Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / arch / ia64 / kernel / smpboot.c
diff --git a/kernel/arch/ia64/kernel/smpboot.c b/kernel/arch/ia64/kernel/smpboot.c
new file mode 100644 (file)
index 0000000..b054c5c
--- /dev/null
@@ -0,0 +1,861 @@
+/*
+ * SMP boot-related support
+ *
+ * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
+ *     David Mosberger-Tang <davidm@hpl.hp.com>
+ * Copyright (C) 2001, 2004-2005 Intel Corp
+ *     Rohit Seth <rohit.seth@intel.com>
+ *     Suresh Siddha <suresh.b.siddha@intel.com>
+ *     Gordon Jin <gordon.jin@intel.com>
+ *     Ashok Raj  <ashok.raj@intel.com>
+ *
+ * 01/05/16 Rohit Seth <rohit.seth@intel.com>  Moved SMP booting functions from smp.c to here.
+ * 01/04/27 David Mosberger <davidm@hpl.hp.com>        Added ITC synching code.
+ * 02/07/31 David Mosberger <davidm@hpl.hp.com>        Switch over to hotplug-CPU boot-sequence.
+ *                                             smp_boot_cpus()/smp_commence() is replaced by
+ *                                             smp_prepare_cpus()/__cpu_up()/smp_cpus_done().
+ * 04/06/21 Ashok Raj          <ashok.raj@intel.com> Added CPU Hotplug Support
+ * 04/12/26 Jin Gordon <gordon.jin@intel.com>
+ * 04/12/26 Rohit Seth <rohit.seth@intel.com>
+ *                                             Add multi-threading and multi-core detection
+ * 05/01/30 Suresh Siddha <suresh.b.siddha@intel.com>
+ *                                             Setup cpu_sibling_map and cpu_core_map
+ */
+
+#include <linux/module.h>
+#include <linux/acpi.h>
+#include <linux/bootmem.h>
+#include <linux/cpu.h>
+#include <linux/delay.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/kernel.h>
+#include <linux/kernel_stat.h>
+#include <linux/mm.h>
+#include <linux/notifier.h>
+#include <linux/smp.h>
+#include <linux/spinlock.h>
+#include <linux/efi.h>
+#include <linux/percpu.h>
+#include <linux/bitops.h>
+
+#include <linux/atomic.h>
+#include <asm/cache.h>
+#include <asm/current.h>
+#include <asm/delay.h>
+#include <asm/io.h>
+#include <asm/irq.h>
+#include <asm/machvec.h>
+#include <asm/mca.h>
+#include <asm/page.h>
+#include <asm/paravirt.h>
+#include <asm/pgalloc.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+#include <asm/ptrace.h>
+#include <asm/sal.h>
+#include <asm/tlbflush.h>
+#include <asm/unistd.h>
+#include <asm/sn/arch.h>
+
+#define SMP_DEBUG 0
+
+#if SMP_DEBUG
+#define Dprintk(x...)  printk(x)
+#else
+#define Dprintk(x...)
+#endif
+
+#ifdef CONFIG_HOTPLUG_CPU
+#ifdef CONFIG_PERMIT_BSP_REMOVE
+#define bsp_remove_ok  1
+#else
+#define bsp_remove_ok  0
+#endif
+
+/*
+ * Global array allocated for NR_CPUS at boot time
+ */
+struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];
+
+/*
+ * start_ap in head.S uses this to store current booting cpu
+ * info.
