X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/d7e50217d7adf6e52786a38bcaa4cd698cb9a79e..060df5ea7c632b1ac8cc8aac1fb59758165c2084:/osfmk/kern/thread.c diff --git a/osfmk/kern/thread.c b/osfmk/kern/thread.c index 9c208da92..adff14820 100644 --- a/osfmk/kern/thread.c +++ b/osfmk/kern/thread.c @@ -1,16 +1,19 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2009 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ - * - * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved. + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in - * compliance with the License. Please obtain a copy of the License at - * http://www.opensource.apple.com/apsl/ and read it before using this - * file. + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. + * + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER @@ -20,7 +23,7 @@ * Please see the License for the specific language governing rights and * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* * @OSF_FREE_COPYRIGHT@ @@ -57,7 +60,7 @@ * Author: Avadis Tevanian, Jr., Michael Wayne Young, David Golub * Date: 1986 * - * Thread/thread_shuttle management primitives implementation. + * Thread management primitives implementation. */ /* * Copyright (c) 1993 The University of Utah and @@ -78,12 +81,7 @@ * */ -#include -#include -#include -#include -#include - +#include #include #include #include @@ -91,10 +89,13 @@ #include #include #include -#include + +#include + +#include +#include #include #include -#include #include #include #include @@ -104,781 +105,638 @@ #include #include #include -#include /*** ??? fix so this can be removed ***/ +#include +#include #include #include -#include -#include #include #include -#include +#include + #include #include -#include /* for MACHINE_STACK */ -#include -#include + +#include +#include + #include +#include + /* * Exported interfaces */ - +#include #include #include +#include -/* - * Per-Cpu stashed global state - */ -vm_offset_t active_stacks[NCPUS]; /* per-cpu active stacks */ -vm_offset_t kernel_stack[NCPUS]; /* top of active stacks */ -thread_act_t active_kloaded[NCPUS]; /* + act if kernel loaded */ -boolean_t first_thread; +static struct zone *thread_zone; +static lck_grp_attr_t thread_lck_grp_attr; +lck_attr_t thread_lck_attr; +lck_grp_t thread_lck_grp; -struct zone *thread_shuttle_zone; +decl_simple_lock_data(static,thread_stack_lock) +static queue_head_t thread_stack_queue; -queue_head_t reaper_queue; -decl_simple_lock_data(,reaper_lock) +decl_simple_lock_data(static,thread_terminate_lock) +static queue_head_t thread_terminate_queue; -extern int tick; +static struct thread thread_template, init_thread; -extern void pcb_module_init(void); +static void sched_call_null( + int type, + thread_t thread); -struct thread_shuttle pageout_thread; +#ifdef MACH_BSD +extern void proc_exit(void *); +extern uint64_t get_dispatchqueue_offset_from_proc(void *); +#endif /* MACH_BSD */ -/* private */ -static struct thread_shuttle thr_sh_template; +extern int debug_task; +int thread_max = CONFIG_THREAD_MAX; /* Max number of threads */ +int task_threadmax = CONFIG_THREAD_MAX; -#if MACH_DEBUG +static uint64_t thread_unique_id = 0; -#ifdef MACHINE_STACK -extern void stack_statistics( - unsigned int *totalp, - vm_size_t *maxusagep); -#endif /* MACHINE_STACK */ -#endif /* MACH_DEBUG */ +void +thread_bootstrap(void) +{ + /* + * Fill in a template thread for fast initialization. + */ -/* Forwards */ -void thread_collect_scan(void); + thread_template.runq = PROCESSOR_NULL; -kern_return_t thread_create_shuttle( - thread_act_t thr_act, - integer_t priority, - void (*start)(void), - thread_t *new_thread); + thread_template.ref_count = 2; -extern void Load_context( - thread_t thread); + thread_template.reason = AST_NONE; + thread_template.at_safe_point = FALSE; + thread_template.wait_event = NO_EVENT64; + thread_template.wait_queue = WAIT_QUEUE_NULL; + thread_template.wait_result = THREAD_WAITING; + thread_template.options = THREAD_ABORTSAFE; + thread_template.state = TH_WAIT | TH_UNINT; + thread_template.wake_active = FALSE; + thread_template.continuation = THREAD_CONTINUE_NULL; + thread_template.parameter = NULL; + thread_template.importance = 0; + thread_template.sched_mode = 0; + thread_template.safe_mode = 0; + thread_template.safe_release = 0; -/* - * Machine-dependent code must define: - * thread_machine_init - * thread_machine_terminate - * thread_machine_collect - * - * The thread->pcb field is reserved for machine-dependent code. - */ + thread_template.priority = 0; + thread_template.sched_pri = 0; + thread_template.max_priority = 0; + thread_template.task_priority = 0; + thread_template.promotions = 0; + thread_template.pending_promoter_index = 0; + thread_template.pending_promoter[0] = + thread_template.pending_promoter[1] = NULL; -#ifdef MACHINE_STACK -/* - * Machine-dependent code must define: - * stack_alloc_try - * stack_alloc - * stack_free - * stack_free_stack - * stack_collect - * and if MACH_DEBUG: - * stack_statistics - */ -#else /* MACHINE_STACK */ -/* - * We allocate stacks from generic kernel VM. - * Machine-dependent code must define: - * machine_kernel_stack_init - * - * The stack_free_list can only be accessed at splsched, - * because stack_alloc_try/thread_invoke operate at splsched. - */ + thread_template.realtime.deadline = UINT64_MAX; -decl_simple_lock_data(,stack_lock_data) /* splsched only */ -#define stack_lock() simple_lock(&stack_lock_data) -#define stack_unlock() simple_unlock(&stack_lock_data) + thread_template.current_quantum = 0; -mutex_t stack_map_lock; /* Lock when allocating stacks maps */ -vm_map_t stack_map; /* Map for allocating stacks */ -vm_offset_t stack_free_list; /* splsched only */ -unsigned int stack_free_max = 0; -unsigned int stack_free_count = 0; /* splsched only */ -unsigned int stack_free_limit = 1; /* Arbitrary */ + thread_template.computation_metered = 0; + thread_template.computation_epoch = 0; -unsigned int stack_alloc_hits = 0; /* debugging */ -unsigned int stack_alloc_misses = 0; /* debugging */ + thread_template.sched_stamp = 0; + thread_template.sched_usage = 0; + thread_template.pri_shift = INT8_MAX; + thread_template.cpu_usage = thread_template.cpu_delta = 0; + thread_template.c_switch = thread_template.p_switch = thread_template.ps_switch = 0; -unsigned int stack_alloc_total = 0; -unsigned int stack_alloc_hiwater = 0; -unsigned int stack_alloc_bndry = 0; + thread_template.bound_processor = PROCESSOR_NULL; + thread_template.last_processor = PROCESSOR_NULL; + thread_template.sched_call = sched_call_null; -/* - * The next field is at the base of the stack, - * so the low end is left unsullied. - */ + timer_init(&thread_template.user_timer); + timer_init(&thread_template.system_timer); + thread_template.user_timer_save = 0; + thread_template.system_timer_save = 0; + thread_template.vtimer_user_save = 0; + thread_template.vtimer_prof_save = 0; + thread_template.vtimer_rlim_save = 0; -#define stack_next(stack) (*((vm_offset_t *)((stack) + KERNEL_STACK_SIZE) - 1)) + thread_template.wait_timer_is_set = FALSE; + thread_template.wait_timer_active = 0; -/* - * stack_alloc: - * - * Allocate a kernel stack for an activation. - * May block. - */ -vm_offset_t -stack_alloc( - thread_t thread, - void (*start_pos)(thread_t)) -{ - vm_offset_t stack = thread->kernel_stack; - spl_t s; + thread_template.depress_timer_active = 0; - if (stack) - return (stack); + thread_template.special_handler.handler = special_handler; + thread_template.special_handler.next = NULL; -/* - * We first try the free list. It is probably empty, or - * stack_alloc_try would have succeeded, but possibly a stack was - * freed before the swapin thread