arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999, 2009
   5  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
   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  * We work with logical cpu numbering everywhere we can. The only
  14  * functions using the real cpu address (got from STAP) are the sigp
  15  * functions. For all other functions we use the identity mapping.
  16  * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
  17  * used e.g. to find the idle task belonging to a logical cpu. Every array
  18  * in the kernel is sorted by the logical cpu number and not by the physical
  19  * one which is causing all the confusion with __cpu_logical_map and
  20  * cpu_number_map in other architectures.
  21  */
  22
  23 #define KMSG_COMPONENT "cpu"
  24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  25
  26 #include <linux/module.h>
  27 #include <linux/init.h>
  28 #include <linux/mm.h>
  29 #include <linux/err.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/kernel_stat.h>
  32 #include <linux/delay.h>
  33 #include <linux/cache.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/irqflags.h>
  36 #include <linux/cpu.h>
  37 #include <linux/timex.h>
  38 #include <linux/bootmem.h>
  39 #include <linux/slab.h>
  40 #include <asm/asm-offsets.h>
  41 #include <asm/ipl.h>
  42 #include <asm/setup.h>
  43 #include <asm/sigp.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/irq.h>
  46 #include <asm/s390_ext.h>
  47 #include <asm/cpcmd.h>
  48 #include <asm/tlbflush.h>
  49 #include <asm/timer.h>
  50 #include <asm/lowcore.h>
  51 #include <asm/sclp.h>
  52 #include <asm/cputime.h>
  53 #include <asm/vdso.h>
  54 #include <asm/cpu.h>
  55 #include "entry.h"
  56
  57 /* logical cpu to cpu address */
  58 unsigned short __cpu_logical_map[NR_CPUS];
  59
  60 static struct task_struct *current_set[NR_CPUS];
  61
  62 static u8 smp_cpu_type;
  63 static int smp_use_sigp_detection;
  64
  65 enum s390_cpu_state {
  66         CPU_STATE_STANDBY,
  67         CPU_STATE_CONFIGURED,
  68 };
  69
  70 DEFINE_MUTEX(smp_cpu_state_mutex);
  71 int smp_cpu_polarization[NR_CPUS];
  72 static int smp_cpu_state[NR_CPUS];
  73 static int cpu_management;
  74
  75 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  76
  77 static void smp_ext_bitcall(int, int);
  78
  79 static int raw_cpu_stopped(int cpu)
  80 {
  81         u32 status;
  82
  83         switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
  84         case sigp_status_stored:
  85                 /* Check for stopped and check stop state */
  86                 if (status & 0x50)
  87                         return 1;
  88                 break;
  89         default:
  90                 break;
  91         }
  92         return 0;
  93 }
  94
  95 static inline int cpu_stopped(int cpu)
  96 {
  97         return raw_cpu_stopped(cpu_logical_map(cpu));
  98 }
  99
 100 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
 101 {
 102         struct _lowcore *lc, *current_lc;
 103         struct stack_frame *sf;
 104         struct pt_regs *regs;
 105         unsigned long sp;
 106
 107         if (smp_processor_id() == 0)
 108                 func(data);
 109         __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
 110         /* Disable lowcore protection */
 111         __ctl_clear_bit(0, 28);
 112         current_lc = lowcore_ptr[smp_processor_id()];
 113         lc = lowcore_ptr[0];
 114         if (!lc)
 115                 lc = current_lc;
 116         lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 117         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
 118         if (!cpu_online(0))
 119                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
 120         while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
 121                 cpu_relax();
 122         sp = lc->panic_stack;
 123         sp -= sizeof(struct pt_regs);
 124         regs = (struct pt_regs *) sp;
 125         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
 126         regs->psw = lc->psw_save_area;
 127         sp -= STACK_FRAME_OVERHEAD;
 128         sf = (struct stack_frame *) sp;
 129         sf->back_chain = regs->gprs[15];
 130         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 131 }
 132
 133 void smp_send_stop(void)
 134 {
 135         int cpu, rc;
 136
 137         /* Disable all interrupts/machine checks */
 138         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 139         trace_hardirqs_off();
 140
 141         /* stop all processors */
 142         for_each_online_cpu(cpu) {
 143                 if (cpu == smp_processor_id())
 144                         continue;
 145                 do {
 146                         rc = sigp(cpu, sigp_stop);
 147                 } while (rc == sigp_busy);
 148
 149                 while (!cpu_stopped(cpu))
 150                         cpu_relax();
 151         }
 152 }
 153
 154 /*
 155  * This is the main routine where commands issued by other
 156  * cpus are handled.
