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_valid(int apicid)
 270{
 271        return 1;
 272}
 273
 274static int uv_apic_id_registered(void)
 275{
 276        return 1;
 277}
 278
 279static void uv_init_apic_ldr(void)
 280{
 281}
 282
 283static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
 284{
 285        /*
 286         * We're using fixed IRQ delivery, can only return one phys APIC ID.
 287         * May as well be the first.
 288         */
 289        int cpu = cpumask_first(cpumask);
 290
 291        if ((unsigned)cpu < nr_cpu_ids)
 292                return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
 293        else
 294                return BAD_APICID;
 295}
 296
 297static unsigned int
 298uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
 299                          const struct cpumask *andmask)
 300{
 301        int cpu;
 302
 303        /*
 304         * We're using fixed IRQ delivery, can only return one phys APIC ID.
 305         * May as well be the first.
 306         */
 307        for_each_cpu_and(cpu, cpumask, andmask) {
 308                if (cpumask_test_cpu(cpu, cpu_online_mask))
 309                        break;
 310        }
 311        return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
 312}
 313
 314static unsigned int x2apic_get_apic_id(unsigned long x)
 315{
 316        unsigned int id;
 317
 318        WARN_ON(preemptible() && num_online_cpus() > 1);
 319        id = x | __this_cpu_read(x2apic_extra_bits);
 320
 321        return id;
 322}
 323
 324static unsigned long set_apic_id(unsigned int id)
 325{
 326        unsigned long x;
 327
 328        /* maskout x2apic_extra_bits ? */
 329        x = id;
 330        return x;
 331}
 332
 333static unsigned int uv_read_apic_id(void)
 334{
 335
 336        return x2apic_get_apic_id(apic_read(APIC_ID));
 337}
 338
 339static int uv_phys_pkg_id(int initial_apicid, int index_msb)
 340{
 341        return uv_read_apic_id() >> index_msb;
 342}
 343
 344static void uv_send_IPI_self(int vector)
 345{
 346        apic_write(APIC_SELF_IPI, vector);
 347}
 348
 349static int uv_probe(void)
 350{
 351        return apic == &apic_x2apic_uv_x;
 352}
 353
 354static struct apic __refdata apic_x2apic_uv_x = {
 355
 356        .name                           = "UV large system",
 357        .probe                          = uv_probe,
 358        .acpi_madt_oem_check            = uv_acpi_madt_oem_check,
 359        .apic_id_valid                  = uv_apic_id_valid,
 360        .apic_id_registered             = uv_apic_id_registered,
 361
 362        .irq_delivery_mode              = dest_Fixed,
 363        .irq_dest_mode                  = 0, /* physical */
 364
 365        .target_cpus                    = uv_target_cpus,
 366        .disable_esr                    = 0,
 367        .dest_logical                   = APIC_DEST_LOGICAL,
 368        .check_apicid_used              = NULL,
 369        .check_apicid_present           = NULL,
 370
 371        .vector_allocation_domain       = uv_vector_allocation_domain,
 372        .init_apic_ldr                  = uv_init_apic_ldr,
 373
 374        .ioapic_phys_id_map             = NULL,
 375        .setup_apic_routing             = NULL,
 376        .multi_timer_check              = NULL,
 377        .cpu_present_to_apicid          = default_cpu_present_to_apicid,
 378        .apicid_to_cpu_present          = NULL,
 379        .setup_portio_remap             = NULL,
 380        .check_phys_apicid_present      = default_check_phys_apicid_present,
 381        .enable_apic_mode               = NULL,
 382        .phys_pkg_id                    = uv_phys_pkg_id,
 383        .mps_oem_check                  = NULL,
 384
 385        .get_apic_id                    = x2apic_get_apic_id,
 386        .set_apic_id                    = set_apic_id,
 387        .apic_id_mask                   = 0xFFFFFFFFu,
 388
 389        .cpu_mask_to_apicid             = uv_cpu_mask_to_apicid,
 390        .cpu_mask_to_apicid_and         = uv_cpu_mask_to_apicid_and,
 391
 392        .send_IPI_mask                  = uv_send_IPI_mask,
 393        .send_IPI_mask_allbutself       = uv_send_IPI_mask_allbutself,
 394        .send_IPI_allbutself            = uv_send_IPI_allbutself,
 395        .send_IPI_all                   = uv_send_IPI_all,
 396        .send_IPI_self                  = uv_send_IPI_self,
 397
 398        .wakeup_secondary_cpu           = uv_wakeup_secondary,
 399        .trampoline_phys_low            = DEFAULT_TRAMPOLINE_PHYS_LOW,
 400        .trampoline_phys_high           = DEFAULT_TRAMPOLINE_PHYS_HIGH,
 401        .wait_for_init_deassert         = NULL,
