linux/arch/x86/platform/uv/uv_nmi.c
<<
>>
Prefs
   1/*
   2 * SGI NMI support routines
   3 *
   4 *  This program is free software; you can redistribute it and/or modify
   5 *  it under the terms of the GNU General Public License as published by
   6 *  the Free Software Foundation; either version 2 of the License, or
   7 *  (at your option) any later version.
   8 *
   9 *  This program is distributed in the hope that it will be useful,
  10 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 *  GNU General Public License for more details.
  13 *
  14 *  You should have received a copy of the GNU General Public License
  15 *  along with this program; if not, write to the Free Software
  16 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  17 *
  18 *  Copyright (c) 2009-2013 Silicon Graphics, Inc.  All Rights Reserved.
  19 *  Copyright (c) Mike Travis
  20 */
  21
  22#include <linux/cpu.h>
  23#include <linux/delay.h>
  24#include <linux/kdb.h>
  25#include <linux/kexec.h>
  26#include <linux/kgdb.h>
  27#include <linux/module.h>
  28#include <linux/nmi.h>
  29#include <linux/sched.h>
  30#include <linux/slab.h>
  31#include <linux/clocksource.h>
  32
  33#include <asm/apic.h>
  34#include <asm/current.h>
  35#include <asm/kdebug.h>
  36#include <asm/local64.h>
  37#include <asm/nmi.h>
  38#include <asm/traps.h>
  39#include <asm/uv/uv.h>
  40#include <asm/uv/uv_hub.h>
  41#include <asm/uv/uv_mmrs.h>
  42
  43/*
  44 * UV handler for NMI
  45 *
  46 * Handle system-wide NMI events generated by the global 'power nmi' command.
  47 *
  48 * Basic operation is to field the NMI interrupt on each cpu and wait
  49 * until all cpus have arrived into the nmi handler.  If some cpus do not
  50 * make it into the handler, try and force them in with the IPI(NMI) signal.
  51 *
  52 * We also have to lessen UV Hub MMR accesses as much as possible as this
  53 * disrupts the UV Hub's primary mission of directing NumaLink traffic and
  54 * can cause system problems to occur.
  55 *
  56 * To do this we register our primary NMI notifier on the NMI_UNKNOWN
  57 * chain.  This reduces the number of false NMI calls when the perf
  58 * tools are running which generate an enormous number of NMIs per
  59 * second (~4M/s for 1024 cpu threads).  Our secondary NMI handler is
  60 * very short as it only checks that if it has been "pinged" with the
  61 * IPI(NMI) signal as mentioned above, and does not read the UV Hub's MMR.
  62 *
  63 */
  64
  65static struct uv_hub_nmi_s **uv_hub_nmi_list;
  66
  67DEFINE_PER_CPU(struct uv_cpu_nmi_s, uv_cpu_nmi);
  68EXPORT_PER_CPU_SYMBOL_GPL(uv_cpu_nmi);
  69
  70static unsigned long nmi_mmr;
  71static unsigned long nmi_mmr_clear;
  72static unsigned long nmi_mmr_pending;
  73
  74static atomic_t uv_in_nmi;
  75static atomic_t uv_nmi_cpu = ATOMIC_INIT(-1);
  76static atomic_t uv_nmi_cpus_in_nmi = ATOMIC_INIT(-1);
  77static atomic_t uv_nmi_slave_continue;
  78static cpumask_var_t uv_nmi_cpu_mask;
  79
  80/* Values for uv_nmi_slave_continue */
  81#define SLAVE_CLEAR     0
  82#define SLAVE_CONTINUE  1
  83#define SLAVE_EXIT      2
  84
  85/*
  86 * Default is all stack dumps go to the console and buffer.
  87 * Lower level to send to log buffer only.
  88 */
  89static int uv_nmi_loglevel = CONSOLE_LOGLEVEL_DEFAULT;
  90module_param_named(dump_loglevel, uv_nmi_loglevel, int, 0644);
  91
  92/*
  93 * The following values show statistics on how perf events are affecting
  94 * this system.
