#include <kern/task.h>
#include <kern/thread.h>
#include <kern/host.h>
+#include <kern/policy_internal.h>
+#include <kern/thread_group.h>
+
+#include <IOKit/IOBSD.h>
+
#include <libkern/libkern.h>
#include <mach/coalition.h>
#include <mach/mach_time.h>
#include <mach/task.h>
#include <mach/host_priv.h>
#include <mach/mach_host.h>
+#include <os/log.h>
#include <pexpert/pexpert.h>
#include <sys/coalition.h>
#include <sys/kern_event.h>
#include <sys/proc.h>
#include <sys/proc_info.h>
+#include <sys/reason.h>
#include <sys/signal.h>
#include <sys/signalvar.h>
#include <sys/sysctl.h>
#include <sys/kern_memorystatus.h>
-#if CONFIG_JETSAM
+#include <mach/machine/sdt.h>
+#include <libkern/section_keywords.h>
+
/* For logging clarity */
-static const char *jetsam_kill_cause_name[] = {
+static const char *memorystatus_kill_cause_name[] = {
"" ,
"jettisoned" , /* kMemorystatusKilled */
"highwater" , /* kMemorystatusKilledHiwat */
"per-process-limit" , /* kMemorystatusKilledPerProcessLimit */
"diagnostic" , /* kMemorystatusKilledDiagnostic */
"idle-exit" , /* kMemorystatusKilledIdleExit */
+ "zone-map-exhaustion" , /* kMemorystatusKilledZoneMapExhaustion */
};
+static const char *
+memorystatus_priority_band_name(int32_t priority)
+{
+ switch (priority) {
+ case JETSAM_PRIORITY_FOREGROUND:
+ return "FOREGROUND";
+ case JETSAM_PRIORITY_AUDIO_AND_ACCESSORY:
+ return "AUDIO_AND_ACCESSORY";
+ case JETSAM_PRIORITY_CONDUCTOR:
+ return "CONDUCTOR";
+ case JETSAM_PRIORITY_HOME:
+ return "HOME";
+ case JETSAM_PRIORITY_EXECUTIVE:
+ return "EXECUTIVE";
+ case JETSAM_PRIORITY_IMPORTANT:
+ return "IMPORTANT";
+ case JETSAM_PRIORITY_CRITICAL:
+ return "CRITICAL";
+ }
+
+ return ("?");
+}
+
/* Does cause indicate vm or fc thrashing? */
static boolean_t
-is_thrashing(unsigned cause)
+is_reason_thrashing(unsigned cause)
{
switch (cause) {
case kMemorystatusKilledVMThrashing:
}
}
-/* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
-extern void vm_thrashing_jetsam_done(void);
-#endif
+/* Is the zone map almost full? */
+static boolean_t
+is_reason_zone_map_exhaustion(unsigned cause)
+{
+ if (cause == kMemorystatusKilledZoneMapExhaustion)
+ return TRUE;
+ return FALSE;
+}
+
+/*
+ * Returns the current zone map size and capacity to include in the jetsam snapshot.
+ * Defined in zalloc.c
+ */
+extern void get_zone_map_size(uint64_t *current_size, uint64_t *capacity);
+
+/*
+ * Returns the name of the largest zone and its size to include in the jetsam snapshot.
+ * Defined in zalloc.c
+ */
+extern void get_largest_zone_info(char *zone_name, size_t zone_name_len, uint64_t *zone_size);
/* These are very verbose printfs(), enable with
* MEMORYSTATUS_DEBUG_LOG
#define SET_ACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
MACRO_BEGIN \
(p)->p_memstat_memlimit_active = (limit); \
- (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED; \
if (is_fatal) { \
(p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
} else { \
#define SET_INACTIVE_LIMITS_LOCKED(p, limit, is_fatal) \
MACRO_BEGIN \
(p)->p_memstat_memlimit_inactive = (limit); \
- (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED; \
if (is_fatal) { \
(p)->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
} else { \
} \
MACRO_END
-#define CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception) \
+#define CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal) \
MACRO_BEGIN \
(p)->p_memstat_memlimit = (p)->p_memstat_memlimit_active; \
if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) { \
(p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
+ is_fatal = TRUE; \
} else { \
(p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
- } \
- if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED) { \
- trigger_exception = FALSE; \
- } else { \
- trigger_exception = TRUE; \
+ is_fatal = FALSE; \
} \
MACRO_END
-#define CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception) \
+#define CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal) \
MACRO_BEGIN \
(p)->p_memstat_memlimit = (p)->p_memstat_memlimit_inactive; \
if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) { \
(p)->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT; \
+ is_fatal = TRUE; \
} else { \
(p)->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT; \
- } \
- if ((p)->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED) { \
- trigger_exception = FALSE; \
- } else { \
- trigger_exception = TRUE; \
+ is_fatal = FALSE; \
} \
MACRO_END
unsigned long idle_offset_percentage = 5;
unsigned long pressure_threshold_percentage = 15;
unsigned long freeze_threshold_percentage = 50;
+unsigned long policy_more_free_offset_percentage = 5;
/* General memorystatus stuff */
static void memorystatus_klist_lock(void);
static void memorystatus_klist_unlock(void);
-static uint64_t memorystatus_idle_delay_time = 0;
+static uint64_t memorystatus_sysprocs_idle_delay_time = 0;
+static uint64_t memorystatus_apps_idle_delay_time = 0;
/*
* Memorystatus kevents
*/
-static int filt_memorystatusattach(struct knote *kn);
+static int filt_memorystatusattach(struct knote *kn, struct kevent_internal_s *kev);
static void filt_memorystatusdetach(struct knote *kn);
static int filt_memorystatus(struct knote *kn, long hint);
+static int filt_memorystatustouch(struct knote *kn, struct kevent_internal_s *kev);
+static int filt_memorystatusprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev);
-struct filterops memorystatus_filtops = {
+SECURITY_READ_ONLY_EARLY(struct filterops) memorystatus_filtops = {
.f_attach = filt_memorystatusattach,
.f_detach = filt_memorystatusdetach,
.f_event = filt_memorystatus,
+ .f_touch = filt_memorystatustouch,
+ .f_process = filt_memorystatusprocess,
};
enum {
kMemorystatusNoPressure = 0x1,
kMemorystatusPressure = 0x2,
- kMemorystatusLowSwap = 0x4
+ kMemorystatusLowSwap = 0x4,
+ kMemorystatusProcLimitWarn = 0x8,
+ kMemorystatusProcLimitCritical = 0x10
};
/* Idle guard handling */
-static int32_t memorystatus_scheduled_idle_demotions = 0;
+static int32_t memorystatus_scheduled_idle_demotions_sysprocs = 0;
+static int32_t memorystatus_scheduled_idle_demotions_apps = 0;
static thread_call_t memorystatus_idle_demotion_call;
static void memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clean_state);
static void memorystatus_reschedule_idle_demotion_locked(void);
-static void memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert);
+static void memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check);
+
+int memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap);
+
+vm_pressure_level_t convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
boolean_t is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
+void memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear);
void memorystatus_send_low_swap_note(void);
int memorystatus_wakeup = 0;
unsigned int memorystatus_level = 0;
-unsigned int memorystatus_early_boot_level = 0;
static int memorystatus_list_count = 0;
memstat_bucket_t memstat_bucket[MEMSTAT_BUCKET_COUNT];
+int memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index);
+
uint64_t memstat_idle_demotion_deadline = 0;
+int system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+int applications_aging_band = JETSAM_PRIORITY_IDLE;
+
+#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)))
+#define isApp(p) (! (p->p_memstat_dirty & P_DIRTY_TRACK))
+#define isSysProc(p) ((p->p_memstat_dirty & P_DIRTY_TRACK))
+
+#define kJetsamAgingPolicyNone (0)
+#define kJetsamAgingPolicyLegacy (1)
+#define kJetsamAgingPolicySysProcsReclaimedFirst (2)
+#define kJetsamAgingPolicyAppsReclaimedFirst (3)
+#define kJetsamAgingPolicyMax kJetsamAgingPolicyAppsReclaimedFirst
+
+unsigned int jetsam_aging_policy = kJetsamAgingPolicyLegacy;
+
+extern int corpse_for_fatal_memkill;
+extern unsigned long total_corpses_count(void) __attribute__((pure));
+extern void task_purge_all_corpses(void);
+boolean_t memorystatus_allowed_vm_map_fork(__unused task_t);
+
+#if 0
+
+/* Keeping around for future use if we need a utility that can do this OR an app that needs a dynamic adjustment. */
+
+static int
+sysctl_set_jetsam_aging_policy SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp, arg1, arg2)
+
+ int error = 0, val = 0;
+ memstat_bucket_t *old_bucket = 0;
+ int old_system_procs_aging_band = 0, new_system_procs_aging_band = 0;
+ int old_applications_aging_band = 0, new_applications_aging_band = 0;
+ proc_t p = NULL, next_proc = NULL;
+
+
+ error = sysctl_io_number(req, jetsam_aging_policy, sizeof(int), &val, NULL);
+ if (error || !req->newptr) {
+ return (error);
+ }
+
+ if ((val < 0) || (val > kJetsamAgingPolicyMax)) {
+ printf("jetsam: ordering policy sysctl has invalid value - %d\n", val);
+ return EINVAL;
+ }
+
+ /*
+ * We need to synchronize with any potential adding/removal from aging bands
+ * that might be in progress currently. We use the proc_list_lock() just for
+ * consistency with all the routines dealing with 'aging' processes. We need
+ * a lighterweight lock.
+ */
+ proc_list_lock();
+
+ old_system_procs_aging_band = system_procs_aging_band;
+ old_applications_aging_band = applications_aging_band;
+
+ switch (val) {
+
+ case kJetsamAgingPolicyNone:
+ new_system_procs_aging_band = JETSAM_PRIORITY_IDLE;
+ new_applications_aging_band = JETSAM_PRIORITY_IDLE;
+ break;
+
+ case kJetsamAgingPolicyLegacy:
+ /*
+ * Legacy behavior where some daemons get a 10s protection once and only before the first clean->dirty->clean transition before going into IDLE band.
+ */
+ new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+ new_applications_aging_band = JETSAM_PRIORITY_IDLE;
+ break;
+
+ case kJetsamAgingPolicySysProcsReclaimedFirst:
+ new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+ new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
+ break;
+
+ case kJetsamAgingPolicyAppsReclaimedFirst:
+ new_system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
+ new_applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+ break;
+
+ default:
+ break;
+ }
+
+ if (old_system_procs_aging_band && (old_system_procs_aging_band != new_system_procs_aging_band)) {
+
+ old_bucket = &memstat_bucket[old_system_procs_aging_band];
+ p = TAILQ_FIRST(&old_bucket->list);
+
+ while (p) {
+
+ next_proc = TAILQ_NEXT(p, p_memstat_list);
+
+ if (isSysProc(p)) {
+ if (new_system_procs_aging_band == JETSAM_PRIORITY_IDLE) {
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ }
+
+ memorystatus_update_priority_locked(p, new_system_procs_aging_band, false, true);
+ }
+
+ p = next_proc;
+ continue;
+ }
+ }
+
+ if (old_applications_aging_band && (old_applications_aging_band != new_applications_aging_band)) {
+
+ old_bucket = &memstat_bucket[old_applications_aging_band];
+ p = TAILQ_FIRST(&old_bucket->list);
+
+ while (p) {
+
+ next_proc = TAILQ_NEXT(p, p_memstat_list);
+
+ if (isApp(p)) {
+ if (new_applications_aging_band == JETSAM_PRIORITY_IDLE) {
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ }
+
+ memorystatus_update_priority_locked(p, new_applications_aging_band, false, true);
+ }
+
+ p = next_proc;
+ continue;
+ }
+ }
+
+ jetsam_aging_policy = val;
+ system_procs_aging_band = new_system_procs_aging_band;
+ applications_aging_band = new_applications_aging_band;
+
+ proc_list_unlock();
+
+ return (0);
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, set_jetsam_aging_policy, CTLTYPE_INT|CTLFLAG_RW,
+ 0, 0, sysctl_set_jetsam_aging_policy, "I", "Jetsam Aging Policy");
+#endif /*0*/
+
+static int
+sysctl_jetsam_set_sysprocs_idle_delay_time SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp, arg1, arg2)
+
+ int error = 0, val = 0, old_time_in_secs = 0;
+ uint64_t old_time_in_ns = 0;
+
+ absolutetime_to_nanoseconds(memorystatus_sysprocs_idle_delay_time, &old_time_in_ns);
+ old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
+
+ error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
+ if (error || !req->newptr) {
+ return (error);
+ }
+
+ if ((val < 0) || (val > INT32_MAX)) {
+ printf("jetsam: new idle delay interval has invalid value.\n");
+ return EINVAL;
+ }
+
+ nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
+
+ return(0);
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_sysprocs_idle_delay_time, CTLTYPE_INT|CTLFLAG_RW,
+ 0, 0, sysctl_jetsam_set_sysprocs_idle_delay_time, "I", "Aging window for system processes");
+
+
+static int
+sysctl_jetsam_set_apps_idle_delay_time SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp, arg1, arg2)
+
+ int error = 0, val = 0, old_time_in_secs = 0;
+ uint64_t old_time_in_ns = 0;
+
+ absolutetime_to_nanoseconds(memorystatus_apps_idle_delay_time, &old_time_in_ns);
+ old_time_in_secs = old_time_in_ns / NSEC_PER_SEC;
+
+ error = sysctl_io_number(req, old_time_in_secs, sizeof(int), &val, NULL);
+ if (error || !req->newptr) {
+ return (error);
+ }
+
+ if ((val < 0) || (val > INT32_MAX)) {
+ printf("jetsam: new idle delay interval has invalid value.\n");
+ return EINVAL;
+ }
+
+ nanoseconds_to_absolutetime((uint64_t)val * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
+
+ return(0);
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_apps_idle_delay_time, CTLTYPE_INT|CTLFLAG_RW,
+ 0, 0, sysctl_jetsam_set_apps_idle_delay_time, "I", "Aging window for applications");
+
+SYSCTL_INT(_kern, OID_AUTO, jetsam_aging_policy, CTLTYPE_INT|CTLFLAG_RD, &jetsam_aging_policy, 0, "");
+
static unsigned int memorystatus_dirty_count = 0;
-#if CONFIG_JETSAM
SYSCTL_INT(_kern, OID_AUTO, max_task_pmem, CTLFLAG_RD|CTLFLAG_LOCKED|CTLFLAG_MASKED, &max_task_footprint_mb, 0, "");
-#endif // CONFIG_JETSAM
+#if CONFIG_EMBEDDED
+
+SYSCTL_INT(_kern, OID_AUTO, memorystatus_level, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_level, 0, "");
+
+#endif /* CONFIG_EMBEDDED */
int
memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
static void memorystatus_thread(void *param __unused, wait_result_t wr __unused);
-/* Jetsam */
+/* Memory Limits */
-#if CONFIG_JETSAM
+static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. */
+
+static boolean_t proc_jetsam_state_is_active_locked(proc_t);
+static boolean_t memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
+static boolean_t memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason);
-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);
static int memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
static int memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval);
-static boolean_t proc_jetsam_state_is_active_locked(proc_t);
+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);
int proc_get_memstat_priority(proc_t, boolean_t);
-/* Kill processes exceeding their limit either under memory pressure (1), or as soon as possible (0) */
-#define LEGACY_HIWATER 1
-
static boolean_t memorystatus_idle_snapshot = 0;
-static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. This should be removed. */
-
unsigned int memorystatus_delta = 0;
-static unsigned int memorystatus_available_pages_critical_base = 0;
-//static unsigned int memorystatus_last_foreground_pressure_pages = (unsigned int)-1;
-static unsigned int memorystatus_available_pages_critical_idle_offset = 0;
-
/* Jetsam Loop Detection */
-static boolean_t memorystatus_jld_enabled = TRUE; /* Enables jetsam loop detection on all devices */
+static boolean_t memorystatus_jld_enabled = FALSE; /* Enable jetsam loop detection */
static uint32_t memorystatus_jld_eval_period_msecs = 0; /* Init pass sets this based on device memory size */
static int memorystatus_jld_eval_aggressive_count = 3; /* Raise the priority max after 'n' aggressive loops */
static int memorystatus_jld_eval_aggressive_priority_band_max = 15; /* Kill aggressively up through this band */
+/*
+ * A FG app can request that the aggressive jetsam mechanism display some leniency in the FG band. This 'lenient' mode is described as:
+ * --- 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.
+ *
+ * RESTRICTIONS:
+ * - Such a request is respected/acknowledged only once while that 'requesting' app is in the FG band i.e. if aggressive jetsam was
+ * needed and the 'lenient' mode was deployed then that's it for this special mode while the app is in the FG band.
+ *
+ * - 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.
+ *
+ * - Also, the transition of the 'requesting' app away from the FG band will void this special behavior.
+ */
+
+#define AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD 25
+boolean_t memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
+boolean_t memorystatus_aggressive_jetsam_lenient = FALSE;
+
#if DEVELOPMENT || DEBUG
/*
* Jetsam Loop Detection tunables.
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_priority_band_max, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_priority_band_max, 0, "");
#endif /* DEVELOPMENT || DEBUG */
-#if DEVELOPMENT || DEBUG
-static unsigned int memorystatus_jetsam_panic_debug = 0;
-
-static unsigned int memorystatus_jetsam_policy = kPolicyDefault;
-static unsigned int memorystatus_jetsam_policy_offset_pages_diagnostic = 0;
-static unsigned int memorystatus_debug_dump_this_bucket = 0;
-#endif
-
-static unsigned int memorystatus_thread_wasted_wakeup = 0;
-
static uint32_t kill_under_pressure_cause = 0;
/*
*/
static memorystatus_jetsam_snapshot_t memorystatus_at_boot_snapshot;
+static void memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count);
+static boolean_t memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount);
+static void memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime);
+
static void memorystatus_clear_errors(void);
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);
+static void memorystatus_get_task_phys_footprint_page_counts(task_t task,
+ uint64_t *internal_pages, uint64_t *internal_compressed_pages,
+ uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
+ uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
+ uint64_t *iokit_mapped_pages, uint64_t *page_table_pages);
+
+static void memorystatus_get_task_memory_region_count(task_t task, uint64_t *count);
+
static uint32_t memorystatus_build_state(proc_t p);
-static void memorystatus_update_levels_locked(boolean_t critical_only);
//static boolean_t memorystatus_issue_pressure_kevent(boolean_t pressured);
-static boolean_t memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause);
-static boolean_t memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, int32_t *priority, uint32_t *errors);
-static boolean_t memorystatus_kill_top_process_aggressive(boolean_t any, uint32_t cause, int aggr_count, int32_t priority_max, uint32_t *errors);
-#if LEGACY_HIWATER
+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);
+static boolean_t memorystatus_kill_top_process_aggressive(uint32_t cause, int aggr_count, int32_t priority_max, uint32_t *errors);
+static boolean_t memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, int aggr_count, uint32_t *errors);
static boolean_t memorystatus_kill_hiwat_proc(uint32_t *errors);
-#endif
static boolean_t memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause);
-static boolean_t memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause);
/* Priority Band Sorting Routines */
static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order);
extern void qsort(void *a, size_t n, size_t es, cmpfunc_t cmp);
static int memstat_asc_cmp(const void *a, const void *b);
-#endif /* CONFIG_JETSAM */
-
/* VM pressure */
extern unsigned int vm_page_free_count;
extern unsigned int vm_page_throttled_count;
extern unsigned int vm_page_purgeable_count;
extern unsigned int vm_page_wire_count;
+#if CONFIG_SECLUDED_MEMORY
+extern unsigned int vm_page_secluded_count;
+#endif /* CONFIG_SECLUDED_MEMORY */
-#if VM_PRESSURE_EVENTS
+#if CONFIG_JETSAM
+unsigned int memorystatus_available_pages = (unsigned int)-1;
+unsigned int memorystatus_available_pages_pressure = 0;
+unsigned int memorystatus_available_pages_critical = 0;
+static unsigned int memorystatus_available_pages_critical_base = 0;
+static unsigned int memorystatus_available_pages_critical_idle_offset = 0;
-#include "vm_pressure.h"
+#if DEVELOPMENT || DEBUG
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
+#else
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD | CTLFLAG_MASKED | CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
+#endif /* DEVELOPMENT || DEBUG */
-extern boolean_t memorystatus_warn_process(pid_t pid, boolean_t critical);
+static unsigned int memorystatus_jetsam_policy = kPolicyDefault;
+unsigned int memorystatus_policy_more_free_offset_pages = 0;
+static void memorystatus_update_levels_locked(boolean_t critical_only);
+static unsigned int memorystatus_thread_wasted_wakeup = 0;
-vm_pressure_level_t memorystatus_vm_pressure_level = kVMPressureNormal;
+/* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
+extern void vm_thrashing_jetsam_done(void);
+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);
+
+int32_t max_kill_priority = JETSAM_PRIORITY_MAX;
+
+#else /* CONFIG_JETSAM */
+
+uint64_t memorystatus_available_pages = (uint64_t)-1;
+uint64_t memorystatus_available_pages_pressure = (uint64_t)-1;
+uint64_t memorystatus_available_pages_critical = (uint64_t)-1;
+
+int32_t max_kill_priority = JETSAM_PRIORITY_IDLE;
+#endif /* CONFIG_JETSAM */
-#if CONFIG_MEMORYSTATUS
-unsigned int memorystatus_available_pages = (unsigned int)-1;
-unsigned int memorystatus_available_pages_pressure = 0;
-unsigned int memorystatus_available_pages_critical = 0;
unsigned int memorystatus_frozen_count = 0;
unsigned int memorystatus_suspended_count = 0;
+#if VM_PRESSURE_EVENTS
+
+boolean_t memorystatus_warn_process(pid_t pid, __unused boolean_t is_active, __unused boolean_t is_fatal, boolean_t exceeded);
+
+vm_pressure_level_t memorystatus_vm_pressure_level = kVMPressureNormal;
+
/*
* We use this flag to signal if we have any HWM offenders
* on the system. This way we can reduce the number of wakeups
boolean_t memorystatus_hwm_candidates = 0;
static int memorystatus_send_note(int event_code, void *data, size_t data_length);
-#endif /* CONFIG_MEMORYSTATUS */
#endif /* VM_PRESSURE_EVENTS */
+
+#if DEVELOPMENT || DEBUG
+
+lck_grp_attr_t *disconnect_page_mappings_lck_grp_attr;
+lck_grp_t *disconnect_page_mappings_lck_grp;
+static lck_mtx_t disconnect_page_mappings_mutex;
+
+extern boolean_t kill_on_no_paging_space;
+#endif /* DEVELOPMENT || DEBUG */
+
+
/* Freeze */
#if CONFIG_FREEZE
static uint64_t memorystatus_freeze_throttle_count = 0;
-static unsigned int memorystatus_suspended_footprint_total = 0;
+static unsigned int memorystatus_suspended_footprint_total = 0; /* pages */
extern uint64_t vm_swap_get_free_space(void);
-static boolean_t memorystatus_freeze_update_throttle();
+static boolean_t memorystatus_freeze_update_throttle(void);
#endif /* CONFIG_FREEZE */
#if DEVELOPMENT || DEBUG
-#if CONFIG_JETSAM
+static unsigned int memorystatus_debug_dump_this_bucket = 0;
static void
memorystatus_debug_dump_bucket_locked (unsigned int bucket_index)
{
proc_t p = NULL;
- uint32_t pages = 0;
- uint32_t pages_in_mb = 0;
+ uint64_t bytes = 0;
+ int ledger_limit = 0;
unsigned int b = bucket_index;
boolean_t traverse_all_buckets = FALSE;
}
/*
- * Missing from this dump is the value actually
- * stored in the ledger... also, format could be better.
+ * footprint reported in [pages / MB ]
+ * limits reported as:
+ * L-limit proc's Ledger limit
+ * C-limit proc's Cached limit, should match Ledger
+ * A-limit proc's Active limit
+ * IA-limit proc's Inactive limit
+ * F==Fatal, NF==NonFatal
*/
- printf("memorystatus_debug_dump ***START***\n");
- printf("bucket [pid] [pages/pages-mb] state [EP / RP] dirty deadline [C-limit / A-limit / IA-limit] name\n");
+
+ printf("memorystatus_debug_dump ***START*(PAGE_SIZE_64=%llu)**\n", PAGE_SIZE_64);
+ printf("bucket [pid] [pages / MB] [state] [EP / RP] dirty deadline [L-limit / C-limit / A-limit / IA-limit] name\n");
p = memorystatus_get_first_proc_locked(&b, traverse_all_buckets);
while (p) {
- memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
- pages_in_mb = (pages * 4096) /1024 / 1024;
- printf("%d [%d] [%d/%dMB] 0x%x [%d / %d] 0x%x %lld [%d%s / %d%s / %d%s] %s\n",
- b, p->p_pid, pages, pages_in_mb,
+ bytes = get_task_phys_footprint(p->task);
+ task_get_phys_footprint_limit(p->task, &ledger_limit);
+ printf("%2d [%5d] [%5lld /%3lldMB] 0x%-8x [%2d / %2d] 0x%-3x %10lld [%3d / %3d%s / %3d%s / %3d%s] %s\n",
+ b, p->p_pid,
+ (bytes / PAGE_SIZE_64), /* task's footprint converted from bytes to pages */
+ (bytes / (1024ULL * 1024ULL)), /* task's footprint converted from bytes to MB */
p->p_memstat_state, p->p_memstat_effectivepriority, p->p_memstat_requestedpriority, p->p_memstat_dirty, p->p_memstat_idledeadline,
+ ledger_limit,
p->p_memstat_memlimit,
(p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"),
p->p_memstat_memlimit_active,
(p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL ? "F " : "NF"),
p->p_memstat_memlimit_inactive,
(p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL ? "F " : "NF"),
- (p->p_comm ? p->p_comm : "unknown"));
+ (*p->p_name ? p->p_name : "unknown"));
p = memorystatus_get_next_proc_locked(&b, p, traverse_all_buckets);
}
printf("memorystatus_debug_dump ***END***\n");
proc_t p;
unsigned int b = 0;
int error, enable = 0;
+ boolean_t use_active; /* use the active limit and active limit attributes */
+ boolean_t is_fatal;
error = SYSCTL_OUT(req, arg1, sizeof(int));
if (error || !req->newptr) {
p = memorystatus_get_first_proc_locked(&b, TRUE);
while (p) {
- boolean_t trigger_exception;
+ use_active = proc_jetsam_state_is_active_locked(p);
if (enable) {
- /*
- * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
- * Background limits are described via the inactive limit slots.
- */
- if (proc_jetsam_state_is_active_locked(p) == TRUE) {
- CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ if (use_active == TRUE) {
+ CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
} else {
- CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
}
} else {
*/
p->p_memstat_memlimit = -1;
p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
- trigger_exception = TRUE;
+ is_fatal = TRUE;
}
/*
* Enforce the cached limit by writing to the ledger.
*/
- task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit: -1, NULL, trigger_exception);
+ task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit: -1, NULL, use_active, is_fatal);
p = memorystatus_get_next_proc_locked(&b, p, TRUE);
}
}
-SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
-
SYSCTL_PROC(_kern, OID_AUTO, memorystatus_highwater_enabled, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_highwater_enabled, 0, sysctl_memorystatus_highwater_enable, "I", "");
-SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
-SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
-SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
-SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
-
-/* Diagnostic code */
-
-enum {
- kJetsamDiagnosticModeNone = 0,
- kJetsamDiagnosticModeAll = 1,
- kJetsamDiagnosticModeStopAtFirstActive = 2,
- kJetsamDiagnosticModeCount
-} jetsam_diagnostic_mode = kJetsamDiagnosticModeNone;
+#if VM_PRESSURE_EVENTS
-static int jetsam_diagnostic_suspended_one_active_proc = 0;
+/*
+ * This routine is used for targeted notifications regardless of system memory pressure
+ * and regardless of whether or not the process has already been notified.
