err.no Git - linux-2.6/blob - arch/i386/kernel/io_apic.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/smp_lock.h>
  29 #include <linux/mc146818rtc.h>
  30 #include <linux/compiler.h>
  31 #include <linux/acpi.h>
  32 #include <linux/module.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/pci.h>
  35
  36 #include <asm/io.h>
  37 #include <asm/smp.h>
  38 #include <asm/desc.h>
  39 #include <asm/timer.h>
  40 #include <asm/i8259.h>
  41 #include <asm/nmi.h>
  42 #include <asm/msidef.h>
  43 #include <asm/hypertransport.h>
  44
  45 #include <mach_apic.h>
  46 #include <mach_apicdef.h>
  47
  48 #include "io_ports.h"
  49
  50 int (*ioapic_renumber_irq)(int ioapic, int irq);
  51 atomic_t irq_mis_count;
  52
  53 /* Where if anywhere is the i8259 connect in external int mode */
  54 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
  55
  56 static DEFINE_SPINLOCK(ioapic_lock);
  57 static DEFINE_SPINLOCK(vector_lock);
  58
  59 int timer_over_8254 __initdata = 1;
  60
  61 /*
  62  *      Is the SiS APIC rmw bug present ?
  63  *      -1 = don't know, 0 = no, 1 = yes
  64  */
  65 int sis_apic_bug = -1;
  66
  67 /*
  68  * # of IRQ routing registers
  69  */
  70 int nr_ioapic_registers[MAX_IO_APICS];
  71
  72 static int disable_timer_pin_1 __initdata;
  73
  74 /*
  75  * Rough estimation of how many shared IRQs there are, can
  76  * be changed anytime.
  77  */
  78 #define MAX_PLUS_SHARED_IRQS NR_IRQS
  79 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
  80
  81 /*
  82  * This is performance-critical, we want to do it O(1)
  83  *
  84  * the indexing order of this array favors 1:1 mappings
  85  * between pins and IRQs.
  86  */
  87
  88 static struct irq_pin_list {
  89         int apic, pin, next;
  90 } irq_2_pin[PIN_MAP_SIZE];
  91
  92 union entry_union {
  93         struct { u32 w1, w2; };
  94         struct IO_APIC_route_entry entry;
  95 };
  96
  97 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
  98 {
  99         union entry_union eu;
 100         unsigned long flags;
 101         spin_lock_irqsave(&ioapic_lock, flags);
 102         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
 103         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
 104         spin_unlock_irqrestore(&ioapic_lock, flags);
 105         return eu.entry;
 106 }
 107
 108 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 109 {
 110         unsigned long flags;
 111         union entry_union eu;
 112         eu.entry = e;
 113         spin_lock_irqsave(&ioapic_lock, flags);
 114         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 115         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 116         spin_unlock_irqrestore(&ioapic_lock, flags);
 117 }
 118
 119 /*
 120  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
 121  * shared ISA-space IRQs, so we have to support them. We are super
 122  * fast in the common case, and fast for shared ISA-space IRQs.
 123  */
 124 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 125 {
 126         static int first_free_entry = NR_IRQS;
 127         struct irq_pin_list *entry = irq_2_pin + irq;
 128
 129         while (entry->next)
 130                 entry = irq_2_pin + entry->next;
 131
 132         if (entry->pin != -1) {
 133                 entry->next = first_free_entry;
 134                 entry = irq_2_pin + entry->next;
 135                 if (++first_free_entry >= PIN_MAP_SIZE)
 136                         panic("io_apic.c: whoops");
 137         }
 138         entry->apic = apic;
 139         entry->pin = pin;
 140 }
 141
 142 /*
 143  * Reroute an IRQ to a different pin.
 144  */
 145 static void __init replace_pin_at_irq(unsigned int irq,
 146                                       int oldapic, int oldpin,
 147                                       int newapic, int newpin)
 148 {
 149         struct irq_pin_list *entry = irq_2_pin + irq;
 150
 151         while (1) {
 152                 if (entry->apic == oldapic && entry->pin == oldpin) {
 153                         entry->apic = newapic;
 154                         entry->pin = newpin;
 155                 }
 156                 if (!entry->next)
 157                         break;
 158                 entry = irq_2_pin + entry->next;
 159         }
 160 }
 161
 162 static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable)
 163 {
 164         struct irq_pin_list *entry = irq_2_pin + irq;
 165         unsigned int pin, reg;
 166
 167         for (;;) {
 168                 pin = entry->pin;
 169                 if (pin == -1)
 170                         break;
 171                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
 172                 reg &= ~disable;
 173                 reg |= enable;
 174                 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
 175                 if (!entry->next)
 176                         break;
 177                 entry = irq_2_pin + entry->next;
 178         }
 179 }
 180
 181 /* mask = 1 */
 182 static void __mask_IO_APIC_irq (unsigned int irq)
 183 {
 184         __modify_IO_APIC_irq(irq, 0x00010000, 0);
 185 }
 186
 187 /* mask = 0 */
 188 static void __unmask_IO_APIC_irq (unsigned int irq)
 189 {
 190         __modify_IO_APIC_irq(irq, 0, 0x00010000);
 191 }
 192
 193 /* mask = 1, trigger = 0 */
 194 static void __mask_and_edge_IO_APIC_irq (unsigned int irq)
 195 {
 196         __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000);
 197 }
 198
 199 /* mask = 0, trigger = 1 */
 200 static void __unmask_and_level_IO_APIC_irq (unsigned int irq)
 201 {
 202         __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000);
 203 }
 204
 205 static void mask_IO_APIC_irq (unsigned int irq)
 206 {
 207         unsigned long flags;
 208
 209         spin_lock_irqsave(&ioapic_lock, flags);
 210         __mask_IO_APIC_irq(irq);
 211         spin_unlock_irqrestore(&ioapic_lock, flags);
 212 }
 213
 214 static void unmask_IO_APIC_irq (unsigned int irq)
 215 {
 216         unsigned long flags;
 217
 218         spin_lock_irqsave(&ioapic_lock, flags);
 219         __unmask_IO_APIC_irq(irq);
 220         spin_unlock_irqrestore(&ioapic_lock, flags);
 221 }
 222
 223 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 224 {
 225         struct IO_APIC_route_entry entry;
 226
 227         /* Check delivery_mode to be sure we're not clearing an SMI pin */
 228         entry = ioapic_read_entry(apic, pin);
 229         if (entry.delivery_mode == dest_SMI)
 230                 return;
 231
 232         /*
 233          * Disable it in the IO-APIC irq-routing table:
 234          */
 235         memset(&entry, 0, sizeof(entry));
 236         entry.mask = 1;
 237         ioapic_write_entry(apic, pin, entry);
 238 }
 239
 240 static void clear_IO_APIC (void)
 241 {
 242         int apic, pin;
 243
 244         for (apic = 0; apic < nr_ioapics; apic++)
 245                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 246                         clear_IO_APIC_pin(apic, pin);
 247 }
 248
 249 #ifdef CONFIG_SMP
 250 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
 251 {
 252         unsigned long flags;
 253         int pin;
 254         struct irq_pin_list *entry = irq_2_pin + irq;
 255         unsigned int apicid_value;
 256         cpumask_t tmp;
 257
 258         cpus_and(tmp, cpumask, cpu_online_map);
 259         if (cpus_empty(tmp))
 260                 tmp = TARGET_CPUS;
 261
 262         cpus_and(cpumask, tmp, CPU_MASK_ALL);
 263
 264         apicid_value = cpu_mask_to_apicid(cpumask);
 265         /* Prepare to do the io_apic_write */
 266         apicid_value = apicid_value << 24;
 267         spin_lock_irqsave(&ioapic_lock, flags);
 268         for (;;) {
 269                 pin = entry->pin;
 270                 if (pin == -1)
 271                         break;
 272                 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
 273                 if (!entry->next)
 274                         break;
 275                 entry = irq_2_pin + entry->next;
 276         }
 277         set_native_irq_info(irq, cpumask);
 278         spin_unlock_irqrestore(&ioapic_lock, flags);
 279 }
 280
 281 #if defined(CONFIG_IRQBALANCE)
 282 # include <asm/processor.h>     /* kernel_thread() */
 283 # include <linux/kernel_stat.h> /* kstat */
 284 # include <linux/slab.h>                /* kmalloc() */
 285 # include <linux/timer.h>       /* time_after() */
 286
 287 #ifdef CONFIG_BALANCED_IRQ_DEBUG
 288 #  define TDprintk(x...) do { printk("<%ld:%s:%d>: ", jiffies, __FILE__, __LINE__); printk(x); } while (0)
 289 #  define Dprintk(x...) do { TDprintk(x); } while (0)
 290 # else
 291 #  define TDprintk(x...)
 292 #  define Dprintk(x...)
