arch/powerpc/kernel/irq.c

   1 /*
   2  *  Derived from arch/i386/kernel/irq.c
   3  *    Copyright (C) 1992 Linus Torvalds
   4  *  Adapted from arch/i386 by Gary Thomas
   5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   6  *  Updated and modified by Cort Dougan <cort@fsmlabs.com>
   7  *    Copyright (C) 1996-2001 Cort Dougan
   8  *  Adapted for Power Macintosh by Paul Mackerras
   9  *    Copyright (C) 1996 Paul Mackerras (paulus@cs.anu.edu.au)
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public License
  13  * as published by the Free Software Foundation; either version
  14  * 2 of the License, or (at your option) any later version.
  15  *
  16  * This file contains the code used by various IRQ handling routines:
  17  * asking for different IRQ's should be done through these routines
  18  * instead of just grabbing them. Thus setups with different IRQ numbers
  19  * shouldn't result in any weird surprises, and installing new handlers
  20  * should be easier.
  21  *
  22  * The MPC8xx has an interrupt mask in the SIU.  If a bit is set, the
  23  * interrupt is _enabled_.  As expected, IRQ0 is bit 0 in the 32-bit
  24  * mask register (of which only 16 are defined), hence the weird shifting
  25  * and complement of the cached_irq_mask.  I want to be able to stuff
  26  * this right into the SIU SMASK register.
  27  * Many of the prep/chrp functions are conditional compiled on CONFIG_8xx
  28  * to reduce code space and undefined function references.
  29  */
  30
  31 #undef DEBUG
  32
  33 #include <linux/export.h>
  34 #include <linux/threads.h>
  35 #include <linux/kernel_stat.h>
  36 #include <linux/signal.h>
  37 #include <linux/sched.h>
  38 #include <linux/ptrace.h>
  39 #include <linux/ioport.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/timex.h>
  42 #include <linux/init.h>
  43 #include <linux/slab.h>
  44 #include <linux/delay.h>
  45 #include <linux/irq.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/cpumask.h>
  48 #include <linux/profile.h>
  49 #include <linux/bitops.h>
  50 #include <linux/list.h>
  51 #include <linux/radix-tree.h>
  52 #include <linux/mutex.h>
  53 #include <linux/bootmem.h>
  54 #include <linux/pci.h>
  55 #include <linux/debugfs.h>
  56 #include <linux/of.h>
  57 #include <linux/of_irq.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/system.h>
  61 #include <asm/io.h>
  62 #include <asm/pgtable.h>
  63 #include <asm/irq.h>
  64 #include <asm/cache.h>
  65 #include <asm/prom.h>
  66 #include <asm/ptrace.h>
  67 #include <asm/machdep.h>
  68 #include <asm/udbg.h>
  69 #include <asm/smp.h>
  70
  71 #ifdef CONFIG_PPC64
  72 #include <asm/paca.h>
  73 #include <asm/firmware.h>
  74 #include <asm/lv1call.h>
  75 #endif
  76 #define CREATE_TRACE_POINTS
  77 #include <asm/trace.h>
  78
  79 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
  80 EXPORT_PER_CPU_SYMBOL(irq_stat);
  81
  82 int __irq_offset_value;
  83
  84 #ifdef CONFIG_PPC32
  85 EXPORT_SYMBOL(__irq_offset_value);
  86 atomic_t ppc_n_lost_interrupts;
  87
  88 #ifdef CONFIG_TAU_INT
  89 extern int tau_initialized;
  90 extern int tau_interrupts(int);
  91 #endif
  92 #endif /* CONFIG_PPC32 */
  93
  94 #ifdef CONFIG_PPC64
  95
  96 int distribute_irqs = 1;
  97
  98 static inline notrace unsigned long get_irq_happened(void)
  99 {
 100         unsigned long happened;
 101
 102         __asm__ __volatile__("lbz %0,%1(13)"
 103         : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
 104
 105         return happened;
 106 }
 107
 108 static inline notrace void set_soft_enabled(unsigned long enable)
 109 {
 110         __asm__ __volatile__("stb %0,%1(13)"
 111         : : "r" (enable), "i" (offsetof(struct paca_struct, soft_enabled)));
 112 }
 113
 114 static inline notrace int decrementer_check_overflow(void)
 115 {
 116         u64 now = get_tb_or_rtc();
 117         u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
 118
 119         if (now >= *next_tb)
 120                 set_dec(1);
 121         return now >= *next_tb;
 122 }
 123
 124 /* This is called whenever we are re-enabling interrupts
 125  * and returns either 0 (nothing to do) or 500/900 if there's
 126  * either an EE or a DEC to generate.
 127  *
 128  * This is called in two contexts: From arch_local_irq_restore()
 129  * before soft-enabling interrupts, and from the exception exit
 130  * path when returning from an interrupt from a soft-disabled to
 131  * a soft enabled context. In both case we have interrupts hard
 132  * disabled.
 133  *
 134  * We take care of only clearing the bits we handled in the
 135  * PACA irq_happened field since we can only re-emit one at a
 136  * time and we don't want to "lose" one.
 137  */
 138 notrace unsigned int __check_irq_replay(void)
 139 {
 140         /*
 141          * We use local_paca rather than get_paca() to avoid all
 142          * the debug_smp_processor_id() business in this low level
 143          * function
 144          */
 145         unsigned char happened = local_paca->irq_happened;
 146
 147         /* Clear bit 0 which we wouldn't clear otherwise */
 148         local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
 149
 150         /*
 151          * Force the delivery of pending soft-disabled interrupts on PS3.
 152          * Any HV call will have this side effect.
