xnu-2422.90.20.tar.gz

[apple/xnu.git] / osfmk / kern / processor.c

diff --git a/osfmk/kern/processor.c b/osfmk/kern/processor.c
index b0771351f96e6b18374d70f3bba7e44aa4b49ac6..709019b9fa6f87522f88b9b993cd4d33e6b845cd 100644 (file)
--- a/osfmk/kern/processor.c
+++ b/osfmk/kern/processor.c
@@ -91,6 +91,7 @@ decl_simple_lock_data(static,pset_node_lock)
 queue_head_t                   tasks;
 queue_head_t                   terminated_tasks;       /* To be used ONLY for stackshot. */
 int                                            tasks_count;
+int                                            terminated_tasks_count;
 queue_head_t                   threads;
 int                                            threads_count;
 decl_lck_mtx_data(,tasks_threads_lock)
@@ -143,6 +144,8 @@ processor_init(
        int                                     cpu_id,
        processor_set_t         pset)
 {
+       spl_t           s;
+
        if (processor != master_processor) {
                /* Scheduler state deferred until sched_init() */
                SCHED(processor_init)(processor);
@@ -162,6 +165,7 @@ processor_init(
        processor_data_init(processor);
        processor->processor_list = NULL;
 
+       s = splsched();
        pset_lock(pset);
        if (pset->cpu_set_count++ == 0)
                pset->cpu_set_low = pset->cpu_set_hi = cpu_id;
@@ -170,6 +174,7 @@ 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)
@@ -255,6 +260,7 @@ pset_init(
        pset_count_init_hint(pset, PROCESSOR_NULL);
        pset->cpu_set_low = pset->cpu_set_hi = 0;
        pset->cpu_set_count = 0;
+       pset->pending_AST_cpu_mask = 0;
        pset_lock_init(pset);
        pset->pset_self = IP_NULL;
        pset->pset_name_self = IP_NULL;
@@ -332,34 +338,71 @@ processor_info(
 
        case PROCESSOR_CPU_LOAD_INFO:
        {
-               register processor_cpu_load_info_t      cpu_load_info;
+               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] =
+               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);
-               {
-               timer_data_t    idle_temp;
-               timer_t         idle_state;
+               } else {
+                       uint64_t tval = timer_grab(&PROCESSOR_DATA(processor, user_state)) +
+                               timer_grab(&PROCESSOR_DATA(processor, system_state));
 
-               idle_state = &PROCESSOR_DATA(processor, idle_state);
-               idle_temp = *idle_state;
+                       cpu_load_info->cpu_ticks[CPU_STATE_USER] = (uint32_t)(tval / hz_tick_interval);
+                       cpu_load_info->cpu_ticks[CPU_STATE_SYSTEM] = 0;
+               }
 
-               if (PROCESSOR_DATA(processor, current_state) != idle_state ||
-                   timer_grab(&idle_temp) != timer_grab(idle_state))
+               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)(timer_grab(&PROCESSOR_DATA(processor, idle_state)) / hz_tick_interval);
-               else {
-                       timer_advance(&idle_temp, mach_absolute_time() - idle_temp.tstamp);
+                                                       (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)(timer_grab(&idle_temp) / hz_tick_interval);
-               }
+                               (uint32_t)(idle_time_snapshot1 / hz_tick_interval);
                }
+
                cpu_load_info->cpu_ticks[CPU_STATE_NICE] = 0;
 
            *count = PROCESSOR_CPU_LOAD_INFO_COUNT;
@@ -525,6 +568,9 @@ 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);
@@ -941,15 +987,6 @@ processor_set_threads(
 {
     return KERN_FAILURE;
 }
-#elif defined(CONFIG_EMBEDDED)
-kern_return_t
-processor_set_threads(
-       __unused processor_set_t                pset,
-       __unused thread_array_t         *thread_list,
-       __unused mach_msg_type_number_t *count)
-{
-    return KERN_NOT_SUPPORTED;
-}
 #else
 kern_return_t
 processor_set_threads(