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         int lc_order;
 479
 480         lc_order = sizeof(long) == 8 ? 1 : 0;
 481         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 482         if (!lowcore)
 483                 return -ENOMEM;
 484         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 485         panic_stack = __get_free_page(GFP_KERNEL);
 486         if (!panic_stack || !async_stack)
 487                 goto out;
 488         memcpy(lowcore, &S390_lowcore, 512);
 489         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 490         lowcore->async_stack = async_stack + ASYNC_SIZE;
 491         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 492
 493 #ifndef CONFIG_64BIT
 494         if (MACHINE_HAS_IEEE) {
 495                 unsigned long save_area;
 496
 497                 save_area = get_zeroed_page(GFP_KERNEL);
 498                 if (!save_area)
 499                         goto out;
 500                 lowcore->extended_save_area_addr = (u32) save_area;
 501         }
 502 #else
 503         if (vdso_alloc_per_cpu(cpu, lowcore))
 504                 goto out;
 505 #endif
 506         lowcore_ptr[cpu] = lowcore;
 507         return 0;
 508
 509 out:
 510         free_page(panic_stack);
 511         free_pages(async_stack, ASYNC_ORDER);
 512         free_pages((unsigned long) lowcore, lc_order);
 513         return -ENOMEM;
 514 }
 515
 516 static void smp_free_lowcore(int cpu)
 517 {
 518         struct _lowcore *lowcore;
 519         int lc_order;
 520
 521         lc_order = sizeof(long) == 8 ? 1 : 0;
 522         lowcore = lowcore_ptr[cpu];
 523 #ifndef CONFIG_64BIT
 524         if (MACHINE_HAS_IEEE)
 525                 free_page((unsigned long) lowcore->extended_save_area_addr);
 526 #else
 527         vdso_free_per_cpu(cpu, lowcore);
 528 #endif
 529         free_page(lowcore->panic_stack - PAGE_SIZE);
 530         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 531         free_pages((unsigned long) lowcore, lc_order);
 532         lowcore_ptr[cpu] = NULL;
 533 }
 534
 535 /* Upping and downing of CPUs */
 536 int __cpuinit __cpu_up(unsigned int cpu)
 537 {
 538         struct task_struct *idle;
 539         struct _lowcore *cpu_lowcore;
 540         struct stack_frame *sf;
 541         sigp_ccode ccode;
 542         u32 lowcore;
 543
 544         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 545                 return -EIO;
 546         if (smp_alloc_lowcore(cpu))
 547                 return -ENOMEM;
 548         do {
 549                 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
 550                 if (ccode == sigp_busy)
 551                         udelay(10);
 552                 if (ccode == sigp_not_operational)
 553                         goto err_out;
 554         } while (ccode == sigp_busy);
 555
 556         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 557         while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 558                 udelay(10);
 559
 560         idle = current_set[cpu];
 561         cpu_lowcore = lowcore_ptr[cpu];
 562         cpu_lowcore->kernel_stack = (unsigned long)
 563                 task_stack_page(idle) + THREAD_SIZE;
 564         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 565         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 566                                      - sizeof(struct pt_regs)
 567                                      - sizeof(struct stack_frame));
 568         memset(sf, 0, sizeof(struct stack_frame));
 569         sf->gprs[9] = (unsigned long) sf;
 570         cpu_lowcore->save_area[15] = (unsigned long) sf;
 571         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 572         asm volatile(
 573                 "       stam    0,15,0(%0)"
 574                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 575         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 576         cpu_lowcore->current_task = (unsigned long) idle;
 577         cpu_lowcore->cpu_nr = cpu;
 578         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 579         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 580         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 581         eieio();
 582
 583         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 584                 udelay(10);
 585
 586         while (!cpu_online(cpu))
 587                 cpu_relax();
 588         return 0;
 589
 590 err_out:
 591         smp_free_lowcore(cpu);
 592         return -EIO;
 593 }
 594
 595 static int __init setup_possible_cpus(char *s)
 596 {
