arch/x86/kernel/io_apic_32.c

   1 /*
   2  *      Intel IO-APIC support for multi-Pentium hosts.
   3  *
   4  *      Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
   5  *
   6  *      Many thanks to Stig Venaas for trying out countless experimental
   7  *      patches and reporting/debugging problems patiently!
   8  *
   9  *      (c) 1999, Multiple IO-APIC support, developed by
  10  *      Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
  11  *      Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
  12  *      further tested and cleaned up by Zach Brown <zab@redhat.com>
  13  *      and Ingo Molnar <mingo@redhat.com>
  14  *
  15  *      Fixes
  16  *      Maciej W. Rozycki       :       Bits for genuine 82489DX APICs;
  17  *                                      thanks to Eric Gilmore
  18  *                                      and Rolf G. Tews
  19  *                                      for testing these extensively
  20  *      Paul Diefenbaugh        :       Added full ACPI support
  21  */
  22
  23 #include <linux/mm.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/init.h>
  26 #include <linux/delay.h>
  27 #include <linux/sched.h>
  28 #include <linux/mc146818rtc.h>
  29 #include <linux/compiler.h>
  30 #include <linux/acpi.h>
  31 #include <linux/module.h>
  32 #include <linux/sysdev.h>
  33 #include <linux/pci.h>
  34 #include <linux/msi.h>
  35 #include <linux/htirq.h>
  36 #include <linux/freezer.h>
  37 #include <linux/kthread.h>
  38
  39 #include <asm/io.h>
  40 #include <asm/smp.h>
  41 #include <asm/desc.h>
  42 #include <asm/timer.h>
  43 #include <asm/i8259.h>
  44 #include <asm/nmi.h>
  45 #include <asm/msidef.h>
  46 #include <asm/hypertransport.h>
  47
  48 #include <mach_apic.h>
  49 #include <mach_apicdef.h>
  50
  51 int (*ioapic_renumber_irq)(int ioapic, int irq);
  52 atomic_t irq_mis_count;
  53
  54 /* Where if anywhere is the i8259 connect in external int mode */
  55 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
  56
  57 static DEFINE_SPINLOCK(ioapic_lock);
  58 static DEFINE_SPINLOCK(vector_lock);
  59
  60 int timer_over_8254 __initdata = 1;
  61
  62 /*
  63  *      Is the SiS APIC rmw bug present ?
  64  *      -1 = don't know, 0 = no, 1 = yes
  65  */
  66 int sis_apic_bug = -1;
  67
  68 /*
  69  * # of IRQ routing registers
  70  */
  71 int nr_ioapic_registers[MAX_IO_APICS];
  72
  73 static int disable_timer_pin_1 __initdata;
  74
  75 /*
  76  * Rough estimation of how many shared IRQs there are, can
  77  * be changed anytime.
  78  */
  79 #define MAX_PLUS_SHARED_IRQS NR_IRQS
  80 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
  81
  82 /*
  83  * This is performance-critical, we want to do it O(1)
  84  *
  85  * the indexing order of this array favors 1:1 mappings
  86  * between pins and IRQs.
  87  */
  88
  89 static struct irq_pin_list {
  90         int apic, pin, next;
  91 } irq_2_pin[PIN_MAP_SIZE];
  92
  93 struct io_apic {
  94         unsigned int index;
  95         unsigned int unused[3];
  96         unsigned int data;
  97 };
  98
  99 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
 100 {
 101         return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
 102                 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
 103 }
 104
 105 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
 106 {
 107         struct io_apic __iomem *io_apic = io_apic_base(apic);
 108         writel(reg, &io_apic->index);
 109         return readl(&io_apic->data);
 110 }
 111
 112 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
 113 {
 114         struct io_apic __iomem *io_apic = io_apic_base(apic);
 115         writel(reg, &io_apic->index);
 116         writel(value, &io_apic->data);
 117 }
 118
 119 /*
 120  * Re-write a value: to be used for read-modify-write
 121  * cycles where the read already set up the index register.
 122  *
 123  * Older SiS APIC requires we rewrite the index register
 124  */
 125 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
 126 {
 127         volatile struct io_apic __iomem *io_apic = io_apic_base(apic);
 128         if (sis_apic_bug)
 129                 writel(reg, &io_apic->index);
 130         writel(value, &io_apic->data);
 131 }
 132
 133 union entry_union {
 134         struct { u32 w1, w2; };
 135         struct IO_APIC_route_entry entry;
 136 };
 137
 138 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
 139 {
 140         union entry_union eu;
 141         unsigned long flags;
 142         spin_lock_irqsave(&ioapic_lock, flags);
 143         eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
 144         eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
 145         spin_unlock_irqrestore(&ioapic_lock, flags);
 146         return eu.entry;
 147 }
 148
 149 /*
 150  * When we write a new IO APIC routing entry, we need to write the high
 151  * word first! If the mask bit in the low word is clear, we will enable
 152  * the interrupt, and we need to make sure the entry is fully populated
 153  * before that happens.
 154  */
 155 static void
 156 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 157 {
 158         union entry_union eu;
 159         eu.entry = e;
 160         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 161         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 162 }
 163
 164 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
 165 {
 166         unsigned long flags;
 167         spin_lock_irqsave(&ioapic_lock, flags);
 168         __ioapic_write_entry(apic, pin, e);
 169         spin_unlock_irqrestore(&ioapic_lock, flags);
 170 }
 171
 172 /*
 173  * When we mask an IO APIC routing entry, we need to write the low
 174  * word first, in order to set the mask bit before we change the
 175  * high bits!
 176  */
 177 static void ioapic_mask_entry(int apic, int pin)
 178 {
 179         unsigned long flags;
 180         union entry_union eu = { .entry.mask = 1 };
 181
 182         spin_lock_irqsave(&ioapic_lock, flags);
 183         io_apic_write(apic, 0x10 + 2*pin, eu.w1);
 184         io_apic_write(apic, 0x11 + 2*pin, eu.w2);
 185         spin_unlock_irqrestore(&ioapic_lock, flags);
 186 }
 187
 188 /*
 189  * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
 190  * shared ISA-space IRQs, so we have to support them. We are super
 191  * fast in the common case, and fast for shared ISA-space IRQs.
 192  */
 193 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
 194 {
 195         static int first_free_entry = NR_IRQS;
 196         struct irq_pin_list *entry = irq_2_pin + irq;
 197
 198         while (entry->next)
 199                 entry = irq_2_pin + entry->next;
 200
 201         if (entry->pin != -1) {
 202                 entry->next = first_free_entry;
 203                 entry = irq_2_pin + entry->next;
 204                 if (++first_free_entry >= PIN_MAP_SIZE)
 205                         panic("io_apic.c: whoops");
 206         }
 207         entry->apic = apic;
 208         entry->pin = pin;
 209 }
 210
 211 /*
 212  * Reroute an IRQ to a different pin.
 213  */
 214 static void __init replace_pin_at_irq(unsigned int irq,
 215                                       int oldapic, int oldpin,
 216                                       int newapic, int newpin)
 217 {
 218         struct irq_pin_list *entry = irq_2_pin + irq;
 219
 220         while (1) {
 221                 if (entry->apic == oldapic && entry->pin == oldpin) {
 222                         entry->apic = newapic;
 223                         entry->pin = newpin;
 224                 }
 225                 if (!entry->next)
 226                         break;
 227                 entry = irq_2_pin + entry->next;
 228         }
 229 }
 230
 231 static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable)
 232 {
 233         struct irq_pin_list *entry = irq_2_pin + irq;
 234         unsigned int pin, reg;
 235
 236         for (;;) {
 237                 pin = entry->pin;
 238                 if (pin == -1)
 239                         break;
 240                 reg = io_apic_read(entry->apic, 0x10 + pin*2);
 241                 reg &= ~disable;
 242                 reg |= enable;
 243                 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
 244                 if (!entry->next)
 245                         break;
 246                 entry = irq_2_pin + entry->next;
 247         }
 248 }
 249
 250 /* mask = 1 */
 251 static void __mask_IO_APIC_irq (unsigned int irq)
 252 {
 253         __modify_IO_APIC_irq(irq, 0x00010000, 0);
 254 }
 255
 256 /* mask = 0 */
 257 static void __unmask_IO_APIC_irq (unsigned int irq)
 258 {
 259         __modify_IO_APIC_irq(irq, 0, 0x00010000);
 260 }
 261
 262 /* mask = 1, trigger = 0 */
 263 static void __mask_and_edge_IO_APIC_irq (unsigned int irq)
 264 {
 265         __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000);
 266 }
 267
 268 /* mask = 0, trigger = 1 */
 269 static void __unmask_and_level_IO_APIC_irq (unsigned int irq)
 270 {
 271         __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000);
 272 }
 273
 274 static void mask_IO_APIC_irq (unsigned int irq)
 275 {
 276         unsigned long flags;
 277
 278         spin_lock_irqsave(&ioapic_lock, flags);
 279         __mask_IO_APIC_irq(irq);
 280         spin_unlock_irqrestore(&ioapic_lock, flags);
 281 }
 282
 283 static void unmask_IO_APIC_irq (unsigned int irq)
 284 {
 285         unsigned long flags;
 286
 287         spin_lock_irqsave(&ioapic_lock, flags);
 288         __unmask_IO_APIC_irq(irq);
 289         spin_unlock_irqrestore(&ioapic_lock, flags);
 290 }
 291
 292 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
 293 {
 294         struct IO_APIC_route_entry entry;
 295
 296         /* Check delivery_mode to be sure we're not clearing an SMI pin */
 297         entry = ioapic_read_entry(apic, pin);
 298         if (entry.delivery_mode == dest_SMI)
 299                 return;
 300
 301         /*
 302          * Disable it in the IO-APIC irq-routing table:
 303          */
 304         ioapic_mask_entry(apic, pin);
 305 }
 306
 307 static void clear_IO_APIC (void)
 308 {
 309         int apic, pin;
 310
 311         for (apic = 0; apic < nr_ioapics; apic++)
 312                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
 313                         clear_IO_APIC_pin(apic, pin);
 314 }
 315
 316 #ifdef CONFIG_SMP
 317 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
 318 {
 319         unsigned long flags;
 320         int pin;
 321         struct irq_pin_list *entry = irq_2_pin + irq;
 322         unsigned int apicid_value;
 323         cpumask_t tmp;
 324
 325         cpus_and(tmp, cpumask, cpu_online_map);
 326         if (cpus_empty(tmp))
 327                 tmp = TARGET_CPUS;
 328
 329         cpus_and(cpumask, tmp, CPU_MASK_ALL);
 330
 331         apicid_value = cpu_mask_to_apicid(cpumask);
 332         /* Prepare to do the io_apic_write */
 333         apicid_value = apicid_value << 24;
 334         spin_lock_irqsave(&ioapic_lock, flags);
 335         for (;;) {
 336                 pin = entry->pin;
 337                 if (pin == -1)
 338                         break;
 339                 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
 340                 if (!entry->next)
 341                         break;
 342                 entry = irq_2_pin + entry->next;
 343         }
 344         irq_desc[irq].affinity = cpumask;
 345         spin_unlock_irqrestore(&ioapic_lock, flags);
 346 }
 347
 348 #if defined(CONFIG_IRQBALANCE)
 349 # include <asm/processor.h>     /* kernel_thread() */
 350 # include <linux/kernel_stat.h> /* kstat */
 351 # include <linux/slab.h>                /* kmalloc() */
 352 # include <linux/timer.h>       /* time_after() */
 353
 354 #define IRQBALANCE_CHECK_ARCH -999
 355 #define MAX_BALANCED_IRQ_INTERVAL       (5*HZ)
 356 #define MIN_BALANCED_IRQ_INTERVAL       (HZ/2)
 357 #define BALANCED_IRQ_MORE_DELTA         (HZ/10)
 358 #define BALANCED_IRQ_LESS_DELTA         (HZ)
 359
 360 static int irqbalance_disabled __read_mostly = IRQBALANCE_CHECK_ARCH;
 361 static int physical_balance __read_mostly;
 362 static long balanced_irq_interval __read_mostly = MAX_BALANCED_IRQ_INTERVAL;
 363
 364 static struct irq_cpu_info {
 365         unsigned long * last_irq;
 366         unsigned long * irq_delta;
 367         unsigned long irq;
 368 } irq_cpu_data[NR_CPUS];
 369
 370 #define CPU_IRQ(cpu)            (irq_cpu_data[cpu].irq)
 371 #define LAST_CPU_IRQ(cpu,irq)   (irq_cpu_data[cpu].last_irq[irq])
 372 #define IRQ_DELTA(cpu,irq)      (irq_cpu_data[cpu].irq_delta[irq])
 373
 374 #define IDLE_ENOUGH(cpu,now) \
 375         (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
 376
 377 #define IRQ_ALLOWED(cpu, allowed_mask)  cpu_isset(cpu, allowed_mask)
 378
 379 #define CPU_TO_PACKAGEINDEX(i) (first_cpu(per_cpu(cpu_sibling_map, i)))
 380
 381 static cpumask_t balance_irq_affinity[NR_IRQS] = {
 382         [0 ... NR_IRQS-1] = CPU_MASK_ALL
 383 };
 384
 385 void set_balance_irq_affinity(unsigned int irq, cpumask_t mask)
 386 {
 387         balance_irq_affinity[irq] = mask;
 388 }
 389
 390 static unsigned long move(int curr_cpu, cpumask_t allowed_mask,
 391                         unsigned long now, int direction)
 392 {
 393         int search_idle = 1;
 394         int cpu = curr_cpu;
 395
 396         goto inside;
 397
 398         do {
 399                 if (unlikely(cpu == curr_cpu))
 400                         search_idle = 0;
 401 inside:
