Use macros for .bss.page_aligned section.
[safe/jmp/linux-2.6] / arch / x86 / kernel / smpboot.c
index af57f88..09c5e07 100644 (file)
@@ -47,6 +47,7 @@
 #include <linux/bootmem.h>
 #include <linux/err.h>
 #include <linux/nmi.h>
+#include <linux/tboot.h>
 
 #include <asm/acpi.h>
 #include <asm/desc.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 <asm/genapic.h>
 #include <asm/smpboot_hooks.h>
 
 #ifdef CONFIG_X86_32
@@ -102,29 +102,20 @@ EXPORT_SYMBOL(smp_num_siblings);
 DEFINE_PER_CPU(u16, cpu_llc_id) = BAD_APICID;
 
 /* representing HT siblings of each logical CPU */
-DEFINE_PER_CPU(cpumask_t, cpu_sibling_map);
+DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
 
 /* representing HT and core siblings of each logical CPU */
-DEFINE_PER_CPU(cpumask_t, cpu_core_map);
+DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
 EXPORT_PER_CPU_SYMBOL(cpu_core_map);
 
 /* Per CPU bogomips and other parameters */
 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
 EXPORT_PER_CPU_SYMBOL(cpu_info);
 
-static atomic_t init_deasserted;
-
-
-/* Set if we find a B stepping CPU */
-static int __cpuinitdata smp_b_stepping;
+atomic_t init_deasserted;
 
 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_32)
-
-/* which logical CPUs are on which nodes */
-cpumask_t node_to_cpumask_map[MAX_NUMNODES] __read_mostly =
-                               { [0 ... MAX_NUMNODES-1] = CPU_MASK_NONE };
-EXPORT_SYMBOL(node_to_cpumask_map);
 /* which node each logical CPU is on */
 int cpu_to_node_map[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = 0 };
 EXPORT_SYMBOL(cpu_to_node_map);
@@ -133,7 +124,7 @@ 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);
-       cpumask_set_cpu(cpu, &node_to_cpumask_map[node]);
+       cpumask_set_cpu(cpu, node_to_cpumask_map[node]);
        cpu_to_node_map[cpu] = node;
 }
 
@@ -144,7 +135,7 @@ 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++)
-               cpumask_clear_cpu(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) */
@@ -272,8 +263,6 @@ static void __cpuinit smp_callin(void)
        cpumask_set_cpu(cpuid, cpu_callin_mask);
 }
 
-static int __cpuinitdata unsafe_smp;
-
 /*
  * Activate a secondary processor.
  */
@@ -308,7 +297,7 @@ notrace static void __cpuinit start_secondary(void *unused)
        __flush_tlb_all();
 #endif
 
-       /* This must be done before setting cpu_online_map */
+       /* This must be done before setting cpu_online_mask */
        set_cpu_sibling_map(raw_smp_processor_id());
        wmb();
 
@@ -335,81 +324,28 @@ notrace static void __cpuinit start_secondary(void *unused)
        /* enable local interrupts */
        local_irq_enable();
 
-       setup_secondary_clock();
+       x86_cpuinit.setup_percpu_clockev();
 
        wmb();
        cpu_idle();
 }
 
-static void __cpuinit smp_apply_quirks(struct cpuinfo_x86 *c)
+#ifdef CONFIG_CPUMASK_OFFSTACK
+/* In this case, llc_shared_map is a pointer to a cpumask. */
+static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
+                                   const struct cpuinfo_x86 *src)
 {
-       /*
-        * 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. */
-               unsafe_smp = 1;
-       }
-
-valid_k7:
-       ;
+       struct cpumask *llc = dst->llc_shared_map;
+       *dst = *src;
+       dst->llc_shared_map = llc;
 }
-
-static void __cpuinit smp_checks(void)
+#else
+static inline void copy_cpuinfo_x86(struct cpuinfo_x86 *dst,
+                                   const struct cpuinfo_x86 *src)
 {
-       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 (unsafe_smp && num_online_cpus() > 1) {
-               printk(KERN_INFO "WARNING: This combination of AMD"
-                       "processors is not suitable for SMP.\n");
-               add_taint(TAINT_UNSAFE_SMP);
-       }
+       *dst = *src;
 }
+#endif /* CONFIG_CPUMASK_OFFSTACK */
 
