]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/pmCPU.c
xnu-1504.9.17.tar.gz
[apple/xnu.git] / osfmk / i386 / pmCPU.c
index 63108e188c57638969d83b5b5da9655eff570c57..91e3799bb0e70032a5efaf61944acc8c079aef60 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2004-2006 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2004-2010 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  *
  * Implements the "wrappers" to the KEXT.
  */
-#include <i386/machine_routines.h>
-#include <i386/machine_cpu.h>
-#include <i386/misc_protos.h>
-#include <i386/pmap.h>
 #include <i386/asm.h>
+#include <i386/machine_cpu.h>
 #include <i386/mp.h>
+#include <i386/machine_routines.h>
 #include <i386/proc_reg.h>
+#include <i386/pmap.h>
+#include <i386/misc_protos.h>
+#include <kern/machine.h>
 #include <kern/pms.h>
 #include <kern/processor.h>
+#include <kern/etimer.h>
+#include <i386/cpu_threads.h>
 #include <i386/pmCPU.h>
 #include <i386/cpuid.h>
 #include <i386/rtclock.h>
-#if MACH_KDB
-#include <i386/db_machdep.h>
-#include <ddb/db_aout.h>
-#include <ddb/db_access.h>
-#include <ddb/db_sym.h>
-#include <ddb/db_variables.h>
-#include <ddb/db_command.h>
-#include <ddb/db_output.h>
-#include <ddb/db_expr.h>
-#endif
+#include <kern/sched_prim.h>
+#include <i386/lapic.h>
 
 extern int disableConsoleOutput;
 
@@ -63,332 +58,642 @@ decl_simple_lock_data(,pm_init_lock);
  */
 pmDispatch_t   *pmDispatch     = NULL;
 
+static uint32_t                pmInitDone      = 0;
+
+
 /*
- * Current power management states (for use until KEXT is loaded).
+ * Initialize the Cstate change code.
  */
-static pmInitState_t   pmInitState;
+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)();
+}
 
 /*
- * Nap control variables:
+ * Called when the CPU is idle.  It calls into the power management kext
+ * to determine the best way to idle the CPU.
  */
-uint32_t napCtl = 0;                   /* Defaults to neither napping
-                                          nor halting */
-uint32_t forcenap = 0;                 /* Force nap (fn) boot-arg controls */
-uint32_t maxBusDelay = 0xFFFFFFFF;     /* Maximum memory bus delay that
-                                          I/O devices can tolerate
-                                          before errors (nanoseconds) */
-uint32_t C4C2SnoopDelay = 0;           /* C4 to C2 transition time -
-                                          time before a C4 system
-                                          can snoop (nanoseconds) */
+void
+machine_idle(void)
+{
+    cpu_data_t         *my_cpu         = current_cpu_datap();
+
+    if (my_cpu == NULL)
+       goto out;
+
+    my_cpu->lcpu.state = LCPU_IDLE;
+    DBGLOG(cpu_handle, cpu_number(), MP_IDLE);
+    MARK_CPU_IDLE(cpu_number());
+
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->MachineIdle != NULL)
+       (*pmDispatch->MachineIdle)(0x7FFFFFFFFFFFFFFFULL);
+    else {
+       /*
+        * If no power management, re-enable interrupts and halt.
+        * This will keep the CPU from spinning through the scheduler
+        * and will allow at least some minimal power savings (but it
+        * cause problems in some MP configurations w.r.t. the APIC
+        * stopping during a GV3 transition).
+        */
+       __asm__ volatile ("sti; hlt");
+    }
+
+    /*
+     * Mark the CPU as running again.
+     */
+    MARK_CPU_ACTIVE(cpu_number());
+    DBGLOG(cpu_handle, cpu_number(), MP_UNIDLE);
+    my_cpu->lcpu.state = LCPU_RUN;
+
+    /*
+     * Re-enable interrupts.
+     */
+  out:
+    __asm__ volatile("sti");
+}
 
