drivers/xen/events.c

   1 /*
   2  * Xen event channels
   3  *
   4  * Xen models interrupts with abstract event channels.  Because each
   5  * domain gets 1024 event channels, but NR_IRQ is not that large, we
   6  * must dynamically map irqs<->event channels.  The event channels
   7  * interface with the rest of the kernel by defining a xen interrupt
   8  * chip.  When an event is recieved, it is mapped to an irq and sent
   9  * through the normal interrupt processing path.
  10  *
  11  * There are four kinds of events which can be mapped to an event
  12  * channel:
  13  *
  14  * 1. Inter-domain notifications.  This includes all the virtual
  15  *    device events, since they're driven by front-ends in another domain
  16  *    (typically dom0).
  17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
  18  * 3. IPIs.
  19  * 4. PIRQs - Hardware interrupts.
  20  *
  21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  22  */
  23
  24 #include <linux/linkage.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/irq.h>
  27 #include <linux/module.h>
  28 #include <linux/string.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/slab.h>
  31 #include <linux/irqnr.h>
  32 #include <linux/pci.h>
  33
  34 #include <asm/desc.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/irq.h>
  37 #include <asm/idle.h>
  38 #include <asm/io_apic.h>
  39 #include <asm/sync_bitops.h>
  40 #include <asm/xen/pci.h>
  41 #include <asm/xen/hypercall.h>
  42 #include <asm/xen/hypervisor.h>
  43
  44 #include <xen/xen.h>
  45 #include <xen/hvm.h>
  46 #include <xen/xen-ops.h>
  47 #include <xen/events.h>
  48 #include <xen/interface/xen.h>
  49 #include <xen/interface/event_channel.h>
  50 #include <xen/interface/hvm/hvm_op.h>
  51 #include <xen/interface/hvm/params.h>
  52
  53 /*
  54  * This lock protects updates to the following mapping and reference-count
  55  * arrays. The lock does not need to be acquired to read the mapping tables.
  56  */
  57 static DEFINE_SPINLOCK(irq_mapping_update_lock);
  58
  59 /* IRQ <-> VIRQ mapping. */
  60 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  61
  62 /* IRQ <-> IPI mapping */
  63 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  64
  65 /* Interrupt types. */
  66 enum xen_irq_type {
  67         IRQT_UNBOUND = 0,
  68         IRQT_PIRQ,
  69         IRQT_VIRQ,
  70         IRQT_IPI,
  71         IRQT_EVTCHN
  72 };
  73
  74 /*
  75  * Packed IRQ information:
  76  * type - enum xen_irq_type
  77  * event channel - irq->event channel mapping
  78  * cpu - cpu this event channel is bound to
  79  * index - type-specific information:
  80  *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
  81  *           guest, or GSI (real passthrough IRQ) of the device.
  82  *    VIRQ - virq number
  83  *    IPI - IPI vector
  84  *    EVTCHN -
  85  */
  86 struct irq_info
  87 {
  88         enum xen_irq_type type; /* type */
  89         unsigned short evtchn;  /* event channel */
  90         unsigned short cpu;     /* cpu bound */
  91
  92         union {
  93                 unsigned short virq;
  94                 enum ipi_vector ipi;
  95                 struct {
  96                         unsigned short pirq;
  97                         unsigned short gsi;
  98                         unsigned char vector;
  99                         unsigned char flags;
 100                 } pirq;
 101         } u;
 102 };
 103 #define PIRQ_NEEDS_EOI  (1 << 0)
 104 #define PIRQ_SHAREABLE  (1 << 1)
 105
 106 static struct irq_info *irq_info;
 107 static int *pirq_to_irq;
 108
 109 static int *evtchn_to_irq;
 110 struct cpu_evtchn_s {
 111         unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
 112 };
 113
 114 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
 115         .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
 116 };
 117 static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
 118
 119 static inline unsigned long *cpu_evtchn_mask(int cpu)
 120 {
 121         return cpu_evtchn_mask_p[cpu].bits;
 122 }
 123
 124 /* Xen will never allocate port zero for any purpose. */
 125 #define VALID_EVTCHN(chn)       ((chn) != 0)
 126
 127 static struct irq_chip xen_dynamic_chip;
 128 static struct irq_chip xen_percpu_chip;
 129 static struct irq_chip xen_pirq_chip;
 130
 131 /* Constructor for packed IRQ information. */
 132 static struct irq_info mk_unbound_info(void)
 133 {
 134         return (struct irq_info) { .type = IRQT_UNBOUND };
 135 }
 136
 137 static struct irq_info mk_evtchn_info(unsigned short evtchn)
 138 {
 139         return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
 140                         .cpu = 0 };
 141 }
 142
 143 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
 144 {
 145         return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
 146                         .cpu = 0, .u.ipi = ipi };
 147 }
 148
 149 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
 150 {
 151         return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
 152                         .cpu = 0, .u.virq = virq };
 153 }
 154
 155 static struct irq_info mk_pirq_info(unsigned short evtchn, unsigned short pirq,
 156                                     unsigned short gsi, unsigned short vector)
 157 {
 158         return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
 159                         .cpu = 0,
 160                         .u.pirq = { .pirq = pirq, .gsi = gsi, .vector = vector } };
 161 }
 162
 163 /*
 164  * Accessors for packed IRQ information.