 402        .smp_callin_clear_local_apic    = NULL,
 403        .inquire_remote_apic            = NULL,
 404
 405        .read                           = native_apic_msr_read,
 406        .write                          = native_apic_msr_write,
 407        .eoi_write                      = native_apic_msr_eoi_write,
 408        .icr_read                       = native_x2apic_icr_read,
 409        .icr_write                      = native_x2apic_icr_write,
 410        .wait_icr_idle                  = native_x2apic_wait_icr_idle,
 411        .safe_wait_icr_idle             = native_safe_x2apic_wait_icr_idle,
 412};
 413
 414static __cpuinit void set_x2apic_extra_bits(int pnode)
 415{
 416        __this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
 417}
 418
 419/*
 420 * Called on boot cpu.
 421 */
 422static __init int boot_pnode_to_blade(int pnode)
 423{
 424        int blade;
 425
 426        for (blade = 0; blade < uv_num_possible_blades(); blade++)
 427                if (pnode == uv_blade_info[blade].pnode)
 428                        return blade;
 429        BUG();
 430}
 431
 432struct redir_addr {
 433        unsigned long redirect;
 434        unsigned long alias;
 435};
 436
 437#define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
 438
 439static __initdata struct redir_addr redir_addrs[] = {
 440        {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
 441        {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
 442        {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
 443};
 444
 445static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
 446{
 447        union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
 448        union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
 449        int i;
 450
 451        for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
 452                alias.v = uv_read_local_mmr(redir_addrs[i].alias);
 453                if (alias.s.enable && alias.s.base == 0) {
 454                        *size = (1UL << alias.s.m_alias);
 455                        redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
 456                        *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
 457                        return;
 458                }
 459        }
 460        *base = *size = 0;
 461}
 462
 463enum map_type {map_wb, map_uc};
 464
 465static __init void map_high(char *id, unsigned long base, int pshift,
 466                        int bshift, int max_pnode, enum map_type map_type)
 467{
 468        unsigned long bytes, paddr;
 469
 470        paddr = base << pshift;
 471        bytes = (1UL << bshift) * (max_pnode + 1);
 472        printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
 473                                                paddr + bytes);
 474        if (map_type == map_uc)
 475                init_extra_mapping_uc(paddr, bytes);
 476        else
 477                init_extra_mapping_wb(paddr, bytes);
 478
 479}
 480static __init void map_gru_high(int max_pnode)
 481{
 482        union uvh_rh_gam_gru_overlay_config_mmr_u gru;
 483        int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
 484
 485        gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
 486        if (gru.s.enable) {
 487                map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
 488                gru_start_paddr = ((u64)gru.s.base << shift);
 489                gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
 490
 491        }
 492}
 493
 494static __init void map_mmr_high(int max_pnode)
 495{
 496        union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
 497        int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
 498
 499        mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
 500        if (mmr.s.enable)
 501                map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
 502}
 503
 504static __init void map_mmioh_high(int max_pnode)
 505{
 506        union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
 507        int shift;
 508
 509        mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
 510        if (is_uv1_hub() && mmioh.s1.enable) {
 511                shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
 512                map_high("MMIOH", mmioh.s1.base, shift, mmioh.s1.m_io,
 513                        max_pnode, map_uc);
 514        }
 515        if (is_uv2_hub() && mmioh.s2.enable) {
 516                shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
 517                map_high("MMIOH", mmioh.s2.base, shift, mmioh.s2.m_io,
 518                        max_pnode, map_uc);
 519        }
 520}
 521