 597         int pcpus, cpu;
 598
 599         pcpus = simple_strtoul(s, NULL, 0);
 600         init_cpu_possible(cpumask_of(0));
 601         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 602                 set_cpu_possible(cpu, true);
 603         return 0;
 604 }
 605 early_param("possible_cpus", setup_possible_cpus);
 606
 607 #ifdef CONFIG_HOTPLUG_CPU
 608
 609 int __cpu_disable(void)
 610 {
 611         struct ec_creg_mask_parms cr_parms;
 612         int cpu = smp_processor_id();
 613
 614         cpu_clear(cpu, cpu_online_map);
 615
 616         /* Disable pfault pseudo page faults on this cpu. */
 617         pfault_fini();
 618
 619         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 620         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 621
 622         /* disable all external interrupts */
 623         cr_parms.orvals[0] = 0;
 624         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 625                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 626         /* disable all I/O interrupts */
 627         cr_parms.orvals[6] = 0;
 628         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 629                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 630         /* disable most machine checks */
 631         cr_parms.orvals[14] = 0;
 632         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 633                                  1 << 25 | 1 << 24);
 634
 635         smp_ctl_bit_callback(&cr_parms);
 636
 637         return 0;
 638 }
 639
 640 void __cpu_die(unsigned int cpu)
 641 {
 642         /* Wait until target cpu is down */
 643         while (!cpu_stopped(cpu))
 644                 cpu_relax();
 645         smp_free_lowcore(cpu);
 646         pr_info("Processor %d stopped\n", cpu);
 647 }
 648
 649 void cpu_die(void)
 650 {
 651         idle_task_exit();
 652         signal_processor(smp_processor_id(), sigp_stop);
 653         BUG();
 654         for (;;);
 655 }
 656
 657 #endif /* CONFIG_HOTPLUG_CPU */
 658
 659 void __init smp_prepare_cpus(unsigned int max_cpus)
 660 {
 661 #ifndef CONFIG_64BIT
 662         unsigned long save_area = 0;
 663 #endif
 664         unsigned long async_stack, panic_stack;
 665         struct _lowcore *lowcore;
 666         unsigned int cpu;
 667         int lc_order;
 668
 669         smp_detect_cpus();
 670
 671         /* request the 0x1201 emergency signal external interrupt */
 672         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 673                 panic("Couldn't request external interrupt 0x1201");
 674         print_cpu_info();
 675
 676         /* Reallocate current lowcore, but keep its contents. */
 677         lc_order = sizeof(long) == 8 ? 1 : 0;
 678         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 679         panic_stack = __get_free_page(GFP_KERNEL);
 680         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 681         BUG_ON(!lowcore || !panic_stack || !async_stack);
 682 #ifndef CONFIG_64BIT
 683         if (MACHINE_HAS_IEEE)
 684                 save_area = get_zeroed_page(GFP_KERNEL);
 685 #endif
 686         local_irq_disable();
 687         local_mcck_disable();
 688         lowcore_ptr[smp_processor_id()] = lowcore;
 689         *lowcore = S390_lowcore;
 690         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 691         lowcore->async_stack = async_stack + ASYNC_SIZE;
 692 #ifndef CONFIG_64BIT
 693         if (MACHINE_HAS_IEEE)
 694                 lowcore->extended_save_area_addr = (u32) save_area;
 695 #endif
 696         set_prefix((u32)(unsigned long) lowcore);
 697         local_mcck_enable();
 698         local_irq_enable();
 699 #ifdef CONFIG_64BIT
 700         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 701                 BUG();
 702 #endif
 703         for_each_possible_cpu(cpu)
 704                 if (cpu != smp_processor_id())
 705                         smp_create_idle(cpu);
 706 }
 707
 708 void __init smp_prepare_boot_cpu(void)
 709 {
 710         BUG_ON(smp_processor_id() != 0);
 711
 712         current_thread_info()->cpu = 0;
 713         cpu_set(0, cpu_present_map);
 714         cpu_set(0, cpu_online_map);
 715         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 716         current_set[0] = current;
 717         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 718         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 719 }
 720
 721 void __init smp_cpus_done(unsigned int max_cpus)
 722 {
 723 }
 724
 725 /*
 726  * the frequency of the profiling timer can be changed
 727  * by writing a multiplier value into /proc/profile.
