| 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 received, 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 | static LIST_HEAD(xen_irq_list_head); |
| 60 | |
| 61 | /* IRQ <-> VIRQ mapping. */ |
| 62 | static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1}; |
| 63 | |
| 64 | /* IRQ <-> IPI mapping */ |
| 65 | static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1}; |
| 66 | |
| 67 | /* Interrupt types. */ |
| 68 | enum xen_irq_type { |
| 69 | IRQT_UNBOUND = 0, |
| 70 | IRQT_PIRQ, |
| 71 | IRQT_VIRQ, |
| 72 | IRQT_IPI, |
| 73 | IRQT_EVTCHN |
| 74 | }; |
| 75 | |
| 76 | /* |
| 77 | * Packed IRQ information: |
| 78 | * type - enum xen_irq_type |
| 79 | * event channel - irq->event channel mapping |
| 80 | * cpu - cpu this event channel is bound to |
| 81 | * index - type-specific information: |
| 82 | * PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM |
| 83 | * guest, or GSI (real passthrough IRQ) of the device. |
| 84 | * VIRQ - virq number |
| 85 | * IPI - IPI vector |
| 86 | * EVTCHN - |
| 87 | */ |
| 88 | struct irq_info |
| 89 | { |
| 90 | struct list_head list; |
| 91 | enum xen_irq_type type; /* type */ |
| 92 | unsigned irq; |
| 93 | unsigned short evtchn; /* event channel */ |
| 94 | unsigned short cpu; /* cpu bound */ |
| 95 | |
| 96 | union { |
| 97 | unsigned short virq; |
| 98 | enum ipi_vector ipi; |
| 99 | struct { |
| 100 | unsigned short pirq; |
| 101 | unsigned short gsi; |
| 102 | unsigned char vector; |
| 103 | unsigned char flags; |
| 104 | } pirq; |
| 105 | } u; |
| 106 | }; |
| 107 | #define PIRQ_NEEDS_EOI (1 << 0) |
| 108 | #define PIRQ_SHAREABLE (1 << 1) |
| 109 | |
| 110 | static int *evtchn_to_irq; |
| 111 | |
| 112 | static DEFINE_PER_CPU(unsigned long [NR_EVENT_CHANNELS/BITS_PER_LONG], |
| 113 | cpu_evtchn_mask); |
| 114 | |
| 115 | /* Xen will never allocate port zero for any purpose. */ |
| 116 | #define VALID_EVTCHN(chn) ((chn) != 0) |
| 117 | |
| 118 | static struct irq_chip xen_dynamic_chip; |
| 119 | static struct irq_chip xen_percpu_chip; |
| 120 | static struct irq_chip xen_pirq_chip; |
| 121 | |
| 122 | /* Get info for IRQ */ |
| 123 | static struct irq_info *info_for_irq(unsigned irq) |
| 124 | { |
| 125 | return irq_get_handler_data(irq); |
| 126 | } |
| 127 | |
| 128 | /* Constructors for packed IRQ information. */ |
| 129 | static void xen_irq_info_common_init(struct irq_info *info, |
| 130 | unsigned irq, |
| 131 | enum xen_irq_type type, |
| 132 | unsigned short evtchn, |
| 133 | unsigned short cpu) |
| 134 | { |
| 135 | |
| 136 | BUG_ON(info->type != IRQT_UNBOUND && info->type != type); |
| 137 | |
| 138 | info->type = type; |
| 139 | info->irq = irq; |
| 140 | info->evtchn = evtchn; |
| 141 | info->cpu = cpu; |
| 142 | |
| 143 | evtchn_to_irq[evtchn] = irq; |
| 144 | } |
| 145 | |
| 146 | static void xen_irq_info_evtchn_init(unsigned irq, |
| 147 | unsigned short evtchn) |
| 148 | { |
| 149 | struct irq_info *info = info_for_irq(irq); |
| 150 | |
| 151 | xen_irq_info_common_init(info, irq, IRQT_EVTCHN, evtchn, 0); |
| 152 | } |
| 153 | |
| 154 | static void xen_irq_info_ipi_init(unsigned cpu, |
| 155 | unsigned irq, |
| 156 | unsigned short evtchn, |
| 157 | enum ipi_vector ipi) |
| 158 | { |
| 159 | struct irq_info *info = info_for_irq(irq); |
| 160 | |
| 161 | xen_irq_info_common_init(info, irq, IRQT_IPI, evtchn, 0); |
| 162 | |
| 163 | info->u.ipi = ipi; |
| 164 | |
| 165 | per_cpu(ipi_to_irq, cpu)[ipi] = irq; |
| 166 | } |
| 167 | |
| 168 | static void xen_irq_info_virq_init(unsigned cpu, |
| 169 | unsigned irq, |
| 170 | unsigned short evtchn, |
| 171 | unsigned short virq) |
| 172 | { |
| 173 | struct irq_info *info = info_for_irq(irq); |
| 174 | |
| 175 | xen_irq_info_common_init(info, irq, IRQT_VIRQ, evtchn, 0); |
| 176 | |
| 177 | info->u.virq = virq; |
| 178 | |
| 179 | per_cpu(virq_to_irq, cpu)[virq] = irq; |
| 180 | } |
| 181 | |
| 182 | static void xen_irq_info_pirq_init(unsigned irq, |
| 183 | unsigned short evtchn, |
| 184 | unsigned short pirq, |
| 185 | unsigned short gsi, |
| 186 | unsigned short vector, |
| 187 | unsigned char flags) |
| 188 | { |
| 189 | struct irq_info *info = info_for_irq(irq); |
| 190 | |
| 191 | xen_irq_info_common_init(info, irq, IRQT_PIRQ, evtchn, 0); |
| 192 | |
| 193 | info->u.pirq.pirq = pirq; |
| 194 | info->u.pirq.gsi = gsi; |
| 195 | info->u.pirq.vector = vector; |
| 196 | info->u.pirq.flags = flags; |
| 197 | } |
| 198 | |
| 199 | /* |
| 200 | * Accessors for packed IRQ information. |
| 201 | */ |
| 202 | static unsigned int evtchn_from_irq(unsigned irq) |
| 203 | { |
| 204 | if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq))) |
| 205 | return 0; |
| 206 | |
| 207 | return info_for_irq(irq)->evtchn; |
| 208 | } |
| 209 | |
| 210 | unsigned irq_from_evtchn(unsigned int evtchn) |
| 211 | { |
| 212 | return evtchn_to_irq[evtchn]; |
| 213 | } |
| 214 | EXPORT_SYMBOL_GPL(irq_from_evtchn); |
| 215 | |
| 216 | static enum ipi_vector ipi_from_irq(unsigned irq) |
| 217 | { |
| 218 | struct irq_info *info = info_for_irq(irq); |
| 219 | |
| 220 | BUG_ON(info == NULL); |
| 221 | BUG_ON(info->type != IRQT_IPI); |
| 222 | |
| 223 | return info->u.ipi; |
| 224 | } |
| 225 | |
| 226 | static unsigned virq_from_irq(unsigned irq) |
| 227 | { |
| 228 | struct irq_info *info = info_for_irq(irq); |
| 229 | |
| 230 | BUG_ON(info == NULL); |
| 231 | BUG_ON(info->type != IRQT_VIRQ); |
| 232 | |
| 233 | return info->u.virq; |
| 234 | } |
| 235 | |
| 236 | static unsigned pirq_from_irq(unsigned irq) |
| 237 | { |
| 238 | struct irq_info *info = info_for_irq(irq); |
| 239 | |
| 240 | BUG_ON(info == NULL); |
| 241 | BUG_ON(info->type != IRQT_PIRQ); |
| 242 | |
| 243 | return info->u.pirq.pirq; |
| 244 | } |
| 245 | |
| 246 | static enum xen_irq_type type_from_irq(unsigned irq) |
| 247 | { |
| 248 | return info_for_irq(irq)->type; |
| 249 | } |
| 250 | |
| 251 | static unsigned cpu_from_irq(unsigned irq) |
| 252 | { |
| 253 | return info_for_irq(irq)->cpu; |
| 254 | } |
| 255 | |
| 256 | static unsigned int cpu_from_evtchn(unsigned int evtchn) |
| 257 | { |
| 258 | int irq = evtchn_to_irq[evtchn]; |
| 259 | unsigned ret = 0; |
| 260 | |
| 261 | if (irq != -1) |
| 262 | ret = cpu_from_irq(irq); |
| 263 | |
| 264 | return ret; |
| 265 | } |
| 266 | |
