arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999, 2009
   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 #define KMSG_COMPONENT "cpu"
  24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  25
  26 #include <linux/module.h>
  27 #include <linux/init.h>
  28 #include <linux/mm.h>
  29 #include <linux/err.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/kernel_stat.h>
  32 #include <linux/delay.h>
  33 #include <linux/cache.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/irqflags.h>
  36 #include <linux/cpu.h>
  37 #include <linux/timex.h>
  38 #include <linux/bootmem.h>
  39 #include <asm/ipl.h>
  40 #include <asm/setup.h>
  41 #include <asm/sigp.h>
  42 #include <asm/pgalloc.h>
  43 #include <asm/irq.h>
  44 #include <asm/s390_ext.h>
  45 #include <asm/cpcmd.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/timer.h>
  48 #include <asm/lowcore.h>
  49 #include <asm/sclp.h>
  50 #include <asm/cputime.h>
  51 #include <asm/vdso.h>
  52 #include "entry.h"
  53
  54 static struct task_struct *current_set[NR_CPUS];
  55
  56 static u8 smp_cpu_type;
  57 static int smp_use_sigp_detection;
  58
  59 enum s390_cpu_state {
  60         CPU_STATE_STANDBY,
  61         CPU_STATE_CONFIGURED,
  62 };
  63
  64 DEFINE_MUTEX(smp_cpu_state_mutex);
  65 int smp_cpu_polarization[NR_CPUS];
  66 static int smp_cpu_state[NR_CPUS];
  67 static int cpu_management;
  68
  69 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  70
  71 static void smp_ext_bitcall(int, ec_bit_sig);
  72
  73 void smp_send_stop(void)
  74 {
  75         int cpu, rc;
  76
  77         /* Disable all interrupts/machine checks */
  78         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  79         trace_hardirqs_off();
  80
  81         /* stop all processors */
  82         for_each_online_cpu(cpu) {
  83                 if (cpu == smp_processor_id())
  84                         continue;
  85                 do {
  86                         rc = signal_processor(cpu, sigp_stop);
  87                 } while (rc == sigp_busy);
  88
  89                 while (!smp_cpu_not_running(cpu))
  90                         cpu_relax();
  91         }
  92 }
  93
  94 /*
  95  * This is the main routine where commands issued by other
  96  * cpus are handled.
  97  */
  98
  99 static void do_ext_call_interrupt(__u16 code)
 100 {
 101         unsigned long bits;
 102
 103         /*
 104          * handle bit signal external calls
 105          *
 106          * For the ec_schedule signal we have to do nothing. All the work
 107          * is done automatically when we return from the interrupt.
 108          */
 109         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 110
 111         if (test_bit(ec_call_function, &bits))
 112                 generic_smp_call_function_interrupt();
 113
 114         if (test_bit(ec_call_function_single, &bits))
 115                 generic_smp_call_function_single_interrupt();
 116 }
 117
 118 /*
 119  * Send an external call sigp to another cpu and return without waiting
 120  * for its completion.
 121  */
 122 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 123 {
 124         /*
 125          * Set signaling bit in lowcore of target cpu and kick it
 126          */
 127         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 128         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 129                 udelay(10);
 130 }
 131
 132 void arch_send_call_function_ipi(cpumask_t mask)
 133 {
 134         int cpu;
 135
 136         for_each_cpu_mask(cpu, mask)
 137                 smp_ext_bitcall(cpu, ec_call_function);
 138 }
 139
 140 void arch_send_call_function_single_ipi(int cpu)
 141 {
 142         smp_ext_bitcall(cpu, ec_call_function_single);
 143 }
 144
 145 #ifndef CONFIG_64BIT
 146 /*
 147  * this function sends a 'purge tlb' signal to another CPU.
 148  */
 149 static void smp_ptlb_callback(void *info)
 150 {
 151         __tlb_flush_local();
 152 }
 153
 154 void smp_ptlb_all(void)
 155 {
 156         on_each_cpu(smp_ptlb_callback, NULL, 1);
 157 }
 158 EXPORT_SYMBOL(smp_ptlb_all);
 159 #endif /* ! CONFIG_64BIT */
 160
 161 /*
 162  * this function sends a 'reschedule' IPI to another CPU.