  95 */
  96static int param_get_local64(char *buffer, const struct kernel_param *kp)
  97{
  98        return sprintf(buffer, "%lu\n", local64_read((local64_t *)kp->arg));
  99}
 100
 101static int param_set_local64(const char *val, const struct kernel_param *kp)
 102{
 103        /* clear on any write */
 104        local64_set((local64_t *)kp->arg, 0);
 105        return 0;
 106}
 107
 108static const struct kernel_param_ops param_ops_local64 = {
 109        .get = param_get_local64,
 110        .set = param_set_local64,
 111};
 112#define param_check_local64(name, p) __param_check(name, p, local64_t)
 113
 114static local64_t uv_nmi_count;
 115module_param_named(nmi_count, uv_nmi_count, local64, 0644);
 116
 117static local64_t uv_nmi_misses;
 118module_param_named(nmi_misses, uv_nmi_misses, local64, 0644);
 119
 120static local64_t uv_nmi_ping_count;
 121module_param_named(ping_count, uv_nmi_ping_count, local64, 0644);
 122
 123static local64_t uv_nmi_ping_misses;
 124module_param_named(ping_misses, uv_nmi_ping_misses, local64, 0644);
 125
 126/*
 127 * Following values allow tuning for large systems under heavy loading
 128 */
 129static int uv_nmi_initial_delay = 100;
 130module_param_named(initial_delay, uv_nmi_initial_delay, int, 0644);
 131
 132static int uv_nmi_slave_delay = 100;
 133module_param_named(slave_delay, uv_nmi_slave_delay, int, 0644);
 134
 135static int uv_nmi_loop_delay = 100;
 136module_param_named(loop_delay, uv_nmi_loop_delay, int, 0644);
 137
 138static int uv_nmi_trigger_delay = 10000;
 139module_param_named(trigger_delay, uv_nmi_trigger_delay, int, 0644);
 140
 141static int uv_nmi_wait_count = 100;
 142module_param_named(wait_count, uv_nmi_wait_count, int, 0644);
 143
 144static int uv_nmi_retry_count = 500;
 145module_param_named(retry_count, uv_nmi_retry_count, int, 0644);
 146
 147/*
 148 * Valid NMI Actions:
 149 *  "dump"      - dump process stack for each cpu
 150 *  "ips"       - dump IP info for each cpu
 151 *  "kdump"     - do crash dump
 152 *  "kdb"       - enter KDB (default)
 153 *  "kgdb"      - enter KGDB
 154 */
 155static char uv_nmi_action[8] = "kdb";
 156module_param_string(action, uv_nmi_action, sizeof(uv_nmi_action), 0644);
 157
 158static inline bool uv_nmi_action_is(const char *action)
 159{
 160        return (strncmp(uv_nmi_action, action, strlen(action)) == 0);
 161}
 162
 163/* Setup which NMI support is present in system */
 164static void uv_nmi_setup_mmrs(void)
 165{
 166        if (uv_read_local_mmr(UVH_NMI_MMRX_SUPPORTED)) {
 167                uv_write_local_mmr(UVH_NMI_MMRX_REQ,
 168                                        1UL << UVH_NMI_MMRX_REQ_SHIFT);
 169                nmi_mmr = UVH_NMI_MMRX;
 170                nmi_mmr_clear = UVH_NMI_MMRX_CLEAR;
 171                nmi_mmr_pending = 1UL << UVH_NMI_MMRX_SHIFT;
 172                pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMRX_TYPE);
 173        } else {
 174                nmi_mmr = UVH_NMI_MMR;
 175                nmi_mmr_clear = UVH_NMI_MMR_CLEAR;
 176                nmi_mmr_pending = 1UL << UVH_NMI_MMR_SHIFT;
 177                pr_info("UV: SMI NMI support: %s\n", UVH_NMI_MMR_TYPE);
 178        }
 179}
 180
 181/* Read NMI MMR and check if NMI flag was set by BMC. */
 182static inline int uv_nmi_test_mmr(struct uv_hub_nmi_s *hub_nmi)
 183{
 184        hub_nmi->nmi_value = uv_read_local_mmr(nmi_mmr);
 185        atomic_inc(&hub_nmi->read_mmr_count);
 186        return !!(hub_nmi->nmi_value & nmi_mmr_pending);
 187}
 188
 189static inline void uv_local_mmr_clear_nmi(void)
 190{
 191        uv_write_local_mmr(nmi_mmr_clear, nmi_mmr_pending);
 192}
 193
 194/*
 195 * If first cpu in on this hub, set hub_nmi "in_nmi" and "owner" values and
 196 * return true.  If first cpu in on the system, set global "in_nmi" flag.
