arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999, 2009
   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 #define KMSG_COMPONENT "cpu"
  24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  25
  26 #include <linux/workqueue.h>
  27 #include <linux/module.h>
  28 #include <linux/init.h>
  29 #include <linux/mm.h>
  30 #include <linux/err.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/kernel_stat.h>
  33 #include <linux/delay.h>
  34 #include <linux/cache.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/irqflags.h>
  37 #include <linux/cpu.h>
  38 #include <linux/timex.h>
  39 #include <linux/bootmem.h>
  40 #include <linux/slab.h>
  41 #include <linux/crash_dump.h>
  42 #include <asm/asm-offsets.h>
  43 #include <asm/ipl.h>
  44 #include <asm/setup.h>
  45 #include <asm/sigp.h>
  46 #include <asm/pgalloc.h>
  47 #include <asm/irq.h>
  48 #include <asm/cpcmd.h>
  49 #include <asm/tlbflush.h>
  50 #include <asm/timer.h>
  51 #include <asm/lowcore.h>
  52 #include <asm/sclp.h>
  53 #include <asm/cputime.h>
  54 #include <asm/vdso.h>
  55 #include <asm/cpu.h>
  56 #include "entry.h"
  57
  58 /* logical cpu to cpu address */
  59 unsigned short __cpu_logical_map[NR_CPUS];
  60
  61 static struct task_struct *current_set[NR_CPUS];
  62
  63 static u8 smp_cpu_type;
  64 static int smp_use_sigp_detection;
  65
  66 enum s390_cpu_state {
  67         CPU_STATE_STANDBY,
  68         CPU_STATE_CONFIGURED,
  69 };
  70
  71 DEFINE_MUTEX(smp_cpu_state_mutex);
  72 static int smp_cpu_state[NR_CPUS];
  73
  74 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  75
  76 static void smp_ext_bitcall(int, int);
  77
  78 static int raw_cpu_stopped(int cpu)
  79 {
  80         u32 status;
  81
  82         switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
  83         case sigp_status_stored:
  84                 /* Check for stopped and check stop state */
  85                 if (status & 0x50)
  86                         return 1;
  87                 break;
  88         default:
  89                 break;
  90         }
  91         return 0;
  92 }
  93
  94 static inline int cpu_stopped(int cpu)
  95 {
  96         return raw_cpu_stopped(cpu_logical_map(cpu));
  97 }
  98
  99 /*
 100  * Ensure that PSW restart is done on an online CPU
 101  */
 102 void smp_restart_with_online_cpu(void)
 103 {
 104         int cpu;
 105
 106         for_each_online_cpu(cpu) {
 107                 if (stap() == __cpu_logical_map[cpu]) {
 108                         /* We are online: Enable DAT again and return */
 109                         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 110                         return;
 111                 }
 112         }
 113         /* We are not online: Do PSW restart on an online CPU */
 114         while (sigp(cpu, sigp_restart) == sigp_busy)
 115                 cpu_relax();
 116         /* And stop ourself */
 117         while (raw_sigp(stap(), sigp_stop) == sigp_busy)
 118                 cpu_relax();
 119         for (;;);
 120 }
 121
 122 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
 123 {
 124         struct _lowcore *lc, *current_lc;
 125         struct stack_frame *sf;
 126         struct pt_regs *regs;
 127         unsigned long sp;
 128
 129         if (smp_processor_id() == 0)
 130                 func(data);
 131         __load_psw_mask(PSW_DEFAULT_KEY | PSW_MASK_BASE |
 132                         PSW_MASK_EA | PSW_MASK_BA);
 133         /* Disable lowcore protection */
 134         __ctl_clear_bit(0, 28);
 135         current_lc = lowcore_ptr[smp_processor_id()];
 136         lc = lowcore_ptr[0];
 137         if (!lc)
 138                 lc = current_lc;
 139         lc->restart_psw.mask =
 140                 PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
 141         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
 142         if (!cpu_online(0))
 143                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
 144         while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
 145                 cpu_relax();
 146         sp = lc->panic_stack;
 147         sp -= sizeof(struct pt_regs);
 148         regs = (struct pt_regs *) sp;
 149         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
 150         regs->psw = current_lc->psw_save_area;
 151         sp -= STACK_FRAME_OVERHEAD;
 152         sf = (struct stack_frame *) sp;
 153         sf->back_chain = 0;
 154         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 155 }
 156
 157 void smp_send_stop(void)
 158 {
 159         int cpu, rc;
 160
 161         /* Disable all interrupts/machine checks */
 162         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 163         trace_hardirqs_off();
 164
 165         /* stop all processors */
 166         for_each_online_cpu(cpu) {
 167                 if (cpu == smp_processor_id())
 168                         continue;
 169                 do {
 170                         rc = sigp(cpu, sigp_stop);
 171                 } while (rc == sigp_busy);
 172
 173                 while (!cpu_stopped(cpu))
 174                         cpu_relax();
 175         }
 176 }
 177
 178 /*
 179  * This is the main routine where commands issued by other
 180  * cpus are handled.