 163  * it goes straight through and wastes no time serializing
 164  * anything. Worst case is that we lose a reschedule ...
 165  */
 166 void smp_send_reschedule(int cpu)
 167 {
 168         smp_ext_bitcall(cpu, ec_schedule);
 169 }
 170
 171 /*
 172  * parameter area for the set/clear control bit callbacks
 173  */
 174 struct ec_creg_mask_parms {
 175         unsigned long orvals[16];
 176         unsigned long andvals[16];
 177 };
 178
 179 /*
 180  * callback for setting/clearing control bits
 181  */
 182 static void smp_ctl_bit_callback(void *info)
 183 {
 184         struct ec_creg_mask_parms *pp = info;
 185         unsigned long cregs[16];
 186         int i;
 187
 188         __ctl_store(cregs, 0, 15);
 189         for (i = 0; i <= 15; i++)
 190                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 191         __ctl_load(cregs, 0, 15);
 192 }
 193
 194 /*
 195  * Set a bit in a control register of all cpus
 196  */
 197 void smp_ctl_set_bit(int cr, int bit)
 198 {
 199         struct ec_creg_mask_parms parms;
 200
 201         memset(&parms.orvals, 0, sizeof(parms.orvals));
 202         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 203         parms.orvals[cr] = 1 << bit;
 204         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 205 }
 206 EXPORT_SYMBOL(smp_ctl_set_bit);
 207
 208 /*
 209  * Clear a bit in a control register of all cpus
 210  */
 211 void smp_ctl_clear_bit(int cr, int bit)
 212 {
 213         struct ec_creg_mask_parms parms;
 214
 215         memset(&parms.orvals, 0, sizeof(parms.orvals));
 216         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 217         parms.andvals[cr] = ~(1L << bit);
 218         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 219 }
 220 EXPORT_SYMBOL(smp_ctl_clear_bit);
 221
 222 /*
 223  * In early ipl state a temp. logically cpu number is needed, so the sigp
 224  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 225  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 226  */
 227 #define CPU_INIT_NO     1
 228
 229 #ifdef CONFIG_ZFCPDUMP
 230
 231 /*
 232  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 233  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 234  * save its prefix registers, since they get lost, when switching from 31 bit
 235  * to 64 bit.
 236  */
 237 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 238         __attribute__((__section__(".data")));
 239
 240 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 241 {
 242         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 243                 return;
 244         if (cpu >= NR_CPUS) {
 245                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 246                            "the dump\n", cpu, NR_CPUS - 1);
 247                 return;
 248         }
 249         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 250         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 251         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 252                sigp_busy)
 253                 cpu_relax();
 254         memcpy(zfcpdump_save_areas[cpu],
 255                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 256                SAVE_AREA_SIZE);
 257 #ifdef CONFIG_64BIT
 258         /* copy original prefix register */
 259         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 260 #endif
 261 }
 262
 263 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 264 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 265
 266 #else
 267
 268 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 269
 270 #endif /* CONFIG_ZFCPDUMP */
 271
 272 static int cpu_stopped(int cpu)
 273 {
 274         __u32 status;
 275
 276         /* Check for stopped state */
 277         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 278             sigp_status_stored) {
 279                 if (status & 0x40)
 280                         return 1;
 281         }
 282         return 0;
 283 }
 284
 285 static int cpu_known(int cpu_id)
 286 {
 287         int cpu;
 288
 289         for_each_present_cpu(cpu) {
 290                 if (__cpu_logical_map[cpu] == cpu_id)
 291                         return 1;
 292         }
 293         return 0;
 294 }
 295
 296 static int smp_rescan_cpus_sigp(cpumask_t avail)
 297 {
 298         int cpu_id, logical_cpu;
 299
 300         logical_cpu = cpumask_first(&avail);
 301         if (logical_cpu >= nr_cpu_ids)
 302                 return 0;
 303         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 304                 if (cpu_known(cpu_id))
 305                         continue;
 306                 __cpu_logical_map[logical_cpu] = cpu_id;
