git.oblomov.eu Git - linux-2.6/blob - arch/x86/kernel/apic/io_apic.c

   1 /*
   2  *      Intel IO-APIC support for multi-Pentium hosts.
   3  *
   4  *      Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
   5  *
   6  *      Many thanks to Stig Venaas for trying out countless experimental
   7  *      patches and reporting/debugging problems patiently!
   8  *
   9  *      (c) 1999, Multiple IO-APIC support, developed by
  10  *      Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
  11  *      Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
  12  *      further tested and cleaned up by Zach Brown <zab@redhat.com>
  13  *      and Ingo Molnar <mingo@redhat.com>
  14  *
  15  *      Fixes
  16  *      Maciej W. Rozycki       :       Bits for genuine 82489DX APICs;
  17  *                                      thanks to Eric Gilmore
  18  *                                      and Rolf G. Tews
  19  *                                      for testing these extensively
  20  *      Paul Diefenbaugh        :       Added full ACPI support
  21  */
  22
  23 #include <linux/mm.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/init.h>
  26 #include <linux/delay.h>
  27 #include <linux/sched.h>
  28 #include <linux/pci.h>
  29 #include <linux/mc146818rtc.h>
  30 #include <linux/compiler.h>
  31 #include <linux/acpi.h>
  32 #include <linux/module.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/msi.h>
  35 #include <linux/htirq.h>
  36 #include <linux/freezer.h>
  37 #include <linux/kthread.h>
  38 #include <linux/jiffies.h>      /* time_after() */
  39 #ifdef CONFIG_ACPI
  40 #include <acpi/acpi_bus.h>
  41 #endif
  42 #include <linux/bootmem.h>
  43 #include <linux/dmar.h>
  44 #include <linux/hpet.h>
  45
  46 #include <asm/idle.h>
  47 #include <asm/io.h>
  48 #include <asm/smp.h>
  49 #include <asm/cpu.h>
  50 #include <asm/desc.h>
  51 #include <asm/proto.h>
  52 #include <asm/acpi.h>
  53 #include <asm/dma.h>
  54 #include <asm/timer.h>
  55 #include <asm/i8259.h>
  56 #include <asm/nmi.h>
  57 #include <asm/msidef.h>
  58 #include <asm/hypertransport.h>
  59 #include <asm/setup.h>
  60 #include <asm/irq_remapping.h>
  61 #include <asm/hpet.h>
  62 #include <asm/uv/uv_hub.h>
  63 #include <asm/uv/uv_irq.h>
  64
  65 #include <asm/apic.h>
  66
  67 #define __apicdebuginit(type) static type __init
  68
  69 /*
  70  *      Is the SiS APIC rmw bug present ?
  71  *      -1 = don't know, 0 = no, 1 = yes
  72  */
  73 int sis_apic_bug = -1;
  74
  75 static DEFINE_SPINLOCK(ioapic_lock);
  76 static DEFINE_SPINLOCK(vector_lock);
  77
  78 /*
  79  * # of IRQ routing registers
  80  */
  81 int nr_ioapic_registers[MAX_IO_APICS];
  82
  83 /* I/O APIC entries */
  84 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
  85 int nr_ioapics;
  86
  87 /* MP IRQ source entries */
  88 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
  89
  90 /* # of MP IRQ source entries */
  91 int mp_irq_entries;
  92
  93 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
  94 int mp_bus_id_to_type[MAX_MP_BUSSES];
  95 #endif
  96
  97 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
  98
  99 int skip_ioapic_setup;
 100
 101 void arch_disable_smp_support(void)
 102 {
 103 #ifdef CONFIG_PCI
 104         noioapicquirk = 1;
 105         noioapicreroute = -1;
 106 #endif
 107         skip_ioapic_setup = 1;
 108 }
 109
 110 static int __init parse_noapic(char *str)
 111 {
 112         /* disable IO-APIC */
 113         arch_disable_smp_support();
 114         return 0;
 115 }
 116 early_param("noapic", parse_noapic);
 117
 118 struct irq_pin_list;
 119
 120 /*
 121  * This is performance-critical, we want to do it O(1)
 122  *
 123  * the indexing order of this array favors 1:1 mappings
 124  * between pins and IRQs.
 125  */
 126
 127 struct irq_pin_list {
 128         int apic, pin;
 129         struct irq_pin_list *next;
 130 };
 131
 132 static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
 133 {
 134         struct irq_pin_list *pin;
 135         int node;
 136
 137         node = cpu_to_node(cpu);
 138
 139         pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
 140
 141         return pin;
 142 }
 143
 144 struct irq_cfg {
 145         struct irq_pin_list *irq_2_pin;
 146         cpumask_var_t domain;
 147         cpumask_var_t old_domain;
 148         unsigned move_cleanup_count;
 149         u8 vector;
 150         u8 move_in_progress : 1;
 151 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
 152         u8 move_desc_pending : 1;
 153 #endif
 154 };
 155
 156 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
 157 #ifdef CONFIG_SPARSE_IRQ
 158 static struct irq_cfg irq_cfgx[] = {
 159 #else
 160 static struct irq_cfg irq_cfgx[NR_IRQS] = {
 161 #endif
 162         [0]  = { .vector = IRQ0_VECTOR,  },
 163         [1]  = { .vector = IRQ1_VECTOR,  },
 164         [2]  = { .vector = IRQ2_VECTOR,  },
 165         [3]  = { .vector = IRQ3_VECTOR,  },
 166         [4]  = { .vector = IRQ4_VECTOR,  },
 167         [5]  = { .vector = IRQ5_VECTOR,  },
 168         [6]  = { .vector = IRQ6_VECTOR,  },
 169         [7]  = { .vector = IRQ7_VECTOR,  },
 170         [8]  = { .vector = IRQ8_VECTOR,  },
 171         [9]  = { .vector = IRQ9_VECTOR,  },
 172         [10] = { .vector = IRQ10_VECTOR, },
 173         [11] = { .vector = IRQ11_VECTOR, },
 174         [12] = { .vector = IRQ12_VECTOR, },
 175         [13] = { .vector = IRQ13_VECTOR, },
 176         [14] = { .vector = IRQ14_VECTOR, },
 177         [15] = { .vector = IRQ15_VECTOR, },
 178 };
 179
 180 int __init arch_early_irq_init(void)
 181 {
 182         struct irq_cfg *cfg;
 183         struct irq_desc *desc;
 184         int count;
 185         int i;
 186
 187         cfg = irq_cfgx;
 188         count = ARRAY_SIZE(irq_cfgx);
 189
 190         for (i = 0; i < count; i++) {
 191                 desc = irq_to_desc(i);
 192                 desc->chip_data = &cfg[i];
 193                 alloc_bootmem_cpumask_var(&cfg[i].domain);
 194                 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
 195                 if (i < NR_IRQS_LEGACY)
 196                         cpumask_setall(cfg[i].domain);
 197         }
 198
 199         return 0;
 200 }
 201
 202 #ifdef CONFIG_SPARSE_IRQ
 203 static struct irq_cfg *irq_cfg(unsigned int irq)
 204 {
 205         struct irq_cfg *cfg = NULL;
 206         struct irq_desc *desc;
 207
 208         desc = irq_to_desc(irq);
 209         if (desc)
 210                 cfg = desc->chip_data;
 211
 212         return cfg;
 213 }
 214
 215 static struct irq_cfg *get_one_free_irq_cfg(int cpu)
 216 {
 217         struct irq_cfg *cfg;
 218         int node;
 219
 220         node = cpu_to_node(cpu);
 221
 222         cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
 223         if (cfg) {
 224                 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
 225                         kfree(cfg);
 226                         cfg = NULL;
 227                 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
 228                                                           GFP_ATOMIC, node)) {
 229                         free_cpumask_var(cfg->domain);
 230                         kfree(cfg);
 231                         cfg = NULL;
 232                 } else {
 233                         cpumask_clear(cfg->domain);
 234                         cpumask_clear(cfg->old_domain);
 235                 }
 236         }
 237
 238         return cfg;
 239 }
 240
 241 int arch_init_chip_data(struct irq_desc *desc, int cpu)
 242 {
 243         struct irq_cfg *cfg;
 244
 245         cfg = desc->chip_data;
 246         if (!cfg) {
 247                 desc->chip_data = get_one_free_irq_cfg(cpu);
 248                 if (!desc->chip_data) {
 249                         printk(KERN_ERR "can not alloc irq_cfg\n");
 250                         BUG_ON(1);
 251                 }
 252         }
 253
 254         return 0;
 255 }
 256
 257 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
 258
 259 static void
 260 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
 261 {
 262         struct irq_pin_list *old_entry, *head, *tail, *entry;
 263
 264         cfg->irq_2_pin = NULL;
 265         old_entry = old_cfg->irq_2_pin;
 266         if (!old_entry)
 267                 return;
 268
 269         entry = get_one_free_irq_2_pin(cpu);
 270         if (!entry)
 271                 return;
 272
 273         entry->apic     = old_entry->apic;
 274         entry->pin      = old_entry->pin;
 275         head            = entry;
 276         tail            = entry;
 277         old_entry       = old_entry->next;
 278         while (old_entry) {
 279                 entry = get_one_free_irq_2_pin(cpu);
 280                 if (!entry) {
 281                         entry = head;
 282                         while (entry) {
 283                                 head = entry->next;
 284                                 kfree(entry);
 285                                 entry = head;
 286                         }
 287                         /* still use the old one */
 288                         return;
 289                 }
 290                 entry->apic     = old_entry->apic;
 291                 entry->pin      = old_entry->pin;
 292                 tail->next      = entry;
 293                 tail            = entry;
 294                 old_entry       = old_entry->next;
 295         }
 296
 297         tail->next = NULL;
 298         cfg->irq_2_pin = head;
 299 }
 300
 301 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
 302 {
 303         struct irq_pin_list *entry, *next;
 304
 305         if (old_cfg->irq_2_pin == cfg->irq_2_pin)
 306                 return;
 307
 308         entry = old_cfg->irq_2_pin;
 309
 310         while (entry) {
 311                 next = entry->next;
 312                 kfree(entry);
 313                 entry = next;
 314         }
 315         old_cfg->irq_2_pin = NULL;
 316 }
 317
 318 void arch_init_copy_chip_data(struct irq_desc *old_desc,
 319                                  struct irq_desc *desc, int cpu)
 320 {
 321         struct irq_cfg *cfg;
 322         struct irq_cfg *old_cfg;
 323
 324         cfg = get_one_free_irq_cfg(cpu);
 325
 326         if (!cfg)
 327                 return;
 328
 329         desc->chip_data = cfg;
 330
 331         old_cfg = old_desc->chip_data;
 332
 333         memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
 334
 335         init_copy_irq_2_pin(old_cfg, cfg, cpu);
 336 }
 337
 338 static void free_irq_cfg(struct irq_cfg *old_cfg)
 339 {
 340         kfree(old_cfg);
 341 }
 342
 343 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
 344 {
 345         struct irq_cfg *old_cfg, *cfg;
 346
 347         old_cfg = old_desc->chip_data;
 348         cfg = desc->chip_data;
 349
 350         if (old_cfg == cfg)
 351                 return;
 352
 353         if (old_cfg) {
 354                 free_irq_2_pin(old_cfg, cfg);
 355                 free_irq_cfg(old_cfg);
 356                 old_desc->chip_data = NULL;
 357         }
 358 }
 359
 360 static void
 361 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
 362 {
 363         struct irq_cfg *cfg = desc->chip_data;
 364
 365         if (!cfg->move_in_progress) {
 366                 /* it means that domain is not changed */
 367                 if (!cpumask_intersects(desc->affinity, mask))
 368                         cfg->move_desc_pending = 1;
 369         }
 370 }
 371 #endif
 372
 373 #else
 374 static struct irq_cfg *irq_cfg(unsigned int irq)
 375 {
 376         return irq < nr_irqs ? irq_cfgx + irq : NULL;
 377 }
 378
 379 #endif
 380
 381 #ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
 382 static inline void
 383 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
 384 {
 385 }
 386 #endif
 387
 388 struct io_apic {
 389         unsigned int index;
 390         unsigned int unused[3];
 391         unsigned int data;
 392 };
 393
 394 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
 395 {
 396         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
 397                 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
 398 }
 399
 400 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
 401 {
 402         struct io_apic __iomem *io_apic = io_apic_base(apic);
 403         writel(reg, &io_apic->index);
 404         return readl(&io_apic->data);
 405 }
 406
 407 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
 408 {
 409         struct io_apic __iomem *io_apic = io_apic_base(apic);
 410         writel(reg, &io_apic->index);
 411         writel(value, &io_apic->data);
 412 }
 413
 414 /*
 415  * Re-write a value: to be used for read-modify-write
 416  * cycles where the read already set up the index register.
 417  *
 418  * Older SiS APIC requires we rewrite the index register
 419  */
 420 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
 421 {
 422         struct io_apic __iomem *io_apic = io_apic_base(apic);
 423
 424         if (sis_apic_bug)
 425                 writel(reg, &io_apic->index);
 426         writel(value, &io_apic->data);
 427 }
 428
 429 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
 430 {
 431         struct irq_pin_list *entry;
 432         unsigned long flags;
 433
 434         spin_lock_irqsave(&ioapic_lock, flags);
 435         entry = cfg->irq_2_pin;
 436         for (;;) {
 437                 unsigned int reg;
 438                 int pin;
 439
 440                 if (!entry)
 441                         break;
 442                 pin = entry->pin;
 443                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
 444                 /* Is the remote IRR bit set? */
 445                 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
 446                         spin_unlock_irqrestore(&ioapic_lock, flags);
 447                         return true;
 448                 }
 449                 if (!entry->next)
 450                         break;
 451                 entry = entry->next;
 452         }
 453         spin_unlock_irqrestore(&ioapic_lock, flags);
 454
 455         return false;
 456 }
 457
 458 union entry_union {
 459         struct { u32 w1, w2; };
 460         struct IO_APIC_route_entry entry;
 461 };
 462
 463 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
 464 {
 465         union entry_union eu;
 466         unsigned long flags;
 467         spin_lock_irqsave(&ioapic_lock, flags);
 468         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
 469         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
 470         spin_unlock_irqrestore(&ioapic_lock, flags);
 471         return eu.entry;
 472 }
 473
 474 /*
 475  * When we write a new IO APIC routing entry, we need to write the high
 476  * word first! If the mask bit in the low word is clear, we will enable
 477  * the interrupt, and we need to make sure the entry is fully populated
 478  * before that happens.
 479  */
 480 static void
 481 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 482 {
 483         union entry_union eu;
 484         eu.entry = e;
 485         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 486         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 487 }
 488
 489 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 490 {
 491         unsigned long flags;
 492         spin_lock_irqsave(&ioapic_lock, flags);
 493         __ioapic_write_entry(apic, pin, e);
 494         spin_unlock_irqrestore(&ioapic_lock, flags);
 495 }
 496
 497 /*
 498  * When we mask an IO APIC routing entry, we need to write the low
 499  * word first, in order to set the mask bit before we change the
 500  * high bits!
 501  */
 502 static void ioapic_mask_entry(int apic, int pin)
 503 {
 504         unsigned long flags;
 505         union entry_union eu = { .entry.mask = 1 };
 506
 507         spin_lock_irqsave(&ioapic_lock, flags);
 508         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 509         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 510         spin_unlock_irqrestore(&ioapic_lock, flags);
 511 }
 512
 513 #ifdef CONFIG_SMP
 514 static void send_cleanup_vector(struct irq_cfg *cfg)
 515 {
 516         cpumask_var_t cleanup_mask;
 517
 518         if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
 519                 unsigned int i;
 520                 cfg->move_cleanup_count = 0;
 521                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
 522                         cfg->move_cleanup_count++;
 523                 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
 524                         apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
 525         } else {
 526                 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
 527                 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
 528                 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
 529                 free_cpumask_var(cleanup_mask);
 530         }
 531         cfg->move_in_progress = 0;
 532 }
 533
 534 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
 535 {
 536         int apic, pin;
 537         struct irq_pin_list *entry;
 538         u8 vector = cfg->vector;
 539
 540         entry = cfg->irq_2_pin;
 541         for (;;) {
 542                 unsigned int reg;
 543
 544                 if (!entry)
 545                         break;
 546
 547                 apic = entry->apic;
 548                 pin = entry->pin;
 549 #ifdef CONFIG_INTR_REMAP
 550                 /*
 551                  * With interrupt-remapping, destination information comes
 552                  * from interrupt-remapping table entry.