| 267 | static bool pirq_needs_eoi(unsigned irq) |
| 268 | { |
| 269 | struct irq_info *info = info_for_irq(irq); |
| 270 | |
| 271 | BUG_ON(info->type != IRQT_PIRQ); |
| 272 | |
| 273 | return info->u.pirq.flags & PIRQ_NEEDS_EOI; |
| 274 | } |
| 275 | |
| 276 | static inline unsigned long active_evtchns(unsigned int cpu, |
| 277 | struct shared_info *sh, |
| 278 | unsigned int idx) |
| 279 | { |
| 280 | return (sh->evtchn_pending[idx] & |
| 281 | per_cpu(cpu_evtchn_mask, cpu)[idx] & |
| 282 | ~sh->evtchn_mask[idx]); |
| 283 | } |
| 284 | |
| 285 | static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu) |
| 286 | { |
| 287 | int irq = evtchn_to_irq[chn]; |
| 288 | |
| 289 | BUG_ON(irq == -1); |
| 290 | #ifdef CONFIG_SMP |
| 291 | cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu)); |
| 292 | #endif |
| 293 | |
| 294 | clear_bit(chn, per_cpu(cpu_evtchn_mask, cpu_from_irq(irq))); |
| 295 | set_bit(chn, per_cpu(cpu_evtchn_mask, cpu)); |
| 296 | |
| 297 | info_for_irq(irq)->cpu = cpu; |
| 298 | } |
| 299 | |
| 300 | static void init_evtchn_cpu_bindings(void) |
| 301 | { |
| 302 | int i; |
| 303 | #ifdef CONFIG_SMP |
| 304 | struct irq_info *info; |
| 305 | |
| 306 | /* By default all event channels notify CPU#0. */ |
| 307 | list_for_each_entry(info, &xen_irq_list_head, list) { |
| 308 | struct irq_desc *desc = irq_to_desc(info->irq); |
| 309 | cpumask_copy(desc->irq_data.affinity, cpumask_of(0)); |
| 310 | } |
| 311 | #endif |
| 312 | |
| 313 | for_each_possible_cpu(i) |
| 314 | memset(per_cpu(cpu_evtchn_mask, i), |
| 315 | (i == 0) ? ~0 : 0, sizeof(*per_cpu(cpu_evtchn_mask, i))); |
| 316 | } |
| 317 | |
| 318 | static inline void clear_evtchn(int port) |
| 319 | { |
| 320 | struct shared_info *s = HYPERVISOR_shared_info; |
| 321 | sync_clear_bit(port, &s->evtchn_pending[0]); |
| 322 | } |
| 323 | |
| 324 | static inline void set_evtchn(int port) |
| 325 | { |
| 326 | struct shared_info *s = HYPERVISOR_shared_info; |
| 327 | sync_set_bit(port, &s->evtchn_pending[0]); |
| 328 | } |
| 329 | |
| 330 | static inline int test_evtchn(int port) |
| 331 | { |
| 332 | struct shared_info *s = HYPERVISOR_shared_info; |
| 333 | return sync_test_bit(port, &s->evtchn_pending[0]); |
| 334 | } |
| 335 | |
| 336 | |
| 337 | /** |
| 338 | * notify_remote_via_irq - send event to remote end of event channel via irq |
| 339 | * @irq: irq of event channel to send event to |
| 340 | * |
| 341 | * Unlike notify_remote_via_evtchn(), this is safe to use across |
| 342 | * save/restore. Notifications on a broken connection are silently |
| 343 | * dropped. |
| 344 | */ |
| 345 | void notify_remote_via_irq(int irq) |
| 346 | { |
| 347 | int evtchn = evtchn_from_irq(irq); |
| 348 | |
| 349 | if (VALID_EVTCHN(evtchn)) |
| 350 | notify_remote_via_evtchn(evtchn); |
| 351 | } |
| 352 | EXPORT_SYMBOL_GPL(notify_remote_via_irq); |
| 353 | |
| 354 | static void mask_evtchn(int port) |
| 355 | { |
| 356 | struct shared_info *s = HYPERVISOR_shared_info; |
| 357 | sync_set_bit(port, &s->evtchn_mask[0]); |
| 358 | } |
| 359 | |
| 360 | static void unmask_evtchn(int port) |
| 361 | { |
| 362 | struct shared_info *s = HYPERVISOR_shared_info; |
| 363 | unsigned int cpu = get_cpu(); |
| 364 | |
| 365 | BUG_ON(!irqs_disabled()); |
| 366 | |
| 367 | /* Slow path (hypercall) if this is a non-local port. */ |
| 368 | if (unlikely(cpu != cpu_from_evtchn(port))) { |
| 369 | struct evtchn_unmask unmask = { .port = port }; |
| 370 | (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask); |
| 371 | } else { |
| 372 | struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu); |
| 373 | |
| 374 | sync_clear_bit(port, &s->evtchn_mask[0]); |
| 375 | |
| 376 | /* |
| 377 | * The following is basically the equivalent of |
| 378 | * 'hw_resend_irq'. Just like a real IO-APIC we 'lose |
| 379 | * the interrupt edge' if the channel is masked. |
| 380 | */ |
| 381 | if (sync_test_bit(port, &s->evtchn_pending[0]) && |
| 382 | !sync_test_and_set_bit(port / BITS_PER_LONG, |
| 383 | &vcpu_info->evtchn_pending_sel)) |
| 384 | vcpu_info->evtchn_upcall_pending = 1; |
| 385 | } |
| 386 | |
| 387 | put_cpu(); |
| 388 | } |
| 389 | |
| 390 | static void xen_irq_init(unsigned irq) |
| 391 | { |
| 392 | struct irq_info *info; |
| 393 | struct irq_desc *desc = irq_to_desc(irq); |
| 394 | |
| 395 | #ifdef CONFIG_SMP |
| 396 | /* By default all event channels notify CPU#0. */ |
| 397 | cpumask_copy(desc->irq_data.affinity, cpumask_of(0)); |
| 398 | #endif |
| 399 | |
| 400 | info = kzalloc(sizeof(*info), GFP_KERNEL); |
| 401 | if (info == NULL) |
| 402 | panic("Unable to allocate metadata for IRQ%d\n", irq); |
| 403 | |
| 404 | info->type = IRQT_UNBOUND; |
| 405 | |
| 406 | irq_set_handler_data(irq, info); |
| 407 | |
| 408 | list_add_tail(&info->list, &xen_irq_list_head); |
| 409 | } |
| 410 | |
| 411 | static int __must_check xen_allocate_irq_dynamic(void) |
| 412 | { |
| 413 | int first = 0; |
| 414 | int irq; |
| 415 | |
| 416 | #ifdef CONFIG_X86_IO_APIC |
| 417 | /* |
| 418 | * For an HVM guest or domain 0 which see "real" (emulated or |
| 419 | * actual respectively) GSIs we allocate dynamic IRQs |
| 420 | * e.g. those corresponding to event channels or MSIs |
| 421 | * etc. from the range above those "real" GSIs to avoid |
| 422 | * collisions. |
| 423 | */ |
| 424 | if (xen_initial_domain() || xen_hvm_domain()) |
| 425 | first = get_nr_irqs_gsi(); |
| 426 | #endif |
| 427 | |
| 428 | irq = irq_alloc_desc_from(first, -1); |
| 429 | |
| 430 | xen_irq_init(irq); |
| 431 | |
| 432 | return irq; |
| 433 | } |
| 434 | |
| 435 | static int __must_check xen_allocate_irq_gsi(unsigned gsi) |
| 436 | { |
| 437 | int irq; |
| 438 | |
| 439 | /* |
| 440 | * A PV guest has no concept of a GSI (since it has no ACPI |
| 441 | * nor access to/knowledge of the physical APICs). Therefore |
| 442 | * all IRQs are dynamically allocated from the entire IRQ |
| 443 | * space. |
| 444 | */ |
| 445 | if (xen_pv_domain() && !xen_initial_domain()) |
| 446 | return xen_allocate_irq_dynamic(); |
| 447 | |
| 448 | /* Legacy IRQ descriptors are already allocated by the arch. */ |
| 449 | if (gsi < NR_IRQS_LEGACY) |
| 450 | irq = gsi; |
| 451 | else |
| 452 | irq = irq_alloc_desc_at(gsi, -1); |
| 453 | |
| 454 | xen_irq_init(irq); |
| 455 | |
| 456 | return irq; |
