arch/xtensa/kernel/smp.c

   1 /*
   2  * Xtensa SMP support functions.
   3  *
   4  * This file is subject to the terms and conditions of the GNU General Public
   5  * License.  See the file "COPYING" in the main directory of this archive
   6  * for more details.
   7  *
   8  * Copyright (C) 2008 - 2013 Tensilica Inc.
   9  *
  10  * Chris Zankel <chris@zankel.net>
  11  * Joe Taylor <joe@tensilica.com>
  12  * Pete Delaney <piet@tensilica.com
  13  */
  14
  15 #include <linux/cpu.h>
  16 #include <linux/cpumask.h>
  17 #include <linux/delay.h>
  18 #include <linux/init.h>
  19 #include <linux/interrupt.h>
  20 #include <linux/irqdomain.h>
  21 #include <linux/irq.h>
  22 #include <linux/kdebug.h>
  23 #include <linux/module.h>
  24 #include <linux/sched/mm.h>
  25 #include <linux/sched/hotplug.h>
  26 #include <linux/sched/task_stack.h>
  27 #include <linux/reboot.h>
  28 #include <linux/seq_file.h>
  29 #include <linux/smp.h>
  30 #include <linux/thread_info.h>
  31
  32 #include <asm/cacheflush.h>
  33 #include <asm/kdebug.h>
  34 #include <asm/mmu_context.h>
  35 #include <asm/mxregs.h>
  36 #include <asm/platform.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/traps.h>
  39
  40 #ifdef CONFIG_SMP
  41 # if XCHAL_HAVE_S32C1I == 0
  42 #  error "The S32C1I option is required for SMP."
  43 # endif
  44 #endif
  45
  46 static void system_invalidate_dcache_range(unsigned long start,
  47                 unsigned long size);
  48 static void system_flush_invalidate_dcache_range(unsigned long start,
  49                 unsigned long size);
  50
  51 /* IPI (Inter Process Interrupt) */
  52
  53 #define IPI_IRQ 0
  54
  55 static irqreturn_t ipi_interrupt(int irq, void *dev_id);
  56
  57 void ipi_init(void)
  58 {
  59         unsigned irq = irq_create_mapping(NULL, IPI_IRQ);
  60         if (request_irq(irq, ipi_interrupt, IRQF_PERCPU, "ipi", NULL))
  61                 pr_err("Failed to request irq %u (ipi)\n", irq);
  62 }
  63
  64 static inline unsigned int get_core_count(void)
  65 {
  66         /* Bits 18..21 of SYSCFGID contain the core count minus 1. */
  67         unsigned int syscfgid = get_er(SYSCFGID);
  68         return ((syscfgid >> 18) & 0xf) + 1;
  69 }
  70
  71 static inline int get_core_id(void)
  72 {
  73         /* Bits 0...18 of SYSCFGID contain the core id  */
  74         unsigned int core_id = get_er(SYSCFGID);
  75         return core_id & 0x3fff;
  76 }
  77
  78 void __init smp_prepare_cpus(unsigned int max_cpus)
  79 {
  80         unsigned i;
  81
  82         for_each_possible_cpu(i)
  83                 set_cpu_present(i, true);
  84 }
  85
  86 void __init smp_init_cpus(void)
  87 {
  88         unsigned i;
  89         unsigned int ncpus = get_core_count();
  90         unsigned int core_id = get_core_id();
  91
  92         pr_info("%s: Core Count = %d\n", __func__, ncpus);
  93         pr_info("%s: Core Id = %d\n", __func__, core_id);
  94
  95         if (ncpus > NR_CPUS) {
  96                 ncpus = NR_CPUS;
  97                 pr_info("%s: limiting core count by %d\n", __func__, ncpus);
  98         }
  99
 100         for (i = 0; i < ncpus; ++i)
 101                 set_cpu_possible(i, true);
 102 }
 103
 104 void __init smp_prepare_boot_cpu(void)
 105 {
 106         unsigned int cpu = smp_processor_id();
 107         BUG_ON(cpu != 0);
 108         cpu_asid_cache(cpu) = ASID_USER_FIRST;
 109 }
 110
 111 void __init smp_cpus_done(unsigned int max_cpus)
 112 {
 113 }
 114