 /*
- * We are being asked to set PState (sel).
+ * Called when the CPU is to be halted.  It will choose the best C-State
+ * to be in.
  */
 void
-pmsCPUSet(uint32_t sel)
+pmCPUHalt(uint32_t reason)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPUSet != NULL)
-       (*pmDispatch->pmsCPUSet)(sel);
-    else
-       pmInitState.PState = sel;
+    cpu_data_t *cpup   = current_cpu_datap();
+
+    switch (reason) {
+    case PM_HALT_DEBUG:
+       cpup->lcpu.state = LCPU_PAUSE;
+       __asm__ volatile ("wbinvd; hlt");
+       break;
+
+    case PM_HALT_PANIC:
+       cpup->lcpu.state = LCPU_PAUSE;
+       __asm__ volatile ("cli; wbinvd; hlt");
+       break;
+
+    case PM_HALT_NORMAL:
+    default:
+       __asm__ volatile ("cli");
+
+    if (pmInitDone
+           && pmDispatch != NULL
+           && pmDispatch->pmCPUHalt != NULL) {
+           /*
+            * Halt the CPU (and put it in a low power state.
+            */
+           (*pmDispatch->pmCPUHalt)();
+
+           /*
+            * We've exited halt, so get the the CPU schedulable again.
+            */
+           i386_init_slave_fast();
+
+           panic("init_slave_fast returned");
+       } else {
+           /*
+            * If no power managment and a processor is taken off-line,
+            * then invalidate the cache and halt it (it will not be able
+            * to be brought back on-line without resetting the CPU).
+            */
+           __asm__ volatile ("wbinvd");
+           cpup->lcpu.state = LCPU_HALT;
+           __asm__ volatile ( "wbinvd; hlt" );
+
+           panic("back from Halt");
+       }
+       break;
+    }
+}
+
+void
+pmMarkAllCPUsOff(void)
+{
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->markAllCPUsOff != NULL)
+       (*pmDispatch->markAllCPUsOff)();
+}
+
+static void
+pmInitComplete(void)
+{
+    pmInitDone = 1;
+}
+
+static x86_lcpu_t *
+pmGetLogicalCPU(int cpu)
+{
+    return(cpu_to_lcpu(cpu));
+}
+
+static x86_lcpu_t *
+pmGetMyLogicalCPU(void)
+{
+    cpu_data_t *cpup   = current_cpu_datap();
+
+    return(&cpup->lcpu);
+}
+
+static x86_core_t *
+pmGetCore(int cpu)
+{
+    return(cpu_to_core(cpu));
+}
+
+static x86_core_t *
+pmGetMyCore(void)
+{
+    cpu_data_t *cpup   = current_cpu_datap();
+
+    return(cpup->lcpu.core);
+}
+
+static x86_die_t *
+pmGetDie(int cpu)
+{
+    return(cpu_to_die(cpu));
+}
+
+static x86_die_t *
+pmGetMyDie(void)
+{
+    cpu_data_t *cpup   = current_cpu_datap();
+
+    return(cpup->lcpu.die);
+}
+
+static x86_pkg_t *
+pmGetPackage(int cpu)
+{
+    return(cpu_to_package(cpu));
+}
+
+static x86_pkg_t *
+pmGetMyPackage(void)
+{
+    cpu_data_t *cpup   = current_cpu_datap();
+
+    return(cpup->lcpu.package);
+}
+
+static void
+pmLockCPUTopology(int lock)
+{
+    if (lock) {
+       simple_lock(&x86_topo_lock);
+    } else {
+       simple_unlock(&x86_topo_lock);
+    }
 }
 
 /*
- * This code configures the initial step tables.  It should be called after
- * the timebase frequency is initialized.
- *
- * Note that this is not used in normal operation.  It is strictly for
- * debugging/testing purposes.
+ * Called to get the next deadline that has been set by the
+ * power management code.
  */
-void
-pmsCPUConf(void)
+uint64_t
+pmCPUGetDeadline(cpu_data_t *cpu)
 {
+    uint64_t   deadline        = EndOfAllTime;
 
