]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/processor.c
xnu-4903.221.2.tar.gz
[apple/xnu.git] / osfmk / kern / processor.c
index ca65ceca63674b73db6d379aa313b2c54f25382e..479094c30569339ba261ba8dc984ec7e199d7c52 100644 (file)
 #include <ipc/ipc_port.h>
 #include <kern/kalloc.h>
 
+#include <security/mac_mach_internal.h>
+
+#if defined(CONFIG_XNUPOST)
+
+#include <tests/xnupost.h>
+
+#endif /* CONFIG_XNUPOST */
+
 /*
  * Exported interface
  */
@@ -89,10 +97,14 @@ struct pset_node            pset_node0;
 decl_simple_lock_data(static,pset_node_lock)
 
 queue_head_t                   tasks;
+queue_head_t                   terminated_tasks;       /* To be used ONLY for stackshot. */
+queue_head_t                   corpse_tasks;
 int                                            tasks_count;
+int                                            terminated_tasks_count;
 queue_head_t                   threads;
 int                                            threads_count;
 decl_lck_mtx_data(,tasks_threads_lock)
+decl_lck_mtx_data(,tasks_corpse_lock)
 
 processor_t                            processor_list;
 unsigned int                   processor_count;
@@ -103,13 +115,41 @@ uint32_t                          processor_avail_count;
 
 processor_t            master_processor;
 int                    master_cpu = 0;
+boolean_t              sched_stats_active = FALSE;
+
+processor_t            processor_array[MAX_SCHED_CPUS] = { 0 };
+
+#if defined(CONFIG_XNUPOST)
+kern_return_t ipi_test(void);
+extern void arm64_ipi_test(void);
+
+kern_return_t
+ipi_test()
+{
+#if __arm64__
+       processor_t p;
+
+       for (p = processor_list; p != NULL; p = p->processor_list) {
+               thread_bind(p);
+               thread_block(THREAD_CONTINUE_NULL);
+               kprintf("Running IPI test on cpu %d\n", p->cpu_id);
+               arm64_ipi_test();
+       }
+
+       /* unbind thread from specific cpu */
+       thread_bind(PROCESSOR_NULL);
+       thread_block(THREAD_CONTINUE_NULL);
+
+       T_PASS("Done running IPI tests");
+#else
+       T_PASS("Unsupported platform. Not running IPI tests");
+
+#endif /* __arm64__ */
+
+       return KERN_SUCCESS;
+}
+#endif /* defined(CONFIG_XNUPOST) */
 
-/* Forwards */
-kern_return_t  processor_set_things(
-               processor_set_t         pset,
-               mach_port_t             **thing_list,
-               mach_msg_type_number_t  *count,
-               int                     type);
 
 void
 processor_bootstrap(void)
@@ -120,7 +160,9 @@ processor_bootstrap(void)
        simple_lock_init(&pset_node_lock, 0);
 
        queue_init(&tasks);
+       queue_init(&terminated_tasks);
        queue_init(&threads);
+       queue_init(&corpse_tasks);
 
        simple_lock_init(&processor_list_lock, 0);
 
@@ -140,22 +182,38 @@ processor_init(
        int                                     cpu_id,
        processor_set_t         pset)
 {
-       run_queue_init(&processor->runq);
+       spl_t           s;
+
+       if (processor != master_processor) {
+               /* Scheduler state for master_processor initialized in sched_init() */
+               SCHED(processor_init)(processor);
+       }
+
+       assert(cpu_id < MAX_SCHED_CPUS);
 
        processor->state = PROCESSOR_OFF_LINE;
        processor->active_thread = processor->next_thread = processor->idle_thread = THREAD_NULL;
        processor->processor_set = pset;
-       processor->current_pri = MINPRI;
+       processor_state_update_idle(processor);
+       processor->starting_pri = MINPRI;
        processor->cpu_id = cpu_id;
        timer_call_setup(&processor->quantum_timer, thread_quantum_expire, processor);
+       processor->quantum_end = UINT64_MAX;
        processor->deadline = UINT64_MAX;
-       processor->timeslice = 0;
-       processor->processor_meta = PROCESSOR_META_NULL;
+       processor->first_timeslice = FALSE;
+       processor->processor_primary = processor; /* no SMT relationship known at this point */
+       processor->processor_secondary = NULL;
+       processor->is_SMT = FALSE;
+       processor->is_recommended = (pset->recommended_bitmask & (1ULL << cpu_id)) ? TRUE : FALSE;
        processor->processor_self = IP_NULL;
        processor_data_init(processor);
        processor->processor_list = NULL;
+       processor->cpu_quiesce_state = CPU_QUIESCE_COUNTER_NONE;
+       processor->cpu_quiesce_last_checkin = 0;
 
