kernel/irq/irqdesc.c

   1 /*
   2  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
   3  * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
   4  *
   5  * This file contains the interrupt descriptor management code
   6  *
   7  * Detailed information is available in Documentation/DocBook/genericirq
   8  *
   9  */
  10 #include <linux/irq.h>
  11 #include <linux/slab.h>
  12 #include <linux/export.h>
  13 #include <linux/interrupt.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/radix-tree.h>
  16 #include <linux/bitmap.h>
  17 #include <linux/irqdomain.h>
  18
  19 #include "internals.h"
  20
  21 /*
  22  * lockdep: we want to handle all irq_desc locks as a single lock-class:
  23  */
  24 static struct lock_class_key irq_desc_lock_class;
  25
  26 #if defined(CONFIG_SMP)
  27 static void __init init_irq_default_affinity(void)
  28 {
  29         alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
  30         cpumask_setall(irq_default_affinity);
  31 }
  32 #else
  33 static void __init init_irq_default_affinity(void)
  34 {
  35 }
  36 #endif
  37
  38 #ifdef CONFIG_SMP
  39 static int alloc_masks(struct irq_desc *desc, gfp_t gfp, int node)
  40 {
  41         if (!zalloc_cpumask_var_node(&desc->irq_common_data.affinity,
  42                                      gfp, node))
  43                 return -ENOMEM;
  44
  45 #ifdef CONFIG_GENERIC_PENDING_IRQ
  46         if (!zalloc_cpumask_var_node(&desc->pending_mask, gfp, node)) {
  47                 free_cpumask_var(desc->irq_common_data.affinity);
  48                 return -ENOMEM;
  49         }
  50 #endif
  51         return 0;
  52 }
  53
  54 static void desc_smp_init(struct irq_desc *desc, int node)
  55 {
  56         cpumask_copy(desc->irq_common_data.affinity, irq_default_affinity);
  57 #ifdef CONFIG_GENERIC_PENDING_IRQ
  58         cpumask_clear(desc->pending_mask);
  59 #endif
  60 #ifdef CONFIG_NUMA
  61         desc->irq_common_data.node = node;
  62 #endif
  63 }
  64
  65 #else
  66 static inline int
  67 alloc_masks(struct irq_desc *desc, gfp_t gfp, int node) { return 0; }
  68 static inline void desc_smp_init(struct irq_desc *desc, int node) { }
  69 #endif
  70
  71 static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node,
  72                 struct module *owner)
  73 {
  74         int cpu;
  75
  76         desc->irq_common_data.handler_data = NULL;
  77         desc->irq_common_data.msi_desc = NULL;
  78
  79         desc->irq_data.common = &desc->irq_common_data;
  80         desc->irq_data.irq = irq;
  81         desc->irq_data.chip = &no_irq_chip;
  82         desc->irq_data.chip_data = NULL;
  83         irq_settings_clr_and_set(desc, ~0, _IRQ_DEFAULT_INIT_FLAGS);
  84         irqd_set(&desc->irq_data, IRQD_IRQ_DISABLED);
  85         desc->handle_irq = handle_bad_irq;
  86         desc->depth = 1;
  87         desc->irq_count = 0;
  88         desc->irqs_unhandled = 0;
  89         desc->name = NULL;
  90         desc->owner = owner;
  91         for_each_possible_cpu(cpu)
  92                 *per_cpu_ptr(desc->kstat_irqs, cpu) = 0;
  93         desc_smp_init(desc, node);
  94 }
  95
  96 int nr_irqs = NR_IRQS;
