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 #include "entry.h"
  48
  49 /*
  50  * An array with a pointer the lowcore of every CPU.
  51  */
  52 struct _lowcore *lowcore_ptr[NR_CPUS];
  53 EXPORT_SYMBOL(lowcore_ptr);
  54
  55 cpumask_t cpu_online_map = CPU_MASK_NONE;
  56 EXPORT_SYMBOL(cpu_online_map);
  57
  58 cpumask_t cpu_possible_map = CPU_MASK_ALL;
  59 EXPORT_SYMBOL(cpu_possible_map);
  60
  61 static struct task_struct *current_set[NR_CPUS];
  62
  63 static u8 smp_cpu_type;
  64 static int smp_use_sigp_detection;
  65
  66 enum s390_cpu_state {
  67         CPU_STATE_STANDBY,
  68         CPU_STATE_CONFIGURED,
  69 };
  70
  71 DEFINE_MUTEX(smp_cpu_state_mutex);
  72 int smp_cpu_polarization[NR_CPUS];
  73 static int smp_cpu_state[NR_CPUS];
  74 static int cpu_management;
  75
  76 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  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 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         call_data = &data;
 143
 144         for_each_cpu_mask(cpu, map)
 145                 smp_ext_bitcall(cpu, ec_call_function);
 146
 147         /* Wait for response */
 148         while (!cpus_equal(map, data.started))
 149                 cpu_relax();
 150         if (wait)
 151                 while (!cpus_equal(map, data.finished))
 152                         cpu_relax();
 153 out:
 154         if (local) {
 155                 local_irq_disable();
 156                 func(info);
 157                 local_irq_enable();
 158         }
 159 }
 160
 161 /*
 162  * smp_call_function:
 163  * @func: the function to run; this must be fast and non-blocking
 164  * @info: an arbitrary pointer to pass to the function
 165  * @wait: if true, wait (atomically) until function has completed on other CPUs
 166  *
 167  * Run a function on all other CPUs.
 168  *
 169  * You must not call this function with disabled interrupts, from a
 170  * hardware interrupt handler or from a bottom half.
 171  */
 172 int smp_call_function(void (*func) (void *info), void *info, int wait)
 173 {
 174         cpumask_t map;
 175
 176         spin_lock(&call_lock);
 177         map = cpu_online_map;
 178         cpu_clear(smp_processor_id(), map);
 179         __smp_call_function_map(func, info, wait, map);
 180         spin_unlock(&call_lock);
 181         return 0;
 182 }
 183 EXPORT_SYMBOL(smp_call_function);
 184
 185 /*
 186  * smp_call_function_single:
 187  * @cpu: the CPU where func should run
 188  * @func: the function to run; this must be fast and non-blocking
 189  * @info: an arbitrary pointer to pass to the function
 190  * @wait: if true, wait (atomically) until function has completed on other CPUs
 191  *
 192  * Run a function on one processor.
 193  *
 194  * You must not call this function with disabled interrupts, from a
 195  * hardware interrupt handler or from a bottom half.
 196  */
 197 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 198                              int wait)
 199 {
 200         spin_lock(&call_lock);
 201         __smp_call_function_map(func, info, wait, cpumask_of_cpu(cpu));
 202         spin_unlock(&call_lock);
 203         return 0;
 204 }
 205 EXPORT_SYMBOL(smp_call_function_single);
 206
 207 /**
 208  * smp_call_function_mask(): Run a function on a set of other CPUs.
 209  * @mask: The set of cpus to run on.  Must not include the current cpu.
 210  * @func: The function to run. This must be fast and non-blocking.
 211  * @info: An arbitrary pointer to pass to the function.
 212  * @wait: If true, wait (atomically) until function has completed on other CPUs.
 213  *
 214  * Returns 0 on success, else a negative status code.
 215  *
 216  * If @wait is true, then returns once @func has returned; otherwise
 217  * it returns just before the target cpu calls @func.
 218  *
 219  * You must not call this function with disabled interrupts or from a
 220  * hardware interrupt handler or from a bottom half handler.
