Pileus Git - ~andy/linux/blob - 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/ipl.h>
  36 #include <asm/setup.h>
  37 #include <asm/irq.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/timer.h>
  40 #include <asm/lowcore.h>
  41 #include <asm/sclp.h>
  42 #include <asm/vdso.h>
  43 #include <asm/debug.h>
  44 #include <asm/os_info.h>
  45 #include "entry.h"
  46
  47 enum {
  48         sigp_sense = 1,
  49         sigp_external_call = 2,
  50         sigp_emergency_signal = 3,
  51         sigp_start = 4,
  52         sigp_stop = 5,
  53         sigp_restart = 6,
  54         sigp_stop_and_store_status = 9,
  55         sigp_initial_cpu_reset = 11,
  56         sigp_cpu_reset = 12,
  57         sigp_set_prefix = 13,
  58         sigp_store_status_at_address = 14,
  59         sigp_store_extended_status_at_address = 15,
  60         sigp_set_architecture = 18,
  61         sigp_conditional_emergency_signal = 19,
  62         sigp_sense_running = 21,
  63 };
  64
  65 enum {
  66         sigp_order_code_accepted = 0,
  67         sigp_status_stored = 1,
  68         sigp_busy = 2,
  69         sigp_not_operational = 3,
  70 };
  71
  72 enum {
  73         ec_schedule = 0,
  74         ec_call_function,
  75         ec_call_function_single,
  76         ec_stop_cpu,
  77 };
  78
  79 enum {
  80         CPU_STATE_STANDBY,
  81         CPU_STATE_CONFIGURED,
  82 };
  83
  84 struct pcpu {
  85         struct cpu cpu;
  86         struct task_struct *idle;       /* idle process for the cpu */
  87         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
  88         unsigned long async_stack;      /* async stack for the cpu */
  89         unsigned long panic_stack;      /* panic stack for the cpu */
  90         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  91         int state;                      /* physical cpu state */
  92         u32 status;                     /* last status received via sigp */
  93         u16 address;                    /* physical cpu address */
  94 };
  95
  96 static u8 boot_cpu_type;
  97 static u16 boot_cpu_address;
  98 static struct pcpu pcpu_devices[NR_CPUS];
  99
 100 DEFINE_MUTEX(smp_cpu_state_mutex);
 101
 102 /*
 103  * Signal processor helper functions.
 104  */
 105 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
 106 {
 107         register unsigned int reg1 asm ("1") = parm;
 108         int cc;
 109
 110         asm volatile(
 111                 "       sigp    %1,%2,0(%3)\n"
 112                 "       ipm     %0\n"
 113                 "       srl     %0,28\n"
 114                 : "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
 115         if (status && cc == 1)
 116                 *status = reg1;
 117         return cc;
 118 }
 119
 120 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
 121 {
 122         int cc;
 123
 124         while (1) {
 125                 cc = __pcpu_sigp(addr, order, parm, status);
 126                 if (cc != sigp_busy)
 127                         return cc;
 128                 cpu_relax();
 129         }
 130 }
 131
 132 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 133 {
 134         int cc, retry;
 135
 136         for (retry = 0; ; retry++) {
 137                 cc = __pcpu_sigp(pcpu->address, order, parm, &pcpu->status);
 138                 if (cc != sigp_busy)
 139                         break;
 140                 if (retry >= 3)
 141                         udelay(10);
 142         }
 143         return cc;
 144 }
 145
 146 static inline int pcpu_stopped(struct pcpu *pcpu)
 147 {
 148         if (__pcpu_sigp(pcpu->address, sigp_sense,
 149                         0, &pcpu->status) != sigp_status_stored)
 150                 return 0;
 151         /* Check for stopped and check stop state */
 152         return !!(pcpu->status & 0x50);
 153 }
 154
 155 static inline int pcpu_running(struct pcpu *pcpu)
 156 {
 157         if (__pcpu_sigp(pcpu->address, sigp_sense_running,
 158                         0, &pcpu->status) != sigp_status_stored)
 159                 return 1;
 160         /* Check for running status */
 161         return !(pcpu->status & 0x400);
 162 }
 163
 164 /*
 165  * Find struct pcpu by cpu address.
 166  */
 167 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
 168 {
 169         int cpu;
 170
 171         for_each_cpu(cpu, mask)
 172                 if (pcpu_devices[cpu].address == address)
 173                         return pcpu_devices + cpu;
 174         return NULL;
 175 }
 176
 177 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 178 {
 179         int order;
 180
 181         set_bit(ec_bit, &pcpu->ec_mask);
 182         order = pcpu_running(pcpu) ?