-    if (pmDispatch != NULL && pmDispatch->pmsCPUConf != NULL)
-       (*pmDispatch->pmsCPUConf)();
+       if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->GetDeadline != NULL)
+       deadline = (*pmDispatch->GetDeadline)(&cpu->lcpu);
+
+    return(deadline);
 }
 
 /*
- * Machine-dependent initialization.
+ * Called to determine if the supplied deadline or the power management
+ * deadline is sooner.  Returns which ever one is first.
  */
-void
-pmsCPUMachineInit(void)
+uint64_t
+pmCPUSetDeadline(cpu_data_t *cpu, uint64_t deadline)
 {
-    /*
-     * Initialize some of the initial state to "uninitialized" until
-     * it gets set with something more useful.  This allows the KEXT
-     * to determine if the initial value was actually set to something.
-     */
-    pmInitState.PState = -1;
-    pmInitState.PLimit = -1;
+   if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->SetDeadline != NULL)
+       deadline = (*pmDispatch->SetDeadline)(&cpu->lcpu, deadline);
 
-    if (pmDispatch != NULL && pmDispatch->pmsCPUMachineInit != NULL)
-       (*pmDispatch->pmsCPUMachineInit)();
+    return(deadline);
 }
 
 /*
- * This function should be called once for each processor to force the
- * processor to the correct initial voltage and frequency.
+ * Called when a power management deadline expires.
  */
 void
-pmsCPUInit(void)
+pmCPUDeadline(cpu_data_t *cpu)
+{
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->Deadline != NULL)
+       (*pmDispatch->Deadline)(&cpu->lcpu);
+}
+
+/*
+ * Called to get a CPU out of idle.
+ */
+boolean_t
+pmCPUExitIdle(cpu_data_t *cpu)
+{
+    boolean_t          do_ipi;
+
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->exitIdle != NULL)
+       do_ipi = (*pmDispatch->exitIdle)(&cpu->lcpu);
+    else
+       do_ipi = TRUE;
+
+    return(do_ipi);
+}
+
+kern_return_t
+pmCPUExitHalt(int cpu)
 {
-    pmsCPUMachineInit();
-    if (pmDispatch != NULL && pmDispatch->pmsCPUInit != NULL)
-       (*pmDispatch->pmsCPUInit)();
+    kern_return_t      rc      = KERN_INVALID_ARGUMENT;
+
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->exitHalt != NULL)
+       rc = pmDispatch->exitHalt(cpu_to_lcpu(cpu));
+
+    return(rc);
+}
+
+kern_return_t
+pmCPUExitHaltToOff(int cpu)
+{
+    kern_return_t      rc      = KERN_INVALID_ARGUMENT;
+
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->exitHaltToOff != NULL)
+       rc = pmDispatch->exitHaltToOff(cpu_to_lcpu(cpu));
+
+    return(rc);
 }
 
 /*
- * Broadcast a change to all processing including ourselves.
+ * Called to initialize the power management structures for the CPUs.
  */
 void
-pmsCPURun(uint32_t nstep)
+pmCPUStateInit(void)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPURun != NULL)
-       (*pmDispatch->pmsCPURun)(nstep);
+    if (pmDispatch != NULL && pmDispatch->pmCPUStateInit != NULL)
+       (*pmDispatch->pmCPUStateInit)();
 }
 
 /*
- * Return the current state of a core.
+ * Called when a CPU is being restarted after being powered off (as in S3).
  */
-uint32_t
-pmsCPUQuery(void)
+void
+pmCPUMarkRunning(cpu_data_t *cpu)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPUQuery != NULL)
-       return((*pmDispatch->pmsCPUQuery)());
+    cpu_data_t *cpup   = current_cpu_datap();
 