 181  */
 182
 183 static void do_ext_call_interrupt(unsigned int ext_int_code,
 184                                   unsigned int param32, unsigned long param64)
 185 {
 186         unsigned long bits;
 187
 188         if ((ext_int_code & 0xffff) == 0x1202)
 189                 kstat_cpu(smp_processor_id()).irqs[EXTINT_EXC]++;
 190         else
 191                 kstat_cpu(smp_processor_id()).irqs[EXTINT_EMS]++;
 192         /*
 193          * handle bit signal external calls
 194          */
 195         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 196
 197         if (test_bit(ec_schedule, &bits))
 198                 scheduler_ipi();
 199
 200         if (test_bit(ec_call_function, &bits))
 201                 generic_smp_call_function_interrupt();
 202
 203         if (test_bit(ec_call_function_single, &bits))
 204                 generic_smp_call_function_single_interrupt();
 205 }
 206
 207 /*
 208  * Send an external call sigp to another cpu and return without waiting
 209  * for its completion.
 210  */
 211 static void smp_ext_bitcall(int cpu, int sig)
 212 {
 213         int order;
 214
 215         /*
 216          * Set signaling bit in lowcore of target cpu and kick it
 217          */
 218         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 219         while (1) {
 220                 order = smp_vcpu_scheduled(cpu) ?
 221                         sigp_external_call : sigp_emergency_signal;
 222                 if (sigp(cpu, order) != sigp_busy)
 223                         break;
 224                 udelay(10);
 225         }
 226 }
 227
 228 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 229 {
 230         int cpu;
 231
 232         for_each_cpu(cpu, mask)
 233                 smp_ext_bitcall(cpu, ec_call_function);
 234 }
 235
 236 void arch_send_call_function_single_ipi(int cpu)
 237 {
 238         smp_ext_bitcall(cpu, ec_call_function_single);
 239 }
 240
 241 #ifndef CONFIG_64BIT
 242 /*
 243  * this function sends a 'purge tlb' signal to another CPU.
 244  */
 245 static void smp_ptlb_callback(void *info)
 246 {
 247         __tlb_flush_local();
 248 }
 249
 250 void smp_ptlb_all(void)
 251 {
 252         on_each_cpu(smp_ptlb_callback, NULL, 1);
 253 }
 254 EXPORT_SYMBOL(smp_ptlb_all);
 255 #endif /* ! CONFIG_64BIT */
 256
 257 /*
 258  * this function sends a 'reschedule' IPI to another CPU.
 259  * it goes straight through and wastes no time serializing
 260  * anything. Worst case is that we lose a reschedule ...