got to us. - * - * We allocate stacks from their own map which is submaps of the - * kernel map. Because we want to have a guard page (at least) in - * front of each stack to catch evil code that overruns its stack, we - * allocate the stack on aligned boundaries. The boundary is - * calculated as the next power of 2 above the stack size. For - * example, a stack of 4 pages would have a boundry of 8, likewise 5 - * would also be 8. - * - * We limit the number of stacks to be one allocation chunk - * (THREAD_CHUNK) more than the maximum number of threads - * (THREAD_MAX). The extra is to allow for priviliged threads that - * can sometimes have 2 stacks. - * - */ + thread_template.funnel_lock = THR_FUNNEL_NULL; + thread_template.funnel_state = 0; + thread_template.recover = (vm_offset_t)NULL; + + thread_template.map = VM_MAP_NULL; - s = splsched(); - stack_lock(); - stack = stack_free_list; - if (stack != 0) { - stack_free_list = stack_next(stack); - stack_free_count--; - } - stack_unlock(); - splx(s); +#if CONFIG_DTRACE + thread_template.t_dtrace_predcache = 0; + thread_template.t_dtrace_vtime = 0; + thread_template.t_dtrace_tracing = 0; +#endif /* CONFIG_DTRACE */ - if (stack != 0) { /* Did we find a free one? */ - stack_attach(thread, stack, start_pos); /* Initialize it */ - return (stack); /* Send it on home */ - } - - if (kernel_memory_allocate( - stack_map, &stack, - KERNEL_STACK_SIZE, stack_alloc_bndry - 1, - KMA_KOBJECT) != KERN_SUCCESS) - panic("stack_alloc: no space left for stack maps"); - - stack_alloc_total++; - if (stack_alloc_total > stack_alloc_hiwater) - stack_alloc_hiwater = stack_alloc_total; - - stack_attach(thread, stack, start_pos); - return (stack); -} + thread_template.t_chud = 0; + thread_template.t_page_creation_count = 0; + thread_template.t_page_creation_time = 0; -/* - * stack_free: - * - * Free a kernel stack. - * Called at splsched. - */ + thread_template.affinity_set = NULL; + + init_thread = thread_template; + machine_set_current_thread(&init_thread); +} void -stack_free( - thread_t thread) +thread_init(void) { - vm_offset_t stack = stack_detach(thread); - - assert(stack); - if (stack != thread->stack_privilege) { - stack_lock(); - stack_next(stack) = stack_free_list; - stack_free_list = stack; - if (++stack_free_count > stack_free_max) - stack_free_max = stack_free_count; - stack_unlock(); - } + thread_zone = zinit( + sizeof(struct thread), + thread_max * sizeof(struct thread), + THREAD_CHUNK * sizeof(struct thread), + "threads"); + zone_change(thread_zone, Z_NOENCRYPT, TRUE); + + lck_grp_attr_setdefault(&thread_lck_grp_attr); + lck_grp_init(&thread_lck_grp, "thread", &thread_lck_grp_attr); + lck_attr_setdefault(&thread_lck_attr); + + stack_init(); + + /* + * Initialize any machine-dependent + * per-thread structures necessary. + */ + machine_thread_init(); } static void -stack_free_stack( - vm_offset_t stack) +thread_terminate_continue(void) { - spl_t s; - - s = splsched(); - stack_lock(); - stack_next(stack) = stack_free_list; - stack_free_list = stack; - if (++stack_free_count > stack_free_max) - stack_free_max = stack_free_count; - stack_unlock(); - splx(s); + panic("thread_terminate_continue"); + /*NOTREACHED*/ } /* - * stack_collect: - * - * Free excess kernel stacks. - * May block. + * thread_terminate_self: */ - void -stack_collect(void) +thread_terminate_self(void) { - vm_offset_t stack; - int i; - spl_t s; + thread_t thread = current_thread(); + task_t task; + spl_t s; + int threadcnt; + + DTRACE_PROC(lwp__exit); + + thread_mtx_lock(thread); + + ulock_release_all(thread); + + ipc_thread_disable(thread); + + thread_mtx_unlock(thread); s = splsched(); - stack_lock(); - while (stack_free_count > stack_free_limit) { - stack = stack_free_list; - stack_free_list = stack_next(stack); - stack_free_count--; - stack_unlock(); + thread_lock(thread); + + /* + * Cancel priority depression, wait for concurrent expirations + * on other processors. + */ + if (thread->sched_mode & TH_MODE_ISDEPRESSED) { + thread->sched_mode &= ~TH_MODE_ISDEPRESSED; + + if (timer_call_cancel(&thread->depress_timer)) + thread->depress_timer_active--; + } + + while (thread->depress_timer_active > 0) { + thread_unlock(thread); splx(s); - if (vm_map_remove( - stack_map, stack, stack + KERNEL_STACK_SIZE, - VM_MAP_REMOVE_KUNWIRE) != KERN_SUCCESS) - panic("stack_collect: vm_map_remove failed"); + delay(1); s = splsched(); - stack_lock(); - stack_alloc_total--; + thread_lock(thread); } - stack_unlock(); - splx(s); -} + thread_sched_call(thread, NULL); -#if MACH_DEBUG -/* - * stack_statistics: - * - * Return statistics on cached kernel stacks. - * *maxusagep must be initialized by the caller. - */ + thread_unlock(thread); + splx(s); -void -stack_statistics( - unsigned int *totalp, - vm_size_t *maxusagep) -{ - spl_t s; + thread_policy_reset(thread); - s = splsched(); - stack_lock(); + task = thread->task; + uthread_cleanup(task, thread->uthread, task->bsd_info); + threadcnt = hw_atomic_sub(&task->active_thread_count, 1); - *totalp = stack_free_count; - *maxusagep = 0; + /* + * If we are the last thread to terminate and the task is + * associated with a BSD process, perform BSD process exit. + */ + if (threadcnt == 0 && task->bsd_info != NULL) + proc_exit(task->bsd_info); - stack_unlock(); - splx(s); -} -#endif /* MACH_DEBUG */ + uthread_cred_free(thread->uthread); -#endif /* MACHINE_STACK */ + s = splsched(); + thread_lock(thread); + /* + * Cancel wait timer, and wait for + * concurrent expirations. + */ + if (thread->wait_timer_is_set) { + thread->wait_timer_is_set = FALSE; -stack_fake_zone_info(int *count, vm_size_t *cur_size, vm_size_t *max_size, vm_size_t *elem_size, - vm_size_t *alloc_size, int *collectable, int *exhaustable) -{ - *count = stack_alloc_total - stack_free_count; - *cur_size = KERNEL_STACK_SIZE * stack_alloc_total; - *max_size = KERNEL_STACK_SIZE * stack_alloc_hiwater; - *elem_size = KERNEL_STACK_SIZE; - *alloc_size = KERNEL_STACK_SIZE; - *collectable = 1; - *exhaustable = 0; -} + if (timer_call_cancel(&thread->wait_timer)) + thread->wait_timer_active--; + } + while (thread->wait_timer_active > 0) { + thread_unlock(thread); + splx(s); -/* - * stack_privilege: - * - * stack_alloc_try on this thread must always succeed. - */ + delay(1); + + s = splsched(); + thread_lock(thread); + } -void -stack_privilege( - register thread_t thread) -{ /* - * This implementation only works for the current thread. + * If there is a reserved stack, release it. */ + if (thread->reserved_stack != 0) { + if (thread->reserved_stack != thread->kernel_stack) + stack_free_stack(thread->reserved_stack); + thread->reserved_stack = 0; + } - if (thread != current_thread()) - panic("stack_privilege"); + /* + * Mark thread as terminating, and block. + */ + thread->state |= TH_TERMINATE; + thread_mark_wait_locked(thread, THREAD_UNINT); + assert(thread->promotions == 0); + thread_unlock(thread); + /* splsched */ - if (thread->stack_privilege == 0) - thread->stack_privilege = current_stack(); + thread_block((thread_continue_t)thread_terminate_continue); + /*NOTREACHED*/ } -/* - * stack_alloc_try: - * - * Non-blocking attempt to allocate a kernel stack. - * Called at splsched with the thread locked. - */ - -boolean_t stack_alloc_try( - thread_t thread, - void (*start_pos)(thread_t)) +void +thread_deallocate( + thread_t thread) { - register vm_offset_t stack = thread->stack_privilege; + task_t task; - if (stack == 0) { - stack_lock(); + if (thread == THREAD_NULL) + return; - stack = stack_free_list; - if (stack != (vm_offset_t)0) { - stack_free_list = stack_next(stack); - stack_free_count--; - } + if (thread_deallocate_internal(thread) > 