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/delay.h> 21#include <linux/init.h> 22#include <linux/spinlock.h> 23#include <linux/sched.h> 24#include <linux/interrupt.h> 25#include <linux/cache.h> 26#include <linux/profile.h> 27#include <linux/errno.h> 28#include <linux/mm.h> 29#include <linux/err.h> 30#include <linux/cpu.h> 31#include <linux/smp.h> 32#include <linux/seq_file.h> 33#include <linux/irq.h> 34#include <linux/percpu.h> 35#include <linux/clockchips.h> 36#include <linux/completion.h> 37#include <linux/of.h> 38#include <linux/irq_work.h> 39 40#include <asm/atomic.h> 41#include <asm/cacheflush.h> 42#include <asm/cpu.h> 43#include <asm/cputype.h> 44#include <asm/cpu_ops.h> 45#include <asm/mmu_context.h> 46#include <asm/pgtable.h> 47#include <asm/pgalloc.h> 48#include <asm/processor.h> 49#include <asm/smp_plat.h> 50#include <asm/sections.h> 51#include <asm/tlbflush.h> 52#include <asm/ptrace.h> 53 54#define CREATE_TRACE_POINTS 55#include <trace/events/ipi.h> 56 57/* 58 * as from 2.5, kernels no longer have an init_tasks structure 59 * so we need some other way of telling a new secondary core 60 * where to place its SVC stack 61 */ 62struct secondary_data secondary_data; 63 64enum ipi_msg_type { 65 IPI_RESCHEDULE, 66 IPI_CALL_FUNC, 67 IPI_CALL_FUNC_SINGLE, 68 IPI_CPU_STOP, 69 IPI_TIMER, 70 IPI_IRQ_WORK, 71}; 72 73/* 74 * Boot a secondary CPU, and assign it the specified idle task. 75 * This also gives us the initial stack to use for this CPU. 76 */ 77static int boot_secondary(unsigned int cpu, struct task_struct *idle) 78{ 79 if (cpu_ops[cpu]->cpu_boot) 80 return cpu_ops[cpu]->cpu_boot(cpu); 81 82 return -EOPNOTSUPP; 83} 84 85static DECLARE_COMPLETION(cpu_running); 86 87int __cpu_up(unsigned int cpu, struct task_struct *idle) 88{ 89 int ret; 90 91 /* 92 * We need to tell the secondary core where to find its stack and the 93 * page tables. 94 */ 95 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP; 96 __flush_dcache_area(&secondary_data, sizeof(secondary_data)); 97 98 /* 99 * Now bring the CPU into our world. 100 */ 101 ret = boot_secondary(cpu, idle); 102 if (ret == 0) { 103 /* 104 * CPU was successfully started, wait for it to come online or 105 * time out. 106 */ 107 wait_for_completion_timeout(&cpu_running, 108 msecs_to_jiffies(1000)); 109 110 if (!cpu_online(cpu)) { 111 pr_crit("CPU%u: failed to come online\n", cpu); 112 ret = -EIO; 113 } 114 } else { 115 pr_err("CPU%u: failed to boot: %d\n", cpu, ret); 116 } 117 118 secondary_data.stack = NULL; 119 120 return ret; 121} 122 123static void smp_store_cpu_info(unsigned int cpuid) 124{ 125 store_cpu_topology(cpuid); 126} 127 128/* 129 * This is the secondary CPU boot entry. We're using this CPUs 130 * idle thread stack, but a set of temporary page tables. 131 */ 132asmlinkage void secondary_start_kernel(void) 133{ 134 struct mm_struct *mm = &init_mm; 135 unsigned int cpu = smp_processor_id(); 136 137 /* 138 * All kernel threads share the same mm context; grab a 139 * reference and switch to it. 140 */ 141 atomic_inc(&mm->mm_count); 142 current->active_mm = mm; 143 cpumask_set_cpu(cpu, mm_cpumask(mm)); 144 145 set_my_cpu_offset(per_cpu_offset(smp_processor_id())); 146 printk("CPU%u: Booted secondary processor\n", cpu); 147 148 /* 149 * TTBR0 is only used for the identity mapping at this stage. Make it 150 * point to zero page to avoid speculatively fetching new entries. 