 /*
  * The bootstrap kernel entry code has set these up. Save them for
@@ -420,11 +356,10 @@ void __cpuinit smp_store_cpu_info(int id)
 {
        struct cpuinfo_x86 *c = &cpu_data(id);
 
-       *c = boot_cpu_data;
+       copy_cpuinfo_x86(c, &boot_cpu_data);
        c->cpu_index = id;
        if (id != 0)
                identify_secondary_cpu(c);
-       smp_apply_quirks(c);
 }
 
 
@@ -445,15 +380,15 @@ void __cpuinit set_cpu_sibling_map(int 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);
+                               cpumask_set_cpu(i, c->llc_shared_map);
+                               cpumask_set_cpu(cpu, o->llc_shared_map);
                        }
                }
        } else {
                cpumask_set_cpu(cpu, cpu_sibling_mask(cpu));
        }
 
-       cpumask_set_cpu(cpu, &c->llc_shared_map);
+       cpumask_set_cpu(cpu, c->llc_shared_map);
 
        if (current_cpu_data.x86_max_cores == 1) {
                cpumask_copy(cpu_core_mask(cpu), cpu_sibling_mask(cpu));
@@ -464,8 +399,8 @@ void __cpuinit set_cpu_sibling_map(int cpu)
        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)) {
-                       cpumask_set_cpu(i, &c->llc_shared_map);
-                       cpumask_set_cpu(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) {
                        cpumask_set_cpu(i, cpu_core_mask(cpu));
@@ -500,15 +435,11 @@ const struct cpumask *cpu_coregroup_mask(int cpu)
         * For perf, we return last level cache shared map.
         * And for power savings, we return cpu_core_map
         */
-       if (sched_mc_power_savings || sched_smt_power_savings)
+       if ((sched_mc_power_savings || sched_smt_power_savings) &&
+           !(cpu_has(c, X86_FEATURE_AMD_DCM)))
                return cpu_core_mask(cpu);
        else
-               return &c->llc_shared_map;
-}
-
-cpumask_t cpu_coregroup_map(int cpu)
-{
-       return *cpu_coregroup_mask(cpu);
+               return c->llc_shared_map;
 }
 
 static void impress_friends(void)
@@ -575,7 +506,7 @@ void __inquire_remote_apic(int apicid)
  * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
  * won't ... remember to clear down the APIC, etc later.
  */
-int __devinit
+int __cpuinit
 wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
 {
        unsigned long send_status, accept_status = 0;
@@ -609,18 +540,12 @@ wakeup_secondary_cpu_via_nmi(int logical_apicid, unsigned long start_eip)
        return (send_status | accept_status);
 }
 
-int __devinit
+static int __cpuinit
 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();
 
        /*
@@ -746,21 +671,22 @@ static void __cpuinit do_fork_idle(struct work_struct *work)
        complete(&c_idle->done);
 }
 
-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_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);
 
        alternatives_smp_switch(1);
@@ -825,9 +751,6 @@ do_rest:
 
                pr_debug("Setting warm reset code and vector.\n");
 
-               if (apic->store_NMI_vector)
-                       apic->store_NMI_vector(&nmi_high, &nmi_low);
-
                smpboot_setup_warm_reset_vector(start_ip);
                /*
                 * Be paranoid about clearing APIC errors.
@@ -839,9 +762,13 @@ do_rest:
        }
 
        /*
-        * 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 = apic->wakeup_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) {
                /*
@@ -897,10 +824,12 @@ do_rest:
        /* mark "stuck" area as not stuck */
        *((volatile unsigned long *)trampoline_base) = 0;
 
-       /*
-        * Cleanup possible dangling ends...
-        */
-       smpboot_restore_warm_reset_vector();
+       if (get_uv_system_type() != UV_NON_UNIQUE_APIC) {
+               /*
+                * Cleanup possible dangling ends...
+                */
+               smpboot_restore_warm_reset_vector();
+       }
 
        return boot_error;
 }
@@ -946,7 +875,7 @@ int __cpuinit native_cpu_up(unsigned int cpu)
 
        err = do_boot_cpu(apicid, cpu);
 