  97 EXPORT_SYMBOL_GPL(nr_irqs);
  98
  99 static DEFINE_MUTEX(sparse_irq_lock);
 100 static DECLARE_BITMAP(allocated_irqs, IRQ_BITMAP_BITS);
 101
 102 #ifdef CONFIG_SPARSE_IRQ
 103
 104 static RADIX_TREE(irq_desc_tree, GFP_KERNEL);
 105
 106 static void irq_insert_desc(unsigned int irq, struct irq_desc *desc)
 107 {
 108         radix_tree_insert(&irq_desc_tree, irq, desc);
 109 }
 110
 111 struct irq_desc *irq_to_desc(unsigned int irq)
 112 {
 113         return radix_tree_lookup(&irq_desc_tree, irq);
 114 }
 115 EXPORT_SYMBOL(irq_to_desc);
 116
 117 static void delete_irq_desc(unsigned int irq)
 118 {
 119         radix_tree_delete(&irq_desc_tree, irq);
 120 }
 121
 122 #ifdef CONFIG_SMP
 123 static void free_masks(struct irq_desc *desc)
 124 {
 125 #ifdef CONFIG_GENERIC_PENDING_IRQ
 126         free_cpumask_var(desc->pending_mask);
 127 #endif
 128         free_cpumask_var(desc->irq_common_data.affinity);
 129 }
 130 #else
 131 static inline void free_masks(struct irq_desc *desc) { }
 132 #endif
 133
 134 void irq_lock_sparse(void)
 135 {
 136         mutex_lock(&sparse_irq_lock);
 137 }
 138
 139 void irq_unlock_sparse(void)
 140 {
 141         mutex_unlock(&sparse_irq_lock);
 142 }
 143
 144 static struct irq_desc *alloc_desc(int irq, int node, struct module *owner)
 145 {
 146         struct irq_desc *desc;
 147         gfp_t gfp = GFP_KERNEL;
 148
 149         desc = kzalloc_node(sizeof(*desc), gfp, node);
 150         if (!desc)
 151                 return NULL;
 152         /* allocate based on nr_cpu_ids */
 153         desc->kstat_irqs = alloc_percpu(unsigned int);
 154         if (!desc->kstat_irqs)
 155                 goto err_desc;
 156
 157         if (alloc_masks(desc, gfp, node))
 158                 goto err_kstat;
 159
 160         raw_spin_lock_init(&desc->lock);
 161         lockdep_set_class(&desc->lock, &irq_desc_lock_class);
 162         init_rcu_head(&desc->rcu);
 163
 164         desc_set_defaults(irq, desc, node, owner);
 165
 166         return desc;
 167
 168 err_kstat:
 169         free_percpu(desc->kstat_irqs);
 170 err_desc:
 171         kfree(desc);
 172         return NULL;
 173 }
 174
 175 static void delayed_free_desc(struct rcu_head *rhp)
 176 {
 177         struct irq_desc *desc = container_of(rhp, struct irq_desc, rcu);
 178
 179         free_masks(desc);
 180         free_percpu(desc->kstat_irqs);
 181         kfree(desc);
 182 }
 183
 184 static void free_desc(unsigned int irq)
 185 {
 186         struct irq_desc *desc = irq_to_desc(irq);
 187
 188         unregister_irq_proc(irq, desc);
 189
 190         /*
 191          * sparse_irq_lock protects also show_interrupts() and
 192          * kstat_irq_usr(). Once we deleted the descriptor from the
 193          * sparse tree we can free it. Access in proc will fail to
 194          * lookup the descriptor.
 195          */
 196         mutex_lock(&sparse_irq_lock);
 197         delete_irq_desc(irq);
 198         mutex_unlock(&sparse_irq_lock);
 199
 200         /*
 201          * We free the descriptor, masks and stat fields via RCU. That
 202          * allows demultiplex interrupts to do rcu based management of
 203          * the child interrupts.