 307                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 308                 if (!cpu_stopped(logical_cpu))
 309                         continue;
 310                 cpu_set(logical_cpu, cpu_present_map);
 311                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 312                 logical_cpu = cpumask_next(logical_cpu, &avail);
 313                 if (logical_cpu >= nr_cpu_ids)
 314                         break;
 315         }
 316         return 0;
 317 }
 318
 319 static int smp_rescan_cpus_sclp(cpumask_t avail)
 320 {
 321         struct sclp_cpu_info *info;
 322         int cpu_id, logical_cpu, cpu;
 323         int rc;
 324
 325         logical_cpu = cpumask_first(&avail);
 326         if (logical_cpu >= nr_cpu_ids)
 327                 return 0;
 328         info = kmalloc(sizeof(*info), GFP_KERNEL);
 329         if (!info)
 330                 return -ENOMEM;
 331         rc = sclp_get_cpu_info(info);
 332         if (rc)
 333                 goto out;
 334         for (cpu = 0; cpu < info->combined; cpu++) {
 335                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 336                         continue;
 337                 cpu_id = info->cpu[cpu].address;
 338                 if (cpu_known(cpu_id))
 339                         continue;
 340                 __cpu_logical_map[logical_cpu] = cpu_id;
 341                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 342                 cpu_set(logical_cpu, cpu_present_map);
 343                 if (cpu >= info->configured)
 344                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 345                 else
 346                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 347                 logical_cpu = cpumask_next(logical_cpu, &avail);
 348                 if (logical_cpu >= nr_cpu_ids)
 349                         break;
 350         }
 351 out:
 352         kfree(info);
 353         return rc;
 354 }
 355
 356 static int __smp_rescan_cpus(void)
 357 {
 358         cpumask_t avail;
 359
 360         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 361         if (smp_use_sigp_detection)
 362                 return smp_rescan_cpus_sigp(avail);
 363         else
 364                 return smp_rescan_cpus_sclp(avail);
 365 }
 366
 367 static void __init smp_detect_cpus(void)
 368 {
 369         unsigned int cpu, c_cpus, s_cpus;
 370         struct sclp_cpu_info *info;
 371         u16 boot_cpu_addr, cpu_addr;
 372
 373         c_cpus = 1;
 374         s_cpus = 0;
 375         boot_cpu_addr = __cpu_logical_map[0];
 376         info = kmalloc(sizeof(*info), GFP_KERNEL);
 377         if (!info)
 378                 panic("smp_detect_cpus failed to allocate memory\n");
 379         /* Use sigp detection algorithm if sclp doesn't work. */
 380         if (sclp_get_cpu_info(info)) {
 381                 smp_use_sigp_detection = 1;
 382                 for (cpu = 0; cpu <= 65535; cpu++) {
 383                         if (cpu == boot_cpu_addr)
 384                                 continue;
 385                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 386                         if (!cpu_stopped(CPU_INIT_NO))
 387                                 continue;
 388                         smp_get_save_area(c_cpus, cpu);
 389                         c_cpus++;
 390                 }
 391                 goto out;
 392         }
 393
 394         if (info->has_cpu_type) {
 395                 for (cpu = 0; cpu < info->combined; cpu++) {
 396                         if (info->cpu[cpu].address == boot_cpu_addr) {
 397                                 smp_cpu_type = info->cpu[cpu].type;
 398                                 break;
 399                         }
 400                 }
 401         }
 402
 403         for (cpu = 0; cpu < info->combined; cpu++) {
 404                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 405                         continue;
 406                 cpu_addr = info->cpu[cpu].address;
 407                 if (cpu_addr == boot_cpu_addr)
 408                         continue;
 409                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 410                 if (!cpu_stopped(CPU_INIT_NO)) {
 411                         s_cpus++;
 412                         continue;
 413                 }
 414                 smp_get_save_area(c_cpus, cpu_addr);
 415                 c_cpus++;
 416         }
 417 out:
 418         kfree(info);
 419         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 420         get_online_cpus();
 421         __smp_rescan_cpus();
 422         put_online_cpus();
 423 }
 424
 425 /*
 426  *      Activate a secondary processor.
