Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 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 | |
| 66 | DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number); |
| 67 | EXPORT_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 | */ |
| 74 | struct secondary_data secondary_data; |
| 75 | /* Number of CPUs which aren't online, but looping in kernel text. */ |
| 76 | int cpus_stuck_in_kernel; |
| 77 | |
| 78 | enum 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_HOTPLUG_CPU |
| 89 | static int op_cpu_kill(unsigned int cpu); |
| 90 | #else |
| 91 | static inline int op_cpu_kill(unsigned int cpu) |
| 92 | { |
| 93 | return -ENOSYS; |
| 94 | } |
| 95 | #endif |
| 96 | |
| 97 | |
| 98 | /* |
| 99 | * Boot a secondary CPU, and assign it the specified idle task. |
| 100 | * This also gives us the initial stack to use for this CPU. |
| 101 | */ |
| 102 | static int boot_secondary(unsigned int cpu, struct task_struct *idle) |
| 103 | { |
| 104 | if (cpu_ops[cpu]->cpu_boot) |
| 105 | return cpu_ops[cpu]->cpu_boot(cpu); |
| 106 | |
| 107 | return -EOPNOTSUPP; |
| 108 | } |
| 109 | |
| 110 | static DECLARE_COMPLETION(cpu_running); |
| 111 | |
| 112 | int __cpu_up(unsigned int cpu, struct task_struct *idle) |
| 113 | { |
| 114 | int ret; |
| 115 | long status; |
| 116 | |
| 117 | /* |
| 118 | * We need to tell the secondary core where to find its stack and the |
| 119 | * page tables. |
| 120 | */ |
| 121 | secondary_data.task = idle; |
| 122 | secondary_data.stack = task_stack_page(idle) + THREAD_SIZE; |
| 123 | update_cpu_boot_status(CPU_MMU_OFF); |
| 124 | __flush_dcache_area(&secondary_data, sizeof(secondary_data)); |
| 125 | |
| 126 | /* |
| 127 | * Now bring the CPU into our world. |
| 128 | */ |
| 129 | ret = boot_secondary(cpu, idle); |
| 130 | if (ret == 0) { |
| 131 | /* |
| 132 | * CPU was successfully started, wait for it to come online or |
| 133 | * time out. |
| 134 | */ |
| 135 | wait_for_completion_timeout(&cpu_running, |
| 136 | msecs_to_jiffies(1000)); |
| 137 | |
| 138 | if (!cpu_online(cpu)) { |
| 139 | pr_crit("CPU%u: failed to come online\n", cpu); |
| 140 | ret = -EIO; |
| 141 | } |
| 142 | } else { |
| 143 | pr_err("CPU%u: failed to boot: %d\n", cpu, ret); |
| 144 | } |
| 145 | |
| 146 | secondary_data.task = NULL; |
| 147 | secondary_data.stack = NULL; |
| 148 | status = READ_ONCE(secondary_data.status); |
| 149 | if (ret && status) { |
| 150 | |
| 151 | if (status == CPU_MMU_OFF) |
| 152 | status = READ_ONCE(__early_cpu_boot_status); |
| 153 | |
| 154 | switch (status) { |
| 155 | default: |
| 156 | pr_err("CPU%u: failed in unknown state : 0x%lx\n", |
| 157 | cpu, status); |
| 158 | break; |
| 159 | case CPU_KILL_ME: |
| 160 | if (!op_cpu_kill(cpu)) { |
| 161 | pr_crit("CPU%u: died during early boot\n", cpu); |
| 162 | break; |
| 163 | } |
| 164 | /* Fall through */ |
| 165 | pr_crit("CPU%u: may not have shut down cleanly\n", cpu); |
| 166 | case CPU_STUCK_IN_KERNEL: |
| 167 | pr_crit("CPU%u: is stuck in kernel\n", cpu); |
| 168 | cpus_stuck_in_kernel++; |
| 169 | break; |
| 170 | case CPU_PANIC_KERNEL: |
| 171 | panic("CPU%u detected unsupported configuration\n", cpu); |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | return ret; |
| 176 | } |
| 177 | |
| 178 | /* |
| 179 | * This is the secondary CPU boot entry. We're using this CPUs |
| 180 | * idle thread stack, but a set of temporary page tables. |
| 181 | */ |
| 182 | asmlinkage notrace void secondary_start_kernel(void) |
| 183 | { |
| 184 | u64 mpidr = read_cpuid_mpidr() & MPIDR_HWID_BITMASK; |
| 185 | struct mm_struct *mm = &init_mm; |
| 186 | unsigned int cpu; |
| 187 | |
| 188 | cpu = task_cpu(current); |
| 189 | set_my_cpu_offset(per_cpu_offset(cpu)); |
| 190 | |
