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 int __init irq_affinity_setup(char *str)
  28 {
  29         zalloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
  30         cpulist_parse(str, irq_default_affinity);
  31         /*
  32          * Set at least the boot cpu. We don't want to end up with
  33          * bugreports caused by random comandline masks
  34          */
  35         cpumask_set_cpu(smp_processor_id(), irq_default_affinity);
  36         return 1;
  37 }
  38 __setup("irqaffinity=", irq_affinity_setup);
  39
  40 static void __init init_irq_default_affinity(void)
  41 {
  42 #ifdef CONFIG_CPUMASK_OFFSTACK
  43         if (!irq_default_affinity)
  44                 zalloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
  45 #endif
  46         if (cpumask_empty(irq_default_affinity))
  47                 cpumask_setall(irq_default_affinity);
  48 }
  49 #else
  50 static void __init init_irq_default_affinity(void)
  51 {
  52 }
  53 #endif
  54
  55 #ifdef CONFIG_SMP
  56 static int alloc_masks(struct irq_desc *desc, gfp_t gfp, int node)
  57 {
  58         if (!zalloc_cpumask_var_node(&desc->irq_common_data.affinity,
  59                                      gfp, node))
  60                 return -ENOMEM;
  61
  62 #ifdef CONFIG_GENERIC_PENDING_IRQ
  63         if (!zalloc_cpumask_var_node(&desc->pending_mask, gfp, node)) {
  64                 free_cpumask_var(desc->irq_common_data.affinity);
  65                 return -ENOMEM;
  66         }
  67 #endif
  68         return 0;
  69 }
  70
  71 static void desc_smp_init(struct irq_desc *desc, int node)
  72 {
  73         cpumask_copy(desc->irq_common_data.affinity, irq_default_affinity);
  74 #ifdef CONFIG_GENERIC_PENDING_IRQ
  75         cpumask_clear(desc->pending_mask);
  76 #endif
  77 #ifdef CONFIG_NUMA
  78         desc->irq_common_data.node = node;
  79 #endif
  80 }
  81
  82 #else
  83 static inline int
  84 alloc_masks(struct irq_desc *desc, gfp_t gfp, int node) { return 0; }
  85 static inline void desc_smp_init(struct irq_desc *desc, int node) { }
  86 #endif
  87
  88 static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node,
  89                 struct module *owner)
  90 {
  91         int cpu;
  92
  93         desc->irq_common_data.handler_data = NULL;
  94         desc->irq_common_data.msi_desc = NULL;
  95
  96         desc->irq_data.common = &desc->irq_common_data;
  97         desc->irq_data.irq = irq;
  98         desc->irq_data.chip = &no_irq_chip;
  99         desc->irq_data.chip_data = NULL;
 100         irq_settings_clr_and_set(desc, ~0, _IRQ_DEFAULT_INIT_FLAGS);
 101         irqd_set(&desc->irq_data, IRQD_IRQ_DISABLED);
 102         desc->handle_irq = handle_bad_irq;
 103         desc->depth = 1;
 104         desc->irq_count = 0;
 105         desc->irqs_unhandled = 0;
 106         desc->name = NULL;
 107         desc->owner = owner;
 108         for_each_possible_cpu(cpu)
 109                 *per_cpu_ptr(desc->kstat_irqs, cpu) = 0;
 110         desc_smp_init(desc, node);
 111 }
 112
 113 int nr_irqs = NR_IRQS;
 114 EXPORT_SYMBOL_GPL(nr_irqs);
 115
