arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999,2007
   5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
   6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
   7  *               Heiko Carstens (heiko.carstens@de.ibm.com)
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * We work with logical cpu numbering everywhere we can. The only
  14  * functions using the real cpu address (got from STAP) are the sigp
  15  * functions. For all other functions we use the identity mapping.
  16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
  17  * used e.g. to find the idle task belonging to a logical cpu. Every array
  18  * in the kernel is sorted by the logical cpu number and not by the physical
  19  * one which is causing all the confusion with __cpu_logical_map and
  20  * cpu_number_map in other architectures.
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/init.h>
  25 #include <linux/mm.h>
  26 #include <linux/err.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/kernel_stat.h>
  29 #include <linux/delay.h>
  30 #include <linux/cache.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/cpu.h>
  33 #include <linux/timex.h>
  34 #include <linux/bootmem.h>
  35 #include <asm/ipl.h>
  36 #include <asm/setup.h>
  37 #include <asm/sigp.h>
  38 #include <asm/pgalloc.h>
  39 #include <asm/irq.h>
  40 #include <asm/s390_ext.h>
  41 #include <asm/cpcmd.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/timer.h>
  44 #include <asm/lowcore.h>
  45 #include <asm/sclp.h>
  46 #include <asm/cpu.h>
  47
  48 /*
  49  * An array with a pointer the lowcore of every CPU.
  50  */
  51 struct _lowcore *lowcore_ptr[NR_CPUS];
  52 EXPORT_SYMBOL(lowcore_ptr);
  53
  54 cpumask_t cpu_online_map = CPU_MASK_NONE;
  55 EXPORT_SYMBOL(cpu_online_map);
  56
  57 cpumask_t cpu_possible_map = CPU_MASK_ALL;
  58 EXPORT_SYMBOL(cpu_possible_map);
  59
  60 static struct task_struct *current_set[NR_CPUS];
  61
  62 static u8 smp_cpu_type;
  63 static int smp_use_sigp_detection;
  64
  65 enum s390_cpu_state {
  66         CPU_STATE_STANDBY,
  67         CPU_STATE_CONFIGURED,
  68 };
  69
  70 #ifdef CONFIG_HOTPLUG_CPU
  71 static DEFINE_MUTEX(smp_cpu_state_mutex);
  72 #endif
  73 static int smp_cpu_state[NR_CPUS];
  74
  75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  76 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
  77
  78 static void smp_ext_bitcall(int, ec_bit_sig);
  79
  80 /*
  81  * Structure and data for __smp_call_function_map(). This is designed to
  82  * minimise static memory requirements. It also looks cleaner.
  83  */
  84 static DEFINE_SPINLOCK(call_lock);
  85
  86 struct call_data_struct {
  87         void (*func) (void *info);
  88         void *info;
  89         cpumask_t started;
  90         cpumask_t finished;
  91         int wait;
  92 };
  93
  94 static struct call_data_struct *call_data;
  95
  96 /*
  97  * 'Call function' interrupt callback
  98  */
  99 static void do_call_function(void)
 100 {
 101         void (*func) (void *info) = call_data->func;
 102         void *info = call_data->info;
 103         int wait = call_data->wait;
 104
 105         cpu_set(smp_processor_id(), call_data->started);
 106         (*func)(info);
 107         if (wait)
 108                 cpu_set(smp_processor_id(), call_data->finished);;
 109 }
 110
 111 static void __smp_call_function_map(void (*func) (void *info), void *info,
 112                                     int nonatomic, int wait, cpumask_t map)
 113 {
 114         struct call_data_struct data;
 115         int cpu, local = 0;
 116
 117         /*
 118          * Can deadlock when interrupts are disabled or if in wrong context.
 119          */
 120         WARN_ON(irqs_disabled() || in_irq());
 121
 122         /*
 123          * Check for local function call. We have to have the same call order
 124          * as in on_each_cpu() because of machine_restart_smp().