 221  */
 222 int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
 223                            int wait)
 224 {
 225         spin_lock(&call_lock);
 226         cpu_clear(smp_processor_id(), mask);
 227         __smp_call_function_map(func, info, wait, mask);
 228         spin_unlock(&call_lock);
 229         return 0;
 230 }
 231 EXPORT_SYMBOL(smp_call_function_mask);
 232
 233 void smp_send_stop(void)
 234 {
 235         int cpu, rc;
 236
 237         /* Disable all interrupts/machine checks */
 238         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 239
 240         /* write magic number to zero page (absolute 0) */
 241         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
 242
 243         /* stop all processors */
 244         for_each_online_cpu(cpu) {
 245                 if (cpu == smp_processor_id())
 246                         continue;
 247                 do {
 248                         rc = signal_processor(cpu, sigp_stop);
 249                 } while (rc == sigp_busy);
 250
 251                 while (!smp_cpu_not_running(cpu))
 252                         cpu_relax();
 253         }
 254 }
 255
 256 /*
 257  * This is the main routine where commands issued by other
 258  * cpus are handled.
 259  */
 260
 261 static void do_ext_call_interrupt(__u16 code)
 262 {
 263         unsigned long bits;
 264
 265         /*
 266          * handle bit signal external calls
 267          *
 268          * For the ec_schedule signal we have to do nothing. All the work
 269          * is done automatically when we return from the interrupt.
 270          */
 271         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 272
 273         if (test_bit(ec_call_function, &bits))
 274                 do_call_function();
 275 }
 276
 277 /*
 278  * Send an external call sigp to another cpu and return without waiting
 279  * for its completion.
 280  */
 281 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 282 {
 283         /*
 284          * Set signaling bit in lowcore of target cpu and kick it
 285          */
 286         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 287         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 288                 udelay(10);
 289 }
 290
 291 #ifndef CONFIG_64BIT
 292 /*
 293  * this function sends a 'purge tlb' signal to another CPU.
 294  */
 295 static void smp_ptlb_callback(void *info)
 296 {
 297         __tlb_flush_local();
 298 }
 299
 300 void smp_ptlb_all(void)
 301 {
 302         on_each_cpu(smp_ptlb_callback, NULL, 1);
 303 }
 304 EXPORT_SYMBOL(smp_ptlb_all);
 305 #endif /* ! CONFIG_64BIT */
 306
 307 /*
 308  * this function sends a 'reschedule' IPI to another CPU.
 309  * it goes straight through and wastes no time serializing
 310  * anything. Worst case is that we lose a reschedule ...
 311  */
 312 void smp_send_reschedule(int cpu)
 313 {
 314         smp_ext_bitcall(cpu, ec_schedule);
 315 }
 316
 317 /*
 318  * parameter area for the set/clear control bit callbacks
 319  */
 320 struct ec_creg_mask_parms {
 321         unsigned long orvals[16];
 322         unsigned long andvals[16];
 323 };
 324
 325 /*
 326  * callback for setting/clearing control bits
 327  */
 328 static void smp_ctl_bit_callback(void *info)
 329 {
 330         struct ec_creg_mask_parms *pp = info;
 331         unsigned long cregs[16];
 332         int i;
 333
 334         __ctl_store(cregs, 0, 15);
 335         for (i = 0; i <= 15; i++)
 336                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 337         __ctl_load(cregs, 0, 15);
 338 }
 339
 340 /*
 341  * Set a bit in a control register of all cpus
 342  */
 343 void smp_ctl_set_bit(int cr, int bit)
 344 {
 345         struct ec_creg_mask_parms parms;
 346
 347         memset(&parms.orvals, 0, sizeof(parms.orvals));
 348         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 349         parms.orvals[cr] = 1 << bit;
 350         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 351 }
 352 EXPORT_SYMBOL(smp_ctl_set_bit);
 353
 354 /*
 355  * Clear a bit in a control register of all cpus
 356  */
 357 void smp_ctl_clear_bit(int cr, int bit)
 358 {
 359         struct ec_creg_mask_parms parms;
 360
 361         memset(&parms.orvals, 0, sizeof(parms.orvals));
 362         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 363         parms.andvals[cr] = ~(1L << bit);
 364         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 365 }
 366 EXPORT_SYMBOL(smp_ctl_clear_bit);
 367
 368 /*
 369  * In early ipl state a temp. logically cpu number is needed, so the sigp
 370  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 371  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 372  */
 373 #define CPU_INIT_NO     1
 374
 375 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 376
 377 /*
 378  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 379  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 380  * save its prefix registers, since they get lost, when switching from 31 bit
 381  * to 64 bit.