 293 # endif
 294
 295 #define IRQBALANCE_CHECK_ARCH -999
 296 #define MAX_BALANCED_IRQ_INTERVAL       (5*HZ)
 297 #define MIN_BALANCED_IRQ_INTERVAL       (HZ/2)
 298 #define BALANCED_IRQ_MORE_DELTA         (HZ/10)
 299 #define BALANCED_IRQ_LESS_DELTA         (HZ)
 300
 301 static int irqbalance_disabled __read_mostly = IRQBALANCE_CHECK_ARCH;
 302 static int physical_balance __read_mostly;
 303 static long balanced_irq_interval __read_mostly = MAX_BALANCED_IRQ_INTERVAL;
 304
 305 static struct irq_cpu_info {
 306         unsigned long * last_irq;
 307         unsigned long * irq_delta;
 308         unsigned long irq;
 309 } irq_cpu_data[NR_CPUS];
 310
 311 #define CPU_IRQ(cpu)            (irq_cpu_data[cpu].irq)
 312 #define LAST_CPU_IRQ(cpu,irq)   (irq_cpu_data[cpu].last_irq[irq])
 313 #define IRQ_DELTA(cpu,irq)      (irq_cpu_data[cpu].irq_delta[irq])
 314
 315 #define IDLE_ENOUGH(cpu,now) \
 316         (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
 317
 318 #define IRQ_ALLOWED(cpu, allowed_mask)  cpu_isset(cpu, allowed_mask)
 319
 320 #define CPU_TO_PACKAGEINDEX(i) (first_cpu(cpu_sibling_map[i]))
 321
 322 static cpumask_t balance_irq_affinity[NR_IRQS] = {
 323         [0 ... NR_IRQS-1] = CPU_MASK_ALL
 324 };
 325
 326 void set_balance_irq_affinity(unsigned int irq, cpumask_t mask)
 327 {
 328         balance_irq_affinity[irq] = mask;
 329 }
 330
 331 static unsigned long move(int curr_cpu, cpumask_t allowed_mask,
 332                         unsigned long now, int direction)
 333 {
 334         int search_idle = 1;
 335         int cpu = curr_cpu;
 336
 337         goto inside;
 338
 339         do {
 340                 if (unlikely(cpu == curr_cpu))
 341                         search_idle = 0;
 342 inside:
 343                 if (direction == 1) {
 344                         cpu++;
 345                         if (cpu >= NR_CPUS)
 346                                 cpu = 0;
 347                 } else {
 348                         cpu--;
 349                         if (cpu == -1)
 350                                 cpu = NR_CPUS-1;
 351                 }
 352         } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
 353                         (search_idle && !IDLE_ENOUGH(cpu,now)));
 354
 355         return cpu;
 356 }
 357
 358 static inline void balance_irq(int cpu, int irq)
 359 {
 360         unsigned long now = jiffies;
 361         cpumask_t allowed_mask;
 362         unsigned int new_cpu;
 363
 364         if (irqbalance_disabled)
 365                 return;
 366
 367         cpus_and(allowed_mask, cpu_online_map, balance_irq_affinity[irq]);
 368         new_cpu = move(cpu, allowed_mask, now, 1);
 369         if (cpu != new_cpu) {
 370                 set_pending_irq(irq, cpumask_of_cpu(new_cpu));
 371         }
 372 }
 373
 374 static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
 375 {
 376         int i, j;
 377         Dprintk("Rotating IRQs among CPUs.\n");
 378         for_each_online_cpu(i) {
 379                 for (j = 0; j < NR_IRQS; j++) {
 380                         if (!irq_desc[j].action)
 381                                 continue;
 382                         /* Is it a significant load ?  */
 383                         if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) <
 384                                                 useful_load_threshold)
 385                                 continue;
 386                         balance_irq(i, j);
 387                 }
 388         }
 389         balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 390                 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 391         return;
 392 }
 393
 394 static void do_irq_balance(void)
 395 {
 396         int i, j;
 397         unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
 398         unsigned long move_this_load = 0;
 399         int max_loaded = 0, min_loaded = 0;
 400         int load;
 401         unsigned long useful_load_threshold = balanced_irq_interval + 10;
 402         int selected_irq;
 403         int tmp_loaded, first_attempt = 1;
 404         unsigned long tmp_cpu_irq;
 405         unsigned long imbalance = 0;
 406         cpumask_t allowed_mask, target_cpu_mask, tmp;
 407
 408         for_each_possible_cpu(i) {
 409                 int package_index;
 410                 CPU_IRQ(i) = 0;
 411                 if (!cpu_online(i))
 412                         continue;
 413                 package_index = CPU_TO_PACKAGEINDEX(i);
 414                 for (j = 0; j < NR_IRQS; j++) {
 415                         unsigned long value_now, delta;
 416                         /* Is this an active IRQ? */
 417                         if (!irq_desc[j].action)
 418                                 continue;
 419                         if ( package_index == i )
 420                                 IRQ_DELTA(package_index,j) = 0;
 421                         /* Determine the total count per processor per IRQ */
 422                         value_now = (unsigned long) kstat_cpu(i).irqs[j];
 423
 424                         /* Determine the activity per processor per IRQ */
 425                         delta = value_now - LAST_CPU_IRQ(i,j);
 426
 427                         /* Update last_cpu_irq[][] for the next time */
 428                         LAST_CPU_IRQ(i,j) = value_now;
 429
 430                         /* Ignore IRQs whose rate is less than the clock */
 431                         if (delta < useful_load_threshold)
 432                                 continue;
 433                         /* update the load for the processor or package total */
 434                         IRQ_DELTA(package_index,j) += delta;
 435
 436                         /* Keep track of the higher numbered sibling as well */
 437                         if (i != package_index)
 438                                 CPU_IRQ(i) += delta;
 439                         /*
 440                          * We have sibling A and sibling B in the package
 441                          *
 442                          * cpu_irq[A] = load for cpu A + load for cpu B
 443                          * cpu_irq[B] = load for cpu B
 444                          */
 445                         CPU_IRQ(package_index) += delta;
 446                 }
 447         }
 448         /* Find the least loaded processor package */
 449         for_each_online_cpu(i) {
 450                 if (i != CPU_TO_PACKAGEINDEX(i))
 451                         continue;
 452                 if (min_cpu_irq > CPU_IRQ(i)) {
 453                         min_cpu_irq = CPU_IRQ(i);
 454                         min_loaded = i;
 455                 }
 456         }
 457         max_cpu_irq = ULONG_MAX;
 458
 459 tryanothercpu:
 460         /* Look for heaviest loaded processor.
 461          * We may come back to get the next heaviest loaded processor.
 462          * Skip processors with trivial loads.
 463          */
 464         tmp_cpu_irq = 0;
 465         tmp_loaded = -1;
 466         for_each_online_cpu(i) {
 467                 if (i != CPU_TO_PACKAGEINDEX(i))
 468                         continue;
 469                 if (max_cpu_irq <= CPU_IRQ(i))
 470                         continue;
 471                 if (tmp_cpu_irq < CPU_IRQ(i)) {
 472                         tmp_cpu_irq = CPU_IRQ(i);
 473                         tmp_loaded = i;
 474                 }
 475         }
 476
 477         if (tmp_loaded == -1) {
 478          /* In the case of small number of heavy interrupt sources,
 479           * loading some of the cpus too much. We use Ingo's original
 480           * approach to rotate them around.
 481           */
 482                 if (!first_attempt && imbalance >= useful_load_threshold) {
 483                         rotate_irqs_among_cpus(useful_load_threshold);
 484                         return;
 485                 }
 486                 goto not_worth_the_effort;
 487         }
 488
 489         first_attempt = 0;              /* heaviest search */
 490         max_cpu_irq = tmp_cpu_irq;      /* load */
 491         max_loaded = tmp_loaded;        /* processor */
 492         imbalance = (max_cpu_irq - min_cpu_irq) / 2;
 493
 494         Dprintk("max_loaded cpu = %d\n", max_loaded);
 495         Dprintk("min_loaded cpu = %d\n", min_loaded);
 496         Dprintk("max_cpu_irq load = %ld\n", max_cpu_irq);
 497         Dprintk("min_cpu_irq load = %ld\n", min_cpu_irq);
 498         Dprintk("load imbalance = %lu\n", imbalance);
 499
 500         /* if imbalance is less than approx 10% of max load, then
 501          * observe diminishing returns action. - quit
 502          */
 503         if (imbalance < (max_cpu_irq >> 3)) {
 504                 Dprintk("Imbalance too trivial\n");
 505                 goto not_worth_the_effort;
 506         }
 507
 508 tryanotherirq:
 509         /* if we select an IRQ to move that can't go where we want, then
 510          * see if there is another one to try.
 511          */
 512         move_this_load = 0;
 513         selected_irq = -1;
 514         for (j = 0; j < NR_IRQS; j++) {
 515                 /* Is this an active IRQ? */
 516                 if (!irq_desc[j].action)
 517                         continue;
 518                 if (imbalance <= IRQ_DELTA(max_loaded,j))
 519                         continue;
 520                 /* Try to find the IRQ that is closest to the imbalance
 521                  * without going over.
 522                  */
 523                 if (move_this_load < IRQ_DELTA(max_loaded,j)) {
 524                         move_this_load = IRQ_DELTA(max_loaded,j);
 525                         selected_irq = j;
 526                 }
 527         }
 528         if (selected_irq == -1) {
 529                 goto tryanothercpu;
 530         }
 531
 532         imbalance = move_this_load;
 533
 534         /* For physical_balance case, we accumlated both load
 535          * values in the one of the siblings cpu_irq[],
 536          * to use the same code for physical and logical processors
 537          * as much as possible.
 538          *
 539          * NOTE: the cpu_irq[] array holds the sum of the load for
 540          * sibling A and sibling B in the slot for the lowest numbered
 541          * sibling (A), _AND_ the load for sibling B in the slot for
 542          * the higher numbered sibling.
 543          *
 544          * We seek the least loaded sibling by making the comparison
 545          * (A+B)/2 vs B
 546          */
 547         load = CPU_IRQ(min_loaded) >> 1;
 548         for_each_cpu_mask(j, cpu_sibling_map[min_loaded]) {
 549                 if (load > CPU_IRQ(j)) {
 550                         /* This won't change cpu_sibling_map[min_loaded] */
 551                         load = CPU_IRQ(j);
 552                         min_loaded = j;
 553                 }
 554         }
 555
 556         cpus_and(allowed_mask,
 557                 cpu_online_map,
 558                 balance_irq_affinity[selected_irq]);
 559         target_cpu_mask = cpumask_of_cpu(min_loaded);
 560         cpus_and(tmp, target_cpu_mask, allowed_mask);
 561
 562         if (!cpus_empty(tmp)) {
 563
 564                 Dprintk("irq = %d moved to cpu = %d\n",
 565                                 selected_irq, min_loaded);
 566                 /* mark for change destination */
 567                 set_pending_irq(selected_irq, cpumask_of_cpu(min_loaded));
 568
 569                 /* Since we made a change, come back sooner to
 570                  * check for more variation.
