cpumask: remove cpu_coregroup_map: x86
[safe/jmp/linux-2.6] / arch / x86 / kernel / smpboot.c
index b25097c..f534257 100644 (file)
@@ -1,8 +1,8 @@
 /*
  *     x86 SMP booting functions
  *
- *     (c) 1995 Alan Cox, Building #3 <alan@redhat.com>
- *     (c) 1998, 1999, 2000 Ingo Molnar <mingo@redhat.com>
+ *     (c) 1995 Alan Cox, Building #3 <alan@lxorguk.ukuu.org.uk>
+ *     (c) 1998, 1999, 2000, 2009 Ingo Molnar <mingo@redhat.com>
  *     Copyright 2001 Andi Kleen, SuSE Labs.
  *
  *     Much of the core SMP work is based on previous work by Thomas Radke, to
@@ -52,7 +52,7 @@
 #include <asm/desc.h>
 #include <asm/nmi.h>
 #include <asm/irq.h>
-#include <asm/smp.h>
+#include <asm/idle.h>
 #include <asm/trampoline.h>
 #include <asm/cpu.h>
 #include <asm/numa.h>
 #include <asm/tlbflush.h>
 #include <asm/mtrr.h>
 #include <asm/vmi.h>
-#include <asm/genapic.h>
+#include <asm/apic.h>
+#include <asm/setup.h>
+#include <asm/uv/uv.h>
 #include <linux/mc146818rtc.h>
 
-#include <mach_apic.h>
-#include <mach_wakecpu.h>
-#include <smpboot_hooks.h>
+#include <asm/smpboot_hooks.h>
 
 #ifdef CONFIG_X86_32
 u8 apicid_2_node[MAX_APICID];
@@ -88,7 +88,7 @@ static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
 #define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
 #define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
 #else
-struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
+static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
 #define get_idle_for_cpu(x)      (idle_thread_array[(x)])
 #define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
 #endif
@@ -100,15 +100,6 @@ EXPORT_SYMBOL(smp_num_siblings);
 /* Last level cache ID of each logical CPU */
 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
 
-/* bitmap of online cpus */
-cpumask_t cpu_online_map __read_mostly;
-EXPORT_SYMBOL(cpu_online_map);
-
-cpumask_t cpu_callin_map;
-cpumask_t cpu_callout_map;
-cpumask_t cpu_possible_map;
-EXPORT_SYMBOL(cpu_possible_map);
-
 /* representing HT siblings of each logical CPU */
 DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
@@ -121,14 +112,7 @@ EXPORT_PER_CPU_SYMBOL(cpu_core_map);
 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
 EXPORT_PER_CPU_SYMBOL(cpu_info);
 
-static atomic_t init_deasserted;
-
-
-/* representing cpus for which sibling maps can be computed */
-static cpumask_t cpu_sibling_setup_map;
-
-/* Set if we find a B stepping CPU */
-int __cpuinitdata smp_b_stepping;
+atomic_t init_deasserted;
 
 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
 
@@ -144,7 +128,7 @@ EXPORT_SYMBOL(cpu_to_node_map);
 static void map_cpu_to_node(int cpu, int node)
 {
        printk(KERN_INFO "Mapping cpu %d to node %d\n", cpu, node);
-       cpu_set(cpu, node_to_cpumask_map[node]);
+       cpumask_set_cpu(cpu, &node_to_cpumask_map[node]);
        cpu_to_node_map[cpu] = node;
 }
 
@@ -155,7 +139,7 @@ static void unmap_cpu_to_node(int cpu)
 
        printk(KERN_INFO "Unmapping cpu %d from all nodes\n", cpu);
        for (node = 0; node < MAX_NUMNODES; node++)
-               cpu_clear(cpu, node_to_cpumask_map[node]);
+               cpumask_clear_cpu(cpu, &node_to_cpumask_map[node]);
        cpu_to_node_map[cpu] = 0;
 }
 #else /* !(CONFIG_NUMA && CONFIG_X86_32) */
@@ -173,7 +157,7 @@ static void map_cpu_to_logical_apicid(void)
 {
        int cpu = smp_processor_id();
        int apicid = logical_smp_processor_id();
-       int node = apicid_to_node(apicid);
+       int node = apic->apicid_to_node(apicid);
 
        if (!node_online(node))
                node = first_online_node;
@@ -206,14 +190,15 @@ static void __cpuinit smp_callin(void)
         * our local APIC.  We have to wait for the IPI or we'll
         * lock up on an APIC access.
         */
-       wait_for_init_deassert(&init_deasserted);
+       if (apic->wait_for_init_deassert)
+               apic->wait_for_init_deassert(&init_deasserted);
 
