SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/s390/kernel/smp.c

   1 /*
   2  *  arch/s390/kernel/smp.c
   3  *
   4  *    Copyright IBM Corp. 1999,2007
   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 #include <linux/module.h>
  24 #include <linux/init.h>
  25 #include <linux/mm.h>
  26 #include <linux/err.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/kernel_stat.h>
  29 #include <linux/delay.h>
  30 #include <linux/cache.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/cpu.h>
  33 #include <linux/timex.h>
  34 #include <linux/bootmem.h>
  35 #include <asm/ipl.h>
  36 #include <asm/setup.h>
  37 #include <asm/sigp.h>
  38 #include <asm/pgalloc.h>
  39 #include <asm/irq.h>
  40 #include <asm/s390_ext.h>
  41 #include <asm/cpcmd.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/timer.h>
  44 #include <asm/lowcore.h>
  45 #include <asm/sclp.h>
  46 #include <asm/cpu.h>
  47 #include "entry.h"
  48
  49 /*
  50  * An array with a pointer the lowcore of every CPU.
  51  */
  52 struct _lowcore *lowcore_ptr[NR_CPUS];
  53 EXPORT_SYMBOL(lowcore_ptr);
  54
  55 cpumask_t cpu_online_map = CPU_MASK_NONE;
  56 EXPORT_SYMBOL(cpu_online_map);
  57
  58 cpumask_t cpu_possible_map = CPU_MASK_ALL;
  59 EXPORT_SYMBOL(cpu_possible_map);
  60
  61 static struct task_struct *current_set[NR_CPUS];
  62
  63 static u8 smp_cpu_type;
  64 static int smp_use_sigp_detection;
  65
  66 enum s390_cpu_state {
  67         CPU_STATE_STANDBY,
  68         CPU_STATE_CONFIGURED,
  69 };
  70
  71 DEFINE_MUTEX(smp_cpu_state_mutex);
  72 int smp_cpu_polarization[NR_CPUS];
  73 static int smp_cpu_state[NR_CPUS];
  74 static int cpu_management;
  75
  76 static DEFINE_PER_CPU(struct cpu, cpu_devices);
  77
  78 static void smp_ext_bitcall(int, ec_bit_sig);
  79
  80 void smp_send_stop(void)
  81 {
  82         int cpu, rc;
  83
  84         /* Disable all interrupts/machine checks */
  85         __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
  86
  87         /* write magic number to zero page (absolute 0) */
  88         lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
  89
  90         /* stop all processors */
  91         for_each_online_cpu(cpu) {
  92                 if (cpu == smp_processor_id())
  93                         continue;
  94                 do {
  95                         rc = signal_processor(cpu, sigp_stop);
  96                 } while (rc == sigp_busy);
  97
  98                 while (!smp_cpu_not_running(cpu))
  99                         cpu_relax();
 100         }
 101 }
 102
 103 /*
 104  * This is the main routine where commands issued by other
 105  * cpus are handled.
 106  */
 107
 108 static void do_ext_call_interrupt(__u16 code)
 109 {
 110         unsigned long bits;
 111
 112         /*
 113          * handle bit signal external calls
 114          *
 115          * For the ec_schedule signal we have to do nothing. All the work
 116          * is done automatically when we return from the interrupt.
 117          */
 118         bits = xchg(&S390_lowcore.ext_call_fast, 0);
 119
 120         if (test_bit(ec_call_function, &bits))
 121                 generic_smp_call_function_interrupt();
 122
 123         if (test_bit(ec_call_function_single, &bits))
 124                 generic_smp_call_function_single_interrupt();
 125 }
 126
 127 /*
 128  * Send an external call sigp to another cpu and return without waiting
 129  * for its completion.
 130  */
 131 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
 132 {
 133         /*
 134          * Set signaling bit in lowcore of target cpu and kick it
 135          */
 136         set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
 137         while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
 138                 udelay(10);
 139 }
 140
 141 void arch_send_call_function_ipi(cpumask_t mask)
 142 {
 143         int cpu;
 144
 145         for_each_cpu_mask(cpu, mask)
 146                 smp_ext_bitcall(cpu, ec_call_function);
 147 }
 148
 149 void arch_send_call_function_single_ipi(int cpu)
 150 {
 151         smp_ext_bitcall(cpu, ec_call_function_single);
 152 }
 153
 154 #ifndef CONFIG_64BIT
 155 /*
 156  * this function sends a 'purge tlb' signal to another CPU.
