git.oblomov.eu Git - linux-2.6/blob - 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
  39 #include <asm/io.h>
  40 #include <asm/smp.h>
  41 #include <asm/desc.h>
  42 #include <asm/timer.h>
  43 #include <asm/i8259.h>
  44 #include <asm/nmi.h>
  45 #include <asm/msidef.h>
  46 #include <asm/hypertransport.h>
  47
  48 #include <mach_apic.h>
  49 #include <mach_apicdef.h>
  50
  51 #include "io_ports.h"
  52
  53 int (*ioapic_renumber_irq)(int ioapic, int irq);
  54 atomic_t irq_mis_count;
  55
  56 /* Where if anywhere is the i8259 connect in external int mode */
  57 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
  58
  59 static DEFINE_SPINLOCK(ioapic_lock);
  60 static DEFINE_SPINLOCK(vector_lock);
  61
  62 int timer_over_8254 __initdata = 1;
  63
  64 /*
  65  *      Is the SiS APIC rmw bug present ?
  66  *      -1 = don't know, 0 = no, 1 = yes
  67  */
  68 int sis_apic_bug = -1;
  69
  70 /*
  71  * # of IRQ routing registers
  72  */
  73 int nr_ioapic_registers[MAX_IO_APICS];
  74
  75 static int disable_timer_pin_1 __initdata;
  76
  77 /*
  78  * Rough estimation of how many shared IRQs there are, can
  79  * be changed anytime.
  80  */
  81 #define MAX_PLUS_SHARED_IRQS NR_IRQS
  82 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
  83
  84 /*
  85  * This is performance-critical, we want to do it O(1)
  86  *
  87  * the indexing order of this array favors 1:1 mappings
  88  * between pins and IRQs.
  89  */
  90
  91 static struct irq_pin_list {
  92         int apic, pin, next;
  93 } irq_2_pin[PIN_MAP_SIZE];
  94
  95 struct io_apic {
  96         unsigned int index;
  97         unsigned int unused[3];
  98         unsigned int data;
  99 };
 100
 101 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
 102 {
 103         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
 104                 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
 105 }
 106
 107 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
 108 {
 109         struct io_apic __iomem *io_apic = io_apic_base(apic);
 110         writel(reg, &io_apic->index);
 111         return readl(&io_apic->data);
 112 }
 113
 114 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
 115 {
 116         struct io_apic __iomem *io_apic = io_apic_base(apic);
 117         writel(reg, &io_apic->index);
 118         writel(value, &io_apic->data);
 119 }
 120
 121 /*
 122  * Re-write a value: to be used for read-modify-write
 123  * cycles where the read already set up the index register.
 124  *
 125  * Older SiS APIC requires we rewrite the index register
 126  */
 127 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
 128 {
 129         volatile struct io_apic __iomem *io_apic = io_apic_base(apic);
 130         if (sis_apic_bug)
 131                 writel(reg, &io_apic->index);
 132         writel(value, &io_apic->data);
 133 }
 134
 135 union entry_union {
 136         struct { u32 w1, w2; };
 137         struct IO_APIC_route_entry entry;
 138 };
 139
 140 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
 141 {
 142         union entry_union eu;
 143         unsigned long flags;
 144         spin_lock_irqsave(&ioapic_lock, flags);
 145         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
 146         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
 147         spin_unlock_irqrestore(&ioapic_lock, flags);
 148         return eu.entry;
 149 }
 150
 151 /*
 152  * When we write a new IO APIC routing entry, we need to write the high
 153  * word first! If the mask bit in the low word is clear, we will enable
 154  * the interrupt, and we need to make sure the entry is fully populated
 155  * before that happens.
 156  */
 157 static void
 158 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 159 {
 160         union entry_union eu;
 161         eu.entry = e;
 162         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 163         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 164 }
 165
 166 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 167 {
 168         unsigned long flags;
 169         spin_lock_irqsave(&ioapic_lock, flags);
 170         __ioapic_write_entry(apic, pin, e);
 171         spin_unlock_irqrestore(&ioapic_lock, flags);
 172 }
 173
 174 /*
 175  * When we mask an IO APIC routing entry, we need to write the low
 176  * word first, in order to set the mask bit before we change the
 177  * high bits!
 178  */
 179 static void ioapic_mask_entry(int apic, int pin)
 180 {
 181         unsigned long flags;
 182         union entry_union eu = { .entry.mask = 1 };
 183
 184         spin_lock_irqsave(&ioapic_lock, flags);
 185         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 186         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 187         spin_unlock_irqrestore(&ioapic_lock, flags);
 188 }
 189
 190 /*
 191  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
 192  * shared ISA-space IRQs, so we have to support them. We are super
 193  * fast in the common case, and fast for shared ISA-space IRQs.
 194  */
 195 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 196 {
 197         static int first_free_entry = NR_IRQS;
 198         struct irq_pin_list *entry = irq_2_pin + irq;
 199
 200         while (entry->next)
 201                 entry = irq_2_pin + entry->next;
 202
 203         if (entry->pin != -1) {
 204                 entry->next = first_free_entry;
 205                 entry = irq_2_pin + entry->next;
 206                 if (++first_free_entry >= PIN_MAP_SIZE)
 207                         panic("io_apic.c: whoops");
 208         }
 209         entry->apic = apic;
 210         entry->pin = pin;
 211 }
 212
 213 /*
 214  * Reroute an IRQ to a different pin.
 215  */
 216 static void __init replace_pin_at_irq(unsigned int irq,
 217                                       int oldapic, int oldpin,
 218                                       int newapic, int newpin)
 219 {
 220         struct irq_pin_list *entry = irq_2_pin + irq;
 221
 222         while (1) {
 223                 if (entry->apic == oldapic && entry->pin == oldpin) {
 224                         entry->apic = newapic;
 225                         entry->pin = newpin;
 226                 }
 227                 if (!entry->next)
 228                         break;
 229                 entry = irq_2_pin + entry->next;
 230         }
 231 }
 232
 233 static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable)
 234 {
 235         struct irq_pin_list *entry = irq_2_pin + irq;
 236         unsigned int pin, reg;
 237
 238         for (;;) {
 239                 pin = entry->pin;
 240                 if (pin == -1)
 241                         break;
 242                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
 243                 reg &= ~disable;
 244                 reg |= enable;
 245                 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
 246                 if (!entry->next)
 247                         break;
 248                 entry = irq_2_pin + entry->next;
 249         }
 250 }
 251
 252 /* mask = 1 */
 253 static void __mask_IO_APIC_irq (unsigned int irq)
 254 {
 255         __modify_IO_APIC_irq(irq, 0x00010000, 0);
 256 }
 257
 258 /* mask = 0 */
 259 static void __unmask_IO_APIC_irq (unsigned int irq)
 260 {
 261         __modify_IO_APIC_irq(irq, 0, 0x00010000);
 262 }
 263
 264 /* mask = 1, trigger = 0 */
 265 static void __mask_and_edge_IO_APIC_irq (unsigned int irq)
 266 {
 267         __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000);
 268 }
 269
 270 /* mask = 0, trigger = 1 */
 271 static void __unmask_and_level_IO_APIC_irq (unsigned int irq)
 272 {
 273         __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000);
 274 }
 275
 276 static void mask_IO_APIC_irq (unsigned int irq)
 277 {
 278         unsigned long flags;
 279
 280         spin_lock_irqsave(&ioapic_lock, flags);
 281         __mask_IO_APIC_irq(irq);
 282         spin_unlock_irqrestore(&ioapic_lock, flags);
 283 }
 284
 285 static void unmask_IO_APIC_irq (unsigned int irq)
 286 {
 287         unsigned long flags;
 288
 289         spin_lock_irqsave(&ioapic_lock, flags);
 290         __unmask_IO_APIC_irq(irq);
 291         spin_unlock_irqrestore(&ioapic_lock, flags);
 292 }
 293
 294 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 295 {
 296         struct IO_APIC_route_entry entry;
 297
 298         /* Check delivery_mode to be sure we're not clearing an SMI pin */
 299         entry = ioapic_read_entry(apic, pin);
 300         if (entry.delivery_mode == dest_SMI)
 301                 return;
 302
 303         /*
 304          * Disable it in the IO-APIC irq-routing table:
 305          */
 306         ioapic_mask_entry(apic, pin);
 307 }
 308
 309 static void clear_IO_APIC (void)
 310 {
 311         int apic, pin;
 312
 313         for (apic = 0; apic < nr_ioapics; apic++)
 314                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 315                         clear_IO_APIC_pin(apic, pin);
 316 }
 317
 318 #ifdef CONFIG_SMP
 319 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
 320 {
 321         unsigned long flags;
 322         int pin;
 323         struct irq_pin_list *entry = irq_2_pin + irq;
 324         unsigned int apicid_value;
 325         cpumask_t tmp;
 326
 327         cpus_and(tmp, cpumask, cpu_online_map);
 328         if (cpus_empty(tmp))
 329                 tmp = TARGET_CPUS;
 330
 331         cpus_and(cpumask, tmp, CPU_MASK_ALL);
 332
 333         apicid_value = cpu_mask_to_apicid(cpumask);
 334         /* Prepare to do the io_apic_write */
 335         apicid_value = apicid_value << 24;
 336         spin_lock_irqsave(&ioapic_lock, flags);
 337         for (;;) {
 338                 pin = entry->pin;
 339                 if (pin == -1)
 340                         break;
 341                 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
 342                 if (!entry->next)
 343                         break;
 344                 entry = irq_2_pin + entry->next;
 345         }
 346         irq_desc[irq].affinity = cpumask;
 347         spin_unlock_irqrestore(&ioapic_lock, flags);
 348 }
 349
 350 #if defined(CONFIG_IRQBALANCE)
 351 # include <asm/processor.h>     /* kernel_thread() */
 352 # include <linux/kernel_stat.h> /* kstat */
 353 # include <linux/slab.h>                /* kmalloc() */
 354 # include <linux/timer.h>       /* time_after() */
 355
 356 #define IRQBALANCE_CHECK_ARCH -999
 357 #define MAX_BALANCED_IRQ_INTERVAL       (5*HZ)
 358 #define MIN_BALANCED_IRQ_INTERVAL       (HZ/2)
 359 #define BALANCED_IRQ_MORE_DELTA         (HZ/10)
 360 #define BALANCED_IRQ_LESS_DELTA         (HZ)
 361
 362 static int irqbalance_disabled __read_mostly = IRQBALANCE_CHECK_ARCH;
 363 static int physical_balance __read_mostly;
 364 static long balanced_irq_interval __read_mostly = MAX_BALANCED_IRQ_INTERVAL;
 365
 366 static struct irq_cpu_info {
 367         unsigned long * last_irq;
 368         unsigned long * irq_delta;
 369         unsigned long irq;
 370 } irq_cpu_data[NR_CPUS];
 371
 372 #define CPU_IRQ(cpu)            (irq_cpu_data[cpu].irq)
 373 #define LAST_CPU_IRQ(cpu,irq)   (irq_cpu_data[cpu].last_irq[irq])
 374 #define IRQ_DELTA(cpu,irq)      (irq_cpu_data[cpu].irq_delta[irq])
 375
 376 #define IDLE_ENOUGH(cpu,now) \
 377         (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
 378
 379 #define IRQ_ALLOWED(cpu, allowed_mask)  cpu_isset(cpu, allowed_mask)
 380
 381 #define CPU_TO_PACKAGEINDEX(i) (first_cpu(per_cpu(cpu_sibling_map, i)))
 382
 383 static cpumask_t balance_irq_affinity[NR_IRQS] = {
 384         [0 ... NR_IRQS-1] = CPU_MASK_ALL
 385 };
 386
 387 void set_balance_irq_affinity(unsigned int irq, cpumask_t mask)
 388 {
 389         balance_irq_affinity[irq] = mask;
 390 }
 391
 392 static unsigned long move(int curr_cpu, cpumask_t allowed_mask,
 393                         unsigned long now, int direction)
 394 {
 395         int search_idle = 1;
 396         int cpu = curr_cpu;
 397
 398         goto inside;
 399
 400         do {
 401                 if (unlikely(cpu == curr_cpu))
 402                         search_idle = 0;
 403 inside:
 404                 if (direction == 1) {
 405                         cpu++;
 406                         if (cpu >= NR_CPUS)
 407                                 cpu = 0;
 408                 } else {
 409                         cpu--;
 410                         if (cpu == -1)
 411                                 cpu = NR_CPUS-1;
 412                 }
 413         } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
 414                         (search_idle && !IDLE_ENOUGH(cpu,now)));
 415
 416         return cpu;
 417 }
 418
 419 static inline void balance_irq(int cpu, int irq)
 420 {
 421         unsigned long now = jiffies;
 422         cpumask_t allowed_mask;
 423         unsigned int new_cpu;
