linux/arch/x86/kernel/apic/vector.c
<<
>>
Prefs
   1/*
   2 * Local APIC related interfaces to support IOAPIC, MSI, HT_IRQ etc.
   3 *
   4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
   5 *      Moved from arch/x86/kernel/apic/io_apic.c.
   6 * Jiang Liu <jiang.liu@linux.intel.com>
   7 *      Enable support of hierarchical irqdomains
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License version 2 as
  11 * published by the Free Software Foundation.
  12 */
  13#include <linux/interrupt.h>
  14#include <linux/init.h>
  15#include <linux/compiler.h>
  16#include <linux/slab.h>
  17#include <asm/irqdomain.h>
  18#include <asm/hw_irq.h>
  19#include <asm/apic.h>
  20#include <asm/i8259.h>
  21#include <asm/desc.h>
  22#include <asm/irq_remapping.h>
  23
  24struct apic_chip_data {
  25        struct irq_cfg          cfg;
  26        cpumask_var_t           domain;
  27        cpumask_var_t           old_domain;
  28        u8                      move_in_progress : 1;
  29};
  30
  31struct irq_domain *x86_vector_domain;
  32EXPORT_SYMBOL_GPL(x86_vector_domain);
  33static DEFINE_RAW_SPINLOCK(vector_lock);
  34static cpumask_var_t vector_cpumask, vector_searchmask, searched_cpumask;
  35static struct irq_chip lapic_controller;
  36#ifdef  CONFIG_X86_IO_APIC
  37static struct apic_chip_data *legacy_irq_data[NR_IRQS_LEGACY];
  38#endif
  39
  40void lock_vector_lock(void)
  41{
  42        /* Used to the online set of cpus does not change
  43         * during assign_irq_vector.
  44         */
  45        raw_spin_lock(&vector_lock);
  46}
  47
  48void unlock_vector_lock(void)
  49{
  50        raw_spin_unlock(&vector_lock);
  51}
  52
  53static struct apic_chip_data *apic_chip_data(struct irq_data *irq_data)
  54{
  55        if (!irq_data)
  56                return NULL;
  57
  58        while (irq_data->parent_data)
  59                irq_data = irq_data->parent_data;
  60
  61        return irq_data->chip_data;
  62}
  63
  64struct irq_cfg *irqd_cfg(struct irq_data *irq_data)
  65{
  66        struct apic_chip_data *data = apic_chip_data(irq_data);
  67
  68        return data ? &data->cfg : NULL;
  69}
  70EXPORT_SYMBOL_GPL(irqd_cfg);
  71
  72struct irq_cfg *irq_cfg(unsigned int irq)
  73{
  74        return irqd_cfg(irq_get_irq_data(irq));
  75}
  76
  77static struct apic_chip_data *alloc_apic_chip_data(int node)
  78{
  79        struct apic_chip_data *data;
  80
  81        data = kzalloc_node(sizeof(*data), GFP_KERNEL, node);
  82        if (!data)
  83                return NULL;
  84        if (!zalloc_cpumask_var_node(&data->domain, GFP_KERNEL, node))
  85                goto out_data;
  86        if (!zalloc_cpumask_var_node(&data->old_domain, GFP_KERNEL, node))
  87                goto out_domain;
  88        return data;
  89out_domain:
  90        free_cpumask_var(data->domain);
  91out_data:
  92        kfree(data);
  93        return NULL;
  94}
  95
  96static void free_apic_chip_data(struct apic_chip_data *data)
  97{
  98        if (data) {
  99                free_cpumask_var(data->domain);
 100                free_cpumask_var(data->old_domain);
 101                kfree(data);
 102        }
 103}
 104
 105static int __assign_irq_vector(int irq, struct apic_chip_data *d,
 106                               const struct cpumask *mask)
 107{
 108        /*
 109         * NOTE! The local APIC isn't very good at handling
 110         * multiple interrupts at the same interrupt level.
 111         * As the interrupt level is determined by taking the
 112         * vector number and shifting that right by 4, we
 113         * want to spread these out a bit so that they don't
 114         * all fall in the same interrupt level.
 115         *
 116         * Also, we've got to be careful not to trash gate
 117         * 0x80, because int 0x80 is hm, kind of importantish. ;)
 118         */
 119        static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
 120        static int current_offset = VECTOR_OFFSET_START % 16;
 121        int cpu, vector;
 122
 123        /*
 124         * If there is still a move in progress or the previous move has not
 125         * been cleaned up completely, tell the caller to come back later.
