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 <asm/asm-offsets.h>
  42 #include <asm/ipl.h>
  43 #include <asm/setup.h>
  44 #include <asm/sigp.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/irq.h>
  47 #include <asm/cpcmd.h>
  48 #include <asm/tlbflush.h>
  49 #include <asm/timer.h>
  50 #include <asm/lowcore.h>
  51 #include <asm/sclp.h>
  52 #include <asm/cputime.h>
  53 #include <asm/vdso.h>
  54 #include <asm/cpu.h>
  55 #include "entry.h"
  56
  57 /* logical cpu to cpu address */
  58 unsigned short __cpu_logical_map[NR_CPUS];
  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 DEFINE_MUTEX(smp_cpu_state_mutex);
  71 int smp_cpu_polarization[NR_CPUS];
  72 static int smp_cpu_state[NR_CPUS];
  73 static int cpu_management;
  74
  75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  76
  77 static void smp_ext_bitcall(int, int);
  78
  79 static int raw_cpu_stopped(int cpu)
  80 {
  81         u32 status;
  82
  83         switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
  84         case sigp_status_stored:
  85                 /* Check for stopped and check stop state */
  86                 if (status & 0x50)
  87                         return 1;
  88                 break;
  89         default:
  90                 break;
  91         }
  92         return 0;
  93 }
  94
  95 static inline int cpu_stopped(int cpu)
  96 {
  97         return raw_cpu_stopped(cpu_logical_map(cpu));
  98 }
  99
 100 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
 101 {
 102         struct _lowcore *lc, *current_lc;
 103         struct stack_frame *sf;
 104         struct pt_regs *regs;
 105         unsigned long sp;
 106
 107         if (smp_processor_id() == 0)
 108                 func(data);
 109         __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
 110         /* Disable lowcore protection */
 111         __ctl_clear_bit(0, 28);
 112         current_lc = lowcore_ptr[smp_processor_id()];
 113         lc = lowcore_ptr[0];
 114         if (!lc)
 115                 lc = current_lc;
 116         lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 117         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
 118         if (!cpu_online(0))
 119                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
 120         while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
 121                 cpu_relax();
 122         sp = lc->panic_stack;
 123         sp -= sizeof(struct pt_regs);
 124         regs = (struct pt_regs *) sp;
 125         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
 126         regs->psw = lc->psw_save_area;
 127         sp -= STACK_FRAME_OVERHEAD;
 128         sf = (struct stack_frame *) sp;
 129         sf->back_chain = regs->gprs[15];
 130         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 131 }
 132
 133 void smp_send_stop(void)
 134 {
 135         int cpu, rc;
 136
 137         /* Disable all interrupts/machine checks */
 138         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 139         trace_hardirqs_off();
 140
 141         /* stop all processors */
 142         for_each_online_cpu(cpu) {
 143                 if (cpu == smp_processor_id())
 144                         continue;
 145                 do {
 146                         rc = sigp(cpu, sigp_stop);
 147                 } while (rc == sigp_busy);
 148
 149                 while (!cpu_stopped(cpu))
 150                         cpu_relax();
 151         }
 152 }
 153
 154 /*
 155  * This is the main routine where commands issued by other
 156  * cpus are handled.
 157  */
 158
 159 static void do_ext_call_interrupt(unsigned int ext_int_code,
 160                                   unsigned int param32, unsigned long param64)
 161 {
 162         unsigned long bits;
 163
 164         kstat_cpu(smp_processor_id()).irqs[EXTINT_IPI]++;
 165         /*
 166          * handle bit signal external calls
 167          */
 168         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 169
 170         if (test_bit(ec_schedule, &bits))
 171                 scheduler_ipi();
 172
 173         if (test_bit(ec_call_function, &bits))
 174                 generic_smp_call_function_interrupt();
 175
 176         if (test_bit(ec_call_function_single, &bits))
 177                 generic_smp_call_function_single_interrupt();
 178 }
 179
 180 /*
 181  * Send an external call sigp to another cpu and return without waiting
 182  * for its completion.
