X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/b7266188b87f3620ec3f9f717e57194a7dd989fe..a39ff7e25e19b3a8c3020042a3872ca9ec9659f1:/osfmk/kern/sched_prim.h diff --git a/osfmk/kern/sched_prim.h b/osfmk/kern/sched_prim.h index 9f1c95347..c6361a9ca 100644 --- a/osfmk/kern/sched_prim.h +++ b/osfmk/kern/sched_prim.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2008 Apple Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -73,19 +73,37 @@ #include #include #include +#include #ifdef MACH_KERNEL_PRIVATE +#include + /* Initialization */ -extern void sched_init(void) __attribute__((section("__TEXT, initcode"))); +extern void sched_init(void); extern void sched_startup(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, + boolean_t until_not_runnable); /* Release a previous stop request */ extern void thread_unstop( @@ -93,7 +111,8 @@ extern void thread_unstop( /* Wait for a thread to stop running */ extern void thread_wait( - thread_t thread); + thread_t thread, + boolean_t until_not_runnable); /* Unblock thread on wake up */ extern boolean_t thread_unblock( @@ -133,26 +152,46 @@ extern void set_sched_pri( int priority); /* Set base priority of the specified thread */ -extern void set_priority( +extern void sched_set_thread_base_priority( thread_t thread, int priority); +/* Set the thread's true scheduling mode */ +extern void sched_set_thread_mode(thread_t thread, + sched_mode_t mode); +/* Demote the true scheduler mode */ +extern void sched_thread_mode_demote(thread_t thread, + uint32_t reason); +/* Un-demote the true scheduler 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_tick_thread(void); +extern void sched_init_thread(void (*)(void)); /* Perform sched_tick housekeeping activities */ -extern void update_priority( +extern boolean_t can_update_priority( thread_t thread); +extern void update_priority( + thread_t thread); + +extern void lightweight_update_priority( + thread_t thread); + +extern void sched_default_quantum_expire(thread_t thread); + /* Idle processor thread */ extern void idle_thread(void); @@ -174,9 +213,13 @@ extern void thread_setrun( 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); @@ -185,13 +228,120 @@ extern processor_set_t task_choose_pset( extern processor_t thread_bind( processor_t processor); +/* Choose the best processor to run a thread */ +extern processor_t choose_processor( + processor_set_t pset, + processor_t processor, + thread_t thread); + +extern void sched_SMT_balance( + processor_t processor, + processor_set_t pset); + +extern void thread_quantum_init( + thread_t thread); + extern void run_queue_init( run_queue_t runq); +extern thread_t run_queue_dequeue( + run_queue_t runq, + integer_t options); + +extern boolean_t run_queue_enqueue( + run_queue_t runq, + thread_t thread, + integer_t options); + +extern void run_queue_remove( + 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 void thread_update_process_threads(void); +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); + +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); +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, void *p1); +extern boolean_t thread_eager_preemption( + thread_t thread); + +extern boolean_t sched_generic_direct_dispatch_to_idle_processors; + /* Set the maximum interrupt level for the thread */ __private_extern__ wait_interrupt_t thread_interrupt_level( wait_interrupt_t interruptible); @@ -205,31 +355,152 @@ __private_extern__ kern_return_t clear_wait_internal( thread_t thread, wait_result_t result); +extern void sched_stats_handle_csw( + processor_t processor, + int reasons, + int selfpri, + int otherpri); + +extern void sched_stats_handle_runq_change( + struct runq_stats *stats, + int old_count); + + +#if DEBUG + +#define SCHED_STATS_CSW(processor, reasons, selfpri, otherpri) \ +do { \ + if (__builtin_expect(sched_stats_active, 0)) { \ + sched_stats_handle_csw((processor), \ + (reasons), (selfpri), (otherpri)); \ + } \ +} while (0) + + +#define SCHED_STATS_RUNQ_CHANGE(stats, old_count) \ +do { \ + if (__builtin_expect(sched_stats_active, 0)) { \ + sched_stats_handle_runq_change((stats), \ + (old_count)); \ + } \ +} 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 */ +#define THREAD_URGENCY_REAL_TIME 3 /* indicates that the thread is marked as a "real-time" or urgent thread */ +#define THREAD_URGENCY_MAX 4 /* Marker */ +/* Returns the "urgency" of a thread (provided by scheduler) */ +extern int thread_get_urgency( + thread_t thread, + uint64_t *rt_period, + uint64_t *rt_deadline); + +/* Tells the "urgency" of the just scheduled thread (provided by CPU PM) */ +extern void thread_tell_urgency( + 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 +extern void