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 /* logical cpu to cpu address */
  56 int __cpu_logical_map[NR_CPUS];
  57
  58 static struct task_struct *current_set[NR_CPUS];
  59
  60 static u8 smp_cpu_type;
  61 static int smp_use_sigp_detection;
  62
  63 enum s390_cpu_state {
  64         CPU_STATE_STANDBY,
  65         CPU_STATE_CONFIGURED,
  66 };
  67
  68 DEFINE_MUTEX(smp_cpu_state_mutex);
  69 int smp_cpu_polarization[NR_CPUS];
  70 static int smp_cpu_state[NR_CPUS];
  71 static int cpu_management;
  72
  73 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  74
  75 static void smp_ext_bitcall(int, ec_bit_sig);
  76
  77 static int cpu_stopped(int cpu)
  78 {
  79         __u32 status;
  80
  81         switch (signal_processor_ps(&status, 0, cpu, sigp_sense)) {
  82         case sigp_status_stored:
  83                 /* Check for stopped and check stop state */
  84                 if (status & 0x50)
  85                         return 1;
  86                 break;
  87         default:
  88                 break;
  89         }
  90         return 0;
  91 }
  92
  93 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
  94 {
  95         struct _lowcore *lc, *current_lc;
  96         struct stack_frame *sf;
  97         struct pt_regs *regs;
  98         unsigned long sp;
  99
 100         if (smp_processor_id() == 0)
 101                 func(data);
 102         __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
 103         /* Disable lowcore protection */
 104         __ctl_clear_bit(0, 28);
 105         current_lc = lowcore_ptr[smp_processor_id()];
 106         lc = lowcore_ptr[0];
 107         if (!lc)
 108                 lc = current_lc;
 109         lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 110         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
 111         if (!cpu_online(0))
 112                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
 113         while (signal_processor(0, sigp_stop_and_store_status) == sigp_busy)
 114                 cpu_relax();
 115         sp = lc->panic_stack;
 116         sp -= sizeof(struct pt_regs);
 117         regs = (struct pt_regs *) sp;
 118         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
 119         memcpy(&regs->psw, &current_lc->st_status_fixed_logout, sizeof(psw_t));
 120         sp -= STACK_FRAME_OVERHEAD;
 121         sf = (struct stack_frame *) sp;
 122         sf->back_chain = regs->gprs[15];
 123         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 124 }
 125
 126 void smp_send_stop(void)
 127 {
 128         int cpu, rc;
 129
 130         /* Disable all interrupts/machine checks */
 131         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 132         trace_hardirqs_off();
 133
 134         /* stop all processors */
 135         for_each_online_cpu(cpu) {
 136                 if (cpu == smp_processor_id())
 137                         continue;
 138                 do {
 139                         rc = signal_processor(cpu, sigp_stop);
 140                 } while (rc == sigp_busy);
 141
 142                 while (!cpu_stopped(cpu))
 143                         cpu_relax();
 144         }
 145 }
 146
 147 /*
 148  * This is the main routine where commands issued by other
 149  * cpus are handled.
 150  */
 151
 152 static void do_ext_call_interrupt(__u16 code)
 153 {
 154         unsigned long bits;
 155
 156         /*
 157          * handle bit signal external calls
 158          *
 159          * For the ec_schedule signal we have to do nothing. All the work
 160          * is done automatically when we return from the interrupt.
 161          */
 162         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 163
 164         if (test_bit(ec_call_function, &bits))
 165                 generic_smp_call_function_interrupt();
 166
 167         if (test_bit(ec_call_function_single, &bits))
 168                 generic_smp_call_function_single_interrupt();
 169 }
 170
 171 /*
 172  * Send an external call sigp to another cpu and return without waiting
 173  * for its completion.
 174  */
 175 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 176 {
 177         /*
 178          * Set signaling bit in lowcore of target cpu and kick it
 179          */
 180         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 181         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 182                 udelay(10);
 183 }
 184
 185 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 186 {
 187         int cpu;
 188
 189         for_each_cpu(cpu, mask)
 190                 smp_ext_bitcall(cpu, ec_call_function);
 191 }
 192
 193 void arch_send_call_function_single_ipi(int cpu)
 194 {
 195         smp_ext_bitcall(cpu, ec_call_function_single);
 196 }
 197
 198 #ifndef CONFIG_64BIT
 199 /*
 200  * this function sends a 'purge tlb' signal to another CPU.
