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 <asm/cpu.h>
  53 #include "entry.h"
  54
  55 /* logical cpu to cpu address */
  56 int __cpu_logical_map[NR_CPUS];
  57
  58 static struct task_struct *current_set[NR_CPUS];
  59
  60 static u8 smp_cpu_type;
  61 static int smp_use_sigp_detection;
  62
  63 enum s390_cpu_state {
  64         CPU_STATE_STANDBY,
  65         CPU_STATE_CONFIGURED,
  66 };
  67
  68 DEFINE_MUTEX(smp_cpu_state_mutex);
  69 int smp_cpu_polarization[NR_CPUS];
  70 static int smp_cpu_state[NR_CPUS];
  71 static int cpu_management;
  72
  73 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  74
  75 static void smp_ext_bitcall(int, ec_bit_sig);
  76
  77 static int cpu_stopped(int cpu)
  78 {
  79         __u32 status;
  80
  81         switch (signal_processor_ps(&status, 0, cpu, sigp_sense)) {
  82         case sigp_status_stored:
  83                 /* Check for stopped and check stop state */
  84                 if (status & 0x50)
  85                         return 1;
  86                 break;
  87         default:
  88                 break;
  89         }
  90         return 0;
  91 }
  92
  93 void smp_send_stop(void)
  94 {
  95         int cpu, rc;
  96
  97         /* Disable all interrupts/machine checks */
  98         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  99         trace_hardirqs_off();
 100
 101         /* stop all processors */
 102         for_each_online_cpu(cpu) {
 103                 if (cpu == smp_processor_id())
 104                         continue;
 105                 do {
 106                         rc = signal_processor(cpu, sigp_stop);
 107                 } while (rc == sigp_busy);
 108
 109                 while (!cpu_stopped(cpu))
 110                         cpu_relax();
 111         }
 112 }
 113
 114 /*
 115  * This is the main routine where commands issued by other
 116  * cpus are handled.
 117  */
 118
 119 static void do_ext_call_interrupt(__u16 code)
 120 {
 121         unsigned long bits;
 122
 123         /*
 124          * handle bit signal external calls
 125          *
 126          * For the ec_schedule signal we have to do nothing. All the work
 127          * is done automatically when we return from the interrupt.
 128          */
 129         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 130
 131         if (test_bit(ec_call_function, &bits))
 132                 generic_smp_call_function_interrupt();
 133
 134         if (test_bit(ec_call_function_single, &bits))
 135                 generic_smp_call_function_single_interrupt();
 136 }
 137
 138 /*
 139  * Send an external call sigp to another cpu and return without waiting
 140  * for its completion.
 141  */
 142 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 143 {
 144         /*
 145          * Set signaling bit in lowcore of target cpu and kick it
 146          */
 147         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 148         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 149                 udelay(10);
 150 }
 151
 152 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 153 {
 154         int cpu;
 155
 156         for_each_cpu(cpu, mask)
 157                 smp_ext_bitcall(cpu, ec_call_function);
 158 }
 159
 160 void arch_send_call_function_single_ipi(int cpu)
 161 {
 162         smp_ext_bitcall(cpu, ec_call_function_single);
 163 }
 164
 165 #ifndef CONFIG_64BIT
 166 /*
 167  * this function sends a 'purge tlb' signal to another CPU.
 168  */
 169 static void smp_ptlb_callback(void *info)
 170 {
 171         __tlb_flush_local();
 172 }
 173
 174 void smp_ptlb_all(void)
 175 {
 176         on_each_cpu(smp_ptlb_callback, NULL, 1);
 177 }
 178 EXPORT_SYMBOL(smp_ptlb_all);
 179 #endif /* ! CONFIG_64BIT */
 180
 181 /*
 182  * this function sends a 'reschedule' IPI to another CPU.
 183  * it goes straight through and wastes no time serializing
 184  * anything. Worst case is that we lose a reschedule ...
