arch/mips/kernel/smp.c

   1 /*
   2  * This program is free software; you can redistribute it and/or
   3  * modify it under the terms of the GNU General Public License
   4  * as published by the Free Software Foundation; either version 2
   5  * of the License, or (at your option) any later version.
   6  *
   7  * This program is distributed in the hope that it will be useful,
   8  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  10  * GNU General Public License for more details.
  11  *
  12  * You should have received a copy of the GNU General Public License
  13  * along with this program; if not, write to the Free Software
  14  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  15  *
  16  * Copyright (C) 2000, 2001 Kanoj Sarcar
  17  * Copyright (C) 2000, 2001 Ralf Baechle
  18  * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
  19  * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
  20  */
  21 #include <linux/cache.h>
  22 #include <linux/delay.h>
  23 #include <linux/init.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/smp.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/threads.h>
  28 #include <linux/module.h>
  29 #include <linux/time.h>
  30 #include <linux/timex.h>
  31 #include <linux/sched.h>
  32 #include <linux/cpumask.h>
  33 #include <linux/cpu.h>
  34 #include <linux/err.h>
  35 #include <linux/ftrace.h>
  36
  37 #include <linux/atomic.h>
  38 #include <asm/cpu.h>
  39 #include <asm/processor.h>
  40 #include <asm/idle.h>
  41 #include <asm/r4k-timer.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/time.h>
  44 #include <asm/setup.h>
  45
  46 cpumask_t cpu_callin_map;               /* Bitmask of started secondaries */
  47
  48 int __cpu_number_map[NR_CPUS];          /* Map physical to logical */
  49 EXPORT_SYMBOL(__cpu_number_map);
  50
  51 int __cpu_logical_map[NR_CPUS];         /* Map logical to physical */
  52 EXPORT_SYMBOL(__cpu_logical_map);
  53
  54 /* Number of TCs (or siblings in Intel speak) per CPU core */
  55 int smp_num_siblings = 1;
  56 EXPORT_SYMBOL(smp_num_siblings);
  57
  58 /* representing the TCs (or siblings in Intel speak) of each logical CPU */
  59 cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
  60 EXPORT_SYMBOL(cpu_sibling_map);
  61
  62 /* representing the core map of multi-core chips of each logical CPU */
  63 cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
  64 EXPORT_SYMBOL(cpu_core_map);
  65
  66 /*
  67  * A logcal cpu mask containing only one VPE per core to
  68  * reduce the number of IPIs on large MT systems.
  69  */
  70 cpumask_t cpu_foreign_map __read_mostly;
  71 EXPORT_SYMBOL(cpu_foreign_map);
  72
  73 /* representing cpus for which sibling maps can be computed */
  74 static cpumask_t cpu_sibling_setup_map;
  75
  76 /* representing cpus for which core maps can be computed */
  77 static cpumask_t cpu_core_setup_map;
  78
  79 cpumask_t cpu_coherent_mask;
  80
  81 static inline void set_cpu_sibling_map(int cpu)
  82 {
  83         int i;
  84
  85         cpumask_set_cpu(cpu, &cpu_sibling_setup_map);
  86
  87         if (smp_num_siblings > 1) {
  88                 for_each_cpu(i, &cpu_sibling_setup_map) {
  89                         if (cpu_data[cpu].package == cpu_data[i].package &&
  90                                     cpu_data[cpu].core == cpu_data[i].core) {
  91                                 cpumask_set_cpu(i, &cpu_sibling_map[cpu]);
  92                                 cpumask_set_cpu(cpu, &cpu_sibling_map[i]);
  93                         }
  94                 }
  95         } else
  96                 cpumask_set_cpu(cpu, &cpu_sibling_map[cpu]);
  97 }
  98
  99 static inline void set_cpu_core_map(int cpu)
 100 {
 101         int i;
 102
 103         cpumask_set_cpu(cpu, &cpu_core_setup_map);
 104
 105         for_each_cpu(i, &cpu_core_setup_map) {
 106                 if (cpu_data[cpu].package == cpu_data[i].package) {
 107                         cpumask_set_cpu(i, &cpu_core_map[cpu]);
 108                         cpumask_set_cpu(cpu, &cpu_core_map[i]);
 109                 }
 110         }
 111 }
 112
 113 /*
 114  * Calculate a new cpu_foreign_map mask whenever a
 115  * new cpu appears or disappears.
