msi.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2014 Intel Corp.
   4 * Author: Jiang Liu <jiang.liu@linux.intel.com>
   5 *
   6 * This file is licensed under GPLv2.
   7 *
   8 * This file contains common code to support Message Signaled Interrupts for
   9 * PCI compatible and non PCI compatible devices.
  10 */
  11#include <linux/device.h>
  12#include <linux/irq.h>
  13#include <linux/irqdomain.h>
  14#include <linux/msi.h>
  15#include <linux/mutex.h>
  16#include <linux/pci.h>
  17#include <linux/slab.h>
  18#include <linux/seq_file.h>
  19#include <linux/sysfs.h>
  20#include <linux/types.h>
  21#include <linux/xarray.h>
  22
  23#include "internals.h"
  24
  25/**
  26 * struct msi_device_data - MSI per device data
  27 * @properties:		MSI properties which are interesting to drivers
  28 * @mutex:		Mutex protecting the MSI descriptor store
  29 * @__domains:		Internal data for per device MSI domains
  30 * @__iter_idx:		Index to search the next entry for iterators
  31 */
  32struct msi_device_data {
  33	unsigned long			properties;
  34	struct mutex			mutex;
  35	struct msi_dev_domain		__domains[MSI_MAX_DEVICE_IRQDOMAINS];
  36	unsigned long			__iter_idx;
  37};
  38
  39/**
  40 * struct msi_ctrl - MSI internal management control structure
  41 * @domid:	ID of the domain on which management operations should be done
  42 * @first:	First (hardware) slot index to operate on
  43 * @last:	Last (hardware) slot index to operate on
  44 * @nirqs:	The number of Linux interrupts to allocate. Can be larger
  45 *		than the range due to PCI/multi-MSI.
  46 */
  47struct msi_ctrl {
  48	unsigned int			domid;
  49	unsigned int			first;
  50	unsigned int			last;
  51	unsigned int			nirqs;
  52};
  53
  54/* Invalid Xarray index which is outside of any searchable range */
  55#define MSI_XA_MAX_INDEX	(ULONG_MAX - 1)
  56/* The maximum domain size */
  57#define MSI_XA_DOMAIN_SIZE	(MSI_MAX_INDEX + 1)
  58
  59static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl);
  60static unsigned int msi_domain_get_hwsize(struct device *dev, unsigned int domid);
  61static inline int msi_sysfs_create_group(struct device *dev);
  62static int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
  63				   int nvec, msi_alloc_info_t *arg);
  64
  65/**
  66 * msi_alloc_desc - Allocate an initialized msi_desc
  67 * @dev:	Pointer to the device for which this is allocated
  68 * @nvec:	The number of vectors used in this entry
  69 * @affinity:	Optional pointer to an affinity mask array size of @nvec
  70 *
  71 * If @affinity is not %NULL then an affinity array[@nvec] is allocated
  72 * and the affinity masks and flags from @affinity are copied.
  73 *
  74 * Return: pointer to allocated &msi_desc on success or %NULL on failure
  75 */
  76static struct msi_desc *msi_alloc_desc(struct device *dev, int nvec,
  77				       const struct irq_affinity_desc *affinity)
  78{
  79	struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
  80
  81	if (!desc)
  82		return NULL;
  83
  84	desc->dev = dev;
  85	desc->nvec_used = nvec;
  86	if (affinity) {
  87		desc->affinity = kmemdup_array(affinity, nvec, sizeof(*desc->affinity), GFP_KERNEL);
  88		if (!desc->affinity) {
  89			kfree(desc);
  90			return NULL;
  91		}
  92	}
  93	return desc;
  94}
  95
  96static void msi_free_desc(struct msi_desc *desc)
  97{
  98	kfree(desc->affinity);
  99	kfree(desc);
 100}
 101
 102static int msi_insert_desc(struct device *dev, struct msi_desc *desc,
 103			   unsigned int domid, unsigned int index)
 104{
 105	struct msi_device_data *md = dev->msi.data;
 106	struct xarray *xa = &md->__domains[domid].store;
 107	unsigned int hwsize;
 108	int ret;
 109
 110	hwsize = msi_domain_get_hwsize(dev, domid);
 111
 112	if (index == MSI_ANY_INDEX) {
 113		struct xa_limit limit = { .min = 0, .max = hwsize - 1 };
 114		unsigned int index;
 115
 116		/* Let the xarray allocate a free index within the limit */
 117		ret = xa_alloc(xa, &index, desc, limit, GFP_KERNEL);
 118		if (ret)
 119			goto fail;
 120
 121		desc->msi_index = index;
 122		return 0;
 123	} else {
 124		if (index >= hwsize) {
 125			ret = -ERANGE;
 126			goto fail;
 127		}
 128
 129		desc->msi_index = index;
 130		ret = xa_insert(xa, index, desc, GFP_KERNEL);
 131		if (ret)
 132			goto fail;
 133		return 0;
 134	}
 135fail:
 136	msi_free_desc(desc);
 137	return ret;
 138}
 139
 140/**
 141 * msi_domain_insert_msi_desc - Allocate and initialize a MSI descriptor and
 142 *				insert it at @init_desc->msi_index
 143 *
 144 * @dev:	Pointer to the device for which the descriptor is allocated
 145 * @domid:	The id of the interrupt domain to which the desriptor is added
 146 * @init_desc:	Pointer to an MSI descriptor to initialize the new descriptor
 147 *
 148 * Return: 0 on success or an appropriate failure code.
 149 */
 150int msi_domain_insert_msi_desc(struct device *dev, unsigned int domid,
 151			       struct msi_desc *init_desc)
 152{
 153	struct msi_desc *desc;
 154
 155	lockdep_assert_held(&dev->msi.data->mutex);
 156
 157	desc = msi_alloc_desc(dev, init_desc->nvec_used, init_desc->affinity);
 158	if (!desc)
 159		return -ENOMEM;
 160
 161	/* Copy type specific data to the new descriptor. */
 162	desc->pci = init_desc->pci;
 163
 164	return msi_insert_desc(dev, desc, domid, init_desc->msi_index);
 165}
 166
 167static bool msi_desc_match(struct msi_desc *desc, enum msi_desc_filter filter)
 168{
 169	switch (filter) {
 170	case MSI_DESC_ALL:
 171		return true;
 172	case MSI_DESC_NOTASSOCIATED:
 173		return !desc->irq;
 174	case MSI_DESC_ASSOCIATED:
 175		return !!desc->irq;
 176	}
 177	WARN_ON_ONCE(1);
 178	return false;
 179}
 180
 181static bool msi_ctrl_valid(struct device *dev, struct msi_ctrl *ctrl)
 182{
 183	unsigned int hwsize;
 184
 185	if (WARN_ON_ONCE(ctrl->domid >= MSI_MAX_DEVICE_IRQDOMAINS ||
 186			 (dev->msi.domain &&
 187			  !dev->msi.data->__domains[ctrl->domid].domain)))
 188		return false;
 189
 190	hwsize = msi_domain_get_hwsize(dev, ctrl->domid);
 191	if (WARN_ON_ONCE(ctrl->first > ctrl->last ||
 192			 ctrl->first >= hwsize ||
 193			 ctrl->last >= hwsize))
 194		return false;
 195	return true;
 196}
 197
 198static void msi_domain_free_descs(struct device *dev, struct msi_ctrl *ctrl)
 199{
 200	struct msi_desc *desc;
 201	struct xarray *xa;
 202	unsigned long idx;
 203
 204	lockdep_assert_held(&dev->msi.data->mutex);
 205
 206	if (!msi_ctrl_valid(dev, ctrl))
 207		return;
 208
 209	xa = &dev->msi.data->__domains[ctrl->domid].store;
 210	xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
 211		xa_erase(xa, idx);
 212
 213		/* Leak the descriptor when it is still referenced */
 214		if (WARN_ON_ONCE(msi_desc_match(desc, MSI_DESC_ASSOCIATED)))
 215			continue;
 216		msi_free_desc(desc);
 217	}
 218}
 219
 220/**
 221 * msi_domain_free_msi_descs_range - Free a range of MSI descriptors of a device in an irqdomain
 222 * @dev:	Device for which to free the descriptors
 223 * @domid:	Id of the domain to operate on
 224 * @first:	Index to start freeing from (inclusive)
 225 * @last:	Last index to be freed (inclusive)
 226 */
 227void msi_domain_free_msi_descs_range(struct device *dev, unsigned int domid,
 228				     unsigned int first, unsigned int last)
 229{
 230	struct msi_ctrl ctrl = {
 231		.domid	= domid,
 232		.first	= first,
 233		.last	= last,
 234	};
 235
 236	msi_domain_free_descs(dev, &ctrl);
 237}
 238
 239/**
 240 * msi_domain_add_simple_msi_descs - Allocate and initialize MSI descriptors
 241 * @dev:	Pointer to the device for which the descriptors are allocated
 242 * @ctrl:	Allocation control struct
 243 *
 244 * Return: 0 on success or an appropriate failure code.