 115 static int boot_secondary_processors = 1; /* Set with xt-gdb via .xt-gdb */
 116 static DECLARE_COMPLETION(cpu_running);
 117
 118 void secondary_start_kernel(void)
 119 {
 120         struct mm_struct *mm = &init_mm;
 121         unsigned int cpu = smp_processor_id();
 122
 123         init_mmu();
 124
 125 #ifdef CONFIG_DEBUG_MISC
 126         if (boot_secondary_processors == 0) {
 127                 pr_debug("%s: boot_secondary_processors:%d; Hanging cpu:%d\n",
 128                         __func__, boot_secondary_processors, cpu);
 129                 for (;;)
 130                         __asm__ __volatile__ ("waiti " __stringify(LOCKLEVEL));
 131         }
 132
 133         pr_debug("%s: boot_secondary_processors:%d; Booting cpu:%d\n",
 134                 __func__, boot_secondary_processors, cpu);
 135 #endif
 136         /* Init EXCSAVE1 */
 137
 138         secondary_trap_init();
 139
 140         /* All kernel threads share the same mm context. */
 141
 142         mmget(mm);
 143         mmgrab(mm);
 144         current->active_mm = mm;
 145         cpumask_set_cpu(cpu, mm_cpumask(mm));
 146         enter_lazy_tlb(mm, current);
 147
 148         preempt_disable();
 149         trace_hardirqs_off();
 150
 151         calibrate_delay();
 152
 153         notify_cpu_starting(cpu);
 154
 155         secondary_init_irq();
 156         local_timer_setup(cpu);
 157
 158         set_cpu_online(cpu, true);
 159
 160         local_irq_enable();
 161
 162         complete(&cpu_running);
 163
 164         cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
 165 }
 166
 167 static void mx_cpu_start(void *p)
 168 {
 169         unsigned cpu = (unsigned)p;
 170         unsigned long run_stall_mask = get_er(MPSCORE);
 171
 172         set_er(run_stall_mask & ~(1u << cpu), MPSCORE);
 173         pr_debug("%s: cpu: %d, run_stall_mask: %lx ---> %lx\n",
 174                         __func__, cpu, run_stall_mask, get_er(MPSCORE));
 175 }
 176
 177 static void mx_cpu_stop(void *p)
 178 {
 179         unsigned cpu = (unsigned)p;
 180         unsigned long run_stall_mask = get_er(MPSCORE);
 181
 182         set_er(run_stall_mask | (1u << cpu), MPSCORE);
 183         pr_debug("%s: cpu: %d, run_stall_mask: %lx ---> %lx\n",
 184                         __func__, cpu, run_stall_mask, get_er(MPSCORE));
 185 }
 186
 187 #ifdef CONFIG_HOTPLUG_CPU
 188 unsigned long cpu_start_id __cacheline_aligned;
 189 #endif
 190 unsigned long cpu_start_ccount;
 191
 192 static int boot_secondary(unsigned int cpu, struct task_struct *ts)
 193 {
 194         unsigned long timeout = jiffies + msecs_to_jiffies(1000);
 195         unsigned long ccount;
 196         int i;
 197
 198 #ifdef CONFIG_HOTPLUG_CPU
 199         WRITE_ONCE(cpu_start_id, cpu);
 200         /* Pairs with the third memw in the cpu_restart */
 201         mb();
 202         system_flush_invalidate_dcache_range((unsigned long)&cpu_start_id,
 203                                              sizeof(cpu_start_id));
 204 #endif
 205         smp_call_function_single(0, mx_cpu_start, (void *)cpu, 1);
 206
 207         for (i = 0; i < 2; ++i) {
 208                 do
 209                         ccount = get_ccount();
 210                 while (!ccount);
 211
 212                 WRITE_ONCE(cpu_start_ccount, ccount);
 213
 214                 do {
 215                         /*
 216                          * Pairs with the first two memws in the
 217                          * .Lboot_secondary.