 427  */
 428 int __cpuinit start_secondary(void *cpuvoid)
 429 {
 430         /* Setup the cpu */
 431         cpu_init();
 432         preempt_disable();
 433         /* Enable TOD clock interrupts on the secondary cpu. */
 434         init_cpu_timer();
 435         /* Enable cpu timer interrupts on the secondary cpu. */
 436         init_cpu_vtimer();
 437         /* Enable pfault pseudo page faults on this cpu. */
 438         pfault_init();
 439
 440         /* call cpu notifiers */
 441         notify_cpu_starting(smp_processor_id());
 442         /* Mark this cpu as online */
 443         ipi_call_lock();
 444         cpu_set(smp_processor_id(), cpu_online_map);
 445         ipi_call_unlock();
 446         /* Switch on interrupts */
 447         local_irq_enable();
 448         /* Print info about this processor */
 449         print_cpu_info();
 450         /* cpu_idle will call schedule for us */
 451         cpu_idle();
 452         return 0;
 453 }
 454
 455 static void __init smp_create_idle(unsigned int cpu)
 456 {
 457         struct task_struct *p;
 458
 459         /*
 460          *  don't care about the psw and regs settings since we'll never
 461          *  reschedule the forked task.
 462          */
 463         p = fork_idle(cpu);
 464         if (IS_ERR(p))
 465                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 466         current_set[cpu] = p;
 467 }
 468
 469 static int __cpuinit smp_alloc_lowcore(int cpu)
 470 {
 471         unsigned long async_stack, panic_stack;
 472         struct _lowcore *lowcore;
 473         int lc_order;
 474
 475         lc_order = sizeof(long) == 8 ? 1 : 0;
 476         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 477         if (!lowcore)
 478                 return -ENOMEM;
 479         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 480         panic_stack = __get_free_page(GFP_KERNEL);
 481         if (!panic_stack || !async_stack)
 482                 goto out;
 483         memcpy(lowcore, &S390_lowcore, 512);
 484         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 485         lowcore->async_stack = async_stack + ASYNC_SIZE;
 486         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 487
 488 #ifndef CONFIG_64BIT
 489         if (MACHINE_HAS_IEEE) {
 490                 unsigned long save_area;
 491
 492                 save_area = get_zeroed_page(GFP_KERNEL);
 493                 if (!save_area)
 494                         goto out;
 495                 lowcore->extended_save_area_addr = (u32) save_area;
 496         }
 497 #else
 498         if (vdso_alloc_per_cpu(cpu, lowcore))
 499                 goto out;
 500 #endif
 501         lowcore_ptr[cpu] = lowcore;
 502         return 0;
 503
 504 out:
 505         free_page(panic_stack);
 506         free_pages(async_stack, ASYNC_ORDER);
 507         free_pages((unsigned long) lowcore, lc_order);
 508         return -ENOMEM;
 509 }
 510
 511 static void smp_free_lowcore(int cpu)
 512 {
 513         struct _lowcore *lowcore;
 514         int lc_order;
 515
 516         lc_order = sizeof(long) == 8 ? 1 : 0;
 517         lowcore = lowcore_ptr[cpu];
 518 #ifndef CONFIG_64BIT
 519         if (MACHINE_HAS_IEEE)
 520                 free_page((unsigned long) lowcore->extended_save_area_addr);
 521 #else
 522         vdso_free_per_cpu(cpu, lowcore);
 523 #endif
 524         free_page(lowcore->panic_stack - PAGE_SIZE);
 525         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 526         free_pages((unsigned long) lowcore, lc_order);
 527         lowcore_ptr[cpu] = NULL;
 528 }
 529
 530 /* Upping and downing of CPUs */
 531 int __cpuinit __cpu_up(unsigned int cpu)
 532 {
 533         struct task_struct *idle;
 534         struct _lowcore *cpu_lowcore;
 535         struct stack_frame *sf;
 536         sigp_ccode ccode;
 537         u32 lowcore;
 538
 539         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 540                 return -EIO;
 541         if (smp_alloc_lowcore(cpu))
 542                 return -ENOMEM;
 543         do {
 544                 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
 545                 if (ccode == sigp_busy)
 546                         udelay(10);
 547                 if (ccode == sigp_not_operational)
 548                         goto err_out;
 549         } while (ccode == sigp_busy);
 550
 551         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 552         while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 553                 udelay(10);
 554
 555         idle = current_set[cpu];
 556         cpu_lowcore = lowcore_ptr[cpu];
 557         cpu_lowcore->kernel_stack = (unsigned long)
 558                 task_stack_page(idle) + THREAD_SIZE;
 559         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 560         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 561                                      - sizeof(struct pt_regs)
 562                                      - sizeof(struct stack_frame));
 563         memset(sf, 0, sizeof(struct stack_frame));
 564         sf->gprs[9] = (unsigned long) sf;
 565         cpu_lowcore->save_area[15] = (unsigned long) sf;
 566         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 567         asm volatile(
 568                 "       stam    0,15,0(%0)"