 204          */
 205         call_rcu(&desc->rcu, delayed_free_desc);
 206 }
 207
 208 static int alloc_descs(unsigned int start, unsigned int cnt, int node,
 209                        struct module *owner)
 210 {
 211         struct irq_desc *desc;
 212         int i;
 213
 214         for (i = 0; i < cnt; i++) {
 215                 desc = alloc_desc(start + i, node, owner);
 216                 if (!desc)
 217                         goto err;
 218                 mutex_lock(&sparse_irq_lock);
 219                 irq_insert_desc(start + i, desc);
 220                 mutex_unlock(&sparse_irq_lock);
 221         }
 222         return start;
 223
 224 err:
 225         for (i--; i >= 0; i--)
 226                 free_desc(start + i);
 227
 228         mutex_lock(&sparse_irq_lock);
 229         bitmap_clear(allocated_irqs, start, cnt);
 230         mutex_unlock(&sparse_irq_lock);
 231         return -ENOMEM;
 232 }
 233
 234 static int irq_expand_nr_irqs(unsigned int nr)
 235 {
 236         if (nr > IRQ_BITMAP_BITS)
 237                 return -ENOMEM;
 238         nr_irqs = nr;
 239         return 0;
 240 }
 241
 242 int __init early_irq_init(void)
 243 {
 244         int i, initcnt, node = first_online_node;
 245         struct irq_desc *desc;
 246
 247         init_irq_default_affinity();
 248
 249         /* Let arch update nr_irqs and return the nr of preallocated irqs */
 250         initcnt = arch_probe_nr_irqs();
 251         printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d %d\n", NR_IRQS, nr_irqs, initcnt);
 252
 253         if (WARN_ON(nr_irqs > IRQ_BITMAP_BITS))
 254                 nr_irqs = IRQ_BITMAP_BITS;
 255
 256         if (WARN_ON(initcnt > IRQ_BITMAP_BITS))
 257                 initcnt = IRQ_BITMAP_BITS;
 258
 259         if (initcnt > nr_irqs)
 260                 nr_irqs = initcnt;
 261
 262         for (i = 0; i < initcnt; i++) {
 263                 desc = alloc_desc(i, node, NULL);
 264                 set_bit(i, allocated_irqs);
 265                 irq_insert_desc(i, desc);
 266         }
 267         return arch_early_irq_init();
 268 }
 269
 270 #else /* !CONFIG_SPARSE_IRQ */
 271
 272 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
 273         [0 ... NR_IRQS-1] = {
 274                 .handle_irq     = handle_bad_irq,
 275                 .depth          = 1,
 276                 .lock           = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
 277         }
 278 };
 279
 280 int __init early_irq_init(void)
 281 {
 282         int count, i, node = first_online_node;
 283         struct irq_desc *desc;
 284
 285         init_irq_default_affinity();
 286
 287         printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
 288
 289         desc = irq_desc;
 290         count = ARRAY_SIZE(irq_desc);
 291
 292         for (i = 0; i < count; i++) {
 293                 desc[i].kstat_irqs = alloc_percpu(unsigned int);
 294                 alloc_masks(&desc[i], GFP_KERNEL, node);
 295                 raw_spin_lock_init(&desc[i].lock);
 296                 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
 297                 desc_set_defaults(i, &desc[i], node, NULL);
 298         }
 299         return arch_early_irq_init();
 300 }
 301
 302 struct irq_desc *irq_to_desc(unsigned int irq)
 303 {
 304         return (irq < NR_IRQS) ? irq_desc + irq : NULL;
 305 }
 306 EXPORT_SYMBOL(irq_to_desc);
 307
 308 static void free_desc(unsigned int irq)
 309 {
 310         struct irq_desc *desc = irq_to_desc(irq);
 311         unsigned long flags;
 312
 313         raw_spin_lock_irqsave(&desc->lock, flags);
 314         desc_set_defaults(irq, desc, irq_desc_get_node(desc), NULL);
