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,
  53         ec_call_function_single,
  54         ec_stop_cpu,
  55 };
  56
  57 enum {
  58         CPU_STATE_STANDBY,
  59         CPU_STATE_CONFIGURED,
  60 };
  61
  62 struct pcpu {
  63         struct cpu cpu;
  64         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
  65         unsigned long async_stack;      /* async stack for the cpu */
  66         unsigned long panic_stack;      /* panic stack for the cpu */
  67         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  68         int state;                      /* physical cpu state */
  69         int polarization;               /* physical polarization */
  70         u16 address;                    /* physical cpu address */
  71 };
  72
  73 static u8 boot_cpu_type;
  74 static u16 boot_cpu_address;
  75 static struct pcpu pcpu_devices[NR_CPUS];
  76
  77 /*
  78  * The smp_cpu_state_mutex must be held when changing the state or polarization
  79  * member of a pcpu data structure within the pcpu_devices arreay.
  80  */
  81 DEFINE_MUTEX(smp_cpu_state_mutex);
  82
  83 /*
  84  * Signal processor helper functions.
  85  */
  86 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
  87 {
  88         register unsigned int reg1 asm ("1") = parm;
  89         int cc;
  90
  91         asm volatile(
  92                 "       sigp    %1,%2,0(%3)\n"
  93                 "       ipm     %0\n"
  94                 "       srl     %0,28\n"
  95                 : "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
  96         if (status && cc == 1)
  97                 *status = reg1;
  98         return cc;
  99 }
 100
 101 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
 102 {
 103         int cc;
 104
 105         while (1) {
 106                 cc = __pcpu_sigp(addr, order, parm, NULL);
 107                 if (cc != SIGP_CC_BUSY)
 108                         return cc;
 109                 cpu_relax();
 110         }
 111 }
 112
 113 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 114 {
 115         int cc, retry;
 116
 117         for (retry = 0; ; retry++) {
 118                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 119                 if (cc != SIGP_CC_BUSY)
 120                         break;
 121                 if (retry >= 3)
 122                         udelay(10);
 123         }
 124         return cc;
 125 }
 126
 127 static inline int pcpu_stopped(struct pcpu *pcpu)
 128 {
 129         u32 uninitialized_var(status);
 130
 131         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 132                         0, &status) != SIGP_CC_STATUS_STORED)
 133                 return 0;
 134         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 135 }
 136
 137 static inline int pcpu_running(struct pcpu *pcpu)
 138 {
 139         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 140                         0, NULL) != SIGP_CC_STATUS_STORED)
 141                 return 1;
 142         /* Status stored condition code is equivalent to cpu not running. */
 143         return 0;
 144 }
 145
 146 /*
 147  * Find struct pcpu by cpu address.
 148  */
 149 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
 150 {
 151         int cpu;
 152
 153         for_each_cpu(cpu, mask)
 154                 if (pcpu_devices[cpu].address == address)
 155                         return pcpu_devices + cpu;
 156         return NULL;
 157 }
 158
 159 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 160 {
 161         int order;
 162
 163         set_bit(ec_bit, &pcpu->ec_mask);
 164         order = pcpu_running(pcpu) ?