sys_override_cpu_throttle(int flag); /* ****************** 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, wait_result_t result); /* Start thread running */ -extern void thread_bootstrap_return(void); +extern void thread_bootstrap_return(void) __attribute__((noreturn)); /* 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); @@ -250,16 +521,33 @@ extern wait_result_t assert_wait_timeout( uint32_t interval, uint32_t scale_factor); +/* Assert that the thread intends to wait with an urgency, timeout and leeway */ +extern wait_result_t assert_wait_timeout_with_leeway( + event_t event, + wait_interrupt_t interruptible, + wait_timeout_urgency_t urgency, + uint32_t interval, + uint32_t leeway, + uint32_t scale_factor); + extern wait_result_t assert_wait_deadline( event_t event, wait_interrupt_t interruptible, uint64_t deadline); +/* Assert that the thread intends to wait with an urgency, deadline, and leeway */ +extern wait_result_t assert_wait_deadline_with_leeway( + event_t event, + wait_interrupt_t interruptible, + wait_timeout_urgency_t urgency, + uint64_t deadline, + uint64_t leeway); + /* Wake up thread (or threads) waiting on a particular event */ extern kern_return_t thread_wakeup_prim( event_t event, boolean_t one_thread, - wait_result_t result); + wait_result_t result); #define thread_wakeup(x) \ thread_wakeup_prim((x), FALSE, THREAD_AWAKENED) @@ -268,40 +556,223 @@ extern kern_return_t thread_wakeup_prim( #define thread_wakeup_one(x) \ thread_wakeup_prim((x), TRUE, THREAD_AWAKENED) -extern boolean_t preemption_enabled(void); +/* Wakeup the specified thread if it is waiting on this event */ +extern kern_return_t thread_wakeup_thread(event_t event, thread_t thread); -#ifdef KERNEL_PRIVATE +extern boolean_t preemption_enabled(void); -#ifndef __LP64__ +#ifdef MACH_KERNEL_PRIVATE /* - * Obsolete interfaces. + * Scheduler algorithm indirection. If only one algorithm is + * enabled at compile-time, a direction function call is used. + * If more than one is enabled, calls are dispatched through + * a function pointer table. */ -extern void thread_set_timer( - uint32_t interval, - uint32_t scale_factor); +#if !defined(CONFIG_SCHED_TRADITIONAL) && !defined(CONFIG_SCHED_PROTO) && !defined(CONFIG_SCHED_GRRR) && !defined(CONFIG_SCHED_MULTIQ) +#error Enable at least one scheduler algorithm in osfmk/conf/MASTER.XXX +#endif -extern void thread_set_timer_deadline( - uint64_t deadline); +#if DEBUG +#define SCHED(f) (sched_current_dispatch->f) +#else /* DEBUG */ -extern void thread_cancel_timer(void); - -#ifndef MACH_KERNEL_PRIVATE - -#ifndef ABSOLUTETIME_SCALAR_TYPE - -#define thread_set_timer_deadline(a) \ - thread_set_timer_deadline(__OSAbsoluteTime(a)) +/* + * 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 + */ + thread_t (*choose_thread)( + 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); + + /* + * Compute priority for a timeshare thread based on base priority. + */ + int (*compute_timeshare_priority)(thread_t thread); + + /* + * Pick the best processor for a thread (any kind of thread) to run on. + */ + processor_t (*choose_processor)( + processor_set_t pset, + processor_t processor, + thread_t thread); + /* + * Enqueue a timeshare or fixed priority thread onto the per-processor + * runqueue + */ + boolean_t (*processor_enqueue)( + 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 + * thread selection + */ + 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 + * to dynamic priority. + */ + boolean_t (*can_update_priority)(thread_t thread); + + /* + * Update both scheduled priority and other persistent state. + * 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); + + /* + * 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); + + boolean_t (*processor_bound_count)(processor_t processor); + + void (*thread_update_scan)(sched_update_scan_context_t scan_context); + + /* + * Use processor->next_thread to pin a thread to an idle + * processor. If FALSE, threads are enqueued and can + * 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) +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; +extern const struct sched_dispatch_table sched_dualq_dispatch; +#endif + +#if defined(CONFIG_SCHED_PROTO) +extern const struct sched_dispatch_table sched_proto_dispatch; +#endif + +#if defined(CONFIG_SCHED_GRRR) +extern const struct sched_dispatch_table sched_grrr_dispatch; +#endif -#endif /* ABSOLUTETIME_SCALAR_TYPE */ +/* + * It is an error to invoke any scheduler-related code + * before this is set up + */ +extern const struct sched_dispatch_table *sched_current_dispatch; #endif /* MACH_KERNEL_PRIVATE */ -#endif /* __LP64__ */ - -#endif /* KERNEL_PRIVATE */ - __END_DECLS #endif /* _KERN_SCHED_PRIM_H_ */