]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/i386/mp.c
xnu-1228.5.18.tar.gz
[apple/xnu.git] / osfmk / i386 / mp.c
index 2f63b5386e9ae80c605eaa634413598b1c43d424..fb7afc4d05b9387fa8a0f798ec7744c4cf37d397 100644 (file)
@@ -1,31 +1,29 @@
 /*
- * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_OSREFERENCE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
- * This file contains Original Code and/or Modifications of Original Code 
- * as defined in and that are subject to the Apple Public Source License 
- * Version 2.0 (the 'License'). You may not use this file except in 
- * compliance with the License.  The rights granted to you under the 
- * License may not be used to create, or enable the creation or 
- * redistribution of, unlawful or unlicensed copies of an Apple operating 
- * system, or to circumvent, violate, or enable the circumvention or 
- * violation of, any terms of an Apple operating system software license 
- * agreement.
- *
- * Please obtain a copy of the License at 
- * http://www.opensource.apple.com/apsl/ and read it before using this 
- * file.
- *
- * The Original Code and all software distributed under the License are 
- * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 
- * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 
- * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 
- * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 
- * Please see the License for the specific language governing rights and 
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
  * limitations under the License.
- *
- * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
+ * 
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 /*
  * @OSF_COPYRIGHT@
@@ -47,6 +45,7 @@
 #include <kern/cpu_data.h>
 #include <kern/assert.h>
 #include <kern/machine.h>
+#include <kern/pms.h>
 
 #include <vm/vm_map.h>
 #include <vm/vm_kern.h>
 #include <i386/apic.h>
 #include <i386/ipl.h>
 #include <i386/fpu.h>
-#include <i386/pio.h>
 #include <i386/cpuid.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/trap.h>
+#include <i386/machine_routines.h>
+#include <i386/pmCPU.h>
+#include <i386/hpet.h>
+#include <i386/machine_check.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)
 #define PAUSE
 #endif /* MP_DEBUG */
 
-/*
- * By default, use high vectors to leave vector space for systems
- * with multiple I/O APIC's. However some systems that boot with
- * local APIC disabled will hang in SMM when vectors greater than
- * 0x5F are used. Those systems are not expected to have I/O APIC
- * so 16 (0x50 - 0x40) vectors for legacy PIC support is perfect.
- */
-#define LAPIC_DEFAULT_INTERRUPT_BASE   0xD0
-#define LAPIC_REDUCED_INTERRUPT_BASE   0x50
-/*
- * Specific lapic interrupts are relative to this base:
- */ 
-#define LAPIC_PERFCNT_INTERRUPT                0xB
-#define LAPIC_TIMER_INTERRUPT          0xC
-#define LAPIC_SPURIOUS_INTERRUPT       0xD     
-#define LAPIC_INTERPROCESSOR_INTERRUPT 0xE
-#define LAPIC_ERROR_INTERRUPT          0xF
-
 /* Initialize lapic_id so cpu_number() works on non SMP systems */
 unsigned long  lapic_id_initdata = 0;
 unsigned long  lapic_id = (unsigned long)&lapic_id_initdata;
@@ -103,6 +104,7 @@ vm_offset_t lapic_start;
 
 static i386_intr_func_t        lapic_timer_func;
 static i386_intr_func_t        lapic_pmi_func;
+static i386_intr_func_t        lapic_thermal_func;
 
 /* TRUE if local APIC was enabled by the OS not by the BIOS */
 static boolean_t lapic_os_enabled = FALSE;
@@ -112,23 +114,47 @@ int lapic_interrupt_base = LAPIC_DEFAULT_INTERRUPT_BASE;
 
 void           slave_boot_init(void);
 
-static void    mp_kdp_wait(void);
+#if MACH_KDB
+static void    mp_kdb_wait(void);
+volatile boolean_t     mp_kdb_trap = FALSE;
+volatile long  mp_kdb_ncpus = 0;
+#endif
+
+static void    mp_kdp_wait(boolean_t flush);
 static void    mp_rendezvous_action(void);
+static void    mp_broadcast_action(void);
+
+static int             NMIInterruptHandler(x86_saved_state_t *regs);
+static boolean_t       cpu_signal_pending(int cpu, mp_event_t event);
+static void            cpu_NMI_interrupt(int cpu);
 
 boolean_t      smp_initialized = FALSE;
+boolean_t      force_immediate_debugger_NMI = FALSE;
 
 decl_simple_lock_data(,mp_kdp_lock);
 
 decl_mutex_data(static, mp_cpu_boot_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 long            mp_rv_waiters[2];
-decl_simple_lock_data(,mp_rv_lock);
+                       /* 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)));
+
+/* 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);
+
+static void    mp_cpus_call_action(void); 
 
 int            lapic_to_cpu[MAX_CPUS];
 int            cpu_to_lapic[MAX_CPUS];
@@ -151,6 +177,24 @@ lapic_cpu_map(int apic_id, int cpu)
        lapic_to_cpu[apic_id] = cpu;
 }
 
+/*
+ * Retrieve the local apic ID a cpu.
+ *
+ * Returns the local apic ID for the given processor.
+ * If the processor does not exist or apic not configured, returns -1.
+ */
+
+uint32_t
+ml_get_apicid(uint32_t cpu)
+{
+       if(cpu >= (uint32_t)MAX_CPUS)
+               return 0xFFFFFFFF;      /* Return -1 if cpu too big */
+       
+       /* Return the apic ID (or -1 if not configured) */
+       return (uint32_t)cpu_to_lapic[cpu];
+
+}
+
 #ifdef MP_DEBUG
 static void
 lapic_cpu_map_dump(void)
@@ -177,18 +221,6 @@ lapic_cpu_map_dump(void)
 #define LAPIC_DUMP()
 #endif /* MP_DEBUG */
 
