]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/kern_memorystatus.c
projects / apple / xnu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xnu-3789.1.32.tar.gz
[apple/xnu.git] / bsd / kern / kern_memorystatus.c
1 /*
2 * Copyright (c) 2006 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 *
28 */
29
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
40 #include <IOKit/IOBSD.h>
41
42 #include <libkern/libkern.h>
43 #include <mach/coalition.h>
44 #include <mach/mach_time.h>
45 #include <mach/task.h>
46 #include <mach/host_priv.h>
47 #include <mach/mach_host.h>
48 #include <pexpert/pexpert.h>
49 #include <sys/coalition.h>
50 #include <sys/kern_event.h>
51 #include <sys/proc.h>
52 #include <sys/proc_info.h>
53 #include <sys/reason.h>
54 #include <sys/signal.h>
55 #include <sys/signalvar.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysproto.h>
58 #include <sys/wait.h>
59 #include <sys/tree.h>
60 #include <sys/priv.h>
61 #include <vm/vm_pageout.h>
62 #include <vm/vm_protos.h>
63
64 #if CONFIG_FREEZE
65 #include <vm/vm_map.h>
66 #endif /* CONFIG_FREEZE */
67
68 #include <sys/kern_memorystatus.h>
69
70 #include <mach/machine/sdt.h>
71
72 /* For logging clarity */
73 static const char *jetsam_kill_cause_name[] = {
74 "" ,
75 "jettisoned" , /* kMemorystatusKilled */
76 "highwater" , /* kMemorystatusKilledHiwat */
77 "vnode-limit" , /* kMemorystatusKilledVnodes */
78 "vm-pageshortage" , /* kMemorystatusKilledVMPageShortage */
79 "vm-thrashing" , /* kMemorystatusKilledVMThrashing */
80 "fc-thrashing" , /* kMemorystatusKilledFCThrashing */
81 "per-process-limit" , /* kMemorystatusKilledPerProcessLimit */
82 "diagnostic" , /* kMemorystatusKilledDiagnostic */
83 "idle-exit" , /* kMemorystatusKilledIdleExit */
84 };
85
86 #if CONFIG_JETSAM
87 /* Does cause indicate vm or fc thrashing? */
88 static boolean_t
89 is_thrashing(unsigned cause)
90 {
91 switch (cause) {
92 case kMemorystatusKilledVMThrashing:
93 case kMemorystatusKilledFCThrashing:
94 return TRUE;
95 default:
96 return FALSE;
97 }
98 }
99
100 /* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
101 extern void vm_thrashing_jetsam_done(void);
102 #endif /* CONFIG_JETSAM */
103
104 /* These are very verbose printfs(), enable with
105 * MEMORYSTATUS_DEBUG_LOG
106 */
107 #if MEMORYSTATUS_DEBUG_LOG
108 #define MEMORYSTATUS_DEBUG(cond, format, ...) \
109 do { \
110 if (cond) { printf(format, ##__VA_ARGS__); } \
111 } while(0)
112 #else
113 #define MEMORYSTATUS_DEBUG(cond, format, ...)
114 #endif
115
116 /*
117 * Active / Inactive limit support
118 * proc list must be locked
119 *
120 * The SET_*** macros are used to initialize a limit
121 * for the first time.
122 *
123 * The CACHE_*** macros are use to cache the limit that will
124 * soon be in effect down in the ledgers.
125 */
126
127 #define SET_ACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
128 MACRO_BEGIN \
129 (p)->p_memstat_memlimit_active = (limit); \
130 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED; \
131 if (is_fatal) { \
132 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
133 } else { \
134 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
135 } \
136 MACRO_END
137
138 #define SET_INACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
139 MACRO_BEGIN \
140 (p)->p_memstat_memlimit_inactive = (limit); \
141 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED; \
142 if (is_fatal) { \
143 (p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
144 } else { \
145 (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
146 } \
147 MACRO_END
148
149 #define CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception) \
150 MACRO_BEGIN \
151 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_active; \
152 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) { \
153 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
154 } else { \
155 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
156 } \
157 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED) { \
158 trigger_exception = FALSE; \
159 } else { \
160 trigger_exception = TRUE; \
161 } \
162 MACRO_END
163
164 #define CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception) \
165 MACRO_BEGIN \
166 (p)->p_memstat_memlimit = (p)->p_memstat_memlimit_inactive; \
167 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) { \
168 (p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
169 } else { \
170 (p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
171 } \
172 if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED) { \
173 trigger_exception = FALSE; \
174 } else { \
175 trigger_exception = TRUE; \
176 } \
177 MACRO_END
178
179
180 /* General tunables */
181
182 unsigned long delta_percentage = 5;
183 unsigned long critical_threshold_percentage = 5;
184 unsigned long idle_offset_percentage = 5;
185 unsigned long pressure_threshold_percentage = 15;
186 unsigned long freeze_threshold_percentage = 50;
187 unsigned long policy_more_free_offset_percentage = 5;
188
189 /* General memorystatus stuff */
190
191 struct klist memorystatus_klist;
192 static lck_mtx_t memorystatus_klist_mutex;
193
194 static void memorystatus_klist_lock(void);
195 static void memorystatus_klist_unlock(void);
196
197 static uint64_t memorystatus_sysprocs_idle_delay_time = 0;
198 static uint64_t memorystatus_apps_idle_delay_time = 0;
199
200 /*
201 * Memorystatus kevents
202 */
203
204 static int filt_memorystatusattach(struct knote *kn);
205 static void filt_memorystatusdetach(struct knote *kn);
206 static int filt_memorystatus(struct knote *kn, long hint);
207 static int filt_memorystatustouch(struct knote *kn, struct kevent_internal_s *kev);
208 static int filt_memorystatusprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev);
209
210 struct filterops memorystatus_filtops = {
211 .f_attach = filt_memorystatusattach,
212 .f_detach = filt_memorystatusdetach,
213 .f_event = filt_memorystatus,
214 .f_touch = filt_memorystatustouch,
215 .f_process = filt_memorystatusprocess,
216 };
217
218 enum {
219 kMemorystatusNoPressure = 0x1,
220 kMemorystatusPressure = 0x2,
221 kMemorystatusLowSwap = 0x4,
222 kMemorystatusProcLimitWarn = 0x8,
223 kMemorystatusProcLimitCritical = 0x10
224 };
225
226 /* Idle guard handling */
227
228 static int32_t memorystatus_scheduled_idle_demotions_sysprocs = 0;
229 static int32_t memorystatus_scheduled_idle_demotions_apps = 0;
230
231 static thread_call_t memorystatus_idle_demotion_call;
232
233 static void memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2);
234 static void memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state);
235 static void memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clean_state);
236 static void memorystatus_reschedule_idle_demotion_locked(void);
237
238 static void memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check);
239
240 vm_pressure_level_t convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
241
242 boolean_t is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
243 void memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear);
244 void memorystatus_send_low_swap_note(void);
245
246 int memorystatus_wakeup = 0;
247
248 unsigned int memorystatus_level = 0;
249
250 static int memorystatus_list_count = 0;
251
252 #define MEMSTAT_BUCKET_COUNT (JETSAM_PRIORITY_MAX + 1)
253
254 typedef struct memstat_bucket {
255 TAILQ_HEAD(, proc) list;
256 int count;
257 } memstat_bucket_t;
258
259 memstat_bucket_t memstat_bucket[MEMSTAT_BUCKET_COUNT];
260
261 uint64_t memstat_idle_demotion_deadline = 0;
262
263 int system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
264 int applications_aging_band = JETSAM_PRIORITY_IDLE;
265
266 #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)))
267 #define isApp(p) (! (p->p_memstat_dirty & P_DIRTY_TRACK))
268 #define isSysProc(p) ((p->p_memstat_dirty & P_DIRTY_TRACK))
269
270 #define kJetsamAgingPolicyNone (0)
271 #define kJetsamAgingPolicyLegacy (1)
272 #define kJetsamAgingPolicySysProcsReclaimedFirst (2)
273 #define kJetsamAgingPolicyAppsReclaimedFirst (3)
274 #define kJetsamAgingPolicyMax kJetsamAgingPolicyAppsReclaimedFirst
275
276 unsigned int jetsam_aging_policy = kJetsamAgingPolicyLegacy;
277
278 extern int corpse_for_fatal_memkill;
279 extern unsigned long total_corpses_count;
280 extern void task_purge_all_corpses(void);
281
282 #if 0
283
284 /* Keeping around for future use if we need a utility that can do this OR an app that needs a dynamic adjustment. */
285
286 static int
287 sysctl_set_jetsam_aging_policy SYSCTL_HANDLER_ARGS
288 {
289 #pragma unused(oidp, arg1, arg2)
290
291 int error = 0, val = 0;
292 memstat_bucket_t *old_bucket = 0;
293 int old_system_procs_aging_band = 0, new_system_procs_aging_band = 0;
294 int old_applications_aging_band = 0, new_applications_aging_band = 0;
295 proc_t p = NULL, next_proc = NULL;
296
297
298 error = sysctl_io_number(req, jetsam_aging_policy, sizeof(int), &val, NULL);
299 if (error || !req->newptr) {
300 return (error);
301 }
302
303 if ((val < 0) || (val > kJetsamAgingPolicyMax)) {
304 printf("jetsam: ordering policy sysctl has invalid value - %d\n", val);
305 return EINVAL;
306 }
307
308 /*
309 * We need to synchronize with any potential adding/removal from aging bands
310 * that might be in progress currently. We use the proc_list_lock() just for
311 * consistency with all the routines dealing with 'aging' processes. We need
312 * a lighterweight lock.
313 */
314 proc_list_lock();
315
316 old_system_procs_aging_band = system_procs_aging_band;
317 old_applications_aging_band = applications_aging_band;
318
319 switch (val) {
320
321 case kJetsamAgingPolicyNone:
322 new_system_procs_aging_band = JETSAM_PRIORITY_IDLE;
323 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
324 break;
325
326 case kJetsamAgingPolicyLegacy:
327 /*
328 * Legacy behavior where some daemons get a 10s protection once and only before the first clean->dirty->clean transition before going into IDLE band.
329 */
330 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
331 new_applications_aging_band = JETSAM_PRIORITY_IDLE;
332 break;
333
334 case kJetsamAgingPolicySysProcsReclaimedFirst:
335 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
336 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
337 break;
338
339 case kJetsamAgingPolicyAppsReclaimedFirst:
340 new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
341 new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
342 break;
343
344 default:
345 break;
346 }
347
348 if (old_system_procs_aging_band && (old_system_procs_aging_band != new_system_procs_aging_band)) {
349
350 old_bucket = &memstat_bucket[old_system_procs_aging_band];
351 p = TAILQ_FIRST(&old_bucket->list);
352
353 while (p) {
354
355 next_proc = TAILQ_NEXT(p, p_memstat_list);
356
357 if (isSysProc(p)) {
358 if (new_system_procs_aging_band == JETSAM_PRIORITY_IDLE) {
359 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
360 }
361
362 memorystatus_update_priority_locked(p, new_system_procs_aging_band, false, true);
363 }
364
365 p = next_proc;
366 continue;
367 }
368 }
369
370 if (old_applications_aging_band && (old_applications_aging_band != new_applications_aging_band)) {
371
372 old_bucket = &memstat_bucket[old_applications_aging_band];
373 p = TAILQ_FIRST(&old_bucket->list);
374
375 while (p) {
376
377 next_proc = TAILQ_NEXT(p, p_memstat_list);
378
379 if (isApp(p)) {
380 if (new_applications_aging_band == JETSAM_PRIORITY_IDLE) {
381 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
382 }
383
384 memorystatus_update_priority_locked(p, new_applications_aging_band, false, true);
385 }
386
387 p = next_proc;
388 continue;
389 }
390 }
391
392 jetsam_aging_policy = val;
393 system_procs_aging_band = new_system_procs_aging_band;
394 applications_aging_band = new_applications_aging_band;
395
396 proc_list_unlock();
397
398 return (0);
399 }
400
401 SYSCTL_PROC(_kern, OID_AUTO, set_jetsam_aging_policy, CTLTYPE_INT|CTLFLAG_RW,
402 0, 0, sysctl_set_jetsam_aging_policy, "I", "Jetsam Aging Policy");
403 #endif /*0*/
404
405 static int
406 sysctl_jetsam_set_sysprocs_idle_delay_time SYSCTL_HANDLER_ARGS
407 {
408 #pragma unused(oidp, arg1, arg2)
409
410 int error = 0, val = 0, old_time_in_secs = 0;
411 uint64_t old_time_in_ns = 0;
412
413 absolutetime_to_nanoseconds(memorystatus_sysprocs_idle_delay_time, &old_time_in_ns);
414 old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
415
416 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
417 if (error || !req->newptr) {
418 return (error);
419 }
420
421 if ((val < 0) || (val > INT32_MAX)) {
422 printf("jetsam: new idle delay interval has invalid value.\n");
423 return EINVAL;
424 }
425
426 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
427
428 return(0);
429 }
430
431 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_sysprocs_idle_delay_time, CTLTYPE_INT|CTLFLAG_RW,
432 0, 0, sysctl_jetsam_set_sysprocs_idle_delay_time, "I", "Aging window for system processes");
433
434
435 static int
436 sysctl_jetsam_set_apps_idle_delay_time SYSCTL_HANDLER_ARGS
437 {
438 #pragma unused(oidp, arg1, arg2)
439
440 int error = 0, val = 0, old_time_in_secs = 0;
441 uint64_t old_time_in_ns = 0;
442
443 absolutetime_to_nanoseconds(memorystatus_apps_idle_delay_time, &old_time_in_ns);
444 old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
445
446 error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
447 if (error || !req->newptr) {
448 return (error);
449 }
450
451 if ((val < 0) || (val > INT32_MAX)) {
452 printf("jetsam: new idle delay interval has invalid value.\n");
453 return EINVAL;
454 }
455
456 nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
457
458 return(0);
459 }
460
461 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_apps_idle_delay_time, CTLTYPE_INT|CTLFLAG_RW,
462 0, 0, sysctl_jetsam_set_apps_idle_delay_time, "I", "Aging window for applications");
463
464 SYSCTL_INT(_kern, OID_AUTO, jetsam_aging_policy, CTLTYPE_INT|CTLFLAG_RD, &jetsam_aging_policy, 0, "");
465
466 static unsigned int memorystatus_dirty_count = 0;
467
468 SYSCTL_INT(_kern, OID_AUTO, max_task_pmem, CTLFLAG_RD|CTLFLAG_LOCKED|CTLFLAG_MASKED, &max_task_footprint_mb, 0, "");
469
470
471 int
472 memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
473 {
474 user_addr_t level = 0;
475
476 level = args->level;
477
478 if (copyout(&memorystatus_level, level, sizeof(memorystatus_level)) != 0) {
479 return EFAULT;
480 }
481
482 return 0;
483 }
484
485 static proc_t memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search);
486 static proc_t memorystatus_get_next_proc_locked(unsigned int *bucket_index, proc_t p, boolean_t search);
487
488 static void memorystatus_thread(void *param __unused, wait_result_t wr __unused);
489
490 /* Memory Limits */
491
492 static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. */
493
494 static boolean_t proc_jetsam_state_is_active_locked(proc_t);
495 static boolean_t memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
496 static boolean_t memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
497
498
499 /* Jetsam */
500
501 #if CONFIG_JETSAM
502
503 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);
504
505 static int memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
506
507 static int memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry);
508
509 static int memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
510
511 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);
512
513 int proc_get_memstat_priority(proc_t, boolean_t);
514
515 static boolean_t memorystatus_idle_snapshot = 0;
516
517 unsigned int memorystatus_delta = 0;
518
519 static unsigned int memorystatus_available_pages_critical_base = 0;
520 //static unsigned int memorystatus_last_foreground_pressure_pages = (unsigned int)-1;
521 static unsigned int memorystatus_available_pages_critical_idle_offset = 0;
522
523 /* Jetsam Loop Detection */
524 static boolean_t memorystatus_jld_enabled = TRUE; /* Enables jetsam loop detection on all devices */
525 static uint32_t memorystatus_jld_eval_period_msecs = 0; /* Init pass sets this based on device memory size */
526 static int memorystatus_jld_eval_aggressive_count = 3; /* Raise the priority max after 'n' aggressive loops */
527 static int memorystatus_jld_eval_aggressive_priority_band_max = 15; /* Kill aggressively up through this band */
528
529 /*
530 * A FG app can request that the aggressive jetsam mechanism display some leniency in the FG band. This 'lenient' mode is described as:
531 * --- 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.
532 *
533 * RESTRICTIONS:
534 * - Such a request is respected/acknowledged only once while that 'requesting' app is in the FG band i.e. if aggressive jetsam was
535 * needed and the 'lenient' mode was deployed then that's it for this special mode while the app is in the FG band.
536 *
537 * - 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.
538 *
539 * - Also, the transition of the 'requesting' app away from the FG band will void this special behavior.
540 */
541
542 #define AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD 25
543 boolean_t memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
544 boolean_t memorystatus_aggressive_jetsam_lenient = FALSE;
545
546 #if DEVELOPMENT || DEBUG
547 /*
548 * Jetsam Loop Detection tunables.
549 */
550
551 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_period_msecs, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jld_eval_period_msecs, 0, "");
552 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_count, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_count, 0, "");
553 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_priority_band_max, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_priority_band_max, 0, "");
554 #endif /* DEVELOPMENT || DEBUG */
555
556 #if DEVELOPMENT || DEBUG
557 static unsigned int memorystatus_jetsam_panic_debug = 0;
558 static unsigned int memorystatus_jetsam_policy_offset_pages_diagnostic = 0;
559 #endif
560
561 static unsigned int memorystatus_jetsam_policy = kPolicyDefault;
562 static unsigned int memorystatus_thread_wasted_wakeup = 0;
563
564 static uint32_t kill_under_pressure_cause = 0;
565
566 /*
567 * default jetsam snapshot support
568 */
569 static memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot;
570 #define memorystatus_jetsam_snapshot_list memorystatus_jetsam_snapshot->entries
571 static unsigned int memorystatus_jetsam_snapshot_count = 0;
572 static unsigned int memorystatus_jetsam_snapshot_max = 0;
573 static uint64_t memorystatus_jetsam_snapshot_last_timestamp = 0;
574 static uint64_t memorystatus_jetsam_snapshot_timeout = 0;
575 #define JETSAM_SNAPSHOT_TIMEOUT_SECS 30
576
577 /*
578 * snapshot support for memstats collected at boot.
579 */
580 static memorystatus_jetsam_snapshot_t memorystatus_at_boot_snapshot;
581
582 static void memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count);
583 static boolean_t memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount);
584 static void memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime);
585
586 static void memorystatus_clear_errors(void);
587 static void memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages);
588 static void memorystatus_get_task_phys_footprint_page_counts(task_t task,
589 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
590 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
591 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
592 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages);
593
594 static void memorystatus_get_task_memory_region_count(task_t task, uint64_t *count);
595
596 static uint32_t memorystatus_build_state(proc_t p);
597 static void memorystatus_update_levels_locked(boolean_t critical_only);
598 //static boolean_t memorystatus_issue_pressure_kevent(boolean_t pressured);
599
600 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);
601 static boolean_t memorystatus_kill_top_process_aggressive(boolean_t any, uint32_t cause, os_reason_t jetsam_reason, int aggr_count, int32_t priority_max, uint32_t *errors);
602 static boolean_t memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, int aggr_count, uint32_t *errors);
603 static boolean_t memorystatus_kill_hiwat_proc(uint32_t *errors);
604
605 static boolean_t memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause);
606
607 /* Priority Band Sorting Routines */
608 static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order);
609 static int memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order);
610 static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index);
611 static int memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz);
612
613 /* qsort routines */
614 typedef int (*cmpfunc_t)(const void *a, const void *b);
615 extern void qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
616 static int memstat_asc_cmp(const void *a, const void *b);
617
618 #endif /* CONFIG_JETSAM */
619
620 /* VM pressure */
621
622 extern unsigned int vm_page_free_count;
623 extern unsigned int vm_page_active_count;
624 extern unsigned int vm_page_inactive_count;
625 extern unsigned int vm_page_throttled_count;
626 extern unsigned int vm_page_purgeable_count;
627 extern unsigned int vm_page_wire_count;
628 #if CONFIG_SECLUDED_MEMORY
629 extern unsigned int vm_page_secluded_count;
630 #endif /* CONFIG_SECLUDED_MEMORY */
631
632 #if VM_PRESSURE_EVENTS
633
634 boolean_t memorystatus_warn_process(pid_t pid, boolean_t exceeded);
635
636 vm_pressure_level_t memorystatus_vm_pressure_level = kVMPressureNormal;
637
638 #if CONFIG_MEMORYSTATUS
639 unsigned int memorystatus_available_pages = (unsigned int)-1;
640 unsigned int memorystatus_available_pages_pressure = 0;
641 unsigned int memorystatus_available_pages_critical = 0;
642 unsigned int memorystatus_frozen_count = 0;
643 unsigned int memorystatus_suspended_count = 0;
644 unsigned int memorystatus_policy_more_free_offset_pages = 0;
645
646 /*
647 * We use this flag to signal if we have any HWM offenders
648 * on the system. This way we can reduce the number of wakeups
649 * of the memorystatus_thread when the system is between the
650 * "pressure" and "critical" threshold.
651 *
652 * The (re-)setting of this variable is done without any locks
653 * or synchronization simply because it is not possible (currently)
654 * to keep track of HWM offenders that drop down below their memory
655 * limit and/or exit. So, we choose to burn a couple of wasted wakeups
656 * by allowing the unguarded modification of this variable.
657 */
658 boolean_t memorystatus_hwm_candidates = 0;
659
660 static int memorystatus_send_note(int event_code, void *data, size_t data_length);
661 #endif /* CONFIG_MEMORYSTATUS */
662
663 #endif /* VM_PRESSURE_EVENTS */
664
665
666 #if DEVELOPMENT || DEBUG
667
668 lck_grp_attr_t *disconnect_page_mappings_lck_grp_attr;
669 lck_grp_t *disconnect_page_mappings_lck_grp;
670 static lck_mtx_t disconnect_page_mappings_mutex;
671
672 #endif
673
674
675 /* Freeze */
676
677 #if CONFIG_FREEZE
678
679 boolean_t memorystatus_freeze_enabled = FALSE;
680 int memorystatus_freeze_wakeup = 0;
681
682 lck_grp_attr_t *freezer_lck_grp_attr;
683 lck_grp_t *freezer_lck_grp;
684 static lck_mtx_t freezer_mutex;
685
686 static inline boolean_t memorystatus_can_freeze_processes(void);
687 static boolean_t memorystatus_can_freeze(boolean_t *memorystatus_freeze_swap_low);
688
689 static void memorystatus_freeze_thread(void *param __unused, wait_result_t wr __unused);
690
691 /* Thresholds */
692 static unsigned int memorystatus_freeze_threshold = 0;
693
694 static unsigned int memorystatus_freeze_pages_min = 0;
695 static unsigned int memorystatus_freeze_pages_max = 0;
696
697 static unsigned int memorystatus_freeze_suspended_threshold = FREEZE_SUSPENDED_THRESHOLD_DEFAULT;
698
699 static unsigned int memorystatus_freeze_daily_mb_max = FREEZE_DAILY_MB_MAX_DEFAULT;
700
701 /* Stats */
702 static uint64_t memorystatus_freeze_count = 0;
703 static uint64_t memorystatus_freeze_pageouts = 0;
704
705 /* Throttling */
706 static throttle_interval_t throttle_intervals[] = {
707 { 60, 8, 0, 0, { 0, 0 }, FALSE }, /* 1 hour intermediate interval, 8x burst */
708 { 24 * 60, 1, 0, 0, { 0, 0 }, FALSE }, /* 24 hour long interval, no burst */
709 };
710
711 static uint64_t memorystatus_freeze_throttle_count = 0;
712
713 static unsigned int memorystatus_suspended_footprint_total = 0; /* pages */
714
715 extern uint64_t vm_swap_get_free_space(void);
716
717 static boolean_t memorystatus_freeze_update_throttle();
718
719 #endif /* CONFIG_FREEZE */
720
721 /* Debug */
722
723 extern struct knote *vm_find_knote_from_pid(pid_t, struct klist *);
724
725 #if DEVELOPMENT || DEBUG
726
727 static unsigned int memorystatus_debug_dump_this_bucket = 0;
728
729 static void
730 memorystatus_debug_dump_bucket_locked (unsigned int bucket_index)
731 {
732 proc_t p = NULL;
733 uint64_t bytes = 0;
734 int ledger_limit = 0;
735 unsigned int b = bucket_index;
736 boolean_t traverse_all_buckets = FALSE;
737
738 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
739 traverse_all_buckets = TRUE;
740 b = 0;
741 } else {
742 traverse_all_buckets = FALSE;
743 b = bucket_index;
744 }
745
746 /*
747 * footprint reported in [pages / MB ]
748 * limits reported as:
749 * L-limit proc's Ledger limit
750 * C-limit proc's Cached limit, should match Ledger
751 * A-limit proc's Active limit
752 * IA-limit proc's Inactive limit
753 * F==Fatal, NF==NonFatal
754 */
755
756 printf("memorystatus_debug_dump ***START*(PAGE_SIZE_64=%llu)**\n", PAGE_SIZE_64);
757 printf("bucket [pid] [pages / MB] [state] [EP / RP] dirty deadline [L-limit / C-limit / A-limit / IA-limit] name\n");
758 p = memorystatus_get_first_proc_locked(&b, traverse_all_buckets);
759 while (p) {
760 bytes = get_task_phys_footprint(p->task);
761 task_get_phys_footprint_limit(p->task, &ledger_limit);
762 printf("%2d [%5d] [%5lld /%3lldMB] 0x%-8x [%2d / %2d] 0x%-3x %10lld [%3d / %3d%s / %3d%s / %3d%s] %s\n",
763 b, p->p_pid,
764 (bytes / PAGE_SIZE_64), /* task's footprint converted from bytes to pages */
765 (bytes / (1024ULL * 1024ULL)), /* task's footprint converted from bytes to MB */
766 p->p_memstat_state, p->p_memstat_effectivepriority, p->p_memstat_requestedpriority, p->p_memstat_dirty, p->p_memstat_idledeadline,
767 ledger_limit,
768 p->p_memstat_memlimit,
769 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"),
770 p->p_memstat_memlimit_active,
771 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL ? "F " : "NF"),
772 p->p_memstat_memlimit_inactive,
773 (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL ? "F " : "NF"),
774 (*p->p_name ? p->p_name : "unknown"));
775 p = memorystatus_get_next_proc_locked(&b, p, traverse_all_buckets);
776 }
777 printf("memorystatus_debug_dump ***END***\n");
778 }
779
780 static int
781 sysctl_memorystatus_debug_dump_bucket SYSCTL_HANDLER_ARGS
782 {
783 #pragma unused(oidp, arg2)
784 int bucket_index = 0;
785 int error;
786 error = SYSCTL_OUT(req, arg1, sizeof(int));
787 if (error || !req->newptr) {
788 return (error);
789 }
790 error = SYSCTL_IN(req, &bucket_index, sizeof(int));
791 if (error || !req->newptr) {
792 return (error);
793 }
794 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
795 /*
796 * All jetsam buckets will be dumped.
797 */
798 } else {
799 /*
800 * Only a single bucket will be dumped.
801 */
802 }
803
804 proc_list_lock();
805 memorystatus_debug_dump_bucket_locked(bucket_index);
806 proc_list_unlock();
807 memorystatus_debug_dump_this_bucket = bucket_index;
808 return (error);
809 }
810
811 /*
812 * Debug aid to look at jetsam buckets and proc jetsam fields.
813 * Use this sysctl to act on a particular jetsam bucket.
814 * Writing the sysctl triggers the dump.
815 * Usage: sysctl kern.memorystatus_debug_dump_this_bucket=<bucket_index>
816 */
817
818 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", "");
819
820
821 /* Debug aid to aid determination of limit */
822
823 static int
824 sysctl_memorystatus_highwater_enable SYSCTL_HANDLER_ARGS
825 {
826 #pragma unused(oidp, arg2)
827 proc_t p;
828 unsigned int b = 0;
829 int error, enable = 0;
830
831 error = SYSCTL_OUT(req, arg1, sizeof(int));
832 if (error || !req->newptr) {
833 return (error);
834 }
835
836 error = SYSCTL_IN(req, &enable, sizeof(int));
837 if (error || !req->newptr) {
838 return (error);
839 }
840
841 if (!(enable == 0 || enable == 1)) {
842 return EINVAL;
843 }
844
845 proc_list_lock();
846
847 p = memorystatus_get_first_proc_locked(&b, TRUE);
848 while (p) {
849 boolean_t trigger_exception;
850
851 if (enable) {
852 /*
853 * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
854 * Background limits are described via the inactive limit slots.
855 */
856
857 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
858 CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
859 } else {
860 CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
861 }
862
863 } else {
864 /*
865 * Disabling limits does not touch the stored variants.
866 * Set the cached limit fields to system_wide defaults.
867 */
868 p->p_memstat_memlimit = -1;
869 p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
870 trigger_exception = TRUE;
871 }
872
873 /*
874 * Enforce the cached limit by writing to the ledger.
875 */
876 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit: -1, NULL, trigger_exception);
877
878 p = memorystatus_get_next_proc_locked(&b, p, TRUE);
879 }
880
881 memorystatus_highwater_enabled = enable;
882
883 proc_list_unlock();
884
885 return 0;
886
887 }
888
889 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_highwater_enabled, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_highwater_enabled, 0, sysctl_memorystatus_highwater_enable, "I", "");
890
891 #if VM_PRESSURE_EVENTS
892
893 /*
894 * This routine is used for targeted notifications
895 * regardless of system memory pressure.
896 * "memnote" is the current user.
897 */
898
899 static int
900 sysctl_memorystatus_vm_pressure_send SYSCTL_HANDLER_ARGS
901 {
902 #pragma unused(arg1, arg2)
903
904 int error = 0, pid = 0;
905 struct knote *kn = NULL;
906 boolean_t found_knote = FALSE;
907 int fflags = 0; /* filter flags for EVFILT_MEMORYSTATUS */
908 uint64_t value = 0;
909
910 error = sysctl_handle_quad(oidp, &value, 0, req);
911 if (error || !req->newptr)
912 return (error);
913
914 /*
915 * Find the pid in the low 32 bits of value passed in.
916 */
917 pid = (int)(value & 0xFFFFFFFF);
918
919 /*
920 * Find notification in the high 32 bits of the value passed in.
921 */
922 fflags = (int)((value >> 32) & 0xFFFFFFFF);
923
924 /*
925 * For backwards compatibility, when no notification is
926 * passed in, default to the NOTE_MEMORYSTATUS_PRESSURE_WARN
927 */
928 if (fflags == 0) {
929 fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
930 // printf("memorystatus_vm_pressure_send: using default notification [0x%x]\n", fflags);
931 }
932
933 /*
934 * See event.h ... fflags for EVFILT_MEMORYSTATUS
935 */
936 if (!((fflags == NOTE_MEMORYSTATUS_PRESSURE_NORMAL)||
937 (fflags == NOTE_MEMORYSTATUS_PRESSURE_WARN) ||
938 (fflags == NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) ||
939 (fflags == NOTE_MEMORYSTATUS_LOW_SWAP) ||
940 (fflags == NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) ||
941 (fflags == NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL))) {
942
943 printf("memorystatus_vm_pressure_send: notification [0x%x] not supported \n", fflags);
944 error = 1;
945 return (error);
946 }
947
948 /*
949 * Forcibly send pid a memorystatus notification.
950 */
951
952 memorystatus_klist_lock();
953
954 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
955 proc_t knote_proc = knote_get_kq(kn)->kq_p;
956 pid_t knote_pid = knote_proc->p_pid;
957
958 if (knote_pid == pid) {
959 /*
960 * Forcibly send this pid a memorystatus notification.
961 */
962 kn->kn_fflags = fflags;
963 found_knote = TRUE;
964 }
965 }
966
967 if (found_knote) {
968 KNOTE(&memorystatus_klist, 0);
969 printf("memorystatus_vm_pressure_send: (value 0x%llx) notification [0x%x] sent to process [%d] \n", value, fflags, pid);
970 error = 0;
971 } else {
972 printf("memorystatus_vm_pressure_send: (value 0x%llx) notification [0x%x] not sent to process [%d] (none registered?)\n", value, fflags, pid);
973 error = 1;
974 }
975
976 memorystatus_klist_unlock();
977
978 return (error);
979 }
980
981 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_send, CTLTYPE_QUAD|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
982 0, 0, &sysctl_memorystatus_vm_pressure_send, "Q", "");
983
984 #endif /* VM_PRESSURE_EVENTS */
985
986 #if CONFIG_JETSAM
987
988 SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
989
990 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
991 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
992 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
993 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
994 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_policy_more_free_offset_pages, CTLFLAG_RW, &memorystatus_policy_more_free_offset_pages, 0, "");
995
996 /* Diagnostic code */
997
998 enum {
999 kJetsamDiagnosticModeNone = 0,
1000 kJetsamDiagnosticModeAll = 1,
1001 kJetsamDiagnosticModeStopAtFirstActive = 2,
1002 kJetsamDiagnosticModeCount
1003 } jetsam_diagnostic_mode = kJetsamDiagnosticModeNone;
1004
1005 static int jetsam_diagnostic_suspended_one_active_proc = 0;
1006
1007 static int
1008 sysctl_jetsam_diagnostic_mode SYSCTL_HANDLER_ARGS
1009 {
1010 #pragma unused(arg1, arg2)
1011
1012 const char *diagnosticStrings[] = {
1013 "jetsam: diagnostic mode: resetting critical level.",
1014 "jetsam: diagnostic mode: will examine all processes",
1015 "jetsam: diagnostic mode: will stop at first active process"
1016 };
1017
1018 int error, val = jetsam_diagnostic_mode;
1019 boolean_t changed = FALSE;
1020
1021 error = sysctl_handle_int(oidp, &val, 0, req);
1022 if (error || !req->newptr)
1023 return (error);
1024 if ((val < 0) || (val >= kJetsamDiagnosticModeCount)) {
1025 printf("jetsam: diagnostic mode: invalid value - %d\n", val);
1026 return EINVAL;
1027 }
1028
1029 proc_list_lock();
1030
1031 if ((unsigned int) val != jetsam_diagnostic_mode) {
1032 jetsam_diagnostic_mode = val;
1033
1034 memorystatus_jetsam_policy &= ~kPolicyDiagnoseActive;
1035
1036 switch (jetsam_diagnostic_mode) {
1037 case kJetsamDiagnosticModeNone:
1038 /* Already cleared */
1039 break;
1040 case kJetsamDiagnosticModeAll:
1041 memorystatus_jetsam_policy |= kPolicyDiagnoseAll;
1042 break;
1043 case kJetsamDiagnosticModeStopAtFirstActive:
1044 memorystatus_jetsam_policy |= kPolicyDiagnoseFirst;
1045 break;
1046 default:
1047 /* Already validated */
1048 break;
1049 }
1050
1051 memorystatus_update_levels_locked(FALSE);
1052 changed = TRUE;
1053 }
1054
1055 proc_list_unlock();
1056
1057 if (changed) {
1058 printf("%s\n", diagnosticStrings[val]);
1059 }
1060
1061 return (0);
1062 }
1063
1064 SYSCTL_PROC(_debug, OID_AUTO, jetsam_diagnostic_mode, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED|CTLFLAG_ANYBODY,
1065 &jetsam_diagnostic_mode, 0, sysctl_jetsam_diagnostic_mode, "I", "Jetsam Diagnostic Mode");
1066
1067 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jetsam_policy_offset_pages_diagnostic, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jetsam_policy_offset_pages_diagnostic, 0, "");
1068
1069 #if VM_PRESSURE_EVENTS
1070
1071 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_pressure, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_pressure, 0, "");
1072
1073 #endif /* VM_PRESSURE_EVENTS */
1074
1075 #endif /* CONFIG_JETSAM */
1076
1077 #if CONFIG_FREEZE
1078
1079 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_daily_mb_max, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_daily_mb_max, 0, "");
1080
1081 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_threshold, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_threshold, 0, "");
1082
1083 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_pages_min, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_pages_min, 0, "");
1084 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_pages_max, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_pages_max, 0, "");
1085
1086 SYSCTL_QUAD(_kern, OID_AUTO, memorystatus_freeze_count, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_freeze_count, "");
1087 SYSCTL_QUAD(_kern, OID_AUTO, memorystatus_freeze_pageouts, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_freeze_pageouts, "");
1088 SYSCTL_QUAD(_kern, OID_AUTO, memorystatus_freeze_throttle_count, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_freeze_throttle_count, "");
1089 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_min_processes, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_suspended_threshold, 0, "");
1090
1091 boolean_t memorystatus_freeze_throttle_enabled = TRUE;
1092 SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_throttle_enabled, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_throttle_enabled, 0, "");
1093
1094 /*
1095 * Manual trigger of freeze and thaw for dev / debug kernels only.
