arch/s390/kernel/smp.c

   1 /*
   2  *  SMP related functions
   3  *
   4  *    Copyright IBM Corp. 1999, 2012
   5  *    Author(s): Denis Joseph Barrow,
   6  *               Martin Schwidefsky <schwidefsky@de.ibm.com>,
   7  *               Heiko Carstens <heiko.carstens@de.ibm.com>,
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14  * the translation of logical to physical cpu ids. All new code that
  15  * operates on physical cpu numbers needs to go into smp.c.
  16  */
  17
  18 #define KMSG_COMPONENT "cpu"
  19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21 #include <linux/workqueue.h>
  22 #include <linux/module.h>
  23 #include <linux/init.h>
  24 #include <linux/mm.h>
  25 #include <linux/err.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/kernel_stat.h>
  28 #include <linux/delay.h>
  29 #include <linux/interrupt.h>
  30 #include <linux/irqflags.h>
  31 #include <linux/cpu.h>
  32 #include <linux/slab.h>
  33 #include <linux/crash_dump.h>
  34 #include <asm/asm-offsets.h>
  35 #include <asm/switch_to.h>
  36 #include <asm/facility.h>
  37 #include <asm/ipl.h>
  38 #include <asm/setup.h>
  39 #include <asm/irq.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/vtimer.h>
  42 #include <asm/lowcore.h>
  43 #include <asm/sclp.h>
  44 #include <asm/vdso.h>
  45 #include <asm/debug.h>
  46 #include <asm/os_info.h>
  47 #include <asm/sigp.h>
  48 #include "entry.h"
  49
  50 enum {
  51         ec_schedule = 0,
  52         ec_call_function_single,
  53         ec_stop_cpu,
  54 };
  55
  56 enum {
  57         CPU_STATE_STANDBY,
  58         CPU_STATE_CONFIGURED,
  59 };
  60
  61 struct pcpu {
  62         struct cpu *cpu;
  63         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
  64         unsigned long async_stack;      /* async stack for the cpu */
  65         unsigned long panic_stack;      /* panic stack for the cpu */
  66         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  67         int state;                      /* physical cpu state */
  68         int polarization;               /* physical polarization */
  69         u16 address;                    /* physical cpu address */
  70 };
  71
  72 static u8 boot_cpu_type;
  73 static u16 boot_cpu_address;
  74 static struct pcpu pcpu_devices[NR_CPUS];
  75
  76 /*
  77  * The smp_cpu_state_mutex must be held when changing the state or polarization
  78  * member of a pcpu data structure within the pcpu_devices arreay.
  79  */
  80 DEFINE_MUTEX(smp_cpu_state_mutex);
  81
  82 /*
  83  * Signal processor helper functions.
  84  */
  85 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
  86 {
  87         register unsigned int reg1 asm ("1") = parm;
  88         int cc;
  89
  90         asm volatile(
  91                 "       sigp    %1,%2,0(%3)\n"
  92                 "       ipm     %0\n"
  93                 "       srl     %0,28\n"
  94                 : "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
  95         if (status && cc == 1)
  96                 *status = reg1;
  97         return cc;
  98 }
  99
 100 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
 101 {
 102         int cc;
 103
 104         while (1) {
 105                 cc = __pcpu_sigp(addr, order, parm, NULL);
 106                 if (cc != SIGP_CC_BUSY)
 107                         return cc;
 108                 cpu_relax();
 109         }
 110 }
 111
 112 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 113 {
 114         int cc, retry;
 115
 116         for (retry = 0; ; retry++) {
 117                 cc = __pcpu_sigp(pcpu->address, order, parm, NULL);
 118                 if (cc != SIGP_CC_BUSY)
 119                         break;
 120                 if (retry >= 3)
 121                         udelay(10);
 122         }
 123         return cc;
 124 }
 125
 126 static inline int pcpu_stopped(struct pcpu *pcpu)
 127 {
 128         u32 uninitialized_var(status);
 129
 130         if (__pcpu_sigp(pcpu->address, SIGP_SENSE,
 131                         0, &status) != SIGP_CC_STATUS_STORED)
 132                 return 0;
 133         return !!(status & (SIGP_STATUS_CHECK_STOP|SIGP_STATUS_STOPPED));
 134 }
 135
 136 static inline int pcpu_running(struct pcpu *pcpu)
 137 {
 138         if (__pcpu_sigp(pcpu->address, SIGP_SENSE_RUNNING,
 139                         0, NULL) != SIGP_CC_STATUS_STORED)
 140                 return 1;
 141         /* Status stored condition code is equivalent to cpu not running. */
 142         return 0;
 143 }
 144
 145 /*
 146  * Find struct pcpu by cpu address.
