Pileus Git - ~andy/linux/blob - arch/s390/kernel/smp.c

   1 /*
   2  *  SMP related functions
   3  *
   4  *    Copyright IBM Corp. 1999,2012
   5  *    Author(s): Denis Joseph Barrow,
   6  *               Martin Schwidefsky <schwidefsky@de.ibm.com>,
   7  *               Heiko Carstens <heiko.carstens@de.ibm.com>,
   8  *
   9  *  based on other smp stuff by
  10  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  11  *    (c) 1998 Ingo Molnar
  12  *
  13  * The code outside of smp.c uses logical cpu numbers, only smp.c does
  14  * the translation of logical to physical cpu ids. All new code that
  15  * operates on physical cpu numbers needs to go into smp.c.
  16  */
  17
  18 #define KMSG_COMPONENT "cpu"
  19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  20
  21 #include <linux/workqueue.h>
  22 #include <linux/module.h>
  23 #include <linux/init.h>
  24 #include <linux/mm.h>
  25 #include <linux/err.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/kernel_stat.h>
  28 #include <linux/delay.h>
  29 #include <linux/interrupt.h>
  30 #include <linux/irqflags.h>
  31 #include <linux/cpu.h>
  32 #include <linux/slab.h>
  33 #include <linux/crash_dump.h>
  34 #include <asm/asm-offsets.h>
  35 #include <asm/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/timer.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 "entry.h"
  48
  49 enum {
  50         sigp_sense = 1,
  51         sigp_external_call = 2,
  52         sigp_emergency_signal = 3,
  53         sigp_start = 4,
  54         sigp_stop = 5,
  55         sigp_restart = 6,
  56         sigp_stop_and_store_status = 9,
  57         sigp_initial_cpu_reset = 11,
  58         sigp_cpu_reset = 12,
  59         sigp_set_prefix = 13,
  60         sigp_store_status_at_address = 14,
  61         sigp_store_extended_status_at_address = 15,
  62         sigp_set_architecture = 18,
  63         sigp_conditional_emergency_signal = 19,
  64         sigp_sense_running = 21,
  65 };
  66
  67 enum {
  68         sigp_order_code_accepted = 0,
  69         sigp_status_stored = 1,
  70         sigp_busy = 2,
  71         sigp_not_operational = 3,
  72 };
  73
  74 enum {
  75         ec_schedule = 0,
  76         ec_call_function,
  77         ec_call_function_single,
  78         ec_stop_cpu,
  79 };
  80
  81 enum {
  82         CPU_STATE_STANDBY,
  83         CPU_STATE_CONFIGURED,
  84 };
  85
  86 struct pcpu {
  87         struct cpu cpu;
  88         struct task_struct *idle;       /* idle process for the cpu */
  89         struct _lowcore *lowcore;       /* lowcore page(s) for the cpu */
  90         unsigned long async_stack;      /* async stack for the cpu */
  91         unsigned long panic_stack;      /* panic stack for the cpu */
  92         unsigned long ec_mask;          /* bit mask for ec_xxx functions */
  93         int state;                      /* physical cpu state */
  94         u32 status;                     /* last status received via sigp */
  95         u16 address;                    /* physical cpu address */
  96 };
  97
  98 static u8 boot_cpu_type;
  99 static u16 boot_cpu_address;
 100 static struct pcpu pcpu_devices[NR_CPUS];
 101
 102 DEFINE_MUTEX(smp_cpu_state_mutex);
 103
 104 /*
 105  * Signal processor helper functions.
 106  */
 107 static inline int __pcpu_sigp(u16 addr, u8 order, u32 parm, u32 *status)
 108 {
 109         register unsigned int reg1 asm ("1") = parm;
 110         int cc;
 111
 112         asm volatile(
 113                 "       sigp    %1,%2,0(%3)\n"
 114                 "       ipm     %0\n"
 115                 "       srl     %0,28\n"
 116                 : "=d" (cc), "+d" (reg1) : "d" (addr), "a" (order) : "cc");
 117         if (status && cc == 1)
 118                 *status = reg1;
 119         return cc;
 120 }
 121
 122 static inline int __pcpu_sigp_relax(u16 addr, u8 order, u32 parm, u32 *status)
 123 {
 124         int cc;
 125
 126         while (1) {
 127                 cc = __pcpu_sigp(addr, order, parm, status);
 128                 if (cc != sigp_busy)
 129                         return cc;
 130                 cpu_relax();
 131         }
 132 }
 133
 134 static int pcpu_sigp_retry(struct pcpu *pcpu, u8 order, u32 parm)
 135 {
 136         int cc, retry;
 137
 138         for (retry = 0; ; retry++) {
 139                 cc = __pcpu_sigp(pcpu->address, order, parm, &pcpu->status);
 140                 if (cc != sigp_busy)
 141                         break;
 142                 if (retry >= 3)
 143                         udelay(10);
 144         }
 145         return cc;
 146 }
 147
 148 static inline int pcpu_stopped(struct pcpu *pcpu)
 149 {
 150         if (__pcpu_sigp(pcpu->address, sigp_sense,
 151                         0, &pcpu->status) != sigp_status_stored)
 152                 return 0;
 153         /* Check for stopped and check stop state */
 154         return !!(pcpu->status & 0x50);
 155 }
 156
 157 static inline int pcpu_running(struct pcpu *pcpu)
 158 {
 159         if (__pcpu_sigp(pcpu->address, sigp_sense_running,
 160                         0, &pcpu->status) != sigp_status_stored)
 161                 return 1;
 162         /* Check for running status */
 163         return !(pcpu->status & 0x400);
 164 }
 165
 166 /*
 167  * Find struct pcpu by cpu address.
