linux/arch/arm64/kernel/smp.c
<<
>>
Prefs
   1/*
   2 * SMP initialisation and IPI support
   3 * Based on arch/arm/kernel/smp.c
   4 *
   5 * Copyright (C) 2012 ARM Ltd.
   6 *
   7 * This program is free software; you can redistribute it and/or modify
   8 * it under the terms of the GNU General Public License version 2 as
   9 * published by the Free Software Foundation.
  10 *
  11 * This program is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  18 */
  19
  20#include <linux/acpi.h>
  21#include <linux/arm_sdei.h>
  22#include <linux/delay.h>
  23#include <linux/init.h>
  24#include <linux/spinlock.h>
  25#include <linux/sched/mm.h>
  26#include <linux/sched/hotplug.h>
  27#include <linux/sched/task_stack.h>
  28#include <linux/interrupt.h>
  29#include <linux/cache.h>
  30#include <linux/profile.h>
  31#include <linux/errno.h>
  32#include <linux/mm.h>
  33#include <linux/err.h>
  34#include <linux/cpu.h>
  35#include <linux/smp.h>
  36#include <linux/seq_file.h>
  37#include <linux/irq.h>
  38#include <linux/irqchip/arm-gic-v3.h>
  39#include <linux/percpu.h>
  40#include <linux/clockchips.h>
  41#include <linux/completion.h>
  42#include <linux/of.h>
  43#include <linux/irq_work.h>
  44#include <linux/kexec.h>
  45
  46#include <asm/alternative.h>
  47#include <asm/atomic.h>
  48#include <asm/cacheflush.h>
  49#include <asm/cpu.h>
  50#include <asm/cputype.h>
  51#include <asm/cpu_ops.h>
  52#include <asm/daifflags.h>
  53#include <asm/mmu_context.h>
  54#include <asm/numa.h>
  55#include <asm/pgtable.h>
  56#include <asm/pgalloc.h>
  57#include <asm/processor.h>
  58#include <asm/smp_plat.h>
  59#include <asm/sections.h>
  60#include <asm/tlbflush.h>
  61#include <asm/ptrace.h>
  62#include <asm/virt.h>
  63
  64#define CREATE_TRACE_POINTS
  65#include <trace/events/ipi.h>
  66
  67DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number);
  68EXPORT_PER_CPU_SYMBOL(cpu_number);
  69
  70/*
  71 * as from 2.5, kernels no longer have an init_tasks structure
  72 * so we need some other way of telling a new secondary core
  73 * where to place its SVC stack
  74 */
  75struct secondary_data secondary_data;
  76/* Number of CPUs which aren't online, but looping in kernel text. */
  77int cpus_stuck_in_kernel;
  78
  79enum ipi_msg_type {
  80        IPI_RESCHEDULE,
  81        IPI_CALL_FUNC,
  82        IPI_CPU_STOP,
  83        IPI_CPU_CRASH_STOP,
  84        IPI_TIMER,
  85        IPI_IRQ_WORK,
  86        IPI_WAKEUP
  87};
  88
  89#ifdef CONFIG_HOTPLUG_CPU
  90static int op_cpu_kill(unsigned int cpu);
  91#else
  92static inline int op_cpu_kill(unsigned int cpu)
  93{
  94        return -ENOSYS;
  95}
  96#endif
  97
  98
  99/*
 100 * Boot a secondary CPU, and assign it the specified idle task.
 101 * This also gives us the initial stack to use for this CPU.
 102 */
 103static int boot_secondary(unsigned int cpu, struct task_struct *idle)
 104{
 105        if (cpu_ops[cpu]->cpu_boot)
 106                return cpu_ops[cpu]->cpu_boot(cpu);
 107
 108        return -EOPNOTSUPP;
 109}
 110
 111static DECLARE_COMPLETION(cpu_running);
 112
 113int __cpu_up(unsigned int cpu, struct task_struct *idle)
 114{
 115        int ret;
 116        long status;
 117
 118        /*
 119         * We need to tell the secondary core where to find its stack and the
 120         * page tables.
 121         */
 122        secondary_data.task = idle;
 123        secondary_data.stack = task_stack_page(idle) + THREAD_SIZE;
 124        update_cpu_boot_status(CPU_MMU_OFF);
 125        __flush_dcache_area(&secondary_data, sizeof(secondary_data));
 126
 127        /*
 128         * Now bring the CPU into our world.