 402                 if (direction == 1) {
 403                         cpu++;
 404                         if (cpu >= NR_CPUS)
 405                                 cpu = 0;
 406                 } else {
 407                         cpu--;
 408                         if (cpu == -1)
 409                                 cpu = NR_CPUS-1;
 410                 }
 411         } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
 412                         (search_idle && !IDLE_ENOUGH(cpu,now)));
 413
 414         return cpu;
 415 }
 416
 417 static inline void balance_irq(int cpu, int irq)
 418 {
 419         unsigned long now = jiffies;
 420         cpumask_t allowed_mask;
 421         unsigned int new_cpu;
 422
 423         if (irqbalance_disabled)
 424                 return;
 425
 426         cpus_and(allowed_mask, cpu_online_map, balance_irq_affinity[irq]);
 427         new_cpu = move(cpu, allowed_mask, now, 1);
 428         if (cpu != new_cpu) {
 429                 set_pending_irq(irq, cpumask_of_cpu(new_cpu));
 430         }
 431 }
 432
 433 static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
 434 {
 435         int i, j;
 436
 437         for_each_online_cpu(i) {
 438                 for (j = 0; j < NR_IRQS; j++) {
 439                         if (!irq_desc[j].action)
 440                                 continue;
 441                         /* Is it a significant load ?  */
 442                         if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) <
 443                                                 useful_load_threshold)
 444                                 continue;
 445                         balance_irq(i, j);
 446                 }
 447         }
 448         balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 449                 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 450         return;
 451 }
 452
 453 static void do_irq_balance(void)
 454 {
 455         int i, j;
 456         unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
 457         unsigned long move_this_load = 0;
 458         int max_loaded = 0, min_loaded = 0;
 459         int load;
 460         unsigned long useful_load_threshold = balanced_irq_interval + 10;
 461         int selected_irq;
 462         int tmp_loaded, first_attempt = 1;
 463         unsigned long tmp_cpu_irq;
 464         unsigned long imbalance = 0;
 465         cpumask_t allowed_mask, target_cpu_mask, tmp;
 466
 467         for_each_possible_cpu(i) {
 468                 int package_index;
 469                 CPU_IRQ(i) = 0;
 470                 if (!cpu_online(i))
 471                         continue;
 472                 package_index = CPU_TO_PACKAGEINDEX(i);
 473                 for (j = 0; j < NR_IRQS; j++) {
 474                         unsigned long value_now, delta;
 475                         /* Is this an active IRQ or balancing disabled ? */
 476                         if (!irq_desc[j].action || irq_balancing_disabled(j))
 477                                 continue;
 478                         if ( package_index == i )
 479                                 IRQ_DELTA(package_index,j) = 0;
 480                         /* Determine the total count per processor per IRQ */
 481                         value_now = (unsigned long) kstat_cpu(i).irqs[j];
 482
 483                         /* Determine the activity per processor per IRQ */
 484                         delta = value_now - LAST_CPU_IRQ(i,j);
 485
 486                         /* Update last_cpu_irq[][] for the next time */
 487                         LAST_CPU_IRQ(i,j) = value_now;
 488
 489                         /* Ignore IRQs whose rate is less than the clock */
 490                         if (delta < useful_load_threshold)
 491                                 continue;
 492                         /* update the load for the processor or package total */
 493                         IRQ_DELTA(package_index,j) += delta;
 494
 495                         /* Keep track of the higher numbered sibling as well */
 496                         if (i != package_index)
 497                                 CPU_IRQ(i) += delta;
 498                         /*
 499                          * We have sibling A and sibling B in the package
 500                          *
 501                          * cpu_irq[A] = load for cpu A + load for cpu B
 502                          * cpu_irq[B] = load for cpu B
 503                          */
 504                         CPU_IRQ(package_index) += delta;
 505                 }
 506         }
 507         /* Find the least loaded processor package */
 508         for_each_online_cpu(i) {
 509                 if (i != CPU_TO_PACKAGEINDEX(i))
 510                         continue;
 511                 if (min_cpu_irq > CPU_IRQ(i)) {
 512                         min_cpu_irq = CPU_IRQ(i);
 513                         min_loaded = i;
 514                 }
 515         }
 516         max_cpu_irq = ULONG_MAX;
 517
 518 tryanothercpu:
 519         /* Look for heaviest loaded processor.
 520          * We may come back to get the next heaviest loaded processor.
 521          * Skip processors with trivial loads.
 522          */
 523         tmp_cpu_irq = 0;
 524         tmp_loaded = -1;
 525         for_each_online_cpu(i) {
 526                 if (i != CPU_TO_PACKAGEINDEX(i))
 527                         continue;
 528                 if (max_cpu_irq <= CPU_IRQ(i))
 529                         continue;
 530                 if (tmp_cpu_irq < CPU_IRQ(i)) {
 531                         tmp_cpu_irq = CPU_IRQ(i);
 532                         tmp_loaded = i;
 533                 }
 534         }
 535
 536         if (tmp_loaded == -1) {
 537          /* In the case of small number of heavy interrupt sources,
 538           * loading some of the cpus too much. We use Ingo's original
 539           * approach to rotate them around.
 540           */
 541                 if (!first_attempt && imbalance >= useful_load_threshold) {
 542                         rotate_irqs_among_cpus(useful_load_threshold);
 543                         return;
 544                 }
 545                 goto not_worth_the_effort;
 546         }
 547
 548         first_attempt = 0;              /* heaviest search */
 549         max_cpu_irq = tmp_cpu_irq;      /* load */
 550         max_loaded = tmp_loaded;        /* processor */
 551         imbalance = (max_cpu_irq - min_cpu_irq) / 2;
 552
 553         /* if imbalance is less than approx 10% of max load, then
 554          * observe diminishing returns action. - quit
 555          */
 556         if (imbalance < (max_cpu_irq >> 3))
 557                 goto not_worth_the_effort;
 558
 559 tryanotherirq:
 560         /* if we select an IRQ to move that can't go where we want, then
 561          * see if there is another one to try.
 562          */
 563         move_this_load = 0;
 564         selected_irq = -1;
 565         for (j = 0; j < NR_IRQS; j++) {
 566                 /* Is this an active IRQ? */
 567                 if (!irq_desc[j].action)
 568                         continue;
 569                 if (imbalance <= IRQ_DELTA(max_loaded,j))
 570                         continue;
 571                 /* Try to find the IRQ that is closest to the imbalance
 572                  * without going over.
 573                  */
 574                 if (move_this_load < IRQ_DELTA(max_loaded,j)) {
 575                         move_this_load = IRQ_DELTA(max_loaded,j);
 576                         selected_irq = j;
 577                 }
 578         }
 579         if (selected_irq == -1) {
 580                 goto tryanothercpu;
 581         }
 582
 583         imbalance = move_this_load;
 584
 585         /* For physical_balance case, we accumulated both load
 586          * values in the one of the siblings cpu_irq[],
 587          * to use the same code for physical and logical processors
 588          * as much as possible.
 589          *
 590          * NOTE: the cpu_irq[] array holds the sum of the load for
 591          * sibling A and sibling B in the slot for the lowest numbered
 592          * sibling (A), _AND_ the load for sibling B in the slot for
 593          * the higher numbered sibling.
 594          *
 595          * We seek the least loaded sibling by making the comparison
 596          * (A+B)/2 vs B
 597          */
 598         load = CPU_IRQ(min_loaded) >> 1;
 599         for_each_cpu_mask(j, per_cpu(cpu_sibling_map, min_loaded)) {
 600                 if (load > CPU_IRQ(j)) {
 601                         /* This won't change cpu_sibling_map[min_loaded] */
 602                         load = CPU_IRQ(j);
 603                         min_loaded = j;
 604                 }
 605         }
 606
 607         cpus_and(allowed_mask,
 608                 cpu_online_map,
 609                 balance_irq_affinity[selected_irq]);
 610         target_cpu_mask = cpumask_of_cpu(min_loaded);
 611         cpus_and(tmp, target_cpu_mask, allowed_mask);
 612
 613         if (!cpus_empty(tmp)) {
 614                 /* mark for change destination */
 615                 set_pending_irq(selected_irq, cpumask_of_cpu(min_loaded));
 616
 617                 /* Since we made a change, come back sooner to
 618                  * check for more variation.
 619                  */
 620                 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
 621                         balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
 622                 return;
 623         }
 624         goto tryanotherirq;
 625
 626 not_worth_the_effort:
 627         /*
 628          * if we did not find an IRQ to move, then adjust the time interval
 629          * upward
 630          */
 631         balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
 632                 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
 633         return;
 634 }
 635
 636 static int balanced_irq(void *unused)
 637 {
 638         int i;
 639         unsigned long prev_balance_time = jiffies;
 640         long time_remaining = balanced_irq_interval;
 641
 642         /* push everything to CPU 0 to give us a starting point.  */
 643         for (i = 0 ; i < NR_IRQS ; i++) {
 644                 irq_desc[i].pending_mask = cpumask_of_cpu(0);
 645                 set_pending_irq(i, cpumask_of_cpu(0));
 646         }
 647
 648         set_freezable();
 649         for ( ; ; ) {
 650                 time_remaining = schedule_timeout_interruptible(time_remaining);
 651                 try_to_freeze();
 652                 if (time_after(jiffies,
 653                                 prev_balance_time+balanced_irq_interval)) {
 654                         preempt_disable();
 655                         do_irq_balance();
 656                         prev_balance_time = jiffies;
 657                         time_remaining = balanced_irq_interval;
 658                         preempt_enable();
 659                 }
 660         }
 661         return 0;
 662 }
 663
 664 static int __init balanced_irq_init(void)
 665 {
 666         int i;
 667         struct cpuinfo_x86 *c;
 668         cpumask_t tmp;
 669
 670         cpus_shift_right(tmp, cpu_online_map, 2);
 671         c = &boot_cpu_data;
 672         /* When not overwritten by the command line ask subarchitecture. */
 673         if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
 674                 irqbalance_disabled = NO_BALANCE_IRQ;
 675         if (irqbalance_disabled)
 676                 return 0;
 677
 678          /* disable irqbalance completely if there is only one processor online */
 679         if (num_online_cpus() < 2) {
 680                 irqbalance_disabled = 1;
 681                 return 0;
 682         }
 683         /*
 684          * Enable physical balance only if more than 1 physical processor
 685          * is present
 686          */
 687         if (smp_num_siblings > 1 && !cpus_empty(tmp))
 688                 physical_balance = 1;
 689
 690         for_each_online_cpu(i) {
 691                 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 692                 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
 693                 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
 694                         printk(KERN_ERR "balanced_irq_init: out of memory");
 695                         goto failed;
 696                 }
 697                 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
 698                 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
 699         }
 700
 701         printk(KERN_INFO "Starting balanced_irq\n");
 702         if (!IS_ERR(kthread_run(balanced_irq, NULL, "kirqd")))
 703                 return 0;
 704         printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
 705 failed:
 706         for_each_possible_cpu(i) {
 707                 kfree(irq_cpu_data[i].irq_delta);
 708                 irq_cpu_data[i].irq_delta = NULL;
 709                 kfree(irq_cpu_data[i].last_irq);
 710                 irq_cpu_data[i].last_irq = NULL;
 711         }
 712         return 0;
 713 }
 714
 715 int __devinit irqbalance_disable(char *str)
 716 {
 717         irqbalance_disabled = 1;
 718         return 1;
 719 }
 720
 721 __setup("noirqbalance", irqbalance_disable);
 722
 723 late_initcall(balanced_irq_init);
 724 #endif /* CONFIG_IRQBALANCE */
 725 #endif /* CONFIG_SMP */
 726
 727 #ifndef CONFIG_SMP
 728 void send_IPI_self(int vector)
 729 {
 730         unsigned int cfg;
 731
 732         /*
 733          * Wait for idle.