        /*
         * (This works even if the APIC is not enabled.)
         */
        phys_id = read_apic_id();
        cpuid = smp_processor_id();
-       if (cpu_isset(cpuid, cpu_callin_map)) {
+       if (cpumask_test_cpu(cpuid, cpu_callin_mask)) {
                panic("%s: phys CPU#%d, CPU#%d already present??\n", __func__,
                                        phys_id, cpuid);
        }
@@ -235,7 +220,7 @@ static void __cpuinit smp_callin(void)
                /*
                 * Has the boot CPU finished it's STARTUP sequence?
                 */
-               if (cpu_isset(cpuid, cpu_callout_map))
+               if (cpumask_test_cpu(cpuid, cpu_callout_mask))
                        break;
                cpu_relax();
        }
@@ -253,11 +238,13 @@ static void __cpuinit smp_callin(void)
         */
 
        pr_debug("CALLIN, before setup_local_APIC().\n");
-       smp_callin_clear_local_apic();
+       if (apic->smp_callin_clear_local_apic)
+               apic->smp_callin_clear_local_apic();
        setup_local_APIC();
        end_local_APIC_setup();
        map_cpu_to_logical_apicid();
 
+       notify_cpu_starting(cpuid);
        /*
         * Get our bogomips.
         *
@@ -277,22 +264,20 @@ static void __cpuinit smp_callin(void)
        /*
         * Allow the master to continue.
         */
-       cpu_set(cpuid, cpu_callin_map);
+       cpumask_set_cpu(cpuid, cpu_callin_mask);
 }
 
 /*
  * Activate a secondary processor.
  */
-static void __cpuinit start_secondary(void *unused)
+notrace static void __cpuinit start_secondary(void *unused)
 {
        /*
         * Don't put *anything* before cpu_init(), SMP booting is too
         * fragile that we want to limit the things done here to the
         * most necessary things.
         */
-#ifdef CONFIG_VMI
        vmi_bringup();
-#endif
        cpu_init();
        preempt_disable();
        smp_callin();
@@ -332,92 +317,23 @@ static void __cpuinit start_secondary(void *unused)
         * does not change while we are assigning vectors to cpus.  Holding
         * this lock ensures we don't half assign or remove an irq from a cpu.
         */
-       ipi_call_lock_irq();
+       ipi_call_lock();
        lock_vector_lock();
        __setup_vector_irq(smp_processor_id());
-       cpu_set(smp_processor_id(), cpu_online_map);
+       set_cpu_online(smp_processor_id(), true);
        unlock_vector_lock();
-       ipi_call_unlock_irq();
+       ipi_call_unlock();
        per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
 
+       /* enable local interrupts */
+       local_irq_enable();
+
        setup_secondary_clock();
 
        wmb();
        cpu_idle();
 }
 
-static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
-{
-       /*
-        * Mask B, Pentium, but not Pentium MMX
-        */
-       if (c->x86_vendor == X86_VENDOR_INTEL &&
-           c->x86 == 5 &&
-           c->x86_mask >= 1 && c->x86_mask <= 4 &&
-           c->x86_model <= 3)
-               /*
-                * Remember we have B step Pentia with bugs
-                */
-               smp_b_stepping = 1;
-
-       /*
-        * Certain Athlons might work (for various values of 'work') in SMP
-        * but they are not certified as MP capable.
-        */
-       if ((c->x86_vendor == X86_VENDOR_AMD) && (c->x86 == 6)) {
-
-               if (num_possible_cpus() == 1)
-                       goto valid_k7;
-
-               /* Athlon 660/661 is valid. */
-               if ((c->x86_model == 6) && ((c->x86_mask == 0) ||
-                   (c->x86_mask == 1)))
-                       goto valid_k7;
-
-               /* Duron 670 is valid */
-               if ((c->x86_model == 7) && (c->x86_mask == 0))
-                       goto valid_k7;
-
-               /*
-                * Athlon 662, Duron 671, and Athlon >model 7 have capability
-                * bit. It's worth noting that the A5 stepping (662) of some
-                * Athlon XP's have the MP bit set.
-                * See http://www.heise.de/newsticker/data/jow-18.10.01-000 for
-                * more.
-                */
-               if (((c->x86_model == 6) && (c->x86_mask >= 2)) ||
-                   ((c->x86_model == 7) && (c->x86_mask >= 1)) ||
-                    (c->x86_model > 7))
-                       if (cpu_has_mp)
-                               goto valid_k7;
-
-               /* If we get here, not a certified SMP capable AMD system. */
-               add_taint(TAINT_UNSAFE_SMP);
-       }
-
-valid_k7:
-       ;
-}
-
-static void __cpuinit smp_checks(void)
-{
-       if (smp_b_stepping)
-               printk(KERN_WARNING "WARNING: SMP operation may be unreliable"
-                                   "with B stepping processors.\n");
-
-       /*
-        * Don't taint if we are running SMP kernel on a single non-MP
-        * approved Athlon
-        */
-       if (tainted & TAINT_UNSAFE_SMP) {
-               if (num_online_cpus())
-                       printk(KERN_INFO "WARNING: This combination of AMD"
-                               "processors is not suitable for SMP.\n");
-               else
-                       tainted &= ~TAINT_UNSAFE_SMP;
-       }
-}
-
 /*
  * The bootstrap kernel entry code has set these up. Save them for
  * a given CPU
@@ -431,7 +347,6 @@ void __cpuinit smp_store_cpu_info(int id)
        c->cpu_index = id;
        if (id != 0)
                identify_secondary_cpu(c);
-       smp_apply_quirks(c);
 }
 
