket_index, p, TRUE);
9187 }
9188
9189 /* We now have ordered list of procs ready to move */
9190 for (i = 0; i < table_count; i++) {
9191 p = table[i].proc;
9192 assert(p != NULL);
9193
9194 /* Allow head inserts -- but relative order is now */
9195 if (table[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
9196 new_priority = JETSAM_PRIORITY_IDLE;
9197 head_insert = true;
9198 } else {
9199 new_priority = table[i].priority;
9200 head_insert = false;
9201 }
9202
9203 /* Not allowed */
9204 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
9205 continue;
9206 }
9207
9208 /*
9209 * Take appropriate steps if moving proc out of
9210 * either of the aging bands.
9211 */
9212 if ((p->p_memstat_effectivepriority == system_procs_aging_band) || (p->p_memstat_effectivepriority == applications_aging_band)) {
9213 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
9214 }
9215
9216 memorystatus_update_priority_locked(p, new_priority, head_insert, false);
9217 }
9218
9219 proc_list_unlock();
9220
9221 /*
9222 * if (table_count != entry_count)
9223 * then some pids were not found in a jetsam band.
9224 * harmless but interesting...
9225 */
9226 out:
9227 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);
9228
9229 if (entries) {
9230 kfree(entries, buffer_size);
9231 }
9232 if (table) {
9233 kfree(table, table_size);
9234 }
9235
9236 return error;
9237 }
9238
9239 static int
9240 memorystatus_cmd_grp_set_probabilities(user_addr_t buffer, size_t buffer_size)
9241 {
9242 int error = 0;
9243 memorystatus_properties_entry_v1_t *entries = NULL;
9244 uint32_t entry_count = 0, i = 0;
9245 memorystatus_internal_probabilities_t *tmp_table_new = NULL, *tmp_table_old = NULL;
9246 size_t tmp_table_new_size = 0, tmp_table_old_size = 0;
9247
9248 /* Verify inputs */
9249 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
9250 error = EINVAL;
9251 goto out;
9252 }
9253
9254 entry_count = (buffer_size / sizeof(memorystatus_properties_entry_v1_t));
9255
9256 if ((entries = (memorystatus_properties_entry_v1_t *) kalloc(buffer_size)) == NULL) {
9257 error = ENOMEM;
9258 goto out;
9259 }
9260
9261 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);
9262
9263 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
9264 goto out;
9265 }
9266
9267 if (entries[0].version == MEMORYSTATUS_MPE_VERSION_1) {
9268 if ((buffer_size % MEMORYSTATUS_MPE_VERSION_1_SIZE) != 0) {
9269 error = EINVAL;
9270 goto out;
9271 }
9272 } else {
9273 error = EINVAL;
9274 goto out;
9275 }
9276
9277 /* Verify sanity of input priorities */
9278 for (i = 0; i < entry_count; i++) {
9279 /*
9280 * 0 - low probability of use.
9281 * 1 - high probability of use.
9282 *
9283 * Keeping this field an int (& not a bool) to allow
9284 * us to experiment with different values/approaches
9285 * later on.
9286 */
9287 if (entries[i].use_probability > 1) {
9288 error = EINVAL;
9289 goto out;
9290 }
9291 }
9292
9293 tmp_table_new_size = sizeof(memorystatus_internal_probabilities_t) * entry_count;
9294
9295 if ((tmp_table_new = (memorystatus_internal_probabilities_t *) kalloc(tmp_table_new_size)) == NULL) {
9296 error = ENOMEM;
9297 goto out;
9298 }
9299 memset(tmp_table_new, 0, tmp_table_new_size);
9300
9301 proc_list_lock();
9302
9303 if (memorystatus_global_probabilities_table) {
9304 tmp_table_old = memorystatus_global_probabilities_table;
9305 tmp_table_old_size = memorystatus_global_probabilities_size;
9306 }
9307
9308 memorystatus_global_probabilities_table = tmp_table_new;
9309 memorystatus_global_probabilities_size = tmp_table_new_size;
9310 tmp_table_new = NULL;
9311
9312 for (i = 0; i < entry_count; i++) {
9313 /* Build the table data */
9314 strlcpy(memorystatus_global_probabilities_table[i].proc_name, entries[i].proc_name, MAXCOMLEN + 1);
9315 memorystatus_global_probabilities_table[i].use_probability = entries[i].use_probability;
9316 }
9317
9318 proc_list_unlock();
9319
9320 out:
9321 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);
9322
9323 if (entries) {
9324 kfree(entries, buffer_size);
9325 entries = NULL;
9326 }
9327
9328 if (tmp_table_old) {
9329 kfree(tmp_table_old, tmp_table_old_size);
9330 tmp_table_old = NULL;
9331 }
9332
9333 return error;
9334 }
9335
9336 static int
9337 memorystatus_cmd_grp_set_properties(int32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
9338 {
9339 int error = 0;
9340
9341 if ((flags & MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY) == MEMORYSTATUS_FLAGS_GRP_SET_PRIORITY) {
9342 error = memorystatus_cmd_grp_set_priorities(buffer, buffer_size);
9343 } else if ((flags & MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY) == MEMORYSTATUS_FLAGS_GRP_SET_PROBABILITY) {
9344 error = memorystatus_cmd_grp_set_probabilities(buffer, buffer_size);
9345 } else {
9346 error = EINVAL;
9347 }
9348
9349 return error;
9350 }
9351
9352 /*
9353 * This routine is used to update a process's jetsam priority position and stored user_data.
9354 * It is not used for the setting of memory limits, which is why the last 6 args to the
9355 * memorystatus_update() call are 0 or FALSE.