 183                 sigp_external_call : sigp_emergency_signal;
 184         pcpu_sigp_retry(pcpu, order, 0);
 185 }
 186
 187 static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 188 {
 189         struct _lowcore *lc;
 190
 191         if (pcpu != &pcpu_devices[0]) {
 192                 pcpu->lowcore = (struct _lowcore *)
 193                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 194                 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 195                 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
 196                 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
 197                         goto out;
 198         }
 199         lc = pcpu->lowcore;
 200         memcpy(lc, &S390_lowcore, 512);
 201         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 202         lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
 203         lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
 204         lc->cpu_nr = cpu;
 205 #ifndef CONFIG_64BIT
 206         if (MACHINE_HAS_IEEE) {
 207                 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
 208                 if (!lc->extended_save_area_addr)
 209                         goto out;
 210         }
 211 #else
 212         if (vdso_alloc_per_cpu(lc))
 213                 goto out;
 214 #endif
 215         lowcore_ptr[cpu] = lc;
 216         pcpu_sigp_retry(pcpu, sigp_set_prefix, (u32)(unsigned long) lc);
 217         return 0;
 218 out:
 219         if (pcpu != &pcpu_devices[0]) {
 220                 free_page(pcpu->panic_stack);
 221                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 222                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 223         }
 224         return -ENOMEM;
 225 }
 226
 227 static void pcpu_free_lowcore(struct pcpu *pcpu)
 228 {
 229         pcpu_sigp_retry(pcpu, sigp_set_prefix, 0);
 230         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 231 #ifndef CONFIG_64BIT
 232         if (MACHINE_HAS_IEEE) {
 233                 struct _lowcore *lc = pcpu->lowcore;
 234
 235                 free_page((unsigned long) lc->extended_save_area_addr);
 236                 lc->extended_save_area_addr = 0;
 237         }
 238 #else
 239         vdso_free_per_cpu(pcpu->lowcore);
 240 #endif
 241         if (pcpu != &pcpu_devices[0]) {
 242                 free_page(pcpu->panic_stack);
 243                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 244                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 245         }
 246 }
 247
 248 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 249 {
 250         struct _lowcore *lc = pcpu->lowcore;
 251
 252         atomic_inc(&init_mm.context.attach_count);
 253         lc->cpu_nr = cpu;
 254         lc->percpu_offset = __per_cpu_offset[cpu];
 255         lc->kernel_asce = S390_lowcore.kernel_asce;
 256         lc->machine_flags = S390_lowcore.machine_flags;
 257         lc->ftrace_func = S390_lowcore.ftrace_func;
 258         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 259         __ctl_store(lc->cregs_save_area, 0, 15);
 260         save_access_regs((unsigned int *) lc->access_regs_save_area);
 261         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 262                MAX_FACILITY_BIT/8);
 263 }
 264
 265 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 266 {
 267         struct _lowcore *lc = pcpu->lowcore;
 268         struct thread_info *ti = task_thread_info(tsk);
 269
 270         lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
 271         lc->thread_info = (unsigned long) task_thread_info(tsk);
 272         lc->current_task = (unsigned long) tsk;
 273         lc->user_timer = ti->user_timer;
 274         lc->system_timer = ti->system_timer;
 275         lc->steal_timer = 0;
 276 }
 277
 278 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 279 {
 280         struct _lowcore *lc = pcpu->lowcore;
 281
 282         lc->restart_stack = lc->kernel_stack;
 283         lc->restart_fn = (unsigned long) func;
 284         lc->restart_data = (unsigned long) data;
 285         lc->restart_source = -1UL;
 286         pcpu_sigp_retry(pcpu, sigp_restart, 0);
 287 }
 288
 289 /*
 290  * Call function via PSW restart on pcpu and stop the current cpu.
