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 <asm/ipl.h>
  40 #include <asm/setup.h>
  41 #include <asm/sigp.h>
  42 #include <asm/pgalloc.h>
  43 #include <asm/irq.h>
  44 #include <asm/s390_ext.h>
  45 #include <asm/cpcmd.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/timer.h>
  48 #include <asm/lowcore.h>
  49 #include <asm/sclp.h>
  50 #include <asm/cputime.h>
  51 #include <asm/vdso.h>
  52 #include <asm/cpu.h>
  53 #include "entry.h"
  54
  55 /* logical cpu to cpu address */
  56 unsigned short __cpu_logical_map[NR_CPUS];
  57
  58 static struct task_struct *current_set[NR_CPUS];
  59
  60 static u8 smp_cpu_type;
  61 static int smp_use_sigp_detection;
  62
  63 enum s390_cpu_state {
  64         CPU_STATE_STANDBY,
  65         CPU_STATE_CONFIGURED,
  66 };
  67
  68 DEFINE_MUTEX(smp_cpu_state_mutex);
  69 int smp_cpu_polarization[NR_CPUS];
  70 static int smp_cpu_state[NR_CPUS];
  71 static int cpu_management;
  72
  73 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  74
  75 static void smp_ext_bitcall(int, int);
  76
  77 static int raw_cpu_stopped(int cpu)
  78 {
  79         u32 status;
  80
  81         switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
  82         case sigp_status_stored:
  83                 /* Check for stopped and check stop state */
  84                 if (status & 0x50)
  85                         return 1;
  86                 break;
  87         default:
  88                 break;
  89         }
  90         return 0;
  91 }
  92
  93 static inline int cpu_stopped(int cpu)
  94 {
  95         return raw_cpu_stopped(cpu_logical_map(cpu));
  96 }
  97
  98 void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
  99 {
 100         struct _lowcore *lc, *current_lc;
 101         struct stack_frame *sf;
 102         struct pt_regs *regs;
 103         unsigned long sp;
 104
 105         if (smp_processor_id() == 0)
 106                 func(data);
 107         __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
 108         /* Disable lowcore protection */
 109         __ctl_clear_bit(0, 28);
 110         current_lc = lowcore_ptr[smp_processor_id()];
 111         lc = lowcore_ptr[0];
 112         if (!lc)
 113                 lc = current_lc;
 114         lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
 115         lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
 116         if (!cpu_online(0))
 117                 smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
 118         while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
 119                 cpu_relax();
 120         sp = lc->panic_stack;
 121         sp -= sizeof(struct pt_regs);
 122         regs = (struct pt_regs *) sp;
 123         memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
 124         memcpy(&regs->psw, &current_lc->st_status_fixed_logout, sizeof(psw_t));
 125         sp -= STACK_FRAME_OVERHEAD;
 126         sf = (struct stack_frame *) sp;
 127         sf->back_chain = regs->gprs[15];
 128         smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 129 }
 130
 131 void smp_send_stop(void)
 132 {
 133         int cpu, rc;
 134
 135         /* Disable all interrupts/machine checks */
 136         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
 137         trace_hardirqs_off();
 138
 139         /* stop all processors */
 140         for_each_online_cpu(cpu) {
 141                 if (cpu == smp_processor_id())
 142                         continue;
 143                 do {
 144                         rc = sigp(cpu, sigp_stop);
 145                 } while (rc == sigp_busy);
 146
 147                 while (!cpu_stopped(cpu))
 148                         cpu_relax();
 149         }
 150 }
 151
 152 /*
 153  * This is the main routine where commands issued by other
 154  * cpus are handled.
 155  */
 156
 157 static void do_ext_call_interrupt(__u16 code)
 158 {
 159         unsigned long bits;
 160
 161         /*
 162          * handle bit signal external calls
 163          *
 164          * For the ec_schedule signal we have to do nothing. All the work
 165          * is done automatically when we return from the interrupt.
 166          */
 167         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 168
 169         if (test_bit(ec_call_function, &bits))
 170                 generic_smp_call_function_interrupt();
 171
 172         if (test_bit(ec_call_function_single, &bits))
 173                 generic_smp_call_function_single_interrupt();
 174 }
 175
 176 /*
 177  * Send an external call sigp to another cpu and return without waiting
 178  * for its completion.