 147  */
 148 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
 149 {
 150         int cpu;
 151
 152         for_each_cpu(cpu, mask)
 153                 if (pcpu_devices[cpu].address == address)
 154                         return pcpu_devices + cpu;
 155         return NULL;
 156 }
 157
 158 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 159 {
 160         int order;
 161
 162         if (test_and_set_bit(ec_bit, &pcpu->ec_mask))
 163                 return;
 164         order = pcpu_running(pcpu) ? SIGP_EXTERNAL_CALL : SIGP_EMERGENCY_SIGNAL;
 165         pcpu_sigp_retry(pcpu, order, 0);
 166 }
 167
 168 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 169 {
 170         struct _lowcore *lc;
 171
 172         if (pcpu != &pcpu_devices[0]) {
 173                 pcpu->lowcore = (struct _lowcore *)
 174                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 175                 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 176                 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
 177                 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
 178                         goto out;
 179         }
 180         lc = pcpu->lowcore;
 181         memcpy(lc, &S390_lowcore, 512);
 182         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 183         lc->async_stack = pcpu->async_stack + ASYNC_SIZE
 184                 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 185         lc->panic_stack = pcpu->panic_stack + PAGE_SIZE
 186                 - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 187         lc->cpu_nr = cpu;
 188 #ifndef CONFIG_64BIT
 189         if (MACHINE_HAS_IEEE) {
 190                 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
 191                 if (!lc->extended_save_area_addr)
 192                         goto out;
 193         }
 194 #else
 195         if (vdso_alloc_per_cpu(lc))
 196                 goto out;
 197 #endif
 198         lowcore_ptr[cpu] = lc;
 199         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 200         return 0;
 201 out:
 202         if (pcpu != &pcpu_devices[0]) {
 203                 free_page(pcpu->panic_stack);
 204                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 205                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 206         }
 207         return -ENOMEM;
 208 }
 209
 210 #ifdef CONFIG_HOTPLUG_CPU
 211
 212 static void pcpu_free_lowcore(struct pcpu *pcpu)
 213 {
 214         pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 215         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 216 #ifndef CONFIG_64BIT
 217         if (MACHINE_HAS_IEEE) {
 218                 struct _lowcore *lc = pcpu->lowcore;
 219
 220                 free_page((unsigned long) lc->extended_save_area_addr);
 221                 lc->extended_save_area_addr = 0;
 222         }
 223 #else
 224         vdso_free_per_cpu(pcpu->lowcore);
 225 #endif
 226         if (pcpu != &pcpu_devices[0]) {
 227                 free_page(pcpu->panic_stack);
 228                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 229                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 230         }
 231 }
 232
 233 #endif /* CONFIG_HOTPLUG_CPU */
 234
 235 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 236 {
 237         struct _lowcore *lc = pcpu->lowcore;
 238
 239         if (MACHINE_HAS_TLB_LC)
 240                 cpumask_set_cpu(cpu, &init_mm.context.cpu_attach_mask);
 241         cpumask_set_cpu(cpu, mm_cpumask(&init_mm));
 242         atomic_inc(&init_mm.context.attach_count);
