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 <linux/slab.h>
  40 #include <asm/asm-offsets.h>
  41 #include <asm/ipl.h>
  42 #include <asm/setup.h>
  43 #include <asm/sigp.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/irq.h>
  46 #include <asm/s390_ext.h>
  47 #include <asm/cpcmd.h>
  48 #include <asm/tlbflush.h>
  49 #include <asm/timer.h>
  50 #include <asm/lowcore.h>
  51 #include <asm/sclp.h>
  52 #include <asm/cputime.h>
  53 #include <asm/vdso.h>
  54 #include <asm/cpu.h>
  55 #include "entry.h"
  56
  57 /* logical cpu to cpu address */
  58 unsigned short __cpu_logical_map[NR_CPUS];
  59
  60 static struct task_struct *current_set[NR_CPUS];
  61
  62 static u8 smp_cpu_type;
  63 static int smp_use_sigp_detection;
  64
  65 enum s390_cpu_state {
  66         CPU_STATE_STANDBY,
  67         CPU_STATE_CONFIGURED,
  68 };
  69
  70 DEFINE_MUTEX(smp_cpu_state_mutex);
  71 int smp_cpu_polarization[NR_CPUS];
  72 static int smp_cpu_state[NR_CPUS];
  73 static int cpu_management;
  74
  75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  76
  77 static void smp_ext_bitcall(int, int);
  78
  79 static int raw_cpu_stopped(int cpu)
  80 {
  81         u32 status;
  82
  83         switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
  84         case sigp_status_stored:
  85                 /* Check for stopped and check stop state */
  86                 if (status & 0x50)
  87                         return 1;
  88                 break;
  89         default:
  90                 break;
  91         }
  92         return 0;
  93 }
  94
  95 static inline int cpu_stopped(int cpu)
  96 {
  97         return raw_cpu_stopped(cpu_logical_map(cpu));
  98 }
  99
 100 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
 101 {
 102         struct _lowcore *lc, *current_lc;
 103         struct stack_frame *sf;
 104         struct pt_regs *regs;
 105         unsigned long sp;
 106
 107         if (smp_processor_id() == 0)
 108                 func(data);
 109         __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
 110         /* Disable lowcore protection */
 111         __ctl_clear_bit(0, 28);
 112         current_lc = lowcore_ptr[smp_processor_id()];
 113         lc = lowcore_ptr[0];
 114         if (!lc)
 115                 lc = current_lc;
 116         lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 117         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
 118         if (!cpu_online(0))
 119                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
 120         while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
 121                 cpu_relax();
 122         sp = lc->panic_stack;
 123         sp -= sizeof(struct pt_regs);
 124         regs = (struct pt_regs *) sp;
 125         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
 126         regs->psw = lc->psw_save_area;
 127         sp -= STACK_FRAME_OVERHEAD;
 128         sf = (struct stack_frame *) sp;
 129         sf->back_chain = regs->gprs[15];
 130         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 131 }
 132
 133 void smp_send_stop(void)
 134 {
 135         int cpu, rc;
 136
 137         /* Disable all interrupts/machine checks */
 138         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 139         trace_hardirqs_off();
 140
 141         /* stop all processors */
 142         for_each_online_cpu(cpu) {
 143                 if (cpu == smp_processor_id())
 144                         continue;
 145                 do {
 146                         rc = sigp(cpu, sigp_stop);
 147                 } while (rc == sigp_busy);
 148
 149                 while (!cpu_stopped(cpu))
 150                         cpu_relax();
 151         }
 152 }
 153
 154 /*
 155  * This is the main routine where commands issued by other
 156  * cpus are handled.
 157  */
 158
 159 static void do_ext_call_interrupt(unsigned int ext_int_code,
 160                                   unsigned int param32, unsigned long param64)
 161 {
 162         unsigned long bits;
 163
 164         /*
 165          * handle bit signal external calls
 166          *
 167          * For the ec_schedule signal we have to do nothing. All the work
 168          * is done automatically when we return from the interrupt.
 169          */
 170         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 171
 172         if (test_bit(ec_call_function, &bits))
 173                 generic_smp_call_function_interrupt();
 174
 175         if (test_bit(ec_call_function_single, &bits))
 176                 generic_smp_call_function_single_interrupt();
 177 }
 178
 179 /*
 180  * Send an external call sigp to another cpu and return without waiting
 181  * for its completion.