-#define LAPIC_REG(reg) \
-       (*((volatile int *)(lapic_start + LAPIC_##reg)))
-#define LAPIC_REG_OFFSET(reg,off) \
-       (*((volatile int *)(lapic_start + LAPIC_##reg + (off))))
-
-#define LAPIC_VECTOR(src) \
-       (lapic_interrupt_base + LAPIC_##src##_INTERRUPT)
-
-#define LAPIC_ISR_IS_SET(base,src) \
-       (LAPIC_REG_OFFSET(ISR_BASE,((base+LAPIC_##src##_INTERRUPT)/32)*0x10) & \
-               (1 <<((base + LAPIC_##src##_INTERRUPT)%32)))
-
 #if GPROF
 /*
  * Initialize dummy structs for profiling. These aren't used but
@@ -208,8 +240,6 @@ struct profile_vars *_profile_vars_cpus[MAX_CPUS] = { &_profile_vars };
 #define GPROF_INIT()
 #endif /* GPROF */
 
-extern void    master_up(void);
-
 void
 smp_init(void)
 {
@@ -224,6 +254,7 @@ 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);
        console_init();
 
        /* Local APIC? */
@@ -243,17 +274,22 @@ smp_init(void)
 
        /* Establish a map to the local apic */
        lapic_start = vm_map_min(kernel_map);
-       result = vm_map_find_space(kernel_map, &lapic_start,
-                                  round_page(LAPIC_SIZE), 0, &entry);
+       result = vm_map_find_space(kernel_map,
+                                  (vm_map_address_t *) &lapic_start,
+                                  round_page(LAPIC_SIZE), 0,
+                                  VM_MAKE_TAG(VM_MEMORY_IOKIT), &entry);
        if (result != KERN_SUCCESS) {
                panic("smp_init: vm_map_find_entry FAILED (err=%d)", result);
        }
        vm_map_unlock(kernel_map);
+/* Map in the local APIC non-cacheable, as recommended by Intel
+ * in section 8.4.1 of the "System Programming Guide".
+ */
        pmap_enter(pmap_kernel(),
                        lapic_start,
                        (ppnum_t) i386_btop(lapic_base),
-                       VM_PROT_READ|VM_PROT_WRITE,
-                       VM_WIMG_USE_DEFAULT,
+                       VM_PROT_READ|VM_PROT_WRITE,
+                       VM_WIMG_IO,
                        TRUE);
        lapic_id = (unsigned long)(lapic_start + LAPIC_ID);
 
@@ -264,19 +300,16 @@ smp_init(void)
        /* Set up the lapic_id <-> cpu_number map and add this boot processor */
        lapic_cpu_map_init();
        lapic_cpu_map((LAPIC_REG(ID)>>LAPIC_ID_SHIFT)&LAPIC_ID_MASK, 0);
+       kprintf("Boot cpu local APIC id 0x%x\n", cpu_to_lapic[0]);
 
        lapic_init();
 
        cpu_thread_init();
 
-       if (pmc_init() != KERN_SUCCESS)
-               printf("Performance counters not available\n");
-
        GPROF_INIT();
        DBGLOG_CPU_INIT(master_cpu);
 
        slave_boot_init();
-       master_up();
 
        smp_initialized = TRUE;
 
@@ -344,6 +377,11 @@ lapic_dump(void)
                DM[(LAPIC_REG(LVT_PERFCNT)>>LAPIC_LVT_DM_SHIFT)&LAPIC_LVT_DM_MASK],
                (LAPIC_REG(LVT_PERFCNT)&LAPIC_LVT_DS_PENDING)?"SendPending":"Idle",
                BOOL(LAPIC_REG(LVT_PERFCNT)&LAPIC_LVT_MASKED));
+       kprintf("LVT_THERMAL: Vector 0x%02x [%s] %s %cmasked\n",
+               LAPIC_REG(LVT_THERMAL)&LAPIC_LVT_VECTOR_MASK,
+               DM[(LAPIC_REG(LVT_THERMAL)>>LAPIC_LVT_DM_SHIFT)&LAPIC_LVT_DM_MASK],
+               (LAPIC_REG(LVT_THERMAL)&LAPIC_LVT_DS_PENDING)?"SendPending":"Idle",
+               BOOL(LAPIC_REG(LVT_THERMAL)&LAPIC_LVT_MASKED));
        kprintf("LVT_LINT0:   Vector 0x%02x [%s][%s][%s] %s %cmasked\n",
                LAPIC_REG(LVT_LINT0)&LAPIC_LVT_VECTOR_MASK,
                DM[(LAPIC_REG(LVT_LINT0)>>LAPIC_LVT_DM_SHIFT)&LAPIC_LVT_DM_MASK],
@@ -381,6 +419,26 @@ lapic_dump(void)
        kprintf("\n");
 }
 