1096 */
1097 static int
1098 sysctl_memorystatus_freeze SYSCTL_HANDLER_ARGS
1099 {
1100 #pragma unused(arg1, arg2)
1101 int error, pid = 0;
1102 proc_t p;
1103
1104 if (memorystatus_freeze_enabled == FALSE) {
1105 return ENOTSUP;
1106 }
1107
1108 error = sysctl_handle_int(oidp, &pid, 0, req);
1109 if (error || !req->newptr)
1110 return (error);
1111
1112 if (pid == 2) {
1113 vm_pageout_anonymous_pages();
1114
1115 return 0;
1116 }
1117
1118 lck_mtx_lock(&freezer_mutex);
1119
1120 p = proc_find(pid);
1121 if (p != NULL) {
1122 uint32_t purgeable, wired, clean, dirty;
1123 boolean_t shared;
1124 uint32_t max_pages = 0;
1125
1126 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
1127
1128 unsigned int avail_swap_space = 0; /* in pages. */
1129
1130 /*
1131 * Freezer backed by the compressor and swap file(s)
1132 * while will hold compressed data.
1133 */
1134 avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
1135
1136 max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
1137
1138 } else {
1139 /*
1140 * We only have the compressor without any swap.
1141 */
1142 max_pages = UINT32_MAX - 1;
1143 }
1144
1145 error = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
1146 proc_rele(p);
1147
1148 if (error)
1149 error = EIO;
1150
1151 lck_mtx_unlock(&freezer_mutex);
1152 return error;
1153 }
1154
1155 lck_mtx_unlock(&freezer_mutex);
1156 return EINVAL;
1157 }
1158
1159 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_freeze, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
1160 0, 0, &sysctl_memorystatus_freeze, "I", "");
1161
1162 static int
1163 sysctl_memorystatus_available_pages_thaw SYSCTL_HANDLER_ARGS
1164 {
1165 #pragma unused(arg1, arg2)
1166
1167 int error, pid = 0;
1168 proc_t p;
1169
1170 if (memorystatus_freeze_enabled == FALSE) {
1171 return ENOTSUP;
1172 }
1173
1174 error = sysctl_handle_int(oidp, &pid, 0, req);
1175 if (error || !req->newptr)
1176 return (error);
1177
1178 p = proc_find(pid);
1179 if (p != NULL) {
1180 error = task_thaw(p->task);
1181 proc_rele(p);
1182
1183 if (error)
1184 error = EIO;
1185 return error;
1186 }
1187
1188 return EINVAL;
1189 }
1190
1191 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_thaw, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
1192 0, 0, &sysctl_memorystatus_available_pages_thaw, "I", "");
1193
1194 #endif /* CONFIG_FREEZE */
1195
1196 #endif /* DEVELOPMENT || DEBUG */
1197
1198 extern kern_return_t kernel_thread_start_priority(thread_continue_t continuation,
1199 void *parameter,
1200 integer_t priority,
1201 thread_t *new_thread);
1202
1203 #if DEVELOPMENT || DEBUG
1204
1205 static int
1206 sysctl_memorystatus_disconnect_page_mappings SYSCTL_HANDLER_ARGS
1207 {
1208 #pragma unused(arg1, arg2)
1209 int error = 0, pid = 0;
1210 proc_t p;
1211
1212 error = sysctl_handle_int(oidp, &pid, 0, req);
1213 if (error || !req->newptr)
1214 return (error);
1215
1216 lck_mtx_lock(&disconnect_page_mappings_mutex);
1217
1218 if (pid == -1) {
1219 vm_pageout_disconnect_all_pages();
1220 } else {
1221 p = proc_find(pid);
1222
1223 if (p != NULL) {
1224 error = task_disconnect_page_mappings(p->task);
1225
1226 proc_rele(p);
1227
1228 if (error)
1229 error = EIO;
1230 } else
1231 error = EINVAL;
1232 }
1233 lck_mtx_unlock(&disconnect_page_mappings_mutex);
1234
1235 return error;
1236 }
1237
1238 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_disconnect_page_mappings, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
1239 0, 0, &sysctl_memorystatus_disconnect_page_mappings, "I", "");
1240
1241 #endif /* DEVELOPMENT || DEBUG */
1242
1243
1244
1245 #if CONFIG_JETSAM
1246 /*
1247 * Picks the sorting routine for a given jetsam priority band.
1248 *
1249 * Input:
1250 * bucket_index - jetsam priority band to be sorted.
1251 * sort_order - JETSAM_SORT_xxx from kern_memorystatus.h
1252 * Currently sort_order is only meaningful when handling
1253 * coalitions.
1254 *
1255 * Return:
1256 * 0 on success
1257 * non-0 on failure
1258 */
1259 static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order)
1260 {
1261 int coal_sort_order;
1262
1263 /*
1264 * Verify the jetsam priority
1265 */
1266 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1267 return(EINVAL);
1268 }
1269
1270 #if DEVELOPMENT || DEBUG
1271 if (sort_order == JETSAM_SORT_DEFAULT) {
1272 coal_sort_order = COALITION_SORT_DEFAULT;
1273 } else {
1274 coal_sort_order = sort_order; /* only used for testing scenarios */
1275 }
1276 #else
1277 /* Verify default */
1278 if (sort_order == JETSAM_SORT_DEFAULT) {
1279 coal_sort_order = COALITION_SORT_DEFAULT;
1280 } else {
1281 return(EINVAL);
1282 }
1283 #endif
1284
1285 proc_list_lock();
1286 switch (bucket_index) {
1287 case JETSAM_PRIORITY_FOREGROUND:
1288 if (memorystatus_sort_by_largest_coalition_locked(bucket_index, coal_sort_order) == 0) {
1289 /*
1290 * Fall back to per process sorting when zero coalitions are found.
1291 */
1292 memorystatus_sort_by_largest_process_locked(bucket_index);
1293 }
1294 break;
1295 default:
1296 memorystatus_sort_by_largest_process_locked(bucket_index);
1297 break;
1298 }
1299 proc_list_unlock();
1300
1301 return(0);
1302 }
1303
1304 /*
1305 * Sort processes by size for a single jetsam bucket.
1306 */
1307
1308 static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index)
1309 {
1310 proc_t p = NULL, insert_after_proc = NULL, max_proc = NULL;
1311 proc_t next_p = NULL, prev_max_proc = NULL;
1312 uint32_t pages = 0, max_pages = 0;
1313 memstat_bucket_t *current_bucket;
1314
1315 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
1316 return;
1317 }
1318
1319 current_bucket = &memstat_bucket[bucket_index];
1320
1321 p = TAILQ_FIRST(&current_bucket->list);
1322
1323 while (p) {
1324 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
1325 max_pages = pages;
1326 max_proc = p;
1327 prev_max_proc = p;
1328
1329 while ((next_p = TAILQ_NEXT(p, p_memstat_list)) != NULL) {
1330 /* traversing list until we find next largest process */
1331 p=next_p;
1332 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
1333 if (pages > max_pages) {
1334 max_pages = pages;
1335 max_proc = p;
1336 }
1337 }
1338
1339 if (prev_max_proc != max_proc) {
1340 /* found a larger process, place it in the list */
1341 TAILQ_REMOVE(&current_bucket->list, max_proc, p_memstat_list);
1342 if (insert_after_proc == NULL) {
1343 TAILQ_INSERT_HEAD(&current_bucket->list, max_proc, p_memstat_list);
1344 } else {
1345 TAILQ_INSERT_AFTER(&current_bucket->list, insert_after_proc, max_proc, p_memstat_list);
1346 }
1347 prev_max_proc = max_proc;
1348 }
1349
1350 insert_after_proc = max_proc;
1351
1352 p = TAILQ_NEXT(max_proc, p_memstat_list);
1353 }
1354 }
1355
1356 #endif /* CONFIG_JETSAM */
1357
1358 static proc_t memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search) {
1359 memstat_bucket_t *current_bucket;
1360 proc_t next_p;
1361
1362 if ((*bucket_index) >= MEMSTAT_BUCKET_COUNT) {
1363 return NULL;
1364 }
1365
1366 current_bucket = &memstat_bucket[*bucket_index];
1367 next_p = TAILQ_FIRST(&current_bucket->list);
1368 if (!next_p && search) {
1369 while (!next_p && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1370 current_bucket = &memstat_bucket[*bucket_index];
1371 next_p = TAILQ_FIRST(&current_bucket->list);
1372 }
1373 }
1374
1375 return next_p;
1376 }
1377
1378 static proc_t memorystatus_get_next_proc_locked(unsigned int *bucket_index, proc_t p, boolean_t search) {
1379 memstat_bucket_t *current_bucket;
1380 proc_t next_p;
1381
1382 if (!p || ((*bucket_index) >= MEMSTAT_BUCKET_COUNT)) {
1383 return NULL;
1384 }
1385
1386 next_p = TAILQ_NEXT(p, p_memstat_list);
1387 while (!next_p && search && (++(*bucket_index) < MEMSTAT_BUCKET_COUNT)) {
1388 current_bucket = &memstat_bucket[*bucket_index];
1389 next_p = TAILQ_FIRST(&current_bucket->list);
1390 }
1391
1392 return next_p;
1393 }
1394
1395 __private_extern__ void
1396 memorystatus_init(void)
1397 {
1398 thread_t thread = THREAD_NULL;
1399 kern_return_t result;
1400 int i;
1401
1402 #if CONFIG_FREEZE
1403 memorystatus_freeze_pages_min = FREEZE_PAGES_MIN;
1404 memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
1405 #endif
1406
1407 #if DEVELOPMENT || DEBUG
1408 disconnect_page_mappings_lck_grp_attr = lck_grp_attr_alloc_init();
1409 disconnect_page_mappings_lck_grp = lck_grp_alloc_init("disconnect_page_mappings", disconnect_page_mappings_lck_grp_attr);
1410
1411 lck_mtx_init(&disconnect_page_mappings_mutex, disconnect_page_mappings_lck_grp, NULL);
1412 #endif
1413
1414 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
1415 nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
1416
1417 /* Init buckets */
1418 for (i = 0; i < MEMSTAT_BUCKET_COUNT; i++) {
1419 TAILQ_INIT(&memstat_bucket[i].list);
1420 memstat_bucket[i].count = 0;
1421 }
1422
1423 memorystatus_idle_demotion_call = thread_call_allocate((thread_call_func_t)memorystatus_perform_idle_demotion, NULL);
1424
1425 /* Apply overrides */
1426 PE_get_default("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage));
1427 if (delta_percentage == 0) {
1428 delta_percentage = 5;
1429 }
1430 assert(delta_percentage < 100);
1431 PE_get_default("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage));
1432 assert(critical_threshold_percentage < 100);
1433 PE_get_default("kern.jetsam_idle_offset", &idle_offset_percentage, sizeof(idle_offset_percentage));
1434 assert(idle_offset_percentage < 100);
1435 PE_get_default("kern.jetsam_pressure_threshold", &pressure_threshold_percentage, sizeof(pressure_threshold_percentage));
1436 assert(pressure_threshold_percentage < 100);
1437 PE_get_default("kern.jetsam_freeze_threshold", &freeze_threshold_percentage, sizeof(freeze_threshold_percentage));
1438 assert(freeze_threshold_percentage < 100);
1439
1440 if (!PE_parse_boot_argn("jetsam_aging_policy", &jetsam_aging_policy,
1441 sizeof (jetsam_aging_policy))) {
1442
1443 if (!PE_get_default("kern.jetsam_aging_policy", &jetsam_aging_policy,
1444 sizeof(jetsam_aging_policy))) {
1445
1446 jetsam_aging_policy = kJetsamAgingPolicyLegacy;
1447 }
1448 }
1449
1450 if (jetsam_aging_policy > kJetsamAgingPolicyMax) {
1451 jetsam_aging_policy = kJetsamAgingPolicyLegacy;
1452 }
1453
1454 switch (jetsam_aging_policy) {
1455
1456 case kJetsamAgingPolicyNone:
1457 system_procs_aging_band = JETSAM_PRIORITY_IDLE;
1458 applications_aging_band = JETSAM_PRIORITY_IDLE;
1459 break;
1460
1461 case kJetsamAgingPolicyLegacy:
1462 /*
1463 * Legacy behavior where some daemons get a 10s protection once
1464 * AND only before the first clean->dirty->clean transition before
1465 * going into IDLE band.
1466 */
1467 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1468 applications_aging_band = JETSAM_PRIORITY_IDLE;
1469 break;
1470
1471 case kJetsamAgingPolicySysProcsReclaimedFirst:
1472 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1473 applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1474 break;
1475
1476 case kJetsamAgingPolicyAppsReclaimedFirst:
1477 system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
1478 applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
1479 break;
1480
1481 default:
1482 break;
1483 }
1484
1485 /*
1486 * The aging bands cannot overlap with the JETSAM_PRIORITY_ELEVATED_INACTIVE
1487 * band and must be below it in priority. This is so that we don't have to make
1488 * our 'aging' code worry about a mix of processes, some of which need to age
1489 * and some others that need to stay elevated in the jetsam bands.
1490 */
1491 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > system_procs_aging_band);
1492 assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > applications_aging_band);
1493
1494 #if CONFIG_JETSAM
1495 /* Take snapshots for idle-exit kills by default? First check the boot-arg... */
1496 if (!PE_parse_boot_argn("jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof (memorystatus_idle_snapshot))) {
1497 /* ...no boot-arg, so check the device tree */
1498 PE_get_default("kern.jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot));
1499 }
1500
1501 memorystatus_delta = delta_percentage * atop_64(max_mem) / 100;
1502 memorystatus_available_pages_critical_idle_offset = idle_offset_percentage * atop_64(max_mem) / 100;
1503 memorystatus_available_pages_critical_base = (critical_threshold_percentage / delta_percentage) * memorystatus_delta;
1504 memorystatus_policy_more_free_offset_pages = (policy_more_free_offset_percentage / delta_percentage) * memorystatus_delta;
1505
1506 memorystatus_jetsam_snapshot_max = maxproc;
1507 memorystatus_jetsam_snapshot =
1508 (memorystatus_jetsam_snapshot_t*)kalloc(sizeof(memorystatus_jetsam_snapshot_t) +
1509 sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_max);
1510 if (!memorystatus_jetsam_snapshot) {
1511 panic("Could not allocate memorystatus_jetsam_snapshot");
1512 }
1513
1514 nanoseconds_to_absolutetime((uint64_t)JETSAM_SNAPSHOT_TIMEOUT_SECS * NSEC_PER_SEC, &memorystatus_jetsam_snapshot_timeout);
1515
1516 memset(&memorystatus_at_boot_snapshot, 0, sizeof(memorystatus_jetsam_snapshot_t));
1517
1518 /* No contention at this point */
1519 memorystatus_update_levels_locked(FALSE);
1520
1521 /* Jetsam Loop Detection */
1522 if (max_mem <= (512 * 1024 * 1024)) {
1523 /* 512 MB devices */
1524 memorystatus_jld_eval_period_msecs = 8000; /* 8000 msecs == 8 second window */
1525 } else {
1526 /* 1GB and larger devices */
1527 memorystatus_jld_eval_period_msecs = 6000; /* 6000 msecs == 6 second window */
1528 }
1529 #endif
1530
1531 #if CONFIG_FREEZE
1532 memorystatus_freeze_threshold = (freeze_threshold_percentage / delta_percentage) * memorystatus_delta;
1533 #endif
1534
1535 result = kernel_thread_start_priority(memorystatus_thread, NULL, 95 /* MAXPRI_KERNEL */, &thread);
1536 if (result == KERN_SUCCESS) {
1537 thread_deallocate(thread);
1538 } else {
1539 panic("Could not create memorystatus_thread");
1540 }
1541 }
1542
1543 /* Centralised for the purposes of allowing panic-on-jetsam */
1544 extern void
1545 vm_run_compactor(void);
1546
1547 /*
1548 * The jetsam no frills kill call
1549 * Return: 0 on success
1550 * error code on failure (EINVAL...)
1551 */
1552 static int
1553 jetsam_do_kill(proc_t p, int jetsam_flags, os_reason_t jetsam_reason) {
1554 int error = 0;
1555 error = exit_with_reason(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags, jetsam_reason);
1556 return(error);
1557 }
1558
1559 /*
1560 * Wrapper for processes exiting with memorystatus details
1561 */
1562 static boolean_t
1563 memorystatus_do_kill(proc_t p, uint32_t cause, os_reason_t jetsam_reason) {
1564
1565 int error = 0;
1566 __unused pid_t victim_pid = p->p_pid;
1567
1568 KERNEL_DEBUG_CONSTANT( (BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_START,
1569 victim_pid, cause, vm_page_free_count, 0, 0);
1570
1571 DTRACE_MEMORYSTATUS3(memorystatus_do_kill, proc_t, p, os_reason_t, jetsam_reason, uint32_t, cause);
1572 #if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
1573 if (memorystatus_jetsam_panic_debug & (1 << cause)) {
1574 panic("memorystatus_do_kill(): jetsam debug panic (cause: %d)", cause);
1575 }
1576 #else
1577 #pragma unused(cause)
1578 #endif
1579 int jetsam_flags = P_LTERM_JETSAM;
1580 switch (cause) {
1581 case kMemorystatusKilledHiwat: jetsam_flags |= P_JETSAM_HIWAT; break;
1582 case kMemorystatusKilledVnodes: jetsam_flags |= P_JETSAM_VNODE; break;
1583 case kMemorystatusKilledVMPageShortage: jetsam_flags |= P_JETSAM_VMPAGESHORTAGE; break;
1584 case kMemorystatusKilledVMThrashing: jetsam_flags |= P_JETSAM_VMTHRASHING; break;
1585 case kMemorystatusKilledFCThrashing: jetsam_flags |= P_JETSAM_FCTHRASHING; break;
1586 case kMemorystatusKilledPerProcessLimit: jetsam_flags |= P_JETSAM_PID; break;
1587 case kMemorystatusKilledIdleExit: jetsam_flags |= P_JETSAM_IDLEEXIT; break;
1588 }
1589 error = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
1590
1591 KERNEL_DEBUG_CONSTANT( (BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_END,
1592 victim_pid, cause, vm_page_free_count, error, 0);
1593
1594 vm_run_compactor();
1595
1596 return (error == 0);
1597 }
1598
1599 /*
1600 * Node manipulation
1601 */
1602
1603 static void
1604 memorystatus_check_levels_locked(void) {
1605 #if CONFIG_JETSAM
1606 /* Update levels */
1607 memorystatus_update_levels_locked(TRUE);
1608 #endif
1609 }
1610
1611 /*
1612 * Pin a process to a particular jetsam band when it is in the background i.e. not doing active work.
1613 * For an application: that means no longer in the FG band
1614 * For a daemon: that means no longer in its 'requested' jetsam priority band
1615 */
1616
1617 int
1618 memorystatus_update_inactive_jetsam_priority_band(pid_t pid, uint32_t op_flags, boolean_t effective_now)
1619 {
1620 int error = 0;
1621 boolean_t enable = FALSE;
1622 proc_t p = NULL;
1623
1624 if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE) {
1625 enable = TRUE;
1626 } else if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE) {
1627 enable = FALSE;
1628 } else {
1629 return EINVAL;
1630 }
1631
1632 p = proc_find(pid);
1633 if (p != NULL) {
1634
1635 if ((enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) ||
1636 (!enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == 0))) {
1637 /*
1638 * No change in state.
1639 */
1640
1641 } else {
1642
1643 proc_list_lock();
1644
1645 if (enable) {
1646 p->p_memstat_state |= P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
1647 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1648
1649 if (effective_now) {
1650 if (p->p_memstat_effectivepriority < JETSAM_PRIORITY_ELEVATED_INACTIVE) {
1651 boolean_t trigger_exception;
1652 CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
1653 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, trigger_exception);
1654 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_ELEVATED_INACTIVE, FALSE, FALSE);
1655 }
1656 } else {
1657 if (isProcessInAgingBands(p)) {
1658 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1659 }
1660 }
1661 } else {
1662
1663 p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
1664 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1665
1666 if (effective_now) {
1667 if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_ELEVATED_INACTIVE) {
1668 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1669 }
1670 } else {
1671 if (isProcessInAgingBands(p)) {
1672 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
1673 }
1674 }
1675 }
1676
1677 proc_list_unlock();
1678 }
1679 proc_rele(p);
1680 error = 0;
1681
1682 } else {
1683 error = ESRCH;
1684 }
1685
1686 return error;
1687 }
1688
1689 static void
1690 memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2)
1691 {
1692 proc_t p;
1693 uint64_t current_time = 0, idle_delay_time = 0;
1694 int demote_prio_band = 0;
1695 memstat_bucket_t *demotion_bucket;
1696
1697 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion()\n");
1698
1699 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_START, 0, 0, 0, 0, 0);
1700
1701 current_time = mach_absolute_time();
1702
1703 proc_list_lock();
1704
1705 demote_prio_band = JETSAM_PRIORITY_IDLE + 1;
1706
1707 for (; demote_prio_band < JETSAM_PRIORITY_MAX; demote_prio_band++) {
1708
1709 if (demote_prio_band != system_procs_aging_band && demote_prio_band != applications_aging_band)
1710 continue;
1711
1712 demotion_bucket = &memstat_bucket[demote_prio_band];
1713 p = TAILQ_FIRST(&demotion_bucket->list);
1714
1715 while (p) {
1716 MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion() found %d\n", p->p_pid);
1717
1718 assert(p->p_memstat_idledeadline);
1719
1720 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
1721
1722 if (current_time >= p->p_memstat_idledeadline) {
1723
1724 if ((isSysProc(p) &&
1725 ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_IS_DIRTY)) != P_DIRTY_IDLE_EXIT_ENABLED)) || /* system proc marked dirty*/
1726 task_has_assertions((struct task *)(p->task))) { /* has outstanding assertions which might indicate outstanding work too */
1727 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_delay_time : memorystatus_apps_idle_delay_time;
1728
1729 p->p_memstat_idledeadline += idle_delay_time;
1730 p = TAILQ_NEXT(p, p_memstat_list);
1731
1732 } else {
1733
1734 proc_t next_proc = NULL;
1735
1736 next_proc = TAILQ_NEXT(p, p_memstat_list);
1737 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
1738
1739 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false, true);
1740
1741 p = next_proc;
1742 continue;
1743
1744 }
1745 } else {
1746 // No further candidates
1747 break;
1748 }
1749 }
1750
1751 }
1752
1753 memorystatus_reschedule_idle_demotion_locked();
1754
1755 proc_list_unlock();
1756
1757 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_IDLE_DEMOTE) | DBG_FUNC_END, 0, 0, 0, 0, 0);
1758 }
1759
1760 static void
1761 memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state)
1762 {
1763 boolean_t present_in_sysprocs_aging_bucket = FALSE;
1764 boolean_t present_in_apps_aging_bucket = FALSE;
1765 uint64_t idle_delay_time = 0;
1766
1767 if (jetsam_aging_policy == kJetsamAgingPolicyNone) {
1768 return;
1769 }
1770
1771 if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
1772 /*
1773 * This process isn't going to be making the trip to the lower bands.
1774 */
1775 return;
1776 }
1777
1778 if (isProcessInAgingBands(p)){
1779
1780 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
1781 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) != P_DIRTY_AGING_IN_PROGRESS);
1782 }
1783
1784 if (isSysProc(p) && system_procs_aging_band) {
1785 present_in_sysprocs_aging_bucket = TRUE;
1786
1787 } else if (isApp(p) && applications_aging_band) {
1788 present_in_apps_aging_bucket = TRUE;
1789 }
1790 }
1791
1792 assert(!present_in_sysprocs_aging_bucket);
1793 assert(!present_in_apps_aging_bucket);
1794
1795 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",
1796 p->p_pid, p->p_memstat_dirty, set_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
1797
1798 if(isSysProc(p)) {
1799 assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
1800 }
1801
1802 idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_delay_time : memorystatus_apps_idle_delay_time;
1803
1804 if (set_state) {
1805 p->p_memstat_dirty |= P_DIRTY_AGING_IN_PROGRESS;
1806 p->p_memstat_idledeadline = mach_absolute_time() + idle_delay_time;
1807 }
1808
1809 assert(p->p_memstat_idledeadline);
1810
1811 if (isSysProc(p) && present_in_sysprocs_aging_bucket == FALSE) {
1812 memorystatus_scheduled_idle_demotions_sysprocs++;
1813
1814 } else if (isApp(p) && present_in_apps_aging_bucket == FALSE) {
1815 memorystatus_scheduled_idle_demotions_apps++;
1816 }
1817 }
1818
1819 static void
1820 memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clear_state)
1821 {
1822 boolean_t present_in_sysprocs_aging_bucket = FALSE;
1823 boolean_t present_in_apps_aging_bucket = FALSE;
1824
1825 if (!system_procs_aging_band && !applications_aging_band) {
1826 return;
1827 }
1828
1829 if ((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == 0) {
1830 return;
1831 }
1832
1833 if (isProcessInAgingBands(p)) {
1834
1835 if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
1836 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == P_DIRTY_AGING_IN_PROGRESS);
1837 }
1838
1839 if (isSysProc(p) && system_procs_aging_band) {
1840 assert(p->p_memstat_effectivepriority == system_procs_aging_band);
1841 assert(p->p_memstat_idledeadline);
1842 present_in_sysprocs_aging_bucket = TRUE;
1843
1844 } else if (isApp(p) && applications_aging_band) {
1845 assert(p->p_memstat_effectivepriority == applications_aging_band);
1846 assert(p->p_memstat_idledeadline);
1847 present_in_apps_aging_bucket = TRUE;
1848 }
1849 }
1850
1851 MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for pid %d (clear_state %d, demotions %d).\n",
1852 p->p_pid, clear_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
1853
1854
1855 if (clear_state) {
1856 p->p_memstat_idledeadline = 0;
1857 p->p_memstat_dirty &= ~P_DIRTY_AGING_IN_PROGRESS;
1858 }
1859
1860 if (isSysProc(p) &&present_in_sysprocs_aging_bucket == TRUE) {
1861 memorystatus_scheduled_idle_demotions_sysprocs--;
1862 assert(memorystatus_scheduled_idle_demotions_sysprocs >= 0);
1863
1864 } else if (isApp(p) && present_in_apps_aging_bucket == TRUE) {
1865 memorystatus_scheduled_idle_demotions_apps--;
1866 assert(memorystatus_scheduled_idle_demotions_apps >= 0);
1867 }
1868
1869 assert((memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps) >= 0);
1870 }
1871
1872 static void
1873 memorystatus_reschedule_idle_demotion_locked(void) {
1874 if (0 == (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps)) {
1875 if (memstat_idle_demotion_deadline) {
1876 /* Transitioned 1->0, so cancel next call */
1877 thread_call_cancel(memorystatus_idle_demotion_call);
1878 memstat_idle_demotion_deadline = 0;
1879 }
1880 } else {
1881 memstat_bucket_t *demotion_bucket;
1882 proc_t p = NULL, p1 = NULL, p2 = NULL;
1883
1884 if (system_procs_aging_band) {
1885
1886 demotion_bucket = &memstat_bucket[system_procs_aging_band];
1887 p1 = TAILQ_FIRST(&demotion_bucket->list);
1888
1889 p = p1;
1890 }
1891
1892 if (applications_aging_band) {
1893
1894 demotion_bucket = &memstat_bucket[applications_aging_band];
1895 p2 = TAILQ_FIRST(&demotion_bucket->list);
1896
1897 if (p1 && p2) {
1898 p = (p1->p_memstat_idledeadline > p2->p_memstat_idledeadline) ? p2 : p1;
1899 } else {
1900 p = (p1 == NULL) ? p2 : p1;
1901 }
1902
1903 }
1904
1905 assert(p);
1906
1907 if (p != NULL) {
1908 assert(p && p->p_memstat_idledeadline);
1909 if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline){
1910 thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
1911 memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
1912 }
1913 }
1914 }
1915 }
1916
1917 /*
1918 * List manipulation
1919 */
1920
1921 int
1922 memorystatus_add(proc_t p, boolean_t locked)
1923 {
1924 memstat_bucket_t *bucket;
1925
1926 MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding pid %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
1927
1928 if (!locked) {
1929 proc_list_lock();
1930 }
1931
1932 DTRACE_MEMORYSTATUS2(memorystatus_add, proc_t, p, int32_t, p->p_memstat_effectivepriority);
1933
1934 /* Processes marked internal do not have priority tracked */
1935 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
1936 goto exit;
1937 }
1938
1939 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
1940
1941 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
1942 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs - 1);
1943
1944 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
1945 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps - 1);
1946
1947 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
1948 /*
1949 * Entering the idle band.
1950 * Record idle start time.
1951 */
1952 p->p_memstat_idle_start = mach_absolute_time();
1953 }
1954
1955 TAILQ_INSERT_TAIL(&bucket->list, p, p_memstat_list);
1956 bucket->count++;
1957
1958 memorystatus_list_count++;
1959
1960 memorystatus_check_levels_locked();
1961
1962 exit:
1963 if (!locked) {
1964 proc_list_unlock();
1965 }
1966
1967 return 0;
1968 }
1969
1970 /*
1971 * Description:
1972 * Moves a process from one jetsam bucket to another.
1973 * which changes the LRU position of the process.
1974 *
1975 * Monitors transition between buckets and if necessary
1976 * will update cached memory limits accordingly.
1977 *
1978 * skip_demotion_check:
1979 * - if the 'jetsam aging policy' is NOT 'legacy':
1980 * When this flag is TRUE, it means we are going
1981 * to age the ripe processes out of the aging bands and into the
1982 * IDLE band and apply their inactive memory limits.
1983 *
1984 * - if the 'jetsam aging policy' is 'legacy':
1985 * When this flag is TRUE, it might mean the above aging mechanism
1986 * OR
1987 * It might be that we have a process that has used up its 'idle deferral'
1988 * stay that is given to it once per lifetime. And in this case, the process
1989 * won't be going through any aging codepaths. But we still need to apply
1990 * the right inactive limits and so we explicitly set this to TRUE if the
1991 * new priority for the process is the IDLE band.
1992 */
1993 void
1994 memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check)
1995 {
1996 memstat_bucket_t *old_bucket, *new_bucket;
1997
1998 assert(priority < MEMSTAT_BUCKET_COUNT);
1999
2000 /* Ensure that exit isn't underway, leaving the proc retained but removed from its bucket */
2001 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2002 return;
2003 }
2004
2005 MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting %s(%d) to priority %d, inserting at %s\n",
2006 (*p->p_name ? p->p_name : "unknown"), p->p_pid, priority, head_insert ? "head" : "tail");
2007
2008 DTRACE_MEMORYSTATUS3(memorystatus_update_priority, proc_t, p, int32_t, p->p_memstat_effectivepriority, int, priority);
2009
2010 #if DEVELOPMENT || DEBUG
2011 if (priority == JETSAM_PRIORITY_IDLE && /* if the process is on its way into the IDLE band */
2012 skip_demotion_check == FALSE && /* and it isn't via the path that will set the INACTIVE memlimits */
2013 (p->p_memstat_dirty & P_DIRTY_TRACK) && /* and it has 'DIRTY' tracking enabled */
2014 ((p->p_memstat_memlimit != p->p_memstat_memlimit_inactive) || /* and we notice that the current limit isn't the right value (inactive) */
2015 ((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) */
2016 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 */
2017 #endif /* DEVELOPMENT || DEBUG */
2018
2019 old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2020
2021 if (skip_demotion_check == FALSE) {
2022
2023 if (isSysProc(p)) {
2024 /*
2025 * For system processes, the memorystatus_dirty_* routines take care of adding/removing
2026 * the processes from the aging bands and balancing the demotion counts.
2027 * We can, however, override that if the process has an 'elevated inactive jetsam band' attribute.
2028 */
2029
2030 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE && (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
2031 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2032
2033 assert(! (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2034 }
2035 } else if (isApp(p)) {
2036
2037 /*
2038 * Check to see if the application is being lowered in jetsam priority. If so, and:
2039 * - it has an 'elevated inactive jetsam band' attribute, then put it in the JETSAM_PRIORITY_ELEVATED_INACTIVE band.
2040 * - it is a normal application, then let it age in the aging band if that policy is in effect.
2041 */
2042
2043 if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE && (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
2044 priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
2045 } else {
2046
2047 if (applications_aging_band) {
2048 if (p->p_memstat_effectivepriority == applications_aging_band) {
2049 assert(old_bucket->count == (memorystatus_scheduled_idle_demotions_apps + 1));
2050 }
2051
2052 if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && (priority <= applications_aging_band)) {
2053 assert(! (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
2054 priority = applications_aging_band;
2055 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2056 }
2057 }
2058 }
2059 }
2060 }
2061
2062 if ((system_procs_aging_band && (priority == system_procs_aging_band)) || (applications_aging_band && (priority == applications_aging_band))) {
2063 assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
2064 }
2065
2066 TAILQ_REMOVE(&old_bucket->list, p, p_memstat_list);
2067 old_bucket->count--;
2068
2069 new_bucket = &memstat_bucket[priority];
2070 if (head_insert)
2071 TAILQ_INSERT_HEAD(&new_bucket->list, p, p_memstat_list);
2072 else
2073 TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
2074 new_bucket->count++;
2075
2076 if (memorystatus_highwater_enabled) {
2077 boolean_t trigger_exception;
2078
2079 /*
2080 * If cached limit data is updated, then the limits
2081 * will be enforced by writing to the ledgers.