 165                 SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 166         pcpu_sigp_retry(pcpu, order, 0);
 167 }
 168
 169 static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 170 {
 171         struct _lowcore *lc;
 172
 173         if (pcpu != &pcpu_devices[0]) {
 174                 pcpu->lowcore = (struct _lowcore *)
 175                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 176                 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 177                 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
 178                 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
 179                         goto out;
 180         }
 181         lc = pcpu->lowcore;
 182         memcpy(lc, &S390_lowcore, 512);
 183         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 184         lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
 185         lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
 186         lc->cpu_nr = cpu;
 187 #ifndef CONFIG_64BIT
 188         if (MACHINE_HAS_IEEE) {
 189                 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
 190                 if (!lc->extended_save_area_addr)
 191                         goto out;
 192         }
 193 #else
 194         if (vdso_alloc_per_cpu(lc))
 195                 goto out;
 196 #endif
 197         lowcore_ptr[cpu] = lc;
 198         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 199         return 0;
 200 out:
 201         if (pcpu != &pcpu_devices[0]) {
 202                 free_page(pcpu->panic_stack);
 203                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 204                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 205         }
 206         return -ENOMEM;
 207 }
 208
 209 #ifdef CONFIG_HOTPLUG_CPU
 210
 211 static void pcpu_free_lowcore(struct pcpu *pcpu)
 212 {
 213         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 214         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 215 #ifndef CONFIG_64BIT
 216         if (MACHINE_HAS_IEEE) {
 217                 struct _lowcore *lc = pcpu->lowcore;
 218
 219                 free_page((unsigned long) lc->extended_save_area_addr);
 220                 lc->extended_save_area_addr = 0;
 221         }
 222 #else
 223         vdso_free_per_cpu(pcpu->lowcore);
 224 #endif
 225         if (pcpu != &pcpu_devices[0]) {
 226                 free_page(pcpu->panic_stack);
 227                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 228                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 229         }
 230 }
 231
 232 #endif /* CONFIG_HOTPLUG_CPU */
 233
 234 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 235 {
 236         struct _lowcore *lc = pcpu->lowcore;
 237
 238         atomic_inc(&init_mm.context.attach_count);
 239         lc->cpu_nr = cpu;
 240         lc->percpu_offset = __per_cpu_offset[cpu];
 241         lc->kernel_asce = S390_lowcore.kernel_asce;
 242         lc->machine_flags = S390_lowcore.machine_flags;
 243         lc->ftrace_func = S390_lowcore.ftrace_func;
 244         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 245         __ctl_store(lc->cregs_save_area, 0, 15);
 246         save_access_regs((unsigned int *) lc->access_regs_save_area);
 247         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 248                MAX_FACILITY_BIT/8);
 249 }
 250
 251 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 252 {
 253         struct _lowcore *lc = pcpu->lowcore;
 254         struct thread_info *ti = task_thread_info(tsk);
 255
 256         lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
 257         lc->thread_info = (unsigned long) task_thread_info(tsk);
 258         lc->current_task = (unsigned long) tsk;
 259         lc->user_timer = ti->user_timer;
 260         lc->system_timer = ti->system_timer;
 261         lc->steal_timer = 0;
 262 }
 263
 264 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 265 {
 266         struct _lowcore *lc = pcpu->lowcore;
 267
 268         lc->restart_stack = lc->kernel_stack;
 269         lc->restart_fn = (unsigned long) func;
 270         lc->restart_data = (unsigned long) data;
 271         lc->restart_source = -1UL;
 272         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 273 }
 274
 275 /*
 276  * Call function via PSW restart on pcpu and stop the current cpu.
 277  */
 278 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 279                           void *data, unsigned long stack)
 280 {
 281         struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 282         unsigned long source_cpu = stap();
 283
 284         __load_psw_mask(psw_kernel_bits);
 285         if (pcpu->address == source_cpu)
 286                 func(data);     /* should not return */
 287         /* Stop target cpu (if func returns this stops the current cpu). */
 288         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 289         /* Restart func on the target cpu and stop the current cpu. */
 290         mem_assign_absolute(lc->restart_stack, stack);
 291         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 292         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 293         mem_assign_absolute(lc->restart_source, source_cpu);
 294         asm volatile(
 295                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 296                 "       brc     2,0b    # busy, try again\n"
 297                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 298                 "       brc     2,1b    # busy, try again\n"
 299                 : : "d" (pcpu->address), "d" (source_cpu),
 300                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 301                 : "0", "1", "cc");
 302         for (;;) ;
 303 }
 304
 305 /*
 306  * Call function on an online CPU.