 129         */
 130        ret = boot_secondary(cpu, idle);
 131        if (ret == 0) {
 132                /*
 133                 * CPU was successfully started, wait for it to come online or
 134                 * time out.
 135                 */
 136                wait_for_completion_timeout(&cpu_running,
 137                                            msecs_to_jiffies(1000));
 138
 139                if (!cpu_online(cpu)) {
 140                        pr_crit("CPU%u: failed to come online\n", cpu);
 141                        ret = -EIO;
 142                }
 143        } else {
 144                pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
 145                return ret;
 146        }
 147
 148        secondary_data.task = NULL;
 149        secondary_data.stack = NULL;
 150        status = READ_ONCE(secondary_data.status);
 151        if (ret && status) {
 152
 153                if (status == CPU_MMU_OFF)
 154                        status = READ_ONCE(__early_cpu_boot_status);
 155
 156                switch (status & CPU_BOOT_STATUS_MASK) {
 157                default:
 158                        pr_err("CPU%u: failed in unknown state : 0x%lx\n",
 159                                        cpu, status);
 160                        break;
 161                case CPU_KILL_ME:
 162                        if (!op_cpu_kill(cpu)) {
 163                                pr_crit("CPU%u: died during early boot\n", cpu);
 164                                break;
 165                        }
 166                        /* Fall through */
 167                        pr_crit("CPU%u: may not have shut down cleanly\n", cpu);
 168                case CPU_STUCK_IN_KERNEL:
 169                        pr_crit("CPU%u: is stuck in kernel\n", cpu);
 170                        if (status & CPU_STUCK_REASON_52_BIT_VA)
 171                                pr_crit("CPU%u: does not support 52-bit VAs\n", cpu);
 172                        if (status & CPU_STUCK_REASON_NO_GRAN)
 173                                pr_crit("CPU%u: does not support %luK granule \n", cpu, PAGE_SIZE / SZ_1K);
 174                        cpus_stuck_in_kernel++;
 175                        break;
 176                case CPU_PANIC_KERNEL:
 177                        panic("CPU%u detected unsupported configuration\n", cpu);
 178                }
 179        }
 180
 181        return ret;
 182}
 183
 184static void init_gic_priority_masking(void)
 185{
 186        u32 cpuflags;
 187
 188        if (WARN_ON(!gic_enable_sre()))
 189                return;
 190
 191        cpuflags = read_sysreg(daif);
 192
 193        WARN_ON(!(cpuflags & PSR_I_BIT));
 194
 195        gic_write_pmr(GIC_PRIO_IRQOFF);
 196
 197        /* We can only unmask PSR.I if we can take aborts */
 198        if (!(cpuflags & PSR_A_BIT))
 199                write_sysreg(cpuflags & ~PSR_I_BIT, daif);
 200}
 201
 202/*
 203 * This is the secondary CPU boot entry.  We're using this CPUs
 204 * idle thread stack, but a set of temporary page tables.
 205 */
 206asmlinkage notrace void secondary_start_kernel(void)
 207{
 208        u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
 209        struct mm_struct *mm = &init_mm;
 210        unsigned int cpu;
 211
 212        cpu = task_cpu(current);
 213        set_my_cpu_offset(per_cpu_offset(cpu));
 214
 215        /*
 216         * All kernel threads share the same mm context; grab a
 217         * reference and switch to it.
 218         */
 219        mmgrab(mm);
 220        current->active_mm = mm;
 221
 222        /*
 223         * TTBR0 is only used for the identity mapping at this stage. Make it
 224         * point to zero page to avoid speculatively fetching new entries.
 225         */
 226        cpu_uninstall_idmap();
 227
 228        if (system_uses_irq_prio_masking())
 229                init_gic_priority_masking();
 230
 231        preempt_disable();
 232        trace_hardirqs_off();
 233
 234        /*
 235         * If the system has established the capabilities, make sure
 236         * this CPU ticks all of those. If it doesn't, the CPU will
 237         * fail to come online.
 238         */
 239        check_local_cpu_capabilities();
 240
 241        if (cpu_ops[cpu]->cpu_postboot)
 242                cpu_ops[cpu]->cpu_postboot();
 243
 244        /*
 245         * Log the CPU info before it is marked online and might get read.
 246         */
 247        cpuinfo_store_cpu();
 248
 249        /*
 250         * Enable GIC and timers.