+#if MACH_KDB
+/*
+ *     Displays apic junk
+ *
+ *     da
+ */
+void 
+db_apic(__unused db_expr_t addr,
+       __unused int have_addr,
+       __unused db_expr_t count,
+       __unused char *modif)
+{
+
+       lapic_dump();
+
+       return;
+}
+
+#endif
+
 boolean_t
 lapic_probe(void)
 {
@@ -405,7 +463,7 @@ lapic_probe(void)
                /*
                 * Re-initialize cpu features info and re-check.
                 */
-               set_cpu_model();
+               cpuid_set_info();
                if (cpuid_features() & CPUID_FEATURE_APIC) {
                        printf("Local APIC discovered and enabled\n");
                        lapic_os_enabled = TRUE;
@@ -453,7 +511,7 @@ lapic_shutdown(void)
        rdmsr(MSR_IA32_APIC_BASE, lo, hi);
        lo &= ~MSR_IA32_APIC_BASE_ENABLE;
        wrmsr(MSR_IA32_APIC_BASE, lo, hi);
-       set_cpu_model();
+       cpuid_set_info();
 
        mp_enable_preemption();
 }
@@ -486,10 +544,12 @@ lapic_init(void)
        /* Perfmon: unmasked */
        LAPIC_REG(LVT_PERFCNT) = LAPIC_VECTOR(PERFCNT);
 
+       /* Thermal: unmasked */
+       LAPIC_REG(LVT_THERMAL) = LAPIC_VECTOR(THERMAL);
+
        lapic_esr_clear();
 
        LAPIC_REG(LVT_ERROR) = LAPIC_VECTOR(ERROR);
-
 }
 
 void
@@ -547,6 +607,12 @@ lapic_set_pmi_func(i386_intr_func_t func)
        lapic_pmi_func = func;
 }
 
+void
+lapic_set_thermal_func(i386_intr_func_t func)
+{
+        lapic_thermal_func = func;
+}
+
 static inline void
 _lapic_end_of_interrupt(void)
 {
@@ -560,42 +626,77 @@ lapic_end_of_interrupt(void)
 }
 
 int
-lapic_interrupt(int interrupt, void *state)
+lapic_interrupt(int interrupt, x86_saved_state_t *state)
 {
+       int     retval = 0;
+
+       /* Did we just field an interruption for the HPET comparator? */
+       if(x86_core()->HpetVec == ((uint32_t)interrupt - 0x40)) {
+               /* Yes, go handle it... */
+               retval = HPETInterrupt();
+               /* Was it really handled? */
+               if(retval) {
+                       /* If so, EOI the 'rupt */
+                       _lapic_end_of_interrupt();
+                       /*
+                        * and then leave,
+                        * indicating that this has been handled
+                        */
+                       return 1;
+               }
+       }
+
        interrupt -= lapic_interrupt_base;
-       if (interrupt < 0)
-               return 0;
+       if (interrupt < 0) {
+               if (interrupt == (LAPIC_NMI_INTERRUPT - lapic_interrupt_base)) {
+                       retval = NMIInterruptHandler(state);
+                       _lapic_end_of_interrupt();
+                       return retval;
+               }
+               else
+                       return 0;
+       }
 
        switch(interrupt) {
        case LAPIC_PERFCNT_INTERRUPT:
                if (lapic_pmi_func != NULL)
-                       (*lapic_pmi_func)(
-                               (struct i386_interrupt_state *) state);
+                       (*lapic_pmi_func)(NULL);
                /* Clear interrupt masked */
                LAPIC_REG(LVT_PERFCNT) = LAPIC_VECTOR(PERFCNT);
                _lapic_end_of_interrupt();
-               return 1;
+               retval = 1;
+               break;
        case LAPIC_TIMER_INTERRUPT:
                _lapic_end_of_interrupt();
                if (lapic_timer_func != NULL)
-                       (*lapic_timer_func)(
-                               (struct i386_interrupt_state *) state);
-               return 1;
+                       (*lapic_timer_func)(state);
+               retval = 1;
+               break;
+       case LAPIC_THERMAL_INTERRUPT:
+               if (lapic_thermal_func != NULL)
+                       (*lapic_thermal_func)(NULL);
+               _lapic_end_of_interrupt();
+               retval = 1;
+               break;
        case LAPIC_ERROR_INTERRUPT:
                lapic_dump();
                panic("Local APIC error\n");
                _lapic_end_of_interrupt();
-               return 1;
+               retval = 1;
+               break;
        case LAPIC_SPURIOUS_INTERRUPT:
                kprintf("SPIV\n");
                /* No EOI required here */
-               return 1;
+               retval = 1;
+               break;
        case LAPIC_INTERPROCESSOR_INTERRUPT:
-               cpu_signal_handler((struct i386_interrupt_state *) state);
                _lapic_end_of_interrupt();
-               return 1;
+               cpu_signal_handler(state);
+               retval = 1;
+               break;
        }
-       return 0;
+
+       return retval;
 }
 
 void
@@ -646,8 +747,14 @@ intel_startCPU(
        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. */
-       mp_desc_init(cpu_datap(slot_num), FALSE);
+       /*
+        * 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);
@@ -685,21 +792,20 @@ intel_startCPU(
        mutex_unlock(&mp_cpu_boot_lock);
 
        if (!cpu_datap(slot_num)->cpu_running) {
-               DBG("Failed to start CPU %02d\n", slot_num);
+               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 {
-               DBG("Started CPU %02d\n", slot_num);
-               printf("Started CPU %02d\n", slot_num);
+               kprintf("Started cpu %d (lapic id %08x)\n", slot_num, lapic);
                return KERN_SUCCESS;
        }
 }
 
 extern char    slave_boot_base[];
 extern char    slave_boot_end[];
-extern void    pstart(void);
+extern void    slave_pstart(void);
 
 void
 slave_boot_init(void)
@@ -717,7 +823,7 @@ slave_boot_init(void)
         * The slave boot code is responsible for switching to protected
         * mode and then jumping to the common startup, _start().
         */
-       bcopy_phys((addr64_t) kvtophys((vm_offset_t) slave_boot_base),
+       bcopy_phys(kvtophys((vm_offset_t) slave_boot_base),
                   (addr64_t) MP_BOOT,
                   slave_boot_end-slave_boot_base);
 