 125          */
 126         if (cpu_isset(smp_processor_id(), map)) {
 127                 local = 1;
 128                 cpu_clear(smp_processor_id(), map);
 129         }
 130
 131         cpus_and(map, map, cpu_online_map);
 132         if (cpus_empty(map))
 133                 goto out;
 134
 135         data.func = func;
 136         data.info = info;
 137         data.started = CPU_MASK_NONE;
 138         data.wait = wait;
 139         if (wait)
 140                 data.finished = CPU_MASK_NONE;
 141
 142         spin_lock(&call_lock);
 143         call_data = &data;
 144
 145         for_each_cpu_mask(cpu, map)
 146                 smp_ext_bitcall(cpu, ec_call_function);
 147
 148         /* Wait for response */
 149         while (!cpus_equal(map, data.started))
 150                 cpu_relax();
 151         if (wait)
 152                 while (!cpus_equal(map, data.finished))
 153                         cpu_relax();
 154         spin_unlock(&call_lock);
 155 out:
 156         if (local) {
 157                 local_irq_disable();
 158                 func(info);
 159                 local_irq_enable();
 160         }
 161 }
 162
 163 /*
 164  * smp_call_function:
 165  * @func: the function to run; this must be fast and non-blocking
 166  * @info: an arbitrary pointer to pass to the function
 167  * @nonatomic: unused
 168  * @wait: if true, wait (atomically) until function has completed on other CPUs
 169  *
 170  * Run a function on all other CPUs.
 171  *
 172  * You must not call this function with disabled interrupts, from a
 173  * hardware interrupt handler or from a bottom half.
 174  */
 175 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
 176                       int wait)
 177 {
 178         cpumask_t map;
 179
 180         preempt_disable();
 181         map = cpu_online_map;
 182         cpu_clear(smp_processor_id(), map);
 183         __smp_call_function_map(func, info, nonatomic, wait, map);
 184         preempt_enable();
 185         return 0;
 186 }
 187 EXPORT_SYMBOL(smp_call_function);
 188
 189 /*
 190  * smp_call_function_single:
 191  * @cpu: the CPU where func should run
 192  * @func: the function to run; this must be fast and non-blocking
 193  * @info: an arbitrary pointer to pass to the function
 194  * @nonatomic: unused
 195  * @wait: if true, wait (atomically) until function has completed on other CPUs
 196  *
 197  * Run a function on one processor.
 198  *
 199  * You must not call this function with disabled interrupts, from a
 200  * hardware interrupt handler or from a bottom half.
 201  */
 202 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 203                              int nonatomic, int wait)
 204 {
 205         preempt_disable();
 206         __smp_call_function_map(func, info, nonatomic, wait,
 207                                 cpumask_of_cpu(cpu));
 208         preempt_enable();
 209         return 0;
 210 }
 211 EXPORT_SYMBOL(smp_call_function_single);
 212
 213 /**
 214  * smp_call_function_mask(): Run a function on a set of other CPUs.
 215  * @mask: The set of cpus to run on.  Must not include the current cpu.
 216  * @func: The function to run. This must be fast and non-blocking.
 217  * @info: An arbitrary pointer to pass to the function.
 218  * @wait: If true, wait (atomically) until function has completed on other CPUs.
 219  *
 220  * Returns 0 on success, else a negative status code.
 221  *
 222  * If @wait is true, then returns once @func has returned; otherwise
 223  * it returns just before the target cpu calls @func.
 224  *
 225  * You must not call this function with disabled interrupts or from a
 226  * hardware interrupt handler or from a bottom half handler.
 227  */
 228 int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
 229                            int wait)
 230 {
 231         preempt_disable();
 232         cpu_clear(smp_processor_id(), mask);
 233         __smp_call_function_map(func, info, 0, wait, mask);
 234         preempt_enable();
 235         return 0;
 236 }
 237 EXPORT_SYMBOL(smp_call_function_mask);
 238
 239 void smp_send_stop(void)
 240 {
 241         int cpu, rc;
 242
 243         /* Disable all interrupts/machine checks */
 244         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 245
 246         /* write magic number to zero page (absolute 0) */
 247         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
 248
 249         /* stop all processors */
 250         for_each_online_cpu(cpu) {
 251                 if (cpu == smp_processor_id())
 252                         continue;
 253                 do {
 254                         rc = signal_processor(cpu, sigp_stop);
 255                 } while (rc == sigp_busy);
 256
 257                 while (!smp_cpu_not_running(cpu))
 258                         cpu_relax();
 259         }
 260 }
 261
 262 /*
 263  * This is the main routine where commands issued by other
 264  * cpus are handled.