 291  */
 292 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 293                           void *data, unsigned long stack)
 294 {
 295         struct _lowcore *lc = pcpu->lowcore;
 296         unsigned short this_cpu;
 297
 298         __load_psw_mask(psw_kernel_bits);
 299         this_cpu = stap();
 300         if (pcpu->address == this_cpu)
 301                 func(data);     /* should not return */
 302         /* Stop target cpu (if func returns this stops the current cpu). */
 303         pcpu_sigp_retry(pcpu, sigp_stop, 0);
 304         /* Restart func on the target cpu and stop the current cpu. */
 305         lc->restart_stack = stack;
 306         lc->restart_fn = (unsigned long) func;
 307         lc->restart_data = (unsigned long) data;
 308         lc->restart_source = (unsigned long) this_cpu;
 309         asm volatile(
 310                 "0:     sigp    0,%0,6  # sigp restart to target cpu\n"
 311                 "       brc     2,0b    # busy, try again\n"
 312                 "1:     sigp    0,%1,5  # sigp stop to current cpu\n"
 313                 "       brc     2,1b    # busy, try again\n"
 314                 : : "d" (pcpu->address), "d" (this_cpu) : "0", "1", "cc");
 315         for (;;) ;
 316 }
 317
 318 /*
 319  * Call function on an online CPU.
 320  */
 321 void smp_call_online_cpu(void (*func)(void *), void *data)
 322 {
 323         struct pcpu *pcpu;
 324
 325         /* Use the current cpu if it is online. */
 326         pcpu = pcpu_find_address(cpu_online_mask, stap());
 327         if (!pcpu)
 328                 /* Use the first online cpu. */
 329                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 330         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 331 }
 332
 333 /*
 334  * Call function on the ipl CPU.
 335  */
 336 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 337 {
 338         pcpu_delegate(&pcpu_devices[0], func, data,
 339                       pcpu_devices->panic_stack + PAGE_SIZE);
 340 }
 341
 342 int smp_find_processor_id(u16 address)
 343 {
 344         int cpu;
 345
 346         for_each_present_cpu(cpu)
 347                 if (pcpu_devices[cpu].address == address)
 348                         return cpu;
 349         return -1;
 350 }
 351
 352 int smp_vcpu_scheduled(int cpu)
 353 {
 354         return pcpu_running(pcpu_devices + cpu);
 355 }
 356
 357 void smp_yield(void)
 358 {
 359         if (MACHINE_HAS_DIAG44)
 360                 asm volatile("diag 0,0,0x44");
 361 }
 362
 363 void smp_yield_cpu(int cpu)
 364 {
 365         if (MACHINE_HAS_DIAG9C)
 366                 asm volatile("diag %0,0,0x9c"
 367                              : : "d" (pcpu_devices[cpu].address));
 368         else if (MACHINE_HAS_DIAG44)
 369                 asm volatile("diag 0,0,0x44");
 370 }
 371
 372 /*
 373  * Send cpus emergency shutdown signal. This gives the cpus the
 374  * opportunity to complete outstanding interrupts.
 375  */
 376 void smp_emergency_stop(cpumask_t *cpumask)
 377 {
 378         u64 end;
 379         int cpu;
 380
 381         end = get_clock() + (1000000UL << 12);
 382         for_each_cpu(cpu, cpumask) {
 383                 struct pcpu *pcpu = pcpu_devices + cpu;
 384                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 385                 while (__pcpu_sigp(pcpu->address, sigp_emergency_signal,
 386                                    0, NULL) == sigp_busy &&
 387                        get_clock() < end)
 388                         cpu_relax();
 389         }
 390         while (get_clock() < end) {
 391                 for_each_cpu(cpu, cpumask)
 392                         if (pcpu_stopped(pcpu_devices + cpu))
 393                                 cpumask_clear_cpu(cpu, cpumask);
 394                 if (cpumask_empty(cpumask))
 395                         break;
 396                 cpu_relax();
 397         }
 398 }
 399
 400 /*
 401  * Stop all cpus but the current one.
 402  */
 403 void smp_send_stop(void)
 404 {
 405         cpumask_t cpumask;
 406         int cpu;
 407
 408         /* Disable all interrupts/machine checks */
 409         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 410         trace_hardirqs_off();
 411
 412         debug_set_critical();
 413         cpumask_copy(&cpumask, cpu_online_mask);
 414         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 415
 416         if (oops_in_progress)
 417                 smp_emergency_stop(&cpumask);
 418
 419         /* stop all processors */
 420         for_each_cpu(cpu, &cpumask) {
 421                 struct pcpu *pcpu = pcpu_devices + cpu;
 422                 pcpu_sigp_retry(pcpu, sigp_stop, 0);
 423                 while (!pcpu_stopped(pcpu))
 424                         cpu_relax();
 425         }
 426 }
 427
 428 /*
 429  * Stop the current cpu.
 430  */
 431 void smp_stop_cpu(void)
 432 {
 433         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
 434         for (;;) ;
 435 }
 436
 437 /*
 438  * This is the main routine where commands issued by other
 439  * cpus are handled.
