pci-driver.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * (C) Copyright 2002-2004, 2007 Greg Kroah-Hartman <greg@kroah.com>
   4 * (C) Copyright 2007 Novell Inc.
   5 */
   6
   7#include <linux/pci.h>
   8#include <linux/module.h>
   9#include <linux/init.h>
  10#include <linux/device.h>
  11#include <linux/mempolicy.h>
  12#include <linux/string.h>
  13#include <linux/slab.h>
  14#include <linux/sched.h>
  15#include <linux/sched/isolation.h>
  16#include <linux/cpu.h>
  17#include <linux/pm_runtime.h>
  18#include <linux/suspend.h>
  19#include <linux/kexec.h>
  20#include <linux/of_device.h>
  21#include <linux/acpi.h>
  22#include <linux/dma-map-ops.h>
  23#include <linux/iommu.h>
  24#include "pci.h"
  25#include "pcie/portdrv.h"
  26
  27struct pci_dynid {
  28	struct list_head node;
  29	struct pci_device_id id;
  30};
  31
  32/**
  33 * pci_add_dynid - add a new PCI device ID to this driver and re-probe devices
  34 * @drv: target pci driver
  35 * @vendor: PCI vendor ID
  36 * @device: PCI device ID
  37 * @subvendor: PCI subvendor ID
  38 * @subdevice: PCI subdevice ID
  39 * @class: PCI class
  40 * @class_mask: PCI class mask
  41 * @driver_data: private driver data
  42 *
  43 * Adds a new dynamic pci device ID to this driver and causes the
  44 * driver to probe for all devices again.  @drv must have been
  45 * registered prior to calling this function.
  46 *
  47 * CONTEXT:
  48 * Does GFP_KERNEL allocation.
  49 *
  50 * RETURNS:
  51 * 0 on success, -errno on failure.
  52 */
  53int pci_add_dynid(struct pci_driver *drv,
  54		  unsigned int vendor, unsigned int device,
  55		  unsigned int subvendor, unsigned int subdevice,
  56		  unsigned int class, unsigned int class_mask,
  57		  unsigned long driver_data)
  58{
  59	struct pci_dynid *dynid;
  60
  61	dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
  62	if (!dynid)
  63		return -ENOMEM;
  64
  65	dynid->id.vendor = vendor;
  66	dynid->id.device = device;
  67	dynid->id.subvendor = subvendor;
  68	dynid->id.subdevice = subdevice;
  69	dynid->id.class = class;
  70	dynid->id.class_mask = class_mask;
  71	dynid->id.driver_data = driver_data;
  72
  73	spin_lock(&drv->dynids.lock);
  74	list_add_tail(&dynid->node, &drv->dynids.list);
  75	spin_unlock(&drv->dynids.lock);
  76
  77	return driver_attach(&drv->driver);
  78}
  79EXPORT_SYMBOL_GPL(pci_add_dynid);
  80
  81static void pci_free_dynids(struct pci_driver *drv)
  82{
  83	struct pci_dynid *dynid, *n;
  84
  85	spin_lock(&drv->dynids.lock);
  86	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
  87		list_del(&dynid->node);
  88		kfree(dynid);
  89	}
  90	spin_unlock(&drv->dynids.lock);
  91}
  92
  93/**
  94 * pci_match_id - See if a PCI device matches a given pci_id table
  95 * @ids: array of PCI device ID structures to search in
  96 * @dev: the PCI device structure to match against.
  97 *
  98 * Used by a driver to check whether a PCI device is in its list of
  99 * supported devices.  Returns the matching pci_device_id structure or
 100 * %NULL if there is no match.
 101 *
 102 * Deprecated; don't use this as it will not catch any dynamic IDs
 103 * that a driver might want to check for.
 104 */
 105const struct pci_device_id *pci_match_id(const struct pci_device_id *ids,
 106					 struct pci_dev *dev)
 107{
 108	if (ids) {
 109		while (ids->vendor || ids->subvendor || ids->class_mask) {
 110			if (pci_match_one_device(ids, dev))
 111				return ids;
 112			ids++;
 113		}
 114	}
 115	return NULL;
 116}
 117EXPORT_SYMBOL(pci_match_id);
 118
 119static const struct pci_device_id pci_device_id_any = {
 120	.vendor = PCI_ANY_ID,
 121	.device = PCI_ANY_ID,
 122	.subvendor = PCI_ANY_ID,
 123	.subdevice = PCI_ANY_ID,
 124};
 125
 126/**
 127 * pci_match_device - See if a device matches a driver's list of IDs
 128 * @drv: the PCI driver to match against
 129 * @dev: the PCI device structure to match against
 130 *
 131 * Used by a driver to check whether a PCI device is in its list of
 132 * supported devices or in the dynids list, which may have been augmented
 133 * via the sysfs "new_id" file.  Returns the matching pci_device_id
 134 * structure or %NULL if there is no match.
 135 */
 136static const struct pci_device_id *pci_match_device(struct pci_driver *drv,
 137						    struct pci_dev *dev)
 138{
 139	struct pci_dynid *dynid;
 140	const struct pci_device_id *found_id = NULL, *ids;
 141
 142	/* When driver_override is set, only bind to the matching driver */
 143	if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 144		return NULL;
 145
 146	/* Look at the dynamic ids first, before the static ones */
 147	spin_lock(&drv->dynids.lock);
 148	list_for_each_entry(dynid, &drv->dynids.list, node) {
 149		if (pci_match_one_device(&dynid->id, dev)) {
 150			found_id = &dynid->id;
 151			break;
 152		}
 153	}
 154	spin_unlock(&drv->dynids.lock);
 155
 156	if (found_id)
 157		return found_id;
 158
 159	for (ids = drv->id_table; (found_id = pci_match_id(ids, dev));
 160	     ids = found_id + 1) {
 161		/*
 162		 * The match table is split based on driver_override.
 163		 * In case override_only was set, enforce driver_override
 164		 * matching.
 165		 */
 166		if (found_id->override_only) {
 167			if (dev->driver_override)
 168				return found_id;
 169		} else {
 170			return found_id;
 171		}
 172	}
 173
 174	/* driver_override will always match, send a dummy id */
 175	if (dev->driver_override)
 176		return &pci_device_id_any;
 177	return NULL;
 178}
 179
 180/**
 181 * new_id_store - sysfs frontend to pci_add_dynid()
 182 * @driver: target device driver
 183 * @buf: buffer for scanning device ID data
 184 * @count: input size
 185 *
 186 * Allow PCI IDs to be added to an existing driver via sysfs.
 187 */
 188static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 189			    size_t count)
 190{
 191	struct pci_driver *pdrv = to_pci_driver(driver);
 192	const struct pci_device_id *ids = pdrv->id_table;
 193	u32 vendor, device, subvendor = PCI_ANY_ID,
 194		subdevice = PCI_ANY_ID, class = 0, class_mask = 0;
 195	unsigned long driver_data = 0;
 196	int fields;
 197	int retval = 0;
 198
 199	fields = sscanf(buf, "%x %x %x %x %x %x %lx",
 200			&vendor, &device, &subvendor, &subdevice,
 201			&class, &class_mask, &driver_data);
 202	if (fields < 2)
 203		return -EINVAL;
 204
 205	if (fields != 7) {
 206		struct pci_dev *pdev = kzalloc(sizeof(*pdev), GFP_KERNEL);
 207		if (!pdev)
 208			return -ENOMEM;
 209
 210		pdev->vendor = vendor;
 211		pdev->device = device;
 212		pdev->subsystem_vendor = subvendor;
 213		pdev->subsystem_device = subdevice;
 214		pdev->class = class;
 215
 216		if (pci_match_device(pdrv, pdev))
 217			retval = -EEXIST;
 218
 219		kfree(pdev);
 220
 221		if (retval)
 222			return retval;
 223	}
 224
 225	/* Only accept driver_data values that match an existing id_table
 226	   entry */
 227	if (ids) {
 228		retval = -EINVAL;
 229		while (ids->vendor || ids->subvendor || ids->class_mask) {
 230			if (driver_data == ids->driver_data) {
 231				retval = 0;
 232				break;
 233			}
 234			ids++;
 235		}
 236		if (retval)	/* No match */
 237			return retval;
 238	}
 239
 240	retval = pci_add_dynid(pdrv, vendor, device, subvendor, subdevice,
 241			       class, class_mask, driver_data);
 242	if (retval)
 243		return retval;
 244	return count;
 245}
 246static DRIVER_ATTR_WO(new_id);
 247
 248/**
 249 * remove_id_store - remove a PCI device ID from this driver
 250 * @driver: target device driver
 251 * @buf: buffer for scanning device ID data
 252 * @count: input size
 253 *
 254 * Removes a dynamic pci device ID to this driver.
