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 DEFINE_MUTEX(smp_cpu_state_mutex);
  71 int smp_cpu_polarization[NR_CPUS];
  72 static int smp_cpu_state[NR_CPUS];
  73 static int cpu_management;
  74
  75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  76
  77 static void smp_ext_bitcall(int, ec_bit_sig);
  78
  79 /*
  80  * Structure and data for __smp_call_function_map(). This is designed to
  81  * minimise static memory requirements. It also looks cleaner.
  82  */
  83 static DEFINE_SPINLOCK(call_lock);
  84
  85 struct call_data_struct {
  86         void (*func) (void *info);
  87         void *info;
  88         cpumask_t started;
  89         cpumask_t finished;
  90         int wait;
  91 };
  92
  93 static struct call_data_struct *call_data;
  94
  95 /*
  96  * 'Call function' interrupt callback
  97  */
  98 static void do_call_function(void)
  99 {
 100         void (*func) (void *info) = call_data->func;
 101         void *info = call_data->info;
 102         int wait = call_data->wait;
 103
 104         cpu_set(smp_processor_id(), call_data->started);
 105         (*func)(info);
 106         if (wait)
 107                 cpu_set(smp_processor_id(), call_data->finished);;
 108 }
 109
 110 static void __smp_call_function_map(void (*func) (void *info), void *info,
 111                                     int nonatomic, int wait, cpumask_t map)
 112 {
 113         struct call_data_struct data;
 114         int cpu, local = 0;
 115
 116         /*
 117          * Can deadlock when interrupts are disabled or if in wrong context.
 118          */
 119         WARN_ON(irqs_disabled() || in_irq());
 120
 121         /*
 122          * Check for local function call. We have to have the same call order
 123          * as in on_each_cpu() because of machine_restart_smp().
 124          */
 125         if (cpu_isset(smp_processor_id(), map)) {
 126                 local = 1;
 127                 cpu_clear(smp_processor_id(), map);
 128         }
 129
 130         cpus_and(map, map, cpu_online_map);
 131         if (cpus_empty(map))
 132                 goto out;
 133
 134         data.func = func;
 135         data.info = info;
 136         data.started = CPU_MASK_NONE;
 137         data.wait = wait;
 138         if (wait)
 139                 data.finished = CPU_MASK_NONE;
 140
 141         spin_lock(&call_lock);
 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         spin_unlock(&call_lock);
 154 out:
 155         if (local) {
 156                 local_irq_disable();
 157                 func(info);
 158                 local_irq_enable();
 159         }
 160 }
 161
 162 /*
 163  * smp_call_function:
 164  * @func: the function to run; this must be fast and non-blocking
 165  * @info: an arbitrary pointer to pass to the function
 166  * @nonatomic: unused
 167  * @wait: if true, wait (atomically) until function has completed on other CPUs
 168  *
 169  * Run a function on all other CPUs.
 170  *
 171  * You must not call this function with disabled interrupts, from a
 172  * hardware interrupt handler or from a bottom half.
 173  */
 174 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
 175                       int wait)
 176 {
 177         cpumask_t map;
 178
 179         preempt_disable();
 180         map = cpu_online_map;
 181         cpu_clear(smp_processor_id(), map);
 182         __smp_call_function_map(func, info, nonatomic, wait, map);
 183         preempt_enable();
 184         return 0;
 185 }
 186 EXPORT_SYMBOL(smp_call_function);
 187
 188 /*
 189  * smp_call_function_single:
 190  * @cpu: the CPU where func should run
 191  * @func: the function to run; this must be fast and non-blocking
 192  * @info: an arbitrary pointer to pass to the function
 193  * @nonatomic: unused
 194  * @wait: if true, wait (atomically) until function has completed on other CPUs
 195  *
 196  * Run a function on one processor.
 197  *
 198  * You must not call this function with disabled interrupts, from a
 199  * hardware interrupt handler or from a bottom half.