 571                  */
 572                 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 573                         balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 574                 return;
 575         }
 576         goto tryanotherirq;
 577
 578 not_worth_the_effort:
 579         /*
 580          * if we did not find an IRQ to move, then adjust the time interval
 581          * upward
 582          */
 583         balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
 584                 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
 585         Dprintk("IRQ worth rotating not found\n");
 586         return;
 587 }
 588
 589 static int balanced_irq(void *unused)
 590 {
 591         int i;
 592         unsigned long prev_balance_time = jiffies;
 593         long time_remaining = balanced_irq_interval;
 594
 595         daemonize("kirqd");
 596
 597         /* push everything to CPU 0 to give us a starting point.  */
 598         for (i = 0 ; i < NR_IRQS ; i++) {
 599                 irq_desc[i].pending_mask = cpumask_of_cpu(0);
 600                 set_pending_irq(i, cpumask_of_cpu(0));
 601         }
 602
 603         for ( ; ; ) {
 604                 time_remaining = schedule_timeout_interruptible(time_remaining);
 605                 try_to_freeze();
 606                 if (time_after(jiffies,
 607                                 prev_balance_time+balanced_irq_interval)) {
 608                         preempt_disable();
 609                         do_irq_balance();
 610                         prev_balance_time = jiffies;
 611                         time_remaining = balanced_irq_interval;
 612                         preempt_enable();
 613                 }
 614         }
 615         return 0;
 616 }
 617
 618 static int __init balanced_irq_init(void)
 619 {
 620         int i;
 621         struct cpuinfo_x86 *c;
 622         cpumask_t tmp;
 623
 624         cpus_shift_right(tmp, cpu_online_map, 2);
 625         c = &boot_cpu_data;
 626         /* When not overwritten by the command line ask subarchitecture. */
 627         if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
 628                 irqbalance_disabled = NO_BALANCE_IRQ;
 629         if (irqbalance_disabled)
 630                 return 0;
 631
 632          /* disable irqbalance completely if there is only one processor online */
 633         if (num_online_cpus() < 2) {
 634                 irqbalance_disabled = 1;
 635                 return 0;
 636         }
 637         /*
 638          * Enable physical balance only if more than 1 physical processor
 639          * is present
 640          */
 641         if (smp_num_siblings > 1 && !cpus_empty(tmp))
 642                 physical_balance = 1;
 643
 644         for_each_online_cpu(i) {
 645                 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 646                 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 647                 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
 648                         printk(KERN_ERR "balanced_irq_init: out of memory");
 649                         goto failed;
 650                 }
 651                 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
 652                 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
 653         }
 654
 655         printk(KERN_INFO "Starting balanced_irq\n");
 656         if (kernel_thread(balanced_irq, NULL, CLONE_KERNEL) >= 0)
 657                 return 0;
 658         else
 659                 printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
 660 failed:
 661         for_each_possible_cpu(i) {
 662                 kfree(irq_cpu_data[i].irq_delta);
 663                 irq_cpu_data[i].irq_delta = NULL;
 664                 kfree(irq_cpu_data[i].last_irq);
 665                 irq_cpu_data[i].last_irq = NULL;
 666         }
 667         return 0;
 668 }
 669
 670 int __init irqbalance_disable(char *str)
 671 {
 672         irqbalance_disabled = 1;
 673         return 1;
 674 }
 675
 676 __setup("noirqbalance", irqbalance_disable);
 677
 678 late_initcall(balanced_irq_init);
 679 #endif /* CONFIG_IRQBALANCE */
 680 #endif /* CONFIG_SMP */
 681
 682 #ifndef CONFIG_SMP
 683 void fastcall send_IPI_self(int vector)
 684 {
 685         unsigned int cfg;
 686
 687         /*
 688          * Wait for idle.
 689          */
 690         apic_wait_icr_idle();
 691         cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
 692         /*
 693          * Send the IPI. The write to APIC_ICR fires this off.
 694          */
 695         apic_write_around(APIC_ICR, cfg);
 696 }
 697 #endif /* !CONFIG_SMP */
 698
 699
 700 /*
 701  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
 702  * specific CPU-side IRQs.
 703  */
 704
 705 #define MAX_PIRQS 8
 706 static int pirq_entries [MAX_PIRQS];
 707 static int pirqs_enabled;
 708 int skip_ioapic_setup;
 709
 710 static int __init ioapic_setup(char *str)
 711 {
 712         skip_ioapic_setup = 1;
 713         return 1;
 714 }
 715
 716 __setup("noapic", ioapic_setup);
 717
 718 static int __init ioapic_pirq_setup(char *str)
 719 {
 720         int i, max;
 721         int ints[MAX_PIRQS+1];
 722
 723         get_options(str, ARRAY_SIZE(ints), ints);
 724
 725         for (i = 0; i < MAX_PIRQS; i++)
 726                 pirq_entries[i] = -1;
 727
 728         pirqs_enabled = 1;
 729         apic_printk(APIC_VERBOSE, KERN_INFO
 730                         "PIRQ redirection, working around broken MP-BIOS.\n");
 731         max = MAX_PIRQS;
 732         if (ints[0] < MAX_PIRQS)
 733                 max = ints[0];
 734
 735         for (i = 0; i < max; i++) {
 736                 apic_printk(APIC_VERBOSE, KERN_DEBUG
 737                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
 738                 /*
 739                  * PIRQs are mapped upside down, usually.
 740                  */
 741                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
 742         }
 743         return 1;
 744 }
 745
 746 __setup("pirq=", ioapic_pirq_setup);
 747
 748 /*
 749  * Find the IRQ entry number of a certain pin.
 750  */
 751 static int find_irq_entry(int apic, int pin, int type)
 752 {
 753         int i;
 754
 755         for (i = 0; i < mp_irq_entries; i++)
 756                 if (mp_irqs[i].mpc_irqtype == type &&
 757                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
 758                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
 759                     mp_irqs[i].mpc_dstirq == pin)
 760                         return i;
 761
 762         return -1;
 763 }
 764
 765 /*
 766  * Find the pin to which IRQ[irq] (ISA) is connected
 767  */
 768 static int __init find_isa_irq_pin(int irq, int type)
 769 {
 770         int i;
 771
 772         for (i = 0; i < mp_irq_entries; i++) {
 773                 int lbus = mp_irqs[i].mpc_srcbus;
 774
 775                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
 776                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
 777                      mp_bus_id_to_type[lbus] == MP_BUS_MCA ||
 778                      mp_bus_id_to_type[lbus] == MP_BUS_NEC98
 779                     ) &&
 780                     (mp_irqs[i].mpc_irqtype == type) &&
 781                     (mp_irqs[i].mpc_srcbusirq == irq))
 782
 783                         return mp_irqs[i].mpc_dstirq;
 784         }
 785         return -1;
 786 }
 787
 788 static int __init find_isa_irq_apic(int irq, int type)
 789 {
 790         int i;
 791
 792         for (i = 0; i < mp_irq_entries; i++) {
 793                 int lbus = mp_irqs[i].mpc_srcbus;
 794
 795                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
 796                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
 797                      mp_bus_id_to_type[lbus] == MP_BUS_MCA ||
 798                      mp_bus_id_to_type[lbus] == MP_BUS_NEC98
 799                     ) &&
 800                     (mp_irqs[i].mpc_irqtype == type) &&
 801                     (mp_irqs[i].mpc_srcbusirq == irq))
 802                         break;
 803         }
 804         if (i < mp_irq_entries) {
 805                 int apic;
 806                 for(apic = 0; apic < nr_ioapics; apic++) {
 807                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
 808                                 return apic;
 809                 }
 810         }
 811
 812         return -1;
 813 }
 814
 815 /*
 816  * Find a specific PCI IRQ entry.
 817  * Not an __init, possibly needed by modules
 818  */
 819 static int pin_2_irq(int idx, int apic, int pin);
 820
 821 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
 822 {
 823         int apic, i, best_guess = -1;
 824
 825         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
 826                 "slot:%d, pin:%d.\n", bus, slot, pin);
 827         if (mp_bus_id_to_pci_bus[bus] == -1) {
 828                 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
 829                 return -1;
 830         }
 831         for (i = 0; i < mp_irq_entries; i++) {
 832                 int lbus = mp_irqs[i].mpc_srcbus;
 833
 834                 for (apic = 0; apic < nr_ioapics; apic++)
 835                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
 836                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
 837                                 break;
 838
 839                 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
 840                     !mp_irqs[i].mpc_irqtype &&
 841                     (bus == lbus) &&
 842                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
 843                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
 844
 845                         if (!(apic || IO_APIC_IRQ(irq)))
 846                                 continue;
 847
 848                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
 849                                 return irq;
 850                         /*
 851                          * Use the first all-but-pin matching entry as a
 852                          * best-guess fuzzy result for broken mptables.