 197 */
 198static int uv_set_in_nmi(int cpu, struct uv_hub_nmi_s *hub_nmi)
 199{
 200        int first = atomic_add_unless(&hub_nmi->in_nmi, 1, 1);
 201
 202        if (first) {
 203                atomic_set(&hub_nmi->cpu_owner, cpu);
 204                if (atomic_add_unless(&uv_in_nmi, 1, 1))
 205                        atomic_set(&uv_nmi_cpu, cpu);
 206
 207                atomic_inc(&hub_nmi->nmi_count);
 208        }
 209        return first;
 210}
 211
 212/* Check if this is a system NMI event */
 213static int uv_check_nmi(struct uv_hub_nmi_s *hub_nmi)
 214{
 215        int cpu = smp_processor_id();
 216        int nmi = 0;
 217
 218        local64_inc(&uv_nmi_count);
 219        this_cpu_inc(uv_cpu_nmi.queries);
 220
 221        do {
 222                nmi = atomic_read(&hub_nmi->in_nmi);
 223                if (nmi)
 224                        break;
 225
 226                if (raw_spin_trylock(&hub_nmi->nmi_lock)) {
 227
 228                        /* check hub MMR NMI flag */
 229                        if (uv_nmi_test_mmr(hub_nmi)) {
 230                                uv_set_in_nmi(cpu, hub_nmi);
 231                                nmi = 1;
 232                                break;
 233                        }
 234
 235                        /* MMR NMI flag is clear */
 236                        raw_spin_unlock(&hub_nmi->nmi_lock);
 237
 238                } else {
 239                        /* wait a moment for the hub nmi locker to set flag */
 240                        cpu_relax();
 241                        udelay(uv_nmi_slave_delay);
 242
 243                        /* re-check hub in_nmi flag */
 244                        nmi = atomic_read(&hub_nmi->in_nmi);
 245                        if (nmi)
 246                                break;
 247                }
 248
 249                /* check if this BMC missed setting the MMR NMI flag */
 250                if (!nmi) {
 251                        nmi = atomic_read(&uv_in_nmi);
 252                        if (nmi)
 253                                uv_set_in_nmi(cpu, hub_nmi);
 254                }
 255
 256        } while (0);
 257
 258        if (!nmi)
 259                local64_inc(&uv_nmi_misses);
 260
 261        return nmi;
 262}
 263
 264/* Need to reset the NMI MMR register, but only once per hub. */
 265static inline void uv_clear_nmi(int cpu)
 266{
 267        struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi;
 268
 269        if (cpu == atomic_read(&hub_nmi->cpu_owner)) {
 270                atomic_set(&hub_nmi->cpu_owner, -1);
 271                atomic_set(&hub_nmi->in_nmi, 0);
 272                uv_local_mmr_clear_nmi();
 273                raw_spin_unlock(&hub_nmi->nmi_lock);
 274        }
 275}
 276
 277/* Ping non-responding cpus attemping to force them into the NMI handler */
 278static void uv_nmi_nr_cpus_ping(void)
 279{
 280        int cpu;
 281
 282        for_each_cpu(cpu, uv_nmi_cpu_mask)
 283                uv_cpu_nmi_per(cpu).pinging = 1;
 284
 285        apic->send_IPI_mask(uv_nmi_cpu_mask, APIC_DM_NMI);
 286}
 287
 288/* Clean up flags for cpus that ignored both NMI and ping */
 289static void uv_nmi_cleanup_mask(void)
 290{
 291        int cpu;
 292
 293        for_each_cpu(cpu, uv_nmi_cpu_mask) {
 294                uv_cpu_nmi_per(cpu).pinging =  0;
 295                uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_OUT;
 296                cpumask_clear_cpu(cpu, uv_nmi_cpu_mask);
 297        }
 298}
 299
 300/* Loop waiting as cpus enter nmi handler */
 301static int uv_nmi_wait_cpus(int first)
 302{
 303        int i, j, k, n = num_online_cpus();
 304        int last_k = 0, waiting = 0;
 305
 306        if (first) {