 569                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 570         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 571         cpu_lowcore->current_task = (unsigned long) idle;
 572         cpu_lowcore->cpu_nr = cpu;
 573         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 574         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 575         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 576         eieio();
 577
 578         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 579                 udelay(10);
 580
 581         while (!cpu_online(cpu))
 582                 cpu_relax();
 583         return 0;
 584
 585 err_out:
 586         smp_free_lowcore(cpu);
 587         return -EIO;
 588 }
 589
 590 static int __init setup_possible_cpus(char *s)
 591 {
 592         int pcpus, cpu;
 593
 594         pcpus = simple_strtoul(s, NULL, 0);
 595         init_cpu_possible(cpumask_of(0));
 596         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 597                 set_cpu_possible(cpu, true);
 598         return 0;
 599 }
 600 early_param("possible_cpus", setup_possible_cpus);
 601
 602 #ifdef CONFIG_HOTPLUG_CPU
 603
 604 int __cpu_disable(void)
 605 {
 606         struct ec_creg_mask_parms cr_parms;
 607         int cpu = smp_processor_id();
 608
 609         cpu_clear(cpu, cpu_online_map);
 610
 611         /* Disable pfault pseudo page faults on this cpu. */
 612         pfault_fini();
 613
 614         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 615         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 616
 617         /* disable all external interrupts */
 618         cr_parms.orvals[0] = 0;
 619         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 620                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 621         /* disable all I/O interrupts */
 622         cr_parms.orvals[6] = 0;
 623         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 624                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 625         /* disable most machine checks */
 626         cr_parms.orvals[14] = 0;
 627         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 628                                  1 << 25 | 1 << 24);
 629
 630         smp_ctl_bit_callback(&cr_parms);
 631
 632         return 0;
 633 }
 634
 635 void __cpu_die(unsigned int cpu)
 636 {
 637         /* Wait until target cpu is down */
 638         while (!smp_cpu_not_running(cpu))
 639                 cpu_relax();
 640         smp_free_lowcore(cpu);
 641         pr_info("Processor %d stopped\n", cpu);
 642 }
 643
 644 void cpu_die(void)
 645 {
 646         idle_task_exit();
 647         signal_processor(smp_processor_id(), sigp_stop);
 648         BUG();
 649         for (;;);
 650 }
 651
 652 #endif /* CONFIG_HOTPLUG_CPU */
 653
 654 void __init smp_prepare_cpus(unsigned int max_cpus)
 655 {
 656 #ifndef CONFIG_64BIT
 657         unsigned long save_area = 0;
 658 #endif
 659         unsigned long async_stack, panic_stack;
 660         struct _lowcore *lowcore;
 661         unsigned int cpu;
 662         int lc_order;
 663
 664         smp_detect_cpus();
 665
 666         /* request the 0x1201 emergency signal external interrupt */
 667         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 668                 panic("Couldn't request external interrupt 0x1201");
 669         print_cpu_info();
 670
 671         /* Reallocate current lowcore, but keep its contents. */
 672         lc_order = sizeof(long) == 8 ? 1 : 0;
 673         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 674         panic_stack = __get_free_page(GFP_KERNEL);
 675         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 676         BUG_ON(!lowcore || !panic_stack || !async_stack);
 677 #ifndef CONFIG_64BIT
 678         if (MACHINE_HAS_IEEE)
 679                 save_area = get_zeroed_page(GFP_KERNEL);
 680 #endif
 681         local_irq_disable();
 682         local_mcck_disable();
 683         lowcore_ptr[smp_processor_id()] = lowcore;
 684         *lowcore = S390_lowcore;
 685         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 686         lowcore->async_stack = async_stack + ASYNC_SIZE;
 687 #ifndef CONFIG_64BIT
 688         if (MACHINE_HAS_IEEE)
 689                 lowcore->extended_save_area_addr = (u32) save_area;
 690 #else
 691         if (vdso_alloc_per_cpu(smp_processor_id(), lowcore))
 692                 BUG();
 693 #endif
 694         set_prefix((u32)(unsigned long) lowcore);
 695         local_mcck_enable();
 696         local_irq_enable();
 697         for_each_possible_cpu(cpu)
 698                 if (cpu != smp_processor_id())
 699                         smp_create_idle(cpu);
 700 }
 701
 702 void __init smp_prepare_boot_cpu(void)
 703 {
 704         BUG_ON(smp_processor_id() != 0);
 705
 706         current_thread_info()->cpu = 0;
 707         cpu_set(0, cpu_present_map);
 708         cpu_set(0, cpu_online_map);
 709         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 710         current_set[0] = current;
 711         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 712         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 713 }
 714
 715 void __init smp_cpus_done(unsigned int max_cpus)
 716 {
 717 }
 718
 719 /*
 720  * the frequency of the profiling timer can be changed
 721  * by writing a multiplier value into /proc/profile.
