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 void smp_send_stop(void)
  75 {
  76         int cpu, rc;
  77
  78         /* Disable all interrupts/machine checks */
  79         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  80         trace_hardirqs_off();
  81
  82         /* stop all processors */
  83         for_each_online_cpu(cpu) {
  84                 if (cpu == smp_processor_id())
  85                         continue;
  86                 do {
  87                         rc = signal_processor(cpu, sigp_stop);
  88                 } while (rc == sigp_busy);
  89
  90                 while (!smp_cpu_not_running(cpu))
  91                         cpu_relax();
  92         }
  93 }
  94
  95 /*
  96  * This is the main routine where commands issued by other
  97  * cpus are handled.
  98  */
  99
 100 static void do_ext_call_interrupt(__u16 code)
 101 {
 102         unsigned long bits;
 103
 104         /*
 105          * handle bit signal external calls
 106          *
 107          * For the ec_schedule signal we have to do nothing. All the work
 108          * is done automatically when we return from the interrupt.
 109          */
 110         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 111
 112         if (test_bit(ec_call_function, &bits))
 113                 generic_smp_call_function_interrupt();
 114
 115         if (test_bit(ec_call_function_single, &bits))
 116                 generic_smp_call_function_single_interrupt();
 117 }
 118
 119 /*
 120  * Send an external call sigp to another cpu and return without waiting
 121  * for its completion.
 122  */
 123 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 124 {
 125         /*
 126          * Set signaling bit in lowcore of target cpu and kick it
 127          */
 128         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 129         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 130                 udelay(10);
 131 }
 132
 133 void arch_send_call_function_ipi(cpumask_t mask)
 134 {
 135         int cpu;
 136
 137         for_each_cpu_mask(cpu, mask)
 138                 smp_ext_bitcall(cpu, ec_call_function);
 139 }
 140
 141 void arch_send_call_function_single_ipi(int cpu)
 142 {
 143         smp_ext_bitcall(cpu, ec_call_function_single);
 144 }
 145
 146 #ifndef CONFIG_64BIT
 147 /*
 148  * this function sends a 'purge tlb' signal to another CPU.
 149  */
 150 static void smp_ptlb_callback(void *info)
 151 {
 152         __tlb_flush_local();
 153 }
 154
 155 void smp_ptlb_all(void)
 156 {
 157         on_each_cpu(smp_ptlb_callback, NULL, 1);
 158 }
 159 EXPORT_SYMBOL(smp_ptlb_all);
 160 #endif /* ! CONFIG_64BIT */
 161
 162 /*
 163  * this function sends a 'reschedule' IPI to another CPU.
 164  * it goes straight through and wastes no time serializing
 165  * anything. Worst case is that we lose a reschedule ...
 166  */
 167 void smp_send_reschedule(int cpu)
 168 {
 169         smp_ext_bitcall(cpu, ec_schedule);
 170 }
 171
 172 /*
 173  * parameter area for the set/clear control bit callbacks
 174  */
 175 struct ec_creg_mask_parms {
 176         unsigned long orvals[16];
 177         unsigned long andvals[16];
 178 };
 179
 180 /*
 181  * callback for setting/clearing control bits
 182  */
 183 static void smp_ctl_bit_callback(void *info)
 184 {
 185         struct ec_creg_mask_parms *pp = info;
 186         unsigned long cregs[16];
 187         int i;
 188
 189         __ctl_store(cregs, 0, 15);
 190         for (i = 0; i <= 15; i++)
 191                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 192         __ctl_load(cregs, 0, 15);
 193 }
 194
 195 /*
 196  * Set a bit in a control register of all cpus
 197  */
 198 void smp_ctl_set_bit(int cr, int bit)
 199 {
 200         struct ec_creg_mask_parms parms;
 201
 202         memset(&parms.orvals, 0, sizeof(parms.orvals));
 203         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 204         parms.orvals[cr] = 1 << bit;
 205         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 206 }
 207 EXPORT_SYMBOL(smp_ctl_set_bit);
 208
 209 /*
 210  * Clear a bit in a control register of all cpus
 211  */
 212 void smp_ctl_clear_bit(int cr, int bit)
 213 {
 214         struct ec_creg_mask_parms parms;
 215
 216         memset(&parms.orvals, 0, sizeof(parms.orvals));
 217         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 218         parms.andvals[cr] = ~(1L << bit);
 219         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 220 }
 221 EXPORT_SYMBOL(smp_ctl_clear_bit);
 222
 223 /*
 224  * In early ipl state a temp. logically cpu number is needed, so the sigp
 225  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 226  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 227  */
 228 #define CPU_INIT_NO     1
 229
 230 #ifdef CONFIG_ZFCPDUMP
 231
 232 /*
 233  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 234  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 235  * save its prefix registers, since they get lost, when switching from 31 bit
 236  * to 64 bit.
