SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999,2007
   5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
   6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
   7  *               Heiko Carstens (heiko.carstens@de.ibm.com)
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * We work with logical cpu numbering everywhere we can. The only
  14  * functions using the real cpu address (got from STAP) are the sigp
  15  * functions. For all other functions we use the identity mapping.
  16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
  17  * used e.g. to find the idle task belonging to a logical cpu. Every array
  18  * in the kernel is sorted by the logical cpu number and not by the physical
  19  * one which is causing all the confusion with __cpu_logical_map and
  20  * cpu_number_map in other architectures.
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/init.h>
  25 #include <linux/mm.h>
  26 #include <linux/err.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/kernel_stat.h>
  29 #include <linux/delay.h>
  30 #include <linux/cache.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/cpu.h>
  33 #include <linux/timex.h>
  34 #include <linux/bootmem.h>
  35 #include <asm/ipl.h>
  36 #include <asm/setup.h>
  37 #include <asm/sigp.h>
  38 #include <asm/pgalloc.h>
  39 #include <asm/irq.h>
  40 #include <asm/s390_ext.h>
  41 #include <asm/cpcmd.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/timer.h>
  44 #include <asm/lowcore.h>
  45 #include <asm/sclp.h>
  46 #include <asm/cpu.h>
  47
  48 /*
  49  * An array with a pointer the lowcore of every CPU.
  50  */
  51 struct _lowcore *lowcore_ptr[NR_CPUS];
  52 EXPORT_SYMBOL(lowcore_ptr);
  53
  54 cpumask_t cpu_online_map = CPU_MASK_NONE;
  55 EXPORT_SYMBOL(cpu_online_map);
  56
  57 cpumask_t cpu_possible_map = CPU_MASK_ALL;
  58 EXPORT_SYMBOL(cpu_possible_map);
  59
  60 static struct task_struct *current_set[NR_CPUS];
  61
  62 static u8 smp_cpu_type;
  63 static int smp_use_sigp_detection;
  64
  65 enum s390_cpu_state {
  66         CPU_STATE_STANDBY,
  67         CPU_STATE_CONFIGURED,
  68 };
  69
  70 #ifdef CONFIG_HOTPLUG_CPU
  71 static DEFINE_MUTEX(smp_cpu_state_mutex);
  72 #endif
  73 static int smp_cpu_state[NR_CPUS];
  74
  75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  76 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
  77
  78 static void smp_ext_bitcall(int, ec_bit_sig);
  79
  80 /*
  81  * Structure and data for __smp_call_function_map(). This is designed to
  82  * minimise static memory requirements. It also looks cleaner.
  83  */
  84 static DEFINE_SPINLOCK(call_lock);
  85
  86 struct call_data_struct {
  87         void (*func) (void *info);
  88         void *info;
  89         cpumask_t started;
  90         cpumask_t finished;
  91         int wait;
  92 };
  93
  94 static struct call_data_struct *call_data;
  95
  96 /*
  97  * 'Call function' interrupt callback
  98  */
  99 static void do_call_function(void)
 100 {
 101         void (*func) (void *info) = call_data->func;
 102         void *info = call_data->info;
 103         int wait = call_data->wait;
 104
 105         cpu_set(smp_processor_id(), call_data->started);
 106         (*func)(info);
 107         if (wait)
 108                 cpu_set(smp_processor_id(), call_data->finished);;
 109 }
 110
 111 static void __smp_call_function_map(void (*func) (void *info), void *info,
 112                                     int nonatomic, int wait, cpumask_t map)
 113 {
 114         struct call_data_struct data;
 115         int cpu, local = 0;
 116
 117         /*
 118          * Can deadlock when interrupts are disabled or if in wrong context.
 119          */
 120         WARN_ON(irqs_disabled() || in_irq());
 121
 122         /*
 123          * Check for local function call. We have to have the same call order
 124          * as in on_each_cpu() because of machine_restart_smp().