 722  *
 723  * usually you want to run this on all CPUs ;)
 724  */
 725 int setup_profiling_timer(unsigned int multiplier)
 726 {
 727         return 0;
 728 }
 729
 730 #ifdef CONFIG_HOTPLUG_CPU
 731 static ssize_t cpu_configure_show(struct sys_device *dev,
 732                                 struct sysdev_attribute *attr, char *buf)
 733 {
 734         ssize_t count;
 735
 736         mutex_lock(&smp_cpu_state_mutex);
 737         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 738         mutex_unlock(&smp_cpu_state_mutex);
 739         return count;
 740 }
 741
 742 static ssize_t cpu_configure_store(struct sys_device *dev,
 743                                   struct sysdev_attribute *attr,
 744                                   const char *buf, size_t count)
 745 {
 746         int cpu = dev->id;
 747         int val, rc;
 748         char delim;
 749
 750         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 751                 return -EINVAL;
 752         if (val != 0 && val != 1)
 753                 return -EINVAL;
 754
 755         get_online_cpus();
 756         mutex_lock(&smp_cpu_state_mutex);
 757         rc = -EBUSY;
 758         if (cpu_online(cpu))
 759                 goto out;
 760         rc = 0;
 761         switch (val) {
 762         case 0:
 763                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 764                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 765                         if (!rc) {
 766                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 767                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 768                         }
 769                 }
 770                 break;
 771         case 1:
 772                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 773                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 774                         if (!rc) {
 775                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 776                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 777                         }
 778                 }
 779                 break;
 780         default:
 781                 break;
 782         }
 783 out:
 784         mutex_unlock(&smp_cpu_state_mutex);
 785         put_online_cpus();
 786         return rc ? rc : count;
 787 }
 788 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 789 #endif /* CONFIG_HOTPLUG_CPU */
 790
 791 static ssize_t cpu_polarization_show(struct sys_device *dev,
 792                                      struct sysdev_attribute *attr, char *buf)
 793 {
 794         int cpu = dev->id;
 795         ssize_t count;
 796
 797         mutex_lock(&smp_cpu_state_mutex);
 798         switch (smp_cpu_polarization[cpu]) {
 799         case POLARIZATION_HRZ:
 800                 count = sprintf(buf, "horizontal\n");
 801                 break;
 802         case POLARIZATION_VL:
 803                 count = sprintf(buf, "vertical:low\n");
 804                 break;
 805         case POLARIZATION_VM:
 806                 count = sprintf(buf, "vertical:medium\n");
 807                 break;
 808         case POLARIZATION_VH:
 809                 count = sprintf(buf, "vertical:high\n");
 810                 break;
 811         default:
 812                 count = sprintf(buf, "unknown\n");
 813                 break;
 814         }
 815         mutex_unlock(&smp_cpu_state_mutex);
 816         return count;
 817 }
 818 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 819
 820 static ssize_t show_cpu_address(struct sys_device *dev,
 821                                 struct sysdev_attribute *attr, char *buf)
 822 {
 823         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 824 }
 825 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 826
 827
 828 static struct attribute *cpu_common_attrs[] = {
 829 #ifdef CONFIG_HOTPLUG_CPU
 830         &attr_configure.attr,
 831 #endif
 832         &attr_address.attr,
 833         &attr_polarization.attr,
 834         NULL,
 835 };
 836
 837 static struct attribute_group cpu_common_attr_group = {
 838         .attrs = cpu_common_attrs,
 839 };
 840
 841 static ssize_t show_capability(struct sys_device *dev,