 
@@ -440,50 +355,52 @@ void __cpuinit set_cpu_sibling_map(int cpu)
        int i;
        struct cpuinfo_x86 *c = &cpu_data(cpu);
 
-       cpu_set(cpu, cpu_sibling_setup_map);
+       cpumask_set_cpu(cpu, cpu_sibling_setup_mask);
 
        if (smp_num_siblings > 1) {
-               for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
-                       if (c->phys_proc_id == cpu_data(i).phys_proc_id &&
-                           c->cpu_core_id == cpu_data(i).cpu_core_id) {
-                               cpu_set(i, per_cpu(cpu_sibling_map, cpu));
-                               cpu_set(cpu, per_cpu(cpu_sibling_map, i));
-                               cpu_set(i, per_cpu(cpu_core_map, cpu));
-                               cpu_set(cpu, per_cpu(cpu_core_map, i));
-                               cpu_set(i, c->llc_shared_map);
-                               cpu_set(cpu, cpu_data(i).llc_shared_map);
+               for_each_cpu(i, cpu_sibling_setup_mask) {
+                       struct cpuinfo_x86 *o = &cpu_data(i);
+
+                       if (c->phys_proc_id == o->phys_proc_id &&
+                           c->cpu_core_id == o->cpu_core_id) {
+                               cpumask_set_cpu(i, cpu_sibling_mask(cpu));
+                               cpumask_set_cpu(cpu, cpu_sibling_mask(i));
+                               cpumask_set_cpu(i, cpu_core_mask(cpu));
+                               cpumask_set_cpu(cpu, cpu_core_mask(i));
+                               cpumask_set_cpu(i, &c->llc_shared_map);
+                               cpumask_set_cpu(cpu, &o->llc_shared_map);
                        }
                }
        } else {
-               cpu_set(cpu, per_cpu(cpu_sibling_map, cpu));
+               cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
        }
 
-       cpu_set(cpu, c->llc_shared_map);
+       cpumask_set_cpu(cpu, &c->llc_shared_map);
 
        if (current_cpu_data.x86_max_cores == 1) {
-               per_cpu(cpu_core_map, cpu) = per_cpu(cpu_sibling_map, cpu);
+               cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
                c->booted_cores = 1;
                return;
        }
 
-       for_each_cpu_mask_nr(i, cpu_sibling_setup_map) {
+       for_each_cpu(i, cpu_sibling_setup_mask) {
                if (per_cpu(cpu_llc_id, cpu) != BAD_APICID &&
                    per_cpu(cpu_llc_id, cpu) == per_cpu(cpu_llc_id, i)) {
-                       cpu_set(i, c->llc_shared_map);
-                       cpu_set(cpu, cpu_data(i).llc_shared_map);
+                       cpumask_set_cpu(i, &c->llc_shared_map);
+                       cpumask_set_cpu(cpu, &cpu_data(i).llc_shared_map);
                }
                if (c->phys_proc_id == cpu_data(i).phys_proc_id) {
-                       cpu_set(i, per_cpu(cpu_core_map, cpu));
-                       cpu_set(cpu, per_cpu(cpu_core_map, i));
+                       cpumask_set_cpu(i, cpu_core_mask(cpu));
+                       cpumask_set_cpu(cpu, cpu_core_mask(i));
                        /*
                         *  Does this new cpu bringup a new core?
                         */
-                       if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1) {
+                       if (cpumask_weight(cpu_sibling_mask(cpu)) == 1) {
                                /*
                                 * for each core in package, increment
                                 * the booted_cores for this new cpu
                                 */
-                               if (first_cpu(per_cpu(cpu_sibling_map, i)) == i)
+                               if (cpumask_first(cpu_sibling_mask(i)) == i)
                                        c->booted_cores++;
                                /*
                                 * increment the core count for all
@@ -498,7 +415,7 @@ void __cpuinit set_cpu_sibling_map(int cpu)
 }
 
 /* maps the cpu to the sched domain representing multi-core */
-cpumask_t cpu_coregroup_map(int cpu)
+const struct cpumask *cpu_coregroup_mask(int cpu)
 {
        struct cpuinfo_x86 *c = &cpu_data(cpu);
        /*
@@ -506,9 +423,9 @@ cpumask_t cpu_coregroup_map(int cpu)
         * And for power savings, we return cpu_core_map
         */
        if (sched_mc_power_savings || sched_smt_power_savings)
-               return per_cpu(cpu_core_map, cpu);
+               return cpu_core_mask(cpu);
        else
-               return c->llc_shared_map;
+               return &c->llc_shared_map;
 }
 
 static void impress_friends(void)
@@ -520,7 +437,7 @@ static void impress_friends(void)
         */
        pr_debug("Before bogomips.\n");
        for_each_possible_cpu(cpu)
-               if (cpu_isset(cpu, cpu_callout_map))
+               if (cpumask_test_cpu(cpu, cpu_callout_mask))
                        bogosum += cpu_data(cpu).loops_per_jiffy;
        printk(KERN_INFO
                "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
@@ -531,17 +448,17 @@ static void impress_friends(void)
        pr_debug("Before bogocount - setting activated=1.\n");
 }
 
