arch/s390/kernel/smp.c

   1 /*
   2  *  SMP related functions
   3  *
   4  *    Copyright IBM Corp. 1999, 2012
   5  *    Author(s): Denis Joseph Barrow,
   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  * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14  * the translation of logical to physical cpu ids. All new code that
  15  * operates on physical cpu numbers needs to go into smp.c.
  16  */
  17
  18 #define KMSG_COMPONENT "cpu"
  19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21 #include <linux/workqueue.h>
  22 #include <linux/module.h>
  23 #include <linux/init.h>
  24 #include <linux/mm.h>
  25 #include <linux/err.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/kernel_stat.h>
  28 #include <linux/delay.h>
  29 #include <linux/interrupt.h>
  30 #include <linux/irqflags.h>
  31 #include <linux/cpu.h>
  32 #include <linux/slab.h>
  33 #include <linux/crash_dump.h>
  34 #include <asm/asm-offsets.h>
  35 #include <asm/switch_to.h>
  36 #include <asm/facility.h>
  37 #include <asm/ipl.h>
  38 #include <asm/setup.h>
  39 #include <asm/irq.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/vtimer.h>
  42 #include <asm/lowcore.h>
  43 #include <asm/sclp.h>
  44 #include <asm/vdso.h>
  45 #include <asm/debug.h>
  46 #include <asm/os_info.h>
  47 #include <asm/sigp.h>
  48 #include "entry.h"
  49
  50 enum {
  51         ec_schedule = 0,
  52         ec_call_function_single,
  53         ec_stop_cpu,
  54 };
  55
  56 enum {
  57         CPU_STATE_STANDBY,
  58         CPU_STATE_CONFIGURED,
  59 };
  60
  61 struct pcpu {
  62         struct cpu *cpu;
  63         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
  64         unsigned long async_stack;      /* async stack for the cpu */
  65         unsigned long panic_stack;      /* panic stack for the cpu */
  66         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  67         int state;                      /* physical cpu state */
  68         int polarization;               /* physical polarization */
  69         u16 address;                    /* physical cpu address */
  70 };
  71
  72 static u8 boot_cpu_type;
  73 static u16 boot_cpu_address;
  74 static struct pcpu pcpu_devices[NR_CPUS];
  75
  76 /*
  77  * The smp_cpu_state_mutex must be held when changing the state or polarization
  78  * member of a pcpu data structure within the pcpu_devices arreay.
  79  */
  80 DEFINE_MUTEX(smp_cpu_state_mutex);
  81
  82 /*
  83  * Signal processor helper functions.
  84  */
  85 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
  86 {
  87         int cc;
  88
  89         while (1) {
  90                 cc = __pcpu_sigp(addr, order, parm, NULL);
  91                 if (cc != SIGP_CC_BUSY)
  92                         return cc;
  93                 cpu_relax();
  94         }
  95 }
  96
  97 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
  98 {
  99         int cc, retry;
 100
 101         for (retry = 0; ; retry++) {
 102                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 103                 if (cc != SIGP_CC_BUSY)
 104                         break;
 105                 if (retry >= 3)
 106                         udelay(10);
 107         }
 108         return cc;
 109 }
 110
 111 static inline int pcpu_stopped(struct pcpu *pcpu)
 112 {
 113         u32 uninitialized_var(status);
 114
 115         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 116                         0, &status) != SIGP_CC_STATUS_STORED)
 117                 return 0;
 118         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 119 }
 120
 121 static inline int pcpu_running(struct pcpu *pcpu)
 122 {
 123         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 124                         0, NULL) != SIGP_CC_STATUS_STORED)
 125                 return 1;
 126         /* Status stored condition code is equivalent to cpu not running. */
 127         return 0;
 128 }
 129
 130 /*
 131  * Find struct pcpu by cpu address.
