#include <kern/kalloc.h>
#include <kern/assert.h>
#include <kern/debug.h>
-#include <kern/lock.h>
+#include <kern/locks.h>
#include <kern/task.h>
#include <kern/thread.h>
#include <kern/host.h>
#include <sys/kern_memorystatus.h>
+#if CONFIG_JETSAM
+/* For logging clarity */
+static const char *jetsam_kill_cause_name[] = {
+ "" ,
+ "jettisoned" , /* kMemorystatusKilled */
+ "highwater" , /* kMemorystatusKilledHiwat */
+ "vnode-limit" , /* kMemorystatusKilledVnodes */
+ "vm-pageshortage" , /* kMemorystatusKilledVMPageShortage */
+ "vm-thrashing" , /* kMemorystatusKilledVMThrashing */
+ "fc-thrashing" , /* kMemorystatusKilledFCThrashing */
+ "per-process-limit" , /* kMemorystatusKilledPerProcessLimit */
+ "diagnostic" , /* kMemorystatusKilledDiagnostic */
+ "idle-exit" , /* kMemorystatusKilledIdleExit */
+};
+
+/* Does cause indicate vm or fc thrashing? */
+static boolean_t
+is_thrashing(unsigned cause)
+{
+ switch (cause) {
+ case kMemorystatusKilledVMThrashing:
+ case kMemorystatusKilledFCThrashing:
+ return TRUE;
+ default:
+ return FALSE;
+ }
+}
+
+/* Callback into vm_compressor.c to signal that thrashing has been mitigated. */
+extern void vm_thrashing_jetsam_done(void);
+#endif
+
/* These are very verbose printfs(), enable with
* MEMORYSTATUS_DEBUG_LOG
*/
};
enum {
- kMemorystatusNoPressure = 1,
- kMemorystatusPressure = 2
+ kMemorystatusNoPressure = 0x1,
+ kMemorystatusPressure = 0x2,
+ kMemorystatusLowSwap = 0x4
};
/* Idle guard handling */
static void memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clean_state);
static void memorystatus_reschedule_idle_demotion_locked(void);
-static void memorystatus_update_priority_locked(proc_t p, int priority);
+static void memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert);
+
+boolean_t is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
+void memorystatus_send_low_swap_note(void);
int memorystatus_wakeup = 0;
static unsigned int memorystatus_dirty_count = 0;
-#if !CONFIG_JETSAM
-static boolean_t kill_idle_exit = FALSE;
-#endif
-
int
memorystatus_get_level(__unused struct proc *p, struct memorystatus_get_level_args *args, __unused int *ret)
#if CONFIG_JETSAM
+int proc_get_memstat_priority(proc_t, boolean_t);
+
/* Kill processes exceeding their limit either under memory pressure (1), or as soon as possible (0) */
#define LEGACY_HIWATER 1
-static int memorystatus_highwater_enabled = 1;
+static boolean_t memorystatus_idle_snapshot = 0;
-extern unsigned int vm_page_free_count;
-extern unsigned int vm_page_active_count;
-extern unsigned int vm_page_inactive_count;
-extern unsigned int vm_page_throttled_count;
-extern unsigned int vm_page_purgeable_count;
-extern unsigned int vm_page_wire_count;
+static int memorystatus_highwater_enabled = 1;
unsigned int memorystatus_delta = 0;
-static unsigned int memorystatus_available_pages = (unsigned int)-1;
-static unsigned int memorystatus_available_pages_pressure = 0;
-static unsigned int memorystatus_available_pages_critical = 0;
static unsigned int memorystatus_available_pages_critical_base = 0;
-static unsigned int memorystatus_last_foreground_pressure_pages = (unsigned int)-1;
-#if !LATENCY_JETSAM
+//static unsigned int memorystatus_last_foreground_pressure_pages = (unsigned int)-1;
static unsigned int memorystatus_available_pages_critical_idle_offset = 0;
-#endif
#if DEVELOPMENT || DEBUG
static unsigned int memorystatus_jetsam_panic_debug = 0;
static unsigned int memorystatus_jetsam_policy_offset_pages_diagnostic = 0;
#endif
-static boolean_t kill_under_pressure = FALSE;
+static unsigned int memorystatus_thread_wasted_wakeup = 0;
+
+static uint32_t kill_under_pressure_cause = 0;
static memorystatus_jetsam_snapshot_t *memorystatus_jetsam_snapshot;
#define memorystatus_jetsam_snapshot_list memorystatus_jetsam_snapshot->entries
static unsigned int memorystatus_jetsam_snapshot_max = 0;
static void memorystatus_clear_errors(void);
-static void memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint);
-static int memorystatus_send_note(int event_code, void *data, size_t data_length);
+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 uint32_t memorystatus_build_state(proc_t p);
static void memorystatus_update_levels_locked(boolean_t critical_only);
-static boolean_t memorystatus_issue_pressure_kevent(boolean_t pressured);
+//static boolean_t memorystatus_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);
/* VM pressure */
+extern unsigned int vm_page_free_count;
+extern unsigned int vm_page_active_count;
+extern unsigned int vm_page_inactive_count;
+extern unsigned int vm_page_throttled_count;
+extern unsigned int vm_page_purgeable_count;
+extern unsigned int vm_page_wire_count;
+
#if VM_PRESSURE_EVENTS
#include "vm_pressure.h"
-extern boolean_t memorystatus_warn_process(pid_t pid);
+extern boolean_t memorystatus_warn_process(pid_t pid, boolean_t critical);
vm_pressure_level_t memorystatus_vm_pressure_level = kVMPressureNormal;
+#if CONFIG_MEMORYSTATUS
+unsigned int memorystatus_available_pages = (unsigned int)-1;
+unsigned int memorystatus_available_pages_pressure = 0;
+unsigned int memorystatus_available_pages_critical = 0;
+unsigned int memorystatus_frozen_count = 0;
+unsigned int memorystatus_suspended_count = 0;
+
+/*
+ * We use this flag to signal if we have any HWM offenders
+ * on the system. This way we can reduce the number of wakeups
+ * of the memorystatus_thread when the system is between the
+ * "pressure" and "critical" threshold.
+ *
+ * The (re-)setting of this variable is done without any locks
+ * or synchronization simply because it is not possible (currently)
+ * to keep track of HWM offenders that drop down below their memory
+ * limit and/or exit. So, we choose to burn a couple of wasted wakeups
+ * by allowing the unguarded modification of this variable.
+ */
+boolean_t memorystatus_hwm_candidates = 0;
+
+static int memorystatus_send_note(int event_code, void *data, size_t data_length);
+#endif /* CONFIG_MEMORYSTATUS */
+
#endif /* VM_PRESSURE_EVENTS */
/* Freeze */
/* Thresholds */
static unsigned int memorystatus_freeze_threshold = 0;
-static unsigned int memorystatus_freeze_pages_min = FREEZE_PAGES_MIN;
-static unsigned int memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
-
-static unsigned int memorystatus_frozen_count = 0;
+static unsigned int memorystatus_freeze_pages_min = 0;
+static unsigned int memorystatus_freeze_pages_max = 0;
static unsigned int memorystatus_freeze_suspended_threshold = FREEZE_SUSPENDED_THRESHOLD_DEFAULT;
static uint64_t memorystatus_freeze_throttle_count = 0;
-static unsigned int memorystatus_suspended_count = 0;
static unsigned int memorystatus_suspended_footprint_total = 0;
#endif /* CONFIG_FREEZE */
/* Debug */
+extern struct knote *vm_find_knote_from_pid(pid_t, struct klist *);
+
#if DEVELOPMENT || DEBUG
#if CONFIG_JETSAM
}
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;
+ }
+ }
+
p = memorystatus_get_next_proc_locked(&b, p, TRUE);
}
return 0;
}
+SYSCTL_INT(_kern, OID_AUTO, memorystatus_idle_snapshot, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_idle_snapshot, 0, "");
+
SYSCTL_PROC(_kern, OID_AUTO, memorystatus_highwater_enabled, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_highwater_enabled, 0, sysctl_memorystatus_highwater_enable, "I", "");
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages, 0, "");
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical, CTLFLAG_RD|CTLFLAG_LOCKED, &memorystatus_available_pages_critical, 0, "");
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_base, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_base, 0, "");
-#if !LATENCY_JETSAM
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_critical_idle_offset, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_critical_idle_offset, 0, "");
-#endif
/* Diagnostic code */
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_available_pages_pressure, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_available_pages_pressure, 0, "");
-static int
-sysctl_memorystatus_vm_pressure_level SYSCTL_HANDLER_ARGS
-{
-#pragma unused(arg1, arg2, oidp)
- int error = 0;
-
- error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
- if (error)
- return (error);
-
- return SYSCTL_OUT(req, &memorystatus_vm_pressure_level, sizeof(memorystatus_vm_pressure_level));
-}
-SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_LOCKED|CTLFLAG_MASKED,
- 0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
+/*
+ * This routine is used for targeted notifications
+ * regardless of system memory pressure.
+ * "memnote" is the current user.
+ */
static int
sysctl_memorystatus_vm_pressure_send SYSCTL_HANDLER_ARGS
{
#pragma unused(arg1, arg2)
- int error, pid = 0;
+ int error = 0, pid = 0;
+ int ret = 0;
+ struct knote *kn = NULL;
error = sysctl_handle_int(oidp, &pid, 0, req);
if (error || !req->newptr)
return (error);
- return vm_dispatch_pressure_note_to_pid(pid, FALSE);
+ /*
+ * We inspect 3 lists here for targeted notifications:
+ * - memorystatus_klist
+ * - vm_pressure_klist
+ * - vm_pressure_dormant_klist
+ *
+ * The vm_pressure_* lists are tied to the old VM_PRESSURE
+ * notification mechanism. We intend to stop using that
+ * mechanism and, in turn, get rid of the 2 lists and
+ * vm_dispatch_pressure_note_to_pid() too.