 553                  */
 554                 if (!irq_remapped(irq))
 555                         io_apic_write(apic, 0x11 + pin*2, dest);
 556 #else
 557                 io_apic_write(apic, 0x11 + pin*2, dest);
 558 #endif
 559                 reg = io_apic_read(apic, 0x10 + pin*2);
 560                 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
 561                 reg |= vector;
 562                 io_apic_modify(apic, 0x10 + pin*2, reg);
 563                 if (!entry->next)
 564                         break;
 565                 entry = entry->next;
 566         }
 567 }
 568
 569 static int
 570 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
 571
 572 /*
 573  * Either sets desc->affinity to a valid value, and returns
 574  * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
 575  * leaves desc->affinity untouched.
 576  */
 577 static unsigned int
 578 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
 579 {
 580         struct irq_cfg *cfg;
 581         unsigned int irq;
 582
 583         if (!cpumask_intersects(mask, cpu_online_mask))
 584                 return BAD_APICID;
 585
 586         irq = desc->irq;
 587         cfg = desc->chip_data;
 588         if (assign_irq_vector(irq, cfg, mask))
 589                 return BAD_APICID;
 590
 591         cpumask_and(desc->affinity, cfg->domain, mask);
 592         set_extra_move_desc(desc, mask);
 593
 594         return apic->cpu_mask_to_apicid_and(desc->affinity, cpu_online_mask);
 595 }
 596
 597 static void
 598 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
 599 {
 600         struct irq_cfg *cfg;
 601         unsigned long flags;
 602         unsigned int dest;
 603         unsigned int irq;
 604
 605         irq = desc->irq;
 606         cfg = desc->chip_data;
 607
 608         spin_lock_irqsave(&ioapic_lock, flags);
 609         dest = set_desc_affinity(desc, mask);
 610         if (dest != BAD_APICID) {
 611                 /* Only the high 8 bits are valid. */
 612                 dest = SET_APIC_LOGICAL_ID(dest);
 613                 __target_IO_APIC_irq(irq, dest, cfg);
 614         }
 615         spin_unlock_irqrestore(&ioapic_lock, flags);
 616 }
 617
 618 static void
 619 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
 620 {
 621         struct irq_desc *desc;
 622
 623         desc = irq_to_desc(irq);
 624
 625         set_ioapic_affinity_irq_desc(desc, mask);
 626 }
 627 #endif /* CONFIG_SMP */
 628
 629 /*
 630  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
 631  * shared ISA-space IRQs, so we have to support them. We are super
 632  * fast in the common case, and fast for shared ISA-space IRQs.
 633  */
 634 static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
 635 {
 636         struct irq_pin_list *entry;
 637
 638         entry = cfg->irq_2_pin;
 639         if (!entry) {
 640                 entry = get_one_free_irq_2_pin(cpu);
 641                 if (!entry) {
 642                         printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
 643                                         apic, pin);
 644                         return;
 645                 }
 646                 cfg->irq_2_pin = entry;
 647                 entry->apic = apic;
 648                 entry->pin = pin;
 649                 return;
 650         }
 651
 652         while (entry->next) {
 653                 /* not again, please */
 654                 if (entry->apic == apic && entry->pin == pin)
 655                         return;
 656
 657                 entry = entry->next;
 658         }
 659
 660         entry->next = get_one_free_irq_2_pin(cpu);
 661         entry = entry->next;
 662         entry->apic = apic;
 663         entry->pin = pin;
 664 }
 665
 666 /*
 667  * Reroute an IRQ to a different pin.
 668  */
 669 static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
 670                                       int oldapic, int oldpin,
 671                                       int newapic, int newpin)
 672 {
 673         struct irq_pin_list *entry = cfg->irq_2_pin;
 674         int replaced = 0;
 675
 676         while (entry) {
 677                 if (entry->apic == oldapic && entry->pin == oldpin) {
 678                         entry->apic = newapic;
 679                         entry->pin = newpin;
 680                         replaced = 1;
 681                         /* every one is different, right? */
 682                         break;
 683                 }
 684                 entry = entry->next;
 685         }
 686
 687         /* why? call replace before add? */
 688         if (!replaced)
 689                 add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
 690 }
 691
 692 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
 693                                 int mask_and, int mask_or,
 694                                 void (*final)(struct irq_pin_list *entry))
 695 {
 696         int pin;
 697         struct irq_pin_list *entry;
 698
 699         for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
 700                 unsigned int reg;
 701                 pin = entry->pin;
 702                 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
 703                 reg &= mask_and;
 704                 reg |= mask_or;
 705                 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
 706                 if (final)
 707                         final(entry);
 708         }
 709 }
 710
 711 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
 712 {
 713         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
 714 }
 715
 716 #ifdef CONFIG_X86_64
 717 static void io_apic_sync(struct irq_pin_list *entry)
 718 {
 719         /*
 720          * Synchronize the IO-APIC and the CPU by doing
 721          * a dummy read from the IO-APIC
 722          */
 723         struct io_apic __iomem *io_apic;
 724         io_apic = io_apic_base(entry->apic);
 725         readl(&io_apic->data);
 726 }
 727
 728 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
 729 {
 730         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
 731 }
 732 #else /* CONFIG_X86_32 */
 733 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
 734 {
 735         io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
 736 }
 737
 738 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
 739 {
 740         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
 741                         IO_APIC_REDIR_MASKED, NULL);
 742 }
 743
 744 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
 745 {
 746         io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
 747                         IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
 748 }
 749 #endif /* CONFIG_X86_32 */
 750
 751 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
 752 {
 753         struct irq_cfg *cfg = desc->chip_data;
 754         unsigned long flags;
 755
 756         BUG_ON(!cfg);
 757
 758         spin_lock_irqsave(&ioapic_lock, flags);
 759         __mask_IO_APIC_irq(cfg);
 760         spin_unlock_irqrestore(&ioapic_lock, flags);
 761 }
 762
 763 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
 764 {
 765         struct irq_cfg *cfg = desc->chip_data;
 766         unsigned long flags;
 767
 768         spin_lock_irqsave(&ioapic_lock, flags);
 769         __unmask_IO_APIC_irq(cfg);
 770         spin_unlock_irqrestore(&ioapic_lock, flags);
 771 }
 772
 773 static void mask_IO_APIC_irq(unsigned int irq)
 774 {
 775         struct irq_desc *desc = irq_to_desc(irq);
 776
 777         mask_IO_APIC_irq_desc(desc);
 778 }
 779 static void unmask_IO_APIC_irq(unsigned int irq)
 780 {
 781         struct irq_desc *desc = irq_to_desc(irq);
 782
 783         unmask_IO_APIC_irq_desc(desc);
 784 }
 785
 786 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 787 {
 788         struct IO_APIC_route_entry entry;
 789
 790         /* Check delivery_mode to be sure we're not clearing an SMI pin */
 791         entry = ioapic_read_entry(apic, pin);
 792         if (entry.delivery_mode == dest_SMI)
 793                 return;
 794         /*
 795          * Disable it in the IO-APIC irq-routing table:
 796          */
 797         ioapic_mask_entry(apic, pin);
 798 }
 799
 800 static void clear_IO_APIC (void)
 801 {
 802         int apic, pin;
 803
 804         for (apic = 0; apic < nr_ioapics; apic++)
 805                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 806                         clear_IO_APIC_pin(apic, pin);
 807 }
 808
 809 #ifdef CONFIG_X86_32
 810 /*
 811  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
 812  * specific CPU-side IRQs.
 813  */
 814
 815 #define MAX_PIRQS 8
 816 static int pirq_entries[MAX_PIRQS] = {
 817         [0 ... MAX_PIRQS - 1] = -1
 818 };
 819
 820 static int __init ioapic_pirq_setup(char *str)
 821 {
 822         int i, max;
 823         int ints[MAX_PIRQS+1];
 824
 825         get_options(str, ARRAY_SIZE(ints), ints);
 826
 827         apic_printk(APIC_VERBOSE, KERN_INFO
 828                         "PIRQ redirection, working around broken MP-BIOS.\n");
 829         max = MAX_PIRQS;
 830         if (ints[0] < MAX_PIRQS)
 831                 max = ints[0];
 832
 833         for (i = 0; i < max; i++) {
 834                 apic_printk(APIC_VERBOSE, KERN_DEBUG
 835                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
 836                 /*
 837                  * PIRQs are mapped upside down, usually.
 838                  */
 839                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
 840         }
 841         return 1;
 842 }
 843
 844 __setup("pirq=", ioapic_pirq_setup);
 845 #endif /* CONFIG_X86_32 */
 846
 847 #ifdef CONFIG_INTR_REMAP
 848 /* I/O APIC RTE contents at the OS boot up */
 849 static struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
 850
 851 /*
 852  * Saves and masks all the unmasked IO-APIC RTE's
 853  */
 854 int save_mask_IO_APIC_setup(void)
 855 {
 856         union IO_APIC_reg_01 reg_01;
 857         unsigned long flags;
 858         int apic, pin;
 859
 860         /*
 861          * The number of IO-APIC IRQ registers (== #pins):
 862          */
 863         for (apic = 0; apic < nr_ioapics; apic++) {
 864                 spin_lock_irqsave(&ioapic_lock, flags);
 865                 reg_01.raw = io_apic_read(apic, 1);
 866                 spin_unlock_irqrestore(&ioapic_lock, flags);
 867                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
 868         }
 869
 870         for (apic = 0; apic < nr_ioapics; apic++) {
 871                 early_ioapic_entries[apic] =
 872                         kzalloc(sizeof(struct IO_APIC_route_entry) *
 873                                 nr_ioapic_registers[apic], GFP_KERNEL);
 874                 if (!early_ioapic_entries[apic])
 875                         goto nomem;
 876         }
 877
 878         for (apic = 0; apic < nr_ioapics; apic++)
 879                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
 880                         struct IO_APIC_route_entry entry;
 881
 882                         entry = early_ioapic_entries[apic][pin] =
 883                                 ioapic_read_entry(apic, pin);
 884                         if (!entry.mask) {
 885                                 entry.mask = 1;
 886                                 ioapic_write_entry(apic, pin, entry);
 887                         }
 888                 }
 889
 890         return 0;
 891
 892 nomem:
 893         while (apic >= 0)
 894                 kfree(early_ioapic_entries[apic--]);
 895         memset(early_ioapic_entries, 0,
 896                 ARRAY_SIZE(early_ioapic_entries));
 897
 898         return -ENOMEM;
 899 }
 900
 901 void restore_IO_APIC_setup(void)
 902 {
 903         int apic, pin;
 904
 905         for (apic = 0; apic < nr_ioapics; apic++) {
 906                 if (!early_ioapic_entries[apic])
 907                         break;
 908                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 909                         ioapic_write_entry(apic, pin,
 910                                            early_ioapic_entries[apic][pin]);
 911                 kfree(early_ioapic_entries[apic]);
 912                 early_ioapic_entries[apic] = NULL;
 913         }
 914 }
 915
 916 void reinit_intr_remapped_IO_APIC(int intr_remapping)
 917 {
 918         /*
 919          * for now plain restore of previous settings.
 920          * TBD: In the case of OS enabling interrupt-remapping,
 921          * IO-APIC RTE's need to be setup to point to interrupt-remapping
 922          * table entries. for now, do a plain restore, and wait for
 923          * the setup_IO_APIC_irqs() to do proper initialization.
 924          */
 925         restore_IO_APIC_setup();
 926 }
 927 #endif
 928
 929 /*
 930  * Find the IRQ entry number of a certain pin.
 931  */
 932 static int find_irq_entry(int apic, int pin, int type)
 933 {
 934         int i;
 935
 936         for (i = 0; i < mp_irq_entries; i++)
 937                 if (mp_irqs[i].irqtype == type &&
 938                     (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
 939                      mp_irqs[i].dstapic == MP_APIC_ALL) &&
 940                     mp_irqs[i].dstirq == pin)
 941                         return i;
 942
 943         return -1;
 944 }
 945
 946 /*
 947  * Find the pin to which IRQ[irq] (ISA) is connected
 948  */
 949 static int __init find_isa_irq_pin(int irq, int type)
 950 {
 951         int i;
 952
 953         for (i = 0; i < mp_irq_entries; i++) {
 954                 int lbus = mp_irqs[i].srcbus;
 955
 956                 if (test_bit(lbus, mp_bus_not_pci) &&
 957                     (mp_irqs[i].irqtype == type) &&
 958                     (mp_irqs[i].srcbusirq == irq))
 959
 960                         return mp_irqs[i].dstirq;
 961         }
 962         return -1;
 963 }
 964
 965 static int __init find_isa_irq_apic(int irq, int type)
 966 {
 967         int i;
 968
 969         for (i = 0; i < mp_irq_entries; i++) {
 970                 int lbus = mp_irqs[i].srcbus;
 971
 972                 if (test_bit(lbus, mp_bus_not_pci) &&
 973                     (mp_irqs[i].irqtype == type) &&
 974                     (mp_irqs[i].srcbusirq == irq))
 975                         break;
 976         }
 977         if (i < mp_irq_entries) {
 978                 int apic;
 979                 for(apic = 0; apic < nr_ioapics; apic++) {
 980                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
 981                                 return apic;
 982                 }
 983         }
 984
 985         return -1;
 986 }
 987
 988 /*
 989  * Find a specific PCI IRQ entry.
 990  * Not an __init, possibly needed by modules
 991  */
 992 static int pin_2_irq(int idx, int apic, int pin);
 993
 994 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
 995 {
 996         int apic, i, best_guess = -1;
 997
 998         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
 999                 bus, slot, pin);
1000         if (test_bit(bus, mp_bus_not_pci)) {
1001                 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1002                 return -1;
1003         }
1004         for (i = 0; i < mp_irq_entries; i++) {
1005                 int lbus = mp_irqs[i].srcbus;
1006
1007                 for (apic = 0; apic < nr_ioapics; apic++)
1008                         if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1009                             mp_irqs[i].dstapic == MP_APIC_ALL)
1010                                 break;
1011
1012                 if (!test_bit(lbus, mp_bus_not_pci) &&
1013                     !mp_irqs[i].irqtype &&
1014                     (bus == lbus) &&
1015                     (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1016                         int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1017
1018                         if (!(apic || IO_APIC_IRQ(irq)))
1019                                 continue;
1020
1021                         if (pin == (mp_irqs[i].srcbusirq & 3))
1022                                 return irq;
1023                         /*
1024                          * Use the first all-but-pin matching entry as a
1025                          * best-guess fuzzy result for broken mptables.