 132  */
 133 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
 134 {
 135         int cpu;
 136
 137         for_each_cpu(cpu, mask)
 138                 if (pcpu_devices[cpu].address == address)
 139                         return pcpu_devices + cpu;
 140         return NULL;
 141 }
 142
 143 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 144 {
 145         int order;
 146
 147         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 148                 return;
 149         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 150         pcpu_sigp_retry(pcpu, order, 0);
 151 }
 152
 153 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 154 {
 155         struct _lowcore *lc;
 156
 157         if (pcpu != &pcpu_devices[0]) {
 158                 pcpu->lowcore = (struct _lowcore *)
 159                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 160                 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 161                 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
 162                 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
 163                         goto out;
 164         }
 165         lc = pcpu->lowcore;
 166         memcpy(lc, &S390_lowcore, 512);
 167         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 168         lc->async_stack = pcpu->async_stack + ASYNC_SIZE
 169                 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 170         lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
 171                 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 172         lc->cpu_nr = cpu;
 173 #ifndef CONFIG_64BIT
 174         if (MACHINE_HAS_IEEE) {
 175                 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
 176                 if (!lc->extended_save_area_addr)
 177                         goto out;
 178         }
 179 #else
 180         if (vdso_alloc_per_cpu(lc))
 181                 goto out;
 182 #endif
 183         lowcore_ptr[cpu] = lc;
 184         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 185         return 0;
 186 out:
 187         if (pcpu != &pcpu_devices[0]) {
 188                 free_page(pcpu->panic_stack);
 189                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 190                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 191         }
 192         return -ENOMEM;
 193 }
 194
 195 #ifdef CONFIG_HOTPLUG_CPU
 196
 197 static void pcpu_free_lowcore(struct pcpu *pcpu)
 198 {
 199         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 200         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 201 #ifndef CONFIG_64BIT
 202         if (MACHINE_HAS_IEEE) {
 203                 struct _lowcore *lc = pcpu->lowcore;
 204
 205                 free_page((unsigned long) lc->extended_save_area_addr);
 206                 lc->extended_save_area_addr = 0;
 207         }
 208 #else
 209         vdso_free_per_cpu(pcpu->lowcore);
 210 #endif
 211         if (pcpu != &pcpu_devices[0]) {
 212                 free_page(pcpu->panic_stack);
 213                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 214                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 215         }
 216 }
 217
 218 #endif /* CONFIG_HOTPLUG_CPU */
 219
 220 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 221 {
 222         struct _lowcore *lc = pcpu->lowcore;
 223
 224         if (MACHINE_HAS_TLB_LC)
 225                 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 226         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 227         atomic_inc(&init_mm.context.attach_count);
 228         lc->cpu_nr = cpu;
 229         lc->percpu_offset = __per_cpu_offset[cpu];
 230         lc->kernel_asce = S390_lowcore.kernel_asce;
 231         lc->machine_flags = S390_lowcore.machine_flags;
 232         lc->ftrace_func = S390_lowcore.ftrace_func;
 233         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 234         __ctl_store(lc->cregs_save_area, 0, 15);
 235         save_access_regs((unsigned int *) lc->access_regs_save_area);
 236         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 237                MAX_FACILITY_BIT/8);
 238 }
 239
 240 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 241 {
 242         struct _lowcore *lc = pcpu->lowcore;
 243         struct thread_info *ti = task_thread_info(tsk);
 244
 245         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 246                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 247         lc->thread_info = (unsigned long) task_thread_info(tsk);
 248         lc->current_task = (unsigned long) tsk;
 249         lc->user_timer = ti->user_timer;
 250         lc->system_timer = ti->system_timer;
 251         lc->steal_timer = 0;
 252 }
 253
 254 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 255 {
 256         struct _lowcore *lc = pcpu->lowcore;
 257
 258         lc->restart_stack = lc->kernel_stack;
 259         lc->restart_fn = (unsigned long) func;
 260         lc->restart_data = (unsigned long) data;
 261         lc->restart_source = -1UL;
 262         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 263 }
 264
 265 /*
 266  * Call function via PSW restart on pcpu and stop the current cpu.