 245 */
 246static int msi_domain_add_simple_msi_descs(struct device *dev, struct msi_ctrl *ctrl)
 247{
 248	struct msi_desc *desc;
 249	unsigned int idx;
 250	int ret;
 251
 252	lockdep_assert_held(&dev->msi.data->mutex);
 253
 254	if (!msi_ctrl_valid(dev, ctrl))
 255		return -EINVAL;
 256
 257	for (idx = ctrl->first; idx <= ctrl->last; idx++) {
 258		desc = msi_alloc_desc(dev, 1, NULL);
 259		if (!desc)
 260			goto fail_mem;
 261		ret = msi_insert_desc(dev, desc, ctrl->domid, idx);
 262		if (ret)
 263			goto fail;
 264	}
 265	return 0;
 266
 267fail_mem:
 268	ret = -ENOMEM;
 269fail:
 270	msi_domain_free_descs(dev, ctrl);
 271	return ret;
 272}
 273
 274void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
 275{
 276	*msg = entry->msg;
 277}
 278
 279void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
 280{
 281	struct msi_desc *entry = irq_get_msi_desc(irq);
 282
 283	__get_cached_msi_msg(entry, msg);
 284}
 285EXPORT_SYMBOL_GPL(get_cached_msi_msg);
 286
 287static void msi_device_data_release(struct device *dev, void *res)
 288{
 289	struct msi_device_data *md = res;
 290	int i;
 291
 292	for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++) {
 293		msi_remove_device_irq_domain(dev, i);
 294		WARN_ON_ONCE(!xa_empty(&md->__domains[i].store));
 295		xa_destroy(&md->__domains[i].store);
 296	}
 297	dev->msi.data = NULL;
 298}
 299
 300/**
 301 * msi_setup_device_data - Setup MSI device data
 302 * @dev:	Device for which MSI device data should be set up
 303 *
 304 * Return: 0 on success, appropriate error code otherwise
 305 *
 306 * This can be called more than once for @dev. If the MSI device data is
 307 * already allocated the call succeeds. The allocated memory is
 308 * automatically released when the device is destroyed.
 309 */
 310int msi_setup_device_data(struct device *dev)
 311{
 312	struct msi_device_data *md;
 313	int ret, i;
 314
 315	if (dev->msi.data)
 316		return 0;
 317
 318	md = devres_alloc(msi_device_data_release, sizeof(*md), GFP_KERNEL);
 319	if (!md)
 320		return -ENOMEM;
 321
 322	ret = msi_sysfs_create_group(dev);
 323	if (ret) {
 324		devres_free(md);
 325		return ret;
 326	}
 327
 328	for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++)
 329		xa_init_flags(&md->__domains[i].store, XA_FLAGS_ALLOC);
 330
 331	/*
 332	 * If @dev::msi::domain is set and is a global MSI domain, copy the
 333	 * pointer into the domain array so all code can operate on domain
 334	 * ids. The NULL pointer check is required to keep the legacy
 335	 * architecture specific PCI/MSI support working.
 336	 */
 337	if (dev->msi.domain && !irq_domain_is_msi_parent(dev->msi.domain))
 338		md->__domains[MSI_DEFAULT_DOMAIN].domain = dev->msi.domain;
 339
 340	mutex_init(&md->mutex);
 341	dev->msi.data = md;
 342	devres_add(dev, md);
 343	return 0;
 344}
 345
 346/**
 347 * __msi_lock_descs - Lock the MSI descriptor storage of a device
 348 * @dev:	Device to operate on
 349 *
 350 * Internal function for guard(msi_descs_lock). Don't use in code.
 351 */
 352void __msi_lock_descs(struct device *dev)
 353{
 354	mutex_lock(&dev->msi.data->mutex);
 355}
 356EXPORT_SYMBOL_GPL(__msi_lock_descs);
 357
 358/**
 359 * __msi_unlock_descs - Unlock the MSI descriptor storage of a device
 360 * @dev:	Device to operate on
 361 *
 362 * Internal function for guard(msi_descs_lock). Don't use in code.
 363 */
 364void __msi_unlock_descs(struct device *dev)
 365{
 366	/* Invalidate the index which was cached by the iterator */
 367	dev->msi.data->__iter_idx = MSI_XA_MAX_INDEX;
 368	mutex_unlock(&dev->msi.data->mutex);
 369}
 370EXPORT_SYMBOL_GPL(__msi_unlock_descs);
 371
 372static struct msi_desc *msi_find_desc(struct msi_device_data *md, unsigned int domid,
 373				      enum msi_desc_filter filter)
 374{
 375	struct xarray *xa = &md->__domains[domid].store;
 376	struct msi_desc *desc;
 377
 378	xa_for_each_start(xa, md->__iter_idx, desc, md->__iter_idx) {
 379		if (msi_desc_match(desc, filter))
 380			return desc;
 381	}
 382	md->__iter_idx = MSI_XA_MAX_INDEX;
 383	return NULL;
 384}
 385
 386/**
 387 * msi_domain_first_desc - Get the first MSI descriptor of an irqdomain associated to a device
 388 * @dev:	Device to operate on
 389 * @domid:	The id of the interrupt domain which should be walked.
 390 * @filter:	Descriptor state filter
 391 *
 392 * Must be called with the MSI descriptor mutex held, i.e. msi_lock_descs()
 393 * must be invoked before the call.
 394 *
 395 * Return: Pointer to the first MSI descriptor matching the search
 396 *	   criteria, NULL if none found.
 397 */
 398struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
 399				       enum msi_desc_filter filter)
 400{
 401	struct msi_device_data *md = dev->msi.data;
 402
 403	if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
 404		return NULL;
 405
 406	lockdep_assert_held(&md->mutex);
 407
 408	md->__iter_idx = 0;
 409	return msi_find_desc(md, domid, filter);
 410}
 411EXPORT_SYMBOL_GPL(msi_domain_first_desc);
 412
 413/**
 414 * msi_next_desc - Get the next MSI descriptor of a device
 415 * @dev:	Device to operate on
 416 * @domid:	The id of the interrupt domain which should be walked.
 417 * @filter:	Descriptor state filter
 418 *
 419 * The first invocation of msi_next_desc() has to be preceeded by a
 420 * successful invocation of __msi_first_desc(). Consecutive invocations are
 421 * only valid if the previous one was successful. All these operations have
 422 * to be done within the same MSI mutex held region.
 423 *
 424 * Return: Pointer to the next MSI descriptor matching the search
 425 *	   criteria, NULL if none found.
 426 */
 427struct msi_desc *msi_next_desc(struct device *dev, unsigned int domid,
 428			       enum msi_desc_filter filter)
 429{
 430	struct msi_device_data *md = dev->msi.data;
 431
 432	if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
 433		return NULL;
 434
 435	lockdep_assert_held(&md->mutex);
 436
 437	if (md->__iter_idx >= (unsigned long)MSI_MAX_INDEX)
 438		return NULL;
 439
 440	md->__iter_idx++;
 441	return msi_find_desc(md, domid, filter);
 442}
 443EXPORT_SYMBOL_GPL(msi_next_desc);
 444
 445/**
 446 * msi_domain_get_virq - Lookup the Linux interrupt number for a MSI index on a interrupt domain
 447 * @dev:	Device to operate on
 448 * @domid:	Domain ID of the interrupt domain associated to the device
 449 * @index:	MSI interrupt index to look for (0-based)
 450 *
 451 * Return: The Linux interrupt number on success (> 0), 0 if not found
 452 */
 453unsigned int msi_domain_get_virq(struct device *dev, unsigned int domid, unsigned int index)
 454{
 455	struct msi_desc *desc;
 456	bool pcimsi = false;
 457	struct xarray *xa;
 458
 459	if (!dev->msi.data)
 460		return 0;
 461
 462	if (WARN_ON_ONCE(index > MSI_MAX_INDEX || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
 463		return 0;
 464
 465	/* This check is only valid for the PCI default MSI domain */
 466	if (dev_is_pci(dev) && domid == MSI_DEFAULT_DOMAIN)
 467		pcimsi = to_pci_dev(dev)->msi_enabled;
 468
 469	guard(msi_descs_lock)(dev);
 470	xa = &dev->msi.data->__domains[domid].store;
 471	desc = xa_load(xa, pcimsi ? 0 : index);
 472	if (desc && desc->irq) {
 473		/*
 474		 * PCI-MSI has only one descriptor for multiple interrupts.