2082 */
2083 boolean_t ledger_update_needed = TRUE;
2084
2085 /*
2086 * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
2087 * Background limits are described via the inactive limit slots.
2088 *
2089 * Here, we must update the cached memory limit if the task
2090 * is transitioning between:
2091 * active <--> inactive
2092 * FG <--> BG
2093 * but:
2094 * dirty <--> clean is ignored
2095 *
2096 * We bypass non-idle processes that have opted into dirty tracking because
2097 * a move between buckets does not imply a transition between the
2098 * dirty <--> clean state.
2099 */
2100
2101 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
2102
2103 if (skip_demotion_check == TRUE && priority == JETSAM_PRIORITY_IDLE) {
2104 CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
2105 } else {
2106 ledger_update_needed = FALSE;
2107 }
2108
2109 } else if ((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) {
2110 /*
2111 * inactive --> active
2112 * BG --> FG
2113 * assign active state
2114 */
2115 CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
2116
2117 } else if ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
2118 /*
2119 * active --> inactive
2120 * FG --> BG
2121 * assign inactive state
2122 */
2123 CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
2124 } else {
2125 /*
2126 * The transition between jetsam priority buckets apparently did
2127 * not affect active/inactive state.
2128 * This is not unusual... especially during startup when
2129 * processes are getting established in their respective bands.
2130 */
2131 ledger_update_needed = FALSE;
2132 }
2133
2134 /*
2135 * Enforce the new limits by writing to the ledger
2136 */
2137 if (ledger_update_needed) {
2138 task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, trigger_exception);
2139
2140 MEMORYSTATUS_DEBUG(3, "memorystatus_update_priority_locked: new limit on pid %d (%dMB %s) priority old --> new (%d --> %d) dirty?=0x%x %s\n",
2141 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2142 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, priority, p->p_memstat_dirty,
2143 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2144 }
2145 }
2146
2147 /*
2148 * Record idle start or idle delta.
2149 */
2150 if (p->p_memstat_effectivepriority == priority) {
2151 /*
2152 * This process is not transitioning between
2153 * jetsam priority buckets. Do nothing.
2154 */
2155 } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2156 uint64_t now;
2157 /*
2158 * Transitioning out of the idle priority bucket.
2159 * Record idle delta.
2160 */
2161 assert(p->p_memstat_idle_start != 0);
2162 now = mach_absolute_time();
2163 if (now > p->p_memstat_idle_start) {
2164 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2165 }
2166 } else if (priority == JETSAM_PRIORITY_IDLE) {
2167 /*
2168 * Transitioning into the idle priority bucket.
2169 * Record idle start.
2170 */
2171 p->p_memstat_idle_start = mach_absolute_time();
2172 }
2173
2174 p->p_memstat_effectivepriority = priority;
2175
2176 #if CONFIG_SECLUDED_MEMORY
2177 if (secluded_for_apps &&
2178 task_could_use_secluded_mem(p->task)) {
2179 task_set_can_use_secluded_mem(
2180 p->task,
2181 (priority >= JETSAM_PRIORITY_FOREGROUND));
2182 }
2183 #endif /* CONFIG_SECLUDED_MEMORY */
2184
2185 memorystatus_check_levels_locked();
2186 }
2187
2188 /*
2189 *
2190 * Description: Update the jetsam priority and memory limit attributes for a given process.
2191 *
2192 * Parameters:
2193 * p init this process's jetsam information.
2194 * priority The jetsam priority band
2195 * user_data user specific data, unused by the kernel
2196 * effective guards against race if process's update already occurred
2197 * update_memlimit When true we know this is the init step via the posix_spawn path.
2198 *
2199 * memlimit_active Value in megabytes; The monitored footprint level while the
2200 * process is active. Exceeding it may result in termination
2201 * based on it's associated fatal flag.
2202 *
2203 * memlimit_active_is_fatal When a process is active and exceeds its memory footprint,
2204 * this describes whether or not it should be immediately fatal.
2205 *
2206 * memlimit_inactive Value in megabytes; The monitored footprint level while the
2207 * process is inactive. Exceeding it may result in termination
2208 * based on it's associated fatal flag.
2209 *
2210 * memlimit_inactive_is_fatal When a process is inactive and exceeds its memory footprint,
2211 * this describes whether or not it should be immediatly fatal.
2212 *
2213 * memlimit_background This process has a high-water-mark while in the background.
2214 * No longer meaningful. Background limits are described via
2215 * the inactive slots. Flag is ignored.
2216 *
2217 *
2218 * Returns: 0 Success
2219 * non-0 Failure
2220 */
2221
2222 int
2223 memorystatus_update(proc_t p, int priority, uint64_t user_data, boolean_t effective, boolean_t update_memlimit,
2224 int32_t memlimit_active, boolean_t memlimit_active_is_fatal,
2225 int32_t memlimit_inactive, boolean_t memlimit_inactive_is_fatal,
2226 __unused boolean_t memlimit_background)
2227 {
2228 int ret;
2229 boolean_t head_insert = false;
2230
2231 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);
2232
2233 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_START, p->p_pid, priority, user_data, effective, 0);
2234
2235 if (priority == -1) {
2236 /* Use as shorthand for default priority */
2237 priority = JETSAM_PRIORITY_DEFAULT;
2238 } else if ((priority == system_procs_aging_band) || (priority == applications_aging_band)) {
2239 /* Both the aging bands are reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
2240 priority = JETSAM_PRIORITY_IDLE;
2241 } else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
2242 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
2243 priority = JETSAM_PRIORITY_IDLE;
2244 head_insert = TRUE;
2245 } else if ((priority < 0) || (priority >= MEMSTAT_BUCKET_COUNT)) {
2246 /* Sanity check */
2247 ret = EINVAL;
2248 goto out;
2249 }
2250
2251 proc_list_lock();
2252
2253 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2254
2255 if (effective && (p->p_memstat_state & P_MEMSTAT_PRIORITYUPDATED)) {
2256 ret = EALREADY;
2257 proc_list_unlock();
2258 MEMORYSTATUS_DEBUG(1, "memorystatus_update: effective change specified for pid %d, but change already occurred.\n", p->p_pid);
2259 goto out;
2260 }
2261
2262 if ((p->p_memstat_state & P_MEMSTAT_TERMINATED) || ((p->p_listflag & P_LIST_EXITED) != 0)) {
2263 /*
2264 * This could happen when a process calling posix_spawn() is exiting on the jetsam thread.
2265 */
2266 ret = EBUSY;
2267 proc_list_unlock();
2268 goto out;
2269 }
2270
2271 p->p_memstat_state |= P_MEMSTAT_PRIORITYUPDATED;
2272 p->p_memstat_userdata = user_data;
2273 p->p_memstat_requestedpriority = priority;
2274
2275 if (update_memlimit) {
2276 boolean_t trigger_exception;
2277
2278 /*
2279 * Posix_spawn'd processes come through this path to instantiate ledger limits.
2280 * Forked processes do not come through this path, so no ledger limits exist.
2281 * (That's why forked processes can consume unlimited memory.)
2282 */
2283
2284 MEMORYSTATUS_DEBUG(3, "memorystatus_update(enter): pid %d, priority %d, dirty=0x%x, Active(%dMB %s), Inactive(%dMB, %s)\n",
2285 p->p_pid, priority, p->p_memstat_dirty,
2286 memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
2287 memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
2288
2289 if (memlimit_background) {
2290
2291 /*
2292 * With 2-level HWM support, we no longer honor P_MEMSTAT_MEMLIMIT_BACKGROUND.
2293 * Background limits are described via the inactive limit slots.
2294 */
2295
2296 // p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_BACKGROUND;
2297
2298 #if DEVELOPMENT || DEBUG
2299 printf("memorystatus_update: WARNING %s[%d] set unused flag P_MEMSTAT_MEMLIMIT_BACKGROUND [A==%dMB %s] [IA==%dMB %s]\n",
2300 (*p->p_name ? p->p_name : "unknown"), p->p_pid,
2301 memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
2302 memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
2303 #endif /* DEVELOPMENT || DEBUG */
2304 }
2305
2306 if (memlimit_active <= 0) {
2307 /*
2308 * This process will have a system_wide task limit when active.
2309 * System_wide task limit is always fatal.
2310 * It's quite common to see non-fatal flag passed in here.
2311 * It's not an error, we just ignore it.
2312 */
2313
2314 /*
2315 * For backward compatibility with some unexplained launchd behavior,
2316 * we allow a zero sized limit. But we still enforce system_wide limit
2317 * when written to the ledgers.
2318 */
2319
2320 if (memlimit_active < 0) {
2321 memlimit_active = -1; /* enforces system_wide task limit */
2322 }
2323 memlimit_active_is_fatal = TRUE;
2324 }
2325
2326 if (memlimit_inactive <= 0) {
2327 /*
2328 * This process will have a system_wide task limit when inactive.
2329 * System_wide task limit is always fatal.
2330 */
2331
2332 memlimit_inactive = -1;
2333 memlimit_inactive_is_fatal = TRUE;
2334 }
2335
2336 /*
2337 * Initialize the active limit variants for this process.
2338 */
2339 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
2340
2341 /*
2342 * Initialize the inactive limit variants for this process.
2343 */
2344 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
2345
2346 /*
2347 * Initialize the cached limits for target process.
2348 * When the target process is dirty tracked, it's typically
2349 * in a clean state. Non dirty tracked processes are
2350 * typically active (Foreground or above).
2351 * But just in case, we don't make assumptions...
2352 */
2353
2354 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
2355 CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
2356 } else {
2357 CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
2358 }
2359
2360 /*
2361 * Enforce the cached limit by writing to the ledger.
2362 */
2363 if (memorystatus_highwater_enabled) {
2364 /* apply now */
2365 assert(trigger_exception == TRUE);
2366 task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, trigger_exception);
2367
2368 MEMORYSTATUS_DEBUG(3, "memorystatus_update: init: limit on pid %d (%dMB %s) targeting priority(%d) dirty?=0x%x %s\n",
2369 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2370 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), priority, p->p_memstat_dirty,
2371 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2372 }
2373 }
2374
2375 /*
2376 * We can't add to the aging bands buckets here.
2377 * But, we could be removing it from those buckets.
2378 * Check and take appropriate steps if so.
2379 */
2380
2381 if (isProcessInAgingBands(p)) {
2382
2383 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2384 memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
2385 } else {
2386 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy && priority == JETSAM_PRIORITY_IDLE) {
2387 /*
2388 * Daemons with 'inactive' limits will go through the dirty tracking codepath.
2389 * This path deals with apps that may have 'inactive' limits e.g. WebContent processes.
2390 * If this is the legacy aging policy we explicitly need to apply those limits. If it
2391 * is any other aging policy, then we don't need to worry because all processes
2392 * will go through the aging bands and then the demotion thread will take care to
2393 * move them into the IDLE band and apply the required limits.
2394 */
2395 memorystatus_update_priority_locked(p, priority, head_insert, TRUE);
2396 }
2397 }
2398
2399 memorystatus_update_priority_locked(p, priority, head_insert, FALSE);
2400
2401 proc_list_unlock();
2402 ret = 0;
2403
2404 out:
2405 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_END, ret, 0, 0, 0, 0);
2406
2407 return ret;
2408 }
2409
2410 int
2411 memorystatus_remove(proc_t p, boolean_t locked)
2412 {
2413 int ret;
2414 memstat_bucket_t *bucket;
2415 boolean_t reschedule = FALSE;
2416
2417 MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing pid %d\n", p->p_pid);
2418
2419 if (!locked) {
2420 proc_list_lock();
2421 }
2422
2423 assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
2424
2425 bucket = &memstat_bucket[p->p_memstat_effectivepriority];
2426
2427 if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
2428
2429 assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs);
2430 reschedule = TRUE;
2431
2432 } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
2433
2434 assert(bucket->count == memorystatus_scheduled_idle_demotions_apps);
2435 reschedule = TRUE;
2436 }
2437
2438 /*
2439 * Record idle delta
2440 */
2441
2442 if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
2443 uint64_t now = mach_absolute_time();
2444 if (now > p->p_memstat_idle_start) {
2445 p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
2446 }
2447 }
2448
2449 TAILQ_REMOVE(&bucket->list, p, p_memstat_list);
2450 bucket->count--;
2451
2452 memorystatus_list_count--;
2453
2454 /* If awaiting demotion to the idle band, clean up */
2455 if (reschedule) {
2456 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2457 memorystatus_reschedule_idle_demotion_locked();
2458 }
2459
2460 memorystatus_check_levels_locked();
2461
2462 #if CONFIG_FREEZE
2463 if (p->p_memstat_state & (P_MEMSTAT_FROZEN)) {
2464 memorystatus_frozen_count--;
2465 }
2466
2467 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
2468 memorystatus_suspended_footprint_total -= p->p_memstat_suspendedfootprint;
2469 memorystatus_suspended_count--;
2470 }
2471 #endif
2472
2473 if (!locked) {
2474 proc_list_unlock();
2475 }
2476
2477 if (p) {
2478 ret = 0;
2479 } else {
2480 ret = ESRCH;
2481 }
2482
2483 return ret;
2484 }
2485
2486 /*
2487 * Validate dirty tracking flags with process state.
2488 *
2489 * Return:
2490 * 0 on success
2491 * non-0 on failure
2492 *
2493 * The proc_list_lock is held by the caller.
2494 */
2495
2496 static int
2497 memorystatus_validate_track_flags(struct proc *target_p, uint32_t pcontrol) {
2498 /* See that the process isn't marked for termination */
2499 if (target_p->p_memstat_dirty & P_DIRTY_TERMINATED) {
2500 return EBUSY;
2501 }
2502
2503 /* Idle exit requires that process be tracked */
2504 if ((pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) &&
2505 !(pcontrol & PROC_DIRTY_TRACK)) {
2506 return EINVAL;
2507 }
2508
2509 /* 'Launch in progress' tracking requires that process have enabled dirty tracking too. */
2510 if ((pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) &&
2511 !(pcontrol & PROC_DIRTY_TRACK)) {
2512 return EINVAL;
2513 }
2514
2515 /* Deferral is only relevant if idle exit is specified */
2516 if ((pcontrol & PROC_DIRTY_DEFER) &&
2517 !(pcontrol & PROC_DIRTY_ALLOWS_IDLE_EXIT)) {
2518 return EINVAL;
2519 }
2520
2521 return(0);
2522 }
2523
2524 static void
2525 memorystatus_update_idle_priority_locked(proc_t p) {
2526 int32_t priority;
2527
2528 MEMORYSTATUS_DEBUG(1, "memorystatus_update_idle_priority_locked(): pid %d dirty 0x%X\n", p->p_pid, p->p_memstat_dirty);
2529
2530 assert(isSysProc(p));
2531
2532 if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
2533
2534 priority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
2535 } else {
2536 priority = p->p_memstat_requestedpriority;
2537 }
2538
2539 if (priority != p->p_memstat_effectivepriority) {
2540
2541 if ((jetsam_aging_policy == kJetsamAgingPolicyLegacy) &&
2542 (priority == JETSAM_PRIORITY_IDLE)) {
2543
2544 /*
2545 * This process is on its way into the IDLE band. The system is
2546 * using 'legacy' jetsam aging policy. That means, this process
2547 * has already used up its idle-deferral aging time that is given
2548 * once per its lifetime. So we need to set the INACTIVE limits
2549 * explicitly because it won't be going through the demotion paths
2550 * that take care to apply the limits appropriately.
2551 */
2552 assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == 0);
2553 memorystatus_update_priority_locked(p, priority, false, true);
2554
2555 } else {
2556 memorystatus_update_priority_locked(p, priority, false, false);
2557 }
2558 }
2559 }
2560
2561 /*
2562 * Processes can opt to have their state tracked by the kernel, indicating when they are busy (dirty) or idle
2563 * (clean). They may also indicate that they support termination when idle, with the result that they are promoted
2564 * to their desired, higher, jetsam priority when dirty (and are therefore killed later), and demoted to the low
2565 * priority idle band when clean (and killed earlier, protecting higher priority procesess).
2566 *
2567 * If the deferral flag is set, then newly tracked processes will be protected for an initial period (as determined by
2568 * memorystatus_sysprocs_idle_delay_time); if they go clean during this time, then they will be moved to a deferred-idle band
2569 * with a slightly higher priority, guarding against immediate termination under memory pressure and being unable to
2570 * make forward progress. Finally, when the guard expires, they will be moved to the standard, lowest-priority, idle
2571 * band. The deferral can be cleared early by clearing the appropriate flag.
2572 *
2573 * The deferral timer is active only for the duration that the process is marked as guarded and clean; if the process
2574 * is marked dirty, the timer will be cancelled. Upon being subsequently marked clean, the deferment will either be
2575 * re-enabled or the guard state cleared, depending on whether the guard deadline has passed.
2576 */
2577
2578 int
2579 memorystatus_dirty_track(proc_t p, uint32_t pcontrol) {
2580 unsigned int old_dirty;
2581 boolean_t reschedule = FALSE;
2582 boolean_t already_deferred = FALSE;
2583 boolean_t defer_now = FALSE;
2584 int ret = 0;
2585
2586 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_TRACK),
2587 p->p_pid, p->p_memstat_dirty, pcontrol, 0, 0);
2588
2589 proc_list_lock();
2590
2591 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2592 /*
2593 * Process is on its way out.
2594 */
2595 ret = EBUSY;
2596 goto exit;
2597 }
2598
2599 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2600 ret = EPERM;
2601 goto exit;
2602 }
2603
2604 if ((ret = memorystatus_validate_track_flags(p, pcontrol)) != 0) {
2605 /* error */
2606 goto exit;
2607 }
2608
2609 old_dirty = p->p_memstat_dirty;
2610
2611 /* These bits are cumulative, as per <rdar://problem/11159924> */
2612 if (pcontrol & PROC_DIRTY_TRACK) {
2613 p->p_memstat_dirty |= P_DIRTY_TRACK;
2614 }
2615
2616 if (pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) {
2617 p->p_memstat_dirty |= P_DIRTY_ALLOW_IDLE_EXIT;
2618 }
2619
2620 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
2621 p->p_memstat_dirty |= P_DIRTY_LAUNCH_IN_PROGRESS;
2622 }
2623
2624 if (old_dirty & P_DIRTY_AGING_IN_PROGRESS) {
2625 already_deferred = TRUE;
2626 }
2627
2628
2629 /* This can be set and cleared exactly once. */
2630 if (pcontrol & PROC_DIRTY_DEFER) {
2631
2632 if ( !(old_dirty & P_DIRTY_DEFER)) {
2633 p->p_memstat_dirty |= P_DIRTY_DEFER;
2634 }
2635
2636 defer_now = TRUE;
2637 }
2638
2639 MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for pid %d\n",
2640 ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) ? "Y" : "N",
2641 defer_now ? "Y" : "N",
2642 p->p_memstat_dirty & P_DIRTY ? "Y" : "N",
2643 p->p_pid);
2644
2645 /* Kick off or invalidate the idle exit deferment if there's a state transition. */
2646 if (!(p->p_memstat_dirty & P_DIRTY_IS_DIRTY)) {
2647 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
2648
2649 if (defer_now && !already_deferred) {
2650
2651 /*
2652 * Request to defer a clean process that's idle-exit enabled
2653 * and not already in the jetsam deferred band. Most likely a
2654 * new launch.
2655 */
2656 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2657 reschedule = TRUE;
2658
2659 } else if (!defer_now) {
2660
2661 /*
2662 * The process isn't asking for the 'aging' facility.
2663 * Could be that it is:
2664 */
2665
2666 if (already_deferred) {
2667 /*
2668 * already in the aging bands. Traditionally,
2669 * some processes have tried to use this to
2670 * opt out of the 'aging' facility.
2671 */
2672
2673 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2674 } else {
2675 /*
2676 * agnostic to the 'aging' facility. In that case,
2677 * we'll go ahead and opt it in because this is likely
2678 * a new launch (clean process, dirty tracking enabled)
2679 */
2680
2681 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2682 }
2683
2684 reschedule = TRUE;
2685 }
2686 }
2687 } else {
2688
2689 /*
2690 * We are trying to operate on a dirty process. Dirty processes have to
2691 * be removed from the deferred band. The question is do we reset the
2692 * deferred state or not?
2693 *
2694 * This could be a legal request like:
2695 * - this process had opted into the 'aging' band
2696 * - but it's now dirty and requests to opt out.
2697 * In this case, we remove the process from the band and reset its
2698 * state too. It'll opt back in properly when needed.
2699 *
2700 * OR, this request could be a user-space bug. E.g.:
2701 * - this process had opted into the 'aging' band when clean
2702 * - and, then issues another request to again put it into the band except
2703 * this time the process is dirty.
2704 * The process going dirty, as a transition in memorystatus_dirty_set(), will pull the process out of
2705 * the deferred band with its state intact. So our request below is no-op.
2706 * But we do it here anyways for coverage.
2707 *
2708 * memorystatus_update_idle_priority_locked()
2709 * single-mindedly treats a dirty process as "cannot be in the aging band".
2710 */
2711
2712 if (!defer_now && already_deferred) {
2713 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2714 reschedule = TRUE;
2715 } else {
2716
2717 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
2718
2719 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
2720 reschedule = TRUE;
2721 }
2722 }
2723
2724 memorystatus_update_idle_priority_locked(p);
2725
2726 if (reschedule) {
2727 memorystatus_reschedule_idle_demotion_locked();
2728 }
2729
2730 ret = 0;
2731
2732 exit:
2733 proc_list_unlock();
2734
2735 return ret;
2736 }
2737
2738 int
2739 memorystatus_dirty_set(proc_t p, boolean_t self, uint32_t pcontrol) {
2740 int ret;
2741 boolean_t kill = false;
2742 boolean_t reschedule = FALSE;
2743 boolean_t was_dirty = FALSE;
2744 boolean_t now_dirty = FALSE;
2745
2746 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_set(): %d %d 0x%x 0x%x\n", self, p->p_pid, pcontrol, p->p_memstat_dirty);
2747 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_SET), p->p_pid, self, pcontrol, 0, 0);
2748
2749 proc_list_lock();
2750
2751 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2752 /*
2753 * Process is on its way out.
2754 */
2755 ret = EBUSY;
2756 goto exit;
2757 }
2758
2759 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2760 ret = EPERM;
2761 goto exit;
2762 }
2763
2764 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY)
2765 was_dirty = TRUE;
2766
2767 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
2768 /* Dirty tracking not enabled */
2769 ret = EINVAL;
2770 } else if (pcontrol && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
2771 /*
2772 * Process is set to be terminated and we're attempting to mark it dirty.
2773 * Set for termination and marking as clean is OK - see <rdar://problem/10594349>.
2774 */
2775 ret = EBUSY;
2776 } else {
2777 int flag = (self == TRUE) ? P_DIRTY : P_DIRTY_SHUTDOWN;
2778 if (pcontrol && !(p->p_memstat_dirty & flag)) {
2779 /* Mark the process as having been dirtied at some point */
2780 p->p_memstat_dirty |= (flag | P_DIRTY_MARKED);
2781 memorystatus_dirty_count++;
2782 ret = 0;
2783 } else if ((pcontrol == 0) && (p->p_memstat_dirty & flag)) {
2784 if ((flag == P_DIRTY_SHUTDOWN) && (!(p->p_memstat_dirty & P_DIRTY))) {
2785 /* Clearing the dirty shutdown flag, and the process is otherwise clean - kill */
2786 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
2787 kill = true;
2788 } else if ((flag == P_DIRTY) && (p->p_memstat_dirty & P_DIRTY_TERMINATED)) {
2789 /* Kill previously terminated processes if set clean */
2790 kill = true;
2791 }
2792 p->p_memstat_dirty &= ~flag;
2793 memorystatus_dirty_count--;
2794 ret = 0;
2795 } else {
2796 /* Already set */
2797 ret = EALREADY;
2798 }
2799 }
2800
2801 if (ret != 0) {
2802 goto exit;
2803 }
2804
2805 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY)
2806 now_dirty = TRUE;
2807
2808 if ((was_dirty == TRUE && now_dirty == FALSE) ||
2809 (was_dirty == FALSE && now_dirty == TRUE)) {
2810
2811 /* Manage idle exit deferral, if applied */
2812 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
2813
2814 /*
2815 * Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band OR it might be heading back
2816 * there once it's clean again. For the legacy case, this only applies if it has some protection window left.
2817 *
2818 * Non-Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band. It will always stop over
2819 * in that band on it's way to IDLE.
2820 */
2821
2822 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
2823 /*
2824 * New dirty process i.e. "was_dirty == FALSE && now_dirty == TRUE"
2825 *
2826 * The process will move from its aging band to its higher requested
2827 * jetsam band.
2828 */
2829 boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
2830
2831 memorystatus_invalidate_idle_demotion_locked(p, reset_state);
2832 reschedule = TRUE;
2833 } else {
2834
2835 /*
2836 * Process is back from "dirty" to "clean".
2837 */
2838
2839 if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
2840 if (mach_absolute_time() >= p->p_memstat_idledeadline) {
2841 /*
2842 * The process' deadline has expired. It currently
2843 * does not reside in any of the aging buckets.
2844 *
2845 * It's on its way to the JETSAM_PRIORITY_IDLE
2846 * bucket via memorystatus_update_idle_priority_locked()
2847 * below.
2848
2849 * So all we need to do is reset all the state on the
2850 * process that's related to the aging bucket i.e.
2851 * the AGING_IN_PROGRESS flag and the timer deadline.
2852 */
2853
2854 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
2855 reschedule = TRUE;
2856 } else {
2857 /*
2858 * It still has some protection window left and so
2859 * we just re-arm the timer without modifying any
2860 * state on the process iff it still wants into that band.
2861 */
2862
2863 if (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) {
2864 memorystatus_schedule_idle_demotion_locked(p, FALSE);
2865 reschedule = TRUE;
2866 }
2867 }
2868 } else {
2869
2870 memorystatus_schedule_idle_demotion_locked(p, TRUE);
2871 reschedule = TRUE;
2872 }
2873 }
2874 }
2875
2876 memorystatus_update_idle_priority_locked(p);
2877
2878 if (memorystatus_highwater_enabled) {
2879 boolean_t trigger_exception = FALSE, ledger_update_needed = TRUE;
2880 /*
2881 * We are in this path because this process transitioned between
2882 * dirty <--> clean state. Update the cached memory limits.
2883 */
2884
2885 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
2886 /*
2887 * process is dirty
2888 */
2889 CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
2890 ledger_update_needed = TRUE;
2891 } else {
2892 /*
2893 * process is clean...but if it has opted into pressured-exit
2894 * we don't apply the INACTIVE limit till the process has aged
2895 * out and is entering the IDLE band.
2896 * See memorystatus_update_priority_locked() for that.
2897 */
2898
2899 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
2900 ledger_update_needed = FALSE;
2901 } else {
2902 CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
2903 ledger_update_needed = TRUE;
2904 }
2905 }
2906
2907 /*
2908 * Enforce the new limits by writing to the ledger.
2909 *
2910 * This is a hot path and holding the proc_list_lock while writing to the ledgers,
2911 * (where the task lock is taken) is bad. So, we temporarily drop the proc_list_lock.
2912 * We aren't traversing the jetsam bucket list here, so we should be safe.
2913 * See rdar://21394491.
2914 */
2915
2916 if (ledger_update_needed && proc_ref_locked(p) == p) {
2917 int ledger_limit;
2918 if (p->p_memstat_memlimit > 0) {
2919 ledger_limit = p->p_memstat_memlimit;
2920 } else {
2921 ledger_limit = -1;
2922 }
2923 proc_list_unlock();
2924 task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, trigger_exception);
2925 proc_list_lock();
2926 proc_rele_locked(p);
2927
2928 MEMORYSTATUS_DEBUG(3, "memorystatus_dirty_set: new limit on pid %d (%dMB %s) priority(%d) dirty?=0x%x %s\n",
2929 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
2930 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
2931 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
2932 }
2933
2934 }
2935
2936 /* If the deferral state changed, reschedule the demotion timer */
2937 if (reschedule) {
2938 memorystatus_reschedule_idle_demotion_locked();
2939 }
2940 }
2941
2942 if (kill) {
2943 if (proc_ref_locked(p) == p) {
2944 proc_list_unlock();
2945 psignal(p, SIGKILL);
2946 proc_list_lock();
2947 proc_rele_locked(p);
2948 }
2949 }
2950
2951 exit:
2952 proc_list_unlock();
2953
2954 return ret;
2955 }
2956
2957 int
2958 memorystatus_dirty_clear(proc_t p, uint32_t pcontrol) {
2959
2960 int ret = 0;
2961
2962 MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_clear(): %d 0x%x 0x%x\n", p->p_pid, pcontrol, p->p_memstat_dirty);
2963
2964 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_CLEAR), p->p_pid, pcontrol, 0, 0, 0);
2965
2966 proc_list_lock();
2967
2968 if ((p->p_listflag & P_LIST_EXITED) != 0) {
2969 /*
2970 * Process is on its way out.
2971 */
2972 ret = EBUSY;
2973 goto exit;
2974 }
2975
2976 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
2977 ret = EPERM;
2978 goto exit;
2979 }
2980
2981 if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
2982 /* Dirty tracking not enabled */
2983 ret = EINVAL;
2984 goto exit;
2985 }
2986
2987 if (!pcontrol || (pcontrol & (PROC_DIRTY_LAUNCH_IN_PROGRESS | PROC_DIRTY_DEFER)) == 0) {
2988 ret = EINVAL;
2989 goto exit;
2990 }
2991
2992 if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
2993 p->p_memstat_dirty &= ~P_DIRTY_LAUNCH_IN_PROGRESS;
2994 }
2995
2996 /* This can be set and cleared exactly once. */
2997 if (pcontrol & PROC_DIRTY_DEFER) {
2998
2999 if (p->p_memstat_dirty & P_DIRTY_DEFER) {
3000
3001 p->p_memstat_dirty &= ~P_DIRTY_DEFER;
3002
3003 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
3004 memorystatus_update_idle_priority_locked(p);
3005 memorystatus_reschedule_idle_demotion_locked();
3006 }
3007 }
3008
3009 ret = 0;
3010 exit:
3011 proc_list_unlock();
3012
3013 return ret;
3014 }
3015
3016 int
3017 memorystatus_dirty_get(proc_t p) {
3018 int ret = 0;
3019
3020 proc_list_lock();
3021
3022 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3023 ret |= PROC_DIRTY_TRACKED;
3024 if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
3025 ret |= PROC_DIRTY_ALLOWS_IDLE_EXIT;
3026 }
3027 if (p->p_memstat_dirty & P_DIRTY) {
3028 ret |= PROC_DIRTY_IS_DIRTY;
3029 }
3030 if (p->p_memstat_dirty & P_DIRTY_LAUNCH_IN_PROGRESS) {
3031 ret |= PROC_DIRTY_LAUNCH_IS_IN_PROGRESS;
3032 }
3033 }
3034
3035 proc_list_unlock();
3036
3037 return ret;
3038 }
3039
3040 int
3041 memorystatus_on_terminate(proc_t p) {
3042 int sig;
3043
3044 proc_list_lock();
3045
3046 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3047
3048 if ((p->p_memstat_dirty & (P_DIRTY_TRACK|P_DIRTY_IS_DIRTY)) == P_DIRTY_TRACK) {
3049 /* Clean; mark as terminated and issue SIGKILL */
3050 sig = SIGKILL;
3051 } else {
3052 /* Dirty, terminated, or state tracking is unsupported; issue SIGTERM to allow cleanup */
3053 sig = SIGTERM;
3054 }
3055
3056 proc_list_unlock();
3057
3058 return sig;
3059 }
3060
3061 void
3062 memorystatus_on_suspend(proc_t p)
3063 {
3064 #if CONFIG_FREEZE
3065 uint32_t pages;
3066 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
3067 #endif
3068 proc_list_lock();
3069 #if CONFIG_FREEZE
3070 p->p_memstat_suspendedfootprint = pages;
3071 memorystatus_suspended_footprint_total += pages;
3072 memorystatus_suspended_count++;
3073 #endif
3074 p->p_memstat_state |= P_MEMSTAT_SUSPENDED;
3075 proc_list_unlock();
3076 }
3077
3078 void
3079 memorystatus_on_resume(proc_t p)
3080 {
3081 #if CONFIG_FREEZE
3082 boolean_t frozen;
3083 pid_t pid;
3084 #endif
3085
3086 proc_list_lock();
3087
3088 #if CONFIG_FREEZE
3089 frozen = (p->p_memstat_state & P_MEMSTAT_FROZEN);
3090 if (frozen) {
3091 memorystatus_frozen_count--;
3092 p->p_memstat_state |= P_MEMSTAT_PRIOR_THAW;
3093 }
3094
3095 memorystatus_suspended_footprint_total -= p->p_memstat_suspendedfootprint;
3096 memorystatus_suspended_count--;
3097
3098 pid = p->p_pid;
3099 #endif
3100
3101 p->p_memstat_state &= ~(P_MEMSTAT_SUSPENDED | P_MEMSTAT_FROZEN);
3102
3103 proc_list_unlock();
3104
3105 #if CONFIG_FREEZE
3106 if (frozen) {
3107 memorystatus_freeze_entry_t data = { pid, FALSE, 0 };
3108 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
3109 }
3110 #endif
3111 }
3112
3113 void
3114 memorystatus_on_inactivity(proc_t p)
3115 {
3116 #pragma unused(p)
3117 #if CONFIG_FREEZE
3118 /* Wake the freeze thread */
3119 thread_wakeup((event_t)&memorystatus_freeze_wakeup);
3120 #endif
3121 }
3122
3123 /*
3124 * The proc_list_lock is held by the caller.