 185  */
 186 void smp_send_reschedule(int cpu)
 187 {
 188         smp_ext_bitcall(cpu, ec_schedule);
 189 }
 190
 191 /*
 192  * parameter area for the set/clear control bit callbacks
 193  */
 194 struct ec_creg_mask_parms {
 195         unsigned long orvals[16];
 196         unsigned long andvals[16];
 197 };
 198
 199 /*
 200  * callback for setting/clearing control bits
 201  */
 202 static void smp_ctl_bit_callback(void *info)
 203 {
 204         struct ec_creg_mask_parms *pp = info;
 205         unsigned long cregs[16];
 206         int i;
 207
 208         __ctl_store(cregs, 0, 15);
 209         for (i = 0; i <= 15; i++)
 210                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 211         __ctl_load(cregs, 0, 15);
 212 }
 213
 214 /*
 215  * Set a bit in a control register of all cpus
 216  */
 217 void smp_ctl_set_bit(int cr, int bit)
 218 {
 219         struct ec_creg_mask_parms parms;
 220
 221         memset(&parms.orvals, 0, sizeof(parms.orvals));
 222         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 223         parms.orvals[cr] = 1 << bit;
 224         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 225 }
 226 EXPORT_SYMBOL(smp_ctl_set_bit);
 227
 228 /*
 229  * Clear a bit in a control register of all cpus
 230  */
 231 void smp_ctl_clear_bit(int cr, int bit)
 232 {
 233         struct ec_creg_mask_parms parms;
 234
 235         memset(&parms.orvals, 0, sizeof(parms.orvals));
 236         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 237         parms.andvals[cr] = ~(1L << bit);
 238         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 239 }
 240 EXPORT_SYMBOL(smp_ctl_clear_bit);
 241
 242 /*
 243  * In early ipl state a temp. logically cpu number is needed, so the sigp
 244  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 245  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 246  */
 247 #define CPU_INIT_NO     1
 248
 249 #ifdef CONFIG_ZFCPDUMP
 250
 251 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 252 {
 253         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 254                 return;
 255         if (cpu >= NR_CPUS) {
 256                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 257                            "the dump\n", cpu, NR_CPUS - 1);
 258                 return;
 259         }
 260         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 261         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 262         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 263                sigp_busy)
 264                 cpu_relax();
 265         memcpy(zfcpdump_save_areas[cpu],
 266                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 267                sizeof(struct save_area));
 268 }
 269
 270 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 271 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 272
 273 #else
 274
 275 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 276
 277 #endif /* CONFIG_ZFCPDUMP */
 278
 279 static int cpu_known(int cpu_id)
 280 {
 281         int cpu;
 282
 283         for_each_present_cpu(cpu) {
 284                 if (__cpu_logical_map[cpu] == cpu_id)
 285                         return 1;
 286         }
 287         return 0;
 288 }
 289
 290 static int smp_rescan_cpus_sigp(cpumask_t avail)
 291 {
 292         int cpu_id, logical_cpu;
 293
 294         logical_cpu = cpumask_first(&avail);
 295         if (logical_cpu >= nr_cpu_ids)
 296                 return 0;
 297         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 298                 if (cpu_known(cpu_id))
 299                         continue;
 300                 __cpu_logical_map[logical_cpu] = cpu_id;
 301                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 302                 if (!cpu_stopped(logical_cpu))
 303                         continue;
 304                 cpu_set(logical_cpu, cpu_present_map);
 305                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 306                 logical_cpu = cpumask_next(logical_cpu, &avail);
 307                 if (logical_cpu >= nr_cpu_ids)
 308                         break;
 309         }
 310         return 0;
 311 }
 312