+       s = splsched();
        pset_lock(pset);
+       bit_set(pset->cpu_bitmask, cpu_id);
        if (pset->cpu_set_count++ == 0)
                pset->cpu_set_low = pset->cpu_set_hi = cpu_id;
        else {
@@ -163,6 +221,7 @@ processor_init(
                pset->cpu_set_hi = (cpu_id > pset->cpu_set_hi)? cpu_id: pset->cpu_set_hi;
        }
        pset_unlock(pset);
+       splx(s);
 
        simple_lock(&processor_list_lock);
        if (processor_list == NULL)
@@ -171,25 +230,34 @@ processor_init(
                processor_list_tail->processor_list = processor;
        processor_list_tail = processor;
        processor_count++;
+       processor_array[cpu_id] = processor;
        simple_unlock(&processor_list_lock);
 }
 
 void
-processor_meta_init(
+processor_set_primary(
        processor_t             processor,
        processor_t             primary)
 {
-       processor_meta_t        pmeta = primary->processor_meta;
-
-       if (pmeta == PROCESSOR_META_NULL) {
-               pmeta = kalloc(sizeof (*pmeta));
-
-               queue_init(&pmeta->idle_queue);
-
-               pmeta->primary = primary;
+       assert(processor->processor_primary == primary || processor->processor_primary == processor);
+       /* Re-adjust primary point for this (possibly) secondary processor */
+       processor->processor_primary = primary;
+
+       assert(primary->processor_secondary == NULL || primary->processor_secondary == processor);
+       if (primary != processor) {
+               /* Link primary to secondary, assumes a 2-way SMT model
+                * We'll need to move to a queue if any future architecture
+                * requires otherwise.
+                */
+               assert(processor->processor_secondary == NULL);
+               primary->processor_secondary = processor;
+               /* Mark both processors as SMT siblings */
+               primary->is_SMT = TRUE;
+               processor->is_SMT = TRUE;
+
+               processor_set_t pset = processor->processor_set;
+               atomic_bit_clear(&pset->primary_map, processor->cpu_id, memory_order_relaxed);
        }
-
-       processor->processor_meta = pmeta;
 }
 
 processor_set_t
@@ -199,6 +267,34 @@ processor_pset(
        return (processor->processor_set);
 }
 
+void
+processor_state_update_idle(processor_t processor)
+{
+    processor->current_pri = IDLEPRI;
+    processor->current_sfi_class = SFI_CLASS_KERNEL;
+    processor->current_recommended_pset_type = PSET_SMP;
+    processor->current_perfctl_class = PERFCONTROL_CLASS_IDLE;
+}
+
+void
+processor_state_update_from_thread(processor_t processor, thread_t thread)
+{
+    processor->current_pri = thread->sched_pri;
+    processor->current_sfi_class = thread->sfi_class;
+    processor->current_recommended_pset_type = recommended_pset_type(thread);
+    processor->current_perfctl_class = thread_get_perfcontrol_class(thread);
+}
+
+void
+processor_state_update_explicit(processor_t processor, int pri, sfi_class_id_t sfi_class, 
+       pset_cluster_type_t pset_type, perfcontrol_class_t perfctl_class)
+{
+    processor->current_pri = pri;
+    processor->current_sfi_class = sfi_class;
+    processor->current_recommended_pset_type = pset_type;
+    processor->current_perfctl_class = perfctl_class;
+}
+
 pset_node_t
 pset_node_root(void)
 {
@@ -209,6 +305,10 @@ processor_set_t
 pset_create(
        pset_node_t                     node)
 {
+       /* some schedulers do not support multiple psets */
+       if (SCHED(multiple_psets_enabled) == FALSE)
+               return processor_pset(master_processor);
+
        processor_set_t         *prev, pset = kalloc(sizeof (*pset));
 
        if (pset != PROCESSOR_SET_NULL) {
@@ -228,6 +328,33 @@ pset_create(
        return (pset);
 }
 
