arch/x86/kernel/io_apic_32.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/mc146818rtc.h>
  29 #include <linux/compiler.h>
  30 #include <linux/acpi.h>
  31 #include <linux/module.h>
  32 #include <linux/sysdev.h>
  33 #include <linux/pci.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
  40 #include <asm/io.h>
  41 #include <asm/smp.h>
  42 #include <asm/desc.h>
  43 #include <asm/timer.h>
  44 #include <asm/i8259.h>
  45 #include <asm/nmi.h>
  46 #include <asm/msidef.h>
  47 #include <asm/hypertransport.h>
  48
  49 #include <mach_apic.h>
  50 #include <mach_apicdef.h>
  51
  52 int (*ioapic_renumber_irq)(int ioapic, int irq);
  53 atomic_t irq_mis_count;
  54
  55 /* Where if anywhere is the i8259 connect in external int mode */
  56 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
  57
  58 static DEFINE_SPINLOCK(ioapic_lock);
  59 static DEFINE_SPINLOCK(vector_lock);
  60
  61 int timer_over_8254 __initdata = 1;
  62
  63 /*
  64  *      Is the SiS APIC rmw bug present ?
  65  *      -1 = don't know, 0 = no, 1 = yes
  66  */
  67 int sis_apic_bug = -1;
  68
  69 /*
  70  * # of IRQ routing registers
  71  */
  72 int nr_ioapic_registers[MAX_IO_APICS];
  73
  74 /* I/O APIC entries */
  75 struct mpc_config_ioapic mp_ioapics[MAX_IO_APICS];
  76 int nr_ioapics;
  77
  78 /* MP IRQ source entries */
  79 struct mpc_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
  80
  81 /* # of MP IRQ source entries */
  82 int mp_irq_entries;
  83
  84 static int disable_timer_pin_1 __initdata;
  85
  86 /*
  87  * Rough estimation of how many shared IRQs there are, can
  88  * be changed anytime.
  89  */
  90 #define MAX_PLUS_SHARED_IRQS NR_IRQS
  91 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
  92
  93 /*
  94  * This is performance-critical, we want to do it O(1)
  95  *
  96  * the indexing order of this array favors 1:1 mappings
  97  * between pins and IRQs.
  98  */
  99
 100 static struct irq_pin_list {
 101         int apic, pin, next;
 102 } irq_2_pin[PIN_MAP_SIZE];
 103
 104 struct io_apic {
 105         unsigned int index;
 106         unsigned int unused[3];
 107         unsigned int data;
 108 };
 109
 110 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
 111 {
 112         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
 113                 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
 114 }
 115
 116 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
 117 {
 118         struct io_apic __iomem *io_apic = io_apic_base(apic);
 119         writel(reg, &io_apic->index);
 120         return readl(&io_apic->data);
 121 }
 122
 123 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
 124 {
 125         struct io_apic __iomem *io_apic = io_apic_base(apic);
 126         writel(reg, &io_apic->index);
 127         writel(value, &io_apic->data);
 128 }
 129
 130 /*
 131  * Re-write a value: to be used for read-modify-write
 132  * cycles where the read already set up the index register.
 133  *
 134  * Older SiS APIC requires we rewrite the index register
 135  */
 136 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
 137 {
 138         volatile struct io_apic __iomem *io_apic = io_apic_base(apic);
 139         if (sis_apic_bug)
 140                 writel(reg, &io_apic->index);
 141         writel(value, &io_apic->data);
 142 }
 143
 144 union entry_union {
 145         struct { u32 w1, w2; };
 146         struct IO_APIC_route_entry entry;
 147 };
 148
 149 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
 150 {
 151         union entry_union eu;
 152         unsigned long flags;
 153         spin_lock_irqsave(&ioapic_lock, flags);
 154         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
 155         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
 156         spin_unlock_irqrestore(&ioapic_lock, flags);
 157         return eu.entry;
 158 }
 159
 160 /*
 161  * When we write a new IO APIC routing entry, we need to write the high
 162  * word first! If the mask bit in the low word is clear, we will enable
 163  * the interrupt, and we need to make sure the entry is fully populated
 164  * before that happens.
 165  */
 166 static void
 167 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 168 {
 169         union entry_union eu;
 170         eu.entry = e;
 171         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 172         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 173 }
 174
 175 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 176 {
 177         unsigned long flags;
 178         spin_lock_irqsave(&ioapic_lock, flags);
 179         __ioapic_write_entry(apic, pin, e);
 180         spin_unlock_irqrestore(&ioapic_lock, flags);
 181 }
 182
 183 /*
 184  * When we mask an IO APIC routing entry, we need to write the low
 185  * word first, in order to set the mask bit before we change the
 186  * high bits!
 187  */
 188 static void ioapic_mask_entry(int apic, int pin)
 189 {
 190         unsigned long flags;
 191         union entry_union eu = { .entry.mask = 1 };
 192
 193         spin_lock_irqsave(&ioapic_lock, flags);
 194         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 195         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 196         spin_unlock_irqrestore(&ioapic_lock, flags);
 197 }
 198
 199 /*
 200  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
 201  * shared ISA-space IRQs, so we have to support them. We are super
 202  * fast in the common case, and fast for shared ISA-space IRQs.
 203  */
 204 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 205 {
 206         static int first_free_entry = NR_IRQS;
 207         struct irq_pin_list *entry = irq_2_pin + irq;
 208
 209         while (entry->next)
 210                 entry = irq_2_pin + entry->next;
 211
 212         if (entry->pin != -1) {
 213                 entry->next = first_free_entry;
 214                 entry = irq_2_pin + entry->next;
 215                 if (++first_free_entry >= PIN_MAP_SIZE)
 216                         panic("io_apic.c: whoops");
 217         }
 218         entry->apic = apic;
 219         entry->pin = pin;
 220 }
 221
 222 /*
 223  * Reroute an IRQ to a different pin.
 224  */
 225 static void __init replace_pin_at_irq(unsigned int irq,
 226                                       int oldapic, int oldpin,
 227                                       int newapic, int newpin)
 228 {
 229         struct irq_pin_list *entry = irq_2_pin + irq;
 230
 231         while (1) {
 232                 if (entry->apic == oldapic && entry->pin == oldpin) {
 233                         entry->apic = newapic;
 234                         entry->pin = newpin;
 235                 }
 236                 if (!entry->next)
 237                         break;
 238                 entry = irq_2_pin + entry->next;
 239         }
 240 }
 241
 242 static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable)
 243 {
 244         struct irq_pin_list *entry = irq_2_pin + irq;
 245         unsigned int pin, reg;
 246
 247         for (;;) {
 248                 pin = entry->pin;
 249                 if (pin == -1)
 250                         break;
 251                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
 252                 reg &= ~disable;
 253                 reg |= enable;
 254                 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
 255                 if (!entry->next)
 256                         break;
 257                 entry = irq_2_pin + entry->next;
 258         }
 259 }
 260
 261 /* mask = 1 */
 262 static void __mask_IO_APIC_irq (unsigned int irq)
 263 {
 264         __modify_IO_APIC_irq(irq, 0x00010000, 0);
 265 }
 266
 267 /* mask = 0 */
 268 static void __unmask_IO_APIC_irq (unsigned int irq)
 269 {
 270         __modify_IO_APIC_irq(irq, 0, 0x00010000);
 271 }
 272
 273 /* mask = 1, trigger = 0 */
 274 static void __mask_and_edge_IO_APIC_irq (unsigned int irq)
 275 {
 276         __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000);
 277 }
 278
 279 /* mask = 0, trigger = 1 */
 280 static void __unmask_and_level_IO_APIC_irq (unsigned int irq)
 281 {
 282         __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000);
 283 }
 284
 285 static void mask_IO_APIC_irq (unsigned int irq)
 286 {
 287         unsigned long flags;
 288
 289         spin_lock_irqsave(&ioapic_lock, flags);
 290         __mask_IO_APIC_irq(irq);
 291         spin_unlock_irqrestore(&ioapic_lock, flags);
 292 }
 293
 294 static void unmask_IO_APIC_irq (unsigned int irq)
 295 {
 296         unsigned long flags;
 297
 298         spin_lock_irqsave(&ioapic_lock, flags);
 299         __unmask_IO_APIC_irq(irq);
 300         spin_unlock_irqrestore(&ioapic_lock, flags);
 301 }
 302
 303 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 304 {
 305         struct IO_APIC_route_entry entry;
 306
 307         /* Check delivery_mode to be sure we're not clearing an SMI pin */
 308         entry = ioapic_read_entry(apic, pin);
 309         if (entry.delivery_mode == dest_SMI)
 310                 return;
 311
 312         /*
 313          * Disable it in the IO-APIC irq-routing table:
 314          */
 315         ioapic_mask_entry(apic, pin);
 316 }
 317
 318 static void clear_IO_APIC (void)
 319 {
 320         int apic, pin;
 321
 322         for (apic = 0; apic < nr_ioapics; apic++)
 323                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 324                         clear_IO_APIC_pin(apic, pin);
 325 }
 326
 327 #ifdef CONFIG_SMP
 328 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
 329 {
 330         unsigned long flags;
 331         int pin;
 332         struct irq_pin_list *entry = irq_2_pin + irq;
 333         unsigned int apicid_value;
 334         cpumask_t tmp;
 335
 336         cpus_and(tmp, cpumask, cpu_online_map);
 337         if (cpus_empty(tmp))
 338                 tmp = TARGET_CPUS;
 339
 340         cpus_and(cpumask, tmp, CPU_MASK_ALL);
 341
 342         apicid_value = cpu_mask_to_apicid(cpumask);
 343         /* Prepare to do the io_apic_write */
 344         apicid_value = apicid_value << 24;
 345         spin_lock_irqsave(&ioapic_lock, flags);
 346         for (;;) {
 347                 pin = entry->pin;
 348                 if (pin == -1)
 349                         break;
 350                 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
 351                 if (!entry->next)
 352                         break;
 353                 entry = irq_2_pin + entry->next;
 354         }
 355         irq_desc[irq].affinity = cpumask;
 356         spin_unlock_irqrestore(&ioapic_lock, flags);
 357 }
 358
 359 #if defined(CONFIG_IRQBALANCE)
 360 # include <asm/processor.h>     /* kernel_thread() */
 361 # include <linux/kernel_stat.h> /* kstat */
 362 # include <linux/slab.h>                /* kmalloc() */
 363 # include <linux/timer.h>
 364
 365 #define IRQBALANCE_CHECK_ARCH -999
 366 #define MAX_BALANCED_IRQ_INTERVAL       (5*HZ)
 367 #define MIN_BALANCED_IRQ_INTERVAL       (HZ/2)
 368 #define BALANCED_IRQ_MORE_DELTA         (HZ/10)
 369 #define BALANCED_IRQ_LESS_DELTA         (HZ)
 370
 371 static int irqbalance_disabled __read_mostly = IRQBALANCE_CHECK_ARCH;
 372 static int physical_balance __read_mostly;
 373 static long balanced_irq_interval __read_mostly = MAX_BALANCED_IRQ_INTERVAL;
 374
 375 static struct irq_cpu_info {
 376         unsigned long * last_irq;
 377         unsigned long * irq_delta;
 378         unsigned long irq;
 379 } irq_cpu_data[NR_CPUS];
 380
 381 #define CPU_IRQ(cpu)            (irq_cpu_data[cpu].irq)
 382 #define LAST_CPU_IRQ(cpu,irq)   (irq_cpu_data[cpu].last_irq[irq])
 383 #define IRQ_DELTA(cpu,irq)      (irq_cpu_data[cpu].irq_delta[irq])
 384
 385 #define IDLE_ENOUGH(cpu,now) \
 386         (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
 387
 388 #define IRQ_ALLOWED(cpu, allowed_mask)  cpu_isset(cpu, allowed_mask)
 389
 390 #define CPU_TO_PACKAGEINDEX(i) (first_cpu(per_cpu(cpu_sibling_map, i)))
 391
 392 static cpumask_t balance_irq_affinity[NR_IRQS] = {
 393         [0 ... NR_IRQS-1] = CPU_MASK_ALL
 394 };
 395
 396 void set_balance_irq_affinity(unsigned int irq, cpumask_t mask)
 397 {
 398         balance_irq_affinity[irq] = mask;
 399 }
 400
 401 static unsigned long move(int curr_cpu, cpumask_t allowed_mask,
 402                         unsigned long now, int direction)
 403 {
 404         int search_idle = 1;
 405         int cpu = curr_cpu;
 406
 407         goto inside;
 408
 409         do {
 410                 if (unlikely(cpu == curr_cpu))
 411                         search_idle = 0;
 412 inside:
 413                 if (direction == 1) {
 414                         cpu++;
 415                         if (cpu >= NR_CPUS)
 416                                 cpu = 0;
 417                 } else {
 418                         cpu--;
 419                         if (cpu == -1)
 420                                 cpu = NR_CPUS-1;
 421                 }
 422         } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
 423                         (search_idle && !IDLE_ENOUGH(cpu,now)));
