linux/arch/x86/kernel/apic/x2apic_uv_x.c
<<
>>
Prefs
   1/*
   2 * This file is subject to the terms and conditions of the GNU General Public
   3 * License.  See the file "COPYING" in the main directory of this archive
   4 * for more details.
   5 *
   6 * SGI UV APIC functions (note: not an Intel compatible APIC)
   7 *
   8 * Copyright (C) 2007-2010 Silicon Graphics, Inc. All rights reserved.
   9 */
  10#include <linux/cpumask.h>
  11#include <linux/hardirq.h>
  12#include <linux/proc_fs.h>
  13#include <linux/threads.h>
  14#include <linux/kernel.h>
  15#include <linux/module.h>
  16#include <linux/string.h>
  17#include <linux/ctype.h>
  18#include <linux/sched.h>
  19#include <linux/timer.h>
  20#include <linux/slab.h>
  21#include <linux/cpu.h>
  22#include <linux/init.h>
  23#include <linux/io.h>
  24#include <linux/pci.h>
  25#include <linux/kdebug.h>
  26#include <linux/delay.h>
  27#include <linux/crash_dump.h>
  28
  29#include <asm/uv/uv_mmrs.h>
  30#include <asm/uv/uv_hub.h>
  31#include <asm/current.h>
  32#include <asm/pgtable.h>
  33#include <asm/uv/bios.h>
  34#include <asm/uv/uv.h>
  35#include <asm/apic.h>
  36#include <asm/ipi.h>
  37#include <asm/smp.h>
  38#include <asm/x86_init.h>
  39#include <asm/emergency-restart.h>
  40#include <asm/nmi.h>
  41
  42/* BMC sets a bit this MMR non-zero before sending an NMI */
  43#define UVH_NMI_MMR                             UVH_SCRATCH5
  44#define UVH_NMI_MMR_CLEAR                       (UVH_NMI_MMR + 8)
  45#define UV_NMI_PENDING_MASK                     (1UL << 63)
  46DEFINE_PER_CPU(unsigned long, cpu_last_nmi_count);
  47
  48DEFINE_PER_CPU(int, x2apic_extra_bits);
  49
  50#define PR_DEVEL(fmt, args...)  pr_devel("%s: " fmt, __func__, args)
  51
  52static enum uv_system_type uv_system_type;
  53static u64 gru_start_paddr, gru_end_paddr;
  54static union uvh_apicid uvh_apicid;
  55int uv_min_hub_revision_id;
  56EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
  57unsigned int uv_apicid_hibits;
  58EXPORT_SYMBOL_GPL(uv_apicid_hibits);
  59static DEFINE_SPINLOCK(uv_nmi_lock);
  60
  61static struct apic apic_x2apic_uv_x;
  62
  63static unsigned long __init uv_early_read_mmr(unsigned long addr)
  64{
  65        unsigned long val, *mmr;
  66
  67        mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
  68        val = *mmr;
  69        early_iounmap(mmr, sizeof(*mmr));
  70        return val;
  71}
  72
  73static inline bool is_GRU_range(u64 start, u64 end)
  74{
  75        return start >= gru_start_paddr && end <= gru_end_paddr;
  76}
  77
  78static bool uv_is_untracked_pat_range(u64 start, u64 end)
  79{
  80        return is_ISA_range(start, end) || is_GRU_range(start, end);
  81}
  82
  83static int __init early_get_pnodeid(void)
  84{
  85        union uvh_node_id_u node_id;
  86        union uvh_rh_gam_config_mmr_u  m_n_config;
  87        int pnode;
  88
  89        /* Currently, all blades have same revision number */
  90        node_id.v = uv_early_read_mmr(UVH_NODE_ID);
  91        m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
  92        uv_min_hub_revision_id = node_id.s.revision;
  93
  94        if (node_id.s.part_number == UV2_HUB_PART_NUMBER)
  95                uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
  96        if (node_id.s.part_number == UV2_HUB_PART_NUMBER_X)
  97                uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
  98
  99        uv_hub_info->hub_revision = uv_min_hub_revision_id;
 100        pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
 101        return pnode;
 102}
 103
 104static void __init early_get_apic_pnode_shift(void)
 105{
 106        uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
 107        if (!uvh_apicid.v)
 108                /*
 109                 * Old bios, use default value
 110                 */
 111                uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
 112}
 113
 114/*
 115 * Add an extra bit as dictated by bios to the destination apicid of
 116 * interrupts potentially passing through the UV HUB.  This prevents
 117 * a deadlock between interrupts and IO port operations.