1026                          */
1027                         if (best_guess < 0)
1028                                 best_guess = irq;
1029                 }
1030         }
1031         return best_guess;
1032 }
1033
1034 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1035
1036 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1037 /*
1038  * EISA Edge/Level control register, ELCR
1039  */
1040 static int EISA_ELCR(unsigned int irq)
1041 {
1042         if (irq < NR_IRQS_LEGACY) {
1043                 unsigned int port = 0x4d0 + (irq >> 3);
1044                 return (inb(port) >> (irq & 7)) & 1;
1045         }
1046         apic_printk(APIC_VERBOSE, KERN_INFO
1047                         "Broken MPtable reports ISA irq %d\n", irq);
1048         return 0;
1049 }
1050
1051 #endif
1052
1053 /* ISA interrupts are always polarity zero edge triggered,
1054  * when listed as conforming in the MP table. */
1055
1056 #define default_ISA_trigger(idx)        (0)
1057 #define default_ISA_polarity(idx)       (0)
1058
1059 /* EISA interrupts are always polarity zero and can be edge or level
1060  * trigger depending on the ELCR value.  If an interrupt is listed as
1061  * EISA conforming in the MP table, that means its trigger type must
1062  * be read in from the ELCR */
1063
1064 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].srcbusirq))
1065 #define default_EISA_polarity(idx)      default_ISA_polarity(idx)
1066
1067 /* PCI interrupts are always polarity one level triggered,
1068  * when listed as conforming in the MP table. */
1069
1070 #define default_PCI_trigger(idx)        (1)
1071 #define default_PCI_polarity(idx)       (1)
1072
1073 /* MCA interrupts are always polarity zero level triggered,
1074  * when listed as conforming in the MP table. */
1075
1076 #define default_MCA_trigger(idx)        (1)
1077 #define default_MCA_polarity(idx)       default_ISA_polarity(idx)
1078
1079 static int MPBIOS_polarity(int idx)
1080 {
1081         int bus = mp_irqs[idx].srcbus;
1082         int polarity;
1083
1084         /*
1085          * Determine IRQ line polarity (high active or low active):
1086          */
1087         switch (mp_irqs[idx].irqflag & 3)
1088         {
1089                 case 0: /* conforms, ie. bus-type dependent polarity */
1090                         if (test_bit(bus, mp_bus_not_pci))
1091                                 polarity = default_ISA_polarity(idx);
1092                         else
1093                                 polarity = default_PCI_polarity(idx);
1094                         break;
1095                 case 1: /* high active */
1096                 {
1097                         polarity = 0;
1098                         break;
1099                 }
1100                 case 2: /* reserved */
1101                 {
1102                         printk(KERN_WARNING "broken BIOS!!\n");
1103                         polarity = 1;
1104                         break;
1105                 }
1106                 case 3: /* low active */
1107                 {
1108                         polarity = 1;
1109                         break;
1110                 }
1111                 default: /* invalid */
1112                 {
1113                         printk(KERN_WARNING "broken BIOS!!\n");
1114                         polarity = 1;
1115                         break;
1116                 }
1117         }
1118         return polarity;
1119 }
1120
1121 static int MPBIOS_trigger(int idx)
1122 {
1123         int bus = mp_irqs[idx].srcbus;
1124         int trigger;
1125
1126         /*
1127          * Determine IRQ trigger mode (edge or level sensitive):
1128          */
1129         switch ((mp_irqs[idx].irqflag>>2) & 3)
1130         {
1131                 case 0: /* conforms, ie. bus-type dependent */
1132                         if (test_bit(bus, mp_bus_not_pci))
1133                                 trigger = default_ISA_trigger(idx);
1134                         else
1135                                 trigger = default_PCI_trigger(idx);
1136 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1137                         switch (mp_bus_id_to_type[bus]) {
1138                                 case MP_BUS_ISA: /* ISA pin */
1139                                 {
1140                                         /* set before the switch */
1141                                         break;
1142                                 }
1143                                 case MP_BUS_EISA: /* EISA pin */
1144                                 {
1145                                         trigger = default_EISA_trigger(idx);
1146                                         break;
1147                                 }
1148                                 case MP_BUS_PCI: /* PCI pin */
1149                                 {
1150                                         /* set before the switch */
1151                                         break;
1152                                 }
1153                                 case MP_BUS_MCA: /* MCA pin */
1154                                 {
1155                                         trigger = default_MCA_trigger(idx);
1156                                         break;
1157                                 }
1158                                 default:
1159                                 {
1160                                         printk(KERN_WARNING "broken BIOS!!\n");
1161                                         trigger = 1;
1162                                         break;
1163                                 }
1164                         }
1165 #endif
1166                         break;
1167                 case 1: /* edge */
1168                 {
1169                         trigger = 0;
1170                         break;
1171                 }
1172                 case 2: /* reserved */
1173                 {
1174                         printk(KERN_WARNING "broken BIOS!!\n");
1175                         trigger = 1;
1176                         break;
1177                 }
1178                 case 3: /* level */
1179                 {
1180                         trigger = 1;
1181                         break;
1182                 }
1183                 default: /* invalid */
1184                 {
1185                         printk(KERN_WARNING "broken BIOS!!\n");
1186                         trigger = 0;
1187                         break;
1188                 }
1189         }
1190         return trigger;
1191 }
1192
1193 static inline int irq_polarity(int idx)
1194 {
1195         return MPBIOS_polarity(idx);
1196 }
1197
1198 static inline int irq_trigger(int idx)
1199 {
1200         return MPBIOS_trigger(idx);
1201 }
1202
1203 int (*ioapic_renumber_irq)(int ioapic, int irq);
1204 static int pin_2_irq(int idx, int apic, int pin)
1205 {
1206         int irq, i;
1207         int bus = mp_irqs[idx].srcbus;
1208
1209         /*
1210          * Debugging check, we are in big trouble if this message pops up!
1211          */
1212         if (mp_irqs[idx].dstirq != pin)
1213                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1214
1215         if (test_bit(bus, mp_bus_not_pci)) {
1216                 irq = mp_irqs[idx].srcbusirq;
1217         } else {
1218                 /*
1219                  * PCI IRQs are mapped in order
1220                  */
1221                 i = irq = 0;
1222                 while (i < apic)
1223                         irq += nr_ioapic_registers[i++];
1224                 irq += pin;
1225                 /*
1226                  * For MPS mode, so far only needed by ES7000 platform
1227                  */
1228                 if (ioapic_renumber_irq)
1229                         irq = ioapic_renumber_irq(apic, irq);
1230         }
1231
1232 #ifdef CONFIG_X86_32
1233         /*
1234          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1235          */
1236         if ((pin >= 16) && (pin <= 23)) {
1237                 if (pirq_entries[pin-16] != -1) {
1238                         if (!pirq_entries[pin-16]) {
1239                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1240                                                 "disabling PIRQ%d\n", pin-16);
1241                         } else {
1242                                 irq = pirq_entries[pin-16];
1243                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1244                                                 "using PIRQ%d -> IRQ %d\n",
1245                                                 pin-16, irq);
1246                         }
1247                 }
1248         }
1249 #endif
1250
1251         return irq;
1252 }
1253
1254 void lock_vector_lock(void)
1255 {
1256         /* Used to the online set of cpus does not change
1257          * during assign_irq_vector.
1258          */
1259         spin_lock(&vector_lock);
1260 }
1261
1262 void unlock_vector_lock(void)
1263 {
1264         spin_unlock(&vector_lock);
1265 }
1266
1267 static int
1268 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1269 {
1270         /*
1271          * NOTE! The local APIC isn't very good at handling
1272          * multiple interrupts at the same interrupt level.
1273          * As the interrupt level is determined by taking the
1274          * vector number and shifting that right by 4, we
1275          * want to spread these out a bit so that they don't
1276          * all fall in the same interrupt level.
1277          *
1278          * Also, we've got to be careful not to trash gate
1279          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1280          */
1281         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1282         unsigned int old_vector;
1283         int cpu, err;
1284         cpumask_var_t tmp_mask;
1285
1286         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1287                 return -EBUSY;
1288
1289         if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1290                 return -ENOMEM;
1291
1292         old_vector = cfg->vector;
1293         if (old_vector) {
1294                 cpumask_and(tmp_mask, mask, cpu_online_mask);
1295                 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1296                 if (!cpumask_empty(tmp_mask)) {
1297                         free_cpumask_var(tmp_mask);
1298                         return 0;
1299                 }
1300         }
1301
1302         /* Only try and allocate irqs on cpus that are present */
1303         err = -ENOSPC;
1304         for_each_cpu_and(cpu, mask, cpu_online_mask) {
1305                 int new_cpu;
1306                 int vector, offset;
1307
1308                 apic->vector_allocation_domain(cpu, tmp_mask);
1309
1310                 vector = current_vector;
1311                 offset = current_offset;
1312 next:
1313                 vector += 8;
1314                 if (vector >= first_system_vector) {
1315                         /* If out of vectors on large boxen, must share them. */
1316                         offset = (offset + 1) % 8;
1317                         vector = FIRST_DEVICE_VECTOR + offset;
1318                 }
1319                 if (unlikely(current_vector == vector))
1320                         continue;
1321
1322                 if (test_bit(vector, used_vectors))
1323                         goto next;
1324
1325                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1326                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1327                                 goto next;
1328                 /* Found one! */
1329                 current_vector = vector;
1330                 current_offset = offset;
1331                 if (old_vector) {
1332                         cfg->move_in_progress = 1;
1333                         cpumask_copy(cfg->old_domain, cfg->domain);
1334                 }
1335                 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1336                         per_cpu(vector_irq, new_cpu)[vector] = irq;
1337                 cfg->vector = vector;
1338                 cpumask_copy(cfg->domain, tmp_mask);
1339                 err = 0;
1340                 break;
1341         }
1342         free_cpumask_var(tmp_mask);
1343         return err;
1344 }
1345
1346 static int
1347 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1348 {
1349         int err;
1350         unsigned long flags;
1351
1352         spin_lock_irqsave(&vector_lock, flags);
1353         err = __assign_irq_vector(irq, cfg, mask);
1354         spin_unlock_irqrestore(&vector_lock, flags);
1355         return err;
1356 }
1357
1358 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1359 {
1360         int cpu, vector;
1361
1362         BUG_ON(!cfg->vector);
1363
1364         vector = cfg->vector;
1365         for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1366                 per_cpu(vector_irq, cpu)[vector] = -1;
1367
1368         cfg->vector = 0;
1369         cpumask_clear(cfg->domain);
1370
1371         if (likely(!cfg->move_in_progress))
1372                 return;
1373         for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1374                 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1375                                                                 vector++) {
1376                         if (per_cpu(vector_irq, cpu)[vector] != irq)
1377                                 continue;
1378                         per_cpu(vector_irq, cpu)[vector] = -1;
1379                         break;
1380                 }
1381         }
1382         cfg->move_in_progress = 0;
1383 }
1384
1385 void __setup_vector_irq(int cpu)
1386 {
1387         /* Initialize vector_irq on a new cpu */
1388         /* This function must be called with vector_lock held */
1389         int irq, vector;
1390         struct irq_cfg *cfg;
1391         struct irq_desc *desc;
1392
1393         /* Mark the inuse vectors */
1394         for_each_irq_desc(irq, desc) {
1395                 cfg = desc->chip_data;
1396                 if (!cpumask_test_cpu(cpu, cfg->domain))
1397                         continue;
1398                 vector = cfg->vector;
1399                 per_cpu(vector_irq, cpu)[vector] = irq;
1400         }
1401         /* Mark the free vectors */
1402         for (vector = 0; vector < NR_VECTORS; ++vector) {
1403                 irq = per_cpu(vector_irq, cpu)[vector];
1404                 if (irq < 0)
1405                         continue;
1406
1407                 cfg = irq_cfg(irq);
1408                 if (!cpumask_test_cpu(cpu, cfg->domain))
1409                         per_cpu(vector_irq, cpu)[vector] = -1;
1410         }
1411 }
1412
1413 static struct irq_chip ioapic_chip;
1414 #ifdef CONFIG_INTR_REMAP
1415 static struct irq_chip ir_ioapic_chip;
1416 #endif
1417
1418 #define IOAPIC_AUTO     -1
1419 #define IOAPIC_EDGE     0
1420 #define IOAPIC_LEVEL    1
1421
1422 #ifdef CONFIG_X86_32
1423 static inline int IO_APIC_irq_trigger(int irq)
1424 {
1425         int apic, idx, pin;
1426
1427         for (apic = 0; apic < nr_ioapics; apic++) {
1428                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1429                         idx = find_irq_entry(apic, pin, mp_INT);
1430                         if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1431                                 return irq_trigger(idx);
1432                 }
1433         }
1434         /*
1435          * nonexistent IRQs are edge default
1436          */
1437         return 0;
1438 }
1439 #else
1440 static inline int IO_APIC_irq_trigger(int irq)
1441 {
1442         return 1;
1443 }
1444 #endif
1445
1446 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1447 {
1448
1449         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1450             trigger == IOAPIC_LEVEL)
1451                 desc->status |= IRQ_LEVEL;
1452         else
1453                 desc->status &= ~IRQ_LEVEL;
1454
1455 #ifdef CONFIG_INTR_REMAP
1456         if (irq_remapped(irq)) {
1457                 desc->status |= IRQ_MOVE_PCNTXT;
1458                 if (trigger)
1459                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1460                                                       handle_fasteoi_irq,
1461                                                      "fasteoi");
1462                 else
1463                         set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1464                                                       handle_edge_irq, "edge");
1465                 return;
1466         }
1467 #endif
1468         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1469             trigger == IOAPIC_LEVEL)
1470                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1471                                               handle_fasteoi_irq,
1472                                               "fasteoi");
1473         else
1474                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1475                                               handle_edge_irq, "edge");
1476 }
1477
1478 int setup_ioapic_entry(int apic_id, int irq,
1479                        struct IO_APIC_route_entry *entry,
1480                        unsigned int destination, int trigger,
1481                        int polarity, int vector)
1482 {
1483         /*
1484          * add it to the IO-APIC irq-routing table:
1485          */
1486         memset(entry,0,sizeof(*entry));
1487
1488 #ifdef CONFIG_INTR_REMAP
1489         if (intr_remapping_enabled) {
1490                 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1491                 struct irte irte;
1492                 struct IR_IO_APIC_route_entry *ir_entry =
1493                         (struct IR_IO_APIC_route_entry *) entry;
1494                 int index;
1495
1496                 if (!iommu)
1497                         panic("No mapping iommu for ioapic %d\n", apic_id);
1498
1499                 index = alloc_irte(iommu, irq, 1);
1500                 if (index < 0)
1501                         panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1502
1503                 memset(&irte, 0, sizeof(irte));
1504
1505                 irte.present = 1;
1506                 irte.dst_mode = apic->irq_dest_mode;
1507                 irte.trigger_mode = trigger;
1508                 irte.dlvry_mode = apic->irq_delivery_mode;
1509                 irte.vector = vector;
1510                 irte.dest_id = IRTE_DEST(destination);
1511
1512                 modify_irte(irq, &irte);
1513
1514                 ir_entry->index2 = (index >> 15) & 0x1;
1515                 ir_entry->zero = 0;
1516                 ir_entry->format = 1;
1517                 ir_entry->index = (index & 0x7fff);
1518         } else
1519 #endif
1520         {
1521                 entry->delivery_mode = apic->irq_delivery_mode;
1522                 entry->dest_mode = apic->irq_dest_mode;
1523                 entry->dest = destination;
1524         }
1525
1526         entry->mask = 0;                                /* enable IRQ */
1527         entry->trigger = trigger;
1528         entry->polarity = polarity;
1529         entry->vector = vector;
1530
1531         /* Mask level triggered irqs.
1532          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1533          */
1534         if (trigger)
1535                 entry->mask = 1;
1536         return 0;
1537 }
1538
1539 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1540                               int trigger, int polarity)
1541 {
1542         struct irq_cfg *cfg;
1543         struct IO_APIC_route_entry entry;
1544         unsigned int dest;
1545
1546         if (!IO_APIC_IRQ(irq))
1547                 return;
1548
1549         cfg = desc->chip_data;
1550
1551         if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1552                 return;
1553
1554         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1555
1556         apic_printk(APIC_VERBOSE,KERN_DEBUG
1557                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1558                     "IRQ %d Mode:%i Active:%i)\n",
1559                     apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1560                     irq, trigger, polarity);
1561
1562
1563         if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1564                                dest, trigger, polarity, cfg->vector)) {
1565                 printk("Failed to setup ioapic entry for ioapic  %d, pin %d\n",
1566                        mp_ioapics[apic_id].apicid, pin);
1567                 __clear_irq_vector(irq, cfg);
1568                 return;
1569         }
1570
1571         ioapic_register_intr(irq, desc, trigger);
1572         if (irq < NR_IRQS_LEGACY)
1573                 disable_8259A_irq(irq);
1574
1575         ioapic_write_entry(apic_id, pin, entry);
1576 }
1577
1578 static void __init setup_IO_APIC_irqs(void)
1579 {
1580         int apic_id, pin, idx, irq;
1581         int notcon = 0;
1582         struct irq_desc *desc;
1583         struct irq_cfg *cfg;
1584         int cpu = boot_cpu_id;
1585
1586         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1587
1588         for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
1589                 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1590
1591                         idx = find_irq_entry(apic_id, pin, mp_INT);
1592                         if (idx == -1) {
1593                                 if (!notcon) {
1594                                         notcon = 1;
1595                                         apic_printk(APIC_VERBOSE,
1596                                                 KERN_DEBUG " %d-%d",
1597                                                 mp_ioapics[apic_id].apicid, pin);
1598                                 } else
1599                                         apic_printk(APIC_VERBOSE, " %d-%d",
1600                                                 mp_ioapics[apic_id].apicid, pin);
1601                                 continue;
1602                         }
1603                         if (notcon) {
1604                                 apic_printk(APIC_VERBOSE,
1605                                         " (apicid-pin) not connected\n");
1606                                 notcon = 0;
1607                         }
1608
1609                         irq = pin_2_irq(idx, apic_id, pin);
1610
1611                         /*
1612                          * Skip the timer IRQ if there's a quirk handler
1613                          * installed and if it returns 1:
1614                          */
1615                         if (apic->multi_timer_check &&
1616                                         apic->multi_timer_check(apic_id, irq))
1617                                 continue;
1618
1619                         desc = irq_to_desc_alloc_cpu(irq, cpu);
1620                         if (!desc) {
1621                                 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1622                                 continue;
1623                         }
1624                         cfg = desc->chip_data;
1625                         add_pin_to_irq_cpu(cfg, cpu, apic_id, pin);
1626
1627                         setup_IO_APIC_irq(apic_id, pin, irq, desc,
1628                                         irq_trigger(idx), irq_polarity(idx));
1629                 }
1630         }
1631
1632         if (notcon)
1633                 apic_printk(APIC_VERBOSE,
1634                         " (apicid-pin) not connected\n");
1635 }
1636
1637 /*
1638  * Set up the timer pin, possibly with the 8259A-master behind.
