arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999,2007
   5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
   6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
   7  *               Heiko Carstens (heiko.carstens@de.ibm.com)
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * We work with logical cpu numbering everywhere we can. The only
  14  * functions using the real cpu address (got from STAP) are the sigp
  15  * functions. For all other functions we use the identity mapping.
  16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
  17  * used e.g. to find the idle task belonging to a logical cpu. Every array
  18  * in the kernel is sorted by the logical cpu number and not by the physical
  19  * one which is causing all the confusion with __cpu_logical_map and
  20  * cpu_number_map in other architectures.
  21  */
  22
  23 #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/cpu.h>
  36 #include <linux/timex.h>
  37 #include <linux/bootmem.h>
  38 #include <asm/ipl.h>
  39 #include <asm/setup.h>
  40 #include <asm/sigp.h>
  41 #include <asm/pgalloc.h>
  42 #include <asm/irq.h>
  43 #include <asm/s390_ext.h>
  44 #include <asm/cpcmd.h>
  45 #include <asm/tlbflush.h>
  46 #include <asm/timer.h>
  47 #include <asm/lowcore.h>
  48 #include <asm/sclp.h>
  49 #include <asm/cpu.h>
  50 #include "entry.h"
  51
  52 /*
  53  * An array with a pointer the lowcore of every CPU.
  54  */
  55 struct _lowcore *lowcore_ptr[NR_CPUS];
  56 EXPORT_SYMBOL(lowcore_ptr);
  57
  58 cpumask_t cpu_online_map = CPU_MASK_NONE;
  59 EXPORT_SYMBOL(cpu_online_map);
  60
  61 cpumask_t cpu_possible_map = CPU_MASK_ALL;
  62 EXPORT_SYMBOL(cpu_possible_map);
  63
  64 static struct task_struct *current_set[NR_CPUS];
  65
  66 static u8 smp_cpu_type;
  67 static int smp_use_sigp_detection;
  68
  69 enum s390_cpu_state {
  70         CPU_STATE_STANDBY,
  71         CPU_STATE_CONFIGURED,
  72 };
  73
  74 DEFINE_MUTEX(smp_cpu_state_mutex);
  75 int smp_cpu_polarization[NR_CPUS];
  76 static int smp_cpu_state[NR_CPUS];
  77 static int cpu_management;
  78
  79 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  80
  81 static void smp_ext_bitcall(int, ec_bit_sig);
  82
  83 void smp_send_stop(void)
  84 {
  85         int cpu, rc;
  86
  87         /* Disable all interrupts/machine checks */
  88         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  89
  90         /* write magic number to zero page (absolute 0) */
  91         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
  92
  93         /* stop all processors */
  94         for_each_online_cpu(cpu) {
  95                 if (cpu == smp_processor_id())
  96                         continue;
  97                 do {
  98                         rc = signal_processor(cpu, sigp_stop);
  99                 } while (rc == sigp_busy);
 100
 101                 while (!smp_cpu_not_running(cpu))
 102                         cpu_relax();
 103         }
 104 }
 105
 106 /*
 107  * This is the main routine where commands issued by other
 108  * cpus are handled.
 109  */
 110
 111 static void do_ext_call_interrupt(__u16 code)
 112 {
 113         unsigned long bits;
 114
 115         /*
 116          * handle bit signal external calls
 117          *
 118          * For the ec_schedule signal we have to do nothing. All the work
 119          * is done automatically when we return from the interrupt.
 120          */
 121         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 122
 123         if (test_bit(ec_call_function, &bits))
 124                 generic_smp_call_function_interrupt();
 125
 126         if (test_bit(ec_call_function_single, &bits))
 127                 generic_smp_call_function_single_interrupt();
 128 }
 129
 130 /*
 131  * Send an external call sigp to another cpu and return without waiting
 132  * for its completion.