 125          */
 126         if (cpu_isset(smp_processor_id(), map)) {
 127                 local = 1;
 128                 cpu_clear(smp_processor_id(), map);
 129         }
 130
 131         cpus_and(map, map, cpu_online_map);
 132         if (cpus_empty(map))
 133                 goto out;
 134
 135         data.func = func;
 136         data.info = info;
 137         data.started = CPU_MASK_NONE;
 138         data.wait = wait;
 139         if (wait)
 140                 data.finished = CPU_MASK_NONE;
 141
 142         spin_lock(&call_lock);
 143         call_data = &data;
 144
 145         for_each_cpu_mask(cpu, map)
 146                 smp_ext_bitcall(cpu, ec_call_function);
 147
 148         /* Wait for response */
 149         while (!cpus_equal(map, data.started))
 150                 cpu_relax();
 151         if (wait)
 152                 while (!cpus_equal(map, data.finished))
 153                         cpu_relax();
 154         spin_unlock(&call_lock);
 155 out:
 156         if (local) {
 157                 local_irq_disable();
 158                 func(info);
 159                 local_irq_enable();
 160         }
 161 }
 162
 163 /*
 164  * smp_call_function:
 165  * @func: the function to run; this must be fast and non-blocking
 166  * @info: an arbitrary pointer to pass to the function
 167  * @nonatomic: unused
 168  * @wait: if true, wait (atomically) until function has completed on other CPUs
 169  *
 170  * Run a function on all other CPUs.
 171  *
 172  * You must not call this function with disabled interrupts, from a
 173  * hardware interrupt handler or from a bottom half.
 174  */
 175 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
 176                       int wait)
 177 {
 178         cpumask_t map;
 179
 180         preempt_disable();
 181         map = cpu_online_map;
 182         cpu_clear(smp_processor_id(), map);
 183         __smp_call_function_map(func, info, nonatomic, wait, map);
 184         preempt_enable();
 185         return 0;
 186 }
 187 EXPORT_SYMBOL(smp_call_function);
 188
 189 /*
 190  * smp_call_function_single:
 191  * @cpu: the CPU where func should run
 192  * @func: the function to run; this must be fast and non-blocking
 193  * @info: an arbitrary pointer to pass to the function
 194  * @nonatomic: unused
 195  * @wait: if true, wait (atomically) until function has completed on other CPUs
 196  *
 197  * Run a function on one processor.
 198  *
 199  * You must not call this function with disabled interrupts, from a
 200  * hardware interrupt handler or from a bottom half.
 201  */
 202 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 203                              int nonatomic, int wait)
 204 {
 205         preempt_disable();
 206         __smp_call_function_map(func, info, nonatomic, wait,
 207                                 cpumask_of_cpu(cpu));
 208         preempt_enable();
 209         return 0;
 210 }
 211 EXPORT_SYMBOL(smp_call_function_single);
 212
 213 void smp_send_stop(void)
 214 {
 215         int cpu, rc;
 216
 217         /* Disable all interrupts/machine checks */
 218         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 219
 220         /* write magic number to zero page (absolute 0) */
 221         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
 222
 223         /* stop all processors */
 224         for_each_online_cpu(cpu) {
 225                 if (cpu == smp_processor_id())
 226                         continue;
 227                 do {
 228                         rc = signal_processor(cpu, sigp_stop);
 229                 } while (rc == sigp_busy);
 230
 231                 while (!smp_cpu_not_running(cpu))
 232                         cpu_relax();
 233         }
 234 }
 235
 236 /*
 237  * Reboot, halt and power_off routines for SMP.
 238  */
 239 void machine_restart_smp(char *__unused)
 240 {
 241         smp_send_stop();
 242         do_reipl();
 243 }
 244
 245 void machine_halt_smp(void)
 246 {
 247         smp_send_stop();
 248         if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
 249                 __cpcmd(vmhalt_cmd, NULL, 0, NULL);
 250         signal_processor(smp_processor_id(), sigp_stop_and_store_status);
 251         for (;;);
 252 }
 253
 254 void machine_power_off_smp(void)
 255 {
 256         smp_send_stop();
 257         if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
 258                 __cpcmd(vmpoff_cmd, NULL, 0, NULL);
 259         signal_processor(smp_processor_id(), sigp_stop_and_store_status);
 260         for (;;);
 261 }
 262
 263 /*
 264  * This is the main routine where commands issued by other
 265  * cpus are handled.