1639  */
1640 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1641                                         int vector)
1642 {
1643         struct IO_APIC_route_entry entry;
1644
1645 #ifdef CONFIG_INTR_REMAP
1646         if (intr_remapping_enabled)
1647                 return;
1648 #endif
1649
1650         memset(&entry, 0, sizeof(entry));
1651
1652         /*
1653          * We use logical delivery to get the timer IRQ
1654          * to the first CPU.
1655          */
1656         entry.dest_mode = apic->irq_dest_mode;
1657         entry.mask = 0;                 /* don't mask IRQ for edge */
1658         entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1659         entry.delivery_mode = apic->irq_delivery_mode;
1660         entry.polarity = 0;
1661         entry.trigger = 0;
1662         entry.vector = vector;
1663
1664         /*
1665          * The timer IRQ doesn't have to know that behind the
1666          * scene we may have a 8259A-master in AEOI mode ...
1667          */
1668         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1669
1670         /*
1671          * Add it to the IO-APIC irq-routing table:
1672          */
1673         ioapic_write_entry(apic_id, pin, entry);
1674 }
1675
1676
1677 __apicdebuginit(void) print_IO_APIC(void)
1678 {
1679         int apic, i;
1680         union IO_APIC_reg_00 reg_00;
1681         union IO_APIC_reg_01 reg_01;
1682         union IO_APIC_reg_02 reg_02;
1683         union IO_APIC_reg_03 reg_03;
1684         unsigned long flags;
1685         struct irq_cfg *cfg;
1686         struct irq_desc *desc;
1687         unsigned int irq;
1688
1689         if (apic_verbosity == APIC_QUIET)
1690                 return;
1691
1692         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1693         for (i = 0; i < nr_ioapics; i++)
1694                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1695                        mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1696
1697         /*
1698          * We are a bit conservative about what we expect.  We have to
1699          * know about every hardware change ASAP.
1700          */
1701         printk(KERN_INFO "testing the IO APIC.......................\n");
1702
1703         for (apic = 0; apic < nr_ioapics; apic++) {
1704
1705         spin_lock_irqsave(&ioapic_lock, flags);
1706         reg_00.raw = io_apic_read(apic, 0);
1707         reg_01.raw = io_apic_read(apic, 1);
1708         if (reg_01.bits.version >= 0x10)
1709                 reg_02.raw = io_apic_read(apic, 2);
1710         if (reg_01.bits.version >= 0x20)
1711                 reg_03.raw = io_apic_read(apic, 3);
1712         spin_unlock_irqrestore(&ioapic_lock, flags);
1713
1714         printk("\n");
1715         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1716         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1717         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1718         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1719         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1720
1721         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1722         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1723
1724         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1725         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1726
1727         /*
1728          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1729          * but the value of reg_02 is read as the previous read register
1730          * value, so ignore it if reg_02 == reg_01.
1731          */
1732         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1733                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1734                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1735         }
1736
1737         /*
1738          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1739          * or reg_03, but the value of reg_0[23] is read as the previous read
1740          * register value, so ignore it if reg_03 == reg_0[12].
1741          */
1742         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1743             reg_03.raw != reg_01.raw) {
1744                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1745                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1746         }
1747
1748         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1749
1750         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1751                           " Stat Dmod Deli Vect:   \n");
1752
1753         for (i = 0; i <= reg_01.bits.entries; i++) {
1754                 struct IO_APIC_route_entry entry;
1755
1756                 entry = ioapic_read_entry(apic, i);
1757
1758                 printk(KERN_DEBUG " %02x %03X ",
1759                         i,
1760                         entry.dest
1761                 );
1762
1763                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1764                         entry.mask,
1765                         entry.trigger,
1766                         entry.irr,
1767                         entry.polarity,
1768                         entry.delivery_status,
1769                         entry.dest_mode,
1770                         entry.delivery_mode,
1771                         entry.vector
1772                 );
1773         }
1774         }
1775         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1776         for_each_irq_desc(irq, desc) {
1777                 struct irq_pin_list *entry;
1778
1779                 cfg = desc->chip_data;
1780                 entry = cfg->irq_2_pin;
1781                 if (!entry)
1782                         continue;
1783                 printk(KERN_DEBUG "IRQ%d ", irq);
1784                 for (;;) {
1785                         printk("-> %d:%d", entry->apic, entry->pin);
1786                         if (!entry->next)
1787                                 break;
1788                         entry = entry->next;
1789                 }
1790                 printk("\n");
1791         }
1792
1793         printk(KERN_INFO ".................................... done.\n");
1794
1795         return;
1796 }
1797
1798 __apicdebuginit(void) print_APIC_bitfield(int base)
1799 {
1800         unsigned int v;
1801         int i, j;
1802
1803         if (apic_verbosity == APIC_QUIET)
1804                 return;
1805
1806         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1807         for (i = 0; i < 8; i++) {
1808                 v = apic_read(base + i*0x10);
1809                 for (j = 0; j < 32; j++) {
1810                         if (v & (1<<j))
1811                                 printk("1");
1812                         else
1813                                 printk("0");
1814                 }
1815                 printk("\n");
1816         }
1817 }
1818
1819 __apicdebuginit(void) print_local_APIC(void *dummy)
1820 {
1821         unsigned int v, ver, maxlvt;
1822         u64 icr;
1823
1824         if (apic_verbosity == APIC_QUIET)
1825                 return;
1826
1827         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1828                 smp_processor_id(), hard_smp_processor_id());
1829         v = apic_read(APIC_ID);
1830         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, read_apic_id());
1831         v = apic_read(APIC_LVR);
1832         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1833         ver = GET_APIC_VERSION(v);
1834         maxlvt = lapic_get_maxlvt();
1835
1836         v = apic_read(APIC_TASKPRI);
1837         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1838
1839         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1840                 if (!APIC_XAPIC(ver)) {
1841                         v = apic_read(APIC_ARBPRI);
1842                         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1843                                v & APIC_ARBPRI_MASK);
1844                 }
1845                 v = apic_read(APIC_PROCPRI);
1846                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1847         }
1848
1849         /*
1850          * Remote read supported only in the 82489DX and local APIC for
1851          * Pentium processors.
1852          */
1853         if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1854                 v = apic_read(APIC_RRR);
1855                 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1856         }
1857
1858         v = apic_read(APIC_LDR);
1859         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1860         if (!x2apic_enabled()) {
1861                 v = apic_read(APIC_DFR);
1862                 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1863         }
1864         v = apic_read(APIC_SPIV);
1865         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1866
1867         printk(KERN_DEBUG "... APIC ISR field:\n");
1868         print_APIC_bitfield(APIC_ISR);
1869         printk(KERN_DEBUG "... APIC TMR field:\n");
1870         print_APIC_bitfield(APIC_TMR);
1871         printk(KERN_DEBUG "... APIC IRR field:\n");
1872         print_APIC_bitfield(APIC_IRR);
1873
1874         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1875                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1876                         apic_write(APIC_ESR, 0);
1877
1878                 v = apic_read(APIC_ESR);
1879                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1880         }
1881
1882         icr = apic_icr_read();
1883         printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1884         printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1885
1886         v = apic_read(APIC_LVTT);
1887         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1888
1889         if (maxlvt > 3) {                       /* PC is LVT#4. */
1890                 v = apic_read(APIC_LVTPC);
1891                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1892         }
1893         v = apic_read(APIC_LVT0);
1894         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1895         v = apic_read(APIC_LVT1);
1896         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1897
1898         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1899                 v = apic_read(APIC_LVTERR);
1900                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1901         }
1902
1903         v = apic_read(APIC_TMICT);
1904         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1905         v = apic_read(APIC_TMCCT);
1906         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1907         v = apic_read(APIC_TDCR);
1908         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1909         printk("\n");
1910 }
1911
1912 __apicdebuginit(void) print_all_local_APICs(void)
1913 {
1914         int cpu;
1915
1916         preempt_disable();
1917         for_each_online_cpu(cpu)
1918                 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1919         preempt_enable();
1920 }
1921
1922 __apicdebuginit(void) print_PIC(void)
1923 {
1924         unsigned int v;
1925         unsigned long flags;
1926
1927         if (apic_verbosity == APIC_QUIET)
1928                 return;
1929
1930         printk(KERN_DEBUG "\nprinting PIC contents\n");
1931
1932         spin_lock_irqsave(&i8259A_lock, flags);
1933
1934         v = inb(0xa1) << 8 | inb(0x21);
1935         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1936
1937         v = inb(0xa0) << 8 | inb(0x20);
1938         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1939
1940         outb(0x0b,0xa0);
1941         outb(0x0b,0x20);
1942         v = inb(0xa0) << 8 | inb(0x20);
1943         outb(0x0a,0xa0);
1944         outb(0x0a,0x20);
1945
1946         spin_unlock_irqrestore(&i8259A_lock, flags);
1947
1948         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1949
1950         v = inb(0x4d1) << 8 | inb(0x4d0);
1951         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1952 }
1953
1954 __apicdebuginit(int) print_all_ICs(void)
1955 {
1956         print_PIC();
1957         print_all_local_APICs();
1958         print_IO_APIC();
1959
1960         return 0;
1961 }
1962
1963 fs_initcall(print_all_ICs);
1964
1965
1966 /* Where if anywhere is the i8259 connect in external int mode */
1967 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1968
1969 void __init enable_IO_APIC(void)
1970 {
1971         union IO_APIC_reg_01 reg_01;
1972         int i8259_apic, i8259_pin;
1973         int apic;
1974         unsigned long flags;
1975
1976         /*
1977          * The number of IO-APIC IRQ registers (== #pins):
1978          */
1979         for (apic = 0; apic < nr_ioapics; apic++) {
1980                 spin_lock_irqsave(&ioapic_lock, flags);
1981                 reg_01.raw = io_apic_read(apic, 1);
1982                 spin_unlock_irqrestore(&ioapic_lock, flags);
1983                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1984         }
1985         for(apic = 0; apic < nr_ioapics; apic++) {
1986                 int pin;
1987                 /* See if any of the pins is in ExtINT mode */
1988                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1989                         struct IO_APIC_route_entry entry;
1990                         entry = ioapic_read_entry(apic, pin);
1991
1992                         /* If the interrupt line is enabled and in ExtInt mode
1993                          * I have found the pin where the i8259 is connected.
1994                          */
1995                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1996                                 ioapic_i8259.apic = apic;
1997                                 ioapic_i8259.pin  = pin;
1998                                 goto found_i8259;
1999                         }
2000                 }
2001         }
2002  found_i8259:
2003         /* Look to see what if the MP table has reported the ExtINT */
2004         /* If we could not find the appropriate pin by looking at the ioapic
2005          * the i8259 probably is not connected the ioapic but give the
2006          * mptable a chance anyway.
2007          */
2008         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
2009         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
2010         /* Trust the MP table if nothing is setup in the hardware */
2011         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
2012                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
2013                 ioapic_i8259.pin  = i8259_pin;
2014                 ioapic_i8259.apic = i8259_apic;
2015         }
2016         /* Complain if the MP table and the hardware disagree */
2017         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2018                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2019         {
2020                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
2021         }
2022
2023         /*
2024          * Do not trust the IO-APIC being empty at bootup
2025          */
2026         clear_IO_APIC();
2027 }
2028
2029 /*
2030  * Not an __init, needed by the reboot code
2031  */
2032 void disable_IO_APIC(void)
2033 {
2034         /*
2035          * Clear the IO-APIC before rebooting:
2036          */
2037         clear_IO_APIC();
2038
2039         /*
2040          * If the i8259 is routed through an IOAPIC
2041          * Put that IOAPIC in virtual wire mode
2042          * so legacy interrupts can be delivered.
2043          */
2044         if (ioapic_i8259.pin != -1) {
2045                 struct IO_APIC_route_entry entry;
2046
2047                 memset(&entry, 0, sizeof(entry));
2048                 entry.mask            = 0; /* Enabled */
2049                 entry.trigger         = 0; /* Edge */
2050                 entry.irr             = 0;
2051                 entry.polarity        = 0; /* High */
2052                 entry.delivery_status = 0;
2053                 entry.dest_mode       = 0; /* Physical */
2054                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
2055                 entry.vector          = 0;
2056                 entry.dest            = read_apic_id();
2057
2058                 /*
2059                  * Add it to the IO-APIC irq-routing table:
2060                  */
2061                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2062         }
2063
2064         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
2065 }
2066
2067 #ifdef CONFIG_X86_32
2068 /*
2069  * function to set the IO-APIC physical IDs based on the
2070  * values stored in the MPC table.
2071  *
2072  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
2073  */
2074
2075 static void __init setup_ioapic_ids_from_mpc(void)
2076 {
2077         union IO_APIC_reg_00 reg_00;
2078         physid_mask_t phys_id_present_map;
2079         int apic_id;
2080         int i;
2081         unsigned char old_id;
2082         unsigned long flags;
2083
2084         if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2085                 return;
2086
2087         /*
2088          * Don't check I/O APIC IDs for xAPIC systems.  They have
2089          * no meaning without the serial APIC bus.
2090          */
2091         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2092                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2093                 return;
2094         /*
2095          * This is broken; anything with a real cpu count has to
2096          * circumvent this idiocy regardless.
2097          */
2098         phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2099
2100         /*
2101          * Set the IOAPIC ID to the value stored in the MPC table.
2102          */
2103         for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2104
2105                 /* Read the register 0 value */
2106                 spin_lock_irqsave(&ioapic_lock, flags);
2107                 reg_00.raw = io_apic_read(apic_id, 0);
2108                 spin_unlock_irqrestore(&ioapic_lock, flags);
2109
2110                 old_id = mp_ioapics[apic_id].apicid;
2111
2112                 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2113                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2114                                 apic_id, mp_ioapics[apic_id].apicid);
2115                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2116                                 reg_00.bits.ID);
2117                         mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2118                 }
2119
2120                 /*
2121                  * Sanity check, is the ID really free? Every APIC in a
2122                  * system must have a unique ID or we get lots of nice
2123                  * 'stuck on smp_invalidate_needed IPI wait' messages.
2124                  */
2125                 if (apic->check_apicid_used(phys_id_present_map,
2126                                         mp_ioapics[apic_id].apicid)) {
2127                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2128                                 apic_id, mp_ioapics[apic_id].apicid);
2129                         for (i = 0; i < get_physical_broadcast(); i++)
2130                                 if (!physid_isset(i, phys_id_present_map))
2131                                         break;
2132                         if (i >= get_physical_broadcast())
2133                                 panic("Max APIC ID exceeded!\n");
2134                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2135                                 i);
2136                         physid_set(i, phys_id_present_map);
2137                         mp_ioapics[apic_id].apicid = i;
2138                 } else {
2139                         physid_mask_t tmp;
2140                         tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2141                         apic_printk(APIC_VERBOSE, "Setting %d in the "
2142                                         "phys_id_present_map\n",
2143                                         mp_ioapics[apic_id].apicid);
2144                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
2145                 }
2146
2147
2148                 /*
2149                  * We need to adjust the IRQ routing table
2150                  * if the ID changed.