 382  */
 383 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 384         __attribute__((__section__(".data")));
 385
 386 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 387 {
 388         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 389                 return;
 390         if (cpu >= NR_CPUS) {
 391                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
 392                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
 393                 return;
 394         }
 395         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 396         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 397         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 398                sigp_busy)
 399                 cpu_relax();
 400         memcpy(zfcpdump_save_areas[cpu],
 401                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 402                SAVE_AREA_SIZE);
 403 #ifdef CONFIG_64BIT
 404         /* copy original prefix register */
 405         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 406 #endif
 407 }
 408
 409 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 410 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 411
 412 #else
 413
 414 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 415
 416 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 417
 418 static int cpu_stopped(int cpu)
 419 {
 420         __u32 status;
 421
 422         /* Check for stopped state */
 423         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 424             sigp_status_stored) {
 425                 if (status & 0x40)
 426                         return 1;
 427         }
 428         return 0;
 429 }
 430
 431 static int cpu_known(int cpu_id)
 432 {
 433         int cpu;
 434
 435         for_each_present_cpu(cpu) {
 436                 if (__cpu_logical_map[cpu] == cpu_id)
 437                         return 1;
 438         }
 439         return 0;
 440 }
 441
 442 static int smp_rescan_cpus_sigp(cpumask_t avail)
 443 {
 444         int cpu_id, logical_cpu;
 445
 446         logical_cpu = first_cpu(avail);
 447         if (logical_cpu == NR_CPUS)
 448                 return 0;
 449         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 450                 if (cpu_known(cpu_id))
 451                         continue;
 452                 __cpu_logical_map[logical_cpu] = cpu_id;
 453                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 454                 if (!cpu_stopped(logical_cpu))
 455                         continue;
 456                 cpu_set(logical_cpu, cpu_present_map);
 457                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 458                 logical_cpu = next_cpu(logical_cpu, avail);
 459                 if (logical_cpu == NR_CPUS)
 460                         break;
 461         }
 462         return 0;
 463 }
 464
 465 static int smp_rescan_cpus_sclp(cpumask_t avail)
 466 {
 467         struct sclp_cpu_info *info;
 468         int cpu_id, logical_cpu, cpu;
 469         int rc;
 470
 471         logical_cpu = first_cpu(avail);
 472         if (logical_cpu == NR_CPUS)
 473                 return 0;
 474         info = kmalloc(sizeof(*info), GFP_KERNEL);
 475         if (!info)
 476                 return -ENOMEM;
 477         rc = sclp_get_cpu_info(info);
 478         if (rc)
 479                 goto out;
 480         for (cpu = 0; cpu < info->combined; cpu++) {
 481                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 482                         continue;
 483                 cpu_id = info->cpu[cpu].address;
 484                 if (cpu_known(cpu_id))
 485                         continue;
 486                 __cpu_logical_map[logical_cpu] = cpu_id;
 487                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 488                 cpu_set(logical_cpu, cpu_present_map);
 489                 if (cpu >= info->configured)
 490                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 491                 else
 492                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 493                 logical_cpu = next_cpu(logical_cpu, avail);
 494                 if (logical_cpu == NR_CPUS)
 495                         break;
 496         }
 497 out:
 498         kfree(info);
 499         return rc;
 500 }
 501
 502 static int __smp_rescan_cpus(void)
 503 {
 504         cpumask_t avail;
 505
 506         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 507         if (smp_use_sigp_detection)
 508                 return smp_rescan_cpus_sigp(avail);
 509         else
 510                 return smp_rescan_cpus_sclp(avail);
 511 }
 512
 513 static void __init smp_detect_cpus(void)
 514 {
 515         unsigned int cpu, c_cpus, s_cpus;
 516         struct sclp_cpu_info *info;