 126         */
 127        if (d->move_in_progress ||
 128            cpumask_intersects(d->old_domain, cpu_online_mask))
 129                return -EBUSY;
 130
 131        /* Only try and allocate irqs on cpus that are present */
 132        cpumask_clear(d->old_domain);
 133        cpumask_clear(searched_cpumask);
 134        cpu = cpumask_first_and(mask, cpu_online_mask);
 135        while (cpu < nr_cpu_ids) {
 136                int new_cpu, offset;
 137
 138                /* Get the possible target cpus for @mask/@cpu from the apic */
 139                apic->vector_allocation_domain(cpu, vector_cpumask, mask);
 140
 141                /*
 142                 * Clear the offline cpus from @vector_cpumask for searching
 143                 * and verify whether the result overlaps with @mask. If true,
 144                 * then the call to apic->cpu_mask_to_apicid_and() will
 145                 * succeed as well. If not, no point in trying to find a
 146                 * vector in this mask.
 147                 */
 148                cpumask_and(vector_searchmask, vector_cpumask, cpu_online_mask);
 149                if (!cpumask_intersects(vector_searchmask, mask))
 150                        goto next_cpu;
 151
 152                if (cpumask_subset(vector_cpumask, d->domain)) {
 153                        if (cpumask_equal(vector_cpumask, d->domain))
 154                                goto success;
 155                        /*
 156                         * Mark the cpus which are not longer in the mask for
 157                         * cleanup.
 158                         */
 159                        cpumask_andnot(d->old_domain, d->domain, vector_cpumask);
 160                        vector = d->cfg.vector;
 161                        goto update;
 162                }
 163
 164                vector = current_vector;
 165                offset = current_offset;
 166next:
 167                vector += 16;
 168                if (vector >= first_system_vector) {
 169                        offset = (offset + 1) % 16;
 170                        vector = FIRST_EXTERNAL_VECTOR + offset;
 171                }
 172
 173                /* If the search wrapped around, try the next cpu */
 174                if (unlikely(current_vector == vector))
 175                        goto next_cpu;
 176
 177                if (test_bit(vector, used_vectors))
 178                        goto next;
 179
 180                for_each_cpu(new_cpu, vector_searchmask) {
 181                        if (!IS_ERR_OR_NULL(per_cpu(vector_irq, new_cpu)[vector]))
 182                                goto next;
 183                }
 184                /* Found one! */
 185                current_vector = vector;
 186                current_offset = offset;
 187                /* Schedule the old vector for cleanup on all cpus */
 188                if (d->cfg.vector)
 189                        cpumask_copy(d->old_domain, d->domain);
 190                for_each_cpu(new_cpu, vector_searchmask)
 191                        per_cpu(vector_irq, new_cpu)[vector] = irq_to_desc(irq);
 192                goto update;
 193
 194next_cpu:
 195                /*
 196                 * We exclude the current @vector_cpumask from the requested
 197                 * @mask and try again with the next online cpu in the
 198                 * result. We cannot modify @mask, so we use @vector_cpumask
 199                 * as a temporary buffer here as it will be reassigned when
 200                 * calling apic->vector_allocation_domain() above.
 201                 */
 202                cpumask_or(searched_cpumask, searched_cpumask, vector_cpumask);
 203                cpumask_andnot(vector_cpumask, mask, searched_cpumask);
 204                cpu = cpumask_first_and(vector_cpumask, cpu_online_mask);
 205                continue;
 206        }
 207        return -ENOSPC;
 208
 209update:
 210        /*
 211         * Exclude offline cpus from the cleanup mask and set the
 212         * move_in_progress flag when the result is not empty.
 213         */
 214        cpumask_and(d->old_domain, d->old_domain, cpu_online_mask);
 215        d->move_in_progress = !cpumask_empty(d->old_domain);
 216        d->cfg.old_vector = d->move_in_progress ? d->cfg.vector : 0;
 217        d->cfg.vector = vector;
 218        cpumask_copy(d->domain, vector_cpumask);
 219success:
 220        /*
 221         * Cache destination APIC IDs into cfg->dest_apicid. This cannot fail
 222         * as we already established, that mask & d->domain & cpu_online_mask
 223         * is not empty.