 201  */
 202 static void smp_ptlb_callback(void *info)
 203 {
 204         __tlb_flush_local();
 205 }
 206
 207 void smp_ptlb_all(void)
 208 {
 209         on_each_cpu(smp_ptlb_callback, NULL, 1);
 210 }
 211 EXPORT_SYMBOL(smp_ptlb_all);
 212 #endif /* ! CONFIG_64BIT */
 213
 214 /*
 215  * this function sends a 'reschedule' IPI to another CPU.
 216  * it goes straight through and wastes no time serializing
 217  * anything. Worst case is that we lose a reschedule ...
 218  */
 219 void smp_send_reschedule(int cpu)
 220 {
 221         smp_ext_bitcall(cpu, ec_schedule);
 222 }
 223
 224 /*
 225  * parameter area for the set/clear control bit callbacks
 226  */
 227 struct ec_creg_mask_parms {
 228         unsigned long orvals[16];
 229         unsigned long andvals[16];
 230 };
 231
 232 /*
 233  * callback for setting/clearing control bits
 234  */
 235 static void smp_ctl_bit_callback(void *info)
 236 {
 237         struct ec_creg_mask_parms *pp = info;
 238         unsigned long cregs[16];
 239         int i;
 240
 241         __ctl_store(cregs, 0, 15);
 242         for (i = 0; i <= 15; i++)
 243                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 244         __ctl_load(cregs, 0, 15);
 245 }
 246
 247 /*
 248  * Set a bit in a control register of all cpus
 249  */
 250 void smp_ctl_set_bit(int cr, int bit)
 251 {
 252         struct ec_creg_mask_parms parms;
 253
 254         memset(&parms.orvals, 0, sizeof(parms.orvals));
 255         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 256         parms.orvals[cr] = 1 << bit;
 257         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 258 }
 259 EXPORT_SYMBOL(smp_ctl_set_bit);
 260
 261 /*
 262  * Clear a bit in a control register of all cpus
 263  */
 264 void smp_ctl_clear_bit(int cr, int bit)
 265 {
 266         struct ec_creg_mask_parms parms;
 267
 268         memset(&parms.orvals, 0, sizeof(parms.orvals));
 269         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 270         parms.andvals[cr] = ~(1L << bit);
 271         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 272 }
 273 EXPORT_SYMBOL(smp_ctl_clear_bit);
 274
 275 /*
 276  * In early ipl state a temp. logically cpu number is needed, so the sigp
 277  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 278  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 279  */
 280 #define CPU_INIT_NO     1
 281
 282 #ifdef CONFIG_ZFCPDUMP
 283
 284 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 285 {
 286         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 287                 return;
 288         if (cpu >= NR_CPUS) {
 289                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 290                            "the dump\n", cpu, NR_CPUS - 1);
 291                 return;
 292         }
 293         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 294         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 295         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 296                sigp_busy)
 297                 cpu_relax();
 298         memcpy(zfcpdump_save_areas[cpu],
 299                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 300                sizeof(struct save_area));
 301 }
 302
 303 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 304 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 305
 306 #else
 307
 308 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 309
 310 #endif /* CONFIG_ZFCPDUMP */
 311
 312 static int cpu_known(int cpu_id)
 313 {
 314         int cpu;
 315
 316         for_each_present_cpu(cpu) {
 317                 if (__cpu_logical_map[cpu] == cpu_id)
 318                         return 1;
 319         }
 320         return 0;
 321 }
 322
 323 static int smp_rescan_cpus_sigp(cpumask_t avail)
 324 {
 325         int cpu_id, logical_cpu;
 326
 327         logical_cpu = cpumask_first(&avail);
 328         if (logical_cpu >= nr_cpu_ids)
 329                 return 0;
 330         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 331                 if (cpu_known(cpu_id))
 332                         continue;
 333                 __cpu_logical_map[logical_cpu] = cpu_id;
 334                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 335                 if (!cpu_stopped(logical_cpu))
 336                         continue;
 337                 cpu_set(logical_cpu, cpu_present_map);