+ * It bypasses and has no effect on the only-one-notification per soft-limit policy.
+ *
+ * "memnote" is the current user.
+ */
static int
-sysctl_jetsam_diagnostic_mode SYSCTL_HANDLER_ARGS
+sysctl_memorystatus_vm_pressure_send SYSCTL_HANDLER_ARGS
{
#pragma unused(arg1, arg2)
- const char *diagnosticStrings[] = {
- "jetsam: diagnostic mode: resetting critical level.",
- "jetsam: diagnostic mode: will examine all processes",
- "jetsam: diagnostic mode: will stop at first active process"
- };
-
- int error, val = jetsam_diagnostic_mode;
- boolean_t changed = FALSE;
-
- error = sysctl_handle_int(oidp, &val, 0, req);
+ int error = 0, pid = 0;
+ struct knote *kn = NULL;
+ boolean_t found_knote = FALSE;
+ int fflags = 0; /* filter flags for EVFILT_MEMORYSTATUS */
+ uint64_t value = 0;
+
+ error = sysctl_handle_quad(oidp, &value, 0, req);
+ if (error || !req->newptr)
+ return (error);
+
+ /*
+ * Find the pid in the low 32 bits of value passed in.
+ */
+ pid = (int)(value & 0xFFFFFFFF);
+
+ /*
+ * Find notification in the high 32 bits of the value passed in.
+ */
+ fflags = (int)((value >> 32) & 0xFFFFFFFF);
+
+ /*
+ * For backwards compatibility, when no notification is
+ * passed in, default to the NOTE_MEMORYSTATUS_PRESSURE_WARN
+ */
+ if (fflags == 0) {
+ fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ // printf("memorystatus_vm_pressure_send: using default notification [0x%x]\n", fflags);
+ }
+
+ /*
+ * See event.h ... fflags for EVFILT_MEMORYSTATUS
+ */
+ if (!((fflags == NOTE_MEMORYSTATUS_PRESSURE_NORMAL)||
+ (fflags == NOTE_MEMORYSTATUS_PRESSURE_WARN) ||
+ (fflags == NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) ||
+ (fflags == NOTE_MEMORYSTATUS_LOW_SWAP) ||
+ (fflags == NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) ||
+ (fflags == NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) ||
+ (((fflags & NOTE_MEMORYSTATUS_MSL_STATUS) != 0 &&
+ ((fflags & ~NOTE_MEMORYSTATUS_MSL_STATUS) == 0))))) {
+
+ printf("memorystatus_vm_pressure_send: notification [0x%x] not supported \n", fflags);
+ error = 1;
+ return (error);
+ }
+
+ /*
+ * Forcibly send pid a memorystatus notification.
+ */
+
+ memorystatus_klist_lock();
+
+ SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
+ proc_t knote_proc = knote_get_kq(kn)->kq_p;
+ pid_t knote_pid = knote_proc->p_pid;
+
+ if (knote_pid == pid) {
+ /*
+ * Forcibly send this pid a memorystatus notification.
+ */
+ kn->kn_fflags = fflags;
+ found_knote = TRUE;
+ }
+ }
+
+ if (found_knote) {
+ KNOTE(&memorystatus_klist, 0);
+ printf("memorystatus_vm_pressure_send: (value 0x%llx) notification [0x%x] sent to process [%d] \n", value, fflags, pid);
+ error = 0;
+ } else {
+ printf("memorystatus_vm_pressure_send: (value 0x%llx) notification [0x%x] not sent to process [%d] (none registered?)\n", value, fflags, pid);
+ error = 1;
+ }
+
+ memorystatus_klist_unlock();
+
+ return (error);
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_send, CTLTYPE_QUAD|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
+ 0, 0, &sysctl_memorystatus_vm_pressure_send, "Q", "");
+
+#endif /* VM_PRESSURE_EVENTS */
+
+SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
+
+#if CONFIG_JETSAM
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_policy_more_free_offset_pages, CTLFLAG_RW, &memorystatus_policy_more_free_offset_pages, 0, "");
+
+static unsigned int memorystatus_jetsam_panic_debug = 0;
+static unsigned int memorystatus_jetsam_policy_offset_pages_diagnostic = 0;
+
+/* Diagnostic code */
+
+enum {
+ kJetsamDiagnosticModeNone = 0,
+ kJetsamDiagnosticModeAll = 1,
+ kJetsamDiagnosticModeStopAtFirstActive = 2,
+ kJetsamDiagnosticModeCount
+} jetsam_diagnostic_mode = kJetsamDiagnosticModeNone;
+
+static int jetsam_diagnostic_suspended_one_active_proc = 0;
+
+static int
+sysctl_jetsam_diagnostic_mode SYSCTL_HANDLER_ARGS
+{
+#pragma unused(arg1, arg2)
+
+ const char *diagnosticStrings[] = {
+ "jetsam: diagnostic mode: resetting critical level.",
+ "jetsam: diagnostic mode: will examine all processes",
+ "jetsam: diagnostic mode: will stop at first active process"
+ };
+
+ int error, val = jetsam_diagnostic_mode;
+ boolean_t changed = FALSE;
+
+ error = sysctl_handle_int(oidp, &val, 0, req);
if (error || !req->newptr)
return (error);
if ((val < 0) || (val >= kJetsamDiagnosticModeCount)) {
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_pressure, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_pressure, 0, "");
-
-/*
- * This routine is used for targeted notifications
- * regardless of system memory pressure.
- * "memnote" is the current user.
- */
-
-static int
-sysctl_memorystatus_vm_pressure_send SYSCTL_HANDLER_ARGS
-{
-#pragma unused(arg1, arg2)
-
- int error = 0, pid = 0;
- int ret = 0;
- struct knote *kn = NULL;
- boolean_t found_knote = FALSE;
-
- error = sysctl_handle_int(oidp, &pid, 0, req);
- if (error || !req->newptr)
- return (error);
-
- /*
- * We inspect 3 lists here for targeted notifications:
- * - memorystatus_klist
- * - vm_pressure_klist
- * - vm_pressure_dormant_klist
- *
- * The vm_pressure_* lists are tied to the old VM_PRESSURE
- * notification mechanism. We intend to stop using that
- * mechanism and, in turn, get rid of the 2 lists and
- * vm_dispatch_pressure_note_to_pid() too.
- */
-
- memorystatus_klist_lock();
-
- SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
- proc_t knote_proc = kn->kn_kq->kq_p;
- pid_t knote_pid = knote_proc->p_pid;
-
- if (knote_pid == pid) {
- /*
- * Forcibly send this pid a "warning" memory pressure notification.
- */
- kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
- found_knote = TRUE;
- }
- }
-
- if (found_knote) {
- KNOTE(&memorystatus_klist, 0);
- ret = 0;
- } else {
- ret = vm_dispatch_pressure_note_to_pid(pid, FALSE);
- }
-
- memorystatus_klist_unlock();
-
- return ret;
-}
-
-SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_send, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
- 0, 0, &sysctl_memorystatus_vm_pressure_send, "I", "");
-
#endif /* VM_PRESSURE_EVENTS */
#endif /* CONFIG_JETSAM */
boolean_t memorystatus_freeze_throttle_enabled = TRUE;
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_throttle_enabled, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_throttle_enabled, 0, "");
+#define VM_PAGES_FOR_ALL_PROCS (2)
/*
* Manual trigger of freeze and thaw for dev / debug kernels only.
*/
if (error || !req->newptr)
return (error);
- if (pid == 2) {
+ if (pid == VM_PAGES_FOR_ALL_PROCS) {
vm_pageout_anonymous_pages();
return 0;
boolean_t shared;
uint32_t max_pages = 0;
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
unsigned int avail_swap_space = 0; /* in pages. */
- if (DEFAULT_FREEZER_IS_ACTIVE) {
- /*
- * Freezer backed by default pager and swap file(s).
- */
- avail_swap_space = default_pager_swap_pages_free();
- } else {
- /*
- * Freezer backed by the compressor and swap file(s)
- * while will hold compressed data.
- */
- avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
- }
+ /*
+ * Freezer backed by the compressor and swap file(s)
+ * while will hold compressed data.
+ */
+ avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
if (error || !req->newptr)
return (error);
- p = proc_find(pid);
- if (p != NULL) {
- error = task_thaw(p->task);
- proc_rele(p);
-
- if (error)
- error = EIO;
- return error;
+ if (pid == VM_PAGES_FOR_ALL_PROCS) {
+ do_fastwake_warmup_all();
+ return 0;
+ } else {
+ p = proc_find(pid);
+ if (p != NULL) {
+ error = task_thaw(p->task);
+ proc_rele(p);
+
+ if (error)
+ error = EIO;
+ return error;
+ }
}
return EINVAL;
integer_t priority,
thread_t *new_thread);
-#if CONFIG_JETSAM
+#if DEVELOPMENT || DEBUG
+
+static int
+sysctl_memorystatus_disconnect_page_mappings SYSCTL_HANDLER_ARGS
+{
+#pragma unused(arg1, arg2)
+ int error = 0, pid = 0;
+ proc_t p;
+
+ error = sysctl_handle_int(oidp, &pid, 0, req);
+ if (error || !req->newptr)
+ return (error);
+
+ lck_mtx_lock(&disconnect_page_mappings_mutex);
+
+ if (pid == -1) {
+ vm_pageout_disconnect_all_pages();
+ } else {
+ p = proc_find(pid);
+
+ if (p != NULL) {
+ error = task_disconnect_page_mappings(p->task);
+
+ proc_rele(p);
+
+ if (error)
+ error = EIO;
+ } else
+ error = EINVAL;
+ }
+ lck_mtx_unlock(&disconnect_page_mappings_mutex);
+
+ return error;
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_disconnect_page_mappings, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
+ 0, 0, &sysctl_memorystatus_disconnect_page_mappings, "I", "");
+
+#endif /* DEVELOPMENT || DEBUG */
+
+
/*
* Picks the sorting routine for a given jetsam priority band.
*
#endif
proc_list_lock();
+
+ if (memstat_bucket[bucket_index].count == 0) {
+ proc_list_unlock();
+ return (0);
+ }
+
switch (bucket_index) {
case JETSAM_PRIORITY_FOREGROUND:
if (memorystatus_sort_by_largest_coalition_locked(bucket_index, coal_sort_order) == 0) {
}
}
-#endif /* CONFIG_JETSAM */
-
static proc_t memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search) {
memstat_bucket_t *current_bucket;
proc_t next_p;
memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
#endif
- nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_idle_delay_time);
+#if DEVELOPMENT || DEBUG
+ disconnect_page_mappings_lck_grp_attr = lck_grp_attr_alloc_init();
+ disconnect_page_mappings_lck_grp = lck_grp_alloc_init("disconnect_page_mappings", disconnect_page_mappings_lck_grp_attr);
+
+ lck_mtx_init(&disconnect_page_mappings_mutex, disconnect_page_mappings_lck_grp, NULL);
+
+ if (kill_on_no_paging_space == TRUE) {
+ max_kill_priority = JETSAM_PRIORITY_MAX;
+ }
+#endif
+
/* Init buckets */
for (i = 0; i < MEMSTAT_BUCKET_COUNT; i++) {
TAILQ_INIT(&memstat_bucket[i].list);
memstat_bucket[i].count = 0;
}
-
memorystatus_idle_demotion_call = thread_call_allocate((thread_call_func_t)memorystatus_perform_idle_demotion, NULL);
+#if CONFIG_JETSAM
+ nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_sysprocs_idle_delay_time);
+ nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_apps_idle_delay_time);
+
/* Apply overrides */
PE_get_default("kern.jetsam_delta", &delta_percentage, sizeof(delta_percentage));
+ if (delta_percentage == 0) {
+ delta_percentage = 5;
+ }
assert(delta_percentage < 100);
PE_get_default("kern.jetsam_critical_threshold", &critical_threshold_percentage, sizeof(critical_threshold_percentage));
assert(critical_threshold_percentage < 100);
PE_get_default("kern.jetsam_freeze_threshold", &freeze_threshold_percentage, sizeof(freeze_threshold_percentage));
assert(freeze_threshold_percentage < 100);
-#if CONFIG_JETSAM
- /* device tree can request to take snapshots for idle-exit kills by default */
- PE_get_default("kern.jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot));
+ if (!PE_parse_boot_argn("jetsam_aging_policy", &jetsam_aging_policy,
+ sizeof (jetsam_aging_policy))) {
+
+ if (!PE_get_default("kern.jetsam_aging_policy", &jetsam_aging_policy,
+ sizeof(jetsam_aging_policy))) {
+
+ jetsam_aging_policy = kJetsamAgingPolicyLegacy;
+ }
+ }
+
+ if (jetsam_aging_policy > kJetsamAgingPolicyMax) {
+ jetsam_aging_policy = kJetsamAgingPolicyLegacy;
+ }
+
+ switch (jetsam_aging_policy) {
+
+ case kJetsamAgingPolicyNone:
+ system_procs_aging_band = JETSAM_PRIORITY_IDLE;
+ applications_aging_band = JETSAM_PRIORITY_IDLE;
+ break;
+
+ case kJetsamAgingPolicyLegacy:
+ /*
+ * Legacy behavior where some daemons get a 10s protection once
+ * AND only before the first clean->dirty->clean transition before
+ * going into IDLE band.
+ */
+ system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+ applications_aging_band = JETSAM_PRIORITY_IDLE;
+ break;
+
+ case kJetsamAgingPolicySysProcsReclaimedFirst:
+ system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+ applications_aging_band = JETSAM_PRIORITY_AGING_BAND2;
+ break;
+
+ case kJetsamAgingPolicyAppsReclaimedFirst:
+ system_procs_aging_band = JETSAM_PRIORITY_AGING_BAND2;
+ applications_aging_band = JETSAM_PRIORITY_AGING_BAND1;
+ break;
+
+ default:
+ break;
+ }
+
+ /*
+ * The aging bands cannot overlap with the JETSAM_PRIORITY_ELEVATED_INACTIVE
+ * band and must be below it in priority. This is so that we don't have to make
+ * our 'aging' code worry about a mix of processes, some of which need to age
+ * and some others that need to stay elevated in the jetsam bands.
+ */
+ assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > system_procs_aging_band);
+ assert(JETSAM_PRIORITY_ELEVATED_INACTIVE > applications_aging_band);
+
+ /* Take snapshots for idle-exit kills by default? First check the boot-arg... */
+ if (!PE_parse_boot_argn("jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof (memorystatus_idle_snapshot))) {
+ /* ...no boot-arg, so check the device tree */
+ PE_get_default("kern.jetsam_idle_snapshot", &memorystatus_idle_snapshot, sizeof(memorystatus_idle_snapshot));
+ }
memorystatus_delta = delta_percentage * atop_64(max_mem) / 100;
memorystatus_available_pages_critical_idle_offset = idle_offset_percentage * atop_64(max_mem) / 100;
memorystatus_available_pages_critical_base = (critical_threshold_percentage / delta_percentage) * memorystatus_delta;
+ memorystatus_policy_more_free_offset_pages = (policy_more_free_offset_percentage / delta_percentage) * memorystatus_delta;
+ /* Jetsam Loop Detection */
+ if (max_mem <= (512 * 1024 * 1024)) {
+ /* 512 MB devices */
+ memorystatus_jld_eval_period_msecs = 8000; /* 8000 msecs == 8 second window */
+ } else {
+ /* 1GB and larger devices */
+ memorystatus_jld_eval_period_msecs = 6000; /* 6000 msecs == 6 second window */
+ }
+
+ memorystatus_jld_enabled = TRUE;
+
+ /* No contention at this point */
+ memorystatus_update_levels_locked(FALSE);
+
+#endif /* CONFIG_JETSAM */
+
memorystatus_jetsam_snapshot_max = maxproc;
memorystatus_jetsam_snapshot =
(memorystatus_jetsam_snapshot_t*)kalloc(sizeof(memorystatus_jetsam_snapshot_t) +
memset(&memorystatus_at_boot_snapshot, 0, sizeof(memorystatus_jetsam_snapshot_t));
- /* No contention at this point */
- memorystatus_update_levels_locked(FALSE);
-
- /* Jetsam Loop Detection */
- if (max_mem <= (512 * 1024 * 1024)) {
- /* 512 MB devices */
- memorystatus_jld_eval_period_msecs = 8000; /* 8000 msecs == 8 second window */
- } else {
- /* 1GB and larger devices */
- memorystatus_jld_eval_period_msecs = 6000; /* 6000 msecs == 6 second window */
- }
-#endif
-
#if CONFIG_FREEZE
memorystatus_freeze_threshold = (freeze_threshold_percentage / delta_percentage) * memorystatus_delta;
#endif
/* Centralised for the purposes of allowing panic-on-jetsam */
extern void
-vm_wake_compactor_swapper(void);
+vm_run_compactor(void);
/*
* The jetsam no frills kill call
* error code on failure (EINVAL...)
*/
static int
-jetsam_do_kill(proc_t p, int jetsam_flags) {
+jetsam_do_kill(proc_t p, int jetsam_flags, os_reason_t jetsam_reason) {
int error = 0;
- error = exit1_internal(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags);
+ error = exit_with_reason(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags, jetsam_reason);
return(error);
}
* Wrapper for processes exiting with memorystatus details
*/
static boolean_t
-memorystatus_do_kill(proc_t p, uint32_t cause) {
+memorystatus_do_kill(proc_t p, uint32_t cause, os_reason_t jetsam_reason) {
int error = 0;
__unused pid_t victim_pid = p->p_pid;
KERNEL_DEBUG_CONSTANT( (BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_START,
victim_pid, cause, vm_page_free_count, 0, 0);
+ DTRACE_MEMORYSTATUS3(memorystatus_do_kill, proc_t, p, os_reason_t, jetsam_reason, uint32_t, cause);
#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if (memorystatus_jetsam_panic_debug & (1 << cause)) {
panic("memorystatus_do_kill(): jetsam debug panic (cause: %d)", cause);
#else
#pragma unused(cause)
#endif
+
+ if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
+ printf("memorystatus: killing process %d [%s] in high band %s (%d) - memorystatus_available_pages: %llu\n", p->p_pid,
+ (*p->p_name ? p->p_name : "unknown"),
+ memorystatus_priority_band_name(p->p_memstat_effectivepriority), p->p_memstat_effectivepriority,
+ (uint64_t)memorystatus_available_pages);
+ }
+
int jetsam_flags = P_LTERM_JETSAM;
switch (cause) {
case kMemorystatusKilledHiwat: jetsam_flags |= P_JETSAM_HIWAT; break;
case kMemorystatusKilledPerProcessLimit: jetsam_flags |= P_JETSAM_PID; break;
case kMemorystatusKilledIdleExit: jetsam_flags |= P_JETSAM_IDLEEXIT; break;
}
- error = jetsam_do_kill(p, jetsam_flags);
+ error = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
KERNEL_DEBUG_CONSTANT( (BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_END,
victim_pid, cause, vm_page_free_count, error, 0);
- vm_wake_compactor_swapper();
+ vm_run_compactor();
return (error == 0);
}
#if CONFIG_JETSAM
/* Update levels */
memorystatus_update_levels_locked(TRUE);
-#endif
+#else /* CONFIG_JETSAM */
+ /*
+ * Nothing to do here currently since we update
+ * memorystatus_available_pages in vm_pressure_response.
+ */
+#endif /* CONFIG_JETSAM */
+}
+
+/*
+ * Pin a process to a particular jetsam band when it is in the background i.e. not doing active work.
+ * For an application: that means no longer in the FG band
+ * For a daemon: that means no longer in its 'requested' jetsam priority band
+ */
+
+int
+memorystatus_update_inactive_jetsam_priority_band(pid_t pid, uint32_t op_flags, boolean_t effective_now)
+{
+ int error = 0;
+ boolean_t enable = FALSE;
+ proc_t p = NULL;
+
+ if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE) {
+ enable = TRUE;
+ } else if (op_flags == MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE) {
+ enable = FALSE;
+ } else {
+ return EINVAL;
+ }
+
+ p = proc_find(pid);
+ if (p != NULL) {
+
+ if ((enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) ||
+ (!enable && ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) == 0))) {
+ /*
+ * No change in state.
+ */
+
+ } else {
+
+ proc_list_lock();
+
+ if (enable) {
+ p->p_memstat_state |= P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+
+ if (effective_now) {
+ if (p->p_memstat_effectivepriority < JETSAM_PRIORITY_ELEVATED_INACTIVE) {
+ if(memorystatus_highwater_enabled) {
+ /*
+ * Process is about to transition from
+ * inactive --> active
+ * assign active state
+ */
+ boolean_t is_fatal;
+ boolean_t use_active = TRUE;
+ CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
+ task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
+ }
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_ELEVATED_INACTIVE, FALSE, FALSE);
+ }
+ } else {
+ if (isProcessInAgingBands(p)) {
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
+ }
+ }
+ } else {
+
+ p->p_memstat_state &= ~P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND;
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+
+ if (effective_now) {
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_ELEVATED_INACTIVE) {
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
+ }
+ } else {
+ if (isProcessInAgingBands(p)) {
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
+ }
+ }
+ }
+
+ proc_list_unlock();
+ }
+ proc_rele(p);
+ error = 0;
+
+ } else {
+ error = ESRCH;
+ }
+
+ return error;
}
static void
memorystatus_perform_idle_demotion(__unused void *spare1, __unused void *spare2)
{
proc_t p;
- uint64_t current_time;
+ uint64_t current_time = 0, idle_delay_time = 0;
+ int demote_prio_band = 0;
memstat_bucket_t *demotion_bucket;
MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion()\n");
proc_list_lock();
- demotion_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE_DEFERRED];
- p = TAILQ_FIRST(&demotion_bucket->list);
-
- while (p) {
- MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion() found %d\n", p->p_pid);
-
- assert(p->p_memstat_idledeadline);
- assert(p->p_memstat_dirty & P_DIRTY_DEFER_IN_PROGRESS);
- assert((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED);
-
- if (current_time >= p->p_memstat_idledeadline) {
-#if DEBUG || DEVELOPMENT
- if (!(p->p_memstat_dirty & P_DIRTY_MARKED)) {
- printf("memorystatus_perform_idle_demotion: moving process %d [%s] to idle band, but never dirtied (0x%x)!\n",
- p->p_pid, (p->p_comm ? p->p_comm : "(unknown)"), p->p_memstat_dirty);
- }
-#endif
- memorystatus_invalidate_idle_demotion_locked(p, TRUE);
- memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false);
-
- // The prior process has moved out of the demotion bucket, so grab the new head and continue
- p = TAILQ_FIRST(&demotion_bucket->list);
+ demote_prio_band = JETSAM_PRIORITY_IDLE + 1;
+
+ for (; demote_prio_band < JETSAM_PRIORITY_MAX; demote_prio_band++) {
+
+ if (demote_prio_band != system_procs_aging_band && demote_prio_band != applications_aging_band)
continue;
+
+ demotion_bucket = &memstat_bucket[demote_prio_band];
+ p = TAILQ_FIRST(&demotion_bucket->list);
+
+ while (p) {
+ MEMORYSTATUS_DEBUG(1, "memorystatus_perform_idle_demotion() found %d\n", p->p_pid);
+
+ assert(p->p_memstat_idledeadline);
+
+ assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
+
+ if (current_time >= p->p_memstat_idledeadline) {
+
+ if ((isSysProc(p) &&
+ ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_IS_DIRTY)) != P_DIRTY_IDLE_EXIT_ENABLED)) || /* system proc marked dirty*/
+ task_has_assertions((struct task *)(p->task))) { /* has outstanding assertions which might indicate outstanding work too */
+ idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_delay_time : memorystatus_apps_idle_delay_time;
+
+ p->p_memstat_idledeadline += idle_delay_time;
+ p = TAILQ_NEXT(p, p_memstat_list);
+
+ } else {
+
+ proc_t next_proc = NULL;
+
+ next_proc = TAILQ_NEXT(p, p_memstat_list);
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false, true);
+
+ p = next_proc;
+ continue;
+
+ }
+ } else {
+ // No further candidates
+ break;
+ }
}
-
- // No further candidates
- break;
+
}
-
+
memorystatus_reschedule_idle_demotion_locked();
proc_list_unlock();
static void
memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state)
{
- boolean_t present_in_deferred_bucket = FALSE;
-
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- present_in_deferred_bucket = TRUE;
+ boolean_t present_in_sysprocs_aging_bucket = FALSE;
+ boolean_t present_in_apps_aging_bucket = FALSE;
+ uint64_t idle_delay_time = 0;
+
+ if (jetsam_aging_policy == kJetsamAgingPolicyNone) {
+ return;
+ }
+
+ if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
+ /*
+ * This process isn't going to be making the trip to the lower bands.