 116  */
 117 static inline void calculate_cpu_foreign_map(void)
 118 {
 119         int i, k, core_present;
 120         cpumask_t temp_foreign_map;
 121
 122         /* Re-calculate the mask */
 123         for_each_online_cpu(i) {
 124                 core_present = 0;
 125                 for_each_cpu(k, &temp_foreign_map)
 126                         if (cpu_data[i].package == cpu_data[k].package &&
 127                             cpu_data[i].core == cpu_data[k].core)
 128                                 core_present = 1;
 129                 if (!core_present)
 130                         cpumask_set_cpu(i, &temp_foreign_map);
 131         }
 132
 133         cpumask_copy(&cpu_foreign_map, &temp_foreign_map);
 134 }
 135
 136 struct plat_smp_ops *mp_ops;
 137 EXPORT_SYMBOL(mp_ops);
 138
 139 void register_smp_ops(struct plat_smp_ops *ops)
 140 {
 141         if (mp_ops)
 142                 printk(KERN_WARNING "Overriding previously set SMP ops\n");
 143
 144         mp_ops = ops;
 145 }
 146
 147 /*
 148  * First C code run on the secondary CPUs after being started up by
 149  * the master.
 150  */
 151 asmlinkage void start_secondary(void)
 152 {
 153         unsigned int cpu;
 154
 155         cpu_probe();
 156         per_cpu_trap_init(false);
 157         mips_clockevent_init();
 158         mp_ops->init_secondary();
 159         cpu_report();
 160
 161         /*
 162          * XXX parity protection should be folded in here when it's converted
 163          * to an option instead of something based on .cputype
 164          */
 165
 166         calibrate_delay();
 167         preempt_disable();
 168         cpu = smp_processor_id();
 169         cpu_data[cpu].udelay_val = loops_per_jiffy;
 170
 171         cpumask_set_cpu(cpu, &cpu_coherent_mask);
 172         notify_cpu_starting(cpu);
 173
 174         set_cpu_online(cpu, true);
 175
 176         set_cpu_sibling_map(cpu);
 177         set_cpu_core_map(cpu);
 178
 179         calculate_cpu_foreign_map();
 180
 181         cpumask_set_cpu(cpu, &cpu_callin_map);
 182
 183         synchronise_count_slave(cpu);
 184
 185         /*
 186          * irq will be enabled in ->smp_finish(), enabling it too early
 187          * is dangerous.
 188          */
 189         WARN_ON_ONCE(!irqs_disabled());
 190         mp_ops->smp_finish();
 191
 192         cpu_startup_entry(CPUHP_ONLINE);
 193 }
 194
 195 static void stop_this_cpu(void *dummy)
 196 {
 197         /*
 198          * Remove this CPU. Be a bit slow here and
 199          * set the bits for every online CPU so we don't miss
 200          * any IPI whilst taking this VPE down.
 201          */
 202
 203         cpumask_copy(&cpu_foreign_map, cpu_online_mask);
 204
 205         /* Make it visible to every other CPU */
 206         smp_mb();
 207
 208         set_cpu_online(smp_processor_id(), false);
 209         calculate_cpu_foreign_map();
 210         local_irq_disable();
 211         while (1);
 212 }
 213
 214 void smp_send_stop(void)
 215 {
 216         smp_call_function(stop_this_cpu, NULL, 0);
 217 }
 218
 219 void __init smp_cpus_done(unsigned int max_cpus)
 220 {
 221 }
 222
 223 /* called from main before smp_init() */
 224 void __init smp_prepare_cpus(unsigned int max_cpus)
 225 {
 226         init_new_context(current, &init_mm);
 227         current_thread_info()->cpu = 0;
 228         mp_ops->prepare_cpus(max_cpus);
 229         set_cpu_sibling_map(0);
 230         set_cpu_core_map(0);
 231         calculate_cpu_foreign_map();
 232 #ifndef CONFIG_HOTPLUG_CPU
 233         init_cpu_present(cpu_possible_mask);
 234 #endif
 235         cpumask_copy(&cpu_coherent_mask, cpu_possible_mask);
 236 }
 237
 238 /* preload SMP state for boot cpu */
 239 void smp_prepare_boot_cpu(void)
 240 {
 241         set_cpu_possible(0, true);
 242         set_cpu_online(0, true);
 243         cpumask_set_cpu(0, &cpu_callin_map);
 244 }
 245
 246 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 247 {
 248         mp_ops->boot_secondary(cpu, tidle);
 249
 250         /*
 251          * Trust is futile.  We should really have timeouts ...