 157  */
 158
 159 static void do_ext_call_interrupt(__u16 code)
 160 {
 161         unsigned long bits;
 162
 163         /*
 164          * handle bit signal external calls
 165          *
 166          * For the ec_schedule signal we have to do nothing. All the work
 167          * is done automatically when we return from the interrupt.
 168          */
 169         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 170
 171         if (test_bit(ec_call_function, &bits))
 172                 generic_smp_call_function_interrupt();
 173
 174         if (test_bit(ec_call_function_single, &bits))
 175                 generic_smp_call_function_single_interrupt();
 176 }
 177
 178 /*
 179  * Send an external call sigp to another cpu and return without waiting
 180  * for its completion.
 181  */
 182 static void smp_ext_bitcall(int cpu, int sig)
 183 {
 184         /*
 185          * Set signaling bit in lowcore of target cpu and kick it
 186          */
 187         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 188         while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
 189                 udelay(10);
 190 }
 191
 192 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 193 {
 194         int cpu;
 195
 196         for_each_cpu(cpu, mask)
 197                 smp_ext_bitcall(cpu, ec_call_function);
 198 }
 199
 200 void arch_send_call_function_single_ipi(int cpu)
 201 {
 202         smp_ext_bitcall(cpu, ec_call_function_single);
 203 }
 204
 205 #ifndef CONFIG_64BIT
 206 /*
 207  * this function sends a 'purge tlb' signal to another CPU.
 208  */
 209 static void smp_ptlb_callback(void *info)
 210 {
 211         __tlb_flush_local();
 212 }
 213
 214 void smp_ptlb_all(void)
 215 {
 216         on_each_cpu(smp_ptlb_callback, NULL, 1);
 217 }
 218 EXPORT_SYMBOL(smp_ptlb_all);
 219 #endif /* ! CONFIG_64BIT */
 220
 221 /*
 222  * this function sends a 'reschedule' IPI to another CPU.
 223  * it goes straight through and wastes no time serializing
 224  * anything. Worst case is that we lose a reschedule ...
 225  */
 226 void smp_send_reschedule(int cpu)
 227 {
 228         smp_ext_bitcall(cpu, ec_schedule);
 229 }
 230
 231 /*
 232  * parameter area for the set/clear control bit callbacks
 233  */
 234 struct ec_creg_mask_parms {
 235         unsigned long orvals[16];
 236         unsigned long andvals[16];
 237 };
 238
 239 /*
 240  * callback for setting/clearing control bits
 241  */
 242 static void smp_ctl_bit_callback(void *info)
 243 {
 244         struct ec_creg_mask_parms *pp = info;
 245         unsigned long cregs[16];
 246         int i;
 247
 248         __ctl_store(cregs, 0, 15);
 249         for (i = 0; i <= 15; i++)
 250                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 251         __ctl_load(cregs, 0, 15);
 252 }
 253
 254 /*
 255  * Set a bit in a control register of all cpus
 256  */
 257 void smp_ctl_set_bit(int cr, int bit)
 258 {
 259         struct ec_creg_mask_parms parms;
 260
 261         memset(&parms.orvals, 0, sizeof(parms.orvals));
 262         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 263         parms.orvals[cr] = 1 << bit;
 264         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 265 }
 266 EXPORT_SYMBOL(smp_ctl_set_bit);
 267
 268 /*
 269  * Clear a bit in a control register of all cpus
 270  */
 271 void smp_ctl_clear_bit(int cr, int bit)
 272 {
 273         struct ec_creg_mask_parms parms;
 274
 275         memset(&parms.orvals, 0, sizeof(parms.orvals));
 276         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 277         parms.andvals[cr] = ~(1L << bit);
 278         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 279 }
 280 EXPORT_SYMBOL(smp_ctl_clear_bit);
 281
 282 #ifdef CONFIG_ZFCPDUMP
 283
 284 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 285 {
 286         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 287                 return;
 288         if (cpu >= NR_CPUS) {
 289                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 290                            "the dump\n", cpu, NR_CPUS - 1);
 291                 return;
 292         }
 293         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 294         while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
 295                 cpu_relax();
 296         memcpy_real(zfcpdump_save_areas[cpu],
 297                     (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 298                     sizeof(struct save_area));
 299 }
 300
 301 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 302 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 303
 304 #else
 305
 306 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 307
 308 #endif /* CONFIG_ZFCPDUMP */
 309
 310 static int cpu_known(int cpu_id)
 311 {
 312         int cpu;
 313
 314         for_each_present_cpu(cpu) {
 315                 if (__cpu_logical_map[cpu] == cpu_id)
 316                         return 1;
 317         }
 318         return 0;
 319 }
 320
