xnu-2050.7.9.tar.gz

[apple/xnu.git] / osfmk / i386 / mp.c

diff --git a/osfmk/i386/mp.c b/osfmk/i386/mp.c
index 5073997835e1fe7d83e6e90389b7edbb99839800..b66399d2d7e2d66bd76703d4f505792bb9349874 100644 (file)
--- a/osfmk/i386/mp.c
+++ b/osfmk/i386/mp.c
@@ -1,5 +1,4 @@
 /*
 /*

- * Copyright (c) 2000-2008 Apple Inc. All rights reserved.

  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 


@@ -30,7 +29,6 @@
  */
 
 #include <mach_rt.h>
  */
 
 #include <mach_rt.h>

-#include <mach_kdb.h>

 #include <mach_kdp.h>
 #include <mach_ldebug.h>
 #include <gprof.h>
 #include <mach_kdp.h>
 #include <mach_ldebug.h>
 #include <gprof.h>


@@ -40,6 +38,7 @@
 
 #include <kern/kern_types.h>
 #include <kern/startup.h>
 
 #include <kern/kern_types.h>
 #include <kern/startup.h>

+#include <kern/timer_queue.h>

 #include <kern/processor.h>
 #include <kern/cpu_number.h>
 #include <kern/cpu_data.h>
 #include <kern/processor.h>
 #include <kern/cpu_number.h>
 #include <kern/cpu_data.h>


@@ -47,47 +46,38 @@
 #include <kern/machine.h>
 #include <kern/pms.h>
 #include <kern/misc_protos.h>
 #include <kern/machine.h>
 #include <kern/pms.h>
 #include <kern/misc_protos.h>

+#include <kern/etimer.h>
+#include <kern/kalloc.h>
+#include <kern/queue.h>

 
 #include <vm/vm_map.h>
 #include <vm/vm_kern.h>
 
 #include <profiling/profile-mk.h>
 
 
 #include <vm/vm_map.h>
 #include <vm/vm_kern.h>
 
 #include <profiling/profile-mk.h>
 

-#include <i386/mp.h>
-#include <i386/mp_events.h>
-#include <i386/mp_slave_boot.h>
-#include <i386/lapic.h>
-#include <i386/ipl.h>
-#include <i386/fpu.h>
-#include <i386/cpuid.h>

 #include <i386/proc_reg.h>
 #include <i386/proc_reg.h>

-#include <i386/machine_cpu.h>
-#include <i386/misc_protos.h>
-#include <i386/mtrr.h>
-#include <i386/vmx/vmx_cpu.h>
-#include <i386/postcode.h>
-#include <i386/perfmon.h>

 #include <i386/cpu_threads.h>
 #include <i386/mp_desc.h>
 #include <i386/cpu_threads.h>
 #include <i386/mp_desc.h>

+#include <i386/misc_protos.h>

 #include <i386/trap.h>
 #include <i386/trap.h>

+#include <i386/postcode.h>

 #include <i386/machine_routines.h>
 #include <i386/machine_routines.h>

+#include <i386/mp.h>
+#include <i386/mp_events.h>
+#include <i386/lapic.h>
+#include <i386/cpuid.h>
+#include <i386/fpu.h>
+#include <i386/machine_cpu.h>

 #include <i386/pmCPU.h>
 #include <i386/pmCPU.h>

+#if CONFIG_MCA

 #include <i386/machine_check.h>
 #include <i386/machine_check.h>

+#endif
+#include <i386/acpi.h>

 
 #include <chud/chud_xnu.h>
 #include <chud/chud_xnu_private.h>
 
 #include <sys/kdebug.h>
 
 #include <chud/chud_xnu.h>
 #include <chud/chud_xnu_private.h>
 
 #include <sys/kdebug.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

 
 #if    MP_DEBUG
 #define PAUSE          delay(1000000)
 
 #if    MP_DEBUG
 #define PAUSE          delay(1000000)


@@ -97,54 +87,84 @@
 #define PAUSE
 #endif /* MP_DEBUG */
 
 #define PAUSE
 #endif /* MP_DEBUG */
 

-#define FULL_SLAVE_INIT        (NULL)
-#define FAST_SLAVE_INIT        ((void *)(uintptr_t)1)
+/* Debugging/test trace events: */
+#define        TRACE_MP_TLB_FLUSH              MACHDBG_CODE(DBG_MACH_MP, 0)
+#define        TRACE_MP_CPUS_CALL              MACHDBG_CODE(DBG_MACH_MP, 1)
+#define        TRACE_MP_CPUS_CALL_LOCAL        MACHDBG_CODE(DBG_MACH_MP, 2)
+#define        TRACE_MP_CPUS_CALL_ACTION       MACHDBG_CODE(DBG_MACH_MP, 3)
+#define        TRACE_MP_CPUS_CALL_NOBUF        MACHDBG_CODE(DBG_MACH_MP, 4)

 
 

-void           slave_boot_init(void);
+#define ABS(v)         (((v) > 0)?(v):-(v))

 
 

-#if MACH_KDB
-static void    mp_kdb_wait(void);
-volatile boolean_t     mp_kdb_trap = FALSE;
-volatile long  mp_kdb_ncpus = 0;
-#endif
+void           slave_boot_init(void);
+void           i386_cpu_IPI(int cpu);

 
 

-static void    mp_kdp_wait(boolean_t flush);
+static void    mp_kdp_wait(boolean_t flush, boolean_t isNMI);

 static void    mp_rendezvous_action(void);
 static void    mp_broadcast_action(void);
 
 static boolean_t       cpu_signal_pending(int cpu, mp_event_t event);
 static void    mp_rendezvous_action(void);
 static void    mp_broadcast_action(void);
 
 static boolean_t       cpu_signal_pending(int cpu, mp_event_t event);

-static int             cpu_signal_handler(x86_saved_state_t *regs);

 static int             NMIInterruptHandler(x86_saved_state_t *regs);
 
 static int             NMIInterruptHandler(x86_saved_state_t *regs);
 

-boolean_t      smp_initialized = FALSE;
+boolean_t              smp_initialized = FALSE;
+uint32_t               TSC_sync_margin = 0xFFF;

 volatile boolean_t     force_immediate_debugger_NMI = FALSE;
 volatile boolean_t     pmap_tlb_flush_timeout = FALSE;
 volatile boolean_t     force_immediate_debugger_NMI = FALSE;
 volatile boolean_t     pmap_tlb_flush_timeout = FALSE;

-

 decl_simple_lock_data(,mp_kdp_lock);
 
 decl_simple_lock_data(,mp_kdp_lock);
 

-decl_mutex_data(static, mp_cpu_boot_lock);
+decl_lck_mtx_data(static, mp_cpu_boot_lock);
+lck_mtx_ext_t  mp_cpu_boot_lock_ext;

 
 /* Variables needed for MP rendezvous. */
 decl_simple_lock_data(,mp_rv_lock);
 
 /* Variables needed for MP rendezvous. */
 decl_simple_lock_data(,mp_rv_lock);

-static void            (*mp_rv_setup_func)(void *arg);
-static void            (*mp_rv_action_func)(void *arg);
-static void            (*mp_rv_teardown_func)(void *arg);
-static void            *mp_rv_func_arg;
-static int             mp_rv_ncpus;
+static void    (*mp_rv_setup_func)(void *arg);
+static void    (*mp_rv_action_func)(void *arg);
+static void    (*mp_rv_teardown_func)(void *arg);
+static void    *mp_rv_func_arg;
+static volatile int    mp_rv_ncpus;

                        /* Cache-aligned barriers: */
 static volatile long   mp_rv_entry    __attribute__((aligned(64)));
 static volatile long   mp_rv_exit     __attribute__((aligned(64)));
 static volatile long   mp_rv_complete __attribute__((aligned(64)));
 
                        /* Cache-aligned barriers: */
 static volatile long   mp_rv_entry    __attribute__((aligned(64)));
 static volatile long   mp_rv_exit     __attribute__((aligned(64)));
 static volatile long   mp_rv_complete __attribute__((aligned(64)));
 

+volatile       uint64_t        debugger_entry_time;
+volatile       uint64_t        debugger_exit_time;
+#if MACH_KDP
+#include <kdp/kdp.h>
+extern int kdp_snapshot;
+static struct _kdp_xcpu_call_func {
+       kdp_x86_xcpu_func_t func;
+       void     *arg0, *arg1;
+       volatile long     ret;
+       volatile uint16_t cpu;
+} kdp_xcpu_call_func = {
+       .cpu  = KDP_XCPU_NONE
+};
+
+#endif
+

 /* Variables needed for MP broadcast. */
 static void        (*mp_bc_action_func)(void *arg);
 static void        *mp_bc_func_arg;
 static int             mp_bc_ncpus;
 static volatile long   mp_bc_count;
 /* Variables needed for MP broadcast. */
 static void        (*mp_bc_action_func)(void *arg);
 static void        *mp_bc_func_arg;
 static int             mp_bc_ncpus;
 static volatile long   mp_bc_count;

-decl_mutex_data(static, mp_bc_lock);
+decl_lck_mtx_data(static, mp_bc_lock);
+lck_mtx_ext_t  mp_bc_lock_ext;

 static volatile int    debugger_cpu = -1;
 static volatile int    debugger_cpu = -1;

+volatile long NMIPI_acks = 0;

 
 

+static void    mp_cpus_call_init(void); 
+static void    mp_cpus_call_cpu_init(void); 

 static void    mp_cpus_call_action(void); 
 static void    mp_cpus_call_action(void); 

+static void    mp_call_PM(void);
+
+char           mp_slave_stack[PAGE_SIZE] __attribute__((aligned(PAGE_SIZE))); // Temp stack for slave init
+
+/* PAL-related routines */
+boolean_t i386_smp_init(int nmi_vector, i386_intr_func_t nmi_handler, 
+               int ipi_vector, i386_intr_func_t ipi_handler);
+void i386_start_cpu(int lapic_id, int cpu_num);
+void i386_send_NMI(int cpu);

 
 #if GPROF
 /*
 
 #if GPROF
 /*


@@ -165,123 +185,97 @@ struct profile_vars *_profile_vars_cpus[MAX_CPUS] = { &_profile_vars };
 #define GPROF_INIT()
 #endif /* GPROF */
 