 224         */
 225        BUG_ON(apic->cpu_mask_to_apicid_and(mask, d->domain,
 226                                            &d->cfg.dest_apicid));
 227        return 0;
 228}
 229
 230static int assign_irq_vector(int irq, struct apic_chip_data *data,
 231                             const struct cpumask *mask)
 232{
 233        int err;
 234        unsigned long flags;
 235
 236        raw_spin_lock_irqsave(&vector_lock, flags);
 237        err = __assign_irq_vector(irq, data, mask);
 238        raw_spin_unlock_irqrestore(&vector_lock, flags);
 239        return err;
 240}
 241
 242static int assign_irq_vector_policy(int irq, int node,
 243                                    struct apic_chip_data *data,
 244                                    struct irq_alloc_info *info)
 245{
 246        if (info && info->mask)
 247                return assign_irq_vector(irq, data, info->mask);
 248        if (node != NUMA_NO_NODE &&
 249            assign_irq_vector(irq, data, cpumask_of_node(node)) == 0)
 250                return 0;
 251        return assign_irq_vector(irq, data, apic->target_cpus());
 252}
 253
 254static void clear_irq_vector(int irq, struct apic_chip_data *data)
 255{
 256        struct irq_desc *desc;
 257        int cpu, vector;
 258
 259        if (!data->cfg.vector)
 260                return;
 261
 262        vector = data->cfg.vector;
 263        for_each_cpu_and(cpu, data->domain, cpu_online_mask)
 264                per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
 265
 266        data->cfg.vector = 0;
 267        cpumask_clear(data->domain);
 268
 269        /*
 270         * If move is in progress or the old_domain mask is not empty,
 271         * i.e. the cleanup IPI has not been processed yet, we need to remove
 272         * the old references to desc from all cpus vector tables.
 273         */
 274        if (!data->move_in_progress && cpumask_empty(data->old_domain))
 275                return;
 276
 277        desc = irq_to_desc(irq);
 278        for_each_cpu_and(cpu, data->old_domain, cpu_online_mask) {
 279                for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
 280                     vector++) {
 281                        if (per_cpu(vector_irq, cpu)[vector] != desc)
 282                                continue;
 283                        per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
 284                        break;
 285                }
 286        }
 287        data->move_in_progress = 0;
 288}
 289
 290void init_irq_alloc_info(struct irq_alloc_info *info,
 291                         const struct cpumask *mask)
 292{
 293        memset(info, 0, sizeof(*info));
 294        info->mask = mask;
 295}
 296
 297void copy_irq_alloc_info(struct irq_alloc_info *dst, struct irq_alloc_info *src)
 298{
 299        if (src)
 300                *dst = *src;
 301        else
 302                memset(dst, 0, sizeof(*dst));
 303}
 304
 305static void x86_vector_free_irqs(struct irq_domain *domain,
 306                                 unsigned int virq, unsigned int nr_irqs)
 307{
 308        struct apic_chip_data *apic_data;
 309        struct irq_data *irq_data;
 310        unsigned long flags;
 311        int i;
 312
 313        for (i = 0; i < nr_irqs; i++) {
 314                irq_data = irq_domain_get_irq_data(x86_vector_domain, virq + i);
 315                if (irq_data && irq_data->chip_data) {
 316                        raw_spin_lock_irqsave(&vector_lock, flags);
 317                        clear_irq_vector(virq + i, irq_data->chip_data);
 318                        apic_data = irq_data->chip_data;
 319                        irq_domain_reset_irq_data(irq_data);
 320                        raw_spin_unlock_irqrestore(&vector_lock, flags);
 321                        free_apic_chip_data(apic_data);
 322#ifdef  CONFIG_X86_IO_APIC
 323                        if (virq + i < nr_legacy_irqs())
 324                                legacy_irq_data[virq + i] = NULL;
 325#endif
 326                }
 327        }
 328}
 329
 330static int x86_vector_alloc_irqs(struct irq_domain *domain, unsigned int virq,
 331                                 unsigned int nr_irqs, void *arg)
 332{
 333        struct irq_alloc_info *info = arg;
 334        struct apic_chip_data *data;
 335        struct irq_data *irq_data;
 336        int i, err, node;
 337
 338        if (disable_apic)
 339                return -ENXIO;
 340
 341        /* Currently vector allocator can't guarantee contiguous allocations */
 342        if ((info->flags & X86_IRQ_ALLOC_CONTIGUOUS_VECTORS) && nr_irqs > 1)
 343                return -ENOSYS;
 344
 345        for (i = 0; i < nr_irqs; i++) {