+ */
+
+ memorystatus_klist_lock();
+ 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;
+ } else {
+ ret = vm_dispatch_pressure_note_to_pid(pid, FALSE);
+ }
+ memorystatus_klist_unlock();
+
+ return ret;
}
SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_send, CTLTYPE_INT|CTLFLAG_WR|CTLFLAG_LOCKED|CTLFLAG_MASKED,
#endif /* CONFIG_JETSAM */
-#endif /* DEVELOPMENT || DEBUG */
-
#if CONFIG_FREEZE
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_threshold, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_threshold, 0, "");
SYSCTL_UINT(_kern, OID_AUTO, memorystatus_freeze_throttle_enabled, CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_freeze_throttle_enabled, 0, "");
/*
- * Enabled via: <rdar://problem/13248767> Enable the sysctl_memorystatus_freeze/thaw sysctls on Release KC
- *
- * TODO: Manual trigger of freeze and thaw for dev / debug kernels only.
- * <rdar://problem/13248795> Disable/restrict the sysctl_memorystatus_freeze/thaw sysctls on Release KC
+ * Manual trigger of freeze and thaw for dev / debug kernels only.
*/
static int
sysctl_memorystatus_freeze SYSCTL_HANDLER_ARGS
int error, pid = 0;
proc_t p;
+ if (memorystatus_freeze_enabled == FALSE) {
+ return ENOTSUP;
+ }
+
error = sysctl_handle_int(oidp, &pid, 0, req);
if (error || !req->newptr)
return (error);
boolean_t shared;
uint32_t max_pages = 0;
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED) {
max_pages = MIN(default_pager_swap_pages_free(), memorystatus_freeze_pages_max);
} else {
max_pages = UINT32_MAX - 1;
int error, pid = 0;
proc_t p;
+ if (memorystatus_freeze_enabled == FALSE) {
+ return ENOTSUP;
+ }
+
error = sysctl_handle_int(oidp, &pid, 0, req);
if (error || !req->newptr)
return (error);
#endif /* CONFIG_FREEZE */
+#endif /* DEVELOPMENT || DEBUG */
+
extern kern_return_t kernel_thread_start_priority(thread_continue_t continuation,
void *parameter,
integer_t priority,
thread_t *new_thread);
+#if CONFIG_JETSAM
+/*
+ * 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;
+ uint32_t pages = 0, max_pages = 0;
+ memstat_bucket_t *current_bucket;
+
+ if (bucket_index >= MEMSTAT_BUCKET_COUNT) {
+ return;
+ }
+
+ current_bucket = &memstat_bucket[bucket_index];
+
+ p = TAILQ_FIRST(¤t_bucket->list);
+
+ if (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) {
+
+ 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) {
+
+ 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);
+ }
+
+ 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;
+
+ goto restart;
+ }
+ }
+}
+
+#endif /* CONFIG_JETSAM */
+
static proc_t memorystatus_get_first_proc_locked(unsigned int *bucket_index, boolean_t search) {
memstat_bucket_t *current_bucket;
proc_t next_p;
kern_return_t result;
int i;
+#if CONFIG_FREEZE
+ memorystatus_freeze_pages_min = FREEZE_PAGES_MIN;
+ memorystatus_freeze_pages_max = FREEZE_PAGES_MAX;
+#endif
+
nanoseconds_to_absolutetime((uint64_t)DEFERRED_IDLE_EXIT_TIME_SECS * NSEC_PER_SEC, &memorystatus_idle_delay_time);
/* Init buckets */
#if CONFIG_JETSAM
memorystatus_delta = delta_percentage * atop_64(max_mem) / 100;
-#if !LATENCY_JETSAM
memorystatus_available_pages_critical_idle_offset = idle_offset_percentage * atop_64(max_mem) / 100;
-#endif
-
memorystatus_available_pages_critical_base = (critical_threshold_percentage / delta_percentage) * memorystatus_delta;
memorystatus_jetsam_snapshot_max = maxproc;
extern void
vm_wake_compactor_swapper(void);
+/*
+ * The jetsam no frills kill call
+ * Return: 0 on success
+ * error code on failure (EINVAL...)
+ */
+static int
+jetsam_do_kill(proc_t p, int jetsam_flags) {
+ int error = 0;
+ error = exit1_internal(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags);
+ return(error);
+}
+
+/*
+ * Wrapper for processes exiting with memorystatus details
+ */
static boolean_t
memorystatus_do_kill(proc_t p, uint32_t cause) {
- int retval = 0;
+ int error = 0;
+ __unused pid_t victim_pid = p->p_pid;
+
+ KERNEL_DEBUG_CONSTANT( (BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DO_KILL)) | DBG_FUNC_START,
+ victim_pid, cause, vm_page_free_count, 0, 0);
#if CONFIG_JETSAM && (DEVELOPMENT || DEBUG)
if (memorystatus_jetsam_panic_debug & (1 << cause)) {
case kMemorystatusKilledVnodes: jetsam_flags |= P_JETSAM_VNODE; break;
case kMemorystatusKilledVMPageShortage: jetsam_flags |= P_JETSAM_VMPAGESHORTAGE; break;
case kMemorystatusKilledVMThrashing: jetsam_flags |= P_JETSAM_VMTHRASHING; break;
+ case kMemorystatusKilledFCThrashing: jetsam_flags |= P_JETSAM_FCTHRASHING; break;
case kMemorystatusKilledPerProcessLimit: jetsam_flags |= P_JETSAM_PID; break;
case kMemorystatusKilledIdleExit: jetsam_flags |= P_JETSAM_IDLEEXIT; break;
}
- retval = exit1_internal(p, W_EXITCODE(0, SIGKILL), (int *)NULL, FALSE, FALSE, jetsam_flags);
+ error = jetsam_do_kill(p, jetsam_flags);
+
+ 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();
}
-
- return (retval == 0);
+
+ return (error == 0);
}
/*
if (current_time >= p->p_memstat_idledeadline) {
#if DEBUG || DEVELOPMENT
if (!(p->p_memstat_dirty & P_DIRTY_MARKED)) {
- printf("memorystatus_perform_idle_demotion: moving process %d to idle band, but never dirtied (0x%x)!\n", p->p_pid, p->p_memstat_dirty);
+ printf("memorystatus_perform_idle_demotion: moving process %d [%s] to idle band, but never dirtied (0x%x)!\n",
+ p->p_pid, (p->p_comm ? p->p_comm : "(unknown)"), p->p_memstat_dirty);
}
#endif
memorystatus_invalidate_idle_demotion_locked(p, TRUE);
- memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE);
+ memorystatus_update_priority_locked(p, JETSAM_PRIORITY_IDLE, false);
// The prior process has moved out of the demotion bucket, so grab the new head and continue
p = TAILQ_FIRST(&demotion_bucket->list);
static void
memorystatus_schedule_idle_demotion_locked(proc_t p, boolean_t set_state)
{
+ boolean_t present_in_deferred_bucket = FALSE;
+
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ present_in_deferred_bucket = TRUE;
+ }
+
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",
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_DEFER_IN_PROGRESS)) == (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_DEFER_IN_PROGRESS));
+ assert((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED);
if (set_state) {
assert(p->p_memstat_idledeadline == 0);
+ p->p_memstat_dirty |= P_DIRTY_DEFER_IN_PROGRESS;
p->p_memstat_idledeadline = mach_absolute_time() + memorystatus_idle_delay_time;
}
- assert(p->p_memstat_idledeadline);
+ assert(p->p_memstat_idledeadline);
- memorystatus_scheduled_idle_demotions++;
+ if (present_in_deferred_bucket == FALSE) {
+ memorystatus_scheduled_idle_demotions++;
+ }
}
static void
memorystatus_invalidate_idle_demotion_locked(proc_t p, boolean_t clear_state)
{
+ boolean_t present_in_deferred_bucket = FALSE;
+
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ present_in_deferred_bucket = TRUE;
+ assert(p->p_memstat_idledeadline);
+ }
+
MEMORYSTATUS_DEBUG(1, "memorystatus_invalidate_idle_demotion(): invalidating demotion to idle band for process %d (clear_state %d, demotions %d).\n",
p->p_pid, clear_state, memorystatus_scheduled_idle_demotions);
- assert(p->p_memstat_idledeadline);
if (clear_state) {
p->p_memstat_idledeadline = 0;
p->p_memstat_dirty &= ~P_DIRTY_DEFER_IN_PROGRESS;
}
- memorystatus_scheduled_idle_demotions--;
+ if (present_in_deferred_bucket == TRUE) {
+ memorystatus_scheduled_idle_demotions--;
+ }
+
assert(memorystatus_scheduled_idle_demotions >= 0);
}
proc_t p;
demotion_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE_DEFERRED];
p = TAILQ_FIRST(&demotion_bucket->list);
- assert(p && p->p_memstat_idledeadline);
- if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline){
- thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
- memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
+ assert(p && p->p_memstat_idledeadline);
+
+ if (memstat_idle_demotion_deadline != p->p_memstat_idledeadline){
+ thread_call_enter_delayed(memorystatus_idle_demotion_call, p->p_memstat_idledeadline);
+ memstat_idle_demotion_deadline = p->p_memstat_idledeadline;
}
}
}
{
memstat_bucket_t *bucket;
- MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding process %d with priority %d.\n", p->pid, priority);
+ MEMORYSTATUS_DEBUG(1, "memorystatus_list_add(): adding process %d with priority %d.\n", p->p_pid, p->p_memstat_effectivepriority);
if (!locked) {
proc_list_lock();
bucket = &memstat_bucket[p->p_memstat_effectivepriority];
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ assert(bucket->count == memorystatus_scheduled_idle_demotions);
+ }
+
TAILQ_INSERT_TAIL(&bucket->list, p, p_memstat_list);
bucket->count++;
}
static void
-memorystatus_update_priority_locked(proc_t p, int priority)
+memorystatus_update_priority_locked(proc_t p, int priority, boolean_t head_insert)
{
memstat_bucket_t *old_bucket, *new_bucket;
return;
}
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting process %d to priority %d\n", p->p_pid, priority);
+ MEMORYSTATUS_DEBUG(1, "memorystatus_update_priority_locked(): setting process %d to priority %d, inserting at %s\n",
+ p->p_pid, priority, head_insert ? "head" : "tail");
old_bucket = &memstat_bucket[p->p_memstat_effectivepriority];
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ assert(old_bucket->count == (memorystatus_scheduled_idle_demotions + 1));
+ }
+
TAILQ_REMOVE(&old_bucket->list, p, p_memstat_list);
old_bucket->count--;
new_bucket = &memstat_bucket[priority];
- TAILQ_INSERT_TAIL(&new_bucket->list, p, p_memstat_list);
+ if (head_insert)
+ TAILQ_INSERT_HEAD(&new_bucket->list, p, p_memstat_list);
+ 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)) {
+
+ /*
+ * Adjust memory limit based on if the task is going to/from foreground and background.
