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 #define KMSG_COMPONENT "cpu"
  24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  25
  26 #include <linux/module.h>
  27 #include <linux/init.h>
  28 #include <linux/mm.h>
  29 #include <linux/err.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/kernel_stat.h>
  32 #include <linux/delay.h>
  33 #include <linux/cache.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/cpu.h>
  36 #include <linux/timex.h>
  37 #include <linux/bootmem.h>
  38 #include <asm/ipl.h>
  39 #include <asm/setup.h>
  40 #include <asm/sigp.h>
  41 #include <asm/pgalloc.h>
  42 #include <asm/irq.h>
  43 #include <asm/s390_ext.h>
  44 #include <asm/cpcmd.h>
  45 #include <asm/tlbflush.h>
  46 #include <asm/timer.h>
  47 #include <asm/lowcore.h>
  48 #include <asm/sclp.h>
  49 #include <asm/cpu.h>
  50 #include <asm/vdso.h>
  51 #include "entry.h"
  52
  53 /*
  54  * An array with a pointer the lowcore of every CPU.
  55  */
  56 struct _lowcore *lowcore_ptr[NR_CPUS];
  57 EXPORT_SYMBOL(lowcore_ptr);
  58
  59 cpumask_t cpu_online_map = CPU_MASK_NONE;
  60 EXPORT_SYMBOL(cpu_online_map);
  61
  62 cpumask_t cpu_possible_map = CPU_MASK_ALL;
  63 EXPORT_SYMBOL(cpu_possible_map);
  64
  65 static struct task_struct *current_set[NR_CPUS];
  66
  67 static u8 smp_cpu_type;
  68 static int smp_use_sigp_detection;
  69
  70 enum s390_cpu_state {
  71         CPU_STATE_STANDBY,
  72         CPU_STATE_CONFIGURED,
  73 };
  74
  75 DEFINE_MUTEX(smp_cpu_state_mutex);
  76 int smp_cpu_polarization[NR_CPUS];
  77 static int smp_cpu_state[NR_CPUS];
  78 static int cpu_management;
  79
  80 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  81
  82 static void smp_ext_bitcall(int, ec_bit_sig);
  83
  84 void smp_send_stop(void)
  85 {
  86         int cpu, rc;
  87
  88         /* Disable all interrupts/machine checks */
  89         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  90
  91         /* write magic number to zero page (absolute 0) */
  92         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
  93
  94         /* stop all processors */
  95         for_each_online_cpu(cpu) {
  96                 if (cpu == smp_processor_id())
  97                         continue;
  98                 do {
  99                         rc = signal_processor(cpu, sigp_stop);
 100                 } while (rc == sigp_busy);
 101
 102                 while (!smp_cpu_not_running(cpu))
 103                         cpu_relax();
 104         }
 105 }
 106
 107 /*
 108  * This is the main routine where commands issued by other
 109  * cpus are handled.
 110  */
 111
 112 static void do_ext_call_interrupt(__u16 code)
 113 {
 114         unsigned long bits;
 115
 116         /*
 117          * handle bit signal external calls
 118          *
 119          * For the ec_schedule signal we have to do nothing. All the work
 120          * is done automatically when we return from the interrupt.
 121          */
 122         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 123
 124         if (test_bit(ec_call_function, &bits))
 125                 generic_smp_call_function_interrupt();
 126
 127         if (test_bit(ec_call_function_single, &bits))
 128                 generic_smp_call_function_single_interrupt();
 129 }
 130
 131 /*
 132  * Send an external call sigp to another cpu and return without waiting
 133  * for its completion.
 134  */
 135 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 136 {
 137         /*
 138          * Set signaling bit in lowcore of target cpu and kick it
 139          */
 140         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 141         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 142                 udelay(10);
 143 }
 144
 145 void arch_send_call_function_ipi(cpumask_t mask)
 146 {
 147         int cpu;
 148
 149         for_each_cpu_mask(cpu, mask)
 150                 smp_ext_bitcall(cpu, ec_call_function);
 151 }
 152
 153 void arch_send_call_function_single_ipi(int cpu)
 154 {
 155         smp_ext_bitcall(cpu, ec_call_function_single);
 156 }
 157
 158 #ifndef CONFIG_64BIT
 159 /*
 160  * this function sends a 'purge tlb' signal to another CPU.