 218                          */
 219                         mb();
 220                         ccount = READ_ONCE(cpu_start_ccount);
 221                 } while (ccount && time_before(jiffies, timeout));
 222
 223                 if (ccount) {
 224                         smp_call_function_single(0, mx_cpu_stop,
 225                                                  (void *)cpu, 1);
 226                         WRITE_ONCE(cpu_start_ccount, 0);
 227                         return -EIO;
 228                 }
 229         }
 230         return 0;
 231 }
 232
 233 int __cpu_up(unsigned int cpu, struct task_struct *idle)
 234 {
 235         int ret = 0;
 236
 237         if (cpu_asid_cache(cpu) == 0)
 238                 cpu_asid_cache(cpu) = ASID_USER_FIRST;
 239
 240         start_info.stack = (unsigned long)task_pt_regs(idle);
 241         wmb();
 242
 243         pr_debug("%s: Calling wakeup_secondary(cpu:%d, idle:%p, sp: %08lx)\n",
 244                         __func__, cpu, idle, start_info.stack);
 245
 246         init_completion(&cpu_running);
 247         ret = boot_secondary(cpu, idle);
 248         if (ret == 0) {
 249                 wait_for_completion_timeout(&cpu_running,
 250                                 msecs_to_jiffies(1000));
 251                 if (!cpu_online(cpu))
 252                         ret = -EIO;
 253         }
 254
 255         if (ret)
 256                 pr_err("CPU %u failed to boot\n", cpu);
 257
 258         return ret;
 259 }
 260
 261 #ifdef CONFIG_HOTPLUG_CPU
 262
 263 /*
 264  * __cpu_disable runs on the processor to be shutdown.
 265  */
 266 int __cpu_disable(void)
 267 {
 268         unsigned int cpu = smp_processor_id();
 269
 270         /*
 271          * Take this CPU offline.  Once we clear this, we can't return,
 272          * and we must not schedule until we're ready to give up the cpu.
 273          */
 274         set_cpu_online(cpu, false);
 275
 276         /*
 277          * OK - migrate IRQs away from this CPU
 278          */
 279         migrate_irqs();
 280
 281         /*
 282          * Flush user cache and TLB mappings, and then remove this CPU
 283          * from the vm mask set of all processes.
 284          */
 285         local_flush_cache_all();
 286         local_flush_tlb_all();
 287         invalidate_page_directory();
 288
 289         clear_tasks_mm_cpumask(cpu);
 290
 291         return 0;
 292 }
 293
 294 static void platform_cpu_kill(unsigned int cpu)
 295 {
 296         smp_call_function_single(0, mx_cpu_stop, (void *)cpu, true);
 297 }
 298
 299 /*
 300  * called on the thread which is asking for a CPU to be shutdown -
 301  * waits until shutdown has completed, or it is timed out.
 302  */
 303 void __cpu_die(unsigned int cpu)
 304 {
 305         unsigned long timeout = jiffies + msecs_to_jiffies(1000);
 306         while (time_before(jiffies, timeout)) {
 307                 system_invalidate_dcache_range((unsigned long)&cpu_start_id,
 308                                                sizeof(cpu_start_id));
 309                 /* Pairs with the second memw in the cpu_restart */
 310                 mb();
 311                 if (READ_ONCE(cpu_start_id) == -cpu) {
 312                         platform_cpu_kill(cpu);
 313                         return;
 314                 }
 315         }
 316         pr_err("CPU%u: unable to kill\n", cpu);
 317 }
 318
 319 void arch_cpu_idle_dead(void)
 320 {
 321         cpu_die();
 322 }
 323 /*
 324  * Called from the idle thread for the CPU which has been shutdown.
 325  *
 326  * Note that we disable IRQs here, but do not re-enable them
 327  * before returning to the caller. This is also the behaviour
 328  * of the other hotplug-cpu capable cores, so presumably coming
 329  * out of idle fixes this.