 338                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 339                 logical_cpu = cpumask_next(logical_cpu, &avail);
 340                 if (logical_cpu >= nr_cpu_ids)
 341                         break;
 342         }
 343         return 0;
 344 }
 345
 346 static int smp_rescan_cpus_sclp(cpumask_t avail)
 347 {
 348         struct sclp_cpu_info *info;
 349         int cpu_id, logical_cpu, cpu;
 350         int rc;
 351
 352         logical_cpu = cpumask_first(&avail);
 353         if (logical_cpu >= nr_cpu_ids)
 354                 return 0;
 355         info = kmalloc(sizeof(*info), GFP_KERNEL);
 356         if (!info)
 357                 return -ENOMEM;
 358         rc = sclp_get_cpu_info(info);
 359         if (rc)
 360                 goto out;
 361         for (cpu = 0; cpu < info->combined; cpu++) {
 362                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 363                         continue;
 364                 cpu_id = info->cpu[cpu].address;
 365                 if (cpu_known(cpu_id))
 366                         continue;
 367                 __cpu_logical_map[logical_cpu] = cpu_id;
 368                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 369                 cpu_set(logical_cpu, cpu_present_map);
 370                 if (cpu >= info->configured)
 371                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 372                 else
 373                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 374                 logical_cpu = cpumask_next(logical_cpu, &avail);
 375                 if (logical_cpu >= nr_cpu_ids)
 376                         break;
 377         }
 378 out:
 379         kfree(info);
 380         return rc;
 381 }
 382
 383 static int __smp_rescan_cpus(void)
 384 {
 385         cpumask_t avail;
 386
 387         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 388         if (smp_use_sigp_detection)
 389                 return smp_rescan_cpus_sigp(avail);
 390         else
 391                 return smp_rescan_cpus_sclp(avail);
 392 }
 393
 394 static void __init smp_detect_cpus(void)
 395 {
 396         unsigned int cpu, c_cpus, s_cpus;
 397         struct sclp_cpu_info *info;
 398         u16 boot_cpu_addr, cpu_addr;
 399
 400         c_cpus = 1;
 401         s_cpus = 0;
 402         boot_cpu_addr = __cpu_logical_map[0];
 403         info = kmalloc(sizeof(*info), GFP_KERNEL);
 404         if (!info)
 405                 panic("smp_detect_cpus failed to allocate memory\n");
 406         /* Use sigp detection algorithm if sclp doesn't work. */
 407         if (sclp_get_cpu_info(info)) {
 408                 smp_use_sigp_detection = 1;
 409                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 410                         if (cpu == boot_cpu_addr)
 411                                 continue;
 412                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 413                         if (!cpu_stopped(CPU_INIT_NO))
 414                                 continue;
 415                         smp_get_save_area(c_cpus, cpu);
 416                         c_cpus++;
 417                 }
 418                 goto out;
 419         }
 420
 421         if (info->has_cpu_type) {
 422                 for (cpu = 0; cpu < info->combined; cpu++) {
 423                         if (info->cpu[cpu].address == boot_cpu_addr) {
 424                                 smp_cpu_type = info->cpu[cpu].type;
 425                                 break;
 426                         }
 427                 }
 428         }
 429
 430         for (cpu = 0; cpu < info->combined; cpu++) {
 431                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 432                         continue;
 433                 cpu_addr = info->cpu[cpu].address;
 434                 if (cpu_addr == boot_cpu_addr)
 435                         continue;
 436                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 437                 if (!cpu_stopped(CPU_INIT_NO)) {
 438                         s_cpus++;
 439                         continue;
 440                 }
 441                 smp_get_save_area(c_cpus, cpu_addr);
 442                 c_cpus++;
 443         }
 444 out:
 445         kfree(info);
 446         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 447         get_online_cpus();
 448         __smp_rescan_cpus();
 449         put_online_cpus();
 450 }
 451
 452 /*
 453  *      Activate a secondary processor.
