#include <kern/kern_types.h>
#include <kern/thread.h>
#include <sys/cdefs.h>
+#include <kern/block_hint.h>
#ifdef MACH_KERNEL_PRIVATE
+#include <mach/branch_predicates.h>
+
/* Initialization */
extern void sched_init(void);
extern void sched_timebase_init(void);
+extern void pset_rt_init(processor_set_t pset);
+
+extern void sched_rtglobal_init(processor_set_t pset);
+
+extern rt_queue_t sched_rtglobal_runq(processor_set_t pset);
+
+extern void sched_rtglobal_queue_shutdown(processor_t processor);
+
+extern int64_t sched_rtglobal_runq_count_sum(void);
+
+extern void sched_check_spill(processor_set_t pset, thread_t thread);
+
+extern bool sched_thread_should_yield(processor_t processor, thread_t thread);
+
/* Force a preemption point for a thread and wait for it to stop running */
extern boolean_t thread_stop(
thread_t thread,
thread_t thread,
int priority);
-/* Set the thread to be categorized as 'background' */
-extern void sched_set_thread_throttled(thread_t thread,
- boolean_t wants_throttle);
-
/* Set the thread's true scheduling mode */
extern void sched_set_thread_mode(thread_t thread,
sched_mode_t mode);
extern void sched_thread_mode_undemote(thread_t thread,
uint32_t reason);
+/* Re-evaluate base priority of thread (thread locked) */
+void thread_recompute_priority(thread_t thread);
+
+/* Re-evaluate base priority of thread (thread unlocked) */
+void thread_recompute_qos(thread_t thread);
+
/* Reset scheduled priority of thread */
-extern void compute_priority(
+extern void thread_recompute_sched_pri(
thread_t thread,
boolean_t override_depress);
-/* Adjust scheduled priority of thread during execution */
-extern void compute_my_priority(
- thread_t thread);
-
/* Periodic scheduler activity */
extern void sched_init_thread(void (*)(void));
extern void lightweight_update_priority(
thread_t thread);
-extern void sched_traditional_quantum_expire(thread_t thread);
+extern void sched_default_quantum_expire(thread_t thread);
/* Idle processor thread */
extern void idle_thread(void);
thread_t thread,
integer_t options);
-#define SCHED_TAILQ 1
-#define SCHED_HEADQ 2
-#define SCHED_PREEMPT 4
+typedef enum {
+ SCHED_NONE = 0x0,
+ SCHED_TAILQ = 0x1,
+ SCHED_HEADQ = 0x2,
+ SCHED_PREEMPT = 0x4,
+ SCHED_REBALANCE = 0x8,
+} sched_options_t;
extern processor_set_t task_choose_pset(
task_t task);
processor_t processor,
thread_t thread);
-/* Choose a thread from a processor's priority-based runq */
-extern thread_t choose_thread_from_runq(
- processor_t processor,
- run_queue_t runq,
- int priority);
-
+extern void sched_SMT_balance(
+ processor_t processor,
+ processor_set_t pset);
extern void thread_quantum_init(
thread_t thread);
run_queue_t runq,
thread_t thread);
+struct sched_update_scan_context
+{
+ uint64_t earliest_bg_make_runnable_time;
+ uint64_t earliest_normal_make_runnable_time;
+ uint64_t earliest_rt_make_runnable_time;
+};
+typedef struct sched_update_scan_context *sched_update_scan_context_t;
+
+extern void sched_rtglobal_runq_scan(sched_update_scan_context_t scan_context);
+
+/*
+ * Enum to define various events which need IPIs. The IPI policy
+ * engine decides what kind of IPI to use based on destination
+ * processor state, thread and one of the following scheduling events.
+ */
+typedef enum {
+ SCHED_IPI_EVENT_BOUND_THR = 0x1,
+ SCHED_IPI_EVENT_PREEMPT = 0x2,
+ SCHED_IPI_EVENT_SMT_REBAL = 0x3,
+ SCHED_IPI_EVENT_SPILL = 0x4,
+ SCHED_IPI_EVENT_REBALANCE = 0x5,
+} sched_ipi_event_t;
+
+
+/* Enum to define various IPI types used by the scheduler */
+typedef enum {
+ SCHED_IPI_NONE = 0x0,
+ SCHED_IPI_IMMEDIATE = 0x1,
+ SCHED_IPI_IDLE = 0x2,
+ SCHED_IPI_DEFERRED = 0x3,
+} sched_ipi_type_t;
+
+/* The IPI policy engine behaves in the following manner:
+ * - All scheduler events which need an IPI invoke sched_ipi_action() with
+ * the appropriate destination processor, thread and event.
