]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/pmCPU.c
xnu-2050.22.13.tar.gz
[apple/xnu.git] / osfmk / i386 / pmCPU.c
index 91e3799bb0e70032a5efaf61944acc8c079aef60..b22749df79e3ce64935e3ba068baf6815d4611f2 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2004-2010 Apple Inc. All rights reserved.
+ * Copyright (c) 2004-2011 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
 #include <i386/cpu_threads.h>
 #include <i386/pmCPU.h>
 #include <i386/cpuid.h>
-#include <i386/rtclock.h>
+#include <i386/rtclock_protos.h>
 #include <kern/sched_prim.h>
 #include <i386/lapic.h>
+#include <i386/pal_routines.h>
+
+#include <sys/kdebug.h>
 
 extern int disableConsoleOutput;
 
-decl_simple_lock_data(,pm_init_lock);
+#define DELAY_UNSET            0xFFFFFFFFFFFFFFFFULL
 
 /*
  * The following is set when the KEXT loads and initializes.
  */
 pmDispatch_t   *pmDispatch     = NULL;
 
-static uint32_t                pmInitDone      = 0;
-
+static uint32_t                pmInitDone              = 0;
+static boolean_t       earlyTopology           = FALSE;
+static uint64_t                earlyMaxBusDelay        = DELAY_UNSET;
+static uint64_t                earlyMaxIntDelay        = DELAY_UNSET;
 
 /*
  * Initialize the Cstate change code.
@@ -67,16 +72,6 @@ static uint32_t              pmInitDone      = 0;
 void
 power_management_init(void)
 {
-    static boolean_t   initialized     = FALSE;
-
-    /*
-     * Initialize the lock for the KEXT initialization.
-     */
-    if (!initialized) {
-       simple_lock_init(&pm_init_lock, 0);
-       initialized = TRUE;
-    }
-
     if (pmDispatch != NULL && pmDispatch->cstateInit != NULL)
        (*pmDispatch->cstateInit)();
 }
@@ -97,6 +92,20 @@ machine_idle(void)
     DBGLOG(cpu_handle, cpu_number(), MP_IDLE);
     MARK_CPU_IDLE(cpu_number());
 
+    if (pmInitDone) {
+       /*
+        * Handle case where ml_set_maxbusdelay() or ml_set_maxintdelay()
+        * were called prior to the CPU PM kext being registered.  We do
+        * this here since we know at this point the values will be first
+        * used since idle is where the decisions using these values is made.
+        */
+       if (earlyMaxBusDelay != DELAY_UNSET)
+           ml_set_maxbusdelay((uint32_t)(earlyMaxBusDelay & 0xFFFFFFFF));
+
+       if (earlyMaxIntDelay != DELAY_UNSET)
+           ml_set_maxintdelay(earlyMaxIntDelay);
+    }
+
     if (pmInitDone
        && pmDispatch != NULL
        && pmDispatch->MachineIdle != NULL)
@@ -109,7 +118,10 @@ machine_idle(void)
         * cause problems in some MP configurations w.r.t. the APIC
         * stopping during a GV3 transition).
         */
-       __asm__ volatile ("sti; hlt");
+       pal_hlt();
+
+       /* Once woken, re-disable interrupts. */
+       pal_cli();
     }
 
     /*
@@ -123,7 +135,7 @@ machine_idle(void)
      * Re-enable interrupts.
      */
   out:
-    __asm__ volatile("sti");
+    pal_sti();
 }
 
 /*
@@ -138,19 +150,19 @@ pmCPUHalt(uint32_t reason)
     switch (reason) {
     case PM_HALT_DEBUG:
        cpup->lcpu.state = LCPU_PAUSE;
-       __asm__ volatile ("wbinvd; hlt");
+       pal_stop_cpu(FALSE);
        break;
 
     case PM_HALT_PANIC:
        cpup->lcpu.state = LCPU_PAUSE;
-       __asm__ volatile ("cli; wbinvd; hlt");
+       pal_stop_cpu(TRUE);
        break;
 
     case PM_HALT_NORMAL:
     default:
-       __asm__ volatile ("cli");
+        pal_cli();
 
-    if (pmInitDone
+       if (pmInitDone
            && pmDispatch != NULL
            && pmDispatch->pmCPUHalt != NULL) {
            /*
@@ -164,7 +176,8 @@ pmCPUHalt(uint32_t reason)
            i386_init_slave_fast();
 
            panic("init_slave_fast returned");
-       } else {
+       } else
+       {
            /*
             * If no power managment and a processor is taken off-line,
             * then invalidate the cache and halt it (it will not be able
@@ -172,10 +185,11 @@ pmCPUHalt(uint32_t reason)
             */
            __asm__ volatile ("wbinvd");
            cpup->lcpu.state = LCPU_HALT;
-           __asm__ volatile ( "wbinvd; hlt" );
+           pal_stop_cpu(FALSE);
 
            panic("back from Halt");
        }
+
        break;
     }
 }
@@ -192,6 +206,13 @@ pmMarkAllCPUsOff(void)
 static void
 pmInitComplete(void)
 {
+    if (earlyTopology
+       && pmDispatch != NULL
+       && pmDispatch->pmCPUStateInit != NULL) {
+       (*pmDispatch->pmCPUStateInit)();
+       earlyTopology = FALSE;
+    }
+
     pmInitDone = 1;
 }
 