 266  */
 267
 268 static void do_ext_call_interrupt(__u16 code)
 269 {
 270         unsigned long bits;
 271
 272         /*
 273          * handle bit signal external calls
 274          *
 275          * For the ec_schedule signal we have to do nothing. All the work
 276          * is done automatically when we return from the interrupt.
 277          */
 278         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 279
 280         if (test_bit(ec_call_function, &bits))
 281                 do_call_function();
 282 }
 283
 284 /*
 285  * Send an external call sigp to another cpu and return without waiting
 286  * for its completion.
 287  */
 288 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 289 {
 290         /*
 291          * Set signaling bit in lowcore of target cpu and kick it
 292          */
 293         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 294         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 295                 udelay(10);
 296 }
 297
 298 #ifndef CONFIG_64BIT
 299 /*
 300  * this function sends a 'purge tlb' signal to another CPU.
 301  */
 302 void smp_ptlb_callback(void *info)
 303 {
 304         __tlb_flush_local();
 305 }
 306
 307 void smp_ptlb_all(void)
 308 {
 309         on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
 310 }
 311 EXPORT_SYMBOL(smp_ptlb_all);
 312 #endif /* ! CONFIG_64BIT */
 313
 314 /*
 315  * this function sends a 'reschedule' IPI to another CPU.
 316  * it goes straight through and wastes no time serializing
 317  * anything. Worst case is that we lose a reschedule ...
 318  */
 319 void smp_send_reschedule(int cpu)
 320 {
 321         smp_ext_bitcall(cpu, ec_schedule);
 322 }
 323
 324 /*
 325  * parameter area for the set/clear control bit callbacks
 326  */
 327 struct ec_creg_mask_parms {
 328         unsigned long orvals[16];
 329         unsigned long andvals[16];
 330 };
 331
 332 /*
 333  * callback for setting/clearing control bits
 334  */
 335 static void smp_ctl_bit_callback(void *info)
 336 {
 337         struct ec_creg_mask_parms *pp = info;
 338         unsigned long cregs[16];
 339         int i;
 340
 341         __ctl_store(cregs, 0, 15);
 342         for (i = 0; i <= 15; i++)
 343                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 344         __ctl_load(cregs, 0, 15);
 345 }
 346
 347 /*
 348  * Set a bit in a control register of all cpus
 349  */
 350 void smp_ctl_set_bit(int cr, int bit)
 351 {
 352         struct ec_creg_mask_parms parms;
 353
 354         memset(&parms.orvals, 0, sizeof(parms.orvals));
 355         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 356         parms.orvals[cr] = 1 << bit;
 357         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 358 }
 359 EXPORT_SYMBOL(smp_ctl_set_bit);
 360
 361 /*
 362  * Clear a bit in a control register of all cpus
 363  */
 364 void smp_ctl_clear_bit(int cr, int bit)
 365 {
 366         struct ec_creg_mask_parms parms;
 367
 368         memset(&parms.orvals, 0, sizeof(parms.orvals));
 369         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 370         parms.andvals[cr] = ~(1L << bit);
 371         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 372 }
 373 EXPORT_SYMBOL(smp_ctl_clear_bit);
 374
 375 /*
 376  * In early ipl state a temp. logically cpu number is needed, so the sigp
 377  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 378  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 379  */
 380 #define CPU_INIT_NO     1
 381
 382 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 383
 384 /*
 385  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 386  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 387  * save its prefix registers, since they get lost, when switching from 31 bit
 388  * to 64 bit.