| 457 | } |
| 458 | |
| 459 | static void xen_free_irq(unsigned irq) |
| 460 | { |
| 461 | struct irq_info *info = irq_get_handler_data(irq); |
| 462 | |
| 463 | list_del(&info->list); |
| 464 | |
| 465 | irq_set_handler_data(irq, NULL); |
| 466 | |
| 467 | kfree(info); |
| 468 | |
| 469 | /* Legacy IRQ descriptors are managed by the arch. */ |
| 470 | if (irq < NR_IRQS_LEGACY) |
| 471 | return; |
| 472 | |
| 473 | irq_free_desc(irq); |
| 474 | } |
| 475 | |
| 476 | static void pirq_unmask_notify(int irq) |
| 477 | { |
| 478 | struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) }; |
| 479 | |
| 480 | if (unlikely(pirq_needs_eoi(irq))) { |
| 481 | int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi); |
| 482 | WARN_ON(rc); |
| 483 | } |
| 484 | } |
| 485 | |
| 486 | static void pirq_query_unmask(int irq) |
| 487 | { |
| 488 | struct physdev_irq_status_query irq_status; |
| 489 | struct irq_info *info = info_for_irq(irq); |
| 490 | |
| 491 | BUG_ON(info->type != IRQT_PIRQ); |
| 492 | |
| 493 | irq_status.irq = pirq_from_irq(irq); |
| 494 | if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status)) |
| 495 | irq_status.flags = 0; |
| 496 | |
| 497 | info->u.pirq.flags &= ~PIRQ_NEEDS_EOI; |
| 498 | if (irq_status.flags & XENIRQSTAT_needs_eoi) |
| 499 | info->u.pirq.flags |= PIRQ_NEEDS_EOI; |
| 500 | } |
| 501 | |
| 502 | static bool probing_irq(int irq) |
| 503 | { |
| 504 | struct irq_desc *desc = irq_to_desc(irq); |
| 505 | |
| 506 | return desc && desc->action == NULL; |
| 507 | } |
| 508 | |
| 509 | static unsigned int __startup_pirq(unsigned int irq) |
| 510 | { |
| 511 | struct evtchn_bind_pirq bind_pirq; |
| 512 | struct irq_info *info = info_for_irq(irq); |
| 513 | int evtchn = evtchn_from_irq(irq); |
| 514 | int rc; |
| 515 | |
| 516 | BUG_ON(info->type != IRQT_PIRQ); |
| 517 | |
| 518 | if (VALID_EVTCHN(evtchn)) |
| 519 | goto out; |
| 520 | |
| 521 | bind_pirq.pirq = pirq_from_irq(irq); |
| 522 | /* NB. We are happy to share unless we are probing. */ |
| 523 | bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ? |
| 524 | BIND_PIRQ__WILL_SHARE : 0; |
| 525 | rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq); |
| 526 | if (rc != 0) { |
| 527 | if (!probing_irq(irq)) |
| 528 | printk(KERN_INFO "Failed to obtain physical IRQ %d\n", |
| 529 | irq); |
| 530 | return 0; |
| 531 | } |
| 532 | evtchn = bind_pirq.port; |
| 533 | |
| 534 | pirq_query_unmask(irq); |
| 535 | |
| 536 | evtchn_to_irq[evtchn] = irq; |
| 537 | bind_evtchn_to_cpu(evtchn, 0); |
| 538 | info->evtchn = evtchn; |
| 539 | |
| 540 | out: |
| 541 | unmask_evtchn(evtchn); |
| 542 | pirq_unmask_notify(irq); |
| 543 | |
| 544 | return 0; |
| 545 | } |
| 546 | |
| 547 | static unsigned int startup_pirq(struct irq_data *data) |
| 548 | { |
| 549 | return __startup_pirq(data->irq); |
| 550 | } |
| 551 | |
| 552 | static void shutdown_pirq(struct irq_data *data) |
| 553 | { |
| 554 | struct evtchn_close close; |
| 555 | unsigned int irq = data->irq; |
| 556 | struct irq_info *info = info_for_irq(irq); |
| 557 | int evtchn = evtchn_from_irq(irq); |
| 558 | |
| 559 | BUG_ON(info->type != IRQT_PIRQ); |
| 560 | |
| 561 | if (!VALID_EVTCHN(evtchn)) |
| 562 | return; |
| 563 | |
| 564 | mask_evtchn(evtchn); |
| 565 | |
| 566 | close.port = evtchn; |
| 567 | if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0) |
| 568 | BUG(); |
| 569 | |
| 570 | bind_evtchn_to_cpu(evtchn, 0); |
| 571 | evtchn_to_irq[evtchn] = -1; |
| 572 | info->evtchn = 0; |
| 573 | } |
| 574 | |
| 575 | static void enable_pirq(struct irq_data *data) |
| 576 | { |
| 577 | startup_pirq(data); |
| 578 | } |
| 579 | |
| 580 | static void disable_pirq(struct irq_data *data) |
| 581 | { |
| 582 | } |
| 583 | |
| 584 | static void ack_pirq(struct irq_data *data) |
| 585 | { |
| 586 | int evtchn = evtchn_from_irq(data->irq); |
| 587 | |
| 588 | irq_move_irq(data); |
| 589 | |
| 590 | if (VALID_EVTCHN(evtchn)) { |
| 591 | mask_evtchn(evtchn); |
| 592 | clear_evtchn(evtchn); |
| 593 | } |
| 594 | } |
| 595 | |
| 596 | static int find_irq_by_gsi(unsigned gsi) |
| 597 | { |
| 598 | struct irq_info *info; |
| 599 | |
| 600 | list_for_each_entry(info, &xen_irq_list_head, list) { |
| 601 | if (info->type != IRQT_PIRQ) |
| 602 | continue; |
| 603 | |
| 604 | if (info->u.pirq.gsi == gsi) |
| 605 | return info->irq; |
| 606 | } |
| 607 | |
| 608 | return -1; |
| 609 | } |
| 610 | |
| 611 | int xen_allocate_pirq_gsi(unsigned gsi) |
| 612 | { |
| 613 | return gsi; |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * Do not make any assumptions regarding the relationship between the |
| 618 | * IRQ number returned here and the Xen pirq argument. |
| 619 | * |
| 620 | * Note: We don't assign an event channel until the irq actually started |
| 621 | * up. Return an existing irq if we've already got one for the gsi. |
| 622 | */ |
| 623 | int xen_bind_pirq_gsi_to_irq(unsigned gsi, |
| 624 | unsigned pirq, int shareable, char *name) |
| 625 | { |
| 626 | int irq = -1; |
| 627 | struct physdev_irq irq_op; |
| 628 | |
| 629 | spin_lock(&irq_mapping_update_lock); |
| 630 | |
| 631 | irq = find_irq_by_gsi(gsi); |
| 632 | if (irq != -1) { |
| 633 | printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n", |
| 634 | irq, gsi); |
| 635 | goto out; /* XXX need refcount? */ |
| 636 | } |
| 637 | |
| 638 | irq = xen_allocate_irq_gsi(gsi); |
| 639 | if (irq < 0) |
| 640 | goto out; |
| 641 | |
| 642 | irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_level_irq, |
| 643 | name); |
| 644 | |
| 645 | irq_op.irq = irq; |
| 646 | irq_op.vector = 0; |
| 647 | |
| 648 | /* Only the privileged domain can do this. For non-priv, the pcifront |
| 649 | * driver provides a PCI bus that does the call to do exactly |
| 650 | * this in the priv domain. */ |
| 651 | if (xen_initial_domain() && |
| 652 | HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) { |
| 653 | xen_free_irq(irq); |
| 654 | irq = -ENOSPC; |
| 655 | goto out; |
| 656 | } |
| 657 | |
| 658 | xen_irq_info_pirq_init(irq, 0, pirq, gsi, irq_op.vector, |
| 659 | shareable ? PIRQ_SHAREABLE : 0); |
| 660 | |
| 661 | out: |
| 662 | spin_unlock(&irq_mapping_update_lock); |
| 663 | |
| 664 | return irq; |
| 665 | } |
| 666 | |
| 667 | #ifdef CONFIG_PCI_MSI |
| 668 | int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc) |