 153          */
 154         if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
 155                 u64 tmp, tmp2;
 156                 lv1_get_version_info(&tmp, &tmp2);
 157         }
 158
 159         /*
 160          * We may have missed a decrementer interrupt. We check the
 161          * decrementer itself rather than the paca irq_happened field
 162          * in case we also had a rollover while hard disabled
 163          */
 164         local_paca->irq_happened &= ~PACA_IRQ_DEC;
 165         if (decrementer_check_overflow())
 166                 return 0x900;
 167
 168         /* Finally check if an external interrupt happened */
 169         local_paca->irq_happened &= ~PACA_IRQ_EE;
 170         if (happened & PACA_IRQ_EE)
 171                 return 0x500;
 172
 173 #ifdef CONFIG_PPC_BOOK3E
 174         /* Finally check if an EPR external interrupt happened
 175          * this bit is typically set if we need to handle another
 176          * "edge" interrupt from within the MPIC "EPR" handler
 177          */
 178         local_paca->irq_happened &= ~PACA_IRQ_EE_EDGE;
 179         if (happened & PACA_IRQ_EE_EDGE)
 180                 return 0x500;
 181
 182         local_paca->irq_happened &= ~PACA_IRQ_DBELL;
 183         if (happened & PACA_IRQ_DBELL)
 184                 return 0x280;
 185 #endif /* CONFIG_PPC_BOOK3E */
 186
 187         /* There should be nothing left ! */
 188         BUG_ON(local_paca->irq_happened != 0);
 189
 190         return 0;
 191 }
 192
 193 notrace void arch_local_irq_restore(unsigned long en)
 194 {
 195         unsigned char irq_happened;
 196         unsigned int replay;
 197
 198         /* Write the new soft-enabled value */
 199         set_soft_enabled(en);
 200         if (!en)
 201                 return;
 202         /*
 203          * From this point onward, we can take interrupts, preempt,
 204          * etc... unless we got hard-disabled. We check if an event
 205          * happened. If none happened, we know we can just return.
 206          *
 207          * We may have preempted before the check below, in which case
 208          * we are checking the "new" CPU instead of the old one. This
 209          * is only a problem if an event happened on the "old" CPU.
 210          *
 211          * External interrupt events on non-iseries will have caused
 212          * interrupts to be hard-disabled, so there is no problem, we
 213          * cannot have preempted.
 214          */
 215         irq_happened = get_irq_happened();
 216         if (!irq_happened)
 217                 return;
 218
 219         /*
 220          * We need to hard disable to get a trusted value from
 221          * __check_irq_replay(). We also need to soft-disable
 222          * again to avoid warnings in there due to the use of
 223          * per-cpu variables.
 224          *
 225          * We know that if the value in irq_happened is exactly 0x01
 226          * then we are already hard disabled (there are other less
 227          * common cases that we'll ignore for now), so we skip the
 228          * (expensive) mtmsrd.
 229          */
 230         if (unlikely(irq_happened != PACA_IRQ_HARD_DIS))
 231                 __hard_irq_disable();
 232         set_soft_enabled(0);
 233
 234         /*
 235          * Check if anything needs to be re-emitted. We haven't
 236          * soft-enabled yet to avoid warnings in decrementer_check_overflow
 237          * accessing per-cpu variables
 238          */
 239         replay = __check_irq_replay();
 240
 241         /* We can soft-enable now */
 242         set_soft_enabled(1);
 243
 244         /*
 245          * And replay if we have to. This will return with interrupts
 246          * hard-enabled.
 247          */
 248         if (replay) {
 249                 __replay_interrupt(replay);
 250                 return;
 251         }
 252
 253         /* Finally, let's ensure we are hard enabled */
 254         __hard_irq_enable();
 255 }
 256 EXPORT_SYMBOL(arch_local_irq_restore);
 257
 258 /*
 259  * This is specifically called by assembly code to re-enable interrupts
 260  * if they are currently disabled. This is typically called before
 261  * schedule() or do_signal() when returning to userspace. We do it
 262  * in C to avoid the burden of dealing with lockdep etc...