 315         raw_spin_unlock_irqrestore(&desc->lock, flags);
 316 }
 317
 318 static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
 319                               struct module *owner)
 320 {
 321         u32 i;
 322
 323         for (i = 0; i < cnt; i++) {
 324                 struct irq_desc *desc = irq_to_desc(start + i);
 325
 326                 desc->owner = owner;
 327         }
 328         return start;
 329 }
 330
 331 static int irq_expand_nr_irqs(unsigned int nr)
 332 {
 333         return -ENOMEM;
 334 }
 335
 336 void irq_mark_irq(unsigned int irq)
 337 {
 338         mutex_lock(&sparse_irq_lock);
 339         bitmap_set(allocated_irqs, irq, 1);
 340         mutex_unlock(&sparse_irq_lock);
 341 }
 342
 343 #ifdef CONFIG_GENERIC_IRQ_LEGACY
 344 void irq_init_desc(unsigned int irq)
 345 {
 346         free_desc(irq);
 347 }
 348 #endif
 349
 350 #endif /* !CONFIG_SPARSE_IRQ */
 351
 352 /**
 353  * generic_handle_irq - Invoke the handler for a particular irq
 354  * @irq:        The irq number to handle
 355  *
 356  */
 357 int generic_handle_irq(unsigned int irq)
 358 {
 359         struct irq_desc *desc = irq_to_desc(irq);
 360
 361         if (!desc)
 362                 return -EINVAL;
 363         generic_handle_irq_desc(desc);
 364         return 0;
 365 }
 366 EXPORT_SYMBOL_GPL(generic_handle_irq);
 367
 368 #ifdef CONFIG_HANDLE_DOMAIN_IRQ
 369 /**
 370  * __handle_domain_irq - Invoke the handler for a HW irq belonging to a domain
 371  * @domain:     The domain where to perform the lookup
 372  * @hwirq:      The HW irq number to convert to a logical one
 373  * @lookup:     Whether to perform the domain lookup or not
 374  * @regs:       Register file coming from the low-level handling code
 375  *
 376  * Returns:     0 on success, or -EINVAL if conversion has failed
 377  */
 378 int __handle_domain_irq(struct irq_domain *domain, unsigned int hwirq,
 379                         bool lookup, struct pt_regs *regs)
 380 {
 381         struct pt_regs *old_regs = set_irq_regs(regs);
 382         unsigned int irq = hwirq;
 383         int ret = 0;
 384
 385         irq_enter();
 386
 387 #ifdef CONFIG_IRQ_DOMAIN
 388         if (lookup)
 389                 irq = irq_find_mapping(domain, hwirq);
 390 #endif
 391
 392         /*
 393          * Some hardware gives randomly wrong interrupts.  Rather
 394          * than crashing, do something sensible.
 395          */
 396         if (unlikely(!irq || irq >= nr_irqs)) {
 397                 ack_bad_irq(irq);
 398                 ret = -EINVAL;
 399         } else {
 400                 generic_handle_irq(irq);
 401         }
 402
 403         irq_exit();
 404         set_irq_regs(old_regs);
 405         return ret;
 406 }
 407 #endif
 408
 409 /* Dynamic interrupt handling */
 410
 411 /**
 412  * irq_free_descs - free irq descriptors
 413  * @from:       Start of descriptor range
 414  * @cnt:        Number of consecutive irqs to free
 415  */
 416 void irq_free_descs(unsigned int from, unsigned int cnt)
 417 {
 418         int i;
 419
 420         if (from >= nr_irqs || (from + cnt) > nr_irqs)
 421                 return;
 422
 423         for (i = 0; i < cnt; i++)
 424                 free_desc(from + i);
 425
 426         mutex_lock(&sparse_irq_lock);
 427         bitmap_clear(allocated_irqs, from, cnt);
 428         mutex_unlock(&sparse_irq_lock);
 429 }
 430 EXPORT_SYMBOL_GPL(irq_free_descs);
 431
 432 /**
 433  * irq_alloc_descs - allocate and initialize a range of irq descriptors
 434  * @irq:        Allocate for specific irq number if irq >= 0
 435  * @from:       Start the search from this irq number
 436  * @cnt:        Number of consecutive irqs to allocate.