| 669 | { |
| 670 | int rc; |
| 671 | struct physdev_get_free_pirq op_get_free_pirq; |
| 672 | |
| 673 | op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI; |
| 674 | rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq); |
| 675 | |
| 676 | WARN_ONCE(rc == -ENOSYS, |
| 677 | "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n"); |
| 678 | |
| 679 | return rc ? -1 : op_get_free_pirq.pirq; |
| 680 | } |
| 681 | |
| 682 | int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc, |
| 683 | int pirq, int vector, const char *name) |
| 684 | { |
| 685 | int irq, ret; |
| 686 | |
| 687 | spin_lock(&irq_mapping_update_lock); |
| 688 | |
| 689 | irq = xen_allocate_irq_dynamic(); |
| 690 | if (irq == -1) |
| 691 | goto out; |
| 692 | |
| 693 | irq_set_chip_and_handler_name(irq, &xen_pirq_chip, handle_level_irq, |
| 694 | name); |
| 695 | |
| 696 | xen_irq_info_pirq_init(irq, 0, pirq, 0, vector, 0); |
| 697 | ret = irq_set_msi_desc(irq, msidesc); |
| 698 | if (ret < 0) |
| 699 | goto error_irq; |
| 700 | out: |
| 701 | spin_unlock(&irq_mapping_update_lock); |
| 702 | return irq; |
| 703 | error_irq: |
| 704 | spin_unlock(&irq_mapping_update_lock); |
| 705 | xen_free_irq(irq); |
| 706 | return -1; |
| 707 | } |
| 708 | #endif |
| 709 | |
| 710 | int xen_destroy_irq(int irq) |
| 711 | { |
| 712 | struct irq_desc *desc; |
| 713 | struct physdev_unmap_pirq unmap_irq; |
| 714 | struct irq_info *info = info_for_irq(irq); |
| 715 | int rc = -ENOENT; |
| 716 | |
| 717 | spin_lock(&irq_mapping_update_lock); |
| 718 | |
| 719 | desc = irq_to_desc(irq); |
| 720 | if (!desc) |
| 721 | goto out; |
| 722 | |
| 723 | if (xen_initial_domain()) { |
| 724 | unmap_irq.pirq = info->u.pirq.pirq; |
| 725 | unmap_irq.domid = DOMID_SELF; |
| 726 | rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq); |
| 727 | if (rc) { |
| 728 | printk(KERN_WARNING "unmap irq failed %d\n", rc); |
| 729 | goto out; |
| 730 | } |
| 731 | } |
| 732 | |
| 733 | xen_free_irq(irq); |
| 734 | |
| 735 | out: |
| 736 | spin_unlock(&irq_mapping_update_lock); |
| 737 | return rc; |
| 738 | } |
| 739 | |
| 740 | int xen_irq_from_pirq(unsigned pirq) |
| 741 | { |
| 742 | int irq; |
| 743 | |
| 744 | struct irq_info *info; |
| 745 | |
| 746 | spin_lock(&irq_mapping_update_lock); |
| 747 | |
| 748 | list_for_each_entry(info, &xen_irq_list_head, list) { |
| 749 | if (info == NULL || info->type != IRQT_PIRQ) |
| 750 | continue; |
| 751 | irq = info->irq; |
| 752 | if (info->u.pirq.pirq == pirq) |
| 753 | goto out; |
| 754 | } |
| 755 | irq = -1; |
| 756 | out: |
| 757 | spin_unlock(&irq_mapping_update_lock); |
| 758 | |
| 759 | return irq; |
| 760 | } |
| 761 | |
| 762 | int bind_evtchn_to_irq(unsigned int evtchn) |
| 763 | { |
| 764 | int irq; |
| 765 | |
| 766 | spin_lock(&irq_mapping_update_lock); |
| 767 | |
| 768 | irq = evtchn_to_irq[evtchn]; |
| 769 | |
| 770 | if (irq == -1) { |
| 771 | irq = xen_allocate_irq_dynamic(); |
| 772 | if (irq == -1) |
| 773 | goto out; |
| 774 | |
| 775 | irq_set_chip_and_handler_name(irq, &xen_dynamic_chip, |
| 776 | handle_fasteoi_irq, "event"); |
| 777 | |
| 778 | xen_irq_info_evtchn_init(irq, evtchn); |
| 779 | } |
| 780 | |
| 781 | out: |
| 782 | spin_unlock(&irq_mapping_update_lock); |
| 783 | |
| 784 | return irq; |
| 785 | } |
| 786 | EXPORT_SYMBOL_GPL(bind_evtchn_to_irq); |
| 787 | |
| 788 | static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu) |
| 789 | { |
| 790 | struct evtchn_bind_ipi bind_ipi; |
| 791 | int evtchn, irq; |
| 792 | |
| 793 | spin_lock(&irq_mapping_update_lock); |
| 794 | |
| 795 | irq = per_cpu(ipi_to_irq, cpu)[ipi]; |
| 796 | |
| 797 | if (irq == -1) { |
| 798 | irq = xen_allocate_irq_dynamic(); |
| 799 | if (irq < 0) |
| 800 | goto out; |
| 801 | |
| 802 | irq_set_chip_and_handler_name(irq, &xen_percpu_chip, |
| 803 | handle_percpu_irq, "ipi"); |
| 804 | |
| 805 | bind_ipi.vcpu = cpu; |
| 806 | if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi, |
| 807 | &bind_ipi) != 0) |
| 808 | BUG(); |
| 809 | evtchn = bind_ipi.port; |
| 810 | |
| 811 | xen_irq_info_ipi_init(cpu, irq, evtchn, ipi); |
| 812 | |
| 813 | bind_evtchn_to_cpu(evtchn, cpu); |
| 814 | } |
| 815 | |
| 816 | out: |
| 817 | spin_unlock(&irq_mapping_update_lock); |
| 818 | return irq; |
| 819 | } |
| 820 | |
| 821 | static int bind_interdomain_evtchn_to_irq(unsigned int remote_domain, |
| 822 | unsigned int remote_port) |
| 823 | { |
| 824 | struct evtchn_bind_interdomain bind_interdomain; |
| 825 | int err; |
| 826 | |
| 827 | bind_interdomain.remote_dom = remote_domain; |
| 828 | bind_interdomain.remote_port = remote_port; |
| 829 | |
| 830 | err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain, |
| 831 | &bind_interdomain); |
| 832 | |
| 833 | return err ? : bind_evtchn_to_irq(bind_interdomain.local_port); |
| 834 | } |
| 835 | |
| 836 | |
| 837 | int bind_virq_to_irq(unsigned int virq, unsigned int cpu) |
| 838 | { |
| 839 | struct evtchn_bind_virq bind_virq; |
| 840 | int evtchn, irq; |
| 841 | |
| 842 | spin_lock(&irq_mapping_update_lock); |
| 843 | |
| 844 | irq = per_cpu(virq_to_irq, cpu)[virq]; |
| 845 | |
| 846 | if (irq == -1) { |
| 847 | irq = xen_allocate_irq_dynamic(); |
| 848 | if (irq == -1) |
| 849 | goto out; |
| 850 | |
| 851 | irq_set_chip_and_handler_name(irq, &xen_percpu_chip, |
| 852 | handle_percpu_irq, "virq"); |
| 853 | |
| 854 | bind_virq.virq = virq; |
| 855 | bind_virq.vcpu = cpu; |
| 856 | if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq, |
| 857 | &bind_virq) != 0) |
| 858 | BUG(); |
| 859 | evtchn = bind_virq.port; |
| 860 | |
| 861 | xen_irq_info_virq_init(cpu, irq, evtchn, virq); |
| 862 | |
| 863 | bind_evtchn_to_cpu(evtchn, cpu); |
| 864 | } |
| 865 | |
| 866 | out: |
| 867 | spin_unlock(&irq_mapping_update_lock); |
| 868 | |
| 869 | return irq; |
| 870 | } |
| 871 | |
| 872 | static void unbind_from_irq(unsigned int irq) |
| 873 | { |
| 874 | struct evtchn_close close; |
| 875 | int evtchn = evtchn_from_irq(irq); |
| 876 | |
| 877 | spin_lock(&irq_mapping_update_lock); |
| 878 | |
| 879 | if (VALID_EVTCHN(evtchn)) { |
| 880 | close.port = evtchn; |
| 881 | if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0) |
| 882 | BUG(); |
| 883 | |
| 884 | switch (type_from_irq(irq)) { |