 243         lc->cpu_nr = cpu;
 244         lc->percpu_offset = __per_cpu_offset[cpu];
 245         lc->kernel_asce = S390_lowcore.kernel_asce;
 246         lc->machine_flags = S390_lowcore.machine_flags;
 247         lc->ftrace_func = S390_lowcore.ftrace_func;
 248         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 249         __ctl_store(lc->cregs_save_area, 0, 15);
 250         save_access_regs((unsigned int *) lc->access_regs_save_area);
 251         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 252                MAX_FACILITY_BIT/8);
 253 }
 254
 255 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 256 {
 257         struct _lowcore *lc = pcpu->lowcore;
 258         struct thread_info *ti = task_thread_info(tsk);
 259
 260         lc->kernel_stack = (unsigned long) task_stack_page(tsk)
 261                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 262         lc->thread_info = (unsigned long) task_thread_info(tsk);
 263         lc->current_task = (unsigned long) tsk;
 264         lc->user_timer = ti->user_timer;
 265         lc->system_timer = ti->system_timer;
 266         lc->steal_timer = 0;
 267 }
 268
 269 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 270 {
 271         struct _lowcore *lc = pcpu->lowcore;
 272
 273         lc->restart_stack = lc->kernel_stack;
 274         lc->restart_fn = (unsigned long) func;
 275         lc->restart_data = (unsigned long) data;
 276         lc->restart_source = -1UL;
 277         pcpu_sigp_retry(pcpu, SIGP_RESTART, 0);
 278 }
 279
 280 /*
 281  * Call function via PSW restart on pcpu and stop the current cpu.
 282  */
 283 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 284                           void *data, unsigned long stack)
 285 {
 286         struct _lowcore *lc = lowcore_ptr[pcpu - pcpu_devices];
 287         unsigned long source_cpu = stap();
 288
 289         __load_psw_mask(PSW_KERNEL_BITS);
 290         if (pcpu->address == source_cpu)
 291                 func(data);     /* should not return */
 292         /* Stop target cpu (if func returns this stops the current cpu). */
 293         pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 294         /* Restart func on the target cpu and stop the current cpu. */
 295         mem_assign_absolute(lc->restart_stack, stack);
 296         mem_assign_absolute(lc->restart_fn, (unsigned long) func);
 297         mem_assign_absolute(lc->restart_data, (unsigned long) data);
 298         mem_assign_absolute(lc->restart_source, source_cpu);
 299         asm volatile(
 300                 "0:     sigp    0,%0,%2 # sigp restart to target cpu\n"
 301                 "       brc     2,0b    # busy, try again\n"
 302                 "1:     sigp    0,%1,%3 # sigp stop to current cpu\n"
 303                 "       brc     2,1b    # busy, try again\n"
 304                 : : "d" (pcpu->address), "d" (source_cpu),
 305                     "K" (SIGP_RESTART), "K" (SIGP_STOP)
 306                 : "0", "1", "cc");
 307         for (;;) ;
 308 }
 309
 310 /*
 311  * Call function on an online CPU.
 312  */
 313 void smp_call_online_cpu(void (*func)(void *), void *data)
 314 {
 315         struct pcpu *pcpu;
 316
 317         /* Use the current cpu if it is online. */
 318         pcpu = pcpu_find_address(cpu_online_mask, stap());
 319         if (!pcpu)
 320                 /* Use the first online cpu. */
 321                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 322         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 323 }
 324
 325 /*
 326  * Call function on the ipl CPU.
