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
 229 smp_call_function_mask(cpumask_t mask,
 230                         void (*func)(void *), void *info,
 231                         int wait)
 232 {
 233         preempt_disable();
 234         __smp_call_function_map(func, info, 0, wait, mask);
 235         preempt_enable();
 236         return 0;
 237 }
 238 EXPORT_SYMBOL(smp_call_function_mask);
 239
 240 void smp_send_stop(void)
 241 {
 242         int cpu, rc;
 243
 244         /* Disable all interrupts/machine checks */
 245         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 246
 247         /* write magic number to zero page (absolute 0) */
 248         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
 249
 250         /* stop all processors */
 251         for_each_online_cpu(cpu) {
 252                 if (cpu == smp_processor_id())
 253                         continue;
 254                 do {
 255                         rc = signal_processor(cpu, sigp_stop);
 256                 } while (rc == sigp_busy);
 257
 258                 while (!smp_cpu_not_running(cpu))
 259                         cpu_relax();
 260         }
 261 }
 262
 263 /*
 264  * This is the main routine where commands issued by other
 265  * cpus are handled.
 266  */
 267
 268 static void do_ext_call_interrupt(__u16 code)
 269 {
 270         unsigned long bits;
 271
 272         /*
 273          * handle bit signal external calls
 274          *
 275          * For the ec_schedule signal we have to do nothing. All the work
 276          * is done automatically when we return from the interrupt.
 277          */
 278         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 279
 280         if (test_bit(ec_call_function, &bits))
 281                 do_call_function();
 282 }
 283
 284 /*
 285  * Send an external call sigp to another cpu and return without waiting
 286  * for its completion.
 287  */
 288 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 289 {
 290         /*
 291          * Set signaling bit in lowcore of target cpu and kick it
 292          */
 293         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 294         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 295                 udelay(10);
 296 }
 297
 298 #ifndef CONFIG_64BIT
 299 /*
 300  * this function sends a 'purge tlb' signal to another CPU.
 301  */
 302 void smp_ptlb_callback(void *info)
 303 {
 304         __tlb_flush_local();
 305 }
 306
 307 void smp_ptlb_all(void)
 308 {
 309         on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
 310 }
 311 EXPORT_SYMBOL(smp_ptlb_all);
 312 #endif /* ! CONFIG_64BIT */
 313
 314 /*
 315  * this function sends a 'reschedule' IPI to another CPU.
 316  * it goes straight through and wastes no time serializing
 317  * anything. Worst case is that we lose a reschedule ...
 318  */
 319 void smp_send_reschedule(int cpu)
 320 {
 321         smp_ext_bitcall(cpu, ec_schedule);
 322 }
 323
 324 /*
 325  * parameter area for the set/clear control bit callbacks
 326  */
 327 struct ec_creg_mask_parms {
 328         unsigned long orvals[16];
 329         unsigned long andvals[16];
 330 };
 331
 332 /*
 333  * callback for setting/clearing control bits
 334  */
 335 static void smp_ctl_bit_callback(void *info)
 336 {
 337         struct ec_creg_mask_parms *pp = info;
 338         unsigned long cregs[16];
 339         int i;
 340
 341         __ctl_store(cregs, 0, 15);
 342         for (i = 0; i <= 15; i++)
 343                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 344         __ctl_load(cregs, 0, 15);
 345 }
 346
 347 /*
 348  * Set a bit in a control register of all cpus
 349  */
 350 void smp_ctl_set_bit(int cr, int bit)
 351 {
 352         struct ec_creg_mask_parms parms;
 353
 354         memset(&parms.orvals, 0, sizeof(parms.orvals));
 355         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 356         parms.orvals[cr] = 1 << bit;
 357         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 358 }
 359 EXPORT_SYMBOL(smp_ctl_set_bit);
 360
 361 /*
 362  * Clear a bit in a control register of all cpus
 363  */
 364 void smp_ctl_clear_bit(int cr, int bit)
 365 {
 366         struct ec_creg_mask_parms parms;
 367
 368         memset(&parms.orvals, 0, sizeof(parms.orvals));
 369         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 370         parms.andvals[cr] = ~(1L << bit);
 371         on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
 372 }
 373 EXPORT_SYMBOL(smp_ctl_clear_bit);
 374
 375 /*
 376  * In early ipl state a temp. logically cpu number is needed, so the sigp
 377  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 378  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 379  */
 380 #define CPU_INIT_NO     1
 381
 382 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 383
 384 /*
 385  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 386  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 387  * save its prefix registers, since they get lost, when switching from 31 bit
 388  * to 64 bit.