 734          */
 735         apic_wait_icr_idle();
 736         cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
 737         /*
 738          * Send the IPI. The write to APIC_ICR fires this off.
 739          */
 740         apic_write_around(APIC_ICR, cfg);
 741 }
 742 #endif /* !CONFIG_SMP */
 743
 744
 745 /*
 746  * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
 747  * specific CPU-side IRQs.
 748  */
 749
 750 #define MAX_PIRQS 8
 751 static int pirq_entries [MAX_PIRQS];
 752 static int pirqs_enabled;
 753 int skip_ioapic_setup;
 754
 755 static int __init ioapic_pirq_setup(char *str)
 756 {
 757         int i, max;
 758         int ints[MAX_PIRQS+1];
 759
 760         get_options(str, ARRAY_SIZE(ints), ints);
 761
 762         for (i = 0; i < MAX_PIRQS; i++)
 763                 pirq_entries[i] = -1;
 764
 765         pirqs_enabled = 1;
 766         apic_printk(APIC_VERBOSE, KERN_INFO
 767                         "PIRQ redirection, working around broken MP-BIOS.\n");
 768         max = MAX_PIRQS;
 769         if (ints[0] < MAX_PIRQS)
 770                 max = ints[0];
 771
 772         for (i = 0; i < max; i++) {
 773                 apic_printk(APIC_VERBOSE, KERN_DEBUG
 774                                 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
 775                 /*
 776                  * PIRQs are mapped upside down, usually.
 777                  */
 778                 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
 779         }
 780         return 1;
 781 }
 782
 783 __setup("pirq=", ioapic_pirq_setup);
 784
 785 /*
 786  * Find the IRQ entry number of a certain pin.
 787  */
 788 static int find_irq_entry(int apic, int pin, int type)
 789 {
 790         int i;
 791
 792         for (i = 0; i < mp_irq_entries; i++)
 793                 if (mp_irqs[i].mpc_irqtype == type &&
 794                     (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
 795                      mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
 796                     mp_irqs[i].mpc_dstirq == pin)
 797                         return i;
 798
 799         return -1;
 800 }
 801
 802 /*
 803  * Find the pin to which IRQ[irq] (ISA) is connected
 804  */
 805 static int __init find_isa_irq_pin(int irq, int type)
 806 {
 807         int i;
 808
 809         for (i = 0; i < mp_irq_entries; i++) {
 810                 int lbus = mp_irqs[i].mpc_srcbus;
 811
 812                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
 813                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
 814                      mp_bus_id_to_type[lbus] == MP_BUS_MCA
 815                     ) &&
 816                     (mp_irqs[i].mpc_irqtype == type) &&
 817                     (mp_irqs[i].mpc_srcbusirq == irq))
 818
 819                         return mp_irqs[i].mpc_dstirq;
 820         }
 821         return -1;
 822 }
 823
 824 static int __init find_isa_irq_apic(int irq, int type)
 825 {
 826         int i;
 827
 828         for (i = 0; i < mp_irq_entries; i++) {
 829                 int lbus = mp_irqs[i].mpc_srcbus;
 830
 831                 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
 832                      mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
 833                      mp_bus_id_to_type[lbus] == MP_BUS_MCA
 834                     ) &&
 835                     (mp_irqs[i].mpc_irqtype == type) &&
 836                     (mp_irqs[i].mpc_srcbusirq == irq))
 837                         break;
 838         }
 839         if (i < mp_irq_entries) {
 840                 int apic;
 841                 for(apic = 0; apic < nr_ioapics; apic++) {
 842                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
 843                                 return apic;
 844                 }
 845         }
 846
 847         return -1;
 848 }
 849
 850 /*
 851  * Find a specific PCI IRQ entry.
 852  * Not an __init, possibly needed by modules
 853  */
 854 static int pin_2_irq(int idx, int apic, int pin);
 855
 856 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
 857 {
 858         int apic, i, best_guess = -1;
 859
 860         apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
 861                 "slot:%d, pin:%d.\n", bus, slot, pin);
 862         if (mp_bus_id_to_pci_bus[bus] == -1) {
 863                 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
 864                 return -1;
 865         }
 866         for (i = 0; i < mp_irq_entries; i++) {
 867                 int lbus = mp_irqs[i].mpc_srcbus;
 868
 869                 for (apic = 0; apic < nr_ioapics; apic++)
 870                         if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
 871                             mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
 872                                 break;
 873
 874                 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
 875                     !mp_irqs[i].mpc_irqtype &&
 876                     (bus == lbus) &&
 877                     (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
 878                         int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
 879
 880                         if (!(apic || IO_APIC_IRQ(irq)))
 881                                 continue;
 882
 883                         if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
 884                                 return irq;
 885                         /*
 886                          * Use the first all-but-pin matching entry as a
 887                          * best-guess fuzzy result for broken mptables.
 888                          */
 889                         if (best_guess < 0)
 890                                 best_guess = irq;
 891                 }
 892         }
 893         return best_guess;
 894 }
 895 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
 896
 897 /*
 898  * This function currently is only a helper for the i386 smp boot process where
 899  * we need to reprogram the ioredtbls to cater for the cpus which have come online
 900  * so mask in all cases should simply be TARGET_CPUS
 901  */
 902 #ifdef CONFIG_SMP
 903 void __init setup_ioapic_dest(void)
 904 {
 905         int pin, ioapic, irq, irq_entry;
 906
 907         if (skip_ioapic_setup == 1)
 908                 return;
 909
 910         for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
 911                 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
 912                         irq_entry = find_irq_entry(ioapic, pin, mp_INT);
 913                         if (irq_entry == -1)
 914                                 continue;
 915                         irq = pin_2_irq(irq_entry, ioapic, pin);
 916                         set_ioapic_affinity_irq(irq, TARGET_CPUS);
 917                 }
 918
 919         }
 920 }
 921 #endif
 922
 923 /*
 924  * EISA Edge/Level control register, ELCR
 925  */
 926 static int EISA_ELCR(unsigned int irq)
 927 {
 928         if (irq < 16) {
 929                 unsigned int port = 0x4d0 + (irq >> 3);
 930                 return (inb(port) >> (irq & 7)) & 1;
 931         }
 932         apic_printk(APIC_VERBOSE, KERN_INFO
 933                         "Broken MPtable reports ISA irq %d\n", irq);
 934         return 0;
 935 }
 936
 937 /* EISA interrupts are always polarity zero and can be edge or level
 938  * trigger depending on the ELCR value.  If an interrupt is listed as
 939  * EISA conforming in the MP table, that means its trigger type must
 940  * be read in from the ELCR */
 941
 942 #define default_EISA_trigger(idx)       (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
 943 #define default_EISA_polarity(idx)      (0)
 944
 945 /* ISA interrupts are always polarity zero edge triggered,
 946  * when listed as conforming in the MP table. */
 947
 948 #define default_ISA_trigger(idx)        (0)
 949 #define default_ISA_polarity(idx)       (0)
 950
 951 /* PCI interrupts are always polarity one level triggered,
 952  * when listed as conforming in the MP table. */
 953
 954 #define default_PCI_trigger(idx)        (1)
 955 #define default_PCI_polarity(idx)       (1)
 956
 957 /* MCA interrupts are always polarity zero level triggered,
 958  * when listed as conforming in the MP table. */
 959
 960 #define default_MCA_trigger(idx)        (1)
 961 #define default_MCA_polarity(idx)       (0)
 962
 963 static int MPBIOS_polarity(int idx)
 964 {
 965         int bus = mp_irqs[idx].mpc_srcbus;
 966         int polarity;
 967
 968         /*
 969          * Determine IRQ line polarity (high active or low active):
 970          */
 971         switch (mp_irqs[idx].mpc_irqflag & 3)
 972         {
 973                 case 0: /* conforms, ie. bus-type dependent polarity */
 974                 {
 975                         switch (mp_bus_id_to_type[bus])
 976                         {
 977                                 case MP_BUS_ISA: /* ISA pin */
 978                                 {
 979                                         polarity = default_ISA_polarity(idx);
 980                                         break;
 981                                 }
 982                                 case MP_BUS_EISA: /* EISA pin */
 983                                 {
 984                                         polarity = default_EISA_polarity(idx);
 985                                         break;
 986                                 }
 987                                 case MP_BUS_PCI: /* PCI pin */
 988                                 {
 989                                         polarity = default_PCI_polarity(idx);
 990                                         break;
 991                                 }
 992                                 case MP_BUS_MCA: /* MCA pin */
 993                                 {
 994                                         polarity = default_MCA_polarity(idx);
 995                                         break;
 996                                 }
 997                                 default:
 998                                 {
 999                                         printk(KERN_WARNING "broken BIOS!!\n");
1000                                         polarity = 1;
1001                                         break;
1002                                 }
1003                         }
1004                         break;
1005                 }
1006                 case 1: /* high active */
1007                 {
1008                         polarity = 0;
1009                         break;
1010                 }
1011                 case 2: /* reserved */
1012                 {
1013                         printk(KERN_WARNING "broken BIOS!!\n");
1014                         polarity = 1;
1015                         break;
1016                 }
1017                 case 3: /* low active */
1018                 {
1019                         polarity = 1;
1020                         break;
1021                 }
1022                 default: /* invalid */
1023                 {
1024                         printk(KERN_WARNING "broken BIOS!!\n");
1025                         polarity = 1;
1026                         break;
1027                 }
1028         }
1029         return polarity;
1030 }
1031
1032 static int MPBIOS_trigger(int idx)
1033 {
1034         int bus = mp_irqs[idx].mpc_srcbus;
1035         int trigger;
1036
1037         /*
1038          * Determine IRQ trigger mode (edge or level sensitive):
1039          */
1040         switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
1041         {
1042                 case 0: /* conforms, ie. bus-type dependent */
1043                 {
1044                         switch (mp_bus_id_to_type[bus])
1045                         {
1046                                 case MP_BUS_ISA: /* ISA pin */
1047                                 {
1048                                         trigger = default_ISA_trigger(idx);
1049                                         break;
1050                                 }
1051                                 case MP_BUS_EISA: /* EISA pin */
1052                                 {
1053                                         trigger = default_EISA_trigger(idx);
1054                                         break;
1055                                 }
1056                                 case MP_BUS_PCI: /* PCI pin */
1057                                 {
1058                                         trigger = default_PCI_trigger(idx);
1059                                         break;
1060                                 }
1061                                 case MP_BUS_MCA: /* MCA pin */
1062                                 {
1063                                         trigger = default_MCA_trigger(idx);
1064                                         break;
1065                                 }
1066                                 default:
1067                                 {
1068                                         printk(KERN_WARNING "broken BIOS!!\n");
1069                                         trigger = 1;
1070                                         break;
1071                                 }
1072                         }
1073                         break;
1074                 }
1075                 case 1: /* edge */
1076                 {
1077                         trigger = 0;
1078                         break;
1079                 }
1080                 case 2: /* reserved */
1081                 {
1082                         printk(KERN_WARNING "broken BIOS!!\n");
1083                         trigger = 1;
1084                         break;
1085                 }
1086                 case 3: /* level */
1087                 {
1088                         trigger = 1;
1089                         break;
1090                 }
1091                 default: /* invalid */
1092                 {
1093                         printk(KERN_WARNING "broken BIOS!!\n");
1094                         trigger = 0;
1095                         break;
1096                 }
1097         }
1098         return trigger;
1099 }
1100
1101 static inline int irq_polarity(int idx)
1102 {
1103         return MPBIOS_polarity(idx);
1104 }
1105
1106 static inline int irq_trigger(int idx)
1107 {
1108         return MPBIOS_trigger(idx);
1109 }
1110
1111 static int pin_2_irq(int idx, int apic, int pin)
1112 {
1113         int irq, i;
1114         int bus = mp_irqs[idx].mpc_srcbus;
1115
1116         /*
1117          * Debugging check, we are in big trouble if this message pops up!