+ */
+
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;
+ }
}
}
#endif
}
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)
+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)
{
int ret;
+ boolean_t head_insert = false;
#if !CONFIG_JETSAM
-#pragma unused(update_memlimit, memlimit, memlimit_background)
+#pragma unused(update_memlimit, memlimit, memlimit_background, is_fatal_limit)
#endif
MEMORYSTATUS_DEBUG(1, "memorystatus_update: changing process %d: priority %d, user_data 0x%llx\n", p->p_pid, priority, user_data);
} else if (priority == JETSAM_PRIORITY_IDLE_DEFERRED) {
/* JETSAM_PRIORITY_IDLE_DEFERRED is reserved for internal use; if requested, adjust to JETSAM_PRIORITY_IDLE. */
priority = JETSAM_PRIORITY_IDLE;
+ } else if (priority == JETSAM_PRIORITY_IDLE_HEAD) {
+ /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle queue */
+ priority = JETSAM_PRIORITY_IDLE;
+ head_insert = true;
} else if ((priority < 0) || (priority >= MEMSTAT_BUCKET_COUNT)) {
/* Sanity check */
ret = EINVAL;
if (effective && (p->p_memstat_state & P_MEMSTAT_PRIORITYUPDATED)) {
ret = EALREADY;
proc_list_unlock();
- MEMORYSTATUS_DEBUG(1, "memorystatus_update: effective change specified for pid %d, but change already occurred.\n", pid);
+ MEMORYSTATUS_DEBUG(1, "memorystatus_update: effective change specified for pid %d, but change already occurred.\n", p->p_pid);
+ goto out;
+ }
+
+ if ((p->p_memstat_state & P_MEMSTAT_TERMINATED) || ((p->p_listflag & P_LIST_EXITED) != 0)) {
+ /*
+ * This could happen when a process calling posix_spawn() is exiting on the jetsam thread.
+ */
+ ret = EBUSY;
+ proc_list_unlock();
goto out;
}
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;
+
} else {
/* Otherwise, apply now */
if (memorystatus_highwater_enabled) {
task_set_phys_footprint_limit_internal(p->task, (memlimit > 0) ? memlimit : -1, NULL, TRUE);
}
}
+
+ if (is_fatal_limit || memlimit <= 0) {
+ p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
+ } else {
+ p->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT;
+ }
}
#endif
- memorystatus_update_priority_locked(p, priority);
+ /*
+ * 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.
+ */
+
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ }
+
+ memorystatus_update_priority_locked(p, priority, head_insert);
proc_list_unlock();
ret = 0;
int ret;
memstat_bucket_t *bucket;
- MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing process %d\n", pid);
+ MEMORYSTATUS_DEBUG(1, "memorystatus_list_remove: removing process %d\n", p->p_pid);
if (!locked) {
proc_list_lock();
}
assert(!(p->p_memstat_state & P_MEMSTAT_INTERNAL));
-
+
bucket = &memstat_bucket[p->p_memstat_effectivepriority];
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ assert(bucket->count == memorystatus_scheduled_idle_demotions);
+ }
+
TAILQ_REMOVE(&bucket->list, p, p_memstat_list);
bucket->count--;
return FALSE;
}
+ /* '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;
+ }
+
/* Deferral is only relevant if idle exit is specified */
if ((pcontrol & PROC_DIRTY_DEFER) &&
!(pcontrol & PROC_DIRTY_ALLOWS_IDLE_EXIT)) {
priority = p->p_memstat_requestedpriority;
}
- memorystatus_update_priority_locked(p, priority);
+ if (priority != p->p_memstat_effectivepriority) {
+ memorystatus_update_priority_locked(p, priority, false);
+ }
}
/*
memorystatus_dirty_track(proc_t p, uint32_t pcontrol) {
unsigned int old_dirty;
boolean_t reschedule = FALSE;
+ boolean_t already_deferred = FALSE;
+ boolean_t defer_now = FALSE;
int ret;
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_TRACK),
+ p->p_pid, p->p_memstat_dirty, pcontrol, 0, 0);
+
proc_list_lock();
+ if ((p->p_listflag & P_LIST_EXITED) != 0) {
+ /*
+ * Process is on its way out.
+ */
+ ret = EBUSY;
+ goto exit;
+ }
+
if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
ret = EPERM;
goto exit;
p->p_memstat_dirty |= P_DIRTY_ALLOW_IDLE_EXIT;
}
+ if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
+ p->p_memstat_dirty |= P_DIRTY_LAUNCH_IN_PROGRESS;
+ }
+
+ if (old_dirty & P_DIRTY_DEFER_IN_PROGRESS) {
+ already_deferred = TRUE;
+ }
+
/* This can be set and cleared exactly once. */
- if ((pcontrol & PROC_DIRTY_DEFER) && !(old_dirty & P_DIRTY_DEFER)) {
- p->p_memstat_dirty |= (P_DIRTY_DEFER|P_DIRTY_DEFER_IN_PROGRESS);
- } else {
- p->p_memstat_dirty &= ~P_DIRTY_DEFER_IN_PROGRESS;
+ if (pcontrol & PROC_DIRTY_DEFER) {
+
+ if ( !(old_dirty & P_DIRTY_DEFER)) {
+ p->p_memstat_dirty |= P_DIRTY_DEFER;
+ }
+
+ defer_now = TRUE;
}
- MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / deferred %s / dirty %s for process %d\n",
+ MEMORYSTATUS_DEBUG(1, "memorystatus_on_track_dirty(): set idle-exit %s / defer %s / dirty %s for process %d\n",
((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) ? "Y" : "N",
- p->p_memstat_dirty & P_DIRTY_DEFER_IN_PROGRESS ? "Y" : "N",
+ defer_now ? "Y" : "N",
p->p_memstat_dirty & P_DIRTY ? "Y" : "N",
p->p_pid);
/* Kick off or invalidate the idle exit deferment if there's a state transition. */
if (!(p->p_memstat_dirty & P_DIRTY_IS_DIRTY)) {
if (((p->p_memstat_dirty & P_DIRTY_IDLE_EXIT_ENABLED) == P_DIRTY_IDLE_EXIT_ENABLED) &&
- (p->p_memstat_dirty & P_DIRTY_DEFER_IN_PROGRESS) && !(old_dirty & P_DIRTY_DEFER_IN_PROGRESS)) {
+ defer_now && !already_deferred) {
+
+ /*
+ * Request to defer a clean process that's idle-exit enabled
+ * and not already in the jetsam deferred band.
+ */
memorystatus_schedule_idle_demotion_locked(p, TRUE);
reschedule = TRUE;
- } else if (!(p->p_memstat_dirty & P_DIRTY_DEFER_IN_PROGRESS) && (old_dirty & P_DIRTY_DEFER_IN_PROGRESS)) {
+
+ } else if (!defer_now && already_deferred) {
+
+ /*
+ * Either the process is no longer idle-exit enabled OR
+ * there's a request to cancel a currently active deferral.
+ */
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ reschedule = TRUE;
+ }
+ } else {
+
+ /*
+ * We are trying to operate on a dirty process. Dirty processes have to
+ * be removed from the deferred band. The question is do we reset the
+ * deferred state or not?
+ *
+ * This could be a legal request like:
+ * - this process had opted into the JETSAM_DEFERRED band
+ * - but it's now dirty and requests to opt out.
+ * In this case, we remove the process from the band and reset its
+ * state too. It'll opt back in properly when needed.
+ *
+ * OR, this request could be a user-space bug. E.g.:
+ * - this process had opted into the JETSAM_DEFERRED band when clean
+ * - and, then issues another request to again put it into the band except
+ * this time the process is dirty.
+ * The process going dirty, as a transition in memorystatus_dirty_set(), will pull the process out of
+ * the deferred band with its state intact. So our request below is no-op.
+ * But we do it here anyways for coverage.
+ *
+ * memorystatus_update_idle_priority_locked()
+ * single-mindedly treats a dirty process as "cannot be in the deferred band".
+ */
+
+ if (!defer_now && already_deferred) {
memorystatus_invalidate_idle_demotion_locked(p, TRUE);
reschedule = TRUE;
+ } else {
+ memorystatus_invalidate_idle_demotion_locked(p, FALSE);
+ reschedule = TRUE;
}
}
boolean_t now_dirty = FALSE;
MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_set(): %d %d 0x%x 0x%x\n", self, p->p_pid, pcontrol, p->p_memstat_dirty);
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_SET), p->p_pid, self, pcontrol, 0, 0);
proc_list_lock();
+ if ((p->p_listflag & P_LIST_EXITED) != 0) {
+ /*
+ * Process is on its way out.
+ */
+ ret = EBUSY;
+ goto exit;
+ }
+
if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
ret = EPERM;
goto exit;
/* Manage idle exit deferral, if applied */
if ((p->p_memstat_dirty & (P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_DEFER_IN_PROGRESS)) ==
(P_DIRTY_IDLE_EXIT_ENABLED|P_DIRTY_DEFER_IN_PROGRESS)) {
+
+ /*
+ * P_DIRTY_DEFER_IN_PROGRESS means the process is in the deferred band OR it might be heading back
+ * there once it's clean again and has some protection window left.
+ */
+
if (p->p_memstat_dirty & P_DIRTY_IS_DIRTY) {
+ /*
+ * New dirty process i.e. "was_dirty == FALSE && now_dirty == TRUE"
+ *
+ * The process will move from the deferred band to its higher requested
+ * jetsam band. But we don't clear its state i.e. we want to remember that
+ * this process was part of the "deferred" band and will return to it.
