1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * PCI Bus Services, see include/linux/pci.h for further explanation.
   4 *
   5 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
   6 * David Mosberger-Tang
   7 *
   8 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
   9 */
  10
  11#include <linux/acpi.h>
  12#include <linux/kernel.h>
  13#include <linux/delay.h>
  14#include <linux/dmi.h>
  15#include <linux/init.h>
  16#include <linux/msi.h>
  17#include <linux/of.h>
  18#include <linux/of_pci.h>
  19#include <linux/pci.h>
  20#include <linux/pm.h>
  21#include <linux/slab.h>
  22#include <linux/module.h>
  23#include <linux/spinlock.h>
  24#include <linux/string.h>
  25#include <linux/log2.h>
  26#include <linux/logic_pio.h>
  27#include <linux/pm_wakeup.h>
  28#include <linux/interrupt.h>
  29#include <linux/device.h>
  30#include <linux/pm_runtime.h>
  31#include <linux/pci_hotplug.h>
  32#include <linux/vmalloc.h>
  33#include <linux/pci-ats.h>
  34#include <asm/setup.h>
  35#include <asm/dma.h>
  36#include <linux/aer.h>
  37#include "pci.h"
  38
  39DEFINE_MUTEX(pci_slot_mutex);
  40
  41const char *pci_power_names[] = {
  42	"error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
  43};
  44EXPORT_SYMBOL_GPL(pci_power_names);
  45
  46int isa_dma_bridge_buggy;
  47EXPORT_SYMBOL(isa_dma_bridge_buggy);
  48
  49int pci_pci_problems;
  50EXPORT_SYMBOL(pci_pci_problems);
  51
  52unsigned int pci_pm_d3_delay;
  53
  54static void pci_pme_list_scan(struct work_struct *work);
  55
  56static LIST_HEAD(pci_pme_list);
  57static DEFINE_MUTEX(pci_pme_list_mutex);
  58static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
  59
  60struct pci_pme_device {
  61	struct list_head list;
  62	struct pci_dev *dev;
  63};
  64
  65#define PME_TIMEOUT 1000 /* How long between PME checks */
  66
  67static void pci_dev_d3_sleep(struct pci_dev *dev)
  68{
  69	unsigned int delay = dev->d3_delay;
  70
  71	if (delay < pci_pm_d3_delay)
  72		delay = pci_pm_d3_delay;
  73
  74	if (delay)
  75		msleep(delay);
  76}
  77
  78#ifdef CONFIG_PCI_DOMAINS
  79int pci_domains_supported = 1;
  80#endif
  81
  82#define DEFAULT_CARDBUS_IO_SIZE		(256)
  83#define DEFAULT_CARDBUS_MEM_SIZE	(64*1024*1024)
  84/* pci=cbmemsize=nnM,cbiosize=nn can override this */
  85unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
  86unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
  87
  88#define DEFAULT_HOTPLUG_IO_SIZE		(256)
  89#define DEFAULT_HOTPLUG_MMIO_SIZE	(2*1024*1024)
  90#define DEFAULT_HOTPLUG_MMIO_PREF_SIZE	(2*1024*1024)
  91/* hpiosize=nn can override this */
  92unsigned long pci_hotplug_io_size  = DEFAULT_HOTPLUG_IO_SIZE;
  93/*
  94 * pci=hpmmiosize=nnM overrides non-prefetchable MMIO size,
  95 * pci=hpmmioprefsize=nnM overrides prefetchable MMIO size;
  96 * pci=hpmemsize=nnM overrides both
  97 */
  98unsigned long pci_hotplug_mmio_size = DEFAULT_HOTPLUG_MMIO_SIZE;
  99unsigned long pci_hotplug_mmio_pref_size = DEFAULT_HOTPLUG_MMIO_PREF_SIZE;
 100
 101#define DEFAULT_HOTPLUG_BUS_SIZE	1
 102unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
 103
 104enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
 105
 106/*
 107 * The default CLS is used if arch didn't set CLS explicitly and not
 108 * all pci devices agree on the same value.  Arch can override either
 109 * the dfl or actual value as it sees fit.  Don't forget this is
 110 * measured in 32-bit words, not bytes.
 111 */
 112u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
 113u8 pci_cache_line_size;
 114
 115/*
 116 * If we set up a device for bus mastering, we need to check the latency
 117 * timer as certain BIOSes forget to set it properly.
 118 */
 119unsigned int pcibios_max_latency = 255;
 120
 121/* If set, the PCIe ARI capability will not be used. */
 122static bool pcie_ari_disabled;
 123
 124/* If set, the PCIe ATS capability will not be used. */
 125static bool pcie_ats_disabled;
 126
 127/* If set, the PCI config space of each device is printed during boot. */
 128bool pci_early_dump;
 129
 130bool pci_ats_disabled(void)
 131{
 132	return pcie_ats_disabled;
 133}
 134EXPORT_SYMBOL_GPL(pci_ats_disabled);
 135
 136/* Disable bridge_d3 for all PCIe ports */
 137static bool pci_bridge_d3_disable;
 138/* Force bridge_d3 for all PCIe ports */
 139static bool pci_bridge_d3_force;
 140
 141static int __init pcie_port_pm_setup(char *str)
 142{
 143	if (!strcmp(str, "off"))
 144		pci_bridge_d3_disable = true;
 145	else if (!strcmp(str, "force"))
 146		pci_bridge_d3_force = true;
 147	return 1;
 148}
 149__setup("pcie_port_pm=", pcie_port_pm_setup);
 150
 151/* Time to wait after a reset for device to become responsive */
 152#define PCIE_RESET_READY_POLL_MS 60000
 153
 154/**
 155 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
 156 * @bus: pointer to PCI bus structure to search
 157 *
 158 * Given a PCI bus, returns the highest PCI bus number present in the set
 159 * including the given PCI bus and its list of child PCI buses.
 160 */
 161unsigned char pci_bus_max_busnr(struct pci_bus *bus)
 162{
 163	struct pci_bus *tmp;
 164	unsigned char max, n;
 165
 166	max = bus->busn_res.end;
 167	list_for_each_entry(tmp, &bus->children, node) {
 168		n = pci_bus_max_busnr(tmp);
 169		if (n > max)
 170			max = n;
 171	}
 172	return max;
 173}
 174EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
 175
 176/**
 177 * pci_status_get_and_clear_errors - return and clear error bits in PCI_STATUS
 178 * @pdev: the PCI device
 179 *
 180 * Returns error bits set in PCI_STATUS and clears them.
 181 */
 182int pci_status_get_and_clear_errors(struct pci_dev *pdev)
 183{
 184	u16 status;
 185	int ret;
 186
 187	ret = pci_read_config_word(pdev, PCI_STATUS, &status);
 188	if (ret != PCIBIOS_SUCCESSFUL)
 189		return -EIO;
 190
 191	status &= PCI_STATUS_ERROR_BITS;
 192	if (status)
 193		pci_write_config_word(pdev, PCI_STATUS, status);
 194
 195	return status;
 196}
 197EXPORT_SYMBOL_GPL(pci_status_get_and_clear_errors);
 198
 199#ifdef CONFIG_HAS_IOMEM
 200void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
 201{
 202	struct resource *res = &pdev->resource[bar];
 203
 204	/*
 205	 * Make sure the BAR is actually a memory resource, not an IO resource
 206	 */
 207	if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
 208		pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
 209		return NULL;
 210	}
 211	return ioremap(res->start, resource_size(res));
 212}
 213EXPORT_SYMBOL_GPL(pci_ioremap_bar);
 214
 215void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
 216{
 217	/*
 218	 * Make sure the BAR is actually a memory resource, not an IO resource
 219	 */
 220	if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
 221		WARN_ON(1);
 222		return NULL;
 223	}
 224	return ioremap_wc(pci_resource_start(pdev, bar),
 225			  pci_resource_len(pdev, bar));
 226}
 227EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
 228#endif
 229
 230/**
 231 * pci_dev_str_match_path - test if a path string matches a device
 232 * @dev: the PCI device to test
 233 * @path: string to match the device against
 234 * @endptr: pointer to the string after the match
 235 *
 236 * Test if a string (typically from a kernel parameter) formatted as a
 237 * path of device/function addresses matches a PCI device. The string must
 238 * be of the form:
 239 *
 240 *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
 241 *
 242 * A path for a device can be obtained using 'lspci -t'.  Using a path
 243 * is more robust against bus renumbering than using only a single bus,
 244 * device and function address.
 245 *
 246 * Returns 1 if the string matches the device, 0 if it does not and
 247 * a negative error code if it fails to parse the string.
 248 */
 249static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
 250				  const char **endptr)
 251{
 252	int ret;
 253	int seg, bus, slot, func;
 254	char *wpath, *p;
 255	char end;
 256
 257	*endptr = strchrnul(path, ';');
 258
 259	wpath = kmemdup_nul(path, *endptr - path, GFP_KERNEL);
 260	if (!wpath)
 261		return -ENOMEM;
 262
 263	while (1) {
 264		p = strrchr(wpath, '/');
 265		if (!p)
 266			break;
 267		ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
 268		if (ret != 2) {
 269			ret = -EINVAL;
 270			goto free_and_exit;
 271		}
 272
 273		if (dev->devfn != PCI_DEVFN(slot, func)) {
 274			ret = 0;
 275			goto free_and_exit;
 276		}
 277
 278		/*
 279		 * Note: we don't need to get a reference to the upstream
 280		 * bridge because we hold a reference to the top level
 281		 * device which should hold a reference to the bridge,
 282		 * and so on.
 283		 */
 284		dev = pci_upstream_bridge(dev);
 285		if (!dev) {
 286			ret = 0;
 287			goto free_and_exit;
 288		}
 289
 290		*p = 0;
 291	}
 292
 293	ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
 294		     &func, &end);
 295	if (ret != 4) {
 296		seg = 0;
 297		ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
 298		if (ret != 3) {
 299			ret = -EINVAL;
 300			goto free_and_exit;
 301		}
 302	}
 303
 304	ret = (seg == pci_domain_nr(dev->bus) &&
 305	       bus == dev->bus->number &&
 306	       dev->devfn == PCI_DEVFN(slot, func));
 307
 308free_and_exit:
 309	kfree(wpath);
 310	return ret;
 311}
 312
 313/**
 314 * pci_dev_str_match - test if a string matches a device
 315 * @dev: the PCI device to test
 316 * @p: string to match the device against
 317 * @endptr: pointer to the string after the match
 318 *
 319 * Test if a string (typically from a kernel parameter) matches a specified
 320 * PCI device. The string may be of one of the following formats:
 321 *
 322 *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
 323 *   pci:<vendor>:<device>[:<subvendor>:<subdevice>]
 324 *
 325 * The first format specifies a PCI bus/device/function address which
 326 * may change if new hardware is inserted, if motherboard firmware changes,
 327 * or due to changes caused in kernel parameters. If the domain is
 328 * left unspecified, it is taken to be 0.  In order to be robust against
 329 * bus renumbering issues, a path of PCI device/function numbers may be used
 330 * to address the specific device.  The path for a device can be determined
 331 * through the use of 'lspci -t'.
 332 *
 333 * The second format matches devices using IDs in the configuration
 334 * space which may match multiple devices in the system. A value of 0
 335 * for any field will match all devices. (Note: this differs from
 336 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
 337 * legacy reasons and convenience so users don't have to specify
 338 * FFFFFFFFs on the command line.)
 339 *
 340 * Returns 1 if the string matches the device, 0 if it does not and
 341 * a negative error code if the string cannot be parsed.
 342 */
 343static int pci_dev_str_match(struct pci_dev *dev, const char *p,
 344			     const char **endptr)
 345{
 346	int ret;
 347	int count;
 348	unsigned short vendor, device, subsystem_vendor, subsystem_device;
 349
 350	if (strncmp(p, "pci:", 4) == 0) {
 351		/* PCI vendor/device (subvendor/subdevice) IDs are specified */
 352		p += 4;
 353		ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
 354			     &subsystem_vendor, &subsystem_device, &count);
 355		if (ret != 4) {
 356			ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
 357			if (ret != 2)
 358				return -EINVAL;
 359
 360			subsystem_vendor = 0;
 361			subsystem_device = 0;
 362		}
 363
 364		p += count;
 365
 366		if ((!vendor || vendor == dev->vendor) &&
 367		    (!device || device == dev->device) &&
 368		    (!subsystem_vendor ||
 369			    subsystem_vendor == dev->subsystem_vendor) &&
 370		    (!subsystem_device ||
 371			    subsystem_device == dev->subsystem_device))
 372			goto found;
 373	} else {
 374		/*
 375		 * PCI Bus, Device, Function IDs are specified
 376		 * (optionally, may include a path of devfns following it)
 377		 */
 378		ret = pci_dev_str_match_path(dev, p, &p);
 379		if (ret < 0)
 380			return ret;
 381		else if (ret)
 382			goto found;
 383	}
 384
 385	*endptr = p;
 386	return 0;
 387
 388found:
 389	*endptr = p;
 390	return 1;
 391}
 392
 393static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
 394				   u8 pos, int cap, int *ttl)
 395{
 396	u8 id;
 397	u16 ent;
 398
 399	pci_bus_read_config_byte(bus, devfn, pos, &pos);
 400
 401	while ((*ttl)--) {
 402		if (pos < 0x40)
 403			break;
 404		pos &= ~3;
 405		pci_bus_read_config_word(bus, devfn, pos, &ent);
 406
 407		id = ent & 0xff;
 408		if (id == 0xff)
 409			break;
 410		if (id == cap)
 411			return pos;
 412		pos = (ent >> 8);
 413	}
 414	return 0;
 415}
 416
 417static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
 418			       u8 pos, int cap)
 419{
 420	int ttl = PCI_FIND_CAP_TTL;
 421
 422	return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
 423}
 424
 425int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
 426{
 427	return __pci_find_next_cap(dev->bus, dev->devfn,
 428				   pos + PCI_CAP_LIST_NEXT, cap);
 429}
 430EXPORT_SYMBOL_GPL(pci_find_next_capability);
 431
 432static int __pci_bus_find_cap_start(struct pci_bus *bus,
 433				    unsigned int devfn, u8 hdr_type)
 434{
 435	u16 status;
 436
 437	pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
 438	if (!(status & PCI_STATUS_CAP_LIST))
 439		return 0;
 440
 441	switch (hdr_type) {
 442	case PCI_HEADER_TYPE_NORMAL:
 443	case PCI_HEADER_TYPE_BRIDGE:
 444		return PCI_CAPABILITY_LIST;
 445	case PCI_HEADER_TYPE_CARDBUS:
 446		return PCI_CB_CAPABILITY_LIST;
 447	}
 448
 449	return 0;
 450}
 451
 452/**
 453 * pci_find_capability - query for devices' capabilities
 454 * @dev: PCI device to query
 455 * @cap: capability code
 456 *
 457 * Tell if a device supports a given PCI capability.
 458 * Returns the address of the requested capability structure within the
 459 * device's PCI configuration space or 0 in case the device does not
 460 * support it.  Possible values for @cap include:
 461 *
 462 *  %PCI_CAP_ID_PM           Power Management
 463 *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
 464 *  %PCI_CAP_ID_VPD          Vital Product Data
 465 *  %PCI_CAP_ID_SLOTID       Slot Identification
 466 *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
 467 *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
 468 *  %PCI_CAP_ID_PCIX         PCI-X
 469 *  %PCI_CAP_ID_EXP          PCI Express
 470 */
 471int pci_find_capability(struct pci_dev *dev, int cap)
 472{
 473	int pos;
 474
 475	pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 476	if (pos)
 477		pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
 478
 479	return pos;
 480}
 481EXPORT_SYMBOL(pci_find_capability);
 482
 483/**
 484 * pci_bus_find_capability - query for devices' capabilities
 485 * @bus: the PCI bus to query
 486 * @devfn: PCI device to query
 487 * @cap: capability code
 488 *
 489 * Like pci_find_capability() but works for PCI devices that do not have a
 490 * pci_dev structure set up yet.
 491 *
 492 * Returns the address of the requested capability structure within the
 493 * device's PCI configuration space or 0 in case the device does not
 494 * support it.
 495 */
 496int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
 497{
 498	int pos;
 499	u8 hdr_type;
 500
 501	pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
 502
 503	pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
 504	if (pos)
 505		pos = __pci_find_next_cap(bus, devfn, pos, cap);
 506
 507	return pos;
 508}
 509EXPORT_SYMBOL(pci_bus_find_capability);
 510
 511/**
 512 * pci_find_next_ext_capability - Find an extended capability
 513 * @dev: PCI device to query
 514 * @start: address at which to start looking (0 to start at beginning of list)
 515 * @cap: capability code
 516 *
 517 * Returns the address of the next matching extended capability structure
 518 * within the device's PCI configuration space or 0 if the device does
 519 * not support it.  Some capabilities can occur several times, e.g., the
 520 * vendor-specific capability, and this provides a way to find them all.
 521 */
 522int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
 523{
 524	u32 header;
 525	int ttl;
 526	int pos = PCI_CFG_SPACE_SIZE;
 527
 528	/* minimum 8 bytes per capability */
 529	ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
 530
 531	if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
 532		return 0;
 533
 534	if (start)
 535		pos = start;
 536
 537	if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 538		return 0;
 539
 540	/*
 541	 * If we have no capabilities, this is indicated by cap ID,
 542	 * cap version and next pointer all being 0.
 543	 */
 544	if (header == 0)
 545		return 0;
 546
 547	while (ttl-- > 0) {
 548		if (PCI_EXT_CAP_ID(header) == cap && pos != start)
 549			return pos;
 550
 551		pos = PCI_EXT_CAP_NEXT(header);
 552		if (pos < PCI_CFG_SPACE_SIZE)
 553			break;
 554
 555		if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 556			break;
 557	}
 558
 559	return 0;
 560}
 561EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
 562
 563/**
 564 * pci_find_ext_capability - Find an extended capability
 565 * @dev: PCI device to query
 566 * @cap: capability code
 567 *
 568 * Returns the address of the requested extended capability structure
 569 * within the device's PCI configuration space or 0 if the device does
 570 * not support it.  Possible values for @cap include:
 571 *
 572 *  %PCI_EXT_CAP_ID_ERR		Advanced Error Reporting
 573 *  %PCI_EXT_CAP_ID_VC		Virtual Channel
 574 *  %PCI_EXT_CAP_ID_DSN		Device Serial Number
 575 *  %PCI_EXT_CAP_ID_PWR		Power Budgeting
 576 */
 577int pci_find_ext_capability(struct pci_dev *dev, int cap)
 578{
 579	return pci_find_next_ext_capability(dev, 0, cap);
 580}
 581EXPORT_SYMBOL_GPL(pci_find_ext_capability);
 582
 583/**
 584 * pci_get_dsn - Read and return the 8-byte Device Serial Number
 585 * @dev: PCI device to query
 586 *
 587 * Looks up the PCI_EXT_CAP_ID_DSN and reads the 8 bytes of the Device Serial
 588 * Number.
 589 *
 590 * Returns the DSN, or zero if the capability does not exist.
 591 */
 592u64 pci_get_dsn(struct pci_dev *dev)
 593{
 594	u32 dword;
 595	u64 dsn;
 596	int pos;
 597
 598	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DSN);
 599	if (!pos)
 600		return 0;
 601
 602	/*
 603	 * The Device Serial Number is two dwords offset 4 bytes from the
 604	 * capability position. The specification says that the first dword is
 605	 * the lower half, and the second dword is the upper half.