9356 */
9357
9358 static int
9359 memorystatus_cmd_set_priority_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
9360 {
9361 int error = 0;
9362 memorystatus_priority_properties_t mpp_entry;
9363
9364 /* Validate inputs */
9365 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_properties_t))) {
9366 return EINVAL;
9367 }
9368
9369 error = copyin(buffer, &mpp_entry, buffer_size);
9370
9371 if (error == 0) {
9372 proc_t p;
9373
9374 p = proc_find(pid);
9375 if (!p) {
9376 return ESRCH;
9377 }
9378
9379 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
9380 proc_rele(p);
9381 return EPERM;
9382 }
9383
9384 error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, FALSE, FALSE, 0, 0, FALSE, FALSE);
9385 proc_rele(p);
9386 }
9387
9388 return error;
9389 }
9390
9391 static int
9392 memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
9393 {
9394 int error = 0;
9395 memorystatus_memlimit_properties_t mmp_entry;
9396
9397 /* Validate inputs */
9398 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
9399 return EINVAL;
9400 }
9401
9402 error = copyin(buffer, &mmp_entry, buffer_size);
9403
9404 if (error == 0) {
9405 error = memorystatus_set_memlimit_properties(pid, &mmp_entry);
9406 }
9407
9408 return error;
9409 }
9410
9411 /*
9412 * When getting the memlimit settings, we can't simply call task_get_phys_footprint_limit().
9413 * That gets the proc's cached memlimit and there is no guarantee that the active/inactive
9414 * limits will be the same in the no-limit case. Instead we convert limits <= 0 using
9415 * task_convert_phys_footprint_limit(). It computes the same limit value that would be written
9416 * to the task's ledgers via task_set_phys_footprint_limit().
9417 */
9418 static int
9419 memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
9420 {
9421 int error = 0;
9422 memorystatus_memlimit_properties_t mmp_entry;
9423
9424 /* Validate inputs */
9425 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
9426 return EINVAL;
9427 }
9428
9429 memset(&mmp_entry, 0, sizeof(memorystatus_memlimit_properties_t));
9430
9431 proc_t p = proc_find(pid);
9432 if (!p) {
9433 return ESRCH;
9434 }
9435
9436 /*
9437 * Get the active limit and attributes.
9438 * No locks taken since we hold a reference to the proc.
9439 */
9440
9441 if (p->p_memstat_memlimit_active > 0) {
9442 mmp_entry.memlimit_active = p->p_memstat_memlimit_active;
9443 } else {
9444 task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_active);
9445 }
9446
9447 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) {
9448 mmp_entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
9449 }
9450
9451 /*
9452 * Get the inactive limit and attributes
9453 */
9454 if (p->p_memstat_memlimit_inactive <= 0) {
9455 task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_inactive);
9456 } else {
9457 mmp_entry.memlimit_inactive = p->p_memstat_memlimit_inactive;
9458 }
9459 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) {
9460 mmp_entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
9461 }
9462 proc_rele(p);
9463
9464 error = copyout(&mmp_entry, buffer, buffer_size);
9465
9466 return error;
9467 }
9468
9469
9470 /*
9471 * SPI for kbd - pr24956468
9472 * This is a very simple snapshot that calculates how much a
9473 * process's phys_footprint exceeds a specific memory limit.
9474 * Only the inactive memory limit is supported for now.
9475 * The delta is returned as bytes in excess or zero.
9476 */
9477 static int
9478 memorystatus_cmd_get_memlimit_excess_np(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval)
9479 {
9480 int error = 0;
9481 uint64_t footprint_in_bytes = 0;
9482 uint64_t delta_in_bytes = 0;
9483 int32_t memlimit_mb = 0;
9484 uint64_t memlimit_bytes = 0;
9485
9486 /* Validate inputs */
9487 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(uint64_t)) || (flags != 0)) {
9488 return EINVAL;
9489 }
9490
9491 proc_t p = proc_find(pid);
9492 if (!p) {
9493 return ESRCH;
9494 }
9495
9496 /*
9497 * Get the inactive limit.
9498 * No locks taken since we hold a reference to the proc.
9499 */
9500
9501 if (p->p_memstat_memlimit_inactive <= 0) {
9502 task_convert_phys_footprint_limit(-1, &memlimit_mb);
9503 } else {
9504 memlimit_mb = p->p_memstat_memlimit_inactive;
9505 }
9506
9507 footprint_in_bytes = get_task_phys_footprint(p->task);
9508
9509 proc_rele(p);
9510
9511 memlimit_bytes = memlimit_mb * 1024 * 1024; /* MB to bytes */
9512
9513 /*
9514 * Computed delta always returns >= 0 bytes
9515 */
9516 if (footprint_in_bytes > memlimit_bytes) {
9517 delta_in_bytes = footprint_in_bytes - memlimit_bytes;
9518 }
9519
9520 error = copyout(&delta_in_bytes, buffer, sizeof(delta_in_bytes));
9521
9522 return error;
9523 }
9524
9525
9526 static int
9527 memorystatus_cmd_get_pressure_status(int32_t *retval)
9528 {
9529 int error;
9530
9531 /* Need privilege for check */
9532 error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
9533 if (error) {
9534 return error;
9535 }
9536
9537 /* Inherently racy, so it's not worth taking a lock here */
9538 *retval = (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
9539
9540 return error;
9541 }
9542
9543 int
9544 memorystatus_get_pressure_status_kdp()
9545 {
9546 return (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
9547 }
9548
9549 /*
9550 * Every process, including a P_MEMSTAT_INTERNAL process (currently only pid 1), is allowed to set a HWM.
9551 *
9552 * This call is inflexible -- it does not distinguish between active/inactive, fatal/non-fatal
9553 * So, with 2-level HWM preserving previous behavior will map as follows.
9554 * - treat the limit passed in as both an active and inactive limit.
9555 * - treat the is_fatal_limit flag as though it applies to both active and inactive limits.