 454  */
 455 int __cpuinit start_secondary(void *cpuvoid)
 456 {
 457         /* Setup the cpu */
 458         cpu_init();
 459         preempt_disable();
 460         /* Enable TOD clock interrupts on the secondary cpu. */
 461         init_cpu_timer();
 462         /* Enable cpu timer interrupts on the secondary cpu. */
 463         init_cpu_vtimer();
 464         /* Enable pfault pseudo page faults on this cpu. */
 465         pfault_init();
 466
 467         /* call cpu notifiers */
 468         notify_cpu_starting(smp_processor_id());
 469         /* Mark this cpu as online */
 470         ipi_call_lock();
 471         cpu_set(smp_processor_id(), cpu_online_map);
 472         ipi_call_unlock();
 473         /* Switch on interrupts */
 474         local_irq_enable();
 475         /* Print info about this processor */
 476         print_cpu_info();
 477         /* cpu_idle will call schedule for us */
 478         cpu_idle();
 479         return 0;
 480 }
 481
 482 static void __init smp_create_idle(unsigned int cpu)
 483 {
 484         struct task_struct *p;
 485
 486         /*
 487          *  don't care about the psw and regs settings since we'll never
 488          *  reschedule the forked task.
 489          */
 490         p = fork_idle(cpu);
 491         if (IS_ERR(p))
 492                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 493         current_set[cpu] = p;
 494 }
 495
 496 static int __cpuinit smp_alloc_lowcore(int cpu)
 497 {
 498         unsigned long async_stack, panic_stack;
 499         struct _lowcore *lowcore;
 500
 501         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 502         if (!lowcore)
 503                 return -ENOMEM;
 504         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 505         panic_stack = __get_free_page(GFP_KERNEL);
 506         if (!panic_stack || !async_stack)
 507                 goto out;
 508         memcpy(lowcore, &S390_lowcore, 512);
 509         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 510         lowcore->async_stack = async_stack + ASYNC_SIZE;
 511         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 512
 513 #ifndef CONFIG_64BIT
 514         if (MACHINE_HAS_IEEE) {
 515                 unsigned long save_area;
 516
 517                 save_area = get_zeroed_page(GFP_KERNEL);
 518                 if (!save_area)
 519                         goto out;
 520                 lowcore->extended_save_area_addr = (u32) save_area;
 521         }
 522 #else
 523         if (vdso_alloc_per_cpu(cpu, lowcore))
 524                 goto out;
 525 #endif
 526         lowcore_ptr[cpu] = lowcore;
 527         return 0;
 528
 529 out:
 530         free_page(panic_stack);
 531         free_pages(async_stack, ASYNC_ORDER);
 532         free_pages((unsigned long) lowcore, LC_ORDER);
 533         return -ENOMEM;
 534 }
 535
 536 static void smp_free_lowcore(int cpu)
 537 {
 538         struct _lowcore *lowcore;
 539
 540         lowcore = lowcore_ptr[cpu];
 541 #ifndef CONFIG_64BIT
 542         if (MACHINE_HAS_IEEE)
 543                 free_page((unsigned long) lowcore->extended_save_area_addr);
 544 #else
 545         vdso_free_per_cpu(cpu, lowcore);
 546 #endif
 547         free_page(lowcore->panic_stack - PAGE_SIZE);
 548         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 549         free_pages((unsigned long) lowcore, LC_ORDER);
 550         lowcore_ptr[cpu] = NULL;
 551 }
 552
 553 /* Upping and downing of CPUs */
 554 int __cpuinit __cpu_up(unsigned int cpu)
 555 {
 556         struct task_struct *idle;
 557         struct _lowcore *cpu_lowcore;
 558         struct stack_frame *sf;
 559         sigp_ccode ccode;
 560         u32 lowcore;
 561
 562         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 563                 return -EIO;
 564         if (smp_alloc_lowcore(cpu))
 565                 return -ENOMEM;
 566         do {
 567                 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
 568                 if (ccode == sigp_busy)
 569                         udelay(10);
 570                 if (ccode == sigp_not_operational)
 571                         goto err_out;
 572         } while (ccode == sigp_busy);
 573
 574         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 575         while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 576                 udelay(10);
 577
 578         idle = current_set[cpu];
 579         cpu_lowcore = lowcore_ptr[cpu];