 200  */
 201 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 202                              int nonatomic, int wait)
 203 {
 204         preempt_disable();
 205         __smp_call_function_map(func, info, nonatomic, wait,
 206                                 cpumask_of_cpu(cpu));
 207         preempt_enable();
 208         return 0;
 209 }
 210 EXPORT_SYMBOL(smp_call_function_single);
 211
 212 /**
 213  * smp_call_function_mask(): Run a function on a set of other CPUs.
 214  * @mask: The set of cpus to run on.  Must not include the current cpu.
 215  * @func: The function to run. This must be fast and non-blocking.
 216  * @info: An arbitrary pointer to pass to the function.
 217  * @wait: If true, wait (atomically) until function has completed on other CPUs.
 218  *
 219  * Returns 0 on success, else a negative status code.
 220  *
 221  * If @wait is true, then returns once @func has returned; otherwise
 222  * it returns just before the target cpu calls @func.
 223  *
 224  * You must not call this function with disabled interrupts or from a
 225  * hardware interrupt handler or from a bottom half handler.
 226  */
 227 int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
 228                            int wait)
 229 {
 230         preempt_disable();
 231         cpu_clear(smp_processor_id(), mask);
 232         __smp_call_function_map(func, info, 0, wait, mask);
 233         preempt_enable();
 234         return 0;
 235 }
 236 EXPORT_SYMBOL(smp_call_function_mask);
 237
 238 void smp_send_stop(void)
 239 {
 240         int cpu, rc;
 241
 242         /* Disable all interrupts/machine checks */
 243         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 244
 245         /* write magic number to zero page (absolute 0) */
 246         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
 247
 248         /* stop all processors */
 249         for_each_online_cpu(cpu) {
 250                 if (cpu == smp_processor_id())
 251                         continue;
 252                 do {
 253                         rc = signal_processor(cpu, sigp_stop);
 254                 } while (rc == sigp_busy);
 255
 256                 while (!smp_cpu_not_running(cpu))
 257                         cpu_relax();
 258         }
 259 }
 260
 261 /*
 262  * This is the main routine where commands issued by other
 263  * cpus are handled.
 264  */
 265
 266 static void do_ext_call_interrupt(__u16 code)
 267 {
 268         unsigned long bits;
 269
 270         /*
 271          * handle bit signal external calls
 272          *
 273          * For the ec_schedule signal we have to do nothing. All the work
 274          * is done automatically when we return from the interrupt.
 275          */
 276         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 277
 278         if (test_bit(ec_call_function, &bits))
 279                 do_call_function();
 280 }
 281
 282 /*
 283  * Send an external call sigp to another cpu and return without waiting
 284  * for its completion.
 285  */
 286 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 287 {
 288         /*
 289          * Set signaling bit in lowcore of target cpu and kick it
 290          */
 291         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 292         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 293                 udelay(10);
 294 }
 295
 296 #ifndef CONFIG_64BIT
 297 /*
 298  * this function sends a 'purge tlb' signal to another CPU.
 299  */
 300 void smp_ptlb_callback(void *info)
 301 {
 302         __tlb_flush_local();
 303 }
 304
 305 void smp_ptlb_all(void)
 306 {
 307         on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
 308 }
 309 EXPORT_SYMBOL(smp_ptlb_all);
 310 #endif /* ! CONFIG_64BIT */
 311
 312 /*
 313  * this function sends a 'reschedule' IPI to another CPU.
 314  * it goes straight through and wastes no time serializing
 315  * anything. Worst case is that we lose a reschedule ...
