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/percpu.h>
  39#include <linux/clockchips.h>
  40#include <linux/completion.h>
  41#include <linux/of.h>
  42#include <linux/irq_work.h>
  43#include <linux/kexec.h>
  44
  45#include <asm/alternative.h>
  46#include <asm/atomic.h>
  47#include <asm/cacheflush.h>
  48#include <asm/cpu.h>
  49#include <asm/cputype.h>
  50#include <asm/cpu_ops.h>
  51#include <asm/daifflags.h>
  52#include <asm/mmu_context.h>
  53#include <asm/numa.h>
  54#include <asm/pgtable.h>
  55#include <asm/pgalloc.h>
  56#include <asm/processor.h>
  57#include <asm/smp_plat.h>
  58#include <asm/sections.h>
  59#include <asm/tlbflush.h>
  60#include <asm/ptrace.h>
  61#include <asm/virt.h>
  62
  63#define CREATE_TRACE_POINTS
  64#include <trace/events/ipi.h>
  65
  66DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number);
  67EXPORT_PER_CPU_SYMBOL(cpu_number);
  68
  69/*
  70 * as from 2.5, kernels no longer have an init_tasks structure
  71 * so we need some other way of telling a new secondary core
  72 * where to place its SVC stack
  73 */
  74struct secondary_data secondary_data;
  75/* Number of CPUs which aren't online, but looping in kernel text. */
  76int cpus_stuck_in_kernel;
  77
  78enum ipi_msg_type {
  79        IPI_RESCHEDULE,
  80        IPI_CALL_FUNC,
  81        IPI_CPU_STOP,
  82        IPI_CPU_CRASH_STOP,
  83        IPI_TIMER,
  84        IPI_IRQ_WORK,
  85        IPI_WAKEUP
  86};
  87
  88#ifdef CONFIG_ARM64_VHE
  89
  90/* Whether the boot CPU is running in HYP mode or not*/
  91static bool boot_cpu_hyp_mode;
  92
  93static inline void save_boot_cpu_run_el(void)
  94{
  95        boot_cpu_hyp_mode = is_kernel_in_hyp_mode();
  96}
  97
  98static inline bool is_boot_cpu_in_hyp_mode(void)
  99{
 100        return boot_cpu_hyp_mode;
 101}
 102
 103/*
 104 * Verify that a secondary CPU is running the kernel at the same
 105 * EL as that of the boot CPU.
 106 */
 107void verify_cpu_run_el(void)
 108{
 109        bool in_el2 = is_kernel_in_hyp_mode();
 110        bool boot_cpu_el2 = is_boot_cpu_in_hyp_mode();
 111
 112        if (in_el2 ^ boot_cpu_el2) {
 113                pr_crit("CPU%d: mismatched Exception Level(EL%d) with boot CPU(EL%d)\n",
 114                                        smp_processor_id(),
 115                                        in_el2 ? 2 : 1,
 116                                        boot_cpu_el2 ? 2 : 1);
 117                cpu_panic_kernel();
 118        }
 119}
 120
 121#else
 122static inline void save_boot_cpu_run_el(void) {}
 123#endif
 124
 125#ifdef CONFIG_HOTPLUG_CPU
 126static int op_cpu_kill(unsigned int cpu);
 127#else
 128static inline int op_cpu_kill(unsigned int cpu)
 129{
 130        return -ENOSYS;
 131}
 132#endif
 133
 134
 135/*
 136 * Boot a secondary CPU, and assign it the specified idle task.
 137 * This also gives us the initial stack to use for this CPU.
 138 */
 139static int boot_secondary(unsigned int cpu, struct task_struct *idle)
 140{
 141        if (cpu_ops[cpu]->cpu_boot)
 142                return cpu_ops[cpu]->cpu_boot(cpu);
 143
 144        return -EOPNOTSUPP;
 145}
 146
 147static DECLARE_COMPLETION(cpu_running);
 148
 149int __cpu_up(unsigned int cpu, struct task_struct *idle)
 150{
 151        int ret;
 152        long status;
 153
 154        /*
 155         * We need to tell the secondary core where to find its stack and the
 156         * page tables.