 168  */
 169 static struct pcpu *pcpu_find_address(const struct cpumask *mask, int address)
 170 {
 171         int cpu;
 172
 173         for_each_cpu(cpu, mask)
 174                 if (pcpu_devices[cpu].address == address)
 175                         return pcpu_devices + cpu;
 176         return NULL;
 177 }
 178
 179 static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 180 {
 181         int order;
 182
 183         set_bit(ec_bit, &pcpu->ec_mask);
 184         order = pcpu_running(pcpu) ?
 185                 sigp_external_call : sigp_emergency_signal;
 186         pcpu_sigp_retry(pcpu, order, 0);
 187 }
 188
 189 static int __cpuinit pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 190 {
 191         struct _lowcore *lc;
 192
 193         if (pcpu != &pcpu_devices[0]) {
 194                 pcpu->lowcore = (struct _lowcore *)
 195                         __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 196                 pcpu->async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 197                 pcpu->panic_stack = __get_free_page(GFP_KERNEL);
 198                 if (!pcpu->lowcore || !pcpu->panic_stack || !pcpu->async_stack)
 199                         goto out;
 200         }
 201         lc = pcpu->lowcore;
 202         memcpy(lc, &S390_lowcore, 512);
 203         memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 204         lc->async_stack = pcpu->async_stack + ASYNC_SIZE;
 205         lc->panic_stack = pcpu->panic_stack + PAGE_SIZE;
 206         lc->cpu_nr = cpu;
 207 #ifndef CONFIG_64BIT
 208         if (MACHINE_HAS_IEEE) {
 209                 lc->extended_save_area_addr = get_zeroed_page(GFP_KERNEL);
 210                 if (!lc->extended_save_area_addr)
 211                         goto out;
 212         }
 213 #else
 214         if (vdso_alloc_per_cpu(lc))
 215                 goto out;
 216 #endif
 217         lowcore_ptr[cpu] = lc;
 218         pcpu_sigp_retry(pcpu, sigp_set_prefix, (u32)(unsigned long) lc);
 219         return 0;
 220 out:
 221         if (pcpu != &pcpu_devices[0]) {
 222                 free_page(pcpu->panic_stack);
 223                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 224                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 225         }
 226         return -ENOMEM;
 227 }
 228
 229 static void pcpu_free_lowcore(struct pcpu *pcpu)
 230 {
 231         pcpu_sigp_retry(pcpu, sigp_set_prefix, 0);
 232         lowcore_ptr[pcpu - pcpu_devices] = NULL;
 233 #ifndef CONFIG_64BIT
 234         if (MACHINE_HAS_IEEE) {
 235                 struct _lowcore *lc = pcpu->lowcore;
 236
 237                 free_page((unsigned long) lc->extended_save_area_addr);
 238                 lc->extended_save_area_addr = 0;
 239         }
 240 #else
 241         vdso_free_per_cpu(pcpu->lowcore);
 242 #endif
 243         if (pcpu != &pcpu_devices[0]) {
 244                 free_page(pcpu->panic_stack);
 245                 free_pages(pcpu->async_stack, ASYNC_ORDER);
 246                 free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
 247         }
 248 }
 249
 250 static void pcpu_prepare_secondary(struct pcpu *pcpu, int cpu)
 251 {
 252         struct _lowcore *lc = pcpu->lowcore;
 253
 254         atomic_inc(&init_mm.context.attach_count);
 255         lc->cpu_nr = cpu;
 256         lc->percpu_offset = __per_cpu_offset[cpu];
 257         lc->kernel_asce = S390_lowcore.kernel_asce;
 258         lc->machine_flags = S390_lowcore.machine_flags;
 259         lc->ftrace_func = S390_lowcore.ftrace_func;
 260         lc->user_timer = lc->system_timer = lc->steal_timer = 0;
 261         __ctl_store(lc->cregs_save_area, 0, 15);
 262         save_access_regs((unsigned int *) lc->access_regs_save_area);
 263         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 264                MAX_FACILITY_BIT/8);
 265 }
 266
 267 static void pcpu_attach_task(struct pcpu *pcpu, struct task_struct *tsk)
 268 {
 269         struct _lowcore *lc = pcpu->lowcore;
 270         struct thread_info *ti = task_thread_info(tsk);
 271
 272         lc->kernel_stack = (unsigned long) task_stack_page(tsk) + THREAD_SIZE;
 273         lc->thread_info = (unsigned long) task_thread_info(tsk);
 274         lc->current_task = (unsigned long) tsk;
 275         lc->user_timer = ti->user_timer;
 276         lc->system_timer = ti->system_timer;
 277         lc->steal_timer = 0;
 278 }
 279
 280 static void pcpu_start_fn(struct pcpu *pcpu, void (*func)(void *), void *data)
 281 {
 282         struct _lowcore *lc = pcpu->lowcore;
 283
 284         lc->restart_stack = lc->kernel_stack;
 285         lc->restart_fn = (unsigned long) func;
 286         lc->restart_data = (unsigned long) data;
 287         lc->restart_source = -1UL;
 288         pcpu_sigp_retry(pcpu, sigp_restart, 0);
 289 }
 290
 291 /*
 292  * Call function via PSW restart on pcpu and stop the current cpu.