 118 */
 119static void __init uv_set_apicid_hibit(void)
 120{
 121        union uv1h_lb_target_physical_apic_id_mask_u apicid_mask;
 122
 123        if (is_uv1_hub()) {
 124                apicid_mask.v =
 125                        uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK);
 126                uv_apicid_hibits =
 127                        apicid_mask.s1.bit_enables & UV_APICID_HIBIT_MASK;
 128        }
 129}
 130
 131static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
 132{
 133        int pnodeid, is_uv1, is_uv2;
 134
 135        is_uv1 = !strcmp(oem_id, "SGI");
 136        is_uv2 = !strcmp(oem_id, "SGI2");
 137        if (is_uv1 || is_uv2) {
 138                uv_hub_info->hub_revision =
 139                        is_uv1 ? UV1_HUB_REVISION_BASE : UV2_HUB_REVISION_BASE;
 140                pnodeid = early_get_pnodeid();
 141                early_get_apic_pnode_shift();
 142                x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
 143                x86_platform.nmi_init = uv_nmi_init;
 144                if (!strcmp(oem_table_id, "UVL"))
 145                        uv_system_type = UV_LEGACY_APIC;
 146                else if (!strcmp(oem_table_id, "UVX"))
 147                        uv_system_type = UV_X2APIC;
 148                else if (!strcmp(oem_table_id, "UVH")) {
 149                        __this_cpu_write(x2apic_extra_bits,
 150                                pnodeid << uvh_apicid.s.pnode_shift);
 151                        uv_system_type = UV_NON_UNIQUE_APIC;
 152                        uv_set_apicid_hibit();
 153                        return 1;
 154                }
 155        }
 156        return 0;
 157}
 158
 159enum uv_system_type get_uv_system_type(void)
 160{
 161        return uv_system_type;
 162}
 163
 164int is_uv_system(void)
 165{
 166        return uv_system_type != UV_NONE;
 167}
 168EXPORT_SYMBOL_GPL(is_uv_system);
 169
 170DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
 171EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
 172
 173struct uv_blade_info *uv_blade_info;
 174EXPORT_SYMBOL_GPL(uv_blade_info);
 175
 176short *uv_node_to_blade;
 177EXPORT_SYMBOL_GPL(uv_node_to_blade);
 178
 179short *uv_cpu_to_blade;
 180EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
 181
 182short uv_possible_blades;
 183EXPORT_SYMBOL_GPL(uv_possible_blades);
 184
 185unsigned long sn_rtc_cycles_per_second;
 186EXPORT_SYMBOL(sn_rtc_cycles_per_second);
 187
 188static const struct cpumask *uv_target_cpus(void)
 189{
 190        return cpu_online_mask;
 191}
 192
 193static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
 194{
 195        cpumask_clear(retmask);
 196        cpumask_set_cpu(cpu, retmask);
 197}
 198
 199static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
 200{
 201#ifdef CONFIG_SMP
 202        unsigned long val;
 203        int pnode;
 204
 205        pnode = uv_apicid_to_pnode(phys_apicid);
 206        phys_apicid |= uv_apicid_hibits;
 207        val = (1UL << UVH_IPI_INT_SEND_SHFT) |
 208            (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
 209            ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
 210            APIC_DM_INIT;
 211        uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
 212
 213        val = (1UL << UVH_IPI_INT_SEND_SHFT) |
 214            (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
 215            ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
 216            APIC_DM_STARTUP;
 217        uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
 218
 219        atomic_set(&init_deasserted, 1);
 220#endif
 221        return 0;
 222}
 223
 224static void uv_send_IPI_one(int cpu, int vector)
 225{
 226        unsigned long apicid;
 227        int pnode;
 228
 229        apicid = per_cpu(x86_cpu_to_apicid, cpu);
 230        pnode = uv_apicid_to_pnode(apicid);
 231        uv_hub_send_ipi(pnode, apicid, vector);
 232}
 233
 234static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
 235{
 236        unsigned int cpu;
 237
 238        for_each_cpu(cpu, mask)
 239                uv_send_IPI_one(cpu, vector);
 240}
 241
 242static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
 243{
 244        unsigned int this_cpu = smp_processor_id();
 245        unsigned int cpu;
 246
 247        for_each_cpu(cpu, mask) {
 248                if (cpu != this_cpu)
 249                        uv_send_IPI_one(cpu, vector);
 250        }
 251}
 252
 253static void uv_send_IPI_allbutself(int vector)
 254{
 255        unsigned int this_cpu = smp_processor_id();
 256        unsigned int cpu;
 257
 258        for_each_online_cpu(cpu) {
 259                if (cpu != this_cpu)
 260                        uv_send_IPI_one(cpu, vector);
 261        }
 262}
 263
 264static void uv_send_IPI_all(int vector)
 265{
 266        uv_send_IPI_mask(cpu_online_mask, vector);
 267}
 268
 269static int uv_apic_id_registered(void)
 270{
 271        return 1;
 272}
 273
 274static void uv_init_apic_ldr(void)
 275{
 276}
 277
 278static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
 279{
 280        /*
 281         * We're using fixed IRQ delivery, can only return one phys APIC ID.