 157  */
 158 static void smp_ptlb_callback(void *info)
 159 {
 160         __tlb_flush_local();
 161 }
 162
 163 void smp_ptlb_all(void)
 164 {
 165         on_each_cpu(smp_ptlb_callback, NULL, 1);
 166 }
 167 EXPORT_SYMBOL(smp_ptlb_all);
 168 #endif /* ! CONFIG_64BIT */
 169
 170 /*
 171  * this function sends a 'reschedule' IPI to another CPU.
 172  * it goes straight through and wastes no time serializing
 173  * anything. Worst case is that we lose a reschedule ...
 174  */
 175 void smp_send_reschedule(int cpu)
 176 {
 177         smp_ext_bitcall(cpu, ec_schedule);
 178 }
 179
 180 /*
 181  * parameter area for the set/clear control bit callbacks
 182  */
 183 struct ec_creg_mask_parms {
 184         unsigned long orvals[16];
 185         unsigned long andvals[16];
 186 };
 187
 188 /*
 189  * callback for setting/clearing control bits
 190  */
 191 static void smp_ctl_bit_callback(void *info)
 192 {
 193         struct ec_creg_mask_parms *pp = info;
 194         unsigned long cregs[16];
 195         int i;
 196
 197         __ctl_store(cregs, 0, 15);
 198         for (i = 0; i <= 15; i++)
 199                 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
 200         __ctl_load(cregs, 0, 15);
 201 }
 202
 203 /*
 204  * Set a bit in a control register of all cpus
 205  */
 206 void smp_ctl_set_bit(int cr, int bit)
 207 {
 208         struct ec_creg_mask_parms parms;
 209
 210         memset(&parms.orvals, 0, sizeof(parms.orvals));
 211         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 212         parms.orvals[cr] = 1 << bit;
 213         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 214 }
 215 EXPORT_SYMBOL(smp_ctl_set_bit);
 216
 217 /*
 218  * Clear a bit in a control register of all cpus
 219  */
 220 void smp_ctl_clear_bit(int cr, int bit)
 221 {
 222         struct ec_creg_mask_parms parms;
 223
 224         memset(&parms.orvals, 0, sizeof(parms.orvals));
 225         memset(&parms.andvals, 0xff, sizeof(parms.andvals));
 226         parms.andvals[cr] = ~(1L << bit);
 227         on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 228 }
 229 EXPORT_SYMBOL(smp_ctl_clear_bit);
 230
 231 /*
 232  * In early ipl state a temp. logically cpu number is needed, so the sigp
 233  * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
 234  * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
 235  */
 236 #define CPU_INIT_NO     1
 237
 238 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 239
 240 /*
 241  * zfcpdump_prefix_array holds prefix registers for the following scenario:
 242  * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
 243  * save its prefix registers, since they get lost, when switching from 31 bit
 244  * to 64 bit.