| 191 | /* |
| 192 | * All kernel threads share the same mm context; grab a |
| 193 | * reference and switch to it. |
| 194 | */ |
| 195 | mmgrab(mm); |
| 196 | current->active_mm = mm; |
| 197 | |
| 198 | /* |
| 199 | * TTBR0 is only used for the identity mapping at this stage. Make it |
| 200 | * point to zero page to avoid speculatively fetching new entries. |
| 201 | */ |
| 202 | cpu_uninstall_idmap(); |
| 203 | |
| 204 | preempt_disable(); |
| 205 | trace_hardirqs_off(); |
| 206 | |
| 207 | /* |
| 208 | * If the system has established the capabilities, make sure |
| 209 | * this CPU ticks all of those. If it doesn't, the CPU will |
| 210 | * fail to come online. |
| 211 | */ |
| 212 | check_local_cpu_capabilities(); |
| 213 | |
| 214 | if (cpu_ops[cpu]->cpu_postboot) |
| 215 | cpu_ops[cpu]->cpu_postboot(); |
| 216 | |
| 217 | /* |
| 218 | * Log the CPU info before it is marked online and might get read. |
| 219 | */ |
| 220 | cpuinfo_store_cpu(); |
| 221 | |
| 222 | /* |
| 223 | * Enable GIC and timers. |
| 224 | */ |
| 225 | notify_cpu_starting(cpu); |
| 226 | |
| 227 | store_cpu_topology(cpu); |
| 228 | numa_add_cpu(cpu); |
| 229 | |
| 230 | /* |
| 231 | * OK, now it's safe to let the boot CPU continue. Wait for |
| 232 | * the CPU migration code to notice that the CPU is online |
| 233 | * before we continue. |
| 234 | */ |
| 235 | pr_info("CPU%u: Booted secondary processor 0x%010lx [0x%08x]\n", |
| 236 | cpu, (unsigned long)mpidr, |
| 237 | read_cpuid_id()); |
| 238 | update_cpu_boot_status(CPU_BOOT_SUCCESS); |
| 239 | set_cpu_online(cpu, true); |
| 240 | complete(&cpu_running); |
| 241 | |
| 242 | local_daif_restore(DAIF_PROCCTX); |
| 243 | |
| 244 | /* |
| 245 | * OK, it's off to the idle thread for us |
| 246 | */ |
| 247 | cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); |
| 248 | } |
| 249 | |
| 250 | #ifdef CONFIG_HOTPLUG_CPU |
| 251 | static int op_cpu_disable(unsigned int cpu) |
| 252 | { |
| 253 | /* |
| 254 | * If we don't have a cpu_die method, abort before we reach the point |
| 255 | * of no return. CPU0 may not have an cpu_ops, so test for it. |
| 256 | */ |
| 257 | if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die) |
| 258 | return -EOPNOTSUPP; |
| 259 | |
| 260 | /* |
| 261 | * We may need to abort a hot unplug for some other mechanism-specific |
| 262 | * reason. |
| 263 | */ |
| 264 | if (cpu_ops[cpu]->cpu_disable) |
| 265 | return cpu_ops[cpu]->cpu_disable(cpu); |
| 266 | |
| 267 | return 0; |
| 268 | } |
| 269 | |
| 270 | /* |
| 271 | * __cpu_disable runs on the processor to be shutdown. |
| 272 | */ |
| 273 | int __cpu_disable(void) |
| 274 | { |
| 275 | unsigned int cpu = smp_processor_id(); |
| 276 | int ret; |
| 277 | |
| 278 | ret = op_cpu_disable(cpu); |
| 279 | if (ret) |
| 280 | return ret; |
| 281 | |
| 282 | remove_cpu_topology(cpu); |
| 283 | numa_remove_cpu(cpu); |
| 284 | |
| 285 | /* |
| 286 | * Take this CPU offline. Once we clear this, we can't return, |
| 287 | * and we must not schedule until we're ready to give up the cpu. |
| 288 | */ |
| 289 | set_cpu_online(cpu, false); |
| 290 | |
| 291 | /* |
| 292 | * OK - migrate IRQs away from this CPU |
| 293 | */ |
| 294 | irq_migrate_all_off_this_cpu(); |
| 295 | |
| 296 | return 0; |
| 297 | } |
| 298 | |
| 299 | static int op_cpu_kill(unsigned int cpu) |
| 300 | { |
| 301 | /* |
| 302 | * If we have no means of synchronising with the dying CPU, then assume |
| 303 | * that it is really dead. We can only wait for an arbitrary length of |
| 304 | * time and hope that it's dead, so let's skip the wait and just hope. |
| 305 | */ |
| 306 | if (!cpu_ops[cpu]->cpu_kill) |
| 307 | return 0; |
| 308 | |
| 309 | return cpu_ops[cpu]->cpu_kill(cpu); |
| 310 | } |
| 311 | |