 282         * May as well be the first.
 283         */
 284        int cpu = cpumask_first(cpumask);
 285
 286        if ((unsigned)cpu < nr_cpu_ids)
 287                return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
 288        else
 289                return BAD_APICID;
 290}
 291
 292static unsigned int
 293uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
 294                          const struct cpumask *andmask)
 295{
 296        int cpu;
 297
 298        /*
 299         * We're using fixed IRQ delivery, can only return one phys APIC ID.
 300         * May as well be the first.
 301         */
 302        for_each_cpu_and(cpu, cpumask, andmask) {
 303                if (cpumask_test_cpu(cpu, cpu_online_mask))
 304                        break;
 305        }
 306        return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
 307}
 308
 309static unsigned int x2apic_get_apic_id(unsigned long x)
 310{
 311        unsigned int id;
 312
 313        WARN_ON(preemptible() && num_online_cpus() > 1);
 314        id = x | __this_cpu_read(x2apic_extra_bits);
 315
 316        return id;
 317}
 318
 319static unsigned long set_apic_id(unsigned int id)
 320{
 321        unsigned long x;
 322
 323        /* maskout x2apic_extra_bits ? */
 324        x = id;
 325        return x;
 326}
 327
 328static unsigned int uv_read_apic_id(void)
 329{
 330
 331        return x2apic_get_apic_id(apic_read(APIC_ID));
 332}
 333
 334static int uv_phys_pkg_id(int initial_apicid, int index_msb)
 335{
 336        return uv_read_apic_id() >> index_msb;
 337}
 338
 339static void uv_send_IPI_self(int vector)
 340{
 341        apic_write(APIC_SELF_IPI, vector);
 342}
 343
 344static int uv_probe(void)
 345{
 346        return apic == &apic_x2apic_uv_x;
 347}
 348
 349static struct apic __refdata apic_x2apic_uv_x = {
 350
 351        .name                           = "UV large system",
 352        .probe                          = uv_probe,
 353        .acpi_madt_oem_check            = uv_acpi_madt_oem_check,
 354        .apic_id_registered             = uv_apic_id_registered,
 355
 356        .irq_delivery_mode              = dest_Fixed,
 357        .irq_dest_mode                  = 0, /* physical */
 358
 359        .target_cpus                    = uv_target_cpus,
 360        .disable_esr                    = 0,
 361        .dest_logical                   = APIC_DEST_LOGICAL,
 362        .check_apicid_used              = NULL,
 363        .check_apicid_present           = NULL,
 364
 365        .vector_allocation_domain       = uv_vector_allocation_domain,
 366        .init_apic_ldr                  = uv_init_apic_ldr,
 367
 368        .ioapic_phys_id_map             = NULL,
 369        .setup_apic_routing             = NULL,
 370        .multi_timer_check              = NULL,
 371        .cpu_present_to_apicid          = default_cpu_present_to_apicid,
 372        .apicid_to_cpu_present          = NULL,
 373        .setup_portio_remap             = NULL,
 374        .check_phys_apicid_present      = default_check_phys_apicid_present,
 375        .enable_apic_mode               = NULL,
 376        .phys_pkg_id                    = uv_phys_pkg_id,
 377        .mps_oem_check                  = NULL,
 378
 379        .get_apic_id                    = x2apic_get_apic_id,
 380        .set_apic_id                    = set_apic_id,
 381        .apic_id_mask                   = 0xFFFFFFFFu,
 382
 383        .cpu_mask_to_apicid             = uv_cpu_mask_to_apicid,
 384        .cpu_mask_to_apicid_and         = uv_cpu_mask_to_apicid_and,
 385
 386        .send_IPI_mask                  = uv_send_IPI_mask,
 387        .send_IPI_mask_allbutself       = uv_send_IPI_mask_allbutself,
 388        .send_IPI_allbutself            = uv_send_IPI_allbutself,
 389        .send_IPI_all                   = uv_send_IPI_all,
 390        .send_IPI_self                  = uv_send_IPI_self,
 391
 392        .wakeup_secondary_cpu           = uv_wakeup_secondary,
 393        .trampoline_phys_low            = DEFAULT_TRAMPOLINE_PHYS_LOW,
 394        .trampoline_phys_high           = DEFAULT_TRAMPOLINE_PHYS_HIGH,
 395        .wait_for_init_deassert         = NULL,
 396        .smp_callin_clear_local_apic    = NULL,
 397        .inquire_remote_apic            = NULL,
 398
 399        .read                           = native_apic_msr_read,
 400        .write                          = native_apic_msr_write,
 401        .icr_read                       = native_x2apic_icr_read,