@@ -732,9 +838,9 @@ slave_boot_init(void)
         * common startup entry.
         */
        DBG("writing 0x%x at phys 0x%x\n",
-               kvtophys((vm_offset_t) &pstart), MP_MACH_START+MP_BOOT);
+               kvtophys((vm_offset_t) &slave_pstart), MP_MACH_START+MP_BOOT);
        ml_phys_write_word(MP_MACH_START+MP_BOOT,
-                          kvtophys((vm_offset_t) &pstart));
+                          (unsigned int)kvtophys((vm_offset_t) &slave_pstart));
        
        /* Flush caches */
        __asm__("wbinvd");
@@ -749,7 +855,7 @@ MP_EVENT_NAME_DECL();
 #endif /* MP_DEBUG */
 
 void
-cpu_signal_handler(__unused struct i386_interrupt_state *regs)
+cpu_signal_handler(x86_saved_state_t *regs)
 {
        int             my_cpu;
        volatile int    *my_word;
@@ -765,13 +871,20 @@ cpu_signal_handler(__unused struct i386_interrupt_state *regs)
        do {
 #if    MACH_KDB && MACH_ASSERT
                if (i-- <= 0)
-                   Debugger("cpu_signal_handler");
+                   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);
                        i_bit_clear(MP_KDP, my_word);
-                       mp_kdp_wait();
+/* Ensure that the i386_kernel_state at the base of the
+ * 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);
                } else
 #endif /* MACH_KDP */
                if (i_bit(MP_TLB_FLUSH, my_word)) {
@@ -784,16 +897,28 @@ cpu_signal_handler(__unused struct i386_interrupt_state *regs)
                        ast_check(cpu_to_processor(my_cpu));
 #if    MACH_KDB
                } else if (i_bit(MP_KDB, my_word)) {
-                       extern kdb_is_slave[];
 
                        i_bit_clear(MP_KDB, my_word);
-                       kdb_is_slave[my_cpu]++;
-                       kdb_kintr();
+                       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);
                        mp_rendezvous_action();
+               } else if (i_bit(MP_BROADCAST, my_word)) {
+                       DBGLOG(cpu_handle,my_cpu,MP_BROADCAST);
+                       i_bit_clear(MP_BROADCAST, my_word);
+                       mp_broadcast_action();
+               } else if (i_bit(MP_CHUD, my_word)) {
+                       DBGLOG(cpu_handle,my_cpu,MP_CHUD);
+                       i_bit_clear(MP_CHUD, my_word);
+                       chudxnu_cpu_signal_handler();
+               } else if (i_bit(MP_CALL, my_word)) {
+                       DBGLOG(cpu_handle,my_cpu,MP_CALL);
+                       i_bit_clear(MP_CALL, my_word);
+                       mp_cpus_call_action();
                }
        } while (*my_word);
 
@@ -801,36 +926,100 @@ cpu_signal_handler(__unused struct i386_interrupt_state *regs)
 
 }
 
+/* We want this to show up in backtraces, hence marked noinline.
+ */
+static int __attribute__((noinline))
+NMIInterruptHandler(x86_saved_state_t *regs)
+{
+       boolean_t state = ml_set_interrupts_enabled(FALSE);
+       sync_iss_to_iks_unconditionally(regs);
+       mp_kdp_wait(FALSE);
+       (void) ml_set_interrupts_enabled(state);
+       return 1;
+}
+
 #ifdef MP_DEBUG
 extern int     max_lock_loops;
+int            trappedalready = 0;     /* (BRINGUP */
 #endif /* MP_DEBUG */
-void
-cpu_interrupt(int cpu)
+
+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 (smp_initialized) {
+#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... */
+       /* Wait for previous interrupt to be delivered... */
 #ifdef MP_DEBUG
-               int     pending_busy_count = 0;
-               while (LAPIC_REG(ICR) & LAPIC_ICR_DS_PENDING) {
-                       if (++pending_busy_count > max_lock_loops)
-                               panic("cpus_interrupt() deadlock\n");
+       int     pending_busy_count = 0;
+       while (LAPIC_REG(ICR) & LAPIC_ICR_DS_PENDING) {
+               if (++pending_busy_count > max_lock_loops)
+                       panic("i386_cpu_IPI() deadlock\n");
 #else
-               while (LAPIC_REG(ICR) & LAPIC_ICR_DS_PENDING) {
+       while (LAPIC_REG(ICR) & LAPIC_ICR_DS_PENDING) {
 #endif /* MP_DEBUG */
-                       cpu_pause();
-               }
+               cpu_pause();
+       }
+
+       state = ml_set_interrupts_enabled(FALSE);
+       LAPIC_REG(ICRD) =
+               cpu_to_lapic[cpu] << LAPIC_ICRD_DEST_SHIFT;
+       LAPIC_REG(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
+ * get a CPU's attention it may not always issue an IPI.  If an
+ * IPI is always needed then use i386_cpu_IPI.
+ */
+void
+cpu_interrupt(int cpu)
+{
+       if (smp_initialized
+           && pmCPUExitIdle(cpu_datap(cpu))) {
+               i386_cpu_IPI(cpu);
+       }
+}
 