 #define GPROF_INIT()
 #endif /* GPROF */
 

+static lck_grp_t       smp_lck_grp;
+static lck_grp_attr_t  smp_lck_grp_attr;
+
+#define NUM_CPU_WARM_CALLS     20
+struct timer_call      cpu_warm_call_arr[NUM_CPU_WARM_CALLS];
+queue_head_t           cpu_warm_call_list;
+decl_simple_lock_data(static, cpu_warm_lock);
+
+typedef struct cpu_warm_data {
+       timer_call_t    cwd_call;
+       uint64_t        cwd_deadline;
+       int             cwd_result;
+} *cpu_warm_data_t;
+
+static void            cpu_prewarm_init(void);
+static void            cpu_warm_timer_call_func(call_entry_param_t p0, call_entry_param_t p1);
+static void            _cpu_warm_setup(void *arg);
+static timer_call_t    grab_warm_timer_call(void);
+static void            free_warm_timer_call(timer_call_t call);
+

 void
 smp_init(void)
 {
        simple_lock_init(&mp_kdp_lock, 0);
        simple_lock_init(&mp_rv_lock, 0);
 void
 smp_init(void)
 {
        simple_lock_init(&mp_kdp_lock, 0);
        simple_lock_init(&mp_rv_lock, 0);

-       mutex_init(&mp_cpu_boot_lock, 0);
-       mutex_init(&mp_bc_lock, 0);
+       lck_grp_attr_setdefault(&smp_lck_grp_attr);
+       lck_grp_init(&smp_lck_grp, "i386_smp", &smp_lck_grp_attr);
+       lck_mtx_init_ext(&mp_cpu_boot_lock, &mp_cpu_boot_lock_ext, &smp_lck_grp, LCK_ATTR_NULL);
+       lck_mtx_init_ext(&mp_bc_lock, &mp_bc_lock_ext, &smp_lck_grp, LCK_ATTR_NULL);

        console_init();
 
        console_init();
 

-       /* Local APIC? */
-       if (!lapic_probe())
+       if(!i386_smp_init(LAPIC_NMI_INTERRUPT, NMIInterruptHandler, 
+                               LAPIC_VECTOR(INTERPROCESSOR), cpu_signal_handler))

                return;
 
                return;
 

-       lapic_init();
-       lapic_configure();
-       lapic_set_intr_func(LAPIC_NMI_INTERRUPT,  NMIInterruptHandler);
-       lapic_set_intr_func(LAPIC_VECTOR(INTERPROCESSOR), cpu_signal_handler);
-

        cpu_thread_init();
 
        GPROF_INIT();
        DBGLOG_CPU_INIT(master_cpu);
 
        cpu_thread_init();
 
        GPROF_INIT();
        DBGLOG_CPU_INIT(master_cpu);
 

-       slave_boot_init();
+       mp_cpus_call_init();
+       mp_cpus_call_cpu_init();

 
 

+       if (PE_parse_boot_argn("TSC_sync_margin",
+                                       &TSC_sync_margin, sizeof(TSC_sync_margin))) {
+               kprintf("TSC sync Margin 0x%x\n", TSC_sync_margin);
+       } else if (cpuid_vmm_present()) {
+               kprintf("TSC sync margin disabled\n");
+               TSC_sync_margin = 0;
+       }

        smp_initialized = TRUE;
 
        smp_initialized = TRUE;
 

+       cpu_prewarm_init();
+

        return;
 }
 
        return;
 }
 

+typedef struct {
+       int                     target_cpu;
+       int                     target_lapic;
+       int                     starter_cpu;
+} processor_start_info_t;
+static processor_start_info_t  start_info        __attribute__((aligned(64)));
+
+/* 
+ * Cache-alignment is to avoid cross-cpu false-sharing interference.
+ */
+static volatile long           tsc_entry_barrier __attribute__((aligned(64)));
+static volatile long           tsc_exit_barrier  __attribute__((aligned(64)));
+static volatile uint64_t       tsc_target        __attribute__((aligned(64)));
+

 /*
  * Poll a CPU to see when it has marked itself as running.
  */
 static void
 mp_wait_for_cpu_up(int slot_num, unsigned int iters, unsigned int usecdelay)
 {
 /*
  * Poll a CPU to see when it has marked itself as running.
  */
 static void
 mp_wait_for_cpu_up(int slot_num, unsigned int iters, unsigned int usecdelay)
 {

-       while (iters-- > 0) {
+       while (iters-- > 0) {

                if (cpu_datap(slot_num)->cpu_running)
                if (cpu_datap(slot_num)->cpu_running)

-                       break;
+                       break;

                delay(usecdelay);
        }
 }
 
                delay(usecdelay);
        }
 }
 

-kern_return_t
-intel_startCPU(
-       int     slot_num)
-{
-
-       int     i = 1000;
-       int     lapic = cpu_to_lapic[slot_num];
-
-       assert(lapic != -1);
-
-       DBGLOG_CPU_INIT(slot_num);
-
-       DBG("intel_startCPU(%d) lapic_id=%d\n", slot_num, lapic);
-       DBG("IdlePTD(%p): 0x%x\n", &IdlePTD, (int) IdlePTD);
-
-       /*
-        * Initialize (or re-initialize) the descriptor tables for this cpu.
-        * Propagate processor mode to slave.
-        */
-       if (cpu_mode_is64bit())
-               cpu_desc_init64(cpu_datap(slot_num), FALSE);
-       else
-               cpu_desc_init(cpu_datap(slot_num), FALSE);
-
-       /* Serialize use of the slave boot stack. */
-       mutex_lock(&mp_cpu_boot_lock);
-
-       mp_disable_preemption();
-       if (slot_num == get_cpu_number()) {
-               mp_enable_preemption();
-               mutex_unlock(&mp_cpu_boot_lock);
-               return KERN_SUCCESS;
-       }
-
-       LAPIC_WRITE(ICRD, lapic << LAPIC_ICRD_DEST_SHIFT);
-       LAPIC_WRITE(ICR, LAPIC_ICR_DM_INIT);
-       delay(10000);
-
-       LAPIC_WRITE(ICRD, lapic << LAPIC_ICRD_DEST_SHIFT);
-       LAPIC_WRITE(ICR, LAPIC_ICR_DM_STARTUP|(MP_BOOT>>12));
-       delay(200);
-
-       LAPIC_WRITE(ICRD, lapic << LAPIC_ICRD_DEST_SHIFT);
-       LAPIC_WRITE(ICR, LAPIC_ICR_DM_STARTUP|(MP_BOOT>>12));
-       delay(200);
-
-#ifdef POSTCODE_DELAY
-       /* Wait much longer if postcodes are displayed for a delay period. */
-       i *= 10000;
-#endif
-       mp_wait_for_cpu_up(slot_num, i, 10000);
-
-       mp_enable_preemption();
-       mutex_unlock(&mp_cpu_boot_lock);
-
-       if (!cpu_datap(slot_num)->cpu_running) {
-               kprintf("Failed to start CPU %02d\n", slot_num);
-               printf("Failed to start CPU %02d, rebooting...\n", slot_num);
-               delay(1000000);
-               cpu_shutdown();
-               return KERN_SUCCESS;
-       } else {
-               kprintf("Started cpu %d (lapic id %08x)\n", slot_num, lapic);
-               return KERN_SUCCESS;
-       }
-}
-

 /*
  * Quickly bring a CPU back online which has been halted.
  */
 kern_return_t
 intel_startCPU_fast(int slot_num)
 {
 /*
  * Quickly bring a CPU back online which has been halted.
  */
 kern_return_t
 intel_startCPU_fast(int slot_num)
 {

-       kern_return_t   rc;
+       kern_return_t   rc;

 
        /*
         * Try to perform a fast restart
 
        /*
         * Try to perform a fast restart


@@ -303,6 +297,7 @@ intel_startCPU_fast(int slot_num)
         * longer than a full restart would require so it should be more
         * than long enough.
         */
         * longer than a full restart would require so it should be more
         * than long enough.
         */

+

        mp_wait_for_cpu_up(slot_num, 30000, 1);
        mp_enable_preemption();
 
        mp_wait_for_cpu_up(slot_num, 30000, 1);
        mp_enable_preemption();
 


@@ -312,51 +307,141 @@ intel_startCPU_fast(int slot_num)
         */
        if (cpu_datap(slot_num)->cpu_running)
                return(KERN_SUCCESS);
         */
        if (cpu_datap(slot_num)->cpu_running)
                return(KERN_SUCCESS);

-       else
+       else

                return(KERN_FAILURE);
 }
 
                return(KERN_FAILURE);
 }
 

-extern char    slave_boot_base[];
-extern char    slave_boot_end[];
-extern void    slave_pstart(void);
+static void
+started_cpu(void)
+{
+       /* Here on the started cpu with cpu_running set TRUE */
+
+       if (TSC_sync_margin &&
+           start_info.target_cpu == cpu_number()) {
+               /*
+                * I've just started-up, synchronize again with the starter cpu
+                * and then snap my TSC.
+                */
+               tsc_target   = 0;
+               atomic_decl(&tsc_entry_barrier, 1);
+               while (tsc_entry_barrier != 0)
+                       ;       /* spin for starter and target at barrier */
+               tsc_target = rdtsc64();
+               atomic_decl(&tsc_exit_barrier, 1);
+       }
+}

 
 

-void
-slave_boot_init(void)
+static void
+start_cpu(void *arg)

 {
 {

-       DBG("V(slave_boot_base)=%p P(slave_boot_base)=%p MP_BOOT=%p sz=0x%x\n",
-               slave_boot_base,
-               kvtophys((vm_offset_t) slave_boot_base),
-               MP_BOOT,
-               slave_boot_end-slave_boot_base);
+       int                     i = 1000;
+       processor_start_info_t  *psip = (processor_start_info_t *) arg;

 
 

