linux/arch/ia64/kernel/smpboot.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * SMP boot-related support
   4 *
   5 * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
   6 *      David Mosberger-Tang <davidm@hpl.hp.com>
   7 * Copyright (C) 2001, 2004-2005 Intel Corp
   8 *      Rohit Seth <rohit.seth@intel.com>
   9 *      Suresh Siddha <suresh.b.siddha@intel.com>
  10 *      Gordon Jin <gordon.jin@intel.com>
  11 *      Ashok Raj  <ashok.raj@intel.com>
  12 *
  13 * 01/05/16 Rohit Seth <rohit.seth@intel.com>   Moved SMP booting functions from smp.c to here.
  14 * 01/04/27 David Mosberger <davidm@hpl.hp.com> Added ITC synching code.
  15 * 02/07/31 David Mosberger <davidm@hpl.hp.com> Switch over to hotplug-CPU boot-sequence.
  16 *                                              smp_boot_cpus()/smp_commence() is replaced by
  17 *                                              smp_prepare_cpus()/__cpu_up()/smp_cpus_done().
  18 * 04/06/21 Ashok Raj           <ashok.raj@intel.com> Added CPU Hotplug Support
  19 * 04/12/26 Jin Gordon <gordon.jin@intel.com>
  20 * 04/12/26 Rohit Seth <rohit.seth@intel.com>
  21 *                                              Add multi-threading and multi-core detection
  22 * 05/01/30 Suresh Siddha <suresh.b.siddha@intel.com>
  23 *                                              Setup cpu_sibling_map and cpu_core_map
  24 */
  25
  26#include <linux/module.h>
  27#include <linux/acpi.h>
  28#include <linux/memblock.h>
  29#include <linux/cpu.h>
  30#include <linux/delay.h>
  31#include <linux/init.h>
  32#include <linux/interrupt.h>
  33#include <linux/irq.h>
  34#include <linux/kernel.h>
  35#include <linux/kernel_stat.h>
  36#include <linux/mm.h>
  37#include <linux/notifier.h>
  38#include <linux/smp.h>
  39#include <linux/spinlock.h>
  40#include <linux/efi.h>
  41#include <linux/percpu.h>
  42#include <linux/bitops.h>
  43
  44#include <linux/atomic.h>
  45#include <asm/cache.h>
  46#include <asm/current.h>
  47#include <asm/delay.h>
  48#include <asm/efi.h>
  49#include <asm/io.h>
  50#include <asm/irq.h>
  51#include <asm/mca.h>
  52#include <asm/page.h>
  53#include <asm/processor.h>
  54#include <asm/ptrace.h>
  55#include <asm/sal.h>
  56#include <asm/tlbflush.h>
  57#include <asm/unistd.h>
  58
  59#define SMP_DEBUG 0
  60
  61#if SMP_DEBUG
  62#define Dprintk(x...)  printk(x)
  63#else
  64#define Dprintk(x...)
  65#endif
  66
  67#ifdef CONFIG_HOTPLUG_CPU
  68#ifdef CONFIG_PERMIT_BSP_REMOVE
  69#define bsp_remove_ok   1
  70#else
  71#define bsp_remove_ok   0
  72#endif
  73
  74/*
  75 * Global array allocated for NR_CPUS at boot time
  76 */
  77struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];
  78
  79/*
  80 * start_ap in head.S uses this to store current booting cpu
  81 * info.