 517         u16 boot_cpu_addr, cpu_addr;
 518
 519         c_cpus = 1;
 520         s_cpus = 0;
 521         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 522         info = kmalloc(sizeof(*info), GFP_KERNEL);
 523         if (!info)
 524                 panic("smp_detect_cpus failed to allocate memory\n");
 525         /* Use sigp detection algorithm if sclp doesn't work. */
 526         if (sclp_get_cpu_info(info)) {
 527                 smp_use_sigp_detection = 1;
 528                 for (cpu = 0; cpu <= 65535; cpu++) {
 529                         if (cpu == boot_cpu_addr)
 530                                 continue;
 531                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 532                         if (!cpu_stopped(CPU_INIT_NO))
 533                                 continue;
 534                         smp_get_save_area(c_cpus, cpu);
 535                         c_cpus++;
 536                 }
 537                 goto out;
 538         }
 539
 540         if (info->has_cpu_type) {
 541                 for (cpu = 0; cpu < info->combined; cpu++) {
 542                         if (info->cpu[cpu].address == boot_cpu_addr) {
 543                                 smp_cpu_type = info->cpu[cpu].type;
 544                                 break;
 545                         }
 546                 }
 547         }
 548
 549         for (cpu = 0; cpu < info->combined; cpu++) {
 550                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 551                         continue;
 552                 cpu_addr = info->cpu[cpu].address;
 553                 if (cpu_addr == boot_cpu_addr)
 554                         continue;
 555                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 556                 if (!cpu_stopped(CPU_INIT_NO)) {
 557                         s_cpus++;
 558                         continue;
 559                 }
 560                 smp_get_save_area(c_cpus, cpu_addr);
 561                 c_cpus++;
 562         }
 563 out:
 564         kfree(info);
 565         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
 566         get_online_cpus();
 567         __smp_rescan_cpus();
 568         put_online_cpus();
 569 }
 570
 571 /*
 572  *      Activate a secondary processor.
 573  */
 574 int __cpuinit start_secondary(void *cpuvoid)
 575 {
 576         /* Setup the cpu */
 577         cpu_init();
 578         preempt_disable();
 579         /* Enable TOD clock interrupts on the secondary cpu. */
 580         init_cpu_timer();
 581 #ifdef CONFIG_VIRT_TIMER
 582         /* Enable cpu timer interrupts on the secondary cpu. */
 583         init_cpu_vtimer();
 584 #endif
 585         /* Enable pfault pseudo page faults on this cpu. */
 586         pfault_init();
 587
 588         /* Mark this cpu as online */
 589         spin_lock(&call_lock);
 590         cpu_set(smp_processor_id(), cpu_online_map);
 591         spin_unlock(&call_lock);
 592         /* Switch on interrupts */
 593         local_irq_enable();
 594         /* Print info about this processor */
 595         print_cpu_info(&S390_lowcore.cpu_data);
 596         /* cpu_idle will call schedule for us */
 597         cpu_idle();
 598         return 0;
 599 }
 600
 601 static void __init smp_create_idle(unsigned int cpu)
 602 {
 603         struct task_struct *p;
 604
 605         /*
 606          *  don't care about the psw and regs settings since we'll never
 607          *  reschedule the forked task.
 608          */
 609         p = fork_idle(cpu);
 610         if (IS_ERR(p))
 611                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 612         current_set[cpu] = p;
 613         spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
 614 }
 615
 616 static int __cpuinit smp_alloc_lowcore(int cpu)
 617 {
 618         unsigned long async_stack, panic_stack;
 619         struct _lowcore *lowcore;
 620         int lc_order;
 621
 622         lc_order = sizeof(long) == 8 ? 1 : 0;
 623         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 624         if (!lowcore)
 625                 return -ENOMEM;
 626         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 627         panic_stack = __get_free_page(GFP_KERNEL);
 628         if (!panic_stack || !async_stack)
 629                 goto out;
 630         memcpy(lowcore, &S390_lowcore, 512);
 631         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 632         lowcore->async_stack = async_stack + ASYNC_SIZE;
 633         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 634
 635 #ifndef CONFIG_64BIT
 636         if (MACHINE_HAS_IEEE) {
 637                 unsigned long save_area;
 638
 639                 save_area = get_zeroed_page(GFP_KERNEL);
 640                 if (!save_area)
 641                         goto out_save_area;