 522static __init void map_low_mmrs(void)
 523{
 524        init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
 525        init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
 526}
 527
 528static __init void uv_rtc_init(void)
 529{
 530        long status;
 531        u64 ticks_per_sec;
 532
 533        status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
 534                                        &ticks_per_sec);
 535        if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
 536                printk(KERN_WARNING
 537                        "unable to determine platform RTC clock frequency, "
 538                        "guessing.\n");
 539                /* BIOS gives wrong value for clock freq. so guess */
 540                sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
 541        } else
 542                sn_rtc_cycles_per_second = ticks_per_sec;
 543}
 544
 545/*
 546 * percpu heartbeat timer
 547 */
 548static void uv_heartbeat(unsigned long ignored)
 549{
 550        struct timer_list *timer = &uv_hub_info->scir.timer;
 551        unsigned char bits = uv_hub_info->scir.state;
 552
 553        /* flip heartbeat bit */
 554        bits ^= SCIR_CPU_HEARTBEAT;
 555
 556        /* is this cpu idle? */
 557        if (idle_cpu(raw_smp_processor_id()))
 558                bits &= ~SCIR_CPU_ACTIVITY;
 559        else
 560                bits |= SCIR_CPU_ACTIVITY;
 561
 562        /* update system controller interface reg */
 563        uv_set_scir_bits(bits);
 564
 565        /* enable next timer period */
 566        mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
 567}
 568
 569static void __cpuinit uv_heartbeat_enable(int cpu)
 570{
 571        while (!uv_cpu_hub_info(cpu)->scir.enabled) {
 572                struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
 573
 574                uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
 575                setup_timer(timer, uv_heartbeat, cpu);
 576                timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
 577                add_timer_on(timer, cpu);
 578                uv_cpu_hub_info(cpu)->scir.enabled = 1;
 579
 580                /* also ensure that boot cpu is enabled */
 581                cpu = 0;
 582        }
 583}
 584
 585#ifdef CONFIG_HOTPLUG_CPU
 586static void __cpuinit uv_heartbeat_disable(int cpu)
 587{
 588        if (uv_cpu_hub_info(cpu)->scir.enabled) {
 589                uv_cpu_hub_info(cpu)->scir.enabled = 0;
 590                del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
 591        }
 592        uv_set_cpu_scir_bits(cpu, 0xff);
 593}
 594
 595/*
 596 * cpu hotplug notifier
 597 */
 598static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
 599                                       unsigned long action, void *hcpu)
 600{
 601        long cpu = (long)hcpu;
 602
 603        switch (action) {
 604        case CPU_ONLINE:
 605                uv_heartbeat_enable(cpu);
 606                break;
 607        case CPU_DOWN_PREPARE:
 608                uv_heartbeat_disable(cpu);
 609                break;
 610        default:
 611                break;
 612        }
 613        return NOTIFY_OK;
 614}
 615
 616static __init void uv_scir_register_cpu_notifier(void)
 617{
 618        hotcpu_notifier(uv_scir_cpu_notify, 0);
 619}
 620
 621#else /* !CONFIG_HOTPLUG_CPU */
 622
 623static __init void uv_scir_register_cpu_notifier(void)
 624{
 625}
 626
 627static __init int uv_init_heartbeat(void)
 628{
 629        int cpu;
 630
 631        if (is_uv_system())
 632                for_each_online_cpu(cpu)
 633                        uv_heartbeat_enable(cpu);
 634        return 0;
 635}
 636
 637late_initcall(uv_init_heartbeat);
 638
 639#endif /* !CONFIG_HOTPLUG_CPU */
 640
 641/* Direct Legacy VGA I/O traffic to designated IOH */
 642int uv_set_vga_state(struct pci_dev *pdev, bool decode,
 643                      unsigned int command_bits, u32 flags)
 644{
 645        int domain, bus, rc;
 646
 647        PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
 648                        pdev->devfn, decode, command_bits, flags);
 649
 650        if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
 651                return 0;
 652
 653        if ((command_bits & PCI_COMMAND_IO) == 0)
 654                return 0;
 655
 656        domain = pci_domain_nr(pdev->bus);
 657        bus = pdev->bus->number;
 658
 659        rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
 660        PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
 661
 662        return rc;
 663}
 664
 665/*
 666 * Called on each cpu to initialize the per_cpu UV data area.