 424
 425         if (irqbalance_disabled)
 426                 return;
 427
 428         cpus_and(allowed_mask, cpu_online_map, balance_irq_affinity[irq]);
 429         new_cpu = move(cpu, allowed_mask, now, 1);
 430         if (cpu != new_cpu) {
 431                 set_pending_irq(irq, cpumask_of_cpu(new_cpu));
 432         }
 433 }
 434
 435 static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
 436 {
 437         int i, j;
 438
 439         for_each_online_cpu(i) {
 440                 for (j = 0; j < NR_IRQS; j++) {
 441                         if (!irq_desc[j].action)
 442                                 continue;
 443                         /* Is it a significant load ?  */
 444                         if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) <
 445                                                 useful_load_threshold)
 446                                 continue;
 447                         balance_irq(i, j);
 448                 }
 449         }
 450         balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 451                 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 452         return;
 453 }
 454
 455 static void do_irq_balance(void)
 456 {
 457         int i, j;
 458         unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
 459         unsigned long move_this_load = 0;
 460         int max_loaded = 0, min_loaded = 0;
 461         int load;
 462         unsigned long useful_load_threshold = balanced_irq_interval + 10;
 463         int selected_irq;
 464         int tmp_loaded, first_attempt = 1;
 465         unsigned long tmp_cpu_irq;
 466         unsigned long imbalance = 0;
 467         cpumask_t allowed_mask, target_cpu_mask, tmp;
 468
 469         for_each_possible_cpu(i) {
 470                 int package_index;
 471                 CPU_IRQ(i) = 0;
 472                 if (!cpu_online(i))
 473                         continue;
 474                 package_index = CPU_TO_PACKAGEINDEX(i);
 475                 for (j = 0; j < NR_IRQS; j++) {
 476                         unsigned long value_now, delta;
 477                         /* Is this an active IRQ or balancing disabled ? */
 478                         if (!irq_desc[j].action || irq_balancing_disabled(j))
 479                                 continue;
 480                         if ( package_index == i )
 481                                 IRQ_DELTA(package_index,j) = 0;
 482                         /* Determine the total count per processor per IRQ */
 483                         value_now = (unsigned long) kstat_cpu(i).irqs[j];
 484
 485                         /* Determine the activity per processor per IRQ */
 486                         delta = value_now - LAST_CPU_IRQ(i,j);
 487
 488                         /* Update last_cpu_irq[][] for the next time */
 489                         LAST_CPU_IRQ(i,j) = value_now;
 490
 491                         /* Ignore IRQs whose rate is less than the clock */
 492                         if (delta < useful_load_threshold)
 493                                 continue;
 494                         /* update the load for the processor or package total */
 495                         IRQ_DELTA(package_index,j) += delta;
 496
 497                         /* Keep track of the higher numbered sibling as well */
 498                         if (i != package_index)
 499                                 CPU_IRQ(i) += delta;
 500                         /*
 501                          * We have sibling A and sibling B in the package
 502                          *
 503                          * cpu_irq[A] = load for cpu A + load for cpu B
 504                          * cpu_irq[B] = load for cpu B
 505                          */
 506                         CPU_IRQ(package_index) += delta;
 507                 }
 508         }
 509         /* Find the least loaded processor package */
 510         for_each_online_cpu(i) {
 511                 if (i != CPU_TO_PACKAGEINDEX(i))
 512                         continue;
 513                 if (min_cpu_irq > CPU_IRQ(i)) {
 514                         min_cpu_irq = CPU_IRQ(i);
 515                         min_loaded = i;
 516                 }
 517         }
 518         max_cpu_irq = ULONG_MAX;
 519
 520 tryanothercpu:
 521         /* Look for heaviest loaded processor.
 522          * We may come back to get the next heaviest loaded processor.
 523          * Skip processors with trivial loads.
 524          */
 525         tmp_cpu_irq = 0;
 526         tmp_loaded = -1;
 527         for_each_online_cpu(i) {
 528                 if (i != CPU_TO_PACKAGEINDEX(i))
 529                         continue;
 530                 if (max_cpu_irq <= CPU_IRQ(i))
 531                         continue;
 532                 if (tmp_cpu_irq < CPU_IRQ(i)) {
 533                         tmp_cpu_irq = CPU_IRQ(i);
 534                         tmp_loaded = i;
 535                 }
 536         }
 537
 538         if (tmp_loaded == -1) {
 539          /* In the case of small number of heavy interrupt sources,
 540           * loading some of the cpus too much. We use Ingo's original
 541           * approach to rotate them around.
 542           */
 543                 if (!first_attempt && imbalance >= useful_load_threshold) {
 544                         rotate_irqs_among_cpus(useful_load_threshold);
 545                         return;
 546                 }
 547                 goto not_worth_the_effort;
 548         }
 549
 550         first_attempt = 0;              /* heaviest search */
 551         max_cpu_irq = tmp_cpu_irq;      /* load */
 552         max_loaded = tmp_loaded;        /* processor */
 553         imbalance = (max_cpu_irq - min_cpu_irq) / 2;
 554
 555         /* if imbalance is less than approx 10% of max load, then
 556          * observe diminishing returns action. - quit
 557          */
 558         if (imbalance < (max_cpu_irq >> 3))
 559                 goto not_worth_the_effort;
 560
 561 tryanotherirq:
 562         /* if we select an IRQ to move that can't go where we want, then
 563          * see if there is another one to try.
 564          */
 565         move_this_load = 0;
 566         selected_irq = -1;
 567         for (j = 0; j < NR_IRQS; j++) {
 568                 /* Is this an active IRQ? */
 569                 if (!irq_desc[j].action)
 570                         continue;
 571                 if (imbalance <= IRQ_DELTA(max_loaded,j))
 572                         continue;
 573                 /* Try to find the IRQ that is closest to the imbalance
 574                  * without going over.
 575                  */
 576                 if (move_this_load < IRQ_DELTA(max_loaded,j)) {
 577                         move_this_load = IRQ_DELTA(max_loaded,j);
 578                         selected_irq = j;
 579                 }
 580         }
 581         if (selected_irq == -1) {
 582                 goto tryanothercpu;
 583         }
 584
 585         imbalance = move_this_load;
 586
 587         /* For physical_balance case, we accumulated both load
 588          * values in the one of the siblings cpu_irq[],
 589          * to use the same code for physical and logical processors
 590          * as much as possible.
 591          *
 592          * NOTE: the cpu_irq[] array holds the sum of the load for
 593          * sibling A and sibling B in the slot for the lowest numbered
 594          * sibling (A), _AND_ the load for sibling B in the slot for
 595          * the higher numbered sibling.
 596          *
 597          * We seek the least loaded sibling by making the comparison
 598          * (A+B)/2 vs B
 599          */
 600         load = CPU_IRQ(min_loaded) >> 1;
 601         for_each_cpu_mask(j, per_cpu(cpu_sibling_map, min_loaded)) {
 602                 if (load > CPU_IRQ(j)) {
 603                         /* This won't change cpu_sibling_map[min_loaded] */
 604                         load = CPU_IRQ(j);
 605                         min_loaded = j;
 606                 }
 607         }
 608
 609         cpus_and(allowed_mask,
 610                 cpu_online_map,
 611                 balance_irq_affinity[selected_irq]);
 612         target_cpu_mask = cpumask_of_cpu(min_loaded);
 613         cpus_and(tmp, target_cpu_mask, allowed_mask);
 614
 615         if (!cpus_empty(tmp)) {
 616                 /* mark for change destination */
 617                 set_pending_irq(selected_irq, cpumask_of_cpu(min_loaded));
 618
 619                 /* Since we made a change, come back sooner to
 620                  * check for more variation.
 621                  */
 622                 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 623                         balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 624                 return;
 625         }
 626         goto tryanotherirq;
 627
 628 not_worth_the_effort:
 629         /*
 630          * if we did not find an IRQ to move, then adjust the time interval
 631          * upward
 632          */
 633         balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
 634                 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
 635         return;
 636 }
 637
 638 static int balanced_irq(void *unused)
 639 {
 640         int i;
 641         unsigned long prev_balance_time = jiffies;
 642         long time_remaining = balanced_irq_interval;
 643
 644         /* push everything to CPU 0 to give us a starting point.  */
 645         for (i = 0 ; i < NR_IRQS ; i++) {
 646                 irq_desc[i].pending_mask = cpumask_of_cpu(0);
 647                 set_pending_irq(i, cpumask_of_cpu(0));
 648         }
 649
 650         set_freezable();
 651         for ( ; ; ) {
 652                 time_remaining = schedule_timeout_interruptible(time_remaining);
 653                 try_to_freeze();
 654                 if (time_after(jiffies,
 655                                 prev_balance_time+balanced_irq_interval)) {
 656                         preempt_disable();
 657                         do_irq_balance();
 658                         prev_balance_time = jiffies;
 659                         time_remaining = balanced_irq_interval;
 660                         preempt_enable();
 661                 }
 662         }
 663         return 0;
 664 }
 665
 666 static int __init balanced_irq_init(void)
 667 {
 668         int i;
 669         struct cpuinfo_x86 *c;
 670         cpumask_t tmp;
 671
 672         cpus_shift_right(tmp, cpu_online_map, 2);
 673         c = &boot_cpu_data;
 674         /* When not overwritten by the command line ask subarchitecture. */
 675         if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
 676                 irqbalance_disabled = NO_BALANCE_IRQ;
 677         if (irqbalance_disabled)
 678                 return 0;
 679
 680          /* disable irqbalance completely if there is only one processor online */
 681         if (num_online_cpus() < 2) {
 682                 irqbalance_disabled = 1;
 683                 return 0;
 684         }
 685         /*
 686          * Enable physical balance only if more than 1 physical processor
 687          * is present
 688          */
 689         if (smp_num_siblings > 1 && !cpus_empty(tmp))
 690                 physical_balance = 1;
 691
 692         for_each_online_cpu(i) {
 693                 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 694                 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 695                 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
 696                         printk(KERN_ERR "balanced_irq_init: out of memory");
 697                         goto failed;
 698                 }
 699                 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
 700                 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
 701         }
 702
 703         printk(KERN_INFO "Starting balanced_irq\n");
 704         if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd")))
 705                 return 0;
 706         printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
 707 failed:
 708         for_each_possible_cpu(i) {
 709                 kfree(irq_cpu_data[i].irq_delta);
 710                 irq_cpu_data[i].irq_delta = NULL;
 711                 kfree(irq_cpu_data[i].last_irq);
 712                 irq_cpu_data[i].last_irq = NULL;
 713         }
 714         return 0;
 715 }
 716
 717 int __devinit irqbalance_disable(char *str)
 718 {
 719         irqbalance_disabled = 1;
 720         return 1;
 721 }
 722
 723 __setup("noirqbalance", irqbalance_disable);
 724
 725 late_initcall(balanced_irq_init);
 726 #endif /* CONFIG_IRQBALANCE */
 727 #endif /* CONFIG_SMP */
 728
 729 #ifndef CONFIG_SMP
 730 void fastcall send_IPI_self(int vector)
 731 {
 732         unsigned int cfg;
 733
 734         /*
 735          * Wait for idle.