 237  */
 238 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 239         __attribute__((__section__(".data")));
 240
 241 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 242 {
 243         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 244                 return;
 245         if (cpu >= NR_CPUS) {
 246                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 247                            "the dump\n", cpu, NR_CPUS - 1);
 248                 return;
 249         }
 250         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 251         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 252         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 253                sigp_busy)
 254                 cpu_relax();
 255         memcpy(zfcpdump_save_areas[cpu],
 256                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 257                SAVE_AREA_SIZE);
 258 #ifdef CONFIG_64BIT
 259         /* copy original prefix register */
 260         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 261 #endif
 262 }
 263
 264 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 265 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 266
 267 #else
 268
 269 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 270
 271 #endif /* CONFIG_ZFCPDUMP */
 272
 273 static int cpu_stopped(int cpu)
 274 {
 275         __u32 status;
 276
 277         /* Check for stopped state */
 278         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 279             sigp_status_stored) {
 280                 if (status & 0x40)
 281                         return 1;
 282         }
 283         return 0;
 284 }
 285
 286 static int cpu_known(int cpu_id)
 287 {
 288         int cpu;
 289
 290         for_each_present_cpu(cpu) {
 291                 if (__cpu_logical_map[cpu] == cpu_id)
 292                         return 1;
 293         }
 294         return 0;
 295 }
 296
 297 static int smp_rescan_cpus_sigp(cpumask_t avail)
 298 {
 299         int cpu_id, logical_cpu;
 300
 301         logical_cpu = cpumask_first(&avail);
 302         if (logical_cpu >= nr_cpu_ids)
 303                 return 0;
 304         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 305                 if (cpu_known(cpu_id))
 306                         continue;
 307                 __cpu_logical_map[logical_cpu] = cpu_id;
 308                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 309                 if (!cpu_stopped(logical_cpu))
 310                         continue;
 311                 cpu_set(logical_cpu, cpu_present_map);
 312                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 313                 logical_cpu = cpumask_next(logical_cpu, &avail);
 314                 if (logical_cpu >= nr_cpu_ids)
 315                         break;
 316         }
 317         return 0;
 318 }
 319
 320 static int smp_rescan_cpus_sclp(cpumask_t avail)
 321 {
 322         struct sclp_cpu_info *info;
 323         int cpu_id, logical_cpu, cpu;
 324         int rc;
 325
 326         logical_cpu = cpumask_first(&avail);
 327         if (logical_cpu >= nr_cpu_ids)
 328                 return 0;
 329         info = kmalloc(sizeof(*info), GFP_KERNEL);
 330         if (!info)
 331                 return -ENOMEM;
 332         rc = sclp_get_cpu_info(info);
 333         if (rc)
 334                 goto out;
 335         for (cpu = 0; cpu < info->combined; cpu++) {
 336                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 337                         continue;
 338                 cpu_id = info->cpu[cpu].address;
 339                 if (cpu_known(cpu_id))
 340                         continue;
 341                 __cpu_logical_map[logical_cpu] = cpu_id;
 342                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 343                 cpu_set(logical_cpu, cpu_present_map);
 344                 if (cpu >= info->configured)
 345                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 346                 else
 347                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 348                 logical_cpu = cpumask_next(logical_cpu, &avail);