 245  */
 246 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
 247         __attribute__((__section__(".data")));
 248
 249 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 250 {
 251         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 252                 return;
 253         if (cpu >= NR_CPUS) {
 254                 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
 255                        "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
 256                 return;
 257         }
 258         zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
 259         __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
 260         while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
 261                sigp_busy)
 262                 cpu_relax();
 263         memcpy(zfcpdump_save_areas[cpu],
 264                (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
 265                SAVE_AREA_SIZE);
 266 #ifdef CONFIG_64BIT
 267         /* copy original prefix register */
 268         zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
 269 #endif
 270 }
 271
 272 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
 273 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 274
 275 #else
 276
 277 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 278
 279 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 280
 281 static int cpu_stopped(int cpu)
 282 {
 283         __u32 status;
 284
 285         /* Check for stopped state */
 286         if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
 287             sigp_status_stored) {
 288                 if (status & 0x40)
 289                         return 1;
 290         }
 291         return 0;
 292 }
 293
 294 static int cpu_known(int cpu_id)
 295 {
 296         int cpu;
 297
 298         for_each_present_cpu(cpu) {
 299                 if (__cpu_logical_map[cpu] == cpu_id)
 300                         return 1;
 301         }
 302         return 0;
 303 }
 304
 305 static int smp_rescan_cpus_sigp(cpumask_t avail)
 306 {
 307         int cpu_id, logical_cpu;
 308
 309         logical_cpu = first_cpu(avail);
 310         if (logical_cpu == NR_CPUS)
 311                 return 0;
 312         for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
 313                 if (cpu_known(cpu_id))
 314                         continue;
 315                 __cpu_logical_map[logical_cpu] = cpu_id;
 316                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 317                 if (!cpu_stopped(logical_cpu))
 318                         continue;
 319                 cpu_set(logical_cpu, cpu_present_map);
 320                 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 321                 logical_cpu = next_cpu(logical_cpu, avail);
 322                 if (logical_cpu == NR_CPUS)
 323                         break;
 324         }
 325         return 0;
 326 }
 327
 328 static int smp_rescan_cpus_sclp(cpumask_t avail)
 329 {
 330         struct sclp_cpu_info *info;
 331         int cpu_id, logical_cpu, cpu;
 332         int rc;
 333
 334         logical_cpu = first_cpu(avail);
 335         if (logical_cpu == NR_CPUS)
 336                 return 0;
 337         info = kmalloc(sizeof(*info), GFP_KERNEL);
 338         if (!info)
 339                 return -ENOMEM;
 340         rc = sclp_get_cpu_info(info);
 341         if (rc)
 342                 goto out;
 343         for (cpu = 0; cpu < info->combined; cpu++) {
 344                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 345                         continue;
 346                 cpu_id = info->cpu[cpu].address;
 347                 if (cpu_known(cpu_id))
 348                         continue;
 349                 __cpu_logical_map[logical_cpu] = cpu_id;
 350                 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
 351                 cpu_set(logical_cpu, cpu_present_map);
 352                 if (cpu >= info->configured)
 353                         smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
 354                 else
 355                         smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
 356                 logical_cpu = next_cpu(logical_cpu, avail);
 357                 if (logical_cpu == NR_CPUS)
 358                         break;
 359         }
 360 out:
 361         kfree(info);
 362         return rc;
 363 }
 364
 365 static int __smp_rescan_cpus(void)
 366 {
 367         cpumask_t avail;
 368
 369         cpus_xor(avail, cpu_possible_map, cpu_present_map);
 370         if (smp_use_sigp_detection)
 371                 return smp_rescan_cpus_sigp(avail);
 372         else
 373                 return smp_rescan_cpus_sclp(avail);
 374 }
 375
 376 static void __init smp_detect_cpus(void)
 377 {
 378         unsigned int cpu, c_cpus, s_cpus;
 379         struct sclp_cpu_info *info;
 380         u16 boot_cpu_addr, cpu_addr;
 381
 382         c_cpus = 1;
 383         s_cpus = 0;
 384         boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
 385         info = kmalloc(sizeof(*info), GFP_KERNEL);
 386         if (!info)
 387                 panic("smp_detect_cpus failed to allocate memory\n");
 388         /* Use sigp detection algorithm if sclp doesn't work. */
 389         if (sclp_get_cpu_info(info)) {
 390                 smp_use_sigp_detection = 1;
 391                 for (cpu = 0; cpu <= 65535; cpu++) {
 392                         if (cpu == boot_cpu_addr)
 393                                 continue;
 394                         __cpu_logical_map[CPU_INIT_NO] = cpu;
 395                         if (!cpu_stopped(CPU_INIT_NO))
 396                                 continue;
 397                         smp_get_save_area(c_cpus, cpu);
 398                         c_cpus++;
 399                 }
 400                 goto out;
 401         }
 402
 403         if (info->has_cpu_type) {
 404                 for (cpu = 0; cpu < info->combined; cpu++) {
 405                         if (info->cpu[cpu].address == boot_cpu_addr) {
 406                                 smp_cpu_type = info->cpu[cpu].type;
 407                                 break;
 408                         }
 409                 }
 410         }
 411
 412         for (cpu = 0; cpu < info->combined; cpu++) {
 413                 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
 414                         continue;
 415                 cpu_addr = info->cpu[cpu].address;
 416                 if (cpu_addr == boot_cpu_addr)
 417                         continue;
 418                 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
 419                 if (!cpu_stopped(CPU_INIT_NO)) {
 420                         s_cpus++;
 421                         continue;
 422                 }
 423                 smp_get_save_area(c_cpus, cpu_addr);
 424                 c_cpus++;
 425         }
 426 out:
 427         kfree(info);
 428         printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
 429         get_online_cpus();
 430         __smp_rescan_cpus();
 431         put_online_cpus();
 432 }
 433
 434 /*
 435  *      Activate a secondary processor.