 252          */
 253         while (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
 254                 udelay(100);
 255                 schedule();
 256         }
 257
 258         synchronise_count_master(cpu);
 259         return 0;
 260 }
 261
 262 /* Not really SMP stuff ... */
 263 int setup_profiling_timer(unsigned int multiplier)
 264 {
 265         return 0;
 266 }
 267
 268 static void flush_tlb_all_ipi(void *info)
 269 {
 270         local_flush_tlb_all();
 271 }
 272
 273 void flush_tlb_all(void)
 274 {
 275         on_each_cpu(flush_tlb_all_ipi, NULL, 1);
 276 }
 277
 278 static void flush_tlb_mm_ipi(void *mm)
 279 {
 280         local_flush_tlb_mm((struct mm_struct *)mm);
 281 }
 282
 283 /*
 284  * Special Variant of smp_call_function for use by TLB functions:
 285  *
 286  *  o No return value
 287  *  o collapses to normal function call on UP kernels
 288  *  o collapses to normal function call on systems with a single shared
 289  *    primary cache.
 290  */
 291 static inline void smp_on_other_tlbs(void (*func) (void *info), void *info)
 292 {
 293         smp_call_function(func, info, 1);
 294 }
 295
 296 static inline void smp_on_each_tlb(void (*func) (void *info), void *info)
 297 {
 298         preempt_disable();
 299
 300         smp_on_other_tlbs(func, info);
 301         func(info);
 302
 303         preempt_enable();
 304 }
 305
 306 /*
 307  * The following tlb flush calls are invoked when old translations are
 308  * being torn down, or pte attributes are changing. For single threaded
 309  * address spaces, a new context is obtained on the current cpu, and tlb
 310  * context on other cpus are invalidated to force a new context allocation
 311  * at switch_mm time, should the mm ever be used on other cpus. For
 312  * multithreaded address spaces, intercpu interrupts have to be sent.
 313  * Another case where intercpu interrupts are required is when the target
 314  * mm might be active on another cpu (eg debuggers doing the flushes on
 315  * behalf of debugees, kswapd stealing pages from another process etc).
 316  * Kanoj 07/00.
 317  */
 318
 319 void flush_tlb_mm(struct mm_struct *mm)
 320 {
 321         preempt_disable();
 322
 323         if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
 324                 smp_on_other_tlbs(flush_tlb_mm_ipi, mm);
 325         } else {
 326                 unsigned int cpu;
 327
 328                 for_each_online_cpu(cpu) {
 329                         if (cpu != smp_processor_id() && cpu_context(cpu, mm))
 330                                 cpu_context(cpu, mm) = 0;
 331                 }
 332         }
 333         local_flush_tlb_mm(mm);
 334
 335         preempt_enable();
 336 }
 337
 338 struct flush_tlb_data {
 339         struct vm_area_struct *vma;
 340         unsigned long addr1;
 341         unsigned long addr2;
 342 };
 343
 344 static void flush_tlb_range_ipi(void *info)
 345 {
 346         struct flush_tlb_data *fd = info;
 347
 348         local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
 349 }
 350
 351 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
 352 {
 353         struct mm_struct *mm = vma->vm_mm;
 354
 355         preempt_disable();
 356         if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
 357                 struct flush_tlb_data fd = {
 358                         .vma = vma,
 359                         .addr1 = start,
 360                         .addr2 = end,
 361                 };
 362
 363                 smp_on_other_tlbs(flush_tlb_range_ipi, &fd);
 364         } else {
 365                 unsigned int cpu;
 366
 367                 for_each_online_cpu(cpu) {
 368                         if (cpu != smp_processor_id() && cpu_context(cpu, mm))
 369                                 cpu_context(cpu, mm) = 0;