151 */ 152 cpu_set_reserved_ttbr0(); 153 flush_tlb_all(); 154 155 preempt_disable(); 156 trace_hardirqs_off(); 157 158 if (cpu_ops[cpu]->cpu_postboot) 159 cpu_ops[cpu]->cpu_postboot(); 160 161 /* 162 * Log the CPU info before it is marked online and might get read. 163 */ 164 cpuinfo_store_cpu(); 165 166 /* 167 * Enable GIC and timers. 168 */ 169 notify_cpu_starting(cpu); 170 171 smp_store_cpu_info(cpu); 172 173 /* 174 * OK, now it's safe to let the boot CPU continue. Wait for 175 * the CPU migration code to notice that the CPU is online 176 * before we continue. 177 */ 178 set_cpu_online(cpu, true); 179 complete(&cpu_running); 180 181 local_dbg_enable(); 182 local_irq_enable(); 183 local_async_enable(); 184 185 /* 186 * OK, it's off to the idle thread for us 187 */ 188 cpu_startup_entry(CPUHP_ONLINE); 189} 190 191#ifdef CONFIG_HOTPLUG_CPU 192static int op_cpu_disable(unsigned int cpu) 193{ 194 /* 195 * If we don't have a cpu_die method, abort before we reach the point 196 * of no return. CPU0 may not have an cpu_ops, so test for it. 197 */ 198 if (!cpu_ops[cpu] || !cpu_ops[cpu]->cpu_die) 199 return -EOPNOTSUPP; 200 201 /* 202 * We may need to abort a hot unplug for some other mechanism-specific 203 * reason. 204 */ 205 if (cpu_ops[cpu]->cpu_disable) 206 return cpu_ops[cpu]->cpu_disable(cpu); 207 208 return 0; 209} 210 211/* 212 * __cpu_disable runs on the processor to be shutdown. 213 */ 214int __cpu_disable(void) 215{ 216 unsigned int cpu = smp_processor_id(); 217 int ret; 218 219 ret = op_cpu_disable(cpu); 220 if (ret) 221 return ret; 222 223 /* 224 * Take this CPU offline. Once we clear this, we can't return, 225 * and we must not schedule until we're ready to give up the cpu. 226 */ 227 set_cpu_online(cpu, false); 228 229 /* 230 * OK - migrate IRQs away from this CPU 231 */ 232 migrate_irqs(); 233 234 /* 235 * Remove this CPU from the vm mask set of all processes. 236 */ 237 clear_tasks_mm_cpumask(cpu); 238 239 return 0; 240} 241 242static int op_cpu_kill(unsigned int cpu) 243{ 244 /* 245 * If we have no means of synchronising with the dying CPU, then assume 246 * that it is really dead. We can only wait for an arbitrary length of 247 * time and hope that it's dead, so let's skip the wait and just hope. 248 */ 249 if (!cpu_ops[cpu]->cpu_kill) 250 return 1; 251 252 return cpu_ops[cpu]->cpu_kill(cpu); 253} 254 255static DECLARE_COMPLETION(cpu_died); 256 257/* 258 * called on the thread which is asking for a CPU to be shutdown - 259 * waits until shutdown has completed, or it is timed out. 260 */ 261void __cpu_die(unsigned int cpu) 262{ 263 if (!wait_for_completion_timeout(&cpu_died, msecs_to_jiffies(5000))) { 264 pr_crit("CPU%u: cpu didn't die\n", cpu); 265 return; 266 } 267 pr_notice("CPU%u: shutdown\n", cpu); 268 269 /* 270 * Now that the dying CPU is beyond the point of no return w.r.t. 271 * in-kernel synchronisation, try to get the firwmare to help us to 272 * verify that it has really left the kernel before we consider 273 * clobbering anything it might still be using. 