arch/x86/kernel/io_apic_64.c

   1 /*
   2  *      Intel IO-APIC support for multi-Pentium hosts.
   3  *
   4  *      Copyright (C) 1997, 1998, 1999, 2000 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/acpi.h>
  31 #include <linux/sysdev.h>
  32 #include <linux/msi.h>
  33 #include <linux/htirq.h>
  34 #ifdef CONFIG_ACPI
  35 #include <acpi/acpi_bus.h>
  36 #endif
  37
  38 #include <asm/idle.h>
  39 #include <asm/io.h>
  40 #include <asm/smp.h>
  41 #include <asm/desc.h>
  42 #include <asm/proto.h>
  43 #include <asm/mach_apic.h>
  44 #include <asm/acpi.h>
  45 #include <asm/dma.h>
  46 #include <asm/nmi.h>
  47 #include <asm/msidef.h>
  48 #include <asm/hypertransport.h>
  49
  50 struct irq_cfg {
  51         cpumask_t domain;
  52         cpumask_t old_domain;
  53         unsigned move_cleanup_count;
  54         u8 vector;
  55         u8 move_in_progress : 1;
  56 };
  57
  58 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
  59 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
  60         [0]  = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR,  },
  61         [1]  = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR,  },
  62         [2]  = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR,  },
  63         [3]  = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR,  },
  64         [4]  = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR,  },
  65         [5]  = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR,  },
  66         [6]  = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR,  },
  67         [7]  = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR,  },
  68         [8]  = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR,  },
  69         [9]  = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR,  },
  70         [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
  71         [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
  72         [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
  73         [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
  74         [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
  75         [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
  76 };
  77
  78 static int assign_irq_vector(int irq, cpumask_t mask);
  79
  80 #define __apicdebuginit  __init
  81
  82 int sis_apic_bug; /* not actually supported, dummy for compile */
  83
  84 static int no_timer_check;
  85
  86 static int disable_timer_pin_1 __initdata;
  87
  88 int timer_over_8254 __initdata = 1;
  89
  90 /* Where if anywhere is the i8259 connect in external int mode */
  91 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
  92
  93 static DEFINE_SPINLOCK(ioapic_lock);
  94 DEFINE_SPINLOCK(vector_lock);
  95
  96 /*
  97  * # of IRQ routing registers
  98  */
  99 int nr_ioapic_registers[MAX_IO_APICS];
 100
 101 /*
 102  * Rough estimation of how many shared IRQs there are, can
 103  * be changed anytime.
 104  */
 105 #define MAX_PLUS_SHARED_IRQS NR_IRQS
 106 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
 107
 108 /*
 109  * This is performance-critical, we want to do it O(1)
 110  *
 111  * the indexing order of this array favors 1:1 mappings
 112  * between pins and IRQs.
 113  */
 114
 115 static struct irq_pin_list {
 116         short apic, pin, next;
 117 } irq_2_pin[PIN_MAP_SIZE];
 118
 119 struct io_apic {
 120         unsigned int index;
 121         unsigned int unused[3];
 122         unsigned int data;
 123 };
 124
 125 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
 126 {
 127         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
 128                 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
 129 }
 130
 131 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
 132 {
 133         struct io_apic __iomem *io_apic = io_apic_base(apic);
 134         writel(reg, &io_apic->index);
 135         return readl(&io_apic->data);
 136 }
 137
 138 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
 139 {
 140         struct io_apic __iomem *io_apic = io_apic_base(apic);
 141         writel(reg, &io_apic->index);
 142         writel(value, &io_apic->data);
 143 }
 144
 145 /*
 146  * Re-write a value: to be used for read-modify-write
 147  * cycles where the read already set up the index register.
 148  */
 149 static inline void io_apic_modify(unsigned int apic, unsigned int value)
 150 {
 151         struct io_apic __iomem *io_apic = io_apic_base(apic);
 152         writel(value, &io_apic->data);
 153 }
 154
 155 static int io_apic_level_ack_pending(unsigned int irq)
 156 {
 157         struct irq_pin_list *entry;
 158         unsigned long flags;
 159         int pending = 0;
 160
 161         spin_lock_irqsave(&ioapic_lock, flags);
 162         entry = irq_2_pin + irq;
 163         for (;;) {
 164                 unsigned int reg;
 165                 int pin;
 166
 167                 pin = entry->pin;
 168                 if (pin == -1)
 169                         break;
 170                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
 171                 /* Is the remote IRR bit set? */
 172                 pending |= (reg >> 14) & 1;
 173                 if (!entry->next)
 174                         break;
 175                 entry = irq_2_pin + entry->next;
 176         }
 177         spin_unlock_irqrestore(&ioapic_lock, flags);
 178         return pending;
 179 }
 180
 181 /*
 182  * Synchronize the IO-APIC and the CPU by doing
 183  * a dummy read from the IO-APIC
 184  */
 185 static inline void io_apic_sync(unsigned int apic)
 186 {
 187         struct io_apic __iomem *io_apic = io_apic_base(apic);
 188         readl(&io_apic->data);
 189 }
 190
 191 #define __DO_ACTION(R, ACTION, FINAL)                                   \
 192                                                                         \
 193 {                                                                       \
 194         int pin;                                                        \
 195         struct irq_pin_list *entry = irq_2_pin + irq;                   \
 196                                                                         \
 197         BUG_ON(irq >= NR_IRQS);                                         \
 198         for (;;) {                                                      \
 199                 unsigned int reg;                                       \
 200                 pin = entry->pin;                                       \
 201                 if (pin == -1)                                          \
 202                         break;                                          \
 203                 reg = io_apic_read(entry->apic, 0x10 + R + pin*2);      \
 204                 reg ACTION;                                             \
 205                 io_apic_modify(entry->apic, reg);                       \
 206                 FINAL;                                                  \
 207                 if (!entry->next)                                       \
 208                         break;                                          \
 209                 entry = irq_2_pin + entry->next;                        \
 210         }                                                               \
 211 }
 212
 213 union entry_union {
 214         struct { u32 w1, w2; };
 215         struct IO_APIC_route_entry entry;
 216 };
 217
 218 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
 219 {
 220         union entry_union eu;
 221         unsigned long flags;
 222         spin_lock_irqsave(&ioapic_lock, flags);
 223         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
 224         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
 225         spin_unlock_irqrestore(&ioapic_lock, flags);
 226         return eu.entry;
 227 }
 228
 229 /*
 230  * When we write a new IO APIC routing entry, we need to write the high
 231  * word first! If the mask bit in the low word is clear, we will enable
 232  * the interrupt, and we need to make sure the entry is fully populated
 233  * before that happens.
 234  */
 235 static void
 236 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 237 {
 238         union entry_union eu;
 239         eu.entry = e;
 240         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 241         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 242 }
 243
 244 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 245 {
 246         unsigned long flags;
 247         spin_lock_irqsave(&ioapic_lock, flags);
 248         __ioapic_write_entry(apic, pin, e);
 249         spin_unlock_irqrestore(&ioapic_lock, flags);
 250 }
 251
 252 /*
 253  * When we mask an IO APIC routing entry, we need to write the low
 254  * word first, in order to set the mask bit before we change the
 255  * high bits!
 256  */
 257 static void ioapic_mask_entry(int apic, int pin)
 258 {
 259         unsigned long flags;
 260         union entry_union eu = { .entry.mask = 1 };
 261
 262         spin_lock_irqsave(&ioapic_lock, flags);
 263         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 264         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 265         spin_unlock_irqrestore(&ioapic_lock, flags);
 266 }
 267
 268 #ifdef CONFIG_SMP
 269 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
 270 {
 271         int apic, pin;
 272         struct irq_pin_list *entry = irq_2_pin + irq;
 273
 274         BUG_ON(irq >= NR_IRQS);
 275         for (;;) {
 276                 unsigned int reg;
 277                 apic = entry->apic;
 278                 pin = entry->pin;
 279                 if (pin == -1)
 280                         break;
 281                 io_apic_write(apic, 0x11 + pin*2, dest);
 282                 reg = io_apic_read(apic, 0x10 + pin*2);
 283                 reg &= ~0x000000ff;
 284                 reg |= vector;
 285                 io_apic_modify(apic, reg);
 286                 if (!entry->next)
 287                         break;
 288                 entry = irq_2_pin + entry->next;
 289         }
 290 }
 291
 292 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
 293 {
 294         struct irq_cfg *cfg = irq_cfg + irq;
 295         unsigned long flags;
 296         unsigned int dest;
 297         cpumask_t tmp;
 298
 299         cpus_and(tmp, mask, cpu_online_map);
 300         if (cpus_empty(tmp))
 301                 return;
 302
 303         if (assign_irq_vector(irq, mask))
 304                 return;
 305
 306         cpus_and(tmp, cfg->domain, mask);
 307         dest = cpu_mask_to_apicid(tmp);
 308
 309         /*
 310          * Only the high 8 bits are valid.