-       /*
-        * Copy the boot entry code to the real-mode vector area MP_BOOT.
-        * This is in page 1 which has been reserved for this purpose by
-        * machine_startup() from the boot processor.
-        * The slave boot code is responsible for switching to protected
-        * mode and then jumping to the common startup, _start().
-        */
-       bcopy_phys(kvtophys((vm_offset_t) slave_boot_base),
-                  (addr64_t) MP_BOOT,
-                  slave_boot_end-slave_boot_base);
+       /* Ignore this if the current processor is not the starter */
+       if (cpu_number() != psip->starter_cpu)
+               return;
+
+       i386_start_cpu(psip->target_lapic, psip->target_cpu);
+
+#ifdef POSTCODE_DELAY
+       /* Wait much longer if postcodes are displayed for a delay period. */
+       i *= 10000;
+#endif
+       mp_wait_for_cpu_up(psip->target_cpu, i*100, 100);
+       if (TSC_sync_margin &&
+           cpu_datap(psip->target_cpu)->cpu_running) {
+               /*
+                * Compare the TSC from the started processor with ours.
+                * Report and log/panic if it diverges by more than
+                * TSC_sync_margin (TSC_SYNC_MARGIN) ticks. This margin
+                * can be overriden by boot-arg (with 0 meaning no checking).
+                */
+               uint64_t        tsc_starter;
+               int64_t         tsc_delta;
+               atomic_decl(&tsc_entry_barrier, 1);
+               while (tsc_entry_barrier != 0)
+                       ;       /* spin for both processors at barrier */
+               tsc_starter = rdtsc64();
+               atomic_decl(&tsc_exit_barrier, 1);
+               while (tsc_exit_barrier != 0)
+                       ;       /* spin for target to store its TSC */
+               tsc_delta = tsc_target - tsc_starter;
+               kprintf("TSC sync for cpu %d: 0x%016llx delta 0x%llx (%lld)\n",
+                       psip->target_cpu, tsc_target, tsc_delta, tsc_delta);
+               if (ABS(tsc_delta) > (int64_t) TSC_sync_margin) { 
+#if DEBUG
+                       panic(
+#else
+                       printf(
+#endif
+                               "Unsynchronized  TSC for cpu %d: "
+                                       "0x%016llx, delta 0x%llx\n",
+                               psip->target_cpu, tsc_target, tsc_delta);
+               }
+       }
+}
+
+kern_return_t
+intel_startCPU(
+       int     slot_num)
+{
+       int             lapic = cpu_to_lapic[slot_num];
+       boolean_t       istate;
+
+       assert(lapic != -1);
+
+       DBGLOG_CPU_INIT(slot_num);
+
+       DBG("intel_startCPU(%d) lapic_id=%d\n", slot_num, lapic);
+       DBG("IdlePTD(%p): 0x%x\n", &IdlePTD, (int) (uintptr_t)IdlePTD);

 
        /*
 
        /*

-        * Zero a stack area above the boot code.
+        * Initialize (or re-initialize) the descriptor tables for this cpu.
+        * Propagate processor mode to slave.

         */
         */

-       DBG("bzero_phys 0x%x sz 0x%x\n",MP_BOOTSTACK+MP_BOOT-0x400, 0x400);
-       bzero_phys((addr64_t)MP_BOOTSTACK+MP_BOOT-0x400, 0x400);
+       if (cpu_mode_is64bit())
+               cpu_desc_init64(cpu_datap(slot_num));
+       else
+               cpu_desc_init(cpu_datap(slot_num));
+
+       /* Serialize use of the slave boot stack, etc. */
+       lck_mtx_lock(&mp_cpu_boot_lock);
+
+       istate = ml_set_interrupts_enabled(FALSE);
+       if (slot_num == get_cpu_number()) {
+               ml_set_interrupts_enabled(istate);
+               lck_mtx_unlock(&mp_cpu_boot_lock);
+               return KERN_SUCCESS;
+       }
+
+       start_info.starter_cpu  = cpu_number();
+       start_info.target_cpu   = slot_num;
+       start_info.target_lapic = lapic;
+       tsc_entry_barrier = 2;
+       tsc_exit_barrier = 2;

 
        /*
 
        /*

-        * Set the location at the base of the stack to point to the
-        * common startup entry.
+        * Perform the processor startup sequence with all running
+        * processors rendezvous'ed. This is required during periods when
+        * the cache-disable bit is set for MTRR/PAT initialization.

         */
         */

-       DBG("writing 0x%x at phys 0x%x\n",
-               kvtophys((vm_offset_t) &slave_pstart), MP_MACH_START+MP_BOOT);
-       ml_phys_write_word(MP_MACH_START+MP_BOOT,
-                          (unsigned int)kvtophys((vm_offset_t) &slave_pstart));
-       
-       /* Flush caches */
-       __asm__("wbinvd");
+       mp_rendezvous_no_intrs(start_cpu, (void *) &start_info);
+
+       start_info.target_cpu = 0;
+
+       ml_set_interrupts_enabled(istate);
+       lck_mtx_unlock(&mp_cpu_boot_lock);
+
+       if (!cpu_datap(slot_num)->cpu_running) {
+               kprintf("Failed to start CPU %02d\n", slot_num);
+               printf("Failed to start CPU %02d, rebooting...\n", slot_num);
+               delay(1000000);
+               halt_cpu();
+               return KERN_SUCCESS;
+       } else {
+               kprintf("Started cpu %d (lapic id %08x)\n", slot_num, lapic);
+               return KERN_SUCCESS;
+       }

 }
 
 #if    MP_DEBUG
 }
 
 #if    MP_DEBUG


@@ -372,20 +457,19 @@ cpu_signal_handler(x86_saved_state_t *regs)
 {
        int             my_cpu;
        volatile int    *my_word;
 {
        int             my_cpu;
        volatile int    *my_word;

-#if    MACH_KDB && MACH_ASSERT
-       int             i=100;
-#endif /* MACH_KDB && MACH_ASSERT */

 
 

-       mp_disable_preemption();
+       SCHED_STATS_IPI(current_processor());

 
        my_cpu = cpu_number();
 
        my_cpu = cpu_number();

-       my_word = &current_cpu_datap()->cpu_signals;
+       my_word = &cpu_data_ptr[my_cpu]->cpu_signals;
+       /* Store the initial set of signals for diagnostics. New
+        * signals could arrive while these are being processed
+        * so it's no more than a hint.
+        */
+
+       cpu_data_ptr[my_cpu]->cpu_prior_signals = *my_word;

 
        do {
 
        do {

-#if    MACH_KDB && MACH_ASSERT
-               if (i-- <= 0)
-                   Debugger("cpu_signal_handler: signals did not clear");
-#endif /* MACH_KDB && MACH_ASSERT */

 #if    MACH_KDP
                if (i_bit(MP_KDP, my_word)) {
                        DBGLOG(cpu_handle,my_cpu,MP_KDP);
 #if    MACH_KDP
                if (i_bit(MP_KDP, my_word)) {
                        DBGLOG(cpu_handle,my_cpu,MP_KDP);


@@ -394,10 +478,13 @@ cpu_signal_handler(x86_saved_state_t *regs)
  * current thread's stack (if any) is synchronized with the
  * context at the moment of the interrupt, to facilitate
  * access through the debugger.
  * current thread's stack (if any) is synchronized with the
  * context at the moment of the interrupt, to facilitate
  * access through the debugger.

- * XXX 64-bit state?

  */
  */

-                       sync_iss_to_iks(saved_state32(regs));
-                       mp_kdp_wait(TRUE);
+                       sync_iss_to_iks(regs);
+                       if (pmsafe_debug && !kdp_snapshot)
+                               pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
+                       mp_kdp_wait(TRUE, FALSE);
+                       if (pmsafe_debug && !kdp_snapshot)
+                               pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_NORMAL);

                } else
 #endif /* MACH_KDP */
                if (i_bit(MP_TLB_FLUSH, my_word)) {
                } else
 #endif /* MACH_KDP */
                if (i_bit(MP_TLB_FLUSH, my_word)) {


@@ -408,14 +495,6 @@ cpu_signal_handler(x86_saved_state_t *regs)
                        DBGLOG(cpu_handle,my_cpu,MP_AST);
                        i_bit_clear(MP_AST, my_word);
                        ast_check(cpu_to_processor(my_cpu));
                        DBGLOG(cpu_handle,my_cpu,MP_AST);
                        i_bit_clear(MP_AST, my_word);
                        ast_check(cpu_to_processor(my_cpu));

-#if    MACH_KDB
-               } else if (i_bit(MP_KDB, my_word)) {
-
-                       i_bit_clear(MP_KDB, my_word);
-                       current_cpu_datap()->cpu_kdb_is_slave++;
-                       mp_kdb_wait();
-                       current_cpu_datap()->cpu_kdb_is_slave--;
-#endif /* MACH_KDB */

                } else if (i_bit(MP_RENDEZVOUS, my_word)) {
                        DBGLOG(cpu_handle,my_cpu,MP_RENDEZVOUS);
                        i_bit_clear(MP_RENDEZVOUS, my_word);
                } else if (i_bit(MP_RENDEZVOUS, my_word)) {
                        DBGLOG(cpu_handle,my_cpu,MP_RENDEZVOUS);
                        i_bit_clear(MP_RENDEZVOUS, my_word);


@@ -432,11 +511,13 @@ cpu_signal_handler(x86_saved_state_t *regs)
                        DBGLOG(cpu_handle,my_cpu,MP_CALL);
                        i_bit_clear(MP_CALL, my_word);
                        mp_cpus_call_action();
                        DBGLOG(cpu_handle,my_cpu,MP_CALL);
                        i_bit_clear(MP_CALL, my_word);
                        mp_cpus_call_action();

+               } else if (i_bit(MP_CALL_PM, my_word)) {
+                       DBGLOG(cpu_handle,my_cpu,MP_CALL_PM);
+                       i_bit_clear(MP_CALL_PM, my_word);
+                       mp_call_PM();

                }
        } while (*my_word);
 
                }
        } while (*my_word);
 

-       mp_enable_preemption();
-

        return 0;
 }
 
        return 0;
 }
 


@@ -444,75 +525,47 @@ static int
 NMIInterruptHandler(x86_saved_state_t *regs)
 {
        void    *stackptr;
 NMIInterruptHandler(x86_saved_state_t *regs)
 {
        void    *stackptr;

-       
+
+       if (panic_active() && !panicDebugging) {
+               if (pmsafe_debug)
+                       pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
+               for(;;)
+                       cpu_pause();
+       }
+
+       atomic_incl(&NMIPI_acks, 1);

        sync_iss_to_iks_unconditionally(regs);
        sync_iss_to_iks_unconditionally(regs);

+#if defined (__i386__)

        __asm__ volatile("movl %%ebp, %0" : "=m" (stackptr));
        __asm__ volatile("movl %%ebp, %0" : "=m" (stackptr));

+#elif defined (__x86_64__)
+       __asm__ volatile("movq %%rbp, %0" : "=m" (stackptr));
+#endif

 
        if (cpu_number() == debugger_cpu)
                        goto NMExit;
 
 
        if (cpu_number() == debugger_cpu)
                        goto NMExit;
 