1118          */
1119         if (mp_irqs[idx].mpc_dstirq != pin)
1120                 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1121
1122         switch (mp_bus_id_to_type[bus])
1123         {
1124                 case MP_BUS_ISA: /* ISA pin */
1125                 case MP_BUS_EISA:
1126                 case MP_BUS_MCA:
1127                 {
1128                         irq = mp_irqs[idx].mpc_srcbusirq;
1129                         break;
1130                 }
1131                 case MP_BUS_PCI: /* PCI pin */
1132                 {
1133                         /*
1134                          * PCI IRQs are mapped in order
1135                          */
1136                         i = irq = 0;
1137                         while (i < apic)
1138                                 irq += nr_ioapic_registers[i++];
1139                         irq += pin;
1140
1141                         /*
1142                          * For MPS mode, so far only needed by ES7000 platform
1143                          */
1144                         if (ioapic_renumber_irq)
1145                                 irq = ioapic_renumber_irq(apic, irq);
1146
1147                         break;
1148                 }
1149                 default:
1150                 {
1151                         printk(KERN_ERR "unknown bus type %d.\n",bus);
1152                         irq = 0;
1153                         break;
1154                 }
1155         }
1156
1157         /*
1158          * PCI IRQ command line redirection. Yes, limits are hardcoded.
1159          */
1160         if ((pin >= 16) && (pin <= 23)) {
1161                 if (pirq_entries[pin-16] != -1) {
1162                         if (!pirq_entries[pin-16]) {
1163                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1164                                                 "disabling PIRQ%d\n", pin-16);
1165                         } else {
1166                                 irq = pirq_entries[pin-16];
1167                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1168                                                 "using PIRQ%d -> IRQ %d\n",
1169                                                 pin-16, irq);
1170                         }
1171                 }
1172         }
1173         return irq;
1174 }
1175
1176 static inline int IO_APIC_irq_trigger(int irq)
1177 {
1178         int apic, idx, pin;
1179
1180         for (apic = 0; apic < nr_ioapics; apic++) {
1181                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1182                         idx = find_irq_entry(apic,pin,mp_INT);
1183                         if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1184                                 return irq_trigger(idx);
1185                 }
1186         }
1187         /*
1188          * nonexistent IRQs are edge default
1189          */
1190         return 0;
1191 }
1192
1193 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1194 static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 };
1195
1196 static int __assign_irq_vector(int irq)
1197 {
1198         static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1199         int vector, offset;
1200
1201         BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
1202
1203         if (irq_vector[irq] > 0)
1204                 return irq_vector[irq];
1205
1206         vector = current_vector;
1207         offset = current_offset;
1208 next:
1209         vector += 8;
1210         if (vector >= FIRST_SYSTEM_VECTOR) {
1211                 offset = (offset + 1) % 8;
1212                 vector = FIRST_DEVICE_VECTOR + offset;
1213         }
1214         if (vector == current_vector)
1215                 return -ENOSPC;
1216         if (test_and_set_bit(vector, used_vectors))
1217                 goto next;
1218
1219         current_vector = vector;
1220         current_offset = offset;
1221         irq_vector[irq] = vector;
1222
1223         return vector;
1224 }
1225
1226 static int assign_irq_vector(int irq)
1227 {
1228         unsigned long flags;
1229         int vector;
1230
1231         spin_lock_irqsave(&vector_lock, flags);
1232         vector = __assign_irq_vector(irq);
1233         spin_unlock_irqrestore(&vector_lock, flags);
1234
1235         return vector;
1236 }
1237 static struct irq_chip ioapic_chip;
1238
1239 #define IOAPIC_AUTO     -1
1240 #define IOAPIC_EDGE     0
1241 #define IOAPIC_LEVEL    1
1242
1243 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1244 {
1245         if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1246             trigger == IOAPIC_LEVEL) {
1247                 irq_desc[irq].status |= IRQ_LEVEL;
1248                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1249                                          handle_fasteoi_irq, "fasteoi");
1250         } else {
1251                 irq_desc[irq].status &= ~IRQ_LEVEL;
1252                 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1253                                          handle_edge_irq, "edge");
1254         }
1255         set_intr_gate(vector, interrupt[irq]);
1256 }
1257
1258 static void __init setup_IO_APIC_irqs(void)
1259 {
1260         struct IO_APIC_route_entry entry;
1261         int apic, pin, idx, irq, first_notcon = 1, vector;
1262         unsigned long flags;
1263
1264         apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1265
1266         for (apic = 0; apic < nr_ioapics; apic++) {
1267         for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1268
1269                 /*
1270                  * add it to the IO-APIC irq-routing table:
1271                  */
1272                 memset(&entry,0,sizeof(entry));
1273
1274                 entry.delivery_mode = INT_DELIVERY_MODE;
1275                 entry.dest_mode = INT_DEST_MODE;
1276                 entry.mask = 0;                         /* enable IRQ */
1277                 entry.dest.logical.logical_dest =
1278                                         cpu_mask_to_apicid(TARGET_CPUS);
1279
1280                 idx = find_irq_entry(apic,pin,mp_INT);
1281                 if (idx == -1) {
1282                         if (first_notcon) {
1283                                 apic_printk(APIC_VERBOSE, KERN_DEBUG
1284                                                 " IO-APIC (apicid-pin) %d-%d",
1285                                                 mp_ioapics[apic].mpc_apicid,
1286                                                 pin);
1287                                 first_notcon = 0;
1288                         } else
1289                                 apic_printk(APIC_VERBOSE, ", %d-%d",
1290                                         mp_ioapics[apic].mpc_apicid, pin);
1291                         continue;
1292                 }
1293
1294                 if (!first_notcon) {
1295                         apic_printk(APIC_VERBOSE, " not connected.\n");
1296                         first_notcon = 1;
1297                 }
1298
1299                 entry.trigger = irq_trigger(idx);
1300                 entry.polarity = irq_polarity(idx);
1301
1302                 if (irq_trigger(idx)) {
1303                         entry.trigger = 1;
1304                         entry.mask = 1;
1305                 }
1306
1307                 irq = pin_2_irq(idx, apic, pin);
1308                 /*
1309                  * skip adding the timer int on secondary nodes, which causes
1310                  * a small but painful rift in the time-space continuum
1311                  */
1312                 if (multi_timer_check(apic, irq))
1313                         continue;
1314                 else
1315                         add_pin_to_irq(irq, apic, pin);
1316
1317                 if (!apic && !IO_APIC_IRQ(irq))
1318                         continue;
1319
1320                 if (IO_APIC_IRQ(irq)) {
1321                         vector = assign_irq_vector(irq);
1322                         entry.vector = vector;
1323                         ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1324
1325                         if (!apic && (irq < 16))
1326                                 disable_8259A_irq(irq);
1327                 }
1328                 spin_lock_irqsave(&ioapic_lock, flags);
1329                 __ioapic_write_entry(apic, pin, entry);
1330                 spin_unlock_irqrestore(&ioapic_lock, flags);
1331         }
1332         }
1333
1334         if (!first_notcon)
1335                 apic_printk(APIC_VERBOSE, " not connected.\n");
1336 }
1337
1338 /*
1339  * Set up the 8259A-master output pin:
1340  */
1341 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
1342 {
1343         struct IO_APIC_route_entry entry;
1344
1345         memset(&entry,0,sizeof(entry));
1346
1347         disable_8259A_irq(0);
1348
1349         /* mask LVT0 */
1350         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1351
1352         /*
1353          * We use logical delivery to get the timer IRQ
1354          * to the first CPU.
1355          */
1356         entry.dest_mode = INT_DEST_MODE;
1357         entry.mask = 0;                                 /* unmask IRQ now */
1358         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1359         entry.delivery_mode = INT_DELIVERY_MODE;
1360         entry.polarity = 0;
1361         entry.trigger = 0;
1362         entry.vector = vector;
1363
1364         /*
1365          * The timer IRQ doesn't have to know that behind the
1366          * scene we have a 8259A-master in AEOI mode ...
1367          */
1368         irq_desc[0].chip = &ioapic_chip;
1369         set_irq_handler(0, handle_edge_irq);
1370
1371         /*
1372          * Add it to the IO-APIC irq-routing table:
1373          */
1374         ioapic_write_entry(apic, pin, entry);
1375
1376         enable_8259A_irq(0);
1377 }
1378
1379 void __init print_IO_APIC(void)
1380 {
1381         int apic, i;
1382         union IO_APIC_reg_00 reg_00;
1383         union IO_APIC_reg_01 reg_01;
1384         union IO_APIC_reg_02 reg_02;
1385         union IO_APIC_reg_03 reg_03;
1386         unsigned long flags;
1387
1388         if (apic_verbosity == APIC_QUIET)
1389                 return;
1390
1391         printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1392         for (i = 0; i < nr_ioapics; i++)
1393                 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1394                        mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1395
1396         /*
1397          * We are a bit conservative about what we expect.  We have to
1398          * know about every hardware change ASAP.
1399          */
1400         printk(KERN_INFO "testing the IO APIC.......................\n");
1401
1402         for (apic = 0; apic < nr_ioapics; apic++) {
1403
1404         spin_lock_irqsave(&ioapic_lock, flags);
1405         reg_00.raw = io_apic_read(apic, 0);
1406         reg_01.raw = io_apic_read(apic, 1);
1407         if (reg_01.bits.version >= 0x10)
1408                 reg_02.raw = io_apic_read(apic, 2);
1409         if (reg_01.bits.version >= 0x20)
1410                 reg_03.raw = io_apic_read(apic, 3);
1411         spin_unlock_irqrestore(&ioapic_lock, flags);
1412
1413         printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1414         printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1415         printk(KERN_DEBUG ".......    : physical APIC id: %02X\n", reg_00.bits.ID);
1416         printk(KERN_DEBUG ".......    : Delivery Type: %X\n", reg_00.bits.delivery_type);
1417         printk(KERN_DEBUG ".......    : LTS          : %X\n", reg_00.bits.LTS);
1418
1419         printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1420         printk(KERN_DEBUG ".......     : max redirection entries: %04X\n", reg_01.bits.entries);
1421
1422         printk(KERN_DEBUG ".......     : PRQ implemented: %X\n", reg_01.bits.PRQ);
1423         printk(KERN_DEBUG ".......     : IO APIC version: %04X\n", reg_01.bits.version);
1424
1425         /*
1426          * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1427          * but the value of reg_02 is read as the previous read register
1428          * value, so ignore it if reg_02 == reg_01.