 475		 * PCI-MSIX and platform MSI use a descriptor per
 476		 * interrupt.
 477		 */
 478		if (!pcimsi)
 479			return desc->irq;
 480		if (index < desc->nvec_used)
 481			return desc->irq + index;
 482	}
 483	return 0;
 484}
 485EXPORT_SYMBOL_GPL(msi_domain_get_virq);
 486
 487#ifdef CONFIG_SYSFS
 488static struct attribute *msi_dev_attrs[] = {
 489	NULL
 490};
 491
 492static const struct attribute_group msi_irqs_group = {
 493	.name	= "msi_irqs",
 494	.attrs	= msi_dev_attrs,
 495};
 496
 497static inline int msi_sysfs_create_group(struct device *dev)
 498{
 499	return devm_device_add_group(dev, &msi_irqs_group);
 500}
 501
 502static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
 503			     char *buf)
 504{
 505	/* MSI vs. MSIX is per device not per interrupt */
 506	bool is_msix = dev_is_pci(dev) ? to_pci_dev(dev)->msix_enabled : false;
 507
 508	return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
 509}
 510
 511static void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc)
 512{
 513	struct device_attribute *attrs = desc->sysfs_attrs;
 514	int i;
 515
 516	if (!attrs)
 517		return;
 518
 519	desc->sysfs_attrs = NULL;
 520	for (i = 0; i < desc->nvec_used; i++) {
 521		if (attrs[i].show)
 522			sysfs_remove_file_from_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
 523		kfree(attrs[i].attr.name);
 524	}
 525	kfree(attrs);
 526}
 527
 528static int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc)
 529{
 530	struct device_attribute *attrs;
 531	int ret, i;
 532
 533	attrs = kcalloc(desc->nvec_used, sizeof(*attrs), GFP_KERNEL);
 534	if (!attrs)
 535		return -ENOMEM;
 536
 537	desc->sysfs_attrs = attrs;
 538	for (i = 0; i < desc->nvec_used; i++) {
 539		sysfs_attr_init(&attrs[i].attr);
 540		attrs[i].attr.name = kasprintf(GFP_KERNEL, "%d", desc->irq + i);
 541		if (!attrs[i].attr.name) {
 542			ret = -ENOMEM;
 543			goto fail;
 544		}
 545
 546		attrs[i].attr.mode = 0444;
 547		attrs[i].show = msi_mode_show;
 548
 549		ret = sysfs_add_file_to_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
 550		if (ret) {
 551			attrs[i].show = NULL;
 552			goto fail;
 553		}
 554	}
 555	return 0;
 556
 557fail:
 558	msi_sysfs_remove_desc(dev, desc);
 559	return ret;
 560}
 561
 562#if defined(CONFIG_PCI_MSI_ARCH_FALLBACKS) || defined(CONFIG_PCI_XEN)
 563/**
 564 * msi_device_populate_sysfs - Populate msi_irqs sysfs entries for a device
 565 * @dev:	The device (PCI, platform etc) which will get sysfs entries
 566 */
 567int msi_device_populate_sysfs(struct device *dev)
 568{
 569	struct msi_desc *desc;
 570	int ret;
 571
 572	msi_for_each_desc(desc, dev, MSI_DESC_ASSOCIATED) {
 573		if (desc->sysfs_attrs)
 574			continue;
 575		ret = msi_sysfs_populate_desc(dev, desc);
 576		if (ret)
 577			return ret;
 578	}
 579	return 0;
 580}
 581
 582/**
 583 * msi_device_destroy_sysfs - Destroy msi_irqs sysfs entries for a device
 584 * @dev:		The device (PCI, platform etc) for which to remove
 585 *			sysfs entries
 586 */
 587void msi_device_destroy_sysfs(struct device *dev)
 588{
 589	struct msi_desc *desc;
 590
 591	msi_for_each_desc(desc, dev, MSI_DESC_ALL)
 592		msi_sysfs_remove_desc(dev, desc);
 593}
 594#endif /* CONFIG_PCI_MSI_ARCH_FALLBACK || CONFIG_PCI_XEN */
 595#else /* CONFIG_SYSFS */
 596static inline int msi_sysfs_create_group(struct device *dev) { return 0; }
 597static inline int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc) { return 0; }
 598static inline void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc) { }
 599#endif /* !CONFIG_SYSFS */
 600
 601static struct irq_domain *msi_get_device_domain(struct device *dev, unsigned int domid)
 602{
 603	struct irq_domain *domain;
 604
 605	lockdep_assert_held(&dev->msi.data->mutex);
 606
 607	if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
 608		return NULL;
 609
 610	domain = dev->msi.data->__domains[domid].domain;
 611	if (!domain)
 612		return NULL;
 613
 614	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
 615		return NULL;
 616
 617	return domain;
 618}
 619
 620static unsigned int msi_domain_get_hwsize(struct device *dev, unsigned int domid)
 621{
 622	struct msi_domain_info *info;
 623	struct irq_domain *domain;
 624
 625	domain = msi_get_device_domain(dev, domid);
 626	if (domain) {
 627		info = domain->host_data;
 628		return info->hwsize;
 629	}
 630	/* No domain, default to MSI_XA_DOMAIN_SIZE */
 631	return MSI_XA_DOMAIN_SIZE;
 632}
 633
 634static inline void irq_chip_write_msi_msg(struct irq_data *data,
 635					  struct msi_msg *msg)
 636{
 637	data->chip->irq_write_msi_msg(data, msg);
 638}
 639
 640static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
 641{
 642	struct msi_domain_info *info = domain->host_data;
 643
 644	/*
 645	 * If the MSI provider has messed with the second message and
 646	 * not advertized that it is level-capable, signal the breakage.
 647	 */
 648	WARN_ON(!((info->flags & MSI_FLAG_LEVEL_CAPABLE) &&
 649		  (info->chip->flags & IRQCHIP_SUPPORTS_LEVEL_MSI)) &&
 650		(msg[1].address_lo || msg[1].address_hi || msg[1].data));
 651}
 652
 653/**
 654 * msi_domain_set_affinity - Generic affinity setter function for MSI domains
 655 * @irq_data:	The irq data associated to the interrupt
 656 * @mask:	The affinity mask to set
 657 * @force:	Flag to enforce setting (disable online checks)
 658 *
 659 * Intended to be used by MSI interrupt controllers which are
 660 * implemented with hierarchical domains.
 661 *
 662 * Return: IRQ_SET_MASK_* result code
 663 */
 664int msi_domain_set_affinity(struct irq_data *irq_data,
 665			    const struct cpumask *mask, bool force)
 666{
 667	struct irq_data *parent = irq_data->parent_data;
 668	struct msi_msg msg[2] = { [1] = { }, };
 669	int ret;
 670
 671	ret = parent->chip->irq_set_affinity(parent, mask, force);
 672	if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
 673		BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
 674		msi_check_level(irq_data->domain, msg);
 675		irq_chip_write_msi_msg(irq_data, msg);
 676	}
 677
 678	return ret;
 679}
 680
 681static int msi_domain_activate(struct irq_domain *domain,
 682			       struct irq_data *irq_data, bool early)
 683{
 684	struct msi_msg msg[2] = { [1] = { }, };
 685
 686	BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
 687	msi_check_level(irq_data->domain, msg);
 688	irq_chip_write_msi_msg(irq_data, msg);
 689	return 0;
 690}
 691
 692static void msi_domain_deactivate(struct irq_domain *domain,
 693				  struct irq_data *irq_data)
 694{
 695	struct msi_msg msg[2];
 696
 697	memset(msg, 0, sizeof(msg));
 698	irq_chip_write_msi_msg(irq_data, msg);
 699}
 700
 701static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
 702			    unsigned int nr_irqs, void *arg)
 703{
 704	struct msi_domain_info *info = domain->host_data;
 705	struct msi_domain_ops *ops = info->ops;
 706	irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
 707	int i, ret;
 708
 709	if (irq_resolve_mapping(domain, hwirq))
 710		return -EEXIST;
 711
 712	if (domain->parent) {
 713		ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
 714		if (ret < 0)
 715			return ret;
 716	}
 717
 718	for (i = 0; i < nr_irqs; i++) {
 719		ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
 720		if (ret < 0) {
 721			if (ops->msi_free) {
 722				for (i--; i >= 0; i--)
 723					ops->msi_free(domain, info, virq + i);
 724			}
 725			irq_domain_free_irqs_top(domain, virq, nr_irqs);
 726			return ret;
 727		}
 728	}
 729
 730	return 0;
 731}
 732
 733static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
 734			    unsigned int nr_irqs)
 735{
 736	struct msi_domain_info *info = domain->host_data;
 737	int i;
 738
 739	if (info->ops->msi_free) {
 740		for (i = 0; i < nr_irqs; i++)
 741			info->ops->msi_free(domain, info, virq + i);
 742	}
 743	irq_domain_free_irqs_top(domain, virq, nr_irqs);
 744}
 745
 746static int msi_domain_translate(struct irq_domain *domain, struct irq_fwspec *fwspec,
 747				irq_hw_number_t *hwirq, unsigned int *type)
 748{
 749	struct msi_domain_info *info = domain->host_data;
 750
 751	/*
 752	 * This will catch allocations through the regular irqdomain path except
 753	 * for MSI domains which really support this, e.g. MBIGEN.
