#include <kern/thread.h>
#include <kern/host.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 <pexpert/pexpert.h>
+#include <sys/coalition.h>
#include <sys/kern_event.h>
#include <sys/proc.h>
#include <sys/proc_info.h>
#define MEMORYSTATUS_DEBUG(cond, format, ...)
#endif
+/*
+ * Active / Inactive limit support
+ * proc list must be locked
+ *
+ * The SET_*** macros are used to initialize a limit
+ * for the first time.
+ *
+ * The CACHE_*** macros are use to cache the limit that will
+ * soon be in effect down in the ledgers.
+ */
+
+#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 { \
+ (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_ACTIVE_FATAL; \
+ } \
+MACRO_END
+
+#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 { \
+ (p)->p_memstat_state &= ~P_MEMSTAT_MEMLIMIT_INACTIVE_FATAL; \
+ } \
+MACRO_END
+
+#define CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception) \
+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; \
+ } 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; \
+ } \
+MACRO_END
+
+#define CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception) \
+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; \
+ } 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; \
+ } \
+MACRO_END
+
+
/* General tunables */
unsigned long delta_percentage = 5;
int memorystatus_wakeup = 0;
unsigned int memorystatus_level = 0;
+unsigned int memorystatus_early_boot_level = 0;
static int memorystatus_list_count = 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
+
int
memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
#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);
+
+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_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry);
+
+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);
+
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) */
static boolean_t memorystatus_idle_snapshot = 0;
-static int memorystatus_highwater_enabled = 1;
+static int memorystatus_highwater_enabled = 1; /* Update the cached memlimit data. This should be removed. */
unsigned int memorystatus_delta = 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 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_period_msecs, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jld_eval_period_msecs, 0, "");
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_jld_eval_aggressive_count, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_jld_eval_aggressive_count, 0, "");
+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;
+/*
+ * default jetsam snapshot support
+ */
static memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot;
#define memorystatus_jetsam_snapshot_list memorystatus_jetsam_snapshot->entries
-
static unsigned int memorystatus_jetsam_snapshot_count = 0;
static unsigned int memorystatus_jetsam_snapshot_max = 0;
+static uint64_t memorystatus_jetsam_snapshot_last_timestamp = 0;
+static uint64_t memorystatus_jetsam_snapshot_timeout = 0;
+#define JETSAM_SNAPSHOT_TIMEOUT_SECS 30
+
+/*
+ * snapshot support for memstats collected at boot.
+ */
+static memorystatus_jetsam_snapshot_t memorystatus_at_boot_snapshot;
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 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, uint32_t cause, int32_t *priority, uint32_t *errors);
+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_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);
+static int memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order);
+static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index);
+static int memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz);
+
+/* qsort routines */
+typedef int (*cmpfunc_t)(const void *a, const void *b);
+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 */
boolean_t memorystatus_freeze_enabled = FALSE;
int memorystatus_freeze_wakeup = 0;
+lck_grp_attr_t *freezer_lck_grp_attr;
+lck_grp_t *freezer_lck_grp;
+static lck_mtx_t freezer_mutex;
+
static inline boolean_t memorystatus_can_freeze_processes(void);
static boolean_t memorystatus_can_freeze(boolean_t *memorystatus_freeze_swap_low);
static unsigned int memorystatus_freeze_suspended_threshold = FREEZE_SUSPENDED_THRESHOLD_DEFAULT;
+static unsigned int memorystatus_freeze_daily_mb_max = FREEZE_DAILY_MB_MAX_DEFAULT;
+
/* Stats */
static uint64_t memorystatus_freeze_count = 0;
static uint64_t memorystatus_freeze_pageouts = 0;
static unsigned int memorystatus_suspended_footprint_total = 0;
+extern uint64_t vm_swap_get_free_space(void);
+
+static boolean_t memorystatus_freeze_update_throttle();
+
#endif /* CONFIG_FREEZE */
/* Debug */
#if CONFIG_JETSAM
+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;
+ unsigned int b = bucket_index;
+ boolean_t traverse_all_buckets = FALSE;
+
+ if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ traverse_all_buckets = TRUE;
+ b = 0;
+ } else {
+ traverse_all_buckets = FALSE;
+ b = bucket_index;
+ }
+
+ /*
+ * Missing from this dump is the value actually
+ * stored in the ledger... also, format could be better.
+ */
+ 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");
+ 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,
+ p->p_memstat_state, p->p_memstat_effectivepriority, p->p_memstat_requestedpriority, p->p_memstat_dirty, p->p_memstat_idledeadline,
+ 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 = memorystatus_get_next_proc_locked(&b, p, traverse_all_buckets);
+ }
+ printf("memorystatus_debug_dump ***END***\n");
+}
+
+static int
+sysctl_memorystatus_debug_dump_bucket SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp, arg2)
+ int bucket_index = 0;
+ int error;
+ error = SYSCTL_OUT(req, arg1, sizeof(int));
+ if (error || !req->newptr) {
+ return (error);
+ }
+ error = SYSCTL_IN(req, &bucket_index, sizeof(int));
+ if (error || !req->newptr) {
+ return (error);
+ }
+ if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ /*
+ * All jetsam buckets will be dumped.
+ */
+ } else {
+ /*
+ * Only a single bucket will be dumped.
+ */
+ }
+
+ proc_list_lock();
+ memorystatus_debug_dump_bucket_locked(bucket_index);
+ proc_list_unlock();
+ memorystatus_debug_dump_this_bucket = bucket_index;
+ return (error);
+}
+
+/*
+ * Debug aid to look at jetsam buckets and proc jetsam fields.
+ * Use this sysctl to act on a particular jetsam bucket.
+ * Writing the sysctl triggers the dump.
+ * Usage: sysctl kern.memorystatus_debug_dump_this_bucket=<bucket_index>
+ */
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_debug_dump_this_bucket, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_debug_dump_this_bucket, 0, sysctl_memorystatus_debug_dump_bucket, "I", "");
+
+
/* Debug aid to aid determination of limit */
static int
proc_t p;
unsigned int b = 0;
int error, enable = 0;
- int32_t memlimit;
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;
+
if (enable) {
- if ((p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
- memlimit = -1;
+ /*
+ * 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);
} else {
- memlimit = p->p_memstat_memlimit;
+ CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
}
+
} else {
- memlimit = -1;
- }
- task_set_phys_footprint_limit_internal(p->task, (memlimit > 0) ? memlimit : -1, NULL, TRUE);
-
- if (memlimit == -1) {
- p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
- } else {
- if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND) {
- p->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT;
- }
+ /*
+ * Disabling limits does not touch the stored variants.
+ * Set the cached limit fields to system_wide defaults.
+ */
+ p->p_memstat_memlimit = -1;
+ p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
+ trigger_exception = 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);
+
p = memorystatus_get_next_proc_locked(&b, p, TRUE);
}
proc_list_unlock();
return 0;
+
}
SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
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)
*/
memorystatus_klist_lock();
- kn = vm_find_knote_from_pid(pid, &memorystatus_klist);
- if (kn) {
- /*
- * Forcibly send this pid a "warning" memory pressure notification.
- */
- kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_WARN;
- KNOTE(&memorystatus_klist, kMemorystatusPressure);
- ret = 0;
+
+ 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;
#if CONFIG_FREEZE
+SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_daily_mb_max, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_daily_mb_max, 0, "");
+
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_threshold, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_threshold, 0, "");
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_pages_min, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_pages_min, 0, "");
sysctl_memorystatus_freeze SYSCTL_HANDLER_ARGS
{
#pragma unused(arg1, arg2)
-
int error, pid = 0;
proc_t p;
if (error || !req->newptr)
return (error);
+ if (pid == 2) {
+ vm_pageout_anonymous_pages();
+
+ return 0;
+ }
+
+ lck_mtx_lock(&freezer_mutex);
+
p = proc_find(pid);
if (p != NULL) {
uint32_t purgeable, wired, clean, dirty;
uint32_t max_pages = 0;
if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
- max_pages = MIN(default_pager_swap_pages_free(), memorystatus_freeze_pages_max);
+
+ 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;
+ }
+
+ max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
+
} else {
+ /*
+ * We only have the compressor without any swap.
+ */
max_pages = UINT32_MAX - 1;
}
+
error = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
proc_rele(p);
if (error)
error = EIO;
+
+ lck_mtx_unlock(&freezer_mutex);
return error;
}
+
+ lck_mtx_unlock(&freezer_mutex);
return EINVAL;
}
thread_t *new_thread);
#if CONFIG_JETSAM
+/*
+ * Picks the sorting routine for a given jetsam priority band.
+ *
+ * Input:
+ * bucket_index - jetsam priority band to be sorted.
+ * sort_order - JETSAM_SORT_xxx from kern_memorystatus.h
+ * Currently sort_order is only meaningful when handling
+ * coalitions.
+ *
+ * Return:
+ * 0 on success
+ * non-0 on failure
+ */
+static int memorystatus_sort_bucket(unsigned int bucket_index, int sort_order)
+{
+ int coal_sort_order;
+
+ /*
+ * Verify the jetsam priority
+ */
+ if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ return(EINVAL);
+ }
+
+#if DEVELOPMENT || DEBUG
+ if (sort_order == JETSAM_SORT_DEFAULT) {
+ coal_sort_order = COALITION_SORT_DEFAULT;
+ } else {
+ coal_sort_order = sort_order; /* only used for testing scenarios */
+ }
+#else
+ /* Verify default */
+ if (sort_order == JETSAM_SORT_DEFAULT) {
+ coal_sort_order = COALITION_SORT_DEFAULT;
+ } else {
+ return(EINVAL);
+ }
+#endif
+
+ proc_list_lock();
+ switch (bucket_index) {
+ case JETSAM_PRIORITY_FOREGROUND:
+ if (memorystatus_sort_by_largest_coalition_locked(bucket_index, coal_sort_order) == 0) {
+ /*
+ * Fall back to per process sorting when zero coalitions are found.
+ */
+ memorystatus_sort_by_largest_process_locked(bucket_index);
+ }
+ break;
+ default:
+ memorystatus_sort_by_largest_process_locked(bucket_index);
+ break;
+ }
+ proc_list_unlock();
+
+ return(0);
+}
+
/*
* Sort processes by size for a single jetsam bucket.