 606	 */
 607	pos += 4;
 608	pci_read_config_dword(dev, pos, &dword);
 609	dsn = (u64)dword;
 610	pci_read_config_dword(dev, pos + 4, &dword);
 611	dsn |= ((u64)dword) << 32;
 612
 613	return dsn;
 614}
 615EXPORT_SYMBOL_GPL(pci_get_dsn);
 616
 617static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
 618{
 619	int rc, ttl = PCI_FIND_CAP_TTL;
 620	u8 cap, mask;
 621
 622	if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
 623		mask = HT_3BIT_CAP_MASK;
 624	else
 625		mask = HT_5BIT_CAP_MASK;
 626
 627	pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
 628				      PCI_CAP_ID_HT, &ttl);
 629	while (pos) {
 630		rc = pci_read_config_byte(dev, pos + 3, &cap);
 631		if (rc != PCIBIOS_SUCCESSFUL)
 632			return 0;
 633
 634		if ((cap & mask) == ht_cap)
 635			return pos;
 636
 637		pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
 638					      pos + PCI_CAP_LIST_NEXT,
 639					      PCI_CAP_ID_HT, &ttl);
 640	}
 641
 642	return 0;
 643}
 644/**
 645 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
 646 * @dev: PCI device to query
 647 * @pos: Position from which to continue searching
 648 * @ht_cap: Hypertransport capability code
 649 *
 650 * To be used in conjunction with pci_find_ht_capability() to search for
 651 * all capabilities matching @ht_cap. @pos should always be a value returned
 652 * from pci_find_ht_capability().
 653 *
 654 * NB. To be 100% safe against broken PCI devices, the caller should take
 655 * steps to avoid an infinite loop.
 656 */
 657int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
 658{
 659	return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
 660}
 661EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
 662
 663/**
 664 * pci_find_ht_capability - query a device's Hypertransport capabilities
 665 * @dev: PCI device to query
 666 * @ht_cap: Hypertransport capability code
 667 *
 668 * Tell if a device supports a given Hypertransport capability.
 669 * Returns an address within the device's PCI configuration space
 670 * or 0 in case the device does not support the request capability.
 671 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
 672 * which has a Hypertransport capability matching @ht_cap.
 673 */
 674int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
 675{
 676	int pos;
 677
 678	pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 679	if (pos)
 680		pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
 681
 682	return pos;
 683}
 684EXPORT_SYMBOL_GPL(pci_find_ht_capability);
 685
 686/**
 687 * pci_find_parent_resource - return resource region of parent bus of given
 688 *			      region
 689 * @dev: PCI device structure contains resources to be searched
 690 * @res: child resource record for which parent is sought
 691 *
 692 * For given resource region of given device, return the resource region of
 693 * parent bus the given region is contained in.
 694 */
 695struct resource *pci_find_parent_resource(const struct pci_dev *dev,
 696					  struct resource *res)
 697{
 698	const struct pci_bus *bus = dev->bus;
 699	struct resource *r;
 700	int i;
 701
 702	pci_bus_for_each_resource(bus, r, i) {
 703		if (!r)
 704			continue;
 705		if (resource_contains(r, res)) {
 706
 707			/*
 708			 * If the window is prefetchable but the BAR is
 709			 * not, the allocator made a mistake.
 710			 */
 711			if (r->flags & IORESOURCE_PREFETCH &&
 712			    !(res->flags & IORESOURCE_PREFETCH))
 713				return NULL;
 714
 715			/*
 716			 * If we're below a transparent bridge, there may
 717			 * be both a positively-decoded aperture and a
 718			 * subtractively-decoded region that contain the BAR.
 719			 * We want the positively-decoded one, so this depends
 720			 * on pci_bus_for_each_resource() giving us those
 721			 * first.
 722			 */
 723			return r;
 724		}
 725	}
 726	return NULL;
 727}
 728EXPORT_SYMBOL(pci_find_parent_resource);
 729
 730/**
 731 * pci_find_resource - Return matching PCI device resource
 732 * @dev: PCI device to query
 733 * @res: Resource to look for
 734 *
 735 * Goes over standard PCI resources (BARs) and checks if the given resource
 736 * is partially or fully contained in any of them. In that case the
 737 * matching resource is returned, %NULL otherwise.
 738 */
 739struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
 740{
 741	int i;
 742
 743	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
 744		struct resource *r = &dev->resource[i];
 745
 746		if (r->start && resource_contains(r, res))
 747			return r;
 748	}
 749
 750	return NULL;
 751}
 752EXPORT_SYMBOL(pci_find_resource);
 753
 754/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 755 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
 756 * @dev: the PCI device to operate on
 757 * @pos: config space offset of status word
 758 * @mask: mask of bit(s) to care about in status word
 759 *
 760 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
 761 */
 762int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
 763{
 764	int i;
 765
 766	/* Wait for Transaction Pending bit clean */
 767	for (i = 0; i < 4; i++) {
 768		u16 status;
 769		if (i)
 770			msleep((1 << (i - 1)) * 100);
 771
 772		pci_read_config_word(dev, pos, &status);
 773		if (!(status & mask))
 774			return 1;
 775	}
 776
 777	return 0;
 778}
 779
 780static int pci_acs_enable;
 781
 782/**
 783 * pci_request_acs - ask for ACS to be enabled if supported
 784 */
 785void pci_request_acs(void)
 786{
 787	pci_acs_enable = 1;
 788}
 789
 790static const char *disable_acs_redir_param;
 791
 792/**
 793 * pci_disable_acs_redir - disable ACS redirect capabilities
 794 * @dev: the PCI device
 795 *
 796 * For only devices specified in the disable_acs_redir parameter.
 797 */
 798static void pci_disable_acs_redir(struct pci_dev *dev)
 799{
 800	int ret = 0;
 801	const char *p;
 802	int pos;
 803	u16 ctrl;
 804
 805	if (!disable_acs_redir_param)
 806		return;
 807
 808	p = disable_acs_redir_param;
 809	while (*p) {
 810		ret = pci_dev_str_match(dev, p, &p);
 811		if (ret < 0) {
 812			pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
 813				     disable_acs_redir_param);
 814
 815			break;
 816		} else if (ret == 1) {
 817			/* Found a match */
 818			break;
 819		}
 820
 821		if (*p != ';' && *p != ',') {
 822			/* End of param or invalid format */
 823			break;
 824		}
 825		p++;
 826	}
 827
 828	if (ret != 1)
 829		return;
 830
 831	if (!pci_dev_specific_disable_acs_redir(dev))
 832		return;
 833
 834	pos = dev->acs_cap;
 835	if (!pos) {
 836		pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
 837		return;
 838	}
 839
 840	pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
 841
 842	/* P2P Request & Completion Redirect */
 843	ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
 844
 845	pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
 846
 847	pci_info(dev, "disabled ACS redirect\n");
 848}
 849
 850/**
 851 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
 852 * @dev: the PCI device
 853 */
 854static void pci_std_enable_acs(struct pci_dev *dev)
 855{
 856	int pos;
 857	u16 cap;
 858	u16 ctrl;
 859
 860	pos = dev->acs_cap;
 861	if (!pos)
 862		return;
 863
 864	pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
 865	pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
 866
 867	/* Source Validation */
 868	ctrl |= (cap & PCI_ACS_SV);
 869
 870	/* P2P Request Redirect */
 871	ctrl |= (cap & PCI_ACS_RR);
 872
 873	/* P2P Completion Redirect */
 874	ctrl |= (cap & PCI_ACS_CR);
 875
 876	/* Upstream Forwarding */
 877	ctrl |= (cap & PCI_ACS_UF);
 878
 879	pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
 880}
 881
 882/**
 883 * pci_enable_acs - enable ACS if hardware support it
 884 * @dev: the PCI device
 885 */
 886static void pci_enable_acs(struct pci_dev *dev)
 887{
 888	if (!pci_acs_enable)
 889		goto disable_acs_redir;
 890
 891	if (!pci_dev_specific_enable_acs(dev))
 892		goto disable_acs_redir;
 893
 894	pci_std_enable_acs(dev);
 895
 896disable_acs_redir:
 897	/*
 898	 * Note: pci_disable_acs_redir() must be called even if ACS was not
 899	 * enabled by the kernel because it may have been enabled by
 900	 * platform firmware.  So if we are told to disable it, we should
 901	 * always disable it after setting the kernel's default
 902	 * preferences.
 903	 */
 904	pci_disable_acs_redir(dev);
 905}
 906
 907/**
 908 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
 909 * @dev: PCI device to have its BARs restored
 910 *
 911 * Restore the BAR values for a given device, so as to make it
 912 * accessible by its driver.
 913 */
 914static void pci_restore_bars(struct pci_dev *dev)
 915{
 916	int i;
 917
 918	for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
 919		pci_update_resource(dev, i);
 920}
 921
 922static const struct pci_platform_pm_ops *pci_platform_pm;
 923
 924int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
 925{
 926	if (!ops->is_manageable || !ops->set_state  || !ops->get_state ||
 927	    !ops->choose_state  || !ops->set_wakeup || !ops->need_resume)
 928		return -EINVAL;
 929	pci_platform_pm = ops;
 930	return 0;
 931}
 932
 933static inline bool platform_pci_power_manageable(struct pci_dev *dev)
 934{
 935	return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
 936}
 937
 938static inline int platform_pci_set_power_state(struct pci_dev *dev,
 939					       pci_power_t t)
 940{
 941	return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
 942}
 943
 944static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
 945{
 946	return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
 947}
 948
 949static inline void platform_pci_refresh_power_state(struct pci_dev *dev)
 950{
 951	if (pci_platform_pm && pci_platform_pm->refresh_state)
 952		pci_platform_pm->refresh_state(dev);
 953}
 954
 955static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
 956{
 957	return pci_platform_pm ?
 958			pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
 959}
 960
 961static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
 962{
 963	return pci_platform_pm ?
 964			pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
 965}
 966
 967static inline bool platform_pci_need_resume(struct pci_dev *dev)
 968{
 969	return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
 970}
 971
 972static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
 973{
 974	if (pci_platform_pm && pci_platform_pm->bridge_d3)
 975		return pci_platform_pm->bridge_d3(dev);
 976	return false;
 977}
 978
 979/**
 980 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
 981 *			     given PCI device
 982 * @dev: PCI device to handle.
 983 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
 984 *
 985 * RETURN VALUE:
 986 * -EINVAL if the requested state is invalid.
 987 * -EIO if device does not support PCI PM or its PM capabilities register has a
 988 * wrong version, or device doesn't support the requested state.
 989 * 0 if device already is in the requested state.
 990 * 0 if device's power state has been successfully changed.
 991 */
 992static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
 993{
 994	u16 pmcsr;
 995	bool need_restore = false;
 996
 997	/* Check if we're already there */
 998	if (dev->current_state == state)
 999		return 0;
1000
1001	if (!dev->pm_cap)
1002		return -EIO;
1003
1004	if (state < PCI_D0 || state > PCI_D3hot)
1005		return -EINVAL;
1006
1007	/*
1008	 * Validate transition: We can enter D0 from any state, but if
1009	 * we're already in a low-power state, we can only go deeper.  E.g.,
1010	 * we can go from D1 to D3, but we can't go directly from D3 to D1;
1011	 * we'd have to go from D3 to D0, then to D1.
1012	 */
1013	if (state != PCI_D0 && dev->current_state <= PCI_D3cold
1014	    && dev->current_state > state) {
1015		pci_err(dev, "invalid power transition (from %s to %s)\n",
1016			pci_power_name(dev->current_state),
1017			pci_power_name(state));
1018		return -EINVAL;
1019	}
1020
1021	/* Check if this device supports the desired state */
1022	if ((state == PCI_D1 && !dev->d1_support)
1023	   || (state == PCI_D2 && !dev->d2_support))
1024		return -EIO;
1025
1026	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1027	if (pmcsr == (u16) ~0) {
1028		pci_err(dev, "can't change power state from %s to %s (config space inaccessible)\n",
1029			pci_power_name(dev->current_state),
1030			pci_power_name(state));
1031		return -EIO;
1032	}
1033
1034	/*
1035	 * If we're (effectively) in D3, force entire word to 0.
1036	 * This doesn't affect PME_Status, disables PME_En, and
1037	 * sets PowerState to 0.
1038	 */
1039	switch (dev->current_state) {
1040	case PCI_D0:
1041	case PCI_D1:
1042	case PCI_D2:
1043		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1044		pmcsr |= state;
1045		break;
1046	case PCI_D3hot:
1047	case PCI_D3cold:
1048	case PCI_UNKNOWN: /* Boot-up */
1049		if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
1050		 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
1051			need_restore = true;
1052		fallthrough;	/* force to D0 */
1053	default:
1054		pmcsr = 0;
1055		break;
1056	}
1057
1058	/* Enter specified state */
1059	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1060
1061	/*
1062	 * Mandatory power management transition delays; see PCI PM 1.1
1063	 * 5.6.1 table 18
1064	 */
1065	if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
1066		pci_dev_d3_sleep(dev);
1067	else if (state == PCI_D2 || dev->current_state == PCI_D2)
1068		msleep(PCI_PM_D2_DELAY);
1069
1070	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1071	dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1072	if (dev->current_state != state)
1073		pci_info_ratelimited(dev, "refused to change power state from %s to %s\n",
1074			 pci_power_name(dev->current_state),
1075			 pci_power_name(state));
1076
1077	/*
1078	 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
1079	 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
1080	 * from D3hot to D0 _may_ perform an internal reset, thereby
1081	 * going to "D0 Uninitialized" rather than "D0 Initialized".
1082	 * For example, at least some versions of the 3c905B and the
1083	 * 3c556B exhibit this behaviour.
1084	 *
1085	 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
1086	 * devices in a D3hot state at boot.  Consequently, we need to
1087	 * restore at least the BARs so that the device will be
1088	 * accessible to its driver.
1089	 */
1090	if (need_restore)
1091		pci_restore_bars(dev);
1092
1093	if (dev->bus->self)
1094		pcie_aspm_pm_state_change(dev->bus->self);
1095
1096	return 0;
1097}
1098
1099/**
1100 * pci_update_current_state - Read power state of given device and cache it
1101 * @dev: PCI device to handle.
1102 * @state: State to cache in case the device doesn't have the PM capability
1103 *
1104 * The power state is read from the PMCSR register, which however is
1105 * inaccessible in D3cold.  The platform firmware is therefore queried first
1106 * to detect accessibility of the register.  In case the platform firmware
1107 * reports an incorrect state or the device isn't power manageable by the
1108 * platform at all, we try to detect D3cold by testing accessibility of the
1109 * vendor ID in config space.
1110 */
1111void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
1112{
1113	if (platform_pci_get_power_state(dev) == PCI_D3cold ||
1114	    !pci_device_is_present(dev)) {
1115		dev->current_state = PCI_D3cold;
1116	} else if (dev->pm_cap) {
1117		u16 pmcsr;
1118
1119		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1120		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1121	} else {
1122		dev->current_state = state;
1123	}
1124}
1125
1126/**
1127 * pci_refresh_power_state - Refresh the given device's power state data
1128 * @dev: Target PCI device.
1129 *
1130 * Ask the platform to refresh the devices power state information and invoke
1131 * pci_update_current_state() to update its current PCI power state.
1132 */
1133void pci_refresh_power_state(struct pci_dev *dev)
1134{
1135	if (platform_pci_power_manageable(dev))
1136		platform_pci_refresh_power_state(dev);
1137
1138	pci_update_current_state(dev, dev->current_state);
1139}
1140
1141/**
1142 * pci_platform_power_transition - Use platform to change device power state
1143 * @dev: PCI device to handle.
1144 * @state: State to put the device into.
1145 */
1146int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
1147{
1148	int error;
1149
1150	if (platform_pci_power_manageable(dev)) {
1151		error = platform_pci_set_power_state(dev, state);
1152		if (!error)
1153			pci_update_current_state(dev, state);
1154	} else
1155		error = -ENODEV;
1156
1157	if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
1158		dev->current_state = PCI_D0;
1159
1160	return error;
1161}
1162EXPORT_SYMBOL_GPL(pci_platform_power_transition);
1163
1164/**
1165 * pci_wakeup - Wake up a PCI device
1166 * @pci_dev: Device to handle.
1167 * @ign: ignored parameter
1168 */
1169static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
1170{
1171	pci_wakeup_event(pci_dev);
1172	pm_request_resume(&pci_dev->dev);
1173	return 0;
1174}
1175
1176/**
1177 * pci_wakeup_bus - Walk given bus and wake up devices on it
1178 * @bus: Top bus of the subtree to walk.
1179 */
1180void pci_wakeup_bus(struct pci_bus *bus)
1181{
1182	if (bus)
1183		pci_walk_bus(bus, pci_wakeup, NULL);
1184}
1185
1186static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
1187{
1188	int delay = 1;
1189	u32 id;
1190
1191	/*
1192	 * After reset, the device should not silently discard config
1193	 * requests, but it may still indicate that it needs more time by
1194	 * responding to them with CRS completions.  The Root Port will
1195	 * generally synthesize ~0 data to complete the read (except when
1196	 * CRS SV is enabled and the read was for the Vendor ID; in that
1197	 * case it synthesizes 0x0001 data).
1198	 *
1199	 * Wait for the device to return a non-CRS completion.  Read the
1200	 * Command register instead of Vendor ID so we don't have to
1201	 * contend with the CRS SV value.
1202	 */
1203	pci_read_config_dword(dev, PCI_COMMAND, &id);
1204	while (id == ~0) {
1205		if (delay > timeout) {
1206			pci_warn(dev, "not ready %dms after %s; giving up\n",
1207				 delay - 1, reset_type);
1208			return -ENOTTY;
1209		}
1210
1211		if (delay > 1000)
1212			pci_info(dev, "not ready %dms after %s; waiting\n",
1213				 delay - 1, reset_type);
1214
1215		msleep(delay);
1216		delay *= 2;
1217		pci_read_config_dword(dev, PCI_COMMAND, &id);
1218	}
1219
1220	if (delay > 1000)
1221		pci_info(dev, "ready %dms after %s\n", delay - 1,
1222			 reset_type);
1223
1224	return 0;
1225}
1226
1227/**
1228 * pci_power_up - Put the given device into D0
1229 * @dev: PCI device to power up
 
1230 */
1231int pci_power_up(struct pci_dev *dev)
1232{
1233	pci_platform_power_transition(dev, PCI_D0);
1234
1235	/*
1236	 * Mandatory power management transition delays are handled in
1237	 * pci_pm_resume_noirq() and pci_pm_runtime_resume() of the
1238	 * corresponding bridge.
1239	 */
1240	if (dev->runtime_d3cold) {
1241		/*
1242		 * When powering on a bridge from D3cold, the whole hierarchy
1243		 * may be powered on into D0uninitialized state, resume them to
1244		 * give them a chance to suspend again
 
 
1245		 */
1246		pci_wakeup_bus(dev->subordinate);
 
 
 
 
 
 
 
 
 
 
1247	}
1248
1249	return pci_raw_set_power_state(dev, PCI_D0);
1250}
1251
1252/**
1253 * __pci_dev_set_current_state - Set current state of a PCI device
1254 * @dev: Device to handle
1255 * @data: pointer to state to be set
1256 */
1257static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1258{
1259	pci_power_t state = *(pci_power_t *)data;
1260
1261	dev->current_state = state;
1262	return 0;
1263}
1264
1265/**
1266 * pci_bus_set_current_state - Walk given bus and set current state of devices
1267 * @bus: Top bus of the subtree to walk.
1268 * @state: state to be set
1269 */
1270void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1271{
1272	if (bus)
1273		pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1274}
1275
1276/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1277 * pci_set_power_state - Set the power state of a PCI device
1278 * @dev: PCI device to handle.