 330  */
 331 void __ref cpu_die(void)
 332 {
 333         idle_task_exit();
 334         local_irq_disable();
 335         __asm__ __volatile__(
 336                         "       movi    a2, cpu_restart\n"
 337                         "       jx      a2\n");
 338 }
 339
 340 #endif /* CONFIG_HOTPLUG_CPU */
 341
 342 enum ipi_msg_type {
 343         IPI_RESCHEDULE = 0,
 344         IPI_CALL_FUNC,
 345         IPI_CPU_STOP,
 346         IPI_MAX
 347 };
 348
 349 static const struct {
 350         const char *short_text;
 351         const char *long_text;
 352 } ipi_text[] = {
 353         { .short_text = "RES", .long_text = "Rescheduling interrupts" },
 354         { .short_text = "CAL", .long_text = "Function call interrupts" },
 355         { .short_text = "DIE", .long_text = "CPU shutdown interrupts" },
 356 };
 357
 358 struct ipi_data {
 359         unsigned long ipi_count[IPI_MAX];
 360 };
 361
 362 static DEFINE_PER_CPU(struct ipi_data, ipi_data);
 363
 364 static void send_ipi_message(const struct cpumask *callmask,
 365                 enum ipi_msg_type msg_id)
 366 {
 367         int index;
 368         unsigned long mask = 0;
 369
 370         for_each_cpu(index, callmask)
 371                 mask |= 1 << index;
 372
 373         set_er(mask, MIPISET(msg_id));
 374 }
 375
 376 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 377 {
 378         send_ipi_message(mask, IPI_CALL_FUNC);
 379 }
 380
 381 void arch_send_call_function_single_ipi(int cpu)
 382 {
 383         send_ipi_message(cpumask_of(cpu), IPI_CALL_FUNC);
 384 }
 385
 386 void smp_send_reschedule(int cpu)
 387 {
 388         send_ipi_message(cpumask_of(cpu), IPI_RESCHEDULE);
 389 }
 390
 391 void smp_send_stop(void)
 392 {
 393         struct cpumask targets;
 394
 395         cpumask_copy(&targets, cpu_online_mask);
 396         cpumask_clear_cpu(smp_processor_id(), &targets);
 397         send_ipi_message(&targets, IPI_CPU_STOP);
 398 }
 399
 400 static void ipi_cpu_stop(unsigned int cpu)
 401 {
 402         set_cpu_online(cpu, false);
 403         machine_halt();
 404 }
 405
 406 irqreturn_t ipi_interrupt(int irq, void *dev_id)
 407 {
 408         unsigned int cpu = smp_processor_id();
 409         struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
 410
 411         for (;;) {
 412                 unsigned int msg;
 413
 414                 msg = get_er(MIPICAUSE(cpu));
 415                 set_er(msg, MIPICAUSE(cpu));
 416
 417                 if (!msg)
 418                         break;
 419
 420                 if (msg & (1 << IPI_CALL_FUNC)) {
 421                         ++ipi->ipi_count[IPI_CALL_FUNC];
 422                         generic_smp_call_function_interrupt();
 423                 }
 424
 425                 if (msg & (1 << IPI_RESCHEDULE)) {
 426                         ++ipi->ipi_count[IPI_RESCHEDULE];
 427                         scheduler_ipi();
 428                 }
 429
 430                 if (msg & (1 << IPI_CPU_STOP)) {
 431                         ++ipi->ipi_count[IPI_CPU_STOP];
 432                         ipi_cpu_stop(cpu);
 433                 }
 434         }
 435
 436         return IRQ_HANDLED;
 437 }
 438
 439 void show_ipi_list(struct seq_file *p, int prec)
 440 {
 441         unsigned int cpu;
 442         unsigned i;
 443
 444         for (i = 0; i < IPI_MAX; ++i) {
 445                 seq_printf(p, "%*s:", prec, ipi_text[i].short_text);
 446                 for_each_online_cpu(cpu)
 447                         seq_printf(p, " %10lu",
 448                                         per_cpu(ipi_data, cpu).ipi_count[i]);
 449                 seq_printf(p, "   %s\n", ipi_text[i].long_text);
 450         }
 451 }
 452
 453 int setup_profiling_timer(unsigned int multiplier)
 454 {
 455         pr_debug("setup_profiling_timer %d\n", multiplier);
 456         return 0;
 457 }
 458
 459 /* TLB flush functions */
 460
 461 struct flush_data {
 462         struct vm_area_struct *vma;
 463         unsigned long addr1;
 464         unsigned long addr2;
 465 };
 466
 467 static void ipi_flush_tlb_all(void *arg)
 468 {
 469         local_flush_tlb_all();
 470 }
 471
 472 void flush_tlb_all(void)
 473 {
 474         on_each_cpu(ipi_flush_tlb_all, NULL, 1);
 475 }
 476
 477 static void ipi_flush_tlb_mm(void *arg)
 478 {
 479         local_flush_tlb_mm(arg);
 480 }
 481
 482 void flush_tlb_mm(struct mm_struct *mm)