+/*
+ *     Find processor set in specified node with specified cluster_id.
+ *     Returns default_pset if not found.
+ */
+processor_set_t
+pset_find(
+       uint32_t cluster_id,
+       processor_set_t default_pset)
+{
+       simple_lock(&pset_node_lock);
+       pset_node_t node = &pset_node0;
+       processor_set_t pset = NULL;
+
+       do {
+               pset = node->psets;
+               while (pset != NULL) {
+                       if (pset->pset_cluster_id == cluster_id)
+                               break;
+                       pset = pset->pset_list;
+               }
+       } while ((node = node->node_list) != NULL);
+       simple_unlock(&pset_node_lock);
+       if (pset == NULL)
+               return default_pset;
+       return (pset);
+}
+
 /*
  *     Initialize the given processor_set structure.
  */
@@ -236,17 +363,36 @@ pset_init(
        processor_set_t         pset,
        pset_node_t                     node)
 {
-       queue_init(&pset->active_queue);
-       queue_init(&pset->idle_queue);
-       pset->processor_count = 0;
-       pset->low_pri = pset->low_count = PROCESSOR_NULL;
+       if (pset != &pset0) {
+               /* Scheduler state for pset0 initialized in sched_init() */
+               SCHED(pset_init)(pset);
+               SCHED(rt_init)(pset);
+       }
+
+       pset->online_processor_count = 0;
+       pset->load_average = 0;
        pset->cpu_set_low = pset->cpu_set_hi = 0;
        pset->cpu_set_count = 0;
+       pset->last_chosen = -1;
+       pset->cpu_bitmask = 0;
+       pset->recommended_bitmask = ~0ULL;
+       pset->primary_map = ~0ULL;
+       pset->cpu_state_map[PROCESSOR_OFF_LINE] = ~0ULL;
+       for (uint i = PROCESSOR_SHUTDOWN; i < PROCESSOR_STATE_LEN; i++) {
+               pset->cpu_state_map[i] = 0;
+       }
+       pset->pending_AST_cpu_mask = 0;
+#if defined(CONFIG_SCHED_DEFERRED_AST)
+       pset->pending_deferred_AST_cpu_mask = 0;
+#endif
+       pset->pending_spill_cpu_mask = 0;
        pset_lock_init(pset);
        pset->pset_self = IP_NULL;
        pset->pset_name_self = IP_NULL;
        pset->pset_list = PROCESSOR_SET_NULL;
        pset->node = node;
+       pset->pset_cluster_type = PSET_SMP;
+       pset->pset_cluster_id = 0;
 }
 
 kern_return_t
@@ -274,13 +420,13 @@ processor_info_count(
 
 kern_return_t
 processor_info(
-       register processor_t    processor,
+       processor_t     processor,
        processor_flavor_t              flavor,
        host_t                                  *host,
        processor_info_t                info,
        mach_msg_type_number_t  *count)
 {
-       register int    cpu_id, state;
+       int     cpu_id, state;
        kern_return_t   result;
 
        if (processor == PROCESSOR_NULL)
@@ -292,7 +438,7 @@ processor_info(
 
        case PROCESSOR_BASIC_INFO:
        {
-               register processor_basic_info_t         basic_info;
+               processor_basic_info_t          basic_info;
 
                if (*count < PROCESSOR_BASIC_INFO_COUNT)
                        return (KERN_FAILURE);
@@ -319,18 +465,71 @@ processor_info(
 
        case PROCESSOR_CPU_LOAD_INFO:
        {
-               register processor_cpu_load_info_t      cpu_load_info;
-
-           if (*count < PROCESSOR_CPU_LOAD_INFO_COUNT)
+               processor_cpu_load_info_t       cpu_load_info;
+               timer_t         idle_state;
+               uint64_t        idle_time_snapshot1, idle_time_snapshot2;
+               uint64_t        idle_time_tstamp1, idle_time_tstamp2;
+
+               /*
+                * We capture the accumulated idle time twice over
+                * the course of this function, as well as the timestamps
+                * when each were last updated. Since these are
+                * all done using non-atomic racy mechanisms, the
+                * most we can infer is whether values are stable.
+                * timer_grab() is the only function that can be
+                * used reliably on another processor's per-processor
+                * data.
+                */
+
+               if (*count < PROCESSOR_CPU_LOAD_INFO_COUNT)
                        return (KERN_FAILURE);
 