 311          */
 312         dest = SET_APIC_LOGICAL_ID(dest);
 313
 314         spin_lock_irqsave(&ioapic_lock, flags);
 315         __target_IO_APIC_irq(irq, dest, cfg->vector);
 316         irq_desc[irq].affinity = mask;
 317         spin_unlock_irqrestore(&ioapic_lock, flags);
 318 }
 319 #endif
 320
 321 /*
 322  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
 323  * shared ISA-space IRQs, so we have to support them. We are super
 324  * fast in the common case, and fast for shared ISA-space IRQs.
 325  */
 326 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 327 {
 328         static int first_free_entry = NR_IRQS;
 329         struct irq_pin_list *entry = irq_2_pin + irq;
 330
 331         BUG_ON(irq >= NR_IRQS);
 332         while (entry->next)
 333                 entry = irq_2_pin + entry->next;
 334
 335         if (entry->pin != -1) {
 336                 entry->next = first_free_entry;
 337                 entry = irq_2_pin + entry->next;
 338                 if (++first_free_entry >= PIN_MAP_SIZE)
 339                         panic("io_apic.c: ran out of irq_2_pin entries!");
 340         }
 341         entry->apic = apic;
 342         entry->pin = pin;
 343 }
 344
 345
 346 #define DO_ACTION(name,R,ACTION, FINAL)                                 \
 347                                                                         \
 348         static void name##_IO_APIC_irq (unsigned int irq)               \
 349         __DO_ACTION(R, ACTION, FINAL)
 350
 351 DO_ACTION( __mask,             0, |= 0x00010000, io_apic_sync(entry->apic) )
 352                                                 /* mask = 1 */
 353 DO_ACTION( __unmask,           0, &= 0xfffeffff, )
 354                                                 /* mask = 0 */
 355
 356 static void mask_IO_APIC_irq (unsigned int irq)
 357 {
 358         unsigned long flags;
 359
 360         spin_lock_irqsave(&ioapic_lock, flags);
 361         __mask_IO_APIC_irq(irq);
 362         spin_unlock_irqrestore(&ioapic_lock, flags);
 363 }
 364
 365 static void unmask_IO_APIC_irq (unsigned int irq)
 366 {
 367         unsigned long flags;
 368
 369         spin_lock_irqsave(&ioapic_lock, flags);
 370         __unmask_IO_APIC_irq(irq);
 371         spin_unlock_irqrestore(&ioapic_lock, flags);
 372 }
 373
 374 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 375 {
 376         struct IO_APIC_route_entry entry;
 377
 378         /* Check delivery_mode to be sure we're not clearing an SMI pin */
 379         entry = ioapic_read_entry(apic, pin);
 380         if (entry.delivery_mode == dest_SMI)
 381                 return;
 382         /*
 383          * Disable it in the IO-APIC irq-routing table:
 384          */
 385         ioapic_mask_entry(apic, pin);
 386 }
 387
 388 static void clear_IO_APIC (void)
 389 {
 390         int apic, pin;
 391
 392         for (apic = 0; apic < nr_ioapics; apic++)
 393                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 394                         clear_IO_APIC_pin(apic, pin);
 395 }
 396
 397 int skip_ioapic_setup;
 398 int ioapic_force;
 399
 400 static int __init parse_noapic(char *str)
 401 {
 402         disable_ioapic_setup();
 403         return 0;
 404 }
 405 early_param("noapic", parse_noapic);
 406
 407 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
 408 static int __init disable_timer_pin_setup(char *arg)
 409 {
 410         disable_timer_pin_1 = 1;
 411         return 1;
 412 }
 413 __setup("disable_timer_pin_1", disable_timer_pin_setup);
 414
 415 static int __init setup_disable_8254_timer(char *s)
 416 {
 417         timer_over_8254 = -1;
 418         return 1;
 419 }
 420 static int __init setup_enable_8254_timer(char *s)
 421 {
 422         timer_over_8254 = 2;
 423         return 1;
 424 }
 425
 426 __setup("disable_8254_timer", setup_disable_8254_timer);
 427 __setup("enable_8254_timer", setup_enable_8254_timer);
 428
 429
 430 /*
 431  * Find the IRQ entry number of a certain pin.
 432  */
 433 static int find_irq_entry(int apic, int pin, int type)
 434 {
 435         int i;
 436
 437         for (i = 0; i < mp_irq_entries; i++)
 438                 if (mp_irqs[i].mpc_irqtype == type &&
 439                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
 440                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
 441                     mp_irqs[i].mpc_dstirq == pin)
 442                         return i;
 443
 444         return -1;
 445 }
 446
 447 /*
 448  * Find the pin to which IRQ[irq] (ISA) is connected
 449  */
 450 static int __init find_isa_irq_pin(int irq, int type)
 451 {
 452         int i;
 453
 454         for (i = 0; i < mp_irq_entries; i++) {
 455                 int lbus = mp_irqs[i].mpc_srcbus;
 456
 457                 if (test_bit(lbus, mp_bus_not_pci) &&
 458                     (mp_irqs[i].mpc_irqtype == type) &&
 459                     (mp_irqs[i].mpc_srcbusirq == irq))
 460
 461                         return mp_irqs[i].mpc_dstirq;
 462         }
 463         return -1;
 464 }
 465
 466 static int __init find_isa_irq_apic(int irq, int type)
 467 {
 468         int i;
 469
 470         for (i = 0; i < mp_irq_entries; i++) {
 471                 int lbus = mp_irqs[i].mpc_srcbus;
 472
 473                 if (test_bit(lbus, mp_bus_not_pci) &&
 474                     (mp_irqs[i].mpc_irqtype == type) &&
 475                     (mp_irqs[i].mpc_srcbusirq == irq))
 476                         break;
 477         }
 478         if (i < mp_irq_entries) {
 479                 int apic;
 480                 for(apic = 0; apic < nr_ioapics; apic++) {
 481                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
 482                                 return apic;
 483                 }
 484         }
 485
 486         return -1;
 487 }
 488
 489 /*
 490  * Find a specific PCI IRQ entry.
 491  * Not an __init, possibly needed by modules
 492  */
 493 static int pin_2_irq(int idx, int apic, int pin);
 494
 495 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
 496 {
 497         int apic, i, best_guess = -1;
 498
 499         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
 500                 bus, slot, pin);
 501         if (mp_bus_id_to_pci_bus[bus] == -1) {
 502                 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
 503                 return -1;
 504         }
 505         for (i = 0; i < mp_irq_entries; i++) {
 506                 int lbus = mp_irqs[i].mpc_srcbus;
 507
 508                 for (apic = 0; apic < nr_ioapics; apic++)
 509                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
 510                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
 511                                 break;
 512
 513                 if (!test_bit(lbus, mp_bus_not_pci) &&
 514                     !mp_irqs[i].mpc_irqtype &&
 515                     (bus == lbus) &&
 516                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
 517                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
 518
 519                         if (!(apic || IO_APIC_IRQ(irq)))
 520                                 continue;
 521
 522                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
 523                                 return irq;
 524                         /*
 525                          * Use the first all-but-pin matching entry as a
 526                          * best-guess fuzzy result for broken mptables.