 389  */
 390 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 391         __attribute__((__section__(".data")));
 392
 393 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 394 {
 395         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 396                 return;
 397         if (cpu >= NR_CPUS) {
 398                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
 399                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
 400                 return;
 401         }
 402         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 403         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 404         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 405                sigp_busy)
 406                 cpu_relax();
 407         memcpy(zfcpdump_save_areas[cpu],
 408                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 409                SAVE_AREA_SIZE);
 410 #ifdef CONFIG_64BIT
 411         /* copy original prefix register */
 412         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 413 #endif
 414 }
 415
 416 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 417 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 418
 419 #else
 420
 421 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 422
 423 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 424
 425 static int cpu_stopped(int cpu)
 426 {
 427         __u32 status;
 428
 429         /* Check for stopped state */
 430         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 431             sigp_status_stored) {
 432                 if (status & 0x40)
 433                         return 1;
 434         }
 435         return 0;
 436 }
 437
 438 static int cpu_known(int cpu_id)
 439 {
 440         int cpu;
 441
 442         for_each_present_cpu(cpu) {
 443                 if (__cpu_logical_map[cpu] == cpu_id)
 444                         return 1;
 445         }
 446         return 0;
 447 }
 448
 449 static int smp_rescan_cpus_sigp(cpumask_t avail)
 450 {
 451         int cpu_id, logical_cpu;
 452
 453         logical_cpu = first_cpu(avail);
 454         if (logical_cpu == NR_CPUS)
 455                 return 0;
 456         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 457                 if (cpu_known(cpu_id))
 458                         continue;
 459                 __cpu_logical_map[logical_cpu] = cpu_id;
 460                 if (!cpu_stopped(logical_cpu))
 461                         continue;
 462                 cpu_set(logical_cpu, cpu_present_map);
 463                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 464                 logical_cpu = next_cpu(logical_cpu, avail);
 465                 if (logical_cpu == NR_CPUS)
 466                         break;
 467         }
 468         return 0;
 469 }
 470
 471 static int smp_rescan_cpus_sclp(cpumask_t avail)
 472 {
 473         struct sclp_cpu_info *info;
 474         int cpu_id, logical_cpu, cpu;
 475         int rc;
 476
 477         logical_cpu = first_cpu(avail);
 478         if (logical_cpu == NR_CPUS)
 479                 return 0;
 480         info = kmalloc(sizeof(*info), GFP_KERNEL);
 481         if (!info)
 482                 return -ENOMEM;
 483         rc = sclp_get_cpu_info(info);
 484         if (rc)
 485                 goto out;
 486         for (cpu = 0; cpu < info->combined; cpu++) {
 487                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 488                         continue;
 489                 cpu_id = info->cpu[cpu].address;
 490                 if (cpu_known(cpu_id))
 491                         continue;
 492                 __cpu_logical_map[logical_cpu] = cpu_id;
 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         if (!async_stack)
 631                 goto out_async_stack;
 632         panic_stack = __get_free_page(GFP_KERNEL);
 633         if (!panic_stack)
 634                 goto out_panic_stack;
 635
 636         *lowcore = S390_lowcore;
 637         lowcore->async_stack = async_stack + ASYNC_SIZE;
 638         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 639
 640 #ifndef CONFIG_64BIT
 641         if (MACHINE_HAS_IEEE) {
 642                 unsigned long save_area;
 643
 644                 save_area = get_zeroed_page(GFP_KERNEL);
 645                 if (!save_area)
 646                         goto out_save_area;
 647                 lowcore->extended_save_area_addr = (u32) save_area;
 648         }
 649 #endif
 650         lowcore_ptr[cpu] = lowcore;
 651         return 0;
 652
 653 #ifndef CONFIG_64BIT
 654 out_save_area:
 655         free_page(panic_stack);
 656 #endif
 657 out_panic_stack:
 658         free_pages(async_stack, ASYNC_ORDER);
 659 out_async_stack:
 660         free_pages((unsigned long) lowcore, lc_order);