0) + return; - stack_unlock(); - } + ipc_thread_terminate(thread); - if (stack != 0) { - stack_attach(thread, stack, start_pos); - stack_alloc_hits++; + task = thread->task; - return (TRUE); - } - else { - stack_alloc_misses++; +#ifdef MACH_BSD + { + void *ut = thread->uthread; - return (FALSE); + thread->uthread = NULL; + uthread_zone_free(ut); } -} +#endif /* MACH_BSD */ -uint64_t max_unsafe_computation; -extern int max_unsafe_quanta; + task_deallocate(task); -uint32_t sched_safe_duration; + if (thread->kernel_stack != 0) + stack_free(thread); -uint64_t max_poll_computation; -extern int max_poll_quanta; + lck_mtx_destroy(&thread->mutex, &thread_lck_grp); + machine_thread_destroy(thread); -uint32_t std_quantum; -uint32_t min_std_quantum; - -uint32_t max_rt_quantum; -uint32_t min_rt_quantum; + zfree(thread_zone, thread); +} -void -thread_init(void) +/* + * thread_terminate_daemon: + * + * Perform final clean up for terminating threads. + */ +static void +thread_terminate_daemon(void) { - kern_return_t ret; - unsigned int stack; - - thread_shuttle_zone = zinit( - sizeof(struct thread_shuttle), - THREAD_MAX * sizeof(struct thread_shuttle), - THREAD_CHUNK * sizeof(struct thread_shuttle), - "threads"); + thread_t thread; + task_t task; - /* - * Fill in a template thread_shuttle for fast initialization. - * [Fields that must be (or are typically) reset at - * time of creation are so noted.] - */ + (void)splsched(); + simple_lock(&thread_terminate_lock); - /* thr_sh_template.links (none) */ - thr_sh_template.runq = RUN_QUEUE_NULL; - - - /* thr_sh_template.task (later) */ - /* thr_sh_template.thread_list (later) */ - /* thr_sh_template.pset_threads (later) */ - - /* reference for activation */ - thr_sh_template.ref_count = 1; - - thr_sh_template.reason = AST_NONE; - thr_sh_template.at_safe_point = FALSE; - thr_sh_template.wait_event = NO_EVENT64; - thr_sh_template.wait_queue = WAIT_QUEUE_NULL; - thr_sh_template.wait_result = THREAD_WAITING; - thr_sh_template.interrupt_level = THREAD_ABORTSAFE; - thr_sh_template.state = TH_STACK_HANDOFF | TH_WAIT | TH_UNINT; - thr_sh_template.wake_active = FALSE; - thr_sh_template.active_callout = FALSE; - thr_sh_template.continuation = (void (*)(void))0; - thr_sh_template.top_act = THR_ACT_NULL; - - thr_sh_template.importance = 0; - thr_sh_template.sched_mode = 0; - thr_sh_template.safe_mode = 0; - - thr_sh_template.priority = 0; - thr_sh_template.sched_pri = 0; - thr_sh_template.max_priority = 0; - thr_sh_template.task_priority = 0; - thr_sh_template.promotions = 0; - thr_sh_template.pending_promoter_index = 0; - thr_sh_template.pending_promoter[0] = - thr_sh_template.pending_promoter[1] = NULL; - - thr_sh_template.current_quantum = 0; - - thr_sh_template.computation_metered = 0; - thr_sh_template.computation_epoch = 0; - - thr_sh_template.cpu_usage = 0; - thr_sh_template.cpu_delta = 0; - thr_sh_template.sched_usage = 0; - thr_sh_template.sched_delta = 0; - thr_sh_template.sched_stamp = 0; - thr_sh_template.sleep_stamp = 0; - thr_sh_template.safe_release = 0; - - thr_sh_template.bound_processor = PROCESSOR_NULL; - thr_sh_template.last_processor = PROCESSOR_NULL; - thr_sh_template.last_switch = 0; - - thr_sh_template.vm_privilege = FALSE; - - timer_init(&(thr_sh_template.user_timer)); - timer_init(&(thr_sh_template.system_timer)); - thr_sh_template.user_timer_save.low = 0; - thr_sh_template.user_timer_save.high = 0; - thr_sh_template.system_timer_save.low = 0; - thr_sh_template.system_timer_save.high = 0; - - thr_sh_template.active = FALSE; /* reset */ - - thr_sh_template.processor_set = PROCESSOR_SET_NULL; -#if MACH_HOST - thr_sh_template.may_assign = TRUE; - thr_sh_template.assign_active = FALSE; -#endif /* MACH_HOST */ - thr_sh_template.funnel_state = 0; + while ((thread = (thread_t)dequeue_head(&thread_terminate_queue)) != THREAD_NULL) { + simple_unlock(&thread_terminate_lock); + (void)spllo(); - /* - * Initialize other data structures used in - * this module. - */ + task = thread->task; - queue_init(&reaper_queue); - simple_lock_init(&reaper_lock, ETAP_THREAD_REAPER); - thr_sh_template.funnel_lock = THR_FUNNEL_NULL; + task_lock(task); + task->total_user_time += timer_grab(&thread->user_timer); + task->total_system_time += timer_grab(&thread->system_timer); -#ifndef MACHINE_STACK - simple_lock_init(&stack_lock_data, ETAP_THREAD_STACK); /* Initialize the stack lock */ - - if (KERNEL_STACK_SIZE < round_page_32(KERNEL_STACK_SIZE)) { /* Kernel stacks must be multiples of pages */ - panic("thread_init: kernel stack size (%08X) must be a multiple of page size (%08X)\n", - KERNEL_STACK_SIZE, PAGE_SIZE); - } - - for(stack_alloc_bndry = PAGE_SIZE; stack_alloc_bndry <= KERNEL_STACK_SIZE; stack_alloc_bndry <<= 1); /* Find next power of 2 above stack size */ + task->c_switch += thread->c_switch; + task->p_switch += thread->p_switch; + task->ps_switch += thread->ps_switch; - ret = kmem_suballoc(kernel_map, /* Suballocate from the kernel map */ + queue_remove(&task->threads, thread, thread_t, task_threads); + task->thread_count--; - &stack, - (stack_alloc_bndry * (2*THREAD_MAX + 64)), /* Allocate enough for all of it */ - FALSE, /* Say not pageable so that it is wired */ - TRUE, /* Allocate from anywhere */ - &stack_map); /* Allocate a submap */ - - if(ret != KERN_SUCCESS) { /* Did we get one? */ - panic("thread_init: kmem_suballoc for stacks failed - ret = %d\n", ret); /* Die */ - } - stack = vm_map_min(stack_map); /* Make sure we skip the first hunk */ - ret = vm_map_enter(stack_map, &stack, PAGE_SIZE, 0, /* Make sure there is nothing at the start */ - 0, /* Force it at start */ - VM_OBJECT_NULL, 0, /* No object yet */ - FALSE, /* No copy */ - VM_PROT_NONE, /* Allow no access */ - VM_PROT_NONE, /* Allow no access */ - VM_INHERIT_DEFAULT); /* Just be normal */ - - if(ret != KERN_SUCCESS) { /* Did it work? */ - panic("thread_init: dummy alignment allocation failed; ret = %d\n", ret); - } - -#endif /* MACHINE_STACK */ + /* + * If the task is being halted, and there is only one thread + * left in the task after this one, then wakeup that thread. + */ + if (task->thread_count == 1 && task->halting) + thread_wakeup((event_t)&task->halting); -#if MACH_LDEBUG - thr_sh_template.mutex_count = 0; -#endif /* MACH_LDEBUG */ + task_unlock(task); - { - uint64_t abstime; - - clock_interval_to_absolutetime_interval( - std_quantum_us, NSEC_PER_USEC, &abstime); - assert((abstime >> 32) == 0 && (uint32_t)abstime != 0); - std_quantum = abstime; - - /* 250 us */ - clock_interval_to_absolutetime_interval(250, NSEC_PER_USEC, &abstime); - assert((abstime >> 32) == 0 && (uint32_t)abstime != 0); - min_std_quantum = abstime; - - /* 50 us */ - clock_interval_to_absolutetime_interval(50, NSEC_PER_USEC, &abstime); - assert((abstime >> 32) == 0 && (uint32_t)abstime != 0); - min_rt_quantum = abstime; - - /* 50 ms */ - clock_interval_to_absolutetime_interval( - 50, 1000*NSEC_PER_USEC, &abstime); - assert((abstime >> 32) == 0 && (uint32_t)abstime != 0); - max_rt_quantum = abstime; - - max_unsafe_computation = max_unsafe_quanta * std_quantum; - max_poll_computation = max_poll_quanta * std_quantum; - - sched_safe_duration = 2 * max_unsafe_quanta * - (std_quantum_us / (1000 * 1000)) * - (1 << SCHED_TICK_SHIFT); + lck_mtx_lock(&tasks_threads_lock); + queue_remove(&threads, thread, thread_t, threads); + threads_count--; + lck_mtx_unlock(&tasks_threads_lock); + + thread_deallocate(thread); + + (void)splsched(); + simple_lock(&thread_terminate_lock); } - first_thread = TRUE; - /* - * Initialize any machine-dependent - * per-thread structures necessary. - */ - thread_machine_init(); + assert_wait((event_t)&thread_terminate_queue, THREAD_UNINT); + simple_unlock(&thread_terminate_lock); + /* splsched */ + + thread_block((thread_continue_t)thread_terminate_daemon); + /*NOTREACHED*/ } /* - * Called at splsched. + * thread_terminate_enqueue: + * + * Enqueue a terminating thread for final disposition. + * + * Called at splsched. */ void -thread_reaper_enqueue( +thread_terminate_enqueue( thread_t thread) { - simple_lock(&reaper_lock); - enqueue_tail(&reaper_queue, (queue_entry_t)thread); - simple_unlock(&reaper_lock); - - thread_wakeup((event_t)&reaper_queue); -} + simple_lock(&thread_terminate_lock); + enqueue_tail(&thread_terminate_queue, (queue_entry_t)thread); + simple_unlock(&thread_terminate_lock); -void -thread_termination_continue(void) -{ - panic("thread_termination_continue"); - /*NOTREACHED*/ + thread_wakeup((event_t)&thread_terminate_queue); } /* - * Routine: thread_terminate_self - * - * This routine is called by a thread which has unwound from - * its current RPC and kernel contexts and found that it's - * root activation has been marked for extinction. This lets - * it clean up the last few things that can only be cleaned - * up in this context and then impale itself on the reaper - * queue. + * thread_stack_daemon: * - * When the reaper gets the thread, it will deallocate the - * thread_act's reference on itself, which in turn will release - * its own reference on this thread. By doing things in that - * order, a thread_act will always have a valid thread - but the - * thread may persist beyond having a thread_act (but must never - * run like that). + * Perform stack allocation as required due to + * invoke failures. */ -void -thread_terminate_self(void) +static void +thread_stack_daemon(void) { - thread_act_t thr_act = current_act(); thread_t thread; - task_t task = thr_act->task; - long active_acts; - spl_t s; - /* - * We should be at the base of the inheritance chain. - */ - thread = act_lock_thread(thr_act); - assert(thr_act->thread == thread); + simple_lock(&thread_stack_lock); - /* This will allow no more control ops on this thr_act. */ - ipc_thr_act_disable(thr_act); + while ((thread = (thread_t)dequeue_head(&thread_stack_queue)) != THREAD_NULL) { + simple_unlock(&thread_stack_lock); - /* Clean-up any ulocks that are still owned by the thread - * activation (acquired but not released or handed-off). - */ - act_ulock_release_all(thr_act); - - act_unlock_thread(thr_act); + stack_alloc(thread); + + (void)splsched(); + thread_lock(thread); + thread_setrun(thread, SCHED_PREEMPT | SCHED_TAILQ); + thread_unlock(thread); + (void)spllo(); - _mk_sp_thread_depress_abort(thread, TRUE); + simple_lock(&thread_stack_lock); + } - /* - * Check to see if this is the last active activation. By - * this we mean the last activation to call thread_terminate_self. - * If so, and the task is associated with a BSD process, we - * need to call BSD and let them clean up. - */ - active_acts = hw_atomic_sub(&task->active_act_count, 1); + assert_wait((event_t)&thread_stack_queue, THREAD_UNINT); + simple_unlock(&thread_stack_lock); - if (active_acts == 0 && task->bsd_info) - proc_exit(task->bsd_info); + thread_block((thread_continue_t)thread_stack_daemon); + /*NOTREACHED*/ +} - /* JMM - for now, no migration */ - assert(!thr_act->lower); +/* + * thread_stack_enqueue: + * + * Enqueue a thread for stack allocation. + * + * Called at splsched. + */ +void +thread_stack_enqueue( + thread_t thread) +{ + simple_lock(&thread_stack_lock); + enqueue_tail(&thread_stack_queue, (queue_entry_t)thread); + simple_unlock(&thread_stack_lock); - s = splsched(); - thread_lock(thread); - thread->active = FALSE; - thread_unlock(thread); - splx(s); + thread_wakeup((event_t)&thread_stack_queue); +} - thread_timer_terminate(); +void +thread_daemon_init(void) +{ + kern_return_t result; + thread_t thread; - /* flush any lazy HW state while in own context */ - thread_machine_flush(thr_act); + simple_lock_init(&thread_terminate_lock, 0); + queue_init(&thread_terminate_queue); - ipc_thread_terminate(thread); + result = kernel_thread_start_priority((thread_continue_t)thread_terminate_daemon, NULL, MINPRI_KERNEL, &thread); + if (result != KERN_SUCCESS) + panic("thread_daemon_init: thread_terminate_daemon"); - s = splsched(); - thread_lock(thread); - thread->state |= TH_TERMINATE; - assert((thread->state & TH_UNINT) == 0); - thread_mark_wait_locked(thread, THREAD_UNINT); - assert(thread->promotions == 0); - thread_unlock(thread); - /* splx(s); */ + thread_deallocate(thread); - ETAP_SET_REASON(thread, BLOCKED_ON_TERMINATION); - thread_block(thread_termination_continue); - /*NOTREACHED*/ + simple_lock_init(&thread_stack_lock, 0); + queue_init(&thread_stack_queue); + + result = kernel_thread_start_priority((thread_continue_t)thread_stack_daemon, NULL, BASEPRI_PREEMPT, &thread); + if (result != KERN_SUCCESS) + panic("thread_daemon_init: thread_stack_daemon"); + + thread_deallocate(thread); } /* * Create a new thread. - * Doesn't start the thread running; It first must be attached to - * an activation - then use thread_go to start it. + * Doesn't start the thread running. */ -kern_return_t -thread_create_shuttle( - thread_act_t thr_act, +static kern_return_t +thread_create_internal( + task_t parent_task, integer_t priority, - void (*start)(void), - thread_t *new_thread) + thread_continue_t continuation, + int options, +#define TH_OPTION_NONE 0x00 +#define TH_OPTION_NOCRED 0x01 +#define TH_OPTION_NOSUSP 0x02 + thread_t *out_thread) { - kern_return_t result; - thread_t new_shuttle; - task_t parent_task = thr_act->task; - processor_set_t pset; + thread_t new_thread; + static thread_t first_thread; /* * Allocate a thread and initialize static fields */ - if (first_thread) { - new_shuttle = &pageout_thread; - first_thread = FALSE; - } else - new_shuttle = (thread_t)zalloc(thread_shuttle_zone); - if (new_shuttle == THREAD_NULL) + if (first_thread == THREAD_NULL) + new_thread = first_thread = current_thread(); + else + new_thread = (thread_t)zalloc(thread_zone); + if (new_thread == THREAD_NULL) return (KERN_RESOURCE_SHORTAGE); -#ifdef DEBUG - if (new_shuttle != &pageout_thread) - assert(!thr_act->thread); -#endif + if (new_thread != first_thread) + *new_thread = thread_template; - *new_shuttle = thr_sh_template; +#ifdef MACH_BSD + new_thread->uthread = uthread_alloc(parent_task, new_thread, (options & TH_OPTION_NOCRED) != 0); + if (new_thread->uthread == NULL) { + zfree(thread_zone, new_thread); + return (KERN_RESOURCE_SHORTAGE); + } +#endif /* MACH_BSD */ - thread_lock_init(new_shuttle); - wake_lock_init(new_shuttle); - new_shuttle->sleep_stamp = sched_tick; + if (machine_thread_create(new_thread, parent_task) != KERN_SUCCESS) { +#ifdef MACH_BSD + void *ut = new_thread->uthread; - /* - * Thread still isn't runnable yet (our caller will do - * that). Initialize runtime-dependent fields here. - */ - result = thread_machine_create(new_shuttle, thr_act, thread_continue); - assert (result == KERN_SUCCESS); + new_thread->uthread = NULL; + /* cred free may not be necessary */ + uthread_cleanup(parent_task, ut, parent_task->bsd_info); + uthread_cred_free(ut); + uthread_zone_free(ut); +#endif /* MACH_BSD */ - thread_start(new_shuttle, start); - thread_timer_setup(new_shuttle); - ipc_thread_init(new_shuttle); + zfree(thread_zone, new_thread); + return (KERN_FAILURE); + } - pset = parent_task->processor_set; - assert(pset == &default_pset); - pset_lock(pset); + new_thread->task = parent_task; + thread_lock_init(new_thread); + wake_lock_init(new_thread); + + lck_mtx_init(&new_thread->mutex, &thread_lck_grp, &thread_lck_attr); + + ipc_thread_init(new_thread); + queue_init(&new_thread->held_ulocks); + + new_thread->continuation = continuation; + + lck_mtx_lock(&tasks_threads_lock); task_lock(parent_task); - assert(parent_task->processor_set == pset); - /* - * Don't need to initialize because the context switch - * code will set it before it can be used. - */ - if (!parent_task->active) { + if ( !parent_task->active || parent_task->halting || + ((options & TH_OPTION_NOSUSP) != 0 && + parent_task->suspend_count > 0) || + (parent_task->thread_count >= task_threadmax && + parent_task != kernel_task) ) { task_unlock(parent_task); - pset_unlock(pset); - thread_machine_destroy(new_shuttle); - zfree(thread_shuttle_zone, (vm_offset_t) new_shuttle); + lck_mtx_unlock(&tasks_threads_lock); + +#ifdef MACH_BSD + { + void *ut = new_thread->uthread; + + new_thread->uthread = NULL; + uthread_cleanup(parent_task, ut, parent_task->bsd_info); + /* cred free may not be necessary */ + uthread_cred_free(ut); + uthread_zone_free(ut); + } +#endif /* MACH_BSD */ + ipc_thread_disable(new_thread); + ipc_thread_terminate(new_thread); + lck_mtx_destroy(&new_thread->mutex, &thread_lck_grp); + machine_thread_destroy(new_thread); + zfree(thread_zone, new_thread); return (KERN_FAILURE); } - act_attach(thr_act, new_shuttle, 0); + /* New threads inherit any default state on the task */ + machine_thread_inherit_taskwide(new_thread, parent_task); + + task_reference_internal(parent_task); + + /* Cache the task's map */ + new_thread->map = parent_task->map; - /* Chain the thr_act onto the task's list */ - queue_enter(&parent_task->thr_acts, thr_act, thread_act_t, thr_acts); - parent_task->thr_act_count++; - parent_task->res_act_count++; + /* Chain the thread onto the task's list */ + queue_enter(&parent_task->threads, new_thread, thread_t, task_threads); + parent_task->thread_count++; /* So terminating threads don't need to take the task lock to decrement */ - hw_atomic_add(&parent_task->active_act_count, 1); + hw_atomic_add(&parent_task->active_thread_count, 1); - /* Associate the thread with the processor set */ - pset_add_thread(pset, new_shuttle); + /* Protected by the tasks_threads_lock */ + new_thread->thread_id = ++thread_unique_id; + + queue_enter(&threads, new_thread, thread_t, threads); + threads_count++; + + timer_call_setup(&new_thread->wait_timer, thread_timer_expire, new_thread); + timer_call_setup(&new_thread->depress_timer, thread_depress_expire, new_thread); + +#if CONFIG_COUNTERS + /* + * If parent task has any reservations, they need to be propagated to this + * thread. + */ + new_thread->t_chud = (TASK_PMC_FLAG == (parent_task->t_chud & TASK_PMC_FLAG)) ? + THREAD_PMC_FLAG : 0U; +#endif /* Set the thread's scheduling parameters */ if (parent_task != kernel_task) - new_shuttle->sched_mode |= TH_MODE_TIMESHARE; - new_shuttle->max_priority = parent_task->max_priority; - new_shuttle->task_priority = parent_task->priority; - new_shuttle->priority = (priority < 0)? parent_task->priority: priority; - if (new_shuttle->priority > new_shuttle->max_priority) - new_shuttle->priority = new_shuttle->max_priority; - new_shuttle->importance = - new_shuttle->priority - new_shuttle->task_priority; - new_shuttle->sched_stamp = sched_tick; - compute_priority(new_shuttle, FALSE); - -#if ETAP_EVENT_MONITOR - new_thread->etap_reason = 0; - new_thread->etap_trace = FALSE; -#endif /* ETAP_EVENT_MONITOR */ - - new_shuttle->active = TRUE; - thr_act->active = TRUE; - - *new_thread = new_shuttle; + new_thread->sched_mode |= TH_MODE_TIMESHARE; + new_thread->max_priority = parent_task->max_priority; + new_thread->task_priority = parent_task->priority; + new_thread->priority = (priority < 0)? parent_task->priority: priority; + if (new_thread->priority > new_thread->max_priority) + new_thread->priority = new_thread->max_priority; + new_thread->importance = + new_thread->priority - new_thread->task_priority; + new_thread->sched_stamp = sched_tick; + new_thread->pri_shift = sched_pri_shift; + compute_priority(new_thread, FALSE); + + new_thread->active = TRUE; + + *out_thread = new_thread; { long dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4; + kdbg_trace_data(parent_task->bsd_info, &dbg_arg2); + KERNEL_DEBUG_CONSTANT( TRACEDBG_CODE(DBG_TRACE_DATA, 1) | DBG_FUNC_NONE, - (vm_address_t)new_shuttle, 0, 0, 0, 0); + (vm_address_t)(uintptr_t)thread_tid(new_thread), dbg_arg2, 0, 0, 0); kdbg_trace_string(parent_task->bsd_info, &dbg_arg1, &dbg_arg2, &dbg_arg3, &dbg_arg4); @@ -888,42 +746,35 @@ thread_create_shuttle( dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, 0); } + DTRACE_PROC1(lwp__create, thread_t, *out_thread); + return (KERN_SUCCESS); } -extern void thread_bootstrap_return(void); - kern_return_t thread_create( task_t task, - thread_act_t *new_act) + thread_t *new_thread) { kern_return_t result; thread_t thread; - thread_act_t act; - if (task == TASK_NULL) - return KERN_INVALID_ARGUMENT; + if (task == TASK_NULL || task == kernel_task) + return (KERN_INVALID_ARGUMENT); - result = act_create(task, &act); + result = thread_create_internal(task, -1, (thread_continue_t)thread_bootstrap_return, TH_OPTION_NONE, &thread); if (result != KERN_SUCCESS) return (result); - result = thread_create_shuttle(act, -1, thread_bootstrap_return, &thread); - if (result != KERN_SUCCESS) { - act_deallocate(act); - return (result); - } - - act->user_stop_count = 1; - thread_hold(act); + thread->user_stop_count = 1; + thread_hold(thread); if (task->suspend_count > 0) - thread_hold(act); + thread_hold(thread); - pset_unlock(task->processor_set); task_unlock(task); + lck_mtx_unlock(&tasks_threads_lock); - *new_act = act; + *new_thread = thread; return (KERN_SUCCESS); } @@ -934,202 +785,173 @@ thread_create_running( int flavor, thread_state_t new_state, mach_msg_type_number_t new_state_count, - thread_act_t *new_act) /* OUT */ + thread_t *new_thread) { register kern_return_t result; thread_t thread; - thread_act_t act; - if (task == TASK_NULL) - return KERN_INVALID_ARGUMENT; + if (task == TASK_NULL || task == kernel_task) + return (KERN_INVALID_ARGUMENT); - result = act_create(task, &act); + result = thread_create_internal(task, -1, (thread_continue_t)thread_bootstrap_return, TH_OPTION_NONE, &thread); if (result != KERN_SUCCESS) return (result); - result = thread_create_shuttle(act, -1, thread_bootstrap_return, &thread); - if (result != KERN_SUCCESS) { - act_deallocate(act); - return (result); - } - - act_lock(act); - result = act_machine_set_state(act, flavor, new_state, new_state_count); + result = machine_thread_set_state( + thread, flavor, new_state, new_state_count); if (result != KERN_SUCCESS) { - act_unlock(act); - pset_unlock(task->processor_set); task_unlock(task); + lck_mtx_unlock(&tasks_threads_lock); - (void)thread_terminate(act); + thread_terminate(thread); + thread_deallocate(thread); return (result); } - clear_wait(thread, THREAD_AWAKENED); - act->inited = TRUE; - act_unlock(act); - pset_unlock(task->processor_set); + thread_mtx_lock(thread); + thread_start_internal(thread); + thread_mtx_unlock(thread); + task_unlock(task); + lck_mtx_unlock(&tasks_threads_lock); - *new_act = act; + *new_thread = thread; return (result); } -/* - * kernel_thread: - * - * Create and kernel thread in the specified task, and - * optionally start it running. - */ -thread_t -kernel_thread_with_priority( +kern_return_t +thread_create_workq( task_t task, - integer_t priority, - void (*start)(void), - boolean_t alloc_stack, - boolean_t start_running) + thread_continue_t thread_return, + thread_t *new_thread) { kern_return_t result; thread_t thread; - thread_act_t act; - result = act_create(task, &act); + if (task == TASK_NULL || task == kernel_task) + return (KERN_INVALID_ARGUMENT); + + result = thread_create_internal(task, -1, thread_return, TH_OPTION_NOCRED | TH_OPTION_NOSUSP, &thread); if (result != KERN_SUCCESS) - return (THREAD_NULL); + return (result); - result = thread_create_shuttle(act, priority, start, &thread); - if (result != KERN_SUCCESS) { - act_deallocate(act); - return (THREAD_NULL); - } + thread->user_stop_count = 1; + thread_hold(thread); + if (task->suspend_count > 0) + thread_hold(thread); - pset_unlock(task->processor_set); task_unlock(task); + lck_mtx_unlock(&tasks_threads_lock); + + *new_thread = thread; - if (alloc_stack) - thread_doswapin(thread); - - act_lock(act); - if (start_running) - clear_wait(thread, THREAD_AWAKENED); - act->inited = TRUE; - act_unlock(act); - - act_deallocate(act); - - return (thread); + return (KERN_SUCCESS); } -thread_t -kernel_thread( - task_t task, - void (*start)(void)) +/* + * kernel_thread_create: + * + * Create a thread in the kernel task + * to execute in kernel context. + */ +kern_return_t +kernel_thread_create( + thread_continue_t continuation, + void *parameter, + integer_t priority, + thread_t *new_thread) { - return kernel_thread_with_priority(task, -1, start, FALSE, TRUE); -} - -unsigned int c_weird_pset_ref_exit = 0; /* pset code raced us */ - -#if MACH_HOST -/* Preclude thread processor set assignement */ -#define thread_freeze(thread) assert((thread)->processor_set == &default_pset) - -/* Allow thread processor set assignement */ -#define thread_unfreeze(thread) assert((thread)->processor_set == &default_pset) - -#endif /* MACH_HOST */ + kern_return_t result; + thread_t thread; + task_t task = kernel_task; -void -thread_deallocate( - thread_t thread) -{ - task_t task; - processor_set_t pset; - int refs; - spl_t s; + result = thread_create_internal(task, priority, continuation, TH_OPTION_NONE, &thread); + if (result != KERN_SUCCESS) + return (result); - if (thread == THREAD_NULL) - return; + task_unlock(task); + lck_mtx_unlock(&tasks_threads_lock); - /* - * First, check for new count > 0 (the common case). - * Only the thread needs to be locked. - */ - s = splsched(); - thread_lock(thread); - refs = --thread->ref_count; - thread_unlock(thread); - splx(s); + stack_alloc(thread); + assert(thread->kernel_stack != 0); +#if CONFIG_EMBEDDED + if (priority > BASEPRI_KERNEL) +#endif + thread->reserved_stack = thread->kernel_stack; - if (refs > 0) - return; + thread->parameter = parameter; - if (thread == current_thread()) - panic("thread deallocating itself"); +if(debug_task & 1) + kprintf("kernel_thread_create: thread = %p continuation = %p\n", thread, continuation); + *new_thread = thread; - /* - * There is a dangling pointer to the thread from the - * processor_set. To clean it up, we freeze the thread - * in the pset (because pset destruction can cause even - * reference-less threads to be reassigned to the default - * pset) and then remove it. - */ + return (result); +} -#if MACH_HOST - thread_freeze(thread); -#endif +kern_return_t +kernel_thread_start_priority( + thread_continue_t continuation, + void *parameter, + integer_t priority, + thread_t *new_thread) +{ + kern_return_t result; + thread_t thread; - pset = thread->processor_set; - pset_lock(pset); - pset_remove_thread(pset, thread); - pset_unlock(pset); + result = kernel_thread_create(continuation, parameter, priority, &thread); + if (result != KERN_SUCCESS) + return (result); -#if MACH_HOST - thread_unfreeze(thread); -#endif + *new_thread = thread; - pset_deallocate(pset); + thread_mtx_lock(thread); + thread_start_internal(thread); + thread_mtx_unlock(thread); - if (thread->stack_privilege != 0) { - if (thread->stack_privilege != thread->kernel_stack) - stack_free_stack(thread->stack_privilege); - thread->stack_privilege = 0; - } - /* frees kernel stack & other MD resources */ - thread_machine_destroy(thread); + return (result); +} - zfree(thread_shuttle_zone, (vm_offset_t) thread); +kern_return_t +kernel_thread_start( + thread_continue_t continuation, + void *parameter, + thread_t *new_thread) +{ + return kernel_thread_start_priority(continuation, parameter, -1, new_thread); } -void -thread_reference( - thread_t thread) +#ifndef __LP64__ + +thread_t +kernel_thread( + task_t task, + void (*start)(void)) { - spl_t s; + kern_return_t result; + thread_t thread; - if (thread == THREAD_NULL) - return; + if (task != kernel_task) + panic("kernel_thread"); - s = splsched(); - thread_lock(thread); - thread_reference_locked(thread); - thread_unlock(thread); - splx(s); + result = kernel_thread_start_priority((thread_continue_t)start, NULL, -1, &thread); + if (result != KERN_SUCCESS) + return (THREAD_NULL); + + thread_deallocate(thread); + + return (thread); } -/* - * Called with "appropriate" thread-related locks held on - * thread and its top_act for synchrony with RPC (see - * act_lock_thread()). - */ +#endif /* __LP64__ */ + kern_return_t -thread_info_shuttle( - register thread_act_t thr_act, +thread_info_internal( + register thread_t thread, thread_flavor_t flavor, thread_info_t thread_info_out, /* ptr to OUT array */ mach_msg_type_number_t *thread_info_count) /*IN/OUT*/ { - register thread_t thread = thr_act->thread; int state, flags; spl_t s; @@ -1165,24 +987,21 @@ thread_info_shuttle( * then for 5/8 ageing. The correction factor [3/5] is * (1/(5/8) - 1). */ - basic_info->cpu_usage = (thread->cpu_usage << SCHED_TICK_SHIFT) / - (TIMER_RATE / TH_USAGE_SCALE); + basic_info->cpu_usage = (integer_t)(((uint64_t)thread->cpu_usage + * TH_USAGE_SCALE) / sched_tick_interval); basic_info->cpu_usage = (basic_info->cpu_usage * 3) / 5; -#if SIMPLE_CLOCK - /* - * Clock drift compensation. - */ - basic_info->cpu_usage = (basic_info->cpu_usage * 1000000) / sched_usec; -#endif /* SIMPLE_CLOCK */ + + if (basic_info->cpu_usage > TH_USAGE_SCALE) + basic_info->cpu_usage = TH_USAGE_SCALE; basic_info->policy = ((thread->sched_mode & TH_MODE_TIMESHARE)? POLICY_TIMESHARE: POLICY_RR); flags = 0; - if (thread->state & TH_IDLE) + if (thread->bound_processor != PROCESSOR_NULL && thread->bound_processor->idle_thread == thread) flags |= TH_FLAGS_IDLE; - if (thread->state & TH_STACK_HANDOFF) + if (!thread->kernel_stack) flags |= TH_FLAGS_SWAPPED; state = 0; @@ -1204,7 +1023,7 @@ thread_info_shuttle( basic_info->run_state = state; basic_info->flags = flags; - basic_info->suspend_count = thr_act->user_stop_count; + basic_info->suspend_count = thread->user_stop_count; thread_unlock(thread); splx(s); @@ -1214,6 +1033,36 @@ thread_info_shuttle( return (KERN_SUCCESS); } else + if (flavor == THREAD_IDENTIFIER_INFO) { + register thread_identifier_info_t identifier_info; + + if (*thread_info_count < THREAD_IDENTIFIER_INFO_COUNT) + return (KERN_INVALID_ARGUMENT); + + identifier_info = (thread_identifier_info_t) thread_info_out; + + s = splsched(); + thread_lock(thread); + + identifier_info->thread_id = thread->thread_id; +#if defined(__ppc__) || defined(__arm__) + identifier_info->thread_handle = thread->machine.cthread_self; +#else + identifier_info->thread_handle = thread->machine.pcb->cthread_self; +#endif + if(thread->task->bsd_info) { + identifier_info->dispatch_qaddr = identifier_info->thread_handle + get_dispatchqueue_offset_from_proc(thread->task->bsd_info); + } else { + thread_unlock(thread); + splx(s); + return KERN_INVALID_ARGUMENT; + } + + thread_unlock(thread); + splx(s); + return KERN_SUCCESS; + } + else if (flavor == THREAD_SCHED_TIMESHARE_INFO) { policy_timeshare_info_t ts_info; @@ -1303,87 +1152,29 @@ thread_info_shuttle( } void -thread_doreap( - register thread_t thread) -{ - thread_act_t thr_act; - - - thr_act = thread_lock_act(thread); - assert(thr_act && thr_act->thread == thread); - - act_locked_act_reference(thr_act); - - /* - * Replace `act_unlock_thread()' with individual - * calls. (`act_detach()' can change fields used - * to determine which locks are held, confusing - * `act_unlock_thread()'.) - */ - act_unlock(thr_act); - - /* Remove the reference held by a rooted thread */ - act_deallocate(thr_act); - - /* Remove the reference held by the thread: */ - act_deallocate(thr_act); -} - -/* - * reaper_thread: - * - * This kernel thread runs forever looking for terminating - * threads, releasing their "self" references. - */ -static void -reaper_thread_continue(void) +thread_read_times( + thread_t thread, + time_value_t *user_time, + time_value_t *system_time) { - register thread_t thread; - - (void)splsched(); - simple_lock(&reaper_lock); - - while ((thread = (thread_t) dequeue_head(&reaper_queue)) != THREAD_NULL) { - simple_unlock(&reaper_lock); - (void)spllo(); - - thread_doreap(thread); - - (void)splsched(); - simple_lock(&reaper_lock); - } - - assert_wait((event_t)&reaper_queue, THREAD_UNINT); - simple_unlock(&reaper_lock); - (void)spllo(); - - thread_block(reaper_thread_continue); - /*NOTREACHED*/ -} - -static void -reaper_thread(void) -{ - thread_t self = current_thread(); - - stack_privilege(self); + clock_sec_t secs; + clock_usec_t usecs; - reaper_thread_continue(); - /*NOTREACHED*/ -} + absolutetime_to_microtime(timer_grab(&thread->user_timer), &secs, &usecs); + user_time->seconds = (typeof(user_time->seconds))secs; + user_time->microseconds = usecs; -void -thread_reaper_init(void) -{ - kernel_thread(kernel_task, reaper_thread); + absolutetime_to_microtime(timer_grab(&thread->system_timer), &secs, &usecs); + system_time->seconds = (typeof(system_time->seconds))secs; + system_time->microseconds = usecs; } kern_return_t thread_assign( - thread_act_t thr_act, - processor_set_t new_pset) + __unused thread_t thread, + __unused processor_set_t new_pset) { - return(KERN_FAILURE); + return (KERN_FAILURE); } /* @@ -1394,9 +1185,9 @@ thread_assign( */ kern_return_t thread_assign_default( - thread_act_t thr_act) + thread_t thread) { - return (thread_assign(thr_act, &default_pset)); + return (thread_assign(thread, &pset0)); } /* @@ -1406,292 +1197,90 @@ thread_assign_default( */ kern_return_t thread_get_assignment( - thread_act_t thr_act, + thread_t thread, processor_set_t *pset) { - thread_t thread; - - if (thr_act == THR_ACT_NULL) - return(KERN_INVALID_ARGUMENT); - thread = act_lock_thread(thr_act); - if (thread == THREAD_NULL) { - act_unlock_thread(thr_act); - return(KERN_INVALID_ARGUMENT); - } - *pset = thread->processor_set; - act_unlock_thread(thr_act); - pset_reference(*pset); - return(KERN_SUCCESS); + if (thread == NULL) + return (KERN_INVALID_ARGUMENT); + + *pset = &pset0; + + return (KERN_SUCCESS); } /* - * thread_wire: + * thread_wire_internal: * * Specify that the target thread must always be able * to run and to allocate memory. */ kern_return_t -thread_wire( - host_priv_t host_priv, - thread_act_t thr_act, - boolean_t wired) +thread_wire_internal( + host_priv_t host_priv, + thread_t thread, + boolean_t wired, + boolean_t *prev_state) { - spl_t s; - thread_t thread; - extern void vm_page_free_reserve(int pages); - - if (thr_act == THR_ACT_NULL || host_priv == HOST_PRIV_NULL) + if (host_priv == NULL || thread != current_thread()) return (KERN_INVALID_ARGUMENT); assert(host_priv == &realhost); - thread = act_lock_thread(thr_act); - if (thread ==THREAD_NULL) { - act_unlock_thread(thr_act); - return(KERN_INVALID_ARGUMENT); - } - - /* - * This implementation only works for the current thread. - * See stack_privilege. - */ - if (thr_act != current_act()) - return KERN_INVALID_ARGUMENT; - - s = splsched(); - thread_lock(thread); - + if (prev_state) + *prev_state = (thread->options & TH_OPT_VMPRIV) != 0; + if (wired) { - if (thread->vm_privilege == FALSE) + if (!(thread->options & TH_OPT_VMPRIV)) vm_page_free_reserve(1); /* XXX */ - thread->vm_privilege = TRUE; - } else { - if (thread->vm_privilege == TRUE) + thread->options |= TH_OPT_VMPRIV; + } + else { + if (thread->options & TH_OPT_VMPRIV) vm_page_free_reserve(-1); /* XXX */ - thread->vm_privilege = FALSE; + thread->options &= ~TH_OPT_VMPRIV; } - thread_unlock(thread); - splx(s); - act_unlock_thread(thr_act); - - return KERN_SUCCESS; -} - -/* - * thread_collect_scan: - * - * Attempt to free resources owned by threads. - */ - -void -thread_collect_scan(void) -{ - /* This code runs very quickly! */ + return (KERN_SUCCESS); } -/* Also disabled in vm/vm_pageout.c */ -boolean_t thread_collect_allowed = FALSE; -unsigned thread_collect_last_tick = 0; -unsigned thread_collect_max_rate = 0; /* in ticks */ /* - * consider_thread_collect: + * thread_wire: * - * Called by the pageout daemon when the system needs more free pages. + * User-api wrapper for thread_wire_internal() */ - -void -consider_thread_collect(void) -{ - /* - * By default, don't attempt thread collection more frequently - * than once a second. - */ - - if (thread_collect_max_rate == 0) - thread_collect_max_rate = (1 << SCHED_TICK_SHIFT) + 1; - - if (thread_collect_allowed && - (sched_tick > - (thread_collect_last_tick + thread_collect_max_rate))) { - thread_collect_last_tick = sched_tick; - thread_collect_scan(); - } -} - kern_return_t -host_stack_usage( - host_t host, - vm_size_t *reservedp, - unsigned int *totalp, - vm_size_t *spacep, - vm_size_t *residentp, - vm_size_t *maxusagep, - vm_offset_t *maxstackp) +thread_wire( + host_priv_t host_priv, + thread_t thread, + boolean_t wired) { -#if !MACH_DEBUG - return KERN_NOT_SUPPORTED; -#else - unsigned int total; - vm_size_t maxusage; - - if (host == HOST_NULL) - return KERN_INVALID_HOST; - - maxusage = 0; - - stack_statistics(&total, &maxusage); - - *reservedp = 0; - *totalp = total; - *spacep = *residentp = total * round_page_32(KERNEL_STACK_SIZE); - *maxusagep = maxusage; - *maxstackp = 0; - return KERN_SUCCESS; - -#endif /* MACH_DEBUG */ + return (thread_wire_internal(host_priv, thread, wired, NULL)); } -/* - * Return info on stack usage for threads in a specific processor set - */ -kern_return_t -processor_set_stack_usage( - processor_set_t pset, - unsigned int *totalp, - vm_size_t *spacep, - vm_size_t *residentp, - vm_size_t *maxusagep, - vm_offset_t *maxstackp) -{ -#if !MACH_DEBUG - return KERN_NOT_SUPPORTED; -#else - unsigned int total; - vm_size_t maxusage; - vm_offset_t maxstack; - - register thread_t *threads; - register thread_t thread; - - unsigned int actual; /* this many things */ - unsigned int i; - - vm_size_t size, size_needed; - vm_offset_t addr; - - spl_t s; - - if (pset == PROCESSOR_SET_NULL) - return KERN_INVALID_ARGUMENT; - - size = 0; addr = 0; - - for (;;) { - pset_lock(pset); - if (!pset->active) { - pset_unlock(pset); - return KERN_INVALID_ARGUMENT; - } - - actual = pset->thread_count; - - /* do we have the memory we need? */ - - size_needed = actual * sizeof(thread_t); - if (size_needed <= size) - break; - - /* unlock the pset and allocate more memory */ - pset_unlock(pset); - - if (size != 0) - kfree(addr, size); - - assert(size_needed > 0); - size = size_needed; - - addr = kalloc(size); - if (addr == 0) - return KERN_RESOURCE_SHORTAGE; - } - - /* OK, have memory and the processor_set is locked & active */ - s = splsched(); - threads = (thread_t *) addr; - for (i = 0, thread = (thread_t) queue_first(&pset->threads); - !queue_end(&pset->threads, (queue_entry_t) thread); - thread = (thread_t) queue_next(&thread->pset_threads)) { - thread_lock(thread); - if (thread->ref_count > 0) { - thread_reference_locked(thread); - threads[i++] = thread; - } - thread_unlock(thread); - } - splx(s); - assert(i <= actual); - - /* can unlock processor set now that we have the thread refs */ - pset_unlock(pset); - - /* calculate maxusage and free thread references */ - - total = 0; - maxusage = 