 346                irq_data = irq_domain_get_irq_data(domain, virq + i);
 347                BUG_ON(!irq_data);
 348                node = irq_data_get_node(irq_data);
 349#ifdef  CONFIG_X86_IO_APIC
 350                if (virq + i < nr_legacy_irqs() && legacy_irq_data[virq + i])
 351                        data = legacy_irq_data[virq + i];
 352                else
 353#endif
 354                        data = alloc_apic_chip_data(node);
 355                if (!data) {
 356                        err = -ENOMEM;
 357                        goto error;
 358                }
 359
 360                irq_data->chip = &lapic_controller;
 361                irq_data->chip_data = data;
 362                irq_data->hwirq = virq + i;
 363                err = assign_irq_vector_policy(virq + i, node, data, info);
 364                if (err)
 365                        goto error;
 366        }
 367
 368        return 0;
 369
 370error:
 371        x86_vector_free_irqs(domain, virq, i + 1);
 372        return err;
 373}
 374
 375static const struct irq_domain_ops x86_vector_domain_ops = {
 376        .alloc  = x86_vector_alloc_irqs,
 377        .free   = x86_vector_free_irqs,
 378};
 379
 380int __init arch_probe_nr_irqs(void)
 381{
 382        int nr;
 383
 384        if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
 385                nr_irqs = NR_VECTORS * nr_cpu_ids;
 386
 387        nr = (gsi_top + nr_legacy_irqs()) + 8 * nr_cpu_ids;
 388#if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
 389        /*
 390         * for MSI and HT dyn irq
 391         */
 392        if (gsi_top <= NR_IRQS_LEGACY)
 393                nr +=  8 * nr_cpu_ids;
 394        else
 395                nr += gsi_top * 16;
 396#endif
 397        if (nr < nr_irqs)
 398                nr_irqs = nr;
 399
 400        /*
 401         * We don't know if PIC is present at this point so we need to do
 402         * probe() to get the right number of legacy IRQs.
 403         */
 404        return legacy_pic->probe();
 405}
 406
 407#ifdef  CONFIG_X86_IO_APIC
 408static void init_legacy_irqs(void)
 409{
 410        int i, node = cpu_to_node(0);
 411        struct apic_chip_data *data;
 412
 413        /*
 414         * For legacy IRQ's, start with assigning irq0 to irq15 to
 415         * ISA_IRQ_VECTOR(i) for all cpu's.
 416         */
 417        for (i = 0; i < nr_legacy_irqs(); i++) {
 418                data = legacy_irq_data[i] = alloc_apic_chip_data(node);
 419                BUG_ON(!data);
 420
 421                data->cfg.vector = ISA_IRQ_VECTOR(i);
 422                cpumask_setall(data->domain);
 423                irq_set_chip_data(i, data);
 424        }
 425}
 426#else
 427static void init_legacy_irqs(void) { }
 428#endif
 429
 430int __init arch_early_irq_init(void)
 431{
 432        init_legacy_irqs();
 433
 434        x86_vector_domain = irq_domain_add_tree(NULL, &x86_vector_domain_ops,
 435                                                NULL);
 436        BUG_ON(x86_vector_domain == NULL);
 437        irq_set_default_host(x86_vector_domain);
 438
 439        arch_init_msi_domain(x86_vector_domain);
 440        arch_init_htirq_domain(x86_vector_domain);
 441
 442        BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
 443        BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
 444        BUG_ON(!alloc_cpumask_var(&searched_cpumask, GFP_KERNEL));
 445
 446        return arch_early_ioapic_init();
 447}
 448
 449/* Initialize vector_irq on a new cpu */
 450static void __setup_vector_irq(int cpu)
 451{
 452        struct apic_chip_data *data;
 453        struct irq_desc *desc;
 454        int irq, vector;
 455
 456        /* Mark the inuse vectors */
 457        for_each_irq_desc(irq, desc) {
 458                struct irq_data *idata = irq_desc_get_irq_data(desc);
 459
 460                data = apic_chip_data(idata);
 461                if (!data || !cpumask_test_cpu(cpu, data->domain))
 462                        continue;
 463                vector = data->cfg.vector;
 464                per_cpu(vector_irq, cpu)[vector] = desc;
 465        }
 466        /* Mark the free vectors */
 467        for (vector = 0; vector < NR_VECTORS; ++vector) {
 468                desc = per_cpu(vector_irq, cpu)[vector];
 469                if (IS_ERR_OR_NULL(desc))
 470                        continue;
 471
 472                data = apic_chip_data(irq_desc_get_irq_data(desc));
 473                if (!cpumask_test_cpu(cpu, data->domain))
 474                        per_cpu(vector_irq, cpu)[vector] = VECTOR_UNUSED;
 475        }
 476}
 477
 478/*
 479 * Setup the vector to irq mappings. Must be called with vector_lock held.