@@ -264,13 +285,15 @@ pmLockCPUTopology(int lock)
 /*
  * Called to get the next deadline that has been set by the
  * power management code.
+ * Note: a return of 0 from AICPM and this routine signifies
+ * that no deadline is set.
  */
 uint64_t
 pmCPUGetDeadline(cpu_data_t *cpu)
 {
-    uint64_t   deadline        = EndOfAllTime;
+    uint64_t   deadline        = 0;
 
-       if (pmInitDone
+    if (pmInitDone
        && pmDispatch != NULL
        && pmDispatch->GetDeadline != NULL)
        deadline = (*pmDispatch->GetDeadline)(&cpu->lcpu);
@@ -357,6 +380,8 @@ pmCPUStateInit(void)
 {
     if (pmDispatch != NULL && pmDispatch->pmCPUStateInit != NULL)
        (*pmDispatch->pmCPUStateInit)();
+    else
+       earlyTopology = TRUE;
 }
 
 /*
@@ -432,7 +457,8 @@ ml_get_maxsnoop(void)
 {
     uint64_t   max_snoop       = 0;
 
-    if (pmDispatch != NULL
+    if (pmInitDone
+       && pmDispatch != NULL
        && pmDispatch->getMaxSnoop != NULL)
        max_snoop = pmDispatch->getMaxSnoop();
 
@@ -445,7 +471,8 @@ ml_get_maxbusdelay(void)
 {
     uint64_t   max_delay       = 0;
 
-    if (pmDispatch != NULL
+    if (pmInitDone
+       && pmDispatch != NULL
        && pmDispatch->getMaxBusDelay != NULL)
        max_delay = pmDispatch->getMaxBusDelay();
 
@@ -466,8 +493,11 @@ ml_set_maxbusdelay(uint32_t mdelay)
     uint64_t   maxdelay        = mdelay;
 
     if (pmDispatch != NULL
-       && pmDispatch->setMaxBusDelay != NULL)
+       && pmDispatch->setMaxBusDelay != NULL) {
+       earlyMaxBusDelay = DELAY_UNSET;
        pmDispatch->setMaxBusDelay(maxdelay);
+    } else
+       earlyMaxBusDelay = maxdelay;
 }
 
 uint64_t
@@ -489,8 +519,24 @@ void
 ml_set_maxintdelay(uint64_t mdelay)
 {
     if (pmDispatch != NULL
-       && pmDispatch->setMaxIntDelay != NULL)
+       && pmDispatch->setMaxIntDelay != NULL) {
+       earlyMaxIntDelay = DELAY_UNSET;
        pmDispatch->setMaxIntDelay(mdelay);
+    } else
+       earlyMaxIntDelay = mdelay;
+}
+
+boolean_t
+ml_get_interrupt_prewake_applicable()
+{
+    boolean_t applicable = FALSE;
+
+    if (pmInitDone 
+       && pmDispatch != NULL
+       && pmDispatch->pmInterruptPrewakeApplicable != NULL)
+       applicable = pmDispatch->pmInterruptPrewakeApplicable();
+
+    return applicable;
 }
 
 /*
@@ -596,6 +642,72 @@ machine_choose_processor(processor_set_t pset,
     return(preferred);
 }
 
