X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/1c79356b52d46aa6b508fb032f5ae709b1f2897b..ccc36f2f2d89f9115c479db4439aa5c88de5b44a:/osfmk/kern/thread.c diff --git a/osfmk/kern/thread.c b/osfmk/kern/thread.c index 8d4523b15..85ebb7dcf 100644 --- a/osfmk/kern/thread.c +++ b/osfmk/kern/thread.c @@ -80,7 +80,6 @@ #include #include #include -#include #include #include @@ -102,7 +101,6 @@ #include #include #include -#include #include /*** ??? fix so this can be removed ***/ #include #include @@ -125,83 +123,32 @@ #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 */ +static struct zone *thread_zone; -struct zone *thread_shuttle_zone; - -queue_head_t reaper_queue; -decl_simple_lock_data(,reaper_lock) -thread_call_t thread_reaper_call; +static queue_head_t reaper_queue; +decl_simple_lock_data(static,reaper_lock) extern int tick; -extern void pcb_module_init(void); - /* private */ -static struct thread_shuttle thr_sh_template; +static struct thread thread_template, init_thread; #if MACH_DEBUG -#if STACK_USAGE -static void stack_init(vm_offset_t stack, unsigned int bytes); -void stack_finalize(vm_offset_t stack); -vm_size_t stack_usage(vm_offset_t stack); -#else /*STACK_USAGE*/ -#define stack_init(stack, size) -#define stack_finalize(stack) -#define stack_usage(stack) (vm_size_t)0 -#endif /*STACK_USAGE*/ #ifdef MACHINE_STACK -extern -#endif - void stack_statistics( +extern void stack_statistics( unsigned int *totalp, vm_size_t *maxusagep); - -#define STACK_MARKER 0xdeadbeef -#if STACK_USAGE -boolean_t stack_check_usage = TRUE; -#else /* STACK_USAGE */ -boolean_t stack_check_usage = FALSE; -#endif /* STACK_USAGE */ -decl_simple_lock_data(,stack_usage_lock) -vm_size_t stack_max_usage = 0; -vm_size_t stack_max_use = KERNEL_STACK_SIZE - 64; +#endif /* MACHINE_STACK */ #endif /* MACH_DEBUG */ -/* Forwards */ -void thread_collect_scan(void); - -kern_return_t thread_create_shuttle( - thread_act_t thr_act, - integer_t priority, - void (*start)(void), - thread_t *new_thread); - -extern void Load_context( - thread_t thread); - - -/* - * Machine-dependent code must define: - * thread_machine_init - * thread_machine_terminate - * thread_machine_collect - * - * The thread->pcb field is reserved for machine-dependent code. - */ - #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 @@ -216,20 +163,28 @@ extern void Load_context( * because stack_alloc_try/thread_invoke operate at splsched. */ -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) +decl_simple_lock_data(static,stack_lock_data) +#define stack_lock() simple_lock(&stack_lock_data) +#define stack_unlock() simple_unlock(&stack_lock_data) + +static vm_map_t stack_map; +static vm_offset_t stack_free_list; + +static vm_offset_t stack_free_cache[NCPUS]; -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; /* patchable */ +unsigned int stack_free_count = 0; /* splsched only */ +unsigned int stack_free_limit = 1; /* Arbitrary */ -unsigned int stack_alloc_hits = 0; /* debugging */ +unsigned int stack_cache_hits = 0; /* debugging */ + +unsigned int stack_alloc_hits = 0; /* debugging */ unsigned int stack_alloc_misses = 0; /* debugging */ unsigned int stack_alloc_total = 0; unsigned int stack_alloc_hiwater = 0; +unsigned int stack_alloc_bndry = 0; + /* * The next field is at the base of the stack, @@ -241,7 +196,7 @@ unsigned int stack_alloc_hiwater = 0; /* * stack_alloc: * - * Allocate a kernel stack for an activation. + * Allocate a kernel stack for a thread. * May block. */ vm_offset_t @@ -249,14 +204,11 @@ stack_alloc( thread_t thread, void (*start_pos)(thread_t)) { - vm_offset_t stack; - spl_t s; + vm_offset_t stack = thread->kernel_stack; + spl_t s; - /* - * 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. - */ + if (stack) + return (stack); s = splsched(); stack_lock(); @@ -268,50 +220,22 @@ stack_alloc( stack_unlock(); splx(s); - if (stack == 0) { - /* - * Kernel stacks should be naturally aligned, - * so that it is easy to find the starting/ending - * addresses of a stack given an address in the middle. - */ - - if (kmem_alloc_aligned(kernel_map, &stack, - round_page(KERNEL_STACK_SIZE)) != KERN_SUCCESS) - panic("stack_alloc"); - - stack_alloc_total++; - if (stack_alloc_total > stack_alloc_hiwater) - stack_alloc_hiwater = stack_alloc_total; - -#if MACH_DEBUG - stack_init(stack, round_page(KERNEL_STACK_SIZE)); -#endif /* MACH_DEBUG */ - - /* - * If using fractional pages, free the remainder(s) - */ - if (KERNEL_STACK_SIZE < round_page(KERNEL_STACK_SIZE)) { - vm_offset_t ptr = stack + KERNEL_STACK_SIZE; - vm_offset_t endp = stack + round_page(KERNEL_STACK_SIZE); - while (ptr < endp) { -#if MACH_DEBUG - /* - * We need to initialize just the end of the - * region. - */ - stack_init(ptr, (unsigned int) (endp - ptr)); -#endif - 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(); - ptr += KERNEL_STACK_SIZE; - } - } + if (stack != 0) { + machine_stack_attach(thread, stack, start_pos); + return (stack); } - stack_attach(thread, stack, start_pos); + + 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; + + machine_stack_attach(thread, stack, start_pos); return (stack); } @@ -319,25 +243,61 @@ stack_alloc( * stack_free: * * Free a kernel stack. - * Called at splsched. */ void stack_free( thread_t thread) { - vm_offset_t stack = stack_detach(thread); + vm_offset_t stack = machine_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(); + if (stack != thread->reserved_stack) { + spl_t s = splsched(); + vm_offset_t *cache; + + cache = &stack_free_cache[cpu_number()]; + if (*cache == 0) { + *cache = stack; + splx(s); + + return; + } + + 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); } } +void +stack_free_stack( + vm_offset_t stack) +{ + spl_t s = splsched(); + vm_offset_t *cache; + + cache = &stack_free_cache[cpu_number()]; + if (*cache == 0) { + *cache = stack; + splx(s); + + return; + } + + 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); +} + /* * stack_collect: * @@ -348,37 +308,75 @@ stack_free( void stack_collect(void) { - register vm_offset_t stack; - spl_t s; - - /* If using fractional pages, Cannot just call kmem_free(), - * and we're too lazy to coalesce small chunks. - */ - if (KERNEL_STACK_SIZE < round_page(KERNEL_STACK_SIZE)) - return; + spl_t s = splsched(); - s = splsched(); stack_lock(); while (stack_free_count > stack_free_limit) { - stack = stack_free_list; + vm_offset_t stack = stack_free_list; + stack_free_list = stack_next(stack); stack_free_count--; stack_unlock(); splx(s); -#if MACH_DEBUG - stack_finalize(stack); -#endif /* MACH_DEBUG */ - kmem_free(kernel_map, stack, KERNEL_STACK_SIZE); + if (vm_map_remove( + stack_map, stack, stack + KERNEL_STACK_SIZE, + VM_MAP_REMOVE_KUNWIRE) != KERN_SUCCESS) + panic("stack_collect: vm_map_remove failed"); s = splsched(); - stack_alloc_total--; stack_lock(); + stack_alloc_total--; } stack_unlock(); splx(s); } +/* + * 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)(thread_t)) +{ + register vm_offset_t stack, *cache; + + cache = &stack_free_cache[cpu_number()]; + if (stack = *cache) { + *cache = 0; + machine_stack_attach(thread, stack, start); + stack_cache_hits++; + + return (TRUE); + } + + stack_lock(); + stack = stack_free_list; + if (stack != (vm_offset_t)0) { + stack_free_list = stack_next(stack); + stack_free_count--; + } + stack_unlock(); + + if (stack == 0) + stack = thread->reserved_stack; + + if (stack != 0) { + machine_stack_attach(thread, stack, start); + stack_alloc_hits++; + + return (TRUE); + } + else { + stack_alloc_misses++; + + return (FALSE); + } +} #if MACH_DEBUG /* @@ -398,27 +396,9 @@ stack_statistics( s = splsched(); stack_lock(); -#if STACK_USAGE - if (stack_check_usage) { - vm_offset_t stack; - - /* - * This is pretty expensive to do at splsched, - * but it only happens when someone makes - * a debugging call, so it should be OK. - */ - - for (stack = stack_free_list; stack != 0; - stack = stack_next(stack)) { - vm_size_t usage = stack_usage(stack); - - if (usage > *maxusagep) - *maxusagep = usage; - } - } -#endif /* STACK_USAGE */ - *totalp = stack_free_count; + *maxusagep = 0; + stack_unlock(); splx(s); } @@ -439,142 +419,112 @@ stack_fake_zone_info(int *count, vm_size_t *cur_size, vm_size_t *max_size, vm_si *exhaustable = 0; } - -/* - * stack_privilege: - * - * stack_alloc_try on this thread must always succeed. - */ - void stack_privilege( - register thread_t thread) -{ - /* - * This implementation only works for the current thread. - */ - - if (thread != current_thread()) - panic("stack_privilege"); - - if (thread->stack_privilege == 0) - thread->stack_privilege = current_stack(); -} - -/* - * 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)) + register thread_t thread) { - register vm_offset_t stack; - - if ((stack = thread->stack_privilege) == (vm_offset_t)0) { - stack_lock(); - stack = stack_free_list; - if (stack != (vm_offset_t)0) { - stack_free_list = stack_next(stack); - stack_free_count--; - } - stack_unlock(); - } - - if (stack != 0) { - stack_attach(thread, stack, start_pos); - stack_alloc_hits++; - return TRUE; - } else { - stack_alloc_misses++; - return FALSE; - } + /* OBSOLETE */ } -natural_t min_quantum_abstime; -extern natural_t min_quantum_ms; - void -thread_init(void) +thread_bootstrap(void) { - thread_shuttle_zone = zinit( - sizeof(struct thread_shuttle), - THREAD_MAX * sizeof(struct thread_shuttle), - THREAD_CHUNK * sizeof(struct thread_shuttle), - "threads"); - /* - * Fill in a template thread_shuttle for fast initialization. - * [Fields that must be (or are typically) reset at - * time of creation are so noted.] + * Fill in a template thread for fast initialization. */ - /* thr_sh_template.links (none) */ - thr_sh_template.runq = RUN_QUEUE_NULL; + thread_template.runq = RUN_QUEUE_NULL; + + thread_template.ref_count = 1; + + 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.interrupt_level = THREAD_ABORTSAFE; + thread_template.state = TH_STACK_HANDOFF | TH_WAIT | TH_UNINT; + thread_template.wake_active = FALSE; + thread_template.active_callout = FALSE; + thread_template.continuation = (void (*)(void))0; + thread_template.top_act = THR_ACT_NULL; + + thread_template.importance = 0; + thread_template.sched_mode = 0; + thread_template.safe_mode = 0; + 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; - /* thr_sh_template.task (later) */ - /* thr_sh_template.thread_list (later) */ - /* thr_sh_template.pset_threads (later) */ + thread_template.realtime.deadline = UINT64_MAX; - /* one ref for pset, one for activation */ - thr_sh_template.ref_count = 2; + thread_template.current_quantum = 0; - thr_sh_template.wait_event = NO_EVENT; - thr_sh_template.wait_result = KERN_SUCCESS; - thr_sh_template.wait_queue = WAIT_QUEUE_NULL; - thr_sh_template.wake_active = FALSE; - thr_sh_template.state = TH_WAIT|TH_UNINT; - thr_sh_template.interruptible = TRUE; - thr_sh_template.continuation = (void (*)(void))0; - thr_sh_template.top_act = THR_ACT_NULL; + thread_template.computation_metered = 0; + thread_template.computation_epoch = 0; - thr_sh_template.importance = 0; - thr_sh_template.sched_mode = 0; + thread_template.cpu_usage = 0; + thread_template.cpu_delta = 0; + thread_template.sched_usage = 0; + thread_template.sched_delta = 0; + thread_template.sched_stamp = 0; + thread_template.sleep_stamp = 0; + thread_template.safe_release = 0; - thr_sh_template.priority = 0; - thr_sh_template.sched_pri = 0; - thr_sh_template.depress_priority = -1; - thr_sh_template.max_priority = 0; + thread_template.bound_processor = PROCESSOR_NULL; + thread_template.last_processor = PROCESSOR_NULL; + thread_template.last_switch = 0; - thr_sh_template.cpu_usage = 0; - thr_sh_template.sched_usage = 0; - thr_sh_template.sched_stamp = 0; - thr_sh_template.sleep_stamp = 0; + thread_template.vm_privilege = FALSE; - thr_sh_template.policy = POLICY_NULL; - thr_sh_template.sp_state = 0; - thr_sh_template.unconsumed_quantum = 0; + timer_init(&(thread_template.user_timer)); + timer_init(&(thread_template.system_timer)); + thread_template.user_timer_save.low = 0; + thread_template.user_timer_save.high = 0; + thread_template.system_timer_save.low = 0; + thread_template.system_timer_save.high = 0; - thr_sh_template.vm_privilege = FALSE; + thread_template.processor_set = PROCESSOR_SET_NULL; - 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.cpu_delta = 0; - thr_sh_template.sched_delta = 0; + thread_template.act_ref_count = 2; - thr_sh_template.active = FALSE; /* reset */ + thread_template.special_handler.handler = special_handler; + thread_template.special_handler.next = 0; - /* thr_sh_template.processor_set (later) */ -#if NCPUS > 1 - thr_sh_template.bound_processor = PROCESSOR_NULL; -#endif /*NCPUS > 1*/ #if MACH_HOST - thr_sh_template.may_assign = TRUE; - thr_sh_template.assign_active = FALSE; + thread_template.may_assign = TRUE; + thread_template.assign_active = FALSE; #endif /* MACH_HOST */ - thr_sh_template.funnel_state = 0; + thread_template.funnel_lock = THR_FUNNEL_NULL; + thread_template.funnel_state = 0; +#if MACH_LDEBUG + thread_template.mutex_count = 0; +#endif /* MACH_LDEBUG */ -#if NCPUS > 1 - /* thr_sh_template.last_processor (later) */ -#endif /* NCPUS > 1 */ + init_thread = thread_template; + + init_thread.top_act = &init_thread; + init_thread.thread = &init_thread; + machine_thread_set_current(&init_thread); +} + +void +thread_init(void) +{ + kern_return_t ret; + unsigned int stack; + + thread_zone = zinit( + sizeof(struct thread), + THREAD_MAX * sizeof(struct thread), + THREAD_CHUNK * sizeof(struct thread), + "threads"); /* * Initialize other data structures used in @@ -583,61 +533,70 @@ thread_init(void) queue_init(&reaper_queue); simple_lock_init(&reaper_lock, ETAP_THREAD_REAPER); - thr_sh_template.funnel_lock = THR_FUNNEL_NULL; #ifndef MACHINE_STACK - simple_lock_init(&stack_lock_data, ETAP_THREAD_STACK); -#endif /* MACHINE_STACK */ - -#if MACH_DEBUG - simple_lock_init(&stack_usage_lock, ETAP_THREAD_STACK_USAGE); -#endif /* MACH_DEBUG */ - -#if MACH_LDEBUG - thr_sh_template.kthread = FALSE; - thr_sh_template.mutex_count = 0; -#endif /* MACH_LDEBUG */ - - { - AbsoluteTime abstime; - - clock_interval_to_absolutetime_interval( - min_quantum_ms, 1000*NSEC_PER_USEC, &abstime); - assert(abstime.hi == 0 && abstime.lo != 0); - min_quantum_abstime = abstime.lo; + 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 */ + + ret = kmem_suballoc(kernel_map, /* Suballocate from the kernel map */ + + &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 */ /* * Initialize any machine-dependent * per-thread structures necessary. */ - thread_machine_init(); + machine_thread_init(); } +/* + * Called at splsched. + */ void thread_reaper_enqueue( thread_t thread) { - /* - * thread lock is already held, splsched() - * not necessary here. - */ simple_lock(&reaper_lock); - enqueue_tail(&reaper_queue, (queue_entry_t)thread); -#if 0 /* CHECKME! */ - /* - * Since thread has been put in the reaper_queue, it must no longer - * be preempted (otherwise, it could be put back in a run queue). - */ - thread->preempt = TH_NOT_PREEMPTABLE; -#endif - simple_unlock(&reaper_lock); - thread_call_enter(thread_reaper_call); + thread_wakeup((event_t)&reaper_queue); } +void +thread_termination_continue(void) +{ + panic("thread_termination_continue"); + /*NOTREACHED*/ +} /* * Routine: thread_terminate_self @@ -659,367 +618,343 @@ thread_reaper_enqueue( void thread_terminate_self(void) { - register thread_t thread = current_thread(); - thread_act_t thr_act = thread->top_act; + thread_act_t thr_act = current_act(); + thread_t thread; task_t task = thr_act->task; - int active_acts; + 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); + /* This will allow no more control ops on this thr_act. */ + ipc_thr_act_disable(thr_act); + + /* 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); + + _mk_sp_thread_depress_abort(thread, TRUE); + /* * 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. */ - task_lock(task); - active_acts = --task->active_act_count; - task_unlock(task); - if (!active_acts && task->bsd_info) - proc_exit(task->bsd_info); - -#ifdef CALLOUT_RPC_MODEL - if (thr_act->lower) { - /* - * JMM - RPC will not be using a callout/stack manipulation - * mechanism. instead we will let it return normally as if - * from a continuation. Accordingly, these need to be cleaned - * up a bit. - */ - act_switch_swapcheck(thread, (ipc_port_t)0); - act_lock(thr_act); /* hierarchy violation XXX */ - (void) switch_act(THR_ACT_NULL); - assert(thr_act->ref_count == 1); /* XXX */ - /* act_deallocate(thr_act); XXX */ - prev_act = thread->top_act; - /* - * disable preemption to protect kernel stack changes - * disable_preemption(); - * MACH_RPC_RET(prev_act) = KERN_RPC_SERVER_TERMINATED; - * machine_kernel_stack_init(thread, mach_rpc_return_error); - */ - act_unlock(thr_act); - - /* - * Load_context(thread); - */ - /* NOTREACHED */ - } + active_acts = hw_atomic_sub(&task->active_thread_count, 1); -#else /* !CALLOUT_RPC_MODEL */ + if (active_acts == 0 && task->bsd_info) + proc_exit(task->bsd_info); + /* JMM - for now, no migration */ assert(!thr_act->lower); -#endif /* CALLOUT_RPC_MODEL */ - - s = splsched(); - thread_lock(thread); - thread->active = FALSE; - thread_unlock(thread); - splx(s); - thread_timer_terminate(); - /* flush any lazy HW state while in own context */ - thread_machine_flush(thr_act); - ipc_thread_terminate(thread); s = splsched(); thread_lock(thread); - thread->state |= (TH_HALTED|TH_TERMINATE); + thread->state |= TH_TERMINATE; assert((thread->state & TH_UNINT) == 0); -#if 0 /* CHECKME! */ - /* - * Since thread has been put in the reaper_queue, it must no longer - * be preempted (otherwise, it could be put back in a run queue). - */ - thread->preempt = TH_NOT_PREEMPTABLE; -#endif thread_mark_wait_locked(thread, THREAD_UNINT); + assert(thread->promotions == 0); thread_unlock(thread); /* splx(s); */ ETAP_SET_REASON(thread, BLOCKED_ON_TERMINATION); - thread_block((void (*)(void)) 0); - panic("the zombie walks!"); + thread_block(thread_termination_continue); /*NOTREACHED*/ } - /* * 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_t *out_thread) { - thread_t new_shuttle; - task_t parent_task = thr_act->task; + thread_t new_thread; processor_set_t pset; - kern_return_t result; - sched_policy_t *policy; - sf_return_t sfr; - int suspcnt; - - assert(!thr_act->thread); - assert(!thr_act->pool_port); + static thread_t first_thread; /* * Allocate a thread and initialize static fields */ - new_shuttle = (thread_t)zalloc(thread_shuttle_zone); - if (new_shuttle == THREAD_NULL) + if (first_thread == NULL) + new_thread = first_thread = current_act(); + else + new_thread = (thread_t)zalloc(thread_zone); + if (new_thread == NULL) return (KERN_RESOURCE_SHORTAGE); - *new_shuttle = thr_sh_template; + if (new_thread != first_thread) + *new_thread = thread_template; - thread_lock_init(new_shuttle); - rpc_lock_init(new_shuttle); - wake_lock_init(new_shuttle); - new_shuttle->sleep_stamp = sched_tick; +#ifdef MACH_BSD + { + extern void *uthread_alloc(task_t, thread_act_t); - pset = parent_task->processor_set; - if (!pset->active) { - pset = &default_pset; + new_thread->uthread = uthread_alloc(parent_task, new_thread); + if (new_thread->uthread == NULL) { + zfree(thread_zone, (vm_offset_t)new_thread); + return (KERN_RESOURCE_SHORTAGE); + } + } +#endif /* MACH_BSD */ + + if (machine_thread_create(new_thread, parent_task) != KERN_SUCCESS) { +#ifdef MACH_BSD + { + extern void uthread_free(task_t, void *, void *, void *); + void *ut = new_thread->uthread; + + new_thread->uthread = NULL; + uthread_free(parent_task, (void *)new_thread, ut, parent_task->bsd_info); + } +#endif /* MACH_BSD */ + zfree(thread_zone, (vm_offset_t)new_thread); + return (KERN_FAILURE); } + + new_thread->task = parent_task; + + thread_lock_init(new_thread); + wake_lock_init(new_thread); + + mutex_init(&new_thread->lock, ETAP_THREAD_ACT); + + ipc_thr_act_init(parent_task, new_thread); + + ipc_thread_init(new_thread); + queue_init(&new_thread->held_ulocks); + act_prof_init(new_thread, parent_task); + + new_thread->continuation = start; + new_thread->sleep_stamp = sched_tick; + + pset = parent_task->processor_set; + assert(pset == &default_pset); pset_lock(pset); 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->thread_count >= THREAD_MAX && + parent_task != kernel_task)) { task_unlock(parent_task); pset_unlock(pset); - zfree(thread_shuttle_zone, (vm_offset_t) new_shuttle); + +#ifdef MACH_BSD + { + extern void uthread_free(task_t, void *, void *, void *); + void *ut = new_thread->uthread; + + new_thread->uthread = NULL; + uthread_free(parent_task, (void *)new_thread, ut, parent_task->bsd_info); + } +#endif /* MACH_BSD */ + act_prof_deallocate(new_thread); + ipc_thr_act_terminate(new_thread); + machine_thread_destroy(new_thread); + zfree(thread_zone, (vm_offset_t) new_thread); return (KERN_FAILURE); } - act_attach(thr_act, new_shuttle, 0); + act_attach(new_thread, new_thread); + + task_reference_locked(parent_task); - /* 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++; - parent_task->active_act_count++; + /* Cache the task's map */ + new_thread->map = parent_task->map; - /* Associate the thread with that scheduling policy */ - new_shuttle->policy = parent_task->policy; - policy = &sched_policy[new_shuttle->policy]; - sfr = policy->sp_ops.sp_thread_attach(policy, new_shuttle); - if (sfr != SF_SUCCESS) - panic("thread_create_shuttle: sp_thread_attach"); + /* Chain the thread onto the task's list */ + queue_enter(&parent_task->threads, new_thread, thread_act_t, task_threads); + parent_task->thread_count++; + parent_task->res_thread_count++; + + /* So terminating threads don't need to take the task lock to decrement */ + hw_atomic_add(&parent_task->active_thread_count, 1); /* Associate the thread with the processor set */ - sfr = policy->sp_ops.sp_thread_processor_set(policy, new_shuttle, pset); - if (sfr != SF_SUCCESS) - panic("thread_create_shuttle: sp_thread_proceessor_set"); + pset_add_thread(pset, new_thread); + + thread_timer_setup(new_thread); /* Set the thread's scheduling parameters */ - new_shuttle->max_priority = parent_task->max_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; - sfr = policy->sp_ops.sp_thread_setup(policy, new_shuttle); - if (sfr != SF_SUCCESS) - panic("thread_create_shuttle: sp_thread_setup"); + if (parent_task != kernel_task) + 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; + compute_priority(new_thread, 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; - pset_unlock(pset); + new_thread->active = TRUE; + *out_thread = new_thread; - /* - * No need to lock thr_act, since it can't be known to anyone -- - * we set its suspend_count to one more than the task suspend_count - * by calling thread_hold. - */ - thr_act->user_stop_count = 1; - for (suspcnt = thr_act->task->suspend_count + 1; suspcnt; --suspcnt) - thread_hold(thr_act); - task_unlock(parent_task); - - /* - * 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); - - machine_kernel_stack_init(new_shuttle, thread_continue); - ipc_thread_init(new_shuttle); - thread_start(new_shuttle, start); - thread_timer_setup(new_shuttle); + { + long dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4; - *new_thread = new_shuttle; + kdbg_trace_data(parent_task->bsd_info, &dbg_arg2); - { - long dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4; + KERNEL_DEBUG_CONSTANT( + TRACEDBG_CODE(DBG_TRACE_DATA, 1) | DBG_FUNC_NONE, + (vm_address_t)new_thread, dbg_arg2, 0, 0, 0); - KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_DATA, 1)) | DBG_FUNC_NONE, - (vm_address_t)new_shuttle, 0,0,0,0); + kdbg_trace_string(parent_task->bsd_info, + &dbg_arg1, &dbg_arg2, &dbg_arg3, &dbg_arg4); - kdbg_trace_string(parent_task->bsd_info, &dbg_arg1, &dbg_arg2, &dbg_arg3, - &dbg_arg4); - KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_STRING, 1)) | DBG_FUNC_NONE, - dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, 0); + KERNEL_DEBUG_CONSTANT( + TRACEDBG_CODE(DBG_TRACE_STRING, 1) | DBG_FUNC_NONE, + dbg_arg1, dbg_arg2, dbg_arg3, dbg_arg4, 0); } return (KERN_SUCCESS); } +extern void thread_bootstrap_return(void); + kern_return_t thread_create( task_t task, - thread_act_t *new_act) + thread_act_t *new_thread) { - thread_act_t thr_act; - thread_t thread; kern_return_t result; - sched_policy_t *policy; - sf_return_t sfr; - spl_t s; - extern void thread_bootstrap_return(void); + thread_t thread; - if (task == TASK_NULL) - return KERN_INVALID_ARGUMENT; + if (task == TASK_NULL || task == kernel_task) + return (KERN_INVALID_ARGUMENT); - result = act_create(task, &thr_act); + result = thread_create_internal(task, -1, thread_bootstrap_return, &thread); if (result != KERN_SUCCESS) return (result); - result = thread_create_shuttle(thr_act, -1, thread_bootstrap_return, &thread); - if (result != KERN_SUCCESS) { - act_deallocate(thr_act); - return (result); - } - - if (task->kernel_loaded) - thread_user_to_kernel(thread); + thread->user_stop_count = 1; + thread_hold(thread); + if (task->suspend_count > 0) + thread_hold(thread); - /* Start the thread running (it will immediately suspend itself). */ - s = splsched(); - thread_ast_set(thr_act, AST_APC); - thread_lock(thread); - thread_go_locked(thread, THREAD_AWAKENED); - thread_unlock(thread); - splx(s); + pset_unlock(task->processor_set); + task_unlock(task); - *new_act = thr_act; + *new_thread = thread; return (KERN_SUCCESS); } -/* - * Update thread that belongs to a task created via kernel_task_create(). - */ -void -thread_user_to_kernel( - thread_t thread) -{ - /* - * Used to set special swap_func here... - */ -} - kern_return_t thread_create_running( - register task_t parent_task, + register task_t task, int flavor, thread_state_t new_state, mach_msg_type_number_t new_state_count, - thread_act_t *child_act) /* OUT */ + thread_act_t *new_thread) { register kern_return_t result; + thread_t thread; + + if (task == TASK_NULL || task == kernel_task) + return (KERN_INVALID_ARGUMENT); - result = thread_create(parent_task, child_act); + result = thread_create_internal(task, -1, thread_bootstrap_return, &thread); if (result != KERN_SUCCESS) return (result); - result = act_machine_set_state(*child_act, flavor, - new_state, new_state_count); + result = machine_thread_set_state(thread, flavor, new_state, new_state_count); if (result != KERN_SUCCESS) { - (void) thread_terminate(*child_act); - return (result); - } + pset_unlock(task->processor_set); + task_unlock(task); - result = thread_resume(*child_act); - if (result != KERN_SUCCESS) { - (void) thread_terminate(*child_act); + thread_terminate(thread); + act_deallocate(thread); return (result); } + act_lock(thread); + clear_wait(thread, THREAD_AWAKENED); + thread->started = TRUE; + act_unlock(thread); + pset_unlock(task->processor_set); + task_unlock(task); + + *new_thread = thread; + return (result); } /* * kernel_thread: * - * Create and kernel thread in the specified task, and - * optionally start it running. + * Create a thread in the kernel task + * to execute in kernel context. */ thread_t -kernel_thread_with_priority( - task_t task, - integer_t priority, +kernel_thread_create( void (*start)(void), - boolean_t start_running) + integer_t priority) { kern_return_t result; + task_t task = kernel_task; thread_t thread; - thread_act_t thr_act; - sched_policy_t *policy; - sf_return_t sfr; - spl_t s; - result = act_create(task, &thr_act); - if (result != KERN_SUCCESS) { - return THREAD_NULL; - } + result = thread_create_internal(task, priority, start, &thread); + if (result != KERN_SUCCESS) + return (THREAD_NULL); - result = thread_create_shuttle(thr_act, priority, start, &thread); - if (result != KERN_SUCCESS) { - act_deallocate(thr_act); - return THREAD_NULL; - } + pset_unlock(task->processor_set); + task_unlock(task); - thread_swappable(thr_act, FALSE); + thread_doswapin(thread); + assert(thread->kernel_stack != 0); + thread->reserved_stack = thread->kernel_stack; - s = splsched(); - thread_lock(thread); + act_deallocate(thread); - thr_act = thread->top_act; -#if MACH_LDEBUG - thread->kthread = TRUE; -#endif /* MACH_LDEBUG */ + return (thread); +} - if (start_running) - thread_go_locked(thread, THREAD_AWAKENED); +thread_t +kernel_thread_with_priority( + void (*start)(void), + integer_t priority) +{ + thread_t thread; - thread_unlock(thread); - splx(s); + thread = kernel_thread_create(start, priority); + if (thread == THREAD_NULL) + return (THREAD_NULL); - if (start_running) - thread_resume(thr_act); + act_lock(thread); + clear_wait(thread, THREAD_AWAKENED); + thread->started = TRUE; + act_unlock(thread); - act_deallocate(thr_act); +#ifdef i386 + thread_bind(thread, master_processor); +#endif /* i386 */ return (thread); } @@ -1028,117 +963,86 @@ kernel_thread( task_t task, void (*start)(void)) { - return kernel_thread_with_priority(task, -1, start, TRUE); + if (task != kernel_task) + panic("kernel_thread"); + + return