 293  */
 294 static void pcpu_delegate(struct pcpu *pcpu, void (*func)(void *),
 295                           void *data, unsigned long stack)
 296 {
 297         struct _lowcore *lc = pcpu->lowcore;
 298         unsigned short this_cpu;
 299
 300         __load_psw_mask(psw_kernel_bits);
 301         this_cpu = stap();
 302         if (pcpu->address == this_cpu)
 303                 func(data);     /* should not return */
 304         /* Stop target cpu (if func returns this stops the current cpu). */
 305         pcpu_sigp_retry(pcpu, sigp_stop, 0);
 306         /* Restart func on the target cpu and stop the current cpu. */
 307         lc->restart_stack = stack;
 308         lc->restart_fn = (unsigned long) func;
 309         lc->restart_data = (unsigned long) data;
 310         lc->restart_source = (unsigned long) this_cpu;
 311         asm volatile(
 312                 "0:     sigp    0,%0,6  # sigp restart to target cpu\n"
 313                 "       brc     2,0b    # busy, try again\n"
 314                 "1:     sigp    0,%1,5  # sigp stop to current cpu\n"
 315                 "       brc     2,1b    # busy, try again\n"
 316                 : : "d" (pcpu->address), "d" (this_cpu) : "0", "1", "cc");
 317         for (;;) ;
 318 }
 319
 320 /*
 321  * Call function on an online CPU.
 322  */
 323 void smp_call_online_cpu(void (*func)(void *), void *data)
 324 {
 325         struct pcpu *pcpu;
 326
 327         /* Use the current cpu if it is online. */
 328         pcpu = pcpu_find_address(cpu_online_mask, stap());
 329         if (!pcpu)
 330                 /* Use the first online cpu. */
 331                 pcpu = pcpu_devices + cpumask_first(cpu_online_mask);
 332         pcpu_delegate(pcpu, func, data, (unsigned long) restart_stack);
 333 }
 334
 335 /*
 336  * Call function on the ipl CPU.
 337  */
 338 void smp_call_ipl_cpu(void (*func)(void *), void *data)
 339 {
 340         pcpu_delegate(&pcpu_devices[0], func, data,
 341                       pcpu_devices->panic_stack + PAGE_SIZE);
 342 }
 343
 344 int smp_find_processor_id(u16 address)
 345 {
 346         int cpu;
 347
 348         for_each_present_cpu(cpu)
 349                 if (pcpu_devices[cpu].address == address)
 350                         return cpu;
 351         return -1;
 352 }
 353
 354 int smp_vcpu_scheduled(int cpu)
 355 {
 356         return pcpu_running(pcpu_devices + cpu);
 357 }
 358
 359 void smp_yield(void)
 360 {
 361         if (MACHINE_HAS_DIAG44)
 362                 asm volatile("diag 0,0,0x44");
 363 }
 364
 365 void smp_yield_cpu(int cpu)
 366 {
 367         if (MACHINE_HAS_DIAG9C)
 368                 asm volatile("diag %0,0,0x9c"
 369                              : : "d" (pcpu_devices[cpu].address));
 370         else if (MACHINE_HAS_DIAG44)
 371                 asm volatile("diag 0,0,0x44");
 372 }
 373
 374 /*
 375  * Send cpus emergency shutdown signal. This gives the cpus the
 376  * opportunity to complete outstanding interrupts.
 377  */
 378 void smp_emergency_stop(cpumask_t *cpumask)
 379 {
 380         u64 end;
 381         int cpu;
 382
 383         end = get_clock() + (1000000UL << 12);
 384         for_each_cpu(cpu, cpumask) {
 385                 struct pcpu *pcpu = pcpu_devices + cpu;
 386                 set_bit(ec_stop_cpu, &pcpu->ec_mask);
 387                 while (__pcpu_sigp(pcpu->address, sigp_emergency_signal,
 388                                    0, NULL) == sigp_busy &&
 389                        get_clock() < end)
 390                         cpu_relax();
 391         }
 392         while (get_clock() < end) {
 393                 for_each_cpu(cpu, cpumask)
 394                         if (pcpu_stopped(pcpu_devices + cpu))
 395                                 cpumask_clear_cpu(cpu, cpumask);
 396                 if (cpumask_empty(cpumask))
 397                         break;
 398                 cpu_relax();
 399         }
 400 }
 401
 402 /*
 403  * Stop all cpus but the current one.
 404  */
 405 void smp_send_stop(void)
 406 {
 407         cpumask_t cpumask;
 408         int cpu;
 409
 410         /* Disable all interrupts/machine checks */
 411         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 412         trace_hardirqs_off();
 413
 414         debug_set_critical();
 415         cpumask_copy(&cpumask, cpu_online_mask);
 416         cpumask_clear_cpu(smp_processor_id(), &cpumask);
 417
 418         if (oops_in_progress)
 419                 smp_emergency_stop(&cpumask);
 420
 421         /* stop all processors */
 422         for_each_cpu(cpu, &cpumask) {
 423                 struct pcpu *pcpu = pcpu_devices + cpu;
 424                 pcpu_sigp_retry(pcpu, sigp_stop, 0);
 425                 while (!pcpu_stopped(pcpu))
 426                         cpu_relax();
 427         }
 428 }
 429
 430 /*
 431  * Stop the current cpu.