9556 *
9557 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK
9558 * - the is_fatal_limit is FALSE, meaning the active and inactive limits are non-fatal/soft
9559 * - so mapping is (active/non-fatal, inactive/non-fatal)
9560 *
9561 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT
9562 * - the is_fatal_limit is TRUE, meaning the process's active and inactive limits are fatal/hard
9563 * - so mapping is (active/fatal, inactive/fatal)
9564 */
9565
9566 #if CONFIG_JETSAM
9567 static int
9568 memorystatus_cmd_set_jetsam_memory_limit(pid_t pid, int32_t high_water_mark, __unused int32_t *retval, boolean_t is_fatal_limit)
9569 {
9570 int error = 0;
9571 memorystatus_memlimit_properties_t entry;
9572
9573 entry.memlimit_active = high_water_mark;
9574 entry.memlimit_active_attr = 0;
9575 entry.memlimit_inactive = high_water_mark;
9576 entry.memlimit_inactive_attr = 0;
9577
9578 if (is_fatal_limit == TRUE) {
9579 entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
9580 entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
9581 }
9582
9583 error = memorystatus_set_memlimit_properties(pid, &entry);
9584 return error;
9585 }
9586 #endif /* CONFIG_JETSAM */
9587
9588 static int
9589 memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry)
9590 {
9591 int32_t memlimit_active;
9592 boolean_t memlimit_active_is_fatal;
9593 int32_t memlimit_inactive;
9594 boolean_t memlimit_inactive_is_fatal;
9595 uint32_t valid_attrs = 0;
9596 int error = 0;
9597
9598 proc_t p = proc_find(pid);
9599 if (!p) {
9600 return ESRCH;
9601 }
9602
9603 /*
9604 * Check for valid attribute flags.
9605 */
9606 valid_attrs |= (MEMORYSTATUS_MEMLIMIT_ATTR_FATAL);
9607 if ((entry->memlimit_active_attr & (~valid_attrs)) != 0) {
9608 proc_rele(p);
9609 return EINVAL;
9610 }
9611 if ((entry->memlimit_inactive_attr & (~valid_attrs)) != 0) {
9612 proc_rele(p);
9613 return EINVAL;
9614 }
9615
9616 /*
9617 * Setup the active memlimit properties
9618 */
9619 memlimit_active = entry->memlimit_active;
9620 if (entry->memlimit_active_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL) {
9621 memlimit_active_is_fatal = TRUE;
9622 } else {
9623 memlimit_active_is_fatal = FALSE;
9624 }
9625
9626 /*
9627 * Setup the inactive memlimit properties
9628 */
9629 memlimit_inactive = entry->memlimit_inactive;
9630 if (entry->memlimit_inactive_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL) {
9631 memlimit_inactive_is_fatal = TRUE;
9632 } else {
9633 memlimit_inactive_is_fatal = FALSE;
9634 }
9635
9636 /*
9637 * Setting a limit of <= 0 implies that the process has no
9638 * high-water-mark and has no per-task-limit. That means
9639 * the system_wide task limit is in place, which by the way,
9640 * is always fatal.
9641 */
9642
9643 if (memlimit_active <= 0) {
9644 /*
9645 * Enforce the fatal system_wide task limit while process is active.
9646 */
9647 memlimit_active = -1;
9648 memlimit_active_is_fatal = TRUE;
9649 }
9650
9651 if (memlimit_inactive <= 0) {
9652 /*
9653 * Enforce the fatal system_wide task limit while process is inactive.
9654 */
9655 memlimit_inactive = -1;
9656 memlimit_inactive_is_fatal = TRUE;
9657 }
9658
9659 proc_list_lock();
9660
9661 /*
9662 * Store the active limit variants in the proc.
9663 */
9664 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
9665
9666 /*
9667 * Store the inactive limit variants in the proc.
9668 */
9669 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
9670
9671 /*
9672 * Enforce appropriate limit variant by updating the cached values
9673 * and writing the ledger.
9674 * Limit choice is based on process active/inactive state.
9675 */
9676
9677 if (memorystatus_highwater_enabled) {
9678 boolean_t is_fatal;
9679 boolean_t use_active;
9680
9681 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
9682 CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
9683 use_active = TRUE;
9684 } else {
9685 CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
9686 use_active = FALSE;
9687 }
9688
9689 /* Enforce the limit by writing to the ledgers */
9690 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;
9691
9692 MEMORYSTATUS_DEBUG(3, "memorystatus_set_memlimit_properties: new limit on pid %d (%dMB %s) current priority (%d) dirty_state?=0x%x %s\n",
9693 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
9694 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
9695 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
9696 DTRACE_MEMORYSTATUS2(memorystatus_set_memlimit, proc_t, p, int32_t, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1));
9697 }
9698
9699 proc_list_unlock();
9700 proc_rele(p);
9701
9702 return error;
9703 }
9704
9705 /*
9706 * Returns the jetsam priority (effective or requested) of the process
9707 * associated with this task.
9708 */
9709 int
9710 proc_get_memstat_priority(proc_t p, boolean_t effective_priority)
9711 {
9712 if (p) {
9713 if (effective_priority) {
9714 return p->p_memstat_effectivepriority;
9715 } else {
9716 return p->p_memstat_requestedpriority;
9717 }
9718 }
9719 return 0;
9720 }
9721
9722 static int
9723 memorystatus_get_process_is_managed(pid_t pid, int *is_managed)
9724 {
9725 proc_t p = NULL;
9726
9727 /* Validate inputs */
9728 if (pid == 0) {
9729 return EINVAL;
9730 }
9731
9732 p = proc_find(pid);
9733 if (!p) {
9734 return ESRCH;
9735 }
9736
9737 proc_list_lock();
9738 *is_managed = ((p->p_memstat_state & P_MEMSTAT_MANAGED) ? 1 : 0);
9739 proc_rele_locked(p);
9740 proc_list_unlock();
9741
9742 return 0;
9743 }
9744
9745 static int
9746 memorystatus_set_process_is_managed(pid_t pid, boolean_t set_managed)
9747 {
9748 proc_t p = NULL;
9749
9750 /* Validate inputs */
9751 if (pid == 0) {
9752 return EINVAL;
9753 }
9754
9755 p = proc_find(pid);
9756 if (!p) {
9757 return ESRCH;
9758 }
9759
9760 proc_list_lock();
9761 if (set_managed == TRUE) {
9762 p->p_memstat_state |= P_MEMSTAT_MANAGED;
9763 } else {
9764 p->p_memstat_state &= ~P_MEMSTAT_MANAGED;
9765 }
9766 proc_rele_locked(p);
9767 proc_list_unlock();
9768
9769 return 0;
9770 }
9771
9772 static int
9773 memorystatus_get_process_is_freezable(pid_t pid, int *is_freezable)
9774 {
9775 proc_t p = PROC_NULL;
9776
9777 if (pid == 0) {
9778 return EINVAL;
9779 }
9780
9781 p = proc_find(pid);
9782 if (!p) {
9783 return ESRCH;
9784 }
9785
9786 /*
9787 * Only allow this on the current proc for now.