 157         */
 158        secondary_data.task = idle;
 159        secondary_data.stack = task_stack_page(idle) + THREAD_SIZE;
 160        update_cpu_boot_status(CPU_MMU_OFF);
 161        __flush_dcache_area(&secondary_data, sizeof(secondary_data));
 162
 163        /*
 164         * Now bring the CPU into our world.
 165         */
 166        ret = boot_secondary(cpu, idle);
 167        if (ret == 0) {
 168                /*
 169                 * CPU was successfully started, wait for it to come online or
 170                 * time out.
 171                 */
 172                wait_for_completion_timeout(&cpu_running,
 173                                            msecs_to_jiffies(1000));
 174
 175                if (!cpu_online(cpu)) {
 176                        pr_crit("CPU%u: failed to come online\n", cpu);
 177                        ret = -EIO;
 178                }
 179        } else {
 180                pr_err("CPU%u: failed to boot: %d\n", cpu, ret);
 181        }
 182
 183        secondary_data.task = NULL;
 184        secondary_data.stack = NULL;
 185        status = READ_ONCE(secondary_data.status);
 186        if (ret && status) {
 187
 188                if (status == CPU_MMU_OFF)
 189                        status = READ_ONCE(__early_cpu_boot_status);
 190
 191                switch (status) {
 192                default:
 193                        pr_err("CPU%u: failed in unknown state : 0x%lx\n",
 194                                        cpu, status);
 195                        break;
 196                case CPU_KILL_ME:
 197                        if (!op_cpu_kill(cpu)) {
 198                                pr_crit("CPU%u: died during early boot\n", cpu);
 199                                break;
 200                        }
 201                        /* Fall through */
 202                        pr_crit("CPU%u: may not have shut down cleanly\n", cpu);
 203                case CPU_STUCK_IN_KERNEL:
 204                        pr_crit("CPU%u: is stuck in kernel\n", cpu);
 205                        cpus_stuck_in_kernel++;
 206                        break;
 207                case CPU_PANIC_KERNEL:
 208                        panic("CPU%u detected unsupported configuration\n", cpu);
 209                }
 210        }
 211
 212        return ret;
 213}
 214
 215/*
 216 * This is the secondary CPU boot entry.  We're using this CPUs
 217 * idle thread stack, but a set of temporary page tables.
 218 */
 219asmlinkage void secondary_start_kernel(void)
 220{
 221        u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK;
 222        struct mm_struct *mm = &init_mm;
 223        unsigned int cpu;
 224
 225        cpu = task_cpu(current);
 226        set_my_cpu_offset(per_cpu_offset(cpu));
 227
 228        /*
 229         * All kernel threads share the same mm context; grab a
 230         * reference and switch to it.
 231         */
 232        mmgrab(mm);
 233        current->active_mm = mm;
 234
 235        /*
 236         * TTBR0 is only used for the identity mapping at this stage. Make it
 237         * point to zero page to avoid speculatively fetching new entries.
 238         */
 239        cpu_uninstall_idmap();
 240
 241        preempt_disable();
 242        trace_hardirqs_off();
 243
 244        /*
 245         * If the system has established the capabilities, make sure
 246         * this CPU ticks all of those. If it doesn't, the CPU will
 247         * fail to come online.
 248         */
 249        check_local_cpu_capabilities();
 250
 251        if (cpu_ops[cpu]->cpu_postboot)
 252                cpu_ops[cpu]->cpu_postboot();
 253
 254        /*
 255         * Log the CPU info before it is marked online and might get read.
 256         */
 257        cpuinfo_store_cpu();
 258
 259        /*
 260         * Enable GIC and timers.
 261         */
 262        notify_cpu_starting(cpu);
 263
 264        store_cpu_topology(cpu);
 265
 266        /*
 267         * OK, now it's safe to let the boot CPU continue.  Wait for
 268         * the CPU migration code to notice that the CPU is online
 269         * before we continue.