 402        .icr_write                      = native_x2apic_icr_write,
 403        .wait_icr_idle                  = native_x2apic_wait_icr_idle,
 404        .safe_wait_icr_idle             = native_safe_x2apic_wait_icr_idle,
 405};
 406
 407static __cpuinit void set_x2apic_extra_bits(int pnode)
 408{
 409        __this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
 410}
 411
 412/*
 413 * Called on boot cpu.
 414 */
 415static __init int boot_pnode_to_blade(int pnode)
 416{
 417        int blade;
 418
 419        for (blade = 0; blade < uv_num_possible_blades(); blade++)
 420                if (pnode == uv_blade_info[blade].pnode)
 421                        return blade;
 422        BUG();
 423}
 424
 425struct redir_addr {
 426        unsigned long redirect;
 427        unsigned long alias;
 428};
 429
 430#define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
 431
 432static __initdata struct redir_addr redir_addrs[] = {
 433        {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
 434        {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
 435        {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
 436};
 437
 438static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
 439{
 440        union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
 441        union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
 442        int i;
 443
 444        for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
 445                alias.v = uv_read_local_mmr(redir_addrs[i].alias);
 446                if (alias.s.enable && alias.s.base == 0) {
 447                        *size = (1UL << alias.s.m_alias);
 448                        redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
 449                        *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
 450                        return;
 451                }
 452        }
 453        *base = *size = 0;
 454}
 455
 456enum map_type {map_wb, map_uc};
 457
 458static __init void map_high(char *id, unsigned long base, int pshift,
 459                        int bshift, int max_pnode, enum map_type map_type)
 460{
 461        unsigned long bytes, paddr;
 462
 463        paddr = base << pshift;
 464        bytes = (1UL << bshift) * (max_pnode + 1);
 465        printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
 466                                                paddr + bytes);
 467        if (map_type == map_uc)
 468                init_extra_mapping_uc(paddr, bytes);
 469        else
 470                init_extra_mapping_wb(paddr, bytes);
 471
 472}
 473static __init void map_gru_high(int max_pnode)
 474{
 475        union uvh_rh_gam_gru_overlay_config_mmr_u gru;
 476        int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
 477
 478        gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
 479        if (gru.s.enable) {
 480                map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
 481                gru_start_paddr = ((u64)gru.s.base << shift);
 482                gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
 483
 484        }
 485}
 486
 487static __init void map_mmr_high(int max_pnode)
 488{
 489        union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
 490        int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
 491
 492        mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
 493        if (mmr.s.enable)
 494                map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
 495}
 496
 497static __init void map_mmioh_high(int max_pnode)
 498{
 499        union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
 500        int shift;
 501
 502        mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
 503        if (is_uv1_hub() && mmioh.s1.enable) {
 504                shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
 505                map_high("MMIOH", mmioh.s1.base, shift, mmioh.s1.m_io,
 506                        max_pnode, map_uc);
 507        }
 508        if (is_uv2_hub() && mmioh.s2.enable) {
 509                shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
 510                map_high("MMIOH", mmioh.s2.base, shift, mmioh.s2.m_io,
 511                        max_pnode, map_uc);
 512        }
 513}
 514
 515static __init void map_low_mmrs(void)
 516{
 517        init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
 518        init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