 251         */
 252        notify_cpu_starting(cpu);
 253
 254        store_cpu_topology(cpu);
 255        numa_add_cpu(cpu);
 256
 257        /*
 258         * OK, now it's safe to let the boot CPU continue.  Wait for
 259         * the CPU migration code to notice that the CPU is online
 260         * before we continue.
 261         */
 262        pr_info("CPU%u: Booted secondary processor 0x%010lx [0x%08x]\n",
 263                                         cpu, (unsigned long)mpidr,
 264                                         read_cpuid_id());
 265        update_cpu_boot_status(CPU_BOOT_SUCCESS);
 266        set_cpu_online(cpu, true);
 267        complete(&cpu_running);
 268
 269        local_daif_restore(DAIF_PROCCTX);
 270
 271        /*
 272         * OK, it's off to the idle thread for us
 273         */
 274        cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 275}
 276
 277#ifdef CONFIG_HOTPLUG_CPU
 278static int op_cpu_disable(unsigned int cpu)
 279{
 280        /*
 281         * If we don't have a cpu_die method, abort before we reach the point
 282         * of no return. CPU0 may not have an cpu_ops, so test for it.
 283         */
 284        if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die)
 285                return -EOPNOTSUPP;
 286
 287        /*
 288         * We may need to abort a hot unplug for some other mechanism-specific
 289         * reason.
 290         */
 291        if (cpu_ops[cpu]->cpu_disable)
 292                return cpu_ops[cpu]->cpu_disable(cpu);
 293
 294        return 0;
 295}
 296
 297/*
 298 * __cpu_disable runs on the processor to be shutdown.
 299 */
 300int __cpu_disable(void)
 301{
 302        unsigned int cpu = smp_processor_id();
 303        int ret;
 304
 305        ret = op_cpu_disable(cpu);
 306        if (ret)
 307                return ret;
 308
 309        remove_cpu_topology(cpu);
 310        numa_remove_cpu(cpu);
 311
 312        /*
 313         * Take this CPU offline.  Once we clear this, we can't return,
 314         * and we must not schedule until we're ready to give up the cpu.
 315         */
 316        set_cpu_online(cpu, false);
 317
 318        /*
 319         * OK - migrate IRQs away from this CPU
 320         */
 321        irq_migrate_all_off_this_cpu();
 322
 323        return 0;
 324}
 325
 326static int op_cpu_kill(unsigned int cpu)
 327{
 328        /*
 329         * If we have no means of synchronising with the dying CPU, then assume
 330         * that it is really dead. We can only wait for an arbitrary length of
 331         * time and hope that it's dead, so let's skip the wait and just hope.
 332         */
 333        if (!cpu_ops[cpu]->cpu_kill)
 334                return 0;
 335
 336        return cpu_ops[cpu]->cpu_kill(cpu);
 337}
 338
 339/*
 340 * called on the thread which is asking for a CPU to be shutdown -
 341 * waits until shutdown has completed, or it is timed out.
 342 */
 343void __cpu_die(unsigned int cpu)
 344{
 345        int err;
 346
 347        if (!cpu_wait_death(cpu, 5)) {
 348                pr_crit("CPU%u: cpu didn't die\n", cpu);
 349                return;
 350        }
 351        pr_notice("CPU%u: shutdown\n", cpu);
 352
 353        /*
 354         * Now that the dying CPU is beyond the point of no return w.r.t.
 355         * in-kernel synchronisation, try to get the firwmare to help us to
 356         * verify that it has really left the kernel before we consider
 357         * clobbering anything it might still be using.
 358         */
 359        err = op_cpu_kill(cpu);
 360        if (err)
 361                pr_warn("CPU%d may not have shut down cleanly: %d\n",
 362                        cpu, err);
 363}
 364
 365/*
 366 * Called from the idle thread for the CPU which has been shutdown.
 367 *
 368 */
 369void cpu_die(void)
 370{
 371        unsigned int cpu = smp_processor_id();
 372
 373        idle_task_exit();
 374
 375        local_daif_mask();
 376
 377        /* Tell __cpu_die() that this CPU is now safe to dispose of */
 378        (void)cpu_report_death();
 379
 380        /*
 381         * Actually shutdown the CPU. This must never fail. The specific hotplug
 382         * mechanism must perform all required cache maintenance to ensure that
 383         * no dirty lines are lost in the process of shutting down the CPU.