| 885 | case IRQT_VIRQ: |
| 886 | per_cpu(virq_to_irq, cpu_from_evtchn(evtchn)) |
| 887 | [virq_from_irq(irq)] = -1; |
| 888 | break; |
| 889 | case IRQT_IPI: |
| 890 | per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn)) |
| 891 | [ipi_from_irq(irq)] = -1; |
| 892 | break; |
| 893 | default: |
| 894 | break; |
| 895 | } |
| 896 | |
| 897 | /* Closed ports are implicitly re-bound to VCPU0. */ |
| 898 | bind_evtchn_to_cpu(evtchn, 0); |
| 899 | |
| 900 | evtchn_to_irq[evtchn] = -1; |
| 901 | } |
| 902 | |
| 903 | BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND); |
| 904 | |
| 905 | xen_free_irq(irq); |
| 906 | |
| 907 | spin_unlock(&irq_mapping_update_lock); |
| 908 | } |
| 909 | |
| 910 | int bind_evtchn_to_irqhandler(unsigned int evtchn, |
| 911 | irq_handler_t handler, |
| 912 | unsigned long irqflags, |
| 913 | const char *devname, void *dev_id) |
| 914 | { |
| 915 | unsigned int irq; |
| 916 | int retval; |
| 917 | |
| 918 | irq = bind_evtchn_to_irq(evtchn); |
| 919 | if (irq < 0) |
| 920 | return irq; |
| 921 | retval = request_irq(irq, handler, irqflags, devname, dev_id); |
| 922 | if (retval != 0) { |
| 923 | unbind_from_irq(irq); |
| 924 | return retval; |
| 925 | } |
| 926 | |
| 927 | return irq; |
| 928 | } |
| 929 | EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler); |
| 930 | |
| 931 | int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain, |
| 932 | unsigned int remote_port, |
| 933 | irq_handler_t handler, |
| 934 | unsigned long irqflags, |
| 935 | const char *devname, |
| 936 | void *dev_id) |
| 937 | { |
| 938 | int irq, retval; |
| 939 | |
| 940 | irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port); |
| 941 | if (irq < 0) |
| 942 | return irq; |
| 943 | |
| 944 | retval = request_irq(irq, handler, irqflags, devname, dev_id); |
| 945 | if (retval != 0) { |
| 946 | unbind_from_irq(irq); |
| 947 | return retval; |
| 948 | } |
| 949 | |
| 950 | return irq; |
| 951 | } |
| 952 | EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler); |
| 953 | |
| 954 | int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu, |
| 955 | irq_handler_t handler, |
| 956 | unsigned long irqflags, const char *devname, void *dev_id) |
| 957 | { |
| 958 | unsigned int irq; |
| 959 | int retval; |
| 960 | |
| 961 | irq = bind_virq_to_irq(virq, cpu); |
| 962 | if (irq < 0) |
| 963 | return irq; |
| 964 | retval = request_irq(irq, handler, irqflags, devname, dev_id); |
| 965 | if (retval != 0) { |
| 966 | unbind_from_irq(irq); |
| 967 | return retval; |
| 968 | } |
| 969 | |
| 970 | return irq; |
| 971 | } |
| 972 | EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler); |
| 973 | |
| 974 | int bind_ipi_to_irqhandler(enum ipi_vector ipi, |
| 975 | unsigned int cpu, |
| 976 | irq_handler_t handler, |
| 977 | unsigned long irqflags, |
| 978 | const char *devname, |
| 979 | void *dev_id) |
| 980 | { |
| 981 | int irq, retval; |
| 982 | |
| 983 | irq = bind_ipi_to_irq(ipi, cpu); |
| 984 | if (irq < 0) |
| 985 | return irq; |
| 986 | |
| 987 | irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME; |
| 988 | retval = request_irq(irq, handler, irqflags, devname, dev_id); |
| 989 | if (retval != 0) { |
| 990 | unbind_from_irq(irq); |
| 991 | return retval; |
| 992 | } |
| 993 | |
| 994 | return irq; |
| 995 | } |
| 996 | |
| 997 | void unbind_from_irqhandler(unsigned int irq, void *dev_id) |
| 998 | { |
| 999 | free_irq(irq, dev_id); |
| 1000 | unbind_from_irq(irq); |
| 1001 | } |
| 1002 | EXPORT_SYMBOL_GPL(unbind_from_irqhandler); |
| 1003 | |
| 1004 | void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector) |
| 1005 | { |
| 1006 | int irq = per_cpu(ipi_to_irq, cpu)[vector]; |
| 1007 | BUG_ON(irq < 0); |
| 1008 | notify_remote_via_irq(irq); |
| 1009 | } |
| 1010 | |
| 1011 | irqreturn_t xen_debug_interrupt(int irq, void *dev_id) |
| 1012 | { |
| 1013 | struct shared_info *sh = HYPERVISOR_shared_info; |
| 1014 | int cpu = smp_processor_id(); |
| 1015 | unsigned long *cpu_evtchn = per_cpu(cpu_evtchn_mask, cpu); |
| 1016 | int i; |
| 1017 | unsigned long flags; |
| 1018 | static DEFINE_SPINLOCK(debug_lock); |
| 1019 | struct vcpu_info *v; |
| 1020 | |
| 1021 | spin_lock_irqsave(&debug_lock, flags); |
| 1022 | |
| 1023 | printk("\nvcpu %d\n ", cpu); |
| 1024 | |
| 1025 | for_each_online_cpu(i) { |
| 1026 | int pending; |
| 1027 | v = per_cpu(xen_vcpu, i); |
| 1028 | pending = (get_irq_regs() && i == cpu) |
| 1029 | ? xen_irqs_disabled(get_irq_regs()) |
| 1030 | : v->evtchn_upcall_mask; |
| 1031 | printk("%d: masked=%d pending=%d event_sel %0*lx\n ", i, |
| 1032 | pending, v->evtchn_upcall_pending, |
| 1033 | (int)(sizeof(v->evtchn_pending_sel)*2), |
| 1034 | v->evtchn_pending_sel); |
| 1035 | } |
| 1036 | v = per_cpu(xen_vcpu, cpu); |
| 1037 | |
| 1038 | printk("\npending:\n "); |
| 1039 | for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--) |
| 1040 | printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2, |
| 1041 | sh->evtchn_pending[i], |
| 1042 | i % 8 == 0 ? "\n " : " "); |
| 1043 | printk("\nglobal mask:\n "); |
| 1044 | for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) |
| 1045 | printk("%0*lx%s", |
| 1046 | (int)(sizeof(sh->evtchn_mask[0])*2), |
| 1047 | sh->evtchn_mask[i], |
| 1048 | i % 8 == 0 ? "\n " : " "); |
| 1049 | |
| 1050 | printk("\nglobally unmasked:\n "); |
| 1051 | for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) |
| 1052 | printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2), |
| 1053 | sh->evtchn_pending[i] & ~sh->evtchn_mask[i], |
| 1054 | i % 8 == 0 ? "\n " : " "); |
| 1055 | |
| 1056 | printk("\nlocal cpu%d mask:\n ", cpu); |
| 1057 | for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--) |
| 1058 | printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2), |
| 1059 | cpu_evtchn[i], |
| 1060 | i % 8 == 0 ? "\n " : " "); |
| 1061 | |
| 1062 | printk("\nlocally unmasked:\n "); |
| 1063 | for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) { |
| 1064 | unsigned long pending = sh->evtchn_pending[i] |
| 1065 | & ~sh->evtchn_mask[i] |
| 1066 | & cpu_evtchn[i]; |
| 1067 | printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2), |
| 1068 | pending, i % 8 == 0 ? "\n " : " "); |
| 1069 | } |
| 1070 | |
| 1071 | printk("\npending list:\n"); |
| 1072 | for (i = 0; i < NR_EVENT_CHANNELS; i++) { |