 853                          */
 854                         if (best_guess < 0)
 855                                 best_guess = irq;
 856                 }
 857         }
 858         return best_guess;
 859 }
 860 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
 861
 862 /*
 863  * This function currently is only a helper for the i386 smp boot process where
 864  * we need to reprogram the ioredtbls to cater for the cpus which have come online
 865  * so mask in all cases should simply be TARGET_CPUS
 866  */
 867 #ifdef CONFIG_SMP
 868 void __init setup_ioapic_dest(void)
 869 {
 870         int pin, ioapic, irq, irq_entry;
 871
 872         if (skip_ioapic_setup == 1)
 873                 return;
 874
 875         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
 876                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
 877                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
 878                         if (irq_entry == -1)
 879                                 continue;
 880                         irq = pin_2_irq(irq_entry, ioapic, pin);
 881                         set_ioapic_affinity_irq(irq, TARGET_CPUS);
 882                 }
 883
 884         }
 885 }
 886 #endif
 887
 888 /*
 889  * EISA Edge/Level control register, ELCR
 890  */
 891 static int EISA_ELCR(unsigned int irq)
 892 {
 893         if (irq < 16) {
 894                 unsigned int port = 0x4d0 + (irq >> 3);
 895                 return (inb(port) >> (irq & 7)) & 1;
 896         }
 897         apic_printk(APIC_VERBOSE, KERN_INFO
 898                         "Broken MPtable reports ISA irq %d\n", irq);
 899         return 0;
 900 }
 901
 902 /* EISA interrupts are always polarity zero and can be edge or level
 903  * trigger depending on the ELCR value.  If an interrupt is listed as
 904  * EISA conforming in the MP table, that means its trigger type must
 905  * be read in from the ELCR */
 906
 907 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
 908 #define default_EISA_polarity(idx)      (0)
 909
 910 /* ISA interrupts are always polarity zero edge triggered,
 911  * when listed as conforming in the MP table. */
 912
 913 #define default_ISA_trigger(idx)        (0)
 914 #define default_ISA_polarity(idx)       (0)
 915
 916 /* PCI interrupts are always polarity one level triggered,
 917  * when listed as conforming in the MP table. */
 918
 919 #define default_PCI_trigger(idx)        (1)
 920 #define default_PCI_polarity(idx)       (1)
 921
 922 /* MCA interrupts are always polarity zero level triggered,
 923  * when listed as conforming in the MP table. */
 924
 925 #define default_MCA_trigger(idx)        (1)
 926 #define default_MCA_polarity(idx)       (0)
 927
 928 /* NEC98 interrupts are always polarity zero edge triggered,
 929  * when listed as conforming in the MP table. */
 930
 931 #define default_NEC98_trigger(idx)     (0)
 932 #define default_NEC98_polarity(idx)    (0)
 933
 934 static int __init MPBIOS_polarity(int idx)
 935 {
 936         int bus = mp_irqs[idx].mpc_srcbus;
 937         int polarity;
 938
 939         /*
 940          * Determine IRQ line polarity (high active or low active):
 941          */
 942         switch (mp_irqs[idx].mpc_irqflag & 3)
 943         {
 944                 case 0: /* conforms, ie. bus-type dependent polarity */
 945                 {
 946                         switch (mp_bus_id_to_type[bus])
 947                         {
 948                                 case MP_BUS_ISA: /* ISA pin */
 949                                 {
 950                                         polarity = default_ISA_polarity(idx);
 951                                         break;
 952                                 }
 953                                 case MP_BUS_EISA: /* EISA pin */
 954                                 {
 955                                         polarity = default_EISA_polarity(idx);
 956                                         break;
 957                                 }
 958                                 case MP_BUS_PCI: /* PCI pin */
 959                                 {
 960                                         polarity = default_PCI_polarity(idx);
 961                                         break;
 962                                 }
 963                                 case MP_BUS_MCA: /* MCA pin */
 964                                 {
 965                                         polarity = default_MCA_polarity(idx);
 966                                         break;
 967                                 }
 968                                 case MP_BUS_NEC98: /* NEC 98 pin */
 969                                 {
 970                                         polarity = default_NEC98_polarity(idx);
 971                                         break;
 972                                 }
 973                                 default:
 974                                 {
 975                                         printk(KERN_WARNING "broken BIOS!!\n");
 976                                         polarity = 1;
 977                                         break;
 978                                 }
 979                         }
 980                         break;
 981                 }
 982                 case 1: /* high active */
 983                 {
 984                         polarity = 0;
 985                         break;
 986                 }
 987                 case 2: /* reserved */
 988                 {
 989                         printk(KERN_WARNING "broken BIOS!!\n");
 990                         polarity = 1;
 991                         break;
 992                 }
 993                 case 3: /* low active */
 994                 {
 995                         polarity = 1;
 996                         break;
 997                 }
 998                 default: /* invalid */
 999                 {
1000                         printk(KERN_WARNING "broken BIOS!!\n");
1001                         polarity = 1;
1002                         break;
1003                 }
1004         }
1005         return polarity;
1006 }
1007
1008 static int MPBIOS_trigger(int idx)
1009 {
1010         int bus = mp_irqs[idx].mpc_srcbus;
1011         int trigger;
1012
1013         /*
1014          * Determine IRQ trigger mode (edge or level sensitive):
1015          */
1016         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
1017         {
1018                 case 0: /* conforms, ie. bus-type dependent */
1019                 {
1020                         switch (mp_bus_id_to_type[bus])
1021                         {
1022                                 case MP_BUS_ISA: /* ISA pin */
1023                                 {
1024                                         trigger = default_ISA_trigger(idx);
1025                                         break;
1026                                 }
1027                                 case MP_BUS_EISA: /* EISA pin */
1028                                 {
1029                                         trigger = default_EISA_trigger(idx);
1030                                         break;
1031                                 }
1032                                 case MP_BUS_PCI: /* PCI pin */
1033                                 {
1034                                         trigger = default_PCI_trigger(idx);
1035                                         break;
1036                                 }
1037                                 case MP_BUS_MCA: /* MCA pin */
1038                                 {
1039                                         trigger = default_MCA_trigger(idx);
1040                                         break;
1041                                 }
1042                                 case MP_BUS_NEC98: /* NEC 98 pin */
1043                                 {
1044                                         trigger = default_NEC98_trigger(idx);
1045                                         break;
1046                                 }
1047                                 default:
1048                                 {
1049                                         printk(KERN_WARNING "broken BIOS!!\n");
1050                                         trigger = 1;
1051                                         break;
1052                                 }
1053                         }
1054                         break;
1055                 }
1056                 case 1: /* edge */
1057                 {
1058                         trigger = 0;
1059                         break;
1060                 }
1061                 case 2: /* reserved */
1062                 {
1063                         printk(KERN_WARNING "broken BIOS!!\n");
1064                         trigger = 1;
1065                         break;
1066                 }
1067                 case 3: /* level */
1068                 {
1069                         trigger = 1;
1070                         break;
1071                 }
1072                 default: /* invalid */
1073                 {
1074                         printk(KERN_WARNING "broken BIOS!!\n");
1075                         trigger = 0;
1076                         break;
1077                 }
1078         }
1079         return trigger;
1080 }
1081
1082 static inline int irq_polarity(int idx)
1083 {
1084         return MPBIOS_polarity(idx);
1085 }
1086
1087 static inline int irq_trigger(int idx)
1088 {
1089         return MPBIOS_trigger(idx);
1090 }
1091
1092 static int pin_2_irq(int idx, int apic, int pin)
1093 {
1094         int irq, i;
1095         int bus = mp_irqs[idx].mpc_srcbus;
1096
1097         /*
1098          * Debugging check, we are in big trouble if this message pops up!