 440  */
 441 static void do_ext_call_interrupt(struct ext_code ext_code,
 442                                   unsigned int param32, unsigned long param64)
 443 {
 444         unsigned long bits;
 445         int cpu;
 446
 447         cpu = smp_processor_id();
 448         if (ext_code.code == 0x1202)
 449                 kstat_cpu(cpu).irqs[EXTINT_EXC]++;
 450         else
 451                 kstat_cpu(cpu).irqs[EXTINT_EMS]++;
 452         /*
 453          * handle bit signal external calls
 454          */
 455         bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
 456
 457         if (test_bit(ec_stop_cpu, &bits))
 458                 smp_stop_cpu();
 459
 460         if (test_bit(ec_schedule, &bits))
 461                 scheduler_ipi();
 462
 463         if (test_bit(ec_call_function, &bits))
 464                 generic_smp_call_function_interrupt();
 465
 466         if (test_bit(ec_call_function_single, &bits))
 467                 generic_smp_call_function_single_interrupt();
 468
 469 }
 470
 471 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 472 {
 473         int cpu;
 474
 475         for_each_cpu(cpu, mask)
 476                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
 477 }
 478
 479 void arch_send_call_function_single_ipi(int cpu)
 480 {
 481         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 482 }
 483
 484 #ifndef CONFIG_64BIT
 485 /*
 486  * this function sends a 'purge tlb' signal to another CPU.
 487  */
 488 static void smp_ptlb_callback(void *info)
 489 {
 490         __tlb_flush_local();
 491 }
 492
 493 void smp_ptlb_all(void)
 494 {
 495         on_each_cpu(smp_ptlb_callback, NULL, 1);
 496 }
 497 EXPORT_SYMBOL(smp_ptlb_all);
 498 #endif /* ! CONFIG_64BIT */
 499
 500 /*
 501  * this function sends a 'reschedule' IPI to another CPU.
 502  * it goes straight through and wastes no time serializing
 503  * anything. Worst case is that we lose a reschedule ...
 504  */
 505 void smp_send_reschedule(int cpu)
 506 {
 507         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 508 }
 509
 510 /*
 511  * parameter area for the set/clear control bit callbacks
 512  */
 513 struct ec_creg_mask_parms {
 514         unsigned long orval;
 515         unsigned long andval;
 516         int cr;
 517 };
 518
 519 /*
 520  * callback for setting/clearing control bits
 521  */
 522 static void smp_ctl_bit_callback(void *info)
 523 {
 524         struct ec_creg_mask_parms *pp = info;
 525         unsigned long cregs[16];
 526
 527         __ctl_store(cregs, 0, 15);
 528         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 529         __ctl_load(cregs, 0, 15);
 530 }
 531
 532 /*
 533  * Set a bit in a control register of all cpus
 534  */
 535 void smp_ctl_set_bit(int cr, int bit)
 536 {
 537         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 538
 539         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 540 }
 541 EXPORT_SYMBOL(smp_ctl_set_bit);
 542
 543 /*
 544  * Clear a bit in a control register of all cpus
 545  */
 546 void smp_ctl_clear_bit(int cr, int bit)
 547 {
 548         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 549
 550         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 551 }
 552 EXPORT_SYMBOL(smp_ctl_clear_bit);
 553
 554 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
 555
 556 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 557 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 558
 559 static void __init smp_get_save_area(int cpu, u16 address)
 560 {
 561         void *lc = pcpu_devices[0].lowcore;
 562         struct save_area *save_area;
 563
 564         if (is_kdump_kernel())
 565                 return;
 566         if (!OLDMEM_BASE && (address == boot_cpu_address ||
 567                              ipl_info.type != IPL_TYPE_FCP_DUMP))
 568                 return;
 569         if (cpu >= NR_CPUS) {
 570                 pr_warning("CPU %i exceeds the maximum %i and is excluded "
 571                            "from the dump\n", cpu, NR_CPUS - 1);
 572                 return;
 573         }
 574         save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 575         if (!save_area)
 576                 panic("could not allocate memory for save area\n");
 577         zfcpdump_save_areas[cpu] = save_area;
 578 #ifdef CONFIG_CRASH_DUMP
 579         if (address == boot_cpu_address) {
 580                 /* Copy the registers of the boot cpu. */
 581                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 582                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