 661         return -ENOMEM;
 662 }
 663
 664 #ifdef CONFIG_HOTPLUG_CPU
 665 static void smp_free_lowcore(int cpu)
 666 {
 667         struct _lowcore *lowcore;
 668         int lc_order;
 669
 670         lc_order = sizeof(long) == 8 ? 1 : 0;
 671         lowcore = lowcore_ptr[cpu];
 672 #ifndef CONFIG_64BIT
 673         if (MACHINE_HAS_IEEE)
 674                 free_page((unsigned long) lowcore->extended_save_area_addr);
 675 #endif
 676         free_page(lowcore->panic_stack - PAGE_SIZE);
 677         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 678         free_pages((unsigned long) lowcore, lc_order);
 679         lowcore_ptr[cpu] = NULL;
 680 }
 681 #endif /* CONFIG_HOTPLUG_CPU */
 682
 683 /* Upping and downing of CPUs */
 684 int __cpuinit __cpu_up(unsigned int cpu)
 685 {
 686         struct task_struct *idle;
 687         struct _lowcore *cpu_lowcore;
 688         struct stack_frame *sf;
 689         sigp_ccode ccode;
 690
 691         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 692                 return -EIO;
 693         if (smp_alloc_lowcore(cpu))
 694                 return -ENOMEM;
 695
 696         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 697                                    cpu, sigp_set_prefix);
 698         if (ccode) {
 699                 printk("sigp_set_prefix failed for cpu %d "
 700                        "with condition code %d\n",
 701                        (int) cpu, (int) ccode);
 702                 return -EIO;
 703         }
 704
 705         idle = current_set[cpu];
 706         cpu_lowcore = lowcore_ptr[cpu];
 707         cpu_lowcore->kernel_stack = (unsigned long)
 708                 task_stack_page(idle) + THREAD_SIZE;
 709         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 710         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 711                                      - sizeof(struct pt_regs)
 712                                      - sizeof(struct stack_frame));
 713         memset(sf, 0, sizeof(struct stack_frame));
 714         sf->gprs[9] = (unsigned long) sf;
 715         cpu_lowcore->save_area[15] = (unsigned long) sf;
 716         __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
 717         asm volatile(
 718                 "       stam    0,15,0(%0)"
 719                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 720         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 721         cpu_lowcore->current_task = (unsigned long) idle;
 722         cpu_lowcore->cpu_data.cpu_nr = cpu;
 723         cpu_lowcore->softirq_pending = 0;
 724         cpu_lowcore->ext_call_fast = 0;
 725         eieio();
 726
 727         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 728                 udelay(10);
 729
 730         while (!cpu_online(cpu))
 731                 cpu_relax();
 732         return 0;
 733 }
 734
 735 static int __init setup_possible_cpus(char *s)
 736 {
 737         int pcpus, cpu;
 738
 739         pcpus = simple_strtoul(s, NULL, 0);
 740         cpu_possible_map = cpumask_of_cpu(0);
 741         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 742                 cpu_set(cpu, cpu_possible_map);
 743         return 0;
 744 }
 745 early_param("possible_cpus", setup_possible_cpus);
 746
 747 #ifdef CONFIG_HOTPLUG_CPU
 748
 749 int __cpu_disable(void)
 750 {
 751         struct ec_creg_mask_parms cr_parms;
 752         int cpu = smp_processor_id();
 753
 754         cpu_clear(cpu, cpu_online_map);
 755
 756         /* Disable pfault pseudo page faults on this cpu. */
 757         pfault_fini();
 758
 759         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 760         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 761
 762         /* disable all external interrupts */
 763         cr_parms.orvals[0] = 0;
 764         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 765                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 766         /* disable all I/O interrupts */
 767         cr_parms.orvals[6] = 0;
 768         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 769                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 770         /* disable most machine checks */
 771         cr_parms.orvals[14] = 0;
 772         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 773                                  1 << 25 | 1 << 24);
 774
 775         smp_ctl_bit_callback(&cr_parms);
 776
 777         return 0;
 778 }
 779
 780 void __cpu_die(unsigned int cpu)
 781 {
 782         /* Wait until target cpu is down */
 783         while (!smp_cpu_not_running(cpu))
 784                 cpu_relax();
 785         smp_free_lowcore(cpu);
 786         printk(KERN_INFO "Processor %d spun down\n", cpu);
 787 }
 788
 789 void cpu_die(void)