 321 static int smp_rescan_cpus_sigp(cpumask_t avail)
 322 {
 323         int cpu_id, logical_cpu;
 324
 325         logical_cpu = cpumask_first(&avail);
 326         if (logical_cpu >= nr_cpu_ids)
 327                 return 0;
 328         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 329                 if (cpu_known(cpu_id))
 330                         continue;
 331                 __cpu_logical_map[logical_cpu] = cpu_id;
 332                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 333                 if (!cpu_stopped(logical_cpu))
 334                         continue;
 335                 cpu_set(logical_cpu, cpu_present_map);
 336                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 337                 logical_cpu = cpumask_next(logical_cpu, &avail);
 338                 if (logical_cpu >= nr_cpu_ids)
 339                         break;
 340         }
 341         return 0;
 342 }
 343
 344 static int smp_rescan_cpus_sclp(cpumask_t avail)
 345 {
 346         struct sclp_cpu_info *info;
 347         int cpu_id, logical_cpu, cpu;
 348         int rc;
 349
 350         logical_cpu = cpumask_first(&avail);
 351         if (logical_cpu >= nr_cpu_ids)
 352                 return 0;
 353         info = kmalloc(sizeof(*info), GFP_KERNEL);
 354         if (!info)
 355                 return -ENOMEM;
 356         rc = sclp_get_cpu_info(info);
 357         if (rc)
 358                 goto out;
 359         for (cpu = 0; cpu < info->combined; cpu++) {
 360                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 361                         continue;
 362                 cpu_id = info->cpu[cpu].address;
 363                 if (cpu_known(cpu_id))
 364                         continue;
 365                 __cpu_logical_map[logical_cpu] = cpu_id;
 366                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 367                 cpu_set(logical_cpu, cpu_present_map);
 368                 if (cpu >= info->configured)
 369                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 370                 else
 371                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 372                 logical_cpu = cpumask_next(logical_cpu, &avail);
 373                 if (logical_cpu >= nr_cpu_ids)
 374                         break;
 375         }
 376 out:
 377         kfree(info);
 378         return rc;
 379 }
 380
 381 static int __smp_rescan_cpus(void)
 382 {
 383         cpumask_t avail;
 384
 385         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 386         if (smp_use_sigp_detection)
 387                 return smp_rescan_cpus_sigp(avail);
 388         else
 389                 return smp_rescan_cpus_sclp(avail);
 390 }
 391
 392 static void __init smp_detect_cpus(void)
 393 {
 394         unsigned int cpu, c_cpus, s_cpus;
 395         struct sclp_cpu_info *info;
 396         u16 boot_cpu_addr, cpu_addr;
 397
 398         c_cpus = 1;
 399         s_cpus = 0;
 400         boot_cpu_addr = __cpu_logical_map[0];
 401         info = kmalloc(sizeof(*info), GFP_KERNEL);
 402         if (!info)
 403                 panic("smp_detect_cpus failed to allocate memory\n");
 404         /* Use sigp detection algorithm if sclp doesn't work. */
 405         if (sclp_get_cpu_info(info)) {
 406                 smp_use_sigp_detection = 1;
 407                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 408                         if (cpu == boot_cpu_addr)
 409                                 continue;
 410                         if (!raw_cpu_stopped(cpu))
 411                                 continue;
 412                         smp_get_save_area(c_cpus, cpu);
 413                         c_cpus++;
 414                 }
 415                 goto out;
 416         }
 417
 418         if (info->has_cpu_type) {
 419                 for (cpu = 0; cpu < info->combined; cpu++) {
 420                         if (info->cpu[cpu].address == boot_cpu_addr) {
 421                                 smp_cpu_type = info->cpu[cpu].type;
 422                                 break;
 423                         }
 424                 }
 425         }
 426
 427         for (cpu = 0; cpu < info->combined; cpu++) {
 428                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 429                         continue;
 430                 cpu_addr = info->cpu[cpu].address;
 431                 if (cpu_addr == boot_cpu_addr)
 432                         continue;
 433                 if (!raw_cpu_stopped(cpu_addr)) {
 434                         s_cpus++;
 435                         continue;
 436                 }
 437                 smp_get_save_area(c_cpus, cpu_addr);
 438                 c_cpus++;
 439         }
 440 out:
 441         kfree(info);
 442         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 443         get_online_cpus();
 444         __smp_rescan_cpus();
 445         put_online_cpus();
 446 }
 447
 448 /*
 449  *      Activate a secondary processor.