-       zap_low_mappings();
+       zap_low_mappings(false);
        low_mappings = 0;
 #else
        err = do_boot_cpu(apicid, cpu);
@@ -979,9 +908,8 @@ int __cpuinit native_cpu_up(unsigned int cpu)
  */
 static __init void disable_smp(void)
 {
-       /* 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));
+       init_cpu_present(cpumask_of(0));
+       init_cpu_possible(cpumask_of(0));
        smpboot_clear_io_apic_irqs();
 
        if (smp_found_config)
@@ -1066,10 +994,12 @@ static int __init smp_sanity_check(unsigned max_cpus)
         */
        if (APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid]) &&
            !cpu_has_apic) {
-               printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
-                       boot_cpu_physical_apicid);
-               printk(KERN_ERR "... forcing use of dummy APIC emulation."
+               if (!disable_apic) {
+                       pr_err("BIOS bug, local APIC #%d not detected!...\n",
+                               boot_cpu_physical_apicid);
+                       pr_err("... forcing use of dummy APIC emulation."
                                "(tell your hw vendor)\n");
+               }
                smpboot_clear_io_apic();
                arch_disable_smp_support();
                return -1;
@@ -1113,6 +1043,8 @@ static void __init smp_cpu_index_default(void)
  */
 void __init native_smp_prepare_cpus(unsigned int max_cpus)
 {
+       unsigned int i;
+
        preempt_disable();
        smp_cpu_index_default();
        current_cpu_data = boot_cpu_data;
@@ -1126,10 +1058,18 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
        boot_cpu_logical_apicid = logical_smp_processor_id();
 #endif
        current_thread_info()->cpu = 0;  /* needed? */
+       for_each_possible_cpu(i) {
+               alloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
+               alloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
+               alloc_cpumask_var(&cpu_data(i).llc_shared_map, GFP_KERNEL);
+               cpumask_clear(per_cpu(cpu_core_map, i));
+               cpumask_clear(per_cpu(cpu_sibling_map, i));
+               cpumask_clear(cpu_data(i).llc_shared_map);
+       }
        set_cpu_sibling_map(0);
 
-#ifdef CONFIG_X86_64
        enable_IR_x2apic();
+#ifdef CONFIG_X86_64
        default_setup_apic_routing();
 #endif
 
@@ -1154,13 +1094,12 @@ 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();
@@ -1175,13 +1114,26 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
 
        printk(KERN_INFO "CPU%d: ", 0);
        print_cpu_info(&cpu_data(0));
-       setup_boot_clock();
+       x86_init.timers.setup_percpu_clockev();
 
        if (is_uv_system())
                uv_system_init();
+
+       set_mtrr_aps_delayed_init();
 out:
        preempt_enable();
 }
+
+void arch_enable_nonboot_cpus_begin(void)
+{
+       set_mtrr_aps_delayed_init();
+}
+
+void arch_enable_nonboot_cpus_end(void)
+{
+       mtrr_aps_init();
+}
+
 /*
  * Early setup to make printk work.
  */
@@ -1199,11 +1151,11 @@ 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();
+       mtrr_aps_init();
 }
 
 static int __initdata setup_possible_cpus = -1;
@@ -1216,11 +1168,11 @@ early_param("possible_cpus", _setup_possible_cpus);
 
 
 /*
- * cpu_possible_map should be static, it cannot change as cpu's
+ * cpu_possible_mask should be static, it cannot change as cpu's
  * are onlined, or offlined. The reason is per-cpu data-structures
  * are allocated by some modules at init time, and dont expect to
  * do this dynamically on cpu arrival/departure.
- * cpu_present_map on the other hand can change dynamically.
+ * cpu_present_mask on the other hand can change dynamically.
  * In case when cpu_hotplug is not compiled, then we resort to current
  * behaviour, which is cpu_possible == cpu_present.
  * - Ashok Raj
@@ -1381,6 +1333,7 @@ void play_dead_common(void)
 void native_play_dead(void)
 {
        play_dead_common();
+       tboot_shutdown(TB_SHUTDOWN_WFS);
        wbinvd_halt();
 }