 267  */
 268 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 269                           void *data, unsigned long stack)
 270 {
 271         struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 272         unsigned long source_cpu = stap();
 273
 274         __load_psw_mask(PSW_KERNEL_BITS);
 275         if (pcpu->address == source_cpu)
 276                 func(data);     /* should not return */
 277         /* Stop target cpu (if func returns this stops the current cpu). */
 278         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 279         /* Restart func on the target cpu and stop the current cpu. */
 280         mem_assign_absolute(lc->restart_stack, stack);
 281         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 282         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 283         mem_assign_absolute(lc->restart_source, source_cpu);
 284         asm volatile(
 285                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 286                 "       brc     2,0b    # busy, try again\n"
 287                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 288                 "       brc     2,1b    # busy, try again\n"
 289                 : : "d" (pcpu->address), "d" (source_cpu),
 290                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 291                 : "0", "1", "cc");
 292         for (;;) ;
 293 }
 294
 295 /*
 296  * Call function on an online CPU.
 297  */
 298 void smp_call_online_cpu(void (*func)(void *), void *data)
 299 {
 300         struct pcpu *pcpu;
 301
 302         /* Use the current cpu if it is online. */
 303         pcpu = pcpu_find_address(cpu_online_mask, stap());
 304         if (!pcpu)
 305                 /* Use the first online cpu. */
 306                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 307         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 308 }
 309
 310 /*
 311  * Call function on the ipl CPU.
 312  */
 313 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 314 {
 315         pcpu_delegate(&pcpu_devices[0], func, data,
 316                       pcpu_devices->panic_stack + PAGE_SIZE);
 317 }
 318
 319 int smp_find_processor_id(u16 address)
 320 {
 321         int cpu;
 322
 323         for_each_present_cpu(cpu)
 324                 if (pcpu_devices[cpu].address == address)
 325                         return cpu;
 326         return -1;
 327 }
 328
 329 int smp_vcpu_scheduled(int cpu)
 330 {
 331         return pcpu_running(pcpu_devices + cpu);
 332 }
 333
 334 void smp_yield(void)
 335 {
 336         if (MACHINE_HAS_DIAG44)
 337                 asm volatile("diag 0,0,0x44");
 338 }
 339
 340 void smp_yield_cpu(int cpu)
 341 {
 342         if (MACHINE_HAS_DIAG9C)
 343                 asm volatile("diag %0,0,0x9c"
 344                              : : "d" (pcpu_devices[cpu].address));
 345         else if (MACHINE_HAS_DIAG44)
 346                 asm volatile("diag 0,0,0x44");
 347 }
 348
 349 /*
 350  * Send cpus emergency shutdown signal. This gives the cpus the
 351  * opportunity to complete outstanding interrupts.
 352  */
 353 static void smp_emergency_stop(cpumask_t *cpumask)
 354 {
 355         u64 end;
 356         int cpu;
 357
 358         end = get_tod_clock() + (1000000UL << 12);
 359         for_each_cpu(cpu, cpumask) {
 360                 struct pcpu *pcpu = pcpu_devices + cpu;
 361                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 362                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 363                                    0, NULL) == SIGP_CC_BUSY &&
 364                        get_tod_clock() < end)
 365                         cpu_relax();
 366         }
 367         while (get_tod_clock() < end) {
 368                 for_each_cpu(cpu, cpumask)
 369                         if (pcpu_stopped(pcpu_devices + cpu))
 370                                 cpumask_clear_cpu(cpu, cpumask);
 371                 if (cpumask_empty(cpumask))
 372                         break;
 373                 cpu_relax();
 374         }
 375 }
 376
 377 /*
 378  * Stop all cpus but the current one.
 379  */
 380 void smp_send_stop(void)
 381 {
 382         cpumask_t cpumask;
 383         int cpu;
 384
 385         /* Disable all interrupts/machine checks */
 386         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 387         trace_hardirqs_off();
 388
 389         debug_set_critical();
 390         cpumask_copy(&cpumask, cpu_online_mask);
 391         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 392
 393         if (oops_in_progress)
 394                 smp_emergency_stop(&cpumask);
 395
 396         /* stop all processors */
 397         for_each_cpu(cpu, &cpumask) {
 398                 struct pcpu *pcpu = pcpu_devices + cpu;
 399                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 400                 while (!pcpu_stopped(pcpu))
 401                         cpu_relax();
 402         }
 403 }
 404
 405 /*
 406  * Stop the current cpu.