 736          */
 737         apic_wait_icr_idle();
 738         cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
 739         /*
 740          * Send the IPI. The write to APIC_ICR fires this off.
 741          */
 742         apic_write_around(APIC_ICR, cfg);
 743 }
 744 #endif /* !CONFIG_SMP */
 745
 746
 747 /*
 748  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
 749  * specific CPU-side IRQs.
 750  */
 751
 752 #define MAX_PIRQS 8
 753 static int pirq_entries [MAX_PIRQS];
 754 static int pirqs_enabled;
 755 int skip_ioapic_setup;
 756
 757 static int __init ioapic_pirq_setup(char *str)
 758 {
 759         int i, max;
 760         int ints[MAX_PIRQS+1];
 761
 762         get_options(str, ARRAY_SIZE(ints), ints);
 763
 764         for (i = 0; i < MAX_PIRQS; i++)
 765                 pirq_entries[i] = -1;
 766
 767         pirqs_enabled = 1;
 768         apic_printk(APIC_VERBOSE, KERN_INFO
 769                         "PIRQ redirection, working around broken MP-BIOS.\n");
 770         max = MAX_PIRQS;
 771         if (ints[0] < MAX_PIRQS)
 772                 max = ints[0];
 773
 774         for (i = 0; i < max; i++) {
 775                 apic_printk(APIC_VERBOSE, KERN_DEBUG
 776                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
 777                 /*
 778                  * PIRQs are mapped upside down, usually.
 779                  */
 780                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
 781         }
 782         return 1;
 783 }
 784
 785 __setup("pirq=", ioapic_pirq_setup);
 786
 787 /*
 788  * Find the IRQ entry number of a certain pin.
 789  */
 790 static int find_irq_entry(int apic, int pin, int type)
 791 {
 792         int i;
 793
 794         for (i = 0; i < mp_irq_entries; i++)
 795                 if (mp_irqs[i].mpc_irqtype == type &&
 796                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
 797                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
 798                     mp_irqs[i].mpc_dstirq == pin)
 799                         return i;
 800
 801         return -1;
 802 }
 803
 804 /*
 805  * Find the pin to which IRQ[irq] (ISA) is connected
 806  */
 807 static int __init find_isa_irq_pin(int irq, int type)
 808 {
 809         int i;
 810
 811         for (i = 0; i < mp_irq_entries; i++) {
 812                 int lbus = mp_irqs[i].mpc_srcbus;
 813
 814                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
 815                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
 816                      mp_bus_id_to_type[lbus] == MP_BUS_MCA
 817                     ) &&
 818                     (mp_irqs[i].mpc_irqtype == type) &&
 819                     (mp_irqs[i].mpc_srcbusirq == irq))
 820
 821                         return mp_irqs[i].mpc_dstirq;
 822         }
 823         return -1;
 824 }
 825
 826 static int __init find_isa_irq_apic(int irq, int type)
 827 {
 828         int i;
 829
 830         for (i = 0; i < mp_irq_entries; i++) {
 831                 int lbus = mp_irqs[i].mpc_srcbus;
 832
 833                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
 834                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
 835                      mp_bus_id_to_type[lbus] == MP_BUS_MCA
 836                     ) &&
 837                     (mp_irqs[i].mpc_irqtype == type) &&
 838                     (mp_irqs[i].mpc_srcbusirq == irq))
 839                         break;
 840         }
 841         if (i < mp_irq_entries) {
 842                 int apic;
 843                 for(apic = 0; apic < nr_ioapics; apic++) {
 844                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
 845                                 return apic;
 846                 }
 847         }
 848
 849         return -1;
 850 }
 851
 852 /*
 853  * Find a specific PCI IRQ entry.
 854  * Not an __init, possibly needed by modules
 855  */
 856 static int pin_2_irq(int idx, int apic, int pin);
 857
 858 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
 859 {
 860         int apic, i, best_guess = -1;
 861
 862         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
 863                 "slot:%d, pin:%d.\n", bus, slot, pin);
 864         if (mp_bus_id_to_pci_bus[bus] == -1) {
 865                 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
 866                 return -1;
 867         }
 868         for (i = 0; i < mp_irq_entries; i++) {
 869                 int lbus = mp_irqs[i].mpc_srcbus;
 870
 871                 for (apic = 0; apic < nr_ioapics; apic++)
 872                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
 873                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
 874                                 break;
 875
 876                 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
 877                     !mp_irqs[i].mpc_irqtype &&
 878                     (bus == lbus) &&
 879                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
 880                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
 881
 882                         if (!(apic || IO_APIC_IRQ(irq)))
 883                                 continue;
 884
 885                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
 886                                 return irq;
 887                         /*
 888                          * Use the first all-but-pin matching entry as a
 889                          * best-guess fuzzy result for broken mptables.
 890                          */
 891                         if (best_guess < 0)
 892                                 best_guess = irq;
 893                 }
 894         }
 895         return best_guess;
 896 }
 897 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
 898
 899 /*
 900  * This function currently is only a helper for the i386 smp boot process where
 901  * we need to reprogram the ioredtbls to cater for the cpus which have come online
 902  * so mask in all cases should simply be TARGET_CPUS
 903  */
 904 #ifdef CONFIG_SMP
 905 void __init setup_ioapic_dest(void)
 906 {
 907         int pin, ioapic, irq, irq_entry;
 908
 909         if (skip_ioapic_setup == 1)
 910                 return;
 911
 912         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
 913                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
 914                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
 915                         if (irq_entry == -1)
 916                                 continue;
 917                         irq = pin_2_irq(irq_entry, ioapic, pin);
 918                         set_ioapic_affinity_irq(irq, TARGET_CPUS);
 919                 }
 920
 921         }
 922 }
 923 #endif
 924
 925 /*
 926  * EISA Edge/Level control register, ELCR
 927  */
 928 static int EISA_ELCR(unsigned int irq)
 929 {
 930         if (irq < 16) {
 931                 unsigned int port = 0x4d0 + (irq >> 3);
 932                 return (inb(port) >> (irq & 7)) & 1;
 933         }
 934         apic_printk(APIC_VERBOSE, KERN_INFO
 935                         "Broken MPtable reports ISA irq %d\n", irq);
 936         return 0;
 937 }
 938
 939 /* EISA interrupts are always polarity zero and can be edge or level
 940  * trigger depending on the ELCR value.  If an interrupt is listed as
 941  * EISA conforming in the MP table, that means its trigger type must
 942  * be read in from the ELCR */
 943
 944 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
 945 #define default_EISA_polarity(idx)      (0)
 946
 947 /* ISA interrupts are always polarity zero edge triggered,
 948  * when listed as conforming in the MP table. */
 949
 950 #define default_ISA_trigger(idx)        (0)
 951 #define default_ISA_polarity(idx)       (0)
 952
 953 /* PCI interrupts are always polarity one level triggered,
 954  * when listed as conforming in the MP table. */
 955
 956 #define default_PCI_trigger(idx)        (1)
 957 #define default_PCI_polarity(idx)       (1)
 958
 959 /* MCA interrupts are always polarity zero level triggered,
 960  * when listed as conforming in the MP table. */
 961
 962 #define default_MCA_trigger(idx)        (1)
 963 #define default_MCA_polarity(idx)       (0)
 964
 965 static int MPBIOS_polarity(int idx)
 966 {
 967         int bus = mp_irqs[idx].mpc_srcbus;
 968         int polarity;
 969
 970         /*
 971          * Determine IRQ line polarity (high active or low active):
 972          */
 973         switch (mp_irqs[idx].mpc_irqflag & 3)
 974         {
 975                 case 0: /* conforms, ie. bus-type dependent polarity */
 976                 {
 977                         switch (mp_bus_id_to_type[bus])
 978                         {
 979                                 case MP_BUS_ISA: /* ISA pin */
 980                                 {
 981                                         polarity = default_ISA_polarity(idx);
 982                                         break;
 983                                 }
 984                                 case MP_BUS_EISA: /* EISA pin */
 985                                 {
 986                                         polarity = default_EISA_polarity(idx);
 987                                         break;
 988                                 }
 989                                 case MP_BUS_PCI: /* PCI pin */
 990                                 {
 991                                         polarity = default_PCI_polarity(idx);
 992                                         break;
 993                                 }
 994                                 case MP_BUS_MCA: /* MCA pin */
 995                                 {
 996                                         polarity = default_MCA_polarity(idx);
 997                                         break;
 998                                 }
 999                                 default:
1000                                 {
1001                                         printk(KERN_WARNING "broken BIOS!!\n");
1002                                         polarity = 1;
1003                                         break;
1004                                 }
1005                         }
1006                         break;
1007                 }
1008                 case 1: /* high active */
1009                 {
1010                         polarity = 0;
1011                         break;
1012                 }
1013                 case 2: /* reserved */
1014                 {
1015                         printk(KERN_WARNING "broken BIOS!!\n");
1016                         polarity = 1;
1017                         break;
1018                 }
1019                 case 3: /* low active */
1020                 {
1021                         polarity = 1;
1022                         break;
1023                 }
1024                 default: /* invalid */
1025                 {
1026                         printk(KERN_WARNING "broken BIOS!!\n");
1027                         polarity = 1;
1028                         break;
1029                 }
1030         }
1031         return polarity;
1032 }
1033
1034 static int MPBIOS_trigger(int idx)
1035 {
1036         int bus = mp_irqs[idx].mpc_srcbus;
1037         int trigger;
1038
1039         /*
1040          * Determine IRQ trigger mode (edge or level sensitive):
1041          */
1042         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
1043         {
1044                 case 0: /* conforms, ie. bus-type dependent */
1045                 {
1046                         switch (mp_bus_id_to_type[bus])
1047                         {
1048                                 case MP_BUS_ISA: /* ISA pin */
1049                                 {
1050                                         trigger = default_ISA_trigger(idx);
1051                                         break;
1052                                 }
1053                                 case MP_BUS_EISA: /* EISA pin */
1054                                 {
1055                                         trigger = default_EISA_trigger(idx);
1056                                         break;
1057                                 }
1058                                 case MP_BUS_PCI: /* PCI pin */
1059                                 {
1060                                         trigger = default_PCI_trigger(idx);
1061                                         break;
1062                                 }
1063                                 case MP_BUS_MCA: /* MCA pin */
1064                                 {
1065                                         trigger = default_MCA_trigger(idx);
1066                                         break;
1067                                 }
1068                                 default:
1069                                 {
1070                                         printk(KERN_WARNING "broken BIOS!!\n");
1071                                         trigger = 1;
1072                                         break;
1073                                 }
1074                         }
1075                         break;
1076                 }
1077                 case 1: /* edge */
1078                 {
1079                         trigger = 0;
1080                         break;
1081                 }
1082                 case 2: /* reserved */
1083                 {
1084                         printk(KERN_WARNING "broken BIOS!!\n");
1085                         trigger = 1;
1086                         break;
1087                 }
1088                 case 3: /* level */
1089                 {
1090                         trigger = 1;
1091                         break;
1092                 }
1093                 default: /* invalid */
1094                 {
1095                         printk(KERN_WARNING "broken BIOS!!\n");
1096                         trigger = 0;
1097                         break;
1098                 }
1099         }
1100         return trigger;
1101 }
1102
1103 static inline int irq_polarity(int idx)
1104 {
1105         return MPBIOS_polarity(idx);
1106 }
1107
1108 static inline int irq_trigger(int idx)
1109 {
1110         return MPBIOS_trigger(idx);
1111 }
1112
1113 static int pin_2_irq(int idx, int apic, int pin)
1114 {
1115         int irq, i;
1116         int bus = mp_irqs[idx].mpc_srcbus;
1117
1118         /*
1119          * Debugging check, we are in big trouble if this message pops up!