 265  */
 266
 267 static void do_ext_call_interrupt(__u16 code)
 268 {
 269         unsigned long bits;
 270
 271         /*
 272          * handle bit signal external calls
 273          *
 274          * For the ec_schedule signal we have to do nothing. All the work
 275          * is done automatically when we return from the interrupt.
 276          */
 277         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 278
 279         if (test_bit(ec_call_function, &bits))
 280                 do_call_function();
 281 }
 282
 283 /*
 284  * Send an external call sigp to another cpu and return without waiting
 285  * for its completion.
 286  */
 287 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 288 {
 289         /*
 290          * Set signaling bit in lowcore of target cpu and kick it
 291          */
 292         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 293         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 294                 udelay(10);
 295 }
 296
 297 #ifndef CONFIG_64BIT
 298 /*
 299  * this function sends a 'purge tlb' signal to another CPU.
 300  */
 301 void smp_ptlb_callback(void *info)
 302 {
 303         __tlb_flush_local();
 304 }
 305
 306 void smp_ptlb_all(void)
 307 {
 308         on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
 309 }
 310 EXPORT_SYMBOL(smp_ptlb_all);
 311 #endif /* ! CONFIG_64BIT */
 312
 313 /*
 314  * this function sends a 'reschedule' IPI to another CPU.
 315  * it goes straight through and wastes no time serializing
 316  * anything. Worst case is that we lose a reschedule ...
 317  */
 318 void smp_send_reschedule(int cpu)
 319 {
 320         smp_ext_bitcall(cpu, ec_schedule);
 321 }
 322
 323 /*
 324  * parameter area for the set/clear control bit callbacks
 325  */
 326 struct ec_creg_mask_parms {
 327         unsigned long orvals[16];
 328         unsigned long andvals[16];
 329 };
 330
 331 /*
 332  * callback for setting/clearing control bits
 333  */
 334 static void smp_ctl_bit_callback(void *info)
 335 {
 336         struct ec_creg_mask_parms *pp = info;
 337         unsigned long cregs[16];
 338         int i;
 339
 340         __ctl_store(cregs, 0, 15);
 341         for (i = 0; i <= 15; i++)
 342                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 343         __ctl_load(cregs, 0, 15);
 344 }
 345
 346 /*
 347  * Set a bit in a control register of all cpus
 348  */
 349 void smp_ctl_set_bit(int cr, int bit)
 350 {
 351         struct ec_creg_mask_parms parms;
 352
 353         memset(&parms.orvals, 0, sizeof(parms.orvals));
 354         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 355         parms.orvals[cr] = 1 << bit;
 356         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 357 }
 358 EXPORT_SYMBOL(smp_ctl_set_bit);
 359
 360 /*
 361  * Clear a bit in a control register of all cpus
 362  */
 363 void smp_ctl_clear_bit(int cr, int bit)
 364 {
 365         struct ec_creg_mask_parms parms;
 366
 367         memset(&parms.orvals, 0, sizeof(parms.orvals));
 368         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 369         parms.andvals[cr] = ~(1L << bit);
 370         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 371 }
 372 EXPORT_SYMBOL(smp_ctl_clear_bit);
 373
 374 /*
 375  * In early ipl state a temp. logically cpu number is needed, so the sigp
 376  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 377  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 378  */
 379 #define CPU_INIT_NO     1
 380
 381 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 382
 383 /*
 384  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 385  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 386  * save its prefix registers, since they get lost, when switching from 31 bit
 387  * to 64 bit.