 182  */
 183 static void smp_ext_bitcall(int cpu, int sig)
 184 {
 185         /*
 186          * Set signaling bit in lowcore of target cpu and kick it
 187          */
 188         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 189         while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
 190                 udelay(10);
 191 }
 192
 193 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 194 {
 195         int cpu;
 196
 197         for_each_cpu(cpu, mask)
 198                 smp_ext_bitcall(cpu, ec_call_function);
 199 }
 200
 201 void arch_send_call_function_single_ipi(int cpu)
 202 {
 203         smp_ext_bitcall(cpu, ec_call_function_single);
 204 }
 205
 206 #ifndef CONFIG_64BIT
 207 /*
 208  * this function sends a 'purge tlb' signal to another CPU.
 209  */
 210 static void smp_ptlb_callback(void *info)
 211 {
 212         __tlb_flush_local();
 213 }
 214
 215 void smp_ptlb_all(void)
 216 {
 217         on_each_cpu(smp_ptlb_callback, NULL, 1);
 218 }
 219 EXPORT_SYMBOL(smp_ptlb_all);
 220 #endif /* ! CONFIG_64BIT */
 221
 222 /*
 223  * this function sends a 'reschedule' IPI to another CPU.
 224  * it goes straight through and wastes no time serializing
 225  * anything. Worst case is that we lose a reschedule ...
 226  */
 227 void smp_send_reschedule(int cpu)
 228 {
 229         smp_ext_bitcall(cpu, ec_schedule);
 230 }
 231
 232 /*
 233  * parameter area for the set/clear control bit callbacks
 234  */
 235 struct ec_creg_mask_parms {
 236         unsigned long orvals[16];
 237         unsigned long andvals[16];
 238 };
 239
 240 /*
 241  * callback for setting/clearing control bits
 242  */
 243 static void smp_ctl_bit_callback(void *info)
 244 {
 245         struct ec_creg_mask_parms *pp = info;
 246         unsigned long cregs[16];
 247         int i;
 248
 249         __ctl_store(cregs, 0, 15);
 250         for (i = 0; i <= 15; i++)
 251                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 252         __ctl_load(cregs, 0, 15);
 253 }
 254
 255 /*
 256  * Set a bit in a control register of all cpus
 257  */
 258 void smp_ctl_set_bit(int cr, int bit)
 259 {
 260         struct ec_creg_mask_parms parms;
 261
 262         memset(&parms.orvals, 0, sizeof(parms.orvals));
 263         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 264         parms.orvals[cr] = 1 << bit;
 265         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 266 }
 267 EXPORT_SYMBOL(smp_ctl_set_bit);
 268
 269 /*
 270  * Clear a bit in a control register of all cpus
 271  */
 272 void smp_ctl_clear_bit(int cr, int bit)
 273 {
 274         struct ec_creg_mask_parms parms;
 275
 276         memset(&parms.orvals, 0, sizeof(parms.orvals));
 277         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 278         parms.andvals[cr] = ~(1L << bit);
 279         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 280 }
 281 EXPORT_SYMBOL(smp_ctl_clear_bit);
 282
 283 #ifdef CONFIG_ZFCPDUMP
 284
 285 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 286 {
 287         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 288                 return;
 289         if (cpu >= NR_CPUS) {
 290                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 291                            "the dump\n", cpu, NR_CPUS - 1);
 292                 return;
 293         }
 294         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 295         while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
 296                 cpu_relax();
 297         memcpy_real(zfcpdump_save_areas[cpu],
 298                     (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 299                     sizeof(struct save_area));
 300 }
 301
 302 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 303 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 304
 305 #else
 306
 307 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 308
 309 #endif /* CONFIG_ZFCPDUMP */
 310
 311 static int cpu_known(int cpu_id)
 312 {
 313         int cpu;
 314
 315         for_each_present_cpu(cpu) {
 316                 if (__cpu_logical_map[cpu] == cpu_id)
 317                         return 1;
 318         }
 319         return 0;
 320 }
 321
 322 static int smp_rescan_cpus_sigp(cpumask_t avail)
 323 {
 324         int cpu_id, logical_cpu;
 325
 326         logical_cpu = cpumask_first(&avail);
 327         if (logical_cpu >= nr_cpu_ids)
 328                 return 0;
 329         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 330                 if (cpu_known(cpu_id))
 331                         continue;