+ * - sched_ipi_action() performs basic checks, invokes the scheduler specific
+ * ipi_policy routine and sets pending_AST bits based on the result.
+ * - Once the pset lock is dropped, the scheduler invokes sched_ipi_perform()
+ * routine which actually sends the appropriate IPI to the destination core.
+ */
+extern sched_ipi_type_t sched_ipi_action(processor_t dst, thread_t thread,
+ boolean_t dst_idle, sched_ipi_event_t event);
+extern void sched_ipi_perform(processor_t dst, sched_ipi_type_t ipi);
+
+/* sched_ipi_policy() is the global default IPI policy for all schedulers */
+extern sched_ipi_type_t sched_ipi_policy(processor_t dst, thread_t thread,
+ boolean_t dst_idle, sched_ipi_event_t event);
+
+/* sched_ipi_deferred_policy() is the global default deferred IPI policy for all schedulers */
+extern sched_ipi_type_t sched_ipi_deferred_policy(processor_set_t pset,
+ processor_t dst, sched_ipi_event_t event);
#if defined(CONFIG_SCHED_TIMESHARE_CORE)
-extern boolean_t thread_update_add_thread(
- thread_t thread);
+extern boolean_t thread_update_add_thread(thread_t thread);
extern void thread_update_process_threads(void);
-extern boolean_t runq_scan(
- run_queue_t runq);
+extern boolean_t runq_scan(run_queue_t runq, sched_update_scan_context_t scan_context);
+
+extern void sched_timeshare_init(void);
+extern void sched_timeshare_timebase_init(void);
+extern void sched_timeshare_maintenance_continue(void);
+
+extern boolean_t priority_is_urgent(int priority);
+extern uint32_t sched_timeshare_initial_quantum_size(thread_t thread);
-void sched_traditional_timebase_init(void);
-void sched_traditional_maintenance_continue(void);
-boolean_t priority_is_urgent(
- int priority);
-uint32_t sched_traditional_initial_quantum_size(
- thread_t thread);
-void sched_traditional_init(void);
+extern int sched_compute_timeshare_priority(thread_t thread);
#endif /* CONFIG_SCHED_TIMESHARE_CORE */
/* Remove thread from its run queue */
-extern boolean_t thread_run_queue_remove(
- thread_t thread);
+extern boolean_t thread_run_queue_remove(thread_t thread);
+thread_t thread_run_queue_remove_for_handoff(thread_t thread);
+
+/* Put a thread back in the run queue after being yanked */
+extern void thread_run_queue_reinsert(thread_t thread, integer_t options);
extern void thread_timer_expire(
void *thread,
extern boolean_t thread_eager_preemption(
thread_t thread);
-/* Fair Share routines */
-#if defined(CONFIG_SCHED_FAIRSHARE_CORE)
-void sched_traditional_fairshare_init(void);
-
-int sched_traditional_fairshare_runq_count(void);
-
-uint64_t sched_traditional_fairshare_runq_stats_count_sum(void);
-
-void sched_traditional_fairshare_enqueue(thread_t thread);
-
-thread_t sched_traditional_fairshare_dequeue(void);
-
-boolean_t sched_traditional_fairshare_queue_remove(thread_t thread);
-#endif /* CONFIG_SCHED_FAIRSHARE_CORE */
-
-#if defined(CONFIG_SCHED_GRRR)
-void sched_grrr_fairshare_init(void);
-
-int sched_grrr_fairshare_runq_count(void);
-
-uint64_t sched_grrr_fairshare_runq_stats_count_sum(void);
-
-void sched_grrr_fairshare_enqueue(thread_t thread);
-
-thread_t sched_grrr_fairshare_dequeue(void);
-
-boolean_t sched_grrr_fairshare_queue_remove(thread_t thread);
-#endif
-
extern boolean_t sched_generic_direct_dispatch_to_idle_processors;
/* Set the maximum interrupt level for the thread */
int old_count);
+#if DEBUG
#define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri) \
do { \
} \
} while (0)
+#else /* DEBUG */
+
+#define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri) do { }while(0)
+#define SCHED_STATS_RUNQ_CHANGE(stats, old_count) do { }while(0)
+
+#endif /* DEBUG */
+
+extern uint32_t sched_debug_flags;
+#define SCHED_DEBUG_FLAG_PLATFORM_TRACEPOINTS 0x00000001
+#define SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS 0x00000002
+
+#define SCHED_DEBUG_PLATFORM_KERNEL_DEBUG_CONSTANT(...) do { \
+ if (__improbable(sched_debug_flags & SCHED_DEBUG_FLAG_PLATFORM_TRACEPOINTS)) { \