  82 */
  83struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];
  84
  85#define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);
  86
  87#else
  88#define set_brendez_area(x)
  89#endif
  90
  91
  92/*
  93 * ITC synchronization related stuff:
  94 */
  95#define MASTER  (0)
  96#define SLAVE   (SMP_CACHE_BYTES/8)
  97
  98#define NUM_ROUNDS      64      /* magic value */
  99#define NUM_ITERS       5       /* likewise */
 100
 101static DEFINE_SPINLOCK(itc_sync_lock);
 102static volatile unsigned long go[SLAVE + 1];
 103
 104#define DEBUG_ITC_SYNC  0
 105
 106extern void start_ap (void);
 107extern unsigned long ia64_iobase;
 108
 109struct task_struct *task_for_booting_cpu;
 110
 111/*
 112 * State for each CPU
 113 */
 114DEFINE_PER_CPU(int, cpu_state);
 115
 116cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
 117EXPORT_SYMBOL(cpu_core_map);
 118DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map);
 119EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
 120
 121int smp_num_siblings = 1;
 122
 123/* which logical CPU number maps to which CPU (physical APIC ID) */
 124volatile int ia64_cpu_to_sapicid[NR_CPUS];
 125EXPORT_SYMBOL(ia64_cpu_to_sapicid);
 126
 127static cpumask_t cpu_callin_map;
 128
 129struct smp_boot_data smp_boot_data __initdata;
 130
 131unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */
 132
 133char __initdata no_int_routing;
 134
 135unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */
 136
 137#ifdef CONFIG_FORCE_CPEI_RETARGET
 138#define CPEI_OVERRIDE_DEFAULT   (1)
 139#else
 140#define CPEI_OVERRIDE_DEFAULT   (0)
 141#endif
 142
 143unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT;
 144
 145static int __init
 146cmdl_force_cpei(char *str)
 147{
 148        int value=0;
 149
 150        get_option (&str, &value);
 151        force_cpei_retarget = value;
 152
 153        return 1;
 154}
 155
 156__setup("force_cpei=", cmdl_force_cpei);
 157
 158static int __init
 159nointroute (char *str)
 160{
 161        no_int_routing = 1;
 162        printk ("no_int_routing on\n");
 163        return 1;
 164}
 165
 166__setup("nointroute", nointroute);
 167
 168static void fix_b0_for_bsp(void)
 169{
 170#ifdef CONFIG_HOTPLUG_CPU
 171        int cpuid;
 172        static int fix_bsp_b0 = 1;
 173
 174        cpuid = smp_processor_id();
 175
 176        /*
 177         * Cache the b0 value on the first AP that comes up
 178         */
 179        if (!(fix_bsp_b0 && cpuid))
 180                return;
 181
 182        sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
 183        printk ("Fixed BSP b0 value from CPU %d\n", cpuid);
 184
 185        fix_bsp_b0 = 0;
 186#endif
 187}
 188
 189void
 190sync_master (void *arg)
 191{
 192        unsigned long flags, i;
 193
 194        go[MASTER] = 0;
 195
 196        local_irq_save(flags);
 197        {
 198                for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
 199                        while (!go[MASTER])
 200                                cpu_relax();
 201                        go[MASTER] = 0;
 202                        go[SLAVE] = ia64_get_itc();
 203                }
 204        }
 205        local_irq_restore(flags);
 206}
 207
 208/*
 209 * Return the number of cycles by which our itc differs from the itc on the master
 210 * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
 211 * negative that it is behind.
 212 */
 213static inline long
 214get_delta (long *rt, long *master)
 215{
 216        unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
 217        unsigned long tcenter, t0, t1, tm;
 218        long i;
 219
 220        for (i = 0; i < NUM_ITERS; ++i) {
 221                t0 = ia64_get_itc();
 222                go[MASTER] = 1;
 223                while (!(tm = go[SLAVE]))
 224                        cpu_relax();
 225                go[SLAVE] = 0;
 226                t1 = ia64_get_itc();
 227
 228                if (t1 - t0 < best_t1 - best_t0)
 229                        best_t0 = t0, best_t1 = t1, best_tm = tm;
 230        }
 231
 232        *rt = best_t1 - best_t0;
 233        *master = best_tm - best_t0;
 234
 235        /* average best_t0 and best_t1 without overflow: */
 236        tcenter = (best_t0/2 + best_t1/2);
 237        if (best_t0 % 2 + best_t1 % 2 == 2)
 238                ++tcenter;
 239        return tcenter - best_tm;
 240}
 241
 242/*
 243 * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
 244 * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
 245 * unaccounted-for errors (such as getting a machine check in the middle of a calibration
 246 * step).  The basic idea is for the slave to ask the master what itc value it has and to
 247 * read its own itc before and after the master responds.  Each iteration gives us three
 248 * timestamps:
 249 *
 250 *      slave           master
 251 *
 252 *      t0 ---\
 253 *             ---\
 254 *                 --->
 255 *                      tm
 256 *                 /---
 257 *             /---
 258 *      t1 <---
 259 *
 260 *
 261 * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
 262 * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
 263 * between the slave and the master is symmetric.  Even if the interconnect were
 264 * asymmetric, we would still know that the synchronization error is smaller than the
 265 * roundtrip latency (t0 - t1).