1279 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1280 *
1281 * Transition a device to a new power state, using the platform firmware and/or
1282 * the device's PCI PM registers.
1283 *
1284 * RETURN VALUE:
1285 * -EINVAL if the requested state is invalid.
1286 * -EIO if device does not support PCI PM or its PM capabilities register has a
1287 * wrong version, or device doesn't support the requested state.
1288 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1289 * 0 if device already is in the requested state.
1290 * 0 if the transition is to D3 but D3 is not supported.
1291 * 0 if device's power state has been successfully changed.
1292 */
1293int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1294{
1295	int error;
1296
1297	/* Bound the state we're entering */
1298	if (state > PCI_D3cold)
1299		state = PCI_D3cold;
1300	else if (state < PCI_D0)
1301		state = PCI_D0;
1302	else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1303
1304		/*
1305		 * If the device or the parent bridge do not support PCI
1306		 * PM, ignore the request if we're doing anything other
1307		 * than putting it into D0 (which would only happen on
1308		 * boot).
1309		 */
1310		return 0;
1311
1312	/* Check if we're already there */
1313	if (dev->current_state == state)
1314		return 0;
1315
1316	if (state == PCI_D0)
1317		return pci_power_up(dev);
1318
1319	/*
1320	 * This device is quirked not to be put into D3, so don't put it in
1321	 * D3
1322	 */
1323	if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1324		return 0;
1325
1326	/*
1327	 * To put device in D3cold, we put device into D3hot in native
1328	 * way, then put device into D3cold with platform ops
1329	 */
1330	error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1331					PCI_D3hot : state);
1332
1333	if (pci_platform_power_transition(dev, state))
1334		return error;
1335
1336	/* Powering off a bridge may power off the whole hierarchy */
1337	if (state == PCI_D3cold)
1338		pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1339
1340	return 0;
1341}
1342EXPORT_SYMBOL(pci_set_power_state);
1343
1344/**
 
 
 
 
 
 
 
 
 
 
 
1345 * pci_choose_state - Choose the power state of a PCI device
1346 * @dev: PCI device to be suspended
1347 * @state: target sleep state for the whole system. This is the value
1348 *	   that is passed to suspend() function.
1349 *
1350 * Returns PCI power state suitable for given device and given system
1351 * message.
1352 */
1353pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
1354{
1355	pci_power_t ret;
1356
1357	if (!dev->pm_cap)
1358		return PCI_D0;
1359
1360	ret = platform_pci_choose_state(dev);
1361	if (ret != PCI_POWER_ERROR)
1362		return ret;
1363
1364	switch (state.event) {
1365	case PM_EVENT_ON:
1366		return PCI_D0;
1367	case PM_EVENT_FREEZE:
1368	case PM_EVENT_PRETHAW:
1369		/* REVISIT both freeze and pre-thaw "should" use D0 */
1370	case PM_EVENT_SUSPEND:
1371	case PM_EVENT_HIBERNATE:
1372		return PCI_D3hot;
1373	default:
1374		pci_info(dev, "unrecognized suspend event %d\n",
1375			 state.event);
1376		BUG();
1377	}
1378	return PCI_D0;
1379}
1380EXPORT_SYMBOL(pci_choose_state);
1381
1382#define PCI_EXP_SAVE_REGS	7
1383
1384static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1385						       u16 cap, bool extended)
1386{
1387	struct pci_cap_saved_state *tmp;
1388
1389	hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1390		if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1391			return tmp;
1392	}
1393	return NULL;
1394}
1395
1396struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1397{
1398	return _pci_find_saved_cap(dev, cap, false);
1399}
1400
1401struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1402{
1403	return _pci_find_saved_cap(dev, cap, true);
1404}
1405
1406static int pci_save_pcie_state(struct pci_dev *dev)
1407{
1408	int i = 0;
1409	struct pci_cap_saved_state *save_state;
1410	u16 *cap;
1411
1412	if (!pci_is_pcie(dev))
1413		return 0;
1414
1415	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1416	if (!save_state) {
1417		pci_err(dev, "buffer not found in %s\n", __func__);
1418		return -ENOMEM;
1419	}
1420
1421	cap = (u16 *)&save_state->cap.data[0];
1422	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1423	pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1424	pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1425	pcie_capability_read_word(dev, PCI_EXP_RTCTL,  &cap[i++]);
1426	pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1427	pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1428	pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1429
1430	return 0;
1431}
1432
1433static void pci_restore_pcie_state(struct pci_dev *dev)
1434{
1435	int i = 0;
1436	struct pci_cap_saved_state *save_state;
1437	u16 *cap;
1438
1439	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1440	if (!save_state)
1441		return;
1442
1443	cap = (u16 *)&save_state->cap.data[0];
1444	pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1445	pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1446	pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1447	pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1448	pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1449	pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1450	pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1451}
1452
1453static int pci_save_pcix_state(struct pci_dev *dev)
1454{
1455	int pos;
1456	struct pci_cap_saved_state *save_state;
1457
1458	pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1459	if (!pos)
1460		return 0;
1461
1462	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1463	if (!save_state) {
1464		pci_err(dev, "buffer not found in %s\n", __func__);
1465		return -ENOMEM;
1466	}
1467
1468	pci_read_config_word(dev, pos + PCI_X_CMD,
1469			     (u16 *)save_state->cap.data);
1470
1471	return 0;
1472}
1473
1474static void pci_restore_pcix_state(struct pci_dev *dev)
1475{
1476	int i = 0, pos;
1477	struct pci_cap_saved_state *save_state;
1478	u16 *cap;
1479
1480	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1481	pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1482	if (!save_state || !pos)
1483		return;
1484	cap = (u16 *)&save_state->cap.data[0];
1485
1486	pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1487}
1488
1489static void pci_save_ltr_state(struct pci_dev *dev)
1490{
1491	int ltr;
1492	struct pci_cap_saved_state *save_state;
1493	u16 *cap;
1494
1495	if (!pci_is_pcie(dev))
1496		return;
1497
1498	ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1499	if (!ltr)
1500		return;
1501
1502	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1503	if (!save_state) {
1504		pci_err(dev, "no suspend buffer for LTR; ASPM issues possible after resume\n");
1505		return;
1506	}
1507
1508	cap = (u16 *)&save_state->cap.data[0];
1509	pci_read_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, cap++);
1510	pci_read_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, cap++);
1511}
1512
1513static void pci_restore_ltr_state(struct pci_dev *dev)
1514{
1515	struct pci_cap_saved_state *save_state;
1516	int ltr;
1517	u16 *cap;
1518
1519	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1520	ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1521	if (!save_state || !ltr)
1522		return;
1523
1524	cap = (u16 *)&save_state->cap.data[0];
1525	pci_write_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, *cap++);
1526	pci_write_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, *cap++);
1527}
1528
1529/**
1530 * pci_save_state - save the PCI configuration space of a device before
1531 *		    suspending
1532 * @dev: PCI device that we're dealing with
1533 */
1534int pci_save_state(struct pci_dev *dev)
1535{
1536	int i;
1537	/* XXX: 100% dword access ok here? */
1538	for (i = 0; i < 16; i++) {
1539		pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1540		pci_dbg(dev, "saving config space at offset %#x (reading %#x)\n",
1541			i * 4, dev->saved_config_space[i]);
1542	}
1543	dev->state_saved = true;
1544
1545	i = pci_save_pcie_state(dev);
1546	if (i != 0)
1547		return i;
1548
1549	i = pci_save_pcix_state(dev);
1550	if (i != 0)
1551		return i;
1552
1553	pci_save_ltr_state(dev);
1554	pci_save_dpc_state(dev);
1555	pci_save_aer_state(dev);
1556	return pci_save_vc_state(dev);
1557}
1558EXPORT_SYMBOL(pci_save_state);
1559
1560static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1561				     u32 saved_val, int retry, bool force)
1562{
1563	u32 val;
1564
1565	pci_read_config_dword(pdev, offset, &val);
1566	if (!force && val == saved_val)
1567		return;
1568
1569	for (;;) {
1570		pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1571			offset, val, saved_val);
1572		pci_write_config_dword(pdev, offset, saved_val);
1573		if (retry-- <= 0)
1574			return;
1575
1576		pci_read_config_dword(pdev, offset, &val);
1577		if (val == saved_val)
1578			return;
1579
1580		mdelay(1);
1581	}
1582}
1583
1584static void pci_restore_config_space_range(struct pci_dev *pdev,
1585					   int start, int end, int retry,
1586					   bool force)
1587{
1588	int index;
1589
1590	for (index = end; index >= start; index--)
1591		pci_restore_config_dword(pdev, 4 * index,
1592					 pdev->saved_config_space[index],
1593					 retry, force);
1594}
1595
1596static void pci_restore_config_space(struct pci_dev *pdev)
1597{
1598	if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1599		pci_restore_config_space_range(pdev, 10, 15, 0, false);
1600		/* Restore BARs before the command register. */
1601		pci_restore_config_space_range(pdev, 4, 9, 10, false);
1602		pci_restore_config_space_range(pdev, 0, 3, 0, false);
1603	} else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1604		pci_restore_config_space_range(pdev, 12, 15, 0, false);
1605
1606		/*
1607		 * Force rewriting of prefetch registers to avoid S3 resume
1608		 * issues on Intel PCI bridges that occur when these
1609		 * registers are not explicitly written.
1610		 */
1611		pci_restore_config_space_range(pdev, 9, 11, 0, true);
1612		pci_restore_config_space_range(pdev, 0, 8, 0, false);
1613	} else {
1614		pci_restore_config_space_range(pdev, 0, 15, 0, false);
1615	}
1616}
1617
1618static void pci_restore_rebar_state(struct pci_dev *pdev)
1619{
1620	unsigned int pos, nbars, i;
1621	u32 ctrl;
1622
1623	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1624	if (!pos)
1625		return;
1626
1627	pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1628	nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1629		    PCI_REBAR_CTRL_NBAR_SHIFT;
1630
1631	for (i = 0; i < nbars; i++, pos += 8) {
1632		struct resource *res;
1633		int bar_idx, size;
1634
1635		pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1636		bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1637		res = pdev->resource + bar_idx;
1638		size = ilog2(resource_size(res)) - 20;
1639		ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1640		ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1641		pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1642	}
1643}
1644
1645/**
1646 * pci_restore_state - Restore the saved state of a PCI device
1647 * @dev: PCI device that we're dealing with
1648 */
1649void pci_restore_state(struct pci_dev *dev)
1650{
1651	if (!dev->state_saved)
1652		return;
1653
1654	/*
1655	 * Restore max latencies (in the LTR capability) before enabling
1656	 * LTR itself (in the PCIe capability).
1657	 */
1658	pci_restore_ltr_state(dev);
1659
1660	pci_restore_pcie_state(dev);
1661	pci_restore_pasid_state(dev);
1662	pci_restore_pri_state(dev);
1663	pci_restore_ats_state(dev);
1664	pci_restore_vc_state(dev);
1665	pci_restore_rebar_state(dev);
1666	pci_restore_dpc_state(dev);
1667
1668	pci_aer_clear_status(dev);
1669	pci_restore_aer_state(dev);
1670
1671	pci_restore_config_space(dev);
1672
1673	pci_restore_pcix_state(dev);
1674	pci_restore_msi_state(dev);
1675
1676	/* Restore ACS and IOV configuration state */
1677	pci_enable_acs(dev);
1678	pci_restore_iov_state(dev);
1679
1680	dev->state_saved = false;
1681}
1682EXPORT_SYMBOL(pci_restore_state);
1683
1684struct pci_saved_state {
1685	u32 config_space[16];
1686	struct pci_cap_saved_data cap[];
1687};
1688
1689/**
1690 * pci_store_saved_state - Allocate and return an opaque struct containing
1691 *			   the device saved state.
1692 * @dev: PCI device that we're dealing with
1693 *
1694 * Return NULL if no state or error.
1695 */
1696struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1697{
1698	struct pci_saved_state *state;
1699	struct pci_cap_saved_state *tmp;
1700	struct pci_cap_saved_data *cap;
1701	size_t size;
1702
1703	if (!dev->state_saved)
1704		return NULL;
1705
1706	size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1707
1708	hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1709		size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1710
1711	state = kzalloc(size, GFP_KERNEL);
1712	if (!state)
1713		return NULL;
1714
1715	memcpy(state->config_space, dev->saved_config_space,
1716	       sizeof(state->config_space));
1717
1718	cap = state->cap;
1719	hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1720		size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1721		memcpy(cap, &tmp->cap, len);
1722		cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1723	}
1724	/* Empty cap_save terminates list */
1725
1726	return state;
1727}
1728EXPORT_SYMBOL_GPL(pci_store_saved_state);
1729
1730/**
1731 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1732 * @dev: PCI device that we're dealing with
1733 * @state: Saved state returned from pci_store_saved_state()
1734 */
1735int pci_load_saved_state(struct pci_dev *dev,
1736			 struct pci_saved_state *state)
1737{
1738	struct pci_cap_saved_data *cap;
1739
1740	dev->state_saved = false;
1741
1742	if (!state)
1743		return 0;
1744
1745	memcpy(dev->saved_config_space, state->config_space,
1746	       sizeof(state->config_space));
1747
1748	cap = state->cap;
1749	while (cap->size) {
1750		struct pci_cap_saved_state *tmp;
1751
1752		tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1753		if (!tmp || tmp->cap.size != cap->size)
1754			return -EINVAL;
1755
1756		memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1757		cap = (struct pci_cap_saved_data *)((u8 *)cap +
1758		       sizeof(struct pci_cap_saved_data) + cap->size);
1759	}
1760
1761	dev->state_saved = true;
1762	return 0;
1763}
1764EXPORT_SYMBOL_GPL(pci_load_saved_state);
1765
1766/**
1767 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1768 *				   and free the memory allocated for it.
1769 * @dev: PCI device that we're dealing with
1770 * @state: Pointer to saved state returned from pci_store_saved_state()
1771 */
1772int pci_load_and_free_saved_state(struct pci_dev *dev,
1773				  struct pci_saved_state **state)
1774{
1775	int ret = pci_load_saved_state(dev, *state);
1776	kfree(*state);
1777	*state = NULL;
1778	return ret;
1779}
1780EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1781
1782int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1783{
1784	return pci_enable_resources(dev, bars);
1785}
1786
1787static int do_pci_enable_device(struct pci_dev *dev, int bars)
1788{
1789	int err;
1790	struct pci_dev *bridge;
1791	u16 cmd;
1792	u8 pin;
1793
1794	err = pci_set_power_state(dev, PCI_D0);
1795	if (err < 0 && err != -EIO)
1796		return err;
1797
1798	bridge = pci_upstream_bridge(dev);
1799	if (bridge)
1800		pcie_aspm_powersave_config_link(bridge);
1801
1802	err = pcibios_enable_device(dev, bars);
1803	if (err < 0)
1804		return err;
1805	pci_fixup_device(pci_fixup_enable, dev);
1806
1807	if (dev->msi_enabled || dev->msix_enabled)
1808		return 0;
1809
1810	pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1811	if (pin) {
1812		pci_read_config_word(dev, PCI_COMMAND, &cmd);
1813		if (cmd & PCI_COMMAND_INTX_DISABLE)
1814			pci_write_config_word(dev, PCI_COMMAND,
1815					      cmd & ~PCI_COMMAND_INTX_DISABLE);
1816	}
1817
1818	return 0;
1819}
1820
1821/**
1822 * pci_reenable_device - Resume abandoned device
1823 * @dev: PCI device to be resumed
1824 *
1825 * NOTE: This function is a backend of pci_default_resume() and is not supposed
1826 * to be called by normal code, write proper resume handler and use it instead.
1827 */
1828int pci_reenable_device(struct pci_dev *dev)
1829{
1830	if (pci_is_enabled(dev))
1831		return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1832	return 0;
1833}
1834EXPORT_SYMBOL(pci_reenable_device);
1835
1836static void pci_enable_bridge(struct pci_dev *dev)
1837{
1838	struct pci_dev *bridge;
1839	int retval;
1840
1841	bridge = pci_upstream_bridge(dev);
1842	if (bridge)
1843		pci_enable_bridge(bridge);
1844
1845	if (pci_is_enabled(dev)) {
1846		if (!dev->is_busmaster)
1847			pci_set_master(dev);
1848		return;
1849	}
1850
1851	retval = pci_enable_device(dev);
1852	if (retval)
1853		pci_err(dev, "Error enabling bridge (%d), continuing\n",
1854			retval);
1855	pci_set_master(dev);
1856}
1857
1858static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1859{
1860	struct pci_dev *bridge;
1861	int err;
1862	int i, bars = 0;
1863
1864	/*
1865	 * Power state could be unknown at this point, either due to a fresh
1866	 * boot or a device removal call.  So get the current power state
1867	 * so that things like MSI message writing will behave as expected
1868	 * (e.g. if the device really is in D0 at enable time).
1869	 */
1870	if (dev->pm_cap) {
1871		u16 pmcsr;
1872		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1873		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1874	}
1875
1876	if (atomic_inc_return(&dev->enable_cnt) > 1)
1877		return 0;		/* already enabled */
1878
1879	bridge = pci_upstream_bridge(dev);
1880	if (bridge)
1881		pci_enable_bridge(bridge);
1882
1883	/* only skip sriov related */
1884	for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1885		if (dev->resource[i].flags & flags)
1886			bars |= (1 << i);
1887	for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1888		if (dev->resource[i].flags & flags)
1889			bars |= (1 << i);
1890
1891	err = do_pci_enable_device(dev, bars);
1892	if (err < 0)
1893		atomic_dec(&dev->enable_cnt);
1894	return err;
1895}
1896
1897/**
1898 * pci_enable_device_io - Initialize a device for use with IO space
1899 * @dev: PCI device to be initialized
1900 *
1901 * Initialize device before it's used by a driver. Ask low-level code
1902 * to enable I/O resources. Wake up the device if it was suspended.
1903 * Beware, this function can fail.
1904 */
1905int pci_enable_device_io(struct pci_dev *dev)
1906{
1907	return pci_enable_device_flags(dev, IORESOURCE_IO);
1908}
1909EXPORT_SYMBOL(pci_enable_device_io);
1910
1911/**
1912 * pci_enable_device_mem - Initialize a device for use with Memory space
1913 * @dev: PCI device to be initialized
1914 *
1915 * Initialize device before it's used by a driver. Ask low-level code
1916 * to enable Memory resources. Wake up the device if it was suspended.
1917 * Beware, this function can fail.
1918 */
1919int pci_enable_device_mem(struct pci_dev *dev)
1920{
1921	return pci_enable_device_flags(dev, IORESOURCE_MEM);
1922}
1923EXPORT_SYMBOL(pci_enable_device_mem);
1924
1925/**
1926 * pci_enable_device - Initialize device before it's used by a driver.
1927 * @dev: PCI device to be initialized
1928 *
1929 * Initialize device before it's used by a driver. Ask low-level code
1930 * to enable I/O and memory. Wake up the device if it was suspended.
1931 * Beware, this function can fail.
1932 *
1933 * Note we don't actually enable the device many times if we call
1934 * this function repeatedly (we just increment the count).
1935 */
1936int pci_enable_device(struct pci_dev *dev)
1937{
1938	return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1939}
1940EXPORT_SYMBOL(pci_enable_device);
1941
1942/*
1943 * Managed PCI resources.  This manages device on/off, INTx/MSI/MSI-X
1944 * on/off and BAR regions.  pci_dev itself records MSI/MSI-X status, so
1945 * there's no need to track it separately.  pci_devres is initialized
1946 * when a device is enabled using managed PCI device enable interface.