9788 * We can check for privileges and allow targeting another process in the future.
9789 */
9790 if (p != current_proc()) {
9791 proc_rele(p);
9792 return EPERM;
9793 }
9794
9795 proc_list_lock();
9796 *is_freezable = ((p->p_memstat_state & P_MEMSTAT_FREEZE_DISABLED) ? 0 : 1);
9797 proc_rele_locked(p);
9798 proc_list_unlock();
9799
9800 return 0;
9801 }
9802
9803 static int
9804 memorystatus_set_process_is_freezable(pid_t pid, boolean_t is_freezable)
9805 {
9806 proc_t p = PROC_NULL;
9807
9808 if (pid == 0) {
9809 return EINVAL;
9810 }
9811
9812 p = proc_find(pid);
9813 if (!p) {
9814 return ESRCH;
9815 }
9816
9817 /*
9818 * Only allow this on the current proc for now.
9819 * We can check for privileges and allow targeting another process in the future.
9820 */
9821 if (p != current_proc()) {
9822 proc_rele(p);
9823 return EPERM;
9824 }
9825
9826 proc_list_lock();
9827 if (is_freezable == FALSE) {
9828 /* Freeze preference set to FALSE. Set the P_MEMSTAT_FREEZE_DISABLED bit. */
9829 p->p_memstat_state |= P_MEMSTAT_FREEZE_DISABLED;
9830 printf("memorystatus_set_process_is_freezable: disabling freeze for pid %d [%s]\n",
9831 p->p_pid, (*p->p_name ? p->p_name : "unknown"));
9832 } else {
9833 p->p_memstat_state &= ~P_MEMSTAT_FREEZE_DISABLED;
9834 printf("memorystatus_set_process_is_freezable: enabling freeze for pid %d [%s]\n",
9835 p->p_pid, (*p->p_name ? p->p_name : "unknown"));
9836 }
9837 proc_rele_locked(p);
9838 proc_list_unlock();
9839
9840 return 0;
9841 }
9842
9843 int
9844 memorystatus_control(struct proc *p __unused, struct memorystatus_control_args *args, int *ret)
9845 {
9846 int error = EINVAL;
9847 boolean_t skip_auth_check = FALSE;
9848 os_reason_t jetsam_reason = OS_REASON_NULL;
9849
9850 #if !CONFIG_JETSAM
9851 #pragma unused(ret)
9852 #pragma unused(jetsam_reason)
9853 #endif
9854
9855 /* We don't need entitlements if we're setting/ querying the freeze preference for a process. Skip the check below. */
9856 if (args->command == MEMORYSTATUS_CMD_SET_PROCESS_IS_FREEZABLE || args->command == MEMORYSTATUS_CMD_GET_PROCESS_IS_FREEZABLE) {
9857 skip_auth_check = TRUE;
9858 }
9859
9860 /* Need to be root or have entitlement. */
9861 if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT) && !skip_auth_check) {
9862 error = EPERM;
9863 goto out;
9864 }
9865
9866 /*
9867 * Sanity check.
9868 * Do not enforce it for snapshots.
9869 */
9870 if (args->command != MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT) {
9871 if (args->buffersize > MEMORYSTATUS_BUFFERSIZE_MAX) {
9872 error = EINVAL;
9873 goto out;
9874 }
9875 }
9876
9877 switch (args->command) {
9878 case MEMORYSTATUS_CMD_GET_PRIORITY_LIST:
9879 error = memorystatus_cmd_get_priority_list(args->pid, args->buffer, args->buffersize, ret);
9880 break;
9881 case MEMORYSTATUS_CMD_SET_PRIORITY_PROPERTIES:
9882 error = memorystatus_cmd_set_priority_properties(args->pid, args->buffer, args->buffersize, ret);
9883 break;
9884 case MEMORYSTATUS_CMD_SET_MEMLIMIT_PROPERTIES:
9885 error = memorystatus_cmd_set_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
9886 break;
9887 case MEMORYSTATUS_CMD_GET_MEMLIMIT_PROPERTIES:
9888 error = memorystatus_cmd_get_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
9889 break;
9890 case MEMORYSTATUS_CMD_GET_MEMLIMIT_EXCESS:
9891 error = memorystatus_cmd_get_memlimit_excess_np(args->pid, args->flags, args->buffer, args->buffersize, ret);
9892 break;
9893 case MEMORYSTATUS_CMD_GRP_SET_PROPERTIES:
9894 error = memorystatus_cmd_grp_set_properties((int32_t)args->flags, args->buffer, args->buffersize, ret);
9895 break;
9896 case MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT:
9897 error = memorystatus_cmd_get_jetsam_snapshot((int32_t)args->flags, args->buffer, args->buffersize, ret);
9898 break;
9899 case MEMORYSTATUS_CMD_GET_PRESSURE_STATUS:
9900 error = memorystatus_cmd_get_pressure_status(ret);
9901 break;
9902 #if CONFIG_JETSAM
9903 case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
9904 /*
9905 * This call does not distinguish between active and inactive limits.
9906 * Default behavior in 2-level HWM world is to set both.
9907 * Non-fatal limit is also assumed for both.
9908 */
9909 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, FALSE);
9910 break;
9911 case MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT:
9912 /*
9913 * This call does not distinguish between active and inactive limits.
9914 * Default behavior in 2-level HWM world is to set both.
9915 * Fatal limit is also assumed for both.