-       if (pmap_tlb_flush_timeout == TRUE && current_cpu_datap()->cpu_tlb_invalid) {
+       if (spinlock_timed_out) {
+               char pstr[192];
+               snprintf(&pstr[0], sizeof(pstr), "Panic(CPU %d): NMIPI for spinlock acquisition timeout, spinlock: %p, spinlock owner: %p, current_thread: %p, spinlock_owner_cpu: 0x%x\n", cpu_number(), spinlock_timed_out, (void *) spinlock_timed_out->interlock.lock_data, current_thread(), spinlock_owner_cpu);
+               panic_i386_backtrace(stackptr, 64, &pstr[0], TRUE, regs);
+       } else if (pmap_tlb_flush_timeout == TRUE) {

                char pstr[128];
                char pstr[128];

-               snprintf(&pstr[0], sizeof(pstr), "Panic(CPU %d): Unresponsive processor\n", cpu_number());
-               panic_i386_backtrace(stackptr, 10, &pstr[0], TRUE, regs);
-               panic_io_port_read();
-               mca_check_save();
-               if (pmsafe_debug)
-                       pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
-               for(;;) {
-                       cpu_pause();
-               }
+               snprintf(&pstr[0], sizeof(pstr), "Panic(CPU %d): Unresponsive processor (this CPU did not acknowledge interrupts) TLB state:0x%x\n", cpu_number(), current_cpu_datap()->cpu_tlb_invalid);
+               panic_i386_backtrace(stackptr, 48, &pstr[0], TRUE, regs);

        }
        }

-       mp_kdp_wait(FALSE);
+
+#if MACH_KDP
+       if (pmsafe_debug && !kdp_snapshot)
+               pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
+       current_cpu_datap()->cpu_NMI_acknowledged = TRUE;
+       mp_kdp_wait(FALSE, pmap_tlb_flush_timeout || spinlock_timed_out || panic_active());
+       if (pmsafe_debug && !kdp_snapshot)
+               pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_NORMAL);
+#endif

 NMExit:        
        return 1;
 }
 
 NMExit:        
        return 1;
 }
 

-#ifdef MP_DEBUG
-int            max_lock_loops = 1000000;
-int            trappedalready = 0;     /* (BRINGUP */
-#endif /* MP_DEBUG */
-
-static void
-i386_cpu_IPI(int cpu)
-{
-       boolean_t       state;
-       
-#ifdef MP_DEBUG
-       if(cpu_datap(cpu)->cpu_signals & 6) {   /* (BRINGUP) */
-               kprintf("i386_cpu_IPI: sending enter debugger signal (%08X) to cpu %d\n", cpu_datap(cpu)->cpu_signals, cpu);
-       }
-#endif /* MP_DEBUG */
-
-#if MACH_KDB
-#ifdef MP_DEBUG
-       if(!trappedalready && (cpu_datap(cpu)->cpu_signals & 6)) {      /* (BRINGUP) */
-               if(kdb_cpu != cpu_number()) {
-                       trappedalready = 1;
-                       panic("i386_cpu_IPI: sending enter debugger signal (%08X) to cpu %d and I do not own debugger, owner = %08X\n", 
-                               cpu_datap(cpu)->cpu_signals, cpu, kdb_cpu);
-               }
-       }
-#endif /* MP_DEBUG */
-#endif
-
-       /* Wait for previous interrupt to be delivered... */
-#ifdef MP_DEBUG
-       int     pending_busy_count = 0;
-       while (LAPIC_READ(ICR) & LAPIC_ICR_DS_PENDING) {
-               if (++pending_busy_count > max_lock_loops)
-                       panic("i386_cpu_IPI() deadlock\n");
-#else
-       while (LAPIC_READ(ICR) & LAPIC_ICR_DS_PENDING) {
-#endif /* MP_DEBUG */
-               cpu_pause();
-       }
-
-       state = ml_set_interrupts_enabled(FALSE);
-       LAPIC_WRITE(ICRD, cpu_to_lapic[cpu] << LAPIC_ICRD_DEST_SHIFT);
-       LAPIC_WRITE(ICR, LAPIC_VECTOR(INTERPROCESSOR) | LAPIC_ICR_DM_FIXED);
-       (void) ml_set_interrupts_enabled(state);
-}

 
 /*
  * cpu_interrupt is really just to be used by the scheduler to
 
 /*
  * cpu_interrupt is really just to be used by the scheduler to


@@ -522,10 +575,15 @@ i386_cpu_IPI(int cpu)
 void
 cpu_interrupt(int cpu)
 {
 void
 cpu_interrupt(int cpu)
 {

+       boolean_t did_IPI = FALSE;
+

        if (smp_initialized
            && pmCPUExitIdle(cpu_datap(cpu))) {
                i386_cpu_IPI(cpu);
        if (smp_initialized
            && pmCPUExitIdle(cpu_datap(cpu))) {
                i386_cpu_IPI(cpu);

+               did_IPI = TRUE;

        }
        }

+
+       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_REMOTE_AST), cpu, did_IPI, 0, 0, 0);

 }
 
 /*
 }
 
 /*


@@ -534,20 +592,41 @@ cpu_interrupt(int cpu)
 void
 cpu_NMI_interrupt(int cpu)
 {
 void
 cpu_NMI_interrupt(int cpu)
 {

-       boolean_t       state;
-

        if (smp_initialized) {
        if (smp_initialized) {

-               state = ml_set_interrupts_enabled(FALSE);
-/* Program the interrupt command register */
-               LAPIC_WRITE(ICRD, cpu_to_lapic[cpu] << LAPIC_ICRD_DEST_SHIFT);
-/* The vector is ignored in this case--the target CPU will enter on the
- * NMI vector.
- */
-               LAPIC_WRITE(ICR, LAPIC_VECTOR(INTERPROCESSOR)|LAPIC_ICR_DM_NMI);
-               (void) ml_set_interrupts_enabled(state);
+               i386_send_NMI(cpu);
+       }
+}
+
+static void    (* volatile mp_PM_func)(void) = NULL;
+
+static void
+mp_call_PM(void)
+{
+       assert(!ml_get_interrupts_enabled());
+
+       if (mp_PM_func != NULL)
+               mp_PM_func();
+}
+
+void
+cpu_PM_interrupt(int cpu)
+{
+       assert(!ml_get_interrupts_enabled());
+
+       if (mp_PM_func != NULL) {
+               if (cpu == cpu_number())
+                       mp_PM_func();
+               else
+                       i386_signal_cpu(cpu, MP_CALL_PM, ASYNC);

        }
 }
 
        }
 }
 

+void
+PM_interrupt_register(void (*fn)(void))
+{
+       mp_PM_func = fn;
+}
+

 void
 i386_signal_cpu(int cpu, mp_event_t event, mp_sync_t mode)
 {
 void
 i386_signal_cpu(int cpu, mp_event_t event, mp_sync_t mode)
 {


@@ -559,7 +638,7 @@ i386_signal_cpu(int cpu, mp_event_t event, mp_sync_t mode)
                return;
 
        if (event == MP_TLB_FLUSH)
                return;
 
        if (event == MP_TLB_FLUSH)

-               KERNEL_DEBUG(0xef800020 | DBG_FUNC_START, cpu, 0, 0, 0, 0);
+               KERNEL_DEBUG(TRACE_MP_TLB_FLUSH | DBG_FUNC_START, cpu, 0, 0, 0, 0);

 
        DBGLOG(cpu_signal, cpu, event);
        
 
        DBGLOG(cpu_signal, cpu, event);
        


@@ -578,7 +657,7 @@ i386_signal_cpu(int cpu, mp_event_t event, mp_sync_t mode)
                }
        }
        if (event == MP_TLB_FLUSH)
                }
        }
        if (event == MP_TLB_FLUSH)

-               KERNEL_DEBUG(0xef800020 | DBG_FUNC_END, cpu, 0, 0, 0, 0);
+               KERNEL_DEBUG(TRACE_MP_TLB_FLUSH | DBG_FUNC_END, cpu, 0, 0, 0, 0);

 }
 
 /*
 }
 
 /*


@@ -591,7 +670,7 @@ i386_signal_cpus(mp_event_t event, mp_sync_t mode)
        unsigned int    cpu;
        unsigned int    my_cpu = cpu_number();
 
        unsigned int    cpu;
        unsigned int    my_cpu = cpu_number();
 

-       assert(hw_lock_held(&x86_topo_lock));
+       assert(hw_lock_held((hw_lock_t)&x86_topo_lock));

 
        for (cpu = 0; cpu < real_ncpus; cpu++) {
                if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
 
        for (cpu = 0; cpu < real_ncpus; cpu++) {
                if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)


@@ -610,7 +689,7 @@ i386_active_cpus(void)
        unsigned int    cpu;
        unsigned int    ncpus = 0;
 
        unsigned int    cpu;
        unsigned int    ncpus = 0;
 

-       assert(hw_lock_held(&x86_topo_lock));
+       assert(hw_lock_held((hw_lock_t)&x86_topo_lock));

 
        for (cpu = 0; cpu < real_ncpus; cpu++) {
                if (cpu_datap(cpu)->cpu_running)
 
        for (cpu = 0; cpu < real_ncpus; cpu++) {
                if (cpu_datap(cpu)->cpu_running)


@@ -652,9 +731,11 @@ mp_rendezvous_action(void)
                        handle_pending_TLB_flushes();
                cpu_pause();
        }
                        handle_pending_TLB_flushes();
                cpu_pause();
        }

+

        /* action function */
        if (mp_rv_action_func != NULL)
                mp_rv_action_func(mp_rv_func_arg);
        /* action function */
        if (mp_rv_action_func != NULL)
                mp_rv_action_func(mp_rv_func_arg);

+

        /* spin on exit rendezvous */
        atomic_incl(&mp_rv_exit, 1);
        while (mp_rv_exit < mp_rv_ncpus) {
        /* spin on exit rendezvous */
        atomic_incl(&mp_rv_exit, 1);
        while (mp_rv_exit < mp_rv_ncpus) {