+ *
+ * This way, we don't let it age beyond the protection
+ * window when it returns to "clean". All the while giving
+ * it a chance to perform its work while "dirty".
+ *
+ */
memorystatus_invalidate_idle_demotion_locked(p, FALSE);
reschedule = TRUE;
} else {
- /* We evaluate lazily, so reset the idle-deadline if it's expired by the time the process becomes clean. */
+
+ /*
+ * Process is back from "dirty" to "clean".
+ *
+ * Is its timer up OR does it still have some protection
+ * window left?
+ */
+
if (mach_absolute_time() >= p->p_memstat_idledeadline) {
- p->p_memstat_idledeadline = 0;
- p->p_memstat_dirty &= ~P_DIRTY_DEFER_IN_PROGRESS;
+ /*
+ * The process' deadline has expired. It currently
+ * does not reside in the DEFERRED bucket.
+ *
+ * It's on its way to the JETSAM_PRIORITY_IDLE
+ * bucket via memorystatus_update_idle_priority_locked()
+ * below.
+
+ * So all we need to do is reset all the state on the
+ * process that's related to the DEFERRED bucket i.e.
+ * the DIRTY_DEFER_IN_PROGRESS flag and the timer deadline.
+ *
+ */
+
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ reschedule = TRUE;
} else {
+ /*
+ * It still has some protection window left and so
+ * we just re-arm the timer without modifying any
+ * state on the process.
+ */
memorystatus_schedule_idle_demotion_locked(p, FALSE);
reschedule = TRUE;
}
}
int
-memorystatus_dirty_get(proc_t p) {
+memorystatus_dirty_clear(proc_t p, uint32_t pcontrol) {
+
int ret = 0;
-
- proc_list_lock();
+
+ MEMORYSTATUS_DEBUG(1, "memorystatus_dirty_clear(): %d 0x%x 0x%x\n", p->p_pid, pcontrol, p->p_memstat_dirty);
- if (p->p_memstat_dirty & P_DIRTY_TRACK) {
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_DIRTY_CLEAR), p->p_pid, pcontrol, 0, 0, 0);
+
+ proc_list_lock();
+
+ if ((p->p_listflag & P_LIST_EXITED) != 0) {
+ /*
+ * Process is on its way out.
+ */
+ ret = EBUSY;
+ goto exit;
+ }
+
+ if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
+ ret = EPERM;
+ goto exit;
+ }
+
+ if (!(p->p_memstat_dirty & P_DIRTY_TRACK)) {
+ /* Dirty tracking not enabled */
+ ret = EINVAL;
+ goto exit;
+ }
+
+ if (!pcontrol || (pcontrol & (PROC_DIRTY_LAUNCH_IN_PROGRESS | PROC_DIRTY_DEFER)) == 0) {
+ ret = EINVAL;
+ goto exit;
+ }
+
+ if (pcontrol & PROC_DIRTY_LAUNCH_IN_PROGRESS) {
+ p->p_memstat_dirty &= ~P_DIRTY_LAUNCH_IN_PROGRESS;
+ }
+
+ /* This can be set and cleared exactly once. */
+ if (pcontrol & PROC_DIRTY_DEFER) {
+
+ if (p->p_memstat_dirty & P_DIRTY_DEFER) {
+
+ p->p_memstat_dirty &= ~P_DIRTY_DEFER;
+
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ memorystatus_update_idle_priority_locked(p);
+ memorystatus_reschedule_idle_demotion_locked();
+ }
+ }
+
+ ret = 0;
+exit:
+ proc_list_unlock();
+
+ return ret;
+}
+
+int
+memorystatus_dirty_get(proc_t p) {
+ int ret = 0;
+
+ proc_list_lock();
+
+ if (p->p_memstat_dirty & P_DIRTY_TRACK) {
ret |= PROC_DIRTY_TRACKED;
if (p->p_memstat_dirty & P_DIRTY_ALLOW_IDLE_EXIT) {
ret |= PROC_DIRTY_ALLOWS_IDLE_EXIT;
if (p->p_memstat_dirty & P_DIRTY) {
ret |= PROC_DIRTY_IS_DIRTY;
}
+ if (p->p_memstat_dirty & P_DIRTY_LAUNCH_IN_PROGRESS) {
+ ret |= PROC_DIRTY_LAUNCH_IS_IN_PROGRESS;
+ }
}
proc_list_unlock();
{
#if CONFIG_FREEZE
uint32_t pages;
- memorystatus_get_task_page_counts(p->task, &pages, NULL);
+ memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
#endif
proc_list_lock();
#if CONFIG_FREEZE
}
#endif
+#if CONFIG_JETSAM
static void
memorystatus_thread_wake(void) {
thread_wakeup((event_t)&memorystatus_wakeup);
}
+#endif /* CONFIG_JETSAM */
+
+extern void vm_pressure_response(void);
static int
memorystatus_thread_block(uint32_t interval_ms, thread_continue_t continuation)
return thread_block(continuation);
}
-extern boolean_t vm_compressor_thrashing_detected;
-extern uint64_t vm_compressor_total_compressions(void);
-
static void
memorystatus_thread(void *param __unused, wait_result_t wr __unused)
{
#if CONFIG_JETSAM
boolean_t post_snapshot = FALSE;
uint32_t errors = 0;
+ uint32_t hwm_kill = 0;
#endif
if (is_vm_privileged == FALSE) {
KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_SCAN) | DBG_FUNC_START,
memorystatus_available_pages, 0, 0, 0, 0);
- uint32_t cause = vm_compressor_thrashing_detected ? kMemorystatusKilledVMThrashing : kMemorystatusKilledVMPageShortage;
-
- /* Jetsam aware version.
+ /*
+ * Jetsam aware version.
+ *
+ * The VM pressure notification thread is working it's way through clients in parallel.
*
- * If woken under pressure, go down the path of killing:
+ * So, while the pressure notification thread is targeting processes in order of
+ * increasing jetsam priority, we can hopefully reduce / stop it's work by killing
+ * any processes that have exceeded their highwater mark.
*
- * - processes exceeding their highwater mark if no clean victims available
- * - the least recently used process if no highwater mark victims available
+ * If we run out of HWM processes and our available pages drops below the critical threshold, then,
+ * we target the least recently used process in order of increasing jetsam priority (exception: the FG band).
*/
-#if !LATENCY_JETSAM
- while (vm_compressor_thrashing_detected || memorystatus_available_pages <= memorystatus_available_pages_critical) {
-#else
- while (kill_under_pressure) {
- const uint32_t SNAPSHOT_WAIT_TIMEOUT_MS = 100;
- wait_result_t wait_result;
-#endif
+ while (is_thrashing(kill_under_pressure_cause) ||
+ memorystatus_available_pages <= memorystatus_available_pages_pressure) {
boolean_t killed;
int32_t priority;
+ uint32_t cause;
+
+ if (kill_under_pressure_cause) {
+ cause = kill_under_pressure_cause;
+ } else {
+ cause = kMemorystatusKilledVMPageShortage;
+ }
#if LEGACY_HIWATER
/* Highwater */
killed = memorystatus_kill_hiwat_proc(&errors);
if (killed) {
+ hwm_kill++;
post_snapshot = TRUE;
goto done;
+ } else {
+ memorystatus_hwm_candidates = FALSE;
+ }
+
+ /* No highwater processes to kill. Continue or stop for now? */
+ if (!is_thrashing(kill_under_pressure_cause) &&
+ (memorystatus_available_pages > memorystatus_available_pages_critical)) {
+ /*
+ * We are _not_ out of pressure but we are above the critical threshold and there's:
+ * - no compressor thrashing
+ * - no more HWM processes left.
+ * For now, don't kill any other processes.
+ */
+
+ if (hwm_kill == 0) {
+ memorystatus_thread_wasted_wakeup++;
+ }
+
+ break;
}
#endif
/* LRU */
killed = memorystatus_kill_top_process(TRUE, cause, &priority, &errors);
if (killed) {
- if (!kill_under_pressure && (priority != JETSAM_PRIORITY_IDLE)) {
- /* Don't generate logs for steady-state idle-exit kills */
+ /* Don't generate logs for steady-state idle-exit kills (unless overridden for debug) */
+ if ((priority != JETSAM_PRIORITY_IDLE) || memorystatus_idle_snapshot) {
post_snapshot = TRUE;
}
goto done;
}
-
- /* Under pressure and unable to kill a process - panic */
- panic("memorystatus_jetsam_thread: no victim! available pages:%d\n", memorystatus_available_pages);
+
+ if (memorystatus_available_pages <= memorystatus_available_pages_critical) {
+ /* Under pressure and unable to kill a process - panic */
+ panic("memorystatus_jetsam_thread: no victim! available pages:%d\n", memorystatus_available_pages);
+ }
done:
- kill_under_pressure = FALSE;
- vm_compressor_thrashing_detected = FALSE;
-
-#if LATENCY_JETSAM
- KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_LATENCY_COALESCE) | DBG_FUNC_START,
- memorystatus_available_pages, 0, 0, 0, 0);
- thread_wakeup((event_t)&latency_jetsam_wakeup);
- /*
- * Coalesce snapshot reports in the face of repeated jetsams by blocking here with a timeout.
- * If the wait expires, issue the note.
+
+ /*
+ * We do not want to over-kill when thrashing has been detected.
+ * To avoid that, we reset the flag here and notify the
+ * compressor.
*/
- wait_result = memorystatus_thread_block(SNAPSHOT_WAIT_TIMEOUT_MS, THREAD_CONTINUE_NULL);
- KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_LATENCY_COALESCE) | DBG_FUNC_END,
- memorystatus_available_pages, 0, 0, 0, 0);
- if (wait_result != THREAD_AWAKENED) {
- /* Catch-all */
- break;
+ if (is_thrashing(kill_under_pressure_cause)) {
+ kill_under_pressure_cause = 0;
+ vm_thrashing_jetsam_done();
}
-#endif
}
-
+
+ kill_under_pressure_cause = 0;
+
if (errors) {
memorystatus_clear_errors();
}
#if VM_PRESSURE_EVENTS
- memorystatus_update_vm_pressure(TRUE);
+ /*
+ * LD: We used to target the foreground process first and foremost here.