 388  */
 389 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 390         __attribute__((__section__(".data")));
 391
 392 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 393 {
 394         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 395                 return;
 396         if (cpu >= NR_CPUS) {
 397                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
 398                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
 399                 return;
 400         }
 401         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 402         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 403         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 404                sigp_busy)
 405                 cpu_relax();
 406         memcpy(zfcpdump_save_areas[cpu],
 407                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 408                SAVE_AREA_SIZE);
 409 #ifdef CONFIG_64BIT
 410         /* copy original prefix register */
 411         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 412 #endif
 413 }
 414
 415 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 416 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 417
 418 #else
 419
 420 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 421
 422 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 423
 424 static int cpu_stopped(int cpu)
 425 {
 426         __u32 status;
 427
 428         /* Check for stopped state */
 429         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 430             sigp_status_stored) {
 431                 if (status & 0x40)
 432                         return 1;
 433         }
 434         return 0;
 435 }
 436
 437 static int cpu_known(int cpu_id)
 438 {
 439         int cpu;
 440
 441         for_each_present_cpu(cpu) {
 442                 if (__cpu_logical_map[cpu] == cpu_id)
 443                         return 1;
 444         }
 445         return 0;
 446 }
 447
 448 static int smp_rescan_cpus_sigp(cpumask_t avail)
 449 {
 450         int cpu_id, logical_cpu;
 451
 452         logical_cpu = first_cpu(avail);
 453         if (logical_cpu == NR_CPUS)
 454                 return 0;
 455         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 456                 if (cpu_known(cpu_id))
 457                         continue;
 458                 __cpu_logical_map[logical_cpu] = cpu_id;
 459                 if (!cpu_stopped(logical_cpu))
 460                         continue;
 461                 cpu_set(logical_cpu, cpu_present_map);
 462                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 463                 logical_cpu = next_cpu(logical_cpu, avail);
 464                 if (logical_cpu == NR_CPUS)
 465                         break;
 466         }
 467         return 0;
 468 }
 469
 470 static int smp_rescan_cpus_sclp(cpumask_t avail)
 471 {
 472         struct sclp_cpu_info *info;
 473         int cpu_id, logical_cpu, cpu;
 474         int rc;
 475
 476         logical_cpu = first_cpu(avail);
 477         if (logical_cpu == NR_CPUS)
 478                 return 0;
 479         info = kmalloc(sizeof(*info), GFP_KERNEL);
 480         if (!info)
 481                 return -ENOMEM;
 482         rc = sclp_get_cpu_info(info);
 483         if (rc)
 484                 goto out;
 485         for (cpu = 0; cpu < info->combined; cpu++) {
 486                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 487                         continue;
 488                 cpu_id = info->cpu[cpu].address;
 489                 if (cpu_known(cpu_id))
 490                         continue;
 491                 __cpu_logical_map[logical_cpu] = cpu_id;
 492                 cpu_set(logical_cpu, cpu_present_map);
 493                 if (cpu >= info->configured)
 494                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 495                 else
 496                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 497                 logical_cpu = next_cpu(logical_cpu, avail);
 498                 if (logical_cpu == NR_CPUS)
 499                         break;
 500         }
 501 out:
 502         kfree(info);
 503         return rc;
 504 }
 505
 506 static int smp_rescan_cpus(void)
 507 {
 508         cpumask_t avail;
 509
 510         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 511         if (smp_use_sigp_detection)
 512                 return smp_rescan_cpus_sigp(avail);
 513         else
 514                 return smp_rescan_cpus_sclp(avail);
 515 }
 516
 517 static void __init smp_detect_cpus(void)
 518 {
 519         unsigned int cpu, c_cpus, s_cpus;
 520         struct sclp_cpu_info *info;
 521         u16 boot_cpu_addr, cpu_addr;
 522
 523         c_cpus = 1;
 524         s_cpus = 0;
 525         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 526         info = kmalloc(sizeof(*info), GFP_KERNEL);
 527         if (!info)
 528                 panic("smp_detect_cpus failed to allocate memory\n");
 529         /* Use sigp detection algorithm if sclp doesn't work. */
 530         if (sclp_get_cpu_info(info)) {
 531                 smp_use_sigp_detection = 1;
 532                 for (cpu = 0; cpu <= 65535; cpu++) {
 533                         if (cpu == boot_cpu_addr)
 534                                 continue;
 535                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 536                         if (!cpu_stopped(CPU_INIT_NO))
 537                                 continue;
 538                         smp_get_save_area(c_cpus, cpu);
 539                         c_cpus++;
 540                 }
 541                 goto out;
 542         }
 543
 544         if (info->has_cpu_type) {
 545                 for (cpu = 0; cpu < info->combined; cpu++) {
 546                         if (info->cpu[cpu].address == boot_cpu_addr) {
 547                                 smp_cpu_type = info->cpu[cpu].type;
 548                                 break;
 549                         }
 550                 }
 551         }
 552
 553         for (cpu = 0; cpu < info->combined; cpu++) {
 554                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 555                         continue;
 556                 cpu_addr = info->cpu[cpu].address;
 557                 if (cpu_addr == boot_cpu_addr)
 558                         continue;
 559                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 560                 if (!cpu_stopped(CPU_INIT_NO)) {
 561                         s_cpus++;
 562                         continue;
 563                 }
 564                 smp_get_save_area(c_cpus, cpu_addr);
 565                 c_cpus++;
 566         }
 567 out:
 568         kfree(info);
 569         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
 570         get_online_cpus();
 571         smp_rescan_cpus();
 572         put_online_cpus();
 573 }
 574
 575 /*
 576  *      Activate a secondary processor.