 480 */
 481void setup_vector_irq(int cpu)
 482{
 483        int irq;
 484
 485        lockdep_assert_held(&vector_lock);
 486        /*
 487         * On most of the platforms, legacy PIC delivers the interrupts on the
 488         * boot cpu. But there are certain platforms where PIC interrupts are
 489         * delivered to multiple cpu's. If the legacy IRQ is handled by the
 490         * legacy PIC, for the new cpu that is coming online, setup the static
 491         * legacy vector to irq mapping:
 492         */
 493        for (irq = 0; irq < nr_legacy_irqs(); irq++)
 494                per_cpu(vector_irq, cpu)[ISA_IRQ_VECTOR(irq)] = irq_to_desc(irq);
 495
 496        __setup_vector_irq(cpu);
 497}
 498
 499static int apic_retrigger_irq(struct irq_data *irq_data)
 500{
 501        struct apic_chip_data *data = apic_chip_data(irq_data);
 502        unsigned long flags;
 503        int cpu;
 504
 505        raw_spin_lock_irqsave(&vector_lock, flags);
 506        cpu = cpumask_first_and(data->domain, cpu_online_mask);
 507        apic->send_IPI_mask(cpumask_of(cpu), data->cfg.vector);
 508        raw_spin_unlock_irqrestore(&vector_lock, flags);
 509
 510        return 1;
 511}
 512
 513void apic_ack_edge(struct irq_data *data)
 514{
 515        irq_complete_move(irqd_cfg(data));
 516        irq_move_irq(data);
 517        ack_APIC_irq();
 518}
 519
 520static int apic_set_affinity(struct irq_data *irq_data,
 521                             const struct cpumask *dest, bool force)
 522{
 523        struct apic_chip_data *data = irq_data->chip_data;
 524        int err, irq = irq_data->irq;
 525
 526        if (!config_enabled(CONFIG_SMP))
 527                return -EPERM;
 528
 529        if (!cpumask_intersects(dest, cpu_online_mask))
 530                return -EINVAL;
 531
 532        err = assign_irq_vector(irq, data, dest);
 533        return err ? err : IRQ_SET_MASK_OK;
 534}
 535
 536static struct irq_chip lapic_controller = {
 537        .irq_ack                = apic_ack_edge,
 538        .irq_set_affinity       = apic_set_affinity,
 539        .irq_retrigger          = apic_retrigger_irq,
 540};
 541
 542#ifdef CONFIG_SMP
 543static void __send_cleanup_vector(struct apic_chip_data *data)
 544{
 545        raw_spin_lock(&vector_lock);
 546        cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
 547        data->move_in_progress = 0;
 548        if (!cpumask_empty(data->old_domain))
 549                apic->send_IPI_mask(data->old_domain, IRQ_MOVE_CLEANUP_VECTOR);
 550        raw_spin_unlock(&vector_lock);
 551}
 552
 553void send_cleanup_vector(struct irq_cfg *cfg)
 554{
 555        struct apic_chip_data *data;
 556
 557        data = container_of(cfg, struct apic_chip_data, cfg);
 558        if (data->move_in_progress)
 559                __send_cleanup_vector(data);
 560}
 561
 562asmlinkage __visible void smp_irq_move_cleanup_interrupt(void)
 563{
 564        unsigned vector, me;
 565
 566        entering_ack_irq();
 567
 568        /* Prevent vectors vanishing under us */
 569        raw_spin_lock(&vector_lock);
 570
 571        me = smp_processor_id();
 572        for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
 573                struct apic_chip_data *data;
 574                struct irq_desc *desc;
 575                unsigned int irr;
 576
 577        retry:
 578                desc = __this_cpu_read(vector_irq[vector]);
 579                if (IS_ERR_OR_NULL(desc))
 580                        continue;
 581
 582                if (!raw_spin_trylock(&desc->lock)) {
 583                        raw_spin_unlock(&vector_lock);
 584                        cpu_relax();
 585                        raw_spin_lock(&vector_lock);
 586                        goto retry;
 587                }
 588
 589                data = apic_chip_data(irq_desc_get_irq_data(desc));
 590                if (!data)
 591                        goto unlock;
 592
 593                /*
 594                 * Nothing to cleanup if irq migration is in progress
 595                 * or this cpu is not set in the cleanup mask.