 270         */
 271        pr_info("CPU%u: Booted secondary processor 0x%010lx [0x%08x]\n",
 272                                         cpu, (unsigned long)mpidr,
 273                                         read_cpuid_id());
 274        update_cpu_boot_status(CPU_BOOT_SUCCESS);
 275        set_cpu_online(cpu, true);
 276        complete(&cpu_running);
 277
 278        local_daif_restore(DAIF_PROCCTX);
 279
 280        /*
 281         * OK, it's off to the idle thread for us
 282         */
 283        cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 284}
 285
 286#ifdef CONFIG_HOTPLUG_CPU
 287static int op_cpu_disable(unsigned int cpu)
 288{
 289        /*
 290         * If we don't have a cpu_die method, abort before we reach the point
 291         * of no return. CPU0 may not have an cpu_ops, so test for it.
 292         */
 293        if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die)
 294                return -EOPNOTSUPP;
 295
 296        /*
 297         * We may need to abort a hot unplug for some other mechanism-specific
 298         * reason.
 299         */
 300        if (cpu_ops[cpu]->cpu_disable)
 301                return cpu_ops[cpu]->cpu_disable(cpu);
 302
 303        return 0;
 304}
 305
 306/*
 307 * __cpu_disable runs on the processor to be shutdown.
 308 */
 309int __cpu_disable(void)
 310{
 311        unsigned int cpu = smp_processor_id();
 312        int ret;
 313
 314        ret = op_cpu_disable(cpu);
 315        if (ret)
 316                return ret;
 317
 318        /*
 319         * Take this CPU offline.  Once we clear this, we can't return,
 320         * and we must not schedule until we're ready to give up the cpu.
 321         */
 322        set_cpu_online(cpu, false);
 323
 324        /*
 325         * OK - migrate IRQs away from this CPU
 326         */
 327        irq_migrate_all_off_this_cpu();
 328
 329        return 0;
 330}
 331
 332static int op_cpu_kill(unsigned int cpu)
 333{
 334        /*
 335         * If we have no means of synchronising with the dying CPU, then assume
 336         * that it is really dead. We can only wait for an arbitrary length of
 337         * time and hope that it's dead, so let's skip the wait and just hope.
 338         */
 339        if (!cpu_ops[cpu]->cpu_kill)
 340                return 0;
 341
 342        return cpu_ops[cpu]->cpu_kill(cpu);
 343}
 344
 345/*
 346 * called on the thread which is asking for a CPU to be shutdown -
 347 * waits until shutdown has completed, or it is timed out.
 348 */
 349void __cpu_die(unsigned int cpu)
 350{
 351        int err;
 352
 353        if (!cpu_wait_death(cpu, 5)) {
 354                pr_crit("CPU%u: cpu didn't die\n", cpu);
 355                return;
 356        }
 357        pr_notice("CPU%u: shutdown\n", cpu);
 358
 359        /*
 360         * Now that the dying CPU is beyond the point of no return w.r.t.
 361         * in-kernel synchronisation, try to get the firwmare to help us to
 362         * verify that it has really left the kernel before we consider
 363         * clobbering anything it might still be using.
 364         */
 365        err = op_cpu_kill(cpu);
 366        if (err)
 367                pr_warn("CPU%d may not have shut down cleanly: %d\n",
 368                        cpu, err);
 369}
 370
 371/*
 372 * Called from the idle thread for the CPU which has been shutdown.
 373 *
 374 */
 375void cpu_die(void)
 376{
 377        unsigned int cpu = smp_processor_id();
 378
 379        idle_task_exit();
 380
 381        local_daif_mask();
 382
 383        /* Tell __cpu_die() that this CPU is now safe to dispose of */
 384        (void)cpu_report_death();
 385
 386        /*
 387         * Actually shutdown the CPU. This must never fail. The specific hotplug
 388         * mechanism must perform all required cache maintenance to ensure that
 389         * no dirty lines are lost in the process of shutting down the CPU.
 390         */
 391        cpu_ops[cpu]->cpu_die(cpu);
 392
 393        BUG();
 394}
 395#endif
 396
 397/*
 398 * Kill the calling secondary CPU, early in bringup before it is turned
 399 * online.