 307                cpumask_copy(uv_nmi_cpu_mask, cpu_online_mask);
 308                k = 0;
 309        } else {
 310                k = n - cpumask_weight(uv_nmi_cpu_mask);
 311        }
 312
 313        udelay(uv_nmi_initial_delay);
 314        for (i = 0; i < uv_nmi_retry_count; i++) {
 315                int loop_delay = uv_nmi_loop_delay;
 316
 317                for_each_cpu(j, uv_nmi_cpu_mask) {
 318                        if (uv_cpu_nmi_per(j).state) {
 319                                cpumask_clear_cpu(j, uv_nmi_cpu_mask);
 320                                if (++k >= n)
 321                                        break;
 322                        }
 323                }
 324                if (k >= n) {           /* all in? */
 325                        k = n;
 326                        break;
 327                }
 328                if (last_k != k) {      /* abort if no new cpus coming in */
 329                        last_k = k;
 330                        waiting = 0;
 331                } else if (++waiting > uv_nmi_wait_count)
 332                        break;
 333
 334                /* extend delay if waiting only for cpu 0 */
 335                if (waiting && (n - k) == 1 &&
 336                    cpumask_test_cpu(0, uv_nmi_cpu_mask))
 337                        loop_delay *= 100;
 338
 339                udelay(loop_delay);
 340        }
 341        atomic_set(&uv_nmi_cpus_in_nmi, k);
 342        return n - k;
 343}
 344
 345/* Wait until all slave cpus have entered UV NMI handler */
 346static void uv_nmi_wait(int master)
 347{
 348        /* indicate this cpu is in */
 349        this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_IN);
 350
 351        /* if not the first cpu in (the master), then we are a slave cpu */
 352        if (!master)
 353                return;
 354
 355        do {
 356                /* wait for all other cpus to gather here */
 357                if (!uv_nmi_wait_cpus(1))
 358                        break;
 359
 360                /* if not all made it in, send IPI NMI to them */
 361                pr_alert("UV: Sending NMI IPI to %d non-responding CPUs: %*pbl\n",
 362                         cpumask_weight(uv_nmi_cpu_mask),
 363                         cpumask_pr_args(uv_nmi_cpu_mask));
 364
 365                uv_nmi_nr_cpus_ping();
 366
 367                /* if all cpus are in, then done */
 368                if (!uv_nmi_wait_cpus(0))
 369                        break;
 370
 371                pr_alert("UV: %d CPUs not in NMI loop: %*pbl\n",
 372                         cpumask_weight(uv_nmi_cpu_mask),
 373                         cpumask_pr_args(uv_nmi_cpu_mask));
 374        } while (0);
 375
 376        pr_alert("UV: %d of %d CPUs in NMI\n",
 377                atomic_read(&uv_nmi_cpus_in_nmi), num_online_cpus());
 378}
 379
 380/* Dump Instruction Pointer header */
 381static void uv_nmi_dump_cpu_ip_hdr(void)
 382{
 383        pr_info("\nUV: %4s %6s %-32s %s   (Note: PID 0 not listed)\n",
 384                "CPU", "PID", "COMMAND", "IP");
 385}
 386
 387/* Dump Instruction Pointer info */
 388static void uv_nmi_dump_cpu_ip(int cpu, struct pt_regs *regs)
 389{
 390        pr_info("UV: %4d %6d %-32.32s ", cpu, current->pid, current->comm);
 391        printk_address(regs->ip);
 392}
 393
 394/*
 395 * Dump this CPU's state.  If action was set to "kdump" and the crash_kexec
 396 * failed, then we provide "dump" as an alternate action.  Action "dump" now
 397 * also includes the show "ips" (instruction pointers) action whereas the
 398 * action "ips" only displays instruction pointers for the non-idle CPU's.
 399 * This is an abbreviated form of the "ps" command.
