Merge branch 'generic-ipi' into generic-ipi-for-linus
[safe/jmp/linux-2.6] / arch / x86 / kernel / smpboot.c
index 8964789..687376a 100644 (file)
@@ -59,7 +59,6 @@
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>
 #include <asm/mtrr.h>
-#include <asm/nmi.h>
 #include <asm/vmi.h>
 #include <asm/genapic.h>
 #include <linux/mc146818rtc.h>
 #include <mach_wakecpu.h>
 #include <smpboot_hooks.h>
 
-/*
- * FIXME: For x86_64, those are defined in other files. But moving them here,
- * would make the setup areas dependent on smp, which is a loss. When we
- * integrate apic between arches, we can probably do a better job, but
- * right now, they'll stay here -- glommer
- */
-
-/* which logical CPU number maps to which CPU (physical APIC ID) */
-u16 x86_cpu_to_apicid_init[NR_CPUS] __initdata =
-                       { [0 ... NR_CPUS-1] = BAD_APICID };
-void *x86_cpu_to_apicid_early_ptr;
-
-u16 x86_bios_cpu_apicid_init[NR_CPUS] __initdata
-                               = { [0 ... NR_CPUS-1] = BAD_APICID };
-void *x86_bios_cpu_apicid_early_ptr;
-
 #ifdef CONFIG_X86_32
 u8 apicid_2_node[MAX_APICID];
 static int low_mappings;
@@ -198,13 +181,12 @@ static void map_cpu_to_logical_apicid(void)
        map_cpu_to_node(cpu, node);
 }
 
-static void unmap_cpu_to_logical_apicid(int cpu)
+void numa_remove_cpu(int cpu)
 {
        cpu_2_logical_apicid[cpu] = BAD_APICID;
        unmap_cpu_to_node(cpu);
 }
 #else
-#define unmap_cpu_to_logical_apicid(cpu) do {} while (0)
 #define map_cpu_to_logical_apicid()  do {} while (0)
 #endif
 
@@ -346,15 +328,8 @@ static void __cpuinit start_secondary(void *unused)
         * smp_call_function().
         */
        ipi_call_lock_irq();
-#ifdef CONFIG_X86_64
-       spin_lock(&vector_lock);
-
-       /* Setup the per cpu irq handling data structures */
-       __setup_vector_irq(smp_processor_id());
-       /*
-        * Allow the master to continue.
-        */
-       spin_unlock(&vector_lock);
+#ifdef CONFIG_X86_IO_APIC
+       setup_vector_irq(smp_processor_id());
 #endif
        cpu_set(smp_processor_id(), cpu_online_map);
        ipi_call_unlock_irq();
@@ -366,31 +341,8 @@ static void __cpuinit start_secondary(void *unused)
        cpu_idle();
 }
 
-#ifdef CONFIG_X86_32
-/*
- * Everything has been set up for the secondary
- * CPUs - they just need to reload everything
- * from the task structure
- * This function must not return.
- */
-void __devinit initialize_secondary(void)
-{
-       /*
-        * We don't actually need to load the full TSS,
-        * basically just the stack pointer and the ip.
-        */
-
-       asm volatile(
-               "movl %0,%%esp\n\t"
-               "jmp *%1"
-               :
-               :"m" (current->thread.sp), "m" (current->thread.ip));
-}
-#endif
-
 static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
 {
-#ifdef CONFIG_X86_32
        /*
         * Mask B, Pentium, but not Pentium MMX
         */
@@ -440,7 +392,6 @@ static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
 
 valid_k7:
        ;
-#endif
 }
 
 static void __cpuinit smp_checks(void)
@@ -555,23 +506,6 @@ cpumask_t cpu_coregroup_map(int cpu)
                return c->llc_shared_map;
 }
 