 116 static DEFINE_MUTEX(sparse_irq_lock);
 117 static DECLARE_BITMAP(allocated_irqs, IRQ_BITMAP_BITS);
 118
 119 #ifdef CONFIG_SPARSE_IRQ
 120
 121 static RADIX_TREE(irq_desc_tree, GFP_KERNEL);
 122
 123 static void irq_insert_desc(unsigned int irq, struct irq_desc *desc)
 124 {
 125         radix_tree_insert(&irq_desc_tree, irq, desc);
 126 }
 127
 128 struct irq_desc *irq_to_desc(unsigned int irq)
 129 {
 130         return radix_tree_lookup(&irq_desc_tree, irq);
 131 }
 132 EXPORT_SYMBOL(irq_to_desc);
 133
 134 static void delete_irq_desc(unsigned int irq)
 135 {
 136         radix_tree_delete(&irq_desc_tree, irq);
 137 }
 138
 139 #ifdef CONFIG_SMP
 140 static void free_masks(struct irq_desc *desc)
 141 {
 142 #ifdef CONFIG_GENERIC_PENDING_IRQ
 143         free_cpumask_var(desc->pending_mask);
 144 #endif
 145         free_cpumask_var(desc->irq_common_data.affinity);
 146 }
 147 #else
 148 static inline void free_masks(struct irq_desc *desc) { }
 149 #endif
 150
 151 void irq_lock_sparse(void)
 152 {
 153         mutex_lock(&sparse_irq_lock);
 154 }
 155
 156 void irq_unlock_sparse(void)
 157 {
 158         mutex_unlock(&sparse_irq_lock);
 159 }
 160
 161 static struct irq_desc *alloc_desc(int irq, int node, struct module *owner)
 162 {
 163         struct irq_desc *desc;
 164         gfp_t gfp = GFP_KERNEL;
 165
 166         desc = kzalloc_node(sizeof(*desc), gfp, node);
 167         if (!desc)
 168                 return NULL;
 169         /* allocate based on nr_cpu_ids */
 170         desc->kstat_irqs = alloc_percpu(unsigned int);
 171         if (!desc->kstat_irqs)
 172                 goto err_desc;
 173
 174         if (alloc_masks(desc, gfp, node))
 175                 goto err_kstat;
 176
 177         raw_spin_lock_init(&desc->lock);
 178         lockdep_set_class(&desc->lock, &irq_desc_lock_class);
 179         init_rcu_head(&desc->rcu);
 180
 181         desc_set_defaults(irq, desc, node, owner);
 182
 183         return desc;
 184
 185 err_kstat:
 186         free_percpu(desc->kstat_irqs);
 187 err_desc:
 188         kfree(desc);
 189         return NULL;
 190 }
 191
 192 static void delayed_free_desc(struct rcu_head *rhp)
 193 {
 194         struct irq_desc *desc = container_of(rhp, struct irq_desc, rcu);
 195
 196         free_masks(desc);
 197         free_percpu(desc->kstat_irqs);
 198         kfree(desc);
 199 }
 200
 201 static void free_desc(unsigned int irq)
 202 {
 203         struct irq_desc *desc = irq_to_desc(irq);
 204
 205         unregister_irq_proc(irq, desc);
 206
 207         /*
 208          * sparse_irq_lock protects also show_interrupts() and
 209          * kstat_irq_usr(). Once we deleted the descriptor from the
 210          * sparse tree we can free it. Access in proc will fail to
 211          * lookup the descriptor.
 212          */
 213         mutex_lock(&sparse_irq_lock);
 214         delete_irq_desc(irq);
 215         mutex_unlock(&sparse_irq_lock);
 216
 217         /*
 218          * We free the descriptor, masks and stat fields via RCU. That
 219          * allows demultiplex interrupts to do rcu based management of
 220          * the child interrupts.