 667 * FIXME: hotplug not supported yet
 668 */
 669void __cpuinit uv_cpu_init(void)
 670{
 671        /* CPU 0 initilization will be done via uv_system_init. */
 672        if (!uv_blade_info)
 673                return;
 674
 675        uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
 676
 677        if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
 678                set_x2apic_extra_bits(uv_hub_info->pnode);
 679}
 680
 681/*
 682 * When NMI is received, print a stack trace.
 683 */
 684int uv_handle_nmi(unsigned int reason, struct pt_regs *regs)
 685{
 686        unsigned long real_uv_nmi;
 687        int bid;
 688
 689        /*
 690         * Each blade has an MMR that indicates when an NMI has been sent
 691         * to cpus on the blade. If an NMI is detected, atomically
 692         * clear the MMR and update a per-blade NMI count used to
 693         * cause each cpu on the blade to notice a new NMI.
 694         */
 695        bid = uv_numa_blade_id();
 696        real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
 697
 698        if (unlikely(real_uv_nmi)) {
 699                spin_lock(&uv_blade_info[bid].nmi_lock);
 700                real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
 701                if (real_uv_nmi) {
 702                        uv_blade_info[bid].nmi_count++;
 703                        uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
 704                }
 705                spin_unlock(&uv_blade_info[bid].nmi_lock);
 706        }
 707
 708        if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
 709                return NMI_DONE;
 710
 711        __get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
 712
 713        /*
 714         * Use a lock so only one cpu prints at a time.
 715         * This prevents intermixed output.
 716         */
 717        spin_lock(&uv_nmi_lock);
 718        pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
 719        dump_stack();
 720        spin_unlock(&uv_nmi_lock);
 721
 722        return NMI_HANDLED;
 723}
 724
 725void uv_register_nmi_notifier(void)
 726{
 727        if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv"))
 728                printk(KERN_WARNING "UV NMI handler failed to register\n");
 729}
 730
 731void uv_nmi_init(void)
 732{
 733        unsigned int value;
 734
 735        /*
 736         * Unmask NMI on all cpus
 737         */
 738        value = apic_read(APIC_LVT1) | APIC_DM_NMI;
 739        value &= ~APIC_LVT_MASKED;
 740        apic_write(APIC_LVT1, value);
 741}
 742
 743void __init uv_system_init(void)
 744{
 745        union uvh_rh_gam_config_mmr_u  m_n_config;
 746        union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
 747        union uvh_node_id_u node_id;
 748        unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
 749        int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
 750        int gnode_extra, max_pnode = 0;
 751        unsigned long mmr_base, present, paddr;
 752        unsigned short pnode_mask, pnode_io_mask;
 753
 754        printk(KERN_INFO "UV: Found %s hub\n", is_uv1_hub() ? "UV1" : "UV2");
 755        map_low_mmrs();
 756
 757        m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
 758        m_val = m_n_config.s.m_skt;
 759        n_val = m_n_config.s.n_skt;
 760        mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
 761        n_io = is_uv1_hub() ? mmioh.s1.n_io : mmioh.s2.n_io;
 762        mmr_base =
 763            uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
 764            ~UV_MMR_ENABLE;
 765        pnode_mask = (1 << n_val) - 1;
 766        pnode_io_mask = (1 << n_io) - 1;
 767
 768        node_id.v = uv_read_local_mmr(UVH_NODE_ID);
 769        gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
 770        gnode_upper = ((unsigned long)gnode_extra  << m_val);
 771        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",
 772                        n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
 773
 774        printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
 775
 776        for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
 777                uv_possible_blades +=
 778                  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
 779
 780        /* uv_num_possible_blades() is really the hub count */
 781        printk(KERN_INFO "UV: Found %d blades, %d hubs\n",
 782                        is_uv1_hub() ? uv_num_possible_blades() :
 783                        (uv_num_possible_blades() + 1) / 2,
 784                        uv_num_possible_blades());
 785
 786        bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
 787        uv_blade_info = kzalloc(bytes, GFP_KERNEL);
 788        BUG_ON(!uv_blade_info);
 789
 790        for (blade = 0; blade < uv_num_possible_blades(); blade++)