 255 */
 256static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 257			       size_t count)
 258{
 259	struct pci_dynid *dynid, *n;
 260	struct pci_driver *pdrv = to_pci_driver(driver);
 261	u32 vendor, device, subvendor = PCI_ANY_ID,
 262		subdevice = PCI_ANY_ID, class = 0, class_mask = 0;
 263	int fields;
 264	size_t retval = -ENODEV;
 265
 266	fields = sscanf(buf, "%x %x %x %x %x %x",
 267			&vendor, &device, &subvendor, &subdevice,
 268			&class, &class_mask);
 269	if (fields < 2)
 270		return -EINVAL;
 271
 272	spin_lock(&pdrv->dynids.lock);
 273	list_for_each_entry_safe(dynid, n, &pdrv->dynids.list, node) {
 274		struct pci_device_id *id = &dynid->id;
 275		if ((id->vendor == vendor) &&
 276		    (id->device == device) &&
 277		    (subvendor == PCI_ANY_ID || id->subvendor == subvendor) &&
 278		    (subdevice == PCI_ANY_ID || id->subdevice == subdevice) &&
 279		    !((id->class ^ class) & class_mask)) {
 280			list_del(&dynid->node);
 281			kfree(dynid);
 282			retval = count;
 283			break;
 284		}
 285	}
 286	spin_unlock(&pdrv->dynids.lock);
 287
 288	return retval;
 289}
 290static DRIVER_ATTR_WO(remove_id);
 291
 292static struct attribute *pci_drv_attrs[] = {
 293	&driver_attr_new_id.attr,
 294	&driver_attr_remove_id.attr,
 295	NULL,
 296};
 297ATTRIBUTE_GROUPS(pci_drv);
 298
 299struct drv_dev_and_id {
 300	struct pci_driver *drv;
 301	struct pci_dev *dev;
 302	const struct pci_device_id *id;
 303};
 304
 305static long local_pci_probe(void *_ddi)
 306{
 307	struct drv_dev_and_id *ddi = _ddi;
 308	struct pci_dev *pci_dev = ddi->dev;
 309	struct pci_driver *pci_drv = ddi->drv;
 310	struct device *dev = &pci_dev->dev;
 311	int rc;
 312
 313	/*
 314	 * Unbound PCI devices are always put in D0, regardless of
 315	 * runtime PM status.  During probe, the device is set to
 316	 * active and the usage count is incremented.  If the driver
 317	 * supports runtime PM, it should call pm_runtime_put_noidle(),
 318	 * or any other runtime PM helper function decrementing the usage
 319	 * count, in its probe routine and pm_runtime_get_noresume() in
 320	 * its remove routine.
 321	 */
 322	pm_runtime_get_sync(dev);
 323	pci_dev->driver = pci_drv;
 324	rc = pci_drv->probe(pci_dev, ddi->id);
 325	if (!rc)
 326		return rc;
 327	if (rc < 0) {
 328		pci_dev->driver = NULL;
 329		pm_runtime_put_sync(dev);
 330		return rc;
 331	}
 332	/*
 333	 * Probe function should return < 0 for failure, 0 for success
 334	 * Treat values > 0 as success, but warn.
 335	 */
 336	pci_warn(pci_dev, "Driver probe function unexpectedly returned %d\n",
 337		 rc);
 338	return 0;
 339}
 340
 341static bool pci_physfn_is_probed(struct pci_dev *dev)
 342{
 343#ifdef CONFIG_PCI_IOV
 344	return dev->is_virtfn && dev->physfn->is_probed;
 345#else
 346	return false;
 347#endif
 348}
 349
 350static int pci_call_probe(struct pci_driver *drv, struct pci_dev *dev,
 351			  const struct pci_device_id *id)
 352{
 353	int error, node, cpu;
 354	struct drv_dev_and_id ddi = { drv, dev, id };
 355
 356	/*
 357	 * Execute driver initialization on node where the device is
 358	 * attached.  This way the driver likely allocates its local memory
 359	 * on the right node.
 360	 */
 361	node = dev_to_node(&dev->dev);
 362	dev->is_probed = 1;
 363
 364	cpu_hotplug_disable();
 365
 366	/*
 367	 * Prevent nesting work_on_cpu() for the case where a Virtual Function
 368	 * device is probed from work_on_cpu() of the Physical device.
 369	 */
 370	if (node < 0 || node >= MAX_NUMNODES || !node_online(node) ||
 371	    pci_physfn_is_probed(dev)) {
 372		cpu = nr_cpu_ids;
 373	} else {
 374		cpumask_var_t wq_domain_mask;
 375
 376		if (!zalloc_cpumask_var(&wq_domain_mask, GFP_KERNEL)) {
 377			error = -ENOMEM;
 378			goto out;
 379		}
 380		cpumask_and(wq_domain_mask,
 381			    housekeeping_cpumask(HK_TYPE_WQ),
 382			    housekeeping_cpumask(HK_TYPE_DOMAIN));
 383
 384		cpu = cpumask_any_and(cpumask_of_node(node),
 385				      wq_domain_mask);
 386		free_cpumask_var(wq_domain_mask);
 387	}
 388
 389	if (cpu < nr_cpu_ids)
 390		error = work_on_cpu(cpu, local_pci_probe, &ddi);
 391	else
 392		error = local_pci_probe(&ddi);
 393out:
 394	dev->is_probed = 0;
 395	cpu_hotplug_enable();
 396	return error;
 397}
 398
 399/**
 400 * __pci_device_probe - check if a driver wants to claim a specific PCI device
 401 * @drv: driver to call to check if it wants the PCI device
 402 * @pci_dev: PCI device being probed
 403 *
 404 * returns 0 on success, else error.
 405 * side-effect: pci_dev->driver is set to drv when drv claims pci_dev.
 406 */
 407static int __pci_device_probe(struct pci_driver *drv, struct pci_dev *pci_dev)
 408{
 409	const struct pci_device_id *id;
 410	int error = 0;
 411
 412	if (drv->probe) {
 413		error = -ENODEV;
 414
 415		id = pci_match_device(drv, pci_dev);
 416		if (id)
 417			error = pci_call_probe(drv, pci_dev, id);
 418	}
 419	return error;
 420}
 421
 422#ifdef CONFIG_PCI_IOV
 423static inline bool pci_device_can_probe(struct pci_dev *pdev)
 424{
 425	return (!pdev->is_virtfn || pdev->physfn->sriov->drivers_autoprobe ||
 426		pdev->driver_override);
 427}
 428#else
 429static inline bool pci_device_can_probe(struct pci_dev *pdev)
 430{
 431	return true;
 432}
 433#endif
 434
 435static int pci_device_probe(struct device *dev)
 436{
 437	int error;
 438	struct pci_dev *pci_dev = to_pci_dev(dev);
 439	struct pci_driver *drv = to_pci_driver(dev->driver);
 440
 441	if (!pci_device_can_probe(pci_dev))
 442		return -ENODEV;
 443
 444	pci_assign_irq(pci_dev);
 445
 446	error = pcibios_alloc_irq(pci_dev);
 447	if (error < 0)
 448		return error;
 449
 450	pci_dev_get(pci_dev);
 451	error = __pci_device_probe(drv, pci_dev);
 452	if (error) {
 453		pcibios_free_irq(pci_dev);
 454		pci_dev_put(pci_dev);
 455	}
 456
 457	return error;
 458}
 459
 460static void pci_device_remove(struct device *dev)
 461{
 462	struct pci_dev *pci_dev = to_pci_dev(dev);
 463	struct pci_driver *drv = pci_dev->driver;
 464
 465	if (drv->remove) {
 466		pm_runtime_get_sync(dev);
 467		/*
 468		 * If the driver provides a .runtime_idle() callback and it has
 469		 * started to run already, it may continue to run in parallel
 470		 * with the code below, so wait until all of the runtime PM
 471		 * activity has completed.