-    /*
-     * Return a non-sense value.
-     */
-    return((~0) << 16);
+    if (pmInitDone
+       && pmDispatch != NULL
+       && pmDispatch->markCPURunning != NULL)
+       (*pmDispatch->markCPURunning)(&cpu->lcpu);
+    else
+       cpup->lcpu.state = LCPU_RUN;
 }
 
 /*
- * Return the current state of the package.
+ * Called to get/set CPU power management state.
  */
-uint32_t
-pmsCPUPackageQuery(void)
+int
+pmCPUControl(uint32_t cmd, void *datap)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPUPackageQuery != NULL)
-       return((*pmDispatch->pmsCPUPackageQuery)());
+    int                rc      = -1;
 
-    /*
-     * Return a non-sense value.
-     */
-    return((~0) << 16);
+    if (pmDispatch != NULL
+       && pmDispatch->pmCPUControl != NULL)
+       rc = (*pmDispatch->pmCPUControl)(cmd, datap);
+
+    return(rc);
 }
 
 /*
- * Force the CPU package to the lowest power level.  This is a low-level
- * interface meant to be called from the panic or debugger code to bring
- * the CPU to a safe power level for unmanaged operation.
- *
- * Note that while this will bring an entire package to a safe level, it
- * cannot affect other packages.  As a general rule, this should be run on
- * every code as part of entering the debugger or on the panic path.
+ * Called to save the timer state used by power management prior
+ * to "sleeping".
  */
 void
-pmsCPUYellowFlag(void)
+pmTimerSave(void)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPUYellowFlag != NULL)
-       (*pmDispatch->pmsCPUYellowFlag)();
+    if (pmDispatch != NULL
+       && pmDispatch->pmTimerStateSave != NULL)
+       (*pmDispatch->pmTimerStateSave)();
 }
 
 /*
- * Restore the CPU to the power state it was in before a yellow flag.
+ * Called to restore the timer state used by power management after
+ * waking from "sleep".
  */
 void
-pmsCPUGreenFlag(void)
+pmTimerRestore(void)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPUGreenFlag != NULL)
-       (*pmDispatch->pmsCPUGreenFlag)();
+    if (pmDispatch != NULL
+       && pmDispatch->pmTimerStateRestore != NULL)
+       (*pmDispatch->pmTimerStateRestore)();
 }
 
 /*
- * Load a new ratio/VID table.
- *
- * Note that this interface is specific to the Intel SpeedStep implementation.
- * It is expected that this will only be called once to override the default
- * ratio/VID table when the platform starts.
- *
- * Normally, the table will need to be replaced at the same time that the
- * stepper program proper is replaced, as the PState indices from an old
- * program may no longer be valid.  When replacing the default program this
- * should not be a problem as any new table will have at least two PState
- * entries and the default program only references P0 and P1.
+ * Set the worst-case time for the C4 to C2 transition.
+ * No longer does anything.
  */
-kern_return_t
-pmsCPULoadVIDTable(uint16_t *tablep, int nstates)
+void
+ml_set_maxsnoop(__unused uint32_t maxdelay)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPULoadVIDTable != NULL)
-       return((*pmDispatch->pmsCPULoadVIDTable)(tablep, nstates));
-    else {
-       int     i;
-
-       if (nstates > MAX_PSTATES)
-           return(KERN_FAILURE);
-
-       for (i = 0; i < nstates; i += 1)
-           pmInitState.VIDTable[i] = tablep[i];
-    }
-    return(KERN_SUCCESS);
 }
 
+
 /*
- * Set the (global) PState limit.  CPUs will not be permitted to run at
- * a lower (more performant) PState than this.
+ * Get the worst-case time for the C4 to C2 transition.  Returns nanoseconds.
  */
-kern_return_t
-pmsCPUSetPStateLimit(uint32_t limit)
+unsigned
+ml_get_maxsnoop(void)
 {
-    if (pmDispatch != NULL && pmDispatch->pmsCPUSetPStateLimit != NULL)
-       return((*pmDispatch->pmsCPUSetPStateLimit)(limit));
+    uint64_t   max_snoop       = 0;
 