 179  */
 180 static void smp_ext_bitcall(int cpu, int sig)
 181 {
 182         /*
 183          * Set signaling bit in lowcore of target cpu and kick it
 184          */
 185         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 186         while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
 187                 udelay(10);
 188 }
 189
 190 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 191 {
 192         int cpu;
 193
 194         for_each_cpu(cpu, mask)
 195                 smp_ext_bitcall(cpu, ec_call_function);
 196 }
 197
 198 void arch_send_call_function_single_ipi(int cpu)
 199 {
 200         smp_ext_bitcall(cpu, ec_call_function_single);
 201 }
 202
 203 #ifndef CONFIG_64BIT
 204 /*
 205  * this function sends a 'purge tlb' signal to another CPU.
 206  */
 207 static void smp_ptlb_callback(void *info)
 208 {
 209         __tlb_flush_local();
 210 }
 211
 212 void smp_ptlb_all(void)
 213 {
 214         on_each_cpu(smp_ptlb_callback, NULL, 1);
 215 }
 216 EXPORT_SYMBOL(smp_ptlb_all);
 217 #endif /* ! CONFIG_64BIT */
 218
 219 /*
 220  * this function sends a 'reschedule' IPI to another CPU.
 221  * it goes straight through and wastes no time serializing
 222  * anything. Worst case is that we lose a reschedule ...
 223  */
 224 void smp_send_reschedule(int cpu)
 225 {
 226         smp_ext_bitcall(cpu, ec_schedule);
 227 }
 228
 229 /*
 230  * parameter area for the set/clear control bit callbacks
 231  */
 232 struct ec_creg_mask_parms {
 233         unsigned long orvals[16];
 234         unsigned long andvals[16];
 235 };
 236
 237 /*
 238  * callback for setting/clearing control bits
 239  */
 240 static void smp_ctl_bit_callback(void *info)
 241 {
 242         struct ec_creg_mask_parms *pp = info;
 243         unsigned long cregs[16];
 244         int i;
 245
 246         __ctl_store(cregs, 0, 15);
 247         for (i = 0; i <= 15; i++)
 248                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 249         __ctl_load(cregs, 0, 15);
 250 }
 251
 252 /*
 253  * Set a bit in a control register of all cpus
 254  */
 255 void smp_ctl_set_bit(int cr, int bit)
 256 {
 257         struct ec_creg_mask_parms parms;
 258
 259         memset(&parms.orvals, 0, sizeof(parms.orvals));
 260         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 261         parms.orvals[cr] = 1 << bit;
 262         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 263 }
 264 EXPORT_SYMBOL(smp_ctl_set_bit);
 265
 266 /*
 267  * Clear a bit in a control register of all cpus
 268  */
 269 void smp_ctl_clear_bit(int cr, int bit)
 270 {
 271         struct ec_creg_mask_parms parms;
 272
 273         memset(&parms.orvals, 0, sizeof(parms.orvals));
 274         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 275         parms.andvals[cr] = ~(1L << bit);
 276         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 277 }
 278 EXPORT_SYMBOL(smp_ctl_clear_bit);
 279
 280 #ifdef CONFIG_ZFCPDUMP
 281
 282 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 283 {
 284         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 285                 return;
 286         if (cpu >= NR_CPUS) {
 287                 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
 288                            "the dump\n", cpu, NR_CPUS - 1);
 289                 return;
 290         }
 291         zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
 292         while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
 293                 cpu_relax();
 294         memcpy(zfcpdump_save_areas[cpu],
 295                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 296                sizeof(struct save_area));
 297 }
 298
 299 struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
 300 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 301
 302 #else
 303
 304 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 305
 306 #endif /* CONFIG_ZFCPDUMP */
 307
 308 static int cpu_known(int cpu_id)
 309 {
 310         int cpu;
 311
 312         for_each_present_cpu(cpu) {
 313                 if (__cpu_logical_map[cpu] == cpu_id)
 314                         return 1;
 315         }
 316         return 0;
 317 }
 318
 319 static int smp_rescan_cpus_sigp(cpumask_t avail)
 320 {
 321         int cpu_id, logical_cpu;
 322
 323         logical_cpu = cpumask_first(&avail);
 324         if (logical_cpu >= nr_cpu_ids)
 325                 return 0;
 326         for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
 327                 if (cpu_known(cpu_id))
 328                         continue;