1429          */
1430         if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1431                 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1432                 printk(KERN_DEBUG ".......     : arbitration: %02X\n", reg_02.bits.arbitration);
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         }
1445
1446         printk(KERN_DEBUG ".... IRQ redirection table:\n");
1447
1448         printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1449                           " Stat Dest Deli Vect:   \n");
1450
1451         for (i = 0; i <= reg_01.bits.entries; i++) {
1452                 struct IO_APIC_route_entry entry;
1453
1454                 entry = ioapic_read_entry(apic, i);
1455
1456                 printk(KERN_DEBUG " %02x %03X %02X  ",
1457                         i,
1458                         entry.dest.logical.logical_dest,
1459                         entry.dest.physical.physical_dest
1460                 );
1461
1462                 printk("%1d    %1d    %1d   %1d   %1d    %1d    %1d    %02X\n",
1463                         entry.mask,
1464                         entry.trigger,
1465                         entry.irr,
1466                         entry.polarity,
1467                         entry.delivery_status,
1468                         entry.dest_mode,
1469                         entry.delivery_mode,
1470                         entry.vector
1471                 );
1472         }
1473         }
1474         printk(KERN_DEBUG "IRQ to pin mappings:\n");
1475         for (i = 0; i < NR_IRQS; i++) {
1476                 struct irq_pin_list *entry = irq_2_pin + i;
1477                 if (entry->pin < 0)
1478                         continue;
1479                 printk(KERN_DEBUG "IRQ%d ", i);
1480                 for (;;) {
1481                         printk("-> %d:%d", entry->apic, entry->pin);
1482                         if (!entry->next)
1483                                 break;
1484                         entry = irq_2_pin + entry->next;
1485                 }
1486                 printk("\n");
1487         }
1488
1489         printk(KERN_INFO ".................................... done.\n");
1490
1491         return;
1492 }
1493
1494 #if 0
1495
1496 static void print_APIC_bitfield (int base)
1497 {
1498         unsigned int v;
1499         int i, j;
1500
1501         if (apic_verbosity == APIC_QUIET)
1502                 return;
1503
1504         printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1505         for (i = 0; i < 8; i++) {
1506                 v = apic_read(base + i*0x10);
1507                 for (j = 0; j < 32; j++) {
1508                         if (v & (1<<j))
1509                                 printk("1");
1510                         else
1511                                 printk("0");
1512                 }
1513                 printk("\n");
1514         }
1515 }
1516
1517 void /*__init*/ print_local_APIC(void * dummy)
1518 {
1519         unsigned int v, ver, maxlvt;
1520
1521         if (apic_verbosity == APIC_QUIET)
1522                 return;
1523
1524         printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1525                 smp_processor_id(), hard_smp_processor_id());
1526         v = apic_read(APIC_ID);
1527         printk(KERN_INFO "... APIC ID:      %08x (%01x)\n", v, GET_APIC_ID(v));
1528         v = apic_read(APIC_LVR);
1529         printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1530         ver = GET_APIC_VERSION(v);
1531         maxlvt = lapic_get_maxlvt();
1532
1533         v = apic_read(APIC_TASKPRI);
1534         printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1535
1536         if (APIC_INTEGRATED(ver)) {                     /* !82489DX */
1537                 v = apic_read(APIC_ARBPRI);
1538                 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1539                         v & APIC_ARBPRI_MASK);
1540                 v = apic_read(APIC_PROCPRI);
1541                 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1542         }
1543
1544         v = apic_read(APIC_EOI);
1545         printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1546         v = apic_read(APIC_RRR);
1547         printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1548         v = apic_read(APIC_LDR);
1549         printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1550         v = apic_read(APIC_DFR);
1551         printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1552         v = apic_read(APIC_SPIV);
1553         printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1554
1555         printk(KERN_DEBUG "... APIC ISR field:\n");
1556         print_APIC_bitfield(APIC_ISR);
1557         printk(KERN_DEBUG "... APIC TMR field:\n");
1558         print_APIC_bitfield(APIC_TMR);
1559         printk(KERN_DEBUG "... APIC IRR field:\n");
1560         print_APIC_bitfield(APIC_IRR);
1561
1562         if (APIC_INTEGRATED(ver)) {             /* !82489DX */
1563                 if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
1564                         apic_write(APIC_ESR, 0);
1565                 v = apic_read(APIC_ESR);
1566                 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1567         }
1568
1569         v = apic_read(APIC_ICR);
1570         printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1571         v = apic_read(APIC_ICR2);
1572         printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1573
1574         v = apic_read(APIC_LVTT);
1575         printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1576
1577         if (maxlvt > 3) {                       /* PC is LVT#4. */
1578                 v = apic_read(APIC_LVTPC);
1579                 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1580         }
1581         v = apic_read(APIC_LVT0);
1582         printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1583         v = apic_read(APIC_LVT1);
1584         printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1585
1586         if (maxlvt > 2) {                       /* ERR is LVT#3. */
1587                 v = apic_read(APIC_LVTERR);
1588                 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1589         }
1590
1591         v = apic_read(APIC_TMICT);
1592         printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1593         v = apic_read(APIC_TMCCT);
1594         printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1595         v = apic_read(APIC_TDCR);
1596         printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1597         printk("\n");
1598 }
1599
1600 void print_all_local_APICs (void)
1601 {
1602         on_each_cpu(print_local_APIC, NULL, 1, 1);
1603 }
1604
1605 void /*__init*/ print_PIC(void)
1606 {
1607         unsigned int v;
1608         unsigned long flags;
1609
1610         if (apic_verbosity == APIC_QUIET)
1611                 return;
1612
1613         printk(KERN_DEBUG "\nprinting PIC contents\n");
1614
1615         spin_lock_irqsave(&i8259A_lock, flags);
1616
1617         v = inb(0xa1) << 8 | inb(0x21);
1618         printk(KERN_DEBUG "... PIC  IMR: %04x\n", v);
1619
1620         v = inb(0xa0) << 8 | inb(0x20);
1621         printk(KERN_DEBUG "... PIC  IRR: %04x\n", v);
1622
1623         outb(0x0b,0xa0);
1624         outb(0x0b,0x20);
1625         v = inb(0xa0) << 8 | inb(0x20);
1626         outb(0x0a,0xa0);
1627         outb(0x0a,0x20);
1628
1629         spin_unlock_irqrestore(&i8259A_lock, flags);
1630
1631         printk(KERN_DEBUG "... PIC  ISR: %04x\n", v);
1632
1633         v = inb(0x4d1) << 8 | inb(0x4d0);
1634         printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1635 }
1636
1637 #endif  /*  0  */
1638
1639 static void __init enable_IO_APIC(void)
1640 {
1641         union IO_APIC_reg_01 reg_01;
1642         int i8259_apic, i8259_pin;
1643         int i, apic;
1644         unsigned long flags;
1645
1646         for (i = 0; i < PIN_MAP_SIZE; i++) {
1647                 irq_2_pin[i].pin = -1;
1648                 irq_2_pin[i].next = 0;
1649         }
1650         if (!pirqs_enabled)
1651                 for (i = 0; i < MAX_PIRQS; i++)
1652                         pirq_entries[i] = -1;
1653
1654         /*
1655          * The number of IO-APIC IRQ registers (== #pins):
1656          */
1657         for (apic = 0; apic < nr_ioapics; apic++) {
1658                 spin_lock_irqsave(&ioapic_lock, flags);
1659                 reg_01.raw = io_apic_read(apic, 1);
1660                 spin_unlock_irqrestore(&ioapic_lock, flags);
1661                 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1662         }
1663         for(apic = 0; apic < nr_ioapics; apic++) {
1664                 int pin;
1665                 /* See if any of the pins is in ExtINT mode */
1666                 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1667                         struct IO_APIC_route_entry entry;
1668                         entry = ioapic_read_entry(apic, pin);
1669
1670
1671                         /* If the interrupt line is enabled and in ExtInt mode
1672                          * I have found the pin where the i8259 is connected.
1673                          */
1674                         if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1675                                 ioapic_i8259.apic = apic;
1676                                 ioapic_i8259.pin  = pin;
1677                                 goto found_i8259;
1678                         }
1679                 }
1680         }
1681  found_i8259:
1682         /* Look to see what if the MP table has reported the ExtINT */
1683         /* If we could not find the appropriate pin by looking at the ioapic
1684          * the i8259 probably is not connected the ioapic but give the
1685          * mptable a chance anyway.
1686          */
1687         i8259_pin  = find_isa_irq_pin(0, mp_ExtINT);
1688         i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1689         /* Trust the MP table if nothing is setup in the hardware */
1690         if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1691                 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1692                 ioapic_i8259.pin  = i8259_pin;
1693                 ioapic_i8259.apic = i8259_apic;
1694         }
1695         /* Complain if the MP table and the hardware disagree */
1696         if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1697                 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1698         {
1699                 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1700         }
1701
1702         /*
1703          * Do not trust the IO-APIC being empty at bootup
1704          */
1705         clear_IO_APIC();
1706 }
1707
1708 /*
1709  * Not an __init, needed by the reboot code
1710  */
1711 void disable_IO_APIC(void)
1712 {
1713         /*
1714          * Clear the IO-APIC before rebooting:
1715          */
1716         clear_IO_APIC();
1717
1718         /*
1719          * If the i8259 is routed through an IOAPIC
1720          * Put that IOAPIC in virtual wire mode
1721          * so legacy interrupts can be delivered.
1722          */
1723         if (ioapic_i8259.pin != -1) {
1724                 struct IO_APIC_route_entry entry;
1725
1726                 memset(&entry, 0, sizeof(entry));
1727                 entry.mask            = 0; /* Enabled */
1728                 entry.trigger         = 0; /* Edge */
1729                 entry.irr             = 0;
1730                 entry.polarity        = 0; /* High */
1731                 entry.delivery_status = 0;
1732                 entry.dest_mode       = 0; /* Physical */
1733                 entry.delivery_mode   = dest_ExtINT; /* ExtInt */
1734                 entry.vector          = 0;
1735                 entry.dest.physical.physical_dest =
1736                                         GET_APIC_ID(apic_read(APIC_ID));
1737
1738                 /*
1739                  * Add it to the IO-APIC irq-routing table:
1740                  */
1741                 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1742         }
1743         disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1744 }
1745
1746 /*
1747  * function to set the IO-APIC physical IDs based on the
1748  * values stored in the MPC table.
1749  *
1750  * by Matt Domsch <Matt_Domsch@dell.com>  Tue Dec 21 12:25:05 CST 1999
1751  */
1752
1753 #ifndef CONFIG_X86_NUMAQ
1754 static void __init setup_ioapic_ids_from_mpc(void)
1755 {
1756         union IO_APIC_reg_00 reg_00;
1757         physid_mask_t phys_id_present_map;
1758         int apic;
1759         int i;
1760         unsigned char old_id;
1761         unsigned long flags;
1762
1763         /*
1764          * Don't check I/O APIC IDs for xAPIC systems.  They have
1765          * no meaning without the serial APIC bus.
1766          */
1767         if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1768                 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
1769                 return;
1770         /*
1771          * This is broken; anything with a real cpu count has to
1772          * circumvent this idiocy regardless.
1773          */
1774         phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1775
1776         /*
1777          * Set the IOAPIC ID to the value stored in the MPC table.
1778          */
1779         for (apic = 0; apic < nr_ioapics; apic++) {
1780
1781                 /* Read the register 0 value */
1782                 spin_lock_irqsave(&ioapic_lock, flags);
1783                 reg_00.raw = io_apic_read(apic, 0);
1784                 spin_unlock_irqrestore(&ioapic_lock, flags);
1785
1786                 old_id = mp_ioapics[apic].mpc_apicid;
1787
1788                 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) {
1789                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1790                                 apic, mp_ioapics[apic].mpc_apicid);
1791                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1792                                 reg_00.bits.ID);
1793                         mp_ioapics[apic].mpc_apicid = reg_00.bits.ID;
1794                 }
1795
1796                 /*
1797                  * Sanity check, is the ID really free? Every APIC in a
1798                  * system must have a unique ID or we get lots of nice
1799                  * 'stuck on smp_invalidate_needed IPI wait' messages.