 450  */
 451 int __cpuinit start_secondary(void *cpuvoid)
 452 {
 453         /* Setup the cpu */
 454         cpu_init();
 455         preempt_disable();
 456         /* Enable TOD clock interrupts on the secondary cpu. */
 457         init_cpu_timer();
 458         /* Enable cpu timer interrupts on the secondary cpu. */
 459         init_cpu_vtimer();
 460         /* Enable pfault pseudo page faults on this cpu. */
 461         pfault_init();
 462
 463         /* call cpu notifiers */
 464         notify_cpu_starting(smp_processor_id());
 465         /* Mark this cpu as online */
 466         ipi_call_lock();
 467         cpu_set(smp_processor_id(), cpu_online_map);
 468         ipi_call_unlock();
 469         /* Switch on interrupts */
 470         local_irq_enable();
 471         /* Print info about this processor */
 472         print_cpu_info();
 473         /* cpu_idle will call schedule for us */
 474         cpu_idle();
 475         return 0;
 476 }
 477
 478 static void __init smp_create_idle(unsigned int cpu)
 479 {
 480         struct task_struct *p;
 481
 482         /*
 483          *  don't care about the psw and regs settings since we'll never
 484          *  reschedule the forked task.
 485          */
 486         p = fork_idle(cpu);
 487         if (IS_ERR(p))
 488                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 489         current_set[cpu] = p;
 490 }
 491
 492 static int __cpuinit smp_alloc_lowcore(int cpu)
 493 {
 494         unsigned long async_stack, panic_stack;
 495         struct _lowcore *lowcore;
 496
 497         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 498         if (!lowcore)
 499                 return -ENOMEM;
 500         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 501         panic_stack = __get_free_page(GFP_KERNEL);
 502         if (!panic_stack || !async_stack)
 503                 goto out;
 504         memcpy(lowcore, &S390_lowcore, 512);
 505         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 506         lowcore->async_stack = async_stack + ASYNC_SIZE;
 507         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 508
 509 #ifndef CONFIG_64BIT
 510         if (MACHINE_HAS_IEEE) {
 511                 unsigned long save_area;
 512
 513                 save_area = get_zeroed_page(GFP_KERNEL);
 514                 if (!save_area)
 515                         goto out;
 516                 lowcore->extended_save_area_addr = (u32) save_area;
 517         }
 518 #else
 519         if (vdso_alloc_per_cpu(cpu, lowcore))
 520                 goto out;
 521 #endif
 522         lowcore_ptr[cpu] = lowcore;
 523         return 0;
 524
 525 out:
 526         free_page(panic_stack);
 527         free_pages(async_stack, ASYNC_ORDER);
 528         free_pages((unsigned long) lowcore, LC_ORDER);
 529         return -ENOMEM;
 530 }
 531
 532 static void smp_free_lowcore(int cpu)
 533 {
 534         struct _lowcore *lowcore;
 535
 536         lowcore = lowcore_ptr[cpu];
 537 #ifndef CONFIG_64BIT
 538         if (MACHINE_HAS_IEEE)
 539                 free_page((unsigned long) lowcore->extended_save_area_addr);
 540 #else
 541         vdso_free_per_cpu(cpu, lowcore);
 542 #endif
 543         free_page(lowcore->panic_stack - PAGE_SIZE);
 544         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 545         free_pages((unsigned long) lowcore, LC_ORDER);
 546         lowcore_ptr[cpu] = NULL;
 547 }
 548
 549 /* Upping and downing of CPUs */
 550 int __cpuinit __cpu_up(unsigned int cpu)
 551 {
 552         struct _lowcore *cpu_lowcore;
 553         struct task_struct *idle;
 554         struct stack_frame *sf;
 555         u32 lowcore;
 556         int ccode;
 557
 558         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 559                 return -EIO;
 560         if (smp_alloc_lowcore(cpu))
 561                 return -ENOMEM;
 562         do {
 563                 ccode = sigp(cpu, sigp_initial_cpu_reset);
 564                 if (ccode == sigp_busy)
 565                         udelay(10);
 566                 if (ccode == sigp_not_operational)
 567                         goto err_out;
 568         } while (ccode == sigp_busy);
 569
 570         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 571         while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 572                 udelay(10);
 573
 574         idle = current_set[cpu];
 575         cpu_lowcore = lowcore_ptr[cpu];
 576         cpu_lowcore->kernel_stack = (unsigned long)
 577                 task_stack_page(idle) + THREAD_SIZE;