 472		 */
 473		pm_runtime_barrier(dev);
 474		drv->remove(pci_dev);
 475		pm_runtime_put_noidle(dev);
 476	}
 477	pcibios_free_irq(pci_dev);
 478	pci_dev->driver = NULL;
 479	pci_iov_remove(pci_dev);
 480
 481	/* Undo the runtime PM settings in local_pci_probe() */
 482	pm_runtime_put_sync(dev);
 483
 484	/*
 485	 * If the device is still on, set the power state as "unknown",
 486	 * since it might change by the next time we load the driver.
 487	 */
 488	if (pci_dev->current_state == PCI_D0)
 489		pci_dev->current_state = PCI_UNKNOWN;
 490
 491	/*
 492	 * We would love to complain here if pci_dev->is_enabled is set, that
 493	 * the driver should have called pci_disable_device(), but the
 494	 * unfortunate fact is there are too many odd BIOS and bridge setups
 495	 * that don't like drivers doing that all of the time.
 496	 * Oh well, we can dream of sane hardware when we sleep, no matter how
 497	 * horrible the crap we have to deal with is when we are awake...
 498	 */
 499
 500	pci_dev_put(pci_dev);
 501}
 502
 503static void pci_device_shutdown(struct device *dev)
 504{
 505	struct pci_dev *pci_dev = to_pci_dev(dev);
 506	struct pci_driver *drv = pci_dev->driver;
 507
 508	pm_runtime_resume(dev);
 509
 510	if (drv && drv->shutdown)
 511		drv->shutdown(pci_dev);
 512
 513	/*
 514	 * If this is a kexec reboot, turn off Bus Master bit on the
 515	 * device to tell it to not continue to do DMA. Don't touch
 516	 * devices in D3cold or unknown states.
 517	 * If it is not a kexec reboot, firmware will hit the PCI
 518	 * devices with big hammer and stop their DMA any way.
 519	 */
 520	if (kexec_in_progress && (pci_dev->current_state <= PCI_D3hot))
 521		pci_clear_master(pci_dev);
 522}
 523
 524#ifdef CONFIG_PM_SLEEP
 525
 526/* Auxiliary functions used for system resume */
 527
 528/**
 529 * pci_restore_standard_config - restore standard config registers of PCI device
 530 * @pci_dev: PCI device to handle
 531 */
 532static int pci_restore_standard_config(struct pci_dev *pci_dev)
 533{
 534	pci_update_current_state(pci_dev, PCI_UNKNOWN);
 535
 536	if (pci_dev->current_state != PCI_D0) {
 537		int error = pci_set_power_state(pci_dev, PCI_D0);
 538		if (error)
 539			return error;
 540	}
 541
 542	pci_restore_state(pci_dev);
 543	pci_pme_restore(pci_dev);
 544	return 0;
 545}
 546#endif /* CONFIG_PM_SLEEP */
 547
 548#ifdef CONFIG_PM
 549
 550/* Auxiliary functions used for system resume and run-time resume */
 551
 552static void pci_pm_default_resume(struct pci_dev *pci_dev)
 553{
 554	pci_fixup_device(pci_fixup_resume, pci_dev);
 555	pci_enable_wake(pci_dev, PCI_D0, false);
 556}
 557
 558static void pci_pm_default_resume_early(struct pci_dev *pci_dev)
 559{
 560	pci_pm_power_up_and_verify_state(pci_dev);
 561	pci_restore_state(pci_dev);
 562	pci_pme_restore(pci_dev);
 563}
 564
 565static void pci_pm_bridge_power_up_actions(struct pci_dev *pci_dev)
 566{
 567	int ret;
 568
 569	ret = pci_bridge_wait_for_secondary_bus(pci_dev, "resume");
 570	if (ret) {
 571		/*
 572		 * The downstream link failed to come up, so mark the
 573		 * devices below as disconnected to make sure we don't
 574		 * attempt to resume them.
 575		 */
 576		pci_walk_bus(pci_dev->subordinate, pci_dev_set_disconnected,
 577			     NULL);
 578		return;
 579	}
 580
 581	/*
 582	 * When powering on a bridge from D3cold, the whole hierarchy may be
 583	 * powered on into D0uninitialized state, resume them to give them a
 584	 * chance to suspend again
 585	 */
 586	pci_resume_bus(pci_dev->subordinate);
 587}
 588
 589#endif /* CONFIG_PM */
 590
 591#ifdef CONFIG_PM_SLEEP
 592
 593/*
 594 * Default "suspend" method for devices that have no driver provided suspend,
 595 * or not even a driver at all (second part).
 596 */
 597static void pci_pm_set_unknown_state(struct pci_dev *pci_dev)
 598{
 599	/*
 600	 * mark its power state as "unknown", since we don't know if
 601	 * e.g. the BIOS will change its device state when we suspend.
 602	 */
 603	if (pci_dev->current_state == PCI_D0)
 604		pci_dev->current_state = PCI_UNKNOWN;
 605}
 606
 607/*
 608 * Default "resume" method for devices that have no driver provided resume,
 609 * or not even a driver at all (second part).
 610 */
 611static int pci_pm_reenable_device(struct pci_dev *pci_dev)
 612{
 613	int retval;
 614
 615	/* if the device was enabled before suspend, re-enable */
 616	retval = pci_reenable_device(pci_dev);
 617	/*
 618	 * if the device was busmaster before the suspend, make it busmaster
 619	 * again
 620	 */
 621	if (pci_dev->is_busmaster)
 622		pci_set_master(pci_dev);
 623
 624	return retval;
 625}
 626
 627static int pci_legacy_suspend(struct device *dev, pm_message_t state)
 628{
 629	struct pci_dev *pci_dev = to_pci_dev(dev);
 630	struct pci_driver *drv = pci_dev->driver;
 631
 632	pci_dev->state_saved = false;
 633
 634	if (drv && drv->suspend) {
 635		pci_power_t prev = pci_dev->current_state;
 636		int error;
 637
 638		error = drv->suspend(pci_dev, state);
 639		suspend_report_result(dev, drv->suspend, error);
 640		if (error)
 641			return error;
 642
 643		if (!pci_dev->state_saved && pci_dev->current_state != PCI_D0
 644		    && pci_dev->current_state != PCI_UNKNOWN) {
 645			pci_WARN_ONCE(pci_dev, pci_dev->current_state != prev,
 646				      "PCI PM: Device state not saved by %pS\n",
 647				      drv->suspend);
 648		}
 649	}
 650
 651	pci_fixup_device(pci_fixup_suspend, pci_dev);
 652
 653	return 0;
 654}
 655
 656static int pci_legacy_suspend_late(struct device *dev)
 657{
 658	struct pci_dev *pci_dev = to_pci_dev(dev);
 659
 660	if (!pci_dev->state_saved)
 661		pci_save_state(pci_dev);
 662
 663	pci_pm_set_unknown_state(pci_dev);
 664
 665	pci_fixup_device(pci_fixup_suspend_late, pci_dev);
 666
 667	return 0;
 668}
 669
 670static int pci_legacy_resume(struct device *dev)
 671{
 672	struct pci_dev *pci_dev = to_pci_dev(dev);
 673	struct pci_driver *drv = pci_dev->driver;
 674
 675	pci_fixup_device(pci_fixup_resume, pci_dev);
 676
 677	return drv && drv->resume ?