1947 */
1948struct pci_devres {
1949	unsigned int enabled:1;
1950	unsigned int pinned:1;
1951	unsigned int orig_intx:1;
1952	unsigned int restore_intx:1;
1953	unsigned int mwi:1;
1954	u32 region_mask;
1955};
1956
1957static void pcim_release(struct device *gendev, void *res)
1958{
1959	struct pci_dev *dev = to_pci_dev(gendev);
1960	struct pci_devres *this = res;
1961	int i;
1962
1963	if (dev->msi_enabled)
1964		pci_disable_msi(dev);
1965	if (dev->msix_enabled)
1966		pci_disable_msix(dev);
1967
1968	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1969		if (this->region_mask & (1 << i))
1970			pci_release_region(dev, i);
1971
1972	if (this->mwi)
1973		pci_clear_mwi(dev);
1974
1975	if (this->restore_intx)
1976		pci_intx(dev, this->orig_intx);
1977
1978	if (this->enabled && !this->pinned)
1979		pci_disable_device(dev);
1980}
1981
1982static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1983{
1984	struct pci_devres *dr, *new_dr;
1985
1986	dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1987	if (dr)
1988		return dr;
1989
1990	new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1991	if (!new_dr)
1992		return NULL;
1993	return devres_get(&pdev->dev, new_dr, NULL, NULL);
1994}
1995
1996static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1997{
1998	if (pci_is_managed(pdev))
1999		return devres_find(&pdev->dev, pcim_release, NULL, NULL);
2000	return NULL;
2001}
2002
2003/**
2004 * pcim_enable_device - Managed pci_enable_device()
2005 * @pdev: PCI device to be initialized
2006 *
2007 * Managed pci_enable_device().
2008 */
2009int pcim_enable_device(struct pci_dev *pdev)
2010{
2011	struct pci_devres *dr;
2012	int rc;
2013
2014	dr = get_pci_dr(pdev);
2015	if (unlikely(!dr))
2016		return -ENOMEM;
2017	if (dr->enabled)
2018		return 0;
2019
2020	rc = pci_enable_device(pdev);
2021	if (!rc) {
2022		pdev->is_managed = 1;
2023		dr->enabled = 1;
2024	}
2025	return rc;
2026}
2027EXPORT_SYMBOL(pcim_enable_device);
2028
2029/**
2030 * pcim_pin_device - Pin managed PCI device
2031 * @pdev: PCI device to pin
2032 *
2033 * Pin managed PCI device @pdev.  Pinned device won't be disabled on
2034 * driver detach.  @pdev must have been enabled with
2035 * pcim_enable_device().
2036 */
2037void pcim_pin_device(struct pci_dev *pdev)
2038{
2039	struct pci_devres *dr;
2040
2041	dr = find_pci_dr(pdev);
2042	WARN_ON(!dr || !dr->enabled);
2043	if (dr)
2044		dr->pinned = 1;
2045}
2046EXPORT_SYMBOL(pcim_pin_device);
2047
2048/*
2049 * pcibios_add_device - provide arch specific hooks when adding device dev
2050 * @dev: the PCI device being added
2051 *
2052 * Permits the platform to provide architecture specific functionality when
2053 * devices are added. This is the default implementation. Architecture
2054 * implementations can override this.
2055 */
2056int __weak pcibios_add_device(struct pci_dev *dev)
2057{
2058	return 0;
2059}
2060
2061/**
2062 * pcibios_release_device - provide arch specific hooks when releasing
2063 *			    device dev
2064 * @dev: the PCI device being released
2065 *
2066 * Permits the platform to provide architecture specific functionality when
2067 * devices are released. This is the default implementation. Architecture
2068 * implementations can override this.
2069 */
2070void __weak pcibios_release_device(struct pci_dev *dev) {}
2071
2072/**
2073 * pcibios_disable_device - disable arch specific PCI resources for device dev
2074 * @dev: the PCI device to disable
2075 *
2076 * Disables architecture specific PCI resources for the device. This
2077 * is the default implementation. Architecture implementations can
2078 * override this.
2079 */
2080void __weak pcibios_disable_device(struct pci_dev *dev) {}
2081
2082/**
2083 * pcibios_penalize_isa_irq - penalize an ISA IRQ
2084 * @irq: ISA IRQ to penalize
2085 * @active: IRQ active or not
2086 *
2087 * Permits the platform to provide architecture-specific functionality when
2088 * penalizing ISA IRQs. This is the default implementation. Architecture
2089 * implementations can override this.
2090 */
2091void __weak pcibios_penalize_isa_irq(int irq, int active) {}
2092
2093static void do_pci_disable_device(struct pci_dev *dev)
2094{
2095	u16 pci_command;
2096
2097	pci_read_config_word(dev, PCI_COMMAND, &pci_command);
2098	if (pci_command & PCI_COMMAND_MASTER) {
2099		pci_command &= ~PCI_COMMAND_MASTER;
2100		pci_write_config_word(dev, PCI_COMMAND, pci_command);
2101	}
2102
2103	pcibios_disable_device(dev);
2104}
2105
2106/**
2107 * pci_disable_enabled_device - Disable device without updating enable_cnt
2108 * @dev: PCI device to disable
2109 *
2110 * NOTE: This function is a backend of PCI power management routines and is
2111 * not supposed to be called drivers.
2112 */
2113void pci_disable_enabled_device(struct pci_dev *dev)
2114{
2115	if (pci_is_enabled(dev))
2116		do_pci_disable_device(dev);
2117}
2118
2119/**
2120 * pci_disable_device - Disable PCI device after use
2121 * @dev: PCI device to be disabled
2122 *
2123 * Signal to the system that the PCI device is not in use by the system
2124 * anymore.  This only involves disabling PCI bus-mastering, if active.
2125 *
2126 * Note we don't actually disable the device until all callers of
2127 * pci_enable_device() have called pci_disable_device().
2128 */
2129void pci_disable_device(struct pci_dev *dev)
2130{
2131	struct pci_devres *dr;
2132
2133	dr = find_pci_dr(dev);
2134	if (dr)
2135		dr->enabled = 0;
2136
2137	dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
2138		      "disabling already-disabled device");
2139
2140	if (atomic_dec_return(&dev->enable_cnt) != 0)
2141		return;
2142
2143	do_pci_disable_device(dev);
2144
2145	dev->is_busmaster = 0;
2146}
2147EXPORT_SYMBOL(pci_disable_device);
2148
2149/**
2150 * pcibios_set_pcie_reset_state - set reset state for device dev
2151 * @dev: the PCIe device reset
2152 * @state: Reset state to enter into
2153 *
2154 * Set the PCIe reset state for the device. This is the default
2155 * implementation. Architecture implementations can override this.
2156 */
2157int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
2158					enum pcie_reset_state state)
2159{
2160	return -EINVAL;
2161}
2162
2163/**
2164 * pci_set_pcie_reset_state - set reset state for device dev
2165 * @dev: the PCIe device reset
2166 * @state: Reset state to enter into
2167 *
2168 * Sets the PCI reset state for the device.
2169 */
2170int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
2171{
2172	return pcibios_set_pcie_reset_state(dev, state);
2173}
2174EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
2175
2176void pcie_clear_device_status(struct pci_dev *dev)
2177{
2178	u16 sta;
2179
2180	pcie_capability_read_word(dev, PCI_EXP_DEVSTA, &sta);
2181	pcie_capability_write_word(dev, PCI_EXP_DEVSTA, sta);
2182}
2183
2184/**
2185 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
2186 * @dev: PCIe root port or event collector.
2187 */
2188void pcie_clear_root_pme_status(struct pci_dev *dev)
2189{
2190	pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
2191}
2192
2193/**
2194 * pci_check_pme_status - Check if given device has generated PME.
2195 * @dev: Device to check.
2196 *
2197 * Check the PME status of the device and if set, clear it and clear PME enable
2198 * (if set).  Return 'true' if PME status and PME enable were both set or
2199 * 'false' otherwise.
2200 */
2201bool pci_check_pme_status(struct pci_dev *dev)
2202{
2203	int pmcsr_pos;
2204	u16 pmcsr;
2205	bool ret = false;
2206
2207	if (!dev->pm_cap)
2208		return false;
2209
2210	pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
2211	pci_read_config_word(dev, pmcsr_pos, &pmcsr);
2212	if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
2213		return false;
2214
2215	/* Clear PME status. */
2216	pmcsr |= PCI_PM_CTRL_PME_STATUS;
2217	if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
2218		/* Disable PME to avoid interrupt flood. */
2219		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2220		ret = true;
2221	}
2222
2223	pci_write_config_word(dev, pmcsr_pos, pmcsr);
2224
2225	return ret;
2226}
2227
2228/**
2229 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2230 * @dev: Device to handle.
2231 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2232 *
2233 * Check if @dev has generated PME and queue a resume request for it in that
2234 * case.
2235 */
2236static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2237{
2238	if (pme_poll_reset && dev->pme_poll)
2239		dev->pme_poll = false;
2240
2241	if (pci_check_pme_status(dev)) {
2242		pci_wakeup_event(dev);
2243		pm_request_resume(&dev->dev);
2244	}
2245	return 0;
2246}
2247
2248/**
2249 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2250 * @bus: Top bus of the subtree to walk.
2251 */
2252void pci_pme_wakeup_bus(struct pci_bus *bus)
2253{
2254	if (bus)
2255		pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2256}
2257
2258
2259/**
2260 * pci_pme_capable - check the capability of PCI device to generate PME#
2261 * @dev: PCI device to handle.
2262 * @state: PCI state from which device will issue PME#.
2263 */
2264bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2265{
2266	if (!dev->pm_cap)
2267		return false;
2268
2269	return !!(dev->pme_support & (1 << state));
2270}
2271EXPORT_SYMBOL(pci_pme_capable);
2272
2273static void pci_pme_list_scan(struct work_struct *work)
2274{
2275	struct pci_pme_device *pme_dev, *n;
2276
2277	mutex_lock(&pci_pme_list_mutex);
2278	list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2279		if (pme_dev->dev->pme_poll) {
2280			struct pci_dev *bridge;
2281
2282			bridge = pme_dev->dev->bus->self;
2283			/*
2284			 * If bridge is in low power state, the
2285			 * configuration space of subordinate devices
2286			 * may be not accessible
2287			 */
2288			if (bridge && bridge->current_state != PCI_D0)
2289				continue;
2290			/*
2291			 * If the device is in D3cold it should not be
2292			 * polled either.
2293			 */
2294			if (pme_dev->dev->current_state == PCI_D3cold)
2295				continue;
2296
2297			pci_pme_wakeup(pme_dev->dev, NULL);
2298		} else {
2299			list_del(&pme_dev->list);
2300			kfree(pme_dev);
2301		}
2302	}
2303	if (!list_empty(&pci_pme_list))
2304		queue_delayed_work(system_freezable_wq, &pci_pme_work,
2305				   msecs_to_jiffies(PME_TIMEOUT));
2306	mutex_unlock(&pci_pme_list_mutex);
2307}
2308
2309static void __pci_pme_active(struct pci_dev *dev, bool enable)
2310{
2311	u16 pmcsr;
2312
2313	if (!dev->pme_support)
2314		return;
2315
2316	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2317	/* Clear PME_Status by writing 1 to it and enable PME# */
2318	pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2319	if (!enable)
2320		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2321
2322	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2323}
2324
2325/**
2326 * pci_pme_restore - Restore PME configuration after config space restore.
2327 * @dev: PCI device to update.
2328 */
2329void pci_pme_restore(struct pci_dev *dev)
2330{
2331	u16 pmcsr;
2332
2333	if (!dev->pme_support)
2334		return;
2335
2336	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2337	if (dev->wakeup_prepared) {
2338		pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2339		pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2340	} else {
2341		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2342		pmcsr |= PCI_PM_CTRL_PME_STATUS;
2343	}
2344	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2345}
2346
2347/**
2348 * pci_pme_active - enable or disable PCI device's PME# function
2349 * @dev: PCI device to handle.
2350 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2351 *
2352 * The caller must verify that the device is capable of generating PME# before
2353 * calling this function with @enable equal to 'true'.
2354 */
2355void pci_pme_active(struct pci_dev *dev, bool enable)
2356{
2357	__pci_pme_active(dev, enable);
2358
2359	/*
2360	 * PCI (as opposed to PCIe) PME requires that the device have
2361	 * its PME# line hooked up correctly. Not all hardware vendors
2362	 * do this, so the PME never gets delivered and the device
2363	 * remains asleep. The easiest way around this is to
2364	 * periodically walk the list of suspended devices and check
2365	 * whether any have their PME flag set. The assumption is that
2366	 * we'll wake up often enough anyway that this won't be a huge
2367	 * hit, and the power savings from the devices will still be a
2368	 * win.
2369	 *
2370	 * Although PCIe uses in-band PME message instead of PME# line
2371	 * to report PME, PME does not work for some PCIe devices in
2372	 * reality.  For example, there are devices that set their PME
2373	 * status bits, but don't really bother to send a PME message;
2374	 * there are PCI Express Root Ports that don't bother to
2375	 * trigger interrupts when they receive PME messages from the
2376	 * devices below.  So PME poll is used for PCIe devices too.
2377	 */
2378
2379	if (dev->pme_poll) {
2380		struct pci_pme_device *pme_dev;
2381		if (enable) {
2382			pme_dev = kmalloc(sizeof(struct pci_pme_device),
2383					  GFP_KERNEL);
2384			if (!pme_dev) {
2385				pci_warn(dev, "can't enable PME#\n");
2386				return;
2387			}
2388			pme_dev->dev = dev;
2389			mutex_lock(&pci_pme_list_mutex);
2390			list_add(&pme_dev->list, &pci_pme_list);
2391			if (list_is_singular(&pci_pme_list))
2392				queue_delayed_work(system_freezable_wq,
2393						   &pci_pme_work,
2394						   msecs_to_jiffies(PME_TIMEOUT));
2395			mutex_unlock(&pci_pme_list_mutex);
2396		} else {
2397			mutex_lock(&pci_pme_list_mutex);
2398			list_for_each_entry(pme_dev, &pci_pme_list, list) {
2399				if (pme_dev->dev == dev) {
2400					list_del(&pme_dev->list);
2401					kfree(pme_dev);
2402					break;
2403				}
2404			}
2405			mutex_unlock(&pci_pme_list_mutex);
2406		}
2407	}
2408
2409	pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2410}
2411EXPORT_SYMBOL(pci_pme_active);
2412
2413/**
2414 * __pci_enable_wake - enable PCI device as wakeup event source
2415 * @dev: PCI device affected
2416 * @state: PCI state from which device will issue wakeup events
2417 * @enable: True to enable event generation; false to disable
2418 *
2419 * This enables the device as a wakeup event source, or disables it.
2420 * When such events involves platform-specific hooks, those hooks are
2421 * called automatically by this routine.
2422 *
2423 * Devices with legacy power management (no standard PCI PM capabilities)
2424 * always require such platform hooks.
2425 *
2426 * RETURN VALUE:
2427 * 0 is returned on success
2428 * -EINVAL is returned if device is not supposed to wake up the system
2429 * Error code depending on the platform is returned if both the platform and
2430 * the native mechanism fail to enable the generation of wake-up events
2431 */
2432static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2433{
2434	int ret = 0;
2435
2436	/*
2437	 * Bridges that are not power-manageable directly only signal
2438	 * wakeup on behalf of subordinate devices which is set up
2439	 * elsewhere, so skip them. However, bridges that are
2440	 * power-manageable may signal wakeup for themselves (for example,
2441	 * on a hotplug event) and they need to be covered here.
2442	 */
2443	if (!pci_power_manageable(dev))
2444		return 0;
2445
2446	/* Don't do the same thing twice in a row for one device. */
2447	if (!!enable == !!dev->wakeup_prepared)
2448		return 0;
2449
2450	/*
2451	 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2452	 * Anderson we should be doing PME# wake enable followed by ACPI wake
2453	 * enable.  To disable wake-up we call the platform first, for symmetry.
2454	 */
2455
2456	if (enable) {
2457		int error;
2458
2459		if (pci_pme_capable(dev, state))
2460			pci_pme_active(dev, true);
2461		else
2462			ret = 1;
2463		error = platform_pci_set_wakeup(dev, true);
2464		if (ret)
2465			ret = error;
2466		if (!ret)
2467			dev->wakeup_prepared = true;
2468	} else {
2469		platform_pci_set_wakeup(dev, false);
2470		pci_pme_active(dev, false);
2471		dev->wakeup_prepared = false;
2472	}
2473
2474	return ret;
2475}
2476
2477/**
2478 * pci_enable_wake - change wakeup settings for a PCI device
2479 * @pci_dev: Target device
2480 * @state: PCI state from which device will issue wakeup events
2481 * @enable: Whether or not to enable event generation
2482 *
2483 * If @enable is set, check device_may_wakeup() for the device before calling
2484 * __pci_enable_wake() for it.
2485 */
2486int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2487{
2488	if (enable && !device_may_wakeup(&pci_dev->dev))
2489		return -EINVAL;
2490
2491	return __pci_enable_wake(pci_dev, state, enable);
2492}
2493EXPORT_SYMBOL(pci_enable_wake);
2494
2495/**
2496 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2497 * @dev: PCI device to prepare
2498 * @enable: True to enable wake-up event generation; false to disable
2499 *
2500 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2501 * and this function allows them to set that up cleanly - pci_enable_wake()
2502 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2503 * ordering constraints.
2504 *
2505 * This function only returns error code if the device is not allowed to wake
2506 * up the system from sleep or it is not capable of generating PME# from both
2507 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2508 */
2509int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2510{
2511	return pci_pme_capable(dev, PCI_D3cold) ?
2512			pci_enable_wake(dev, PCI_D3cold, enable) :
2513			pci_enable_wake(dev, PCI_D3hot, enable);
2514}
2515EXPORT_SYMBOL(pci_wake_from_d3);
2516
2517/**
2518 * pci_target_state - find an appropriate low power state for a given PCI dev
2519 * @dev: PCI device
2520 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2521 *
2522 * Use underlying platform code to find a supported low power state for @dev.
2523 * If the platform can't manage @dev, return the deepest state from which it
2524 * can generate wake events, based on any available PME info.
2525 */
2526static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2527{
2528	pci_power_t target_state = PCI_D3hot;
2529
2530	if (platform_pci_power_manageable(dev)) {
2531		/*
2532		 * Call the platform to find the target state for the device.
2533		 */
2534		pci_power_t state = platform_pci_choose_state(dev);
2535
2536		switch (state) {
2537		case PCI_POWER_ERROR:
2538		case PCI_UNKNOWN:
2539			break;
2540		case PCI_D1:
2541		case PCI_D2:
2542			if (pci_no_d1d2(dev))
2543				break;
2544			fallthrough;
2545		default:
2546			target_state = state;
2547		}
2548
2549		return target_state;
2550	}
2551
2552	if (!dev->pm_cap)
2553		target_state = PCI_D0;
2554
2555	/*
2556	 * If the device is in D3cold even though it's not power-manageable by
2557	 * the platform, it may have been powered down by non-standard means.
2558	 * Best to let it slumber.
2559	 */
2560	if (dev->current_state == PCI_D3cold)
2561		target_state = PCI_D3cold;
2562
2563	if (wakeup) {
2564		/*
2565		 * Find the deepest state from which the device can generate
2566		 * PME#.
2567		 */
2568		if (dev->pme_support) {
2569			while (target_state
2570			      && !(dev->pme_support & (1 << target_state)))
2571				target_state--;
2572		}
2573	}
2574
2575	return target_state;
2576}
2577
2578/**
2579 * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2580 *			  into a sleep state
2581 * @dev: Device to handle.