1800                  */
1801                 if (check_apicid_used(phys_id_present_map,
1802                                         mp_ioapics[apic].mpc_apicid)) {
1803                         printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1804                                 apic, mp_ioapics[apic].mpc_apicid);
1805                         for (i = 0; i < get_physical_broadcast(); i++)
1806                                 if (!physid_isset(i, phys_id_present_map))
1807                                         break;
1808                         if (i >= get_physical_broadcast())
1809                                 panic("Max APIC ID exceeded!\n");
1810                         printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1811                                 i);
1812                         physid_set(i, phys_id_present_map);
1813                         mp_ioapics[apic].mpc_apicid = i;
1814                 } else {
1815                         physid_mask_t tmp;
1816                         tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid);
1817                         apic_printk(APIC_VERBOSE, "Setting %d in the "
1818                                         "phys_id_present_map\n",
1819                                         mp_ioapics[apic].mpc_apicid);
1820                         physids_or(phys_id_present_map, phys_id_present_map, tmp);
1821                 }
1822
1823
1824                 /*
1825                  * We need to adjust the IRQ routing table
1826                  * if the ID changed.
1827                  */
1828                 if (old_id != mp_ioapics[apic].mpc_apicid)
1829                         for (i = 0; i < mp_irq_entries; i++)
1830                                 if (mp_irqs[i].mpc_dstapic == old_id)
1831                                         mp_irqs[i].mpc_dstapic
1832                                                 = mp_ioapics[apic].mpc_apicid;
1833
1834                 /*
1835                  * Read the right value from the MPC table and
1836                  * write it into the ID register.
1837                  */
1838                 apic_printk(APIC_VERBOSE, KERN_INFO
1839                         "...changing IO-APIC physical APIC ID to %d ...",
1840                         mp_ioapics[apic].mpc_apicid);
1841
1842                 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1843                 spin_lock_irqsave(&ioapic_lock, flags);
1844                 io_apic_write(apic, 0, reg_00.raw);
1845                 spin_unlock_irqrestore(&ioapic_lock, flags);
1846
1847                 /*
1848                  * Sanity check
1849                  */
1850                 spin_lock_irqsave(&ioapic_lock, flags);
1851                 reg_00.raw = io_apic_read(apic, 0);
1852                 spin_unlock_irqrestore(&ioapic_lock, flags);
1853                 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1854                         printk("could not set ID!\n");
1855                 else
1856                         apic_printk(APIC_VERBOSE, " ok.\n");
1857         }
1858 }
1859 #else
1860 static void __init setup_ioapic_ids_from_mpc(void) { }
1861 #endif
1862
1863 int no_timer_check __initdata;
1864
1865 static int __init notimercheck(char *s)
1866 {
1867         no_timer_check = 1;
1868         return 1;
1869 }
1870 __setup("no_timer_check", notimercheck);
1871
1872 /*
1873  * There is a nasty bug in some older SMP boards, their mptable lies
1874  * about the timer IRQ. We do the following to work around the situation:
1875  *
1876  *      - timer IRQ defaults to IO-APIC IRQ
1877  *      - if this function detects that timer IRQs are defunct, then we fall
1878  *        back to ISA timer IRQs
1879  */
1880 static int __init timer_irq_works(void)
1881 {
1882         unsigned long t1 = jiffies;
1883         unsigned long flags;
1884
1885         if (no_timer_check)
1886                 return 1;
1887
1888         local_save_flags(flags);
1889         local_irq_enable();
1890         /* Let ten ticks pass... */
1891         mdelay((10 * 1000) / HZ);
1892         local_irq_restore(flags);
1893
1894         /*
1895          * Expect a few ticks at least, to be sure some possible
1896          * glue logic does not lock up after one or two first
1897          * ticks in a non-ExtINT mode.  Also the local APIC
1898          * might have cached one ExtINT interrupt.  Finally, at
1899          * least one tick may be lost due to delays.
1900          */
1901         if (jiffies - t1 > 4)
1902                 return 1;
1903
1904         return 0;
1905 }
1906
1907 /*
1908  * In the SMP+IOAPIC case it might happen that there are an unspecified
1909  * number of pending IRQ events unhandled. These cases are very rare,
1910  * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1911  * better to do it this way as thus we do not have to be aware of
1912  * 'pending' interrupts in the IRQ path, except at this point.
1913  */
1914 /*
1915  * Edge triggered needs to resend any interrupt
1916  * that was delayed but this is now handled in the device
1917  * independent code.
1918  */
1919
1920 /*
1921  * Startup quirk:
1922  *
1923  * Starting up a edge-triggered IO-APIC interrupt is
1924  * nasty - we need to make sure that we get the edge.
1925  * If it is already asserted for some reason, we need
1926  * return 1 to indicate that is was pending.
1927  *
1928  * This is not complete - we should be able to fake
1929  * an edge even if it isn't on the 8259A...
1930  *
1931  * (We do this for level-triggered IRQs too - it cannot hurt.)
1932  */
1933 static unsigned int startup_ioapic_irq(unsigned int irq)
1934 {
1935         int was_pending = 0;
1936         unsigned long flags;
1937
1938         spin_lock_irqsave(&ioapic_lock, flags);
1939         if (irq < 16) {
1940                 disable_8259A_irq(irq);
1941                 if (i8259A_irq_pending(irq))
1942                         was_pending = 1;
1943         }
1944         __unmask_IO_APIC_irq(irq);
1945         spin_unlock_irqrestore(&ioapic_lock, flags);
1946
1947         return was_pending;
1948 }
1949
1950 static void ack_ioapic_irq(unsigned int irq)
1951 {
1952         move_native_irq(irq);
1953         ack_APIC_irq();
1954 }
1955
1956 static void ack_ioapic_quirk_irq(unsigned int irq)
1957 {
1958         unsigned long v;
1959         int i;
1960
1961         move_native_irq(irq);
1962 /*
1963  * It appears there is an erratum which affects at least version 0x11
1964  * of I/O APIC (that's the 82093AA and cores integrated into various
1965  * chipsets).  Under certain conditions a level-triggered interrupt is
1966  * erroneously delivered as edge-triggered one but the respective IRR
1967  * bit gets set nevertheless.  As a result the I/O unit expects an EOI
1968  * message but it will never arrive and further interrupts are blocked
1969  * from the source.  The exact reason is so far unknown, but the
1970  * phenomenon was observed when two consecutive interrupt requests
1971  * from a given source get delivered to the same CPU and the source is
1972  * temporarily disabled in between.
1973  *
1974  * A workaround is to simulate an EOI message manually.  We achieve it
1975  * by setting the trigger mode to edge and then to level when the edge
1976  * trigger mode gets detected in the TMR of a local APIC for a
1977  * level-triggered interrupt.  We mask the source for the time of the
1978  * operation to prevent an edge-triggered interrupt escaping meanwhile.
1979  * The idea is from Manfred Spraul.  --macro
1980  */
1981         i = irq_vector[irq];
1982
1983         v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1984
1985         ack_APIC_irq();
1986
1987         if (!(v & (1 << (i & 0x1f)))) {
1988                 atomic_inc(&irq_mis_count);
1989                 spin_lock(&ioapic_lock);
1990                 __mask_and_edge_IO_APIC_irq(irq);
1991                 __unmask_and_level_IO_APIC_irq(irq);
1992                 spin_unlock(&ioapic_lock);
1993         }
1994 }
1995
1996 static int ioapic_retrigger_irq(unsigned int irq)
1997 {
1998         send_IPI_self(irq_vector[irq]);
1999
2000         return 1;
2001 }
2002
2003 static struct irq_chip ioapic_chip __read_mostly = {
2004         .name           = "IO-APIC",
2005         .startup        = startup_ioapic_irq,
2006         .mask           = mask_IO_APIC_irq,
2007         .unmask         = unmask_IO_APIC_irq,
2008         .ack            = ack_ioapic_irq,
2009         .eoi            = ack_ioapic_quirk_irq,
2010 #ifdef CONFIG_SMP
2011         .set_affinity   = set_ioapic_affinity_irq,
2012 #endif
2013         .retrigger      = ioapic_retrigger_irq,
2014 };
2015
2016
2017 static inline void init_IO_APIC_traps(void)
2018 {
2019         int irq;
2020
2021         /*
2022          * NOTE! The local APIC isn't very good at handling
2023          * multiple interrupts at the same interrupt level.
2024          * As the interrupt level is determined by taking the
2025          * vector number and shifting that right by 4, we
2026          * want to spread these out a bit so that they don't
2027          * all fall in the same interrupt level.
2028          *
2029          * Also, we've got to be careful not to trash gate
2030          * 0x80, because int 0x80 is hm, kind of importantish. ;)
2031          */
2032         for (irq = 0; irq < NR_IRQS ; irq++) {
2033                 int tmp = irq;
2034                 if (IO_APIC_IRQ(tmp) && !irq_vector[tmp]) {
2035                         /*
2036                          * Hmm.. We don't have an entry for this,
2037                          * so default to an old-fashioned 8259
2038                          * interrupt if we can..
2039                          */
2040                         if (irq < 16)
2041                                 make_8259A_irq(irq);
2042                         else
2043                                 /* Strange. Oh, well.. */
2044                                 irq_desc[irq].chip = &no_irq_chip;
2045                 }
2046         }
2047 }
2048
2049 /*
2050  * The local APIC irq-chip implementation:
2051  */
2052
2053 static void ack_apic(unsigned int irq)
2054 {
2055         ack_APIC_irq();
2056 }
2057
2058 static void mask_lapic_irq (unsigned int irq)
2059 {
2060         unsigned long v;
2061
2062         v = apic_read(APIC_LVT0);
2063         apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2064 }
2065
2066 static void unmask_lapic_irq (unsigned int irq)
2067 {
2068         unsigned long v;
2069
2070         v = apic_read(APIC_LVT0);
2071         apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2072 }
2073
2074 static struct irq_chip lapic_chip __read_mostly = {
2075         .name           = "local-APIC-edge",
2076         .mask           = mask_lapic_irq,
2077         .unmask         = unmask_lapic_irq,
2078         .eoi            = ack_apic,
2079 };
2080
2081 static void __init setup_nmi(void)
2082 {
2083         /*
2084          * Dirty trick to enable the NMI watchdog ...
2085          * We put the 8259A master into AEOI mode and
2086          * unmask on all local APICs LVT0 as NMI.
2087          *
2088          * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2089          * is from Maciej W. Rozycki - so we do not have to EOI from
2090          * the NMI handler or the timer interrupt.
2091          */
2092         apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2093
2094         enable_NMI_through_LVT0();
2095
2096         apic_printk(APIC_VERBOSE, " done.\n");
2097 }
2098
2099 /*
2100  * This looks a bit hackish but it's about the only one way of sending
2101  * a few INTA cycles to 8259As and any associated glue logic.  ICR does
2102  * not support the ExtINT mode, unfortunately.  We need to send these
2103  * cycles as some i82489DX-based boards have glue logic that keeps the
2104  * 8259A interrupt line asserted until INTA.  --macro
2105  */
2106 static inline void unlock_ExtINT_logic(void)
2107 {
2108         int apic, pin, i;
2109         struct IO_APIC_route_entry entry0, entry1;
2110         unsigned char save_control, save_freq_select;
2111
2112         pin  = find_isa_irq_pin(8, mp_INT);
2113         if (pin == -1) {
2114                 WARN_ON_ONCE(1);
2115                 return;
2116         }
2117         apic = find_isa_irq_apic(8, mp_INT);
2118         if (apic == -1) {
2119                 WARN_ON_ONCE(1);
2120                 return;
2121         }
2122
2123         entry0 = ioapic_read_entry(apic, pin);
2124         clear_IO_APIC_pin(apic, pin);
2125
2126         memset(&entry1, 0, sizeof(entry1));
2127
2128         entry1.dest_mode = 0;                   /* physical delivery */
2129         entry1.mask = 0;                        /* unmask IRQ now */
2130         entry1.dest.physical.physical_dest = hard_smp_processor_id();
2131         entry1.delivery_mode = dest_ExtINT;
2132         entry1.polarity = entry0.polarity;
2133         entry1.trigger = 0;
2134         entry1.vector = 0;
2135
2136         ioapic_write_entry(apic, pin, entry1);
2137
2138         save_control = CMOS_READ(RTC_CONTROL);
2139         save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2140         CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2141                    RTC_FREQ_SELECT);
2142         CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2143
2144         i = 100;
2145         while (i-- > 0) {
2146                 mdelay(10);
2147                 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2148                         i -= 10;
2149         }
2150
2151         CMOS_WRITE(save_control, RTC_CONTROL);
2152         CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2153         clear_IO_APIC_pin(apic, pin);
2154
2155         ioapic_write_entry(apic, pin, entry0);
2156 }
2157
2158 int timer_uses_ioapic_pin_0;
2159
2160 /*
2161  * This code may look a bit paranoid, but it's supposed to cooperate with
2162  * a wide range of boards and BIOS bugs.  Fortunately only the timer IRQ
2163  * is so screwy.  Thanks to Brian Perkins for testing/hacking this beast
2164  * fanatically on his truly buggy board.