+ */
+ return;
+ }
+
+ if (isProcessInAgingBands(p)){
+
+ if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
+ assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) != P_DIRTY_AGING_IN_PROGRESS);
+ }
+
+ if (isSysProc(p) && system_procs_aging_band) {
+ present_in_sysprocs_aging_bucket = TRUE;
+
+ } else if (isApp(p) && applications_aging_band) {
+ present_in_apps_aging_bucket = TRUE;
+ }
}
+ assert(!present_in_sysprocs_aging_bucket);
+ assert(!present_in_apps_aging_bucket);
+
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",
- p->p_pid, p->p_memstat_dirty, set_state, memorystatus_scheduled_idle_demotions);
+ p->p_pid, p->p_memstat_dirty, set_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
+
+ if(isSysProc(p)) {
+ assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
+ }
- assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
+ idle_delay_time = (isSysProc(p)) ? memorystatus_sysprocs_idle_delay_time : memorystatus_apps_idle_delay_time;
if (set_state) {
- assert(p->p_memstat_idledeadline == 0);
- p->p_memstat_dirty |= P_DIRTY_DEFER_IN_PROGRESS;
- p->p_memstat_idledeadline = mach_absolute_time() + memorystatus_idle_delay_time;
+ p->p_memstat_dirty |= P_DIRTY_AGING_IN_PROGRESS;
+ p->p_memstat_idledeadline = mach_absolute_time() + idle_delay_time;
}
assert(p->p_memstat_idledeadline);
- if (present_in_deferred_bucket == FALSE) {
- memorystatus_scheduled_idle_demotions++;
+ if (isSysProc(p) && present_in_sysprocs_aging_bucket == FALSE) {
+ memorystatus_scheduled_idle_demotions_sysprocs++;
+
+ } else if (isApp(p) && present_in_apps_aging_bucket == FALSE) {
+ memorystatus_scheduled_idle_demotions_apps++;
}
}
static void
memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clear_state)
{
- boolean_t present_in_deferred_bucket = FALSE;
-
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- present_in_deferred_bucket = TRUE;
- assert(p->p_memstat_idledeadline);
+ boolean_t present_in_sysprocs_aging_bucket = FALSE;
+ boolean_t present_in_apps_aging_bucket = FALSE;
+
+ if (!system_procs_aging_band && !applications_aging_band) {
+ return;
+ }
+
+ if ((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == 0) {
+ return;
+ }
+
+ if (isProcessInAgingBands(p)) {
+
+ if (jetsam_aging_policy != kJetsamAgingPolicyLegacy) {
+ assert((p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) == P_DIRTY_AGING_IN_PROGRESS);
+ }
+
+ if (isSysProc(p) && system_procs_aging_band) {
+ assert(p->p_memstat_effectivepriority == system_procs_aging_band);
+ assert(p->p_memstat_idledeadline);
+ present_in_sysprocs_aging_bucket = TRUE;
+
+ } else if (isApp(p) && applications_aging_band) {
+ assert(p->p_memstat_effectivepriority == applications_aging_band);
+ assert(p->p_memstat_idledeadline);
+ present_in_apps_aging_bucket = TRUE;
+ }
}
MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for pid %d (clear_state %d, demotions %d).\n",
- p->p_pid, clear_state, memorystatus_scheduled_idle_demotions);
+ p->p_pid, clear_state, (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps));
if (clear_state) {
p->p_memstat_idledeadline = 0;
- p->p_memstat_dirty &= ~P_DIRTY_DEFER_IN_PROGRESS;
+ p->p_memstat_dirty &= ~P_DIRTY_AGING_IN_PROGRESS;
}
- if (present_in_deferred_bucket == TRUE) {
- memorystatus_scheduled_idle_demotions--;
+ if (isSysProc(p) &&present_in_sysprocs_aging_bucket == TRUE) {
+ memorystatus_scheduled_idle_demotions_sysprocs--;
+ assert(memorystatus_scheduled_idle_demotions_sysprocs >= 0);
+
+ } else if (isApp(p) && present_in_apps_aging_bucket == TRUE) {
+ memorystatus_scheduled_idle_demotions_apps--;
+ assert(memorystatus_scheduled_idle_demotions_apps >= 0);
}
- assert(memorystatus_scheduled_idle_demotions >= 0);
+ assert((memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps) >= 0);
}
static void
memorystatus_reschedule_idle_demotion_locked(void) {
- if (0 == memorystatus_scheduled_idle_demotions) {
+ if (0 == (memorystatus_scheduled_idle_demotions_sysprocs + memorystatus_scheduled_idle_demotions_apps)) {
if (memstat_idle_demotion_deadline) {
/* Transitioned 1->0, so cancel next call */
thread_call_cancel(memorystatus_idle_demotion_call);
}
} else {
memstat_bucket_t *demotion_bucket;
- proc_t p;
- demotion_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE_DEFERRED];
- p = TAILQ_FIRST(&demotion_bucket->list);
-
- assert(p && p->p_memstat_idledeadline);
-
- if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline){
- thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
- memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
+ proc_t p = NULL, p1 = NULL, p2 = NULL;
+
+ if (system_procs_aging_band) {
+
+ demotion_bucket = &memstat_bucket[system_procs_aging_band];
+ p1 = TAILQ_FIRST(&demotion_bucket->list);
+
+ p = p1;
+ }
+
+ if (applications_aging_band) {
+
+ demotion_bucket = &memstat_bucket[applications_aging_band];
+ p2 = TAILQ_FIRST(&demotion_bucket->list);
+
+ if (p1 && p2) {
+ p = (p1->p_memstat_idledeadline > p2->p_memstat_idledeadline) ? p2 : p1;
+ } else {
+ p = (p1 == NULL) ? p2 : p1;
+ }
+
+ }
+
+ assert(p);
+
+ if (p != NULL) {
+ assert(p && p->p_memstat_idledeadline);
+ if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline){
+ thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
+ memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
+ }
}
}
}
memstat_bucket_t *bucket;
MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding pid %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
-
+
if (!locked) {
proc_list_lock();
}
-
+
+ DTRACE_MEMORYSTATUS2(memorystatus_add, proc_t, p, int32_t, p->p_memstat_effectivepriority);
+
/* Processes marked internal do not have priority tracked */
if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
goto exit;
bucket = &memstat_bucket[p->p_memstat_effectivepriority];
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- assert(bucket->count == memorystatus_scheduled_idle_demotions);
+ if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
+ assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs - 1);
+
+ } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
+ assert(bucket->count == memorystatus_scheduled_idle_demotions_apps - 1);
+
+ } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
+ /*
+ * Entering the idle band.
+ * Record idle start time.
+ */
+ p->p_memstat_idle_start = mach_absolute_time();
}
TAILQ_INSERT_TAIL(&bucket->list, p, p_memstat_list);
*
* Monitors transition between buckets and if necessary
* will update cached memory limits accordingly.
+ *
+ * skip_demotion_check:
+ * - if the 'jetsam aging policy' is NOT 'legacy':
+ * When this flag is TRUE, it means we are going
+ * to age the ripe processes out of the aging bands and into the
+ * IDLE band and apply their inactive memory limits.
+ *
+ * - if the 'jetsam aging policy' is 'legacy':
+ * When this flag is TRUE, it might mean the above aging mechanism
+ * OR
+ * It might be that we have a process that has used up its 'idle deferral'
+ * stay that is given to it once per lifetime. And in this case, the process
+ * won't be going through any aging codepaths. But we still need to apply
+ * the right inactive limits and so we explicitly set this to TRUE if the
+ * new priority for the process is the IDLE band.
*/
-static void
-memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert)
+void
+memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert, boolean_t skip_demotion_check)
{
memstat_bucket_t *old_bucket, *new_bucket;
if ((p->p_listflag & P_LIST_EXITED) != 0) {
return;
}
-
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting pid %d to priority %d, inserting at %s\n",
- p->p_pid, priority, head_insert ? "head" : "tail");
+
+ MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting %s(%d) to priority %d, inserting at %s\n",
+ (*p->p_name ? p->p_name : "unknown"), p->p_pid, priority, head_insert ? "head" : "tail");
+
+ DTRACE_MEMORYSTATUS3(memorystatus_update_priority, proc_t, p, int32_t, p->p_memstat_effectivepriority, int, priority);
+
+#if DEVELOPMENT || DEBUG
+ if (priority == JETSAM_PRIORITY_IDLE && /* if the process is on its way into the IDLE band */
+ skip_demotion_check == FALSE && /* and it isn't via the path that will set the INACTIVE memlimits */
+ (p->p_memstat_dirty & P_DIRTY_TRACK) && /* and it has 'DIRTY' tracking enabled */
+ ((p->p_memstat_memlimit != p->p_memstat_memlimit_inactive) || /* and we notice that the current limit isn't the right value (inactive) */
+ ((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) */
+ 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 */
+#endif /* DEVELOPMENT || DEBUG */
old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- assert(old_bucket->count == (memorystatus_scheduled_idle_demotions + 1));
+
+ if (skip_demotion_check == FALSE) {
+
+ if (isSysProc(p)) {
+ /*
+ * For system processes, the memorystatus_dirty_* routines take care of adding/removing
+ * the processes from the aging bands and balancing the demotion counts.
+ * We can, however, override that if the process has an 'elevated inactive jetsam band' attribute.
+ */
+
+ if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE && (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
+ priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
+
+ assert(! (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
+ }
+ } else if (isApp(p)) {
+
+ /*
+ * Check to see if the application is being lowered in jetsam priority. If so, and:
+ * - it has an 'elevated inactive jetsam band' attribute, then put it in the JETSAM_PRIORITY_ELEVATED_INACTIVE band.
+ * - it is a normal application, then let it age in the aging band if that policy is in effect.
+ */
+
+ if (priority <= JETSAM_PRIORITY_ELEVATED_INACTIVE && (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
+ priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
+ } else {
+
+ if (applications_aging_band) {
+ if (p->p_memstat_effectivepriority == applications_aging_band) {
+ assert(old_bucket->count == (memorystatus_scheduled_idle_demotions_apps + 1));
+ }
+
+ if ((jetsam_aging_policy != kJetsamAgingPolicyLegacy) && (priority <= applications_aging_band)) {
+ assert(! (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS));
+ priority = applications_aging_band;
+ memorystatus_schedule_idle_demotion_locked(p, TRUE);
+ }
+ }
+ }
+ }
+ }
+
+ if ((system_procs_aging_band && (priority == system_procs_aging_band)) || (applications_aging_band && (priority == applications_aging_band))) {
+ assert(p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS);
}
TAILQ_REMOVE(&old_bucket->list, p, p_memstat_list);
old_bucket->count--;
-
+
new_bucket = &memstat_bucket[priority];
if (head_insert)
TAILQ_INSERT_HEAD(&new_bucket->list, p, p_memstat_list);
TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
new_bucket->count++;
-#if CONFIG_JETSAM
if (memorystatus_highwater_enabled) {
- boolean_t trigger_exception;
+ boolean_t is_fatal;
+ boolean_t use_active;
/*
* If cached limit data is updated, then the limits
boolean_t ledger_update_needed = TRUE;
/*
- * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
- * Background limits are described via the inactive limit slots.
- *
* Here, we must update the cached memory limit if the task
* is transitioning between:
* active <--> inactive
* but:
* dirty <--> clean is ignored
*
- * We bypass processes that have opted into dirty tracking because
+ * We bypass non-idle processes that have opted into dirty tracking because
* a move between buckets does not imply a transition between the
* dirty <--> clean state.
- * Setting limits on processes opted into dirty tracking is handled
- * in memorystatus_dirty_set() where the transition is very clear.
*/
if (p->p_memstat_dirty & P_DIRTY_TRACK) {
- ledger_update_needed = FALSE;
+ if (skip_demotion_check == TRUE && priority == JETSAM_PRIORITY_IDLE) {
+ CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = FALSE;
+ } else {
+ ledger_update_needed = FALSE;
+ }
} else if ((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) {
/*
* BG --> FG
* assign active state
*/
- CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = TRUE;
} else if ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
/*
* FG --> BG
* assign inactive state
*/
- CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = FALSE;
} else {
/*
* The transition between jetsam priority buckets apparently did
* Enforce the new limits by writing to the ledger
*/
if (ledger_update_needed) {
- task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, trigger_exception);
+ task_set_phys_footprint_limit_internal(p->task, (p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1, NULL, use_active, is_fatal);
MEMORYSTATUS_DEBUG(3, "memorystatus_update_priority_locked: new limit on pid %d (%dMB %s) priority old --> new (%d --> %d) dirty?=0x%x %s\n",
p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
}
}
-#endif /* CONFIG_JETSAM */
-
+ /*
+ * Record idle start or idle delta.
+ */
+ if (p->p_memstat_effectivepriority == priority) {
+ /*
+ * This process is not transitioning between
+ * jetsam priority buckets. Do nothing.
+ */
+ } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
+ uint64_t now;
+ /*
+ * Transitioning out of the idle priority bucket.
+ * Record idle delta.
+ */
+ assert(p->p_memstat_idle_start != 0);
+ now = mach_absolute_time();
+ if (now > p->p_memstat_idle_start) {
+ p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
+ }
+ } else if (priority == JETSAM_PRIORITY_IDLE) {
+ /*
+ * Transitioning into the idle priority bucket.
+ * Record idle start.
+ */
+ p->p_memstat_idle_start = mach_absolute_time();
+ }
+
p->p_memstat_effectivepriority = priority;
+
+#if CONFIG_SECLUDED_MEMORY
+ if (secluded_for_apps &&
+ task_could_use_secluded_mem(p->task)) {
+ task_set_can_use_secluded_mem(
+ p->task,
+ (priority >= JETSAM_PRIORITY_FOREGROUND));
+ }
+#endif /* CONFIG_SECLUDED_MEMORY */
memorystatus_check_levels_locked();
}
* memlimit_inactive_is_fatal When a process is inactive and exceeds its memory footprint,
* this describes whether or not it should be immediatly fatal.
*
- * memlimit_background This process has a high-water-mark while in the background.
- * No longer meaningful. Background limits are described via
- * the inactive slots. Flag is ignored.
- *
- *
* Returns: 0 Success
* non-0 Failure
*/
int
memorystatus_update(proc_t p, int priority, uint64_t user_data, boolean_t effective, boolean_t update_memlimit,
int32_t memlimit_active, boolean_t memlimit_active_is_fatal,
- int32_t memlimit_inactive, boolean_t memlimit_inactive_is_fatal,
- __unused boolean_t memlimit_background)
+ int32_t memlimit_inactive, boolean_t memlimit_inactive_is_fatal)
{
int ret;
boolean_t head_insert = false;
-
-#if !CONFIG_JETSAM
-#pragma unused(update_memlimit, memlimit_active, memlimit_inactive)
-#pragma unused(memlimit_active_is_fatal, memlimit_inactive_is_fatal)
-#endif /* !CONFIG_JETSAM */
- MEMORYSTATUS_DEBUG(1, "memorystatus_update: changing pid %d: priority %d, user_data 0x%llx\n", p->p_pid, priority, user_data);
+ 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);
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_UPDATE) | DBG_FUNC_START, p->p_pid, priority, user_data, effective, 0);
if (priority == -1) {
/* Use as shorthand for default priority */
priority = JETSAM_PRIORITY_DEFAULT;
- } else if (priority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- /* JETSAM_PRIORITY_IDLE_DEFERRED is reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
+ } else if ((priority == system_procs_aging_band) || (priority == applications_aging_band)) {
+ /* Both the aging bands are reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
priority = JETSAM_PRIORITY_IDLE;
} else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
/* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
p->p_memstat_state |= P_MEMSTAT_PRIORITYUPDATED;
p->p_memstat_userdata = user_data;
p->p_memstat_requestedpriority = priority;
-
-#if CONFIG_JETSAM
+
if (update_memlimit) {
- boolean_t trigger_exception;
+ boolean_t is_fatal;
+ boolean_t use_active;
/*
* Posix_spawn'd processes come through this path to instantiate ledger limits.
memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
- if (memlimit_background) {
-
- /*
- * With 2-level HWM support, we no longer honor P_MEMSTAT_MEMLIMIT_BACKGROUND.
- * Background limits are described via the inactive limit slots.
- */
-
- // p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_BACKGROUND;
-
-#if DEVELOPMENT || DEBUG
- printf("memorystatus_update: WARNING %s[%d] set unused flag P_MEMSTAT_MEMLIMIT_BACKGROUND [A==%dMB %s] [IA==%dMB %s]\n",
- (p->p_comm ? p->p_comm : "unknown"), p->p_pid,
- memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
- memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
-#endif /* DEVELOPMENT || DEBUG */
- }
-
if (memlimit_active <= 0) {
/*
* This process will have a system_wide task limit when active.
*/
if (proc_jetsam_state_is_active_locked(p) == TRUE) {
- CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = TRUE;
} else {
- CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = FALSE;
}
/*
*/
if (memorystatus_highwater_enabled) {
/* apply now */
- assert(trigger_exception == TRUE);
- task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, trigger_exception);
+ task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, use_active, is_fatal);
MEMORYSTATUS_DEBUG(3, "memorystatus_update: init: limit on pid %d (%dMB %s) targeting priority(%d) dirty?=0x%x %s\n",
p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
(p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
}
}
-#endif /* CONFIG_JETSAM */
/*
- * We can't add to the JETSAM_PRIORITY_IDLE_DEFERRED bucket here.
- * But, we could be removing it from the bucket.
+ * We can't add to the aging bands buckets here.
+ * But, we could be removing it from those buckets.
* Check and take appropriate steps if so.
*/
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ if (isProcessInAgingBands(p)) {
memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, FALSE, TRUE);
+ } else {
+ if (jetsam_aging_policy == kJetsamAgingPolicyLegacy && priority == JETSAM_PRIORITY_IDLE) {
+ /*
+ * Daemons with 'inactive' limits will go through the dirty tracking codepath.
+ * This path deals with apps that may have 'inactive' limits e.g. WebContent processes.
+ * If this is the legacy aging policy we explicitly need to apply those limits. If it
+ * is any other aging policy, then we don't need to worry because all processes
+ * will go through the aging bands and then the demotion thread will take care to
+ * move them into the IDLE band and apply the required limits.
+ */
+ memorystatus_update_priority_locked(p, priority, head_insert, TRUE);
+ }
}
-
- memorystatus_update_priority_locked(p, priority, head_insert);
-
+
+ memorystatus_update_priority_locked(p, priority, head_insert, FALSE);
+
proc_list_unlock();
ret = 0;
{
int ret;
memstat_bucket_t *bucket;
+ boolean_t reschedule = FALSE;
MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing pid %d\n", p->p_pid);
assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
bucket = &memstat_bucket[p->p_memstat_effectivepriority];
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- assert(bucket->count == memorystatus_scheduled_idle_demotions);
+
+ if (isSysProc(p) && system_procs_aging_band && (p->p_memstat_effectivepriority == system_procs_aging_band)) {
+
+ assert(bucket->count == memorystatus_scheduled_idle_demotions_sysprocs);
+ reschedule = TRUE;
+
+ } else if (isApp(p) && applications_aging_band && (p->p_memstat_effectivepriority == applications_aging_band)) {
+
+ assert(bucket->count == memorystatus_scheduled_idle_demotions_apps);
+ reschedule = TRUE;
+ }
+
+ /*
+ * Record idle delta
+ */
+
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
+ uint64_t now = mach_absolute_time();
+ if (now > p->p_memstat_idle_start) {
+ p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
+ }
}
TAILQ_REMOVE(&bucket->list, p, p_memstat_list);
memorystatus_list_count--;
/* If awaiting demotion to the idle band, clean up */
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ if (reschedule) {
memorystatus_invalidate_idle_demotion_locked(p, TRUE);
memorystatus_reschedule_idle_demotion_locked();
}
* Return:
* 0 on success
* non-0 on failure
+ *
+ * The proc_list_lock is held by the caller.
*/
static int
int32_t priority;
MEMORYSTATUS_DEBUG(1, "memorystatus_update_idle_priority_locked(): pid %d dirty 0x%X\n", p->p_pid, p->p_memstat_dirty);
-
+
+ assert(isSysProc(p));
+
if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
- priority = (p->p_memstat_dirty & P_DIRTY_DEFER_IN_PROGRESS) ? JETSAM_PRIORITY_IDLE_DEFERRED : JETSAM_PRIORITY_IDLE;
+
+ priority = (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) ? system_procs_aging_band : JETSAM_PRIORITY_IDLE;
} else {
priority = p->p_memstat_requestedpriority;
}
if (priority != p->p_memstat_effectivepriority) {
- memorystatus_update_priority_locked(p, priority, false);
+
+ if ((jetsam_aging_policy == kJetsamAgingPolicyLegacy) &&
+ (priority == JETSAM_PRIORITY_IDLE)) {
+
+ /*
+ * This process is on its way into the IDLE band. The system is
+ * using 'legacy' jetsam aging policy. That means, this process
+ * has already used up its idle-deferral aging time that is given
+ * once per its lifetime. So we need to set the INACTIVE limits
+ * explicitly because it won't be going through the demotion paths
+ * that take care to apply the limits appropriately.
+ */
+
+ if (p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) {
+
+ /*
+ * This process has the 'elevated inactive jetsam band' attribute.
+ * So, there will be no trip to IDLE after all.
+ * Instead, we pin the process in the elevated band,
+ * where its ACTIVE limits will apply.
+ */
+
+ priority = JETSAM_PRIORITY_ELEVATED_INACTIVE;
+ }
+
+ memorystatus_update_priority_locked(p, priority, false, true);
+
+ } else {
+ memorystatus_update_priority_locked(p, priority, false, false);
+ }
}
}
* priority idle band when clean (and killed earlier, protecting higher priority procesess).
*
* If the deferral flag is set, then newly tracked processes will be protected for an initial period (as determined by
- * memorystatus_idle_delay_time); if they go clean during this time, then they will be moved to a deferred-idle band
+ * memorystatus_sysprocs_idle_delay_time); if they go clean during this time, then they will be moved to a deferred-idle band
* with a slightly higher priority, guarding against immediate termination under memory pressure and being unable to
* make forward progress. Finally, when the guard expires, they will be moved to the standard, lowest-priority, idle
* band. The deferral can be cleared early by clearing the appropriate flag.
p->p_memstat_dirty |= P_DIRTY_LAUNCH_IN_PROGRESS;
}
- if (old_dirty & P_DIRTY_DEFER_IN_PROGRESS) {
+ if (old_dirty & P_DIRTY_AGING_IN_PROGRESS) {
already_deferred = TRUE;
}
+
/* This can be set and cleared exactly once. */
if (pcontrol & PROC_DIRTY_DEFER) {
/* Kick off or invalidate the idle exit deferment if there's a state transition. */
if (!(p->p_memstat_dirty & P_DIRTY_IS_DIRTY)) {
- if (((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) &&
- defer_now && !already_deferred) {
-
- /*
- * Request to defer a clean process that's idle-exit enabled
- * and not already in the jetsam deferred band.
- */
- memorystatus_schedule_idle_demotion_locked(p, TRUE);
- reschedule = TRUE;
+ if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
- } else if (!defer_now && already_deferred) {
+ if (defer_now && !already_deferred) {
+
+ /*
+ * Request to defer a clean process that's idle-exit enabled
+ * and not already in the jetsam deferred band. Most likely a
+ * new launch.
+ */
+ memorystatus_schedule_idle_demotion_locked(p, TRUE);
+ reschedule = TRUE;
- /*
- * Either the process is no longer idle-exit enabled OR
- * there's a request to cancel a currently active deferral.
- */
- memorystatus_invalidate_idle_demotion_locked(p, TRUE);
- reschedule = TRUE;
+ } else if (!defer_now) {
+
+ /*
+ * The process isn't asking for the 'aging' facility.
+ * Could be that it is:
+ */
+
+ if (already_deferred) {
+ /*
+ * already in the aging bands. Traditionally,
+ * some processes have tried to use this to
+ * opt out of the 'aging' facility.
+ */
+
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ } else {
+ /*
+ * agnostic to the 'aging' facility. In that case,
+ * we'll go ahead and opt it in because this is likely
+ * a new launch (clean process, dirty tracking enabled)
+ */
+
+ memorystatus_schedule_idle_demotion_locked(p, TRUE);
+ }
+
+ reschedule = TRUE;
+ }
}
} else {
* deferred state or not?
*
* This could be a legal request like:
- * - this process had opted into the JETSAM_DEFERRED band
+ * - this process had opted into the 'aging' band
* - but it's now dirty and requests to opt out.
* In this case, we remove the process from the band and reset its
* state too. It'll opt back in properly when needed.
*
* OR, this request could be a user-space bug. E.g.:
- * - this process had opted into the JETSAM_DEFERRED band when clean
+ * - this process had opted into the 'aging' band when clean
* - and, then issues another request to again put it into the band except
* this time the process is dirty.
* The process going dirty, as a transition in memorystatus_dirty_set(), will pull the process out of
* But we do it here anyways for coverage.
*
* memorystatus_update_idle_priority_locked()
- * single-mindedly treats a dirty process as "cannot be in the deferred band".
+ * single-mindedly treats a dirty process as "cannot be in the aging band".
*/
if (!defer_now && already_deferred) {
memorystatus_invalidate_idle_demotion_locked(p, TRUE);
reschedule = TRUE;
} else {
- memorystatus_invalidate_idle_demotion_locked(p, FALSE);
+
+ boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
+
+ memorystatus_invalidate_idle_demotion_locked(p, reset_state);
reschedule = TRUE;
}
}
boolean_t now_dirty = FALSE;
MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_set(): %d %d 0x%x 0x%x\n", self, p->p_pid, pcontrol, p->p_memstat_dirty);
-
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_SET), p->p_pid, self, pcontrol, 0, 0);
proc_list_lock();
(was_dirty == FALSE && now_dirty == TRUE)) {
/* Manage idle exit deferral, if applied */
- if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_DEFER_IN_PROGRESS)) ==
- (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_DEFER_IN_PROGRESS)) {
+ if ((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) {
/*
- * P_DIRTY_DEFER_IN_PROGRESS means the process is in the deferred band OR it might be heading back
- * there once it's clean again and has some protection window left.
+ * Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band OR it might be heading back
+ * there once it's clean again. For the legacy case, this only applies if it has some protection window left.
+ *
+ * Non-Legacy mode: P_DIRTY_AGING_IN_PROGRESS means the process is in the aging band. It will always stop over
+ * in that band on it's way to IDLE.
*/
if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
/*
* New dirty process i.e. "was_dirty == FALSE && now_dirty == TRUE"
*
- * The process will move from the deferred band to its higher requested
- * jetsam band. But we don't clear its state i.e. we want to remember that
- * this process was part of the "deferred" band and will return to it.
- *
- * This way, we don't let it age beyond the protection
- * window when it returns to "clean". All the while giving
- * it a chance to perform its work while "dirty".
- *
+ * The process will move from its aging band to its higher requested
+ * jetsam band.
*/
- memorystatus_invalidate_idle_demotion_locked(p, FALSE);
+ boolean_t reset_state = (jetsam_aging_policy != kJetsamAgingPolicyLegacy) ? TRUE : FALSE;
+
+ memorystatus_invalidate_idle_demotion_locked(p, reset_state);
reschedule = TRUE;
} else {
/*
* Process is back from "dirty" to "clean".
- *
- * Is its timer up OR does it still have some protection
- * window left?
*/
- if (mach_absolute_time() >= p->p_memstat_idledeadline) {
- /*
- * The process' deadline has expired. It currently
- * does not reside in the DEFERRED bucket.
- *
- * It's on its way to the JETSAM_PRIORITY_IDLE
- * bucket via memorystatus_update_idle_priority_locked()
- * below.
-
- * So all we need to do is reset all the state on the
- * process that's related to the DEFERRED bucket i.e.
- * the DIRTY_DEFER_IN_PROGRESS flag and the timer deadline.
- *
- */
+ if (jetsam_aging_policy == kJetsamAgingPolicyLegacy) {
+ if (mach_absolute_time() >= p->p_memstat_idledeadline) {
+ /*
+ * The process' deadline has expired. It currently
+ * does not reside in any of the aging buckets.
+ *
+ * It's on its way to the JETSAM_PRIORITY_IDLE
+ * bucket via memorystatus_update_idle_priority_locked()
+ * below.
+
+ * So all we need to do is reset all the state on the
+ * process that's related to the aging bucket i.e.
+ * the AGING_IN_PROGRESS flag and the timer deadline.
+ */
- memorystatus_invalidate_idle_demotion_locked(p, TRUE);
- reschedule = TRUE;
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ reschedule = TRUE;
+ } else {
+ /*
+ * It still has some protection window left and so
+ * we just re-arm the timer without modifying any
+ * state on the process iff it still wants into that band.
+ */
+
+ if (p->p_memstat_dirty & P_DIRTY_AGING_IN_PROGRESS) {
+ memorystatus_schedule_idle_demotion_locked(p, FALSE);
+ reschedule = TRUE;
+ }
+ }
} else {
- /*
- * It still has some protection window left and so
- * we just re-arm the timer without modifying any
- * state on the process.
- */
- memorystatus_schedule_idle_demotion_locked(p, FALSE);
+
+ memorystatus_schedule_idle_demotion_locked(p, TRUE);
reschedule = TRUE;
}
}
memorystatus_update_idle_priority_locked(p);
-#if CONFIG_JETSAM
if (memorystatus_highwater_enabled) {
- boolean_t trigger_exception;
+ boolean_t ledger_update_needed = TRUE;
+ boolean_t use_active;
+ boolean_t is_fatal;
/*
* We are in this path because this process transitioned between
* dirty <--> clean state. Update the cached memory limits.
if (proc_jetsam_state_is_active_locked(p) == TRUE) {
/*
+ * process is pinned in elevated band
+ * or
* process is dirty
*/
- CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = TRUE;
+ ledger_update_needed = TRUE;
} else {
/*
- * process is clean
+ * process is clean...but if it has opted into pressured-exit
+ * we don't apply the INACTIVE limit till the process has aged
+ * out and is entering the IDLE band.
+ * See memorystatus_update_priority_locked() for that.