 332                 __cpu_logical_map[logical_cpu] = cpu_id;
 333                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 334                 if (!cpu_stopped(logical_cpu))
 335                         continue;
 336                 cpu_set(logical_cpu, cpu_present_map);
 337                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 338                 logical_cpu = cpumask_next(logical_cpu, &avail);
 339                 if (logical_cpu >= nr_cpu_ids)
 340                         break;
 341         }
 342         return 0;
 343 }
 344
 345 static int smp_rescan_cpus_sclp(cpumask_t avail)
 346 {
 347         struct sclp_cpu_info *info;
 348         int cpu_id, logical_cpu, cpu;
 349         int rc;
 350
 351         logical_cpu = cpumask_first(&avail);
 352         if (logical_cpu >= nr_cpu_ids)
 353                 return 0;
 354         info = kmalloc(sizeof(*info), GFP_KERNEL);
 355         if (!info)
 356                 return -ENOMEM;
 357         rc = sclp_get_cpu_info(info);
 358         if (rc)
 359                 goto out;
 360         for (cpu = 0; cpu < info->combined; cpu++) {
 361                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 362                         continue;
 363                 cpu_id = info->cpu[cpu].address;
 364                 if (cpu_known(cpu_id))
 365                         continue;
 366                 __cpu_logical_map[logical_cpu] = cpu_id;
 367                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 368                 cpu_set(logical_cpu, cpu_present_map);
 369                 if (cpu >= info->configured)
 370                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 371                 else
 372                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 373                 logical_cpu = cpumask_next(logical_cpu, &avail);
 374                 if (logical_cpu >= nr_cpu_ids)
 375                         break;
 376         }
 377 out:
 378         kfree(info);
 379         return rc;
 380 }
 381
 382 static int __smp_rescan_cpus(void)
 383 {
 384         cpumask_t avail;
 385
 386         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 387         if (smp_use_sigp_detection)
 388                 return smp_rescan_cpus_sigp(avail);
 389         else
 390                 return smp_rescan_cpus_sclp(avail);
 391 }
 392
 393 static void __init smp_detect_cpus(void)
 394 {
 395         unsigned int cpu, c_cpus, s_cpus;
 396         struct sclp_cpu_info *info;
 397         u16 boot_cpu_addr, cpu_addr;
 398
 399         c_cpus = 1;
 400         s_cpus = 0;
 401         boot_cpu_addr = __cpu_logical_map[0];
 402         info = kmalloc(sizeof(*info), GFP_KERNEL);
 403         if (!info)
 404                 panic("smp_detect_cpus failed to allocate memory\n");
 405         /* Use sigp detection algorithm if sclp doesn't work. */
 406         if (sclp_get_cpu_info(info)) {
 407                 smp_use_sigp_detection = 1;
 408                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 409                         if (cpu == boot_cpu_addr)
 410                                 continue;
 411                         if (!raw_cpu_stopped(cpu))
 412                                 continue;
 413                         smp_get_save_area(c_cpus, cpu);
 414                         c_cpus++;
 415                 }
 416                 goto out;
 417         }
 418
 419         if (info->has_cpu_type) {
 420                 for (cpu = 0; cpu < info->combined; cpu++) {
 421                         if (info->cpu[cpu].address == boot_cpu_addr) {
 422                                 smp_cpu_type = info->cpu[cpu].type;
 423                                 break;
 424                         }
 425                 }
 426         }
 427
 428         for (cpu = 0; cpu < info->combined; cpu++) {
 429                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 430                         continue;
 431                 cpu_addr = info->cpu[cpu].address;
 432                 if (cpu_addr == boot_cpu_addr)
 433                         continue;
 434                 if (!raw_cpu_stopped(cpu_addr)) {
 435                         s_cpus++;
 436                         continue;
 437                 }
 438                 smp_get_save_area(c_cpus, cpu_addr);
 439                 c_cpus++;
 440         }
 441 out:
 442         kfree(info);
 443         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 444         get_online_cpus();
 445         __smp_rescan_cpus();
 446         put_online_cpus();
 447 }
 448
 449 /*
 450  *      Activate a secondary processor.