 389  */
 390 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 391         __attribute__((__section__(".data")));
 392
 393 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 394 {
 395         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 396                 return;
 397         if (cpu >= NR_CPUS) {
 398                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
 399                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
 400                 return;
 401         }
 402         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 403         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 404         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 405                sigp_busy)
 406                 cpu_relax();
 407         memcpy(zfcpdump_save_areas[cpu],
 408                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 409                SAVE_AREA_SIZE);
 410 #ifdef CONFIG_64BIT
 411         /* copy original prefix register */
 412         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 413 #endif
 414 }
 415
 416 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 417 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 418
 419 #else
 420
 421 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 422
 423 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 424
 425 static int cpu_stopped(int cpu)
 426 {
 427         __u32 status;
 428
 429         /* Check for stopped state */
 430         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 431             sigp_status_stored) {
 432                 if (status & 0x40)
 433                         return 1;
 434         }
 435         return 0;
 436 }
 437
 438 static int cpu_known(int cpu_id)
 439 {
 440         int cpu;
 441
 442         for_each_present_cpu(cpu) {
 443                 if (__cpu_logical_map[cpu] == cpu_id)
 444                         return 1;
 445         }
 446         return 0;
 447 }
 448
 449 static int smp_rescan_cpus_sigp(cpumask_t avail)
 450 {
 451         int cpu_id, logical_cpu;
 452
 453         logical_cpu = first_cpu(avail);
 454         if (logical_cpu == NR_CPUS)
 455                 return 0;
 456         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 457                 if (cpu_known(cpu_id))
 458                         continue;
 459                 __cpu_logical_map[logical_cpu] = cpu_id;
 460                 if (!cpu_stopped(logical_cpu))
 461                         continue;
 462                 cpu_set(logical_cpu, cpu_present_map);
 463                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 464                 logical_cpu = next_cpu(logical_cpu, avail);
 465                 if (logical_cpu == NR_CPUS)
 466                         break;
 467         }
 468         return 0;
 469 }
 470
 471 static int smp_rescan_cpus_sclp(cpumask_t avail)
 472 {
 473         struct sclp_cpu_info *info;
 474         int cpu_id, logical_cpu, cpu;
 475         int rc;
 476
 477         logical_cpu = first_cpu(avail);
 478         if (logical_cpu == NR_CPUS)
 479                 return 0;
 480         info = kmalloc(sizeof(*info), GFP_KERNEL);
 481         if (!info)
 482                 return -ENOMEM;
 483         rc = sclp_get_cpu_info(info);
 484         if (rc)
 485                 goto out;
 486         for (cpu = 0; cpu < info->combined; cpu++) {
 487                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 488                         continue;
 489                 cpu_id = info->cpu[cpu].address;
 490                 if (cpu_known(cpu_id))
 491                         continue;
 492                 __cpu_logical_map[logical_cpu] = cpu_id;
 493                 cpu_set(logical_cpu, cpu_present_map);
 494                 if (cpu >= info->configured)
 495                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 496                 else
 497                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 498                 logical_cpu = next_cpu(logical_cpu, avail);
 499                 if (logical_cpu == NR_CPUS)
 500                         break;
 501         }
 502 out:
 503         kfree(info);
 504         return rc;
 505 }
 506
 507 static int smp_rescan_cpus(void)
 508 {
 509         cpumask_t avail;
 510
 511         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 512         if (smp_use_sigp_detection)
 513                 return smp_rescan_cpus_sigp(avail);
 514         else
 515                 return smp_rescan_cpus_sclp(avail);
 516 }
 517
 518 static void __init smp_detect_cpus(void)
 519 {
 520         unsigned int cpu, c_cpus, s_cpus;
 521         struct sclp_cpu_info *info;
 522         u16 boot_cpu_addr, cpu_addr;
 523
 524         c_cpus = 1;
 525         s_cpus = 0;
 526         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 527         info = kmalloc(sizeof(*info), GFP_KERNEL);
 528         if (!info)
 529                 panic("smp_detect_cpus failed to allocate memory\n");
 530         /* Use sigp detection algorithm if sclp doesn't work. */
 531         if (sclp_get_cpu_info(info)) {
 532                 smp_use_sigp_detection = 1;