 483 {
 484         on_each_cpu(ipi_flush_tlb_mm, mm, 1);
 485 }
 486
 487 static void ipi_flush_tlb_page(void *arg)
 488 {
 489         struct flush_data *fd = arg;
 490         local_flush_tlb_page(fd->vma, fd->addr1);
 491 }
 492
 493 void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
 494 {
 495         struct flush_data fd = {
 496                 .vma = vma,
 497                 .addr1 = addr,
 498         };
 499         on_each_cpu(ipi_flush_tlb_page, &fd, 1);
 500 }
 501
 502 static void ipi_flush_tlb_range(void *arg)
 503 {
 504         struct flush_data *fd = arg;
 505         local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
 506 }
 507
 508 void flush_tlb_range(struct vm_area_struct *vma,
 509                      unsigned long start, unsigned long end)
 510 {
 511         struct flush_data fd = {
 512                 .vma = vma,
 513                 .addr1 = start,
 514                 .addr2 = end,
 515         };
 516         on_each_cpu(ipi_flush_tlb_range, &fd, 1);
 517 }
 518
 519 static void ipi_flush_tlb_kernel_range(void *arg)
 520 {
 521         struct flush_data *fd = arg;
 522         local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
 523 }
 524
 525 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 526 {
 527         struct flush_data fd = {
 528                 .addr1 = start,
 529                 .addr2 = end,
 530         };
 531         on_each_cpu(ipi_flush_tlb_kernel_range, &fd, 1);
 532 }
 533
 534 /* Cache flush functions */
 535
 536 static void ipi_flush_cache_all(void *arg)
 537 {
 538         local_flush_cache_all();
 539 }
 540
 541 void flush_cache_all(void)
 542 {
 543         on_each_cpu(ipi_flush_cache_all, NULL, 1);
 544 }
 545
 546 static void ipi_flush_cache_page(void *arg)
 547 {
 548         struct flush_data *fd = arg;
 549         local_flush_cache_page(fd->vma, fd->addr1, fd->addr2);
 550 }
 551
 552 void flush_cache_page(struct vm_area_struct *vma,
 553                      unsigned long address, unsigned long pfn)
 554 {
 555         struct flush_data fd = {
 556                 .vma = vma,
 557                 .addr1 = address,
 558                 .addr2 = pfn,
 559         };
 560         on_each_cpu(ipi_flush_cache_page, &fd, 1);
 561 }
 562
 563 static void ipi_flush_cache_range(void *arg)
 564 {
 565         struct flush_data *fd = arg;
 566         local_flush_cache_range(fd->vma, fd->addr1, fd->addr2);
 567 }
 568
 569 void flush_cache_range(struct vm_area_struct *vma,
 570                      unsigned long start, unsigned long end)
 571 {
 572         struct flush_data fd = {
 573                 .vma = vma,
 574                 .addr1 = start,
 575                 .addr2 = end,
 576         };
 577         on_each_cpu(ipi_flush_cache_range, &fd, 1);
 578 }
 579
 580 static void ipi_flush_icache_range(void *arg)
 581 {
 582         struct flush_data *fd = arg;
 583         local_flush_icache_range(fd->addr1, fd->addr2);
 584 }
 585
 586 void flush_icache_range(unsigned long start, unsigned long end)
 587 {
 588         struct flush_data fd = {
 589                 .addr1 = start,
 590                 .addr2 = end,
 591         };
 592         on_each_cpu(ipi_flush_icache_range, &fd, 1);
 593 }
 594 EXPORT_SYMBOL(flush_icache_range);
 595
 596 /* ------------------------------------------------------------------------- */
 597
 598 static void ipi_invalidate_dcache_range(void *arg)
 599 {
 600         struct flush_data *fd = arg;
 601         __invalidate_dcache_range(fd->addr1, fd->addr2);
 602 }
 603
 604 static void system_invalidate_dcache_range(unsigned long start,
 605                 unsigned long size)
 606 {
 607         struct flush_data fd = {
 608                 .addr1 = start,
 609                 .addr2 = size,
 610         };
 611         on_each_cpu(ipi_invalidate_dcache_range, &fd, 1);
 612 }
 613
 614 static void ipi_flush_invalidate_dcache_range(void *arg)
 615 {
 616         struct flush_data *fd = arg;
 617         __flush_invalidate_dcache_range(fd->addr1, fd->addr2);
 618 }
 619
 620 static void system_flush_invalidate_dcache_range(unsigned long start,
 621                 unsigned long size)
 622 {
 623         struct flush_data fd = {
 624                 .addr1 = start,
 625                 .addr2 = size,
 626         };
 627         on_each_cpu(ipi_flush_invalidate_dcache_range, &fd, 1);
 628 }