-static inline void __inquire_remote_apic(int apicid)
+void __inquire_remote_apic(int apicid)
 {
        unsigned i, regs[] = { APIC_ID >> 4, APIC_LVR >> 4, APIC_SPIV >> 4 };
        char *names[] = { "ID", "VERSION", "SPIV" };
        int timeout;
        u32 status;
 
-       printk(KERN_INFO "Inquiring remote APIC #%d...\n", apicid);
+       printk(KERN_INFO "Inquiring remote APIC 0x%x...\n", apicid);
 
        for (i = 0; i < ARRAY_SIZE(regs); i++) {
-               printk(KERN_INFO "... APIC #%d %s: ", apicid, names[i]);
+               printk(KERN_INFO "... APIC 0x%x %s: ", apicid, names[i]);
 
                /*
                 * Wait for idle.
@@ -570,14 +487,13 @@ static inline void __inquire_remote_apic(int apicid)
        }
 }
 
-#ifdef WAKE_SECONDARY_VIA_NMI
 /*
  * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
  * won't ... remember to clear down the APIC, etc later.
  */
-static int __devinit
-wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
+int __devinit
+wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
 {
        unsigned long send_status, accept_status = 0;
        int maxlvt;
@@ -585,7 +501,7 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
        /* Target chip */
        /* Boot on the stack */
        /* Kick the second */
-       apic_icr_write(APIC_DM_NMI | APIC_DEST_LOGICAL, logical_apicid);
+       apic_icr_write(APIC_DM_NMI | apic->dest_logical, logical_apicid);
 
        pr_debug("Waiting for send to finish...\n");
        send_status = safe_apic_wait_icr_idle();
@@ -594,10 +510,12 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
         * Give the other CPU some time to accept the IPI.
         */
        udelay(200);
-       maxlvt = lapic_get_maxlvt();
-       if (maxlvt > 3)                 /* Due to the Pentium erratum 3AP.  */
-               apic_write(APIC_ESR, 0);
-       accept_status = (apic_read(APIC_ESR) & 0xEF);
+       if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
+               maxlvt = lapic_get_maxlvt();
+               if (maxlvt > 3)                 /* Due to the Pentium erratum 3AP.  */
+                       apic_write(APIC_ESR, 0);
+               accept_status = (apic_read(APIC_ESR) & 0xEF);
+       }
        pr_debug("NMI sent.\n");
 
        if (send_status)
@@ -607,21 +525,13 @@ wakeup_secondary_cpu(int logical_apicid, unsigned long start_eip)
 
        return (send_status | accept_status);
 }
-#endif /* WAKE_SECONDARY_VIA_NMI */
 
-#ifdef WAKE_SECONDARY_VIA_INIT
-static int __devinit
-wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
+int __devinit
+wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
 {
        unsigned long send_status, accept_status = 0;
        int maxlvt, num_starts, j;
 
-       if (get_uv_system_type() == UV_NON_UNIQUE_APIC) {
-               send_status = uv_wakeup_secondary(phys_apicid, start_eip);
-               atomic_set(&init_deasserted, 1);
-               return send_status;
-       }
-
        maxlvt = lapic_get_maxlvt();
 
        /*
@@ -730,7 +640,6 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
 
        return (send_status | accept_status);
 }
-#endif /* WAKE_SECONDARY_VIA_INIT */
 
 struct create_idle {
        struct work_struct work;
@@ -748,78 +657,23 @@ static void __cpuinit do_fork_idle(struct work_struct *work)
        complete(&c_idle->done);
 }
 
-#ifdef CONFIG_X86_64
-
-/* __ref because it's safe to call free_bootmem when after_bootmem == 0. */
-static void __ref free_bootmem_pda(struct x8664_pda *oldpda)
-{
-       if (!after_bootmem)
-               free_bootmem((unsigned long)oldpda, sizeof(*oldpda));
-}
-
-/*
- * Allocate node local memory for the AP pda.
- *
- * Must be called after the _cpu_pda pointer table is initialized.
- */
-int __cpuinit get_local_pda(int cpu)
-{
-       struct x8664_pda *oldpda, *newpda;
-       unsigned long size = sizeof(struct x8664_pda);
-       int node = cpu_to_node(cpu);
-
-       if (cpu_pda(cpu) && !cpu_pda(cpu)->in_bootmem)
-               return 0;
-
-       oldpda = cpu_pda(cpu);
-       newpda = kmalloc_node(size, GFP_ATOMIC, node);
-       if (!newpda) {
-               printk(KERN_ERR "Could not allocate node local PDA "
-                       "for CPU %d on node %d\n", cpu, node);
-
-               if (oldpda)
-                       return 0;       /* have a usable pda */
-               else
-                       return -1;
-       }
-
-       if (oldpda) {
-               memcpy(newpda, oldpda, size);
-               free_bootmem_pda(oldpda);
-       }
-
-       newpda->in_bootmem = 0;
-       cpu_pda(cpu) = newpda;
-       return 0;
-}
-#endif /* CONFIG_X86_64 */
-
-static int __cpuinit do_boot_cpu(int apicid, int cpu)
 /*
  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
- * Returns zero if CPU booted OK, else error code from wakeup_secondary_cpu.
+ * Returns zero if CPU booted OK, else error code from
+ * ->wakeup_secondary_cpu.
  */
+static int __cpuinit do_boot_cpu(int apicid, int cpu)
 {
        unsigned long boot_error = 0;
-       int timeout;
        unsigned long start_ip;
-       unsigned short nmi_high = 0, nmi_low = 0;
+       int timeout;
        struct create_idle c_idle = {
-               .cpu = cpu,
-               .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
+               .cpu    = cpu,
+               .done   = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
        };
-       INIT_WORK(&c_idle.work, do_fork_idle);
 