2165  */
2166 static inline void __init check_timer(void)
2167 {
2168         int apic1, pin1, apic2, pin2;
2169         int vector;
2170         unsigned long flags;
2171
2172         local_irq_save(flags);
2173
2174         /*
2175          * get/set the timer IRQ vector:
2176          */
2177         disable_8259A_irq(0);
2178         vector = assign_irq_vector(0);
2179         set_intr_gate(vector, interrupt[0]);
2180
2181         /*
2182          * Subtle, code in do_timer_interrupt() expects an AEOI
2183          * mode for the 8259A whenever interrupts are routed
2184          * through I/O APICs.  Also IRQ0 has to be enabled in
2185          * the 8259A which implies the virtual wire has to be
2186          * disabled in the local APIC.
2187          */
2188         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2189         init_8259A(1);
2190         timer_ack = 1;
2191         if (timer_over_8254 > 0)
2192                 enable_8259A_irq(0);
2193
2194         pin1  = find_isa_irq_pin(0, mp_INT);
2195         apic1 = find_isa_irq_apic(0, mp_INT);
2196         pin2  = ioapic_i8259.pin;
2197         apic2 = ioapic_i8259.apic;
2198
2199         if (pin1 == 0)
2200                 timer_uses_ioapic_pin_0 = 1;
2201
2202         printk(KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
2203                 vector, apic1, pin1, apic2, pin2);
2204
2205         if (pin1 != -1) {
2206                 /*
2207                  * Ok, does IRQ0 through the IOAPIC work?
2208                  */
2209                 unmask_IO_APIC_irq(0);
2210                 if (timer_irq_works()) {
2211                         if (nmi_watchdog == NMI_IO_APIC) {
2212                                 disable_8259A_irq(0);
2213                                 setup_nmi();
2214                                 enable_8259A_irq(0);
2215                         }
2216                         if (disable_timer_pin_1 > 0)
2217                                 clear_IO_APIC_pin(0, pin1);
2218                         goto out;
2219                 }
2220                 clear_IO_APIC_pin(apic1, pin1);
2221                 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to "
2222                                 "IO-APIC\n");
2223         }
2224
2225         printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2226         if (pin2 != -1) {
2227                 printk("\n..... (found pin %d) ...", pin2);
2228                 /*
2229                  * legacy devices should be connected to IO APIC #0
2230                  */
2231                 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
2232                 if (timer_irq_works()) {
2233                         printk("works.\n");
2234                         if (pin1 != -1)
2235                                 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2236                         else
2237                                 add_pin_to_irq(0, apic2, pin2);
2238                         if (nmi_watchdog == NMI_IO_APIC) {
2239                                 setup_nmi();
2240                         }
2241                         goto out;
2242                 }
2243                 /*
2244                  * Cleanup, just in case ...
2245                  */
2246                 clear_IO_APIC_pin(apic2, pin2);
2247         }
2248         printk(" failed.\n");
2249
2250         if (nmi_watchdog == NMI_IO_APIC) {
2251                 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2252                 nmi_watchdog = 0;
2253         }
2254
2255         printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2256
2257         disable_8259A_irq(0);
2258         set_irq_chip_and_handler_name(0, &lapic_chip, handle_fasteoi_irq,
2259                                       "fasteoi");
2260         apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector);   /* Fixed mode */
2261         enable_8259A_irq(0);
2262
2263         if (timer_irq_works()) {
2264                 printk(" works.\n");
2265                 goto out;
2266         }
2267         apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2268         printk(" failed.\n");
2269
2270         printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2271
2272         timer_ack = 0;
2273         init_8259A(0);
2274         make_8259A_irq(0);
2275         apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2276
2277         unlock_ExtINT_logic();
2278
2279         if (timer_irq_works()) {
2280                 printk(" works.\n");
2281                 goto out;
2282         }
2283         printk(" failed :(.\n");
2284         panic("IO-APIC + timer doesn't work!  Boot with apic=debug and send a "
2285                 "report.  Then try booting with the 'noapic' option");
2286 out:
2287         local_irq_restore(flags);
2288 }
2289
2290 /*
2291  *
2292  * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2293  * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2294  *   Linux doesn't really care, as it's not actually used
2295  *   for any interrupt handling anyway.
2296  */
2297 #define PIC_IRQS        (1 << PIC_CASCADE_IR)
2298
2299 void __init setup_IO_APIC(void)
2300 {
2301         int i;
2302
2303         /* Reserve all the system vectors. */
2304         for (i = FIRST_SYSTEM_VECTOR; i < NR_VECTORS; i++)
2305                 set_bit(i, used_vectors);
2306
2307         enable_IO_APIC();
2308
2309         if (acpi_ioapic)
2310                 io_apic_irqs = ~0;      /* all IRQs go through IOAPIC */
2311         else
2312                 io_apic_irqs = ~PIC_IRQS;
2313
2314         printk("ENABLING IO-APIC IRQs\n");
2315
2316         /*
2317          * Set up IO-APIC IRQ routing.
2318          */
2319         if (!acpi_ioapic)
2320                 setup_ioapic_ids_from_mpc();
2321         sync_Arb_IDs();
2322         setup_IO_APIC_irqs();
2323         init_IO_APIC_traps();
2324         check_timer();
2325         if (!acpi_ioapic)
2326                 print_IO_APIC();
2327 }
2328
2329 static int __init setup_disable_8254_timer(char *s)
2330 {
2331         timer_over_8254 = -1;
2332         return 1;
2333 }
2334 static int __init setup_enable_8254_timer(char *s)
2335 {
2336         timer_over_8254 = 2;
2337         return 1;
2338 }
2339
2340 __setup("disable_8254_timer", setup_disable_8254_timer);
2341 __setup("enable_8254_timer", setup_enable_8254_timer);
2342
2343 /*
2344  *      Called after all the initialization is done. If we didnt find any
2345  *      APIC bugs then we can allow the modify fast path
2346  */
2347
2348 static int __init io_apic_bug_finalize(void)
2349 {
2350         if(sis_apic_bug == -1)
2351                 sis_apic_bug = 0;
2352         return 0;
2353 }
2354
2355 late_initcall(io_apic_bug_finalize);
2356
2357 struct sysfs_ioapic_data {
2358         struct sys_device dev;
2359         struct IO_APIC_route_entry entry[0];
2360 };
2361 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2362
2363 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2364 {
2365         struct IO_APIC_route_entry *entry;
2366         struct sysfs_ioapic_data *data;
2367         int i;
2368
2369         data = container_of(dev, struct sysfs_ioapic_data, dev);
2370         entry = data->entry;
2371         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2372                 entry[i] = ioapic_read_entry(dev->id, i);
2373
2374         return 0;
2375 }
2376
2377 static int ioapic_resume(struct sys_device *dev)
2378 {
2379         struct IO_APIC_route_entry *entry;
2380         struct sysfs_ioapic_data *data;
2381         unsigned long flags;
2382         union IO_APIC_reg_00 reg_00;
2383         int i;
2384
2385         data = container_of(dev, struct sysfs_ioapic_data, dev);
2386         entry = data->entry;
2387
2388         spin_lock_irqsave(&ioapic_lock, flags);
2389         reg_00.raw = io_apic_read(dev->id, 0);
2390         if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2391                 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2392                 io_apic_write(dev->id, 0, reg_00.raw);
2393         }
2394         spin_unlock_irqrestore(&ioapic_lock, flags);
2395         for (i = 0; i < nr_ioapic_registers[dev->id]; i ++)
2396                 ioapic_write_entry(dev->id, i, entry[i]);
2397
2398         return 0;
2399 }
2400
2401 static struct sysdev_class ioapic_sysdev_class = {
2402         .name = "ioapic",
2403         .suspend = ioapic_suspend,
2404         .resume = ioapic_resume,
2405 };
2406
2407 static int __init ioapic_init_sysfs(void)
2408 {
2409         struct sys_device * dev;
2410         int i, size, error = 0;
2411
2412         error = sysdev_class_register(&ioapic_sysdev_class);
2413         if (error)
2414                 return error;
2415
2416         for (i = 0; i < nr_ioapics; i++ ) {
2417                 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2418                         * sizeof(struct IO_APIC_route_entry);
2419                 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2420                 if (!mp_ioapic_data[i]) {
2421                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2422                         continue;
2423                 }
2424                 memset(mp_ioapic_data[i], 0, size);
2425                 dev = &mp_ioapic_data[i]->dev;
2426                 dev->id = i;
2427                 dev->cls = &ioapic_sysdev_class;
2428                 error = sysdev_register(dev);
2429                 if (error) {
2430                         kfree(mp_ioapic_data[i]);
2431                         mp_ioapic_data[i] = NULL;
2432                         printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2433                         continue;
2434                 }
2435         }
2436
2437         return 0;
2438 }
2439
2440 device_initcall(ioapic_init_sysfs);
2441
2442 /*
2443  * Dynamic irq allocate and deallocation
2444  */
2445 int create_irq(void)
2446 {
2447         /* Allocate an unused irq */
2448         int irq, new, vector = 0;
2449         unsigned long flags;
2450
2451         irq = -ENOSPC;
2452         spin_lock_irqsave(&vector_lock, flags);
2453         for (new = (NR_IRQS - 1); new >= 0; new--) {
2454                 if (platform_legacy_irq(new))
2455                         continue;
2456                 if (irq_vector[new] != 0)
2457                         continue;
2458                 vector = __assign_irq_vector(new);
2459                 if (likely(vector > 0))
2460                         irq = new;
2461                 break;
2462         }
2463         spin_unlock_irqrestore(&vector_lock, flags);
2464
2465         if (irq >= 0) {
2466                 set_intr_gate(vector, interrupt[irq]);
2467                 dynamic_irq_init(irq);
2468         }
2469         return irq;
2470 }
2471
2472 void destroy_irq(unsigned int irq)
2473 {
2474         unsigned long flags;
2475
2476         dynamic_irq_cleanup(irq);
2477
2478         spin_lock_irqsave(&vector_lock, flags);
2479         irq_vector[irq] = 0;
2480         spin_unlock_irqrestore(&vector_lock, flags);
2481 }
2482
2483 /*
2484  * MSI message composition
2485  */
2486 #ifdef CONFIG_PCI_MSI
2487 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2488 {
2489         int vector;
2490         unsigned dest;
2491
2492         vector = assign_irq_vector(irq);
2493         if (vector >= 0) {
2494                 dest = cpu_mask_to_apicid(TARGET_CPUS);
2495
2496                 msg->address_hi = MSI_ADDR_BASE_HI;
2497                 msg->address_lo =
2498                         MSI_ADDR_BASE_LO |
2499                         ((INT_DEST_MODE == 0) ?