 384         */
 385        cpu_ops[cpu]->cpu_die(cpu);
 386
 387        BUG();
 388}
 389#endif
 390
 391/*
 392 * Kill the calling secondary CPU, early in bringup before it is turned
 393 * online.
 394 */
 395void cpu_die_early(void)
 396{
 397        int cpu = smp_processor_id();
 398
 399        pr_crit("CPU%d: will not boot\n", cpu);
 400
 401        /* Mark this CPU absent */
 402        set_cpu_present(cpu, 0);
 403
 404#ifdef CONFIG_HOTPLUG_CPU
 405        update_cpu_boot_status(CPU_KILL_ME);
 406        /* Check if we can park ourselves */
 407        if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die)
 408                cpu_ops[cpu]->cpu_die(cpu);
 409#endif
 410        update_cpu_boot_status(CPU_STUCK_IN_KERNEL);
 411
 412        cpu_park_loop();
 413}
 414
 415static void __init hyp_mode_check(void)
 416{
 417        if (is_hyp_mode_available())
 418                pr_info("CPU: All CPU(s) started at EL2\n");
 419        else if (is_hyp_mode_mismatched())
 420                WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC,
 421                           "CPU: CPUs started in inconsistent modes");
 422        else
 423                pr_info("CPU: All CPU(s) started at EL1\n");
 424}
 425
 426void __init smp_cpus_done(unsigned int max_cpus)
 427{
 428        pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
 429        setup_cpu_features();
 430        hyp_mode_check();
 431        apply_alternatives_all();
 432        mark_linear_text_alias_ro();
 433}
 434
 435void __init smp_prepare_boot_cpu(void)
 436{
 437        set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
 438        /*
 439         * Initialise the static keys early as they may be enabled by the
 440         * cpufeature code.
 441         */
 442        jump_label_init();
 443        cpuinfo_store_boot_cpu();
 444
 445        /*
 446         * We now know enough about the boot CPU to apply the
 447         * alternatives that cannot wait until interrupt handling
 448         * and/or scheduling is enabled.
 449         */
 450        apply_boot_alternatives();
 451
 452        /* Conditionally switch to GIC PMR for interrupt masking */
 453        if (system_uses_irq_prio_masking())
 454                init_gic_priority_masking();
 455}
 456
 457static u64 __init of_get_cpu_mpidr(struct device_node *dn)
 458{
 459        const __be32 *cell;
 460        u64 hwid;
 461
 462        /*
 463         * A cpu node with missing "reg" property is
 464         * considered invalid to build a cpu_logical_map
 465         * entry.
 466         */
 467        cell = of_get_property(dn, "reg", NULL);
 468        if (!cell) {
 469                pr_err("%pOF: missing reg property\n", dn);
 470                return INVALID_HWID;
 471        }
 472
 473        hwid = of_read_number(cell, of_n_addr_cells(dn));
 474        /*
 475         * Non affinity bits must be set to 0 in the DT
 476         */
 477        if (hwid & ~MPIDR_HWID_BITMASK) {
 478                pr_err("%pOF: invalid reg property\n", dn);
 479                return INVALID_HWID;
 480        }
 481        return hwid;
 482}
 483
 484/*
 485 * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
 486 * entries and check for duplicates. If any is found just ignore the
 487 * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
 488 * matching valid MPIDR values.