-#ifdef CONFIG_X86_64
-       /* Allocate node local memory for AP pdas */
-       if (cpu > 0) {
-               boot_error = get_local_pda(cpu);
-               if (boot_error)
-                       goto restore_state;
-                       /* if can't get pda memory, can't start cpu */
-       }
-#endif
+       INIT_WORK(&c_idle.work, do_fork_idle);
 
        alternatives_smp_switch(1);
 
@@ -850,14 +704,16 @@ static int __cpuinit do_boot_cpu(int apicid, int cpu)
 
        set_idle_for_cpu(cpu, c_idle.idle);
 do_rest:
-#ifdef CONFIG_X86_32
        per_cpu(current_task, cpu) = c_idle.idle;
-       init_gdt(cpu);
+#ifdef CONFIG_X86_32
        /* Stack for startup_32 can be just as for start_secondary onwards */
        irq_ctx_init(cpu);
 #else
-       cpu_pda(cpu)->pcurrent = c_idle.idle;
        clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
+       initial_gs = per_cpu_offset(cpu);
+       per_cpu(kernel_stack, cpu) =
+               (unsigned long)task_stack_page(c_idle.idle) -
+               KERNEL_STACK_OFFSET + THREAD_SIZE;
 #endif
        early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
        initial_code = (unsigned long)start_secondary;
@@ -867,7 +723,7 @@ do_rest:
        start_ip = setup_trampoline();
 
        /* So we see what's up   */
-       printk(KERN_INFO "Booting processor %d/%d ip %lx\n",
+       printk(KERN_INFO "Booting processor %d APIC 0x%x ip 0x%lx\n",
                          cpu, apicid, start_ip);
 
        /*
@@ -881,39 +737,43 @@ do_rest:
 
                pr_debug("Setting warm reset code and vector.\n");
 
-               store_NMI_vector(&nmi_high, &nmi_low);
-
                smpboot_setup_warm_reset_vector(start_ip);
                /*
                 * Be paranoid about clearing APIC errors.
-               */
-               apic_write(APIC_ESR, 0);
-               apic_read(APIC_ESR);
+               */
+               if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
+                       apic_write(APIC_ESR, 0);
+                       apic_read(APIC_ESR);
+               }
        }
 
        /*
-        * Starting actual IPI sequence...
+        * Kick the secondary CPU. Use the method in the APIC driver
+        * if it's defined - or use an INIT boot APIC message otherwise:
         */
-       boot_error = wakeup_secondary_cpu(apicid, start_ip);
+       if (apic->wakeup_secondary_cpu)
+               boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
+       else
+               boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
 
        if (!boot_error) {
                /*
                 * allow APs to start initializing.
                 */
                pr_debug("Before Callout %d.\n", cpu);
-               cpu_set(cpu, cpu_callout_map);
+               cpumask_set_cpu(cpu, cpu_callout_mask);
                pr_debug("After Callout %d.\n", cpu);
 
                /*
                 * Wait 5s total for a response
                 */
                for (timeout = 0; timeout < 50000; timeout++) {
-                       if (cpu_isset(cpu, cpu_callin_map))
+                       if (cpumask_test_cpu(cpu, cpu_callin_mask))
                                break;  /* It has booted */
                        udelay(100);
                }
 
-               if (cpu_isset(cpu, cpu_callin_map)) {
+               if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
                        /* number CPUs logically, starting from 1 (BSP is 0) */
                        pr_debug("OK.\n");
                        printk(KERN_INFO "CPU%d: ", cpu);
@@ -928,19 +788,22 @@ do_rest:
                        else
                                /* trampoline code not run */
                                printk(KERN_ERR "Not responding.\n");
-                       if (get_uv_system_type() != UV_NON_UNIQUE_APIC)
-                               inquire_remote_apic(apicid);
+                       if (apic->inquire_remote_apic)
+                               apic->inquire_remote_apic(apicid);
                }
        }
-#ifdef CONFIG_X86_64
-restore_state:
-#endif
+
        if (boot_error) {
                /* Try to put things back the way they were before ... */
                numa_remove_cpu(cpu); /* was set by numa_add_cpu */
-               cpu_clear(cpu, cpu_callout_map); /* was set by do_boot_cpu() */
-               cpu_clear(cpu, cpu_initialized); /* was set by cpu_init() */
-               cpu_clear(cpu, cpu_present_map);
+
+               /* was set by do_boot_cpu() */
+               cpumask_clear_cpu(cpu, cpu_callout_mask);
+
+               /* was set by cpu_init() */
+               cpumask_clear_cpu(cpu, cpu_initialized_mask);
+
+               set_cpu_present(cpu, false);
                per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
        }
 