2151                  */
2152                 if (old_id != mp_ioapics[apic_id].apicid)
2153                         for (i = 0; i < mp_irq_entries; i++)
2154                                 if (mp_irqs[i].dstapic == old_id)
2155                                         mp_irqs[i].dstapic
2156                                                 = mp_ioapics[apic_id].apicid;
2157
2158                 /*
2159                  * Read the right value from the MPC table and
2160                  * write it into the ID register.
2161                  */
2162                 apic_printk(APIC_VERBOSE, KERN_INFO
2163                         "...changing IO-APIC physical APIC ID to %d ...",
2164                         mp_ioapics[apic_id].apicid);
2165
2166                 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2167                 spin_lock_irqsave(&ioapic_lock, flags);
2168                 io_apic_write(apic_id, 0, reg_00.raw);
2169                 spin_unlock_irqrestore(&ioapic_lock, flags);
2170
2171                 /*
2172                  * Sanity check
2173                  */
2174                 spin_lock_irqsave(&ioapic_lock, flags);
2175                 reg_00.raw = io_apic_read(apic_id, 0);
2176                 spin_unlock_irqrestore(&ioapic_lock, flags);
2177                 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2178                         printk("could not set ID!\n");
2179                 else
2180                         apic_printk(APIC_VERBOSE, " ok.\n");
2181         }
2182 }
2183 #endif
2184
2185 int no_timer_check __initdata;
2186
2187 static int __init notimercheck(char *s)
2188 {
2189         no_timer_check = 1;
2190         return 1;
2191 }
2192 __setup("no_timer_check", notimercheck);
2193
2194 /*
2195  * There is a nasty bug in some older SMP boards, their mptable lies
2196  * about the timer IRQ. We do the following to work around the situation:
2197  *
2198  *      - timer IRQ defaults to IO-APIC IRQ
2199  *      - if this function detects that timer IRQs are defunct, then we fall
2200  *        back to ISA timer IRQs
2201  */
2202 static int __init timer_irq_works(void)
2203 {
2204         unsigned long t1 = jiffies;
2205         unsigned long flags;
2206
2207         if (no_timer_check)
2208                 return 1;
2209
2210         local_save_flags(flags);
2211         local_irq_enable();
2212         /* Let ten ticks pass... */
2213         mdelay((10 * 1000) / HZ);
2214         local_irq_restore(flags);
2215
2216         /*
2217          * Expect a few ticks at least, to be sure some possible
2218          * glue logic does not lock up after one or two first
2219          * ticks in a non-ExtINT mode.  Also the local APIC
2220          * might have cached one ExtINT interrupt.  Finally, at
2221          * least one tick may be lost due to delays.
2222          */
2223
2224         /* jiffies wrap? */
2225         if (time_after(jiffies, t1 + 4))
2226                 return 1;
2227         return 0;
2228 }
2229
2230 /*
2231  * In the SMP+IOAPIC case it might happen that there are an unspecified
2232  * number of pending IRQ events unhandled. These cases are very rare,
2233  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2234  * better to do it this way as thus we do not have to be aware of
2235  * 'pending' interrupts in the IRQ path, except at this point.
2236  */
2237 /*
2238  * Edge triggered needs to resend any interrupt
2239  * that was delayed but this is now handled in the device
2240  * independent code.
2241  */
2242
2243 /*
2244  * Starting up a edge-triggered IO-APIC interrupt is
2245  * nasty - we need to make sure that we get the edge.
2246  * If it is already asserted for some reason, we need
2247  * return 1 to indicate that is was pending.
2248  *
2249  * This is not complete - we should be able to fake
2250  * an edge even if it isn't on the 8259A...
2251  */
2252
2253 static unsigned int startup_ioapic_irq(unsigned int irq)
2254 {
2255         int was_pending = 0;
2256         unsigned long flags;
2257         struct irq_cfg *cfg;
2258
2259         spin_lock_irqsave(&ioapic_lock, flags);
2260         if (irq < NR_IRQS_LEGACY) {
2261                 disable_8259A_irq(irq);
2262                 if (i8259A_irq_pending(irq))
2263                         was_pending = 1;
2264         }
2265         cfg = irq_cfg(irq);
2266         __unmask_IO_APIC_irq(cfg);
2267         spin_unlock_irqrestore(&ioapic_lock, flags);
2268
2269         return was_pending;
2270 }
2271
2272 #ifdef CONFIG_X86_64
2273 static int ioapic_retrigger_irq(unsigned int irq)
2274 {
2275
2276         struct irq_cfg *cfg = irq_cfg(irq);
2277         unsigned long flags;
2278
2279         spin_lock_irqsave(&vector_lock, flags);
2280         apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2281         spin_unlock_irqrestore(&vector_lock, flags);
2282
2283         return 1;
2284 }
2285 #else
2286 static int ioapic_retrigger_irq(unsigned int irq)
2287 {
2288         apic->send_IPI_self(irq_cfg(irq)->vector);
2289
2290         return 1;
2291 }
2292 #endif
2293
2294 /*
2295  * Level and edge triggered IO-APIC interrupts need different handling,
2296  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2297  * handled with the level-triggered descriptor, but that one has slightly
2298  * more overhead. Level-triggered interrupts cannot be handled with the
2299  * edge-triggered handler, without risking IRQ storms and other ugly
2300  * races.
2301  */
2302
2303 #ifdef CONFIG_SMP
2304
2305 #ifdef CONFIG_INTR_REMAP
2306 static void ir_irq_migration(struct work_struct *work);
2307
2308 static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
2309
2310 /*
2311  * Migrate the IO-APIC irq in the presence of intr-remapping.
2312  *
2313  * For edge triggered, irq migration is a simple atomic update(of vector
2314  * and cpu destination) of IRTE and flush the hardware cache.
2315  *
2316  * For level triggered, we need to modify the io-apic RTE aswell with the update
2317  * vector information, along with modifying IRTE with vector and destination.
2318  * So irq migration for level triggered is little  bit more complex compared to
2319  * edge triggered migration. But the good news is, we use the same algorithm
2320  * for level triggered migration as we have today, only difference being,
2321  * we now initiate the irq migration from process context instead of the
2322  * interrupt context.
2323  *
2324  * In future, when we do a directed EOI (combined with cpu EOI broadcast
2325  * suppression) to the IO-APIC, level triggered irq migration will also be
2326  * as simple as edge triggered migration and we can do the irq migration
2327  * with a simple atomic update to IO-APIC RTE.
2328  */
2329 static void
2330 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2331 {
2332         struct irq_cfg *cfg;
2333         struct irte irte;
2334         int modify_ioapic_rte;
2335         unsigned int dest;
2336         unsigned long flags;
2337         unsigned int irq;
2338
2339         if (!cpumask_intersects(mask, cpu_online_mask))
2340                 return;
2341
2342         irq = desc->irq;
2343         if (get_irte(irq, &irte))
2344                 return;
2345
2346         cfg = desc->chip_data;
2347         if (assign_irq_vector(irq, cfg, mask))
2348                 return;
2349
2350         set_extra_move_desc(desc, mask);
2351
2352         dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2353
2354         modify_ioapic_rte = desc->status & IRQ_LEVEL;
2355         if (modify_ioapic_rte) {
2356                 spin_lock_irqsave(&ioapic_lock, flags);
2357                 __target_IO_APIC_irq(irq, dest, cfg);
2358                 spin_unlock_irqrestore(&ioapic_lock, flags);
2359         }
2360
2361         irte.vector = cfg->vector;
2362         irte.dest_id = IRTE_DEST(dest);
2363
2364         /*
2365          * Modified the IRTE and flushes the Interrupt entry cache.
2366          */
2367         modify_irte(irq, &irte);
2368
2369         if (cfg->move_in_progress)
2370                 send_cleanup_vector(cfg);
2371
2372         cpumask_copy(desc->affinity, mask);
2373 }
2374
2375 static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
2376 {
2377         int ret = -1;
2378         struct irq_cfg *cfg = desc->chip_data;
2379
2380         mask_IO_APIC_irq_desc(desc);
2381
2382         if (io_apic_level_ack_pending(cfg)) {
2383                 /*
2384                  * Interrupt in progress. Migrating irq now will change the
2385                  * vector information in the IO-APIC RTE and that will confuse
2386                  * the EOI broadcast performed by cpu.
2387                  * So, delay the irq migration to the next instance.
2388                  */
2389                 schedule_delayed_work(&ir_migration_work, 1);
2390                 goto unmask;
2391         }
2392
2393         /* everthing is clear. we have right of way */
2394         migrate_ioapic_irq_desc(desc, desc->pending_mask);
2395
2396         ret = 0;
2397         desc->status &= ~IRQ_MOVE_PENDING;
2398         cpumask_clear(desc->pending_mask);
2399
2400 unmask:
2401         unmask_IO_APIC_irq_desc(desc);
2402
2403         return ret;
2404 }
2405
2406 static void ir_irq_migration(struct work_struct *work)
2407 {
2408         unsigned int irq;
2409         struct irq_desc *desc;
2410
2411         for_each_irq_desc(irq, desc) {
2412                 if (desc->status & IRQ_MOVE_PENDING) {
2413                         unsigned long flags;
2414
2415                         spin_lock_irqsave(&desc->lock, flags);
2416                         if (!desc->chip->set_affinity ||
2417                             !(desc->status & IRQ_MOVE_PENDING)) {
2418                                 desc->status &= ~IRQ_MOVE_PENDING;
2419                                 spin_unlock_irqrestore(&desc->lock, flags);
2420                                 continue;
2421                         }
2422
2423                         desc->chip->set_affinity(irq, desc->pending_mask);
2424                         spin_unlock_irqrestore(&desc->lock, flags);
2425                 }
2426         }
2427 }
2428
2429 /*
2430  * Migrates the IRQ destination in the process context.
2431  */
2432 static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2433                                             const struct cpumask *mask)
2434 {
2435         if (desc->status & IRQ_LEVEL) {
2436                 desc->status |= IRQ_MOVE_PENDING;
2437                 cpumask_copy(desc->pending_mask, mask);
2438                 migrate_irq_remapped_level_desc(desc);
2439                 return;
2440         }
2441
2442         migrate_ioapic_irq_desc(desc, mask);
2443 }
2444 static void set_ir_ioapic_affinity_irq(unsigned int irq,
2445                                        const struct cpumask *mask)
2446 {
2447         struct irq_desc *desc = irq_to_desc(irq);
2448
2449         set_ir_ioapic_affinity_irq_desc(desc, mask);
2450 }
2451 #endif
2452
2453 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2454 {
2455         unsigned vector, me;
2456
2457         ack_APIC_irq();
2458         exit_idle();
2459         irq_enter();
2460
2461         me = smp_processor_id();
2462         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2463                 unsigned int irq;
2464                 struct irq_desc *desc;
2465                 struct irq_cfg *cfg;
2466                 irq = __get_cpu_var(vector_irq)[vector];
2467
2468                 if (irq == -1)
2469                         continue;
2470
2471                 desc = irq_to_desc(irq);
2472                 if (!desc)
2473                         continue;
2474
2475                 cfg = irq_cfg(irq);
2476                 spin_lock(&desc->lock);
2477                 if (!cfg->move_cleanup_count)
2478                         goto unlock;
2479
2480                 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2481                         goto unlock;
2482
2483                 __get_cpu_var(vector_irq)[vector] = -1;
2484                 cfg->move_cleanup_count--;
2485 unlock:
2486                 spin_unlock(&desc->lock);
2487         }
2488
2489         irq_exit();
2490 }
2491
2492 static void irq_complete_move(struct irq_desc **descp)
2493 {
2494         struct irq_desc *desc = *descp;
2495         struct irq_cfg *cfg = desc->chip_data;
2496         unsigned vector, me;
2497
2498         if (likely(!cfg->move_in_progress)) {
2499 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2500                 if (likely(!cfg->move_desc_pending))
2501                         return;
2502
2503                 /* domain has not changed, but affinity did */
2504                 me = smp_processor_id();
2505                 if (cpumask_test_cpu(me, desc->affinity)) {
2506                         *descp = desc = move_irq_desc(desc, me);
2507                         /* get the new one */
2508                         cfg = desc->chip_data;
2509                         cfg->move_desc_pending = 0;
2510                 }
2511 #endif
2512                 return;
2513         }
2514
2515         vector = ~get_irq_regs()->orig_ax;
2516         me = smp_processor_id();
2517
2518         if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain)) {
2519 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2520                 *descp = desc = move_irq_desc(desc, me);
2521                 /* get the new one */
2522                 cfg = desc->chip_data;
2523 #endif
2524                 send_cleanup_vector(cfg);
2525         }
2526 }
2527 #else
2528 static inline void irq_complete_move(struct irq_desc **descp) {}
2529 #endif
2530
2531 #ifdef CONFIG_INTR_REMAP
2532 static void ack_x2apic_level(unsigned int irq)
2533 {
2534         ack_x2APIC_irq();
2535 }
2536
2537 static void ack_x2apic_edge(unsigned int irq)
2538 {
2539         ack_x2APIC_irq();
2540 }
2541
2542 #endif
2543
2544 static void ack_apic_edge(unsigned int irq)
2545 {
2546         struct irq_desc *desc = irq_to_desc(irq);
2547
2548         irq_complete_move(&desc);
2549         move_native_irq(irq);
2550         ack_APIC_irq();
2551 }
2552
2553 atomic_t irq_mis_count;
2554
2555 static void ack_apic_level(unsigned int irq)
2556 {
2557         struct irq_desc *desc = irq_to_desc(irq);
2558
2559 #ifdef CONFIG_X86_32
2560         unsigned long v;
2561         int i;
2562 #endif
2563         struct irq_cfg *cfg;
2564         int do_unmask_irq = 0;
2565
2566         irq_complete_move(&desc);
2567 #ifdef CONFIG_GENERIC_PENDING_IRQ
2568         /* If we are moving the irq we need to mask it */
2569         if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2570                 do_unmask_irq = 1;
2571                 mask_IO_APIC_irq_desc(desc);
2572         }
2573 #endif
2574
2575 #ifdef CONFIG_X86_32
2576         /*
2577         * It appears there is an erratum which affects at least version 0x11
2578         * of I/O APIC (that's the 82093AA and cores integrated into various
2579         * chipsets).  Under certain conditions a level-triggered interrupt is
2580         * erroneously delivered as edge-triggered one but the respective IRR
2581         * bit gets set nevertheless.  As a result the I/O unit expects an EOI
2582         * message but it will never arrive and further interrupts are blocked
2583         * from the source.  The exact reason is so far unknown, but the
2584         * phenomenon was observed when two consecutive interrupt requests
2585         * from a given source get delivered to the same CPU and the source is
2586         * temporarily disabled in between.
2587         *
2588         * A workaround is to simulate an EOI message manually.  We achieve it
2589         * by setting the trigger mode to edge and then to level when the edge
2590         * trigger mode gets detected in the TMR of a local APIC for a
2591         * level-triggered interrupt.  We mask the source for the time of the
2592         * operation to prevent an edge-triggered interrupt escaping meanwhile.
2593         * The idea is from Manfred Spraul.  --macro
2594         */
2595         cfg = desc->chip_data;
2596         i = cfg->vector;
2597
2598         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2599 #endif
2600
2601         /*
2602          * We must acknowledge the irq before we move it or the acknowledge will
2603          * not propagate properly.
2604          */
2605         ack_APIC_irq();
2606
2607         /* Now we can move and renable the irq */
2608         if (unlikely(do_unmask_irq)) {
2609                 /* Only migrate the irq if the ack has been received.
2610                  *
2611                  * On rare occasions the broadcast level triggered ack gets
2612                  * delayed going to ioapics, and if we reprogram the
2613                  * vector while Remote IRR is still set the irq will never
2614                  * fire again.
2615                  *
2616                  * To prevent this scenario we read the Remote IRR bit
2617                  * of the ioapic.  This has two effects.
2618                  * - On any sane system the read of the ioapic will
2619                  *   flush writes (and acks) going to the ioapic from
2620                  *   this cpu.
2621                  * - We get to see if the ACK has actually been delivered.
2622                  *
2623                  * Based on failed experiments of reprogramming the
2624                  * ioapic entry from outside of irq context starting
2625                  * with masking the ioapic entry and then polling until
2626                  * Remote IRR was clear before reprogramming the
2627                  * ioapic I don't trust the Remote IRR bit to be
2628                  * completey accurate.