 642                 lowcore->extended_save_area_addr = (u32) save_area;
 643         }
 644 #endif
 645         lowcore_ptr[cpu] = lowcore;
 646         return 0;
 647
 648 #ifndef CONFIG_64BIT
 649 out_save_area:
 650         free_page(panic_stack);
 651 #endif
 652 out:
 653         free_pages(async_stack, ASYNC_ORDER);
 654         free_pages((unsigned long) lowcore, lc_order);
 655         return -ENOMEM;
 656 }
 657
 658 #ifdef CONFIG_HOTPLUG_CPU
 659 static void smp_free_lowcore(int cpu)
 660 {
 661         struct _lowcore *lowcore;
 662         int lc_order;
 663
 664         lc_order = sizeof(long) == 8 ? 1 : 0;
 665         lowcore = lowcore_ptr[cpu];
 666 #ifndef CONFIG_64BIT
 667         if (MACHINE_HAS_IEEE)
 668                 free_page((unsigned long) lowcore->extended_save_area_addr);
 669 #endif
 670         free_page(lowcore->panic_stack - PAGE_SIZE);
 671         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 672         free_pages((unsigned long) lowcore, lc_order);
 673         lowcore_ptr[cpu] = NULL;
 674 }
 675 #endif /* CONFIG_HOTPLUG_CPU */
 676
 677 /* Upping and downing of CPUs */
 678 int __cpuinit __cpu_up(unsigned int cpu)
 679 {
 680         struct task_struct *idle;
 681         struct _lowcore *cpu_lowcore;
 682         struct stack_frame *sf;
 683         sigp_ccode ccode;
 684
 685         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 686                 return -EIO;
 687         if (smp_alloc_lowcore(cpu))
 688                 return -ENOMEM;
 689
 690         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 691                                    cpu, sigp_set_prefix);
 692         if (ccode) {
 693                 printk("sigp_set_prefix failed for cpu %d "
 694                        "with condition code %d\n",
 695                        (int) cpu, (int) ccode);
 696                 return -EIO;
 697         }
 698
 699         idle = current_set[cpu];
 700         cpu_lowcore = lowcore_ptr[cpu];
 701         cpu_lowcore->kernel_stack = (unsigned long)
 702                 task_stack_page(idle) + THREAD_SIZE;
 703         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 704         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 705                                      - sizeof(struct pt_regs)
 706                                      - sizeof(struct stack_frame));
 707         memset(sf, 0, sizeof(struct stack_frame));
 708         sf->gprs[9] = (unsigned long) sf;
 709         cpu_lowcore->save_area[15] = (unsigned long) sf;
 710         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 711         asm volatile(
 712                 "       stam    0,15,0(%0)"
 713                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 714         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 715         cpu_lowcore->current_task = (unsigned long) idle;
 716         cpu_lowcore->cpu_data.cpu_nr = cpu;
 717         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 718         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
 719         eieio();
 720
 721         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 722                 udelay(10);
 723
 724         while (!cpu_online(cpu))
 725                 cpu_relax();
 726         return 0;
 727 }
 728
 729 static int __init setup_possible_cpus(char *s)
 730 {
 731         int pcpus, cpu;
 732
 733         pcpus = simple_strtoul(s, NULL, 0);
 734         cpu_possible_map = cpumask_of_cpu(0);
 735         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 736                 cpu_set(cpu, cpu_possible_map);
 737         return 0;
 738 }
 739 early_param("possible_cpus", setup_possible_cpus);
 740
 741 #ifdef CONFIG_HOTPLUG_CPU
 742
 743 int __cpu_disable(void)
 744 {
 745         struct ec_creg_mask_parms cr_parms;
 746         int cpu = smp_processor_id();
 747
 748         cpu_clear(cpu, cpu_online_map);
 749
 750         /* Disable pfault pseudo page faults on this cpu. */
 751         pfault_fini();
 752
 753         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 754         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 755
 756         /* disable all external interrupts */
 757         cr_parms.orvals[0] = 0;
 758         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 759                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 760         /* disable all I/O interrupts */
 761         cr_parms.orvals[6] = 0;
 762         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 763                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 764         /* disable most machine checks */
 765         cr_parms.orvals[14] = 0;
 766         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 767                                  1 << 25 | 1 << 24);
 768
 769         smp_ctl_bit_callback(&cr_parms);