 583                 return;
 584         }
 585 #endif
 586         /* Get the registers of a non-boot cpu. */
 587         __pcpu_sigp_relax(address, sigp_stop_and_store_status, 0, NULL);
 588         memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
 589 }
 590
 591 int smp_store_status(int cpu)
 592 {
 593         struct pcpu *pcpu;
 594
 595         pcpu = pcpu_devices + cpu;
 596         if (__pcpu_sigp_relax(pcpu->address, sigp_stop_and_store_status,
 597                               0, NULL) != sigp_order_code_accepted)
 598                 return -EIO;
 599         return 0;
 600 }
 601
 602 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 603
 604 static inline void smp_get_save_area(int cpu, u16 address) { }
 605
 606 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 607
 608 static struct sclp_cpu_info *smp_get_cpu_info(void)
 609 {
 610         static int use_sigp_detection;
 611         struct sclp_cpu_info *info;
 612         int address;
 613
 614         info = kzalloc(sizeof(*info), GFP_KERNEL);
 615         if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
 616                 use_sigp_detection = 1;
 617                 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
 618                         if (__pcpu_sigp_relax(address, sigp_sense, 0, NULL) ==
 619                             sigp_not_operational)
 620                                 continue;
 621                         info->cpu[info->configured].address = address;
 622                         info->configured++;
 623                 }
 624                 info->combined = info->configured;
 625         }
 626         return info;
 627 }
 628
 629 static int __devinit smp_add_present_cpu(int cpu);
 630
 631 static int __devinit __smp_rescan_cpus(struct sclp_cpu_info *info,
 632                                        int sysfs_add)
 633 {
 634         struct pcpu *pcpu;
 635         cpumask_t avail;
 636         int cpu, nr, i;
 637
 638         nr = 0;
 639         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 640         cpu = cpumask_first(&avail);
 641         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 642                 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
 643                         continue;
 644                 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
 645                         continue;
 646                 pcpu = pcpu_devices + cpu;
 647                 pcpu->address = info->cpu[i].address;
 648                 pcpu->state = (cpu >= info->configured) ?
 649                         CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 650                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 651                 set_cpu_present(cpu, true);
 652                 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 653                         set_cpu_present(cpu, false);
 654                 else
 655                         nr++;
 656                 cpu = cpumask_next(cpu, &avail);
 657         }
 658         return nr;
 659 }
 660
 661 static void __init smp_detect_cpus(void)
 662 {
 663         unsigned int cpu, c_cpus, s_cpus;
 664         struct sclp_cpu_info *info;
 665
 666         info = smp_get_cpu_info();
 667         if (!info)
 668                 panic("smp_detect_cpus failed to allocate memory\n");
 669         if (info->has_cpu_type) {
 670                 for (cpu = 0; cpu < info->combined; cpu++) {
 671                         if (info->cpu[cpu].address != boot_cpu_address)
 672                                 continue;
 673                         /* The boot cpu dictates the cpu type. */
 674                         boot_cpu_type = info->cpu[cpu].type;
 675                         break;
 676                 }
 677         }
 678         c_cpus = s_cpus = 0;
 679         for (cpu = 0; cpu < info->combined; cpu++) {
 680                 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
 681                         continue;
 682                 if (cpu < info->configured) {
 683                         smp_get_save_area(c_cpus, info->cpu[cpu].address);
 684                         c_cpus++;
 685                 } else
 686                         s_cpus++;
 687         }
 688         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 689         get_online_cpus();
 690         __smp_rescan_cpus(info, 0);
 691         put_online_cpus();
 692         kfree(info);
 693 }
 694
 695 /*
 696  *      Activate a secondary processor.