 533                 for (cpu = 0; cpu <= 65535; cpu++) {
 534                         if (cpu == boot_cpu_addr)
 535                                 continue;
 536                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 537                         if (!cpu_stopped(CPU_INIT_NO))
 538                                 continue;
 539                         smp_get_save_area(c_cpus, cpu);
 540                         c_cpus++;
 541                 }
 542                 goto out;
 543         }
 544
 545         if (info->has_cpu_type) {
 546                 for (cpu = 0; cpu < info->combined; cpu++) {
 547                         if (info->cpu[cpu].address == boot_cpu_addr) {
 548                                 smp_cpu_type = info->cpu[cpu].type;
 549                                 break;
 550                         }
 551                 }
 552         }
 553
 554         for (cpu = 0; cpu < info->combined; cpu++) {
 555                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 556                         continue;
 557                 cpu_addr = info->cpu[cpu].address;
 558                 if (cpu_addr == boot_cpu_addr)
 559                         continue;
 560                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 561                 if (!cpu_stopped(CPU_INIT_NO)) {
 562                         s_cpus++;
 563                         continue;
 564                 }
 565                 smp_get_save_area(c_cpus, cpu_addr);
 566                 c_cpus++;
 567         }
 568 out:
 569         kfree(info);
 570         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
 571         lock_cpu_hotplug();
 572         smp_rescan_cpus();
 573         unlock_cpu_hotplug();
 574 }
 575
 576 /*
 577  *      Activate a secondary processor.
 578  */
 579 int __cpuinit start_secondary(void *cpuvoid)
 580 {
 581         /* Setup the cpu */
 582         cpu_init();
 583         preempt_disable();
 584         /* Enable TOD clock interrupts on the secondary cpu. */
 585         init_cpu_timer();
 586 #ifdef CONFIG_VIRT_TIMER
 587         /* Enable cpu timer interrupts on the secondary cpu. */
 588         init_cpu_vtimer();
 589 #endif
 590         /* Enable pfault pseudo page faults on this cpu. */
 591         pfault_init();
 592
 593         /* Mark this cpu as online */
 594         cpu_set(smp_processor_id(), cpu_online_map);
 595         /* Switch on interrupts */
 596         local_irq_enable();
 597         /* Print info about this processor */
 598         print_cpu_info(&S390_lowcore.cpu_data);
 599         /* cpu_idle will call schedule for us */
 600         cpu_idle();
 601         return 0;
 602 }
 603
 604 static void __init smp_create_idle(unsigned int cpu)
 605 {
 606         struct task_struct *p;
 607
 608         /*
 609          *  don't care about the psw and regs settings since we'll never
 610          *  reschedule the forked task.
 611          */
 612         p = fork_idle(cpu);
 613         if (IS_ERR(p))
 614                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 615         current_set[cpu] = p;
 616         spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
 617 }
 618
 619 /* Upping and downing of CPUs */
 620 int __cpu_up(unsigned int cpu)
 621 {
 622         struct task_struct *idle;
 623         struct _lowcore *cpu_lowcore;
 624         struct stack_frame *sf;
 625         sigp_ccode ccode;
 626
 627         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 628                 return -EIO;
 629
 630         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 631                                    cpu, sigp_set_prefix);
 632         if (ccode) {
 633                 printk("sigp_set_prefix failed for cpu %d "
 634                        "with condition code %d\n",
 635                        (int) cpu, (int) ccode);
 636                 return -EIO;
 637         }
 638
 639         idle = current_set[cpu];
 640         cpu_lowcore = lowcore_ptr[cpu];
 641         cpu_lowcore->kernel_stack = (unsigned long)
 642                 task_stack_page(idle) + THREAD_SIZE;
 643         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 644                                      - sizeof(struct pt_regs)
 645                                      - sizeof(struct stack_frame));
 646         memset(sf, 0, sizeof(struct stack_frame));
 647         sf->gprs[9] = (unsigned long) sf;
 648         cpu_lowcore->save_area[15] = (unsigned long) sf;
 649         __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
 650         asm volatile(
 651                 "       stam    0,15,0(%0)"
 652                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 653         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 654         cpu_lowcore->current_task = (unsigned long) idle;
 655         cpu_lowcore->cpu_data.cpu_nr = cpu;
 656         eieio();
 657
 658         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 659                 udelay(10);
 660
 661         while (!cpu_online(cpu))
 662                 cpu_relax();
 663         return 0;
 664 }
 665