 770
 771         return 0;
 772 }
 773
 774 void __cpu_die(unsigned int cpu)
 775 {
 776         /* Wait until target cpu is down */
 777         while (!smp_cpu_not_running(cpu))
 778                 cpu_relax();
 779         smp_free_lowcore(cpu);
 780         printk(KERN_INFO "Processor %d spun down\n", cpu);
 781 }
 782
 783 void cpu_die(void)
 784 {
 785         idle_task_exit();
 786         signal_processor(smp_processor_id(), sigp_stop);
 787         BUG();
 788         for (;;);
 789 }
 790
 791 #endif /* CONFIG_HOTPLUG_CPU */
 792
 793 void __init smp_prepare_cpus(unsigned int max_cpus)
 794 {
 795 #ifndef CONFIG_64BIT
 796         unsigned long save_area = 0;
 797 #endif
 798         unsigned long async_stack, panic_stack;
 799         struct _lowcore *lowcore;
 800         unsigned int cpu;
 801         int lc_order;
 802
 803         smp_detect_cpus();
 804
 805         /* request the 0x1201 emergency signal external interrupt */
 806         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 807                 panic("Couldn't request external interrupt 0x1201");
 808         print_cpu_info(&S390_lowcore.cpu_data);
 809
 810         /* Reallocate current lowcore, but keep its contents. */
 811         lc_order = sizeof(long) == 8 ? 1 : 0;
 812         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 813         panic_stack = __get_free_page(GFP_KERNEL);
 814         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 815 #ifndef CONFIG_64BIT
 816         if (MACHINE_HAS_IEEE)
 817                 save_area = get_zeroed_page(GFP_KERNEL);
 818 #endif
 819         local_irq_disable();
 820         local_mcck_disable();
 821         lowcore_ptr[smp_processor_id()] = lowcore;
 822         *lowcore = S390_lowcore;
 823         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 824         lowcore->async_stack = async_stack + ASYNC_SIZE;
 825 #ifndef CONFIG_64BIT
 826         if (MACHINE_HAS_IEEE)
 827                 lowcore->extended_save_area_addr = (u32) save_area;
 828 #endif
 829         set_prefix((u32)(unsigned long) lowcore);
 830         local_mcck_enable();
 831         local_irq_enable();
 832         for_each_possible_cpu(cpu)
 833                 if (cpu != smp_processor_id())
 834                         smp_create_idle(cpu);
 835 }
 836
 837 void __init smp_prepare_boot_cpu(void)
 838 {
 839         BUG_ON(smp_processor_id() != 0);
 840
 841         current_thread_info()->cpu = 0;
 842         cpu_set(0, cpu_present_map);
 843         cpu_set(0, cpu_online_map);
 844         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 845         current_set[0] = current;
 846         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 847         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 848         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 849 }
 850
 851 void __init smp_cpus_done(unsigned int max_cpus)
 852 {
 853 }
 854
 855 /*
 856  * the frequency of the profiling timer can be changed
 857  * by writing a multiplier value into /proc/profile.
 858  *
 859  * usually you want to run this on all CPUs ;)
 860  */
 861 int setup_profiling_timer(unsigned int multiplier)
 862 {
 863         return 0;
 864 }
 865
 866 #ifdef CONFIG_HOTPLUG_CPU
 867 static ssize_t cpu_configure_show(struct sys_device *dev,
 868                                 struct sysdev_attribute *attr, 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,
 879                                   struct sysdev_attribute *attr,
 880                                   const char *buf, size_t count)
 881 {
 882         int cpu = dev->id;
 883         int val, rc;
 884         char delim;
 885
 886         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 887                 return -EINVAL;
 888         if (val != 0 && val != 1)
 889                 return -EINVAL;
 890
 891         get_online_cpus();
 892         mutex_lock(&smp_cpu_state_mutex);
 893         rc = -EBUSY;
 894         if (cpu_online(cpu))
 895                 goto out;
 896         rc = 0;
 897         switch (val) {
 898         case 0:
 899                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 900                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 901                         if (!rc) {
 902                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 903                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 904                         }
 905                 }
 906                 break;
 907         case 1:
 908                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 909                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 910                         if (!rc) {
 911                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 912                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 913                         }