 424
 425         return cpu;
 426 }
 427
 428 static inline void balance_irq(int cpu, int irq)
 429 {
 430         unsigned long now = jiffies;
 431         cpumask_t allowed_mask;
 432         unsigned int new_cpu;
 433
 434         if (irqbalance_disabled)
 435                 return;
 436
 437         cpus_and(allowed_mask, cpu_online_map, balance_irq_affinity[irq]);
 438         new_cpu = move(cpu, allowed_mask, now, 1);
 439         if (cpu != new_cpu) {
 440                 set_pending_irq(irq, cpumask_of_cpu(new_cpu));
 441         }
 442 }
 443
 444 static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
 445 {
 446         int i, j;
 447
 448         for_each_online_cpu(i) {
 449                 for (j = 0; j < NR_IRQS; j++) {
 450                         if (!irq_desc[j].action)
 451                                 continue;
 452                         /* Is it a significant load ?  */
 453                         if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) <
 454                                                 useful_load_threshold)
 455                                 continue;
 456                         balance_irq(i, j);
 457                 }
 458         }
 459         balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 460                 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 461         return;
 462 }
 463
 464 static void do_irq_balance(void)
 465 {
 466         int i, j;
 467         unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
 468         unsigned long move_this_load = 0;
 469         int max_loaded = 0, min_loaded = 0;
 470         int load;
 471         unsigned long useful_load_threshold = balanced_irq_interval + 10;
 472         int selected_irq;
 473         int tmp_loaded, first_attempt = 1;
 474         unsigned long tmp_cpu_irq;
 475         unsigned long imbalance = 0;
 476         cpumask_t allowed_mask, target_cpu_mask, tmp;
 477
 478         for_each_possible_cpu(i) {
 479                 int package_index;
 480                 CPU_IRQ(i) = 0;
 481                 if (!cpu_online(i))
 482                         continue;
 483                 package_index = CPU_TO_PACKAGEINDEX(i);
 484                 for (j = 0; j < NR_IRQS; j++) {
 485                         unsigned long value_now, delta;
 486                         /* Is this an active IRQ or balancing disabled ? */
 487                         if (!irq_desc[j].action || irq_balancing_disabled(j))
 488                                 continue;
 489                         if ( package_index == i )
 490                                 IRQ_DELTA(package_index,j) = 0;
 491                         /* Determine the total count per processor per IRQ */
 492                         value_now = (unsigned long) kstat_cpu(i).irqs[j];
 493
 494                         /* Determine the activity per processor per IRQ */
 495                         delta = value_now - LAST_CPU_IRQ(i,j);
 496
 497                         /* Update last_cpu_irq[][] for the next time */
 498                         LAST_CPU_IRQ(i,j) = value_now;
 499
 500                         /* Ignore IRQs whose rate is less than the clock */
 501                         if (delta < useful_load_threshold)
 502                                 continue;
 503                         /* update the load for the processor or package total */
 504                         IRQ_DELTA(package_index,j) += delta;
 505
 506                         /* Keep track of the higher numbered sibling as well */
 507                         if (i != package_index)
 508                                 CPU_IRQ(i) += delta;
 509                         /*
 510                          * We have sibling A and sibling B in the package
 511                          *
 512                          * cpu_irq[A] = load for cpu A + load for cpu B
 513                          * cpu_irq[B] = load for cpu B
 514                          */
 515                         CPU_IRQ(package_index) += delta;
 516                 }
 517         }
 518         /* Find the least loaded processor package */
 519         for_each_online_cpu(i) {
 520                 if (i != CPU_TO_PACKAGEINDEX(i))
 521                         continue;
 522                 if (min_cpu_irq > CPU_IRQ(i)) {
 523                         min_cpu_irq = CPU_IRQ(i);
 524                         min_loaded = i;
 525                 }
 526         }
 527         max_cpu_irq = ULONG_MAX;
 528
 529 tryanothercpu:
 530         /* Look for heaviest loaded processor.
 531          * We may come back to get the next heaviest loaded processor.
 532          * Skip processors with trivial loads.
 533          */
 534         tmp_cpu_irq = 0;
 535         tmp_loaded = -1;
 536         for_each_online_cpu(i) {
 537                 if (i != CPU_TO_PACKAGEINDEX(i))
 538                         continue;
 539                 if (max_cpu_irq <= CPU_IRQ(i))
 540                         continue;
 541                 if (tmp_cpu_irq < CPU_IRQ(i)) {
 542                         tmp_cpu_irq = CPU_IRQ(i);
 543                         tmp_loaded = i;
 544                 }
 545         }
 546
 547         if (tmp_loaded == -1) {
 548          /* In the case of small number of heavy interrupt sources,
 549           * loading some of the cpus too much. We use Ingo's original
 550           * approach to rotate them around.
 551           */
 552                 if (!first_attempt && imbalance >= useful_load_threshold) {
 553                         rotate_irqs_among_cpus(useful_load_threshold);
 554                         return;
 555                 }
 556                 goto not_worth_the_effort;
 557         }
 558
 559         first_attempt = 0;              /* heaviest search */
 560         max_cpu_irq = tmp_cpu_irq;      /* load */
 561         max_loaded = tmp_loaded;        /* processor */
 562         imbalance = (max_cpu_irq - min_cpu_irq) / 2;
 563
 564         /* if imbalance is less than approx 10% of max load, then
 565          * observe diminishing returns action. - quit
 566          */
 567         if (imbalance < (max_cpu_irq >> 3))
 568                 goto not_worth_the_effort;
 569
 570 tryanotherirq:
 571         /* if we select an IRQ to move that can't go where we want, then
 572          * see if there is another one to try.
 573          */
 574         move_this_load = 0;
 575         selected_irq = -1;
 576         for (j = 0; j < NR_IRQS; j++) {
 577                 /* Is this an active IRQ? */
 578                 if (!irq_desc[j].action)
 579                         continue;
 580                 if (imbalance <= IRQ_DELTA(max_loaded,j))
 581                         continue;
 582                 /* Try to find the IRQ that is closest to the imbalance
 583                  * without going over.
 584                  */
 585                 if (move_this_load < IRQ_DELTA(max_loaded,j)) {
 586                         move_this_load = IRQ_DELTA(max_loaded,j);
 587                         selected_irq = j;
 588                 }
 589         }
 590         if (selected_irq == -1) {
 591                 goto tryanothercpu;
 592         }
 593
 594         imbalance = move_this_load;
 595
 596         /* For physical_balance case, we accumulated both load
 597          * values in the one of the siblings cpu_irq[],
 598          * to use the same code for physical and logical processors
 599          * as much as possible.
 600          *
 601          * NOTE: the cpu_irq[] array holds the sum of the load for
 602          * sibling A and sibling B in the slot for the lowest numbered
 603          * sibling (A), _AND_ the load for sibling B in the slot for
 604          * the higher numbered sibling.
 605          *
 606          * We seek the least loaded sibling by making the comparison
 607          * (A+B)/2 vs B
 608          */
 609         load = CPU_IRQ(min_loaded) >> 1;
 610         for_each_cpu_mask(j, per_cpu(cpu_sibling_map, min_loaded)) {
 611                 if (load > CPU_IRQ(j)) {
 612                         /* This won't change cpu_sibling_map[min_loaded] */
 613                         load = CPU_IRQ(j);
 614                         min_loaded = j;
 615                 }
 616         }
 617
 618         cpus_and(allowed_mask,
 619                 cpu_online_map,
 620                 balance_irq_affinity[selected_irq]);
 621         target_cpu_mask = cpumask_of_cpu(min_loaded);
 622         cpus_and(tmp, target_cpu_mask, allowed_mask);
 623
 624         if (!cpus_empty(tmp)) {
 625                 /* mark for change destination */
 626                 set_pending_irq(selected_irq, cpumask_of_cpu(min_loaded));
 627
 628                 /* Since we made a change, come back sooner to
 629                  * check for more variation.
 630                  */
 631                 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 632                         balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 633                 return;
 634         }
 635         goto tryanotherirq;
 636
 637 not_worth_the_effort:
 638         /*
 639          * if we did not find an IRQ to move, then adjust the time interval
 640          * upward
 641          */
 642         balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
 643                 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
 644         return;
 645 }
 646
 647 static int balanced_irq(void *unused)
 648 {
 649         int i;
 650         unsigned long prev_balance_time = jiffies;
 651         long time_remaining = balanced_irq_interval;
 652
 653         /* push everything to CPU 0 to give us a starting point.  */
 654         for (i = 0 ; i < NR_IRQS ; i++) {
 655                 irq_desc[i].pending_mask = cpumask_of_cpu(0);
 656                 set_pending_irq(i, cpumask_of_cpu(0));
 657         }
 658
 659         set_freezable();
 660         for ( ; ; ) {
 661                 time_remaining = schedule_timeout_interruptible(time_remaining);
 662                 try_to_freeze();
 663                 if (time_after(jiffies,
 664                                 prev_balance_time+balanced_irq_interval)) {
 665                         preempt_disable();
 666                         do_irq_balance();
 667                         prev_balance_time = jiffies;
 668                         time_remaining = balanced_irq_interval;
 669                         preempt_enable();
 670                 }
 671         }
 672         return 0;
 673 }
 674
 675 static int __init balanced_irq_init(void)
 676 {
 677         int i;
 678         struct cpuinfo_x86 *c;
 679         cpumask_t tmp;
 680
 681         cpus_shift_right(tmp, cpu_online_map, 2);
 682         c = &boot_cpu_data;
 683         /* When not overwritten by the command line ask subarchitecture. */
 684         if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
 685                 irqbalance_disabled = NO_BALANCE_IRQ;
 686         if (irqbalance_disabled)
 687                 return 0;
 688
 689          /* disable irqbalance completely if there is only one processor online */
 690         if (num_online_cpus() < 2) {
 691                 irqbalance_disabled = 1;
 692                 return 0;
 693         }
 694         /*
 695          * Enable physical balance only if more than 1 physical processor
 696          * is present
 697          */
 698         if (smp_num_siblings > 1 && !cpus_empty(tmp))
 699                 physical_balance = 1;
 700
 701         for_each_online_cpu(i) {
 702                 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 703                 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 704                 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
 705                         printk(KERN_ERR "balanced_irq_init: out of memory");
 706                         goto failed;
 707                 }
 708                 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
 709                 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
 710         }
 711
 712         printk(KERN_INFO "Starting balanced_irq\n");
 713         if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd")))
 714                 return 0;
 715         printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
 716 failed:
 717         for_each_possible_cpu(i) {
 718                 kfree(irq_cpu_data[i].irq_delta);
 719                 irq_cpu_data[i].irq_delta = NULL;
 720                 kfree(irq_cpu_data[i].last_irq);
 721                 irq_cpu_data[i].last_irq = NULL;
 722         }
 723         return 0;
 724 }
 725
 726 int __devinit irqbalance_disable(char *str)
 727 {
 728         irqbalance_disabled = 1;
 729         return 1;
 730 }
 731
 732 __setup("noirqbalance", irqbalance_disable);
 733
 734 late_initcall(balanced_irq_init);
 735 #endif /* CONFIG_IRQBALANCE */
 736 #endif /* CONFIG_SMP */
 737
 738 #ifndef CONFIG_SMP
 739 void send_IPI_self(int vector)
 740 {
 741         unsigned int cfg;
 742
 743         /*
 744          * Wait for idle.