-#ifdef CONFIG_X86_32
-/*
- * We are called very early to get the low memory for the
- * SMP bootup trampoline page.
- */
-void __init smp_alloc_memory(void)
-{
-       trampoline_base = alloc_bootmem_low_pages(PAGE_SIZE);
-       /*
-        * Has to be in very low memory so we can execute
-        * real-mode AP code.
-        */
-       if (__pa(trampoline_base) >= 0x9F000)
-               BUG();
-}
-#endif
-
 static void impress_friends(void)
 {
        int cpu;
@@ -748,11 +682,7 @@ wakeup_secondary_cpu(int phys_apicid, unsigned long start_eip)
         * target processor state.
         */
        startup_ipi_hook(phys_apicid, (unsigned long) start_secondary,
-#ifdef CONFIG_X86_64
-                        (unsigned long)init_rsp);
-#else
                         (unsigned long)stack_start.sp);
-#endif
 
        /*
         * Run STARTUP IPI loop.
@@ -832,6 +762,45 @@ static void __cpuinit do_fork_idle(struct work_struct *work)
        complete(&c_idle->done);
 }
 
+#ifdef CONFIG_X86_64
+/*
+ * Allocate node local memory for the AP pda.
+ *
+ * Must be called after the _cpu_pda pointer table is initialized.
+ */
+static 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);
+               if (!after_bootmem)
+                       free_bootmem((unsigned long)oldpda, size);
+       }
+
+       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
@@ -848,28 +817,14 @@ static int __cpuinit do_boot_cpu(int apicid, int cpu)
                .done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
        };
        INIT_WORK(&c_idle.work, do_fork_idle);
-#ifdef CONFIG_X86_64
-       /* allocate memory for gdts of secondary cpus. Hotplug is considered */
-       if (!cpu_gdt_descr[cpu].address &&
-               !(cpu_gdt_descr[cpu].address = get_zeroed_page(GFP_KERNEL))) {
-               printk(KERN_ERR "Failed to allocate GDT for CPU %d\n", cpu);
-               return -1;
-       }
 
+#ifdef CONFIG_X86_64
        /* Allocate node local memory for AP pdas */
-       if (cpu_pda(cpu) == &boot_cpu_pda[cpu]) {
-               struct x8664_pda *newpda, *pda;
-               int node = cpu_to_node(cpu);
-               pda = cpu_pda(cpu);
-               newpda = kmalloc_node(sizeof(struct x8664_pda), GFP_ATOMIC,
-                                     node);
-               if (newpda) {
-                       memcpy(newpda, pda, sizeof(struct x8664_pda));
-                       cpu_pda(cpu) = newpda;
-               } else
-                       printk(KERN_ERR
-               "Could not allocate node local PDA for CPU %d on node %d\n",
-                               cpu, node);
+       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
 
@@ -905,18 +860,15 @@ do_rest:
 #ifdef CONFIG_X86_32
        per_cpu(current_task, cpu) = c_idle.idle;
        init_gdt(cpu);
-       early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
-       c_idle.idle->thread.ip = (unsigned long) start_secondary;
        /* Stack for startup_32 can be just as for start_secondary onwards */
-       stack_start.sp = (void *) c_idle.idle->thread.sp;
        irq_ctx_init(cpu);
 #else
        cpu_pda(cpu)->pcurrent = c_idle.idle;
-       init_rsp = c_idle.idle->thread.sp;
-       load_sp0(&per_cpu(init_tss, cpu), &c_idle.idle->thread);
-       initial_code = (unsigned long)start_secondary;
        clear_tsk_thread_flag(c_idle.idle, TIF_FORK);
 #endif
+       early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
+       initial_code = (unsigned long)start_secondary;
+       stack_start.sp = (void *) c_idle.idle->thread.sp;
 