 578         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 579         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 580                                      - sizeof(struct pt_regs)
 581                                      - sizeof(struct stack_frame));
 582         memset(sf, 0, sizeof(struct stack_frame));
 583         sf->gprs[9] = (unsigned long) sf;
 584         cpu_lowcore->save_area[15] = (unsigned long) sf;
 585         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 586         atomic_inc(&init_mm.context.attach_count);
 587         asm volatile(
 588                 "       stam    0,15,0(%0)"
 589                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 590         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 591         cpu_lowcore->current_task = (unsigned long) idle;
 592         cpu_lowcore->cpu_nr = cpu;
 593         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 594         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 595         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 596         eieio();
 597
 598         while (sigp(cpu, sigp_restart) == sigp_busy)
 599                 udelay(10);
 600
 601         while (!cpu_online(cpu))
 602                 cpu_relax();
 603         return 0;
 604
 605 err_out:
 606         smp_free_lowcore(cpu);
 607         return -EIO;
 608 }
 609
 610 static int __init setup_possible_cpus(char *s)
 611 {
 612         int pcpus, cpu;
 613
 614         pcpus = simple_strtoul(s, NULL, 0);
 615         init_cpu_possible(cpumask_of(0));
 616         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 617                 set_cpu_possible(cpu, true);
 618         return 0;
 619 }
 620 early_param("possible_cpus", setup_possible_cpus);
 621
 622 #ifdef CONFIG_HOTPLUG_CPU
 623
 624 int __cpu_disable(void)
 625 {
 626         struct ec_creg_mask_parms cr_parms;
 627         int cpu = smp_processor_id();
 628
 629         cpu_clear(cpu, cpu_online_map);
 630
 631         /* Disable pfault pseudo page faults on this cpu. */
 632         pfault_fini();
 633
 634         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 635         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 636
 637         /* disable all external interrupts */
 638         cr_parms.orvals[0] = 0;
 639         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 640                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 641         /* disable all I/O interrupts */
 642         cr_parms.orvals[6] = 0;
 643         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 644                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 645         /* disable most machine checks */
 646         cr_parms.orvals[14] = 0;
 647         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 648                                  1 << 25 | 1 << 24);
 649
 650         smp_ctl_bit_callback(&cr_parms);
 651
 652         return 0;
 653 }
 654
 655 void __cpu_die(unsigned int cpu)
 656 {
 657         /* Wait until target cpu is down */
 658         while (!cpu_stopped(cpu))
 659                 cpu_relax();
 660         while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
 661                 udelay(10);
 662         smp_free_lowcore(cpu);
 663         atomic_dec(&init_mm.context.attach_count);
 664         pr_info("Processor %d stopped\n", cpu);
 665 }
 666
 667 void cpu_die(void)
 668 {
 669         idle_task_exit();
 670         while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
 671                 cpu_relax();
 672         for (;;);
 673 }
 674
 675 #endif /* CONFIG_HOTPLUG_CPU */
 676
 677 void __init smp_prepare_cpus(unsigned int max_cpus)
 678 {
 679 #ifndef CONFIG_64BIT
 680         unsigned long save_area = 0;
 681 #endif
 682         unsigned long async_stack, panic_stack;
 683         struct _lowcore *lowcore;
 684         unsigned int cpu;
 685
 686         smp_detect_cpus();
 687
 688         /* request the 0x1201 emergency signal external interrupt */
 689         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 690                 panic("Couldn't request external interrupt 0x1201");
 691         print_cpu_info();
 692
 693         /* Reallocate current lowcore, but keep its contents. */
 694         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 695         panic_stack = __get_free_page(GFP_KERNEL);
 696         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 697         BUG_ON(!lowcore || !panic_stack || !async_stack);
 698 #ifndef CONFIG_64BIT