 745          */
 746         apic_wait_icr_idle();
 747         cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
 748         /*
 749          * Send the IPI. The write to APIC_ICR fires this off.
 750          */
 751         apic_write_around(APIC_ICR, cfg);
 752 }
 753 #endif /* !CONFIG_SMP */
 754
 755
 756 /*
 757  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
 758  * specific CPU-side IRQs.
 759  */
 760
 761 #define MAX_PIRQS 8
 762 static int pirq_entries [MAX_PIRQS];
 763 static int pirqs_enabled;
 764 int skip_ioapic_setup;
 765
 766 static int __init ioapic_pirq_setup(char *str)
 767 {
 768         int i, max;
 769         int ints[MAX_PIRQS+1];
 770
 771         get_options(str, ARRAY_SIZE(ints), ints);
 772
 773         for (i = 0; i < MAX_PIRQS; i++)
 774                 pirq_entries[i] = -1;
 775
 776         pirqs_enabled = 1;
 777         apic_printk(APIC_VERBOSE, KERN_INFO
 778                         "PIRQ redirection, working around broken MP-BIOS.\n");
 779         max = MAX_PIRQS;
 780         if (ints[0] < MAX_PIRQS)
 781                 max = ints[0];
 782
 783         for (i = 0; i < max; i++) {
 784                 apic_printk(APIC_VERBOSE, KERN_DEBUG
 785                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
 786                 /*
 787                  * PIRQs are mapped upside down, usually.
 788                  */
 789                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
 790         }
 791         return 1;
 792 }
 793
 794 __setup("pirq=", ioapic_pirq_setup);
 795
 796 /*
 797  * Find the IRQ entry number of a certain pin.
 798  */
 799 static int find_irq_entry(int apic, int pin, int type)
 800 {
 801         int i;
 802
 803         for (i = 0; i < mp_irq_entries; i++)
 804                 if (mp_irqs[i].mpc_irqtype == type &&
 805                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
 806                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
 807                     mp_irqs[i].mpc_dstirq == pin)
 808                         return i;
 809
 810         return -1;
 811 }
 812
 813 /*
 814  * Find the pin to which IRQ[irq] (ISA) is connected
 815  */
 816 static int __init find_isa_irq_pin(int irq, int type)
 817 {
 818         int i;
 819
 820         for (i = 0; i < mp_irq_entries; i++) {
 821                 int lbus = mp_irqs[i].mpc_srcbus;
 822
 823                 if (test_bit(lbus, mp_bus_not_pci) &&
 824                     (mp_irqs[i].mpc_irqtype == type) &&
 825                     (mp_irqs[i].mpc_srcbusirq == irq))
 826
 827                         return mp_irqs[i].mpc_dstirq;
 828         }
 829         return -1;
 830 }
 831
 832 static int __init find_isa_irq_apic(int irq, int type)
 833 {
 834         int i;
 835
 836         for (i = 0; i < mp_irq_entries; i++) {
 837                 int lbus = mp_irqs[i].mpc_srcbus;
 838
 839                 if (test_bit(lbus, mp_bus_not_pci) &&
 840                     (mp_irqs[i].mpc_irqtype == type) &&
 841                     (mp_irqs[i].mpc_srcbusirq == irq))
 842                         break;
 843         }
 844         if (i < mp_irq_entries) {
 845                 int apic;
 846                 for(apic = 0; apic < nr_ioapics; apic++) {
 847                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
 848                                 return apic;
 849                 }
 850         }
 851
 852         return -1;
 853 }
 854
 855 /*
 856  * Find a specific PCI IRQ entry.
 857  * Not an __init, possibly needed by modules
 858  */
 859 static int pin_2_irq(int idx, int apic, int pin);
 860
 861 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
 862 {
 863         int apic, i, best_guess = -1;
 864
 865         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
 866                 "slot:%d, pin:%d.\n", bus, slot, pin);
 867         if (mp_bus_id_to_pci_bus[bus] == -1) {
 868                 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
 869                 return -1;
 870         }
 871         for (i = 0; i < mp_irq_entries; i++) {
 872                 int lbus = mp_irqs[i].mpc_srcbus;
 873
 874                 for (apic = 0; apic < nr_ioapics; apic++)
 875                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
 876                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
 877                                 break;
 878
 879                 if (!test_bit(lbus, mp_bus_not_pci) &&
 880                     !mp_irqs[i].mpc_irqtype &&
 881                     (bus == lbus) &&
 882                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
 883                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
 884
 885                         if (!(apic || IO_APIC_IRQ(irq)))
 886                                 continue;
 887
 888                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
 889                                 return irq;
 890                         /*
 891                          * Use the first all-but-pin matching entry as a
 892                          * best-guess fuzzy result for broken mptables.
 893                          */
 894                         if (best_guess < 0)
 895                                 best_guess = irq;
 896                 }
 897         }
 898         return best_guess;
 899 }
 900 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
 901
 902 /*
 903  * This function currently is only a helper for the i386 smp boot process where
 904  * we need to reprogram the ioredtbls to cater for the cpus which have come online
 905  * so mask in all cases should simply be TARGET_CPUS
 906  */
 907 #ifdef CONFIG_SMP
 908 void __init setup_ioapic_dest(void)
 909 {
 910         int pin, ioapic, irq, irq_entry;
 911
 912         if (skip_ioapic_setup == 1)
 913                 return;
 914
 915         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
 916                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
 917                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
 918                         if (irq_entry == -1)
 919                                 continue;
 920                         irq = pin_2_irq(irq_entry, ioapic, pin);
 921                         set_ioapic_affinity_irq(irq, TARGET_CPUS);
 922                 }
 923
 924         }
 925 }
 926 #endif
 927
 928 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
 929 /*
 930  * EISA Edge/Level control register, ELCR
 931  */
 932 static int EISA_ELCR(unsigned int irq)
 933 {
 934         if (irq < 16) {
 935                 unsigned int port = 0x4d0 + (irq >> 3);
 936                 return (inb(port) >> (irq & 7)) & 1;
 937         }
 938         apic_printk(APIC_VERBOSE, KERN_INFO
 939                         "Broken MPtable reports ISA irq %d\n", irq);
 940         return 0;
 941 }
 942 #endif
 943
 944 /* ISA interrupts are always polarity zero edge triggered,
 945  * when listed as conforming in the MP table. */
 946
 947 #define default_ISA_trigger(idx)        (0)
 948 #define default_ISA_polarity(idx)       (0)
 949
 950 /* EISA interrupts are always polarity zero and can be edge or level
 951  * trigger depending on the ELCR value.  If an interrupt is listed as
 952  * EISA conforming in the MP table, that means its trigger type must
 953  * be read in from the ELCR */
 954
 955 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
 956 #define default_EISA_polarity(idx)      default_ISA_polarity(idx)
 957
 958 /* PCI interrupts are always polarity one level triggered,
 959  * when listed as conforming in the MP table. */
 960
 961 #define default_PCI_trigger(idx)        (1)
 962 #define default_PCI_polarity(idx)       (1)
 963
 964 /* MCA interrupts are always polarity zero level triggered,
 965  * when listed as conforming in the MP table. */
 966
 967 #define default_MCA_trigger(idx)        (1)
 968 #define default_MCA_polarity(idx)       default_ISA_polarity(idx)
 969
 970 static int MPBIOS_polarity(int idx)
 971 {
 972         int bus = mp_irqs[idx].mpc_srcbus;
 973         int polarity;
 974
 975         /*
 976          * Determine IRQ line polarity (high active or low active):
 977          */
 978         switch (mp_irqs[idx].mpc_irqflag & 3)
 979         {
 980                 case 0: /* conforms, ie. bus-type dependent polarity */
 981                 {
 982                         polarity = test_bit(bus, mp_bus_not_pci)?
 983                                 default_ISA_polarity(idx):
 984                                 default_PCI_polarity(idx);
 985                         break;
 986                 }
 987                 case 1: /* high active */
 988                 {
 989                         polarity = 0;
 990                         break;
 991                 }
 992                 case 2: /* reserved */
 993                 {
 994                         printk(KERN_WARNING "broken BIOS!!\n");
 995                         polarity = 1;
 996                         break;
 997                 }
 998                 case 3: /* low active */
 999                 {
1000                         polarity = 1;
1001                         break;
1002                 }
1003                 default: /* invalid */
1004                 {
1005                         printk(KERN_WARNING "broken BIOS!!\n");
1006                         polarity = 1;
1007                         break;
1008                 }
1009         }
1010         return polarity;
1011 }
1012
1013 static int MPBIOS_trigger(int idx)
1014 {
1015         int bus = mp_irqs[idx].mpc_srcbus;
1016         int trigger;
1017
1018         /*
1019          * Determine IRQ trigger mode (edge or level sensitive):
1020          */
1021         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
1022         {
1023                 case 0: /* conforms, ie. bus-type dependent */
1024                 {
1025                         trigger = test_bit(bus, mp_bus_not_pci)?