*/
- CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+
+ if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
+ ledger_update_needed = FALSE;
+ } else {
+ CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = FALSE;
+ ledger_update_needed = TRUE;
+ }
}
/*
* See rdar://21394491.
*/
- if (proc_ref_locked(p) == p) {
+ if (ledger_update_needed && proc_ref_locked(p) == p) {
int ledger_limit;
if (p->p_memstat_memlimit > 0) {
ledger_limit = p->p_memstat_memlimit;
ledger_limit = -1;
}
proc_list_unlock();
- task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, trigger_exception);
+ task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, use_active, is_fatal);
proc_list_lock();
proc_rele_locked(p);
}
}
-#endif /* CONFIG_JETSAM */
/* If the deferral state changed, reschedule the demotion timer */
if (reschedule) {
#endif
}
+/*
+ * The proc_list_lock is held by the caller.
+*/
static uint32_t
memorystatus_build_state(proc_t p) {
uint32_t snapshot_state = 0;
return snapshot_state;
}
-#if !CONFIG_JETSAM
-
static boolean_t
kill_idle_exit_proc(void)
{
uint64_t current_time;
boolean_t killed = FALSE;
unsigned int i = 0;
+ os_reason_t jetsam_reason = OS_REASON_NULL;
/* Pick next idle exit victim. */
current_time = mach_absolute_time();
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_IDLE_EXIT);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("kill_idle_exit_proc: failed to allocate jetsam reason\n");
+ }
+
proc_list_lock();
p = memorystatus_get_first_proc_locked(&i, FALSE);
proc_list_unlock();
if (victim_p) {
- printf("memorystatus_thread: idle exiting pid %d [%s]\n", victim_p->p_pid, (victim_p->p_comm ? victim_p->p_comm : "(unknown)"));
- killed = memorystatus_do_kill(victim_p, kMemorystatusKilledIdleExit);
+ printf("memorystatus: killing_idle_process pid %d [%s]\n", victim_p->p_pid, (*victim_p->p_name ? victim_p->p_name : "unknown"));
+ killed = memorystatus_do_kill(victim_p, kMemorystatusKilledIdleExit, jetsam_reason);
proc_rele(victim_p);
+ } else {
+ os_reason_free(jetsam_reason);
}
return killed;
}
-#endif
-#if CONFIG_JETSAM
static void
memorystatus_thread_wake(void) {
thread_wakeup((event_t)&memorystatus_wakeup);
}
-#endif /* CONFIG_JETSAM */
extern void vm_pressure_response(void);
return thread_block(continuation);
}
+static boolean_t
+memorystatus_avail_pages_below_pressure(void)
+{
+#if CONFIG_EMBEDDED
+/*
+ * Instead of CONFIG_EMBEDDED for these *avail_pages* routines, we should
+ * key off of the system having dynamic swap support. With full swap support,
+ * the system shouldn't really need to worry about various page thresholds.
+ */
+ return (memorystatus_available_pages <= memorystatus_available_pages_pressure);
+#else /* CONFIG_EMBEDDED */
+ return FALSE;
+#endif /* CONFIG_EMBEDDED */
+}
+
+static boolean_t
+memorystatus_avail_pages_below_critical(void)
+{
+#if CONFIG_EMBEDDED
+ return (memorystatus_available_pages <= memorystatus_available_pages_critical);
+#else /* CONFIG_EMBEDDED */
+ return FALSE;
+#endif /* CONFIG_EMBEDDED */
+}
+
+static boolean_t
+memorystatus_post_snapshot(int32_t priority, uint32_t cause)
+{
+#if CONFIG_EMBEDDED
+#pragma unused(cause)
+ /*
+ * Don't generate logs for steady-state idle-exit kills,
+ * unless it is overridden for debug or by the device
+ * tree.
+ */
+
+ return ((priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot);
+
+#else /* CONFIG_EMBEDDED */
+ /*
+ * Don't generate logs for steady-state idle-exit kills,
+ * unless
+ * - it is overridden for debug or by the device
+ * tree.
+ * OR
+ * - the kill causes are important i.e. not kMemorystatusKilledIdleExit
+ */
+
+ boolean_t snapshot_eligible_kill_cause = (is_reason_thrashing(cause) || is_reason_zone_map_exhaustion(cause));
+ return ((priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot || snapshot_eligible_kill_cause);
+#endif /* CONFIG_EMBEDDED */
+}
+
+static boolean_t
+memorystatus_action_needed(void)
+{
+#if CONFIG_EMBEDDED
+ return (is_reason_thrashing(kill_under_pressure_cause) ||
+ is_reason_zone_map_exhaustion(kill_under_pressure_cause) ||
+ memorystatus_available_pages <= memorystatus_available_pages_pressure);
+#else /* CONFIG_EMBEDDED */
+ return (is_reason_thrashing(kill_under_pressure_cause) ||
+ is_reason_zone_map_exhaustion(kill_under_pressure_cause));
+#endif /* CONFIG_EMBEDDED */
+}
+
+static boolean_t
+memorystatus_act_on_hiwat_processes(uint32_t *errors, uint32_t *hwm_kill, boolean_t *post_snapshot, __unused boolean_t *is_critical)
+{
+ boolean_t killed = memorystatus_kill_hiwat_proc(errors);
+
+ if (killed) {
+ *hwm_kill = *hwm_kill + 1;
+ *post_snapshot = TRUE;
+ return TRUE;
+ } else {
+ memorystatus_hwm_candidates = FALSE;
+ }
+
+#if CONFIG_JETSAM
+ /* No highwater processes to kill. Continue or stop for now? */
+ if (!is_reason_thrashing(kill_under_pressure_cause) &&
+ !is_reason_zone_map_exhaustion(kill_under_pressure_cause) &&
+ (memorystatus_available_pages > memorystatus_available_pages_critical)) {
+ /*
+ * We are _not_ out of pressure but we are above the critical threshold and there's:
+ * - no compressor thrashing
+ * - enough zone memory
+ * - no more HWM processes left.
+ * For now, don't kill any other processes.
+ */
+
+ if (*hwm_kill == 0) {
+ memorystatus_thread_wasted_wakeup++;
+ }
+
+ *is_critical = FALSE;
+
+ return TRUE;
+ }
+#endif /* CONFIG_JETSAM */
+
+ return FALSE;
+}
+
+static boolean_t
+memorystatus_act_aggressive(uint32_t cause, os_reason_t jetsam_reason, int *jld_idle_kills, boolean_t *corpse_list_purged, boolean_t *post_snapshot)
+{
+ if (memorystatus_jld_enabled == TRUE) {
+
+ boolean_t killed;
+ uint32_t errors = 0;
+
+ /* Jetsam Loop Detection - locals */
+ memstat_bucket_t *bucket;
+ int jld_bucket_count = 0;
+ struct timeval jld_now_tstamp = {0,0};
+ uint64_t jld_now_msecs = 0;
+ int elevated_bucket_count = 0;
+
+ /* Jetsam Loop Detection - statics */
+ static uint64_t jld_timestamp_msecs = 0;
+ static int jld_idle_kill_candidates = 0; /* Number of available processes in band 0,1 at start */
+ static int jld_eval_aggressive_count = 0; /* Bumps the max priority in aggressive loop */
+ static int32_t jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
+ /*
+ * Jetsam Loop Detection: attempt to detect
+ * rapid daemon relaunches in the lower bands.
+ */
+
+ microuptime(&jld_now_tstamp);
+
+ /*
+ * Ignore usecs in this calculation.
+ * msecs granularity is close enough.
+ */
+ jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
+
+ proc_list_lock();
+ switch (jetsam_aging_policy) {
+ case kJetsamAgingPolicyLegacy:
+ bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
+ jld_bucket_count = bucket->count;
+ bucket = &memstat_bucket[JETSAM_PRIORITY_AGING_BAND1];
+ jld_bucket_count += bucket->count;
+ break;
+ case kJetsamAgingPolicySysProcsReclaimedFirst:
+ case kJetsamAgingPolicyAppsReclaimedFirst:
+ bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
+ jld_bucket_count = bucket->count;
+ bucket = &memstat_bucket[system_procs_aging_band];
+ jld_bucket_count += bucket->count;
+ bucket = &memstat_bucket[applications_aging_band];
+ jld_bucket_count += bucket->count;
+ break;
+ case kJetsamAgingPolicyNone:
+ default:
+ bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
+ jld_bucket_count = bucket->count;
+ break;
+ }
+
+ bucket = &memstat_bucket[JETSAM_PRIORITY_ELEVATED_INACTIVE];
+ elevated_bucket_count = bucket->count;
+
+ proc_list_unlock();
+
+ /*
+ * memorystatus_jld_eval_period_msecs is a tunable
+ * memorystatus_jld_eval_aggressive_count is a tunable
+ * memorystatus_jld_eval_aggressive_priority_band_max is a tunable
+ */
+ if ( (jld_bucket_count == 0) ||
+ (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
+
+ /*
+ * Refresh evaluation parameters
+ */
+ jld_timestamp_msecs = jld_now_msecs;
+ jld_idle_kill_candidates = jld_bucket_count;
+ *jld_idle_kills = 0;
+ jld_eval_aggressive_count = 0;
+ jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
+ }
+
+ if (*jld_idle_kills > jld_idle_kill_candidates) {
+ jld_eval_aggressive_count++;
+
+#if DEVELOPMENT || DEBUG
+ printf("memorystatus: aggressive%d: beginning of window: %lld ms, : timestamp now: %lld ms\n",
+ jld_eval_aggressive_count,
+ jld_timestamp_msecs,
+ jld_now_msecs);
+ printf("memorystatus: aggressive%d: idle candidates: %d, idle kills: %d\n",
+ jld_eval_aggressive_count,
+ jld_idle_kill_candidates,
+ *jld_idle_kills);
+#endif /* DEVELOPMENT || DEBUG */
+
+ if ((jld_eval_aggressive_count == memorystatus_jld_eval_aggressive_count) &&
+ (total_corpses_count() > 0) && (*corpse_list_purged == FALSE)) {
+ /*
+ * If we reach this aggressive cycle, corpses might be causing memory pressure.
+ * So, in an effort to avoid jetsams in the FG band, we will attempt to purge
+ * corpse memory prior to this final march through JETSAM_PRIORITY_UI_SUPPORT.
+ */
+ task_purge_all_corpses();
+ *corpse_list_purged = TRUE;
+ }
+ else if (jld_eval_aggressive_count > memorystatus_jld_eval_aggressive_count) {
+ /*
+ * Bump up the jetsam priority limit (eg: the bucket index)
+ * Enforce bucket index sanity.
+ */
+ if ((memorystatus_jld_eval_aggressive_priority_band_max < 0) ||
+ (memorystatus_jld_eval_aggressive_priority_band_max >= MEMSTAT_BUCKET_COUNT)) {
+ /*
+ * Do nothing. Stick with the default level.
+ */
+ } else {
+ jld_priority_band_max = memorystatus_jld_eval_aggressive_priority_band_max;
+ }
+ }
+
+ /* Visit elevated processes first */
+ while (elevated_bucket_count) {
+
+ elevated_bucket_count--;
+
+ /*
+ * memorystatus_kill_elevated_process() drops a reference,
+ * so take another one so we can continue to use this exit reason
+ * even after it returns.
+ */
+
+ os_reason_ref(jetsam_reason);
+ killed = memorystatus_kill_elevated_process(
+ cause,
+ jetsam_reason,
+ jld_eval_aggressive_count,
+ &errors);
+
+ if (killed) {
+ *post_snapshot = TRUE;
+ if (memorystatus_avail_pages_below_pressure()) {
+ /*
+ * Still under pressure.
+ * Find another pinned processes.
+ */
+ continue;
+ } else {
+ return TRUE;
+ }
+ } else {
+ /*
+ * No pinned processes left to kill.
+ * Abandon elevated band.
+ */
+ break;
+ }
+ }
+
+ /*
+ * memorystatus_kill_top_process_aggressive() allocates its own
+ * jetsam_reason so the kMemorystatusKilledVMThrashing cause
+ * is consistent throughout the aggressive march.
+ */
+ killed = memorystatus_kill_top_process_aggressive(
+ kMemorystatusKilledVMThrashing,
+ jld_eval_aggressive_count,
+ jld_priority_band_max,
+ &errors);
+
+ if (killed) {
+ /* Always generate logs after aggressive kill */
+ *post_snapshot = TRUE;
+ *jld_idle_kills = 0;
+ return TRUE;
+ }
+ }
+
+ return FALSE;
+ }
+
+ return FALSE;
+}
+
+
static void
memorystatus_thread(void *param __unused, wait_result_t wr __unused)
{
static boolean_t is_vm_privileged = FALSE;
-#if CONFIG_JETSAM
boolean_t post_snapshot = FALSE;
uint32_t errors = 0;
uint32_t hwm_kill = 0;
boolean_t sort_flag = TRUE;
-
- /* Jetsam Loop Detection - locals */
- memstat_bucket_t *bucket;
- int jld_bucket_count = 0;
- struct timeval jld_now_tstamp = {0,0};
- uint64_t jld_now_msecs = 0;
-
- /* Jetsam Loop Detection - statics */
- static uint64_t jld_timestamp_msecs = 0;
- static int jld_idle_kill_candidates = 0; /* Number of available processes in band 0,1 at start */
- static int jld_idle_kills = 0; /* Number of procs killed during eval period */
- static int jld_eval_aggressive_count = 0; /* Bumps the max priority in aggressive loop */
- static int32_t jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
-#endif
+ boolean_t corpse_list_purged = FALSE;
+ int jld_idle_kills = 0;
if (is_vm_privileged == FALSE) {
/*
if (vm_restricted_to_single_processor == TRUE)
thread_vm_bind_group_add();
-
+ thread_set_thread_name(current_thread(), "VM_memorystatus");
memorystatus_thread_block(0, memorystatus_thread);
}
-#if CONFIG_JETSAM
-
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_START,
memorystatus_available_pages, memorystatus_jld_enabled, memorystatus_jld_eval_period_msecs, memorystatus_jld_eval_aggressive_count,0);
* If we run out of HWM processes and our available pages drops below the critical threshold, then,
* we target the least recently used process in order of increasing jetsam priority (exception: the FG band).
*/
- while (is_thrashing(kill_under_pressure_cause) ||
- memorystatus_available_pages <= memorystatus_available_pages_pressure) {
+ while (memorystatus_action_needed()) {
boolean_t killed;
int32_t priority;
uint32_t cause;
+ uint64_t jetsam_reason_code = JETSAM_REASON_INVALID;
+ os_reason_t jetsam_reason = OS_REASON_NULL;
- if (kill_under_pressure_cause) {
- cause = kill_under_pressure_cause;
- } else {
- cause = kMemorystatusKilledVMPageShortage;
+ cause = kill_under_pressure_cause;
+ switch (cause) {
+ case kMemorystatusKilledFCThrashing:
+ jetsam_reason_code = JETSAM_REASON_MEMORY_FCTHRASHING;
+ break;
+ case kMemorystatusKilledVMThrashing:
+ jetsam_reason_code = JETSAM_REASON_MEMORY_VMTHRASHING;
+ break;
+ case kMemorystatusKilledZoneMapExhaustion:
+ jetsam_reason_code = JETSAM_REASON_ZONE_MAP_EXHAUSTION;
+ break;
+ case kMemorystatusKilledVMPageShortage:
+ /* falls through */
+ default:
+ jetsam_reason_code = JETSAM_REASON_MEMORY_VMPAGESHORTAGE;
+ cause = kMemorystatusKilledVMPageShortage;
+ break;
}
-#if LEGACY_HIWATER
/* Highwater */
- killed = memorystatus_kill_hiwat_proc(&errors);
- if (killed) {
- hwm_kill++;
- post_snapshot = TRUE;
- goto done;
- } else {
- memorystatus_hwm_candidates = FALSE;
- }
-
- /* No highwater processes to kill. Continue or stop for now? */
- if (!is_thrashing(kill_under_pressure_cause) &&
- (memorystatus_available_pages > memorystatus_available_pages_critical)) {
- /*
- * We are _not_ out of pressure but we are above the critical threshold and there's:
- * - no compressor thrashing
- * - no more HWM processes left.
- * For now, don't kill any other processes.
- */
-
- if (hwm_kill == 0) {
- memorystatus_thread_wasted_wakeup++;
+ boolean_t is_critical = TRUE;
+ if (memorystatus_act_on_hiwat_processes(&errors, &hwm_kill, &post_snapshot, &is_critical)) {
+ if (is_critical == FALSE) {
+ /*
+ * For now, don't kill any other processes.
+ */
+ break;
+ } else {
+ goto done;
}
-
- break;
}
-#endif
- if (memorystatus_jld_enabled == TRUE) {
-
- /*
- * Jetsam Loop Detection: attempt to detect
- * rapid daemon relaunches in the lower bands.
- */
-
- microuptime(&jld_now_tstamp);
-
- /*
- * Ignore usecs in this calculation.
- * msecs granularity is close enough.
- */
- jld_now_msecs = (jld_now_tstamp.tv_sec * 1000);
-
- proc_list_lock();
- bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
- jld_bucket_count = bucket->count;
- bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE_DEFERRED];
- jld_bucket_count += bucket->count;
- proc_list_unlock();
-
- /*
- * memorystatus_jld_eval_period_msecs is a tunable
- * memorystatus_jld_eval_aggressive_count is a tunable
- * memorystatus_jld_eval_aggressive_priority_band_max is a tunable
- */
- if ( (jld_bucket_count == 0) ||
- (jld_now_msecs > (jld_timestamp_msecs + memorystatus_jld_eval_period_msecs))) {
- /*
- * Refresh evaluation parameters
- */
- jld_timestamp_msecs = jld_now_msecs;
- jld_idle_kill_candidates = jld_bucket_count;
- jld_idle_kills = 0;
- jld_eval_aggressive_count = 0;
- jld_priority_band_max = JETSAM_PRIORITY_UI_SUPPORT;
- }
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, jetsam_reason_code);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_thread: failed to allocate jetsam reason\n");
+ }
- if (jld_idle_kills > jld_idle_kill_candidates) {
- jld_eval_aggressive_count++;
- if (jld_eval_aggressive_count > memorystatus_jld_eval_aggressive_count) {
- /*
- * Bump up the jetsam priority limit (eg: the bucket index)
- * Enforce bucket index sanity.
- */
- if ((memorystatus_jld_eval_aggressive_priority_band_max < 0) ||
- (memorystatus_jld_eval_aggressive_priority_band_max >= MEMSTAT_BUCKET_COUNT)) {
- /*
- * Do nothing. Stick with the default level.
- */
- } else {
- jld_priority_band_max = memorystatus_jld_eval_aggressive_priority_band_max;
- }
- }
+ if (memorystatus_act_aggressive(cause, jetsam_reason, &jld_idle_kills, &corpse_list_purged, &post_snapshot)) {
+ goto done;
+ }
- killed = memorystatus_kill_top_process_aggressive(
- TRUE,
- kMemorystatusKilledVMThrashing,
- jld_eval_aggressive_count,
- jld_priority_band_max,
- &errors);
+ /*
+ * memorystatus_kill_top_process() drops a reference,
+ * so take another one so we can continue to use this exit reason
+ * even after it returns
+ */
+ os_reason_ref(jetsam_reason);
-
- if (killed) {
- /* Always generate logs after aggressive kill */
- post_snapshot = TRUE;
- goto done;
- }
- }
- }
-
/* LRU */
- killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, &priority, &errors);
+ killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, jetsam_reason, &priority, &errors);
sort_flag = FALSE;
if (killed) {
- /*
- * Don't generate logs for steady-state idle-exit kills,
- * unless it is overridden for debug or by the device
- * tree.
- */
- if ((priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot) {
+ if (memorystatus_post_snapshot(priority, cause) == TRUE) {
+
post_snapshot = TRUE;
}
/* Jetsam Loop Detection */
if (memorystatus_jld_enabled == TRUE) {
- if ((priority == JETSAM_PRIORITY_IDLE) || (priority == JETSAM_PRIORITY_IDLE_DEFERRED)) {
+ if ((priority == JETSAM_PRIORITY_IDLE) || (priority == system_procs_aging_band) || (priority == applications_aging_band)) {
jld_idle_kills++;
} else {
/*
*/
}
}
- goto done;
- }
-
- if (memorystatus_available_pages <= memorystatus_available_pages_critical) {
- /* Under pressure and unable to kill a process - panic */
- panic("memorystatus_jetsam_thread: no victim! available pages:%d\n", memorystatus_available_pages);
+
+ if ((priority >= JETSAM_PRIORITY_UI_SUPPORT) && (total_corpses_count() > 0) && (corpse_list_purged == FALSE)) {
+ /*
+ * If we have jetsammed a process in or above JETSAM_PRIORITY_UI_SUPPORT
+ * then we attempt to relieve pressure by purging corpse memory.
+ */
+ task_purge_all_corpses();
+ corpse_list_purged = TRUE;
+ }
+ goto done;
+ }
+
+ if (memorystatus_avail_pages_below_critical()) {
+ /*
+ * Still under pressure and unable to kill a process - purge corpse memory
+ */
+ if (total_corpses_count() > 0) {
+ task_purge_all_corpses();
+ corpse_list_purged = TRUE;
+ }
+
+ if (memorystatus_avail_pages_below_critical()) {
+ /*
+ * Still under pressure and unable to kill a process - panic
+ */
+ panic("memorystatus_jetsam_thread: no victim! available pages:%llu\n", (uint64_t)memorystatus_available_pages);
+ }
}
done:
* To avoid that, we reset the flag here and notify the
* compressor.
*/
- if (is_thrashing(kill_under_pressure_cause)) {
+ if (is_reason_thrashing(kill_under_pressure_cause)) {
kill_under_pressure_cause = 0;
+#if CONFIG_JETSAM
vm_thrashing_jetsam_done();
+#endif /* CONFIG_JETSAM */
+ } else if (is_reason_zone_map_exhaustion(kill_under_pressure_cause)) {
+ kill_under_pressure_cause = 0;
}
+
+ os_reason_free(jetsam_reason);
}
kill_under_pressure_cause = 0;
memorystatus_clear_errors();
}
-#if VM_PRESSURE_EVENTS
- /*
- * LD: We used to target the foreground process first and foremost here.
- * Now, we target all processes, starting from the non-suspended, background
- * processes first. We will target foreground too.
- *
- * memorystatus_update_vm_pressure(TRUE);
- */
- //vm_pressure_response();
-#endif
-
if (post_snapshot) {
+ proc_list_lock();
size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
uint64_t timestamp_now = mach_absolute_time();
memorystatus_jetsam_snapshot->notification_time = timestamp_now;
- if (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
- timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout) {
+ memorystatus_jetsam_snapshot->js_gencount++;
+ if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
+ timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
+ proc_list_unlock();
int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
if (!ret) {
proc_list_lock();
memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
proc_list_unlock();
}
+ } else {
+ proc_list_unlock();
}
}
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_END,
memorystatus_available_pages, 0, 0, 0, 0);
-#else /* CONFIG_JETSAM */
-
- /*
- * Jetsam not enabled
- */
-
-#endif /* CONFIG_JETSAM */
-
memorystatus_thread_block(0, memorystatus_thread);
}
-#if !CONFIG_JETSAM
/*
* Returns TRUE:
* when an idle-exitable proc was killed
* when the attempt to kill an idle-exitable proc failed
*/
boolean_t memorystatus_idle_exit_from_VM(void) {
+
+ /*
+ * This routine should no longer be needed since we are
+ * now using jetsam bands on all platforms and so will deal
+ * with IDLE processes within the memorystatus thread itself.
+ *
+ * But we still use it because we observed that macos systems
+ * started heavy compression/swapping with a bunch of
+ * idle-exitable processes alive and doing nothing. We decided
+ * to rather kill those processes than start swapping earlier.
+ */
+
return(kill_idle_exit_proc());
}
-#endif /* !CONFIG_JETSAM */
-
-#if CONFIG_JETSAM
/*
* Callback invoked when allowable physical memory footprint exceeded
* as well as the fatal task memory limits.
*/
void
-memorystatus_on_ledger_footprint_exceeded(boolean_t warning, const int max_footprint_mb)
+memorystatus_on_ledger_footprint_exceeded(boolean_t warning, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
{
- boolean_t is_active;
- boolean_t is_fatal;
+ os_reason_t jetsam_reason = OS_REASON_NULL;
proc_t p = current_proc();
- proc_list_lock();
-
- is_active = proc_jetsam_state_is_active_locked(p);
- is_fatal = (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT);
-
- if (warning == FALSE) {
- /*
- * We only want the EXC_RESOURCE to trigger once per lifetime
- * of the active/inactive limit state. So, here, we detect the
- * active/inactive state of the process and mark the
- * state as exception has been triggered.
- */
- if (is_active == TRUE) {
- /*
- * turn off exceptions for active state
- */
- p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_ACTIVE_EXC_TRIGGERED;
- } else {
- /*
- * turn off exceptions for inactive state
- */
- p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_INACTIVE_EXC_TRIGGERED;
- }
-
- /*
- * Soft memory limit is a non-fatal high-water-mark
- * Hard memory limit is a fatal custom-task-limit or system-wide per-task memory limit.
- */
- printf("process %d (%s) exceeded physical memory footprint, the %s%sMemoryLimit of %d MB\n",
- p->p_pid, p->p_comm, (is_active ? "Active" : "Inactive"),
- (is_fatal ? "Hard" : "Soft"), max_footprint_mb);
-
- }
-
- proc_list_unlock();
-
#if VM_PRESSURE_EVENTS
if (warning == TRUE) {
- if (memorystatus_warn_process(p->p_pid, TRUE /* critical? */) != TRUE) {
+ /*
+ * This is a warning path which implies that the current process is close, but has
+ * not yet exceeded its per-process memory limit.
+ */
+ if (memorystatus_warn_process(p->p_pid, memlimit_is_active, memlimit_is_fatal, FALSE /* not exceeded */) != TRUE) {
/* Print warning, since it's possible that task has not registered for pressure notifications */
- printf("task_exceeded_footprint: failed to warn the current task (exiting, or no handler registered?).\n");
+ os_log(OS_LOG_DEFAULT, "memorystatus_on_ledger_footprint_exceeded: failed to warn the current task (%d exiting, or no handler registered?).\n", p->p_pid);
}
return;
}
#endif /* VM_PRESSURE_EVENTS */
- if (is_fatal) {
+ if (memlimit_is_fatal) {
/*
* If this process has no high watermark or has a fatal task limit, then we have been invoked because the task
* has violated either the system-wide per-task memory limit OR its own task limit.
*/
- if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit) != TRUE) {
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_PERPROCESSLIMIT);
+ if (jetsam_reason == NULL) {
+ printf("task_exceeded footprint: failed to allocate jetsam reason\n");
+ } else if (corpse_for_fatal_memkill != 0) {
+ /* Set OS_REASON_FLAG_GENERATE_CRASH_REPORT to generate corpse */
+ jetsam_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
+ }
+
+ if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit, jetsam_reason) != TRUE) {
printf("task_exceeded_footprint: failed to kill the current task (exiting?).\n");
}
} else {
* See comment near its declaration for more details.
*/
memorystatus_hwm_candidates = TRUE;
+
+#if VM_PRESSURE_EVENTS
+ /*
+ * The current process is not in the warning path.
+ * This path implies the current process has exceeded a non-fatal (soft) memory limit.
+ * Failure to send note is ignored here.
+ */
+ (void)memorystatus_warn_process(p->p_pid, memlimit_is_active, memlimit_is_fatal, TRUE /* exceeded */);
+
+#endif /* VM_PRESSURE_EVENTS */
}
}
+void
+memorystatus_log_exception(const int max_footprint_mb, boolean_t memlimit_is_active, boolean_t memlimit_is_fatal)
+{
+ proc_t p = current_proc();
+
+ /*
+ * The limit violation is logged here, but only once per process per limit.