2500                                 MSI_ADDR_DEST_MODE_PHYSICAL:
2501                                 MSI_ADDR_DEST_MODE_LOGICAL) |
2502                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2503                                 MSI_ADDR_REDIRECTION_CPU:
2504                                 MSI_ADDR_REDIRECTION_LOWPRI) |
2505                         MSI_ADDR_DEST_ID(dest);
2506
2507                 msg->data =
2508                         MSI_DATA_TRIGGER_EDGE |
2509                         MSI_DATA_LEVEL_ASSERT |
2510                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2511                                 MSI_DATA_DELIVERY_FIXED:
2512                                 MSI_DATA_DELIVERY_LOWPRI) |
2513                         MSI_DATA_VECTOR(vector);
2514         }
2515         return vector;
2516 }
2517
2518 #ifdef CONFIG_SMP
2519 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2520 {
2521         struct msi_msg msg;
2522         unsigned int dest;
2523         cpumask_t tmp;
2524         int vector;
2525
2526         cpus_and(tmp, mask, cpu_online_map);
2527         if (cpus_empty(tmp))
2528                 tmp = TARGET_CPUS;
2529
2530         vector = assign_irq_vector(irq);
2531         if (vector < 0)
2532                 return;
2533
2534         dest = cpu_mask_to_apicid(mask);
2535
2536         read_msi_msg(irq, &msg);
2537
2538         msg.data &= ~MSI_DATA_VECTOR_MASK;
2539         msg.data |= MSI_DATA_VECTOR(vector);
2540         msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2541         msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2542
2543         write_msi_msg(irq, &msg);
2544         irq_desc[irq].affinity = mask;
2545 }
2546 #endif /* CONFIG_SMP */
2547
2548 /*
2549  * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2550  * which implement the MSI or MSI-X Capability Structure.
2551  */
2552 static struct irq_chip msi_chip = {
2553         .name           = "PCI-MSI",
2554         .unmask         = unmask_msi_irq,
2555         .mask           = mask_msi_irq,
2556         .ack            = ack_ioapic_irq,
2557 #ifdef CONFIG_SMP
2558         .set_affinity   = set_msi_irq_affinity,
2559 #endif
2560         .retrigger      = ioapic_retrigger_irq,
2561 };
2562
2563 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2564 {
2565         struct msi_msg msg;
2566         int irq, ret;
2567         irq = create_irq();
2568         if (irq < 0)
2569                 return irq;
2570
2571         ret = msi_compose_msg(dev, irq, &msg);
2572         if (ret < 0) {
2573                 destroy_irq(irq);
2574                 return ret;
2575         }
2576
2577         set_irq_msi(irq, desc);
2578         write_msi_msg(irq, &msg);
2579
2580         set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq,
2581                                       "edge");
2582
2583         return 0;
2584 }
2585
2586 void arch_teardown_msi_irq(unsigned int irq)
2587 {
2588         destroy_irq(irq);
2589 }
2590
2591 #endif /* CONFIG_PCI_MSI */
2592
2593 /*
2594  * Hypertransport interrupt support
2595  */
2596 #ifdef CONFIG_HT_IRQ
2597
2598 #ifdef CONFIG_SMP
2599
2600 static void target_ht_irq(unsigned int irq, unsigned int dest)
2601 {
2602         struct ht_irq_msg msg;
2603         fetch_ht_irq_msg(irq, &msg);
2604
2605         msg.address_lo &= ~(HT_IRQ_LOW_DEST_ID_MASK);
2606         msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2607
2608         msg.address_lo |= HT_IRQ_LOW_DEST_ID(dest);
2609         msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2610
2611         write_ht_irq_msg(irq, &msg);
2612 }
2613
2614 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2615 {
2616         unsigned int dest;
2617         cpumask_t tmp;
2618
2619         cpus_and(tmp, mask, cpu_online_map);
2620         if (cpus_empty(tmp))
2621                 tmp = TARGET_CPUS;
2622
2623         cpus_and(mask, tmp, CPU_MASK_ALL);
2624
2625         dest = cpu_mask_to_apicid(mask);
2626
2627         target_ht_irq(irq, dest);
2628         irq_desc[irq].affinity = mask;
2629 }
2630 #endif
2631
2632 static struct irq_chip ht_irq_chip = {
2633         .name           = "PCI-HT",
2634         .mask           = mask_ht_irq,
2635         .unmask         = unmask_ht_irq,
2636         .ack            = ack_ioapic_irq,
2637 #ifdef CONFIG_SMP
2638         .set_affinity   = set_ht_irq_affinity,
2639 #endif
2640         .retrigger      = ioapic_retrigger_irq,
2641 };
2642
2643 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2644 {
2645         int vector;
2646
2647         vector = assign_irq_vector(irq);
2648         if (vector >= 0) {
2649                 struct ht_irq_msg msg;
2650                 unsigned dest;
2651                 cpumask_t tmp;
2652
2653                 cpus_clear(tmp);
2654                 cpu_set(vector >> 8, tmp);
2655                 dest = cpu_mask_to_apicid(tmp);
2656
2657                 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2658
2659                 msg.address_lo =
2660                         HT_IRQ_LOW_BASE |
2661                         HT_IRQ_LOW_DEST_ID(dest) |
2662                         HT_IRQ_LOW_VECTOR(vector) |
2663                         ((INT_DEST_MODE == 0) ?
2664                                 HT_IRQ_LOW_DM_PHYSICAL :
2665                                 HT_IRQ_LOW_DM_LOGICAL) |
2666                         HT_IRQ_LOW_RQEOI_EDGE |
2667                         ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2668                                 HT_IRQ_LOW_MT_FIXED :
2669                                 HT_IRQ_LOW_MT_ARBITRATED) |
2670                         HT_IRQ_LOW_IRQ_MASKED;
2671
2672                 write_ht_irq_msg(irq, &msg);
2673
2674                 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2675                                               handle_edge_irq, "edge");
2676         }
2677         return vector;
2678 }
2679 #endif /* CONFIG_HT_IRQ */
2680
2681 /* --------------------------------------------------------------------------
2682                           ACPI-based IOAPIC Configuration
2683    -------------------------------------------------------------------------- */
2684
2685 #ifdef CONFIG_ACPI
2686
2687 int __init io_apic_get_unique_id (int ioapic, int apic_id)
2688 {
2689         union IO_APIC_reg_00 reg_00;
2690         static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2691         physid_mask_t tmp;
2692         unsigned long flags;
2693         int i = 0;
2694
2695         /*
2696          * The P4 platform supports up to 256 APIC IDs on two separate APIC
2697          * buses (one for LAPICs, one for IOAPICs), where predecessors only
2698          * supports up to 16 on one shared APIC bus.
2699          *
2700          * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2701          *      advantage of new APIC bus architecture.
2702          */
2703
2704         if (physids_empty(apic_id_map))
2705                 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2706
2707         spin_lock_irqsave(&ioapic_lock, flags);
2708         reg_00.raw = io_apic_read(ioapic, 0);
2709         spin_unlock_irqrestore(&ioapic_lock, flags);
2710
2711         if (apic_id >= get_physical_broadcast()) {
2712                 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2713                         "%d\n", ioapic, apic_id, reg_00.bits.ID);
2714                 apic_id = reg_00.bits.ID;
2715         }
2716
2717         /*
2718          * Every APIC in a system must have a unique ID or we get lots of nice
2719          * 'stuck on smp_invalidate_needed IPI wait' messages.
2720          */
2721         if (check_apicid_used(apic_id_map, apic_id)) {
2722
2723                 for (i = 0; i < get_physical_broadcast(); i++) {
2724                         if (!check_apicid_used(apic_id_map, i))
2725                                 break;
2726                 }
2727
2728                 if (i == get_physical_broadcast())
2729                         panic("Max apic_id exceeded!\n");
2730
2731                 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2732                         "trying %d\n", ioapic, apic_id, i);
2733
2734                 apic_id = i;
2735         }
2736
2737         tmp = apicid_to_cpu_present(apic_id);
2738         physids_or(apic_id_map, apic_id_map, tmp);
2739
2740         if (reg_00.bits.ID != apic_id) {
2741                 reg_00.bits.ID = apic_id;
2742
2743                 spin_lock_irqsave(&ioapic_lock, flags);
2744                 io_apic_write(ioapic, 0, reg_00.raw);
2745                 reg_00.raw = io_apic_read(ioapic, 0);
2746                 spin_unlock_irqrestore(&ioapic_lock, flags);
2747
2748                 /* Sanity check */
2749                 if (reg_00.bits.ID != apic_id) {
2750                         printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
2751                         return -1;
2752                 }
2753         }
2754
2755         apic_printk(APIC_VERBOSE, KERN_INFO
2756                         "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2757
2758         return apic_id;
2759 }
2760
2761
2762 int __init io_apic_get_version (int ioapic)
2763 {
2764         union IO_APIC_reg_01    reg_01;
2765         unsigned long flags;
2766
2767         spin_lock_irqsave(&ioapic_lock, flags);
2768         reg_01.raw = io_apic_read(ioapic, 1);
2769         spin_unlock_irqrestore(&ioapic_lock, flags);
2770
2771         return reg_01.bits.version;
2772 }
2773
2774
2775 int __init io_apic_get_redir_entries (int ioapic)
2776 {
2777         union IO_APIC_reg_01    reg_01;
2778         unsigned long flags;
2779
2780         spin_lock_irqsave(&ioapic_lock, flags);
2781         reg_01.raw = io_apic_read(ioapic, 1);
2782         spin_unlock_irqrestore(&ioapic_lock, flags);
2783
2784         return reg_01.bits.entries;
2785 }
2786
2787
2788 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2789 {
2790         struct IO_APIC_route_entry entry;
2791         unsigned long flags;
2792
2793         if (!IO_APIC_IRQ(irq)) {
2794                 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2795                         ioapic);
2796                 return -EINVAL;
2797         }
2798
2799         /*
2800          * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2801          * Note that we mask (disable) IRQs now -- these get enabled when the
2802          * corresponding device driver registers for this IRQ.
2803          */
2804
2805         memset(&entry,0,sizeof(entry));
2806
2807         entry.delivery_mode = INT_DELIVERY_MODE;
2808         entry.dest_mode = INT_DEST_MODE;
2809         entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2810         entry.trigger = edge_level;
2811         entry.polarity = active_high_low;
2812         entry.mask  = 1;
2813
2814         /*
2815          * IRQs < 16 are already in the irq_2_pin[] map
2816          */
2817         if (irq >= 16)
2818                 add_pin_to_irq(irq, ioapic, pin);
2819
2820         entry.vector = assign_irq_vector(irq);
2821
2822         apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2823                 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2824                 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2825                 edge_level, active_high_low);
2826
2827         ioapic_register_intr(irq, entry.vector, edge_level);
2828
2829         if (!ioapic && (irq < 16))
2830                 disable_8259A_irq(irq);
2831
2832         spin_lock_irqsave(&ioapic_lock, flags);
2833         __ioapic_write_entry(ioapic, pin, entry);
2834         spin_unlock_irqrestore(&ioapic_lock, flags);
2835
2836         return 0;
2837 }
2838
2839 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2840 {
2841         int i;
2842
2843         if (skip_ioapic_setup)
2844                 return -1;
2845
2846         for (i = 0; i < mp_irq_entries; i++)
2847                 if (mp_irqs[i].mpc_irqtype == mp_INT &&
2848                     mp_irqs[i].mpc_srcbusirq == bus_irq)
2849                         break;
2850         if (i >= mp_irq_entries)
2851                 return -1;
2852
2853         *trigger = irq_trigger(i);
2854         *polarity = irq_polarity(i);
2855         return 0;
2856 }
2857
2858 #endif /* CONFIG_ACPI */
2859
2860 static int __init parse_disable_timer_pin_1(char *arg)
2861 {
2862         disable_timer_pin_1 = 1;
2863         return 0;
2864 }
2865 early_param("disable_timer_pin_1", parse_disable_timer_pin_1);
2866
2867 static int __init parse_enable_timer_pin_1(char *arg)
2868 {
2869         disable_timer_pin_1 = -1;
2870         return 0;
2871 }
2872 early_param("enable_timer_pin_1", parse_enable_timer_pin_1);
2873
2874 static int __init parse_noapic(char *arg)
2875 {
2876         /* disable IO-APIC */
2877         disable_ioapic_setup();
2878         return 0;
2879 }
2880 early_param("noapic", parse_noapic);