git.oblomov.eu Git - linux-2.6/blob - arch/s390/kernel/smp.c

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