1120          */
1121         if (mp_irqs[idx].mpc_dstirq != pin)
1122                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1123
1124         switch (mp_bus_id_to_type[bus])
1125         {
1126                 case MP_BUS_ISA: /* ISA pin */
1127                 case MP_BUS_EISA:
1128                 case MP_BUS_MCA:
1129                 {
1130                         irq = mp_irqs[idx].mpc_srcbusirq;
1131                         break;
1132                 }
1133                 case MP_BUS_PCI: /* PCI pin */
1134                 {
1135                         /*
1136                          * PCI IRQs are mapped in order
1137                          */
1138                         i = irq = 0;
1139                         while (i < apic)
1140                                 irq += nr_ioapic_registers[i++];
1141                         irq += pin;
1142
1143                         /*
1144                          * For MPS mode, so far only needed by ES7000 platform
1145                          */
1146                         if (ioapic_renumber_irq)
1147                                 irq = ioapic_renumber_irq(apic, irq);
1148
1149                         break;
1150                 }
1151                 default:
1152                 {
1153                         printk(KERN_ERR "unknown bus type %d.\n",bus);
1154                         irq = 0;
1155                         break;
1156                 }
1157         }
1158
1159         /*
1160          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1161          */
1162         if ((pin >= 16) && (pin <= 23)) {
1163                 if (pirq_entries[pin-16] != -1) {
1164                         if (!pirq_entries[pin-16]) {
1165                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1166                                                 "disabling PIRQ%d\n", pin-16);
1167                         } else {
1168                                 irq = pirq_entries[pin-16];
1169                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1170                                                 "using PIRQ%d -> IRQ %d\n",
1171                                                 pin-16, irq);
1172                         }
1173                 }
1174         }
1175         return irq;
1176 }
1177
1178 static inline int IO_APIC_irq_trigger(int irq)
1179 {
1180         int apic, idx, pin;
1181
1182         for (apic = 0; apic < nr_ioapics; apic++) {
1183                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1184                         idx = find_irq_entry(apic,pin,mp_INT);
1185                         if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1186                                 return irq_trigger(idx);
1187                 }
1188         }
1189         /*
1190          * nonexistent IRQs are edge default
1191          */
1192         return 0;
1193 }
1194
1195 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1196 static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 };
1197
1198 static int __assign_irq_vector(int irq)
1199 {
1200         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1201         int vector, offset;
1202
1203         BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
1204
1205         if (irq_vector[irq] > 0)
1206                 return irq_vector[irq];
1207
1208         vector = current_vector;
1209         offset = current_offset;
1210 next:
1211         vector += 8;
1212         if (vector >= FIRST_SYSTEM_VECTOR) {
1213                 offset = (offset + 1) % 8;
1214                 vector = FIRST_DEVICE_VECTOR + offset;
1215         }
1216         if (vector == current_vector)
1217                 return -ENOSPC;
1218         if (test_and_set_bit(vector, used_vectors))
1219                 goto next;
1220
1221         current_vector = vector;
1222         current_offset = offset;
1223         irq_vector[irq] = vector;
1224
1225         return vector;
1226 }
1227
1228 static int assign_irq_vector(int irq)
1229 {
1230         unsigned long flags;
1231         int vector;
1232
1233         spin_lock_irqsave(&vector_lock, flags);
1234         vector = __assign_irq_vector(irq);
1235         spin_unlock_irqrestore(&vector_lock, flags);
1236
1237         return vector;
1238 }
1239 static struct irq_chip ioapic_chip;
1240
1241 #define IOAPIC_AUTO     -1
1242 #define IOAPIC_EDGE     0
1243 #define IOAPIC_LEVEL    1
1244
1245 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1246 {
1247         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1248             trigger == IOAPIC_LEVEL) {
1249                 irq_desc[irq].status |= IRQ_LEVEL;
1250                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1251                                          handle_fasteoi_irq, "fasteoi");
1252         } else {
1253                 irq_desc[irq].status &= ~IRQ_LEVEL;
1254                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1255                                          handle_edge_irq, "edge");
1256         }
1257         set_intr_gate(vector, interrupt[irq]);
1258 }
1259
1260 static void __init setup_IO_APIC_irqs(void)
1261 {
1262         struct IO_APIC_route_entry entry;
1263         int apic, pin, idx, irq, first_notcon = 1, vector;
1264         unsigned long flags;
1265
1266         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1267
1268         for (apic = 0; apic < nr_ioapics; apic++) {
1269         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1270
1271                 /*
1272                  * add it to the IO-APIC irq-routing table:
1273                  */
1274                 memset(&entry,0,sizeof(entry));
1275
1276                 entry.delivery_mode = INT_DELIVERY_MODE;
1277                 entry.dest_mode = INT_DEST_MODE;
1278                 entry.mask = 0;                         /* enable IRQ */
1279                 entry.dest.logical.logical_dest =
1280                                         cpu_mask_to_apicid(TARGET_CPUS);
1281
1282                 idx = find_irq_entry(apic,pin,mp_INT);
1283                 if (idx == -1) {
1284                         if (first_notcon) {
1285                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1286                                                 " IO-APIC (apicid-pin) %d-%d",
1287                                                 mp_ioapics[apic].mpc_apicid,
1288                                                 pin);
1289                                 first_notcon = 0;
1290                         } else
1291                                 apic_printk(APIC_VERBOSE, ", %d-%d",
1292                                         mp_ioapics[apic].mpc_apicid, pin);
1293                         continue;
1294                 }
1295
1296                 if (!first_notcon) {
1297                         apic_printk(APIC_VERBOSE, " not connected.\n");
1298                         first_notcon = 1;
1299                 }
1300
1301                 entry.trigger = irq_trigger(idx);
1302                 entry.polarity = irq_polarity(idx);
1303
1304                 if (irq_trigger(idx)) {
1305                         entry.trigger = 1;
1306                         entry.mask = 1;
1307                 }
1308
1309                 irq = pin_2_irq(idx, apic, pin);
1310                 /*
1311                  * skip adding the timer int on secondary nodes, which causes
1312                  * a small but painful rift in the time-space continuum
1313                  */
1314                 if (multi_timer_check(apic, irq))
1315                         continue;
1316                 else
1317                         add_pin_to_irq(irq, apic, pin);
1318
1319                 if (!apic && !IO_APIC_IRQ(irq))
1320                         continue;
1321
1322                 if (IO_APIC_IRQ(irq)) {
1323                         vector = assign_irq_vector(irq);
1324                         entry.vector = vector;
1325                         ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1326
1327                         if (!apic && (irq < 16))
1328                                 disable_8259A_irq(irq);
1329                 }
1330                 spin_lock_irqsave(&ioapic_lock, flags);
1331                 __ioapic_write_entry(apic, pin, entry);
1332                 spin_unlock_irqrestore(&ioapic_lock, flags);
1333         }
1334         }
1335
1336         if (!first_notcon)
1337                 apic_printk(APIC_VERBOSE, " not connected.\n");
1338 }
1339
1340 /*
1341  * Set up the 8259A-master output pin:
1342  */
1343 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
1344 {
1345         struct IO_APIC_route_entry entry;
1346
1347         memset(&entry,0,sizeof(entry));
1348
1349         disable_8259A_irq(0);
1350
1351         /* mask LVT0 */
1352         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1353
1354         /*
1355          * We use logical delivery to get the timer IRQ
1356          * to the first CPU.
1357          */
1358         entry.dest_mode = INT_DEST_MODE;
1359         entry.mask = 0;                                 /* unmask IRQ now */
1360         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1361         entry.delivery_mode = INT_DELIVERY_MODE;
1362         entry.polarity = 0;
1363         entry.trigger = 0;
1364         entry.vector = vector;
1365
1366         /*
1367          * The timer IRQ doesn't have to know that behind the
1368          * scene we have a 8259A-master in AEOI mode ...
1369          */
1370         irq_desc[0].chip = &ioapic_chip;
1371         set_irq_handler(0, handle_edge_irq);
1372
1373         /*
1374          * Add it to the IO-APIC irq-routing table:
1375          */
1376         ioapic_write_entry(apic, pin, entry);
1377
1378         enable_8259A_irq(0);
1379 }
1380
1381 void __init print_IO_APIC(void)
1382 {
1383         int apic, i;
1384         union IO_APIC_reg_00 reg_00;
1385         union IO_APIC_reg_01 reg_01;
1386         union IO_APIC_reg_02 reg_02;
1387         union IO_APIC_reg_03 reg_03;
1388         unsigned long flags;
1389
1390         if (apic_verbosity == APIC_QUIET)
1391                 return;
1392
1393         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1394         for (i = 0; i < nr_ioapics; i++)
1395                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1396                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1397
1398         /*
1399          * We are a bit conservative about what we expect.  We have to
1400          * know about every hardware change ASAP.
1401          */
1402         printk(KERN_INFO "testing the IO APIC.......................\n");
1403
1404         for (apic = 0; apic < nr_ioapics; apic++) {
1405
1406         spin_lock_irqsave(&ioapic_lock, flags);
1407         reg_00.raw = io_apic_read(apic, 0);
1408         reg_01.raw = io_apic_read(apic, 1);
1409         if (reg_01.bits.version >= 0x10)
1410                 reg_02.raw = io_apic_read(apic, 2);
1411         if (reg_01.bits.version >= 0x20)
1412                 reg_03.raw = io_apic_read(apic, 3);
1413         spin_unlock_irqrestore(&ioapic_lock, flags);
1414
1415         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1416         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1417         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1418         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1419         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1420
1421         printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1422         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1423
1424         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1425         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1426
1427         /*
1428          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1429          * but the value of reg_02 is read as the previous read register
1430          * value, so ignore it if reg_02 == reg_01.
1431          */
1432         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1433                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1434                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1435         }
1436
1437         /*
1438          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1439          * or reg_03, but the value of reg_0[23] is read as the previous read
1440          * register value, so ignore it if reg_03 == reg_0[12].
