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