| 1073 | if (sync_test_bit(i, sh->evtchn_pending)) { |
| 1074 | int word_idx = i / BITS_PER_LONG; |
| 1075 | printk(" %d: event %d -> irq %d%s%s%s\n", |
| 1076 | cpu_from_evtchn(i), i, |
| 1077 | evtchn_to_irq[i], |
| 1078 | sync_test_bit(word_idx, &v->evtchn_pending_sel) |
| 1079 | ? "" : " l2-clear", |
| 1080 | !sync_test_bit(i, sh->evtchn_mask) |
| 1081 | ? "" : " globally-masked", |
| 1082 | sync_test_bit(i, cpu_evtchn) |
| 1083 | ? "" : " locally-masked"); |
| 1084 | } |
| 1085 | } |
| 1086 | |
| 1087 | spin_unlock_irqrestore(&debug_lock, flags); |
| 1088 | |
| 1089 | return IRQ_HANDLED; |
| 1090 | } |
| 1091 | |
| 1092 | static DEFINE_PER_CPU(unsigned, xed_nesting_count); |
| 1093 | static DEFINE_PER_CPU(unsigned int, current_word_idx); |
| 1094 | static DEFINE_PER_CPU(unsigned int, current_bit_idx); |
| 1095 | |
| 1096 | /* |
| 1097 | * Mask out the i least significant bits of w |
| 1098 | */ |
| 1099 | #define MASK_LSBS(w, i) (w & ((~0UL) << i)) |
| 1100 | |
| 1101 | /* |
| 1102 | * Search the CPUs pending events bitmasks. For each one found, map |
| 1103 | * the event number to an irq, and feed it into do_IRQ() for |
| 1104 | * handling. |
| 1105 | * |
| 1106 | * Xen uses a two-level bitmap to speed searching. The first level is |
| 1107 | * a bitset of words which contain pending event bits. The second |
| 1108 | * level is a bitset of pending events themselves. |
| 1109 | */ |
| 1110 | static void __xen_evtchn_do_upcall(void) |
| 1111 | { |
| 1112 | int start_word_idx, start_bit_idx; |
| 1113 | int word_idx, bit_idx; |
| 1114 | int i; |
| 1115 | int cpu = get_cpu(); |
| 1116 | struct shared_info *s = HYPERVISOR_shared_info; |
| 1117 | struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu); |
| 1118 | unsigned count; |
| 1119 | |
| 1120 | do { |
| 1121 | unsigned long pending_words; |
| 1122 | |
| 1123 | vcpu_info->evtchn_upcall_pending = 0; |
| 1124 | |
| 1125 | if (__this_cpu_inc_return(xed_nesting_count) - 1) |
| 1126 | goto out; |
| 1127 | |
| 1128 | #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */ |
| 1129 | /* Clear master flag /before/ clearing selector flag. */ |
| 1130 | wmb(); |
| 1131 | #endif |
| 1132 | pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0); |
| 1133 | |
| 1134 | start_word_idx = __this_cpu_read(current_word_idx); |
| 1135 | start_bit_idx = __this_cpu_read(current_bit_idx); |
| 1136 | |
| 1137 | word_idx = start_word_idx; |
| 1138 | |
| 1139 | for (i = 0; pending_words != 0; i++) { |
| 1140 | unsigned long pending_bits; |
| 1141 | unsigned long words; |
| 1142 | |
| 1143 | words = MASK_LSBS(pending_words, word_idx); |
| 1144 | |
| 1145 | /* |
| 1146 | * If we masked out all events, wrap to beginning. |
| 1147 | */ |
| 1148 | if (words == 0) { |
| 1149 | word_idx = 0; |
| 1150 | bit_idx = 0; |
| 1151 | continue; |
| 1152 | } |
| 1153 | word_idx = __ffs(words); |
| 1154 | |
| 1155 | pending_bits = active_evtchns(cpu, s, word_idx); |
| 1156 | bit_idx = 0; /* usually scan entire word from start */ |
| 1157 | if (word_idx == start_word_idx) { |
| 1158 | /* We scan the starting word in two parts */ |
| 1159 | if (i == 0) |
| 1160 | /* 1st time: start in the middle */ |
| 1161 | bit_idx = start_bit_idx; |
| 1162 | else |
| 1163 | /* 2nd time: mask bits done already */ |
| 1164 | bit_idx &= (1UL << start_bit_idx) - 1; |
| 1165 | } |
| 1166 | |
| 1167 | do { |
| 1168 | unsigned long bits; |
| 1169 | int port, irq; |
| 1170 | struct irq_desc *desc; |
| 1171 | |
| 1172 | bits = MASK_LSBS(pending_bits, bit_idx); |
| 1173 | |
| 1174 | /* If we masked out all events, move on. */ |
| 1175 | if (bits == 0) |
| 1176 | break; |
| 1177 | |
| 1178 | bit_idx = __ffs(bits); |
| 1179 | |
| 1180 | /* Process port. */ |
| 1181 | port = (word_idx * BITS_PER_LONG) + bit_idx; |
| 1182 | irq = evtchn_to_irq[port]; |
| 1183 | |
| 1184 | mask_evtchn(port); |
| 1185 | clear_evtchn(port); |
| 1186 | |
| 1187 | if (irq != -1) { |
| 1188 | desc = irq_to_desc(irq); |
| 1189 | if (desc) |
| 1190 | generic_handle_irq_desc(irq, desc); |
| 1191 | } |
| 1192 | |
| 1193 | bit_idx = (bit_idx + 1) % BITS_PER_LONG; |
| 1194 | |
| 1195 | /* Next caller starts at last processed + 1 */ |
| 1196 | __this_cpu_write(current_word_idx, |
| 1197 | bit_idx ? word_idx : |
| 1198 | (word_idx+1) % BITS_PER_LONG); |
| 1199 | __this_cpu_write(current_bit_idx, bit_idx); |
| 1200 | } while (bit_idx != 0); |
| 1201 | |
| 1202 | /* Scan start_l1i twice; all others once. */ |
| 1203 | if ((word_idx != start_word_idx) || (i != 0)) |
| 1204 | pending_words &= ~(1UL << word_idx); |
| 1205 | |
| 1206 | word_idx = (word_idx + 1) % BITS_PER_LONG; |
| 1207 | } |
| 1208 | |
| 1209 | BUG_ON(!irqs_disabled()); |
| 1210 | |
| 1211 | count = __this_cpu_read(xed_nesting_count); |
| 1212 | __this_cpu_write(xed_nesting_count, 0); |
| 1213 | } while (count != 1 || vcpu_info->evtchn_upcall_pending); |
| 1214 | |
| 1215 | out: |
| 1216 | |
| 1217 | put_cpu(); |
| 1218 | } |
| 1219 | |
| 1220 | void xen_evtchn_do_upcall(struct pt_regs *regs) |
| 1221 | { |
| 1222 | struct pt_regs *old_regs = set_irq_regs(regs); |
| 1223 | |
| 1224 | exit_idle(); |
| 1225 | irq_enter(); |
| 1226 | |
| 1227 | __xen_evtchn_do_upcall(); |
| 1228 | |
| 1229 | irq_exit(); |
| 1230 | set_irq_regs(old_regs); |
| 1231 | } |
| 1232 | |
| 1233 | void xen_hvm_evtchn_do_upcall(void) |
| 1234 | { |
| 1235 | __xen_evtchn_do_upcall(); |
| 1236 | } |
| 1237 | EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall); |
| 1238 | |
| 1239 | /* Rebind a new event channel to an existing irq. */ |
| 1240 | void rebind_evtchn_irq(int evtchn, int irq) |
| 1241 | { |
| 1242 | struct irq_info *info = info_for_irq(irq); |
| 1243 | |
| 1244 | /* Make sure the irq is masked, since the new event channel |
| 1245 | will also be masked. */ |
| 1246 | disable_irq(irq); |
| 1247 | |
| 1248 | spin_lock(&irq_mapping_update_lock); |
| 1249 | |
| 1250 | /* After resume the irq<->evtchn mappings are all cleared out */ |
| 1251 | BUG_ON(evtchn_to_irq[evtchn] != -1); |
| 1252 | /* Expect irq to have been bound before, |
| 1253 | so there should be a proper type */ |
| 1254 | BUG_ON(info->type == IRQT_UNBOUND); |
| 1255 | |
| 1256 | xen_irq_info_evtchn_init(irq, evtchn); |
| 1257 | |
| 1258 | spin_unlock(&irq_mapping_update_lock); |