 678			drv->resume(pci_dev) : pci_pm_reenable_device(pci_dev);
 679}
 680
 681/* Auxiliary functions used by the new power management framework */
 682
 683static void pci_pm_default_suspend(struct pci_dev *pci_dev)
 684{
 685	/* Disable non-bridge devices without PM support */
 686	if (!pci_has_subordinate(pci_dev))
 687		pci_disable_enabled_device(pci_dev);
 688}
 689
 690static bool pci_has_legacy_pm_support(struct pci_dev *pci_dev)
 691{
 692	struct pci_driver *drv = pci_dev->driver;
 693	bool ret = drv && (drv->suspend || drv->resume);
 694
 695	/*
 696	 * Legacy PM support is used by default, so warn if the new framework is
 697	 * supported as well.  Drivers are supposed to support either the
 698	 * former, or the latter, but not both at the same time.
 699	 */
 700	pci_WARN(pci_dev, ret && drv->driver.pm, "device %04x:%04x\n",
 701		 pci_dev->vendor, pci_dev->device);
 702
 703	return ret;
 704}
 705
 706/* New power management framework */
 707
 708static int pci_pm_prepare(struct device *dev)
 709{
 710	struct pci_dev *pci_dev = to_pci_dev(dev);
 711	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
 712
 713	dev_pm_set_strict_midlayer(dev, true);
 714
 715	if (pm && pm->prepare) {
 716		int error = pm->prepare(dev);
 717		if (error < 0)
 718			return error;
 719
 720		if (!error && dev_pm_test_driver_flags(dev, DPM_FLAG_SMART_PREPARE))
 721			return 0;
 722	}
 723	if (pci_dev_need_resume(pci_dev))
 724		return 0;
 725
 726	/*
 727	 * The PME setting needs to be adjusted here in case the direct-complete
 728	 * optimization is used with respect to this device.
 729	 */
 730	pci_dev_adjust_pme(pci_dev);
 731	return 1;
 732}
 733
 734static void pci_pm_complete(struct device *dev)
 735{
 736	struct pci_dev *pci_dev = to_pci_dev(dev);
 737
 738	pci_dev_complete_resume(pci_dev);
 739	pm_generic_complete(dev);
 740
 741	/* Resume device if platform firmware has put it in reset-power-on */
 742	if (pm_runtime_suspended(dev) && pm_resume_via_firmware()) {
 743		pci_power_t pre_sleep_state = pci_dev->current_state;
 744
 745		pci_refresh_power_state(pci_dev);
 746		/*
 747		 * On platforms with ACPI this check may also trigger for
 748		 * devices sharing power resources if one of those power
 749		 * resources has been activated as a result of a change of the
 750		 * power state of another device sharing it.  However, in that
 751		 * case it is also better to resume the device, in general.
 752		 */
 753		if (pci_dev->current_state < pre_sleep_state)
 754			pm_request_resume(dev);
 755	}
 756
 757	dev_pm_set_strict_midlayer(dev, false);
 758}
 759
 760#else /* !CONFIG_PM_SLEEP */
 761
 762#define pci_pm_prepare	NULL
 763#define pci_pm_complete	NULL
 764
 765#endif /* !CONFIG_PM_SLEEP */
 766
 767#ifdef CONFIG_SUSPEND
 768static void pcie_pme_root_status_cleanup(struct pci_dev *pci_dev)
 769{
 770	/*
 771	 * Some BIOSes forget to clear Root PME Status bits after system
 772	 * wakeup, which breaks ACPI-based runtime wakeup on PCI Express.
 773	 * Clear those bits now just in case (shouldn't hurt).
 774	 */
 775	if (pci_is_pcie(pci_dev) &&
 776	    (pci_pcie_type(pci_dev) == PCI_EXP_TYPE_ROOT_PORT ||
 777	     pci_pcie_type(pci_dev) == PCI_EXP_TYPE_RC_EC))
 778		pcie_clear_root_pme_status(pci_dev);
 779}
 780
 781static int pci_pm_suspend(struct device *dev)
 782{
 783	struct pci_dev *pci_dev = to_pci_dev(dev);
 784	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
 785
 786	pci_dev->skip_bus_pm = false;
 787
 788	/*
 789	 * Disabling PTM allows some systems, e.g., Intel mobile chips
 790	 * since Coffee Lake, to enter a lower-power PM state.
 791	 */
 792	pci_suspend_ptm(pci_dev);
 793
 794	if (pci_has_legacy_pm_support(pci_dev))
 795		return pci_legacy_suspend(dev, PMSG_SUSPEND);
 796
 797	if (!pm) {
 798		pci_pm_default_suspend(pci_dev);
 799		return 0;
 800	}
 801
 802	/*
 803	 * PCI devices suspended at run time may need to be resumed at this
 804	 * point, because in general it may be necessary to reconfigure them for
 805	 * system suspend.  Namely, if the device is expected to wake up the
 806	 * system from the sleep state, it may have to be reconfigured for this
 807	 * purpose, or if the device is not expected to wake up the system from
 808	 * the sleep state, it should be prevented from signaling wakeup events
 809	 * going forward.
 810	 *
 811	 * Also if the driver of the device does not indicate that its system
 812	 * suspend callbacks can cope with runtime-suspended devices, it is
 813	 * better to resume the device from runtime suspend here.
 814	 */
 815	if (!dev_pm_smart_suspend(dev) || pci_dev_need_resume(pci_dev)) {
 816		pm_runtime_resume(dev);
 817		pci_dev->state_saved = false;
 818	} else {
 819		pci_dev_adjust_pme(pci_dev);
 820	}
 821
 822	if (pm->suspend) {
 823		pci_power_t prev = pci_dev->current_state;
 824		int error;
 825
 826		error = pm->suspend(dev);
 827		suspend_report_result(dev, pm->suspend, error);
 828		if (error)
 829			return error;
 830
 831		if (!pci_dev->state_saved && pci_dev->current_state != PCI_D0
 832		    && pci_dev->current_state != PCI_UNKNOWN) {
 833			pci_WARN_ONCE(pci_dev, pci_dev->current_state != prev,
 834				      "PCI PM: State of device not saved by %pS\n",
 835				      pm->suspend);
 836		}
 837	}
 838
 839	return 0;
 840}
 841
 842static int pci_pm_suspend_late(struct device *dev)
 843{
 844	if (dev_pm_skip_suspend(dev))
 845		return 0;
 846
 847	pci_fixup_device(pci_fixup_suspend, to_pci_dev(dev));
 848
 849	return pm_generic_suspend_late(dev);
 850}
 851
 852static int pci_pm_suspend_noirq(struct device *dev)
 853{
 854	struct pci_dev *pci_dev = to_pci_dev(dev);
 855	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
 856
 857	if (dev_pm_skip_suspend(dev))
 858		return 0;
 859
 860	if (pci_has_legacy_pm_support(pci_dev))
 861		return pci_legacy_suspend_late(dev);
 862
 863	if (!pm) {
 864		pci_save_state(pci_dev);
 865		goto Fixup;
 866	}
 867
 868	if (pm->suspend_noirq) {
 869		pci_power_t prev = pci_dev->current_state;
 870		int error;
 871
 872		error = pm->suspend_noirq(dev);
 873		suspend_report_result(dev, pm->suspend_noirq, error);
 874		if (error)
 875			return error;
 876
 877		if (!pci_dev->state_saved && pci_dev->current_state != PCI_D0
 878		    && pci_dev->current_state != PCI_UNKNOWN) {
 879			pci_WARN_ONCE(pci_dev, pci_dev->current_state != prev,
 880				      "PCI PM: State of device not saved by %pS\n",
 881				      pm->suspend_noirq);
 882			goto Fixup;
 883		}
 884	}
 885
 886	if (!pci_dev->state_saved) {
 887		pci_save_state(pci_dev);
 888
 889		/*
 890		 * If the device is a bridge with a child in D0 below it,
 891		 * it needs to stay in D0, so check skip_bus_pm to avoid
 892		 * putting it into a low-power state in that case.
