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 static struct task_struct *current_set[NR_CPUS];
  56
  57 static u8 smp_cpu_type;
  58 static int smp_use_sigp_detection;
  59
  60 enum s390_cpu_state {
  61         CPU_STATE_STANDBY,
  62         CPU_STATE_CONFIGURED,
  63 };
  64
  65 DEFINE_MUTEX(smp_cpu_state_mutex);
  66 int smp_cpu_polarization[NR_CPUS];
  67 static int smp_cpu_state[NR_CPUS];
  68 static int cpu_management;
  69
  70 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  71
  72 static void smp_ext_bitcall(int, ec_bit_sig);
  73
  74 static int cpu_stopped(int cpu)
  75 {
  76         __u32 status;
  77
  78         switch (signal_processor_ps(&status, 0, cpu, sigp_sense)) {
  79         case sigp_order_code_accepted:
  80         case sigp_status_stored:
  81                 /* Check for stopped and check stop state */
  82                 if (status & 0x50)
  83                         return 1;
  84                 break;
  85         default:
  86                 break;
  87         }
  88         return 0;
  89 }
  90
  91 void smp_send_stop(void)
  92 {
  93         int cpu, rc;
  94
  95         /* Disable all interrupts/machine checks */
  96         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  97         trace_hardirqs_off();
  98
  99         /* stop all processors */
 100         for_each_online_cpu(cpu) {
 101                 if (cpu == smp_processor_id())
 102                         continue;
 103                 do {
 104                         rc = signal_processor(cpu, sigp_stop);
 105                 } while (rc == sigp_busy);
 106
 107                 while (!cpu_stopped(cpu))
 108                         cpu_relax();
 109         }
 110 }
 111
 112 /*
 113  * This is the main routine where commands issued by other
 114  * cpus are handled.
 115  */
 116
 117 static void do_ext_call_interrupt(__u16 code)
 118 {
 119         unsigned long bits;
 120
 121         /*
 122          * handle bit signal external calls
 123          *
 124          * For the ec_schedule signal we have to do nothing. All the work
 125          * is done automatically when we return from the interrupt.
 126          */
 127         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 128
 129         if (test_bit(ec_call_function, &bits))
 130                 generic_smp_call_function_interrupt();
 131
 132         if (test_bit(ec_call_function_single, &bits))
 133                 generic_smp_call_function_single_interrupt();
 134 }
 135
 136 /*
 137  * Send an external call sigp to another cpu and return without waiting
 138  * for its completion.
 139  */
 140 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 141 {
 142         /*
 143          * Set signaling bit in lowcore of target cpu and kick it
 144          */
 145         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 146         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 147                 udelay(10);
 148 }
 149
 150 void arch_send_call_function_ipi(cpumask_t mask)
 151 {
 152         int cpu;
 153
 154         for_each_cpu_mask(cpu, mask)
 155                 smp_ext_bitcall(cpu, ec_call_function);
 156 }
 157
 158 void arch_send_call_function_single_ipi(int cpu)
 159 {
 160         smp_ext_bitcall(cpu, ec_call_function_single);
 161 }
 162
 163 #ifndef CONFIG_64BIT
 164 /*
 165  * this function sends a 'purge tlb' signal to another CPU.
 166  */
 167 static void smp_ptlb_callback(void *info)
 168 {
 169         __tlb_flush_local();
 170 }
 171
 172 void smp_ptlb_all(void)
 173 {
 174         on_each_cpu(smp_ptlb_callback, NULL, 1);
 175 }
 176 EXPORT_SYMBOL(smp_ptlb_all);
 177 #endif /* ! CONFIG_64BIT */
 178
 179 /*
 180  * this function sends a 'reschedule' IPI to another CPU.
 181  * it goes straight through and wastes no time serializing
 182  * anything. Worst case is that we lose a reschedule ...