 519}
 520
 521static __init void uv_rtc_init(void)
 522{
 523        long status;
 524        u64 ticks_per_sec;
 525
 526        status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
 527                                        &ticks_per_sec);
 528        if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
 529                printk(KERN_WARNING
 530                        "unable to determine platform RTC clock frequency, "
 531                        "guessing.\n");
 532                /* BIOS gives wrong value for clock freq. so guess */
 533                sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
 534        } else
 535                sn_rtc_cycles_per_second = ticks_per_sec;
 536}
 537
 538/*
 539 * percpu heartbeat timer
 540 */
 541static void uv_heartbeat(unsigned long ignored)
 542{
 543        struct timer_list *timer = &uv_hub_info->scir.timer;
 544        unsigned char bits = uv_hub_info->scir.state;
 545
 546        /* flip heartbeat bit */
 547        bits ^= SCIR_CPU_HEARTBEAT;
 548
 549        /* is this cpu idle? */
 550        if (idle_cpu(raw_smp_processor_id()))
 551                bits &= ~SCIR_CPU_ACTIVITY;
 552        else
 553                bits |= SCIR_CPU_ACTIVITY;
 554
 555        /* update system controller interface reg */
 556        uv_set_scir_bits(bits);
 557
 558        /* enable next timer period */
 559        mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
 560}
 561
 562static void __cpuinit uv_heartbeat_enable(int cpu)
 563{
 564        while (!uv_cpu_hub_info(cpu)->scir.enabled) {
 565                struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
 566
 567                uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
 568                setup_timer(timer, uv_heartbeat, cpu);
 569                timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
 570                add_timer_on(timer, cpu);
 571                uv_cpu_hub_info(cpu)->scir.enabled = 1;
 572
 573                /* also ensure that boot cpu is enabled */
 574                cpu = 0;
 575        }
 576}
 577
 578#ifdef CONFIG_HOTPLUG_CPU
 579static void __cpuinit uv_heartbeat_disable(int cpu)
 580{
 581        if (uv_cpu_hub_info(cpu)->scir.enabled) {
 582                uv_cpu_hub_info(cpu)->scir.enabled = 0;
 583                del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
 584        }
 585        uv_set_cpu_scir_bits(cpu, 0xff);
 586}
 587
 588/*
 589 * cpu hotplug notifier
 590 */
 591static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
 592                                       unsigned long action, void *hcpu)
 593{
 594        long cpu = (long)hcpu;
 595
 596        switch (action) {
 597        case CPU_ONLINE:
 598                uv_heartbeat_enable(cpu);
 599                break;
 600        case CPU_DOWN_PREPARE:
 601                uv_heartbeat_disable(cpu);
 602                break;
 603        default:
 604                break;
 605        }
 606        return NOTIFY_OK;
 607}
 608
 609static __init void uv_scir_register_cpu_notifier(void)
 610{
 611        hotcpu_notifier(uv_scir_cpu_notify, 0);
 612}
 613
 614#else /* !CONFIG_HOTPLUG_CPU */
 615
 616static __init void uv_scir_register_cpu_notifier(void)
 617{
 618}
 619
 620static __init int uv_init_heartbeat(void)
 621{
 622        int cpu;
 623
 624        if (is_uv_system())
 625                for_each_online_cpu(cpu)
 626                        uv_heartbeat_enable(cpu);
 627        return 0;
 628}
 629
 630late_initcall(uv_init_heartbeat);
 631
 632#endif /* !CONFIG_HOTPLUG_CPU */
 633
 634/* Direct Legacy VGA I/O traffic to designated IOH */
 635int uv_set_vga_state(struct pci_dev *pdev, bool decode,
 636                      unsigned int command_bits, u32 flags)
 637{
 638        int domain, bus, rc;
 639
 640        PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
 641                        pdev->devfn, decode, command_bits, flags);
 642
 643        if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
 644                return 0;
 645
 646        if ((command_bits & PCI_COMMAND_IO) == 0)
 647                return 0;
 648
 649        domain = pci_domain_nr(pdev->bus);
 650        bus = pdev->bus->number;
 651
 652        rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
 653        PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
 654
 655        return rc;
 656}
 657
 658/*
 659 * Called on each cpu to initialize the per_cpu UV data area.