 221          */
 222         call_rcu(&desc->rcu, delayed_free_desc);
 223 }
 224
 225 static int alloc_descs(unsigned int start, unsigned int cnt, int node,
 226                        struct module *owner)
 227 {
 228         struct irq_desc *desc;
 229         int i;
 230
 231         for (i = 0; i < cnt; i++) {
 232                 desc = alloc_desc(start + i, node, owner);
 233                 if (!desc)
 234                         goto err;
 235                 mutex_lock(&sparse_irq_lock);
 236                 irq_insert_desc(start + i, desc);
 237                 mutex_unlock(&sparse_irq_lock);
 238         }
 239         return start;
 240
 241 err:
 242         for (i--; i >= 0; i--)
 243                 free_desc(start + i);
 244
 245         mutex_lock(&sparse_irq_lock);
 246         bitmap_clear(allocated_irqs, start, cnt);
 247         mutex_unlock(&sparse_irq_lock);
 248         return -ENOMEM;
 249 }
 250
 251 static int irq_expand_nr_irqs(unsigned int nr)
 252 {
 253         if (nr > IRQ_BITMAP_BITS)
 254                 return -ENOMEM;
 255         nr_irqs = nr;
 256         return 0;
 257 }
 258
 259 int __init early_irq_init(void)
 260 {
 261         int i, initcnt, node = first_online_node;
 262         struct irq_desc *desc;
 263
 264         init_irq_default_affinity();
 265
 266         /* Let arch update nr_irqs and return the nr of preallocated irqs */
 267         initcnt = arch_probe_nr_irqs();
 268         printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d %d\n", NR_IRQS, nr_irqs, initcnt);
 269
 270         if (WARN_ON(nr_irqs > IRQ_BITMAP_BITS))
 271                 nr_irqs = IRQ_BITMAP_BITS;
 272
 273         if (WARN_ON(initcnt > IRQ_BITMAP_BITS))
 274                 initcnt = IRQ_BITMAP_BITS;
 275
 276         if (initcnt > nr_irqs)
 277                 nr_irqs = initcnt;
 278
 279         for (i = 0; i < initcnt; i++) {
 280                 desc = alloc_desc(i, node, NULL);
 281                 set_bit(i, allocated_irqs);
 282                 irq_insert_desc(i, desc);
 283         }
 284         return arch_early_irq_init();
 285 }
 286
 287 #else /* !CONFIG_SPARSE_IRQ */
 288
 289 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
 290         [0 ... NR_IRQS-1] = {
 291                 .handle_irq     = handle_bad_irq,
 292                 .depth          = 1,
 293                 .lock           = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
 294         }
 295 };
 296
 297 int __init early_irq_init(void)
 298 {
 299         int count, i, node = first_online_node;
 300         struct irq_desc *desc;
 301
 302         init_irq_default_affinity();
 303
 304         printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
 305
 306         desc = irq_desc;
 307         count = ARRAY_SIZE(irq_desc);
 308
 309         for (i = 0; i < count; i++) {
 310                 desc[i].kstat_irqs = alloc_percpu(unsigned int);
 311                 alloc_masks(&desc[i], GFP_KERNEL, node);
 312                 raw_spin_lock_init(&desc[i].lock);
 313                 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
 314                 desc_set_defaults(i, &desc[i], node, NULL);
 315         }
 316         return arch_early_irq_init();
 317 }
 318
 319 struct irq_desc *irq_to_desc(unsigned int irq)
 320 {
 321         return (irq < NR_IRQS) ? irq_desc + irq : NULL;
 322 }
 323 EXPORT_SYMBOL(irq_to_desc);
 324
 325 static void free_desc(unsigned int irq)
 326 {
 327         struct irq_desc *desc = irq_to_desc(irq);
 328         unsigned long flags;
 329
 330         raw_spin_lock_irqsave(&desc->lock, flags);
 331         desc_set_defaults(irq, desc, irq_desc_get_node(desc), NULL);