 489 */
 490static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid)
 491{
 492        unsigned int i;
 493
 494        for (i = 1; (i < cpu) && (i < NR_CPUS); i++)
 495                if (cpu_logical_map(i) == hwid)
 496                        return true;
 497        return false;
 498}
 499
 500/*
 501 * Initialize cpu operations for a logical cpu and
 502 * set it in the possible mask on success
 503 */
 504static int __init smp_cpu_setup(int cpu)
 505{
 506        if (cpu_read_ops(cpu))
 507                return -ENODEV;
 508
 509        if (cpu_ops[cpu]->cpu_init(cpu))
 510                return -ENODEV;
 511
 512        set_cpu_possible(cpu, true);
 513
 514        return 0;
 515}
 516
 517static bool bootcpu_valid __initdata;
 518static unsigned int cpu_count = 1;
 519
 520#ifdef CONFIG_ACPI
 521static struct acpi_madt_generic_interrupt cpu_madt_gicc[NR_CPUS];
 522
 523struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu)
 524{
 525        return &cpu_madt_gicc[cpu];
 526}
 527
 528/*
 529 * acpi_map_gic_cpu_interface - parse processor MADT entry
 530 *
 531 * Carry out sanity checks on MADT processor entry and initialize
 532 * cpu_logical_map on success
 533 */
 534static void __init
 535acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
 536{
 537        u64 hwid = processor->arm_mpidr;
 538
 539        if (!(processor->flags & ACPI_MADT_ENABLED)) {
 540                pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid);
 541                return;
 542        }
 543
 544        if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) {
 545                pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid);
 546                return;
 547        }
 548
 549        if (is_mpidr_duplicate(cpu_count, hwid)) {
 550                pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid);
 551                return;
 552        }
 553
 554        /* Check if GICC structure of boot CPU is available in the MADT */
 555        if (cpu_logical_map(0) == hwid) {
 556                if (bootcpu_valid) {
 557                        pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
 558                               hwid);
 559                        return;
 560                }
 561                bootcpu_valid = true;
 562                cpu_madt_gicc[0] = *processor;
 563                return;
 564        }
 565
 566        if (cpu_count >= NR_CPUS)
 567                return;
 568
 569        /* map the logical cpu id to cpu MPIDR */
 570        cpu_logical_map(cpu_count) = hwid;
 571
 572        cpu_madt_gicc[cpu_count] = *processor;
 573
 574        /*
 575         * Set-up the ACPI parking protocol cpu entries
 576         * while initializing the cpu_logical_map to
 577         * avoid parsing MADT entries multiple times for
 578         * nothing (ie a valid cpu_logical_map entry should
 579         * contain a valid parking protocol data set to
 580         * initialize the cpu if the parking protocol is
 581         * the only available enable method).
 582         */
 583        acpi_set_mailbox_entry(cpu_count, processor);
 584
 585        cpu_count++;
 586}
 587
 588static int __init
 589acpi_parse_gic_cpu_interface(struct acpi_subtable_header *header,
 590                             const unsigned long end)
 591{
 592        struct acpi_madt_generic_interrupt *processor;
 593
 594        processor = (struct acpi_madt_generic_interrupt *)header;
 595        if (BAD_MADT_GICC_ENTRY(processor, end))
 596                return -EINVAL;
 597
 598        acpi_table_print_madt_entry(header);
 599
 600        acpi_map_gic_cpu_interface(processor);
 601
 602        return 0;
 603}
 604
 605static void __init acpi_parse_and_init_cpus(void)
 606{
 607        int i;
 608
 609        /*
 610         * do a walk of MADT to determine how many CPUs
 611         * we have including disabled CPUs, and get information
 612         * we need for SMP init.
 613         */
 614        acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT,
 615                                      acpi_parse_gic_cpu_interface, 0);
 616
 617        /*
 618         * In ACPI, SMP and CPU NUMA information is provided in separate
 619         * static tables, namely the MADT and the SRAT.
 620         *
 621         * Thus, it is simpler to first create the cpu logical map through
 622         * an MADT walk and then map the logical cpus to their node ids
 623         * as separate steps.
 624         */
 625        acpi_map_cpus_to_nodes();
 626
 627        for (i = 0; i < nr_cpu_ids; i++)
 628                early_map_cpu_to_node(i, acpi_numa_get_nid(i));
 629}
 630#else
 631#define acpi_parse_and_init_cpus(...)   do { } while (0)
 632#endif
 633
 634/*
 635 * Enumerate the possible CPU set from the device tree and build the
 636 * cpu logical map array containing MPIDR values related to logical
 637 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
 638 */
 639static void __init of_parse_and_init_cpus(void)
 640{
 641        struct device_node *dn;
 642
 643        for_each_of_cpu_node(dn) {
 644                u64 hwid = of_get_cpu_mpidr(dn);
 645
 646                if (hwid == INVALID_HWID)
 647                        goto next;
 648
 649                if (is_mpidr_duplicate(cpu_count, hwid)) {
 650                        pr_err("%pOF: duplicate cpu reg properties in the DT\n",
 651                                dn);
 652                        goto next;
 653                }
 654
 655                /*
 656                 * The numbering scheme requires that the boot CPU
 657                 * must be assigned logical id 0. Record it so that
 658                 * the logical map built from DT is validated and can
 659                 * be used.