 400 */
 401void cpu_die_early(void)
 402{
 403        int cpu = smp_processor_id();
 404
 405        pr_crit("CPU%d: will not boot\n", cpu);
 406
 407        /* Mark this CPU absent */
 408        set_cpu_present(cpu, 0);
 409
 410#ifdef CONFIG_HOTPLUG_CPU
 411        update_cpu_boot_status(CPU_KILL_ME);
 412        /* Check if we can park ourselves */
 413        if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die)
 414                cpu_ops[cpu]->cpu_die(cpu);
 415#endif
 416        update_cpu_boot_status(CPU_STUCK_IN_KERNEL);
 417
 418        cpu_park_loop();
 419}
 420
 421static void __init hyp_mode_check(void)
 422{
 423        if (is_hyp_mode_available())
 424                pr_info("CPU: All CPU(s) started at EL2\n");
 425        else if (is_hyp_mode_mismatched())
 426                WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC,
 427                           "CPU: CPUs started in inconsistent modes");
 428        else
 429                pr_info("CPU: All CPU(s) started at EL1\n");
 430}
 431
 432void __init smp_cpus_done(unsigned int max_cpus)
 433{
 434        pr_info("SMP: Total of %d processors activated.\n", num_online_cpus());
 435        setup_cpu_features();
 436        hyp_mode_check();
 437        apply_alternatives_all();
 438        mark_linear_text_alias_ro();
 439}
 440
 441void __init smp_prepare_boot_cpu(void)
 442{
 443        set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
 444        /*
 445         * Initialise the static keys early as they may be enabled by the
 446         * cpufeature code.
 447         */
 448        jump_label_init();
 449        cpuinfo_store_boot_cpu();
 450        save_boot_cpu_run_el();
 451        /*
 452         * Run the errata work around checks on the boot CPU, once we have
 453         * initialised the cpu feature infrastructure from
 454         * cpuinfo_store_boot_cpu() above.
 455         */
 456        update_cpu_errata_workarounds();
 457}
 458
 459static u64 __init of_get_cpu_mpidr(struct device_node *dn)
 460{
 461        const __be32 *cell;
 462        u64 hwid;
 463
 464        /*
 465         * A cpu node with missing "reg" property is
 466         * considered invalid to build a cpu_logical_map
 467         * entry.
 468         */
 469        cell = of_get_property(dn, "reg", NULL);
 470        if (!cell) {
 471                pr_err("%pOF: missing reg property\n", dn);
 472                return INVALID_HWID;
 473        }
 474
 475        hwid = of_read_number(cell, of_n_addr_cells(dn));
 476        /*
 477         * Non affinity bits must be set to 0 in the DT
 478         */
 479        if (hwid & ~MPIDR_HWID_BITMASK) {
 480                pr_err("%pOF: invalid reg property\n", dn);
 481                return INVALID_HWID;
 482        }
 483        return hwid;
 484}
 485
 486/*
 487 * Duplicate MPIDRs are a recipe for disaster. Scan all initialized
 488 * entries and check for duplicates. If any is found just ignore the
 489 * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid
 490 * matching valid MPIDR values.
 491 */
 492static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid)
 493{
 494        unsigned int i;
 495
 496        for (i = 1; (i < cpu) && (i < NR_CPUS); i++)
 497                if (cpu_logical_map(i) == hwid)
 498                        return true;
 499        return false;
 500}
 501
 502/*
 503 * Initialize cpu operations for a logical cpu and
 504 * set it in the possible mask on success
 505 */
 506static int __init smp_cpu_setup(int cpu)
 507{
 508        if (cpu_read_ops(cpu))
 509                return -ENODEV;
 510
 511        if (cpu_ops[cpu]->cpu_init(cpu))
 512                return -ENODEV;
 513
 514        set_cpu_possible(cpu, true);
 515
 516        return 0;
 517}
 518
 519static bool bootcpu_valid __initdata;
 520static unsigned int cpu_count = 1;
 521
 522#ifdef CONFIG_ACPI
 523static struct acpi_madt_generic_interrupt cpu_madt_gicc[NR_CPUS];
 524
 525struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu)
 526{
 527        return &cpu_madt_gicc[cpu];
 528}
 529
 530/*
 531 * acpi_map_gic_cpu_interface - parse processor MADT entry
 532 *
 533 * Carry out sanity checks on MADT processor entry and initialize
 534 * cpu_logical_map on success
 535 */
 536static void __init
 537acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
 538{
 539        u64 hwid = processor->arm_mpidr;
 540
 541        if (!(processor->flags & ACPI_MADT_ENABLED)) {
 542                pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid);
 543                return;
 544        }
 545
 546        if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) {
 547                pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid);