 580         cpu_lowcore->kernel_stack = (unsigned long)
 581                 task_stack_page(idle) + THREAD_SIZE;
 582         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 583         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 584                                      - sizeof(struct pt_regs)
 585                                      - sizeof(struct stack_frame));
 586         memset(sf, 0, sizeof(struct stack_frame));
 587         sf->gprs[9] = (unsigned long) sf;
 588         cpu_lowcore->save_area[15] = (unsigned long) sf;
 589         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 590         asm volatile(
 591                 "       stam    0,15,0(%0)"
 592                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 593         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 594         cpu_lowcore->current_task = (unsigned long) idle;
 595         cpu_lowcore->cpu_nr = cpu;
 596         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 597         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 598         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 599         eieio();
 600
 601         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 602                 udelay(10);
 603
 604         while (!cpu_online(cpu))
 605                 cpu_relax();
 606         return 0;
 607
 608 err_out:
 609         smp_free_lowcore(cpu);
 610         return -EIO;
 611 }
 612
 613 static int __init setup_possible_cpus(char *s)
 614 {
 615         int pcpus, cpu;
 616
 617         pcpus = simple_strtoul(s, NULL, 0);
 618         init_cpu_possible(cpumask_of(0));
 619         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 620                 set_cpu_possible(cpu, true);
 621         return 0;
 622 }
 623 early_param("possible_cpus", setup_possible_cpus);
 624
 625 #ifdef CONFIG_HOTPLUG_CPU
 626
 627 int __cpu_disable(void)
 628 {
 629         struct ec_creg_mask_parms cr_parms;
 630         int cpu = smp_processor_id();
 631
 632         cpu_clear(cpu, cpu_online_map);
 633
 634         /* Disable pfault pseudo page faults on this cpu. */
 635         pfault_fini();
 636
 637         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 638         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 639
 640         /* disable all external interrupts */
 641         cr_parms.orvals[0] = 0;
 642         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 643                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 644         /* disable all I/O interrupts */
 645         cr_parms.orvals[6] = 0;
 646         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 647                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 648         /* disable most machine checks */
 649         cr_parms.orvals[14] = 0;
 650         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 651                                  1 << 25 | 1 << 24);
 652
 653         smp_ctl_bit_callback(&cr_parms);
 654
 655         return 0;
 656 }
 657
 658 void __cpu_die(unsigned int cpu)
 659 {
 660         /* Wait until target cpu is down */
 661         while (!cpu_stopped(cpu))
 662                 cpu_relax();
 663         while (signal_processor_p(0, cpu, sigp_set_prefix) == sigp_busy)
 664                 udelay(10);
 665         smp_free_lowcore(cpu);
 666         pr_info("Processor %d stopped\n", cpu);
 667 }
 668
 669 void cpu_die(void)
 670 {
 671         idle_task_exit();
 672         while (signal_processor(smp_processor_id(), sigp_stop) == sigp_busy)
 673                 cpu_relax();
 674         for (;;);
 675 }
 676
 677 #endif /* CONFIG_HOTPLUG_CPU */
 678
 679 void __init smp_prepare_cpus(unsigned int max_cpus)
 680 {
 681 #ifndef CONFIG_64BIT
 682         unsigned long save_area = 0;
 683 #endif
 684         unsigned long async_stack, panic_stack;
 685         struct _lowcore *lowcore;
 686         unsigned int cpu;
 687
 688         smp_detect_cpus();
 689
 690         /* request the 0x1201 emergency signal external interrupt */
 691         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 692                 panic("Couldn't request external interrupt 0x1201");
 693         print_cpu_info();
 694
 695         /* Reallocate current lowcore, but keep its contents. */
 696         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 697         panic_stack = __get_free_page(GFP_KERNEL);
 698         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 699         BUG_ON(!lowcore || !panic_stack || !async_stack);
