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