 332         raw_spin_unlock_irqrestore(&desc->lock, flags);
 333 }
 334
 335 static inline int alloc_descs(unsigned int start, unsigned int cnt, int node,
 336                               struct module *owner)
 337 {
 338         u32 i;
 339
 340         for (i = 0; i < cnt; i++) {
 341                 struct irq_desc *desc = irq_to_desc(start + i);
 342
 343                 desc->owner = owner;
 344         }
 345         return start;
 346 }
 347
 348 static int irq_expand_nr_irqs(unsigned int nr)
 349 {
 350         return -ENOMEM;
 351 }
 352
 353 void irq_mark_irq(unsigned int irq)
 354 {
 355         mutex_lock(&sparse_irq_lock);
 356         bitmap_set(allocated_irqs, irq, 1);
 357         mutex_unlock(&sparse_irq_lock);
 358 }
 359
 360 #ifdef CONFIG_GENERIC_IRQ_LEGACY
 361 void irq_init_desc(unsigned int irq)
 362 {
 363         free_desc(irq);
 364 }
 365 #endif
 366
 367 #endif /* !CONFIG_SPARSE_IRQ */
 368
 369 /**
 370  * generic_handle_irq - Invoke the handler for a particular irq
 371  * @irq:        The irq number to handle
 372  *
 373  */
 374 int generic_handle_irq(unsigned int irq)
 375 {
 376         struct irq_desc *desc = irq_to_desc(irq);
 377
 378         if (!desc)
 379                 return -EINVAL;
 380         generic_handle_irq_desc(desc);
 381         return 0;
 382 }
 383 EXPORT_SYMBOL_GPL(generic_handle_irq);
 384
 385 #ifdef CONFIG_HANDLE_DOMAIN_IRQ
 386 /**
 387  * __handle_domain_irq - Invoke the handler for a HW irq belonging to a domain
 388  * @domain:     The domain where to perform the lookup
 389  * @hwirq:      The HW irq number to convert to a logical one
 390  * @lookup:     Whether to perform the domain lookup or not
 391  * @regs:       Register file coming from the low-level handling code
 392  *
 393  * Returns:     0 on success, or -EINVAL if conversion has failed
 394  */
 395 int __handle_domain_irq(struct irq_domain *domain, unsigned int hwirq,
 396                         bool lookup, struct pt_regs *regs)
 397 {
 398         struct pt_regs *old_regs = set_irq_regs(regs);
 399         unsigned int irq = hwirq;
 400         int ret = 0;
 401
 402         irq_enter();
 403
 404 #ifdef CONFIG_IRQ_DOMAIN
 405         if (lookup)
 406                 irq = irq_find_mapping(domain, hwirq);
 407 #endif
 408
 409         /*
 410          * Some hardware gives randomly wrong interrupts.  Rather
 411          * than crashing, do something sensible.
 412          */
 413         if (unlikely(!irq || irq >= nr_irqs)) {
 414                 ack_bad_irq(irq);
 415                 ret = -EINVAL;
 416         } else {
 417                 generic_handle_irq(irq);
 418         }
 419
 420         irq_exit();
 421         set_irq_regs(old_regs);
 422         return ret;
 423 }
 424 #endif
 425
 426 /* Dynamic interrupt handling */
 427
 428 /**
 429  * irq_free_descs - free irq descriptors
 430  * @from:       Start of descriptor range
 431  * @cnt:        Number of consecutive irqs to free
 432  */
 433 void irq_free_descs(unsigned int from, unsigned int cnt)
 434 {
 435         int i;
 436
 437         if (from >= nr_irqs || (from + cnt) > nr_irqs)
 438                 return;
 439
 440         for (i = 0; i < cnt; i++)
 441                 free_desc(from + i);
 442
 443         mutex_lock(&sparse_irq_lock);
 444         bitmap_clear(allocated_irqs, from, cnt);
 445         mutex_unlock(&sparse_irq_lock);
 446 }
 447 EXPORT_SYMBOL_GPL(irq_free_descs);
 448
 449 /**
 450  * irq_alloc_descs - allocate and initialize a range of irq descriptors
 451  * @irq:        Allocate for specific irq number if irq >= 0
 452  * @from:       Start the search from this irq number
 453  * @cnt:        Number of consecutive irqs to allocate.