 307  */
 308 void smp_call_online_cpu(void (*func)(void *), void *data)
 309 {
 310         struct pcpu *pcpu;
 311
 312         /* Use the current cpu if it is online. */
 313         pcpu = pcpu_find_address(cpu_online_mask, stap());
 314         if (!pcpu)
 315                 /* Use the first online cpu. */
 316                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 317         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 318 }
 319
 320 /*
 321  * Call function on the ipl CPU.
 322  */
 323 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 324 {
 325         pcpu_delegate(&pcpu_devices[0], func, data,
 326                       pcpu_devices->panic_stack + PAGE_SIZE);
 327 }
 328
 329 int smp_find_processor_id(u16 address)
 330 {
 331         int cpu;
 332
 333         for_each_present_cpu(cpu)
 334                 if (pcpu_devices[cpu].address == address)
 335                         return cpu;
 336         return -1;
 337 }
 338
 339 int smp_vcpu_scheduled(int cpu)
 340 {
 341         return pcpu_running(pcpu_devices + cpu);
 342 }
 343
 344 void smp_yield(void)
 345 {
 346         if (MACHINE_HAS_DIAG44)
 347                 asm volatile("diag 0,0,0x44");
 348 }
 349
 350 void smp_yield_cpu(int cpu)
 351 {
 352         if (MACHINE_HAS_DIAG9C)
 353                 asm volatile("diag %0,0,0x9c"
 354                              : : "d" (pcpu_devices[cpu].address));
 355         else if (MACHINE_HAS_DIAG44)
 356                 asm volatile("diag 0,0,0x44");
 357 }
 358
 359 /*
 360  * Send cpus emergency shutdown signal. This gives the cpus the
 361  * opportunity to complete outstanding interrupts.
 362  */
 363 void smp_emergency_stop(cpumask_t *cpumask)
 364 {
 365         u64 end;
 366         int cpu;
 367
 368         end = get_clock() + (1000000UL << 12);
 369         for_each_cpu(cpu, cpumask) {
 370                 struct pcpu *pcpu = pcpu_devices + cpu;
 371                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 372                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 373                                    0, NULL) == SIGP_CC_BUSY &&
 374                        get_clock() < end)
 375                         cpu_relax();
 376         }
 377         while (get_clock() < end) {
 378                 for_each_cpu(cpu, cpumask)
 379                         if (pcpu_stopped(pcpu_devices + cpu))
 380                                 cpumask_clear_cpu(cpu, cpumask);
 381                 if (cpumask_empty(cpumask))
 382                         break;
 383                 cpu_relax();
 384         }
 385 }
 386
 387 /*
 388  * Stop all cpus but the current one.
 389  */
 390 void smp_send_stop(void)
 391 {
 392         cpumask_t cpumask;
 393         int cpu;
 394
 395         /* Disable all interrupts/machine checks */
 396         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 397         trace_hardirqs_off();
 398
 399         debug_set_critical();
 400         cpumask_copy(&cpumask, cpu_online_mask);
 401         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 402
 403         if (oops_in_progress)
 404                 smp_emergency_stop(&cpumask);
 405
 406         /* stop all processors */
 407         for_each_cpu(cpu, &cpumask) {
 408                 struct pcpu *pcpu = pcpu_devices + cpu;
 409                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 410                 while (!pcpu_stopped(pcpu))
 411                         cpu_relax();
 412         }
 413 }
 414
 415 /*
 416  * Stop the current cpu.
 417  */
 418 void smp_stop_cpu(void)
 419 {
 420         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 421         for (;;) ;
 422 }
 423
 424 /*
 425  * This is the main routine where commands issued by other
 426  * cpus are handled.