 183  */
 184 void smp_send_reschedule(int cpu)
 185 {
 186         smp_ext_bitcall(cpu, ec_schedule);
 187 }
 188
 189 /*
 190  * parameter area for the set/clear control bit callbacks
 191  */
 192 struct ec_creg_mask_parms {
 193         unsigned long orvals[16];
 194         unsigned long andvals[16];
 195 };
 196
 197 /*
 198  * callback for setting/clearing control bits
 199  */
 200 static void smp_ctl_bit_callback(void *info)
 201 {
 202         struct ec_creg_mask_parms *pp = info;
 203         unsigned long cregs[16];
 204         int i;
 205
 206         __ctl_store(cregs, 0, 15);
 207         for (i = 0; i <= 15; i++)
 208                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 209         __ctl_load(cregs, 0, 15);
 210 }
 211
 212 /*
 213  * Set a bit in a control register of all cpus
 214  */
 215 void smp_ctl_set_bit(int cr, int bit)
 216 {
 217         struct ec_creg_mask_parms parms;
 218
 219         memset(&parms.orvals, 0, sizeof(parms.orvals));
 220         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 221         parms.orvals[cr] = 1 << bit;
 222         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 223 }
 224 EXPORT_SYMBOL(smp_ctl_set_bit);
 225
 226 /*
 227  * Clear a bit in a control register of all cpus
 228  */
 229 void smp_ctl_clear_bit(int cr, int bit)
 230 {
 231         struct ec_creg_mask_parms parms;
 232
 233         memset(&parms.orvals, 0, sizeof(parms.orvals));
 234         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 235         parms.andvals[cr] = ~(1L << bit);
 236         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 237 }
 238 EXPORT_SYMBOL(smp_ctl_clear_bit);
 239
 240 /*
 241  * In early ipl state a temp. logically cpu number is needed, so the sigp
 242  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 243  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 244  */
 245 #define CPU_INIT_NO     1
 246
 247 #ifdef CONFIG_ZFCPDUMP
 248
 249 /*
 250  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 251  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 252  * save its prefix registers, since they get lost, when switching from 31 bit
 253  * to 64 bit.
 254  */
 255 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 256         __attribute__((__section__(".data")));
 257
 258 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 259 {
 260         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 261                 return;
 262         if (cpu >= NR_CPUS) {
 263                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 264                            "the dump\n", cpu, NR_CPUS - 1);
 265                 return;
 266         }
 267         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 268         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 269         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 270                sigp_busy)
 271                 cpu_relax();
 272         memcpy(zfcpdump_save_areas[cpu],
 273                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 274                SAVE_AREA_SIZE);
 275 #ifdef CONFIG_64BIT
 276         /* copy original prefix register */
 277         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 278 #endif
 279 }
 280
 281 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 282 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 283
 284 #else
 285
 286 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 287
 288 #endif /* CONFIG_ZFCPDUMP */
 289
 290 static int cpu_known(int cpu_id)
 291 {
 292         int cpu;
 293
 294         for_each_present_cpu(cpu) {
 295                 if (__cpu_logical_map[cpu] == cpu_id)
 296                         return 1;
 297         }
 298         return 0;
 299 }
 300
 301 static int smp_rescan_cpus_sigp(cpumask_t avail)
 302 {
 303         int cpu_id, logical_cpu;
 304
 305         logical_cpu = cpumask_first(&avail);
 306         if (logical_cpu >= nr_cpu_ids)
 307                 return 0;
 308         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 309                 if (cpu_known(cpu_id))
 310                         continue;
 311                 __cpu_logical_map[logical_cpu] = cpu_id;
 312                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 313                 if (!cpu_stopped(logical_cpu))
 314                         continue;
 315                 cpu_set(logical_cpu, cpu_present_map);
 316                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 317                 logical_cpu = cpumask_next(logical_cpu, &avail);
 318                 if (logical_cpu >= nr_cpu_ids)
 319                         break;
 320         }
 321         return 0;
 322 }
 323
 324 static int smp_rescan_cpus_sclp(cpumask_t avail)
 325 {
 326         struct sclp_cpu_info *info;
 327         int cpu_id, logical_cpu, cpu;
 328         int rc;
 329
 330         logical_cpu = cpumask_first(&avail);
 331         if (logical_cpu >= nr_cpu_ids)
 332                 return 0;
 333         info = kmalloc(sizeof(*info), GFP_KERNEL);
 334         if (!info)