 914                 }
 915                 break;
 916         default:
 917                 break;
 918         }
 919 out:
 920         mutex_unlock(&smp_cpu_state_mutex);
 921         put_online_cpus();
 922         return rc ? rc : count;
 923 }
 924 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 925 #endif /* CONFIG_HOTPLUG_CPU */
 926
 927 static ssize_t cpu_polarization_show(struct sys_device *dev,
 928                                      struct sysdev_attribute *attr, char *buf)
 929 {
 930         int cpu = dev->id;
 931         ssize_t count;
 932
 933         mutex_lock(&smp_cpu_state_mutex);
 934         switch (smp_cpu_polarization[cpu]) {
 935         case POLARIZATION_HRZ:
 936                 count = sprintf(buf, "horizontal\n");
 937                 break;
 938         case POLARIZATION_VL:
 939                 count = sprintf(buf, "vertical:low\n");
 940                 break;
 941         case POLARIZATION_VM:
 942                 count = sprintf(buf, "vertical:medium\n");
 943                 break;
 944         case POLARIZATION_VH:
 945                 count = sprintf(buf, "vertical:high\n");
 946                 break;
 947         default:
 948                 count = sprintf(buf, "unknown\n");
 949                 break;
 950         }
 951         mutex_unlock(&smp_cpu_state_mutex);
 952         return count;
 953 }
 954 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 955
 956 static ssize_t show_cpu_address(struct sys_device *dev,
 957                                 struct sysdev_attribute *attr, char *buf)
 958 {
 959         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 960 }
 961 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 962
 963
 964 static struct attribute *cpu_common_attrs[] = {
 965 #ifdef CONFIG_HOTPLUG_CPU
 966         &attr_configure.attr,
 967 #endif
 968         &attr_address.attr,
 969         &attr_polarization.attr,
 970         NULL,
 971 };
 972
 973 static struct attribute_group cpu_common_attr_group = {
 974         .attrs = cpu_common_attrs,
 975 };
 976
 977 static ssize_t show_capability(struct sys_device *dev,
 978                                 struct sysdev_attribute *attr, char *buf)
 979 {
 980         unsigned int capability;
 981         int rc;
 982
 983         rc = get_cpu_capability(&capability);
 984         if (rc)
 985                 return rc;
 986         return sprintf(buf, "%u\n", capability);
 987 }
 988 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 989
 990 static ssize_t show_idle_count(struct sys_device *dev,
 991                                 struct sysdev_attribute *attr, char *buf)
 992 {
 993         struct s390_idle_data *idle;
 994         unsigned long long idle_count;
 995
 996         idle = &per_cpu(s390_idle, dev->id);
 997         spin_lock_irq(&idle->lock);
 998         idle_count = idle->idle_count;
 999         spin_unlock_irq(&idle->lock);
1000         return sprintf(buf, "%llu\n", idle_count);
1001 }
1002 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
1003
1004 static ssize_t show_idle_time(struct sys_device *dev,
1005                                 struct sysdev_attribute *attr, char *buf)
1006 {
1007         struct s390_idle_data *idle;
1008         unsigned long long new_time;
1009
1010         idle = &per_cpu(s390_idle, dev->id);
1011         spin_lock_irq(&idle->lock);
1012         if (idle->in_idle) {
1013                 new_time = get_clock();
1014                 idle->idle_time += new_time - idle->idle_enter;
1015                 idle->idle_enter = new_time;
1016         }
1017         new_time = idle->idle_time;
1018         spin_unlock_irq(&idle->lock);
1019         return sprintf(buf, "%llu\n", new_time >> 12);
1020 }
1021 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1022
1023 static struct attribute *cpu_online_attrs[] = {
1024         &attr_capability.attr,
1025         &attr_idle_count.attr,
1026         &attr_idle_time_us.attr,
1027         NULL,
1028 };
1029
1030 static struct attribute_group cpu_online_attr_group = {
1031         .attrs = cpu_online_attrs,
1032 };
1033
1034 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1035                                     unsigned long action, void *hcpu)
1036 {
1037         unsigned int cpu = (unsigned int)(long)hcpu;
1038         struct cpu *c = &per_cpu(cpu_devices, cpu);
1039         struct sys_device *s = &c->sysdev;
1040         struct s390_idle_data *idle;
1041
1042         switch (action) {
1043         case CPU_ONLINE:
1044         case CPU_ONLINE_FROZEN:
1045                 idle = &per_cpu(s390_idle, cpu);
1046                 spin_lock_irq(&idle->lock);
1047                 idle->idle_enter = 0;
1048                 idle->idle_time = 0;
1049                 idle->idle_count = 0;