@@ -771,38 +852,175 @@ mp_rendezvous_no_intrs(
                      arg);     
 }
 
                      arg);     
 }
 

-void
-handle_pending_TLB_flushes(void)
+
+typedef struct {
+       queue_chain_t   link;                   /* queue linkage */
+       void            (*func)(void *,void *); /* routine to call */
+       void            *arg0;                  /* routine's 1st arg */
+       void            *arg1;                  /* routine's 2nd arg */
+       volatile long   *countp;                /* completion counter */
+} mp_call_t;
+
+
+typedef struct {
+       queue_head_t            queue;
+       decl_simple_lock_data(, lock);
+} mp_call_queue_t;
+#define MP_CPUS_CALL_BUFS_PER_CPU      MAX_CPUS
+static mp_call_queue_t mp_cpus_call_freelist;
+static mp_call_queue_t mp_cpus_call_head[MAX_CPUS];
+
+static inline boolean_t
+mp_call_head_lock(mp_call_queue_t *cqp)
+{
+       boolean_t       intrs_enabled;
+
+       intrs_enabled = ml_set_interrupts_enabled(FALSE);
+       simple_lock(&cqp->lock);
+
+       return intrs_enabled;
+}
+
+static inline boolean_t
+mp_call_head_is_locked(mp_call_queue_t *cqp)
+{
+       return !ml_get_interrupts_enabled() &&
+               hw_lock_held((hw_lock_t)&cqp->lock);
+}
+
+static inline void
+mp_call_head_unlock(mp_call_queue_t *cqp, boolean_t intrs_enabled)
+{
+       simple_unlock(&cqp->lock);
+       ml_set_interrupts_enabled(intrs_enabled);
+}
+
+static inline mp_call_t *
+mp_call_alloc(void)
+{
+       mp_call_t       *callp = NULL;
+       boolean_t       intrs_enabled;
+       mp_call_queue_t *cqp = &mp_cpus_call_freelist;
+
+       intrs_enabled = mp_call_head_lock(cqp);
+       if (!queue_empty(&cqp->queue))
+               queue_remove_first(&cqp->queue, callp, typeof(callp), link);
+       mp_call_head_unlock(cqp, intrs_enabled);
+
+       return callp;
+}
+
+static inline void
+mp_call_free(mp_call_t *callp)
+{
+       boolean_t       intrs_enabled;
+       mp_call_queue_t *cqp = &mp_cpus_call_freelist;
+
+       intrs_enabled = mp_call_head_lock(cqp);
+       queue_enter_first(&cqp->queue, callp, typeof(callp), link);
+       mp_call_head_unlock(cqp, intrs_enabled);
+}
+
+static inline mp_call_t *
+mp_call_dequeue_locked(mp_call_queue_t *cqp)

 {
 {

-       volatile int    *my_word = &current_cpu_datap()->cpu_signals;
+       mp_call_t       *callp = NULL;

 
 

-       if (i_bit(MP_TLB_FLUSH, my_word)) {
-               DBGLOG(cpu_handle, cpu_number(), MP_TLB_FLUSH);
-               i_bit_clear(MP_TLB_FLUSH, my_word);
-               pmap_update_interrupt();
+       assert(mp_call_head_is_locked(cqp));
+       if (!queue_empty(&cqp->queue))
+               queue_remove_first(&cqp->queue, callp, typeof(callp), link);
+       return callp;
+}
+
+static inline void
+mp_call_enqueue_locked(
+       mp_call_queue_t *cqp,
+       mp_call_t       *callp)
+{
+       queue_enter(&cqp->queue, callp, typeof(callp), link);
+}
+
+/* Called on the boot processor to initialize global structures */
+static void
+mp_cpus_call_init(void)
+{
+       mp_call_queue_t *cqp = &mp_cpus_call_freelist;
+
+       DBG("mp_cpus_call_init()\n");
+       simple_lock_init(&cqp->lock, 0);
+       queue_init(&cqp->queue);
+}
+
+/*
+ * Called by each processor to add call buffers to the free list
+ * and to initialize the per-cpu call queue.
+ * Also called but ignored on slave processors on re-start/wake.
+ */
+static void
+mp_cpus_call_cpu_init(void)
+{
+       int             i;
+       mp_call_queue_t *cqp = &mp_cpus_call_head[cpu_number()];
+       mp_call_t       *callp;
+
+       if (cqp->queue.next != NULL)
+               return; /* restart/wake case: called already */
+
+       simple_lock_init(&cqp->lock, 0);
+       queue_init(&cqp->queue);
+       for (i = 0; i < MP_CPUS_CALL_BUFS_PER_CPU; i++) {
+               callp = (mp_call_t *) kalloc(sizeof(mp_call_t));
+               mp_call_free(callp);

        }
        }

+
+       DBG("mp_cpus_call_init() done on cpu %d\n", cpu_number());

 }
 
 /*
  * This is called from cpu_signal_handler() to process an MP_CALL signal.
 }
 
 /*
  * This is called from cpu_signal_handler() to process an MP_CALL signal.

+ * And also from i386_deactivate_cpu() when a cpu is being taken offline.

  */
 static void
 mp_cpus_call_action(void)
 {
  */
 static void
 mp_cpus_call_action(void)
 {

-       if (mp_rv_action_func != NULL)
-               mp_rv_action_func(mp_rv_func_arg);
-       atomic_incl(&mp_rv_complete, 1);
+       mp_call_queue_t *cqp;
+       boolean_t       intrs_enabled;
+       mp_call_t       *callp;
+       mp_call_t       call;
+
+       assert(!ml_get_interrupts_enabled());
+       cqp = &mp_cpus_call_head[cpu_number()];
+       intrs_enabled = mp_call_head_lock(cqp);
+       while ((callp = mp_call_dequeue_locked(cqp)) != NULL) {
+               /* Copy call request to the stack to free buffer */
+               call = *callp;
+               mp_call_free(callp);
+               if (call.func != NULL) {
+                       mp_call_head_unlock(cqp, intrs_enabled);
+                       KERNEL_DEBUG_CONSTANT(
+                               TRACE_MP_CPUS_CALL_ACTION,
+                               call.func, call.arg0, call.arg1, call.countp, 0);
+                       call.func(call.arg0, call.arg1);
+                       (void) mp_call_head_lock(cqp);
+               }
+               if (call.countp != NULL)
+                       atomic_incl(call.countp, 1);
+       }
+       mp_call_head_unlock(cqp, intrs_enabled);

 }
 
 /*
  * mp_cpus_call() runs a given function on cpus specified in a given cpu mask.
 }
 
 /*
  * mp_cpus_call() runs a given function on cpus specified in a given cpu mask.

- * If the mode is SYNC, the function is called serially on the target cpus
- * in logical cpu order. If the mode is ASYNC, the function is called in
- * parallel over the specified cpus.
+ * Possible modes are:
+ *  SYNC:   function is called serially on target cpus in logical cpu order
+ *         waiting for each call to be acknowledged before proceeding
+ *  ASYNC:  function call is queued to the specified cpus
+ *         waiting for all calls to complete in parallel before returning
+ *  NOSYNC: function calls are queued
+ *         but we return before confirmation of calls completing. 

  * The action function may be NULL.
  * The cpu mask may include the local cpu. Offline cpus are ignored.
  * The action function may be NULL.
  * The cpu mask may include the local cpu. Offline cpus are ignored.

- * Return does not occur until the function has completed on all cpus.
- * The return value is the number of cpus on which the function was called.
+ * The return value is the number of cpus on which the call was made or queued.

  */
 cpu_t
 mp_cpus_call(
  */
 cpu_t
 mp_cpus_call(


@@ -810,32 +1028,77 @@ mp_cpus_call(
        mp_sync_t       mode,
         void           (*action_func)(void *),
         void           *arg)
        mp_sync_t       mode,
         void           (*action_func)(void *),
         void           *arg)

+{
+       return mp_cpus_call1(
+                       cpus,
+                       mode,
+                       (void (*)(void *,void *))action_func,
+                       arg,
+                       NULL,
+                       NULL,
+                       NULL);
+}
+
+static void
+mp_cpus_call_wait(boolean_t    intrs_enabled,
+                 long          mp_cpus_signals,
+                 volatile long *mp_cpus_calls)
+{
+       mp_call_queue_t         *cqp;
+
+       cqp = &mp_cpus_call_head[cpu_number()];
+
+       while (*mp_cpus_calls < mp_cpus_signals) {
+               if (!intrs_enabled) {
+                       /* Sniffing w/o locking */
+                       if (!queue_empty(&cqp->queue))
+                               mp_cpus_call_action();
+                       handle_pending_TLB_flushes();
+               }
+               cpu_pause();
+       }
+}
+
+cpu_t
+mp_cpus_call1(
+       cpumask_t       cpus,
+       mp_sync_t       mode,
+        void           (*action_func)(void *, void *),
+        void           *arg0,
+        void           *arg1,
+       cpumask_t       *cpus_calledp,
+       cpumask_t       *cpus_notcalledp)

 {
        cpu_t           cpu;
 {
        cpu_t           cpu;

-       boolean_t       intrs_enabled = ml_get_interrupts_enabled();
+       boolean_t       intrs_enabled = FALSE;

        boolean_t       call_self = FALSE;
        boolean_t       call_self = FALSE;

+       cpumask_t       cpus_called = 0;
+       cpumask_t       cpus_notcalled = 0;
+       long            mp_cpus_signals = 0;
+       volatile long   mp_cpus_calls = 0;
+
+       KERNEL_DEBUG_CONSTANT(
+               TRACE_MP_CPUS_CALL | DBG_FUNC_START,
+               cpus, mode, VM_KERNEL_UNSLIDE(action_func), arg0, arg1);

 
        if (!smp_initialized) {
                if ((cpus & CPUMASK_SELF) == 0)
 
        if (!smp_initialized) {
                if ((cpus & CPUMASK_SELF) == 0)

-                       return 0;
+                       goto out;

                if (action_func != NULL) {
                if (action_func != NULL) {

-                       (void) ml_set_interrupts_enabled(FALSE);
-                       action_func(arg);
+                       intrs_enabled = ml_set_interrupts_enabled(FALSE);
+                       action_func(arg0, arg1);

                        ml_set_interrupts_enabled(intrs_enabled);
                }
                        ml_set_interrupts_enabled(intrs_enabled);
                }

-               return 1;
+               call_self = TRUE;
+               goto out;

        }
        }

-               
-       /* obtain rendezvous lock */
-       simple_lock(&mp_rv_lock);

 
 