 790 {
 791         idle_task_exit();
 792         signal_processor(smp_processor_id(), sigp_stop);
 793         BUG();
 794         for (;;);
 795 }
 796
 797 #endif /* CONFIG_HOTPLUG_CPU */
 798
 799 void __init smp_prepare_cpus(unsigned int max_cpus)
 800 {
 801         unsigned int cpu;
 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         memset(lowcore_ptr, 0, sizeof(lowcore_ptr));
 809         print_cpu_info(&S390_lowcore.cpu_data);
 810         smp_alloc_lowcore(smp_processor_id());
 811
 812 #ifndef CONFIG_64BIT
 813         if (MACHINE_HAS_IEEE)
 814                 ctl_set_bit(14, 29); /* enable extended save area */
 815 #endif
 816         set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
 817
 818         for_each_possible_cpu(cpu)
 819                 if (cpu != smp_processor_id())
 820                         smp_create_idle(cpu);
 821 }
 822
 823 void __init smp_prepare_boot_cpu(void)
 824 {
 825         BUG_ON(smp_processor_id() != 0);
 826
 827         current_thread_info()->cpu = 0;
 828         cpu_set(0, cpu_present_map);
 829         cpu_set(0, cpu_online_map);
 830         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 831         current_set[0] = current;
 832         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 833         spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 834 }
 835
 836 void __init smp_cpus_done(unsigned int max_cpus)
 837 {
 838 }
 839
 840 /*
 841  * the frequency of the profiling timer can be changed
 842  * by writing a multiplier value into /proc/profile.
 843  *
 844  * usually you want to run this on all CPUs ;)
 845  */
 846 int setup_profiling_timer(unsigned int multiplier)
 847 {
 848         return 0;
 849 }
 850
 851 #ifdef CONFIG_HOTPLUG_CPU
 852 static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
 853 {
 854         ssize_t count;
 855
 856         mutex_lock(&smp_cpu_state_mutex);
 857         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 858         mutex_unlock(&smp_cpu_state_mutex);
 859         return count;
 860 }
 861
 862 static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
 863                                    size_t count)
 864 {
 865         int cpu = dev->id;
 866         int val, rc;
 867         char delim;
 868
 869         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 870                 return -EINVAL;
 871         if (val != 0 && val != 1)
 872                 return -EINVAL;
 873
 874         mutex_lock(&smp_cpu_state_mutex);
 875         get_online_cpus();
 876         rc = -EBUSY;
 877         if (cpu_online(cpu))
 878                 goto out;
 879         rc = 0;
 880         switch (val) {
 881         case 0:
 882                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 883                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 884                         if (!rc)
 885                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 886                 }
 887                 break;
 888         case 1:
 889                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 890                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 891                         if (!rc)
 892                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 893                 }
 894                 break;
 895         default:
 896                 break;
 897         }
 898 out:
 899         put_online_cpus();
 900         mutex_unlock(&smp_cpu_state_mutex);
 901         return rc ? rc : count;
 902 }
 903 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 904 #endif /* CONFIG_HOTPLUG_CPU */
 905
 906 static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
 907 {
 908         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 909 }
 910 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 911
 912
 913 static struct attribute *cpu_common_attrs[] = {
 914 #ifdef CONFIG_HOTPLUG_CPU
 915         &attr_configure.attr,
 916 #endif
 917         &attr_address.attr,
 918         NULL,
 919 };
 920
 921 static struct attribute_group cpu_common_attr_group = {
 922         .attrs = cpu_common_attrs,
 923 };
 924
 925 static ssize_t show_capability(struct sys_device *dev, char *buf)
 926 {
 927         unsigned int capability;
 928         int rc;
 929
 930         rc = get_cpu_capability(&capability);
 931         if (rc)
 932                 return rc;
 933         return sprintf(buf, "%u\n", capability);
 934 }
 935 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 936
 937 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
 938 {
 939         struct s390_idle_data *idle;
 940         unsigned long long idle_count;
 941
 942         idle = &per_cpu(s390_idle, dev->id);
 943         spin_lock_irq(&idle->lock);