 407  */
 408 void smp_stop_cpu(void)
 409 {
 410         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 411         for (;;) ;
 412 }
 413
 414 /*
 415  * This is the main routine where commands issued by other
 416  * cpus are handled.
 417  */
 418 static void smp_handle_ext_call(void)
 419 {
 420         unsigned long bits;
 421
 422         /* handle bit signal external calls */
 423         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 424         if (test_bit(ec_stop_cpu, &bits))
 425                 smp_stop_cpu();
 426         if (test_bit(ec_schedule, &bits))
 427                 scheduler_ipi();
 428         if (test_bit(ec_call_function_single, &bits))
 429                 generic_smp_call_function_single_interrupt();
 430 }
 431
 432 static void do_ext_call_interrupt(struct ext_code ext_code,
 433                                   unsigned int param32, unsigned long param64)
 434 {
 435         inc_irq_stat(ext_code.code == 0x1202 ? IRQEXT_EXC : IRQEXT_EMS);
 436         smp_handle_ext_call();
 437 }
 438
 439 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 440 {
 441         int cpu;
 442
 443         for_each_cpu(cpu, mask)
 444                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 445 }
 446
 447 void arch_send_call_function_single_ipi(int cpu)
 448 {
 449         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 450 }
 451
 452 #ifndef CONFIG_64BIT
 453 /*
 454  * this function sends a 'purge tlb' signal to another CPU.
 455  */
 456 static void smp_ptlb_callback(void *info)
 457 {
 458         __tlb_flush_local();
 459 }
 460
 461 void smp_ptlb_all(void)
 462 {
 463         on_each_cpu(smp_ptlb_callback, NULL, 1);
 464 }
 465 EXPORT_SYMBOL(smp_ptlb_all);
 466 #endif /* ! CONFIG_64BIT */
 467
 468 /*
 469  * this function sends a 'reschedule' IPI to another CPU.
 470  * it goes straight through and wastes no time serializing
 471  * anything. Worst case is that we lose a reschedule ...
 472  */
 473 void smp_send_reschedule(int cpu)
 474 {
 475         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 476 }
 477
 478 /*
 479  * parameter area for the set/clear control bit callbacks
 480  */
 481 struct ec_creg_mask_parms {
 482         unsigned long orval;
 483         unsigned long andval;
 484         int cr;
 485 };
 486
 487 /*
 488  * callback for setting/clearing control bits
 489  */
 490 static void smp_ctl_bit_callback(void *info)
 491 {
 492         struct ec_creg_mask_parms *pp = info;
 493         unsigned long cregs[16];
 494
 495         __ctl_store(cregs, 0, 15);
 496         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 497         __ctl_load(cregs, 0, 15);
 498 }
 499
 500 /*
 501  * Set a bit in a control register of all cpus
 502  */
 503 void smp_ctl_set_bit(int cr, int bit)
 504 {
 505         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 506
 507         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 508 }
 509 EXPORT_SYMBOL(smp_ctl_set_bit);
 510
 511 /*
 512  * Clear a bit in a control register of all cpus
 513  */
 514 void smp_ctl_clear_bit(int cr, int bit)
 515 {
 516         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 517
 518         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 519 }
 520 EXPORT_SYMBOL(smp_ctl_clear_bit);
 521
 522 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
 523
 524 static void __init smp_get_save_area(int cpu, u16 address)
 525 {
 526         void *lc = pcpu_devices[0].lowcore;
 527         struct save_area *save_area;
 528
 529         if (is_kdump_kernel())
 530                 return;
 531         if (!OLDMEM_BASE && (address == boot_cpu_address ||
 532                              ipl_info.type != IPL_TYPE_FCP_DUMP))
 533                 return;
 534         save_area = dump_save_area_create(cpu);
 535         if (!save_area)
 536                 panic("could not allocate memory for save area\n");
 537 #ifdef CONFIG_CRASH_DUMP
 538         if (address == boot_cpu_address) {
 539                 /* Copy the registers of the boot cpu. */
 540                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 541                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
 542                 return;
 543         }
 544 #endif
 545         /* Get the registers of a non-boot cpu. */
 546         __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
 547         memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