 161  */
 162 static void smp_ptlb_callback(void *info)
 163 {
 164         __tlb_flush_local();
 165 }
 166
 167 void smp_ptlb_all(void)
 168 {
 169         on_each_cpu(smp_ptlb_callback, NULL, 1);
 170 }
 171 EXPORT_SYMBOL(smp_ptlb_all);
 172 #endif /* ! CONFIG_64BIT */
 173
 174 /*
 175  * this function sends a 'reschedule' IPI to another CPU.
 176  * it goes straight through and wastes no time serializing
 177  * anything. Worst case is that we lose a reschedule ...
 178  */
 179 void smp_send_reschedule(int cpu)
 180 {
 181         smp_ext_bitcall(cpu, ec_schedule);
 182 }
 183
 184 /*
 185  * parameter area for the set/clear control bit callbacks
 186  */
 187 struct ec_creg_mask_parms {
 188         unsigned long orvals[16];
 189         unsigned long andvals[16];
 190 };
 191
 192 /*
 193  * callback for setting/clearing control bits
 194  */
 195 static void smp_ctl_bit_callback(void *info)
 196 {
 197         struct ec_creg_mask_parms *pp = info;
 198         unsigned long cregs[16];
 199         int i;
 200
 201         __ctl_store(cregs, 0, 15);
 202         for (i = 0; i <= 15; i++)
 203                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 204         __ctl_load(cregs, 0, 15);
 205 }
 206
 207 /*
 208  * Set a bit in a control register of all cpus
 209  */
 210 void smp_ctl_set_bit(int cr, int bit)
 211 {
 212         struct ec_creg_mask_parms parms;
 213
 214         memset(&parms.orvals, 0, sizeof(parms.orvals));
 215         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 216         parms.orvals[cr] = 1 << bit;
 217         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 218 }
 219 EXPORT_SYMBOL(smp_ctl_set_bit);
 220
 221 /*
 222  * Clear a bit in a control register of all cpus
 223  */
 224 void smp_ctl_clear_bit(int cr, int bit)
 225 {
 226         struct ec_creg_mask_parms parms;
 227
 228         memset(&parms.orvals, 0, sizeof(parms.orvals));
 229         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 230         parms.andvals[cr] = ~(1L << bit);
 231         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 232 }
 233 EXPORT_SYMBOL(smp_ctl_clear_bit);
 234
 235 /*
 236  * In early ipl state a temp. logically cpu number is needed, so the sigp
 237  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 238  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 239  */
 240 #define CPU_INIT_NO     1
 241
 242 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 243
 244 /*
 245  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 246  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 247  * save its prefix registers, since they get lost, when switching from 31 bit
 248  * to 64 bit.