 454  * @node:       Preferred node on which the irq descriptor should be allocated
 455  * @owner:      Owning module (can be NULL)
 456  *
 457  * Returns the first irq number or error code
 458  */
 459 int __ref
 460 __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
 461                   struct module *owner)
 462 {
 463         int start, ret;
 464
 465         if (!cnt)
 466                 return -EINVAL;
 467
 468         if (irq >= 0) {
 469                 if (from > irq)
 470                         return -EINVAL;
 471                 from = irq;
 472         } else {
 473                 /*
 474                  * For interrupts which are freely allocated the
 475                  * architecture can force a lower bound to the @from
 476                  * argument. x86 uses this to exclude the GSI space.
 477                  */
 478                 from = arch_dynirq_lower_bound(from);
 479         }
 480
 481         mutex_lock(&sparse_irq_lock);
 482
 483         start = bitmap_find_next_zero_area(allocated_irqs, IRQ_BITMAP_BITS,
 484                                            from, cnt, 0);
 485         ret = -EEXIST;
 486         if (irq >=0 && start != irq)
 487                 goto err;
 488
 489         if (start + cnt > nr_irqs) {
 490                 ret = irq_expand_nr_irqs(start + cnt);
 491                 if (ret)
 492                         goto err;
 493         }
 494
 495         bitmap_set(allocated_irqs, start, cnt);
 496         mutex_unlock(&sparse_irq_lock);
 497         return alloc_descs(start, cnt, node, owner);
 498
 499 err:
 500         mutex_unlock(&sparse_irq_lock);
 501         return ret;
 502 }
 503 EXPORT_SYMBOL_GPL(__irq_alloc_descs);
 504
 505 #ifdef CONFIG_GENERIC_IRQ_LEGACY_ALLOC_HWIRQ
 506 /**
 507  * irq_alloc_hwirqs - Allocate an irq descriptor and initialize the hardware
 508  * @cnt:        number of interrupts to allocate
 509  * @node:       node on which to allocate
 510  *
 511  * Returns an interrupt number > 0 or 0, if the allocation fails.
 512  */
 513 unsigned int irq_alloc_hwirqs(int cnt, int node)
 514 {
 515         int i, irq = __irq_alloc_descs(-1, 0, cnt, node, NULL);
 516
 517         if (irq < 0)
 518                 return 0;
 519
 520         for (i = irq; cnt > 0; i++, cnt--) {
 521                 if (arch_setup_hwirq(i, node))
 522                         goto err;
 523                 irq_clear_status_flags(i, _IRQ_NOREQUEST);
 524         }
 525         return irq;
 526
 527 err:
 528         for (i--; i >= irq; i--) {
 529                 irq_set_status_flags(i, _IRQ_NOREQUEST | _IRQ_NOPROBE);
 530                 arch_teardown_hwirq(i);
 531         }
 532         irq_free_descs(irq, cnt);
 533         return 0;
 534 }
 535 EXPORT_SYMBOL_GPL(irq_alloc_hwirqs);
 536
 537 /**
 538  * irq_free_hwirqs - Free irq descriptor and cleanup the hardware
 539  * @from:       Free from irq number
 540  * @cnt:        number of interrupts to free
 541  *
 542  */
 543 void irq_free_hwirqs(unsigned int from, int cnt)
 544 {
 545         int i, j;
 546
 547         for (i = from, j = cnt; j > 0; i++, j--) {
 548                 irq_set_status_flags(i, _IRQ_NOREQUEST | _IRQ_NOPROBE);
 549                 arch_teardown_hwirq(i);
 550         }
 551         irq_free_descs(from, cnt);
 552 }
 553 EXPORT_SYMBOL_GPL(irq_free_hwirqs);
 554 #endif
 555
 556 /**
 557  * irq_get_next_irq - get next allocated irq number
 558  * @offset:     where to start the search
 559  *
 560  * Returns next irq number after offset or nr_irqs if none is found.