 327  */
 328 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 329 {
 330         pcpu_delegate(&pcpu_devices[0], func, data,
 331                       pcpu_devices->panic_stack + PAGE_SIZE);
 332 }
 333
 334 int smp_find_processor_id(u16 address)
 335 {
 336         int cpu;
 337
 338         for_each_present_cpu(cpu)
 339                 if (pcpu_devices[cpu].address == address)
 340                         return cpu;
 341         return -1;
 342 }
 343
 344 int smp_vcpu_scheduled(int cpu)
 345 {
 346         return pcpu_running(pcpu_devices + cpu);
 347 }
 348
 349 void smp_yield(void)
 350 {
 351         if (MACHINE_HAS_DIAG44)
 352                 asm volatile("diag 0,0,0x44");
 353 }
 354
 355 void smp_yield_cpu(int cpu)
 356 {
 357         if (MACHINE_HAS_DIAG9C)
 358                 asm volatile("diag %0,0,0x9c"
 359                              : : "d" (pcpu_devices[cpu].address));
 360         else if (MACHINE_HAS_DIAG44)
 361                 asm volatile("diag 0,0,0x44");
 362 }
 363
 364 /*
 365  * Send cpus emergency shutdown signal. This gives the cpus the
 366  * opportunity to complete outstanding interrupts.
 367  */
 368 static void smp_emergency_stop(cpumask_t *cpumask)
 369 {
 370         u64 end;
 371         int cpu;
 372
 373         end = get_tod_clock() + (1000000UL << 12);
 374         for_each_cpu(cpu, cpumask) {
 375                 struct pcpu *pcpu = pcpu_devices + cpu;
 376                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 377                 while (__pcpu_sigp(pcpu->address, SIGP_EMERGENCY_SIGNAL,
 378                                    0, NULL) == SIGP_CC_BUSY &&
 379                        get_tod_clock() < end)
 380                         cpu_relax();
 381         }
 382         while (get_tod_clock() < end) {
 383                 for_each_cpu(cpu, cpumask)
 384                         if (pcpu_stopped(pcpu_devices + cpu))
 385                                 cpumask_clear_cpu(cpu, cpumask);
 386                 if (cpumask_empty(cpumask))
 387                         break;
 388                 cpu_relax();
 389         }
 390 }
 391
 392 /*
 393  * Stop all cpus but the current one.
 394  */
 395 void smp_send_stop(void)
 396 {
 397         cpumask_t cpumask;
 398         int cpu;
 399
 400         /* Disable all interrupts/machine checks */
 401         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 402         trace_hardirqs_off();
 403
 404         debug_set_critical();
 405         cpumask_copy(&cpumask, cpu_online_mask);
 406         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 407
 408         if (oops_in_progress)
 409                 smp_emergency_stop(&cpumask);
 410
 411         /* stop all processors */
 412         for_each_cpu(cpu, &cpumask) {
 413                 struct pcpu *pcpu = pcpu_devices + cpu;
 414                 pcpu_sigp_retry(pcpu, SIGP_STOP, 0);
 415                 while (!pcpu_stopped(pcpu))
 416                         cpu_relax();
 417         }
 418 }
 419
 420 /*
 421  * Stop the current cpu.
 422  */
 423 void smp_stop_cpu(void)
 424 {
 425         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 426         for (;;) ;
 427 }
 428
 429 /*
 430  * This is the main routine where commands issued by other
 431  * cpus are handled.