 451  */
 452 int __cpuinit start_secondary(void *cpuvoid)
 453 {
 454         /* Setup the cpu */
 455         cpu_init();
 456         preempt_disable();
 457         /* Enable TOD clock interrupts on the secondary cpu. */
 458         init_cpu_timer();
 459         /* Enable cpu timer interrupts on the secondary cpu. */
 460         init_cpu_vtimer();
 461         /* Enable pfault pseudo page faults on this cpu. */
 462         pfault_init();
 463
 464         /* call cpu notifiers */
 465         notify_cpu_starting(smp_processor_id());
 466         /* Mark this cpu as online */
 467         ipi_call_lock();
 468         cpu_set(smp_processor_id(), cpu_online_map);
 469         ipi_call_unlock();
 470         /* Switch on interrupts */
 471         local_irq_enable();
 472         /* Print info about this processor */
 473         print_cpu_info();
 474         /* cpu_idle will call schedule for us */
 475         cpu_idle();
 476         return 0;
 477 }
 478
 479 static void __init smp_create_idle(unsigned int cpu)
 480 {
 481         struct task_struct *p;
 482
 483         /*
 484          *  don't care about the psw and regs settings since we'll never
 485          *  reschedule the forked task.
 486          */
 487         p = fork_idle(cpu);
 488         if (IS_ERR(p))
 489                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 490         current_set[cpu] = p;
 491 }
 492
 493 static int __cpuinit smp_alloc_lowcore(int cpu)
 494 {
 495         unsigned long async_stack, panic_stack;
 496         struct _lowcore *lowcore;
 497
 498         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 499         if (!lowcore)
 500                 return -ENOMEM;
 501         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 502         panic_stack = __get_free_page(GFP_KERNEL);
 503         if (!panic_stack || !async_stack)
 504                 goto out;
 505         memcpy(lowcore, &S390_lowcore, 512);
 506         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 507         lowcore->async_stack = async_stack + ASYNC_SIZE;
 508         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 509
 510 #ifndef CONFIG_64BIT
 511         if (MACHINE_HAS_IEEE) {
 512                 unsigned long save_area;
 513
 514                 save_area = get_zeroed_page(GFP_KERNEL);
 515                 if (!save_area)
 516                         goto out;
 517                 lowcore->extended_save_area_addr = (u32) save_area;
 518         }
 519 #else
 520         if (vdso_alloc_per_cpu(cpu, lowcore))
 521                 goto out;
 522 #endif
 523         lowcore_ptr[cpu] = lowcore;
 524         return 0;
 525
 526 out:
 527         free_page(panic_stack);
 528         free_pages(async_stack, ASYNC_ORDER);
 529         free_pages((unsigned long) lowcore, LC_ORDER);
 530         return -ENOMEM;
 531 }
 532
 533 static void smp_free_lowcore(int cpu)
 534 {
 535         struct _lowcore *lowcore;
 536
 537         lowcore = lowcore_ptr[cpu];
 538 #ifndef CONFIG_64BIT
 539         if (MACHINE_HAS_IEEE)
 540                 free_page((unsigned long) lowcore->extended_save_area_addr);
 541 #else
 542         vdso_free_per_cpu(cpu, lowcore);
 543 #endif
 544         free_page(lowcore->panic_stack - PAGE_SIZE);
 545         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 546         free_pages((unsigned long) lowcore, LC_ORDER);
 547         lowcore_ptr[cpu] = NULL;
 548 }
 549
 550 /* Upping and downing of CPUs */
 551 int __cpuinit __cpu_up(unsigned int cpu)
 552 {
 553         struct _lowcore *cpu_lowcore;
 554         struct task_struct *idle;
 555         struct stack_frame *sf;
 556         u32 lowcore;
 557         int ccode;
 558
 559         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 560                 return -EIO;
 561         if (smp_alloc_lowcore(cpu))
 562                 return -ENOMEM;
 563         do {
 564                 ccode = sigp(cpu, sigp_initial_cpu_reset);
 565                 if (ccode == sigp_busy)
 566                         udelay(10);
 567                 if (ccode == sigp_not_operational)
 568                         goto err_out;
 569         } while (ccode == sigp_busy);
 570
 571         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 572         while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 573                 udelay(10);
 574
 575         idle = current_set[cpu];
 576         cpu_lowcore = lowcore_ptr[cpu];
 577         cpu_lowcore->kernel_stack = (unsigned long)
 578                 task_stack_page(idle) + THREAD_SIZE;