+ * Soft memory limit is a non-fatal high-water-mark
+ * Hard memory limit is a fatal custom-task-limit or system-wide per-task memory limit.
+ */
+
+ os_log_with_startup_serial(OS_LOG_DEFAULT, "EXC_RESOURCE -> %s[%d] exceeded mem limit: %s%s %d MB (%s)\n",
+ (*p->p_name ? p->p_name : "unknown"), p->p_pid, (memlimit_is_active ? "Active" : "Inactive"),
+ (memlimit_is_fatal ? "Hard" : "Soft"), max_footprint_mb,
+ (memlimit_is_fatal ? "fatal" : "non-fatal"));
+
+ return;
+}
+
+
+/*
+ * Description:
+ * Evaluates process state to determine which limit
+ * should be applied (active vs. inactive limit).
+ *
+ * Processes that have the 'elevated inactive jetsam band' attribute
+ * are first evaluated based on their current priority band.
+ * presently elevated ==> active
+ *
+ * Processes that opt into dirty tracking are evaluated
+ * based on clean vs dirty state.
+ * dirty ==> active
+ * clean ==> inactive
+ *
+ * Process that do not opt into dirty tracking are
+ * evalulated based on priority level.
+ * Foreground or above ==> active
+ * Below Foreground ==> inactive
+ *
+ * Return: TRUE if active
+ * False if inactive
+ */
+
+static boolean_t
+proc_jetsam_state_is_active_locked(proc_t p) {
+
+ if ((p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND) &&
+ (p->p_memstat_effectivepriority == JETSAM_PRIORITY_ELEVATED_INACTIVE)) {
+ /*
+ * process has the 'elevated inactive jetsam band' attribute
+ * and process is present in the elevated band
+ * implies active state
+ */
+ return TRUE;
+ } else if (p->p_memstat_dirty & P_DIRTY_TRACK) {
+ /*
+ * process has opted into dirty tracking
+ * active state is based on dirty vs. clean
+ */
+ if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
+ /*
+ * process is dirty
+ * implies active state
+ */
+ return TRUE;
+ } else {
+ /*
+ * process is clean
+ * implies inactive state
+ */
+ return FALSE;
+ }
+ } else if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
+ /*
+ * process is Foreground or higher
+ * implies active state
+ */
+ return TRUE;
+ } else {
+ /*
+ * process found below Foreground
+ * implies inactive state
+ */
+ return FALSE;
+ }
+}
+
+static boolean_t
+memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason) {
+ boolean_t res;
+
+ uint32_t errors = 0;
+
+ if (victim_pid == -1) {
+ /* No pid, so kill first process */
+ res = memorystatus_kill_top_process(TRUE, TRUE, cause, jetsam_reason, NULL, &errors);
+ } else {
+ res = memorystatus_kill_specific_process(victim_pid, cause, jetsam_reason);
+ }
+
+ if (errors) {
+ memorystatus_clear_errors();
+ }
+
+ if (res == TRUE) {
+ /* Fire off snapshot notification */
+ proc_list_lock();
+ size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
+ sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
+ uint64_t timestamp_now = mach_absolute_time();
+ memorystatus_jetsam_snapshot->notification_time = timestamp_now;
+ if (memorystatus_jetsam_snapshot_count > 0 && (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
+ timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout)) {
+ proc_list_unlock();
+ int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
+ if (!ret) {
+ proc_list_lock();
+ memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
+ proc_list_unlock();
+ }
+ } else {
+ proc_list_unlock();
+ }
+ }
+
+ return res;
+}
+
+/*
+ * Jetsam a specific process.
+ */
+static boolean_t
+memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause, os_reason_t jetsam_reason) {
+ boolean_t killed;
+ proc_t p;
+ uint64_t killtime = 0;
+ clock_sec_t tv_sec;
+ clock_usec_t tv_usec;
+ uint32_t tv_msec;
+
+ /* TODO - add a victim queue and push this into the main jetsam thread */
+
+ p = proc_find(victim_pid);
+ if (!p) {
+ os_reason_free(jetsam_reason);
+ return FALSE;
+ }
+
+ proc_list_lock();
+
+ if (memorystatus_jetsam_snapshot_count == 0) {
+ memorystatus_init_jetsam_snapshot_locked(NULL,0);
+ }
+
+ killtime = mach_absolute_time();
+ absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
+ tv_msec = tv_usec / 1000;
+
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
+
+ proc_list_unlock();
+
+ os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_specific_process pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
+ (unsigned long)tv_sec, tv_msec, victim_pid, (*p->p_name ? p->p_name : "unknown"),
+ memorystatus_kill_cause_name[cause], p->p_memstat_effectivepriority, (uint64_t)memorystatus_available_pages);
+
+ killed = memorystatus_do_kill(p, cause, jetsam_reason);
+ proc_rele(p);
+
+ return killed;
+}
+
+
/*
* Toggle the P_MEMSTAT_TERMINATED state.
* Takes the proc_list_lock.
return;
}
+
+#if CONFIG_JETSAM
/*
* This is invoked when cpulimits have been exceeded while in fatal mode.
* The jetsam_flags do not apply as those are for memory related kills.
int retval = 0;
int jetsam_flags = 0; /* make it obvious */
proc_t p = current_proc();
+ os_reason_t jetsam_reason = OS_REASON_NULL;
printf("task_exceeded_cpulimit: killing pid %d [%s]\n",
- p->p_pid, (p->p_comm ? p->p_comm : "(unknown)"));
+ p->p_pid, (*p->p_name ? p->p_name : "(unknown)"));
+
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_CPULIMIT);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("task_exceeded_cpulimit: unable to allocate memory for jetsam reason\n");
+ }
- retval = jetsam_do_kill(p, jetsam_flags);
+ retval = jetsam_do_kill(p, jetsam_flags, jetsam_reason);
if (retval) {
printf("task_exceeded_cpulimit: failed to kill current task (exiting?).\n");
}
}
+#endif /* CONFIG_JETSAM */
+
+static void
+memorystatus_get_task_memory_region_count(task_t task, uint64_t *count)
+{
+ assert(task);
+ assert(count);
+
+ *count = get_task_memory_region_count(task);
+}
+
+#if DEVELOPMENT || DEBUG
+
+/*
+ * Sysctl only used to test memorystatus_allowed_vm_map_fork() path.
+ * set a new pidwatch value
+ * or
+ * get the current pidwatch value
+ */
+
+uint64_t memorystatus_vm_map_fork_pidwatch_val = 0;
+#define MEMORYSTATUS_VM_MAP_FORK_ALLOWED 0x100000000
+#define MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED 0x200000000
+
+static int sysctl_memorystatus_vm_map_fork_pidwatch SYSCTL_HANDLER_ARGS {
+#pragma unused(oidp, arg1, arg2)
+
+ uint64_t new_value = 0;
+ uint64_t old_value = 0;
+ int error = 0;
+
+ /*
+ * The pid is held in the low 32 bits.
+ * The 'allowed' flags are in the upper 32 bits.
+ */
+ old_value = memorystatus_vm_map_fork_pidwatch_val;
+
+ error = sysctl_io_number(req, old_value, sizeof(old_value), &new_value, NULL);
+
+ if (error || !req->newptr) {
+ /*
+ * No new value passed in.
+ */
+ return(error);
+ }
+
+ /*
+ * A new pid was passed in via req->newptr.
+ * Ignore any attempt to set the higher order bits.
+ */
+ memorystatus_vm_map_fork_pidwatch_val = new_value & 0xFFFFFFFF;
+ printf("memorystatus: pidwatch old_value = 0x%llx, new_value = 0x%llx \n", old_value, new_value);
+
+ return(error);
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_map_fork_pidwatch, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED| CTLFLAG_MASKED,
+ 0, 0, sysctl_memorystatus_vm_map_fork_pidwatch, "Q", "get/set pid watched for in vm_map_fork");
+
+
+#define SET_VM_MAP_FORK_PIDWATCH_ALLOWED(task) \
+MACRO_BEGIN \
+if (memorystatus_vm_map_fork_pidwatch_val != 0) { \
+ proc_t p = get_bsdtask_info(task); \
+ if (p && (memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid)) { \
+ memorystatus_vm_map_fork_pidwatch_val |= MEMORYSTATUS_VM_MAP_FORK_ALLOWED; \
+ } \
+} \
+MACRO_END
+
+#define SET_VM_MAP_FORK_PIDWATCH_NOT_ALLOWED(task) \
+MACRO_BEGIN \
+if (memorystatus_vm_map_fork_pidwatch_val != 0) { \
+ proc_t p = get_bsdtask_info(task); \
+ if (p && (memorystatus_vm_map_fork_pidwatch_val == (uint64_t)p->p_pid)) { \
+ memorystatus_vm_map_fork_pidwatch_val |= MEMORYSTATUS_VM_MAP_FORK_NOT_ALLOWED; \
+ } \
+} \
+MACRO_END
+
+#else /* DEVELOPMENT || DEBUG */
+
+#define SET_VM_MAP_FORK_PIDWATCH_ALLOWED(task)
+#define SET_VM_MAP_FORK_PIDWATCH_NOT_ALLOWED(task)
+
+#endif /* DEVELOPMENT || DEBUG */
+
+/*
+ * Called during EXC_RESOURCE handling when a process exceeds a soft
+ * memory limit. This is the corpse fork path and here we decide if
+ * vm_map_fork will be allowed when creating the corpse.
+ * The task being considered is suspended.
+ *
+ * By default, a vm_map_fork is allowed to proceed.
+ *
+ * A few simple policy assumptions:
+ * Desktop platform is not considered in this path.
+ * The vm_map_fork is always allowed.
+ *
+ * If the device has a zero system-wide task limit,
+ * then the vm_map_fork is allowed.
+ *
+ * And if a process's memory footprint calculates less
+ * than or equal to half of the system-wide task limit,
+ * then the vm_map_fork is allowed. This calculation
+ * is based on the assumption that a process can
+ * munch memory up to the system-wide task limit.
+ */
+boolean_t
+memorystatus_allowed_vm_map_fork(__unused task_t task)
+{
+ boolean_t is_allowed = TRUE; /* default */
+
+#if CONFIG_EMBEDDED
+
+ uint64_t footprint_in_bytes = 0;
+ uint64_t purgeable_in_bytes = 0;
+ uint64_t max_allowed_bytes = 0;
+
+ if (max_task_footprint_mb == 0) {
+ SET_VM_MAP_FORK_PIDWATCH_ALLOWED(task);
+ return (is_allowed);
+ }
+
+ purgeable_in_bytes = get_task_purgeable_size(task);
+ footprint_in_bytes = get_task_phys_footprint(task);
+
+ /*
+ * Maximum is half the system-wide task limit.
+ */
+ max_allowed_bytes = ((((uint64_t)max_task_footprint_mb) * 1024ULL * 1024ULL) >> 1);
+
+ if (footprint_in_bytes > purgeable_in_bytes) {
+ footprint_in_bytes -= purgeable_in_bytes;
+ }
+
+ if (footprint_in_bytes <= max_allowed_bytes) {
+ SET_VM_MAP_FORK_PIDWATCH_ALLOWED(task);
+ return (is_allowed);
+ } else {
+ printf("memorystatus disallowed vm_map_fork %lld %lld\n", footprint_in_bytes, max_allowed_bytes);
+ SET_VM_MAP_FORK_PIDWATCH_NOT_ALLOWED(task);
+ return (!is_allowed);
+ }
+
+#else /* CONFIG_EMBEDDED */
+
+ SET_VM_MAP_FORK_PIDWATCH_ALLOWED(task);
+ return (is_allowed);
+
+#endif /* CONFIG_EMBEDDED */
+
+}
+
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)
{
assert(task);
assert(footprint);
-
- *footprint = (uint32_t)(get_task_phys_footprint(task) / PAGE_SIZE_64);
+
+ uint64_t pages;
+
+ pages = (get_task_phys_footprint(task) / PAGE_SIZE_64);
+ assert(((uint32_t)pages) == pages);
+ *footprint = (uint32_t)pages;
+
if (max_footprint) {
- *max_footprint = (uint32_t)(get_task_phys_footprint_max(task) / PAGE_SIZE_64);
+ pages = (get_task_phys_footprint_recent_max(task) / PAGE_SIZE_64);
+ assert(((uint32_t)pages) == pages);
+ *max_footprint = (uint32_t)pages;
}
if (max_footprint_lifetime) {
- *max_footprint_lifetime = (uint32_t)(get_task_resident_max(task) / PAGE_SIZE_64);
+ pages = (get_task_resident_max(task) / PAGE_SIZE_64);
+ assert(((uint32_t)pages) == pages);
+ *max_footprint_lifetime = (uint32_t)pages;
}
if (purgeable_pages) {
- *purgeable_pages = (uint32_t)(get_task_purgeable_size(task) / PAGE_SIZE_64);
+ pages = (get_task_purgeable_size(task) / PAGE_SIZE_64);
+ assert(((uint32_t)pages) == pages);
+ *purgeable_pages = (uint32_t)pages;
+ }
+}
+
+static void
+memorystatus_get_task_phys_footprint_page_counts(task_t task,
+ uint64_t *internal_pages, uint64_t *internal_compressed_pages,
+ uint64_t *purgeable_nonvolatile_pages, uint64_t *purgeable_nonvolatile_compressed_pages,
+ uint64_t *alternate_accounting_pages, uint64_t *alternate_accounting_compressed_pages,
+ uint64_t *iokit_mapped_pages, uint64_t *page_table_pages)
+{
+ assert(task);
+
+ if (internal_pages) {
+ *internal_pages = (get_task_internal(task) / PAGE_SIZE_64);
+ }
+
+ if (internal_compressed_pages) {
+ *internal_compressed_pages = (get_task_internal_compressed(task) / PAGE_SIZE_64);
+ }
+
+ if (purgeable_nonvolatile_pages) {
+ *purgeable_nonvolatile_pages = (get_task_purgeable_nonvolatile(task) / PAGE_SIZE_64);
+ }
+
+ if (purgeable_nonvolatile_compressed_pages) {
+ *purgeable_nonvolatile_compressed_pages = (get_task_purgeable_nonvolatile_compressed(task) / PAGE_SIZE_64);
+ }
+
+ if (alternate_accounting_pages) {
+ *alternate_accounting_pages = (get_task_alternate_accounting(task) / PAGE_SIZE_64);
+ }
+
+ if (alternate_accounting_compressed_pages) {
+ *alternate_accounting_compressed_pages = (get_task_alternate_accounting_compressed(task) / PAGE_SIZE_64);
+ }
+
+ if (iokit_mapped_pages) {
+ *iokit_mapped_pages = (get_task_iokit_mapped(task) / PAGE_SIZE_64);
+ }
+
+ if (page_table_pages) {
+ *page_table_pages = (get_task_page_table(task) / PAGE_SIZE_64);
}
}
+/*
+ * This routine only acts on the global jetsam event snapshot.
+ * Updating the process's entry can race when the memorystatus_thread
+ * has chosen to kill a process that is racing to exit on another core.
+ */
static void
-memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause)
+memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause, uint64_t killtime)
{
+ memorystatus_jetsam_snapshot_entry_t *entry = NULL;
+ memorystatus_jetsam_snapshot_t *snapshot = NULL;
+ memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
+
unsigned int i;
+ if (memorystatus_jetsam_snapshot_count == 0) {
+ /*
+ * No active snapshot.
+ * Nothing to do.
+ */
+ return;
+ }
+
+ /*
+ * Sanity check as this routine should only be called
+ * from a jetsam kill path.
+ */
+ assert(kill_cause != 0 && killtime != 0);
+
+ snapshot = memorystatus_jetsam_snapshot;
+ snapshot_list = memorystatus_jetsam_snapshot->entries;
+
for (i = 0; i < memorystatus_jetsam_snapshot_count; i++) {
- if (memorystatus_jetsam_snapshot_list[i].pid == p->p_pid) {
- /* Update if the priority has changed since the snapshot was taken */
- if (memorystatus_jetsam_snapshot_list[i].priority != p->p_memstat_effectivepriority) {
- memorystatus_jetsam_snapshot_list[i].priority = p->p_memstat_effectivepriority;
- strlcpy(memorystatus_jetsam_snapshot_list[i].name, p->p_comm, MAXCOMLEN+1);
- memorystatus_jetsam_snapshot_list[i].state = memorystatus_build_state(p);
- memorystatus_jetsam_snapshot_list[i].user_data = p->p_memstat_userdata;
- memorystatus_jetsam_snapshot_list[i].fds = p->p_fd->fd_nfiles;
+ if (snapshot_list[i].pid == p->p_pid) {
+
+ entry = &snapshot_list[i];
+
+ if (entry->killed || entry->jse_killtime) {
+ /*
+ * We apparently raced on the exit path
+ * for this process, as it's snapshot entry
+ * has already recorded a kill.
+ */
+ assert(entry->killed && entry->jse_killtime);
+ break;
}
- memorystatus_jetsam_snapshot_list[i].killed = kill_cause;
- return;
+
+ /*
+ * Update the entry we just found in the snapshot.
+ */
+
+ entry->killed = kill_cause;
+ entry->jse_killtime = killtime;
+ entry->jse_gencount = snapshot->js_gencount;
+ entry->jse_idle_delta = p->p_memstat_idle_delta;
+
+ /*
+ * If a process has moved between bands since snapshot was
+ * initialized, then likely these fields changed too.
+ */
+ if (entry->priority != p->p_memstat_effectivepriority) {
+
+ strlcpy(entry->name, p->p_name, sizeof(entry->name));
+ entry->priority = p->p_memstat_effectivepriority;
+ entry->state = memorystatus_build_state(p);
+ entry->user_data = p->p_memstat_userdata;
+ entry->fds = p->p_fd->fd_nfiles;
+ }
+
+ /*
+ * Always update the page counts on a kill.
+ */
+
+ uint32_t pages = 0;
+ uint32_t max_pages = 0;
+ uint32_t max_pages_lifetime = 0;
+ uint32_t purgeable_pages = 0;
+
+ memorystatus_get_task_page_counts(p->task, &pages, &max_pages, &max_pages_lifetime, &purgeable_pages);
+ entry->pages = (uint64_t)pages;
+ entry->max_pages = (uint64_t)max_pages;
+ entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
+ entry->purgeable_pages = (uint64_t)purgeable_pages;
+
+ uint64_t internal_pages = 0;
+ uint64_t internal_compressed_pages = 0;
+ uint64_t purgeable_nonvolatile_pages = 0;
+ uint64_t purgeable_nonvolatile_compressed_pages = 0;
+ uint64_t alternate_accounting_pages = 0;
+ uint64_t alternate_accounting_compressed_pages = 0;
+ uint64_t iokit_mapped_pages = 0;
+ uint64_t page_table_pages = 0;
+
+ memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
+ &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
+ &alternate_accounting_pages, &alternate_accounting_compressed_pages,
+ &iokit_mapped_pages, &page_table_pages);
+
+ entry->jse_internal_pages = internal_pages;
+ entry->jse_internal_compressed_pages = internal_compressed_pages;
+ entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
+ entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
+ entry->jse_alternate_accounting_pages = alternate_accounting_pages;
+ entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
+ entry->jse_iokit_mapped_pages = iokit_mapped_pages;
+ entry->jse_page_table_pages = page_table_pages;
+
+ uint64_t region_count = 0;
+ memorystatus_get_task_memory_region_count(p->task, ®ion_count);
+ entry->jse_memory_region_count = region_count;
+
+ goto exit;
}
}
+
+ if (entry == NULL) {
+ /*
+ * The entry was not found in the snapshot, so the process must have
+ * launched after the snapshot was initialized.
+ * Let's try to append the new entry.
+ */
+ if (memorystatus_jetsam_snapshot_count < memorystatus_jetsam_snapshot_max) {
+ /*
+ * A populated snapshot buffer exists
+ * and there is room to init a new entry.
+ */
+ assert(memorystatus_jetsam_snapshot_count == snapshot->entry_count);
+
+ unsigned int next = memorystatus_jetsam_snapshot_count;
+
+ if(memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[next], (snapshot->js_gencount)) == TRUE) {
+
+ entry = &snapshot_list[next];
+ entry->killed = kill_cause;
+ entry->jse_killtime = killtime;
+
+ snapshot->entry_count = ++next;
+ memorystatus_jetsam_snapshot_count = next;
+
+ if (memorystatus_jetsam_snapshot_count >= memorystatus_jetsam_snapshot_max) {
+ /*
+ * We just used the last slot in the snapshot buffer.
+ * We only want to log it once... so we do it here
+ * when we notice we've hit the max.
+ */
+ printf("memorystatus: WARNING snapshot buffer is full, count %d\n",
+ memorystatus_jetsam_snapshot_count);
+ }
+ }
+ }
+ }
+
+exit:
+ if (entry == NULL) {
+ /*
+ * If we reach here, the snapshot buffer could not be updated.
+ * Most likely, the buffer is full, in which case we would have
+ * logged a warning in the previous call.
+ *
+ * For now, we will stop appending snapshot entries.
+ * When the buffer is consumed, the snapshot state will reset.
+ */
+
+ MEMORYSTATUS_DEBUG(4, "memorystatus_update_jetsam_snapshot_entry_locked: failed to update pid %d, priority %d, count %d\n",
+ p->p_pid, p->p_memstat_effectivepriority, memorystatus_jetsam_snapshot_count);
+ }
+
+ return;
}
+#if CONFIG_JETSAM
void memorystatus_pages_update(unsigned int pages_avail)
{
memorystatus_available_pages = pages_avail;
|| (memorystatus_available_pages >= (pages_avail + memorystatus_delta))) ? TRUE : FALSE;
if (critical || delta) {
- memorystatus_level = memorystatus_available_pages * 100 / atop_64(max_mem);
+ unsigned int total_pages;
+
+ total_pages = (unsigned int) atop_64(max_mem);
+#if CONFIG_SECLUDED_MEMORY
+ total_pages -= vm_page_secluded_count;
+#endif /* CONFIG_SECLUDED_MEMORY */
+ memorystatus_level = memorystatus_available_pages * 100 / total_pages;
memorystatus_thread_wake();
}
#endif /* VM_PRESSURE_EVENTS */
}
+#endif /* CONFIG_JETSAM */
static boolean_t
-memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry)
+memorystatus_init_jetsam_snapshot_entry_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry, uint64_t gencount)
{
clock_sec_t tv_sec;
clock_usec_t tv_usec;
+ uint32_t pages = 0;
+ uint32_t max_pages = 0;
+ uint32_t max_pages_lifetime = 0;
+ uint32_t purgeable_pages = 0;
+ uint64_t internal_pages = 0;
+ uint64_t internal_compressed_pages = 0;
+ uint64_t purgeable_nonvolatile_pages = 0;
+ uint64_t purgeable_nonvolatile_compressed_pages = 0;
+ uint64_t alternate_accounting_pages = 0;
+ uint64_t alternate_accounting_compressed_pages = 0;
+ uint64_t iokit_mapped_pages = 0;
+ uint64_t page_table_pages =0;
+ uint64_t region_count = 0;
+ uint64_t cids[COALITION_NUM_TYPES];
memset(entry, 0, sizeof(memorystatus_jetsam_snapshot_entry_t));
-
+
entry->pid = p->p_pid;
- strlcpy(&entry->name[0], p->p_comm, MAXCOMLEN+1);
+ strlcpy(&entry->name[0], p->p_name, sizeof(entry->name));
entry->priority = p->p_memstat_effectivepriority;
- memorystatus_get_task_page_counts(p->task, &entry->pages, &entry->max_pages, &entry->max_pages_lifetime, &entry->purgeable_pages);
- entry->state = memorystatus_build_state(p);
+
+ memorystatus_get_task_page_counts(p->task, &pages, &max_pages, &max_pages_lifetime, &purgeable_pages);
+ entry->pages = (uint64_t)pages;
+ entry->max_pages = (uint64_t)max_pages;
+ entry->max_pages_lifetime = (uint64_t)max_pages_lifetime;
+ entry->purgeable_pages = (uint64_t)purgeable_pages;
+
+ memorystatus_get_task_phys_footprint_page_counts(p->task, &internal_pages, &internal_compressed_pages,
+ &purgeable_nonvolatile_pages, &purgeable_nonvolatile_compressed_pages,
+ &alternate_accounting_pages, &alternate_accounting_compressed_pages,
+ &iokit_mapped_pages, &page_table_pages);
+
+ entry->jse_internal_pages = internal_pages;
+ entry->jse_internal_compressed_pages = internal_compressed_pages;
+ entry->jse_purgeable_nonvolatile_pages = purgeable_nonvolatile_pages;
+ entry->jse_purgeable_nonvolatile_compressed_pages = purgeable_nonvolatile_compressed_pages;
+ entry->jse_alternate_accounting_pages = alternate_accounting_pages;
+ entry->jse_alternate_accounting_compressed_pages = alternate_accounting_compressed_pages;
+ entry->jse_iokit_mapped_pages = iokit_mapped_pages;
+ entry->jse_page_table_pages = page_table_pages;
+
+ memorystatus_get_task_memory_region_count(p->task, ®ion_count);
+ entry->jse_memory_region_count = region_count;
+
+ entry->state = memorystatus_build_state(p);
entry->user_data = p->p_memstat_userdata;
memcpy(&entry->uuid[0], &p->p_uuid[0], sizeof(p->p_uuid));
- entry->fds = p->p_fd->fd_nfiles;
+ entry->fds = p->p_fd->fd_nfiles;
absolutetime_to_microtime(get_task_cpu_time(p->task), &tv_sec, &tv_usec);
entry->cpu_time.tv_sec = tv_sec;
entry->cpu_time.tv_usec = tv_usec;
+ assert(p->p_stats != NULL);
+ entry->jse_starttime = p->p_stats->ps_start; /* abstime process started */
+ entry->jse_killtime = 0; /* abstime jetsam chose to kill process */
+ entry->killed = 0; /* the jetsam kill cause */
+ entry->jse_gencount = gencount; /* indicates a pass through jetsam thread, when process was targeted to be killed */
+
+ entry->jse_idle_delta = p->p_memstat_idle_delta; /* Most recent timespan spent in idle-band */
+
+ proc_coalitionids(p, cids);
+ entry->jse_coalition_jetsam_id = cids[COALITION_TYPE_JETSAM];
+
return TRUE;
}
snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
}
+
+ get_zone_map_size(&snapshot->stats.zone_map_size, &snapshot->stats.zone_map_capacity);
+ get_largest_zone_info(snapshot->stats.largest_zone_name, sizeof(snapshot->stats.largest_zone_name),
+ &snapshot->stats.largest_zone_size);
}
/*
snapshot_max = memorystatus_jetsam_snapshot_max;
}
+ /*
+ * Init the snapshot header information
+ */
memorystatus_init_snapshot_vmstats(snapshot);
+ snapshot->snapshot_time = mach_absolute_time();
+ snapshot->notification_time = 0;
+ snapshot->js_gencount = 0;
next_p = memorystatus_get_first_proc_locked(&b, TRUE);
while (next_p) {
p = next_p;
next_p = memorystatus_get_next_proc_locked(&b, p, TRUE);
- if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i])) {
+ if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i], snapshot->js_gencount)) {
continue;
}
}
}
- snapshot->snapshot_time = mach_absolute_time();
snapshot->entry_count = i;
if (!od_snapshot) {
#if DEVELOPMENT || DEBUG
+#if CONFIG_JETSAM
static int
memorystatus_cmd_set_panic_bits(user_addr_t buffer, uint32_t buffer_size) {
int ret;
return ret;
}
+#endif /* CONFIG_JETSAM */
/*
* Triggers a sort_order on a specified jetsam priority band.
int error = 0;
- unsigned int bucket_index = 0;
-
- if (priority == -1) {
- /* Use as shorthand for default priority */
- bucket_index = JETSAM_PRIORITY_DEFAULT;
- } else {
- bucket_index = (unsigned int)priority;
- }
-
- error = memorystatus_sort_bucket(bucket_index, sort_order);
-
- return (error);
-}
-
-#endif
-
-/*
- * Jetsam a specific process.
