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

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