 728  *
 729  * usually you want to run this on all CPUs ;)
 730  */
 731 int setup_profiling_timer(unsigned int multiplier)
 732 {
 733         return 0;
 734 }
 735
 736 #ifdef CONFIG_HOTPLUG_CPU
 737 static ssize_t cpu_configure_show(struct sys_device *dev,
 738                                 struct sysdev_attribute *attr, char *buf)
 739 {
 740         ssize_t count;
 741
 742         mutex_lock(&smp_cpu_state_mutex);
 743         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 744         mutex_unlock(&smp_cpu_state_mutex);
 745         return count;
 746 }
 747
 748 static ssize_t cpu_configure_store(struct sys_device *dev,
 749                                   struct sysdev_attribute *attr,
 750                                   const char *buf, size_t count)
 751 {
 752         int cpu = dev->id;
 753         int val, rc;
 754         char delim;
 755
 756         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 757                 return -EINVAL;
 758         if (val != 0 && val != 1)
 759                 return -EINVAL;
 760
 761         get_online_cpus();
 762         mutex_lock(&smp_cpu_state_mutex);
 763         rc = -EBUSY;
 764         if (cpu_online(cpu))
 765                 goto out;
 766         rc = 0;
 767         switch (val) {
 768         case 0:
 769                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 770                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 771                         if (!rc) {
 772                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 773                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 774                         }
 775                 }
 776                 break;
 777         case 1:
 778                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 779                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 780                         if (!rc) {
 781                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 782                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 783                         }
 784                 }
 785                 break;
 786         default:
 787                 break;
 788         }
 789 out:
 790         mutex_unlock(&smp_cpu_state_mutex);
 791         put_online_cpus();
 792         return rc ? rc : count;
 793 }
 794 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 795 #endif /* CONFIG_HOTPLUG_CPU */
 796
 797 static ssize_t cpu_polarization_show(struct sys_device *dev,
 798                                      struct sysdev_attribute *attr, char *buf)
 799 {
 800         int cpu = dev->id;
 801         ssize_t count;
 802
 803         mutex_lock(&smp_cpu_state_mutex);
 804         switch (smp_cpu_polarization[cpu]) {
 805         case POLARIZATION_HRZ:
 806                 count = sprintf(buf, "horizontal\n");
 807                 break;
 808         case POLARIZATION_VL:
 809                 count = sprintf(buf, "vertical:low\n");
 810                 break;
 811         case POLARIZATION_VM:
 812                 count = sprintf(buf, "vertical:medium\n");
 813                 break;
 814         case POLARIZATION_VH:
 815                 count = sprintf(buf, "vertical:high\n");
 816                 break;
 817         default:
 818                 count = sprintf(buf, "unknown\n");
 819                 break;
 820         }
 821         mutex_unlock(&smp_cpu_state_mutex);
 822         return count;
 823 }
 824 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 825
 826 static ssize_t show_cpu_address(struct sys_device *dev,
 827                                 struct sysdev_attribute *attr, char *buf)
 828 {
 829         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 830 }
 831 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 832
 833
 834 static struct attribute *cpu_common_attrs[] = {
 835 #ifdef CONFIG_HOTPLUG_CPU
 836         &attr_configure.attr,
 837 #endif
 838         &attr_address.attr,
 839         &attr_polarization.attr,
 840         NULL,
 841 };
 842
 843 static struct attribute_group cpu_common_attr_group = {
 844         .attrs = cpu_common_attrs,
 845 };
 846
 847 static ssize_t show_capability(struct sys_device *dev,
 848                                 struct sysdev_attribute *attr, char *buf)
 849 {