274 */ 275 if (!op_cpu_kill(cpu)) 276 pr_warn("CPU%d may not have shut down cleanly\n", cpu); 277} 278 279/* 280 * Called from the idle thread for the CPU which has been shutdown. 281 * 282 * Note that we disable IRQs here, but do not re-enable them 283 * before returning to the caller. This is also the behaviour 284 * of the other hotplug-cpu capable cores, so presumably coming 285 * out of idle fixes this. 286 */ 287void cpu_die(void) 288{ 289 unsigned int cpu = smp_processor_id(); 290 291 idle_task_exit(); 292 293 local_irq_disable(); 294 295 /* Tell __cpu_die() that this CPU is now safe to dispose of */ 296 complete(&cpu_died); 297 298 /* 299 * Actually shutdown the CPU. This must never fail. The specific hotplug 300 * mechanism must perform all required cache maintenance to ensure that 301 * no dirty lines are lost in the process of shutting down the CPU. 302 */ 303 cpu_ops[cpu]->cpu_die(cpu); 304 305 BUG(); 306} 307#endif 308 309void __init smp_cpus_done(unsigned int max_cpus) 310{ 311 pr_info("SMP: Total of %d processors activated.\n", num_online_cpus()); 312} 313 314void __init smp_prepare_boot_cpu(void) 315{ 316 set_my_cpu_offset(per_cpu_offset(smp_processor_id())); 317} 318 319/* 320 * Enumerate the possible CPU set from the device tree and build the 321 * cpu logical map array containing MPIDR values related to logical 322 * cpus. Assumes that cpu_logical_map(0) has already been initialized. 323 */ 324void __init smp_init_cpus(void) 325{ 326 struct device_node *dn = NULL; 327 unsigned int i, cpu = 1; 328 bool bootcpu_valid = false; 329 330 while ((dn = of_find_node_by_type(dn, "cpu"))) { 331 const u32 *cell; 332 u64 hwid; 333 334 /* 335 * A cpu node with missing "reg" property is 336 * considered invalid to build a cpu_logical_map 337 * entry. 338 */ 339 cell = of_get_property(dn, "reg", NULL); 340 if (!cell) { 341 pr_err("%s: missing reg property\n", dn->full_name); 342 goto next; 343 } 344 hwid = of_read_number(cell, of_n_addr_cells(dn)); 345 346 /* 347 * Non affinity bits must be set to 0 in the DT 348 */ 349 if (hwid & ~MPIDR_HWID_BITMASK) { 350 pr_err("%s: invalid reg property\n", dn->full_name); 351 goto next; 352 } 353 354 /* 355 * Duplicate MPIDRs are a recipe for disaster. Scan 356 * all initialized entries and check for 357 * duplicates. If any is found just ignore the cpu. 358 * cpu_logical_map was initialized to INVALID_HWID to 359 * avoid matching valid MPIDR values. 360 */ 361 for (i = 1; (i < cpu) && (i < NR_CPUS); i++) { 362 if (cpu_logical_map(i) == hwid) { 363 pr_err("%s: duplicate cpu reg properties in the DT\n", 364 dn->full_name); 365 goto next; 366 } 367 } 368 369 /* 370 * The numbering scheme requires that the boot CPU 371 * must be assigned logical id 0. Record it so that 372 * the logical map built from DT is validated and can 373 * be used. 374 */ 375 if (hwid == cpu_logical_map(0)) { 376 if (bootcpu_valid) { 377 pr_err("%s: duplicate boot cpu reg property in DT\n", 378 dn->full_name); 379 goto next; 380 } 381 382 bootcpu_valid = true; 383 384 /* 385 * cpu_logical_map