 700 #ifndef CONFIG_64BIT
 701         if (MACHINE_HAS_IEEE)
 702                 save_area = get_zeroed_page(GFP_KERNEL);
 703 #endif
 704         local_irq_disable();
 705         local_mcck_disable();
 706         lowcore_ptr[smp_processor_id()] = lowcore;
 707         *lowcore = S390_lowcore;
 708         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 709         lowcore->async_stack = async_stack + ASYNC_SIZE;
 710 #ifndef CONFIG_64BIT
 711         if (MACHINE_HAS_IEEE)
 712                 lowcore->extended_save_area_addr = (u32) save_area;
 713 #endif
 714         set_prefix((u32)(unsigned long) lowcore);
 715         local_mcck_enable();
 716         local_irq_enable();
 717 #ifdef CONFIG_64BIT
 718         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 719                 BUG();
 720 #endif
 721         for_each_possible_cpu(cpu)
 722                 if (cpu != smp_processor_id())
 723                         smp_create_idle(cpu);
 724 }
 725
 726 void __init smp_prepare_boot_cpu(void)
 727 {
 728         BUG_ON(smp_processor_id() != 0);
 729
 730         current_thread_info()->cpu = 0;
 731         cpu_set(0, cpu_present_map);
 732         cpu_set(0, cpu_online_map);
 733         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 734         current_set[0] = current;
 735         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 736         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 737 }
 738
 739 void __init smp_cpus_done(unsigned int max_cpus)
 740 {
 741 }
 742
 743 void __init smp_setup_processor_id(void)
 744 {
 745         S390_lowcore.cpu_nr = 0;
 746         __cpu_logical_map[0] = stap();
 747 }
 748
 749 /*
 750  * the frequency of the profiling timer can be changed
 751  * by writing a multiplier value into /proc/profile.
 752  *
 753  * usually you want to run this on all CPUs ;)
 754  */
 755 int setup_profiling_timer(unsigned int multiplier)
 756 {
 757         return 0;
 758 }
 759
 760 #ifdef CONFIG_HOTPLUG_CPU
 761 static ssize_t cpu_configure_show(struct sys_device *dev,
 762                                 struct sysdev_attribute *attr, char *buf)
 763 {
 764         ssize_t count;
 765
 766         mutex_lock(&smp_cpu_state_mutex);
 767         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 768         mutex_unlock(&smp_cpu_state_mutex);
 769         return count;
 770 }
 771
 772 static ssize_t cpu_configure_store(struct sys_device *dev,
 773                                   struct sysdev_attribute *attr,
 774                                   const char *buf, size_t count)
 775 {
 776         int cpu = dev->id;
 777         int val, rc;
 778         char delim;
 779
 780         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 781                 return -EINVAL;
 782         if (val != 0 && val != 1)
 783                 return -EINVAL;
 784
 785         get_online_cpus();
 786         mutex_lock(&smp_cpu_state_mutex);
 787         rc = -EBUSY;
 788         /* disallow configuration changes of online cpus and cpu 0 */
 789         if (cpu_online(cpu) || cpu == 0)
 790                 goto out;
 791         rc = 0;
 792         switch (val) {
 793         case 0:
 794                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 795                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 796                         if (!rc) {
 797                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 798                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 799                         }
 800                 }
 801                 break;
 802         case 1:
 803                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 804                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 805                         if (!rc) {
 806                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 807                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 808                         }
 809                 }
 810                 break;
 811         default:
 812                 break;
 813         }
 814 out:
 815         mutex_unlock(&smp_cpu_state_mutex);
 816         put_online_cpus();
 817         return rc ? rc : count;
 818 }
 819 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 820 #endif /* CONFIG_HOTPLUG_CPU */
 821
 822 static ssize_t cpu_polarization_show(struct sys_device *dev,
 823                                      struct sysdev_attribute *attr, char *buf)