 329                 __cpu_logical_map[logical_cpu] = cpu_id;
 330                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 331                 if (!cpu_stopped(logical_cpu))
 332                         continue;
 333                 cpu_set(logical_cpu, cpu_present_map);
 334                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 335                 logical_cpu = cpumask_next(logical_cpu, &avail);
 336                 if (logical_cpu >= nr_cpu_ids)
 337                         break;
 338         }
 339         return 0;
 340 }
 341
 342 static int smp_rescan_cpus_sclp(cpumask_t avail)
 343 {
 344         struct sclp_cpu_info *info;
 345         int cpu_id, logical_cpu, cpu;
 346         int rc;
 347
 348         logical_cpu = cpumask_first(&avail);
 349         if (logical_cpu >= nr_cpu_ids)
 350                 return 0;
 351         info = kmalloc(sizeof(*info), GFP_KERNEL);
 352         if (!info)
 353                 return -ENOMEM;
 354         rc = sclp_get_cpu_info(info);
 355         if (rc)
 356                 goto out;
 357         for (cpu = 0; cpu < info->combined; cpu++) {
 358                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 359                         continue;
 360                 cpu_id = info->cpu[cpu].address;
 361                 if (cpu_known(cpu_id))
 362                         continue;
 363                 __cpu_logical_map[logical_cpu] = cpu_id;
 364                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 365                 cpu_set(logical_cpu, cpu_present_map);
 366                 if (cpu >= info->configured)
 367                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 368                 else
 369                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 370                 logical_cpu = cpumask_next(logical_cpu, &avail);
 371                 if (logical_cpu >= nr_cpu_ids)
 372                         break;
 373         }
 374 out:
 375         kfree(info);
 376         return rc;
 377 }
 378
 379 static int __smp_rescan_cpus(void)
 380 {
 381         cpumask_t avail;
 382
 383         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 384         if (smp_use_sigp_detection)
 385                 return smp_rescan_cpus_sigp(avail);
 386         else
 387                 return smp_rescan_cpus_sclp(avail);
 388 }
 389
 390 static void __init smp_detect_cpus(void)
 391 {
 392         unsigned int cpu, c_cpus, s_cpus;
 393         struct sclp_cpu_info *info;
 394         u16 boot_cpu_addr, cpu_addr;
 395
 396         c_cpus = 1;
 397         s_cpus = 0;
 398         boot_cpu_addr = __cpu_logical_map[0];
 399         info = kmalloc(sizeof(*info), GFP_KERNEL);
 400         if (!info)
 401                 panic("smp_detect_cpus failed to allocate memory\n");
 402         /* Use sigp detection algorithm if sclp doesn't work. */
 403         if (sclp_get_cpu_info(info)) {
 404                 smp_use_sigp_detection = 1;
 405                 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
 406                         if (cpu == boot_cpu_addr)
 407                                 continue;
 408                         if (!raw_cpu_stopped(cpu))
 409                                 continue;
 410                         smp_get_save_area(c_cpus, cpu);
 411                         c_cpus++;
 412                 }
 413                 goto out;
 414         }
 415
 416         if (info->has_cpu_type) {
 417                 for (cpu = 0; cpu < info->combined; cpu++) {
 418                         if (info->cpu[cpu].address == boot_cpu_addr) {
 419                                 smp_cpu_type = info->cpu[cpu].type;
 420                                 break;
 421                         }
 422                 }
 423         }
 424
 425         for (cpu = 0; cpu < info->combined; cpu++) {
 426                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 427                         continue;
 428                 cpu_addr = info->cpu[cpu].address;
 429                 if (cpu_addr == boot_cpu_addr)
 430                         continue;
 431                 if (!raw_cpu_stopped(cpu_addr)) {
 432                         s_cpus++;
 433                         continue;
 434                 }
 435                 smp_get_save_area(c_cpus, cpu_addr);
 436                 c_cpus++;
 437         }
 438 out:
 439         kfree(info);
 440         pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
 441         get_online_cpus();
 442         __smp_rescan_cpus();
 443         put_online_cpus();
 444 }
 445
 446 /*
 447  *      Activate a secondary processor.