 548 }
 549
 550 int smp_store_status(int cpu)
 551 {
 552         struct pcpu *pcpu;
 553
 554         pcpu = pcpu_devices + cpu;
 555         if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
 556                               0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
 557                 return -EIO;
 558         return 0;
 559 }
 560
 561 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 562
 563 static inline void smp_get_save_area(int cpu, u16 address) { }
 564
 565 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 566
 567 void smp_cpu_set_polarization(int cpu, int val)
 568 {
 569         pcpu_devices[cpu].polarization = val;
 570 }
 571
 572 int smp_cpu_get_polarization(int cpu)
 573 {
 574         return pcpu_devices[cpu].polarization;
 575 }
 576
 577 static struct sclp_cpu_info *smp_get_cpu_info(void)
 578 {
 579         static int use_sigp_detection;
 580         struct sclp_cpu_info *info;
 581         int address;
 582
 583         info = kzalloc(sizeof(*info), GFP_KERNEL);
 584         if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
 585                 use_sigp_detection = 1;
 586                 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
 587                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
 588                             SIGP_CC_NOT_OPERATIONAL)
 589                                 continue;
 590                         info->cpu[info->configured].address = address;
 591                         info->configured++;
 592                 }
 593                 info->combined = info->configured;
 594         }
 595         return info;
 596 }
 597
 598 static int smp_add_present_cpu(int cpu);
 599
 600 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
 601 {
 602         struct pcpu *pcpu;
 603         cpumask_t avail;
 604         int cpu, nr, i;
 605
 606         nr = 0;
 607         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 608         cpu = cpumask_first(&avail);
 609         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 610                 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
 611                         continue;
 612                 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
 613                         continue;
 614                 pcpu = pcpu_devices + cpu;
 615                 pcpu->address = info->cpu[i].address;
 616                 pcpu->state = (i >= info->configured) ?
 617                         CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 618                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 619                 set_cpu_present(cpu, true);
 620                 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 621                         set_cpu_present(cpu, false);
 622                 else
 623                         nr++;
 624                 cpu = cpumask_next(cpu, &avail);
 625         }
 626         return nr;
 627 }
 628
 629 static void __init smp_detect_cpus(void)
 630 {
 631         unsigned int cpu, c_cpus, s_cpus;
 632         struct sclp_cpu_info *info;
 633
 634         info = smp_get_cpu_info();
 635         if (!info)
 636                 panic("smp_detect_cpus failed to allocate memory\n");
 637         if (info->has_cpu_type) {
 638                 for (cpu = 0; cpu < info->combined; cpu++) {
 639                         if (info->cpu[cpu].address != boot_cpu_address)
 640                                 continue;
 641                         /* The boot cpu dictates the cpu type. */
 642                         boot_cpu_type = info->cpu[cpu].type;
 643                         break;
 644                 }
 645         }
 646         c_cpus = s_cpus = 0;
 647         for (cpu = 0; cpu < info->combined; cpu++) {
 648                 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
 649                         continue;
 650                 if (cpu < info->configured) {
 651                         smp_get_save_area(c_cpus, info->cpu[cpu].address);
 652                         c_cpus++;
 653                 } else
 654                         s_cpus++;
 655         }
 656         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 657         get_online_cpus();
 658         __smp_rescan_cpus(info, 0);
 659         put_online_cpus();
 660         kfree(info);
 661 }
 662
 663 /*
 664  *      Activate a secondary processor.