 349                 if (logical_cpu >= nr_cpu_ids)
 350                         break;
 351         }
 352 out:
 353         kfree(info);
 354         return rc;
 355 }
 356
 357 static int __smp_rescan_cpus(void)
 358 {
 359         cpumask_t avail;
 360
 361         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 362         if (smp_use_sigp_detection)
 363                 return smp_rescan_cpus_sigp(avail);
 364         else
 365                 return smp_rescan_cpus_sclp(avail);
 366 }
 367
 368 static void __init smp_detect_cpus(void)
 369 {
 370         unsigned int cpu, c_cpus, s_cpus;
 371         struct sclp_cpu_info *info;
 372         u16 boot_cpu_addr, cpu_addr;
 373
 374         c_cpus = 1;
 375         s_cpus = 0;
 376         boot_cpu_addr = __cpu_logical_map[0];
 377         info = kmalloc(sizeof(*info), GFP_KERNEL);
 378         if (!info)
 379                 panic("smp_detect_cpus failed to allocate memory\n");
 380         /* Use sigp detection algorithm if sclp doesn't work. */
 381         if (sclp_get_cpu_info(info)) {
 382                 smp_use_sigp_detection = 1;
 383                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 384                         if (cpu == boot_cpu_addr)
 385                                 continue;
 386                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 387                         if (!cpu_stopped(CPU_INIT_NO))
 388                                 continue;
 389                         smp_get_save_area(c_cpus, cpu);
 390                         c_cpus++;
 391                 }
 392                 goto out;
 393         }
 394
 395         if (info->has_cpu_type) {
 396                 for (cpu = 0; cpu < info->combined; cpu++) {
 397                         if (info->cpu[cpu].address == boot_cpu_addr) {
 398                                 smp_cpu_type = info->cpu[cpu].type;
 399                                 break;
 400                         }
 401                 }
 402         }
 403
 404         for (cpu = 0; cpu < info->combined; cpu++) {
 405                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 406                         continue;
 407                 cpu_addr = info->cpu[cpu].address;
 408                 if (cpu_addr == boot_cpu_addr)
 409                         continue;
 410                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 411                 if (!cpu_stopped(CPU_INIT_NO)) {
 412                         s_cpus++;
 413                         continue;
 414                 }
 415                 smp_get_save_area(c_cpus, cpu_addr);
 416                 c_cpus++;
 417         }
 418 out:
 419         kfree(info);
 420         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 421         get_online_cpus();
 422         __smp_rescan_cpus();
 423         put_online_cpus();
 424 }
 425
 426 /*
 427  *      Activate a secondary processor.
 428  */
 429 int __cpuinit start_secondary(void *cpuvoid)
 430 {
 431         /* Setup the cpu */
 432         cpu_init();
 433         preempt_disable();
 434         /* Enable TOD clock interrupts on the secondary cpu. */
 435         init_cpu_timer();
 436         /* Enable cpu timer interrupts on the secondary cpu. */
 437         init_cpu_vtimer();
 438         /* Enable pfault pseudo page faults on this cpu. */
 439         pfault_init();
 440
 441         /* call cpu notifiers */
 442         notify_cpu_starting(smp_processor_id());
 443         /* Mark this cpu as online */
 444         ipi_call_lock();
 445         cpu_set(smp_processor_id(), cpu_online_map);
 446         ipi_call_unlock();
 447         /* Switch on interrupts */
 448         local_irq_enable();
 449         /* Print info about this processor */
 450         print_cpu_info();
 451         /* cpu_idle will call schedule for us */
 452         cpu_idle();
 453         return 0;
 454 }
 455
 456 static void __init smp_create_idle(unsigned int cpu)
 457 {
 458         struct task_struct *p;
 459
 460         /*
 461          *  don't care about the psw and regs settings since we'll never
 462          *  reschedule the forked task.