 577  */
 578 int __cpuinit start_secondary(void *cpuvoid)
 579 {
 580         /* Setup the cpu */
 581         cpu_init();
 582         preempt_disable();
 583         /* Enable TOD clock interrupts on the secondary cpu. */
 584         init_cpu_timer();
 585 #ifdef CONFIG_VIRT_TIMER
 586         /* Enable cpu timer interrupts on the secondary cpu. */
 587         init_cpu_vtimer();
 588 #endif
 589         /* Enable pfault pseudo page faults on this cpu. */
 590         pfault_init();
 591
 592         /* Mark this cpu as online */
 593         cpu_set(smp_processor_id(), cpu_online_map);
 594         /* Switch on interrupts */
 595         local_irq_enable();
 596         /* Print info about this processor */
 597         print_cpu_info(&S390_lowcore.cpu_data);
 598         /* cpu_idle will call schedule for us */
 599         cpu_idle();
 600         return 0;
 601 }
 602
 603 static void __init smp_create_idle(unsigned int cpu)
 604 {
 605         struct task_struct *p;
 606
 607         /*
 608          *  don't care about the psw and regs settings since we'll never
 609          *  reschedule the forked task.
 610          */
 611         p = fork_idle(cpu);
 612         if (IS_ERR(p))
 613                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 614         current_set[cpu] = p;
 615         spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
 616 }
 617
 618 static int __cpuinit smp_alloc_lowcore(int cpu)
 619 {
 620         unsigned long async_stack, panic_stack;
 621         struct _lowcore *lowcore;
 622         int lc_order;
 623
 624         lc_order = sizeof(long) == 8 ? 1 : 0;
 625         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 626         if (!lowcore)
 627                 return -ENOMEM;
 628         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 629         panic_stack = __get_free_page(GFP_KERNEL);
 630         if (!panic_stack || !async_stack)
 631                 goto out;
 632         memcpy(lowcore, &S390_lowcore, 512);
 633         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 634         lowcore->async_stack = async_stack + ASYNC_SIZE;
 635         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 636
 637 #ifndef CONFIG_64BIT
 638         if (MACHINE_HAS_IEEE) {
 639                 unsigned long save_area;
 640
 641                 save_area = get_zeroed_page(GFP_KERNEL);
 642                 if (!save_area)
 643                         goto out_save_area;
 644                 lowcore->extended_save_area_addr = (u32) save_area;
 645         }
 646 #endif
 647         lowcore_ptr[cpu] = lowcore;
 648         return 0;
 649
 650 #ifndef CONFIG_64BIT
 651 out_save_area:
 652         free_page(panic_stack);
 653 #endif
 654 out:
 655         free_pages(async_stack, ASYNC_ORDER);
 656         free_pages((unsigned long) lowcore, lc_order);
 657         return -ENOMEM;
 658 }
 659
 660 #ifdef CONFIG_HOTPLUG_CPU
 661 static void smp_free_lowcore(int cpu)
 662 {
 663         struct _lowcore *lowcore;
 664         int lc_order;
 665
 666         lc_order = sizeof(long) == 8 ? 1 : 0;
 667         lowcore = lowcore_ptr[cpu];
 668 #ifndef CONFIG_64BIT
 669         if (MACHINE_HAS_IEEE)
 670                 free_page((unsigned long) lowcore->extended_save_area_addr);
 671 #endif
 672         free_page(lowcore->panic_stack - PAGE_SIZE);
 673         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 674         free_pages((unsigned long) lowcore, lc_order);
 675         lowcore_ptr[cpu] = NULL;
 676 }
 677 #endif /* CONFIG_HOTPLUG_CPU */
 678
 679 /* Upping and downing of CPUs */
 680 int __cpuinit __cpu_up(unsigned int cpu)
 681 {
 682         struct task_struct *idle;
 683         struct _lowcore *cpu_lowcore;
 684         struct stack_frame *sf;
 685         sigp_ccode ccode;
 686
 687         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 688                 return -EIO;
 689         if (smp_alloc_lowcore(cpu))
 690                 return -ENOMEM;
 691
 692         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 693                                    cpu, sigp_set_prefix);
 694         if (ccode) {
 695                 printk("sigp_set_prefix failed for cpu %d "
 696                        "with condition code %d\n",
 697                        (int) cpu, (int) ccode);
 698                 return -EIO;
 699         }
 700
 701         idle = current_set[cpu];
 702         cpu_lowcore = lowcore_ptr[cpu];
 703         cpu_lowcore->kernel_stack = (unsigned long)
 704                 task_stack_page(idle) + THREAD_SIZE;