 313 static int smp_rescan_cpus_sclp(cpumask_t avail)
 314 {
 315         struct sclp_cpu_info *info;
 316         int cpu_id, logical_cpu, cpu;
 317         int rc;
 318
 319         logical_cpu = cpumask_first(&avail);
 320         if (logical_cpu >= nr_cpu_ids)
 321                 return 0;
 322         info = kmalloc(sizeof(*info), GFP_KERNEL);
 323         if (!info)
 324                 return -ENOMEM;
 325         rc = sclp_get_cpu_info(info);
 326         if (rc)
 327                 goto out;
 328         for (cpu = 0; cpu < info->combined; cpu++) {
 329                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 330                         continue;
 331                 cpu_id = info->cpu[cpu].address;
 332                 if (cpu_known(cpu_id))
 333                         continue;
 334                 __cpu_logical_map[logical_cpu] = cpu_id;
 335                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 336                 cpu_set(logical_cpu, cpu_present_map);
 337                 if (cpu >= info->configured)
 338                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 339                 else
 340                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 341                 logical_cpu = cpumask_next(logical_cpu, &avail);
 342                 if (logical_cpu >= nr_cpu_ids)
 343                         break;
 344         }
 345 out:
 346         kfree(info);
 347         return rc;
 348 }
 349
 350 static int __smp_rescan_cpus(void)
 351 {
 352         cpumask_t avail;
 353
 354         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 355         if (smp_use_sigp_detection)
 356                 return smp_rescan_cpus_sigp(avail);
 357         else
 358                 return smp_rescan_cpus_sclp(avail);
 359 }
 360
 361 static void __init smp_detect_cpus(void)
 362 {
 363         unsigned int cpu, c_cpus, s_cpus;
 364         struct sclp_cpu_info *info;
 365         u16 boot_cpu_addr, cpu_addr;
 366
 367         c_cpus = 1;
 368         s_cpus = 0;
 369         boot_cpu_addr = __cpu_logical_map[0];
 370         info = kmalloc(sizeof(*info), GFP_KERNEL);
 371         if (!info)
 372                 panic("smp_detect_cpus failed to allocate memory\n");
 373         /* Use sigp detection algorithm if sclp doesn't work. */
 374         if (sclp_get_cpu_info(info)) {
 375                 smp_use_sigp_detection = 1;
 376                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 377                         if (cpu == boot_cpu_addr)
 378                                 continue;
 379                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 380                         if (!cpu_stopped(CPU_INIT_NO))
 381                                 continue;
 382                         smp_get_save_area(c_cpus, cpu);
 383                         c_cpus++;
 384                 }
 385                 goto out;
 386         }
 387
 388         if (info->has_cpu_type) {
 389                 for (cpu = 0; cpu < info->combined; cpu++) {
 390                         if (info->cpu[cpu].address == boot_cpu_addr) {
 391                                 smp_cpu_type = info->cpu[cpu].type;
 392                                 break;
 393                         }
 394                 }
 395         }
 396
 397         for (cpu = 0; cpu < info->combined; cpu++) {
 398                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 399                         continue;
 400                 cpu_addr = info->cpu[cpu].address;
 401                 if (cpu_addr == boot_cpu_addr)
 402                         continue;
 403                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 404                 if (!cpu_stopped(CPU_INIT_NO)) {
 405                         s_cpus++;
 406                         continue;
 407                 }
 408                 smp_get_save_area(c_cpus, cpu_addr);
 409                 c_cpus++;
 410         }
 411 out:
 412         kfree(info);
 413         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 414         get_online_cpus();
 415         __smp_rescan_cpus();
 416         put_online_cpus();
 417 }
 418
 419 /*
 420  *      Activate a secondary processor.