has already been 386 * initialized and the boot cpu doesn't need 387 * the enable-method so continue without 388 * incrementing cpu. 389 */ 390 continue; 391 } 392 393 if (cpu >= NR_CPUS) 394 goto next; 395 396 if (cpu_read_ops(dn, cpu) != 0) 397 goto next; 398 399 if (cpu_ops[cpu]->cpu_init(dn, cpu)) 400 goto next; 401 402 pr_debug("cpu logical map 0x%llx\n", hwid); 403 cpu_logical_map(cpu) = hwid; 404next: 405 cpu++; 406 } 407 408 /* sanity check */ 409 if (cpu > NR_CPUS) 410 pr_warning("no. of cores (%d) greater than configured maximum of %d - clipping\n", 411 cpu, NR_CPUS); 412 413 if (!bootcpu_valid) { 414 pr_err("DT missing boot CPU MPIDR, not enabling secondaries\n"); 415 return; 416 } 417 418 /* 419 * All the cpus that made it to the cpu_logical_map have been 420 * validated so set them as possible cpus. 421 */ 422 for (i = 0; i < NR_CPUS; i++) 423 if (cpu_logical_map(i) != INVALID_HWID) 424 set_cpu_possible(i, true); 425} 426 427void __init smp_prepare_cpus(unsigned int max_cpus) 428{ 429 int err; 430 unsigned int cpu, ncores = num_possible_cpus(); 431 432 init_cpu_topology(); 433 434 smp_store_cpu_info(smp_processor_id()); 435 436 /* 437 * are we trying to boot more cores than exist? 438 */ 439 if (max_cpus > ncores) 440 max_cpus = ncores; 441 442 /* Don't bother if we're effectively UP */ 443 if (max_cpus <= 1) 444 return; 445 446 /* 447 * Initialise the present map (which describes the set of CPUs 448 * actually populated at the present time) and release the 449 * secondaries from the bootloader. 450 * 451 * Make sure we online at most (max_cpus - 1) additional CPUs. 452 */ 453 max_cpus--; 454 for_each_possible_cpu(cpu) { 455 if (max_cpus == 0) 456 break; 457 458 if (cpu == smp_processor_id()) 459 continue; 460 461 if (!cpu_ops[cpu]) 462 continue; 463 464 err = cpu_ops[cpu]->cpu_prepare(cpu); 465 if (err) 466 continue; 467 468 set_cpu_present(cpu, true); 469 max_cpus--; 470 } 471} 472 473void (*__smp_cross_call)(const struct cpumask *, unsigned int); 474 475void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int)) 476{ 477 __smp_cross_call = fn; 478} 479 480static const char *ipi_types[NR_IPI] __tracepoint_string = { 481#define S(x,s) [x] = s 482 S(IPI_RESCHEDULE, "Rescheduling interrupts"), 483 S(IPI_CALL_FUNC, "Function call interrupts"), 484 S(IPI_CALL_FUNC_SINGLE, "Single function call interrupts"), 485 S(IPI_CPU_STOP, "CPU stop interrupts"), 486 S(IPI_TIMER, "Timer broadcast interrupts"), 487 S(IPI_IRQ_WORK, "IRQ work interrupts"), 488}; 489 490static void smp_cross_call(const struct cpumask *target, unsigned int ipinr) 491{ 492 trace_ipi_raise(target, ipi_types[ipinr]); 493 __smp_cross_call(target, ipinr); 494} 495 496void show_ipi_list(struct seq_file *p, int prec) 497{ 498 unsigned int cpu, i; 499 500 for (i = 0; i < NR_IPI; i++) { 501 seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i, 502 prec >= 4 ? " " : ""); 503 for_each_online_cpu(cpu) 504 seq_printf(p, "%10u ", 505 __get_irq_stat(cpu, ipi_irqs[i])); 506 seq_printf(p, " %s\n", ipi_types[i]); 507 } 508} 509 510u64 smp_irq_stat_cpu(unsigned int cpu) 511{ 512 u64 sum = 0; 513 int i; 514 515 for (i = 0; i < NR_IPI; i++) 