 665  */
 666 static void smp_start_secondary(void *cpuvoid)
 667 {
 668         S390_lowcore.last_update_clock = get_tod_clock();
 669         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 670         S390_lowcore.restart_fn = (unsigned long) do_restart;
 671         S390_lowcore.restart_data = 0;
 672         S390_lowcore.restart_source = -1UL;
 673         restore_access_regs(S390_lowcore.access_regs_save_area);
 674         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 675         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 676         cpu_init();
 677         preempt_disable();
 678         init_cpu_timer();
 679         init_cpu_vtimer();
 680         pfault_init();
 681         notify_cpu_starting(smp_processor_id());
 682         set_cpu_online(smp_processor_id(), true);
 683         inc_irq_stat(CPU_RST);
 684         local_irq_enable();
 685         cpu_startup_entry(CPUHP_ONLINE);
 686 }
 687
 688 /* Upping and downing of CPUs */
 689 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 690 {
 691         struct pcpu *pcpu;
 692         int rc;
 693
 694         pcpu = pcpu_devices + cpu;
 695         if (pcpu->state != CPU_STATE_CONFIGURED)
 696                 return -EIO;
 697         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 698             SIGP_CC_ORDER_CODE_ACCEPTED)
 699                 return -EIO;
 700
 701         rc = pcpu_alloc_lowcore(pcpu, cpu);
 702         if (rc)
 703                 return rc;
 704         pcpu_prepare_secondary(pcpu, cpu);
 705         pcpu_attach_task(pcpu, tidle);
 706         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 707         while (!cpu_online(cpu))
 708                 cpu_relax();
 709         return 0;
 710 }
 711
 712 static unsigned int setup_possible_cpus __initdata;
 713
 714 static int __init _setup_possible_cpus(char *s)
 715 {
 716         get_option(&s, &setup_possible_cpus);
 717         return 0;
 718 }
 719 early_param("possible_cpus", _setup_possible_cpus);
 720
 721 #ifdef CONFIG_HOTPLUG_CPU
 722
 723 int __cpu_disable(void)
 724 {
 725         unsigned long cregs[16];
 726
 727         /* Handle possible pending IPIs */
 728         smp_handle_ext_call();
 729         set_cpu_online(smp_processor_id(), false);
 730         /* Disable pseudo page faults on this cpu. */
 731         pfault_fini();
 732         /* Disable interrupt sources via control register. */
 733         __ctl_store(cregs, 0, 15);
 734         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 735         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 736         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 737         __ctl_load(cregs, 0, 15);
 738         return 0;
 739 }
 740
 741 void __cpu_die(unsigned int cpu)
 742 {
 743         struct pcpu *pcpu;
 744
 745         /* Wait until target cpu is down */
 746         pcpu = pcpu_devices + cpu;
 747         while (!pcpu_stopped(pcpu))
 748                 cpu_relax();
 749         pcpu_free_lowcore(pcpu);
 750         atomic_dec(&init_mm.context.attach_count);
 751         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 752         if (MACHINE_HAS_TLB_LC)
 753                 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 754 }
 755
 756 void __noreturn cpu_die(void)
 757 {
 758         idle_task_exit();
 759         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 760         for (;;) ;
 761 }
 762
 763 #endif /* CONFIG_HOTPLUG_CPU */
 764
 765 void __init smp_fill_possible_mask(void)
 766 {
 767         unsigned int possible, sclp, cpu;
 768
 769         sclp = sclp_get_max_cpu() ?: nr_cpu_ids;
 770         possible = setup_possible_cpus ?: nr_cpu_ids;
 771         possible = min(possible, sclp);
 772         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 773                 set_cpu_possible(cpu, true);
 774 }
 775
 776 void __init smp_prepare_cpus(unsigned int max_cpus)
 777 {
 778         /* request the 0x1201 emergency signal external interrupt */
 779         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 780                 panic("Couldn't request external interrupt 0x1201");
 781         /* request the 0x1202 external call external interrupt */
 782         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 783                 panic("Couldn't request external interrupt 0x1202");
 784         smp_detect_cpus();
 785 }
 786
 787 void __init smp_prepare_boot_cpu(void)
 788 {
 789         struct pcpu *pcpu = pcpu_devices;
 790
 791         boot_cpu_address = stap();
 792         pcpu->state = CPU_STATE_CONFIGURED;
 793         pcpu->address = boot_cpu_address;
 794         pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
 795         pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE
 796                 + STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
 797         pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE
 798                 + STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
 799         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 800         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 801         set_cpu_present(0, true);
 802         set_cpu_online(0, true);
 803 }
 804
 805 void __init smp_cpus_done(unsigned int max_cpus)
 806 {
 807 }
 808
 809 void __init smp_setup_processor_id(void)
 810 {
 811         S390_lowcore.cpu_nr = 0;
 812 }
 813
 814 /*
 815  * the frequency of the profiling timer can be changed
 816  * by writing a multiplier value into /proc/profile.