2582 *
2583 * Choose the power state appropriate for the device depending on whether
2584 * it can wake up the system and/or is power manageable by the platform
2585 * (PCI_D3hot is the default) and put the device into that state.
2586 */
2587int pci_prepare_to_sleep(struct pci_dev *dev)
2588{
2589	bool wakeup = device_may_wakeup(&dev->dev);
2590	pci_power_t target_state = pci_target_state(dev, wakeup);
2591	int error;
2592
2593	if (target_state == PCI_POWER_ERROR)
2594		return -EIO;
2595
2596	pci_enable_wake(dev, target_state, wakeup);
2597
2598	error = pci_set_power_state(dev, target_state);
2599
2600	if (error)
2601		pci_enable_wake(dev, target_state, false);
2602
2603	return error;
2604}
2605EXPORT_SYMBOL(pci_prepare_to_sleep);
2606
2607/**
2608 * pci_back_from_sleep - turn PCI device on during system-wide transition
2609 *			 into working state
2610 * @dev: Device to handle.
2611 *
2612 * Disable device's system wake-up capability and put it into D0.
2613 */
2614int pci_back_from_sleep(struct pci_dev *dev)
2615{
2616	pci_enable_wake(dev, PCI_D0, false);
2617	return pci_set_power_state(dev, PCI_D0);
2618}
2619EXPORT_SYMBOL(pci_back_from_sleep);
2620
2621/**
2622 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2623 * @dev: PCI device being suspended.
2624 *
2625 * Prepare @dev to generate wake-up events at run time and put it into a low
2626 * power state.
2627 */
2628int pci_finish_runtime_suspend(struct pci_dev *dev)
2629{
2630	pci_power_t target_state;
2631	int error;
2632
2633	target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2634	if (target_state == PCI_POWER_ERROR)
2635		return -EIO;
2636
2637	dev->runtime_d3cold = target_state == PCI_D3cold;
2638
2639	__pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2640
2641	error = pci_set_power_state(dev, target_state);
2642
2643	if (error) {
2644		pci_enable_wake(dev, target_state, false);
2645		dev->runtime_d3cold = false;
2646	}
2647
2648	return error;
2649}
2650
2651/**
2652 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2653 * @dev: Device to check.
2654 *
2655 * Return true if the device itself is capable of generating wake-up events
2656 * (through the platform or using the native PCIe PME) or if the device supports
2657 * PME and one of its upstream bridges can generate wake-up events.
2658 */
2659bool pci_dev_run_wake(struct pci_dev *dev)
2660{
2661	struct pci_bus *bus = dev->bus;
2662
2663	if (!dev->pme_support)
2664		return false;
2665
2666	/* PME-capable in principle, but not from the target power state */
2667	if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2668		return false;
2669
2670	if (device_can_wakeup(&dev->dev))
2671		return true;
2672
2673	while (bus->parent) {
2674		struct pci_dev *bridge = bus->self;
2675
2676		if (device_can_wakeup(&bridge->dev))
2677			return true;
2678
2679		bus = bus->parent;
2680	}
2681
2682	/* We have reached the root bus. */
2683	if (bus->bridge)
2684		return device_can_wakeup(bus->bridge);
2685
2686	return false;
2687}
2688EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2689
2690/**
2691 * pci_dev_need_resume - Check if it is necessary to resume the device.
2692 * @pci_dev: Device to check.
2693 *
2694 * Return 'true' if the device is not runtime-suspended or it has to be
2695 * reconfigured due to wakeup settings difference between system and runtime
2696 * suspend, or the current power state of it is not suitable for the upcoming
2697 * (system-wide) transition.
2698 */
2699bool pci_dev_need_resume(struct pci_dev *pci_dev)
2700{
2701	struct device *dev = &pci_dev->dev;
2702	pci_power_t target_state;
2703
2704	if (!pm_runtime_suspended(dev) || platform_pci_need_resume(pci_dev))
2705		return true;
2706
2707	target_state = pci_target_state(pci_dev, device_may_wakeup(dev));
2708
2709	/*
2710	 * If the earlier platform check has not triggered, D3cold is just power
2711	 * removal on top of D3hot, so no need to resume the device in that
2712	 * case.
2713	 */
2714	return target_state != pci_dev->current_state &&
2715		target_state != PCI_D3cold &&
2716		pci_dev->current_state != PCI_D3hot;
2717}
2718
2719/**
2720 * pci_dev_adjust_pme - Adjust PME setting for a suspended device.
2721 * @pci_dev: Device to check.
2722 *
2723 * If the device is suspended and it is not configured for system wakeup,
2724 * disable PME for it to prevent it from waking up the system unnecessarily.
2725 *
2726 * Note that if the device's power state is D3cold and the platform check in
2727 * pci_dev_need_resume() has not triggered, the device's configuration need not
2728 * be changed.
2729 */
2730void pci_dev_adjust_pme(struct pci_dev *pci_dev)
2731{
2732	struct device *dev = &pci_dev->dev;
2733
2734	spin_lock_irq(&dev->power.lock);
2735
2736	if (pm_runtime_suspended(dev) && !device_may_wakeup(dev) &&
2737	    pci_dev->current_state < PCI_D3cold)
2738		__pci_pme_active(pci_dev, false);
2739
2740	spin_unlock_irq(&dev->power.lock);
2741}
2742
2743/**
2744 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2745 * @pci_dev: Device to handle.
2746 *
2747 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2748 * it might have been disabled during the prepare phase of system suspend if
2749 * the device was not configured for system wakeup.
2750 */
2751void pci_dev_complete_resume(struct pci_dev *pci_dev)
2752{
2753	struct device *dev = &pci_dev->dev;
2754
2755	if (!pci_dev_run_wake(pci_dev))
2756		return;
2757
2758	spin_lock_irq(&dev->power.lock);
2759
2760	if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2761		__pci_pme_active(pci_dev, true);
2762
2763	spin_unlock_irq(&dev->power.lock);
2764}
2765
2766void pci_config_pm_runtime_get(struct pci_dev *pdev)
2767{
2768	struct device *dev = &pdev->dev;
2769	struct device *parent = dev->parent;
2770
2771	if (parent)
2772		pm_runtime_get_sync(parent);
2773	pm_runtime_get_noresume(dev);
2774	/*
2775	 * pdev->current_state is set to PCI_D3cold during suspending,
2776	 * so wait until suspending completes
2777	 */
2778	pm_runtime_barrier(dev);
2779	/*
2780	 * Only need to resume devices in D3cold, because config
2781	 * registers are still accessible for devices suspended but
2782	 * not in D3cold.
2783	 */
2784	if (pdev->current_state == PCI_D3cold)
2785		pm_runtime_resume(dev);
2786}
2787
2788void pci_config_pm_runtime_put(struct pci_dev *pdev)
2789{
2790	struct device *dev = &pdev->dev;
2791	struct device *parent = dev->parent;
2792
2793	pm_runtime_put(dev);
2794	if (parent)
2795		pm_runtime_put_sync(parent);
2796}
2797
2798static const struct dmi_system_id bridge_d3_blacklist[] = {
2799#ifdef CONFIG_X86
2800	{
2801		/*
2802		 * Gigabyte X299 root port is not marked as hotplug capable
2803		 * which allows Linux to power manage it.  However, this
2804		 * confuses the BIOS SMI handler so don't power manage root
2805		 * ports on that system.
2806		 */
2807		.ident = "X299 DESIGNARE EX-CF",
2808		.matches = {
2809			DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2810			DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2811		},
2812	},
2813#endif
2814	{ }
2815};
2816
2817/**
2818 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2819 * @bridge: Bridge to check
2820 *
2821 * This function checks if it is possible to move the bridge to D3.
2822 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2823 */
2824bool pci_bridge_d3_possible(struct pci_dev *bridge)
2825{
2826	if (!pci_is_pcie(bridge))
2827		return false;
2828
2829	switch (pci_pcie_type(bridge)) {
2830	case PCI_EXP_TYPE_ROOT_PORT:
2831	case PCI_EXP_TYPE_UPSTREAM:
2832	case PCI_EXP_TYPE_DOWNSTREAM:
2833		if (pci_bridge_d3_disable)
2834			return false;
2835
2836		/*
2837		 * Hotplug ports handled by firmware in System Management Mode
2838		 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2839		 */
2840		if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2841			return false;
2842
2843		if (pci_bridge_d3_force)
2844			return true;
2845
2846		/* Even the oldest 2010 Thunderbolt controller supports D3. */
2847		if (bridge->is_thunderbolt)
2848			return true;
2849
2850		/* Platform might know better if the bridge supports D3 */
2851		if (platform_pci_bridge_d3(bridge))
2852			return true;
2853
2854		/*
2855		 * Hotplug ports handled natively by the OS were not validated
2856		 * by vendors for runtime D3 at least until 2018 because there
2857		 * was no OS support.
2858		 */
2859		if (bridge->is_hotplug_bridge)
2860			return false;
2861
2862		if (dmi_check_system(bridge_d3_blacklist))
2863			return false;
2864
2865		/*
2866		 * It should be safe to put PCIe ports from 2015 or newer
2867		 * to D3.
2868		 */
2869		if (dmi_get_bios_year() >= 2015)
2870			return true;
2871		break;
2872	}
2873
2874	return false;
2875}
2876
2877static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2878{
2879	bool *d3cold_ok = data;
2880
2881	if (/* The device needs to be allowed to go D3cold ... */
2882	    dev->no_d3cold || !dev->d3cold_allowed ||
2883
2884	    /* ... and if it is wakeup capable to do so from D3cold. */
2885	    (device_may_wakeup(&dev->dev) &&
2886	     !pci_pme_capable(dev, PCI_D3cold)) ||
2887
2888	    /* If it is a bridge it must be allowed to go to D3. */
2889	    !pci_power_manageable(dev))
2890
2891		*d3cold_ok = false;
2892
2893	return !*d3cold_ok;
2894}
2895
2896/*
2897 * pci_bridge_d3_update - Update bridge D3 capabilities
2898 * @dev: PCI device which is changed
2899 *
2900 * Update upstream bridge PM capabilities accordingly depending on if the
2901 * device PM configuration was changed or the device is being removed.  The
2902 * change is also propagated upstream.
2903 */
2904void pci_bridge_d3_update(struct pci_dev *dev)
2905{
2906	bool remove = !device_is_registered(&dev->dev);
2907	struct pci_dev *bridge;
2908	bool d3cold_ok = true;
2909
2910	bridge = pci_upstream_bridge(dev);
2911	if (!bridge || !pci_bridge_d3_possible(bridge))
2912		return;
2913
2914	/*
2915	 * If D3 is currently allowed for the bridge, removing one of its
2916	 * children won't change that.
2917	 */
2918	if (remove && bridge->bridge_d3)
2919		return;
2920
2921	/*
2922	 * If D3 is currently allowed for the bridge and a child is added or
2923	 * changed, disallowance of D3 can only be caused by that child, so
2924	 * we only need to check that single device, not any of its siblings.
2925	 *
2926	 * If D3 is currently not allowed for the bridge, checking the device
2927	 * first may allow us to skip checking its siblings.
2928	 */
2929	if (!remove)
2930		pci_dev_check_d3cold(dev, &d3cold_ok);
2931
2932	/*
2933	 * If D3 is currently not allowed for the bridge, this may be caused
2934	 * either by the device being changed/removed or any of its siblings,
2935	 * so we need to go through all children to find out if one of them
2936	 * continues to block D3.
2937	 */
2938	if (d3cold_ok && !bridge->bridge_d3)
2939		pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2940			     &d3cold_ok);
2941
2942	if (bridge->bridge_d3 != d3cold_ok) {
2943		bridge->bridge_d3 = d3cold_ok;
2944		/* Propagate change to upstream bridges */
2945		pci_bridge_d3_update(bridge);
2946	}
2947}
2948
2949/**
2950 * pci_d3cold_enable - Enable D3cold for device
2951 * @dev: PCI device to handle
2952 *
2953 * This function can be used in drivers to enable D3cold from the device
2954 * they handle.  It also updates upstream PCI bridge PM capabilities
2955 * accordingly.
2956 */
2957void pci_d3cold_enable(struct pci_dev *dev)
2958{
2959	if (dev->no_d3cold) {
2960		dev->no_d3cold = false;
2961		pci_bridge_d3_update(dev);
2962	}
2963}
2964EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2965
2966/**
2967 * pci_d3cold_disable - Disable D3cold for device
2968 * @dev: PCI device to handle
2969 *
2970 * This function can be used in drivers to disable D3cold from the device
2971 * they handle.  It also updates upstream PCI bridge PM capabilities
2972 * accordingly.
2973 */
2974void pci_d3cold_disable(struct pci_dev *dev)
2975{
2976	if (!dev->no_d3cold) {
2977		dev->no_d3cold = true;
2978		pci_bridge_d3_update(dev);
2979	}
2980}
2981EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2982
2983/**
2984 * pci_pm_init - Initialize PM functions of given PCI device
2985 * @dev: PCI device to handle.
2986 */
2987void pci_pm_init(struct pci_dev *dev)
2988{
2989	int pm;
2990	u16 status;
2991	u16 pmc;
2992
2993	pm_runtime_forbid(&dev->dev);
2994	pm_runtime_set_active(&dev->dev);
2995	pm_runtime_enable(&dev->dev);
2996	device_enable_async_suspend(&dev->dev);
2997	dev->wakeup_prepared = false;
2998
2999	dev->pm_cap = 0;
3000	dev->pme_support = 0;
3001
3002	/* find PCI PM capability in list */
3003	pm = pci_find_capability(dev, PCI_CAP_ID_PM);
3004	if (!pm)
3005		return;
3006	/* Check device's ability to generate PME# */
3007	pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
3008
3009	if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
3010		pci_err(dev, "unsupported PM cap regs version (%u)\n",
3011			pmc & PCI_PM_CAP_VER_MASK);
3012		return;
3013	}
3014
3015	dev->pm_cap = pm;
3016	dev->d3_delay = PCI_PM_D3_WAIT;
3017	dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
3018	dev->bridge_d3 = pci_bridge_d3_possible(dev);
3019	dev->d3cold_allowed = true;
3020
3021	dev->d1_support = false;
3022	dev->d2_support = false;
3023	if (!pci_no_d1d2(dev)) {
3024		if (pmc & PCI_PM_CAP_D1)
3025			dev->d1_support = true;
3026		if (pmc & PCI_PM_CAP_D2)
3027			dev->d2_support = true;
3028
3029		if (dev->d1_support || dev->d2_support)
3030			pci_info(dev, "supports%s%s\n",
3031				   dev->d1_support ? " D1" : "",
3032				   dev->d2_support ? " D2" : "");
3033	}
3034
3035	pmc &= PCI_PM_CAP_PME_MASK;
3036	if (pmc) {
3037		pci_info(dev, "PME# supported from%s%s%s%s%s\n",
3038			 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
3039			 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
3040			 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
3041			 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
3042			 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
3043		dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
3044		dev->pme_poll = true;
3045		/*
3046		 * Make device's PM flags reflect the wake-up capability, but
3047		 * let the user space enable it to wake up the system as needed.
3048		 */
3049		device_set_wakeup_capable(&dev->dev, true);
3050		/* Disable the PME# generation functionality */
3051		pci_pme_active(dev, false);
3052	}
3053
3054	pci_read_config_word(dev, PCI_STATUS, &status);
3055	if (status & PCI_STATUS_IMM_READY)
3056		dev->imm_ready = 1;
3057}
3058
3059static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
3060{
3061	unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
3062
3063	switch (prop) {
3064	case PCI_EA_P_MEM:
3065	case PCI_EA_P_VF_MEM:
3066		flags |= IORESOURCE_MEM;
3067		break;
3068	case PCI_EA_P_MEM_PREFETCH:
3069	case PCI_EA_P_VF_MEM_PREFETCH:
3070		flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
3071		break;
3072	case PCI_EA_P_IO:
3073		flags |= IORESOURCE_IO;
3074		break;
3075	default:
3076		return 0;
3077	}
3078
3079	return flags;
3080}
3081
3082static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
3083					    u8 prop)
3084{
3085	if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
3086		return &dev->resource[bei];
3087#ifdef CONFIG_PCI_IOV
3088	else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
3089		 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
3090		return &dev->resource[PCI_IOV_RESOURCES +
3091				      bei - PCI_EA_BEI_VF_BAR0];
3092#endif
3093	else if (bei == PCI_EA_BEI_ROM)
3094		return &dev->resource[PCI_ROM_RESOURCE];
3095	else
3096		return NULL;
3097}
3098
3099/* Read an Enhanced Allocation (EA) entry */
3100static int pci_ea_read(struct pci_dev *dev, int offset)
3101{
3102	struct resource *res;
3103	int ent_size, ent_offset = offset;
3104	resource_size_t start, end;
3105	unsigned long flags;
3106	u32 dw0, bei, base, max_offset;
3107	u8 prop;
3108	bool support_64 = (sizeof(resource_size_t) >= 8);
3109
3110	pci_read_config_dword(dev, ent_offset, &dw0);
3111	ent_offset += 4;
3112
3113	/* Entry size field indicates DWORDs after 1st */
3114	ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
3115
3116	if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
3117		goto out;
3118
3119	bei = (dw0 & PCI_EA_BEI) >> 4;
3120	prop = (dw0 & PCI_EA_PP) >> 8;
3121
3122	/*
3123	 * If the Property is in the reserved range, try the Secondary
3124	 * Property instead.