 249  */
 250 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 251         __attribute__((__section__(".data")));
 252
 253 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 254 {
 255         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 256                 return;
 257         if (cpu >= NR_CPUS) {
 258                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 259                            "the dump\n", cpu, NR_CPUS - 1);
 260                 return;
 261         }
 262         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 263         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 264         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 265                sigp_busy)
 266                 cpu_relax();
 267         memcpy(zfcpdump_save_areas[cpu],
 268                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 269                SAVE_AREA_SIZE);
 270 #ifdef CONFIG_64BIT
 271         /* copy original prefix register */
 272         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 273 #endif
 274 }
 275
 276 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 277 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 278
 279 #else
 280
 281 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 282
 283 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 284
 285 static int cpu_stopped(int cpu)
 286 {
 287         __u32 status;
 288
 289         /* Check for stopped state */
 290         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 291             sigp_status_stored) {
 292                 if (status & 0x40)
 293                         return 1;
 294         }
 295         return 0;
 296 }
 297
 298 static int cpu_known(int cpu_id)
 299 {
 300         int cpu;
 301
 302         for_each_present_cpu(cpu) {
 303                 if (__cpu_logical_map[cpu] == cpu_id)
 304                         return 1;
 305         }
 306         return 0;
 307 }
 308
 309 static int smp_rescan_cpus_sigp(cpumask_t avail)
 310 {
 311         int cpu_id, logical_cpu;
 312
 313         logical_cpu = first_cpu(avail);
 314         if (logical_cpu == NR_CPUS)
 315                 return 0;
 316         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 317                 if (cpu_known(cpu_id))
 318                         continue;
 319                 __cpu_logical_map[logical_cpu] = cpu_id;
 320                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 321                 if (!cpu_stopped(logical_cpu))
 322                         continue;
 323                 cpu_set(logical_cpu, cpu_present_map);
 324                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 325                 logical_cpu = next_cpu(logical_cpu, avail);
 326                 if (logical_cpu == NR_CPUS)
 327                         break;
 328         }
 329         return 0;
 330 }
 331
 332 static int smp_rescan_cpus_sclp(cpumask_t avail)
 333 {
 334         struct sclp_cpu_info *info;
 335         int cpu_id, logical_cpu, cpu;
 336         int rc;
 337
 338         logical_cpu = first_cpu(avail);
 339         if (logical_cpu == NR_CPUS)
 340                 return 0;
 341         info = kmalloc(sizeof(*info), GFP_KERNEL);
 342         if (!info)
 343                 return -ENOMEM;
 344         rc = sclp_get_cpu_info(info);
 345         if (rc)
 346                 goto out;
 347         for (cpu = 0; cpu < info->combined; cpu++) {
 348                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 349                         continue;
 350                 cpu_id = info->cpu[cpu].address;
 351                 if (cpu_known(cpu_id))
 352                         continue;
 353                 __cpu_logical_map[logical_cpu] = cpu_id;
 354                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 355                 cpu_set(logical_cpu, cpu_present_map);
 356                 if (cpu >= info->configured)
 357                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 358                 else
 359                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 360                 logical_cpu = next_cpu(logical_cpu, avail);
 361                 if (logical_cpu == NR_CPUS)
 362                         break;
 363         }
 364 out:
 365         kfree(info);
 366         return rc;
 367 }
 368
 369 static int __smp_rescan_cpus(void)
 370 {
 371         cpumask_t avail;
 372
 373         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 374         if (smp_use_sigp_detection)
 375                 return smp_rescan_cpus_sigp(avail);
 376         else
 377                 return smp_rescan_cpus_sclp(avail);
 378 }
 379
 380 static void __init smp_detect_cpus(void)
 381 {
 382         unsigned int cpu, c_cpus, s_cpus;
 383         struct sclp_cpu_info *info;
 384         u16 boot_cpu_addr, cpu_addr;
 385
 386         c_cpus = 1;
 387         s_cpus = 0;
 388         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 389         info = kmalloc(sizeof(*info), GFP_KERNEL);
 390         if (!info)
 391                 panic("smp_detect_cpus failed to allocate memory\n");
 392         /* Use sigp detection algorithm if sclp doesn't work. */
 393         if (sclp_get_cpu_info(info)) {
 394                 smp_use_sigp_detection = 1;
 395                 for (cpu = 0; cpu <= 65535; cpu++) {
 396                         if (cpu == boot_cpu_addr)
 397                                 continue;
 398                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 399                         if (!cpu_stopped(CPU_INIT_NO))
 400                                 continue;
 401                         smp_get_save_area(c_cpus, cpu);
 402                         c_cpus++;
 403                 }
 404                 goto out;
 405         }
 406
 407         if (info->has_cpu_type) {
 408                 for (cpu = 0; cpu < info->combined; cpu++) {
 409                         if (info->cpu[cpu].address == boot_cpu_addr) {
 410                                 smp_cpu_type = info->cpu[cpu].type;
 411                                 break;
 412                         }
 413                 }
 414         }
 415
 416         for (cpu = 0; cpu < info->combined; cpu++) {
 417                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 418                         continue;
 419                 cpu_addr = info->cpu[cpu].address;
 420                 if (cpu_addr == boot_cpu_addr)
 421                         continue;
 422                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 423                 if (!cpu_stopped(CPU_INIT_NO)) {
 424                         s_cpus++;
 425                         continue;
 426                 }
 427                 smp_get_save_area(c_cpus, cpu_addr);
 428                 c_cpus++;
 429         }
 430 out:
 431         kfree(info);
 432         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 433         get_online_cpus();
 434         __smp_rescan_cpus();
 435         put_online_cpus();
 436 }
 437
 438 /*
 439  *      Activate a secondary processor.