-           cpu_load_info = (processor_cpu_load_info_t) info;
-               cpu_load_info->cpu_ticks[CPU_STATE_USER] =
+               cpu_load_info = (processor_cpu_load_info_t) info;
+               if (precise_user_kernel_time) {
+                       cpu_load_info->cpu_ticks[CPU_STATE_USER] =
                                                        (uint32_t)(timer_grab(&PROCESSOR_DATA(processor, user_state)) / hz_tick_interval);
-               cpu_load_info->cpu_ticks[CPU_STATE_SYSTEM] =
+                       cpu_load_info->cpu_ticks[CPU_STATE_SYSTEM] =
                                                        (uint32_t)(timer_grab(&PROCESSOR_DATA(processor, system_state)) / hz_tick_interval);
-               cpu_load_info->cpu_ticks[CPU_STATE_IDLE] =
-                                                       (uint32_t)(timer_grab(&PROCESSOR_DATA(processor, idle_state)) / hz_tick_interval);
+               } else {
+                       uint64_t tval = timer_grab(&PROCESSOR_DATA(processor, user_state)) +
+                               timer_grab(&PROCESSOR_DATA(processor, system_state));
+
+                       cpu_load_info->cpu_ticks[CPU_STATE_USER] = (uint32_t)(tval / hz_tick_interval);
+                       cpu_load_info->cpu_ticks[CPU_STATE_SYSTEM] = 0;
+               }
+
+               idle_state = &PROCESSOR_DATA(processor, idle_state);
+               idle_time_snapshot1 = timer_grab(idle_state);
+               idle_time_tstamp1 = idle_state->tstamp;
+
+               /*
+                * Idle processors are not continually updating their
+                * per-processor idle timer, so it may be extremely
+                * out of date, resulting in an over-representation
+                * of non-idle time between two measurement
+                * intervals by e.g. top(1). If we are non-idle, or
+                * have evidence that the timer is being updated
+                * concurrently, we consider its value up-to-date.
+                */
+               if (PROCESSOR_DATA(processor, current_state) != idle_state) {
+                       cpu_load_info->cpu_ticks[CPU_STATE_IDLE] =
+                                                       (uint32_t)(idle_time_snapshot1 / hz_tick_interval);
+               } else if ((idle_time_snapshot1 != (idle_time_snapshot2 = timer_grab(idle_state))) ||
+                                  (idle_time_tstamp1 != (idle_time_tstamp2 = idle_state->tstamp))){
+                       /* Idle timer is being updated concurrently, second stamp is good enough */
+                       cpu_load_info->cpu_ticks[CPU_STATE_IDLE] =
+                                                       (uint32_t)(idle_time_snapshot2 / hz_tick_interval);
+               } else {
+                       /*
+                        * Idle timer may be very stale. Fortunately we have established
+                        * that idle_time_snapshot1 and idle_time_tstamp1 are unchanging
+                        */
+                       idle_time_snapshot1 += mach_absolute_time() - idle_time_tstamp1;
+                               
+                       cpu_load_info->cpu_ticks[CPU_STATE_IDLE] =
+                               (uint32_t)(idle_time_snapshot1 / hz_tick_interval);
+               }
+
                cpu_load_info->cpu_ticks[CPU_STATE_NICE] = 0;
 
            *count = PROCESSOR_CPU_LOAD_INFO_COUNT;
@@ -383,7 +582,7 @@ processor_start(
                return (KERN_FAILURE);
        }
 
-       processor->state = PROCESSOR_START;
+       pset_update_processor_state(pset, processor, PROCESSOR_START);
        pset_unlock(pset);
        splx(s);
 
@@ -395,7 +594,7 @@ processor_start(
                if (result != KERN_SUCCESS) {
                        s = splsched();
                        pset_lock(pset);
-                       processor->state = PROCESSOR_OFF_LINE;
+                       pset_update_processor_state(pset, processor, PROCESSOR_OFF_LINE);
                        pset_unlock(pset);
                        splx(s);
 