 266 *
 267 * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
 268 * within one or two cycles.  However, we can only *guarantee* that the synchronization is
 269 * accurate to within a round-trip time, which is typically in the range of several
 270 * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
 271 * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
 272 * than half a micro second or so.
 273 */
 274void
 275ia64_sync_itc (unsigned int master)
 276{
 277        long i, delta, adj, adjust_latency = 0, done = 0;
 278        unsigned long flags, rt, master_time_stamp, bound;
 279#if DEBUG_ITC_SYNC
 280        struct {
 281                long rt;        /* roundtrip time */
 282                long master;    /* master's timestamp */
 283                long diff;      /* difference between midpoint and master's timestamp */
 284                long lat;       /* estimate of itc adjustment latency */
 285        } t[NUM_ROUNDS];
 286#endif
 287
 288        /*
 289         * Make sure local timer ticks are disabled while we sync.  If
 290         * they were enabled, we'd have to worry about nasty issues
 291         * like setting the ITC ahead of (or a long time before) the
 292         * next scheduled tick.
 293         */
 294        BUG_ON((ia64_get_itv() & (1 << 16)) == 0);
 295
 296        go[MASTER] = 1;
 297
 298        if (smp_call_function_single(master, sync_master, NULL, 0) < 0) {
 299                printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
 300                return;
 301        }
 302
 303        while (go[MASTER])
 304                cpu_relax();    /* wait for master to be ready */
 305
 306        spin_lock_irqsave(&itc_sync_lock, flags);
 307        {
 308                for (i = 0; i < NUM_ROUNDS; ++i) {
 309                        delta = get_delta(&rt, &master_time_stamp);
 310                        if (delta == 0) {
 311                                done = 1;       /* let's lock on to this... */
 312                                bound = rt;
 313                        }
 314
 315                        if (!done) {
 316                                if (i > 0) {
 317                                        adjust_latency += -delta;
 318                                        adj = -delta + adjust_latency/4;
 319                                } else
 320                                        adj = -delta;
 321
 322                                ia64_set_itc(ia64_get_itc() + adj);
 323                        }
 324#if DEBUG_ITC_SYNC
 325                        t[i].rt = rt;
 326                        t[i].master = master_time_stamp;
 327                        t[i].diff = delta;
 328                        t[i].lat = adjust_latency/4;
 329#endif
 330                }
 331        }
 332        spin_unlock_irqrestore(&itc_sync_lock, flags);
 333
 334#if DEBUG_ITC_SYNC
 335        for (i = 0; i < NUM_ROUNDS; ++i)
 336                printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
 337                       t[i].rt, t[i].master, t[i].diff, t[i].lat);
 338#endif
 339
 340        printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
 341               "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
 342}
 343
 344/*
 345 * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
 346 */
 347static inline void smp_setup_percpu_timer(void)
 348{
 349}
 350
 351static void
 352smp_callin (void)
 353{
 354        int cpuid, phys_id, itc_master;
 355        struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo;
 356        extern void ia64_init_itm(void);
 357        extern volatile int time_keeper_id;
 358
 359        cpuid = smp_processor_id();
 360        phys_id = hard_smp_processor_id();
 361        itc_master = time_keeper_id;
 362
 363        if (cpu_online(cpuid)) {
 364                printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
 365                       phys_id, cpuid);
 366                BUG();
 367        }
 368
 369        fix_b0_for_bsp();
 370
 371        /*
 372         * numa_node_id() works after this.