 165  */
 166 static struct irq_info *info_for_irq(unsigned irq)
 167 {
 168         return &irq_info[irq];
 169 }
 170
 171 static unsigned int evtchn_from_irq(unsigned irq)
 172 {
 173         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 174                 return 0;
 175
 176         return info_for_irq(irq)->evtchn;
 177 }
 178
 179 unsigned irq_from_evtchn(unsigned int evtchn)
 180 {
 181         return evtchn_to_irq[evtchn];
 182 }
 183 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 184
 185 static enum ipi_vector ipi_from_irq(unsigned irq)
 186 {
 187         struct irq_info *info = info_for_irq(irq);
 188
 189         BUG_ON(info == NULL);
 190         BUG_ON(info->type != IRQT_IPI);
 191
 192         return info->u.ipi;
 193 }
 194
 195 static unsigned virq_from_irq(unsigned irq)
 196 {
 197         struct irq_info *info = info_for_irq(irq);
 198
 199         BUG_ON(info == NULL);
 200         BUG_ON(info->type != IRQT_VIRQ);
 201
 202         return info->u.virq;
 203 }
 204
 205 static unsigned pirq_from_irq(unsigned irq)
 206 {
 207         struct irq_info *info = info_for_irq(irq);
 208
 209         BUG_ON(info == NULL);
 210         BUG_ON(info->type != IRQT_PIRQ);
 211
 212         return info->u.pirq.pirq;
 213 }
 214
 215 static unsigned gsi_from_irq(unsigned irq)
 216 {
 217         struct irq_info *info = info_for_irq(irq);
 218
 219         BUG_ON(info == NULL);
 220         BUG_ON(info->type != IRQT_PIRQ);
 221
 222         return info->u.pirq.gsi;
 223 }
 224
 225 static unsigned vector_from_irq(unsigned irq)
 226 {
 227         struct irq_info *info = info_for_irq(irq);
 228
 229         BUG_ON(info == NULL);
 230         BUG_ON(info->type != IRQT_PIRQ);
 231
 232         return info->u.pirq.vector;
 233 }
 234
 235 static enum xen_irq_type type_from_irq(unsigned irq)
 236 {
 237         return info_for_irq(irq)->type;
 238 }
 239
 240 static unsigned cpu_from_irq(unsigned irq)
 241 {
 242         return info_for_irq(irq)->cpu;
 243 }
 244
 245 static unsigned int cpu_from_evtchn(unsigned int evtchn)
 246 {
 247         int irq = evtchn_to_irq[evtchn];
 248         unsigned ret = 0;
 249
 250         if (irq != -1)
 251                 ret = cpu_from_irq(irq);
 252
 253         return ret;
 254 }
 255
 256 static bool pirq_needs_eoi(unsigned irq)
 257 {
 258         struct irq_info *info = info_for_irq(irq);
 259
 260         BUG_ON(info->type != IRQT_PIRQ);
 261
 262         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 263 }
 264
 265 static inline unsigned long active_evtchns(unsigned int cpu,
 266                                            struct shared_info *sh,
 267                                            unsigned int idx)
 268 {
 269         return (sh->evtchn_pending[idx] &
 270                 cpu_evtchn_mask(cpu)[idx] &
 271                 ~sh->evtchn_mask[idx]);
 272 }
 273
 274 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 275 {
 276         int irq = evtchn_to_irq[chn];
 277
 278         BUG_ON(irq == -1);
 279 #ifdef CONFIG_SMP
 280         cpumask_copy(irq_to_desc(irq)->affinity, cpumask_of(cpu));
 281 #endif
 282
 283         clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
 284         set_bit(chn, cpu_evtchn_mask(cpu));
 285
 286         irq_info[irq].cpu = cpu;
 287 }
 288
 289 static void init_evtchn_cpu_bindings(void)
 290 {
 291         int i;
 292 #ifdef CONFIG_SMP
 293         struct irq_desc *desc;
 294
 295         /* By default all event channels notify CPU#0. */
 296         for_each_irq_desc(i, desc) {
 297                 cpumask_copy(desc->affinity, cpumask_of(0));
 298         }
 299 #endif
 300
 301         for_each_possible_cpu(i)
 302                 memset(cpu_evtchn_mask(i),
 303                        (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
 304
 305 }
 306
 307 static inline void clear_evtchn(int port)
 308 {
 309         struct shared_info *s = HYPERVISOR_shared_info;
 310         sync_clear_bit(port, &s->evtchn_pending[0]);
 311 }
 312
 313 static inline void set_evtchn(int port)
 314 {
 315         struct shared_info *s = HYPERVISOR_shared_info;
 316         sync_set_bit(port, &s->evtchn_pending[0]);
 317 }
 318
 319 static inline int test_evtchn(int port)
 320 {
 321         struct shared_info *s = HYPERVISOR_shared_info;
 322         return sync_test_bit(port, &s->evtchn_pending[0]);
 323 }
 324
 325
 326 /**
 327  * notify_remote_via_irq - send event to remote end of event channel via irq
 328  * @irq: irq of event channel to send event to
 329  *
 330  * Unlike notify_remote_via_evtchn(), this is safe to use across
 331  * save/restore. Notifications on a broken connection are silently
 332  * dropped.
 333  */
 334 void notify_remote_via_irq(int irq)
 335 {
 336         int evtchn = evtchn_from_irq(irq);
 337
 338         if (VALID_EVTCHN(evtchn))
 339                 notify_remote_via_evtchn(evtchn);
 340 }
 341 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 342
 343 static void mask_evtchn(int port)
 344 {
 345         struct shared_info *s = HYPERVISOR_shared_info;
 346         sync_set_bit(port, &s->evtchn_mask[0]);
 347 }
 348
 349 static void unmask_evtchn(int port)
 350 {
 351         struct shared_info *s = HYPERVISOR_shared_info;
 352         unsigned int cpu = get_cpu();
 353
 354         BUG_ON(!irqs_disabled());
 355
 356         /* Slow path (hypercall) if this is a non-local port. */
 357         if (unlikely(cpu != cpu_from_evtchn(port))) {
 358                 struct evtchn_unmask unmask = { .port = port };
 359                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
 360         } else {
 361                 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
 362
 363                 sync_clear_bit(port, &s->evtchn_mask[0]);
 364
 365                 /*
 366                  * The following is basically the equivalent of
 367                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
 368                  * the interrupt edge' if the channel is masked.
 369                  */
 370                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
 371                     !sync_test_and_set_bit(port / BITS_PER_LONG,
 372                                            &vcpu_info->evtchn_pending_sel))
 373                         vcpu_info->evtchn_upcall_pending = 1;
 374         }
 375
 376         put_cpu();
 377 }
 378
 379 static int xen_allocate_irq_dynamic(void)
 380 {
 381         int first = 0;
 382         int irq;
 383
 384 #ifdef CONFIG_X86_IO_APIC
 385         /*
 386          * For an HVM guest or domain 0 which see "real" (emulated or
 387          * actual repectively) GSIs we allocate dynamic IRQs
 388          * e.g. those corresponding to event channels or MSIs
 389          * etc. from the range above those "real" GSIs to avoid
 390          * collisions.