 432  */
 433 void smp_stop_cpu(void)
 434 {
 435         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
 436         for (;;) ;
 437 }
 438
 439 /*
 440  * This is the main routine where commands issued by other
 441  * cpus are handled.
 442  */
 443 static void do_ext_call_interrupt(struct ext_code ext_code,
 444                                   unsigned int param32, unsigned long param64)
 445 {
 446         unsigned long bits;
 447         int cpu;
 448
 449         cpu = smp_processor_id();
 450         if (ext_code.code == 0x1202)
 451                 kstat_cpu(cpu).irqs[EXTINT_EXC]++;
 452         else
 453                 kstat_cpu(cpu).irqs[EXTINT_EMS]++;
 454         /*
 455          * handle bit signal external calls
 456          */
 457         bits = xchg(&pcpu_devices[cpu].ec_mask, 0);
 458
 459         if (test_bit(ec_stop_cpu, &bits))
 460                 smp_stop_cpu();
 461
 462         if (test_bit(ec_schedule, &bits))
 463                 scheduler_ipi();
 464
 465         if (test_bit(ec_call_function, &bits))
 466                 generic_smp_call_function_interrupt();
 467
 468         if (test_bit(ec_call_function_single, &bits))
 469                 generic_smp_call_function_single_interrupt();
 470
 471 }
 472
 473 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 474 {
 475         int cpu;
 476
 477         for_each_cpu(cpu, mask)
 478                 pcpu_ec_call(pcpu_devices + cpu, ec_call_function);
 479 }
 480
 481 void arch_send_call_function_single_ipi(int cpu)
 482 {
 483         pcpu_ec_call(pcpu_devices + cpu, ec_call_function_single);
 484 }
 485
 486 #ifndef CONFIG_64BIT
 487 /*
 488  * this function sends a 'purge tlb' signal to another CPU.
 489  */
 490 static void smp_ptlb_callback(void *info)
 491 {
 492         __tlb_flush_local();
 493 }
 494
 495 void smp_ptlb_all(void)
 496 {
 497         on_each_cpu(smp_ptlb_callback, NULL, 1);
 498 }
 499 EXPORT_SYMBOL(smp_ptlb_all);
 500 #endif /* ! CONFIG_64BIT */
 501
 502 /*
 503  * this function sends a 'reschedule' IPI to another CPU.
 504  * it goes straight through and wastes no time serializing
 505  * anything. Worst case is that we lose a reschedule ...
 506  */
 507 void smp_send_reschedule(int cpu)
 508 {
 509         pcpu_ec_call(pcpu_devices + cpu, ec_schedule);
 510 }
 511
 512 /*
 513  * parameter area for the set/clear control bit callbacks
 514  */
 515 struct ec_creg_mask_parms {
 516         unsigned long orval;
 517         unsigned long andval;
 518         int cr;
 519 };
 520
 521 /*
 522  * callback for setting/clearing control bits
 523  */
 524 static void smp_ctl_bit_callback(void *info)
 525 {
 526         struct ec_creg_mask_parms *pp = info;
 527         unsigned long cregs[16];
 528
 529         __ctl_store(cregs, 0, 15);
 530         cregs[pp->cr] = (cregs[pp->cr] & pp->andval) | pp->orval;
 531         __ctl_load(cregs, 0, 15);
 532 }
 533
 534 /*
 535  * Set a bit in a control register of all cpus
 536  */
 537 void smp_ctl_set_bit(int cr, int bit)
 538 {
 539         struct ec_creg_mask_parms parms = { 1UL << bit, -1UL, cr };
 540
 541         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 542 }
 543 EXPORT_SYMBOL(smp_ctl_set_bit);
 544
 545 /*
 546  * Clear a bit in a control register of all cpus
 547  */
 548 void smp_ctl_clear_bit(int cr, int bit)
 549 {
 550         struct ec_creg_mask_parms parms = { 0, ~(1UL << bit), cr };
 551
 552         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 553 }
 554 EXPORT_SYMBOL(smp_ctl_clear_bit);
 555
 556 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_CRASH_DUMP)
 557
 558 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 559 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 560
 561 static void __init smp_get_save_area(int cpu, u16 address)
 562 {
 563         void *lc = pcpu_devices[0].lowcore;
 564         struct save_area *save_area;
 565
 566         if (is_kdump_kernel())
 567                 return;
 568         if (!OLDMEM_BASE && (address == boot_cpu_address ||
 569                              ipl_info.type != IPL_TYPE_FCP_DUMP))
 570                 return;
 571         if (cpu >= NR_CPUS) {
 572                 pr_warning("CPU %i exceeds the maximum %i and is excluded "
 573                            "from the dump\n", cpu, NR_CPUS - 1);
 574                 return;
 575         }
 576         save_area = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 577         if (!save_area)
 578                 panic("could not allocate memory for save area\n");
 579         zfcpdump_save_areas[cpu] = save_area;
 580 #ifdef CONFIG_CRASH_DUMP
 581         if (address == boot_cpu_address) {
 582                 /* Copy the registers of the boot cpu. */