+ * Now, we target all processes, starting from the non-suspended, background
+ * processes first. We will target foreground too.
+ *
+ * memorystatus_update_vm_pressure(TRUE);
+ */
+ //vm_pressure_response();
#endif
if (post_snapshot) {
#else /* CONFIG_JETSAM */
- /* Simple version.
- *
- * Jetsam not enabled, so just kill the first suitable clean process
- * and sleep.
+ /*
+ * Jetsam not enabled
*/
- if (kill_idle_exit) {
- kill_idle_exit_proc();
- kill_idle_exit = FALSE;
- }
-
#endif /* CONFIG_JETSAM */
memorystatus_thread_block(0, memorystatus_thread);
}
#if !CONFIG_JETSAM
+/*
+ * Returns TRUE:
+ * when an idle-exitable proc was killed
+ * Returns FALSE:
+ * when there are no more idle-exitable procs found
+ * when the attempt to kill an idle-exitable proc failed
+ */
boolean_t memorystatus_idle_exit_from_VM(void) {
- kill_idle_exit = TRUE;
- memorystatus_thread_wake();
- return TRUE;
+ return(kill_idle_exit_proc());
}
-#endif
+#endif /* !CONFIG_JETSAM */
#if CONFIG_JETSAM
* (dirty pages + IOKit mappings)
*
* This is invoked for both advisory, non-fatal per-task high watermarks,
- * as well as the fatal system-wide task memory limit.
+ * as well as the fatal task memory limits.
*/
void
memorystatus_on_ledger_footprint_exceeded(boolean_t warning, const int max_footprint_mb)
{
proc_t p = current_proc();
-
- printf("process %d (%s) %s physical memory footprint limit of %d MB\n",
- p->p_pid, p->p_comm,
- warning ? "approaching" : "exceeded",
- max_footprint_mb);
+
+ 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);
+ }
#if VM_PRESSURE_EVENTS
if (warning == TRUE) {
- if (memorystatus_warn_process(p->p_pid) != 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?).\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_memlimit <= 0) {
+ if ((p->p_memstat_state & P_MEMSTAT_FATAL_MEMLIMIT) == P_MEMSTAT_FATAL_MEMLIMIT) {
/*
- * If this process has no high watermark, then we have been invoked because the task
- * has violated the system-wide per-task memory limit.
+ * If this process has no high watermark or has a fatal task limit, then we have been invoked because the task
+ * has violated either the system-wide per-task memory limit OR its own task limit.
*/
if (memorystatus_kill_process_sync(p->p_pid, kMemorystatusKilledPerProcessLimit) != TRUE) {
printf("task_exceeded_footprint: failed to kill the current task (exiting?).\n");
}
+ } else {
+ /*
+ * HWM offender exists. Done without locks or synchronization.
+ * See comment near its declaration for more details.
+ */
+ memorystatus_hwm_candidates = TRUE;
+ }
+}
+
+/*
+ * 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.
+ * We call this routine so that the offending process is killed with
+ * a non-zero exit status.
+ */
+void
+jetsam_on_ledger_cpulimit_exceeded(void)
+{
+ int retval = 0;
+ int jetsam_flags = 0; /* make it obvious */
+ proc_t p = current_proc();
+
+ printf("task_exceeded_cpulimit: killing pid %d [%s]\n",
+ p->p_pid, (p->p_comm ? p->p_comm : "(unknown)"));
+
+ retval = jetsam_do_kill(p, jetsam_flags);
+
+ if (retval) {
+ printf("task_exceeded_cpulimit: failed to kill current task (exiting?).\n");
}
}
static void
-memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint)
+memorystatus_get_task_page_counts(task_t task, uint32_t *footprint, uint32_t *max_footprint, uint32_t *max_footprint_lifetime, uint32_t *purgeable_pages)
{
assert(task);
assert(footprint);
if (max_footprint) {
*max_footprint = (uint32_t)(get_task_phys_footprint_max(task) / PAGE_SIZE_64);
}
-}
-
-static int
-memorystatus_send_note(int event_code, void *data, size_t data_length) {
- int ret;
- struct kev_msg ev_msg;
-
- ev_msg.vendor_code = KEV_VENDOR_APPLE;
- ev_msg.kev_class = KEV_SYSTEM_CLASS;
- ev_msg.kev_subclass = KEV_MEMORYSTATUS_SUBCLASS;
-
- ev_msg.event_code = event_code;
-
- ev_msg.dv[0].data_length = data_length;
- ev_msg.dv[0].data_ptr = data;
- ev_msg.dv[1].data_length = 0;
-
- ret = kev_post_msg(&ev_msg);
- if (ret) {
- printf("%s: kev_post_msg() failed, err %d\n", __func__, ret);
+ if (max_footprint_lifetime) {
+ *max_footprint_lifetime = (uint32_t)(get_task_resident_max(task) / PAGE_SIZE_64);
+ }
+ if (purgeable_pages) {
+ *purgeable_pages = (uint32_t)(get_task_purgeable_size(task) / PAGE_SIZE_64);
}
-
- return ret;
}
+
static void
memorystatus_update_snapshot_locked(proc_t p, uint32_t kill_cause)
{
void memorystatus_pages_update(unsigned int pages_avail)
{
+ memorystatus_available_pages = pages_avail;
+
+#if VM_PRESSURE_EVENTS
+ /*
+ * Since memorystatus_available_pages changes, we should
+ * re-evaluate the pressure levels on the system and
+ * check if we need to wake the pressure thread.
+ * We also update memorystatus_level in that routine.
+ */
+ vm_pressure_response();
+
+ if (memorystatus_available_pages <= memorystatus_available_pages_pressure) {
+
+ if (memorystatus_hwm_candidates || (memorystatus_available_pages <= memorystatus_available_pages_critical)) {
+ memorystatus_thread_wake();
+ }
+ }
+#else /* VM_PRESSURE_EVENTS */
+
boolean_t critical, delta;
if (!memorystatus_delta) {
|| (memorystatus_available_pages >= (pages_avail + memorystatus_delta))) ? TRUE : FALSE;
if (critical || delta) {
- memorystatus_available_pages = pages_avail;
memorystatus_level = memorystatus_available_pages * 100 / atop_64(max_mem);
-
-#if LATENCY_JETSAM
- /* Bail early to avoid excessive wake-ups */
- if (critical) {
- return;
- }
-#endif
-
memorystatus_thread_wake();
}
+#endif /* VM_PRESSURE_EVENTS */
}
static boolean_t
memorystatus_get_snapshot_properties_for_proc_locked(proc_t p, memorystatus_jetsam_snapshot_entry_t *entry)
{
+ clock_sec_t tv_sec;
+ clock_usec_t tv_usec;
+
memset(entry, 0, sizeof(memorystatus_jetsam_snapshot_entry_t));
entry->pid = p->p_pid;
strlcpy(&entry->name[0], p->p_comm, MAXCOMLEN+1);
entry->priority = p->p_memstat_effectivepriority;
- memorystatus_get_task_page_counts(p->task, &entry->pages, &entry->max_pages);
+ memorystatus_get_task_page_counts(p->task, &entry->pages, &entry->max_pages, &entry->max_pages_lifetime, &entry->purgeable_pages);
entry->state = memorystatus_build_state(p);
entry->user_data = p->p_memstat_userdata;
memcpy(&entry->uuid[0], &p->p_uuid[0], sizeof(p->p_uuid));
+ entry->fds = p->p_fd->fd_nfiles;
+
+ absolutetime_to_microtime(get_task_cpu_time(p->task), &tv_sec, &tv_usec);
+ entry->cpu_time.tv_sec = tv_sec;
+ entry->cpu_time.tv_usec = tv_usec;
return TRUE;
}
return FALSE;
}
- printf("memorystatus: specifically killing pid %d [%s] - memorystatus_available_pages: %d\n",
- victim_pid, (p->p_comm ? p->p_comm : "(unknown)"), memorystatus_available_pages);
+ 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();
proc_list_lock();
+ memorystatus_sort_by_largest_process_locked(JETSAM_PRIORITY_FOREGROUND);
+
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
while (next_p) {
#if DEVELOPMENT || DEBUG
}
#endif /* DEVELOPMENT || DEBUG */
+ if (cause == kMemorystatusKilledVnodes)
+ {
+ /*
+ * If the system runs out of vnodes, we systematically jetsam
+ * processes in hopes of stumbling onto a vnode gain that helps
+ * the system recover. The process that happens to trigger
+ * this path has no known relationship to the vnode consumption.
+ * We attempt to safeguard that process e.g: do not jetsam it.