 432  */
 433 static void smp_handle_ext_call(void)
 434 {
 435         unsigned long bits;
 436
 437         /* handle bit signal external calls */
 438         bits = xchg(&pcpu_devices[smp_processor_id()].ec_mask, 0);
 439         if (test_bit(ec_stop_cpu, &bits))
 440                 smp_stop_cpu();
 441         if (test_bit(ec_schedule, &bits))
 442                 scheduler_ipi();
 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_single);
 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 static void __init smp_get_save_area(int cpu, u16 address)
 540 {
 541         void *lc = pcpu_devices[0].lowcore;
 542         struct save_area *save_area;
 543
 544         if (is_kdump_kernel())
 545                 return;
 546         if (!OLDMEM_BASE && (address == boot_cpu_address ||
 547                              ipl_info.type != IPL_TYPE_FCP_DUMP))
 548                 return;
 549         save_area = dump_save_area_create(cpu);
 550         if (!save_area)
 551                 panic("could not allocate memory for save area\n");
 552 #ifdef CONFIG_CRASH_DUMP
 553         if (address == boot_cpu_address) {
 554                 /* Copy the registers of the boot cpu. */
 555                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 556                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
 557                 return;
 558         }
 559 #endif
 560         /* Get the registers of a non-boot cpu. */
 561         __pcpu_sigp_relax(address, SIGP_STOP_AND_STORE_STATUS, 0, NULL);
 562         memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
 563 }
 564
 565 int smp_store_status(int cpu)
 566 {
 567         struct pcpu *pcpu;
 568
 569         pcpu = pcpu_devices + cpu;
 570         if (__pcpu_sigp_relax(pcpu->address, SIGP_STOP_AND_STORE_STATUS,
 571                               0, NULL) != SIGP_CC_ORDER_CODE_ACCEPTED)
 572                 return -EIO;
 573         return 0;
 574 }
 575
 576 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 577
 578 static inline void smp_get_save_area(int cpu, u16 address) { }
 579
 580 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 581
 582 void smp_cpu_set_polarization(int cpu, int val)
 583 {
 584         pcpu_devices[cpu].polarization = val;
 585 }
 586
 587 int smp_cpu_get_polarization(int cpu)
 588 {
 589         return pcpu_devices[cpu].polarization;
 590 }
 591
 592 static struct sclp_cpu_info *smp_get_cpu_info(void)
 593 {
 594         static int use_sigp_detection;
 595         struct sclp_cpu_info *info;
 596         int address;
 597
 598         info = kzalloc(sizeof(*info), GFP_KERNEL);
 599         if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
 600                 use_sigp_detection = 1;
 601                 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
 602                         if (__pcpu_sigp_relax(address, SIGP_SENSE, 0, NULL) ==
 603                             SIGP_CC_NOT_OPERATIONAL)
 604                                 continue;
 605                         info->cpu[info->configured].address = address;
 606                         info->configured++;
 607                 }
 608                 info->combined = info->configured;
 609         }
 610         return info;
 611 }
 612
 613 static int smp_add_present_cpu(int cpu);
 614
 615 static int __smp_rescan_cpus(struct sclp_cpu_info *info, int sysfs_add)
 616 {
 617         struct pcpu *pcpu;
 618         cpumask_t avail;
 619         int cpu, nr, i;
 620
 621         nr = 0;
 622         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 623         cpu = cpumask_first(&avail);
 624         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 625                 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
 626                         continue;
 627                 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
 628                         continue;
 629                 pcpu = pcpu_devices + cpu;
 630                 pcpu->address = info->cpu[i].address;
 631                 pcpu->state = (i >= info->configured) ?
 632                         CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 633                 smp_cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 634                 set_cpu_present(cpu, true);
 635                 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 636                         set_cpu_present(cpu, false);
 637                 else
 638                         nr++;
 639                 cpu = cpumask_next(cpu, &avail);
 640         }
 641         return nr;
 642 }
 643
 644 static void __init smp_detect_cpus(void)
 645 {
 646         unsigned int cpu, c_cpus, s_cpus;
 647         struct sclp_cpu_info *info;
 648
 649         info = smp_get_cpu_info();
 650         if (!info)
 651                 panic("smp_detect_cpus failed to allocate memory\n");
 652         if (info->has_cpu_type) {
 653                 for (cpu = 0; cpu < info->combined; cpu++) {
 654                         if (info->cpu[cpu].address != boot_cpu_address)
 655                                 continue;
 656                         /* The boot cpu dictates the cpu type. */
 657                         boot_cpu_type = info->cpu[cpu].type;
 658                         break;
 659                 }
 660         }
 661         c_cpus = s_cpus = 0;
 662         for (cpu = 0; cpu < info->combined; cpu++) {
 663                 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
 664                         continue;
 665                 if (cpu < info->configured) {
 666                         smp_get_save_area(c_cpus, info->cpu[cpu].address);
 667                         c_cpus++;
 668                 } else
 669                         s_cpus++;
 670         }
 671         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 672         get_online_cpus();
 673         __smp_rescan_cpus(info, 0);
 674         put_online_cpus();
 675         kfree(info);
 676 }
 677
 678 /*
 679  *      Activate a secondary processor.