 316  */
 317 void smp_send_reschedule(int cpu)
 318 {
 319         smp_ext_bitcall(cpu, ec_schedule);
 320 }
 321
 322 /*
 323  * parameter area for the set/clear control bit callbacks
 324  */
 325 struct ec_creg_mask_parms {
 326         unsigned long orvals[16];
 327         unsigned long andvals[16];
 328 };
 329
 330 /*
 331  * callback for setting/clearing control bits
 332  */
 333 static void smp_ctl_bit_callback(void *info)
 334 {
 335         struct ec_creg_mask_parms *pp = info;
 336         unsigned long cregs[16];
 337         int i;
 338
 339         __ctl_store(cregs, 0, 15);
 340         for (i = 0; i <= 15; i++)
 341                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 342         __ctl_load(cregs, 0, 15);
 343 }
 344
 345 /*
 346  * Set a bit in a control register of all cpus
 347  */
 348 void smp_ctl_set_bit(int cr, int bit)
 349 {
 350         struct ec_creg_mask_parms parms;
 351
 352         memset(&parms.orvals, 0, sizeof(parms.orvals));
 353         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 354         parms.orvals[cr] = 1 << bit;
 355         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 356 }
 357 EXPORT_SYMBOL(smp_ctl_set_bit);
 358
 359 /*
 360  * Clear a bit in a control register of all cpus
 361  */
 362 void smp_ctl_clear_bit(int cr, int bit)
 363 {
 364         struct ec_creg_mask_parms parms;
 365
 366         memset(&parms.orvals, 0, sizeof(parms.orvals));
 367         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 368         parms.andvals[cr] = ~(1L << bit);
 369         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 370 }
 371 EXPORT_SYMBOL(smp_ctl_clear_bit);
 372
 373 /*
 374  * In early ipl state a temp. logically cpu number is needed, so the sigp
 375  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 376  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 377  */
 378 #define CPU_INIT_NO     1
 379
 380 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 381
 382 /*
 383  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 384  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 385  * save its prefix registers, since they get lost, when switching from 31 bit
 386  * to 64 bit.
 387  */
 388 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 389         __attribute__((__section__(".data")));
 390
 391 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 392 {
 393         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 394                 return;
 395         if (cpu >= NR_CPUS) {
 396                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
 397                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
 398                 return;
 399         }
 400         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 401         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 402         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 403                sigp_busy)
 404                 cpu_relax();
 405         memcpy(zfcpdump_save_areas[cpu],
 406                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 407                SAVE_AREA_SIZE);
 408 #ifdef CONFIG_64BIT
 409         /* copy original prefix register */
 410         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 411 #endif
 412 }
 413
 414 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 415 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 416
 417 #else
 418
 419 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 420
 421 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 422
 423 static int cpu_stopped(int cpu)
 424 {
 425         __u32 status;
 426
 427         /* Check for stopped state */
 428         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 429             sigp_status_stored) {
 430                 if (status & 0x40)
 431                         return 1;
 432         }
 433         return 0;
 434 }
 435
 436 static int cpu_known(int cpu_id)
 437 {
 438         int cpu;
 439
 440         for_each_present_cpu(cpu) {
 441                 if (__cpu_logical_map[cpu] == cpu_id)
 442                         return 1;
 443         }
 444         return 0;
 445 }
 446
 447 static int smp_rescan_cpus_sigp(cpumask_t avail)
 448 {
 449         int cpu_id, logical_cpu;
 450
 451         logical_cpu = first_cpu(avail);
 452         if (logical_cpu == NR_CPUS)
 453                 return 0;
 454         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 455                 if (cpu_known(cpu_id))
 456                         continue;
 457                 __cpu_logical_map[logical_cpu] = cpu_id;