 335                 return -ENOMEM;
 336         rc = sclp_get_cpu_info(info);
 337         if (rc)
 338                 goto out;
 339         for (cpu = 0; cpu < info->combined; cpu++) {
 340                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 341                         continue;
 342                 cpu_id = info->cpu[cpu].address;
 343                 if (cpu_known(cpu_id))
 344                         continue;
 345                 __cpu_logical_map[logical_cpu] = cpu_id;
 346                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 347                 cpu_set(logical_cpu, cpu_present_map);
 348                 if (cpu >= info->configured)
 349                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 350                 else
 351                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 352                 logical_cpu = cpumask_next(logical_cpu, &avail);
 353                 if (logical_cpu >= nr_cpu_ids)
 354                         break;
 355         }
 356 out:
 357         kfree(info);
 358         return rc;
 359 }
 360
 361 static int __smp_rescan_cpus(void)
 362 {
 363         cpumask_t avail;
 364
 365         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 366         if (smp_use_sigp_detection)
 367                 return smp_rescan_cpus_sigp(avail);
 368         else
 369                 return smp_rescan_cpus_sclp(avail);
 370 }
 371
 372 static void __init smp_detect_cpus(void)
 373 {
 374         unsigned int cpu, c_cpus, s_cpus;
 375         struct sclp_cpu_info *info;
 376         u16 boot_cpu_addr, cpu_addr;
 377
 378         c_cpus = 1;
 379         s_cpus = 0;
 380         boot_cpu_addr = __cpu_logical_map[0];
 381         info = kmalloc(sizeof(*info), GFP_KERNEL);
 382         if (!info)
 383                 panic("smp_detect_cpus failed to allocate memory\n");
 384         /* Use sigp detection algorithm if sclp doesn't work. */
 385         if (sclp_get_cpu_info(info)) {
 386                 smp_use_sigp_detection = 1;
 387                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 388                         if (cpu == boot_cpu_addr)
 389                                 continue;
 390                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 391                         if (!cpu_stopped(CPU_INIT_NO))
 392                                 continue;
 393                         smp_get_save_area(c_cpus, cpu);
 394                         c_cpus++;
 395                 }
 396                 goto out;
 397         }
 398
 399         if (info->has_cpu_type) {
 400                 for (cpu = 0; cpu < info->combined; cpu++) {
 401                         if (info->cpu[cpu].address == boot_cpu_addr) {
 402                                 smp_cpu_type = info->cpu[cpu].type;
 403                                 break;
 404                         }
 405                 }
 406         }
 407
 408         for (cpu = 0; cpu < info->combined; cpu++) {
 409                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 410                         continue;
 411                 cpu_addr = info->cpu[cpu].address;
 412                 if (cpu_addr == boot_cpu_addr)
 413                         continue;
 414                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 415                 if (!cpu_stopped(CPU_INIT_NO)) {
 416                         s_cpus++;
 417                         continue;
 418                 }
 419                 smp_get_save_area(c_cpus, cpu_addr);
 420                 c_cpus++;
 421         }
 422 out:
 423         kfree(info);
 424         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 425         get_online_cpus();
 426         __smp_rescan_cpus();
 427         put_online_cpus();
 428 }
 429
 430 /*
 431  *      Activate a secondary processor.
 432  */
 433 int __cpuinit start_secondary(void *cpuvoid)
 434 {
 435         /* Setup the cpu */
 436         cpu_init();
 437         preempt_disable();
 438         /* Enable TOD clock interrupts on the secondary cpu. */
 439         init_cpu_timer();
 440         /* Enable cpu timer interrupts on the secondary cpu. */
 441         init_cpu_vtimer();
 442         /* Enable pfault pseudo page faults on this cpu. */
 443         pfault_init();
 444
 445         /* call cpu notifiers */
 446         notify_cpu_starting(smp_processor_id());
 447         /* Mark this cpu as online */
 448         ipi_call_lock();
 449         cpu_set(smp_processor_id(), cpu_online_map);
 450         ipi_call_unlock();
 451         /* Switch on interrupts */
 452         local_irq_enable();
 453         /* Print info about this processor */
 454         print_cpu_info();
 455         /* cpu_idle will call schedule for us */
 456         cpu_idle();
 457         return 0;
 458 }
 459
 460 static void __init smp_create_idle(unsigned int cpu)
 461 {
 462         struct task_struct *p;
 463
 464         /*
 465          *  don't care about the psw and regs settings since we'll never
 466          *  reschedule the forked task.