3125	 */
3126	if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
3127		prop = (dw0 & PCI_EA_SP) >> 16;
3128	if (prop > PCI_EA_P_BRIDGE_IO)
3129		goto out;
3130
3131	res = pci_ea_get_resource(dev, bei, prop);
3132	if (!res) {
3133		pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
3134		goto out;
3135	}
3136
3137	flags = pci_ea_flags(dev, prop);
3138	if (!flags) {
3139		pci_err(dev, "Unsupported EA properties: %#x\n", prop);
3140		goto out;
3141	}
3142
3143	/* Read Base */
3144	pci_read_config_dword(dev, ent_offset, &base);
3145	start = (base & PCI_EA_FIELD_MASK);
3146	ent_offset += 4;
3147
3148	/* Read MaxOffset */
3149	pci_read_config_dword(dev, ent_offset, &max_offset);
3150	ent_offset += 4;
3151
3152	/* Read Base MSBs (if 64-bit entry) */
3153	if (base & PCI_EA_IS_64) {
3154		u32 base_upper;
3155
3156		pci_read_config_dword(dev, ent_offset, &base_upper);
3157		ent_offset += 4;
3158
3159		flags |= IORESOURCE_MEM_64;
3160
3161		/* entry starts above 32-bit boundary, can't use */
3162		if (!support_64 && base_upper)
3163			goto out;
3164
3165		if (support_64)
3166			start |= ((u64)base_upper << 32);
3167	}
3168
3169	end = start + (max_offset | 0x03);
3170
3171	/* Read MaxOffset MSBs (if 64-bit entry) */
3172	if (max_offset & PCI_EA_IS_64) {
3173		u32 max_offset_upper;
3174
3175		pci_read_config_dword(dev, ent_offset, &max_offset_upper);
3176		ent_offset += 4;
3177
3178		flags |= IORESOURCE_MEM_64;
3179
3180		/* entry too big, can't use */
3181		if (!support_64 && max_offset_upper)
3182			goto out;
3183
3184		if (support_64)
3185			end += ((u64)max_offset_upper << 32);
3186	}
3187
3188	if (end < start) {
3189		pci_err(dev, "EA Entry crosses address boundary\n");
3190		goto out;
3191	}
3192
3193	if (ent_size != ent_offset - offset) {
3194		pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
3195			ent_size, ent_offset - offset);
3196		goto out;
3197	}
3198
3199	res->name = pci_name(dev);
3200	res->start = start;
3201	res->end = end;
3202	res->flags = flags;
3203
3204	if (bei <= PCI_EA_BEI_BAR5)
3205		pci_info(dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3206			   bei, res, prop);
3207	else if (bei == PCI_EA_BEI_ROM)
3208		pci_info(dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
3209			   res, prop);
3210	else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
3211		pci_info(dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3212			   bei - PCI_EA_BEI_VF_BAR0, res, prop);
3213	else
3214		pci_info(dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
3215			   bei, res, prop);
3216
3217out:
3218	return offset + ent_size;
3219}
3220
3221/* Enhanced Allocation Initialization */
3222void pci_ea_init(struct pci_dev *dev)
3223{
3224	int ea;
3225	u8 num_ent;
3226	int offset;
3227	int i;
3228
3229	/* find PCI EA capability in list */
3230	ea = pci_find_capability(dev, PCI_CAP_ID_EA);
3231	if (!ea)
3232		return;
3233
3234	/* determine the number of entries */
3235	pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
3236					&num_ent);
3237	num_ent &= PCI_EA_NUM_ENT_MASK;
3238
3239	offset = ea + PCI_EA_FIRST_ENT;
3240
3241	/* Skip DWORD 2 for type 1 functions */
3242	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3243		offset += 4;
3244
3245	/* parse each EA entry */
3246	for (i = 0; i < num_ent; ++i)
3247		offset = pci_ea_read(dev, offset);
3248}
3249
3250static void pci_add_saved_cap(struct pci_dev *pci_dev,
3251	struct pci_cap_saved_state *new_cap)
3252{
3253	hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3254}
3255
3256/**
3257 * _pci_add_cap_save_buffer - allocate buffer for saving given
3258 *			      capability registers
3259 * @dev: the PCI device
3260 * @cap: the capability to allocate the buffer for
3261 * @extended: Standard or Extended capability ID
3262 * @size: requested size of the buffer
3263 */
3264static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3265				    bool extended, unsigned int size)
3266{
3267	int pos;
3268	struct pci_cap_saved_state *save_state;
3269
3270	if (extended)
3271		pos = pci_find_ext_capability(dev, cap);
3272	else
3273		pos = pci_find_capability(dev, cap);
3274
3275	if (!pos)
3276		return 0;
3277
3278	save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3279	if (!save_state)
3280		return -ENOMEM;
3281
3282	save_state->cap.cap_nr = cap;
3283	save_state->cap.cap_extended = extended;
3284	save_state->cap.size = size;
3285	pci_add_saved_cap(dev, save_state);
3286
3287	return 0;
3288}
3289
3290int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3291{
3292	return _pci_add_cap_save_buffer(dev, cap, false, size);
3293}
3294
3295int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3296{
3297	return _pci_add_cap_save_buffer(dev, cap, true, size);
3298}
3299
3300/**
3301 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3302 * @dev: the PCI device
3303 */
3304void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3305{
3306	int error;
3307
3308	error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3309					PCI_EXP_SAVE_REGS * sizeof(u16));
3310	if (error)
3311		pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3312
3313	error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3314	if (error)
3315		pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3316
3317	error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3318					    2 * sizeof(u16));
3319	if (error)
3320		pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3321
3322	pci_allocate_vc_save_buffers(dev);
3323}
3324
3325void pci_free_cap_save_buffers(struct pci_dev *dev)
3326{
3327	struct pci_cap_saved_state *tmp;
3328	struct hlist_node *n;
3329
3330	hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3331		kfree(tmp);
3332}
3333
3334/**
3335 * pci_configure_ari - enable or disable ARI forwarding
3336 * @dev: the PCI device
3337 *
3338 * If @dev and its upstream bridge both support ARI, enable ARI in the
3339 * bridge.  Otherwise, disable ARI in the bridge.
3340 */
3341void pci_configure_ari(struct pci_dev *dev)
3342{
3343	u32 cap;
3344	struct pci_dev *bridge;
3345
3346	if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3347		return;
3348
3349	bridge = dev->bus->self;
3350	if (!bridge)
3351		return;
3352
3353	pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3354	if (!(cap & PCI_EXP_DEVCAP2_ARI))
3355		return;
3356
3357	if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3358		pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3359					 PCI_EXP_DEVCTL2_ARI);
3360		bridge->ari_enabled = 1;
3361	} else {
3362		pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3363					   PCI_EXP_DEVCTL2_ARI);
3364		bridge->ari_enabled = 0;
3365	}
3366}
3367
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3368static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3369{
3370	int pos;
3371	u16 cap, ctrl;
3372
3373	pos = pdev->acs_cap;
3374	if (!pos)
3375		return false;
3376
3377	/*
3378	 * Except for egress control, capabilities are either required
3379	 * or only required if controllable.  Features missing from the
3380	 * capability field can therefore be assumed as hard-wired enabled.
3381	 */
3382	pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3383	acs_flags &= (cap | PCI_ACS_EC);
3384
3385	pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3386	return (ctrl & acs_flags) == acs_flags;
3387}
3388
3389/**
3390 * pci_acs_enabled - test ACS against required flags for a given device
3391 * @pdev: device to test
3392 * @acs_flags: required PCI ACS flags
3393 *
3394 * Return true if the device supports the provided flags.  Automatically
3395 * filters out flags that are not implemented on multifunction devices.
3396 *
3397 * Note that this interface checks the effective ACS capabilities of the
3398 * device rather than the actual capabilities.  For instance, most single
3399 * function endpoints are not required to support ACS because they have no
3400 * opportunity for peer-to-peer access.  We therefore return 'true'
3401 * regardless of whether the device exposes an ACS capability.  This makes
3402 * it much easier for callers of this function to ignore the actual type
3403 * or topology of the device when testing ACS support.
3404 */
3405bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3406{
3407	int ret;
3408
3409	ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3410	if (ret >= 0)
3411		return ret > 0;
3412
3413	/*
3414	 * Conventional PCI and PCI-X devices never support ACS, either
3415	 * effectively or actually.  The shared bus topology implies that
3416	 * any device on the bus can receive or snoop DMA.
3417	 */
3418	if (!pci_is_pcie(pdev))
3419		return false;
3420
3421	switch (pci_pcie_type(pdev)) {
3422	/*
3423	 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3424	 * but since their primary interface is PCI/X, we conservatively
3425	 * handle them as we would a non-PCIe device.
3426	 */
3427	case PCI_EXP_TYPE_PCIE_BRIDGE:
3428	/*
3429	 * PCIe 3.0, 6.12.1 excludes ACS on these devices.  "ACS is never
3430	 * applicable... must never implement an ACS Extended Capability...".
3431	 * This seems arbitrary, but we take a conservative interpretation
3432	 * of this statement.
3433	 */
3434	case PCI_EXP_TYPE_PCI_BRIDGE:
3435	case PCI_EXP_TYPE_RC_EC:
3436		return false;
3437	/*
3438	 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3439	 * implement ACS in order to indicate their peer-to-peer capabilities,
3440	 * regardless of whether they are single- or multi-function devices.
3441	 */
3442	case PCI_EXP_TYPE_DOWNSTREAM:
3443	case PCI_EXP_TYPE_ROOT_PORT:
3444		return pci_acs_flags_enabled(pdev, acs_flags);
3445	/*
3446	 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3447	 * implemented by the remaining PCIe types to indicate peer-to-peer
3448	 * capabilities, but only when they are part of a multifunction
3449	 * device.  The footnote for section 6.12 indicates the specific
3450	 * PCIe types included here.
3451	 */
3452	case PCI_EXP_TYPE_ENDPOINT:
3453	case PCI_EXP_TYPE_UPSTREAM:
3454	case PCI_EXP_TYPE_LEG_END:
3455	case PCI_EXP_TYPE_RC_END:
3456		if (!pdev->multifunction)
3457			break;
3458
3459		return pci_acs_flags_enabled(pdev, acs_flags);
3460	}
3461
3462	/*
3463	 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3464	 * to single function devices with the exception of downstream ports.
3465	 */
3466	return true;
3467}
3468
3469/**
3470 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3471 * @start: starting downstream device
3472 * @end: ending upstream device or NULL to search to the root bus
3473 * @acs_flags: required flags
3474 *
3475 * Walk up a device tree from start to end testing PCI ACS support.  If
3476 * any step along the way does not support the required flags, return false.
3477 */
3478bool pci_acs_path_enabled(struct pci_dev *start,
3479			  struct pci_dev *end, u16 acs_flags)
3480{
3481	struct pci_dev *pdev, *parent = start;
3482
3483	do {
3484		pdev = parent;
3485
3486		if (!pci_acs_enabled(pdev, acs_flags))
3487			return false;
3488
3489		if (pci_is_root_bus(pdev->bus))
3490			return (end == NULL);
3491
3492		parent = pdev->bus->self;
3493	} while (pdev != end);
3494
3495	return true;
3496}
3497
3498/**
3499 * pci_acs_init - Initialize ACS if hardware supports it
3500 * @dev: the PCI device
3501 */
3502void pci_acs_init(struct pci_dev *dev)
3503{
3504	dev->acs_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3505
3506	if (dev->acs_cap)
3507		pci_enable_acs(dev);
3508}
3509
3510/**
3511 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3512 * @pdev: PCI device
3513 * @bar: BAR to find
3514 *
3515 * Helper to find the position of the ctrl register for a BAR.
3516 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3517 * Returns -ENOENT if no ctrl register for the BAR could be found.
3518 */
3519static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3520{
3521	unsigned int pos, nbars, i;
3522	u32 ctrl;
3523
3524	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3525	if (!pos)
3526		return -ENOTSUPP;
3527
3528	pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3529	nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3530		    PCI_REBAR_CTRL_NBAR_SHIFT;
3531
3532	for (i = 0; i < nbars; i++, pos += 8) {
3533		int bar_idx;
3534
3535		pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3536		bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3537		if (bar_idx == bar)
3538			return pos;
3539	}
3540
3541	return -ENOENT;
3542}
3543
3544/**
3545 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3546 * @pdev: PCI device
3547 * @bar: BAR to query
3548 *
3549 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3550 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3551 */
3552u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3553{
3554	int pos;
3555	u32 cap;
3556
3557	pos = pci_rebar_find_pos(pdev, bar);
3558	if (pos < 0)
3559		return 0;
3560
3561	pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3562	return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3563}
3564
3565/**
3566 * pci_rebar_get_current_size - get the current size of a BAR
3567 * @pdev: PCI device
3568 * @bar: BAR to set size to
3569 *
3570 * Read the size of a BAR from the resizable BAR config.
3571 * Returns size if found or negative error code.
3572 */
3573int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3574{
3575	int pos;
3576	u32 ctrl;
3577
3578	pos = pci_rebar_find_pos(pdev, bar);
3579	if (pos < 0)
3580		return pos;
3581
3582	pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3583	return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3584}
3585
3586/**
3587 * pci_rebar_set_size - set a new size for a BAR
3588 * @pdev: PCI device
3589 * @bar: BAR to set size to
3590 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3591 *
3592 * Set the new size of a BAR as defined in the spec.
3593 * Returns zero if resizing was successful, error code otherwise.
3594 */
3595int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3596{
3597	int pos;
3598	u32 ctrl;
3599
3600	pos = pci_rebar_find_pos(pdev, bar);
3601	if (pos < 0)
3602		return pos;
3603
3604	pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3605	ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3606	ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3607	pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3608	return 0;
3609}
3610
3611/**
3612 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3613 * @dev: the PCI device
3614 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3615 *	PCI_EXP_DEVCAP2_ATOMIC_COMP32
3616 *	PCI_EXP_DEVCAP2_ATOMIC_COMP64
3617 *	PCI_EXP_DEVCAP2_ATOMIC_COMP128
3618 *
3619 * Return 0 if all upstream bridges support AtomicOp routing, egress
3620 * blocking is disabled on all upstream ports, and the root port supports
3621 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3622 * AtomicOp completion), or negative otherwise.
3623 */
3624int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3625{
3626	struct pci_bus *bus = dev->bus;
3627	struct pci_dev *bridge;
3628	u32 cap, ctl2;
3629
3630	if (!pci_is_pcie(dev))
3631		return -EINVAL;
3632
3633	/*
3634	 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3635	 * AtomicOp requesters.  For now, we only support endpoints as
3636	 * requesters and root ports as completers.  No endpoints as
3637	 * completers, and no peer-to-peer.
3638	 */
3639
3640	switch (pci_pcie_type(dev)) {
3641	case PCI_EXP_TYPE_ENDPOINT:
3642	case PCI_EXP_TYPE_LEG_END:
3643	case PCI_EXP_TYPE_RC_END:
3644		break;
3645	default:
3646		return -EINVAL;
3647	}
3648
3649	while (bus->parent) {
3650		bridge = bus->self;
3651
3652		pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3653
3654		switch (pci_pcie_type(bridge)) {
3655		/* Ensure switch ports support AtomicOp routing */
3656		case PCI_EXP_TYPE_UPSTREAM:
3657		case PCI_EXP_TYPE_DOWNSTREAM:
3658			if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3659				return -EINVAL;
3660			break;
3661
3662		/* Ensure root port supports all the sizes we care about */
3663		case PCI_EXP_TYPE_ROOT_PORT:
3664			if ((cap & cap_mask) != cap_mask)
3665				return -EINVAL;
3666			break;
3667		}
3668
3669		/* Ensure upstream ports don't block AtomicOps on egress */
3670		if (pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM) {
3671			pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3672						   &ctl2);
3673			if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3674				return -EINVAL;
3675		}
3676
3677		bus = bus->parent;
3678	}
3679
3680	pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3681				 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3682	return 0;
3683}
3684EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3685
3686/**
3687 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3688 * @dev: the PCI device
3689 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3690 *
3691 * Perform INTx swizzling for a device behind one level of bridge.  This is
3692 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3693 * behind bridges on add-in cards.  For devices with ARI enabled, the slot
3694 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3695 * the PCI Express Base Specification, Revision 2.1)
3696 */
3697u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3698{
3699	int slot;
3700
3701	if (pci_ari_enabled(dev->bus))
3702		slot = 0;
3703	else
3704		slot = PCI_SLOT(dev->devfn);
3705
3706	return (((pin - 1) + slot) % 4) + 1;
3707}
3708
3709int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3710{
3711	u8 pin;
3712
3713	pin = dev->pin;
3714	if (!pin)
3715		return -1;
3716
3717	while (!pci_is_root_bus(dev->bus)) {
3718		pin = pci_swizzle_interrupt_pin(dev, pin);
3719		dev = dev->bus->self;
3720	}
3721	*bridge = dev;
3722	return pin;
3723}
3724
3725/**
3726 * pci_common_swizzle - swizzle INTx all the way to root bridge
3727 * @dev: the PCI device
3728 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3729 *
3730 * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
3731 * bridges all the way up to a PCI root bus.
3732 */
3733u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3734{
3735	u8 pin = *pinp;
3736
3737	while (!pci_is_root_bus(dev->bus)) {
3738		pin = pci_swizzle_interrupt_pin(dev, pin);
3739		dev = dev->bus->self;
3740	}
3741	*pinp = pin;
3742	return PCI_SLOT(dev->devfn);
3743}
3744EXPORT_SYMBOL_GPL(pci_common_swizzle);
3745
3746/**
3747 * pci_release_region - Release a PCI bar
3748 * @pdev: PCI device whose resources were previously reserved by
3749 *	  pci_request_region()
3750 * @bar: BAR to release
3751 *
3752 * Releases the PCI I/O and memory resources previously reserved by a
3753 * successful call to pci_request_region().  Call this function only
3754 * after all use of the PCI regions has ceased.
3755 */
3756void pci_release_region(struct pci_dev *pdev, int bar)
3757{
3758	struct pci_devres *dr;
3759
3760	if (pci_resource_len(pdev, bar) == 0)
3761		return;
3762	if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3763		release_region(pci_resource_start(pdev, bar),
3764				pci_resource_len(pdev, bar));
3765	else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3766		release_mem_region(pci_resource_start(pdev, bar),
3767				pci_resource_len(pdev, bar));
3768
3769	dr = find_pci_dr(pdev);
3770	if (dr)
3771		dr->region_mask &= ~(1 << bar);
3772}
3773EXPORT_SYMBOL(pci_release_region);
3774
3775/**
3776 * __pci_request_region - Reserved PCI I/O and memory resource
3777 * @pdev: PCI device whose resources are to be reserved
3778 * @bar: BAR to be reserved
3779 * @res_name: Name to be associated with resource.
3780 * @exclusive: whether the region access is exclusive or not
3781 *
3782 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3783 * being reserved by owner @res_name.  Do not access any
3784 * address inside the PCI regions unless this call returns
3785 * successfully.
3786 *
3787 * If @exclusive is set, then the region is marked so that userspace
3788 * is explicitly not allowed to map the resource via /dev/mem or
3789 * sysfs MMIO access.
3790 *
3791 * Returns 0 on success, or %EBUSY on error.  A warning
3792 * message is also printed on failure.
3793 */
3794static int __pci_request_region(struct pci_dev *pdev, int bar,
3795				const char *res_name, int exclusive)
3796{
3797	struct pci_devres *dr;
3798
3799	if (pci_resource_len(pdev, bar) == 0)
3800		return 0;
3801
3802	if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3803		if (!request_region(pci_resource_start(pdev, bar),
3804			    pci_resource_len(pdev, bar), res_name))
3805			goto err_out;
3806	} else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3807		if (!__request_mem_region(pci_resource_start(pdev, bar),
3808					pci_resource_len(pdev, bar), res_name,
3809					exclusive))
3810			goto err_out;
3811	}
3812
3813	dr = find_pci_dr(pdev);
3814	if (dr)
3815		dr->region_mask |= 1 << bar;
3816
3817	return 0;
3818
3819err_out:
3820	pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3821		 &pdev->resource[bar]);
3822	return -EBUSY;
3823}
3824
3825/**
3826 * pci_request_region - Reserve PCI I/O and memory resource
3827 * @pdev: PCI device whose resources are to be reserved
3828 * @bar: BAR to be reserved
3829 * @res_name: Name to be associated with resource
3830 *
3831 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3832 * being reserved by owner @res_name.  Do not access any
3833 * address inside the PCI regions unless this call returns
3834 * successfully.
3835 *
3836 * Returns 0 on success, or %EBUSY on error.  A warning
3837 * message is also printed on failure.
3838 */
3839int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3840{
3841	return __pci_request_region(pdev, bar, res_name, 0);
3842}
3843EXPORT_SYMBOL(pci_request_region);
3844
3845/**
3846 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3847 * @pdev: PCI device whose resources were previously reserved
3848 * @bars: Bitmask of BARs to be released
3849 *
3850 * Release selected PCI I/O and memory resources previously reserved.
3851 * Call this function only after all use of the PCI regions has ceased.
3852 */
3853void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3854{
3855	int i;
3856
3857	for (i = 0; i < PCI_STD_NUM_BARS; i++)
3858		if (bars & (1 << i))
3859			pci_release_region(pdev, i);
3860}
3861EXPORT_SYMBOL(pci_release_selected_regions);
3862
3863static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3864					  const char *res_name, int excl)
3865{
3866	int i;
3867
3868	for (i = 0; i < PCI_STD_NUM_BARS; i++)
3869		if (bars & (1 << i))
3870			if (__pci_request_region(pdev, i, res_name, excl))
3871				goto err_out;
3872	return 0;
3873
3874err_out:
3875	while (--i >= 0)
3876		if (bars & (1 << i))
3877			pci_release_region(pdev, i);
3878
3879	return -EBUSY;
3880}
3881
3882
3883/**
3884 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3885 * @pdev: PCI device whose resources are to be reserved
3886 * @bars: Bitmask of BARs to be requested
3887 * @res_name: Name to be associated with resource
3888 */
3889int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3890				 const char *res_name)
3891{
3892	return __pci_request_selected_regions(pdev, bars, res_name, 0);
3893}
3894EXPORT_SYMBOL(pci_request_selected_regions);
3895
3896int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3897					   const char *res_name)
3898{
3899	return __pci_request_selected_regions(pdev, bars, res_name,
3900			IORESOURCE_EXCLUSIVE);
3901}
3902EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3903
3904/**
3905 * pci_release_regions - Release reserved PCI I/O and memory resources
3906 * @pdev: PCI device whose resources were previously reserved by
3907 *	  pci_request_regions()
3908 *
3909 * Releases all PCI I/O and memory resources previously reserved by a
3910 * successful call to pci_request_regions().  Call this function only
3911 * after all use of the PCI regions has ceased.
3912 */
3913
3914void pci_release_regions(struct pci_dev *pdev)
3915{
3916	pci_release_selected_regions(pdev, (1 << PCI_STD_NUM_BARS) - 1);
3917}
3918EXPORT_SYMBOL(pci_release_regions);
3919
3920/**
3921 * pci_request_regions - Reserve PCI I/O and memory resources
3922 * @pdev: PCI device whose resources are to be reserved
3923 * @res_name: Name to be associated with resource.
3924 *
3925 * Mark all PCI regions associated with PCI device @pdev as
3926 * being reserved by owner @res_name.  Do not access any
3927 * address inside the PCI regions unless this call returns
3928 * successfully.
3929 *
3930 * Returns 0 on success, or %EBUSY on error.  A warning
3931 * message is also printed on failure.
3932 */
3933int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3934{
3935	return pci_request_selected_regions(pdev,
3936			((1 << PCI_STD_NUM_BARS) - 1), res_name);
3937}
3938EXPORT_SYMBOL(pci_request_regions);
3939
3940/**
3941 * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
3942 * @pdev: PCI device whose resources are to be reserved
3943 * @res_name: Name to be associated with resource.
3944 *
3945 * Mark all PCI regions associated with PCI device @pdev as being reserved
3946 * by owner @res_name.  Do not access any address inside the PCI regions
3947 * unless this call returns successfully.
3948 *
3949 * pci_request_regions_exclusive() will mark the region so that /dev/mem
3950 * and the sysfs MMIO access will not be allowed.
3951 *
3952 * Returns 0 on success, or %EBUSY on error.  A warning message is also
3953 * printed on failure.
3954 */
3955int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3956{
3957	return pci_request_selected_regions_exclusive(pdev,
3958				((1 << PCI_STD_NUM_BARS) - 1), res_name);
3959}
3960EXPORT_SYMBOL(pci_request_regions_exclusive);
3961
3962/*
3963 * Record the PCI IO range (expressed as CPU physical address + size).
3964 * Return a negative value if an error has occurred, zero otherwise
3965 */
3966int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3967			resource_size_t	size)
3968{
3969	int ret = 0;
3970#ifdef PCI_IOBASE
3971	struct logic_pio_hwaddr *range;
3972
3973	if (!size || addr + size < addr)
3974		return -EINVAL;
3975
3976	range = kzalloc(sizeof(*range), GFP_ATOMIC);
3977	if (!range)
3978		return -ENOMEM;
3979
3980	range->fwnode = fwnode;
3981	range->size = size;
3982	range->hw_start = addr;
3983	range->flags = LOGIC_PIO_CPU_MMIO;
3984
3985	ret = logic_pio_register_range(range);
3986	if (ret)
3987		kfree(range);
3988#endif
3989
3990	return ret;
3991}
3992
3993phys_addr_t pci_pio_to_address(unsigned long pio)
3994{
3995	phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3996
3997#ifdef PCI_IOBASE
3998	if (pio >= MMIO_UPPER_LIMIT)
3999		return address;
4000
4001	address = logic_pio_to_hwaddr(pio);
4002#endif
4003
4004	return address;
4005}
4006
4007unsigned long __weak pci_address_to_pio(phys_addr_t address)
4008{
4009#ifdef PCI_IOBASE
4010	return logic_pio_trans_cpuaddr(address);
4011#else
4012	if (address > IO_SPACE_LIMIT)
4013		return (unsigned long)-1;
4014
4015	return (unsigned long) address;
4016#endif
4017}
4018
4019/**
4020 * pci_remap_iospace - Remap the memory mapped I/O space
4021 * @res: Resource describing the I/O space
4022 * @phys_addr: physical address of range to be mapped
4023 *
4024 * Remap the memory mapped I/O space described by the @res and the CPU
4025 * physical address @phys_addr into virtual address space.  Only
4026 * architectures that have memory mapped IO functions defined (and the
4027 * PCI_IOBASE value defined) should call this function.
4028 */
4029int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
4030{
4031#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4032	unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4033
4034	if (!(res->flags & IORESOURCE_IO))
4035		return -EINVAL;
4036
4037	if (res->end > IO_SPACE_LIMIT)
4038		return -EINVAL;
4039
4040	return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
4041				  pgprot_device(PAGE_KERNEL));
4042#else
4043	/*
4044	 * This architecture does not have memory mapped I/O space,
4045	 * so this function should never be called
4046	 */
4047	WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
4048	return -ENODEV;
4049#endif
4050}
4051EXPORT_SYMBOL(pci_remap_iospace);
4052
4053/**
4054 * pci_unmap_iospace - Unmap the memory mapped I/O space
4055 * @res: resource to be unmapped
4056 *
4057 * Unmap the CPU virtual address @res from virtual address space.  Only
4058 * architectures that have memory mapped IO functions defined (and the
4059 * PCI_IOBASE value defined) should call this function.
4060 */
4061void pci_unmap_iospace(struct resource *res)
4062{
4063#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4064	unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4065
4066	unmap_kernel_range(vaddr, resource_size(res));
4067#endif
4068}
4069EXPORT_SYMBOL(pci_unmap_iospace);
4070
4071static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
4072{
4073	struct resource **res = ptr;
4074
4075	pci_unmap_iospace(*res);
4076}
4077
4078/**
4079 * devm_pci_remap_iospace - Managed pci_remap_iospace()
4080 * @dev: Generic device to remap IO address for
4081 * @res: Resource describing the I/O space
4082 * @phys_addr: physical address of range to be mapped
4083 *
4084 * Managed pci_remap_iospace().  Map is automatically unmapped on driver
4085 * detach.
4086 */
4087int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
4088			   phys_addr_t phys_addr)
4089{
4090	const struct resource **ptr;
4091	int error;
4092
4093	ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
4094	if (!ptr)
4095		return -ENOMEM;
4096
4097	error = pci_remap_iospace(res, phys_addr);
4098	if (error) {
4099		devres_free(ptr);
4100	} else	{
4101		*ptr = res;
4102		devres_add(dev, ptr);
4103	}
4104
4105	return error;
4106}
4107EXPORT_SYMBOL(devm_pci_remap_iospace);
4108
4109/**
4110 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
4111 * @dev: Generic device to remap IO address for
4112 * @offset: Resource address to map
4113 * @size: Size of map
4114 *
4115 * Managed pci_remap_cfgspace().  Map is automatically unmapped on driver
4116 * detach.
4117 */
4118void __iomem *devm_pci_remap_cfgspace(struct device *dev,
4119				      resource_size_t offset,
4120				      resource_size_t size)
4121{
4122	void __iomem **ptr, *addr;
4123
4124	ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
4125	if (!ptr)
4126		return NULL;
4127
4128	addr = pci_remap_cfgspace(offset, size);
4129	if (addr) {
4130		*ptr = addr;
4131		devres_add(dev, ptr);
4132	} else
4133		devres_free(ptr);
4134
4135	return addr;
4136}
4137EXPORT_SYMBOL(devm_pci_remap_cfgspace);
4138
4139/**
4140 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
4141 * @dev: generic device to handle the resource for
4142 * @res: configuration space resource to be handled
4143 *
4144 * Checks that a resource is a valid memory region, requests the memory
4145 * region and ioremaps with pci_remap_cfgspace() API that ensures the
4146 * proper PCI configuration space memory attributes are guaranteed.
4147 *
4148 * All operations are managed and will be undone on driver detach.
4149 *
4150 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
4151 * on failure. Usage example::
4152 *
4153 *	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
4154 *	base = devm_pci_remap_cfg_resource(&pdev->dev, res);
4155 *	if (IS_ERR(base))
4156 *		return PTR_ERR(base);
4157 */
4158void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
4159					  struct resource *res)
4160{
4161	resource_size_t size;
4162	const char *name;
4163	void __iomem *dest_ptr;
4164
4165	BUG_ON(!dev);
4166
4167	if (!res || resource_type(res) != IORESOURCE_MEM) {
4168		dev_err(dev, "invalid resource\n");
4169		return IOMEM_ERR_PTR(-EINVAL);
4170	}
4171
4172	size = resource_size(res);
4173	name = res->name ?: dev_name(dev);
4174
4175	if (!devm_request_mem_region(dev, res->start, size, name)) {
4176		dev_err(dev, "can't request region for resource %pR\n", res);
4177		return IOMEM_ERR_PTR(-EBUSY);
4178	}
4179
4180	dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4181	if (!dest_ptr) {
4182		dev_err(dev, "ioremap failed for resource %pR\n", res);
4183		devm_release_mem_region(dev, res->start, size);
4184		dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4185	}
4186
4187	return dest_ptr;
4188}
4189EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4190
4191static void __pci_set_master(struct pci_dev *dev, bool enable)
4192{
4193	u16 old_cmd, cmd;
4194
4195	pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4196	if (enable)
4197		cmd = old_cmd | PCI_COMMAND_MASTER;
4198	else
4199		cmd = old_cmd & ~PCI_COMMAND_MASTER;
4200	if (cmd != old_cmd) {
4201		pci_dbg(dev, "%s bus mastering\n",
4202			enable ? "enabling" : "disabling");
4203		pci_write_config_word(dev, PCI_COMMAND, cmd);
4204	}
4205	dev->is_busmaster = enable;
4206}
4207
4208/**
4209 * pcibios_setup - process "pci=" kernel boot arguments
4210 * @str: string used to pass in "pci=" kernel boot arguments
4211 *
4212 * Process kernel boot arguments.  This is the default implementation.
4213 * Architecture specific implementations can override this as necessary.
4214 */
4215char * __weak __init pcibios_setup(char *str)
4216{
4217	return str;
4218}
4219
4220/**
4221 * pcibios_set_master - enable PCI bus-mastering for device dev
4222 * @dev: the PCI device to enable
4223 *
4224 * Enables PCI bus-mastering for the device.  This is the default
4225 * implementation.  Architecture specific implementations can override
4226 * this if necessary.
4227 */
4228void __weak pcibios_set_master(struct pci_dev *dev)
4229{
4230	u8 lat;
4231
4232	/* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4233	if (pci_is_pcie(dev))
4234		return;
4235
4236	pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4237	if (lat < 16)
4238		lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4239	else if (lat > pcibios_max_latency)
4240		lat = pcibios_max_latency;
4241	else
4242		return;
4243
4244	pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4245}
4246
4247/**
4248 * pci_set_master - enables bus-mastering for device dev
4249 * @dev: the PCI device to enable
4250 *
4251 * Enables bus-mastering on the device and calls pcibios_set_master()
4252 * to do the needed arch specific settings.
4253 */
4254void pci_set_master(struct pci_dev *dev)
4255{
4256	__pci_set_master(dev, true);
4257	pcibios_set_master(dev);
4258}
4259EXPORT_SYMBOL(pci_set_master);
4260
4261/**
4262 * pci_clear_master - disables bus-mastering for device dev
4263 * @dev: the PCI device to disable
4264 */
4265void pci_clear_master(struct pci_dev *dev)
4266{
4267	__pci_set_master(dev, false);
4268}
4269EXPORT_SYMBOL(pci_clear_master);
4270
4271/**
4272 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4273 * @dev: the PCI device for which MWI is to be enabled
4274 *
4275 * Helper function for pci_set_mwi.
4276 * Originally copied from drivers/net/acenic.c.
4277 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4278 *
4279 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4280 */
4281int pci_set_cacheline_size(struct pci_dev *dev)
4282{
4283	u8 cacheline_size;
4284
4285	if (!pci_cache_line_size)
4286		return -EINVAL;
4287
4288	/* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4289	   equal to or multiple of the right value. */
4290	pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4291	if (cacheline_size >= pci_cache_line_size &&
4292	    (cacheline_size % pci_cache_line_size) == 0)
4293		return 0;
4294
4295	/* Write the correct value. */
4296	pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4297	/* Read it back. */
4298	pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4299	if (cacheline_size == pci_cache_line_size)
4300		return 0;
4301
4302	pci_info(dev, "cache line size of %d is not supported\n",
4303		   pci_cache_line_size << 2);
4304
4305	return -EINVAL;
4306}
4307EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4308
4309/**
4310 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4311 * @dev: the PCI device for which MWI is enabled
4312 *
4313 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4314 *
4315 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4316 */
4317int pci_set_mwi(struct pci_dev *dev)
4318{
4319#ifdef PCI_DISABLE_MWI
4320	return 0;
4321#else
4322	int rc;
4323	u16 cmd;
4324
4325	rc = pci_set_cacheline_size(dev);
4326	if (rc)
4327		return rc;
4328
4329	pci_read_config_word(dev, PCI_COMMAND, &cmd);
4330	if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4331		pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4332		cmd |= PCI_COMMAND_INVALIDATE;
4333		pci_write_config_word(dev, PCI_COMMAND, cmd);
4334	}
4335	return 0;
4336#endif
4337}
4338EXPORT_SYMBOL(pci_set_mwi);
4339
4340/**
4341 * pcim_set_mwi - a device-managed pci_set_mwi()
4342 * @dev: the PCI device for which MWI is enabled
4343 *
4344 * Managed pci_set_mwi().
4345 *
4346 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4347 */
4348int pcim_set_mwi(struct pci_dev *dev)
4349{
4350	struct pci_devres *dr;
4351
4352	dr = find_pci_dr(dev);
4353	if (!dr)
4354		return -ENOMEM;
4355
4356	dr->mwi = 1;
4357	return pci_set_mwi(dev);
4358}
4359EXPORT_SYMBOL(pcim_set_mwi);
4360
4361/**
4362 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4363 * @dev: the PCI device for which MWI is enabled
4364 *
4365 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4366 * Callers are not required to check the return value.
4367 *
4368 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4369 */
4370int pci_try_set_mwi(struct pci_dev *dev)
4371{
4372#ifdef PCI_DISABLE_MWI
4373	return 0;
4374#else
4375	return pci_set_mwi(dev);
4376#endif
4377}
4378EXPORT_SYMBOL(pci_try_set_mwi);
4379
4380/**
4381 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4382 * @dev: the PCI device to disable
4383 *
4384 * Disables PCI Memory-Write-Invalidate transaction on the device
4385 */
4386void pci_clear_mwi(struct pci_dev *dev)
4387{
4388#ifndef PCI_DISABLE_MWI
4389	u16 cmd;
4390
4391	pci_read_config_word(dev, PCI_COMMAND, &cmd);
4392	if (cmd & PCI_COMMAND_INVALIDATE) {
4393		cmd &= ~PCI_COMMAND_INVALIDATE;
4394		pci_write_config_word(dev, PCI_COMMAND, cmd);
4395	}
4396#endif
4397}
4398EXPORT_SYMBOL(pci_clear_mwi);
4399
4400/**
4401 * pci_intx - enables/disables PCI INTx for device dev
4402 * @pdev: the PCI device to operate on
4403 * @enable: boolean: whether to enable or disable PCI INTx
4404 *
4405 * Enables/disables PCI INTx for device @pdev
4406 */
4407void pci_intx(struct pci_dev *pdev, int enable)
4408{
4409	u16 pci_command, new;
4410
4411	pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4412
4413	if (enable)
4414		new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4415	else
4416		new = pci_command | PCI_COMMAND_INTX_DISABLE;
4417
4418	if (new != pci_command) {
4419		struct pci_devres *dr;
4420
4421		pci_write_config_word(pdev, PCI_COMMAND, new);
4422
4423		dr = find_pci_dr(pdev);
4424		if (dr && !dr->restore_intx) {
4425			dr->restore_intx = 1;
4426			dr->orig_intx = !enable;
4427		}
4428	}
4429}
4430EXPORT_SYMBOL_GPL(pci_intx);
4431
4432static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4433{
4434	struct pci_bus *bus = dev->bus;
4435	bool mask_updated = true;
4436	u32 cmd_status_dword;
4437	u16 origcmd, newcmd;
4438	unsigned long flags;
4439	bool irq_pending;
4440
4441	/*
4442	 * We do a single dword read to retrieve both command and status.
4443	 * Document assumptions that make this possible.
4444	 */
4445	BUILD_BUG_ON(PCI_COMMAND % 4);
4446	BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4447
4448	raw_spin_lock_irqsave(&pci_lock, flags);
4449
4450	bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4451
4452	irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4453
4454	/*
4455	 * Check interrupt status register to see whether our device
4456	 * triggered the interrupt (when masking) or the next IRQ is
4457	 * already pending (when unmasking).
4458	 */
4459	if (mask != irq_pending) {
4460		mask_updated = false;
4461		goto done;
4462	}
4463
4464	origcmd = cmd_status_dword;
4465	newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4466	if (mask)
4467		newcmd |= PCI_COMMAND_INTX_DISABLE;
4468	if (newcmd != origcmd)
4469		bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4470
4471done:
4472	raw_spin_unlock_irqrestore(&pci_lock, flags);
4473
4474	return mask_updated;
4475}
4476
4477/**
4478 * pci_check_and_mask_intx - mask INTx on pending interrupt
4479 * @dev: the PCI device to operate on
4480 *
4481 * Check if the device dev has its INTx line asserted, mask it and return
4482 * true in that case. False is returned if no interrupt was pending.
4483 */
4484bool pci_check_and_mask_intx(struct pci_dev *dev)
4485{
4486	return pci_check_and_set_intx_mask(dev, true);
4487}
4488EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4489
4490/**
4491 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4492 * @dev: the PCI device to operate on
4493 *
4494 * Check if the device dev has its INTx line asserted, unmask it if not and
4495 * return true. False is returned and the mask remains active if there was
4496 * still an interrupt pending.
4497 */
4498bool pci_check_and_unmask_intx(struct pci_dev *dev)
4499{
4500	return pci_check_and_set_intx_mask(dev, false);
4501}
4502EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4503
4504/**
4505 * pci_wait_for_pending_transaction - wait for pending transaction
4506 * @dev: the PCI device to operate on
4507 *
4508 * Return 0 if transaction is pending 1 otherwise.
4509 */
4510int pci_wait_for_pending_transaction(struct pci_dev *dev)
4511{
4512	if (!pci_is_pcie(dev))
4513		return 1;
4514
4515	return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4516				    PCI_EXP_DEVSTA_TRPND);
4517}
4518EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4519
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4520/**
4521 * pcie_has_flr - check if a device supports function level resets
4522 * @dev: device to check
4523 *
4524 * Returns true if the device advertises support for PCIe function level
4525 * resets.