1099          */
1100         if (mp_irqs[idx].mpc_dstirq != pin)
1101                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1102
1103         switch (mp_bus_id_to_type[bus])
1104         {
1105                 case MP_BUS_ISA: /* ISA pin */
1106                 case MP_BUS_EISA:
1107                 case MP_BUS_MCA:
1108                 case MP_BUS_NEC98:
1109                 {
1110                         irq = mp_irqs[idx].mpc_srcbusirq;
1111                         break;
1112                 }
1113                 case MP_BUS_PCI: /* PCI pin */
1114                 {
1115                         /*
1116                          * PCI IRQs are mapped in order
1117                          */
1118                         i = irq = 0;
1119                         while (i < apic)
1120                                 irq += nr_ioapic_registers[i++];
1121                         irq += pin;
1122
1123                         /*
1124                          * For MPS mode, so far only needed by ES7000 platform
1125                          */
1126                         if (ioapic_renumber_irq)
1127                                 irq = ioapic_renumber_irq(apic, irq);
1128
1129                         break;
1130                 }
1131                 default:
1132                 {
1133                         printk(KERN_ERR "unknown bus type %d.\n",bus);
1134                         irq = 0;
1135                         break;
1136                 }
1137         }
1138
1139         /*
1140          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1141          */
1142         if ((pin >= 16) && (pin <= 23)) {
1143                 if (pirq_entries[pin-16] != -1) {
1144                         if (!pirq_entries[pin-16]) {
1145                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1146                                                 "disabling PIRQ%d\n", pin-16);
1147                         } else {
1148                                 irq = pirq_entries[pin-16];
1149                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1150                                                 "using PIRQ%d -> IRQ %d\n",
1151                                                 pin-16, irq);
1152                         }
1153                 }
1154         }
1155         return irq;
1156 }
1157
1158 static inline int IO_APIC_irq_trigger(int irq)
1159 {
1160         int apic, idx, pin;
1161
1162         for (apic = 0; apic < nr_ioapics; apic++) {
1163                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1164                         idx = find_irq_entry(apic,pin,mp_INT);
1165                         if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1166                                 return irq_trigger(idx);
1167                 }
1168         }
1169         /*
1170          * nonexistent IRQs are edge default
1171          */
1172         return 0;
1173 }
1174
1175 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1176 u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 };
1177
1178 static int __assign_irq_vector(int irq)
1179 {
1180         static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
1181         int vector;
1182
1183         BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
1184
1185         if (IO_APIC_VECTOR(irq) > 0)
1186                 return IO_APIC_VECTOR(irq);
1187
1188         current_vector += 8;
1189         if (current_vector == SYSCALL_VECTOR)
1190                 current_vector += 8;
1191
1192         if (current_vector >= FIRST_SYSTEM_VECTOR) {
1193                 offset++;
1194                 if (!(offset % 8))
1195                         return -ENOSPC;
1196                 current_vector = FIRST_DEVICE_VECTOR + offset;
1197         }
1198
1199         vector = current_vector;
1200         IO_APIC_VECTOR(irq) = vector;
1201
1202         return vector;
1203 }
1204
1205 static int assign_irq_vector(int irq)
1206 {
1207         unsigned long flags;
1208         int vector;
1209
1210         spin_lock_irqsave(&vector_lock, flags);
1211         vector = __assign_irq_vector(irq);
1212         spin_unlock_irqrestore(&vector_lock, flags);
1213
1214         return vector;
1215 }
1216 static struct irq_chip ioapic_chip;
1217
1218 #define IOAPIC_AUTO     -1
1219 #define IOAPIC_EDGE     0
1220 #define IOAPIC_LEVEL    1
1221
1222 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1223 {
1224         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1225                         trigger == IOAPIC_LEVEL)
1226                 set_irq_chip_and_handler(irq, &ioapic_chip,
1227                                          handle_fasteoi_irq);
1228         else
1229                 set_irq_chip_and_handler(irq, &ioapic_chip,
1230                                          handle_edge_irq);
1231         set_intr_gate(vector, interrupt[irq]);
1232 }
1233
1234 static void __init setup_IO_APIC_irqs(void)
1235 {
1236         struct IO_APIC_route_entry entry;
1237         int apic, pin, idx, irq, first_notcon = 1, vector;
1238         unsigned long flags;
1239
1240         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1241
1242         for (apic = 0; apic < nr_ioapics; apic++) {
1243         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1244
1245                 /*
1246                  * add it to the IO-APIC irq-routing table:
1247                  */
1248                 memset(&entry,0,sizeof(entry));
1249
1250                 entry.delivery_mode = INT_DELIVERY_MODE;
1251                 entry.dest_mode = INT_DEST_MODE;
1252                 entry.mask = 0;                         /* enable IRQ */
1253                 entry.dest.logical.logical_dest =
1254                                         cpu_mask_to_apicid(TARGET_CPUS);
1255
1256                 idx = find_irq_entry(apic,pin,mp_INT);
1257                 if (idx == -1) {
1258                         if (first_notcon) {
1259                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1260                                                 " IO-APIC (apicid-pin) %d-%d",
1261                                                 mp_ioapics[apic].mpc_apicid,
1262                                                 pin);
1263                                 first_notcon = 0;
1264                         } else
1265                                 apic_printk(APIC_VERBOSE, ", %d-%d",
1266                                         mp_ioapics[apic].mpc_apicid, pin);
1267                         continue;
1268                 }
1269
1270                 entry.trigger = irq_trigger(idx);
1271                 entry.polarity = irq_polarity(idx);
1272
1273                 if (irq_trigger(idx)) {
1274                         entry.trigger = 1;
1275                         entry.mask = 1;
1276                 }
1277
1278                 irq = pin_2_irq(idx, apic, pin);
1279                 /*
1280                  * skip adding the timer int on secondary nodes, which causes
1281                  * a small but painful rift in the time-space continuum
1282                  */
1283                 if (multi_timer_check(apic, irq))
1284                         continue;
1285                 else
1286                         add_pin_to_irq(irq, apic, pin);
1287
1288                 if (!apic && !IO_APIC_IRQ(irq))
1289                         continue;
1290
1291                 if (IO_APIC_IRQ(irq)) {
1292                         vector = assign_irq_vector(irq);
1293                         entry.vector = vector;
1294                         ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1295
1296                         if (!apic && (irq < 16))
1297                                 disable_8259A_irq(irq);
1298                 }
1299                 ioapic_write_entry(apic, pin, entry);
1300                 spin_lock_irqsave(&ioapic_lock, flags);
1301                 set_native_irq_info(irq, TARGET_CPUS);
1302                 spin_unlock_irqrestore(&ioapic_lock, flags);
1303         }
1304         }
1305
1306         if (!first_notcon)
1307                 apic_printk(APIC_VERBOSE, " not connected.\n");
1308 }
1309
1310 /*
1311  * Set up the 8259A-master output pin:
1312  */
1313 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
1314 {
1315         struct IO_APIC_route_entry entry;
1316
1317         memset(&entry,0,sizeof(entry));
1318
1319         disable_8259A_irq(0);
1320
1321         /* mask LVT0 */
1322         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1323
1324         /*
1325          * We use logical delivery to get the timer IRQ
1326          * to the first CPU.
1327          */
1328         entry.dest_mode = INT_DEST_MODE;
1329         entry.mask = 0;                                 /* unmask IRQ now */
1330         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1331         entry.delivery_mode = INT_DELIVERY_MODE;
1332         entry.polarity = 0;
1333         entry.trigger = 0;
1334         entry.vector = vector;
1335
1336         /*
1337          * The timer IRQ doesn't have to know that behind the
1338          * scene we have a 8259A-master in AEOI mode ...
1339          */
1340         irq_desc[0].chip = &ioapic_chip;
1341         set_irq_handler(0, handle_edge_irq);
1342
1343         /*
1344          * Add it to the IO-APIC irq-routing table:
1345          */
1346         ioapic_write_entry(apic, pin, entry);
1347
1348         enable_8259A_irq(0);
1349 }
1350
1351 static inline void UNEXPECTED_IO_APIC(void)
1352 {
1353 }
1354
1355 void __init print_IO_APIC(void)
1356 {
1357         int apic, i;
1358         union IO_APIC_reg_00 reg_00;
1359         union IO_APIC_reg_01 reg_01;
1360         union IO_APIC_reg_02 reg_02;
1361         union IO_APIC_reg_03 reg_03;
1362         unsigned long flags;
1363
1364         if (apic_verbosity == APIC_QUIET)
1365                 return;
1366
1367         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1368         for (i = 0; i < nr_ioapics; i++)
1369                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1370                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1371
1372         /*
1373          * We are a bit conservative about what we expect.  We have to
1374          * know about every hardware change ASAP.
1375          */
1376         printk(KERN_INFO "testing the IO APIC.......................\n");
1377
1378         for (apic = 0; apic < nr_ioapics; apic++) {
1379
1380         spin_lock_irqsave(&ioapic_lock, flags);
1381         reg_00.raw = io_apic_read(apic, 0);
1382         reg_01.raw = io_apic_read(apic, 1);
1383         if (reg_01.bits.version >= 0x10)
1384                 reg_02.raw = io_apic_read(apic, 2);
1385         if (reg_01.bits.version >= 0x20)
1386                 reg_03.raw = io_apic_read(apic, 3);
1387         spin_unlock_irqrestore(&ioapic_lock, flags);
1388
1389         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1390         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1391         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1392         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1393         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1394         if (reg_00.bits.ID >= get_physical_broadcast())
1395                 UNEXPECTED_IO_APIC();
1396         if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1397                 UNEXPECTED_IO_APIC();
1398
1399         printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1400         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1401         if (    (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1402                 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1403                 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1404                 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1405                 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1406                 (reg_01.bits.entries != 0x2E) &&
1407                 (reg_01.bits.entries != 0x3F)
1408         )
1409                 UNEXPECTED_IO_APIC();
1410
1411         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1412         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1413         if (    (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1414                 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1415                 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1416                 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1417                 (reg_01.bits.version != 0x20)    /* Intel P64H (82806 AA) */
1418         )
1419                 UNEXPECTED_IO_APIC();
1420         if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1421                 UNEXPECTED_IO_APIC();
1422
1423         /*
1424          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1425          * but the value of reg_02 is read as the previous read register
1426          * value, so ignore it if reg_02 == reg_01.
1427          */
1428         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1429                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1430                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
1431                 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1432                         UNEXPECTED_IO_APIC();
1433         }
1434
1435         /*
1436          * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1437          * or reg_03, but the value of reg_0[23] is read as the previous read
1438          * register value, so ignore it if reg_03 == reg_0[12].
1439          */
1440         if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1441             reg_03.raw != reg_01.raw) {
1442                 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1443                 printk(KERN_DEBUG ".......     : Boot DT    : %X\n", reg_03.bits.boot_DT);
1444                 if (reg_03.bits.__reserved_1)
1445                         UNEXPECTED_IO_APIC();
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 = 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 /*
1866  * There is a nasty bug in some older SMP boards, their mptable lies
1867  * about the timer IRQ. We do the following to work around the situation:
1868  *
1869  *      - timer IRQ defaults to IO-APIC IRQ
1870  *      - if this function detects that timer IRQs are defunct, then we fall
1871  *        back to ISA timer IRQs
1872  */
1873 static int __init timer_irq_works(void)
1874 {
1875         unsigned long t1 = jiffies;
1876
1877         local_irq_enable();
1878         /* Let ten ticks pass... */
1879         mdelay((10 * 1000) / HZ);
1880
1881         /*
1882          * Expect a few ticks at least, to be sure some possible
1883          * glue logic does not lock up after one or two first
1884          * ticks in a non-ExtINT mode.  Also the local APIC
1885          * might have cached one ExtINT interrupt.  Finally, at
1886          * least one tick may be lost due to delays.
1887          */
1888         if (jiffies - t1 > 4)
1889                 return 1;
1890
1891         return 0;
1892 }
1893
1894 /*
1895  * In the SMP+IOAPIC case it might happen that there are an unspecified
1896  * number of pending IRQ events unhandled. These cases are very rare,
1897  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1898  * better to do it this way as thus we do not have to be aware of
1899  * 'pending' interrupts in the IRQ path, except at this point.
1900  */
1901 /*
1902  * Edge triggered needs to resend any interrupt
1903  * that was delayed but this is now handled in the device
1904  * independent code.
1905  */
1906
1907 /*
1908  * Startup quirk:
1909  *
1910  * Starting up a edge-triggered IO-APIC interrupt is
1911  * nasty - we need to make sure that we get the edge.