 261  */
 262 void smp_send_reschedule(int cpu)
 263 {
 264         smp_ext_bitcall(cpu, ec_schedule);
 265 }
 266
 267 /*
 268  * parameter area for the set/clear control bit callbacks
 269  */
 270 struct ec_creg_mask_parms {
 271         unsigned long orvals[16];
 272         unsigned long andvals[16];
 273 };
 274
 275 /*
 276  * callback for setting/clearing control bits
 277  */
 278 static void smp_ctl_bit_callback(void *info)
 279 {
 280         struct ec_creg_mask_parms *pp = info;
 281         unsigned long cregs[16];
 282         int i;
 283
 284         __ctl_store(cregs, 0, 15);
 285         for (i = 0; i <= 15; i++)
 286                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 287         __ctl_load(cregs, 0, 15);
 288 }
 289
 290 /*
 291  * Set a bit in a control register of all cpus
 292  */
 293 void smp_ctl_set_bit(int cr, int bit)
 294 {
 295         struct ec_creg_mask_parms parms;
 296
 297         memset(&parms.orvals, 0, sizeof(parms.orvals));
 298         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 299         parms.orvals[cr] = 1UL << bit;
 300         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 301 }
 302 EXPORT_SYMBOL(smp_ctl_set_bit);
 303
 304 /*
 305  * Clear a bit in a control register of all cpus
 306  */
 307 void smp_ctl_clear_bit(int cr, int bit)
 308 {
 309         struct ec_creg_mask_parms parms;
 310
 311         memset(&parms.orvals, 0, sizeof(parms.orvals));
 312         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 313         parms.andvals[cr] = ~(1UL << bit);
 314         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 315 }
 316 EXPORT_SYMBOL(smp_ctl_clear_bit);
 317
 318 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
 319
 320 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 321 {
 322         if (ipl_info.type != IPL_TYPE_FCP_DUMP && !OLDMEM_BASE)
 323                 return;
 324         if (is_kdump_kernel())
 325                 return;
 326         if (cpu >= NR_CPUS) {
 327                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 328                            "the dump\n", cpu, NR_CPUS - 1);
 329                 return;
 330         }
 331         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 332         while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
 333                 cpu_relax();
 334         memcpy_real(zfcpdump_save_areas[cpu],
 335                     (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 336                     sizeof(struct save_area));
 337 }
 338
 339 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 340 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 341
 342 #else
 343
 344 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 345
 346 #endif /* CONFIG_ZFCPDUMP */
 347
 348 static int cpu_known(int cpu_id)
 349 {
 350         int cpu;
 351
 352         for_each_present_cpu(cpu) {
 353                 if (__cpu_logical_map[cpu] == cpu_id)
 354                         return 1;
 355         }
 356         return 0;
 357 }
 358
 359 static int smp_rescan_cpus_sigp(cpumask_t avail)
 360 {
 361         int cpu_id, logical_cpu;
 362
 363         logical_cpu = cpumask_first(&avail);
 364         if (logical_cpu >= nr_cpu_ids)
 365                 return 0;
 366         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 367                 if (cpu_known(cpu_id))
 368                         continue;
 369                 __cpu_logical_map[logical_cpu] = cpu_id;
 370                 cpu_set_polarization(logical_cpu, POLARIZATION_UNKNOWN);
 371                 if (!cpu_stopped(logical_cpu))
 372                         continue;
 373                 set_cpu_present(logical_cpu, true);
 374                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 375                 logical_cpu = cpumask_next(logical_cpu, &avail);
 376                 if (logical_cpu >= nr_cpu_ids)
 377                         break;
 378         }
 379         return 0;
 380 }
 381
 382 static int smp_rescan_cpus_sclp(cpumask_t avail)
 383 {
 384         struct sclp_cpu_info *info;
 385         int cpu_id, logical_cpu, cpu;
 386         int rc;
 387
 388         logical_cpu = cpumask_first(&avail);
 389         if (logical_cpu >= nr_cpu_ids)
 390                 return 0;
 391         info = kmalloc(sizeof(*info), GFP_KERNEL);
 392         if (!info)
 393                 return -ENOMEM;
 394         rc = sclp_get_cpu_info(info);
 395         if (rc)
 396                 goto out;
 397         for (cpu = 0; cpu < info->combined; cpu++) {
 398                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 399                         continue;
 400                 cpu_id = info->cpu[cpu].address;
 401                 if (cpu_known(cpu_id))