 458                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 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                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 493                 cpu_set(logical_cpu, cpu_present_map);
 494                 if (cpu >= info->configured)
 495                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 496                 else
 497                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 498                 logical_cpu = next_cpu(logical_cpu, avail);
 499                 if (logical_cpu == NR_CPUS)
 500                         break;
 501         }
 502 out:
 503         kfree(info);
 504         return rc;
 505 }
 506
 507 static int smp_rescan_cpus(void)
 508 {
 509         cpumask_t avail;
 510
 511         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 512         if (smp_use_sigp_detection)
 513                 return smp_rescan_cpus_sigp(avail);
 514         else
 515                 return smp_rescan_cpus_sclp(avail);
 516 }
 517
 518 static void __init smp_detect_cpus(void)
 519 {
 520         unsigned int cpu, c_cpus, s_cpus;
 521         struct sclp_cpu_info *info;
 522         u16 boot_cpu_addr, cpu_addr;
 523
 524         c_cpus = 1;
 525         s_cpus = 0;
 526         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 527         info = kmalloc(sizeof(*info), GFP_KERNEL);
 528         if (!info)
 529                 panic("smp_detect_cpus failed to allocate memory\n");
 530         /* Use sigp detection algorithm if sclp doesn't work. */
 531         if (sclp_get_cpu_info(info)) {
 532                 smp_use_sigp_detection = 1;
 533                 for (cpu = 0; cpu <= 65535; cpu++) {
 534                         if (cpu == boot_cpu_addr)
 535                                 continue;
 536                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 537                         if (!cpu_stopped(CPU_INIT_NO))
 538                                 continue;
 539                         smp_get_save_area(c_cpus, cpu);
 540                         c_cpus++;
 541                 }
 542                 goto out;
 543         }
 544
 545         if (info->has_cpu_type) {
 546                 for (cpu = 0; cpu < info->combined; cpu++) {
 547                         if (info->cpu[cpu].address == boot_cpu_addr) {
 548                                 smp_cpu_type = info->cpu[cpu].type;
 549                                 break;
 550                         }
 551                 }
 552         }
 553
 554         for (cpu = 0; cpu < info->combined; cpu++) {
 555                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 556                         continue;
 557                 cpu_addr = info->cpu[cpu].address;
 558                 if (cpu_addr == boot_cpu_addr)
 559                         continue;
 560                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 561                 if (!cpu_stopped(CPU_INIT_NO)) {
 562                         s_cpus++;
 563                         continue;
 564                 }
 565                 smp_get_save_area(c_cpus, cpu_addr);
 566                 c_cpus++;
 567         }
 568 out:
 569         kfree(info);
 570         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
 571         get_online_cpus();
 572         smp_rescan_cpus();
 573         put_online_cpus();
 574 }
 575
 576 /*
 577  *      Activate a secondary processor.
 578  */
 579 int __cpuinit start_secondary(void *cpuvoid)
 580 {
 581         /* Setup the cpu */
 582         cpu_init();
 583         preempt_disable();
 584         /* Enable TOD clock interrupts on the secondary cpu. */
 585         init_cpu_timer();
 586 #ifdef CONFIG_VIRT_TIMER
 587         /* Enable cpu timer interrupts on the secondary cpu. */
 588         init_cpu_vtimer();
 589 #endif
 590         /* Enable pfault pseudo page faults on this cpu. */
 591         pfault_init();
 592
 593         /* Mark this cpu as online */
 594         cpu_set(smp_processor_id(), cpu_online_map);
 595         /* Switch on interrupts */
 596         local_irq_enable();
 597         /* Print info about this processor */
 598         print_cpu_info(&S390_lowcore.cpu_data);
 599         /* cpu_idle will call schedule for us */
 600         cpu_idle();
 601         return 0;
 602 }
 603
 604 static void __init smp_create_idle(unsigned int cpu)
 605 {
 606         struct task_struct *p;
 607
 608         /*
 609          *  don't care about the psw and regs settings since we'll never
 610          *  reschedule the forked task.