 391          */
 392         if (xen_initial_domain() || xen_hvm_domain())
 393                 first = get_nr_irqs_gsi();
 394 #endif
 395
 396 retry:
 397         irq = irq_alloc_desc_from(first, -1);
 398
 399         if (irq == -ENOMEM && first > NR_IRQS_LEGACY) {
 400                 printk(KERN_ERR "Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
 401                 first = max(NR_IRQS_LEGACY, first - NR_IRQS_LEGACY);
 402                 goto retry;
 403         }
 404
 405         if (irq < 0)
 406                 panic("No available IRQ to bind to: increase nr_irqs!\n");
 407
 408         return irq;
 409 }
 410
 411 static int xen_allocate_irq_gsi(unsigned gsi)
 412 {
 413         int irq;
 414
 415         /*
 416          * A PV guest has no concept of a GSI (since it has no ACPI
 417          * nor access to/knowledge of the physical APICs). Therefore
 418          * all IRQs are dynamically allocated from the entire IRQ
 419          * space.
 420          */
 421         if (xen_pv_domain() && !xen_initial_domain())
 422                 return xen_allocate_irq_dynamic();
 423
 424         /* Legacy IRQ descriptors are already allocated by the arch. */
 425         if (gsi < NR_IRQS_LEGACY)
 426                 return gsi;
 427
 428         irq = irq_alloc_desc_at(gsi, -1);
 429         if (irq < 0)
 430                 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
 431
 432         return irq;
 433 }
 434
 435 static void xen_free_irq(unsigned irq)
 436 {
 437         /* Legacy IRQ descriptors are managed by the arch. */
 438         if (irq < NR_IRQS_LEGACY)
 439                 return;
 440
 441         irq_free_desc(irq);
 442 }
 443
 444 static void pirq_unmask_notify(int irq)
 445 {
 446         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
 447
 448         if (unlikely(pirq_needs_eoi(irq))) {
 449                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 450                 WARN_ON(rc);
 451         }
 452 }
 453
 454 static void pirq_query_unmask(int irq)
 455 {
 456         struct physdev_irq_status_query irq_status;
 457         struct irq_info *info = info_for_irq(irq);
 458
 459         BUG_ON(info->type != IRQT_PIRQ);
 460
 461         irq_status.irq = pirq_from_irq(irq);
 462         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 463                 irq_status.flags = 0;
 464
 465         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 466         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 467                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 468 }
 469
 470 static bool probing_irq(int irq)
 471 {
 472         struct irq_desc *desc = irq_to_desc(irq);
 473
 474         return desc && desc->action == NULL;
 475 }
 476
 477 static unsigned int startup_pirq(unsigned int irq)
 478 {
 479         struct evtchn_bind_pirq bind_pirq;
 480         struct irq_info *info = info_for_irq(irq);
 481         int evtchn = evtchn_from_irq(irq);
 482         int rc;
 483
 484         BUG_ON(info->type != IRQT_PIRQ);
 485
 486         if (VALID_EVTCHN(evtchn))
 487                 goto out;
 488
 489         bind_pirq.pirq = pirq_from_irq(irq);
 490         /* NB. We are happy to share unless we are probing. */
 491         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 492                                         BIND_PIRQ__WILL_SHARE : 0;
 493         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 494         if (rc != 0) {
 495                 if (!probing_irq(irq))
 496                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
 497                                irq);
 498                 return 0;
 499         }
 500         evtchn = bind_pirq.port;
 501
 502         pirq_query_unmask(irq);
 503
 504         evtchn_to_irq[evtchn] = irq;
 505         bind_evtchn_to_cpu(evtchn, 0);
 506         info->evtchn = evtchn;
 507
 508 out:
 509         unmask_evtchn(evtchn);
 510         pirq_unmask_notify(irq);
 511
 512         return 0;
 513 }
 514
 515 static void shutdown_pirq(unsigned int irq)
 516 {
 517         struct evtchn_close close;
 518         struct irq_info *info = info_for_irq(irq);
 519         int evtchn = evtchn_from_irq(irq);
 520
 521         BUG_ON(info->type != IRQT_PIRQ);
 522
 523         if (!VALID_EVTCHN(evtchn))
 524                 return;
 525
 526         mask_evtchn(evtchn);
 527
 528         close.port = evtchn;
 529         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 530                 BUG();
 531
 532         bind_evtchn_to_cpu(evtchn, 0);
 533         evtchn_to_irq[evtchn] = -1;
 534         info->evtchn = 0;
 535 }
 536
 537 static void enable_pirq(unsigned int irq)
 538 {
 539         startup_pirq(irq);
 540 }
 541
 542 static void disable_pirq(unsigned int irq)
 543 {
 544 }
 545
 546 static void ack_pirq(unsigned int irq)
 547 {
 548         int evtchn = evtchn_from_irq(irq);
 549
 550         move_native_irq(irq);
 551
 552         if (VALID_EVTCHN(evtchn)) {
 553                 mask_evtchn(evtchn);
 554                 clear_evtchn(evtchn);
 555         }
 556 }
 557
 558 static void end_pirq(unsigned int irq)
 559 {
 560         int evtchn = evtchn_from_irq(irq);
 561         struct irq_desc *desc = irq_to_desc(irq);
 562
 563         if (WARN_ON(!desc))
 564                 return;
 565
 566         if ((desc->status & (IRQ_DISABLED|IRQ_PENDING)) ==
 567             (IRQ_DISABLED|IRQ_PENDING)) {
 568                 shutdown_pirq(irq);
 569         } else if (VALID_EVTCHN(evtchn)) {
 570                 unmask_evtchn(evtchn);
 571                 pirq_unmask_notify(irq);
 572         }
 573 }
 574
 575 static int find_irq_by_gsi(unsigned gsi)
 576 {
 577         int irq;
 578
 579         for (irq = 0; irq < nr_irqs; irq++) {
 580                 struct irq_info *info = info_for_irq(irq);
 581
 582                 if (info == NULL || info->type != IRQT_PIRQ)
 583                         continue;
 584
 585                 if (gsi_from_irq(irq) == gsi)
 586                         return irq;
 587         }
 588
 589         return -1;
 590 }
 591
 592 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
 593 {
 594         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
 595 }
 596
 597 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
 598  * consequence don't assume that the irq number returned has a low value
 599  * or can be used as a pirq number unless you know otherwise.