 400 */
 401static void uv_nmi_dump_state_cpu(int cpu, struct pt_regs *regs)
 402{
 403        const char *dots = " ................................. ";
 404
 405        if (cpu == 0)
 406                uv_nmi_dump_cpu_ip_hdr();
 407
 408        if (current->pid != 0 || !uv_nmi_action_is("ips"))
 409                uv_nmi_dump_cpu_ip(cpu, regs);
 410
 411        if (uv_nmi_action_is("dump")) {
 412                pr_info("UV:%sNMI process trace for CPU %d\n", dots, cpu);
 413                show_regs(regs);
 414        }
 415
 416        this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_DUMP_DONE);
 417}
 418
 419/* Trigger a slave cpu to dump it's state */
 420static void uv_nmi_trigger_dump(int cpu)
 421{
 422        int retry = uv_nmi_trigger_delay;
 423
 424        if (uv_cpu_nmi_per(cpu).state != UV_NMI_STATE_IN)
 425                return;
 426
 427        uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_DUMP;
 428        do {
 429                cpu_relax();
 430                udelay(10);
 431                if (uv_cpu_nmi_per(cpu).state
 432                                != UV_NMI_STATE_DUMP)
 433                        return;
 434        } while (--retry > 0);
 435
 436        pr_crit("UV: CPU %d stuck in process dump function\n", cpu);
 437        uv_cpu_nmi_per(cpu).state = UV_NMI_STATE_DUMP_DONE;
 438}
 439
 440/* Wait until all cpus ready to exit */
 441static void uv_nmi_sync_exit(int master)
 442{
 443        atomic_dec(&uv_nmi_cpus_in_nmi);
 444        if (master) {
 445                while (atomic_read(&uv_nmi_cpus_in_nmi) > 0)
 446                        cpu_relax();
 447                atomic_set(&uv_nmi_slave_continue, SLAVE_CLEAR);
 448        } else {
 449                while (atomic_read(&uv_nmi_slave_continue))
 450                        cpu_relax();
 451        }
 452}
 453
 454/* Walk through cpu list and dump state of each */
 455static void uv_nmi_dump_state(int cpu, struct pt_regs *regs, int master)
 456{
 457        if (master) {
 458                int tcpu;
 459                int ignored = 0;
 460                int saved_console_loglevel = console_loglevel;
 461
 462                pr_alert("UV: tracing %s for %d CPUs from CPU %d\n",
 463                        uv_nmi_action_is("ips") ? "IPs" : "processes",
 464                        atomic_read(&uv_nmi_cpus_in_nmi), cpu);
 465
 466                console_loglevel = uv_nmi_loglevel;
 467                atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT);
 468                for_each_online_cpu(tcpu) {
 469                        if (cpumask_test_cpu(tcpu, uv_nmi_cpu_mask))
 470                                ignored++;
 471                        else if (tcpu == cpu)
 472                                uv_nmi_dump_state_cpu(tcpu, regs);
 473                        else
 474                                uv_nmi_trigger_dump(tcpu);
 475                }
 476                if (ignored)
 477                        pr_alert("UV: %d CPUs ignored NMI\n", ignored);
 478
 479                console_loglevel = saved_console_loglevel;
 480                pr_alert("UV: process trace complete\n");
 481        } else {
 482                while (!atomic_read(&uv_nmi_slave_continue))
 483                        cpu_relax();
 484                while (this_cpu_read(uv_cpu_nmi.state) != UV_NMI_STATE_DUMP)
 485                        cpu_relax();
 486                uv_nmi_dump_state_cpu(cpu, regs);
 487        }
 488        uv_nmi_sync_exit(master);
 489}
 490
 491static void uv_nmi_touch_watchdogs(void)
 492{
 493        touch_softlockup_watchdog_sync();
 494        clocksource_touch_watchdog();
 495        rcu_cpu_stall_reset();
 496        touch_nmi_watchdog();
 497}
 498
 499static atomic_t uv_nmi_kexec_failed;
 500
 501#if defined(CONFIG_KEXEC_CORE)
 502static void uv_nmi_kdump(int cpu, int master, struct pt_regs *regs)
 503{
 504        /* Call crash to dump system state */