 893		 */
 894		if (!pci_dev->skip_bus_pm && pci_power_manageable(pci_dev))
 895			pci_prepare_to_sleep(pci_dev);
 896	}
 897
 898	pci_dbg(pci_dev, "PCI PM: Suspend power state: %s\n",
 899		pci_power_name(pci_dev->current_state));
 900
 901	if (pci_dev->current_state == PCI_D0) {
 902		pci_dev->skip_bus_pm = true;
 903		/*
 904		 * Per PCI PM r1.2, table 6-1, a bridge must be in D0 if any
 905		 * downstream device is in D0, so avoid changing the power state
 906		 * of the parent bridge by setting the skip_bus_pm flag for it.
 907		 */
 908		if (pci_dev->bus->self)
 909			pci_dev->bus->self->skip_bus_pm = true;
 910	}
 911
 912	if (pci_dev->skip_bus_pm && pm_suspend_no_platform()) {
 913		pci_dbg(pci_dev, "PCI PM: Skipped\n");
 914		goto Fixup;
 915	}
 916
 917	pci_pm_set_unknown_state(pci_dev);
 918
 919	/*
 920	 * Some BIOSes from ASUS have a bug: If a USB EHCI host controller's
 921	 * PCI COMMAND register isn't 0, the BIOS assumes that the controller
 922	 * hasn't been quiesced and tries to turn it off.  If the controller
 923	 * is already in D3, this can hang or cause memory corruption.
 924	 *
 925	 * Since the value of the COMMAND register doesn't matter once the
 926	 * device has been suspended, we can safely set it to 0 here.
 927	 */
 928	if (pci_dev->class == PCI_CLASS_SERIAL_USB_EHCI)
 929		pci_write_config_word(pci_dev, PCI_COMMAND, 0);
 930
 931Fixup:
 932	pci_fixup_device(pci_fixup_suspend_late, pci_dev);
 933
 934	/*
 935	 * If the target system sleep state is suspend-to-idle, it is sufficient
 936	 * to check whether or not the device's wakeup settings are good for
 937	 * runtime PM.  Otherwise, the pm_resume_via_firmware() check will cause
 938	 * pci_pm_complete() to take care of fixing up the device's state
 939	 * anyway, if need be.
 940	 */
 941	if (device_can_wakeup(dev) && !device_may_wakeup(dev))
 942		dev->power.may_skip_resume = false;
 943
 944	return 0;
 945}
 946
 947static int pci_pm_resume_noirq(struct device *dev)
 948{
 949	struct pci_dev *pci_dev = to_pci_dev(dev);
 950	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
 951	pci_power_t prev_state = pci_dev->current_state;
 952	bool skip_bus_pm = pci_dev->skip_bus_pm;
 953
 954	if (dev_pm_skip_resume(dev))
 955		return 0;
 956
 957	/*
 958	 * In the suspend-to-idle case, devices left in D0 during suspend will
 959	 * stay in D0, so it is not necessary to restore or update their
 960	 * configuration here and attempting to put them into D0 again is
 961	 * pointless, so avoid doing that.
 962	 */
 963	if (!(skip_bus_pm && pm_suspend_no_platform()))
 964		pci_pm_default_resume_early(pci_dev);
 965
 966	pci_fixup_device(pci_fixup_resume_early, pci_dev);
 967	pcie_pme_root_status_cleanup(pci_dev);
 968
 969	if (!skip_bus_pm && prev_state == PCI_D3cold)
 970		pci_pm_bridge_power_up_actions(pci_dev);
 971
 972	if (pci_has_legacy_pm_support(pci_dev))
 973		return 0;
 974
 975	if (pm && pm->resume_noirq)
 976		return pm->resume_noirq(dev);
 977
 978	return 0;
 979}
 980
 981static int pci_pm_resume_early(struct device *dev)
 982{
 983	if (dev_pm_skip_resume(dev))
 984		return 0;
 985
 986	return pm_generic_resume_early(dev);
 987}
 988
 989static int pci_pm_resume(struct device *dev)
 990{
 991	struct pci_dev *pci_dev = to_pci_dev(dev);
 992	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
 993
 994	/*
 995	 * This is necessary for the suspend error path in which resume is
 996	 * called without restoring the standard config registers of the device.
 997	 */
 998	if (pci_dev->state_saved)
 999		pci_restore_standard_config(pci_dev);
1000
1001	pci_resume_ptm(pci_dev);
1002
1003	if (pci_has_legacy_pm_support(pci_dev))
1004		return pci_legacy_resume(dev);
1005
1006	pci_pm_default_resume(pci_dev);
1007
1008	if (pm) {
1009		if (pm->resume)
1010			return pm->resume(dev);
1011	} else {
1012		pci_pm_reenable_device(pci_dev);
1013	}
1014
1015	return 0;
1016}
1017
1018#else /* !CONFIG_SUSPEND */
1019
1020#define pci_pm_suspend		NULL
1021#define pci_pm_suspend_late	NULL
1022#define pci_pm_suspend_noirq	NULL
1023#define pci_pm_resume		NULL
1024#define pci_pm_resume_early	NULL
1025#define pci_pm_resume_noirq	NULL
1026
1027#endif /* !CONFIG_SUSPEND */
1028
1029#ifdef CONFIG_HIBERNATE_CALLBACKS
1030
1031static int pci_pm_freeze(struct device *dev)
1032{
1033	struct pci_dev *pci_dev = to_pci_dev(dev);
1034	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1035
1036	if (pci_has_legacy_pm_support(pci_dev))
1037		return pci_legacy_suspend(dev, PMSG_FREEZE);
1038
1039	if (!pm) {
1040		pci_pm_default_suspend(pci_dev);
1041		if (!pm_runtime_suspended(dev))
1042			pci_dev->state_saved = false;
1043		return 0;
1044	}
1045
1046	/*
1047	 * Resume all runtime-suspended devices before creating a snapshot
1048	 * image of system memory, because the restore kernel generally cannot
1049	 * be expected to always handle them consistently and they need to be
1050	 * put into the runtime-active metastate during system resume anyway,
1051	 * so it is better to ensure that the state saved in the image will be
1052	 * always consistent with that.
1053	 */
1054	pm_runtime_resume(dev);
1055	pci_dev->state_saved = false;
1056
1057	if (pm->freeze) {
1058		int error;
1059
1060		error = pm->freeze(dev);
1061		suspend_report_result(dev, pm->freeze, error);
1062		if (error)
1063			return error;
1064	}
1065
1066	return 0;
1067}
1068
1069static int pci_pm_freeze_noirq(struct device *dev)
1070{
1071	struct pci_dev *pci_dev = to_pci_dev(dev);
1072	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1073
1074	if (pci_has_legacy_pm_support(pci_dev))
1075		return pci_legacy_suspend_late(dev);
1076
1077	if (pm && pm->freeze_noirq) {
1078		int error;
1079
1080		error = pm->freeze_noirq(dev);
1081		suspend_report_result(dev, pm->freeze_noirq, error);
1082		if (error)
1083			return error;
1084	}
1085
1086	if (!pci_dev->state_saved)
1087		pci_save_state(pci_dev);
1088
1089	pci_pm_set_unknown_state(pci_dev);
1090
1091	return 0;
1092}
1093
1094static int pci_pm_thaw_noirq(struct device *dev)
1095{
1096	struct pci_dev *pci_dev = to_pci_dev(dev);
1097	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1098
1099	/*
1100	 * The pm->thaw_noirq() callback assumes the device has been
1101	 * returned to D0 and its config state has been restored.