 436  */
 437 int __cpuinit start_secondary(void *cpuvoid)
 438 {
 439         /* Setup the cpu */
 440         cpu_init();
 441         preempt_disable();
 442         /* Enable TOD clock interrupts on the secondary cpu. */
 443         init_cpu_timer();
 444 #ifdef CONFIG_VIRT_TIMER
 445         /* Enable cpu timer interrupts on the secondary cpu. */
 446         init_cpu_vtimer();
 447 #endif
 448         /* Enable pfault pseudo page faults on this cpu. */
 449         pfault_init();
 450
 451         /* call cpu notifiers */
 452         notify_cpu_starting(smp_processor_id());
 453         /* Mark this cpu as online */
 454         ipi_call_lock();
 455         cpu_set(smp_processor_id(), cpu_online_map);
 456         ipi_call_unlock();
 457         /* Switch on interrupts */
 458         local_irq_enable();
 459         /* Print info about this processor */
 460         print_cpu_info(&S390_lowcore.cpu_data);
 461         /* cpu_idle will call schedule for us */
 462         cpu_idle();
 463         return 0;
 464 }
 465
 466 static void __init smp_create_idle(unsigned int cpu)
 467 {
 468         struct task_struct *p;
 469
 470         /*
 471          *  don't care about the psw and regs settings since we'll never
 472          *  reschedule the forked task.
 473          */
 474         p = fork_idle(cpu);
 475         if (IS_ERR(p))
 476                 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
 477         current_set[cpu] = p;
 478 }
 479
 480 static int __cpuinit smp_alloc_lowcore(int cpu)
 481 {
 482         unsigned long async_stack, panic_stack;
 483         struct _lowcore *lowcore;
 484         int lc_order;
 485
 486         lc_order = sizeof(long) == 8 ? 1 : 0;
 487         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 488         if (!lowcore)
 489                 return -ENOMEM;
 490         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 491         panic_stack = __get_free_page(GFP_KERNEL);
 492         if (!panic_stack || !async_stack)
 493                 goto out;
 494         memcpy(lowcore, &S390_lowcore, 512);
 495         memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
 496         lowcore->async_stack = async_stack + ASYNC_SIZE;
 497         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 498
 499 #ifndef CONFIG_64BIT
 500         if (MACHINE_HAS_IEEE) {
 501                 unsigned long save_area;
 502
 503                 save_area = get_zeroed_page(GFP_KERNEL);
 504                 if (!save_area)
 505                         goto out_save_area;
 506                 lowcore->extended_save_area_addr = (u32) save_area;
 507         }
 508 #endif
 509         lowcore_ptr[cpu] = lowcore;
 510         return 0;
 511
 512 #ifndef CONFIG_64BIT
 513 out_save_area:
 514         free_page(panic_stack);
 515 #endif
 516 out:
 517         free_pages(async_stack, ASYNC_ORDER);
 518         free_pages((unsigned long) lowcore, lc_order);
 519         return -ENOMEM;
 520 }
 521
 522 #ifdef CONFIG_HOTPLUG_CPU
 523 static void smp_free_lowcore(int cpu)
 524 {
 525         struct _lowcore *lowcore;
 526         int lc_order;
 527
 528         lc_order = sizeof(long) == 8 ? 1 : 0;
 529         lowcore = lowcore_ptr[cpu];
 530 #ifndef CONFIG_64BIT
 531         if (MACHINE_HAS_IEEE)
 532                 free_page((unsigned long) lowcore->extended_save_area_addr);
 533 #endif
 534         free_page(lowcore->panic_stack - PAGE_SIZE);
 535         free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
 536         free_pages((unsigned long) lowcore, lc_order);
 537         lowcore_ptr[cpu] = NULL;
 538 }
 539 #endif /* CONFIG_HOTPLUG_CPU */
 540
 541 /* Upping and downing of CPUs */
 542 int __cpuinit __cpu_up(unsigned int cpu)
 543 {
 544         struct task_struct *idle;
 545         struct _lowcore *cpu_lowcore;
 546         struct stack_frame *sf;
 547         sigp_ccode ccode;
 548
 549         if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
 550                 return -EIO;
 551         if (smp_alloc_lowcore(cpu))
 552                 return -ENOMEM;
 553
 554         ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
 555                                    cpu, sigp_set_prefix);
 556         if (ccode) {
 557                 printk("sigp_set_prefix failed for cpu %d "
 558                        "with condition code %d\n",