 600  *
 601  * One notable exception is when xen_map_pirq_gsi is called passing an
 602  * hardware gsi as argument, in that case the irq number returned
 603  * matches the gsi number passed as second argument.
 604  *
 605  * Note: We don't assign an event channel until the irq actually started
 606  * up.  Return an existing irq if we've already got one for the gsi.
 607  */
 608 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
 609 {
 610         int irq = 0;
 611         struct physdev_irq irq_op;
 612
 613         spin_lock(&irq_mapping_update_lock);
 614
 615         if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
 616                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
 617                         pirq > nr_irqs ? "pirq" :"",
 618                         gsi > nr_irqs ? "gsi" : "");
 619                 goto out;
 620         }
 621
 622         irq = find_irq_by_gsi(gsi);
 623         if (irq != -1) {
 624                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
 625                        irq, gsi);
 626                 goto out;       /* XXX need refcount? */
 627         }
 628
 629         irq = xen_allocate_irq_gsi(gsi);
 630
 631         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 632                                       handle_level_irq, name);
 633
 634         irq_op.irq = irq;
 635         irq_op.vector = 0;
 636
 637         /* Only the privileged domain can do this. For non-priv, the pcifront
 638          * driver provides a PCI bus that does the call to do exactly
 639          * this in the priv domain. */
 640         if (xen_initial_domain() &&
 641             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 642                 xen_free_irq(irq);
 643                 irq = -ENOSPC;
 644                 goto out;
 645         }
 646
 647         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
 648         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
 649         pirq_to_irq[pirq] = irq;
 650
 651 out:
 652         spin_unlock(&irq_mapping_update_lock);
 653
 654         return irq;
 655 }
 656
 657 #ifdef CONFIG_PCI_MSI
 658 #include <linux/msi.h>
 659 #include "../pci/msi.h"
 660
 661 static int find_unbound_pirq(int type)
 662 {
 663         int rc, i;
 664         struct physdev_get_free_pirq op_get_free_pirq;
 665         op_get_free_pirq.type = type;
 666
 667         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 668         if (!rc)
 669                 return op_get_free_pirq.pirq;
 670
 671         for (i = 0; i < nr_irqs; i++) {
 672                 if (pirq_to_irq[i] < 0)
 673                         return i;
 674         }
 675         return -1;
 676 }
 677
 678 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc)
 679 {
 680         spin_lock(&irq_mapping_update_lock);
 681
 682         if (alloc & XEN_ALLOC_IRQ) {
 683                 *irq = xen_allocate_irq_dynamic();
 684                 if (*irq == -1)
 685                         goto out;
 686         }
 687
 688         if (alloc & XEN_ALLOC_PIRQ) {
 689                 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
 690                 if (*pirq == -1)
 691                         goto out;
 692         }
 693
 694         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
 695                                       handle_level_irq, name);
 696
 697         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
 698         pirq_to_irq[*pirq] = *irq;
 699
 700 out:
 701         spin_unlock(&irq_mapping_update_lock);
 702 }
 703
 704 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
 705 {
 706         int irq = -1;
 707         struct physdev_map_pirq map_irq;
 708         int rc;
 709         int pos;
 710         u32 table_offset, bir;
 711
 712         memset(&map_irq, 0, sizeof(map_irq));
 713         map_irq.domid = DOMID_SELF;
 714         map_irq.type = MAP_PIRQ_TYPE_MSI;
 715         map_irq.index = -1;
 716         map_irq.pirq = -1;
 717         map_irq.bus = dev->bus->number;
 718         map_irq.devfn = dev->devfn;
 719
 720         if (type == PCI_CAP_ID_MSIX) {
 721                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 722
 723                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
 724                                         &table_offset);
 725                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 726
 727                 map_irq.table_base = pci_resource_start(dev, bir);
 728                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 729         }
 730
 731         spin_lock(&irq_mapping_update_lock);
 732
 733         irq = xen_allocate_irq_dynamic();
 734
 735         if (irq == -1)
 736                 goto out;
 737
 738         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 739         if (rc) {
 740                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
 741
 742                 xen_free_irq(irq);
 743
 744                 irq = -1;
 745                 goto out;
 746         }
 747         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
 748
 749         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 750                         handle_level_irq,
 751                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
 752
 753 out:
 754         spin_unlock(&irq_mapping_update_lock);
 755         return irq;
 756 }
 757 #endif
 758
 759 int xen_destroy_irq(int irq)
 760 {
 761         struct irq_desc *desc;
 762         struct physdev_unmap_pirq unmap_irq;
 763         struct irq_info *info = info_for_irq(irq);
 764         int rc = -ENOENT;
 765
 766         spin_lock(&irq_mapping_update_lock);
 767
 768         desc = irq_to_desc(irq);
 769         if (!desc)
 770                 goto out;
 771
 772         if (xen_initial_domain()) {
 773                 unmap_irq.pirq = info->u.pirq.pirq;
 774                 unmap_irq.domid = DOMID_SELF;
 775                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 776                 if (rc) {
 777                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
 778                         goto out;
 779                 }
 780                 pirq_to_irq[info->u.pirq.pirq] = -1;
 781         }
 782         irq_info[irq] = mk_unbound_info();
 783
 784         xen_free_irq(irq);
 785
 786 out:
 787         spin_unlock(&irq_mapping_update_lock);
 788         return rc;
 789 }
 790
 791 int xen_vector_from_irq(unsigned irq)
 792 {
 793         return vector_from_irq(irq);
 794 }
 795
 796 int xen_gsi_from_irq(unsigned irq)
 797 {
 798         return gsi_from_irq(irq);
 799 }
 800
 801 int xen_irq_from_pirq(unsigned pirq)
 802 {
 803         return pirq_to_irq[pirq];
 804 }
 805
 806 int bind_evtchn_to_irq(unsigned int evtchn)
 807 {
 808         int irq;
 809
 810         spin_lock(&irq_mapping_update_lock);
 811
 812         irq = evtchn_to_irq[evtchn];
 813
 814         if (irq == -1) {
 815                 irq = xen_allocate_irq_dynamic();
 816
 817                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
 818                                               handle_fasteoi_irq, "event");
 819
 820                 evtchn_to_irq[evtchn] = irq;
 821                 irq_info[irq] = mk_evtchn_info(evtchn);
 822         }
 823
 824         spin_unlock(&irq_mapping_update_lock);
 825
 826         return irq;
 827 }
 828 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 829