 754	 */
 755	if (!info->ops->msi_translate)
 756		return -ENOTSUPP;
 757	return info->ops->msi_translate(domain, fwspec, hwirq, type);
 758}
 759
 760#ifdef CONFIG_GENERIC_IRQ_DEBUGFS
 761static void msi_domain_debug_show(struct seq_file *m, struct irq_domain *d,
 762				  struct irq_data *irqd, int ind)
 763{
 764	struct msi_desc *desc = irqd ? irq_data_get_msi_desc(irqd) : NULL;
 765
 766	if (!desc)
 767		return;
 768
 769	seq_printf(m, "\n%*saddress_hi: 0x%08x", ind + 1, "", desc->msg.address_hi);
 770	seq_printf(m, "\n%*saddress_lo: 0x%08x", ind + 1, "", desc->msg.address_lo);
 771	seq_printf(m, "\n%*smsg_data:   0x%08x\n", ind + 1, "", desc->msg.data);
 772}
 773#endif
 774
 775static const struct irq_domain_ops msi_domain_ops = {
 776	.alloc		= msi_domain_alloc,
 777	.free		= msi_domain_free,
 778	.activate	= msi_domain_activate,
 779	.deactivate	= msi_domain_deactivate,
 780	.translate	= msi_domain_translate,
 781#ifdef CONFIG_GENERIC_IRQ_DEBUGFS
 782	.debug_show     = msi_domain_debug_show,
 783#endif
 784};
 785
 786static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
 787						msi_alloc_info_t *arg)
 788{
 789	return arg->hwirq;
 790}
 791
 792static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
 793				  int nvec, msi_alloc_info_t *arg)
 794{
 795	memset(arg, 0, sizeof(*arg));
 796	return 0;
 797}
 798
 799static void msi_domain_ops_teardown(struct irq_domain *domain, msi_alloc_info_t *arg)
 800{
 801}
 802
 803static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
 804				    struct msi_desc *desc)
 805{
 806	arg->desc = desc;
 807}
 808
 809static int msi_domain_ops_init(struct irq_domain *domain,
 810			       struct msi_domain_info *info,
 811			       unsigned int virq, irq_hw_number_t hwirq,
 812			       msi_alloc_info_t *arg)
 813{
 814	irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
 815				      info->chip_data);
 816	if (info->handler && info->handler_name) {
 817		__irq_set_handler(virq, info->handler, 0, info->handler_name);
 818		if (info->handler_data)
 819			irq_set_handler_data(virq, info->handler_data);
 820	}
 821	return 0;
 822}
 823
 824static struct msi_domain_ops msi_domain_ops_default = {
 825	.get_hwirq		= msi_domain_ops_get_hwirq,
 826	.msi_init		= msi_domain_ops_init,
 827	.msi_prepare		= msi_domain_ops_prepare,
 828	.msi_teardown		= msi_domain_ops_teardown,
 829	.set_desc		= msi_domain_ops_set_desc,
 830};
 831
 832static void msi_domain_update_dom_ops(struct msi_domain_info *info)
 833{
 834	struct msi_domain_ops *ops = info->ops;
 835
 836	if (ops == NULL) {
 837		info->ops = &msi_domain_ops_default;
 838		return;
 839	}
 840
 841	if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
 842		return;
 843
 844	if (ops->get_hwirq == NULL)
 845		ops->get_hwirq = msi_domain_ops_default.get_hwirq;
 846	if (ops->msi_init == NULL)
 847		ops->msi_init = msi_domain_ops_default.msi_init;
 848	if (ops->msi_prepare == NULL)
 849		ops->msi_prepare = msi_domain_ops_default.msi_prepare;
 850	if (ops->msi_teardown == NULL)
 851		ops->msi_teardown = msi_domain_ops_default.msi_teardown;
 852	if (ops->set_desc == NULL)
 853		ops->set_desc = msi_domain_ops_default.set_desc;
 854}
 855
 856static void msi_domain_update_chip_ops(struct msi_domain_info *info)
 857{
 858	struct irq_chip *chip = info->chip;
 859
 860	BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
 861	if (!chip->irq_set_affinity && !(info->flags & MSI_FLAG_NO_AFFINITY))
 862		chip->irq_set_affinity = msi_domain_set_affinity;
 863}
 864
 865static struct irq_domain *__msi_create_irq_domain(struct fwnode_handle *fwnode,
 866						  struct msi_domain_info *info,
 867						  unsigned int flags,
 868						  struct irq_domain *parent)
 869{
 870	struct irq_domain *domain;
 871
 872	if (info->hwsize > MSI_XA_DOMAIN_SIZE)
 873		return NULL;
 874
 875	/*
 876	 * Hardware size 0 is valid for backwards compatibility and for
 877	 * domains which are not backed by a hardware table. Grant the
 878	 * maximum index space.
 879	 */
 880	if (!info->hwsize)
 881		info->hwsize = MSI_XA_DOMAIN_SIZE;
 882
 883	msi_domain_update_dom_ops(info);
 884	if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
 885		msi_domain_update_chip_ops(info);
 886
 887	domain = irq_domain_create_hierarchy(parent, flags | IRQ_DOMAIN_FLAG_MSI, 0,
 888					     fwnode, &msi_domain_ops, info);
 889
 890	if (domain) {
 891		irq_domain_update_bus_token(domain, info->bus_token);
 892		domain->dev = info->dev;
 893		if (info->flags & MSI_FLAG_PARENT_PM_DEV)
 894			domain->pm_dev = parent->pm_dev;
 895	}
 896
 897	return domain;
 898}
 899
 900/**
 901 * msi_create_irq_domain - Create an MSI interrupt domain
 902 * @fwnode:	Optional fwnode of the interrupt controller
 903 * @info:	MSI domain info
 904 * @parent:	Parent irq domain
 905 *
 906 * Return: pointer to the created &struct irq_domain or %NULL on failure
 907 */
 908struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
 909					 struct msi_domain_info *info,
 910					 struct irq_domain *parent)
 911{
 912	return __msi_create_irq_domain(fwnode, info, 0, parent);
 913}
 914
 915/**
 916 * msi_create_parent_irq_domain - Create an MSI-parent interrupt domain
 917 * @info:		MSI irqdomain creation info
 918 * @msi_parent_ops:	MSI parent callbacks and configuration
 919 *
 920 * Return: pointer to the created &struct irq_domain or %NULL on failure
 921 */
 922struct irq_domain *msi_create_parent_irq_domain(struct irq_domain_info *info,
 923						const struct msi_parent_ops *msi_parent_ops)
 924{
 925	struct irq_domain *d;
 926
 927	info->hwirq_max		= max(info->hwirq_max, info->size);
 928	info->size		= info->hwirq_max;
 929	info->domain_flags	|= IRQ_DOMAIN_FLAG_MSI_PARENT;
 930	info->bus_token		= msi_parent_ops->bus_select_token;
 931
 932	d = irq_domain_instantiate(info);
 933	if (IS_ERR(d))
 934		return NULL;
 935
 936	d->msi_parent_ops = msi_parent_ops;
 937	return d;
 938}
 939EXPORT_SYMBOL_GPL(msi_create_parent_irq_domain);
 940
 941/**
 942 * msi_parent_init_dev_msi_info - Delegate initialization of device MSI info down
 943 *				  in the domain hierarchy
 944 * @dev:		The device for which the domain should be created
 945 * @domain:		The domain in the hierarchy this op is being called on
 946 * @msi_parent_domain:	The IRQ_DOMAIN_FLAG_MSI_PARENT domain for the child to
 947 *			be created
 948 * @msi_child_info:	The MSI domain info of the IRQ_DOMAIN_FLAG_MSI_DEVICE
 949 *			domain to be created
 950 *
 951 * Return: true on success, false otherwise
 952 *
 953 * This is the most complex problem of per device MSI domains and the
 954 * underlying interrupt domain hierarchy:
 955 *
 956 * The device domain to be initialized requests the broadest feature set
 957 * possible and the underlying domain hierarchy puts restrictions on it.