 373         */
 374        set_numa_node(cpu_to_node_map[cpuid]);
 375        set_numa_mem(local_memory_node(cpu_to_node_map[cpuid]));
 376
 377        spin_lock(&vector_lock);
 378        /* Setup the per cpu irq handling data structures */
 379        __setup_vector_irq(cpuid);
 380        notify_cpu_starting(cpuid);
 381        set_cpu_online(cpuid, true);
 382        per_cpu(cpu_state, cpuid) = CPU_ONLINE;
 383        spin_unlock(&vector_lock);
 384
 385        smp_setup_percpu_timer();
 386
 387        ia64_mca_cmc_vector_setup();    /* Setup vector on AP */
 388
 389        local_irq_enable();
 390
 391        if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
 392                /*
 393                 * Synchronize the ITC with the BP.  Need to do this after irqs are
 394                 * enabled because ia64_sync_itc() calls smp_call_function_single(), which
 395                 * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
 396                 * local_bh_enable(), which bugs out if irqs are not enabled...
 397                 */
 398                Dprintk("Going to syncup ITC with ITC Master.\n");
 399                ia64_sync_itc(itc_master);
 400        }
 401
 402        /*
 403         * Get our bogomips.
 404         */
 405        ia64_init_itm();
 406
 407        /*
 408         * Delay calibration can be skipped if new processor is identical to the
 409         * previous processor.
 410         */
 411        last_cpuinfo = cpu_data(cpuid - 1);
 412        this_cpuinfo = local_cpu_data;
 413        if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq ||
 414            last_cpuinfo->proc_freq != this_cpuinfo->proc_freq ||
 415            last_cpuinfo->features != this_cpuinfo->features ||
 416            last_cpuinfo->revision != this_cpuinfo->revision ||
 417            last_cpuinfo->family != this_cpuinfo->family ||
 418            last_cpuinfo->archrev != this_cpuinfo->archrev ||
 419            last_cpuinfo->model != this_cpuinfo->model)
 420                calibrate_delay();
 421        local_cpu_data->loops_per_jiffy = loops_per_jiffy;
 422
 423        /*
 424         * Allow the master to continue.
 425         */
 426        cpumask_set_cpu(cpuid, &cpu_callin_map);
 427        Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
 428}
 429
 430
 431/*
 432 * Activate a secondary processor.  head.S calls this.
 433 */
 434int
 435start_secondary (void *unused)
 436{
 437        /* Early console may use I/O ports */
 438        ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
 439#ifndef CONFIG_PRINTK_TIME
 440        Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
 441#endif
 442        efi_map_pal_code();
 443        cpu_init();
 444        smp_callin();
 445
 446        cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 447        return 0;
 448}
 449
 450static int
 451do_boot_cpu (int sapicid, int cpu, struct task_struct *idle)
 452{
 453        int timeout;
 454
 455        task_for_booting_cpu = idle;
 456        Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid);
 457
 458        set_brendez_area(cpu);
 459        ia64_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0);
 460
 461        /*
 462         * Wait 10s total for the AP to start
 463         */
 464        Dprintk("Waiting on callin_map ...");
 465        for (timeout = 0; timeout < 100000; timeout++) {
 466                if (cpumask_test_cpu(cpu, &cpu_callin_map))
 467                        break;  /* It has booted */
 468                barrier(); /* Make sure we re-read cpu_callin_map */
 469                udelay(100);
 470        }
 471        Dprintk("\n");
 472
 473        if (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
 474                printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid);
 475                ia64_cpu_to_sapicid[cpu] = -1;