 402                         continue;
 403                 __cpu_logical_map[logical_cpu] = cpu_id;
 404                 cpu_set_polarization(logical_cpu, POLARIZATION_UNKNOWN);
 405                 set_cpu_present(logical_cpu, true);
 406                 if (cpu >= info->configured)
 407                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 408                 else
 409                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 410                 logical_cpu = cpumask_next(logical_cpu, &avail);
 411                 if (logical_cpu >= nr_cpu_ids)
 412                         break;
 413         }
 414 out:
 415         kfree(info);
 416         return rc;
 417 }
 418
 419 static int __smp_rescan_cpus(void)
 420 {
 421         cpumask_t avail;
 422
 423         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 424         if (smp_use_sigp_detection)
 425                 return smp_rescan_cpus_sigp(avail);
 426         else
 427                 return smp_rescan_cpus_sclp(avail);
 428 }
 429
 430 static void __init smp_detect_cpus(void)
 431 {
 432         unsigned int cpu, c_cpus, s_cpus;
 433         struct sclp_cpu_info *info;
 434         u16 boot_cpu_addr, cpu_addr;
 435
 436         c_cpus = 1;
 437         s_cpus = 0;
 438         boot_cpu_addr = __cpu_logical_map[0];
 439         info = kmalloc(sizeof(*info), GFP_KERNEL);
 440         if (!info)
 441                 panic("smp_detect_cpus failed to allocate memory\n");
 442 #ifdef CONFIG_CRASH_DUMP
 443         if (OLDMEM_BASE && !is_kdump_kernel()) {
 444                 struct save_area *save_area;
 445
 446                 save_area = kmalloc(sizeof(*save_area), GFP_KERNEL);
 447                 if (!save_area)
 448                         panic("could not allocate memory for save area\n");
 449                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 450                                  0x200, 0);
 451                 zfcpdump_save_areas[0] = save_area;
 452         }
 453 #endif
 454         /* Use sigp detection algorithm if sclp doesn't work. */
 455         if (sclp_get_cpu_info(info)) {
 456                 smp_use_sigp_detection = 1;
 457                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 458                         if (cpu == boot_cpu_addr)
 459                                 continue;
 460                         if (!raw_cpu_stopped(cpu))
 461                                 continue;
 462                         smp_get_save_area(c_cpus, cpu);
 463                         c_cpus++;
 464                 }
 465                 goto out;
 466         }
 467
 468         if (info->has_cpu_type) {
 469                 for (cpu = 0; cpu < info->combined; cpu++) {
 470                         if (info->cpu[cpu].address == boot_cpu_addr) {
 471                                 smp_cpu_type = info->cpu[cpu].type;
 472                                 break;
 473                         }
 474                 }
 475         }
 476
 477         for (cpu = 0; cpu < info->combined; cpu++) {
 478                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 479                         continue;
 480                 cpu_addr = info->cpu[cpu].address;
 481                 if (cpu_addr == boot_cpu_addr)
 482                         continue;
 483                 if (!raw_cpu_stopped(cpu_addr)) {
 484                         s_cpus++;
 485                         continue;
 486                 }
 487                 smp_get_save_area(c_cpus, cpu_addr);
 488                 c_cpus++;
 489         }
 490 out:
 491         kfree(info);
 492         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 493         get_online_cpus();
 494         __smp_rescan_cpus();
 495         put_online_cpus();
 496 }
 497
 498 /*
 499  *      Activate a secondary processor.
 500  */
 501 int __cpuinit start_secondary(void *cpuvoid)
 502 {
 503         cpu_init();
 504         preempt_disable();
 505         init_cpu_timer();
 506         init_cpu_vtimer();
 507         pfault_init();
 508
 509         notify_cpu_starting(smp_processor_id());
 510         ipi_call_lock();
 511         set_cpu_online(smp_processor_id(), true);
 512         ipi_call_unlock();
 513         __ctl_clear_bit(0, 28); /* Disable lowcore protection */
 514         S390_lowcore.restart_psw.mask =
 515                 PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
 516         S390_lowcore.restart_psw.addr =
 517                 PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
 518         __ctl_set_bit(0, 28); /* Enable lowcore protection */
 519         /*
 520          * Wait until the cpu which brought this one up marked it
 521          * active before enabling interrupts.