*/
static void memorystatus_sort_by_largest_process_locked(unsigned int bucket_index)
{
proc_t p = NULL, insert_after_proc = NULL, max_proc = NULL;
+ proc_t next_p = NULL, prev_max_proc = NULL;
uint32_t pages = 0, max_pages = 0;
memstat_bucket_t *current_bucket;
p = TAILQ_FIRST(¤t_bucket->list);
- if (p) {
+ while (p) {
memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
max_pages = pages;
- insert_after_proc = NULL;
-
- p = TAILQ_NEXT(p, p_memstat_list);
-
-restart:
- while (p) {
-
+ max_proc = p;
+ prev_max_proc = p;
+
+ while ((next_p = TAILQ_NEXT(p, p_memstat_list)) != NULL) {
+ /* traversing list until we find next largest process */
+ p=next_p;
memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
-
if (pages > max_pages) {
max_pages = pages;
max_proc = p;
}
-
- p = TAILQ_NEXT(p, p_memstat_list);
}
- if (max_proc) {
-
+ if (prev_max_proc != max_proc) {
+ /* found a larger process, place it in the list */
TAILQ_REMOVE(¤t_bucket->list, max_proc, p_memstat_list);
-
if (insert_after_proc == NULL) {
TAILQ_INSERT_HEAD(¤t_bucket->list, max_proc, p_memstat_list);
} else {
TAILQ_INSERT_AFTER(¤t_bucket->list, insert_after_proc, max_proc, p_memstat_list);
}
+ prev_max_proc = max_proc;
+ }
- insert_after_proc = max_proc;
-
- /* Reset parameters for the new search. */
- p = TAILQ_NEXT(max_proc, p_memstat_list);
- if (p) {
- memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
- max_pages = pages;
- }
- max_proc = NULL;
+ insert_after_proc = max_proc;
- goto restart;
- }
+ p = TAILQ_NEXT(max_proc, p_memstat_list);
}
}
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));
+
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;
panic("Could not allocate memorystatus_jetsam_snapshot");
}
+ nanoseconds_to_absolutetime((uint64_t)JETSAM_SNAPSHOT_TIMEOUT_SECS * NSEC_PER_SEC, &memorystatus_jetsam_snapshot_timeout);
+
+ 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
KERNEL_DEBUG_CONSTANT( (BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_END,
victim_pid, cause, vm_page_free_count, error, 0);
- if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
- vm_wake_compactor_swapper();
- }
+ vm_wake_compactor_swapper();
return (error == 0);
}
present_in_deferred_bucket = TRUE;
}
- MEMORYSTATUS_DEBUG(1, "memorystatus_schedule_idle_demotion_locked: scheduling demotion to idle band for process %d (dirty:0x%x, set_state %d, demotions %d).\n",
+ 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);
assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
assert(p->p_memstat_idledeadline);
}
- MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for process %d (clear_state %d, demotions %d).\n",
+ 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);
{
memstat_bucket_t *bucket;
- MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding process %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
+ 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();
return 0;
}
+/*
+ * Description:
+ * Moves a process from one jetsam bucket to another.
+ * which changes the LRU position of the process.
+ *
+ * Monitors transition between buckets and if necessary
+ * will update cached memory limits accordingly.
+ */
static void
memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert)
{
return;
}
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting process %d to priority %d, inserting at %s\n",
+ 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");
old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
else
TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
new_bucket->count++;
-
+
#if CONFIG_JETSAM
- if (memorystatus_highwater_enabled && (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND)) {
+ if (memorystatus_highwater_enabled) {
+ boolean_t trigger_exception;
+
+ /*
+ * If cached limit data is updated, then the limits
+ * will be enforced by writing to the ledgers.
+ */
+ boolean_t ledger_update_needed = TRUE;
/*
- * Adjust memory limit based on if the task is going to/from foreground and background.
+ * 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
+ * FG <--> BG
+ * but:
+ * dirty <--> clean is ignored
+ *
+ * We bypass 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 (((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) ||
- ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND))) {
- int32_t memlimit = (priority >= JETSAM_PRIORITY_FOREGROUND) ? -1 : p->p_memstat_memlimit;
- task_set_phys_footprint_limit_internal(p->task, (memlimit > 0) ? memlimit : -1, NULL, TRUE);
-
- if (memlimit <= 0) {
- p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
- } else {
- p->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT;
- }
+ if (p->p_memstat_dirty & P_DIRTY_TRACK) {
+
+ ledger_update_needed = FALSE;
+
+ } else if ((priority >= JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority < JETSAM_PRIORITY_FOREGROUND)) {
+ /*
+ * inactive --> active
+ * BG --> FG
+ * assign active state
+ */
+ CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+
+ } else if ((priority < JETSAM_PRIORITY_FOREGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) {
+ /*
+ * active --> inactive
+ * FG --> BG
+ * assign inactive state
+ */
+ CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ } else {
+ /*
+ * The transition between jetsam priority buckets apparently did
+ * not affect active/inactive state.
+ * This is not unusual... especially during startup when
+ * processes are getting established in their respective bands.
+ */
+ ledger_update_needed = FALSE;
+ }
+
+ /*
+ * 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);
+
+ 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),
+ (p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), p->p_memstat_effectivepriority, priority, p->p_memstat_dirty,
+ (p->p_memstat_dirty ? ((p->p_memstat_dirty & P_DIRTY) ? "isdirty" : "isclean") : ""));
}
}
-#endif
+
+#endif /* CONFIG_JETSAM */
p->p_memstat_effectivepriority = priority;
memorystatus_check_levels_locked();
}
+/*
+ *
+ * Description: Update the jetsam priority and memory limit attributes for a given process.
+ *
+ * Parameters:
+ * p init this process's jetsam information.
+ * priority The jetsam priority band
+ * user_data user specific data, unused by the kernel
+ * effective guards against race if process's update already occurred
+ * update_memlimit When true we know this is the init step via the posix_spawn path.
+ *
+ * memlimit_active Value in megabytes; The monitored footprint level while the
+ * process is active. Exceeding it may result in termination
+ * based on it's associated fatal flag.
+ *
+ * memlimit_active_is_fatal When a process is active and exceeds its memory footprint,
+ * this describes whether or not it should be immediately fatal.
+ *
+ * memlimit_inactive Value in megabytes; The monitored footprint level while the
+ * process is inactive. Exceeding it may result in termination
+ * based on it's associated fatal flag.
+ *
+ * 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, boolean_t memlimit_background, boolean_t is_fatal_limit)
+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)
{
int ret;
boolean_t head_insert = false;
#if !CONFIG_JETSAM
-#pragma unused(update_memlimit, memlimit, memlimit_background, is_fatal_limit)
-#endif
+#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 process %d: priority %d, user_data 0x%llx\n", 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) {
} else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
/* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
priority = JETSAM_PRIORITY_IDLE;
- head_insert = true;
+ head_insert = TRUE;
} else if ((priority < 0) || (priority >= MEMSTAT_BUCKET_COUNT)) {
/* Sanity check */
ret = EINVAL;
goto out;
}
-
+
proc_list_lock();
assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
#if CONFIG_JETSAM
if (update_memlimit) {
- p->p_memstat_memlimit = memlimit;
+ boolean_t trigger_exception;
+
+ /*
+ * Posix_spawn'd processes come through this path to instantiate ledger limits.
+ * Forked processes do not come through this path, so no ledger limits exist.
+ * (That's why forked processes can consume unlimited memory.)
+ */
+
+ MEMORYSTATUS_DEBUG(3, "memorystatus_update(enter): pid %d, priority %d, dirty=0x%x, Active(%dMB %s), Inactive(%dMB, %s)\n",
+ p->p_pid, priority, p->p_memstat_dirty,
+ memlimit_active, (memlimit_active_is_fatal ? "F " : "NF"),
+ memlimit_inactive, (memlimit_inactive_is_fatal ? "F " : "NF"));
+
if (memlimit_background) {
- /* Will be set as priority is updated */
- p->p_memstat_state |= P_MEMSTAT_MEMLIMIT_BACKGROUND;
- /* Cannot have a background memory limit and be fatal. */
- is_fatal_limit = FALSE;
+ /*
+ * With 2-level HWM support, we no longer honor P_MEMSTAT_MEMLIMIT_BACKGROUND.
+ * Background limits are described via the inactive limit slots.
+ */
- } else {
- /* Otherwise, apply now */
- if (memorystatus_highwater_enabled) {
- task_set_phys_footprint_limit_internal(p->task, (memlimit > 0) ? memlimit : -1, NULL, TRUE);
+ // 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.
+ * System_wide task limit is always fatal.
+ * It's quite common to see non-fatal flag passed in here.
+ * It's not an error, we just ignore it.
+ */
+
+ /*
+ * For backward compatibility with some unexplained launchd behavior,
+ * we allow a zero sized limit. But we still enforce system_wide limit
+ * when written to the ledgers.
+ */
+
+ if (memlimit_active < 0) {
+ memlimit_active = -1; /* enforces system_wide task limit */
}
+ memlimit_active_is_fatal = TRUE;
}
-
- if (is_fatal_limit || memlimit <= 0) {
- p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
- } else {
- p->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT;
+
+ if (memlimit_inactive <= 0) {
+ /*
+ * This process will have a system_wide task limit when inactive.
+ * System_wide task limit is always fatal.
+ */
+
+ memlimit_inactive = -1;
+ memlimit_inactive_is_fatal = TRUE;
}
- }
-#endif
- /*
+ /*
+ * Initialize the active limit variants for this process.
+ */
+ SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
+
+ /*
+ * Initialize the inactive limit variants for this process.
+ */
+ SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
+
+ /*
+ * Initialize the cached limits for target process.
+ * When the target process is dirty tracked, it's typically
+ * in a clean state. Non dirty tracked processes are
+ * typically active (Foreground or above).
+ * But just in case, we don't make assumptions...
+ */
+
+ if (proc_jetsam_state_is_active_locked(p) == TRUE) {
+ CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ } else {
+ CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ }
+
+ /*
+ * Enforce the cached limit by writing to the ledger.
+ */
+ 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);
+
+ 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_state & P_MEMSTAT_FATAL_MEMLIMIT ? "F " : "NF"), priority, p->p_memstat_dirty,
+ (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.
* Check and take appropriate steps if so.
int ret;
memstat_bucket_t *bucket;
- MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing process %d\n", p->p_pid);
+ MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing pid %d\n", p->p_pid);
if (!locked) {
proc_list_lock();
return ret;
}
-static boolean_t
+/*
+ * Validate dirty tracking flags with process state.
+ *
+ * Return:
+ * 0 on success
+ * non-0 on failure
+ */
+
+static int
memorystatus_validate_track_flags(struct proc *target_p, uint32_t pcontrol) {
/* See that the process isn't marked for termination */
if (target_p->p_memstat_dirty & P_DIRTY_TERMINATED) {
- return FALSE;
+ return EBUSY;
}
/* Idle exit requires that process be tracked */
if ((pcontrol & PROC_DIRTY_ALLOW_IDLE_EXIT) &&
!(pcontrol & PROC_DIRTY_TRACK)) {
- return FALSE;
+ return EINVAL;
}
/* 'Launch in progress' tracking requires that process have enabled dirty tracking too. */
if ((pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) &&
!(pcontrol & PROC_DIRTY_TRACK)) {
- return FALSE;
+ return EINVAL;
}
/* Deferral is only relevant if idle exit is specified */
if ((pcontrol & PROC_DIRTY_DEFER) &&
!(pcontrol & PROC_DIRTY_ALLOWS_IDLE_EXIT)) {
- return FALSE;
+ return EINVAL;
}
- return TRUE;
+ return(0);
}
static void
memorystatus_update_idle_priority_locked(proc_t p) {
int32_t priority;
-
+
MEMORYSTATUS_DEBUG(1, "memorystatus_update_idle_priority_locked(): pid %d dirty 0x%X\n", p->p_pid, p->p_memstat_dirty);
if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_IS_DIRTY)) == P_DIRTY_IDLE_EXIT_ENABLED) {
boolean_t reschedule = FALSE;
boolean_t already_deferred = FALSE;
boolean_t defer_now = FALSE;
- int ret;
+ int ret = 0;
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_TRACK),
p->p_pid, p->p_memstat_dirty, pcontrol, 0, 0);
goto exit;
}
- if (!memorystatus_validate_track_flags(p, pcontrol)) {
- ret = EINVAL;
+ if ((ret = memorystatus_validate_track_flags(p, pcontrol)) != 0) {
+ /* error */
goto exit;
- }
+ }
old_dirty = p->p_memstat_dirty;
defer_now = TRUE;
}
- MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for process %d\n",
+ MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for pid %d\n",
((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) ? "Y" : "N",
defer_now ? "Y" : "N",
p->p_memstat_dirty & P_DIRTY ? "Y" : "N",
memorystatus_dirty_count++;
ret = 0;
} else if ((pcontrol == 0) && (p->p_memstat_dirty & flag)) {
- if ((flag == P_DIRTY_SHUTDOWN) && (!p->p_memstat_dirty & P_DIRTY)) {
+ if ((flag == P_DIRTY_SHUTDOWN) && (!(p->p_memstat_dirty & P_DIRTY))) {
/* Clearing the dirty shutdown flag, and the process is otherwise clean - kill */
p->p_memstat_dirty |= P_DIRTY_TERMINATED;
kill = true;
if (ret != 0) {
goto exit;
}
-
+
if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY)
now_dirty = TRUE;
}
}
}
-
+
memorystatus_update_idle_priority_locked(p);
+
+#if CONFIG_JETSAM
+ if (memorystatus_highwater_enabled) {
+ boolean_t trigger_exception;
+ /*
+ * 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 dirty
+ */
+ CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ } else {
+ /*
+ * process is clean
+ */
+ CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
+ }
+
+ /*
+ * Enforce the new limits by writing to the ledger.