 660 * FIXME: hotplug not supported yet
 661 */
 662void __cpuinit uv_cpu_init(void)
 663{
 664        /* CPU 0 initilization will be done via uv_system_init. */
 665        if (!uv_blade_info)
 666                return;
 667
 668        uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
 669
 670        if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
 671                set_x2apic_extra_bits(uv_hub_info->pnode);
 672}
 673
 674/*
 675 * When NMI is received, print a stack trace.
 676 */
 677int uv_handle_nmi(unsigned int reason, struct pt_regs *regs)
 678{
 679        unsigned long real_uv_nmi;
 680        int bid;
 681
 682        /*
 683         * Each blade has an MMR that indicates when an NMI has been sent
 684         * to cpus on the blade. If an NMI is detected, atomically
 685         * clear the MMR and update a per-blade NMI count used to
 686         * cause each cpu on the blade to notice a new NMI.
 687         */
 688        bid = uv_numa_blade_id();
 689        real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
 690
 691        if (unlikely(real_uv_nmi)) {
 692                spin_lock(&uv_blade_info[bid].nmi_lock);
 693                real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
 694                if (real_uv_nmi) {
 695                        uv_blade_info[bid].nmi_count++;
 696                        uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
 697                }
 698                spin_unlock(&uv_blade_info[bid].nmi_lock);
 699        }
 700
 701        if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
 702                return NMI_DONE;
 703
 704        __get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
 705
 706        /*
 707         * Use a lock so only one cpu prints at a time.
 708         * This prevents intermixed output.
 709         */
 710        spin_lock(&uv_nmi_lock);
 711        pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
 712        dump_stack();
 713        spin_unlock(&uv_nmi_lock);
 714
 715        return NMI_HANDLED;
 716}
 717
 718void uv_register_nmi_notifier(void)
 719{
 720        if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv"))
 721                printk(KERN_WARNING "UV NMI handler failed to register\n");
 722}
 723
 724void uv_nmi_init(void)
 725{
 726        unsigned int value;
 727
 728        /*
 729         * Unmask NMI on all cpus
 730         */
 731        value = apic_read(APIC_LVT1) | APIC_DM_NMI;
 732        value &= ~APIC_LVT_MASKED;
 733        apic_write(APIC_LVT1, value);
 734}
 735
 736void __init uv_system_init(void)
 737{
 738        union uvh_rh_gam_config_mmr_u  m_n_config;
 739        union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
 740        union uvh_node_id_u node_id;
 741        unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
 742        int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
 743        int gnode_extra, max_pnode = 0;
 744        unsigned long mmr_base, present, paddr;
 745        unsigned short pnode_mask, pnode_io_mask;
 746
 747        printk(KERN_INFO "UV: Found %s hub\n", is_uv1_hub() ? "UV1" : "UV2");
 748        map_low_mmrs();
 749
 750        m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
 751        m_val = m_n_config.s.m_skt;
 752        n_val = m_n_config.s.n_skt;
 753        mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
 754        n_io = is_uv1_hub() ? mmioh.s1.n_io : mmioh.s2.n_io;
 755        mmr_base =
 756            uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
 757            ~UV_MMR_ENABLE;
 758        pnode_mask = (1 << n_val) - 1;
 759        pnode_io_mask = (1 << n_io) - 1;
 760
 761        node_id.v = uv_read_local_mmr(UVH_NODE_ID);
 762        gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
 763        gnode_upper = ((unsigned long)gnode_extra  << m_val);
 764        printk(KERN_INFO "UV: N %d, M %d, N_IO: %d, gnode_upper 0x%lx, gnode_extra 0x%x, pnode_mask 0x%x, pnode_io_mask 0x%x\n",
 765                        n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
 766
 767        printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
 768
 769        for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
 770                uv_possible_blades +=
 771                  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
 772
 773        /* uv_num_possible_blades() is really the hub count */
 774        printk(KERN_INFO "UV: Found %d blades, %d hubs\n",
 775                        is_uv1_hub() ? uv_num_possible_blades() :
 776                        (uv_num_possible_blades() + 1) / 2,
 777                        uv_num_possible_blades());
 778
 779        bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
 780        uv_blade_info = kzalloc(bytes, GFP_KERNEL);
 781        BUG_ON(!uv_blade_info);
 782
 783        for (blade = 0; blade < uv_num_possible_blades(); blade++)
