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