 463          */
 464         p = fork_idle(cpu);
 465         if (IS_ERR(p))
 466                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 467         current_set[cpu] = p;
 468 }
 469
 470 static int __cpuinit smp_alloc_lowcore(int cpu)
 471 {
 472         unsigned long async_stack, panic_stack;
 473         struct _lowcore *lowcore;
 474         int lc_order;
 475
 476         lc_order = sizeof(long) == 8 ? 1 : 0;
 477         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 478         if (!lowcore)
 479                 return -ENOMEM;
 480         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 481         panic_stack = __get_free_page(GFP_KERNEL);
 482         if (!panic_stack || !async_stack)
 483                 goto out;
 484         memcpy(lowcore, &S390_lowcore, 512);
 485         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 486         lowcore->async_stack = async_stack + ASYNC_SIZE;
 487         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 488
 489 #ifndef CONFIG_64BIT
 490         if (MACHINE_HAS_IEEE) {
 491                 unsigned long save_area;
 492
 493                 save_area = get_zeroed_page(GFP_KERNEL);
 494                 if (!save_area)
 495                         goto out;
 496                 lowcore->extended_save_area_addr = (u32) save_area;
 497         }
 498 #else
 499         if (vdso_alloc_per_cpu(cpu, lowcore))
 500                 goto out;
 501 #endif
 502         lowcore_ptr[cpu] = lowcore;
 503         return 0;
 504
 505 out:
 506         free_page(panic_stack);
 507         free_pages(async_stack, ASYNC_ORDER);
 508         free_pages((unsigned long) lowcore, lc_order);
 509         return -ENOMEM;
 510 }
 511
 512 static void smp_free_lowcore(int cpu)
 513 {
 514         struct _lowcore *lowcore;
 515         int lc_order;
 516
 517         lc_order = sizeof(long) == 8 ? 1 : 0;
 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 (!smp_cpu_not_running(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         int lc_order;
 664
 665         smp_detect_cpus();
 666
 667         /* request the 0x1201 emergency signal external interrupt */
 668         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 669                 panic("Couldn't request external interrupt 0x1201");
 670         print_cpu_info();
 671
 672         /* Reallocate current lowcore, but keep its contents. */
 673         lc_order = sizeof(long) == 8 ? 1 : 0;
 674         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 675         panic_stack = __get_free_page(GFP_KERNEL);
 676         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 677         BUG_ON(!lowcore || !panic_stack || !async_stack);
 678 #ifndef CONFIG_64BIT
 679         if (MACHINE_HAS_IEEE)
 680                 save_area = get_zeroed_page(GFP_KERNEL);
 681 #endif
 682         local_irq_disable();
 683         local_mcck_disable();
 684         lowcore_ptr[smp_processor_id()] = lowcore;
 685         *lowcore = S390_lowcore;
 686         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 687         lowcore->async_stack = async_stack + ASYNC_SIZE;
 688 #ifndef CONFIG_64BIT
 689         if (MACHINE_HAS_IEEE)
 690                 lowcore->extended_save_area_addr = (u32) save_area;
 691 #endif
 692         set_prefix((u32)(unsigned long) lowcore);
 693         local_mcck_enable();
 694         local_irq_enable();
 695 #ifdef CONFIG_64BIT
 696         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 697                 BUG();
 698 #endif
 699         for_each_possible_cpu(cpu)
 700                 if (cpu != smp_processor_id())
 701                         smp_create_idle(cpu);
 702 }
 703
 704 void __init smp_prepare_boot_cpu(void)
 705 {
 706         BUG_ON(smp_processor_id() != 0);
 707
 708         current_thread_info()->cpu = 0;
 709         cpu_set(0, cpu_present_map);
 710         cpu_set(0, cpu_online_map);
 711         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 712         current_set[0] = current;
 713         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 714         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 715 }
 716
 717 void __init smp_cpus_done(unsigned int max_cpus)
 718 {
 719 }
 720
 721 /*
 722  * the frequency of the profiling timer can be changed
 723  * by writing a multiplier value into /proc/profile.