 583                 copy_oldmem_page(1, (void *) save_area, sizeof(*save_area),
 584                                  SAVE_AREA_BASE - PAGE_SIZE, 0);
 585                 return;
 586         }
 587 #endif
 588         /* Get the registers of a non-boot cpu. */
 589         __pcpu_sigp_relax(address, sigp_stop_and_store_status, 0, NULL);
 590         memcpy_real(save_area, lc + SAVE_AREA_BASE, sizeof(*save_area));
 591 }
 592
 593 int smp_store_status(int cpu)
 594 {
 595         struct pcpu *pcpu;
 596
 597         pcpu = pcpu_devices + cpu;
 598         if (__pcpu_sigp_relax(pcpu->address, sigp_stop_and_store_status,
 599                               0, NULL) != sigp_order_code_accepted)
 600                 return -EIO;
 601         return 0;
 602 }
 603
 604 #else /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 605
 606 static inline void smp_get_save_area(int cpu, u16 address) { }
 607
 608 #endif /* CONFIG_ZFCPDUMP || CONFIG_CRASH_DUMP */
 609
 610 static struct sclp_cpu_info *smp_get_cpu_info(void)
 611 {
 612         static int use_sigp_detection;
 613         struct sclp_cpu_info *info;
 614         int address;
 615
 616         info = kzalloc(sizeof(*info), GFP_KERNEL);
 617         if (info && (use_sigp_detection || sclp_get_cpu_info(info))) {
 618                 use_sigp_detection = 1;
 619                 for (address = 0; address <= MAX_CPU_ADDRESS; address++) {
 620                         if (__pcpu_sigp_relax(address, sigp_sense, 0, NULL) ==
 621                             sigp_not_operational)
 622                                 continue;
 623                         info->cpu[info->configured].address = address;
 624                         info->configured++;
 625                 }
 626                 info->combined = info->configured;
 627         }
 628         return info;
 629 }
 630
 631 static int __devinit smp_add_present_cpu(int cpu);
 632
 633 static int __devinit __smp_rescan_cpus(struct sclp_cpu_info *info,
 634                                        int sysfs_add)
 635 {
 636         struct pcpu *pcpu;
 637         cpumask_t avail;
 638         int cpu, nr, i;
 639
 640         nr = 0;
 641         cpumask_xor(&avail, cpu_possible_mask, cpu_present_mask);
 642         cpu = cpumask_first(&avail);
 643         for (i = 0; (i < info->combined) && (cpu < nr_cpu_ids); i++) {
 644                 if (info->has_cpu_type && info->cpu[i].type != boot_cpu_type)
 645                         continue;
 646                 if (pcpu_find_address(cpu_present_mask, info->cpu[i].address))
 647                         continue;
 648                 pcpu = pcpu_devices + cpu;
 649                 pcpu->address = info->cpu[i].address;
 650                 pcpu->state = (cpu >= info->configured) ?
 651                         CPU_STATE_STANDBY : CPU_STATE_CONFIGURED;
 652                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 653                 set_cpu_present(cpu, true);
 654                 if (sysfs_add && smp_add_present_cpu(cpu) != 0)
 655                         set_cpu_present(cpu, false);
 656                 else
 657                         nr++;
 658                 cpu = cpumask_next(cpu, &avail);
 659         }
 660         return nr;
 661 }
 662
 663 static void __init smp_detect_cpus(void)
 664 {
 665         unsigned int cpu, c_cpus, s_cpus;
 666         struct sclp_cpu_info *info;
 667
 668         info = smp_get_cpu_info();
 669         if (!info)
 670                 panic("smp_detect_cpus failed to allocate memory\n");
 671         if (info->has_cpu_type) {
 672                 for (cpu = 0; cpu < info->combined; cpu++) {
 673                         if (info->cpu[cpu].address != boot_cpu_address)
 674                                 continue;
 675                         /* The boot cpu dictates the cpu type. */
 676                         boot_cpu_type = info->cpu[cpu].type;
 677                         break;
 678                 }
 679         }
 680         c_cpus = s_cpus = 0;
 681         for (cpu = 0; cpu < info->combined; cpu++) {
 682                 if (info->has_cpu_type && info->cpu[cpu].type != boot_cpu_type)
 683                         continue;
 684                 if (cpu < info->configured) {
 685                         smp_get_save_area(c_cpus, info->cpu[cpu].address);
 686                         c_cpus++;
 687                 } else
 688                         s_cpus++;
 689         }
 690         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 691         get_online_cpus();
 692         __smp_rescan_cpus(info, 0);
 693         put_online_cpus();
 694         kfree(info);
 695 }
 696
 697 /*
 698  *      Activate a secondary processor.