kernel_thread_with_priority(start, -1); } 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 */ + void thread_deallocate( thread_t thread) { task_t task; processor_set_t pset; - sched_policy_t *policy; - sf_return_t sfr; + int refs; spl_t s; if (thread == THREAD_NULL) return; /* - * First, check for new count > 1 (the common case). + * First, check for new count > 0 (the common case). * Only the thread needs to be locked. */ s = splsched(); thread_lock(thread); - if (--thread->ref_count > 1) { - thread_unlock(thread); - splx(s); - return; - } - - /* - * Down to pset reference, lets try to clean up. - * However, the processor set may make more. Its lock - * also dominate the thread lock. So, reverse the - * order of the locks and see if its still the last - * reference; - */ - assert(thread->ref_count == 1); /* Else this is an extra dealloc! */ + refs = --thread->ref_count; thread_unlock(thread); splx(s); -#if MACH_HOST - thread_freeze(thread); -#endif /* MACH_HOST */ - - pset = thread->processor_set; - pset_lock(pset); - - s = splsched(); - thread_lock(thread); - - if (thread->ref_count > 1) { -#if MACH_HOST - boolean_t need_wakeup = FALSE; - /* - * processor_set made extra reference. - */ - /* Inline the unfreeze */ - thread->may_assign = TRUE; - if (thread->assign_active) { - need_wakeup = TRUE; - thread->assign_active = FALSE; - } -#endif /* MACH_HOST */ - thread_unlock(thread); - splx(s); - pset_unlock(pset); -#if MACH_HOST - if (need_wakeup) - thread_wakeup((event_t)&thread->assign_active); -#endif /* MACH_HOST */ - c_weird_pset_ref_exit++; + if (refs > 0) return; - } -#if MACH_HOST - assert(thread->assign_active == FALSE); -#endif /* MACH_HOST */ + + if (thread == current_thread()) + panic("thread_deallocate"); /* - * Thread only had pset reference - we can remove it. + * 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. */ - if (thread == current_thread()) - panic("thread deallocating itself"); - /* Detach thread (shuttle) from its sched policy */ - policy = &sched_policy[thread->policy]; - sfr = policy->sp_ops.sp_thread_detach(policy, thread); - if (sfr != SF_SUCCESS) - panic("thread_deallocate: sp_thread_detach"); +#if MACH_HOST + thread_freeze(thread); +#endif + pset = thread->processor_set; + pset_lock(pset); pset_remove_thread(pset, thread); - thread->ref_count = 0; - thread_unlock(thread); /* no more references - safe */ - splx(s); pset_unlock(pset); - pset_deallocate(thread->processor_set); +#if MACH_HOST + thread_unfreeze(thread); +#endif + + pset_deallocate(pset); - /* frees kernel stack & other MD resources */ - if (thread->stack_privilege && (thread->stack_privilege != thread->kernel_stack)) { - vm_offset_t stack; - int s = splsched(); - stack = thread->stack_privilege; - stack_free(thread); - thread->kernel_stack = stack; - splx(s); + if (thread->reserved_stack != 0) { + if (thread->reserved_stack != thread->kernel_stack) + stack_free_stack(thread->reserved_stack); + thread->reserved_stack = 0; } - thread->stack_privilege = 0; - thread_machine_destroy(thread); - zfree(thread_shuttle_zone, (vm_offset_t) thread); + if (thread->kernel_stack != 0) + stack_free(thread); + + machine_thread_destroy(thread); + + zfree(thread_zone, (vm_offset_t) thread); } void @@ -1152,7 +1056,7 @@ thread_reference( s = splsched(); thread_lock(thread); - thread->ref_count++; + thread_reference_locked(thread); thread_unlock(thread); splx(s); } @@ -1192,45 +1096,41 @@ thread_info_shuttle( thread_read_times(thread, &basic_info->user_time, &basic_info->system_time); - if (thread->policy & (POLICY_TIMESHARE|POLICY_RR|POLICY_FIFO)) { - /* - * Update lazy-evaluated scheduler info because someone wants it. - */ - if (thread->sched_stamp != sched_tick) - update_priority(thread); - - basic_info->sleep_time = 0; - - /* - * To calculate cpu_usage, first correct for timer rate, - * 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 = (basic_info->cpu_usage * 3) / 5; + /* + * Update lazy-evaluated scheduler info because someone wants it. + */ + if (thread->sched_stamp != sched_tick) + update_priority(thread); + + basic_info->sleep_time = 0; + + /* + * To calculate cpu_usage, first correct for timer rate, + * 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 = (basic_info->cpu_usage * 3) / 5; #if SIMPLE_CLOCK - /* - * Clock drift compensation. - */ - basic_info->cpu_usage = - (basic_info->cpu_usage * 1000000) / sched_usec; + /* + * Clock drift compensation. + */ + basic_info->cpu_usage = (basic_info->cpu_usage * 1000000) / sched_usec; #endif /* SIMPLE_CLOCK */ - } - else - basic_info->sleep_time = basic_info->cpu_usage = 0; - basic_info->policy = thread->policy; + basic_info->policy = ((thread->sched_mode & TH_MODE_TIMESHARE)? + POLICY_TIMESHARE: POLICY_RR); flags = 0; - if (thread->state & TH_SWAPPED_OUT) - flags = TH_FLAGS_SWAPPED; - else if (thread->state & TH_IDLE) - flags = TH_FLAGS_IDLE; + flags |= TH_FLAGS_IDLE; + + if (thread->state & TH_STACK_HANDOFF) + flags |= TH_FLAGS_SWAPPED; state = 0; - if (thread->state & TH_HALTED) + if (thread->state & TH_TERMINATE) state = TH_STATE_HALTED; else if (thread->state & TH_RUN) @@ -1269,19 +1169,25 @@ thread_info_shuttle( s = splsched(); thread_lock(thread); - if (thread->policy != POLICY_TIMESHARE) { + if (!