 421  */
 422 int __cpuinit start_secondary(void *cpuvoid)
 423 {
 424         /* Setup the cpu */
 425         cpu_init();
 426         preempt_disable();
 427         /* Enable TOD clock interrupts on the secondary cpu. */
 428         init_cpu_timer();
 429         /* Enable cpu timer interrupts on the secondary cpu. */
 430         init_cpu_vtimer();
 431         /* Enable pfault pseudo page faults on this cpu. */
 432         pfault_init();
 433
 434         /* call cpu notifiers */
 435         notify_cpu_starting(smp_processor_id());
 436         /* Mark this cpu as online */
 437         ipi_call_lock();
 438         cpu_set(smp_processor_id(), cpu_online_map);
 439         ipi_call_unlock();
 440         /* Switch on interrupts */
 441         local_irq_enable();
 442         /* Print info about this processor */
 443         print_cpu_info();
 444         /* cpu_idle will call schedule for us */
 445         cpu_idle();
 446         return 0;
 447 }
 448
 449 static void __init smp_create_idle(unsigned int cpu)
 450 {
 451         struct task_struct *p;
 452
 453         /*
 454          *  don't care about the psw and regs settings since we'll never
 455          *  reschedule the forked task.
 456          */
 457         p = fork_idle(cpu);
 458         if (IS_ERR(p))
 459                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 460         current_set[cpu] = p;
 461 }
 462
 463 static int __cpuinit smp_alloc_lowcore(int cpu)
 464 {
 465         unsigned long async_stack, panic_stack;
 466         struct _lowcore *lowcore;
 467
 468         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 469         if (!lowcore)
 470                 return -ENOMEM;
 471         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 472         panic_stack = __get_free_page(GFP_KERNEL);
 473         if (!panic_stack || !async_stack)
 474                 goto out;
 475         memcpy(lowcore, &S390_lowcore, 512);
 476         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 477         lowcore->async_stack = async_stack + ASYNC_SIZE;
 478         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 479
 480 #ifndef CONFIG_64BIT
 481         if (MACHINE_HAS_IEEE) {
 482                 unsigned long save_area;
 483
 484                 save_area = get_zeroed_page(GFP_KERNEL);
 485                 if (!save_area)
 486                         goto out;
 487                 lowcore->extended_save_area_addr = (u32) save_area;
 488         }
 489 #else
 490         if (vdso_alloc_per_cpu(cpu, lowcore))
 491                 goto out;
 492 #endif
 493         lowcore_ptr[cpu] = lowcore;
 494         return 0;
 495
 496 out:
 497         free_page(panic_stack);
 498         free_pages(async_stack, ASYNC_ORDER);
 499         free_pages((unsigned long) lowcore, LC_ORDER);
 500         return -ENOMEM;
 501 }
 502
 503 static void smp_free_lowcore(int cpu)
 504 {
 505         struct _lowcore *lowcore;
 506
 507         lowcore = lowcore_ptr[cpu];
 508 #ifndef CONFIG_64BIT
 509         if (MACHINE_HAS_IEEE)
 510                 free_page((unsigned long) lowcore->extended_save_area_addr);
 511 #else
 512         vdso_free_per_cpu(cpu, lowcore);
 513 #endif
 514         free_page(lowcore->panic_stack - PAGE_SIZE);
 515         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 516         free_pages((unsigned long) lowcore, LC_ORDER);
 517         lowcore_ptr[cpu] = NULL;
 518 }
 519
 520 /* Upping and downing of CPUs */
 521 int __cpuinit __cpu_up(unsigned int cpu)
 522 {
 523         struct task_struct *idle;
 524         struct _lowcore *cpu_lowcore;
 525         struct stack_frame *sf;
 526         sigp_ccode ccode;
 527         u32 lowcore;
 528
 529         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 530                 return -EIO;
 531         if (smp_alloc_lowcore(cpu))
 532                 return -ENOMEM;
 533         do {
 534                 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
 535                 if (ccode == sigp_busy)
 536                         udelay(10);
 537                 if (ccode == sigp_not_operational)
 538                         goto err_out;
 539         } while (ccode == sigp_busy);
 540
 541         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 542         while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 543                 udelay(10);