-       /* Use the rendezvous data structures for this call */
-       mp_rv_action_func = action_func;
-       mp_rv_func_arg = arg;
-       mp_rv_ncpus = 0;
-       mp_rv_complete = 0;
-
-       simple_lock(&x86_topo_lock);
+       /*
+        * Queue the call for each non-local requested cpu.
+        * The topo lock is not taken. Instead we sniff the cpu_running state
+        * and then re-check it after taking the call lock. A cpu being taken
+        * offline runs the action function after clearing the cpu_running.
+        */ 

        for (cpu = 0; cpu < (cpu_t) real_ncpus; cpu++) {
                if (((cpu_to_cpumask(cpu) & cpus) == 0) ||
                    !cpu_datap(cpu)->cpu_running)
        for (cpu = 0; cpu < (cpu_t) real_ncpus; cpu++) {
                if (((cpu_to_cpumask(cpu) & cpus) == 0) ||
                    !cpu_datap(cpu)->cpu_running)


@@ -846,61 +1109,104 @@ mp_cpus_call(
                         * we defer our call until we have signalled all others.
                         */
                        call_self = TRUE;
                         * we defer our call until we have signalled all others.
                         */
                        call_self = TRUE;

+                       cpus_called |= cpu_to_cpumask(cpu);

                        if (mode == SYNC && action_func != NULL) {
                        if (mode == SYNC && action_func != NULL) {

-                               (void) ml_set_interrupts_enabled(FALSE);
-                               action_func(arg);
-                               ml_set_interrupts_enabled(intrs_enabled);
+                               KERNEL_DEBUG_CONSTANT(
+                                       TRACE_MP_CPUS_CALL_LOCAL,
+                                       VM_KERNEL_UNSLIDE(action_func),
+                                       arg0, arg1, 0, 0);
+                               action_func(arg0, arg1);

                        }
                } else {
                        /*
                        }
                } else {
                        /*

-                        * Bump count of other cpus called and signal this cpu.
-                        * Note: we signal asynchronously regardless of mode
-                        * because we wait on mp_rv_complete either here
-                        * (if mode == SYNC) or later (if mode == ASYNC).
-                        * While spinning, poll for TLB flushes if interrupts
-                        * are disabled.
+                        * Here to queue a call to cpu and IPI.
+                        * Spinning for request buffer unless NOSYNC.

                         */
                         */

-                       mp_rv_ncpus++;
-                       i386_signal_cpu(cpu, MP_CALL, ASYNC);
-                       if (mode == SYNC) {
-                               simple_unlock(&x86_topo_lock);
-                               while (mp_rv_complete < mp_rv_ncpus) {
-                                       if (!intrs_enabled)
+                       mp_call_t       *callp = NULL;
+                       mp_call_queue_t *cqp = &mp_cpus_call_head[cpu];
+
+               queue_call:
+                       if (callp == NULL)
+                               callp = mp_call_alloc();
+                       intrs_enabled = mp_call_head_lock(cqp);
+                       if (!cpu_datap(cpu)->cpu_running) {
+                               mp_call_head_unlock(cqp, intrs_enabled);
+                               continue;
+                       }
+                       if (mode == NOSYNC) {
+                               if (callp == NULL) {
+                                       cpus_notcalled |= cpu_to_cpumask(cpu);
+                                       mp_call_head_unlock(cqp, intrs_enabled);
+                                       KERNEL_DEBUG_CONSTANT(
+                                               TRACE_MP_CPUS_CALL_NOBUF,
+                                               cpu, 0, 0, 0, 0);
+                                       continue;
+                               }
+                               callp->countp = NULL;
+                       } else {
+                               if (callp == NULL) {
+                                       mp_call_head_unlock(cqp, intrs_enabled);
+                                       KERNEL_DEBUG_CONSTANT(
+                                               TRACE_MP_CPUS_CALL_NOBUF,
+                                               cpu, 0, 0, 0, 0);
+                                       if (!intrs_enabled) {
+                                               /* Sniffing w/o locking */
+                                               if (!queue_empty(&cqp->queue))
+                                                       mp_cpus_call_action();

                                                handle_pending_TLB_flushes();
                                                handle_pending_TLB_flushes();

+                                       }

                                        cpu_pause();
                                        cpu_pause();

+                                       goto queue_call;

                                }
                                }

-                               simple_lock(&x86_topo_lock);
+                               callp->countp = &mp_cpus_calls;
+                       }
+                       callp->func = action_func;
+                       callp->arg0 = arg0;
+                       callp->arg1 = arg1;
+                       mp_call_enqueue_locked(cqp, callp);
+                       mp_cpus_signals++;
+                       cpus_called |= cpu_to_cpumask(cpu);
+                       i386_signal_cpu(cpu, MP_CALL, ASYNC);
+                       mp_call_head_unlock(cqp, intrs_enabled);
+                       if (mode == SYNC) {
+                               mp_cpus_call_wait(intrs_enabled, mp_cpus_signals, &mp_cpus_calls);

                        }
                }
        }
                        }
                }
        }

-       simple_unlock(&x86_topo_lock);

 
 

-       /*
-        * If calls are being made asynchronously,
-        * make the local call now if needed, and then
-        * wait for all other cpus to finish their calls.
-        */
-       if (mode == ASYNC) {
-               if (call_self && action_func != NULL) {
-                       (void) ml_set_interrupts_enabled(FALSE);
-                       action_func(arg);
+       /* Call locally if mode not SYNC */
+       if (mode != SYNC && call_self ) {
+               KERNEL_DEBUG_CONSTANT(
+                       TRACE_MP_CPUS_CALL_LOCAL,
+                       VM_KERNEL_UNSLIDE(action_func), arg0, arg1, 0, 0);
+               if (action_func != NULL) {
+                       ml_set_interrupts_enabled(FALSE);
+                       action_func(arg0, arg1);

                        ml_set_interrupts_enabled(intrs_enabled);
                }
                        ml_set_interrupts_enabled(intrs_enabled);
                }

-               while (mp_rv_complete < mp_rv_ncpus) {
-                       if (!intrs_enabled)
-                               handle_pending_TLB_flushes();
-                       cpu_pause();
-               }

        }
        }

-       
-       /* Determine the number of cpus called */
-       cpu = mp_rv_ncpus + (call_self ? 1 : 0);

 
 

-       simple_unlock(&mp_rv_lock);
+       /* For ASYNC, now wait for all signaled cpus to complete their calls */
+       if (mode == ASYNC) {
+               mp_cpus_call_wait(intrs_enabled, mp_cpus_signals, &mp_cpus_calls);
+       }
+
+out:
+       cpu = (cpu_t) mp_cpus_signals + (call_self ? 1 : 0);
+
+       if (cpus_calledp)
+               *cpus_calledp = cpus_called;
+       if (cpus_notcalledp)
+               *cpus_notcalledp = cpus_notcalled;
+
+       KERNEL_DEBUG_CONSTANT(
+               TRACE_MP_CPUS_CALL | DBG_FUNC_END,
+               cpu, cpus_called, cpus_notcalled, 0, 0);

 
        return cpu;
 }
 
 
        return cpu;
 }
 

+

 static void
 mp_broadcast_action(void)
 {
 static void
 mp_broadcast_action(void)
 {


@@ -909,8 +1215,8 @@ mp_broadcast_action(void)
        mp_bc_action_func(mp_bc_func_arg);
 
    /* if we're the last one through, wake up the instigator */
        mp_bc_action_func(mp_bc_func_arg);
 
    /* if we're the last one through, wake up the instigator */

-   if (atomic_decl_and_test((volatile long *)&mp_bc_count, 1))
-       thread_wakeup(((event_t)(unsigned int *) &mp_bc_count));
+   if (atomic_decl_and_test(&mp_bc_count, 1))
+       thread_wakeup(((event_t)(uintptr_t) &mp_bc_count));

 }
 
 /*
 }
 
 /*


@@ -930,13 +1236,13 @@ mp_broadcast(
    }
        
    /* obtain broadcast lock */
    }
        
    /* obtain broadcast lock */

-   mutex_lock(&mp_bc_lock);
+   lck_mtx_lock(&mp_bc_lock);

 
    /* set static function pointers */
    mp_bc_action_func = action_func;
    mp_bc_func_arg = arg;
 
 
    /* set static function pointers */
    mp_bc_action_func = action_func;
    mp_bc_func_arg = arg;
 

-   assert_wait(&mp_bc_count, THREAD_UNINT);
+   assert_wait((event_t)(uintptr_t)&mp_bc_count, THREAD_UNINT);

 
    /*
     * signal other processors, which will call mp_broadcast_action()
 
    /*
     * signal other processors, which will call mp_broadcast_action()


@@ -957,7 +1263,7 @@ mp_broadcast(
        clear_wait(current_thread(), THREAD_AWAKENED);
        
    /* release lock */
        clear_wait(current_thread(), THREAD_AWAKENED);
        
    /* release lock */

-   mutex_unlock(&mp_bc_lock);
+   lck_mtx_unlock(&mp_bc_lock);

 }
 
 void
 }
 
 void


@@ -974,9 +1280,13 @@ i386_activate_cpu(void)
 
        simple_lock(&x86_topo_lock);
        cdp->cpu_running = TRUE;
 
        simple_lock(&x86_topo_lock);
        cdp->cpu_running = TRUE;

+       started_cpu();

        simple_unlock(&x86_topo_lock);
        simple_unlock(&x86_topo_lock);

+       flush_tlb_raw();

 }
 
 }
 

+extern void etimer_timer_expire(void   *arg);
+

 void
 i386_deactivate_cpu(void)
 {
 void
 i386_deactivate_cpu(void)
 {


@@ -988,6 +1298,10 @@ i386_deactivate_cpu(void)
        cdp->cpu_running = FALSE;
        simple_unlock(&x86_topo_lock);
 
        cdp->cpu_running = FALSE;
        simple_unlock(&x86_topo_lock);
 

+       timer_queue_shutdown(&cdp->rtclock_timer.queue);
+       cdp->rtclock_timer.deadline = EndOfAllTime;
+       mp_cpus_call(cpu_to_cpumask(master_cpu), ASYNC, etimer_timer_expire, NULL);
+

        /*
         * In case a rendezvous/braodcast/call was initiated to this cpu
         * before we cleared cpu_running, we must perform any actions due.
        /*
         * In case a rendezvous/braodcast/call was initiated to this cpu
         * before we cleared cpu_running, we must perform any actions due.