2629                  *
2630                  * However there appears to be no other way to plug
2631                  * this race, so if the Remote IRR bit is not
2632                  * accurate and is causing problems then it is a hardware bug
2633                  * and you can go talk to the chipset vendor about it.
2634                  */
2635                 cfg = desc->chip_data;
2636                 if (!io_apic_level_ack_pending(cfg))
2637                         move_masked_irq(irq);
2638                 unmask_IO_APIC_irq_desc(desc);
2639         }
2640
2641 #ifdef CONFIG_X86_32
2642         if (!(v & (1 << (i & 0x1f)))) {
2643                 atomic_inc(&irq_mis_count);
2644                 spin_lock(&ioapic_lock);
2645                 __mask_and_edge_IO_APIC_irq(cfg);
2646                 __unmask_and_level_IO_APIC_irq(cfg);
2647                 spin_unlock(&ioapic_lock);
2648         }
2649 #endif
2650 }
2651
2652 static struct irq_chip ioapic_chip __read_mostly = {
2653         .name           = "IO-APIC",
2654         .startup        = startup_ioapic_irq,
2655         .mask           = mask_IO_APIC_irq,
2656         .unmask         = unmask_IO_APIC_irq,
2657         .ack            = ack_apic_edge,
2658         .eoi            = ack_apic_level,
2659 #ifdef CONFIG_SMP
2660         .set_affinity   = set_ioapic_affinity_irq,
2661 #endif
2662         .retrigger      = ioapic_retrigger_irq,
2663 };
2664
2665 #ifdef CONFIG_INTR_REMAP
2666 static struct irq_chip ir_ioapic_chip __read_mostly = {
2667         .name           = "IR-IO-APIC",
2668         .startup        = startup_ioapic_irq,
2669         .mask           = mask_IO_APIC_irq,
2670         .unmask         = unmask_IO_APIC_irq,
2671         .ack            = ack_x2apic_edge,
2672         .eoi            = ack_x2apic_level,
2673 #ifdef CONFIG_SMP
2674         .set_affinity   = set_ir_ioapic_affinity_irq,
2675 #endif
2676         .retrigger      = ioapic_retrigger_irq,
2677 };
2678 #endif
2679
2680 static inline void init_IO_APIC_traps(void)
2681 {
2682         int irq;
2683         struct irq_desc *desc;
2684         struct irq_cfg *cfg;
2685
2686         /*
2687          * NOTE! The local APIC isn't very good at handling
2688          * multiple interrupts at the same interrupt level.
2689          * As the interrupt level is determined by taking the
2690          * vector number and shifting that right by 4, we
2691          * want to spread these out a bit so that they don't
2692          * all fall in the same interrupt level.
2693          *
2694          * Also, we've got to be careful not to trash gate
2695          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2696          */
2697         for_each_irq_desc(irq, desc) {
2698                 cfg = desc->chip_data;
2699                 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2700                         /*
2701                          * Hmm.. We don't have an entry for this,
2702                          * so default to an old-fashioned 8259
2703                          * interrupt if we can..
2704                          */
2705                         if (irq < NR_IRQS_LEGACY)
2706                                 make_8259A_irq(irq);
2707                         else
2708                                 /* Strange. Oh, well.. */
2709                                 desc->chip = &no_irq_chip;
2710                 }
2711         }
2712 }
2713
2714 /*
2715  * The local APIC irq-chip implementation:
2716  */
2717
2718 static void mask_lapic_irq(unsigned int irq)
2719 {
2720         unsigned long v;
2721
2722         v = apic_read(APIC_LVT0);
2723         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2724 }
2725
2726 static void unmask_lapic_irq(unsigned int irq)
2727 {
2728         unsigned long v;
2729
2730         v = apic_read(APIC_LVT0);
2731         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2732 }
2733
2734 static void ack_lapic_irq(unsigned int irq)
2735 {
2736         ack_APIC_irq();
2737 }
2738
2739 static struct irq_chip lapic_chip __read_mostly = {
2740         .name           = "local-APIC",
2741         .mask           = mask_lapic_irq,
2742         .unmask         = unmask_lapic_irq,
2743         .ack            = ack_lapic_irq,
2744 };
2745
2746 static void lapic_register_intr(int irq, struct irq_desc *desc)
2747 {
2748         desc->status &= ~IRQ_LEVEL;
2749         set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2750                                       "edge");
2751 }
2752
2753 static void __init setup_nmi(void)
2754 {
2755         /*
2756          * Dirty trick to enable the NMI watchdog ...
2757          * We put the 8259A master into AEOI mode and
2758          * unmask on all local APICs LVT0 as NMI.
2759          *
2760          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2761          * is from Maciej W. Rozycki - so we do not have to EOI from
2762          * the NMI handler or the timer interrupt.
2763          */
2764         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2765
2766         enable_NMI_through_LVT0();
2767
2768         apic_printk(APIC_VERBOSE, " done.\n");
2769 }
2770
2771 /*
2772  * This looks a bit hackish but it's about the only one way of sending
2773  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2774  * not support the ExtINT mode, unfortunately.  We need to send these
2775  * cycles as some i82489DX-based boards have glue logic that keeps the
2776  * 8259A interrupt line asserted until INTA.  --macro
2777  */
2778 static inline void __init unlock_ExtINT_logic(void)
2779 {
2780         int apic, pin, i;
2781         struct IO_APIC_route_entry entry0, entry1;
2782         unsigned char save_control, save_freq_select;
2783
2784         pin  = find_isa_irq_pin(8, mp_INT);
2785         if (pin == -1) {
2786                 WARN_ON_ONCE(1);
2787                 return;
2788         }
2789         apic = find_isa_irq_apic(8, mp_INT);
2790         if (apic == -1) {
2791                 WARN_ON_ONCE(1);
2792                 return;
2793         }
2794
2795         entry0 = ioapic_read_entry(apic, pin);
2796         clear_IO_APIC_pin(apic, pin);
2797
2798         memset(&entry1, 0, sizeof(entry1));
2799
2800         entry1.dest_mode = 0;                   /* physical delivery */
2801         entry1.mask = 0;                        /* unmask IRQ now */
2802         entry1.dest = hard_smp_processor_id();
2803         entry1.delivery_mode = dest_ExtINT;
2804         entry1.polarity = entry0.polarity;
2805         entry1.trigger = 0;
2806         entry1.vector = 0;
2807
2808         ioapic_write_entry(apic, pin, entry1);
2809
2810         save_control = CMOS_READ(RTC_CONTROL);
2811         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2812         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2813                    RTC_FREQ_SELECT);
2814         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2815
2816         i = 100;
2817         while (i-- > 0) {
2818                 mdelay(10);
2819                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2820                         i -= 10;
2821         }
2822
2823         CMOS_WRITE(save_control, RTC_CONTROL);
2824         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2825         clear_IO_APIC_pin(apic, pin);
2826
2827         ioapic_write_entry(apic, pin, entry0);
2828 }
2829
2830 static int disable_timer_pin_1 __initdata;
2831 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2832 static int __init disable_timer_pin_setup(char *arg)
2833 {
2834         disable_timer_pin_1 = 1;
2835         return 0;
2836 }
2837 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2838
2839 int timer_through_8259 __initdata;
2840
2841 /*
2842  * This code may look a bit paranoid, but it's supposed to cooperate with
2843  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2844  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2845  * fanatically on his truly buggy board.
2846  *
2847  * FIXME: really need to revamp this for all platforms.
2848  */
2849 static inline void __init check_timer(void)
2850 {
2851         struct irq_desc *desc = irq_to_desc(0);
2852         struct irq_cfg *cfg = desc->chip_data;
2853         int cpu = boot_cpu_id;
2854         int apic1, pin1, apic2, pin2;
2855         unsigned long flags;
2856         int no_pin1 = 0;
2857
2858         local_irq_save(flags);
2859
2860         /*
2861          * get/set the timer IRQ vector:
2862          */
2863         disable_8259A_irq(0);
2864         assign_irq_vector(0, cfg, apic->target_cpus());
2865
2866         /*
2867          * As IRQ0 is to be enabled in the 8259A, the virtual
2868          * wire has to be disabled in the local APIC.  Also
2869          * timer interrupts need to be acknowledged manually in
2870          * the 8259A for the i82489DX when using the NMI
2871          * watchdog as that APIC treats NMIs as level-triggered.
2872          * The AEOI mode will finish them in the 8259A
2873          * automatically.
2874          */
2875         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2876         init_8259A(1);
2877 #ifdef CONFIG_X86_32
2878         {
2879                 unsigned int ver;
2880
2881                 ver = apic_read(APIC_LVR);
2882                 ver = GET_APIC_VERSION(ver);
2883                 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2884         }
2885 #endif
2886
2887         pin1  = find_isa_irq_pin(0, mp_INT);
2888         apic1 = find_isa_irq_apic(0, mp_INT);
2889         pin2  = ioapic_i8259.pin;
2890         apic2 = ioapic_i8259.apic;
2891
2892         apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2893                     "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2894                     cfg->vector, apic1, pin1, apic2, pin2);
2895
2896         /*
2897          * Some BIOS writers are clueless and report the ExtINTA
2898          * I/O APIC input from the cascaded 8259A as the timer
2899          * interrupt input.  So just in case, if only one pin
2900          * was found above, try it both directly and through the
2901          * 8259A.
2902          */
2903         if (pin1 == -1) {
2904 #ifdef CONFIG_INTR_REMAP
2905                 if (intr_remapping_enabled)
2906                         panic("BIOS bug: timer not connected to IO-APIC");
2907 #endif
2908                 pin1 = pin2;
2909                 apic1 = apic2;
2910                 no_pin1 = 1;
2911         } else if (pin2 == -1) {
2912                 pin2 = pin1;
2913                 apic2 = apic1;
2914         }
2915
2916         if (pin1 != -1) {
2917                 /*
2918                  * Ok, does IRQ0 through the IOAPIC work?
2919                  */
2920                 if (no_pin1) {
2921                         add_pin_to_irq_cpu(cfg, cpu, apic1, pin1);
2922                         setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2923                 } else {
2924                         /* for edge trigger, setup_IO_APIC_irq already
2925                          * leave it unmasked.
2926                          * so only need to unmask if it is level-trigger
2927                          * do we really have level trigger timer?
2928                          */
2929                         int idx;
2930                         idx = find_irq_entry(apic1, pin1, mp_INT);
2931                         if (idx != -1 && irq_trigger(idx))
2932                                 unmask_IO_APIC_irq_desc(desc);
2933                 }
2934                 if (timer_irq_works()) {
2935                         if (nmi_watchdog == NMI_IO_APIC) {
2936                                 setup_nmi();
2937                                 enable_8259A_irq(0);
2938                         }
2939                         if (disable_timer_pin_1 > 0)
2940                                 clear_IO_APIC_pin(0, pin1);
2941                         goto out;
2942                 }
2943 #ifdef CONFIG_INTR_REMAP
2944                 if (intr_remapping_enabled)
2945                         panic("timer doesn't work through Interrupt-remapped IO-APIC");
2946 #endif
2947                 local_irq_disable();
2948                 clear_IO_APIC_pin(apic1, pin1);
2949                 if (!no_pin1)
2950                         apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2951                                     "8254 timer not connected to IO-APIC\n");
2952
2953                 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2954                             "(IRQ0) through the 8259A ...\n");
2955                 apic_printk(APIC_QUIET, KERN_INFO
2956                             "..... (found apic %d pin %d) ...\n", apic2, pin2);
2957                 /*
2958                  * legacy devices should be connected to IO APIC #0
2959                  */
2960                 replace_pin_at_irq_cpu(cfg, cpu, apic1, pin1, apic2, pin2);
2961                 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2962                 enable_8259A_irq(0);
2963                 if (timer_irq_works()) {
2964                         apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2965                         timer_through_8259 = 1;
2966                         if (nmi_watchdog == NMI_IO_APIC) {
2967                                 disable_8259A_irq(0);
2968                                 setup_nmi();
2969                                 enable_8259A_irq(0);
2970                         }
2971                         goto out;
2972                 }
2973                 /*
2974                  * Cleanup, just in case ...
2975                  */
2976                 local_irq_disable();
2977                 disable_8259A_irq(0);
2978                 clear_IO_APIC_pin(apic2, pin2);
2979                 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2980         }
2981
2982         if (nmi_watchdog == NMI_IO_APIC) {
2983                 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2984                             "through the IO-APIC - disabling NMI Watchdog!\n");
2985                 nmi_watchdog = NMI_NONE;
2986         }
2987 #ifdef CONFIG_X86_32
2988         timer_ack = 0;
2989 #endif
2990
2991         apic_printk(APIC_QUIET, KERN_INFO
2992                     "...trying to set up timer as Virtual Wire IRQ...\n");
2993
2994         lapic_register_intr(0, desc);
2995         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
2996         enable_8259A_irq(0);
2997
2998         if (timer_irq_works()) {
2999                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3000                 goto out;
3001         }
3002         local_irq_disable();
3003         disable_8259A_irq(0);
3004         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3005         apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3006
3007         apic_printk(APIC_QUIET, KERN_INFO
3008                     "...trying to set up timer as ExtINT IRQ...\n");
3009
3010         init_8259A(0);
3011         make_8259A_irq(0);
3012         apic_write(APIC_LVT0, APIC_DM_EXTINT);
3013
3014         unlock_ExtINT_logic();
3015
3016         if (timer_irq_works()) {
3017                 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3018                 goto out;
3019         }
3020         local_irq_disable();
3021         apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3022         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
3023                 "report.  Then try booting with the 'noapic' option.\n");
3024 out:
3025         local_irq_restore(flags);
3026 }
3027
3028 /*
3029  * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3030  * to devices.  However there may be an I/O APIC pin available for
3031  * this interrupt regardless.  The pin may be left unconnected, but
3032  * typically it will be reused as an ExtINT cascade interrupt for
3033  * the master 8259A.  In the MPS case such a pin will normally be
3034  * reported as an ExtINT interrupt in the MP table.  With ACPI
3035  * there is no provision for ExtINT interrupts, and in the absence
3036  * of an override it would be treated as an ordinary ISA I/O APIC
3037  * interrupt, that is edge-triggered and unmasked by default.  We
3038  * used to do this, but it caused problems on some systems because
3039  * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3040  * the same ExtINT cascade interrupt to drive the local APIC of the
3041  * bootstrap processor.  Therefore we refrain from routing IRQ2 to
3042  * the I/O APIC in all cases now.  No actual device should request
3043  * it anyway.  --macro
3044  */
3045 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
3046
3047 void __init setup_IO_APIC(void)
3048 {
3049
3050         /*
3051          * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3052          */
3053
3054         io_apic_irqs = ~PIC_IRQS;
3055
3056         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3057         /*
3058          * Set up IO-APIC IRQ routing.