1441          */
1442         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1443             reg_03.raw != reg_01.raw) {
1444                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1445                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1446         }
1447
1448         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1449
1450         printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1451                           " Stat Dest Deli Vect:   \n");
1452
1453         for (i = 0; i <= reg_01.bits.entries; i++) {
1454                 struct IO_APIC_route_entry entry;
1455
1456                 entry = ioapic_read_entry(apic, i);
1457
1458                 printk(KERN_DEBUG " %02x %03X %02X  ",
1459                         i,
1460                         entry.dest.logical.logical_dest,
1461                         entry.dest.physical.physical_dest
1462                 );
1463
1464                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1465                         entry.mask,
1466                         entry.trigger,
1467                         entry.irr,
1468                         entry.polarity,
1469                         entry.delivery_status,
1470                         entry.dest_mode,
1471                         entry.delivery_mode,
1472                         entry.vector
1473                 );
1474         }
1475         }
1476         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1477         for (i = 0; i < NR_IRQS; i++) {
1478                 struct irq_pin_list *entry = irq_2_pin + i;
1479                 if (entry->pin < 0)
1480                         continue;
1481                 printk(KERN_DEBUG "IRQ%d ", i);
1482                 for (;;) {
1483                         printk("-> %d:%d", entry->apic, entry->pin);
1484                         if (!entry->next)
1485                                 break;
1486                         entry = irq_2_pin + entry->next;
1487                 }
1488                 printk("\n");
1489         }
1490
1491         printk(KERN_INFO ".................................... done.\n");
1492
1493         return;
1494 }
1495
1496 #if 0
1497
1498 static void print_APIC_bitfield (int base)
1499 {
1500         unsigned int v;
1501         int i, j;
1502
1503         if (apic_verbosity == APIC_QUIET)
1504                 return;
1505
1506         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1507         for (i = 0; i < 8; i++) {
1508                 v = apic_read(base + i*0x10);
1509                 for (j = 0; j < 32; j++) {
1510                         if (v & (1<<j))
1511                                 printk("1");
1512                         else
1513                                 printk("0");
1514                 }
1515                 printk("\n");
1516         }
1517 }
1518
1519 void /*__init*/ print_local_APIC(void * dummy)
1520 {
1521         unsigned int v, ver, maxlvt;
1522
1523         if (apic_verbosity == APIC_QUIET)
1524                 return;
1525
1526         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1527                 smp_processor_id(), hard_smp_processor_id());
1528         v = apic_read(APIC_ID);
1529         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, GET_APIC_ID(v));
1530         v = apic_read(APIC_LVR);
1531         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1532         ver = GET_APIC_VERSION(v);
1533         maxlvt = lapic_get_maxlvt();
1534
1535         v = apic_read(APIC_TASKPRI);
1536         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1537
1538         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1539                 v = apic_read(APIC_ARBPRI);
1540                 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1541                         v & APIC_ARBPRI_MASK);
1542                 v = apic_read(APIC_PROCPRI);
1543                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1544         }
1545
1546         v = apic_read(APIC_EOI);
1547         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1548         v = apic_read(APIC_RRR);
1549         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1550         v = apic_read(APIC_LDR);
1551         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1552         v = apic_read(APIC_DFR);
1553         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1554         v = apic_read(APIC_SPIV);
1555         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1556
1557         printk(KERN_DEBUG "... APIC ISR field:\n");
1558         print_APIC_bitfield(APIC_ISR);
1559         printk(KERN_DEBUG "... APIC TMR field:\n");
1560         print_APIC_bitfield(APIC_TMR);
1561         printk(KERN_DEBUG "... APIC IRR field:\n");
1562         print_APIC_bitfield(APIC_IRR);
1563
1564         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1565                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1566                         apic_write(APIC_ESR, 0);
1567                 v = apic_read(APIC_ESR);
1568                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1569         }
1570
1571         v = apic_read(APIC_ICR);
1572         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1573         v = apic_read(APIC_ICR2);
1574         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1575
1576         v = apic_read(APIC_LVTT);
1577         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1578
1579         if (maxlvt > 3) {                       /* PC is LVT#4. */
1580                 v = apic_read(APIC_LVTPC);
1581                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1582         }
1583         v = apic_read(APIC_LVT0);
1584         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1585         v = apic_read(APIC_LVT1);
1586         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1587
1588         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1589                 v = apic_read(APIC_LVTERR);
1590                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1591         }
1592
1593         v = apic_read(APIC_TMICT);
1594         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1595         v = apic_read(APIC_TMCCT);
1596         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1597         v = apic_read(APIC_TDCR);
1598         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1599         printk("\n");
1600 }
1601
1602 void print_all_local_APICs (void)
1603 {
1604         on_each_cpu(print_local_APIC, NULL, 1, 1);
1605 }
1606
1607 void /*__init*/ print_PIC(void)
1608 {
1609         unsigned int v;
1610         unsigned long flags;
1611
1612         if (apic_verbosity == APIC_QUIET)
1613                 return;
1614
1615         printk(KERN_DEBUG "\nprinting PIC contents\n");
1616
1617         spin_lock_irqsave(&i8259A_lock, flags);
1618
1619         v = inb(0xa1) << 8 | inb(0x21);
1620         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1621
1622         v = inb(0xa0) << 8 | inb(0x20);
1623         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1624
1625         outb(0x0b,0xa0);
1626         outb(0x0b,0x20);
1627         v = inb(0xa0) << 8 | inb(0x20);
1628         outb(0x0a,0xa0);
1629         outb(0x0a,0x20);
1630
1631         spin_unlock_irqrestore(&i8259A_lock, flags);
1632
1633         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1634
1635         v = inb(0x4d1) << 8 | inb(0x4d0);
1636         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1637 }
1638
1639 #endif  /*  0  */
1640
1641 static void __init enable_IO_APIC(void)
1642 {
1643         union IO_APIC_reg_01 reg_01;
1644         int i8259_apic, i8259_pin;
1645         int i, apic;
1646         unsigned long flags;
1647
1648         for (i = 0; i < PIN_MAP_SIZE; i++) {
1649                 irq_2_pin[i].pin = -1;
1650                 irq_2_pin[i].next = 0;
1651         }
1652         if (!pirqs_enabled)
1653                 for (i = 0; i < MAX_PIRQS; i++)
1654                         pirq_entries[i] = -1;
1655
1656         /*
1657          * The number of IO-APIC IRQ registers (== #pins):
1658          */
1659         for (apic = 0; apic < nr_ioapics; apic++) {
1660                 spin_lock_irqsave(&ioapic_lock, flags);
1661                 reg_01.raw = io_apic_read(apic, 1);
1662                 spin_unlock_irqrestore(&ioapic_lock, flags);
1663                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1664         }
1665         for(apic = 0; apic < nr_ioapics; apic++) {
1666                 int pin;
1667                 /* See if any of the pins is in ExtINT mode */
1668                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1669                         struct IO_APIC_route_entry entry;
1670                         entry = ioapic_read_entry(apic, pin);
1671
1672
1673                         /* If the interrupt line is enabled and in ExtInt mode
1674                          * I have found the pin where the i8259 is connected.
1675                          */
1676                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1677                                 ioapic_i8259.apic = apic;
1678                                 ioapic_i8259.pin  = pin;
1679                                 goto found_i8259;
1680                         }
1681                 }
1682         }
1683  found_i8259:
1684         /* Look to see what if the MP table has reported the ExtINT */
1685         /* If we could not find the appropriate pin by looking at the ioapic
1686          * the i8259 probably is not connected the ioapic but give the
1687          * mptable a chance anyway.
1688          */
1689         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1690         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1691         /* Trust the MP table if nothing is setup in the hardware */
1692         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1693                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1694                 ioapic_i8259.pin  = i8259_pin;
1695                 ioapic_i8259.apic = i8259_apic;
1696         }
1697         /* Complain if the MP table and the hardware disagree */
1698         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1699                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1700         {
1701                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1702         }
1703
1704         /*
1705          * Do not trust the IO-APIC being empty at bootup
1706          */
1707         clear_IO_APIC();
1708 }
1709
1710 /*
1711  * Not an __init, needed by the reboot code
1712  */
1713 void disable_IO_APIC(void)
1714 {
1715         /*
1716          * Clear the IO-APIC before rebooting:
1717          */
1718         clear_IO_APIC();
1719
1720         /*
1721          * If the i8259 is routed through an IOAPIC
1722          * Put that IOAPIC in virtual wire mode
1723          * so legacy interrupts can be delivered.
1724          */
1725         if (ioapic_i8259.pin != -1) {
1726                 struct IO_APIC_route_entry entry;
1727
1728                 memset(&entry, 0, sizeof(entry));
1729                 entry.mask            = 0; /* Enabled */
1730                 entry.trigger         = 0; /* Edge */
1731                 entry.irr             = 0;
1732                 entry.polarity        = 0; /* High */
1733                 entry.delivery_status = 0;
1734                 entry.dest_mode       = 0; /* Physical */
1735                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1736                 entry.vector          = 0;
1737                 entry.dest.physical.physical_dest =
1738                                         GET_APIC_ID(apic_read(APIC_ID));
1739
1740                 /*
1741                  * Add it to the IO-APIC irq-routing table:
1742                  */
1743                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1744         }
1745         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1746 }
1747
1748 /*
1749  * function to set the IO-APIC physical IDs based on the
1750  * values stored in the MPC table.
1751  *
1752  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
1753  */
1754
1755 #ifndef CONFIG_X86_NUMAQ
1756 static void __init setup_ioapic_ids_from_mpc(void)
1757 {
1758         union IO_APIC_reg_00 reg_00;
1759         physid_mask_t phys_id_present_map;
1760         int apic;
1761         int i;
1762         unsigned char old_id;
1763         unsigned long flags;
1764
1765         /*
1766          * Don't check I/O APIC IDs for xAPIC systems.  They have
1767          * no meaning without the serial APIC bus.
1768          */
1769         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1770                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
1771                 return;
1772         /*
1773          * This is broken; anything with a real cpu count has to
1774          * circumvent this idiocy regardless.
1775          */
1776         phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1777
1778         /*
1779          * Set the IOAPIC ID to the value stored in the MPC table.
1780          */
1781         for (apic = 0; apic < nr_ioapics; apic++) {
1782
1783                 /* Read the register 0 value */
1784                 spin_lock_irqsave(&ioapic_lock, flags);
1785                 reg_00.raw = io_apic_read(apic, 0);
1786                 spin_unlock_irqrestore(&ioapic_lock, flags);
1787
1788                 old_id = mp_ioapics[apic].mpc_apicid;
1789
1790                 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) {
1791                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1792                                 apic, mp_ioapics[apic].mpc_apicid);
1793                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1794                                 reg_00.bits.ID);
1795                         mp_ioapics[apic].mpc_apicid = reg_00.bits.ID;
1796                 }
1797
1798                 /*
1799                  * Sanity check, is the ID really free? Every APIC in a
1800                  * system must have a unique ID or we get lots of nice
1801                  * 'stuck on smp_invalidate_needed IPI wait' messages.
1802                  */
1803                 if (check_apicid_used(phys_id_present_map,
1804                                         mp_ioapics[apic].mpc_apicid)) {
1805                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1806                                 apic, mp_ioapics[apic].mpc_apicid);
1807                         for (i = 0; i < get_physical_broadcast(); i++)
1808                                 if (!physid_isset(i, phys_id_present_map))
1809                                         break;
1810                         if (i >= get_physical_broadcast())
1811                                 panic("Max APIC ID exceeded!\n");
1812                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1813                                 i);
1814                         physid_set(i, phys_id_present_map);
1815                         mp_ioapics[apic].mpc_apicid = i;
1816                 } else {
1817                         physid_mask_t tmp;
1818                         tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid);
1819                         apic_printk(APIC_VERBOSE, "Setting %d in the "
1820                                         "phys_id_present_map\n",
1821                                         mp_ioapics[apic].mpc_apicid);
1822                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
1823                 }
1824
1825
1826                 /*
1827                  * We need to adjust the IRQ routing table
1828                  * if the ID changed.