@@ -414,7 +613,7 @@ processor_start(
                if (result != KERN_SUCCESS) {
                        s = splsched();
                        pset_lock(pset);
-                       processor->state = PROCESSOR_OFF_LINE;
+                       pset_update_processor_state(pset, processor, PROCESSOR_OFF_LINE);
                        pset_unlock(pset);
                        splx(s);
 
@@ -426,6 +625,7 @@ processor_start(
                thread->bound_processor = processor;
                processor->next_thread = thread;
                thread->state = TH_RUN;
+               thread->last_made_runnable_time = mach_absolute_time();
                thread_unlock(thread);
                splx(s);
 
@@ -439,7 +639,7 @@ processor_start(
        if (result != KERN_SUCCESS) {
                s = splsched();
                pset_lock(pset);
-               processor->state = PROCESSOR_OFF_LINE;
+               pset_update_processor_state(pset, processor, PROCESSOR_OFF_LINE);
                pset_unlock(pset);
                splx(s);
 
@@ -496,6 +696,9 @@ processor_get_assignment(
 {
        int state;
 
+       if (processor == PROCESSOR_NULL)
+               return(KERN_INVALID_ARGUMENT);
+
        state = processor->state;
        if (state == PROCESSOR_SHUTDOWN || state == PROCESSOR_OFF_LINE)
                return(KERN_FAILURE);
@@ -517,7 +720,7 @@ processor_set_info(
                return(KERN_INVALID_ARGUMENT);
 
        if (flavor == PROCESSOR_SET_BASIC_INFO) {
-               register processor_set_basic_info_t     basic_info;
+               processor_set_basic_info_t      basic_info;
 
                if (*count < PROCESSOR_SET_BASIC_INFO_COUNT)
                        return(KERN_FAILURE);
@@ -531,7 +734,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_TIMESHARE_DEFAULT) {
-               register policy_timeshare_base_t        ts_base;
+               policy_timeshare_base_t ts_base;
 
                if (*count < POLICY_TIMESHARE_BASE_COUNT)
                        return(KERN_FAILURE);
@@ -544,7 +747,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_FIFO_DEFAULT) {
-               register policy_fifo_base_t             fifo_base;
+               policy_fifo_base_t              fifo_base;
 
                if (*count < POLICY_FIFO_BASE_COUNT)
                        return(KERN_FAILURE);
@@ -557,7 +760,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_RR_DEFAULT) {
-               register policy_rr_base_t               rr_base;
+               policy_rr_base_t                rr_base;
 
                if (*count < POLICY_RR_BASE_COUNT)
                        return(KERN_FAILURE);
@@ -571,7 +774,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_TIMESHARE_LIMITS) {
-               register policy_timeshare_limit_t       ts_limit;
+               policy_timeshare_limit_t        ts_limit;
 
                if (*count < POLICY_TIMESHARE_LIMIT_COUNT)
                        return(KERN_FAILURE);
@@ -584,7 +787,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_FIFO_LIMITS) {
-               register policy_fifo_limit_t            fifo_limit;
+               policy_fifo_limit_t             fifo_limit;
 
                if (*count < POLICY_FIFO_LIMIT_COUNT)
                        return(KERN_FAILURE);
@@ -597,7 +800,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_RR_LIMITS) {
-               register policy_rr_limit_t              rr_limit;
+               policy_rr_limit_t               rr_limit;
 
                if (*count < POLICY_RR_LIMIT_COUNT)
                        return(KERN_FAILURE);
@@ -610,7 +813,7 @@ processor_set_info(
                return(KERN_SUCCESS);
        }
        else if (flavor == PROCESSOR_SET_ENABLED_POLICIES) {
-               register int                            *enabled;
+               int                             *enabled;
 
                if (*count < (sizeof(*enabled)/sizeof(int)))
                        return(KERN_FAILURE);
@@ -644,7 +847,7 @@ processor_set_statistics(
                return (KERN_INVALID_PROCESSOR_SET);
 
        if (flavor == PROCESSOR_SET_LOAD_INFO) {
-               register processor_set_load_info_t     load_info;
+               processor_set_load_info_t     load_info;
 
                if (*count < PROCESSOR_SET_LOAD_INFO_COUNT)
                        return(KERN_FAILURE);
@@ -709,9 +912,6 @@ processor_set_policy_disable(
        return (KERN_INVALID_ARGUMENT);
 }
 