 437  * @node:       Preferred node on which the irq descriptor should be allocated
 438  * @owner:      Owning module (can be NULL)
 439  *
 440  * Returns the first irq number or error code
 441  */
 442 int __ref
 443 __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
 444                   struct module *owner)
 445 {
 446         int start, ret;
 447
 448         if (!cnt)
 449                 return -EINVAL;
 450
 451         if (irq >= 0) {
 452                 if (from > irq)
 453                         return -EINVAL;
 454                 from = irq;
 455         } else {
 456                 /*
 457                  * For interrupts which are freely allocated the
 458                  * architecture can force a lower bound to the @from
 459                  * argument. x86 uses this to exclude the GSI space.
 460                  */
 461                 from = arch_dynirq_lower_bound(from);
 462         }
 463
 464         mutex_lock(&sparse_irq_lock);
 465
 466         start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
 467                                            from, cnt, 0);
 468         ret = -EEXIST;
 469         if (irq >=0 && start != irq)
 470                 goto err;
 471
 472         if (start + cnt > nr_irqs) {
 473                 ret = irq_expand_nr_irqs(start + cnt);
 474                 if (ret)
 475                         goto err;
 476         }
 477
 478         bitmap_set(allocated_irqs, start, cnt);
 479         mutex_unlock(&sparse_irq_lock);
 480         return alloc_descs(start, cnt, node, owner);
 481
 482 err:
 483         mutex_unlock(&sparse_irq_lock);
 484         return ret;
 485 }
 486 EXPORT_SYMBOL_GPL(__irq_alloc_descs);
 487
 488 #ifdef CONFIG_GENERIC_IRQ_LEGACY_ALLOC_HWIRQ
 489 /**
 490  * irq_alloc_hwirqs - Allocate an irq descriptor and initialize the hardware
 491  * @cnt:        number of interrupts to allocate
 492  * @node:       node on which to allocate
 493  *
 494  * Returns an interrupt number > 0 or 0, if the allocation fails.
 495  */
 496 unsigned int irq_alloc_hwirqs(int cnt, int node)
 497 {
 498         int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL);
 499
 500         if (irq < 0)
 501                 return 0;
 502
 503         for (i = irq; cnt > 0; i++, cnt--) {
 504                 if (arch_setup_hwirq(i, node))
 505                         goto err;
 506                 irq_clear_status_flags(i, _IRQ_NOREQUEST);
 507         }
 508         return irq;
 509
 510 err:
 511         for (i--; i >= irq; i--) {
 512                 irq_set_status_flags(i, _IRQ_NOREQUEST | _IRQ_NOPROBE);
 513                 arch_teardown_hwirq(i);
 514         }
 515         irq_free_descs(irq, cnt);
 516         return 0;
 517 }
 518 EXPORT_SYMBOL_GPL(irq_alloc_hwirqs);
 519
 520 /**
 521  * irq_free_hwirqs - Free irq descriptor and cleanup the hardware
 522  * @from:       Free from irq number
 523  * @cnt:        number of interrupts to free
 524  *
 525  */
 526 void irq_free_hwirqs(unsigned int from, int cnt)
 527 {
 528         int i, j;
 529
 530         for (i = from, j = cnt; j > 0; i++, j--) {
 531                 irq_set_status_flags(i, _IRQ_NOREQUEST | _IRQ_NOPROBE);
 532                 arch_teardown_hwirq(i);
 533         }
 534         irq_free_descs(from, cnt);
 535 }
 536 EXPORT_SYMBOL_GPL(irq_free_hwirqs);
 537 #endif
 538
 539 /**
 540  * irq_get_next_irq - get next allocated irq number
 541  * @offset:     where to start the search
 542  *
 543  * Returns next irq number after offset or nr_irqs if none is found.