 680  */
 681 static void smp_start_secondary(void *cpuvoid)
 682 {
 683         S390_lowcore.last_update_clock = get_tod_clock();
 684         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 685         S390_lowcore.restart_fn = (unsigned long) do_restart;
 686         S390_lowcore.restart_data = 0;
 687         S390_lowcore.restart_source = -1UL;
 688         restore_access_regs(S390_lowcore.access_regs_save_area);
 689         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 690         __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_DAT);
 691         cpu_init();
 692         preempt_disable();
 693         init_cpu_timer();
 694         init_cpu_vtimer();
 695         pfault_init();
 696         notify_cpu_starting(smp_processor_id());
 697         set_cpu_online(smp_processor_id(), true);
 698         inc_irq_stat(CPU_RST);
 699         local_irq_enable();
 700         cpu_startup_entry(CPUHP_ONLINE);
 701 }
 702
 703 /* Upping and downing of CPUs */
 704 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 705 {
 706         struct pcpu *pcpu;
 707         int rc;
 708
 709         pcpu = pcpu_devices + cpu;
 710         if (pcpu->state != CPU_STATE_CONFIGURED)
 711                 return -EIO;
 712         if (pcpu_sigp_retry(pcpu, SIGP_INITIAL_CPU_RESET, 0) !=
 713             SIGP_CC_ORDER_CODE_ACCEPTED)
 714                 return -EIO;
 715
 716         rc = pcpu_alloc_lowcore(pcpu, cpu);
 717         if (rc)
 718                 return rc;
 719         pcpu_prepare_secondary(pcpu, cpu);
 720         pcpu_attach_task(pcpu, tidle);
 721         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 722         while (!cpu_online(cpu))
 723                 cpu_relax();
 724         return 0;
 725 }
 726
 727 static unsigned int setup_possible_cpus __initdata;
 728
 729 static int __init _setup_possible_cpus(char *s)
 730 {
 731         get_option(&s, &setup_possible_cpus);
 732         return 0;
 733 }
 734 early_param("possible_cpus", _setup_possible_cpus);
 735
 736 #ifdef CONFIG_HOTPLUG_CPU
 737
 738 int __cpu_disable(void)
 739 {
 740         unsigned long cregs[16];
 741
 742         /* Handle possible pending IPIs */
 743         smp_handle_ext_call();
 744         set_cpu_online(smp_processor_id(), false);
 745         /* Disable pseudo page faults on this cpu. */
 746         pfault_fini();
 747         /* Disable interrupt sources via control register. */
 748         __ctl_store(cregs, 0, 15);
 749         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 750         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 751         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 752         __ctl_load(cregs, 0, 15);
 753         return 0;
 754 }
 755
 756 void __cpu_die(unsigned int cpu)
 757 {
 758         struct pcpu *pcpu;
 759
 760         /* Wait until target cpu is down */
 761         pcpu = pcpu_devices + cpu;
 762         while (!pcpu_stopped(pcpu))
 763                 cpu_relax();
 764         pcpu_free_lowcore(pcpu);
 765         atomic_dec(&init_mm.context.attach_count);
 766         cpumask_clear_cpu(cpu, mm_cpumask(&init_mm));
 767         if (MACHINE_HAS_TLB_LC)
 768                 cpumask_clear_cpu(cpu, &init_mm.context.cpu_attach_mask);
 769 }
 770
 771 void __noreturn cpu_die(void)
 772 {
 773         idle_task_exit();
 774         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), SIGP_STOP, 0);
 775         for (;;) ;
 776 }
 777
 778 #endif /* CONFIG_HOTPLUG_CPU */
 779
 780 void __init smp_fill_possible_mask(void)
 781 {
 782         unsigned int possible, sclp, cpu;
 783
 784         sclp = sclp_get_max_cpu() ?: nr_cpu_ids;
 785         possible = setup_possible_cpus ?: nr_cpu_ids;
 786         possible = min(possible, sclp);