| 312 | /* |
| 313 | * called on the thread which is asking for a CPU to be shutdown - |
| 314 | * waits until shutdown has completed, or it is timed out. |
| 315 | */ |
| 316 | void __cpu_die(unsigned int cpu) |
| 317 | { |
| 318 | int err; |
| 319 | |
| 320 | if (!cpu_wait_death(cpu, 5)) { |
| 321 | pr_crit("CPU%u: cpu didn't die\n", cpu); |
| 322 | return; |
| 323 | } |
| 324 | pr_notice("CPU%u: shutdown\n", cpu); |
| 325 | |
| 326 | /* |
| 327 | * Now that the dying CPU is beyond the point of no return w.r.t. |
| 328 | * in-kernel synchronisation, try to get the firwmare to help us to |
| 329 | * verify that it has really left the kernel before we consider |
| 330 | * clobbering anything it might still be using. |
| 331 | */ |
| 332 | err = op_cpu_kill(cpu); |
| 333 | if (err) |
| 334 | pr_warn("CPU%d may not have shut down cleanly: %d\n", |
| 335 | cpu, err); |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * Called from the idle thread for the CPU which has been shutdown. |
| 340 | * |
| 341 | */ |
| 342 | void cpu_die(void) |
| 343 | { |
| 344 | unsigned int cpu = smp_processor_id(); |
| 345 | |
| 346 | idle_task_exit(); |
| 347 | |
| 348 | local_daif_mask(); |
| 349 | |
| 350 | /* Tell __cpu_die() that this CPU is now safe to dispose of */ |
| 351 | (void)cpu_report_death(); |
| 352 | |
| 353 | /* |
| 354 | * Actually shutdown the CPU. This must never fail. The specific hotplug |
| 355 | * mechanism must perform all required cache maintenance to ensure that |
| 356 | * no dirty lines are lost in the process of shutting down the CPU. |
| 357 | */ |
| 358 | cpu_ops[cpu]->cpu_die(cpu); |
| 359 | |
| 360 | BUG(); |
| 361 | } |
| 362 | #endif |
| 363 | |
| 364 | /* |
| 365 | * Kill the calling secondary CPU, early in bringup before it is turned |
| 366 | * online. |
| 367 | */ |
| 368 | void cpu_die_early(void) |
| 369 | { |
| 370 | int cpu = smp_processor_id(); |
| 371 | |
| 372 | pr_crit("CPU%d: will not boot\n", cpu); |
| 373 | |
| 374 | /* Mark this CPU absent */ |
| 375 | set_cpu_present(cpu, 0); |
| 376 | |
| 377 | #ifdef CONFIG_HOTPLUG_CPU |
| 378 | update_cpu_boot_status(CPU_KILL_ME); |
| 379 | /* Check if we can park ourselves */ |
| 380 | if (cpu_ops[cpu] && cpu_ops[cpu]->cpu_die) |
| 381 | cpu_ops[cpu]->cpu_die(cpu); |
| 382 | #endif |
| 383 | update_cpu_boot_status(CPU_STUCK_IN_KERNEL); |
| 384 | |
| 385 | cpu_park_loop(); |
| 386 | } |
| 387 | |
| 388 | static void __init hyp_mode_check(void) |
| 389 | { |
| 390 | if (is_hyp_mode_available()) |
| 391 | pr_info("CPU: All CPU(s) started at EL2\n"); |
| 392 | else if (is_hyp_mode_mismatched()) |
| 393 | WARN_TAINT(1, TAINT_CPU_OUT_OF_SPEC, |
| 394 | "CPU: CPUs started in inconsistent modes"); |
| 395 | else |
| 396 | pr_info("CPU: All CPU(s) started at EL1\n"); |
| 397 | } |
| 398 | |
| 399 | void __init smp_cpus_done(unsigned int max_cpus) |
| 400 | { |
| 401 | pr_info("SMP: Total of %d processors activated.\n", num_online_cpus()); |
| 402 | setup_cpu_features(); |
| 403 | hyp_mode_check(); |
| 404 | apply_alternatives_all(); |
| 405 | mark_linear_text_alias_ro(); |
| 406 | } |
| 407 | |
| 408 | void __init smp_prepare_boot_cpu(void) |
| 409 | { |
| 410 | set_my_cpu_offset(per_cpu_offset(smp_processor_id())); |
| 411 | /* |
| 412 | * Initialise the static keys early as they may be enabled by the |
| 413 | * cpufeature code. |
| 414 | */ |
| 415 | jump_label_init(); |
| 416 | cpuinfo_store_boot_cpu(); |
| 417 | } |
| 418 | |
| 419 | static u64 __init of_get_cpu_mpidr(struct device_node *dn) |
| 420 | { |
| 421 | const __be32 *cell; |
| 422 | u64 hwid; |
| 423 | |
| 424 | /* |
| 425 | * A cpu node with missing "reg" property is |
| 426 | * considered invalid to build a cpu_logical_map |
| 427 | * entry. |