 830 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 831 {
 832         struct evtchn_bind_ipi bind_ipi;
 833         int evtchn, irq;
 834
 835         spin_lock(&irq_mapping_update_lock);
 836
 837         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 838
 839         if (irq == -1) {
 840                 irq = xen_allocate_irq_dynamic();
 841                 if (irq < 0)
 842                         goto out;
 843
 844                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 845                                               handle_percpu_irq, "ipi");
 846
 847                 bind_ipi.vcpu = cpu;
 848                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 849                                                 &bind_ipi) != 0)
 850                         BUG();
 851                 evtchn = bind_ipi.port;
 852
 853                 evtchn_to_irq[evtchn] = irq;
 854                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
 855                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 856
 857                 bind_evtchn_to_cpu(evtchn, cpu);
 858         }
 859
 860  out:
 861         spin_unlock(&irq_mapping_update_lock);
 862         return irq;
 863 }
 864
 865
 866 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 867 {
 868         struct evtchn_bind_virq bind_virq;
 869         int evtchn, irq;
 870
 871         spin_lock(&irq_mapping_update_lock);
 872
 873         irq = per_cpu(virq_to_irq, cpu)[virq];
 874
 875         if (irq == -1) {
 876                 irq = xen_allocate_irq_dynamic();
 877
 878                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 879                                               handle_percpu_irq, "virq");
 880
 881                 bind_virq.virq = virq;
 882                 bind_virq.vcpu = cpu;
 883                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 884                                                 &bind_virq) != 0)
 885                         BUG();
 886                 evtchn = bind_virq.port;
 887
 888                 evtchn_to_irq[evtchn] = irq;
 889                 irq_info[irq] = mk_virq_info(evtchn, virq);
 890
 891                 per_cpu(virq_to_irq, cpu)[virq] = irq;
 892
 893                 bind_evtchn_to_cpu(evtchn, cpu);
 894         }
 895
 896         spin_unlock(&irq_mapping_update_lock);
 897
 898         return irq;
 899 }
 900
 901 static void unbind_from_irq(unsigned int irq)
 902 {
 903         struct evtchn_close close;
 904         int evtchn = evtchn_from_irq(irq);
 905
 906         spin_lock(&irq_mapping_update_lock);
 907
 908         if (VALID_EVTCHN(evtchn)) {
 909                 close.port = evtchn;
 910                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 911                         BUG();
 912
 913                 switch (type_from_irq(irq)) {
 914                 case IRQT_VIRQ:
 915                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 916                                 [virq_from_irq(irq)] = -1;
 917                         break;
 918                 case IRQT_IPI:
 919                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 920                                 [ipi_from_irq(irq)] = -1;
 921                         break;
 922                 default:
 923                         break;
 924                 }
 925
 926                 /* Closed ports are implicitly re-bound to VCPU0. */
 927                 bind_evtchn_to_cpu(evtchn, 0);
 928
 929                 evtchn_to_irq[evtchn] = -1;
 930         }
 931
 932         if (irq_info[irq].type != IRQT_UNBOUND) {
 933                 irq_info[irq] = mk_unbound_info();
 934
 935                 xen_free_irq(irq);
 936         }
 937
 938         spin_unlock(&irq_mapping_update_lock);
 939 }
 940
 941 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 942                               irq_handler_t handler,
 943                               unsigned long irqflags,
 944                               const char *devname, void *dev_id)
 945 {
 946         unsigned int irq;
 947         int retval;
 948
 949         irq = bind_evtchn_to_irq(evtchn);
 950         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 951         if (retval != 0) {
 952                 unbind_from_irq(irq);
 953                 return retval;
 954         }
 955
 956         return irq;
 957 }
 958 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 959
 960 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 961                             irq_handler_t handler,
 962                             unsigned long irqflags, const char *devname, void *dev_id)
 963 {
 964         unsigned int irq;
 965         int retval;
 966
 967         irq = bind_virq_to_irq(virq, cpu);
 968         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 969         if (retval != 0) {
 970                 unbind_from_irq(irq);
 971                 return retval;
 972         }
 973
 974         return irq;
 975 }
 976 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 977
 978 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 979                            unsigned int cpu,
 980                            irq_handler_t handler,
 981                            unsigned long irqflags,
 982                            const char *devname,
 983                            void *dev_id)
 984 {
 985         int irq, retval;
 986
 987         irq = bind_ipi_to_irq(ipi, cpu);
 988         if (irq < 0)
 989                 return irq;
 990
 991         irqflags |= IRQF_NO_SUSPEND;
 992         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 993         if (retval != 0) {
 994                 unbind_from_irq(irq);
 995                 return retval;
 996         }
 997
 998         return irq;
 999 }
1000
1001 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1002 {
1003         free_irq(irq, dev_id);
1004         unbind_from_irq(irq);
1005 }
1006 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1007
1008 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1009 {
1010         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1011         BUG_ON(irq < 0);
1012         notify_remote_via_irq(irq);
1013 }
1014
1015 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1016 {
1017         struct shared_info *sh = HYPERVISOR_shared_info;
1018         int cpu = smp_processor_id();
1019         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1020         int i;
1021         unsigned long flags;
1022         static DEFINE_SPINLOCK(debug_lock);
1023         struct vcpu_info *v;
1024
1025         spin_lock_irqsave(&debug_lock, flags);
1026
1027         printk("\nvcpu %d\n  ", cpu);
1028
1029         for_each_online_cpu(i) {
1030                 int pending;
1031                 v = per_cpu(xen_vcpu, i);
1032                 pending = (get_irq_regs() && i == cpu)
1033                         ? xen_irqs_disabled(get_irq_regs())
1034                         : v->evtchn_upcall_mask;
1035                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1036                        pending, v->evtchn_upcall_pending,
1037                        (int)(sizeof(v->evtchn_pending_sel)*2),
1038                        v->evtchn_pending_sel);
1039         }
1040         v = per_cpu(xen_vcpu, cpu);
1041
1042         printk("\npending:\n   ");
1043         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1044                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1045                        sh->evtchn_pending[i],