 263  */
 264 void restore_interrupts(void)
 265 {
 266         if (irqs_disabled())
 267                 local_irq_enable();
 268 }
 269
 270 #endif /* CONFIG_PPC64 */
 271
 272 int arch_show_interrupts(struct seq_file *p, int prec)
 273 {
 274         int j;
 275
 276 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
 277         if (tau_initialized) {
 278                 seq_printf(p, "%*s: ", prec, "TAU");
 279                 for_each_online_cpu(j)
 280                         seq_printf(p, "%10u ", tau_interrupts(j));
 281                 seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
 282         }
 283 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
 284
 285         seq_printf(p, "%*s: ", prec, "LOC");
 286         for_each_online_cpu(j)
 287                 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs);
 288         seq_printf(p, "  Local timer interrupts\n");
 289
 290         seq_printf(p, "%*s: ", prec, "SPU");
 291         for_each_online_cpu(j)
 292                 seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
 293         seq_printf(p, "  Spurious interrupts\n");
 294
 295         seq_printf(p, "%*s: ", prec, "CNT");
 296         for_each_online_cpu(j)
 297                 seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
 298         seq_printf(p, "  Performance monitoring interrupts\n");
 299
 300         seq_printf(p, "%*s: ", prec, "MCE");
 301         for_each_online_cpu(j)
 302                 seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
 303         seq_printf(p, "  Machine check exceptions\n");
 304
 305         return 0;
 306 }
 307
 308 /*
 309  * /proc/stat helpers
 310  */
 311 u64 arch_irq_stat_cpu(unsigned int cpu)
 312 {
 313         u64 sum = per_cpu(irq_stat, cpu).timer_irqs;
 314
 315         sum += per_cpu(irq_stat, cpu).pmu_irqs;
 316         sum += per_cpu(irq_stat, cpu).mce_exceptions;
 317         sum += per_cpu(irq_stat, cpu).spurious_irqs;
 318
 319         return sum;
 320 }
 321
 322 #ifdef CONFIG_HOTPLUG_CPU
 323 void migrate_irqs(void)
 324 {
 325         struct irq_desc *desc;
 326         unsigned int irq;
 327         static int warned;
 328         cpumask_var_t mask;
 329         const struct cpumask *map = cpu_online_mask;
 330
 331         alloc_cpumask_var(&mask, GFP_KERNEL);
 332
 333         for_each_irq(irq) {
 334                 struct irq_data *data;
 335                 struct irq_chip *chip;
 336
 337                 desc = irq_to_desc(irq);
 338                 if (!desc)
 339                         continue;
 340
 341                 data = irq_desc_get_irq_data(desc);
 342                 if (irqd_is_per_cpu(data))
 343                         continue;
 344
 345                 chip = irq_data_get_irq_chip(data);
 346
 347                 cpumask_and(mask, data->affinity, map);
 348                 if (cpumask_any(mask) >= nr_cpu_ids) {
 349                         printk("Breaking affinity for irq %i\n", irq);
 350                         cpumask_copy(mask, map);
 351                 }
 352                 if (chip->irq_set_affinity)
 353                         chip->irq_set_affinity(data, mask, true);
 354                 else if (desc->action && !(warned++))
 355                         printk("Cannot set affinity for irq %i\n", irq);
 356         }
 357
 358         free_cpumask_var(mask);
 359
 360         local_irq_enable();
 361         mdelay(1);
 362         local_irq_disable();
 363 }
 364 #endif
 365
 366 static inline void handle_one_irq(unsigned int irq)
 367 {
 368         struct thread_info *curtp, *irqtp;
 369         unsigned long saved_sp_limit;
 370         struct irq_desc *desc;
 371
 372         desc = irq_to_desc(irq);
 373         if (!desc)
 374                 return;
 375
 376         /* Switch to the irq stack to handle this */
 377         curtp = current_thread_info();
 378         irqtp = hardirq_ctx[smp_processor_id()];
 379
 380         if (curtp == irqtp) {
 381                 /* We're already on the irq stack, just handle it */
 382                 desc->handle_irq(irq, desc);
 383                 return;
 384         }
 385
 386         saved_sp_limit = current->thread.ksp_limit;
 387
 388         irqtp->task = curtp->task;
 389         irqtp->flags = 0;
 390
 391         /* Copy the softirq bits in preempt_count so that the
 392          * softirq checks work in the hardirq context. */
 393         irqtp->preempt_count = (irqtp->preempt_count & ~SOFTIRQ_MASK) |
 394                                (curtp->preempt_count & SOFTIRQ_MASK);
 395
 396         current->thread.ksp_limit = (unsigned long)irqtp +
 397                 _ALIGN_UP(sizeof(struct thread_info), 16);
 398
 399         call_handle_irq(irq, desc, irqtp, desc->handle_irq);
 400         current->thread.ksp_limit = saved_sp_limit;
 401         irqtp->task = NULL;
 402
 403         /* Set any flag that may have been set on the
 404          * alternate stack
 405          */
 406         if (irqtp->flags)
 407                 set_bits(irqtp->flags, &curtp->flags);
 408 }
 409
 410 static inline void check_stack_overflow(void)
 411 {
 412 #ifdef CONFIG_DEBUG_STACKOVERFLOW
 413         long sp;
 414
 415         sp = __get_SP() & (THREAD_SIZE-1);
 416
 417         /* check for stack overflow: is there less than 2KB free? */
 418         if (unlikely(sp < (sizeof(struct thread_info) + 2048))) {
 419                 printk("do_IRQ: stack overflow: %ld\n",
 420                         sp - sizeof(struct thread_info));
 421                 dump_stack();
 422         }
 423 #endif
 424 }
 425
 426 void do_IRQ(struct pt_regs *regs)
 427 {
 428         struct pt_regs *old_regs = set_irq_regs(regs);
 429         unsigned int irq;
 430
 431         trace_irq_entry(regs);
 432
 433         irq_enter();
 434
 435         check_stack_overflow();
 436
 437         /*
 438          * Query the platform PIC for the interrupt & ack it.