- */
-static boolean_t
-memorystatus_kill_specific_process(pid_t victim_pid, uint32_t cause) {
- boolean_t killed;
- proc_t p;
-
- /* TODO - add a victim queue and push this into the main jetsam thread */
-
- p = proc_find(victim_pid);
- if (!p) {
- return FALSE;
- }
-
- printf("memorystatus: specifically killing pid %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
- victim_pid, (p->p_comm ? p->p_comm : "(unknown)"),
- jetsam_kill_cause_name[cause], p->p_memstat_effectivepriority, memorystatus_available_pages);
-
- proc_list_lock();
+ unsigned int bucket_index = 0;
- if (memorystatus_jetsam_snapshot_count == 0) {
- memorystatus_init_jetsam_snapshot_locked(NULL,0);
+ if (priority == -1) {
+ /* Use as shorthand for default priority */
+ bucket_index = JETSAM_PRIORITY_DEFAULT;
+ } else {
+ bucket_index = (unsigned int)priority;
}
- memorystatus_update_jetsam_snapshot_entry_locked(p, cause);
- proc_list_unlock();
-
- killed = memorystatus_do_kill(p, cause);
- proc_rele(p);
-
- return killed;
+ error = memorystatus_sort_bucket(bucket_index, sort_order);
+
+ return (error);
}
+#endif /* DEVELOPMENT || DEBUG */
+
/*
* Jetsam the first process in the queue.
*/
static boolean_t
-memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, int32_t *priority, uint32_t *errors)
+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)
{
pid_t aPid;
proc_t p = PROC_NULL, next_p = PROC_NULL;
- boolean_t new_snapshot = FALSE, killed = FALSE;
+ boolean_t new_snapshot = FALSE, force_new_snapshot = FALSE, killed = FALSE;
int kill_count = 0;
unsigned int i = 0;
uint32_t aPid_ep;
+ uint64_t killtime = 0;
+ clock_sec_t tv_sec;
+ clock_usec_t tv_usec;
+ uint32_t tv_msec;
+ int32_t local_max_kill_prio = JETSAM_PRIORITY_IDLE;
#ifndef CONFIG_FREEZE
#pragma unused(any)
memorystatus_available_pages, 0, 0, 0, 0);
+#if CONFIG_JETSAM
if (sort_flag == TRUE) {
(void)memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
}
+ local_max_kill_prio = max_kill_priority;
+
+ force_new_snapshot = FALSE;
+
+#else /* CONFIG_JETSAM */
+
+ if (sort_flag == TRUE) {
+ (void)memorystatus_sort_bucket(JETSAM_PRIORITY_IDLE, JETSAM_SORT_DEFAULT);
+ }
+
+ /*
+ * On macos, we currently only have 2 reasons to be here:
+ *
+ * kMemorystatusKilledZoneMapExhaustion
+ * AND
+ * kMemorystatusKilledVMThrashing
+ *
+ * If we are here because of kMemorystatusKilledZoneMapExhaustion, we will consider
+ * any and all processes as eligible kill candidates since we need to avoid a panic.
+ *
+ * Since this function can be called async. it is harder to toggle the max_kill_priority
+ * value before and after a call. And so we use this local variable to set the upper band
+ * on the eligible kill bands.
+ */
+ if (cause == kMemorystatusKilledZoneMapExhaustion) {
+ local_max_kill_prio = JETSAM_PRIORITY_MAX;
+ } else {
+ local_max_kill_prio = max_kill_priority;
+ }
+
+ /*
+ * And, because we are here under extreme circumstances, we force a snapshot even for
+ * IDLE kills.
+ */
+ force_new_snapshot = TRUE;
+
+#endif /* CONFIG_JETSAM */
+
proc_list_lock();
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
- while (next_p) {
+ while (next_p && (next_p->p_memstat_effectivepriority <= local_max_kill_prio)) {
#if DEVELOPMENT || DEBUG
int activeProcess;
int procSuspendedForDiagnosis;
aPid_ep = p->p_memstat_effectivepriority;
if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
- continue;
+ continue; /* with lock held */
}
-#if DEVELOPMENT || DEBUG
+#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && procSuspendedForDiagnosis) {
printf("jetsam: continuing after ignoring proc suspended already for diagnosis - %d\n", aPid);
continue;
}
-#endif /* DEVELOPMENT || DEBUG */
+#endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
if (cause == kMemorystatusKilledVnodes)
{
* If the system runs out of vnodes, we systematically jetsam
* processes in hopes of stumbling onto a vnode gain that helps
* the system recover. The process that happens to trigger
- * this path has no known relationship to the vnode consumption.
- * We attempt to safeguard that process e.g: do not jetsam it.
+ * this path has no known relationship to the vnode shortage.
+ * Deadlock avoidance: attempt to safeguard the caller.
*/
if (p == current_proc()) {
{
/*
* Capture a snapshot if none exists and:
+ * - we are forcing a new snapshot creation, either because:
+ * - on a particular platform we need these snapshots every time, OR
+ * - a boot-arg/embedded device tree property has been set.
* - priority was not requested (this is something other than an ambient kill)
* - the priority was requested *and* the targeted process is not at idle priority
*/
if ((memorystatus_jetsam_snapshot_count == 0) &&
- (memorystatus_idle_snapshot || ((!priority) || (priority && (*priority != JETSAM_PRIORITY_IDLE))))) {
+ (force_new_snapshot || memorystatus_idle_snapshot || ((!priority) || (priority && (aPid_ep != JETSAM_PRIORITY_IDLE))))) {
memorystatus_init_jetsam_snapshot_locked(NULL,0);
new_snapshot = TRUE;
}
* acquisition of the proc lock.
*/
p->p_memstat_state |= P_MEMSTAT_TERMINATED;
+
+ killtime = mach_absolute_time();
+ absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
+ tv_msec = tv_usec / 1000;
-#if DEVELOPMENT || DEBUG
+#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && activeProcess) {
MEMORYSTATUS_DEBUG(1, "jetsam: suspending pid %d [%s] (active) for diagnosis - memory_status_level: %d\n",
- aPid, (p->p_comm ? p->p_comm: "(unknown)"), memorystatus_level);
- memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic);
+ aPid, (*p->p_name ? p->p_name: "(unknown)"), memorystatus_level);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic, killtime);
p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
if (memorystatus_jetsam_policy & kPolicyDiagnoseFirst) {
jetsam_diagnostic_suspended_one_active_proc = 1;
goto exit;
} else
-#endif /* DEVELOPMENT || DEBUG */
+#endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
{
/* Shift queue, update stats */
- memorystatus_update_jetsam_snapshot_entry_locked(p, cause);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
if (proc_ref_locked(p) == p) {
proc_list_unlock();
- printf("memorystatus: %s %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
- ((aPid_ep == JETSAM_PRIORITY_IDLE) ?
- "idle exiting pid" : "jetsam killing pid"),
- aPid, (p->p_comm ? p->p_comm : "(unknown)"),
- jetsam_kill_cause_name[cause], aPid_ep, memorystatus_available_pages);
+ os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: %s pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
+ (unsigned long)tv_sec, tv_msec,
+ ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "killing_idle_process" : "killing_top_process"),
+ aPid, (*p->p_name ? p->p_name : "unknown"),
+ memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
+
+ /*
+ * memorystatus_do_kill() drops a reference, so take another one so we can
+ * continue to use this exit reason even after memorystatus_do_kill()
+ * returns.
+ */
+ os_reason_ref(jetsam_reason);
- killed = memorystatus_do_kill(p, cause);
+ killed = memorystatus_do_kill(p, cause, jetsam_reason);
/* Success? */
if (killed) {
proc_list_unlock();
exit:
+ os_reason_free(jetsam_reason);
+
/* Clear snapshot if freshly captured and no target was found */
if (new_snapshot && !killed) {
- memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
+ proc_list_lock();
+ memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
+ proc_list_unlock();
}
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
* Jetsam aggressively
*/
static boolean_t
-memorystatus_kill_top_process_aggressive(boolean_t any, uint32_t cause, int aggr_count, int32_t priority_max,
- uint32_t *errors)
+memorystatus_kill_top_process_aggressive(uint32_t cause, int aggr_count,
+ int32_t priority_max, uint32_t *errors)
{
pid_t aPid;
proc_t p = PROC_NULL, next_p = PROC_NULL;
int kill_count = 0;
unsigned int i = 0;
int32_t aPid_ep = 0;
-
-#pragma unused(any)
+ unsigned int memorystatus_level_snapshot = 0;
+ uint64_t killtime = 0;
+ clock_sec_t tv_sec;
+ clock_usec_t tv_usec;
+ uint32_t tv_msec;
+ os_reason_t jetsam_reason = OS_REASON_NULL;
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
memorystatus_available_pages, priority_max, 0, 0, 0);
+ memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
+
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, cause);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_top_process_aggressive: failed to allocate exit reason\n");
+ }
+
proc_list_lock();
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
int procSuspendedForDiagnosis;
#endif /* DEVELOPMENT || DEBUG */
- if ((unsigned int)(next_p->p_memstat_effectivepriority) != i) {
+ if (((next_p->p_listflag & P_LIST_EXITED) != 0) ||
+ ((unsigned int)(next_p->p_memstat_effectivepriority) != i)) {
/*
- * We have raced with next_p running on another core, as it has
- * moved to a different jetsam priority band. This means we have
- * lost our place in line while traversing the jetsam list. We
+ * We have raced with next_p running on another core.
+ * It may be exiting or it may have moved to a different
+ * jetsam priority band. This means we have lost our
+ * place in line while traversing the jetsam list. We
* attempt to recover by rewinding to the beginning of the band
* we were already traversing. By doing this, we do not guarantee
* that no process escapes this aggressive march, but we can make
* skipping an entire range of processes less likely. (PR-21069019)
*/
- MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: rewinding %s moved from band %d --> %d\n",
- aggr_count, next_p->p_comm, i, next_p->p_memstat_effectivepriority);
+ MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: rewinding band %d, %s(%d) moved or exiting.\n",
+ aggr_count, i, (*next_p->p_name ? next_p->p_name : "unknown"), next_p->p_pid);
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
continue;
continue;
}
-#if DEVELOPMENT || DEBUG
+#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if ((memorystatus_jetsam_policy & kPolicyDiagnoseActive) && procSuspendedForDiagnosis) {
printf("jetsam: continuing after ignoring proc suspended already for diagnosis - %d\n", aPid);
continue;
}
-#endif /* DEVELOPMENT || DEBUG */
+#endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
/*
* Capture a snapshot if none exists.
* acquisition of the proc lock.
*/
p->p_memstat_state |= P_MEMSTAT_TERMINATED;
+
+ killtime = mach_absolute_time();
+ absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
+ tv_msec = tv_usec / 1000;
/* Shift queue, update stats */
- memorystatus_update_jetsam_snapshot_entry_locked(p, cause);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
/*
* In order to kill the target process, we will drop the proc_list_lock.
*/
MEMORYSTATUS_DEBUG(1, "memorystatus: aggressive%d: skipping %d [%s] (exiting?)\n",
- aggr_count, next_p->p_pid, (next_p->p_comm ? next_p->p_comm : "(unknown)"));
+ aggr_count, next_p->p_pid, (*next_p->p_name ? next_p->p_name : "(unknown)"));
temp_p = next_p;
next_p = memorystatus_get_next_proc_locked(&i, temp_p, TRUE);
}
proc_list_unlock();
- printf("memorystatus: aggressive%d: %s %d [%s] (%s %d) - memorystatus_available_pages: %d\n",
- aggr_count,
- ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "idle exiting pid" : "jetsam killing pid"),
- aPid, (p->p_comm ? p->p_comm : "(unknown)"),
- jetsam_kill_cause_name[cause], aPid_ep, memorystatus_available_pages);
+ printf("%lu.%03d memorystatus: %s%d pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
+ (unsigned long)tv_sec, tv_msec,
+ ((aPid_ep == JETSAM_PRIORITY_IDLE) ? "killing_idle_process_aggressive" : "killing_top_process_aggressive"),
+ aggr_count, aPid, (*p->p_name ? p->p_name : "unknown"),
+ memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
+
+ memorystatus_level_snapshot = memorystatus_level;
+
+ /*
+ * memorystatus_do_kill() drops a reference, so take another one so we can
+ * continue to use this exit reason even after memorystatus_do_kill()
+ * returns.
+ */
+ os_reason_ref(jetsam_reason);
+ killed = memorystatus_do_kill(p, cause, jetsam_reason);
- killed = memorystatus_do_kill(p, cause);
-
/* Success? */
if (killed) {
proc_rele(p);
if (next_p) {
proc_rele_locked(next_p);
}
+
+ if (aPid_ep == JETSAM_PRIORITY_FOREGROUND && memorystatus_aggressive_jetsam_lenient == TRUE) {
+ if (memorystatus_level > memorystatus_level_snapshot && ((memorystatus_level - memorystatus_level_snapshot) >= AGGRESSIVE_JETSAM_LENIENT_MODE_THRESHOLD)) {
+#if DEVELOPMENT || DEBUG
+ printf("Disabling Lenient mode after one-time deployment.\n");
+#endif /* DEVELOPMENT || DEBUG */
+ memorystatus_aggressive_jetsam_lenient = FALSE;
+ break;
+ }
+ }
+
continue;
}
proc_list_unlock();
exit:
+ os_reason_free(jetsam_reason);
+
/* Clear snapshot if freshly captured and no target was found */
if (new_snapshot && (kill_count == 0)) {
memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
}
}
-#if LEGACY_HIWATER
-
static boolean_t
memorystatus_kill_hiwat_proc(uint32_t *errors)
{
int kill_count = 0;
unsigned int i = 0;
uint32_t aPid_ep;
-
+ uint64_t killtime = 0;
+ clock_sec_t tv_sec;
+ clock_usec_t tv_usec;
+ uint32_t tv_msec;
+ os_reason_t jetsam_reason = OS_REASON_NULL;
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
memorystatus_available_pages, 0, 0, 0, 0);
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_HIGHWATER);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_hiwat_proc: failed to allocate exit reason\n");
+ }
+
proc_list_lock();
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
while (next_p) {
- uint32_t footprint;
- boolean_t skip;
+ uint64_t footprint_in_bytes = 0;
+ uint64_t memlimit_in_bytes = 0;
+ boolean_t skip = 0;
p = next_p;
next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
continue;
}
-#if 0
- /*
- * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
- * Background limits are described via the inactive limit slots.
- * Their fatal/non-fatal setting will drive whether or not to be
- * considered in this kill path.
- */
-
- /* skip if a currently inapplicable limit is encountered */
- if ((p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
- continue;
- }
-#endif
-
- footprint = (uint32_t)(get_task_phys_footprint(p->task) / (1024 * 1024));
- skip = (((int32_t)footprint) <= p->p_memstat_memlimit);
+ footprint_in_bytes = get_task_phys_footprint(p->task);
+ memlimit_in_bytes = (((uint64_t)p->p_memstat_memlimit) * 1024ULL * 1024ULL); /* convert MB to bytes */
+ skip = (footprint_in_bytes <= memlimit_in_bytes);
-#if DEVELOPMENT || DEBUG
+#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if (!skip && (memorystatus_jetsam_policy & kPolicyDiagnoseActive)) {
if (p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED) {
continue;
}
}
-#endif /* DEVELOPMENT || DEBUG */
+#endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
#if CONFIG_FREEZE
if (!skip) {
if (skip) {
continue;
} else {
- MEMORYSTATUS_DEBUG(1, "jetsam: %s pid %d [%s] - %d Mb > 1 (%d Mb)\n",
- (memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing", aPid, p->p_comm, footprint, p->p_memstat_memlimit);
+#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
+ MEMORYSTATUS_DEBUG(1, "jetsam: %s pid %d [%s] - %lld Mb > 1 (%d Mb)\n",
+ (memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing",
+ aPid, (*p->p_name ? p->p_name : "unknown"),
+ (footprint_in_bytes / (1024ULL * 1024ULL)), /* converted bytes to MB */
+ p->p_memstat_memlimit);
+#endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
if (memorystatus_jetsam_snapshot_count == 0) {
memorystatus_init_jetsam_snapshot_locked(NULL,0);
}
p->p_memstat_state |= P_MEMSTAT_TERMINATED;
+
+ killtime = mach_absolute_time();
+ absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
+ tv_msec = tv_usec / 1000;
-#if DEVELOPMENT || DEBUG
+#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
MEMORYSTATUS_DEBUG(1, "jetsam: pid %d suspended for diagnosis - memorystatus_available_pages: %d\n", aPid, memorystatus_available_pages);
- memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic, killtime);
p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
p = proc_ref_locked(p);
goto exit;
} else
-#endif /* DEVELOPMENT || DEBUG */
+#endif /* CONFIG_JETSAM && (DEVELOPMENT || DEBUG) */
{
- memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledHiwat);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledHiwat, killtime);
if (proc_ref_locked(p) == p) {
proc_list_unlock();
- printf("memorystatus: jetsam killing pid %d [%s] (highwater %d) - memorystatus_available_pages: %d\n",
- aPid, (p->p_comm ? p->p_comm : "(unknown)"), aPid_ep, memorystatus_available_pages);
+ os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_highwater_process pid %d [%s] (highwater %d) - memorystatus_available_pages: %llu\n",
+ (unsigned long)tv_sec, tv_msec, aPid, (*p->p_name ? p->p_name : "unknown"), aPid_ep, (uint64_t)memorystatus_available_pages);
+
+ /*
+ * memorystatus_do_kill drops a reference, so take another one so we can
+ * continue to use this exit reason even after memorystatus_do_kill()
+ * returns
+ */
+ os_reason_ref(jetsam_reason);
+
+ killed = memorystatus_do_kill(p, kMemorystatusKilledHiwat, jetsam_reason);
- killed = memorystatus_do_kill(p, kMemorystatusKilledHiwat);
-
/* Success? */
if (killed) {
proc_rele(p);
proc_list_unlock();
exit:
+ os_reason_free(jetsam_reason);
+
/* Clear snapshot if freshly captured and no target was found */
if (new_snapshot && !killed) {
+ proc_list_lock();
memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
+ proc_list_unlock();
}
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_END,
return killed;
}
-#endif /* LEGACY_HIWATER */
+/*
+ * Jetsam a process pinned in the elevated band.
+ *
+ * Return: true -- at least one pinned process was jetsammed
+ * false -- no pinned process was jetsammed
+ */
+static boolean_t
+memorystatus_kill_elevated_process(uint32_t cause, os_reason_t jetsam_reason, int aggr_count, uint32_t *errors)
+{
+ pid_t aPid = 0;
+ proc_t p = PROC_NULL, next_p = PROC_NULL;
+ boolean_t new_snapshot = FALSE, killed = FALSE;
+ int kill_count = 0;
+ unsigned int i = JETSAM_PRIORITY_ELEVATED_INACTIVE;
+ uint32_t aPid_ep;
+ uint64_t killtime = 0;
+ clock_sec_t tv_sec;
+ clock_usec_t tv_usec;
+ uint32_t tv_msec;
+
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
+ memorystatus_available_pages, 0, 0, 0, 0);
+
+ proc_list_lock();
+
+ next_p = memorystatus_get_first_proc_locked(&i, FALSE);
+ while (next_p) {
+
+ p = next_p;
+ next_p = memorystatus_get_next_proc_locked(&i, p, FALSE);
+
+ aPid = p->p_pid;
+ aPid_ep = p->p_memstat_effectivepriority;
+
+ /*
+ * Only pick a process pinned in this elevated band
+ */
+ if (!(p->p_memstat_state & P_MEMSTAT_USE_ELEVATED_INACTIVE_BAND)) {
+ continue;
+ }
+
+ if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
+ continue;
+ }
+
+#if CONFIG_FREEZE
+ if (p->p_memstat_state & P_MEMSTAT_LOCKED) {
+ continue;
+ }
+#endif
+
+#if DEVELOPMENT || DEBUG
+ MEMORYSTATUS_DEBUG(1, "jetsam: elevated%d process pid %d [%s] - memorystatus_available_pages: %d\n",
+ aggr_count,
+ aPid, (*p->p_name ? p->p_name : "unknown"),
+ memorystatus_available_pages);
+#endif /* DEVELOPMENT || DEBUG */
+
+ if (memorystatus_jetsam_snapshot_count == 0) {
+ memorystatus_init_jetsam_snapshot_locked(NULL,0);
+ new_snapshot = TRUE;
+ }
+
+ p->p_memstat_state |= P_MEMSTAT_TERMINATED;
+
+ killtime = mach_absolute_time();
+ absolutetime_to_microtime(killtime, &tv_sec, &tv_usec);
+ tv_msec = tv_usec / 1000;
+
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause, killtime);
+
+ if (proc_ref_locked(p) == p) {
+
+ proc_list_unlock();
+
+ os_log_with_startup_serial(OS_LOG_DEFAULT, "%lu.%03d memorystatus: killing_top_process_elevated%d pid %d [%s] (%s %d) - memorystatus_available_pages: %llu\n",
+ (unsigned long)tv_sec, tv_msec,
+ aggr_count,
+ aPid, (*p->p_name ? p->p_name : "unknown"),
+ memorystatus_kill_cause_name[cause], aPid_ep, (uint64_t)memorystatus_available_pages);
+
+ /*
+ * memorystatus_do_kill drops a reference, so take another one so we can
+ * continue to use this exit reason even after memorystatus_do_kill()
+ * returns
+ */
+ os_reason_ref(jetsam_reason);
+ killed = memorystatus_do_kill(p, cause, jetsam_reason);
+
+ /* Success? */
+ if (killed) {
+ proc_rele(p);
+ kill_count++;
+ goto exit;
+ }
+
+ /*
+ * Failure - first unwind the state,
+ * then fall through to restart the search.
+ */
+ proc_list_lock();
+ proc_rele_locked(p);
+ p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
+ p->p_memstat_state |= P_MEMSTAT_ERROR;
+ *errors += 1;
+ }
+
+ /*
+ * Failure - restart the search.
+ *
+ * We might have raced with "p" exiting on another core, resulting in no
+ * ref on "p". Or, we may have failed to kill "p".
+ *
+ * Either way, we fall thru to here, leaving the proc in the
+ * P_MEMSTAT_TERMINATED state or P_MEMSTAT_ERROR state.
+ *
+ * And, we hold the the proc_list_lock at this point.
+ */
+
+ next_p = memorystatus_get_first_proc_locked(&i, FALSE);
+ }
+
+ proc_list_unlock();
+
+exit:
+ os_reason_free(jetsam_reason);
+
+ /* Clear snapshot if freshly captured and no target was found */
+ if (new_snapshot && (kill_count == 0)) {
+ proc_list_lock();
+ memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
+ proc_list_unlock();
+ }
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
+ memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
+
+ return (killed);
+}
static boolean_t
memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause) {
- /* TODO: allow a general async path */
+ /*
+ * TODO: allow a general async path
+ *
+ * NOTE: If a new async kill cause is added, make sure to update memorystatus_thread() to
+ * add the appropriate exit reason code mapping.
+ */
if ((victim_pid != -1) || (cause != kMemorystatusKilledVMPageShortage && cause != kMemorystatusKilledVMThrashing &&
- cause != kMemorystatusKilledFCThrashing)) {
+ cause != kMemorystatusKilledFCThrashing && cause != kMemorystatusKilledZoneMapExhaustion)) {
return FALSE;
}
return TRUE;
}
-static boolean_t
-memorystatus_kill_process_sync(pid_t victim_pid, uint32_t cause) {
- boolean_t res;
- uint32_t errors = 0;
-
- if (victim_pid == -1) {
- /* No pid, so kill first process */
- res = memorystatus_kill_top_process(TRUE, TRUE, cause, NULL, &errors);
+boolean_t
+memorystatus_kill_on_VM_thrashing(boolean_t async) {
+ if (async) {
+ return memorystatus_kill_process_async(-1, kMemorystatusKilledVMThrashing);
} else {
- res = memorystatus_kill_specific_process(victim_pid, cause);
- }
-
- if (errors) {
- memorystatus_clear_errors();
- }
-
- if (res == TRUE) {
- /* Fire off snapshot notification */
- size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
- sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
- uint64_t timestamp_now = mach_absolute_time();
- memorystatus_jetsam_snapshot->notification_time = timestamp_now;
- if (memorystatus_jetsam_snapshot_last_timestamp == 0 ||
- timestamp_now > memorystatus_jetsam_snapshot_last_timestamp + memorystatus_jetsam_snapshot_timeout) {
- int ret = memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
- if (!ret) {
- proc_list_lock();
- memorystatus_jetsam_snapshot_last_timestamp = timestamp_now;
- proc_list_unlock();
- }
+ os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMTHRASHING);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_on_VM_thrashing -- sync: failed to allocate jetsam reason\n");
}
+
+ return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMThrashing, jetsam_reason);
}
-
- return res;
}
+#if CONFIG_JETSAM
boolean_t
memorystatus_kill_on_VM_page_shortage(boolean_t async) {
if (async) {
- return memorystatus_kill_process_async(-1, kMemorystatusKilledVMPageShortage);
- } else {
- return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMPageShortage);
- }
-}
-
-boolean_t
-memorystatus_kill_on_VM_thrashing(boolean_t async) {
- if (async) {
- return memorystatus_kill_process_async(-1, kMemorystatusKilledVMThrashing);
+ return memorystatus_kill_process_async(-1, kMemorystatusKilledVMPageShortage);
} else {
- return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMThrashing);
+ os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_VMPAGESHORTAGE);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_on_VM_page_shortage -- sync: failed to allocate jetsam reason\n");
+ }
+
+ return memorystatus_kill_process_sync(-1, kMemorystatusKilledVMPageShortage, jetsam_reason);
}
}
boolean_t
memorystatus_kill_on_FC_thrashing(boolean_t async) {
+
+
if (async) {
return memorystatus_kill_process_async(-1, kMemorystatusKilledFCThrashing);
} else {
- return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing);
+ os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_MEMORY_FCTHRASHING);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_on_FC_thrashing -- sync: failed to allocate jetsam reason\n");
+ }
+
+ return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing, jetsam_reason);
}
}
boolean_t
memorystatus_kill_on_vnode_limit(void) {
- return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes);
+ os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_VNODE);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_on_vnode_limit: failed to allocate jetsam reason\n");
+ }
+
+ return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes, jetsam_reason);
}
#endif /* CONFIG_JETSAM */
+boolean_t
+memorystatus_kill_on_zone_map_exhaustion(pid_t pid) {
+ boolean_t res = FALSE;
+ if (pid == -1) {
+ res = memorystatus_kill_process_async(-1, kMemorystatusKilledZoneMapExhaustion);
+ } else {
+ os_reason_t jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_ZONE_MAP_EXHAUSTION);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_kill_on_zone_map_exhaustion: failed to allocate jetsam reason\n");
+ }
+
+ res = memorystatus_kill_process_sync(pid, kMemorystatusKilledZoneMapExhaustion, jetsam_reason);
+ }
+ return res;
+}
+
#if CONFIG_FREEZE
__private_extern__ void
goto exit;
}
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
unsigned int avail_swap_space = 0; /* in pages. */
- if (DEFAULT_FREEZER_IS_ACTIVE) {
- /*
- * Freezer backed by default pager and swap file(s).