 699         if (MACHINE_HAS_IEEE)
 700                 save_area = get_zeroed_page(GFP_KERNEL);
 701 #endif
 702         local_irq_disable();
 703         local_mcck_disable();
 704         lowcore_ptr[smp_processor_id()] = lowcore;
 705         *lowcore = S390_lowcore;
 706         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 707         lowcore->async_stack = async_stack + ASYNC_SIZE;
 708 #ifndef CONFIG_64BIT
 709         if (MACHINE_HAS_IEEE)
 710                 lowcore->extended_save_area_addr = (u32) save_area;
 711 #endif
 712         set_prefix((u32)(unsigned long) lowcore);
 713         local_mcck_enable();
 714         local_irq_enable();
 715 #ifdef CONFIG_64BIT
 716         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 717                 BUG();
 718 #endif
 719         for_each_possible_cpu(cpu)
 720                 if (cpu != smp_processor_id())
 721                         smp_create_idle(cpu);
 722 }
 723
 724 void __init smp_prepare_boot_cpu(void)
 725 {
 726         BUG_ON(smp_processor_id() != 0);
 727
 728         current_thread_info()->cpu = 0;
 729         cpu_set(0, cpu_present_map);
 730         cpu_set(0, cpu_online_map);
 731         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 732         current_set[0] = current;
 733         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 734         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 735 }
 736
 737 void __init smp_cpus_done(unsigned int max_cpus)
 738 {
 739 }
 740
 741 void __init smp_setup_processor_id(void)
 742 {
 743         S390_lowcore.cpu_nr = 0;
 744         __cpu_logical_map[0] = stap();
 745 }
 746
 747 /*
 748  * the frequency of the profiling timer can be changed
 749  * by writing a multiplier value into /proc/profile.
 750  *
 751  * usually you want to run this on all CPUs ;)
 752  */
 753 int setup_profiling_timer(unsigned int multiplier)
 754 {
 755         return 0;
 756 }
 757
 758 #ifdef CONFIG_HOTPLUG_CPU
 759 static ssize_t cpu_configure_show(struct sys_device *dev,
 760                                 struct sysdev_attribute *attr, char *buf)
 761 {
 762         ssize_t count;
 763
 764         mutex_lock(&smp_cpu_state_mutex);
 765         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 766         mutex_unlock(&smp_cpu_state_mutex);
 767         return count;
 768 }
 769
 770 static ssize_t cpu_configure_store(struct sys_device *dev,
 771                                   struct sysdev_attribute *attr,
 772                                   const char *buf, size_t count)
 773 {
 774         int cpu = dev->id;
 775         int val, rc;
 776         char delim;
 777
 778         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 779                 return -EINVAL;
 780         if (val != 0 && val != 1)
 781                 return -EINVAL;
 782
 783         get_online_cpus();
 784         mutex_lock(&smp_cpu_state_mutex);
 785         rc = -EBUSY;
 786         /* disallow configuration changes of online cpus and cpu 0 */
 787         if (cpu_online(cpu) || cpu == 0)
 788                 goto out;
 789         rc = 0;
 790         switch (val) {
 791         case 0:
 792                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 793                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 794                         if (!rc) {
 795                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 796                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 797                         }
 798                 }
 799                 break;
 800         case 1:
 801                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 802                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 803                         if (!rc) {
 804                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 805                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 806                         }
 807                 }
 808                 break;
 809         default:
 810                 break;
 811         }
 812 out:
 813         mutex_unlock(&smp_cpu_state_mutex);
 814         put_online_cpus();
 815         return rc ? rc : count;
 816 }
 817 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 818 #endif /* CONFIG_HOTPLUG_CPU */
 819
 820 static ssize_t cpu_polarization_show(struct sys_device *dev,
 821                                      struct sysdev_attribute *attr, char *buf)
 822 {
 823         int cpu = dev->id;
 824         ssize_t count;
 825
 826         mutex_lock(&smp_cpu_state_mutex);