+ KERNEL_DEBUG_CONSTANT(__VA_ARGS__); \
+ } \
+ } while(0)
+
+#define SCHED_DEBUG_CHOOSE_PROCESSOR_KERNEL_DEBUG_CONSTANT(...) do { \
+ if (__improbable(sched_debug_flags & SCHED_DEBUG_FLAG_CHOOSE_PROCESSOR_TRACEPOINTS)) { \
+ KERNEL_DEBUG_CONSTANT(__VA_ARGS__); \
+ } \
+ } while(0)
+
#define THREAD_URGENCY_NONE 0 /* indicates that there is no currently runnable */
#define THREAD_URGENCY_BACKGROUND 1 /* indicates that the thread is marked as a "background" thread */
#define THREAD_URGENCY_NORMAL 2 /* indicates that the thread is marked as a "normal" thread */
int urgency,
uint64_t rt_period,
uint64_t rt_deadline,
+ uint64_t sched_latency,
thread_t nthread);
/* Tells if there are "active" RT threads in the system (provided by CPU PM) */
extern void active_rt_threads(
boolean_t active);
+/* Returns the perfcontrol attribute for the thread */
+extern perfcontrol_class_t thread_get_perfcontrol_class(
+ thread_t thread);
+
+#define PSET_LOAD_NUMERATOR_SHIFT 16
+#define PSET_LOAD_FRACTIONAL_SHIFT 4
+
+extern int sched_get_pset_load_average(processor_set_t pset);
+extern void sched_update_pset_load_average(processor_set_t pset);
+
+/* Generic routine for Non-AMP schedulers to calculate parallelism */
+extern uint32_t sched_qos_max_parallelism(int qos, uint64_t options);
+
#endif /* MACH_KERNEL_PRIVATE */
__BEGIN_DECLS
#ifdef XNU_KERNEL_PRIVATE
-extern boolean_t assert_wait_possible(void);
-
/* Toggles a global override to turn off CPU Throttling */
#define CPU_THROTTLE_DISABLE 0
#define CPU_THROTTLE_ENABLE 1
****************** Only exported until BSD stops using ********************
*/
+extern void thread_vm_bind_group_add(void);
+
/* Wake up thread directly, passing result */
extern kern_return_t clear_wait(
thread_t thread,
/* Return from exception (BSD-visible interface) */
extern void thread_exception_return(void) __dead2;
+#define SCHED_STRING_MAX_LENGTH (48)
+/* String declaring the name of the current scheduler */
+extern char sched_string[SCHED_STRING_MAX_LENGTH];
+
+extern thread_t port_name_to_thread_for_ulock(mach_port_name_t thread_name);
+
+/* Attempt to context switch to a specific runnable thread */
+extern wait_result_t thread_handoff(thread_t thread);
+
+extern struct waitq *assert_wait_queue(event_t event);
+
+extern kern_return_t thread_wakeup_one_with_pri(event_t event, int priority);
+
+extern thread_t thread_wakeup_identify(event_t event, int priority);
+
#endif /* XNU_KERNEL_PRIVATE */
+#ifdef KERNEL_PRIVATE
+/* Set pending block hint for a particular object before we go into a wait state */
+extern void thread_set_pending_block_hint(
+ thread_t thread,
+ block_hint_t block_hint);
+
+#define QOS_PARALLELISM_COUNT_LOGICAL 0x1
+#define QOS_PARALLELISM_REALTIME 0x2
+extern uint32_t qos_max_parallelism(int qos, uint64_t options);
+
+#endif /* KERNEL_PRIVATE */
+
/* Context switch */
extern wait_result_t thread_block(
thread_continue_t continuation);
boolean_t one_thread,
wait_result_t result);
-extern kern_return_t thread_wakeup_prim_internal(
- event_t event,
- boolean_t one_thread,
- wait_result_t result,
- int priority);
-
-
#define thread_wakeup(x) \
thread_wakeup_prim((x), FALSE, THREAD_AWAKENED)
#define thread_wakeup_with_result(x, z) \
#define thread_wakeup_one(x) \
thread_wakeup_prim((x), TRUE, THREAD_AWAKENED)
-#ifdef MACH_KERNEL_PRIVATE
-#define thread_wakeup_one_with_pri(x, pri) \
- thread_wakeup_prim_internal((x), TRUE, THREAD_AWAKENED, pri)
-#endif
+/* Wakeup the specified thread if it is waiting on this event */
+extern kern_return_t thread_wakeup_thread(event_t event, thread_t thread);
-extern boolean_t preemption_enabled(void);
+extern boolean_t preemption_enabled(void);
#ifdef MACH_KERNEL_PRIVATE
#error Enable at least one scheduler algorithm in osfmk/conf/MASTER.XXX
#endif
+#if DEBUG
#define SCHED(f) (sched_current_dispatch->f)
+#else /* DEBUG */
+
+/*
+ * For DEV & REL kernels, use a static dispatch table instead of
+ * using the indirect function table.