 505        if (master) {
 506                pr_emerg("UV: NMI executing crash_kexec on CPU%d\n", cpu);
 507                crash_kexec(regs);
 508
 509                pr_emerg("UV: crash_kexec unexpectedly returned, ");
 510                atomic_set(&uv_nmi_kexec_failed, 1);
 511                if (!kexec_crash_image) {
 512                        pr_cont("crash kernel not loaded\n");
 513                        return;
 514                }
 515                pr_cont("kexec busy, stalling cpus while waiting\n");
 516        }
 517
 518        /* If crash exec fails the slaves should return, otherwise stall */
 519        while (atomic_read(&uv_nmi_kexec_failed) == 0)
 520                mdelay(10);
 521}
 522
 523#else /* !CONFIG_KEXEC_CORE */
 524static inline void uv_nmi_kdump(int cpu, int master, struct pt_regs *regs)
 525{
 526        if (master)
 527                pr_err("UV: NMI kdump: KEXEC not supported in this kernel\n");
 528        atomic_set(&uv_nmi_kexec_failed, 1);
 529}
 530#endif /* !CONFIG_KEXEC_CORE */
 531
 532#ifdef CONFIG_KGDB
 533#ifdef CONFIG_KGDB_KDB
 534static inline int uv_nmi_kdb_reason(void)
 535{
 536        return KDB_REASON_SYSTEM_NMI;
 537}
 538#else /* !CONFIG_KGDB_KDB */
 539static inline int uv_nmi_kdb_reason(void)
 540{
 541        /* Insure user is expecting to attach gdb remote */
 542        if (uv_nmi_action_is("kgdb"))
 543                return 0;
 544
 545        pr_err("UV: NMI error: KDB is not enabled in this kernel\n");
 546        return -1;
 547}
 548#endif /* CONFIG_KGDB_KDB */
 549
 550/*
 551 * Call KGDB/KDB from NMI handler
 552 *
 553 * Note that if both KGDB and KDB are configured, then the action of 'kgdb' or
 554 * 'kdb' has no affect on which is used.  See the KGDB documention for further
 555 * information.
 556 */
 557static void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master)
 558{
 559        if (master) {
 560                int reason = uv_nmi_kdb_reason();
 561                int ret;
 562
 563                if (reason < 0)
 564                        return;
 565
 566                /* call KGDB NMI handler as MASTER */
 567                ret = kgdb_nmicallin(cpu, X86_TRAP_NMI, regs, reason,
 568                                &uv_nmi_slave_continue);
 569                if (ret) {
 570                        pr_alert("KGDB returned error, is kgdboc set?\n");
 571                        atomic_set(&uv_nmi_slave_continue, SLAVE_EXIT);
 572                }
 573        } else {
 574                /* wait for KGDB signal that it's ready for slaves to enter */
 575                int sig;
 576
 577                do {
 578                        cpu_relax();
 579                        sig = atomic_read(&uv_nmi_slave_continue);
 580                } while (!sig);
 581
 582                /* call KGDB as slave */
 583                if (sig == SLAVE_CONTINUE)
 584                        kgdb_nmicallback(cpu, regs);
 585        }
 586        uv_nmi_sync_exit(master);
 587}
 588
 589#else /* !CONFIG_KGDB */
 590static inline void uv_call_kgdb_kdb(int cpu, struct pt_regs *regs, int master)
 591{
 592        pr_err("UV: NMI error: KGDB is not enabled in this kernel\n");
 593}
 594#endif /* !CONFIG_KGDB */
 595
 596/*
 597 * UV NMI handler
 598 */
 599int uv_handle_nmi(unsigned int reason, struct pt_regs *regs)
 600{
 601        struct uv_hub_nmi_s *hub_nmi = uv_hub_nmi;
 602        int cpu = smp_processor_id();
 603        int master = 0;
 604        unsigned long flags;
 605
 606        local_irq_save(flags);
 607
 608        /* If not a UV System NMI, ignore */
 609        if (!this_cpu_read(uv_cpu_nmi.pinging) && !uv_check_nmi(hub_nmi)) {
 610                local_irq_restore(flags);
 611                return NMI_DONE;
 612        }
 613
 614        /* Indicate we are the first CPU into the NMI handler */
 615        master = (atomic_read(&uv_nmi_cpu) == cpu);
 616