 440  */
 441 int __cpuinit start_secondary(void *cpuvoid)
 442 {
 443         /* Setup the cpu */
 444         cpu_init();
 445         preempt_disable();
 446         /* Enable TOD clock interrupts on the secondary cpu. */
 447         init_cpu_timer();
 448         /* Enable cpu timer interrupts on the secondary cpu. */
 449         init_cpu_vtimer();
 450         /* Enable pfault pseudo page faults on this cpu. */
 451         pfault_init();
 452
 453         /* call cpu notifiers */
 454         notify_cpu_starting(smp_processor_id());
 455         /* Mark this cpu as online */
 456         ipi_call_lock();
 457         cpu_set(smp_processor_id(), cpu_online_map);
 458         ipi_call_unlock();
 459         /* Switch on interrupts */
 460         local_irq_enable();
 461         /* Print info about this processor */
 462         print_cpu_info(&S390_lowcore.cpu_data);
 463         /* cpu_idle will call schedule for us */
 464         cpu_idle();
 465         return 0;
 466 }
 467
 468 static void __init smp_create_idle(unsigned int cpu)
 469 {
 470         struct task_struct *p;
 471
 472         /*
 473          *  don't care about the psw and regs settings since we'll never
 474          *  reschedule the forked task.
 475          */
 476         p = fork_idle(cpu);
 477         if (IS_ERR(p))
 478                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 479         current_set[cpu] = p;
 480 }
 481
 482 static int __cpuinit smp_alloc_lowcore(int cpu)
 483 {
 484         unsigned long async_stack, panic_stack;
 485         struct _lowcore *lowcore;
 486         int lc_order;
 487
 488         lc_order = sizeof(long) == 8 ? 1 : 0;
 489         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 490         if (!lowcore)
 491                 return -ENOMEM;
 492         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 493         panic_stack = __get_free_page(GFP_KERNEL);
 494         if (!panic_stack || !async_stack)
 495                 goto out;
 496         memcpy(lowcore, &S390_lowcore, 512);
 497         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 498         lowcore->async_stack = async_stack + ASYNC_SIZE;
 499         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 500
 501 #ifndef CONFIG_64BIT
 502         if (MACHINE_HAS_IEEE) {
 503                 unsigned long save_area;
 504
 505                 save_area = get_zeroed_page(GFP_KERNEL);
 506                 if (!save_area)
 507                         goto out;
 508                 lowcore->extended_save_area_addr = (u32) save_area;
 509         }
 510 #else
 511         if (vdso_alloc_per_cpu(cpu, lowcore))
 512                 goto out;
 513 #endif
 514         lowcore_ptr[cpu] = lowcore;
 515         return 0;
 516
 517 out:
 518         free_page(panic_stack);
 519         free_pages(async_stack, ASYNC_ORDER);
 520         free_pages((unsigned long) lowcore, lc_order);
 521         return -ENOMEM;
 522 }
 523
 524 #ifdef CONFIG_HOTPLUG_CPU
 525 static void smp_free_lowcore(int cpu)
 526 {
 527         struct _lowcore *lowcore;
 528         int lc_order;
 529
 530         lc_order = sizeof(long) == 8 ? 1 : 0;
 531         lowcore = lowcore_ptr[cpu];
 532 #ifndef CONFIG_64BIT
 533         if (MACHINE_HAS_IEEE)
 534                 free_page((unsigned long) lowcore->extended_save_area_addr);
 535 #else
 536         vdso_free_per_cpu(cpu, lowcore);
 537 #endif
 538         free_page(lowcore->panic_stack - PAGE_SIZE);
 539         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 540         free_pages((unsigned long) lowcore, lc_order);
 541         lowcore_ptr[cpu] = NULL;
 542 }
 543 #endif /* CONFIG_HOTPLUG_CPU */
 544
 545 /* Upping and downing of CPUs */
 546 int __cpuinit __cpu_up(unsigned int cpu)
 547 {
 548         struct task_struct *idle;
 549         struct _lowcore *cpu_lowcore;
 550         struct stack_frame *sf;
 551         sigp_ccode ccode;
 552
 553         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 554                 return -EIO;
 555         if (smp_alloc_lowcore(cpu))
 556                 return -ENOMEM;
 557
 558         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 559                                    cpu, sigp_set_prefix);
 560         if (ccode)
 561                 return -EIO;
 562
 563         idle = current_set[cpu];