3059          */
3060 #ifdef CONFIG_X86_32
3061         if (!acpi_ioapic)
3062                 setup_ioapic_ids_from_mpc();
3063 #endif
3064         sync_Arb_IDs();
3065         setup_IO_APIC_irqs();
3066         init_IO_APIC_traps();
3067         check_timer();
3068 }
3069
3070 /*
3071  *      Called after all the initialization is done. If we didnt find any
3072  *      APIC bugs then we can allow the modify fast path
3073  */
3074
3075 static int __init io_apic_bug_finalize(void)
3076 {
3077         if (sis_apic_bug == -1)
3078                 sis_apic_bug = 0;
3079         return 0;
3080 }
3081
3082 late_initcall(io_apic_bug_finalize);
3083
3084 struct sysfs_ioapic_data {
3085         struct sys_device dev;
3086         struct IO_APIC_route_entry entry[0];
3087 };
3088 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3089
3090 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3091 {
3092         struct IO_APIC_route_entry *entry;
3093         struct sysfs_ioapic_data *data;
3094         int i;
3095
3096         data = container_of(dev, struct sysfs_ioapic_data, dev);
3097         entry = data->entry;
3098         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3099                 *entry = ioapic_read_entry(dev->id, i);
3100
3101         return 0;
3102 }
3103
3104 static int ioapic_resume(struct sys_device *dev)
3105 {
3106         struct IO_APIC_route_entry *entry;
3107         struct sysfs_ioapic_data *data;
3108         unsigned long flags;
3109         union IO_APIC_reg_00 reg_00;
3110         int i;
3111
3112         data = container_of(dev, struct sysfs_ioapic_data, dev);
3113         entry = data->entry;
3114
3115         spin_lock_irqsave(&ioapic_lock, flags);
3116         reg_00.raw = io_apic_read(dev->id, 0);
3117         if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3118                 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3119                 io_apic_write(dev->id, 0, reg_00.raw);
3120         }
3121         spin_unlock_irqrestore(&ioapic_lock, flags);
3122         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3123                 ioapic_write_entry(dev->id, i, entry[i]);
3124
3125         return 0;
3126 }
3127
3128 static struct sysdev_class ioapic_sysdev_class = {
3129         .name = "ioapic",
3130         .suspend = ioapic_suspend,
3131         .resume = ioapic_resume,
3132 };
3133
3134 static int __init ioapic_init_sysfs(void)
3135 {
3136         struct sys_device * dev;
3137         int i, size, error;
3138
3139         error = sysdev_class_register(&ioapic_sysdev_class);
3140         if (error)
3141                 return error;
3142
3143         for (i = 0; i < nr_ioapics; i++ ) {
3144                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3145                         * sizeof(struct IO_APIC_route_entry);
3146                 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3147                 if (!mp_ioapic_data[i]) {
3148                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3149                         continue;
3150                 }
3151                 dev = &mp_ioapic_data[i]->dev;
3152                 dev->id = i;
3153                 dev->cls = &ioapic_sysdev_class;
3154                 error = sysdev_register(dev);
3155                 if (error) {
3156                         kfree(mp_ioapic_data[i]);
3157                         mp_ioapic_data[i] = NULL;
3158                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3159                         continue;
3160                 }
3161         }
3162
3163         return 0;
3164 }
3165
3166 device_initcall(ioapic_init_sysfs);
3167
3168 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3169 /*
3170  * Dynamic irq allocate and deallocation
3171  */
3172 unsigned int create_irq_nr(unsigned int irq_want)
3173 {
3174         /* Allocate an unused irq */
3175         unsigned int irq;
3176         unsigned int new;
3177         unsigned long flags;
3178         struct irq_cfg *cfg_new = NULL;
3179         int cpu = boot_cpu_id;
3180         struct irq_desc *desc_new = NULL;
3181
3182         irq = 0;
3183         if (irq_want < nr_irqs_gsi)
3184                 irq_want = nr_irqs_gsi;
3185
3186         spin_lock_irqsave(&vector_lock, flags);
3187         for (new = irq_want; new < nr_irqs; new++) {
3188                 desc_new = irq_to_desc_alloc_cpu(new, cpu);
3189                 if (!desc_new) {
3190                         printk(KERN_INFO "can not get irq_desc for %d\n", new);
3191                         continue;
3192                 }
3193                 cfg_new = desc_new->chip_data;
3194
3195                 if (cfg_new->vector != 0)
3196                         continue;
3197                 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3198                         irq = new;
3199                 break;
3200         }
3201         spin_unlock_irqrestore(&vector_lock, flags);
3202
3203         if (irq > 0) {
3204                 dynamic_irq_init(irq);
3205                 /* restore it, in case dynamic_irq_init clear it */
3206                 if (desc_new)
3207                         desc_new->chip_data = cfg_new;
3208         }
3209         return irq;
3210 }
3211
3212 int create_irq(void)
3213 {
3214         unsigned int irq_want;
3215         int irq;
3216
3217         irq_want = nr_irqs_gsi;
3218         irq = create_irq_nr(irq_want);
3219
3220         if (irq == 0)
3221                 irq = -1;
3222
3223         return irq;
3224 }
3225
3226 void destroy_irq(unsigned int irq)
3227 {
3228         unsigned long flags;
3229         struct irq_cfg *cfg;
3230         struct irq_desc *desc;
3231
3232         /* store it, in case dynamic_irq_cleanup clear it */
3233         desc = irq_to_desc(irq);
3234         cfg = desc->chip_data;
3235         dynamic_irq_cleanup(irq);
3236         /* connect back irq_cfg */
3237         if (desc)
3238                 desc->chip_data = cfg;
3239
3240 #ifdef CONFIG_INTR_REMAP
3241         free_irte(irq);
3242 #endif
3243         spin_lock_irqsave(&vector_lock, flags);
3244         __clear_irq_vector(irq, cfg);
3245         spin_unlock_irqrestore(&vector_lock, flags);
3246 }
3247
3248 /*
3249  * MSI message composition
3250  */
3251 #ifdef CONFIG_PCI_MSI
3252 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3253 {
3254         struct irq_cfg *cfg;
3255         int err;
3256         unsigned dest;
3257
3258         if (disable_apic)
3259                 return -ENXIO;
3260
3261         cfg = irq_cfg(irq);
3262         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3263         if (err)
3264                 return err;
3265
3266         dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3267
3268 #ifdef CONFIG_INTR_REMAP
3269         if (irq_remapped(irq)) {
3270                 struct irte irte;
3271                 int ir_index;
3272                 u16 sub_handle;
3273
3274                 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3275                 BUG_ON(ir_index == -1);
3276
3277                 memset (&irte, 0, sizeof(irte));
3278
3279                 irte.present = 1;
3280                 irte.dst_mode = apic->irq_dest_mode;
3281                 irte.trigger_mode = 0; /* edge */
3282                 irte.dlvry_mode = apic->irq_delivery_mode;
3283                 irte.vector = cfg->vector;
3284                 irte.dest_id = IRTE_DEST(dest);
3285
3286                 modify_irte(irq, &irte);
3287
3288                 msg->address_hi = MSI_ADDR_BASE_HI;
3289                 msg->data = sub_handle;
3290                 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3291                                   MSI_ADDR_IR_SHV |
3292                                   MSI_ADDR_IR_INDEX1(ir_index) |
3293                                   MSI_ADDR_IR_INDEX2(ir_index);
3294         } else
3295 #endif
3296         {
3297                 msg->address_hi = MSI_ADDR_BASE_HI;
3298                 msg->address_lo =
3299                         MSI_ADDR_BASE_LO |
3300                         ((apic->irq_dest_mode == 0) ?
3301                                 MSI_ADDR_DEST_MODE_PHYSICAL:
3302                                 MSI_ADDR_DEST_MODE_LOGICAL) |
3303                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3304                                 MSI_ADDR_REDIRECTION_CPU:
3305                                 MSI_ADDR_REDIRECTION_LOWPRI) |
3306                         MSI_ADDR_DEST_ID(dest);
3307
3308                 msg->data =
3309                         MSI_DATA_TRIGGER_EDGE |
3310                         MSI_DATA_LEVEL_ASSERT |
3311                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3312                                 MSI_DATA_DELIVERY_FIXED:
3313                                 MSI_DATA_DELIVERY_LOWPRI) |
3314                         MSI_DATA_VECTOR(cfg->vector);
3315         }
3316         return err;
3317 }
3318
3319 #ifdef CONFIG_SMP
3320 static void set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3321 {
3322         struct irq_desc *desc = irq_to_desc(irq);
3323         struct irq_cfg *cfg;
3324         struct msi_msg msg;
3325         unsigned int dest;
3326
3327         dest = set_desc_affinity(desc, mask);
3328         if (dest == BAD_APICID)
3329                 return;
3330
3331         cfg = desc->chip_data;
3332
3333         read_msi_msg_desc(desc, &msg);
3334
3335         msg.data &= ~MSI_DATA_VECTOR_MASK;
3336         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3337         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3338         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3339
3340         write_msi_msg_desc(desc, &msg);
3341 }
3342 #ifdef CONFIG_INTR_REMAP
3343 /*
3344  * Migrate the MSI irq to another cpumask. This migration is
3345  * done in the process context using interrupt-remapping hardware.
3346  */
3347 static void
3348 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3349 {
3350         struct irq_desc *desc = irq_to_desc(irq);
3351         struct irq_cfg *cfg = desc->chip_data;
3352         unsigned int dest;
3353         struct irte irte;
3354
3355         if (get_irte(irq, &irte))
3356                 return;
3357
3358         dest = set_desc_affinity(desc, mask);
3359         if (dest == BAD_APICID)
3360                 return;
3361
3362         irte.vector = cfg->vector;
3363         irte.dest_id = IRTE_DEST(dest);
3364
3365         /*
3366          * atomically update the IRTE with the new destination and vector.
3367          */
3368         modify_irte(irq, &irte);
3369
3370         /*
3371          * After this point, all the interrupts will start arriving
3372          * at the new destination. So, time to cleanup the previous
3373          * vector allocation.
3374          */
3375         if (cfg->move_in_progress)
3376                 send_cleanup_vector(cfg);
3377 }
3378
3379 #endif
3380 #endif /* CONFIG_SMP */
3381
3382 /*
3383  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3384  * which implement the MSI or MSI-X Capability Structure.
3385  */
3386 static struct irq_chip msi_chip = {
3387         .name           = "PCI-MSI",
3388         .unmask         = unmask_msi_irq,
3389         .mask           = mask_msi_irq,
3390         .ack            = ack_apic_edge,
3391 #ifdef CONFIG_SMP
3392         .set_affinity   = set_msi_irq_affinity,
3393 #endif
3394         .retrigger      = ioapic_retrigger_irq,
3395 };
3396
3397 #ifdef CONFIG_INTR_REMAP
3398 static struct irq_chip msi_ir_chip = {
3399         .name           = "IR-PCI-MSI",
3400         .unmask         = unmask_msi_irq,
3401         .mask           = mask_msi_irq,
3402         .ack            = ack_x2apic_edge,
3403 #ifdef CONFIG_SMP
3404         .set_affinity   = ir_set_msi_irq_affinity,
3405 #endif
3406         .retrigger      = ioapic_retrigger_irq,
3407 };
3408
3409 /*
3410  * Map the PCI dev to the corresponding remapping hardware unit
3411  * and allocate 'nvec' consecutive interrupt-remapping table entries
3412  * in it.
3413  */
3414 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3415 {
3416         struct intel_iommu *iommu;
3417         int index;
3418
3419         iommu = map_dev_to_ir(dev);
3420         if (!iommu) {
3421                 printk(KERN_ERR
3422                        "Unable to map PCI %s to iommu\n", pci_name(dev));
3423                 return -ENOENT;
3424         }
3425
3426         index = alloc_irte(iommu, irq, nvec);
3427         if (index < 0) {
3428                 printk(KERN_ERR
3429                        "Unable to allocate %d IRTE for PCI %s\n", nvec,
3430                        pci_name(dev));
3431                 return -ENOSPC;
3432         }
3433         return index;
3434 }
3435 #endif
3436
3437 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3438 {
3439         int ret;
3440         struct msi_msg msg;
3441
3442         ret = msi_compose_msg(dev, irq, &msg);
3443         if (ret < 0)
3444                 return ret;
3445
3446         set_irq_msi(irq, msidesc);
3447         write_msi_msg(irq, &msg);
3448
3449 #ifdef CONFIG_INTR_REMAP
3450         if (irq_remapped(irq)) {
3451                 struct irq_desc *desc = irq_to_desc(irq);
3452                 /*
3453                  * irq migration in process context
3454                  */
3455                 desc->status |= IRQ_MOVE_PCNTXT;
3456                 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3457         } else
3458 #endif
3459                 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3460
3461         dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3462
3463         return 0;
3464 }
3465
3466 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3467 {
3468         unsigned int irq;
3469         int ret, sub_handle;
3470         struct msi_desc *msidesc;
3471         unsigned int irq_want;
3472
3473 #ifdef CONFIG_INTR_REMAP
3474         struct intel_iommu *iommu = 0;
3475         int index = 0;
3476 #endif
3477
3478         irq_want = nr_irqs_gsi;
3479         sub_handle = 0;
3480         list_for_each_entry(msidesc, &dev->msi_list, list) {
3481                 irq = create_irq_nr(irq_want);
3482                 if (irq == 0)
3483                         return -1;
3484                 irq_want = irq + 1;
3485 #ifdef CONFIG_INTR_REMAP
3486                 if (!intr_remapping_enabled)
3487                         goto no_ir;
3488
3489                 if (!sub_handle) {
3490                         /*
3491                          * allocate the consecutive block of IRTE's
3492                          * for 'nvec'
3493                          */
3494                         index = msi_alloc_irte(dev, irq, nvec);
3495                         if (index < 0) {
3496                                 ret = index;
3497                                 goto error;
3498                         }
3499                 } else {
3500                         iommu = map_dev_to_ir(dev);
3501                         if (!iommu) {
3502                                 ret = -ENOENT;
3503                                 goto error;
3504                         }
3505                         /*
3506                          * setup the mapping between the irq and the IRTE
3507                          * base index, the sub_handle pointing to the
3508                          * appropriate interrupt remap table entry.
3509                          */
3510                         set_irte_irq(irq, iommu, index, sub_handle);
3511                 }
3512 no_ir:
3513 #endif
3514                 ret = setup_msi_irq(dev, msidesc, irq);
3515                 if (ret < 0)
3516                         goto error;
3517                 sub_handle++;
3518         }
3519         return 0;
3520
3521 error:
3522         destroy_irq(irq);
3523         return ret;
3524 }
3525
3526 void arch_teardown_msi_irq(unsigned int irq)
3527 {
3528         destroy_irq(irq);
3529 }
3530
3531 #ifdef CONFIG_DMAR
3532 #ifdef CONFIG_SMP
3533 static void dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3534 {
3535         struct irq_desc *desc = irq_to_desc(irq);
3536         struct irq_cfg *cfg;
3537         struct msi_msg msg;
3538         unsigned int dest;
3539
3540         dest = set_desc_affinity(desc, mask);
3541         if (dest == BAD_APICID)
3542                 return;
3543
3544         cfg = desc->chip_data;
3545
3546         dmar_msi_read(irq, &msg);
3547
3548         msg.data &= ~MSI_DATA_VECTOR_MASK;
3549         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3550         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3551         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3552
3553         dmar_msi_write(irq, &msg);
3554 }
3555
3556 #endif /* CONFIG_SMP */
3557
3558 struct irq_chip dmar_msi_type = {
3559         .name = "DMAR_MSI",
3560         .unmask = dmar_msi_unmask,
3561         .mask = dmar_msi_mask,
3562         .ack = ack_apic_edge,
3563 #ifdef CONFIG_SMP
3564         .set_affinity = dmar_msi_set_affinity,
3565 #endif
3566         .retrigger = ioapic_retrigger_irq,
3567 };
3568
3569 int arch_setup_dmar_msi(unsigned int irq)
3570 {
3571         int ret;
3572         struct msi_msg msg;
3573
3574         ret = msi_compose_msg(NULL, irq, &msg);
3575         if (ret < 0)
3576                 return ret;
3577         dmar_msi_write(irq, &msg);
3578         set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3579                 "edge");
3580         return 0;
3581 }
3582 #endif
3583
3584 #ifdef CONFIG_HPET_TIMER
3585
3586 #ifdef CONFIG_SMP
3587 static void hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3588 {
3589         struct irq_desc *desc = irq_to_desc(irq);
3590         struct irq_cfg *cfg;
3591         struct msi_msg msg;
3592         unsigned int dest;
3593
3594         dest = set_desc_affinity(desc, mask);
3595         if (dest == BAD_APICID)
3596                 return;
3597
3598         cfg = desc->chip_data;
3599
3600         hpet_msi_read(irq, &msg);
3601
3602         msg.data &= ~MSI_DATA_VECTOR_MASK;
3603         msg.data |= MSI_DATA_VECTOR(cfg->vector);
3604         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3605         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3606
3607         hpet_msi_write(irq, &msg);
3608 }
3609
3610 #endif /* CONFIG_SMP */
3611
3612 struct irq_chip hpet_msi_type = {
3613         .name = "HPET_MSI",
3614         .unmask = hpet_msi_unmask,
3615         .mask = hpet_msi_mask,
3616         .ack = ack_apic_edge,
3617 #ifdef CONFIG_SMP
3618         .set_affinity = hpet_msi_set_affinity,
3619 #endif
3620         .retrigger = ioapic_retrigger_irq,
3621 };
3622
3623 int arch_setup_hpet_msi(unsigned int irq)
3624 {
3625         int ret;
3626         struct msi_msg msg;
3627
3628         ret = msi_compose_msg(NULL, irq, &msg);
3629         if (ret < 0)
3630                 return ret;
3631
3632         hpet_msi_write(irq, &msg);
3633         set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3634                 "edge");
3635
3636         return 0;
3637 }
3638 #endif
3639
3640 #endif /* CONFIG_PCI_MSI */
3641 /*
3642  * Hypertransport interrupt support
3643  */
3644 #ifdef CONFIG_HT_IRQ
3645
3646 #ifdef CONFIG_SMP
3647
3648 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3649 {
3650         struct ht_irq_msg msg;
3651         fetch_ht_irq_msg(irq, &msg);
3652
3653         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3654         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3655
3656         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3657         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3658
3659         write_ht_irq_msg(irq, &msg);
3660 }
3661
3662 static void set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3663 {
3664         struct irq_desc *desc = irq_to_desc(irq);
3665         struct irq_cfg *cfg;
3666         unsigned int dest;
3667
3668         dest = set_desc_affinity(desc, mask);
3669         if (dest == BAD_APICID)
3670                 return;
3671
3672         cfg = desc->chip_data;
3673
3674         target_ht_irq(irq, dest, cfg->vector);
3675 }
3676
3677 #endif
3678
3679 static struct irq_chip ht_irq_chip = {
3680         .name           = "PCI-HT",
3681         .mask           = mask_ht_irq,
3682         .unmask         = unmask_ht_irq,
3683         .ack            = ack_apic_edge,
3684 #ifdef CONFIG_SMP
3685         .set_affinity   = set_ht_irq_affinity,
3686 #endif
3687         .retrigger      = ioapic_retrigger_irq,
3688 };
3689
3690 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3691 {
3692         struct irq_cfg *cfg;
3693         int err;
3694
3695         if (disable_apic)
3696                 return -ENXIO;
3697
3698         cfg = irq_cfg(irq);
3699         err = assign_irq_vector(irq, cfg, apic->target_cpus());
3700         if (!err) {
3701                 struct ht_irq_msg msg;
3702                 unsigned dest;
3703
3704                 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3705                                                     apic->target_cpus());
3706
3707                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3708
3709                 msg.address_lo =
3710                         HT_IRQ_LOW_BASE |
3711                         HT_IRQ_LOW_DEST_ID(dest) |
3712                         HT_IRQ_LOW_VECTOR(cfg->vector) |
3713                         ((apic->irq_dest_mode == 0) ?