 596                 */
 597                if (data->move_in_progress ||
 598                    !cpumask_test_cpu(me, data->old_domain))
 599                        goto unlock;
 600
 601                /*
 602                 * We have two cases to handle here:
 603                 * 1) vector is unchanged but the target mask got reduced
 604                 * 2) vector and the target mask has changed
 605                 *
 606                 * #1 is obvious, but in #2 we have two vectors with the same
 607                 * irq descriptor: the old and the new vector. So we need to
 608                 * make sure that we only cleanup the old vector. The new
 609                 * vector has the current @vector number in the config and
 610                 * this cpu is part of the target mask. We better leave that
 611                 * one alone.
 612                 */
 613                if (vector == data->cfg.vector &&
 614                    cpumask_test_cpu(me, data->domain))
 615                        goto unlock;
 616
 617                irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
 618                /*
 619                 * Check if the vector that needs to be cleanedup is
 620                 * registered at the cpu's IRR. If so, then this is not
 621                 * the best time to clean it up. Lets clean it up in the
 622                 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
 623                 * to myself.
 624                 */
 625                if (irr  & (1 << (vector % 32))) {
 626                        apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
 627                        goto unlock;
 628                }
 629                __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
 630                cpumask_clear_cpu(me, data->old_domain);
 631unlock:
 632                raw_spin_unlock(&desc->lock);
 633        }
 634
 635        raw_spin_unlock(&vector_lock);
 636
 637        exiting_irq();
 638}
 639
 640static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
 641{
 642        unsigned me;
 643        struct apic_chip_data *data;
 644
 645        data = container_of(cfg, struct apic_chip_data, cfg);
 646        if (likely(!data->move_in_progress))
 647                return;
 648
 649        me = smp_processor_id();
 650        if (vector == data->cfg.vector && cpumask_test_cpu(me, data->domain))
 651                __send_cleanup_vector(data);
 652}
 653
 654void irq_complete_move(struct irq_cfg *cfg)
 655{
 656        __irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
 657}
 658
 659/*
 660 * Called from fixup_irqs() with @desc->lock held and interrupts disabled.
 661 */
 662void irq_force_complete_move(struct irq_desc *desc)
 663{
 664        struct irq_data *irqdata = irq_desc_get_irq_data(desc);
 665        struct apic_chip_data *data = apic_chip_data(irqdata);
 666        struct irq_cfg *cfg = data ? &data->cfg : NULL;
 667        unsigned int cpu;
 668
 669        if (!cfg)
 670                return;
 671
 672        /*
 673         * This is tricky. If the cleanup of @data->old_domain has not been
 674         * done yet, then the following setaffinity call will fail with
 675         * -EBUSY. This can leave the interrupt in a stale state.
 676         *
 677         * All CPUs are stuck in stop machine with interrupts disabled so
 678         * calling __irq_complete_move() would be completely pointless.
 679         */
 680        raw_spin_lock(&vector_lock);
 681        /*
 682         * Clean out all offline cpus (including the outgoing one) from the
 683         * old_domain mask.
 684         */
 685        cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
 686
 687        /*
 688         * If move_in_progress is cleared and the old_domain mask is empty,
 689         * then there is nothing to cleanup. fixup_irqs() will take care of
 690         * the stale vectors on the outgoing cpu.
 691         */
 692        if (!data->move_in_progress && cpumask_empty(data->old_domain)) {
 693                raw_spin_unlock(&vector_lock);
 694                return;
 695        }
 696
 697        /*
 698         * 1) The interrupt is in move_in_progress state. That means that we
 699         *    have not seen an interrupt since the io_apic was reprogrammed to
 700         *    the new vector.
 701         *
 702         * 2) The interrupt has fired on the new vector, but the cleanup IPIs
 703         *    have not been processed yet.
 704         */
 705        if (data->move_in_progress) {
 706                /*
 707                 * In theory there is a race:
 708                 *
 709                 * set_ioapic(new_vector) <-- Interrupt is raised before update
 710                 *                            is effective, i.e. it's raised on
 711                 *                            the old vector.
 712                 *
 713                 * So if the target cpu cannot handle that interrupt before
 714                 * the old vector is cleaned up, we get a spurious interrupt
 715                 * and in the worst case the ioapic irq line becomes stale.
 716                 *
 717                 * But in case of cpu hotplug this should be a non issue
 718                 * because if the affinity update happens right before all
 719                 * cpus rendevouz in stop machine, there is no way that the
 720                 * interrupt can be blocked on the target cpu because all cpus
 721                 * loops first with interrupts enabled in stop machine, so the
 722                 * old vector is not yet cleaned up when the interrupt fires.