 183  */
 184 static void smp_ext_bitcall(int cpu, int sig)
 185 {
 186         /*
 187          * Set signaling bit in lowcore of target cpu and kick it
 188          */
 189         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 190         while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
 191                 udelay(10);
 192 }
 193
 194 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 195 {
 196         int cpu;
 197
 198         for_each_cpu(cpu, mask)
 199                 smp_ext_bitcall(cpu, ec_call_function);
 200 }
 201
 202 void arch_send_call_function_single_ipi(int cpu)
 203 {
 204         smp_ext_bitcall(cpu, ec_call_function_single);
 205 }
 206
 207 #ifndef CONFIG_64BIT
 208 /*
 209  * this function sends a 'purge tlb' signal to another CPU.
 210  */
 211 static void smp_ptlb_callback(void *info)
 212 {
 213         __tlb_flush_local();
 214 }
 215
 216 void smp_ptlb_all(void)
 217 {
 218         on_each_cpu(smp_ptlb_callback, NULL, 1);
 219 }
 220 EXPORT_SYMBOL(smp_ptlb_all);
 221 #endif /* ! CONFIG_64BIT */
 222
 223 /*
 224  * this function sends a 'reschedule' IPI to another CPU.
 225  * it goes straight through and wastes no time serializing
 226  * anything. Worst case is that we lose a reschedule ...
 227  */
 228 void smp_send_reschedule(int cpu)
 229 {
 230         smp_ext_bitcall(cpu, ec_schedule);
 231 }
 232
 233 /*
 234  * parameter area for the set/clear control bit callbacks
 235  */
 236 struct ec_creg_mask_parms {
 237         unsigned long orvals[16];
 238         unsigned long andvals[16];
 239 };
 240
 241 /*
 242  * callback for setting/clearing control bits
 243  */
 244 static void smp_ctl_bit_callback(void *info)
 245 {
 246         struct ec_creg_mask_parms *pp = info;
 247         unsigned long cregs[16];
 248         int i;
 249
 250         __ctl_store(cregs, 0, 15);
 251         for (i = 0; i <= 15; i++)
 252                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 253         __ctl_load(cregs, 0, 15);
 254 }
 255
 256 /*
 257  * Set a bit in a control register of all cpus
 258  */
 259 void smp_ctl_set_bit(int cr, int bit)
 260 {
 261         struct ec_creg_mask_parms parms;
 262
 263         memset(&parms.orvals, 0, sizeof(parms.orvals));
 264         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 265         parms.orvals[cr] = 1UL << bit;
 266         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 267 }
 268 EXPORT_SYMBOL(smp_ctl_set_bit);
 269
 270 /*
 271  * Clear a bit in a control register of all cpus
 272  */
 273 void smp_ctl_clear_bit(int cr, int bit)
 274 {
 275         struct ec_creg_mask_parms parms;
 276
 277         memset(&parms.orvals, 0, sizeof(parms.orvals));
 278         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 279         parms.andvals[cr] = ~(1UL << bit);
 280         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 281 }
 282 EXPORT_SYMBOL(smp_ctl_clear_bit);
 283
 284 #ifdef CONFIG_ZFCPDUMP
 285
 286 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 287 {
 288         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 289                 return;
 290         if (cpu >= NR_CPUS) {
 291                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 292                            "the dump\n", cpu, NR_CPUS - 1);
 293                 return;
 294         }
 295         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 296         while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
 297                 cpu_relax();
 298         memcpy_real(zfcpdump_save_areas[cpu],
 299                     (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 300                     sizeof(struct save_area));
 301 }
 302
 303 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 304 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 305
 306 #else
 307
 308 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 309
 310 #endif /* CONFIG_ZFCPDUMP */
 311
 312 static int cpu_known(int cpu_id)
 313 {
 314         int cpu;
 315
 316         for_each_present_cpu(cpu) {
 317                 if (__cpu_logical_map[cpu] == cpu_id)
 318                         return 1;
 319         }
 320         return 0;
 321 }
 322
 323 static int smp_rescan_cpus_sigp(cpumask_t avail)
 324 {
 325         int cpu_id, logical_cpu;
 326
 327         logical_cpu = cpumask_first(&avail);
 328         if (logical_cpu >= nr_cpu_ids)
 329                 return 0;
 330         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 331                 if (cpu_known(cpu_id))
 332                         continue;