 544
 545         idle = current_set[cpu];
 546         cpu_lowcore = lowcore_ptr[cpu];
 547         cpu_lowcore->kernel_stack = (unsigned long)
 548                 task_stack_page(idle) + THREAD_SIZE;
 549         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 550         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 551                                      - sizeof(struct pt_regs)
 552                                      - sizeof(struct stack_frame));
 553         memset(sf, 0, sizeof(struct stack_frame));
 554         sf->gprs[9] = (unsigned long) sf;
 555         cpu_lowcore->save_area[15] = (unsigned long) sf;
 556         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 557         asm volatile(
 558                 "       stam    0,15,0(%0)"
 559                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 560         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 561         cpu_lowcore->current_task = (unsigned long) idle;
 562         cpu_lowcore->cpu_nr = cpu;
 563         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 564         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 565         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 566         eieio();
 567
 568         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 569                 udelay(10);
 570
 571         while (!cpu_online(cpu))
 572                 cpu_relax();
 573         return 0;
 574
 575 err_out:
 576         smp_free_lowcore(cpu);
 577         return -EIO;
 578 }
 579
 580 static int __init setup_possible_cpus(char *s)
 581 {
 582         int pcpus, cpu;
 583
 584         pcpus = simple_strtoul(s, NULL, 0);
 585         init_cpu_possible(cpumask_of(0));
 586         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 587                 set_cpu_possible(cpu, true);
 588         return 0;
 589 }
 590 early_param("possible_cpus", setup_possible_cpus);
 591
 592 #ifdef CONFIG_HOTPLUG_CPU
 593
 594 int __cpu_disable(void)
 595 {
 596         struct ec_creg_mask_parms cr_parms;
 597         int cpu = smp_processor_id();
 598
 599         cpu_clear(cpu, cpu_online_map);
 600
 601         /* Disable pfault pseudo page faults on this cpu. */
 602         pfault_fini();
 603
 604         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 605         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 606
 607         /* disable all external interrupts */
 608         cr_parms.orvals[0] = 0;
 609         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 610                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 611         /* disable all I/O interrupts */
 612         cr_parms.orvals[6] = 0;
 613         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 614                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 615         /* disable most machine checks */
 616         cr_parms.orvals[14] = 0;
 617         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 618                                  1 << 25 | 1 << 24);
 619
 620         smp_ctl_bit_callback(&cr_parms);
 621
 622         return 0;
 623 }
 624
 625 void __cpu_die(unsigned int cpu)
 626 {
 627         /* Wait until target cpu is down */
 628         while (!cpu_stopped(cpu))
 629                 cpu_relax();
 630         while (signal_processor_p(0, cpu, sigp_set_prefix) == sigp_busy)
 631                 udelay(10);
 632         smp_free_lowcore(cpu);
 633         pr_info("Processor %d stopped\n", cpu);
 634 }
 635
 636 void cpu_die(void)
 637 {
 638         idle_task_exit();
 639         while (signal_processor(smp_processor_id(), sigp_stop) == sigp_busy)
 640                 cpu_relax();
 641         for (;;);
 642 }
 643
 644 #endif /* CONFIG_HOTPLUG_CPU */
 645
 646 void __init smp_prepare_cpus(unsigned int max_cpus)
 647 {
 648 #ifndef CONFIG_64BIT
 649         unsigned long save_area = 0;
 650 #endif
 651         unsigned long async_stack, panic_stack;
 652         struct _lowcore *lowcore;
 653         unsigned int cpu;
 654
 655         smp_detect_cpus();
 656
 657         /* request the 0x1201 emergency signal external interrupt */
 658         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 659                 panic("Couldn't request external interrupt 0x1201");
 660         print_cpu_info();
 661
 662         /* Reallocate current lowcore, but keep its contents. */