| 428 | */ |
| 429 | cell = of_get_property(dn, "reg", NULL); |
| 430 | if (!cell) { |
| 431 | pr_err("%pOF: missing reg property\n", dn); |
| 432 | return INVALID_HWID; |
| 433 | } |
| 434 | |
| 435 | hwid = of_read_number(cell, of_n_addr_cells(dn)); |
| 436 | /* |
| 437 | * Non affinity bits must be set to 0 in the DT |
| 438 | */ |
| 439 | if (hwid & ~MPIDR_HWID_BITMASK) { |
| 440 | pr_err("%pOF: invalid reg property\n", dn); |
| 441 | return INVALID_HWID; |
| 442 | } |
| 443 | return hwid; |
| 444 | } |
| 445 | |
| 446 | /* |
| 447 | * Duplicate MPIDRs are a recipe for disaster. Scan all initialized |
| 448 | * entries and check for duplicates. If any is found just ignore the |
| 449 | * cpu. cpu_logical_map was initialized to INVALID_HWID to avoid |
| 450 | * matching valid MPIDR values. |
| 451 | */ |
| 452 | static bool __init is_mpidr_duplicate(unsigned int cpu, u64 hwid) |
| 453 | { |
| 454 | unsigned int i; |
| 455 | |
| 456 | for (i = 1; (i < cpu) && (i < NR_CPUS); i++) |
| 457 | if (cpu_logical_map(i) == hwid) |
| 458 | return true; |
| 459 | return false; |
| 460 | } |
| 461 | |
| 462 | /* |
| 463 | * Initialize cpu operations for a logical cpu and |
| 464 | * set it in the possible mask on success |
| 465 | */ |
| 466 | static int __init smp_cpu_setup(int cpu) |
| 467 | { |
| 468 | if (cpu_read_ops(cpu)) |
| 469 | return -ENODEV; |
| 470 | |
| 471 | if (cpu_ops[cpu]->cpu_init(cpu)) |
| 472 | return -ENODEV; |
| 473 | |
| 474 | set_cpu_possible(cpu, true); |
| 475 | |
| 476 | return 0; |
| 477 | } |
| 478 | |
| 479 | static bool bootcpu_valid __initdata; |
| 480 | static unsigned int cpu_count = 1; |
| 481 | |
| 482 | #ifdef CONFIG_ACPI |
| 483 | static struct acpi_madt_generic_interrupt cpu_madt_gicc[NR_CPUS]; |
| 484 | |
| 485 | struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu) |
| 486 | { |
| 487 | return &cpu_madt_gicc[cpu]; |
| 488 | } |
| 489 | |
| 490 | /* |
| 491 | * acpi_map_gic_cpu_interface - parse processor MADT entry |
| 492 | * |
| 493 | * Carry out sanity checks on MADT processor entry and initialize |
| 494 | * cpu_logical_map on success |
| 495 | */ |
| 496 | static void __init |
| 497 | acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor) |
| 498 | { |
| 499 | u64 hwid = processor->arm_mpidr; |
| 500 | |
| 501 | if (!(processor->flags & ACPI_MADT_ENABLED)) { |
| 502 | pr_debug("skipping disabled CPU entry with 0x%llx MPIDR\n", hwid); |
| 503 | return; |
| 504 | } |
| 505 | |
| 506 | if (hwid & ~MPIDR_HWID_BITMASK || hwid == INVALID_HWID) { |
| 507 | pr_err("skipping CPU entry with invalid MPIDR 0x%llx\n", hwid); |
| 508 | return; |
| 509 | } |
| 510 | |
| 511 | if (is_mpidr_duplicate(cpu_count, hwid)) { |
| 512 | pr_err("duplicate CPU MPIDR 0x%llx in MADT\n", hwid); |
| 513 | return; |
| 514 | } |
| 515 | |
| 516 | /* Check if GICC structure of boot CPU is available in the MADT */ |
| 517 | if (cpu_logical_map(0) == hwid) { |
| 518 | if (bootcpu_valid) { |
| 519 | pr_err("duplicate boot CPU MPIDR: 0x%llx in MADT\n", |
| 520 | hwid); |
| 521 | return; |
| 522 | } |
| 523 | bootcpu_valid = true; |
| 524 | cpu_madt_gicc[0] = *processor; |
| 525 | return; |
| 526 | } |
| 527 | |
| 528 | if (cpu_count >= NR_CPUS) |
| 529 | return; |
| 530 | |
| 531 | /* map the logical cpu id to cpu MPIDR */ |
| 532 | cpu_logical_map(cpu_count) = hwid; |
| 533 | |
| 534 | cpu_madt_gicc[cpu_count] = *processor; |
| 535 | |
| 536 | /* |
| 537 | * Set-up the ACPI parking protocol cpu entries |
| 538 | * while initializing the cpu_logical_map to |
| 539 | * avoid parsing MADT entries multiple times for |
| 540 | * nothing (ie a valid cpu_logical_map entry should |
| 541 | * contain a valid parking protocol data set to |
| 542 | * initialize the cpu if the parking protocol is |