 850         unsigned int capability;
 851         int rc;
 852
 853         rc = get_cpu_capability(&capability);
 854         if (rc)
 855                 return rc;
 856         return sprintf(buf, "%u\n", capability);
 857 }
 858 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 859
 860 static ssize_t show_idle_count(struct sys_device *dev,
 861                                 struct sysdev_attribute *attr, char *buf)
 862 {
 863         struct s390_idle_data *idle;
 864         unsigned long long idle_count;
 865         unsigned int sequence;
 866
 867         idle = &per_cpu(s390_idle, dev->id);
 868 repeat:
 869         sequence = idle->sequence;
 870         smp_rmb();
 871         if (sequence & 1)
 872                 goto repeat;
 873         idle_count = idle->idle_count;
 874         if (idle->idle_enter)
 875                 idle_count++;
 876         smp_rmb();
 877         if (idle->sequence != sequence)
 878                 goto repeat;
 879         return sprintf(buf, "%llu\n", idle_count);
 880 }
 881 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 882
 883 static ssize_t show_idle_time(struct sys_device *dev,
 884                                 struct sysdev_attribute *attr, char *buf)
 885 {
 886         struct s390_idle_data *idle;
 887         unsigned long long now, idle_time, idle_enter;
 888         unsigned int sequence;
 889
 890         idle = &per_cpu(s390_idle, dev->id);
 891         now = get_clock();
 892 repeat:
 893         sequence = idle->sequence;
 894         smp_rmb();
 895         if (sequence & 1)
 896                 goto repeat;
 897         idle_time = idle->idle_time;
 898         idle_enter = idle->idle_enter;
 899         if (idle_enter != 0ULL && idle_enter < now)
 900                 idle_time += now - idle_enter;
 901         smp_rmb();
 902         if (idle->sequence != sequence)
 903                 goto repeat;
 904         return sprintf(buf, "%llu\n", idle_time >> 12);
 905 }
 906 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 907
 908 static struct attribute *cpu_online_attrs[] = {
 909         &attr_capability.attr,
 910         &attr_idle_count.attr,
 911         &attr_idle_time_us.attr,
 912         NULL,
 913 };
 914
 915 static struct attribute_group cpu_online_attr_group = {
 916         .attrs = cpu_online_attrs,
 917 };
 918
 919 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 920                                     unsigned long action, void *hcpu)
 921 {
 922         unsigned int cpu = (unsigned int)(long)hcpu;
 923         struct cpu *c = &per_cpu(cpu_devices, cpu);
 924         struct sys_device *s = &c->sysdev;
 925         struct s390_idle_data *idle;
 926
 927         switch (action) {
 928         case CPU_ONLINE:
 929         case CPU_ONLINE_FROZEN:
 930                 idle = &per_cpu(s390_idle, cpu);
 931                 memset(idle, 0, sizeof(struct s390_idle_data));
 932                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 933                         return NOTIFY_BAD;
 934                 break;
 935         case CPU_DEAD:
 936         case CPU_DEAD_FROZEN:
 937                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 938                 break;
 939         }
 940         return NOTIFY_OK;
 941 }
 942
 943 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 944         .notifier_call = smp_cpu_notify,
 945 };
 946
 947 static int __devinit smp_add_present_cpu(int cpu)
 948 {
 949         struct cpu *c = &per_cpu(cpu_devices, cpu);
 950         struct sys_device *s = &c->sysdev;
 951         int rc;
 952
 953         c->hotpluggable = 1;
 954         rc = register_cpu(c, cpu);
 955         if (rc)
 956                 goto out;
 957         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 958         if (rc)
 959                 goto out_cpu;
 960         if (!cpu_online(cpu))
 961                 goto out;
 962         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 963         if (!rc)
 964                 return 0;
 965         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 966 out_cpu:
 967 #ifdef CONFIG_HOTPLUG_CPU
 968         unregister_cpu(c);
 969 #endif
 970 out:
 971         return rc;
 972 }
 973
 974 #ifdef CONFIG_HOTPLUG_CPU
 975
 976 int __ref smp_rescan_cpus(void)
 977 {
 978         cpumask_t newcpus;