 476                set_cpu_online(cpu, false);  /* was set in smp_callin() */
 477                return -EINVAL;
 478        }
 479        return 0;
 480}
 481
 482static int __init
 483decay (char *str)
 484{
 485        int ticks;
 486        get_option (&str, &ticks);
 487        return 1;
 488}
 489
 490__setup("decay=", decay);
 491
 492/*
 493 * Initialize the logical CPU number to SAPICID mapping
 494 */
 495void __init
 496smp_build_cpu_map (void)
 497{
 498        int sapicid, cpu, i;
 499        int boot_cpu_id = hard_smp_processor_id();
 500
 501        for (cpu = 0; cpu < NR_CPUS; cpu++) {
 502                ia64_cpu_to_sapicid[cpu] = -1;
 503        }
 504
 505        ia64_cpu_to_sapicid[0] = boot_cpu_id;
 506        init_cpu_present(cpumask_of(0));
 507        set_cpu_possible(0, true);
 508        for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) {
 509                sapicid = smp_boot_data.cpu_phys_id[i];
 510                if (sapicid == boot_cpu_id)
 511                        continue;
 512                set_cpu_present(cpu, true);
 513                set_cpu_possible(cpu, true);
 514                ia64_cpu_to_sapicid[cpu] = sapicid;
 515                cpu++;
 516        }
 517}
 518
 519/*
 520 * Cycle through the APs sending Wakeup IPIs to boot each.
 521 */
 522void __init
 523smp_prepare_cpus (unsigned int max_cpus)
 524{
 525        int boot_cpu_id = hard_smp_processor_id();
 526
 527        /*
 528         * Initialize the per-CPU profiling counter/multiplier
 529         */
 530
 531        smp_setup_percpu_timer();
 532
 533        cpumask_set_cpu(0, &cpu_callin_map);
 534
 535        local_cpu_data->loops_per_jiffy = loops_per_jiffy;
 536        ia64_cpu_to_sapicid[0] = boot_cpu_id;
 537
 538        printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id);
 539
 540        current_thread_info()->cpu = 0;
 541
 542        /*
 543         * If SMP should be disabled, then really disable it!
 544         */
 545        if (!max_cpus) {
 546                printk(KERN_INFO "SMP mode deactivated.\n");
 547                init_cpu_online(cpumask_of(0));
 548                init_cpu_present(cpumask_of(0));
 549                init_cpu_possible(cpumask_of(0));
 550                return;
 551        }
 552}
 553
 554void smp_prepare_boot_cpu(void)
 555{
 556        set_cpu_online(smp_processor_id(), true);
 557        cpumask_set_cpu(smp_processor_id(), &cpu_callin_map);
 558        set_numa_node(cpu_to_node_map[smp_processor_id()]);
 559        per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
 560}
 561
 562#ifdef CONFIG_HOTPLUG_CPU
 563static inline void
 564clear_cpu_sibling_map(int cpu)
 565{
 566        int i;
 567
 568        for_each_cpu(i, &per_cpu(cpu_sibling_map, cpu))
 569                cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, i));
 570        for_each_cpu(i, &cpu_core_map[cpu])
 571                cpumask_clear_cpu(cpu, &cpu_core_map[i]);
 572
 573        per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE;
 574}
 575
 576static void
 577remove_siblinginfo(int cpu)
 578{
 579        int last = 0;
 580
 581        if (cpu_data(cpu)->threads_per_core == 1 &&
 582            cpu_data(cpu)->cores_per_socket == 1) {
 583                cpumask_clear_cpu(cpu, &cpu_core_map[cpu]);
 584                cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
 585                return;
 586        }
 587
 588        last = (cpumask_weight(&cpu_core_map[cpu]) == 1 ? 1 : 0);
 589
 590        /* remove it from all sibling map's */
 591        clear_cpu_sibling_map(cpu);
 592}
 593
 594extern void fixup_irqs(void);
 595
 596int migrate_platform_irqs(unsigned int cpu)
 597{
 598        int new_cpei_cpu;
 599        struct irq_data *data = NULL;
 600        const struct cpumask *mask;
 601        int             retval = 0;
 602
 603        /*
 604         * dont permit CPEI target to removed.