1102	 *
1103	 * In addition, pci_restore_state() restores MSI-X state in MMIO
1104	 * space, which requires the device to be in D0, so return it to D0
1105	 * in case the driver's "freeze" callbacks put it into a low-power
1106	 * state.
1107	 */
1108	pci_pm_power_up_and_verify_state(pci_dev);
1109	pci_restore_state(pci_dev);
1110
1111	if (pci_has_legacy_pm_support(pci_dev))
1112		return 0;
1113
1114	if (pm && pm->thaw_noirq)
1115		return pm->thaw_noirq(dev);
1116
1117	return 0;
1118}
1119
1120static int pci_pm_thaw(struct device *dev)
1121{
1122	struct pci_dev *pci_dev = to_pci_dev(dev);
1123	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1124	int error = 0;
1125
1126	if (pci_has_legacy_pm_support(pci_dev))
1127		return pci_legacy_resume(dev);
1128
1129	if (pm) {
1130		if (pm->thaw)
1131			error = pm->thaw(dev);
1132	} else {
1133		pci_pm_reenable_device(pci_dev);
1134	}
1135
1136	return error;
1137}
1138
1139static int pci_pm_poweroff(struct device *dev)
1140{
1141	struct pci_dev *pci_dev = to_pci_dev(dev);
1142	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1143
1144	if (pci_has_legacy_pm_support(pci_dev))
1145		return pci_legacy_suspend(dev, PMSG_HIBERNATE);
1146
1147	if (!pm) {
1148		pci_pm_default_suspend(pci_dev);
1149		return 0;
1150	}
1151
1152	/* The reason to do that is the same as in pci_pm_suspend(). */
1153	if (!dev_pm_smart_suspend(dev) || pci_dev_need_resume(pci_dev)) {
1154		pm_runtime_resume(dev);
1155		pci_dev->state_saved = false;
1156	} else {
1157		pci_dev_adjust_pme(pci_dev);
1158	}
1159
1160	if (pm->poweroff) {
1161		int error;
1162
1163		error = pm->poweroff(dev);
1164		suspend_report_result(dev, pm->poweroff, error);
1165		if (error)
1166			return error;
1167	}
1168
1169	return 0;
1170}
1171
1172static int pci_pm_poweroff_late(struct device *dev)
1173{
1174	if (dev_pm_skip_suspend(dev))
1175		return 0;
1176
1177	pci_fixup_device(pci_fixup_suspend, to_pci_dev(dev));
1178
1179	return pm_generic_poweroff_late(dev);
1180}
1181
1182static int pci_pm_poweroff_noirq(struct device *dev)
1183{
1184	struct pci_dev *pci_dev = to_pci_dev(dev);
1185	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1186
1187	if (dev_pm_skip_suspend(dev))
1188		return 0;
1189
1190	if (pci_has_legacy_pm_support(pci_dev))
1191		return pci_legacy_suspend_late(dev);
1192
1193	if (!pm) {
1194		pci_fixup_device(pci_fixup_suspend_late, pci_dev);
1195		return 0;
1196	}
1197
1198	if (pm->poweroff_noirq) {
1199		int error;
1200
1201		error = pm->poweroff_noirq(dev);
1202		suspend_report_result(dev, pm->poweroff_noirq, error);
1203		if (error)
1204			return error;
1205	}
1206
1207	if (!pci_dev->state_saved && !pci_has_subordinate(pci_dev))
1208		pci_prepare_to_sleep(pci_dev);
1209
1210	/*
1211	 * The reason for doing this here is the same as for the analogous code
1212	 * in pci_pm_suspend_noirq().
1213	 */
1214	if (pci_dev->class == PCI_CLASS_SERIAL_USB_EHCI)
1215		pci_write_config_word(pci_dev, PCI_COMMAND, 0);
1216
1217	pci_fixup_device(pci_fixup_suspend_late, pci_dev);
1218
1219	return 0;
1220}
1221
1222static int pci_pm_restore_noirq(struct device *dev)
1223{
1224	struct pci_dev *pci_dev = to_pci_dev(dev);
1225	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1226
1227	pci_pm_default_resume_early(pci_dev);
1228	pci_fixup_device(pci_fixup_resume_early, pci_dev);
1229
1230	if (pci_has_legacy_pm_support(pci_dev))
1231		return 0;
1232
1233	if (pm && pm->restore_noirq)
1234		return pm->restore_noirq(dev);
1235
1236	return 0;
1237}
1238
1239static int pci_pm_restore(struct device *dev)
1240{
1241	struct pci_dev *pci_dev = to_pci_dev(dev);
1242	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1243
1244	/*
1245	 * This is necessary for the hibernation error path in which restore is
1246	 * called without restoring the standard config registers of the device.
1247	 */
1248	if (pci_dev->state_saved)
1249		pci_restore_standard_config(pci_dev);
1250
1251	if (pci_has_legacy_pm_support(pci_dev))
1252		return pci_legacy_resume(dev);
1253
1254	pci_pm_default_resume(pci_dev);
1255
1256	if (pm) {
1257		if (pm->restore)
1258			return pm->restore(dev);
1259	} else {
1260		pci_pm_reenable_device(pci_dev);
1261	}
1262
1263	return 0;
1264}
1265
1266#else /* !CONFIG_HIBERNATE_CALLBACKS */
1267
1268#define pci_pm_freeze		NULL
1269#define pci_pm_freeze_noirq	NULL
1270#define pci_pm_thaw		NULL
1271#define pci_pm_thaw_noirq	NULL
1272#define pci_pm_poweroff		NULL
1273#define pci_pm_poweroff_late	NULL
1274#define pci_pm_poweroff_noirq	NULL
1275#define pci_pm_restore		NULL
1276#define pci_pm_restore_noirq	NULL
1277
1278#endif /* !CONFIG_HIBERNATE_CALLBACKS */
1279
1280#ifdef CONFIG_PM
1281
1282static int pci_pm_runtime_suspend(struct device *dev)
1283{
1284	struct pci_dev *pci_dev = to_pci_dev(dev);
1285	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1286	pci_power_t prev = pci_dev->current_state;
1287	int error;
1288
1289	pci_suspend_ptm(pci_dev);
1290
1291	/*
1292	 * If pci_dev->driver is not set (unbound), we leave the device in D0,
1293	 * but it may go to D3cold when the bridge above it runtime suspends.
1294	 * Save its config space in case that happens.
1295	 */
1296	if (!pci_dev->driver) {
1297		pci_save_state(pci_dev);
1298		return 0;
1299	}
1300
1301	pci_dev->state_saved = false;
1302	if (pm && pm->runtime_suspend) {
1303		error = pm->runtime_suspend(dev);
1304		/*
1305		 * -EBUSY and -EAGAIN is used to request the runtime PM core
1306		 * to schedule a new suspend, so log the event only with debug
1307		 * log level.
1308		 */
1309		if (error == -EBUSY || error == -EAGAIN) {
1310			pci_dbg(pci_dev, "can't suspend now (%ps returned %d)\n",
1311				pm->runtime_suspend, error);
1312			return error;
1313		} else if (error) {
1314			pci_err(pci_dev, "can't suspend (%ps returned %d)\n",
1315				pm->runtime_suspend, error);
1316			return error;
1317		}
1318	}
1319
1320	pci_fixup_device(pci_fixup_suspend, pci_dev);
1321
1322	if (pm && pm->runtime_suspend
1323	    && !pci_dev->state_saved && pci_dev->current_state != PCI_D0
1324	    && pci_dev->current_state != PCI_UNKNOWN) {
1325		pci_WARN_ONCE(pci_dev, pci_dev->current_state != prev,
1326			      "PCI PM: State of device not saved by %pS\n",
1327			      pm->runtime_suspend);
1328		return 0;
1329	}
1330
1331	if (!pci_dev->state_saved) {
1332		pci_save_state(pci_dev);
1333		pci_finish_runtime_suspend(pci_dev);
1334	}
1335
1336	return 0;
1337}
1338
1339static int pci_pm_runtime_resume(struct device *dev)
1340{
1341	struct pci_dev *pci_dev = to_pci_dev(dev);
1342	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1343	pci_power_t prev_state = pci_dev->current_state;
1344	int error = 0;
1345
1346	/*
1347	 * Restoring config space is necessary even if the device is not bound
1348	 * to a driver because although we left it in D0, it may have gone to
1349	 * D3cold when the bridge above it runtime suspended.