1046                        i % 8 == 0 ? "\n   " : " ");
1047         printk("\nglobal mask:\n   ");
1048         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1049                 printk("%0*lx%s",
1050                        (int)(sizeof(sh->evtchn_mask[0])*2),
1051                        sh->evtchn_mask[i],
1052                        i % 8 == 0 ? "\n   " : " ");
1053
1054         printk("\nglobally unmasked:\n   ");
1055         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1056                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1057                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1058                        i % 8 == 0 ? "\n   " : " ");
1059
1060         printk("\nlocal cpu%d mask:\n   ", cpu);
1061         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1062                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1063                        cpu_evtchn[i],
1064                        i % 8 == 0 ? "\n   " : " ");
1065
1066         printk("\nlocally unmasked:\n   ");
1067         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1068                 unsigned long pending = sh->evtchn_pending[i]
1069                         & ~sh->evtchn_mask[i]
1070                         & cpu_evtchn[i];
1071                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1072                        pending, i % 8 == 0 ? "\n   " : " ");
1073         }
1074
1075         printk("\npending list:\n");
1076         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1077                 if (sync_test_bit(i, sh->evtchn_pending)) {
1078                         int word_idx = i / BITS_PER_LONG;
1079                         printk("  %d: event %d -> irq %d%s%s%s\n",
1080                                cpu_from_evtchn(i), i,
1081                                evtchn_to_irq[i],
1082                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1083                                              ? "" : " l2-clear",
1084                                !sync_test_bit(i, sh->evtchn_mask)
1085                                              ? "" : " globally-masked",
1086                                sync_test_bit(i, cpu_evtchn)
1087                                              ? "" : " locally-masked");
1088                 }
1089         }
1090
1091         spin_unlock_irqrestore(&debug_lock, flags);
1092
1093         return IRQ_HANDLED;
1094 }
1095
1096 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1097
1098 /*
1099  * Search the CPUs pending events bitmasks.  For each one found, map
1100  * the event number to an irq, and feed it into do_IRQ() for
1101  * handling.
1102  *
1103  * Xen uses a two-level bitmap to speed searching.  The first level is
1104  * a bitset of words which contain pending event bits.  The second
1105  * level is a bitset of pending events themselves.
1106  */
1107 static void __xen_evtchn_do_upcall(void)
1108 {
1109         int cpu = get_cpu();
1110         struct shared_info *s = HYPERVISOR_shared_info;
1111         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1112         unsigned count;
1113
1114         do {
1115                 unsigned long pending_words;
1116
1117                 vcpu_info->evtchn_upcall_pending = 0;
1118
1119                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1120                         goto out;
1121
1122 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1123                 /* Clear master flag /before/ clearing selector flag. */
1124                 wmb();
1125 #endif
1126                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1127                 while (pending_words != 0) {
1128                         unsigned long pending_bits;
1129                         int word_idx = __ffs(pending_words);
1130                         pending_words &= ~(1UL << word_idx);
1131
1132                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1133                                 int bit_idx = __ffs(pending_bits);
1134                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1135                                 int irq = evtchn_to_irq[port];
1136                                 struct irq_desc *desc;
1137
1138                                 mask_evtchn(port);
1139                                 clear_evtchn(port);
1140
1141                                 if (irq != -1) {
1142                                         desc = irq_to_desc(irq);
1143                                         if (desc)
1144                                                 generic_handle_irq_desc(irq, desc);
1145                                 }
1146                         }
1147                 }
1148
1149                 BUG_ON(!irqs_disabled());
1150
1151                 count = __this_cpu_read(xed_nesting_count);
1152                 __this_cpu_write(xed_nesting_count, 0);
1153         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1154
1155 out:
1156
1157         put_cpu();
1158 }
1159
1160 void xen_evtchn_do_upcall(struct pt_regs *regs)
1161 {
1162         struct pt_regs *old_regs = set_irq_regs(regs);
1163
1164         exit_idle();
1165         irq_enter();
1166
1167         __xen_evtchn_do_upcall();
1168
1169         irq_exit();
1170         set_irq_regs(old_regs);
1171 }
1172
1173 void xen_hvm_evtchn_do_upcall(void)
1174 {
1175         __xen_evtchn_do_upcall();
1176 }
1177 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1178
1179 /* Rebind a new event channel to an existing irq. */
1180 void rebind_evtchn_irq(int evtchn, int irq)
1181 {
1182         struct irq_info *info = info_for_irq(irq);
1183
1184         /* Make sure the irq is masked, since the new event channel
1185            will also be masked. */
1186         disable_irq(irq);
1187
1188         spin_lock(&irq_mapping_update_lock);
1189
1190         /* After resume the irq<->evtchn mappings are all cleared out */
1191         BUG_ON(evtchn_to_irq[evtchn] != -1);
1192         /* Expect irq to have been bound before,
1193            so there should be a proper type */
1194         BUG_ON(info->type == IRQT_UNBOUND);
1195
1196         evtchn_to_irq[evtchn] = irq;
1197         irq_info[irq] = mk_evtchn_info(evtchn);
1198
1199         spin_unlock(&irq_mapping_update_lock);
1200
1201         /* new event channels are always bound to cpu 0 */
1202         irq_set_affinity(irq, cpumask_of(0));
1203
1204         /* Unmask the event channel. */
1205         enable_irq(irq);
1206 }
1207
1208 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1209 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1210 {
1211         struct evtchn_bind_vcpu bind_vcpu;
1212         int evtchn = evtchn_from_irq(irq);
1213
1214         /* events delivered via platform PCI interrupts are always
1215          * routed to vcpu 0 */
1216         if (!VALID_EVTCHN(evtchn) ||
1217                 (xen_hvm_domain() && !xen_have_vector_callback))
1218                 return -1;
1219
1220         /* Send future instances of this interrupt to other vcpu. */
1221         bind_vcpu.port = evtchn;
1222         bind_vcpu.vcpu = tcpu;
1223
1224         /*
1225          * If this fails, it usually just indicates that we're dealing with a
1226          * virq or IPI channel, which don't actually need to be rebound. Ignore
1227          * it, but don't do the xenlinux-level rebind in that case.