-    pmInitState.PLimit = limit;
-    return(KERN_SUCCESS);
+    if (pmDispatch != NULL
+       && pmDispatch->getMaxSnoop != NULL)
+       max_snoop = pmDispatch->getMaxSnoop();
+
+    return((unsigned)(max_snoop & 0xffffffff));
+}
+
+
+uint32_t
+ml_get_maxbusdelay(void)
+{
+    uint64_t   max_delay       = 0;
+
+    if (pmDispatch != NULL
+       && pmDispatch->getMaxBusDelay != NULL)
+       max_delay = pmDispatch->getMaxBusDelay();
+
+    return((uint32_t)(max_delay & 0xffffffff));
 }
 
 /*
- * Initialize the Cstate change code.
+ * Set the maximum delay time allowed for snoop on the bus.
+ *
+ * Note that this value will be compared to the amount of time that it takes
+ * to transition from a non-snooping power state (C4) to a snooping state (C2).
+ * If maxBusDelay is less than C4C2SnoopDelay,
+ * we will not enter the lowest power state.
  */
 void
-power_management_init(void)
+ml_set_maxbusdelay(uint32_t mdelay)
 {
-    uint32_t   cpuModel;
-    uint32_t   cpuFamily;
-    uint32_t   xcpuid[4];
+    uint64_t   maxdelay        = mdelay;
 
-    /*
-     * Initialize the lock for the KEXT initialization.
-     */
-    simple_lock_init(&pm_init_lock, 0);
+    if (pmDispatch != NULL
+       && pmDispatch->setMaxBusDelay != NULL)
+       pmDispatch->setMaxBusDelay(maxdelay);
+}
 
-    /*
-     * XXX
-     *
-     * The following is a hack to disable power management on some systems
-     * until the KEXT is done.  This is strictly temporary!!!
-     */
-    do_cpuid(1, xcpuid);
-    cpuFamily = (xcpuid[eax] >> 8) & 0xf;
-    cpuModel  = (xcpuid[eax] >> 4) & 0xf;
+uint64_t
+ml_get_maxintdelay(void)
+{
+    uint64_t   max_delay       = 0;
 
-    if (cpuFamily != 0x6 || cpuModel < 0xe)
-       pmDispatch = NULL;
+    if (pmDispatch != NULL
+       && pmDispatch->getMaxIntDelay != NULL)
+       max_delay = pmDispatch->getMaxIntDelay();
 
-    if (pmDispatch != NULL && pmDispatch->cstateInit != NULL)
-       (*pmDispatch->cstateInit)();
+    return(max_delay);
 }
 
 /*
- * This function will update the system nap policy.  It should be called
- * whenever conditions change: when the system is ready to being napping
- * and if something changes the rules (e.g. a sysctl altering the policy
- * for debugging).
+ * Set the maximum delay allowed for an interrupt.
  */
 void
-machine_nap_policy(void)
+ml_set_maxintdelay(uint64_t mdelay)
 {
-    if (pmDispatch != NULL && pmDispatch->cstateNapPolicy != NULL)
-       napCtl = (*pmDispatch->cstateNapPolicy)(forcenap, napCtl);
+    if (pmDispatch != NULL
+       && pmDispatch->setMaxIntDelay != NULL)
+       pmDispatch->setMaxIntDelay(mdelay);
 }
 