1912  * If it is already asserted for some reason, we need
1913  * return 1 to indicate that is was pending.
1914  *
1915  * This is not complete - we should be able to fake
1916  * an edge even if it isn't on the 8259A...
1917  *
1918  * (We do this for level-triggered IRQs too - it cannot hurt.)
1919  */
1920 static unsigned int startup_ioapic_irq(unsigned int irq)
1921 {
1922         int was_pending = 0;
1923         unsigned long flags;
1924
1925         spin_lock_irqsave(&ioapic_lock, flags);
1926         if (irq < 16) {
1927                 disable_8259A_irq(irq);
1928                 if (i8259A_irq_pending(irq))
1929                         was_pending = 1;
1930         }
1931         __unmask_IO_APIC_irq(irq);
1932         spin_unlock_irqrestore(&ioapic_lock, flags);
1933
1934         return was_pending;
1935 }
1936
1937 static void ack_ioapic_irq(unsigned int irq)
1938 {
1939         move_native_irq(irq);
1940         ack_APIC_irq();
1941 }
1942
1943 static void ack_ioapic_quirk_irq(unsigned int irq)
1944 {
1945         unsigned long v;
1946         int i;
1947
1948         move_native_irq(irq);
1949 /*
1950  * It appears there is an erratum which affects at least version 0x11
1951  * of I/O APIC (that's the 82093AA and cores integrated into various
1952  * chipsets).  Under certain conditions a level-triggered interrupt is
1953  * erroneously delivered as edge-triggered one but the respective IRR
1954  * bit gets set nevertheless.  As a result the I/O unit expects an EOI
1955  * message but it will never arrive and further interrupts are blocked
1956  * from the source.  The exact reason is so far unknown, but the
1957  * phenomenon was observed when two consecutive interrupt requests
1958  * from a given source get delivered to the same CPU and the source is
1959  * temporarily disabled in between.
1960  *
1961  * A workaround is to simulate an EOI message manually.  We achieve it
1962  * by setting the trigger mode to edge and then to level when the edge
1963  * trigger mode gets detected in the TMR of a local APIC for a
1964  * level-triggered interrupt.  We mask the source for the time of the
1965  * operation to prevent an edge-triggered interrupt escaping meanwhile.
1966  * The idea is from Manfred Spraul.  --macro
1967  */
1968         i = IO_APIC_VECTOR(irq);
1969
1970         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1971
1972         ack_APIC_irq();
1973
1974         if (!(v & (1 << (i & 0x1f)))) {
1975                 atomic_inc(&irq_mis_count);
1976                 spin_lock(&ioapic_lock);
1977                 __mask_and_edge_IO_APIC_irq(irq);
1978                 __unmask_and_level_IO_APIC_irq(irq);
1979                 spin_unlock(&ioapic_lock);
1980         }
1981 }
1982
1983 static int ioapic_retrigger_irq(unsigned int irq)
1984 {
1985         send_IPI_self(IO_APIC_VECTOR(irq));
1986
1987         return 1;
1988 }
1989
1990 static struct irq_chip ioapic_chip __read_mostly = {
1991         .name           = "IO-APIC",
1992         .startup        = startup_ioapic_irq,
1993         .mask           = mask_IO_APIC_irq,
1994         .unmask         = unmask_IO_APIC_irq,
1995         .ack            = ack_ioapic_irq,
1996         .eoi            = ack_ioapic_quirk_irq,
1997 #ifdef CONFIG_SMP
1998         .set_affinity   = set_ioapic_affinity_irq,
1999 #endif
2000         .retrigger      = ioapic_retrigger_irq,
2001 };
2002
2003
2004 static inline void init_IO_APIC_traps(void)
2005 {
2006         int irq;
2007
2008         /*
2009          * NOTE! The local APIC isn't very good at handling
2010          * multiple interrupts at the same interrupt level.
2011          * As the interrupt level is determined by taking the
2012          * vector number and shifting that right by 4, we
2013          * want to spread these out a bit so that they don't
2014          * all fall in the same interrupt level.
2015          *
2016          * Also, we've got to be careful not to trash gate
2017          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2018          */
2019         for (irq = 0; irq < NR_IRQS ; irq++) {
2020                 int tmp = irq;
2021                 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
2022                         /*
2023                          * Hmm.. We don't have an entry for this,
2024                          * so default to an old-fashioned 8259
2025                          * interrupt if we can..
2026                          */
2027                         if (irq < 16)
2028                                 make_8259A_irq(irq);
2029                         else
2030                                 /* Strange. Oh, well.. */
2031                                 irq_desc[irq].chip = &no_irq_chip;
2032                 }
2033         }
2034 }
2035
2036 /*
2037  * The local APIC irq-chip implementation:
2038  */
2039
2040 static void ack_apic(unsigned int irq)
2041 {
2042         ack_APIC_irq();
2043 }
2044
2045 static void mask_lapic_irq (unsigned int irq)
2046 {
2047         unsigned long v;
2048
2049         v = apic_read(APIC_LVT0);
2050         apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2051 }
2052
2053 static void unmask_lapic_irq (unsigned int irq)
2054 {
2055         unsigned long v;
2056
2057         v = apic_read(APIC_LVT0);
2058         apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2059 }
2060
2061 static struct irq_chip lapic_chip __read_mostly = {
2062         .name           = "local-APIC-edge",
2063         .mask           = mask_lapic_irq,
2064         .unmask         = unmask_lapic_irq,
2065         .eoi            = ack_apic,
2066 };
2067
2068 static void setup_nmi (void)
2069 {
2070         /*
2071          * Dirty trick to enable the NMI watchdog ...
2072          * We put the 8259A master into AEOI mode and
2073          * unmask on all local APICs LVT0 as NMI.
2074          *
2075          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2076          * is from Maciej W. Rozycki - so we do not have to EOI from
2077          * the NMI handler or the timer interrupt.
2078          */
2079         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2080
2081         on_each_cpu(enable_NMI_through_LVT0, NULL, 1, 1);
2082
2083         apic_printk(APIC_VERBOSE, " done.\n");
2084 }
2085
2086 /*
2087  * This looks a bit hackish but it's about the only one way of sending
2088  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2089  * not support the ExtINT mode, unfortunately.  We need to send these
2090  * cycles as some i82489DX-based boards have glue logic that keeps the
2091  * 8259A interrupt line asserted until INTA.  --macro
2092  */
2093 static inline void unlock_ExtINT_logic(void)
2094 {
2095         int apic, pin, i;
2096         struct IO_APIC_route_entry entry0, entry1;
2097         unsigned char save_control, save_freq_select;
2098
2099         pin  = find_isa_irq_pin(8, mp_INT);
2100         apic = find_isa_irq_apic(8, mp_INT);
2101         if (pin == -1)
2102                 return;
2103
2104         entry0 = ioapic_read_entry(apic, pin);
2105         clear_IO_APIC_pin(apic, pin);
2106
2107         memset(&entry1, 0, sizeof(entry1));
2108
2109         entry1.dest_mode = 0;                   /* physical delivery */
2110         entry1.mask = 0;                        /* unmask IRQ now */
2111         entry1.dest.physical.physical_dest = hard_smp_processor_id();
2112         entry1.delivery_mode = dest_ExtINT;
2113         entry1.polarity = entry0.polarity;
2114         entry1.trigger = 0;
2115         entry1.vector = 0;
2116
2117         ioapic_write_entry(apic, pin, entry1);
2118
2119         save_control = CMOS_READ(RTC_CONTROL);
2120         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2121         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2122                    RTC_FREQ_SELECT);
2123         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2124
2125         i = 100;
2126         while (i-- > 0) {
2127                 mdelay(10);
2128                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2129                         i -= 10;
2130         }
2131
2132         CMOS_WRITE(save_control, RTC_CONTROL);
2133         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2134         clear_IO_APIC_pin(apic, pin);
2135
2136         ioapic_write_entry(apic, pin, entry0);
2137 }
2138
2139 int timer_uses_ioapic_pin_0;
2140
2141 /*
2142  * This code may look a bit paranoid, but it's supposed to cooperate with
2143  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2144  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2145  * fanatically on his truly buggy board.
2146  */
2147 static inline void check_timer(void)
2148 {
2149         int apic1, pin1, apic2, pin2;
2150         int vector;
2151
2152         /*
2153          * get/set the timer IRQ vector:
2154          */
2155         disable_8259A_irq(0);
2156         vector = assign_irq_vector(0);
2157         set_intr_gate(vector, interrupt[0]);
2158
2159         /*
2160          * Subtle, code in do_timer_interrupt() expects an AEOI
2161          * mode for the 8259A whenever interrupts are routed
2162          * through I/O APICs.  Also IRQ0 has to be enabled in
2163          * the 8259A which implies the virtual wire has to be
2164          * disabled in the local APIC.
2165          */
2166         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2167         init_8259A(1);
2168         timer_ack = 1;
2169         if (timer_over_8254 > 0)
2170                 enable_8259A_irq(0);
2171
2172         pin1  = find_isa_irq_pin(0, mp_INT);
2173         apic1 = find_isa_irq_apic(0, mp_INT);
2174         pin2  = ioapic_i8259.pin;
2175         apic2 = ioapic_i8259.apic;
2176
2177         if (pin1 == 0)
2178                 timer_uses_ioapic_pin_0 = 1;
2179
2180         printk(KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
2181                 vector, apic1, pin1, apic2, pin2);
2182
2183         if (pin1 != -1) {
2184                 /*
2185                  * Ok, does IRQ0 through the IOAPIC work?