 611          */
 612         p = fork_idle(cpu);
 613         if (IS_ERR(p))
 614                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 615         current_set[cpu] = p;
 616         spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
 617 }
 618
 619 static int __cpuinit smp_alloc_lowcore(int cpu)
 620 {
 621         unsigned long async_stack, panic_stack;
 622         struct _lowcore *lowcore;
 623         int lc_order;
 624
 625         lc_order = sizeof(long) == 8 ? 1 : 0;
 626         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 627         if (!lowcore)
 628                 return -ENOMEM;
 629         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 630         panic_stack = __get_free_page(GFP_KERNEL);
 631         if (!panic_stack || !async_stack)
 632                 goto out;
 633         memcpy(lowcore, &S390_lowcore, 512);
 634         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 635         lowcore->async_stack = async_stack + ASYNC_SIZE;
 636         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 637
 638 #ifndef CONFIG_64BIT
 639         if (MACHINE_HAS_IEEE) {
 640                 unsigned long save_area;
 641
 642                 save_area = get_zeroed_page(GFP_KERNEL);
 643                 if (!save_area)
 644                         goto out_save_area;
 645                 lowcore->extended_save_area_addr = (u32) save_area;
 646         }
 647 #endif
 648         lowcore_ptr[cpu] = lowcore;
 649         return 0;
 650
 651 #ifndef CONFIG_64BIT
 652 out_save_area:
 653         free_page(panic_stack);
 654 #endif
 655 out:
 656         free_pages(async_stack, ASYNC_ORDER);
 657         free_pages((unsigned long) lowcore, lc_order);
 658         return -ENOMEM;
 659 }
 660
 661 #ifdef CONFIG_HOTPLUG_CPU
 662 static void smp_free_lowcore(int cpu)
 663 {
 664         struct _lowcore *lowcore;
 665         int lc_order;
 666
 667         lc_order = sizeof(long) == 8 ? 1 : 0;
 668         lowcore = lowcore_ptr[cpu];
 669 #ifndef CONFIG_64BIT
 670         if (MACHINE_HAS_IEEE)
 671                 free_page((unsigned long) lowcore->extended_save_area_addr);
 672 #endif
 673         free_page(lowcore->panic_stack - PAGE_SIZE);
 674         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 675         free_pages((unsigned long) lowcore, lc_order);
 676         lowcore_ptr[cpu] = NULL;
 677 }
 678 #endif /* CONFIG_HOTPLUG_CPU */
 679
 680 /* Upping and downing of CPUs */
 681 int __cpuinit __cpu_up(unsigned int cpu)
 682 {
 683         struct task_struct *idle;
 684         struct _lowcore *cpu_lowcore;
 685         struct stack_frame *sf;
 686         sigp_ccode ccode;
 687
 688         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 689                 return -EIO;
 690         if (smp_alloc_lowcore(cpu))
 691                 return -ENOMEM;
 692
 693         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 694                                    cpu, sigp_set_prefix);
 695         if (ccode) {
 696                 printk("sigp_set_prefix failed for cpu %d "
 697                        "with condition code %d\n",
 698                        (int) cpu, (int) ccode);
 699                 return -EIO;
 700         }
 701
 702         idle = current_set[cpu];
 703         cpu_lowcore = lowcore_ptr[cpu];
 704         cpu_lowcore->kernel_stack = (unsigned long)
 705                 task_stack_page(idle) + THREAD_SIZE;
 706         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 707         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 708                                      - sizeof(struct pt_regs)
 709                                      - sizeof(struct stack_frame));
 710         memset(sf, 0, sizeof(struct stack_frame));
 711         sf->gprs[9] = (unsigned long) sf;
 712         cpu_lowcore->save_area[15] = (unsigned long) sf;
 713         __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
 714         asm volatile(
 715                 "       stam    0,15,0(%0)"
 716                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 717         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 718         cpu_lowcore->current_task = (unsigned long) idle;
 719         cpu_lowcore->cpu_data.cpu_nr = cpu;
 720         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 721         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
 722         eieio();
 723
 724         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 725                 udelay(10);
 726
 727         while (!cpu_online(cpu))
 728                 cpu_relax();
 729         return 0;