 660                 */
 661                if (hwid == cpu_logical_map(0)) {
 662                        if (bootcpu_valid) {
 663                                pr_err("%pOF: duplicate boot cpu reg property in DT\n",
 664                                        dn);
 665                                goto next;
 666                        }
 667
 668                        bootcpu_valid = true;
 669                        early_map_cpu_to_node(0, of_node_to_nid(dn));
 670
 671                        /*
 672                         * cpu_logical_map has already been
 673                         * initialized and the boot cpu doesn't need
 674                         * the enable-method so continue without
 675                         * incrementing cpu.
 676                         */
 677                        continue;
 678                }
 679
 680                if (cpu_count >= NR_CPUS)
 681                        goto next;
 682
 683                pr_debug("cpu logical map 0x%llx\n", hwid);
 684                cpu_logical_map(cpu_count) = hwid;
 685
 686                early_map_cpu_to_node(cpu_count, of_node_to_nid(dn));
 687next:
 688                cpu_count++;
 689        }
 690}
 691
 692/*
 693 * Enumerate the possible CPU set from the device tree or ACPI and build the
 694 * cpu logical map array containing MPIDR values related to logical
 695 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
 696 */
 697void __init smp_init_cpus(void)
 698{
 699        int i;
 700
 701        if (acpi_disabled)
 702                of_parse_and_init_cpus();
 703        else
 704                acpi_parse_and_init_cpus();
 705
 706        if (cpu_count > nr_cpu_ids)
 707                pr_warn("Number of cores (%d) exceeds configured maximum of %u - clipping\n",
 708                        cpu_count, nr_cpu_ids);
 709
 710        if (!bootcpu_valid) {
 711                pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
 712                return;
 713        }
 714
 715        /*
 716         * We need to set the cpu_logical_map entries before enabling
 717         * the cpus so that cpu processor description entries (DT cpu nodes
 718         * and ACPI MADT entries) can be retrieved by matching the cpu hwid
 719         * with entries in cpu_logical_map while initializing the cpus.
 720         * If the cpu set-up fails, invalidate the cpu_logical_map entry.
 721         */
 722        for (i = 1; i < nr_cpu_ids; i++) {
 723                if (cpu_logical_map(i) != INVALID_HWID) {
 724                        if (smp_cpu_setup(i))
 725                                cpu_logical_map(i) = INVALID_HWID;
 726                }
 727        }
 728}
 729
 730void __init smp_prepare_cpus(unsigned int max_cpus)
 731{
 732        int err;
 733        unsigned int cpu;
 734        unsigned int this_cpu;
 735
 736        init_cpu_topology();
 737
 738        this_cpu = smp_processor_id();
 739        store_cpu_topology(this_cpu);
 740        numa_store_cpu_info(this_cpu);
 741        numa_add_cpu(this_cpu);
 742
 743        /*
 744         * If UP is mandated by "nosmp" (which implies "maxcpus=0"), don't set
 745         * secondary CPUs present.
 746         */
 747        if (max_cpus == 0)
 748                return;
 749
 750        /*
 751         * Initialise the present map (which describes the set of CPUs
 752         * actually populated at the present time) and release the
 753         * secondaries from the bootloader.