 705         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 706         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 707                                      - sizeof(struct pt_regs)
 708                                      - sizeof(struct stack_frame));
 709         memset(sf, 0, sizeof(struct stack_frame));
 710         sf->gprs[9] = (unsigned long) sf;
 711         cpu_lowcore->save_area[15] = (unsigned long) sf;
 712         __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
 713         asm volatile(
 714                 "       stam    0,15,0(%0)"
 715                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 716         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 717         cpu_lowcore->current_task = (unsigned long) idle;
 718         cpu_lowcore->cpu_data.cpu_nr = cpu;
 719         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 720         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
 721         eieio();
 722
 723         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 724                 udelay(10);
 725
 726         while (!cpu_online(cpu))
 727                 cpu_relax();
 728         return 0;
 729 }
 730
 731 static int __init setup_possible_cpus(char *s)
 732 {
 733         int pcpus, cpu;
 734
 735         pcpus = simple_strtoul(s, NULL, 0);
 736         cpu_possible_map = cpumask_of_cpu(0);
 737         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 738                 cpu_set(cpu, cpu_possible_map);
 739         return 0;
 740 }
 741 early_param("possible_cpus", setup_possible_cpus);
 742
 743 #ifdef CONFIG_HOTPLUG_CPU
 744
 745 int __cpu_disable(void)
 746 {
 747         struct ec_creg_mask_parms cr_parms;
 748         int cpu = smp_processor_id();
 749
 750         cpu_clear(cpu, cpu_online_map);
 751
 752         /* Disable pfault pseudo page faults on this cpu. */
 753         pfault_fini();
 754
 755         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 756         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 757
 758         /* disable all external interrupts */
 759         cr_parms.orvals[0] = 0;
 760         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 761                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 762         /* disable all I/O interrupts */
 763         cr_parms.orvals[6] = 0;
 764         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 765                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 766         /* disable most machine checks */
 767         cr_parms.orvals[14] = 0;
 768         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 769                                  1 << 25 | 1 << 24);
 770
 771         smp_ctl_bit_callback(&cr_parms);
 772
 773         return 0;
 774 }
 775
 776 void __cpu_die(unsigned int cpu)
 777 {
 778         /* Wait until target cpu is down */
 779         while (!smp_cpu_not_running(cpu))
 780                 cpu_relax();
 781         smp_free_lowcore(cpu);
 782         printk(KERN_INFO "Processor %d spun down\n", cpu);
 783 }
 784
 785 void cpu_die(void)
 786 {
 787         idle_task_exit();
 788         signal_processor(smp_processor_id(), sigp_stop);
 789         BUG();
 790         for (;;);
 791 }
 792
 793 #endif /* CONFIG_HOTPLUG_CPU */
 794
 795 void __init smp_prepare_cpus(unsigned int max_cpus)
 796 {
 797 #ifndef CONFIG_64BIT
 798         unsigned long save_area = 0;
 799 #endif
 800         unsigned long async_stack, panic_stack;
 801         struct _lowcore *lowcore;
 802         unsigned int cpu;
 803         int lc_order;
 804
 805         smp_detect_cpus();
 806
 807         /* request the 0x1201 emergency signal external interrupt */
 808         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 809                 panic("Couldn't request external interrupt 0x1201");
 810         print_cpu_info(&S390_lowcore.cpu_data);
 811
 812         /* Reallocate current lowcore, but keep its contents. */
 813         lc_order = sizeof(long) == 8 ? 1 : 0;
 814         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 815         panic_stack = __get_free_page(GFP_KERNEL);
 816         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 817 #ifndef CONFIG_64BIT
 818         if (MACHINE_HAS_IEEE)
 819                 save_area = get_zeroed_page(GFP_KERNEL);
 820 #endif
 821         local_irq_disable();
 822         local_mcck_disable();
 823         lowcore_ptr[smp_processor_id()] = lowcore;
 824         *lowcore = S390_lowcore;
 825         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 826         lowcore->async_stack = async_stack + ASYNC_SIZE;