 697  */
 698 static void __cpuinit smp_start_secondary(void *cpuvoid)
 699 {
 700         S390_lowcore.last_update_clock = get_clock();
 701         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 702         S390_lowcore.restart_fn = (unsigned long) do_restart;
 703         S390_lowcore.restart_data = 0;
 704         S390_lowcore.restart_source = -1UL;
 705         restore_access_regs(S390_lowcore.access_regs_save_area);
 706         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 707         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 708         cpu_init();
 709         preempt_disable();
 710         init_cpu_timer();
 711         init_cpu_vtimer();
 712         pfault_init();
 713         notify_cpu_starting(smp_processor_id());
 714         ipi_call_lock();
 715         set_cpu_online(smp_processor_id(), true);
 716         ipi_call_unlock();
 717         local_irq_enable();
 718         /* cpu_idle will call schedule for us */
 719         cpu_idle();
 720 }
 721
 722 struct create_idle {
 723         struct work_struct work;
 724         struct task_struct *idle;
 725         struct completion done;
 726         int cpu;
 727 };
 728
 729 static void __cpuinit smp_fork_idle(struct work_struct *work)
 730 {
 731         struct create_idle *c_idle;
 732
 733         c_idle = container_of(work, struct create_idle, work);
 734         c_idle->idle = fork_idle(c_idle->cpu);
 735         complete(&c_idle->done);
 736 }
 737
 738 /* Upping and downing of CPUs */
 739 int __cpuinit __cpu_up(unsigned int cpu)
 740 {
 741         struct create_idle c_idle;
 742         struct pcpu *pcpu;
 743         int rc;
 744
 745         pcpu = pcpu_devices + cpu;
 746         if (pcpu->state != CPU_STATE_CONFIGURED)
 747                 return -EIO;
 748         if (pcpu_sigp_retry(pcpu, sigp_initial_cpu_reset, 0) !=
 749             sigp_order_code_accepted)
 750                 return -EIO;
 751         if (!pcpu->idle) {
 752                 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
 753                 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
 754                 c_idle.cpu = cpu;
 755                 schedule_work(&c_idle.work);
 756                 wait_for_completion(&c_idle.done);
 757                 if (IS_ERR(c_idle.idle))
 758                         return PTR_ERR(c_idle.idle);
 759                 pcpu->idle = c_idle.idle;
 760         }
 761         init_idle(pcpu->idle, cpu);
 762         rc = pcpu_alloc_lowcore(pcpu, cpu);
 763         if (rc)
 764                 return rc;
 765         pcpu_prepare_secondary(pcpu, cpu);
 766         pcpu_attach_task(pcpu, pcpu->idle);
 767         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 768         while (!cpu_online(cpu))
 769                 cpu_relax();
 770         return 0;
 771 }
 772
 773 static int __init setup_possible_cpus(char *s)
 774 {
 775         int max, cpu;
 776
 777         if (kstrtoint(s, 0, &max) < 0)
 778                 return 0;
 779         init_cpu_possible(cpumask_of(0));
 780         for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
 781                 set_cpu_possible(cpu, true);
 782         return 0;
 783 }
 784 early_param("possible_cpus", setup_possible_cpus);
 785
 786 #ifdef CONFIG_HOTPLUG_CPU
 787
 788 int __cpu_disable(void)
 789 {
 790         unsigned long cregs[16];
 791
 792         set_cpu_online(smp_processor_id(), false);
 793         /* Disable pseudo page faults on this cpu. */
 794         pfault_fini();
 795         /* Disable interrupt sources via control register. */
 796         __ctl_store(cregs, 0, 15);
 797         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 798         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 799         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 800         __ctl_load(cregs, 0, 15);
 801         return 0;
 802 }
 803
 804 void __cpu_die(unsigned int cpu)
 805 {
 806         struct pcpu *pcpu;
 807
 808         /* Wait until target cpu is down */
 809         pcpu = pcpu_devices + cpu;
 810         while (!pcpu_stopped(pcpu))
 811                 cpu_relax();
 812         pcpu_free_lowcore(pcpu);
 813         atomic_dec(&init_mm.context.attach_count);
 814 }
 815
 816 void __noreturn cpu_die(void)
 817 {
 818         idle_task_exit();
 819         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
 820         for (;;) ;
 821 }
 822
 823 #endif /* CONFIG_HOTPLUG_CPU */
 824
 825 static void smp_call_os_info_init_fn(void)
 826 {
 827         int (*init_fn)(void);
 828         unsigned long size;
 829
 830         init_fn = os_info_old_entry(OS_INFO_INIT_FN, &size);
 831         if (!init_fn)
 832                 return;
 833         init_fn();
 834 }
 835
 836 void __init smp_prepare_cpus(unsigned int max_cpus)
 837 {
 838         /* request the 0x1201 emergency signal external interrupt */
 839         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 840                 panic("Couldn't request external interrupt 0x1201");
 841         /* request the 0x1202 external call external interrupt */
 842         if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
 843                 panic("Couldn't request external interrupt 0x1202");
 844         smp_call_os_info_init_fn();
 845         smp_detect_cpus();
 846 }
 847
 848 void __init smp_prepare_boot_cpu(void)
 849 {
 850         struct pcpu *pcpu = pcpu_devices;
 851
 852         boot_cpu_address = stap();
 853         pcpu->idle = current;
 854         pcpu->state = CPU_STATE_CONFIGURED;
 855         pcpu->address = boot_cpu_address;
 856         pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
 857         pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
 858         pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
 859         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 860         cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 861         set_cpu_present(0, true);
 862         set_cpu_online(0, true);
 863 }
 864
 865 void __init smp_cpus_done(unsigned int max_cpus)
 866 {
 867 }
 868
 869 void __init smp_setup_processor_id(void)
 870 {
 871         S390_lowcore.cpu_nr = 0;
 872 }
 873
 874 /*
 875  * the frequency of the profiling timer can be changed
 876  * by writing a multiplier value into /proc/profile.