 827         switch (smp_cpu_polarization[cpu]) {
 828         case POLARIZATION_HRZ:
 829                 count = sprintf(buf, "horizontal\n");
 830                 break;
 831         case POLARIZATION_VL:
 832                 count = sprintf(buf, "vertical:low\n");
 833                 break;
 834         case POLARIZATION_VM:
 835                 count = sprintf(buf, "vertical:medium\n");
 836                 break;
 837         case POLARIZATION_VH:
 838                 count = sprintf(buf, "vertical:high\n");
 839                 break;
 840         default:
 841                 count = sprintf(buf, "unknown\n");
 842                 break;
 843         }
 844         mutex_unlock(&smp_cpu_state_mutex);
 845         return count;
 846 }
 847 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 848
 849 static ssize_t show_cpu_address(struct sys_device *dev,
 850                                 struct sysdev_attribute *attr, char *buf)
 851 {
 852         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 853 }
 854 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 855
 856
 857 static struct attribute *cpu_common_attrs[] = {
 858 #ifdef CONFIG_HOTPLUG_CPU
 859         &attr_configure.attr,
 860 #endif
 861         &attr_address.attr,
 862         &attr_polarization.attr,
 863         NULL,
 864 };
 865
 866 static struct attribute_group cpu_common_attr_group = {
 867         .attrs = cpu_common_attrs,
 868 };
 869
 870 static ssize_t show_capability(struct sys_device *dev,
 871                                 struct sysdev_attribute *attr, char *buf)
 872 {
 873         unsigned int capability;
 874         int rc;
 875
 876         rc = get_cpu_capability(&capability);
 877         if (rc)
 878                 return rc;
 879         return sprintf(buf, "%u\n", capability);
 880 }
 881 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 882
 883 static ssize_t show_idle_count(struct sys_device *dev,
 884                                 struct sysdev_attribute *attr, char *buf)
 885 {
 886         struct s390_idle_data *idle;
 887         unsigned long long idle_count;
 888         unsigned int sequence;
 889
 890         idle = &per_cpu(s390_idle, dev->id);
 891 repeat:
 892         sequence = idle->sequence;
 893         smp_rmb();
 894         if (sequence & 1)
 895                 goto repeat;
 896         idle_count = idle->idle_count;
 897         if (idle->idle_enter)
 898                 idle_count++;
 899         smp_rmb();
 900         if (idle->sequence != sequence)
 901                 goto repeat;
 902         return sprintf(buf, "%llu\n", idle_count);
 903 }
 904 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 905
 906 static ssize_t show_idle_time(struct sys_device *dev,
 907                                 struct sysdev_attribute *attr, char *buf)
 908 {
 909         struct s390_idle_data *idle;
 910         unsigned long long now, idle_time, idle_enter;
 911         unsigned int sequence;
 912
 913         idle = &per_cpu(s390_idle, dev->id);
 914         now = get_clock();
 915 repeat:
 916         sequence = idle->sequence;
 917         smp_rmb();
 918         if (sequence & 1)
 919                 goto repeat;
 920         idle_time = idle->idle_time;
 921         idle_enter = idle->idle_enter;
 922         if (idle_enter != 0ULL && idle_enter < now)
 923                 idle_time += now - idle_enter;
 924         smp_rmb();
 925         if (idle->sequence != sequence)
 926                 goto repeat;
 927         return sprintf(buf, "%llu\n", idle_time >> 12);
 928 }
 929 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 930
 931 static struct attribute *cpu_online_attrs[] = {
 932         &attr_capability.attr,
 933         &attr_idle_count.attr,
 934         &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 __cpuinit smp_cpu_notify(struct notifier_block *self,
 943                                     unsigned long action, void *hcpu)
 944 {
 945         unsigned int cpu = (unsigned int)(long)hcpu;
 946         struct cpu *c = &per_cpu(cpu_devices, cpu);
 947         struct sys_device *s = &c->sysdev;
 948         struct s390_idle_data *idle;
 949         int err = 0;
 950
 951         switch (action) {
 952         case CPU_ONLINE:
 953         case CPU_ONLINE_FROZEN:
 954                 idle = &per_cpu(s390_idle, cpu);
 955                 memset(idle, 0, sizeof(struct s390_idle_data));
 956                 err = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 957                 break;
 958         case CPU_DEAD:
 959         case CPU_DEAD_FROZEN:
 960                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 961                 break;
 962         }
 963         return notifier_from_errno(err);
 964 }
 965
 966 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 967         .notifier_call = smp_cpu_notify,
 968 };
 969
 970 static int __devinit smp_add_present_cpu(int cpu)
 971 {
 972         struct cpu *c = &per_cpu(cpu_devices, cpu);
 973         struct sys_device *s = &c->sysdev;
 974         int rc;
 975
 976         c->hotpluggable = 1;
 977         rc = register_cpu(c, cpu);
 978         if (rc)
 979                 goto out;
 980         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 981         if (rc)
 982                 goto out_cpu;
 983         if (!cpu_online(cpu))
 984                 goto out;
 985         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 986         if (!rc)
 987                 return 0;
 988         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 989 out_cpu:
 990 #ifdef CONFIG_HOTPLUG_CPU
 991         unregister_cpu(c);
 992 #endif
 993 out:
 994         return rc;
 995 }
 996
 997 #ifdef CONFIG_HOTPLUG_CPU
 998
 999 int __ref smp_rescan_cpus(void)
1000 {
1001         cpumask_t newcpus;
1002         int cpu;
1003         int rc;
1004
1005         get_online_cpus();
1006         mutex_lock(&smp_cpu_state_mutex);
1007         newcpus = cpu_present_map;
1008         rc = __smp_rescan_cpus();
1009         if (rc)
1010                 goto out;
1011         cpus_andnot(newcpus, cpu_present_map, newcpus);
1012         for_each_cpu_mask(cpu, newcpus) {
1013                 rc = smp_add_present_cpu(cpu);
1014                 if (rc)
1015                         cpu_clear(cpu, cpu_present_map);
1016         }
1017         rc = 0;
1018 out:
1019         mutex_unlock(&smp_cpu_state_mutex);
1020         put_online_cpus();
1021         if (!cpus_empty(newcpus))
1022                 topology_schedule_update();
1023         return rc;
1024 }
1025
1026 static ssize_t __ref rescan_store(struct sysdev_class *class,
1027                                   struct sysdev_class_attribute *attr,
1028                                   const char *buf,
1029                                   size_t count)
1030 {
1031         int rc;
1032
1033         rc = smp_rescan_cpus();
1034         return rc ? rc : count;
1035 }
1036 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1037 #endif /* CONFIG_HOTPLUG_CPU */
1038
1039 static ssize_t dispatching_show(struct sysdev_class *class,
1040                                 struct sysdev_class_attribute *attr,
1041                                 char *buf)
1042 {
1043         ssize_t count;
1044
1045         mutex_lock(&smp_cpu_state_mutex);
1046         count = sprintf(buf, "%d\n", cpu_management);
1047         mutex_unlock(&smp_cpu_state_mutex);
1048         return count;
1049 }
1050
1051 static ssize_t dispatching_store(struct sysdev_class *dev,
1052                                  struct sysdev_class_attribute *attr,
1053                                  const char *buf,
1054                                  size_t count)
1055 {
1056         int val, rc;
1057         char delim;
1058
1059         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1060                 return -EINVAL;
1061         if (val != 0 && val != 1)
1062                 return -EINVAL;
1063         rc = 0;
1064         get_online_cpus();
1065         mutex_lock(&smp_cpu_state_mutex);
1066         if (cpu_management == val)
1067                 goto out;
1068         rc = topology_set_cpu_management(val);
1069         if (!rc)
1070                 cpu_management = val;
1071 out:
1072         mutex_unlock(&smp_cpu_state_mutex);
1073         put_online_cpus();
1074         return rc ? rc : count;
1075 }
1076 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1077                          dispatching_store);
1078
1079 static int __init topology_init(void)
1080 {
1081         int cpu;
1082         int rc;
1083
1084         register_cpu_notifier(&smp_cpu_nb);
1085
1086 #ifdef CONFIG_HOTPLUG_CPU
1087         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1088         if (rc)
1089                 return rc;
1090 #endif
1091         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1092         if (rc)
1093                 return rc;
1094         for_each_present_cpu(cpu) {
1095                 rc = smp_add_present_cpu(cpu);
1096                 if (rc)
1097                         return rc;
1098         }
1099         return 0;
1100 }
1101 subsys_initcall(topology_init);