 559                        (int) cpu, (int) ccode);
 560                 return -EIO;
 561         }
 562
 563         idle = current_set[cpu];
 564         cpu_lowcore = lowcore_ptr[cpu];
 565         cpu_lowcore->kernel_stack = (unsigned long)
 566                 task_stack_page(idle) + THREAD_SIZE;
 567         cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
 568         sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
 569                                      - sizeof(struct pt_regs)
 570                                      - sizeof(struct stack_frame));
 571         memset(sf, 0, sizeof(struct stack_frame));
 572         sf->gprs[9] = (unsigned long) sf;
 573         cpu_lowcore->save_area[15] = (unsigned long) sf;
 574         __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
 575         asm volatile(
 576                 "       stam    0,15,0(%0)"
 577                 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
 578         cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
 579         cpu_lowcore->current_task = (unsigned long) idle;
 580         cpu_lowcore->cpu_data.cpu_nr = cpu;
 581         cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
 582         cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
 583         eieio();
 584
 585         while (signal_processor(cpu, sigp_restart) == sigp_busy)
 586                 udelay(10);
 587
 588         while (!cpu_online(cpu))
 589                 cpu_relax();
 590         return 0;
 591 }
 592
 593 static int __init setup_possible_cpus(char *s)
 594 {
 595         int pcpus, cpu;
 596
 597         pcpus = simple_strtoul(s, NULL, 0);
 598         cpu_possible_map = cpumask_of_cpu(0);
 599         for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
 600                 cpu_set(cpu, cpu_possible_map);
 601         return 0;
 602 }
 603 early_param("possible_cpus", setup_possible_cpus);
 604
 605 #ifdef CONFIG_HOTPLUG_CPU
 606
 607 int __cpu_disable(void)
 608 {
 609         struct ec_creg_mask_parms cr_parms;
 610         int cpu = smp_processor_id();
 611
 612         cpu_clear(cpu, cpu_online_map);
 613
 614         /* Disable pfault pseudo page faults on this cpu. */
 615         pfault_fini();
 616
 617         memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
 618         memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
 619
 620         /* disable all external interrupts */
 621         cr_parms.orvals[0] = 0;
 622         cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
 623                                 1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
 624         /* disable all I/O interrupts */
 625         cr_parms.orvals[6] = 0;
 626         cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
 627                                 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
 628         /* disable most machine checks */
 629         cr_parms.orvals[14] = 0;
 630         cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
 631                                  1 << 25 | 1 << 24);
 632
 633         smp_ctl_bit_callback(&cr_parms);
 634
 635         return 0;
 636 }
 637
 638 void __cpu_die(unsigned int cpu)
 639 {
 640         /* Wait until target cpu is down */
 641         while (!smp_cpu_not_running(cpu))
 642                 cpu_relax();
 643         smp_free_lowcore(cpu);
 644         printk(KERN_INFO "Processor %d spun down\n", cpu);
 645 }
 646
 647 void cpu_die(void)
 648 {
 649         idle_task_exit();
 650         signal_processor(smp_processor_id(), sigp_stop);
 651         BUG();
 652         for (;;);
 653 }
 654
 655 #endif /* CONFIG_HOTPLUG_CPU */
 656
 657 void __init smp_prepare_cpus(unsigned int max_cpus)
 658 {
 659 #ifndef CONFIG_64BIT
 660         unsigned long save_area = 0;
 661 #endif
 662         unsigned long async_stack, panic_stack;
 663         struct _lowcore *lowcore;
 664         unsigned int cpu;
 665         int lc_order;
 666
 667         smp_detect_cpus();
 668
 669         /* request the 0x1201 emergency signal external interrupt */
 670         if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
 671                 panic("Couldn't request external interrupt 0x1201");
 672         print_cpu_info(&S390_lowcore.cpu_data);
 673
 674         /* Reallocate current lowcore, but keep its contents. */