@@ -1013,7 +1327,7 @@ void
 mp_kdp_enter(void)
 {
        unsigned int    cpu;
 mp_kdp_enter(void)
 {
        unsigned int    cpu;

-       unsigned int    ncpus;
+       unsigned int    ncpus = 0;

        unsigned int    my_cpu;
        uint64_t        tsc_timeout;
 
        unsigned int    my_cpu;
        uint64_t        tsc_timeout;
 


@@ -1025,21 +1339,33 @@ mp_kdp_enter(void)
         * stopping others.
         */
        mp_kdp_state = ml_set_interrupts_enabled(FALSE);
         * stopping others.
         */
        mp_kdp_state = ml_set_interrupts_enabled(FALSE);

+       my_cpu = cpu_number();
+
+       if (my_cpu == (unsigned) debugger_cpu) {
+               kprintf("\n\nRECURSIVE DEBUGGER ENTRY DETECTED\n\n");
+               kdp_reset();
+               return;
+       }
+
+       cpu_datap(my_cpu)->debugger_entry_time = mach_absolute_time();

        simple_lock(&mp_kdp_lock);
 
        simple_lock(&mp_kdp_lock);
 

-       if (pmsafe_debug)
+       if (pmsafe_debug && !kdp_snapshot)

            pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
 
        while (mp_kdp_trap) {
                simple_unlock(&mp_kdp_lock);
                DBG("mp_kdp_enter() race lost\n");
            pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
 
        while (mp_kdp_trap) {
                simple_unlock(&mp_kdp_lock);
                DBG("mp_kdp_enter() race lost\n");

-               mp_kdp_wait(TRUE);
+#if MACH_KDP
+               mp_kdp_wait(TRUE, FALSE);
+#endif

                simple_lock(&mp_kdp_lock);
        }
                simple_lock(&mp_kdp_lock);
        }

-       my_cpu = cpu_number();

        debugger_cpu = my_cpu;
        debugger_cpu = my_cpu;

+       ncpus = 1;

        mp_kdp_ncpus = 1;       /* self */
        mp_kdp_trap = TRUE;
        mp_kdp_ncpus = 1;       /* self */
        mp_kdp_trap = TRUE;

+       debugger_entry_time = cpu_datap(my_cpu)->debugger_entry_time;

        simple_unlock(&mp_kdp_lock);
 
        /*
        simple_unlock(&mp_kdp_lock);
 
        /*


@@ -1047,7 +1373,7 @@ mp_kdp_enter(void)
         */
        DBG("mp_kdp_enter() signaling other processors\n");
        if (force_immediate_debugger_NMI == FALSE) {
         */
        DBG("mp_kdp_enter() signaling other processors\n");
        if (force_immediate_debugger_NMI == FALSE) {

-               for (ncpus = 1, cpu = 0; cpu < real_ncpus; cpu++) {
+               for (cpu = 0; cpu < real_ncpus; cpu++) {

                        if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
                                continue;
                        ncpus++;
                        if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
                                continue;
                        ncpus++;


@@ -1064,7 +1390,10 @@ mp_kdp_enter(void)
                 * "unsafe-to-interrupt" points such as the trampolines,
                 * but neither do we want to lose state by waiting too long.
                 */
                 * "unsafe-to-interrupt" points such as the trampolines,
                 * but neither do we want to lose state by waiting too long.
                 */

-               tsc_timeout = rdtsc64() + (ncpus * 1000 * 1000);
+               tsc_timeout = rdtsc64() + (ncpus * 1000 * 1000 * 10ULL);
+
+               if (virtualized)
+                       tsc_timeout = ~0ULL;

 
                while (mp_kdp_ncpus != ncpus && rdtsc64() < tsc_timeout) {
                        /*
 
                while (mp_kdp_ncpus != ncpus && rdtsc64() < tsc_timeout) {
                        /*


@@ -1096,7 +1425,7 @@ mp_kdp_enter(void)
                }
 
        DBG("mp_kdp_enter() %u processors done %s\n",
                }
 
        DBG("mp_kdp_enter() %u processors done %s\n",

-           mp_kdp_ncpus, (mp_kdp_ncpus == ncpus) ? "OK" : "timed out");
+           (int)mp_kdp_ncpus, (mp_kdp_ncpus == ncpus) ? "OK" : "timed out");

        
        postcode(MP_KDP_ENTER);
 }
        
        postcode(MP_KDP_ENTER);
 }


@@ -1111,23 +1440,53 @@ cpu_signal_pending(int cpu, mp_event_t event)
                retval = TRUE;
        return retval;
 }
                retval = TRUE;
        return retval;
 }

- 
+
+long kdp_x86_xcpu_invoke(const uint16_t lcpu, kdp_x86_xcpu_func_t func,
+                        void *arg0, void *arg1)
+{
+       if (lcpu > (real_ncpus - 1))
+               return -1;
+
+        if (func == NULL)
+               return -1;
+
+       kdp_xcpu_call_func.func = func;
+        kdp_xcpu_call_func.ret  = -1;
+       kdp_xcpu_call_func.arg0 = arg0;
+       kdp_xcpu_call_func.arg1 = arg1;
+       kdp_xcpu_call_func.cpu  = lcpu;
+       DBG("Invoking function %p on CPU %d\n", func, (int32_t)lcpu);
+       while (kdp_xcpu_call_func.cpu != KDP_XCPU_NONE)
+               cpu_pause();
+        return kdp_xcpu_call_func.ret;
+}
+
+static void
+kdp_x86_xcpu_poll(void)
+{
+       if ((uint16_t)cpu_number() == kdp_xcpu_call_func.cpu) {
+            kdp_xcpu_call_func.ret = 
+                   kdp_xcpu_call_func.func(kdp_xcpu_call_func.arg0,
+                                           kdp_xcpu_call_func.arg1,
+                                           cpu_number());
+               kdp_xcpu_call_func.cpu = KDP_XCPU_NONE;
+       }
+}

 
 static void
 
 static void

-mp_kdp_wait(boolean_t flush)
+mp_kdp_wait(boolean_t flush, boolean_t isNMI)

 {
        DBG("mp_kdp_wait()\n");
        /* If an I/O port has been specified as a debugging aid, issue a read */
        panic_io_port_read();
 
 {
        DBG("mp_kdp_wait()\n");
        /* If an I/O port has been specified as a debugging aid, issue a read */
        panic_io_port_read();
 

+#if CONFIG_MCA

        /* If we've trapped due to a machine-check, save MCA registers */
        mca_check_save();
        /* If we've trapped due to a machine-check, save MCA registers */
        mca_check_save();

-
-       if (pmsafe_debug)
-           pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_SAFE);
+#endif

 
        atomic_incl((volatile long *)&mp_kdp_ncpus, 1);
 
        atomic_incl((volatile long *)&mp_kdp_ncpus, 1);

-       while (mp_kdp_trap) {
+       while (mp_kdp_trap || (isNMI == TRUE)) {

                /*
                 * A TLB shootdown request may be pending--this would result
                 * in the requesting processor waiting in PMAP_UPDATE_TLBS()
                /*
                 * A TLB shootdown request may be pending--this would result
                 * in the requesting processor waiting in PMAP_UPDATE_TLBS()


@@ -1136,12 +1495,11 @@ mp_kdp_wait(boolean_t flush)
                 */
                if (flush)
                        handle_pending_TLB_flushes();
                 */
                if (flush)
                        handle_pending_TLB_flushes();

+
+               kdp_x86_xcpu_poll();

                cpu_pause();
        }
 
                cpu_pause();
        }
 

-       if (pmsafe_debug)
-           pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_NORMAL);
-

        atomic_decl((volatile long *)&mp_kdp_ncpus, 1);
        DBG("mp_kdp_wait() done\n");
 }
        atomic_decl((volatile long *)&mp_kdp_ncpus, 1);
        DBG("mp_kdp_wait() done\n");
 }


@@ -1152,6 +1510,9 @@ mp_kdp_exit(void)
        DBG("mp_kdp_exit()\n");
        debugger_cpu = -1;
        atomic_decl((volatile long *)&mp_kdp_ncpus, 1);
        DBG("mp_kdp_exit()\n");
        debugger_cpu = -1;
        atomic_decl((volatile long *)&mp_kdp_ncpus, 1);

+
+       debugger_exit_time = mach_absolute_time();
+

        mp_kdp_trap = FALSE;
        __asm__ volatile("mfence");
 
        mp_kdp_trap = FALSE;
        __asm__ volatile("mfence");
 


@@ -1168,15 +1529,24 @@ mp_kdp_exit(void)
                cpu_pause();
        }
 
                cpu_pause();
        }
 

-       if (pmsafe_debug)
+       if (pmsafe_debug && !kdp_snapshot)

            pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_NORMAL);
 
            pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_NORMAL);
 

+       debugger_exit_time = mach_absolute_time();
+

        DBG("mp_kdp_exit() done\n");
        (void) ml_set_interrupts_enabled(mp_kdp_state);
        postcode(0);
 }
 #endif /* MACH_KDP */
 
        DBG("mp_kdp_exit() done\n");
        (void) ml_set_interrupts_enabled(mp_kdp_state);
        postcode(0);
 }
 #endif /* MACH_KDP */
 

+boolean_t
+mp_recent_debugger_activity() {
+       uint64_t abstime = mach_absolute_time();
+       return (((abstime - debugger_entry_time) < LastDebuggerEntryAllowance) ||
+           ((abstime - debugger_exit_time) < LastDebuggerEntryAllowance));
+}
+

 /*ARGSUSED*/
 void
 init_ast_check(
 /*ARGSUSED*/
 void
 init_ast_check(


@@ -1188,249 +1558,149 @@ void
 cause_ast_check(
        processor_t     processor)
 {
 cause_ast_check(
        processor_t     processor)
 {

-       int     cpu = PROCESSOR_DATA(processor, slot_num);
+       int     cpu = processor->cpu_id;

 
        if (cpu != cpu_number()) {
                i386_signal_cpu(cpu, MP_AST, ASYNC);
 
        if (cpu != cpu_number()) {
                i386_signal_cpu(cpu, MP_AST, ASYNC);

+               KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_REMOTE_AST), cpu, 1, 0, 0, 0);

        }
 }
 
        }
 }
 

-#if MACH_KDB
-/*
- * invoke kdb on slave processors 
- */
-

 void
 void

-remote_kdb(void)
+slave_machine_init(void *param)