+ */
+struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];
+
+#define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);
+
+#else
+#define set_brendez_area(x)
+#endif
+
+
+/*
+ * ITC synchronization related stuff:
+ */
+#define MASTER (0)
+#define SLAVE  (SMP_CACHE_BYTES/8)
+
+#define NUM_ROUNDS     64      /* magic value */
+#define NUM_ITERS      5       /* likewise */
+
+static DEFINE_SPINLOCK(itc_sync_lock);
+static volatile unsigned long go[SLAVE + 1];
+
+#define DEBUG_ITC_SYNC 0
+
+extern void start_ap (void);
+extern unsigned long ia64_iobase;
+
+struct task_struct *task_for_booting_cpu;
+
+/*
+ * State for each CPU
+ */
+DEFINE_PER_CPU(int, cpu_state);
+
+cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
+EXPORT_SYMBOL(cpu_core_map);
+DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map);
+EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
+
+int smp_num_siblings = 1;
+
+/* which logical CPU number maps to which CPU (physical APIC ID) */
+volatile int ia64_cpu_to_sapicid[NR_CPUS];
+EXPORT_SYMBOL(ia64_cpu_to_sapicid);
+
+static cpumask_t cpu_callin_map;
+
+struct smp_boot_data smp_boot_data __initdata;
+
+unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */
+
+char __initdata no_int_routing;
+
+unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */
+
+#ifdef CONFIG_FORCE_CPEI_RETARGET
+#define CPEI_OVERRIDE_DEFAULT  (1)
+#else
+#define CPEI_OVERRIDE_DEFAULT  (0)
+#endif
+
+unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT;
+
+static int __init
+cmdl_force_cpei(char *str)
+{
+       int value=0;
+
+       get_option (&str, &value);
+       force_cpei_retarget = value;
+
+       return 1;
+}
+
+__setup("force_cpei=", cmdl_force_cpei);
+
+static int __init
+nointroute (char *str)
+{
+       no_int_routing = 1;
+       printk ("no_int_routing on\n");
+       return 1;
+}
+
+__setup("nointroute", nointroute);
+
+static void fix_b0_for_bsp(void)
+{
+#ifdef CONFIG_HOTPLUG_CPU
+       int cpuid;
+       static int fix_bsp_b0 = 1;
+
+       cpuid = smp_processor_id();
+
+       /*
+        * Cache the b0 value on the first AP that comes up
+        */
+       if (!(fix_bsp_b0 && cpuid))
+               return;
+
+       sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
+       printk ("Fixed BSP b0 value from CPU %d\n", cpuid);
+
+       fix_bsp_b0 = 0;
+#endif
+}
+
+void
+sync_master (void *arg)
+{
+       unsigned long flags, i;
+
+       go[MASTER] = 0;
+
+       local_irq_save(flags);
+       {
+               for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
+                       while (!go[MASTER])
+                               cpu_relax();
+                       go[MASTER] = 0;
+                       go[SLAVE] = ia64_get_itc();
+               }
+       }
+       local_irq_restore(flags);
+}
+
+/*
+ * Return the number of cycles by which our itc differs from the itc on the master
+ * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
+ * negative that it is behind.
+ */
+static inline long
+get_delta (long *rt, long *master)
+{
+       unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
+       unsigned long tcenter, t0, t1, tm;
+       long i;
+
+       for (i = 0; i < NUM_ITERS; ++i) {
+               t0 = ia64_get_itc();
+               go[MASTER] = 1;
+               while (!(tm = go[SLAVE]))
+                       cpu_relax();
+               go[SLAVE] = 0;
+               t1 = ia64_get_itc();
+
+               if (t1 - t0 < best_t1 - best_t0)
+                       best_t0 = t0, best_t1 = t1, best_tm = tm;
+       }
+
+       *rt = best_t1 - best_t0;
+       *master = best_tm - best_t0;
+
+       /* average best_t0 and best_t1 without overflow: */
+       tcenter = (best_t0/2 + best_t1/2);
+       if (best_t0 % 2 + best_t1 % 2 == 2)
+               ++tcenter;
+       return tcenter - best_tm;
+}
+
+/*
+ * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
+ * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
+ * unaccounted-for errors (such as getting a machine check in the middle of a calibration
+ * step).  The basic idea is for the slave to ask the master what itc value it has and to
+ * read its own itc before and after the master responds.  Each iteration gives us three
+ * timestamps:
+ *
+ *     slave           master
+ *
+ *     t0 ---\
+ *             ---\
+ *                --->
+ *                     tm
+ *                /---
+ *            /---
+ *     t1 <---
+ *
+ *
+ * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
+ * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
+ * between the slave and the master is symmetric.  Even if the interconnect were
+ * asymmetric, we would still know that the synchronization error is smaller than the
+ * roundtrip latency (t0 - t1).
+ *
+ * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
+ * within one or two cycles.  However, we can only *guarantee* that the synchronization is
+ * accurate to within a round-trip time, which is typically in the range of several
+ * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
+ * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
+ * than half a micro second or so.