 754         */
 755        for_each_possible_cpu(cpu) {
 756
 757                per_cpu(cpu_number, cpu) = cpu;
 758
 759                if (cpu == smp_processor_id())
 760                        continue;
 761
 762                if (!cpu_ops[cpu])
 763                        continue;
 764
 765                err = cpu_ops[cpu]->cpu_prepare(cpu);
 766                if (err)
 767                        continue;
 768
 769                set_cpu_present(cpu, true);
 770                numa_store_cpu_info(cpu);
 771        }
 772}
 773
 774void (*__smp_cross_call)(const struct cpumask *, unsigned int);
 775
 776void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
 777{
 778        __smp_cross_call = fn;
 779}
 780
 781static const char *ipi_types[NR_IPI] __tracepoint_string = {
 782#define S(x,s)  [x] = s
 783        S(IPI_RESCHEDULE, "Rescheduling interrupts"),
 784        S(IPI_CALL_FUNC, "Function call interrupts"),
 785        S(IPI_CPU_STOP, "CPU stop interrupts"),
 786        S(IPI_CPU_CRASH_STOP, "CPU stop (for crash dump) interrupts"),
 787        S(IPI_TIMER, "Timer broadcast interrupts"),
 788        S(IPI_IRQ_WORK, "IRQ work interrupts"),
 789        S(IPI_WAKEUP, "CPU wake-up interrupts"),
 790};
 791
 792static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
 793{
 794        trace_ipi_raise(target, ipi_types[ipinr]);
 795        __smp_cross_call(target, ipinr);
 796}
 797
 798void show_ipi_list(struct seq_file *p, int prec)
 799{
 800        unsigned int cpu, i;
 801
 802        for (i = 0; i < NR_IPI; i++) {
 803                seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
 804                           prec >= 4 ? " " : "");
 805                for_each_online_cpu(cpu)
 806                        seq_printf(p, "%10u ",
 807                                   __get_irq_stat(cpu, ipi_irqs[i]));
 808                seq_printf(p, "      %s\n", ipi_types[i]);
 809        }
 810}
 811
 812u64 smp_irq_stat_cpu(unsigned int cpu)
 813{
 814        u64 sum = 0;
 815        int i;
 816
 817        for (i = 0; i < NR_IPI; i++)
 818                sum += __get_irq_stat(cpu, ipi_irqs[i]);
 819
 820        return sum;
 821}
 822
 823void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 824{
 825        smp_cross_call(mask, IPI_CALL_FUNC);
 826}
 827
 828void arch_send_call_function_single_ipi(int cpu)
 829{
 830        smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
 831}
 832
 833#ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
 834void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
 835{
 836        smp_cross_call(mask, IPI_WAKEUP);
 837}
 838#endif
 839
 840#ifdef CONFIG_IRQ_WORK
 841void arch_irq_work_raise(void)
 842{
 843        if (__smp_cross_call)
 844                smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
 845}
 846#endif
 847
 848/*
 849 * ipi_cpu_stop - handle IPI from smp_send_stop()
 850 */
 851static void ipi_cpu_stop(unsigned int cpu)
 852{
 853        set_cpu_online(cpu, false);
 854
 855        local_daif_mask();
 856        sdei_mask_local_cpu();
 857
 858        while (1)
 859                cpu_relax();
 860}
 861
 862#ifdef CONFIG_KEXEC_CORE
 863static atomic_t waiting_for_crash_ipi = ATOMIC_INIT(0);
 864#endif
 865
 866static void ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs)
 867{
 868#ifdef CONFIG_KEXEC_CORE
 869        crash_save_cpu(regs, cpu);
 870
 871        atomic_dec(&waiting_for_crash_ipi);
 872
 873        local_irq_disable();
 874        sdei_mask_local_cpu();
 875
 876#ifdef CONFIG_HOTPLUG_CPU
 877        if (cpu_ops[cpu]->cpu_die)
 878                cpu_ops[cpu]->cpu_die(cpu);
 879#endif
 880
 881        /* just in case */
 882        cpu_park_loop();
 883#endif
 884}
 885
 886/*
 887 * Main handler for inter-processor interrupts
 888 */
 889void handle_IPI(int ipinr, struct pt_regs *regs)
 890{
 891        unsigned int cpu = smp_processor_id();
 892        struct pt_regs *old_regs = set_irq_regs(regs);
 893
 894        if ((unsigned)ipinr < NR_IPI) {
 895                trace_ipi_entry_rcuidle(ipi_types[ipinr]);
 896                __inc_irq_stat(cpu, ipi_irqs[ipinr]);
 897        }
 898
 899        switch (ipinr) {
 900        case IPI_RESCHEDULE:
 901                scheduler_ipi();
 902                break;
 903
 904        case IPI_CALL_FUNC:
 905                irq_enter();
 906                generic_smp_call_function_interrupt();
 907                irq_exit();
 908                break;
 909
 910        case IPI_CPU_STOP:
 911                irq_enter();
 912                ipi_cpu_stop(cpu);
 913                irq_exit();