 958 *
 959 * That's trivial for a simple parent->child relationship, but it gets
 960 * interesting with an intermediate domain: root->parent->child.  The
 961 * intermediate 'parent' can expand the capabilities which the 'root'
 962 * domain is providing. So that creates a classic hen and egg problem:
 963 * Which entity is doing the restrictions/expansions?
 964 *
 965 * One solution is to let the root domain handle the initialization that's
 966 * why there is the @domain and the @msi_parent_domain pointer.
 967 */
 968bool msi_parent_init_dev_msi_info(struct device *dev, struct irq_domain *domain,
 969				  struct irq_domain *msi_parent_domain,
 970				  struct msi_domain_info *msi_child_info)
 971{
 972	struct irq_domain *parent = domain->parent;
 973
 974	if (WARN_ON_ONCE(!parent || !parent->msi_parent_ops ||
 975			 !parent->msi_parent_ops->init_dev_msi_info))
 976		return false;
 977
 978	return parent->msi_parent_ops->init_dev_msi_info(dev, parent, msi_parent_domain,
 979							 msi_child_info);
 980}
 981
 982/**
 983 * msi_create_device_irq_domain - Create a device MSI interrupt domain
 984 * @dev:		Pointer to the device
 985 * @domid:		Domain id
 986 * @template:		MSI domain info bundle used as template
 987 * @hwsize:		Maximum number of MSI table entries (0 if unknown or unlimited)
 988 * @domain_data:	Optional pointer to domain specific data which is set in
 989 *			msi_domain_info::data
 990 * @chip_data:		Optional pointer to chip specific data which is set in
 991 *			msi_domain_info::chip_data
 992 *
 993 * Return: True on success, false otherwise
 994 *
 995 * There is no firmware node required for this interface because the per
 996 * device domains are software constructs which are actually closer to the
 997 * hardware reality than any firmware can describe them.
 998 *
 999 * The domain name and the irq chip name for a MSI device domain are
1000 * composed by: "$(PREFIX)$(CHIPNAME)-$(DEVNAME)"
1001 *
1002 * $PREFIX:   Optional prefix provided by the underlying MSI parent domain
1003 *	      via msi_parent_ops::prefix. If that pointer is NULL the prefix
1004 *	      is empty.
1005 * $CHIPNAME: The name of the irq_chip in @template
1006 * $DEVNAME:  The name of the device
1007 *
1008 * This results in understandable chip names and hardware interrupt numbers
1009 * in e.g. /proc/interrupts
1010 *
1011 * PCI-MSI-0000:00:1c.0     0-edge  Parent domain has no prefix
1012 * IR-PCI-MSI-0000:00:1c.4  0-edge  Same with interrupt remapping prefix 'IR-'
1013 *
1014 * IR-PCI-MSIX-0000:3d:00.0 0-edge  Hardware interrupt numbers reflect
1015 * IR-PCI-MSIX-0000:3d:00.0 1-edge  the real MSI-X index on that device
1016 * IR-PCI-MSIX-0000:3d:00.0 2-edge
1017 *
1018 * On IMS domains the hardware interrupt number is either a table entry
1019 * index or a purely software managed index but it is guaranteed to be
1020 * unique.
1021 *
1022 * The domain pointer is stored in @dev::msi::data::__irqdomains[]. All
1023 * subsequent operations on the domain depend on the domain id.
1024 *
1025 * The domain is automatically freed when the device is removed via devres
1026 * in the context of @dev::msi::data freeing, but it can also be
1027 * independently removed via @msi_remove_device_irq_domain().
1028 */
1029bool msi_create_device_irq_domain(struct device *dev, unsigned int domid,
1030				  const struct msi_domain_template *template,
1031				  unsigned int hwsize, void *domain_data,
1032				  void *chip_data)
1033{
1034	struct irq_domain *domain, *parent = dev->msi.domain;
1035	const struct msi_parent_ops *pops;
1036	struct fwnode_handle *fwnode;
1037
1038	if (!irq_domain_is_msi_parent(parent))
1039		return false;
1040
1041	if (domid >= MSI_MAX_DEVICE_IRQDOMAINS)
1042		return false;
1043
1044	struct msi_domain_template *bundle __free(kfree) =
1045		kmemdup(template, sizeof(*bundle), GFP_KERNEL);
1046	if (!bundle)
1047		return false;
1048
1049	bundle->info.hwsize = hwsize;
1050	bundle->info.chip = &bundle->chip;
1051	bundle->info.ops = &bundle->ops;
1052	bundle->info.data = domain_data;
1053	bundle->info.chip_data = chip_data;
1054	bundle->info.alloc_data = &bundle->alloc_info;
1055	bundle->info.dev = dev;
1056
1057	pops = parent->msi_parent_ops;
1058	snprintf(bundle->name, sizeof(bundle->name), "%s%s-%s",
1059		 pops->prefix ? : "", bundle->chip.name, dev_name(dev));
1060	bundle->chip.name = bundle->name;
1061
1062	/*
1063	 * Using the device firmware node is required for wire to MSI
1064	 * device domains so that the existing firmware results in a domain
1065	 * match.
1066	 * All other device domains like PCI/MSI use the named firmware
1067	 * node as they are not guaranteed to have a fwnode. They are never
1068	 * looked up and always handled in the context of the device.
1069	 */
1070	struct fwnode_handle *fwnode_alloced __free(irq_domain_free_fwnode) = NULL;
1071
1072	if (!(bundle->info.flags & MSI_FLAG_USE_DEV_FWNODE))
1073		fwnode = fwnode_alloced = irq_domain_alloc_named_fwnode(bundle->name);
1074	else
1075		fwnode = dev->fwnode;
1076
1077	if (!fwnode)
1078		return false;
1079
1080	if (msi_setup_device_data(dev))
1081		return false;
1082
1083	guard(msi_descs_lock)(dev);
1084	if (WARN_ON_ONCE(msi_get_device_domain(dev, domid)))
1085		return false;
1086
1087	if (!pops->init_dev_msi_info(dev, parent, parent, &bundle->info))
1088		return false;
1089
1090	domain = __msi_create_irq_domain(fwnode, &bundle->info, IRQ_DOMAIN_FLAG_MSI_DEVICE, parent);
1091	if (!domain)
1092		return false;
1093
1094	dev->msi.data->__domains[domid].domain = domain;
1095
1096	if (msi_domain_prepare_irqs(domain, dev, hwsize, &bundle->alloc_info)) {
1097		dev->msi.data->__domains[domid].domain = NULL;
1098		irq_domain_remove(domain);
1099		return false;
1100	}
1101
1102	/* @bundle and @fwnode_alloced are now in use. Prevent cleanup */
1103	retain_and_null_ptr(bundle);
1104	retain_and_null_ptr(fwnode_alloced);
1105	return true;
1106}
1107
1108/**
1109 * msi_remove_device_irq_domain - Free a device MSI interrupt domain
1110 * @dev:	Pointer to the device
1111 * @domid:	Domain id
1112 */
1113void msi_remove_device_irq_domain(struct device *dev, unsigned int domid)
1114{
1115	struct fwnode_handle *fwnode = NULL;
1116	struct msi_domain_info *info;
1117	struct irq_domain *domain;
1118
1119	guard(msi_descs_lock)(dev);
1120	domain = msi_get_device_domain(dev, domid);
1121	if (!domain || !irq_domain_is_msi_device(domain))
1122		return;
1123
1124	dev->msi.data->__domains[domid].domain = NULL;
1125	info = domain->host_data;
1126
1127	info->ops->msi_teardown(domain, info->alloc_data);
1128
1129	if (irq_domain_is_msi_device(domain))
1130		fwnode = domain->fwnode;
1131	irq_domain_remove(domain);
1132	irq_domain_free_fwnode(fwnode);
1133	kfree(container_of(info, struct msi_domain_template, info));
1134}
1135
1136/**
1137 * msi_match_device_irq_domain - Match a device irq domain against a bus token
1138 * @dev:	Pointer to the device
1139 * @domid:	Domain id
1140 * @bus_token:	Bus token to match against the domain bus token
1141 *
1142 * Return: True if device domain exists and bus tokens match.