| 543 | * the only available enable method). |
| 544 | */ |
| 545 | acpi_set_mailbox_entry(cpu_count, processor); |
| 546 | |
| 547 | cpu_count++; |
| 548 | } |
| 549 | |
| 550 | static int __init |
| 551 | acpi_parse_gic_cpu_interface(struct acpi_subtable_header *header, |
| 552 | const unsigned long end) |
| 553 | { |
| 554 | struct acpi_madt_generic_interrupt *processor; |
| 555 | |
| 556 | processor = (struct acpi_madt_generic_interrupt *)header; |
| 557 | if (BAD_MADT_GICC_ENTRY(processor, end)) |
| 558 | return -EINVAL; |
| 559 | |
| 560 | acpi_table_print_madt_entry(header); |
| 561 | |
| 562 | acpi_map_gic_cpu_interface(processor); |
| 563 | |
| 564 | return 0; |
| 565 | } |
| 566 | |
| 567 | static void __init acpi_parse_and_init_cpus(void) |
| 568 | { |
| 569 | int i; |
| 570 | |
| 571 | /* |
| 572 | * do a walk of MADT to determine how many CPUs |
| 573 | * we have including disabled CPUs, and get information |
| 574 | * we need for SMP init. |
| 575 | */ |
| 576 | acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_INTERRUPT, |
| 577 | acpi_parse_gic_cpu_interface, 0); |
| 578 | |
| 579 | /* |
| 580 | * In ACPI, SMP and CPU NUMA information is provided in separate |
| 581 | * static tables, namely the MADT and the SRAT. |
| 582 | * |
| 583 | * Thus, it is simpler to first create the cpu logical map through |
| 584 | * an MADT walk and then map the logical cpus to their node ids |
| 585 | * as separate steps. |
| 586 | */ |
| 587 | acpi_map_cpus_to_nodes(); |
| 588 | |
| 589 | for (i = 0; i < nr_cpu_ids; i++) |
| 590 | early_map_cpu_to_node(i, acpi_numa_get_nid(i)); |
| 591 | } |
| 592 | #else |
| 593 | #define acpi_parse_and_init_cpus(...) do { } while (0) |
| 594 | #endif |
| 595 | |
| 596 | /* |
| 597 | * Enumerate the possible CPU set from the device tree and build the |
| 598 | * cpu logical map array containing MPIDR values related to logical |
| 599 | * cpus. Assumes that cpu_logical_map(0) has already been initialized. |
| 600 | */ |
| 601 | static void __init of_parse_and_init_cpus(void) |
| 602 | { |
| 603 | struct device_node *dn; |
| 604 | |
| 605 | for_each_node_by_type(dn, "cpu") { |
| 606 | u64 hwid = of_get_cpu_mpidr(dn); |
| 607 | |
| 608 | if (hwid == INVALID_HWID) |
| 609 | goto next; |
| 610 | |
| 611 | if (is_mpidr_duplicate(cpu_count, hwid)) { |
| 612 | pr_err("%pOF: duplicate cpu reg properties in the DT\n", |
| 613 | dn); |
| 614 | goto next; |
| 615 | } |
| 616 | |
| 617 | /* |
| 618 | * The numbering scheme requires that the boot CPU |
| 619 | * must be assigned logical id 0. Record it so that |
| 620 | * the logical map built from DT is validated and can |
| 621 | * be used. |
| 622 | */ |
| 623 | if (hwid == cpu_logical_map(0)) { |
| 624 | if (bootcpu_valid) { |
| 625 | pr_err("%pOF: duplicate boot cpu reg property in DT\n", |
| 626 | dn); |
| 627 | goto next; |
| 628 | } |
| 629 | |
| 630 | bootcpu_valid = true; |
| 631 | early_map_cpu_to_node(0, of_node_to_nid(dn)); |
| 632 | |
| 633 | /* |
| 634 | * cpu_logical_map has already been |
| 635 | * initialized and the boot cpu doesn't need |
| 636 | * the enable-method so continue without |
| 637 | * incrementing cpu. |
| 638 | */ |
| 639 | continue; |
| 640 | } |
| 641 | |
| 642 | if (cpu_count >= NR_CPUS) |
| 643 | goto next; |
| 644 | |
| 645 | pr_debug("cpu logical map 0x%llx\n", hwid); |
| 646 | cpu_logical_map(cpu_count) = hwid; |
| 647 | |
| 648 | early_map_cpu_to_node(cpu_count, of_node_to_nid(dn)); |
| 649 | next: |
| 650 | cpu_count++; |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | /* |
| 655 | * Enumerate the possible CPU set from the device tree or ACPI and build the |
| 656 | * cpu logical map array containing MPIDR values related to logical |