 666 static int __init setup_possible_cpus(char *s)
 667 {
 668         int pcpus, cpu;
 669
 670         pcpus = simple_strtoul(s, NULL, 0);
 671         cpu_possible_map = cpumask_of_cpu(0);
 672         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 673                 cpu_set(cpu, cpu_possible_map);
 674         return 0;
 675 }
 676 early_param("possible_cpus", setup_possible_cpus);
 677
 678 #ifdef CONFIG_HOTPLUG_CPU
 679
 680 int __cpu_disable(void)
 681 {
 682         struct ec_creg_mask_parms cr_parms;
 683         int cpu = smp_processor_id();
 684
 685         cpu_clear(cpu, cpu_online_map);
 686
 687         /* Disable pfault pseudo page faults on this cpu. */
 688         pfault_fini();
 689
 690         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 691         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 692
 693         /* disable all external interrupts */
 694         cr_parms.orvals[0] = 0;
 695         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 696                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 697         /* disable all I/O interrupts */
 698         cr_parms.orvals[6] = 0;
 699         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 700                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 701         /* disable most machine checks */
 702         cr_parms.orvals[14] = 0;
 703         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 704                                  1 << 25 | 1 << 24);
 705
 706         smp_ctl_bit_callback(&cr_parms);
 707
 708         return 0;
 709 }
 710
 711 void __cpu_die(unsigned int cpu)
 712 {
 713         /* Wait until target cpu is down */
 714         while (!smp_cpu_not_running(cpu))
 715                 cpu_relax();
 716         printk(KERN_INFO "Processor %d spun down\n", cpu);
 717 }
 718
 719 void cpu_die(void)
 720 {
 721         idle_task_exit();
 722         signal_processor(smp_processor_id(), sigp_stop);
 723         BUG();
 724         for (;;);
 725 }
 726
 727 #endif /* CONFIG_HOTPLUG_CPU */
 728
 729 /*
 730  *      Cycle through the processors and setup structures.
 731  */
 732
 733 void __init smp_prepare_cpus(unsigned int max_cpus)
 734 {
 735         unsigned long stack;
 736         unsigned int cpu;
 737         int i;
 738
 739         smp_detect_cpus();
 740
 741         /* request the 0x1201 emergency signal external interrupt */
 742         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 743                 panic("Couldn't request external interrupt 0x1201");
 744         memset(lowcore_ptr, 0, sizeof(lowcore_ptr));
 745         /*
 746          *  Initialize prefix pages and stacks for all possible cpus
 747          */
 748         print_cpu_info(&S390_lowcore.cpu_data);
 749
 750         for_each_possible_cpu(i) {
 751                 lowcore_ptr[i] = (struct _lowcore *)
 752                         __get_free_pages(GFP_KERNEL | GFP_DMA,
 753                                          sizeof(void*) == 8 ? 1 : 0);
 754                 stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 755                 if (!lowcore_ptr[i] || !stack)
 756                         panic("smp_boot_cpus failed to allocate memory\n");
 757
 758                 *(lowcore_ptr[i]) = S390_lowcore;
 759                 lowcore_ptr[i]->async_stack = stack + ASYNC_SIZE;
 760                 stack = __get_free_pages(GFP_KERNEL, 0);
 761                 if (!stack)
 762                         panic("smp_boot_cpus failed to allocate memory\n");
 763                 lowcore_ptr[i]->panic_stack = stack + PAGE_SIZE;
 764 #ifndef CONFIG_64BIT
 765                 if (MACHINE_HAS_IEEE) {
 766                         lowcore_ptr[i]->extended_save_area_addr =
 767                                 (__u32) __get_free_pages(GFP_KERNEL, 0);
 768                         if (!lowcore_ptr[i]->extended_save_area_addr)
 769                                 panic("smp_boot_cpus failed to "
 770                                       "allocate memory\n");
 771                 }
 772 #endif
 773         }
 774 #ifndef CONFIG_64BIT
 775         if (MACHINE_HAS_IEEE)
 776                 ctl_set_bit(14, 29); /* enable extended save area */
 777 #endif
 778         set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
 779
 780         for_each_possible_cpu(cpu)
 781                 if (cpu != smp_processor_id())
 782                         smp_create_idle(cpu);
 783 }
 784
 785 void __init smp_prepare_boot_cpu(void)
 786 {
 787         BUG_ON(smp_processor_id() != 0);
 788
 789         current_thread_info()->cpu = 0;
 790         cpu_set(0, cpu_present_map);
 791         cpu_set(0, cpu_online_map);
 792         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 793         current_set[0] = current;
 794         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 795         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 796 }
 797
 798 void __init smp_cpus_done(unsigned int max_cpus)
 799 {
 800 }
 801
 802 /*
 803  * the frequency of the profiling timer can be changed
 804  * by writing a multiplier value into /proc/profile.