 527                          */
 528                         if (best_guess < 0)
 529                                 best_guess = irq;
 530                 }
 531         }
 532         BUG_ON(best_guess >= NR_IRQS);
 533         return best_guess;
 534 }
 535
 536 /* ISA interrupts are always polarity zero edge triggered,
 537  * when listed as conforming in the MP table. */
 538
 539 #define default_ISA_trigger(idx)        (0)
 540 #define default_ISA_polarity(idx)       (0)
 541
 542 /* PCI interrupts are always polarity one level triggered,
 543  * when listed as conforming in the MP table. */
 544
 545 #define default_PCI_trigger(idx)        (1)
 546 #define default_PCI_polarity(idx)       (1)
 547
 548 static int __init MPBIOS_polarity(int idx)
 549 {
 550         int bus = mp_irqs[idx].mpc_srcbus;
 551         int polarity;
 552
 553         /*
 554          * Determine IRQ line polarity (high active or low active):
 555          */
 556         switch (mp_irqs[idx].mpc_irqflag & 3)
 557         {
 558                 case 0: /* conforms, ie. bus-type dependent polarity */
 559                         if (test_bit(bus, mp_bus_not_pci))
 560                                 polarity = default_ISA_polarity(idx);
 561                         else
 562                                 polarity = default_PCI_polarity(idx);
 563                         break;
 564                 case 1: /* high active */
 565                 {
 566                         polarity = 0;
 567                         break;
 568                 }
 569                 case 2: /* reserved */
 570                 {
 571                         printk(KERN_WARNING "broken BIOS!!\n");
 572                         polarity = 1;
 573                         break;
 574                 }
 575                 case 3: /* low active */
 576                 {
 577                         polarity = 1;
 578                         break;
 579                 }
 580                 default: /* invalid */
 581                 {
 582                         printk(KERN_WARNING "broken BIOS!!\n");
 583                         polarity = 1;
 584                         break;
 585                 }
 586         }
 587         return polarity;
 588 }
 589
 590 static int MPBIOS_trigger(int idx)
 591 {
 592         int bus = mp_irqs[idx].mpc_srcbus;
 593         int trigger;
 594
 595         /*
 596          * Determine IRQ trigger mode (edge or level sensitive):
 597          */
 598         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
 599         {
 600                 case 0: /* conforms, ie. bus-type dependent */
 601                         if (test_bit(bus, mp_bus_not_pci))
 602                                 trigger = default_ISA_trigger(idx);
 603                         else
 604                                 trigger = default_PCI_trigger(idx);
 605                         break;
 606                 case 1: /* edge */
 607                 {
 608                         trigger = 0;
 609                         break;
 610                 }
 611                 case 2: /* reserved */
 612                 {
 613                         printk(KERN_WARNING "broken BIOS!!\n");
 614                         trigger = 1;
 615                         break;
 616                 }
 617                 case 3: /* level */
 618                 {
 619                         trigger = 1;
 620                         break;
 621                 }
 622                 default: /* invalid */
 623                 {
 624                         printk(KERN_WARNING "broken BIOS!!\n");
 625                         trigger = 0;
 626                         break;
 627                 }
 628         }
 629         return trigger;
 630 }
 631
 632 static inline int irq_polarity(int idx)
 633 {
 634         return MPBIOS_polarity(idx);
 635 }
 636
 637 static inline int irq_trigger(int idx)
 638 {
 639         return MPBIOS_trigger(idx);
 640 }
 641
 642 static int pin_2_irq(int idx, int apic, int pin)
 643 {
 644         int irq, i;
 645         int bus = mp_irqs[idx].mpc_srcbus;
 646
 647         /*
 648          * Debugging check, we are in big trouble if this message pops up!
 649          */
 650         if (mp_irqs[idx].mpc_dstirq != pin)
 651                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
 652
 653         if (test_bit(bus, mp_bus_not_pci)) {
 654                 irq = mp_irqs[idx].mpc_srcbusirq;
 655         } else {
 656                 /*
 657                  * PCI IRQs are mapped in order
 658                  */
 659                 i = irq = 0;
 660                 while (i < apic)
 661                         irq += nr_ioapic_registers[i++];
 662                 irq += pin;
 663         }
 664         BUG_ON(irq >= NR_IRQS);
 665         return irq;
 666 }
 667
 668 static int __assign_irq_vector(int irq, cpumask_t mask)
 669 {
 670         /*
 671          * NOTE! The local APIC isn't very good at handling
 672          * multiple interrupts at the same interrupt level.
 673          * As the interrupt level is determined by taking the
 674          * vector number and shifting that right by 4, we
 675          * want to spread these out a bit so that they don't
 676          * all fall in the same interrupt level.
 677          *
 678          * Also, we've got to be careful not to trash gate
 679          * 0x80, because int 0x80 is hm, kind of importantish. ;)
 680          */
 681         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
 682         unsigned int old_vector;
 683         int cpu;
 684         struct irq_cfg *cfg;
 685
 686         BUG_ON((unsigned)irq >= NR_IRQS);
 687         cfg = &irq_cfg[irq];
 688
 689         /* Only try and allocate irqs on cpus that are present */
 690         cpus_and(mask, mask, cpu_online_map);
 691
 692         if ((cfg->move_in_progress) || cfg->move_cleanup_count)
 693                 return -EBUSY;
 694
 695         old_vector = cfg->vector;
 696         if (old_vector) {
 697                 cpumask_t tmp;
 698                 cpus_and(tmp, cfg->domain, mask);
 699                 if (!cpus_empty(tmp))
 700                         return 0;
 701         }
 702
 703         for_each_cpu_mask(cpu, mask) {
 704                 cpumask_t domain, new_mask;
 705                 int new_cpu;
 706                 int vector, offset;
 707
 708                 domain = vector_allocation_domain(cpu);
 709                 cpus_and(new_mask, domain, cpu_online_map);
 710
 711                 vector = current_vector;
 712                 offset = current_offset;
 713 next:
 714                 vector += 8;
 715                 if (vector >= FIRST_SYSTEM_VECTOR) {
 716                         /* If we run out of vectors on large boxen, must share them. */
 717                         offset = (offset + 1) % 8;
 718                         vector = FIRST_DEVICE_VECTOR + offset;
 719                 }
 720                 if (unlikely(current_vector == vector))
 721                         continue;
 722                 if (vector == IA32_SYSCALL_VECTOR)
 723                         goto next;
 724                 for_each_cpu_mask(new_cpu, new_mask)
 725                         if (per_cpu(vector_irq, new_cpu)[vector] != -1)
 726                                 goto next;
 727                 /* Found one! */
 728                 current_vector = vector;
 729                 current_offset = offset;
 730                 if (old_vector) {
 731                         cfg->move_in_progress = 1;
 732                         cfg->old_domain = cfg->domain;
 733                 }
 734                 for_each_cpu_mask(new_cpu, new_mask)
 735                         per_cpu(vector_irq, new_cpu)[vector] = irq;
 736                 cfg->vector = vector;
 737                 cfg->domain = domain;
 738                 return 0;
 739         }
 740         return -ENOSPC;
 741 }
 742
 743 static int assign_irq_vector(int irq, cpumask_t mask)
 744 {
 745         int err;
 746         unsigned long flags;
 747
 748         spin_lock_irqsave(&vector_lock, flags);
 749         err = __assign_irq_vector(irq, mask);
 750         spin_unlock_irqrestore(&vector_lock, flags);
 751         return err;
 752 }
 753
 754 static void __clear_irq_vector(int irq)
 755 {
 756         struct irq_cfg *cfg;
 757         cpumask_t mask;
 758         int cpu, vector;
 759
 760         BUG_ON((unsigned)irq >= NR_IRQS);
 761         cfg = &irq_cfg[irq];
 762         BUG_ON(!cfg->vector);
 763
 764         vector = cfg->vector;
 765         cpus_and(mask, cfg->domain, cpu_online_map);
 766         for_each_cpu_mask(cpu, mask)
 767                 per_cpu(vector_irq, cpu)[vector] = -1;
 768
 769         cfg->vector = 0;
 770         cfg->domain = CPU_MASK_NONE;
 771 }
 772
 773 void __setup_vector_irq(int cpu)
 774 {
 775         /* Initialize vector_irq on a new cpu */
 776         /* This function must be called with vector_lock held */
 777         int irq, vector;
 778
 779         /* Mark the inuse vectors */
 780         for (irq = 0; irq < NR_IRQS; ++irq) {
 781                 if (!cpu_isset(cpu, irq_cfg[irq].domain))
 782                         continue;
 783                 vector = irq_cfg[irq].vector;
 784                 per_cpu(vector_irq, cpu)[vector] = irq;
 785         }
 786         /* Mark the free vectors */
 787         for (vector = 0; vector < NR_VECTORS; ++vector) {
 788                 irq = per_cpu(vector_irq, cpu)[vector];
 789                 if (irq < 0)
 790                         continue;
 791                 if (!cpu_isset(cpu, irq_cfg[irq].domain))
 792                         per_cpu(vector_irq, cpu)[vector] = -1;
 793         }
 794 }
 795
 796
 797 static struct irq_chip ioapic_chip;
 798
 799 static void ioapic_register_intr(int irq, unsigned long trigger)
 800 {
 801         if (trigger) {
 802                 irq_desc[irq].status |= IRQ_LEVEL;
 803                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
 804                                               handle_fasteoi_irq, "fasteoi");
 805         } else {
 806                 irq_desc[irq].status &= ~IRQ_LEVEL;
 807                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
 808                                               handle_edge_irq, "edge");
 809         }
 810 }
 811
 812 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
 813                               int trigger, int polarity)
 814 {
 815         struct irq_cfg *cfg = irq_cfg + irq;
 816         struct IO_APIC_route_entry entry;
 817         cpumask_t mask;
 818
 819         if (!IO_APIC_IRQ(irq))
 820                 return;
 821
 822         mask = TARGET_CPUS;
 823         if (assign_irq_vector(irq, mask))
 824                 return;
 825
 826         cpus_and(mask, cfg->domain, mask);
 827
 828         apic_printk(APIC_VERBOSE,KERN_DEBUG
 829                     "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
 830                     "IRQ %d Mode:%i Active:%i)\n",
 831                     apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector,
 832                     irq, trigger, polarity);
 833
 834         /*
 835          * add it to the IO-APIC irq-routing table:
 836          */
 837         memset(&entry,0,sizeof(entry));
 838
 839         entry.delivery_mode = INT_DELIVERY_MODE;
 840         entry.dest_mode = INT_DEST_MODE;
 841         entry.dest = cpu_mask_to_apicid(mask);
 842         entry.mask = 0;                         /* enable IRQ */
 843         entry.trigger = trigger;
 844         entry.polarity = polarity;
 845         entry.vector = cfg->vector;
 846
 847         /* Mask level triggered irqs.