 544  */
 545 unsigned int irq_get_next_irq(unsigned int offset)
 546 {
 547         return find_next_bit(allocated_irqs, nr_irqs, offset);
 548 }
 549
 550 struct irq_desc *
 551 __irq_get_desc_lock(unsigned int irq, unsigned long *flags, bool bus,
 552                     unsigned int check)
 553 {
 554         struct irq_desc *desc = irq_to_desc(irq);
 555
 556         if (desc) {
 557                 if (check & _IRQ_DESC_CHECK) {
 558                         if ((check & _IRQ_DESC_PERCPU) &&
 559                             !irq_settings_is_per_cpu_devid(desc))
 560                                 return NULL;
 561
 562                         if (!(check & _IRQ_DESC_PERCPU) &&
 563                             irq_settings_is_per_cpu_devid(desc))
 564                                 return NULL;
 565                 }
 566
 567                 if (bus)
 568                         chip_bus_lock(desc);
 569                 raw_spin_lock_irqsave(&desc->lock, *flags);
 570         }
 571         return desc;
 572 }
 573
 574 void __irq_put_desc_unlock(struct irq_desc *desc, unsigned long flags, bool bus)
 575 {
 576         raw_spin_unlock_irqrestore(&desc->lock, flags);
 577         if (bus)
 578                 chip_bus_sync_unlock(desc);
 579 }
 580
 581 int irq_set_percpu_devid(unsigned int irq)
 582 {
 583         struct irq_desc *desc = irq_to_desc(irq);
 584
 585         if (!desc)
 586                 return -EINVAL;
 587
 588         if (desc->percpu_enabled)
 589                 return -EINVAL;
 590
 591         desc->percpu_enabled = kzalloc(sizeof(*desc->percpu_enabled), GFP_KERNEL);
 592
 593         if (!desc->percpu_enabled)
 594                 return -ENOMEM;
 595
 596         irq_set_percpu_devid_flags(irq);
 597         return 0;
 598 }
 599
 600 void kstat_incr_irq_this_cpu(unsigned int irq)
 601 {
 602         kstat_incr_irqs_this_cpu(irq_to_desc(irq));
 603 }
 604
 605 /**
 606  * kstat_irqs_cpu - Get the statistics for an interrupt on a cpu
 607  * @irq:        The interrupt number
 608  * @cpu:        The cpu number
 609  *
 610  * Returns the sum of interrupt counts on @cpu since boot for
 611  * @irq. The caller must ensure that the interrupt is not removed
 612  * concurrently.
 613  */
 614 unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
 615 {
 616         struct irq_desc *desc = irq_to_desc(irq);
 617
 618         return desc && desc->kstat_irqs ?
 619                         *per_cpu_ptr(desc->kstat_irqs, cpu) : 0;
 620 }
 621
 622 /**
 623  * kstat_irqs - Get the statistics for an interrupt
 624  * @irq:        The interrupt number
 625  *
 626  * Returns the sum of interrupt counts on all cpus since boot for
 627  * @irq. The caller must ensure that the interrupt is not removed
 628  * concurrently.
 629  */
 630 unsigned int kstat_irqs(unsigned int irq)
 631 {
 632         struct irq_desc *desc = irq_to_desc(irq);
 633         int cpu;
 634         unsigned int sum = 0;
 635
 636         if (!desc || !desc->kstat_irqs)
 637                 return 0;
 638         for_each_possible_cpu(cpu)
 639                 sum += *per_cpu_ptr(desc->kstat_irqs, cpu);
 640         return sum;
 641 }
 642
 643 /**
 644  * kstat_irqs_usr - Get the statistics for an interrupt
 645  * @irq:        The interrupt number
 646  *
 647  * Returns the sum of interrupt counts on all cpus since boot for
 648  * @irq. Contrary to kstat_irqs() this can be called from any
 649  * preemptible context. It's protected against concurrent removal of
 650  * an interrupt descriptor when sparse irqs are enabled.
 651  */
 652 unsigned int kstat_irqs_usr(unsigned int irq)
 653 {
 654         unsigned int sum;
 655
 656         irq_lock_sparse();
 657         sum = kstat_irqs(irq);
 658         irq_unlock_sparse();
 659         return sum;
 660 }