+ *
+ * This is a hot path and holding the proc_list_lock while writing to the ledgers,
+ * (where the task lock is taken) is bad. So, we temporarily drop the proc_list_lock.
+ * We aren't traversing the jetsam bucket list here, so we should be safe.
+ * See rdar://21394491.
+ */
+
+ if (proc_ref_locked(p) == p) {
+ int ledger_limit;
+ if (p->p_memstat_memlimit > 0) {
+ ledger_limit = p->p_memstat_memlimit;
+ } else {
+ ledger_limit = -1;
+ }
+ proc_list_unlock();
+ task_set_phys_footprint_limit_internal(p->task, ledger_limit, NULL, trigger_exception);
+ proc_list_lock();
+ proc_rele_locked(p);
+
+ MEMORYSTATUS_DEBUG(3, "memorystatus_dirty_set: new limit on pid %d (%dMB %s) priority(%d) dirty?=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") : ""));
+ }
+
+ }
+#endif /* CONFIG_JETSAM */
/* If the deferral state changed, reschedule the demotion timer */
if (reschedule) {
memorystatus_reschedule_idle_demotion_locked();
}
}
-
+
if (kill) {
- psignal(p, SIGKILL);
+ if (proc_ref_locked(p) == p) {
+ proc_list_unlock();
+ psignal(p, SIGKILL);
+ proc_list_lock();
+ proc_rele_locked(p);
+ }
}
exit:
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
if (is_vm_privileged == FALSE) {
thread_wire(host_priv_self(), current_thread(), TRUE);
is_vm_privileged = TRUE;
+ if (vm_restricted_to_single_processor == TRUE)
+ thread_vm_bind_group_add();
+
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, 0, 0, 0, 0);
+ memorystatus_available_pages, memorystatus_jld_enabled, memorystatus_jld_eval_period_msecs, memorystatus_jld_eval_aggressive_count,0);
/*
* Jetsam aware version.
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;
+ }
+
+ 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;
+ }
+ }
+
+ killed = memorystatus_kill_top_process_aggressive(
+ TRUE,
+ kMemorystatusKilledVMThrashing,
+ jld_eval_aggressive_count,
+ jld_priority_band_max,
+ &errors);
+
+
+ if (killed) {
+ /* Always generate logs after aggressive kill */
+ post_snapshot = TRUE;
+ goto done;
+ }
+ }
+ }
/* LRU */
- killed = memorystatus_kill_top_process(TRUE, cause, &priority, &errors);
+ killed = memorystatus_kill_top_process(TRUE, sort_flag, cause, &priority, &errors);
+ sort_flag = FALSE;
+
if (killed) {
- /* Don't generate logs for steady-state idle-exit kills (unless overridden for debug) */
+ /*
+ * 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) {
post_snapshot = TRUE;
}
+
+ /* Jetsam Loop Detection */
+ if (memorystatus_jld_enabled == TRUE) {
+ if ((priority == JETSAM_PRIORITY_IDLE) || (priority == JETSAM_PRIORITY_IDLE_DEFERRED)) {
+ jld_idle_kills++;
+ } else {
+ /*
+ * We've reached into bands beyond idle deferred.
+ * We make no attempt to monitor them
+ */
+ }
+ }
goto done;
}
if (post_snapshot) {
size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count);
- memorystatus_jetsam_snapshot->notification_time = mach_absolute_time();
- memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
+ 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();
+ }
+ }
}
-
+
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_END,
memorystatus_available_pages, 0, 0, 0, 0);
void
memorystatus_on_ledger_footprint_exceeded(boolean_t warning, const int max_footprint_mb)
{
+ boolean_t is_active;
+ boolean_t is_fatal;
+
proc_t p = current_proc();
- if (warning == FALSE) {
- printf("process %d (%s) exceeded physical memory footprint limit of %d MB\n",
- p->p_pid, p->p_comm, max_footprint_mb);
+ 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) {
/* 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");
+ printf("task_exceeded_footprint: failed to warn the current task (exiting, or no handler registered?).\n");
}
return;
}
#endif /* VM_PRESSURE_EVENTS */
- if ((p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT) == P_MEMSTAT_FATAL_MEMLIMIT) {
+ if (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.
}
}
+/*
+ * Toggle the P_MEMSTAT_TERMINATED state.
+ * Takes the proc_list_lock.
+ */
+void
+proc_memstat_terminated(proc_t p, boolean_t set)
+{
+#if DEVELOPMENT || DEBUG
+ if (p) {
+ proc_list_lock();
+ if (set == TRUE) {
+ p->p_memstat_state |= P_MEMSTAT_TERMINATED;
+ } else {
+ p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
+ }
+ proc_list_unlock();
+ }
+#else
+#pragma unused(p, set)
+ /*
+ * do nothing
+ */
+#endif /* DEVELOPMENT || DEBUG */
+ return;
+}
+
/*
* 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.
}
}
-
static void
-memorystatus_update_snapshot_locked(proc_t p, uint32_t kill_cause)
+memorystatus_update_jetsam_snapshot_entry_locked(proc_t p, uint32_t kill_cause)
{
unsigned int i;
}
static boolean_t
-memorystatus_get_snapshot_properties_for_proc_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)
{
clock_sec_t tv_sec;
clock_usec_t tv_usec;
}
static void
-memorystatus_jetsam_snapshot_procs_locked(void)
+memorystatus_init_snapshot_vmstats(memorystatus_jetsam_snapshot_t *snapshot)
{
- proc_t p, next_p;
- unsigned int b = 0, i = 0;
kern_return_t kr = KERN_SUCCESS;
-
mach_msg_type_number_t count = HOST_VM_INFO64_COUNT;
vm_statistics64_data_t vm_stat;
if ((kr = host_statistics64(host_self(), HOST_VM_INFO64, (host_info64_t)&vm_stat, &count) != KERN_SUCCESS)) {
- printf("memorystatus_jetsam_snapshot_procs_locked: host_statistics64 failed with %d\n", kr);
- memset(&memorystatus_jetsam_snapshot->stats, 0, sizeof(memorystatus_jetsam_snapshot->stats));
+ printf("memorystatus_init_jetsam_snapshot_stats: host_statistics64 failed with %d\n", kr);
+ memset(&snapshot->stats, 0, sizeof(snapshot->stats));
} else {
- memorystatus_jetsam_snapshot->stats.free_pages = vm_stat.free_count;
- memorystatus_jetsam_snapshot->stats.active_pages = vm_stat.active_count;
- memorystatus_jetsam_snapshot->stats.inactive_pages = vm_stat.inactive_count;
- memorystatus_jetsam_snapshot->stats.throttled_pages = vm_stat.throttled_count;
- memorystatus_jetsam_snapshot->stats.purgeable_pages = vm_stat.purgeable_count;
- memorystatus_jetsam_snapshot->stats.wired_pages = vm_stat.wire_count;
-
- memorystatus_jetsam_snapshot->stats.speculative_pages = vm_stat.speculative_count;
- memorystatus_jetsam_snapshot->stats.filebacked_pages = vm_stat.external_page_count;
- memorystatus_jetsam_snapshot->stats.anonymous_pages = vm_stat.internal_page_count;
- memorystatus_jetsam_snapshot->stats.compressions = vm_stat.compressions;
- memorystatus_jetsam_snapshot->stats.decompressions = vm_stat.decompressions;
- memorystatus_jetsam_snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
- memorystatus_jetsam_snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
+ snapshot->stats.free_pages = vm_stat.free_count;
+ snapshot->stats.active_pages = vm_stat.active_count;
+ snapshot->stats.inactive_pages = vm_stat.inactive_count;
+ snapshot->stats.throttled_pages = vm_stat.throttled_count;
+ snapshot->stats.purgeable_pages = vm_stat.purgeable_count;
+ snapshot->stats.wired_pages = vm_stat.wire_count;
+
+ snapshot->stats.speculative_pages = vm_stat.speculative_count;
+ snapshot->stats.filebacked_pages = vm_stat.external_page_count;
+ snapshot->stats.anonymous_pages = vm_stat.internal_page_count;
+ snapshot->stats.compressions = vm_stat.compressions;
+ snapshot->stats.decompressions = vm_stat.decompressions;
+ snapshot->stats.compressor_pages = vm_stat.compressor_page_count;
+ snapshot->stats.total_uncompressed_pages_in_compressor = vm_stat.total_uncompressed_pages_in_compressor;
+ }
+}
+
+/*
+ * Collect vm statistics at boot.
+ * Called only once (see kern_exec.c)
+ * Data can be consumed at any time.
+ */
+void
+memorystatus_init_at_boot_snapshot() {
+ memorystatus_init_snapshot_vmstats(&memorystatus_at_boot_snapshot);
+ memorystatus_at_boot_snapshot.entry_count = 0;
+ memorystatus_at_boot_snapshot.notification_time = 0; /* updated when consumed */
+ memorystatus_at_boot_snapshot.snapshot_time = mach_absolute_time();
+}
+
+static void
+memorystatus_init_jetsam_snapshot_locked(memorystatus_jetsam_snapshot_t *od_snapshot, uint32_t ods_list_count )
+{
+ proc_t p, next_p;
+ unsigned int b = 0, i = 0;
+
+ memorystatus_jetsam_snapshot_t *snapshot = NULL;
+ memorystatus_jetsam_snapshot_entry_t *snapshot_list = NULL;
+ unsigned int snapshot_max = 0;
+
+ if (od_snapshot) {
+ /*
+ * This is an on_demand snapshot
+ */
+ snapshot = od_snapshot;
+ snapshot_list = od_snapshot->entries;
+ snapshot_max = ods_list_count;
+ } else {
+ /*
+ * This is a jetsam event snapshot
+ */
+ snapshot = memorystatus_jetsam_snapshot;
+ snapshot_list = memorystatus_jetsam_snapshot->entries;
+ snapshot_max = memorystatus_jetsam_snapshot_max;
}
+ memorystatus_init_snapshot_vmstats(snapshot);
+
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_get_snapshot_properties_for_proc_locked(p, &memorystatus_jetsam_snapshot_list[i])) {
+ if (FALSE == memorystatus_init_jetsam_snapshot_entry_locked(p, &snapshot_list[i])) {
continue;
}
- MEMORYSTATUS_DEBUG(0, "jetsam snapshot pid = %d, uuid = %02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
+ MEMORYSTATUS_DEBUG(0, "jetsam snapshot pid %d, uuid = %02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
p->p_pid,
p->p_uuid[0], p->p_uuid[1], p->p_uuid[2], p->p_uuid[3], p->p_uuid[4], p->p_uuid[5], p->p_uuid[6], p->p_uuid[7],
p->p_uuid[8], p->p_uuid[9], p->p_uuid[10], p->p_uuid[11], p->p_uuid[12], p->p_uuid[13], p->p_uuid[14], p->p_uuid[15]);
- if (++i == memorystatus_jetsam_snapshot_max) {
+ if (++i == snapshot_max) {
break;
}
}
- memorystatus_jetsam_snapshot->snapshot_time = mach_absolute_time();
- memorystatus_jetsam_snapshot->entry_count = memorystatus_jetsam_snapshot_count = i;
+ snapshot->snapshot_time = mach_absolute_time();
+ snapshot->entry_count = i;
+
+ if (!od_snapshot) {
+ /* update the system buffer count */
+ memorystatus_jetsam_snapshot_count = i;
+ }
}
#if DEVELOPMENT || DEBUG
return ret;
}
+/*
+ * Triggers a sort_order on a specified jetsam priority band.