| 1259 | |
| 1260 | /* new event channels are always bound to cpu 0 */ |
| 1261 | irq_set_affinity(irq, cpumask_of(0)); |
| 1262 | |
| 1263 | /* Unmask the event channel. */ |
| 1264 | enable_irq(irq); |
| 1265 | } |
| 1266 | |
| 1267 | /* Rebind an evtchn so that it gets delivered to a specific cpu */ |
| 1268 | static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu) |
| 1269 | { |
| 1270 | struct evtchn_bind_vcpu bind_vcpu; |
| 1271 | int evtchn = evtchn_from_irq(irq); |
| 1272 | |
| 1273 | if (!VALID_EVTCHN(evtchn)) |
| 1274 | return -1; |
| 1275 | |
| 1276 | /* |
| 1277 | * Events delivered via platform PCI interrupts are always |
| 1278 | * routed to vcpu 0 and hence cannot be rebound. |
| 1279 | */ |
| 1280 | if (xen_hvm_domain() && !xen_have_vector_callback) |
| 1281 | return -1; |
| 1282 | |
| 1283 | /* Send future instances of this interrupt to other vcpu. */ |
| 1284 | bind_vcpu.port = evtchn; |
| 1285 | bind_vcpu.vcpu = tcpu; |
| 1286 | |
| 1287 | /* |
| 1288 | * If this fails, it usually just indicates that we're dealing with a |
| 1289 | * virq or IPI channel, which don't actually need to be rebound. Ignore |
| 1290 | * it, but don't do the xenlinux-level rebind in that case. |
| 1291 | */ |
| 1292 | if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0) |
| 1293 | bind_evtchn_to_cpu(evtchn, tcpu); |
| 1294 | |
| 1295 | return 0; |
| 1296 | } |
| 1297 | |
| 1298 | static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest, |
| 1299 | bool force) |
| 1300 | { |
| 1301 | unsigned tcpu = cpumask_first(dest); |
| 1302 | |
| 1303 | return rebind_irq_to_cpu(data->irq, tcpu); |
| 1304 | } |
| 1305 | |
| 1306 | int resend_irq_on_evtchn(unsigned int irq) |
| 1307 | { |
| 1308 | int masked, evtchn = evtchn_from_irq(irq); |
| 1309 | struct shared_info *s = HYPERVISOR_shared_info; |
| 1310 | |
| 1311 | if (!VALID_EVTCHN(evtchn)) |
| 1312 | return 1; |
| 1313 | |
| 1314 | masked = sync_test_and_set_bit(evtchn, s->evtchn_mask); |
| 1315 | sync_set_bit(evtchn, s->evtchn_pending); |
| 1316 | if (!masked) |
| 1317 | unmask_evtchn(evtchn); |
| 1318 | |
| 1319 | return 1; |
| 1320 | } |
| 1321 | |
| 1322 | static void enable_dynirq(struct irq_data *data) |
| 1323 | { |
| 1324 | int evtchn = evtchn_from_irq(data->irq); |
| 1325 | |
| 1326 | if (VALID_EVTCHN(evtchn)) |
| 1327 | unmask_evtchn(evtchn); |
| 1328 | } |
| 1329 | |
| 1330 | static void disable_dynirq(struct irq_data *data) |
| 1331 | { |
| 1332 | int evtchn = evtchn_from_irq(data->irq); |
| 1333 | |
| 1334 | if (VALID_EVTCHN(evtchn)) |
| 1335 | mask_evtchn(evtchn); |
| 1336 | } |
| 1337 | |
| 1338 | static void ack_dynirq(struct irq_data *data) |
| 1339 | { |
| 1340 | int evtchn = evtchn_from_irq(data->irq); |
| 1341 | |
| 1342 | irq_move_masked_irq(data); |
| 1343 | |
| 1344 | if (VALID_EVTCHN(evtchn)) |
| 1345 | unmask_evtchn(evtchn); |
| 1346 | } |
| 1347 | |
| 1348 | static int retrigger_dynirq(struct irq_data *data) |
| 1349 | { |
| 1350 | int evtchn = evtchn_from_irq(data->irq); |
| 1351 | struct shared_info *sh = HYPERVISOR_shared_info; |
| 1352 | int ret = 0; |
| 1353 | |
| 1354 | if (VALID_EVTCHN(evtchn)) { |
| 1355 | int masked; |
| 1356 | |
| 1357 | masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask); |
| 1358 | sync_set_bit(evtchn, sh->evtchn_pending); |
| 1359 | if (!masked) |
| 1360 | unmask_evtchn(evtchn); |
| 1361 | ret = 1; |
| 1362 | } |
| 1363 | |
| 1364 | return ret; |
| 1365 | } |
| 1366 | |
| 1367 | static void restore_pirqs(void) |
| 1368 | { |
| 1369 | int pirq, rc, irq, gsi; |
| 1370 | struct physdev_map_pirq map_irq; |
| 1371 | struct irq_info *info; |
| 1372 | |
| 1373 | list_for_each_entry(info, &xen_irq_list_head, list) { |
| 1374 | if (info->type != IRQT_PIRQ) |
| 1375 | continue; |
| 1376 | |
| 1377 | pirq = info->u.pirq.pirq; |
| 1378 | gsi = info->u.pirq.gsi; |
| 1379 | irq = info->irq; |
| 1380 | |
| 1381 | /* save/restore of PT devices doesn't work, so at this point the |
| 1382 | * only devices present are GSI based emulated devices */ |
| 1383 | if (!gsi) |
| 1384 | continue; |
| 1385 | |
| 1386 | map_irq.domid = DOMID_SELF; |
| 1387 | map_irq.type = MAP_PIRQ_TYPE_GSI; |
| 1388 | map_irq.index = gsi; |
| 1389 | map_irq.pirq = pirq; |
| 1390 | |
| 1391 | rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq); |
| 1392 | if (rc) { |
| 1393 | printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n", |
| 1394 | gsi, irq, pirq, rc); |
| 1395 | xen_free_irq(irq); |
| 1396 | continue; |
| 1397 | } |
| 1398 | |
| 1399 | printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq); |
| 1400 | |
| 1401 | __startup_pirq(irq); |
| 1402 | } |
| 1403 | } |
| 1404 | |
| 1405 | static void restore_cpu_virqs(unsigned int cpu) |
| 1406 | { |
| 1407 | struct evtchn_bind_virq bind_virq; |
| 1408 | int virq, irq, evtchn; |
| 1409 | |
| 1410 | for (virq = 0; virq < NR_VIRQS; virq++) { |
| 1411 | if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1) |
| 1412 | continue; |
| 1413 | |
| 1414 | BUG_ON(virq_from_irq(irq) != virq); |
| 1415 | |
| 1416 | /* Get a new binding from Xen. */ |
| 1417 | bind_virq.virq = virq; |
| 1418 | bind_virq.vcpu = cpu; |
| 1419 | if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq, |
| 1420 | &bind_virq) != 0) |
| 1421 | BUG(); |
| 1422 | evtchn = bind_virq.port; |
| 1423 | |
| 1424 | /* Record the new mapping. */ |
| 1425 | xen_irq_info_virq_init(cpu, irq, evtchn, virq); |
| 1426 | bind_evtchn_to_cpu(evtchn, cpu); |
| 1427 | } |
| 1428 | } |
| 1429 | |
| 1430 | static void restore_cpu_ipis(unsigned int cpu) |
| 1431 | { |
| 1432 | struct evtchn_bind_ipi bind_ipi; |
| 1433 | int ipi, irq, evtchn; |
| 1434 | |
| 1435 | for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) { |
| 1436 | if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1) |
| 1437 | continue; |
| 1438 | |
| 1439 | BUG_ON(ipi_from_irq(irq) != ipi); |
| 1440 | |
| 1441 | /* Get a new binding from Xen. */ |
| 1442 | bind_ipi.vcpu = cpu; |
| 1443 | if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi, |
| 1444 | &bind_ipi) != 0) |
| 1445 | BUG(); |
| 1446 | evtchn = bind_ipi.port; |
| 1447 | |
| 1448 | /* Record the new mapping. */ |
| 1449 | xen_irq_info_ipi_init(cpu, irq, evtchn, ipi); |
| 1450 | bind_evtchn_to_cpu(evtchn, cpu); |
| 1451 | } |
| 1452 | } |
| 1453 | |