 439          *
 440          * This will typically lower the interrupt line to the CPU
 441          */
 442         irq = ppc_md.get_irq();
 443
 444         /* We can hard enable interrupts now */
 445         may_hard_irq_enable();
 446
 447         /* And finally process it */
 448         if (irq != NO_IRQ && irq != NO_IRQ_IGNORE)
 449                 handle_one_irq(irq);
 450         else if (irq != NO_IRQ_IGNORE)
 451                 __get_cpu_var(irq_stat).spurious_irqs++;
 452
 453         irq_exit();
 454         set_irq_regs(old_regs);
 455
 456         trace_irq_exit(regs);
 457 }
 458
 459 void __init init_IRQ(void)
 460 {
 461         if (ppc_md.init_IRQ)
 462                 ppc_md.init_IRQ();
 463
 464         exc_lvl_ctx_init();
 465
 466         irq_ctx_init();
 467 }
 468
 469 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
 470 struct thread_info   *critirq_ctx[NR_CPUS] __read_mostly;
 471 struct thread_info    *dbgirq_ctx[NR_CPUS] __read_mostly;
 472 struct thread_info *mcheckirq_ctx[NR_CPUS] __read_mostly;
 473
 474 void exc_lvl_ctx_init(void)
 475 {
 476         struct thread_info *tp;
 477         int i, cpu_nr;
 478
 479         for_each_possible_cpu(i) {
 480 #ifdef CONFIG_PPC64
 481                 cpu_nr = i;
 482 #else
 483                 cpu_nr = get_hard_smp_processor_id(i);
 484 #endif
 485                 memset((void *)critirq_ctx[cpu_nr], 0, THREAD_SIZE);
 486                 tp = critirq_ctx[cpu_nr];
 487                 tp->cpu = cpu_nr;
 488                 tp->preempt_count = 0;
 489
 490 #ifdef CONFIG_BOOKE
 491                 memset((void *)dbgirq_ctx[cpu_nr], 0, THREAD_SIZE);
 492                 tp = dbgirq_ctx[cpu_nr];
 493                 tp->cpu = cpu_nr;
 494                 tp->preempt_count = 0;
 495
 496                 memset((void *)mcheckirq_ctx[cpu_nr], 0, THREAD_SIZE);
 497                 tp = mcheckirq_ctx[cpu_nr];
 498                 tp->cpu = cpu_nr;
 499                 tp->preempt_count = HARDIRQ_OFFSET;
 500 #endif
 501         }
 502 }
 503 #endif
 504
 505 struct thread_info *softirq_ctx[NR_CPUS] __read_mostly;
 506 struct thread_info *hardirq_ctx[NR_CPUS] __read_mostly;
 507
 508 void irq_ctx_init(void)
 509 {
 510         struct thread_info *tp;
 511         int i;
 512
 513         for_each_possible_cpu(i) {
 514                 memset((void *)softirq_ctx[i], 0, THREAD_SIZE);
 515                 tp = softirq_ctx[i];
 516                 tp->cpu = i;
 517                 tp->preempt_count = 0;
 518
 519                 memset((void *)hardirq_ctx[i], 0, THREAD_SIZE);
 520                 tp = hardirq_ctx[i];
 521                 tp->cpu = i;
 522                 tp->preempt_count = HARDIRQ_OFFSET;
 523         }
 524 }
 525
 526 static inline void do_softirq_onstack(void)
 527 {
 528         struct thread_info *curtp, *irqtp;
 529         unsigned long saved_sp_limit = current->thread.ksp_limit;
 530
 531         curtp = current_thread_info();
 532         irqtp = softirq_ctx[smp_processor_id()];
 533         irqtp->task = curtp->task;
 534         irqtp->flags = 0;
 535         current->thread.ksp_limit = (unsigned long)irqtp +
 536                                     _ALIGN_UP(sizeof(struct thread_info), 16);
 537         call_do_softirq(irqtp);
 538         current->thread.ksp_limit = saved_sp_limit;
 539         irqtp->task = NULL;
 540
 541         /* Set any flag that may have been set on the
 542          * alternate stack
 543          */
 544         if (irqtp->flags)
 545                 set_bits(irqtp->flags, &curtp->flags);
 546 }
 547
 548 void do_softirq(void)
 549 {
 550         unsigned long flags;
 551
 552         if (in_interrupt())
 553                 return;
 554
 555         local_irq_save(flags);
 556
 557         if (local_softirq_pending())
 558                 do_softirq_onstack();
 559
 560         local_irq_restore(flags);
 561 }
 562
 563
 564 /*
 565  * IRQ controller and virtual interrupts
 566  */
 567
 568 /* The main irq map itself is an array of NR_IRQ entries containing the
 569  * associate host and irq number. An entry with a host of NULL is free.
 570  * An entry can be allocated if it's free, the allocator always then sets
 571  * hwirq first to the host's invalid irq number and then fills ops.
 572  */
 573 struct irq_map_entry {
 574         irq_hw_number_t hwirq;
 575         struct irq_host *host;
 576 };
 577
 578 static LIST_HEAD(irq_hosts);
 579 static DEFINE_RAW_SPINLOCK(irq_big_lock);
 580 static DEFINE_MUTEX(revmap_trees_mutex);
 581 static struct irq_map_entry irq_map[NR_IRQS];
 582 static unsigned int irq_virq_count = NR_IRQS;
 583 static struct irq_host *irq_default_host;
 584
 585 irq_hw_number_t irqd_to_hwirq(struct irq_data *d)
 586 {
 587         return irq_map[d->irq].hwirq;
 588 }
 589 EXPORT_SYMBOL_GPL(irqd_to_hwirq);
 590
 591 irq_hw_number_t virq_to_hw(unsigned int virq)
 592 {
 593         return irq_map[virq].hwirq;
 594 }
 595 EXPORT_SYMBOL_GPL(virq_to_hw);
 596
 597 bool virq_is_host(unsigned int virq, struct irq_host *host)
 598 {
 599         return irq_map[virq].host == host;
 600 }
 601 EXPORT_SYMBOL_GPL(virq_is_host);
 602
 603 static int default_irq_host_match(struct irq_host *h, struct device_node *np)
 604 {
 605         return h->of_node != NULL && h->of_node == np;
 606 }
 607
 608 struct irq_host *irq_alloc_host(struct device_node *of_node,
 609                                 unsigned int revmap_type,
 610                                 unsigned int revmap_arg,
 611                                 struct irq_host_ops *ops,