 724  *
 725  * usually you want to run this on all CPUs ;)
 726  */
 727 int setup_profiling_timer(unsigned int multiplier)
 728 {
 729         return 0;
 730 }
 731
 732 #ifdef CONFIG_HOTPLUG_CPU
 733 static ssize_t cpu_configure_show(struct sys_device *dev,
 734                                 struct sysdev_attribute *attr, char *buf)
 735 {
 736         ssize_t count;
 737
 738         mutex_lock(&smp_cpu_state_mutex);
 739         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 740         mutex_unlock(&smp_cpu_state_mutex);
 741         return count;
 742 }
 743
 744 static ssize_t cpu_configure_store(struct sys_device *dev,
 745                                   struct sysdev_attribute *attr,
 746                                   const char *buf, size_t count)
 747 {
 748         int cpu = dev->id;
 749         int val, rc;
 750         char delim;
 751
 752         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 753                 return -EINVAL;
 754         if (val != 0 && val != 1)
 755                 return -EINVAL;
 756
 757         get_online_cpus();
 758         mutex_lock(&smp_cpu_state_mutex);
 759         rc = -EBUSY;
 760         if (cpu_online(cpu))
 761                 goto out;
 762         rc = 0;
 763         switch (val) {
 764         case 0:
 765                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 766                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 767                         if (!rc) {
 768                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 769                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 770                         }
 771                 }
 772                 break;
 773         case 1:
 774                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 775                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 776                         if (!rc) {
 777                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 778                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 779                         }
 780                 }
 781                 break;
 782         default:
 783                 break;
 784         }
 785 out:
 786         mutex_unlock(&smp_cpu_state_mutex);
 787         put_online_cpus();
 788         return rc ? rc : count;
 789 }
 790 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 791 #endif /* CONFIG_HOTPLUG_CPU */
 792
 793 static ssize_t cpu_polarization_show(struct sys_device *dev,
 794                                      struct sysdev_attribute *attr, char *buf)
 795 {
 796         int cpu = dev->id;
 797         ssize_t count;
 798
 799         mutex_lock(&smp_cpu_state_mutex);
 800         switch (smp_cpu_polarization[cpu]) {
 801         case POLARIZATION_HRZ:
 802                 count = sprintf(buf, "horizontal\n");
 803                 break;
 804         case POLARIZATION_VL:
 805                 count = sprintf(buf, "vertical:low\n");
 806                 break;
 807         case POLARIZATION_VM:
 808                 count = sprintf(buf, "vertical:medium\n");
 809                 break;
 810         case POLARIZATION_VH:
 811                 count = sprintf(buf, "vertical:high\n");
 812                 break;
 813         default:
 814                 count = sprintf(buf, "unknown\n");
 815                 break;
 816         }
 817         mutex_unlock(&smp_cpu_state_mutex);
 818         return count;
 819 }
 820 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 821
 822 static ssize_t show_cpu_address(struct sys_device *dev,
 823                                 struct sysdev_attribute *attr, char *buf)
 824 {
 825         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 826 }
 827 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 828
 829
 830 static struct attribute *cpu_common_attrs[] = {
 831 #ifdef CONFIG_HOTPLUG_CPU
 832         &attr_configure.attr,
 833 #endif
 834         &attr_address.attr,
 835         &attr_polarization.attr,
 836         NULL,
 837 };
 838
 839 static struct attribute_group cpu_common_attr_group = {
 840         .attrs = cpu_common_attrs,
 841 };
 842
 843 static ssize_t show_capability(struct sys_device *dev,
 844                                 struct sysdev_attribute *attr, char *buf)
 845 {