 467          */
 468         p = fork_idle(cpu);
 469         if (IS_ERR(p))
 470                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 471         current_set[cpu] = p;
 472 }
 473
 474 static int __cpuinit smp_alloc_lowcore(int cpu)
 475 {
 476         unsigned long async_stack, panic_stack;
 477         struct _lowcore *lowcore;
 478
 479         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 480         if (!lowcore)
 481                 return -ENOMEM;
 482         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 483         panic_stack = __get_free_page(GFP_KERNEL);
 484         if (!panic_stack || !async_stack)
 485                 goto out;
 486         memcpy(lowcore, &S390_lowcore, 512);
 487         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 488         lowcore->async_stack = async_stack + ASYNC_SIZE;
 489         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 490
 491 #ifndef CONFIG_64BIT
 492         if (MACHINE_HAS_IEEE) {
 493                 unsigned long save_area;
 494
 495                 save_area = get_zeroed_page(GFP_KERNEL);
 496                 if (!save_area)
 497                         goto out;
 498                 lowcore->extended_save_area_addr = (u32) save_area;
 499         }
 500 #else
 501         if (vdso_alloc_per_cpu(cpu, lowcore))
 502                 goto out;
 503 #endif
 504         lowcore_ptr[cpu] = lowcore;
 505         return 0;
 506
 507 out:
 508         free_page(panic_stack);
 509         free_pages(async_stack, ASYNC_ORDER);
 510         free_pages((unsigned long) lowcore, LC_ORDER);
 511         return -ENOMEM;
 512 }
 513
 514 static void smp_free_lowcore(int cpu)
 515 {
 516         struct _lowcore *lowcore;
 517
 518         lowcore = lowcore_ptr[cpu];
 519 #ifndef CONFIG_64BIT
 520         if (MACHINE_HAS_IEEE)
 521                 free_page((unsigned long) lowcore->extended_save_area_addr);
 522 #else
 523         vdso_free_per_cpu(cpu, lowcore);
 524 #endif
 525         free_page(lowcore->panic_stack - PAGE_SIZE);
 526         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 527         free_pages((unsigned long) lowcore, LC_ORDER);
 528         lowcore_ptr[cpu] = NULL;
 529 }
 530
 531 /* Upping and downing of CPUs */
 532 int __cpuinit __cpu_up(unsigned int cpu)
 533 {
 534         struct task_struct *idle;
 535         struct _lowcore *cpu_lowcore;
 536         struct stack_frame *sf;
 537         sigp_ccode ccode;
 538         u32 lowcore;
 539
 540         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 541                 return -EIO;
 542         if (smp_alloc_lowcore(cpu))
 543                 return -ENOMEM;
 544         do {
 545                 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
 546                 if (ccode == sigp_busy)
 547                         udelay(10);
 548                 if (ccode == sigp_not_operational)
 549                         goto err_out;
 550         } while (ccode == sigp_busy);
 551
 552         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 553         while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 554                 udelay(10);
 555
 556         idle = current_set[cpu];
 557         cpu_lowcore = lowcore_ptr[cpu];
 558         cpu_lowcore->kernel_stack = (unsigned long)
 559                 task_stack_page(idle) + THREAD_SIZE;
 560         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 561         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 562                                      - sizeof(struct pt_regs)
 563                                      - sizeof(struct stack_frame));
 564         memset(sf, 0, sizeof(struct stack_frame));
 565         sf->gprs[9] = (unsigned long) sf;
 566         cpu_lowcore->save_area[15] = (unsigned long) sf;
 567         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 568         asm volatile(
 569                 "       stam    0,15,0(%0)"
 570                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 571         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 572         cpu_lowcore->current_task = (unsigned long) idle;
 573         cpu_lowcore->cpu_nr = cpu;
 574         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 575         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 576         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 577         eieio();
 578
 579         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 580                 udelay(10);
 581
 582         while (!cpu_online(cpu))
 583                 cpu_relax();
 584         return 0;
 585
 586 err_out:
 587         smp_free_lowcore(cpu);
 588         return -EIO;
 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         init_cpu_possible(cpumask_of(0));
 597         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 598                 set_cpu_possible(cpu, true);