1143 */
1144bool msi_match_device_irq_domain(struct device *dev, unsigned int domid,
1145				 enum irq_domain_bus_token bus_token)
1146{
1147	struct msi_domain_info *info;
1148	struct irq_domain *domain;
1149
1150	guard(msi_descs_lock)(dev);
1151	domain = msi_get_device_domain(dev, domid);
1152	if (domain && irq_domain_is_msi_device(domain)) {
1153		info = domain->host_data;
1154		return info->bus_token == bus_token;
1155	}
1156	return false;
1157}
1158
1159static int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
1160				   int nvec, msi_alloc_info_t *arg)
1161{
1162	struct msi_domain_info *info = domain->host_data;
1163	struct msi_domain_ops *ops = info->ops;
1164
1165	return ops->msi_prepare(domain, dev, nvec, arg);
1166}
1167
1168/*
1169 * Carefully check whether the device can use reservation mode. If
1170 * reservation mode is enabled then the early activation will assign a
1171 * dummy vector to the device. If the PCI/MSI device does not support
1172 * masking of the entry then this can result in spurious interrupts when
1173 * the device driver is not absolutely careful. But even then a malfunction
1174 * of the hardware could result in a spurious interrupt on the dummy vector
1175 * and render the device unusable. If the entry can be masked then the core
1176 * logic will prevent the spurious interrupt and reservation mode can be
1177 * used. For now reservation mode is restricted to PCI/MSI.
1178 */
1179static bool msi_check_reservation_mode(struct irq_domain *domain,
1180				       struct msi_domain_info *info,
1181				       struct device *dev)
1182{
1183	struct msi_desc *desc;
1184
1185	switch(domain->bus_token) {
1186	case DOMAIN_BUS_PCI_MSI:
1187	case DOMAIN_BUS_PCI_DEVICE_MSI:
1188	case DOMAIN_BUS_PCI_DEVICE_MSIX:
1189	case DOMAIN_BUS_VMD_MSI:
1190		break;
1191	default:
1192		return false;
1193	}
1194
1195	if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
1196		return false;
1197
1198	if (info->flags & MSI_FLAG_NO_MASK)
1199		return false;
1200
1201	/*
1202	 * Checking the first MSI descriptor is sufficient. MSIX supports
1203	 * masking and MSI does so when the can_mask attribute is set.
1204	 */
1205	desc = msi_first_desc(dev, MSI_DESC_ALL);
1206	return desc->pci.msi_attrib.is_msix || desc->pci.msi_attrib.can_mask;
1207}
1208
1209static int msi_handle_pci_fail(struct irq_domain *domain, struct msi_desc *desc,
1210			       int allocated)
1211{
1212	switch(domain->bus_token) {
1213	case DOMAIN_BUS_PCI_MSI:
1214	case DOMAIN_BUS_PCI_DEVICE_MSI:
1215	case DOMAIN_BUS_PCI_DEVICE_MSIX:
1216	case DOMAIN_BUS_VMD_MSI:
1217		if (IS_ENABLED(CONFIG_PCI_MSI))
1218			break;
1219		fallthrough;
1220	default:
1221		return -ENOSPC;
1222	}
1223
1224	/* Let a failed PCI multi MSI allocation retry */
1225	if (desc->nvec_used > 1)
1226		return 1;
1227
1228	/* If there was a successful allocation let the caller know */
1229	return allocated ? allocated : -ENOSPC;
1230}
1231
1232#define VIRQ_CAN_RESERVE	0x01
1233#define VIRQ_ACTIVATE		0x02
1234
1235static int msi_init_virq(struct irq_domain *domain, int virq, unsigned int vflags)
1236{
1237	struct irq_data *irqd = irq_domain_get_irq_data(domain, virq);
1238	int ret;
1239
1240	if (!(vflags & VIRQ_CAN_RESERVE)) {
1241		irqd_clr_can_reserve(irqd);
1242
1243		/*
1244		 * If the interrupt is managed but no CPU is available to
1245		 * service it, shut it down until better times. Note that
1246		 * we only do this on the !RESERVE path as x86 (the only
1247		 * architecture using this flag) deals with this in a
1248		 * different way by using a catch-all vector.
1249		 */
1250		if ((vflags & VIRQ_ACTIVATE) &&
1251		    irqd_affinity_is_managed(irqd) &&
1252		    !cpumask_intersects(irq_data_get_affinity_mask(irqd),
1253					cpu_online_mask)) {
1254			    irqd_set_managed_shutdown(irqd);
1255			    return 0;
1256		    }
1257	}
1258
1259	if (!(vflags & VIRQ_ACTIVATE))
1260		return 0;
1261
1262	ret = irq_domain_activate_irq(irqd, vflags & VIRQ_CAN_RESERVE);
1263	if (ret)
1264		return ret;
1265	/*
1266	 * If the interrupt uses reservation mode, clear the activated bit
1267	 * so request_irq() will assign the final vector.
1268	 */
1269	if (vflags & VIRQ_CAN_RESERVE)
1270		irqd_clr_activated(irqd);
1271	return 0;
1272}
1273
1274static int populate_alloc_info(struct irq_domain *domain, struct device *dev,
1275			       unsigned int nirqs, msi_alloc_info_t *arg)
1276{
1277	struct msi_domain_info *info = domain->host_data;
1278
1279	/*
1280	 * If the caller has provided a template alloc info, use that. Once
1281	 * all users of msi_create_irq_domain() have been eliminated, this
1282	 * should be the only source of allocation information, and the
1283	 * prepare call below should be finally removed.
1284	 */
1285	if (!info->alloc_data)
1286		return msi_domain_prepare_irqs(domain, dev, nirqs, arg);
1287
1288	*arg = *info->alloc_data;
1289	return 0;
1290}
1291
1292static int __msi_domain_alloc_irqs(struct device *dev, struct irq_domain *domain,
1293				   struct msi_ctrl *ctrl)
1294{
1295	struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1296	struct msi_domain_info *info = domain->host_data;
1297	struct msi_domain_ops *ops = info->ops;
1298	unsigned int vflags = 0, allocated = 0;
1299	msi_alloc_info_t arg = { };
1300	struct msi_desc *desc;
1301	unsigned long idx;
1302	int i, ret, virq;
1303
1304	ret = populate_alloc_info(domain, dev, ctrl->nirqs, &arg);
1305	if (ret)
1306		return ret;
1307
1308	/*
1309	 * This flag is set by the PCI layer as we need to activate
1310	 * the MSI entries before the PCI layer enables MSI in the
1311	 * card. Otherwise the card latches a random msi message.
1312	 */
1313	if (info->flags & MSI_FLAG_ACTIVATE_EARLY)
1314		vflags |= VIRQ_ACTIVATE;
1315
1316	/*
1317	 * Interrupt can use a reserved vector and will not occupy
1318	 * a real device vector until the interrupt is requested.