 846         unsigned int capability;
 847         int rc;
 848
 849         rc = get_cpu_capability(&capability);
 850         if (rc)
 851                 return rc;
 852         return sprintf(buf, "%u\n", capability);
 853 }
 854 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 855
 856 static ssize_t show_idle_count(struct sys_device *dev,
 857                                 struct sysdev_attribute *attr, char *buf)
 858 {
 859         struct s390_idle_data *idle;
 860         unsigned long long idle_count;
 861         unsigned int sequence;
 862
 863         idle = &per_cpu(s390_idle, dev->id);
 864 repeat:
 865         sequence = idle->sequence;
 866         smp_rmb();
 867         if (sequence & 1)
 868                 goto repeat;
 869         idle_count = idle->idle_count;
 870         if (idle->idle_enter)
 871                 idle_count++;
 872         smp_rmb();
 873         if (idle->sequence != sequence)
 874                 goto repeat;
 875         return sprintf(buf, "%llu\n", idle_count);
 876 }
 877 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 878
 879 static ssize_t show_idle_time(struct sys_device *dev,
 880                                 struct sysdev_attribute *attr, char *buf)
 881 {
 882         struct s390_idle_data *idle;
 883         unsigned long long now, idle_time, idle_enter;
 884         unsigned int sequence;
 885
 886         idle = &per_cpu(s390_idle, dev->id);
 887         now = get_clock();
 888 repeat:
 889         sequence = idle->sequence;
 890         smp_rmb();
 891         if (sequence & 1)
 892                 goto repeat;
 893         idle_time = idle->idle_time;
 894         idle_enter = idle->idle_enter;
 895         if (idle_enter != 0ULL && idle_enter < now)
 896                 idle_time += now - idle_enter;
 897         smp_rmb();
 898         if (idle->sequence != sequence)
 899                 goto repeat;
 900         return sprintf(buf, "%llu\n", idle_time >> 12);
 901 }
 902 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 903
 904 static struct attribute *cpu_online_attrs[] = {
 905         &attr_capability.attr,
 906         &attr_idle_count.attr,
 907         &attr_idle_time_us.attr,
 908         NULL,
 909 };
 910
 911 static struct attribute_group cpu_online_attr_group = {
 912         .attrs = cpu_online_attrs,
 913 };
 914
 915 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 916                                     unsigned long action, void *hcpu)
 917 {
 918         unsigned int cpu = (unsigned int)(long)hcpu;
 919         struct cpu *c = &per_cpu(cpu_devices, cpu);
 920         struct sys_device *s = &c->sysdev;
 921         struct s390_idle_data *idle;
 922
 923         switch (action) {
 924         case CPU_ONLINE:
 925         case CPU_ONLINE_FROZEN:
 926                 idle = &per_cpu(s390_idle, cpu);
 927                 memset(idle, 0, sizeof(struct s390_idle_data));
 928                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 929                         return NOTIFY_BAD;
 930                 break;
 931         case CPU_DEAD:
 932         case CPU_DEAD_FROZEN:
 933                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 934                 break;
 935         }
 936         return NOTIFY_OK;
 937 }
 938
 939 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 940         .notifier_call = smp_cpu_notify,
 941 };
 942
 943 static int __devinit smp_add_present_cpu(int cpu)
 944 {
 945         struct cpu *c = &per_cpu(cpu_devices, cpu);
 946         struct sys_device *s = &c->sysdev;
 947         int rc;
 948
 949         c->hotpluggable = 1;
 950         rc = register_cpu(c, cpu);
 951         if (rc)
 952                 goto out;
 953         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 954         if (rc)
 955                 goto out_cpu;
 956         if (!cpu_online(cpu))
 957                 goto out;
 958         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 959         if (!rc)
 960                 return 0;
 961         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 962 out_cpu:
 963 #ifdef CONFIG_HOTPLUG_CPU
 964         unregister_cpu(c);
 965 #endif
 966 out:
 967         return rc;
 968 }
 969
 970 #ifdef CONFIG_HOTPLUG_CPU
 971
 972 int __ref smp_rescan_cpus(void)
 973 {
 974         cpumask_t newcpus;