 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 (!cpu_stopped(cpu))
 640                 cpu_relax();
 641         smp_free_lowcore(cpu);
 642         pr_info("Processor %d stopped\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
 664         smp_detect_cpus();
 665
 666         /* request the 0x1201 emergency signal external interrupt */
 667         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 668                 panic("Couldn't request external interrupt 0x1201");
 669         print_cpu_info();
 670
 671         /* Reallocate current lowcore, but keep its contents. */
 672         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 673         panic_stack = __get_free_page(GFP_KERNEL);
 674         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 675         BUG_ON(!lowcore || !panic_stack || !async_stack);
 676 #ifndef CONFIG_64BIT
 677         if (MACHINE_HAS_IEEE)
 678                 save_area = get_zeroed_page(GFP_KERNEL);
 679 #endif
 680         local_irq_disable();
 681         local_mcck_disable();
 682         lowcore_ptr[smp_processor_id()] = lowcore;
 683         *lowcore = S390_lowcore;
 684         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 685         lowcore->async_stack = async_stack + ASYNC_SIZE;
 686 #ifndef CONFIG_64BIT
 687         if (MACHINE_HAS_IEEE)
 688                 lowcore->extended_save_area_addr = (u32) save_area;
 689 #endif
 690         set_prefix((u32)(unsigned long) lowcore);
 691         local_mcck_enable();
 692         local_irq_enable();
 693 #ifdef CONFIG_64BIT
 694         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 695                 BUG();
 696 #endif
 697         for_each_possible_cpu(cpu)
 698                 if (cpu != smp_processor_id())
 699                         smp_create_idle(cpu);
 700 }
 701
 702 void __init smp_prepare_boot_cpu(void)
 703 {
 704         BUG_ON(smp_processor_id() != 0);
 705
 706         current_thread_info()->cpu = 0;
 707         cpu_set(0, cpu_present_map);
 708         cpu_set(0, cpu_online_map);
 709         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 710         current_set[0] = current;
 711         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 712         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 713 }
 714
 715 void __init smp_cpus_done(unsigned int max_cpus)
 716 {
 717 }
 718
 719 /*
 720  * the frequency of the profiling timer can be changed
 721  * by writing a multiplier value into /proc/profile.
 722  *
 723  * usually you want to run this on all CPUs ;)
 724  */
 725 int setup_profiling_timer(unsigned int multiplier)
 726 {
 727         return 0;
 728 }
 729
 730 #ifdef CONFIG_HOTPLUG_CPU
 731 static ssize_t cpu_configure_show(struct sys_device *dev,
 732                                 struct sysdev_attribute *attr, char *buf)
 733 {
 734         ssize_t count;
 735
 736         mutex_lock(&smp_cpu_state_mutex);
 737         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 738         mutex_unlock(&smp_cpu_state_mutex);
 739         return count;
 740 }
 741
 742 static ssize_t cpu_configure_store(struct sys_device *dev,
 743                                   struct sysdev_attribute *attr,
 744                                   const char *buf, size_t count)
 745 {
 746         int cpu = dev->id;
 747         int val, rc;
 748         char delim;
 749
 750         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 751                 return -EINVAL;
 752         if (val != 0 && val != 1)
 753                 return -EINVAL;
 754
 755         get_online_cpus();
 756         mutex_lock(&smp_cpu_state_mutex);
 757         rc = -EBUSY;
 758         if (cpu_online(cpu))
 759                 goto out;
 760         rc = 0;
 761         switch (val) {
 762         case 0:
 763                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 764                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 765                         if (!rc) {
 766                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 767                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 768                         }
 769                 }
 770                 break;
 771         case 1:
 772                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 773                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 774                         if (!rc) {
 775                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 776                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 777                         }
 778                 }
 779                 break;
 780         default:
 781                 break;
 782         }
 783 out:
 784         mutex_unlock(&smp_cpu_state_mutex);
 785         put_online_cpus();
 786         return rc ? rc : count;
 787 }
 788 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 789 #endif /* CONFIG_HOTPLUG_CPU */
 790