 /*
- * ACPI calls the following routine to set/update mwait hints.  A table
- * (possibly null) specifies the available Cstates and their hints, all
- * other states are assumed to be invalid.  ACPI may update available
- * states to change the nap policy (for example, while AC power is
- * available).
+ * Put a CPU into "safe" mode with respect to power.
+ *
+ * Some systems cannot operate at a continuous "normal" speed without
+ * exceeding the thermal design.  This is called per-CPU to place the
+ * CPUs into a "safe" operating mode.
  */
-kern_return_t
-Cstate_table_set(Cstate_hint_t *tablep, unsigned int nstates)
+void
+pmSafeMode(x86_lcpu_t *lcpu, uint32_t flags)
 {
-    if (forcenap)
-       return(KERN_SUCCESS);
-
-    if (pmDispatch != NULL && pmDispatch->cstateTableSet != NULL)
-       return((*pmDispatch->cstateTableSet)(tablep, nstates));
+    if (pmDispatch != NULL
+       && pmDispatch->pmCPUSafeMode != NULL)
+       pmDispatch->pmCPUSafeMode(lcpu, flags);
     else {
-       unsigned int    i;
-
-       for (i = 0; i < nstates; i += 1) {
-           pmInitState.CStates[i].number = tablep[i].number;
-           pmInitState.CStates[i].hint   = tablep[i].hint;
+       /*
+        * Do something reasonable if the KEXT isn't present.
+        *
+        * We only look at the PAUSE and RESUME flags.  The other flag(s)
+        * will not make any sense without the KEXT, so just ignore them.
+        *
+        * We set the CPU's state to indicate that it's halted.  If this
+        * is the CPU we're currently running on, then spin until the
+        * state becomes non-halted.
+        */
+       if (flags & PM_SAFE_FL_PAUSE) {
+           lcpu->state = LCPU_PAUSE;
+           if (lcpu == x86_lcpu()) {
+               while (lcpu->state == LCPU_PAUSE)
+                   cpu_pause();
+           }
+       }
+       
+       /*
+        * Clear the halted flag for the specified CPU, that will
+        * get it out of it's spin loop.
+        */
+       if (flags & PM_SAFE_FL_RESUME) {
+           lcpu->state = LCPU_RUN;
        }
+    }
+}
+
+static uint32_t                saved_run_count = 0;
 
-       pmInitState.CStatesCount = nstates;
+void
+machine_run_count(uint32_t count)
+{
+    if (pmDispatch != NULL
+       && pmDispatch->pmSetRunCount != NULL)
+       pmDispatch->pmSetRunCount(count);
+    else
+       saved_run_count = count;
+}
+
+boolean_t
+machine_processor_is_inactive(processor_t processor)
+{
+    int                cpu = processor->cpu_id;
+
+    if (pmDispatch != NULL
+       && pmDispatch->pmIsCPUUnAvailable != NULL)
+       return(pmDispatch->pmIsCPUUnAvailable(cpu_to_lcpu(cpu)));
+    else
+       return(FALSE);
+}
+
+processor_t
+machine_choose_processor(processor_set_t pset,
+                        processor_t preferred)
+{
+    int                startCPU;
+    int                endCPU;
+    int                preferredCPU;
+    int                chosenCPU;
+
+    if (!pmInitDone)
+       return(preferred);
+
+    if (pset == NULL) {
+       startCPU = -1;
+       endCPU = -1;
+    } else {
+       startCPU = pset->cpu_set_low;
+       endCPU = pset->cpu_set_hi;
     }
-    return(KERN_SUCCESS);
+
+    if (preferred == NULL)
+       preferredCPU = -1;
+    else
+       preferredCPU = preferred->cpu_id;
+
+    if (pmDispatch != NULL
+       && pmDispatch->pmChooseCPU != NULL) {
+       chosenCPU = pmDispatch->pmChooseCPU(startCPU, endCPU, preferredCPU);
+
+       if (chosenCPU == -1)
+           return(NULL);
+       return(cpu_datap(chosenCPU)->cpu_processor);
+    }
+
+    return(preferred);
 }
 
-static inline void
-sti(void) {
-       __asm__ volatile ( "sti" : : : "memory");
+static uint32_t
+pmGetSavedRunCount(void)
+{
+    return(saved_run_count);
 }
 