 522          */
 523         while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
 524                 cpu_relax();
 525         local_irq_enable();
 526         /* cpu_idle will call schedule for us */
 527         cpu_idle();
 528         return 0;
 529 }
 530
 531 struct create_idle {
 532         struct work_struct work;
 533         struct task_struct *idle;
 534         struct completion done;
 535         int cpu;
 536 };
 537
 538 static void __cpuinit smp_fork_idle(struct work_struct *work)
 539 {
 540         struct create_idle *c_idle;
 541
 542         c_idle = container_of(work, struct create_idle, work);
 543         c_idle->idle = fork_idle(c_idle->cpu);
 544         complete(&c_idle->done);
 545 }
 546
 547 static int __cpuinit smp_alloc_lowcore(int cpu)
 548 {
 549         unsigned long async_stack, panic_stack;
 550         struct _lowcore *lowcore;
 551
 552         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 553         if (!lowcore)
 554                 return -ENOMEM;
 555         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 556         panic_stack = __get_free_page(GFP_KERNEL);
 557         if (!panic_stack || !async_stack)
 558                 goto out;
 559         memcpy(lowcore, &S390_lowcore, 512);
 560         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 561         lowcore->async_stack = async_stack + ASYNC_SIZE;
 562         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 563         lowcore->restart_psw.mask =
 564                 PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_EA | PSW_MASK_BA;
 565         lowcore->restart_psw.addr =
 566                 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
 567         if (user_mode != HOME_SPACE_MODE)
 568                 lowcore->restart_psw.mask |= PSW_ASC_HOME;
 569 #ifndef CONFIG_64BIT
 570         if (MACHINE_HAS_IEEE) {
 571                 unsigned long save_area;
 572
 573                 save_area = get_zeroed_page(GFP_KERNEL);
 574                 if (!save_area)
 575                         goto out;
 576                 lowcore->extended_save_area_addr = (u32) save_area;
 577         }
 578 #else
 579         if (vdso_alloc_per_cpu(cpu, lowcore))
 580                 goto out;
 581 #endif
 582         lowcore_ptr[cpu] = lowcore;
 583         return 0;
 584
 585 out:
 586         free_page(panic_stack);
 587         free_pages(async_stack, ASYNC_ORDER);
 588         free_pages((unsigned long) lowcore, LC_ORDER);
 589         return -ENOMEM;
 590 }
 591
 592 static void smp_free_lowcore(int cpu)
 593 {
 594         struct _lowcore *lowcore;
 595
 596         lowcore = lowcore_ptr[cpu];
 597 #ifndef CONFIG_64BIT
 598         if (MACHINE_HAS_IEEE)
 599                 free_page((unsigned long) lowcore->extended_save_area_addr);
 600 #else
 601         vdso_free_per_cpu(cpu, lowcore);
 602 #endif
 603         free_page(lowcore->panic_stack - PAGE_SIZE);
 604         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 605         free_pages((unsigned long) lowcore, LC_ORDER);
 606         lowcore_ptr[cpu] = NULL;
 607 }
 608
 609 /* Upping and downing of CPUs */
 610 int __cpuinit __cpu_up(unsigned int cpu)
 611 {
 612         struct _lowcore *cpu_lowcore;
 613         struct create_idle c_idle;
 614         struct task_struct *idle;
 615         struct stack_frame *sf;
 616         u32 lowcore;
 617         int ccode;
 618
 619         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 620                 return -EIO;
 621         idle = current_set[cpu];
 622         if (!idle) {
 623                 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
 624                 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
 625                 c_idle.cpu = cpu;
 626                 schedule_work(&c_idle.work);
 627                 wait_for_completion(&c_idle.done);
 628                 if (IS_ERR(c_idle.idle))
 629                         return PTR_ERR(c_idle.idle);
 630                 idle = c_idle.idle;
 631                 current_set[cpu] = c_idle.idle;
 632         }
 633         init_idle(idle, cpu);
 634         if (smp_alloc_lowcore(cpu))
 635                 return -ENOMEM;
 636         do {
 637                 ccode = sigp(cpu, sigp_initial_cpu_reset);
 638                 if (ccode == sigp_busy)
 639                         udelay(10);
 640                 if (ccode == sigp_not_operational)
 641                         goto err_out;
 642         } while (ccode == sigp_busy);
 643
 644         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 645         while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 646                 udelay(10);
 647
 648         cpu_lowcore = lowcore_ptr[cpu];
 649         cpu_lowcore->kernel_stack = (unsigned long)
 650                 task_stack_page(idle) + THREAD_SIZE;
 651         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 652         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 653                                      - sizeof(struct pt_regs)
 654                                      - sizeof(struct stack_frame));
 655         memset(sf, 0, sizeof(struct stack_frame));
 656         sf->gprs[9] = (unsigned long) sf;
 657         cpu_lowcore->save_area[15] = (unsigned long) sf;
 658         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 659         atomic_inc(&init_mm.context.attach_count);
 660         asm volatile(
 661                 "       stam    0,15,0(%0)"
 662                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 663         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 664         cpu_lowcore->current_task = (unsigned long) idle;
