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