 {
 {

-       unsigned int    my_cpu = cpu_number();
-       unsigned int    cpu;
-       int kdb_ncpus;
-       uint64_t tsc_timeout = 0;
-       
-       mp_kdb_trap = TRUE;
-       mp_kdb_ncpus = 1;
-       for (kdb_ncpus = 1, cpu = 0; cpu < real_ncpus; cpu++) {
-               if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
-                       continue;
-               kdb_ncpus++;
-               i386_signal_cpu(cpu, MP_KDB, ASYNC);
-       }
-       DBG("remote_kdb() waiting for (%d) processors to suspend\n",kdb_ncpus);
-
-       tsc_timeout = rdtsc64() + (kdb_ncpus * 100 * 1000 * 1000);
+       /*
+        * Here in process context, but with interrupts disabled.
+        */
+       DBG("slave_machine_init() CPU%d\n", get_cpu_number());

 
 

-       while (mp_kdb_ncpus != kdb_ncpus && rdtsc64() < tsc_timeout) {
-               /*
-                * a TLB shootdown request may be pending... this would result in the requesting
-                * processor waiting in PMAP_UPDATE_TLBS() until this processor deals with it.
-                * Process it, so it can now enter mp_kdp_wait()
+       if (param == FULL_SLAVE_INIT) {
+               /*
+                * Cold start

                 */
                 */

-               handle_pending_TLB_flushes();
-
-               cpu_pause();
+               clock_init();
+               cpu_machine_init();     /* Interrupts enabled hereafter */
+               mp_cpus_call_cpu_init();

        }
        }

-       DBG("mp_kdp_enter() %d processors done %s\n",
-               mp_kdb_ncpus, (mp_kdb_ncpus == kdb_ncpus) ? "OK" : "timed out");

 }
 
 }
 

-static void
-mp_kdb_wait(void)
+#undef cpu_number
+int cpu_number(void)

 {
 {

-       DBG("mp_kdb_wait()\n");
-
-       /* If an I/O port has been specified as a debugging aid, issue a read */
-       panic_io_port_read();
-
-       atomic_incl(&mp_kdb_ncpus, 1);
-       while (mp_kdb_trap) {
-               /*
-                * a TLB shootdown request may be pending... this would result in the requesting
-                * processor waiting in PMAP_UPDATE_TLBS() until this processor deals with it.
-                * Process it, so it can now enter mp_kdp_wait()
-                */
-               handle_pending_TLB_flushes();
-
-               cpu_pause();
-       }
-       atomic_decl((volatile long *)&mp_kdb_ncpus, 1);
-       DBG("mp_kdb_wait() done\n");
+       return get_cpu_number();

 }
 
 }
 

-/*
- * Clear kdb interrupt
- */
-
-void
-clear_kdb_intr(void)
+static void
+cpu_prewarm_init()

 {
 {

-       mp_disable_preemption();
-       i_bit_clear(MP_KDB, &current_cpu_datap()->cpu_signals);
-       mp_enable_preemption();
+       int i;
+
+       simple_lock_init(&cpu_warm_lock, 0);
+       queue_init(&cpu_warm_call_list);
+       for (i = 0; i < NUM_CPU_WARM_CALLS; i++) {
+               enqueue_head(&cpu_warm_call_list, (queue_entry_t)&cpu_warm_call_arr[i]);
+       }

 }
 
 }
 

-void
-mp_kdb_exit(void)
+static timer_call_t
+grab_warm_timer_call()

 {
 {

-       DBG("mp_kdb_exit()\n");
-       atomic_decl((volatile long *)&mp_kdb_ncpus, 1);
-       mp_kdb_trap = FALSE;
-       __asm__ volatile("mfence");
-
-       while (mp_kdb_ncpus > 0) {
-               /*
-                * a TLB shootdown request may be pending... this would result in the requesting
-                * processor waiting in PMAP_UPDATE_TLBS() until this processor deals with it.
-                * Process it, so it can now enter mp_kdp_wait()
-                */
-               handle_pending_TLB_flushes();
+       spl_t x;
+       timer_call_t call = NULL;

 
 

-               cpu_pause();
+       x = splsched();
+       simple_lock(&cpu_warm_lock);
+       if (!queue_empty(&cpu_warm_call_list)) {
+               call = (timer_call_t) dequeue_head(&cpu_warm_call_list);

        }
        }

+       simple_unlock(&cpu_warm_lock);
+       splx(x);

 
 

-       DBG("mp_kdb_exit() done\n");
+       return call;

 }
 
 }
 

-#endif /* MACH_KDB */
-

 static void
 static void

-do_init_slave(boolean_t fast_restart)
+free_warm_timer_call(timer_call_t call)

 {
 {

-       void    *init_param     = FULL_SLAVE_INIT;
-
-       postcode(I386_INIT_SLAVE);
-
-       if (!fast_restart) {
-               /* Ensure that caching and write-through are enabled */
-               set_cr0(get_cr0() & ~(CR0_NW|CR0_CD));
-
-               DBG("i386_init_slave() CPU%d: phys (%d) active.\n",
-                   get_cpu_number(), get_cpu_phys_number());
-
-               assert(!ml_get_interrupts_enabled());
+       spl_t x;

 
 

-               cpu_mode_init(current_cpu_datap());
-
-               mca_cpu_init();
-
-               lapic_configure();
-               LAPIC_DUMP();
-               LAPIC_CPU_MAP_DUMP();
-
-               init_fpu();
-
-               mtrr_update_cpu();
-       } else
-               init_param = FAST_SLAVE_INIT;
-
-       /* resume VT operation */
-       vmx_resume();
-
-       if (!fast_restart)
-               pat_init();
-
-       cpu_thread_init();      /* not strictly necessary */
-
-       cpu_init();     /* Sets cpu_running which starter cpu waits for */ 
-
-       slave_main(init_param);
-
-       panic("do_init_slave() returned from slave_main()");
+       x = splsched();
+       simple_lock(&cpu_warm_lock);
+       enqueue_head(&cpu_warm_call_list, (queue_entry_t)call);
+       simple_unlock(&cpu_warm_lock);
+       splx(x);

 }
 
 /*
 }
 
 /*

- * i386_init_slave() is called from pstart.
- * We're in the cpu's interrupt stack with interrupts disabled.
- * At this point we are in legacy mode. We need to switch on IA32e
- * if the mode is set to 64-bits.
+ * Runs in timer call context (interrupts disabled).

  */
  */

-void
-i386_init_slave(void)
+static void
+cpu_warm_timer_call_func(
+               call_entry_param_t p0,
+               __unused call_entry_param_t p1)

 {
 {

-       do_init_slave(FALSE);
+       free_warm_timer_call((timer_call_t)p0);
+       return;

 }
 
 /*
 }
 
 /*

- * i386_init_slave_fast() is called from pmCPUHalt.
- * We're running on the idle thread and need to fix up
- * some accounting and get it so that the scheduler sees this
- * CPU again.
+ * Runs with interrupts disabled on the CPU we wish to warm (i.e. CPU 0).

  */
  */

-void
-i386_init_slave_fast(void)
-{
-       do_init_slave(TRUE);
-}
-
-void
-slave_machine_init(void *param)
+static void
+_cpu_warm_setup(
+               void *arg)

 {
 {

-       /*
-        * Here in process context, but with interrupts disabled.
-        */
-       DBG("slave_machine_init() CPU%d\n", get_cpu_number());
+       cpu_warm_data_t cwdp = (cpu_warm_data_t)arg;

 
 

-       if (param == FULL_SLAVE_INIT) {
-               /*
-                * Cold start
-                */
-               clock_init();
+       timer_call_enter(cwdp->cwd_call, cwdp->cwd_deadline, TIMER_CALL_CRITICAL | TIMER_CALL_LOCAL);
+       cwdp->cwd_result = 0;

 
 

-               cpu_machine_init();     /* Interrupts enabled hereafter */
-       }
+       return;

 }
 
 }
 

-#undef cpu_number()
-int cpu_number(void)
+/*
+ * Not safe to call with interrupts disabled.
+ */
+kern_return_t
+ml_interrupt_prewarm(
+       uint64_t        deadline)

 {
 {

-       return get_cpu_number();
-}
-
-#if    MACH_KDB
-#include <ddb/db_output.h>
-
-#define TRAP_DEBUG 0 /* Must match interrupt.s and spl.s */
+       struct cpu_warm_data cwd;
+       timer_call_t call;
+       cpu_t ct;

 
 

+       if (ml_get_interrupts_enabled() == FALSE) {
+               panic("%s: Interrupts disabled?\n", __FUNCTION__);
+       }

 
 

-#if    TRAP_DEBUG
-#define MTRAPS 100
-struct mp_trap_hist_struct {
-       unsigned char type;
-       unsigned char data[5];
-} trap_hist[MTRAPS], *cur_trap_hist = trap_hist,
-    *max_trap_hist = &trap_hist[MTRAPS];
+       /* 
+        * If the platform doesn't need our help, say that we succeeded. 
+        */
+       if (!ml_get_interrupt_prewake_applicable()) {
+               return KERN_SUCCESS;
+       }

 
 

-void db_trap_hist(void);
+       /*
+        * Grab a timer call to use.
+        */
+       call = grab_warm_timer_call();
+       if (call == NULL) {
+               return KERN_RESOURCE_SHORTAGE;
+       }

 
 

-/*
- * SPL:
- *     1: new spl
- *     2: old spl
- *     3: new tpr
- *     4: old tpr
- * INT:
- *     1: int vec
- *     2: old spl
- *     3: new spl
- *     4: post eoi tpr
- *     5: exit tpr
- */
+       timer_call_setup(call, cpu_warm_timer_call_func, call);
+       cwd.cwd_call = call;
+       cwd.cwd_deadline = deadline;
+       cwd.cwd_result = 0;

 
 

-void
-db_trap_hist(void)
-{
-       int i,j;
-       for(i=0;i<MTRAPS;i++)
-           if (trap_hist[i].type == 1 || trap_hist[i].type == 2) {
-                   db_printf("%s%s",
-                             (&trap_hist[i]>=cur_trap_hist)?"*":" ",
-                             (trap_hist[i].type == 1)?"SPL":"INT");
-                   for(j=0;j<5;j++)
-                       db_printf(" %02x", trap_hist[i].data[j]);
-                   db_printf("\n");
-           }
-               
+       /*
+        * For now, non-local interrupts happen on the master processor.
+        */
+       ct = mp_cpus_call(cpu_to_cpumask(master_cpu), SYNC, _cpu_warm_setup, &cwd);
+       if (ct == 0) {
+               free_warm_timer_call(call);
+               return KERN_FAILURE;
+       } else {
+               return cwd.cwd_result;
+       }

 }
 }

-#endif /* TRAP_DEBUG */
-#endif /* MACH_KDB */
-