 370                 }
 371         }
 372         local_flush_tlb_range(vma, start, end);
 373         preempt_enable();
 374 }
 375
 376 static void flush_tlb_kernel_range_ipi(void *info)
 377 {
 378         struct flush_tlb_data *fd = info;
 379
 380         local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
 381 }
 382
 383 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 384 {
 385         struct flush_tlb_data fd = {
 386                 .addr1 = start,
 387                 .addr2 = end,
 388         };
 389
 390         on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1);
 391 }
 392
 393 static void flush_tlb_page_ipi(void *info)
 394 {
 395         struct flush_tlb_data *fd = info;
 396
 397         local_flush_tlb_page(fd->vma, fd->addr1);
 398 }
 399
 400 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 401 {
 402         preempt_disable();
 403         if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
 404                 struct flush_tlb_data fd = {
 405                         .vma = vma,
 406                         .addr1 = page,
 407                 };
 408
 409                 smp_on_other_tlbs(flush_tlb_page_ipi, &fd);
 410         } else {
 411                 unsigned int cpu;
 412
 413                 for_each_online_cpu(cpu) {
 414                         if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
 415                                 cpu_context(cpu, vma->vm_mm) = 0;
 416                 }
 417         }
 418         local_flush_tlb_page(vma, page);
 419         preempt_enable();
 420 }
 421
 422 static void flush_tlb_one_ipi(void *info)
 423 {
 424         unsigned long vaddr = (unsigned long) info;
 425
 426         local_flush_tlb_one(vaddr);
 427 }
 428
 429 void flush_tlb_one(unsigned long vaddr)
 430 {
 431         smp_on_each_tlb(flush_tlb_one_ipi, (void *) vaddr);
 432 }
 433
 434 EXPORT_SYMBOL(flush_tlb_page);
 435 EXPORT_SYMBOL(flush_tlb_one);
 436
 437 #if defined(CONFIG_KEXEC)
 438 void (*dump_ipi_function_ptr)(void *) = NULL;
 439 void dump_send_ipi(void (*dump_ipi_callback)(void *))
 440 {
 441         int i;
 442         int cpu = smp_processor_id();
 443
 444         dump_ipi_function_ptr = dump_ipi_callback;
 445         smp_mb();
 446         for_each_online_cpu(i)
 447                 if (i != cpu)
 448                         mp_ops->send_ipi_single(i, SMP_DUMP);
 449
 450 }
 451 EXPORT_SYMBOL(dump_send_ipi);
 452 #endif
 453
 454 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 455
 456 static DEFINE_PER_CPU(atomic_t, tick_broadcast_count);
 457 static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd);
 458
 459 void tick_broadcast(const struct cpumask *mask)
 460 {
 461         atomic_t *count;
 462         struct call_single_data *csd;
 463         int cpu;
 464
 465         for_each_cpu(cpu, mask) {
 466                 count = &per_cpu(tick_broadcast_count, cpu);
 467                 csd = &per_cpu(tick_broadcast_csd, cpu);
 468
 469                 if (atomic_inc_return(count) == 1)
 470                         smp_call_function_single_async(cpu, csd);
 471         }
 472 }
 473
 474 static void tick_broadcast_callee(void *info)
 475 {
 476         int cpu = smp_processor_id();
 477         tick_receive_broadcast();
 478         atomic_set(&per_cpu(tick_broadcast_count, cpu), 0);
 479 }
 480
 481 static int __init tick_broadcast_init(void)
 482 {
 483         struct call_single_data *csd;
 484         int cpu;
 485
 486         for (cpu = 0; cpu < NR_CPUS; cpu++) {
 487                 csd = &per_cpu(tick_broadcast_csd, cpu);
 488                 csd->func = tick_broadcast_callee;
 489         }
 490
 491         return 0;
 492 }
 493 early_initcall(tick_broadcast_init);
 494
 495 #endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */