Linux Audio

Check our new training course

Loading...
   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#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  25
  26#include <linux/linkage.h>
  27#include <linux/interrupt.h>
  28#include <linux/irq.h>
  29#include <linux/moduleparam.h>
  30#include <linux/string.h>
  31#include <linux/bootmem.h>
  32#include <linux/slab.h>
  33#include <linux/irqnr.h>
  34#include <linux/pci.h>
  35
  36#ifdef CONFIG_X86
  37#include <asm/desc.h>
  38#include <asm/ptrace.h>
  39#include <asm/irq.h>
  40#include <asm/idle.h>
  41#include <asm/io_apic.h>
  42#include <asm/i8259.h>
  43#include <asm/xen/pci.h>
  44#endif
  45#include <asm/sync_bitops.h>
  46#include <asm/xen/hypercall.h>
  47#include <asm/xen/hypervisor.h>
  48#include <xen/page.h>
  49
  50#include <xen/xen.h>
  51#include <xen/hvm.h>
  52#include <xen/xen-ops.h>
  53#include <xen/events.h>
  54#include <xen/interface/xen.h>
  55#include <xen/interface/event_channel.h>
  56#include <xen/interface/hvm/hvm_op.h>
  57#include <xen/interface/hvm/params.h>
  58#include <xen/interface/physdev.h>
  59#include <xen/interface/sched.h>
  60#include <xen/interface/vcpu.h>
  61#include <asm/hw_irq.h>
  62
  63#include "events_internal.h"
  64
  65const struct evtchn_ops *evtchn_ops;
  66
  67/*
  68 * This lock protects updates to the following mapping and reference-count
  69 * arrays. The lock does not need to be acquired to read the mapping tables.
  70 */
  71static DEFINE_MUTEX(irq_mapping_update_lock);
  72
  73static LIST_HEAD(xen_irq_list_head);
  74
  75/* IRQ <-> VIRQ mapping. */
  76static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  77
  78/* IRQ <-> IPI mapping */
  79static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  80
  81int **evtchn_to_irq;
  82#ifdef CONFIG_X86
  83static unsigned long *pirq_eoi_map;
  84#endif
  85static bool (*pirq_needs_eoi)(unsigned irq);
  86
  87#define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88#define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  89#define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  90
  91/* Xen will never allocate port zero for any purpose. */
  92#define VALID_EVTCHN(chn)	((chn) != 0)
  93
  94static struct irq_chip xen_dynamic_chip;
  95static struct irq_chip xen_percpu_chip;
  96static struct irq_chip xen_pirq_chip;
  97static void enable_dynirq(struct irq_data *data);
  98static void disable_dynirq(struct irq_data *data);
  99
 100static void clear_evtchn_to_irq_row(unsigned row)
 101{
 102	unsigned col;
 103
 104	for (col = 0; col < EVTCHN_PER_ROW; col++)
 105		evtchn_to_irq[row][col] = -1;
 106}
 107
 108static void clear_evtchn_to_irq_all(void)
 109{
 110	unsigned row;
 111
 112	for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 113		if (evtchn_to_irq[row] == NULL)
 114			continue;
 115		clear_evtchn_to_irq_row(row);
 116	}
 117}
 118
 119static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 120{
 121	unsigned row;
 122	unsigned col;
 123
 124	if (evtchn >= xen_evtchn_max_channels())
 125		return -EINVAL;
 126
 127	row = EVTCHN_ROW(evtchn);
 128	col = EVTCHN_COL(evtchn);
 129
 130	if (evtchn_to_irq[row] == NULL) {
 131		/* Unallocated irq entries return -1 anyway */
 132		if (irq == -1)
 133			return 0;
 134
 135		evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 136		if (evtchn_to_irq[row] == NULL)
 137			return -ENOMEM;
 138
 139		clear_evtchn_to_irq_row(row);
 140	}
 141
 142	evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 143	return 0;
 144}
 145
 146int get_evtchn_to_irq(unsigned evtchn)
 147{
 148	if (evtchn >= xen_evtchn_max_channels())
 149		return -1;
 150	if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 151		return -1;
 152	return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 153}
 154
 155/* Get info for IRQ */
 156struct irq_info *info_for_irq(unsigned irq)
 157{
 158	return irq_get_handler_data(irq);
 159}
 160
 161/* Constructors for packed IRQ information. */
 162static int xen_irq_info_common_setup(struct irq_info *info,
 163				     unsigned irq,
 164				     enum xen_irq_type type,
 165				     unsigned evtchn,
 166				     unsigned short cpu)
 167{
 168	int ret;
 169
 170	BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 171
 172	info->type = type;
 173	info->irq = irq;
 174	info->evtchn = evtchn;
 175	info->cpu = cpu;
 176
 177	ret = set_evtchn_to_irq(evtchn, irq);
 178	if (ret < 0)
 179		return ret;
 180
 181	irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 182
 183	return xen_evtchn_port_setup(info);
 184}
 185
 186static int xen_irq_info_evtchn_setup(unsigned irq,
 187				     unsigned evtchn)
 188{
 189	struct irq_info *info = info_for_irq(irq);
 190
 191	return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 192}
 193
 194static int xen_irq_info_ipi_setup(unsigned cpu,
 195				  unsigned irq,
 196				  unsigned evtchn,
 197				  enum ipi_vector ipi)
 198{
 199	struct irq_info *info = info_for_irq(irq);
 200
 201	info->u.ipi = ipi;
 202
 203	per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 204
 205	return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 206}
 207
 208static int xen_irq_info_virq_setup(unsigned cpu,
 209				   unsigned irq,
 210				   unsigned evtchn,
 211				   unsigned virq)
 212{
 213	struct irq_info *info = info_for_irq(irq);
 214
 215	info->u.virq = virq;
 216
 217	per_cpu(virq_to_irq, cpu)[virq] = irq;
 218
 219	return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 220}
 221
 222static int xen_irq_info_pirq_setup(unsigned irq,
 223				   unsigned evtchn,
 224				   unsigned pirq,
 225				   unsigned gsi,
 226				   uint16_t domid,
 227				   unsigned char flags)
 228{
 229	struct irq_info *info = info_for_irq(irq);
 230
 231	info->u.pirq.pirq = pirq;
 232	info->u.pirq.gsi = gsi;
 233	info->u.pirq.domid = domid;
 234	info->u.pirq.flags = flags;
 235
 236	return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 237}
 238
 239static void xen_irq_info_cleanup(struct irq_info *info)
 240{
 241	set_evtchn_to_irq(info->evtchn, -1);
 242	info->evtchn = 0;
 243}
 244
 245/*
 246 * Accessors for packed IRQ information.
 247 */
 248unsigned int evtchn_from_irq(unsigned irq)
 249{
 250	if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 251		return 0;
 252
 253	return info_for_irq(irq)->evtchn;
 254}
 255
 256unsigned irq_from_evtchn(unsigned int evtchn)
 257{
 258	return get_evtchn_to_irq(evtchn);
 259}
 260EXPORT_SYMBOL_GPL(irq_from_evtchn);
 261
 262int irq_from_virq(unsigned int cpu, unsigned int virq)
 263{
 264	return per_cpu(virq_to_irq, cpu)[virq];
 265}
 266
 267static enum ipi_vector ipi_from_irq(unsigned irq)
 268{
 269	struct irq_info *info = info_for_irq(irq);
 270
 271	BUG_ON(info == NULL);
 272	BUG_ON(info->type != IRQT_IPI);
 273
 274	return info->u.ipi;
 275}
 276
 277static unsigned virq_from_irq(unsigned irq)
 278{
 279	struct irq_info *info = info_for_irq(irq);
 280
 281	BUG_ON(info == NULL);
 282	BUG_ON(info->type != IRQT_VIRQ);
 283
 284	return info->u.virq;
 285}
 286
 287static unsigned pirq_from_irq(unsigned irq)
 288{
 289	struct irq_info *info = info_for_irq(irq);
 290
 291	BUG_ON(info == NULL);
 292	BUG_ON(info->type != IRQT_PIRQ);
 293
 294	return info->u.pirq.pirq;
 295}
 296
 297static enum xen_irq_type type_from_irq(unsigned irq)
 298{
 299	return info_for_irq(irq)->type;
 300}
 301
 302unsigned cpu_from_irq(unsigned irq)
 303{
 304	return info_for_irq(irq)->cpu;
 305}
 306
 307unsigned int cpu_from_evtchn(unsigned int evtchn)
 308{
 309	int irq = get_evtchn_to_irq(evtchn);
 310	unsigned ret = 0;
 311
 312	if (irq != -1)
 313		ret = cpu_from_irq(irq);
 314
 315	return ret;
 316}
 317
 318#ifdef CONFIG_X86
 319static bool pirq_check_eoi_map(unsigned irq)
 320{
 321	return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 322}
 323#endif
 324
 325static bool pirq_needs_eoi_flag(unsigned irq)
 326{
 327	struct irq_info *info = info_for_irq(irq);
 328	BUG_ON(info->type != IRQT_PIRQ);
 329
 330	return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 331}
 332
 333static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 334{
 335	int irq = get_evtchn_to_irq(chn);
 336	struct irq_info *info = info_for_irq(irq);
 337
 338	BUG_ON(irq == -1);
 339#ifdef CONFIG_SMP
 340	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 341#endif
 342	xen_evtchn_port_bind_to_cpu(info, cpu);
 343
 344	info->cpu = cpu;
 345}
 346
 347static void xen_evtchn_mask_all(void)
 348{
 349	unsigned int evtchn;
 350
 351	for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
 352		mask_evtchn(evtchn);
 353}
 354
 355/**
 356 * notify_remote_via_irq - send event to remote end of event channel via irq
 357 * @irq: irq of event channel to send event to
 358 *
 359 * Unlike notify_remote_via_evtchn(), this is safe to use across
 360 * save/restore. Notifications on a broken connection are silently
 361 * dropped.
 362 */
 363void notify_remote_via_irq(int irq)
 364{
 365	int evtchn = evtchn_from_irq(irq);
 366
 367	if (VALID_EVTCHN(evtchn))
 368		notify_remote_via_evtchn(evtchn);
 369}
 370EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 371
 372static void xen_irq_init(unsigned irq)
 373{
 374	struct irq_info *info;
 375#ifdef CONFIG_SMP
 376	/* By default all event channels notify CPU#0. */
 377	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 378#endif
 379
 380	info = kzalloc(sizeof(*info), GFP_KERNEL);
 381	if (info == NULL)
 382		panic("Unable to allocate metadata for IRQ%d\n", irq);
 383
 384	info->type = IRQT_UNBOUND;
 385	info->refcnt = -1;
 386
 387	irq_set_handler_data(irq, info);
 388
 389	list_add_tail(&info->list, &xen_irq_list_head);
 390}
 391
 392static int __must_check xen_allocate_irqs_dynamic(int nvec)
 393{
 394	int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 395
 396	if (irq >= 0) {
 397		for (i = 0; i < nvec; i++)
 398			xen_irq_init(irq + i);
 399	}
 400
 401	return irq;
 402}
 403
 404static inline int __must_check xen_allocate_irq_dynamic(void)
 405{
 406
 407	return xen_allocate_irqs_dynamic(1);
 408}
 409
 410static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 411{
 412	int irq;
 413
 414	/*
 415	 * A PV guest has no concept of a GSI (since it has no ACPI
 416	 * nor access to/knowledge of the physical APICs). Therefore
 417	 * all IRQs are dynamically allocated from the entire IRQ
 418	 * space.
 419	 */
 420	if (xen_pv_domain() && !xen_initial_domain())
 421		return xen_allocate_irq_dynamic();
 422
 423	/* Legacy IRQ descriptors are already allocated by the arch. */
 424	if (gsi < nr_legacy_irqs())
 425		irq = gsi;
 426	else
 427		irq = irq_alloc_desc_at(gsi, -1);
 428
 429	xen_irq_init(irq);
 430
 431	return irq;
 432}
 433
 434static void xen_free_irq(unsigned irq)
 435{
 436	struct irq_info *info = irq_get_handler_data(irq);
 437
 438	if (WARN_ON(!info))
 439		return;
 440
 441	list_del(&info->list);
 442
 443	irq_set_handler_data(irq, NULL);
 444
 445	WARN_ON(info->refcnt > 0);
 446
 447	kfree(info);
 448
 449	/* Legacy IRQ descriptors are managed by the arch. */
 450	if (irq < nr_legacy_irqs())
 451		return;
 452
 453	irq_free_desc(irq);
 454}
 455
 456static void xen_evtchn_close(unsigned int port)
 457{
 458	struct evtchn_close close;
 459
 460	close.port = port;
 461	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 462		BUG();
 463}
 464
 465static void pirq_query_unmask(int irq)
 466{
 467	struct physdev_irq_status_query irq_status;
 468	struct irq_info *info = info_for_irq(irq);
 469
 470	BUG_ON(info->type != IRQT_PIRQ);
 471
 472	irq_status.irq = pirq_from_irq(irq);
 473	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 474		irq_status.flags = 0;
 475
 476	info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 477	if (irq_status.flags & XENIRQSTAT_needs_eoi)
 478		info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 479}
 480
 481static void eoi_pirq(struct irq_data *data)
 482{
 483	int evtchn = evtchn_from_irq(data->irq);
 484	struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 485	int rc = 0;
 486
 487	if (!VALID_EVTCHN(evtchn))
 488		return;
 489
 490	if (unlikely(irqd_is_setaffinity_pending(data))) {
 
 491		int masked = test_and_set_mask(evtchn);
 492
 493		clear_evtchn(evtchn);
 494
 495		irq_move_masked_irq(data);
 496
 497		if (!masked)
 498			unmask_evtchn(evtchn);
 499	} else
 500		clear_evtchn(evtchn);
 501
 502	if (pirq_needs_eoi(data->irq)) {
 503		rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 504		WARN_ON(rc);
 505	}
 506}
 507
 508static void mask_ack_pirq(struct irq_data *data)
 509{
 510	disable_dynirq(data);
 511	eoi_pirq(data);
 512}
 513
 514static unsigned int __startup_pirq(unsigned int irq)
 515{
 516	struct evtchn_bind_pirq bind_pirq;
 517	struct irq_info *info = info_for_irq(irq);
 518	int evtchn = evtchn_from_irq(irq);
 519	int rc;
 520
 521	BUG_ON(info->type != IRQT_PIRQ);
 522
 523	if (VALID_EVTCHN(evtchn))
 524		goto out;
 525
 526	bind_pirq.pirq = pirq_from_irq(irq);
 527	/* NB. We are happy to share unless we are probing. */
 528	bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 529					BIND_PIRQ__WILL_SHARE : 0;
 530	rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 531	if (rc != 0) {
 532		pr_warn("Failed to obtain physical IRQ %d\n", irq);
 533		return 0;
 534	}
 535	evtchn = bind_pirq.port;
 536
 537	pirq_query_unmask(irq);
 538
 539	rc = set_evtchn_to_irq(evtchn, irq);
 540	if (rc)
 541		goto err;
 542
 543	info->evtchn = evtchn;
 544	bind_evtchn_to_cpu(evtchn, 0);
 545
 546	rc = xen_evtchn_port_setup(info);
 547	if (rc)
 548		goto err;
 549
 550out:
 551	unmask_evtchn(evtchn);
 552	eoi_pirq(irq_get_irq_data(irq));
 553
 554	return 0;
 555
 556err:
 557	pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 558	xen_evtchn_close(evtchn);
 559	return 0;
 560}
 561
 562static unsigned int startup_pirq(struct irq_data *data)
 563{
 564	return __startup_pirq(data->irq);
 565}
 566
 567static void shutdown_pirq(struct irq_data *data)
 568{
 569	unsigned int irq = data->irq;
 570	struct irq_info *info = info_for_irq(irq);
 571	unsigned evtchn = evtchn_from_irq(irq);
 572
 573	BUG_ON(info->type != IRQT_PIRQ);
 574
 575	if (!VALID_EVTCHN(evtchn))
 576		return;
 577
 578	mask_evtchn(evtchn);
 579	xen_evtchn_close(evtchn);
 580	xen_irq_info_cleanup(info);
 581}
 582
 583static void enable_pirq(struct irq_data *data)
 584{
 585	startup_pirq(data);
 586}
 587
 588static void disable_pirq(struct irq_data *data)
 589{
 590	disable_dynirq(data);
 591}
 592
 593int xen_irq_from_gsi(unsigned gsi)
 594{
 595	struct irq_info *info;
 596
 597	list_for_each_entry(info, &xen_irq_list_head, list) {
 598		if (info->type != IRQT_PIRQ)
 599			continue;
 600
 601		if (info->u.pirq.gsi == gsi)
 602			return info->irq;
 603	}
 604
 605	return -1;
 606}
 607EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 608
 609static void __unbind_from_irq(unsigned int irq)
 610{
 611	int evtchn = evtchn_from_irq(irq);
 612	struct irq_info *info = irq_get_handler_data(irq);
 613
 614	if (info->refcnt > 0) {
 615		info->refcnt--;
 616		if (info->refcnt != 0)
 617			return;
 618	}
 619
 620	if (VALID_EVTCHN(evtchn)) {
 621		unsigned int cpu = cpu_from_irq(irq);
 622
 623		xen_evtchn_close(evtchn);
 624
 625		switch (type_from_irq(irq)) {
 626		case IRQT_VIRQ:
 627			per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 628			break;
 629		case IRQT_IPI:
 630			per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 631			break;
 632		default:
 633			break;
 634		}
 635
 636		xen_irq_info_cleanup(info);
 637	}
 638
 639	BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 640
 641	xen_free_irq(irq);
 642}
 643
 644/*
 645 * Do not make any assumptions regarding the relationship between the
 646 * IRQ number returned here and the Xen pirq argument.
 647 *
 648 * Note: We don't assign an event channel until the irq actually started
 649 * up.  Return an existing irq if we've already got one for the gsi.
 650 *
 651 * Shareable implies level triggered, not shareable implies edge
 652 * triggered here.
 653 */
 654int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 655			     unsigned pirq, int shareable, char *name)
 656{
 657	int irq = -1;
 658	struct physdev_irq irq_op;
 659	int ret;
 660
 661	mutex_lock(&irq_mapping_update_lock);
 662
 663	irq = xen_irq_from_gsi(gsi);
 664	if (irq != -1) {
 665		pr_info("%s: returning irq %d for gsi %u\n",
 666			__func__, irq, gsi);
 667		goto out;
 668	}
 669
 670	irq = xen_allocate_irq_gsi(gsi);
 671	if (irq < 0)
 672		goto out;
 673
 674	irq_op.irq = irq;
 675	irq_op.vector = 0;
 676
 677	/* Only the privileged domain can do this. For non-priv, the pcifront
 678	 * driver provides a PCI bus that does the call to do exactly
 679	 * this in the priv domain. */
 680	if (xen_initial_domain() &&
 681	    HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 682		xen_free_irq(irq);
 683		irq = -ENOSPC;
 684		goto out;
 685	}
 686
 687	ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 688			       shareable ? PIRQ_SHAREABLE : 0);
 689	if (ret < 0) {
 690		__unbind_from_irq(irq);
 691		irq = ret;
 692		goto out;
 693	}
 694
 695	pirq_query_unmask(irq);
 696	/* We try to use the handler with the appropriate semantic for the
 697	 * type of interrupt: if the interrupt is an edge triggered
 698	 * interrupt we use handle_edge_irq.
 699	 *
 700	 * On the other hand if the interrupt is level triggered we use
 701	 * handle_fasteoi_irq like the native code does for this kind of
 702	 * interrupts.
 703	 *
 704	 * Depending on the Xen version, pirq_needs_eoi might return true
 705	 * not only for level triggered interrupts but for edge triggered
 706	 * interrupts too. In any case Xen always honors the eoi mechanism,
 707	 * not injecting any more pirqs of the same kind if the first one
 708	 * hasn't received an eoi yet. Therefore using the fasteoi handler
 709	 * is the right choice either way.
 710	 */
 711	if (shareable)
 712		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 713				handle_fasteoi_irq, name);
 714	else
 715		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 716				handle_edge_irq, name);
 717
 718out:
 719	mutex_unlock(&irq_mapping_update_lock);
 720
 721	return irq;
 722}
 723
 724#ifdef CONFIG_PCI_MSI
 725int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 726{
 727	int rc;
 728	struct physdev_get_free_pirq op_get_free_pirq;
 729
 730	op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 731	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 732
 733	WARN_ONCE(rc == -ENOSYS,
 734		  "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 735
 736	return rc ? -1 : op_get_free_pirq.pirq;
 737}
 738
 739int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 740			     int pirq, int nvec, const char *name, domid_t domid)
 741{
 742	int i, irq, ret;
 743
 744	mutex_lock(&irq_mapping_update_lock);
 745
 746	irq = xen_allocate_irqs_dynamic(nvec);
 747	if (irq < 0)
 748		goto out;
 749
 750	for (i = 0; i < nvec; i++) {
 751		irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 752
 753		ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 754					      i == 0 ? 0 : PIRQ_MSI_GROUP);
 755		if (ret < 0)
 756			goto error_irq;
 757	}
 758
 759	ret = irq_set_msi_desc(irq, msidesc);
 760	if (ret < 0)
 761		goto error_irq;
 762out:
 763	mutex_unlock(&irq_mapping_update_lock);
 764	return irq;
 765error_irq:
 766	for (; i >= 0; i--)
 767		__unbind_from_irq(irq + i);
 768	mutex_unlock(&irq_mapping_update_lock);
 769	return ret;
 770}
 771#endif
 772
 773int xen_destroy_irq(int irq)
 774{
 775	struct physdev_unmap_pirq unmap_irq;
 776	struct irq_info *info = info_for_irq(irq);
 777	int rc = -ENOENT;
 778
 779	mutex_lock(&irq_mapping_update_lock);
 780
 781	/*
 782	 * If trying to remove a vector in a MSI group different
 783	 * than the first one skip the PIRQ unmap unless this vector
 784	 * is the first one in the group.
 785	 */
 786	if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 787		unmap_irq.pirq = info->u.pirq.pirq;
 788		unmap_irq.domid = info->u.pirq.domid;
 789		rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 790		/* If another domain quits without making the pci_disable_msix
 791		 * call, the Xen hypervisor takes care of freeing the PIRQs
 792		 * (free_domain_pirqs).
 793		 */
 794		if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 795			pr_info("domain %d does not have %d anymore\n",
 796				info->u.pirq.domid, info->u.pirq.pirq);
 797		else if (rc) {
 798			pr_warn("unmap irq failed %d\n", rc);
 799			goto out;
 800		}
 801	}
 802
 803	xen_free_irq(irq);
 804
 805out:
 806	mutex_unlock(&irq_mapping_update_lock);
 807	return rc;
 808}
 809
 810int xen_irq_from_pirq(unsigned pirq)
 811{
 812	int irq;
 813
 814	struct irq_info *info;
 815
 816	mutex_lock(&irq_mapping_update_lock);
 817
 818	list_for_each_entry(info, &xen_irq_list_head, list) {
 819		if (info->type != IRQT_PIRQ)
 820			continue;
 821		irq = info->irq;
 822		if (info->u.pirq.pirq == pirq)
 823			goto out;
 824	}
 825	irq = -1;
 826out:
 827	mutex_unlock(&irq_mapping_update_lock);
 828
 829	return irq;
 830}
 831
 832
 833int xen_pirq_from_irq(unsigned irq)
 834{
 835	return pirq_from_irq(irq);
 836}
 837EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 838
 839int bind_evtchn_to_irq(unsigned int evtchn)
 840{
 841	int irq;
 842	int ret;
 843
 844	if (evtchn >= xen_evtchn_max_channels())
 845		return -ENOMEM;
 846
 847	mutex_lock(&irq_mapping_update_lock);
 848
 849	irq = get_evtchn_to_irq(evtchn);
 850
 851	if (irq == -1) {
 852		irq = xen_allocate_irq_dynamic();
 853		if (irq < 0)
 854			goto out;
 855
 856		irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 857					      handle_edge_irq, "event");
 858
 859		ret = xen_irq_info_evtchn_setup(irq, evtchn);
 860		if (ret < 0) {
 861			__unbind_from_irq(irq);
 862			irq = ret;
 863			goto out;
 864		}
 865		/* New interdomain events are bound to VCPU 0. */
 866		bind_evtchn_to_cpu(evtchn, 0);
 867	} else {
 868		struct irq_info *info = info_for_irq(irq);
 869		WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 870	}
 871
 872out:
 873	mutex_unlock(&irq_mapping_update_lock);
 874
 875	return irq;
 876}
 877EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 878
 879static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 880{
 881	struct evtchn_bind_ipi bind_ipi;
 882	int evtchn, irq;
 883	int ret;
 884
 885	mutex_lock(&irq_mapping_update_lock);
 886
 887	irq = per_cpu(ipi_to_irq, cpu)[ipi];
 888
 889	if (irq == -1) {
 890		irq = xen_allocate_irq_dynamic();
 891		if (irq < 0)
 892			goto out;
 893
 894		irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 895					      handle_percpu_irq, "ipi");
 896
 897		bind_ipi.vcpu = cpu;
 898		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 899						&bind_ipi) != 0)
 900			BUG();
 901		evtchn = bind_ipi.port;
 902
 903		ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 904		if (ret < 0) {
 905			__unbind_from_irq(irq);
 906			irq = ret;
 907			goto out;
 908		}
 909		bind_evtchn_to_cpu(evtchn, cpu);
 910	} else {
 911		struct irq_info *info = info_for_irq(irq);
 912		WARN_ON(info == NULL || info->type != IRQT_IPI);
 913	}
 914
 915 out:
 916	mutex_unlock(&irq_mapping_update_lock);
 917	return irq;
 918}
 919
 920int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 921				   unsigned int remote_port)
 922{
 923	struct evtchn_bind_interdomain bind_interdomain;
 924	int err;
 925
 926	bind_interdomain.remote_dom  = remote_domain;
 927	bind_interdomain.remote_port = remote_port;
 928
 929	err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 930					  &bind_interdomain);
 931
 932	return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 933}
 934EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 935
 936static int find_virq(unsigned int virq, unsigned int cpu)
 937{
 938	struct evtchn_status status;
 939	int port, rc = -ENOENT;
 940
 941	memset(&status, 0, sizeof(status));
 942	for (port = 0; port < xen_evtchn_max_channels(); port++) {
 943		status.dom = DOMID_SELF;
 944		status.port = port;
 945		rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 946		if (rc < 0)
 947			continue;
 948		if (status.status != EVTCHNSTAT_virq)
 949			continue;
 950		if (status.u.virq == virq && status.vcpu == cpu) {
 951			rc = port;
 952			break;
 953		}
 954	}
 955	return rc;
 956}
 957
 958/**
 959 * xen_evtchn_nr_channels - number of usable event channel ports
 960 *
 961 * This may be less than the maximum supported by the current
 962 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 963 * supported.
 964 */
 965unsigned xen_evtchn_nr_channels(void)
 966{
 967        return evtchn_ops->nr_channels();
 968}
 969EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 970
 971int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 972{
 973	struct evtchn_bind_virq bind_virq;
 974	int evtchn, irq, ret;
 975
 976	mutex_lock(&irq_mapping_update_lock);
 977
 978	irq = per_cpu(virq_to_irq, cpu)[virq];
 979
 980	if (irq == -1) {
 981		irq = xen_allocate_irq_dynamic();
 982		if (irq < 0)
 983			goto out;
 984
 985		if (percpu)
 986			irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 987						      handle_percpu_irq, "virq");
 988		else
 989			irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 990						      handle_edge_irq, "virq");
 991
 992		bind_virq.virq = virq;
 993		bind_virq.vcpu = cpu;
 994		ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 995						&bind_virq);
 996		if (ret == 0)
 997			evtchn = bind_virq.port;
 998		else {
 999			if (ret == -EEXIST)
1000				ret = find_virq(virq, cpu);
1001			BUG_ON(ret < 0);
1002			evtchn = ret;
1003		}
1004
1005		ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1006		if (ret < 0) {
1007			__unbind_from_irq(irq);
1008			irq = ret;
1009			goto out;
1010		}
1011
1012		bind_evtchn_to_cpu(evtchn, cpu);
1013	} else {
1014		struct irq_info *info = info_for_irq(irq);
1015		WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1016	}
1017
1018out:
1019	mutex_unlock(&irq_mapping_update_lock);
1020
1021	return irq;
1022}
1023
1024static void unbind_from_irq(unsigned int irq)
1025{
1026	mutex_lock(&irq_mapping_update_lock);
1027	__unbind_from_irq(irq);
1028	mutex_unlock(&irq_mapping_update_lock);
1029}
1030
1031int bind_evtchn_to_irqhandler(unsigned int evtchn,
1032			      irq_handler_t handler,
1033			      unsigned long irqflags,
1034			      const char *devname, void *dev_id)
1035{
1036	int irq, retval;
1037
1038	irq = bind_evtchn_to_irq(evtchn);
1039	if (irq < 0)
1040		return irq;
1041	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1042	if (retval != 0) {
1043		unbind_from_irq(irq);
1044		return retval;
1045	}
1046
1047	return irq;
1048}
1049EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1050
1051int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1052					  unsigned int remote_port,
1053					  irq_handler_t handler,
1054					  unsigned long irqflags,
1055					  const char *devname,
1056					  void *dev_id)
1057{
1058	int irq, retval;
1059
1060	irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1061	if (irq < 0)
1062		return irq;
1063
1064	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1065	if (retval != 0) {
1066		unbind_from_irq(irq);
1067		return retval;
1068	}
1069
1070	return irq;
1071}
1072EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1073
1074int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1075			    irq_handler_t handler,
1076			    unsigned long irqflags, const char *devname, void *dev_id)
1077{
1078	int irq, retval;
1079
1080	irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1081	if (irq < 0)
1082		return irq;
1083	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1084	if (retval != 0) {
1085		unbind_from_irq(irq);
1086		return retval;
1087	}
1088
1089	return irq;
1090}
1091EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1092
1093int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1094			   unsigned int cpu,
1095			   irq_handler_t handler,
1096			   unsigned long irqflags,
1097			   const char *devname,
1098			   void *dev_id)
1099{
1100	int irq, retval;
1101
1102	irq = bind_ipi_to_irq(ipi, cpu);
1103	if (irq < 0)
1104		return irq;
1105
1106	irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1107	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1108	if (retval != 0) {
1109		unbind_from_irq(irq);
1110		return retval;
1111	}
1112
1113	return irq;
1114}
1115
1116void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1117{
1118	struct irq_info *info = irq_get_handler_data(irq);
1119
1120	if (WARN_ON(!info))
1121		return;
1122	free_irq(irq, dev_id);
1123	unbind_from_irq(irq);
1124}
1125EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1126
1127/**
1128 * xen_set_irq_priority() - set an event channel priority.
1129 * @irq:irq bound to an event channel.
1130 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1131 */
1132int xen_set_irq_priority(unsigned irq, unsigned priority)
1133{
1134	struct evtchn_set_priority set_priority;
1135
1136	set_priority.port = evtchn_from_irq(irq);
1137	set_priority.priority = priority;
1138
1139	return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1140					   &set_priority);
1141}
1142EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1143
1144int evtchn_make_refcounted(unsigned int evtchn)
1145{
1146	int irq = get_evtchn_to_irq(evtchn);
1147	struct irq_info *info;
1148
1149	if (irq == -1)
1150		return -ENOENT;
1151
1152	info = irq_get_handler_data(irq);
1153
1154	if (!info)
1155		return -ENOENT;
1156
1157	WARN_ON(info->refcnt != -1);
1158
1159	info->refcnt = 1;
1160
1161	return 0;
1162}
1163EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1164
1165int evtchn_get(unsigned int evtchn)
1166{
1167	int irq;
1168	struct irq_info *info;
1169	int err = -ENOENT;
1170
1171	if (evtchn >= xen_evtchn_max_channels())
1172		return -EINVAL;
1173
1174	mutex_lock(&irq_mapping_update_lock);
1175
1176	irq = get_evtchn_to_irq(evtchn);
1177	if (irq == -1)
1178		goto done;
1179
1180	info = irq_get_handler_data(irq);
1181
1182	if (!info)
1183		goto done;
1184
1185	err = -EINVAL;
1186	if (info->refcnt <= 0)
1187		goto done;
1188
1189	info->refcnt++;
1190	err = 0;
1191 done:
1192	mutex_unlock(&irq_mapping_update_lock);
1193
1194	return err;
1195}
1196EXPORT_SYMBOL_GPL(evtchn_get);
1197
1198void evtchn_put(unsigned int evtchn)
1199{
1200	int irq = get_evtchn_to_irq(evtchn);
1201	if (WARN_ON(irq == -1))
1202		return;
1203	unbind_from_irq(irq);
1204}
1205EXPORT_SYMBOL_GPL(evtchn_put);
1206
1207void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1208{
1209	int irq;
1210
1211#ifdef CONFIG_X86
1212	if (unlikely(vector == XEN_NMI_VECTOR)) {
1213		int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, cpu, NULL);
 
1214		if (rc < 0)
1215			printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1216		return;
1217	}
1218#endif
1219	irq = per_cpu(ipi_to_irq, cpu)[vector];
1220	BUG_ON(irq < 0);
1221	notify_remote_via_irq(irq);
1222}
1223
1224static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1225
1226static void __xen_evtchn_do_upcall(void)
1227{
1228	struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1229	int cpu = get_cpu();
1230	unsigned count;
1231
1232	do {
1233		vcpu_info->evtchn_upcall_pending = 0;
1234
1235		if (__this_cpu_inc_return(xed_nesting_count) - 1)
1236			goto out;
1237
1238		xen_evtchn_handle_events(cpu);
1239
1240		BUG_ON(!irqs_disabled());
1241
1242		count = __this_cpu_read(xed_nesting_count);
1243		__this_cpu_write(xed_nesting_count, 0);
1244	} while (count != 1 || vcpu_info->evtchn_upcall_pending);
1245
1246out:
1247
1248	put_cpu();
1249}
1250
1251void xen_evtchn_do_upcall(struct pt_regs *regs)
1252{
1253	struct pt_regs *old_regs = set_irq_regs(regs);
1254
1255	irq_enter();
1256#ifdef CONFIG_X86
1257	exit_idle();
1258	inc_irq_stat(irq_hv_callback_count);
1259#endif
1260
1261	__xen_evtchn_do_upcall();
1262
1263	irq_exit();
1264	set_irq_regs(old_regs);
1265}
1266
1267void xen_hvm_evtchn_do_upcall(void)
1268{
1269	__xen_evtchn_do_upcall();
1270}
1271EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1272
1273/* Rebind a new event channel to an existing irq. */
1274void rebind_evtchn_irq(int evtchn, int irq)
1275{
1276	struct irq_info *info = info_for_irq(irq);
1277
1278	if (WARN_ON(!info))
1279		return;
1280
1281	/* Make sure the irq is masked, since the new event channel
1282	   will also be masked. */
1283	disable_irq(irq);
1284
1285	mutex_lock(&irq_mapping_update_lock);
1286
1287	/* After resume the irq<->evtchn mappings are all cleared out */
1288	BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1289	/* Expect irq to have been bound before,
1290	   so there should be a proper type */
1291	BUG_ON(info->type == IRQT_UNBOUND);
1292
1293	(void)xen_irq_info_evtchn_setup(irq, evtchn);
1294
1295	mutex_unlock(&irq_mapping_update_lock);
1296
1297        bind_evtchn_to_cpu(evtchn, info->cpu);
1298	/* This will be deferred until interrupt is processed */
1299	irq_set_affinity(irq, cpumask_of(info->cpu));
1300
1301	/* Unmask the event channel. */
1302	enable_irq(irq);
1303}
1304
1305/* Rebind an evtchn so that it gets delivered to a specific cpu */
1306static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1307{
1308	struct evtchn_bind_vcpu bind_vcpu;
1309	int evtchn = evtchn_from_irq(irq);
1310	int masked;
1311
1312	if (!VALID_EVTCHN(evtchn))
1313		return -1;
1314
1315	if (!xen_support_evtchn_rebind())
1316		return -1;
1317
1318	/* Send future instances of this interrupt to other vcpu. */
1319	bind_vcpu.port = evtchn;
1320	bind_vcpu.vcpu = tcpu;
1321
1322	/*
1323	 * Mask the event while changing the VCPU binding to prevent
1324	 * it being delivered on an unexpected VCPU.
1325	 */
1326	masked = test_and_set_mask(evtchn);
1327
1328	/*
1329	 * If this fails, it usually just indicates that we're dealing with a
1330	 * virq or IPI channel, which don't actually need to be rebound. Ignore
1331	 * it, but don't do the xenlinux-level rebind in that case.
1332	 */
1333	if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1334		bind_evtchn_to_cpu(evtchn, tcpu);
1335
1336	if (!masked)
1337		unmask_evtchn(evtchn);
1338
1339	return 0;
1340}
 
1341
1342static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1343			    bool force)
1344{
1345	unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
 
 
 
 
1346
1347	return rebind_irq_to_cpu(data->irq, tcpu);
1348}
1349
1350static void enable_dynirq(struct irq_data *data)
1351{
1352	int evtchn = evtchn_from_irq(data->irq);
1353
1354	if (VALID_EVTCHN(evtchn))
1355		unmask_evtchn(evtchn);
1356}
1357
1358static void disable_dynirq(struct irq_data *data)
1359{
1360	int evtchn = evtchn_from_irq(data->irq);
1361
1362	if (VALID_EVTCHN(evtchn))
1363		mask_evtchn(evtchn);
1364}
1365
1366static void ack_dynirq(struct irq_data *data)
1367{
1368	int evtchn = evtchn_from_irq(data->irq);
1369
1370	if (!VALID_EVTCHN(evtchn))
1371		return;
1372
1373	if (unlikely(irqd_is_setaffinity_pending(data))) {
 
1374		int masked = test_and_set_mask(evtchn);
1375
1376		clear_evtchn(evtchn);
1377
1378		irq_move_masked_irq(data);
1379
1380		if (!masked)
1381			unmask_evtchn(evtchn);
1382	} else
1383		clear_evtchn(evtchn);
1384}
1385
1386static void mask_ack_dynirq(struct irq_data *data)
1387{
1388	disable_dynirq(data);
1389	ack_dynirq(data);
1390}
1391
1392static int retrigger_dynirq(struct irq_data *data)
1393{
1394	unsigned int evtchn = evtchn_from_irq(data->irq);
1395	int masked;
1396
1397	if (!VALID_EVTCHN(evtchn))
1398		return 0;
1399
1400	masked = test_and_set_mask(evtchn);
1401	set_evtchn(evtchn);
1402	if (!masked)
1403		unmask_evtchn(evtchn);
1404
1405	return 1;
1406}
1407
1408static void restore_pirqs(void)
1409{
1410	int pirq, rc, irq, gsi;
1411	struct physdev_map_pirq map_irq;
1412	struct irq_info *info;
1413
1414	list_for_each_entry(info, &xen_irq_list_head, list) {
1415		if (info->type != IRQT_PIRQ)
1416			continue;
1417
1418		pirq = info->u.pirq.pirq;
1419		gsi = info->u.pirq.gsi;
1420		irq = info->irq;
1421
1422		/* save/restore of PT devices doesn't work, so at this point the
1423		 * only devices present are GSI based emulated devices */
1424		if (!gsi)
1425			continue;
1426
1427		map_irq.domid = DOMID_SELF;
1428		map_irq.type = MAP_PIRQ_TYPE_GSI;
1429		map_irq.index = gsi;
1430		map_irq.pirq = pirq;
1431
1432		rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1433		if (rc) {
1434			pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1435				gsi, irq, pirq, rc);
1436			xen_free_irq(irq);
1437			continue;
1438		}
1439
1440		printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1441
1442		__startup_pirq(irq);
1443	}
1444}
1445
1446static void restore_cpu_virqs(unsigned int cpu)
1447{
1448	struct evtchn_bind_virq bind_virq;
1449	int virq, irq, evtchn;
1450
1451	for (virq = 0; virq < NR_VIRQS; virq++) {
1452		if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1453			continue;
1454
1455		BUG_ON(virq_from_irq(irq) != virq);
1456
1457		/* Get a new binding from Xen. */
1458		bind_virq.virq = virq;
1459		bind_virq.vcpu = cpu;
1460		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1461						&bind_virq) != 0)
1462			BUG();
1463		evtchn = bind_virq.port;
1464
1465		/* Record the new mapping. */
1466		(void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1467		bind_evtchn_to_cpu(evtchn, cpu);
1468	}
1469}
1470
1471static void restore_cpu_ipis(unsigned int cpu)
1472{
1473	struct evtchn_bind_ipi bind_ipi;
1474	int ipi, irq, evtchn;
1475
1476	for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1477		if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1478			continue;
1479
1480		BUG_ON(ipi_from_irq(irq) != ipi);
1481
1482		/* Get a new binding from Xen. */
1483		bind_ipi.vcpu = cpu;
1484		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1485						&bind_ipi) != 0)
1486			BUG();
1487		evtchn = bind_ipi.port;
1488
1489		/* Record the new mapping. */
1490		(void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1491		bind_evtchn_to_cpu(evtchn, cpu);
1492	}
1493}
1494
1495/* Clear an irq's pending state, in preparation for polling on it */
1496void xen_clear_irq_pending(int irq)
1497{
1498	int evtchn = evtchn_from_irq(irq);
1499
1500	if (VALID_EVTCHN(evtchn))
1501		clear_evtchn(evtchn);
1502}
1503EXPORT_SYMBOL(xen_clear_irq_pending);
1504void xen_set_irq_pending(int irq)
1505{
1506	int evtchn = evtchn_from_irq(irq);
1507
1508	if (VALID_EVTCHN(evtchn))
1509		set_evtchn(evtchn);
1510}
1511
1512bool xen_test_irq_pending(int irq)
1513{
1514	int evtchn = evtchn_from_irq(irq);
1515	bool ret = false;
1516
1517	if (VALID_EVTCHN(evtchn))
1518		ret = test_evtchn(evtchn);
1519
1520	return ret;
1521}
1522
1523/* Poll waiting for an irq to become pending with timeout.  In the usual case,
1524 * the irq will be disabled so it won't deliver an interrupt. */
1525void xen_poll_irq_timeout(int irq, u64 timeout)
1526{
1527	evtchn_port_t evtchn = evtchn_from_irq(irq);
1528
1529	if (VALID_EVTCHN(evtchn)) {
1530		struct sched_poll poll;
1531
1532		poll.nr_ports = 1;
1533		poll.timeout = timeout;
1534		set_xen_guest_handle(poll.ports, &evtchn);
1535
1536		if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1537			BUG();
1538	}
1539}
1540EXPORT_SYMBOL(xen_poll_irq_timeout);
1541/* Poll waiting for an irq to become pending.  In the usual case, the
1542 * irq will be disabled so it won't deliver an interrupt. */
1543void xen_poll_irq(int irq)
1544{
1545	xen_poll_irq_timeout(irq, 0 /* no timeout */);
1546}
1547
1548/* Check whether the IRQ line is shared with other guests. */
1549int xen_test_irq_shared(int irq)
1550{
1551	struct irq_info *info = info_for_irq(irq);
1552	struct physdev_irq_status_query irq_status;
1553
1554	if (WARN_ON(!info))
1555		return -ENOENT;
1556
1557	irq_status.irq = info->u.pirq.pirq;
1558
1559	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1560		return 0;
1561	return !(irq_status.flags & XENIRQSTAT_shared);
1562}
1563EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1564
1565void xen_irq_resume(void)
1566{
1567	unsigned int cpu;
1568	struct irq_info *info;
1569
1570	/* New event-channel space is not 'live' yet. */
1571	xen_evtchn_mask_all();
1572	xen_evtchn_resume();
1573
1574	/* No IRQ <-> event-channel mappings. */
1575	list_for_each_entry(info, &xen_irq_list_head, list)
1576		info->evtchn = 0; /* zap event-channel binding */
1577
1578	clear_evtchn_to_irq_all();
1579
1580	for_each_possible_cpu(cpu) {
1581		restore_cpu_virqs(cpu);
1582		restore_cpu_ipis(cpu);
1583	}
1584
1585	restore_pirqs();
1586}
1587
1588static struct irq_chip xen_dynamic_chip __read_mostly = {
1589	.name			= "xen-dyn",
1590
1591	.irq_disable		= disable_dynirq,
1592	.irq_mask		= disable_dynirq,
1593	.irq_unmask		= enable_dynirq,
1594
1595	.irq_ack		= ack_dynirq,
1596	.irq_mask_ack		= mask_ack_dynirq,
1597
1598	.irq_set_affinity	= set_affinity_irq,
1599	.irq_retrigger		= retrigger_dynirq,
1600};
1601
1602static struct irq_chip xen_pirq_chip __read_mostly = {
1603	.name			= "xen-pirq",
1604
1605	.irq_startup		= startup_pirq,
1606	.irq_shutdown		= shutdown_pirq,
1607	.irq_enable		= enable_pirq,
1608	.irq_disable		= disable_pirq,
1609
1610	.irq_mask		= disable_dynirq,
1611	.irq_unmask		= enable_dynirq,
1612
1613	.irq_ack		= eoi_pirq,
1614	.irq_eoi		= eoi_pirq,
1615	.irq_mask_ack		= mask_ack_pirq,
1616
1617	.irq_set_affinity	= set_affinity_irq,
1618
1619	.irq_retrigger		= retrigger_dynirq,
1620};
1621
1622static struct irq_chip xen_percpu_chip __read_mostly = {
1623	.name			= "xen-percpu",
1624
1625	.irq_disable		= disable_dynirq,
1626	.irq_mask		= disable_dynirq,
1627	.irq_unmask		= enable_dynirq,
1628
1629	.irq_ack		= ack_dynirq,
1630};
1631
1632int xen_set_callback_via(uint64_t via)
1633{
1634	struct xen_hvm_param a;
1635	a.domid = DOMID_SELF;
1636	a.index = HVM_PARAM_CALLBACK_IRQ;
1637	a.value = via;
1638	return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1639}
1640EXPORT_SYMBOL_GPL(xen_set_callback_via);
1641
1642#ifdef CONFIG_XEN_PVHVM
1643/* Vector callbacks are better than PCI interrupts to receive event
1644 * channel notifications because we can receive vector callbacks on any
1645 * vcpu and we don't need PCI support or APIC interactions. */
1646void xen_callback_vector(void)
1647{
1648	int rc;
1649	uint64_t callback_via;
 
1650	if (xen_have_vector_callback) {
1651		callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1652		rc = xen_set_callback_via(callback_via);
1653		if (rc) {
1654			pr_err("Request for Xen HVM callback vector failed\n");
1655			xen_have_vector_callback = 0;
1656			return;
1657		}
1658		pr_info("Xen HVM callback vector for event delivery is enabled\n");
1659		/* in the restore case the vector has already been allocated */
1660		if (!test_bit(HYPERVISOR_CALLBACK_VECTOR, used_vectors))
1661			alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1662					xen_hvm_callback_vector);
1663	}
1664}
1665#else
1666void xen_callback_vector(void) {}
1667#endif
1668
1669#undef MODULE_PARAM_PREFIX
1670#define MODULE_PARAM_PREFIX "xen."
1671
1672static bool fifo_events = true;
1673module_param(fifo_events, bool, 0);
1674
1675void __init xen_init_IRQ(void)
1676{
1677	int ret = -EINVAL;
 
1678
1679	if (fifo_events)
1680		ret = xen_evtchn_fifo_init();
1681	if (ret < 0)
1682		xen_evtchn_2l_init();
1683
1684	evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1685				sizeof(*evtchn_to_irq), GFP_KERNEL);
1686	BUG_ON(!evtchn_to_irq);
1687
1688	/* No event channels are 'live' right now. */
1689	xen_evtchn_mask_all();
 
1690
1691	pirq_needs_eoi = pirq_needs_eoi_flag;
1692
1693#ifdef CONFIG_X86
1694	if (xen_pv_domain()) {
1695		irq_ctx_init(smp_processor_id());
1696		if (xen_initial_domain())
1697			pci_xen_initial_domain();
1698	}
1699	if (xen_feature(XENFEAT_hvm_callback_vector))
1700		xen_callback_vector();
1701
1702	if (xen_hvm_domain()) {
1703		native_init_IRQ();
1704		/* pci_xen_hvm_init must be called after native_init_IRQ so that
1705		 * __acpi_register_gsi can point at the right function */
1706		pci_xen_hvm_init();
1707	} else {
1708		int rc;
1709		struct physdev_pirq_eoi_gmfn eoi_gmfn;
1710
1711		pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1712		eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1713		rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
1714		/* TODO: No PVH support for PIRQ EOI */
1715		if (rc != 0) {
1716			free_page((unsigned long) pirq_eoi_map);
1717			pirq_eoi_map = NULL;
1718		} else
1719			pirq_needs_eoi = pirq_check_eoi_map;
1720	}
1721#endif
1722}
   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#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  25
  26#include <linux/linkage.h>
  27#include <linux/interrupt.h>
  28#include <linux/irq.h>
  29#include <linux/moduleparam.h>
  30#include <linux/string.h>
  31#include <linux/bootmem.h>
  32#include <linux/slab.h>
  33#include <linux/irqnr.h>
  34#include <linux/pci.h>
  35
  36#ifdef CONFIG_X86
  37#include <asm/desc.h>
  38#include <asm/ptrace.h>
  39#include <asm/irq.h>
 
  40#include <asm/io_apic.h>
  41#include <asm/i8259.h>
  42#include <asm/xen/pci.h>
  43#endif
  44#include <asm/sync_bitops.h>
  45#include <asm/xen/hypercall.h>
  46#include <asm/xen/hypervisor.h>
  47#include <xen/page.h>
  48
  49#include <xen/xen.h>
  50#include <xen/hvm.h>
  51#include <xen/xen-ops.h>
  52#include <xen/events.h>
  53#include <xen/interface/xen.h>
  54#include <xen/interface/event_channel.h>
  55#include <xen/interface/hvm/hvm_op.h>
  56#include <xen/interface/hvm/params.h>
  57#include <xen/interface/physdev.h>
  58#include <xen/interface/sched.h>
  59#include <xen/interface/vcpu.h>
  60#include <asm/hw_irq.h>
  61
  62#include "events_internal.h"
  63
  64const struct evtchn_ops *evtchn_ops;
  65
  66/*
  67 * This lock protects updates to the following mapping and reference-count
  68 * arrays. The lock does not need to be acquired to read the mapping tables.
  69 */
  70static DEFINE_MUTEX(irq_mapping_update_lock);
  71
  72static LIST_HEAD(xen_irq_list_head);
  73
  74/* IRQ <-> VIRQ mapping. */
  75static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  76
  77/* IRQ <-> IPI mapping */
  78static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  79
  80int **evtchn_to_irq;
  81#ifdef CONFIG_X86
  82static unsigned long *pirq_eoi_map;
  83#endif
  84static bool (*pirq_needs_eoi)(unsigned irq);
  85
  86#define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  87#define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
  88#define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
  89
  90/* Xen will never allocate port zero for any purpose. */
  91#define VALID_EVTCHN(chn)	((chn) != 0)
  92
  93static struct irq_chip xen_dynamic_chip;
  94static struct irq_chip xen_percpu_chip;
  95static struct irq_chip xen_pirq_chip;
  96static void enable_dynirq(struct irq_data *data);
  97static void disable_dynirq(struct irq_data *data);
  98
  99static void clear_evtchn_to_irq_row(unsigned row)
 100{
 101	unsigned col;
 102
 103	for (col = 0; col < EVTCHN_PER_ROW; col++)
 104		evtchn_to_irq[row][col] = -1;
 105}
 106
 107static void clear_evtchn_to_irq_all(void)
 108{
 109	unsigned row;
 110
 111	for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 112		if (evtchn_to_irq[row] == NULL)
 113			continue;
 114		clear_evtchn_to_irq_row(row);
 115	}
 116}
 117
 118static int set_evtchn_to_irq(unsigned evtchn, unsigned irq)
 119{
 120	unsigned row;
 121	unsigned col;
 122
 123	if (evtchn >= xen_evtchn_max_channels())
 124		return -EINVAL;
 125
 126	row = EVTCHN_ROW(evtchn);
 127	col = EVTCHN_COL(evtchn);
 128
 129	if (evtchn_to_irq[row] == NULL) {
 130		/* Unallocated irq entries return -1 anyway */
 131		if (irq == -1)
 132			return 0;
 133
 134		evtchn_to_irq[row] = (int *)get_zeroed_page(GFP_KERNEL);
 135		if (evtchn_to_irq[row] == NULL)
 136			return -ENOMEM;
 137
 138		clear_evtchn_to_irq_row(row);
 139	}
 140
 141	evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)] = irq;
 142	return 0;
 143}
 144
 145int get_evtchn_to_irq(unsigned evtchn)
 146{
 147	if (evtchn >= xen_evtchn_max_channels())
 148		return -1;
 149	if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 150		return -1;
 151	return evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)];
 152}
 153
 154/* Get info for IRQ */
 155struct irq_info *info_for_irq(unsigned irq)
 156{
 157	return irq_get_handler_data(irq);
 158}
 159
 160/* Constructors for packed IRQ information. */
 161static int xen_irq_info_common_setup(struct irq_info *info,
 162				     unsigned irq,
 163				     enum xen_irq_type type,
 164				     unsigned evtchn,
 165				     unsigned short cpu)
 166{
 167	int ret;
 168
 169	BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 170
 171	info->type = type;
 172	info->irq = irq;
 173	info->evtchn = evtchn;
 174	info->cpu = cpu;
 175
 176	ret = set_evtchn_to_irq(evtchn, irq);
 177	if (ret < 0)
 178		return ret;
 179
 180	irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 181
 182	return xen_evtchn_port_setup(info);
 183}
 184
 185static int xen_irq_info_evtchn_setup(unsigned irq,
 186				     unsigned evtchn)
 187{
 188	struct irq_info *info = info_for_irq(irq);
 189
 190	return xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 191}
 192
 193static int xen_irq_info_ipi_setup(unsigned cpu,
 194				  unsigned irq,
 195				  unsigned evtchn,
 196				  enum ipi_vector ipi)
 197{
 198	struct irq_info *info = info_for_irq(irq);
 199
 200	info->u.ipi = ipi;
 201
 202	per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 203
 204	return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 205}
 206
 207static int xen_irq_info_virq_setup(unsigned cpu,
 208				   unsigned irq,
 209				   unsigned evtchn,
 210				   unsigned virq)
 211{
 212	struct irq_info *info = info_for_irq(irq);
 213
 214	info->u.virq = virq;
 215
 216	per_cpu(virq_to_irq, cpu)[virq] = irq;
 217
 218	return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 219}
 220
 221static int xen_irq_info_pirq_setup(unsigned irq,
 222				   unsigned evtchn,
 223				   unsigned pirq,
 224				   unsigned gsi,
 225				   uint16_t domid,
 226				   unsigned char flags)
 227{
 228	struct irq_info *info = info_for_irq(irq);
 229
 230	info->u.pirq.pirq = pirq;
 231	info->u.pirq.gsi = gsi;
 232	info->u.pirq.domid = domid;
 233	info->u.pirq.flags = flags;
 234
 235	return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 236}
 237
 238static void xen_irq_info_cleanup(struct irq_info *info)
 239{
 240	set_evtchn_to_irq(info->evtchn, -1);
 241	info->evtchn = 0;
 242}
 243
 244/*
 245 * Accessors for packed IRQ information.
 246 */
 247unsigned int evtchn_from_irq(unsigned irq)
 248{
 249	if (unlikely(WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 250		return 0;
 251
 252	return info_for_irq(irq)->evtchn;
 253}
 254
 255unsigned irq_from_evtchn(unsigned int evtchn)
 256{
 257	return get_evtchn_to_irq(evtchn);
 258}
 259EXPORT_SYMBOL_GPL(irq_from_evtchn);
 260
 261int irq_from_virq(unsigned int cpu, unsigned int virq)
 262{
 263	return per_cpu(virq_to_irq, cpu)[virq];
 264}
 265
 266static enum ipi_vector ipi_from_irq(unsigned irq)
 267{
 268	struct irq_info *info = info_for_irq(irq);
 269
 270	BUG_ON(info == NULL);
 271	BUG_ON(info->type != IRQT_IPI);
 272
 273	return info->u.ipi;
 274}
 275
 276static unsigned virq_from_irq(unsigned irq)
 277{
 278	struct irq_info *info = info_for_irq(irq);
 279
 280	BUG_ON(info == NULL);
 281	BUG_ON(info->type != IRQT_VIRQ);
 282
 283	return info->u.virq;
 284}
 285
 286static unsigned pirq_from_irq(unsigned irq)
 287{
 288	struct irq_info *info = info_for_irq(irq);
 289
 290	BUG_ON(info == NULL);
 291	BUG_ON(info->type != IRQT_PIRQ);
 292
 293	return info->u.pirq.pirq;
 294}
 295
 296static enum xen_irq_type type_from_irq(unsigned irq)
 297{
 298	return info_for_irq(irq)->type;
 299}
 300
 301unsigned cpu_from_irq(unsigned irq)
 302{
 303	return info_for_irq(irq)->cpu;
 304}
 305
 306unsigned int cpu_from_evtchn(unsigned int evtchn)
 307{
 308	int irq = get_evtchn_to_irq(evtchn);
 309	unsigned ret = 0;
 310
 311	if (irq != -1)
 312		ret = cpu_from_irq(irq);
 313
 314	return ret;
 315}
 316
 317#ifdef CONFIG_X86
 318static bool pirq_check_eoi_map(unsigned irq)
 319{
 320	return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 321}
 322#endif
 323
 324static bool pirq_needs_eoi_flag(unsigned irq)
 325{
 326	struct irq_info *info = info_for_irq(irq);
 327	BUG_ON(info->type != IRQT_PIRQ);
 328
 329	return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 330}
 331
 332static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 333{
 334	int irq = get_evtchn_to_irq(chn);
 335	struct irq_info *info = info_for_irq(irq);
 336
 337	BUG_ON(irq == -1);
 338#ifdef CONFIG_SMP
 339	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 340#endif
 341	xen_evtchn_port_bind_to_cpu(info, cpu);
 342
 343	info->cpu = cpu;
 344}
 345
 
 
 
 
 
 
 
 
 346/**
 347 * notify_remote_via_irq - send event to remote end of event channel via irq
 348 * @irq: irq of event channel to send event to
 349 *
 350 * Unlike notify_remote_via_evtchn(), this is safe to use across
 351 * save/restore. Notifications on a broken connection are silently
 352 * dropped.
 353 */
 354void notify_remote_via_irq(int irq)
 355{
 356	int evtchn = evtchn_from_irq(irq);
 357
 358	if (VALID_EVTCHN(evtchn))
 359		notify_remote_via_evtchn(evtchn);
 360}
 361EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 362
 363static void xen_irq_init(unsigned irq)
 364{
 365	struct irq_info *info;
 366#ifdef CONFIG_SMP
 367	/* By default all event channels notify CPU#0. */
 368	cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(0));
 369#endif
 370
 371	info = kzalloc(sizeof(*info), GFP_KERNEL);
 372	if (info == NULL)
 373		panic("Unable to allocate metadata for IRQ%d\n", irq);
 374
 375	info->type = IRQT_UNBOUND;
 376	info->refcnt = -1;
 377
 378	irq_set_handler_data(irq, info);
 379
 380	list_add_tail(&info->list, &xen_irq_list_head);
 381}
 382
 383static int __must_check xen_allocate_irqs_dynamic(int nvec)
 384{
 385	int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 386
 387	if (irq >= 0) {
 388		for (i = 0; i < nvec; i++)
 389			xen_irq_init(irq + i);
 390	}
 391
 392	return irq;
 393}
 394
 395static inline int __must_check xen_allocate_irq_dynamic(void)
 396{
 397
 398	return xen_allocate_irqs_dynamic(1);
 399}
 400
 401static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 402{
 403	int irq;
 404
 405	/*
 406	 * A PV guest has no concept of a GSI (since it has no ACPI
 407	 * nor access to/knowledge of the physical APICs). Therefore
 408	 * all IRQs are dynamically allocated from the entire IRQ
 409	 * space.
 410	 */
 411	if (xen_pv_domain() && !xen_initial_domain())
 412		return xen_allocate_irq_dynamic();
 413
 414	/* Legacy IRQ descriptors are already allocated by the arch. */
 415	if (gsi < nr_legacy_irqs())
 416		irq = gsi;
 417	else
 418		irq = irq_alloc_desc_at(gsi, -1);
 419
 420	xen_irq_init(irq);
 421
 422	return irq;
 423}
 424
 425static void xen_free_irq(unsigned irq)
 426{
 427	struct irq_info *info = irq_get_handler_data(irq);
 428
 429	if (WARN_ON(!info))
 430		return;
 431
 432	list_del(&info->list);
 433
 434	irq_set_handler_data(irq, NULL);
 435
 436	WARN_ON(info->refcnt > 0);
 437
 438	kfree(info);
 439
 440	/* Legacy IRQ descriptors are managed by the arch. */
 441	if (irq < nr_legacy_irqs())
 442		return;
 443
 444	irq_free_desc(irq);
 445}
 446
 447static void xen_evtchn_close(unsigned int port)
 448{
 449	struct evtchn_close close;
 450
 451	close.port = port;
 452	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 453		BUG();
 454}
 455
 456static void pirq_query_unmask(int irq)
 457{
 458	struct physdev_irq_status_query irq_status;
 459	struct irq_info *info = info_for_irq(irq);
 460
 461	BUG_ON(info->type != IRQT_PIRQ);
 462
 463	irq_status.irq = pirq_from_irq(irq);
 464	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 465		irq_status.flags = 0;
 466
 467	info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 468	if (irq_status.flags & XENIRQSTAT_needs_eoi)
 469		info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 470}
 471
 472static void eoi_pirq(struct irq_data *data)
 473{
 474	int evtchn = evtchn_from_irq(data->irq);
 475	struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 476	int rc = 0;
 477
 478	if (!VALID_EVTCHN(evtchn))
 479		return;
 480
 481	if (unlikely(irqd_is_setaffinity_pending(data)) &&
 482	    likely(!irqd_irq_disabled(data))) {
 483		int masked = test_and_set_mask(evtchn);
 484
 485		clear_evtchn(evtchn);
 486
 487		irq_move_masked_irq(data);
 488
 489		if (!masked)
 490			unmask_evtchn(evtchn);
 491	} else
 492		clear_evtchn(evtchn);
 493
 494	if (pirq_needs_eoi(data->irq)) {
 495		rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 496		WARN_ON(rc);
 497	}
 498}
 499
 500static void mask_ack_pirq(struct irq_data *data)
 501{
 502	disable_dynirq(data);
 503	eoi_pirq(data);
 504}
 505
 506static unsigned int __startup_pirq(unsigned int irq)
 507{
 508	struct evtchn_bind_pirq bind_pirq;
 509	struct irq_info *info = info_for_irq(irq);
 510	int evtchn = evtchn_from_irq(irq);
 511	int rc;
 512
 513	BUG_ON(info->type != IRQT_PIRQ);
 514
 515	if (VALID_EVTCHN(evtchn))
 516		goto out;
 517
 518	bind_pirq.pirq = pirq_from_irq(irq);
 519	/* NB. We are happy to share unless we are probing. */
 520	bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 521					BIND_PIRQ__WILL_SHARE : 0;
 522	rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 523	if (rc != 0) {
 524		pr_warn("Failed to obtain physical IRQ %d\n", irq);
 525		return 0;
 526	}
 527	evtchn = bind_pirq.port;
 528
 529	pirq_query_unmask(irq);
 530
 531	rc = set_evtchn_to_irq(evtchn, irq);
 532	if (rc)
 533		goto err;
 534
 535	info->evtchn = evtchn;
 536	bind_evtchn_to_cpu(evtchn, 0);
 537
 538	rc = xen_evtchn_port_setup(info);
 539	if (rc)
 540		goto err;
 541
 542out:
 543	unmask_evtchn(evtchn);
 544	eoi_pirq(irq_get_irq_data(irq));
 545
 546	return 0;
 547
 548err:
 549	pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 550	xen_evtchn_close(evtchn);
 551	return 0;
 552}
 553
 554static unsigned int startup_pirq(struct irq_data *data)
 555{
 556	return __startup_pirq(data->irq);
 557}
 558
 559static void shutdown_pirq(struct irq_data *data)
 560{
 561	unsigned int irq = data->irq;
 562	struct irq_info *info = info_for_irq(irq);
 563	unsigned evtchn = evtchn_from_irq(irq);
 564
 565	BUG_ON(info->type != IRQT_PIRQ);
 566
 567	if (!VALID_EVTCHN(evtchn))
 568		return;
 569
 570	mask_evtchn(evtchn);
 571	xen_evtchn_close(evtchn);
 572	xen_irq_info_cleanup(info);
 573}
 574
 575static void enable_pirq(struct irq_data *data)
 576{
 577	enable_dynirq(data);
 578}
 579
 580static void disable_pirq(struct irq_data *data)
 581{
 582	disable_dynirq(data);
 583}
 584
 585int xen_irq_from_gsi(unsigned gsi)
 586{
 587	struct irq_info *info;
 588
 589	list_for_each_entry(info, &xen_irq_list_head, list) {
 590		if (info->type != IRQT_PIRQ)
 591			continue;
 592
 593		if (info->u.pirq.gsi == gsi)
 594			return info->irq;
 595	}
 596
 597	return -1;
 598}
 599EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 600
 601static void __unbind_from_irq(unsigned int irq)
 602{
 603	int evtchn = evtchn_from_irq(irq);
 604	struct irq_info *info = irq_get_handler_data(irq);
 605
 606	if (info->refcnt > 0) {
 607		info->refcnt--;
 608		if (info->refcnt != 0)
 609			return;
 610	}
 611
 612	if (VALID_EVTCHN(evtchn)) {
 613		unsigned int cpu = cpu_from_irq(irq);
 614
 615		xen_evtchn_close(evtchn);
 616
 617		switch (type_from_irq(irq)) {
 618		case IRQT_VIRQ:
 619			per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 620			break;
 621		case IRQT_IPI:
 622			per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 623			break;
 624		default:
 625			break;
 626		}
 627
 628		xen_irq_info_cleanup(info);
 629	}
 630
 631	BUG_ON(info_for_irq(irq)->type == IRQT_UNBOUND);
 632
 633	xen_free_irq(irq);
 634}
 635
 636/*
 637 * Do not make any assumptions regarding the relationship between the
 638 * IRQ number returned here and the Xen pirq argument.
 639 *
 640 * Note: We don't assign an event channel until the irq actually started
 641 * up.  Return an existing irq if we've already got one for the gsi.
 642 *
 643 * Shareable implies level triggered, not shareable implies edge
 644 * triggered here.
 645 */
 646int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 647			     unsigned pirq, int shareable, char *name)
 648{
 649	int irq = -1;
 650	struct physdev_irq irq_op;
 651	int ret;
 652
 653	mutex_lock(&irq_mapping_update_lock);
 654
 655	irq = xen_irq_from_gsi(gsi);
 656	if (irq != -1) {
 657		pr_info("%s: returning irq %d for gsi %u\n",
 658			__func__, irq, gsi);
 659		goto out;
 660	}
 661
 662	irq = xen_allocate_irq_gsi(gsi);
 663	if (irq < 0)
 664		goto out;
 665
 666	irq_op.irq = irq;
 667	irq_op.vector = 0;
 668
 669	/* Only the privileged domain can do this. For non-priv, the pcifront
 670	 * driver provides a PCI bus that does the call to do exactly
 671	 * this in the priv domain. */
 672	if (xen_initial_domain() &&
 673	    HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 674		xen_free_irq(irq);
 675		irq = -ENOSPC;
 676		goto out;
 677	}
 678
 679	ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
 680			       shareable ? PIRQ_SHAREABLE : 0);
 681	if (ret < 0) {
 682		__unbind_from_irq(irq);
 683		irq = ret;
 684		goto out;
 685	}
 686
 687	pirq_query_unmask(irq);
 688	/* We try to use the handler with the appropriate semantic for the
 689	 * type of interrupt: if the interrupt is an edge triggered
 690	 * interrupt we use handle_edge_irq.
 691	 *
 692	 * On the other hand if the interrupt is level triggered we use
 693	 * handle_fasteoi_irq like the native code does for this kind of
 694	 * interrupts.
 695	 *
 696	 * Depending on the Xen version, pirq_needs_eoi might return true
 697	 * not only for level triggered interrupts but for edge triggered
 698	 * interrupts too. In any case Xen always honors the eoi mechanism,
 699	 * not injecting any more pirqs of the same kind if the first one
 700	 * hasn't received an eoi yet. Therefore using the fasteoi handler
 701	 * is the right choice either way.
 702	 */
 703	if (shareable)
 704		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 705				handle_fasteoi_irq, name);
 706	else
 707		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
 708				handle_edge_irq, name);
 709
 710out:
 711	mutex_unlock(&irq_mapping_update_lock);
 712
 713	return irq;
 714}
 715
 716#ifdef CONFIG_PCI_MSI
 717int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
 718{
 719	int rc;
 720	struct physdev_get_free_pirq op_get_free_pirq;
 721
 722	op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
 723	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 724
 725	WARN_ONCE(rc == -ENOSYS,
 726		  "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
 727
 728	return rc ? -1 : op_get_free_pirq.pirq;
 729}
 730
 731int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
 732			     int pirq, int nvec, const char *name, domid_t domid)
 733{
 734	int i, irq, ret;
 735
 736	mutex_lock(&irq_mapping_update_lock);
 737
 738	irq = xen_allocate_irqs_dynamic(nvec);
 739	if (irq < 0)
 740		goto out;
 741
 742	for (i = 0; i < nvec; i++) {
 743		irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
 744
 745		ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
 746					      i == 0 ? 0 : PIRQ_MSI_GROUP);
 747		if (ret < 0)
 748			goto error_irq;
 749	}
 750
 751	ret = irq_set_msi_desc(irq, msidesc);
 752	if (ret < 0)
 753		goto error_irq;
 754out:
 755	mutex_unlock(&irq_mapping_update_lock);
 756	return irq;
 757error_irq:
 758	while (nvec--)
 759		__unbind_from_irq(irq + nvec);
 760	mutex_unlock(&irq_mapping_update_lock);
 761	return ret;
 762}
 763#endif
 764
 765int xen_destroy_irq(int irq)
 766{
 767	struct physdev_unmap_pirq unmap_irq;
 768	struct irq_info *info = info_for_irq(irq);
 769	int rc = -ENOENT;
 770
 771	mutex_lock(&irq_mapping_update_lock);
 772
 773	/*
 774	 * If trying to remove a vector in a MSI group different
 775	 * than the first one skip the PIRQ unmap unless this vector
 776	 * is the first one in the group.
 777	 */
 778	if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
 779		unmap_irq.pirq = info->u.pirq.pirq;
 780		unmap_irq.domid = info->u.pirq.domid;
 781		rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 782		/* If another domain quits without making the pci_disable_msix
 783		 * call, the Xen hypervisor takes care of freeing the PIRQs
 784		 * (free_domain_pirqs).
 785		 */
 786		if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
 787			pr_info("domain %d does not have %d anymore\n",
 788				info->u.pirq.domid, info->u.pirq.pirq);
 789		else if (rc) {
 790			pr_warn("unmap irq failed %d\n", rc);
 791			goto out;
 792		}
 793	}
 794
 795	xen_free_irq(irq);
 796
 797out:
 798	mutex_unlock(&irq_mapping_update_lock);
 799	return rc;
 800}
 801
 802int xen_irq_from_pirq(unsigned pirq)
 803{
 804	int irq;
 805
 806	struct irq_info *info;
 807
 808	mutex_lock(&irq_mapping_update_lock);
 809
 810	list_for_each_entry(info, &xen_irq_list_head, list) {
 811		if (info->type != IRQT_PIRQ)
 812			continue;
 813		irq = info->irq;
 814		if (info->u.pirq.pirq == pirq)
 815			goto out;
 816	}
 817	irq = -1;
 818out:
 819	mutex_unlock(&irq_mapping_update_lock);
 820
 821	return irq;
 822}
 823
 824
 825int xen_pirq_from_irq(unsigned irq)
 826{
 827	return pirq_from_irq(irq);
 828}
 829EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
 830
 831int bind_evtchn_to_irq(unsigned int evtchn)
 832{
 833	int irq;
 834	int ret;
 835
 836	if (evtchn >= xen_evtchn_max_channels())
 837		return -ENOMEM;
 838
 839	mutex_lock(&irq_mapping_update_lock);
 840
 841	irq = get_evtchn_to_irq(evtchn);
 842
 843	if (irq == -1) {
 844		irq = xen_allocate_irq_dynamic();
 845		if (irq < 0)
 846			goto out;
 847
 848		irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 849					      handle_edge_irq, "event");
 850
 851		ret = xen_irq_info_evtchn_setup(irq, evtchn);
 852		if (ret < 0) {
 853			__unbind_from_irq(irq);
 854			irq = ret;
 855			goto out;
 856		}
 857		/* New interdomain events are bound to VCPU 0. */
 858		bind_evtchn_to_cpu(evtchn, 0);
 859	} else {
 860		struct irq_info *info = info_for_irq(irq);
 861		WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
 862	}
 863
 864out:
 865	mutex_unlock(&irq_mapping_update_lock);
 866
 867	return irq;
 868}
 869EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 870
 871static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 872{
 873	struct evtchn_bind_ipi bind_ipi;
 874	int evtchn, irq;
 875	int ret;
 876
 877	mutex_lock(&irq_mapping_update_lock);
 878
 879	irq = per_cpu(ipi_to_irq, cpu)[ipi];
 880
 881	if (irq == -1) {
 882		irq = xen_allocate_irq_dynamic();
 883		if (irq < 0)
 884			goto out;
 885
 886		irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 887					      handle_percpu_irq, "ipi");
 888
 889		bind_ipi.vcpu = xen_vcpu_nr(cpu);
 890		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 891						&bind_ipi) != 0)
 892			BUG();
 893		evtchn = bind_ipi.port;
 894
 895		ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
 896		if (ret < 0) {
 897			__unbind_from_irq(irq);
 898			irq = ret;
 899			goto out;
 900		}
 901		bind_evtchn_to_cpu(evtchn, cpu);
 902	} else {
 903		struct irq_info *info = info_for_irq(irq);
 904		WARN_ON(info == NULL || info->type != IRQT_IPI);
 905	}
 906
 907 out:
 908	mutex_unlock(&irq_mapping_update_lock);
 909	return irq;
 910}
 911
 912int bind_interdomain_evtchn_to_irq(unsigned int remote_domain,
 913				   unsigned int remote_port)
 914{
 915	struct evtchn_bind_interdomain bind_interdomain;
 916	int err;
 917
 918	bind_interdomain.remote_dom  = remote_domain;
 919	bind_interdomain.remote_port = remote_port;
 920
 921	err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
 922					  &bind_interdomain);
 923
 924	return err ? : bind_evtchn_to_irq(bind_interdomain.local_port);
 925}
 926EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq);
 927
 928static int find_virq(unsigned int virq, unsigned int cpu)
 929{
 930	struct evtchn_status status;
 931	int port, rc = -ENOENT;
 932
 933	memset(&status, 0, sizeof(status));
 934	for (port = 0; port < xen_evtchn_max_channels(); port++) {
 935		status.dom = DOMID_SELF;
 936		status.port = port;
 937		rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
 938		if (rc < 0)
 939			continue;
 940		if (status.status != EVTCHNSTAT_virq)
 941			continue;
 942		if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
 943			rc = port;
 944			break;
 945		}
 946	}
 947	return rc;
 948}
 949
 950/**
 951 * xen_evtchn_nr_channels - number of usable event channel ports
 952 *
 953 * This may be less than the maximum supported by the current
 954 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
 955 * supported.
 956 */
 957unsigned xen_evtchn_nr_channels(void)
 958{
 959        return evtchn_ops->nr_channels();
 960}
 961EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
 962
 963int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
 964{
 965	struct evtchn_bind_virq bind_virq;
 966	int evtchn, irq, ret;
 967
 968	mutex_lock(&irq_mapping_update_lock);
 969
 970	irq = per_cpu(virq_to_irq, cpu)[virq];
 971
 972	if (irq == -1) {
 973		irq = xen_allocate_irq_dynamic();
 974		if (irq < 0)
 975			goto out;
 976
 977		if (percpu)
 978			irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
 979						      handle_percpu_irq, "virq");
 980		else
 981			irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
 982						      handle_edge_irq, "virq");
 983
 984		bind_virq.virq = virq;
 985		bind_virq.vcpu = xen_vcpu_nr(cpu);
 986		ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 987						&bind_virq);
 988		if (ret == 0)
 989			evtchn = bind_virq.port;
 990		else {
 991			if (ret == -EEXIST)
 992				ret = find_virq(virq, cpu);
 993			BUG_ON(ret < 0);
 994			evtchn = ret;
 995		}
 996
 997		ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
 998		if (ret < 0) {
 999			__unbind_from_irq(irq);
1000			irq = ret;
1001			goto out;
1002		}
1003
1004		bind_evtchn_to_cpu(evtchn, cpu);
1005	} else {
1006		struct irq_info *info = info_for_irq(irq);
1007		WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1008	}
1009
1010out:
1011	mutex_unlock(&irq_mapping_update_lock);
1012
1013	return irq;
1014}
1015
1016static void unbind_from_irq(unsigned int irq)
1017{
1018	mutex_lock(&irq_mapping_update_lock);
1019	__unbind_from_irq(irq);
1020	mutex_unlock(&irq_mapping_update_lock);
1021}
1022
1023int bind_evtchn_to_irqhandler(unsigned int evtchn,
1024			      irq_handler_t handler,
1025			      unsigned long irqflags,
1026			      const char *devname, void *dev_id)
1027{
1028	int irq, retval;
1029
1030	irq = bind_evtchn_to_irq(evtchn);
1031	if (irq < 0)
1032		return irq;
1033	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1034	if (retval != 0) {
1035		unbind_from_irq(irq);
1036		return retval;
1037	}
1038
1039	return irq;
1040}
1041EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1042
1043int bind_interdomain_evtchn_to_irqhandler(unsigned int remote_domain,
1044					  unsigned int remote_port,
1045					  irq_handler_t handler,
1046					  unsigned long irqflags,
1047					  const char *devname,
1048					  void *dev_id)
1049{
1050	int irq, retval;
1051
1052	irq = bind_interdomain_evtchn_to_irq(remote_domain, remote_port);
1053	if (irq < 0)
1054		return irq;
1055
1056	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1057	if (retval != 0) {
1058		unbind_from_irq(irq);
1059		return retval;
1060	}
1061
1062	return irq;
1063}
1064EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler);
1065
1066int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1067			    irq_handler_t handler,
1068			    unsigned long irqflags, const char *devname, void *dev_id)
1069{
1070	int irq, retval;
1071
1072	irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1073	if (irq < 0)
1074		return irq;
1075	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1076	if (retval != 0) {
1077		unbind_from_irq(irq);
1078		return retval;
1079	}
1080
1081	return irq;
1082}
1083EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1084
1085int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1086			   unsigned int cpu,
1087			   irq_handler_t handler,
1088			   unsigned long irqflags,
1089			   const char *devname,
1090			   void *dev_id)
1091{
1092	int irq, retval;
1093
1094	irq = bind_ipi_to_irq(ipi, cpu);
1095	if (irq < 0)
1096		return irq;
1097
1098	irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1099	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1100	if (retval != 0) {
1101		unbind_from_irq(irq);
1102		return retval;
1103	}
1104
1105	return irq;
1106}
1107
1108void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1109{
1110	struct irq_info *info = irq_get_handler_data(irq);
1111
1112	if (WARN_ON(!info))
1113		return;
1114	free_irq(irq, dev_id);
1115	unbind_from_irq(irq);
1116}
1117EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1118
1119/**
1120 * xen_set_irq_priority() - set an event channel priority.
1121 * @irq:irq bound to an event channel.
1122 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1123 */
1124int xen_set_irq_priority(unsigned irq, unsigned priority)
1125{
1126	struct evtchn_set_priority set_priority;
1127
1128	set_priority.port = evtchn_from_irq(irq);
1129	set_priority.priority = priority;
1130
1131	return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1132					   &set_priority);
1133}
1134EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1135
1136int evtchn_make_refcounted(unsigned int evtchn)
1137{
1138	int irq = get_evtchn_to_irq(evtchn);
1139	struct irq_info *info;
1140
1141	if (irq == -1)
1142		return -ENOENT;
1143
1144	info = irq_get_handler_data(irq);
1145
1146	if (!info)
1147		return -ENOENT;
1148
1149	WARN_ON(info->refcnt != -1);
1150
1151	info->refcnt = 1;
1152
1153	return 0;
1154}
1155EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1156
1157int evtchn_get(unsigned int evtchn)
1158{
1159	int irq;
1160	struct irq_info *info;
1161	int err = -ENOENT;
1162
1163	if (evtchn >= xen_evtchn_max_channels())
1164		return -EINVAL;
1165
1166	mutex_lock(&irq_mapping_update_lock);
1167
1168	irq = get_evtchn_to_irq(evtchn);
1169	if (irq == -1)
1170		goto done;
1171
1172	info = irq_get_handler_data(irq);
1173
1174	if (!info)
1175		goto done;
1176
1177	err = -EINVAL;
1178	if (info->refcnt <= 0)
1179		goto done;
1180
1181	info->refcnt++;
1182	err = 0;
1183 done:
1184	mutex_unlock(&irq_mapping_update_lock);
1185
1186	return err;
1187}
1188EXPORT_SYMBOL_GPL(evtchn_get);
1189
1190void evtchn_put(unsigned int evtchn)
1191{
1192	int irq = get_evtchn_to_irq(evtchn);
1193	if (WARN_ON(irq == -1))
1194		return;
1195	unbind_from_irq(irq);
1196}
1197EXPORT_SYMBOL_GPL(evtchn_put);
1198
1199void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1200{
1201	int irq;
1202
1203#ifdef CONFIG_X86
1204	if (unlikely(vector == XEN_NMI_VECTOR)) {
1205		int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1206					     NULL);
1207		if (rc < 0)
1208			printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1209		return;
1210	}
1211#endif
1212	irq = per_cpu(ipi_to_irq, cpu)[vector];
1213	BUG_ON(irq < 0);
1214	notify_remote_via_irq(irq);
1215}
1216
1217static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1218
1219static void __xen_evtchn_do_upcall(void)
1220{
1221	struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1222	int cpu = get_cpu();
1223	unsigned count;
1224
1225	do {
1226		vcpu_info->evtchn_upcall_pending = 0;
1227
1228		if (__this_cpu_inc_return(xed_nesting_count) - 1)
1229			goto out;
1230
1231		xen_evtchn_handle_events(cpu);
1232
1233		BUG_ON(!irqs_disabled());
1234
1235		count = __this_cpu_read(xed_nesting_count);
1236		__this_cpu_write(xed_nesting_count, 0);
1237	} while (count != 1 || vcpu_info->evtchn_upcall_pending);
1238
1239out:
1240
1241	put_cpu();
1242}
1243
1244void xen_evtchn_do_upcall(struct pt_regs *regs)
1245{
1246	struct pt_regs *old_regs = set_irq_regs(regs);
1247
1248	irq_enter();
1249#ifdef CONFIG_X86
 
1250	inc_irq_stat(irq_hv_callback_count);
1251#endif
1252
1253	__xen_evtchn_do_upcall();
1254
1255	irq_exit();
1256	set_irq_regs(old_regs);
1257}
1258
1259void xen_hvm_evtchn_do_upcall(void)
1260{
1261	__xen_evtchn_do_upcall();
1262}
1263EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1264
1265/* Rebind a new event channel to an existing irq. */
1266void rebind_evtchn_irq(int evtchn, int irq)
1267{
1268	struct irq_info *info = info_for_irq(irq);
1269
1270	if (WARN_ON(!info))
1271		return;
1272
1273	/* Make sure the irq is masked, since the new event channel
1274	   will also be masked. */
1275	disable_irq(irq);
1276
1277	mutex_lock(&irq_mapping_update_lock);
1278
1279	/* After resume the irq<->evtchn mappings are all cleared out */
1280	BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1281	/* Expect irq to have been bound before,
1282	   so there should be a proper type */
1283	BUG_ON(info->type == IRQT_UNBOUND);
1284
1285	(void)xen_irq_info_evtchn_setup(irq, evtchn);
1286
1287	mutex_unlock(&irq_mapping_update_lock);
1288
1289        bind_evtchn_to_cpu(evtchn, info->cpu);
1290	/* This will be deferred until interrupt is processed */
1291	irq_set_affinity(irq, cpumask_of(info->cpu));
1292
1293	/* Unmask the event channel. */
1294	enable_irq(irq);
1295}
1296
1297/* Rebind an evtchn so that it gets delivered to a specific cpu */
1298int xen_rebind_evtchn_to_cpu(int evtchn, unsigned tcpu)
1299{
1300	struct evtchn_bind_vcpu bind_vcpu;
 