 333                 __cpu_logical_map[logical_cpu] = cpu_id;
 334                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 335                 if (!cpu_stopped(logical_cpu))
 336                         continue;
 337                 set_cpu_present(logical_cpu, true);
 338                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 339                 logical_cpu = cpumask_next(logical_cpu, &avail);
 340                 if (logical_cpu >= nr_cpu_ids)
 341                         break;
 342         }
 343         return 0;
 344 }
 345
 346 static int smp_rescan_cpus_sclp(cpumask_t avail)
 347 {
 348         struct sclp_cpu_info *info;
 349         int cpu_id, logical_cpu, cpu;
 350         int rc;
 351
 352         logical_cpu = cpumask_first(&avail);
 353         if (logical_cpu >= nr_cpu_ids)
 354                 return 0;
 355         info = kmalloc(sizeof(*info), GFP_KERNEL);
 356         if (!info)
 357                 return -ENOMEM;
 358         rc = sclp_get_cpu_info(info);
 359         if (rc)
 360                 goto out;
 361         for (cpu = 0; cpu < info->combined; cpu++) {
 362                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 363                         continue;
 364                 cpu_id = info->cpu[cpu].address;
 365                 if (cpu_known(cpu_id))
 366                         continue;
 367                 __cpu_logical_map[logical_cpu] = cpu_id;
 368                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 369                 set_cpu_present(logical_cpu, true);
 370                 if (cpu >= info->configured)
 371                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 372                 else
 373                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 374                 logical_cpu = cpumask_next(logical_cpu, &avail);
 375                 if (logical_cpu >= nr_cpu_ids)
 376                         break;
 377         }
 378 out:
 379         kfree(info);
 380         return rc;
 381 }
 382
 383 static int __smp_rescan_cpus(void)
 384 {
 385         cpumask_t avail;
 386
 387         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 388         if (smp_use_sigp_detection)
 389                 return smp_rescan_cpus_sigp(avail);
 390         else
 391                 return smp_rescan_cpus_sclp(avail);
 392 }
 393
 394 static void __init smp_detect_cpus(void)
 395 {
 396         unsigned int cpu, c_cpus, s_cpus;
 397         struct sclp_cpu_info *info;
 398         u16 boot_cpu_addr, cpu_addr;
 399
 400         c_cpus = 1;
 401         s_cpus = 0;
 402         boot_cpu_addr = __cpu_logical_map[0];
 403         info = kmalloc(sizeof(*info), GFP_KERNEL);
 404         if (!info)
 405                 panic("smp_detect_cpus failed to allocate memory\n");
 406         /* Use sigp detection algorithm if sclp doesn't work. */
 407         if (sclp_get_cpu_info(info)) {
 408                 smp_use_sigp_detection = 1;
 409                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 410                         if (cpu == boot_cpu_addr)
 411                                 continue;
 412                         if (!raw_cpu_stopped(cpu))
 413                                 continue;
 414                         smp_get_save_area(c_cpus, cpu);
 415                         c_cpus++;
 416                 }
 417                 goto out;
 418         }
 419
 420         if (info->has_cpu_type) {
 421                 for (cpu = 0; cpu < info->combined; cpu++) {
 422                         if (info->cpu[cpu].address == boot_cpu_addr) {
 423                                 smp_cpu_type = info->cpu[cpu].type;
 424                                 break;
 425                         }
 426                 }
 427         }
 428
 429         for (cpu = 0; cpu < info->combined; cpu++) {
 430                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 431                         continue;
 432                 cpu_addr = info->cpu[cpu].address;
 433                 if (cpu_addr == boot_cpu_addr)
 434                         continue;
 435                 if (!raw_cpu_stopped(cpu_addr)) {
 436                         s_cpus++;
 437                         continue;
 438                 }
 439                 smp_get_save_area(c_cpus, cpu_addr);
 440                 c_cpus++;
 441         }
 442 out:
 443         kfree(info);
 444         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 445         get_online_cpus();
 446         __smp_rescan_cpus();
 447         put_online_cpus();
 448 }
 449
 450 /*
 451  *      Activate a secondary processor.