 848          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
 849          */
 850         if (trigger)
 851                 entry.mask = 1;
 852
 853         ioapic_register_intr(irq, trigger);
 854         if (irq < 16)
 855                 disable_8259A_irq(irq);
 856
 857         ioapic_write_entry(apic, pin, entry);
 858 }
 859
 860 static void __init setup_IO_APIC_irqs(void)
 861 {
 862         int apic, pin, idx, irq, first_notcon = 1;
 863
 864         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
 865
 866         for (apic = 0; apic < nr_ioapics; apic++) {
 867         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
 868
 869                 idx = find_irq_entry(apic,pin,mp_INT);
 870                 if (idx == -1) {
 871                         if (first_notcon) {
 872                                 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
 873                                 first_notcon = 0;
 874                         } else
 875                                 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
 876                         continue;
 877                 }
 878                 if (!first_notcon) {
 879                         apic_printk(APIC_VERBOSE, " not connected.\n");
 880                         first_notcon = 1;
 881                 }
 882
 883                 irq = pin_2_irq(idx, apic, pin);
 884                 add_pin_to_irq(irq, apic, pin);
 885
 886                 setup_IO_APIC_irq(apic, pin, irq,
 887                                   irq_trigger(idx), irq_polarity(idx));
 888         }
 889         }
 890
 891         if (!first_notcon)
 892                 apic_printk(APIC_VERBOSE, " not connected.\n");
 893 }
 894
 895 /*
 896  * Set up the 8259A-master output pin as broadcast to all
 897  * CPUs.
 898  */
 899 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
 900 {
 901         struct IO_APIC_route_entry entry;
 902         unsigned long flags;
 903
 904         memset(&entry,0,sizeof(entry));
 905
 906         disable_8259A_irq(0);
 907
 908         /* mask LVT0 */
 909         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
 910
 911         /*
 912          * We use logical delivery to get the timer IRQ
 913          * to the first CPU.
 914          */
 915         entry.dest_mode = INT_DEST_MODE;
 916         entry.mask = 0;                                 /* unmask IRQ now */
 917         entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
 918         entry.delivery_mode = INT_DELIVERY_MODE;
 919         entry.polarity = 0;
 920         entry.trigger = 0;
 921         entry.vector = vector;
 922
 923         /*
 924          * The timer IRQ doesn't have to know that behind the
 925          * scene we have a 8259A-master in AEOI mode ...
 926          */
 927         set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
 928
 929         /*
 930          * Add it to the IO-APIC irq-routing table:
 931          */
 932         spin_lock_irqsave(&ioapic_lock, flags);
 933         io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
 934         io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
 935         spin_unlock_irqrestore(&ioapic_lock, flags);
 936
 937         enable_8259A_irq(0);
 938 }
 939
 940 void __apicdebuginit print_IO_APIC(void)
 941 {
 942         int apic, i;
 943         union IO_APIC_reg_00 reg_00;
 944         union IO_APIC_reg_01 reg_01;
 945         union IO_APIC_reg_02 reg_02;
 946         unsigned long flags;
 947
 948         if (apic_verbosity == APIC_QUIET)
 949                 return;
 950
 951         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
 952         for (i = 0; i < nr_ioapics; i++)
 953                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
 954                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
 955
 956         /*
 957          * We are a bit conservative about what we expect.  We have to
 958          * know about every hardware change ASAP.
 959          */
 960         printk(KERN_INFO "testing the IO APIC.......................\n");
 961
 962         for (apic = 0; apic < nr_ioapics; apic++) {
 963
 964         spin_lock_irqsave(&ioapic_lock, flags);
 965         reg_00.raw = io_apic_read(apic, 0);
 966         reg_01.raw = io_apic_read(apic, 1);
 967         if (reg_01.bits.version >= 0x10)
 968                 reg_02.raw = io_apic_read(apic, 2);
 969         spin_unlock_irqrestore(&ioapic_lock, flags);
 970
 971         printk("\n");
 972         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
 973         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
 974         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
 975
 976         printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
 977         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
 978
 979         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
 980         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
 981
 982         if (reg_01.bits.version >= 0x10) {
 983                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
 984                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
 985         }
 986
 987         printk(KERN_DEBUG ".... IRQ redirection table:\n");
 988
 989         printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
 990                           " Stat Dmod Deli Vect:   \n");
 991
 992         for (i = 0; i <= reg_01.bits.entries; i++) {
 993                 struct IO_APIC_route_entry entry;
 994
 995                 entry = ioapic_read_entry(apic, i);
 996
 997                 printk(KERN_DEBUG " %02x %03X ",
 998                         i,
 999                         entry.dest
1000                 );
1001
1002                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1003                         entry.mask,
1004                         entry.trigger,
1005                         entry.irr,
1006                         entry.polarity,
1007                         entry.delivery_status,
1008                         entry.dest_mode,
1009                         entry.delivery_mode,
1010                         entry.vector
1011                 );
1012         }
1013         }
1014         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1015         for (i = 0; i < NR_IRQS; i++) {
1016                 struct irq_pin_list *entry = irq_2_pin + i;
1017                 if (entry->pin < 0)
1018                         continue;
1019                 printk(KERN_DEBUG "IRQ%d ", i);
1020                 for (;;) {
1021                         printk("-> %d:%d", entry->apic, entry->pin);
1022                         if (!entry->next)
1023                                 break;
1024                         entry = irq_2_pin + entry->next;
1025                 }
1026                 printk("\n");
1027         }
1028
1029         printk(KERN_INFO ".................................... done.\n");
1030
1031         return;
1032 }
1033
1034 #if 0
1035
1036 static __apicdebuginit void print_APIC_bitfield (int base)
1037 {
1038         unsigned int v;
1039         int i, j;
1040
1041         if (apic_verbosity == APIC_QUIET)
1042                 return;
1043
1044         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1045         for (i = 0; i < 8; i++) {
1046                 v = apic_read(base + i*0x10);
1047                 for (j = 0; j < 32; j++) {
1048                         if (v & (1<<j))
1049                                 printk("1");
1050                         else
1051                                 printk("0");
1052                 }
1053                 printk("\n");
1054         }
1055 }
1056
1057 void __apicdebuginit print_local_APIC(void * dummy)
1058 {
1059         unsigned int v, ver, maxlvt;
1060
1061         if (apic_verbosity == APIC_QUIET)
1062                 return;
1063
1064         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1065                 smp_processor_id(), hard_smp_processor_id());
1066         v = apic_read(APIC_ID);
1067         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, GET_APIC_ID(v));
1068         v = apic_read(APIC_LVR);
1069         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1070         ver = GET_APIC_VERSION(v);
1071         maxlvt = get_maxlvt();
1072
1073         v = apic_read(APIC_TASKPRI);
1074         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1075
1076         v = apic_read(APIC_ARBPRI);
1077         printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1078                 v & APIC_ARBPRI_MASK);
1079         v = apic_read(APIC_PROCPRI);
1080         printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1081
1082         v = apic_read(APIC_EOI);
1083         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1084         v = apic_read(APIC_RRR);
1085         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1086         v = apic_read(APIC_LDR);
1087         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1088         v = apic_read(APIC_DFR);
1089         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1090         v = apic_read(APIC_SPIV);
1091         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1092
1093         printk(KERN_DEBUG "... APIC ISR field:\n");
1094         print_APIC_bitfield(APIC_ISR);
1095         printk(KERN_DEBUG "... APIC TMR field:\n");
1096         print_APIC_bitfield(APIC_TMR);
1097         printk(KERN_DEBUG "... APIC IRR field:\n");
1098         print_APIC_bitfield(APIC_IRR);
1099
1100         v = apic_read(APIC_ESR);
1101         printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1102
1103         v = apic_read(APIC_ICR);
1104         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1105         v = apic_read(APIC_ICR2);