 805  *
 806  * usually you want to run this on all CPUs ;)
 807  */
 808 int setup_profiling_timer(unsigned int multiplier)
 809 {
 810         return 0;
 811 }
 812
 813 #ifdef CONFIG_HOTPLUG_CPU
 814 static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
 815 {
 816         ssize_t count;
 817
 818         mutex_lock(&smp_cpu_state_mutex);
 819         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 820         mutex_unlock(&smp_cpu_state_mutex);
 821         return count;
 822 }
 823
 824 static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
 825                                    size_t count)
 826 {
 827         int cpu = dev->id;
 828         int val, rc;
 829         char delim;
 830
 831         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 832                 return -EINVAL;
 833         if (val != 0 && val != 1)
 834                 return -EINVAL;
 835
 836         mutex_lock(&smp_cpu_state_mutex);
 837         lock_cpu_hotplug();
 838         rc = -EBUSY;
 839         if (cpu_online(cpu))
 840                 goto out;
 841         rc = 0;
 842         switch (val) {
 843         case 0:
 844                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 845                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 846                         if (!rc)
 847                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 848                 }
 849                 break;
 850         case 1:
 851                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 852                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 853                         if (!rc)
 854                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 855                 }
 856                 break;
 857         default:
 858                 break;
 859         }
 860 out:
 861         unlock_cpu_hotplug();
 862         mutex_unlock(&smp_cpu_state_mutex);
 863         return rc ? rc : count;
 864 }
 865 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 866 #endif /* CONFIG_HOTPLUG_CPU */
 867
 868 static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
 869 {
 870         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 871 }
 872 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 873
 874
 875 static struct attribute *cpu_common_attrs[] = {
 876 #ifdef CONFIG_HOTPLUG_CPU
 877         &attr_configure.attr,
 878 #endif
 879         &attr_address.attr,
 880         NULL,
 881 };
 882
 883 static struct attribute_group cpu_common_attr_group = {
 884         .attrs = cpu_common_attrs,
 885 };
 886
 887 static ssize_t show_capability(struct sys_device *dev, char *buf)
 888 {
 889         unsigned int capability;
 890         int rc;
 891
 892         rc = get_cpu_capability(&capability);
 893         if (rc)
 894                 return rc;
 895         return sprintf(buf, "%u\n", capability);
 896 }
 897 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 898
 899 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
 900 {
 901         struct s390_idle_data *idle;
 902         unsigned long long idle_count;
 903
 904         idle = &per_cpu(s390_idle, dev->id);
 905         spin_lock_irq(&idle->lock);
 906         idle_count = idle->idle_count;
 907         spin_unlock_irq(&idle->lock);
 908         return sprintf(buf, "%llu\n", idle_count);
 909 }
 910 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 911
 912 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
 913 {
 914         struct s390_idle_data *idle;
 915         unsigned long long new_time;
 916
 917         idle = &per_cpu(s390_idle, dev->id);
 918         spin_lock_irq(&idle->lock);
 919         if (idle->in_idle) {
 920                 new_time = get_clock();
 921                 idle->idle_time += new_time - idle->idle_enter;
 922                 idle->idle_enter = new_time;
 923         }
 924         new_time = idle->idle_time;
 925         spin_unlock_irq(&idle->lock);
 926         return sprintf(buf, "%llu\n", new_time >> 12);