@@ -957,7 +820,7 @@ restore_state:
 
 int __cpuinit native_cpu_up(unsigned int cpu)
 {
-       int apicid = cpu_present_to_apicid(cpu);
+       int apicid = apic->cpu_present_to_apicid(cpu);
        unsigned long flags;
        int err;
 
@@ -974,7 +837,7 @@ int __cpuinit native_cpu_up(unsigned int cpu)
        /*
         * Already booted CPU?
         */
-       if (cpu_isset(cpu, cpu_callin_map)) {
+       if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
                pr_debug("do_boot_cpu %d Already started\n", cpu);
                return -ENOSYS;
        }
@@ -1029,8 +892,9 @@ int __cpuinit native_cpu_up(unsigned int cpu)
  */
 static __init void disable_smp(void)
 {
-       cpu_present_map = cpumask_of_cpu(0);
-       cpu_possible_map = cpumask_of_cpu(0);
+       /* use the read/write pointers to the present and possible maps */
+       cpumask_copy(&cpu_present_map, cpumask_of(0));
+       cpumask_copy(&cpu_possible_map, cpumask_of(0));
        smpboot_clear_io_apic_irqs();
 
        if (smp_found_config)
@@ -1038,8 +902,8 @@ static __init void disable_smp(void)
        else
                physid_set_mask_of_physid(0, &phys_cpu_present_map);
        map_cpu_to_logical_apicid();
-       cpu_set(0, per_cpu(cpu_sibling_map, 0));
-       cpu_set(0, per_cpu(cpu_core_map, 0));
+       cpumask_set_cpu(0, cpu_sibling_mask(0));
+       cpumask_set_cpu(0, cpu_core_mask(0));
 }
 
 /*
@@ -1049,26 +913,26 @@ static int __init smp_sanity_check(unsigned max_cpus)
 {
        preempt_disable();
 
-#if defined(CONFIG_X86_PC) && defined(CONFIG_X86_32)
+#if !defined(CONFIG_X86_BIGSMP) && defined(CONFIG_X86_32)
        if (def_to_bigsmp && nr_cpu_ids > 8) {
                unsigned int cpu;
                unsigned nr;
 
                printk(KERN_WARNING
                       "More than 8 CPUs detected - skipping them.\n"
-                      "Use CONFIG_X86_GENERICARCH and CONFIG_X86_BIGSMP.\n");
+                      "Use CONFIG_X86_BIGSMP.\n");
 
                nr = 0;
                for_each_present_cpu(cpu) {
                        if (nr >= 8)
-                               cpu_clear(cpu, cpu_present_map);
+                               set_cpu_present(cpu, false);
                        nr++;
                }
 
                nr = 0;
                for_each_possible_cpu(cpu) {
                        if (nr >= 8)
-                               cpu_clear(cpu, cpu_possible_map);
+                               set_cpu_possible(cpu, false);
                        nr++;
                }
 
@@ -1077,8 +941,10 @@ static int __init smp_sanity_check(unsigned max_cpus)
 #endif
 
        if (!physid_isset(hard_smp_processor_id(), phys_cpu_present_map)) {
-               printk(KERN_WARNING "weird, boot CPU (#%d) not listed"
-                                   "by the BIOS.\n", hard_smp_processor_id());
+               printk(KERN_WARNING
+                       "weird, boot CPU (#%d) not listed by the BIOS.\n",
+                       hard_smp_processor_id());
+
                physid_set(hard_smp_processor_id(), phys_cpu_present_map);
        }
 
@@ -1100,7 +966,7 @@ static int __init smp_sanity_check(unsigned max_cpus)
         * Should not be necessary because the MP table should list the boot
         * CPU too, but we do it for the sake of robustness anyway.
         */
-       if (!check_phys_apicid_present(boot_cpu_physical_apicid)) {
+       if (!apic->check_phys_apicid_present(boot_cpu_physical_apicid)) {
                printk(KERN_NOTICE
                        "weird, boot CPU (#%d) not listed by the BIOS.\n",
                        boot_cpu_physical_apicid);
@@ -1118,6 +984,7 @@ static int __init smp_sanity_check(unsigned max_cpus)
                printk(KERN_ERR "... forcing use of dummy APIC emulation."
                                "(tell your hw vendor)\n");
                smpboot_clear_io_apic();
+               arch_disable_smp_support();
                return -1;
        }
 
@@ -1149,7 +1016,7 @@ static void __init smp_cpu_index_default(void)
        for_each_possible_cpu(i) {
                c = &cpu_data(i);
                /* mark all to hotplug */
-               c->cpu_index = NR_CPUS;
+               c->cpu_index = nr_cpu_ids;
        }
 }
 
@@ -1162,7 +1029,7 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
        preempt_disable();
        smp_cpu_index_default();
        current_cpu_data = boot_cpu_data;
-       cpu_callin_map = cpumask_of_cpu(0);
+       cpumask_copy(cpu_callin_mask, cpumask_of(0));
        mb();
        /*
         * Setup boot CPU information
@@ -1174,9 +1041,9 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
        current_thread_info()->cpu = 0;  /* needed? */
        set_cpu_sibling_map(0);
 
-#ifdef CONFIG_X86_64
        enable_IR_x2apic();
-       setup_apic_routing();
+#ifdef CONFIG_X86_64
+       default_setup_apic_routing();
 #endif
 
        if (smp_sanity_check(max_cpus) < 0) {
@@ -1200,18 +1067,18 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
         */
        setup_local_APIC();
 
-#ifdef CONFIG_X86_64
        /*
         * Enable IO APIC before setting up error vector
         */
        if (!skip_ioapic_setup && nr_ioapics)
                enable_IO_APIC();
-#endif
+
        end_local_APIC_setup();
 
        map_cpu_to_logical_apicid();
 
-       setup_portio_remap();
+       if (apic->setup_portio_remap)
+               apic->setup_portio_remap();
 
        smpboot_setup_io_apic();
        /*
@@ -1233,12 +1100,9 @@ out:
 void __init native_smp_prepare_boot_cpu(void)
 {
        int me = smp_processor_id();
-#ifdef CONFIG_X86_32
-       init_gdt(me);
-#endif
-       switch_to_new_gdt();
-       /* already set me in cpu_online_map in boot_cpu_init() */
-       cpu_set(me, cpu_callout_map);
+       switch_to_new_gdt(me);
+       /* already set me in cpu_online_mask in boot_cpu_init() */
+       cpumask_set_cpu(me, cpu_callout_mask);
        per_cpu(cpu_state, me) = CPU_ONLINE;
 }
 
@@ -1247,45 +1111,20 @@ void __init native_smp_cpus_done(unsigned int max_cpus)
        pr_debug("Boot done.\n");
 
        impress_friends();
-       smp_checks();
 #ifdef CONFIG_X86_IO_APIC
        setup_ioapic_dest();
 #endif
        check_nmi_watchdog();
 }
 
-#ifdef CONFIG_HOTPLUG_CPU
-
-static void remove_siblinginfo(int cpu)
+static int __initdata setup_possible_cpus = -1;
+static int __init _setup_possible_cpus(char *str)
 {
-       int sibling;
-       struct cpuinfo_x86 *c = &cpu_data(cpu);
-
-       for_each_cpu_mask_nr(sibling, per_cpu(cpu_core_map, cpu)) {
-               cpu_clear(cpu, per_cpu(cpu_core_map, sibling));
-               /*/
-                * last thread sibling in this cpu core going down
-                */
-               if (cpus_weight(per_cpu(cpu_sibling_map, cpu)) == 1)
-                       cpu_data(sibling).booted_cores--;
-       }
-
-       for_each_cpu_mask_nr(sibling, per_cpu(cpu_sibling_map, cpu))
-               cpu_clear(cpu, per_cpu(cpu_sibling_map, sibling));
-       cpus_clear(per_cpu(cpu_sibling_map, cpu));
-       cpus_clear(per_cpu(cpu_core_map, cpu));
-       c->phys_proc_id = 0;
-       c->cpu_core_id = 0;
-       cpu_clear(cpu, cpu_sibling_setup_map);
+       get_option(&str, &setup_possible_cpus);
+       return 0;
 }
+early_param("possible_cpus", _setup_possible_cpus);
 
-static int additional_cpus __initdata = -1;
-
-static __init int setup_additional_cpus(char *s)
-{
-       return s && get_option(&s, &additional_cpus) ? 0 : -EINVAL;
-}
-early_param("additional_cpus", setup_additional_cpus);
 
 /*
  * cpu_possible_map should be static, it cannot change as cpu's
@@ -1299,72 +1138,80 @@ early_param("additional_cpus", setup_additional_cpus);
  *
  * Three ways to find out the number of additional hotplug CPUs:
  * - If the BIOS specified disabled CPUs in ACPI/mptables use that.
- * - The user can overwrite it with additional_cpus=NUM
+ * - The user can overwrite it with possible_cpus=NUM
  * - Otherwise don't reserve additional CPUs.
  * We do this because additional CPUs waste a lot of memory.
  * -AK
  */
 __init void prefill_possible_map(void)
 {
-       int i;
-       int possible;
+       int i, possible;
 
        /* no processor from mptable or madt */
        if (!num_processors)
                num_processors = 1;
 
-#ifdef CONFIG_HOTPLUG_CPU
-       if (additional_cpus == -1) {
-               if (disabled_cpus > 0)
-                       additional_cpus = disabled_cpus;
-               else
-                       additional_cpus = 0;
+       if (setup_possible_cpus == -1)
+               possible = num_processors + disabled_cpus;
+       else
+               possible = setup_possible_cpus;
+
+       total_cpus = max_t(int, possible, num_processors + disabled_cpus);
+
+       if (possible > CONFIG_NR_CPUS) {
+               printk(KERN_WARNING
+                       "%d Processors exceeds NR_CPUS limit of %d\n",
+                       possible, CONFIG_NR_CPUS);
+               possible = CONFIG_NR_CPUS;
        }
-#else
-       additional_cpus = 0;
-#endif
-       possible = num_processors + additional_cpus;
-       if (possible > NR_CPUS)
-               possible = NR_CPUS;
 