 723                 *
 724                 * So the only way to run into this issue is if the delivery
 725                 * of the interrupt on the apic/system bus would be delayed
 726                 * beyond the point where the target cpu disables interrupts
 727                 * in stop machine. I doubt that it can happen, but at least
 728                 * there is a theroretical chance. Virtualization might be
 729                 * able to expose this, but AFAICT the IOAPIC emulation is not
 730                 * as stupid as the real hardware.
 731                 *
 732                 * Anyway, there is nothing we can do about that at this point
 733                 * w/o refactoring the whole fixup_irq() business completely.
 734                 * We print at least the irq number and the old vector number,
 735                 * so we have the necessary information when a problem in that
 736                 * area arises.
 737                 */
 738                pr_warn("IRQ fixup: irq %d move in progress, old vector %d\n",
 739                        irqdata->irq, cfg->old_vector);
 740        }
 741        /*
 742         * If old_domain is not empty, then other cpus still have the irq
 743         * descriptor set in their vector array. Clean it up.
 744         */
 745        for_each_cpu(cpu, data->old_domain)
 746                per_cpu(vector_irq, cpu)[cfg->old_vector] = VECTOR_UNUSED;
 747
 748        /* Cleanup the left overs of the (half finished) move */
 749        cpumask_clear(data->old_domain);
 750        data->move_in_progress = 0;
 751        raw_spin_unlock(&vector_lock);
 752}
 753#endif
 754
 755static void __init print_APIC_field(int base)
 756{
 757        int i;
 758
 759        printk(KERN_DEBUG);
 760
 761        for (i = 0; i < 8; i++)
 762                pr_cont("%08x", apic_read(base + i*0x10));
 763
 764        pr_cont("\n");
 765}
 766
 767static void __init print_local_APIC(void *dummy)
 768{
 769        unsigned int i, v, ver, maxlvt;
 770        u64 icr;
 771
 772        pr_debug("printing local APIC contents on CPU#%d/%d:\n",
 773                 smp_processor_id(), hard_smp_processor_id());
 774        v = apic_read(APIC_ID);
 775        pr_info("... APIC ID:      %08x (%01x)\n", v, read_apic_id());
 776        v = apic_read(APIC_LVR);
 777        pr_info("... APIC VERSION: %08x\n", v);
 778        ver = GET_APIC_VERSION(v);
 779        maxlvt = lapic_get_maxlvt();
 780
 781        v = apic_read(APIC_TASKPRI);
 782        pr_debug("... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
 783
 784        /* !82489DX */
 785        if (APIC_INTEGRATED(ver)) {
 786                if (!APIC_XAPIC(ver)) {
 787                        v = apic_read(APIC_ARBPRI);
 788                        pr_debug("... APIC ARBPRI: %08x (%02x)\n",
 789                                 v, v & APIC_ARBPRI_MASK);
 790                }
 791                v = apic_read(APIC_PROCPRI);
 792                pr_debug("... APIC PROCPRI: %08x\n", v);
 793        }
 794
 795        /*
 796         * Remote read supported only in the 82489DX and local APIC for
 797         * Pentium processors.
 798         */
 799        if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
 800                v = apic_read(APIC_RRR);
 801                pr_debug("... APIC RRR: %08x\n", v);
 802        }
 803
 804        v = apic_read(APIC_LDR);
 805        pr_debug("... APIC LDR: %08x\n", v);
 806        if (!x2apic_enabled()) {
 807                v = apic_read(APIC_DFR);
 808                pr_debug("... APIC DFR: %08x\n", v);
 809        }
 810        v = apic_read(APIC_SPIV);
 811        pr_debug("... APIC SPIV: %08x\n", v);
 812
 813        pr_debug("... APIC ISR field:\n");
 814        print_APIC_field(APIC_ISR);
 815        pr_debug("... APIC TMR field:\n");
 816        print_APIC_field(APIC_TMR);
 817        pr_debug("... APIC IRR field:\n");
 818        print_APIC_field(APIC_IRR);
 819
 820        /* !82489DX */
 821        if (APIC_INTEGRATED(ver)) {
 822                /* Due to the Pentium erratum 3AP. */