+ * This is for testing only, used to force a path through the sort
+ * function.
+ */
+static int
+memorystatus_cmd_test_jetsam_sort(int priority, int sort_order) {
+
+ 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
/*
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) - memorystatus_available_pages: %d\n",
- victim_pid, (p->p_comm ? p->p_comm : "(unknown)"),
- jetsam_kill_cause_name[cause], memorystatus_available_pages);
-
proc_list_lock();
if (memorystatus_jetsam_snapshot_count == 0) {
- memorystatus_jetsam_snapshot_procs_locked();
+ memorystatus_init_jetsam_snapshot_locked(NULL,0);
}
- memorystatus_update_snapshot_locked(p, cause);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause);
proc_list_unlock();
+
+ 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);
+
killed = memorystatus_do_kill(p, cause);
proc_rele(p);
* Jetsam the first process in the queue.
*/
static boolean_t
-memorystatus_kill_top_process(boolean_t any, uint32_t cause, int32_t *priority, uint32_t *errors)
+memorystatus_kill_top_process(boolean_t any, boolean_t sort_flag, uint32_t cause, 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;
+ int kill_count = 0;
unsigned int i = 0;
+ uint32_t aPid_ep;
#ifndef CONFIG_FREEZE
#pragma unused(any)
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_START,
memorystatus_available_pages, 0, 0, 0, 0);
- proc_list_lock();
- memorystatus_sort_by_largest_process_locked(JETSAM_PRIORITY_FOREGROUND);
+ if (sort_flag == TRUE) {
+ (void)memorystatus_sort_bucket(JETSAM_PRIORITY_FOREGROUND, JETSAM_SORT_DEFAULT);
+ }
+
+ proc_list_lock();
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
while (next_p) {
#endif /* DEVELOPMENT || DEBUG */
aPid = p->p_pid;
+ aPid_ep = p->p_memstat_effectivepriority;
if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
continue;
} else
#endif
{
- if (priority) {
- *priority = p->p_memstat_effectivepriority;
- }
-
/*
* Capture a snapshot if none exists and:
* - priority was not requested (this is something other than an ambient kill)
*/
if ((memorystatus_jetsam_snapshot_count == 0) &&
(memorystatus_idle_snapshot || ((!priority) || (priority && (*priority != JETSAM_PRIORITY_IDLE))))) {
- memorystatus_jetsam_snapshot_procs_locked();
+ memorystatus_init_jetsam_snapshot_locked(NULL,0);
new_snapshot = TRUE;
}
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_snapshot_locked(p, kMemorystatusKilledDiagnostic);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic);
p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
if (memorystatus_jetsam_policy & kPolicyDiagnoseFirst) {
jetsam_diagnostic_suspended_one_active_proc = 1;
proc_list_unlock();
if (p) {
task_suspend(p->task);
+ if (priority) {
+ *priority = aPid_ep;
+ }
proc_rele(p);
killed = TRUE;
}
#endif /* DEVELOPMENT || DEBUG */
{
/* Shift queue, update stats */
- memorystatus_update_snapshot_locked(p, cause);
-
- p = proc_ref_locked(p);
- proc_list_unlock();
- if (p) {
- printf("memorystatus: %s %d [%s] (%s) - memorystatus_available_pages: %d\n",
- ((p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE) ?
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause);
+
+ 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], memorystatus_available_pages);
+ jetsam_kill_cause_name[cause], aPid_ep, memorystatus_available_pages);
+
killed = memorystatus_do_kill(p, cause);
- }
+
+ /* Success? */
+ if (killed) {
+ if (priority) {
+ *priority = aPid_ep;
+ }
+ proc_rele(p);
+ kill_count++;
+ goto exit;
+ }
- /* Success? */
- if (killed) {
- proc_rele(p);
- 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 - unwind and restart. */
- 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.
+ *
+ * And, we hold the the proc_list_lock at this point.
+ */
+
i = 0;
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
}
}
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
- memorystatus_available_pages, killed ? aPid : 0, 0, 0, 0);
+ memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
return killed;
}
-#if LEGACY_HIWATER
-
+/*
+ * Jetsam aggressively
+ */
static boolean_t
-memorystatus_kill_hiwat_proc(uint32_t *errors)
+memorystatus_kill_top_process_aggressive(boolean_t any, uint32_t cause, int aggr_count, int32_t priority_max,
+ uint32_t *errors)
{
- pid_t aPid = 0;
+ pid_t aPid;
proc_t p = PROC_NULL, next_p = PROC_NULL;
boolean_t new_snapshot = FALSE, killed = FALSE;
+ int kill_count = 0;
unsigned int i = 0;
-
- KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
- memorystatus_available_pages, 0, 0, 0, 0);
-
+ int32_t aPid_ep = 0;
+ unsigned int memorystatus_level_snapshot = 0;
+
+#pragma unused(any)
+
+ 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);
+
proc_list_lock();
- memorystatus_sort_by_largest_process_locked(JETSAM_PRIORITY_FOREGROUND);
-
+
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
while (next_p) {
- uint32_t footprint;
- boolean_t skip;
+#if DEVELOPMENT || DEBUG
+ int activeProcess;
+ int procSuspendedForDiagnosis;
+#endif /* DEVELOPMENT || DEBUG */
- p = next_p;
- next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
-
+ if ((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
+ * 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);
+
+ next_p = memorystatus_get_first_proc_locked(&i, TRUE);
+ continue;
+ }
+
+ p = next_p;
+ next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
+
+ if (p->p_memstat_effectivepriority > priority_max) {
+ /*
+ * Bail out of this killing spree if we have
+ * reached beyond the priority_max jetsam band.
+ * That is, we kill up to and through the
+ * priority_max jetsam band.
+ */
+ proc_list_unlock();
+ goto exit;
+ }
+
+#if DEVELOPMENT || DEBUG
+ activeProcess = p->p_memstat_state & P_MEMSTAT_FOREGROUND;
+ procSuspendedForDiagnosis = p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED;
+#endif /* DEVELOPMENT || DEBUG */
+
+ aPid = p->p_pid;
+ aPid_ep = p->p_memstat_effectivepriority;
+
+ if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
+ continue;
+ }
+
+#if 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 */
+
+ /*
+ * Capture a snapshot if none exists.
+ */
+ if (memorystatus_jetsam_snapshot_count == 0) {
+ memorystatus_init_jetsam_snapshot_locked(NULL,0);
+ new_snapshot = TRUE;
+ }
+
+ /*
+ * Mark as terminated so that if exit1() indicates success, but the process (for example)
+ * is blocked in task_exception_notify(), it'll be skipped if encountered again - see
+ * <rdar://problem/13553476>. This is cheaper than examining P_LEXIT, which requires the
+ * acquisition of the proc lock.
+ */
+ p->p_memstat_state |= P_MEMSTAT_TERMINATED;
+
+ /* Shift queue, update stats */
+ memorystatus_update_jetsam_snapshot_entry_locked(p, cause);
+
+ /*
+ * In order to kill the target process, we will drop the proc_list_lock.
+ * To guaranteee that p and next_p don't disappear out from under the lock,
+ * we must take a ref on both.
+ * If we cannot get a reference, then it's likely we've raced with
+ * that process exiting on another core.
+ */
+ if (proc_ref_locked(p) == p) {
+ if (next_p) {
+ while (next_p && (proc_ref_locked(next_p) != next_p)) {
+ proc_t temp_p;
+
+ /*
+ * We must have raced with next_p exiting on another core.
+ * Recover by getting the next eligible process in the band.
+ */
+
+ 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)"));
+
+ 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);
+
+ memorystatus_level_snapshot = memorystatus_level;
+
+ killed = memorystatus_do_kill(p, cause);
+
+ /* Success? */
+ if (killed) {
+ proc_rele(p);
+ kill_count++;
+ p = NULL;
+ killed = FALSE;
+
+ /*
+ * Continue the killing spree.
+ */
+ proc_list_lock();
+ 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;
+ }
+
+ /*
+ * Failure - first unwind the state,
+ * then fall through to restart the search.
+ */
+ proc_list_lock();
+ proc_rele_locked(p);
+ if (next_p) {
+ proc_rele_locked(next_p);
+ }
+ p->p_memstat_state &= ~P_MEMSTAT_TERMINATED;
+ p->p_memstat_state |= P_MEMSTAT_ERROR;
+ *errors += 1;
+ p = NULL;
+ }
+
+ /*
+ * Failure - restart the search at the beginning of
+ * the band we were already traversing.
+ *
+ * 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 or P_MEMSTAT_ERROR state.
+ *
+ * And, we hold the the proc_list_lock at this point.
+ */
+
+ next_p = memorystatus_get_first_proc_locked(&i, TRUE);
+ }
+
+ proc_list_unlock();
+
+exit:
+ /* 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;
+ }
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM) | DBG_FUNC_END,
+ memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
+
+ if (kill_count > 0) {
+ return(TRUE);
+ }
+ else {
+ return(FALSE);
+ }
+}
+
+#if LEGACY_HIWATER
+
+static boolean_t
+memorystatus_kill_hiwat_proc(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 = 0;
+ uint32_t aPid_ep;
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_START,
+ memorystatus_available_pages, 0, 0, 0, 0);
+
+ proc_list_lock();
+
+ next_p = memorystatus_get_first_proc_locked(&i, TRUE);
+ while (next_p) {
+ uint32_t footprint;
+ boolean_t skip;
+
+ p = next_p;
+ next_p = memorystatus_get_next_proc_locked(&i, p, TRUE);
+
aPid = p->p_pid;
+ aPid_ep = p->p_memstat_effectivepriority;
if (p->p_memstat_state & (P_MEMSTAT_ERROR | P_MEMSTAT_TERMINATED)) {
continue;
if (p->p_memstat_memlimit <= 0) {
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);
+
#if DEVELOPMENT || DEBUG
if (!skip && (memorystatus_jetsam_policy & kPolicyDiagnoseActive)) {
if (p->p_memstat_state & P_MEMSTAT_DIAG_SUSPENDED) {
(memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing", aPid, p->p_comm, footprint, p->p_memstat_memlimit);
if (memorystatus_jetsam_snapshot_count == 0) {
- memorystatus_jetsam_snapshot_procs_locked();
+ memorystatus_init_jetsam_snapshot_locked(NULL,0);
new_snapshot = TRUE;
}
#if 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_snapshot_locked(p, kMemorystatusKilledDiagnostic);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledDiagnostic);
p->p_memstat_state |= P_MEMSTAT_DIAG_SUSPENDED;
p = proc_ref_locked(p);
} else
#endif /* DEVELOPMENT || DEBUG */
{
- memorystatus_update_snapshot_locked(p, kMemorystatusKilledHiwat);
+ memorystatus_update_jetsam_snapshot_entry_locked(p, kMemorystatusKilledHiwat);
- p = proc_ref_locked(p);
- proc_list_unlock();
- if (p) {
- printf("memorystatus: jetsam killing pid %d [%s] (highwater) - memorystatus_available_pages: %d\n",
- aPid, (p->p_comm ? p->p_comm : "(unknown)"), memorystatus_available_pages);
- killed = memorystatus_do_kill(p, kMemorystatusKilledHiwat);
- }
+ 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);
+
+ killed = memorystatus_do_kill(p, kMemorystatusKilledHiwat);
- /* Success? */
- if (killed) {
- proc_rele(p);
- goto exit;
+ /* 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 - unwind and restart. */
- 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.