 579         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 580         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 581                                      - sizeof(struct pt_regs)
 582                                      - sizeof(struct stack_frame));
 583         memset(sf, 0, sizeof(struct stack_frame));
 584         sf->gprs[9] = (unsigned long) sf;
 585         cpu_lowcore->save_area[15] = (unsigned long) sf;
 586         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 587         atomic_inc(&init_mm.context.attach_count);
 588         asm volatile(
 589                 "       stam    0,15,0(%0)"
 590                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 591         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 592         cpu_lowcore->current_task = (unsigned long) idle;
 593         cpu_lowcore->cpu_nr = cpu;
 594         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 595         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 596         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 597         eieio();
 598
 599         while (sigp(cpu, sigp_restart) == sigp_busy)
 600                 udelay(10);
 601
 602         while (!cpu_online(cpu))
 603                 cpu_relax();
 604         return 0;
 605
 606 err_out:
 607         smp_free_lowcore(cpu);
 608         return -EIO;
 609 }
 610
 611 static int __init setup_possible_cpus(char *s)
 612 {
 613         int pcpus, cpu;
 614
 615         pcpus = simple_strtoul(s, NULL, 0);
 616         init_cpu_possible(cpumask_of(0));
 617         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 618                 set_cpu_possible(cpu, true);
 619         return 0;
 620 }
 621 early_param("possible_cpus", setup_possible_cpus);
 622
 623 #ifdef CONFIG_HOTPLUG_CPU
 624
 625 int __cpu_disable(void)
 626 {
 627         struct ec_creg_mask_parms cr_parms;
 628         int cpu = smp_processor_id();
 629
 630         cpu_clear(cpu, cpu_online_map);
 631
 632         /* Disable pfault pseudo page faults on this cpu. */
 633         pfault_fini();
 634
 635         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 636         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 637
 638         /* disable all external interrupts */
 639         cr_parms.orvals[0] = 0;
 640         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 641                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 642         /* disable all I/O interrupts */
 643         cr_parms.orvals[6] = 0;
 644         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 645                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 646         /* disable most machine checks */
 647         cr_parms.orvals[14] = 0;
 648         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 649                                  1 << 25 | 1 << 24);
 650
 651         smp_ctl_bit_callback(&cr_parms);
 652
 653         return 0;
 654 }
 655
 656 void __cpu_die(unsigned int cpu)
 657 {
 658         /* Wait until target cpu is down */
 659         while (!cpu_stopped(cpu))
 660                 cpu_relax();
 661         while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
 662                 udelay(10);
 663         smp_free_lowcore(cpu);
 664         atomic_dec(&init_mm.context.attach_count);
 665         pr_info("Processor %d stopped\n", cpu);
 666 }
 667
 668 void cpu_die(void)
 669 {
 670         idle_task_exit();
 671         while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
 672                 cpu_relax();
 673         for (;;);
 674 }
 675
 676 #endif /* CONFIG_HOTPLUG_CPU */
 677
 678 void __init smp_prepare_cpus(unsigned int max_cpus)
 679 {
 680 #ifndef CONFIG_64BIT
 681         unsigned long save_area = 0;
 682 #endif
 683         unsigned long async_stack, panic_stack;
 684         struct _lowcore *lowcore;
 685         unsigned int cpu;
 686
 687         smp_detect_cpus();
 688
 689         /* request the 0x1201 emergency signal external interrupt */
 690         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 691                 panic("Couldn't request external interrupt 0x1201");
 692         print_cpu_info();
 693
 694         /* Reallocate current lowcore, but keep its contents. */
 695         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 696         panic_stack = __get_free_page(GFP_KERNEL);
 697         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 698         BUG_ON(!lowcore || !panic_stack || !async_stack);
 699 #ifndef CONFIG_64BIT
 700         if (MACHINE_HAS_IEEE)