 663         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 664         panic_stack = __get_free_page(GFP_KERNEL);
 665         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 666         BUG_ON(!lowcore || !panic_stack || !async_stack);
 667 #ifndef CONFIG_64BIT
 668         if (MACHINE_HAS_IEEE)
 669                 save_area = get_zeroed_page(GFP_KERNEL);
 670 #endif
 671         local_irq_disable();
 672         local_mcck_disable();
 673         lowcore_ptr[smp_processor_id()] = lowcore;
 674         *lowcore = S390_lowcore;
 675         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 676         lowcore->async_stack = async_stack + ASYNC_SIZE;
 677 #ifndef CONFIG_64BIT
 678         if (MACHINE_HAS_IEEE)
 679                 lowcore->extended_save_area_addr = (u32) save_area;
 680 #endif
 681         set_prefix((u32)(unsigned long) lowcore);
 682         local_mcck_enable();
 683         local_irq_enable();
 684 #ifdef CONFIG_64BIT
 685         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 686                 BUG();
 687 #endif
 688         for_each_possible_cpu(cpu)
 689                 if (cpu != smp_processor_id())
 690                         smp_create_idle(cpu);
 691 }
 692
 693 void __init smp_prepare_boot_cpu(void)
 694 {
 695         BUG_ON(smp_processor_id() != 0);
 696
 697         current_thread_info()->cpu = 0;
 698         cpu_set(0, cpu_present_map);
 699         cpu_set(0, cpu_online_map);
 700         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 701         current_set[0] = current;
 702         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 703         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 704 }
 705
 706 void __init smp_cpus_done(unsigned int max_cpus)
 707 {
 708 }
 709
 710 void __init smp_setup_processor_id(void)
 711 {
 712         S390_lowcore.cpu_nr = 0;
 713         __cpu_logical_map[0] = stap();
 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         unsigned int sequence;
 857
 858         idle = &per_cpu(s390_idle, dev->id);
 859 repeat:
 860         sequence = idle->sequence;
 861         smp_rmb();
 862         if (sequence & 1)
 863                 goto repeat;
 864         idle_count = idle->idle_count;
 865         if (idle->idle_enter)
 866                 idle_count++;
 867         smp_rmb();
 868         if (idle->sequence != sequence)
 869                 goto repeat;
 870         return sprintf(buf, "%llu\n", idle_count);
 871 }
 872 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 873
 874 static ssize_t show_idle_time(struct sys_device *dev,
 875                                 struct sysdev_attribute *attr, char *buf)
 876 {
 877         struct s390_idle_data *idle;
 878         unsigned long long now, idle_time, idle_enter;
 879         unsigned int sequence;
 880
 881         idle = &per_cpu(s390_idle, dev->id);
 882         now = get_clock();
 883 repeat:
 884         sequence = idle->sequence;
 885         smp_rmb();
 886         if (sequence & 1)
 887                 goto repeat;
 888         idle_time = idle->idle_time;
 889         idle_enter = idle->idle_enter;
 890         if (idle_enter != 0ULL && idle_enter < now)
 891                 idle_time += now - idle_enter;
 892         smp_rmb();
 893         if (idle->sequence != sequence)
 894                 goto repeat;
 895         return sprintf(buf, "%llu\n", idle_time >> 12);
 896 }
 897 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 898
 899 static struct attribute *cpu_online_attrs[] = {
 900         &attr_capability.attr,
 901         &attr_idle_count.attr,
 902         &attr_idle_time_us.attr,
 903         NULL,
 904 };
 905
 906 static struct attribute_group cpu_online_attr_group = {
 907         .attrs = cpu_online_attrs,
 908 };
 909
 910 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 911                                     unsigned long action, void *hcpu)
 912 {
 913         unsigned int cpu = (unsigned int)(long)hcpu;
 914         struct cpu *c = &per_cpu(cpu_devices, cpu);
 915         struct sys_device *s = &c->sysdev;
 916         struct s390_idle_data *idle;
 917
 918         switch (action) {
 919         case CPU_ONLINE:
 920         case CPU_ONLINE_FROZEN:
 921                 idle = &per_cpu(s390_idle, cpu);
 922                 memset(idle, 0, sizeof(struct s390_idle_data));