        /* start_ip had better be page-aligned! */
        start_ip = setup_trampoline();
@@ -987,16 +939,14 @@ do_rest:
                                inquire_remote_apic(apicid);
                }
        }
-
-       if (boot_error) {
-               /* Try to put things back the way they were before ... */
-               unmap_cpu_to_logical_apicid(cpu);
 #ifdef CONFIG_X86_64
-               clear_node_cpumask(cpu); /* was set by numa_add_cpu */
+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_possible_map);
                cpu_clear(cpu, cpu_present_map);
                per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
        }
@@ -1088,14 +1038,12 @@ static __init void disable_smp(void)
 {
        cpu_present_map = cpumask_of_cpu(0);
        cpu_possible_map = cpumask_of_cpu(0);
-#ifdef CONFIG_X86_32
        smpboot_clear_io_apic_irqs();
-#endif
+
        if (smp_found_config)
-               phys_cpu_present_map =
-                               physid_mask_of_physid(boot_cpu_physical_apicid);
+               physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
        else
-               phys_cpu_present_map = physid_mask_of_physid(0);
+               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));
@@ -1158,12 +1106,12 @@ static int __init smp_sanity_check(unsigned max_cpus)
         * If SMP should be disabled, then really disable it!
         */
        if (!max_cpus) {
-               printk(KERN_INFO "SMP mode deactivated,"
-                                "forcing use of dummy APIC emulation.\n");
+               printk(KERN_INFO "SMP mode deactivated.\n");
                smpboot_clear_io_apic();
-#ifdef CONFIG_X86_32
+
+               localise_nmi_watchdog();
+
                connect_bsp_APIC();
-#endif
                setup_local_APIC();
                end_local_APIC_setup();
                return -1;
@@ -1191,7 +1139,6 @@ static void __init smp_cpu_index_default(void)
 void __init native_smp_prepare_cpus(unsigned int max_cpus)
 {
        preempt_disable();
-       nmi_watchdog_default();
        smp_cpu_index_default();
        current_cpu_data = boot_cpu_data;
        cpu_callin_map = cpumask_of_cpu(0);
@@ -1218,9 +1165,8 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
        }
        preempt_enable();
 
-#ifdef CONFIG_X86_32
        connect_bsp_APIC();
-#endif
+
        /*
         * Switch from PIC to APIC mode.
         */
@@ -1258,8 +1204,8 @@ void __init native_smp_prepare_boot_cpu(void)
        int me = smp_processor_id();
 #ifdef CONFIG_X86_32
        init_gdt(me);
-       switch_to_new_gdt();
 #endif
+       switch_to_new_gdt();
        /* already set me in cpu_online_map in boot_cpu_init() */
        cpu_set(me, cpu_callout_map);
        per_cpu(cpu_state, me) = CPU_ONLINE;
@@ -1279,23 +1225,6 @@ void __init native_smp_cpus_done(unsigned int max_cpus)
 
 #ifdef CONFIG_HOTPLUG_CPU
 
-#  ifdef CONFIG_X86_32
-void cpu_exit_clear(void)
-{
-       int cpu = raw_smp_processor_id();
-
-       idle_task_exit();
-
-       cpu_uninit();
-       irq_ctx_exit(cpu);
-
-       cpu_clear(cpu, cpu_callout_map);
-       cpu_clear(cpu, cpu_callin_map);
-
-       unmap_cpu_to_logical_apicid(cpu);
-}
-#  endif /* CONFIG_X86_32 */
-
 static void remove_siblinginfo(int cpu)
 {
        int sibling;
@@ -1349,12 +1278,20 @@ __init void prefill_possible_map(void)
        int i;
        int 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;
        }
+#else
+       additional_cpus = 0;
+#endif
        possible = num_processors + additional_cpus;
        if (possible > NR_CPUS)
                possible = NR_CPUS;
@@ -1364,18 +1301,18 @@ __init void prefill_possible_map(void)
 
        for (i = 0; i < possible; i++)
                cpu_set(i, cpu_possible_map);
+
+       nr_cpu_ids = possible;
 }
 
 static void __ref remove_cpu_from_maps(int cpu)
 {
        cpu_clear(cpu, cpu_online_map);
-#ifdef CONFIG_X86_64
        cpu_clear(cpu, cpu_callout_map);
        cpu_clear(cpu, cpu_callin_map);
        /* was set by cpu_init() */
        clear_bit(cpu, (unsigned long *)&cpu_initialized);
-       clear_node_cpumask(cpu);
-#endif
+       numa_remove_cpu(cpu);
 }
 
 int __cpu_disable(void)