+ *
+ * And, we hold the the proc_list_lock at this point.
+ */
+
i = 0;
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
}
}
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_JETSAM_HIWAT) | DBG_FUNC_END,
- memorystatus_available_pages, killed ? aPid : 0, 0, 0, 0);
+ memorystatus_available_pages, killed ? aPid : 0, kill_count, 0, 0);
return killed;
}
if (victim_pid == -1) {
/* No pid, so kill first process */
- res = memorystatus_kill_top_process(TRUE, cause, NULL, &errors);
+ res = memorystatus_kill_top_process(TRUE, TRUE, cause, NULL, &errors);
} else {
res = memorystatus_kill_specific_process(victim_pid, cause);
}
/* Fire off snapshot notification */
size_t snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) +
sizeof(memorystatus_jetsam_snapshot_entry_t) * memorystatus_jetsam_snapshot_count;
- memorystatus_jetsam_snapshot->notification_time = mach_absolute_time();
- memorystatus_send_note(kMemorystatusSnapshotNote, &snapshot_size, sizeof(snapshot_size));
+ 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();
+ }
+ }
}
return res;
{
kern_return_t result;
thread_t thread;
+
+ freezer_lck_grp_attr = lck_grp_attr_alloc_init();
+ freezer_lck_grp = lck_grp_alloc_init("freezer", freezer_lck_grp_attr);
+
+ lck_mtx_init(&freezer_mutex, freezer_lck_grp, NULL);
result = kernel_thread_start(memorystatus_freeze_thread, NULL, &thread);
if (result == KERN_SUCCESS) {
}
}
+/*
+ * Synchronously freeze the passed proc. Called with a reference to the proc held.
+ *
+ * Returns EINVAL or the value returned by task_freeze().
+ */
+int
+memorystatus_freeze_process_sync(proc_t p)
+{
+ int ret = EINVAL;
+ pid_t aPid = 0;
+ boolean_t memorystatus_freeze_swap_low = FALSE;
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_START,
+ memorystatus_available_pages, 0, 0, 0, 0);
+
+ lck_mtx_lock(&freezer_mutex);
+
+ if (p == NULL) {
+ goto exit;
+ }
+
+ if (memorystatus_freeze_enabled == FALSE) {
+ goto exit;
+ }
+
+ if (!memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
+ goto exit;
+ }
+
+ if (memorystatus_freeze_update_throttle()) {
+ printf("memorystatus_freeze_process_sync: in throttle, ignorning freeze\n");
+ memorystatus_freeze_throttle_count++;
+ goto exit;
+ }
+
+ proc_list_lock();
+
+ if (p != NULL) {
+ uint32_t purgeable, wired, clean, dirty, state;
+ uint32_t max_pages, pages, i;
+ boolean_t shared;
+
+ aPid = p->p_pid;
+ state = p->p_memstat_state;
+
+ /* Ensure the process is eligible for freezing */
+ if ((state & (P_MEMSTAT_TERMINATED | P_MEMSTAT_LOCKED | P_MEMSTAT_FROZEN)) || !(state & P_MEMSTAT_SUSPENDED)) {
+ proc_list_unlock();
+ goto exit;
+ }
+
+ /* Only freeze processes meeting our minimum resident page criteria */
+ memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
+ if (pages < memorystatus_freeze_pages_min) {
+ proc_list_unlock();
+ goto exit;
+ }
+
+ if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+
+ 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;
+ }
+
+ max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
+
+ if (max_pages < memorystatus_freeze_pages_min) {
+ proc_list_unlock();
+ goto exit;
+ }
+ } else {
+ /*
+ * We only have the compressor without any swap.
+ */
+ max_pages = UINT32_MAX - 1;
+ }
+
+ /* Mark as locked temporarily to avoid kill */
+ p->p_memstat_state |= P_MEMSTAT_LOCKED;
+ proc_list_unlock();
+
+ ret = task_freeze(p->task, &purgeable, &wired, &clean, &dirty, max_pages, &shared, FALSE);
+
+ 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());
+
+ proc_list_lock();
+ p->p_memstat_state &= ~P_MEMSTAT_LOCKED;
+
+ if (ret == KERN_SUCCESS) {
+ memorystatus_freeze_entry_t data = { aPid, TRUE, dirty };
+
+ 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) {
+ /* Update stats */
+ for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
+ throttle_intervals[i].pageouts += dirty;
+ }
+ }
+
+ memorystatus_freeze_pageouts += dirty;
+ memorystatus_freeze_count++;
+
+ proc_list_unlock();
+
+ memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
+ } else {
+ proc_list_unlock();
+ }
+ }
+
+exit:
+ lck_mtx_unlock(&freezer_mutex);
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_FREEZE) | DBG_FUNC_END,
+ memorystatus_available_pages, aPid, 0, 0, 0);
+
+ return ret;
+}
+
static int
memorystatus_freeze_top_process(boolean_t *memorystatus_freeze_swap_low)
{
}
if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
- /* Ensure there's enough free space to freeze this process. */
- max_pages = MIN(default_pager_swap_pages_free(), memorystatus_freeze_pages_max);
+
+ /* 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;
+ }
+
+ max_pages = MIN(avail_swap_space, memorystatus_freeze_pages_max);
+
if (max_pages < memorystatus_freeze_pages_min) {
*memorystatus_freeze_swap_low = TRUE;
proc_list_unlock();
goto exit;
}
} else {
+ /*
+ * We only have the compressor pool.
+ */
max_pages = UINT32_MAX - 1;
}
p->p_memstat_state |= (P_MEMSTAT_FROZEN | (shared ? 0: P_MEMSTAT_NORECLAIM));
- /* Update stats */
- for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
- throttle_intervals[i].pageouts += dirty;
+ if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
+ /* Update stats */
+ for (i = 0; i < sizeof(throttle_intervals) / sizeof(struct throttle_interval_t); i++) {
+ throttle_intervals[i].pageouts += dirty;
+ }
}
-
+
memorystatus_freeze_pageouts += dirty;
memorystatus_freeze_count++;
memorystatus_send_note(kMemorystatusFreezeNote, &data, sizeof(data));
- /* Return the number of reclaimed pages */
- ret = dirty;
+ /* Return KERN_SUCESS */
+ ret = kr;
} else {
proc_list_unlock();
static boolean_t
memorystatus_can_freeze(boolean_t *memorystatus_freeze_swap_low)
{
+ boolean_t can_freeze = TRUE;
+
/* Only freeze if we're sufficiently low on memory; this holds off freeze right
after boot, and is generally is a no-op once we've reached steady state. */
if (memorystatus_available_pages > memorystatus_freeze_threshold) {
return FALSE;
}
- /* Is swap running low? */
- if (*memorystatus_freeze_swap_low) {
- /* If there's been no movement in free swap pages since we last attempted freeze, return. */
- if (default_pager_swap_pages_free() < memorystatus_freeze_pages_min) {
- return FALSE;
+ if (COMPRESSED_PAGER_IS_SWAPLESS || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS) {
+ /*
+ * 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;
+ }
+
+ can_freeze = FALSE;
+
+ } else {
+ if (*memorystatus_freeze_swap_low) {
+ *memorystatus_freeze_swap_low = FALSE;
+ }
+
+ can_freeze = TRUE;
+ }
+ } else {
+ /*
+ * Freezing WITH on-disk swap support.
+ */
+
+ 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;
+ }
+
+ } 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");
}
-
- /* Pages have been freed - we can retry. */
- *memorystatus_freeze_swap_low = FALSE;
}
- /* OK */
- return TRUE;
+ return can_freeze;
}
static void
memorystatus_freeze_update_throttle_interval(mach_timespec_t *ts, struct throttle_interval_t *interval)
{
+ unsigned int freeze_daily_pageouts_max = memorystatus_freeze_daily_mb_max * (1024 * 1024 / PAGE_SIZE);
if (CMP_MACH_TIMESPEC(ts, &interval->ts) >= 0) {
if (!interval->max_pageouts) {
- interval->max_pageouts = (interval->burst_multiple * (((uint64_t)interval->mins * FREEZE_DAILY_PAGEOUTS_MAX) / (24 * 60)));
+ interval->max_pageouts = (interval->burst_multiple * (((uint64_t)interval->mins * freeze_daily_pageouts_max) / (24 * 60)));
} else {
printf("memorystatus_freeze_update_throttle_interval: %d minute throttle timeout, resetting\n", interval->mins);
}
memorystatus_freeze_thread(void *param __unused, wait_result_t wr __unused)
{
static boolean_t memorystatus_freeze_swap_low = FALSE;
-
+
+ lck_mtx_lock(&freezer_mutex);
if (memorystatus_freeze_enabled) {
if (memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
- /* Only freeze if we've not exceeded our pageout budgets or we're not backed by swap. */
- if (DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPLESS ||
- !memorystatus_freeze_update_throttle()) {
+ /* Only freeze if we've not exceeded our pageout budgets.*/
+ if (!memorystatus_freeze_update_throttle()) {
memorystatus_freeze_top_process(&memorystatus_freeze_swap_low);
} else {
printf("memorystatus_freeze_thread: in throttle, ignoring freeze\n");
}
}
}
+ lck_mtx_unlock(&freezer_mutex);
assert_wait((event_t) &memorystatus_freeze_wakeup, THREAD_UNINT);
thread_block((thread_continue_t) memorystatus_freeze_thread);
memorystatus_warn_process(pid_t pid, boolean_t critical) {
boolean_t ret = FALSE;
+ boolean_t found_knote = FALSE;
struct knote *kn = NULL;
/*
*/
memorystatus_klist_lock();
- kn = vm_find_knote_from_pid(pid, &memorystatus_klist);
- if (kn) {
- /*
- * By setting the "fflags" here, we are forcing
- * a process to deal with the case where it's
- * bumping up into its memory limits. If we don't
- * do this here, we will end up depending on the
- * system pressure snapshot evaluation in
- * filt_memorystatus().
- */
+
+ 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) {
+ /*
+ * By setting the "fflags" here, we are forcing
+ * a process to deal with the case where it's
+ * bumping up into its memory limits. If we don't
+ * do this here, we will end up depending on the
+ * system pressure snapshot evaluation in
+ * filt_memorystatus().