+ */
+void
+ia64_sync_itc (unsigned int master)
+{
+       long i, delta, adj, adjust_latency = 0, done = 0;
+       unsigned long flags, rt, master_time_stamp, bound;
+#if DEBUG_ITC_SYNC
+       struct {
+               long rt;        /* roundtrip time */
+               long master;    /* master's timestamp */
+               long diff;      /* difference between midpoint and master's timestamp */
+               long lat;       /* estimate of itc adjustment latency */
+       } t[NUM_ROUNDS];
+#endif
+
+       /*
+        * Make sure local timer ticks are disabled while we sync.  If
+        * they were enabled, we'd have to worry about nasty issues
+        * like setting the ITC ahead of (or a long time before) the
+        * next scheduled tick.
+        */
+       BUG_ON((ia64_get_itv() & (1 << 16)) == 0);
+
+       go[MASTER] = 1;
+
+       if (smp_call_function_single(master, sync_master, NULL, 0) < 0) {
+               printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
+               return;
+       }
+
+       while (go[MASTER])
+               cpu_relax();    /* wait for master to be ready */
+
+       spin_lock_irqsave(&itc_sync_lock, flags);
+       {
+               for (i = 0; i < NUM_ROUNDS; ++i) {
+                       delta = get_delta(&rt, &master_time_stamp);
+                       if (delta == 0) {
+                               done = 1;       /* let's lock on to this... */
+                               bound = rt;
+                       }
+
+                       if (!done) {
+                               if (i > 0) {
+                                       adjust_latency += -delta;
+                                       adj = -delta + adjust_latency/4;
+                               } else
+                                       adj = -delta;
+
+                               ia64_set_itc(ia64_get_itc() + adj);
+                       }
+#if DEBUG_ITC_SYNC
+                       t[i].rt = rt;
+                       t[i].master = master_time_stamp;
+                       t[i].diff = delta;
+                       t[i].lat = adjust_latency/4;
+#endif
+               }
+       }
+       spin_unlock_irqrestore(&itc_sync_lock, flags);
+
+#if DEBUG_ITC_SYNC
+       for (i = 0; i < NUM_ROUNDS; ++i)
+               printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
+                      t[i].rt, t[i].master, t[i].diff, t[i].lat);
+#endif
+
+       printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
+              "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
+}
+
+/*
+ * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
+ */
+static inline void smp_setup_percpu_timer(void)
+{
+}
+
+static void
+smp_callin (void)
+{
+       int cpuid, phys_id, itc_master;
+       struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo;
+       extern void ia64_init_itm(void);
+       extern volatile int time_keeper_id;
+
+#ifdef CONFIG_PERFMON
+       extern void pfm_init_percpu(void);
+#endif
+
+       cpuid = smp_processor_id();
+       phys_id = hard_smp_processor_id();
+       itc_master = time_keeper_id;
+
+       if (cpu_online(cpuid)) {
+               printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
+                      phys_id, cpuid);
+               BUG();
+       }
+
+       fix_b0_for_bsp();
+
+       /*
+        * numa_node_id() works after this.
+        */
+       set_numa_node(cpu_to_node_map[cpuid]);
+       set_numa_mem(local_memory_node(cpu_to_node_map[cpuid]));
+
+       spin_lock(&vector_lock);
+       /* Setup the per cpu irq handling data structures */
+       __setup_vector_irq(cpuid);
+       notify_cpu_starting(cpuid);
+       set_cpu_online(cpuid, true);
+       per_cpu(cpu_state, cpuid) = CPU_ONLINE;
+       spin_unlock(&vector_lock);
+
+       smp_setup_percpu_timer();
+
+       ia64_mca_cmc_vector_setup();    /* Setup vector on AP */
+
+#ifdef CONFIG_PERFMON
+       pfm_init_percpu();
+#endif
+
+       local_irq_enable();
+
+       if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
+               /*
+                * Synchronize the ITC with the BP.  Need to do this after irqs are
+                * enabled because ia64_sync_itc() calls smp_call_function_single(), which
+                * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
+                * local_bh_enable(), which bugs out if irqs are not enabled...
+                */
+               Dprintk("Going to syncup ITC with ITC Master.\n");
+               ia64_sync_itc(itc_master);
+       }
+
+       /*
+        * Get our bogomips.
+        */
+       ia64_init_itm();
+
+       /*
+        * Delay calibration can be skipped if new processor is identical to the
+        * previous processor.