+ */
+extern const struct sched_dispatch_table sched_multiq_dispatch;
+#define SCHED(f) (sched_multiq_dispatch.f)
+
+#endif /* DEBUG */
struct sched_dispatch_table {
+ const char *sched_name;
void (*init)(void); /* Init global state */
void (*timebase_init)(void); /* Timebase-dependent initialization */
void (*processor_init)(processor_t processor); /* Per-processor scheduler init */
void (*pset_init)(processor_set_t pset); /* Per-processor set scheduler init */
-
+
void (*maintenance_continuation)(void); /* Function called regularly */
-
+
/*
* Choose a thread of greater or equal priority from the per-processor
* runqueue for timeshare/fixed threads
processor_t processor,
int priority,
ast_t reason);
-
+
+ /* True if scheduler supports stealing threads */
+ boolean_t steal_thread_enabled;
+
/*
* Steal a thread from another processor in the pset so that it can run
* immediately
*/
thread_t (*steal_thread)(
processor_set_t pset);
-
+
/*
- * Recalculate sched_pri based on base priority, past running time,
- * and scheduling class.
+ * Compute priority for a timeshare thread based on base priority.
*/
- void (*compute_priority)(
- thread_t thread,
- boolean_t override_depress);
-
+ int (*compute_timeshare_priority)(thread_t thread);
+
/*
* Pick the best processor for a thread (any kind of thread) to run on.
*/
processor_t processor,
thread_t thread,
integer_t options);
-
+
/* Migrate threads away in preparation for processor shutdown */
void (*processor_queue_shutdown)(
processor_t processor);
-
+
/* Remove the specific thread from the per-processor runqueue */
boolean_t (*processor_queue_remove)(
processor_t processor,
thread_t thread);
-
+
/*
* Does the per-processor runqueue have any timeshare or fixed priority
* threads on it? Called without pset lock held, so should
* not assume immutability while executing.
*/
boolean_t (*processor_queue_empty)(processor_t processor);
-
+
/*
* Would this priority trigger an urgent preemption if it's sitting
* on the per-processor runqueue?
*/
boolean_t (*priority_is_urgent)(int priority);
-
+
/*
* Does the per-processor runqueue contain runnable threads that
* should cause the currently-running thread to be preempted?
*/
ast_t (*processor_csw_check)(processor_t processor);
-
+
/*
* Does the per-processor runqueue contain a runnable thread
* of > or >= priority, as a preflight for choose_thread() or other
boolean_t (*processor_queue_has_priority)(processor_t processor,
int priority,
boolean_t gte);
-
+
/* Quantum size for the specified non-realtime thread. */
uint32_t (*initial_quantum_size)(thread_t thread);
/* Scheduler mode for a new thread */
sched_mode_t (*initial_thread_sched_mode)(task_t parent_task);
-
+
/*
* Is it safe to call update_priority, which may change a thread's
* runqueue or other state. This can be used to throttle changes
* Side effects may including migration to another processor's runqueue.
*/
void (*update_priority)(thread_t thread);
-
+
/* Lower overhead update to scheduled priority and state. */
void (*lightweight_update_priority)(thread_t thread);
-
+
/* Callback for non-realtime threads when the quantum timer fires */
void (*quantum_expire)(thread_t thread);
-
- /*
- * Even though we could continue executing on this processor, does the
- * topology (SMT, for instance) indicate that a better processor could be
- * chosen
- */
- boolean_t (*should_current_thread_rechoose_processor)(processor_t processor);
-
+
/*
* Runnable threads on per-processor runqueue. Should only
* be used for relative comparisons of load between processors.