 701                 save_area = get_zeroed_page(GFP_KERNEL);
 702 #endif
 703         local_irq_disable();
 704         local_mcck_disable();
 705         lowcore_ptr[smp_processor_id()] = lowcore;
 706         *lowcore = S390_lowcore;
 707         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 708         lowcore->async_stack = async_stack + ASYNC_SIZE;
 709 #ifndef CONFIG_64BIT
 710         if (MACHINE_HAS_IEEE)
 711                 lowcore->extended_save_area_addr = (u32) save_area;
 712 #endif
 713         set_prefix((u32)(unsigned long) lowcore);
 714         local_mcck_enable();
 715         local_irq_enable();
 716 #ifdef CONFIG_64BIT
 717         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 718                 BUG();
 719 #endif
 720         for_each_possible_cpu(cpu)
 721                 if (cpu != smp_processor_id())
 722                         smp_create_idle(cpu);
 723 }
 724
 725 void __init smp_prepare_boot_cpu(void)
 726 {
 727         BUG_ON(smp_processor_id() != 0);
 728
 729         current_thread_info()->cpu = 0;
 730         cpu_set(0, cpu_present_map);
 731         cpu_set(0, cpu_online_map);
 732         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 733         current_set[0] = current;
 734         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 735         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 736 }
 737
 738 void __init smp_cpus_done(unsigned int max_cpus)
 739 {
 740 }
 741
 742 void __init smp_setup_processor_id(void)
 743 {
 744         S390_lowcore.cpu_nr = 0;
 745         __cpu_logical_map[0] = stap();
 746 }
 747
 748 /*
 749  * the frequency of the profiling timer can be changed
 750  * by writing a multiplier value into /proc/profile.
 751  *
 752  * usually you want to run this on all CPUs ;)
 753  */
 754 int setup_profiling_timer(unsigned int multiplier)
 755 {
 756         return 0;
 757 }
 758
 759 #ifdef CONFIG_HOTPLUG_CPU
 760 static ssize_t cpu_configure_show(struct sys_device *dev,
 761                                 struct sysdev_attribute *attr, char *buf)
 762 {
 763         ssize_t count;
 764
 765         mutex_lock(&smp_cpu_state_mutex);
 766         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 767         mutex_unlock(&smp_cpu_state_mutex);
 768         return count;
 769 }
 770
 771 static ssize_t cpu_configure_store(struct sys_device *dev,
 772                                   struct sysdev_attribute *attr,
 773                                   const char *buf, size_t count)
 774 {
 775         int cpu = dev->id;
 776         int val, rc;
 777         char delim;
 778
 779         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 780                 return -EINVAL;
 781         if (val != 0 && val != 1)
 782                 return -EINVAL;
 783
 784         get_online_cpus();
 785         mutex_lock(&smp_cpu_state_mutex);
 786         rc = -EBUSY;
 787         /* disallow configuration changes of online cpus and cpu 0 */
 788         if (cpu_online(cpu) || cpu == 0)
 789                 goto out;
 790         rc = 0;
 791         switch (val) {
 792         case 0:
 793                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 794                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 795                         if (!rc) {
 796                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 797                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 798                         }
 799                 }
 800                 break;
 801         case 1:
 802                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 803                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 804                         if (!rc) {
 805                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 806                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 807                         }
 808                 }
 809                 break;
 810         default:
 811                 break;
 812         }
 813 out:
 814         mutex_unlock(&smp_cpu_state_mutex);
 815         put_online_cpus();
 816         return rc ? rc : count;
 817 }
 818 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 819 #endif /* CONFIG_HOTPLUG_CPU */
 820
 821 static ssize_t cpu_polarization_show(struct sys_device *dev,
 822                                      struct sysdev_attribute *attr, char *buf)
 823 {
 824         int cpu = dev->id;
 825         ssize_t count;
 826
 827         mutex_lock(&smp_cpu_state_mutex);
 828         switch (smp_cpu_polarization[cpu]) {
 829         case POLARIZATION_HRZ:
 830                 count = sprintf(buf, "horizontal\n");
 831                 break;
 832         case POLARIZATION_VL:
 833                 count = sprintf(buf, "vertical:low\n");
 834                 break;
 835         case POLARIZATION_VM:
 836                 count = sprintf(buf, "vertical:medium\n");
 837                 break;
 838         case POLARIZATION_VH:
 839                 count = sprintf(buf, "vertical:high\n");
 840                 break;
 841         default:
 842                 count = sprintf(buf, "unknown\n");
 843                 break;
 844         }
 845         mutex_unlock(&smp_cpu_state_mutex);
 846         return count;
 847 }
 848 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 849
 850 static ssize_t show_cpu_address(struct sys_device *dev,
 851                                 struct sysdev_attribute *attr, char *buf)
 852 {
 853         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 854 }
 855 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 856
 857
 858 static struct attribute *cpu_common_attrs[] = {
 859 #ifdef CONFIG_HOTPLUG_CPU
 860         &attr_configure.attr,
 861 #endif
 862         &attr_address.attr,
 863         &attr_polarization.attr,
 864         NULL,
 865 };
 866
 867 static struct attribute_group cpu_common_attr_group = {
 868         .attrs = cpu_common_attrs,
 869 };
 870
 871 static ssize_t show_capability(struct sys_device *dev,
 872                                 struct sysdev_attribute *attr, char *buf)
 873 {
 874         unsigned int capability;
 875         int rc;
 876
 877         rc = get_cpu_capability(&capability);
 878         if (rc)
 879                 return rc;
 880         return sprintf(buf, "%u\n", capability);
 881 }
 882 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 883
 884 static ssize_t show_idle_count(struct sys_device *dev,
 885                                 struct sysdev_attribute *attr, char *buf)
 886 {
 887         struct s390_idle_data *idle;
 888         unsigned long long idle_count;
 889         unsigned int sequence;
 890
 891         idle = &per_cpu(s390_idle, dev->id);
 892 repeat:
 893         sequence = idle->sequence;
 894         smp_rmb();
 895         if (sequence & 1)
 896                 goto repeat;
 897         idle_count = idle->idle_count;
 898         if (idle->idle_enter)
 899                 idle_count++;
 900         smp_rmb();
 901         if (idle->sequence != sequence)
 902                 goto repeat;
 903         return sprintf(buf, "%llu\n", idle_count);
 904 }
 905 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 906
 907 static ssize_t show_idle_time(struct sys_device *dev,
 908                                 struct sysdev_attribute *attr, char *buf)
 909 {
 910         struct s390_idle_data *idle;
 911         unsigned long long now, idle_time, idle_enter;
 912         unsigned int sequence;
 913
 914         idle = &per_cpu(s390_idle, dev->id);
 915         now = get_clock();
 916 repeat:
 917         sequence = idle->sequence;
 918         smp_rmb();
 919         if (sequence & 1)
 920                 goto repeat;
 921         idle_time = idle->idle_time;
 922         idle_enter = idle->idle_enter;
 923         if (idle_enter != 0ULL && idle_enter < now)
 924                 idle_time += now - idle_enter;
 925         smp_rmb();
 926         if (idle->sequence != sequence)
 927                 goto repeat;
 928         return sprintf(buf, "%llu\n", idle_time >> 12);
 929 }
 930 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 931
 932 static struct attribute *cpu_online_attrs[] = {
 933         &attr_capability.attr,
 934         &attr_idle_count.attr,
 935         &attr_idle_time_us.attr,
 936         NULL,
 937 };
 938
 939 static struct attribute_group cpu_online_attr_group = {
 940         .attrs = cpu_online_attrs,
 941 };
 942
 943 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 944                                     unsigned long action, void *hcpu)
 945 {
 946         unsigned int cpu = (unsigned int)(long)hcpu;
 947         struct cpu *c = &per_cpu(cpu_devices, cpu);
 948         struct sys_device *s = &c->sysdev;
 949         struct s390_idle_data *idle;
 950         int err = 0;
 951
 952         switch (action) {
 953         case CPU_ONLINE:
 954         case CPU_ONLINE_FROZEN:
 955                 idle = &per_cpu(s390_idle, cpu);
 956                 memset(idle, 0, sizeof(struct s390_idle_data));
 957                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 958                 break;
 959         case CPU_DEAD:
 960         case CPU_DEAD_FROZEN:
 961                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 962                 break;
 963         }
 964         return notifier_from_errno(err);
 965 }
 966
 967 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 968         .notifier_call = smp_cpu_notify,
 969 };
 970
 971 static int __devinit smp_add_present_cpu(int cpu)
 972 {
 973         struct cpu *c = &per_cpu(cpu_devices, cpu);
 974         struct sys_device *s = &c->sysdev;
 975         int rc;
 976
 977         c->hotpluggable = 1;
 978         rc = register_cpu(c, cpu);
 979         if (rc)
 980                 goto out;
 981         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 982         if (rc)
 983                 goto out_cpu;
 984         if (!cpu_online(cpu))
 985                 goto out;
 986         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 987         if (!rc)
 988                 return 0;
 989         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 990 out_cpu:
 991 #ifdef CONFIG_HOTPLUG_CPU
 992         unregister_cpu(c);
 993 #endif
 994 out:
 995         return rc;
 996 }
 997
 998 #ifdef CONFIG_HOTPLUG_CPU
 999
1000 int __ref smp_rescan_cpus(void)
1001 {
1002         cpumask_t newcpus;
1003         int cpu;
1004         int rc;
1005
1006         get_online_cpus();
1007         mutex_lock(&smp_cpu_state_mutex);
1008         newcpus = cpu_present_map;
1009         rc = __smp_rescan_cpus();
1010         if (rc)
1011                 goto out;
1012         cpus_andnot(newcpus, cpu_present_map, newcpus);
1013         for_each_cpu_mask(cpu, newcpus) {
1014                 rc = smp_add_present_cpu(cpu);
1015                 if (rc)
1016                         cpu_clear(cpu, cpu_present_map);
1017         }
1018         rc = 0;
1019 out:
1020         mutex_unlock(&smp_cpu_state_mutex);
1021         put_online_cpus();
1022         if (!cpus_empty(newcpus))
1023                 topology_schedule_update();
1024         return rc;
1025 }
1026
1027 static ssize_t __ref rescan_store(struct sysdev_class *class,
1028                                   struct sysdev_class_attribute *attr,
1029                                   const char *buf,
1030                                   size_t count)
1031 {
1032         int rc;
1033
1034         rc = smp_rescan_cpus();
1035         return rc ? rc : count;
1036 }
1037 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1038 #endif /* CONFIG_HOTPLUG_CPU */
1039
1040 static ssize_t dispatching_show(struct sysdev_class *class,
1041                                 struct sysdev_class_attribute *attr,
1042                                 char *buf)
1043 {
1044         ssize_t count;
1045
1046         mutex_lock(&smp_cpu_state_mutex);
1047         count = sprintf(buf, "%d\n", cpu_management);
1048         mutex_unlock(&smp_cpu_state_mutex);
1049         return count;
1050 }
1051
1052 static ssize_t dispatching_store(struct sysdev_class *dev,
1053                                  struct sysdev_class_attribute *attr,
1054                                  const char *buf,
1055                                  size_t count)
1056 {
1057         int val, rc;
1058         char delim;
1059
1060         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1061                 return -EINVAL;
1062         if (val != 0 && val != 1)
1063                 return -EINVAL;
1064         rc = 0;
1065         get_online_cpus();
1066         mutex_lock(&smp_cpu_state_mutex);
1067         if (cpu_management == val)
1068                 goto out;
1069         rc = topology_set_cpu_management(val);
1070         if (!rc)
1071                 cpu_management = val;
1072 out:
1073         mutex_unlock(&smp_cpu_state_mutex);
1074         put_online_cpus();
1075         return rc ? rc : count;
1076 }
1077 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1078                          dispatching_store);
1079
1080 static int __init topology_init(void)
1081 {
1082         int cpu;
1083         int rc;
1084
1085         register_cpu_notifier(&smp_cpu_nb);
1086
1087 #ifdef CONFIG_HOTPLUG_CPU
1088         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1089         if (rc)
1090                 return rc;
1091 #endif
1092         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1093         if (rc)
1094                 return rc;
1095         for_each_present_cpu(cpu) {
1096                 rc = smp_add_present_cpu(cpu);
1097                 if (rc)
1098                         return rc;
1099         }
1100         return 0;
1101 }
1102 subsys_initcall(topology_init);