9916 */
9917 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
9918 break;
9919 #endif /* CONFIG_JETSAM */
9920 /* Test commands */
9921 #if DEVELOPMENT || DEBUG
9922 case MEMORYSTATUS_CMD_TEST_JETSAM:
9923 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_GENERIC);
9924 if (jetsam_reason == OS_REASON_NULL) {
9925 printf("memorystatus_control: failed to allocate jetsam reason\n");
9926 }
9927
9928 error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled, jetsam_reason) ? 0 : EINVAL;
9929 break;
9930 case MEMORYSTATUS_CMD_TEST_JETSAM_SORT:
9931 error = memorystatus_cmd_test_jetsam_sort(args->pid, (int32_t)args->flags);
9932 break;
9933 #if CONFIG_JETSAM
9934 case MEMORYSTATUS_CMD_SET_JETSAM_PANIC_BITS:
9935 error = memorystatus_cmd_set_panic_bits(args->buffer, args->buffersize);
9936 break;
9937 #endif /* CONFIG_JETSAM */
9938 #else /* DEVELOPMENT || DEBUG */
9939 #pragma unused(jetsam_reason)
9940 #endif /* DEVELOPMENT || DEBUG */
9941 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_ENABLE:
9942 if (memorystatus_aggressive_jetsam_lenient_allowed == FALSE) {
9943 #if DEVELOPMENT || DEBUG
9944 printf("Enabling Lenient Mode\n");
9945 #endif /* DEVELOPMENT || DEBUG */
9946
9947 memorystatus_aggressive_jetsam_lenient_allowed = TRUE;
9948 memorystatus_aggressive_jetsam_lenient = TRUE;
9949 error = 0;
9950 }
9951 break;
9952 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_DISABLE:
9953 #if DEVELOPMENT || DEBUG
9954 printf("Disabling Lenient mode\n");
9955 #endif /* DEVELOPMENT || DEBUG */
9956 memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
9957 memorystatus_aggressive_jetsam_lenient = FALSE;
9958 error = 0;
9959 break;
9960 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE:
9961 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE:
9962 error = memorystatus_low_mem_privileged_listener(args->command);
9963 break;
9964
9965 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE:
9966 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE:
9967 error = memorystatus_update_inactive_jetsam_priority_band(args->pid, args->command, JETSAM_PRIORITY_ELEVATED_INACTIVE, args->flags ? TRUE : FALSE);
9968 break;
9969 case MEMORYSTATUS_CMD_SET_PROCESS_IS_MANAGED:
9970 error = memorystatus_set_process_is_managed(args->pid, args->flags);
9971 break;
9972
9973 case MEMORYSTATUS_CMD_GET_PROCESS_IS_MANAGED:
9974 error = memorystatus_get_process_is_managed(args->pid, ret);
9975 break;
9976
9977 case MEMORYSTATUS_CMD_SET_PROCESS_IS_FREEZABLE:
9978 error = memorystatus_set_process_is_freezable(args->pid, args->flags ? TRUE : FALSE);
9979 break;
9980
9981 case MEMORYSTATUS_CMD_GET_PROCESS_IS_FREEZABLE:
9982 error = memorystatus_get_process_is_freezable(args->pid, ret);
9983 break;
9984
9985 #if CONFIG_FREEZE
9986 #if DEVELOPMENT || DEBUG
9987 case MEMORYSTATUS_CMD_FREEZER_CONTROL:
9988 error = memorystatus_freezer_control(args->flags, args->buffer, args->buffersize, ret);
9989 break;
9990 #endif /* DEVELOPMENT || DEBUG */
9991 #endif /* CONFIG_FREEZE */
9992
9993 default:
9994 break;
9995 }
9996
9997 out:
9998 return error;
9999 }
10000
10001
10002 static int
10003 filt_memorystatusattach(struct knote *kn, __unused struct kevent_internal_s *kev)
10004 {
10005 int error;
10006
10007 kn->kn_flags |= EV_CLEAR;
10008 error = memorystatus_knote_register(kn);
10009 if (error) {
10010 kn->kn_flags = EV_ERROR;
10011 kn->kn_data = error;
10012 }
10013 return 0;
10014 }
10015
10016 static void
10017 filt_memorystatusdetach(struct knote *kn)
10018 {
10019 memorystatus_knote_unregister(kn);
10020 }
10021
10022 static int
10023 filt_memorystatus(struct knote *kn __unused, long hint)
10024 {
10025 if (hint) {
10026 switch (hint) {
10027 case kMemorystatusNoPressure:
10028 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
10029 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
10030 }
10031 break;
10032 case kMemorystatusPressure:
10033 if (memorystatus_vm_pressure_level == kVMPressureWarning || memorystatus_vm_pressure_level == kVMPressureUrgent) {
10034 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
10035 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
10036 }
10037 } else if (memorystatus_vm_pressure_level == kVMPressureCritical) {
10038 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
10039 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
10040 }
10041 }
10042 break;
10043 case kMemorystatusLowSwap:
10044 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_LOW_SWAP) {
10045 kn->kn_fflags = NOTE_MEMORYSTATUS_LOW_SWAP;
10046 }
10047 break;
10048
10049 case kMemorystatusProcLimitWarn:
10050 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
10051 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
10052 }
10053 break;
10054
10055 case kMemorystatusProcLimitCritical:
10056 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
10057 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
10058 }
10059 break;
10060
10061 default:
10062 break;
10063 }
10064 }
10065
10066 #if 0
10067 if (kn->kn_fflags != 0) {
10068 proc_t knote_proc = knote_get_kq(kn)->kq_p;
10069 pid_t knote_pid = knote_proc->p_pid;
10070
10071 printf("filt_memorystatus: sending kn 0x%lx (event 0x%x) for pid (%d)\n",
10072 (unsigned long)kn, kn->kn_fflags, knote_pid);
10073 }
10074 #endif
10075
10076 return kn->kn_fflags != 0;
10077 }
10078
10079 static int
10080 filt_memorystatustouch(struct knote *kn, struct kevent_internal_s *kev)
10081 {
10082 int res;
10083 int prev_kn_sfflags = 0;
10084
10085 memorystatus_klist_lock();
10086
10087 /*
10088 * copy in new kevent settings
10089 * (saving the "desired" data and fflags).
10090 */
10091
10092 prev_kn_sfflags = kn->kn_sfflags;
10093 kn->kn_sfflags = (kev->fflags & EVFILT_MEMORYSTATUS_ALL_MASK);
10094
10095 #if !CONFIG_EMBEDDED
10096 /*
10097 * Only on desktop do we restrict notifications to
10098 * one per active/inactive state (soft limits only).
10099 */
10100 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
10101 /*
10102 * Is there previous state to preserve?