 548                return;
 549        }
 550
 551        if (is_mpidr_duplicate(cpu_count, hwid)) {
 552                pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid);
 553                return;
 554        }
 555
 556        /* Check if GICC structure of boot CPU is available in the MADT */
 557        if (cpu_logical_map(0) == hwid) {
 558                if (bootcpu_valid) {
 559                        pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n",
 560                               hwid);
 561                        return;
 562                }
 563                bootcpu_valid = true;
 564                cpu_madt_gicc[0] = *processor;
 565                early_map_cpu_to_node(0, acpi_numa_get_nid(0, hwid));
 566                return;
 567        }
 568
 569        if (cpu_count >= NR_CPUS)
 570                return;
 571
 572        /* map the logical cpu id to cpu MPIDR */
 573        cpu_logical_map(cpu_count) = hwid;
 574
 575        cpu_madt_gicc[cpu_count] = *processor;
 576
 577        /*
 578         * Set-up the ACPI parking protocol cpu entries
 579         * while initializing the cpu_logical_map to
 580         * avoid parsing MADT entries multiple times for
 581         * nothing (ie a valid cpu_logical_map entry should
 582         * contain a valid parking protocol data set to
 583         * initialize the cpu if the parking protocol is
 584         * the only available enable method).
 585         */
 586        acpi_set_mailbox_entry(cpu_count, processor);
 587
 588        early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count, hwid));
 589
 590        cpu_count++;
 591}
 592
 593static int __init
 594acpi_parse_gic_cpu_interface(struct acpi_subtable_header *header,
 595                             const unsigned long end)
 596{
 597        struct acpi_madt_generic_interrupt *processor;
 598
 599        processor = (struct acpi_madt_generic_interrupt *)header;
 600        if (BAD_MADT_GICC_ENTRY(processor, end))
 601                return -EINVAL;
 602
 603        acpi_table_print_madt_entry(header);
 604
 605        acpi_map_gic_cpu_interface(processor);
 606
 607        return 0;
 608}
 609#else
 610#define acpi_table_parse_madt(...)      do { } while (0)
 611#endif
 612
 613/*
 614 * Enumerate the possible CPU set from the device tree and build the
 615 * cpu logical map array containing MPIDR values related to logical
 616 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
 617 */
 618static void __init of_parse_and_init_cpus(void)
 619{
 620        struct device_node *dn;
 621
 622        for_each_node_by_type(dn, "cpu") {
 623                u64 hwid = of_get_cpu_mpidr(dn);
 624
 625                if (hwid == INVALID_HWID)
 626                        goto next;
 627
 628                if (is_mpidr_duplicate(cpu_count, hwid)) {
 629                        pr_err("%pOF: duplicate cpu reg properties in the DT\n",
 630                                dn);
 631                        goto next;
 632                }
 633
 634                /*
 635                 * The numbering scheme requires that the boot CPU
 636                 * must be assigned logical id 0. Record it so that
 637                 * the logical map built from DT is validated and can
 638                 * be used.
 639                 */
 640                if (hwid == cpu_logical_map(0)) {
 641                        if (bootcpu_valid) {
 642                                pr_err("%pOF: duplicate boot cpu reg property in DT\n",
 643                                        dn);
 644                                goto next;
 645                        }
 646
 647                        bootcpu_valid = true;
 648                        early_map_cpu_to_node(0, of_node_to_nid(dn));
 649
 650                        /*
 651                         * cpu_logical_map has already been
 652                         * initialized and the boot cpu doesn't need
 653                         * the enable-method so continue without
 654                         * incrementing cpu.
 655                         */
 656                        continue;
 657                }
 658
 659                if (cpu_count >= NR_CPUS)
 660                        goto next;
 661
 662                pr_debug("cpu logical map 0x%llx\n", hwid);
 663                cpu_logical_map(cpu_count) = hwid;
 664
 665                early_map_cpu_to_node(cpu_count, of_node_to_nid(dn));
 666next:
 667                cpu_count++;
 668        }
 669}
 670
 671/*
 672 * Enumerate the possible CPU set from the device tree or ACPI and build the
 673 * cpu logical map array containing MPIDR values related to logical
 674 * cpus. Assumes that cpu_logical_map(0) has already been initialized.