 612                                 irq_hw_number_t inval_irq)
 613 {
 614         struct irq_host *host;
 615         unsigned int size = sizeof(struct irq_host);
 616         unsigned int i;
 617         unsigned int *rmap;
 618         unsigned long flags;
 619
 620         /* Allocate structure and revmap table if using linear mapping */
 621         if (revmap_type == IRQ_HOST_MAP_LINEAR)
 622                 size += revmap_arg * sizeof(unsigned int);
 623         host = kzalloc(size, GFP_KERNEL);
 624         if (host == NULL)
 625                 return NULL;
 626
 627         /* Fill structure */
 628         host->revmap_type = revmap_type;
 629         host->inval_irq = inval_irq;
 630         host->ops = ops;
 631         host->of_node = of_node_get(of_node);
 632
 633         if (host->ops->match == NULL)
 634                 host->ops->match = default_irq_host_match;
 635
 636         raw_spin_lock_irqsave(&irq_big_lock, flags);
 637
 638         /* If it's a legacy controller, check for duplicates and
 639          * mark it as allocated (we use irq 0 host pointer for that
 640          */
 641         if (revmap_type == IRQ_HOST_MAP_LEGACY) {
 642                 if (irq_map[0].host != NULL) {
 643                         raw_spin_unlock_irqrestore(&irq_big_lock, flags);
 644                         of_node_put(host->of_node);
 645                         kfree(host);
 646                         return NULL;
 647                 }
 648                 irq_map[0].host = host;
 649         }
 650
 651         list_add(&host->link, &irq_hosts);
 652         raw_spin_unlock_irqrestore(&irq_big_lock, flags);
 653
 654         /* Additional setups per revmap type */
 655         switch(revmap_type) {
 656         case IRQ_HOST_MAP_LEGACY:
 657                 /* 0 is always the invalid number for legacy */
 658                 host->inval_irq = 0;
 659                 /* setup us as the host for all legacy interrupts */
 660                 for (i = 1; i < NUM_ISA_INTERRUPTS; i++) {
 661                         irq_map[i].hwirq = i;
 662                         smp_wmb();
 663                         irq_map[i].host = host;
 664                         smp_wmb();
 665
 666                         /* Legacy flags are left to default at this point,
 667                          * one can then use irq_create_mapping() to
 668                          * explicitly change them
 669                          */
 670                         ops->map(host, i, i);
 671
 672                         /* Clear norequest flags */
 673                         irq_clear_status_flags(i, IRQ_NOREQUEST);
 674                 }
 675                 break;
 676         case IRQ_HOST_MAP_LINEAR:
 677                 rmap = (unsigned int *)(host + 1);
 678                 for (i = 0; i < revmap_arg; i++)
 679                         rmap[i] = NO_IRQ;
 680                 host->revmap_data.linear.size = revmap_arg;
 681                 smp_wmb();
 682                 host->revmap_data.linear.revmap = rmap;
 683                 break;
 684         case IRQ_HOST_MAP_TREE:
 685                 INIT_RADIX_TREE(&host->revmap_data.tree, GFP_KERNEL);
 686                 break;
 687         default:
 688                 break;
 689         }
 690
 691         pr_debug("irq: Allocated host of type %d @0x%p\n", revmap_type, host);
 692
 693         return host;
 694 }
 695
 696 struct irq_host *irq_find_host(struct device_node *node)
 697 {
 698         struct irq_host *h, *found = NULL;
 699         unsigned long flags;
 700
 701         /* We might want to match the legacy controller last since
 702          * it might potentially be set to match all interrupts in
 703          * the absence of a device node. This isn't a problem so far
 704          * yet though...
 705          */
 706         raw_spin_lock_irqsave(&irq_big_lock, flags);
 707         list_for_each_entry(h, &irq_hosts, link)
 708                 if (h->ops->match(h, node)) {
 709                         found = h;
 710                         break;
 711                 }
 712         raw_spin_unlock_irqrestore(&irq_big_lock, flags);
 713         return found;
 714 }
 715 EXPORT_SYMBOL_GPL(irq_find_host);
 716
 717 void irq_set_default_host(struct irq_host *host)
 718 {
 719         pr_debug("irq: Default host set to @0x%p\n", host);
 720
 721         irq_default_host = host;
 722 }
 723
 724 void irq_set_virq_count(unsigned int count)
 725 {
 726         pr_debug("irq: Trying to set virq count to %d\n", count);
 727
 728         BUG_ON(count < NUM_ISA_INTERRUPTS);
 729         if (count < NR_IRQS)
 730                 irq_virq_count = count;
 731 }
 732
 733 static int irq_setup_virq(struct irq_host *host, unsigned int virq,
 734                             irq_hw_number_t hwirq)
 735 {
 736         int res;
 737
 738         res = irq_alloc_desc_at(virq, 0);
 739         if (res != virq) {
 740                 pr_debug("irq: -> allocating desc failed\n");
 741                 goto error;
 742         }
 743
 744         /* map it */
 745         smp_wmb();
 746         irq_map[virq].hwirq = hwirq;
 747         smp_mb();
 748
 749         if (host->ops->map(host, virq, hwirq)) {
 750                 pr_debug("irq: -> mapping failed, freeing\n");
 751                 goto errdesc;
 752         }
 753
 754         irq_clear_status_flags(virq, IRQ_NOREQUEST);
 755
 756         return 0;
 757
 758 errdesc:
 759         irq_free_descs(virq, 1);
 760 error:
 761         irq_free_virt(virq, 1);
 762         return -1;
 763 }
 764