 605         */
 606        if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) {
 607                printk ("CPU (%d) is CPEI Target\n", cpu);
 608                if (can_cpei_retarget()) {
 609                        /*
 610                         * Now re-target the CPEI to a different processor
 611                         */
 612                        new_cpei_cpu = cpumask_any(cpu_online_mask);
 613                        mask = cpumask_of(new_cpei_cpu);
 614                        set_cpei_target_cpu(new_cpei_cpu);
 615                        data = irq_get_irq_data(ia64_cpe_irq);
 616                        /*
 617                         * Switch for now, immediately, we need to do fake intr
 618                         * as other interrupts, but need to study CPEI behaviour with
 619                         * polling before making changes.
 620                         */
 621                        if (data && data->chip) {
 622                                data->chip->irq_disable(data);
 623                                data->chip->irq_set_affinity(data, mask, false);
 624                                data->chip->irq_enable(data);
 625                                printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu);
 626                        }
 627                }
 628                if (!data) {
 629                        printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu);
 630                        retval = -EBUSY;
 631                }
 632        }
 633        return retval;
 634}
 635
 636/* must be called with cpucontrol mutex held */
 637int __cpu_disable(void)
 638{
 639        int cpu = smp_processor_id();
 640
 641        /*
 642         * dont permit boot processor for now
 643         */
 644        if (cpu == 0 && !bsp_remove_ok) {
 645                printk ("Your platform does not support removal of BSP\n");
 646                return (-EBUSY);
 647        }
 648
 649        set_cpu_online(cpu, false);
 650
 651        if (migrate_platform_irqs(cpu)) {
 652                set_cpu_online(cpu, true);
 653                return -EBUSY;
 654        }
 655
 656        remove_siblinginfo(cpu);
 657        fixup_irqs();
 658        local_flush_tlb_all();
 659        cpumask_clear_cpu(cpu, &cpu_callin_map);
 660        return 0;
 661}
 662
 663void __cpu_die(unsigned int cpu)
 664{
 665        unsigned int i;
 666
 667        for (i = 0; i < 100; i++) {
 668                /* They ack this in play_dead by setting CPU_DEAD */
 669                if (per_cpu(cpu_state, cpu) == CPU_DEAD)
 670                {
 671                        printk ("CPU %d is now offline\n", cpu);
 672                        return;
 673                }
 674                msleep(100);
 675        }
 676        printk(KERN_ERR "CPU %u didn't die...\n", cpu);
 677}
 678#endif /* CONFIG_HOTPLUG_CPU */
 679
 680void
 681smp_cpus_done (unsigned int dummy)
 682{
 683        int cpu;
 684        unsigned long bogosum = 0;
 685
 686        /*
 687         * Allow the user to impress friends.
 688         */
 689
 690        for_each_online_cpu(cpu) {
 691                bogosum += cpu_data(cpu)->loops_per_jiffy;
 692        }
 693
 694        printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
 695               (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100);
 696}
 697
 698static inline void set_cpu_sibling_map(int cpu)
 699{
 700        int i;
 701
 702        for_each_online_cpu(i) {
 703                if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) {
 704                        cpumask_set_cpu(i, &cpu_core_map[cpu]);
 705                        cpumask_set_cpu(cpu, &cpu_core_map[i]);
 706                        if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) {
 707                                cpumask_set_cpu(i,
 708                                                &per_cpu(cpu_sibling_map, cpu));
 709                                cpumask_set_cpu(cpu,
 710                                                &per_cpu(cpu_sibling_map, i));
 711                        }
 712                }
 713        }
 714}
 715
 716int
 717__cpu_up(unsigned int cpu, struct task_struct *tidle)
 718{
 719        int ret;
 720        int sapicid;
 721
 722        sapicid = ia64_cpu_to_sapicid[cpu];
 723        if (sapicid == -1)
 724                return -EINVAL;
 725
 726        /*
 727         * Already booted cpu? not valid anymore since we dont
 728         * do idle loop tightspin anymore.