1228          */
1229         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1230                 bind_evtchn_to_cpu(evtchn, tcpu);
1231
1232         return 0;
1233 }
1234
1235 static int set_affinity_irq(unsigned irq, const struct cpumask *dest)
1236 {
1237         unsigned tcpu = cpumask_first(dest);
1238
1239         return rebind_irq_to_cpu(irq, tcpu);
1240 }
1241
1242 int resend_irq_on_evtchn(unsigned int irq)
1243 {
1244         int masked, evtchn = evtchn_from_irq(irq);
1245         struct shared_info *s = HYPERVISOR_shared_info;
1246
1247         if (!VALID_EVTCHN(evtchn))
1248                 return 1;
1249
1250         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1251         sync_set_bit(evtchn, s->evtchn_pending);
1252         if (!masked)
1253                 unmask_evtchn(evtchn);
1254
1255         return 1;
1256 }
1257
1258 static void enable_dynirq(unsigned int irq)
1259 {
1260         int evtchn = evtchn_from_irq(irq);
1261
1262         if (VALID_EVTCHN(evtchn))
1263                 unmask_evtchn(evtchn);
1264 }
1265
1266 static void disable_dynirq(unsigned int irq)
1267 {
1268         int evtchn = evtchn_from_irq(irq);
1269
1270         if (VALID_EVTCHN(evtchn))
1271                 mask_evtchn(evtchn);
1272 }
1273
1274 static void ack_dynirq(unsigned int irq)
1275 {
1276         int evtchn = evtchn_from_irq(irq);
1277
1278         move_masked_irq(irq);
1279
1280         if (VALID_EVTCHN(evtchn))
1281                 unmask_evtchn(evtchn);
1282 }
1283
1284 static int retrigger_dynirq(unsigned int irq)
1285 {
1286         int evtchn = evtchn_from_irq(irq);
1287         struct shared_info *sh = HYPERVISOR_shared_info;
1288         int ret = 0;
1289
1290         if (VALID_EVTCHN(evtchn)) {
1291                 int masked;
1292
1293                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1294                 sync_set_bit(evtchn, sh->evtchn_pending);
1295                 if (!masked)
1296                         unmask_evtchn(evtchn);
1297                 ret = 1;
1298         }
1299
1300         return ret;
1301 }
1302
1303 static void restore_cpu_pirqs(void)
1304 {
1305         int pirq, rc, irq, gsi;
1306         struct physdev_map_pirq map_irq;
1307
1308         for (pirq = 0; pirq < nr_irqs; pirq++) {
1309                 irq = pirq_to_irq[pirq];
1310                 if (irq == -1)
1311                         continue;
1312
1313                 /* save/restore of PT devices doesn't work, so at this point the
1314                  * only devices present are GSI based emulated devices */
1315                 gsi = gsi_from_irq(irq);
1316                 if (!gsi)
1317                         continue;
1318
1319                 map_irq.domid = DOMID_SELF;
1320                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1321                 map_irq.index = gsi;
1322                 map_irq.pirq = pirq;
1323
1324                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1325                 if (rc) {
1326                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1327                                         gsi, irq, pirq, rc);
1328                         irq_info[irq] = mk_unbound_info();
1329                         pirq_to_irq[pirq] = -1;
1330                         continue;
1331                 }
1332
1333                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1334
1335                 startup_pirq(irq);
1336         }
1337 }
1338
1339 static void restore_cpu_virqs(unsigned int cpu)
1340 {
1341         struct evtchn_bind_virq bind_virq;
1342         int virq, irq, evtchn;
1343
1344         for (virq = 0; virq < NR_VIRQS; virq++) {
1345                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1346                         continue;
1347
1348                 BUG_ON(virq_from_irq(irq) != virq);
1349
1350                 /* Get a new binding from Xen. */
1351                 bind_virq.virq = virq;
1352                 bind_virq.vcpu = cpu;
1353                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1354                                                 &bind_virq) != 0)
1355                         BUG();
1356                 evtchn = bind_virq.port;
1357
1358                 /* Record the new mapping. */
1359                 evtchn_to_irq[evtchn] = irq;
1360                 irq_info[irq] = mk_virq_info(evtchn, virq);
1361                 bind_evtchn_to_cpu(evtchn, cpu);
1362         }
1363 }
1364
1365 static void restore_cpu_ipis(unsigned int cpu)
1366 {
1367         struct evtchn_bind_ipi bind_ipi;
1368         int ipi, irq, evtchn;
1369
1370         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1371                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1372                         continue;
1373
1374                 BUG_ON(ipi_from_irq(irq) != ipi);
1375
1376                 /* Get a new binding from Xen. */
1377                 bind_ipi.vcpu = cpu;
1378                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1379                                                 &bind_ipi) != 0)
1380                         BUG();
1381                 evtchn = bind_ipi.port;
1382
1383                 /* Record the new mapping. */
1384                 evtchn_to_irq[evtchn] = irq;
1385                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1386                 bind_evtchn_to_cpu(evtchn, cpu);
1387         }
1388 }
1389
1390 /* Clear an irq's pending state, in preparation for polling on it */
1391 void xen_clear_irq_pending(int irq)
1392 {
1393         int evtchn = evtchn_from_irq(irq);
1394
1395         if (VALID_EVTCHN(evtchn))
1396                 clear_evtchn(evtchn);
1397 }
1398 EXPORT_SYMBOL(xen_clear_irq_pending);
1399 void xen_set_irq_pending(int irq)
1400 {
1401         int evtchn = evtchn_from_irq(irq);
1402
1403         if (VALID_EVTCHN(evtchn))
1404                 set_evtchn(evtchn);
1405 }
1406
1407 bool xen_test_irq_pending(int irq)
1408 {
1409         int evtchn = evtchn_from_irq(irq);
1410         bool ret = false;
1411
1412         if (VALID_EVTCHN(evtchn))
1413                 ret = test_evtchn(evtchn);
1414
1415         return ret;
1416 }
1417
1418 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1419  * the irq will be disabled so it won't deliver an interrupt. */
1420 void xen_poll_irq_timeout(int irq, u64 timeout)
1421 {
1422         evtchn_port_t evtchn = evtchn_from_irq(irq);
1423
1424         if (VALID_EVTCHN(evtchn)) {
1425                 struct sched_poll poll;
1426