 765 unsigned int irq_create_direct_mapping(struct irq_host *host)
 766 {
 767         unsigned int virq;
 768
 769         if (host == NULL)
 770                 host = irq_default_host;
 771
 772         BUG_ON(host == NULL);
 773         WARN_ON(host->revmap_type != IRQ_HOST_MAP_NOMAP);
 774
 775         virq = irq_alloc_virt(host, 1, 0);
 776         if (virq == NO_IRQ) {
 777                 pr_debug("irq: create_direct virq allocation failed\n");
 778                 return NO_IRQ;
 779         }
 780
 781         pr_debug("irq: create_direct obtained virq %d\n", virq);
 782
 783         if (irq_setup_virq(host, virq, virq))
 784                 return NO_IRQ;
 785
 786         return virq;
 787 }
 788
 789 unsigned int irq_create_mapping(struct irq_host *host,
 790                                 irq_hw_number_t hwirq)
 791 {
 792         unsigned int virq, hint;
 793
 794         pr_debug("irq: irq_create_mapping(0x%p, 0x%lx)\n", host, hwirq);
 795
 796         /* Look for default host if nececssary */
 797         if (host == NULL)
 798                 host = irq_default_host;
 799         if (host == NULL) {
 800                 printk(KERN_WARNING "irq_create_mapping called for"
 801                        " NULL host, hwirq=%lx\n", hwirq);
 802                 WARN_ON(1);
 803                 return NO_IRQ;
 804         }
 805         pr_debug("irq: -> using host @%p\n", host);
 806
 807         /* Check if mapping already exists */
 808         virq = irq_find_mapping(host, hwirq);
 809         if (virq != NO_IRQ) {
 810                 pr_debug("irq: -> existing mapping on virq %d\n", virq);
 811                 return virq;
 812         }
 813
 814         /* Get a virtual interrupt number */
 815         if (host->revmap_type == IRQ_HOST_MAP_LEGACY) {
 816                 /* Handle legacy */
 817                 virq = (unsigned int)hwirq;
 818                 if (virq == 0 || virq >= NUM_ISA_INTERRUPTS)
 819                         return NO_IRQ;
 820                 return virq;
 821         } else {
 822                 /* Allocate a virtual interrupt number */
 823                 hint = hwirq % irq_virq_count;
 824                 virq = irq_alloc_virt(host, 1, hint);
 825                 if (virq == NO_IRQ) {
 826                         pr_debug("irq: -> virq allocation failed\n");
 827                         return NO_IRQ;
 828                 }
 829         }
 830
 831         if (irq_setup_virq(host, virq, hwirq))
 832                 return NO_IRQ;
 833
 834         pr_debug("irq: irq %lu on host %s mapped to virtual irq %u\n",
 835                 hwirq, host->of_node ? host->of_node->full_name : "null", virq);
 836
 837         return virq;
 838 }
 839 EXPORT_SYMBOL_GPL(irq_create_mapping);
 840
 841 unsigned int irq_create_of_mapping(struct device_node *controller,
 842                                    const u32 *intspec, unsigned int intsize)
 843 {
 844         struct irq_host *host;
 845         irq_hw_number_t hwirq;
 846         unsigned int type = IRQ_TYPE_NONE;
 847         unsigned int virq;
 848
 849         if (controller == NULL)
 850                 host = irq_default_host;
 851         else
 852                 host = irq_find_host(controller);
 853         if (host == NULL) {
 854                 printk(KERN_WARNING "irq: no irq host found for %s !\n",
 855                        controller->full_name);
 856                 return NO_IRQ;
 857         }
 858
 859         /* If host has no translation, then we assume interrupt line */
 860         if (host->ops->xlate == NULL)
 861                 hwirq = intspec[0];
 862         else {
 863                 if (host->ops->xlate(host, controller, intspec, intsize,
 864                                      &hwirq, &type))
 865                         return NO_IRQ;
 866         }
 867
 868         /* Create mapping */
 869         virq = irq_create_mapping(host, hwirq);
 870         if (virq == NO_IRQ)
 871                 return virq;
 872
 873         /* Set type if specified and different than the current one */
 874         if (type != IRQ_TYPE_NONE &&
 875             type != (irqd_get_trigger_type(irq_get_irq_data(virq))))
 876                 irq_set_irq_type(virq, type);
 877         return virq;
 878 }
 879 EXPORT_SYMBOL_GPL(irq_create_of_mapping);
 880
 881 void irq_dispose_mapping(unsigned int virq)
 882 {
 883         struct irq_host *host;
 884         irq_hw_number_t hwirq;
 885
 886         if (virq == NO_IRQ)
 887                 return;
 888
 889         host = irq_map[virq].host;
 890         if (WARN_ON(host == NULL))
 891                 return;
 892
 893         /* Never unmap legacy interrupts */
 894         if (host->revmap_type == IRQ_HOST_MAP_LEGACY)
 895                 return;
 896
 897         irq_set_status_flags(virq, IRQ_NOREQUEST);
 898
 899         /* remove chip and handler */
 900         irq_set_chip_and_handler(virq, NULL, NULL);
 901
 902         /* Make sure it's completed */
 903         synchronize_irq(virq);
 904
 905         /* Tell the PIC about it */
 906         if (host->ops->unmap)
 907                 host->ops->unmap(host, virq);
 908         smp_mb();
 909
 910         /* Clear reverse map */
 911         hwirq = irq_map[virq].hwirq;
 912         switch(host->revmap_type) {
 913         case IRQ_HOST_MAP_LINEAR:
 914                 if (hwirq < host->revmap_data.linear.size)
 915                         host->revmap_data.linear.revmap[hwirq] = NO_IRQ;
 916                 break;
 917         case IRQ_HOST_MAP_TREE:
 918                 mutex_lock(&revmap_trees_mutex);