+        */
+       last_cpuinfo = cpu_data(cpuid - 1);
+       this_cpuinfo = local_cpu_data;
+       if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq ||
+           last_cpuinfo->proc_freq != this_cpuinfo->proc_freq ||
+           last_cpuinfo->features != this_cpuinfo->features ||
+           last_cpuinfo->revision != this_cpuinfo->revision ||
+           last_cpuinfo->family != this_cpuinfo->family ||
+           last_cpuinfo->archrev != this_cpuinfo->archrev ||
+           last_cpuinfo->model != this_cpuinfo->model)
+               calibrate_delay();
+       local_cpu_data->loops_per_jiffy = loops_per_jiffy;
+
+       /*
+        * Allow the master to continue.
+        */
+       cpumask_set_cpu(cpuid, &cpu_callin_map);
+       Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
+}
+
+
+/*
+ * Activate a secondary processor.  head.S calls this.
+ */
+int
+start_secondary (void *unused)
+{
+       /* Early console may use I/O ports */
+       ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
+#ifndef CONFIG_PRINTK_TIME
+       Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
+#endif
+       efi_map_pal_code();
+       cpu_init();
+       preempt_disable();
+       smp_callin();
+
+       cpu_startup_entry(CPUHP_ONLINE);
+       return 0;
+}
+
+static int
+do_boot_cpu (int sapicid, int cpu, struct task_struct *idle)
+{
+       int timeout;
+
+       task_for_booting_cpu = idle;
+       Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid);
+
+       set_brendez_area(cpu);
+       platform_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0);
+
+       /*
+        * Wait 10s total for the AP to start
+        */
+       Dprintk("Waiting on callin_map ...");
+       for (timeout = 0; timeout < 100000; timeout++) {
+               if (cpumask_test_cpu(cpu, &cpu_callin_map))
+                       break;  /* It has booted */
+               barrier(); /* Make sure we re-read cpu_callin_map */
+               udelay(100);
+       }
+       Dprintk("\n");
+
+       if (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
+               printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid);
+               ia64_cpu_to_sapicid[cpu] = -1;
+               set_cpu_online(cpu, false);  /* was set in smp_callin() */
+               return -EINVAL;
+       }
+       return 0;
+}
+
+static int __init
+decay (char *str)
+{
+       int ticks;
+       get_option (&str, &ticks);
+       return 1;
+}
+
+__setup("decay=", decay);
+
+/*
+ * Initialize the logical CPU number to SAPICID mapping
+ */
+void __init
+smp_build_cpu_map (void)
+{
+       int sapicid, cpu, i;
+       int boot_cpu_id = hard_smp_processor_id();
+
+       for (cpu = 0; cpu < NR_CPUS; cpu++) {
+               ia64_cpu_to_sapicid[cpu] = -1;
+       }
+
+       ia64_cpu_to_sapicid[0] = boot_cpu_id;
+       init_cpu_present(cpumask_of(0));
+       set_cpu_possible(0, true);
+       for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) {
+               sapicid = smp_boot_data.cpu_phys_id[i];
+               if (sapicid == boot_cpu_id)
+                       continue;
+               set_cpu_present(cpu, true);
+               set_cpu_possible(cpu, true);
+               ia64_cpu_to_sapicid[cpu] = sapicid;
+               cpu++;
+       }
+}
+
+/*
+ * Cycle through the APs sending Wakeup IPIs to boot each.
+ */
+void __init
+smp_prepare_cpus (unsigned int max_cpus)
+{
+       int boot_cpu_id = hard_smp_processor_id();
+
+       /*
+        * Initialize the per-CPU profiling counter/multiplier
+        */
+
+       smp_setup_percpu_timer();
+
+       cpumask_set_cpu(0, &cpu_callin_map);
+
+       local_cpu_data->loops_per_jiffy = loops_per_jiffy;
+       ia64_cpu_to_sapicid[0] = boot_cpu_id;
+
+       printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id);
+
+       current_thread_info()->cpu = 0;
+
+       /*
+        * If SMP should be disabled, then really disable it!