+/*
+ * Send a true NMI via the local APIC to the specified CPU.
+ */
+static void
+cpu_NMI_interrupt(int cpu)
+{
+       boolean_t       state;
+
+       if (smp_initialized) {
                state = ml_set_interrupts_enabled(FALSE);
+/* Program the interrupt command register */
                LAPIC_REG(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_REG(ICR)  =
-                       LAPIC_VECTOR(INTERPROCESSOR) | LAPIC_ICR_DM_FIXED;
+                       LAPIC_VECTOR(INTERPROCESSOR) | LAPIC_ICR_DM_NMI;
                (void) ml_set_interrupts_enabled(state);
        }
-
 }
 
 void
@@ -838,15 +1027,18 @@ i386_signal_cpu(int cpu, mp_event_t event, mp_sync_t mode)
 {
        volatile int    *signals = &cpu_datap(cpu)->cpu_signals;
        uint64_t        tsc_timeout;
-       
 
+       
        if (!cpu_datap(cpu)->cpu_running)
                return;
 
-       DBGLOG(cpu_signal, cpu, event);
+       if (event == MP_TLB_FLUSH)
+               KERNEL_DEBUG(0xef800020 | DBG_FUNC_START, cpu, 0, 0, 0, 0);
 
+       DBGLOG(cpu_signal, cpu, event);
+       
        i_bit_set(event, signals);
-       cpu_interrupt(cpu);
+       i386_cpu_IPI(cpu);
        if (mode == SYNC) {
           again:
                tsc_timeout = rdtsc64() + (1000*1000*1000);
@@ -859,14 +1051,22 @@ i386_signal_cpu(int cpu, mp_event_t event, mp_sync_t mode)
                        goto again;
                }
        }
+       if (event == MP_TLB_FLUSH)
+               KERNEL_DEBUG(0xef800020 | DBG_FUNC_END, cpu, 0, 0, 0, 0);
 }
 
+/*
+ * Send event to all running cpus.
+ * Called with the topology locked.
+ */
 void
 i386_signal_cpus(mp_event_t event, mp_sync_t mode)
 {
        unsigned int    cpu;
        unsigned int    my_cpu = cpu_number();
 
+       assert(hw_lock_held(&x86_topo_lock));
+
        for (cpu = 0; cpu < real_ncpus; cpu++) {
                if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
                        continue;
@@ -874,12 +1074,18 @@ i386_signal_cpus(mp_event_t event, mp_sync_t mode)
        }
 }
 
+/*
+ * Return the number of running cpus.
+ * Called with the topology locked.
+ */
 int
 i386_active_cpus(void)
 {
        unsigned int    cpu;
        unsigned int    ncpus = 0;
 
+       assert(hw_lock_held(&x86_topo_lock));
+
        for (cpu = 0; cpu < real_ncpus; cpu++) {
                if (cpu_datap(cpu)->cpu_running)
                        ncpus++;
@@ -904,24 +1110,39 @@ i386_active_cpus(void)
 static void
 mp_rendezvous_action(void)
 {
+       boolean_t intrs_enabled;
 
        /* setup function */
        if (mp_rv_setup_func != NULL)
                mp_rv_setup_func(mp_rv_func_arg);
+
+       intrs_enabled = ml_get_interrupts_enabled();
+
        /* spin on entry rendezvous */
-       atomic_incl(&mp_rv_waiters[0], 1);
-       while (*((volatile long *) &mp_rv_waiters[0]) < mp_rv_ncpus)
+       atomic_incl(&mp_rv_entry, 1);
+       while (mp_rv_entry < mp_rv_ncpus) {
+               /* poll for pesky tlb flushes if interrupts disabled */
+               if (!intrs_enabled)
+                       handle_pending_TLB_flushes();
                cpu_pause();
+       }
        /* action function */
        if (mp_rv_action_func != NULL)
                mp_rv_action_func(mp_rv_func_arg);
        /* spin on exit rendezvous */
-       atomic_incl(&mp_rv_waiters[1], 1);
-       while (*((volatile long *) &mp_rv_waiters[1]) < mp_rv_ncpus)
+       atomic_incl(&mp_rv_exit, 1);
+       while (mp_rv_exit < mp_rv_ncpus) {
+               if (!intrs_enabled)
+                       handle_pending_TLB_flushes();
                cpu_pause();
+       }
+
        /* teardown function */
        if (mp_rv_teardown_func != NULL)
                mp_rv_teardown_func(mp_rv_func_arg);
+
+       /* Bump completion count */
+       atomic_incl(&mp_rv_complete, 1);
 }
 
 void
@@ -950,26 +1171,315 @@ mp_rendezvous(void (*setup_func)(void *),
        mp_rv_teardown_func = teardown_func;
        mp_rv_func_arg = arg;
 
-       mp_rv_waiters[0] = 0;           /* entry rendezvous count */
-       mp_rv_waiters[1] = 0;           /* exit  rendezvous count */
-       mp_rv_ncpus = i386_active_cpus();
+       mp_rv_entry    = 0;
+       mp_rv_exit     = 0;
+       mp_rv_complete = 0;
 
        /*
         * signal other processors, which will call mp_rendezvous_action()
         * with interrupts disabled
         */
+       simple_lock(&x86_topo_lock);
+       mp_rv_ncpus = i386_active_cpus();
        i386_signal_cpus(MP_RENDEZVOUS, ASYNC);
+       simple_unlock(&x86_topo_lock);
 
        /* call executor function on this cpu */
        mp_rendezvous_action();
 
+       /*
+        * Spin for everyone to complete.
+        * This is necessary to ensure that all processors have proceeded
+        * from the exit barrier before we release the rendezvous structure.
+        */
+       while (mp_rv_complete < mp_rv_ncpus) {
+               cpu_pause();
+       }
+       
+       /* Tidy up */
+       mp_rv_setup_func = NULL;
+       mp_rv_action_func = NULL;
+       mp_rv_teardown_func = NULL;
+       mp_rv_func_arg = NULL;
+
        /* release lock */
        simple_unlock(&mp_rv_lock);
 }
 
+void
+mp_rendezvous_break_lock(void)
+{
+       simple_lock_init(&mp_rv_lock, 0);
+}
+
+static void
+setup_disable_intrs(__unused void * param_not_used)
+{
+       /* disable interrupts before the first barrier */
+       boolean_t intr = ml_set_interrupts_enabled(FALSE);
+
+       current_cpu_datap()->cpu_iflag = intr;
+       DBG("CPU%d: %s\n", get_cpu_number(), __FUNCTION__);
+}
+
+static void
+teardown_restore_intrs(__unused void * param_not_used)
+{
+       /* restore interrupt flag following MTRR changes */
+       ml_set_interrupts_enabled(current_cpu_datap()->cpu_iflag);
+       DBG("CPU%d: %s\n", get_cpu_number(), __FUNCTION__);
+}
+
+/*
+ * A wrapper to mp_rendezvous() to call action_func() with interrupts disabled.
+ * This is exported for use by kexts.
+ */
+void
+mp_rendezvous_no_intrs(
+             void (*action_func)(void *),
+             void *arg)
+{
+       mp_rendezvous(setup_disable_intrs,
+                     action_func,
+                     teardown_restore_intrs,
+                     arg);     
+}
+
+void
+handle_pending_TLB_flushes(void)
+{
+       volatile int    *my_word = &current_cpu_datap()->cpu_signals;
+
+       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();
+       }
+}
+
+/*
+ * This is called from cpu_signal_handler() to process an MP_CALL signal.
+ */
+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_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.
+ * 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.
+ */
+cpu_t
+mp_cpus_call(
+       cpumask_t       cpus,
+       mp_sync_t       mode,
+        void           (*action_func)(void *),
+        void           *arg)
+{
+       cpu_t           cpu;
+       boolean_t       intrs_enabled = ml_get_interrupts_enabled();
+       boolean_t       call_self = FALSE;
+
+       if (!smp_initialized) {
+               if ((cpus & CPUMASK_SELF) == 0)
+                       return 0;
+               if (action_func != NULL) {
+                       (void) ml_set_interrupts_enabled(FALSE);
+                       action_func(arg);
+                       ml_set_interrupts_enabled(intrs_enabled);
+               }
+               return 1;
+       }
+               
+       /* 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);
+       for (cpu = 0; cpu < (cpu_t) real_ncpus; cpu++) {
+               if (((cpu_to_cpumask(cpu) & cpus) == 0) ||
+                   !cpu_datap(cpu)->cpu_running)
+                       continue;
+               if (cpu == (cpu_t) cpu_number()) {
+                       /*
+                        * We don't IPI ourself and if calling asynchronously,
+                        * we defer our call until we have signalled all others.
+                        */
+                       call_self = TRUE;
+                       if (mode == SYNC && action_func != NULL) {
+                               (void) ml_set_interrupts_enabled(FALSE);
+                               action_func(arg);
+                               ml_set_interrupts_enabled(intrs_enabled);
+                       }
+               } 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.
+                        */
+                       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)
+                                               handle_pending_TLB_flushes();
+                                       cpu_pause();
+                               }
+                               simple_lock(&x86_topo_lock);
+                       }
+               }
+       }
+       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);
+                       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);
+
+       return cpu;
+}
+
+static void
+mp_broadcast_action(void)
+{
+   /* call action function */
+   if (mp_bc_action_func != NULL)
+       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));
+}
+
+/*
+ * mp_broadcast() runs a given function on all active cpus.
+ * The caller blocks until the functions has run on all cpus.
+ * The caller will also block if there is another pending braodcast.
+ */
+void
+mp_broadcast(
+         void (*action_func)(void *),
+         void *arg)
+{
+   if (!smp_initialized) {
+       if (action_func != NULL)
+                  action_func(arg);
+       return;
+   }
+       
+   /* obtain broadcast lock */
+   mutex_lock(&mp_bc_lock);
+
+   /* set static function pointers */
+   mp_bc_action_func = action_func;
+   mp_bc_func_arg = arg;
+
+   assert_wait(&mp_bc_count, THREAD_UNINT);
+
+   /*
+    * signal other processors, which will call mp_broadcast_action()
+    */
+   simple_lock(&x86_topo_lock);
+   mp_bc_ncpus = i386_active_cpus();   /* total including this cpu */
+   mp_bc_count = mp_bc_ncpus;
+   i386_signal_cpus(MP_BROADCAST, ASYNC);
+
+   /* call executor function on this cpu */
+   mp_broadcast_action();
+   simple_unlock(&x86_topo_lock);
+
+   /* block for all cpus to have run action_func */
+   if (mp_bc_ncpus > 1)
+       thread_block(THREAD_CONTINUE_NULL);
+   else
+       clear_wait(current_thread(), THREAD_AWAKENED);
+       
+   /* release lock */
+   mutex_unlock(&mp_bc_lock);
+}
+
+void
+i386_activate_cpu(void)
+{
+       cpu_data_t      *cdp = current_cpu_datap();
+
+       assert(!ml_get_interrupts_enabled());
+
+       if (!smp_initialized) {
+               cdp->cpu_running = TRUE;
+               return;
+       }
+
+       simple_lock(&x86_topo_lock);
+       cdp->cpu_running = TRUE;
+       simple_unlock(&x86_topo_lock);
+}
+
+void
+i386_deactivate_cpu(void)
+{
+       cpu_data_t      *cdp = current_cpu_datap();
+
+       assert(!ml_get_interrupts_enabled());
+
+       simple_lock(&x86_topo_lock);
+       cdp->cpu_running = FALSE;
+       simple_unlock(&x86_topo_lock);
+
+       /*
+        * In case a rendezvous/braodcast/call was initiated to this cpu
+        * before we cleared cpu_running, we must perform any actions due.
+        */
+       if (i_bit(MP_RENDEZVOUS, &cdp->cpu_signals))
+               mp_rendezvous_action();
+       if (i_bit(MP_BROADCAST, &cdp->cpu_signals))
+               mp_broadcast_action();
+       if (i_bit(MP_CALL, &cdp->cpu_signals))
+               mp_cpus_call_action();
+       cdp->cpu_signals = 0;                   /* all clear */
+}
+
+int    pmsafe_debug    = 1;
+
 #if    MACH_KDP
 volatile boolean_t     mp_kdp_trap = FALSE;
-long                   mp_kdp_ncpus;
+volatile unsigned long         mp_kdp_ncpus;
 boolean_t              mp_kdp_state;
 
 
@@ -990,65 +1500,148 @@ mp_kdp_enter(void)
         */
        mp_kdp_state = ml_set_interrupts_enabled(FALSE);
        simple_lock(&mp_kdp_lock);
+
+       if (pmsafe_debug)
+           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();
+               mp_kdp_wait(TRUE);
                simple_lock(&mp_kdp_lock);
        }
        mp_kdp_ncpus = 1;       /* self */
        mp_kdp_trap = TRUE;
        simple_unlock(&mp_kdp_lock);
 
-       /* Deliver a nudge to other cpus, counting how many */
+       /*
+        * Deliver a nudge to other cpus, counting how many
+        */
        DBG("mp_kdp_enter() signaling other processors\n");
-       for (ncpus = 1, cpu = 0; cpu < real_ncpus; cpu++) {
-               if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
-                       continue;
-               ncpus++;
-               i386_signal_cpu(cpu, MP_KDP, ASYNC); 
-       }
+       if (force_immediate_debugger_NMI == FALSE) {
+               for (ncpus = 1, cpu = 0; cpu < real_ncpus; cpu++) {
+                       if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
+                               continue;
+                       ncpus++;
+                       i386_signal_cpu(cpu, MP_KDP, ASYNC);
+               }
+               /*
+                * Wait other processors to synchronize
+                */
+               DBG("mp_kdp_enter() waiting for (%d) processors to suspend\n", ncpus);
 
-       /* Wait other processors to spin. */
-       DBG("mp_kdp_enter() waiting for (%d) processors to suspend\n", ncpus);
-       tsc_timeout = rdtsc64() + (1000*1000*1000);
-       while (*((volatile unsigned int *) &mp_kdp_ncpus) != ncpus
-               && rdtsc64() < tsc_timeout) {
-               cpu_pause();
+               /*
+                * This timeout is rather arbitrary; we don't want to NMI
+                * processors that are executing at potentially
+                * "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);
+
+               while (mp_kdp_ncpus != 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()
+                        */
+                       handle_pending_TLB_flushes();
+                       cpu_pause();
+               }
+               /* If we've timed out, and some processor(s) are still unresponsive,
+                * interrupt them with an NMI via the local APIC.
+                */
+               if (mp_kdp_ncpus != ncpus) {
+                       for (cpu = 0; cpu < real_ncpus; cpu++) {
+                               if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
+                                       continue;
+                               if (cpu_signal_pending(cpu, MP_KDP))
+                                       cpu_NMI_interrupt(cpu);
+                       }
+               }
        }
-       DBG("mp_kdp_enter() %d processors done %s\n",
-               mp_kdp_ncpus, (mp_kdp_ncpus == ncpus) ? "OK" : "timed out");
+       else
+               for (cpu = 0; cpu < real_ncpus; cpu++) {
+                       if (cpu == my_cpu || !cpu_datap(cpu)->cpu_running)
+                               continue;
+                       cpu_NMI_interrupt(cpu);
+               }
+
+       DBG("mp_kdp_enter() %u processors done %s\n",
+           mp_kdp_ncpus, (mp_kdp_ncpus == ncpus) ? "OK" : "timed out");
+       
        postcode(MP_KDP_ENTER);
 }
 
-static void
-mp_kdp_wait(void)
+static boolean_t
+cpu_signal_pending(int cpu, mp_event_t event)
 {
-       boolean_t       state;
+       volatile int    *signals = &cpu_datap(cpu)->cpu_signals;
+       boolean_t retval = FALSE;
 