 427  */
 428 static void do_ext_call_interrupt(struct ext_code ext_code,
 429                                   unsigned int param32, unsigned long param64)
 430 {
 431         unsigned long bits;
 432         int cpu;
 433
 434         cpu = smp_processor_id();
 435         if (ext_code.code == 0x1202)
 436                 inc_irq_stat(IRQEXT_EXC);
 437         else
 438                 inc_irq_stat(IRQEXT_EMS);
 439         /*
 440          * handle bit signal external calls
 441          */
 442         bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
 443
 444         if (test_bit(ec_stop_cpu, &bits))
 445                 smp_stop_cpu();
 446
 447         if (test_bit(ec_schedule, &bits))
 448                 scheduler_ipi();
 449
 450         if (test_bit(ec_call_function, &bits))
 451                 generic_smp_call_function_interrupt();
 452
 453         if (test_bit(ec_call_function_single, &bits))
 454                 generic_smp_call_function_single_interrupt();
 455
 456 }
 457
 458 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 459 {
 460         int cpu;
 461
 462         for_each_cpu(cpu, mask)
 463                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
 464 }
 465
 466 void arch_send_call_function_single_ipi(int cpu)
 467 {
 468         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 469 }
 470
 471 #ifndef CONFIG_64BIT
 472 /*
 473  * this function sends a 'purge tlb' signal to another CPU.
 474  */
 475 static void smp_ptlb_callback(void *info)
 476 {
 477         __tlb_flush_local();
 478 }
 479
 480 void smp_ptlb_all(void)
 481 {
 482         on_each_cpu(smp_ptlb_callback, NULL, 1);
 483 }
 484 EXPORT_SYMBOL(smp_ptlb_all);
 485 #endif /* ! CONFIG_64BIT */
 486
 487 /*
 488  * this function sends a 'reschedule' IPI to another CPU.
 489  * it goes straight through and wastes no time serializing
 490  * anything. Worst case is that we lose a reschedule ...
 491  */
 492 void smp_send_reschedule(int cpu)
 493 {
 494         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 495 }
 496
 497 /*
 498  * parameter area for the set/clear control bit callbacks
 499  */
 500 struct ec_creg_mask_parms {
 501         unsigned long orval;
 502         unsigned long andval;
 503         int cr;
 504 };
 505
 506 /*
 507  * callback for setting/clearing control bits
 508  */
 509 static void smp_ctl_bit_callback(void *info)
 510 {
 511         struct ec_creg_mask_parms *pp = info;
 512         unsigned long cregs[16];
 513
 514         __ctl_store(cregs, 0, 15);
 515         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 516         __ctl_load(cregs, 0, 15);
 517 }
 518
 519 /*
 520  * Set a bit in a control register of all cpus
 521  */
 522 void smp_ctl_set_bit(int cr, int bit)
 523 {
 524         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 525
 526         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 527 }
 528 EXPORT_SYMBOL(smp_ctl_set_bit);
 529
 530 /*
 531  * Clear a bit in a control register of all cpus
 532  */
 533 void smp_ctl_clear_bit(int cr, int bit)
 534 {
 535         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 536
 537         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 538 }
 539 EXPORT_SYMBOL(smp_ctl_clear_bit);
 540
 541 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
 542
 543 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 544 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 545
 546 static void __init smp_get_save_area(int cpu, u16 address)
 547 {
 548         void *lc = pcpu_devices[0].lowcore;
 549         struct save_area *save_area;
 550
 551         if (is_kdump_kernel())
 552                 return;
 553         if (!OLDMEM_BASE && (address == boot_cpu_address ||
 554                              ipl_info.type != IPL_TYPE_FCP_DUMP))
 555                 return;
 556         if (cpu >= NR_CPUS) {
 557                 pr_warning("CPU %i exceeds the maximum %i and is excluded "
 558                            "from the dump\n", cpu, NR_CPUS - 1);
 559                 return;
 560         }
 561         save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 562         if (!save_area)
 563                 panic("could not allocate memory for save area\n");
 564         zfcpdump_save_areas[cpu] = save_area;
 565 #ifdef CONFIG_CRASH_DUMP
 566         if (address == boot_cpu_address) {
 567                 /* Copy the registers of the boot cpu. */
 568                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 569                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
 570                 return;
 571         }
 572 #endif