 944         idle_count = idle->idle_count;
 945         spin_unlock_irq(&idle->lock);
 946         return sprintf(buf, "%llu\n", idle_count);
 947 }
 948 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 949
 950 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
 951 {
 952         struct s390_idle_data *idle;
 953         unsigned long long new_time;
 954
 955         idle = &per_cpu(s390_idle, dev->id);
 956         spin_lock_irq(&idle->lock);
 957         if (idle->in_idle) {
 958                 new_time = get_clock();
 959                 idle->idle_time += new_time - idle->idle_enter;
 960                 idle->idle_enter = new_time;
 961         }
 962         new_time = idle->idle_time;
 963         spin_unlock_irq(&idle->lock);
 964         return sprintf(buf, "%llu\n", new_time >> 12);
 965 }
 966 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 967
 968 static struct attribute *cpu_online_attrs[] = {
 969         &attr_capability.attr,
 970         &attr_idle_count.attr,
 971         &attr_idle_time_us.attr,
 972         NULL,
 973 };
 974
 975 static struct attribute_group cpu_online_attr_group = {
 976         .attrs = cpu_online_attrs,
 977 };
 978
 979 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 980                                     unsigned long action, void *hcpu)
 981 {
 982         unsigned int cpu = (unsigned int)(long)hcpu;
 983         struct cpu *c = &per_cpu(cpu_devices, cpu);
 984         struct sys_device *s = &c->sysdev;
 985         struct s390_idle_data *idle;
 986
 987         switch (action) {
 988         case CPU_ONLINE:
 989         case CPU_ONLINE_FROZEN:
 990                 idle = &per_cpu(s390_idle, cpu);
 991                 spin_lock_irq(&idle->lock);
 992                 idle->idle_enter = 0;
 993                 idle->idle_time = 0;
 994                 idle->idle_count = 0;
 995                 spin_unlock_irq(&idle->lock);
 996                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 997                         return NOTIFY_BAD;
 998                 break;
 999         case CPU_DEAD:
1000         case CPU_DEAD_FROZEN:
1001                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1002                 break;
1003         }
1004         return NOTIFY_OK;
1005 }
1006
1007 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1008         .notifier_call = smp_cpu_notify,
1009 };
1010
1011 static int smp_add_present_cpu(int cpu)
1012 {
1013         struct cpu *c = &per_cpu(cpu_devices, cpu);
1014         struct sys_device *s = &c->sysdev;
1015         int rc;
1016
1017         c->hotpluggable = 1;
1018         rc = register_cpu(c, cpu);
1019         if (rc)
1020                 goto out;
1021         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1022         if (rc)
1023                 goto out_cpu;
1024         if (!cpu_online(cpu))
1025                 goto out;
1026         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1027         if (!rc)
1028                 return 0;
1029         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1030 out_cpu:
1031 #ifdef CONFIG_HOTPLUG_CPU
1032         unregister_cpu(c);
1033 #endif
1034 out:
1035         return rc;
1036 }
1037
1038 #ifdef CONFIG_HOTPLUG_CPU
1039 static ssize_t rescan_store(struct sys_device *dev, const char *buf,
1040                             size_t count)
1041 {
1042         cpumask_t newcpus;
1043         int cpu;
1044         int rc;
1045
1046         mutex_lock(&smp_cpu_state_mutex);
1047         get_online_cpus();
1048         newcpus = cpu_present_map;
1049         rc = smp_rescan_cpus();
1050         if (rc)
1051                 goto out;
1052         cpus_andnot(newcpus, cpu_present_map, newcpus);
1053         for_each_cpu_mask(cpu, newcpus) {
1054                 rc = smp_add_present_cpu(cpu);
1055                 if (rc)
1056                         cpu_clear(cpu, cpu_present_map);
1057         }
1058         rc = 0;
1059 out:
1060         put_online_cpus();
1061         mutex_unlock(&smp_cpu_state_mutex);
1062         return rc ? rc : count;
1063 }
1064 static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
1065 #endif /* CONFIG_HOTPLUG_CPU */
1066
1067 static int __init topology_init(void)
1068 {
1069         int cpu;
1070         int rc;
1071
1072         register_cpu_notifier(&smp_cpu_nb);
1073
1074 #ifdef CONFIG_HOTPLUG_CPU
1075         rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
1076                                &attr_rescan.attr);
1077         if (rc)
1078                 return rc;
1079 #endif
1080         for_each_present_cpu(cpu) {
1081                 rc = smp_add_present_cpu(cpu);
1082                 if (rc)
1083                         return rc;
1084         }
1085         return 0;
1086 }
1087 subsys_initcall(topology_init);