-       state = ml_set_interrupts_enabled(TRUE);
+       if (i_bit(event, signals))
+               retval = TRUE;
+       return retval;
+}
+
+static void
+mp_kdp_wait(boolean_t flush)
+{
        DBG("mp_kdp_wait()\n");
-       atomic_incl(&mp_kdp_ncpus, 1);
+       /* If an I/O port has been specified as a debugging aid, issue a read */
+       panic_io_port_read();
+
+       /* 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);
+
+       atomic_incl((volatile long *)&mp_kdp_ncpus, 1);
        while (mp_kdp_trap) {
+               /*
+                * A TLB shootdown request may be pending--this would result
+                * in the requesting processor waiting in PMAP_UPDATE_TLBS()
+                * until this processor handles it.
+                * Process it, so it can now enter mp_kdp_wait()
+                */
+               if (flush)
+                       handle_pending_TLB_flushes();
                cpu_pause();
        }
-       atomic_decl(&mp_kdp_ncpus, 1);
+
+       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");
-       (void) ml_set_interrupts_enabled(state);
 }
 
 void
 mp_kdp_exit(void)
 {
        DBG("mp_kdp_exit()\n");
-       atomic_decl(&mp_kdp_ncpus, 1);
+       atomic_decl((volatile long *)&mp_kdp_ncpus, 1);
        mp_kdp_trap = FALSE;
+       __asm__ volatile("mfence");
 
        /* Wait other processors to stop spinning. XXX needs timeout */
        DBG("mp_kdp_exit() waiting for processors to resume\n");
-       while (*((volatile long *) &mp_kdp_ncpus) > 0) {
+       while (mp_kdp_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();
+
                cpu_pause();
        }
+
+       if (pmsafe_debug)
+           pmSafeMode(&current_cpu_datap()->lcpu, PM_SAFE_FL_NORMAL);
+
        DBG("mp_kdp_exit() done\n");
        (void) ml_set_interrupts_enabled(mp_kdp_state);
        postcode(0);
@@ -1073,6 +1666,7 @@ cause_ast_check(
        }
 }
 
+#if MACH_KDB
 /*
  * invoke kdb on slave processors 
  */
@@ -1082,14 +1676,56 @@ remote_kdb(void)
 {
        unsigned int    my_cpu = cpu_number();
        unsigned int    cpu;
+       int kdb_ncpus;
+       uint64_t tsc_timeout = 0;
        
-       mp_disable_preemption();
-       for (cpu = 0; cpu < real_ncpus; cpu++) {
+       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;
-               i386_signal_cpu(cpu, MP_KDB, SYNC);
+               kdb_ncpus++;
+               i386_signal_cpu(cpu, MP_KDB, ASYNC);
        }
-       mp_enable_preemption();
+       DBG("remote_kdb() waiting for (%d) processors to suspend\n",kdb_ncpus);
+
+       tsc_timeout = rdtsc64() + (kdb_ncpus * 100 * 1000 * 1000);
+
+       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()
+                */
+               handle_pending_TLB_flushes();
+
+               cpu_pause();
+       }
+       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)
+{
+       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");
 }
 
 /*
@@ -1104,9 +1740,35 @@ clear_kdb_intr(void)
        mp_enable_preemption();
 }
 
+void
+mp_kdb_exit(void)
+{
+       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();
+
+               cpu_pause();
+       }
+
+       DBG("mp_kdb_exit() done\n");
+}
+
+#endif /* MACH_KDB */
+
 /*
  * 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.
  */
 void
 i386_init_slave(void)
@@ -1119,16 +1781,28 @@ i386_init_slave(void)
        DBG("i386_init_slave() CPU%d: phys (%d) active.\n",
                get_cpu_number(), get_cpu_phys_number());
 
-       lapic_init();
+       assert(!ml_get_interrupts_enabled());
+
+       cpu_mode_init(current_cpu_datap());
 
+       mca_cpu_init();
+
+       lapic_init();
        LAPIC_DUMP();
        LAPIC_CPU_MAP_DUMP();
 
+       init_fpu();
+
        mtrr_update_cpu();
 
+       /* resume VT operation */
+       vmx_resume();
+
        pat_init();
 
-       cpu_init();
+       cpu_thread_init();      /* not strictly necessary */
+
+       cpu_init();     /* Sets cpu_running which starter cpu waits for */ 
 
        slave_main();
 
@@ -1139,19 +1813,13 @@ void
 slave_machine_init(void)
 {
        /*
-        * Here in process context.
+        * Here in process context, but with interrupts disabled.
         */
        DBG("slave_machine_init() CPU%d\n", get_cpu_number());
 
-       init_fpu();
-
-       cpu_thread_init();
-
-       pmc_init();
-
-       cpu_machine_init();
-
        clock_init();
+
+       cpu_machine_init();             /* Interrupts enabled hereafter */
 }
 
 #undef cpu_number()
@@ -1206,44 +1874,5 @@ db_trap_hist(void)
                
 }
 #endif /* TRAP_DEBUG */
-
-void db_lapic(int cpu);
-unsigned int db_remote_read(int cpu, int reg);
-void db_ioapic(unsigned int);
-void kdb_console(void);
-
-void
-kdb_console(void)
-{
-}
-
-#define BOOLP(a) ((a)?' ':'!')
-
-static char *DM[8] = {
-       "Fixed",
-       "Lowest Priority",
-       "Invalid",
-       "Invalid",
-       "NMI",
-       "Reset",
-       "Invalid",
-       "ExtINT"};
-
-unsigned int
-db_remote_read(int cpu, int reg)
-{
-       return -1;
-}
-
-void
-db_lapic(int cpu)
-{
-}
-
-void
-db_ioapic(unsigned int ind)
-{
-}
-
 #endif /* MACH_KDB */