 573         /* Get the registers of a non-boot cpu. */
 574         __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
 575         memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
 576 }
 577
 578 int smp_store_status(int cpu)
 579 {
 580         struct pcpu *pcpu;
 581
 582         pcpu = pcpu_devices + cpu;
 583         if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
 584                               0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
 585                 return -EIO;
 586         return 0;
 587 }
 588
 589 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 590
 591 static inline void smp_get_save_area(int cpu, u16 address) { }
 592
 593 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 594
 595 void smp_cpu_set_polarization(int cpu, int val)
 596 {
 597         pcpu_devices[cpu].polarization = val;
 598 }
 599
 600 int smp_cpu_get_polarization(int cpu)
 601 {
 602         return pcpu_devices[cpu].polarization;
 603 }
 604
 605 static struct sclp_cpu_info *smp_get_cpu_info(void)
 606 {
 607         static int use_sigp_detection;
 608         struct sclp_cpu_info *info;
 609         int address;
 610
 611         info = kzalloc(sizeof(*info), GFP_KERNEL);
 612         if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
 613                 use_sigp_detection = 1;
 614                 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
 615                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
 616                             SIGP_CC_NOT_OPERATIONAL)
 617                                 continue;
 618                         info->cpu[info->configured].address = address;
 619                         info->configured++;
 620                 }
 621                 info->combined = info->configured;
 622         }
 623         return info;
 624 }
 625
 626 static int __cpuinit smp_add_present_cpu(int cpu);
 627
 628 static int __cpuinit __smp_rescan_cpus(struct sclp_cpu_info *info,
 629                                        int sysfs_add)
 630 {
 631         struct pcpu *pcpu;
 632         cpumask_t avail;
 633         int cpu, nr, i;
 634
 635         nr = 0;
 636         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 637         cpu = cpumask_first(&avail);
 638         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 639                 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
 640                         continue;
 641                 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
 642                         continue;
 643                 pcpu = pcpu_devices + cpu;
 644                 pcpu->address = info->cpu[i].address;
 645                 pcpu->state = (cpu >= info->configured) ?
 646                         CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 647                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 648                 set_cpu_present(cpu, true);
 649                 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 650                         set_cpu_present(cpu, false);
 651                 else
 652                         nr++;
 653                 cpu = cpumask_next(cpu, &avail);
 654         }
 655         return nr;
 656 }
 657
 658 static void __init smp_detect_cpus(void)
 659 {
 660         unsigned int cpu, c_cpus, s_cpus;
 661         struct sclp_cpu_info *info;
 662
 663         info = smp_get_cpu_info();
 664         if (!info)
 665                 panic("smp_detect_cpus failed to allocate memory\n");
 666         if (info->has_cpu_type) {
 667                 for (cpu = 0; cpu < info->combined; cpu++) {
 668                         if (info->cpu[cpu].address != boot_cpu_address)
 669                                 continue;
 670                         /* The boot cpu dictates the cpu type. */
 671                         boot_cpu_type = info->cpu[cpu].type;
 672                         break;
 673                 }
 674         }
 675         c_cpus = s_cpus = 0;
 676         for (cpu = 0; cpu < info->combined; cpu++) {
 677                 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
 678                         continue;
 679                 if (cpu < info->configured) {
 680                         smp_get_save_area(c_cpus, info->cpu[cpu].address);
 681                         c_cpus++;
 682                 } else
 683                         s_cpus++;
 684         }
 685         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 686         get_online_cpus();
 687         __smp_rescan_cpus(info, 0);
 688         put_online_cpus();
 689         kfree(info);
 690 }
 691
 692 /*
 693  *      Activate a secondary processor.