1106         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1107
1108         v = apic_read(APIC_LVTT);
1109         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1110
1111         if (maxlvt > 3) {                       /* PC is LVT#4. */
1112                 v = apic_read(APIC_LVTPC);
1113                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1114         }
1115         v = apic_read(APIC_LVT0);
1116         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1117         v = apic_read(APIC_LVT1);
1118         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1119
1120         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1121                 v = apic_read(APIC_LVTERR);
1122                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1123         }
1124
1125         v = apic_read(APIC_TMICT);
1126         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1127         v = apic_read(APIC_TMCCT);
1128         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1129         v = apic_read(APIC_TDCR);
1130         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1131         printk("\n");
1132 }
1133
1134 void print_all_local_APICs (void)
1135 {
1136         on_each_cpu(print_local_APIC, NULL, 1, 1);
1137 }
1138
1139 void __apicdebuginit print_PIC(void)
1140 {
1141         unsigned int v;
1142         unsigned long flags;
1143
1144         if (apic_verbosity == APIC_QUIET)
1145                 return;
1146
1147         printk(KERN_DEBUG "\nprinting PIC contents\n");
1148
1149         spin_lock_irqsave(&i8259A_lock, flags);
1150
1151         v = inb(0xa1) << 8 | inb(0x21);
1152         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1153
1154         v = inb(0xa0) << 8 | inb(0x20);
1155         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1156
1157         outb(0x0b,0xa0);
1158         outb(0x0b,0x20);
1159         v = inb(0xa0) << 8 | inb(0x20);
1160         outb(0x0a,0xa0);
1161         outb(0x0a,0x20);
1162
1163         spin_unlock_irqrestore(&i8259A_lock, flags);
1164
1165         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1166
1167         v = inb(0x4d1) << 8 | inb(0x4d0);
1168         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1169 }
1170
1171 #endif  /*  0  */
1172
1173 static void __init enable_IO_APIC(void)
1174 {
1175         union IO_APIC_reg_01 reg_01;
1176         int i8259_apic, i8259_pin;
1177         int i, apic;
1178         unsigned long flags;
1179
1180         for (i = 0; i < PIN_MAP_SIZE; i++) {
1181                 irq_2_pin[i].pin = -1;
1182                 irq_2_pin[i].next = 0;
1183         }
1184
1185         /*
1186          * The number of IO-APIC IRQ registers (== #pins):
1187          */
1188         for (apic = 0; apic < nr_ioapics; apic++) {
1189                 spin_lock_irqsave(&ioapic_lock, flags);
1190                 reg_01.raw = io_apic_read(apic, 1);
1191                 spin_unlock_irqrestore(&ioapic_lock, flags);
1192                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1193         }
1194         for(apic = 0; apic < nr_ioapics; apic++) {
1195                 int pin;
1196                 /* See if any of the pins is in ExtINT mode */
1197                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1198                         struct IO_APIC_route_entry entry;
1199                         entry = ioapic_read_entry(apic, pin);
1200
1201                         /* If the interrupt line is enabled and in ExtInt mode
1202                          * I have found the pin where the i8259 is connected.
1203                          */
1204                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1205                                 ioapic_i8259.apic = apic;
1206                                 ioapic_i8259.pin  = pin;
1207                                 goto found_i8259;
1208                         }
1209                 }
1210         }
1211  found_i8259:
1212         /* Look to see what if the MP table has reported the ExtINT */
1213         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1214         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1215         /* Trust the MP table if nothing is setup in the hardware */
1216         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1217                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1218                 ioapic_i8259.pin  = i8259_pin;
1219                 ioapic_i8259.apic = i8259_apic;
1220         }
1221         /* Complain if the MP table and the hardware disagree */
1222         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1223                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1224         {
1225                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1226         }
1227
1228         /*
1229          * Do not trust the IO-APIC being empty at bootup
1230          */
1231         clear_IO_APIC();
1232 }
1233
1234 /*
1235  * Not an __init, needed by the reboot code
1236  */
1237 void disable_IO_APIC(void)
1238 {
1239         /*
1240          * Clear the IO-APIC before rebooting:
1241          */
1242         clear_IO_APIC();
1243
1244         /*
1245          * If the i8259 is routed through an IOAPIC
1246          * Put that IOAPIC in virtual wire mode
1247          * so legacy interrupts can be delivered.
1248          */
1249         if (ioapic_i8259.pin != -1) {
1250                 struct IO_APIC_route_entry entry;
1251
1252                 memset(&entry, 0, sizeof(entry));
1253                 entry.mask            = 0; /* Enabled */
1254                 entry.trigger         = 0; /* Edge */
1255                 entry.irr             = 0;
1256                 entry.polarity        = 0; /* High */
1257                 entry.delivery_status = 0;
1258                 entry.dest_mode       = 0; /* Physical */
1259                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1260                 entry.vector          = 0;
1261                 entry.dest          = GET_APIC_ID(apic_read(APIC_ID));
1262
1263                 /*
1264                  * Add it to the IO-APIC irq-routing table:
1265                  */
1266                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1267         }
1268
1269         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1270 }
1271
1272 /*
1273  * There is a nasty bug in some older SMP boards, their mptable lies
1274  * about the timer IRQ. We do the following to work around the situation:
1275  *
1276  *      - timer IRQ defaults to IO-APIC IRQ
1277  *      - if this function detects that timer IRQs are defunct, then we fall
1278  *        back to ISA timer IRQs
1279  */
1280 static int __init timer_irq_works(void)
1281 {
1282         unsigned long t1 = jiffies;
1283
1284         local_irq_enable();
1285         /* Let ten ticks pass... */
1286         mdelay((10 * 1000) / HZ);
1287
1288         /*
1289          * Expect a few ticks at least, to be sure some possible
1290          * glue logic does not lock up after one or two first
1291          * ticks in a non-ExtINT mode.  Also the local APIC
1292          * might have cached one ExtINT interrupt.  Finally, at
1293          * least one tick may be lost due to delays.
1294          */
1295
1296         /* jiffies wrap? */
1297         if (jiffies - t1 > 4)
1298                 return 1;
1299         return 0;
1300 }
1301
1302 /*
1303  * In the SMP+IOAPIC case it might happen that there are an unspecified
1304  * number of pending IRQ events unhandled. These cases are very rare,
1305  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1306  * better to do it this way as thus we do not have to be aware of
1307  * 'pending' interrupts in the IRQ path, except at this point.
1308  */
1309 /*
1310  * Edge triggered needs to resend any interrupt
1311  * that was delayed but this is now handled in the device
1312  * independent code.
1313  */
1314
1315 /*
1316  * Starting up a edge-triggered IO-APIC interrupt is
1317  * nasty - we need to make sure that we get the edge.
1318  * If it is already asserted for some reason, we need
1319  * return 1 to indicate that is was pending.
1320  *
1321  * This is not complete - we should be able to fake
1322  * an edge even if it isn't on the 8259A...
1323  */
1324
1325 static unsigned int startup_ioapic_irq(unsigned int irq)
1326 {
1327         int was_pending = 0;
1328         unsigned long flags;
1329
1330         spin_lock_irqsave(&ioapic_lock, flags);
1331         if (irq < 16) {
1332                 disable_8259A_irq(irq);
1333                 if (i8259A_irq_pending(irq))
1334                         was_pending = 1;
1335         }
1336         __unmask_IO_APIC_irq(irq);
1337         spin_unlock_irqrestore(&ioapic_lock, flags);
1338
1339         return was_pending;
1340 }
1341
1342 static int ioapic_retrigger_irq(unsigned int irq)
1343 {
1344         struct irq_cfg *cfg = &irq_cfg[irq];
1345         cpumask_t mask;
1346         unsigned long flags;
1347
1348         spin_lock_irqsave(&vector_lock, flags);
1349         cpus_clear(mask);
1350         cpu_set(first_cpu(cfg->domain), mask);
1351
1352         send_IPI_mask(mask, cfg->vector);
1353         spin_unlock_irqrestore(&vector_lock, flags);
1354
1355         return 1;
1356 }
1357
1358 /*
1359  * Level and edge triggered IO-APIC interrupts need different handling,
1360  * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1361  * handled with the level-triggered descriptor, but that one has slightly
1362  * more overhead. Level-triggered interrupts cannot be handled with the
1363  * edge-triggered handler, without risking IRQ storms and other ugly
1364  * races.