 675 */
 676void __init smp_init_cpus(void)
 677{
 678        int i;
 679
 680        if (acpi_disabled)
 681                of_parse_and_init_cpus();
 682        else
 683                /*
 684                 * do a walk of MADT to determine how many CPUs
 685                 * we have including disabled CPUs, and get information
 686                 * we need for SMP init
 687                 */
 688                acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT,
 689                                      acpi_parse_gic_cpu_interface, 0);
 690
 691        if (cpu_count > nr_cpu_ids)
 692                pr_warn("Number of cores (%d) exceeds configured maximum of %u - clipping\n",
 693                        cpu_count, nr_cpu_ids);
 694
 695        if (!bootcpu_valid) {
 696                pr_err("missing boot CPU MPIDR, not enabling secondaries\n");
 697                return;
 698        }
 699
 700        /*
 701         * We need to set the cpu_logical_map entries before enabling
 702         * the cpus so that cpu processor description entries (DT cpu nodes
 703         * and ACPI MADT entries) can be retrieved by matching the cpu hwid
 704         * with entries in cpu_logical_map while initializing the cpus.
 705         * If the cpu set-up fails, invalidate the cpu_logical_map entry.
 706         */
 707        for (i = 1; i < nr_cpu_ids; i++) {
 708                if (cpu_logical_map(i) != INVALID_HWID) {
 709                        if (smp_cpu_setup(i))
 710                                cpu_logical_map(i) = INVALID_HWID;
 711                }
 712        }
 713}
 714
 715void __init smp_prepare_cpus(unsigned int max_cpus)
 716{
 717        int err;
 718        unsigned int cpu;
 719        unsigned int this_cpu;
 720
 721        init_cpu_topology();
 722
 723        this_cpu = smp_processor_id();
 724        store_cpu_topology(this_cpu);
 725        numa_store_cpu_info(this_cpu);
 726
 727        /*
 728         * If UP is mandated by "nosmp" (which implies "maxcpus=0"), don't set
 729         * secondary CPUs present.
 730         */
 731        if (max_cpus == 0)
 732                return;
 733
 734        /*
 735         * Initialise the present map (which describes the set of CPUs
 736         * actually populated at the present time) and release the
 737         * secondaries from the bootloader.
 738         */
 739        for_each_possible_cpu(cpu) {
 740
 741                per_cpu(cpu_number, cpu) = cpu;
 742
 743                if (cpu == smp_processor_id())
 744                        continue;
 745
 746                if (!cpu_ops[cpu])
 747                        continue;
 748
 749                err = cpu_ops[cpu]->cpu_prepare(cpu);
 750                if (err)
 751                        continue;
 752
 753                set_cpu_present(cpu, true);
 754                numa_store_cpu_info(cpu);
 755        }
 756}
 757
 758void (*__smp_cross_call)(const struct cpumask *, unsigned int);
 759
 760void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
 761{
 762        __smp_cross_call = fn;
 763}
 764
 765static const char *ipi_types[NR_IPI] __tracepoint_string = {
 766#define S(x,s)  [x] = s
 767        S(IPI_RESCHEDULE, "Rescheduling interrupts"),
 768        S(IPI_CALL_FUNC, "Function call interrupts"),
 769        S(IPI_CPU_STOP, "CPU stop interrupts"),
 770        S(IPI_CPU_CRASH_STOP, "CPU stop (for crash dump) interrupts"),
 771        S(IPI_TIMER, "Timer broadcast interrupts"),
 772        S(IPI_IRQ_WORK, "IRQ work interrupts"),
 773        S(IPI_WAKEUP, "CPU wake-up interrupts"),
 774};
 775
 776static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
 777{
 778        trace_ipi_raise(target, ipi_types[ipinr]);
 779        __smp_cross_call(target, ipinr);
 780}
 781
 782void show_ipi_list(struct seq_file *p, int prec)
 783{
 784        unsigned int cpu, i;
 785