1026                                         default_ISA_trigger(idx):
1027                                         default_PCI_trigger(idx);
1028 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1029                         switch (mp_bus_id_to_type[bus])
1030                         {
1031                                 case MP_BUS_ISA: /* ISA pin */
1032                                 {
1033                                         /* set before the switch */
1034                                         break;
1035                                 }
1036                                 case MP_BUS_EISA: /* EISA pin */
1037                                 {
1038                                         trigger = default_EISA_trigger(idx);
1039                                         break;
1040                                 }
1041                                 case MP_BUS_PCI: /* PCI pin */
1042                                 {
1043                                         /* set before the switch */
1044                                         break;
1045                                 }
1046                                 case MP_BUS_MCA: /* MCA pin */
1047                                 {
1048                                         trigger = default_MCA_trigger(idx);
1049                                         break;
1050                                 }
1051                                 default:
1052                                 {
1053                                         printk(KERN_WARNING "broken BIOS!!\n");
1054                                         trigger = 1;
1055                                         break;
1056                                 }
1057                         }
1058 #endif
1059                         break;
1060                 }
1061                 case 1: /* edge */
1062                 {
1063                         trigger = 0;
1064                         break;
1065                 }
1066                 case 2: /* reserved */
1067                 {
1068                         printk(KERN_WARNING "broken BIOS!!\n");
1069                         trigger = 1;
1070                         break;
1071                 }
1072                 case 3: /* level */
1073                 {
1074                         trigger = 1;
1075                         break;
1076                 }
1077                 default: /* invalid */
1078                 {
1079                         printk(KERN_WARNING "broken BIOS!!\n");
1080                         trigger = 0;
1081                         break;
1082                 }
1083         }
1084         return trigger;
1085 }
1086
1087 static inline int irq_polarity(int idx)
1088 {
1089         return MPBIOS_polarity(idx);
1090 }
1091
1092 static inline int irq_trigger(int idx)
1093 {
1094         return MPBIOS_trigger(idx);
1095 }
1096
1097 static int pin_2_irq(int idx, int apic, int pin)
1098 {
1099         int irq, i;
1100         int bus = mp_irqs[idx].mpc_srcbus;
1101
1102         /*
1103          * Debugging check, we are in big trouble if this message pops up!
1104          */
1105         if (mp_irqs[idx].mpc_dstirq != pin)
1106                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1107
1108         if (test_bit(bus, mp_bus_not_pci))
1109                 irq = mp_irqs[idx].mpc_srcbusirq;
1110         else {
1111                 /*
1112                  * PCI IRQs are mapped in order
1113                  */
1114                 i = irq = 0;
1115                 while (i < apic)
1116                         irq += nr_ioapic_registers[i++];
1117                 irq += pin;
1118
1119                 /*
1120                  * For MPS mode, so far only needed by ES7000 platform
1121                  */
1122                 if (ioapic_renumber_irq)
1123                         irq = ioapic_renumber_irq(apic, irq);
1124         }
1125
1126         /*
1127          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1128          */
1129         if ((pin >= 16) && (pin <= 23)) {
1130                 if (pirq_entries[pin-16] != -1) {
1131                         if (!pirq_entries[pin-16]) {
1132                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1133                                                 "disabling PIRQ%d\n", pin-16);
1134                         } else {
1135                                 irq = pirq_entries[pin-16];
1136                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1137                                                 "using PIRQ%d -> IRQ %d\n",
1138                                                 pin-16, irq);
1139                         }
1140                 }
1141         }
1142         return irq;
1143 }
1144
1145 static inline int IO_APIC_irq_trigger(int irq)
1146 {
1147         int apic, idx, pin;
1148
1149         for (apic = 0; apic < nr_ioapics; apic++) {
1150                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1151                         idx = find_irq_entry(apic,pin,mp_INT);
1152                         if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1153                                 return irq_trigger(idx);
1154                 }
1155         }
1156         /*
1157          * nonexistent IRQs are edge default
1158          */
1159         return 0;
1160 }
1161
1162 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1163 static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 };
1164
1165 static int __assign_irq_vector(int irq)
1166 {
1167         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1168         int vector, offset;
1169
1170         BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
1171
1172         if (irq_vector[irq] > 0)
1173                 return irq_vector[irq];
1174
1175         vector = current_vector;
1176         offset = current_offset;
1177 next:
1178         vector += 8;
1179         if (vector >= FIRST_SYSTEM_VECTOR) {
1180                 offset = (offset + 1) % 8;
1181                 vector = FIRST_DEVICE_VECTOR + offset;
1182         }
1183         if (vector == current_vector)
1184                 return -ENOSPC;
1185         if (test_and_set_bit(vector, used_vectors))
1186                 goto next;
1187
1188         current_vector = vector;
1189         current_offset = offset;
1190         irq_vector[irq] = vector;
1191
1192         return vector;
1193 }
1194
1195 static int assign_irq_vector(int irq)
1196 {
1197         unsigned long flags;
1198         int vector;
1199
1200         spin_lock_irqsave(&vector_lock, flags);
1201         vector = __assign_irq_vector(irq);
1202         spin_unlock_irqrestore(&vector_lock, flags);
1203
1204         return vector;
1205 }
1206 static struct irq_chip ioapic_chip;
1207
1208 #define IOAPIC_AUTO     -1
1209 #define IOAPIC_EDGE     0
1210 #define IOAPIC_LEVEL    1
1211
1212 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1213 {
1214         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1215             trigger == IOAPIC_LEVEL) {
1216                 irq_desc[irq].status |= IRQ_LEVEL;
1217                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1218                                          handle_fasteoi_irq, "fasteoi");
1219         } else {
1220                 irq_desc[irq].status &= ~IRQ_LEVEL;
1221                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1222                                          handle_edge_irq, "edge");
1223         }
1224         set_intr_gate(vector, interrupt[irq]);
1225 }
1226
1227 static void __init setup_IO_APIC_irqs(void)
1228 {
1229         struct IO_APIC_route_entry entry;
1230         int apic, pin, idx, irq, first_notcon = 1, vector;
1231         unsigned long flags;
1232
1233         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1234
1235         for (apic = 0; apic < nr_ioapics; apic++) {
1236         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1237
1238                 /*
1239                  * add it to the IO-APIC irq-routing table:
1240                  */
1241                 memset(&entry,0,sizeof(entry));
1242
1243                 entry.delivery_mode = INT_DELIVERY_MODE;
1244                 entry.dest_mode = INT_DEST_MODE;
1245                 entry.mask = 0;                         /* enable IRQ */
1246                 entry.dest.logical.logical_dest =
1247                                         cpu_mask_to_apicid(TARGET_CPUS);
1248
1249                 idx = find_irq_entry(apic,pin,mp_INT);
1250                 if (idx == -1) {
1251                         if (first_notcon) {
1252                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1253                                                 " IO-APIC (apicid-pin) %d-%d",
1254                                                 mp_ioapics[apic].mpc_apicid,
1255                                                 pin);
1256                                 first_notcon = 0;
1257                         } else
1258                                 apic_printk(APIC_VERBOSE, ", %d-%d",
1259                                         mp_ioapics[apic].mpc_apicid, pin);
1260                         continue;
1261                 }
1262
1263                 if (!first_notcon) {
1264                         apic_printk(APIC_VERBOSE, " not connected.\n");
1265                         first_notcon = 1;
1266                 }
1267
1268                 entry.trigger = irq_trigger(idx);
1269                 entry.polarity = irq_polarity(idx);
1270
1271                 if (irq_trigger(idx)) {
1272                         entry.trigger = 1;
1273                         entry.mask = 1;
1274                 }
1275
1276                 irq = pin_2_irq(idx, apic, pin);
1277                 /*
1278                  * skip adding the timer int on secondary nodes, which causes
1279                  * a small but painful rift in the time-space continuum
1280                  */
1281                 if (multi_timer_check(apic, irq))
1282                         continue;
1283                 else
1284                         add_pin_to_irq(irq, apic, pin);
1285
1286                 if (!apic && !IO_APIC_IRQ(irq))
1287                         continue;
1288
1289                 if (IO_APIC_IRQ(irq)) {
1290                         vector = assign_irq_vector(irq);
1291                         entry.vector = vector;
1292                         ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1293
1294                         if (!apic && (irq < 16))
1295                                 disable_8259A_irq(irq);
1296                 }
1297                 spin_lock_irqsave(&ioapic_lock, flags);
1298                 __ioapic_write_entry(apic, pin, entry);
1299                 spin_unlock_irqrestore(&ioapic_lock, flags);
1300         }
1301         }
1302
1303         if (!first_notcon)
1304                 apic_printk(APIC_VERBOSE, " not connected.\n");
1305 }
1306
1307 /*
1308  * Set up the 8259A-master output pin:
1309  */
1310 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
1311 {
1312         struct IO_APIC_route_entry entry;
1313
1314         memset(&entry,0,sizeof(entry));
1315
1316         disable_8259A_irq(0);
1317
1318         /* mask LVT0 */
1319         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1320
1321         /*
1322          * We use logical delivery to get the timer IRQ
1323          * to the first CPU.
1324          */
1325         entry.dest_mode = INT_DEST_MODE;
1326         entry.mask = 0;                                 /* unmask IRQ now */
1327         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1328         entry.delivery_mode = INT_DELIVERY_MODE;
1329         entry.polarity = 0;
1330         entry.trigger = 0;
1331         entry.vector = vector;
1332
1333         /*
1334          * The timer IRQ doesn't have to know that behind the
1335          * scene we have a 8259A-master in AEOI mode ...
1336          */
1337         irq_desc[0].chip = &ioapic_chip;
1338         set_irq_handler(0, handle_edge_irq);
1339
1340         /*
1341          * Add it to the IO-APIC irq-routing table:
1342          */
1343         ioapic_write_entry(apic, pin, entry);
1344
1345         enable_8259A_irq(0);
1346 }
1347
1348 void __init print_IO_APIC(void)
1349 {
1350         int apic, i;
1351         union IO_APIC_reg_00 reg_00;
1352         union IO_APIC_reg_01 reg_01;
1353         union IO_APIC_reg_02 reg_02;
1354         union IO_APIC_reg_03 reg_03;
1355         unsigned long flags;
1356
1357         if (apic_verbosity == APIC_QUIET)
1358                 return;
1359
1360         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1361         for (i = 0; i < nr_ioapics; i++)
1362                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1363                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1364
1365         /*
1366          * We are a bit conservative about what we expect.  We have to
1367          * know about every hardware change ASAP.
1368          */
1369         printk(KERN_INFO "testing the IO APIC.......................\n");
1370
1371         for (apic = 0; apic < nr_ioapics; apic++) {
1372
1373         spin_lock_irqsave(&ioapic_lock, flags);
1374         reg_00.raw = io_apic_read(apic, 0);
1375         reg_01.raw = io_apic_read(apic, 1);
1376         if (reg_01.bits.version >= 0x10)
1377                 reg_02.raw = io_apic_read(apic, 2);
1378         if (reg_01.bits.version >= 0x20)
1379                 reg_03.raw = io_apic_read(apic, 3);
1380         spin_unlock_irqrestore(&ioapic_lock, flags);
1381
1382         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1383         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1384         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1385         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1386         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1387
1388         printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1389         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1390
1391         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1392         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1393
1394         /*
1395          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1396          * but the value of reg_02 is read as the previous read register
1397          * value, so ignore it if reg_02 == reg_01.
1398          */
1399         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1400                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1401                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1402         }
1403
1404         /*
1405          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1406          * or reg_03, but the value of reg_0[23] is read as the previous read
1407          * register value, so ignore it if reg_03 == reg_0[12].