 564         cpu_lowcore = lowcore_ptr[cpu];
 565         cpu_lowcore->kernel_stack = (unsigned long)
 566                 task_stack_page(idle) + THREAD_SIZE;
 567         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 568         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 569                                      - sizeof(struct pt_regs)
 570                                      - sizeof(struct stack_frame));
 571         memset(sf, 0, sizeof(struct stack_frame));
 572         sf->gprs[9] = (unsigned long) sf;
 573         cpu_lowcore->save_area[15] = (unsigned long) sf;
 574         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 575         asm volatile(
 576                 "       stam    0,15,0(%0)"
 577                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 578         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 579         cpu_lowcore->current_task = (unsigned long) idle;
 580         cpu_lowcore->cpu_data.cpu_nr = cpu;
 581         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 582         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
 583         eieio();
 584
 585         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 586                 udelay(10);
 587
 588         while (!cpu_online(cpu))
 589                 cpu_relax();
 590         return 0;
 591 }
 592
 593 static int __init setup_possible_cpus(char *s)
 594 {
 595         int pcpus, cpu;
 596
 597         pcpus = simple_strtoul(s, NULL, 0);
 598         cpu_possible_map = cpumask_of_cpu(0);
 599         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 600                 cpu_set(cpu, cpu_possible_map);
 601         return 0;
 602 }
 603 early_param("possible_cpus", setup_possible_cpus);
 604
 605 #ifdef CONFIG_HOTPLUG_CPU
 606
 607 int __cpu_disable(void)
 608 {
 609         struct ec_creg_mask_parms cr_parms;
 610         int cpu = smp_processor_id();
 611
 612         cpu_clear(cpu, cpu_online_map);
 613
 614         /* Disable pfault pseudo page faults on this cpu. */
 615         pfault_fini();
 616
 617         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 618         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 619
 620         /* disable all external interrupts */
 621         cr_parms.orvals[0] = 0;
 622         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 623                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 624         /* disable all I/O interrupts */
 625         cr_parms.orvals[6] = 0;
 626         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 627                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 628         /* disable most machine checks */
 629         cr_parms.orvals[14] = 0;
 630         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 631                                  1 << 25 | 1 << 24);
 632
 633         smp_ctl_bit_callback(&cr_parms);
 634
 635         return 0;
 636 }
 637
 638 void __cpu_die(unsigned int cpu)
 639 {
 640         /* Wait until target cpu is down */
 641         while (!smp_cpu_not_running(cpu))
 642                 cpu_relax();
 643         smp_free_lowcore(cpu);
 644         pr_info("Processor %d stopped\n", cpu);
 645 }
 646
 647 void cpu_die(void)
 648 {
 649         idle_task_exit();
 650         signal_processor(smp_processor_id(), sigp_stop);
 651         BUG();
 652         for (;;);
 653 }
 654
 655 #endif /* CONFIG_HOTPLUG_CPU */
 656
 657 void __init smp_prepare_cpus(unsigned int max_cpus)
 658 {
 659 #ifndef CONFIG_64BIT
 660         unsigned long save_area = 0;
 661 #endif
 662         unsigned long async_stack, panic_stack;
 663         struct _lowcore *lowcore;
 664         unsigned int cpu;
 665         int lc_order;
 666
 667         smp_detect_cpus();
 668
 669         /* request the 0x1201 emergency signal external interrupt */
 670         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 671                 panic("Couldn't request external interrupt 0x1201");
 672         print_cpu_info(&S390_lowcore.cpu_data);
 673
 674         /* Reallocate current lowcore, but keep its contents. */
 675         lc_order = sizeof(long) == 8 ? 1 : 0;
 676         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 677         panic_stack = __get_free_page(GFP_KERNEL);