 694  */
 695 static void __cpuinit smp_start_secondary(void *cpuvoid)
 696 {
 697         S390_lowcore.last_update_clock = get_clock();
 698         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 699         S390_lowcore.restart_fn = (unsigned long) do_restart;
 700         S390_lowcore.restart_data = 0;
 701         S390_lowcore.restart_source = -1UL;
 702         restore_access_regs(S390_lowcore.access_regs_save_area);
 703         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 704         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 705         cpu_init();
 706         preempt_disable();
 707         init_cpu_timer();
 708         init_cpu_vtimer();
 709         pfault_init();
 710         notify_cpu_starting(smp_processor_id());
 711         set_cpu_online(smp_processor_id(), true);
 712         inc_irq_stat(CPU_RST);
 713         local_irq_enable();
 714         /* cpu_idle will call schedule for us */
 715         cpu_idle();
 716 }
 717
 718 /* Upping and downing of CPUs */
 719 int __cpuinit __cpu_up(unsigned int cpu, struct task_struct *tidle)
 720 {
 721         struct pcpu *pcpu;
 722         int rc;
 723
 724         pcpu = pcpu_devices + cpu;
 725         if (pcpu->state != CPU_STATE_CONFIGURED)
 726                 return -EIO;
 727         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 728             SIGP_CC_ORDER_CODE_ACCEPTED)
 729                 return -EIO;
 730
 731         rc = pcpu_alloc_lowcore(pcpu, cpu);
 732         if (rc)
 733                 return rc;
 734         pcpu_prepare_secondary(pcpu, cpu);
 735         pcpu_attach_task(pcpu, tidle);
 736         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 737         while (!cpu_online(cpu))
 738                 cpu_relax();
 739         return 0;
 740 }
 741
 742 static int __init setup_possible_cpus(char *s)
 743 {
 744         int max, cpu;
 745
 746         if (kstrtoint(s, 0, &max) < 0)
 747                 return 0;
 748         init_cpu_possible(cpumask_of(0));
 749         for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
 750                 set_cpu_possible(cpu, true);
 751         return 0;
 752 }
 753 early_param("possible_cpus", setup_possible_cpus);
 754
 755 #ifdef CONFIG_HOTPLUG_CPU
 756
 757 int __cpu_disable(void)
 758 {
 759         unsigned long cregs[16];
 760
 761         set_cpu_online(smp_processor_id(), false);
 762         /* Disable pseudo page faults on this cpu. */
 763         pfault_fini();
 764         /* Disable interrupt sources via control register. */
 765         __ctl_store(cregs, 0, 15);
 766         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 767         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 768         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 769         __ctl_load(cregs, 0, 15);
 770         return 0;
 771 }
 772
 773 void __cpu_die(unsigned int cpu)
 774 {
 775         struct pcpu *pcpu;
 776
 777         /* Wait until target cpu is down */
 778         pcpu = pcpu_devices + cpu;
 779         while (!pcpu_stopped(pcpu))
 780                 cpu_relax();
 781         pcpu_free_lowcore(pcpu);
 782         atomic_dec(&init_mm.context.attach_count);
 783 }
 784
 785 void __noreturn cpu_die(void)
 786 {
 787         idle_task_exit();
 788         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 789         for (;;) ;
 790 }
 791
 792 #endif /* CONFIG_HOTPLUG_CPU */
 793
 794 void __init smp_prepare_cpus(unsigned int max_cpus)
 795 {
 796         /* request the 0x1201 emergency signal external interrupt */
 797         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 798                 panic("Couldn't request external interrupt 0x1201");
 799         /* request the 0x1202 external call external interrupt */
 800         if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
 801                 panic("Couldn't request external interrupt 0x1202");
 802         smp_detect_cpus();
 803 }
 804
 805 void __init smp_prepare_boot_cpu(void)
 806 {
 807         struct pcpu *pcpu = pcpu_devices;
 808
 809         boot_cpu_address = stap();
 810         pcpu->state = CPU_STATE_CONFIGURED;
 811         pcpu->address = boot_cpu_address;
 812         pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
 813         pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
 814         pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
 815         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 816         smp_cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 817         set_cpu_present(0, true);
 818         set_cpu_online(0, true);
 819 }
 820
 821 void __init smp_cpus_done(unsigned int max_cpus)
 822 {
 823 }
 824
 825 void __init smp_setup_processor_id(void)
 826 {
 827         S390_lowcore.cpu_nr = 0;
 828 }
 829
 830 /*
 831  * the frequency of the profiling timer can be changed
 832  * by writing a multiplier value into /proc/profile.