 927 }
 928 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 929
 930 static struct attribute *cpu_online_attrs[] = {
 931         &attr_capability.attr,
 932         &attr_idle_count.attr,
 933         &attr_idle_time_us.attr,
 934         NULL,
 935 };
 936
 937 static struct attribute_group cpu_online_attr_group = {
 938         .attrs = cpu_online_attrs,
 939 };
 940
 941 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 942                                     unsigned long action, void *hcpu)
 943 {
 944         unsigned int cpu = (unsigned int)(long)hcpu;
 945         struct cpu *c = &per_cpu(cpu_devices, cpu);
 946         struct sys_device *s = &c->sysdev;
 947         struct s390_idle_data *idle;
 948
 949         switch (action) {
 950         case CPU_ONLINE:
 951         case CPU_ONLINE_FROZEN:
 952                 idle = &per_cpu(s390_idle, cpu);
 953                 spin_lock_irq(&idle->lock);
 954                 idle->idle_enter = 0;
 955                 idle->idle_time = 0;
 956                 idle->idle_count = 0;
 957                 spin_unlock_irq(&idle->lock);
 958                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 959                         return NOTIFY_BAD;
 960                 break;
 961         case CPU_DEAD:
 962         case CPU_DEAD_FROZEN:
 963                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 964                 break;
 965         }
 966         return NOTIFY_OK;
 967 }
 968
 969 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 970         .notifier_call = smp_cpu_notify,
 971 };
 972
 973 static int smp_add_present_cpu(int cpu)
 974 {
 975         struct cpu *c = &per_cpu(cpu_devices, cpu);
 976         struct sys_device *s = &c->sysdev;
 977         int rc;
 978
 979         c->hotpluggable = 1;
 980         rc = register_cpu(c, cpu);
 981         if (rc)
 982                 goto out;
 983         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 984         if (rc)
 985                 goto out_cpu;
 986         if (!cpu_online(cpu))
 987                 goto out;
 988         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 989         if (!rc)
 990                 return 0;
 991         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 992 out_cpu:
 993 #ifdef CONFIG_HOTPLUG_CPU
 994         unregister_cpu(c);
 995 #endif
 996 out:
 997         return rc;
 998 }
 999
1000 #ifdef CONFIG_HOTPLUG_CPU
1001 static ssize_t rescan_store(struct sys_device *dev, const char *buf,
1002                             size_t count)
1003 {
1004         cpumask_t newcpus;
1005         int cpu;
1006         int rc;
1007
1008         mutex_lock(&smp_cpu_state_mutex);
1009         lock_cpu_hotplug();
1010         newcpus = cpu_present_map;
1011         rc = smp_rescan_cpus();
1012         if (rc)
1013                 goto out;
1014         cpus_andnot(newcpus, cpu_present_map, newcpus);
1015         for_each_cpu_mask(cpu, newcpus) {
1016                 rc = smp_add_present_cpu(cpu);
1017                 if (rc)
1018                         cpu_clear(cpu, cpu_present_map);
1019         }
1020         rc = 0;
1021 out:
1022         unlock_cpu_hotplug();
1023         mutex_unlock(&smp_cpu_state_mutex);
1024         return rc ? rc : count;
1025 }
1026 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1027 #endif /* CONFIG_HOTPLUG_CPU */
1028
1029 static int __init topology_init(void)
1030 {
1031         int cpu;
1032         int rc;
1033
1034         register_cpu_notifier(&smp_cpu_nb);
1035
1036 #ifdef CONFIG_HOTPLUG_CPU
1037         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1038                                &attr_rescan.attr);
1039         if (rc)
1040                 return rc;
1041 #endif
1042         for_each_present_cpu(cpu) {
1043                 rc = smp_add_present_cpu(cpu);
1044                 if (rc)
1045                         return rc;
1046         }
1047         return 0;
1048 }
1049 subsys_initcall(topology_init);