3125 */
3126 static uint32_t
3127 memorystatus_build_state(proc_t p) {
3128 uint32_t snapshot_state = 0;
3129
3130 /* General */
3131 if (p->p_memstat_state & P_MEMSTAT_SUSPENDED) {
3132 snapshot_state |= kMemorystatusSuspended;
3133 }
3134 if (p->p_memstat_state & P_MEMSTAT_FROZEN) {
3135 snapshot_state |= kMemorystatusFrozen;
3136 }
3137 if (p->p_memstat_state & P_MEMSTAT_PRIOR_THAW) {
3138 snapshot_state |= kMemorystatusWasThawed;
3139 }
3140
3141 /* Tracking */
3142 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3143 snapshot_state |= kMemorystatusTracked;
3144 }
3145 if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
3146 snapshot_state |= kMemorystatusSupportsIdleExit;
3147 }
3148 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3149 snapshot_state |= kMemorystatusDirty;
3150 }
3151
3152 return snapshot_state;
3153 }
3154
3155 #if !CONFIG_JETSAM
3156
3157 static boolean_t
3158 kill_idle_exit_proc(void)
3159 {
3160 proc_t p, victim_p = PROC_NULL;
3161 uint64_t current_time;
3162 boolean_t killed = FALSE;
3163 unsigned int i = 0;
3164 os_reason_t jetsam_reason = OS_REASON_NULL;
3165
3166 /* Pick next idle exit victim. */
3167 current_time = mach_absolute_time();
3168
3169 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_IDLE_EXIT);
3170 if (jetsam_reason == OS_REASON_NULL) {
3171 printf("kill_idle_exit_proc: failed to allocate jetsam reason\n");
3172 }
3173
3174 proc_list_lock();
3175
3176 p = memorystatus_get_first_proc_locked(&i, FALSE);
3177 while (p) {
3178 /* No need to look beyond the idle band */
3179 if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
3180 break;
3181 }
3182
3183 if ((p->p_memstat_dirty & (P_DIRTY_ALLOW_IDLE_EXIT|P_DIRTY_IS_DIRTY|P_DIRTY_TERMINATED)) == (P_DIRTY_ALLOW_IDLE_EXIT)) {
3184 if (current_time >= p->p_memstat_idledeadline) {
3185 p->p_memstat_dirty |= P_DIRTY_TERMINATED;
3186 victim_p = proc_ref_locked(p);
3187 break;
3188 }
3189 }
3190
3191 p = memorystatus_get_next_proc_locked(&i, p, FALSE);
3192 }
3193
3194 proc_list_unlock();
3195
3196 if (victim_p) {
3197 printf("memorystatus_thread: idle exiting pid %d [%s]\n", victim_p->p_pid, (*victim_p->p_name ? victim_p->p_name : "(unknown)"));
3198 killed = memorystatus_do_kill(victim_p, kMemorystatusKilledIdleExit, jetsam_reason);
3199 proc_rele(victim_p);
3200 } else {
3201 os_reason_free(jetsam_reason);
3202 }
3203
3204 return killed;
3205 }
3206 #endif
3207
3208 #if CONFIG_JETSAM
3209 static void
3210 memorystatus_thread_wake(void) {
3211 thread_wakeup((event_t)&memorystatus_wakeup);
3212 }
3213 #endif /* CONFIG_JETSAM */
3214
3215 extern void vm_pressure_response(void);
3216
3217 static int
3218 memorystatus_thread_block(uint32_t interval_ms, thread_continue_t continuation)
3219 {
3220 if (interval_ms) {
3221 assert_wait_timeout(&memorystatus_wakeup, THREAD_UNINT, interval_ms, 1000 * NSEC_PER_USEC);
3222 } else {
3223 assert_wait(&memorystatus_wakeup, THREAD_UNINT);
3224 }
3225
3226 return thread_block(continuation);
3227 }
3228
3229 static void
3230 memorystatus_thread(void *param __unused, wait_result_t wr __unused)
3231 {
3232 static boolean_t is_vm_privileged = FALSE;
3233
3234 #if CONFIG_JETSAM
3235 boolean_t post_snapshot = FALSE;
3236 uint32_t errors = 0;
3237 uint32_t hwm_kill = 0;
3238 boolean_t sort_flag = TRUE;
3239 boolean_t corpse_list_purged = FALSE;
3240
3241 /* Jetsam Loop Detection - locals */
3242 memstat_bucket_t *bucket;
3243 int jld_bucket_count = 0;
3244 struct timeval jld_now_tstamp = {0,0};
3245 uint64_t jld_now_msecs = 0;
3246 int elevated_bucket_count = 0;
3247
3248 /* Jetsam Loop Detection - statics */
3249 static uint64_t jld_timestamp_msecs = 0;
3250 static int jld_idle_kill_candidates = 0; /* Number of available processes in band 0,1 at start */
3251 static int jld_idle_kills = 0; /* Number of procs killed during eval period */
3252 static int jld_eval_aggressive_count = 0; /* Bumps the max priority in aggressive loop */
3253 static int32_t jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
3254 #endif
3255
3256 if (is_vm_privileged == FALSE) {
3257 /*
3258 * It's the first time the thread has run, so just mark the thread as privileged and block.
3259 * This avoids a spurious pass with unset variables, as set out in <rdar://problem/9609402>.
3260 */
3261 thread_wire(host_priv_self(), current_thread(), TRUE);
3262 is_vm_privileged = TRUE;
3263
3264 if (vm_restricted_to_single_processor == TRUE)
3265 thread_vm_bind_group_add();
3266
3267 memorystatus_thread_block(0, memorystatus_thread);
3268 }
3269
3270 #if CONFIG_JETSAM
3271
3272 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_START,
3273 memorystatus_available_pages, memorystatus_jld_enabled, memorystatus_jld_eval_period_msecs, memorystatus_jld_eval_aggressive_count,0);
3274
3275 /*
3276 * Jetsam aware version.
3277 *
3278 * The VM pressure notification thread is working it's way through clients in parallel.
3279 *
3280 * So, while the pressure notification thread is targeting processes in order of
3281 * increasing jetsam priority, we can hopefully reduce / stop it's work by killing
3282 * any processes that have exceeded their highwater mark.
3283 *
3284 * If we run out of HWM processes and our available pages drops below the critical threshold, then,
3285 * we target the least recently used process in order of increasing jetsam priority (exception: the FG band).
3286 */
3287 while (is_thrashing(kill_under_pressure_cause) ||
3288 memorystatus_available_pages <= memorystatus_available_pages_pressure) {
3289 boolean_t killed;
3290 int32_t priority;
3291 uint32_t cause;
3292 uint64_t jetsam_reason_code = JETSAM_REASON_INVALID;
3293 os_reason_t jetsam_reason = OS_REASON_NULL;
3294
3295 cause = kill_under_pressure_cause;
3296 switch (cause) {
3297 case kMemorystatusKilledFCThrashing:
3298 jetsam_reason_code = JETSAM_REASON_MEMORY_FCTHRASHING;
3299 break;
3300 case kMemorystatusKilledVMThrashing:
3301 jetsam_reason_code = JETSAM_REASON_MEMORY_VMTHRASHING;
3302 break;
3303 case kMemorystatusKilledVMPageShortage:
3304 /* falls through */
3305 default:
3306 jetsam_reason_code = JETSAM_REASON_MEMORY_VMPAGESHORTAGE;
3307 cause = kMemorystatusKilledVMPageShortage;
3308 break;
3309 }
3310
3311 /* Highwater */
3312 killed = memorystatus_kill_hiwat_proc(&errors);
3313 if (killed) {
3314 hwm_kill++;
3315 post_snapshot = TRUE;
3316 goto done;
3317 } else {
3318 memorystatus_hwm_candidates = FALSE;
3319 }
3320
3321 /* No highwater processes to kill. Continue or stop for now? */
3322 if (!is_thrashing(kill_under_pressure_cause) &&
3323 (memorystatus_available_pages > memorystatus_available_pages_critical)) {
3324 /*
3325 * We are _not_ out of pressure but we are above the critical threshold and there's:
3326 * - no compressor thrashing
3327 * - no more HWM processes left.
3328 * For now, don't kill any other processes.
3329 */
3330
3331 if (hwm_kill == 0) {
3332 memorystatus_thread_wasted_wakeup++;
3333 }
3334
3335 break;
3336 }
3337
3338 jetsam_reason = os_reason_create(OS_REASON_JETSAM, jetsam_reason_code);
3339 if (jetsam_reason == OS_REASON_NULL) {
3340 printf("memorystatus_thread: failed to allocate jetsam reason\n");
3341 }
3342
3343 if (memorystatus_jld_enabled == TRUE) {
3344
3345 /*
3346 * Jetsam Loop Detection: attempt to detect
3347 * rapid daemon relaunches in the lower bands.
3348 */
3349
3350 microuptime(&jld_now_tstamp);
3351
3352 /*
3353 * Ignore usecs in this calculation.
3354 * msecs granularity is close enough.
3355 */
3356 jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
3357
3358 proc_list_lock();
3359 switch (jetsam_aging_policy) {
3360 case kJetsamAgingPolicyLegacy:
3361 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3362 jld_bucket_count = bucket->count;
3363 bucket = &memstat_bucket[JETSAM_PRIORITY_AGING_BAND1];
3364 jld_bucket_count += bucket->count;
3365 break;
3366 case kJetsamAgingPolicySysProcsReclaimedFirst:
3367 case kJetsamAgingPolicyAppsReclaimedFirst:
3368 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3369 jld_bucket_count = bucket->count;
3370 bucket = &memstat_bucket[system_procs_aging_band];
3371 jld_bucket_count += bucket->count;
3372 bucket = &memstat_bucket[applications_aging_band];
3373 jld_bucket_count += bucket->count;
3374 break;
3375 case kJetsamAgingPolicyNone:
3376 default:
3377 bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
3378 jld_bucket_count = bucket->count;
3379 break;
3380 }
3381
3382 bucket = &memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE];
3383 elevated_bucket_count = bucket->count;
3384
3385 proc_list_unlock();
3386
3387 /*
3388 * memorystatus_jld_eval_period_msecs is a tunable
3389 * memorystatus_jld_eval_aggressive_count is a tunable
3390 * memorystatus_jld_eval_aggressive_priority_band_max is a tunable
3391 */
3392 if ( (jld_bucket_count == 0) ||
3393 (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
3394
3395 /*
3396 * Refresh evaluation parameters
3397 */
3398 jld_timestamp_msecs = jld_now_msecs;
3399 jld_idle_kill_candidates = jld_bucket_count;
3400 jld_idle_kills = 0;
3401 jld_eval_aggressive_count = 0;
3402 jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
3403 }
3404
3405 if (jld_idle_kills > jld_idle_kill_candidates) {
3406 jld_eval_aggressive_count++;
3407
3408 #if DEVELOPMENT || DEBUG
3409 printf("memorystatus: aggressive%d: beginning of window: %lld ms, : timestamp now: %lld ms\n",
3410 jld_eval_aggressive_count,
3411 jld_timestamp_msecs,
3412 jld_now_msecs);
3413 printf("memorystatus: aggressive%d: idle candidates: %d, idle kills: %d\n",
3414 jld_eval_aggressive_count,
3415 jld_idle_kill_candidates,
3416 jld_idle_kills);
3417 #endif /* DEVELOPMENT || DEBUG */
3418
3419 if ((jld_eval_aggressive_count == memorystatus_jld_eval_aggressive_count) &&
3420 (total_corpses_count > 0) && (corpse_list_purged == FALSE)) {
3421 /*
3422 * If we reach this aggressive cycle, corpses might be causing memory pressure.
3423 * So, in an effort to avoid jetsams in the FG band, we will attempt to purge
3424 * corpse memory prior to this final march through JETSAM_PRIORITY_UI_SUPPORT.
3425 */
3426 task_purge_all_corpses();
3427 corpse_list_purged = TRUE;
3428 }
3429 else if (jld_eval_aggressive_count > memorystatus_jld_eval_aggressive_count) {
3430 /*
3431 * Bump up the jetsam priority limit (eg: the bucket index)
3432 * Enforce bucket index sanity.
3433 */
3434 if ((memorystatus_jld_eval_aggressive_priority_band_max < 0) ||
3435 (memorystatus_jld_eval_aggressive_priority_band_max >= MEMSTAT_BUCKET_COUNT)) {
3436 /*
3437 * Do nothing. Stick with the default level.
3438 */
3439 } else {
3440 jld_priority_band_max = memorystatus_jld_eval_aggressive_priority_band_max;
3441 }
3442 }
3443
3444 /* Visit elevated processes first */
3445 while (elevated_bucket_count) {
3446
3447 elevated_bucket_count--;
3448
3449 /*
3450 * memorystatus_kill_elevated_process() drops a reference,
3451 * so take another one so we can continue to use this exit reason
3452 * even after it returns.
3453 */
3454
3455 os_reason_ref(jetsam_reason);
3456 killed = memorystatus_kill_elevated_process(
3457 kMemorystatusKilledVMThrashing,
3458 jetsam_reason,
3459 jld_eval_aggressive_count,
3460 &errors);
3461
3462 if (killed) {
3463 post_snapshot = TRUE;
3464 if (memorystatus_available_pages <= memorystatus_available_pages_pressure) {
3465 /*
3466 * Still under pressure.
3467 * Find another pinned processes.
3468 */
3469 continue;
3470 } else {
3471 goto done;
3472 }
3473 } else {
3474 /*
3475 * No pinned processes left to kill.
3476 * Abandon elevated band.
3477 */
3478 break;
3479 }
3480 }
3481
3482 /*
3483 * memorystatus_kill_top_process_aggressive() drops a reference,
3484 * so take another one so we can continue to use this exit reason
3485 * even after it returns
3486 */
3487 os_reason_ref(jetsam_reason);
3488 killed = memorystatus_kill_top_process_aggressive(
3489 TRUE,
3490 kMemorystatusKilledVMThrashing,
3491 jetsam_reason,
3492 jld_eval_aggressive_count,
3493 jld_priority_band_max,
3494 &errors);
3495
3496 if (killed) {
3497 /* Always generate logs after aggressive kill */
3498 post_snapshot = TRUE;
3499 jld_idle_kills = 0;
3500 goto done;
3501 }
3502 }
3503 }
3504
3505 /*
3506 * memorystatus_kill_top_process() drops a reference,
3507 * so take another one so we can continue to use this exit reason
3508 * even after it returns
3509 */
3510 os_reason_ref(jetsam_reason);
3511
3512 /* LRU */
3513 killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, jetsam_reason, &priority, &errors);
3514 sort_flag = FALSE;
3515
3516 if (killed) {
3517 /*
3518 * Don't generate logs for steady-state idle-exit kills,
3519 * unless it is overridden for debug or by the device
3520 * tree.
3521 */
3522 if ((priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot) {
3523 post_snapshot = TRUE;
3524 }
3525
3526 /* Jetsam Loop Detection */
3527 if (memorystatus_jld_enabled == TRUE) {
3528 if ((priority == JETSAM_PRIORITY_IDLE) || (priority == system_procs_aging_band) || (priority == applications_aging_band)) {
3529 jld_idle_kills++;
3530 } else {
3531 /*
3532 * We've reached into bands beyond idle deferred.
3533 * We make no attempt to monitor them
3534 */
3535 }
3536 }
3537
3538 if ((priority >= JETSAM_PRIORITY_UI_SUPPORT) && (total_corpses_count > 0) && (corpse_list_purged == FALSE)) {
3539 /*
3540 * If we have jetsammed a process in or above JETSAM_PRIORITY_UI_SUPPORT
3541 * then we attempt to relieve pressure by purging corpse memory.
3542 */
3543 task_purge_all_corpses();
3544 corpse_list_purged = TRUE;
3545 }
3546 goto done;
3547 }
3548
3549 if (memorystatus_available_pages <= memorystatus_available_pages_critical) {
3550 /*
3551 * Still under pressure and unable to kill a process - purge corpse memory
3552 */
3553 if (total_corpses_count > 0) {
3554 task_purge_all_corpses();
3555 corpse_list_purged = TRUE;
3556 }
3557
3558 if (memorystatus_available_pages <= memorystatus_available_pages_critical) {
3559 /*
3560 * Still under pressure and unable to kill a process - panic
3561 */
3562 panic("memorystatus_jetsam_thread: no victim! available pages:%d\n", memorystatus_available_pages);
3563 }
3564 }
3565
3566 done:
3567
3568 /*
3569 * We do not want to over-kill when thrashing has been detected.
3570 * To avoid that, we reset the flag here and notify the
3571 * compressor.
3572 */
3573 if (is_thrashing(kill_under_pressure_cause)) {
3574 kill_under_pressure_cause = 0;
3575 vm_thrashing_jetsam_done();
3576 }
3577
3578 os_reason_free(jetsam_reason);
3579 }
3580
3581 kill_under_pressure_cause = 0;
3582
3583 if (errors) {
3584 memorystatus_clear_errors();
3585 }
3586
3587 #if VM_PRESSURE_EVENTS
3588 /*
3589 * LD: We used to target the foreground process first and foremost here.
3590 * Now, we target all processes, starting from the non-suspended, background
3591 * processes first. We will target foreground too.
3592 *
3593 * memorystatus_update_vm_pressure(TRUE);
3594 */
3595 //vm_pressure_response();
3596 #endif
3597
3598 if (post_snapshot) {
3599 proc_list_lock();
3600 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
3601 sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
3602 uint64_t timestamp_now = mach_absolute_time();
3603 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
3604 memorystatus_jetsam_snapshot->js_gencount++;
3605 if (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
3606 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout) {
3607 proc_list_unlock();
3608 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
3609 if (!ret) {
3610 proc_list_lock();
3611 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
3612 proc_list_unlock();
3613 }
3614 } else {
3615 proc_list_unlock();
3616 }
3617 }
3618
3619 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_END,
3620 memorystatus_available_pages, 0, 0, 0, 0);
3621
3622 #else /* CONFIG_JETSAM */
3623
3624 /*
3625 * Jetsam not enabled
3626 */
3627
3628 #endif /* CONFIG_JETSAM */
3629
3630 memorystatus_thread_block(0, memorystatus_thread);
3631 }
3632
3633 #if !CONFIG_JETSAM
3634 /*
3635 * Returns TRUE:
3636 * when an idle-exitable proc was killed
3637 * Returns FALSE:
3638 * when there are no more idle-exitable procs found
3639 * when the attempt to kill an idle-exitable proc failed
3640 */
3641 boolean_t memorystatus_idle_exit_from_VM(void) {
3642 return(kill_idle_exit_proc());
3643 }
3644 #endif /* !CONFIG_JETSAM */
3645
3646 /*
3647 * Returns TRUE:
3648 * when exceeding ledger footprint is fatal.
3649 * Returns FALSE:
3650 * when exceeding ledger footprint is non fatal.
3651 */
3652 boolean_t
3653 memorystatus_turnoff_exception_and_get_fatalness(boolean_t warning, const int max_footprint_mb)
3654 {
3655 proc_t p = current_proc();
3656 boolean_t is_fatal;
3657
3658 proc_list_lock();
3659
3660 is_fatal = (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT);
3661
3662 if (warning == FALSE) {
3663 boolean_t is_active;
3664 boolean_t state_changed = FALSE;
3665
3666 /*
3667 * We are here because a process has exceeded its ledger limit.
3668 * That is, the process is no longer in the limit warning range.
3669 *
3670 * When a process exceeds its ledger limit, we want an EXC_RESOURCE
3671 * to trigger, but only once per process per limit. We enforce that
3672 * here, by identifying the active/inactive limit type. We then turn
3673 * off the exception state by marking the limit as exception triggered.
3674 */
3675
3676 is_active = proc_jetsam_state_is_active_locked(p);
3677
3678 if (is_active == TRUE) {
3679 /*
3680 * turn off exceptions for active state
3681 */
3682 if (!(p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED)) {
3683 p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED;
3684 state_changed = TRUE;
3685 }
3686 } else {
3687 /*
3688 * turn off exceptions for inactive state
3689 */
3690 if (!(p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED)) {
3691 p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED;
3692 state_changed = TRUE;
3693 }
3694 }
3695
3696 /*
3697 * The limit violation is logged here, but only once per process per limit.
3698 * This avoids excessive logging when a process consistently exceeds a soft limit.
3699 * Soft memory limit is a non-fatal high-water-mark
3700 * Hard memory limit is a fatal custom-task-limit or system-wide per-task memory limit.
3701 */
3702 if(state_changed) {
3703 printf("process %d (%s) exceeded physical memory footprint, the %s%sMemoryLimit of %d MB\n",
3704 p->p_pid, (*p->p_name ? p->p_name : "unknown"), (is_active ? "Active" : "Inactive"),
3705 (is_fatal ? "Hard" : "Soft"), max_footprint_mb);
3706 }
3707
3708 }
3709 proc_list_unlock();
3710
3711 return is_fatal;
3712 }
3713
3714 /*
3715 * Callback invoked when allowable physical memory footprint exceeded
3716 * (dirty pages + IOKit mappings)
3717 *
3718 * This is invoked for both advisory, non-fatal per-task high watermarks,
3719 * as well as the fatal task memory limits.
3720 */
3721 void
3722 memorystatus_on_ledger_footprint_exceeded(boolean_t warning, boolean_t is_fatal)
3723 {
3724 os_reason_t jetsam_reason = OS_REASON_NULL;
3725
3726 proc_t p = current_proc();
3727
3728 #if VM_PRESSURE_EVENTS
3729 if (warning == TRUE) {
3730 /*
3731 * This is a warning path which implies that the current process is close, but has
3732 * not yet exceeded its per-process memory limit.
3733 */
3734 if (memorystatus_warn_process(p->p_pid, FALSE /* not exceeded */) != TRUE) {
3735 /* Print warning, since it's possible that task has not registered for pressure notifications */
3736 printf("task_exceeded_footprint: failed to warn the current task (%d exiting, or no handler registered?).\n", p->p_pid);
3737 }
3738 return;
3739 }
3740 #endif /* VM_PRESSURE_EVENTS */
3741
3742 if (is_fatal) {
3743 /*
3744 * If this process has no high watermark or has a fatal task limit, then we have been invoked because the task
3745 * has violated either the system-wide per-task memory limit OR its own task limit.
3746 */
3747 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_PERPROCESSLIMIT);
3748 if (jetsam_reason == NULL) {
3749 printf("task_exceeded footprint: failed to allocate jetsam reason\n");
3750 } else if (corpse_for_fatal_memkill != 0) {
3751 /* Set OS_REASON_FLAG_GENERATE_CRASH_REPORT to generate corpse */
3752 jetsam_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
3753 }
3754
3755 if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit, jetsam_reason) != TRUE) {
3756 printf("task_exceeded_footprint: failed to kill the current task (exiting?).\n");
3757 }
3758 } else {
3759 /*
3760 * HWM offender exists. Done without locks or synchronization.
3761 * See comment near its declaration for more details.
3762 */
3763 memorystatus_hwm_candidates = TRUE;
3764
3765 #if VM_PRESSURE_EVENTS
3766 /*
3767 * The current process is not in the warning path.
3768 * This path implies the current process has exceeded a non-fatal (soft) memory limit.
3769 * Failure to send note is ignored here.
3770 */
3771 (void)memorystatus_warn_process(p->p_pid, TRUE /* exceeded */);
3772
3773 #endif /* VM_PRESSURE_EVENTS */
3774 }
3775 }
3776
3777 /*
3778 * Description:
3779 * Evaluates active vs. inactive process state.
3780 * Processes that opt into dirty tracking are evaluated
3781 * based on clean vs dirty state.
3782 * dirty ==> active
3783 * clean ==> inactive
3784 *
3785 * Process that do not opt into dirty tracking are
3786 * evalulated based on priority level.
3787 * Foreground or above ==> active
3788 * Below Foreground ==> inactive
3789 *
3790 * Return: TRUE if active
3791 * False if inactive
3792 */
3793
3794 static boolean_t
3795 proc_jetsam_state_is_active_locked(proc_t p) {
3796
3797 if (p->p_memstat_dirty & P_DIRTY_TRACK) {
3798 /*
3799 * process has opted into dirty tracking
3800 * active state is based on dirty vs. clean
3801 */
3802 if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
3803 /*
3804 * process is dirty
3805 * implies active state
3806 */
3807 return TRUE;
3808 } else {
3809 /*
3810 * process is clean
3811 * implies inactive state
3812 */
3813 return FALSE;
3814 }
3815 } else if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
3816 /*
3817 * process is Foreground or higher
3818 * implies active state
3819 */
3820 return TRUE;
3821 } else {
3822 /*
3823 * process found below Foreground
3824 * implies inactive state
3825 */
3826 return FALSE;
3827 }
3828 }
3829
3830 static boolean_t
3831 memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason) {
3832 boolean_t res;
3833
3834 #if CONFIG_JETSAM
3835 uint32_t errors = 0;
3836
3837 if (victim_pid == -1) {
3838 /* No pid, so kill first process */
3839 res = memorystatus_kill_top_process(TRUE, TRUE, cause, jetsam_reason, NULL, &errors);
3840 } else {
3841 res = memorystatus_kill_specific_process(victim_pid, cause, jetsam_reason);
3842 }
3843
3844 if (errors) {
3845 memorystatus_clear_errors();
3846 }
3847
3848 if (res == TRUE) {
3849 /* Fire off snapshot notification */
3850 proc_list_lock();
3851 size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
3852 sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
3853 uint64_t timestamp_now = mach_absolute_time();
3854 memorystatus_jetsam_snapshot->notification_time = timestamp_now;
3855 if (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
3856 timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout) {
3857 proc_list_unlock();
3858 int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
3859 if (!ret) {
3860 proc_list_lock();
3861 memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
3862 proc_list_unlock();
3863 }
3864 } else {
3865 proc_list_unlock();
3866 }
3867 }
3868 #else /* !CONFIG_JETSAM */
3869
3870 res = memorystatus_kill_specific_process(victim_pid, cause, jetsam_reason);
3871
3872 #endif /* CONFIG_JETSAM */
3873
3874 return res;
3875 }
3876
3877 /*
3878 * Jetsam a specific process.
3879 */
3880 static boolean_t
3881 memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason) {
3882 boolean_t killed;
3883 proc_t p;
3884 uint64_t killtime = 0;
3885 clock_sec_t tv_sec;
3886 clock_usec_t tv_usec;
3887 uint32_t tv_msec;
3888
3889 /* TODO - add a victim queue and push this into the main jetsam thread */
3890
3891 p = proc_find(victim_pid);
3892 if (!p) {
3893 os_reason_free(jetsam_reason);
3894 return FALSE;
3895 }
3896
3897 proc_list_lock();
3898
3899 #if CONFIG_JETSAM
3900 if (memorystatus_jetsam_snapshot_count == 0) {
3901 memorystatus_init_jetsam_snapshot_locked(NULL,0);
3902 }
3903
3904 killtime = mach_absolute_time();
3905 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
3906 tv_msec = tv_usec / 1000;
3907
3908 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
3909
3910 proc_list_unlock();
3911
3912 printf("%lu.%02d memorystatus: specifically killing pid %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
3913 (unsigned long)tv_sec, tv_msec, victim_pid, (*p->p_name ? p->p_name : "(unknown)"),
3914 jetsam_kill_cause_name[cause], p->p_memstat_effectivepriority, memorystatus_available_pages);
3915 #else /* !CONFIG_JETSAM */
3916 proc_list_unlock();
3917
3918 killtime = mach_absolute_time();
3919 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
3920 tv_msec = tv_usec / 1000;
3921 printf("%lu.%02d memorystatus: specifically killing pid %d [%s] (%s %d)\n",
3922 (unsigned long)tv_sec, tv_msec, victim_pid, (*p->p_name ? p->p_name : "(unknown)"),
3923 jetsam_kill_cause_name[cause], p->p_memstat_effectivepriority);
3924 #endif /* CONFIG_JETSAM */
3925
3926 killed = memorystatus_do_kill(p, cause, jetsam_reason);
3927 proc_rele(p);
3928
3929 return killed;
3930 }
3931
3932
3933 /*
3934 * Toggle the P_MEMSTAT_TERMINATED state.
3935 * Takes the proc_list_lock.
3936 */
3937 void
3938 proc_memstat_terminated(proc_t p, boolean_t set)
3939 {
3940 #if DEVELOPMENT || DEBUG
3941 if (p) {
3942 proc_list_lock();
3943 if (set == TRUE) {
3944 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
3945 } else {
3946 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
3947 }
3948 proc_list_unlock();
3949 }
3950 #else
3951 #pragma unused(p, set)
3952 /*
3953 * do nothing
3954 */
3955 #endif /* DEVELOPMENT || DEBUG */
3956 return;
3957 }
3958
3959
3960 #if CONFIG_JETSAM
3961 /*
3962 * This is invoked when cpulimits have been exceeded while in fatal mode.
3963 * The jetsam_flags do not apply as those are for memory related kills.
3964 * We call this routine so that the offending process is killed with
3965 * a non-zero exit status.
3966 */
3967 void
3968 jetsam_on_ledger_cpulimit_exceeded(void)
3969 {
3970 int retval = 0;
3971 int jetsam_flags = 0; /* make it obvious */
3972 proc_t p = current_proc();
3973 os_reason_t jetsam_reason = OS_REASON_NULL;
3974
3975 printf("task_exceeded_cpulimit: killing pid %d [%s]\n",
3976 p->p_pid, (*p->p_name ? p->p_name : "(unknown)"));
3977
3978 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_CPULIMIT);
3979 if (jetsam_reason == OS_REASON_NULL) {
3980 printf("task_exceeded_cpulimit: unable to allocate memory for jetsam reason\n");
3981 }
3982
3983 retval = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
3984
3985 if (retval) {
3986 printf("task_exceeded_cpulimit: failed to kill current task (exiting?).\n");
3987 }
3988 }
3989
3990 static void
3991 memorystatus_get_task_memory_region_count(task_t task, uint64_t *count)
3992 {
3993 assert(task);
3994 assert(count);
3995
3996 *count = get_task_memory_region_count(task);
3997 }
3998
3999 static void
4000 memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages)
4001 {
4002 assert(task);
4003 assert(footprint);
4004
4005 uint64_t pages;
4006
4007 pages = (get_task_phys_footprint(task) / PAGE_SIZE_64);
4008 assert(((uint32_t)pages) == pages);
4009 *footprint = (uint32_t)pages;
4010
4011 if (max_footprint) {
4012 pages = (get_task_phys_footprint_max(task) / PAGE_SIZE_64);
4013 assert(((uint32_t)pages) == pages);
4014 *max_footprint = (uint32_t)pages;
4015 }
4016 if (max_footprint_lifetime) {
4017 pages = (get_task_resident_max(task) / PAGE_SIZE_64);
4018 assert(((uint32_t)pages) == pages);
4019 *max_footprint_lifetime = (uint32_t)pages;
4020 }
4021 if (purgeable_pages) {
4022 pages = (get_task_purgeable_size(task) / PAGE_SIZE_64);
4023 assert(((uint32_t)pages) == pages);
4024 *purgeable_pages = (uint32_t)pages;
4025 }
4026 }
4027
4028 static void
4029 memorystatus_get_task_phys_footprint_page_counts(task_t task,
4030 uint64_t *internal_pages, uint64_t *internal_compressed_pages,
4031 uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
4032 uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
4033 uint64_t *iokit_mapped_pages, uint64_t *page_table_pages)
4034 {
4035 assert(task);
4036
4037 if (internal_pages) {
4038 *internal_pages = (get_task_internal(task) / PAGE_SIZE_64);
4039 }
4040
4041 if (internal_compressed_pages) {
4042 *internal_compressed_pages = (get_task_internal_compressed(task) / PAGE_SIZE_64);
4043 }
4044
4045 if (purgeable_nonvolatile_pages) {
4046 *purgeable_nonvolatile_pages = (get_task_purgeable_nonvolatile(task) / PAGE_SIZE_64);
4047 }
4048
4049 if (purgeable_nonvolatile_compressed_pages) {
4050 *purgeable_nonvolatile_compressed_pages = (get_task_purgeable_nonvolatile_compressed(task) / PAGE_SIZE_64);
4051 }
4052
4053 if (alternate_accounting_pages) {
4054 *alternate_accounting_pages = (get_task_alternate_accounting(task) / PAGE_SIZE_64);
4055 }
4056
4057 if (alternate_accounting_compressed_pages) {
4058 *alternate_accounting_compressed_pages = (get_task_alternate_accounting_compressed(task) / PAGE_SIZE_64);
4059 }
4060
4061 if (iokit_mapped_pages) {
4062 *iokit_mapped_pages = (get_task_iokit_mapped(task) / PAGE_SIZE_64);
4063 }
4064
4065 if (page_table_pages) {
4066 *page_table_pages = (get_task_page_table(task) / PAGE_SIZE_64);
4067 }
4068 }
4069
4070 /*
4071 * This routine only acts on the global jetsam event snapshot.
4072 * Updating the process's entry can race when the memorystatus_thread
4073 * has chosen to kill a process that is racing to exit on another core.
4074 */
4075 static void
4076 memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime)
4077 {
4078 memorystatus_jetsam_snapshot_entry_t *entry = NULL;
4079 memorystatus_jetsam_snapshot_t *snapshot = NULL;
4080 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
4081
4082 unsigned int i;
4083
4084 if (memorystatus_jetsam_snapshot_count == 0) {
4085 /*
4086 * No active snapshot.
4087 * Nothing to do.
4088 */
4089 return;
4090 }
4091
4092 /*
4093 * Sanity check as this routine should only be called
4094 * from a jetsam kill path.
4095 */
4096 assert(kill_cause != 0 && killtime != 0);
4097
4098 snapshot = memorystatus_jetsam_snapshot;
4099 snapshot_list = memorystatus_jetsam_snapshot->entries;
4100
4101 for (i = 0; i < memorystatus_jetsam_snapshot_count; i++) {
4102 if (snapshot_list[i].pid == p->p_pid) {
4103
4104 entry = &snapshot_list[i];
4105
4106 if (entry->killed || entry->jse_killtime) {
4107 /*
4108 * We apparently raced on the exit path
4109 * for this process, as it's snapshot entry
4110 * has already recorded a kill.
4111 */
4112 assert(entry->killed && entry->jse_killtime);
4113 break;
4114 }
4115
4116 /*
4117 * Update the entry we just found in the snapshot.
4118 */
4119
4120 entry->killed = kill_cause;
4121 entry->jse_killtime = killtime;
4122 entry->jse_gencount = snapshot->js_gencount;
4123 entry->jse_idle_delta = p->p_memstat_idle_delta;
4124
4125 /*
4126 * If a process has moved between bands since snapshot was
4127 * initialized, then likely these fields changed too.
4128 */
4129 if (entry->priority != p->p_memstat_effectivepriority) {
4130
4131 strlcpy(entry->name, p->p_name, sizeof(entry->name));
4132 entry->priority = p->p_memstat_effectivepriority;
4133 entry->state = memorystatus_build_state(p);
4134 entry->user_data = p->p_memstat_userdata;
4135 entry->fds = p->p_fd->fd_nfiles;
4136 }
4137
4138 /*
4139 * Always update the page counts on a kill.
4140 */
4141
4142 uint32_t pages = 0;
4143 uint32_t max_pages = 0;
4144 uint32_t max_pages_lifetime = 0;
4145 uint32_t purgeable_pages = 0;
4146
4147 memorystatus_get_task_page_counts(p->task, &pages, &max_pages, &max_pages_lifetime, &purgeable_pages);
4148 entry->pages = (uint64_t)pages;
4149 entry->max_pages = (uint64_t)max_pages;
4150 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
4151 entry->purgeable_pages = (uint64_t)purgeable_pages;
4152
4153 uint64_t internal_pages = 0;
4154 uint64_t internal_compressed_pages = 0;
4155 uint64_t purgeable_nonvolatile_pages = 0;
4156 uint64_t purgeable_nonvolatile_compressed_pages = 0;
4157 uint64_t alternate_accounting_pages = 0;
4158 uint64_t alternate_accounting_compressed_pages = 0;
4159 uint64_t iokit_mapped_pages = 0;
4160 uint64_t page_table_pages = 0;
4161
4162 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
4163 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
4164 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
4165 &iokit_mapped_pages, &page_table_pages);
4166
4167 entry->jse_internal_pages = internal_pages;
4168 entry->jse_internal_compressed_pages = internal_compressed_pages;
4169 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
4170 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
4171 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
4172 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
4173 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
4174 entry->jse_page_table_pages = page_table_pages;
4175
4176 uint64_t region_count = 0;
4177 memorystatus_get_task_memory_region_count(p->task, &region_count);
4178 entry->jse_memory_region_count = region_count;
4179
4180 goto exit;
4181 }
4182 }
4183
4184 if (entry == NULL) {
4185 /*
4186 * The entry was not found in the snapshot, so the process must have
4187 * launched after the snapshot was initialized.
4188 * Let's try to append the new entry.
4189 */
4190 if (memorystatus_jetsam_snapshot_count < memorystatus_jetsam_snapshot_max) {
4191 /*
4192 * A populated snapshot buffer exists
4193 * and there is room to init a new entry.
4194 */
4195 assert(memorystatus_jetsam_snapshot_count == snapshot->entry_count);
4196
4197 unsigned int next = memorystatus_jetsam_snapshot_count;
4198
4199 if(memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[next], (snapshot->js_gencount)) == TRUE) {
4200
4201 entry = &snapshot_list[next];
4202 entry->killed = kill_cause;
4203 entry->jse_killtime = killtime;
4204
4205 snapshot->entry_count = ++next;
4206 memorystatus_jetsam_snapshot_count = next;
4207
4208 if (memorystatus_jetsam_snapshot_count >= memorystatus_jetsam_snapshot_max) {
4209 /*
4210 * We just used the last slot in the snapshot buffer.
4211 * We only want to log it once... so we do it here
4212 * when we notice we've hit the max.
4213 */
4214 printf("memorystatus: WARNING snapshot buffer is full, count %d\n",
4215 memorystatus_jetsam_snapshot_count);
4216 }
4217 }
4218 }
4219 }
4220
4221 exit:
4222 if (entry == NULL) {
4223 /*
4224 * If we reach here, the snapshot buffer could not be updated.
4225 * Most likely, the buffer is full, in which case we would have
4226 * logged a warning in the previous call.
4227 *
4228 * For now, we will stop appending snapshot entries.
4229 * When the buffer is consumed, the snapshot state will reset.
4230 */
4231
4232 MEMORYSTATUS_DEBUG(4, "memorystatus_update_jetsam_snapshot_entry_locked: failed to update pid %d, priority %d, count %d\n",
4233 p->p_pid, p->p_memstat_effectivepriority, memorystatus_jetsam_snapshot_count);
4234 }
4235
4236 return;
4237 }
4238
4239 void memorystatus_pages_update(unsigned int pages_avail)
4240 {
4241 memorystatus_available_pages = pages_avail;
4242
4243 #if VM_PRESSURE_EVENTS
4244 /*
4245 * Since memorystatus_available_pages changes, we should
4246 * re-evaluate the pressure levels on the system and
4247 * check if we need to wake the pressure thread.
4248 * We also update memorystatus_level in that routine.
4249 */
4250 vm_pressure_response();
4251
4252 if (memorystatus_available_pages <= memorystatus_available_pages_pressure) {
4253
4254 if (memorystatus_hwm_candidates || (memorystatus_available_pages <= memorystatus_available_pages_critical)) {
4255 memorystatus_thread_wake();
4256 }
4257 }
4258 #else /* VM_PRESSURE_EVENTS */
4259
4260 boolean_t critical, delta;
4261
4262 if (!memorystatus_delta) {
4263 return;
4264 }
4265
4266 critical = (pages_avail < memorystatus_available_pages_critical) ? TRUE : FALSE;
4267 delta = ((pages_avail >= (memorystatus_available_pages + memorystatus_delta))
4268 || (memorystatus_available_pages >= (pages_avail + memorystatus_delta))) ? TRUE : FALSE;
4269
4270 if (critical || delta) {
4271 unsigned int total_pages;
4272
4273 total_pages = (unsigned int) atop_64(max_mem);
4274 #if CONFIG_SECLUDED_MEMORY
4275 total_pages -= vm_page_secluded_count;
4276 #endif /* CONFIG_SECLUDED_MEMORY */
4277 memorystatus_level = memorystatus_available_pages * 100 / total_pages;
4278 memorystatus_thread_wake();
4279 }
4280 #endif /* VM_PRESSURE_EVENTS */
4281 }
4282
4283 static boolean_t
4284 memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount)
4285 {
4286 clock_sec_t tv_sec;
4287 clock_usec_t tv_usec;
4288 uint32_t pages = 0;
4289 uint32_t max_pages = 0;
4290 uint32_t max_pages_lifetime = 0;
4291 uint32_t purgeable_pages = 0;
4292 uint64_t internal_pages = 0;
4293 uint64_t internal_compressed_pages = 0;
4294 uint64_t purgeable_nonvolatile_pages = 0;
4295 uint64_t purgeable_nonvolatile_compressed_pages = 0;
4296 uint64_t alternate_accounting_pages = 0;
4297 uint64_t alternate_accounting_compressed_pages = 0;
4298 uint64_t iokit_mapped_pages = 0;
4299 uint64_t page_table_pages =0;
4300 uint64_t region_count = 0;
4301 uint64_t cids[COALITION_NUM_TYPES];
4302
4303 memset(entry, 0, sizeof(memorystatus_jetsam_snapshot_entry_t));
4304
4305 entry->pid = p->p_pid;
4306 strlcpy(&entry->name[0], p->p_name, sizeof(entry->name));
4307 entry->priority = p->p_memstat_effectivepriority;
4308
4309 memorystatus_get_task_page_counts(p->task, &pages, &max_pages, &max_pages_lifetime, &purgeable_pages);
4310 entry->pages = (uint64_t)pages;
4311 entry->max_pages = (uint64_t)max_pages;
4312 entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
4313 entry->purgeable_pages = (uint64_t)purgeable_pages;
4314
4315 memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
4316 &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
4317 &alternate_accounting_pages, &alternate_accounting_compressed_pages,
4318 &iokit_mapped_pages, &page_table_pages);
4319
4320 entry->jse_internal_pages = internal_pages;
4321 entry->jse_internal_compressed_pages = internal_compressed_pages;
4322 entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
4323 entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
4324 entry->jse_alternate_accounting_pages = alternate_accounting_pages;
4325 entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
4326 entry->jse_iokit_mapped_pages = iokit_mapped_pages;
4327 entry->jse_page_table_pages = page_table_pages;
4328
4329 memorystatus_get_task_memory_region_count(p->task, &region_count);
4330 entry->jse_memory_region_count = region_count;
4331
4332 entry->state = memorystatus_build_state(p);
4333 entry->user_data = p->p_memstat_userdata;
4334 memcpy(&entry->uuid[0], &p->p_uuid[0], sizeof(p->p_uuid));
4335 entry->fds = p->p_fd->fd_nfiles;
4336
4337 absolutetime_to_microtime(get_task_cpu_time(p->task), &tv_sec, &tv_usec);
4338 entry->cpu_time.tv_sec = tv_sec;
4339 entry->cpu_time.tv_usec = tv_usec;
4340
4341 assert(p->p_stats != NULL);
4342 entry->jse_starttime = p->p_stats->ps_start; /* abstime process started */
4343 entry->jse_killtime = 0; /* abstime jetsam chose to kill process */
4344 entry->killed = 0; /* the jetsam kill cause */
4345 entry->jse_gencount = gencount; /* indicates a pass through jetsam thread, when process was targeted to be killed */
4346
4347 entry->jse_idle_delta = p->p_memstat_idle_delta; /* Most recent timespan spent in idle-band */
4348
4349 proc_coalitionids(p, cids);
4350 entry->jse_coalition_jetsam_id = cids[COALITION_TYPE_JETSAM];
4351
4352 return TRUE;
4353 }
4354
4355 static void
4356 memorystatus_init_snapshot_vmstats(memorystatus_jetsam_snapshot_t *snapshot)
4357 {
4358 kern_return_t kr = KERN_SUCCESS;
4359 mach_msg_type_number_t count = HOST_VM_INFO64_COUNT;
4360 vm_statistics64_data_t vm_stat;
4361
4362 if ((kr = host_statistics64(host_self(), HOST_VM_INFO64, (host_info64_t)&vm_stat, &count) != KERN_SUCCESS)) {
4363 printf("memorystatus_init_jetsam_snapshot_stats: host_statistics64 failed with %d\n", kr);
4364 memset(&snapshot->stats, 0, sizeof(snapshot->stats));
4365 } else {
4366 snapshot->stats.free_pages = vm_stat.free_count;
4367 snapshot->stats.active_pages = vm_stat.active_count;
4368 snapshot->stats.inactive_pages = vm_stat.inactive_count;
4369 snapshot->stats.throttled_pages = vm_stat.throttled_count;
4370 snapshot->stats.purgeable_pages = vm_stat.purgeable_count;
4371 snapshot->stats.wired_pages = vm_stat.wire_count;
4372
4373 snapshot->stats.speculative_pages = vm_stat.speculative_count;
4374 snapshot->stats.filebacked_pages = vm_stat.external_page_count;
4375 snapshot->stats.anonymous_pages = vm_stat.internal_page_count;
4376 snapshot->stats.compressions = vm_stat.compressions;
4377 snapshot->stats.decompressions = vm_stat.decompressions;
4378 snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
4379 snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
4380 }
4381 }
4382
4383 /*
4384 * Collect vm statistics at boot.
4385 * Called only once (see kern_exec.c)
4386 * Data can be consumed at any time.
4387 */
4388 void
4389 memorystatus_init_at_boot_snapshot() {
4390 memorystatus_init_snapshot_vmstats(&memorystatus_at_boot_snapshot);
4391 memorystatus_at_boot_snapshot.entry_count = 0;
4392 memorystatus_at_boot_snapshot.notification_time = 0; /* updated when consumed */
4393 memorystatus_at_boot_snapshot.snapshot_time = mach_absolute_time();
4394 }
4395
4396 static void
4397 memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count )
4398 {
4399 proc_t p, next_p;
4400 unsigned int b = 0, i = 0;
4401
4402 memorystatus_jetsam_snapshot_t *snapshot = NULL;
4403 memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
4404 unsigned int snapshot_max = 0;
4405
4406 if (od_snapshot) {
4407 /*
4408 * This is an on_demand snapshot
4409 */
4410 snapshot = od_snapshot;
4411 snapshot_list = od_snapshot->entries;
4412 snapshot_max = ods_list_count;
4413 } else {
4414 /*
4415 * This is a jetsam event snapshot
4416 */
4417 snapshot = memorystatus_jetsam_snapshot;
4418 snapshot_list = memorystatus_jetsam_snapshot->entries;
4419 snapshot_max = memorystatus_jetsam_snapshot_max;
4420 }
4421
4422 /*
4423 * Init the snapshot header information
4424 */
4425 memorystatus_init_snapshot_vmstats(snapshot);
4426 snapshot->snapshot_time = mach_absolute_time();
4427 snapshot->notification_time = 0;
4428 snapshot->js_gencount = 0;
4429
4430 next_p = memorystatus_get_first_proc_locked(&b, TRUE);
4431 while (next_p) {
4432 p = next_p;
4433 next_p = memorystatus_get_next_proc_locked(&b, p, TRUE);
4434
4435 if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i], snapshot->js_gencount)) {
4436 continue;
4437 }
4438
4439 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",
4440 p->p_pid,
4441 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],
4442 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]);
4443
4444 if (++i == snapshot_max) {
4445 break;
4446 }
4447 }
4448
4449 snapshot->entry_count = i;
4450
4451 if (!od_snapshot) {
4452 /* update the system buffer count */
4453 memorystatus_jetsam_snapshot_count = i;
4454 }
4455 }
4456
4457 #if DEVELOPMENT || DEBUG
4458
4459 static int
4460 memorystatus_cmd_set_panic_bits(user_addr_t buffer, uint32_t buffer_size) {
4461 int ret;
4462 memorystatus_jetsam_panic_options_t debug;
4463
4464 if (buffer_size != sizeof(memorystatus_jetsam_panic_options_t)) {
4465 return EINVAL;
4466 }
4467
4468 ret = copyin(buffer, &debug, buffer_size);
4469 if (ret) {
4470 return ret;
4471 }
4472
4473 /* Panic bits match kMemorystatusKilled* enum */
4474 memorystatus_jetsam_panic_debug = (memorystatus_jetsam_panic_debug & ~debug.mask) | (debug.data & debug.mask);
4475
4476 /* Copyout new value */
4477 debug.data = memorystatus_jetsam_panic_debug;
4478 ret = copyout(&debug, buffer, sizeof(memorystatus_jetsam_panic_options_t));
4479
4480 return ret;
4481 }
4482
4483 /*
4484 * Triggers a sort_order on a specified jetsam priority band.
4485 * This is for testing only, used to force a path through the sort
4486 * function.
4487 */
4488 static int
4489 memorystatus_cmd_test_jetsam_sort(int priority, int sort_order) {
4490
4491 int error = 0;
4492
4493 unsigned int bucket_index = 0;
4494
4495 if (priority == -1) {
4496 /* Use as shorthand for default priority */
4497 bucket_index = JETSAM_PRIORITY_DEFAULT;
4498 } else {
4499 bucket_index = (unsigned int)priority;
4500 }
4501
4502 error = memorystatus_sort_bucket(bucket_index, sort_order);
4503
4504 return (error);
4505 }
4506
4507 #endif /* DEVELOPMENT || DEBUG */
4508
4509 /*
4510 * Jetsam the first process in the queue.
4511 */
4512 static boolean_t
4513 memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, os_reason_t jetsam_reason,
4514 int32_t *priority, uint32_t *errors)
4515 {
4516 pid_t aPid;
4517 proc_t p = PROC_NULL, next_p = PROC_NULL;
4518 boolean_t new_snapshot = FALSE, killed = FALSE;
4519 int kill_count = 0;
4520 unsigned int i = 0;
4521 uint32_t aPid_ep;
4522 uint64_t killtime = 0;
4523 clock_sec_t tv_sec;
4524 clock_usec_t tv_usec;
4525 uint32_t tv_msec;
4526
4527 #ifndef CONFIG_FREEZE
4528 #pragma unused(any)
4529 #endif
4530
4531 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
4532 memorystatus_available_pages, 0, 0, 0, 0);
4533
4534
4535 if (sort_flag == TRUE) {
4536 (void)memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
4537 }
4538
4539 proc_list_lock();
4540
4541 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
4542 while (next_p) {
4543 #if DEVELOPMENT || DEBUG
4544 int activeProcess;
4545 int procSuspendedForDiagnosis;
4546 #endif /* DEVELOPMENT || DEBUG */
4547
4548 p = next_p;
4549 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
4550
4551 #if DEVELOPMENT || DEBUG
4552 activeProcess = p->p_memstat_state & P_MEMSTAT_FOREGROUND;
4553 procSuspendedForDiagnosis = p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED;
4554 #endif /* DEVELOPMENT || DEBUG */
4555
4556 aPid = p->p_pid;
4557 aPid_ep = p->p_memstat_effectivepriority;
4558
4559 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
4560 continue; /* with lock held */
4561 }
4562
4563 #if DEVELOPMENT || DEBUG
4564 if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && procSuspendedForDiagnosis) {
4565 printf("jetsam: continuing after ignoring proc suspended already for diagnosis - %d\n", aPid);
4566 continue;
4567 }
4568 #endif /* DEVELOPMENT || DEBUG */
4569
4570 if (cause == kMemorystatusKilledVnodes)
4571 {
4572 /*
4573 * If the system runs out of vnodes, we systematically jetsam
4574 * processes in hopes of stumbling onto a vnode gain that helps
4575 * the system recover. The process that happens to trigger
4576 * this path has no known relationship to the vnode consumption.
4577 * We attempt to safeguard that process e.g: do not jetsam it.
4578 */
4579
4580 if (p == current_proc()) {
4581 /* do not jetsam the current process */
4582 continue;
4583 }
4584 }
4585
4586 #if CONFIG_FREEZE
4587 boolean_t skip;
4588 boolean_t reclaim_proc = !(p->p_memstat_state & (P_MEMSTAT_LOCKED | P_MEMSTAT_NORECLAIM));
4589 if (any || reclaim_proc) {
4590 skip = FALSE;
4591 } else {
4592 skip = TRUE;
4593 }
4594
4595 if (skip) {
4596 continue;
4597 } else
4598 #endif
4599 {
4600 /*
4601 * Capture a snapshot if none exists and:
4602 * - priority was not requested (this is something other than an ambient kill)
4603 * - the priority was requested *and* the targeted process is not at idle priority
4604 */
4605 if ((memorystatus_jetsam_snapshot_count == 0) &&
4606 (memorystatus_idle_snapshot || ((!priority) || (priority && (aPid_ep != JETSAM_PRIORITY_IDLE))))) {
4607 memorystatus_init_jetsam_snapshot_locked(NULL,0);
4608 new_snapshot = TRUE;
4609 }
4610
4611 /*
4612 * Mark as terminated so that if exit1() indicates success, but the process (for example)
4613 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
4614 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
4615 * acquisition of the proc lock.
4616 */
4617 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
4618
4619 killtime = mach_absolute_time();
4620 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
4621 tv_msec = tv_usec / 1000;
4622
4623 #if DEVELOPMENT || DEBUG
4624 if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && activeProcess) {
4625 MEMORYSTATUS_DEBUG(1, "jetsam: suspending pid %d [%s] (active) for diagnosis - memory_status_level: %d\n",
4626 aPid, (*p->p_name ? p->p_name: "(unknown)"), memorystatus_level);
4627 memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic, killtime);
4628 p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
4629 if (memorystatus_jetsam_policy & kPolicyDiagnoseFirst) {
4630 jetsam_diagnostic_suspended_one_active_proc = 1;
4631 printf("jetsam: returning after suspending first active proc - %d\n", aPid);
4632 }
4633
4634 p = proc_ref_locked(p);
4635 proc_list_unlock();
4636 if (p) {
4637 task_suspend(p->task);
4638 if (priority) {
4639 *priority = aPid_ep;
4640 }
4641 proc_rele(p);
4642 killed = TRUE;
4643 }
4644
4645 goto exit;
4646 } else
4647 #endif /* DEVELOPMENT || DEBUG */
4648 {
4649 /* Shift queue, update stats */
4650 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
4651
4652 if (proc_ref_locked(p) == p) {
4653 proc_list_unlock();
4654 printf("%lu.%02d memorystatus: %s %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
4655 (unsigned long)tv_sec, tv_msec,
4656 ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "idle exiting pid" : "jetsam killing top process pid"),
4657 aPid, (*p->p_name ? p->p_name : "(unknown)"),
4658 jetsam_kill_cause_name[cause], aPid_ep, memorystatus_available_pages);
4659
4660 /*
4661 * memorystatus_do_kill() drops a reference, so take another one so we can
4662 * continue to use this exit reason even after memorystatus_do_kill()
4663 * returns.
4664 */
4665 os_reason_ref(jetsam_reason);
4666
4667 killed = memorystatus_do_kill(p, cause, jetsam_reason);
4668
4669 /* Success? */
4670 if (killed) {
4671 if (priority) {
4672 *priority = aPid_ep;
4673 }
4674 proc_rele(p);
4675 kill_count++;
4676 goto exit;
4677 }
4678
4679 /*
4680 * Failure - first unwind the state,
4681 * then fall through to restart the search.
4682 */
4683 proc_list_lock();
4684 proc_rele_locked(p);
4685 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
4686 p->p_memstat_state |= P_MEMSTAT_ERROR;
4687 *errors += 1;
4688 }
4689
4690 /*
4691 * Failure - restart the search.
4692 *
4693 * We might have raced with "p" exiting on another core, resulting in no
4694 * ref on "p". Or, we may have failed to kill "p".
4695 *
4696 * Either way, we fall thru to here, leaving the proc in the
4697 * P_MEMSTAT_TERMINATED state.
4698 *
4699 * And, we hold the the proc_list_lock at this point.
4700 */
4701
4702 i = 0;
4703 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
4704 }
4705 }
4706 }
4707
4708 proc_list_unlock();
4709
4710 exit:
4711 os_reason_free(jetsam_reason);
4712
4713 /* Clear snapshot if freshly captured and no target was found */
4714 if (new_snapshot && !killed) {
4715 proc_list_lock();
4716 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
4717 proc_list_unlock();
4718 }
4719
4720 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
4721 memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
4722
4723 return killed;
4724 }
4725
4726 /*
4727 * Jetsam aggressively
4728 */
4729 static boolean_t
4730 memorystatus_kill_top_process_aggressive(boolean_t any, uint32_t cause, os_reason_t jetsam_reason, int aggr_count,
4731 int32_t priority_max, uint32_t *errors)
4732 {
4733 pid_t aPid;
4734 proc_t p = PROC_NULL, next_p = PROC_NULL;
4735 boolean_t new_snapshot = FALSE, killed = FALSE;
4736 int kill_count = 0;
4737 unsigned int i = 0;
4738 int32_t aPid_ep = 0;
4739 unsigned int memorystatus_level_snapshot = 0;
4740 uint64_t killtime = 0;
4741 clock_sec_t tv_sec;
4742 clock_usec_t tv_usec;
4743 uint32_t tv_msec;
4744
4745 #pragma unused(any)
4746
4747 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
4748 memorystatus_available_pages, priority_max, 0, 0, 0);
4749
4750 memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
4751
4752 proc_list_lock();
4753
4754 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
4755 while (next_p) {
4756 #if DEVELOPMENT || DEBUG
4757 int activeProcess;
4758 int procSuspendedForDiagnosis;
4759 #endif /* DEVELOPMENT || DEBUG */
4760
4761 if ((unsigned int)(next_p->p_memstat_effectivepriority) != i) {
4762
4763 /*
4764 * We have raced with next_p running on another core, as it has
4765 * moved to a different jetsam priority band. This means we have
4766 * lost our place in line while traversing the jetsam list. We
4767 * attempt to recover by rewinding to the beginning of the band
4768 * we were already traversing. By doing this, we do not guarantee
4769 * that no process escapes this aggressive march, but we can make
4770 * skipping an entire range of processes less likely. (PR-21069019)
4771 */
4772
4773 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: rewinding %s moved from band %d --> %d\n",
4774 aggr_count, (*next_p->p_name ? next_p->p_name : "unknown"), i, next_p->p_memstat_effectivepriority);
4775
4776 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
4777 continue;
4778 }
4779
4780 p = next_p;
4781 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
4782
4783 if (p->p_memstat_effectivepriority > priority_max) {
4784 /*
4785 * Bail out of this killing spree if we have
4786 * reached beyond the priority_max jetsam band.
4787 * That is, we kill up to and through the
4788 * priority_max jetsam band.
4789 */
4790 proc_list_unlock();
4791 goto exit;
4792 }
4793
4794 #if DEVELOPMENT || DEBUG
4795 activeProcess = p->p_memstat_state & P_MEMSTAT_FOREGROUND;
4796 procSuspendedForDiagnosis = p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED;
4797 #endif /* DEVELOPMENT || DEBUG */
4798
4799 aPid = p->p_pid;
4800 aPid_ep = p->p_memstat_effectivepriority;
4801
4802 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
4803 continue;
4804 }
4805
4806 #if DEVELOPMENT || DEBUG
4807 if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && procSuspendedForDiagnosis) {
4808 printf("jetsam: continuing after ignoring proc suspended already for diagnosis - %d\n", aPid);
4809 continue;
4810 }
4811 #endif /* DEVELOPMENT || DEBUG */
4812
4813 /*
4814 * Capture a snapshot if none exists.
4815 */
4816 if (memorystatus_jetsam_snapshot_count == 0) {
4817 memorystatus_init_jetsam_snapshot_locked(NULL,0);
4818 new_snapshot = TRUE;
4819 }
4820
4821 /*
4822 * Mark as terminated so that if exit1() indicates success, but the process (for example)
4823 * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
4824 * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
4825 * acquisition of the proc lock.
4826 */
4827 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
4828
4829 killtime = mach_absolute_time();
4830 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
4831 tv_msec = tv_usec / 1000;
4832
4833 /* Shift queue, update stats */
4834 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
4835
4836 /*
4837 * In order to kill the target process, we will drop the proc_list_lock.
4838 * To guaranteee that p and next_p don't disappear out from under the lock,
4839 * we must take a ref on both.
4840 * If we cannot get a reference, then it's likely we've raced with
4841 * that process exiting on another core.
4842 */
4843 if (proc_ref_locked(p) == p) {
4844 if (next_p) {
4845 while (next_p && (proc_ref_locked(next_p) != next_p)) {
4846 proc_t temp_p;
4847
4848 /*
4849 * We must have raced with next_p exiting on another core.
4850 * Recover by getting the next eligible process in the band.
4851 */
4852
4853 MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: skipping %d [%s] (exiting?)\n",
4854 aggr_count, next_p->p_pid, (*next_p->p_name ? next_p->p_name : "(unknown)"));
4855
4856 temp_p = next_p;
4857 next_p = memorystatus_get_next_proc_locked(&i, temp_p, TRUE);
4858 }
4859 }
4860 proc_list_unlock();
4861
4862 printf("%lu.%01d memorystatus: aggressive%d: %s %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
4863 (unsigned long)tv_sec, tv_msec, aggr_count,
4864 ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "idle exiting pid" : "jetsam killing pid"),
4865 aPid, (*p->p_name ? p->p_name : "(unknown)"),
4866 jetsam_kill_cause_name[cause], aPid_ep, memorystatus_available_pages);
4867
4868 memorystatus_level_snapshot = memorystatus_level;
4869
4870 /*
4871 * memorystatus_do_kill() drops a reference, so take another one so we can
4872 * continue to use this exit reason even after memorystatus_do_kill()
4873 * returns.
4874 */
4875 os_reason_ref(jetsam_reason);
4876 killed = memorystatus_do_kill(p, cause, jetsam_reason);
4877
4878 /* Success? */
4879 if (killed) {
4880 proc_rele(p);
4881 kill_count++;
4882 p = NULL;
4883 killed = FALSE;
4884
4885 /*
4886 * Continue the killing spree.
4887 */
4888 proc_list_lock();
4889 if (next_p) {
4890 proc_rele_locked(next_p);
4891 }
4892
4893 if (aPid_ep == JETSAM_PRIORITY_FOREGROUND && memorystatus_aggressive_jetsam_lenient == TRUE) {
4894 if (memorystatus_level > memorystatus_level_snapshot && ((memorystatus_level - memorystatus_level_snapshot) >= AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD)) {
4895 #if DEVELOPMENT || DEBUG
4896 printf("Disabling Lenient mode after one-time deployment.\n");
4897 #endif /* DEVELOPMENT || DEBUG */
4898 memorystatus_aggressive_jetsam_lenient = FALSE;
4899 break;
4900 }
4901 }
4902
4903 continue;
4904 }
4905
4906 /*
4907 * Failure - first unwind the state,
4908 * then fall through to restart the search.
4909 */
4910 proc_list_lock();
4911 proc_rele_locked(p);
4912 if (next_p) {
4913 proc_rele_locked(next_p);
4914 }
4915 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
4916 p->p_memstat_state |= P_MEMSTAT_ERROR;
4917 *errors += 1;
4918 p = NULL;
4919 }
4920
4921 /*
4922 * Failure - restart the search at the beginning of
4923 * the band we were already traversing.
4924 *
4925 * We might have raced with "p" exiting on another core, resulting in no
4926 * ref on "p". Or, we may have failed to kill "p".
4927 *
4928 * Either way, we fall thru to here, leaving the proc in the
4929 * P_MEMSTAT_TERMINATED or P_MEMSTAT_ERROR state.
4930 *
4931 * And, we hold the the proc_list_lock at this point.
4932 */
4933
4934 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
4935 }
4936
4937 proc_list_unlock();
4938
4939 exit:
4940 os_reason_free(jetsam_reason);
4941
4942 /* Clear snapshot if freshly captured and no target was found */
4943 if (new_snapshot && (kill_count == 0)) {
4944 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
4945 }
4946
4947 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
4948 memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
4949
4950 if (kill_count > 0) {
4951 return(TRUE);
4952 }
4953 else {
4954 return(FALSE);
4955 }
4956 }
4957
4958 static boolean_t
4959 memorystatus_kill_hiwat_proc(uint32_t *errors)
4960 {
4961 pid_t aPid = 0;
4962 proc_t p = PROC_NULL, next_p = PROC_NULL;
4963 boolean_t new_snapshot = FALSE, killed = FALSE;
4964 int kill_count = 0;
4965 unsigned int i = 0;
4966 uint32_t aPid_ep;
4967 uint64_t killtime = 0;
4968 clock_sec_t tv_sec;
4969 clock_usec_t tv_usec;
4970 uint32_t tv_msec;
4971 os_reason_t jetsam_reason = OS_REASON_NULL;
4972 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
4973 memorystatus_available_pages, 0, 0, 0, 0);
4974
4975 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_HIGHWATER);
4976 if (jetsam_reason == OS_REASON_NULL) {
4977 printf("memorystatus_kill_hiwat_proc: failed to allocate exit reason\n");
4978 }
4979
4980 proc_list_lock();
4981
4982 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
4983 while (next_p) {
4984 uint64_t footprint_in_bytes = 0;
4985 uint64_t memlimit_in_bytes = 0;
4986 boolean_t skip = 0;
4987
4988 p = next_p;
4989 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
4990
4991 aPid = p->p_pid;
4992 aPid_ep = p->p_memstat_effectivepriority;
4993
4994 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
4995 continue;
4996 }
4997
4998 /* skip if no limit set */
4999 if (p->p_memstat_memlimit <= 0) {
5000 continue;
5001 }
5002
5003 #if 0
5004 /*
5005 * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
5006 * Background limits are described via the inactive limit slots.
5007 * Their fatal/non-fatal setting will drive whether or not to be
5008 * considered in this kill path.
5009 */
5010
5011 /* skip if a currently inapplicable limit is encountered */
5012 if ((p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
5013 continue;
5014 }
5015 #endif
5016 footprint_in_bytes = get_task_phys_footprint(p->task);
5017 memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
5018 skip = (footprint_in_bytes <= memlimit_in_bytes);
5019
5020 #if DEVELOPMENT || DEBUG
5021 if (!skip && (memorystatus_jetsam_policy & kPolicyDiagnoseActive)) {
5022 if (p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED) {
5023 continue;
5024 }
5025 }
5026 #endif /* DEVELOPMENT || DEBUG */
5027
5028 #if CONFIG_FREEZE
5029 if (!skip) {
5030 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
5031 skip = TRUE;
5032 } else {
5033 skip = FALSE;
5034 }
5035 }
5036 #endif
5037
5038 if (skip) {
5039 continue;
5040 } else {
5041 #if DEVELOPMENT || DEBUG
5042 MEMORYSTATUS_DEBUG(1, "jetsam: %s pid %d [%s] - %lld Mb > 1 (%d Mb)\n",
5043 (memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing",
5044 aPid, (*p->p_name ? p->p_name : "unknown"),
5045 (footprint_in_bytes / (1024ULL * 1024ULL)), /* converted bytes to MB */
5046 p->p_memstat_memlimit);
5047 #endif /* DEVELOPMENT || DEBUG */
5048
5049 if (memorystatus_jetsam_snapshot_count == 0) {
5050 memorystatus_init_jetsam_snapshot_locked(NULL,0);
5051 new_snapshot = TRUE;
5052 }
5053
5054 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5055
5056 killtime = mach_absolute_time();
5057 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5058 tv_msec = tv_usec / 1000;
5059
5060 #if DEVELOPMENT || DEBUG
5061 if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
5062 MEMORYSTATUS_DEBUG(1, "jetsam: pid %d suspended for diagnosis - memorystatus_available_pages: %d\n", aPid, memorystatus_available_pages);
5063 memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic, killtime);
5064 p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
5065
5066 p = proc_ref_locked(p);
5067 proc_list_unlock();
5068 if (p) {
5069 task_suspend(p->task);
5070 proc_rele(p);
5071 killed = TRUE;
5072 }
5073
5074 goto exit;
5075 } else
5076 #endif /* DEVELOPMENT || DEBUG */
5077 {
5078 memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledHiwat, killtime);
5079
5080 if (proc_ref_locked(p) == p) {
5081 proc_list_unlock();
5082
5083 printf("%lu.%02d memorystatus: jetsam killing pid %d [%s] (highwater %d) - memorystatus_available_pages: %d\n",
5084 (unsigned long)tv_sec, tv_msec, aPid, (*p->p_name ? p->p_name : "(unknown)"), aPid_ep, memorystatus_available_pages);
5085
5086 /*
5087 * memorystatus_do_kill drops a reference, so take another one so we can
5088 * continue to use this exit reason even after memorystatus_do_kill()
5089 * returns
5090 */
5091 os_reason_ref(jetsam_reason);
5092
5093 killed = memorystatus_do_kill(p, kMemorystatusKilledHiwat, jetsam_reason);
5094
5095 /* Success? */
5096 if (killed) {
5097 proc_rele(p);
5098 kill_count++;
5099 goto exit;
5100 }
5101
5102 /*
5103 * Failure - first unwind the state,
5104 * then fall through to restart the search.
5105 */
5106 proc_list_lock();
5107 proc_rele_locked(p);
5108 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5109 p->p_memstat_state |= P_MEMSTAT_ERROR;
5110 *errors += 1;
5111 }
5112
5113 /*
5114 * Failure - restart the search.
5115 *
5116 * We might have raced with "p" exiting on another core, resulting in no
5117 * ref on "p". Or, we may have failed to kill "p".
5118 *
5119 * Either way, we fall thru to here, leaving the proc in the
5120 * P_MEMSTAT_TERMINATED state.
5121 *
5122 * And, we hold the the proc_list_lock at this point.
5123 */
5124
5125 i = 0;
5126 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5127 }
5128 }
5129 }
5130
5131 proc_list_unlock();
5132
5133 exit:
5134 os_reason_free(jetsam_reason);
5135
5136 /* Clear snapshot if freshly captured and no target was found */
5137 if (new_snapshot && !killed) {
5138 proc_list_lock();
5139 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5140 proc_list_unlock();
5141 }
5142
5143 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_END,
5144 memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
5145
5146 return killed;
5147 }
5148
5149 /*
5150 * Jetsam a process pinned in the elevated band.
5151 *
5152 * Return: true -- at least one pinned process was jetsammed
5153 * false -- no pinned process was jetsammed
5154 */
5155 static boolean_t
5156 memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, int aggr_count, uint32_t *errors)
5157 {
5158 pid_t aPid = 0;
5159 proc_t p = PROC_NULL, next_p = PROC_NULL;
5160 boolean_t new_snapshot = FALSE, killed = FALSE;
5161 int kill_count = 0;
5162 unsigned int i = JETSAM_PRIORITY_ELEVATED_INACTIVE;
5163 uint32_t aPid_ep;
5164 uint64_t killtime = 0;
5165 clock_sec_t tv_sec;
5166 clock_usec_t tv_usec;
5167 uint32_t tv_msec;
5168
5169
5170 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
5171 memorystatus_available_pages, 0, 0, 0, 0);
5172
5173 proc_list_lock();
5174
5175 next_p = memorystatus_get_first_proc_locked(&i, FALSE);
5176 while (next_p) {
5177
5178 p = next_p;
5179 next_p = memorystatus_get_next_proc_locked(&i, p, FALSE);
5180
5181 aPid = p->p_pid;
5182 aPid_ep = p->p_memstat_effectivepriority;
5183
5184 /*
5185 * Only pick a process pinned in this elevated band
5186 */
5187 if (!(p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
5188 continue;
5189 }
5190
5191 if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
5192 continue;
5193 }
5194
5195 #if CONFIG_FREEZE
5196 if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
5197 continue;
5198 }
5199 #endif
5200
5201 #if DEVELOPMENT || DEBUG
5202 MEMORYSTATUS_DEBUG(1, "jetsam: elevated%d process pid %d [%s] - memorystatus_available_pages: %d\n",
5203 aggr_count,
5204 aPid, (*p->p_name ? p->p_name : "unknown"),
5205 memorystatus_available_pages);
5206 #endif /* DEVELOPMENT || DEBUG */
5207
5208 if (memorystatus_jetsam_snapshot_count == 0) {
5209 memorystatus_init_jetsam_snapshot_locked(NULL,0);
5210 new_snapshot = TRUE;
5211 }
5212
5213 p->p_memstat_state |= P_MEMSTAT_TERMINATED;
5214
5215 killtime = mach_absolute_time();
5216 absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
5217 tv_msec = tv_usec / 1000;
5218
5219 memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
5220
5221 if (proc_ref_locked(p) == p) {
5222
5223 proc_list_unlock();
5224
5225 printf("%lu.%01d memorystatus: elevated%d: jetsam killing pid %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
5226 (unsigned long)tv_sec, tv_msec,
5227 aggr_count,
5228 aPid, (*p->p_name ? p->p_name : "(unknown)"),
5229 jetsam_kill_cause_name[cause], aPid_ep, memorystatus_available_pages);
5230
5231 /*
5232 * memorystatus_do_kill drops a reference, so take another one so we can
5233 * continue to use this exit reason even after memorystatus_do_kill()
5234 * returns
5235 */
5236 os_reason_ref(jetsam_reason);
5237 killed = memorystatus_do_kill(p, cause, jetsam_reason);
5238
5239 /* Success? */
5240 if (killed) {
5241 proc_rele(p);
5242 kill_count++;
5243 goto exit;
5244 }
5245
5246 /*
5247 * Failure - first unwind the state,
5248 * then fall through to restart the search.
5249 */
5250 proc_list_lock();
5251 proc_rele_locked(p);
5252 p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
5253 p->p_memstat_state |= P_MEMSTAT_ERROR;
5254 *errors += 1;
5255 }
5256
5257 /*
5258 * Failure - restart the search.
5259 *
5260 * We might have raced with "p" exiting on another core, resulting in no
5261 * ref on "p". Or, we may have failed to kill "p".
5262 *
5263 * Either way, we fall thru to here, leaving the proc in the
5264 * P_MEMSTAT_TERMINATED state or P_MEMSTAT_ERROR state.
5265 *
5266 * And, we hold the the proc_list_lock at this point.
5267 */
5268
5269 next_p = memorystatus_get_first_proc_locked(&i, FALSE);
5270 }
5271
5272 proc_list_unlock();
5273
5274 exit:
5275 os_reason_free(jetsam_reason);
5276
5277 /* Clear snapshot if freshly captured and no target was found */
5278 if (new_snapshot && (kill_count == 0)) {
5279 proc_list_lock();
5280 memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
5281 proc_list_unlock();
5282 }
5283
5284 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
5285 memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
5286
5287 return (killed);
5288 }
5289
5290 static boolean_t
5291 memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause) {
5292 /*
5293 * TODO: allow a general async path
5294 *
5295 * NOTE: If a new async kill cause is added, make sure to update memorystatus_thread() to
5296 * add the appropriate exit reason code mapping.
5297 */
5298 if ((victim_pid != -1) || (cause != kMemorystatusKilledVMPageShortage && cause != kMemorystatusKilledVMThrashing &&
5299 cause != kMemorystatusKilledFCThrashing)) {
5300 return FALSE;
5301 }
5302
5303 kill_under_pressure_cause = cause;
5304 memorystatus_thread_wake();
5305 return TRUE;
5306 }
5307
5308 boolean_t
5309 memorystatus_kill_on_VM_page_shortage(boolean_t async) {
5310 if (async) {
5311 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMPageShortage);
5312 } else {
5313 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMPAGESHORTAGE);
5314 if (jetsam_reason == OS_REASON_NULL) {
5315 printf("memorystatus_kill_on_VM_page_shortage -- sync: failed to allocate jetsam reason\n");
5316 }
5317
5318 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMPageShortage, jetsam_reason);
5319 }
5320 }
5321
5322 boolean_t
5323 memorystatus_kill_on_VM_thrashing(boolean_t async) {
5324 if (async) {
5325 return memorystatus_kill_process_async(-1, kMemorystatusKilledVMThrashing);
5326 } else {
5327 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMTHRASHING);
5328 if (jetsam_reason == OS_REASON_NULL) {
5329 printf("memorystatus_kill_on_VM_thrashing -- sync: failed to allocate jetsam reason\n");
5330 }
5331
5332 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMThrashing, jetsam_reason);
5333 }
5334 }
5335
5336 boolean_t
5337 memorystatus_kill_on_FC_thrashing(boolean_t async) {
5338
5339
5340 if (async) {
5341 return memorystatus_kill_process_async(-1, kMemorystatusKilledFCThrashing);
5342 } else {
5343 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_FCTHRASHING);
5344 if (jetsam_reason == OS_REASON_NULL) {
5345 printf("memorystatus_kill_on_FC_thrashing -- sync: failed to allocate jetsam reason\n");
5346 }
5347
5348 return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing, jetsam_reason);
5349 }
5350 }
5351
5352 boolean_t
5353 memorystatus_kill_on_vnode_limit(void) {
5354 os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_VNODE);
5355 if (jetsam_reason == OS_REASON_NULL) {
5356 printf("memorystatus_kill_on_vnode_limit: failed to allocate jetsam reason\n");
5357 }
5358
5359 return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes, jetsam_reason);
5360 }
5361
5362 #endif /* CONFIG_JETSAM */
5363
5364 #if CONFIG_FREEZE
5365
5366 __private_extern__ void
5367 memorystatus_freeze_init(void)
5368 {
5369 kern_return_t result;
5370 thread_t thread;
5371
5372 freezer_lck_grp_attr = lck_grp_attr_alloc_init();
5373 freezer_lck_grp = lck_grp_alloc_init("freezer", freezer_lck_grp_attr);
5374
5375 lck_mtx_init(&freezer_mutex, freezer_lck_grp, NULL);
5376
5377 result = kernel_thread_start(memorystatus_freeze_thread, NULL, &thread);
5378 if (result == KERN_SUCCESS) {
5379 thread_deallocate(thread);
5380 } else {
5381 panic("Could not create memorystatus_freeze_thread");
5382 }
5383 }
5384
5385 /*
5386 * Synchronously freeze the passed proc. Called with a reference to the proc held.
5387 *
5388 * Returns EINVAL or the value returned by task_freeze().
5389 */
5390 int
5391 memorystatus_freeze_process_sync(proc_t p)
5392 {
5393 int ret = EINVAL;
5394 pid_t aPid = 0;
5395 boolean_t memorystatus_freeze_swap_low = FALSE;
5396
5397 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_START,
5398 memorystatus_available_pages, 0, 0, 0, 0);
5399
5400 lck_mtx_lock(&freezer_mutex);
5401
5402 if (p == NULL) {
5403 goto exit;
5404 }
5405
5406 if (memorystatus_freeze_enabled == FALSE) {
5407 goto exit;
5408 }
5409
5410 if (!memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
5411 goto exit;
5412 }
5413
5414 if (memorystatus_freeze_update_throttle()) {
5415 printf("memorystatus_freeze_process_sync: in throttle, ignorning freeze\n");
5416 memorystatus_freeze_throttle_count++;
5417 goto exit;
5418 }
5419
5420 proc_list_lock();
5421
5422 if (p != NULL) {
5423 uint32_t purgeable, wired, clean, dirty, state;
5424 uint32_t max_pages, pages, i;
5425 boolean_t shared;
5426
5427 aPid = p->p_pid;
5428 state = p->p_memstat_state;
5429
5430 /* Ensure the process is eligible for freezing */
5431 if ((state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FROZEN)) || !(state & P_MEMSTAT_SUSPENDED)) {
5432 proc_list_unlock();
5433 goto exit;
5434 }
5435
5436 /* Only freeze processes meeting our minimum resident page criteria */
5437 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
5438 if (pages < memorystatus_freeze_pages_min) {
5439 proc_list_unlock();
5440 goto exit;
5441 }
5442
5443 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5444
5445 unsigned int avail_swap_space = 0; /* in pages. */
5446
5447 /*
5448 * Freezer backed by the compressor and swap file(s)
5449 * while will hold compressed data.
5450 */
5451 avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
5452
5453 max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
5454
5455 if (max_pages < memorystatus_freeze_pages_min) {
5456 proc_list_unlock();
5457 goto exit;
5458 }
5459 } else {
5460 /*
5461 * We only have the compressor without any swap.
5462 */
5463 max_pages = UINT32_MAX - 1;
5464 }
5465
5466 /* Mark as locked temporarily to avoid kill */
5467 p->p_memstat_state |= P_MEMSTAT_LOCKED;
5468 proc_list_unlock();
5469
5470 ret = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
5471
5472 DTRACE_MEMORYSTATUS6(memorystatus_freeze, proc_t, p, unsigned int, memorystatus_available_pages, boolean_t, purgeable, unsigned int, wired, uint32_t, clean, uint32_t, dirty);
5473
5474 MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_process_sync: task_freeze %s for pid %d [%s] - "
5475 "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, max_pages %d, shared %d\n",
5476 (ret == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (*p->p_name ? p->p_name : "(unknown)"),
5477 memorystatus_available_pages, purgeable, wired, clean, dirty, max_pages, shared);
5478
5479 proc_list_lock();
5480 p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
5481
5482 if (ret == KERN_SUCCESS) {
5483 memorystatus_freeze_entry_t data = { aPid, TRUE, dirty };
5484
5485 memorystatus_frozen_count++;
5486
5487 p->p_memstat_state |= (P_MEMSTAT_FROZEN | (shared ? 0: P_MEMSTAT_NORECLAIM));
5488
5489 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5490 /* Update stats */
5491 for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
5492 throttle_intervals[i].pageouts += dirty;
5493 }
5494 }
5495
5496 memorystatus_freeze_pageouts += dirty;
5497 memorystatus_freeze_count++;
5498
5499 proc_list_unlock();
5500
5501 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
5502 } else {
5503 proc_list_unlock();
5504 }
5505 }
5506
5507 exit:
5508 lck_mtx_unlock(&freezer_mutex);
5509 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_END,
5510 memorystatus_available_pages, aPid, 0, 0, 0);
5511
5512 return ret;
5513 }
5514
5515 static int
5516 memorystatus_freeze_top_process(boolean_t *memorystatus_freeze_swap_low)
5517 {
5518 pid_t aPid = 0;
5519 int ret = -1;
5520 proc_t p = PROC_NULL, next_p = PROC_NULL;
5521 unsigned int i = 0;
5522
5523 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_START,
5524 memorystatus_available_pages, 0, 0, 0, 0);
5525
5526 proc_list_lock();
5527
5528 next_p = memorystatus_get_first_proc_locked(&i, TRUE);
5529 while (next_p) {
5530 kern_return_t kr;
5531 uint32_t purgeable, wired, clean, dirty;
5532 boolean_t shared;
5533 uint32_t pages;
5534 uint32_t max_pages = 0;
5535 uint32_t state;
5536
5537 p = next_p;
5538 next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
5539
5540 aPid = p->p_pid;
5541 state = p->p_memstat_state;
5542
5543 /* Ensure the process is eligible for freezing */
5544 if ((state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FROZEN)) || !(state & P_MEMSTAT_SUSPENDED)) {
5545 continue; // with lock held
5546 }
5547
5548 /* Only freeze processes meeting our minimum resident page criteria */
5549 memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
5550 if (pages < memorystatus_freeze_pages_min) {
5551 continue; // with lock held
5552 }
5553
5554 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5555
5556 /* Ensure there's enough free space to freeze this process. */
5557
5558 unsigned int avail_swap_space = 0; /* in pages. */
5559
5560 /*
5561 * Freezer backed by the compressor and swap file(s)
5562 * while will hold compressed data.
5563 */
5564 avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
5565
5566 max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
5567
5568 if (max_pages < memorystatus_freeze_pages_min) {
5569 *memorystatus_freeze_swap_low = TRUE;
5570 proc_list_unlock();
5571 goto exit;
5572 }
5573 } else {
5574 /*
5575 * We only have the compressor pool.
5576 */
5577 max_pages = UINT32_MAX - 1;
5578 }
5579
5580 /* Mark as locked temporarily to avoid kill */
5581 p->p_memstat_state |= P_MEMSTAT_LOCKED;
5582
5583 p = proc_ref_locked(p);
5584 proc_list_unlock();
5585 if (!p) {
5586 goto exit;
5587 }
5588
5589 kr = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
5590
5591 MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_top_process: task_freeze %s for pid %d [%s] - "
5592 "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, max_pages %d, shared %d\n",
5593 (kr == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (*p->p_name ? p->p_name : "(unknown)"),
5594 memorystatus_available_pages, purgeable, wired, clean, dirty, max_pages, shared);
5595
5596 proc_list_lock();
5597 p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
5598
5599 /* Success? */
5600 if (KERN_SUCCESS == kr) {
5601 memorystatus_freeze_entry_t data = { aPid, TRUE, dirty };
5602
5603 memorystatus_frozen_count++;
5604
5605 p->p_memstat_state |= (P_MEMSTAT_FROZEN | (shared ? 0: P_MEMSTAT_NORECLAIM));
5606
5607 if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5608 /* Update stats */
5609 for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
5610 throttle_intervals[i].pageouts += dirty;
5611 }
5612 }
5613
5614 memorystatus_freeze_pageouts += dirty;
5615 memorystatus_freeze_count++;
5616
5617 proc_list_unlock();
5618
5619 memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
5620
5621 /* Return KERN_SUCESS */
5622 ret = kr;
5623
5624 } else {
5625 proc_list_unlock();
5626 }
5627
5628 proc_rele(p);
5629 goto exit;
5630 }
5631
5632 proc_list_unlock();
5633
5634 exit:
5635 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_END,
5636 memorystatus_available_pages, aPid, 0, 0, 0);
5637
5638 return ret;
5639 }
5640
5641 static inline boolean_t
5642 memorystatus_can_freeze_processes(void)
5643 {
5644 boolean_t ret;
5645
5646 proc_list_lock();
5647
5648 if (memorystatus_suspended_count) {
5649 uint32_t average_resident_pages, estimated_processes;
5650
5651 /* Estimate the number of suspended processes we can fit */
5652 average_resident_pages = memorystatus_suspended_footprint_total / memorystatus_suspended_count;
5653 estimated_processes = memorystatus_suspended_count +
5654 ((memorystatus_available_pages - memorystatus_available_pages_critical) / average_resident_pages);
5655
5656 /* If it's predicted that no freeze will occur, lower the threshold temporarily */
5657 if (estimated_processes <= FREEZE_SUSPENDED_THRESHOLD_DEFAULT) {
5658 memorystatus_freeze_suspended_threshold = FREEZE_SUSPENDED_THRESHOLD_LOW;
5659 } else {
5660 memorystatus_freeze_suspended_threshold = FREEZE_SUSPENDED_THRESHOLD_DEFAULT;
5661 }
5662
5663 MEMORYSTATUS_DEBUG(1, "memorystatus_can_freeze_processes: %d suspended processes, %d average resident pages / process, %d suspended processes estimated\n",
5664 memorystatus_suspended_count, average_resident_pages, estimated_processes);
5665
5666 if ((memorystatus_suspended_count - memorystatus_frozen_count) > memorystatus_freeze_suspended_threshold) {
5667 ret = TRUE;
5668 } else {
5669 ret = FALSE;
5670 }
5671 } else {
5672 ret = FALSE;
5673 }
5674
5675 proc_list_unlock();
5676
5677 return ret;
5678 }
5679
5680 static boolean_t
5681 memorystatus_can_freeze(boolean_t *memorystatus_freeze_swap_low)
5682 {
5683 boolean_t can_freeze = TRUE;
5684
5685 /* Only freeze if we're sufficiently low on memory; this holds off freeze right
5686 after boot, and is generally is a no-op once we've reached steady state. */
5687 if (memorystatus_available_pages > memorystatus_freeze_threshold) {
5688 return FALSE;
5689 }
5690
5691 /* Check minimum suspended process threshold. */
5692 if (!memorystatus_can_freeze_processes()) {
5693 return FALSE;
5694 }
5695 assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
5696
5697 if ( !VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
5698 /*
5699 * In-core compressor used for freezing WITHOUT on-disk swap support.
5700 */
5701 if (vm_compressor_low_on_space()) {
5702 if (*memorystatus_freeze_swap_low) {
5703 *memorystatus_freeze_swap_low = TRUE;
5704 }
5705
5706 can_freeze = FALSE;
5707
5708 } else {
5709 if (*memorystatus_freeze_swap_low) {
5710 *memorystatus_freeze_swap_low = FALSE;
5711 }
5712
5713 can_freeze = TRUE;
5714 }
5715 } else {
5716 /*
5717 * Freezing WITH on-disk swap support.
5718 *
5719 * In-core compressor fronts the swap.
5720 */
5721 if (vm_swap_low_on_space()) {
5722 if (*memorystatus_freeze_swap_low) {
5723 *memorystatus_freeze_swap_low = TRUE;
5724 }
5725
5726 can_freeze = FALSE;
5727 }
5728
5729 }
5730
5731 return can_freeze;
5732 }
5733
5734 static void
5735 memorystatus_freeze_update_throttle_interval(mach_timespec_t *ts, struct throttle_interval_t *interval)
5736 {
5737 unsigned int freeze_daily_pageouts_max = memorystatus_freeze_daily_mb_max * (1024 * 1024 / PAGE_SIZE);
5738 if (CMP_MACH_TIMESPEC(ts, &interval->ts) >= 0) {
5739 if (!interval->max_pageouts) {
5740 interval->max_pageouts = (interval->burst_multiple * (((uint64_t)interval->mins * freeze_daily_pageouts_max) / (24 * 60)));
5741 } else {
5742 printf("memorystatus_freeze_update_throttle_interval: %d minute throttle timeout, resetting\n", interval->mins);
5743 }
5744 interval->ts.tv_sec = interval->mins * 60;
5745 interval->ts.tv_nsec = 0;
5746 ADD_MACH_TIMESPEC(&interval->ts, ts);
5747 /* Since we update the throttle stats pre-freeze, adjust for overshoot here */
5748 if (interval->pageouts > interval->max_pageouts) {
5749 interval->pageouts -= interval->max_pageouts;
5750 } else {
5751 interval->pageouts = 0;
5752 }
5753 interval->throttle = FALSE;
5754 } else if (!interval->throttle && interval->pageouts >= interval->max_pageouts) {
5755 printf("memorystatus_freeze_update_throttle_interval: %d minute pageout limit exceeded; enabling throttle\n", interval->mins);
5756 interval->throttle = TRUE;
5757 }
5758
5759 MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_update_throttle_interval: throttle updated - %d frozen (%d max) within %dm; %dm remaining; throttle %s\n",
5760 interval->pageouts, interval->max_pageouts, interval->mins, (interval->ts.tv_sec - ts->tv_sec) / 60,
5761 interval->throttle ? "on" : "off");
5762 }
5763
5764 static boolean_t
5765 memorystatus_freeze_update_throttle(void)
5766 {
5767 clock_sec_t sec;
5768 clock_nsec_t nsec;
5769 mach_timespec_t ts;
5770 uint32_t i;
5771 boolean_t throttled = FALSE;
5772
5773 #if DEVELOPMENT || DEBUG
5774 if (!memorystatus_freeze_throttle_enabled)
5775 return FALSE;
5776 #endif
5777
5778 clock_get_system_nanotime(&sec, &nsec);
5779 ts.tv_sec = sec;
5780 ts.tv_nsec = nsec;
5781
5782 /* Check freeze pageouts over multiple intervals and throttle if we've exceeded our budget.
5783 *
5784 * This ensures that periods of inactivity can't be used as 'credit' towards freeze if the device has
5785 * remained dormant for a long period. We do, however, allow increased thresholds for shorter intervals in
5786 * order to allow for bursts of activity.
5787 */
5788 for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
5789 memorystatus_freeze_update_throttle_interval(&ts, &throttle_intervals[i]);
5790 if (throttle_intervals[i].throttle == TRUE)
5791 throttled = TRUE;
5792 }
5793
5794 return throttled;
5795 }
5796
5797 static void
5798 memorystatus_freeze_thread(void *param __unused, wait_result_t wr __unused)
5799 {
5800 static boolean_t memorystatus_freeze_swap_low = FALSE;
5801
5802 lck_mtx_lock(&freezer_mutex);
5803 if (memorystatus_freeze_enabled) {
5804 if (memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
5805 /* Only freeze if we've not exceeded our pageout budgets.*/
5806 if (!memorystatus_freeze_update_throttle()) {
5807 memorystatus_freeze_top_process(&memorystatus_freeze_swap_low);
5808 } else {
5809 printf("memorystatus_freeze_thread: in throttle, ignoring freeze\n");
5810 memorystatus_freeze_throttle_count++; /* Throttled, update stats */
5811 }
5812 }
5813 }
5814 lck_mtx_unlock(&freezer_mutex);
5815
5816 assert_wait((event_t) &memorystatus_freeze_wakeup, THREAD_UNINT);
5817 thread_block((thread_continue_t) memorystatus_freeze_thread);
5818 }
5819
5820 #endif /* CONFIG_FREEZE */
5821
5822 #if VM_PRESSURE_EVENTS
5823
5824 #if CONFIG_MEMORYSTATUS
5825
5826 static int
5827 memorystatus_send_note(int event_code, void *data, size_t data_length) {
5828 int ret;
5829 struct kev_msg ev_msg;
5830
5831 ev_msg.vendor_code = KEV_VENDOR_APPLE;
5832 ev_msg.kev_class = KEV_SYSTEM_CLASS;
5833 ev_msg.kev_subclass = KEV_MEMORYSTATUS_SUBCLASS;
5834
5835 ev_msg.event_code = event_code;
5836
5837 ev_msg.dv[0].data_length = data_length;
5838 ev_msg.dv[0].data_ptr = data;
5839 ev_msg.dv[1].data_length = 0;
5840
5841 ret = kev_post_msg(&ev_msg);
5842 if (ret) {
5843 printf("%s: kev_post_msg() failed, err %d\n", __func__, ret);
5844 }
5845
5846 return ret;
5847 }
5848
5849 boolean_t
5850 memorystatus_warn_process(pid_t pid, boolean_t limit_exceeded) {
5851
5852 boolean_t ret = FALSE;
5853 boolean_t found_knote = FALSE;
5854 struct knote *kn = NULL;
5855
5856 /*
5857 * See comment in sysctl_memorystatus_vm_pressure_send.
5858 */
5859
5860 memorystatus_klist_lock();
5861
5862 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
5863 proc_t knote_proc = knote_get_kq(kn)->kq_p;
5864 pid_t knote_pid = knote_proc->p_pid;
5865
5866 if (knote_pid == pid) {
5867 /*
5868 * By setting the "fflags" here, we are forcing
5869 * a process to deal with the case where it's
5870 * bumping up into its memory limits. If we don't
5871 * do this here, we will end up depending on the
5872 * system pressure snapshot evaluation in
5873 * filt_memorystatus().
5874 */
5875
5876 if (!limit_exceeded) {
5877
5878 /*
5879 * Processes on desktop are not expecting to handle a system-wide
5880 * critical or system-wide warning notification from this path.
5881 * Intentionally set only the unambiguous limit warning here.
5882 */
5883
5884 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
5885 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
5886 found_knote = TRUE;
5887 }
5888
5889 } else {
5890 /*
5891 * Send this notification when a process has exceeded a soft limit.
5892 */
5893 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
5894 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
5895 found_knote = TRUE;
5896 }
5897 }
5898 }
5899 }
5900
5901 if (found_knote) {
5902 KNOTE(&memorystatus_klist, 0);
5903 ret = TRUE;
5904 }
5905
5906 memorystatus_klist_unlock();
5907
5908 return ret;
5909 }
5910
5911 /*
5912 * Can only be set by the current task on itself.
5913 */
5914 int
5915 memorystatus_low_mem_privileged_listener(uint32_t op_flags)
5916 {
5917 boolean_t set_privilege = FALSE;
5918 /*
5919 * Need an entitlement check here?
5920 */
5921 if (op_flags == MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE) {
5922 set_privilege = TRUE;
5923 } else if (op_flags == MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE) {
5924 set_privilege = FALSE;
5925 } else {
5926 return EINVAL;
5927 }
5928
5929 return (task_low_mem_privileged_listener(current_task(), set_privilege, NULL));
5930 }
5931
5932 int
5933 memorystatus_send_pressure_note(pid_t pid) {
5934 MEMORYSTATUS_DEBUG(1, "memorystatus_send_pressure_note(): pid %d\n", pid);
5935 return memorystatus_send_note(kMemorystatusPressureNote, &pid, sizeof(pid));
5936 }
5937
5938 void
5939 memorystatus_send_low_swap_note(void) {
5940
5941 struct knote *kn = NULL;
5942
5943 memorystatus_klist_lock();
5944 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
5945 /* We call is_knote_registered_modify_task_pressure_bits to check if the sfflags for the
5946 * current note contain NOTE_MEMORYSTATUS_LOW_SWAP. Once we find one note in the memorystatus_klist
5947 * that has the NOTE_MEMORYSTATUS_LOW_SWAP flags in its sfflags set, we call KNOTE with
5948 * kMemoryStatusLowSwap as the hint to process and update all knotes on the memorystatus_klist accordingly. */
5949 if (is_knote_registered_modify_task_pressure_bits(kn, NOTE_MEMORYSTATUS_LOW_SWAP, NULL, 0, 0) == TRUE) {
5950 KNOTE(&memorystatus_klist, kMemorystatusLowSwap);
5951 break;
5952 }
5953 }
5954
5955 memorystatus_klist_unlock();
5956 }
5957
5958 boolean_t
5959 memorystatus_bg_pressure_eligible(proc_t p) {
5960 boolean_t eligible = FALSE;
5961
5962 proc_list_lock();
5963
5964 MEMORYSTATUS_DEBUG(1, "memorystatus_bg_pressure_eligible: pid %d, state 0x%x\n", p->p_pid, p->p_memstat_state);
5965
5966 /* Foreground processes have already been dealt with at this point, so just test for eligibility */
5967 if (!(p->p_memstat_state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_SUSPENDED | P_MEMSTAT_FROZEN))) {
5968 eligible = TRUE;
5969 }
5970
5971 proc_list_unlock();
5972
5973 return eligible;
5974 }
5975
5976 boolean_t
5977 memorystatus_is_foreground_locked(proc_t p) {
5978 return ((p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND) ||
5979 (p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND_SUPPORT));
5980 }
5981
5982 /*
5983 * This is meant for stackshot and kperf -- it does not take the proc_list_lock
5984 * to access the p_memstat_dirty field.
5985 */
5986 boolean_t
5987 memorystatus_proc_is_dirty_unsafe(void *v)
5988 {
5989 if (!v) {
5990 return FALSE;
5991 }
5992 proc_t p = (proc_t)v;
5993 return (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) != 0;
5994 }
5995
5996 #endif /* CONFIG_MEMORYSTATUS */
5997
5998 /*
5999 * Trigger levels to test the mechanism.
6000 * Can be used via a sysctl.
6001 */
6002 #define TEST_LOW_MEMORY_TRIGGER_ONE 1
6003 #define TEST_LOW_MEMORY_TRIGGER_ALL 2
6004 #define TEST_PURGEABLE_TRIGGER_ONE 3
6005 #define TEST_PURGEABLE_TRIGGER_ALL 4
6006 #define TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ONE 5
6007 #define TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL 6
6008
6009 boolean_t memorystatus_manual_testing_on = FALSE;
6010 vm_pressure_level_t memorystatus_manual_testing_level = kVMPressureNormal;
6011
6012 extern struct knote *
6013 vm_pressure_select_optimal_candidate_to_notify(struct klist *, int, boolean_t);
6014
6015 /*
6016 * This value is the threshold that a process must meet to be considered for scavenging.
6017 */
6018 #define VM_PRESSURE_MINIMUM_RSIZE 10 /* MB */
6019
6020 #define VM_PRESSURE_NOTIFY_WAIT_PERIOD 10000 /* milliseconds */
6021
6022 #if DEBUG
6023 #define VM_PRESSURE_DEBUG(cond, format, ...) \
6024 do { \
6025 if (cond) { printf(format, ##__VA_ARGS__); } \
6026 } while(0)
6027 #else
6028 #define VM_PRESSURE_DEBUG(cond, format, ...)
6029 #endif
6030
6031 #define INTER_NOTIFICATION_DELAY (250000) /* .25 second */
6032
6033 void memorystatus_on_pageout_scan_end(void) {
6034 /* No-op */
6035 }
6036
6037 /*
6038 * kn_max - knote
6039 *
6040 * knote_pressure_level - to check if the knote is registered for this notification level.
6041 *
6042 * task - task whose bits we'll be modifying
6043 *
6044 * 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.
6045 *
6046 * pressure_level_to_set - the task is about to be notified of this new level. Update the task's bit notification information appropriately.
6047 *
6048 */
6049
6050 boolean_t
6051 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)
6052 {
6053 if (kn_max->kn_sfflags & knote_pressure_level) {
6054
6055 if (pressure_level_to_clear && task_has_been_notified(task, pressure_level_to_clear) == TRUE) {
6056
6057 task_clear_has_been_notified(task, pressure_level_to_clear);
6058 }
6059
6060 task_mark_has_been_notified(task, pressure_level_to_set);
6061 return TRUE;
6062 }
6063
6064 return FALSE;
6065 }
6066
6067 void
6068 memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear)
6069 {
6070 struct knote *kn = NULL;
6071
6072 memorystatus_klist_lock();
6073 SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
6074
6075 proc_t p = PROC_NULL;
6076 struct task* t = TASK_NULL;
6077
6078 p = knote_get_kq(kn)->kq_p;
6079 proc_list_lock();
6080 if (p != proc_ref_locked(p)) {
6081 p = PROC_NULL;
6082 proc_list_unlock();
6083 continue;
6084 }
6085 proc_list_unlock();
6086
6087 t = (struct task *)(p->task);
6088
6089 task_clear_has_been_notified(t, pressure_level_to_clear);
6090
6091 proc_rele(p);
6092 }
6093
6094 memorystatus_klist_unlock();
6095 }
6096
6097 extern kern_return_t vm_pressure_notify_dispatch_vm_clients(boolean_t target_foreground_process);
6098
6099 struct knote *
6100 vm_pressure_select_optimal_candidate_to_notify(struct klist *candidate_list, int level, boolean_t target_foreground_process);
6101
6102 /*
6103 * Used by the vm_pressure_thread which is
6104 * signalled from within vm_pageout_scan().
6105 */
6106 static void vm_dispatch_memory_pressure(void);
6107 void consider_vm_pressure_events(void);
6108
6109 void consider_vm_pressure_events(void)
6110 {
6111 vm_dispatch_memory_pressure();
6112 }
6113 static void vm_dispatch_memory_pressure(void)
6114 {
6115 memorystatus_update_vm_pressure(FALSE);
6116 }
6117
6118 extern vm_pressure_level_t
6119 convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
6120
6121 struct knote *
6122 vm_pressure_select_optimal_candidate_to_notify(struct klist *candidate_list, int level, boolean_t target_foreground_process)
6123 {
6124 struct knote *kn = NULL, *kn_max = NULL;
6125 uint64_t resident_max = 0; /* MB */
6126 struct timeval curr_tstamp = {0, 0};
6127 int elapsed_msecs = 0;
6128 int selected_task_importance = 0;
6129 static int pressure_snapshot = -1;
6130 boolean_t pressure_increase = FALSE;
6131
6132 if (pressure_snapshot == -1) {
6133 /*
6134 * Initial snapshot.
6135 */
6136 pressure_snapshot = level;
6137 pressure_increase = TRUE;
6138 } else {
6139
6140 if (level >= pressure_snapshot) {
6141 pressure_increase = TRUE;
6142 } else {
6143 pressure_increase = FALSE;
6144 }
6145
6146 pressure_snapshot = level;
6147 }
6148
6149 if (pressure_increase == TRUE) {
6150 /*
6151 * We'll start by considering the largest
6152 * unimportant task in our list.
6153 */
6154 selected_task_importance = INT_MAX;
6155 } else {
6156 /*
6157 * We'll start by considering the largest
6158 * important task in our list.
6159 */
6160 selected_task_importance = 0;
6161 }
6162
6163 microuptime(&curr_tstamp);
6164
6165 SLIST_FOREACH(kn, candidate_list, kn_selnext) {
6166
6167 uint64_t resident_size = 0; /* MB */
6168 proc_t p = PROC_NULL;
6169 struct task* t = TASK_NULL;
6170 int curr_task_importance = 0;
6171 boolean_t consider_knote = FALSE;
6172 boolean_t privileged_listener = FALSE;
6173
6174 p = knote_get_kq(kn)->kq_p;
6175 proc_list_lock();
6176 if (p != proc_ref_locked(p)) {
6177 p = PROC_NULL;
6178 proc_list_unlock();
6179 continue;
6180 }
6181 proc_list_unlock();
6182
6183 #if CONFIG_MEMORYSTATUS
6184 if (target_foreground_process == TRUE && !memorystatus_is_foreground_locked(p)) {
6185 /*
6186 * Skip process not marked foreground.
6187 */
6188 proc_rele(p);
6189 continue;
6190 }
6191 #endif /* CONFIG_MEMORYSTATUS */
6192
6193 t = (struct task *)(p->task);
6194
6195 timevalsub(&curr_tstamp, &p->vm_pressure_last_notify_tstamp);
6196 elapsed_msecs = curr_tstamp.tv_sec * 1000 + curr_tstamp.tv_usec / 1000;
6197
6198 vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(level);
6199
6200 if ((kn->kn_sfflags & dispatch_level) == 0) {
6201 proc_rele(p);
6202 continue;
6203 }
6204
6205 #if CONFIG_MEMORYSTATUS
6206 if (target_foreground_process == FALSE && !memorystatus_bg_pressure_eligible(p)) {
6207 VM_PRESSURE_DEBUG(1, "[vm_pressure] skipping process %d\n", p->p_pid);
6208 proc_rele(p);
6209 continue;
6210 }
6211 #endif /* CONFIG_MEMORYSTATUS */
6212
6213 curr_task_importance = task_importance_estimate(t);
6214
6215 /*
6216 * Privileged listeners are only considered in the multi-level pressure scheme
6217 * AND only if the pressure is increasing.
6218 */
6219 if (level > 0) {
6220
6221 if (task_has_been_notified(t, level) == FALSE) {
6222
6223 /*
6224 * Is this a privileged listener?
6225 */
6226 if (task_low_mem_privileged_listener(t, FALSE, &privileged_listener) == 0) {
6227
6228 if (privileged_listener) {
6229 kn_max = kn;
6230 proc_rele(p);
6231 goto done_scanning;
6232 }
6233 }
6234 } else {
6235 proc_rele(p);
6236 continue;
6237 }
6238 } else if (level == 0) {
6239
6240 /*
6241 * Task wasn't notified when the pressure was increasing and so
6242 * no need to notify it that the pressure is decreasing.
6243 */
6244 if ((task_has_been_notified(t, kVMPressureWarning) == FALSE) && (task_has_been_notified(t, kVMPressureCritical) == FALSE)) {
6245 proc_rele(p);
6246 continue;
6247 }
6248 }
6249
6250 /*
6251 * We don't want a small process to block large processes from
6252 * being notified again. <rdar://problem/7955532>
6253 */
6254 resident_size = (get_task_phys_footprint(t))/(1024*1024ULL); /* MB */
6255
6256 if (resident_size >= VM_PRESSURE_MINIMUM_RSIZE) {
6257
6258 if (level > 0) {
6259 /*
6260 * Warning or Critical Pressure.
6261 */
6262 if (pressure_increase) {
6263 if ((curr_task_importance < selected_task_importance) ||
6264 ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
6265
6266 /*
6267 * We have found a candidate process which is:
6268 * a) at a lower importance than the current selected process
6269 * OR
6270 * b) has importance equal to that of the current selected process but is larger
6271 */
6272
6273 consider_knote = TRUE;
6274 }
6275 } else {
6276 if ((curr_task_importance > selected_task_importance) ||
6277 ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
6278
6279 /*
6280 * We have found a candidate process which is:
6281 * a) at a higher importance than the current selected process
6282 * OR
6283 * b) has importance equal to that of the current selected process but is larger
6284 */
6285
6286 consider_knote = TRUE;
6287 }
6288 }
6289 } else if (level == 0) {
6290 /*
6291 * Pressure back to normal.
6292 */
6293 if ((curr_task_importance > selected_task_importance) ||
6294 ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
6295
6296 consider_knote = TRUE;
6297 }
6298 }
6299
6300 if (consider_knote) {
6301 resident_max = resident_size;
6302 kn_max = kn;
6303 selected_task_importance = curr_task_importance;
6304 consider_knote = FALSE; /* reset for the next candidate */
6305 }
6306 } else {
6307 /* There was no candidate with enough resident memory to scavenge */
6308 VM_PRESSURE_DEBUG(0, "[vm_pressure] threshold failed for pid %d with %llu resident...\n", p->p_pid, resident_size);
6309 }
6310 proc_rele(p);
6311 }
6312
6313 done_scanning:
6314 if (kn_max) {
6315 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);
6316 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);
6317 }
6318
6319 return kn_max;
6320 }
6321
6322 #define VM_PRESSURE_DECREASED_SMOOTHING_PERIOD 5000 /* milliseconds */
6323 #define WARNING_NOTIFICATION_RESTING_PERIOD 25 /* seconds */
6324 #define CRITICAL_NOTIFICATION_RESTING_PERIOD 25 /* seconds */
6325
6326 uint64_t next_warning_notification_sent_at_ts = 0;
6327 uint64_t next_critical_notification_sent_at_ts = 0;
6328
6329 kern_return_t
6330 memorystatus_update_vm_pressure(boolean_t target_foreground_process)
6331 {
6332 struct knote *kn_max = NULL;
6333 struct knote *kn_cur = NULL, *kn_temp = NULL; /* for safe list traversal */
6334 pid_t target_pid = -1;
6335 struct klist dispatch_klist = { NULL };
6336 proc_t target_proc = PROC_NULL;
6337 struct task *task = NULL;
6338 boolean_t found_candidate = FALSE;
6339
6340 static vm_pressure_level_t level_snapshot = kVMPressureNormal;
6341 static vm_pressure_level_t prev_level_snapshot = kVMPressureNormal;
6342 boolean_t smoothing_window_started = FALSE;
6343 struct timeval smoothing_window_start_tstamp = {0, 0};
6344 struct timeval curr_tstamp = {0, 0};
6345 int elapsed_msecs = 0;
6346 uint64_t curr_ts = mach_absolute_time();
6347
6348 #if !CONFIG_JETSAM
6349 #define MAX_IDLE_KILLS 100 /* limit the number of idle kills allowed */
6350
6351 int idle_kill_counter = 0;
6352
6353 /*
6354 * On desktop we take this opportunity to free up memory pressure
6355 * by immediately killing idle exitable processes. We use a delay
6356 * to avoid overkill. And we impose a max counter as a fail safe
6357 * in case daemons re-launch too fast.
6358 */
6359 while ((memorystatus_vm_pressure_level != kVMPressureNormal) && (idle_kill_counter < MAX_IDLE_KILLS)) {
6360 if (memorystatus_idle_exit_from_VM() == FALSE) {
6361 /* No idle exitable processes left to kill */
6362 break;
6363 }
6364 idle_kill_counter++;
6365
6366 if (memorystatus_manual_testing_on == TRUE) {
6367 /*
6368 * Skip the delay when testing
6369 * the pressure notification scheme.
6370 */
6371 } else {
6372 delay(1000000); /* 1 second */
6373 }
6374 }
6375 #endif /* !CONFIG_JETSAM */
6376
6377 if (level_snapshot != kVMPressureNormal) {
6378
6379 /*
6380 * Check to see if we are still in the 'resting' period
6381 * after having notified all clients interested in
6382 * a particular pressure level.
6383 */
6384
6385 level_snapshot = memorystatus_vm_pressure_level;
6386
6387 if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
6388
6389 if (curr_ts < next_warning_notification_sent_at_ts) {
6390 delay(INTER_NOTIFICATION_DELAY * 4 /* 1 sec */);
6391 return KERN_SUCCESS;
6392 }
6393 } else if (level_snapshot == kVMPressureCritical) {
6394
6395 if (curr_ts < next_critical_notification_sent_at_ts) {
6396 delay(INTER_NOTIFICATION_DELAY * 4 /* 1 sec */);
6397 return KERN_SUCCESS;
6398 }
6399 }
6400 }
6401
6402 while (1) {
6403
6404 /*
6405 * There is a race window here. But it's not clear
6406 * how much we benefit from having extra synchronization.
6407 */
6408 level_snapshot = memorystatus_vm_pressure_level;
6409
6410 if (prev_level_snapshot > level_snapshot) {
6411 /*
6412 * Pressure decreased? Let's take a little breather
6413 * and see if this condition stays.
6414 */
6415 if (smoothing_window_started == FALSE) {
6416
6417 smoothing_window_started = TRUE;
6418 microuptime(&smoothing_window_start_tstamp);
6419 }
6420
6421 microuptime(&curr_tstamp);
6422 timevalsub(&curr_tstamp, &smoothing_window_start_tstamp);
6423 elapsed_msecs = curr_tstamp.tv_sec * 1000 + curr_tstamp.tv_usec / 1000;
6424
6425 if (elapsed_msecs < VM_PRESSURE_DECREASED_SMOOTHING_PERIOD) {
6426
6427 delay(INTER_NOTIFICATION_DELAY);
6428 continue;
6429 }
6430 }
6431
6432 prev_level_snapshot = level_snapshot;
6433 smoothing_window_started = FALSE;
6434
6435 memorystatus_klist_lock();
6436 kn_max = vm_pressure_select_optimal_candidate_to_notify(&memorystatus_klist, level_snapshot, target_foreground_process);
6437
6438 if (kn_max == NULL) {
6439 memorystatus_klist_unlock();
6440
6441 /*
6442 * No more level-based clients to notify.
6443 *
6444 * Start the 'resting' window within which clients will not be re-notified.
6445 */
6446
6447 if (level_snapshot != kVMPressureNormal) {
6448 if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
6449 nanoseconds_to_absolutetime(WARNING_NOTIFICATION_RESTING_PERIOD * NSEC_PER_SEC, &curr_ts);
6450 next_warning_notification_sent_at_ts = mach_absolute_time() + curr_ts;
6451
6452 memorystatus_klist_reset_all_for_level(kVMPressureWarning);
6453 }
6454
6455 if (level_snapshot == kVMPressureCritical) {
6456 nanoseconds_to_absolutetime(CRITICAL_NOTIFICATION_RESTING_PERIOD * NSEC_PER_SEC, &curr_ts);
6457 next_critical_notification_sent_at_ts = mach_absolute_time() + curr_ts;
6458
6459 memorystatus_klist_reset_all_for_level(kVMPressureCritical);
6460 }
6461 }
6462 return KERN_FAILURE;
6463 }
6464
6465 target_proc = knote_get_kq(kn_max)->kq_p;
6466
6467 proc_list_lock();
6468 if (target_proc != proc_ref_locked(target_proc)) {
6469 target_proc = PROC_NULL;
6470 proc_list_unlock();
6471 memorystatus_klist_unlock();
6472 continue;
6473 }
6474 proc_list_unlock();
6475
6476 target_pid = target_proc->p_pid;
6477
6478 task = (struct task *)(target_proc->task);
6479
6480 if (level_snapshot != kVMPressureNormal) {
6481
6482 if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
6483
6484 if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_WARN, task, 0, kVMPressureWarning) == TRUE) {
6485 found_candidate = TRUE;
6486 }
6487 } else {
6488 if (level_snapshot == kVMPressureCritical) {
6489
6490 if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_CRITICAL, task, 0, kVMPressureCritical) == TRUE) {
6491 found_candidate = TRUE;
6492 }
6493 }
6494 }
6495 } else {
6496 if (kn_max->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
6497
6498 task_clear_has_been_notified(task, kVMPressureWarning);
6499 task_clear_has_been_notified(task, kVMPressureCritical);
6500
6501 found_candidate = TRUE;
6502 }
6503 }
6504
6505 if (found_candidate == FALSE) {
6506 proc_rele(target_proc);
6507 memorystatus_klist_unlock();
6508 continue;
6509 }
6510
6511 SLIST_FOREACH_SAFE(kn_cur, &memorystatus_klist, kn_selnext, kn_temp) {
6512
6513 int knote_pressure_level = convert_internal_pressure_level_to_dispatch_level(level_snapshot);
6514
6515 if (is_knote_registered_modify_task_pressure_bits(kn_cur, knote_pressure_level, task, 0, level_snapshot) == TRUE) {
6516 proc_t knote_proc = knote_get_kq(kn_cur)->kq_p;
6517 pid_t knote_pid = knote_proc->p_pid;
6518 if (knote_pid == target_pid) {
6519 KNOTE_DETACH(&memorystatus_klist, kn_cur);
6520 KNOTE_ATTACH(&dispatch_klist, kn_cur);
6521 }
6522 }
6523 }
6524
6525 KNOTE(&dispatch_klist, (level_snapshot != kVMPressureNormal) ? kMemorystatusPressure : kMemorystatusNoPressure);
6526
6527 SLIST_FOREACH_SAFE(kn_cur, &dispatch_klist, kn_selnext, kn_temp) {
6528 KNOTE_DETACH(&dispatch_klist, kn_cur);
6529 KNOTE_ATTACH(&memorystatus_klist, kn_cur);
6530 }
6531
6532 memorystatus_klist_unlock();
6533
6534 microuptime(&target_proc->vm_pressure_last_notify_tstamp);
6535 proc_rele(target_proc);
6536
6537 if (memorystatus_manual_testing_on == TRUE && target_foreground_process == TRUE) {
6538 break;
6539 }
6540
6541 if (memorystatus_manual_testing_on == TRUE) {
6542 /*
6543 * Testing out the pressure notification scheme.
6544 * No need for delays etc.
6545 */
6546 } else {
6547
6548 uint32_t sleep_interval = INTER_NOTIFICATION_DELAY;
6549 #if CONFIG_JETSAM
6550 unsigned int page_delta = 0;
6551 unsigned int skip_delay_page_threshold = 0;
6552
6553 assert(memorystatus_available_pages_pressure >= memorystatus_available_pages_critical_base);
6554
6555 page_delta = (memorystatus_available_pages_pressure - memorystatus_available_pages_critical_base) / 2;
6556 skip_delay_page_threshold = memorystatus_available_pages_pressure - page_delta;
6557
6558 if (memorystatus_available_pages <= skip_delay_page_threshold) {
6559 /*
6560 * We are nearing the critcal mark fast and can't afford to wait between
6561 * notifications.
6562 */
6563 sleep_interval = 0;
6564 }
6565 #endif /* CONFIG_JETSAM */
6566
6567 if (sleep_interval) {
6568 delay(sleep_interval);
6569 }
6570 }
6571 }
6572
6573 return KERN_SUCCESS;
6574 }
6575
6576 vm_pressure_level_t
6577 convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t internal_pressure_level)
6578 {
6579 vm_pressure_level_t dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
6580
6581 switch (internal_pressure_level) {
6582
6583 case kVMPressureNormal:
6584 {
6585 dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
6586 break;
6587 }
6588
6589 case kVMPressureWarning:
6590 case kVMPressureUrgent:
6591 {
6592 dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_WARN;
6593 break;
6594 }
6595
6596 case kVMPressureCritical:
6597 {
6598 dispatch_level = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
6599 break;
6600 }
6601
6602 default:
6603 break;
6604 }
6605
6606 return dispatch_level;
6607 }
6608
6609 static int
6610 sysctl_memorystatus_vm_pressure_level SYSCTL_HANDLER_ARGS
6611 {
6612 #pragma unused(arg1, arg2, oidp)
6613 vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(memorystatus_vm_pressure_level);
6614
6615 return SYSCTL_OUT(req, &dispatch_level, sizeof(dispatch_level));
6616 }
6617
6618 #if DEBUG || DEVELOPMENT
6619
6620 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_LOCKED,
6621 0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
6622
6623 #else /* DEBUG || DEVELOPMENT */
6624
6625 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_LOCKED|CTLFLAG_MASKED,
6626 0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
6627
6628 #endif /* DEBUG || DEVELOPMENT */
6629
6630 extern int memorystatus_purge_on_warning;
6631 extern int memorystatus_purge_on_critical;
6632
6633 static int
6634 sysctl_memorypressure_manual_trigger SYSCTL_HANDLER_ARGS
6635 {
6636 #pragma unused(arg1, arg2)
6637
6638 int level = 0;
6639 int error = 0;
6640 int pressure_level = 0;
6641 int trigger_request = 0;
6642 int force_purge;
6643
6644 error = sysctl_handle_int(oidp, &level, 0, req);
6645 if (error || !req->newptr) {
6646 return (error);
6647 }
6648
6649 memorystatus_manual_testing_on = TRUE;
6650
6651 trigger_request = (level >> 16) & 0xFFFF;
6652 pressure_level = (level & 0xFFFF);
6653
6654 if (trigger_request < TEST_LOW_MEMORY_TRIGGER_ONE ||
6655 trigger_request > TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL) {
6656 return EINVAL;
6657 }
6658 switch (pressure_level) {
6659 case NOTE_MEMORYSTATUS_PRESSURE_NORMAL:
6660 case NOTE_MEMORYSTATUS_PRESSURE_WARN:
6661 case NOTE_MEMORYSTATUS_PRESSURE_CRITICAL:
6662 break;
6663 default:
6664 return EINVAL;
6665 }
6666
6667 /*
6668 * The pressure level is being set from user-space.
6669 * And user-space uses the constants in sys/event.h
6670 * So we translate those events to our internal levels here.
6671 */
6672 if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
6673
6674 memorystatus_manual_testing_level = kVMPressureNormal;
6675 force_purge = 0;
6676
6677 } else if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_WARN) {
6678
6679 memorystatus_manual_testing_level = kVMPressureWarning;
6680 force_purge = memorystatus_purge_on_warning;
6681
6682 } else if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
6683
6684 memorystatus_manual_testing_level = kVMPressureCritical;
6685 force_purge = memorystatus_purge_on_critical;
6686 }
6687
6688 memorystatus_vm_pressure_level = memorystatus_manual_testing_level;
6689
6690 /* purge according to the new pressure level */
6691 switch (trigger_request) {
6692 case TEST_PURGEABLE_TRIGGER_ONE:
6693 case TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ONE:
6694 if (force_purge == 0) {
6695 /* no purging requested */
6696 break;
6697 }
6698 vm_purgeable_object_purge_one_unlocked(force_purge);
6699 break;
6700 case TEST_PURGEABLE_TRIGGER_ALL:
6701 case TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL:
6702 if (force_purge == 0) {
6703 /* no purging requested */
6704 break;
6705 }
6706 while (vm_purgeable_object_purge_one_unlocked(force_purge));
6707 break;
6708 }
6709
6710 if ((trigger_request == TEST_LOW_MEMORY_TRIGGER_ONE) ||
6711 (trigger_request == TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ONE)) {
6712
6713 memorystatus_update_vm_pressure(TRUE);
6714 }
6715
6716 if ((trigger_request == TEST_LOW_MEMORY_TRIGGER_ALL) ||
6717 (trigger_request == TEST_LOW_MEMORY_PURGEABLE_TRIGGER_ALL)) {
6718
6719 while (memorystatus_update_vm_pressure(FALSE) == KERN_SUCCESS) {
6720 continue;
6721 }
6722 }
6723
6724 if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
6725 memorystatus_manual_testing_on = FALSE;
6726 }
6727
6728 return 0;
6729 }
6730
6731 SYSCTL_PROC(_kern, OID_AUTO, memorypressure_manual_trigger, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
6732 0, 0, &sysctl_memorypressure_manual_trigger, "I", "");
6733
6734
6735 extern int memorystatus_purge_on_warning;
6736 extern int memorystatus_purge_on_urgent;
6737 extern int memorystatus_purge_on_critical;
6738
6739 SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_warning, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_purge_on_warning, 0, "");
6740 SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_urgent, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_purge_on_urgent, 0, "");
6741 SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_critical, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_purge_on_critical, 0, "");
6742
6743
6744 #endif /* VM_PRESSURE_EVENTS */
6745
6746 /* Return both allocated and actual size, since there's a race between allocation and list compilation */
6747 static int
6748 memorystatus_get_priority_list(memorystatus_priority_entry_t **list_ptr, size_t *buffer_size, size_t *list_size, boolean_t size_only)
6749 {
6750 uint32_t list_count, i = 0;
6751 memorystatus_priority_entry_t *list_entry;
6752 proc_t p;
6753
6754 list_count = memorystatus_list_count;
6755 *list_size = sizeof(memorystatus_priority_entry_t) * list_count;
6756
6757 /* Just a size check? */
6758 if (size_only) {
6759 return 0;
6760 }
6761
6762 /* Otherwise, validate the size of the buffer */
6763 if (*buffer_size < *list_size) {
6764 return EINVAL;
6765 }
6766
6767 *list_ptr = (memorystatus_priority_entry_t*)kalloc(*list_size);
6768 if (!list_ptr) {
6769 return ENOMEM;
6770 }
6771
6772 memset(*list_ptr, 0, *list_size);
6773
6774 *buffer_size = *list_size;
6775 *list_size = 0;
6776
6777 list_entry = *list_ptr;
6778
6779 proc_list_lock();
6780
6781 p = memorystatus_get_first_proc_locked(&i, TRUE);
6782 while (p && (*list_size < *buffer_size)) {
6783 list_entry->pid = p->p_pid;
6784 list_entry->priority = p->p_memstat_effectivepriority;
6785 list_entry->user_data = p->p_memstat_userdata;
6786
6787 /*
6788 * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
6789 * Background limits are described via the inactive limit slots.
6790 * So, here, the cached limit should always be valid.
6791 */
6792
6793 if (p->p_memstat_memlimit <= 0) {
6794 task_get_phys_footprint_limit(p->task, &list_entry->limit);
6795 } else {
6796 list_entry->limit = p->p_memstat_memlimit;
6797 }
6798
6799 list_entry->state = memorystatus_build_state(p);
6800 list_entry++;
6801
6802 *list_size += sizeof(memorystatus_priority_entry_t);
6803
6804 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6805 }
6806
6807 proc_list_unlock();
6808
6809 MEMORYSTATUS_DEBUG(1, "memorystatus_get_priority_list: returning %lu for size\n", (unsigned long)*list_size);
6810
6811 return 0;
6812 }
6813
6814 static int
6815 memorystatus_cmd_get_priority_list(user_addr_t buffer, size_t buffer_size, int32_t *retval) {
6816 int error = EINVAL;
6817 boolean_t size_only;
6818 memorystatus_priority_entry_t *list = NULL;
6819 size_t list_size;
6820
6821 size_only = ((buffer == USER_ADDR_NULL) ? TRUE: FALSE);
6822
6823 error = memorystatus_get_priority_list(&list, &buffer_size, &list_size, size_only);
6824 if (error) {
6825 goto out;
6826 }
6827
6828 if (!size_only) {
6829 error = copyout(list, buffer, list_size);
6830 }
6831
6832 if (error == 0) {
6833 *retval = list_size;
6834 }
6835 out:
6836
6837 if (list) {
6838 kfree(list, buffer_size);
6839 }
6840
6841 return error;
6842 }
6843
6844 #if CONFIG_JETSAM
6845
6846 static void
6847 memorystatus_clear_errors(void)
6848 {
6849 proc_t p;
6850 unsigned int i = 0;
6851
6852 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_START, 0, 0, 0, 0, 0);
6853
6854 proc_list_lock();
6855
6856 p = memorystatus_get_first_proc_locked(&i, TRUE);
6857 while (p) {
6858 if (p->p_memstat_state & P_MEMSTAT_ERROR) {
6859 p->p_memstat_state &= ~P_MEMSTAT_ERROR;
6860 }
6861 p = memorystatus_get_next_proc_locked(&i, p, TRUE);
6862 }
6863
6864 proc_list_unlock();
6865
6866 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_END, 0, 0, 0, 0, 0);
6867 }
6868
6869 static void
6870 memorystatus_update_levels_locked(boolean_t critical_only) {
6871
6872 memorystatus_available_pages_critical = memorystatus_available_pages_critical_base;
6873
6874 /*
6875 * If there's an entry in the first bucket, we have idle processes.
6876 */
6877
6878 memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
6879 if (first_bucket->count) {
6880 memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
6881
6882 if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure ) {
6883 /*
6884 * The critical threshold must never exceed the pressure threshold
6885 */
6886 memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
6887 }
6888 }
6889
6890 #if DEBUG || DEVELOPMENT
6891 if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
6892 memorystatus_available_pages_critical += memorystatus_jetsam_policy_offset_pages_diagnostic;
6893
6894 if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure ) {
6895 /*
6896 * The critical threshold must never exceed the pressure threshold
6897 */
6898 memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
6899 }
6900 }
6901 #endif
6902
6903 if (memorystatus_jetsam_policy & kPolicyMoreFree) {
6904 memorystatus_available_pages_critical += memorystatus_policy_more_free_offset_pages;
6905 }
6906
6907 if (critical_only) {
6908 return;
6909 }
6910
6911 #if VM_PRESSURE_EVENTS
6912 memorystatus_available_pages_pressure = (pressure_threshold_percentage / delta_percentage) * memorystatus_delta;
6913 #if DEBUG || DEVELOPMENT
6914 if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
6915 memorystatus_available_pages_pressure += memorystatus_jetsam_policy_offset_pages_diagnostic;
6916 }
6917 #endif
6918 #endif
6919 }
6920
6921 static int
6922 sysctl_kern_memorystatus_policy_more_free SYSCTL_HANDLER_ARGS
6923 {
6924 #pragma unused(arg1, arg2, oidp)
6925 int error = 0, more_free = 0;
6926
6927 /*
6928 * TODO: Enable this privilege check?
6929 *
6930 * error = priv_check_cred(kauth_cred_get(), PRIV_VM_JETSAM, 0);
6931 * if (error)
6932 * return (error);
6933 */
6934
6935 error = sysctl_handle_int(oidp, &more_free, 0, req);
6936 if (error || !req->newptr)
6937 return (error);
6938
6939 if ((more_free && ((memorystatus_jetsam_policy & kPolicyMoreFree) == kPolicyMoreFree)) ||
6940 (!more_free && ((memorystatus_jetsam_policy & kPolicyMoreFree) == 0))) {
6941
6942 /*
6943 * No change in state.
6944 */
6945 return 0;
6946 }
6947
6948 proc_list_lock();
6949
6950 if (more_free) {
6951 memorystatus_jetsam_policy |= kPolicyMoreFree;
6952 } else {
6953 memorystatus_jetsam_policy &= ~kPolicyMoreFree;
6954 }
6955
6956 memorystatus_update_levels_locked(TRUE);
6957
6958 proc_list_unlock();
6959
6960 return 0;
6961 }
6962 SYSCTL_PROC(_kern, OID_AUTO, memorystatus_policy_more_free, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
6963 0, 0, &sysctl_kern_memorystatus_policy_more_free, "I", "");
6964
6965 /*
6966 * Get the at_boot snapshot
6967 */
6968 static int
6969 memorystatus_get_at_boot_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
6970 size_t input_size = *snapshot_size;
6971
6972 /*
6973 * The at_boot snapshot has no entry list.
6974 */
6975 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t);
6976
6977 if (size_only) {
6978 return 0;
6979 }
6980
6981 /*
6982 * Validate the size of the snapshot buffer
6983 */
6984 if (input_size < *snapshot_size) {
6985 return EINVAL;
6986 }
6987
6988 /*
6989 * Update the notification_time only
6990 */
6991 memorystatus_at_boot_snapshot.notification_time = mach_absolute_time();
6992 *snapshot = &memorystatus_at_boot_snapshot;
6993
6994 MEMORYSTATUS_DEBUG(7, "memorystatus_get_at_boot_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%d)\n",
6995 (long)input_size, (long)*snapshot_size, 0);
6996 return 0;
6997 }
6998
6999 static int
7000 memorystatus_get_on_demand_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
7001 size_t input_size = *snapshot_size;
7002 uint32_t ods_list_count = memorystatus_list_count;
7003 memorystatus_jetsam_snapshot_t *ods = NULL; /* The on_demand snapshot buffer */
7004
7005 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (ods_list_count));
7006
7007 if (size_only) {
7008 return 0;
7009 }
7010
7011 /*
7012 * Validate the size of the snapshot buffer.
7013 * This is inherently racey. May want to revisit
7014 * this error condition and trim the output when
7015 * it doesn't fit.
7016 */
7017 if (input_size < *snapshot_size) {
7018 return EINVAL;
7019 }
7020
7021 /*
7022 * Allocate and initialize a snapshot buffer.
7023 */
7024 ods = (memorystatus_jetsam_snapshot_t *)kalloc(*snapshot_size);
7025 if (!ods) {
7026 return (ENOMEM);
7027 }
7028
7029 memset(ods, 0, *snapshot_size);
7030
7031 proc_list_lock();
7032 memorystatus_init_jetsam_snapshot_locked(ods, ods_list_count);
7033 proc_list_unlock();
7034
7035 /*
7036 * Return the kernel allocated, on_demand buffer.
7037 * The caller of this routine will copy the data out
7038 * to user space and then free the kernel allocated
7039 * buffer.
7040 */
7041 *snapshot = ods;
7042
7043 MEMORYSTATUS_DEBUG(7, "memorystatus_get_on_demand_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
7044 (long)input_size, (long)*snapshot_size, (long)ods_list_count);
7045
7046 return 0;
7047 }
7048
7049 static int
7050 memorystatus_get_jetsam_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
7051 size_t input_size = *snapshot_size;
7052
7053 if (memorystatus_jetsam_snapshot_count > 0) {
7054 *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count));
7055 } else {
7056 *snapshot_size = 0;
7057 }
7058
7059 if (size_only) {
7060 return 0;
7061 }
7062
7063 if (input_size < *snapshot_size) {
7064 return EINVAL;
7065 }
7066
7067 *snapshot = memorystatus_jetsam_snapshot;
7068
7069 MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
7070 (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_count);
7071
7072 return 0;
7073 }
7074
7075
7076 static int
7077 memorystatus_cmd_get_jetsam_snapshot(int32_t flags, user_addr_t buffer, size_t buffer_size, int32_t *retval) {
7078 int error = EINVAL;
7079 boolean_t size_only;
7080 boolean_t is_default_snapshot = FALSE;
7081 boolean_t is_on_demand_snapshot = FALSE;
7082 boolean_t is_at_boot_snapshot = FALSE;
7083 memorystatus_jetsam_snapshot_t *snapshot;
7084
7085 size_only = ((buffer == USER_ADDR_NULL) ? TRUE : FALSE);
7086
7087 if (flags == 0) {
7088 /* Default */
7089 is_default_snapshot = TRUE;
7090 error = memorystatus_get_jetsam_snapshot(&snapshot, &buffer_size, size_only);
7091 } else {
7092 if (flags & ~(MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT)) {
7093 /*
7094 * Unsupported bit set in flag.
7095 */
7096 return EINVAL;
7097 }
7098
7099 if ((flags & (MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT)) ==
7100 (MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT)) {
7101 /*
7102 * Can't have both set at the same time.
7103 */
7104 return EINVAL;
7105 }
7106
7107 if (flags & MEMORYSTATUS_SNAPSHOT_ON_DEMAND) {
7108 is_on_demand_snapshot = TRUE;
7109 /*
7110 * When not requesting the size only, the following call will allocate
7111 * an on_demand snapshot buffer, which is freed below.
7112 */
7113 error = memorystatus_get_on_demand_snapshot(&snapshot, &buffer_size, size_only);
7114
7115 } else if (flags & MEMORYSTATUS_SNAPSHOT_AT_BOOT) {
7116 is_at_boot_snapshot = TRUE;
7117 error = memorystatus_get_at_boot_snapshot(&snapshot, &buffer_size, size_only);
7118 } else {
7119 /*
7120 * Invalid flag setting.
7121 */
7122 return EINVAL;
7123 }
7124 }
7125
7126 if (error) {
7127 goto out;
7128 }
7129
7130 /*
7131 * Copy the data out to user space and clear the snapshot buffer.
7132 * If working with the jetsam snapshot,
7133 * clearing the buffer means, reset the count.
7134 * If working with an on_demand snapshot
7135 * clearing the buffer means, free it.
7136 * If working with the at_boot snapshot
7137 * there is nothing to clear or update.
7138 */
7139 if (!size_only) {
7140 if ((error = copyout(snapshot, buffer, buffer_size)) == 0) {
7141 if (is_default_snapshot) {
7142 /*
7143 * The jetsam snapshot is never freed, its count is simply reset.
7144 */
7145 proc_list_lock();
7146 snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
7147 memorystatus_jetsam_snapshot_last_timestamp = 0;
7148 proc_list_unlock();
7149 }
7150 }
7151
7152 if (is_on_demand_snapshot) {
7153 /*
7154 * The on_demand snapshot is always freed,
7155 * even if the copyout failed.
7156 */
7157 if(snapshot) {
7158 kfree(snapshot, buffer_size);
7159 }
7160 }
7161 }
7162
7163 if (error == 0) {
7164 *retval = buffer_size;
7165 }
7166 out:
7167 return error;
7168 }
7169
7170 /*
7171 * Routine: memorystatus_cmd_grp_set_properties
7172 * Purpose: Update properties for a group of processes.
7173 *
7174 * Supported Properties:
7175 * [priority]
7176 * Move each process out of its effective priority
7177 * band and into a new priority band.
7178 * Maintains relative order from lowest to highest priority.
7179 * In single band, maintains relative order from head to tail.
7180 *
7181 * eg: before [effectivepriority | pid]
7182 * [18 | p101 ]
7183 * [17 | p55, p67, p19 ]
7184 * [12 | p103 p10 ]
7185 * [ 7 | p25 ]
7186 * [ 0 | p71, p82, ]
7187 *
7188 * after [ new band | pid]
7189 * [ xxx | p71, p82, p25, p103, p10, p55, p67, p19, p101]
7190 *
7191 * Returns: 0 on success, else non-zero.
7192 *
7193 * Caveat: We know there is a race window regarding recycled pids.
7194 * A process could be killed before the kernel can act on it here.
7195 * If a pid cannot be found in any of the jetsam priority bands,
7196 * then we simply ignore it. No harm.
7197 * But, if the pid has been recycled then it could be an issue.
7198 * In that scenario, we might move an unsuspecting process to the new
7199 * priority band. It's not clear how the kernel can safeguard
7200 * against this, but it would be an extremely rare case anyway.
7201 * The caller of this api might avoid such race conditions by
7202 * ensuring that the processes passed in the pid list are suspended.
7203 */
7204
7205
7206 /* This internal structure can expand when we add support for more properties */
7207 typedef struct memorystatus_internal_properties
7208 {
7209 proc_t proc;
7210 int32_t priority; /* see memorytstatus_priority_entry_t : priority */
7211 } memorystatus_internal_properties_t;
7212
7213
7214 static int
7215 memorystatus_cmd_grp_set_properties(int32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
7216
7217 #pragma unused (flags)
7218
7219 /*
7220 * We only handle setting priority
7221 * per process
7222 */
7223
7224 int error = 0;
7225 memorystatus_priority_entry_t *entries = NULL;
7226 uint32_t entry_count = 0;
7227
7228 /* This will be the ordered proc list */
7229 memorystatus_internal_properties_t *table = NULL;
7230 size_t table_size = 0;
7231 uint32_t table_count = 0;
7232
7233 uint32_t i = 0;
7234 uint32_t bucket_index = 0;
7235 boolean_t head_insert;
7236 int32_t new_priority;
7237
7238 proc_t p;
7239
7240 /* Verify inputs */
7241 if ((buffer == USER_ADDR_NULL) || (buffer_size == 0) || ((buffer_size % sizeof(memorystatus_priority_entry_t)) != 0)) {
7242 error = EINVAL;
7243 goto out;
7244 }
7245
7246 entry_count = (buffer_size / sizeof(memorystatus_priority_entry_t));
7247 if ((entries = (memorystatus_priority_entry_t *)kalloc(buffer_size)) == NULL) {
7248 error = ENOMEM;
7249 goto out;
7250 }
7251
7252 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_START, entry_count, 0, 0, 0, 0);
7253
7254 if ((error = copyin(buffer, entries, buffer_size)) != 0) {
7255 goto out;
7256 }
7257
7258 /* Verify sanity of input priorities */
7259 for (i=0; i < entry_count; i++) {
7260 if (entries[i].priority == -1) {
7261 /* Use as shorthand for default priority */
7262 entries[i].priority = JETSAM_PRIORITY_DEFAULT;
7263 } else if ((entries[i].priority == system_procs_aging_band) || (entries[i].priority == applications_aging_band)) {
7264 /* Both the aging bands are reserved for internal use;
7265 * if requested, adjust to JETSAM_PRIORITY_IDLE. */
7266 entries[i].priority = JETSAM_PRIORITY_IDLE;
7267 } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
7268 /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle
7269 * queue */
7270 /* Deal with this later */
7271 } else if ((entries[i].priority < 0) || (entries[i].priority >= MEMSTAT_BUCKET_COUNT)) {
7272 /* Sanity check */
7273 error = EINVAL;
7274 goto out;
7275 }
7276 }
7277
7278 table_size = sizeof(memorystatus_internal_properties_t) * entry_count;
7279 if ( (table = (memorystatus_internal_properties_t *)kalloc(table_size)) == NULL) {
7280 error = ENOMEM;
7281 goto out;
7282 }
7283 memset(table, 0, table_size);
7284
7285
7286 /*
7287 * For each jetsam bucket entry, spin through the input property list.
7288 * When a matching pid is found, populate an adjacent table with the
7289 * appropriate proc pointer and new property values.
7290 * This traversal automatically preserves order from lowest
7291 * to highest priority.
7292 */
7293
7294 bucket_index=0;
7295
7296 proc_list_lock();
7297
7298 /* Create the ordered table */
7299 p = memorystatus_get_first_proc_locked(&bucket_index, TRUE);
7300 while (p && (table_count < entry_count)) {
7301 for (i=0; i < entry_count; i++ ) {
7302 if (p->p_pid == entries[i].pid) {
7303 /* Build the table data */
7304 table[table_count].proc = p;
7305 table[table_count].priority = entries[i].priority;
7306 table_count++;
7307 break;
7308 }
7309 }
7310 p = memorystatus_get_next_proc_locked(&bucket_index, p, TRUE);
7311 }
7312
7313 /* We now have ordered list of procs ready to move */
7314 for (i=0; i < table_count; i++) {
7315 p = table[i].proc;
7316 assert(p != NULL);
7317
7318 /* Allow head inserts -- but relative order is now */
7319 if (table[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
7320 new_priority = JETSAM_PRIORITY_IDLE;
7321 head_insert = true;
7322 } else {
7323 new_priority = table[i].priority;
7324 head_insert = false;
7325 }
7326
7327 /* Not allowed */
7328 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
7329 continue;
7330 }
7331
7332 /*
7333 * Take appropriate steps if moving proc out of
7334 * either of the aging bands.
7335 */
7336 if ((p->p_memstat_effectivepriority == system_procs_aging_band) || (p->p_memstat_effectivepriority == applications_aging_band)) {
7337 memorystatus_invalidate_idle_demotion_locked(p, TRUE);
7338 }
7339
7340 memorystatus_update_priority_locked(p, new_priority, head_insert, false);
7341 }
7342
7343 proc_list_unlock();
7344
7345 /*
7346 * if (table_count != entry_count)
7347 * then some pids were not found in a jetsam band.
7348 * harmless but interesting...
7349 */
7350 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_END, entry_count, table_count, 0, 0, 0);
7351
7352 out:
7353 if (entries)
7354 kfree(entries, buffer_size);
7355 if (table)
7356 kfree(table, table_size);
7357
7358 return (error);
7359 }
7360
7361
7362 /*
7363 * This routine is used to update a process's jetsam priority position and stored user_data.
7364 * It is not used for the setting of memory limits, which is why the last 6 args to the
7365 * memorystatus_update() call are 0 or FALSE.
7366 */
7367
7368 static int
7369 memorystatus_cmd_set_priority_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
7370 int error = 0;
7371 memorystatus_priority_properties_t mpp_entry;
7372
7373 /* Validate inputs */
7374 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_properties_t))) {
7375 return EINVAL;
7376 }
7377
7378 error = copyin(buffer, &mpp_entry, buffer_size);
7379
7380 if (error == 0) {
7381 proc_t p;
7382
7383 p = proc_find(pid);
7384 if (!p) {
7385 return ESRCH;
7386 }
7387
7388 if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
7389 proc_rele(p);
7390 return EPERM;
7391 }
7392
7393 error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, FALSE, FALSE, 0, 0, FALSE, FALSE, FALSE);
7394 proc_rele(p);
7395 }
7396
7397 return(error);
7398 }
7399
7400 static int
7401 memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
7402 int error = 0;
7403 memorystatus_memlimit_properties_t mmp_entry;
7404
7405 /* Validate inputs */
7406 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
7407 return EINVAL;
7408 }
7409
7410 error = copyin(buffer, &mmp_entry, buffer_size);
7411
7412 if (error == 0) {
7413 error = memorystatus_set_memlimit_properties(pid, &mmp_entry);
7414 }
7415
7416 return(error);
7417 }
7418
7419 /*
7420 * When getting the memlimit settings, we can't simply call task_get_phys_footprint_limit().
7421 * That gets the proc's cached memlimit and there is no guarantee that the active/inactive
7422 * limits will be the same in the no-limit case. Instead we convert limits <= 0 using
7423 * task_convert_phys_footprint_limit(). It computes the same limit value that would be written
7424 * to the task's ledgers via task_set_phys_footprint_limit().
7425 */
7426 static int
7427 memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
7428 int error = 0;
7429 memorystatus_memlimit_properties_t mmp_entry;
7430
7431 /* Validate inputs */
7432 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
7433 return EINVAL;
7434 }
7435
7436 memset (&mmp_entry, 0, sizeof(memorystatus_memlimit_properties_t));
7437
7438 proc_t p = proc_find(pid);
7439 if (!p) {
7440 return ESRCH;
7441 }
7442
7443 /*
7444 * Get the active limit and attributes.
7445 * No locks taken since we hold a reference to the proc.
7446 */
7447
7448 if (p->p_memstat_memlimit_active > 0 ) {
7449 mmp_entry.memlimit_active = p->p_memstat_memlimit_active;
7450 } else {
7451 task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_active);
7452 }
7453
7454 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) {
7455 mmp_entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7456 }
7457
7458 /*
7459 * Get the inactive limit and attributes
7460 */
7461 if (p->p_memstat_memlimit_inactive <= 0) {
7462 task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_inactive);
7463 } else {
7464 mmp_entry.memlimit_inactive = p->p_memstat_memlimit_inactive;
7465 }
7466 if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) {
7467 mmp_entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7468 }
7469 proc_rele(p);
7470
7471 error = copyout(&mmp_entry, buffer, buffer_size);
7472
7473 return(error);
7474 }
7475
7476
7477 /*
7478 * SPI for kbd - pr24956468
7479 * This is a very simple snapshot that calculates how much a
7480 * process's phys_footprint exceeds a specific memory limit.
7481 * Only the inactive memory limit is supported for now.
7482 * The delta is returned as bytes in excess or zero.
7483 */
7484 static int
7485 memorystatus_cmd_get_memlimit_excess_np(pid_t pid, uint32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
7486 int error = 0;
7487 uint64_t footprint_in_bytes = 0;
7488 uint64_t delta_in_bytes = 0;
7489 int32_t memlimit_mb = 0;
7490 uint64_t memlimit_bytes = 0;
7491
7492 /* Validate inputs */
7493 if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(uint64_t)) || (flags != 0)) {
7494 return EINVAL;
7495 }
7496
7497 proc_t p = proc_find(pid);
7498 if (!p) {
7499 return ESRCH;
7500 }
7501
7502 /*
7503 * Get the inactive limit.
7504 * No locks taken since we hold a reference to the proc.
7505 */
7506
7507 if (p->p_memstat_memlimit_inactive <= 0) {
7508 task_convert_phys_footprint_limit(-1, &memlimit_mb);
7509 } else {
7510 memlimit_mb = p->p_memstat_memlimit_inactive;
7511 }
7512
7513 footprint_in_bytes = get_task_phys_footprint(p->task);
7514
7515 proc_rele(p);
7516
7517 memlimit_bytes = memlimit_mb * 1024 * 1024; /* MB to bytes */
7518
7519 /*
7520 * Computed delta always returns >= 0 bytes
7521 */
7522 if (footprint_in_bytes > memlimit_bytes) {
7523 delta_in_bytes = footprint_in_bytes - memlimit_bytes;
7524 }
7525
7526 error = copyout(&delta_in_bytes, buffer, sizeof(delta_in_bytes));
7527
7528 return(error);
7529 }
7530
7531
7532 static int
7533 memorystatus_cmd_get_pressure_status(int32_t *retval) {
7534 int error;
7535
7536 /* Need privilege for check */
7537 error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
7538 if (error) {
7539 return (error);
7540 }
7541
7542 /* Inherently racy, so it's not worth taking a lock here */
7543 *retval = (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
7544
7545 return error;
7546 }
7547
7548 int
7549 memorystatus_get_pressure_status_kdp() {
7550 return (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
7551 }
7552
7553 /*
7554 * Every process, including a P_MEMSTAT_INTERNAL process (currently only pid 1), is allowed to set a HWM.
7555 *
7556 * This call is inflexible -- it does not distinguish between active/inactive, fatal/non-fatal
7557 * So, with 2-level HWM preserving previous behavior will map as follows.
7558 * - treat the limit passed in as both an active and inactive limit.
7559 * - treat the is_fatal_limit flag as though it applies to both active and inactive limits.
7560 *
7561 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK
7562 * - the is_fatal_limit is FALSE, meaning the active and inactive limits are non-fatal/soft
7563 * - so mapping is (active/non-fatal, inactive/non-fatal)
7564 *
7565 * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT
7566 * - the is_fatal_limit is TRUE, meaning the process's active and inactive limits are fatal/hard
7567 * - so mapping is (active/fatal, inactive/fatal)
7568 */
7569
7570 static int
7571 memorystatus_cmd_set_jetsam_memory_limit(pid_t pid, int32_t high_water_mark, __unused int32_t *retval, boolean_t is_fatal_limit) {
7572 int error = 0;
7573 memorystatus_memlimit_properties_t entry;
7574
7575 entry.memlimit_active = high_water_mark;
7576 entry.memlimit_active_attr = 0;
7577 entry.memlimit_inactive = high_water_mark;
7578 entry.memlimit_inactive_attr = 0;
7579
7580 if (is_fatal_limit == TRUE) {
7581 entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7582 entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
7583 }
7584
7585 error = memorystatus_set_memlimit_properties(pid, &entry);
7586 return (error);
7587 }
7588
7589 static int
7590 memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry) {
7591
7592 int32_t memlimit_active;
7593 boolean_t memlimit_active_is_fatal;
7594 int32_t memlimit_inactive;
7595 boolean_t memlimit_inactive_is_fatal;
7596 uint32_t valid_attrs = 0;
7597 int error = 0;
7598
7599 proc_t p = proc_find(pid);
7600 if (!p) {
7601 return ESRCH;
7602 }
7603
7604 /*
7605 * Check for valid attribute flags.
7606 */
7607 valid_attrs |= (MEMORYSTATUS_MEMLIMIT_ATTR_FATAL);
7608 if ((entry->memlimit_active_attr & (~valid_attrs)) != 0) {
7609 proc_rele(p);
7610 return EINVAL;
7611 }
7612 if ((entry->memlimit_inactive_attr & (~valid_attrs)) != 0) {
7613 proc_rele(p);
7614 return EINVAL;
7615 }
7616
7617 /*
7618 * Setup the active memlimit properties
7619 */
7620 memlimit_active = entry->memlimit_active;
7621 if (entry->memlimit_active_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL) {
7622 memlimit_active_is_fatal = TRUE;
7623 } else {
7624 memlimit_active_is_fatal = FALSE;
7625 }
7626
7627 /*
7628 * Setup the inactive memlimit properties
7629 */
7630 memlimit_inactive = entry->memlimit_inactive;
7631 if (entry->memlimit_inactive_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL) {
7632 memlimit_inactive_is_fatal = TRUE;
7633 } else {
7634 memlimit_inactive_is_fatal = FALSE;
7635 }
7636
7637 /*
7638 * Setting a limit of <= 0 implies that the process has no
7639 * high-water-mark and has no per-task-limit. That means
7640 * the system_wide task limit is in place, which by the way,
7641 * is always fatal.
7642 */
7643
7644 if (memlimit_active <= 0) {
7645 /*
7646 * Enforce the fatal system_wide task limit while process is active.
7647 */
7648 memlimit_active = -1;
7649 memlimit_active_is_fatal = TRUE;
7650 }
7651
7652 if (memlimit_inactive <= 0) {
7653 /*
7654 * Enforce the fatal system_wide task limit while process is inactive.
7655 */
7656 memlimit_inactive = -1;
7657 memlimit_inactive_is_fatal = TRUE;
7658 }
7659
7660 proc_list_lock();
7661
7662 /*
7663 * Store the active limit variants in the proc.
7664 */
7665 SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
7666
7667 /*
7668 * Store the inactive limit variants in the proc.
7669 */
7670 SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
7671
7672 /*
7673 * Enforce appropriate limit variant by updating the cached values
7674 * and writing the ledger.
7675 * Limit choice is based on process active/inactive state.
7676 */
7677
7678 if (memorystatus_highwater_enabled) {
7679 boolean_t trigger_exception;
7680 /*
7681 * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
7682 * Background limits are described via the inactive limit slots.
7683 */
7684
7685 if (proc_jetsam_state_is_active_locked(p) == TRUE) {
7686 CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
7687 } else {
7688 CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
7689 }
7690
7691 /* Enforce the limit by writing to the ledgers */
7692 assert(trigger_exception == TRUE);
7693 error = (task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, trigger_exception) == 0) ? 0 : EINVAL;
7694
7695 MEMORYSTATUS_DEBUG(3, "memorystatus_set_memlimit_properties: new limit on pid %d (%dMB %s) current priority (%d) dirty_state?=0x%x %s\n",
7696 p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
7697 (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
7698 (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
7699 DTRACE_MEMORYSTATUS2(memorystatus_set_memlimit, proc_t, p, int32_t, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1));
7700 }
7701
7702 proc_list_unlock();
7703 proc_rele(p);
7704
7705 return error;
7706 }
7707
7708 /*
7709 * Returns the jetsam priority (effective or requested) of the process
7710 * associated with this task.
7711 */
7712 int
7713 proc_get_memstat_priority(proc_t p, boolean_t effective_priority)
7714 {
7715 if (p) {
7716 if (effective_priority) {
7717 return p->p_memstat_effectivepriority;
7718 } else {
7719 return p->p_memstat_requestedpriority;
7720 }
7721 }
7722 return 0;
7723 }
7724
7725 #endif /* CONFIG_JETSAM */
7726
7727 int
7728 memorystatus_control(struct proc *p __unused, struct memorystatus_control_args *args, int *ret) {
7729 int error = EINVAL;
7730 os_reason_t jetsam_reason = OS_REASON_NULL;
7731
7732 #if !CONFIG_JETSAM
7733 #pragma unused(ret)
7734 #pragma unused(jetsam_reason)
7735 #endif
7736
7737 /* Need to be root or have entitlement */
7738 if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT)) {
7739 error = EPERM;
7740 goto out;
7741 }
7742
7743 /*
7744 * Sanity check.
7745 * Do not enforce it for snapshots.
7746 */
7747 if (args->command != MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT) {
7748 if (args->buffersize > MEMORYSTATUS_BUFFERSIZE_MAX) {
7749 error = EINVAL;
7750 goto out;
7751 }
7752 }
7753
7754 switch (args->command) {
7755 case MEMORYSTATUS_CMD_GET_PRIORITY_LIST:
7756 error = memorystatus_cmd_get_priority_list(args->buffer, args->buffersize, ret);
7757 break;
7758 #if CONFIG_JETSAM
7759 case MEMORYSTATUS_CMD_SET_PRIORITY_PROPERTIES:
7760 error = memorystatus_cmd_set_priority_properties(args->pid, args->buffer, args->buffersize, ret);
7761 break;
7762 case MEMORYSTATUS_CMD_SET_MEMLIMIT_PROPERTIES:
7763 error = memorystatus_cmd_set_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
7764 break;
7765 case MEMORYSTATUS_CMD_GET_MEMLIMIT_PROPERTIES:
7766 error = memorystatus_cmd_get_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
7767 break;
7768 case MEMORYSTATUS_CMD_GET_MEMLIMIT_EXCESS:
7769 error = memorystatus_cmd_get_memlimit_excess_np(args->pid, args->flags, args->buffer, args->buffersize, ret);
7770 break;
7771 case MEMORYSTATUS_CMD_GRP_SET_PROPERTIES:
7772 error = memorystatus_cmd_grp_set_properties((int32_t)args->flags, args->buffer, args->buffersize, ret);
7773 break;
7774 case MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT:
7775 error = memorystatus_cmd_get_jetsam_snapshot((int32_t)args->flags, args->buffer, args->buffersize, ret);
7776 break;
7777 case MEMORYSTATUS_CMD_GET_PRESSURE_STATUS:
7778 error = memorystatus_cmd_get_pressure_status(ret);
7779 break;
7780 case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
7781 /*
7782 * This call does not distinguish between active and inactive limits.
7783 * Default behavior in 2-level HWM world is to set both.
7784 * Non-fatal limit is also assumed for both.
7785 */
7786 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, FALSE);
7787 break;
7788 case MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT:
7789 /*
7790 * This call does not distinguish between active and inactive limits.
7791 * Default behavior in 2-level HWM world is to set both.
7792 * Fatal limit is also assumed for both.
7793 */
7794 error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
7795 break;
7796 /* Test commands */
7797 #if DEVELOPMENT || DEBUG
7798 case MEMORYSTATUS_CMD_TEST_JETSAM:
7799 jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_GENERIC);
7800 if (jetsam_reason == OS_REASON_NULL) {
7801 printf("memorystatus_control: failed to allocate jetsam reason\n");
7802 }
7803
7804 error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled, jetsam_reason) ? 0 : EINVAL;
7805 break;
7806 case MEMORYSTATUS_CMD_TEST_JETSAM_SORT:
7807 error = memorystatus_cmd_test_jetsam_sort(args->pid, (int32_t)args->flags);
7808 break;
7809 case MEMORYSTATUS_CMD_SET_JETSAM_PANIC_BITS:
7810 error = memorystatus_cmd_set_panic_bits(args->buffer, args->buffersize);
7811 break;
7812 #else /* DEVELOPMENT || DEBUG */
7813 #pragma unused(jetsam_reason)
7814 #endif /* DEVELOPMENT || DEBUG */
7815 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_ENABLE:
7816 if (memorystatus_aggressive_jetsam_lenient_allowed == FALSE) {
7817 #if DEVELOPMENT || DEBUG
7818 printf("Enabling Lenient Mode\n");
7819 #endif /* DEVELOPMENT || DEBUG */
7820
7821 memorystatus_aggressive_jetsam_lenient_allowed = TRUE;
7822 memorystatus_aggressive_jetsam_lenient = TRUE;
7823 error = 0;
7824 }
7825 break;
7826 case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_DISABLE:
7827 #if DEVELOPMENT || DEBUG
7828 printf("Disabling Lenient mode\n");
7829 #endif /* DEVELOPMENT || DEBUG */
7830 memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
7831 memorystatus_aggressive_jetsam_lenient = FALSE;
7832 error = 0;
7833 break;
7834 #endif /* CONFIG_JETSAM */
7835 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE:
7836 case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE:
7837 error = memorystatus_low_mem_privileged_listener(args->command);
7838 break;
7839
7840 #if CONFIG_JETSAM
7841 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE:
7842 case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE:
7843 error = memorystatus_update_inactive_jetsam_priority_band(args->pid, args->command, args->flags ? TRUE : FALSE);
7844 break;
7845 #endif /* CONFIG_JETSAM */
7846
7847 default:
7848 break;
7849 }
7850
7851 out:
7852 return error;
7853 }
7854
7855
7856 static int
7857 filt_memorystatusattach(struct knote *kn)
7858 {
7859 int error;
7860
7861 kn->kn_flags |= EV_CLEAR;
7862 error = memorystatus_knote_register(kn);
7863 if (error) {
7864 kn->kn_flags = EV_ERROR;
7865 kn->kn_data = error;
7866 }
7867 return 0;
7868 }
7869
7870 static void
7871 filt_memorystatusdetach(struct knote *kn)
7872 {
7873 memorystatus_knote_unregister(kn);
7874 }
7875
7876 static int
7877 filt_memorystatus(struct knote *kn __unused, long hint)
7878 {
7879 if (hint) {
7880 switch (hint) {
7881 case kMemorystatusNoPressure:
7882 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
7883 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
7884 }
7885 break;
7886 case kMemorystatusPressure:
7887 if (memorystatus_vm_pressure_level == kVMPressureWarning || memorystatus_vm_pressure_level == kVMPressureUrgent) {
7888 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
7889 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
7890 }
7891 } else if (memorystatus_vm_pressure_level == kVMPressureCritical) {
7892
7893 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
7894 kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
7895 }
7896 }
7897 break;
7898 case kMemorystatusLowSwap:
7899 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_LOW_SWAP) {
7900 kn->kn_fflags = NOTE_MEMORYSTATUS_LOW_SWAP;
7901 }
7902 break;
7903
7904 case kMemorystatusProcLimitWarn:
7905 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
7906 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
7907 }
7908 break;
7909
7910 case kMemorystatusProcLimitCritical:
7911 if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
7912 kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
7913 }
7914 break;
7915
7916 default:
7917 break;
7918 }
7919 }
7920
7921 return (kn->kn_fflags != 0);
7922 }
7923
7924 static int
7925 filt_memorystatustouch(struct knote *kn, struct kevent_internal_s *kev)
7926 {
7927 int res;
7928
7929 memorystatus_klist_lock();
7930
7931 /*
7932 * copy in new kevent settings
7933 * (saving the "desired" data and fflags).
7934 */
7935 kn->kn_sfflags = kev->fflags;
7936
7937 if ((kn->kn_status & KN_UDATA_SPECIFIC) == 0)
7938 kn->kn_udata = kev->udata;
7939
7940 /*
7941 * reset the output flags based on a
7942 * combination of the old events and
7943 * the new desired event list.
7944 */
7945 //kn->kn_fflags &= kn->kn_sfflags;
7946
7947 res = (kn->kn_fflags != 0);
7948
7949 memorystatus_klist_unlock();
7950
7951 return res;
7952 }
7953
7954 static int
7955 filt_memorystatusprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev)
7956 {
7957 #pragma unused(data)
7958 int res;
7959
7960 memorystatus_klist_lock();
7961 res = (kn->kn_fflags != 0);
7962 if (res) {
7963 *kev = kn->kn_kevent;
7964 kn->kn_flags |= EV_CLEAR; /* automatic */
7965 kn->kn_fflags = 0;
7966 kn->kn_data = 0;
7967 }
7968 memorystatus_klist_unlock();
7969
7970 return res;
7971 }
7972
7973 static void
7974 memorystatus_klist_lock(void) {
7975 lck_mtx_lock(&memorystatus_klist_mutex);
7976 }
7977
7978 static void
7979 memorystatus_klist_unlock(void) {
7980 lck_mtx_unlock(&memorystatus_klist_mutex);
7981 }
7982
7983 void
7984 memorystatus_kevent_init(lck_grp_t *grp, lck_attr_t *attr) {
7985 lck_mtx_init(&memorystatus_klist_mutex, grp, attr);
7986 klist_init(&memorystatus_klist);
7987 }
7988
7989 int
7990 memorystatus_knote_register(struct knote *kn) {
7991 int error = 0;
7992
7993 memorystatus_klist_lock();
7994
7995 if (kn->kn_sfflags & (NOTE_MEMORYSTATUS_PRESSURE_NORMAL | NOTE_MEMORYSTATUS_PRESSURE_WARN |
7996 NOTE_MEMORYSTATUS_PRESSURE_CRITICAL | NOTE_MEMORYSTATUS_LOW_SWAP |
7997 NOTE_MEMORYSTATUS_PROC_LIMIT_WARN | NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL)) {
7998
7999 KNOTE_ATTACH(&memorystatus_klist, kn);
8000
8001 } else {
8002 error = ENOTSUP;
8003 }
8004
8005 memorystatus_klist_unlock();
8006
8007 return error;
8008 }
8009
8010 void
8011 memorystatus_knote_unregister(struct knote *kn __unused) {
8012 memorystatus_klist_lock();
8013 KNOTE_DETACH(&memorystatus_klist, kn);
8014 memorystatus_klist_unlock();
8015 }
8016
8017
8018 #if 0
8019 #if CONFIG_JETSAM && VM_PRESSURE_EVENTS
8020 static boolean_t
8021 memorystatus_issue_pressure_kevent(boolean_t pressured) {
8022 memorystatus_klist_lock();
8023 KNOTE(&memorystatus_klist, pressured ? kMemorystatusPressure : kMemorystatusNoPressure);
8024 memorystatus_klist_unlock();
8025 return TRUE;
8026 }
8027 #endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
8028 #endif /* 0 */
8029
8030 #if CONFIG_JETSAM
8031 /* Coalition support */
8032
8033 /* sorting info for a particular priority bucket */
8034 typedef struct memstat_sort_info {
8035 coalition_t msi_coal;
8036 uint64_t msi_page_count;
8037 pid_t msi_pid;
8038 int msi_ntasks;
8039 } memstat_sort_info_t;
8040
8041 /*
8042 * qsort from smallest page count to largest page count
8043 *
8044 * return < 0 for a < b
8045 * 0 for a == b
8046 * > 0 for a > b
8047 */
8048 static int memstat_asc_cmp(const void *a, const void *b)
8049 {
8050 const memstat_sort_info_t *msA = (const memstat_sort_info_t *)a;
8051 const memstat_sort_info_t *msB = (const memstat_sort_info_t *)b;
8052
8053 return (int)((uint64_t)msA->msi_page_count - (uint64_t)msB->msi_page_count);
8054 }
8055
8056 /*
8057 * Return the number of pids rearranged during this sort.
8058 */
8059 static int
8060 memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order)
8061 {
8062 #define MAX_SORT_PIDS 80
8063 #define MAX_COAL_LEADERS 10
8064
8065 unsigned int b = bucket_index;
8066 int nleaders = 0;
8067 int ntasks = 0;
8068 proc_t p = NULL;
8069 coalition_t coal = COALITION_NULL;
8070 int pids_moved = 0;
8071 int total_pids_moved = 0;
8072 int i;
8073
8074 /*
8075 * The system is typically under memory pressure when in this
8076 * path, hence, we want to avoid dynamic memory allocation.
8077 */
8078 memstat_sort_info_t leaders[MAX_COAL_LEADERS];
8079 pid_t pid_list[MAX_SORT_PIDS];
8080
8081 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
8082 return(0);
8083 }
8084
8085 /*
8086 * Clear the array that holds coalition leader information
8087 */
8088 for (i=0; i < MAX_COAL_LEADERS; i++) {
8089 leaders[i].msi_coal = COALITION_NULL;
8090 leaders[i].msi_page_count = 0; /* will hold total coalition page count */
8091 leaders[i].msi_pid = 0; /* will hold coalition leader pid */
8092 leaders[i].msi_ntasks = 0; /* will hold the number of tasks in a coalition */
8093 }
8094
8095 p = memorystatus_get_first_proc_locked(&b, FALSE);
8096 while (p) {
8097 if (coalition_is_leader(p->task, COALITION_TYPE_JETSAM, &coal)) {
8098 if (nleaders < MAX_COAL_LEADERS) {
8099 int coal_ntasks = 0;
8100 uint64_t coal_page_count = coalition_get_page_count(coal, &coal_ntasks);
8101 leaders[nleaders].msi_coal = coal;
8102 leaders[nleaders].msi_page_count = coal_page_count;
8103 leaders[nleaders].msi_pid = p->p_pid; /* the coalition leader */
8104 leaders[nleaders].msi_ntasks = coal_ntasks;
8105 nleaders++;
8106 } else {
8107 /*
8108 * We've hit MAX_COAL_LEADERS meaning we can handle no more coalitions.
8109 * Abandoned coalitions will linger at the tail of the priority band
8110 * when this sort session ends.
8111 * TODO: should this be an assert?
8112 */
8113 printf("%s: WARNING: more than %d leaders in priority band [%d]\n",
8114 __FUNCTION__, MAX_COAL_LEADERS, bucket_index);
8115 break;
8116 }
8117 }
8118 p=memorystatus_get_next_proc_locked(&b, p, FALSE);
8119 }
8120
8121 if (nleaders == 0) {
8122 /* Nothing to sort */
8123 return(0);
8124 }
8125
8126 /*
8127 * Sort the coalition leader array, from smallest coalition page count
8128 * to largest coalition page count. When inserted in the priority bucket,
8129 * smallest coalition is handled first, resulting in the last to be jetsammed.
8130 */
8131 if (nleaders > 1) {
8132 qsort(leaders, nleaders, sizeof(memstat_sort_info_t), memstat_asc_cmp);
8133 }
8134
8135 #if 0
8136 for (i = 0; i < nleaders; i++) {
8137 printf("%s: coal_leader[%d of %d] pid[%d] pages[%llu] ntasks[%d]\n",
8138 __FUNCTION__, i, nleaders, leaders[i].msi_pid, leaders[i].msi_page_count,
8139 leaders[i].msi_ntasks);
8140 }
8141 #endif
8142
8143 /*
8144 * During coalition sorting, processes in a priority band are rearranged
8145 * by being re-inserted at the head of the queue. So, when handling a
8146 * list, the first process that gets moved to the head of the queue,
8147 * ultimately gets pushed toward the queue tail, and hence, jetsams last.
8148 *
8149 * So, for example, the coalition leader is expected to jetsam last,
8150 * after its coalition members. Therefore, the coalition leader is
8151 * inserted at the head of the queue first.
8152 *
8153 * After processing a coalition, the jetsam order is as follows:
8154 * undefs(jetsam first), extensions, xpc services, leader(jetsam last)
8155 */
8156
8157 /*
8158 * Coalition members are rearranged in the priority bucket here,
8159 * based on their coalition role.
8160 */
8161 total_pids_moved = 0;
8162 for (i=0; i < nleaders; i++) {
8163
8164 /* a bit of bookkeeping */
8165 pids_moved = 0;
8166
8167 /* Coalition leaders are jetsammed last, so move into place first */
8168 pid_list[0] = leaders[i].msi_pid;
8169 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list, 1);
8170
8171 /* xpc services should jetsam after extensions */
8172 ntasks = coalition_get_pid_list (leaders[i].msi_coal, COALITION_ROLEMASK_XPC,
8173 coal_sort_order, pid_list, MAX_SORT_PIDS);
8174
8175 if (ntasks > 0) {
8176 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
8177 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
8178 }
8179
8180 /* extensions should jetsam after unmarked processes */
8181 ntasks = coalition_get_pid_list (leaders[i].msi_coal, COALITION_ROLEMASK_EXT,
8182 coal_sort_order, pid_list, MAX_SORT_PIDS);
8183
8184 if (ntasks > 0) {
8185 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
8186 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
8187 }
8188
8189 /* undefined coalition members should be the first to jetsam */
8190 ntasks = coalition_get_pid_list (leaders[i].msi_coal, COALITION_ROLEMASK_UNDEF,
8191 coal_sort_order, pid_list, MAX_SORT_PIDS);
8192
8193 if (ntasks > 0) {
8194 pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
8195 (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
8196 }
8197
8198 #if 0
8199 if (pids_moved == leaders[i].msi_ntasks) {
8200 /*
8201 * All the pids in the coalition were found in this band.
8202 */
8203 printf("%s: pids_moved[%d] equal total coalition ntasks[%d] \n", __FUNCTION__,
8204 pids_moved, leaders[i].msi_ntasks);
8205 } else if (pids_moved > leaders[i].msi_ntasks) {
8206 /*
8207 * Apparently new coalition members showed up during the sort?
8208 */
8209 printf("%s: pids_moved[%d] were greater than expected coalition ntasks[%d] \n", __FUNCTION__,
8210 pids_moved, leaders[i].msi_ntasks);
8211 } else {
8212 /*
8213 * Apparently not all the pids in the coalition were found in this band?
8214 */
8215 printf("%s: pids_moved[%d] were less than expected coalition ntasks[%d] \n", __FUNCTION__,
8216 pids_moved, leaders[i].msi_ntasks);
8217 }
8218 #endif
8219
8220 total_pids_moved += pids_moved;
8221
8222 } /* end for */
8223
8224 return(total_pids_moved);
8225 }
8226
8227
8228 /*
8229 * Traverse a list of pids, searching for each within the priority band provided.
8230 * If pid is found, move it to the front of the priority band.
8231 * Never searches outside the priority band provided.
8232 *
8233 * Input:
8234 * bucket_index - jetsam priority band.
8235 * pid_list - pointer to a list of pids.
8236 * list_sz - number of pids in the list.
8237 *
8238 * Pid list ordering is important in that,
8239 * pid_list[n] is expected to jetsam ahead of pid_list[n+1].
8240 * The sort_order is set by the coalition default.
8241 *
8242 * Return:
8243 * the number of pids found and hence moved within the priority band.
8244 */
8245 static int
8246 memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz)
8247 {
8248 memstat_bucket_t *current_bucket;
8249 int i;
8250 int found_pids = 0;
8251
8252 if ((pid_list == NULL) || (list_sz <= 0)) {
8253 return(0);
8254 }
8255
8256 if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
8257 return(0);
8258 }
8259
8260 current_bucket = &memstat_bucket[bucket_index];
8261 for (i=0; i < list_sz; i++) {
8262 unsigned int b = bucket_index;
8263 proc_t p = NULL;
8264 proc_t aProc = NULL;
8265 pid_t aPid;
8266 int list_index;
8267
8268 list_index = ((list_sz - 1) - i);
8269 aPid = pid_list[list_index];
8270
8271 /* never search beyond bucket_index provided */
8272 p = memorystatus_get_first_proc_locked(&b, FALSE);
8273 while (p) {
8274 if (p->p_pid == aPid) {
8275 aProc = p;
8276 break;
8277 }
8278 p = memorystatus_get_next_proc_locked(&b, p, FALSE);
8279 }
8280
8281 if (aProc == NULL) {
8282 /* pid not found in this band, just skip it */
8283 continue;
8284 } else {
8285 TAILQ_REMOVE(&current_bucket->list, aProc, p_memstat_list);
8286 TAILQ_INSERT_HEAD(&current_bucket->list, aProc, p_memstat_list);
8287 found_pids++;
8288 }
8289 }
8290 return(found_pids);
8291 }
8292 #endif /* CONFIG_JETSAM */
mirror of XNU