 /*
- * Called when the CPU is idle.  It will choose the best C state to
- * be in.
+ * Returns the root of the package tree.
  */
-void
-machine_idle_cstate(void)
+static x86_pkg_t *
+pmGetPkgRoot(void)
 {
-    if (pmDispatch != NULL && pmDispatch->cstateMachineIdle != NULL)
-       (*pmDispatch->cstateMachineIdle)(napCtl);
-    else {
-       sti();
-    }
+    return(x86_pkgs);
 }
 
-static pmStats_t *
-pmsCPUStats(void)
+static boolean_t
+pmCPUGetHibernate(int cpu)
 {
-    cpu_data_t *pp;
+    return(cpu_datap(cpu)->cpu_hibernate);
+}
 
-    pp = current_cpu_datap();
-    return(&pp->cpu_pmStats);
+static processor_t
+pmLCPUtoProcessor(int lcpu)
+{
+    return(cpu_datap(lcpu)->cpu_processor);
 }
 
-static pmsd *
-pmsCPUStepperData(void)
+static void
+pmReSyncDeadlines(int cpu)
 {
-    cpu_data_t *pp;
+    static boolean_t   registered      = FALSE;
+
+    if (!registered) {
+       PM_interrupt_register(&etimer_resync_deadlines);
+       registered = TRUE;
+    }
+
+    if ((uint32_t)cpu == current_cpu_datap()->lcpu.cpu_num)
+       etimer_resync_deadlines();
+    else
+       cpu_PM_interrupt(cpu);
+}
 
-    pp = current_cpu_datap();
-    return(&pp->pms);
+static void
+pmSendIPI(int cpu)
+{
+    lapic_send_ipi(cpu, LAPIC_PM_INTERRUPT);
 }
 
-static uint64_t *
-CPUHPETAddr(void)
+static rtc_nanotime_t *
+pmGetNanotimeInfo(void)
 {
-    cpu_data_t *pp;
-    pp = current_cpu_datap();
-    return(pp->cpu_pmHpet);
+    return(&rtc_nanotime_info);
 }
 
 /*
  * Called by the power management kext to register itself and to get the
- * callbacks it might need into other power management functions.
+ * callbacks it might need into other kernel functions.  This interface
+ * is versioned to allow for slight mis-matches between the kext and the
+ * kernel.
  */
 void
-pmRegister(pmDispatch_t *cpuFuncs, pmCallBacks_t *callbacks)
-{
-    if (callbacks != NULL) {
-       callbacks->Park        = pmsPark;
-       callbacks->Run         = pmsRun;
-       callbacks->RunLocal    = pmsRunLocal;
-       callbacks->SetStep     = pmsSetStep;
-       callbacks->NapPolicy   = machine_nap_policy;
-       callbacks->Build       = pmsBuild;
-       callbacks->Stats       = pmsCPUStats;
-       callbacks->StepperData = pmsCPUStepperData;
-       callbacks->HPETAddr    = CPUHPETAddr;
-       callbacks->InitState   = &pmInitState;
-       callbacks->resetPop    = resetPop;
+pmKextRegister(uint32_t version, pmDispatch_t *cpuFuncs,
+              pmCallBacks_t *callbacks)
+{
+    if (callbacks != NULL && version == PM_DISPATCH_VERSION) {
+       callbacks->setRTCPop            = setPop;
+       callbacks->resyncDeadlines      = pmReSyncDeadlines;
+       callbacks->initComplete         = pmInitComplete;
+       callbacks->GetLCPU              = pmGetLogicalCPU;
+       callbacks->GetCore              = pmGetCore;
+       callbacks->GetDie               = pmGetDie;
+       callbacks->GetPackage           = pmGetPackage;
+       callbacks->GetMyLCPU            = pmGetMyLogicalCPU;
+       callbacks->GetMyCore            = pmGetMyCore;
+       callbacks->GetMyDie             = pmGetMyDie;
+       callbacks->GetMyPackage         = pmGetMyPackage;
+       callbacks->GetPkgRoot           = pmGetPkgRoot;
+       callbacks->LockCPUTopology      = pmLockCPUTopology;
+       callbacks->GetHibernate         = pmCPUGetHibernate;
+       callbacks->LCPUtoProcessor      = pmLCPUtoProcessor;
+       callbacks->ThreadBind           = thread_bind;
+       callbacks->GetSavedRunCount     = pmGetSavedRunCount;
+       callbacks->pmSendIPI            = pmSendIPI;
+       callbacks->GetNanotimeInfo      = pmGetNanotimeInfo;
+       callbacks->RTCClockAdjust       = rtc_clock_adjust;
+       callbacks->topoParms            = &topoParms;
+    } else {
+       panic("Version mis-match between Kernel and CPU PM");
     }
 