 730 }
 731
 732 static int __init setup_possible_cpus(char *s)
 733 {
 734         int pcpus, cpu;
 735
 736         pcpus = simple_strtoul(s, NULL, 0);
 737         cpu_possible_map = cpumask_of_cpu(0);
 738         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 739                 cpu_set(cpu, cpu_possible_map);
 740         return 0;
 741 }
 742 early_param("possible_cpus", setup_possible_cpus);
 743
 744 #ifdef CONFIG_HOTPLUG_CPU
 745
 746 int __cpu_disable(void)
 747 {
 748         struct ec_creg_mask_parms cr_parms;
 749         int cpu = smp_processor_id();
 750
 751         cpu_clear(cpu, cpu_online_map);
 752
 753         /* Disable pfault pseudo page faults on this cpu. */
 754         pfault_fini();
 755
 756         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 757         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 758
 759         /* disable all external interrupts */
 760         cr_parms.orvals[0] = 0;
 761         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 762                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 763         /* disable all I/O interrupts */
 764         cr_parms.orvals[6] = 0;
 765         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 766                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 767         /* disable most machine checks */
 768         cr_parms.orvals[14] = 0;
 769         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 770                                  1 << 25 | 1 << 24);
 771
 772         smp_ctl_bit_callback(&cr_parms);
 773
 774         return 0;
 775 }
 776
 777 void __cpu_die(unsigned int cpu)
 778 {
 779         /* Wait until target cpu is down */
 780         while (!smp_cpu_not_running(cpu))
 781                 cpu_relax();
 782         smp_free_lowcore(cpu);
 783         printk(KERN_INFO "Processor %d spun down\n", cpu);
 784 }
 785
 786 void cpu_die(void)
 787 {
 788         idle_task_exit();
 789         signal_processor(smp_processor_id(), sigp_stop);
 790         BUG();
 791         for (;;);
 792 }
 793
 794 #endif /* CONFIG_HOTPLUG_CPU */
 795
 796 void __init smp_prepare_cpus(unsigned int max_cpus)
 797 {
 798 #ifndef CONFIG_64BIT
 799         unsigned long save_area = 0;
 800 #endif
 801         unsigned long async_stack, panic_stack;
 802         struct _lowcore *lowcore;
 803         unsigned int cpu;
 804         int lc_order;
 805
 806         smp_detect_cpus();
 807
 808         /* request the 0x1201 emergency signal external interrupt */
 809         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 810                 panic("Couldn't request external interrupt 0x1201");
 811         print_cpu_info(&S390_lowcore.cpu_data);
 812
 813         /* Reallocate current lowcore, but keep its contents. */
 814         lc_order = sizeof(long) == 8 ? 1 : 0;
 815         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 816         panic_stack = __get_free_page(GFP_KERNEL);
 817         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 818 #ifndef CONFIG_64BIT
 819         if (MACHINE_HAS_IEEE)
 820                 save_area = get_zeroed_page(GFP_KERNEL);
 821 #endif
 822         local_irq_disable();
 823         local_mcck_disable();
 824         lowcore_ptr[smp_processor_id()] = lowcore;
 825         *lowcore = S390_lowcore;
 826         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 827         lowcore->async_stack = async_stack + ASYNC_SIZE;
 828 #ifndef CONFIG_64BIT
 829         if (MACHINE_HAS_IEEE)
 830                 lowcore->extended_save_area_addr = (u32) save_area;
 831 #endif
 832         set_prefix((u32)(unsigned long) lowcore);
 833         local_mcck_enable();
 834         local_irq_enable();
 835         for_each_possible_cpu(cpu)
 836                 if (cpu != smp_processor_id())
 837                         smp_create_idle(cpu);
 838 }
 839
 840 void __init smp_prepare_boot_cpu(void)
 841 {
 842         BUG_ON(smp_processor_id() != 0);
 843
 844         current_thread_info()->cpu = 0;
 845         cpu_set(0, cpu_present_map);
 846         cpu_set(0, cpu_online_map);
 847         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 848         current_set[0] = current;
 849         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 850         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 851         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 852 }
 853
 854 void __init smp_cpus_done(unsigned int max_cpus)
 855 {
 856 }
 857
 858 /*
 859  * the frequency of the profiling timer can be changed
 860  * by writing a multiplier value into /proc/profile.