 678         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 679         BUG_ON(!lowcore || !panic_stack || !async_stack);
 680 #ifndef CONFIG_64BIT
 681         if (MACHINE_HAS_IEEE)
 682                 save_area = get_zeroed_page(GFP_KERNEL);
 683 #endif
 684         local_irq_disable();
 685         local_mcck_disable();
 686         lowcore_ptr[smp_processor_id()] = lowcore;
 687         *lowcore = S390_lowcore;
 688         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 689         lowcore->async_stack = async_stack + ASYNC_SIZE;
 690 #ifndef CONFIG_64BIT
 691         if (MACHINE_HAS_IEEE)
 692                 lowcore->extended_save_area_addr = (u32) save_area;
 693 #else
 694         BUG_ON(vdso_alloc_per_cpu(smp_processor_id(), lowcore));
 695 #endif
 696         set_prefix((u32)(unsigned long) lowcore);
 697         local_mcck_enable();
 698         local_irq_enable();
 699         for_each_possible_cpu(cpu)
 700                 if (cpu != smp_processor_id())
 701                         smp_create_idle(cpu);
 702 }
 703
 704 void __init smp_prepare_boot_cpu(void)
 705 {
 706         BUG_ON(smp_processor_id() != 0);
 707
 708         current_thread_info()->cpu = 0;
 709         cpu_set(0, cpu_present_map);
 710         cpu_set(0, cpu_online_map);
 711         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 712         current_set[0] = current;
 713         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 714         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 715 }
 716
 717 void __init smp_cpus_done(unsigned int max_cpus)
 718 {
 719 }
 720
 721 /*
 722  * the frequency of the profiling timer can be changed
 723  * by writing a multiplier value into /proc/profile.
 724  *
 725  * usually you want to run this on all CPUs ;)
 726  */
 727 int setup_profiling_timer(unsigned int multiplier)
 728 {
 729         return 0;
 730 }
 731
 732 #ifdef CONFIG_HOTPLUG_CPU
 733 static ssize_t cpu_configure_show(struct sys_device *dev,
 734                                 struct sysdev_attribute *attr, char *buf)
 735 {
 736         ssize_t count;
 737
 738         mutex_lock(&smp_cpu_state_mutex);
 739         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 740         mutex_unlock(&smp_cpu_state_mutex);
 741         return count;
 742 }
 743
 744 static ssize_t cpu_configure_store(struct sys_device *dev,
 745                                   struct sysdev_attribute *attr,
 746                                   const char *buf, size_t count)
 747 {
 748         int cpu = dev->id;
 749         int val, rc;
 750         char delim;
 751
 752         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 753                 return -EINVAL;
 754         if (val != 0 && val != 1)
 755                 return -EINVAL;
 756
 757         get_online_cpus();
 758         mutex_lock(&smp_cpu_state_mutex);
 759         rc = -EBUSY;
 760         if (cpu_online(cpu))
 761                 goto out;
 762         rc = 0;
 763         switch (val) {
 764         case 0:
 765                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 766                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 767                         if (!rc) {
 768                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 769                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 770                         }
 771                 }
 772                 break;
 773         case 1:
 774                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 775                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 776                         if (!rc) {
 777                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 778                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 779                         }
 780                 }
 781                 break;
 782         default:
 783                 break;
 784         }
 785 out:
 786         mutex_unlock(&smp_cpu_state_mutex);
 787         put_online_cpus();
 788         return rc ? rc : count;
 789 }
 790 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 791 #endif /* CONFIG_HOTPLUG_CPU */
 792
 793 static ssize_t cpu_polarization_show(struct sys_device *dev,
 794                                      struct sysdev_attribute *attr, char *buf)
 795 {
 796         int cpu = dev->id;