 791 static ssize_t cpu_polarization_show(struct sys_device *dev,
 792                                      struct sysdev_attribute *attr, char *buf)
 793 {
 794         int cpu = dev->id;
 795         ssize_t count;
 796
 797         mutex_lock(&smp_cpu_state_mutex);
 798         switch (smp_cpu_polarization[cpu]) {
 799         case POLARIZATION_HRZ:
 800                 count = sprintf(buf, "horizontal\n");
 801                 break;
 802         case POLARIZATION_VL:
 803                 count = sprintf(buf, "vertical:low\n");
 804                 break;
 805         case POLARIZATION_VM:
 806                 count = sprintf(buf, "vertical:medium\n");
 807                 break;
 808         case POLARIZATION_VH:
 809                 count = sprintf(buf, "vertical:high\n");
 810                 break;
 811         default:
 812                 count = sprintf(buf, "unknown\n");
 813                 break;
 814         }
 815         mutex_unlock(&smp_cpu_state_mutex);
 816         return count;
 817 }
 818 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 819
 820 static ssize_t show_cpu_address(struct sys_device *dev,
 821                                 struct sysdev_attribute *attr, char *buf)
 822 {
 823         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 824 }
 825 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 826
 827
 828 static struct attribute *cpu_common_attrs[] = {
 829 #ifdef CONFIG_HOTPLUG_CPU
 830         &attr_configure.attr,
 831 #endif
 832         &attr_address.attr,
 833         &attr_polarization.attr,
 834         NULL,
 835 };
 836
 837 static struct attribute_group cpu_common_attr_group = {
 838         .attrs = cpu_common_attrs,
 839 };
 840
 841 static ssize_t show_capability(struct sys_device *dev,
 842                                 struct sysdev_attribute *attr, char *buf)
 843 {
 844         unsigned int capability;
 845         int rc;
 846
 847         rc = get_cpu_capability(&capability);
 848         if (rc)
 849                 return rc;
 850         return sprintf(buf, "%u\n", capability);
 851 }
 852 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 853
 854 static ssize_t show_idle_count(struct sys_device *dev,
 855                                 struct sysdev_attribute *attr, char *buf)
 856 {
 857         struct s390_idle_data *idle;
 858         unsigned long long idle_count;
 859         unsigned int sequence;
 860
 861         idle = &per_cpu(s390_idle, dev->id);
 862 repeat:
 863         sequence = idle->sequence;
 864         smp_rmb();
 865         if (sequence & 1)
 866                 goto repeat;
 867         idle_count = idle->idle_count;
 868         if (idle->idle_enter)
 869                 idle_count++;
 870         smp_rmb();
 871         if (idle->sequence != sequence)
 872                 goto repeat;
 873         return sprintf(buf, "%llu\n", idle_count);
 874 }
 875 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 876
 877 static ssize_t show_idle_time(struct sys_device *dev,
 878                                 struct sysdev_attribute *attr, char *buf)
 879 {
 880         struct s390_idle_data *idle;
 881         unsigned long long now, idle_time, idle_enter;
 882         unsigned int sequence;
 883
 884         idle = &per_cpu(s390_idle, dev->id);
 885         now = get_clock();
 886 repeat:
 887         sequence = idle->sequence;
 888         smp_rmb();
 889         if (sequence & 1)
 890                 goto repeat;
 891         idle_time = idle->idle_time;
 892         idle_enter = idle->idle_enter;
 893         if (idle_enter != 0ULL && idle_enter < now)
 894                 idle_time += now - idle_enter;
 895         smp_rmb();
 896         if (idle->sequence != sequence)
 897                 goto repeat;
 898         return sprintf(buf, "%llu\n", idle_time >> 12);
 899 }
 900 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 901
 902 static struct attribute *cpu_online_attrs[] = {
 903         &attr_capability.attr,
 904         &attr_idle_count.attr,
 905         &attr_idle_time_us.attr,
 906         NULL,
 907 };
 908
 909 static struct attribute_group cpu_online_attr_group = {
 910         .attrs = cpu_online_attrs,
 911 };
 912
 913 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 914                                     unsigned long action, void *hcpu)
 915 {
 916         unsigned int cpu = (unsigned int)(long)hcpu;
 917         struct cpu *c = &per_cpu(cpu_devices, cpu);
 918         struct sys_device *s = &c->sysdev;
 919         struct s390_idle_data *idle;
 920
 921         switch (action) {
 922         case CPU_ONLINE:
 923         case CPU_ONLINE_FROZEN:
 924                 idle = &per_cpu(s390_idle, cpu);
 925                 memset(idle, 0, sizeof(struct s390_idle_data));
 926                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 927                         return NOTIFY_BAD;
 928                 break;
 929         case CPU_DEAD:
 930         case CPU_DEAD_FROZEN:
 931                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 932                 break;
 933         }
 934         return NOTIFY_OK;
 935 }
 936
 937 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 938         .notifier_call = smp_cpu_notify,
 939 };
 940
 941 static int __devinit smp_add_present_cpu(int cpu)
 942 {
 943         struct cpu *c = &per_cpu(cpu_devices, cpu);
 944         struct sys_device *s = &c->sysdev;
 945         int rc;
 946
 947         c->hotpluggable = 1;
 948         rc = register_cpu(c, cpu);
 949         if (rc)
 950                 goto out;
 951         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 952         if (rc)
 953                 goto out_cpu;
 954         if (!cpu_online(cpu))
 955                 goto out;
 956         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 957         if (!rc)
 958                 return 0;
 959         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 960 out_cpu:
 961 #ifdef CONFIG_HOTPLUG_CPU
 962         unregister_cpu(c);
 963 #endif
 964 out:
 965         return rc;
 966 }
 967
 968 #ifdef CONFIG_HOTPLUG_CPU
 969
 970 int __ref smp_rescan_cpus(void)
 971 {
 972         cpumask_t newcpus;
 973         int cpu;
 974         int rc;
 975
 976         get_online_cpus();
 977         mutex_lock(&smp_cpu_state_mutex);
 978         newcpus = cpu_present_map;
 979         rc = __smp_rescan_cpus();
 980         if (rc)
 981                 goto out;
 982         cpus_andnot(newcpus, cpu_present_map, newcpus);
 983         for_each_cpu_mask(cpu, newcpus) {
 984                 rc = smp_add_present_cpu(cpu);
 985                 if (rc)
 986                         cpu_clear(cpu, cpu_present_map);
 987         }
 988         rc = 0;
 989 out:
 990         mutex_unlock(&smp_cpu_state_mutex);
 991         put_online_cpus();
 992         if (!cpus_empty(newcpus))
 993                 topology_schedule_update();
 994         return rc;
 995 }
 996
 997 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
 998                                   size_t count)
 999 {
1000         int rc;
1001
1002         rc = smp_rescan_cpus();
1003         return rc ? rc : count;
1004 }
1005 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1006 #endif /* CONFIG_HOTPLUG_CPU */
1007
1008 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1009 {
1010         ssize_t count;
1011
1012         mutex_lock(&smp_cpu_state_mutex);
1013         count = sprintf(buf, "%d\n", cpu_management);
1014         mutex_unlock(&smp_cpu_state_mutex);
1015         return count;
1016 }
1017
1018 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1019                                  size_t count)
1020 {
1021         int val, rc;
1022         char delim;
1023
1024         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1025                 return -EINVAL;
1026         if (val != 0 && val != 1)
1027                 return -EINVAL;
1028         rc = 0;
1029         get_online_cpus();
1030         mutex_lock(&smp_cpu_state_mutex);
1031         if (cpu_management == val)
1032                 goto out;
1033         rc = topology_set_cpu_management(val);
1034         if (!rc)
1035                 cpu_management = val;
1036 out:
1037         mutex_unlock(&smp_cpu_state_mutex);
1038         put_online_cpus();
1039         return rc ? rc : count;
1040 }
1041 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1042                          dispatching_store);
1043
1044 /*
1045  * If the resume kernel runs on another cpu than the suspended kernel,
1046  * we have to switch the cpu IDs in the logical map.
1047  */
1048 void smp_switch_boot_cpu_in_resume(u32 resume_phys_cpu_id,
1049                                    struct _lowcore *suspend_lowcore)
1050 {
1051         int cpu, suspend_cpu_id, resume_cpu_id;
1052         u32 suspend_phys_cpu_id;
1053
1054         suspend_phys_cpu_id = __cpu_logical_map[suspend_lowcore->cpu_nr];
1055         suspend_cpu_id = suspend_lowcore->cpu_nr;
1056
1057         for_each_present_cpu(cpu) {
1058                 if (__cpu_logical_map[cpu] == resume_phys_cpu_id) {
1059                         resume_cpu_id = cpu;
1060                         goto found;
1061                 }
1062         }
1063         panic("Could not find resume cpu in logical map.\n");
1064
1065 found:
1066         printk("Resume  cpu ID: %i/%i\n", resume_phys_cpu_id, resume_cpu_id);
1067         printk("Suspend cpu ID: %i/%i\n", suspend_phys_cpu_id, suspend_cpu_id);
1068
1069         __cpu_logical_map[resume_cpu_id] = suspend_phys_cpu_id;
1070         __cpu_logical_map[suspend_cpu_id] = resume_phys_cpu_id;
1071
1072         lowcore_ptr[suspend_cpu_id]->cpu_addr = resume_phys_cpu_id;
1073 }
1074
1075 u32 smp_get_phys_cpu_id(void)
1076 {
1077         return __cpu_logical_map[smp_processor_id()];
1078 }
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);