 842                                 struct sysdev_attribute *attr, char *buf)
 843 {
 844         unsigned int capability;
 845         int rc;
 846
 847         rc = get_cpu_capability(&capability);
 848         if (rc)
 849                 return rc;
 850         return sprintf(buf, "%u\n", capability);
 851 }
 852 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 853
 854 static ssize_t show_idle_count(struct sys_device *dev,
 855                                 struct sysdev_attribute *attr, char *buf)
 856 {
 857         struct s390_idle_data *idle;
 858         unsigned long long idle_count;
 859
 860         idle = &per_cpu(s390_idle, dev->id);
 861         spin_lock(&idle->lock);
 862         idle_count = idle->idle_count;
 863         if (idle->idle_enter)
 864                 idle_count++;
 865         spin_unlock(&idle->lock);
 866         return sprintf(buf, "%llu\n", idle_count);
 867 }
 868 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 869
 870 static ssize_t show_idle_time(struct sys_device *dev,
 871                                 struct sysdev_attribute *attr, char *buf)
 872 {
 873         struct s390_idle_data *idle;
 874         unsigned long long now, idle_time, idle_enter;
 875
 876         idle = &per_cpu(s390_idle, dev->id);
 877         spin_lock(&idle->lock);
 878         now = get_clock();
 879         idle_time = idle->idle_time;
 880         idle_enter = idle->idle_enter;
 881         if (idle_enter != 0ULL && idle_enter < now)
 882                 idle_time += now - idle_enter;
 883         spin_unlock(&idle->lock);
 884         return sprintf(buf, "%llu\n", idle_time >> 12);
 885 }
 886 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 887
 888 static struct attribute *cpu_online_attrs[] = {
 889         &attr_capability.attr,
 890         &attr_idle_count.attr,
 891         &attr_idle_time_us.attr,
 892         NULL,
 893 };
 894
 895 static struct attribute_group cpu_online_attr_group = {
 896         .attrs = cpu_online_attrs,
 897 };
 898
 899 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 900                                     unsigned long action, void *hcpu)
 901 {
 902         unsigned int cpu = (unsigned int)(long)hcpu;
 903         struct cpu *c = &per_cpu(cpu_devices, cpu);
 904         struct sys_device *s = &c->sysdev;
 905         struct s390_idle_data *idle;
 906
 907         switch (action) {
 908         case CPU_ONLINE:
 909         case CPU_ONLINE_FROZEN:
 910                 idle = &per_cpu(s390_idle, cpu);
 911                 spin_lock_irq(&idle->lock);
 912                 idle->idle_enter = 0;
 913                 idle->idle_time = 0;
 914                 idle->idle_count = 0;
 915                 spin_unlock_irq(&idle->lock);
 916                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 917                         return NOTIFY_BAD;
 918                 break;
 919         case CPU_DEAD:
 920         case CPU_DEAD_FROZEN:
 921                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 922                 break;
 923         }
 924         return NOTIFY_OK;
 925 }
 926
 927 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 928         .notifier_call = smp_cpu_notify,
 929 };
 930
 931 static int __devinit smp_add_present_cpu(int cpu)
 932 {
 933         struct cpu *c = &per_cpu(cpu_devices, cpu);
 934         struct sys_device *s = &c->sysdev;
 935         int rc;
 936
 937         c->hotpluggable = 1;
 938         rc = register_cpu(c, cpu);
 939         if (rc)
 940                 goto out;
 941         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 942         if (rc)
 943                 goto out_cpu;
 944         if (!cpu_online(cpu))
 945                 goto out;
 946         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 947         if (!rc)
 948                 return 0;
 949         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 950 out_cpu:
 951 #ifdef CONFIG_HOTPLUG_CPU
 952         unregister_cpu(c);
 953 #endif
 954 out:
 955         return rc;
 956 }
 957
 958 #ifdef CONFIG_HOTPLUG_CPU
 959
 960 int __ref smp_rescan_cpus(void)
 961 {
 962         cpumask_t newcpus;
 963         int cpu;
 964         int rc;
 965
 966         get_online_cpus();