1350	 */
1351	pci_pm_default_resume_early(pci_dev);
1352	pci_resume_ptm(pci_dev);
1353
1354	if (!pci_dev->driver)
1355		return 0;
1356
1357	pci_fixup_device(pci_fixup_resume_early, pci_dev);
1358	pci_pm_default_resume(pci_dev);
1359
1360	if (prev_state == PCI_D3cold)
1361		pci_pm_bridge_power_up_actions(pci_dev);
1362
1363	if (pm && pm->runtime_resume)
1364		error = pm->runtime_resume(dev);
1365
1366	return error;
1367}
1368
1369static int pci_pm_runtime_idle(struct device *dev)
1370{
1371	struct pci_dev *pci_dev = to_pci_dev(dev);
1372	const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL;
1373
1374	/*
1375	 * If pci_dev->driver is not set (unbound), the device should
1376	 * always remain in D0 regardless of the runtime PM status
1377	 */
1378	if (!pci_dev->driver)
1379		return 0;
1380
1381	if (pm && pm->runtime_idle)
1382		return pm->runtime_idle(dev);
1383
1384	return 0;
1385}
1386
1387static const struct dev_pm_ops pci_dev_pm_ops = {
1388	.prepare = pci_pm_prepare,
1389	.complete = pci_pm_complete,
1390	.suspend = pci_pm_suspend,
1391	.suspend_late = pci_pm_suspend_late,
1392	.resume = pci_pm_resume,
1393	.resume_early = pci_pm_resume_early,
1394	.freeze = pci_pm_freeze,
1395	.thaw = pci_pm_thaw,
1396	.poweroff = pci_pm_poweroff,
1397	.poweroff_late = pci_pm_poweroff_late,
1398	.restore = pci_pm_restore,
1399	.suspend_noirq = pci_pm_suspend_noirq,
1400	.resume_noirq = pci_pm_resume_noirq,
1401	.freeze_noirq = pci_pm_freeze_noirq,
1402	.thaw_noirq = pci_pm_thaw_noirq,
1403	.poweroff_noirq = pci_pm_poweroff_noirq,
1404	.restore_noirq = pci_pm_restore_noirq,
1405	.runtime_suspend = pci_pm_runtime_suspend,
1406	.runtime_resume = pci_pm_runtime_resume,
1407	.runtime_idle = pci_pm_runtime_idle,
1408};
1409
1410#define PCI_PM_OPS_PTR	(&pci_dev_pm_ops)
1411
1412#else /* !CONFIG_PM */
1413
1414#define pci_pm_runtime_suspend	NULL
1415#define pci_pm_runtime_resume	NULL
1416#define pci_pm_runtime_idle	NULL
1417
1418#define PCI_PM_OPS_PTR	NULL
1419
1420#endif /* !CONFIG_PM */
1421
1422/**
1423 * __pci_register_driver - register a new pci driver
1424 * @drv: the driver structure to register
1425 * @owner: owner module of drv
1426 * @mod_name: module name string
1427 *
1428 * Adds the driver structure to the list of registered drivers.
1429 * Returns a negative value on error, otherwise 0.
1430 * If no error occurred, the driver remains registered even if
1431 * no device was claimed during registration.
1432 */
1433int __pci_register_driver(struct pci_driver *drv, struct module *owner,
1434			  const char *mod_name)
1435{
1436	/* initialize common driver fields */
1437	drv->driver.name = drv->name;
1438	drv->driver.bus = &pci_bus_type;
1439	drv->driver.owner = owner;
1440	drv->driver.mod_name = mod_name;
1441	drv->driver.groups = drv->groups;
1442	drv->driver.dev_groups = drv->dev_groups;
1443
1444	spin_lock_init(&drv->dynids.lock);
1445	INIT_LIST_HEAD(&drv->dynids.list);
1446
1447	/* register with core */
1448	return driver_register(&drv->driver);
1449}
1450EXPORT_SYMBOL(__pci_register_driver);
1451
1452/**
1453 * pci_unregister_driver - unregister a pci driver
1454 * @drv: the driver structure to unregister
1455 *
1456 * Deletes the driver structure from the list of registered PCI drivers,
1457 * gives it a chance to clean up by calling its remove() function for
1458 * each device it was responsible for, and marks those devices as
1459 * driverless.
1460 */
1461
1462void pci_unregister_driver(struct pci_driver *drv)
1463{
1464	driver_unregister(&drv->driver);
1465	pci_free_dynids(drv);
1466}
1467EXPORT_SYMBOL(pci_unregister_driver);
1468
1469static struct pci_driver pci_compat_driver = {
1470	.name = "compat"
1471};
1472
1473/**
1474 * pci_dev_driver - get the pci_driver of a device
1475 * @dev: the device to query
1476 *
1477 * Returns the appropriate pci_driver structure or %NULL if there is no
1478 * registered driver for the device.
1479 */
1480struct pci_driver *pci_dev_driver(const struct pci_dev *dev)
1481{
1482	int i;
1483
1484	if (dev->driver)
1485		return dev->driver;
1486
1487	for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1488		if (dev->resource[i].flags & IORESOURCE_BUSY)
1489			return &pci_compat_driver;
1490
1491	return NULL;
1492}
1493EXPORT_SYMBOL(pci_dev_driver);
1494
1495/**
1496 * pci_bus_match - Tell if a PCI device structure has a matching PCI device id structure
1497 * @dev: the PCI device structure to match against
1498 * @drv: the device driver to search for matching PCI device id structures
1499 *
1500 * Used by a driver to check whether a PCI device present in the
1501 * system is in its list of supported devices. Returns the matching
1502 * pci_device_id structure or %NULL if there is no match.
1503 */
1504static int pci_bus_match(struct device *dev, const struct device_driver *drv)
1505{
1506	struct pci_dev *pci_dev = to_pci_dev(dev);
1507	struct pci_driver *pci_drv;
1508	const struct pci_device_id *found_id;
1509
1510	if (pci_dev_binding_disallowed(pci_dev))
1511		return 0;
1512
1513	pci_drv = (struct pci_driver *)to_pci_driver(drv);
1514	found_id = pci_match_device(pci_drv, pci_dev);
1515	if (found_id)
1516		return 1;
1517
1518	return 0;
1519}
1520
1521/**
1522 * pci_dev_get - increments the reference count of the pci device structure
1523 * @dev: the device being referenced
1524 *
1525 * Each live reference to a device should be refcounted.
1526 *
1527 * Drivers for PCI devices should normally record such references in
1528 * their probe() methods, when they bind to a device, and release
1529 * them by calling pci_dev_put(), in their disconnect() methods.
1530 *
1531 * A pointer to the device with the incremented reference counter is returned.
1532 */
1533struct pci_dev *pci_dev_get(struct pci_dev *dev)
1534{
1535	if (dev)
1536		get_device(&dev->dev);
1537	return dev;
1538}
1539EXPORT_SYMBOL(pci_dev_get);
1540
1541/**
1542 * pci_dev_put - release a use of the pci device structure
1543 * @dev: device that's been disconnected
1544 *
1545 * Must be called when a user of a device is finished with it.  When the last
1546 * user of the device calls this function, the memory of the device is freed.