 133  */
 134 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 135 {
 136         /*
 137          * Set signaling bit in lowcore of target cpu and kick it
 138          */
 139         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 140         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 141                 udelay(10);
 142 }
 143
 144 void arch_send_call_function_ipi(cpumask_t mask)
 145 {
 146         int cpu;
 147
 148         for_each_cpu_mask(cpu, mask)
 149                 smp_ext_bitcall(cpu, ec_call_function);
 150 }
 151
 152 void arch_send_call_function_single_ipi(int cpu)
 153 {
 154         smp_ext_bitcall(cpu, ec_call_function_single);
 155 }
 156
 157 #ifndef CONFIG_64BIT
 158 /*
 159  * this function sends a 'purge tlb' signal to another CPU.
 160  */
 161 static void smp_ptlb_callback(void *info)
 162 {
 163         __tlb_flush_local();
 164 }
 165
 166 void smp_ptlb_all(void)
 167 {
 168         on_each_cpu(smp_ptlb_callback, NULL, 1);
 169 }
 170 EXPORT_SYMBOL(smp_ptlb_all);
 171 #endif /* ! CONFIG_64BIT */
 172
 173 /*
 174  * this function sends a 'reschedule' IPI to another CPU.
 175  * it goes straight through and wastes no time serializing
 176  * anything. Worst case is that we lose a reschedule ...
 177  */
 178 void smp_send_reschedule(int cpu)
 179 {
 180         smp_ext_bitcall(cpu, ec_schedule);
 181 }
 182
 183 /*
 184  * parameter area for the set/clear control bit callbacks
 185  */
 186 struct ec_creg_mask_parms {
 187         unsigned long orvals[16];
 188         unsigned long andvals[16];
 189 };
 190
 191 /*
 192  * callback for setting/clearing control bits
 193  */
 194 static void smp_ctl_bit_callback(void *info)
 195 {
 196         struct ec_creg_mask_parms *pp = info;
 197         unsigned long cregs[16];
 198         int i;
 199
 200         __ctl_store(cregs, 0, 15);
 201         for (i = 0; i <= 15; i++)
 202                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 203         __ctl_load(cregs, 0, 15);
 204 }
 205
 206 /*
 207  * Set a bit in a control register of all cpus
 208  */
 209 void smp_ctl_set_bit(int cr, int bit)
 210 {
 211         struct ec_creg_mask_parms parms;
 212
 213         memset(&parms.orvals, 0, sizeof(parms.orvals));
 214         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 215         parms.orvals[cr] = 1 << bit;
 216         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 217 }
 218 EXPORT_SYMBOL(smp_ctl_set_bit);
 219
 220 /*
 221  * Clear a bit in a control register of all cpus
 222  */
 223 void smp_ctl_clear_bit(int cr, int bit)
 224 {
 225         struct ec_creg_mask_parms parms;
 226
 227         memset(&parms.orvals, 0, sizeof(parms.orvals));
 228         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 229         parms.andvals[cr] = ~(1L << bit);
 230         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 231 }
 232 EXPORT_SYMBOL(smp_ctl_clear_bit);
 233
 234 /*
 235  * In early ipl state a temp. logically cpu number is needed, so the sigp
 236  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 237  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 238  */
 239 #define CPU_INIT_NO     1
 240
 241 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 242
 243 /*
 244  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 245  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 246  * save its prefix registers, since they get lost, when switching from 31 bit
 247  * to 64 bit.