10103 */
10104 if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
10105 /*
10106 * This knote was previously interested in proc_limit_warn,
10107 * so yes, preserve previous state.
10108 */
10109 if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE) {
10110 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
10111 }
10112 if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE) {
10113 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
10114 }
10115 } else {
10116 /*
10117 * This knote was not previously interested in proc_limit_warn,
10118 * but it is now. Set both states.
10119 */
10120 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
10121 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
10122 }
10123 }
10124
10125 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
10126 /*
10127 * Is there previous state to preserve?
10128 */
10129 if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
10130 /*
10131 * This knote was previously interested in proc_limit_critical,
10132 * so yes, preserve previous state.
10133 */
10134 if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE) {
10135 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
10136 }
10137 if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE) {
10138 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
10139 }
10140 } else {
10141 /*
10142 * This knote was not previously interested in proc_limit_critical,
10143 * but it is now. Set both states.
10144 */
10145 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
10146 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
10147 }
10148 }
10149 #endif /* !CONFIG_EMBEDDED */
10150
10151 /*
10152 * reset the output flags based on a
10153 * combination of the old events and
10154 * the new desired event list.
10155 */
10156 //kn->kn_fflags &= kn->kn_sfflags;
10157
10158 res = (kn->kn_fflags != 0);
10159
10160 memorystatus_klist_unlock();
10161
10162 return res;
10163 }
10164
10165 static int
10166 filt_memorystatusprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev)
10167 {
10168 #pragma unused(data)
10169 int res;
10170
10171 memorystatus_klist_lock();
10172 res = (kn->kn_fflags != 0);
10173 if (res) {
10174 *kev = kn->kn_kevent;
10175 kn->kn_flags |= EV_CLEAR; /* automatic */
10176 kn->kn_fflags = 0;
10177 kn->kn_data = 0;
10178 }
10179 memorystatus_klist_unlock();
10180
10181 return res;
10182 }
10183
10184 static void
10185 memorystatus_klist_lock(void)
10186 {
10187 lck_mtx_lock(&memorystatus_klist_mutex);
10188 }
10189
10190 static void
10191 memorystatus_klist_unlock(void)
10192 {
10193 lck_mtx_unlock(&memorystatus_klist_mutex);
10194 }
10195
10196 void
10197 memorystatus_kevent_init(lck_grp_t *grp, lck_attr_t *attr)
10198 {
10199 lck_mtx_init(&memorystatus_klist_mutex, grp, attr);
10200 klist_init(&memorystatus_klist);
10201 }
10202
10203 int
10204 memorystatus_knote_register(struct knote *kn)
10205 {
10206 int error = 0;
10207
10208 memorystatus_klist_lock();
10209
10210 /*
10211 * Support only userspace visible flags.
10212 */
10213 if ((kn->kn_sfflags & EVFILT_MEMORYSTATUS_ALL_MASK) == (unsigned int) kn->kn_sfflags) {
10214 #if !CONFIG_EMBEDDED
10215 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
10216 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
10217 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
10218 }
10219
10220 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
10221 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
10222 kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
10223 }
10224 #endif /* !CONFIG_EMBEDDED */
10225
10226 KNOTE_ATTACH(&memorystatus_klist, kn);
10227 } else {
10228 error = ENOTSUP;
10229 }
10230
10231 memorystatus_klist_unlock();
10232
10233 return error;
10234 }
10235
10236 void
10237 memorystatus_knote_unregister(struct knote *kn __unused)
10238 {
10239 memorystatus_klist_lock();
10240 KNOTE_DETACH(&memorystatus_klist, kn);
10241 memorystatus_klist_unlock();
10242 }
10243
10244
10245 #if 0
10246 #if CONFIG_JETSAM && VM_PRESSURE_EVENTS
10247 static boolean_t
10248 memorystatus_issue_pressure_kevent(boolean_t pressured)
10249 {
10250 memorystatus_klist_lock();
10251 KNOTE(&memorystatus_klist, pressured ? kMemorystatusPressure : kMemorystatusNoPressure);
10252 memorystatus_klist_unlock();
10253 return TRUE;
10254 }
10255 #endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
10256 #endif /* 0 */
10257
10258 /* Coalition support */
10259
10260 /* sorting info for a particular priority bucket */
10261 typedef struct memstat_sort_info {
10262 coalition_t msi_coal;
10263 uint64_t msi_page_count;
10264 pid_t msi_pid;
10265 int msi_ntasks;
10266 } memstat_sort_info_t;
10267
10268 /*
10269 * qsort from smallest page count to largest page count
10270 *
10271 * return < 0 for a < b
10272 * 0 for a == b
10273 * > 0 for a > b
10274 */
10275 static int
10276 memstat_asc_cmp(const void *a, const void *b)
10277 {
10278 const memstat_sort_info_t *msA = (const memstat_sort_info_t *)a;
10279 const memstat_sort_info_t *msB = (const memstat_sort_info_t *)b;
10280
10281 return (int)((uint64_t)msA->msi_page_count - (uint64_t)msB->msi_page_count);
10282 }
10283
10284 /*
10285 * Return the number of pids rearranged during this sort.
10286 */
10287 static int
10288 memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order)
10289 {
10290 #define MAX_SORT_PIDS 80
10291 #define MAX_COAL_LEADERS 10
10292
10293 unsigned int b = bucket_index;
10294 int nleaders = 0;
10295 int ntasks = 0;
10296 proc_t p = NULL;
10297 coalition_t coal = COALITION_NULL;
10298 int pids_moved = 0;
10299 int total_pids_moved = 0;
10300 int i;
10301
10302 /*
10303 * The system is typically under memory pressure when in this
10304 * path, hence, we want to avoid dynamic memory allocation.