 448  */
 449 int __cpuinit start_secondary(void *cpuvoid)
 450 {
 451         /* Setup the cpu */
 452         cpu_init();
 453         preempt_disable();
 454         /* Enable TOD clock interrupts on the secondary cpu. */
 455         init_cpu_timer();
 456         /* Enable cpu timer interrupts on the secondary cpu. */
 457         init_cpu_vtimer();
 458         /* Enable pfault pseudo page faults on this cpu. */
 459         pfault_init();
 460
 461         /* call cpu notifiers */
 462         notify_cpu_starting(smp_processor_id());
 463         /* Mark this cpu as online */
 464         ipi_call_lock();
 465         cpu_set(smp_processor_id(), cpu_online_map);
 466         ipi_call_unlock();
 467         /* Switch on interrupts */
 468         local_irq_enable();
 469         /* Print info about this processor */
 470         print_cpu_info();
 471         /* cpu_idle will call schedule for us */
 472         cpu_idle();
 473         return 0;
 474 }
 475
 476 static void __init smp_create_idle(unsigned int cpu)
 477 {
 478         struct task_struct *p;
 479
 480         /*
 481          *  don't care about the psw and regs settings since we'll never
 482          *  reschedule the forked task.
 483          */
 484         p = fork_idle(cpu);
 485         if (IS_ERR(p))
 486                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 487         current_set[cpu] = p;
 488 }
 489
 490 static int __cpuinit smp_alloc_lowcore(int cpu)
 491 {
 492         unsigned long async_stack, panic_stack;
 493         struct _lowcore *lowcore;
 494
 495         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 496         if (!lowcore)
 497                 return -ENOMEM;
 498         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 499         panic_stack = __get_free_page(GFP_KERNEL);
 500         if (!panic_stack || !async_stack)
 501                 goto out;
 502         memcpy(lowcore, &S390_lowcore, 512);
 503         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 504         lowcore->async_stack = async_stack + ASYNC_SIZE;
 505         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 506
 507 #ifndef CONFIG_64BIT
 508         if (MACHINE_HAS_IEEE) {
 509                 unsigned long save_area;
 510
 511                 save_area = get_zeroed_page(GFP_KERNEL);
 512                 if (!save_area)
 513                         goto out;
 514                 lowcore->extended_save_area_addr = (u32) save_area;
 515         }
 516 #else
 517         if (vdso_alloc_per_cpu(cpu, lowcore))
 518                 goto out;
 519 #endif
 520         lowcore_ptr[cpu] = lowcore;
 521         return 0;
 522
 523 out:
 524         free_page(panic_stack);
 525         free_pages(async_stack, ASYNC_ORDER);
 526         free_pages((unsigned long) lowcore, LC_ORDER);
 527         return -ENOMEM;
 528 }
 529
 530 static void smp_free_lowcore(int cpu)
 531 {
 532         struct _lowcore *lowcore;
 533
 534         lowcore = lowcore_ptr[cpu];
 535 #ifndef CONFIG_64BIT
 536         if (MACHINE_HAS_IEEE)
 537                 free_page((unsigned long) lowcore->extended_save_area_addr);
 538 #else
 539         vdso_free_per_cpu(cpu, lowcore);
 540 #endif
 541         free_page(lowcore->panic_stack - PAGE_SIZE);
 542         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 543         free_pages((unsigned long) lowcore, LC_ORDER);
 544         lowcore_ptr[cpu] = NULL;
 545 }
 546
 547 /* Upping and downing of CPUs */
 548 int __cpuinit __cpu_up(unsigned int cpu)
 549 {
 550         struct _lowcore *cpu_lowcore;
 551         struct task_struct *idle;
 552         struct stack_frame *sf;
 553         u32 lowcore;
 554         int ccode;
 555
 556         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 557                 return -EIO;
 558         if (smp_alloc_lowcore(cpu))
 559                 return -ENOMEM;
 560         do {
 561                 ccode = sigp(cpu, sigp_initial_cpu_reset);
 562                 if (ccode == sigp_busy)
 563                         udelay(10);
 564                 if (ccode == sigp_not_operational)
 565                         goto err_out;
 566         } while (ccode == sigp_busy);
 567
 568         lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
 569         while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
 570                 udelay(10);
 571
 572         idle = current_set[cpu];
 573         cpu_lowcore = lowcore_ptr[cpu];
 574         cpu_lowcore->kernel_stack = (unsigned long)