 877  *
 878  * usually you want to run this on all CPUs ;)
 879  */
 880 int setup_profiling_timer(unsigned int multiplier)
 881 {
 882         return 0;
 883 }
 884
 885 #ifdef CONFIG_HOTPLUG_CPU
 886 static ssize_t cpu_configure_show(struct device *dev,
 887                                   struct device_attribute *attr, char *buf)
 888 {
 889         ssize_t count;
 890
 891         mutex_lock(&smp_cpu_state_mutex);
 892         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 893         mutex_unlock(&smp_cpu_state_mutex);
 894         return count;
 895 }
 896
 897 static ssize_t cpu_configure_store(struct device *dev,
 898                                    struct device_attribute *attr,
 899                                    const char *buf, size_t count)
 900 {
 901         struct pcpu *pcpu;
 902         int cpu, val, rc;
 903         char delim;
 904
 905         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 906                 return -EINVAL;
 907         if (val != 0 && val != 1)
 908                 return -EINVAL;
 909         get_online_cpus();
 910         mutex_lock(&smp_cpu_state_mutex);
 911         rc = -EBUSY;
 912         /* disallow configuration changes of online cpus and cpu 0 */
 913         cpu = dev->id;
 914         if (cpu_online(cpu) || cpu == 0)
 915                 goto out;
 916         pcpu = pcpu_devices + cpu;
 917         rc = 0;
 918         switch (val) {
 919         case 0:
 920                 if (pcpu->state != CPU_STATE_CONFIGURED)
 921                         break;
 922                 rc = sclp_cpu_deconfigure(pcpu->address);
 923                 if (rc)
 924                         break;
 925                 pcpu->state = CPU_STATE_STANDBY;
 926                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 927                 topology_expect_change();
 928                 break;
 929         case 1:
 930                 if (pcpu->state != CPU_STATE_STANDBY)
 931                         break;
 932                 rc = sclp_cpu_configure(pcpu->address);
 933                 if (rc)
 934                         break;
 935                 pcpu->state = CPU_STATE_CONFIGURED;
 936                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 937                 topology_expect_change();
 938                 break;
 939         default:
 940                 break;
 941         }
 942 out:
 943         mutex_unlock(&smp_cpu_state_mutex);
 944         put_online_cpus();
 945         return rc ? rc : count;
 946 }
 947 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 948 #endif /* CONFIG_HOTPLUG_CPU */
 949
 950 static ssize_t show_cpu_address(struct device *dev,
 951                                 struct device_attribute *attr, char *buf)
 952 {
 953         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
 954 }
 955 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
 956
 957 static struct attribute *cpu_common_attrs[] = {
 958 #ifdef CONFIG_HOTPLUG_CPU
 959         &dev_attr_configure.attr,
 960 #endif
 961         &dev_attr_address.attr,
 962         NULL,
 963 };
 964
 965 static struct attribute_group cpu_common_attr_group = {
 966         .attrs = cpu_common_attrs,
 967 };
 968
 969 static ssize_t show_capability(struct device *dev,
 970                                 struct device_attribute *attr, char *buf)
 971 {
 972         unsigned int capability;
 973         int rc;
 974
 975         rc = get_cpu_capability(&capability);
 976         if (rc)
 977                 return rc;
 978         return sprintf(buf, "%u\n", capability);
 979 }
 980 static DEVICE_ATTR(capability, 0444, show_capability, NULL);
 981
 982 static ssize_t show_idle_count(struct device *dev,
 983                                 struct device_attribute *attr, char *buf)
 984 {
 985         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 986         unsigned long long idle_count;
 987         unsigned int sequence;
 988
 989         do {
 990                 sequence = ACCESS_ONCE(idle->sequence);
 991                 idle_count = ACCESS_ONCE(idle->idle_count);
 992                 if (ACCESS_ONCE(idle->idle_enter))
 993                         idle_count++;
 994         } while ((sequence & 1) || (idle->sequence != sequence));
 995         return sprintf(buf, "%llu\n", idle_count);
 996 }
 997 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
 998
 999 static ssize_t show_idle_time(struct device *dev,
1000                                 struct device_attribute *attr, char *buf)
1001 {
1002         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
1003         unsigned long long now, idle_time, idle_enter, idle_exit;