 787         for (cpu = 0; cpu < possible && cpu < nr_cpu_ids; cpu++)
 788                 set_cpu_possible(cpu, true);
 789 }
 790
 791 void __init smp_prepare_cpus(unsigned int max_cpus)
 792 {
 793         /* request the 0x1201 emergency signal external interrupt */
 794         if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 795                 panic("Couldn't request external interrupt 0x1201");
 796         /* request the 0x1202 external call external interrupt */
 797         if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 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) || (ACCESS_ONCE(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) || (ACCESS_ONCE(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 smp_cpu_notify(struct notifier_block *self, unsigned long action,
 971                           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 smp_add_present_cpu(int cpu)
 990 {
 991         struct device *s;
 992         struct cpu *c;
 993         int rc;
 994
 995         c = kzalloc(sizeof(*c), GFP_KERNEL);
 996         if (!c)
 997                 return -ENOMEM;
 998         pcpu_devices[cpu].cpu = c;
 999         s = &c->dev;
1000         c->hotpluggable = 1;
1001         rc = register_cpu(c, cpu);
1002         if (rc)
1003                 goto out;
1004         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1005         if (rc)
1006                 goto out_cpu;
1007         if (cpu_online(cpu)) {
1008                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1009                 if (rc)
1010                         goto out_online;
1011         }
1012         rc = topology_cpu_init(c);
1013         if (rc)
1014                 goto out_topology;
1015         return 0;
1016
1017 out_topology:
1018         if (cpu_online(cpu))
1019                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1020 out_online:
1021         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1022 out_cpu:
1023 #ifdef CONFIG_HOTPLUG_CPU
1024         unregister_cpu(c);
1025 #endif
1026 out:
1027         return rc;
1028 }
1029
1030 #ifdef CONFIG_HOTPLUG_CPU
1031
1032 int __ref smp_rescan_cpus(void)
1033 {
1034         struct sclp_cpu_info *info;
1035         int nr;
1036
1037         info = smp_get_cpu_info();
1038         if (!info)
1039                 return -ENOMEM;
1040         get_online_cpus();
1041         mutex_lock(&smp_cpu_state_mutex);
1042         nr = __smp_rescan_cpus(info, 1);
1043         mutex_unlock(&smp_cpu_state_mutex);
1044         put_online_cpus();
1045         kfree(info);
1046         if (nr)
1047                 topology_schedule_update();
1048         return 0;
1049 }
1050
1051 static ssize_t __ref rescan_store(struct device *dev,
1052                                   struct device_attribute *attr,
1053                                   const char *buf,
1054                                   size_t count)
1055 {
1056         int rc;
1057
1058         rc = smp_rescan_cpus();
1059         return rc ? rc : count;
1060 }
1061 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1062 #endif /* CONFIG_HOTPLUG_CPU */
1063
1064 static int __init s390_smp_init(void)
1065 {
1066         int cpu, rc = 0;
1067
1068 #ifdef CONFIG_HOTPLUG_CPU
1069         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1070         if (rc)
1071                 return rc;
1072 #endif
1073         cpu_notifier_register_begin();
1074         for_each_present_cpu(cpu) {
1075                 rc = smp_add_present_cpu(cpu);
1076                 if (rc)
1077                         goto out;
1078         }
1079
1080         __hotcpu_notifier(smp_cpu_notify, 0);
1081
1082 out:
1083         cpu_notifier_register_done();
1084         return rc;
1085 }
1086 subsys_initcall(s390_smp_init);