 786        for (i = 0; i < NR_IPI; i++) {
 787                seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i,
 788                           prec >= 4 ? " " : "");
 789                for_each_online_cpu(cpu)
 790                        seq_printf(p, "%10u ",
 791                                   __get_irq_stat(cpu, ipi_irqs[i]));
 792                seq_printf(p, "      %s\n", ipi_types[i]);
 793        }
 794}
 795
 796u64 smp_irq_stat_cpu(unsigned int cpu)
 797{
 798        u64 sum = 0;
 799        int i;
 800
 801        for (i = 0; i < NR_IPI; i++)
 802                sum += __get_irq_stat(cpu, ipi_irqs[i]);
 803
 804        return sum;
 805}
 806
 807void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 808{
 809        smp_cross_call(mask, IPI_CALL_FUNC);
 810}
 811
 812void arch_send_call_function_single_ipi(int cpu)
 813{
 814        smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC);
 815}
 816
 817#ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
 818void arch_send_wakeup_ipi_mask(const struct cpumask *mask)
 819{
 820        smp_cross_call(mask, IPI_WAKEUP);
 821}
 822#endif
 823
 824#ifdef CONFIG_IRQ_WORK
 825void arch_irq_work_raise(void)
 826{
 827        if (__smp_cross_call)
 828                smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK);
 829}
 830#endif
 831
 832/*
 833 * ipi_cpu_stop - handle IPI from smp_send_stop()
 834 */
 835static void ipi_cpu_stop(unsigned int cpu)
 836{
 837        set_cpu_online(cpu, false);
 838
 839        local_daif_mask();
 840        sdei_mask_local_cpu();
 841
 842        while (1)
 843                cpu_relax();
 844}
 845
 846#ifdef CONFIG_KEXEC_CORE
 847static atomic_t waiting_for_crash_ipi = ATOMIC_INIT(0);
 848#endif
 849
 850static void ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs)
 851{
 852#ifdef CONFIG_KEXEC_CORE
 853        crash_save_cpu(regs, cpu);
 854
 855        atomic_dec(&waiting_for_crash_ipi);
 856
 857        local_irq_disable();
 858        sdei_mask_local_cpu();
 859
 860#ifdef CONFIG_HOTPLUG_CPU
 861        if (cpu_ops[cpu]->cpu_die)
 862                cpu_ops[cpu]->cpu_die(cpu);
 863#endif
 864
 865        /* just in case */
 866        cpu_park_loop();
 867#endif
 868}
 869
 870/*
 871 * Main handler for inter-processor interrupts
 872 */
 873void handle_IPI(int ipinr, struct pt_regs *regs)
 874{
 875        unsigned int cpu = smp_processor_id();
 876        struct pt_regs *old_regs = set_irq_regs(regs);
 877
 878        if ((unsigned)ipinr < NR_IPI) {
 879                trace_ipi_entry_rcuidle(ipi_types[ipinr]);
 880                __inc_irq_stat(cpu, ipi_irqs[ipinr]);
 881        }
 882
 883        switch (ipinr) {
 884        case IPI_RESCHEDULE:
 885                scheduler_ipi();
 886                break;
 887
 888        case IPI_CALL_FUNC:
 889                irq_enter();
 890                generic_smp_call_function_interrupt();
 891                irq_exit();
 892                break;
 893
 894        case IPI_CPU_STOP:
 895                irq_enter();
 896                ipi_cpu_stop(cpu);
 897                irq_exit();
 898                break;
 899
 900        case IPI_CPU_CRASH_STOP:
 901                if (IS_ENABLED(CONFIG_KEXEC_CORE)) {
 902                        irq_enter();
 903                        ipi_cpu_crash_stop(cpu, regs);
 904
 905                        unreachable();
 906                }
 907                break;
 908
 909#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 910        case IPI_TIMER:
 911                irq_enter();
 912                tick_receive_broadcast();
 913                irq_exit();
 914                break;
 915#endif
 916
 917#ifdef CONFIG_IRQ_WORK
 918        case IPI_IRQ_WORK:
 919                irq_enter();
 920                irq_work_run();
 921                irq_exit();
 922                break;
 923#endif
 924