+ */
- if (critical) {
- kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
- } else {
- kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ if (critical) {
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
+ } else if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ }
+ } else {
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ }
+ }
+
+ found_knote = TRUE;
}
- KNOTE(&memorystatus_klist, kMemorystatusPressure);
- ret = TRUE;
+ }
+
+ if (found_knote) {
+ KNOTE(&memorystatus_klist, 0);
+ ret = TRUE;
} else {
if (vm_dispatch_pressure_note_to_pid(pid, FALSE) == 0) {
ret = TRUE;
}
}
+
memorystatus_klist_unlock();
return ret;
}
+/*
+ * Can only be set by the current task on itself.
+ */
+int
+memorystatus_low_mem_privileged_listener(uint32_t op_flags)
+{
+ boolean_t set_privilege = FALSE;
+ /*
+ * Need an entitlement check here?
+ */
+ if (op_flags == MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE) {
+ set_privilege = TRUE;
+ } else if (op_flags == MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE) {
+ set_privilege = FALSE;
+ } else {
+ return EINVAL;
+ }
+
+ return (task_low_mem_privileged_listener(current_task(), set_privilege, NULL));
+}
+
int
memorystatus_send_pressure_note(pid_t pid) {
MEMORYSTATUS_DEBUG(1, "memorystatus_send_pressure_note(): pid %d\n", pid);
memorystatus_send_low_swap_note(void) {
struct knote *kn = NULL;
-
+
memorystatus_klist_lock();
SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
+ /* We call is_knote_registered_modify_task_pressure_bits to check if the sfflags for the
+ * current note contain NOTE_MEMORYSTATUS_LOW_SWAP. Once we find one note in the memorystatus_klist
+ * that has the NOTE_MEMORYSTATUS_LOW_SWAP flags in its sfflags set, we call KNOTE with
+ * kMemoryStatusLowSwap as the hint to process and update all knotes on the memorystatus_klist accordingly. */
if (is_knote_registered_modify_task_pressure_bits(kn, NOTE_MEMORYSTATUS_LOW_SWAP, NULL, 0, 0) == TRUE) {
- KNOTE(&memorystatus_klist, kMemorystatusLowSwap);
+ KNOTE(&memorystatus_klist, kMemorystatusLowSwap);
+ break;
}
}
+
memorystatus_klist_unlock();
}
memorystatus_update_vm_pressure(boolean_t target_foreground_process)
{
struct knote *kn_max = NULL;
+ struct knote *kn_cur = NULL, *kn_temp = NULL; /* for safe list traversal */
pid_t target_pid = -1;
struct klist dispatch_klist = { NULL };
proc_t target_proc = PROC_NULL;
break;
}
idle_kill_counter++;
- delay(1000000); /* 1 second */
+
+ if (memorystatus_manual_testing_on == TRUE) {
+ /*
+ * Skip the delay when testing
+ * the pressure notification scheme.
+ */
+ } else {
+ delay(1000000); /* 1 second */
+ }
}
#endif /* !CONFIG_JETSAM */
continue;
}
proc_list_unlock();
- memorystatus_klist_unlock();
target_pid = target_proc->p_pid;
}
if (found_candidate == FALSE) {
+ proc_rele(target_proc);
+ memorystatus_klist_unlock();
continue;
}
- memorystatus_klist_lock();
- KNOTE_DETACH(&memorystatus_klist, kn_max);
- KNOTE_ATTACH(&dispatch_klist, kn_max);
- memorystatus_klist_unlock();
+ 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);
+ }
+ }
KNOTE(&dispatch_klist, (level_snapshot != kVMPressureNormal) ? kMemorystatusPressure : kMemorystatusNoPressure);
- memorystatus_klist_lock();
- KNOTE_DETACH(&dispatch_klist, kn_max);
- KNOTE_ATTACH(&memorystatus_klist, kn_max);
+ SLIST_FOREACH_SAFE(kn_cur, &dispatch_klist, kn_selnext, kn_temp) {
+ KNOTE_DETACH(&dispatch_klist, kn_cur);
+ KNOTE_ATTACH(&memorystatus_klist, kn_cur);
+ }
+
memorystatus_klist_unlock();
microuptime(&target_proc->vm_pressure_last_notify_tstamp);
list_entry->priority = p->p_memstat_effectivepriority;
list_entry->user_data = p->p_memstat_userdata;
#if LEGACY_HIWATER
- if (((p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND) && (p->p_memstat_effectivepriority >= JETSAM_PRIORITY_FOREGROUND)) ||
- (p->p_memstat_memlimit <= 0)) {
- task_get_phys_footprint_limit(p->task, &list_entry->limit);
- } else {
- list_entry->limit = p->p_memstat_memlimit;
- }
+
+ /*
+ * 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
#endif
}
+/*
+ * Get the at_boot snapshot
+ */
+static int
+memorystatus_get_at_boot_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
+ size_t input_size = *snapshot_size;
+
+ /*
+ * The at_boot snapshot has no entry list.
+ */
+ *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t);
+
+ if (size_only) {
+ return 0;
+ }
+
+ /*
+ * Validate the size of the snapshot buffer
+ */
+ if (input_size < *snapshot_size) {
+ return EINVAL;
+ }
+
+ /*
+ * Update the notification_time only
+ */
+ memorystatus_at_boot_snapshot.notification_time = mach_absolute_time();
+ *snapshot = &memorystatus_at_boot_snapshot;
+
+ MEMORYSTATUS_DEBUG(7, "memorystatus_get_at_boot_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%d)\n",
+ (long)input_size, (long)*snapshot_size, 0);
+ return 0;
+}
+
static int
-memorystatus_get_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
+memorystatus_get_on_demand_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
size_t input_size = *snapshot_size;
+ uint32_t ods_list_count = memorystatus_list_count;
+ memorystatus_jetsam_snapshot_t *ods = NULL; /* The on_demand snapshot buffer */
+
+ *snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (ods_list_count));
+
+ if (size_only) {
+ return 0;
+ }
+
+ /*
+ * Validate the size of the snapshot buffer.
+ * This is inherently racey. May want to revisit
+ * this error condition and trim the output when
+ * it doesn't fit.
+ */
+ if (input_size < *snapshot_size) {
+ return EINVAL;
+ }
+
+ /*
+ * Allocate and initialize a snapshot buffer.
+ */
+ ods = (memorystatus_jetsam_snapshot_t *)kalloc(*snapshot_size);
+ if (!ods) {
+ return (ENOMEM);
+ }
+
+ memset(ods, 0, *snapshot_size);
+
+ proc_list_lock();
+ memorystatus_init_jetsam_snapshot_locked(ods, ods_list_count);
+ proc_list_unlock();
+
+ /*
+ * Return the kernel allocated, on_demand buffer.
+ * The caller of this routine will copy the data out
+ * to user space and then free the kernel allocated
+ * buffer.
+ */
+ *snapshot = ods;
+
+ MEMORYSTATUS_DEBUG(7, "memorystatus_get_on_demand_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
+ (long)input_size, (long)*snapshot_size, (long)ods_list_count);
+ return 0;
+}
+
+static int
+memorystatus_get_jetsam_snapshot(memorystatus_jetsam_snapshot_t **snapshot, size_t *snapshot_size, boolean_t size_only) {
+ size_t input_size = *snapshot_size;
+
if (memorystatus_jetsam_snapshot_count > 0) {
*snapshot_size = sizeof(memorystatus_jetsam_snapshot_t) + (sizeof(memorystatus_jetsam_snapshot_entry_t) * (memorystatus_jetsam_snapshot_count));
} else {
}
*snapshot = memorystatus_jetsam_snapshot;
-
- MEMORYSTATUS_DEBUG(1, "memorystatus_snapshot: returning %ld for size\n", (long)*snapshot_size);
-
+
+ MEMORYSTATUS_DEBUG(7, "memorystatus_get_jetsam_snapshot: returned inputsize (%ld), snapshot_size(%ld), listcount(%ld)\n",
+ (long)input_size, (long)*snapshot_size, (long)memorystatus_jetsam_snapshot_count);
+
return 0;
}
static int
-memorystatus_cmd_get_jetsam_snapshot(user_addr_t buffer, size_t buffer_size, int32_t *retval) {
+memorystatus_cmd_get_jetsam_snapshot(int32_t flags, user_addr_t buffer, size_t buffer_size, int32_t *retval) {
int error = EINVAL;
boolean_t size_only;
+ boolean_t is_default_snapshot = FALSE;
+ boolean_t is_on_demand_snapshot = FALSE;
+ boolean_t is_at_boot_snapshot = FALSE;
memorystatus_jetsam_snapshot_t *snapshot;
-
+
size_only = ((buffer == USER_ADDR_NULL) ? TRUE : FALSE);
-
- error = memorystatus_get_snapshot(&snapshot, &buffer_size, size_only);
+
+ if (flags == 0) {
+ /* Default */
+ is_default_snapshot = TRUE;
+ error = memorystatus_get_jetsam_snapshot(&snapshot, &buffer_size, size_only);
+ } else {
+ if (flags & ~(MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT)) {
+ /*
+ * Unsupported bit set in flag.
+ */
+ return EINVAL;
+ }
+
+ if ((flags & (MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT)) ==
+ (MEMORYSTATUS_SNAPSHOT_ON_DEMAND | MEMORYSTATUS_SNAPSHOT_AT_BOOT)) {
+ /*
+ * Can't have both set at the same time.
+ */
+ return EINVAL;
+ }
+
+ if (flags & MEMORYSTATUS_SNAPSHOT_ON_DEMAND) {
+ is_on_demand_snapshot = TRUE;
+ /*
+ * When not requesting the size only, the following call will allocate
+ * an on_demand snapshot buffer, which is freed below.
+ */
+ error = memorystatus_get_on_demand_snapshot(&snapshot, &buffer_size, size_only);
+
+ } else if (flags & MEMORYSTATUS_SNAPSHOT_AT_BOOT) {
+ is_at_boot_snapshot = TRUE;
+ error = memorystatus_get_at_boot_snapshot(&snapshot, &buffer_size, size_only);
+ } else {
+ /*
+ * Invalid flag setting.
+ */
+ return EINVAL;
+ }
+ }
+
if (error) {
goto out;
}
- /* Copy out and reset */
+ /*
+ * Copy the data out to user space and clear the snapshot buffer.
+ * If working with the jetsam snapshot,
+ * clearing the buffer means, reset the count.
+ * If working with an on_demand snapshot
+ * clearing the buffer means, free it.
+ * If working with the at_boot snapshot
+ * there is nothing to clear or update.
+ */
if (!size_only) {
if ((error = copyout(snapshot, buffer, buffer_size)) == 0) {
- snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
+ if (is_default_snapshot) {
+ /*
+ * The jetsam snapshot is never freed, its count is simply reset.
+ */
+ snapshot->entry_count = memorystatus_jetsam_snapshot_count = 0;
+
+ proc_list_lock();
+ memorystatus_jetsam_snapshot_last_timestamp = 0;
+ proc_list_unlock();
+ }
+ }
+
+ if (is_on_demand_snapshot) {
+ /*
+ * The on_demand snapshot is always freed,
+ * even if the copyout failed.
+ */
+ if(snapshot) {
+ kfree(snapshot, buffer_size);
+ }
}
}
/*
- * This routine is meant solely for the purpose of adjusting jetsam priorities and bands.