 833  *
 834  * usually you want to run this on all CPUs ;)
 835  */
 836 int setup_profiling_timer(unsigned int multiplier)
 837 {
 838         return 0;
 839 }
 840
 841 #ifdef CONFIG_HOTPLUG_CPU
 842 static ssize_t cpu_configure_show(struct device *dev,
 843                                   struct device_attribute *attr, char *buf)
 844 {
 845         ssize_t count;
 846
 847         mutex_lock(&smp_cpu_state_mutex);
 848         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 849         mutex_unlock(&smp_cpu_state_mutex);
 850         return count;
 851 }
 852
 853 static ssize_t cpu_configure_store(struct device *dev,
 854                                    struct device_attribute *attr,
 855                                    const char *buf, size_t count)
 856 {
 857         struct pcpu *pcpu;
 858         int cpu, val, rc;
 859         char delim;
 860
 861         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 862                 return -EINVAL;
 863         if (val != 0 && val != 1)
 864                 return -EINVAL;
 865         get_online_cpus();
 866         mutex_lock(&smp_cpu_state_mutex);
 867         rc = -EBUSY;
 868         /* disallow configuration changes of online cpus and cpu 0 */
 869         cpu = dev->id;
 870         if (cpu_online(cpu) || cpu == 0)
 871                 goto out;
 872         pcpu = pcpu_devices + cpu;
 873         rc = 0;
 874         switch (val) {
 875         case 0:
 876                 if (pcpu->state != CPU_STATE_CONFIGURED)
 877                         break;
 878                 rc = sclp_cpu_deconfigure(pcpu->address);
 879                 if (rc)
 880                         break;
 881                 pcpu->state = CPU_STATE_STANDBY;
 882                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 883                 topology_expect_change();
 884                 break;
 885         case 1:
 886                 if (pcpu->state != CPU_STATE_STANDBY)
 887                         break;
 888                 rc = sclp_cpu_configure(pcpu->address);
 889                 if (rc)
 890                         break;
 891                 pcpu->state = CPU_STATE_CONFIGURED;
 892                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 893                 topology_expect_change();
 894                 break;
 895         default:
 896                 break;
 897         }
 898 out:
 899         mutex_unlock(&smp_cpu_state_mutex);
 900         put_online_cpus();
 901         return rc ? rc : count;
 902 }
 903 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 904 #endif /* CONFIG_HOTPLUG_CPU */
 905
 906 static ssize_t show_cpu_address(struct device *dev,
 907                                 struct device_attribute *attr, char *buf)
 908 {
 909         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
 910 }
 911 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
 912
 913 static struct attribute *cpu_common_attrs[] = {
 914 #ifdef CONFIG_HOTPLUG_CPU
 915         &dev_attr_configure.attr,
 916 #endif
 917         &dev_attr_address.attr,
 918         NULL,
 919 };
 920
 921 static struct attribute_group cpu_common_attr_group = {
 922         .attrs = cpu_common_attrs,
 923 };
 924
 925 static ssize_t show_idle_count(struct device *dev,
 926                                 struct device_attribute *attr, char *buf)
 927 {
 928         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 929         unsigned long long idle_count;
 930         unsigned int sequence;
 931
 932         do {
 933                 sequence = ACCESS_ONCE(idle->sequence);
 934                 idle_count = ACCESS_ONCE(idle->idle_count);
 935                 if (ACCESS_ONCE(idle->clock_idle_enter))
 936                         idle_count++;
 937         } while ((sequence & 1) || (idle->sequence != sequence));
 938         return sprintf(buf, "%llu\n", idle_count);
 939 }
 940 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
 941
 942 static ssize_t show_idle_time(struct device *dev,
 943                                 struct device_attribute *attr, char *buf)
 944 {
 945         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 946         unsigned long long now, idle_time, idle_enter, idle_exit;
 947         unsigned int sequence;