 823                if (maxlvt > 3)
 824                        apic_write(APIC_ESR, 0);
 825
 826                v = apic_read(APIC_ESR);
 827                pr_debug("... APIC ESR: %08x\n", v);
 828        }
 829
 830        icr = apic_icr_read();
 831        pr_debug("... APIC ICR: %08x\n", (u32)icr);
 832        pr_debug("... APIC ICR2: %08x\n", (u32)(icr >> 32));
 833
 834        v = apic_read(APIC_LVTT);
 835        pr_debug("... APIC LVTT: %08x\n", v);
 836
 837        if (maxlvt > 3) {
 838                /* PC is LVT#4. */
 839                v = apic_read(APIC_LVTPC);
 840                pr_debug("... APIC LVTPC: %08x\n", v);
 841        }
 842        v = apic_read(APIC_LVT0);
 843        pr_debug("... APIC LVT0: %08x\n", v);
 844        v = apic_read(APIC_LVT1);
 845        pr_debug("... APIC LVT1: %08x\n", v);
 846
 847        if (maxlvt > 2) {
 848                /* ERR is LVT#3. */
 849                v = apic_read(APIC_LVTERR);
 850                pr_debug("... APIC LVTERR: %08x\n", v);
 851        }
 852
 853        v = apic_read(APIC_TMICT);
 854        pr_debug("... APIC TMICT: %08x\n", v);
 855        v = apic_read(APIC_TMCCT);
 856        pr_debug("... APIC TMCCT: %08x\n", v);
 857        v = apic_read(APIC_TDCR);
 858        pr_debug("... APIC TDCR: %08x\n", v);
 859
 860        if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
 861                v = apic_read(APIC_EFEAT);
 862                maxlvt = (v >> 16) & 0xff;
 863                pr_debug("... APIC EFEAT: %08x\n", v);
 864                v = apic_read(APIC_ECTRL);
 865                pr_debug("... APIC ECTRL: %08x\n", v);
 866                for (i = 0; i < maxlvt; i++) {
 867                        v = apic_read(APIC_EILVTn(i));
 868                        pr_debug("... APIC EILVT%d: %08x\n", i, v);
 869                }
 870        }
 871        pr_cont("\n");
 872}
 873
 874static void __init print_local_APICs(int maxcpu)
 875{
 876        int cpu;
 877
 878        if (!maxcpu)
 879                return;
 880
 881        preempt_disable();
 882        for_each_online_cpu(cpu) {
 883                if (cpu >= maxcpu)
 884                        break;
 885                smp_call_function_single(cpu, print_local_APIC, NULL, 1);
 886        }
 887        preempt_enable();
 888}
 889
 890static void __init print_PIC(void)
 891{
 892        unsigned int v;
 893        unsigned long flags;
 894
 895        if (!nr_legacy_irqs())
 896                return;
 897
 898        pr_debug("\nprinting PIC contents\n");
 899
 900        raw_spin_lock_irqsave(&i8259A_lock, flags);
 901
 902        v = inb(0xa1) << 8 | inb(0x21);
 903        pr_debug("... PIC  IMR: %04x\n", v);
 904
 905        v = inb(0xa0) << 8 | inb(0x20);
 906        pr_debug("... PIC  IRR: %04x\n", v);
 907
 908        outb(0x0b, 0xa0);
 909        outb(0x0b, 0x20);
 910        v = inb(0xa0) << 8 | inb(0x20);
 911        outb(0x0a, 0xa0);
 912        outb(0x0a, 0x20);
 913
 914        raw_spin_unlock_irqrestore(&i8259A_lock, flags);
 915
 916        pr_debug("... PIC  ISR: %04x\n", v);
 917
 918        v = inb(0x4d1) << 8 | inb(0x4d0);
 919        pr_debug("... PIC ELCR: %04x\n", v);
 920}
 921
 922static int show_lapic __initdata = 1;
 923static __init int setup_show_lapic(char *arg)
 924{
 925        int num = -1;
 926
 927        if (strcmp(arg, "all") == 0) {
 928                show_lapic = CONFIG_NR_CPUS;
 929        } else {
 930                get_option(&arg, &num);
 931                if (num >= 0)
 932                        show_lapic = num;
 933        }
 934
 935        return 1;
 936}
 937__setup("show_lapic=", setup_show_lapic);
 938
 939static int __init print_ICs(void)
 940{
 941        if (apic_verbosity == APIC_QUIET)
 942                return 0;
 943
 944        print_PIC();
 945
 946        /* don't print out if apic is not there */
 947        if (!boot_cpu_has(X86_FEATURE_APIC) && !apic_from_smp_config())
 948                return 0;
 949
 950        print_local_APICs(show_lapic);
 951        print_IO_APICs();
 952
 953        return 0;
 954}
 955
 956late_initcall(print_ICs);
 957