| 657 | * cpus. Assumes that cpu_logical_map(0) has already been initialized. |
| 658 | */ |
| 659 | void __init smp_init_cpus(void) |
| 660 | { |
| 661 | int i; |
| 662 | |
| 663 | if (acpi_disabled) |
| 664 | of_parse_and_init_cpus(); |
| 665 | else |
| 666 | acpi_parse_and_init_cpus(); |
| 667 | |
| 668 | if (cpu_count > nr_cpu_ids) |
| 669 | pr_warn("Number of cores (%d) exceeds configured maximum of %u - clipping\n", |
| 670 | cpu_count, nr_cpu_ids); |
| 671 | |
| 672 | if (!bootcpu_valid) { |
| 673 | pr_err("missing boot CPU MPIDR, not enabling secondaries\n"); |
| 674 | return; |
| 675 | } |
| 676 | |
| 677 | /* |
| 678 | * We need to set the cpu_logical_map entries before enabling |
| 679 | * the cpus so that cpu processor description entries (DT cpu nodes |
| 680 | * and ACPI MADT entries) can be retrieved by matching the cpu hwid |
| 681 | * with entries in cpu_logical_map while initializing the cpus. |
| 682 | * If the cpu set-up fails, invalidate the cpu_logical_map entry. |
| 683 | */ |
| 684 | for (i = 1; i < nr_cpu_ids; i++) { |
| 685 | if (cpu_logical_map(i) != INVALID_HWID) { |
| 686 | if (smp_cpu_setup(i)) |
| 687 | cpu_logical_map(i) = INVALID_HWID; |
| 688 | } |
| 689 | } |
| 690 | } |
| 691 | |
| 692 | void __init smp_prepare_cpus(unsigned int max_cpus) |
| 693 | { |
| 694 | int err; |
| 695 | unsigned int cpu; |
| 696 | unsigned int this_cpu; |
| 697 | |
| 698 | init_cpu_topology(); |
| 699 | |
| 700 | this_cpu = smp_processor_id(); |
| 701 | store_cpu_topology(this_cpu); |
| 702 | numa_store_cpu_info(this_cpu); |
| 703 | numa_add_cpu(this_cpu); |
| 704 | |
| 705 | /* |
| 706 | * If UP is mandated by "nosmp" (which implies "maxcpus=0"), don't set |
| 707 | * secondary CPUs present. |
| 708 | */ |
| 709 | if (max_cpus == 0) |
| 710 | return; |
| 711 | |
| 712 | /* |
| 713 | * Initialise the present map (which describes the set of CPUs |
| 714 | * actually populated at the present time) and release the |
| 715 | * secondaries from the bootloader. |
| 716 | */ |
| 717 | for_each_possible_cpu(cpu) { |
| 718 | |
| 719 | per_cpu(cpu_number, cpu) = cpu; |
| 720 | |
| 721 | if (cpu == smp_processor_id()) |
| 722 | continue; |
| 723 | |
| 724 | if (!cpu_ops[cpu]) |
| 725 | continue; |
| 726 | |
| 727 | err = cpu_ops[cpu]->cpu_prepare(cpu); |
| 728 | if (err) |
| 729 | continue; |
| 730 | |
| 731 | set_cpu_present(cpu, true); |
| 732 | numa_store_cpu_info(cpu); |
| 733 | } |
| 734 | } |
| 735 | |
| 736 | void (*__smp_cross_call)(const struct cpumask *, unsigned int); |
| 737 | |
| 738 | void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int)) |
| 739 | { |
| 740 | __smp_cross_call = fn; |
| 741 | } |
| 742 | |
| 743 | static const char *ipi_types[NR_IPI] __tracepoint_string = { |
| 744 | #define S(x,s) [x] = s |
| 745 | S(IPI_RESCHEDULE, "Rescheduling interrupts"), |
| 746 | S(IPI_CALL_FUNC, "Function call interrupts"), |
| 747 | S(IPI_CPU_STOP, "CPU stop interrupts"), |
| 748 | S(IPI_CPU_CRASH_STOP, "CPU stop (for crash dump) interrupts"), |
| 749 | S(IPI_TIMER, "Timer broadcast interrupts"), |
| 750 | S(IPI_IRQ_WORK, "IRQ work interrupts"), |
| 751 | S(IPI_WAKEUP, "CPU wake-up interrupts"), |
| 752 | }; |
| 753 | |
| 754 | static void smp_cross_call(const struct cpumask *target, unsigned int ipinr) |
| 755 | { |
| 756 | trace_ipi_raise(target, ipi_types[ipinr]); |
| 757 | __smp_cross_call(target, ipinr); |
| 758 | } |
| 759 | |
| 760 | void show_ipi_list(struct seq_file *p, int prec) |
| 761 | { |
| 762 | unsigned int cpu, i; |
| 763 | |
| 764 | for (i = 0; i < NR_IPI; i++) { |
| 765 | seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i, |
| 766 | prec >= 4 ? " " : ""); |
| 767 | for_each_online_cpu(cpu) |
| 768 | seq_printf(p, "%10u ", |