 925#ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
 926        case IPI_WAKEUP:
 927                WARN_ONCE(!acpi_parking_protocol_valid(cpu),
 928                          "CPU%u: Wake-up IPI outside the ACPI parking protocol\n",
 929                          cpu);
 930                break;
 931#endif
 932
 933        default:
 934                pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
 935                break;
 936        }
 937
 938        if ((unsigned)ipinr < NR_IPI)
 939                trace_ipi_exit_rcuidle(ipi_types[ipinr]);
 940        set_irq_regs(old_regs);
 941}
 942
 943void smp_send_reschedule(int cpu)
 944{
 945        smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE);
 946}
 947
 948#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 949void tick_broadcast(const struct cpumask *mask)
 950{
 951        smp_cross_call(mask, IPI_TIMER);
 952}
 953#endif
 954
 955void smp_send_stop(void)
 956{
 957        unsigned long timeout;
 958
 959        if (num_online_cpus() > 1) {
 960                cpumask_t mask;
 961
 962                cpumask_copy(&mask, cpu_online_mask);
 963                cpumask_clear_cpu(smp_processor_id(), &mask);
 964
 965                if (system_state <= SYSTEM_RUNNING)
 966                        pr_crit("SMP: stopping secondary CPUs\n");
 967                smp_cross_call(&mask, IPI_CPU_STOP);
 968        }
 969
 970        /* Wait up to one second for other CPUs to stop */
 971        timeout = USEC_PER_SEC;
 972        while (num_online_cpus() > 1 && timeout--)
 973                udelay(1);
 974
 975        if (num_online_cpus() > 1)
 976                pr_warning("SMP: failed to stop secondary CPUs %*pbl\n",
 977                           cpumask_pr_args(cpu_online_mask));
 978
 979        sdei_mask_local_cpu();
 980}
 981
 982#ifdef CONFIG_KEXEC_CORE
 983void crash_smp_send_stop(void)
 984{
 985        static int cpus_stopped;
 986        cpumask_t mask;
 987        unsigned long timeout;
 988
 989        /*
 990         * This function can be called twice in panic path, but obviously
 991         * we execute this only once.
 992         */
 993        if (cpus_stopped)
 994                return;
 995
 996        cpus_stopped = 1;
 997
 998        if (num_online_cpus() == 1) {
 999                sdei_mask_local_cpu();
1000                return;
1001        }
1002
1003        cpumask_copy(&mask, cpu_online_mask);
1004        cpumask_clear_cpu(smp_processor_id(), &mask);
1005
1006        atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
1007
1008        pr_crit("SMP: stopping secondary CPUs\n");
1009        smp_cross_call(&mask, IPI_CPU_CRASH_STOP);
1010
1011        /* Wait up to one second for other CPUs to stop */
1012        timeout = USEC_PER_SEC;
1013        while ((atomic_read(&waiting_for_crash_ipi) > 0) && timeout--)
1014                udelay(1);
1015
1016        if (atomic_read(&waiting_for_crash_ipi) > 0)
1017                pr_warning("SMP: failed to stop secondary CPUs %*pbl\n",
1018                           cpumask_pr_args(&mask));
1019
1020        sdei_mask_local_cpu();
1021}
1022
1023bool smp_crash_stop_failed(void)
1024{
1025        return (atomic_read(&waiting_for_crash_ipi) > 0);
1026}
1027#endif
1028
1029/*
1030 * not supported here
1031 */
1032int setup_profiling_timer(unsigned int multiplier)
1033{
1034        return -EINVAL;
1035}
1036
1037static bool have_cpu_die(void)
1038{
1039#ifdef CONFIG_HOTPLUG_CPU
1040        int any_cpu = raw_smp_processor_id();
1041
1042        if (cpu_ops[any_cpu] && cpu_ops[any_cpu]->cpu_die)
1043                return true;
1044#endif
1045        return false;
1046}
1047
1048bool cpus_are_stuck_in_kernel(void)
1049{
1050        bool smp_spin_tables = (num_possible_cpus() > 1 && !have_cpu_die());
1051
1052        return !!cpus_stuck_in_kernel || smp_spin_tables;
1053}
1054