 797         ssize_t count;
 798
 799         mutex_lock(&smp_cpu_state_mutex);
 800         switch (smp_cpu_polarization[cpu]) {
 801         case POLARIZATION_HRZ:
 802                 count = sprintf(buf, "horizontal\n");
 803                 break;
 804         case POLARIZATION_VL:
 805                 count = sprintf(buf, "vertical:low\n");
 806                 break;
 807         case POLARIZATION_VM:
 808                 count = sprintf(buf, "vertical:medium\n");
 809                 break;
 810         case POLARIZATION_VH:
 811                 count = sprintf(buf, "vertical:high\n");
 812                 break;
 813         default:
 814                 count = sprintf(buf, "unknown\n");
 815                 break;
 816         }
 817         mutex_unlock(&smp_cpu_state_mutex);
 818         return count;
 819 }
 820 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 821
 822 static ssize_t show_cpu_address(struct sys_device *dev,
 823                                 struct sysdev_attribute *attr, char *buf)
 824 {
 825         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 826 }
 827 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 828
 829
 830 static struct attribute *cpu_common_attrs[] = {
 831 #ifdef CONFIG_HOTPLUG_CPU
 832         &attr_configure.attr,
 833 #endif
 834         &attr_address.attr,
 835         &attr_polarization.attr,
 836         NULL,
 837 };
 838
 839 static struct attribute_group cpu_common_attr_group = {
 840         .attrs = cpu_common_attrs,
 841 };
 842
 843 static ssize_t show_capability(struct sys_device *dev,
 844                                 struct sysdev_attribute *attr, char *buf)
 845 {
 846         unsigned int capability;
 847         int rc;
 848
 849         rc = get_cpu_capability(&capability);
 850         if (rc)
 851                 return rc;
 852         return sprintf(buf, "%u\n", capability);
 853 }
 854 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 855
 856 static ssize_t show_idle_count(struct sys_device *dev,
 857                                 struct sysdev_attribute *attr, char *buf)
 858 {
 859         struct s390_idle_data *idle;
 860         unsigned long long idle_count;
 861
 862         idle = &per_cpu(s390_idle, dev->id);
 863         spin_lock(&idle->lock);
 864         idle_count = idle->idle_count;
 865         if (idle->idle_enter)
 866                 idle_count++;
 867         spin_unlock(&idle->lock);
 868         return sprintf(buf, "%llu\n", idle_count);
 869 }
 870 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 871
 872 static ssize_t show_idle_time(struct sys_device *dev,
 873                                 struct sysdev_attribute *attr, char *buf)
 874 {
 875         struct s390_idle_data *idle;
 876         unsigned long long now, idle_time, idle_enter;
 877
 878         idle = &per_cpu(s390_idle, dev->id);
 879         spin_lock(&idle->lock);
 880         now = get_clock();
 881         idle_time = idle->idle_time;
 882         idle_enter = idle->idle_enter;
 883         if (idle_enter != 0ULL && idle_enter < now)
 884                 idle_time += now - idle_enter;
 885         spin_unlock(&idle->lock);
 886         return sprintf(buf, "%llu\n", idle_time >> 12);
 887 }
 888 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 889
 890 static struct attribute *cpu_online_attrs[] = {
 891         &attr_capability.attr,
 892         &attr_idle_count.attr,
 893         &attr_idle_time_us.attr,
 894         NULL,
 895 };
 896
 897 static struct attribute_group cpu_online_attr_group = {
 898         .attrs = cpu_online_attrs,
 899 };
 900
 901 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 902                                     unsigned long action, void *hcpu)
 903 {
 904         unsigned int cpu = (unsigned int)(long)hcpu;
 905         struct cpu *c = &per_cpu(cpu_devices, cpu);
 906         struct sys_device *s = &c->sysdev;
 907         struct s390_idle_data *idle;
 908
 909         switch (action) {
 910         case CPU_ONLINE:
 911         case CPU_ONLINE_FROZEN:
 912                 idle = &per_cpu(s390_idle, cpu);
 913                 spin_lock_irq(&idle->lock);
 914                 idle->idle_enter = 0;
 915                 idle->idle_time = 0;
 916                 idle->idle_count = 0;
 917                 spin_unlock_irq(&idle->lock);