-#define THING_TASK     0
-#define THING_THREAD   1
-
 /*
  *     processor_set_things:
  *
@@ -719,167 +919,227 @@ processor_set_policy_disable(
  */
 kern_return_t
 processor_set_things(
-       processor_set_t                 pset,
-       mach_port_t                             **thing_list,
-       mach_msg_type_number_t  *count,
-       int                                             type)
+       processor_set_t pset,
+       void **thing_list,
+       mach_msg_type_number_t *count,
+       int type)
 {
-       unsigned int actual;    /* this many things */
-       unsigned int maxthings;
        unsigned int i;
+       task_t task;
+       thread_t thread;
 
+       task_t *task_list;
+       unsigned int actual_tasks;
+       vm_size_t task_size, task_size_needed;
+
+       thread_t *thread_list;
+       unsigned int actual_threads;
+       vm_size_t thread_size, thread_size_needed;
+
+       void *addr, *newaddr;
        vm_size_t size, size_needed;
-       void  *addr;
 
        if (pset == PROCESSOR_SET_NULL || pset != &pset0)
                return (KERN_INVALID_ARGUMENT);
 
-       size = 0;
-       addr = NULL;
+       task_size = 0;
+       task_size_needed = 0;
+       task_list = NULL;
+       actual_tasks = 0;
+
+       thread_size = 0;
+       thread_size_needed = 0;
+       thread_list = NULL;
+       actual_threads = 0;
 
        for (;;) {
                lck_mtx_lock(&tasks_threads_lock);
 
-               if (type == THING_TASK)
-                       maxthings = tasks_count;
-               else
-                       maxthings = threads_count;
-
                /* do we have the memory we need? */
+               if (type == PSET_THING_THREAD)
+                       thread_size_needed = threads_count * sizeof(void *);
+#if !CONFIG_MACF
+               else
+#endif
+                       task_size_needed = tasks_count * sizeof(void *);
 
-               size_needed = maxthings * sizeof (mach_port_t);
-               if (size_needed <= size)
+               if (task_size_needed <= task_size &&
+                   thread_size_needed <= thread_size)
                        break;
 
                /* unlock and allocate more memory */
                lck_mtx_unlock(&tasks_threads_lock);
 
-               if (size != 0)
-                       kfree(addr, size);
+               /* grow task array */
+               if (task_size_needed > task_size) {
+                       if (task_size != 0)
+                               kfree(task_list, task_size);
 
-               assert(size_needed > 0);
-               size = size_needed;
+                       assert(task_size_needed > 0);
+                       task_size = task_size_needed;
 
-               addr = kalloc(size);
-               if (addr == 0)
-                       return (KERN_RESOURCE_SHORTAGE);
-       }
+                       task_list = (task_t *)kalloc(task_size);
+                       if (task_list == NULL) {
+                               if (thread_size != 0)
+                                       kfree(thread_list, thread_size);
+                               return (KERN_RESOURCE_SHORTAGE);
+                       }
+               }
 
-       /* OK, have memory and the list locked */
+               /* grow thread array */
+               if (thread_size_needed > thread_size) {
+                       if (thread_size != 0)
+                               kfree(thread_list, thread_size);
+
+                       assert(thread_size_needed > 0);
+                       thread_size = thread_size_needed;
 
-       actual = 0;
-       switch (type) {
+                       thread_list = (thread_t *)kalloc(thread_size);
+                       if (thread_list == 0) {
+                               if (task_size != 0)
+                                       kfree(task_list, task_size);
+                               return (KERN_RESOURCE_SHORTAGE);
+                       }
+               }
+       }
 
-       case THING_TASK: {
-               task_t          task, *task_list = (task_t *)addr;
+       /* OK, have memory and the list locked */
 