+        */
+       if (!max_cpus) {
+               printk(KERN_INFO "SMP mode deactivated.\n");
+               init_cpu_online(cpumask_of(0));
+               init_cpu_present(cpumask_of(0));
+               init_cpu_possible(cpumask_of(0));
+               return;
+       }
+}
+
+void smp_prepare_boot_cpu(void)
+{
+       set_cpu_online(smp_processor_id(), true);
+       cpumask_set_cpu(smp_processor_id(), &cpu_callin_map);
+       set_numa_node(cpu_to_node_map[smp_processor_id()]);
+       per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
+       paravirt_post_smp_prepare_boot_cpu();
+}
+
+#ifdef CONFIG_HOTPLUG_CPU
+static inline void
+clear_cpu_sibling_map(int cpu)
+{
+       int i;
+
+       for_each_cpu(i, &per_cpu(cpu_sibling_map, cpu))
+               cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, i));
+       for_each_cpu(i, &cpu_core_map[cpu])
+               cpumask_clear_cpu(cpu, &cpu_core_map[i]);
+
+       per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE;
+}
+
+static void
+remove_siblinginfo(int cpu)
+{
+       int last = 0;
+
+       if (cpu_data(cpu)->threads_per_core == 1 &&
+           cpu_data(cpu)->cores_per_socket == 1) {
+               cpumask_clear_cpu(cpu, &cpu_core_map[cpu]);
+               cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
+               return;
+       }
+
+       last = (cpumask_weight(&cpu_core_map[cpu]) == 1 ? 1 : 0);
+
+       /* remove it from all sibling map's */
+       clear_cpu_sibling_map(cpu);
+}
+
+extern void fixup_irqs(void);
+
+int migrate_platform_irqs(unsigned int cpu)
+{
+       int new_cpei_cpu;
+       struct irq_data *data = NULL;
+       const struct cpumask *mask;
+       int             retval = 0;
+
+       /*
+        * dont permit CPEI target to removed.
+        */
+       if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) {
+               printk ("CPU (%d) is CPEI Target\n", cpu);
+               if (can_cpei_retarget()) {
+                       /*
+                        * Now re-target the CPEI to a different processor
+                        */
+                       new_cpei_cpu = cpumask_any(cpu_online_mask);
+                       mask = cpumask_of(new_cpei_cpu);
+                       set_cpei_target_cpu(new_cpei_cpu);
+                       data = irq_get_irq_data(ia64_cpe_irq);
+                       /*
+                        * Switch for now, immediately, we need to do fake intr
+                        * as other interrupts, but need to study CPEI behaviour with
+                        * polling before making changes.
+                        */
+                       if (data && data->chip) {
+                               data->chip->irq_disable(data);
+                               data->chip->irq_set_affinity(data, mask, false);
+                               data->chip->irq_enable(data);
+                               printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu);
+                       }
+               }
+               if (!data) {
+                       printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu);
+                       retval = -EBUSY;
+               }
+       }
+       return retval;
+}
+
+/* must be called with cpucontrol mutex held */
+int __cpu_disable(void)
+{
+       int cpu = smp_processor_id();
+
+       /*
+        * dont permit boot processor for now
+        */
+       if (cpu == 0 && !bsp_remove_ok) {
+               printk ("Your platform does not support removal of BSP\n");
+               return (-EBUSY);
+       }
+
+       if (ia64_platform_is("sn2")) {
+               if (!sn_cpu_disable_allowed(cpu))
+                       return -EBUSY;
+       }
+
+       set_cpu_online(cpu, false);
+
+       if (migrate_platform_irqs(cpu)) {
+               set_cpu_online(cpu, true);
+               return -EBUSY;
+       }
+
+       remove_siblinginfo(cpu);
+       fixup_irqs();
+       local_flush_tlb_all();
+       cpumask_clear_cpu(cpu, &cpu_callin_map);
+       return 0;
+}
+
+void __cpu_die(unsigned int cpu)
+{
+       unsigned int i;
+
+       for (i = 0; i < 100; i++) {
+               /* They ack this in play_dead by setting CPU_DEAD */
+               if (per_cpu(cpu_state, cpu) == CPU_DEAD)
+               {
+                       printk ("CPU %d is now offline\n", cpu);
+                       return;
+               }
+               msleep(100);
+       }
+       printk(KERN_ERR "CPU %u didn't die...\n", cpu);
+}
+#endif /* CONFIG_HOTPLUG_CPU */
+
+void
+smp_cpus_done (unsigned int dummy)
+{
+       int cpu;
+       unsigned long bogosum = 0;
+
+       /*
+        * Allow the user to impress friends.