 784                uv_blade_info[blade].memory_nid = -1;
 785
 786        get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
 787
 788        bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
 789        uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
 790        BUG_ON(!uv_node_to_blade);
 791        memset(uv_node_to_blade, 255, bytes);
 792
 793        bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
 794        uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
 795        BUG_ON(!uv_cpu_to_blade);
 796        memset(uv_cpu_to_blade, 255, bytes);
 797
 798        blade = 0;
 799        for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
 800                present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
 801                for (j = 0; j < 64; j++) {
 802                        if (!test_bit(j, &present))
 803                                continue;
 804                        pnode = (i * 64 + j) & pnode_mask;
 805                        uv_blade_info[blade].pnode = pnode;
 806                        uv_blade_info[blade].nr_possible_cpus = 0;
 807                        uv_blade_info[blade].nr_online_cpus = 0;
 808                        spin_lock_init(&uv_blade_info[blade].nmi_lock);
 809                        max_pnode = max(pnode, max_pnode);
 810                        blade++;
 811                }
 812        }
 813
 814        uv_bios_init();
 815        uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
 816                            &sn_region_size, &system_serial_number);
 817        uv_rtc_init();
 818
 819        for_each_present_cpu(cpu) {
 820                int apicid = per_cpu(x86_cpu_to_apicid, cpu);
 821
 822                nid = cpu_to_node(cpu);
 823                /*
 824                 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
 825                 */
 826                uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
 827                uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
 828                uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
 829
 830                uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
 831                uv_cpu_hub_info(cpu)->n_lshift = is_uv2_1_hub() ?
 832                                (m_val == 40 ? 40 : 39) : m_val;
 833
 834                pnode = uv_apicid_to_pnode(apicid);
 835                blade = boot_pnode_to_blade(pnode);
 836                lcpu = uv_blade_info[blade].nr_possible_cpus;
 837                uv_blade_info[blade].nr_possible_cpus++;
 838
 839                /* Any node on the blade, else will contain -1. */
 840                uv_blade_info[blade].memory_nid = nid;
 841
 842                uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
 843                uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
 844                uv_cpu_hub_info(cpu)->m_val = m_val;
 845                uv_cpu_hub_info(cpu)->n_val = n_val;
 846                uv_cpu_hub_info(cpu)->numa_blade_id = blade;
 847                uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
 848                uv_cpu_hub_info(cpu)->pnode = pnode;
 849                uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
 850                uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
 851                uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
 852                uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
 853                uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
 854                uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
 855                uv_node_to_blade[nid] = blade;
 856                uv_cpu_to_blade[cpu] = blade;
 857        }
 858
 859        /* Add blade/pnode info for nodes without cpus */
 860        for_each_online_node(nid) {
 861                if (uv_node_to_blade[nid] >= 0)
 862                        continue;
 863                paddr = node_start_pfn(nid) << PAGE_SHIFT;
 864                pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
 865                blade = boot_pnode_to_blade(pnode);
 866                uv_node_to_blade[nid] = blade;
 867        }
 868
 869        map_gru_high(max_pnode);
 870        map_mmr_high(max_pnode);
 871        map_mmioh_high(max_pnode & pnode_io_mask);
 872
 873        uv_cpu_init();
 874        uv_scir_register_cpu_notifier();
 875        uv_register_nmi_notifier();
 876        proc_mkdir("sgi_uv", NULL);
 877
 878        /* register Legacy VGA I/O redirection handler */
 879        pci_register_set_vga_state(uv_set_vga_state);
 880
 881        /*
 882         * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
 883         * EFI is not enabled in the kdump kernel.
 884         */
 885        if (is_kdump_kernel())
 886                reboot_type = BOOT_ACPI;
 887}
 888
 889apic_driver(apic_x2apic_uv_x);
 890
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.