 918                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 919                         return NOTIFY_BAD;
 920                 break;
 921         case CPU_DEAD:
 922         case CPU_DEAD_FROZEN:
 923                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 924                 break;
 925         }
 926         return NOTIFY_OK;
 927 }
 928
 929 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 930         .notifier_call = smp_cpu_notify,
 931 };
 932
 933 static int __devinit smp_add_present_cpu(int cpu)
 934 {
 935         struct cpu *c = &per_cpu(cpu_devices, cpu);
 936         struct sys_device *s = &c->sysdev;
 937         int rc;
 938
 939         c->hotpluggable = 1;
 940         rc = register_cpu(c, cpu);
 941         if (rc)
 942                 goto out;
 943         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 944         if (rc)
 945                 goto out_cpu;
 946         if (!cpu_online(cpu))
 947                 goto out;
 948         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 949         if (!rc)
 950                 return 0;
 951         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 952 out_cpu:
 953 #ifdef CONFIG_HOTPLUG_CPU
 954         unregister_cpu(c);
 955 #endif
 956 out:
 957         return rc;
 958 }
 959
 960 #ifdef CONFIG_HOTPLUG_CPU
 961
 962 int __ref smp_rescan_cpus(void)
 963 {
 964         cpumask_t newcpus;
 965         int cpu;
 966         int rc;
 967
 968         get_online_cpus();
 969         mutex_lock(&smp_cpu_state_mutex);
 970         newcpus = cpu_present_map;
 971         rc = __smp_rescan_cpus();
 972         if (rc)
 973                 goto out;
 974         cpus_andnot(newcpus, cpu_present_map, newcpus);
 975         for_each_cpu_mask(cpu, newcpus) {
 976                 rc = smp_add_present_cpu(cpu);
 977                 if (rc)
 978                         cpu_clear(cpu, cpu_present_map);
 979         }
 980         rc = 0;
 981 out:
 982         mutex_unlock(&smp_cpu_state_mutex);
 983         put_online_cpus();
 984         if (!cpus_empty(newcpus))
 985                 topology_schedule_update();
 986         return rc;
 987 }
 988
 989 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
 990                                   size_t count)
 991 {
 992         int rc;
 993
 994         rc = smp_rescan_cpus();
 995         return rc ? rc : count;
 996 }
 997 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
 998 #endif /* CONFIG_HOTPLUG_CPU */
 999
1000 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1001 {
1002         ssize_t count;
1003
1004         mutex_lock(&smp_cpu_state_mutex);
1005         count = sprintf(buf, "%d\n", cpu_management);
1006         mutex_unlock(&smp_cpu_state_mutex);
1007         return count;
1008 }
1009
1010 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1011                                  size_t count)
1012 {
1013         int val, rc;
1014         char delim;
1015
1016         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1017                 return -EINVAL;
1018         if (val != 0 && val != 1)
1019                 return -EINVAL;
1020         rc = 0;
1021         get_online_cpus();
1022         mutex_lock(&smp_cpu_state_mutex);
1023         if (cpu_management == val)
1024                 goto out;
1025         rc = topology_set_cpu_management(val);
1026         if (!rc)
1027                 cpu_management = val;
1028 out:
1029         mutex_unlock(&smp_cpu_state_mutex);
1030         put_online_cpus();
1031         return rc ? rc : count;
1032 }
1033 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1034                          dispatching_store);
1035
1036 static int __init topology_init(void)
1037 {
1038         int cpu;
1039         int rc;
1040
1041         register_cpu_notifier(&smp_cpu_nb);
1042
1043 #ifdef CONFIG_HOTPLUG_CPU
1044         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1045         if (rc)
1046                 return rc;
1047 #endif
1048         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1049         if (rc)
1050                 return rc;
1051         for_each_present_cpu(cpu) {
1052                 rc = smp_add_present_cpu(cpu);
1053                 if (rc)
1054                         return rc;
1055         }
1056         return 0;
1057 }
1058 subsys_initcall(topology_init);