+        */
+
+       for_each_online_cpu(cpu) {
+               bogosum += cpu_data(cpu)->loops_per_jiffy;
+       }
+
+       printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
+              (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100);
+}
+
+static inline void set_cpu_sibling_map(int cpu)
+{
+       int i;
+
+       for_each_online_cpu(i) {
+               if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) {
+                       cpumask_set_cpu(i, &cpu_core_map[cpu]);
+                       cpumask_set_cpu(cpu, &cpu_core_map[i]);
+                       if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) {
+                               cpumask_set_cpu(i,
+                                               &per_cpu(cpu_sibling_map, cpu));
+                               cpumask_set_cpu(cpu,
+                                               &per_cpu(cpu_sibling_map, i));
+                       }
+               }
+       }
+}
+
+int
+__cpu_up(unsigned int cpu, struct task_struct *tidle)
+{
+       int ret;
+       int sapicid;
+
+       sapicid = ia64_cpu_to_sapicid[cpu];
+       if (sapicid == -1)
+               return -EINVAL;
+
+       /*
+        * Already booted cpu? not valid anymore since we dont
+        * do idle loop tightspin anymore.
+        */
+       if (cpumask_test_cpu(cpu, &cpu_callin_map))
+               return -EINVAL;
+
+       per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
+       /* Processor goes to start_secondary(), sets online flag */
+       ret = do_boot_cpu(sapicid, cpu, tidle);
+       if (ret < 0)
+               return ret;
+
+       if (cpu_data(cpu)->threads_per_core == 1 &&
+           cpu_data(cpu)->cores_per_socket == 1) {
+               cpumask_set_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
+               cpumask_set_cpu(cpu, &cpu_core_map[cpu]);
+               return 0;
+       }
+
+       set_cpu_sibling_map(cpu);
+
+       return 0;
+}
+
+/*
+ * Assume that CPUs have been discovered by some platform-dependent interface.  For
+ * SoftSDV/Lion, that would be ACPI.
+ *
+ * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP().
+ */
+void __init
+init_smp_config(void)
+{
+       struct fptr {
+               unsigned long fp;
+               unsigned long gp;
+       } *ap_startup;
+       long sal_ret;
+
+       /* Tell SAL where to drop the APs.  */
+       ap_startup = (struct fptr *) start_ap;
+       sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ,
+                                      ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0);
+       if (sal_ret < 0)
+               printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n",
+                      ia64_sal_strerror(sal_ret));
+}
+
+/*
+ * identify_siblings(cpu) gets called from identify_cpu. This populates the 
+ * information related to logical execution units in per_cpu_data structure.
+ */
+void identify_siblings(struct cpuinfo_ia64 *c)
+{
+       long status;
+       u16 pltid;
+       pal_logical_to_physical_t info;
+
+       status = ia64_pal_logical_to_phys(-1, &info);
+       if (status != PAL_STATUS_SUCCESS) {
+               if (status != PAL_STATUS_UNIMPLEMENTED) {
+                       printk(KERN_ERR
+                               "ia64_pal_logical_to_phys failed with %ld\n",
+                               status);
+                       return;
+               }
+
+               info.overview_ppid = 0;
+               info.overview_cpp  = 1;
+               info.overview_tpc  = 1;
+       }
+
+       status = ia64_sal_physical_id_info(&pltid);
+       if (status != PAL_STATUS_SUCCESS) {
+               if (status != PAL_STATUS_UNIMPLEMENTED)
+                       printk(KERN_ERR
+                               "ia64_sal_pltid failed with %ld\n",
+                               status);
+               return;
+       }
+
+       c->socket_id =  (pltid << 8) | info.overview_ppid;
+
+       if (info.overview_cpp == 1 && info.overview_tpc == 1)
+               return;
+
+       c->cores_per_socket = info.overview_cpp;
+       c->threads_per_core = info.overview_tpc;
+       c->num_log = info.overview_num_log;
+
+       c->core_id = info.log1_cid;
+       c->thread_id = info.log1_tid;
+}
+
+/*
+ * returns non zero, if multi-threading is enabled
+ * on at least one physical package. Due to hotplug cpu
+ * and (maxcpus=), all threads may not necessarily be enabled
+ * even though the processor supports multi-threading.
+ */
+int is_multithreading_enabled(void)
+{
+       int i, j;
+
+       for_each_present_cpu(i) {
+               for_each_present_cpu(j) {
+                       if (j == i)
+                               continue;
+                       if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) {
+                               if (cpu_data(j)->core_id == cpu_data(i)->core_id)
+                                       return 1;
+                       }
+               }
+       }
+       return 0;
+}
+EXPORT_SYMBOL_GPL(is_multithreading_enabled);