+static int
+pmThreadGetUrgency(uint64_t *rt_period, uint64_t *rt_deadline)
+{
+
+    return(thread_get_urgency(rt_period, rt_deadline));
+}
+
+#if    DEBUG
+uint32_t       urgency_stats[64][THREAD_URGENCY_MAX];
+#endif
+
+#define                URGENCY_NOTIFICATION_ASSERT_NS (5 * 1000 * 1000)
+uint64_t       urgency_notification_assert_abstime_threshold, urgency_notification_max_recorded;
+
+void
+thread_tell_urgency(int urgency,
+    uint64_t rt_period,
+    uint64_t rt_deadline)
+{
+       uint64_t        urgency_notification_time_start, delta;
+       boolean_t       urgency_assert = (urgency_notification_assert_abstime_threshold != 0);
+       assert(get_preemption_level() > 0 || ml_get_interrupts_enabled() == FALSE);
+#if    DEBUG
+       urgency_stats[cpu_number() % 64][urgency]++;
+#endif
+       if (!pmInitDone
+           || pmDispatch == NULL
+           || pmDispatch->pmThreadTellUrgency == NULL)
+               return;
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_URGENCY) | DBG_FUNC_START, urgency, rt_period, (rt_deadline >> 32), rt_deadline, 0);
+
+       if (__improbable((urgency_assert == TRUE)))
+               urgency_notification_time_start = mach_absolute_time();
+
+       pmDispatch->pmThreadTellUrgency(urgency, rt_period, rt_deadline);
+
+       if (__improbable((urgency_assert == TRUE))) {
+               delta = mach_absolute_time() - urgency_notification_time_start;
+
+               if (__improbable(delta > urgency_notification_max_recorded)) {
+                       /* This is not synchronized, but it doesn't matter
+                        * if we (rarely) miss an event, as it is statistically
+                        * unlikely that it will never recur.
+                        */
+                       urgency_notification_max_recorded = delta;
+
+                       if (__improbable((delta > urgency_notification_assert_abstime_threshold) && !machine_timeout_suspended()))
+                               panic("Urgency notification callout %p exceeded threshold, 0x%llx abstime units", pmDispatch->pmThreadTellUrgency, delta);
+               }
+       }
+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_URGENCY) | DBG_FUNC_END, urgency, rt_period, (rt_deadline >> 32), rt_deadline, 0);
+}
+
+void
+active_rt_threads(boolean_t active)
+{
+    if (!pmInitDone
+       || pmDispatch == NULL
+       || pmDispatch->pmActiveRTThreads == NULL)
+       return;
+
+    pmDispatch->pmActiveRTThreads(active);
+}
+
 static uint32_t
 pmGetSavedRunCount(void)
 {
@@ -645,12 +757,32 @@ pmSendIPI(int cpu)
     lapic_send_ipi(cpu, LAPIC_PM_INTERRUPT);
 }
 
-static rtc_nanotime_t *
-pmGetNanotimeInfo(void)
+static void
+pmGetNanotimeInfo(pm_rtc_nanotime_t *rtc_nanotime)
+{
+       /*
+        * Make sure that nanotime didn't change while we were reading it.
+        */
+       do {
+               rtc_nanotime->generation = pal_rtc_nanotime_info.generation; /* must be first */
+               rtc_nanotime->tsc_base = pal_rtc_nanotime_info.tsc_base;
+               rtc_nanotime->ns_base = pal_rtc_nanotime_info.ns_base;
+               rtc_nanotime->scale = pal_rtc_nanotime_info.scale;
+               rtc_nanotime->shift = pal_rtc_nanotime_info.shift;
+       } while(pal_rtc_nanotime_info.generation != 0
+               && rtc_nanotime->generation != pal_rtc_nanotime_info.generation);
+}
+
+static uint32_t
+pmTimerQueueMigrate(int target_cpu)
 {
-    return(&rtc_nanotime_info);
+    /* Call the etimer code to do this. */
+    return (target_cpu != cpu_number())
+               ? etimer_queue_migrate(target_cpu)
+               : 0;
 }
 
+
 /*
  * Called by the power management kext to register itself and to get the
  * callbacks it might need into other kernel functions.  This interface
@@ -679,10 +811,16 @@ pmKextRegister(uint32_t version, pmDispatch_t *cpuFuncs,
        callbacks->LCPUtoProcessor      = pmLCPUtoProcessor;
        callbacks->ThreadBind           = thread_bind;
        callbacks->GetSavedRunCount     = pmGetSavedRunCount;
-       callbacks->pmSendIPI            = pmSendIPI;
        callbacks->GetNanotimeInfo      = pmGetNanotimeInfo;
-       callbacks->RTCClockAdjust       = rtc_clock_adjust;
+       callbacks->ThreadGetUrgency     = pmThreadGetUrgency;
+       callbacks->RTCClockAdjust       = rtc_clock_adjust;
+       callbacks->timerQueueMigrate    = pmTimerQueueMigrate;
        callbacks->topoParms            = &topoParms;
+       callbacks->pmSendIPI            = pmSendIPI;
+       callbacks->InterruptPending     = lapic_is_interrupt_pending;
+       callbacks->IsInterrupting       = lapic_is_interrupting;
+       callbacks->InterruptStats       = lapic_interrupt_counts;
+       callbacks->DisableApicTimer     = lapic_disable_timer;
     } else {
        panic("Version mis-match between Kernel and CPU PM");
     }
@@ -690,6 +828,12 @@ pmKextRegister(uint32_t version, pmDispatch_t *cpuFuncs,
     if (cpuFuncs != NULL) {
        pmDispatch = cpuFuncs;
 
+       if (earlyTopology
+           && pmDispatch->pmCPUStateInit != NULL) {
+           (*pmDispatch->pmCPUStateInit)();
+           earlyTopology = FALSE;
+       }
+
        if (pmDispatch->pmIPIHandler != NULL) {
            lapic_set_pm_func((i386_intr_func_t)pmDispatch->pmIPIHandler);
        }