1319	 */
1320	if (msi_check_reservation_mode(domain, info, dev))
1321		vflags |= VIRQ_CAN_RESERVE;
1322
1323	xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1324		if (!msi_desc_match(desc, MSI_DESC_NOTASSOCIATED))
1325			continue;
1326
1327		/* This should return -ECONFUSED... */
1328		if (WARN_ON_ONCE(allocated >= ctrl->nirqs))
1329			return -EINVAL;
1330
1331		if (ops->prepare_desc)
1332			ops->prepare_desc(domain, &arg, desc);
1333
1334		ops->set_desc(&arg, desc);
1335
1336		virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
1337					       dev_to_node(dev), &arg, false,
1338					       desc->affinity);
1339		if (virq < 0)
1340			return msi_handle_pci_fail(domain, desc, allocated);
1341
1342		for (i = 0; i < desc->nvec_used; i++) {
1343			irq_set_msi_desc_off(virq, i, desc);
1344			irq_debugfs_copy_devname(virq + i, dev);
1345			ret = msi_init_virq(domain, virq + i, vflags);
1346			if (ret)
1347				return ret;
1348		}
1349		if (info->flags & MSI_FLAG_DEV_SYSFS) {
1350			ret = msi_sysfs_populate_desc(dev, desc);
1351			if (ret)
1352				return ret;
1353		}
1354		allocated++;
1355	}
1356	return 0;
1357}
1358
1359static int msi_domain_alloc_simple_msi_descs(struct device *dev,
1360					     struct msi_domain_info *info,
1361					     struct msi_ctrl *ctrl)
1362{
1363	if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
1364		return 0;
1365
1366	return msi_domain_add_simple_msi_descs(dev, ctrl);
1367}
1368
1369static int __msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1370{
1371	struct msi_domain_info *info;
1372	struct msi_domain_ops *ops;
1373	struct irq_domain *domain;
1374	int ret;
1375
1376	if (!msi_ctrl_valid(dev, ctrl))
1377		return -EINVAL;
1378
1379	domain = msi_get_device_domain(dev, ctrl->domid);
1380	if (!domain)
1381		return -ENODEV;
1382
1383	info = domain->host_data;
1384
1385	ret = msi_domain_alloc_simple_msi_descs(dev, info, ctrl);
1386	if (ret)
1387		return ret;
1388
1389	ops = info->ops;
1390	if (ops->domain_alloc_irqs)
1391		return ops->domain_alloc_irqs(domain, dev, ctrl->nirqs);
1392
1393	return __msi_domain_alloc_irqs(dev, domain, ctrl);
1394}
1395
1396static int msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1397{
1398	int ret = __msi_domain_alloc_locked(dev, ctrl);
1399
1400	if (ret)
1401		msi_domain_free_locked(dev, ctrl);
1402	return ret;
1403}
1404
1405/**
1406 * msi_domain_alloc_irqs_range_locked - Allocate interrupts from a MSI interrupt domain
1407 * @dev:	Pointer to device struct of the device for which the interrupts
1408 *		are allocated
1409 * @domid:	Id of the interrupt domain to operate on
1410 * @first:	First index to allocate (inclusive)
1411 * @last:	Last index to allocate (inclusive)
1412 *
1413 * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1414 * pair. Use this for MSI irqdomains which implement their own descriptor
1415 * allocation/free.
1416 *
1417 * Return: %0 on success or an error code.
1418 */
1419int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
1420				       unsigned int first, unsigned int last)
1421{
1422	struct msi_ctrl ctrl = {
1423		.domid	= domid,
1424		.first	= first,
1425		.last	= last,
1426		.nirqs	= last + 1 - first,
1427	};
1428
1429	return msi_domain_alloc_locked(dev, &ctrl);
1430}
1431
1432/**
1433 * msi_domain_alloc_irqs_range - Allocate interrupts from a MSI interrupt domain
1434 * @dev:	Pointer to device struct of the device for which the interrupts
1435 *		are allocated
1436 * @domid:	Id of the interrupt domain to operate on
1437 * @first:	First index to allocate (inclusive)
1438 * @last:	Last index to allocate (inclusive)
1439 *
1440 * Return: %0 on success or an error code.
1441 */
1442int msi_domain_alloc_irqs_range(struct device *dev, unsigned int domid,
1443				unsigned int first, unsigned int last)
1444{
1445
1446	guard(msi_descs_lock)(dev);
1447	return msi_domain_alloc_irqs_range_locked(dev, domid, first, last);
1448}
1449EXPORT_SYMBOL_GPL(msi_domain_alloc_irqs_range);
1450
1451/**
1452 * msi_domain_alloc_irqs_all_locked - Allocate all interrupts from a MSI interrupt domain
1453 *
1454 * @dev:	Pointer to device struct of the device for which the interrupts
1455 *		are allocated
1456 * @domid:	Id of the interrupt domain to operate on
1457 * @nirqs:	The number of interrupts to allocate
1458 *
1459 * This function scans all MSI descriptors of the MSI domain and allocates interrupts
1460 * for all unassigned ones. That function is to be used for MSI domain usage where
1461 * the descriptor allocation is handled at the call site, e.g. PCI/MSI[X].
1462 *
1463 * Return: %0 on success or an error code.
1464 */
1465int msi_domain_alloc_irqs_all_locked(struct device *dev, unsigned int domid, int nirqs)
1466{
1467	struct msi_ctrl ctrl = {
1468		.domid	= domid,
1469		.first	= 0,
1470		.last	= msi_domain_get_hwsize(dev, domid) - 1,
1471		.nirqs	= nirqs,
1472	};
1473
1474	return msi_domain_alloc_locked(dev, &ctrl);
1475}
1476
1477static struct msi_map __msi_domain_alloc_irq_at(struct device *dev, unsigned int domid,
1478						unsigned int index,
1479						const struct irq_affinity_desc *affdesc,
1480						union msi_instance_cookie *icookie)
1481{
1482	struct msi_ctrl ctrl = { .domid	= domid, .nirqs = 1, };
1483	struct irq_domain *domain;
1484	struct msi_map map = { };
1485	struct msi_desc *desc;
1486	int ret;
1487
1488	domain = msi_get_device_domain(dev, domid);
1489	if (!domain) {
1490		map.index = -ENODEV;
1491		return map;
1492	}
1493
1494	desc = msi_alloc_desc(dev, 1, affdesc);
1495	if (!desc) {
1496		map.index = -ENOMEM;
1497		return map;
1498	}
1499
1500	if (icookie)
1501		desc->data.icookie = *icookie;
1502
1503	ret = msi_insert_desc(dev, desc, domid, index);
1504	if (ret) {
1505		map.index = ret;
1506		return map;
1507	}
1508
1509	ctrl.first = ctrl.last = desc->msi_index;
1510
1511	ret = __msi_domain_alloc_irqs(dev, domain, &ctrl);
1512	if (ret) {
1513		map.index = ret;
1514		msi_domain_free_locked(dev, &ctrl);
1515	} else {
1516		map.index = desc->msi_index;
1517		map.virq = desc->irq;
1518	}
1519	return map;
1520}
1521
1522/**
1523 * msi_domain_alloc_irq_at - Allocate an interrupt from a MSI interrupt domain at
1524 *			     a given index - or at the next free index
1525 *
1526 * @dev:	Pointer to device struct of the device for which the interrupts
1527 *		are allocated
1528 * @domid:	Id of the interrupt domain to operate on
1529 * @index:	Index for allocation. If @index == %MSI_ANY_INDEX the allocation
1530 *		uses the next free index.
1531 * @affdesc:	Optional pointer to an interrupt affinity descriptor structure
1532 * @icookie:	Optional pointer to a domain specific per instance cookie. If
1533 *		non-NULL the content of the cookie is stored in msi_desc::data.
1534 *		Must be NULL for MSI-X allocations
1535 *
1536 * This requires a MSI interrupt domain which lets the core code manage the
1537 * MSI descriptors.
1538 *
1539 * Return: struct msi_map
1540 *
1541 *	On success msi_map::index contains the allocated index number and
1542 *	msi_map::virq the corresponding Linux interrupt number
1543 *
1544 *	On failure msi_map::index contains the error code and msi_map::virq
1545 *	is %0.
1546 */
1547struct msi_map msi_domain_alloc_irq_at(struct device *dev, unsigned int domid, unsigned int index,
1548				       const struct irq_affinity_desc *affdesc,
1549				       union msi_instance_cookie *icookie)
1550{
1551	guard(msi_descs_lock)(dev);
1552	return __msi_domain_alloc_irq_at(dev, domid, index, affdesc, icookie);
1553}
1554
1555/**
1556 * msi_device_domain_alloc_wired - Allocate a "wired" interrupt on @domain
1557 * @domain:	The domain to allocate on
1558 * @hwirq:	The hardware interrupt number to allocate for
1559 * @type:	The interrupt type
1560 *
1561 * This weirdness supports wire to MSI controllers like MBIGEN.
1562 *
1563 * @hwirq is the hardware interrupt number which is handed in from
1564 * irq_create_fwspec_mapping(). As the wire to MSI domain is sparse, but
1565 * sized in firmware, the hardware interrupt number cannot be used as MSI
1566 * index. For the underlying irq chip the MSI index is irrelevant and
1567 * all it needs is the hardware interrupt number.