 575                 task_stack_page(idle) + THREAD_SIZE;
 576         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 577         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 578                                      - sizeof(struct pt_regs)
 579                                      - sizeof(struct stack_frame));
 580         memset(sf, 0, sizeof(struct stack_frame));
 581         sf->gprs[9] = (unsigned long) sf;
 582         cpu_lowcore->save_area[15] = (unsigned long) sf;
 583         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 584         asm volatile(
 585                 "       stam    0,15,0(%0)"
 586                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 587         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 588         cpu_lowcore->current_task = (unsigned long) idle;
 589         cpu_lowcore->cpu_nr = cpu;
 590         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 591         cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
 592         cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
 593         eieio();
 594
 595         while (sigp(cpu, sigp_restart) == sigp_busy)
 596                 udelay(10);
 597
 598         while (!cpu_online(cpu))
 599                 cpu_relax();
 600         return 0;
 601
 602 err_out:
 603         smp_free_lowcore(cpu);
 604         return -EIO;
 605 }
 606
 607 static int __init setup_possible_cpus(char *s)
 608 {
 609         int pcpus, cpu;
 610
 611         pcpus = simple_strtoul(s, NULL, 0);
 612         init_cpu_possible(cpumask_of(0));
 613         for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
 614                 set_cpu_possible(cpu, true);
 615         return 0;
 616 }
 617 early_param("possible_cpus", setup_possible_cpus);
 618
 619 #ifdef CONFIG_HOTPLUG_CPU
 620
 621 int __cpu_disable(void)
 622 {
 623         struct ec_creg_mask_parms cr_parms;
 624         int cpu = smp_processor_id();
 625
 626         cpu_clear(cpu, cpu_online_map);
 627
 628         /* Disable pfault pseudo page faults on this cpu. */
 629         pfault_fini();
 630
 631         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 632         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 633
 634         /* disable all external interrupts */
 635         cr_parms.orvals[0] = 0;
 636         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 637                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 638         /* disable all I/O interrupts */
 639         cr_parms.orvals[6] = 0;
 640         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 641                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 642         /* disable most machine checks */
 643         cr_parms.orvals[14] = 0;
 644         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 645                                  1 << 25 | 1 << 24);
 646
 647         smp_ctl_bit_callback(&cr_parms);
 648
 649         return 0;
 650 }
 651
 652 void __cpu_die(unsigned int cpu)
 653 {
 654         /* Wait until target cpu is down */
 655         while (!cpu_stopped(cpu))
 656                 cpu_relax();
 657         while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
 658                 udelay(10);
 659         smp_free_lowcore(cpu);
 660         pr_info("Processor %d stopped\n", cpu);
 661 }
 662
 663 void cpu_die(void)
 664 {
 665         idle_task_exit();
 666         while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
 667                 cpu_relax();
 668         for (;;);
 669 }
 670
 671 #endif /* CONFIG_HOTPLUG_CPU */
 672
 673 void __init smp_prepare_cpus(unsigned int max_cpus)
 674 {
 675 #ifndef CONFIG_64BIT
 676         unsigned long save_area = 0;
 677 #endif
 678         unsigned long async_stack, panic_stack;
 679         struct _lowcore *lowcore;
 680         unsigned int cpu;
 681
 682         smp_detect_cpus();
 683
 684         /* request the 0x1201 emergency signal external interrupt */
 685         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 686                 panic("Couldn't request external interrupt 0x1201");
 687         print_cpu_info();
 688
 689         /* Reallocate current lowcore, but keep its contents. */
 690         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
 691         panic_stack = __get_free_page(GFP_KERNEL);
 692         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 693         BUG_ON(!lowcore || !panic_stack || !async_stack);