 861  *
 862  * usually you want to run this on all CPUs ;)
 863  */
 864 int setup_profiling_timer(unsigned int multiplier)
 865 {
 866         return 0;
 867 }
 868
 869 #ifdef CONFIG_HOTPLUG_CPU
 870 static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
 871 {
 872         ssize_t count;
 873
 874         mutex_lock(&smp_cpu_state_mutex);
 875         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 876         mutex_unlock(&smp_cpu_state_mutex);
 877         return count;
 878 }
 879
 880 static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
 881                                    size_t count)
 882 {
 883         int cpu = dev->id;
 884         int val, rc;
 885         char delim;
 886
 887         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 888                 return -EINVAL;
 889         if (val != 0 && val != 1)
 890                 return -EINVAL;
 891
 892         mutex_lock(&smp_cpu_state_mutex);
 893         get_online_cpus();
 894         rc = -EBUSY;
 895         if (cpu_online(cpu))
 896                 goto out;
 897         rc = 0;
 898         switch (val) {
 899         case 0:
 900                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 901                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 902                         if (!rc) {
 903                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 904                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 905                         }
 906                 }
 907                 break;
 908         case 1:
 909                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 910                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 911                         if (!rc) {
 912                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 913                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 914                         }
 915                 }
 916                 break;
 917         default:
 918                 break;
 919         }
 920 out:
 921         put_online_cpus();
 922         mutex_unlock(&smp_cpu_state_mutex);
 923         return rc ? rc : count;
 924 }
 925 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 926 #endif /* CONFIG_HOTPLUG_CPU */
 927
 928 static ssize_t cpu_polarization_show(struct sys_device *dev, 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, char *buf)
 957 {
 958         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 959 }
 960 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 961
 962
 963 static struct attribute *cpu_common_attrs[] = {
 964 #ifdef CONFIG_HOTPLUG_CPU
 965         &attr_configure.attr,
 966 #endif
 967         &attr_address.attr,
 968         &attr_polarization.attr,
 969         NULL,
 970 };
 971
 972 static struct attribute_group cpu_common_attr_group = {
 973         .attrs = cpu_common_attrs,
 974 };
 975
 976 static ssize_t show_capability(struct sys_device *dev, char *buf)
 977 {
 978         unsigned int capability;
 979         int rc;
 980
 981         rc = get_cpu_capability(&capability);
 982         if (rc)
 983                 return rc;
 984         return sprintf(buf, "%u\n", capability);
 985 }
 986 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 987
 988 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
 989 {
 990         struct s390_idle_data *idle;
 991         unsigned long long idle_count;
 992
 993         idle = &per_cpu(s390_idle, dev->id);
 994         spin_lock_irq(&idle->lock);
 995         idle_count = idle->idle_count;
 996         spin_unlock_irq(&idle->lock);
 997         return sprintf(buf, "%llu\n", idle_count);
 998 }
 999 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
1000
1001 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
1002 {
1003         struct s390_idle_data *idle;
1004         unsigned long long new_time;
1005
1006         idle = &per_cpu(s390_idle, dev->id);
1007         spin_lock_irq(&idle->lock);
1008         if (idle->in_idle) {
1009                 new_time = get_clock();
1010                 idle->idle_time += new_time - idle->idle_enter;
1011                 idle->idle_enter = new_time;
1012         }
1013         new_time = idle->idle_time;
1014         spin_unlock_irq(&idle->lock);
1015         return sprintf(buf, "%llu\n", new_time >> 12);
1016 }
1017 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1018
1019 static struct attribute *cpu_online_attrs[] = {
1020         &attr_capability.attr,
1021         &attr_idle_count.attr,
1022         &attr_idle_time_us.attr,
1023         NULL,
1024 };
1025
1026 static struct attribute_group cpu_online_attr_group = {
1027         .attrs = cpu_online_attrs,
1028 };
1029
1030 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1031                                     unsigned long action, void *hcpu)
1032 {
1033         unsigned int cpu = (unsigned int)(long)hcpu;