 675         lc_order = sizeof(long) == 8 ? 1 : 0;
 676         lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
 677         panic_stack = __get_free_page(GFP_KERNEL);
 678         async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
 679 #ifndef CONFIG_64BIT
 680         if (MACHINE_HAS_IEEE)
 681                 save_area = get_zeroed_page(GFP_KERNEL);
 682 #endif
 683         local_irq_disable();
 684         local_mcck_disable();
 685         lowcore_ptr[smp_processor_id()] = lowcore;
 686         *lowcore = S390_lowcore;
 687         lowcore->panic_stack = panic_stack + PAGE_SIZE;
 688         lowcore->async_stack = async_stack + ASYNC_SIZE;
 689 #ifndef CONFIG_64BIT
 690         if (MACHINE_HAS_IEEE)
 691                 lowcore->extended_save_area_addr = (u32) save_area;
 692 #endif
 693         set_prefix((u32)(unsigned long) lowcore);
 694         local_mcck_enable();
 695         local_irq_enable();
 696         for_each_possible_cpu(cpu)
 697                 if (cpu != smp_processor_id())
 698                         smp_create_idle(cpu);
 699 }
 700
 701 void __init smp_prepare_boot_cpu(void)
 702 {
 703         BUG_ON(smp_processor_id() != 0);
 704
 705         current_thread_info()->cpu = 0;
 706         cpu_set(0, cpu_present_map);
 707         cpu_set(0, cpu_online_map);
 708         S390_lowcore.percpu_offset = __per_cpu_offset[0];
 709         current_set[0] = current;
 710         smp_cpu_state[0] = CPU_STATE_CONFIGURED;
 711         smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 712 }
 713
 714 void __init smp_cpus_done(unsigned int max_cpus)
 715 {
 716 }
 717
 718 /*
 719  * the frequency of the profiling timer can be changed
 720  * by writing a multiplier value into /proc/profile.
 721  *
 722  * usually you want to run this on all CPUs ;)
 723  */
 724 int setup_profiling_timer(unsigned int multiplier)
 725 {
 726         return 0;
 727 }
 728
 729 #ifdef CONFIG_HOTPLUG_CPU
 730 static ssize_t cpu_configure_show(struct sys_device *dev,
 731                                 struct sysdev_attribute *attr, char *buf)
 732 {
 733         ssize_t count;
 734
 735         mutex_lock(&smp_cpu_state_mutex);
 736         count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
 737         mutex_unlock(&smp_cpu_state_mutex);
 738         return count;
 739 }
 740
 741 static ssize_t cpu_configure_store(struct sys_device *dev,
 742                                   struct sysdev_attribute *attr,
 743                                   const char *buf, size_t count)
 744 {
 745         int cpu = dev->id;
 746         int val, rc;
 747         char delim;
 748
 749         if (sscanf(buf, "%d %c", &val, &delim) != 1)
 750                 return -EINVAL;
 751         if (val != 0 && val != 1)
 752                 return -EINVAL;
 753
 754         get_online_cpus();
 755         mutex_lock(&smp_cpu_state_mutex);
 756         rc = -EBUSY;
 757         if (cpu_online(cpu))
 758                 goto out;
 759         rc = 0;
 760         switch (val) {
 761         case 0:
 762                 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
 763                         rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
 764                         if (!rc) {
 765                                 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
 766                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 767                         }
 768                 }
 769                 break;
 770         case 1:
 771                 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
 772                         rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
 773                         if (!rc) {
 774                                 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
 775                                 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
 776                         }
 777                 }
 778                 break;
 779         default:
 780                 break;
 781         }
 782 out:
 783         mutex_unlock(&smp_cpu_state_mutex);
 784         put_online_cpus();
 785         return rc ? rc : count;
 786 }
 787 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
 788 #endif /* CONFIG_HOTPLUG_CPU */
 789
 790 static ssize_t cpu_polarization_show(struct sys_device *dev,
 791                                      struct sysdev_attribute *attr, char *buf)