 967         mutex_lock(&smp_cpu_state_mutex);
 968         newcpus = cpu_present_map;
 969         rc = __smp_rescan_cpus();
 970         if (rc)
 971                 goto out;
 972         cpus_andnot(newcpus, cpu_present_map, newcpus);
 973         for_each_cpu_mask(cpu, newcpus) {
 974                 rc = smp_add_present_cpu(cpu);
 975                 if (rc)
 976                         cpu_clear(cpu, cpu_present_map);
 977         }
 978         rc = 0;
 979 out:
 980         mutex_unlock(&smp_cpu_state_mutex);
 981         put_online_cpus();
 982         if (!cpus_empty(newcpus))
 983                 topology_schedule_update();
 984         return rc;
 985 }
 986
 987 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
 988                                   size_t count)
 989 {
 990         int rc;
 991
 992         rc = smp_rescan_cpus();
 993         return rc ? rc : count;
 994 }
 995 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
 996 #endif /* CONFIG_HOTPLUG_CPU */
 997
 998 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
 999 {
1000         ssize_t count;
1001
1002         mutex_lock(&smp_cpu_state_mutex);
1003         count = sprintf(buf, "%d\n", cpu_management);
1004         mutex_unlock(&smp_cpu_state_mutex);
1005         return count;
1006 }
1007
1008 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1009                                  size_t count)
1010 {
1011         int val, rc;
1012         char delim;
1013
1014         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1015                 return -EINVAL;
1016         if (val != 0 && val != 1)
1017                 return -EINVAL;
1018         rc = 0;
1019         get_online_cpus();
1020         mutex_lock(&smp_cpu_state_mutex);
1021         if (cpu_management == val)
1022                 goto out;
1023         rc = topology_set_cpu_management(val);
1024         if (!rc)
1025                 cpu_management = val;
1026 out:
1027         mutex_unlock(&smp_cpu_state_mutex);
1028         put_online_cpus();
1029         return rc ? rc : count;
1030 }
1031 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1032                          dispatching_store);
1033
1034 /*
1035  * If the resume kernel runs on another cpu than the suspended kernel,
1036  * we have to switch the cpu IDs in the logical map.
1037  */
1038 void smp_switch_boot_cpu_in_resume(u32 resume_phys_cpu_id,
1039                                    struct _lowcore *suspend_lowcore)
1040 {
1041         int cpu, suspend_cpu_id, resume_cpu_id;
1042         u32 suspend_phys_cpu_id;
1043
1044         suspend_phys_cpu_id = __cpu_logical_map[suspend_lowcore->cpu_nr];
1045         suspend_cpu_id = suspend_lowcore->cpu_nr;
1046
1047         for_each_present_cpu(cpu) {
1048                 if (__cpu_logical_map[cpu] == resume_phys_cpu_id) {
1049                         resume_cpu_id = cpu;
1050                         goto found;
1051                 }
1052         }
1053         panic("Could not find resume cpu in logical map.\n");
1054
1055 found:
1056         printk("Resume  cpu ID: %i/%i\n", resume_phys_cpu_id, resume_cpu_id);
1057         printk("Suspend cpu ID: %i/%i\n", suspend_phys_cpu_id, suspend_cpu_id);
1058
1059         __cpu_logical_map[resume_cpu_id] = suspend_phys_cpu_id;
1060         __cpu_logical_map[suspend_cpu_id] = resume_phys_cpu_id;
1061
1062         lowcore_ptr[suspend_cpu_id]->cpu_addr = resume_phys_cpu_id;
1063 }
1064
1065 u32 smp_get_phys_cpu_id(void)
1066 {
1067         return __cpu_logical_map[smp_processor_id()];
1068 }
1069
1070 static int __init topology_init(void)
1071 {
1072         int cpu;
1073         int rc;
1074
1075         register_cpu_notifier(&smp_cpu_nb);
1076
1077 #ifdef CONFIG_HOTPLUG_CPU
1078         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1079         if (rc)
1080                 return rc;
1081 #endif
1082         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1083         if (rc)
1084                 return rc;
1085         for_each_present_cpu(cpu) {
1086                 rc = smp_add_present_cpu(cpu);
1087                 if (rc)
1088                         return rc;
1089         }
1090         return 0;
1091 }
1092 subsys_initcall(topology_init);