+       /* If we need it, get the thread list */
+       if (type == PSET_THING_THREAD) {
+               for (thread = (thread_t)queue_first(&threads);
+                    !queue_end(&threads, (queue_entry_t)thread);
+                    thread = (thread_t)queue_next(&thread->threads)) {
+#if defined(SECURE_KERNEL)
+                       if (thread->task != kernel_task) {
+#endif
+                               thread_reference_internal(thread);
+                               thread_list[actual_threads++] = thread;
+#if defined(SECURE_KERNEL)
+                       }
+#endif
+               }
+       }
+#if !CONFIG_MACF
+         else {
+#endif
+               /* get a list of the tasks */
                for (task = (task_t)queue_first(&tasks);
-                                               !queue_end(&tasks, (queue_entry_t)task);
-                                                               task = (task_t)queue_next(&task->tasks)) {
+                    !queue_end(&tasks, (queue_entry_t)task);
+                    task = (task_t)queue_next(&task->tasks)) {
 #if defined(SECURE_KERNEL)
                        if (task != kernel_task) {
 #endif
                                task_reference_internal(task);
-                               task_list[actual++] = task;
+                               task_list[actual_tasks++] = task;
 #if defined(SECURE_KERNEL)
                        }
 #endif
                }
-
-               break;
-       }
-
-       case THING_THREAD: {
-               thread_t        thread, *thread_list = (thread_t *)addr;
-
-               for (thread = (thread_t)queue_first(&threads);
-                                               !queue_end(&threads, (queue_entry_t)thread);
-                                                               thread = (thread_t)queue_next(&thread->threads)) {
-                       thread_reference_internal(thread);
-                       thread_list[actual++] = thread;
-               }
-
-               break;
+#if !CONFIG_MACF
        }
+#endif
 
-       }
-               
        lck_mtx_unlock(&tasks_threads_lock);
 
-       if (actual < maxthings)
-               size_needed = actual * sizeof (mach_port_t);
-
-       if (actual == 0) {
-               /* no things, so return null pointer and deallocate memory */
-               *thing_list = NULL;
-               *count = 0;
+#if CONFIG_MACF
+       unsigned int j, used;
 
-               if (size != 0)
-                       kfree(addr, size);
+       /* for each task, make sure we are allowed to examine it */
+       for (i = used = 0; i < actual_tasks; i++) {
+               if (mac_task_check_expose_task(task_list[i])) {
+                       task_deallocate(task_list[i]);
+                       continue;
+               }
+               task_list[used++] = task_list[i];
        }
-       else {
-               /* if we allocated too much, must copy */
+       actual_tasks = used;
+       task_size_needed = actual_tasks * sizeof(void *);
 
-               if (size_needed < size) {
-                       void *newaddr;
+       if (type == PSET_THING_THREAD) {
 
-                       newaddr = kalloc(size_needed);
-                       if (newaddr == 0) {
-                               switch (type) {
+               /* for each thread (if any), make sure it's task is in the allowed list */
+               for (i = used = 0; i < actual_threads; i++) {
+                       boolean_t found_task = FALSE;
 
-                               case THING_TASK: {
-                                       task_t          *task_list = (task_t *)addr;
-
-                                       for (i = 0; i < actual; i++)
-                                               task_deallocate(task_list[i]);
-                                       break;
-                               }
-
-                               case THING_THREAD: {
-                                       thread_t        *thread_list = (thread_t *)addr;
-
-                                       for (i = 0; i < actual; i++)
-                                               thread_deallocate(thread_list[i]);
+                       task = thread_list[i]->task;
+                       for (j = 0; j < actual_tasks; j++) {
+                               if (task_list[j] == task) {
+                                       found_task = TRUE;
                                        break;
                                }
-
-                               }
-
-                               kfree(addr, size);
-                               return (KERN_RESOURCE_SHORTAGE);
                        }
-
-                       bcopy((void *) addr, (void *) newaddr, size_needed);
-                       kfree(addr, size);
-                       addr = newaddr;
+                       if (found_task)
+                               thread_list[used++] = thread_list[i];
+                       else
+                               thread_deallocate(thread_list[i]);
                }
+               actual_threads = used;
+               thread_size_needed = actual_threads * sizeof(void *);
+
+               /* done with the task list */
+               for (i = 0; i < actual_tasks; i++)
+                       task_deallocate(task_list[i]);
+               kfree(task_list, task_size);
+               task_size = 0;
+               actual_tasks = 0;
+               task_list = NULL;
+       }
+#endif
 