 817  *
 818  * usually you want to run this on all CPUs ;)
 819  */
 820 int setup_profiling_timer(unsigned int multiplier)
 821 {
 822         return 0;
 823 }
 824
 825 #ifdef CONFIG_HOTPLUG_CPU
 826 static ssize_t cpu_configure_show(struct device *dev,
 827                                   struct device_attribute *attr, char *buf)
 828 {
 829         ssize_t count;
 830
 831         mutex_lock(&smp_cpu_state_mutex);
 832         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 833         mutex_unlock(&smp_cpu_state_mutex);
 834         return count;
 835 }
 836
 837 static ssize_t cpu_configure_store(struct device *dev,
 838                                    struct device_attribute *attr,
 839                                    const char *buf, size_t count)
 840 {
 841         struct pcpu *pcpu;
 842         int cpu, val, rc;
 843         char delim;
 844
 845         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 846                 return -EINVAL;
 847         if (val != 0 && val != 1)
 848                 return -EINVAL;
 849         get_online_cpus();
 850         mutex_lock(&smp_cpu_state_mutex);
 851         rc = -EBUSY;
 852         /* disallow configuration changes of online cpus and cpu 0 */
 853         cpu = dev->id;
 854         if (cpu_online(cpu) || cpu == 0)
 855                 goto out;
 856         pcpu = pcpu_devices + cpu;
 857         rc = 0;
 858         switch (val) {
 859         case 0:
 860                 if (pcpu->state != CPU_STATE_CONFIGURED)
 861                         break;
 862                 rc = sclp_cpu_deconfigure(pcpu->address);
 863                 if (rc)
 864                         break;
 865                 pcpu->state = CPU_STATE_STANDBY;
 866                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 867                 topology_expect_change();
 868                 break;
 869         case 1:
 870                 if (pcpu->state != CPU_STATE_STANDBY)
 871                         break;
 872                 rc = sclp_cpu_configure(pcpu->address);
 873                 if (rc)
 874                         break;
 875                 pcpu->state = CPU_STATE_CONFIGURED;
 876                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 877                 topology_expect_change();
 878                 break;
 879         default:
 880                 break;
 881         }
 882 out:
 883         mutex_unlock(&smp_cpu_state_mutex);
 884         put_online_cpus();
 885         return rc ? rc : count;
 886 }
 887 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 888 #endif /* CONFIG_HOTPLUG_CPU */
 889
 890 static ssize_t show_cpu_address(struct device *dev,
 891                                 struct device_attribute *attr, char *buf)
 892 {
 893         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
 894 }
 895 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
 896
 897 static struct attribute *cpu_common_attrs[] = {
 898 #ifdef CONFIG_HOTPLUG_CPU
 899         &dev_attr_configure.attr,
 900 #endif
 901         &dev_attr_address.attr,
 902         NULL,
 903 };
 904
 905 static struct attribute_group cpu_common_attr_group = {
 906         .attrs = cpu_common_attrs,
 907 };
 908
 909 static ssize_t show_idle_count(struct device *dev,
 910                                 struct device_attribute *attr, char *buf)
 911 {
 912         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 913         unsigned long long idle_count;
 914         unsigned int sequence;
 915
 916         do {
 917                 sequence = ACCESS_ONCE(idle->sequence);
 918                 idle_count = ACCESS_ONCE(idle->idle_count);
 919                 if (ACCESS_ONCE(idle->clock_idle_enter))
 920                         idle_count++;
 921         } while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
 922         return sprintf(buf, "%llu\n", idle_count);
 923 }
 924 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
 925
 926 static ssize_t show_idle_time(struct device *dev,
 927                                 struct device_attribute *attr, char *buf)
 928 {
 929         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 930         unsigned long long now, idle_time, idle_enter, idle_exit;
 931         unsigned int sequence;
 932
 933         do {
 934                 now = get_tod_clock();
 935                 sequence = ACCESS_ONCE(idle->sequence);