 665         cpu_lowcore->cpu_nr = cpu;
 666         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 667         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 668         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 669         memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
 670                MAX_FACILITY_BIT/8);
 671         eieio();
 672
 673         while (sigp(cpu, sigp_restart) == sigp_busy)
 674                 udelay(10);
 675
 676         while (!cpu_online(cpu))
 677                 cpu_relax();
 678         return 0;
 679
 680 err_out:
 681         smp_free_lowcore(cpu);
 682         return -EIO;
 683 }
 684
 685 static int __init setup_possible_cpus(char *s)
 686 {
 687         int pcpus, cpu;
 688
 689         pcpus = simple_strtoul(s, NULL, 0);
 690         init_cpu_possible(cpumask_of(0));
 691         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 692                 set_cpu_possible(cpu, true);
 693         return 0;
 694 }
 695 early_param("possible_cpus", setup_possible_cpus);
 696
 697 #ifdef CONFIG_HOTPLUG_CPU
 698
 699 int __cpu_disable(void)
 700 {
 701         struct ec_creg_mask_parms cr_parms;
 702         int cpu = smp_processor_id();
 703
 704         set_cpu_online(cpu, false);
 705
 706         /* Disable pfault pseudo page faults on this cpu. */
 707         pfault_fini();
 708
 709         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 710         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 711
 712         /* disable all external interrupts */
 713         cr_parms.orvals[0] = 0;
 714         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
 715                                 1 << 10 | 1 <<  9 | 1 <<  6 | 1 <<  5 |
 716                                 1 <<  4);
 717         /* disable all I/O interrupts */
 718         cr_parms.orvals[6] = 0;
 719         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 720                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 721         /* disable most machine checks */
 722         cr_parms.orvals[14] = 0;
 723         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 724                                  1 << 25 | 1 << 24);
 725
 726         smp_ctl_bit_callback(&cr_parms);
 727
 728         return 0;
 729 }
 730
 731 void __cpu_die(unsigned int cpu)
 732 {
 733         /* Wait until target cpu is down */
 734         while (!cpu_stopped(cpu))
 735                 cpu_relax();
 736         while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
 737                 udelay(10);
 738         smp_free_lowcore(cpu);
 739         atomic_dec(&init_mm.context.attach_count);
 740 }
 741
 742 void __noreturn cpu_die(void)
 743 {
 744         idle_task_exit();
 745         while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
 746                 cpu_relax();
 747         for (;;);
 748 }
 749
 750 #endif /* CONFIG_HOTPLUG_CPU */
 751
 752 void __init smp_prepare_cpus(unsigned int max_cpus)
 753 {
 754 #ifndef CONFIG_64BIT
 755         unsigned long save_area = 0;
 756 #endif
 757         unsigned long async_stack, panic_stack;
 758         struct _lowcore *lowcore;
 759
 760         smp_detect_cpus();
 761
 762         /* request the 0x1201 emergency signal external interrupt */
 763         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 764                 panic("Couldn't request external interrupt 0x1201");
 765         /* request the 0x1202 external call external interrupt */
 766         if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
 767                 panic("Couldn't request external interrupt 0x1202");
 768
 769         /* Reallocate current lowcore, but keep its contents. */
 770         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 771         panic_stack = __get_free_page(GFP_KERNEL);
 772         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 773         BUG_ON(!lowcore || !panic_stack || !async_stack);
 774 #ifndef CONFIG_64BIT
 775         if (MACHINE_HAS_IEEE)
 776                 save_area = get_zeroed_page(GFP_KERNEL);
 777 #endif
 778         local_irq_disable();
 779         local_mcck_disable();
 780         lowcore_ptr[smp_processor_id()] = lowcore;
 781         *lowcore = S390_lowcore;
 782         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 783         lowcore->async_stack = async_stack + ASYNC_SIZE;
 784 #ifndef CONFIG_64BIT
 785         if (MACHINE_HAS_IEEE)
 786                 lowcore->extended_save_area_addr = (u32) save_area;
 787 #endif
 788         set_prefix((u32)(unsigned long) lowcore);
 789         local_mcck_enable();
 790         local_irq_enable();
 791 #ifdef CONFIG_64BIT
 792         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 793                 BUG();
 794 #endif
 795 }
 796
 797 void __init smp_prepare_boot_cpu(void)
 798 {
 799         BUG_ON(smp_processor_id() != 0);
 800
 801         current_thread_info()->cpu = 0;
 802         set_cpu_present(0, true);
 803         set_cpu_online(0, true);
 804         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 805         current_set[0] = current;
 806         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 807         cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 808 }
 809
 810 void __init smp_cpus_done(unsigned int max_cpus)
 811 {
 812 }
 813
 814 void __init smp_setup_processor_id(void)
 815 {
 816         S390_lowcore.cpu_nr = 0;
 817         __cpu_logical_map[0] = stap();
 818 }
 819
 820 /*
 821  * the frequency of the profiling timer can be changed
 822  * by writing a multiplier value into /proc/profile.