 792 {
 793         int cpu = dev->id;
 794         ssize_t count;
 795
 796         mutex_lock(&smp_cpu_state_mutex);
 797         switch (smp_cpu_polarization[cpu]) {
 798         case POLARIZATION_HRZ:
 799                 count = sprintf(buf, "horizontal\n");
 800                 break;
 801         case POLARIZATION_VL:
 802                 count = sprintf(buf, "vertical:low\n");
 803                 break;
 804         case POLARIZATION_VM:
 805                 count = sprintf(buf, "vertical:medium\n");
 806                 break;
 807         case POLARIZATION_VH:
 808                 count = sprintf(buf, "vertical:high\n");
 809                 break;
 810         default:
 811                 count = sprintf(buf, "unknown\n");
 812                 break;
 813         }
 814         mutex_unlock(&smp_cpu_state_mutex);
 815         return count;
 816 }
 817 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
 818
 819 static ssize_t show_cpu_address(struct sys_device *dev,
 820                                 struct sysdev_attribute *attr, char *buf)
 821 {
 822         return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
 823 }
 824 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
 825
 826
 827 static struct attribute *cpu_common_attrs[] = {
 828 #ifdef CONFIG_HOTPLUG_CPU
 829         &attr_configure.attr,
 830 #endif
 831         &attr_address.attr,
 832         &attr_polarization.attr,
 833         NULL,
 834 };
 835
 836 static struct attribute_group cpu_common_attr_group = {
 837         .attrs = cpu_common_attrs,
 838 };
 839
 840 static ssize_t show_capability(struct sys_device *dev,
 841                                 struct sysdev_attribute *attr, char *buf)
 842 {
 843         unsigned int capability;
 844         int rc;
 845
 846         rc = get_cpu_capability(&capability);
 847         if (rc)
 848                 return rc;
 849         return sprintf(buf, "%u\n", capability);
 850 }
 851 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
 852
 853 static ssize_t show_idle_count(struct sys_device *dev,
 854                                 struct sysdev_attribute *attr, char *buf)
 855 {
 856         struct s390_idle_data *idle;
 857         unsigned long long idle_count;
 858
 859         idle = &per_cpu(s390_idle, dev->id);
 860         spin_lock_irq(&idle->lock);
 861         idle_count = idle->idle_count;
 862         spin_unlock_irq(&idle->lock);
 863         return sprintf(buf, "%llu\n", idle_count);
 864 }
 865 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
 866
 867 static ssize_t show_idle_time(struct sys_device *dev,
 868                                 struct sysdev_attribute *attr, char *buf)
 869 {
 870         struct s390_idle_data *idle;
 871         unsigned long long new_time;
 872
 873         idle = &per_cpu(s390_idle, dev->id);
 874         spin_lock_irq(&idle->lock);
 875         if (idle->in_idle) {
 876                 new_time = get_clock();
 877                 idle->idle_time += new_time - idle->idle_enter;
 878                 idle->idle_enter = new_time;
 879         }
 880         new_time = idle->idle_time;
 881         spin_unlock_irq(&idle->lock);
 882         return sprintf(buf, "%llu\n", new_time >> 12);
 883 }
 884 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 885
 886 static struct attribute *cpu_online_attrs[] = {
 887         &attr_capability.attr,
 888         &attr_idle_count.attr,
 889         &attr_idle_time_us.attr,
 890         NULL,
 891 };
 892
 893 static struct attribute_group cpu_online_attr_group = {
 894         .attrs = cpu_online_attrs,
 895 };
 896
 897 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
 898                                     unsigned long action, void *hcpu)
 899 {
 900         unsigned int cpu = (unsigned int)(long)hcpu;
 901         struct cpu *c = &per_cpu(cpu_devices, cpu);
 902         struct sys_device *s = &c->sysdev;
 903         struct s390_idle_data *idle;
 904
 905         switch (action) {
 906         case CPU_ONLINE:
 907         case CPU_ONLINE_FROZEN:
 908                 idle = &per_cpu(s390_idle, cpu);
 909                 spin_lock_irq(&idle->lock);
 910                 idle->idle_enter = 0;
 911                 idle->idle_time = 0;
 912                 idle->idle_count = 0;
 913                 spin_unlock_irq(&idle->lock);