 827 #ifndef CONFIG_64BIT
 828         if (MACHINE_HAS_IEEE)
 829                 lowcore->extended_save_area_addr = (u32) save_area;
 830 #endif
 831         set_prefix((u32)(unsigned long) lowcore);
 832         local_mcck_enable();
 833         local_irq_enable();
 834         for_each_possible_cpu(cpu)
 835                 if (cpu != smp_processor_id())
 836                         smp_create_idle(cpu);
 837 }
 838
 839 void __init smp_prepare_boot_cpu(void)
 840 {
 841         BUG_ON(smp_processor_id() != 0);
 842
 843         current_thread_info()->cpu = 0;
 844         cpu_set(0, cpu_present_map);
 845         cpu_set(0, cpu_online_map);
 846         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 847         current_set[0] = current;
 848         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 849         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 850 }
 851
 852 void __init smp_cpus_done(unsigned int max_cpus)
 853 {
 854 }
 855
 856 /*
 857  * the frequency of the profiling timer can be changed
 858  * by writing a multiplier value into /proc/profile.
 859  *
 860  * usually you want to run this on all CPUs ;)
 861  */
 862 int setup_profiling_timer(unsigned int multiplier)
 863 {
 864         return 0;
 865 }
 866
 867 #ifdef CONFIG_HOTPLUG_CPU
 868 static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
 869 {
 870         ssize_t count;
 871
 872         mutex_lock(&smp_cpu_state_mutex);
 873         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 874         mutex_unlock(&smp_cpu_state_mutex);
 875         return count;
 876 }
 877
 878 static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
 879                                    size_t count)
 880 {
 881         int cpu = dev->id;
 882         int val, rc;
 883         char delim;
 884
 885         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 886                 return -EINVAL;
 887         if (val != 0 && val != 1)
 888                 return -EINVAL;
 889
 890         mutex_lock(&smp_cpu_state_mutex);
 891         get_online_cpus();
 892         rc = -EBUSY;
 893         if (cpu_online(cpu))
 894                 goto out;
 895         rc = 0;
 896         switch (val) {
 897         case 0:
 898                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 899                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 900                         if (!rc)
 901                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 902                 }
 903                 break;
 904         case 1:
 905                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 906                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 907                         if (!rc)
 908                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 909                 }
 910                 break;
 911         default:
 912                 break;
 913         }
 914 out:
 915         put_online_cpus();
 916         mutex_unlock(&smp_cpu_state_mutex);
 917         return rc ? rc : count;
 918 }
 919 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 920 #endif /* CONFIG_HOTPLUG_CPU */
 921
 922 static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
 923 {
 924         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 925 }
 926 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 927
 928
 929 static struct attribute *cpu_common_attrs[] = {
 930 #ifdef CONFIG_HOTPLUG_CPU
 931         &attr_configure.attr,
 932 #endif
 933         &attr_address.attr,
 934         NULL,
 935 };
 936
 937 static struct attribute_group cpu_common_attr_group = {
 938         .attrs = cpu_common_attrs,
 939 };
 940
 941 static ssize_t show_capability(struct sys_device *dev, char *buf)
 942 {
 943         unsigned int capability;
 944         int rc;
 945
 946         rc = get_cpu_capability(&capability);
 947         if (rc)
 948                 return rc;
 949         return sprintf(buf, "%u\n", capability);
 950 }
 951 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 952
 953 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
 954 {
 955         struct s390_idle_data *idle;
 956         unsigned long long idle_count;
 957
 958         idle = &per_cpu(s390_idle, dev->id);
 959         spin_lock_irq(&idle->lock);
 960         idle_count = idle->idle_count;
 961         spin_unlock_irq(&idle->lock);
 962         return sprintf(buf, "%llu\n", idle_count);
 963 }
 964 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 965