 248  */
 249 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 250         __attribute__((__section__(".data")));
 251
 252 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 253 {
 254         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 255                 return;
 256         if (cpu >= NR_CPUS) {
 257                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 258                            "the dump\n", cpu, NR_CPUS - 1);
 259                 return;
 260         }
 261         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 262         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 263         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 264                sigp_busy)
 265                 cpu_relax();
 266         memcpy(zfcpdump_save_areas[cpu],
 267                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 268                SAVE_AREA_SIZE);
 269 #ifdef CONFIG_64BIT
 270         /* copy original prefix register */
 271         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 272 #endif
 273 }
 274
 275 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 276 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 277
 278 #else
 279
 280 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 281
 282 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 283
 284 static int cpu_stopped(int cpu)
 285 {
 286         __u32 status;
 287
 288         /* Check for stopped state */
 289         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 290             sigp_status_stored) {
 291                 if (status & 0x40)
 292                         return 1;
 293         }
 294         return 0;
 295 }
 296
 297 static int cpu_known(int cpu_id)
 298 {
 299         int cpu;
 300
 301         for_each_present_cpu(cpu) {
 302                 if (__cpu_logical_map[cpu] == cpu_id)
 303                         return 1;
 304         }
 305         return 0;
 306 }
 307
 308 static int smp_rescan_cpus_sigp(cpumask_t avail)
 309 {
 310         int cpu_id, logical_cpu;
 311
 312         logical_cpu = first_cpu(avail);
 313         if (logical_cpu == NR_CPUS)
 314                 return 0;
 315         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 316                 if (cpu_known(cpu_id))
 317                         continue;
 318                 __cpu_logical_map[logical_cpu] = cpu_id;
 319                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 320                 if (!cpu_stopped(logical_cpu))
 321                         continue;
 322                 cpu_set(logical_cpu, cpu_present_map);
 323                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 324                 logical_cpu = next_cpu(logical_cpu, avail);
 325                 if (logical_cpu == NR_CPUS)
 326                         break;
 327         }
 328         return 0;
 329 }
 330
 331 static int smp_rescan_cpus_sclp(cpumask_t avail)
 332 {
 333         struct sclp_cpu_info *info;
 334         int cpu_id, logical_cpu, cpu;
 335         int rc;
 336
 337         logical_cpu = first_cpu(avail);
 338         if (logical_cpu == NR_CPUS)
 339                 return 0;
 340         info = kmalloc(sizeof(*info), GFP_KERNEL);
 341         if (!info)
 342                 return -ENOMEM;
 343         rc = sclp_get_cpu_info(info);
 344         if (rc)
 345                 goto out;
 346         for (cpu = 0; cpu < info->combined; cpu++) {
 347                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 348                         continue;
 349                 cpu_id = info->cpu[cpu].address;
 350                 if (cpu_known(cpu_id))
 351                         continue;
 352                 __cpu_logical_map[logical_cpu] = cpu_id;
 353                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 354                 cpu_set(logical_cpu, cpu_present_map);
 355                 if (cpu >= info->configured)
 356                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 357                 else
 358                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 359                 logical_cpu = next_cpu(logical_cpu, avail);
 360                 if (logical_cpu == NR_CPUS)
 361                         break;
 362         }
 363 out:
 364         kfree(info);
 365         return rc;
 366 }
 367
 368 static int __smp_rescan_cpus(void)
 369 {
 370         cpumask_t avail;
 371
 372         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 373         if (smp_use_sigp_detection)
 374                 return smp_rescan_cpus_sigp(avail);
 375         else
 376                 return smp_rescan_cpus_sclp(avail);
 377 }
 378
 379 static void __init smp_detect_cpus(void)
 380 {
 381         unsigned int cpu, c_cpus, s_cpus;
 382         struct sclp_cpu_info *info;
 383         u16 boot_cpu_addr, cpu_addr;
 384
 385         c_cpus = 1;
 386         s_cpus = 0;
 387         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 388         info = kmalloc(sizeof(*info), GFP_KERNEL);
 389         if (!info)
 390                 panic("smp_detect_cpus failed to allocate memory\n");
 391         /* Use sigp detection algorithm if sclp doesn't work. */
 392         if (sclp_get_cpu_info(info)) {
 393                 smp_use_sigp_detection = 1;
 394                 for (cpu = 0; cpu <= 65535; cpu++) {
 395                         if (cpu == boot_cpu_addr)
 396                                 continue;
 397                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 398                         if (!cpu_stopped(CPU_INIT_NO))
 399                                 continue;
 400                         smp_get_save_area(c_cpus, cpu);
 401                         c_cpus++;
 402                 }
 403                 goto out;
 404         }
 405
 406         if (info->has_cpu_type) {
 407                 for (cpu = 0; cpu < info->combined; cpu++) {
 408                         if (info->cpu[cpu].address == boot_cpu_addr) {
 409                                 smp_cpu_type = info->cpu[cpu].type;
 410                                 break;
 411                         }
 412                 }
 413         }
 414
 415         for (cpu = 0; cpu < info->combined; cpu++) {
 416                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 417                         continue;
 418                 cpu_addr = info->cpu[cpu].address;
 419                 if (cpu_addr == boot_cpu_addr)
 420                         continue;
 421                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 422                 if (!cpu_stopped(CPU_INIT_NO)) {
 423                         s_cpus++;
 424                         continue;
 425                 }
 426                 smp_get_save_area(c_cpus, cpu_addr);
 427                 c_cpus++;
 428         }
 429 out:
 430         kfree(info);
 431         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 432         get_online_cpus();
 433         __smp_rescan_cpus();
 434         put_online_cpus();
 435 }
 436
 437 /*
 438  *      Activate a secondary processor.