 694 #ifndef CONFIG_64BIT
 695         if (MACHINE_HAS_IEEE)
 696                 save_area = get_zeroed_page(GFP_KERNEL);
 697 #endif
 698         local_irq_disable();
 699         local_mcck_disable();
 700         lowcore_ptr[smp_processor_id()] = lowcore;
 701         *lowcore = S390_lowcore;
 702         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 703         lowcore->async_stack = async_stack + ASYNC_SIZE;
 704 #ifndef CONFIG_64BIT
 705         if (MACHINE_HAS_IEEE)
 706                 lowcore->extended_save_area_addr = (u32) save_area;
 707 #endif
 708         set_prefix((u32)(unsigned long) lowcore);
 709         local_mcck_enable();
 710         local_irq_enable();
 711 #ifdef CONFIG_64BIT
 712         if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
 713                 BUG();
 714 #endif
 715         for_each_possible_cpu(cpu)
 716                 if (cpu != smp_processor_id())
 717                         smp_create_idle(cpu);
 718 }
 719
 720 void __init smp_prepare_boot_cpu(void)
 721 {
 722         BUG_ON(smp_processor_id() != 0);
 723
 724         current_thread_info()->cpu = 0;
 725         cpu_set(0, cpu_present_map);
 726         cpu_set(0, cpu_online_map);
 727         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 728         current_set[0] = current;
 729         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 730         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 731 }
 732
 733 void __init smp_cpus_done(unsigned int max_cpus)
 734 {
 735 }
 736
 737 void __init smp_setup_processor_id(void)
 738 {
 739         S390_lowcore.cpu_nr = 0;
 740         __cpu_logical_map[0] = stap();
 741 }
 742
 743 /*
 744  * the frequency of the profiling timer can be changed
 745  * by writing a multiplier value into /proc/profile.
 746  *
 747  * usually you want to run this on all CPUs ;)
 748  */
 749 int setup_profiling_timer(unsigned int multiplier)
 750 {
 751         return 0;
 752 }
 753
 754 #ifdef CONFIG_HOTPLUG_CPU
 755 static ssize_t cpu_configure_show(struct sys_device *dev,
 756                                 struct sysdev_attribute *attr, char *buf)
 757 {
 758         ssize_t count;
 759
 760         mutex_lock(&smp_cpu_state_mutex);
 761         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 762         mutex_unlock(&smp_cpu_state_mutex);
 763         return count;
 764 }
 765
 766 static ssize_t cpu_configure_store(struct sys_device *dev,
 767                                   struct sysdev_attribute *attr,
 768                                   const char *buf, size_t count)
 769 {
 770         int cpu = dev->id;
 771         int val, rc;
 772         char delim;
 773
 774         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 775                 return -EINVAL;
 776         if (val != 0 && val != 1)
 777                 return -EINVAL;
 778
 779         get_online_cpus();
 780         mutex_lock(&smp_cpu_state_mutex);
 781         rc = -EBUSY;
 782         /* disallow configuration changes of online cpus and cpu 0 */
 783         if (cpu_online(cpu) || cpu == 0)
 784                 goto out;
 785         rc = 0;
 786         switch (val) {
 787         case 0:
 788                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 789                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 790                         if (!rc) {
 791                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 792                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 793                         }
 794                 }
 795                 break;
 796         case 1:
 797                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 798                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 799                         if (!rc) {
 800                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 801                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 802                         }
 803                 }
 804                 break;
 805         default:
 806                 break;
 807         }
 808 out:
 809         mutex_unlock(&smp_cpu_state_mutex);
 810         put_online_cpus();
 811         return rc ? rc : count;
 812 }
 813 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 814 #endif /* CONFIG_HOTPLUG_CPU */
 815
 816 static ssize_t cpu_polarization_show(struct sys_device *dev,
 817                                      struct sysdev_attribute *attr, char *buf)