| 769 | __get_irq_stat(cpu, ipi_irqs[i])); |
| 770 | seq_printf(p, " %s\n", ipi_types[i]); |
| 771 | } |
| 772 | } |
| 773 | |
| 774 | u64 smp_irq_stat_cpu(unsigned int cpu) |
| 775 | { |
| 776 | u64 sum = 0; |
| 777 | int i; |
| 778 | |
| 779 | for (i = 0; i < NR_IPI; i++) |
| 780 | sum += __get_irq_stat(cpu, ipi_irqs[i]); |
| 781 | |
| 782 | return sum; |
| 783 | } |
| 784 | |
| 785 | void arch_send_call_function_ipi_mask(const struct cpumask *mask) |
| 786 | { |
| 787 | smp_cross_call(mask, IPI_CALL_FUNC); |
| 788 | } |
| 789 | |
| 790 | void arch_send_call_function_single_ipi(int cpu) |
| 791 | { |
| 792 | smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC); |
| 793 | } |
| 794 | |
| 795 | #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL |
| 796 | void arch_send_wakeup_ipi_mask(const struct cpumask *mask) |
| 797 | { |
| 798 | smp_cross_call(mask, IPI_WAKEUP); |
| 799 | } |
| 800 | #endif |
| 801 | |
| 802 | #ifdef CONFIG_IRQ_WORK |
| 803 | void arch_irq_work_raise(void) |
| 804 | { |
| 805 | if (__smp_cross_call) |
| 806 | smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK); |
| 807 | } |
| 808 | #endif |
| 809 | |
| 810 | /* |
| 811 | * ipi_cpu_stop - handle IPI from smp_send_stop() |
| 812 | */ |
| 813 | static void ipi_cpu_stop(unsigned int cpu) |
| 814 | { |
| 815 | set_cpu_online(cpu, false); |
| 816 | |
| 817 | local_daif_mask(); |
| 818 | sdei_mask_local_cpu(); |
| 819 | |
| 820 | while (1) |
| 821 | cpu_relax(); |
| 822 | } |
| 823 | |
| 824 | #ifdef CONFIG_KEXEC_CORE |
| 825 | static atomic_t waiting_for_crash_ipi = ATOMIC_INIT(0); |
| 826 | #endif |
| 827 | |
| 828 | static void ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs) |
| 829 | { |
| 830 | #ifdef CONFIG_KEXEC_CORE |
| 831 | crash_save_cpu(regs, cpu); |
| 832 | |
| 833 | atomic_dec(&waiting_for_crash_ipi); |
| 834 | |
| 835 | local_irq_disable(); |
| 836 | sdei_mask_local_cpu(); |
| 837 | |
| 838 | #ifdef CONFIG_HOTPLUG_CPU |
| 839 | if (cpu_ops[cpu]->cpu_die) |
| 840 | cpu_ops[cpu]->cpu_die(cpu); |
| 841 | #endif |
| 842 | |
| 843 | /* just in case */ |
| 844 | cpu_park_loop(); |
| 845 | #endif |
| 846 | } |
| 847 | |
| 848 | /* |
| 849 | * Main handler for inter-processor interrupts |
| 850 | */ |
| 851 | void handle_IPI(int ipinr, struct pt_regs *regs) |
| 852 | { |
| 853 | unsigned int cpu = smp_processor_id(); |
| 854 | struct pt_regs *old_regs = set_irq_regs(regs); |
| 855 | |
| 856 | if ((unsigned)ipinr < NR_IPI) { |
| 857 | trace_ipi_entry_rcuidle(ipi_types[ipinr]); |
| 858 | __inc_irq_stat(cpu, ipi_irqs[ipinr]); |
| 859 | } |
| 860 | |
| 861 | switch (ipinr) { |
| 862 | case IPI_RESCHEDULE: |
| 863 | scheduler_ipi(); |
| 864 | break; |
| 865 | |
| 866 | case IPI_CALL_FUNC: |
| 867 | irq_enter(); |
| 868 | generic_smp_call_function_interrupt(); |
| 869 | irq_exit(); |
| 870 | break; |
| 871 | |
| 872 | case IPI_CPU_STOP: |
| 873 | irq_enter(); |
| 874 | ipi_cpu_stop(cpu); |
| 875 | irq_exit(); |
| 876 | break; |
| 877 | |
| 878 | case IPI_CPU_CRASH_STOP: |
| 879 | if (IS_ENABLED(CONFIG_KEXEC_CORE)) { |
| 880 | irq_enter(); |
| 881 | ipi_cpu_crash_stop(cpu, regs); |
| 882 | |
| 883 | unreachable(); |
| 884 | } |
| 885 | break; |
| 886 | |
| 887 | #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST |
| 888 | case IPI_TIMER: |
| 889 | irq_enter(); |
| 890 | tick_receive_broadcast(); |
| 891 | irq_exit(); |
| 892 | break; |
| 893 | #endif |
| 894 | |
| 895 | #ifdef CONFIG_IRQ_WORK |
| 896 | case IPI_IRQ_WORK: |
| 897 | irq_enter(); |
| 898 | irq_work_run(); |
| 899 | irq_exit(); |
| 900 | break; |
| 901 | #endif |
| 902 | |
| 903 | #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL |
| 904 | case IPI_WAKEUP: |
| 905 | WARN_ONCE(!acpi_parking_protocol_valid(cpu), |