516 sum += __get_irq_stat(cpu, ipi_irqs[i]); 517 518 return sum; 519} 520 521void arch_send_call_function_ipi_mask(const struct cpumask *mask) 522{ 523 smp_cross_call(mask, IPI_CALL_FUNC); 524} 525 526void arch_send_call_function_single_ipi(int cpu) 527{ 528 smp_cross_call(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE); 529} 530 531#ifdef CONFIG_IRQ_WORK 532void arch_irq_work_raise(void) 533{ 534 if (__smp_cross_call) 535 smp_cross_call(cpumask_of(smp_processor_id()), IPI_IRQ_WORK); 536} 537#endif 538 539static DEFINE_RAW_SPINLOCK(stop_lock); 540 541/* 542 * ipi_cpu_stop - handle IPI from smp_send_stop() 543 */ 544static void ipi_cpu_stop(unsigned int cpu) 545{ 546 if (system_state == SYSTEM_BOOTING || 547 system_state == SYSTEM_RUNNING) { 548 raw_spin_lock(&stop_lock); 549 pr_crit("CPU%u: stopping\n", cpu); 550 dump_stack(); 551 raw_spin_unlock(&stop_lock); 552 } 553 554 set_cpu_online(cpu, false); 555 556 local_irq_disable(); 557 558 while (1) 559 cpu_relax(); 560} 561 562/* 563 * Main handler for inter-processor interrupts 564 */ 565void handle_IPI(int ipinr, struct pt_regs *regs) 566{ 567 unsigned int cpu = smp_processor_id(); 568 struct pt_regs *old_regs = set_irq_regs(regs); 569 570 if ((unsigned)ipinr < NR_IPI) { 571 trace_ipi_entry(ipi_types[ipinr]); 572 __inc_irq_stat(cpu, ipi_irqs[ipinr]); 573 } 574 575 switch (ipinr) { 576 case IPI_RESCHEDULE: 577 scheduler_ipi(); 578 break; 579 580 case IPI_CALL_FUNC: 581 irq_enter(); 582 generic_smp_call_function_interrupt(); 583 irq_exit(); 584 break; 585 586 case IPI_CALL_FUNC_SINGLE: 587 irq_enter(); 588 generic_smp_call_function_single_interrupt(); 589 irq_exit(); 590 break; 591 592 case IPI_CPU_STOP: 593 irq_enter(); 594 ipi_cpu_stop(cpu); 595 irq_exit(); 596 break; 597 598#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST 599 case IPI_TIMER: 600 irq_enter(); 601 tick_receive_broadcast(); 602 irq_exit(); 603 break; 604#endif 605 606#ifdef CONFIG_IRQ_WORK 607 case IPI_IRQ_WORK: 608 irq_enter(); 609 irq_work_run(); 610 irq_exit(); 611 break; 612#endif 613 614 default: 615 pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr); 616 break; 617 } 618 619 if ((unsigned)ipinr < NR_IPI) 620 trace_ipi_exit(ipi_types[ipinr]); 621 set_irq_regs(old_regs); 622} 623 624void smp_send_reschedule(int cpu) 625{ 626 smp_cross_call(cpumask_of(cpu), IPI_RESCHEDULE); 627} 628 629#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST 630void tick_broadcast(const struct cpumask *mask) 631{ 632 smp_cross_call(mask, IPI_TIMER); 633} 634#endif 635 636void smp_send_stop(void) 637{ 638 unsigned long timeout; 639 640 if (num_online_cpus() > 1) { 641 cpumask_t mask; 642 643 cpumask_copy(&mask, cpu_online_mask); 644 cpu_clear(smp_processor_id(), mask); 645 646 smp_cross_call(&mask, IPI_CPU_STOP); 647 } 648 649 /* Wait up to one second for other CPUs to stop */ 650 timeout = USEC_PER_SEC; 651 while (num_online_cpus() > 1 && timeout--) 652 udelay(1); 653 654 if (num_online_cpus() > 1) 655 pr_warning("SMP: failed to stop secondary CPUs\n"); 656} 657 658/* 659 * not supported here 660 */ 661int setup_profiling_timer(unsigned int multiplier) 662{ 663 return -EINVAL; 664} 665