1004         unsigned int sequence;
1005
1006         do {
1007                 now = get_clock();
1008                 sequence = ACCESS_ONCE(idle->sequence);
1009                 idle_time = ACCESS_ONCE(idle->idle_time);
1010                 idle_enter = ACCESS_ONCE(idle->idle_enter);
1011                 idle_exit = ACCESS_ONCE(idle->idle_exit);
1012         } while ((sequence & 1) || (idle->sequence != sequence));
1013         idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
1014         return sprintf(buf, "%llu\n", idle_time >> 12);
1015 }
1016 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1017
1018 static struct attribute *cpu_online_attrs[] = {
1019         &dev_attr_capability.attr,
1020         &dev_attr_idle_count.attr,
1021         &dev_attr_idle_time_us.attr,
1022         NULL,
1023 };
1024
1025 static struct attribute_group cpu_online_attr_group = {
1026         .attrs = cpu_online_attrs,
1027 };
1028
1029 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1030                                     unsigned long action, void *hcpu)
1031 {
1032         unsigned int cpu = (unsigned int)(long)hcpu;
1033         struct cpu *c = &pcpu_devices[cpu].cpu;
1034         struct device *s = &c->dev;
1035         struct s390_idle_data *idle;
1036         int err = 0;
1037
1038         switch (action) {
1039         case CPU_ONLINE:
1040         case CPU_ONLINE_FROZEN:
1041                 idle = &per_cpu(s390_idle, cpu);
1042                 memset(idle, 0, sizeof(struct s390_idle_data));
1043                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1044                 break;
1045         case CPU_DEAD:
1046         case CPU_DEAD_FROZEN:
1047                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1048                 break;
1049         }
1050         return notifier_from_errno(err);
1051 }
1052
1053 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1054         .notifier_call = smp_cpu_notify,
1055 };
1056
1057 static int __devinit smp_add_present_cpu(int cpu)
1058 {
1059         struct cpu *c = &pcpu_devices[cpu].cpu;
1060         struct device *s = &c->dev;
1061         int rc;
1062
1063         c->hotpluggable = 1;
1064         rc = register_cpu(c, cpu);
1065         if (rc)
1066                 goto out;
1067         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1068         if (rc)
1069                 goto out_cpu;
1070         if (cpu_online(cpu)) {
1071                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1072                 if (rc)
1073                         goto out_online;
1074         }
1075         rc = topology_cpu_init(c);
1076         if (rc)
1077                 goto out_topology;
1078         return 0;
1079
1080 out_topology:
1081         if (cpu_online(cpu))
1082                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1083 out_online:
1084         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1085 out_cpu:
1086 #ifdef CONFIG_HOTPLUG_CPU
1087         unregister_cpu(c);
1088 #endif
1089 out:
1090         return rc;
1091 }
1092
1093 #ifdef CONFIG_HOTPLUG_CPU
1094
1095 int __ref smp_rescan_cpus(void)
1096 {
1097         struct sclp_cpu_info *info;
1098         int nr;
1099
1100         info = smp_get_cpu_info();
1101         if (!info)
1102                 return -ENOMEM;
1103         get_online_cpus();
1104         mutex_lock(&smp_cpu_state_mutex);
1105         nr = __smp_rescan_cpus(info, 1);
1106         mutex_unlock(&smp_cpu_state_mutex);
1107         put_online_cpus();
1108         kfree(info);
1109         if (nr)
1110                 topology_schedule_update();
1111         return 0;
1112 }
1113
1114 static ssize_t __ref rescan_store(struct device *dev,
1115                                   struct device_attribute *attr,
1116                                   const char *buf,
1117                                   size_t count)
1118 {
1119         int rc;
1120
1121         rc = smp_rescan_cpus();
1122         return rc ? rc : count;
1123 }
1124 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1125 #endif /* CONFIG_HOTPLUG_CPU */
1126
1127 static int __init s390_smp_init(void)
1128 {
1129         int cpu, rc;
1130
1131         register_cpu_notifier(&smp_cpu_nb);
1132 #ifdef CONFIG_HOTPLUG_CPU
1133         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1134         if (rc)
1135                 return rc;
1136 #endif
1137         for_each_present_cpu(cpu) {
1138                 rc = smp_add_present_cpu(cpu);
1139                 if (rc)
1140                         return rc;
1141         }
1142         return 0;
1143 }
1144 subsys_initcall(s390_smp_init);