1408          */
1409         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1410             reg_03.raw != reg_01.raw) {
1411                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1412                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1413         }
1414
1415         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1416
1417         printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1418                           " Stat Dest Deli Vect:   \n");
1419
1420         for (i = 0; i <= reg_01.bits.entries; i++) {
1421                 struct IO_APIC_route_entry entry;
1422
1423                 entry = ioapic_read_entry(apic, i);
1424
1425                 printk(KERN_DEBUG " %02x %03X %02X  ",
1426                         i,
1427                         entry.dest.logical.logical_dest,
1428                         entry.dest.physical.physical_dest
1429                 );
1430
1431                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1432                         entry.mask,
1433                         entry.trigger,
1434                         entry.irr,
1435                         entry.polarity,
1436                         entry.delivery_status,
1437                         entry.dest_mode,
1438                         entry.delivery_mode,
1439                         entry.vector
1440                 );
1441         }
1442         }
1443         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1444         for (i = 0; i < NR_IRQS; i++) {
1445                 struct irq_pin_list *entry = irq_2_pin + i;
1446                 if (entry->pin < 0)
1447                         continue;
1448                 printk(KERN_DEBUG "IRQ%d ", i);
1449                 for (;;) {
1450                         printk("-> %d:%d", entry->apic, entry->pin);
1451                         if (!entry->next)
1452                                 break;
1453                         entry = irq_2_pin + entry->next;
1454                 }
1455                 printk("\n");
1456         }
1457
1458         printk(KERN_INFO ".................................... done.\n");
1459
1460         return;
1461 }
1462
1463 #if 0
1464
1465 static void print_APIC_bitfield (int base)
1466 {
1467         unsigned int v;
1468         int i, j;
1469
1470         if (apic_verbosity == APIC_QUIET)
1471                 return;
1472
1473         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1474         for (i = 0; i < 8; i++) {
1475                 v = apic_read(base + i*0x10);
1476                 for (j = 0; j < 32; j++) {
1477                         if (v & (1<<j))
1478                                 printk("1");
1479                         else
1480                                 printk("0");
1481                 }
1482                 printk("\n");
1483         }
1484 }
1485
1486 void /*__init*/ print_local_APIC(void * dummy)
1487 {
1488         unsigned int v, ver, maxlvt;
1489
1490         if (apic_verbosity == APIC_QUIET)
1491                 return;
1492
1493         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1494                 smp_processor_id(), hard_smp_processor_id());
1495         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v,
1496                         GET_APIC_ID(read_apic_id()));
1497         v = apic_read(APIC_LVR);
1498         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1499         ver = GET_APIC_VERSION(v);
1500         maxlvt = lapic_get_maxlvt();
1501
1502         v = apic_read(APIC_TASKPRI);
1503         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1504
1505         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1506                 v = apic_read(APIC_ARBPRI);
1507                 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1508                         v & APIC_ARBPRI_MASK);
1509                 v = apic_read(APIC_PROCPRI);
1510                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1511         }
1512
1513         v = apic_read(APIC_EOI);
1514         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1515         v = apic_read(APIC_RRR);
1516         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1517         v = apic_read(APIC_LDR);
1518         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1519         v = apic_read(APIC_DFR);
1520         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1521         v = apic_read(APIC_SPIV);
1522         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1523
1524         printk(KERN_DEBUG "... APIC ISR field:\n");
1525         print_APIC_bitfield(APIC_ISR);
1526         printk(KERN_DEBUG "... APIC TMR field:\n");
1527         print_APIC_bitfield(APIC_TMR);
1528         printk(KERN_DEBUG "... APIC IRR field:\n");
1529         print_APIC_bitfield(APIC_IRR);
1530
1531         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1532                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1533                         apic_write(APIC_ESR, 0);
1534                 v = apic_read(APIC_ESR);
1535                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1536         }
1537
1538         v = apic_read(APIC_ICR);
1539         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1540         v = apic_read(APIC_ICR2);
1541         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1542
1543         v = apic_read(APIC_LVTT);
1544         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1545
1546         if (maxlvt > 3) {                       /* PC is LVT#4. */
1547                 v = apic_read(APIC_LVTPC);
1548                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1549         }
1550         v = apic_read(APIC_LVT0);
1551         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1552         v = apic_read(APIC_LVT1);
1553         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1554
1555         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1556                 v = apic_read(APIC_LVTERR);
1557                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1558         }
1559
1560         v = apic_read(APIC_TMICT);
1561         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1562         v = apic_read(APIC_TMCCT);
1563         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1564         v = apic_read(APIC_TDCR);
1565         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1566         printk("\n");
1567 }
1568
1569 void print_all_local_APICs (void)
1570 {
1571         on_each_cpu(print_local_APIC, NULL, 1, 1);
1572 }
1573
1574 void /*__init*/ print_PIC(void)
1575 {
1576         unsigned int v;
1577         unsigned long flags;
1578
1579         if (apic_verbosity == APIC_QUIET)
1580                 return;
1581
1582         printk(KERN_DEBUG "\nprinting PIC contents\n");
1583
1584         spin_lock_irqsave(&i8259A_lock, flags);
1585
1586         v = inb(0xa1) << 8 | inb(0x21);
1587         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1588
1589         v = inb(0xa0) << 8 | inb(0x20);
1590         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1591
1592         outb(0x0b,0xa0);
1593         outb(0x0b,0x20);
1594         v = inb(0xa0) << 8 | inb(0x20);
1595         outb(0x0a,0xa0);
1596         outb(0x0a,0x20);
1597
1598         spin_unlock_irqrestore(&i8259A_lock, flags);
1599
1600         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1601
1602         v = inb(0x4d1) << 8 | inb(0x4d0);
1603         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1604 }
1605
1606 #endif  /*  0  */
1607
1608 static void __init enable_IO_APIC(void)
1609 {
1610         union IO_APIC_reg_01 reg_01;
1611         int i8259_apic, i8259_pin;
1612         int i, apic;
1613         unsigned long flags;
1614
1615         for (i = 0; i < PIN_MAP_SIZE; i++) {
1616                 irq_2_pin[i].pin = -1;
1617                 irq_2_pin[i].next = 0;
1618         }
1619         if (!pirqs_enabled)
1620                 for (i = 0; i < MAX_PIRQS; i++)
1621                         pirq_entries[i] = -1;
1622
1623         /*
1624          * The number of IO-APIC IRQ registers (== #pins):
1625          */
1626         for (apic = 0; apic < nr_ioapics; apic++) {
1627                 spin_lock_irqsave(&ioapic_lock, flags);
1628                 reg_01.raw = io_apic_read(apic, 1);
1629                 spin_unlock_irqrestore(&ioapic_lock, flags);
1630                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1631         }
1632         for(apic = 0; apic < nr_ioapics; apic++) {
1633                 int pin;
1634                 /* See if any of the pins is in ExtINT mode */
1635                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1636                         struct IO_APIC_route_entry entry;
1637                         entry = ioapic_read_entry(apic, pin);
1638
1639
1640                         /* If the interrupt line is enabled and in ExtInt mode
1641                          * I have found the pin where the i8259 is connected.
1642                          */
1643                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1644                                 ioapic_i8259.apic = apic;
1645                                 ioapic_i8259.pin  = pin;
1646                                 goto found_i8259;
1647                         }
1648                 }
1649         }
1650  found_i8259:
1651         /* Look to see what if the MP table has reported the ExtINT */
1652         /* If we could not find the appropriate pin by looking at the ioapic
1653          * the i8259 probably is not connected the ioapic but give the
1654          * mptable a chance anyway.
1655          */
1656         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1657         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1658         /* Trust the MP table if nothing is setup in the hardware */
1659         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1660                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1661                 ioapic_i8259.pin  = i8259_pin;
1662                 ioapic_i8259.apic = i8259_apic;
1663         }
1664         /* Complain if the MP table and the hardware disagree */
1665         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1666                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1667         {
1668                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1669         }
1670
1671         /*
1672          * Do not trust the IO-APIC being empty at bootup
1673          */
1674         clear_IO_APIC();
1675 }
1676
1677 /*
1678  * Not an __init, needed by the reboot code
1679  */
1680 void disable_IO_APIC(void)
1681 {
1682         /*
1683          * Clear the IO-APIC before rebooting:
1684          */
1685         clear_IO_APIC();
1686
1687         /*
1688          * If the i8259 is routed through an IOAPIC
1689          * Put that IOAPIC in virtual wire mode
1690          * so legacy interrupts can be delivered.
1691          */
1692         if (ioapic_i8259.pin != -1) {
1693                 struct IO_APIC_route_entry entry;
1694
1695                 memset(&entry, 0, sizeof(entry));
1696                 entry.mask            = 0; /* Enabled */
1697                 entry.trigger         = 0; /* Edge */
1698                 entry.irr             = 0;
1699                 entry.polarity        = 0; /* High */
1700                 entry.delivery_status = 0;
1701                 entry.dest_mode       = 0; /* Physical */
1702                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1703                 entry.vector          = 0;
1704                 entry.dest.physical.physical_dest =
1705                                         GET_APIC_ID(read_apic_id());
1706
1707                 /*
1708                  * Add it to the IO-APIC irq-routing table:
1709                  */
1710                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1711         }
1712         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1713 }
1714
1715 /*
1716  * function to set the IO-APIC physical IDs based on the
1717  * values stored in the MPC table.
1718  *
1719  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
1720  */
1721
1722 #ifndef CONFIG_X86_NUMAQ
1723 static void __init setup_ioapic_ids_from_mpc(void)
1724 {
1725         union IO_APIC_reg_00 reg_00;
1726         physid_mask_t phys_id_present_map;
1727         int apic;
1728         int i;
1729         unsigned char old_id;
1730         unsigned long flags;
1731
1732         /*
1733          * Don't check I/O APIC IDs for xAPIC systems.  They have
1734          * no meaning without the serial APIC bus.
1735          */
1736         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1737                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
1738                 return;
1739         /*
1740          * This is broken; anything with a real cpu count has to
1741          * circumvent this idiocy regardless.
1742          */
1743         phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1744
1745         /*
1746          * Set the IOAPIC ID to the value stored in the MPC table.
1747          */
1748         for (apic = 0; apic < nr_ioapics; apic++) {
1749
1750                 /* Read the register 0 value */
1751                 spin_lock_irqsave(&ioapic_lock, flags);
1752                 reg_00.raw = io_apic_read(apic, 0);
1753                 spin_unlock_irqrestore(&ioapic_lock, flags);
1754
1755                 old_id = mp_ioapics[apic].mpc_apicid;
1756
1757                 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) {
1758                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1759                                 apic, mp_ioapics[apic].mpc_apicid);
1760                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1761                                 reg_00.bits.ID);
1762                         mp_ioapics[apic].mpc_apicid = reg_00.bits.ID;
1763                 }
1764
1765                 /*
1766                  * Sanity check, is the ID really free? Every APIC in a
1767                  * system must have a unique ID or we get lots of nice
1768                  * 'stuck on smp_invalidate_needed IPI wait' messages.
1769                  */
1770                 if (check_apicid_used(phys_id_present_map,
1771                                         mp_ioapics[apic].mpc_apicid)) {
1772                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1773                                 apic, mp_ioapics[apic].mpc_apicid);
1774                         for (i = 0; i < get_physical_broadcast(); i++)
1775                                 if (!physid_isset(i, phys_id_present_map))
1776                                         break;
1777                         if (i >= get_physical_broadcast())
1778                                 panic("Max APIC ID exceeded!\n");
1779                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1780                                 i);
1781                         physid_set(i, phys_id_present_map);
1782                         mp_ioapics[apic].mpc_apicid = i;
1783                 } else {
1784                         physid_mask_t tmp;
1785                         tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid);
1786                         apic_printk(APIC_VERBOSE, "Setting %d in the "
1787                                         "phys_id_present_map\n",
1788                                         mp_ioapics[apic].mpc_apicid);
1789                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
1790                 }
1791
1792
1793                 /*
1794                  * We need to adjust the IRQ routing table
1795                  * if the ID changed.
1796                  */
1797                 if (old_id != mp_ioapics[apic].mpc_apicid)
1798                         for (i = 0; i < mp_irq_entries; i++)
1799                                 if (mp_irqs[i].mpc_dstapic == old_id)
1800                                         mp_irqs[i].mpc_dstapic
1801                                                 = mp_ioapics[apic].mpc_apicid;
1802
1803                 /*
1804                  * Read the right value from the MPC table and
1805                  * write it into the ID register.