 452  */
 453 int __cpuinit start_secondary(void *cpuvoid)
 454 {
 455         /* Setup the cpu */
 456         cpu_init();
 457         preempt_disable();
 458         /* Enable TOD clock interrupts on the secondary cpu. */
 459         init_cpu_timer();
 460         /* Enable cpu timer interrupts on the secondary cpu. */
 461         init_cpu_vtimer();
 462         /* Enable pfault pseudo page faults on this cpu. */
 463         pfault_init();
 464
 465         /* call cpu notifiers */
 466         notify_cpu_starting(smp_processor_id());
 467         /* Mark this cpu as online */
 468         ipi_call_lock();
 469         set_cpu_online(smp_processor_id(), true);
 470         ipi_call_unlock();
 471         __ctl_clear_bit(0, 28); /* Disable lowcore protection */
 472         S390_lowcore.restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 473         S390_lowcore.restart_psw.addr =
 474                 PSW_ADDR_AMODE | (unsigned long) psw_restart_int_handler;
 475         __ctl_set_bit(0, 28); /* Enable lowcore protection */
 476         /*
 477          * Wait until the cpu which brought this one up marked it
 478          * active before enabling interrupts.
 479          */
 480         while (!cpumask_test_cpu(smp_processor_id(), cpu_active_mask))
 481                 cpu_relax();
 482         local_irq_enable();
 483         /* cpu_idle will call schedule for us */
 484         cpu_idle();
 485         return 0;
 486 }
 487
 488 struct create_idle {
 489         struct work_struct work;
 490         struct task_struct *idle;
 491         struct completion done;
 492         int cpu;
 493 };
 494
 495 static void __cpuinit smp_fork_idle(struct work_struct *work)
 496 {
 497         struct create_idle *c_idle;
 498
 499         c_idle = container_of(work, struct create_idle, work);
 500         c_idle->idle = fork_idle(c_idle->cpu);
 501         complete(&c_idle->done);
 502 }
 503
 504 static int __cpuinit smp_alloc_lowcore(int cpu)
 505 {
 506         unsigned long async_stack, panic_stack;
 507         struct _lowcore *lowcore;
 508
 509         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 510         if (!lowcore)
 511                 return -ENOMEM;
 512         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 513         panic_stack = __get_free_page(GFP_KERNEL);
 514         if (!panic_stack || !async_stack)
 515                 goto out;
 516         memcpy(lowcore, &S390_lowcore, 512);
 517         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 518         lowcore->async_stack = async_stack + ASYNC_SIZE;
 519         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 520         lowcore->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 521         lowcore->restart_psw.addr =
 522                 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
 523         if (user_mode != HOME_SPACE_MODE)
 524                 lowcore->restart_psw.mask |= PSW_ASC_HOME;
 525 #ifndef CONFIG_64BIT
 526         if (MACHINE_HAS_IEEE) {
 527                 unsigned long save_area;
 528
 529                 save_area = get_zeroed_page(GFP_KERNEL);
 530                 if (!save_area)
 531                         goto out;
 532                 lowcore->extended_save_area_addr = (u32) save_area;
 533         }
 534 #else
 535         if (vdso_alloc_per_cpu(cpu, lowcore))
 536                 goto out;
 537 #endif
 538         lowcore_ptr[cpu] = lowcore;
 539         return 0;
 540
 541 out:
 542         free_page(panic_stack);
 543         free_pages(async_stack, ASYNC_ORDER);
 544         free_pages((unsigned long) lowcore, LC_ORDER);
 545         return -ENOMEM;
 546 }
 547
 548 static void smp_free_lowcore(int cpu)
 549 {
 550         struct _lowcore *lowcore;
 551
 552         lowcore = lowcore_ptr[cpu];
 553 #ifndef CONFIG_64BIT
 554         if (MACHINE_HAS_IEEE)
 555                 free_page((unsigned long) lowcore->extended_save_area_addr);
 556 #else
 557         vdso_free_per_cpu(cpu, lowcore);
 558 #endif
 559         free_page(lowcore->panic_stack - PAGE_SIZE);
 560         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 561         free_pages((unsigned long) lowcore, LC_ORDER);
 562         lowcore_ptr[cpu] = NULL;
 563 }
 564
 565 /* Upping and downing of CPUs */
 566 int __cpuinit __cpu_up(unsigned int cpu)
 567 {
 568         struct _lowcore *cpu_lowcore;
 569         struct create_idle c_idle;
 570         struct task_struct *idle;
 571         struct stack_frame *sf;
 572         u32 lowcore;
 573         int ccode;
 574
 575         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 576                 return -EIO;
 577         idle = current_set[cpu];
 578         if (!idle) {
 579                 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
 580                 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