*/
int (*processor_runq_count)(processor_t processor);
-
- /* Aggregate runcount statistics for per-processor runqueue */
- uint64_t (*processor_runq_stats_count_sum)(processor_t processor);
-
- /* Initialize structures to track demoted fairshare threads */
- void (*fairshare_init)(void);
-
- /* Number of runnable fairshare threads */
- int (*fairshare_runq_count)(void);
-
- /* Aggregate runcount statistics for fairshare runqueue */
- uint64_t (*fairshare_runq_stats_count_sum)(void);
-
- void (*fairshare_enqueue)(thread_t thread);
-
- thread_t (*fairshare_dequeue)(void);
- boolean_t (*fairshare_queue_remove)(thread_t thread);
+ /* Aggregate runcount statistics for per-processor runqueue */
+ uint64_t (*processor_runq_stats_count_sum)(processor_t processor);
boolean_t (*processor_bound_count)(processor_t processor);
- void (*thread_update_scan)(void);
+ void (*thread_update_scan)(sched_update_scan_context_t scan_context);
/*
* Use processor->next_thread to pin a thread to an idle
* be stolen by other processors.
*/
boolean_t direct_dispatch_to_idle_processors;
+
+ /* Supports more than one pset */
+ boolean_t multiple_psets_enabled;
+ /* Supports scheduler groups */
+ boolean_t sched_groups_enabled;
+
+ /* Supports avoid-processor */
+ boolean_t avoid_processor_enabled;
+
+ /* Returns true if this processor should avoid running this thread. */
+ bool (*thread_avoid_processor)(processor_t processor, thread_t thread);
+
+ /*
+ * Invoked when a processor is about to choose the idle thread
+ * Used to send IPIs to a processor which would be preferred to be idle instead.
+ * Called with pset lock held, returns pset lock unlocked.
+ */
+ void (*processor_balance)(processor_t processor, processor_set_t pset);
+ rt_queue_t (*rt_runq)(processor_set_t pset);
+ void (*rt_init)(processor_set_t pset);
+ void (*rt_queue_shutdown)(processor_t processor);
+ void (*rt_runq_scan)(sched_update_scan_context_t scan_context);
+ int64_t (*rt_runq_count_sum)(void);
+
+ uint32_t (*qos_max_parallelism)(int qos, uint64_t options);
+ void (*check_spill)(processor_set_t pset, thread_t thread);
+ sched_ipi_type_t (*ipi_policy)(processor_t dst, thread_t thread, boolean_t dst_idle, sched_ipi_event_t event);
+ bool (*thread_should_yield)(processor_t processor, thread_t thread);
};
#if defined(CONFIG_SCHED_TRADITIONAL)
-#define kSchedTraditionalString "traditional"
-#define kSchedTraditionalWithPsetRunqueueString "traditional_with_pset_runqueue"
extern const struct sched_dispatch_table sched_traditional_dispatch;
extern const struct sched_dispatch_table sched_traditional_with_pset_runqueue_dispatch;
#endif
#if defined(CONFIG_SCHED_MULTIQ)
extern const struct sched_dispatch_table sched_multiq_dispatch;
-#define kSchedMultiQString "multiq"
extern const struct sched_dispatch_table sched_dualq_dispatch;
-#define kSchedDualQString "dualq"
#endif
#if defined(CONFIG_SCHED_PROTO)
-#define kSchedProtoString "proto"
extern const struct sched_dispatch_table sched_proto_dispatch;
#endif
#if defined(CONFIG_SCHED_GRRR)
-#define kSchedGRRRString "grrr"
extern const struct sched_dispatch_table sched_grrr_dispatch;
#endif
* It is an error to invoke any scheduler-related code
* before this is set up
*/
-enum sched_enum {
- sched_enum_unknown = 0,
-#if defined(CONFIG_SCHED_TRADITIONAL)
- sched_enum_traditional = 1,
- sched_enum_traditional_with_pset_runqueue = 2,
-#endif
-#if defined(CONFIG_SCHED_PROTO)
- sched_enum_proto = 3,
-#endif
-#if defined(CONFIG_SCHED_GRRR)
- sched_enum_grrr = 4,
-#endif
-#if defined(CONFIG_SCHED_MULTIQ)
- sched_enum_multiq = 5,
- sched_enum_dualq = 6,
-#endif
- sched_enum_max = 7,
-};
-
extern const struct sched_dispatch_table *sched_current_dispatch;
#endif /* MACH_KERNEL_PRIVATE */