 439  */
 440 int __cpuinit start_secondary(void *cpuvoid)
 441 {
 442         /* Setup the cpu */
 443         cpu_init();
 444         preempt_disable();
 445         /* Enable TOD clock interrupts on the secondary cpu. */
 446         init_cpu_timer();
 447         /* Enable cpu timer interrupts on the secondary cpu. */
 448         init_cpu_vtimer();
 449         /* Enable pfault pseudo page faults on this cpu. */
 450         pfault_init();
 451
 452         /* call cpu notifiers */
 453         notify_cpu_starting(smp_processor_id());
 454         /* Mark this cpu as online */
 455         ipi_call_lock();
 456         cpu_set(smp_processor_id(), cpu_online_map);
 457         ipi_call_unlock();
 458         /* Switch on interrupts */
 459         local_irq_enable();
 460         /* Print info about this processor */
 461         print_cpu_info(&S390_lowcore.cpu_data);
 462         /* cpu_idle will call schedule for us */
 463         cpu_idle();
 464         return 0;
 465 }
 466
 467 static void __init smp_create_idle(unsigned int cpu)
 468 {
 469         struct task_struct *p;
 470
 471         /*
 472          *  don't care about the psw and regs settings since we'll never
 473          *  reschedule the forked task.
 474          */
 475         p = fork_idle(cpu);
 476         if (IS_ERR(p))
 477                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 478         current_set[cpu] = p;
 479 }
 480
 481 static int __cpuinit smp_alloc_lowcore(int cpu)
 482 {
 483         unsigned long async_stack, panic_stack;
 484         struct _lowcore *lowcore;
 485         int lc_order;
 486
 487         lc_order = sizeof(long) == 8 ? 1 : 0;
 488         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 489         if (!lowcore)
 490                 return -ENOMEM;
 491         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 492         panic_stack = __get_free_page(GFP_KERNEL);
 493         if (!panic_stack || !async_stack)
 494                 goto out;
 495         memcpy(lowcore, &S390_lowcore, 512);
 496         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 497         lowcore->async_stack = async_stack + ASYNC_SIZE;
 498         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 499
 500 #ifndef CONFIG_64BIT
 501         if (MACHINE_HAS_IEEE) {
 502                 unsigned long save_area;
 503
 504                 save_area = get_zeroed_page(GFP_KERNEL);
 505                 if (!save_area)
 506                         goto out;
 507                 lowcore->extended_save_area_addr = (u32) save_area;
 508         }
 509 #endif
 510         lowcore_ptr[cpu] = lowcore;
 511         return 0;
 512
 513 out:
 514         free_page(panic_stack);
 515         free_pages(async_stack, ASYNC_ORDER);
 516         free_pages((unsigned long) lowcore, lc_order);
 517         return -ENOMEM;
 518 }
 519
 520 #ifdef CONFIG_HOTPLUG_CPU
 521 static void smp_free_lowcore(int cpu)
 522 {
 523         struct _lowcore *lowcore;
 524         int lc_order;
 525
 526         lc_order = sizeof(long) == 8 ? 1 : 0;
 527         lowcore = lowcore_ptr[cpu];
 528 #ifndef CONFIG_64BIT
 529         if (MACHINE_HAS_IEEE)
 530                 free_page((unsigned long) lowcore->extended_save_area_addr);
 531 #endif
 532         free_page(lowcore->panic_stack - PAGE_SIZE);
 533         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 534         free_pages((unsigned long) lowcore, lc_order);
 535         lowcore_ptr[cpu] = NULL;
 536 }
 537 #endif /* CONFIG_HOTPLUG_CPU */
 538
 539 /* Upping and downing of CPUs */
 540 int __cpuinit __cpu_up(unsigned int cpu)
 541 {
 542         struct task_struct *idle;
 543         struct _lowcore *cpu_lowcore;
 544         struct stack_frame *sf;
 545         sigp_ccode ccode;
 546
 547         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 548                 return -EIO;
 549         if (smp_alloc_lowcore(cpu))
 550                 return -ENOMEM;
 551
 552         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 553                                    cpu, sigp_set_prefix);
 554         if (ccode)
 555                 return -EIO;
 556
 557         idle = current_set[cpu];
 558         cpu_lowcore = lowcore_ptr[cpu];