 581                 c_idle.cpu = cpu;
 582                 schedule_work(&c_idle.work);
 583                 wait_for_completion(&c_idle.done);
 584                 if (IS_ERR(c_idle.idle))
 585                         return PTR_ERR(c_idle.idle);
 586                 idle = c_idle.idle;
 587                 current_set[cpu] = c_idle.idle;
 588         }
 589         init_idle(idle, cpu);
 590         if (smp_alloc_lowcore(cpu))
 591                 return -ENOMEM;
 592         do {
 593                 ccode = sigp(cpu, sigp_initial_cpu_reset);
 594                 if (ccode == sigp_busy)
 595                         udelay(10);
 596                 if (ccode == sigp_not_operational)
 597                         goto err_out;
 598         } while (ccode == sigp_busy);
 599
 600         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 601         while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 602                 udelay(10);
 603
 604         cpu_lowcore = lowcore_ptr[cpu];
 605         cpu_lowcore->kernel_stack = (unsigned long)
 606                 task_stack_page(idle) + THREAD_SIZE;
 607         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 608         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 609                                      - sizeof(struct pt_regs)
 610                                      - sizeof(struct stack_frame));
 611         memset(sf, 0, sizeof(struct stack_frame));
 612         sf->gprs[9] = (unsigned long) sf;
 613         cpu_lowcore->save_area[15] = (unsigned long) sf;
 614         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 615         atomic_inc(&init_mm.context.attach_count);
 616         asm volatile(
 617                 "       stam    0,15,0(%0)"
 618                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 619         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 620         cpu_lowcore->current_task = (unsigned long) idle;
 621         cpu_lowcore->cpu_nr = cpu;
 622         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 623         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 624         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 625         memcpy(cpu_lowcore->stfle_fac_list, S390_lowcore.stfle_fac_list,
 626                MAX_FACILITY_BIT/8);
 627         eieio();
 628
 629         while (sigp(cpu, sigp_restart) == sigp_busy)
 630                 udelay(10);
 631
 632         while (!cpu_online(cpu))
 633                 cpu_relax();
 634         return 0;
 635
 636 err_out:
 637         smp_free_lowcore(cpu);
 638         return -EIO;
 639 }
 640
 641 static int __init setup_possible_cpus(char *s)
 642 {
 643         int pcpus, cpu;
 644
 645         pcpus = simple_strtoul(s, NULL, 0);
 646         init_cpu_possible(cpumask_of(0));
 647         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 648                 set_cpu_possible(cpu, true);
 649         return 0;
 650 }
 651 early_param("possible_cpus", setup_possible_cpus);
 652
 653 #ifdef CONFIG_HOTPLUG_CPU
 654
 655 int __cpu_disable(void)
 656 {
 657         struct ec_creg_mask_parms cr_parms;
 658         int cpu = smp_processor_id();
 659
 660         set_cpu_online(cpu, false);
 661
 662         /* Disable pfault pseudo page faults on this cpu. */
 663         pfault_fini();
 664
 665         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 666         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 667
 668         /* disable all external interrupts */
 669         cr_parms.orvals[0] = 0;
 670         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 11 |
 671                                 1 << 10 | 1 <<  9 | 1 <<  6 | 1 <<  5 |
 672                                 1 <<  4);
 673         /* disable all I/O interrupts */
 674         cr_parms.orvals[6] = 0;
 675         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 676                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 677         /* disable most machine checks */
 678         cr_parms.orvals[14] = 0;
 679         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 680                                  1 << 25 | 1 << 24);
 681
 682         smp_ctl_bit_callback(&cr_parms);
 683
 684         return 0;
 685 }
 686
 687 void __cpu_die(unsigned int cpu)
 688 {
 689         /* Wait until target cpu is down */
 690         while (!cpu_stopped(cpu))
 691                 cpu_relax();
 692         while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
 693                 udelay(10);
 694         smp_free_lowcore(cpu);
 695         atomic_dec(&init_mm.context.attach_count);
 696 }
 697
 698 void __noreturn cpu_die(void)
 699 {
 700         idle_task_exit();
 701         while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
 702                 cpu_relax();
 703         for (;;);
 704 }
 705