 966 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
 967 {
 968         struct s390_idle_data *idle;
 969         unsigned long long new_time;
 970
 971         idle = &per_cpu(s390_idle, dev->id);
 972         spin_lock_irq(&idle->lock);
 973         if (idle->in_idle) {
 974                 new_time = get_clock();
 975                 idle->idle_time += new_time - idle->idle_enter;
 976                 idle->idle_enter = new_time;
 977         }
 978         new_time = idle->idle_time;
 979         spin_unlock_irq(&idle->lock);
 980         return sprintf(buf, "%llu\n", new_time >> 12);
 981 }
 982 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 983
 984 static struct attribute *cpu_online_attrs[] = {
 985         &attr_capability.attr,
 986         &attr_idle_count.attr,
 987         &attr_idle_time_us.attr,
 988         NULL,
 989 };
 990
 991 static struct attribute_group cpu_online_attr_group = {
 992         .attrs = cpu_online_attrs,
 993 };
 994
 995 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 996                                     unsigned long action, void *hcpu)
 997 {
 998         unsigned int cpu = (unsigned int)(long)hcpu;
 999         struct cpu *c = &per_cpu(cpu_devices, cpu);
1000         struct sys_device *s = &c->sysdev;
1001         struct s390_idle_data *idle;
1002
1003         switch (action) {
1004         case CPU_ONLINE:
1005         case CPU_ONLINE_FROZEN:
1006                 idle = &per_cpu(s390_idle, cpu);
1007                 spin_lock_irq(&idle->lock);
1008                 idle->idle_enter = 0;
1009                 idle->idle_time = 0;
1010                 idle->idle_count = 0;
1011                 spin_unlock_irq(&idle->lock);
1012                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
1013                         return NOTIFY_BAD;
1014                 break;
1015         case CPU_DEAD:
1016         case CPU_DEAD_FROZEN:
1017                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1018                 break;
1019         }
1020         return NOTIFY_OK;
1021 }
1022
1023 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1024         .notifier_call = smp_cpu_notify,
1025 };
1026
1027 static int __devinit smp_add_present_cpu(int cpu)
1028 {
1029         struct cpu *c = &per_cpu(cpu_devices, cpu);
1030         struct sys_device *s = &c->sysdev;
1031         int rc;
1032
1033         c->hotpluggable = 1;
1034         rc = register_cpu(c, cpu);
1035         if (rc)
1036                 goto out;
1037         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1038         if (rc)
1039                 goto out_cpu;
1040         if (!cpu_online(cpu))
1041                 goto out;
1042         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1043         if (!rc)
1044                 return 0;
1045         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1046 out_cpu:
1047 #ifdef CONFIG_HOTPLUG_CPU
1048         unregister_cpu(c);
1049 #endif
1050 out:
1051         return rc;
1052 }
1053
1054 #ifdef CONFIG_HOTPLUG_CPU
1055 static ssize_t __ref rescan_store(struct sys_device *dev,
1056                                   const char *buf, size_t count)
1057 {
1058         cpumask_t newcpus;
1059         int cpu;
1060         int rc;
1061
1062         mutex_lock(&smp_cpu_state_mutex);
1063         get_online_cpus();
1064         newcpus = cpu_present_map;
1065         rc = smp_rescan_cpus();
1066         if (rc)
1067                 goto out;
1068         cpus_andnot(newcpus, cpu_present_map, newcpus);
1069         for_each_cpu_mask(cpu, newcpus) {
1070                 rc = smp_add_present_cpu(cpu);
1071                 if (rc)
1072                         cpu_clear(cpu, cpu_present_map);
1073         }
1074         rc = 0;
1075 out:
1076         put_online_cpus();
1077         mutex_unlock(&smp_cpu_state_mutex);
1078         return rc ? rc : count;
1079 }
1080 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1081 #endif /* CONFIG_HOTPLUG_CPU */
1082
1083 static int __init topology_init(void)
1084 {
1085         int cpu;
1086         int rc;
1087
1088         register_cpu_notifier(&smp_cpu_nb);
1089
1090 #ifdef CONFIG_HOTPLUG_CPU
1091         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1092                                &attr_rescan.attr);
1093         if (rc)
1094                 return rc;
1095 #endif
1096         for_each_present_cpu(cpu) {
1097                 rc = smp_add_present_cpu(cpu);
1098                 if (rc)
1099                         return rc;
1100         }
1101         return 0;
1102 }
1103 subsys_initcall(topology_init);