 919                 radix_tree_delete(&host->revmap_data.tree, hwirq);
 920                 mutex_unlock(&revmap_trees_mutex);
 921                 break;
 922         }
 923
 924         /* Destroy map */
 925         smp_mb();
 926         irq_map[virq].hwirq = host->inval_irq;
 927
 928         irq_free_descs(virq, 1);
 929         /* Free it */
 930         irq_free_virt(virq, 1);
 931 }
 932 EXPORT_SYMBOL_GPL(irq_dispose_mapping);
 933
 934 unsigned int irq_find_mapping(struct irq_host *host,
 935                               irq_hw_number_t hwirq)
 936 {
 937         unsigned int i;
 938         unsigned int hint = hwirq % irq_virq_count;
 939
 940         /* Look for default host if nececssary */
 941         if (host == NULL)
 942                 host = irq_default_host;
 943         if (host == NULL)
 944                 return NO_IRQ;
 945
 946         /* legacy -> bail early */
 947         if (host->revmap_type == IRQ_HOST_MAP_LEGACY)
 948                 return hwirq;
 949
 950         /* Slow path does a linear search of the map */
 951         if (hint < NUM_ISA_INTERRUPTS)
 952                 hint = NUM_ISA_INTERRUPTS;
 953         i = hint;
 954         do  {
 955                 if (irq_map[i].host == host &&
 956                     irq_map[i].hwirq == hwirq)
 957                         return i;
 958                 i++;
 959                 if (i >= irq_virq_count)
 960                         i = NUM_ISA_INTERRUPTS;
 961         } while(i != hint);
 962         return NO_IRQ;
 963 }
 964 EXPORT_SYMBOL_GPL(irq_find_mapping);
 965
 966 #ifdef CONFIG_SMP
 967 int irq_choose_cpu(const struct cpumask *mask)
 968 {
 969         int cpuid;
 970
 971         if (cpumask_equal(mask, cpu_all_mask)) {
 972                 static int irq_rover;
 973                 static DEFINE_RAW_SPINLOCK(irq_rover_lock);
 974                 unsigned long flags;
 975
 976                 /* Round-robin distribution... */
 977 do_round_robin:
 978                 raw_spin_lock_irqsave(&irq_rover_lock, flags);
 979
 980                 irq_rover = cpumask_next(irq_rover, cpu_online_mask);
 981                 if (irq_rover >= nr_cpu_ids)
 982                         irq_rover = cpumask_first(cpu_online_mask);
 983
 984                 cpuid = irq_rover;
 985
 986                 raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
 987         } else {
 988                 cpuid = cpumask_first_and(mask, cpu_online_mask);
 989                 if (cpuid >= nr_cpu_ids)
 990                         goto do_round_robin;
 991         }
 992
 993         return get_hard_smp_processor_id(cpuid);
 994 }
 995 #else
 996 int irq_choose_cpu(const struct cpumask *mask)
 997 {
 998         return hard_smp_processor_id();
 999 }
1000 #endif
1001
1002 unsigned int irq_radix_revmap_lookup(struct irq_host *host,
1003                                      irq_hw_number_t hwirq)
1004 {
1005         struct irq_map_entry *ptr;
1006         unsigned int virq;
1007
1008         if (WARN_ON_ONCE(host->revmap_type != IRQ_HOST_MAP_TREE))
1009                 return irq_find_mapping(host, hwirq);
1010
1011         /*
1012          * The ptr returned references the static global irq_map.
1013          * but freeing an irq can delete nodes along the path to
1014          * do the lookup via call_rcu.
1015          */
1016         rcu_read_lock();
1017         ptr = radix_tree_lookup(&host->revmap_data.tree, hwirq);
1018         rcu_read_unlock();
1019
1020         /*
1021          * If found in radix tree, then fine.
1022          * Else fallback to linear lookup - this should not happen in practice
1023          * as it means that we failed to insert the node in the radix tree.
1024          */
1025         if (ptr)
1026                 virq = ptr - irq_map;
1027         else
1028                 virq = irq_find_mapping(host, hwirq);
1029
1030         return virq;
1031 }
1032
1033 void irq_radix_revmap_insert(struct irq_host *host, unsigned int virq,
1034                              irq_hw_number_t hwirq)
1035 {
1036         if (WARN_ON(host->revmap_type != IRQ_HOST_MAP_TREE))
1037                 return;
1038
1039         if (virq != NO_IRQ) {
1040                 mutex_lock(&revmap_trees_mutex);
1041                 radix_tree_insert(&host->revmap_data.tree, hwirq,
1042                                   &irq_map[virq]);
1043                 mutex_unlock(&revmap_trees_mutex);
1044         }
1045 }
1046
1047 unsigned int irq_linear_revmap(struct irq_host *host,
1048                                irq_hw_number_t hwirq)
1049 {
1050         unsigned int *revmap;
1051
1052         if (WARN_ON_ONCE(host->revmap_type != IRQ_HOST_MAP_LINEAR))
1053                 return irq_find_mapping(host, hwirq);
1054
1055         /* Check revmap bounds */
1056         if (unlikely(hwirq >= host->revmap_data.linear.size))
1057                 return irq_find_mapping(host, hwirq);
1058
1059         /* Check if revmap was allocated */
1060         revmap = host->revmap_data.linear.revmap;
1061         if (unlikely(revmap == NULL))
1062                 return irq_find_mapping(host, hwirq);
1063
1064         /* Fill up revmap with slow path if no mapping found */
1065         if (unlikely(revmap[hwirq] == NO_IRQ))
1066                 revmap[hwirq] = irq_find_mapping(host, hwirq);
1067
1068         return revmap[hwirq];
1069 }
1070
1071 unsigned int irq_alloc_virt(struct irq_host *host,
1072                             unsigned int count,
1073                             unsigned int hint)
1074 {
1075         unsigned long flags;