 823  *
 824  * usually you want to run this on all CPUs ;)
 825  */
 826 int setup_profiling_timer(unsigned int multiplier)
 827 {
 828         return 0;
 829 }
 830
 831 #ifdef CONFIG_HOTPLUG_CPU
 832 static ssize_t cpu_configure_show(struct sys_device *dev,
 833                                 struct sysdev_attribute *attr, char *buf)
 834 {
 835         ssize_t count;
 836
 837         mutex_lock(&smp_cpu_state_mutex);
 838         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 839         mutex_unlock(&smp_cpu_state_mutex);
 840         return count;
 841 }
 842
 843 static ssize_t cpu_configure_store(struct sys_device *dev,
 844                                   struct sysdev_attribute *attr,
 845                                   const char *buf, size_t count)
 846 {
 847         int cpu = dev->id;
 848         int val, rc;
 849         char delim;
 850
 851         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 852                 return -EINVAL;
 853         if (val != 0 && val != 1)
 854                 return -EINVAL;
 855
 856         get_online_cpus();
 857         mutex_lock(&smp_cpu_state_mutex);
 858         rc = -EBUSY;
 859         /* disallow configuration changes of online cpus and cpu 0 */
 860         if (cpu_online(cpu) || cpu == 0)
 861                 goto out;
 862         rc = 0;
 863         switch (val) {
 864         case 0:
 865                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 866                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 867                         if (!rc) {
 868                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 869                                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 870                         }
 871                 }
 872                 break;
 873         case 1:
 874                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 875                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 876                         if (!rc) {
 877                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 878                                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 879                         }
 880                 }
 881                 break;
 882         default:
 883                 break;
 884         }
 885 out:
 886         mutex_unlock(&smp_cpu_state_mutex);
 887         put_online_cpus();
 888         return rc ? rc : count;
 889 }
 890 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 891 #endif /* CONFIG_HOTPLUG_CPU */
 892
 893 static ssize_t show_cpu_address(struct sys_device *dev,
 894                                 struct sysdev_attribute *attr, char *buf)
 895 {
 896         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 897 }
 898 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 899
 900 static struct attribute *cpu_common_attrs[] = {
 901 #ifdef CONFIG_HOTPLUG_CPU
 902         &attr_configure.attr,
 903 #endif
 904         &attr_address.attr,
 905         NULL,
 906 };
 907
 908 static struct attribute_group cpu_common_attr_group = {
 909         .attrs = cpu_common_attrs,
 910 };
 911
 912 static ssize_t show_capability(struct sys_device *dev,
 913                                 struct sysdev_attribute *attr, char *buf)
 914 {
 915         unsigned int capability;
 916         int rc;
 917
 918         rc = get_cpu_capability(&capability);
 919         if (rc)
 920                 return rc;
 921         return sprintf(buf, "%u\n", capability);
 922 }
 923 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 924
 925 static ssize_t show_idle_count(struct sys_device *dev,
 926                                 struct sysdev_attribute *attr, char *buf)
 927 {
 928         struct s390_idle_data *idle;
 929         unsigned long long idle_count;
 930         unsigned int sequence;
 931
 932         idle = &per_cpu(s390_idle, dev->id);
 933 repeat:
 934         sequence = idle->sequence;
 935         smp_rmb();
 936         if (sequence & 1)
 937                 goto repeat;
 938         idle_count = idle->idle_count;
 939         if (idle->idle_enter)
 940                 idle_count++;
 941         smp_rmb();
 942         if (idle->sequence != sequence)
 943                 goto repeat;
 944         return sprintf(buf, "%llu\n", idle_count);
 945 }
 946 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 947
 948 static ssize_t show_idle_time(struct sys_device *dev,
 949                                 struct sysdev_attribute *attr, char *buf)
 950 {
 951         struct s390_idle_data *idle;
 952         unsigned long long now, idle_time, idle_enter;
 953         unsigned int sequence;
 954
 955         idle = &per_cpu(s390_idle, dev->id);
 956         now = get_clock();
 957 repeat:
 958         sequence = idle->sequence;
 959         smp_rmb();