(thread->sched_mode & TH_MODE_TIMESHARE)) { thread_unlock(thread); splx(s); return (KERN_INVALID_POLICY); } - ts_info->base_priority = thread->priority; - ts_info->max_priority = thread->max_priority; - ts_info->cur_priority = thread->sched_pri; + ts_info->depressed = (thread->sched_mode & TH_MODE_ISDEPRESSED) != 0; + if (ts_info->depressed) { + ts_info->base_priority = DEPRESSPRI; + ts_info->depress_priority = thread->priority; + } + else { + ts_info->base_priority = thread->priority; + ts_info->depress_priority = -1; + } - ts_info->depressed = (thread->depress_priority >= 0); - ts_info->depress_priority = thread->depress_priority; + ts_info->cur_priority = thread->sched_pri; + ts_info->max_priority = thread->max_priority; thread_unlock(thread); splx(s); @@ -1292,35 +1198,10 @@ thread_info_shuttle( } else if (flavor == THREAD_SCHED_FIFO_INFO) { - policy_fifo_info_t fifo_info; - if (*thread_info_count < POLICY_FIFO_INFO_COUNT) return (KERN_INVALID_ARGUMENT); - fifo_info = (policy_fifo_info_t)thread_info_out; - - s = splsched(); - thread_lock(thread); - - if (thread->policy != POLICY_FIFO) { - thread_unlock(thread); - splx(s); - - return (KERN_INVALID_POLICY); - } - - fifo_info->base_priority = thread->priority; - fifo_info->max_priority = thread->max_priority; - - fifo_info->depressed = (thread->depress_priority >= 0); - fifo_info->depress_priority = thread->depress_priority; - - thread_unlock(thread); - splx(s); - - *thread_info_count = POLICY_FIFO_INFO_COUNT; - - return (KERN_SUCCESS); + return (KERN_INVALID_POLICY); } else if (flavor == THREAD_SCHED_RR_INFO) { @@ -1334,19 +1215,25 @@ thread_info_shuttle( s = splsched(); thread_lock(thread); - if (thread->policy != POLICY_RR) { + if (thread->sched_mode & TH_MODE_TIMESHARE) { thread_unlock(thread); splx(s); return (KERN_INVALID_POLICY); } - rr_info->base_priority = thread->priority; - rr_info->max_priority = thread->max_priority; - rr_info->quantum = min_quantum_ms; + rr_info->depressed = (thread->sched_mode & TH_MODE_ISDEPRESSED) != 0; + if (rr_info->depressed) { + rr_info->base_priority = DEPRESSPRI; + rr_info->depress_priority = thread->priority; + } + else { + rr_info->base_priority = thread->priority; + rr_info->depress_priority = -1; + } - rr_info->depressed = (thread->depress_priority >= 0); - rr_info->depress_priority = thread->depress_priority; + rr_info->max_priority = thread->max_priority; + rr_info->quantum = std_quantum_us / 1000; thread_unlock(thread); splx(s); @@ -1364,14 +1251,12 @@ thread_doreap( register thread_t thread) { thread_act_t thr_act; - struct ipc_port *pool_port; thr_act = thread_lock_act(thread); assert(thr_act && thr_act->thread == thread); - act_locked_act_reference(thr_act); - pool_port = thr_act->pool_port; + act_reference_locked(thr_act); /* * Replace `act_unlock_thread()' with individual @@ -1379,70 +1264,58 @@ thread_doreap( * to determine which locks are held, confusing * `act_unlock_thread()'.) */ - rpc_unlock(thread); - if (pool_port != IP_NULL) - ip_unlock(pool_port); act_unlock(thr_act); /* Remove the reference held by a rooted thread */ - if (pool_port == IP_NULL) - act_deallocate(thr_act); + act_deallocate(thr_act); /* Remove the reference held by the thread: */ act_deallocate(thr_act); } -static thread_call_data_t thread_reaper_call_data; - /* * reaper_thread: * - * This kernel thread runs forever looking for threads to destroy - * (when they request that they be destroyed, of course). - * - * The reaper thread will disappear in the next revision of thread - * control when it's function will be moved into thread_dispatch. + * This kernel thread runs forever looking for terminating + * threads, releasing their "self" references. */ static void -_thread_reaper( - thread_call_param_t p0, - thread_call_param_t p1) +reaper_thread_continue(void) { register thread_t thread; - spl_t s; - s = splsched(); + (void)splsched(); simple_lock(&reaper_lock); while ((thread = (thread_t) dequeue_head(&reaper_queue)) != THREAD_NULL) { simple_unlock(&reaper_lock); - - /* - * wait for run bit to clear - */ - thread_lock(thread); - if (thread->state & TH_RUN) - panic("thread reaper: TH_RUN"); - thread_unlock(thread); - splx(s); + (void)spllo(); thread_doreap(thread); - s = splsched(); + (void)splsched(); simple_lock(&reaper_lock); } + assert_wait((event_t)&reaper_queue, THREAD_UNINT); simple_unlock(&reaper_lock); - splx(s); + (void)spllo(); + + thread_block(reaper_thread_continue); + /*NOTREACHED*/ } -void -thread_reaper(void) +static void +reaper_thread(void) { - thread_call_setup(&thread_reaper_call_data, _thread_reaper, NULL); - thread_reaper_call = &thread_reaper_call_data; + reaper_thread_continue(); + /*NOTREACHED*/ +} - _thread_reaper(NULL, NULL); +void +thread_reaper_init(void) +{ + kernel_thread_with_priority(reaper_thread, MINPRI_KERNEL); } kern_return_t @@ -1450,9 +1323,6 @@ thread_assign( thread_act_t thr_act, processor_set_t new_pset) { -#ifdef lint - thread++; new_pset++; -#endif /* lint */ return(KERN_FAILURE); } @@ -1495,16 +1365,17 @@ thread_get_assignment( } /* - * 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( +thread_wire_internal( host_priv_t host_priv, thread_act_t thr_act, - boolean_t wired) + boolean_t wired, + boolean_t *prev_state) { spl_t s; thread_t thread; @@ -1523,7 +1394,6 @@ thread_wire( /* * This implementation only works for the current thread. - * See stack_privilege. */ if (thr_act != current_act()) return KERN_INVALID_ARGUMENT; @@ -1531,6 +1401,10 @@ thread_wire( s = splsched(); thread_lock(thread); + if (prev_state) { + *prev_state = thread->vm_privilege; + } + if (wired) { if (thread->vm_privilege == FALSE) vm_page_free_reserve(1); /* XXX */ @@ -1545,116 +1419,25 @@ thread_wire( splx(s); act_unlock_thread(thr_act); - /* - * Make the thread unswappable. - */ - if (wired) - thread_swappable(thr_act, FALSE); - return KERN_SUCCESS; } -/* - * thread_collect_scan: - * - * Attempt to free resources owned by threads. - */ - -void -thread_collect_scan(void) -{ - /* This code runs very quickly! */ -} - -boolean_t thread_collect_allowed = TRUE; -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. - */ - -void -consider_thread_collect(void) -{ - /* - * By default, don't attempt thread collection more frequently - * than once a second (one scheduler tick). - */ - - if (thread_collect_max_rate == 0) - thread_collect_max_rate = 2; /* sched_tick is a 1 second resolution 2 here insures at least 1 second interval */ - - if (thread_collect_allowed && - (sched_tick > - (thread_collect_last_tick + thread_collect_max_rate))) { - thread_collect_last_tick = sched_tick; - thread_collect_scan(); - } -} - -#if MACH_DEBUG -#if STACK_USAGE - -vm_size_t -stack_usage( - register vm_offset_t stack) -{ - int i; - - for (i = 0; i < KERNEL_STACK_SIZE/sizeof(unsigned int); i++) - if (((unsigned int *)stack)[i] != STACK_MARKER) - break; - - return KERNEL_STACK_SIZE - i * sizeof(unsigned int); -} - -/* - * Machine-dependent code should call stack_init - * before doing its own initialization of the stack. - */ - -static void -stack_init( - register vm_offset_t stack, - unsigned int bytes) -{ - if (stack_check_usage) { - int i; - - for (i = 0; i < bytes / sizeof(unsigned int); i++) - ((unsigned int *)stack)[i] = STACK_MARKER; - } -} - -/* - * Machine-dependent code should call stack_finalize - * before releasing the stack memory. + * User-api wrapper for thread_wire_internal() */ +kern_return_t +thread_wire( + host_priv_t host_priv, + thread_act_t thr_act, + boolean_t wired) -void -stack_finalize( - register vm_offset_t stack) { - if (stack_check_usage) { - vm_size_t used = stack_usage(stack); - - simple_lock(&stack_usage_lock); - if (used > stack_max_usage) - stack_max_usage = used; - simple_unlock(&stack_usage_lock); - if (used > stack_max_use) { - printf("stack usage = %x\n", used); - panic("stack overflow"); - } - } + return thread_wire_internal(host_priv, thr_act, wired, NULL); } -#endif /*STACK_USAGE*/ -#endif /* MACH_DEBUG */ - kern_return_t host_stack_usage( host_t host, @@ -1674,15 +1457,13 @@ host_stack_usage( if (host == HOST_NULL) return KERN_INVALID_HOST; - simple_lock(&stack_usage_lock); - maxusage = stack_max_usage; - simple_unlock(&stack_usage_lock); + maxusage = 0; stack_statistics(&total, &maxusage); *reservedp = 0; *totalp = total; - *spacep = *residentp = total * round_page(KERNEL_STACK_SIZE); + *spacep = *residentp = total * round_page_32(KERNEL_STACK_SIZE); *maxusagep = maxusage; *maxstackp = 0; return KERN_SUCCESS; @@ -1718,6 +1499,8 @@ processor_set_stack_usage( vm_size_t size, size_needed; vm_offset_t addr; + spl_t s; + if (pset == PROCESSOR_SET_NULL) return KERN_INVALID_ARGUMENT; @@ -1753,16 +1536,20 @@ processor_set_stack_usage( } /* 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); - i < actual; - i++, + !queue_end(&pset->threads, (queue_entry_t) thread); thread = (thread_t) queue_next(&thread->pset_threads)) { - thread_reference(thread); - threads[i] = thread; + thread_lock(thread); + if (thread->ref_count > 0) { + thread_reference_locked(thread); + threads[i++] = thread; + } + thread_unlock(thread); } - assert(queue_end(&pset->threads, (queue_entry_t) thread)); + splx(s); + assert(i <= actual); /* can unlock processor set now that we have the thread refs */ pset_unlock(pset); @@ -1772,40 +1559,12 @@ processor_set_stack_usage( total = 0; maxusage = 0; maxstack = 0; - for (i = 0; i < actual; i++) { - 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. - */ + while (i > 0) { + thread_t thread = threads[--i]; - stack = thread->kernel_stack; - - for (cpu = 0; cpu < NCPUS; cpu++) - if (cpu_data[cpu].active_thread == thread) { - stack = active_stacks[cpu]; - break; - } - - if (stack != 0) { + if (thread->kernel_stack != 0) total++; - if (stack_check_usage) { - vm_size_t usage = stack_usage(stack); - - if (usage > maxusage) { - maxusage = usage; - maxstack = (vm_offset_t) thread; - } - } - } - thread_deallocate(thread); } @@ -1813,7 +1572,7 @@ processor_set_stack_usage( kfree(addr, size); *totalp = total; - *residentp = *spacep = total * round_page(KERNEL_STACK_SIZE); + *residentp = *spacep = total * round_page_32(KERNEL_STACK_SIZE); *maxusagep = maxusage; *maxstackp = maxstack; return KERN_SUCCESS; @@ -1821,7 +1580,7 @@ processor_set_stack_usage( #endif /* MACH_DEBUG */ } -static int split_funnel_off = 0; +int split_funnel_off = 0; funnel_t * funnel_alloc( int type) @@ -1829,9 +1588,9 @@ funnel_alloc( mutex_t *m; funnel_t * fnl; if ((fnl = (funnel_t *)kalloc(sizeof(funnel_t))) != 0){ - bzero(fnl, sizeof(funnel_t)); + bzero((void *)fnl, sizeof(funnel_t)); if ((m = mutex_alloc(0)) == (mutex_t *)NULL) { - kfree(fnl, sizeof(funnel_t)); + kfree((vm_offset_t)fnl, sizeof(funnel_t)); return(THR_FUNNEL_NULL); } fnl->fnl_mutex = m; @@ -1847,7 +1606,7 @@ funnel_free( mutex_free(fnl->fnl_mutex); if (fnl->fnl_oldmutex) mutex_free(fnl->fnl_oldmutex); - kfree(fnl, sizeof(funnel_t)); + kfree((vm_offset_t)fnl, sizeof(funnel_t)); } void @@ -1875,6 +1634,23 @@ funnel_unlock( fnl->fnl_mtxrelease = current_thread(); } +int refunnel_hint_enabled = 0; + +boolean_t +refunnel_hint( + thread_t thread, + wait_result_t wresult) +{ + if ( !(thread->funnel_state & TH_FN_REFUNNEL) || + wresult != THREAD_AWAKENED ) + return (FALSE); + + if (!refunnel_hint_enabled) + return (FALSE); + + return (mutex_preblock(thread->funnel_lock->fnl_mutex, thread)); +} + funnel_t * thread_funnel_get( void) @@ -1972,17 +1748,20 @@ thread_funnel_merge( } void -thread_set_cont_arg(int arg) +thread_set_cont_arg( + int arg) { - thread_t th = current_thread(); - th->cont_arg = arg; + thread_t self = current_thread(); + + self->saved.misc = arg; } int thread_get_cont_arg(void) { - thread_t th = current_thread(); - return(th->cont_arg); + thread_t self = current_thread(); + + return (self->saved.misc); } /* @@ -1992,8 +1771,17 @@ thread_get_cont_arg(void) #undef thread_should_halt boolean_t thread_should_halt( - thread_shuttle_t th) + thread_t th) { return(thread_should_halt_fast(th)); } +vm_offset_t min_valid_stack_address(void) +{ + return vm_map_min(stack_map); +} + +vm_offset_t max_valid_stack_address(void) +{ + return vm_map_max(stack_map); +}