10305 */
10306 memstat_sort_info_t leaders[MAX_COAL_LEADERS];
10307 pid_t pid_list[MAX_SORT_PIDS];
10308
10309 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
10310 return 0;
10311 }
10312
10313 /*
10314 * Clear the array that holds coalition leader information
10315 */
10316 for (i = 0; i < MAX_COAL_LEADERS; i++) {
10317 leaders[i].msi_coal = COALITION_NULL;
10318 leaders[i].msi_page_count = 0; /* will hold total coalition page count */
10319 leaders[i].msi_pid = 0; /* will hold coalition leader pid */
10320 leaders[i].msi_ntasks = 0; /* will hold the number of tasks in a coalition */
10321 }
10322
10323 p = memorystatus_get_first_proc_locked(&b, FALSE);
10324 while (p) {
10325 if (coalition_is_leader(p->task, COALITION_TYPE_JETSAM, &coal)) {
10326 if (nleaders < MAX_COAL_LEADERS) {
10327 int coal_ntasks = 0;
10328 uint64_t coal_page_count = coalition_get_page_count(coal, &coal_ntasks);
10329 leaders[nleaders].msi_coal = coal;
10330 leaders[nleaders].msi_page_count = coal_page_count;
10331 leaders[nleaders].msi_pid = p->p_pid; /* the coalition leader */
10332 leaders[nleaders].msi_ntasks = coal_ntasks;
10333 nleaders++;
10334 } else {
10335 /*
10336 * We've hit MAX_COAL_LEADERS meaning we can handle no more coalitions.
10337 * Abandoned coalitions will linger at the tail of the priority band
10338 * when this sort session ends.
10339 * TODO: should this be an assert?
10340 */
10341 printf("%s: WARNING: more than %d leaders in priority band [%d]\n",
10342 __FUNCTION__, MAX_COAL_LEADERS, bucket_index);
10343 break;
10344 }
10345 }
10346 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
10347 }
10348
10349 if (nleaders == 0) {
10350 /* Nothing to sort */
10351 return 0;
10352 }
10353
10354 /*
10355 * Sort the coalition leader array, from smallest coalition page count
10356 * to largest coalition page count. When inserted in the priority bucket,
10357 * smallest coalition is handled first, resulting in the last to be jetsammed.
10358 */
10359 if (nleaders > 1) {
10360 qsort(leaders, nleaders, sizeof(memstat_sort_info_t), memstat_asc_cmp);
10361 }
10362
10363 #if 0
10364 for (i = 0; i < nleaders; i++) {
10365 printf("%s: coal_leader[%d of %d] pid[%d] pages[%llu] ntasks[%d]\n",
10366 __FUNCTION__, i, nleaders, leaders[i].msi_pid, leaders[i].msi_page_count,
10367 leaders[i].msi_ntasks);
10368 }
10369 #endif
10370
10371 /*
10372 * During coalition sorting, processes in a priority band are rearranged
10373 * by being re-inserted at the head of the queue. So, when handling a
10374 * list, the first process that gets moved to the head of the queue,
10375 * ultimately gets pushed toward the queue tail, and hence, jetsams last.
10376 *
10377 * So, for example, the coalition leader is expected to jetsam last,
10378 * after its coalition members. Therefore, the coalition leader is
10379 * inserted at the head of the queue first.
10380 *
10381 * After processing a coalition, the jetsam order is as follows:
10382 * undefs(jetsam first), extensions, xpc services, leader(jetsam last)
10383 */
10384
10385 /*
10386 * Coalition members are rearranged in the priority bucket here,
10387 * based on their coalition role.
10388 */
10389 total_pids_moved = 0;
10390 for (i = 0; i < nleaders; i++) {
10391 /* a bit of bookkeeping */
10392 pids_moved = 0;
10393
10394 /* Coalition leaders are jetsammed last, so move into place first */
10395 pid_list[0] = leaders[i].msi_pid;
10396 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list, 1);
10397
10398 /* xpc services should jetsam after extensions */
10399 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_XPC,
10400 coal_sort_order, pid_list, MAX_SORT_PIDS);
10401
10402 if (ntasks > 0) {
10403 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
10404 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
10405 }
10406
10407 /* extensions should jetsam after unmarked processes */
10408 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_EXT,
10409 coal_sort_order, pid_list, MAX_SORT_PIDS);
10410
10411 if (ntasks > 0) {
10412 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
10413 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
10414 }
10415
10416 /* undefined coalition members should be the first to jetsam */
10417 ntasks = coalition_get_pid_list(leaders[i].msi_coal, COALITION_ROLEMASK_UNDEF,
10418 coal_sort_order, pid_list, MAX_SORT_PIDS);
10419
10420 if (ntasks > 0) {
10421 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
10422 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
10423 }
10424
10425 #if 0
10426 if (pids_moved == leaders[i].msi_ntasks) {
10427 /*
10428 * All the pids in the coalition were found in this band.
10429 */
10430 printf("%s: pids_moved[%d] equal total coalition ntasks[%d] \n", __FUNCTION__,
10431 pids_moved, leaders[i].msi_ntasks);
10432 } else if (pids_moved > leaders[i].msi_ntasks) {
10433 /*
10434 * Apparently new coalition members showed up during the sort?
10435 */
10436 printf("%s: pids_moved[%d] were greater than expected coalition ntasks[%d] \n", __FUNCTION__,
10437 pids_moved, leaders[i].msi_ntasks);
10438 } else {
10439 /*
10440 * Apparently not all the pids in the coalition were found in this band?
10441 */
10442 printf("%s: pids_moved[%d] were less than expected coalition ntasks[%d] \n", __FUNCTION__,
10443 pids_moved, leaders[i].msi_ntasks);
10444 }
10445 #endif
10446
10447 total_pids_moved += pids_moved;
10448 } /* end for */
10449
10450 return total_pids_moved;
10451 }
10452
10453
10454 /*
10455 * Traverse a list of pids, searching for each within the priority band provided.
10456 * If pid is found, move it to the front of the priority band.
10457 * Never searches outside the priority band provided.
10458 *
10459 * Input:
10460 * bucket_index - jetsam priority band.
10461 * pid_list - pointer to a list of pids.
10462 * list_sz - number of pids in the list.
10463 *
10464 * Pid list ordering is important in that,
10465 * pid_list[n] is expected to jetsam ahead of pid_list[n+1].
10466 * The sort_order is set by the coalition default.
10467 *
10468 * Return:
10469 * the number of pids found and hence moved within the priority band.