1076         unsigned int i, j, found = NO_IRQ;
1077
1078         if (count == 0 || count > (irq_virq_count - NUM_ISA_INTERRUPTS))
1079                 return NO_IRQ;
1080
1081         raw_spin_lock_irqsave(&irq_big_lock, flags);
1082
1083         /* Use hint for 1 interrupt if any */
1084         if (count == 1 && hint >= NUM_ISA_INTERRUPTS &&
1085             hint < irq_virq_count && irq_map[hint].host == NULL) {
1086                 found = hint;
1087                 goto hint_found;
1088         }
1089
1090         /* Look for count consecutive numbers in the allocatable
1091          * (non-legacy) space
1092          */
1093         for (i = NUM_ISA_INTERRUPTS, j = 0; i < irq_virq_count; i++) {
1094                 if (irq_map[i].host != NULL)
1095                         j = 0;
1096                 else
1097                         j++;
1098
1099                 if (j == count) {
1100                         found = i - count + 1;
1101                         break;
1102                 }
1103         }
1104         if (found == NO_IRQ) {
1105                 raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1106                 return NO_IRQ;
1107         }
1108  hint_found:
1109         for (i = found; i < (found + count); i++) {
1110                 irq_map[i].hwirq = host->inval_irq;
1111                 smp_wmb();
1112                 irq_map[i].host = host;
1113         }
1114         raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1115         return found;
1116 }
1117
1118 void irq_free_virt(unsigned int virq, unsigned int count)
1119 {
1120         unsigned long flags;
1121         unsigned int i;
1122
1123         WARN_ON (virq < NUM_ISA_INTERRUPTS);
1124         WARN_ON (count == 0 || (virq + count) > irq_virq_count);
1125
1126         if (virq < NUM_ISA_INTERRUPTS) {
1127                 if (virq + count < NUM_ISA_INTERRUPTS)
1128                         return;
1129                 count  =- NUM_ISA_INTERRUPTS - virq;
1130                 virq = NUM_ISA_INTERRUPTS;
1131         }
1132
1133         if (count > irq_virq_count || virq > irq_virq_count - count) {
1134                 if (virq > irq_virq_count)
1135                         return;
1136                 count = irq_virq_count - virq;
1137         }
1138
1139         raw_spin_lock_irqsave(&irq_big_lock, flags);
1140         for (i = virq; i < (virq + count); i++) {
1141                 struct irq_host *host;
1142
1143                 host = irq_map[i].host;
1144                 irq_map[i].hwirq = host->inval_irq;
1145                 smp_wmb();
1146                 irq_map[i].host = NULL;
1147         }
1148         raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1149 }
1150
1151 int arch_early_irq_init(void)
1152 {
1153         return 0;
1154 }
1155
1156 #ifdef CONFIG_VIRQ_DEBUG
1157 static int virq_debug_show(struct seq_file *m, void *private)
1158 {
1159         unsigned long flags;
1160         struct irq_desc *desc;
1161         const char *p;
1162         static const char none[] = "none";
1163         void *data;
1164         int i;
1165
1166         seq_printf(m, "%-5s  %-7s  %-15s  %-18s  %s\n", "virq", "hwirq",
1167                       "chip name", "chip data", "host name");
1168
1169         for (i = 1; i < nr_irqs; i++) {
1170                 desc = irq_to_desc(i);
1171                 if (!desc)
1172                         continue;
1173
1174                 raw_spin_lock_irqsave(&desc->lock, flags);
1175
1176                 if (desc->action && desc->action->handler) {
1177                         struct irq_chip *chip;
1178
1179                         seq_printf(m, "%5d  ", i);
1180                         seq_printf(m, "0x%05lx  ", irq_map[i].hwirq);
1181
1182                         chip = irq_desc_get_chip(desc);
1183                         if (chip && chip->name)
1184                                 p = chip->name;
1185                         else
1186                                 p = none;
1187                         seq_printf(m, "%-15s  ", p);
1188
1189                         data = irq_desc_get_chip_data(desc);
1190                         seq_printf(m, "0x%16p  ", data);
1191
1192                         if (irq_map[i].host && irq_map[i].host->of_node)
1193                                 p = irq_map[i].host->of_node->full_name;
1194                         else
1195                                 p = none;
1196                         seq_printf(m, "%s\n", p);
1197                 }
1198
1199                 raw_spin_unlock_irqrestore(&desc->lock, flags);
1200         }
1201
1202         return 0;
1203 }
1204
1205 static int virq_debug_open(struct inode *inode, struct file *file)
1206 {
1207         return single_open(file, virq_debug_show, inode->i_private);
1208 }
1209
1210 static const struct file_operations virq_debug_fops = {
1211         .open = virq_debug_open,
1212         .read = seq_read,
1213         .llseek = seq_lseek,
1214         .release = single_release,
1215 };
1216
1217 static int __init irq_debugfs_init(void)
1218 {
1219         if (debugfs_create_file("virq_mapping", S_IRUGO, powerpc_debugfs_root,
1220                                  NULL, &virq_debug_fops) == NULL)
1221                 return -ENOMEM;
1222
1223         return 0;
1224 }
1225 __initcall(irq_debugfs_init);
1226 #endif /* CONFIG_VIRQ_DEBUG */
1227
1228 #ifdef CONFIG_PPC64
1229 static int __init setup_noirqdistrib(char *str)
1230 {
1231         distribute_irqs = 0;
1232         return 1;
1233 }
1234
1235 __setup("noirqdistrib", setup_noirqdistrib);
1236 #endif /* CONFIG_PPC64 */