 699  */
 700 static void __cpuinit smp_start_secondary(void *cpuvoid)
 701 {
 702         S390_lowcore.last_update_clock = get_clock();
 703         S390_lowcore.restart_stack = (unsigned long) restart_stack;
 704         S390_lowcore.restart_fn = (unsigned long) do_restart;
 705         S390_lowcore.restart_data = 0;
 706         S390_lowcore.restart_source = -1UL;
 707         restore_access_regs(S390_lowcore.access_regs_save_area);
 708         __ctl_load(S390_lowcore.cregs_save_area, 0, 15);
 709         __load_psw_mask(psw_kernel_bits | PSW_MASK_DAT);
 710         cpu_init();
 711         preempt_disable();
 712         init_cpu_timer();
 713         init_cpu_vtimer();
 714         pfault_init();
 715         notify_cpu_starting(smp_processor_id());
 716         ipi_call_lock();
 717         set_cpu_online(smp_processor_id(), true);
 718         ipi_call_unlock();
 719         local_irq_enable();
 720         /* cpu_idle will call schedule for us */
 721         cpu_idle();
 722 }
 723
 724 struct create_idle {
 725         struct work_struct work;
 726         struct task_struct *idle;
 727         struct completion done;
 728         int cpu;
 729 };
 730
 731 static void __cpuinit smp_fork_idle(struct work_struct *work)
 732 {
 733         struct create_idle *c_idle;
 734
 735         c_idle = container_of(work, struct create_idle, work);
 736         c_idle->idle = fork_idle(c_idle->cpu);
 737         complete(&c_idle->done);
 738 }
 739
 740 /* Upping and downing of CPUs */
 741 int __cpuinit __cpu_up(unsigned int cpu)
 742 {
 743         struct create_idle c_idle;
 744         struct pcpu *pcpu;
 745         int rc;
 746
 747         pcpu = pcpu_devices + cpu;
 748         if (pcpu->state != CPU_STATE_CONFIGURED)
 749                 return -EIO;
 750         if (pcpu_sigp_retry(pcpu, sigp_initial_cpu_reset, 0) !=
 751             sigp_order_code_accepted)
 752                 return -EIO;
 753         if (!pcpu->idle) {
 754                 c_idle.done = COMPLETION_INITIALIZER_ONSTACK(c_idle.done);
 755                 INIT_WORK_ONSTACK(&c_idle.work, smp_fork_idle);
 756                 c_idle.cpu = cpu;
 757                 schedule_work(&c_idle.work);
 758                 wait_for_completion(&c_idle.done);
 759                 if (IS_ERR(c_idle.idle))
 760                         return PTR_ERR(c_idle.idle);
 761                 pcpu->idle = c_idle.idle;
 762         }
 763         init_idle(pcpu->idle, cpu);
 764         rc = pcpu_alloc_lowcore(pcpu, cpu);
 765         if (rc)
 766                 return rc;
 767         pcpu_prepare_secondary(pcpu, cpu);
 768         pcpu_attach_task(pcpu, pcpu->idle);
 769         pcpu_start_fn(pcpu, smp_start_secondary, NULL);
 770         while (!cpu_online(cpu))
 771                 cpu_relax();
 772         return 0;
 773 }
 774
 775 static int __init setup_possible_cpus(char *s)
 776 {
 777         int max, cpu;
 778
 779         if (kstrtoint(s, 0, &max) < 0)
 780                 return 0;
 781         init_cpu_possible(cpumask_of(0));
 782         for (cpu = 1; cpu < max && cpu < nr_cpu_ids; cpu++)
 783                 set_cpu_possible(cpu, true);
 784         return 0;
 785 }
 786 early_param("possible_cpus", setup_possible_cpus);
 787
 788 #ifdef CONFIG_HOTPLUG_CPU
 789
 790 int __cpu_disable(void)
 791 {
 792         unsigned long cregs[16];
 793
 794         set_cpu_online(smp_processor_id(), false);
 795         /* Disable pseudo page faults on this cpu. */
 796         pfault_fini();
 797         /* Disable interrupt sources via control register. */
 798         __ctl_store(cregs, 0, 15);
 799         cregs[0]  &= ~0x0000ee70UL;     /* disable all external interrupts */
 800         cregs[6]  &= ~0xff000000UL;     /* disable all I/O interrupts */
 801         cregs[14] &= ~0x1f000000UL;     /* disable most machine checks */
 802         __ctl_load(cregs, 0, 15);
 803         return 0;
 804 }
 805
 806 void __cpu_die(unsigned int cpu)
 807 {
 808         struct pcpu *pcpu;
 809
 810         /* Wait until target cpu is down */
 811         pcpu = pcpu_devices + cpu;
 812         while (!pcpu_stopped(pcpu))
 813                 cpu_relax();
 814         pcpu_free_lowcore(pcpu);
 815         atomic_dec(&init_mm.context.attach_count);
 816 }
 817
 818 void __noreturn cpu_die(void)
 819 {
 820         idle_task_exit();
 821         pcpu_sigp_retry(pcpu_devices + smp_processor_id(), sigp_stop, 0);
 822         for (;;) ;
 823 }
 824
 825 #endif /* CONFIG_HOTPLUG_CPU */
 826
 827 static void smp_call_os_info_init_fn(void)
 828 {
 829         int (*init_fn)(void);
 830         unsigned long size;
 831
 832         init_fn = os_info_old_entry(OS_INFO_INIT_FN, &size);
 833         if (!init_fn)
 834                 return;
 835         init_fn();
 836 }
 837
 838 void __init smp_prepare_cpus(unsigned int max_cpus)
 839 {
 840         /* request the 0x1201 emergency signal external interrupt */
 841         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 842                 panic("Couldn't request external interrupt 0x1201");
 843         /* request the 0x1202 external call external interrupt */
 844         if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
 845                 panic("Couldn't request external interrupt 0x1202");
 846         smp_call_os_info_init_fn();
 847         smp_detect_cpus();
 848 }
 849
 850 void __init smp_prepare_boot_cpu(void)
 851 {
 852         struct pcpu *pcpu = pcpu_devices;
 853
 854         boot_cpu_address = stap();
 855         pcpu->idle = current;
 856         pcpu->state = CPU_STATE_CONFIGURED;
 857         pcpu->address = boot_cpu_address;
 858         pcpu->lowcore = (struct _lowcore *)(unsigned long) store_prefix();
 859         pcpu->async_stack = S390_lowcore.async_stack - ASYNC_SIZE;
 860         pcpu->panic_stack = S390_lowcore.panic_stack - PAGE_SIZE;
 861         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 862         cpu_set_polarization(0, POLARIZATION_UNKNOWN);
 863         set_cpu_present(0, true);
 864         set_cpu_online(0, true);
 865 }
 866
 867 void __init smp_cpus_done(unsigned int max_cpus)
 868 {
 869 }
 870
 871 void __init smp_setup_processor_id(void)
 872 {
 873         S390_lowcore.cpu_nr = 0;
 874 }
 875
 876 /*
 877  * the frequency of the profiling timer can be changed
 878  * by writing a multiplier value into /proc/profile.