1568 *
1569 * To handle this the MSI index is allocated with MSI_ANY_INDEX and the
1570 * hardware interrupt number is stored along with the type information in
1571 * msi_desc::cookie so the underlying interrupt chip and domain code can
1572 * retrieve it.
1573 *
1574 * Return: The Linux interrupt number (> 0) or an error code
1575 */
1576int msi_device_domain_alloc_wired(struct irq_domain *domain, unsigned int hwirq,
1577				  unsigned int type)
1578{
1579	unsigned int domid = MSI_DEFAULT_DOMAIN;
1580	union msi_instance_cookie icookie = { };
1581	struct device *dev = domain->dev;
1582	struct msi_map map = { };
1583
1584	if (WARN_ON_ONCE(!dev || domain->bus_token != DOMAIN_BUS_WIRED_TO_MSI))
1585		return -EINVAL;
1586
1587	icookie.value = ((u64)type << 32) | hwirq;
1588
1589	guard(msi_descs_lock)(dev);
1590	if (WARN_ON_ONCE(msi_get_device_domain(dev, domid) != domain))
1591		map.index = -EINVAL;
1592	else
1593		map = __msi_domain_alloc_irq_at(dev, domid, MSI_ANY_INDEX, NULL, &icookie);
1594	return map.index >= 0 ? map.virq : map.index;
1595}
1596
1597static void __msi_domain_free_irqs(struct device *dev, struct irq_domain *domain,
1598				   struct msi_ctrl *ctrl)
1599{
1600	struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1601	struct msi_domain_info *info = domain->host_data;
1602	struct irq_data *irqd;
1603	struct msi_desc *desc;
1604	unsigned long idx;
1605	int i;
1606
1607	xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1608		/* Only handle MSI entries which have an interrupt associated */
1609		if (!msi_desc_match(desc, MSI_DESC_ASSOCIATED))
1610			continue;
1611
1612		/* Make sure all interrupts are deactivated */
1613		for (i = 0; i < desc->nvec_used; i++) {
1614			irqd = irq_domain_get_irq_data(domain, desc->irq + i);
1615			if (irqd && irqd_is_activated(irqd))
1616				irq_domain_deactivate_irq(irqd);
1617		}
1618
1619		irq_domain_free_irqs(desc->irq, desc->nvec_used);
1620		if (info->flags & MSI_FLAG_DEV_SYSFS)
1621			msi_sysfs_remove_desc(dev, desc);
1622		desc->irq = 0;
1623	}
1624}
1625
1626static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl)
1627{
1628	struct msi_domain_info *info;
1629	struct msi_domain_ops *ops;
1630	struct irq_domain *domain;
1631
1632	if (!msi_ctrl_valid(dev, ctrl))
1633		return;
1634
1635	domain = msi_get_device_domain(dev, ctrl->domid);
1636	if (!domain)
1637		return;
1638
1639	info = domain->host_data;
1640	ops = info->ops;
1641
1642	if (ops->domain_free_irqs)
1643		ops->domain_free_irqs(domain, dev);
1644	else
1645		__msi_domain_free_irqs(dev, domain, ctrl);
1646
1647	if (info->flags & MSI_FLAG_FREE_MSI_DESCS)
1648		msi_domain_free_descs(dev, ctrl);
1649}
1650
1651/**
1652 * msi_domain_free_irqs_range_locked - Free a range of interrupts from a MSI interrupt domain
1653 *				       associated to @dev with msi_lock held
1654 * @dev:	Pointer to device struct of the device for which the interrupts
1655 *		are freed
1656 * @domid:	Id of the interrupt domain to operate on
1657 * @first:	First index to free (inclusive)
1658 * @last:	Last index to free (inclusive)
1659 */
1660void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
1661				       unsigned int first, unsigned int last)
1662{
1663	struct msi_ctrl ctrl = {
1664		.domid	= domid,
1665		.first	= first,
1666		.last	= last,
1667	};
1668	msi_domain_free_locked(dev, &ctrl);
1669}
1670
1671/**
1672 * msi_domain_free_irqs_range - Free a range of interrupts from a MSI interrupt domain
1673 *				associated to @dev
1674 * @dev:	Pointer to device struct of the device for which the interrupts
1675 *		are freed
1676 * @domid:	Id of the interrupt domain to operate on
1677 * @first:	First index to free (inclusive)
1678 * @last:	Last index to free (inclusive)
1679 */
1680void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
1681				unsigned int first, unsigned int last)
1682{
1683	guard(msi_descs_lock)(dev);
1684	msi_domain_free_irqs_range_locked(dev, domid, first, last);
1685}
1686EXPORT_SYMBOL_GPL(msi_domain_free_irqs_all);
1687
1688/**
1689 * msi_domain_free_irqs_all_locked - Free all interrupts from a MSI interrupt domain
1690 *				     associated to a device
1691 * @dev:	Pointer to device struct of the device for which the interrupts
1692 *		are freed
1693 * @domid:	The id of the domain to operate on
1694 *
1695 * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1696 * pair. Use this for MSI irqdomains which implement their own vector
1697 * allocation.
1698 */
1699void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid)
1700{
1701	msi_domain_free_irqs_range_locked(dev, domid, 0,
1702					  msi_domain_get_hwsize(dev, domid) - 1);
1703}
1704
1705/**
1706 * msi_domain_free_irqs_all - Free all interrupts from a MSI interrupt domain
1707 *			      associated to a device
1708 * @dev:	Pointer to device struct of the device for which the interrupts
1709 *		are freed
1710 * @domid:	The id of the domain to operate on
1711 */
1712void msi_domain_free_irqs_all(struct device *dev, unsigned int domid)
1713{
1714	guard(msi_descs_lock)(dev);
1715	msi_domain_free_irqs_all_locked(dev, domid);
1716}
1717
1718/**
1719 * msi_device_domain_free_wired - Free a wired interrupt in @domain
1720 * @domain:	The domain to free the interrupt on
1721 * @virq:	The Linux interrupt number to free
1722 *
1723 * This is the counterpart of msi_device_domain_alloc_wired() for the
1724 * weird wired to MSI converting domains.
1725 */
1726void msi_device_domain_free_wired(struct irq_domain *domain, unsigned int virq)
1727{
1728	struct msi_desc *desc = irq_get_msi_desc(virq);
1729	struct device *dev = domain->dev;
1730
1731	if (WARN_ON_ONCE(!dev || !desc || domain->bus_token != DOMAIN_BUS_WIRED_TO_MSI))
1732		return;
1733
1734	guard(msi_descs_lock)(dev);
1735	if (WARN_ON_ONCE(msi_get_device_domain(dev, MSI_DEFAULT_DOMAIN) != domain))
1736		return;
1737	msi_domain_free_irqs_range_locked(dev, MSI_DEFAULT_DOMAIN, desc->msi_index,
1738					  desc->msi_index);
1739}
1740
1741/**
1742 * msi_get_domain_info - Get the MSI interrupt domain info for @domain
1743 * @domain:	The interrupt domain to retrieve data from
1744 *
1745 * Return: the pointer to the msi_domain_info stored in @domain->host_data.
1746 */
1747struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
1748{
1749	return (struct msi_domain_info *)domain->host_data;
1750}
1751
1752/**
1753 * msi_device_has_isolated_msi - True if the device has isolated MSI
1754 * @dev: The device to check
1755 *
1756 * Isolated MSI means that HW modeled by an irq_domain on the path from the
1757 * initiating device to the CPU will validate that the MSI message specifies an
1758 * interrupt number that the device is authorized to trigger. This must block
1759 * devices from triggering interrupts they are not authorized to trigger.
1760 * Currently authorization means the MSI vector is one assigned to the device.
1761 *
1762 * This is interesting for securing VFIO use cases where a rouge MSI (eg created
1763 * by abusing a normal PCI MemWr DMA) must not allow the VFIO userspace to
1764 * impact outside its security domain, eg userspace triggering interrupts on
1765 * kernel drivers, a VM triggering interrupts on the hypervisor, or a VM
1766 * triggering interrupts on another VM.
1767 */
1768bool msi_device_has_isolated_msi(struct device *dev)
1769{
1770	struct irq_domain *domain = dev_get_msi_domain(dev);
1771
1772	for (; domain; domain = domain->parent)
1773		if (domain->flags & IRQ_DOMAIN_FLAG_ISOLATED_MSI)
1774			return true;
1775	return arch_is_isolated_msi();
1776}
1777EXPORT_SYMBOL_GPL(msi_device_has_isolated_msi);