 791                uv_blade_info[blade].memory_nid = -1;
 792
 793        get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
 794
 795        bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
 796        uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
 797        BUG_ON(!uv_node_to_blade);
 798        memset(uv_node_to_blade, 255, bytes);
 799
 800        bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
 801        uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
 802        BUG_ON(!uv_cpu_to_blade);
 803        memset(uv_cpu_to_blade, 255, bytes);
 804
 805        blade = 0;
 806        for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
 807                present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
 808                for (j = 0; j < 64; j++) {
 809                        if (!test_bit(j, &present))
 810                                continue;
 811                        pnode = (i * 64 + j) & pnode_mask;
 812                        uv_blade_info[blade].pnode = pnode;
 813                        uv_blade_info[blade].nr_possible_cpus = 0;
 814                        uv_blade_info[blade].nr_online_cpus = 0;
 815                        spin_lock_init(&uv_blade_info[blade].nmi_lock);
 816                        max_pnode = max(pnode, max_pnode);
 817                        blade++;
 818                }
 819        }
 820
 821        uv_bios_init();
 822        uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
 823                            &sn_region_size, &system_serial_number);
 824        uv_rtc_init();
 825
 826        for_each_present_cpu(cpu) {
 827                int apicid = per_cpu(x86_cpu_to_apicid, cpu);
 828
 829                nid = cpu_to_node(cpu);
 830                /*
 831                 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
 832                 */
 833                uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
 834                uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
 835                uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
 836
 837                uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
 838                uv_cpu_hub_info(cpu)->n_lshift = is_uv2_1_hub() ?
 839                                (m_val == 40 ? 40 : 39) : m_val;
 840
 841                pnode = uv_apicid_to_pnode(apicid);
 842                blade = boot_pnode_to_blade(pnode);
 843                lcpu = uv_blade_info[blade].nr_possible_cpus;
 844                uv_blade_info[blade].nr_possible_cpus++;
 845
 846                /* Any node on the blade, else will contain -1. */
 847                uv_blade_info[blade].memory_nid = nid;
 848
 849                uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
 850                uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
 851                uv_cpu_hub_info(cpu)->m_val = m_val;
 852                uv_cpu_hub_info(cpu)->n_val = n_val;
 853                uv_cpu_hub_info(cpu)->numa_blade_id = blade;
 854                uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
 855                uv_cpu_hub_info(cpu)->pnode = pnode;
 856                uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
 857                uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
 858                uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
 859                uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
 860                uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
 861                uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
 862                uv_node_to_blade[nid] = blade;
 863                uv_cpu_to_blade[cpu] = blade;
 864        }
 865
 866        /* Add blade/pnode info for nodes without cpus */
 867        for_each_online_node(nid) {
 868                if (uv_node_to_blade[nid] >= 0)
 869                        continue;
 870                paddr = node_start_pfn(nid) << PAGE_SHIFT;
 871                pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
 872                blade = boot_pnode_to_blade(pnode);
 873                uv_node_to_blade[nid] = blade;
 874        }
 875
 876        map_gru_high(max_pnode);
 877        map_mmr_high(max_pnode);
 878        map_mmioh_high(max_pnode & pnode_io_mask);
 879
 880        uv_cpu_init();
 881        uv_scir_register_cpu_notifier();
 882        uv_register_nmi_notifier();
 883        proc_mkdir("sgi_uv", NULL);
 884
 885        /* register Legacy VGA I/O redirection handler */
 886        pci_register_set_vga_state(uv_set_vga_state);
 887
 888        /*
 889         * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
 890         * EFI is not enabled in the kdump kernel.
 891         */
 892        if (is_kdump_kernel())
 893                reboot_type = BOOT_ACPI;
 894}
 895
 896apic_driver(apic_x2apic_uv_x);
 897