1829                  */
1830                 if (old_id != mp_ioapics[apic].mpc_apicid)
1831                         for (i = 0; i < mp_irq_entries; i++)
1832                                 if (mp_irqs[i].mpc_dstapic == old_id)
1833                                         mp_irqs[i].mpc_dstapic
1834                                                 = mp_ioapics[apic].mpc_apicid;
1835
1836                 /*
1837                  * Read the right value from the MPC table and
1838                  * write it into the ID register.
1839                  */
1840                 apic_printk(APIC_VERBOSE, KERN_INFO
1841                         "...changing IO-APIC physical APIC ID to %d ...",
1842                         mp_ioapics[apic].mpc_apicid);
1843
1844                 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1845                 spin_lock_irqsave(&ioapic_lock, flags);
1846                 io_apic_write(apic, 0, reg_00.raw);
1847                 spin_unlock_irqrestore(&ioapic_lock, flags);
1848
1849                 /*
1850                  * Sanity check
1851                  */
1852                 spin_lock_irqsave(&ioapic_lock, flags);
1853                 reg_00.raw = io_apic_read(apic, 0);
1854                 spin_unlock_irqrestore(&ioapic_lock, flags);
1855                 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1856                         printk("could not set ID!\n");
1857                 else
1858                         apic_printk(APIC_VERBOSE, " ok.\n");
1859         }
1860 }
1861 #else
1862 static void __init setup_ioapic_ids_from_mpc(void) { }
1863 #endif
1864
1865 int no_timer_check __initdata;
1866
1867 static int __init notimercheck(char *s)
1868 {
1869         no_timer_check = 1;
1870         return 1;
1871 }
1872 __setup("no_timer_check", notimercheck);
1873
1874 /*
1875  * There is a nasty bug in some older SMP boards, their mptable lies
1876  * about the timer IRQ. We do the following to work around the situation:
1877  *
1878  *      - timer IRQ defaults to IO-APIC IRQ
1879  *      - if this function detects that timer IRQs are defunct, then we fall
1880  *        back to ISA timer IRQs
1881  */
1882 static int __init timer_irq_works(void)
1883 {
1884         unsigned long t1 = jiffies;
1885
1886         if (no_timer_check)
1887                 return 1;
1888
1889         local_irq_enable();
1890         /* Let ten ticks pass... */
1891         mdelay((10 * 1000) / HZ);
1892
1893         /*
1894          * Expect a few ticks at least, to be sure some possible
1895          * glue logic does not lock up after one or two first
1896          * ticks in a non-ExtINT mode.  Also the local APIC
1897          * might have cached one ExtINT interrupt.  Finally, at
1898          * least one tick may be lost due to delays.
1899          */
1900         if (jiffies - t1 > 4)
1901                 return 1;
1902
1903         return 0;
1904 }
1905
1906 /*
1907  * In the SMP+IOAPIC case it might happen that there are an unspecified
1908  * number of pending IRQ events unhandled. These cases are very rare,
1909  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1910  * better to do it this way as thus we do not have to be aware of
1911  * 'pending' interrupts in the IRQ path, except at this point.
1912  */
1913 /*
1914  * Edge triggered needs to resend any interrupt
1915  * that was delayed but this is now handled in the device
1916  * independent code.
1917  */
1918
1919 /*
1920  * Startup quirk:
1921  *
1922  * Starting up a edge-triggered IO-APIC interrupt is
1923  * nasty - we need to make sure that we get the edge.
1924  * If it is already asserted for some reason, we need
1925  * return 1 to indicate that is was pending.
1926  *
1927  * This is not complete - we should be able to fake
1928  * an edge even if it isn't on the 8259A...
1929  *
1930  * (We do this for level-triggered IRQs too - it cannot hurt.)
1931  */
1932 static unsigned int startup_ioapic_irq(unsigned int irq)
1933 {
1934         int was_pending = 0;
1935         unsigned long flags;
1936
1937         spin_lock_irqsave(&ioapic_lock, flags);
1938         if (irq < 16) {
1939                 disable_8259A_irq(irq);
1940                 if (i8259A_irq_pending(irq))
1941                         was_pending = 1;
1942         }
1943         __unmask_IO_APIC_irq(irq);
1944         spin_unlock_irqrestore(&ioapic_lock, flags);
1945
1946         return was_pending;
1947 }
1948
1949 static void ack_ioapic_irq(unsigned int irq)
1950 {
1951         move_native_irq(irq);
1952         ack_APIC_irq();
1953 }
1954
1955 static void ack_ioapic_quirk_irq(unsigned int irq)
1956 {
1957         unsigned long v;
1958         int i;
1959
1960         move_native_irq(irq);
1961 /*
1962  * It appears there is an erratum which affects at least version 0x11
1963  * of I/O APIC (that's the 82093AA and cores integrated into various
1964  * chipsets).  Under certain conditions a level-triggered interrupt is
1965  * erroneously delivered as edge-triggered one but the respective IRR
1966  * bit gets set nevertheless.  As a result the I/O unit expects an EOI
1967  * message but it will never arrive and further interrupts are blocked
1968  * from the source.  The exact reason is so far unknown, but the
1969  * phenomenon was observed when two consecutive interrupt requests
1970  * from a given source get delivered to the same CPU and the source is
1971  * temporarily disabled in between.
1972  *
1973  * A workaround is to simulate an EOI message manually.  We achieve it
1974  * by setting the trigger mode to edge and then to level when the edge
1975  * trigger mode gets detected in the TMR of a local APIC for a
1976  * level-triggered interrupt.  We mask the source for the time of the
1977  * operation to prevent an edge-triggered interrupt escaping meanwhile.
1978  * The idea is from Manfred Spraul.  --macro
1979  */
1980         i = irq_vector[irq];
1981
1982         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1983
1984         ack_APIC_irq();
1985
1986         if (!(v & (1 << (i & 0x1f)))) {
1987                 atomic_inc(&irq_mis_count);
1988                 spin_lock(&ioapic_lock);
1989                 __mask_and_edge_IO_APIC_irq(irq);
1990                 __unmask_and_level_IO_APIC_irq(irq);
1991                 spin_unlock(&ioapic_lock);
1992         }
1993 }
1994
1995 static int ioapic_retrigger_irq(unsigned int irq)
1996 {
1997         send_IPI_self(irq_vector[irq]);
1998
1999         return 1;
2000 }
2001
2002 static struct irq_chip ioapic_chip __read_mostly = {
2003         .name           = "IO-APIC",
2004         .startup        = startup_ioapic_irq,
2005         .mask           = mask_IO_APIC_irq,
2006         .unmask         = unmask_IO_APIC_irq,
2007         .ack            = ack_ioapic_irq,
2008         .eoi            = ack_ioapic_quirk_irq,
2009 #ifdef CONFIG_SMP
2010         .set_affinity   = set_ioapic_affinity_irq,
2011 #endif
2012         .retrigger      = ioapic_retrigger_irq,
2013 };
2014
2015
2016 static inline void init_IO_APIC_traps(void)
2017 {
2018         int irq;
2019
2020         /*
2021          * NOTE! The local APIC isn't very good at handling
2022          * multiple interrupts at the same interrupt level.
2023          * As the interrupt level is determined by taking the
2024          * vector number and shifting that right by 4, we
2025          * want to spread these out a bit so that they don't
2026          * all fall in the same interrupt level.
2027          *
2028          * Also, we've got to be careful not to trash gate
2029          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2030          */
2031         for (irq = 0; irq < NR_IRQS ; irq++) {
2032                 int tmp = irq;
2033                 if (IO_APIC_IRQ(tmp) && !irq_vector[tmp]) {
2034                         /*
2035                          * Hmm.. We don't have an entry for this,
2036                          * so default to an old-fashioned 8259
2037                          * interrupt if we can..
2038                          */
2039                         if (irq < 16)
2040                                 make_8259A_irq(irq);
2041                         else
2042                                 /* Strange. Oh, well.. */
2043                                 irq_desc[irq].chip = &no_irq_chip;
2044                 }
2045         }
2046 }
2047
2048 /*
2049  * The local APIC irq-chip implementation:
2050  */
2051
2052 static void ack_apic(unsigned int irq)
2053 {
2054         ack_APIC_irq();
2055 }
2056
2057 static void mask_lapic_irq (unsigned int irq)
2058 {
2059         unsigned long v;
2060
2061         v = apic_read(APIC_LVT0);
2062         apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2063 }
2064
2065 static void unmask_lapic_irq (unsigned int irq)
2066 {
2067         unsigned long v;
2068
2069         v = apic_read(APIC_LVT0);
2070         apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2071 }
2072
2073 static struct irq_chip lapic_chip __read_mostly = {
2074         .name           = "local-APIC-edge",
2075         .mask           = mask_lapic_irq,
2076         .unmask         = unmask_lapic_irq,
2077         .eoi            = ack_apic,
2078 };
2079
2080 static void setup_nmi (void)
2081 {
2082         /*
2083          * Dirty trick to enable the NMI watchdog ...
2084          * We put the 8259A master into AEOI mode and
2085          * unmask on all local APICs LVT0 as NMI.
2086          *
2087          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2088          * is from Maciej W. Rozycki - so we do not have to EOI from
2089          * the NMI handler or the timer interrupt.
2090          */
2091         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2092
2093         on_each_cpu(enable_NMI_through_LVT0, NULL, 1, 1);
2094
2095         apic_printk(APIC_VERBOSE, " done.\n");
2096 }
2097
2098 /*
2099  * This looks a bit hackish but it's about the only one way of sending
2100  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2101  * not support the ExtINT mode, unfortunately.  We need to send these
2102  * cycles as some i82489DX-based boards have glue logic that keeps the
2103  * 8259A interrupt line asserted until INTA.  --macro
2104  */
2105 static inline void unlock_ExtINT_logic(void)
2106 {
2107         int apic, pin, i;
2108         struct IO_APIC_route_entry entry0, entry1;
2109         unsigned char save_control, save_freq_select;
2110
2111         pin  = find_isa_irq_pin(8, mp_INT);
2112         if (pin == -1) {
2113                 WARN_ON_ONCE(1);
2114                 return;
2115         }
2116         apic = find_isa_irq_apic(8, mp_INT);
2117         if (apic == -1) {
2118                 WARN_ON_ONCE(1);
2119                 return;
2120         }
2121
2122         entry0 = ioapic_read_entry(apic, pin);
2123         clear_IO_APIC_pin(apic, pin);
2124
2125         memset(&entry1, 0, sizeof(entry1));
2126
2127         entry1.dest_mode = 0;                   /* physical delivery */
2128         entry1.mask = 0;                        /* unmask IRQ now */
2129         entry1.dest.physical.physical_dest = hard_smp_processor_id();
2130         entry1.delivery_mode = dest_ExtINT;
2131         entry1.polarity = entry0.polarity;
2132         entry1.trigger = 0;
2133         entry1.vector = 0;
2134
2135         ioapic_write_entry(apic, pin, entry1);
2136
2137         save_control = CMOS_READ(RTC_CONTROL);
2138         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2139         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2140                    RTC_FREQ_SELECT);
2141         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2142
2143         i = 100;
2144         while (i-- > 0) {
2145                 mdelay(10);
2146                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2147                         i -= 10;
2148         }
2149
2150         CMOS_WRITE(save_control, RTC_CONTROL);
2151         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2152         clear_IO_APIC_pin(apic, pin);
2153
2154         ioapic_write_entry(apic, pin, entry0);
2155 }
2156
2157 int timer_uses_ioapic_pin_0;
2158
2159 /*
2160  * This code may look a bit paranoid, but it's supposed to cooperate with
2161  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2162  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2163  * fanatically on his truly buggy board.