 936                 idle_time = ACCESS_ONCE(idle->idle_time);
 937                 idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
 938                 idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
 939         } while ((sequence & 1) || (ACCESS_ONCE(idle->sequence) != sequence));
 940         idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
 941         return sprintf(buf, "%llu\n", idle_time >> 12);
 942 }
 943 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 944
 945 static struct attribute *cpu_online_attrs[] = {
 946         &dev_attr_idle_count.attr,
 947         &dev_attr_idle_time_us.attr,
 948         NULL,
 949 };
 950
 951 static struct attribute_group cpu_online_attr_group = {
 952         .attrs = cpu_online_attrs,
 953 };
 954
 955 static int smp_cpu_notify(struct notifier_block *self, unsigned long action,
 956                           void *hcpu)
 957 {
 958         unsigned int cpu = (unsigned int)(long)hcpu;
 959         struct cpu *c = pcpu_devices[cpu].cpu;
 960         struct device *s = &c->dev;
 961         int err = 0;
 962
 963         switch (action & ~CPU_TASKS_FROZEN) {
 964         case CPU_ONLINE:
 965                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 966                 break;
 967         case CPU_DEAD:
 968                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 969                 break;
 970         }
 971         return notifier_from_errno(err);
 972 }
 973
 974 static int smp_add_present_cpu(int cpu)
 975 {
 976         struct device *s;
 977         struct cpu *c;
 978         int rc;
 979
 980         c = kzalloc(sizeof(*c), GFP_KERNEL);
 981         if (!c)
 982                 return -ENOMEM;
 983         pcpu_devices[cpu].cpu = c;
 984         s = &c->dev;
 985         c->hotpluggable = 1;
 986         rc = register_cpu(c, cpu);
 987         if (rc)
 988                 goto out;
 989         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 990         if (rc)
 991                 goto out_cpu;
 992         if (cpu_online(cpu)) {
 993                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 994                 if (rc)
 995                         goto out_online;
 996         }
 997         rc = topology_cpu_init(c);
 998         if (rc)
 999                 goto out_topology;
1000         return 0;
1001
1002 out_topology:
1003         if (cpu_online(cpu))
1004                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1005 out_online:
1006         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1007 out_cpu:
1008 #ifdef CONFIG_HOTPLUG_CPU
1009         unregister_cpu(c);
1010 #endif
1011 out:
1012         return rc;
1013 }
1014
1015 #ifdef CONFIG_HOTPLUG_CPU
1016
1017 int __ref smp_rescan_cpus(void)
1018 {
1019         struct sclp_cpu_info *info;
1020         int nr;
1021
1022         info = smp_get_cpu_info();
1023         if (!info)
1024                 return -ENOMEM;
1025         get_online_cpus();
1026         mutex_lock(&smp_cpu_state_mutex);
1027         nr = __smp_rescan_cpus(info, 1);
1028         mutex_unlock(&smp_cpu_state_mutex);
1029         put_online_cpus();
1030         kfree(info);
1031         if (nr)
1032                 topology_schedule_update();
1033         return 0;
1034 }
1035
1036 static ssize_t __ref rescan_store(struct device *dev,
1037                                   struct device_attribute *attr,
1038                                   const char *buf,
1039                                   size_t count)
1040 {
1041         int rc;
1042
1043         rc = smp_rescan_cpus();
1044         return rc ? rc : count;
1045 }
1046 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1047 #endif /* CONFIG_HOTPLUG_CPU */
1048
1049 static int __init s390_smp_init(void)
1050 {
1051         int cpu, rc = 0;
1052
1053 #ifdef CONFIG_HOTPLUG_CPU
1054         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1055         if (rc)
1056                 return rc;
1057 #endif
1058         cpu_notifier_register_begin();
1059         for_each_present_cpu(cpu) {
1060                 rc = smp_add_present_cpu(cpu);
1061                 if (rc)
1062                         goto out;
1063         }
1064
1065         __hotcpu_notifier(smp_cpu_notify, 0);
1066
1067 out:
1068         cpu_notifier_register_done();
1069         return rc;
1070 }
1071 subsys_initcall(s390_smp_init);