 706 #endif /* CONFIG_HOTPLUG_CPU */
 707
 708 void __init smp_prepare_cpus(unsigned int max_cpus)
 709 {
 710 #ifndef CONFIG_64BIT
 711         unsigned long save_area = 0;
 712 #endif
 713         unsigned long async_stack, panic_stack;
 714         struct _lowcore *lowcore;
 715
 716         smp_detect_cpus();
 717
 718         /* request the 0x1201 emergency signal external interrupt */
 719         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 720                 panic("Couldn't request external interrupt 0x1201");
 721
 722         /* Reallocate current lowcore, but keep its contents. */
 723         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 724         panic_stack = __get_free_page(GFP_KERNEL);
 725         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 726         BUG_ON(!lowcore || !panic_stack || !async_stack);
 727 #ifndef CONFIG_64BIT
 728         if (MACHINE_HAS_IEEE)
 729                 save_area = get_zeroed_page(GFP_KERNEL);
 730 #endif
 731         local_irq_disable();
 732         local_mcck_disable();
 733         lowcore_ptr[smp_processor_id()] = lowcore;
 734         *lowcore = S390_lowcore;
 735         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 736         lowcore->async_stack = async_stack + ASYNC_SIZE;
 737 #ifndef CONFIG_64BIT
 738         if (MACHINE_HAS_IEEE)
 739                 lowcore->extended_save_area_addr = (u32) save_area;
 740 #endif
 741         set_prefix((u32)(unsigned long) lowcore);
 742         local_mcck_enable();
 743         local_irq_enable();
 744 #ifdef CONFIG_64BIT
 745         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 746                 BUG();
 747 #endif
 748 }
 749
 750 void __init smp_prepare_boot_cpu(void)
 751 {
 752         BUG_ON(smp_processor_id() != 0);
 753
 754         current_thread_info()->cpu = 0;
 755         set_cpu_present(0, true);
 756         set_cpu_online(0, true);
 757         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 758         current_set[0] = current;
 759         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 760         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 761 }
 762
 763 void __init smp_cpus_done(unsigned int max_cpus)
 764 {
 765 }
 766
 767 void __init smp_setup_processor_id(void)
 768 {
 769         S390_lowcore.cpu_nr = 0;
 770         __cpu_logical_map[0] = stap();
 771 }
 772
 773 /*
 774  * the frequency of the profiling timer can be changed
 775  * by writing a multiplier value into /proc/profile.
 776  *
 777  * usually you want to run this on all CPUs ;)
 778  */
 779 int setup_profiling_timer(unsigned int multiplier)
 780 {
 781         return 0;
 782 }
 783
 784 #ifdef CONFIG_HOTPLUG_CPU
 785 static ssize_t cpu_configure_show(struct sys_device *dev,
 786                                 struct sysdev_attribute *attr, char *buf)
 787 {
 788         ssize_t count;
 789
 790         mutex_lock(&smp_cpu_state_mutex);
 791         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 792         mutex_unlock(&smp_cpu_state_mutex);
 793         return count;
 794 }
 795
 796 static ssize_t cpu_configure_store(struct sys_device *dev,
 797                                   struct sysdev_attribute *attr,
 798                                   const char *buf, size_t count)
 799 {
 800         int cpu = dev->id;
 801         int val, rc;
 802         char delim;
 803
 804         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 805                 return -EINVAL;
 806         if (val != 0 && val != 1)
 807                 return -EINVAL;
 808
 809         get_online_cpus();
 810         mutex_lock(&smp_cpu_state_mutex);
 811         rc = -EBUSY;
 812         /* disallow configuration changes of online cpus and cpu 0 */
 813         if (cpu_online(cpu) || cpu == 0)
 814                 goto out;
 815         rc = 0;
 816         switch (val) {
 817         case 0:
 818                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 819                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 820                         if (!rc) {
 821                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 822                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 823                         }
 824                 }
 825                 break;
 826         case 1:
 827                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 828                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 829                         if (!rc) {
 830                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 831                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 832                         }
 833                 }
 834                 break;
 835         default:
 836                 break;
 837         }
 838 out:
 839         mutex_unlock(&smp_cpu_state_mutex);