 729         */
 730        if (cpumask_test_cpu(cpu, &cpu_callin_map))
 731                return -EINVAL;
 732
 733        per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
 734        /* Processor goes to start_secondary(), sets online flag */
 735        ret = do_boot_cpu(sapicid, cpu, tidle);
 736        if (ret < 0)
 737                return ret;
 738
 739        if (cpu_data(cpu)->threads_per_core == 1 &&
 740            cpu_data(cpu)->cores_per_socket == 1) {
 741                cpumask_set_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
 742                cpumask_set_cpu(cpu, &cpu_core_map[cpu]);
 743                return 0;
 744        }
 745
 746        set_cpu_sibling_map(cpu);
 747
 748        return 0;
 749}
 750
 751/*
 752 * Assume that CPUs have been discovered by some platform-dependent interface.  For
 753 * SoftSDV/Lion, that would be ACPI.
 754 *
 755 * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP().
 756 */
 757void __init
 758init_smp_config(void)
 759{
 760        struct fptr {
 761                unsigned long fp;
 762                unsigned long gp;
 763        } *ap_startup;
 764        long sal_ret;
 765
 766        /* Tell SAL where to drop the APs.  */
 767        ap_startup = (struct fptr *) start_ap;
 768        sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ,
 769                                       ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0);
 770        if (sal_ret < 0)
 771                printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n",
 772                       ia64_sal_strerror(sal_ret));
 773}
 774
 775/*
 776 * identify_siblings(cpu) gets called from identify_cpu. This populates the 
 777 * information related to logical execution units in per_cpu_data structure.
 778 */
 779void identify_siblings(struct cpuinfo_ia64 *c)
 780{
 781        long status;
 782        u16 pltid;
 783        pal_logical_to_physical_t info;
 784
 785        status = ia64_pal_logical_to_phys(-1, &info);
 786        if (status != PAL_STATUS_SUCCESS) {
 787                if (status != PAL_STATUS_UNIMPLEMENTED) {
 788                        printk(KERN_ERR
 789                                "ia64_pal_logical_to_phys failed with %ld\n",
 790                                status);
 791                        return;
 792                }
 793
 794                info.overview_ppid = 0;
 795                info.overview_cpp  = 1;
 796                info.overview_tpc  = 1;
 797        }
 798
 799        status = ia64_sal_physical_id_info(&pltid);
 800        if (status != PAL_STATUS_SUCCESS) {
 801                if (status != PAL_STATUS_UNIMPLEMENTED)
 802                        printk(KERN_ERR
 803                                "ia64_sal_pltid failed with %ld\n",
 804                                status);
 805                return;
 806        }
 807
 808        c->socket_id =  (pltid << 8) | info.overview_ppid;
 809
 810        if (info.overview_cpp == 1 && info.overview_tpc == 1)
 811                return;
 812
 813        c->cores_per_socket = info.overview_cpp;
 814        c->threads_per_core = info.overview_tpc;
 815        c->num_log = info.overview_num_log;
 816
 817        c->core_id = info.log1_cid;
 818        c->thread_id = info.log1_tid;
 819}
 820
 821/*
 822 * returns non zero, if multi-threading is enabled
 823 * on at least one physical package. Due to hotplug cpu
 824 * and (maxcpus=), all threads may not necessarily be enabled
 825 * even though the processor supports multi-threading.
 826 */
 827int is_multithreading_enabled(void)
 828{
 829        int i, j;
 830
 831        for_each_present_cpu(i) {
 832                for_each_present_cpu(j) {
 833                        if (j == i)
 834                                continue;
 835                        if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) {
 836                                if (cpu_data(j)->core_id == cpu_data(i)->core_id)
 837                                        return 1;
 838                        }
 839                }
 840        }
 841        return 0;
 842}
 843EXPORT_SYMBOL_GPL(is_multithreading_enabled);
 844