3714                                 HT_IRQ_LOW_DM_PHYSICAL :
3715                                 HT_IRQ_LOW_DM_LOGICAL) |
3716                         HT_IRQ_LOW_RQEOI_EDGE |
3717                         ((apic->irq_delivery_mode != dest_LowestPrio) ?
3718                                 HT_IRQ_LOW_MT_FIXED :
3719                                 HT_IRQ_LOW_MT_ARBITRATED) |
3720                         HT_IRQ_LOW_IRQ_MASKED;
3721
3722                 write_ht_irq_msg(irq, &msg);
3723
3724                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3725                                               handle_edge_irq, "edge");
3726
3727                 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3728         }
3729         return err;
3730 }
3731 #endif /* CONFIG_HT_IRQ */
3732
3733 #ifdef CONFIG_X86_UV
3734 /*
3735  * Re-target the irq to the specified CPU and enable the specified MMR located
3736  * on the specified blade to allow the sending of MSIs to the specified CPU.
3737  */
3738 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3739                        unsigned long mmr_offset)
3740 {
3741         const struct cpumask *eligible_cpu = cpumask_of(cpu);
3742         struct irq_cfg *cfg;
3743         int mmr_pnode;
3744         unsigned long mmr_value;
3745         struct uv_IO_APIC_route_entry *entry;
3746         unsigned long flags;
3747         int err;
3748
3749         cfg = irq_cfg(irq);
3750
3751         err = assign_irq_vector(irq, cfg, eligible_cpu);
3752         if (err != 0)
3753                 return err;
3754
3755         spin_lock_irqsave(&vector_lock, flags);
3756         set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3757                                       irq_name);
3758         spin_unlock_irqrestore(&vector_lock, flags);
3759
3760         mmr_value = 0;
3761         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3762         BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3763
3764         entry->vector = cfg->vector;
3765         entry->delivery_mode = apic->irq_delivery_mode;
3766         entry->dest_mode = apic->irq_dest_mode;
3767         entry->polarity = 0;
3768         entry->trigger = 0;
3769         entry->mask = 0;
3770         entry->dest = apic->cpu_mask_to_apicid(eligible_cpu);
3771
3772         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3773         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3774
3775         return irq;
3776 }
3777
3778 /*
3779  * Disable the specified MMR located on the specified blade so that MSIs are
3780  * longer allowed to be sent.
3781  */
3782 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3783 {
3784         unsigned long mmr_value;
3785         struct uv_IO_APIC_route_entry *entry;
3786         int mmr_pnode;
3787
3788         mmr_value = 0;
3789         entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3790         BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3791
3792         entry->mask = 1;
3793
3794         mmr_pnode = uv_blade_to_pnode(mmr_blade);
3795         uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3796 }
3797 #endif /* CONFIG_X86_64 */
3798
3799 int __init io_apic_get_redir_entries (int ioapic)
3800 {
3801         union IO_APIC_reg_01    reg_01;
3802         unsigned long flags;
3803
3804         spin_lock_irqsave(&ioapic_lock, flags);
3805         reg_01.raw = io_apic_read(ioapic, 1);
3806         spin_unlock_irqrestore(&ioapic_lock, flags);
3807
3808         return reg_01.bits.entries;
3809 }
3810
3811 void __init probe_nr_irqs_gsi(void)
3812 {
3813         int nr = 0;
3814
3815         nr = acpi_probe_gsi();
3816         if (nr > nr_irqs_gsi) {
3817                 nr_irqs_gsi = nr;
3818         } else {
3819                 /* for acpi=off or acpi is not compiled in */
3820                 int idx;
3821
3822                 nr = 0;
3823                 for (idx = 0; idx < nr_ioapics; idx++)
3824                         nr += io_apic_get_redir_entries(idx) + 1;
3825
3826                 if (nr > nr_irqs_gsi)
3827                         nr_irqs_gsi = nr;
3828         }
3829
3830         printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3831 }
3832
3833 #ifdef CONFIG_SPARSE_IRQ
3834 int __init arch_probe_nr_irqs(void)
3835 {
3836         int nr;
3837
3838         if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3839                 nr_irqs = NR_VECTORS * nr_cpu_ids;
3840
3841         nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3842 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3843         /*
3844          * for MSI and HT dyn irq
3845          */
3846         nr += nr_irqs_gsi * 16;
3847 #endif
3848         if (nr < nr_irqs)
3849                 nr_irqs = nr;
3850
3851         return 0;
3852 }
3853 #endif
3854
3855 /* --------------------------------------------------------------------------
3856                           ACPI-based IOAPIC Configuration
3857    -------------------------------------------------------------------------- */
3858
3859 #ifdef CONFIG_ACPI
3860
3861 #ifdef CONFIG_X86_32
3862 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3863 {
3864         union IO_APIC_reg_00 reg_00;
3865         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3866         physid_mask_t tmp;
3867         unsigned long flags;
3868         int i = 0;
3869
3870         /*
3871          * The P4 platform supports up to 256 APIC IDs on two separate APIC
3872          * buses (one for LAPICs, one for IOAPICs), where predecessors only
3873          * supports up to 16 on one shared APIC bus.
3874          *
3875          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3876          *      advantage of new APIC bus architecture.
3877          */
3878
3879         if (physids_empty(apic_id_map))
3880                 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3881
3882         spin_lock_irqsave(&ioapic_lock, flags);
3883         reg_00.raw = io_apic_read(ioapic, 0);
3884         spin_unlock_irqrestore(&ioapic_lock, flags);
3885
3886         if (apic_id >= get_physical_broadcast()) {
3887                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3888                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
3889                 apic_id = reg_00.bits.ID;
3890         }
3891
3892         /*
3893          * Every APIC in a system must have a unique ID or we get lots of nice
3894          * 'stuck on smp_invalidate_needed IPI wait' messages.
3895          */
3896         if (apic->check_apicid_used(apic_id_map, apic_id)) {
3897
3898                 for (i = 0; i < get_physical_broadcast(); i++) {
3899                         if (!apic->check_apicid_used(apic_id_map, i))
3900                                 break;
3901                 }
3902
3903                 if (i == get_physical_broadcast())
3904                         panic("Max apic_id exceeded!\n");
3905
3906                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3907                         "trying %d\n", ioapic, apic_id, i);
3908
3909                 apic_id = i;
3910         }
3911
3912         tmp = apic->apicid_to_cpu_present(apic_id);
3913         physids_or(apic_id_map, apic_id_map, tmp);
3914
3915         if (reg_00.bits.ID != apic_id) {
3916                 reg_00.bits.ID = apic_id;
3917
3918                 spin_lock_irqsave(&ioapic_lock, flags);
3919                 io_apic_write(ioapic, 0, reg_00.raw);
3920                 reg_00.raw = io_apic_read(ioapic, 0);
3921                 spin_unlock_irqrestore(&ioapic_lock, flags);
3922
3923                 /* Sanity check */
3924                 if (reg_00.bits.ID != apic_id) {
3925                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3926                         return -1;
3927                 }
3928         }
3929
3930         apic_printk(APIC_VERBOSE, KERN_INFO
3931                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3932
3933         return apic_id;
3934 }
3935
3936 int __init io_apic_get_version(int ioapic)
3937 {
3938         union IO_APIC_reg_01    reg_01;
3939         unsigned long flags;
3940
3941         spin_lock_irqsave(&ioapic_lock, flags);
3942         reg_01.raw = io_apic_read(ioapic, 1);
3943         spin_unlock_irqrestore(&ioapic_lock, flags);
3944
3945         return reg_01.bits.version;
3946 }
3947 #endif
3948
3949 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
3950 {
3951         struct irq_desc *desc;
3952         struct irq_cfg *cfg;
3953         int cpu = boot_cpu_id;
3954
3955         if (!IO_APIC_IRQ(irq)) {
3956                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3957                         ioapic);
3958                 return -EINVAL;
3959         }
3960
3961         desc = irq_to_desc_alloc_cpu(irq, cpu);
3962         if (!desc) {
3963                 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3964                 return 0;
3965         }
3966
3967         /*
3968          * IRQs < 16 are already in the irq_2_pin[] map
3969          */
3970         if (irq >= NR_IRQS_LEGACY) {
3971                 cfg = desc->chip_data;
3972                 add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
3973         }
3974
3975         setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
3976
3977         return 0;
3978 }
3979
3980
3981 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
3982 {
3983         int i;
3984
3985         if (skip_ioapic_setup)
3986                 return -1;
3987
3988         for (i = 0; i < mp_irq_entries; i++)
3989                 if (mp_irqs[i].irqtype == mp_INT &&
3990                     mp_irqs[i].srcbusirq == bus_irq)
3991                         break;
3992         if (i >= mp_irq_entries)
3993                 return -1;
3994
3995         *trigger = irq_trigger(i);
3996         *polarity = irq_polarity(i);
3997         return 0;
3998 }
3999
4000 #endif /* CONFIG_ACPI */
4001
4002 /*
4003  * This function currently is only a helper for the i386 smp boot process where
4004  * we need to reprogram the ioredtbls to cater for the cpus which have come online
4005  * so mask in all cases should simply be apic->target_cpus()
4006  */
4007 #ifdef CONFIG_SMP
4008 void __init setup_ioapic_dest(void)
4009 {
4010         int pin, ioapic, irq, irq_entry;
4011         struct irq_desc *desc;
4012         struct irq_cfg *cfg;
4013         const struct cpumask *mask;
4014
4015         if (skip_ioapic_setup == 1)
4016                 return;
4017
4018         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
4019                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4020                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4021                         if (irq_entry == -1)
4022                                 continue;
4023                         irq = pin_2_irq(irq_entry, ioapic, pin);
4024
4025                         /* setup_IO_APIC_irqs could fail to get vector for some device
4026                          * when you have too many devices, because at that time only boot
4027                          * cpu is online.
4028                          */
4029                         desc = irq_to_desc(irq);
4030                         cfg = desc->chip_data;
4031                         if (!cfg->vector) {
4032                                 setup_IO_APIC_irq(ioapic, pin, irq, desc,
4033                                                   irq_trigger(irq_entry),
4034                                                   irq_polarity(irq_entry));
4035                                 continue;
4036
4037                         }
4038
4039                         /*
4040                          * Honour affinities which have been set in early boot
4041                          */
4042                         if (desc->status &
4043                             (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4044                                 mask = desc->affinity;
4045                         else
4046                                 mask = apic->target_cpus();
4047
4048 #ifdef CONFIG_INTR_REMAP
4049                         if (intr_remapping_enabled)
4050                                 set_ir_ioapic_affinity_irq_desc(desc, mask);
4051                         else
4052 #endif
4053                                 set_ioapic_affinity_irq_desc(desc, mask);
4054                 }
4055
4056         }
4057 }
4058 #endif
4059
4060 #define IOAPIC_RESOURCE_NAME_SIZE 11
4061
4062 static struct resource *ioapic_resources;
4063
4064 static struct resource * __init ioapic_setup_resources(void)
4065 {
4066         unsigned long n;
4067         struct resource *res;
4068         char *mem;
4069         int i;
4070
4071         if (nr_ioapics <= 0)
4072                 return NULL;
4073
4074         n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4075         n *= nr_ioapics;
4076
4077         mem = alloc_bootmem(n);
4078         res = (void *)mem;
4079
4080         if (mem != NULL) {
4081                 mem += sizeof(struct resource) * nr_ioapics;
4082
4083                 for (i = 0; i < nr_ioapics; i++) {
4084                         res[i].name = mem;
4085                         res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4086                         sprintf(mem,  "IOAPIC %u", i);
4087                         mem += IOAPIC_RESOURCE_NAME_SIZE;
4088                 }
4089         }
4090
4091         ioapic_resources = res;
4092
4093         return res;
4094 }
4095
4096 void __init ioapic_init_mappings(void)
4097 {
4098         unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4099         struct resource *ioapic_res;
4100         int i;
4101
4102         ioapic_res = ioapic_setup_resources();
4103         for (i = 0; i < nr_ioapics; i++) {
4104                 if (smp_found_config) {
4105                         ioapic_phys = mp_ioapics[i].apicaddr;
4106 #ifdef CONFIG_X86_32
4107                         if (!ioapic_phys) {
4108                                 printk(KERN_ERR
4109                                        "WARNING: bogus zero IO-APIC "
4110                                        "address found in MPTABLE, "
4111                                        "disabling IO/APIC support!\n");
4112                                 smp_found_config = 0;
4113                                 skip_ioapic_setup = 1;
4114                                 goto fake_ioapic_page;
4115                         }
4116 #endif
4117                 } else {
4118 #ifdef CONFIG_X86_32
4119 fake_ioapic_page:
4120 #endif
4121                         ioapic_phys = (unsigned long)
4122                                 alloc_bootmem_pages(PAGE_SIZE);
4123                         ioapic_phys = __pa(ioapic_phys);
4124                 }
4125                 set_fixmap_nocache(idx, ioapic_phys);
4126                 apic_printk(APIC_VERBOSE,
4127                             "mapped IOAPIC to %08lx (%08lx)\n",
4128                             __fix_to_virt(idx), ioapic_phys);
4129                 idx++;
4130
4131                 if (ioapic_res != NULL) {
4132                         ioapic_res->start = ioapic_phys;
4133                         ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4134                         ioapic_res++;
4135                 }
4136         }
4137 }
4138
4139 static int __init ioapic_insert_resources(void)
4140 {
4141         int i;
4142         struct resource *r = ioapic_resources;
4143
4144         if (!r) {
4145                 printk(KERN_ERR
4146                        "IO APIC resources could be not be allocated.\n");
4147                 return -1;
4148         }
4149
4150         for (i = 0; i < nr_ioapics; i++) {
4151                 insert_resource(&iomem_resource, r);
4152                 r++;
4153         }
4154
4155         return 0;
4156 }
4157
4158 /* Insert the IO APIC resources after PCI initialization has occured to handle
4159  * IO APICS that are mapped in on a BAR in PCI space. */
4160 late_initcall(ioapic_insert_resources);