1806                  */
1807                 apic_printk(APIC_VERBOSE, KERN_INFO
1808                         "...changing IO-APIC physical APIC ID to %d ...",
1809                         mp_ioapics[apic].mpc_apicid);
1810
1811                 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1812                 spin_lock_irqsave(&ioapic_lock, flags);
1813                 io_apic_write(apic, 0, reg_00.raw);
1814                 spin_unlock_irqrestore(&ioapic_lock, flags);
1815
1816                 /*
1817                  * Sanity check
1818                  */
1819                 spin_lock_irqsave(&ioapic_lock, flags);
1820                 reg_00.raw = io_apic_read(apic, 0);
1821                 spin_unlock_irqrestore(&ioapic_lock, flags);
1822                 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1823                         printk("could not set ID!\n");
1824                 else
1825                         apic_printk(APIC_VERBOSE, " ok.\n");
1826         }
1827 }
1828 #else
1829 static void __init setup_ioapic_ids_from_mpc(void) { }
1830 #endif
1831
1832 int no_timer_check __initdata;
1833
1834 static int __init notimercheck(char *s)
1835 {
1836         no_timer_check = 1;
1837         return 1;
1838 }
1839 __setup("no_timer_check", notimercheck);
1840
1841 /*
1842  * There is a nasty bug in some older SMP boards, their mptable lies
1843  * about the timer IRQ. We do the following to work around the situation:
1844  *
1845  *      - timer IRQ defaults to IO-APIC IRQ
1846  *      - if this function detects that timer IRQs are defunct, then we fall
1847  *        back to ISA timer IRQs
1848  */
1849 static int __init timer_irq_works(void)
1850 {
1851         unsigned long t1 = jiffies;
1852         unsigned long flags;
1853
1854         if (no_timer_check)
1855                 return 1;
1856
1857         local_save_flags(flags);
1858         local_irq_enable();
1859         /* Let ten ticks pass... */
1860         mdelay((10 * 1000) / HZ);
1861         local_irq_restore(flags);
1862
1863         /*
1864          * Expect a few ticks at least, to be sure some possible
1865          * glue logic does not lock up after one or two first
1866          * ticks in a non-ExtINT mode.  Also the local APIC
1867          * might have cached one ExtINT interrupt.  Finally, at
1868          * least one tick may be lost due to delays.
1869          */
1870         if (time_after(jiffies, t1 + 4))
1871                 return 1;
1872
1873         return 0;
1874 }
1875
1876 /*
1877  * In the SMP+IOAPIC case it might happen that there are an unspecified
1878  * number of pending IRQ events unhandled. These cases are very rare,
1879  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1880  * better to do it this way as thus we do not have to be aware of
1881  * 'pending' interrupts in the IRQ path, except at this point.
1882  */
1883 /*
1884  * Edge triggered needs to resend any interrupt
1885  * that was delayed but this is now handled in the device
1886  * independent code.
1887  */
1888
1889 /*
1890  * Startup quirk:
1891  *
1892  * Starting up a edge-triggered IO-APIC interrupt is
1893  * nasty - we need to make sure that we get the edge.
1894  * If it is already asserted for some reason, we need
1895  * return 1 to indicate that is was pending.
1896  *
1897  * This is not complete - we should be able to fake
1898  * an edge even if it isn't on the 8259A...
1899  *
1900  * (We do this for level-triggered IRQs too - it cannot hurt.)
1901  */
1902 static unsigned int startup_ioapic_irq(unsigned int irq)
1903 {
1904         int was_pending = 0;
1905         unsigned long flags;
1906
1907         spin_lock_irqsave(&ioapic_lock, flags);
1908         if (irq < 16) {
1909                 disable_8259A_irq(irq);
1910                 if (i8259A_irq_pending(irq))
1911                         was_pending = 1;
1912         }
1913         __unmask_IO_APIC_irq(irq);
1914         spin_unlock_irqrestore(&ioapic_lock, flags);
1915
1916         return was_pending;
1917 }
1918
1919 static void ack_ioapic_irq(unsigned int irq)
1920 {
1921         move_native_irq(irq);
1922         ack_APIC_irq();
1923 }
1924
1925 static void ack_ioapic_quirk_irq(unsigned int irq)
1926 {
1927         unsigned long v;
1928         int i;
1929
1930         move_native_irq(irq);
1931 /*
1932  * It appears there is an erratum which affects at least version 0x11
1933  * of I/O APIC (that's the 82093AA and cores integrated into various
1934  * chipsets).  Under certain conditions a level-triggered interrupt is
1935  * erroneously delivered as edge-triggered one but the respective IRR
1936  * bit gets set nevertheless.  As a result the I/O unit expects an EOI
1937  * message but it will never arrive and further interrupts are blocked
1938  * from the source.  The exact reason is so far unknown, but the
1939  * phenomenon was observed when two consecutive interrupt requests
1940  * from a given source get delivered to the same CPU and the source is
1941  * temporarily disabled in between.
1942  *
1943  * A workaround is to simulate an EOI message manually.  We achieve it
1944  * by setting the trigger mode to edge and then to level when the edge
1945  * trigger mode gets detected in the TMR of a local APIC for a
1946  * level-triggered interrupt.  We mask the source for the time of the
1947  * operation to prevent an edge-triggered interrupt escaping meanwhile.
1948  * The idea is from Manfred Spraul.  --macro
1949  */
1950         i = irq_vector[irq];
1951
1952         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1953
1954         ack_APIC_irq();
1955
1956         if (!(v & (1 << (i & 0x1f)))) {
1957                 atomic_inc(&irq_mis_count);
1958                 spin_lock(&ioapic_lock);
1959                 __mask_and_edge_IO_APIC_irq(irq);
1960                 __unmask_and_level_IO_APIC_irq(irq);
1961                 spin_unlock(&ioapic_lock);
1962         }
1963 }
1964
1965 static int ioapic_retrigger_irq(unsigned int irq)
1966 {
1967         send_IPI_self(irq_vector[irq]);
1968
1969         return 1;
1970 }
1971
1972 static struct irq_chip ioapic_chip __read_mostly = {
1973         .name           = "IO-APIC",
1974         .startup        = startup_ioapic_irq,
1975         .mask           = mask_IO_APIC_irq,
1976         .unmask         = unmask_IO_APIC_irq,
1977         .ack            = ack_ioapic_irq,
1978         .eoi            = ack_ioapic_quirk_irq,
1979 #ifdef CONFIG_SMP
1980         .set_affinity   = set_ioapic_affinity_irq,
1981 #endif
1982         .retrigger      = ioapic_retrigger_irq,
1983 };
1984
1985
1986 static inline void init_IO_APIC_traps(void)
1987 {
1988         int irq;
1989
1990         /*
1991          * NOTE! The local APIC isn't very good at handling
1992          * multiple interrupts at the same interrupt level.
1993          * As the interrupt level is determined by taking the
1994          * vector number and shifting that right by 4, we
1995          * want to spread these out a bit so that they don't
1996          * all fall in the same interrupt level.
1997          *
1998          * Also, we've got to be careful not to trash gate
1999          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2000          */
2001         for (irq = 0; irq < NR_IRQS ; irq++) {
2002                 int tmp = irq;
2003                 if (IO_APIC_IRQ(tmp) && !irq_vector[tmp]) {
2004                         /*
2005                          * Hmm.. We don't have an entry for this,
2006                          * so default to an old-fashioned 8259
2007                          * interrupt if we can..
2008                          */
2009                         if (irq < 16)
2010                                 make_8259A_irq(irq);
2011                         else
2012                                 /* Strange. Oh, well.. */
2013                                 irq_desc[irq].chip = &no_irq_chip;
2014                 }
2015         }
2016 }
2017
2018 /*
2019  * The local APIC irq-chip implementation:
2020  */
2021
2022 static void ack_apic(unsigned int irq)
2023 {
2024         ack_APIC_irq();
2025 }
2026
2027 static void mask_lapic_irq (unsigned int irq)
2028 {
2029         unsigned long v;
2030
2031         v = apic_read(APIC_LVT0);
2032         apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2033 }
2034
2035 static void unmask_lapic_irq (unsigned int irq)
2036 {
2037         unsigned long v;
2038
2039         v = apic_read(APIC_LVT0);
2040         apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2041 }
2042
2043 static struct irq_chip lapic_chip __read_mostly = {
2044         .name           = "local-APIC-edge",
2045         .mask           = mask_lapic_irq,
2046         .unmask         = unmask_lapic_irq,
2047         .eoi            = ack_apic,
2048 };
2049
2050 static void __init setup_nmi(void)
2051 {
2052         /*
2053          * Dirty trick to enable the NMI watchdog ...
2054          * We put the 8259A master into AEOI mode and
2055          * unmask on all local APICs LVT0 as NMI.
2056          *
2057          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2058          * is from Maciej W. Rozycki - so we do not have to EOI from
2059          * the NMI handler or the timer interrupt.
2060          */
2061         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2062
2063         enable_NMI_through_LVT0();
2064
2065         apic_printk(APIC_VERBOSE, " done.\n");
2066 }
2067
2068 /*
2069  * This looks a bit hackish but it's about the only one way of sending
2070  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2071  * not support the ExtINT mode, unfortunately.  We need to send these
2072  * cycles as some i82489DX-based boards have glue logic that keeps the
2073  * 8259A interrupt line asserted until INTA.  --macro
2074  */
2075 static inline void unlock_ExtINT_logic(void)
2076 {
2077         int apic, pin, i;
2078         struct IO_APIC_route_entry entry0, entry1;
2079         unsigned char save_control, save_freq_select;
2080
2081         pin  = find_isa_irq_pin(8, mp_INT);
2082         if (pin == -1) {
2083                 WARN_ON_ONCE(1);
2084                 return;
2085         }
2086         apic = find_isa_irq_apic(8, mp_INT);
2087         if (apic == -1) {
2088                 WARN_ON_ONCE(1);
2089                 return;
2090         }
2091
2092         entry0 = ioapic_read_entry(apic, pin);
2093         clear_IO_APIC_pin(apic, pin);
2094
2095         memset(&entry1, 0, sizeof(entry1));
2096
2097         entry1.dest_mode = 0;                   /* physical delivery */
2098         entry1.mask = 0;                        /* unmask IRQ now */
2099         entry1.dest.physical.physical_dest = hard_smp_processor_id();
2100         entry1.delivery_mode = dest_ExtINT;
2101         entry1.polarity = entry0.polarity;
2102         entry1.trigger = 0;
2103         entry1.vector = 0;
2104
2105         ioapic_write_entry(apic, pin, entry1);
2106
2107         save_control = CMOS_READ(RTC_CONTROL);
2108         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2109         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2110                    RTC_FREQ_SELECT);
2111         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2112
2113         i = 100;
2114         while (i-- > 0) {
2115                 mdelay(10);
2116                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2117                         i -= 10;
2118         }
2119
2120         CMOS_WRITE(save_control, RTC_CONTROL);
2121         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2122         clear_IO_APIC_pin(apic, pin);
2123
2124         ioapic_write_entry(apic, pin, entry0);
2125 }
2126
2127 /*
2128  * This code may look a bit paranoid, but it's supposed to cooperate with
2129  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2130  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2131  * fanatically on his truly buggy board.