 923                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 924                         return NOTIFY_BAD;
 925                 break;
 926         case CPU_DEAD:
 927         case CPU_DEAD_FROZEN:
 928                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 929                 break;
 930         }
 931         return NOTIFY_OK;
 932 }
 933
 934 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 935         .notifier_call = smp_cpu_notify,
 936 };
 937
 938 static int __devinit smp_add_present_cpu(int cpu)
 939 {
 940         struct cpu *c = &per_cpu(cpu_devices, cpu);
 941         struct sys_device *s = &c->sysdev;
 942         int rc;
 943
 944         c->hotpluggable = 1;
 945         rc = register_cpu(c, cpu);
 946         if (rc)
 947                 goto out;
 948         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 949         if (rc)
 950                 goto out_cpu;
 951         if (!cpu_online(cpu))
 952                 goto out;
 953         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 954         if (!rc)
 955                 return 0;
 956         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 957 out_cpu:
 958 #ifdef CONFIG_HOTPLUG_CPU
 959         unregister_cpu(c);
 960 #endif
 961 out:
 962         return rc;
 963 }
 964
 965 #ifdef CONFIG_HOTPLUG_CPU
 966
 967 int __ref smp_rescan_cpus(void)
 968 {
 969         cpumask_t newcpus;
 970         int cpu;
 971         int rc;
 972
 973         get_online_cpus();
 974         mutex_lock(&smp_cpu_state_mutex);
 975         newcpus = cpu_present_map;
 976         rc = __smp_rescan_cpus();
 977         if (rc)
 978                 goto out;
 979         cpus_andnot(newcpus, cpu_present_map, newcpus);
 980         for_each_cpu_mask(cpu, newcpus) {
 981                 rc = smp_add_present_cpu(cpu);
 982                 if (rc)
 983                         cpu_clear(cpu, cpu_present_map);
 984         }
 985         rc = 0;
 986 out:
 987         mutex_unlock(&smp_cpu_state_mutex);
 988         put_online_cpus();
 989         if (!cpus_empty(newcpus))
 990                 topology_schedule_update();
 991         return rc;
 992 }
 993
 994 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
 995                                   size_t count)
 996 {
 997         int rc;
 998
 999         rc = smp_rescan_cpus();
1000         return rc ? rc : count;
1001 }
1002 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1003 #endif /* CONFIG_HOTPLUG_CPU */
1004
1005 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1006 {
1007         ssize_t count;
1008
1009         mutex_lock(&smp_cpu_state_mutex);
1010         count = sprintf(buf, "%d\n", cpu_management);
1011         mutex_unlock(&smp_cpu_state_mutex);
1012         return count;
1013 }
1014
1015 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1016                                  size_t count)
1017 {
1018         int val, rc;
1019         char delim;
1020
1021         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1022                 return -EINVAL;
1023         if (val != 0 && val != 1)
1024                 return -EINVAL;
1025         rc = 0;
1026         get_online_cpus();
1027         mutex_lock(&smp_cpu_state_mutex);
1028         if (cpu_management == val)
1029                 goto out;
1030         rc = topology_set_cpu_management(val);
1031         if (!rc)
1032                 cpu_management = val;
1033 out:
1034         mutex_unlock(&smp_cpu_state_mutex);
1035         put_online_cpus();
1036         return rc ? rc : count;
1037 }
1038 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1039                          dispatching_store);
1040
1041 static int __init topology_init(void)
1042 {
1043         int cpu;
1044         int rc;
1045
1046         register_cpu_notifier(&smp_cpu_nb);
1047
1048 #ifdef CONFIG_HOTPLUG_CPU
1049         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1050         if (rc)
1051                 return rc;
1052 #endif
1053         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1054         if (rc)
1055                 return rc;
1056         for_each_present_cpu(cpu) {
1057                 rc = smp_add_present_cpu(cpu);
1058                 if (rc)
1059                         return rc;
1060         }
1061         return 0;
1062 }
1063 subsys_initcall(topology_init);