2186                  */
2187                 unmask_IO_APIC_irq(0);
2188                 if (timer_irq_works()) {
2189                         if (nmi_watchdog == NMI_IO_APIC) {
2190                                 disable_8259A_irq(0);
2191                                 setup_nmi();
2192                                 enable_8259A_irq(0);
2193                         }
2194                         if (disable_timer_pin_1 > 0)
2195                                 clear_IO_APIC_pin(0, pin1);
2196                         return;
2197                 }
2198                 clear_IO_APIC_pin(apic1, pin1);
2199                 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to "
2200                                 "IO-APIC\n");
2201         }
2202
2203         printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2204         if (pin2 != -1) {
2205                 printk("\n..... (found pin %d) ...", pin2);
2206                 /*
2207                  * legacy devices should be connected to IO APIC #0
2208                  */
2209                 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
2210                 if (timer_irq_works()) {
2211                         printk("works.\n");
2212                         if (pin1 != -1)
2213                                 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2214                         else
2215                                 add_pin_to_irq(0, apic2, pin2);
2216                         if (nmi_watchdog == NMI_IO_APIC) {
2217                                 setup_nmi();
2218                         }
2219                         return;
2220                 }
2221                 /*
2222                  * Cleanup, just in case ...
2223                  */
2224                 clear_IO_APIC_pin(apic2, pin2);
2225         }
2226         printk(" failed.\n");
2227
2228         if (nmi_watchdog == NMI_IO_APIC) {
2229                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2230                 nmi_watchdog = 0;
2231         }
2232
2233         printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2234
2235         disable_8259A_irq(0);
2236         set_irq_chip_and_handler(0, &lapic_chip, handle_fasteoi_irq);
2237         apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector);   /* Fixed mode */
2238         enable_8259A_irq(0);
2239
2240         if (timer_irq_works()) {
2241                 printk(" works.\n");
2242                 return;
2243         }
2244         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2245         printk(" failed.\n");
2246
2247         printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2248
2249         timer_ack = 0;
2250         init_8259A(0);
2251         make_8259A_irq(0);
2252         apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2253
2254         unlock_ExtINT_logic();
2255
2256         if (timer_irq_works()) {
2257                 printk(" works.\n");
2258                 return;
2259         }
2260         printk(" failed :(.\n");
2261         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
2262                 "report.  Then try booting with the 'noapic' option");
2263 }
2264
2265 /*
2266  *
2267  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2268  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2269  *   Linux doesn't really care, as it's not actually used
2270  *   for any interrupt handling anyway.
2271  */
2272 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
2273
2274 void __init setup_IO_APIC(void)
2275 {
2276         enable_IO_APIC();
2277
2278         if (acpi_ioapic)
2279                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
2280         else
2281                 io_apic_irqs = ~PIC_IRQS;
2282
2283         printk("ENABLING IO-APIC IRQs\n");
2284
2285         /*
2286          * Set up IO-APIC IRQ routing.
2287          */
2288         if (!acpi_ioapic)
2289                 setup_ioapic_ids_from_mpc();
2290         sync_Arb_IDs();
2291         setup_IO_APIC_irqs();
2292         init_IO_APIC_traps();
2293         check_timer();
2294         if (!acpi_ioapic)
2295                 print_IO_APIC();
2296 }
2297
2298 static int __init setup_disable_8254_timer(char *s)
2299 {
2300         timer_over_8254 = -1;
2301         return 1;
2302 }
2303 static int __init setup_enable_8254_timer(char *s)
2304 {
2305         timer_over_8254 = 2;
2306         return 1;
2307 }
2308
2309 __setup("disable_8254_timer", setup_disable_8254_timer);
2310 __setup("enable_8254_timer", setup_enable_8254_timer);
2311
2312 /*
2313  *      Called after all the initialization is done. If we didnt find any
2314  *      APIC bugs then we can allow the modify fast path
2315  */
2316
2317 static int __init io_apic_bug_finalize(void)
2318 {
2319         if(sis_apic_bug == -1)
2320                 sis_apic_bug = 0;
2321         return 0;
2322 }
2323
2324 late_initcall(io_apic_bug_finalize);
2325
2326 struct sysfs_ioapic_data {
2327         struct sys_device dev;
2328         struct IO_APIC_route_entry entry[0];
2329 };
2330 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2331
2332 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2333 {
2334         struct IO_APIC_route_entry *entry;
2335         struct sysfs_ioapic_data *data;
2336         int i;
2337
2338         data = container_of(dev, struct sysfs_ioapic_data, dev);
2339         entry = data->entry;
2340         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2341                 entry[i] = ioapic_read_entry(dev->id, i);
2342
2343         return 0;
2344 }
2345
2346 static int ioapic_resume(struct sys_device *dev)
2347 {
2348         struct IO_APIC_route_entry *entry;
2349         struct sysfs_ioapic_data *data;
2350         unsigned long flags;
2351         union IO_APIC_reg_00 reg_00;
2352         int i;
2353
2354         data = container_of(dev, struct sysfs_ioapic_data, dev);
2355         entry = data->entry;
2356
2357         spin_lock_irqsave(&ioapic_lock, flags);
2358         reg_00.raw = io_apic_read(dev->id, 0);
2359         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2360                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2361                 io_apic_write(dev->id, 0, reg_00.raw);
2362         }
2363         spin_unlock_irqrestore(&ioapic_lock, flags);
2364         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2365                 ioapic_write_entry(dev->id, i, entry[i]);
2366
2367         return 0;
2368 }
2369
2370 static struct sysdev_class ioapic_sysdev_class = {
2371         set_kset_name("ioapic"),
2372         .suspend = ioapic_suspend,
2373         .resume = ioapic_resume,
2374 };
2375
2376 static int __init ioapic_init_sysfs(void)
2377 {
2378         struct sys_device * dev;
2379         int i, size, error = 0;
2380
2381         error = sysdev_class_register(&ioapic_sysdev_class);
2382         if (error)
2383                 return error;
2384
2385         for (i = 0; i < nr_ioapics; i++ ) {
2386                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2387                         * sizeof(struct IO_APIC_route_entry);
2388                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2389                 if (!mp_ioapic_data[i]) {
2390                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2391                         continue;
2392                 }
2393                 memset(mp_ioapic_data[i], 0, size);
2394                 dev = &mp_ioapic_data[i]->dev;
2395                 dev->id = i;
2396                 dev->cls = &ioapic_sysdev_class;
2397                 error = sysdev_register(dev);
2398                 if (error) {
2399                         kfree(mp_ioapic_data[i]);
2400                         mp_ioapic_data[i] = NULL;
2401                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2402                         continue;
2403                 }
2404         }
2405
2406         return 0;
2407 }
2408
2409 device_initcall(ioapic_init_sysfs);
2410
2411 #ifdef CONFIG_PCI_MSI
2412 /*
2413  * Dynamic irq allocate and deallocation for MSI
2414  */
2415 int create_irq(void)
2416 {
2417         /* Allocate an unused irq */
2418         int irq, new, vector;
2419         unsigned long flags;
2420
2421         irq = -ENOSPC;
2422         spin_lock_irqsave(&vector_lock, flags);
2423         for (new = (NR_IRQS - 1); new >= 0; new--) {
2424                 if (platform_legacy_irq(new))
2425                         continue;
2426                 if (irq_vector[new] != 0)
2427                         continue;
2428                 vector = __assign_irq_vector(new);
2429                 if (likely(vector > 0))
2430                         irq = new;
2431                 break;
2432         }
2433         spin_unlock_irqrestore(&vector_lock, flags);
2434
2435         if (irq >= 0) {
2436                 set_intr_gate(vector, interrupt[irq]);
2437                 dynamic_irq_init(irq);
2438         }
2439         return irq;
2440 }
2441
2442 void destroy_irq(unsigned int irq)
2443 {
2444         unsigned long flags;
2445
2446         dynamic_irq_cleanup(irq);
2447
2448         spin_lock_irqsave(&vector_lock, flags);
2449         irq_vector[irq] = 0;
2450         spin_unlock_irqrestore(&vector_lock, flags);
2451 }
2452 #endif /* CONFIG_PCI_MSI */
2453
2454 /*
2455  * MSI mesage composition
2456  */
2457 #ifdef CONFIG_PCI_MSI
2458 static int msi_msg_setup(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2459 {
2460         /* For now always this code always uses physical delivery
2461          * mode.
2462          */
2463         int vector;
2464         unsigned dest;
2465
2466         vector = assign_irq_vector(irq);
2467         if (vector >= 0) {
2468                 dest = cpu_mask_to_apicid(TARGET_CPUS);
2469
2470                 msg->address_hi = MSI_ADDR_BASE_HI;
2471                 msg->address_lo =
2472                         MSI_ADDR_BASE_LO |
2473                         ((INT_DEST_MODE == 0) ?