1365  */
1366
1367 #ifdef CONFIG_SMP
1368 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1369 {
1370         unsigned vector, me;
1371         ack_APIC_irq();
1372         exit_idle();
1373         irq_enter();
1374
1375         me = smp_processor_id();
1376         for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1377                 unsigned int irq;
1378                 struct irq_desc *desc;
1379                 struct irq_cfg *cfg;
1380                 irq = __get_cpu_var(vector_irq)[vector];
1381                 if (irq >= NR_IRQS)
1382                         continue;
1383
1384                 desc = irq_desc + irq;
1385                 cfg = irq_cfg + irq;
1386                 spin_lock(&desc->lock);
1387                 if (!cfg->move_cleanup_count)
1388                         goto unlock;
1389
1390                 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1391                         goto unlock;
1392
1393                 __get_cpu_var(vector_irq)[vector] = -1;
1394                 cfg->move_cleanup_count--;
1395 unlock:
1396                 spin_unlock(&desc->lock);
1397         }
1398
1399         irq_exit();
1400 }
1401
1402 static void irq_complete_move(unsigned int irq)
1403 {
1404         struct irq_cfg *cfg = irq_cfg + irq;
1405         unsigned vector, me;
1406
1407         if (likely(!cfg->move_in_progress))
1408                 return;
1409
1410         vector = ~get_irq_regs()->orig_rax;
1411         me = smp_processor_id();
1412         if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1413                 cpumask_t cleanup_mask;
1414
1415                 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1416                 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1417                 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1418                 cfg->move_in_progress = 0;
1419         }
1420 }
1421 #else
1422 static inline void irq_complete_move(unsigned int irq) {}
1423 #endif
1424
1425 static void ack_apic_edge(unsigned int irq)
1426 {
1427         irq_complete_move(irq);
1428         move_native_irq(irq);
1429         ack_APIC_irq();
1430 }
1431
1432 static void ack_apic_level(unsigned int irq)
1433 {
1434         int do_unmask_irq = 0;
1435
1436         irq_complete_move(irq);
1437 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1438         /* If we are moving the irq we need to mask it */
1439         if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1440                 do_unmask_irq = 1;
1441                 mask_IO_APIC_irq(irq);
1442         }
1443 #endif
1444
1445         /*
1446          * We must acknowledge the irq before we move it or the acknowledge will
1447          * not propagate properly.
1448          */
1449         ack_APIC_irq();
1450
1451         /* Now we can move and renable the irq */
1452         if (unlikely(do_unmask_irq)) {
1453                 /* Only migrate the irq if the ack has been received.
1454                  *
1455                  * On rare occasions the broadcast level triggered ack gets
1456                  * delayed going to ioapics, and if we reprogram the
1457                  * vector while Remote IRR is still set the irq will never
1458                  * fire again.
1459                  *
1460                  * To prevent this scenario we read the Remote IRR bit
1461                  * of the ioapic.  This has two effects.
1462                  * - On any sane system the read of the ioapic will
1463                  *   flush writes (and acks) going to the ioapic from
1464                  *   this cpu.
1465                  * - We get to see if the ACK has actually been delivered.
1466                  *
1467                  * Based on failed experiments of reprogramming the
1468                  * ioapic entry from outside of irq context starting
1469                  * with masking the ioapic entry and then polling until
1470                  * Remote IRR was clear before reprogramming the
1471                  * ioapic I don't trust the Remote IRR bit to be
1472                  * completey accurate.
1473                  *
1474                  * However there appears to be no other way to plug
1475                  * this race, so if the Remote IRR bit is not
1476                  * accurate and is causing problems then it is a hardware bug
1477                  * and you can go talk to the chipset vendor about it.
1478                  */
1479                 if (!io_apic_level_ack_pending(irq))
1480                         move_masked_irq(irq);
1481                 unmask_IO_APIC_irq(irq);
1482         }
1483 }
1484
1485 static struct irq_chip ioapic_chip __read_mostly = {
1486         .name           = "IO-APIC",
1487         .startup        = startup_ioapic_irq,
1488         .mask           = mask_IO_APIC_irq,
1489         .unmask         = unmask_IO_APIC_irq,
1490         .ack            = ack_apic_edge,
1491         .eoi            = ack_apic_level,
1492 #ifdef CONFIG_SMP
1493         .set_affinity   = set_ioapic_affinity_irq,
1494 #endif
1495         .retrigger      = ioapic_retrigger_irq,
1496 };
1497
1498 static inline void init_IO_APIC_traps(void)
1499 {
1500         int irq;
1501
1502         /*
1503          * NOTE! The local APIC isn't very good at handling
1504          * multiple interrupts at the same interrupt level.
1505          * As the interrupt level is determined by taking the
1506          * vector number and shifting that right by 4, we
1507          * want to spread these out a bit so that they don't
1508          * all fall in the same interrupt level.
1509          *
1510          * Also, we've got to be careful not to trash gate
1511          * 0x80, because int 0x80 is hm, kind of importantish. ;)
1512          */
1513         for (irq = 0; irq < NR_IRQS ; irq++) {
1514                 int tmp = irq;
1515                 if (IO_APIC_IRQ(tmp) && !irq_cfg[tmp].vector) {
1516                         /*
1517                          * Hmm.. We don't have an entry for this,
1518                          * so default to an old-fashioned 8259
1519                          * interrupt if we can..
1520                          */
1521                         if (irq < 16)
1522                                 make_8259A_irq(irq);
1523                         else
1524                                 /* Strange. Oh, well.. */
1525                                 irq_desc[irq].chip = &no_irq_chip;
1526                 }
1527         }
1528 }
1529
1530 static void enable_lapic_irq (unsigned int irq)
1531 {
1532         unsigned long v;
1533
1534         v = apic_read(APIC_LVT0);
1535         apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1536 }
1537
1538 static void disable_lapic_irq (unsigned int irq)
1539 {
1540         unsigned long v;
1541
1542         v = apic_read(APIC_LVT0);
1543         apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1544 }
1545
1546 static void ack_lapic_irq (unsigned int irq)
1547 {
1548         ack_APIC_irq();
1549 }
1550
1551 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1552
1553 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1554         .name = "local-APIC",
1555         .typename = "local-APIC-edge",
1556         .startup = NULL, /* startup_irq() not used for IRQ0 */
1557         .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1558         .enable = enable_lapic_irq,
1559         .disable = disable_lapic_irq,
1560         .ack = ack_lapic_irq,
1561         .end = end_lapic_irq,
1562 };
1563
1564 static void setup_nmi (void)
1565 {
1566         /*
1567          * Dirty trick to enable the NMI watchdog ...
1568          * We put the 8259A master into AEOI mode and
1569          * unmask on all local APICs LVT0 as NMI.
1570          *
1571          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1572          * is from Maciej W. Rozycki - so we do not have to EOI from
1573          * the NMI handler or the timer interrupt.
1574          */
1575         printk(KERN_INFO "activating NMI Watchdog ...");
1576
1577         enable_NMI_through_LVT0(NULL);
1578
1579         printk(" done.\n");
1580 }
1581
1582 /*
1583  * This looks a bit hackish but it's about the only one way of sending
1584  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
1585  * not support the ExtINT mode, unfortunately.  We need to send these
1586  * cycles as some i82489DX-based boards have glue logic that keeps the
1587  * 8259A interrupt line asserted until INTA.  --macro
1588  */
1589 static inline void unlock_ExtINT_logic(void)
1590 {
1591         int apic, pin, i;
1592         struct IO_APIC_route_entry entry0, entry1;
1593         unsigned char save_control, save_freq_select;
1594         unsigned long flags;
1595
1596         pin  = find_isa_irq_pin(8, mp_INT);
1597         apic = find_isa_irq_apic(8, mp_INT);
1598         if (pin == -1)
1599                 return;
1600
1601         spin_lock_irqsave(&ioapic_lock, flags);
1602         *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1603         *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1604         spin_unlock_irqrestore(&ioapic_lock, flags);
1605         clear_IO_APIC_pin(apic, pin);
1606
1607         memset(&entry1, 0, sizeof(entry1));
1608
1609         entry1.dest_mode = 0;                   /* physical delivery */
1610         entry1.mask = 0;                        /* unmask IRQ now */
1611         entry1.dest = hard_smp_processor_id();
1612         entry1.delivery_mode = dest_ExtINT;
1613         entry1.polarity = entry0.polarity;
1614         entry1.trigger = 0;
1615         entry1.vector = 0;
1616
1617         spin_lock_irqsave(&ioapic_lock, flags);
1618         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1619         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1620         spin_unlock_irqrestore(&ioapic_lock, flags);
1621
1622         save_control = CMOS_READ(RTC_CONTROL);
1623         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1624         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1625                    RTC_FREQ_SELECT);
1626         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1627
1628         i = 100;
1629         while (i-- > 0) {
1630                 mdelay(10);
1631                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1632                         i -= 10;
1633         }
1634
1635         CMOS_WRITE(save_control, RTC_CONTROL);
1636         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1637         clear_IO_APIC_pin(apic, pin);
1638
1639         spin_lock_irqsave(&ioapic_lock, flags);
1640         io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1641         io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1642         spin_unlock_irqrestore(&ioapic_lock, flags);
1643 }
1644
1645 /*
1646  * This code may look a bit paranoid, but it's supposed to cooperate with
1647  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
1648  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
1649  * fanatically on his truly buggy board.
1650  *
1651  * FIXME: really need to revamp this for modern platforms only.
