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