 561  */
 562 unsigned int irq_get_next_irq(unsigned int offset)
 563 {
 564         return find_next_bit(allocated_irqs, nr_irqs, offset);
 565 }
 566
 567 struct irq_desc *
 568 __irq_get_desc_lock(unsigned int irq, unsigned long *flags, bool bus,
 569                     unsigned int check)
 570 {
 571         struct irq_desc *desc = irq_to_desc(irq);
 572
 573         if (desc) {
 574                 if (check & _IRQ_DESC_CHECK) {
 575                         if ((check & _IRQ_DESC_PERCPU) &&
 576                             !irq_settings_is_per_cpu_devid(desc))
 577                                 return NULL;
 578
 579                         if (!(check & _IRQ_DESC_PERCPU) &&
 580                             irq_settings_is_per_cpu_devid(desc))
 581                                 return NULL;
 582                 }
 583
 584                 if (bus)
 585                         chip_bus_lock(desc);
 586                 raw_spin_lock_irqsave(&desc->lock, *flags);
 587         }
 588         return desc;
 589 }
 590
 591 void __irq_put_desc_unlock(struct irq_desc *desc, unsigned long flags, bool bus)
 592 {
 593         raw_spin_unlock_irqrestore(&desc->lock, flags);
 594         if (bus)
 595                 chip_bus_sync_unlock(desc);
 596 }
 597
 598 int irq_set_percpu_devid(unsigned int irq)
 599 {
 600         struct irq_desc *desc = irq_to_desc(irq);
 601
 602         if (!desc)
 603                 return -EINVAL;
 604
 605         if (desc->percpu_enabled)
 606                 return -EINVAL;
 607
 608         desc->percpu_enabled = kzalloc(sizeof(*desc->percpu_enabled), GFP_KERNEL);
 609
 610         if (!desc->percpu_enabled)
 611                 return -ENOMEM;
 612
 613         irq_set_percpu_devid_flags(irq);
 614         return 0;
 615 }
 616
 617 void kstat_incr_irq_this_cpu(unsigned int irq)
 618 {
 619         kstat_incr_irqs_this_cpu(irq_to_desc(irq));
 620 }
 621
 622 /**
 623  * kstat_irqs_cpu - Get the statistics for an interrupt on a cpu
 624  * @irq:        The interrupt number
 625  * @cpu:        The cpu number
 626  *
 627  * Returns the sum of interrupt counts on @cpu since boot for
 628  * @irq. The caller must ensure that the interrupt is not removed
 629  * concurrently.
 630  */
 631 unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
 632 {
 633         struct irq_desc *desc = irq_to_desc(irq);
 634
 635         return desc && desc->kstat_irqs ?
 636                         *per_cpu_ptr(desc->kstat_irqs, cpu) : 0;
 637 }
 638
 639 /**
 640  * kstat_irqs - Get the statistics for an interrupt
 641  * @irq:        The interrupt number
 642  *
 643  * Returns the sum of interrupt counts on all cpus since boot for
 644  * @irq. The caller must ensure that the interrupt is not removed
 645  * concurrently.
 646  */
 647 unsigned int kstat_irqs(unsigned int irq)
 648 {
 649         struct irq_desc *desc = irq_to_desc(irq);
 650         int cpu;
 651         unsigned int sum = 0;
 652
 653         if (!desc || !desc->kstat_irqs)
 654                 return 0;
 655         for_each_possible_cpu(cpu)
 656                 sum += *per_cpu_ptr(desc->kstat_irqs, cpu);
 657         return sum;
 658 }
 659
 660 /**
 661  * kstat_irqs_usr - Get the statistics for an interrupt
 662  * @irq:        The interrupt number
 663  *
 664  * Returns the sum of interrupt counts on all cpus since boot for
 665  * @irq. Contrary to kstat_irqs() this can be called from any
 666  * preemptible context. It's protected against concurrent removal of
 667  * an interrupt descriptor when sparse irqs are enabled.
 668  */
 669 unsigned int kstat_irqs_usr(unsigned int irq)
 670 {
 671         unsigned int sum;
 672
 673         irq_lock_sparse();
 674         sum = kstat_irqs(irq);
 675         irq_unlock_sparse();
 676         return sum;
 677 }