 559         cpu_lowcore->kernel_stack = (unsigned long)
 560                 task_stack_page(idle) + THREAD_SIZE;
 561         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 562         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 563                                      - sizeof(struct pt_regs)
 564                                      - sizeof(struct stack_frame));
 565         memset(sf, 0, sizeof(struct stack_frame));
 566         sf->gprs[9] = (unsigned long) sf;
 567         cpu_lowcore->save_area[15] = (unsigned long) sf;
 568         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 569         asm volatile(
 570                 "       stam    0,15,0(%0)"
 571                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 572         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 573         cpu_lowcore->current_task = (unsigned long) idle;
 574         cpu_lowcore->cpu_data.cpu_nr = cpu;
 575         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 576         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
 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
 587 static int __init setup_possible_cpus(char *s)
 588 {
 589         int pcpus, cpu;
 590
 591         pcpus = simple_strtoul(s, NULL, 0);
 592         cpu_possible_map = cpumask_of_cpu(0);
 593         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 594                 cpu_set(cpu, cpu_possible_map);
 595         return 0;
 596 }
 597 early_param("possible_cpus", setup_possible_cpus);
 598
 599 #ifdef CONFIG_HOTPLUG_CPU
 600
 601 int __cpu_disable(void)
 602 {
 603         struct ec_creg_mask_parms cr_parms;
 604         int cpu = smp_processor_id();
 605
 606         cpu_clear(cpu, cpu_online_map);
 607
 608         /* Disable pfault pseudo page faults on this cpu. */
 609         pfault_fini();
 610
 611         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 612         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 613
 614         /* disable all external interrupts */
 615         cr_parms.orvals[0] = 0;
 616         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 617                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 618         /* disable all I/O interrupts */
 619         cr_parms.orvals[6] = 0;
 620         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 621                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 622         /* disable most machine checks */
 623         cr_parms.orvals[14] = 0;
 624         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 625                                  1 << 25 | 1 << 24);
 626
 627         smp_ctl_bit_callback(&cr_parms);
 628
 629         return 0;
 630 }
 631
 632 void __cpu_die(unsigned int cpu)
 633 {
 634         /* Wait until target cpu is down */
 635         while (!smp_cpu_not_running(cpu))
 636                 cpu_relax();
 637         smp_free_lowcore(cpu);
 638         pr_info("Processor %d stopped\n", cpu);
 639 }
 640
 641 void cpu_die(void)
 642 {
 643         idle_task_exit();
 644         signal_processor(smp_processor_id(), sigp_stop);
 645         BUG();
 646         for (;;);
 647 }
 648
 649 #endif /* CONFIG_HOTPLUG_CPU */
 650
 651 void __init smp_prepare_cpus(unsigned int max_cpus)
 652 {
 653 #ifndef CONFIG_64BIT
 654         unsigned long save_area = 0;
 655 #endif
 656         unsigned long async_stack, panic_stack;
 657         struct _lowcore *lowcore;
 658         unsigned int cpu;
 659         int lc_order;
 660
 661         smp_detect_cpus();
 662
 663         /* request the 0x1201 emergency signal external interrupt */
 664         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 665                 panic("Couldn't request external interrupt 0x1201");
 666         print_cpu_info(&S390_lowcore.cpu_data);
 667
 668         /* Reallocate current lowcore, but keep its contents. */
 669         lc_order = sizeof(long) == 8 ? 1 : 0;
 670         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 671         panic_stack = __get_free_page(GFP_KERNEL);
 672         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 673 #ifndef CONFIG_64BIT
 674         if (MACHINE_HAS_IEEE)
 675                 save_area = get_zeroed_page(GFP_KERNEL);