2132  */
2133 static inline void __init check_timer(void)
2134 {
2135         int apic1, pin1, apic2, pin2;
2136         int vector;
2137         unsigned int ver;
2138         unsigned long flags;
2139
2140         local_irq_save(flags);
2141
2142         ver = apic_read(APIC_LVR);
2143         ver = GET_APIC_VERSION(ver);
2144
2145         /*
2146          * get/set the timer IRQ vector:
2147          */
2148         disable_8259A_irq(0);
2149         vector = assign_irq_vector(0);
2150         set_intr_gate(vector, interrupt[0]);
2151
2152         /*
2153          * Subtle, code in do_timer_interrupt() expects an AEOI
2154          * mode for the 8259A whenever interrupts are routed
2155          * through I/O APICs.  Also IRQ0 has to be enabled in
2156          * the 8259A which implies the virtual wire has to be
2157          * disabled in the local APIC.  Finally timer interrupts
2158          * need to be acknowledged manually in the 8259A for
2159          * timer_interrupt() and for the i82489DX when using
2160          * the NMI watchdog.
2161          */
2162         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2163         init_8259A(1);
2164         timer_ack = !cpu_has_tsc;
2165         timer_ack |= (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2166         if (timer_over_8254 > 0)
2167                 enable_8259A_irq(0);
2168
2169         pin1  = find_isa_irq_pin(0, mp_INT);
2170         apic1 = find_isa_irq_apic(0, mp_INT);
2171         pin2  = ioapic_i8259.pin;
2172         apic2 = ioapic_i8259.apic;
2173
2174         printk(KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
2175                 vector, apic1, pin1, apic2, pin2);
2176
2177         if (pin1 != -1) {
2178                 /*
2179                  * Ok, does IRQ0 through the IOAPIC work?
2180                  */
2181                 unmask_IO_APIC_irq(0);
2182                 if (timer_irq_works()) {
2183                         if (nmi_watchdog == NMI_IO_APIC) {
2184                                 disable_8259A_irq(0);
2185                                 setup_nmi();
2186                                 enable_8259A_irq(0);
2187                         }
2188                         if (disable_timer_pin_1 > 0)
2189                                 clear_IO_APIC_pin(0, pin1);
2190                         goto out;
2191                 }
2192                 clear_IO_APIC_pin(apic1, pin1);
2193                 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to "
2194                                 "IO-APIC\n");
2195         }
2196
2197         printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2198         if (pin2 != -1) {
2199                 printk("\n..... (found pin %d) ...", pin2);
2200                 /*
2201                  * legacy devices should be connected to IO APIC #0
2202                  */
2203                 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
2204                 if (timer_irq_works()) {
2205                         printk("works.\n");
2206                         if (pin1 != -1)
2207                                 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2208                         else
2209                                 add_pin_to_irq(0, apic2, pin2);
2210                         if (nmi_watchdog == NMI_IO_APIC) {
2211                                 setup_nmi();
2212                         }
2213                         goto out;
2214                 }
2215                 /*
2216                  * Cleanup, just in case ...
2217                  */
2218                 clear_IO_APIC_pin(apic2, pin2);
2219         }
2220         printk(" failed.\n");
2221
2222         if (nmi_watchdog == NMI_IO_APIC) {
2223                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2224                 nmi_watchdog = 0;
2225         }
2226
2227         printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2228
2229         disable_8259A_irq(0);
2230         set_irq_chip_and_handler_name(0, &lapic_chip, handle_fasteoi_irq,
2231                                       "fasteoi");
2232         apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector);   /* Fixed mode */
2233         enable_8259A_irq(0);
2234
2235         if (timer_irq_works()) {
2236                 printk(" works.\n");
2237                 goto out;
2238         }
2239         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2240         printk(" failed.\n");
2241
2242         printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2243
2244         timer_ack = 0;
2245         init_8259A(0);
2246         make_8259A_irq(0);
2247         apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2248
2249         unlock_ExtINT_logic();
2250
2251         if (timer_irq_works()) {
2252                 printk(" works.\n");
2253                 goto out;
2254         }
2255         printk(" failed :(.\n");
2256         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
2257                 "report.  Then try booting with the 'noapic' option");
2258 out:
2259         local_irq_restore(flags);
2260 }
2261
2262 /*
2263  *
2264  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2265  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2266  *   Linux doesn't really care, as it's not actually used
2267  *   for any interrupt handling anyway.
2268  */
2269 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
2270
2271 void __init setup_IO_APIC(void)
2272 {
2273         int i;
2274
2275         /* Reserve all the system vectors. */
2276         for (i = FIRST_SYSTEM_VECTOR; i < NR_VECTORS; i++)
2277                 set_bit(i, used_vectors);
2278
2279         enable_IO_APIC();
2280
2281         if (acpi_ioapic)
2282                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
2283         else
2284                 io_apic_irqs = ~PIC_IRQS;
2285
2286         printk("ENABLING IO-APIC IRQs\n");
2287
2288         /*
2289          * Set up IO-APIC IRQ routing.
2290          */
2291         if (!acpi_ioapic)
2292                 setup_ioapic_ids_from_mpc();
2293         sync_Arb_IDs();
2294         setup_IO_APIC_irqs();
2295         init_IO_APIC_traps();
2296         check_timer();
2297         if (!acpi_ioapic)
2298                 print_IO_APIC();
2299 }
2300
2301 static int __init setup_disable_8254_timer(char *s)
2302 {
2303         timer_over_8254 = -1;
2304         return 1;
2305 }
2306 static int __init setup_enable_8254_timer(char *s)
2307 {
2308         timer_over_8254 = 2;
2309         return 1;
2310 }
2311
2312 __setup("disable_8254_timer", setup_disable_8254_timer);
2313 __setup("enable_8254_timer", setup_enable_8254_timer);
2314
2315 /*
2316  *      Called after all the initialization is done. If we didnt find any
2317  *      APIC bugs then we can allow the modify fast path
2318  */
2319
2320 static int __init io_apic_bug_finalize(void)
2321 {
2322         if(sis_apic_bug == -1)
2323                 sis_apic_bug = 0;
2324         return 0;
2325 }
2326
2327 late_initcall(io_apic_bug_finalize);
2328
2329 struct sysfs_ioapic_data {
2330         struct sys_device dev;
2331         struct IO_APIC_route_entry entry[0];
2332 };
2333 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2334
2335 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2336 {
2337         struct IO_APIC_route_entry *entry;
2338         struct sysfs_ioapic_data *data;
2339         int i;
2340
2341         data = container_of(dev, struct sysfs_ioapic_data, dev);
2342         entry = data->entry;
2343         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2344                 entry[i] = ioapic_read_entry(dev->id, i);
2345
2346         return 0;
2347 }
2348
2349 static int ioapic_resume(struct sys_device *dev)
2350 {
2351         struct IO_APIC_route_entry *entry;
2352         struct sysfs_ioapic_data *data;
2353         unsigned long flags;
2354         union IO_APIC_reg_00 reg_00;
2355         int i;
2356
2357         data = container_of(dev, struct sysfs_ioapic_data, dev);
2358         entry = data->entry;
2359
2360         spin_lock_irqsave(&ioapic_lock, flags);
2361         reg_00.raw = io_apic_read(dev->id, 0);
2362         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2363                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2364                 io_apic_write(dev->id, 0, reg_00.raw);
2365         }
2366         spin_unlock_irqrestore(&ioapic_lock, flags);
2367         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2368                 ioapic_write_entry(dev->id, i, entry[i]);
2369
2370         return 0;
2371 }
2372
2373 static struct sysdev_class ioapic_sysdev_class = {
2374         .name = "ioapic",
2375         .suspend = ioapic_suspend,
2376         .resume = ioapic_resume,
2377 };
2378
2379 static int __init ioapic_init_sysfs(void)
2380 {
2381         struct sys_device * dev;
2382         int i, size, error = 0;
2383
2384         error = sysdev_class_register(&ioapic_sysdev_class);
2385         if (error)
2386                 return error;
2387
2388         for (i = 0; i < nr_ioapics; i++ ) {
2389                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2390                         * sizeof(struct IO_APIC_route_entry);
2391                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2392                 if (!mp_ioapic_data[i]) {
2393                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2394                         continue;
2395                 }
2396                 memset(mp_ioapic_data[i], 0, size);
2397                 dev = &mp_ioapic_data[i]->dev;
2398                 dev->id = i;
2399                 dev->cls = &ioapic_sysdev_class;
2400                 error = sysdev_register(dev);
2401                 if (error) {
2402                         kfree(mp_ioapic_data[i]);
2403                         mp_ioapic_data[i] = NULL;
2404                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2405                         continue;
2406                 }
2407         }
2408
2409         return 0;
2410 }
2411
2412 device_initcall(ioapic_init_sysfs);
2413
2414 /*
2415  * Dynamic irq allocate and deallocation
2416  */
2417 int create_irq(void)
2418 {
2419         /* Allocate an unused irq */
2420         int irq, new, vector = 0;
2421         unsigned long flags;
2422
2423         irq = -ENOSPC;
2424         spin_lock_irqsave(&vector_lock, flags);
2425         for (new = (NR_IRQS - 1); new >= 0; new--) {
2426                 if (platform_legacy_irq(new))
2427                         continue;
2428                 if (irq_vector[new] != 0)
2429                         continue;
2430                 vector = __assign_irq_vector(new);
2431                 if (likely(vector > 0))
2432                         irq = new;
2433                 break;
2434         }
2435         spin_unlock_irqrestore(&vector_lock, flags);
2436
2437         if (irq >= 0) {
2438                 set_intr_gate(vector, interrupt[irq]);
2439                 dynamic_irq_init(irq);
2440         }
2441         return irq;
2442 }
2443
2444 void destroy_irq(unsigned int irq)
2445 {
2446         unsigned long flags;
2447
2448         dynamic_irq_cleanup(irq);
2449
2450         spin_lock_irqsave(&vector_lock, flags);
2451         irq_vector[irq] = 0;
2452         spin_unlock_irqrestore(&vector_lock, flags);
2453 }
2454
2455 /*
2456  * MSI message composition
2457  */
2458 #ifdef CONFIG_PCI_MSI
2459 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2460 {
2461         int vector;
2462         unsigned dest;
2463
2464         vector = assign_irq_vector(irq);
2465         if (vector >= 0) {
2466                 dest = cpu_mask_to_apicid(TARGET_CPUS);
2467
2468                 msg->address_hi = MSI_ADDR_BASE_HI;
2469                 msg->address_lo =
2470                         MSI_ADDR_BASE_LO |
2471                         ((INT_DEST_MODE == 0) ?
2472                                 MSI_ADDR_DEST_MODE_PHYSICAL:
2473                                 MSI_ADDR_DEST_MODE_LOGICAL) |
2474                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2475                                 MSI_ADDR_REDIRECTION_CPU:
2476                                 MSI_ADDR_REDIRECTION_LOWPRI) |
2477                         MSI_ADDR_DEST_ID(dest);
2478
2479                 msg->data =
2480                         MSI_DATA_TRIGGER_EDGE |
2481                         MSI_DATA_LEVEL_ASSERT |
2482                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2483                                 MSI_DATA_DELIVERY_FIXED:
2484                                 MSI_DATA_DELIVERY_LOWPRI) |
2485                         MSI_DATA_VECTOR(vector);
2486         }
2487         return vector;
2488 }
2489
2490 #ifdef CONFIG_SMP
2491 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2492 {
2493         struct msi_msg msg;
2494         unsigned int dest;
2495         cpumask_t tmp;
2496         int vector;
2497
2498         cpus_and(tmp, mask, cpu_online_map);
2499         if (cpus_empty(tmp))
2500                 tmp = TARGET_CPUS;
2501
2502         vector = assign_irq_vector(irq);
2503         if (vector < 0)
2504                 return;
2505
2506         dest = cpu_mask_to_apicid(mask);
2507
2508         read_msi_msg(irq, &msg);
2509
2510         msg.data &= ~MSI_DATA_VECTOR_MASK;
2511         msg.data |= MSI_DATA_VECTOR(vector);
2512         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2513         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2514
2515         write_msi_msg(irq, &msg);
2516         irq_desc[irq].affinity = mask;
2517 }
2518 #endif /* CONFIG_SMP */
2519
2520 /*
2521  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2522  * which implement the MSI or MSI-X Capability Structure.