1301	int masked;
1302
1303	if (!VALID_EVTCHN(evtchn))
1304		return -1;
1305
1306	if (!xen_support_evtchn_rebind())
1307		return -1;
1308
1309	/* Send future instances of this interrupt to other vcpu. */
1310	bind_vcpu.port = evtchn;
1311	bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1312
1313	/*
1314	 * Mask the event while changing the VCPU binding to prevent
1315	 * it being delivered on an unexpected VCPU.
1316	 */
1317	masked = test_and_set_mask(evtchn);
1318
1319	/*
1320	 * If this fails, it usually just indicates that we're dealing with a
1321	 * virq or IPI channel, which don't actually need to be rebound. Ignore
1322	 * it, but don't do the xenlinux-level rebind in that case.
1323	 */
1324	if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1325		bind_evtchn_to_cpu(evtchn, tcpu);
1326
1327	if (!masked)
1328		unmask_evtchn(evtchn);
1329
1330	return 0;
1331}
1332EXPORT_SYMBOL_GPL(xen_rebind_evtchn_to_cpu);
1333
1334static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1335			    bool force)
1336{
1337	unsigned tcpu = cpumask_first_and(dest, cpu_online_mask);
1338	int ret = xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu);
1339
1340	if (!ret)
1341		irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1342
1343	return ret;
1344}
1345
1346static void enable_dynirq(struct irq_data *data)
1347{
1348	int evtchn = evtchn_from_irq(data->irq);
1349
1350	if (VALID_EVTCHN(evtchn))
1351		unmask_evtchn(evtchn);
1352}
1353
1354static void disable_dynirq(struct irq_data *data)
1355{
1356	int evtchn = evtchn_from_irq(data->irq);
1357
1358	if (VALID_EVTCHN(evtchn))
1359		mask_evtchn(evtchn);
1360}
1361
1362static void ack_dynirq(struct irq_data *data)
1363{
1364	int evtchn = evtchn_from_irq(data->irq);
1365
1366	if (!VALID_EVTCHN(evtchn))
1367		return;
1368
1369	if (unlikely(irqd_is_setaffinity_pending(data)) &&
1370	    likely(!irqd_irq_disabled(data))) {
1371		int masked = test_and_set_mask(evtchn);
1372
1373		clear_evtchn(evtchn);
1374
1375		irq_move_masked_irq(data);
1376
1377		if (!masked)
1378			unmask_evtchn(evtchn);
1379	} else
1380		clear_evtchn(evtchn);
1381}
1382
1383static void mask_ack_dynirq(struct irq_data *data)
1384{
1385	disable_dynirq(data);
1386	ack_dynirq(data);
1387}
1388
1389static int retrigger_dynirq(struct irq_data *data)
1390{
1391	unsigned int evtchn = evtchn_from_irq(data->irq);
1392	int masked;
1393
1394	if (!VALID_EVTCHN(evtchn))
1395		return 0;
1396
1397	masked = test_and_set_mask(evtchn);
1398	set_evtchn(evtchn);
1399	if (!masked)
1400		unmask_evtchn(evtchn);
1401
1402	return 1;
1403}
1404
1405static void restore_pirqs(void)
1406{
1407	int pirq, rc, irq, gsi;
1408	struct physdev_map_pirq map_irq;
1409	struct irq_info *info;
1410
1411	list_for_each_entry(info, &xen_irq_list_head, list) {
1412		if (info->type != IRQT_PIRQ)
1413			continue;
1414
1415		pirq = info->u.pirq.pirq;
1416		gsi = info->u.pirq.gsi;
1417		irq = info->irq;
1418
1419		/* save/restore of PT devices doesn't work, so at this point the
1420		 * only devices present are GSI based emulated devices */
1421		if (!gsi)
1422			continue;
1423
1424		map_irq.domid = DOMID_SELF;
1425		map_irq.type = MAP_PIRQ_TYPE_GSI;
1426		map_irq.index = gsi;
1427		map_irq.pirq = pirq;
1428
1429		rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1430		if (rc) {
1431			pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1432				gsi, irq, pirq, rc);
1433			xen_free_irq(irq);
1434			continue;
1435		}
1436
1437		printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1438
1439		__startup_pirq(irq);
1440	}
1441}
1442
1443static void restore_cpu_virqs(unsigned int cpu)
1444{
1445	struct evtchn_bind_virq bind_virq;
1446	int virq, irq, evtchn;
1447
1448	for (virq = 0; virq < NR_VIRQS; virq++) {
1449		if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1450			continue;
1451
1452		BUG_ON(virq_from_irq(irq) != virq);
1453
1454		/* Get a new binding from Xen. */
1455		bind_virq.virq = virq;
1456		bind_virq.vcpu = xen_vcpu_nr(cpu);
1457		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1458						&bind_virq) != 0)
1459			BUG();
1460		evtchn = bind_virq.port;
1461
1462		/* Record the new mapping. */
1463		(void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1464		bind_evtchn_to_cpu(evtchn, cpu);
1465	}
1466}
1467
1468static void restore_cpu_ipis(unsigned int cpu)
1469{
1470	struct evtchn_bind_ipi bind_ipi;
1471	int ipi, irq, evtchn;
1472
1473	for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1474		if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1475			continue;
1476
1477		BUG_ON(ipi_from_irq(irq) != ipi);
1478
1479		/* Get a new binding from Xen. */
1480		bind_ipi.vcpu = xen_vcpu_nr(cpu);
1481		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1482						&bind_ipi) != 0)
1483			BUG();
1484		evtchn = bind_ipi.port;
1485
1486		/* Record the new mapping. */
1487		(void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1488		bind_evtchn_to_cpu(evtchn, cpu);
1489	}
1490}
1491
1492/* Clear an irq's pending state, in preparation for polling on it */
1493void xen_clear_irq_pending(int irq)
1494{
1495	int evtchn = evtchn_from_irq(irq);
1496
1497	if (VALID_EVTCHN(evtchn))
1498		clear_evtchn(evtchn);
1499}
1500EXPORT_SYMBOL(xen_clear_irq_pending);
1501void xen_set_irq_pending(int irq)
1502{
1503	int evtchn = evtchn_from_irq(irq);
1504
1505	if (VALID_EVTCHN(evtchn))
1506		set_evtchn(evtchn);
1507}
1508
1509bool xen_test_irq_pending(int irq)
1510{
1511	int evtchn = evtchn_from_irq(irq);
1512	bool ret = false;
1513
1514	if (VALID_EVTCHN(evtchn))
1515		ret = test_evtchn(evtchn);
1516
1517	return ret;
1518}
1519
1520/* Poll waiting for an irq to become pending with timeout.  In the usual case,
1521 * the irq will be disabled so it won't deliver an interrupt. */
1522void xen_poll_irq_timeout(int irq, u64 timeout)
1523{
1524	evtchn_port_t evtchn = evtchn_from_irq(irq);
1525
1526	if (VALID_EVTCHN(evtchn)) {
1527		struct sched_poll poll;
1528
1529		poll.nr_ports = 1;
1530		poll.timeout = timeout;
1531		set_xen_guest_handle(poll.ports, &evtchn);
1532
1533		if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1534			BUG();
1535	}
1536}
1537EXPORT_SYMBOL(xen_poll_irq_timeout);
1538/* Poll waiting for an irq to become pending.  In the usual case, the
1539 * irq will be disabled so it won't deliver an interrupt. */
1540void xen_poll_irq(int irq)
1541{
1542	xen_poll_irq_timeout(irq, 0 /* no timeout */);
1543}
1544
1545/* Check whether the IRQ line is shared with other guests. */
1546int xen_test_irq_shared(int irq)
1547{
1548	struct irq_info *info = info_for_irq(irq);
1549	struct physdev_irq_status_query irq_status;
1550
1551	if (WARN_ON(!info))
1552		return -ENOENT;
1553
1554	irq_status.irq = info->u.pirq.pirq;
1555
1556	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
1557		return 0;
1558	return !(irq_status.flags & XENIRQSTAT_shared);
1559}
1560EXPORT_SYMBOL_GPL(xen_test_irq_shared);
1561
1562void xen_irq_resume(void)
1563{
1564	unsigned int cpu;
1565	struct irq_info *info;
1566
1567	/* New event-channel space is not 'live' yet. */
 
1568	xen_evtchn_resume();
1569
1570	/* No IRQ <-> event-channel mappings. */
1571	list_for_each_entry(info, &xen_irq_list_head, list)
1572		info->evtchn = 0; /* zap event-channel binding */
1573
1574	clear_evtchn_to_irq_all();
1575
1576	for_each_possible_cpu(cpu) {
1577		restore_cpu_virqs(cpu);
1578		restore_cpu_ipis(cpu);
1579	}
1580
1581	restore_pirqs();
1582}
1583
1584static struct irq_chip xen_dynamic_chip __read_mostly = {
1585	.name			= "xen-dyn",
1586
1587	.irq_disable		= disable_dynirq,
1588	.irq_mask		= disable_dynirq,
1589	.irq_unmask		= enable_dynirq,
1590
1591	.irq_ack		= ack_dynirq,
1592	.irq_mask_ack		= mask_ack_dynirq,
1593
1594	.irq_set_affinity	= set_affinity_irq,
1595	.irq_retrigger		= retrigger_dynirq,
1596};
1597
1598static struct irq_chip xen_pirq_chip __read_mostly = {
1599	.name			= "xen-pirq",
1600
1601	.irq_startup		= startup_pirq,
1602	.irq_shutdown		= shutdown_pirq,
1603	.irq_enable		= enable_pirq,
1604	.irq_disable		= disable_pirq,
1605
1606	.irq_mask		= disable_dynirq,
1607	.irq_unmask		= enable_dynirq,
1608
1609	.irq_ack		= eoi_pirq,
1610	.irq_eoi		= eoi_pirq,
1611	.irq_mask_ack		= mask_ack_pirq,
1612
1613	.irq_set_affinity	= set_affinity_irq,
1614
1615	.irq_retrigger		= retrigger_dynirq,
1616};
1617
1618static struct irq_chip xen_percpu_chip __read_mostly = {
1619	.name			= "xen-percpu",
1620
1621	.irq_disable		= disable_dynirq,
1622	.irq_mask		= disable_dynirq,
1623	.irq_unmask		= enable_dynirq,
1624
1625	.irq_ack		= ack_dynirq,
1626};
1627
1628int xen_set_callback_via(uint64_t via)
1629{
1630	struct xen_hvm_param a;
1631	a.domid = DOMID_SELF;
1632	a.index = HVM_PARAM_CALLBACK_IRQ;
1633	a.value = via;
1634	return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1635}
1636EXPORT_SYMBOL_GPL(xen_set_callback_via);
1637
1638#ifdef CONFIG_XEN_PVHVM
1639/* Vector callbacks are better than PCI interrupts to receive event
1640 * channel notifications because we can receive vector callbacks on any
1641 * vcpu and we don't need PCI support or APIC interactions. */
1642void xen_callback_vector(void)
1643{
1644	int rc;
1645	uint64_t callback_via;
1646
1647	if (xen_have_vector_callback) {
1648		callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
1649		rc = xen_set_callback_via(callback_via);
1650		if (rc) {
1651			pr_err("Request for Xen HVM callback vector failed\n");
1652			xen_have_vector_callback = 0;
1653			return;
1654		}
1655		pr_info("Xen HVM callback vector for event delivery is enabled\n");
1656		alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR,
1657				xen_hvm_callback_vector);
 
 
1658	}
1659}
1660#else
1661void xen_callback_vector(void) {}
1662#endif
1663
1664#undef MODULE_PARAM_PREFIX
1665#define MODULE_PARAM_PREFIX "xen."
1666
1667static bool fifo_events = true;
1668module_param(fifo_events, bool, 0);
1669
1670void __init xen_init_IRQ(void)
1671{
1672	int ret = -EINVAL;
1673	unsigned int evtchn;
1674
1675	if (fifo_events)
1676		ret = xen_evtchn_fifo_init();
1677	if (ret < 0)
1678		xen_evtchn_2l_init();
1679
1680	evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
1681				sizeof(*evtchn_to_irq), GFP_KERNEL);
1682	BUG_ON(!evtchn_to_irq);
1683
1684	/* No event channels are 'live' right now. */
1685	for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
1686		mask_evtchn(evtchn);
1687
1688	pirq_needs_eoi = pirq_needs_eoi_flag;
1689
1690#ifdef CONFIG_X86
1691	if (xen_pv_domain()) {
1692		irq_ctx_init(smp_processor_id());
1693		if (xen_initial_domain())
1694			pci_xen_initial_domain();
1695	}
1696	if (xen_feature(XENFEAT_hvm_callback_vector))
1697		xen_callback_vector();
1698
1699	if (xen_hvm_domain()) {
1700		native_init_IRQ();
1701		/* pci_xen_hvm_init must be called after native_init_IRQ so that
1702		 * __acpi_register_gsi can point at the right function */
1703		pci_xen_hvm_init();
1704	} else {
1705		int rc;
1706		struct physdev_pirq_eoi_gmfn eoi_gmfn;
1707
1708		pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
1709		eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
1710		rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
 
1711		if (rc != 0) {
1712			free_page((unsigned long) pirq_eoi_map);
1713			pirq_eoi_map = NULL;
1714		} else
1715			pirq_needs_eoi = pirq_check_eoi_map;
1716	}
1717#endif
1718}