 676 #endif
 677         local_irq_disable();
 678         local_mcck_disable();
 679         lowcore_ptr[smp_processor_id()] = lowcore;
 680         *lowcore = S390_lowcore;
 681         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 682         lowcore->async_stack = async_stack + ASYNC_SIZE;
 683 #ifndef CONFIG_64BIT
 684         if (MACHINE_HAS_IEEE)
 685                 lowcore->extended_save_area_addr = (u32) save_area;
 686 #endif
 687         set_prefix((u32)(unsigned long) lowcore);
 688         local_mcck_enable();
 689         local_irq_enable();
 690         for_each_possible_cpu(cpu)
 691                 if (cpu != smp_processor_id())
 692                         smp_create_idle(cpu);
 693 }
 694
 695 void __init smp_prepare_boot_cpu(void)
 696 {
 697         BUG_ON(smp_processor_id() != 0);
 698
 699         current_thread_info()->cpu = 0;
 700         cpu_set(0, cpu_present_map);
 701         cpu_set(0, cpu_online_map);
 702         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 703         current_set[0] = current;
 704         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 705         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 706 }
 707
 708 void __init smp_cpus_done(unsigned int max_cpus)
 709 {
 710 }
 711
 712 /*
 713  * the frequency of the profiling timer can be changed
 714  * by writing a multiplier value into /proc/profile.
 715  *
 716  * usually you want to run this on all CPUs ;)
 717  */
 718 int setup_profiling_timer(unsigned int multiplier)
 719 {
 720         return 0;
 721 }
 722
 723 #ifdef CONFIG_HOTPLUG_CPU
 724 static ssize_t cpu_configure_show(struct sys_device *dev,
 725                                 struct sysdev_attribute *attr, char *buf)
 726 {
 727         ssize_t count;
 728
 729         mutex_lock(&smp_cpu_state_mutex);
 730         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 731         mutex_unlock(&smp_cpu_state_mutex);
 732         return count;
 733 }
 734
 735 static ssize_t cpu_configure_store(struct sys_device *dev,
 736                                   struct sysdev_attribute *attr,
 737                                   const char *buf, size_t count)
 738 {
 739         int cpu = dev->id;
 740         int val, rc;
 741         char delim;
 742
 743         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 744                 return -EINVAL;
 745         if (val != 0 && val != 1)
 746                 return -EINVAL;
 747
 748         get_online_cpus();
 749         mutex_lock(&smp_cpu_state_mutex);
 750         rc = -EBUSY;
 751         if (cpu_online(cpu))
 752                 goto out;
 753         rc = 0;
 754         switch (val) {
 755         case 0:
 756                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 757                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 758                         if (!rc) {
 759                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 760                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 761                         }
 762                 }
 763                 break;
 764         case 1:
 765                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 766                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 767                         if (!rc) {
 768                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 769                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 770                         }
 771                 }
 772                 break;
 773         default:
 774                 break;
 775         }
 776 out:
 777         mutex_unlock(&smp_cpu_state_mutex);
 778         put_online_cpus();
 779         return rc ? rc : count;
 780 }
 781 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 782 #endif /* CONFIG_HOTPLUG_CPU */
 783
 784 static ssize_t cpu_polarization_show(struct sys_device *dev,
 785                                      struct sysdev_attribute *attr, char *buf)
 786 {
 787         int cpu = dev->id;
 788         ssize_t count;
 789
 790         mutex_lock(&smp_cpu_state_mutex);
 791         switch (smp_cpu_polarization[cpu]) {
 792         case POLARIZATION_HRZ:
 793                 count = sprintf(buf, "horizontal\n");
 794                 break;
 795         case POLARIZATION_VL:
 796                 count = sprintf(buf, "vertical:low\n");
 797                 break;
 798         case POLARIZATION_VM:
 799                 count = sprintf(buf, "vertical:medium\n");
 800                 break;
 801         case POLARIZATION_VH:
 802                 count = sprintf(buf, "vertical:high\n");
 803                 break;
 804         default:
 805                 count = sprintf(buf, "unknown\n");
 806                 break;
 807         }
 808         mutex_unlock(&smp_cpu_state_mutex);
 809         return count;
 810 }
 811 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 812
 813 static ssize_t show_cpu_address(struct sys_device *dev,
 814                                 struct sysdev_attribute *attr, char *buf)
 815 {
 816         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 817 }
 818 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 819
 820
 821 static struct attribute *cpu_common_attrs[] = {
 822 #ifdef CONFIG_HOTPLUG_CPU
 823         &attr_configure.attr,
 824 #endif
 825         &attr_address.attr,
 826         &attr_polarization.attr,
 827         NULL,
 828 };
 829
 830 static struct attribute_group cpu_common_attr_group = {
 831         .attrs = cpu_common_attrs,
 832 };
 833
 834 static ssize_t show_capability(struct sys_device *dev,
 835                                 struct sysdev_attribute *attr, char *buf)
 836 {
 837         unsigned int capability;
 838         int rc;
 839
 840         rc = get_cpu_capability(&capability);
 841         if (rc)
 842                 return rc;
 843         return sprintf(buf, "%u\n", capability);
 844 }
 845 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 846
 847 static ssize_t show_idle_count(struct sys_device *dev,
 848                                 struct sysdev_attribute *attr, char *buf)
 849 {
 850         struct s390_idle_data *idle;
 851         unsigned long long idle_count;
 852
 853         idle = &per_cpu(s390_idle, dev->id);
 854         spin_lock_irq(&idle->lock);
 855         idle_count = idle->idle_count;
 856         spin_unlock_irq(&idle->lock);
 857         return sprintf(buf, "%llu\n", idle_count);
 858 }
 859 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 860
 861 static ssize_t show_idle_time(struct sys_device *dev,
 862                                 struct sysdev_attribute *attr, char *buf)
 863 {
 864         struct s390_idle_data *idle;
 865         unsigned long long new_time;
 866
 867         idle = &per_cpu(s390_idle, dev->id);
 868         spin_lock_irq(&idle->lock);
 869         if (idle->in_idle) {
 870                 new_time = get_clock();
 871                 idle->idle_time += new_time - idle->idle_enter;
 872                 idle->idle_enter = new_time;
 873         }
 874         new_time = idle->idle_time;
 875         spin_unlock_irq(&idle->lock);
 876         return sprintf(buf, "%llu\n", new_time >> 12);
 877 }
 878 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 879
 880 static struct attribute *cpu_online_attrs[] = {
 881         &attr_capability.attr,
 882         &attr_idle_count.attr,
 883         &attr_idle_time_us.attr,
 884         NULL,
 885 };
 886
 887 static struct attribute_group cpu_online_attr_group = {
 888         .attrs = cpu_online_attrs,
 889 };
 890
 891 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 892                                     unsigned long action, void *hcpu)
 893 {
 894         unsigned int cpu = (unsigned int)(long)hcpu;
 895         struct cpu *c = &per_cpu(cpu_devices, cpu);
 896         struct sys_device *s = &c->sysdev;
 897         struct s390_idle_data *idle;
 898
 899         switch (action) {
 900         case CPU_ONLINE:
 901         case CPU_ONLINE_FROZEN:
 902                 idle = &per_cpu(s390_idle, cpu);
 903                 spin_lock_irq(&idle->lock);
 904                 idle->idle_enter = 0;
 905                 idle->idle_time = 0;
 906                 idle->idle_count = 0;
 907                 spin_unlock_irq(&idle->lock);
 908                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 909                         return NOTIFY_BAD;