1652  */
1653 static inline void check_timer(void)
1654 {
1655         struct irq_cfg *cfg = irq_cfg + 0;
1656         int apic1, pin1, apic2, pin2;
1657
1658         /*
1659          * get/set the timer IRQ vector:
1660          */
1661         disable_8259A_irq(0);
1662         assign_irq_vector(0, TARGET_CPUS);
1663
1664         /*
1665          * Subtle, code in do_timer_interrupt() expects an AEOI
1666          * mode for the 8259A whenever interrupts are routed
1667          * through I/O APICs.  Also IRQ0 has to be enabled in
1668          * the 8259A which implies the virtual wire has to be
1669          * disabled in the local APIC.
1670          */
1671         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1672         init_8259A(1);
1673         if (timer_over_8254 > 0)
1674                 enable_8259A_irq(0);
1675
1676         pin1  = find_isa_irq_pin(0, mp_INT);
1677         apic1 = find_isa_irq_apic(0, mp_INT);
1678         pin2  = ioapic_i8259.pin;
1679         apic2 = ioapic_i8259.apic;
1680
1681         apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1682                 cfg->vector, apic1, pin1, apic2, pin2);
1683
1684         if (pin1 != -1) {
1685                 /*
1686                  * Ok, does IRQ0 through the IOAPIC work?
1687                  */
1688                 unmask_IO_APIC_irq(0);
1689                 if (!no_timer_check && timer_irq_works()) {
1690                         nmi_watchdog_default();
1691                         if (nmi_watchdog == NMI_IO_APIC) {
1692                                 disable_8259A_irq(0);
1693                                 setup_nmi();
1694                                 enable_8259A_irq(0);
1695                         }
1696                         if (disable_timer_pin_1 > 0)
1697                                 clear_IO_APIC_pin(0, pin1);
1698                         return;
1699                 }
1700                 clear_IO_APIC_pin(apic1, pin1);
1701                 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1702                                 "connected to IO-APIC\n");
1703         }
1704
1705         apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1706                                 "through the 8259A ... ");
1707         if (pin2 != -1) {
1708                 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1709                         apic2, pin2);
1710                 /*
1711                  * legacy devices should be connected to IO APIC #0
1712                  */
1713                 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1714                 if (timer_irq_works()) {
1715                         apic_printk(APIC_VERBOSE," works.\n");
1716                         nmi_watchdog_default();
1717                         if (nmi_watchdog == NMI_IO_APIC) {
1718                                 setup_nmi();
1719                         }
1720                         return;
1721                 }
1722                 /*
1723                  * Cleanup, just in case ...
1724                  */
1725                 clear_IO_APIC_pin(apic2, pin2);
1726         }
1727         apic_printk(APIC_VERBOSE," failed.\n");
1728
1729         if (nmi_watchdog == NMI_IO_APIC) {
1730                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1731                 nmi_watchdog = 0;
1732         }
1733
1734         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1735
1736         disable_8259A_irq(0);
1737         irq_desc[0].chip = &lapic_irq_type;
1738         apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector);     /* Fixed mode */
1739         enable_8259A_irq(0);
1740
1741         if (timer_irq_works()) {
1742                 apic_printk(APIC_VERBOSE," works.\n");
1743                 return;
1744         }
1745         apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1746         apic_printk(APIC_VERBOSE," failed.\n");
1747
1748         apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1749
1750         init_8259A(0);
1751         make_8259A_irq(0);
1752         apic_write(APIC_LVT0, APIC_DM_EXTINT);
1753
1754         unlock_ExtINT_logic();
1755
1756         if (timer_irq_works()) {
1757                 apic_printk(APIC_VERBOSE," works.\n");
1758                 return;
1759         }
1760         apic_printk(APIC_VERBOSE," failed :(.\n");
1761         panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1762 }
1763
1764 static int __init notimercheck(char *s)
1765 {
1766         no_timer_check = 1;
1767         return 1;
1768 }
1769 __setup("no_timer_check", notimercheck);
1770
1771 /*
1772  *
1773  * IRQs that are handled by the PIC in the MPS IOAPIC case.
1774  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1775  *   Linux doesn't really care, as it's not actually used
1776  *   for any interrupt handling anyway.
1777  */
1778 #define PIC_IRQS        (1<<2)
1779
1780 void __init setup_IO_APIC(void)
1781 {
1782         enable_IO_APIC();
1783
1784         if (acpi_ioapic)
1785                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
1786         else
1787                 io_apic_irqs = ~PIC_IRQS;
1788
1789         apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1790
1791         sync_Arb_IDs();
1792         setup_IO_APIC_irqs();
1793         init_IO_APIC_traps();
1794         check_timer();
1795         if (!acpi_ioapic)
1796                 print_IO_APIC();
1797 }
1798
1799 struct sysfs_ioapic_data {
1800         struct sys_device dev;
1801         struct IO_APIC_route_entry entry[0];
1802 };
1803 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1804
1805 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1806 {
1807         struct IO_APIC_route_entry *entry;
1808         struct sysfs_ioapic_data *data;
1809         int i;
1810
1811         data = container_of(dev, struct sysfs_ioapic_data, dev);
1812         entry = data->entry;
1813         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1814                 *entry = ioapic_read_entry(dev->id, i);
1815
1816         return 0;
1817 }
1818
1819 static int ioapic_resume(struct sys_device *dev)
1820 {
1821         struct IO_APIC_route_entry *entry;
1822         struct sysfs_ioapic_data *data;
1823         unsigned long flags;
1824         union IO_APIC_reg_00 reg_00;
1825         int i;
1826
1827         data = container_of(dev, struct sysfs_ioapic_data, dev);
1828         entry = data->entry;
1829
1830         spin_lock_irqsave(&ioapic_lock, flags);
1831         reg_00.raw = io_apic_read(dev->id, 0);
1832         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1833                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1834                 io_apic_write(dev->id, 0, reg_00.raw);
1835         }
1836         spin_unlock_irqrestore(&ioapic_lock, flags);
1837         for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1838                 ioapic_write_entry(dev->id, i, entry[i]);
1839
1840         return 0;
1841 }
1842
1843 static struct sysdev_class ioapic_sysdev_class = {
1844         set_kset_name("ioapic"),
1845         .suspend = ioapic_suspend,
1846         .resume = ioapic_resume,
1847 };
1848
1849 static int __init ioapic_init_sysfs(void)
1850 {
1851         struct sys_device * dev;
1852         int i, size, error;
1853
1854         error = sysdev_class_register(&ioapic_sysdev_class);
1855         if (error)
1856                 return error;
1857
1858         for (i = 0; i < nr_ioapics; i++ ) {
1859                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1860                         * sizeof(struct IO_APIC_route_entry);
1861                 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
1862                 if (!mp_ioapic_data[i]) {
1863                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1864                         continue;
1865                 }
1866                 dev = &mp_ioapic_data[i]->dev;
1867                 dev->id = i;
1868                 dev->cls = &ioapic_sysdev_class;
1869                 error = sysdev_register(dev);
1870                 if (error) {
1871                         kfree(mp_ioapic_data[i]);
1872                         mp_ioapic_data[i] = NULL;
1873                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1874                         continue;
1875                 }
1876         }
1877
1878         return 0;
1879 }
1880
1881 device_initcall(ioapic_init_sysfs);
1882
1883 /*
1884  * Dynamic irq allocate and deallocation
1885  */
1886 int create_irq(void)
1887 {
1888         /* Allocate an unused irq */
1889         int irq;
1890         int new;
1891         unsigned long flags;
1892
1893         irq = -ENOSPC;
1894         spin_lock_irqsave(&vector_lock, flags);
1895         for (new = (NR_IRQS - 1); new >= 0; new--) {
1896                 if (platform_legacy_irq(new))
1897                         continue;
1898                 if (irq_cfg[new].vector != 0)
1899                         continue;
1900                 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1901                         irq = new;
1902                 break;
1903         }
1904         spin_unlock_irqrestore(&vector_lock, flags);
1905
1906         if (irq >= 0) {
1907                 dynamic_irq_init(irq);
1908         }
1909         return irq;
1910 }
1911
1912 void destroy_irq(unsigned int irq)
1913 {
1914         unsigned long flags;
1915
1916         dynamic_irq_cleanup(irq);
1917
1918         spin_lock_irqsave(&vector_lock, flags);
1919         __clear_irq_vector(irq);
1920         spin_unlock_irqrestore(&vector_lock, flags);
1921 }
1922
1923 /*
1924  * MSI message composition
1925  */
1926 #ifdef CONFIG_PCI_MSI
1927 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1928 {
1929         struct irq_cfg *cfg = irq_cfg + irq;
1930         int err;
1931         unsigned dest;
1932         cpumask_t tmp;
1933
1934         tmp = TARGET_CPUS;
1935         err = assign_irq_vector(irq, tmp);
1936         if (!err) {
1937                 cpus_and(tmp, cfg->domain, tmp);
1938                 dest = cpu_mask_to_apicid(tmp);
1939
1940                 msg->address_hi = MSI_ADDR_BASE_HI;
1941                 msg->address_lo =
1942                         MSI_ADDR_BASE_LO |
1943                         ((INT_DEST_MODE == 0) ?