- * It is _not_ meant to be used for the setting of memory limits, especially, since we can't
- * tell if the memory limit being set is fatal or not.
- *
- * So the the last 5 args to the memorystatus_update() call below, related to memory limits, are all 0 or FALSE.
+ * This routine is used to update a process's jetsam priority position and stored user_data.
+ * It is not used for the setting of memory limits, which is why the last 6 args to the
+ * memorystatus_update() call are 0 or FALSE.
*/
static int
memorystatus_cmd_set_priority_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
- const uint32_t MAX_ENTRY_COUNT = 2; /* Cap the entry count */
-
- int error;
- uint32_t i;
- uint32_t entry_count;
- memorystatus_priority_properties_t *entries;
-
+ int error = 0;
+ memorystatus_priority_properties_t mpp_entry;
+
/* Validate inputs */
- if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size == 0)) {
+ if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_priority_properties_t))) {
return EINVAL;
}
- /* Make sure the buffer is a multiple of the entry size, and that an excessive size isn't specified */
- entry_count = (buffer_size / sizeof(memorystatus_priority_properties_t));
- if (((buffer_size % sizeof(memorystatus_priority_properties_t)) != 0) || (entry_count > MAX_ENTRY_COUNT)) {
- return EINVAL;
- }
-
- entries = (memorystatus_priority_properties_t *)kalloc(buffer_size);
-
- error = copyin(buffer, entries, buffer_size);
-
- for (i = 0; i < entry_count; i++) {
+ error = copyin(buffer, &mpp_entry, buffer_size);
+
+ if (error == 0) {
proc_t p;
-
- if (error) {
- break;
- }
-
+
p = proc_find(pid);
if (!p) {
- error = ESRCH;
- break;
+ return ESRCH;
}
if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
- error = EPERM;
proc_rele(p);
- break;
+ return EPERM;
}
- error = memorystatus_update(p, entries[i].priority, entries[i].user_data, FALSE, FALSE, 0, 0, FALSE);
+ error = memorystatus_update(p, mpp_entry.priority, mpp_entry.user_data, FALSE, FALSE, 0, 0, FALSE, FALSE, FALSE);
proc_rele(p);
}
- kfree(entries, buffer_size);
-
- return error;
+ return(error);
+}
+
+static int
+memorystatus_cmd_set_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
+ int error = 0;
+ memorystatus_memlimit_properties_t mmp_entry;
+
+ /* Validate inputs */
+ if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
+ return EINVAL;
+ }
+
+ error = copyin(buffer, &mmp_entry, buffer_size);
+
+ if (error == 0) {
+ error = memorystatus_set_memlimit_properties(pid, &mmp_entry);
+ }
+
+ return(error);
+}
+
+/*
+ * When getting the memlimit settings, we can't simply call task_get_phys_footprint_limit().
+ * That gets the proc's cached memlimit and there is no guarantee that the active/inactive
+ * limits will be the same in the no-limit case. Instead we convert limits <= 0 using
+ * task_convert_phys_footprint_limit(). It computes the same limit value that would be written
+ * to the task's ledgers via task_set_phys_footprint_limit().
+ */
+static int
+memorystatus_cmd_get_memlimit_properties(pid_t pid, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
+ int error = 0;
+ memorystatus_memlimit_properties_t mmp_entry;
+
+ /* Validate inputs */
+ if ((pid == 0) || (buffer == USER_ADDR_NULL) || (buffer_size != sizeof(memorystatus_memlimit_properties_t))) {
+ return EINVAL;
+ }
+
+ 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);
}
+
static int
memorystatus_cmd_get_pressure_status(int32_t *retval) {
int error;
return error;
}
+int
+memorystatus_get_pressure_status_kdp() {
+ return (kVMPressureNormal != memorystatus_vm_pressure_level) ? 1 : 0;
+}
+
/*
* Every process, including a P_MEMSTAT_INTERNAL process (currently only pid 1), is allowed to set a HWM.
+ *
+ * This call is inflexible -- it does not distinguish between active/inactive, fatal/non-fatal
+ * So, with 2-level HWM preserving previous behavior will map as follows.
+ * - treat the limit passed in as both an active and inactive limit.
+ * - treat the is_fatal_limit flag as though it applies to both active and inactive limits.
+ *
+ * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK
+ * - the is_fatal_limit is FALSE, meaning the active and inactive limits are non-fatal/soft
+ * - so mapping is (active/non-fatal, inactive/non-fatal)
+ *
+ * When invoked via MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT
+ * - the is_fatal_limit is TRUE, meaning the process's active and inactive limits are fatal/hard
+ * - so mapping is (active/fatal, inactive/fatal)
*/
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;
+ memorystatus_memlimit_properties_t entry;
+
+ entry.memlimit_active = high_water_mark;
+ entry.memlimit_active_attr = 0;
+ entry.memlimit_inactive = high_water_mark;
+ entry.memlimit_inactive_attr = 0;
+
+ if (is_fatal_limit == TRUE) {
+ entry.memlimit_active_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
+ entry.memlimit_inactive_attr |= MEMORYSTATUS_MEMLIMIT_ATTR_FATAL;
+ }
+
+ error = memorystatus_set_memlimit_properties(pid, &entry);
+ return (error);
+}
+
+static int
+memorystatus_set_memlimit_properties(pid_t pid, memorystatus_memlimit_properties_t *entry) {
+
+ int32_t memlimit_active;
+ boolean_t memlimit_active_is_fatal;
+ int32_t memlimit_inactive;
+ boolean_t memlimit_inactive_is_fatal;
+ uint32_t valid_attrs = 0;
+ int error = 0;
proc_t p = proc_find(pid);
if (!p) {
return ESRCH;
}
-
- if (high_water_mark <= 0) {
- high_water_mark = -1; /* Disable */
+
+ /*
+ * Check for valid attribute flags.
+ */
+ valid_attrs |= (MEMORYSTATUS_MEMLIMIT_ATTR_FATAL);
+ if ((entry->memlimit_active_attr & (~valid_attrs)) != 0) {
+ proc_rele(p);
+ return EINVAL;
+ }
+ if ((entry->memlimit_inactive_attr & (~valid_attrs)) != 0) {
+ proc_rele(p);
+ return EINVAL;
}
-
- proc_list_lock();
-
- p->p_memstat_memlimit = high_water_mark;
- if (memorystatus_highwater_enabled) {
- if (p->p_memstat_state & P_MEMSTAT_MEMLIMIT_BACKGROUND) {
- memorystatus_update_priority_locked(p, p->p_memstat_effectivepriority, false);
-
- /*
- * The update priority call above takes care to set/reset the fatal memory limit state
- * IF the process is transitioning between foreground <-> background and has a background
- * memory limit.
- * Here, however, the process won't be doing any such transitions and so we explicitly tackle
- * the fatal limit state.
- */
- is_fatal_limit = FALSE;
+ /*
+ * Setup the active memlimit properties
+ */
+ memlimit_active = entry->memlimit_active;
+ if (entry->memlimit_active_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL) {
+ memlimit_active_is_fatal = TRUE;
+ } else {
+ memlimit_active_is_fatal = FALSE;
+ }
- } else {
- error = (task_set_phys_footprint_limit_internal(p->task, high_water_mark, NULL, TRUE) == 0) ? 0 : EINVAL;
- }
+ /*
+ * Setup the inactive memlimit properties
+ */
+ memlimit_inactive = entry->memlimit_inactive;
+ if (entry->memlimit_inactive_attr & MEMORYSTATUS_MEMLIMIT_ATTR_FATAL) {
+ memlimit_inactive_is_fatal = TRUE;
+ } else {
+ memlimit_inactive_is_fatal = FALSE;
}
- if (error == 0) {
- if (is_fatal_limit == TRUE) {
- p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
+ /*
+ * Setting a limit of <= 0 implies that the process has no
+ * high-water-mark and has no per-task-limit. That means
+ * the system_wide task limit is in place, which by the way,
+ * is always fatal.
+ */
+
+ if (memlimit_active <= 0) {
+ /*
+ * Enforce the fatal system_wide task limit while process is active.
+ */
+ memlimit_active = -1;
+ memlimit_active_is_fatal = TRUE;
+ }
+
+ if (memlimit_inactive <= 0) {
+ /*
+ * Enforce the fatal system_wide task limit while process is inactive.
+ */
+ memlimit_inactive = -1;
+ memlimit_inactive_is_fatal = TRUE;
+ }
+
+ proc_list_lock();
+
+ /*
+ * Store the active limit variants in the proc.
+ */
+ SET_ACTIVE_LIMITS_LOCKED(p, memlimit_active, memlimit_active_is_fatal);
+
+ /*
+ * Store the inactive limit variants in the proc.
+ */
+ SET_INACTIVE_LIMITS_LOCKED(p, memlimit_inactive, memlimit_inactive_is_fatal);
+
+ /*
+ * Enforce appropriate limit variant by updating the cached values
+ * and writing the ledger.
+ * Limit choice is based on process active/inactive state.
+ */
+
+ 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.
+ */
+
+ if (proc_jetsam_state_is_active_locked(p) == TRUE) {
+ CACHE_ACTIVE_LIMITS_LOCKED(p, trigger_exception);
} else {
- p->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT;
+ CACHE_INACTIVE_LIMITS_LOCKED(p, trigger_exception);
}
+
+ /* 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;
+
+ 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") : ""));
}
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
case MEMORYSTATUS_CMD_SET_PRIORITY_PROPERTIES:
error = memorystatus_cmd_set_priority_properties(args->pid, args->buffer, args->buffersize, ret);
break;
+ case MEMORYSTATUS_CMD_SET_MEMLIMIT_PROPERTIES:
+ error = memorystatus_cmd_set_memlimit_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_GRP_SET_PROPERTIES:
error = memorystatus_cmd_grp_set_properties((int32_t)args->flags, args->buffer, args->buffersize, ret);
break;
case MEMORYSTATUS_CMD_GET_JETSAM_SNAPSHOT:
- error = memorystatus_cmd_get_jetsam_snapshot(args->buffer, args->buffersize, ret);
+ error = memorystatus_cmd_get_jetsam_snapshot((int32_t)args->flags, args->buffer, args->buffersize, ret);
break;
case MEMORYSTATUS_CMD_GET_PRESSURE_STATUS:
error = memorystatus_cmd_get_pressure_status(ret);
break;
case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
+ /*
+ * This call does not distinguish between active and inactive limits.
+ * Default behavior in 2-level HWM world is to set both.
+ * Non-fatal limit is also assumed for both.
+ */
error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, FALSE);
break;
case MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT:
+ /*
+ * This call does not distinguish between active and inactive limits.
+ * Default behavior in 2-level HWM world is to set both.
+ * Fatal limit is also assumed for both.