2523  */
2524 static struct irq_chip msi_chip = {
2525         .name           = "PCI-MSI",
2526         .unmask         = unmask_msi_irq,
2527         .mask           = mask_msi_irq,
2528         .ack            = ack_ioapic_irq,
2529 #ifdef CONFIG_SMP
2530         .set_affinity   = set_msi_irq_affinity,
2531 #endif
2532         .retrigger      = ioapic_retrigger_irq,
2533 };
2534
2535 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2536 {
2537         struct msi_msg msg;
2538         int irq, ret;
2539         irq = create_irq();
2540         if (irq < 0)
2541                 return irq;
2542
2543         ret = msi_compose_msg(dev, irq, &msg);
2544         if (ret < 0) {
2545                 destroy_irq(irq);
2546                 return ret;
2547         }
2548
2549         set_irq_msi(irq, desc);
2550         write_msi_msg(irq, &msg);
2551
2552         set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq,
2553                                       "edge");
2554
2555         return 0;
2556 }
2557
2558 void arch_teardown_msi_irq(unsigned int irq)
2559 {
2560         destroy_irq(irq);
2561 }
2562
2563 #endif /* CONFIG_PCI_MSI */
2564
2565 /*
2566  * Hypertransport interrupt support
2567  */
2568 #ifdef CONFIG_HT_IRQ
2569
2570 #ifdef CONFIG_SMP
2571
2572 static void target_ht_irq(unsigned int irq, unsigned int dest)
2573 {
2574         struct ht_irq_msg msg;
2575         fetch_ht_irq_msg(irq, &msg);
2576
2577         msg.address_lo &= ~(HT_IRQ_LOW_DEST_ID_MASK);
2578         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2579
2580         msg.address_lo |= HT_IRQ_LOW_DEST_ID(dest);
2581         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2582
2583         write_ht_irq_msg(irq, &msg);
2584 }
2585
2586 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2587 {
2588         unsigned int dest;
2589         cpumask_t tmp;
2590
2591         cpus_and(tmp, mask, cpu_online_map);
2592         if (cpus_empty(tmp))
2593                 tmp = TARGET_CPUS;
2594
2595         cpus_and(mask, tmp, CPU_MASK_ALL);
2596
2597         dest = cpu_mask_to_apicid(mask);
2598
2599         target_ht_irq(irq, dest);
2600         irq_desc[irq].affinity = mask;
2601 }
2602 #endif
2603
2604 static struct irq_chip ht_irq_chip = {
2605         .name           = "PCI-HT",
2606         .mask           = mask_ht_irq,
2607         .unmask         = unmask_ht_irq,
2608         .ack            = ack_ioapic_irq,
2609 #ifdef CONFIG_SMP
2610         .set_affinity   = set_ht_irq_affinity,
2611 #endif
2612         .retrigger      = ioapic_retrigger_irq,
2613 };
2614
2615 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2616 {
2617         int vector;
2618
2619         vector = assign_irq_vector(irq);
2620         if (vector >= 0) {
2621                 struct ht_irq_msg msg;
2622                 unsigned dest;
2623                 cpumask_t tmp;
2624
2625                 cpus_clear(tmp);
2626                 cpu_set(vector >> 8, tmp);
2627                 dest = cpu_mask_to_apicid(tmp);
2628
2629                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2630
2631                 msg.address_lo =
2632                         HT_IRQ_LOW_BASE |
2633                         HT_IRQ_LOW_DEST_ID(dest) |
2634                         HT_IRQ_LOW_VECTOR(vector) |
2635                         ((INT_DEST_MODE == 0) ?
2636                                 HT_IRQ_LOW_DM_PHYSICAL :
2637                                 HT_IRQ_LOW_DM_LOGICAL) |
2638                         HT_IRQ_LOW_RQEOI_EDGE |
2639                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2640                                 HT_IRQ_LOW_MT_FIXED :
2641                                 HT_IRQ_LOW_MT_ARBITRATED) |
2642                         HT_IRQ_LOW_IRQ_MASKED;
2643
2644                 write_ht_irq_msg(irq, &msg);
2645
2646                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2647                                               handle_edge_irq, "edge");
2648         }
2649         return vector;
2650 }
2651 #endif /* CONFIG_HT_IRQ */
2652
2653 /* --------------------------------------------------------------------------
2654                           ACPI-based IOAPIC Configuration
2655    -------------------------------------------------------------------------- */
2656
2657 #ifdef CONFIG_ACPI
2658
2659 int __init io_apic_get_unique_id (int ioapic, int apic_id)
2660 {
2661         union IO_APIC_reg_00 reg_00;
2662         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2663         physid_mask_t tmp;
2664         unsigned long flags;
2665         int i = 0;
2666
2667         /*
2668          * The P4 platform supports up to 256 APIC IDs on two separate APIC
2669          * buses (one for LAPICs, one for IOAPICs), where predecessors only
2670          * supports up to 16 on one shared APIC bus.
2671          *
2672          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2673          *      advantage of new APIC bus architecture.
2674          */
2675
2676         if (physids_empty(apic_id_map))
2677                 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2678
2679         spin_lock_irqsave(&ioapic_lock, flags);
2680         reg_00.raw = io_apic_read(ioapic, 0);
2681         spin_unlock_irqrestore(&ioapic_lock, flags);
2682
2683         if (apic_id >= get_physical_broadcast()) {
2684                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2685                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
2686                 apic_id = reg_00.bits.ID;
2687         }
2688
2689         /*
2690          * Every APIC in a system must have a unique ID or we get lots of nice
2691          * 'stuck on smp_invalidate_needed IPI wait' messages.
2692          */
2693         if (check_apicid_used(apic_id_map, apic_id)) {
2694
2695                 for (i = 0; i < get_physical_broadcast(); i++) {
2696                         if (!check_apicid_used(apic_id_map, i))
2697                                 break;
2698                 }
2699
2700                 if (i == get_physical_broadcast())
2701                         panic("Max apic_id exceeded!\n");
2702
2703                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2704                         "trying %d\n", ioapic, apic_id, i);
2705
2706                 apic_id = i;
2707         }
2708
2709         tmp = apicid_to_cpu_present(apic_id);
2710         physids_or(apic_id_map, apic_id_map, tmp);
2711
2712         if (reg_00.bits.ID != apic_id) {
2713                 reg_00.bits.ID = apic_id;
2714
2715                 spin_lock_irqsave(&ioapic_lock, flags);
2716                 io_apic_write(ioapic, 0, reg_00.raw);
2717                 reg_00.raw = io_apic_read(ioapic, 0);
2718                 spin_unlock_irqrestore(&ioapic_lock, flags);
2719
2720                 /* Sanity check */
2721                 if (reg_00.bits.ID != apic_id) {
2722                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
2723                         return -1;
2724                 }
2725         }
2726
2727         apic_printk(APIC_VERBOSE, KERN_INFO
2728                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2729
2730         return apic_id;
2731 }
2732
2733
2734 int __init io_apic_get_version (int ioapic)
2735 {
2736         union IO_APIC_reg_01    reg_01;
2737         unsigned long flags;
2738
2739         spin_lock_irqsave(&ioapic_lock, flags);
2740         reg_01.raw = io_apic_read(ioapic, 1);
2741         spin_unlock_irqrestore(&ioapic_lock, flags);
2742
2743         return reg_01.bits.version;
2744 }
2745
2746
2747 int __init io_apic_get_redir_entries (int ioapic)
2748 {
2749         union IO_APIC_reg_01    reg_01;
2750         unsigned long flags;
2751
2752         spin_lock_irqsave(&ioapic_lock, flags);
2753         reg_01.raw = io_apic_read(ioapic, 1);
2754         spin_unlock_irqrestore(&ioapic_lock, flags);
2755
2756         return reg_01.bits.entries;
2757 }
2758
2759
2760 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2761 {
2762         struct IO_APIC_route_entry entry;
2763         unsigned long flags;
2764
2765         if (!IO_APIC_IRQ(irq)) {
2766                 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2767                         ioapic);
2768                 return -EINVAL;
2769         }
2770
2771         /*
2772          * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2773          * Note that we mask (disable) IRQs now -- these get enabled when the
2774          * corresponding device driver registers for this IRQ.
2775          */
2776
2777         memset(&entry,0,sizeof(entry));
2778
2779         entry.delivery_mode = INT_DELIVERY_MODE;
2780         entry.dest_mode = INT_DEST_MODE;
2781         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2782         entry.trigger = edge_level;
2783         entry.polarity = active_high_low;
2784         entry.mask  = 1;
2785
2786         /*
2787          * IRQs < 16 are already in the irq_2_pin[] map
2788          */
2789         if (irq >= 16)
2790                 add_pin_to_irq(irq, ioapic, pin);
2791
2792         entry.vector = assign_irq_vector(irq);
2793
2794         apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2795                 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2796                 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2797                 edge_level, active_high_low);
2798
2799         ioapic_register_intr(irq, entry.vector, edge_level);
2800
2801         if (!ioapic && (irq < 16))
2802                 disable_8259A_irq(irq);
2803
2804         spin_lock_irqsave(&ioapic_lock, flags);
2805         __ioapic_write_entry(ioapic, pin, entry);
2806         spin_unlock_irqrestore(&ioapic_lock, flags);
2807
2808         return 0;
2809 }
2810
2811 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2812 {
2813         int i;
2814
2815         if (skip_ioapic_setup)
2816                 return -1;
2817
2818         for (i = 0; i < mp_irq_entries; i++)
2819                 if (mp_irqs[i].mpc_irqtype == mp_INT &&
2820                     mp_irqs[i].mpc_srcbusirq == bus_irq)
2821                         break;
2822         if (i >= mp_irq_entries)
2823                 return -1;
2824
2825         *trigger = irq_trigger(i);
2826         *polarity = irq_polarity(i);
2827         return 0;
2828 }
2829
2830 #endif /* CONFIG_ACPI */
2831
2832 static int __init parse_disable_timer_pin_1(char *arg)
2833 {
2834         disable_timer_pin_1 = 1;
2835         return 0;
2836 }
2837 early_param("disable_timer_pin_1", parse_disable_timer_pin_1);
2838
2839 static int __init parse_enable_timer_pin_1(char *arg)
2840 {
2841         disable_timer_pin_1 = -1;
2842         return 0;
2843 }
2844 early_param("enable_timer_pin_1", parse_enable_timer_pin_1);
2845
2846 static int __init parse_noapic(char *arg)
2847 {
2848         /* disable IO-APIC */
2849         disable_ioapic_setup();
2850         return 0;
2851 }
2852 early_param("noapic", parse_noapic);