-               *thing_list = (mach_port_t *)addr;
-               *count = actual;
-
-               /* do the conversion that Mig should handle */
-
-               switch (type) {
-
-               case THING_TASK: {
-                       task_t          *task_list = (task_t *)addr;
-
-                       for (i = 0; i < actual; i++)
-                               (*thing_list)[i] = convert_task_to_port(task_list[i]);
-                       break;
+       if (type == PSET_THING_THREAD) {
+               if (actual_threads == 0) {
+                       /* no threads available to return */
+                       assert(task_size == 0);
+                       if (thread_size != 0)
+                               kfree(thread_list, thread_size);
+                       *thing_list = NULL;
+                       *count = 0;
+                       return KERN_SUCCESS;
                }
+               size_needed = actual_threads * sizeof(void *);
+               size = thread_size;
+               addr = thread_list;
+       } else {
+               if (actual_tasks == 0) {
+                       /* no tasks available to return */
+                       assert(thread_size == 0);
+                       if (task_size != 0)
+                               kfree(task_list, task_size);
+                       *thing_list = NULL;
+                       *count = 0;
+                       return KERN_SUCCESS;
+               } 
+               size_needed = actual_tasks * sizeof(void *);
+               size = task_size;
+               addr = task_list;
+       }
 
-               case THING_THREAD: {
-                       thread_t        *thread_list = (thread_t *)addr;
-
-                       for (i = 0; i < actual; i++)
-                               (*thing_list)[i] = convert_thread_to_port(thread_list[i]);
-                       break;
+       /* if we allocated too much, must copy */
+       if (size_needed < size) {
+               newaddr = kalloc(size_needed);
+               if (newaddr == 0) {
+                       for (i = 0; i < actual_tasks; i++) {
+                               if (type == PSET_THING_THREAD)
+                                       thread_deallocate(thread_list[i]);
+                               else
+                                       task_deallocate(task_list[i]);
+                       }
+                       if (size)
+                               kfree(addr, size);
+                       return (KERN_RESOURCE_SHORTAGE);
                }
 
-               }
+               bcopy((void *) addr, (void *) newaddr, size_needed);
+               kfree(addr, size);
+
+               addr = newaddr;
+               size = size_needed;
        }
 
+       *thing_list = (void **)addr;
+       *count = (unsigned int)size / sizeof(void *);
+
        return (KERN_SUCCESS);
 }
 
@@ -895,7 +1155,17 @@ processor_set_tasks(
        task_array_t            *task_list,
        mach_msg_type_number_t  *count)
 {
-    return(processor_set_things(pset, (mach_port_t **)task_list, count, THING_TASK));
+       kern_return_t ret;
+       mach_msg_type_number_t i;
+
+       ret = processor_set_things(pset, (void **)task_list, count, PSET_THING_TASK);
+       if (ret != KERN_SUCCESS)
+               return ret;
+
+       /* do the conversion that Mig should handle */
+       for (i = 0; i < *count; i++)
+               (*task_list)[i] = (task_t)convert_task_to_port((*task_list)[i]);
+       return KERN_SUCCESS;
 }
 
 /*
@@ -928,7 +1198,17 @@ processor_set_threads(
        thread_array_t          *thread_list,
        mach_msg_type_number_t  *count)
 {
-    return(processor_set_things(pset, (mach_port_t **)thread_list, count, THING_THREAD));
+       kern_return_t ret;
+       mach_msg_type_number_t i;
+
+       ret = processor_set_things(pset, (void **)thread_list, count, PSET_THING_THREAD);
+       if (ret != KERN_SUCCESS)
+               return ret;
+
+       /* do the conversion that Mig should handle */
+       for (i = 0; i < *count; i++)
+               (*thread_list)[i] = (thread_t)convert_thread_to_port((*thread_list)[i]);
+       return KERN_SUCCESS;
 }
 #endif
 
@@ -967,3 +1247,10 @@ __unused processor_set_t  pset)
 {
        return;
 }
+
+pset_cluster_type_t
+recommended_pset_type(thread_t thread)
+{
+       (void)thread;
+       return PSET_SMP;
+}