1944                                 MSI_ADDR_DEST_MODE_PHYSICAL:
1945                                 MSI_ADDR_DEST_MODE_LOGICAL) |
1946                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1947                                 MSI_ADDR_REDIRECTION_CPU:
1948                                 MSI_ADDR_REDIRECTION_LOWPRI) |
1949                         MSI_ADDR_DEST_ID(dest);
1950
1951                 msg->data =
1952                         MSI_DATA_TRIGGER_EDGE |
1953                         MSI_DATA_LEVEL_ASSERT |
1954                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1955                                 MSI_DATA_DELIVERY_FIXED:
1956                                 MSI_DATA_DELIVERY_LOWPRI) |
1957                         MSI_DATA_VECTOR(cfg->vector);
1958         }
1959         return err;
1960 }
1961
1962 #ifdef CONFIG_SMP
1963 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1964 {
1965         struct irq_cfg *cfg = irq_cfg + irq;
1966         struct msi_msg msg;
1967         unsigned int dest;
1968         cpumask_t tmp;
1969
1970         cpus_and(tmp, mask, cpu_online_map);
1971         if (cpus_empty(tmp))
1972                 return;
1973
1974         if (assign_irq_vector(irq, mask))
1975                 return;
1976
1977         cpus_and(tmp, cfg->domain, mask);
1978         dest = cpu_mask_to_apicid(tmp);
1979
1980         read_msi_msg(irq, &msg);
1981
1982         msg.data &= ~MSI_DATA_VECTOR_MASK;
1983         msg.data |= MSI_DATA_VECTOR(cfg->vector);
1984         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1985         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1986
1987         write_msi_msg(irq, &msg);
1988         irq_desc[irq].affinity = mask;
1989 }
1990 #endif /* CONFIG_SMP */
1991
1992 /*
1993  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1994  * which implement the MSI or MSI-X Capability Structure.
1995  */
1996 static struct irq_chip msi_chip = {
1997         .name           = "PCI-MSI",
1998         .unmask         = unmask_msi_irq,
1999         .mask           = mask_msi_irq,
2000         .ack            = ack_apic_edge,
2001 #ifdef CONFIG_SMP
2002         .set_affinity   = set_msi_irq_affinity,
2003 #endif
2004         .retrigger      = ioapic_retrigger_irq,
2005 };
2006
2007 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2008 {
2009         struct msi_msg msg;
2010         int irq, ret;
2011         irq = create_irq();
2012         if (irq < 0)
2013                 return irq;
2014
2015         ret = msi_compose_msg(dev, irq, &msg);
2016         if (ret < 0) {
2017                 destroy_irq(irq);
2018                 return ret;
2019         }
2020
2021         set_irq_msi(irq, desc);
2022         write_msi_msg(irq, &msg);
2023
2024         set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2025
2026         return 0;
2027 }
2028
2029 void arch_teardown_msi_irq(unsigned int irq)
2030 {
2031         destroy_irq(irq);
2032 }
2033
2034 #endif /* CONFIG_PCI_MSI */
2035
2036 /*
2037  * Hypertransport interrupt support
2038  */
2039 #ifdef CONFIG_HT_IRQ
2040
2041 #ifdef CONFIG_SMP
2042
2043 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2044 {
2045         struct ht_irq_msg msg;
2046         fetch_ht_irq_msg(irq, &msg);
2047
2048         msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2049         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2050
2051         msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2052         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2053
2054         write_ht_irq_msg(irq, &msg);
2055 }
2056
2057 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2058 {
2059         struct irq_cfg *cfg = irq_cfg + irq;
2060         unsigned int dest;
2061         cpumask_t tmp;
2062
2063         cpus_and(tmp, mask, cpu_online_map);
2064         if (cpus_empty(tmp))
2065                 return;
2066
2067         if (assign_irq_vector(irq, mask))
2068                 return;
2069
2070         cpus_and(tmp, cfg->domain, mask);
2071         dest = cpu_mask_to_apicid(tmp);
2072
2073         target_ht_irq(irq, dest, cfg->vector);
2074         irq_desc[irq].affinity = mask;
2075 }
2076 #endif
2077
2078 static struct irq_chip ht_irq_chip = {
2079         .name           = "PCI-HT",
2080         .mask           = mask_ht_irq,
2081         .unmask         = unmask_ht_irq,
2082         .ack            = ack_apic_edge,
2083 #ifdef CONFIG_SMP
2084         .set_affinity   = set_ht_irq_affinity,
2085 #endif
2086         .retrigger      = ioapic_retrigger_irq,
2087 };
2088
2089 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2090 {
2091         struct irq_cfg *cfg = irq_cfg + irq;
2092         int err;
2093         cpumask_t tmp;
2094
2095         tmp = TARGET_CPUS;
2096         err = assign_irq_vector(irq, tmp);
2097         if (!err) {
2098                 struct ht_irq_msg msg;
2099                 unsigned dest;
2100
2101                 cpus_and(tmp, cfg->domain, tmp);
2102                 dest = cpu_mask_to_apicid(tmp);
2103
2104                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2105
2106                 msg.address_lo =
2107                         HT_IRQ_LOW_BASE |
2108                         HT_IRQ_LOW_DEST_ID(dest) |
2109                         HT_IRQ_LOW_VECTOR(cfg->vector) |
2110                         ((INT_DEST_MODE == 0) ?
2111                                 HT_IRQ_LOW_DM_PHYSICAL :
2112                                 HT_IRQ_LOW_DM_LOGICAL) |
2113                         HT_IRQ_LOW_RQEOI_EDGE |
2114                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2115                                 HT_IRQ_LOW_MT_FIXED :
2116                                 HT_IRQ_LOW_MT_ARBITRATED) |
2117                         HT_IRQ_LOW_IRQ_MASKED;
2118
2119                 write_ht_irq_msg(irq, &msg);
2120
2121                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2122                                               handle_edge_irq, "edge");
2123         }
2124         return err;
2125 }
2126 #endif /* CONFIG_HT_IRQ */
2127
2128 /* --------------------------------------------------------------------------
2129                           ACPI-based IOAPIC Configuration
2130    -------------------------------------------------------------------------- */
2131
2132 #ifdef CONFIG_ACPI
2133
2134 #define IO_APIC_MAX_ID          0xFE
2135
2136 int __init io_apic_get_redir_entries (int ioapic)
2137 {
2138         union IO_APIC_reg_01    reg_01;
2139         unsigned long flags;
2140
2141         spin_lock_irqsave(&ioapic_lock, flags);
2142         reg_01.raw = io_apic_read(ioapic, 1);
2143         spin_unlock_irqrestore(&ioapic_lock, flags);
2144
2145         return reg_01.bits.entries;
2146 }
2147
2148
2149 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2150 {
2151         if (!IO_APIC_IRQ(irq)) {
2152                 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2153                         ioapic);
2154                 return -EINVAL;
2155         }
2156
2157         /*
2158          * IRQs < 16 are already in the irq_2_pin[] map
2159          */
2160         if (irq >= 16)
2161                 add_pin_to_irq(irq, ioapic, pin);
2162
2163         setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2164
2165         return 0;
2166 }
2167
2168 #endif /* CONFIG_ACPI */
2169
2170
2171 /*
2172  * This function currently is only a helper for the i386 smp boot process where
2173  * we need to reprogram the ioredtbls to cater for the cpus which have come online
2174  * so mask in all cases should simply be TARGET_CPUS
2175  */
2176 #ifdef CONFIG_SMP
2177 void __init setup_ioapic_dest(void)
2178 {
2179         int pin, ioapic, irq, irq_entry;
2180
2181         if (skip_ioapic_setup == 1)
2182                 return;
2183
2184         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2185                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2186                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2187                         if (irq_entry == -1)
2188                                 continue;
2189                         irq = pin_2_irq(irq_entry, ioapic, pin);
2190
2191                         /* setup_IO_APIC_irqs could fail to get vector for some device
2192                          * when you have too many devices, because at that time only boot
2193                          * cpu is online.
2194                          */
2195                         if (!irq_cfg[irq].vector)
2196                                 setup_IO_APIC_irq(ioapic, pin, irq,
2197                                                   irq_trigger(irq_entry),
2198                                                   irq_polarity(irq_entry));
2199                         else
2200                                 set_ioapic_affinity_irq(irq, TARGET_CPUS);
2201                 }
2202
2203         }
2204 }
2205 #endif