2474                                 MSI_ADDR_DEST_MODE_PHYSICAL:
2475                                 MSI_ADDR_DEST_MODE_LOGICAL) |
2476                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2477                                 MSI_ADDR_REDIRECTION_CPU:
2478                                 MSI_ADDR_REDIRECTION_LOWPRI) |
2479                         MSI_ADDR_DEST_ID(dest);
2480
2481                 msg->data =
2482                         MSI_DATA_TRIGGER_EDGE |
2483                         MSI_DATA_LEVEL_ASSERT |
2484                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2485                                 MSI_DATA_DELIVERY_FIXED:
2486                                 MSI_DATA_DELIVERY_LOWPRI) |
2487                         MSI_DATA_VECTOR(vector);
2488         }
2489         return vector;
2490 }
2491
2492 static void msi_msg_teardown(unsigned int irq)
2493 {
2494         return;
2495 }
2496
2497 static void msi_msg_set_affinity(unsigned int irq, cpumask_t mask, struct msi_msg *msg)
2498 {
2499         int vector;
2500         unsigned dest;
2501
2502         vector = assign_irq_vector(irq);
2503         if (vector > 0) {
2504                 dest = cpu_mask_to_apicid(mask);
2505
2506                 msg->data &= ~MSI_DATA_VECTOR_MASK;
2507                 msg->data |= MSI_DATA_VECTOR(vector);
2508                 msg->address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2509                 msg->address_lo |= MSI_ADDR_DEST_ID(dest);
2510         }
2511 }
2512
2513 struct msi_ops arch_msi_ops = {
2514         .needs_64bit_address = 0,
2515         .setup = msi_msg_setup,
2516         .teardown = msi_msg_teardown,
2517         .target = msi_msg_set_affinity,
2518 };
2519
2520 #endif /* CONFIG_PCI_MSI */
2521
2522 /*
2523  * Hypertransport interrupt support
2524  */
2525 #ifdef CONFIG_HT_IRQ
2526
2527 #ifdef CONFIG_SMP
2528
2529 static void target_ht_irq(unsigned int irq, unsigned int dest)
2530 {
2531         u32 low, high;
2532         low  = read_ht_irq_low(irq);
2533         high = read_ht_irq_high(irq);
2534
2535         low  &= ~(HT_IRQ_LOW_DEST_ID_MASK);
2536         high &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2537
2538         low  |= HT_IRQ_LOW_DEST_ID(dest);
2539         high |= HT_IRQ_HIGH_DEST_ID(dest);
2540
2541         write_ht_irq_low(irq, low);
2542         write_ht_irq_high(irq, high);
2543 }
2544
2545 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2546 {
2547         unsigned int dest;
2548         cpumask_t tmp;
2549
2550         cpus_and(tmp, mask, cpu_online_map);
2551         if (cpus_empty(tmp))
2552                 tmp = TARGET_CPUS;
2553
2554         cpus_and(mask, tmp, CPU_MASK_ALL);
2555
2556         dest = cpu_mask_to_apicid(mask);
2557
2558         target_ht_irq(irq, dest);
2559         set_native_irq_info(irq, mask);
2560 }
2561 #endif
2562
2563 static struct hw_interrupt_type ht_irq_chip = {
2564         .name           = "PCI-HT",
2565         .mask           = mask_ht_irq,
2566         .unmask         = unmask_ht_irq,
2567         .ack            = ack_ioapic_irq,
2568 #ifdef CONFIG_SMP
2569         .set_affinity   = set_ht_irq_affinity,
2570 #endif
2571         .retrigger      = ioapic_retrigger_irq,
2572 };
2573
2574 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2575 {
2576         int vector;
2577
2578         vector = assign_irq_vector(irq);
2579         if (vector >= 0) {
2580                 u32 low, high;
2581                 unsigned dest;
2582                 cpumask_t tmp;
2583
2584                 cpus_clear(tmp);
2585                 cpu_set(vector >> 8, tmp);
2586                 dest = cpu_mask_to_apicid(tmp);
2587
2588                 high =  HT_IRQ_HIGH_DEST_ID(dest);
2589
2590                 low =   HT_IRQ_LOW_BASE |
2591                         HT_IRQ_LOW_DEST_ID(dest) |
2592                         HT_IRQ_LOW_VECTOR(vector) |
2593                         ((INT_DEST_MODE == 0) ?
2594                                 HT_IRQ_LOW_DM_PHYSICAL :
2595                                 HT_IRQ_LOW_DM_LOGICAL) |
2596                         HT_IRQ_LOW_RQEOI_EDGE |
2597                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2598                                 HT_IRQ_LOW_MT_FIXED :
2599                                 HT_IRQ_LOW_MT_ARBITRATED) |
2600                         HT_IRQ_LOW_IRQ_MASKED;
2601
2602                 write_ht_irq_low(irq, low);
2603                 write_ht_irq_high(irq, high);
2604
2605                 set_irq_chip_and_handler(irq, &ht_irq_chip, handle_edge_irq);
2606         }
2607         return vector;
2608 }
2609 #endif /* CONFIG_HT_IRQ */
2610
2611 /* --------------------------------------------------------------------------
2612                           ACPI-based IOAPIC Configuration
2613    -------------------------------------------------------------------------- */
2614
2615 #ifdef CONFIG_ACPI
2616
2617 int __init io_apic_get_unique_id (int ioapic, int apic_id)
2618 {
2619         union IO_APIC_reg_00 reg_00;
2620         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2621         physid_mask_t tmp;
2622         unsigned long flags;
2623         int i = 0;
2624
2625         /*
2626          * The P4 platform supports up to 256 APIC IDs on two separate APIC
2627          * buses (one for LAPICs, one for IOAPICs), where predecessors only
2628          * supports up to 16 on one shared APIC bus.
2629          *
2630          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2631          *      advantage of new APIC bus architecture.
2632          */
2633
2634         if (physids_empty(apic_id_map))
2635                 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2636
2637         spin_lock_irqsave(&ioapic_lock, flags);
2638         reg_00.raw = io_apic_read(ioapic, 0);
2639         spin_unlock_irqrestore(&ioapic_lock, flags);
2640
2641         if (apic_id >= get_physical_broadcast()) {
2642                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2643                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
2644                 apic_id = reg_00.bits.ID;
2645         }
2646
2647         /*
2648          * Every APIC in a system must have a unique ID or we get lots of nice
2649          * 'stuck on smp_invalidate_needed IPI wait' messages.
2650          */
2651         if (check_apicid_used(apic_id_map, apic_id)) {
2652
2653                 for (i = 0; i < get_physical_broadcast(); i++) {
2654                         if (!check_apicid_used(apic_id_map, i))
2655                                 break;
2656                 }
2657
2658                 if (i == get_physical_broadcast())
2659                         panic("Max apic_id exceeded!\n");
2660
2661                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2662                         "trying %d\n", ioapic, apic_id, i);
2663
2664                 apic_id = i;
2665         }
2666
2667         tmp = apicid_to_cpu_present(apic_id);
2668         physids_or(apic_id_map, apic_id_map, tmp);
2669
2670         if (reg_00.bits.ID != apic_id) {
2671                 reg_00.bits.ID = apic_id;
2672
2673                 spin_lock_irqsave(&ioapic_lock, flags);
2674                 io_apic_write(ioapic, 0, reg_00.raw);
2675                 reg_00.raw = io_apic_read(ioapic, 0);
2676                 spin_unlock_irqrestore(&ioapic_lock, flags);
2677
2678                 /* Sanity check */
2679                 if (reg_00.bits.ID != apic_id) {
2680                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
2681                         return -1;
2682                 }
2683         }
2684
2685         apic_printk(APIC_VERBOSE, KERN_INFO
2686                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2687
2688         return apic_id;
2689 }
2690
2691
2692 int __init io_apic_get_version (int ioapic)
2693 {
2694         union IO_APIC_reg_01    reg_01;
2695         unsigned long flags;
2696
2697         spin_lock_irqsave(&ioapic_lock, flags);
2698         reg_01.raw = io_apic_read(ioapic, 1);
2699         spin_unlock_irqrestore(&ioapic_lock, flags);
2700
2701         return reg_01.bits.version;
2702 }
2703
2704
2705 int __init io_apic_get_redir_entries (int ioapic)
2706 {
2707         union IO_APIC_reg_01    reg_01;
2708         unsigned long flags;
2709
2710         spin_lock_irqsave(&ioapic_lock, flags);
2711         reg_01.raw = io_apic_read(ioapic, 1);
2712         spin_unlock_irqrestore(&ioapic_lock, flags);
2713
2714         return reg_01.bits.entries;
2715 }
2716
2717
2718 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2719 {
2720         struct IO_APIC_route_entry entry;
2721         unsigned long flags;
2722
2723         if (!IO_APIC_IRQ(irq)) {
2724                 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2725                         ioapic);
2726                 return -EINVAL;
2727         }
2728
2729         /*
2730          * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2731          * Note that we mask (disable) IRQs now -- these get enabled when the
2732          * corresponding device driver registers for this IRQ.
2733          */
2734
2735         memset(&entry,0,sizeof(entry));
2736
2737         entry.delivery_mode = INT_DELIVERY_MODE;
2738         entry.dest_mode = INT_DEST_MODE;
2739         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2740         entry.trigger = edge_level;
2741         entry.polarity = active_high_low;
2742         entry.mask  = 1;
2743
2744         /*
2745          * IRQs < 16 are already in the irq_2_pin[] map
2746          */
2747         if (irq >= 16)
2748                 add_pin_to_irq(irq, ioapic, pin);
2749
2750         entry.vector = assign_irq_vector(irq);
2751
2752         apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2753                 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2754                 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2755                 edge_level, active_high_low);
2756
2757         ioapic_register_intr(irq, entry.vector, edge_level);
2758
2759         if (!ioapic && (irq < 16))
2760                 disable_8259A_irq(irq);
2761
2762         ioapic_write_entry(ioapic, pin, entry);
2763         spin_lock_irqsave(&ioapic_lock, flags);
2764         set_native_irq_info(irq, TARGET_CPUS);
2765         spin_unlock_irqrestore(&ioapic_lock, flags);
2766
2767         return 0;
2768 }
2769
2770 #endif /* CONFIG_ACPI */
2771
2772 static int __init parse_disable_timer_pin_1(char *arg)
2773 {
2774         disable_timer_pin_1 = 1;
2775         return 0;
2776 }
2777 early_param("disable_timer_pin_1", parse_disable_timer_pin_1);
2778
2779 static int __init parse_enable_timer_pin_1(char *arg)
2780 {
2781         disable_timer_pin_1 = -1;
2782         return 0;
2783 }
2784 early_param("enable_timer_pin_1", parse_enable_timer_pin_1);
2785
2786 static int __init parse_noapic(char *arg)
2787 {
2788         /* disable IO-APIC */
2789         disable_ioapic_setup();
2790         return 0;
2791 }
2792 early_param("noapic", parse_noapic);