1427                 poll.nr_ports = 1;
1428                 poll.timeout = timeout;
1429                 set_xen_guest_handle(poll.ports, &evtchn);
1430
1431                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1432                         BUG();
1433         }
1434 }
1435 EXPORT_SYMBOL(xen_poll_irq_timeout);
1436 /* Poll waiting for an irq to become pending.  In the usual case, the
1437  * irq will be disabled so it won't deliver an interrupt. */
1438 void xen_poll_irq(int irq)
1439 {
1440         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1441 }
1442
1443 void xen_irq_resume(void)
1444 {
1445         unsigned int cpu, irq, evtchn;
1446         struct irq_desc *desc;
1447
1448         init_evtchn_cpu_bindings();
1449
1450         /* New event-channel space is not 'live' yet. */
1451         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1452                 mask_evtchn(evtchn);
1453
1454         /* No IRQ <-> event-channel mappings. */
1455         for (irq = 0; irq < nr_irqs; irq++)
1456                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1457
1458         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1459                 evtchn_to_irq[evtchn] = -1;
1460
1461         for_each_possible_cpu(cpu) {
1462                 restore_cpu_virqs(cpu);
1463                 restore_cpu_ipis(cpu);
1464         }
1465
1466         /*
1467          * Unmask any IRQF_NO_SUSPEND IRQs which are enabled. These
1468          * are not handled by the IRQ core.
1469          */
1470         for_each_irq_desc(irq, desc) {
1471                 if (!desc->action || !(desc->action->flags & IRQF_NO_SUSPEND))
1472                         continue;
1473                 if (desc->status & IRQ_DISABLED)
1474                         continue;
1475
1476                 evtchn = evtchn_from_irq(irq);
1477                 if (evtchn == -1)
1478                         continue;
1479
1480                 unmask_evtchn(evtchn);
1481         }
1482
1483         restore_cpu_pirqs();
1484 }
1485
1486 static struct irq_chip xen_dynamic_chip __read_mostly = {
1487         .name           = "xen-dyn",
1488
1489         .disable        = disable_dynirq,
1490         .mask           = disable_dynirq,
1491         .unmask         = enable_dynirq,
1492
1493         .eoi            = ack_dynirq,
1494         .set_affinity   = set_affinity_irq,
1495         .retrigger      = retrigger_dynirq,
1496 };
1497
1498 static struct irq_chip xen_pirq_chip __read_mostly = {
1499         .name           = "xen-pirq",
1500
1501         .startup        = startup_pirq,
1502         .shutdown       = shutdown_pirq,
1503
1504         .enable         = enable_pirq,
1505         .unmask         = enable_pirq,
1506
1507         .disable        = disable_pirq,
1508         .mask           = disable_pirq,
1509
1510         .ack            = ack_pirq,
1511         .end            = end_pirq,
1512
1513         .set_affinity   = set_affinity_irq,
1514
1515         .retrigger      = retrigger_dynirq,
1516 };
1517
1518 static struct irq_chip xen_percpu_chip __read_mostly = {
1519         .name           = "xen-percpu",
1520
1521         .disable        = disable_dynirq,
1522         .mask           = disable_dynirq,
1523         .unmask         = enable_dynirq,
1524
1525         .ack            = ack_dynirq,
1526 };
1527
1528 int xen_set_callback_via(uint64_t via)
1529 {
1530         struct xen_hvm_param a;
1531         a.domid = DOMID_SELF;
1532         a.index = HVM_PARAM_CALLBACK_IRQ;
1533         a.value = via;
1534         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1535 }
1536 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1537
1538 #ifdef CONFIG_XEN_PVHVM
1539 /* Vector callbacks are better than PCI interrupts to receive event
1540  * channel notifications because we can receive vector callbacks on any
1541  * vcpu and we don't need PCI support or APIC interactions. */
1542 void xen_callback_vector(void)
1543 {
1544         int rc;
1545         uint64_t callback_via;
1546         if (xen_have_vector_callback) {
1547                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1548                 rc = xen_set_callback_via(callback_via);
1549                 if (rc) {
1550                         printk(KERN_ERR "Request for Xen HVM callback vector"
1551                                         " failed.\n");
1552                         xen_have_vector_callback = 0;
1553                         return;
1554                 }
1555                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1556                                 "enabled\n");
1557                 /* in the restore case the vector has already been allocated */
1558                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1559                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1560         }
1561 }
1562 #else
1563 void xen_callback_vector(void) {}
1564 #endif
1565
1566 void __init xen_init_IRQ(void)
1567 {
1568         int i;
1569
1570         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1571                                     GFP_KERNEL);
1572         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1573
1574         /* We are using nr_irqs as the maximum number of pirq available but
1575          * that number is actually chosen by Xen and we don't know exactly
1576          * what it is. Be careful choosing high pirq numbers. */
1577         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1578         for (i = 0; i < nr_irqs; i++)
1579                 pirq_to_irq[i] = -1;
1580
1581         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1582                                     GFP_KERNEL);
1583         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1584                 evtchn_to_irq[i] = -1;
1585
1586         init_evtchn_cpu_bindings();
1587
1588         /* No event channels are 'live' right now. */
1589         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1590                 mask_evtchn(i);
1591
1592         if (xen_hvm_domain()) {
1593                 xen_callback_vector();
1594                 native_init_IRQ();
1595                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1596                  * __acpi_register_gsi can point at the right function */
1597                 pci_xen_hvm_init();
1598         } else {
1599                 irq_ctx_init(smp_processor_id());
1600                 if (xen_initial_domain())
1601                         xen_setup_pirqs();
1602         }
1603 }