+ */
+
+ if (p == current_proc()) {
+ /* do not jetsam the current process */
+ continue;
+ }
+ }
+
#if CONFIG_FREEZE
boolean_t skip;
boolean_t reclaim_proc = !(p->p_memstat_state & (P_MEMSTAT_LOCKED | P_MEMSTAT_NORECLAIM));
* - the priority was requested *and* the targeted process is not at idle priority
*/
if ((memorystatus_jetsam_snapshot_count == 0) &&
- ((!priority) || (priority && (*priority != JETSAM_PRIORITY_IDLE)))) {
+ (memorystatus_idle_snapshot || ((!priority) || (priority && (*priority != JETSAM_PRIORITY_IDLE))))) {
memorystatus_jetsam_snapshot_procs_locked();
new_snapshot = TRUE;
}
p = proc_ref_locked(p);
proc_list_unlock();
if (p) {
- printf("memorystatus: jetsam killing pid %d [%s] - memorystatus_available_pages: %d\n",
- aPid, (p->p_comm ? p->p_comm : "(unknown)"), memorystatus_available_pages);
+ printf("memorystatus: %s %d [%s] (%s) - memorystatus_available_pages: %d\n",
+ ((p->p_memstat_effectivepriority == 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);
killed = memorystatus_do_kill(p, cause);
}
memorystatus_available_pages, 0, 0, 0, 0);
proc_list_lock();
+ memorystatus_sort_by_largest_process_locked(JETSAM_PRIORITY_FOREGROUND);
next_p = memorystatus_get_first_proc_locked(&i, TRUE);
while (next_p) {
if (skip) {
continue;
} else {
- MEMORYSTATUS_DEBUG(1, "jetsam: %s pid %d [%s] - %d pages > 1 (%d)\n",
- (memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing", aPid, p->p_comm, pages, hiwat);
+ MEMORYSTATUS_DEBUG(1, "jetsam: %s pid %d [%s] - %d Mb > 1 (%d Mb)\n",
+ (memorystatus_jetsam_policy & kPolicyDiagnoseActive) ? "suspending": "killing", aPid, p->p_comm, footprint, p->p_memstat_memlimit);
if (memorystatus_jetsam_snapshot_count == 0) {
memorystatus_jetsam_snapshot_procs_locked();
static boolean_t
memorystatus_kill_process_async(pid_t victim_pid, uint32_t cause) {
/* TODO: allow a general async path */
- if ((victim_pid != -1) || (cause != kMemorystatusKilledVMPageShortage || cause != kMemorystatusKilledVMThrashing)) {
+ if ((victim_pid != -1) || (cause != kMemorystatusKilledVMPageShortage && cause != kMemorystatusKilledVMThrashing &&
+ cause != kMemorystatusKilledFCThrashing)) {
return FALSE;
}
- kill_under_pressure = TRUE;
+ kill_under_pressure_cause = cause;
memorystatus_thread_wake();
return TRUE;
}
}
}
+boolean_t
+memorystatus_kill_on_FC_thrashing(boolean_t async) {
+ if (async) {
+ return memorystatus_kill_process_async(-1, kMemorystatusKilledFCThrashing);
+ } else {
+ return memorystatus_kill_process_sync(-1, kMemorystatusKilledFCThrashing);
+ }
+}
+
boolean_t
memorystatus_kill_on_vnode_limit(void) {
return memorystatus_kill_process_sync(-1, kMemorystatusKilledVnodes);
}
/* Only freeze processes meeting our minimum resident page criteria */
- memorystatus_get_task_page_counts(p->task, &pages, NULL);
+ memorystatus_get_task_page_counts(p->task, &pages, NULL, NULL, NULL);
if (pages < memorystatus_freeze_pages_min) {
continue; // with lock held
}
- if (DEFAULT_FREEZER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ 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);
if (max_pages < memorystatus_freeze_pages_min) {
if (memorystatus_freeze_enabled) {
if (memorystatus_can_freeze(&memorystatus_freeze_swap_low)) {
- /* Only freeze if we've not exceeded our pageout budgets */
- if (!memorystatus_freeze_update_throttle()) {
+ /* 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()) {
memorystatus_freeze_top_process(&memorystatus_freeze_swap_low);
} else {
printf("memorystatus_freeze_thread: in throttle, ignoring freeze\n");
#endif /* CONFIG_FREEZE */
-#if CONFIG_JETSAM && VM_PRESSURE_EVENTS
+#if VM_PRESSURE_EVENTS
-boolean_t
-memorystatus_warn_process(pid_t pid) {
- return (vm_dispatch_pressure_note_to_pid(pid, FALSE) == 0);
-}
+#if CONFIG_MEMORYSTATUS
-static inline boolean_t
-memorystatus_update_pressure_locked(boolean_t *pressured) {
- vm_pressure_level_t old_level, new_level;
-
- old_level = memorystatus_vm_pressure_level;
-
- if (memorystatus_available_pages > memorystatus_available_pages_pressure) {
- /* Too many free pages */
- new_level = kVMPressureNormal;
- }
-#if CONFIG_FREEZE
- else if (memorystatus_frozen_count > 0) {
- /* Frozen processes exist */
- new_level = kVMPressureNormal;
- }
-#endif
- else if (memorystatus_suspended_count > MEMORYSTATUS_SUSPENDED_THRESHOLD) {
- /* Too many supended processes */
- new_level = kVMPressureNormal;
- }
- else if (memorystatus_suspended_count > 0) {
- /* Some suspended processes - warn */
- new_level = kVMPressureWarning;
- }
- else {
- /* Otherwise, pressure level is urgent */
- new_level = kVMPressureUrgent;
- }
-
- *pressured = (new_level != kVMPressureNormal);
+static int
+memorystatus_send_note(int event_code, void *data, size_t data_length) {
+ int ret;
+ struct kev_msg ev_msg;
- /* Did the pressure level change? */
- if (old_level != new_level) {
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_pressure_locked(): memory pressure changed %d -> %d; memorystatus_available_pages: %d\n ",
- old_level, new_level, memorystatus_available_pages);
- memorystatus_vm_pressure_level = new_level;
- return TRUE;
+ ev_msg.vendor_code = KEV_VENDOR_APPLE;
+ ev_msg.kev_class = KEV_SYSTEM_CLASS;
+ ev_msg.kev_subclass = KEV_MEMORYSTATUS_SUBCLASS;
+
+ ev_msg.event_code = event_code;
+
+ ev_msg.dv[0].data_length = data_length;
+ ev_msg.dv[0].data_ptr = data;
+ ev_msg.dv[1].data_length = 0;
+
+ ret = kev_post_msg(&ev_msg);
+ if (ret) {
+ printf("%s: kev_post_msg() failed, err %d\n", __func__, ret);
}
- return FALSE;
+ return ret;
}
-kern_return_t
-memorystatus_update_vm_pressure(boolean_t target_foreground) {
- boolean_t pressure_changed, pressured;
- boolean_t warn = FALSE;
-
- /*
- * Centralised pressure handling routine. Called from:
- * - The main jetsam thread. In this case, we update the pressure level and dispatch warnings to the foreground
- * process *only*, each time the available page % drops.
- * - The pageout scan path. In this scenario, every other registered process is targeted in footprint order.
- *
- * This scheme guarantees delivery to the foreground app, while providing for warnings to the remaining processes
- * driven by the pageout scan.
- */
+boolean_t
+memorystatus_warn_process(pid_t pid, boolean_t critical) {
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_vm_pressure(): foreground %d; available %d, critical %d, pressure %d\n",
- target_foreground, memorystatus_available_pages, memorystatus_available_pages_critical, memorystatus_available_pages_pressure);
+ boolean_t ret = FALSE;
+ struct knote *kn = NULL;
- proc_list_lock();
+ /*
+ * See comment in sysctl_memorystatus_vm_pressure_send.
+ */
- pressure_changed = memorystatus_update_pressure_locked(&pressured);
+ 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().
+ */
- if (pressured) {
- if (target_foreground) {
- if (memorystatus_available_pages != memorystatus_last_foreground_pressure_pages) {
- if (memorystatus_available_pages < memorystatus_last_foreground_pressure_pages) {
- warn = TRUE;
- }
- memorystatus_last_foreground_pressure_pages = memorystatus_available_pages;
- }
+ if (critical) {
+ kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_CRITICAL;
} else {
- warn = TRUE;
+ kn->kn_fflags |= NOTE_MEMORYSTATUS_PRESSURE_WARN;
}
- } else if (pressure_changed) {
- memorystatus_last_foreground_pressure_pages = (unsigned int)-1;
- }
-
- proc_list_unlock();
-
- /* Target foreground processes if specified */
- if (warn) {
- if (target_foreground) {
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_vm_pressure(): invoking vm_find_pressure_foreground_candidates()\n");
- vm_find_pressure_foreground_candidates();
- } else {
- MEMORYSTATUS_DEBUG(1, "memorystatus_update_vm_pressure(): invoking vm_find_pressure_candidate()\n");
- /* Defer to VM code. This can race with the foreground priority, but
- * it's preferable to holding onto locks for an extended period. */
- vm_find_pressure_candidate();
+ KNOTE(&memorystatus_klist, kMemorystatusPressure);
+ ret = TRUE;
+ } else {
+ if (vm_dispatch_pressure_note_to_pid(pid, FALSE) == 0) {
+ ret = TRUE;
}
}
-
- /* Dispatch the global kevent to privileged listeners */
- if (pressure_changed) {
- memorystatus_issue_pressure_kevent(pressured);
- }
+ memorystatus_klist_unlock();
- return KERN_SUCCESS;
+ return ret;
}
int
return memorystatus_send_note(kMemorystatusPressureNote, &pid, sizeof(pid));
}
+void
+memorystatus_send_low_swap_note(void) {
+
+ struct knote *kn = NULL;
+
+ memorystatus_klist_lock();
+ SLIST_FOREACH(kn, &memorystatus_klist, kn_selnext) {
+ if (is_knote_registered_modify_task_pressure_bits(kn, NOTE_MEMORYSTATUS_LOW_SWAP, NULL, 0, 0) == TRUE) {
+ KNOTE(&memorystatus_klist, kMemorystatusLowSwap);
+ }
+ }
+ memorystatus_klist_unlock();
+}
+
boolean_t
memorystatus_bg_pressure_eligible(proc_t p) {
boolean_t eligible = FALSE;
return ((p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND) ||
(p->p_memstat_effectivepriority == JETSAM_PRIORITY_FOREGROUND_SUPPORT));
}
-
-#else /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
+#endif /* CONFIG_MEMORYSTATUS */
/*
* Trigger levels to test the mechanism.
vm_pressure_level_t memorystatus_manual_testing_level = kVMPressureNormal;
extern struct knote *
-vm_pressure_select_optimal_candidate_to_notify(struct klist *, int);
+vm_pressure_select_optimal_candidate_to_notify(struct klist *, int, boolean_t);
extern
-kern_return_t vm_pressure_notification_without_levels(void);
+kern_return_t vm_pressure_notification_without_levels(boolean_t);
extern void vm_pressure_klist_lock(void);
extern void vm_pressure_klist_unlock(void);
* pressure_level_to_set - the task is about to be notified of this new level. Update the task's bit notification information appropriately.