 840         put_online_cpus();
 841         return rc ? rc : count;
 842 }
 843 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 844 #endif /* CONFIG_HOTPLUG_CPU */
 845
 846 static ssize_t cpu_polarization_show(struct sys_device *dev,
 847                                      struct sysdev_attribute *attr, char *buf)
 848 {
 849         int cpu = dev->id;
 850         ssize_t count;
 851
 852         mutex_lock(&smp_cpu_state_mutex);
 853         switch (smp_cpu_polarization[cpu]) {
 854         case POLARIZATION_HRZ:
 855                 count = sprintf(buf, "horizontal\n");
 856                 break;
 857         case POLARIZATION_VL:
 858                 count = sprintf(buf, "vertical:low\n");
 859                 break;
 860         case POLARIZATION_VM:
 861                 count = sprintf(buf, "vertical:medium\n");
 862                 break;
 863         case POLARIZATION_VH:
 864                 count = sprintf(buf, "vertical:high\n");
 865                 break;
 866         default:
 867                 count = sprintf(buf, "unknown\n");
 868                 break;
 869         }
 870         mutex_unlock(&smp_cpu_state_mutex);
 871         return count;
 872 }
 873 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 874
 875 static ssize_t show_cpu_address(struct sys_device *dev,
 876                                 struct sysdev_attribute *attr, char *buf)
 877 {
 878         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 879 }
 880 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 881
 882
 883 static struct attribute *cpu_common_attrs[] = {
 884 #ifdef CONFIG_HOTPLUG_CPU
 885         &attr_configure.attr,
 886 #endif
 887         &attr_address.attr,
 888         &attr_polarization.attr,
 889         NULL,
 890 };
 891
 892 static struct attribute_group cpu_common_attr_group = {
 893         .attrs = cpu_common_attrs,
 894 };
 895
 896 static ssize_t show_capability(struct sys_device *dev,
 897                                 struct sysdev_attribute *attr, char *buf)
 898 {
 899         unsigned int capability;
 900         int rc;
 901
 902         rc = get_cpu_capability(&capability);
 903         if (rc)
 904                 return rc;
 905         return sprintf(buf, "%u\n", capability);
 906 }
 907 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 908
 909 static ssize_t show_idle_count(struct sys_device *dev,
 910                                 struct sysdev_attribute *attr, char *buf)
 911 {
 912         struct s390_idle_data *idle;
 913         unsigned long long idle_count;
 914         unsigned int sequence;
 915
 916         idle = &per_cpu(s390_idle, dev->id);
 917 repeat:
 918         sequence = idle->sequence;
 919         smp_rmb();
 920         if (sequence & 1)
 921                 goto repeat;
 922         idle_count = idle->idle_count;
 923         if (idle->idle_enter)
 924                 idle_count++;
 925         smp_rmb();
 926         if (idle->sequence != sequence)
 927                 goto repeat;
 928         return sprintf(buf, "%llu\n", idle_count);
 929 }
 930 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 931
 932 static ssize_t show_idle_time(struct sys_device *dev,
 933                                 struct sysdev_attribute *attr, char *buf)
 934 {
 935         struct s390_idle_data *idle;
 936         unsigned long long now, idle_time, idle_enter;
 937         unsigned int sequence;
 938
 939         idle = &per_cpu(s390_idle, dev->id);
 940         now = get_clock();
 941 repeat:
 942         sequence = idle->sequence;
 943         smp_rmb();
 944         if (sequence & 1)
 945                 goto repeat;
 946         idle_time = idle->idle_time;
 947         idle_enter = idle->idle_enter;
 948         if (idle_enter != 0ULL && idle_enter < now)
 949                 idle_time += now - idle_enter;
 950         smp_rmb();
 951         if (idle->sequence != sequence)
 952                 goto repeat;
 953         return sprintf(buf, "%llu\n", idle_time >> 12);
 954 }
 955 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 956
 957 static struct attribute *cpu_online_attrs[] = {
 958         &attr_capability.attr,
 959         &attr_idle_count.attr,
 960         &attr_idle_time_us.attr,
 961         NULL,
 962 };
 963
 964 static struct attribute_group cpu_online_attr_group = {
 965         .attrs = cpu_online_attrs,
 966 };
 967
 968 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 969                                     unsigned long action, void *hcpu)
 970 {
 971         unsigned int cpu = (unsigned int)(long)hcpu;
 972         struct cpu *c = &per_cpu(cpu_devices, cpu);
 973         struct sys_device *s = &c->sysdev;
 974         struct s390_idle_data *idle;
 975         int err = 0;
 976
 977         switch (action) {
 978         case CPU_ONLINE:
 979         case CPU_ONLINE_FROZEN:
 980                 idle = &per_cpu(s390_idle, cpu);
 981                 memset(idle, 0, sizeof(struct s390_idle_data));
 982                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 983                 break;
 984         case CPU_DEAD:
 985         case CPU_DEAD_FROZEN:
 986                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 987                 break;
 988         }
 989         return notifier_from_errno(err);
 990 }
 991
 992 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 993         .notifier_call = smp_cpu_notify,
 994 };
 995
 996 static int __devinit smp_add_present_cpu(int cpu)
 997 {
 998         struct cpu *c = &per_cpu(cpu_devices, cpu);
 999         struct sys_device *s = &c->sysdev;
1000         int rc;
1001
1002         c->hotpluggable = 1;
1003         rc = register_cpu(c, cpu);
1004         if (rc)
1005                 goto out;
1006         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1007         if (rc)
1008                 goto out_cpu;
1009         if (!cpu_online(cpu))
1010                 goto out;
1011         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1012         if (!rc)
1013                 return 0;
1014         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1015 out_cpu:
1016 #ifdef CONFIG_HOTPLUG_CPU
1017         unregister_cpu(c);
1018 #endif
1019 out:
1020         return rc;
1021 }
1022
1023 #ifdef CONFIG_HOTPLUG_CPU
1024
1025 int __ref smp_rescan_cpus(void)
1026 {
1027         cpumask_t newcpus;
1028         int cpu;
1029         int rc;
1030
1031         get_online_cpus();
1032         mutex_lock(&smp_cpu_state_mutex);
1033         cpumask_copy(&newcpus, cpu_present_mask);
1034         rc = __smp_rescan_cpus();
1035         if (rc)
1036                 goto out;
1037         cpumask_andnot(&newcpus, cpu_present_mask, &newcpus);
1038         for_each_cpu(cpu, &newcpus) {
1039                 rc = smp_add_present_cpu(cpu);
1040                 if (rc)
1041                         set_cpu_present(cpu, false);
1042         }
1043         rc = 0;
1044 out:
1045         mutex_unlock(&smp_cpu_state_mutex);
1046         put_online_cpus();
1047         if (!cpumask_empty(&newcpus))
1048                 topology_schedule_update();
1049         return rc;
1050 }
1051
1052 static ssize_t __ref rescan_store(struct sysdev_class *class,
1053                                   struct sysdev_class_attribute *attr,
1054                                   const char *buf,
1055                                   size_t count)
1056 {
1057         int rc;
1058
1059         rc = smp_rescan_cpus();
1060         return rc ? rc : count;
1061 }
1062 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1063 #endif /* CONFIG_HOTPLUG_CPU */
1064
1065 static ssize_t dispatching_show(struct sysdev_class *class,
1066                                 struct sysdev_class_attribute *attr,
1067                                 char *buf)
1068 {
1069         ssize_t count;
1070
1071         mutex_lock(&smp_cpu_state_mutex);
1072         count = sprintf(buf, "%d\n", cpu_management);
1073         mutex_unlock(&smp_cpu_state_mutex);
1074         return count;
1075 }
1076
1077 static ssize_t dispatching_store(struct sysdev_class *dev,
1078                                  struct sysdev_class_attribute *attr,
1079                                  const char *buf,
1080                                  size_t count)
1081 {
1082         int val, rc;
1083         char delim;
1084
1085         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1086                 return -EINVAL;
1087         if (val != 0 && val != 1)
1088                 return -EINVAL;
1089         rc = 0;
1090         get_online_cpus();
1091         mutex_lock(&smp_cpu_state_mutex);
1092         if (cpu_management == val)
1093                 goto out;
1094         rc = topology_set_cpu_management(val);
1095         if (!rc)
1096                 cpu_management = val;
1097 out:
1098         mutex_unlock(&smp_cpu_state_mutex);
1099         put_online_cpus();
1100         return rc ? rc : count;
1101 }
1102 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1103                          dispatching_store);
1104
1105 static int __init topology_init(void)
1106 {
1107         int cpu;
1108         int rc;
1109
1110         register_cpu_notifier(&smp_cpu_nb);
1111
1112 #ifdef CONFIG_HOTPLUG_CPU
1113         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1114         if (rc)
1115                 return rc;
1116 #endif
1117         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1118         if (rc)
1119                 return rc;
1120         for_each_present_cpu(cpu) {
1121                 rc = smp_add_present_cpu(cpu);
1122                 if (rc)
1123                         return rc;
1124         }
1125         return 0;
1126 }
1127 subsys_initcall(topology_init);