-    if (cpuFuncs != NULL)
+    if (cpuFuncs != NULL) {
        pmDispatch = cpuFuncs;
+
+       if (pmDispatch->pmIPIHandler != NULL) {
+           lapic_set_pm_func((i386_intr_func_t)pmDispatch->pmIPIHandler);
+       }
+    }
 }
 
 /*
@@ -397,79 +702,47 @@ pmRegister(pmDispatch_t *cpuFuncs, pmCallBacks_t *callbacks)
 void
 pmUnRegister(pmDispatch_t *cpuFuncs)
 {
-    if (cpuFuncs != NULL && pmDispatch == cpuFuncs)
+    if (cpuFuncs != NULL && pmDispatch == cpuFuncs) {
        pmDispatch = NULL;
+    }
 }
 
-#if MACH_KDB
-/*
- * XXX stubs for now
- */
-void
-db_cfg(__unused db_expr_t addr,
-       __unused int have_addr,
-       __unused db_expr_t count,
-       __unused char *modif)
-{
-    return;
-}
-
-void
-db_display_iokit(__unused db_expr_t addr,
-                __unused int have_addr,
-                __unused db_expr_t count,
-                __unused char *modif)
-{
-    return;
-}
-
-void
-db_dtimers(__unused db_expr_t addr,
-          __unused int have_addr,
-          __unused db_expr_t count,
-          __unused char *modif)
+/******************************************************************************
+ *
+ * All of the following are deprecated interfaces and no longer used.
+ *
+ ******************************************************************************/
+kern_return_t
+pmsControl(__unused uint32_t request, __unused user_addr_t reqaddr,
+          __unused uint32_t reqsize)
 {
-    return;
+    return(KERN_SUCCESS);
 }
 
 void
-db_intcnt(__unused db_expr_t addr,
-         __unused int have_addr,
-         __unused db_expr_t count,
-         __unused char *modif)
+pmsInit(void)
 {
-    return;
 }
 
 void
-db_nap(__unused db_expr_t addr,
-       __unused int have_addr,
-       __unused db_expr_t count,
-       __unused char *modif)
+pmsStart(void)
 {
-    return;
 }
 
 void
-db_pmgr(__unused db_expr_t addr,
-       __unused int have_addr,
-       __unused db_expr_t count,
-       __unused char *modif)
+pmsPark(void)
 {
-    return;
 }
 
 void
-db_test(__unused db_expr_t addr,
-       __unused int have_addr,
-       __unused db_expr_t count,
-       __unused char *modif)
+pmsRun(__unused uint32_t nstep)
 {
-    return;
 }
 
-void
-db_getpmgr(__unused pmData_t *pmj)
+kern_return_t
+pmsBuild(__unused pmsDef *pd, __unused uint32_t pdsize,
+        __unused pmsSetFunc_t *functab,
+        __unused uint32_t platformData, __unused pmsQueryFunc_t queryFunc)
 {
+    return(KERN_SUCCESS);
 }
-#endif