 914                break;
 915
 916        case IPI_CPU_CRASH_STOP:
 917                if (IS_ENABLED(CONFIG_KEXEC_CORE)) {
 918                        irq_enter();
 919                        ipi_cpu_crash_stop(cpu, regs);
 920
 921                        unreachable();
 922                }
 923                break;
 924
 925#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 926        case IPI_TIMER:
 927                irq_enter();
 928                tick_receive_broadcast();
 929                irq_exit();
 930                break;
 931#endif
 932
 933#ifdef CONFIG_IRQ_WORK
 934        case IPI_IRQ_WORK:
 935                irq_enter();
 936                irq_work_run();
 937                irq_exit();
 938                break;
 939#endif
 940
 941#ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
 942        case IPI_WAKEUP:
 943                WARN_ONCE(!acpi_parking_protocol_valid(cpu),
 944                          "CPU%u: Wake-up IPI outside the ACPI parking protocol\n",
 945                          cpu);
 946                break;
 947#endif
 948
 949        default:
 950                pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
 951                break;
 952        }
 953
 954        if ((unsigned)ipinr < NR_IPI)
 955                trace_ipi_exit_rcuidle(ipi_types[ipinr]);
 956        set_irq_regs(old_regs);
 957}
 958
 959void smp_send_reschedule(int cpu)
 960{
 961        smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
 962}
 963
 964#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 965void tick_broadcast(const struct cpumask *mask)
 966{
 967        smp_cross_call(mask, IPI_TIMER);
 968}
 969#endif
 970
 971void smp_send_stop(void)
 972{
 973        unsigned long timeout;
 974
 975        if (num_online_cpus() > 1) {
 976                cpumask_t mask;
 977
 978                cpumask_copy(&mask, cpu_online_mask);
 979                cpumask_clear_cpu(smp_processor_id(), &mask);
 980
 981                if (system_state <= SYSTEM_RUNNING)
 982                        pr_crit("SMP: stopping secondary CPUs\n");
 983                smp_cross_call(&mask, IPI_CPU_STOP);
 984        }
 985
 986        /* Wait up to one second for other CPUs to stop */
 987        timeout = USEC_PER_SEC;
 988        while (num_online_cpus() > 1 && timeout--)
 989                udelay(1);
 990
 991        if (num_online_cpus() > 1)
 992                pr_warning("SMP: failed to stop secondary CPUs %*pbl\n",
 993                           cpumask_pr_args(cpu_online_mask));
 994
 995        sdei_mask_local_cpu();
 996}
 997
 998#ifdef CONFIG_KEXEC_CORE
 999void crash_smp_send_stop(void)
1000{
1001        static int cpus_stopped;
1002        cpumask_t mask;
1003        unsigned long timeout;
1004
1005        /*
1006         * This function can be called twice in panic path, but obviously
1007         * we execute this only once.
1008         */
1009        if (cpus_stopped)
1010                return;
1011
1012        cpus_stopped = 1;
1013
1014        if (num_online_cpus() == 1) {
1015                sdei_mask_local_cpu();
1016                return;
1017        }
1018
1019        cpumask_copy(&mask, cpu_online_mask);
1020        cpumask_clear_cpu(smp_processor_id(), &mask);
1021
1022        atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
1023
1024        pr_crit("SMP: stopping secondary CPUs\n");
1025        smp_cross_call(&mask, IPI_CPU_CRASH_STOP);
1026
1027        /* Wait up to one second for other CPUs to stop */
1028        timeout = USEC_PER_SEC;
1029        while ((atomic_read(&waiting_for_crash_ipi) > 0) && timeout--)
1030                udelay(1);
1031
1032        if (atomic_read(&waiting_for_crash_ipi) > 0)
1033                pr_warning("SMP: failed to stop secondary CPUs %*pbl\n",
1034                           cpumask_pr_args(&mask));
1035
1036        sdei_mask_local_cpu();
1037}
1038
1039bool smp_crash_stop_failed(void)
1040{
1041        return (atomic_read(&waiting_for_crash_ipi) > 0);
1042}
1043#endif
1044
1045/*
1046 * not supported here
1047 */
1048int setup_profiling_timer(unsigned int multiplier)
1049{
1050        return -EINVAL;
1051}
1052
1053static bool have_cpu_die(void)
1054{
1055#ifdef CONFIG_HOTPLUG_CPU
1056        int any_cpu = raw_smp_processor_id();
1057
1058        if (cpu_ops[any_cpu] && cpu_ops[any_cpu]->cpu_die)
1059                return true;
1060#endif
1061        return false;
1062}
1063
1064bool cpus_are_stuck_in_kernel(void)
1065{
1066        bool smp_spin_tables = (num_possible_cpus() > 1 && !have_cpu_die());
1067
1068        return !!cpus_stuck_in_kernel || smp_spin_tables;
1069}
1070