10470 */
10471 static int
10472 memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz)
10473 {
10474 memstat_bucket_t *current_bucket;
10475 int i;
10476 int found_pids = 0;
10477
10478 if ((pid_list == NULL) || (list_sz <= 0)) {
10479 return 0;
10480 }
10481
10482 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
10483 return 0;
10484 }
10485
10486 current_bucket = &memstat_bucket[bucket_index];
10487 for (i = 0; i < list_sz; i++) {
10488 unsigned int b = bucket_index;
10489 proc_t p = NULL;
10490 proc_t aProc = NULL;
10491 pid_t aPid;
10492 int list_index;
10493
10494 list_index = ((list_sz - 1) - i);
10495 aPid = pid_list[list_index];
10496
10497 /* never search beyond bucket_index provided */
10498 p = memorystatus_get_first_proc_locked(&b, FALSE);
10499 while (p) {
10500 if (p->p_pid == aPid) {
10501 aProc = p;
10502 break;
10503 }
10504 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
10505 }
10506
10507 if (aProc == NULL) {
10508 /* pid not found in this band, just skip it */
10509 continue;
10510 } else {
10511 TAILQ_REMOVE(&current_bucket->list, aProc, p_memstat_list);
10512 TAILQ_INSERT_HEAD(&current_bucket->list, aProc, p_memstat_list);
10513 found_pids++;
10514 }
10515 }
10516 return found_pids;
10517 }
10518
10519 int
10520 memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index)
10521 {
10522 int32_t i = JETSAM_PRIORITY_IDLE;
10523 int count = 0;
10524
10525 if (max_bucket_index >= MEMSTAT_BUCKET_COUNT) {
10526 return -1;
10527 }
10528
10529 while (i <= max_bucket_index) {
10530 count += memstat_bucket[i++].count;
10531 }
10532
10533 return count;
10534 }
10535
10536 int
10537 memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap)
10538 {
10539 #if !CONFIG_JETSAM
10540 if (!p || (!isApp(p)) || (p->p_memstat_state & (P_MEMSTAT_INTERNAL | P_MEMSTAT_MANAGED))) {
10541 /*
10542 * Ineligible processes OR system processes e.g. launchd.
10543 *
10544 * We also skip processes that have the P_MEMSTAT_MANAGED bit set, i.e.
10545 * they're managed by assertiond. These are iOS apps that have been ported
10546 * to macOS. assertiond might be in the process of modifying the app's
10547 * priority / memory limit - so it might have the proc_list lock, and then try
10548 * to take the task lock. Meanwhile we've entered this function with the task lock
10549 * held, and we need the proc_list lock below. So we'll deadlock with assertiond.
10550 *
10551 * It should be fine to read the P_MEMSTAT_MANAGED bit without the proc_list
10552 * lock here, since assertiond only sets this bit on process launch.
10553 */
10554 return -1;
10555 }
10556
10557 /*
10558 * For macOS only:
10559 * We would like to use memorystatus_update() here to move the processes
10560 * within the bands. Unfortunately memorystatus_update() calls
10561 * memorystatus_update_priority_locked() which uses any band transitions
10562 * as an indication to modify ledgers. For that it needs the task lock
10563 * and since we came into this function with the task lock held, we'll deadlock.
10564 *
10565 * Unfortunately we can't completely disable ledger updates because we still
10566 * need the ledger updates for a subset of processes i.e. daemons.
10567 * When all processes on all platforms support memory limits, we can simply call
10568 * memorystatus_update().
10569 *
10570 * It also has some logic to deal with 'aging' which, currently, is only applicable
10571 * on CONFIG_JETSAM configs. So, till every platform has CONFIG_JETSAM we'll need
10572 * to do this explicit band transition.
10573 */
10574
10575 memstat_bucket_t *current_bucket, *new_bucket;
10576 int32_t priority = 0;
10577
10578 proc_list_lock();
10579
10580 if (((p->p_listflag & P_LIST_EXITED) != 0) ||
10581 (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED))) {
10582 /*
10583 * If the process is on its way out OR
10584 * jetsam has alread tried and failed to kill this process,
10585 * let's skip the whole jetsam band transition.
10586 */
10587 proc_list_unlock();
10588 return 0;
10589 }
10590
10591 if (is_appnap) {
10592 current_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
10593 new_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
10594 priority = JETSAM_PRIORITY_IDLE;
10595 } else {
10596 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
10597 /*
10598 * It is possible that someone pulled this process
10599 * out of the IDLE band without updating its app-nap
10600 * parameters.
10601 */
10602 proc_list_unlock();
10603 return 0;
10604 }
10605
10606 current_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
10607 new_bucket = &memstat_bucket[p->p_memstat_requestedpriority];
10608 priority = p->p_memstat_requestedpriority;
10609 }
10610
10611 TAILQ_REMOVE(&current_bucket->list, p, p_memstat_list);
10612 current_bucket->count--;
10613
10614 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
10615 new_bucket->count++;
10616
10617 /*
10618 * Record idle start or idle delta.
10619 */
10620 if (p->p_memstat_effectivepriority == priority) {
10621 /*
10622 * This process is not transitioning between
10623 * jetsam priority buckets. Do nothing.
10624 */
10625 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
10626 uint64_t now;
10627 /*
10628 * Transitioning out of the idle priority bucket.
10629 * Record idle delta.
10630 */
10631 assert(p->p_memstat_idle_start != 0);
10632 now = mach_absolute_time();
10633 if (now > p->p_memstat_idle_start) {
10634 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
10635 }
10636 } else if (priority == JETSAM_PRIORITY_IDLE) {
10637 /*
10638 * Transitioning into the idle priority bucket.
10639 * Record idle start.
10640 */
10641 p->p_memstat_idle_start = mach_absolute_time();
10642 }
10643
10644 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CHANGE_PRIORITY), p->p_pid, priority, p->p_memstat_effectivepriority, 0, 0);
10645
10646 p->p_memstat_effectivepriority = priority;
10647
10648 proc_list_unlock();
10649
10650 return 0;
10651
10652 #else /* !CONFIG_JETSAM */
10653 #pragma unused(p)
10654 #pragma unused(is_appnap)
10655 return -1;
10656 #endif /* !CONFIG_JETSAM */
10657 }
mirror of XNU