 617        /* If NMI action is "kdump", then attempt to do it */
 618        if (uv_nmi_action_is("kdump")) {
 619                uv_nmi_kdump(cpu, master, regs);
 620
 621                /* Unexpected return, revert action to "dump" */
 622                if (master)
 623                        strncpy(uv_nmi_action, "dump", strlen(uv_nmi_action));
 624        }
 625
 626        /* Pause as all cpus enter the NMI handler */
 627        uv_nmi_wait(master);
 628
 629        /* Dump state of each cpu */
 630        if (uv_nmi_action_is("ips") || uv_nmi_action_is("dump"))
 631                uv_nmi_dump_state(cpu, regs, master);
 632
 633        /* Call KGDB/KDB if enabled */
 634        else if (uv_nmi_action_is("kdb") || uv_nmi_action_is("kgdb"))
 635                uv_call_kgdb_kdb(cpu, regs, master);
 636
 637        /* Clear per_cpu "in nmi" flag */
 638        this_cpu_write(uv_cpu_nmi.state, UV_NMI_STATE_OUT);
 639
 640        /* Clear MMR NMI flag on each hub */
 641        uv_clear_nmi(cpu);
 642
 643        /* Clear global flags */
 644        if (master) {
 645                if (cpumask_weight(uv_nmi_cpu_mask))
 646                        uv_nmi_cleanup_mask();
 647                atomic_set(&uv_nmi_cpus_in_nmi, -1);
 648                atomic_set(&uv_nmi_cpu, -1);
 649                atomic_set(&uv_in_nmi, 0);
 650                atomic_set(&uv_nmi_kexec_failed, 0);
 651        }
 652
 653        uv_nmi_touch_watchdogs();
 654        local_irq_restore(flags);
 655
 656        return NMI_HANDLED;
 657}
 658
 659/*
 660 * NMI handler for pulling in CPUs when perf events are grabbing our NMI
 661 */
 662static int uv_handle_nmi_ping(unsigned int reason, struct pt_regs *regs)
 663{
 664        int ret;
 665
 666        this_cpu_inc(uv_cpu_nmi.queries);
 667        if (!this_cpu_read(uv_cpu_nmi.pinging)) {
 668                local64_inc(&uv_nmi_ping_misses);
 669                return NMI_DONE;
 670        }
 671
 672        this_cpu_inc(uv_cpu_nmi.pings);
 673        local64_inc(&uv_nmi_ping_count);
 674        ret = uv_handle_nmi(reason, regs);
 675        this_cpu_write(uv_cpu_nmi.pinging, 0);
 676        return ret;
 677}
 678
 679static void uv_register_nmi_notifier(void)
 680{
 681        if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv"))
 682                pr_warn("UV: NMI handler failed to register\n");
 683
 684        if (register_nmi_handler(NMI_LOCAL, uv_handle_nmi_ping, 0, "uvping"))
 685                pr_warn("UV: PING NMI handler failed to register\n");
 686}
 687
 688void uv_nmi_init(void)
 689{
 690        unsigned int value;
 691
 692        /*
 693         * Unmask NMI on all cpus
 694         */
 695        value = apic_read(APIC_LVT1) | APIC_DM_NMI;
 696        value &= ~APIC_LVT_MASKED;
 697        apic_write(APIC_LVT1, value);
 698}
 699
 700void uv_nmi_setup(void)
 701{
 702        int size = sizeof(void *) * (1 << NODES_SHIFT);
 703        int cpu, nid;
 704
 705        /* Setup hub nmi info */
 706        uv_nmi_setup_mmrs();
 707        uv_hub_nmi_list = kzalloc(size, GFP_KERNEL);
 708        pr_info("UV: NMI hub list @ 0x%p (%d)\n", uv_hub_nmi_list, size);
 709        BUG_ON(!uv_hub_nmi_list);
 710        size = sizeof(struct uv_hub_nmi_s);
 711        for_each_present_cpu(cpu) {
 712                nid = cpu_to_node(cpu);
 713                if (uv_hub_nmi_list[nid] == NULL) {
 714                        uv_hub_nmi_list[nid] = kzalloc_node(size,
 715                                                            GFP_KERNEL, nid);
 716                        BUG_ON(!uv_hub_nmi_list[nid]);
 717                        raw_spin_lock_init(&(uv_hub_nmi_list[nid]->nmi_lock));
 718                        atomic_set(&uv_hub_nmi_list[nid]->cpu_owner, -1);
 719                }
 720                uv_hub_nmi_per(cpu) = uv_hub_nmi_list[nid];
 721        }
 722        BUG_ON(!alloc_cpumask_var(&uv_nmi_cpu_mask, GFP_KERNEL));
 723        uv_register_nmi_notifier();
 724}
 725