+ */
error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
break;
/* Test commands */
case MEMORYSTATUS_CMD_TEST_JETSAM:
error = memorystatus_kill_process_sync(args->pid, kMemorystatusKilled) ? 0 : EINVAL;
break;
+ case MEMORYSTATUS_CMD_TEST_JETSAM_SORT:
+ error = memorystatus_cmd_test_jetsam_sort(args->pid, (int32_t)args->flags);
+ break;
case MEMORYSTATUS_CMD_SET_JETSAM_PANIC_BITS:
error = memorystatus_cmd_set_panic_bits(args->buffer, args->buffersize);
break;
#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;
+ }
+ 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;
+ break;
#endif /* CONFIG_JETSAM */
+ case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_ENABLE:
+ case MEMORYSTATUS_CMD_PRIVILEGED_LISTENER_DISABLE:
+ error = memorystatus_low_mem_privileged_listener(args->command);
+ break;
default:
break;
}
switch (hint) {
case kMemorystatusNoPressure:
if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_NORMAL) {
- kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_NORMAL;
}
break;
case kMemorystatusPressure:
if (memorystatus_vm_pressure_level == kVMPressureWarning || memorystatus_vm_pressure_level == kVMPressureUrgent) {
if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_WARN) {
- kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_WARN;
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_WARN;
}
} else if (memorystatus_vm_pressure_level == kVMPressureCritical) {
if (kn->kn_sfflags & NOTE_MEMORYSTATUS_PRESSURE_CRITICAL) {
- kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
+ kn->kn_fflags = NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
}
}
break;
case kMemorystatusLowSwap:
if (kn->kn_sfflags & NOTE_MEMORYSTATUS_LOW_SWAP) {
- kn->kn_fflags |= NOTE_MEMORYSTATUS_LOW_SWAP;
+ kn->kn_fflags = NOTE_MEMORYSTATUS_LOW_SWAP;
}
break;
default:
if (kn->kn_sfflags & (NOTE_MEMORYSTATUS_PRESSURE_NORMAL | NOTE_MEMORYSTATUS_PRESSURE_WARN | NOTE_MEMORYSTATUS_PRESSURE_CRITICAL | NOTE_MEMORYSTATUS_LOW_SWAP)) {
- if (kn->kn_sfflags & NOTE_MEMORYSTATUS_LOW_SWAP) {
- error = suser(kauth_cred_get(), 0);
- }
+ KNOTE_ATTACH(&memorystatus_klist, kn);
- if (error == 0) {
- KNOTE_ATTACH(&memorystatus_klist, kn);
- }
} else {
error = ENOTSUP;
}
}
#endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
#endif /* 0 */
+
+#if CONFIG_JETSAM
+/* Coalition support */
+
+/* sorting info for a particular priority bucket */
+typedef struct memstat_sort_info {
+ coalition_t msi_coal;
+ uint64_t msi_page_count;
+ pid_t msi_pid;
+ int msi_ntasks;
+} memstat_sort_info_t;
+
+/*
+ * qsort from smallest page count to largest page count
+ *
+ * return < 0 for a < b
+ * 0 for a == b
+ * > 0 for a > b
+ */
+static int memstat_asc_cmp(const void *a, const void *b)
+{
+ const memstat_sort_info_t *msA = (const memstat_sort_info_t *)a;
+ const memstat_sort_info_t *msB = (const memstat_sort_info_t *)b;
+
+ return (int)((uint64_t)msA->msi_page_count - (uint64_t)msB->msi_page_count);
+}
+
+/*
+ * Return the number of pids rearranged during this sort.
+ */
+static int
+memorystatus_sort_by_largest_coalition_locked(unsigned int bucket_index, int coal_sort_order)
+{
+#define MAX_SORT_PIDS 80
+#define MAX_COAL_LEADERS 10
+
+ unsigned int b = bucket_index;
+ int nleaders = 0;
+ int ntasks = 0;
+ proc_t p = NULL;
+ coalition_t coal = COALITION_NULL;
+ int pids_moved = 0;
+ int total_pids_moved = 0;
+ int i;
+
+ /*
+ * The system is typically under memory pressure when in this
+ * path, hence, we want to avoid dynamic memory allocation.
+ */
+ memstat_sort_info_t leaders[MAX_COAL_LEADERS];
+ pid_t pid_list[MAX_SORT_PIDS];
+
+ if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ return(0);
+ }
+
+ /*
+ * Clear the array that holds coalition leader information
+ */
+ for (i=0; i < MAX_COAL_LEADERS; i++) {
+ leaders[i].msi_coal = COALITION_NULL;
+ leaders[i].msi_page_count = 0; /* will hold total coalition page count */
+ leaders[i].msi_pid = 0; /* will hold coalition leader pid */
+ leaders[i].msi_ntasks = 0; /* will hold the number of tasks in a coalition */
+ }
+
+ p = memorystatus_get_first_proc_locked(&b, FALSE);
+ while (p) {
+ if (coalition_is_leader(p->task, COALITION_TYPE_JETSAM, &coal)) {
+ if (nleaders < MAX_COAL_LEADERS) {
+ int coal_ntasks = 0;
+ uint64_t coal_page_count = coalition_get_page_count(coal, &coal_ntasks);
+ leaders[nleaders].msi_coal = coal;
+ leaders[nleaders].msi_page_count = coal_page_count;
+ leaders[nleaders].msi_pid = p->p_pid; /* the coalition leader */
+ leaders[nleaders].msi_ntasks = coal_ntasks;
+ nleaders++;
+ } else {
+ /*
+ * We've hit MAX_COAL_LEADERS meaning we can handle no more coalitions.
+ * Abandoned coalitions will linger at the tail of the priority band
+ * when this sort session ends.
+ * TODO: should this be an assert?
+ */
+ printf("%s: WARNING: more than %d leaders in priority band [%d]\n",
+ __FUNCTION__, MAX_COAL_LEADERS, bucket_index);
+ break;
+ }
+ }
+ p=memorystatus_get_next_proc_locked(&b, p, FALSE);
+ }
+
+ if (nleaders == 0) {
+ /* Nothing to sort */
+ return(0);
+ }
+
+ /*
+ * Sort the coalition leader array, from smallest coalition page count
+ * to largest coalition page count. When inserted in the priority bucket,
+ * smallest coalition is handled first, resulting in the last to be jetsammed.
+ */
+ if (nleaders > 1) {
+ qsort(leaders, nleaders, sizeof(memstat_sort_info_t), memstat_asc_cmp);
+ }
+
+#if 0
+ for (i = 0; i < nleaders; i++) {
+ printf("%s: coal_leader[%d of %d] pid[%d] pages[%llu] ntasks[%d]\n",
+ __FUNCTION__, i, nleaders, leaders[i].msi_pid, leaders[i].msi_page_count,
+ leaders[i].msi_ntasks);
+ }
+#endif
+
+ /*
+ * During coalition sorting, processes in a priority band are rearranged
+ * by being re-inserted at the head of the queue. So, when handling a
+ * list, the first process that gets moved to the head of the queue,
+ * ultimately gets pushed toward the queue tail, and hence, jetsams last.
+ *
+ * So, for example, the coalition leader is expected to jetsam last,
+ * after its coalition members. Therefore, the coalition leader is
+ * inserted at the head of the queue first.
+ *
+ * After processing a coalition, the jetsam order is as follows:
+ * undefs(jetsam first), extensions, xpc services, leader(jetsam last)
+ */
+
+ /*
+ * Coalition members are rearranged in the priority bucket here,
+ * based on their coalition role.
+ */
+ total_pids_moved = 0;
+ for (i=0; i < nleaders; i++) {
+
+ /* a bit of bookkeeping */
+ pids_moved = 0;
+
+ /* Coalition leaders are jetsammed last, so move into place first */
+ pid_list[0] = leaders[i].msi_pid;
+ pids_moved += memorystatus_move_list_locked(bucket_index, pid_list, 1);
+
+ /* xpc services should jetsam after extensions */
+ ntasks = coalition_get_pid_list (leaders[i].msi_coal, COALITION_ROLEMASK_XPC,
+ coal_sort_order, pid_list, MAX_SORT_PIDS);
+
+ if (ntasks > 0) {
+ pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
+ (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
+ }
+
+ /* extensions should jetsam after unmarked processes */
+ ntasks = coalition_get_pid_list (leaders[i].msi_coal, COALITION_ROLEMASK_EXT,
+ coal_sort_order, pid_list, MAX_SORT_PIDS);
+
+ if (ntasks > 0) {
+ pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
+ (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
+ }
+
+ /* undefined coalition members should be the first to jetsam */
+ ntasks = coalition_get_pid_list (leaders[i].msi_coal, COALITION_ROLEMASK_UNDEF,
+ coal_sort_order, pid_list, MAX_SORT_PIDS);
+
+ if (ntasks > 0) {
+ pids_moved += memorystatus_move_list_locked(bucket_index, pid_list,
+ (ntasks <= MAX_SORT_PIDS ? ntasks : MAX_SORT_PIDS));
+ }
+
+#if 0
+ if (pids_moved == leaders[i].msi_ntasks) {
+ /*
+ * All the pids in the coalition were found in this band.
+ */
+ printf("%s: pids_moved[%d] equal total coalition ntasks[%d] \n", __FUNCTION__,
+ pids_moved, leaders[i].msi_ntasks);
+ } else if (pids_moved > leaders[i].msi_ntasks) {
+ /*
+ * Apparently new coalition members showed up during the sort?
+ */
+ printf("%s: pids_moved[%d] were greater than expected coalition ntasks[%d] \n", __FUNCTION__,
+ pids_moved, leaders[i].msi_ntasks);
+ } else {
+ /*
+ * Apparently not all the pids in the coalition were found in this band?
+ */
+ printf("%s: pids_moved[%d] were less than expected coalition ntasks[%d] \n", __FUNCTION__,
+ pids_moved, leaders[i].msi_ntasks);
+ }
+#endif
+
+ total_pids_moved += pids_moved;
+
+ } /* end for */
+
+ return(total_pids_moved);
+}
+
+
+/*
+ * Traverse a list of pids, searching for each within the priority band provided.
+ * If pid is found, move it to the front of the priority band.
+ * Never searches outside the priority band provided.
+ *
+ * Input:
+ * bucket_index - jetsam priority band.
+ * pid_list - pointer to a list of pids.
+ * list_sz - number of pids in the list.
+ *
+ * Pid list ordering is important in that,
+ * pid_list[n] is expected to jetsam ahead of pid_list[n+1].
+ * The sort_order is set by the coalition default.
+ *
+ * Return:
+ * the number of pids found and hence moved within the priority band.
+ */
+static int
+memorystatus_move_list_locked(unsigned int bucket_index, pid_t *pid_list, int list_sz)
+{
+ memstat_bucket_t *current_bucket;
+ int i;
+ int found_pids = 0;
+
+ if ((pid_list == NULL) || (list_sz <= 0)) {
+ return(0);
+ }
+
+ if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ return(0);
+ }
+
+ current_bucket = &memstat_bucket[bucket_index];
+ for (i=0; i < list_sz; i++) {
+ unsigned int b = bucket_index;
+ proc_t p = NULL;
+ proc_t aProc = NULL;
+ pid_t aPid;
+ int list_index;
+
+ list_index = ((list_sz - 1) - i);
+ aPid = pid_list[list_index];
+
+ /* never search beyond bucket_index provided */
+ p = memorystatus_get_first_proc_locked(&b, FALSE);
+ while (p) {
+ if (p->p_pid == aPid) {
+ aProc = p;
+ break;
+ }
+ p = memorystatus_get_next_proc_locked(&b, p, FALSE);
+ }
+
+ if (aProc == NULL) {
+ /* pid not found in this band, just skip it */
+ continue;
+ } else {
+ TAILQ_REMOVE(¤t_bucket->list, aProc, p_memstat_list);
+ TAILQ_INSERT_HEAD(¤t_bucket->list, aProc, p_memstat_list);
+ found_pids++;
+ }
+ }
+ return(found_pids);
+}
+#endif /* CONFIG_JETSAM */
projects / apple / xnu.git / blobdiff