 979         int cpu;
 980         int rc;
 981
 982         get_online_cpus();
 983         mutex_lock(&smp_cpu_state_mutex);
 984         newcpus = cpu_present_map;
 985         rc = __smp_rescan_cpus();
 986         if (rc)
 987                 goto out;
 988         cpus_andnot(newcpus, cpu_present_map, newcpus);
 989         for_each_cpu_mask(cpu, newcpus) {
 990                 rc = smp_add_present_cpu(cpu);
 991                 if (rc)
 992                         cpu_clear(cpu, cpu_present_map);
 993         }
 994         rc = 0;
 995 out:
 996         mutex_unlock(&smp_cpu_state_mutex);
 997         put_online_cpus();
 998         if (!cpus_empty(newcpus))
 999                 topology_schedule_update();
1000         return rc;
1001 }
1002
1003 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1004                                   size_t count)
1005 {
1006         int rc;
1007
1008         rc = smp_rescan_cpus();
1009         return rc ? rc : count;
1010 }
1011 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1012 #endif /* CONFIG_HOTPLUG_CPU */
1013
1014 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1015 {
1016         ssize_t count;
1017
1018         mutex_lock(&smp_cpu_state_mutex);
1019         count = sprintf(buf, "%d\n", cpu_management);
1020         mutex_unlock(&smp_cpu_state_mutex);
1021         return count;
1022 }
1023
1024 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1025                                  size_t count)
1026 {
1027         int val, rc;
1028         char delim;
1029
1030         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1031                 return -EINVAL;
1032         if (val != 0 && val != 1)
1033                 return -EINVAL;
1034         rc = 0;
1035         get_online_cpus();
1036         mutex_lock(&smp_cpu_state_mutex);
1037         if (cpu_management == val)
1038                 goto out;
1039         rc = topology_set_cpu_management(val);
1040         if (!rc)
1041                 cpu_management = val;
1042 out:
1043         mutex_unlock(&smp_cpu_state_mutex);
1044         put_online_cpus();
1045         return rc ? rc : count;
1046 }
1047 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1048                          dispatching_store);
1049
1050 /*
1051  * If the resume kernel runs on another cpu than the suspended kernel,
1052  * we have to switch the cpu IDs in the logical map.
1053  */
1054 void smp_switch_boot_cpu_in_resume(u32 resume_phys_cpu_id,
1055                                    struct _lowcore *suspend_lowcore)
1056 {
1057         int cpu, suspend_cpu_id, resume_cpu_id;
1058         u32 suspend_phys_cpu_id;
1059
1060         suspend_phys_cpu_id = __cpu_logical_map[suspend_lowcore->cpu_nr];
1061         suspend_cpu_id = suspend_lowcore->cpu_nr;
1062
1063         for_each_present_cpu(cpu) {
1064                 if (__cpu_logical_map[cpu] == resume_phys_cpu_id) {
1065                         resume_cpu_id = cpu;
1066                         goto found;
1067                 }
1068         }
1069         panic("Could not find resume cpu in logical map.\n");
1070
1071 found:
1072         printk("Resume  cpu ID: %i/%i\n", resume_phys_cpu_id, resume_cpu_id);
1073         printk("Suspend cpu ID: %i/%i\n", suspend_phys_cpu_id, suspend_cpu_id);
1074
1075         __cpu_logical_map[resume_cpu_id] = suspend_phys_cpu_id;
1076         __cpu_logical_map[suspend_cpu_id] = resume_phys_cpu_id;
1077
1078         lowcore_ptr[suspend_cpu_id]->cpu_addr = resume_phys_cpu_id;
1079 }
1080
1081 u32 smp_get_phys_cpu_id(void)
1082 {
1083         return __cpu_logical_map[smp_processor_id()];
1084 }
1085
1086 static int __init topology_init(void)
1087 {
1088         int cpu;
1089         int rc;
1090
1091         register_cpu_notifier(&smp_cpu_nb);
1092
1093 #ifdef CONFIG_HOTPLUG_CPU
1094         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1095         if (rc)
1096                 return rc;
1097 #endif
1098         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1099         if (rc)
1100                 return rc;
1101         for_each_present_cpu(cpu) {
1102                 rc = smp_add_present_cpu(cpu);
1103                 if (rc)
1104                         return rc;
1105         }
1106         return 0;
1107 }
1108 subsys_initcall(topology_init);