- */
- avail_swap_space = default_pager_swap_pages_free();
- } else {
- /*
- * Freezer backed by the compressor and swap file(s)
- * while will hold compressed data.
- */
- avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
- }
+ /*
+ * Freezer backed by the compressor and swap file(s)
+ * while will hold compressed data.
+ */
+ avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
ret = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
+ DTRACE_MEMORYSTATUS6(memorystatus_freeze, proc_t, p, unsigned int, memorystatus_available_pages, boolean_t, purgeable, unsigned int, wired, uint32_t, clean, uint32_t, dirty);
+
MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_process_sync: task_freeze %s for pid %d [%s] - "
- "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, shared %d, free swap: %d\n",
- (ret == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (p->p_comm ? p->p_comm : "(unknown)"),
- memorystatus_available_pages, purgeable, wired, clean, dirty, shared, default_pager_swap_pages_free());
+ "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, max_pages %d, shared %d\n",
+ (ret == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (*p->p_name ? p->p_name : "(unknown)"),
+ memorystatus_available_pages, purgeable, wired, clean, dirty, max_pages, shared);
proc_list_lock();
p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
p->p_memstat_state |= (P_MEMSTAT_FROZEN | (shared ? 0: P_MEMSTAT_NORECLAIM));
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
/* Update stats */
for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
throttle_intervals[i].pageouts += dirty;
continue; // with lock held
}
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
/* Ensure there's enough free space to freeze this process. */
unsigned int avail_swap_space = 0; /* in pages. */
- if (DEFAULT_FREEZER_IS_ACTIVE) {
- /*
- * Freezer backed by default pager and swap file(s).
- */
- avail_swap_space = default_pager_swap_pages_free();
- } else {
- /*
- * Freezer backed by the compressor and swap file(s)
- * while will hold compressed data.
- */
- avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
- }
+ /*
+ * Freezer backed by the compressor and swap file(s)
+ * while will hold compressed data.
+ */
+ avail_swap_space = vm_swap_get_free_space() / PAGE_SIZE_64;
max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
kr = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
MEMORYSTATUS_DEBUG(1, "memorystatus_freeze_top_process: task_freeze %s for pid %d [%s] - "
- "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, shared %d, free swap: %d\n",
- (kr == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (p->p_comm ? p->p_comm : "(unknown)"),
- memorystatus_available_pages, purgeable, wired, clean, dirty, shared, default_pager_swap_pages_free());
+ "memorystatus_pages: %d, purgeable: %d, wired: %d, clean: %d, dirty: %d, max_pages %d, shared %d\n",
+ (kr == KERN_SUCCESS) ? "SUCCEEDED" : "FAILED", aPid, (*p->p_name ? p->p_name : "(unknown)"),
+ memorystatus_available_pages, purgeable, wired, clean, dirty, max_pages, shared);
proc_list_lock();
p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
memorystatus_frozen_count++;
p->p_memstat_state |= (P_MEMSTAT_FROZEN | (shared ? 0: P_MEMSTAT_NORECLAIM));
-
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+
+ if (VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
/* Update stats */
for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
throttle_intervals[i].pageouts += dirty;
if (!memorystatus_can_freeze_processes()) {
return FALSE;
}
+ assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
- if (COMPRESSED_PAGER_IS_SWAPLESS || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS) {
+ if ( !VM_CONFIG_FREEZER_SWAP_IS_ACTIVE) {
/*
* In-core compressor used for freezing WITHOUT on-disk swap support.
*/
-
if (vm_compressor_low_on_space()) {
if (*memorystatus_freeze_swap_low) {
*memorystatus_freeze_swap_low = TRUE;
} else {
/*
* Freezing WITH on-disk swap support.
+ *
+ * In-core compressor fronts the swap.
*/
-
- if (DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
- /*
- * In-core compressor fronts the swap.
- */
- if (vm_swap_low_on_space()) {
- if (*memorystatus_freeze_swap_low) {
- *memorystatus_freeze_swap_low = TRUE;
- }
-
- can_freeze = FALSE;
+ if (vm_swap_low_on_space()) {
+ if (*memorystatus_freeze_swap_low) {
+ *memorystatus_freeze_swap_low = TRUE;
}
- } else if (DEFAULT_FREEZER_IS_ACTIVE) {
- /*
- * Legacy freeze mode with no compressor support.
- */
- if (default_pager_swap_pages_free() < memorystatus_freeze_pages_min) {
- if (*memorystatus_freeze_swap_low) {
- *memorystatus_freeze_swap_low = TRUE;
- }
-
- can_freeze = FALSE;
- }
- } else {
- panic("Not a valid freeze configuration.\n");
+ can_freeze = FALSE;
}
+
}
return can_freeze;
thread_block((thread_continue_t) memorystatus_freeze_thread);
}
+static int
+sysctl_memorystatus_do_fastwake_warmup_all SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp, req, arg1, arg2)
+
+ /* Need to be root or have entitlement */
+ if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT)) {
+ return EPERM;
+ }
+
+ if (memorystatus_freeze_enabled == FALSE) {
+ return ENOTSUP;
+ }
+
+ do_fastwake_warmup_all();
+
+ return 0;
+}
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_do_fastwake_warmup_all, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
+ 0, 0, &sysctl_memorystatus_do_fastwake_warmup_all, "I", "");
+
#endif /* CONFIG_FREEZE */
#if VM_PRESSURE_EVENTS
memorystatus_send_note(int event_code, void *data, size_t data_length) {
int ret;
struct kev_msg ev_msg;
-
+
ev_msg.vendor_code = KEV_VENDOR_APPLE;
ev_msg.kev_class = KEV_SYSTEM_CLASS;
ev_msg.kev_subclass = KEV_MEMORYSTATUS_SUBCLASS;
}
boolean_t
-memorystatus_warn_process(pid_t pid, boolean_t critical) {
+memorystatus_warn_process(pid_t pid, __unused boolean_t is_active, __unused boolean_t is_fatal, boolean_t limit_exceeded) {
boolean_t ret = FALSE;
boolean_t found_knote = FALSE;
struct knote *kn = NULL;
+ int send_knote_count = 0;
/*
* See comment in sysctl_memorystatus_vm_pressure_send.
memorystatus_klist_lock();
SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
- proc_t knote_proc = kn->kn_kq->kq_p;
+ proc_t knote_proc = knote_get_kq(kn)->kq_p;
pid_t knote_pid = knote_proc->p_pid;
if (knote_pid == pid) {
* system pressure snapshot evaluation in
* filt_memorystatus().
*/
-
- if (critical) {
- if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
+
+#if CONFIG_EMBEDDED
+ if (!limit_exceeded) {
+ /*
+ * Intentionally set either the unambiguous limit warning,
+ * the system-wide critical or the system-wide warning
+ * notification bit.
+ */
+
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
+ found_knote = TRUE;
+ send_knote_count++;
+ } else if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
+ found_knote = TRUE;
+ send_knote_count++;
} else if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ found_knote = TRUE;
+ send_knote_count++;
}
} else {
- if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
- kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ /*
+ * Send this notification when a process has exceeded a soft limit.
+ */
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
+ found_knote = TRUE;
+ send_knote_count++;
}
}
+#else /* CONFIG_EMBEDDED */
+ if (!limit_exceeded) {
- found_knote = TRUE;
+ /*
+ * Processes on desktop are not expecting to handle a system-wide
+ * critical or system-wide warning notification from this path.
+ * Intentionally set only the unambiguous limit warning here.
+ *
+ * If the limit is soft, however, limit this to one notification per
+ * active/inactive limit (per each registered listener).
+ */
+
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
+ found_knote=TRUE;
+ if (!is_fatal) {
+ /*
+ * Restrict proc_limit_warn notifications when
+ * non-fatal (soft) limit is at play.
+ */
+ if (is_active) {
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE) {
+ /*
+ * Mark this knote for delivery.
+ */
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
+ /*
+ * And suppress it from future notifications.
+ */
+ kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
+ send_knote_count++;
+ }
+ } else {
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE) {
+ /*
+ * Mark this knote for delivery.
+ */
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
+ /*
+ * And suppress it from future notifications.
+ */
+ kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
+ send_knote_count++;
+ }
+ }
+ } else {
+ /*
+ * No restriction on proc_limit_warn notifications when
+ * fatal (hard) limit is at play.
+ */
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
+ send_knote_count++;
+ }
+ }
+ } else {
+ /*
+ * Send this notification when a process has exceeded a soft limit,
+ */
+
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
+ found_knote = TRUE;
+ if (!is_fatal) {
+ /*
+ * Restrict critical notifications for soft limits.
+ */
+
+ if (is_active) {
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE) {
+ /*
+ * Suppress future proc_limit_critical notifications
+ * for the active soft limit.
+ */
+ kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
+ send_knote_count++;
+
+ }
+ } else {
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE) {
+ /*
+ * Suppress future proc_limit_critical_notifications
+ * for the inactive soft limit.
+ */
+ kn->kn_sfflags &= ~NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
+ send_knote_count++;
+ }
+ }
+ } else {
+ /*
+ * We should never be trying to send a critical notification for
+ * a hard limit... the process would be killed before it could be
+ * received.
+ */
+ panic("Caught sending pid %d a critical warning for a fatal limit.\n", pid);
+ }
+ }
+ }
+#endif /* CONFIG_EMBEDDED */
}
}
if (found_knote) {
- KNOTE(&memorystatus_klist, 0);
- ret = TRUE;
- } else {
- if (vm_dispatch_pressure_note_to_pid(pid, FALSE) == 0) {
- ret = TRUE;
+ if (send_knote_count > 0) {
+ KNOTE(&memorystatus_klist, 0);
}
+ ret = TRUE;
}
memorystatus_klist_unlock();
return ((p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND) ||
(p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND_SUPPORT));
}
+
+/*
+ * This is meant for stackshot and kperf -- it does not take the proc_list_lock
+ * to access the p_memstat_dirty field.
+ */
+boolean_t
+memorystatus_proc_is_dirty_unsafe(void *v)
+{
+ if (!v) {
+ return FALSE;
+ }
+ proc_t p = (proc_t)v;
+ return (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) != 0;
+}
+
#endif /* CONFIG_MEMORYSTATUS */
/*
boolean_t memorystatus_manual_testing_on = FALSE;
vm_pressure_level_t memorystatus_manual_testing_level = kVMPressureNormal;
-extern struct knote *
-vm_pressure_select_optimal_candidate_to_notify(struct klist *, int, boolean_t);
+extern struct knote *
+vm_pressure_select_optimal_candidate_to_notify(struct klist *, int, boolean_t);
+
+/*
+ * This value is the threshold that a process must meet to be considered for scavenging.
+ */
+#if CONFIG_EMBEDDED
+#define VM_PRESSURE_MINIMUM_RSIZE 1 /* MB */
+#else /* CONFIG_EMBEDDED */
+#define VM_PRESSURE_MINIMUM_RSIZE 10 /* MB */
+#endif /* CONFIG_EMBEDDED */
+
+#define VM_PRESSURE_NOTIFY_WAIT_PERIOD 10000 /* milliseconds */
+
+#if DEBUG
+#define VM_PRESSURE_DEBUG(cond, format, ...) \
+do { \
+ if (cond) { printf(format, ##__VA_ARGS__); } \
+} while(0)
+#else
+#define VM_PRESSURE_DEBUG(cond, format, ...)
+#endif
+
+#define INTER_NOTIFICATION_DELAY (250000) /* .25 second */
+
+void memorystatus_on_pageout_scan_end(void) {
+ /* No-op */
+}
+
+/*
+ * kn_max - knote
+ *
+ * knote_pressure_level - to check if the knote is registered for this notification level.
+ *
+ * task - task whose bits we'll be modifying
+ *
+ * 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.
+ *
+ * pressure_level_to_set - the task is about to be notified of this new level. Update the task's bit notification information appropriately.
+ *
+ */
+
+boolean_t
+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)
+{
+ if (kn_max->kn_sfflags & knote_pressure_level) {
+
+ if (pressure_level_to_clear && task_has_been_notified(task, pressure_level_to_clear) == TRUE) {
+
+ task_clear_has_been_notified(task, pressure_level_to_clear);
+ }
+
+ task_mark_has_been_notified(task, pressure_level_to_set);
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
+void
+memorystatus_klist_reset_all_for_level(vm_pressure_level_t pressure_level_to_clear)
+{
+ struct knote *kn = NULL;
+
+ memorystatus_klist_lock();
+ SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
+
+ proc_t p = PROC_NULL;
+ struct task* t = TASK_NULL;
+
+ p = knote_get_kq(kn)->kq_p;
+ proc_list_lock();
+ if (p != proc_ref_locked(p)) {
+ p = PROC_NULL;
+ proc_list_unlock();
+ continue;
+ }
+ proc_list_unlock();
+
+ t = (struct task *)(p->task);
+
+ task_clear_has_been_notified(t, pressure_level_to_clear);
+
+ proc_rele(p);
+ }
+
+ memorystatus_klist_unlock();
+}
+
+extern kern_return_t vm_pressure_notify_dispatch_vm_clients(boolean_t target_foreground_process);
+
+struct knote *
+vm_pressure_select_optimal_candidate_to_notify(struct klist *candidate_list, int level, boolean_t target_foreground_process);
+
+/*
+ * Used by the vm_pressure_thread which is
+ * signalled from within vm_pageout_scan().
+ */
+static void vm_dispatch_memory_pressure(void);
+void consider_vm_pressure_events(void);
+
+void consider_vm_pressure_events(void)
+{
+ vm_dispatch_memory_pressure();
+}
+static void vm_dispatch_memory_pressure(void)
+{
+ memorystatus_update_vm_pressure(FALSE);
+}
+
+extern vm_pressure_level_t
+convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
+
+struct knote *
+vm_pressure_select_optimal_candidate_to_notify(struct klist *candidate_list, int level, boolean_t target_foreground_process)
+{
+ struct knote *kn = NULL, *kn_max = NULL;
+ uint64_t resident_max = 0; /* MB */
+ struct timeval curr_tstamp = {0, 0};
+ int elapsed_msecs = 0;
+ int selected_task_importance = 0;
+ static int pressure_snapshot = -1;
+ boolean_t pressure_increase = FALSE;
+
+ if (pressure_snapshot == -1) {
+ /*
+ * Initial snapshot.
+ */
+ pressure_snapshot = level;
+ pressure_increase = TRUE;
+ } else {
+
+ if (level && (level >= pressure_snapshot)) {
+ pressure_increase = TRUE;
+ } else {
+ pressure_increase = FALSE;
+ }
+
+ pressure_snapshot = level;
+ }
+
+ if (pressure_increase == TRUE) {
+ /*
+ * We'll start by considering the largest
+ * unimportant task in our list.
+ */
+ selected_task_importance = INT_MAX;
+ } else {
+ /*
+ * We'll start by considering the largest
+ * important task in our list.
+ */
+ selected_task_importance = 0;
+ }
+
+ microuptime(&curr_tstamp);
+
+ SLIST_FOREACH(kn, candidate_list, kn_selnext) {
+
+ uint64_t resident_size = 0; /* MB */
+ proc_t p = PROC_NULL;
+ struct task* t = TASK_NULL;
+ int curr_task_importance = 0;
+ boolean_t consider_knote = FALSE;
+ boolean_t privileged_listener = FALSE;
+
+ p = knote_get_kq(kn)->kq_p;
+ proc_list_lock();
+ if (p != proc_ref_locked(p)) {
+ p = PROC_NULL;
+ proc_list_unlock();
+ continue;
+ }
+ proc_list_unlock();
+
+#if CONFIG_MEMORYSTATUS
+ if (target_foreground_process == TRUE && !memorystatus_is_foreground_locked(p)) {
+ /*
+ * Skip process not marked foreground.
+ */
+ proc_rele(p);
+ continue;
+ }
+#endif /* CONFIG_MEMORYSTATUS */
+
+ t = (struct task *)(p->task);
+
+ timevalsub(&curr_tstamp, &p->vm_pressure_last_notify_tstamp);
+ elapsed_msecs = curr_tstamp.tv_sec * 1000 + curr_tstamp.tv_usec / 1000;
+
+ vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(level);
+
+ if ((kn->kn_sfflags & dispatch_level) == 0) {
+ proc_rele(p);
+ continue;
+ }
+
+#if CONFIG_MEMORYSTATUS
+ if (target_foreground_process == FALSE && !memorystatus_bg_pressure_eligible(p)) {
+ VM_PRESSURE_DEBUG(1, "[vm_pressure] skipping process %d\n", p->p_pid);
+ proc_rele(p);
+ continue;
+ }
+#endif /* CONFIG_MEMORYSTATUS */
+
+#if CONFIG_EMBEDDED
+ curr_task_importance = p->p_memstat_effectivepriority;
+#else /* CONFIG_EMBEDDED */
+ curr_task_importance = task_importance_estimate(t);
+#endif /* CONFIG_EMBEDDED */
+
+ /*
+ * Privileged listeners are only considered in the multi-level pressure scheme
+ * AND only if the pressure is increasing.
+ */
+ if (level > 0) {
+
+ if (task_has_been_notified(t, level) == FALSE) {
+
+ /*
+ * Is this a privileged listener?
+ */
+ if (task_low_mem_privileged_listener(t, FALSE, &privileged_listener) == 0) {
+
+ if (privileged_listener) {
+ kn_max = kn;
+ proc_rele(p);
+ goto done_scanning;
+ }
+ }
+ } else {
+ proc_rele(p);
+ continue;
+ }
+ } else if (level == 0) {
-extern
-kern_return_t vm_pressure_notification_without_levels(boolean_t);
+ /*
+ * Task wasn't notified when the pressure was increasing and so
+ * no need to notify it that the pressure is decreasing.
+ */
+ if ((task_has_been_notified(t, kVMPressureWarning) == FALSE) && (task_has_been_notified(t, kVMPressureCritical) == FALSE)) {
+ proc_rele(p);
+ continue;
+ }
+ }
-extern void vm_pressure_klist_lock(void);
-extern void vm_pressure_klist_unlock(void);
+ /*
+ * We don't want a small process to block large processes from
+ * being notified again. <rdar://problem/7955532>
+ */
+ resident_size = (get_task_phys_footprint(t))/(1024*1024ULL); /* MB */
-extern void vm_reset_active_list(void);
+ if (resident_size >= VM_PRESSURE_MINIMUM_RSIZE) {
-extern void delay(int);
+ if (level > 0) {
+ /*
+ * Warning or Critical Pressure.
+ */
+ if (pressure_increase) {
+ if ((curr_task_importance < selected_task_importance) ||
+ ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
-#define INTER_NOTIFICATION_DELAY (250000) /* .25 second */
+ /*
+ * We have found a candidate process which is:
+ * a) at a lower importance than the current selected process
+ * OR
+ * b) has importance equal to that of the current selected process but is larger
+ */
-void memorystatus_on_pageout_scan_end(void) {
- /* No-op */
-}
+ consider_knote = TRUE;
+ }
+ } else {
+ if ((curr_task_importance > selected_task_importance) ||
+ ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
-/*
- * kn_max - knote
- *
- * knote_pressure_level - to check if the knote is registered for this notification level.
- *
- * task - task whose bits we'll be modifying
- *
- * 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.
- *
- * pressure_level_to_set - the task is about to be notified of this new level. Update the task's bit notification information appropriately.
- *
- */
+ /*
+ * We have found a candidate process which is:
+ * a) at a higher importance than the current selected process
+ * OR
+ * b) has importance equal to that of the current selected process but is larger
+ */
-boolean_t
-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)
-{
- if (kn_max->kn_sfflags & knote_pressure_level) {
+ consider_knote = TRUE;
+ }
+ }
+ } else if (level == 0) {
+ /*
+ * Pressure back to normal.
+ */
+ if ((curr_task_importance > selected_task_importance) ||
+ ((curr_task_importance == selected_task_importance) && (resident_size > resident_max))) {
- if (task_has_been_notified(task, pressure_level_to_clear) == TRUE) {
+ consider_knote = TRUE;
+ }
+ }
- task_clear_has_been_notified(task, pressure_level_to_clear);
- }
+ if (consider_knote) {
+ resident_max = resident_size;
+ kn_max = kn;
+ selected_task_importance = curr_task_importance;
+ consider_knote = FALSE; /* reset for the next candidate */
+ }
+ } else {
+ /* There was no candidate with enough resident memory to scavenge */
+ VM_PRESSURE_DEBUG(0, "[vm_pressure] threshold failed for pid %d with %llu resident...\n", p->p_pid, resident_size);
+ }
+ proc_rele(p);
+ }
- task_mark_has_been_notified(task, pressure_level_to_set);
- return TRUE;
+done_scanning:
+ if (kn_max) {
+ 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);
+ 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);
}
- return FALSE;
+ return kn_max;
}
-extern kern_return_t vm_pressure_notify_dispatch_vm_clients(boolean_t target_foreground_process);
-
#define VM_PRESSURE_DECREASED_SMOOTHING_PERIOD 5000 /* milliseconds */
+#define WARNING_NOTIFICATION_RESTING_PERIOD 25 /* seconds */
+#define CRITICAL_NOTIFICATION_RESTING_PERIOD 25 /* seconds */
+
+uint64_t next_warning_notification_sent_at_ts = 0;
+uint64_t next_critical_notification_sent_at_ts = 0;
kern_return_t
memorystatus_update_vm_pressure(boolean_t target_foreground_process)
struct timeval smoothing_window_start_tstamp = {0, 0};
struct timeval curr_tstamp = {0, 0};
int elapsed_msecs = 0;
+ uint64_t curr_ts = mach_absolute_time();
#if !CONFIG_JETSAM
#define MAX_IDLE_KILLS 100 /* limit the number of idle kills allowed */
}
#endif /* !CONFIG_JETSAM */
+ if (level_snapshot != kVMPressureNormal) {
+
+ /*
+ * Check to see if we are still in the 'resting' period
+ * after having notified all clients interested in
+ * a particular pressure level.
+ */
+
+ level_snapshot = memorystatus_vm_pressure_level;
+
+ if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
+
+ if (next_warning_notification_sent_at_ts) {
+ if (curr_ts < next_warning_notification_sent_at_ts) {
+ delay(INTER_NOTIFICATION_DELAY * 4 /* 1 sec */);
+ return KERN_SUCCESS;
+ }
+
+ next_warning_notification_sent_at_ts = 0;
+ memorystatus_klist_reset_all_for_level(kVMPressureWarning);
+ }
+ } else if (level_snapshot == kVMPressureCritical) {
+
+ if (next_critical_notification_sent_at_ts) {
+ if (curr_ts < next_critical_notification_sent_at_ts) {
+ delay(INTER_NOTIFICATION_DELAY * 4 /* 1 sec */);
+ return KERN_SUCCESS;
+ }
+ next_critical_notification_sent_at_ts = 0;
+ memorystatus_klist_reset_all_for_level(kVMPressureCritical);
+ }
+ }
+ }
+
while (1) {
/*
/*
* No more level-based clients to notify.
- * Try the non-level based notification clients.
- *
- * However, these non-level clients don't understand
- * the "return-to-normal" notification.
- *
- * So don't consider them for those notifications. Just
- * return instead.
*
+ * Start the 'resting' window within which clients will not be re-notified.
*/
if (level_snapshot != kVMPressureNormal) {
- goto try_dispatch_vm_clients;
- } else {
- return KERN_FAILURE;
- }
+ if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
+ nanoseconds_to_absolutetime(WARNING_NOTIFICATION_RESTING_PERIOD * NSEC_PER_SEC, &curr_ts);
+
+ /* Next warning notification (if nothing changes) won't be sent before...*/
+ next_warning_notification_sent_at_ts = mach_absolute_time() + curr_ts;
+ }
+
+ if (level_snapshot == kVMPressureCritical) {
+ nanoseconds_to_absolutetime(CRITICAL_NOTIFICATION_RESTING_PERIOD * NSEC_PER_SEC, &curr_ts);
+
+ /* Next critical notification (if nothing changes) won't be sent before...*/
+ next_critical_notification_sent_at_ts = mach_absolute_time() + curr_ts;
+ }
+ }
+ return KERN_FAILURE;
}
- target_proc = kn_max->kn_kq->kq_p;
+ target_proc = knote_get_kq(kn_max)->kq_p;
proc_list_lock();
if (target_proc != proc_ref_locked(target_proc)) {
if (level_snapshot == kVMPressureWarning || level_snapshot == kVMPressureUrgent) {
- if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_WARN, task, kVMPressureCritical, kVMPressureWarning) == TRUE) {
+ if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_WARN, task, 0, kVMPressureWarning) == TRUE) {
found_candidate = TRUE;
}
} else {
if (level_snapshot == kVMPressureCritical) {
- if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_CRITICAL, task, kVMPressureWarning, kVMPressureCritical) == TRUE) {
+ if (is_knote_registered_modify_task_pressure_bits(kn_max, NOTE_MEMORYSTATUS_PRESSURE_CRITICAL, task, 0, kVMPressureCritical) == TRUE) {
found_candidate = TRUE;
}
}
}
SLIST_FOREACH_SAFE(kn_cur, &memorystatus_klist, kn_selnext, kn_temp) {
- proc_t knote_proc = kn_cur->kn_kq->kq_p;
- pid_t knote_pid = knote_proc->p_pid;
- if (knote_pid == target_pid) {
- KNOTE_DETACH(&memorystatus_klist, kn_cur);
- KNOTE_ATTACH(&dispatch_klist, kn_cur);
+
+ int knote_pressure_level = convert_internal_pressure_level_to_dispatch_level(level_snapshot);
+
+ if (is_knote_registered_modify_task_pressure_bits(kn_cur, knote_pressure_level, task, 0, level_snapshot) == TRUE) {
+ proc_t knote_proc = knote_get_kq(kn_cur)->kq_p;
+ pid_t knote_pid = knote_proc->p_pid;
+ if (knote_pid == target_pid) {
+ KNOTE_DETACH(&memorystatus_klist, kn_cur);
+ KNOTE_ATTACH(&dispatch_klist, kn_cur);
+ }
}
}
break;
}
-try_dispatch_vm_clients:
- if (kn_max == NULL && level_snapshot != kVMPressureNormal) {
- /*
- * We will exit this loop when we are done with
- * notification clients (level and non-level based).
- */
- if ((vm_pressure_notify_dispatch_vm_clients(target_foreground_process) == KERN_FAILURE) && (kn_max == NULL)) {
- /*
- * kn_max == NULL i.e. we didn't find any eligible clients for the level-based notifications
- * AND
- * we have failed to find any eligible clients for the non-level based notifications too.
- * So, we are done.
- */
-
- return KERN_FAILURE;
- }
- }
-
- /*
- * LD: This block of code below used to be invoked in the older memory notification scheme on embedded everytime
- * a process was sent a memory pressure notification. The "memorystatus_klist" list was used to hold these
- * privileged listeners. But now we have moved to the newer scheme and are trying to move away from the extra
- * notifications. So the code is here in case we break compat. and need to send out notifications to the privileged
- * apps.