 879  *
 880  * usually you want to run this on all CPUs ;)
 881  */
 882 int setup_profiling_timer(unsigned int multiplier)
 883 {
 884         return 0;
 885 }
 886
 887 #ifdef CONFIG_HOTPLUG_CPU
 888 static ssize_t cpu_configure_show(struct device *dev,
 889                                   struct device_attribute *attr, char *buf)
 890 {
 891         ssize_t count;
 892
 893         mutex_lock(&smp_cpu_state_mutex);
 894         count = sprintf(buf, "%d\n", pcpu_devices[dev->id].state);
 895         mutex_unlock(&smp_cpu_state_mutex);
 896         return count;
 897 }
 898
 899 static ssize_t cpu_configure_store(struct device *dev,
 900                                    struct device_attribute *attr,
 901                                    const char *buf, size_t count)
 902 {
 903         struct pcpu *pcpu;
 904         int cpu, val, rc;
 905         char delim;
 906
 907         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 908                 return -EINVAL;
 909         if (val != 0 && val != 1)
 910                 return -EINVAL;
 911         get_online_cpus();
 912         mutex_lock(&smp_cpu_state_mutex);
 913         rc = -EBUSY;
 914         /* disallow configuration changes of online cpus and cpu 0 */
 915         cpu = dev->id;
 916         if (cpu_online(cpu) || cpu == 0)
 917                 goto out;
 918         pcpu = pcpu_devices + cpu;
 919         rc = 0;
 920         switch (val) {
 921         case 0:
 922                 if (pcpu->state != CPU_STATE_CONFIGURED)
 923                         break;
 924                 rc = sclp_cpu_deconfigure(pcpu->address);
 925                 if (rc)
 926                         break;
 927                 pcpu->state = CPU_STATE_STANDBY;
 928                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 929                 topology_expect_change();
 930                 break;
 931         case 1:
 932                 if (pcpu->state != CPU_STATE_STANDBY)
 933                         break;
 934                 rc = sclp_cpu_configure(pcpu->address);
 935                 if (rc)
 936                         break;
 937                 pcpu->state = CPU_STATE_CONFIGURED;
 938                 cpu_set_polarization(cpu, POLARIZATION_UNKNOWN);
 939                 topology_expect_change();
 940                 break;
 941         default:
 942                 break;
 943         }
 944 out:
 945         mutex_unlock(&smp_cpu_state_mutex);
 946         put_online_cpus();
 947         return rc ? rc : count;
 948 }
 949 static DEVICE_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 950 #endif /* CONFIG_HOTPLUG_CPU */
 951
 952 static ssize_t show_cpu_address(struct device *dev,
 953                                 struct device_attribute *attr, char *buf)
 954 {
 955         return sprintf(buf, "%d\n", pcpu_devices[dev->id].address);
 956 }
 957 static DEVICE_ATTR(address, 0444, show_cpu_address, NULL);
 958
 959 static struct attribute *cpu_common_attrs[] = {
 960 #ifdef CONFIG_HOTPLUG_CPU
 961         &dev_attr_configure.attr,
 962 #endif
 963         &dev_attr_address.attr,
 964         NULL,
 965 };
 966
 967 static struct attribute_group cpu_common_attr_group = {
 968         .attrs = cpu_common_attrs,
 969 };
 970
 971 static ssize_t show_capability(struct device *dev,
 972                                 struct device_attribute *attr, char *buf)
 973 {
 974         unsigned int capability;
 975         int rc;
 976
 977         rc = get_cpu_capability(&capability);
 978         if (rc)
 979                 return rc;
 980         return sprintf(buf, "%u\n", capability);
 981 }
 982 static DEVICE_ATTR(capability, 0444, show_capability, NULL);
 983
 984 static ssize_t show_idle_count(struct device *dev,
 985                                 struct device_attribute *attr, char *buf)
 986 {
 987         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
 988         unsigned long long idle_count;
 989         unsigned int sequence;
 990
 991         do {
 992                 sequence = ACCESS_ONCE(idle->sequence);
 993                 idle_count = ACCESS_ONCE(idle->idle_count);
 994                 if (ACCESS_ONCE(idle->idle_enter))
 995                         idle_count++;
 996         } while ((sequence & 1) || (idle->sequence != sequence));
 997         return sprintf(buf, "%llu\n", idle_count);
 998 }
 999 static DEVICE_ATTR(idle_count, 0444, show_idle_count, NULL);
1000
1001 static ssize_t show_idle_time(struct device *dev,
1002                                 struct device_attribute *attr, char *buf)
1003 {
1004         struct s390_idle_data *idle = &per_cpu(s390_idle, dev->id);
1005         unsigned long long now, idle_time, idle_enter, idle_exit;