*
*/
-boolean_t
-is_knote_registered_modify_task_pressure_bits(struct knote*, int, task_t, vm_pressure_level_t, vm_pressure_level_t);
boolean_t
is_knote_registered_modify_task_pressure_bits(struct knote *kn_max, int knote_pressure_level, task_t task, vm_pressure_level_t pressure_level_to_clear, vm_pressure_level_t pressure_level_to_set)
return FALSE;
}
-extern kern_return_t vm_pressure_notify_dispatch_vm_clients(void);
+extern kern_return_t vm_pressure_notify_dispatch_vm_clients(boolean_t target_foreground_process);
+
+#define VM_PRESSURE_DECREASED_SMOOTHING_PERIOD 5000 /* milliseconds */
kern_return_t
-memorystatus_update_vm_pressure(boolean_t target_best_process)
+memorystatus_update_vm_pressure(boolean_t target_foreground_process)
{
struct knote *kn_max = NULL;
pid_t target_pid = -1;
struct klist dispatch_klist = { NULL };
proc_t target_proc = PROC_NULL;
- static vm_pressure_level_t level_snapshot = kVMPressureNormal;
struct task *task = NULL;
boolean_t found_candidate = FALSE;
+ static vm_pressure_level_t level_snapshot = kVMPressureNormal;
+ static vm_pressure_level_t prev_level_snapshot = kVMPressureNormal;
+ boolean_t smoothing_window_started = FALSE;
+ struct timeval smoothing_window_start_tstamp = {0, 0};
+ struct timeval curr_tstamp = {0, 0};
+ int elapsed_msecs = 0;
+
+#if !CONFIG_JETSAM
+#define MAX_IDLE_KILLS 100 /* limit the number of idle kills allowed */
+
+ int idle_kill_counter = 0;
+
+ /*
+ * On desktop we take this opportunity to free up memory pressure
+ * by immediately killing idle exitable processes. We use a delay
+ * to avoid overkill. And we impose a max counter as a fail safe
+ * in case daemons re-launch too fast.
+ */
+ while ((memorystatus_vm_pressure_level != kVMPressureNormal) && (idle_kill_counter < MAX_IDLE_KILLS)) {
+ if (memorystatus_idle_exit_from_VM() == FALSE) {
+ /* No idle exitable processes left to kill */
+ break;
+ }
+ idle_kill_counter++;
+ delay(1000000); /* 1 second */
+ }
+#endif /* !CONFIG_JETSAM */
+
while (1) {
/*
*/
level_snapshot = memorystatus_vm_pressure_level;
+ if (prev_level_snapshot > level_snapshot) {
+ /*
+ * Pressure decreased? Let's take a little breather
+ * and see if this condition stays.
+ */
+ if (smoothing_window_started == FALSE) {
+
+ smoothing_window_started = TRUE;
+ microuptime(&smoothing_window_start_tstamp);
+ }
+
+ microuptime(&curr_tstamp);
+ timevalsub(&curr_tstamp, &smoothing_window_start_tstamp);
+ elapsed_msecs = curr_tstamp.tv_sec * 1000 + curr_tstamp.tv_usec / 1000;
+
+ if (elapsed_msecs < VM_PRESSURE_DECREASED_SMOOTHING_PERIOD) {
+
+ delay(INTER_NOTIFICATION_DELAY);
+ continue;
+ }
+ }
+
+ prev_level_snapshot = level_snapshot;
+ smoothing_window_started = FALSE;
+
memorystatus_klist_lock();
- kn_max = vm_pressure_select_optimal_candidate_to_notify(&memorystatus_klist, level_snapshot);
+ kn_max = vm_pressure_select_optimal_candidate_to_notify(&memorystatus_klist, level_snapshot, target_foreground_process);
if (kn_max == NULL) {
memorystatus_klist_unlock();
microuptime(&target_proc->vm_pressure_last_notify_tstamp);
proc_rele(target_proc);
- if (target_best_process == TRUE) {
+ if (memorystatus_manual_testing_on == TRUE && target_foreground_process == TRUE) {
break;
}
try_dispatch_vm_clients:
- if (level_snapshot != kVMPressureNormal) {
- /*
- * Wake up idle-exit thread.
- * Targets one process per invocation.
- *
- * TODO: memorystatus_idle_exit_from_VM should return FALSE once it's
- * done with all idle-exitable processes. Currently, we will exit this
- * loop when we are done with notification clients (level and non-level based)
- * but we may still have some idle-exitable processes around.
- *
+ if (kn_max == NULL && level_snapshot != kVMPressureNormal) {
+ /*
+ * We will exit this loop when we are done with
+ * notification clients (level and non-level based).
*/
- memorystatus_idle_exit_from_VM();
-
- if ((vm_pressure_notify_dispatch_vm_clients() == KERN_FAILURE) && (kn_max == NULL)) {
+ if ((vm_pressure_notify_dispatch_vm_clients(target_foreground_process) == KERN_FAILURE) && (kn_max == NULL)) {
/*
* kn_max == NULL i.e. we didn't find any eligible clients for the level-based notifications
* AND
}
}
- if (memorystatus_manual_testing_on == FALSE) {
- delay(INTER_NOTIFICATION_DELAY);
+ /*
+ * LD: This block of code below used to be invoked in the older memory notification scheme on embedded everytime
+ * a process was sent a memory pressure notification. The "memorystatus_klist" list was used to hold these
+ * privileged listeners. But now we have moved to the newer scheme and are trying to move away from the extra
+ * notifications. So the code is here in case we break compat. and need to send out notifications to the privileged
+ * apps.
+ */
+#if 0
+#endif /* 0 */
+
+ if (memorystatus_manual_testing_on == TRUE) {
+ /*
+ * Testing out the pressure notification scheme.
+ * No need for delays etc.
+ */
+ } else {
+
+ uint32_t sleep_interval = INTER_NOTIFICATION_DELAY;
+#if CONFIG_JETSAM
+ unsigned int page_delta = 0;
+ unsigned int skip_delay_page_threshold = 0;
+
+ assert(memorystatus_available_pages_pressure >= memorystatus_available_pages_critical_base);
+
+ page_delta = (memorystatus_available_pages_pressure - memorystatus_available_pages_critical_base) / 2;
+ skip_delay_page_threshold = memorystatus_available_pages_pressure - page_delta;
+
+ if (memorystatus_available_pages <= skip_delay_page_threshold) {
+ /*
+ * We are nearing the critcal mark fast and can't afford to wait between
+ * notifications.
+ */
+ sleep_interval = 0;
+ }
+#endif /* CONFIG_JETSAM */
+
+ if (sleep_interval) {
+ delay(sleep_interval);
+ }
}
}
sysctl_memorystatus_vm_pressure_level SYSCTL_HANDLER_ARGS
{
#pragma unused(arg1, arg2, oidp)
-
vm_pressure_level_t dispatch_level = convert_internal_pressure_level_to_dispatch_level(memorystatus_vm_pressure_level);
return SYSCTL_OUT(req, &dispatch_level, sizeof(dispatch_level));
}
+#if DEBUG || DEVELOPMENT
+
SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_LOCKED,
0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
+#else /* DEBUG || DEVELOPMENT */
+
+SYSCTL_PROC(_kern, OID_AUTO, memorystatus_vm_pressure_level, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_LOCKED|CTLFLAG_MASKED,
+ 0, 0, &sysctl_memorystatus_vm_pressure_level, "I", "");
+
+#endif /* DEBUG || DEVELOPMENT */
extern int memorystatus_purge_on_warning;
extern int memorystatus_purge_on_critical;
} else {
vm_pressure_klist_lock();
- vm_pressure_notification_without_levels();
+ vm_pressure_notification_without_levels(FALSE);
vm_pressure_klist_unlock();
}
SYSCTL_INT(_kern, OID_AUTO, memorystatus_purge_on_critical, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_LOCKED, &memorystatus_purge_on_critical, 0, "");
-#endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
+#endif /* VM_PRESSURE_EVENTS */
/* Return both allocated and actual size, since there's a race between allocation and list compilation */
static int
static void
memorystatus_update_levels_locked(boolean_t critical_only) {
+
memorystatus_available_pages_critical = memorystatus_available_pages_critical_base;
-#if !LATENCY_JETSAM
- {
- // If there's an entry in the first bucket, we have idle processes
- memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
- if (first_bucket->count) {
- memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
+
+ /*
+ * If there's an entry in the first bucket, we have idle processes.
+ */
+ memstat_bucket_t *first_bucket = &memstat_bucket[JETSAM_PRIORITY_IDLE];
+ if (first_bucket->count) {
+ memorystatus_available_pages_critical += memorystatus_available_pages_critical_idle_offset;
+
+ if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure ) {
+ /*
+ * The critical threshold must never exceed the pressure threshold
+ */
+ memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
}
}
-#endif
+
#if DEBUG || DEVELOPMENT
if (memorystatus_jetsam_policy & kPolicyDiagnoseActive) {
memorystatus_available_pages_critical += memorystatus_jetsam_policy_offset_pages_diagnostic;
+
+ if (memorystatus_available_pages_critical > memorystatus_available_pages_pressure ) {
+ /*
+ * The critical threshold must never exceed the pressure threshold
+ */
+ memorystatus_available_pages_critical = memorystatus_available_pages_pressure;
+ }
}
#endif
return 0;
}
+
static int
memorystatus_cmd_get_jetsam_snapshot(user_addr_t buffer, size_t buffer_size, int32_t *retval) {
int error = EINVAL;
return error;
}
+/*
+ * Routine: memorystatus_cmd_grp_set_properties
+ * Purpose: Update properties for a group of processes.
+ *
+ * Supported Properties:
+ * [priority]
+ * Move each process out of its effective priority
+ * band and into a new priority band.
+ * Maintains relative order from lowest to highest priority.
+ * In single band, maintains relative order from head to tail.
+ *
+ * eg: before [effectivepriority | pid]
+ * [18 | p101 ]
+ * [17 | p55, p67, p19 ]
+ * [12 | p103 p10 ]
+ * [ 7 | p25 ]
+ * [ 0 | p71, p82, ]
+ *
+ * after [ new band | pid]
+ * [ xxx | p71, p82, p25, p103, p10, p55, p67, p19, p101]
+ *
+ * Returns: 0 on success, else non-zero.