| 906 | "CPU%u: Wake-up IPI outside the ACPI parking protocol\n", |
| 907 | cpu); |
| 908 | break; |
| 909 | #endif |
| 910 | |
| 911 | default: |
| 912 | pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr); |
| 913 | break; |
| 914 | } |
| 915 | |
| 916 | if ((unsigned)ipinr < NR_IPI) |
| 917 | trace_ipi_exit_rcuidle(ipi_types[ipinr]); |
| 918 | set_irq_regs(old_regs); |
| 919 | } |
| 920 | |
| 921 | void smp_send_reschedule(int cpu) |
| 922 | { |
| 923 | smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE); |
| 924 | } |
| 925 | |
| 926 | #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST |
| 927 | void tick_broadcast(const struct cpumask *mask) |
| 928 | { |
| 929 | smp_cross_call(mask, IPI_TIMER); |
| 930 | } |
| 931 | #endif |
| 932 | |
| 933 | void smp_send_stop(void) |
| 934 | { |
| 935 | unsigned long timeout; |
| 936 | |
| 937 | if (num_online_cpus() > 1) { |
| 938 | cpumask_t mask; |
| 939 | |
| 940 | cpumask_copy(&mask, cpu_online_mask); |
| 941 | cpumask_clear_cpu(smp_processor_id(), &mask); |
| 942 | |
| 943 | if (system_state <= SYSTEM_RUNNING) |
| 944 | pr_crit("SMP: stopping secondary CPUs\n"); |
| 945 | smp_cross_call(&mask, IPI_CPU_STOP); |
| 946 | } |
| 947 | |
| 948 | /* Wait up to one second for other CPUs to stop */ |
| 949 | timeout = USEC_PER_SEC; |
| 950 | while (num_online_cpus() > 1 && timeout--) |
| 951 | udelay(1); |
| 952 | |
| 953 | if (num_online_cpus() > 1) |
| 954 | pr_warning("SMP: failed to stop secondary CPUs %*pbl\n", |
| 955 | cpumask_pr_args(cpu_online_mask)); |
| 956 | |
| 957 | sdei_mask_local_cpu(); |
| 958 | } |
| 959 | |
| 960 | #ifdef CONFIG_KEXEC_CORE |
| 961 | void crash_smp_send_stop(void) |
| 962 | { |
| 963 | static int cpus_stopped; |
| 964 | cpumask_t mask; |
| 965 | unsigned long timeout; |
| 966 | |
| 967 | /* |
| 968 | * This function can be called twice in panic path, but obviously |
| 969 | * we execute this only once. |
| 970 | */ |
| 971 | if (cpus_stopped) |
| 972 | return; |
| 973 | |
| 974 | cpus_stopped = 1; |
| 975 | |
| 976 | if (num_online_cpus() == 1) { |
| 977 | sdei_mask_local_cpu(); |
| 978 | return; |
| 979 | } |
| 980 | |
| 981 | cpumask_copy(&mask, cpu_online_mask); |
| 982 | cpumask_clear_cpu(smp_processor_id(), &mask); |
| 983 | |
| 984 | atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1); |
| 985 | |
| 986 | pr_crit("SMP: stopping secondary CPUs\n"); |
| 987 | smp_cross_call(&mask, IPI_CPU_CRASH_STOP); |
| 988 | |
| 989 | /* Wait up to one second for other CPUs to stop */ |
| 990 | timeout = USEC_PER_SEC; |
| 991 | while ((atomic_read(&waiting_for_crash_ipi) > 0) && timeout--) |
| 992 | udelay(1); |
| 993 | |
| 994 | if (atomic_read(&waiting_for_crash_ipi) > 0) |
| 995 | pr_warning("SMP: failed to stop secondary CPUs %*pbl\n", |
| 996 | cpumask_pr_args(&mask)); |
| 997 | |
| 998 | sdei_mask_local_cpu(); |
| 999 | } |
| 1000 | |
| 1001 | bool smp_crash_stop_failed(void) |
| 1002 | { |
| 1003 | return (atomic_read(&waiting_for_crash_ipi) > 0); |
| 1004 | } |
| 1005 | #endif |
| 1006 | |
| 1007 | /* |
| 1008 | * not supported here |
| 1009 | */ |
| 1010 | int setup_profiling_timer(unsigned int multiplier) |
| 1011 | { |
| 1012 | return -EINVAL; |
| 1013 | } |
| 1014 | |
| 1015 | static bool have_cpu_die(void) |
| 1016 | { |
| 1017 | #ifdef CONFIG_HOTPLUG_CPU |
| 1018 | int any_cpu = raw_smp_processor_id(); |
| 1019 | |
| 1020 | if (cpu_ops[any_cpu] && cpu_ops[any_cpu]->cpu_die) |
| 1021 | return true; |
| 1022 | #endif |
| 1023 | return false; |
| 1024 | } |
| 1025 | |
| 1026 | bool cpus_are_stuck_in_kernel(void) |
| 1027 | { |
| 1028 | bool smp_spin_tables = (num_possible_cpus() > 1 && !have_cpu_die()); |
| 1029 | |
| 1030 | return !!cpus_stuck_in_kernel || smp_spin_tables; |
| 1031 | } |