 910                 break;
 911         case CPU_DEAD:
 912         case CPU_DEAD_FROZEN:
 913                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 914                 break;
 915         }
 916         return NOTIFY_OK;
 917 }
 918
 919 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 920         .notifier_call = smp_cpu_notify,
 921 };
 922
 923 static int __devinit smp_add_present_cpu(int cpu)
 924 {
 925         struct cpu *c = &per_cpu(cpu_devices, cpu);
 926         struct sys_device *s = &c->sysdev;
 927         int rc;
 928
 929         c->hotpluggable = 1;
 930         rc = register_cpu(c, cpu);
 931         if (rc)
 932                 goto out;
 933         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 934         if (rc)
 935                 goto out_cpu;
 936         if (!cpu_online(cpu))
 937                 goto out;
 938         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 939         if (!rc)
 940                 return 0;
 941         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 942 out_cpu:
 943 #ifdef CONFIG_HOTPLUG_CPU
 944         unregister_cpu(c);
 945 #endif
 946 out:
 947         return rc;
 948 }
 949
 950 #ifdef CONFIG_HOTPLUG_CPU
 951
 952 int __ref smp_rescan_cpus(void)
 953 {
 954         cpumask_t newcpus;
 955         int cpu;
 956         int rc;
 957
 958         get_online_cpus();
 959         mutex_lock(&smp_cpu_state_mutex);
 960         newcpus = cpu_present_map;
 961         rc = __smp_rescan_cpus();
 962         if (rc)
 963                 goto out;
 964         cpus_andnot(newcpus, cpu_present_map, newcpus);
 965         for_each_cpu_mask(cpu, newcpus) {
 966                 rc = smp_add_present_cpu(cpu);
 967                 if (rc)
 968                         cpu_clear(cpu, cpu_present_map);
 969         }
 970         rc = 0;
 971 out:
 972         mutex_unlock(&smp_cpu_state_mutex);
 973         put_online_cpus();
 974         if (!cpus_empty(newcpus))
 975                 topology_schedule_update();
 976         return rc;
 977 }
 978
 979 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
 980                                   size_t count)
 981 {
 982         int rc;
 983
 984         rc = smp_rescan_cpus();
 985         return rc ? rc : count;
 986 }
 987 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
 988 #endif /* CONFIG_HOTPLUG_CPU */
 989
 990 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
 991 {
 992         ssize_t count;
 993
 994         mutex_lock(&smp_cpu_state_mutex);
 995         count = sprintf(buf, "%d\n", cpu_management);
 996         mutex_unlock(&smp_cpu_state_mutex);
 997         return count;
 998 }
 999
1000 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1001                                  size_t count)
1002 {
1003         int val, rc;
1004         char delim;
1005
1006         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1007                 return -EINVAL;
1008         if (val != 0 && val != 1)
1009                 return -EINVAL;
1010         rc = 0;
1011         get_online_cpus();
1012         mutex_lock(&smp_cpu_state_mutex);
1013         if (cpu_management == val)
1014                 goto out;
1015         rc = topology_set_cpu_management(val);
1016         if (!rc)
1017                 cpu_management = val;
1018 out:
1019         mutex_unlock(&smp_cpu_state_mutex);
1020         put_online_cpus();
1021         return rc ? rc : count;
1022 }
1023 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1024                          dispatching_store);
1025
1026 static int __init topology_init(void)
1027 {
1028         int cpu;
1029         int rc;
1030
1031         register_cpu_notifier(&smp_cpu_nb);
1032
1033 #ifdef CONFIG_HOTPLUG_CPU
1034         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1035         if (rc)
1036                 return rc;
1037 #endif
1038         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1039         if (rc)
1040                 return rc;
1041         for_each_present_cpu(cpu) {
1042                 rc = smp_add_present_cpu(cpu);
1043                 if (rc)
1044                         return rc;
1045         }
1046         return 0;
1047 }
1048 subsys_initcall(topology_init);