+ *
+ * Caveat: We know there is a race window regarding recycled pids.
+ * A process could be killed before the kernel can act on it here.
+ * If a pid cannot be found in any of the jetsam priority bands,
+ * then we simply ignore it. No harm.
+ * But, if the pid has been recycled then it could be an issue.
+ * In that scenario, we might move an unsuspecting process to the new
+ * priority band. It's not clear how the kernel can safeguard
+ * against this, but it would be an extremely rare case anyway.
+ * The caller of this api might avoid such race conditions by
+ * ensuring that the processes passed in the pid list are suspended.
+ */
+
+
+/* This internal structure can expand when we add support for more properties */
+typedef struct memorystatus_internal_properties
+{
+ proc_t proc;
+ int32_t priority; /* see memorytstatus_priority_entry_t : priority */
+} memorystatus_internal_properties_t;
+
+
+static int
+memorystatus_cmd_grp_set_properties(int32_t flags, user_addr_t buffer, size_t buffer_size, __unused int32_t *retval) {
+
+#pragma unused (flags)
+
+ /*
+ * We only handle setting priority
+ * per process
+ */
+
+ int error = 0;
+ memorystatus_priority_entry_t *entries = NULL;
+ uint32_t entry_count = 0;
+
+ /* This will be the ordered proc list */
+ memorystatus_internal_properties_t *table = NULL;
+ size_t table_size = 0;
+ uint32_t table_count = 0;
+
+ uint32_t i = 0;
+ uint32_t bucket_index = 0;
+ boolean_t head_insert;
+ int32_t new_priority;
+
+ proc_t p;
+
+ /* Verify inputs */
+ if ((buffer == USER_ADDR_NULL) || (buffer_size == 0) || ((buffer_size % sizeof(memorystatus_priority_entry_t)) != 0)) {
+ error = EINVAL;
+ goto out;
+ }
+
+ entry_count = (buffer_size / sizeof(memorystatus_priority_entry_t));
+ if ((entries = (memorystatus_priority_entry_t *)kalloc(buffer_size)) == NULL) {
+ error = ENOMEM;
+ goto out;
+ }
+
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_START, entry_count, 0, 0, 0, 0);
+
+ if ((error = copyin(buffer, entries, buffer_size)) != 0) {
+ goto out;
+ }
+
+ /* Verify sanity of input priorities */
+ for (i=0; i < entry_count; i++) {
+ if (entries[i].priority == -1) {
+ /* Use as shorthand for default priority */
+ entries[i].priority = JETSAM_PRIORITY_DEFAULT;
+ } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ /* JETSAM_PRIORITY_IDLE_DEFERRED is reserved for internal use;
+ * if requested, adjust to JETSAM_PRIORITY_IDLE. */
+ entries[i].priority = JETSAM_PRIORITY_IDLE;
+ } else if (entries[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
+ /* JETSAM_PRIORITY_IDLE_HEAD inserts at the head of the idle
+ * queue */
+ /* Deal with this later */
+ } else if ((entries[i].priority < 0) || (entries[i].priority >= MEMSTAT_BUCKET_COUNT)) {
+ /* Sanity check */
+ error = EINVAL;
+ goto out;
+ }
+ }
+
+ table_size = sizeof(memorystatus_internal_properties_t) * entry_count;
+ if ( (table = (memorystatus_internal_properties_t *)kalloc(table_size)) == NULL) {
+ error = ENOMEM;
+ goto out;
+ }
+ memset(table, 0, table_size);
+
+
+ /*
+ * For each jetsam bucket entry, spin through the input property list.
+ * When a matching pid is found, populate an adjacent table with the
+ * appropriate proc pointer and new property values.
+ * This traversal automatically preserves order from lowest
+ * to highest priority.
+ */
+
+ bucket_index=0;
+
+ proc_list_lock();
+
+ /* Create the ordered table */
+ p = memorystatus_get_first_proc_locked(&bucket_index, TRUE);
+ while (p && (table_count < entry_count)) {
+ for (i=0; i < entry_count; i++ ) {
+ if (p->p_pid == entries[i].pid) {
+ /* Build the table data */
+ table[table_count].proc = p;
+ table[table_count].priority = entries[i].priority;
+ table_count++;
+ break;
+ }
+ }
+ p = memorystatus_get_next_proc_locked(&bucket_index, p, TRUE);
+ }
+
+ /* We now have ordered list of procs ready to move */
+ for (i=0; i < table_count; i++) {
+ p = table[i].proc;
+ assert(p != NULL);
+
+ /* Allow head inserts -- but relative order is now */
+ if (table[i].priority == JETSAM_PRIORITY_IDLE_HEAD) {
+ new_priority = JETSAM_PRIORITY_IDLE;
+ head_insert = true;
+ } else {
+ new_priority = table[i].priority;
+ head_insert = false;
+ }
+
+ /* Not allowed */
+ if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
+ continue;
+ }
+
+ /*
+ * Take appropriate steps if moving proc out of the
+ * JETSAM_PRIORITY_IDLE_DEFERRED band.
+ */
+ if (p->p_memstat_effectivepriority == JETSAM_PRIORITY_IDLE_DEFERRED) {
+ memorystatus_invalidate_idle_demotion_locked(p, TRUE);
+ }
+
+ memorystatus_update_priority_locked(p, new_priority, head_insert);
+ }
+
+ proc_list_unlock();
+
+ /*
+ * if (table_count != entry_count)
+ * then some pids were not found in a jetsam band.
+ * harmless but interesting...
+ */
+ KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_MEMSTAT, BSD_MEMSTAT_GRP_SET_PROP) | DBG_FUNC_END, entry_count, table_count, 0, 0, 0);
+
+out:
+ if (entries)
+ kfree(entries, buffer_size);
+ if (table)
+ kfree(table, table_size);
+
+ return (error);
+}
+
+
+/*
+ * 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.
+ */
+
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 */
proc_rele(p);
break;
}
-
- error = memorystatus_update(p, entries[i].priority, entries[i].user_data, FALSE, FALSE, 0, 0);
+
+ error = memorystatus_update(p, entries[i].priority, entries[i].user_data, FALSE, FALSE, 0, 0, FALSE);
proc_rele(p);
}
return error;
}
+/*
+ * Every process, including a P_MEMSTAT_INTERNAL process (currently only pid 1), is allowed to set a HWM.
+ */
+
static int
-memorystatus_cmd_set_jetsam_high_water_mark(pid_t pid, int32_t high_water_mark, __unused int32_t *retval) {
+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;
proc_t p = proc_find(pid);
proc_list_lock();
- if (p->p_memstat_state & P_MEMSTAT_INTERNAL) {
- error = EPERM;
- goto exit;
- }
-
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);
+
+ 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;
+
} else {
error = (task_set_phys_footprint_limit_internal(p->task, high_water_mark, NULL, TRUE) == 0) ? 0 : EINVAL;
}
}
-exit:
+ if (error == 0) {
+ if (is_fatal_limit == TRUE) {
+ p->p_memstat_state |= P_MEMSTAT_FATAL_MEMLIMIT;
+ } else {
+ p->p_memstat_state &= ~P_MEMSTAT_FATAL_MEMLIMIT;
+ }
+ }
+
proc_list_unlock();
proc_rele(p);
return error;
}
+/*
+ * Returns the jetsam priority (effective or requested) of the process
+ * associated with this task.
+ */
+int
+proc_get_memstat_priority(proc_t p, boolean_t effective_priority)
+{
+ if (p) {
+ if (effective_priority) {
+ return p->p_memstat_effectivepriority;
+ } else {
+ return p->p_memstat_requestedpriority;
+ }
+ }
+ return 0;
+}
#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_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);
break;
error = memorystatus_cmd_get_pressure_status(ret);
break;
case MEMORYSTATUS_CMD_SET_JETSAM_HIGH_WATER_MARK:
- /* TODO: deprecate. Keeping it in as there's no pid based way to set the ledger limit right now. */
- error = memorystatus_cmd_set_jetsam_high_water_mark(args->pid, (int32_t)args->flags, ret);
+ error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, FALSE);
+ break;
+ case MEMORYSTATUS_CMD_SET_JETSAM_TASK_LIMIT:
+ error = memorystatus_cmd_set_jetsam_memory_limit(args->pid, (int32_t)args->flags, ret, TRUE);
break;
/* Test commands */
#if DEVELOPMENT || DEBUG
}
}
break;
+ case kMemorystatusLowSwap:
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_LOW_SWAP) {
+ kn->kn_fflags |= NOTE_MEMORYSTATUS_LOW_SWAP;
+ }
+ break;
default:
break;
}
memorystatus_klist_lock();
- if (kn->kn_sfflags & (NOTE_MEMORYSTATUS_PRESSURE_NORMAL | NOTE_MEMORYSTATUS_PRESSURE_WARN | NOTE_MEMORYSTATUS_PRESSURE_CRITICAL)) {
+ if (kn->kn_sfflags & (NOTE_MEMORYSTATUS_PRESSURE_NORMAL | NOTE_MEMORYSTATUS_PRESSURE_WARN | NOTE_MEMORYSTATUS_PRESSURE_CRITICAL | NOTE_MEMORYSTATUS_LOW_SWAP)) {
-#if CONFIG_JETSAM && VM_PRESSURE_EVENTS
- /* Need a privilege to register */
- error = priv_check_cred(kauth_cred_get(), PRIV_VM_PRESSURE, 0);
-#endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
+ if (kn->kn_sfflags & NOTE_MEMORYSTATUS_LOW_SWAP) {
+ error = suser(kauth_cred_get(), 0);
+ }
- if (!error) {
+ if (error == 0) {
KNOTE_ATTACH(&memorystatus_klist, kn);
}
} else {
memorystatus_klist_unlock();
}
+
+#if 0
#if CONFIG_JETSAM && VM_PRESSURE_EVENTS
static boolean_t
memorystatus_issue_pressure_kevent(boolean_t pressured) {
memorystatus_klist_unlock();
return TRUE;
}
-
#endif /* CONFIG_JETSAM && VM_PRESSURE_EVENTS */
+#endif /* 0 */
projects / apple / xnu.git / blobdiff