2164  */
2165 static inline void __init check_timer(void)
2166 {
2167         int apic1, pin1, apic2, pin2;
2168         int vector;
2169
2170         /*
2171          * get/set the timer IRQ vector:
2172          */
2173         disable_8259A_irq(0);
2174         vector = assign_irq_vector(0);
2175         set_intr_gate(vector, interrupt[0]);
2176
2177         /*
2178          * Subtle, code in do_timer_interrupt() expects an AEOI
2179          * mode for the 8259A whenever interrupts are routed
2180          * through I/O APICs.  Also IRQ0 has to be enabled in
2181          * the 8259A which implies the virtual wire has to be
2182          * disabled in the local APIC.
2183          */
2184         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2185         init_8259A(1);
2186         timer_ack = 1;
2187         if (timer_over_8254 > 0)
2188                 enable_8259A_irq(0);
2189
2190         pin1  = find_isa_irq_pin(0, mp_INT);
2191         apic1 = find_isa_irq_apic(0, mp_INT);
2192         pin2  = ioapic_i8259.pin;
2193         apic2 = ioapic_i8259.apic;
2194
2195         if (pin1 == 0)
2196                 timer_uses_ioapic_pin_0 = 1;
2197
2198         printk(KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
2199                 vector, apic1, pin1, apic2, pin2);
2200
2201         if (pin1 != -1) {
2202                 /*
2203                  * Ok, does IRQ0 through the IOAPIC work?
2204                  */
2205                 unmask_IO_APIC_irq(0);
2206                 if (timer_irq_works()) {
2207                         if (nmi_watchdog == NMI_IO_APIC) {
2208                                 disable_8259A_irq(0);
2209                                 setup_nmi();
2210                                 enable_8259A_irq(0);
2211                         }
2212                         if (disable_timer_pin_1 > 0)
2213                                 clear_IO_APIC_pin(0, pin1);
2214                         return;
2215                 }
2216                 clear_IO_APIC_pin(apic1, pin1);
2217                 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to "
2218                                 "IO-APIC\n");
2219         }
2220
2221         printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2222         if (pin2 != -1) {
2223                 printk("\n..... (found pin %d) ...", pin2);
2224                 /*
2225                  * legacy devices should be connected to IO APIC #0
2226                  */
2227                 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
2228                 if (timer_irq_works()) {
2229                         printk("works.\n");
2230                         if (pin1 != -1)
2231                                 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2232                         else
2233                                 add_pin_to_irq(0, apic2, pin2);
2234                         if (nmi_watchdog == NMI_IO_APIC) {
2235                                 setup_nmi();
2236                         }
2237                         return;
2238                 }
2239                 /*
2240                  * Cleanup, just in case ...
2241                  */
2242                 clear_IO_APIC_pin(apic2, pin2);
2243         }
2244         printk(" failed.\n");
2245
2246         if (nmi_watchdog == NMI_IO_APIC) {
2247                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2248                 nmi_watchdog = 0;
2249         }
2250
2251         printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2252
2253         disable_8259A_irq(0);
2254         set_irq_chip_and_handler_name(0, &lapic_chip, handle_fasteoi_irq,
2255                                       "fasteoi");
2256         apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector);   /* Fixed mode */
2257         enable_8259A_irq(0);
2258
2259         if (timer_irq_works()) {
2260                 printk(" works.\n");
2261                 return;
2262         }
2263         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2264         printk(" failed.\n");
2265
2266         printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2267
2268         timer_ack = 0;
2269         init_8259A(0);
2270         make_8259A_irq(0);
2271         apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2272
2273         unlock_ExtINT_logic();
2274
2275         if (timer_irq_works()) {
2276                 printk(" works.\n");
2277                 return;
2278         }
2279         printk(" failed :(.\n");
2280         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
2281                 "report.  Then try booting with the 'noapic' option");
2282 }
2283
2284 /*
2285  *
2286  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2287  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2288  *   Linux doesn't really care, as it's not actually used
2289  *   for any interrupt handling anyway.
2290  */
2291 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
2292
2293 void __init setup_IO_APIC(void)
2294 {
2295         int i;
2296
2297         /* Reserve all the system vectors. */
2298         for (i = FIRST_SYSTEM_VECTOR; i < NR_VECTORS; i++)
2299                 set_bit(i, used_vectors);
2300
2301         enable_IO_APIC();
2302
2303         if (acpi_ioapic)
2304                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
2305         else
2306                 io_apic_irqs = ~PIC_IRQS;
2307
2308         printk("ENABLING IO-APIC IRQs\n");
2309
2310         /*
2311          * Set up IO-APIC IRQ routing.
2312          */
2313         if (!acpi_ioapic)
2314                 setup_ioapic_ids_from_mpc();
2315         sync_Arb_IDs();
2316         setup_IO_APIC_irqs();
2317         init_IO_APIC_traps();
2318         check_timer();
2319         if (!acpi_ioapic)
2320                 print_IO_APIC();
2321 }
2322
2323 static int __init setup_disable_8254_timer(char *s)
2324 {
2325         timer_over_8254 = -1;
2326         return 1;
2327 }
2328 static int __init setup_enable_8254_timer(char *s)
2329 {
2330         timer_over_8254 = 2;
2331         return 1;
2332 }
2333
2334 __setup("disable_8254_timer", setup_disable_8254_timer);
2335 __setup("enable_8254_timer", setup_enable_8254_timer);
2336
2337 /*
2338  *      Called after all the initialization is done. If we didnt find any
2339  *      APIC bugs then we can allow the modify fast path
2340  */
2341
2342 static int __init io_apic_bug_finalize(void)
2343 {
2344         if(sis_apic_bug == -1)
2345                 sis_apic_bug = 0;
2346         return 0;
2347 }
2348
2349 late_initcall(io_apic_bug_finalize);
2350
2351 struct sysfs_ioapic_data {
2352         struct sys_device dev;
2353         struct IO_APIC_route_entry entry[0];
2354 };
2355 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2356
2357 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2358 {
2359         struct IO_APIC_route_entry *entry;
2360         struct sysfs_ioapic_data *data;
2361         int i;
2362
2363         data = container_of(dev, struct sysfs_ioapic_data, dev);
2364         entry = data->entry;
2365         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2366                 entry[i] = ioapic_read_entry(dev->id, i);
2367
2368         return 0;
2369 }
2370
2371 static int ioapic_resume(struct sys_device *dev)
2372 {
2373         struct IO_APIC_route_entry *entry;
2374         struct sysfs_ioapic_data *data;
2375         unsigned long flags;
2376         union IO_APIC_reg_00 reg_00;
2377         int i;
2378
2379         data = container_of(dev, struct sysfs_ioapic_data, dev);
2380         entry = data->entry;
2381
2382         spin_lock_irqsave(&ioapic_lock, flags);
2383         reg_00.raw = io_apic_read(dev->id, 0);
2384         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2385                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2386                 io_apic_write(dev->id, 0, reg_00.raw);
2387         }
2388         spin_unlock_irqrestore(&ioapic_lock, flags);
2389         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2390                 ioapic_write_entry(dev->id, i, entry[i]);
2391
2392         return 0;
2393 }
2394
2395 static struct sysdev_class ioapic_sysdev_class = {
2396         set_kset_name("ioapic"),
2397         .suspend = ioapic_suspend,
2398         .resume = ioapic_resume,
2399 };
2400
2401 static int __init ioapic_init_sysfs(void)
2402 {
2403         struct sys_device * dev;
2404         int i, size, error = 0;
2405
2406         error = sysdev_class_register(&ioapic_sysdev_class);
2407         if (error)
2408                 return error;
2409
2410         for (i = 0; i < nr_ioapics; i++ ) {
2411                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2412                         * sizeof(struct IO_APIC_route_entry);
2413                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2414                 if (!mp_ioapic_data[i]) {
2415                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2416                         continue;
2417                 }
2418                 memset(mp_ioapic_data[i], 0, size);
2419                 dev = &mp_ioapic_data[i]->dev;
2420                 dev->id = i;
2421                 dev->cls = &ioapic_sysdev_class;
2422                 error = sysdev_register(dev);
2423                 if (error) {
2424                         kfree(mp_ioapic_data[i]);
2425                         mp_ioapic_data[i] = NULL;
2426                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2427                         continue;
2428                 }
2429         }
2430
2431         return 0;
2432 }
2433
2434 device_initcall(ioapic_init_sysfs);
2435
2436 /*
2437  * Dynamic irq allocate and deallocation
2438  */
2439 int create_irq(void)
2440 {
2441         /* Allocate an unused irq */
2442         int irq, new, vector = 0;
2443         unsigned long flags;
2444
2445         irq = -ENOSPC;
2446         spin_lock_irqsave(&vector_lock, flags);
2447         for (new = (NR_IRQS - 1); new >= 0; new--) {
2448                 if (platform_legacy_irq(new))
2449                         continue;
2450                 if (irq_vector[new] != 0)
2451                         continue;
2452                 vector = __assign_irq_vector(new);
2453                 if (likely(vector > 0))
2454                         irq = new;
2455                 break;
2456         }
2457         spin_unlock_irqrestore(&vector_lock, flags);
2458
2459         if (irq >= 0) {
2460                 set_intr_gate(vector, interrupt[irq]);
2461                 dynamic_irq_init(irq);
2462         }
2463         return irq;
2464 }
2465
2466 void destroy_irq(unsigned int irq)
2467 {
2468         unsigned long flags;
2469
2470         dynamic_irq_cleanup(irq);
2471
2472         spin_lock_irqsave(&vector_lock, flags);
2473         irq_vector[irq] = 0;
2474         spin_unlock_irqrestore(&vector_lock, flags);
2475 }
2476
2477 /*
2478  * MSI message composition
2479  */
2480 #ifdef CONFIG_PCI_MSI
2481 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2482 {
2483         int vector;
2484         unsigned dest;
2485
2486         vector = assign_irq_vector(irq);
2487         if (vector >= 0) {
2488                 dest = cpu_mask_to_apicid(TARGET_CPUS);
2489
2490                 msg->address_hi = MSI_ADDR_BASE_HI;
2491                 msg->address_lo =
2492                         MSI_ADDR_BASE_LO |
2493                         ((INT_DEST_MODE == 0) ?