 818 {
 819         int cpu = dev->id;
 820         ssize_t count;
 821
 822         mutex_lock(&smp_cpu_state_mutex);
 823         switch (smp_cpu_polarization[cpu]) {
 824         case POLARIZATION_HRZ:
 825                 count = sprintf(buf, "horizontal\n");
 826                 break;
 827         case POLARIZATION_VL:
 828                 count = sprintf(buf, "vertical:low\n");
 829                 break;
 830         case POLARIZATION_VM:
 831                 count = sprintf(buf, "vertical:medium\n");
 832                 break;
 833         case POLARIZATION_VH:
 834                 count = sprintf(buf, "vertical:high\n");
 835                 break;
 836         default:
 837                 count = sprintf(buf, "unknown\n");
 838                 break;
 839         }
 840         mutex_unlock(&smp_cpu_state_mutex);
 841         return count;
 842 }
 843 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 844
 845 static ssize_t show_cpu_address(struct sys_device *dev,
 846                                 struct sysdev_attribute *attr, char *buf)
 847 {
 848         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 849 }
 850 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 851
 852
 853 static struct attribute *cpu_common_attrs[] = {
 854 #ifdef CONFIG_HOTPLUG_CPU
 855         &attr_configure.attr,
 856 #endif
 857         &attr_address.attr,
 858         &attr_polarization.attr,
 859         NULL,
 860 };
 861
 862 static struct attribute_group cpu_common_attr_group = {
 863         .attrs = cpu_common_attrs,
 864 };
 865
 866 static ssize_t show_capability(struct sys_device *dev,
 867                                 struct sysdev_attribute *attr, char *buf)
 868 {
 869         unsigned int capability;
 870         int rc;
 871
 872         rc = get_cpu_capability(&capability);
 873         if (rc)
 874                 return rc;
 875         return sprintf(buf, "%u\n", capability);
 876 }
 877 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 878
 879 static ssize_t show_idle_count(struct sys_device *dev,
 880                                 struct sysdev_attribute *attr, char *buf)
 881 {
 882         struct s390_idle_data *idle;
 883         unsigned long long idle_count;
 884         unsigned int sequence;
 885
 886         idle = &per_cpu(s390_idle, dev->id);
 887 repeat:
 888         sequence = idle->sequence;
 889         smp_rmb();
 890         if (sequence & 1)
 891                 goto repeat;
 892         idle_count = idle->idle_count;
 893         if (idle->idle_enter)
 894                 idle_count++;
 895         smp_rmb();
 896         if (idle->sequence != sequence)
 897                 goto repeat;
 898         return sprintf(buf, "%llu\n", idle_count);
 899 }
 900 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 901
 902 static ssize_t show_idle_time(struct sys_device *dev,
 903                                 struct sysdev_attribute *attr, char *buf)
 904 {
 905         struct s390_idle_data *idle;
 906         unsigned long long now, idle_time, idle_enter;
 907         unsigned int sequence;
 908
 909         idle = &per_cpu(s390_idle, dev->id);
 910         now = get_clock();
 911 repeat:
 912         sequence = idle->sequence;
 913         smp_rmb();
 914         if (sequence & 1)
 915                 goto repeat;
 916         idle_time = idle->idle_time;
 917         idle_enter = idle->idle_enter;
 918         if (idle_enter != 0ULL && idle_enter < now)
 919                 idle_time += now - idle_enter;
 920         smp_rmb();
 921         if (idle->sequence != sequence)
 922                 goto repeat;
 923         return sprintf(buf, "%llu\n", idle_time >> 12);
 924 }
 925 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 926
 927 static struct attribute *cpu_online_attrs[] = {
 928         &attr_capability.attr,
 929         &attr_idle_count.attr,
 930         &attr_idle_time_us.attr,
 931         NULL,
 932 };
 933
 934 static struct attribute_group cpu_online_attr_group = {
 935         .attrs = cpu_online_attrs,
 936 };
 937
 938 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 939                                     unsigned long action, void *hcpu)
 940 {
 941         unsigned int cpu = (unsigned int)(long)hcpu;
 942         struct cpu *c = &per_cpu(cpu_devices, cpu);
 943         struct sys_device *s = &c->sysdev;
 944         struct s390_idle_data *idle;
 945
 946         switch (action) {
 947         case CPU_ONLINE:
 948         case CPU_ONLINE_FROZEN:
 949                 idle = &per_cpu(s390_idle, cpu);
 950                 memset(idle, 0, sizeof(struct s390_idle_data));
 951                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 952                         return NOTIFY_BAD;
 953                 break;
 954         case CPU_DEAD:
 955         case CPU_DEAD_FROZEN:
 956                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 957                 break;
 958         }
 959         return NOTIFY_OK;
 960 }
 961
 962 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 963         .notifier_call = smp_cpu_notify,
 964 };
 965
 966 static int __devinit smp_add_present_cpu(int cpu)
 967 {
 968         struct cpu *c = &per_cpu(cpu_devices, cpu);
 969         struct sys_device *s = &c->sysdev;
 970         int rc;
 971
 972         c->hotpluggable = 1;
 973         rc = register_cpu(c, cpu);
 974         if (rc)
 975                 goto out;
 976         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 977         if (rc)
 978                 goto out_cpu;
 979         if (!cpu_online(cpu))
 980                 goto out;
 981         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 982         if (!rc)
 983                 return 0;
 984         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 985 out_cpu:
 986 #ifdef CONFIG_HOTPLUG_CPU
 987         unregister_cpu(c);
 988 #endif
 989 out:
 990         return rc;
 991 }
 992
 993 #ifdef CONFIG_HOTPLUG_CPU
 994
 995 int __ref smp_rescan_cpus(void)
 996 {
 997         cpumask_t newcpus;
 998         int cpu;
 999         int rc;
1000
1001         get_online_cpus();
1002         mutex_lock(&smp_cpu_state_mutex);
1003         newcpus = cpu_present_map;
1004         rc = __smp_rescan_cpus();
1005         if (rc)
1006                 goto out;
1007         cpus_andnot(newcpus, cpu_present_map, newcpus);
1008         for_each_cpu_mask(cpu, newcpus) {
1009                 rc = smp_add_present_cpu(cpu);
1010                 if (rc)
1011                         cpu_clear(cpu, cpu_present_map);
1012         }
1013         rc = 0;
1014 out:
1015         mutex_unlock(&smp_cpu_state_mutex);
1016         put_online_cpus();
1017         if (!cpus_empty(newcpus))
1018                 topology_schedule_update();
1019         return rc;
1020 }
1021
1022 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1023                                   size_t count)
1024 {
1025         int rc;
1026
1027         rc = smp_rescan_cpus();
1028         return rc ? rc : count;
1029 }
1030 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1031 #endif /* CONFIG_HOTPLUG_CPU */
1032
1033 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1034 {
1035         ssize_t count;
1036
1037         mutex_lock(&smp_cpu_state_mutex);
1038         count = sprintf(buf, "%d\n", cpu_management);
1039         mutex_unlock(&smp_cpu_state_mutex);
1040         return count;
1041 }
1042
1043 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1044                                  size_t count)
1045 {
1046         int val, rc;
1047         char delim;
1048
1049         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1050                 return -EINVAL;
1051         if (val != 0 && val != 1)
1052                 return -EINVAL;
1053         rc = 0;
1054         get_online_cpus();
1055         mutex_lock(&smp_cpu_state_mutex);
1056         if (cpu_management == val)
1057                 goto out;
1058         rc = topology_set_cpu_management(val);
1059         if (!rc)
1060                 cpu_management = val;
1061 out:
1062         mutex_unlock(&smp_cpu_state_mutex);
1063         put_online_cpus();
1064         return rc ? rc : count;
1065 }
1066 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1067                          dispatching_store);
1068
1069 static int __init topology_init(void)
1070 {
1071         int cpu;
1072         int rc;
1073
1074         register_cpu_notifier(&smp_cpu_nb);
1075
1076 #ifdef CONFIG_HOTPLUG_CPU
1077         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1078         if (rc)
1079                 return rc;
1080 #endif
1081         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1082         if (rc)
1083                 return rc;
1084         for_each_present_cpu(cpu) {
1085                 rc = smp_add_present_cpu(cpu);
1086                 if (rc)
1087                         return rc;
1088         }
1089         return 0;
1090 }
1091 subsys_initcall(topology_init);