| 1454 | /* Clear an irq's pending state, in preparation for polling on it */ |
| 1455 | void xen_clear_irq_pending(int irq) |
| 1456 | { |
| 1457 | int evtchn = evtchn_from_irq(irq); |
| 1458 | |
| 1459 | if (VALID_EVTCHN(evtchn)) |
| 1460 | clear_evtchn(evtchn); |
| 1461 | } |
| 1462 | EXPORT_SYMBOL(xen_clear_irq_pending); |
| 1463 | void xen_set_irq_pending(int irq) |
| 1464 | { |
| 1465 | int evtchn = evtchn_from_irq(irq); |
| 1466 | |
| 1467 | if (VALID_EVTCHN(evtchn)) |
| 1468 | set_evtchn(evtchn); |
| 1469 | } |
| 1470 | |
| 1471 | bool xen_test_irq_pending(int irq) |
| 1472 | { |
| 1473 | int evtchn = evtchn_from_irq(irq); |
| 1474 | bool ret = false; |
| 1475 | |
| 1476 | if (VALID_EVTCHN(evtchn)) |
| 1477 | ret = test_evtchn(evtchn); |
| 1478 | |
| 1479 | return ret; |
| 1480 | } |
| 1481 | |
| 1482 | /* Poll waiting for an irq to become pending with timeout. In the usual case, |
| 1483 | * the irq will be disabled so it won't deliver an interrupt. */ |
| 1484 | void xen_poll_irq_timeout(int irq, u64 timeout) |
| 1485 | { |
| 1486 | evtchn_port_t evtchn = evtchn_from_irq(irq); |
| 1487 | |
| 1488 | if (VALID_EVTCHN(evtchn)) { |
| 1489 | struct sched_poll poll; |
| 1490 | |
| 1491 | poll.nr_ports = 1; |
| 1492 | poll.timeout = timeout; |
| 1493 | set_xen_guest_handle(poll.ports, &evtchn); |
| 1494 | |
| 1495 | if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0) |
| 1496 | BUG(); |
| 1497 | } |
| 1498 | } |
| 1499 | EXPORT_SYMBOL(xen_poll_irq_timeout); |
| 1500 | /* Poll waiting for an irq to become pending. In the usual case, the |
| 1501 | * irq will be disabled so it won't deliver an interrupt. */ |
| 1502 | void xen_poll_irq(int irq) |
| 1503 | { |
| 1504 | xen_poll_irq_timeout(irq, 0 /* no timeout */); |
| 1505 | } |
| 1506 | |
| 1507 | void xen_irq_resume(void) |
| 1508 | { |
| 1509 | unsigned int cpu, evtchn; |
| 1510 | struct irq_info *info; |
| 1511 | |
| 1512 | init_evtchn_cpu_bindings(); |
| 1513 | |
| 1514 | /* New event-channel space is not 'live' yet. */ |
| 1515 | for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++) |
| 1516 | mask_evtchn(evtchn); |
| 1517 | |
| 1518 | /* No IRQ <-> event-channel mappings. */ |
| 1519 | list_for_each_entry(info, &xen_irq_list_head, list) |
| 1520 | info->evtchn = 0; /* zap event-channel binding */ |
| 1521 | |
| 1522 | for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++) |
| 1523 | evtchn_to_irq[evtchn] = -1; |
| 1524 | |
| 1525 | for_each_possible_cpu(cpu) { |
| 1526 | restore_cpu_virqs(cpu); |
| 1527 | restore_cpu_ipis(cpu); |
| 1528 | } |
| 1529 | |
| 1530 | restore_pirqs(); |
| 1531 | } |
| 1532 | |
| 1533 | static struct irq_chip xen_dynamic_chip __read_mostly = { |
| 1534 | .name = "xen-dyn", |
| 1535 | |
| 1536 | .irq_disable = disable_dynirq, |
| 1537 | .irq_mask = disable_dynirq, |
| 1538 | .irq_unmask = enable_dynirq, |
| 1539 | |
| 1540 | .irq_eoi = ack_dynirq, |
| 1541 | .irq_set_affinity = set_affinity_irq, |
| 1542 | .irq_retrigger = retrigger_dynirq, |
| 1543 | }; |
| 1544 | |
| 1545 | static struct irq_chip xen_pirq_chip __read_mostly = { |
| 1546 | .name = "xen-pirq", |
| 1547 | |
| 1548 | .irq_startup = startup_pirq, |
| 1549 | .irq_shutdown = shutdown_pirq, |
| 1550 | |
| 1551 | .irq_enable = enable_pirq, |
| 1552 | .irq_unmask = enable_pirq, |
| 1553 | |
| 1554 | .irq_disable = disable_pirq, |
| 1555 | .irq_mask = disable_pirq, |
| 1556 | |
| 1557 | .irq_ack = ack_pirq, |
| 1558 | |
| 1559 | .irq_set_affinity = set_affinity_irq, |
| 1560 | |
| 1561 | .irq_retrigger = retrigger_dynirq, |
| 1562 | }; |
| 1563 | |
| 1564 | static struct irq_chip xen_percpu_chip __read_mostly = { |
| 1565 | .name = "xen-percpu", |
| 1566 | |
| 1567 | .irq_disable = disable_dynirq, |
| 1568 | .irq_mask = disable_dynirq, |
| 1569 | .irq_unmask = enable_dynirq, |
| 1570 | |
| 1571 | .irq_ack = ack_dynirq, |
| 1572 | }; |
| 1573 | |
| 1574 | int xen_set_callback_via(uint64_t via) |
| 1575 | { |
| 1576 | struct xen_hvm_param a; |
| 1577 | a.domid = DOMID_SELF; |
| 1578 | a.index = HVM_PARAM_CALLBACK_IRQ; |
| 1579 | a.value = via; |
| 1580 | return HYPERVISOR_hvm_op(HVMOP_set_param, &a); |
| 1581 | } |
| 1582 | EXPORT_SYMBOL_GPL(xen_set_callback_via); |
| 1583 | |
| 1584 | #ifdef CONFIG_XEN_PVHVM |
| 1585 | /* Vector callbacks are better than PCI interrupts to receive event |
| 1586 | * channel notifications because we can receive vector callbacks on any |
| 1587 | * vcpu and we don't need PCI support or APIC interactions. */ |
| 1588 | void xen_callback_vector(void) |
| 1589 | { |
| 1590 | int rc; |
| 1591 | uint64_t callback_via; |
| 1592 | if (xen_have_vector_callback) { |
| 1593 | callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK); |
| 1594 | rc = xen_set_callback_via(callback_via); |
| 1595 | if (rc) { |
| 1596 | printk(KERN_ERR "Request for Xen HVM callback vector" |
| 1597 | " failed.\n"); |
| 1598 | xen_have_vector_callback = 0; |
| 1599 | return; |
| 1600 | } |
| 1601 | printk(KERN_INFO "Xen HVM callback vector for event delivery is " |
| 1602 | "enabled\n"); |
| 1603 | /* in the restore case the vector has already been allocated */ |
| 1604 | if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors)) |
| 1605 | alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector); |
| 1606 | } |
| 1607 | } |
| 1608 | #else |
| 1609 | void xen_callback_vector(void) {} |
| 1610 | #endif |
| 1611 | |
| 1612 | void __init xen_init_IRQ(void) |
| 1613 | { |
| 1614 | int i; |
| 1615 | |
| 1616 | evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq), |
| 1617 | GFP_KERNEL); |
| 1618 | for (i = 0; i < NR_EVENT_CHANNELS; i++) |
| 1619 | evtchn_to_irq[i] = -1; |
| 1620 | |
| 1621 | init_evtchn_cpu_bindings(); |
| 1622 | |
| 1623 | /* No event channels are 'live' right now. */ |
| 1624 | for (i = 0; i < NR_EVENT_CHANNELS; i++) |
| 1625 | mask_evtchn(i); |
| 1626 | |
| 1627 | if (xen_hvm_domain()) { |
| 1628 | xen_callback_vector(); |
| 1629 | native_init_IRQ(); |
| 1630 | /* pci_xen_hvm_init must be called after native_init_IRQ so that |
| 1631 | * __acpi_register_gsi can point at the right function */ |
| 1632 | pci_xen_hvm_init(); |
| 1633 | } else { |
| 1634 | irq_ctx_init(smp_processor_id()); |
| 1635 | if (xen_initial_domain()) |
| 1636 | xen_setup_pirqs(); |
| 1637 | } |
| 1638 | } |