        printk(KERN_INFO "SMP: Allowing %d CPUs, %d hotplug CPUs\n",
                possible, max_t(int, possible - num_processors, 0));
 
        for (i = 0; i < possible; i++)
-               cpu_set(i, cpu_possible_map);
+               set_cpu_possible(i, true);
 
        nr_cpu_ids = possible;
 }
 
+#ifdef CONFIG_HOTPLUG_CPU
+
+static void remove_siblinginfo(int cpu)
+{
+       int sibling;
+       struct cpuinfo_x86 *c = &cpu_data(cpu);
+
+       for_each_cpu(sibling, cpu_core_mask(cpu)) {
+               cpumask_clear_cpu(cpu, cpu_core_mask(sibling));
+               /*/
+                * last thread sibling in this cpu core going down
+                */
+               if (cpumask_weight(cpu_sibling_mask(cpu)) == 1)
+                       cpu_data(sibling).booted_cores--;
+       }
+
+       for_each_cpu(sibling, cpu_sibling_mask(cpu))
+               cpumask_clear_cpu(cpu, cpu_sibling_mask(sibling));
+       cpumask_clear(cpu_sibling_mask(cpu));
+       cpumask_clear(cpu_core_mask(cpu));
+       c->phys_proc_id = 0;
+       c->cpu_core_id = 0;
+       cpumask_clear_cpu(cpu, cpu_sibling_setup_mask);
+}
+
 static void __ref remove_cpu_from_maps(int cpu)
 {
-       cpu_clear(cpu, cpu_online_map);
-       cpu_clear(cpu, cpu_callout_map);
-       cpu_clear(cpu, cpu_callin_map);
+       set_cpu_online(cpu, false);
+       cpumask_clear_cpu(cpu, cpu_callout_mask);
+       cpumask_clear_cpu(cpu, cpu_callin_mask);
        /* was set by cpu_init() */
-       cpu_clear(cpu, cpu_initialized);
+       cpumask_clear_cpu(cpu, cpu_initialized_mask);
        numa_remove_cpu(cpu);
 }
 
-int __cpu_disable(void)
+void cpu_disable_common(void)
 {
        int cpu = smp_processor_id();
-
-       /*
-        * Perhaps use cpufreq to drop frequency, but that could go
-        * into generic code.
-        *
-        * We won't take down the boot processor on i386 due to some
-        * interrupts only being able to be serviced by the BSP.
-        * Especially so if we're not using an IOAPIC   -zwane
-        */
-       if (cpu == 0)
-               return -EBUSY;
-
-       if (nmi_watchdog == NMI_LOCAL_APIC)
-               stop_apic_nmi_watchdog(NULL);
-       clear_local_APIC();
-
        /*
         * HACK:
         * Allow any queued timer interrupts to get serviced
@@ -1381,11 +1228,33 @@ int __cpu_disable(void)
        lock_vector_lock();
        remove_cpu_from_maps(cpu);
        unlock_vector_lock();
-       fixup_irqs(cpu_online_map);
+       fixup_irqs();
+}
+
+int native_cpu_disable(void)
+{
+       int cpu = smp_processor_id();
+
+       /*
+        * Perhaps use cpufreq to drop frequency, but that could go
+        * into generic code.
+        *
+        * We won't take down the boot processor on i386 due to some
+        * interrupts only being able to be serviced by the BSP.
+        * Especially so if we're not using an IOAPIC   -zwane
+        */
+       if (cpu == 0)
+               return -EBUSY;
+
+       if (nmi_watchdog == NMI_LOCAL_APIC)
+               stop_apic_nmi_watchdog(NULL);
+       clear_local_APIC();
+
+       cpu_disable_common();
        return 0;
 }
 
-void __cpu_die(unsigned int cpu)
+void native_cpu_die(unsigned int cpu)
 {
        /* We don't do anything here: idle task is faking death itself. */
        unsigned int i;
@@ -1402,15 +1271,45 @@ void __cpu_die(unsigned int cpu)
        }
        printk(KERN_ERR "CPU %u didn't die...\n", cpu);
 }
+
+void play_dead_common(void)
+{
+       idle_task_exit();
+       reset_lazy_tlbstate();
+       irq_ctx_exit(raw_smp_processor_id());
+       c1e_remove_cpu(raw_smp_processor_id());
+
+       mb();
+       /* Ack it */
+       __get_cpu_var(cpu_state) = CPU_DEAD;
+
+       /*
+        * With physical CPU hotplug, we should halt the cpu
+        */
+       local_irq_disable();
+}
+
+void native_play_dead(void)
+{
+       play_dead_common();
+       wbinvd_halt();
+}
+
 #else /* ... !CONFIG_HOTPLUG_CPU */
-int __cpu_disable(void)
+int native_cpu_disable(void)
 {
        return -ENOSYS;
 }
 
-void __cpu_die(unsigned int cpu)
+void native_cpu_die(unsigned int cpu)
 {
        /* We said "no" in __cpu_disable */
        BUG();
 }
+
+void native_play_dead(void)
+{
+       BUG();
+}
+
 #endif