2494                                 MSI_ADDR_DEST_MODE_PHYSICAL:
2495                                 MSI_ADDR_DEST_MODE_LOGICAL) |
2496                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2497                                 MSI_ADDR_REDIRECTION_CPU:
2498                                 MSI_ADDR_REDIRECTION_LOWPRI) |
2499                         MSI_ADDR_DEST_ID(dest);
2500
2501                 msg->data =
2502                         MSI_DATA_TRIGGER_EDGE |
2503                         MSI_DATA_LEVEL_ASSERT |
2504                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2505                                 MSI_DATA_DELIVERY_FIXED:
2506                                 MSI_DATA_DELIVERY_LOWPRI) |
2507                         MSI_DATA_VECTOR(vector);
2508         }
2509         return vector;
2510 }
2511
2512 #ifdef CONFIG_SMP
2513 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2514 {
2515         struct msi_msg msg;
2516         unsigned int dest;
2517         cpumask_t tmp;
2518         int vector;
2519
2520         cpus_and(tmp, mask, cpu_online_map);
2521         if (cpus_empty(tmp))
2522                 tmp = TARGET_CPUS;
2523
2524         vector = assign_irq_vector(irq);
2525         if (vector < 0)
2526                 return;
2527
2528         dest = cpu_mask_to_apicid(mask);
2529
2530         read_msi_msg(irq, &msg);
2531
2532         msg.data &= ~MSI_DATA_VECTOR_MASK;
2533         msg.data |= MSI_DATA_VECTOR(vector);
2534         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2535         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2536
2537         write_msi_msg(irq, &msg);
2538         irq_desc[irq].affinity = mask;
2539 }
2540 #endif /* CONFIG_SMP */
2541
2542 /*
2543  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2544  * which implement the MSI or MSI-X Capability Structure.
2545  */
2546 static struct irq_chip msi_chip = {
2547         .name           = "PCI-MSI",
2548         .unmask         = unmask_msi_irq,
2549         .mask           = mask_msi_irq,
2550         .ack            = ack_ioapic_irq,
2551 #ifdef CONFIG_SMP
2552         .set_affinity   = set_msi_irq_affinity,
2553 #endif
2554         .retrigger      = ioapic_retrigger_irq,
2555 };
2556
2557 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2558 {
2559         struct msi_msg msg;
2560         int irq, ret;
2561         irq = create_irq();
2562         if (irq < 0)
2563                 return irq;
2564
2565         ret = msi_compose_msg(dev, irq, &msg);
2566         if (ret < 0) {
2567                 destroy_irq(irq);
2568                 return ret;
2569         }
2570
2571         set_irq_msi(irq, desc);
2572         write_msi_msg(irq, &msg);
2573
2574         set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq,
2575                                       "edge");
2576
2577         return 0;
2578 }
2579
2580 void arch_teardown_msi_irq(unsigned int irq)
2581 {
2582         destroy_irq(irq);
2583 }
2584
2585 #endif /* CONFIG_PCI_MSI */
2586
2587 /*
2588  * Hypertransport interrupt support
2589  */
2590 #ifdef CONFIG_HT_IRQ
2591
2592 #ifdef CONFIG_SMP
2593
2594 static void target_ht_irq(unsigned int irq, unsigned int dest)
2595 {
2596         struct ht_irq_msg msg;
2597         fetch_ht_irq_msg(irq, &msg);
2598
2599         msg.address_lo &= ~(HT_IRQ_LOW_DEST_ID_MASK);
2600         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2601
2602         msg.address_lo |= HT_IRQ_LOW_DEST_ID(dest);
2603         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2604
2605         write_ht_irq_msg(irq, &msg);
2606 }
2607
2608 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2609 {
2610         unsigned int dest;
2611         cpumask_t tmp;
2612
2613         cpus_and(tmp, mask, cpu_online_map);
2614         if (cpus_empty(tmp))
2615                 tmp = TARGET_CPUS;
2616
2617         cpus_and(mask, tmp, CPU_MASK_ALL);
2618
2619         dest = cpu_mask_to_apicid(mask);
2620
2621         target_ht_irq(irq, dest);
2622         irq_desc[irq].affinity = mask;
2623 }
2624 #endif
2625
2626 static struct irq_chip ht_irq_chip = {
2627         .name           = "PCI-HT",
2628         .mask           = mask_ht_irq,
2629         .unmask         = unmask_ht_irq,
2630         .ack            = ack_ioapic_irq,
2631 #ifdef CONFIG_SMP
2632         .set_affinity   = set_ht_irq_affinity,
2633 #endif
2634         .retrigger      = ioapic_retrigger_irq,
2635 };
2636
2637 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2638 {
2639         int vector;
2640
2641         vector = assign_irq_vector(irq);
2642         if (vector >= 0) {
2643                 struct ht_irq_msg msg;
2644                 unsigned dest;
2645                 cpumask_t tmp;
2646
2647                 cpus_clear(tmp);
2648                 cpu_set(vector >> 8, tmp);
2649                 dest = cpu_mask_to_apicid(tmp);
2650
2651                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2652
2653                 msg.address_lo =
2654                         HT_IRQ_LOW_BASE |
2655                         HT_IRQ_LOW_DEST_ID(dest) |
2656                         HT_IRQ_LOW_VECTOR(vector) |
2657                         ((INT_DEST_MODE == 0) ?
2658                                 HT_IRQ_LOW_DM_PHYSICAL :
2659                                 HT_IRQ_LOW_DM_LOGICAL) |
2660                         HT_IRQ_LOW_RQEOI_EDGE |
2661                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2662                                 HT_IRQ_LOW_MT_FIXED :
2663                                 HT_IRQ_LOW_MT_ARBITRATED) |
2664                         HT_IRQ_LOW_IRQ_MASKED;
2665
2666                 write_ht_irq_msg(irq, &msg);
2667
2668                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2669                                               handle_edge_irq, "edge");
2670         }
2671         return vector;
2672 }
2673 #endif /* CONFIG_HT_IRQ */
2674
2675 /* --------------------------------------------------------------------------
2676                           ACPI-based IOAPIC Configuration
2677    -------------------------------------------------------------------------- */
2678
2679 #ifdef CONFIG_ACPI
2680
2681 int __init io_apic_get_unique_id (int ioapic, int apic_id)
2682 {
2683         union IO_APIC_reg_00 reg_00;
2684         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2685         physid_mask_t tmp;
2686         unsigned long flags;
2687         int i = 0;
2688
2689         /*
2690          * The P4 platform supports up to 256 APIC IDs on two separate APIC
2691          * buses (one for LAPICs, one for IOAPICs), where predecessors only
2692          * supports up to 16 on one shared APIC bus.
2693          *
2694          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2695          *      advantage of new APIC bus architecture.
2696          */
2697
2698         if (physids_empty(apic_id_map))
2699                 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2700
2701         spin_lock_irqsave(&ioapic_lock, flags);
2702         reg_00.raw = io_apic_read(ioapic, 0);
2703         spin_unlock_irqrestore(&ioapic_lock, flags);
2704
2705         if (apic_id >= get_physical_broadcast()) {
2706                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2707                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
2708                 apic_id = reg_00.bits.ID;
2709         }
2710
2711         /*
2712          * Every APIC in a system must have a unique ID or we get lots of nice
2713          * 'stuck on smp_invalidate_needed IPI wait' messages.
2714          */
2715         if (check_apicid_used(apic_id_map, apic_id)) {
2716
2717                 for (i = 0; i < get_physical_broadcast(); i++) {
2718                         if (!check_apicid_used(apic_id_map, i))
2719                                 break;
2720                 }
2721
2722                 if (i == get_physical_broadcast())
2723                         panic("Max apic_id exceeded!\n");
2724
2725                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2726                         "trying %d\n", ioapic, apic_id, i);
2727
2728                 apic_id = i;
2729         }
2730
2731         tmp = apicid_to_cpu_present(apic_id);
2732         physids_or(apic_id_map, apic_id_map, tmp);
2733
2734         if (reg_00.bits.ID != apic_id) {
2735                 reg_00.bits.ID = apic_id;
2736
2737                 spin_lock_irqsave(&ioapic_lock, flags);
2738                 io_apic_write(ioapic, 0, reg_00.raw);
2739                 reg_00.raw = io_apic_read(ioapic, 0);
2740                 spin_unlock_irqrestore(&ioapic_lock, flags);
2741
2742                 /* Sanity check */
2743                 if (reg_00.bits.ID != apic_id) {
2744                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
2745                         return -1;
2746                 }
2747         }
2748
2749         apic_printk(APIC_VERBOSE, KERN_INFO
2750                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2751
2752         return apic_id;
2753 }
2754
2755
2756 int __init io_apic_get_version (int ioapic)
2757 {
2758         union IO_APIC_reg_01    reg_01;
2759         unsigned long flags;
2760
2761         spin_lock_irqsave(&ioapic_lock, flags);
2762         reg_01.raw = io_apic_read(ioapic, 1);
2763         spin_unlock_irqrestore(&ioapic_lock, flags);
2764
2765         return reg_01.bits.version;
2766 }
2767
2768
2769 int __init io_apic_get_redir_entries (int ioapic)
2770 {
2771         union IO_APIC_reg_01    reg_01;
2772         unsigned long flags;
2773
2774         spin_lock_irqsave(&ioapic_lock, flags);
2775         reg_01.raw = io_apic_read(ioapic, 1);
2776         spin_unlock_irqrestore(&ioapic_lock, flags);
2777
2778         return reg_01.bits.entries;
2779 }
2780
2781
2782 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2783 {
2784         struct IO_APIC_route_entry entry;
2785         unsigned long flags;
2786
2787         if (!IO_APIC_IRQ(irq)) {
2788                 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2789                         ioapic);
2790                 return -EINVAL;
2791         }
2792
2793         /*
2794          * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2795          * Note that we mask (disable) IRQs now -- these get enabled when the
2796          * corresponding device driver registers for this IRQ.
2797          */
2798
2799         memset(&entry,0,sizeof(entry));
2800
2801         entry.delivery_mode = INT_DELIVERY_MODE;
2802         entry.dest_mode = INT_DEST_MODE;
2803         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2804         entry.trigger = edge_level;
2805         entry.polarity = active_high_low;
2806         entry.mask  = 1;
2807
2808         /*
2809          * IRQs < 16 are already in the irq_2_pin[] map
2810          */
2811         if (irq >= 16)
2812                 add_pin_to_irq(irq, ioapic, pin);
2813
2814         entry.vector = assign_irq_vector(irq);
2815
2816         apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2817                 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2818                 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2819                 edge_level, active_high_low);
2820
2821         ioapic_register_intr(irq, entry.vector, edge_level);
2822
2823         if (!ioapic && (irq < 16))
2824                 disable_8259A_irq(irq);
2825
2826         spin_lock_irqsave(&ioapic_lock, flags);
2827         __ioapic_write_entry(ioapic, pin, entry);
2828         spin_unlock_irqrestore(&ioapic_lock, flags);
2829
2830         return 0;
2831 }
2832
2833 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2834 {
2835         int i;
2836
2837         if (skip_ioapic_setup)
2838                 return -1;
2839
2840         for (i = 0; i < mp_irq_entries; i++)
2841                 if (mp_irqs[i].mpc_irqtype == mp_INT &&
2842                     mp_irqs[i].mpc_srcbusirq == bus_irq)
2843                         break;
2844         if (i >= mp_irq_entries)
2845                 return -1;
2846
2847         *trigger = irq_trigger(i);
2848         *polarity = irq_polarity(i);
2849         return 0;
2850 }
2851
2852 #endif /* CONFIG_ACPI */
2853
2854 static int __init parse_disable_timer_pin_1(char *arg)
2855 {
2856         disable_timer_pin_1 = 1;
2857         return 0;
2858 }
2859 early_param("disable_timer_pin_1", parse_disable_timer_pin_1);
2860
2861 static int __init parse_enable_timer_pin_1(char *arg)
2862 {
2863         disable_timer_pin_1 = -1;
2864         return 0;
2865 }
2866 early_param("enable_timer_pin_1", parse_enable_timer_pin_1);
2867
2868 static int __init parse_noapic(char *arg)
2869 {
2870         /* disable IO-APIC */
2871         disable_ioapic_setup();
2872         return 0;
2873 }
2874 early_param("noapic", parse_noapic);