- */
-#if 0
-#endif /* 0 */
-
if (memorystatus_manual_testing_on == TRUE) {
/*
* Testing out the pressure notification scheme.
return KERN_SUCCESS;
}
-vm_pressure_level_t
-convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t);
-
vm_pressure_level_t
convert_internal_pressure_level_to_dispatch_level(vm_pressure_level_t internal_pressure_level)
{
sysctl_memorystatus_vm_pressure_level SYSCTL_HANDLER_ARGS
{
#pragma unused(arg1, arg2, oidp)
+#if CONFIG_EMBEDDED
+ int error = 0;
+
+ error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
+ if (error)
+ return (error);
+
+#endif /* CONFIG_EMBEDDED */
vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(memorystatus_vm_pressure_level);
return SYSCTL_OUT(req, &dispatch_level, sizeof(dispatch_level));
if (pressure_level == NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
memorystatus_manual_testing_on = FALSE;
-
- vm_pressure_klist_lock();
- vm_reset_active_list();
- vm_pressure_klist_unlock();
- } else {
-
- vm_pressure_klist_lock();
- vm_pressure_notification_without_levels(FALSE);
- vm_pressure_klist_unlock();
}
return 0;
list_entry->pid = p->p_pid;
list_entry->priority = p->p_memstat_effectivepriority;
list_entry->user_data = p->p_memstat_userdata;
-#if LEGACY_HIWATER
-
- /*
- * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
- * Background limits are described via the inactive limit slots.
- * So, here, the cached limit should always be valid.
- */
if (p->p_memstat_memlimit <= 0) {
task_get_phys_footprint_limit(p->task, &list_entry->limit);
} else {
list_entry->limit = p->p_memstat_memlimit;
}
-#else
- task_get_phys_footprint_limit(p->task, &list_entry->limit);
-#endif
+
list_entry->state = memorystatus_build_state(p);
list_entry++;
}
static int
-memorystatus_cmd_get_priority_list(user_addr_t buffer, size_t buffer_size, int32_t *retval) {
- int error = EINVAL;
+memorystatus_get_priority_pid(pid_t pid, user_addr_t buffer, size_t buffer_size) {
+ int error = 0;
+ memorystatus_priority_entry_t mp_entry;
+
+ /* Validate inputs */
+ if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_entry_t))) {
+ return EINVAL;
+ }
+
+ proc_t p = proc_find(pid);
+ if (!p) {
+ return ESRCH;
+ }
+
+ memset (&mp_entry, 0, sizeof(memorystatus_priority_entry_t));
+
+ mp_entry.pid = p->p_pid;
+ mp_entry.priority = p->p_memstat_effectivepriority;
+ mp_entry.user_data = p->p_memstat_userdata;
+ if (p->p_memstat_memlimit <= 0) {
+ task_get_phys_footprint_limit(p->task, &mp_entry.limit);
+ } else {
+ mp_entry.limit = p->p_memstat_memlimit;
+ }
+ mp_entry.state = memorystatus_build_state(p);
+
+ proc_rele(p);
+
+ error = copyout(&mp_entry, buffer, buffer_size);
+
+ return (error);
+}
+
+static int
+memorystatus_cmd_get_priority_list(pid_t pid, user_addr_t buffer, size_t buffer_size, int32_t *retval) {
+ int error = 0;
boolean_t size_only;
- memorystatus_priority_entry_t *list = NULL;
size_t list_size;
+
+ /*
+ * When a non-zero pid is provided, the 'list' has only one entry.
+ */
size_only = ((buffer == USER_ADDR_NULL) ? TRUE: FALSE);
-
- error = memorystatus_get_priority_list(&list, &buffer_size, &list_size, size_only);
- if (error) {
- goto out;
- }
- if (!size_only) {
- error = copyout(list, buffer, list_size);
+ if (pid != 0) {
+ list_size = sizeof(memorystatus_priority_entry_t) * 1;
+ if (!size_only) {
+ error = memorystatus_get_priority_pid(pid, buffer, buffer_size);
+ }
+ } else {
+ memorystatus_priority_entry_t *list = NULL;
+ error = memorystatus_get_priority_list(&list, &buffer_size, &list_size, size_only);
+
+ if (error == 0) {
+ if (!size_only) {
+ error = copyout(list, buffer, list_size);
+ }
+ }
+
+ if (list) {
+ kfree(list, buffer_size);
+ }
}
-
+
if (error == 0) {
*retval = list_size;
}
-out:
-
- if (list) {
- kfree(list, buffer_size);
- }
- return error;
+ return (error);
}
-#if CONFIG_JETSAM
-
static void
memorystatus_clear_errors(void)
{
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_CLEAR_ERRORS) | DBG_FUNC_END, 0, 0, 0, 0, 0);
}
+#if CONFIG_JETSAM
static void
memorystatus_update_levels_locked(boolean_t critical_only) {
/*
* If there's an entry in the first bucket, we have idle processes.
*/
+
memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
if (first_bucket->count) {
memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
}
}
-#endif
-
+#endif /* DEBUG || DEVELOPMENT */
+
+ if (memorystatus_jetsam_policy & kPolicyMoreFree) {
+ memorystatus_available_pages_critical += memorystatus_policy_more_free_offset_pages;
+ }
+
if (critical_only) {
return;
}
#endif
}
+
+static int
+sysctl_kern_memorystatus_policy_more_free SYSCTL_HANDLER_ARGS
+{
+#pragma unused(arg1, arg2, oidp)
+ int error = 0, more_free = 0;
+
+ /*
+ * TODO: Enable this privilege check?
+ *
+ * error = priv_check_cred(kauth_cred_get(), PRIV_VM_JETSAM, 0);
+ * if (error)
+ * return (error);
+ */
+
+ error = sysctl_handle_int(oidp, &more_free, 0, req);
+ if (error || !req->newptr)
+ return (error);
+
+ if ((more_free && ((memorystatus_jetsam_policy & kPolicyMoreFree) == kPolicyMoreFree)) ||
+ (!more_free && ((memorystatus_jetsam_policy & kPolicyMoreFree) == 0))) {
+
+ /*
+ * No change in state.
+ */
+ return 0;
+ }
+
+ proc_list_lock();
+
+ if (more_free) {
+ memorystatus_jetsam_policy |= kPolicyMoreFree;
+ } else {
+ memorystatus_jetsam_policy &= ~kPolicyMoreFree;
+ }
+
+ memorystatus_update_levels_locked(TRUE);
+
+ proc_list_unlock();
+
+ return 0;
+}
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_policy_more_free, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
+ 0, 0, &sysctl_kern_memorystatus_policy_more_free, "I", "");
+
+#endif /* CONFIG_JETSAM */
+
/*
* Get the at_boot snapshot
*/
/*
* The jetsam snapshot is never freed, its count is simply reset.
*/
- snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
-
proc_list_lock();
+ snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
memorystatus_jetsam_snapshot_last_timestamp = 0;
proc_list_unlock();
}
if (entries[i].priority == -1) {
/* Use as shorthand for default priority */
entries[i].priority = JETSAM_PRIORITY_DEFAULT;
- } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_DEFERRED) {
- /* JETSAM_PRIORITY_IDLE_DEFERRED is reserved for internal use;
+ } else if ((entries[i].priority == system_procs_aging_band) || (entries[i].priority == applications_aging_band)) {
+ /* Both the aging bands are reserved for internal use;
* if requested, adjust to JETSAM_PRIORITY_IDLE. */
entries[i].priority = JETSAM_PRIORITY_IDLE;
} else if (entries[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
}
/*
- * Take appropriate steps if moving proc out of the
- * JETSAM_PRIORITY_IDLE_DEFERRED band.
+ * Take appropriate steps if moving proc out of
+ * either of the aging bands.
*/
- if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ if ((p->p_memstat_effectivepriority == system_procs_aging_band) || (p->p_memstat_effectivepriority == applications_aging_band)) {
memorystatus_invalidate_idle_demotion_locked(p, TRUE);
}
- memorystatus_update_priority_locked(p, new_priority, head_insert);
+ memorystatus_update_priority_locked(p, new_priority, head_insert, false);
}
proc_list_unlock();
return EPERM;
}
- error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, FALSE, FALSE, 0, 0, FALSE, FALSE, FALSE);
+ error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, FALSE, FALSE, 0, 0, FALSE, FALSE);
proc_rele(p);
}
return EINVAL;
}
- memset (&mmp_entry, 0, sizeof(memorystatus_memlimit_properties_t));
-
+ memset (&mmp_entry, 0, sizeof(memorystatus_memlimit_properties_t));
+
+ proc_t p = proc_find(pid);
+ if (!p) {
+ return ESRCH;
+ }
+
+ /*
+ * Get the active limit and attributes.
+ * No locks taken since we hold a reference to the proc.
+ */
+
+ if (p->p_memstat_memlimit_active > 0 ) {
+ mmp_entry.memlimit_active = p->p_memstat_memlimit_active;
+ } else {
+ task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_active);
+ }
+
+ if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) {
+ mmp_entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
+ }
+
+ /*
+ * Get the inactive limit and attributes
+ */
+ if (p->p_memstat_memlimit_inactive <= 0) {
+ task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_inactive);
+ } else {
+ mmp_entry.memlimit_inactive = p->p_memstat_memlimit_inactive;
+ }
+ if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) {
+ mmp_entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
+ }
+ proc_rele(p);
+
+ error = copyout(&mmp_entry, buffer, buffer_size);
+
+ return(error);
+}
+
+
+/*
+ * SPI for kbd - pr24956468
+ * This is a very simple snapshot that calculates how much a
+ * process's phys_footprint exceeds a specific memory limit.
+ * Only the inactive memory limit is supported for now.
+ * The delta is returned as bytes in excess or zero.
+ */
+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) {
+ int error = 0;
+ uint64_t footprint_in_bytes = 0;
+ uint64_t delta_in_bytes = 0;
+ int32_t memlimit_mb = 0;
+ uint64_t memlimit_bytes = 0;
+
+ /* Validate inputs */
+ if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(uint64_t)) || (flags != 0)) {
+ return EINVAL;
+ }
+
proc_t p = proc_find(pid);
if (!p) {
return ESRCH;
}
/*
- * Get the active limit and attributes.
+ * Get the inactive limit.
* No locks taken since we hold a reference to the proc.
*/
- if (p->p_memstat_memlimit_active > 0 ) {
- mmp_entry.memlimit_active = p->p_memstat_memlimit_active;
+ if (p->p_memstat_memlimit_inactive <= 0) {
+ task_convert_phys_footprint_limit(-1, &memlimit_mb);
} else {
- task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_active);
+ memlimit_mb = p->p_memstat_memlimit_inactive;
}
- if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL) {
- mmp_entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
- }
+ footprint_in_bytes = get_task_phys_footprint(p->task);
+
+ proc_rele(p);
+
+ memlimit_bytes = memlimit_mb * 1024 * 1024; /* MB to bytes */
/*
- * Get the inactive limit and attributes
+ * Computed delta always returns >= 0 bytes
*/
- if (p->p_memstat_memlimit_inactive <= 0) {
- task_convert_phys_footprint_limit(-1, &mmp_entry.memlimit_inactive);
- } else {
- mmp_entry.memlimit_inactive = p->p_memstat_memlimit_inactive;
- }
- if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL) {
- mmp_entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
+ if (footprint_in_bytes > memlimit_bytes) {
+ delta_in_bytes = footprint_in_bytes - memlimit_bytes;
}
- proc_rele(p);
- error = copyout(&mmp_entry, buffer, buffer_size);
+ error = copyout(&delta_in_bytes, buffer, sizeof(delta_in_bytes));
return(error);
}
* - so mapping is (active/fatal, inactive/fatal)
*/
+#if CONFIG_JETSAM
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) {
int error = 0;
error = memorystatus_set_memlimit_properties(pid, &entry);
return (error);
}
+#endif /* CONFIG_JETSAM */
static int
memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry) {
*/
if (memorystatus_highwater_enabled) {
- boolean_t trigger_exception;
- /*
- * No need to consider P_MEMSTAT_MEMLIMIT_BACKGROUND anymore.
- * Background limits are described via the inactive limit slots.
- */
+ boolean_t is_fatal;
+ boolean_t use_active;
if (proc_jetsam_state_is_active_locked(p) == TRUE) {
- CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_ACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = TRUE;
} else {
- CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ CACHE_INACTIVE_LIMITS_LOCKED(p, is_fatal);
+ use_active = FALSE;
}
/* Enforce the limit by writing to the ledgers */
- assert(trigger_exception == TRUE);
- 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;
+ error = (task_set_phys_footprint_limit_internal(p->task, ((p->p_memstat_memlimit > 0) ? p->p_memstat_memlimit : -1), NULL, use_active, is_fatal) == 0) ? 0 : EINVAL;
MEMORYSTATUS_DEBUG(3, "memorystatus_set_memlimit_properties: new limit on pid %d (%dMB %s) current priority (%d) dirty_state?=0x%x %s\n",
p->p_pid, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1),
(p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, p->p_memstat_dirty,
(p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
+ DTRACE_MEMORYSTATUS2(memorystatus_set_memlimit, proc_t, p, int32_t, (p->p_memstat_memlimit > 0 ? p->p_memstat_memlimit : -1));
}
proc_list_unlock();
return 0;
}
-/*
- * Description:
- * Evaluates active vs. inactive process state.
- * Processes that opt into dirty tracking are evaluated
- * based on clean vs dirty state.
- * dirty ==> active
- * clean ==> inactive
- *
- * Process that do not opt into dirty tracking are
- * evalulated based on priority level.
- * Foreground or above ==> active
- * Below Foreground ==> inactive
- *
- * Return: TRUE if active
- * False if inactive
- */
-
-static boolean_t
-proc_jetsam_state_is_active_locked(proc_t p) {
-
- if (p->p_memstat_dirty & P_DIRTY_TRACK) {
- /*
- * process has opted into dirty tracking
- * active state is based on dirty vs. clean
- */
- if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
- /*
- * process is dirty
- * implies active state
- */
- return TRUE;
- } else {
- /*
- * process is clean
- * implies inactive state
- */
- return FALSE;
- }
- } else if (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND) {
- /*
- * process is Foreground or higher
- * implies active state
- */
- return TRUE;
- } else {
- /*
- * process found below Foreground
- * implies inactive state
- */
- return FALSE;
- }
-}
-
-#endif /* CONFIG_JETSAM */
-
int
memorystatus_control(struct proc *p __unused, struct memorystatus_control_args *args, int *ret) {
int error = EINVAL;
+ os_reason_t jetsam_reason = OS_REASON_NULL;
#if !CONFIG_JETSAM
#pragma unused(ret)
+ #pragma unused(jetsam_reason)
#endif
- /* Root only for now */
- if (!kauth_cred_issuser(kauth_cred_get())) {
+ /* Need to be root or have entitlement */
+ if (!kauth_cred_issuser(kauth_cred_get()) && !IOTaskHasEntitlement(current_task(), MEMORYSTATUS_ENTITLEMENT)) {
error = EPERM;
goto out;
}
-
- /* Sanity check */
- if (args->buffersize > MEMORYSTATUS_BUFFERSIZE_MAX) {
- error = EINVAL;
- goto out;
+
+ /*
+ * Sanity check.
+ * Do not enforce it for snapshots.
+ */
+ if (args->command != MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT) {
+ if (args->buffersize > MEMORYSTATUS_BUFFERSIZE_MAX) {
+ error = EINVAL;
+ goto out;
+ }
}
switch (args->command) {
case MEMORYSTATUS_CMD_GET_PRIORITY_LIST:
- error = memorystatus_cmd_get_priority_list(args->buffer, args->buffersize, ret);
+ error = memorystatus_cmd_get_priority_list(args->pid, args->buffer, args->buffersize, ret);
break;
-#if CONFIG_JETSAM
case MEMORYSTATUS_CMD_SET_PRIORITY_PROPERTIES:
error = memorystatus_cmd_set_priority_properties(args->pid, args->buffer, args->buffersize, ret);
break;
case MEMORYSTATUS_CMD_GET_MEMLIMIT_PROPERTIES:
error = memorystatus_cmd_get_memlimit_properties(args->pid, args->buffer, args->buffersize, ret);
break;
+ case MEMORYSTATUS_CMD_GET_MEMLIMIT_EXCESS:
+ error = memorystatus_cmd_get_memlimit_excess_np(args->pid, args->flags, args->buffer, args->buffersize, ret);
+ break;
case MEMORYSTATUS_CMD_GRP_SET_PROPERTIES:
error = memorystatus_cmd_grp_set_properties((int32_t)args->flags, args->buffer, args->buffersize, ret);
break;
case MEMORYSTATUS_CMD_GET_PRESSURE_STATUS:
error = memorystatus_cmd_get_pressure_status(ret);
break;
+#if CONFIG_JETSAM
case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
/*
* This call does not distinguish between active and inactive limits.
*/
error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
break;
+#endif /* CONFIG_JETSAM */
/* Test commands */
#if DEVELOPMENT || DEBUG
case MEMORYSTATUS_CMD_TEST_JETSAM:
- error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled) ? 0 : EINVAL;
+ jetsam_reason = os_reason_create(OS_REASON_JETSAM, JETSAM_REASON_GENERIC);
+ if (jetsam_reason == OS_REASON_NULL) {
+ printf("memorystatus_control: failed to allocate jetsam reason\n");
+ }
+
+ error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled, jetsam_reason) ? 0 : EINVAL;
break;
case MEMORYSTATUS_CMD_TEST_JETSAM_SORT:
error = memorystatus_cmd_test_jetsam_sort(args->pid, (int32_t)args->flags);
break;
+#if CONFIG_JETSAM
case MEMORYSTATUS_CMD_SET_JETSAM_PANIC_BITS:
error = memorystatus_cmd_set_panic_bits(args->buffer, args->buffersize);
break;
-#endif /* DEVELOPMENT || DEBUG */
#endif /* CONFIG_JETSAM */
+#else /* DEVELOPMENT || DEBUG */
+ #pragma unused(jetsam_reason)
+#endif /* DEVELOPMENT || DEBUG */
+ case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_ENABLE:
+ if (memorystatus_aggressive_jetsam_lenient_allowed == FALSE) {
+#if DEVELOPMENT || DEBUG
+ printf("Enabling Lenient Mode\n");
+#endif /* DEVELOPMENT || DEBUG */
+
+ memorystatus_aggressive_jetsam_lenient_allowed = TRUE;
+ memorystatus_aggressive_jetsam_lenient = TRUE;
+ error = 0;
+ }
+ break;
+ case MEMORYSTATUS_CMD_AGGRESSIVE_JETSAM_LENIENT_MODE_DISABLE:
+#if DEVELOPMENT || DEBUG
+ printf("Disabling Lenient mode\n");
+#endif /* DEVELOPMENT || DEBUG */
+ memorystatus_aggressive_jetsam_lenient_allowed = FALSE;
+ memorystatus_aggressive_jetsam_lenient = FALSE;
+ error = 0;
+ break;
case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE:
case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE:
error = memorystatus_low_mem_privileged_listener(args->command);
break;
+
+ case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_ENABLE:
+ case MEMORYSTATUS_CMD_ELEVATED_INACTIVEJETSAMPRIORITY_DISABLE:
+ error = memorystatus_update_inactive_jetsam_priority_band(args->pid, args->command, args->flags ? TRUE : FALSE);
+ break;
+
default:
break;
}
static int
-filt_memorystatusattach(struct knote *kn)
+filt_memorystatusattach(struct knote *kn, __unused struct kevent_internal_s *kev)
{
+ int error;
+
kn->kn_flags |= EV_CLEAR;
- return memorystatus_knote_register(kn);
+ error = memorystatus_knote_register(kn);
+ if (error) {
+ kn->kn_flags = EV_ERROR;
+ kn->kn_data = error;
+ }
+ return 0;
}
static void
kn->kn_fflags = NOTE_MEMORYSTATUS_LOW_SWAP;
}
break;
+
+ case kMemorystatusProcLimitWarn:
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_WARN;
+ }
+ break;
+
+ case kMemorystatusProcLimitCritical:
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL;
+ }
+ break;
+
default:
break;
}
}
-
+
+#if 0
+ if (kn->kn_fflags != 0) {
+ proc_t knote_proc = knote_get_kq(kn)->kq_p;
+ pid_t knote_pid = knote_proc->p_pid;
+
+ printf("filt_memorystatus: sending kn 0x%lx (event 0x%x) for pid (%d)\n",
+ (unsigned long)kn, kn->kn_fflags, knote_pid);
+ }
+#endif
+
return (kn->kn_fflags != 0);
}
+static int
+filt_memorystatustouch(struct knote *kn, struct kevent_internal_s *kev)
+{
+ int res;
+ int prev_kn_sfflags = 0;
+
+ memorystatus_klist_lock();
+
+ /*
+ * copy in new kevent settings
+ * (saving the "desired" data and fflags).
+ */
+
+ prev_kn_sfflags = kn->kn_sfflags;
+ kn->kn_sfflags = (kev->fflags & EVFILT_MEMORYSTATUS_ALL_MASK);
+
+#if !CONFIG_EMBEDDED
+ /*
+ * Only on desktop do we restrict notifications to
+ * one per active/inactive state (soft limits only).
+ */
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
+ /*
+ * Is there previous state to preserve?
+ */
+ if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
+ /*
+ * This knote was previously interested in proc_limit_warn,
+ * so yes, preserve previous state.
+ */
+ if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE) {
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
+ }
+ if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE) {
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
+ }
+ } else {
+ /*
+ * This knote was not previously interested in proc_limit_warn,
+ * but it is now. Set both states.
+ */
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
+ }
+ }
+
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
+ /*
+ * Is there previous state to preserve?
+ */
+ if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
+ /*
+ * This knote was previously interested in proc_limit_critical,
+ * so yes, preserve previous state.
+ */
+ if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE) {
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
+ }
+ if (prev_kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE) {
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
+ }
+ } else {
+ /*
+ * This knote was not previously interested in proc_limit_critical,
+ * but it is now. Set both states.
+ */
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
+ }
+ }
+#endif /* !CONFIG_EMBEDDED */
+
+ if ((kn->kn_status & KN_UDATA_SPECIFIC) == 0)
+ kn->kn_udata = kev->udata;
+
+ /*
+ * reset the output flags based on a
+ * combination of the old events and
+ * the new desired event list.
+ */
+ //kn->kn_fflags &= kn->kn_sfflags;
+
+ res = (kn->kn_fflags != 0);
+
+ memorystatus_klist_unlock();
+
+ return res;
+}
+
+static int
+filt_memorystatusprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev)
+{
+#pragma unused(data)
+ int res;
+
+ memorystatus_klist_lock();
+ res = (kn->kn_fflags != 0);
+ if (res) {
+ *kev = kn->kn_kevent;
+ kn->kn_flags |= EV_CLEAR; /* automatic */
+ kn->kn_fflags = 0;
+ kn->kn_data = 0;
+ }
+ memorystatus_klist_unlock();
+
+ return res;
+}
+
static void
memorystatus_klist_lock(void) {
lck_mtx_lock(&memorystatus_klist_mutex);
int
memorystatus_knote_register(struct knote *kn) {
int error = 0;
-
+
memorystatus_klist_lock();
-
- if (kn->kn_sfflags & (NOTE_MEMORYSTATUS_PRESSURE_NORMAL | NOTE_MEMORYSTATUS_PRESSURE_WARN | NOTE_MEMORYSTATUS_PRESSURE_CRITICAL | NOTE_MEMORYSTATUS_LOW_SWAP)) {
+
+ /*
+ * Support only userspace visible flags.
+ */
+ if ((kn->kn_sfflags & EVFILT_MEMORYSTATUS_ALL_MASK) == (unsigned int) kn->kn_sfflags) {
+
+#if !CONFIG_EMBEDDED
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_WARN) {
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE;
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE;
+ }
+
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL) {
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE;
+ kn->kn_sfflags |= NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE;
+ }
+#endif /* !CONFIG_EMBEDDED */
KNOTE_ATTACH(&memorystatus_klist, kn);
#endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
#endif /* 0 */
-#if CONFIG_JETSAM
/* Coalition support */
/* sorting info for a particular priority bucket */
}
return(found_pids);
}
-#endif /* CONFIG_JETSAM */
+
+int
+memorystatus_get_proccnt_upto_priority(int32_t max_bucket_index)
+{
+ int32_t i = JETSAM_PRIORITY_IDLE;
+ int count = 0;
+
+ if (max_bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ return(-1);
+ }
+
+ while(i <= max_bucket_index) {
+ count += memstat_bucket[i++].count;
+ }
+
+ return count;
+}
+
+int
+memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap)
+{
+#if !CONFIG_JETSAM
+ if (!p || (!isApp(p)) || (p->p_memstat_state & P_MEMSTAT_INTERNAL)) {
+ /*
+ * Ineligible processes OR system processes e.g. launchd.
+ */
+ return -1;
+ }
+
+ /*
+ * For macOS only:
+ * We would like to use memorystatus_update() here to move the processes
+ * within the bands. Unfortunately memorystatus_update() calls
+ * memorystatus_update_priority_locked() which uses any band transitions
+ * as an indication to modify ledgers. For that it needs the task lock
+ * and since we came into this function with the task lock held, we'll deadlock.
+ *
+ * Unfortunately we can't completely disable ledger updates because we still
+ * need the ledger updates for a subset of processes i.e. daemons.
+ * When all processes on all platforms support memory limits, we can simply call
+ * memorystatus_update().
+
+ * It also has some logic to deal with 'aging' which, currently, is only applicable
+ * on CONFIG_JETSAM configs. So, till every platform has CONFIG_JETSAM we'll need
+ * to do this explicit band transition.
+ */
+
+ memstat_bucket_t *current_bucket, *new_bucket;
+ int32_t priority = 0;
+
+ proc_list_lock();
+
+ if (((p->p_listflag & P_LIST_EXITED) != 0) ||
+ (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED))) {
+ /*
+ * If the process is on its way out OR
+ * jetsam has alread tried and failed to kill this process,
+ * let's skip the whole jetsam band transition.
+ */
+ proc_list_unlock();
+ return(0);
+ }
+
+ if (is_appnap) {
+ current_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
+ new_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
+ priority = JETSAM_PRIORITY_IDLE;
+ } else {
+ if (p->p_memstat_effectivepriority != JETSAM_PRIORITY_IDLE) {
+ /*
+ * It is possible that someone pulled this process
+ * out of the IDLE band without updating its app-nap
+ * parameters.
+ */
+ proc_list_unlock();
+ return (0);
+ }
+
+ current_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
+ new_bucket = &memstat_bucket[p->p_memstat_requestedpriority];
+ priority = p->p_memstat_requestedpriority;
+ }
+
+ TAILQ_REMOVE(¤t_bucket->list, p, p_memstat_list);
+ current_bucket->count--;
+
+ TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
+ new_bucket->count++;
+
+ /*
+ * Record idle start or idle delta.
+ */
+ if (p->p_memstat_effectivepriority == priority) {
+ /*
+ * This process is not transitioning between
+ * jetsam priority buckets. Do nothing.
+ */
+ } else if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) {
+ uint64_t now;
+ /*
+ * Transitioning out of the idle priority bucket.
+ * Record idle delta.
+ */
+ assert(p->p_memstat_idle_start != 0);
+ now = mach_absolute_time();
+ if (now > p->p_memstat_idle_start) {
+ p->p_memstat_idle_delta = now - p->p_memstat_idle_start;
+ }
+ } else if (priority == JETSAM_PRIORITY_IDLE) {
+ /*
+ * Transitioning into the idle priority bucket.
+ * Record idle start.
+ */
+ p->p_memstat_idle_start = mach_absolute_time();
+ }
+
+ p->p_memstat_effectivepriority = priority;
+
+ proc_list_unlock();
+
+ return (0);
+
+#else /* !CONFIG_JETSAM */
+ #pragma unused(p)
+ #pragma unused(is_appnap)
+ return -1;
+#endif /* !CONFIG_JETSAM */
+}
projects / apple / xnu.git / blobdiff