1034         struct cpu *c = &per_cpu(cpu_devices, cpu);
1035         struct sys_device *s = &c->sysdev;
1036         struct s390_idle_data *idle;
1037
1038         switch (action) {
1039         case CPU_ONLINE:
1040         case CPU_ONLINE_FROZEN:
1041                 idle = &per_cpu(s390_idle, cpu);
1042                 spin_lock_irq(&idle->lock);
1043                 idle->idle_enter = 0;
1044                 idle->idle_time = 0;
1045                 idle->idle_count = 0;
1046                 spin_unlock_irq(&idle->lock);
1047                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
1048                         return NOTIFY_BAD;
1049                 break;
1050         case CPU_DEAD:
1051         case CPU_DEAD_FROZEN:
1052                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1053                 break;
1054         }
1055         return NOTIFY_OK;
1056 }
1057
1058 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1059         .notifier_call = smp_cpu_notify,
1060 };
1061
1062 static int __devinit smp_add_present_cpu(int cpu)
1063 {
1064         struct cpu *c = &per_cpu(cpu_devices, cpu);
1065         struct sys_device *s = &c->sysdev;
1066         int rc;
1067
1068         c->hotpluggable = 1;
1069         rc = register_cpu(c, cpu);
1070         if (rc)
1071                 goto out;
1072         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1073         if (rc)
1074                 goto out_cpu;
1075         if (!cpu_online(cpu))
1076                 goto out;
1077         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1078         if (!rc)
1079                 return 0;
1080         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1081 out_cpu:
1082 #ifdef CONFIG_HOTPLUG_CPU
1083         unregister_cpu(c);
1084 #endif
1085 out:
1086         return rc;
1087 }
1088
1089 #ifdef CONFIG_HOTPLUG_CPU
1090 static ssize_t __ref rescan_store(struct sys_device *dev,
1091                                   const char *buf, size_t count)
1092 {
1093         cpumask_t newcpus;
1094         int cpu;
1095         int rc;
1096
1097         mutex_lock(&smp_cpu_state_mutex);
1098         get_online_cpus();
1099         newcpus = cpu_present_map;
1100         rc = smp_rescan_cpus();
1101         if (rc)
1102                 goto out;
1103         cpus_andnot(newcpus, cpu_present_map, newcpus);
1104         for_each_cpu_mask(cpu, newcpus) {
1105                 rc = smp_add_present_cpu(cpu);
1106                 if (rc)
1107                         cpu_clear(cpu, cpu_present_map);
1108         }
1109         rc = 0;
1110 out:
1111         put_online_cpus();
1112         mutex_unlock(&smp_cpu_state_mutex);
1113         if (!cpus_empty(newcpus))
1114                 topology_schedule_update();
1115         return rc ? rc : count;
1116 }
1117 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1118 #endif /* CONFIG_HOTPLUG_CPU */
1119
1120 static ssize_t dispatching_show(struct sys_device *dev, char *buf)
1121 {
1122         ssize_t count;
1123
1124         mutex_lock(&smp_cpu_state_mutex);
1125         count = sprintf(buf, "%d\n", cpu_management);
1126         mutex_unlock(&smp_cpu_state_mutex);
1127         return count;
1128 }
1129
1130 static ssize_t dispatching_store(struct sys_device *dev, const char *buf,
1131                                  size_t count)
1132 {
1133         int val, rc;
1134         char delim;
1135
1136         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1137                 return -EINVAL;
1138         if (val != 0 && val != 1)
1139                 return -EINVAL;
1140         rc = 0;
1141         mutex_lock(&smp_cpu_state_mutex);
1142         get_online_cpus();
1143         if (cpu_management == val)
1144                 goto out;
1145         rc = topology_set_cpu_management(val);
1146         if (!rc)
1147                 cpu_management = val;
1148 out:
1149         put_online_cpus();
1150         mutex_unlock(&smp_cpu_state_mutex);
1151         return rc ? rc : count;
1152 }
1153 static SYSDEV_ATTR(dispatching, 0644, dispatching_show, dispatching_store);
1154
1155 static int __init topology_init(void)
1156 {
1157         int cpu;
1158         int rc;
1159
1160         register_cpu_notifier(&smp_cpu_nb);
1161
1162 #ifdef CONFIG_HOTPLUG_CPU
1163         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1164                                &attr_rescan.attr);
1165         if (rc)
1166                 return rc;
1167 #endif
1168         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1169                                &attr_dispatching.attr);
1170         if (rc)
1171                 return rc;
1172         for_each_present_cpu(cpu) {
1173                 rc = smp_add_present_cpu(cpu);
1174                 if (rc)
1175                         return rc;
1176         }
1177         return 0;
1178 }
1179 subsys_initcall(topology_init);