1006         unsigned int sequence;
1007
1008         do {
1009                 now = get_clock();
1010                 sequence = ACCESS_ONCE(idle->sequence);
1011                 idle_time = ACCESS_ONCE(idle->idle_time);
1012                 idle_enter = ACCESS_ONCE(idle->idle_enter);
1013                 idle_exit = ACCESS_ONCE(idle->idle_exit);
1014         } while ((sequence & 1) || (idle->sequence != sequence));
1015         idle_time += idle_enter ? ((idle_exit ? : now) - idle_enter) : 0;
1016         return sprintf(buf, "%llu\n", idle_time >> 12);
1017 }
1018 static DEVICE_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1019
1020 static struct attribute *cpu_online_attrs[] = {
1021         &dev_attr_capability.attr,
1022         &dev_attr_idle_count.attr,
1023         &dev_attr_idle_time_us.attr,
1024         NULL,
1025 };
1026
1027 static struct attribute_group cpu_online_attr_group = {
1028         .attrs = cpu_online_attrs,
1029 };
1030
1031 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1032                                     unsigned long action, void *hcpu)
1033 {
1034         unsigned int cpu = (unsigned int)(long)hcpu;
1035         struct cpu *c = &pcpu_devices[cpu].cpu;
1036         struct device *s = &c->dev;
1037         struct s390_idle_data *idle;
1038         int err = 0;
1039
1040         switch (action) {
1041         case CPU_ONLINE:
1042         case CPU_ONLINE_FROZEN:
1043                 idle = &per_cpu(s390_idle, cpu);
1044                 memset(idle, 0, sizeof(struct s390_idle_data));
1045                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1046                 break;
1047         case CPU_DEAD:
1048         case CPU_DEAD_FROZEN:
1049                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1050                 break;
1051         }
1052         return notifier_from_errno(err);
1053 }
1054
1055 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1056         .notifier_call = smp_cpu_notify,
1057 };
1058
1059 static int __devinit smp_add_present_cpu(int cpu)
1060 {
1061         struct cpu *c = &pcpu_devices[cpu].cpu;
1062         struct device *s = &c->dev;
1063         int rc;
1064
1065         c->hotpluggable = 1;
1066         rc = register_cpu(c, cpu);
1067         if (rc)
1068                 goto out;
1069         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1070         if (rc)
1071                 goto out_cpu;
1072         if (cpu_online(cpu)) {
1073                 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1074                 if (rc)
1075                         goto out_online;
1076         }
1077         rc = topology_cpu_init(c);
1078         if (rc)
1079                 goto out_topology;
1080         return 0;
1081
1082 out_topology:
1083         if (cpu_online(cpu))
1084                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1085 out_online:
1086         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1087 out_cpu:
1088 #ifdef CONFIG_HOTPLUG_CPU
1089         unregister_cpu(c);
1090 #endif
1091 out:
1092         return rc;
1093 }
1094
1095 #ifdef CONFIG_HOTPLUG_CPU
1096
1097 int __ref smp_rescan_cpus(void)
1098 {
1099         struct sclp_cpu_info *info;
1100         int nr;
1101
1102         info = smp_get_cpu_info();
1103         if (!info)
1104                 return -ENOMEM;
1105         get_online_cpus();
1106         mutex_lock(&smp_cpu_state_mutex);
1107         nr = __smp_rescan_cpus(info, 1);
1108         mutex_unlock(&smp_cpu_state_mutex);
1109         put_online_cpus();
1110         kfree(info);
1111         if (nr)
1112                 topology_schedule_update();
1113         return 0;
1114 }
1115
1116 static ssize_t __ref rescan_store(struct device *dev,
1117                                   struct device_attribute *attr,
1118                                   const char *buf,
1119                                   size_t count)
1120 {
1121         int rc;
1122
1123         rc = smp_rescan_cpus();
1124         return rc ? rc : count;
1125 }
1126 static DEVICE_ATTR(rescan, 0200, NULL, rescan_store);
1127 #endif /* CONFIG_HOTPLUG_CPU */
1128
1129 static int __init s390_smp_init(void)
1130 {
1131         int cpu, rc;
1132
1133         register_cpu_notifier(&smp_cpu_nb);
1134 #ifdef CONFIG_HOTPLUG_CPU
1135         rc = device_create_file(cpu_subsys.dev_root, &dev_attr_rescan);
1136         if (rc)
1137                 return rc;
1138 #endif
1139         for_each_present_cpu(cpu) {
1140                 rc = smp_add_present_cpu(cpu);
1141                 if (rc)
1142                         return rc;
1143         }
1144         return 0;
1145 }
1146 subsys_initcall(s390_smp_init);