 824 {
 825         int cpu = dev->id;
 826         ssize_t count;
 827
 828         mutex_lock(&smp_cpu_state_mutex);
 829         switch (smp_cpu_polarization[cpu]) {
 830         case POLARIZATION_HRZ:
 831                 count = sprintf(buf, "horizontal\n");
 832                 break;
 833         case POLARIZATION_VL:
 834                 count = sprintf(buf, "vertical:low\n");
 835                 break;
 836         case POLARIZATION_VM:
 837                 count = sprintf(buf, "vertical:medium\n");
 838                 break;
 839         case POLARIZATION_VH:
 840                 count = sprintf(buf, "vertical:high\n");
 841                 break;
 842         default:
 843                 count = sprintf(buf, "unknown\n");
 844                 break;
 845         }
 846         mutex_unlock(&smp_cpu_state_mutex);
 847         return count;
 848 }
 849 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 850
 851 static ssize_t show_cpu_address(struct sys_device *dev,
 852                                 struct sysdev_attribute *attr, char *buf)
 853 {
 854         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 855 }
 856 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 857
 858
 859 static struct attribute *cpu_common_attrs[] = {
 860 #ifdef CONFIG_HOTPLUG_CPU
 861         &attr_configure.attr,
 862 #endif
 863         &attr_address.attr,
 864         &attr_polarization.attr,
 865         NULL,
 866 };
 867
 868 static struct attribute_group cpu_common_attr_group = {
 869         .attrs = cpu_common_attrs,
 870 };
 871
 872 static ssize_t show_capability(struct sys_device *dev,
 873                                 struct sysdev_attribute *attr, char *buf)
 874 {
 875         unsigned int capability;
 876         int rc;
 877
 878         rc = get_cpu_capability(&capability);
 879         if (rc)
 880                 return rc;
 881         return sprintf(buf, "%u\n", capability);
 882 }
 883 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 884
 885 static ssize_t show_idle_count(struct sys_device *dev,
 886                                 struct sysdev_attribute *attr, char *buf)
 887 {
 888         struct s390_idle_data *idle;
 889         unsigned long long idle_count;
 890         unsigned int sequence;
 891
 892         idle = &per_cpu(s390_idle, dev->id);
 893 repeat:
 894         sequence = idle->sequence;
 895         smp_rmb();
 896         if (sequence & 1)
 897                 goto repeat;
 898         idle_count = idle->idle_count;
 899         if (idle->idle_enter)
 900                 idle_count++;
 901         smp_rmb();
 902         if (idle->sequence != sequence)
 903                 goto repeat;
 904         return sprintf(buf, "%llu\n", idle_count);
 905 }
 906 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 907
 908 static ssize_t show_idle_time(struct sys_device *dev,
 909                                 struct sysdev_attribute *attr, char *buf)
 910 {
 911         struct s390_idle_data *idle;
 912         unsigned long long now, idle_time, idle_enter;
 913         unsigned int sequence;
 914
 915         idle = &per_cpu(s390_idle, dev->id);
 916         now = get_clock();
 917 repeat:
 918         sequence = idle->sequence;
 919         smp_rmb();
 920         if (sequence & 1)
 921                 goto repeat;
 922         idle_time = idle->idle_time;
 923         idle_enter = idle->idle_enter;
 924         if (idle_enter != 0ULL && idle_enter < now)
 925                 idle_time += now - idle_enter;
 926         smp_rmb();
 927         if (idle->sequence != sequence)
 928                 goto repeat;
 929         return sprintf(buf, "%llu\n", idle_time >> 12);
 930 }
 931 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 932
 933 static struct attribute *cpu_online_attrs[] = {
 934         &attr_capability.attr,
 935         &attr_idle_count.attr,
 936         &attr_idle_time_us.attr,
 937         NULL,
 938 };
 939
 940 static struct attribute_group cpu_online_attr_group = {
 941         .attrs = cpu_online_attrs,
 942 };
 943
 944 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 945                                     unsigned long action, void *hcpu)
 946 {
 947         unsigned int cpu = (unsigned int)(long)hcpu;
 948         struct cpu *c = &per_cpu(cpu_devices, cpu);
 949         struct sys_device *s = &c->sysdev;
 950         struct s390_idle_data *idle;
 951
 952         switch (action) {
 953         case CPU_ONLINE:
 954         case CPU_ONLINE_FROZEN:
 955                 idle = &per_cpu(s390_idle, cpu);
 956                 memset(idle, 0, sizeof(struct s390_idle_data));
 957                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 958                         return NOTIFY_BAD;
 959                 break;
 960         case CPU_DEAD:
 961         case CPU_DEAD_FROZEN:
 962                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 963                 break;
 964         }
 965         return NOTIFY_OK;
 966 }
 967
 968 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 969         .notifier_call = smp_cpu_notify,
 970 };
 971
 972 static int __devinit smp_add_present_cpu(int cpu)
 973 {
 974         struct cpu *c = &per_cpu(cpu_devices, cpu);
 975         struct sys_device *s = &c->sysdev;
 976         int rc;
 977
 978         c->hotpluggable = 1;
 979         rc = register_cpu(c, cpu);
 980         if (rc)
 981                 goto out;
 982         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 983         if (rc)
 984                 goto out_cpu;
 985         if (!cpu_online(cpu))
 986                 goto out;
 987         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 988         if (!rc)
 989                 return 0;
 990         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 991 out_cpu:
 992 #ifdef CONFIG_HOTPLUG_CPU
 993         unregister_cpu(c);
 994 #endif
 995 out:
 996         return rc;
 997 }
 998
 999 #ifdef CONFIG_HOTPLUG_CPU
1000
1001 int __ref smp_rescan_cpus(void)
1002 {
1003         cpumask_t newcpus;
1004         int cpu;
1005         int rc;
1006
1007         get_online_cpus();
1008         mutex_lock(&smp_cpu_state_mutex);
1009         newcpus = cpu_present_map;
1010         rc = __smp_rescan_cpus();
1011         if (rc)
1012                 goto out;
1013         cpus_andnot(newcpus, cpu_present_map, newcpus);
1014         for_each_cpu_mask(cpu, newcpus) {
1015                 rc = smp_add_present_cpu(cpu);
1016                 if (rc)
1017                         cpu_clear(cpu, cpu_present_map);
1018         }
1019         rc = 0;
1020 out:
1021         mutex_unlock(&smp_cpu_state_mutex);
1022         put_online_cpus();
1023         if (!cpus_empty(newcpus))
1024                 topology_schedule_update();
1025         return rc;
1026 }
1027
1028 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1029                                   size_t count)
1030 {
1031         int rc;
1032
1033         rc = smp_rescan_cpus();
1034         return rc ? rc : count;
1035 }
1036 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1037 #endif /* CONFIG_HOTPLUG_CPU */
1038
1039 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1040 {
1041         ssize_t count;
1042
1043         mutex_lock(&smp_cpu_state_mutex);
1044         count = sprintf(buf, "%d\n", cpu_management);
1045         mutex_unlock(&smp_cpu_state_mutex);
1046         return count;
1047 }
1048
1049 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1050                                  size_t count)
1051 {
1052         int val, rc;
1053         char delim;
1054
1055         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1056                 return -EINVAL;
1057         if (val != 0 && val != 1)
1058                 return -EINVAL;
1059         rc = 0;
1060         get_online_cpus();
1061         mutex_lock(&smp_cpu_state_mutex);
1062         if (cpu_management == val)
1063                 goto out;
1064         rc = topology_set_cpu_management(val);
1065         if (!rc)
1066                 cpu_management = val;
1067 out:
1068         mutex_unlock(&smp_cpu_state_mutex);
1069         put_online_cpus();
1070         return rc ? rc : count;
1071 }
1072 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1073                          dispatching_store);
1074
1075 static int __init topology_init(void)
1076 {
1077         int cpu;
1078         int rc;
1079
1080         register_cpu_notifier(&smp_cpu_nb);
1081
1082 #ifdef CONFIG_HOTPLUG_CPU
1083         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1084         if (rc)
1085                 return rc;
1086 #endif
1087         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1088         if (rc)
1089                 return rc;
1090         for_each_present_cpu(cpu) {
1091                 rc = smp_add_present_cpu(cpu);
1092                 if (rc)
1093                         return rc;
1094         }
1095         return 0;
1096 }
1097 subsys_initcall(topology_init);