 960         if (sequence & 1)
 961                 goto repeat;
 962         idle_time = idle->idle_time;
 963         idle_enter = idle->idle_enter;
 964         if (idle_enter != 0ULL && idle_enter < now)
 965                 idle_time += now - idle_enter;
 966         smp_rmb();
 967         if (idle->sequence != sequence)
 968                 goto repeat;
 969         return sprintf(buf, "%llu\n", idle_time >> 12);
 970 }
 971 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 972
 973 static struct attribute *cpu_online_attrs[] = {
 974         &attr_capability.attr,
 975         &attr_idle_count.attr,
 976         &attr_idle_time_us.attr,
 977         NULL,
 978 };
 979
 980 static struct attribute_group cpu_online_attr_group = {
 981         .attrs = cpu_online_attrs,
 982 };
 983
 984 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 985                                     unsigned long action, void *hcpu)
 986 {
 987         unsigned int cpu = (unsigned int)(long)hcpu;
 988         struct cpu *c = &per_cpu(cpu_devices, cpu);
 989         struct sys_device *s = &c->sysdev;
 990         struct s390_idle_data *idle;
 991         int err = 0;
 992
 993         switch (action) {
 994         case CPU_ONLINE:
 995         case CPU_ONLINE_FROZEN:
 996                 idle = &per_cpu(s390_idle, cpu);
 997                 memset(idle, 0, sizeof(struct s390_idle_data));
 998                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 999                 break;
1000         case CPU_DEAD:
1001         case CPU_DEAD_FROZEN:
1002                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1003                 break;
1004         }
1005         return notifier_from_errno(err);
1006 }
1007
1008 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1009         .notifier_call = smp_cpu_notify,
1010 };
1011
1012 static int __devinit smp_add_present_cpu(int cpu)
1013 {
1014         struct cpu *c = &per_cpu(cpu_devices, cpu);
1015         struct sys_device *s = &c->sysdev;
1016         int rc;
1017
1018         c->hotpluggable = 1;
1019         rc = register_cpu(c, cpu);
1020         if (rc)
1021                 goto out;
1022         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1023         if (rc)
1024                 goto out_cpu;
1025         if (cpu_online(cpu)) {
1026                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1027                 if (rc)
1028                         goto out_online;
1029         }
1030         rc = topology_cpu_init(c);
1031         if (rc)
1032                 goto out_topology;
1033         return 0;
1034
1035 out_topology:
1036         if (cpu_online(cpu))
1037                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1038 out_online:
1039         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1040 out_cpu:
1041 #ifdef CONFIG_HOTPLUG_CPU
1042         unregister_cpu(c);
1043 #endif
1044 out:
1045         return rc;
1046 }
1047
1048 #ifdef CONFIG_HOTPLUG_CPU
1049
1050 int __ref smp_rescan_cpus(void)
1051 {
1052         cpumask_t newcpus;
1053         int cpu;
1054         int rc;
1055
1056         get_online_cpus();
1057         mutex_lock(&smp_cpu_state_mutex);
1058         cpumask_copy(&newcpus, cpu_present_mask);
1059         rc = __smp_rescan_cpus();
1060         if (rc)
1061                 goto out;
1062         cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1063         for_each_cpu(cpu, &newcpus) {
1064                 rc = smp_add_present_cpu(cpu);
1065                 if (rc)
1066                         set_cpu_present(cpu, false);
1067         }
1068         rc = 0;
1069 out:
1070         mutex_unlock(&smp_cpu_state_mutex);
1071         put_online_cpus();
1072         if (!cpumask_empty(&newcpus))
1073                 topology_schedule_update();
1074         return rc;
1075 }
1076
1077 static ssize_t __ref rescan_store(struct sysdev_class *class,
1078                                   struct sysdev_class_attribute *attr,
1079                                   const char *buf,
1080                                   size_t count)
1081 {
1082         int rc;
1083
1084         rc = smp_rescan_cpus();
1085         return rc ? rc : count;
1086 }
1087 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1088 #endif /* CONFIG_HOTPLUG_CPU */
1089
1090 static int __init s390_smp_init(void)
1091 {
1092         int cpu, rc;
1093
1094         register_cpu_notifier(&smp_cpu_nb);
1095 #ifdef CONFIG_HOTPLUG_CPU
1096         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1097         if (rc)
1098                 return rc;
1099 #endif
1100         for_each_present_cpu(cpu) {
1101                 rc = smp_add_present_cpu(cpu);
1102                 if (rc)
1103                         return rc;
1104         }
1105         return 0;
1106 }
1107 subsys_initcall(s390_smp_init);