 914                 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
 915                         return NOTIFY_BAD;
 916                 break;
 917         case CPU_DEAD:
 918         case CPU_DEAD_FROZEN:
 919                 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
 920                 break;
 921         }
 922         return NOTIFY_OK;
 923 }
 924
 925 static struct notifier_block __cpuinitdata smp_cpu_nb = {
 926         .notifier_call = smp_cpu_notify,
 927 };
 928
 929 static int __devinit smp_add_present_cpu(int cpu)
 930 {
 931         struct cpu *c = &per_cpu(cpu_devices, cpu);
 932         struct sys_device *s = &c->sysdev;
 933         int rc;
 934
 935         c->hotpluggable = 1;
 936         rc = register_cpu(c, cpu);
 937         if (rc)
 938                 goto out;
 939         rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
 940         if (rc)
 941                 goto out_cpu;
 942         if (!cpu_online(cpu))
 943                 goto out;
 944         rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
 945         if (!rc)
 946                 return 0;
 947         sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
 948 out_cpu:
 949 #ifdef CONFIG_HOTPLUG_CPU
 950         unregister_cpu(c);
 951 #endif
 952 out:
 953         return rc;
 954 }
 955
 956 #ifdef CONFIG_HOTPLUG_CPU
 957
 958 int __ref smp_rescan_cpus(void)
 959 {
 960         cpumask_t newcpus;
 961         int cpu;
 962         int rc;
 963
 964         get_online_cpus();
 965         mutex_lock(&smp_cpu_state_mutex);
 966         newcpus = cpu_present_map;
 967         rc = __smp_rescan_cpus();
 968         if (rc)
 969                 goto out;
 970         cpus_andnot(newcpus, cpu_present_map, newcpus);
 971         for_each_cpu_mask(cpu, newcpus) {
 972                 rc = smp_add_present_cpu(cpu);
 973                 if (rc)
 974                         cpu_clear(cpu, cpu_present_map);
 975         }
 976         rc = 0;
 977 out:
 978         mutex_unlock(&smp_cpu_state_mutex);
 979         put_online_cpus();
 980         if (!cpus_empty(newcpus))
 981                 topology_schedule_update();
 982         return rc;
 983 }
 984
 985 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
 986                                   size_t count)
 987 {
 988         int rc;
 989
 990         rc = smp_rescan_cpus();
 991         return rc ? rc : count;
 992 }
 993 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
 994 #endif /* CONFIG_HOTPLUG_CPU */
 995
 996 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
 997 {
 998         ssize_t count;
 999
1000         mutex_lock(&smp_cpu_state_mutex);
1001         count = sprintf(buf, "%d\n", cpu_management);
1002         mutex_unlock(&smp_cpu_state_mutex);
1003         return count;
1004 }
1005
1006 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1007                                  size_t count)
1008 {
1009         int val, rc;
1010         char delim;
1011
1012         if (sscanf(buf, "%d %c", &val, &delim) != 1)
1013                 return -EINVAL;
1014         if (val != 0 && val != 1)
1015                 return -EINVAL;
1016         rc = 0;
1017         get_online_cpus();
1018         mutex_lock(&smp_cpu_state_mutex);
1019         if (cpu_management == val)
1020                 goto out;
1021         rc = topology_set_cpu_management(val);
1022         if (!rc)
1023                 cpu_management = val;
1024 out:
1025         mutex_unlock(&smp_cpu_state_mutex);
1026         put_online_cpus();
1027         return rc ? rc : count;
1028 }
1029 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1030                          dispatching_store);
1031
1032 static int __init topology_init(void)
1033 {
1034         int cpu;
1035         int rc;
1036
1037         register_cpu_notifier(&smp_cpu_nb);
1038
1039 #ifdef CONFIG_HOTPLUG_CPU
1040         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1041         if (rc)
1042                 return rc;
1043 #endif
1044         rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1045         if (rc)
1046                 return rc;
1047         for_each_present_cpu(cpu) {
1048                 rc = smp_add_present_cpu(cpu);
1049                 if (rc)
1050                         return rc;
1051         }
1052         return 0;
1053 }
1054 subsys_initcall(topology_init);