1547 */
1548void pci_dev_put(struct pci_dev *dev)
1549{
1550	if (dev)
1551		put_device(&dev->dev);
1552}
1553EXPORT_SYMBOL(pci_dev_put);
1554
1555static int pci_uevent(const struct device *dev, struct kobj_uevent_env *env)
1556{
1557	const struct pci_dev *pdev;
1558
1559	if (!dev)
1560		return -ENODEV;
1561
1562	pdev = to_pci_dev(dev);
1563
1564	if (add_uevent_var(env, "PCI_CLASS=%04X", pdev->class))
1565		return -ENOMEM;
1566
1567	if (add_uevent_var(env, "PCI_ID=%04X:%04X", pdev->vendor, pdev->device))
1568		return -ENOMEM;
1569
1570	if (add_uevent_var(env, "PCI_SUBSYS_ID=%04X:%04X", pdev->subsystem_vendor,
1571			   pdev->subsystem_device))
1572		return -ENOMEM;
1573
1574	if (add_uevent_var(env, "PCI_SLOT_NAME=%s", pci_name(pdev)))
1575		return -ENOMEM;
1576
1577	if (add_uevent_var(env, "MODALIAS=pci:v%08Xd%08Xsv%08Xsd%08Xbc%02Xsc%02Xi%02X",
1578			   pdev->vendor, pdev->device,
1579			   pdev->subsystem_vendor, pdev->subsystem_device,
1580			   (u8)(pdev->class >> 16), (u8)(pdev->class >> 8),
1581			   (u8)(pdev->class)))
1582		return -ENOMEM;
1583
1584	return 0;
1585}
1586
1587#if defined(CONFIG_PCIEAER) || defined(CONFIG_EEH) || defined(CONFIG_S390)
1588/**
1589 * pci_uevent_ers - emit a uevent during recovery path of PCI device
1590 * @pdev: PCI device undergoing error recovery
1591 * @err_type: type of error event
1592 */
1593void pci_uevent_ers(struct pci_dev *pdev, enum pci_ers_result err_type)
1594{
1595	int idx = 0;
1596	char *envp[3];
1597
1598	switch (err_type) {
1599	case PCI_ERS_RESULT_NONE:
1600	case PCI_ERS_RESULT_CAN_RECOVER:
1601	case PCI_ERS_RESULT_NEED_RESET:
1602		envp[idx++] = "ERROR_EVENT=BEGIN_RECOVERY";
1603		envp[idx++] = "DEVICE_ONLINE=0";
1604		break;
1605	case PCI_ERS_RESULT_RECOVERED:
1606		envp[idx++] = "ERROR_EVENT=SUCCESSFUL_RECOVERY";
1607		envp[idx++] = "DEVICE_ONLINE=1";
1608		break;
1609	case PCI_ERS_RESULT_DISCONNECT:
1610		envp[idx++] = "ERROR_EVENT=FAILED_RECOVERY";
1611		envp[idx++] = "DEVICE_ONLINE=0";
1612		break;
1613	default:
1614		break;
1615	}
1616
1617	if (idx > 0) {
1618		envp[idx++] = NULL;
1619		kobject_uevent_env(&pdev->dev.kobj, KOBJ_CHANGE, envp);
1620	}
1621}
1622#endif
1623
1624static int pci_bus_num_vf(struct device *dev)
1625{
1626	return pci_num_vf(to_pci_dev(dev));
1627}
1628
1629/**
1630 * pci_dma_configure - Setup DMA configuration
1631 * @dev: ptr to dev structure
1632 *
1633 * Function to update PCI devices's DMA configuration using the same
1634 * info from the OF node or ACPI node of host bridge's parent (if any).
1635 */
1636static int pci_dma_configure(struct device *dev)
1637{
1638	const struct device_driver *drv = READ_ONCE(dev->driver);
1639	struct device *bridge;
1640	int ret = 0;
1641
1642	bridge = pci_get_host_bridge_device(to_pci_dev(dev));
1643
1644	if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
1645	    bridge->parent->of_node) {
1646		ret = of_dma_configure(dev, bridge->parent->of_node, true);
1647	} else if (has_acpi_companion(bridge)) {
1648		struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
1649
1650		ret = acpi_dma_configure(dev, acpi_get_dma_attr(adev));
1651	}
1652
1653	/*
1654	 * Attempt to enable ACS regardless of capability because some Root
1655	 * Ports (e.g. those quirked with *_intel_pch_acs_*) do not have
1656	 * the standard ACS capability but still support ACS via those
1657	 * quirks.
1658	 */
1659	pci_enable_acs(to_pci_dev(dev));
1660
1661	pci_put_host_bridge_device(bridge);
1662
1663	/* @drv may not be valid when we're called from the IOMMU layer */
1664	if (!ret && drv && !to_pci_driver(drv)->driver_managed_dma) {
1665		ret = iommu_device_use_default_domain(dev);
1666		if (ret)
1667			arch_teardown_dma_ops(dev);
1668	}
1669
1670	return ret;
1671}
1672
1673static void pci_dma_cleanup(struct device *dev)
1674{
1675	struct pci_driver *driver = to_pci_driver(dev->driver);
1676
1677	if (!driver->driver_managed_dma)
1678		iommu_device_unuse_default_domain(dev);
1679}
1680
1681/*
1682 * pci_device_irq_get_affinity - get IRQ affinity mask for device
1683 * @dev: ptr to dev structure
1684 * @irq_vec: interrupt vector number
1685 *
1686 * Return the CPU affinity mask for @dev and @irq_vec.
1687 */
1688static const struct cpumask *pci_device_irq_get_affinity(struct device *dev,
1689					unsigned int irq_vec)
1690{
1691	return pci_irq_get_affinity(to_pci_dev(dev), irq_vec);
1692}
1693
1694const struct bus_type pci_bus_type = {
1695	.name		= "pci",
1696	.match		= pci_bus_match,
1697	.uevent		= pci_uevent,
1698	.probe		= pci_device_probe,
1699	.remove		= pci_device_remove,
1700	.shutdown	= pci_device_shutdown,
1701	.irq_get_affinity = pci_device_irq_get_affinity,
1702	.dev_groups	= pci_dev_groups,
1703	.bus_groups	= pci_bus_groups,
1704	.drv_groups	= pci_drv_groups,
1705	.pm		= PCI_PM_OPS_PTR,
1706	.num_vf		= pci_bus_num_vf,
1707	.dma_configure	= pci_dma_configure,
1708	.dma_cleanup	= pci_dma_cleanup,
1709};
1710EXPORT_SYMBOL(pci_bus_type);
1711
1712#ifdef CONFIG_PCIEPORTBUS
1713static int pcie_port_bus_match(struct device *dev, const struct device_driver *drv)
1714{
1715	struct pcie_device *pciedev;
1716	const struct pcie_port_service_driver *driver;
1717
1718	if (drv->bus != &pcie_port_bus_type || dev->bus != &pcie_port_bus_type)
1719		return 0;
1720
1721	pciedev = to_pcie_device(dev);
1722	driver = to_service_driver(drv);
1723
1724	if (driver->service != pciedev->service)
1725		return 0;
1726
1727	if (driver->port_type != PCIE_ANY_PORT &&
1728	    driver->port_type != pci_pcie_type(pciedev->port))
1729		return 0;
1730
1731	return 1;
1732}
1733
1734const struct bus_type pcie_port_bus_type = {
1735	.name		= "pci_express",
1736	.match		= pcie_port_bus_match,
1737};
1738#endif
1739
1740static int __init pci_driver_init(void)
1741{
1742	int ret;
1743
1744	ret = bus_register(&pci_bus_type);
1745	if (ret)
1746		return ret;
1747
1748#ifdef CONFIG_PCIEPORTBUS
1749	ret = bus_register(&pcie_port_bus_type);
1750	if (ret)
1751		return ret;
1752#endif
1753	dma_debug_add_bus(&pci_bus_type);
1754	return 0;
1755}
1756postcore_initcall(pci_driver_init);