1050                 spin_unlock_irq(&idle->lock);
1051                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
1052                         return NOTIFY_BAD;
1053                 break;
1054         case CPU_DEAD:
1055         case CPU_DEAD_FROZEN:
1056                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1057                 break;
1058         }
1059         return NOTIFY_OK;
1060 }
1061
1062 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1063         .notifier_call = smp_cpu_notify,
1064 };
1065
1066 static int __devinit smp_add_present_cpu(int cpu)
1067 {
1068         struct cpu *c = &per_cpu(cpu_devices, cpu);
1069         struct sys_device *s = &c->sysdev;
1070         int rc;
1071
1072         c->hotpluggable = 1;
1073         rc = register_cpu(c, cpu);
1074         if (rc)
1075                 goto out;
1076         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1077         if (rc)
1078                 goto out_cpu;
1079         if (!cpu_online(cpu))
1080                 goto out;
1081         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1082         if (!rc)
1083                 return 0;
1084         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1085 out_cpu:
1086 #ifdef CONFIG_HOTPLUG_CPU
1087         unregister_cpu(c);
1088 #endif
1089 out:
1090         return rc;
1091 }
1092
1093 #ifdef CONFIG_HOTPLUG_CPU
1094
1095 int __ref smp_rescan_cpus(void)
1096 {
1097         cpumask_t newcpus;
1098         int cpu;
1099         int rc;
1100
1101         get_online_cpus();
1102         mutex_lock(&smp_cpu_state_mutex);
1103         newcpus = cpu_present_map;
1104         rc = __smp_rescan_cpus();
1105         if (rc)
1106                 goto out;
1107         cpus_andnot(newcpus, cpu_present_map, newcpus);
1108         for_each_cpu_mask(cpu, newcpus) {
1109                 rc = smp_add_present_cpu(cpu);
1110                 if (rc)
1111                         cpu_clear(cpu, cpu_present_map);
1112         }
1113         rc = 0;
1114 out:
1115         mutex_unlock(&smp_cpu_state_mutex);
1116         put_online_cpus();
1117         if (!cpus_empty(newcpus))
1118                 topology_schedule_update();
1119         return rc;
1120 }
1121
1122 static ssize_t __ref rescan_store(struct sys_device *dev,
1123                                   struct sysdev_attribute *attr,
1124                                   const char *buf,
1125                                   size_t count)
1126 {
1127         int rc;
1128
1129         rc = smp_rescan_cpus();
1130         return rc ? rc : count;
1131 }
1132 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1133 #endif /* CONFIG_HOTPLUG_CPU */
1134
1135 static ssize_t dispatching_show(struct sys_device *dev,
1136                                 struct sysdev_attribute *attr,
1137                                 char *buf)
1138 {
1139         ssize_t count;
1140
1141         mutex_lock(&smp_cpu_state_mutex);
1142         count = sprintf(buf, "%d\n", cpu_management);
1143         mutex_unlock(&smp_cpu_state_mutex);
1144         return count;
1145 }
1146
1147 static ssize_t dispatching_store(struct sys_device *dev,
1148                                  struct sysdev_attribute *attr,
1149                                  const char *buf, size_t count)
1150 {
1151         int val, rc;
1152         char delim;
1153
1154         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1155                 return -EINVAL;
1156         if (val != 0 && val != 1)
1157                 return -EINVAL;
1158         rc = 0;
1159         get_online_cpus();
1160         mutex_lock(&smp_cpu_state_mutex);
1161         if (cpu_management == val)
1162                 goto out;
1163         rc = topology_set_cpu_management(val);
1164         if (!rc)
1165                 cpu_management = val;
1166 out:
1167         mutex_unlock(&smp_cpu_state_mutex);
1168         put_online_cpus();
1169         return rc ? rc : count;
1170 }
1171 static SYSDEV_ATTR(dispatching, 0644, dispatching_show, dispatching_store);
1172
1173 static int __init topology_init(void)
1174 {
1175         int cpu;
1176         int rc;
1177
1178         register_cpu_notifier(&smp_cpu_nb);
1179
1180 #ifdef CONFIG_HOTPLUG_CPU
1181         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1182                                &attr_rescan.attr);
1183         if (rc)
1184                 return rc;
1185 #endif
1186         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1187                                &attr_dispatching.attr);
1188         if (rc)
1189                 return rc;
1190         for_each_present_cpu(cpu) {
1191                 rc = smp_add_present_cpu(cpu);
1192                 if (rc)
1193                         return rc;
1194         }
1195         return 0;
1196 }
1197 subsys_initcall(topology_init);