0; - maxstack = 0; - while (i > 0) { - int cpu; - thread_t thread = threads[--i]; - vm_offset_t stack = 0; - - /* - * thread->kernel_stack is only accurate if the - * thread isn't swapped and is not executing. - * - * Of course, we don't have the appropriate locks - * for these shenanigans. - */ - - stack = thread->kernel_stack; - - for (cpu = 0; cpu < NCPUS; cpu++) - if (cpu_to_processor(cpu)->cpu_data->active_thread == thread) { - stack = active_stacks[cpu]; - break; - } +int split_funnel_off = 0; +lck_grp_t *funnel_lck_grp = LCK_GRP_NULL; +lck_grp_attr_t *funnel_lck_grp_attr; +lck_attr_t *funnel_lck_attr; - if (stack != 0) { - total++; - } - - thread_deallocate(thread); - } - - if (size != 0) - kfree(addr, size); - - *totalp = total; - *residentp = *spacep = total * round_page_32(KERNEL_STACK_SIZE); - *maxusagep = maxusage; - *maxstackp = maxstack; - return KERN_SUCCESS; - -#endif /* MACH_DEBUG */ -} - -int split_funnel_off = 0; funnel_t * funnel_alloc( int type) { - mutex_t *m; - funnel_t * fnl; + lck_mtx_t *m; + funnel_t *fnl; + + if (funnel_lck_grp == LCK_GRP_NULL) { + funnel_lck_grp_attr = lck_grp_attr_alloc_init(); + + funnel_lck_grp = lck_grp_alloc_init("Funnel", funnel_lck_grp_attr); + + funnel_lck_attr = lck_attr_alloc_init(); + } if ((fnl = (funnel_t *)kalloc(sizeof(funnel_t))) != 0){ bzero((void *)fnl, sizeof(funnel_t)); - if ((m = mutex_alloc(0)) == (mutex_t *)NULL) { - kfree((vm_offset_t)fnl, sizeof(funnel_t)); + if ((m = lck_mtx_alloc_init(funnel_lck_grp, funnel_lck_attr)) == (lck_mtx_t *)NULL) { + kfree(fnl, sizeof(funnel_t)); return(THR_FUNNEL_NULL); } fnl->fnl_mutex = m; @@ -1704,34 +1293,26 @@ void funnel_free( funnel_t * fnl) { - mutex_free(fnl->fnl_mutex); + lck_mtx_free(fnl->fnl_mutex, funnel_lck_grp); if (fnl->fnl_oldmutex) - mutex_free(fnl->fnl_oldmutex); - kfree((vm_offset_t)fnl, sizeof(funnel_t)); + lck_mtx_free(fnl->fnl_oldmutex, funnel_lck_grp); + kfree(fnl, sizeof(funnel_t)); } void funnel_lock( funnel_t * fnl) { - mutex_t * m; - - m = fnl->fnl_mutex; -restart: - mutex_lock(m); + lck_mtx_lock(fnl->fnl_mutex); fnl->fnl_mtxholder = current_thread(); - if (split_funnel_off && (m != fnl->fnl_mutex)) { - mutex_unlock(m); - m = fnl->fnl_mutex; - goto restart; - } } void funnel_unlock( funnel_t * fnl) { - mutex_unlock(fnl->fnl_mutex); + lck_mtx_unlock(fnl->fnl_mutex); + fnl->fnl_mtxholder = NULL; fnl->fnl_mtxrelease = current_thread(); } @@ -1764,7 +1345,7 @@ thread_funnel_set( if (funneled == TRUE) { if (cur_thread->funnel_lock) - panic("Funnel lock called when holding one %x", cur_thread->funnel_lock); + panic("Funnel lock called when holding one %p", cur_thread->funnel_lock); KERNEL_DEBUG(0x6032428 | DBG_FUNC_NONE, fnl, 1, 0, 0, 0); funnel_lock(fnl); @@ -1794,68 +1375,193 @@ thread_funnel_set( return(funnel_state_prev); } -boolean_t -thread_funnel_merge( - funnel_t * fnl, - funnel_t * otherfnl) +static void +sched_call_null( +__unused int type, +__unused thread_t thread) { - mutex_t * m; - mutex_t * otherm; - funnel_t * gfnl; - extern int disable_funnel; - - if ((gfnl = thread_funnel_get()) == THR_FUNNEL_NULL) - panic("thread_funnel_merge called with no funnels held"); - - if (gfnl->fnl_type != 1) - panic("thread_funnel_merge called from non kernel funnel"); - - if (gfnl != fnl) - panic("thread_funnel_merge incorrect invocation"); - - if (disable_funnel || split_funnel_off) - return (KERN_FAILURE); - - m = fnl->fnl_mutex; - otherm = otherfnl->fnl_mutex; - - /* Acquire other funnel mutex */ - mutex_lock(otherm); - split_funnel_off = 1; - disable_funnel = 1; - otherfnl->fnl_mutex = m; - otherfnl->fnl_type = fnl->fnl_type; - otherfnl->fnl_oldmutex = otherm; /* save this for future use */ + return; +} - mutex_unlock(otherm); - return(KERN_SUCCESS); +void +thread_sched_call( + thread_t thread, + sched_call_t call) +{ + thread->sched_call = (call != NULL)? call: sched_call_null; } void -thread_set_cont_arg( - int arg) +thread_static_param( + thread_t thread, + boolean_t state) { - thread_t self = current_thread(); + thread_mtx_lock(thread); + thread->static_param = state; + thread_mtx_unlock(thread); +} - self->saved.misc = arg; +uint64_t +thread_tid( + thread_t thread) +{ + return (thread != THREAD_NULL? thread->thread_id: 0); } -int -thread_get_cont_arg(void) +uint64_t +thread_dispatchqaddr( + thread_t thread) { - thread_t self = current_thread(); + uint64_t dispatchqueue_addr = 0; + uint64_t thread_handle = 0; + + if (thread != THREAD_NULL) { +#if defined(__ppc__) || defined(__arm__) + thread_handle = thread->machine.cthread_self; +#else + thread_handle = thread->machine.pcb->cthread_self; +#endif + + if (thread->task->bsd_info) + dispatchqueue_addr = thread_handle + get_dispatchqueue_offset_from_proc(thread->task->bsd_info); + } - return (self->saved.misc); + return (dispatchqueue_addr); } /* * Export routines to other components for things that are done as macros * within the osfmk component. */ + +#undef thread_reference +void thread_reference(thread_t thread); +void +thread_reference( + thread_t thread) +{ + if (thread != THREAD_NULL) + thread_reference_internal(thread); +} + #undef thread_should_halt + boolean_t thread_should_halt( - thread_shuttle_t th) + thread_t th) +{ + return (thread_should_halt_fast(th)); +} + +#if CONFIG_DTRACE +uint32_t dtrace_get_thread_predcache(thread_t thread) +{ + if (thread != THREAD_NULL) + return thread->t_dtrace_predcache; + else + return 0; +} + +int64_t dtrace_get_thread_vtime(thread_t thread) +{ + if (thread != THREAD_NULL) + return thread->t_dtrace_vtime; + else + return 0; +} + +int64_t dtrace_get_thread_tracing(thread_t thread) +{ + if (thread != THREAD_NULL) + return thread->t_dtrace_tracing; + else + return 0; +} + +boolean_t dtrace_get_thread_reentering(thread_t thread) +{ + if (thread != THREAD_NULL) + return (thread->options & TH_OPT_DTRACE) ? TRUE : FALSE; + else + return 0; +} + +vm_offset_t dtrace_get_kernel_stack(thread_t thread) +{ + if (thread != THREAD_NULL) + return thread->kernel_stack; + else + return 0; +} + +int64_t dtrace_calc_thread_recent_vtime(thread_t thread) +{ +#if STAT_TIME + if (thread != THREAD_NULL) { + return timer_grab(&(thread->system_timer)) + timer_grab(&(thread->user_timer)); + } else + return 0; +#else + if (thread != THREAD_NULL) { + processor_t processor = current_processor(); + uint64_t abstime = mach_absolute_time(); + timer_t timer; + + timer = PROCESSOR_DATA(processor, thread_timer); + + return timer_grab(&(thread->system_timer)) + timer_grab(&(thread->user_timer)) + + (abstime - timer->tstamp); /* XXX need interrupts off to prevent missed time? */ + } else + return 0; +#endif +} + +void dtrace_set_thread_predcache(thread_t thread, uint32_t predcache) +{ + if (thread != THREAD_NULL) + thread->t_dtrace_predcache = predcache; +} + +void dtrace_set_thread_vtime(thread_t thread, int64_t vtime) +{ + if (thread != THREAD_NULL) + thread->t_dtrace_vtime = vtime; +} + +void dtrace_set_thread_tracing(thread_t thread, int64_t accum) +{ + if (thread != THREAD_NULL) + thread->t_dtrace_tracing = accum; +} + +void dtrace_set_thread_reentering(thread_t thread, boolean_t vbool) +{ + if (thread != THREAD_NULL) { + if (vbool) + thread->options |= TH_OPT_DTRACE; + else + thread->options &= (~TH_OPT_DTRACE); + } +} + +vm_offset_t dtrace_set_thread_recover(thread_t thread, vm_offset_t recover) +{ + vm_offset_t prev = 0; + + if (thread != THREAD_NULL) { + prev = thread->recover; + thread->recover = recover; + } + return prev; +} + +void dtrace_thread_bootstrap(void) { - return(thread_should_halt_fast(th)); + task_t task = current_task(); + if(task->thread_count == 1) { + DTRACE_PROC(start); + } + DTRACE_PROC(lwp__start); + } +#endif /* CONFIG_DTRACE */