 975         int cpu;
 976         int rc;
 977
 978         get_online_cpus();
 979         mutex_lock(&smp_cpu_state_mutex);
 980         newcpus = cpu_present_map;
 981         rc = __smp_rescan_cpus();
 982         if (rc)
 983                 goto out;
 984         cpus_andnot(newcpus, cpu_present_map, newcpus);
 985         for_each_cpu_mask(cpu, newcpus) {
 986                 rc = smp_add_present_cpu(cpu);
 987                 if (rc)
 988                         cpu_clear(cpu, cpu_present_map);
 989         }
 990         rc = 0;
 991 out:
 992         mutex_unlock(&smp_cpu_state_mutex);
 993         put_online_cpus();
 994         if (!cpus_empty(newcpus))
 995                 topology_schedule_update();
 996         return rc;
 997 }
 998
 999 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1000                                   size_t count)
1001 {
1002         int rc;
1003
1004         rc = smp_rescan_cpus();
1005         return rc ? rc : count;
1006 }
1007 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1008 #endif /* CONFIG_HOTPLUG_CPU */
1009
1010 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1011 {
1012         ssize_t count;
1013
1014         mutex_lock(&smp_cpu_state_mutex);
1015         count = sprintf(buf, "%d\n", cpu_management);
1016         mutex_unlock(&smp_cpu_state_mutex);
1017         return count;
1018 }
1019
1020 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1021                                  size_t count)
1022 {
1023         int val, rc;
1024         char delim;
1025
1026         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1027                 return -EINVAL;
1028         if (val != 0 && val != 1)
1029                 return -EINVAL;
1030         rc = 0;
1031         get_online_cpus();
1032         mutex_lock(&smp_cpu_state_mutex);
1033         if (cpu_management == val)
1034                 goto out;
1035         rc = topology_set_cpu_management(val);
1036         if (!rc)
1037                 cpu_management = val;
1038 out:
1039         mutex_unlock(&smp_cpu_state_mutex);
1040         put_online_cpus();
1041         return rc ? rc : count;
1042 }
1043 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1044                          dispatching_store);
1045
1046 /*
1047  * If the resume kernel runs on another cpu than the suspended kernel,
1048  * we have to switch the cpu IDs in the logical map.
1049  */
1050 void smp_switch_boot_cpu_in_resume(u32 resume_phys_cpu_id,
1051                                    struct _lowcore *suspend_lowcore)
1052 {
1053         int cpu, suspend_cpu_id, resume_cpu_id;
1054         u32 suspend_phys_cpu_id;
1055
1056         suspend_phys_cpu_id = __cpu_logical_map[suspend_lowcore->cpu_nr];
1057         suspend_cpu_id = suspend_lowcore->cpu_nr;
1058
1059         for_each_present_cpu(cpu) {
1060                 if (__cpu_logical_map[cpu] == resume_phys_cpu_id) {
1061                         resume_cpu_id = cpu;
1062                         goto found;
1063                 }
1064         }
1065         panic("Could not find resume cpu in logical map.\n");
1066
1067 found:
1068         printk("Resume  cpu ID: %i/%i\n", resume_phys_cpu_id, resume_cpu_id);
1069         printk("Suspend cpu ID: %i/%i\n", suspend_phys_cpu_id, suspend_cpu_id);
1070
1071         __cpu_logical_map[resume_cpu_id] = suspend_phys_cpu_id;
1072         __cpu_logical_map[suspend_cpu_id] = resume_phys_cpu_id;
1073
1074         lowcore_ptr[suspend_cpu_id]->cpu_addr = resume_phys_cpu_id;
1075 }
1076
1077 u32 smp_get_phys_cpu_id(void)
1078 {
1079         return __cpu_logical_map[smp_processor_id()];
1080 }
1081
1082 static int __init topology_init(void)
1083 {
1084         int cpu;
1085         int rc;
1086
1087         register_cpu_notifier(&smp_cpu_nb);
1088
1089 #ifdef CONFIG_HOTPLUG_CPU
1090         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1091         if (rc)
1092                 return rc;
1093 #endif
1094         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1095         if (rc)
1096                 return rc;
1097         for_each_present_cpu(cpu) {
1098                 rc = smp_add_present_cpu(cpu);
1099                 if (rc)
1100                         return rc;
1101         }
1102         return 0;
1103 }
1104 subsys_initcall(topology_init);