 948
 949         do {
 950                 now = get_clock();
 951                 sequence = ACCESS_ONCE(idle->sequence);
 952                 idle_time = ACCESS_ONCE(idle->idle_time);
 953                 idle_enter = ACCESS_ONCE(idle->clock_idle_enter);
 954                 idle_exit = ACCESS_ONCE(idle->clock_idle_exit);
 955         } while ((sequence & 1) || (idle->sequence != sequence));
 956         idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
 957         return sprintf(buf, "%llu\n", idle_time >> 12);
 958 }
 959 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 960
 961 static struct attribute *cpu_online_attrs[] = {
 962         &dev_attr_idle_count.attr,
 963         &dev_attr_idle_time_us.attr,
 964         NULL,
 965 };
 966
 967 static struct attribute_group cpu_online_attr_group = {
 968         .attrs = cpu_online_attrs,
 969 };
 970
 971 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 972                                     unsigned long action, void *hcpu)
 973 {
 974         unsigned int cpu = (unsigned int)(long)hcpu;
 975         struct cpu *c = &pcpu_devices[cpu].cpu;
 976         struct device *s = &c->dev;
 977         int err = 0;
 978
 979         switch (action & ~CPU_TASKS_FROZEN) {
 980         case CPU_ONLINE:
 981                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 982                 break;
 983         case CPU_DEAD:
 984                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 985                 break;
 986         }
 987         return notifier_from_errno(err);
 988 }
 989
 990 static int __cpuinit smp_add_present_cpu(int cpu)
 991 {
 992         struct cpu *c = &pcpu_devices[cpu].cpu;
 993         struct device *s = &c->dev;
 994         int rc;
 995
 996         c->hotpluggable = 1;
 997         rc = register_cpu(c, cpu);
 998         if (rc)
 999                 goto out;
1000         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1001         if (rc)
1002                 goto out_cpu;
1003         if (cpu_online(cpu)) {
1004                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1005                 if (rc)
1006                         goto out_online;
1007         }
1008         rc = topology_cpu_init(c);
1009         if (rc)
1010                 goto out_topology;
1011         return 0;
1012
1013 out_topology:
1014         if (cpu_online(cpu))
1015                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1016 out_online:
1017         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1018 out_cpu:
1019 #ifdef CONFIG_HOTPLUG_CPU
1020         unregister_cpu(c);
1021 #endif
1022 out:
1023         return rc;
1024 }
1025
1026 #ifdef CONFIG_HOTPLUG_CPU
1027
1028 int __ref smp_rescan_cpus(void)
1029 {
1030         struct sclp_cpu_info *info;
1031         int nr;
1032
1033         info = smp_get_cpu_info();
1034         if (!info)
1035                 return -ENOMEM;
1036         get_online_cpus();
1037         mutex_lock(&smp_cpu_state_mutex);
1038         nr = __smp_rescan_cpus(info, 1);
1039         mutex_unlock(&smp_cpu_state_mutex);
1040         put_online_cpus();
1041         kfree(info);
1042         if (nr)
1043                 topology_schedule_update();
1044         return 0;
1045 }
1046
1047 static ssize_t __ref rescan_store(struct device *dev,
1048                                   struct device_attribute *attr,
1049                                   const char *buf,
1050                                   size_t count)
1051 {
1052         int rc;
1053
1054         rc = smp_rescan_cpus();
1055         return rc ? rc : count;
1056 }
1057 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1058 #endif /* CONFIG_HOTPLUG_CPU */
1059
1060 static int __init s390_smp_init(void)
1061 {
1062         int cpu, rc;
1063
1064         hotcpu_notifier(smp_cpu_notify, 0);
1065 #ifdef CONFIG_HOTPLUG_CPU
1066         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1067         if (rc)
1068                 return rc;
1069 #endif
1070         for_each_present_cpu(cpu) {
1071                 rc = smp_add_present_cpu(cpu);
1072                 if (rc)
1073                         return rc;
1074         }
1075         return 0;
1076 }
1077 subsys_initcall(s390_smp_init);