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// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Xen event channels
   4 *
   5 * Xen models interrupts with abstract event channels.  Because each
   6 * domain gets 1024 event channels, but NR_IRQ is not that large, we
   7 * must dynamically map irqs<->event channels.  The event channels
   8 * interface with the rest of the kernel by defining a xen interrupt
   9 * chip.  When an event is received, it is mapped to an irq and sent
  10 * through the normal interrupt processing path.
  11 *
  12 * There are four kinds of events which can be mapped to an event
  13 * channel:
  14 *
  15 * 1. Inter-domain notifications.  This includes all the virtual
  16 *    device events, since they're driven by front-ends in another domain
  17 *    (typically dom0).
  18 * 2. VIRQs, typically used for timers.  These are per-cpu events.
  19 * 3. IPIs.
  20 * 4. PIRQs - Hardware interrupts.
  21 *
  22 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  23 */
  24
  25#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  26
  27#include <linux/linkage.h>
  28#include <linux/interrupt.h>
  29#include <linux/irq.h>
  30#include <linux/moduleparam.h>
  31#include <linux/string.h>
  32#include <linux/memblock.h>
  33#include <linux/slab.h>
  34#include <linux/irqnr.h>
  35#include <linux/pci.h>
  36#include <linux/spinlock.h>
  37#include <linux/cpuhotplug.h>
  38#include <linux/atomic.h>
  39#include <linux/ktime.h>
  40
  41#ifdef CONFIG_X86
  42#include <asm/desc.h>
  43#include <asm/ptrace.h>
  44#include <asm/idtentry.h>
  45#include <asm/irq.h>
  46#include <asm/io_apic.h>
  47#include <asm/i8259.h>
  48#include <asm/xen/pci.h>
  49#endif
  50#include <asm/sync_bitops.h>
  51#include <asm/xen/hypercall.h>
  52#include <asm/xen/hypervisor.h>
  53#include <xen/page.h>
  54
  55#include <xen/xen.h>
  56#include <xen/hvm.h>
  57#include <xen/xen-ops.h>
  58#include <xen/events.h>
  59#include <xen/interface/xen.h>
  60#include <xen/interface/event_channel.h>
  61#include <xen/interface/hvm/hvm_op.h>
  62#include <xen/interface/hvm/params.h>
  63#include <xen/interface/physdev.h>
  64#include <xen/interface/sched.h>
  65#include <xen/interface/vcpu.h>
  66#include <xen/xenbus.h>
  67#include <asm/hw_irq.h>
  68
  69#include "events_internal.h"
  70
  71#undef MODULE_PARAM_PREFIX
  72#define MODULE_PARAM_PREFIX "xen."
  73
  74/* Interrupt types. */
  75enum xen_irq_type {
  76	IRQT_UNBOUND = 0,
  77	IRQT_PIRQ,
  78	IRQT_VIRQ,
  79	IRQT_IPI,
  80	IRQT_EVTCHN
  81};
  82
  83/*
  84 * Packed IRQ information:
  85 * type - enum xen_irq_type
  86 * event channel - irq->event channel mapping
  87 * cpu - cpu this event channel is bound to
  88 * index - type-specific information:
  89 *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
  90 *           guest, or GSI (real passthrough IRQ) of the device.
  91 *    VIRQ - virq number
  92 *    IPI - IPI vector
  93 *    EVTCHN -
  94 */
  95struct irq_info {
  96	struct list_head list;
  97	struct list_head eoi_list;
  98	short refcnt;
  99	u8 spurious_cnt;
 100	u8 is_accounted;
 101	short type;		/* type: IRQT_* */
 102	u8 mask_reason;		/* Why is event channel masked */
 103#define EVT_MASK_REASON_EXPLICIT	0x01
 104#define EVT_MASK_REASON_TEMPORARY	0x02
 105#define EVT_MASK_REASON_EOI_PENDING	0x04
 106	u8 is_active;		/* Is event just being handled? */
 107	unsigned irq;
 108	evtchn_port_t evtchn;   /* event channel */
 109	unsigned short cpu;     /* cpu bound */
 110	unsigned short eoi_cpu; /* EOI must happen on this cpu-1 */
 111	unsigned int irq_epoch; /* If eoi_cpu valid: irq_epoch of event */
 112	u64 eoi_time;           /* Time in jiffies when to EOI. */
 113	raw_spinlock_t lock;
 114
 115	union {
 116		unsigned short virq;
 117		enum ipi_vector ipi;
 118		struct {
 119			unsigned short pirq;
 120			unsigned short gsi;
 121			unsigned char vector;
 122			unsigned char flags;
 123			uint16_t domid;
 124		} pirq;
 125		struct xenbus_device *interdomain;
 126	} u;
 127};
 128
 129#define PIRQ_NEEDS_EOI	(1 << 0)
 130#define PIRQ_SHAREABLE	(1 << 1)
 131#define PIRQ_MSI_GROUP	(1 << 2)
 132
 133static uint __read_mostly event_loop_timeout = 2;
 134module_param(event_loop_timeout, uint, 0644);
 135
 136static uint __read_mostly event_eoi_delay = 10;
 137module_param(event_eoi_delay, uint, 0644);
 138
 139const struct evtchn_ops *evtchn_ops;
 140
 141/*
 142 * This lock protects updates to the following mapping and reference-count
 143 * arrays. The lock does not need to be acquired to read the mapping tables.
 144 */
 145static DEFINE_MUTEX(irq_mapping_update_lock);
 146
 147/*
 148 * Lock protecting event handling loop against removing event channels.
 149 * Adding of event channels is no issue as the associated IRQ becomes active
 150 * only after everything is setup (before request_[threaded_]irq() the handler
 151 * can't be entered for an event, as the event channel will be unmasked only
 152 * then).
 153 */
 154static DEFINE_RWLOCK(evtchn_rwlock);
 155
 156/*
 157 * Lock hierarchy:
 158 *
 159 * irq_mapping_update_lock
 160 *   evtchn_rwlock
 161 *     IRQ-desc lock
 162 *       percpu eoi_list_lock
 163 *         irq_info->lock
 164 */
 165
 166static LIST_HEAD(xen_irq_list_head);
 167
 168/* IRQ <-> VIRQ mapping. */
 169static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
 170
 171/* IRQ <-> IPI mapping */
 172static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
 173
 174/* Event channel distribution data */
 175static atomic_t channels_on_cpu[NR_CPUS];
 176
 177static int **evtchn_to_irq;
 178#ifdef CONFIG_X86
 179static unsigned long *pirq_eoi_map;
 180#endif
 181static bool (*pirq_needs_eoi)(unsigned irq);
 182
 183#define EVTCHN_ROW(e)  (e / (PAGE_SIZE/sizeof(**evtchn_to_irq)))
 184#define EVTCHN_COL(e)  (e % (PAGE_SIZE/sizeof(**evtchn_to_irq)))
 185#define EVTCHN_PER_ROW (PAGE_SIZE / sizeof(**evtchn_to_irq))
 186
 187/* Xen will never allocate port zero for any purpose. */
 188#define VALID_EVTCHN(chn)	((chn) != 0)
 189
 190static struct irq_info *legacy_info_ptrs[NR_IRQS_LEGACY];
 191
 192static struct irq_chip xen_dynamic_chip;
 193static struct irq_chip xen_lateeoi_chip;
 194static struct irq_chip xen_percpu_chip;
 195static struct irq_chip xen_pirq_chip;
 196static void enable_dynirq(struct irq_data *data);
 197static void disable_dynirq(struct irq_data *data);
 198
 199static DEFINE_PER_CPU(unsigned int, irq_epoch);
 200
 201static void clear_evtchn_to_irq_row(int *evtchn_row)
 202{
 203	unsigned col;
 204
 205	for (col = 0; col < EVTCHN_PER_ROW; col++)
 206		WRITE_ONCE(evtchn_row[col], -1);
 207}
 208
 209static void clear_evtchn_to_irq_all(void)
 210{
 211	unsigned row;
 212
 213	for (row = 0; row < EVTCHN_ROW(xen_evtchn_max_channels()); row++) {
 214		if (evtchn_to_irq[row] == NULL)
 215			continue;
 216		clear_evtchn_to_irq_row(evtchn_to_irq[row]);
 217	}
 218}
 219
 220static int set_evtchn_to_irq(evtchn_port_t evtchn, unsigned int irq)
 221{
 222	unsigned row;
 223	unsigned col;
 224	int *evtchn_row;
 225
 226	if (evtchn >= xen_evtchn_max_channels())
 227		return -EINVAL;
 228
 229	row = EVTCHN_ROW(evtchn);
 230	col = EVTCHN_COL(evtchn);
 231
 232	if (evtchn_to_irq[row] == NULL) {
 233		/* Unallocated irq entries return -1 anyway */
 234		if (irq == -1)
 235			return 0;
 236
 237		evtchn_row = (int *) __get_free_pages(GFP_KERNEL, 0);
 238		if (evtchn_row == NULL)
 239			return -ENOMEM;
 240
 241		clear_evtchn_to_irq_row(evtchn_row);
 242
 243		/*
 244		 * We've prepared an empty row for the mapping. If a different
 245		 * thread was faster inserting it, we can drop ours.
 246		 */
 247		if (cmpxchg(&evtchn_to_irq[row], NULL, evtchn_row) != NULL)
 248			free_page((unsigned long) evtchn_row);
 249	}
 250
 251	WRITE_ONCE(evtchn_to_irq[row][col], irq);
 252	return 0;
 253}
 254
 255int get_evtchn_to_irq(evtchn_port_t evtchn)
 256{
 257	if (evtchn >= xen_evtchn_max_channels())
 258		return -1;
 259	if (evtchn_to_irq[EVTCHN_ROW(evtchn)] == NULL)
 260		return -1;
 261	return READ_ONCE(evtchn_to_irq[EVTCHN_ROW(evtchn)][EVTCHN_COL(evtchn)]);
 262}
 263
 264/* Get info for IRQ */
 265static struct irq_info *info_for_irq(unsigned irq)
 266{
 267	if (irq < nr_legacy_irqs())
 268		return legacy_info_ptrs[irq];
 269	else
 270		return irq_get_chip_data(irq);
 271}
 272
 273static void set_info_for_irq(unsigned int irq, struct irq_info *info)
 274{
 275	if (irq < nr_legacy_irqs())
 276		legacy_info_ptrs[irq] = info;
 277	else
 278		irq_set_chip_data(irq, info);
 279}
 280
 281/* Per CPU channel accounting */
 282static void channels_on_cpu_dec(struct irq_info *info)
 283{
 284	if (!info->is_accounted)
 285		return;
 286
 287	info->is_accounted = 0;
 288
 289	if (WARN_ON_ONCE(info->cpu >= nr_cpu_ids))
 290		return;
 291
 292	WARN_ON_ONCE(!atomic_add_unless(&channels_on_cpu[info->cpu], -1 , 0));
 293}
 294
 295static void channels_on_cpu_inc(struct irq_info *info)
 296{
 297	if (WARN_ON_ONCE(info->cpu >= nr_cpu_ids))
 298		return;
 299
 300	if (WARN_ON_ONCE(!atomic_add_unless(&channels_on_cpu[info->cpu], 1,
 301					    INT_MAX)))
 302		return;
 303
 304	info->is_accounted = 1;
 305}
 306
 307/* Constructors for packed IRQ information. */
 308static int xen_irq_info_common_setup(struct irq_info *info,
 309				     unsigned irq,
 310				     enum xen_irq_type type,
 311				     evtchn_port_t evtchn,
 312				     unsigned short cpu)
 313{
 314	int ret;
 315
 316	BUG_ON(info->type != IRQT_UNBOUND && info->type != type);
 317
 318	info->type = type;
 319	info->irq = irq;
 320	info->evtchn = evtchn;
 321	info->cpu = cpu;
 322	info->mask_reason = EVT_MASK_REASON_EXPLICIT;
 323	raw_spin_lock_init(&info->lock);
 324
 325	ret = set_evtchn_to_irq(evtchn, irq);
 326	if (ret < 0)
 327		return ret;
 328
 329	irq_clear_status_flags(irq, IRQ_NOREQUEST|IRQ_NOAUTOEN);
 330
 331	return xen_evtchn_port_setup(evtchn);
 332}
 333
 334static int xen_irq_info_evtchn_setup(unsigned irq,
 335				     evtchn_port_t evtchn,
 336				     struct xenbus_device *dev)
 337{
 338	struct irq_info *info = info_for_irq(irq);
 339	int ret;
 340
 341	ret = xen_irq_info_common_setup(info, irq, IRQT_EVTCHN, evtchn, 0);
 342	info->u.interdomain = dev;
 343	if (dev)
 344		atomic_inc(&dev->event_channels);
 345
 346	return ret;
 347}
 348
 349static int xen_irq_info_ipi_setup(unsigned cpu,
 350				  unsigned irq,
 351				  evtchn_port_t evtchn,
 352				  enum ipi_vector ipi)
 353{
 354	struct irq_info *info = info_for_irq(irq);
 355
 356	info->u.ipi = ipi;
 357
 358	per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 359
 360	return xen_irq_info_common_setup(info, irq, IRQT_IPI, evtchn, 0);
 361}
 362
 363static int xen_irq_info_virq_setup(unsigned cpu,
 364				   unsigned irq,
 365				   evtchn_port_t evtchn,
 366				   unsigned virq)
 367{
 368	struct irq_info *info = info_for_irq(irq);
 369
 370	info->u.virq = virq;
 371
 372	per_cpu(virq_to_irq, cpu)[virq] = irq;
 373
 374	return xen_irq_info_common_setup(info, irq, IRQT_VIRQ, evtchn, 0);
 375}
 376
 377static int xen_irq_info_pirq_setup(unsigned irq,
 378				   evtchn_port_t evtchn,
 379				   unsigned pirq,
 380				   unsigned gsi,
 381				   uint16_t domid,
 382				   unsigned char flags)
 383{
 384	struct irq_info *info = info_for_irq(irq);
 385
 386	info->u.pirq.pirq = pirq;
 387	info->u.pirq.gsi = gsi;
 388	info->u.pirq.domid = domid;
 389	info->u.pirq.flags = flags;
 390
 391	return xen_irq_info_common_setup(info, irq, IRQT_PIRQ, evtchn, 0);
 392}
 393
 394static void xen_irq_info_cleanup(struct irq_info *info)
 395{
 396	set_evtchn_to_irq(info->evtchn, -1);
 397	xen_evtchn_port_remove(info->evtchn, info->cpu);
 398	info->evtchn = 0;
 399	channels_on_cpu_dec(info);
 400}
 401
 402/*
 403 * Accessors for packed IRQ information.
 404 */
 405evtchn_port_t evtchn_from_irq(unsigned irq)
 406{
 407	const struct irq_info *info = NULL;
 408
 409	if (likely(irq < nr_irqs))
 410		info = info_for_irq(irq);
 411	if (!info)
 412		return 0;
 413
 414	return info->evtchn;
 415}
 416
 417unsigned int irq_from_evtchn(evtchn_port_t evtchn)
 418{
 419	return get_evtchn_to_irq(evtchn);
 420}
 421EXPORT_SYMBOL_GPL(irq_from_evtchn);
 422
 423int irq_from_virq(unsigned int cpu, unsigned int virq)
 424{
 425	return per_cpu(virq_to_irq, cpu)[virq];
 426}
 427
 428static enum ipi_vector ipi_from_irq(unsigned irq)
 429{
 430	struct irq_info *info = info_for_irq(irq);
 431
 432	BUG_ON(info == NULL);
 433	BUG_ON(info->type != IRQT_IPI);
 434
 435	return info->u.ipi;
 436}
 437
 438static unsigned virq_from_irq(unsigned irq)
 439{
 440	struct irq_info *info = info_for_irq(irq);
 441
 442	BUG_ON(info == NULL);
 443	BUG_ON(info->type != IRQT_VIRQ);
 444
 445	return info->u.virq;
 446}
 447
 448static unsigned pirq_from_irq(unsigned irq)
 449{
 450	struct irq_info *info = info_for_irq(irq);
 451
 452	BUG_ON(info == NULL);
 453	BUG_ON(info->type != IRQT_PIRQ);
 454
 455	return info->u.pirq.pirq;
 456}
 457
 458static enum xen_irq_type type_from_irq(unsigned irq)
 459{
 460	return info_for_irq(irq)->type;
 461}
 462
 463static unsigned cpu_from_irq(unsigned irq)
 464{
 465	return info_for_irq(irq)->cpu;
 466}
 467
 468unsigned int cpu_from_evtchn(evtchn_port_t evtchn)
 469{
 470	int irq = get_evtchn_to_irq(evtchn);
 471	unsigned ret = 0;
 472
 473	if (irq != -1)
 474		ret = cpu_from_irq(irq);
 475
 476	return ret;
 477}
 478
 479static void do_mask(struct irq_info *info, u8 reason)
 480{
 481	unsigned long flags;
 482
 483	raw_spin_lock_irqsave(&info->lock, flags);
 484
 485	if (!info->mask_reason)
 486		mask_evtchn(info->evtchn);
 487
 488	info->mask_reason |= reason;
 489
 490	raw_spin_unlock_irqrestore(&info->lock, flags);
 491}
 492
 493static void do_unmask(struct irq_info *info, u8 reason)
 494{
 495	unsigned long flags;
 496
 497	raw_spin_lock_irqsave(&info->lock, flags);
 498
 499	info->mask_reason &= ~reason;
 500
 501	if (!info->mask_reason)
 502		unmask_evtchn(info->evtchn);
 503
 504	raw_spin_unlock_irqrestore(&info->lock, flags);
 505}
 506
 507#ifdef CONFIG_X86
 508static bool pirq_check_eoi_map(unsigned irq)
 509{
 510	return test_bit(pirq_from_irq(irq), pirq_eoi_map);
 511}
 512#endif
 513
 514static bool pirq_needs_eoi_flag(unsigned irq)
 515{
 516	struct irq_info *info = info_for_irq(irq);
 517	BUG_ON(info->type != IRQT_PIRQ);
 518
 519	return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 520}
 521
 522static void bind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int cpu,
 523			       bool force_affinity)
 524{
 525	int irq = get_evtchn_to_irq(evtchn);
 526	struct irq_info *info = info_for_irq(irq);
 527
 528	BUG_ON(irq == -1);
 529
 530	if (IS_ENABLED(CONFIG_SMP) && force_affinity) {
 531		cpumask_copy(irq_get_affinity_mask(irq), cpumask_of(cpu));
 532		cpumask_copy(irq_get_effective_affinity_mask(irq),
 533			     cpumask_of(cpu));
 534	}
 535
 536	xen_evtchn_port_bind_to_cpu(evtchn, cpu, info->cpu);
 537
 538	channels_on_cpu_dec(info);
 539	info->cpu = cpu;
 540	channels_on_cpu_inc(info);
 541}
 542
 543/**
 544 * notify_remote_via_irq - send event to remote end of event channel via irq
 545 * @irq: irq of event channel to send event to
 546 *
 547 * Unlike notify_remote_via_evtchn(), this is safe to use across
 548 * save/restore. Notifications on a broken connection are silently
 549 * dropped.
 550 */
 551void notify_remote_via_irq(int irq)
 552{
 553	evtchn_port_t evtchn = evtchn_from_irq(irq);
 554
 555	if (VALID_EVTCHN(evtchn))
 556		notify_remote_via_evtchn(evtchn);
 557}
 558EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 559
 560struct lateeoi_work {
 561	struct delayed_work delayed;
 562	spinlock_t eoi_list_lock;
 563	struct list_head eoi_list;
 564};
 565
 566static DEFINE_PER_CPU(struct lateeoi_work, lateeoi);
 567
 568static void lateeoi_list_del(struct irq_info *info)
 569{
 570	struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
 571	unsigned long flags;
 572
 573	spin_lock_irqsave(&eoi->eoi_list_lock, flags);
 574	list_del_init(&info->eoi_list);
 575	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
 576}
 577
 578static void lateeoi_list_add(struct irq_info *info)
 579{
 580	struct lateeoi_work *eoi = &per_cpu(lateeoi, info->eoi_cpu);
 581	struct irq_info *elem;
 582	u64 now = get_jiffies_64();
 583	unsigned long delay;
 584	unsigned long flags;
 585
 586	if (now < info->eoi_time)
 587		delay = info->eoi_time - now;
 588	else
 589		delay = 1;
 590
 591	spin_lock_irqsave(&eoi->eoi_list_lock, flags);
 592
 593	if (list_empty(&eoi->eoi_list)) {
 594		list_add(&info->eoi_list, &eoi->eoi_list);
 595		mod_delayed_work_on(info->eoi_cpu, system_wq,
 596				    &eoi->delayed, delay);
 597	} else {
 598		list_for_each_entry_reverse(elem, &eoi->eoi_list, eoi_list) {
 599			if (elem->eoi_time <= info->eoi_time)
 600				break;
 601		}
 602		list_add(&info->eoi_list, &elem->eoi_list);
 603	}
 604
 605	spin_unlock_irqrestore(&eoi->eoi_list_lock, flags);
 606}
 607
 608static void xen_irq_lateeoi_locked(struct irq_info *info, bool spurious)
 609{
 610	evtchn_port_t evtchn;
 611	unsigned int cpu;
 612	unsigned int delay = 0;
 613
 614	evtchn = info->evtchn;
 615	if (!VALID_EVTCHN(evtchn) || !list_empty(&info->eoi_list))
 616		return;
 617
 618	if (spurious) {
 619		struct xenbus_device *dev = info->u.interdomain;
 620		unsigned int threshold = 1;
 621
 622		if (dev && dev->spurious_threshold)
 623			threshold = dev->spurious_threshold;
 624
 625		if ((1 << info->spurious_cnt) < (HZ << 2)) {
 626			if (info->spurious_cnt != 0xFF)
 627				info->spurious_cnt++;
 628		}
 629		if (info->spurious_cnt > threshold) {
 630			delay = 1 << (info->spurious_cnt - 1 - threshold);
 631			if (delay > HZ)
 632				delay = HZ;
 633			if (!info->eoi_time)
 634				info->eoi_cpu = smp_processor_id();
 635			info->eoi_time = get_jiffies_64() + delay;
 636			if (dev)
 637				atomic_add(delay, &dev->jiffies_eoi_delayed);
 638		}
 639		if (dev)
 640			atomic_inc(&dev->spurious_events);
 641	} else {
 642		info->spurious_cnt = 0;
 643	}
 644
 645	cpu = info->eoi_cpu;
 646	if (info->eoi_time &&
 647	    (info->irq_epoch == per_cpu(irq_epoch, cpu) || delay)) {
 648		lateeoi_list_add(info);
 649		return;
 650	}
 651
 652	info->eoi_time = 0;
 653
 654	/* is_active hasn't been reset yet, do it now. */
 655	smp_store_release(&info->is_active, 0);
 656	do_unmask(info, EVT_MASK_REASON_EOI_PENDING);
 657}
 658
 659static void xen_irq_lateeoi_worker(struct work_struct *work)
 660{
 661	struct lateeoi_work *eoi;
 662	struct irq_info *info;
 663	u64 now = get_jiffies_64();
 664	unsigned long flags;
 665
 666	eoi = container_of(to_delayed_work(work), struct lateeoi_work, delayed);
 667
 668	read_lock_irqsave(&evtchn_rwlock, flags);
 669
 670	while (true) {
 671		spin_lock(&eoi->eoi_list_lock);
 672
 673		info = list_first_entry_or_null(&eoi->eoi_list, struct irq_info,
 674						eoi_list);
 675
 676		if (info == NULL || now < info->eoi_time) {
 677			spin_unlock(&eoi->eoi_list_lock);
 678			break;
 679		}
 680
 681		list_del_init(&info->eoi_list);
 682
 683		spin_unlock(&eoi->eoi_list_lock);
 684
 685		info->eoi_time = 0;
 686
 687		xen_irq_lateeoi_locked(info, false);
 688	}
 689
 690	if (info)
 691		mod_delayed_work_on(info->eoi_cpu, system_wq,
 692				    &eoi->delayed, info->eoi_time - now);
 693
 694	read_unlock_irqrestore(&evtchn_rwlock, flags);
 695}
 696
 697static void xen_cpu_init_eoi(unsigned int cpu)
 698{
 699	struct lateeoi_work *eoi = &per_cpu(lateeoi, cpu);
 700
 701	INIT_DELAYED_WORK(&eoi->delayed, xen_irq_lateeoi_worker);
 702	spin_lock_init(&eoi->eoi_list_lock);
 703	INIT_LIST_HEAD(&eoi->eoi_list);
 704}
 705
 706void xen_irq_lateeoi(unsigned int irq, unsigned int eoi_flags)
 707{
 708	struct irq_info *info;
 709	unsigned long flags;
 710
 711	read_lock_irqsave(&evtchn_rwlock, flags);
 712
 713	info = info_for_irq(irq);
 714
 715	if (info)
 716		xen_irq_lateeoi_locked(info, eoi_flags & XEN_EOI_FLAG_SPURIOUS);
 717
 718	read_unlock_irqrestore(&evtchn_rwlock, flags);
 719}
 720EXPORT_SYMBOL_GPL(xen_irq_lateeoi);
 721
 722static void xen_irq_init(unsigned irq)
 723{
 724	struct irq_info *info;
 725
 726	info = kzalloc(sizeof(*info), GFP_KERNEL);
 727	if (info == NULL)
 728		panic("Unable to allocate metadata for IRQ%d\n", irq);
 729
 730	info->type = IRQT_UNBOUND;
 731	info->refcnt = -1;
 732
 733	set_info_for_irq(irq, info);
 734	/*
 735	 * Interrupt affinity setting can be immediate. No point
 736	 * in delaying it until an interrupt is handled.
 737	 */
 738	irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
 739
 740	INIT_LIST_HEAD(&info->eoi_list);
 741	list_add_tail(&info->list, &xen_irq_list_head);
 742}
 743
 744static int __must_check xen_allocate_irqs_dynamic(int nvec)
 745{
 746	int i, irq = irq_alloc_descs(-1, 0, nvec, -1);
 747
 748	if (irq >= 0) {
 749		for (i = 0; i < nvec; i++)
 750			xen_irq_init(irq + i);
 751	}
 752
 753	return irq;
 754}
 755
 756static inline int __must_check xen_allocate_irq_dynamic(void)
 757{
 758
 759	return xen_allocate_irqs_dynamic(1);
 760}
 761
 762static int __must_check xen_allocate_irq_gsi(unsigned gsi)
 763{
 764	int irq;
 765
 766	/*
 767	 * A PV guest has no concept of a GSI (since it has no ACPI
 768	 * nor access to/knowledge of the physical APICs). Therefore
 769	 * all IRQs are dynamically allocated from the entire IRQ
 770	 * space.
 771	 */
 772	if (xen_pv_domain() && !xen_initial_domain())
 773		return xen_allocate_irq_dynamic();
 774
 775	/* Legacy IRQ descriptors are already allocated by the arch. */
 776	if (gsi < nr_legacy_irqs())
 777		irq = gsi;
 778	else
 779		irq = irq_alloc_desc_at(gsi, -1);
 780
 781	xen_irq_init(irq);
 782
 783	return irq;
 784}
 785
 786static void xen_free_irq(unsigned irq)
 787{
 788	struct irq_info *info = info_for_irq(irq);
 789	unsigned long flags;
 790
 791	if (WARN_ON(!info))
 792		return;
 793
 794	write_lock_irqsave(&evtchn_rwlock, flags);
 795
 796	if (!list_empty(&info->eoi_list))
 797		lateeoi_list_del(info);
 798
 799	list_del(&info->list);
 800
 801	set_info_for_irq(irq, NULL);
 802
 803	WARN_ON(info->refcnt > 0);
 804
 805	write_unlock_irqrestore(&evtchn_rwlock, flags);
 806
 807	kfree(info);
 808
 809	/* Legacy IRQ descriptors are managed by the arch. */
 810	if (irq < nr_legacy_irqs())
 811		return;
 812
 813	irq_free_desc(irq);
 814}
 815
 816static void xen_evtchn_close(evtchn_port_t port)
 817{
 818	struct evtchn_close close;
 819
 820	close.port = port;
 821	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 822		BUG();
 823}
 824
 825/* Not called for lateeoi events. */
 826static void event_handler_exit(struct irq_info *info)
 827{
 828	smp_store_release(&info->is_active, 0);
 829	clear_evtchn(info->evtchn);
 830}
 831
 832static void pirq_query_unmask(int irq)
 833{
 834	struct physdev_irq_status_query irq_status;
 835	struct irq_info *info = info_for_irq(irq);
 836
 837	BUG_ON(info->type != IRQT_PIRQ);
 838
 839	irq_status.irq = pirq_from_irq(irq);
 840	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 841		irq_status.flags = 0;
 842
 843	info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 844	if (irq_status.flags & XENIRQSTAT_needs_eoi)
 845		info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 846}
 847
 848static void eoi_pirq(struct irq_data *data)
 849{
 850	struct irq_info *info = info_for_irq(data->irq);
 851	evtchn_port_t evtchn = info ? info->evtchn : 0;
 852	struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) };
 853	int rc = 0;
 854
 855	if (!VALID_EVTCHN(evtchn))
 856		return;
 857
 858	event_handler_exit(info);
 859
 860	if (pirq_needs_eoi(data->irq)) {
 861		rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 862		WARN_ON(rc);
 863	}
 864}
 865
 866static void mask_ack_pirq(struct irq_data *data)
 867{
 868	disable_dynirq(data);
 869	eoi_pirq(data);
 870}
 871
 872static unsigned int __startup_pirq(unsigned int irq)
 873{
 874	struct evtchn_bind_pirq bind_pirq;
 875	struct irq_info *info = info_for_irq(irq);
 876	evtchn_port_t evtchn = evtchn_from_irq(irq);
 877	int rc;
 878
 879	BUG_ON(info->type != IRQT_PIRQ);
 880
 881	if (VALID_EVTCHN(evtchn))
 882		goto out;
 883
 884	bind_pirq.pirq = pirq_from_irq(irq);
 885	/* NB. We are happy to share unless we are probing. */
 886	bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 887					BIND_PIRQ__WILL_SHARE : 0;
 888	rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 889	if (rc != 0) {
 890		pr_warn("Failed to obtain physical IRQ %d\n", irq);
 891		return 0;
 892	}
 893	evtchn = bind_pirq.port;
 894
 895	pirq_query_unmask(irq);
 896
 897	rc = set_evtchn_to_irq(evtchn, irq);
 898	if (rc)
 899		goto err;
 900
 901	info->evtchn = evtchn;
 902	bind_evtchn_to_cpu(evtchn, 0, false);
 903
 904	rc = xen_evtchn_port_setup(evtchn);
 905	if (rc)
 906		goto err;
 907
 908out:
 909	do_unmask(info, EVT_MASK_REASON_EXPLICIT);
 910
 911	eoi_pirq(irq_get_irq_data(irq));
 912
 913	return 0;
 914
 915err:
 916	pr_err("irq%d: Failed to set port to irq mapping (%d)\n", irq, rc);
 917	xen_evtchn_close(evtchn);
 918	return 0;
 919}
 920
 921static unsigned int startup_pirq(struct irq_data *data)
 922{
 923	return __startup_pirq(data->irq);
 924}
 925
 926static void shutdown_pirq(struct irq_data *data)
 927{
 928	unsigned int irq = data->irq;
 929	struct irq_info *info = info_for_irq(irq);
 930	evtchn_port_t evtchn = evtchn_from_irq(irq);
 931
 932	BUG_ON(info->type != IRQT_PIRQ);
 933
 934	if (!VALID_EVTCHN(evtchn))
 935		return;
 936
 937	do_mask(info, EVT_MASK_REASON_EXPLICIT);
 938	xen_evtchn_close(evtchn);
 939	xen_irq_info_cleanup(info);
 940}
 941
 942static void enable_pirq(struct irq_data *data)
 943{
 944	enable_dynirq(data);
 945}
 946
 947static void disable_pirq(struct irq_data *data)
 948{
 949	disable_dynirq(data);
 950}
 951
 952int xen_irq_from_gsi(unsigned gsi)
 953{
 954	struct irq_info *info;
 955
 956	list_for_each_entry(info, &xen_irq_list_head, list) {
 957		if (info->type != IRQT_PIRQ)
 958			continue;
 959
 960		if (info->u.pirq.gsi == gsi)
 961			return info->irq;
 962	}
 963
 964	return -1;
 965}
 966EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 967
 968static void __unbind_from_irq(unsigned int irq)
 969{
 970	evtchn_port_t evtchn = evtchn_from_irq(irq);
 971	struct irq_info *info = info_for_irq(irq);
 972
 973	if (info->refcnt > 0) {
 974		info->refcnt--;
 975		if (info->refcnt != 0)
 976			return;
 977	}
 978
 979	if (VALID_EVTCHN(evtchn)) {
 980		unsigned int cpu = cpu_from_irq(irq);
 981		struct xenbus_device *dev;
 982
 983		xen_evtchn_close(evtchn);
 984
 985		switch (type_from_irq(irq)) {
 986		case IRQT_VIRQ:
 987			per_cpu(virq_to_irq, cpu)[virq_from_irq(irq)] = -1;
 988			break;
 989		case IRQT_IPI:
 990			per_cpu(ipi_to_irq, cpu)[ipi_from_irq(irq)] = -1;
 991			break;
 992		case IRQT_EVTCHN:
 993			dev = info->u.interdomain;
 994			if (dev)
 995				atomic_dec(&dev->event_channels);
 996			break;
 997		default:
 998			break;
 999		}
1000
1001		xen_irq_info_cleanup(info);
1002	}
1003
1004	xen_free_irq(irq);
1005}
1006
1007/*
1008 * Do not make any assumptions regarding the relationship between the
1009 * IRQ number returned here and the Xen pirq argument.
1010 *
1011 * Note: We don't assign an event channel until the irq actually started
1012 * up.  Return an existing irq if we've already got one for the gsi.
1013 *
1014 * Shareable implies level triggered, not shareable implies edge
1015 * triggered here.
1016 */
1017int xen_bind_pirq_gsi_to_irq(unsigned gsi,
1018			     unsigned pirq, int shareable, char *name)
1019{
1020	int irq;
1021	struct physdev_irq irq_op;
1022	int ret;
1023
1024	mutex_lock(&irq_mapping_update_lock);
1025
1026	irq = xen_irq_from_gsi(gsi);
1027	if (irq != -1) {
1028		pr_info("%s: returning irq %d for gsi %u\n",
1029			__func__, irq, gsi);
1030		goto out;
1031	}
1032
1033	irq = xen_allocate_irq_gsi(gsi);
1034	if (irq < 0)
1035		goto out;
1036
1037	irq_op.irq = irq;
1038	irq_op.vector = 0;
1039
1040	/* Only the privileged domain can do this. For non-priv, the pcifront
1041	 * driver provides a PCI bus that does the call to do exactly
1042	 * this in the priv domain. */
1043	if (xen_initial_domain() &&
1044	    HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
1045		xen_free_irq(irq);
1046		irq = -ENOSPC;
1047		goto out;
1048	}
1049
1050	ret = xen_irq_info_pirq_setup(irq, 0, pirq, gsi, DOMID_SELF,
1051			       shareable ? PIRQ_SHAREABLE : 0);
1052	if (ret < 0) {
1053		__unbind_from_irq(irq);
1054		irq = ret;
1055		goto out;
1056	}
1057
1058	pirq_query_unmask(irq);
1059	/* We try to use the handler with the appropriate semantic for the
1060	 * type of interrupt: if the interrupt is an edge triggered
1061	 * interrupt we use handle_edge_irq.
1062	 *
1063	 * On the other hand if the interrupt is level triggered we use
1064	 * handle_fasteoi_irq like the native code does for this kind of
1065	 * interrupts.
1066	 *
1067	 * Depending on the Xen version, pirq_needs_eoi might return true
1068	 * not only for level triggered interrupts but for edge triggered
1069	 * interrupts too. In any case Xen always honors the eoi mechanism,
1070	 * not injecting any more pirqs of the same kind if the first one
1071	 * hasn't received an eoi yet. Therefore using the fasteoi handler
1072	 * is the right choice either way.
1073	 */
1074	if (shareable)
1075		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
1076				handle_fasteoi_irq, name);
1077	else
1078		irq_set_chip_and_handler_name(irq, &xen_pirq_chip,
1079				handle_edge_irq, name);
1080
1081out:
1082	mutex_unlock(&irq_mapping_update_lock);
1083
1084	return irq;
1085}
1086
1087#ifdef CONFIG_PCI_MSI
1088int xen_allocate_pirq_msi(struct pci_dev *dev, struct msi_desc *msidesc)
1089{
1090	int rc;
1091	struct physdev_get_free_pirq op_get_free_pirq;
1092
1093	op_get_free_pirq.type = MAP_PIRQ_TYPE_MSI;
1094	rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
1095
1096	WARN_ONCE(rc == -ENOSYS,
1097		  "hypervisor does not support the PHYSDEVOP_get_free_pirq interface\n");
1098
1099	return rc ? -1 : op_get_free_pirq.pirq;
1100}
1101
1102int xen_bind_pirq_msi_to_irq(struct pci_dev *dev, struct msi_desc *msidesc,
1103			     int pirq, int nvec, const char *name, domid_t domid)
1104{
1105	int i, irq, ret;
1106
1107	mutex_lock(&irq_mapping_update_lock);
1108
1109	irq = xen_allocate_irqs_dynamic(nvec);
1110	if (irq < 0)
1111		goto out;
1112
1113	for (i = 0; i < nvec; i++) {
1114		irq_set_chip_and_handler_name(irq + i, &xen_pirq_chip, handle_edge_irq, name);
1115
1116		ret = xen_irq_info_pirq_setup(irq + i, 0, pirq + i, 0, domid,
1117					      i == 0 ? 0 : PIRQ_MSI_GROUP);
1118		if (ret < 0)
1119			goto error_irq;
1120	}
1121
1122	ret = irq_set_msi_desc(irq, msidesc);
1123	if (ret < 0)
1124		goto error_irq;
1125out:
1126	mutex_unlock(&irq_mapping_update_lock);
1127	return irq;
1128error_irq:
1129	while (nvec--)
1130		__unbind_from_irq(irq + nvec);
1131	mutex_unlock(&irq_mapping_update_lock);
1132	return ret;
1133}
1134#endif
1135
1136int xen_destroy_irq(int irq)
1137{
1138	struct physdev_unmap_pirq unmap_irq;
1139	struct irq_info *info = info_for_irq(irq);
1140	int rc = -ENOENT;
1141
1142	mutex_lock(&irq_mapping_update_lock);
1143
1144	/*
1145	 * If trying to remove a vector in a MSI group different
1146	 * than the first one skip the PIRQ unmap unless this vector
1147	 * is the first one in the group.
1148	 */
1149	if (xen_initial_domain() && !(info->u.pirq.flags & PIRQ_MSI_GROUP)) {
1150		unmap_irq.pirq = info->u.pirq.pirq;
1151		unmap_irq.domid = info->u.pirq.domid;
1152		rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
1153		/* If another domain quits without making the pci_disable_msix
1154		 * call, the Xen hypervisor takes care of freeing the PIRQs
1155		 * (free_domain_pirqs).
1156		 */
1157		if ((rc == -ESRCH && info->u.pirq.domid != DOMID_SELF))
1158			pr_info("domain %d does not have %d anymore\n",
1159				info->u.pirq.domid, info->u.pirq.pirq);
1160		else if (rc) {
1161			pr_warn("unmap irq failed %d\n", rc);
1162			goto out;
1163		}
1164	}
1165
1166	xen_free_irq(irq);
1167
1168out:
1169	mutex_unlock(&irq_mapping_update_lock);
1170	return rc;
1171}
1172
1173int xen_irq_from_pirq(unsigned pirq)
1174{
1175	int irq;
1176
1177	struct irq_info *info;
1178
1179	mutex_lock(&irq_mapping_update_lock);
1180
1181	list_for_each_entry(info, &xen_irq_list_head, list) {
1182		if (info->type != IRQT_PIRQ)
1183			continue;
1184		irq = info->irq;
1185		if (info->u.pirq.pirq == pirq)
1186			goto out;
1187	}
1188	irq = -1;
1189out:
1190	mutex_unlock(&irq_mapping_update_lock);
1191
1192	return irq;
1193}
1194
1195
1196int xen_pirq_from_irq(unsigned irq)
1197{
1198	return pirq_from_irq(irq);
1199}
1200EXPORT_SYMBOL_GPL(xen_pirq_from_irq);
1201
1202static int bind_evtchn_to_irq_chip(evtchn_port_t evtchn, struct irq_chip *chip,
1203				   struct xenbus_device *dev)
1204{
1205	int irq;
1206	int ret;
1207
1208	if (evtchn >= xen_evtchn_max_channels())
1209		return -ENOMEM;
1210
1211	mutex_lock(&irq_mapping_update_lock);
1212
1213	irq = get_evtchn_to_irq(evtchn);
1214
1215	if (irq == -1) {
1216		irq = xen_allocate_irq_dynamic();
1217		if (irq < 0)
1218			goto out;
1219
1220		irq_set_chip_and_handler_name(irq, chip,
1221					      handle_edge_irq, "event");
1222
1223		ret = xen_irq_info_evtchn_setup(irq, evtchn, dev);
1224		if (ret < 0) {
1225			__unbind_from_irq(irq);
1226			irq = ret;
1227			goto out;
1228		}
1229		/*
1230		 * New interdomain events are initially bound to vCPU0 This
1231		 * is required to setup the event channel in the first
1232		 * place and also important for UP guests because the
1233		 * affinity setting is not invoked on them so nothing would
1234		 * bind the channel.
1235		 */
1236		bind_evtchn_to_cpu(evtchn, 0, false);
1237	} else {
1238		struct irq_info *info = info_for_irq(irq);
1239		WARN_ON(info == NULL || info->type != IRQT_EVTCHN);
1240	}
1241
1242out:
1243	mutex_unlock(&irq_mapping_update_lock);
1244
1245	return irq;
1246}
1247
1248int bind_evtchn_to_irq(evtchn_port_t evtchn)
1249{
1250	return bind_evtchn_to_irq_chip(evtchn, &xen_dynamic_chip, NULL);
1251}
1252EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
1253
1254static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
1255{
1256	struct evtchn_bind_ipi bind_ipi;
1257	evtchn_port_t evtchn;
1258	int ret, irq;
1259
1260	mutex_lock(&irq_mapping_update_lock);
1261
1262	irq = per_cpu(ipi_to_irq, cpu)[ipi];
1263
1264	if (irq == -1) {
1265		irq = xen_allocate_irq_dynamic();
1266		if (irq < 0)
1267			goto out;
1268
1269		irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1270					      handle_percpu_irq, "ipi");
1271
1272		bind_ipi.vcpu = xen_vcpu_nr(cpu);
1273		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1274						&bind_ipi) != 0)
1275			BUG();
1276		evtchn = bind_ipi.port;
1277
1278		ret = xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
1279		if (ret < 0) {
1280			__unbind_from_irq(irq);
1281			irq = ret;
1282			goto out;
1283		}
1284		/*
1285		 * Force the affinity mask to the target CPU so proc shows
1286		 * the correct target.
1287		 */
1288		bind_evtchn_to_cpu(evtchn, cpu, true);
1289	} else {
1290		struct irq_info *info = info_for_irq(irq);
1291		WARN_ON(info == NULL || info->type != IRQT_IPI);
1292	}
1293
1294 out:
1295	mutex_unlock(&irq_mapping_update_lock);
1296	return irq;
1297}
1298
1299static int bind_interdomain_evtchn_to_irq_chip(struct xenbus_device *dev,
1300					       evtchn_port_t remote_port,
1301					       struct irq_chip *chip)
1302{
1303	struct evtchn_bind_interdomain bind_interdomain;
1304	int err;
1305
1306	bind_interdomain.remote_dom  = dev->otherend_id;
1307	bind_interdomain.remote_port = remote_port;
1308
1309	err = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
1310					  &bind_interdomain);
1311
1312	return err ? : bind_evtchn_to_irq_chip(bind_interdomain.local_port,
1313					       chip, dev);
1314}
1315
1316int bind_interdomain_evtchn_to_irq_lateeoi(struct xenbus_device *dev,
1317					   evtchn_port_t remote_port)
1318{
1319	return bind_interdomain_evtchn_to_irq_chip(dev, remote_port,
1320						   &xen_lateeoi_chip);
1321}
1322EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irq_lateeoi);
1323
1324static int find_virq(unsigned int virq, unsigned int cpu, evtchn_port_t *evtchn)
1325{
1326	struct evtchn_status status;
1327	evtchn_port_t port;
1328	int rc = -ENOENT;
1329
1330	memset(&status, 0, sizeof(status));
1331	for (port = 0; port < xen_evtchn_max_channels(); port++) {
1332		status.dom = DOMID_SELF;
1333		status.port = port;
1334		rc = HYPERVISOR_event_channel_op(EVTCHNOP_status, &status);
1335		if (rc < 0)
1336			continue;
1337		if (status.status != EVTCHNSTAT_virq)
1338			continue;
1339		if (status.u.virq == virq && status.vcpu == xen_vcpu_nr(cpu)) {
1340			*evtchn = port;
1341			break;
1342		}
1343	}
1344	return rc;
1345}
1346
1347/**
1348 * xen_evtchn_nr_channels - number of usable event channel ports
1349 *
1350 * This may be less than the maximum supported by the current
1351 * hypervisor ABI. Use xen_evtchn_max_channels() for the maximum
1352 * supported.
1353 */
1354unsigned xen_evtchn_nr_channels(void)
1355{
1356        return evtchn_ops->nr_channels();
1357}
1358EXPORT_SYMBOL_GPL(xen_evtchn_nr_channels);
1359
1360int bind_virq_to_irq(unsigned int virq, unsigned int cpu, bool percpu)
1361{
1362	struct evtchn_bind_virq bind_virq;
1363	evtchn_port_t evtchn = 0;
1364	int irq, ret;
1365
1366	mutex_lock(&irq_mapping_update_lock);
1367
1368	irq = per_cpu(virq_to_irq, cpu)[virq];
1369
1370	if (irq == -1) {
1371		irq = xen_allocate_irq_dynamic();
1372		if (irq < 0)
1373			goto out;
1374
1375		if (percpu)
1376			irq_set_chip_and_handler_name(irq, &xen_percpu_chip,
1377						      handle_percpu_irq, "virq");
1378		else
1379			irq_set_chip_and_handler_name(irq, &xen_dynamic_chip,
1380						      handle_edge_irq, "virq");
1381
1382		bind_virq.virq = virq;
1383		bind_virq.vcpu = xen_vcpu_nr(cpu);
1384		ret = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1385						&bind_virq);
1386		if (ret == 0)
1387			evtchn = bind_virq.port;
1388		else {
1389			if (ret == -EEXIST)
1390				ret = find_virq(virq, cpu, &evtchn);
1391			BUG_ON(ret < 0);
1392		}
1393
1394		ret = xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1395		if (ret < 0) {
1396			__unbind_from_irq(irq);
1397			irq = ret;
1398			goto out;
1399		}
1400
1401		/*
1402		 * Force the affinity mask for percpu interrupts so proc
1403		 * shows the correct target.
1404		 */
1405		bind_evtchn_to_cpu(evtchn, cpu, percpu);
1406	} else {
1407		struct irq_info *info = info_for_irq(irq);
1408		WARN_ON(info == NULL || info->type != IRQT_VIRQ);
1409	}
1410
1411out:
1412	mutex_unlock(&irq_mapping_update_lock);
1413
1414	return irq;
1415}
1416
1417static void unbind_from_irq(unsigned int irq)
1418{
1419	mutex_lock(&irq_mapping_update_lock);
1420	__unbind_from_irq(irq);
1421	mutex_unlock(&irq_mapping_update_lock);
1422}
1423
1424static int bind_evtchn_to_irqhandler_chip(evtchn_port_t evtchn,
1425					  irq_handler_t handler,
1426					  unsigned long irqflags,
1427					  const char *devname, void *dev_id,
1428					  struct irq_chip *chip)
1429{
1430	int irq, retval;
1431
1432	irq = bind_evtchn_to_irq_chip(evtchn, chip, NULL);
1433	if (irq < 0)
1434		return irq;
1435	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1436	if (retval != 0) {
1437		unbind_from_irq(irq);
1438		return retval;
1439	}
1440
1441	return irq;
1442}
1443
1444int bind_evtchn_to_irqhandler(evtchn_port_t evtchn,
1445			      irq_handler_t handler,
1446			      unsigned long irqflags,
1447			      const char *devname, void *dev_id)
1448{
1449	return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1450					      devname, dev_id,
1451					      &xen_dynamic_chip);
1452}
1453EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
1454
1455int bind_evtchn_to_irqhandler_lateeoi(evtchn_port_t evtchn,
1456				      irq_handler_t handler,
1457				      unsigned long irqflags,
1458				      const char *devname, void *dev_id)
1459{
1460	return bind_evtchn_to_irqhandler_chip(evtchn, handler, irqflags,
1461					      devname, dev_id,
1462					      &xen_lateeoi_chip);
1463}
1464EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler_lateeoi);
1465
1466static int bind_interdomain_evtchn_to_irqhandler_chip(
1467		struct xenbus_device *dev, evtchn_port_t remote_port,
1468		irq_handler_t handler, unsigned long irqflags,
1469		const char *devname, void *dev_id, struct irq_chip *chip)
1470{
1471	int irq, retval;
1472
1473	irq = bind_interdomain_evtchn_to_irq_chip(dev, remote_port, chip);
1474	if (irq < 0)
1475		return irq;
1476
1477	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1478	if (retval != 0) {
1479		unbind_from_irq(irq);
1480		return retval;
1481	}
1482
1483	return irq;
1484}
1485
1486int bind_interdomain_evtchn_to_irqhandler_lateeoi(struct xenbus_device *dev,
1487						  evtchn_port_t remote_port,
1488						  irq_handler_t handler,
1489						  unsigned long irqflags,
1490						  const char *devname,
1491						  void *dev_id)
1492{
1493	return bind_interdomain_evtchn_to_irqhandler_chip(dev,
1494				remote_port, handler, irqflags, devname,
1495				dev_id, &xen_lateeoi_chip);
1496}
1497EXPORT_SYMBOL_GPL(bind_interdomain_evtchn_to_irqhandler_lateeoi);
1498
1499int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
1500			    irq_handler_t handler,
1501			    unsigned long irqflags, const char *devname, void *dev_id)
1502{
1503	int irq, retval;
1504
1505	irq = bind_virq_to_irq(virq, cpu, irqflags & IRQF_PERCPU);
1506	if (irq < 0)
1507		return irq;
1508	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1509	if (retval != 0) {
1510		unbind_from_irq(irq);
1511		return retval;
1512	}
1513
1514	return irq;
1515}
1516EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
1517
1518int bind_ipi_to_irqhandler(enum ipi_vector ipi,
1519			   unsigned int cpu,
1520			   irq_handler_t handler,
1521			   unsigned long irqflags,
1522			   const char *devname,
1523			   void *dev_id)
1524{
1525	int irq, retval;
1526
1527	irq = bind_ipi_to_irq(ipi, cpu);
1528	if (irq < 0)
1529		return irq;
1530
1531	irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME | IRQF_EARLY_RESUME;
1532	retval = request_irq(irq, handler, irqflags, devname, dev_id);
1533	if (retval != 0) {
1534		unbind_from_irq(irq);
1535		return retval;
1536	}
1537
1538	return irq;
1539}
1540
1541void unbind_from_irqhandler(unsigned int irq, void *dev_id)
1542{
1543	struct irq_info *info = info_for_irq(irq);
1544
1545	if (WARN_ON(!info))
1546		return;
1547	free_irq(irq, dev_id);
1548	unbind_from_irq(irq);
1549}
1550EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
1551
1552/**
1553 * xen_set_irq_priority() - set an event channel priority.
1554 * @irq:irq bound to an event channel.
1555 * @priority: priority between XEN_IRQ_PRIORITY_MAX and XEN_IRQ_PRIORITY_MIN.
1556 */
1557int xen_set_irq_priority(unsigned irq, unsigned priority)
1558{
1559	struct evtchn_set_priority set_priority;
1560
1561	set_priority.port = evtchn_from_irq(irq);
1562	set_priority.priority = priority;
1563
1564	return HYPERVISOR_event_channel_op(EVTCHNOP_set_priority,
1565					   &set_priority);
1566}
1567EXPORT_SYMBOL_GPL(xen_set_irq_priority);
1568
1569int evtchn_make_refcounted(evtchn_port_t evtchn)
1570{
1571	int irq = get_evtchn_to_irq(evtchn);
1572	struct irq_info *info;
1573
1574	if (irq == -1)
1575		return -ENOENT;
1576
1577	info = info_for_irq(irq);
1578
1579	if (!info)
1580		return -ENOENT;
1581
1582	WARN_ON(info->refcnt != -1);
1583
1584	info->refcnt = 1;
1585
1586	return 0;
1587}
1588EXPORT_SYMBOL_GPL(evtchn_make_refcounted);
1589
1590int evtchn_get(evtchn_port_t evtchn)
1591{
1592	int irq;
1593	struct irq_info *info;
1594	int err = -ENOENT;
1595
1596	if (evtchn >= xen_evtchn_max_channels())
1597		return -EINVAL;
1598
1599	mutex_lock(&irq_mapping_update_lock);
1600
1601	irq = get_evtchn_to_irq(evtchn);
1602	if (irq == -1)
1603		goto done;
1604
1605	info = info_for_irq(irq);
1606
1607	if (!info)
1608		goto done;
1609
1610	err = -EINVAL;
1611	if (info->refcnt <= 0 || info->refcnt == SHRT_MAX)
1612		goto done;
1613
1614	info->refcnt++;
1615	err = 0;
1616 done:
1617	mutex_unlock(&irq_mapping_update_lock);
1618
1619	return err;
1620}
1621EXPORT_SYMBOL_GPL(evtchn_get);
1622
1623void evtchn_put(evtchn_port_t evtchn)
1624{
1625	int irq = get_evtchn_to_irq(evtchn);
1626	if (WARN_ON(irq == -1))
1627		return;
1628	unbind_from_irq(irq);
1629}
1630EXPORT_SYMBOL_GPL(evtchn_put);
1631
1632void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
1633{
1634	int irq;
1635
1636#ifdef CONFIG_X86
1637	if (unlikely(vector == XEN_NMI_VECTOR)) {
1638		int rc =  HYPERVISOR_vcpu_op(VCPUOP_send_nmi, xen_vcpu_nr(cpu),
1639					     NULL);
1640		if (rc < 0)
1641			printk(KERN_WARNING "Sending nmi to CPU%d failed (rc:%d)\n", cpu, rc);
1642		return;
1643	}
1644#endif
1645	irq = per_cpu(ipi_to_irq, cpu)[vector];
1646	BUG_ON(irq < 0);
1647	notify_remote_via_irq(irq);
1648}
1649
1650struct evtchn_loop_ctrl {
1651	ktime_t timeout;
1652	unsigned count;
1653	bool defer_eoi;
1654};
1655
1656void handle_irq_for_port(evtchn_port_t port, struct evtchn_loop_ctrl *ctrl)
1657{
1658	int irq;
1659	struct irq_info *info;
1660	struct xenbus_device *dev;
1661
1662	irq = get_evtchn_to_irq(port);
1663	if (irq == -1)
1664		return;
1665
1666	/*
1667	 * Check for timeout every 256 events.
1668	 * We are setting the timeout value only after the first 256
1669	 * events in order to not hurt the common case of few loop
1670	 * iterations. The 256 is basically an arbitrary value.
1671	 *
1672	 * In case we are hitting the timeout we need to defer all further
1673	 * EOIs in order to ensure to leave the event handling loop rather
1674	 * sooner than later.
1675	 */
1676	if (!ctrl->defer_eoi && !(++ctrl->count & 0xff)) {
1677		ktime_t kt = ktime_get();
1678
1679		if (!ctrl->timeout) {
1680			kt = ktime_add_ms(kt,
1681					  jiffies_to_msecs(event_loop_timeout));
1682			ctrl->timeout = kt;
1683		} else if (kt > ctrl->timeout) {
1684			ctrl->defer_eoi = true;
1685		}
1686	}
1687
1688	info = info_for_irq(irq);
1689	if (xchg_acquire(&info->is_active, 1))
1690		return;
1691
1692	dev = (info->type == IRQT_EVTCHN) ? info->u.interdomain : NULL;
1693	if (dev)
1694		atomic_inc(&dev->events);
1695
1696	if (ctrl->defer_eoi) {
1697		info->eoi_cpu = smp_processor_id();
1698		info->irq_epoch = __this_cpu_read(irq_epoch);
1699		info->eoi_time = get_jiffies_64() + event_eoi_delay;
1700	}
1701
1702	generic_handle_irq(irq);
1703}
1704
1705static void __xen_evtchn_do_upcall(void)
1706{
1707	struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1708	int cpu = smp_processor_id();
1709	struct evtchn_loop_ctrl ctrl = { 0 };
1710
1711	read_lock(&evtchn_rwlock);
1712
1713	do {
1714		vcpu_info->evtchn_upcall_pending = 0;
1715
1716		xen_evtchn_handle_events(cpu, &ctrl);
1717
1718		BUG_ON(!irqs_disabled());
1719
1720		virt_rmb(); /* Hypervisor can set upcall pending. */
1721
1722	} while (vcpu_info->evtchn_upcall_pending);
1723
1724	read_unlock(&evtchn_rwlock);
1725
1726	/*
1727	 * Increment irq_epoch only now to defer EOIs only for
1728	 * xen_irq_lateeoi() invocations occurring from inside the loop
1729	 * above.
1730	 */
1731	__this_cpu_inc(irq_epoch);
1732}
1733
1734void xen_evtchn_do_upcall(struct pt_regs *regs)
1735{
1736	struct pt_regs *old_regs = set_irq_regs(regs);
1737
1738	irq_enter();
1739
1740	__xen_evtchn_do_upcall();
1741
1742	irq_exit();
1743	set_irq_regs(old_regs);
1744}
1745
1746void xen_hvm_evtchn_do_upcall(void)
1747{
1748	__xen_evtchn_do_upcall();
1749}
1750EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1751
1752/* Rebind a new event channel to an existing irq. */
1753void rebind_evtchn_irq(evtchn_port_t evtchn, int irq)
1754{
1755	struct irq_info *info = info_for_irq(irq);
1756
1757	if (WARN_ON(!info))
1758		return;
1759
1760	/* Make sure the irq is masked, since the new event channel
1761	   will also be masked. */
1762	disable_irq(irq);
1763
1764	mutex_lock(&irq_mapping_update_lock);
1765
1766	/* After resume the irq<->evtchn mappings are all cleared out */
1767	BUG_ON(get_evtchn_to_irq(evtchn) != -1);
1768	/* Expect irq to have been bound before,
1769	   so there should be a proper type */
1770	BUG_ON(info->type == IRQT_UNBOUND);
1771
1772	(void)xen_irq_info_evtchn_setup(irq, evtchn, NULL);
1773
1774	mutex_unlock(&irq_mapping_update_lock);
1775
1776	bind_evtchn_to_cpu(evtchn, info->cpu, false);
1777
1778	/* Unmask the event channel. */
1779	enable_irq(irq);
1780}
1781
1782/* Rebind an evtchn so that it gets delivered to a specific cpu */
1783static int xen_rebind_evtchn_to_cpu(struct irq_info *info, unsigned int tcpu)
1784{
1785	struct evtchn_bind_vcpu bind_vcpu;
1786	evtchn_port_t evtchn = info ? info->evtchn : 0;
1787
1788	if (!VALID_EVTCHN(evtchn))
1789		return -1;
1790
1791	if (!xen_support_evtchn_rebind())
1792		return -1;
1793
1794	/* Send future instances of this interrupt to other vcpu. */
1795	bind_vcpu.port = evtchn;
1796	bind_vcpu.vcpu = xen_vcpu_nr(tcpu);
1797
1798	/*
1799	 * Mask the event while changing the VCPU binding to prevent
1800	 * it being delivered on an unexpected VCPU.
1801	 */
1802	do_mask(info, EVT_MASK_REASON_TEMPORARY);
1803
1804	/*
1805	 * If this fails, it usually just indicates that we're dealing with a
1806	 * virq or IPI channel, which don't actually need to be rebound. Ignore
1807	 * it, but don't do the xenlinux-level rebind in that case.
1808	 */
1809	if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1810		bind_evtchn_to_cpu(evtchn, tcpu, false);
1811
1812	do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1813
1814	return 0;
1815}
1816
1817/*
1818 * Find the CPU within @dest mask which has the least number of channels
1819 * assigned. This is not precise as the per cpu counts can be modified
1820 * concurrently.
1821 */
1822static unsigned int select_target_cpu(const struct cpumask *dest)
1823{
1824	unsigned int cpu, best_cpu = UINT_MAX, minch = UINT_MAX;
1825
1826	for_each_cpu_and(cpu, dest, cpu_online_mask) {
1827		unsigned int curch = atomic_read(&channels_on_cpu[cpu]);
1828
1829		if (curch < minch) {
1830			minch = curch;
1831			best_cpu = cpu;
1832		}
1833	}
1834
1835	/*
1836	 * Catch the unlikely case that dest contains no online CPUs. Can't
1837	 * recurse.
1838	 */
1839	if (best_cpu == UINT_MAX)
1840		return select_target_cpu(cpu_online_mask);
1841
1842	return best_cpu;
1843}
1844
1845static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1846			    bool force)
1847{
1848	unsigned int tcpu = select_target_cpu(dest);
1849	int ret;
1850
1851	ret = xen_rebind_evtchn_to_cpu(info_for_irq(data->irq), tcpu);
1852	if (!ret)
1853		irq_data_update_effective_affinity(data, cpumask_of(tcpu));
1854
1855	return ret;
1856}
1857
1858static void enable_dynirq(struct irq_data *data)
1859{
1860	struct irq_info *info = info_for_irq(data->irq);
1861	evtchn_port_t evtchn = info ? info->evtchn : 0;
1862
1863	if (VALID_EVTCHN(evtchn))
1864		do_unmask(info, EVT_MASK_REASON_EXPLICIT);
1865}
1866
1867static void disable_dynirq(struct irq_data *data)
1868{
1869	struct irq_info *info = info_for_irq(data->irq);
1870	evtchn_port_t evtchn = info ? info->evtchn : 0;
1871
1872	if (VALID_EVTCHN(evtchn))
1873		do_mask(info, EVT_MASK_REASON_EXPLICIT);
1874}
1875
1876static void ack_dynirq(struct irq_data *data)
1877{
1878	struct irq_info *info = info_for_irq(data->irq);
1879	evtchn_port_t evtchn = info ? info->evtchn : 0;
1880
1881	if (VALID_EVTCHN(evtchn))
1882		event_handler_exit(info);
1883}
1884
1885static void mask_ack_dynirq(struct irq_data *data)
1886{
1887	disable_dynirq(data);
1888	ack_dynirq(data);
1889}
1890
1891static void lateeoi_ack_dynirq(struct irq_data *data)
1892{
1893	struct irq_info *info = info_for_irq(data->irq);
1894	evtchn_port_t evtchn = info ? info->evtchn : 0;
1895
1896	if (VALID_EVTCHN(evtchn)) {
1897		do_mask(info, EVT_MASK_REASON_EOI_PENDING);
1898		/*
1899		 * Don't call event_handler_exit().
1900		 * Need to keep is_active non-zero in order to ignore re-raised
1901		 * events after cpu affinity changes while a lateeoi is pending.
1902		 */
1903		clear_evtchn(evtchn);
1904	}
1905}
1906
1907static void lateeoi_mask_ack_dynirq(struct irq_data *data)
1908{
1909	struct irq_info *info = info_for_irq(data->irq);
1910	evtchn_port_t evtchn = info ? info->evtchn : 0;
1911
1912	if (VALID_EVTCHN(evtchn)) {
1913		do_mask(info, EVT_MASK_REASON_EXPLICIT);
1914		event_handler_exit(info);
1915	}
1916}
1917
1918static int retrigger_dynirq(struct irq_data *data)
1919{
1920	struct irq_info *info = info_for_irq(data->irq);
1921	evtchn_port_t evtchn = info ? info->evtchn : 0;
1922
1923	if (!VALID_EVTCHN(evtchn))
1924		return 0;
1925
1926	do_mask(info, EVT_MASK_REASON_TEMPORARY);
1927	set_evtchn(evtchn);
1928	do_unmask(info, EVT_MASK_REASON_TEMPORARY);
1929
1930	return 1;
1931}
1932
1933static void restore_pirqs(void)
1934{
1935	int pirq, rc, irq, gsi;
1936	struct physdev_map_pirq map_irq;
1937	struct irq_info *info;
1938
1939	list_for_each_entry(info, &xen_irq_list_head, list) {
1940		if (info->type != IRQT_PIRQ)
1941			continue;
1942
1943		pirq = info->u.pirq.pirq;
1944		gsi = info->u.pirq.gsi;
1945		irq = info->irq;
1946
1947		/* save/restore of PT devices doesn't work, so at this point the
1948		 * only devices present are GSI based emulated devices */
1949		if (!gsi)
1950			continue;
1951
1952		map_irq.domid = DOMID_SELF;
1953		map_irq.type = MAP_PIRQ_TYPE_GSI;
1954		map_irq.index = gsi;
1955		map_irq.pirq = pirq;
1956
1957		rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1958		if (rc) {
1959			pr_warn("xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1960				gsi, irq, pirq, rc);
1961			xen_free_irq(irq);
1962			continue;
1963		}
1964
1965		printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1966
1967		__startup_pirq(irq);
1968	}
1969}
1970
1971static void restore_cpu_virqs(unsigned int cpu)
1972{
1973	struct evtchn_bind_virq bind_virq;
1974	evtchn_port_t evtchn;
1975	int virq, irq;
1976
1977	for (virq = 0; virq < NR_VIRQS; virq++) {
1978		if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1979			continue;
1980
1981		BUG_ON(virq_from_irq(irq) != virq);
1982
1983		/* Get a new binding from Xen. */
1984		bind_virq.virq = virq;
1985		bind_virq.vcpu = xen_vcpu_nr(cpu);
1986		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1987						&bind_virq) != 0)
1988			BUG();
1989		evtchn = bind_virq.port;
1990
1991		/* Record the new mapping. */
1992		(void)xen_irq_info_virq_setup(cpu, irq, evtchn, virq);
1993		/* The affinity mask is still valid */
1994		bind_evtchn_to_cpu(evtchn, cpu, false);
1995	}
1996}
1997
1998static void restore_cpu_ipis(unsigned int cpu)
1999{
2000	struct evtchn_bind_ipi bind_ipi;
2001	evtchn_port_t evtchn;
2002	int ipi, irq;
2003
2004	for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
2005		if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
2006			continue;
2007
2008		BUG_ON(ipi_from_irq(irq) != ipi);
2009
2010		/* Get a new binding from Xen. */
2011		bind_ipi.vcpu = xen_vcpu_nr(cpu);
2012		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
2013						&bind_ipi) != 0)
2014			BUG();
2015		evtchn = bind_ipi.port;
2016
2017		/* Record the new mapping. */
2018		(void)xen_irq_info_ipi_setup(cpu, irq, evtchn, ipi);
2019		/* The affinity mask is still valid */
2020		bind_evtchn_to_cpu(evtchn, cpu, false);
2021	}
2022}
2023
2024/* Clear an irq's pending state, in preparation for polling on it */
2025void xen_clear_irq_pending(int irq)
2026{
2027	struct irq_info *info = info_for_irq(irq);
2028	evtchn_port_t evtchn = info ? info->evtchn : 0;
2029
2030	if (VALID_EVTCHN(evtchn))
2031		event_handler_exit(info);
2032}
2033EXPORT_SYMBOL(xen_clear_irq_pending);
2034void xen_set_irq_pending(int irq)
2035{
2036	evtchn_port_t evtchn = evtchn_from_irq(irq);
2037
2038	if (VALID_EVTCHN(evtchn))
2039		set_evtchn(evtchn);
2040}
2041
2042bool xen_test_irq_pending(int irq)
2043{
2044	evtchn_port_t evtchn = evtchn_from_irq(irq);
2045	bool ret = false;
2046
2047	if (VALID_EVTCHN(evtchn))
2048		ret = test_evtchn(evtchn);
2049
2050	return ret;
2051}
2052
2053/* Poll waiting for an irq to become pending with timeout.  In the usual case,
2054 * the irq will be disabled so it won't deliver an interrupt. */
2055void xen_poll_irq_timeout(int irq, u64 timeout)
2056{
2057	evtchn_port_t evtchn = evtchn_from_irq(irq);
2058
2059	if (VALID_EVTCHN(evtchn)) {
2060		struct sched_poll poll;
2061
2062		poll.nr_ports = 1;
2063		poll.timeout = timeout;
2064		set_xen_guest_handle(poll.ports, &evtchn);
2065
2066		if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
2067			BUG();
2068	}
2069}
2070EXPORT_SYMBOL(xen_poll_irq_timeout);
2071/* Poll waiting for an irq to become pending.  In the usual case, the
2072 * irq will be disabled so it won't deliver an interrupt. */
2073void xen_poll_irq(int irq)
2074{
2075	xen_poll_irq_timeout(irq, 0 /* no timeout */);
2076}
2077
2078/* Check whether the IRQ line is shared with other guests. */
2079int xen_test_irq_shared(int irq)
2080{
2081	struct irq_info *info = info_for_irq(irq);
2082	struct physdev_irq_status_query irq_status;
2083
2084	if (WARN_ON(!info))
2085		return -ENOENT;
2086
2087	irq_status.irq = info->u.pirq.pirq;
2088
2089	if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
2090		return 0;
2091	return !(irq_status.flags & XENIRQSTAT_shared);
2092}
2093EXPORT_SYMBOL_GPL(xen_test_irq_shared);
2094
2095void xen_irq_resume(void)
2096{
2097	unsigned int cpu;
2098	struct irq_info *info;
2099
2100	/* New event-channel space is not 'live' yet. */
2101	xen_evtchn_resume();
2102
2103	/* No IRQ <-> event-channel mappings. */
2104	list_for_each_entry(info, &xen_irq_list_head, list) {
2105		/* Zap event-channel binding */
2106		info->evtchn = 0;
2107		/* Adjust accounting */
2108		channels_on_cpu_dec(info);
2109	}
2110
2111	clear_evtchn_to_irq_all();
2112
2113	for_each_possible_cpu(cpu) {
2114		restore_cpu_virqs(cpu);
2115		restore_cpu_ipis(cpu);
2116	}
2117
2118	restore_pirqs();
2119}
2120
2121static struct irq_chip xen_dynamic_chip __read_mostly = {
2122	.name			= "xen-dyn",
2123
2124	.irq_disable		= disable_dynirq,
2125	.irq_mask		= disable_dynirq,
2126	.irq_unmask		= enable_dynirq,
2127
2128	.irq_ack		= ack_dynirq,
2129	.irq_mask_ack		= mask_ack_dynirq,
2130
2131	.irq_set_affinity	= set_affinity_irq,
2132	.irq_retrigger		= retrigger_dynirq,
2133};
2134
2135static struct irq_chip xen_lateeoi_chip __read_mostly = {
2136	/* The chip name needs to contain "xen-dyn" for irqbalance to work. */
2137	.name			= "xen-dyn-lateeoi",
2138
2139	.irq_disable		= disable_dynirq,
2140	.irq_mask		= disable_dynirq,
2141	.irq_unmask		= enable_dynirq,
2142
2143	.irq_ack		= lateeoi_ack_dynirq,
2144	.irq_mask_ack		= lateeoi_mask_ack_dynirq,
2145
2146	.irq_set_affinity	= set_affinity_irq,
2147	.irq_retrigger		= retrigger_dynirq,
2148};
2149
2150static struct irq_chip xen_pirq_chip __read_mostly = {
2151	.name			= "xen-pirq",
2152
2153	.irq_startup		= startup_pirq,
2154	.irq_shutdown		= shutdown_pirq,
2155	.irq_enable		= enable_pirq,
2156	.irq_disable		= disable_pirq,
2157
2158	.irq_mask		= disable_dynirq,
2159	.irq_unmask		= enable_dynirq,
2160
2161	.irq_ack		= eoi_pirq,
2162	.irq_eoi		= eoi_pirq,
2163	.irq_mask_ack		= mask_ack_pirq,
2164
2165	.irq_set_affinity	= set_affinity_irq,
2166
2167	.irq_retrigger		= retrigger_dynirq,
2168};
2169
2170static struct irq_chip xen_percpu_chip __read_mostly = {
2171	.name			= "xen-percpu",
2172
2173	.irq_disable		= disable_dynirq,
2174	.irq_mask		= disable_dynirq,
2175	.irq_unmask		= enable_dynirq,
2176
2177	.irq_ack		= ack_dynirq,
2178};
2179
2180#ifdef CONFIG_XEN_PVHVM
2181/* Vector callbacks are better than PCI interrupts to receive event
2182 * channel notifications because we can receive vector callbacks on any
2183 * vcpu and we don't need PCI support or APIC interactions. */
2184void xen_setup_callback_vector(void)
2185{
2186	uint64_t callback_via;
2187
2188	if (xen_have_vector_callback) {
2189		callback_via = HVM_CALLBACK_VECTOR(HYPERVISOR_CALLBACK_VECTOR);
2190		if (xen_set_callback_via(callback_via)) {
2191			pr_err("Request for Xen HVM callback vector failed\n");
2192			xen_have_vector_callback = 0;
2193		}
2194	}
2195}
2196
2197static __init void xen_alloc_callback_vector(void)
2198{
2199	if (!xen_have_vector_callback)
2200		return;
2201
2202	pr_info("Xen HVM callback vector for event delivery is enabled\n");
2203	alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_xen_hvm_callback);
2204}
2205#else
2206void xen_setup_callback_vector(void) {}
2207static inline void xen_alloc_callback_vector(void) {}
2208#endif
2209
2210bool xen_fifo_events = true;
2211module_param_named(fifo_events, xen_fifo_events, bool, 0);
2212
2213static int xen_evtchn_cpu_prepare(unsigned int cpu)
2214{
2215	int ret = 0;
2216
2217	xen_cpu_init_eoi(cpu);
2218
2219	if (evtchn_ops->percpu_init)
2220		ret = evtchn_ops->percpu_init(cpu);
2221
2222	return ret;
2223}
2224
2225static int xen_evtchn_cpu_dead(unsigned int cpu)
2226{
2227	int ret = 0;
2228
2229	if (evtchn_ops->percpu_deinit)
2230		ret = evtchn_ops->percpu_deinit(cpu);
2231
2232	return ret;
2233}
2234
2235void __init xen_init_IRQ(void)
2236{
2237	int ret = -EINVAL;
2238	evtchn_port_t evtchn;
2239
2240	if (xen_fifo_events)
2241		ret = xen_evtchn_fifo_init();
2242	if (ret < 0) {
2243		xen_evtchn_2l_init();
2244		xen_fifo_events = false;
2245	}
2246
2247	xen_cpu_init_eoi(smp_processor_id());
2248
2249	cpuhp_setup_state_nocalls(CPUHP_XEN_EVTCHN_PREPARE,
2250				  "xen/evtchn:prepare",
2251				  xen_evtchn_cpu_prepare, xen_evtchn_cpu_dead);
2252
2253	evtchn_to_irq = kcalloc(EVTCHN_ROW(xen_evtchn_max_channels()),
2254				sizeof(*evtchn_to_irq), GFP_KERNEL);
2255	BUG_ON(!evtchn_to_irq);
2256
2257	/* No event channels are 'live' right now. */
2258	for (evtchn = 0; evtchn < xen_evtchn_nr_channels(); evtchn++)
2259		mask_evtchn(evtchn);
2260
2261	pirq_needs_eoi = pirq_needs_eoi_flag;
2262
2263#ifdef CONFIG_X86
2264	if (xen_pv_domain()) {
2265		if (xen_initial_domain())
2266			pci_xen_initial_domain();
2267	}
2268	if (xen_feature(XENFEAT_hvm_callback_vector)) {
2269		xen_setup_callback_vector();
2270		xen_alloc_callback_vector();
2271	}
2272
2273	if (xen_hvm_domain()) {
2274		native_init_IRQ();
2275		/* pci_xen_hvm_init must be called after native_init_IRQ so that
2276		 * __acpi_register_gsi can point at the right function */
2277		pci_xen_hvm_init();
2278	} else {
2279		int rc;
2280		struct physdev_pirq_eoi_gmfn eoi_gmfn;
2281
2282		pirq_eoi_map = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
2283		eoi_gmfn.gmfn = virt_to_gfn(pirq_eoi_map);
2284		rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
2285		if (rc != 0) {
2286			free_page((unsigned long) pirq_eoi_map);
2287			pirq_eoi_map = NULL;
2288		} else
2289			pirq_needs_eoi = pirq_check_eoi_map;
2290	}
2291#endif
2292}