4526 */
4527bool pcie_has_flr(struct pci_dev *dev)
4528{
4529	u32 cap;
4530
4531	if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4532		return false;
4533
4534	pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4535	return cap & PCI_EXP_DEVCAP_FLR;
4536}
4537EXPORT_SYMBOL_GPL(pcie_has_flr);
4538
4539/**
4540 * pcie_flr - initiate a PCIe function level reset
4541 * @dev: device to reset
4542 *
4543 * Initiate a function level reset on @dev.  The caller should ensure the
4544 * device supports FLR before calling this function, e.g. by using the
4545 * pcie_has_flr() helper.
4546 */
4547int pcie_flr(struct pci_dev *dev)
4548{
4549	if (!pci_wait_for_pending_transaction(dev))
4550		pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4551
4552	pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4553
4554	if (dev->imm_ready)
4555		return 0;
4556
4557	/*
4558	 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4559	 * 100ms, but may silently discard requests while the FLR is in
4560	 * progress.  Wait 100ms before trying to access the device.
4561	 */
4562	msleep(100);
4563
4564	return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4565}
4566EXPORT_SYMBOL_GPL(pcie_flr);
4567
4568static int pci_af_flr(struct pci_dev *dev, int probe)
4569{
4570	int pos;
4571	u8 cap;
4572
4573	pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4574	if (!pos)
4575		return -ENOTTY;
4576
4577	if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4578		return -ENOTTY;
4579
4580	pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4581	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4582		return -ENOTTY;
4583
4584	if (probe)
4585		return 0;
4586
4587	/*
4588	 * Wait for Transaction Pending bit to clear.  A word-aligned test
4589	 * is used, so we use the control offset rather than status and shift
4590	 * the test bit to match.
4591	 */
4592	if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4593				 PCI_AF_STATUS_TP << 8))
4594		pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4595
4596	pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4597
4598	if (dev->imm_ready)
4599		return 0;
4600
4601	/*
4602	 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4603	 * updated 27 July 2006; a device must complete an FLR within
4604	 * 100ms, but may silently discard requests while the FLR is in
4605	 * progress.  Wait 100ms before trying to access the device.
4606	 */
4607	msleep(100);
4608
4609	return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4610}
4611
4612/**
4613 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4614 * @dev: Device to reset.
4615 * @probe: If set, only check if the device can be reset this way.
4616 *
4617 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4618 * unset, it will be reinitialized internally when going from PCI_D3hot to
4619 * PCI_D0.  If that's the case and the device is not in a low-power state
4620 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4621 *
4622 * NOTE: This causes the caller to sleep for twice the device power transition
4623 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4624 * by default (i.e. unless the @dev's d3_delay field has a different value).
4625 * Moreover, only devices in D0 can be reset by this function.
4626 */
4627static int pci_pm_reset(struct pci_dev *dev, int probe)
4628{
4629	u16 csr;
4630
4631	if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4632		return -ENOTTY;
4633
4634	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4635	if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4636		return -ENOTTY;
4637
4638	if (probe)
4639		return 0;
4640
4641	if (dev->current_state != PCI_D0)
4642		return -EINVAL;
4643
4644	csr &= ~PCI_PM_CTRL_STATE_MASK;
4645	csr |= PCI_D3hot;
4646	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4647	pci_dev_d3_sleep(dev);
4648
4649	csr &= ~PCI_PM_CTRL_STATE_MASK;
4650	csr |= PCI_D0;
4651	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4652	pci_dev_d3_sleep(dev);
4653
4654	return pci_dev_wait(dev, "PM D3hot->D0", PCIE_RESET_READY_POLL_MS);
4655}
4656
4657/**
4658 * pcie_wait_for_link_delay - Wait until link is active or inactive
4659 * @pdev: Bridge device
4660 * @active: waiting for active or inactive?
4661 * @delay: Delay to wait after link has become active (in ms)
4662 *
4663 * Use this to wait till link becomes active or inactive.
4664 */
4665static bool pcie_wait_for_link_delay(struct pci_dev *pdev, bool active,
4666				     int delay)
4667{
4668	int timeout = 1000;
4669	bool ret;
4670	u16 lnk_status;
4671
4672	/*
4673	 * Some controllers might not implement link active reporting. In this
4674	 * case, we wait for 1000 ms + any delay requested by the caller.
4675	 */
4676	if (!pdev->link_active_reporting) {
4677		msleep(timeout + delay);
4678		return true;
4679	}
4680
4681	/*
4682	 * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4683	 * after which we should expect an link active if the reset was
4684	 * successful. If so, software must wait a minimum 100ms before sending
4685	 * configuration requests to devices downstream this port.
4686	 *
4687	 * If the link fails to activate, either the device was physically
4688	 * removed or the link is permanently failed.
4689	 */
4690	if (active)
4691		msleep(20);
4692	for (;;) {
4693		pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4694		ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4695		if (ret == active)
4696			break;
4697		if (timeout <= 0)
4698			break;
4699		msleep(10);
4700		timeout -= 10;
4701	}
4702	if (active && ret)
4703		msleep(delay);
4704	else if (ret != active)
4705		pci_info(pdev, "Data Link Layer Link Active not %s in 1000 msec\n",
4706			active ? "set" : "cleared");
4707	return ret == active;
4708}
4709
4710/**
4711 * pcie_wait_for_link - Wait until link is active or inactive
4712 * @pdev: Bridge device
4713 * @active: waiting for active or inactive?
4714 *
4715 * Use this to wait till link becomes active or inactive.
4716 */
4717bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4718{
4719	return pcie_wait_for_link_delay(pdev, active, 100);
4720}
4721
4722/*
4723 * Find maximum D3cold delay required by all the devices on the bus.  The
4724 * spec says 100 ms, but firmware can lower it and we allow drivers to
4725 * increase it as well.
4726 *
4727 * Called with @pci_bus_sem locked for reading.
4728 */
4729static int pci_bus_max_d3cold_delay(const struct pci_bus *bus)
4730{
4731	const struct pci_dev *pdev;
4732	int min_delay = 100;
4733	int max_delay = 0;
4734
4735	list_for_each_entry(pdev, &bus->devices, bus_list) {
4736		if (pdev->d3cold_delay < min_delay)
4737			min_delay = pdev->d3cold_delay;
4738		if (pdev->d3cold_delay > max_delay)
4739			max_delay = pdev->d3cold_delay;
4740	}
4741
4742	return max(min_delay, max_delay);
4743}
4744
4745/**
4746 * pci_bridge_wait_for_secondary_bus - Wait for secondary bus to be accessible
4747 * @dev: PCI bridge
4748 *
4749 * Handle necessary delays before access to the devices on the secondary
4750 * side of the bridge are permitted after D3cold to D0 transition.
4751 *
4752 * For PCIe this means the delays in PCIe 5.0 section 6.6.1. For
4753 * conventional PCI it means Tpvrh + Trhfa specified in PCI 3.0 section
4754 * 4.3.2.
4755 */
4756void pci_bridge_wait_for_secondary_bus(struct pci_dev *dev)
4757{
4758	struct pci_dev *child;
4759	int delay;
4760
4761	if (pci_dev_is_disconnected(dev))
4762		return;
4763
4764	if (!pci_is_bridge(dev) || !dev->bridge_d3)
4765		return;
4766
4767	down_read(&pci_bus_sem);
4768
4769	/*
4770	 * We only deal with devices that are present currently on the bus.
4771	 * For any hot-added devices the access delay is handled in pciehp
4772	 * board_added(). In case of ACPI hotplug the firmware is expected
4773	 * to configure the devices before OS is notified.
4774	 */
4775	if (!dev->subordinate || list_empty(&dev->subordinate->devices)) {
4776		up_read(&pci_bus_sem);
4777		return;
4778	}
4779
4780	/* Take d3cold_delay requirements into account */
4781	delay = pci_bus_max_d3cold_delay(dev->subordinate);
4782	if (!delay) {
4783		up_read(&pci_bus_sem);
4784		return;
4785	}
4786
4787	child = list_first_entry(&dev->subordinate->devices, struct pci_dev,
4788				 bus_list);
4789	up_read(&pci_bus_sem);
4790
4791	/*
4792	 * Conventional PCI and PCI-X we need to wait Tpvrh + Trhfa before
4793	 * accessing the device after reset (that is 1000 ms + 100 ms). In
4794	 * practice this should not be needed because we don't do power
4795	 * management for them (see pci_bridge_d3_possible()).
4796	 */
4797	if (!pci_is_pcie(dev)) {
4798		pci_dbg(dev, "waiting %d ms for secondary bus\n", 1000 + delay);
4799		msleep(1000 + delay);
4800		return;
4801	}
4802
4803	/*
4804	 * For PCIe downstream and root ports that do not support speeds
4805	 * greater than 5 GT/s need to wait minimum 100 ms. For higher
4806	 * speeds (gen3) we need to wait first for the data link layer to
4807	 * become active.
4808	 *
4809	 * However, 100 ms is the minimum and the PCIe spec says the
4810	 * software must allow at least 1s before it can determine that the
4811	 * device that did not respond is a broken device. There is
4812	 * evidence that 100 ms is not always enough, for example certain
4813	 * Titan Ridge xHCI controller does not always respond to
4814	 * configuration requests if we only wait for 100 ms (see
4815	 * https://bugzilla.kernel.org/show_bug.cgi?id=203885).
4816	 *
4817	 * Therefore we wait for 100 ms and check for the device presence.
4818	 * If it is still not present give it an additional 100 ms.
4819	 */
4820	if (!pcie_downstream_port(dev))
4821		return;
4822
4823	if (pcie_get_speed_cap(dev) <= PCIE_SPEED_5_0GT) {
4824		pci_dbg(dev, "waiting %d ms for downstream link\n", delay);
4825		msleep(delay);
4826	} else {
4827		pci_dbg(dev, "waiting %d ms for downstream link, after activation\n",
4828			delay);
4829		if (!pcie_wait_for_link_delay(dev, true, delay)) {
4830			/* Did not train, no need to wait any further */
4831			return;
4832		}
4833	}
4834
4835	if (!pci_device_is_present(child)) {
4836		pci_dbg(child, "waiting additional %d ms to become accessible\n", delay);
4837		msleep(delay);
4838	}
4839}
4840
4841void pci_reset_secondary_bus(struct pci_dev *dev)
4842{
4843	u16 ctrl;
4844
4845	pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4846	ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4847	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4848
4849	/*
4850	 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms.  Double
4851	 * this to 2ms to ensure that we meet the minimum requirement.
4852	 */
4853	msleep(2);
4854
4855	ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4856	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4857
4858	/*
4859	 * Trhfa for conventional PCI is 2^25 clock cycles.
4860	 * Assuming a minimum 33MHz clock this results in a 1s
4861	 * delay before we can consider subordinate devices to
4862	 * be re-initialized.  PCIe has some ways to shorten this,
4863	 * but we don't make use of them yet.
4864	 */
4865	ssleep(1);
4866}
4867
4868void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4869{
4870	pci_reset_secondary_bus(dev);
4871}
4872
4873/**
4874 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4875 * @dev: Bridge device
4876 *
4877 * Use the bridge control register to assert reset on the secondary bus.
4878 * Devices on the secondary bus are left in power-on state.
4879 */
4880int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4881{
4882	pcibios_reset_secondary_bus(dev);
4883
4884	return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4885}
4886EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4887
4888static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4889{
4890	struct pci_dev *pdev;
4891
4892	if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4893	    !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4894		return -ENOTTY;
4895
4896	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4897		if (pdev != dev)
4898			return -ENOTTY;
4899
4900	if (probe)
4901		return 0;
4902
4903	return pci_bridge_secondary_bus_reset(dev->bus->self);
4904}
4905
4906static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4907{
4908	int rc = -ENOTTY;
4909
4910	if (!hotplug || !try_module_get(hotplug->owner))
4911		return rc;
4912
4913	if (hotplug->ops->reset_slot)
4914		rc = hotplug->ops->reset_slot(hotplug, probe);
4915
4916	module_put(hotplug->owner);
4917
4918	return rc;
4919}
4920
4921static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4922{
4923	struct pci_dev *pdev;
4924
4925	if (dev->subordinate || !dev->slot ||
4926	    dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4927		return -ENOTTY;
4928
4929	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4930		if (pdev != dev && pdev->slot == dev->slot)
4931			return -ENOTTY;
4932
4933	return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4934}
4935
4936static void pci_dev_lock(struct pci_dev *dev)
4937{
4938	pci_cfg_access_lock(dev);
4939	/* block PM suspend, driver probe, etc. */
4940	device_lock(&dev->dev);
4941}
4942
4943/* Return 1 on successful lock, 0 on contention */
4944static int pci_dev_trylock(struct pci_dev *dev)
4945{
4946	if (pci_cfg_access_trylock(dev)) {
4947		if (device_trylock(&dev->dev))
4948			return 1;
4949		pci_cfg_access_unlock(dev);
4950	}
4951
4952	return 0;
4953}
4954
4955static void pci_dev_unlock(struct pci_dev *dev)
4956{
4957	device_unlock(&dev->dev);
4958	pci_cfg_access_unlock(dev);
4959}
4960
4961static void pci_dev_save_and_disable(struct pci_dev *dev)
4962{
4963	const struct pci_error_handlers *err_handler =
4964			dev->driver ? dev->driver->err_handler : NULL;
4965
4966	/*
4967	 * dev->driver->err_handler->reset_prepare() is protected against
4968	 * races with ->remove() by the device lock, which must be held by
4969	 * the caller.
4970	 */
4971	if (err_handler && err_handler->reset_prepare)
4972		err_handler->reset_prepare(dev);
4973
4974	/*
4975	 * Wake-up device prior to save.  PM registers default to D0 after
4976	 * reset and a simple register restore doesn't reliably return
4977	 * to a non-D0 state anyway.
4978	 */
4979	pci_set_power_state(dev, PCI_D0);
4980
4981	pci_save_state(dev);
4982	/*
4983	 * Disable the device by clearing the Command register, except for
4984	 * INTx-disable which is set.  This not only disables MMIO and I/O port
4985	 * BARs, but also prevents the device from being Bus Master, preventing
4986	 * DMA from the device including MSI/MSI-X interrupts.  For PCI 2.3
4987	 * compliant devices, INTx-disable prevents legacy interrupts.
4988	 */
4989	pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4990}
4991
4992static void pci_dev_restore(struct pci_dev *dev)
4993{
4994	const struct pci_error_handlers *err_handler =
4995			dev->driver ? dev->driver->err_handler : NULL;
4996
4997	pci_restore_state(dev);
4998
4999	/*
5000	 * dev->driver->err_handler->reset_done() is protected against
5001	 * races with ->remove() by the device lock, which must be held by
5002	 * the caller.
5003	 */
5004	if (err_handler && err_handler->reset_done)
5005		err_handler->reset_done(dev);
5006}
5007
5008/**
5009 * __pci_reset_function_locked - reset a PCI device function while holding
5010 * the @dev mutex lock.
5011 * @dev: PCI device to reset
5012 *
5013 * Some devices allow an individual function to be reset without affecting
5014 * other functions in the same device.  The PCI device must be responsive
5015 * to PCI config space in order to use this function.
5016 *
5017 * The device function is presumed to be unused and the caller is holding
5018 * the device mutex lock when this function is called.
5019 *
5020 * Resetting the device will make the contents of PCI configuration space
5021 * random, so any caller of this must be prepared to reinitialise the
5022 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
5023 * etc.
5024 *
5025 * Returns 0 if the device function was successfully reset or negative if the
5026 * device doesn't support resetting a single function.
5027 */
5028int __pci_reset_function_locked(struct pci_dev *dev)
5029{
5030	int rc;
5031
5032	might_sleep();
5033
5034	/*
5035	 * A reset method returns -ENOTTY if it doesn't support this device
5036	 * and we should try the next method.
5037	 *
5038	 * If it returns 0 (success), we're finished.  If it returns any
5039	 * other error, we're also finished: this indicates that further
5040	 * reset mechanisms might be broken on the device.
5041	 */
5042	rc = pci_dev_specific_reset(dev, 0);
5043	if (rc != -ENOTTY)
5044		return rc;
5045	if (pcie_has_flr(dev)) {
5046		rc = pcie_flr(dev);
5047		if (rc != -ENOTTY)
5048			return rc;
5049	}
5050	rc = pci_af_flr(dev, 0);
5051	if (rc != -ENOTTY)
5052		return rc;
5053	rc = pci_pm_reset(dev, 0);
5054	if (rc != -ENOTTY)
5055		return rc;
5056	rc = pci_dev_reset_slot_function(dev, 0);
5057	if (rc != -ENOTTY)
5058		return rc;
5059	return pci_parent_bus_reset(dev, 0);
5060}
5061EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
5062
5063/**
5064 * pci_probe_reset_function - check whether the device can be safely reset
5065 * @dev: PCI device to reset
5066 *
5067 * Some devices allow an individual function to be reset without affecting
5068 * other functions in the same device.  The PCI device must be responsive
5069 * to PCI config space in order to use this function.
5070 *
5071 * Returns 0 if the device function can be reset or negative if the
5072 * device doesn't support resetting a single function.
5073 */
5074int pci_probe_reset_function(struct pci_dev *dev)
5075{
5076	int rc;
5077
5078	might_sleep();
5079
5080	rc = pci_dev_specific_reset(dev, 1);
5081	if (rc != -ENOTTY)
5082		return rc;
5083	if (pcie_has_flr(dev))
5084		return 0;
5085	rc = pci_af_flr(dev, 1);
5086	if (rc != -ENOTTY)
5087		return rc;
5088	rc = pci_pm_reset(dev, 1);
5089	if (rc != -ENOTTY)
5090		return rc;
5091	rc = pci_dev_reset_slot_function(dev, 1);
5092	if (rc != -ENOTTY)
5093		return rc;
5094
5095	return pci_parent_bus_reset(dev, 1);
5096}
5097
5098/**
5099 * pci_reset_function - quiesce and reset a PCI device function
5100 * @dev: PCI device to reset
5101 *
5102 * Some devices allow an individual function to be reset without affecting
5103 * other functions in the same device.  The PCI device must be responsive
5104 * to PCI config space in order to use this function.
5105 *
5106 * This function does not just reset the PCI portion of a device, but
5107 * clears all the state associated with the device.  This function differs
5108 * from __pci_reset_function_locked() in that it saves and restores device state
5109 * over the reset and takes the PCI device lock.
5110 *
5111 * Returns 0 if the device function was successfully reset or negative if the
5112 * device doesn't support resetting a single function.
5113 */
5114int pci_reset_function(struct pci_dev *dev)
5115{
5116	int rc;
5117
5118	if (!dev->reset_fn)
5119		return -ENOTTY;
5120
5121	pci_dev_lock(dev);
5122	pci_dev_save_and_disable(dev);
5123
5124	rc = __pci_reset_function_locked(dev);
5125
5126	pci_dev_restore(dev);
5127	pci_dev_unlock(dev);
5128
5129	return rc;
5130}
5131EXPORT_SYMBOL_GPL(pci_reset_function);
5132
5133/**
5134 * pci_reset_function_locked - quiesce and reset a PCI device function
5135 * @dev: PCI device to reset
5136 *
5137 * Some devices allow an individual function to be reset without affecting
5138 * other functions in the same device.  The PCI device must be responsive
5139 * to PCI config space in order to use this function.
5140 *
5141 * This function does not just reset the PCI portion of a device, but
5142 * clears all the state associated with the device.  This function differs
5143 * from __pci_reset_function_locked() in that it saves and restores device state
5144 * over the reset.  It also differs from pci_reset_function() in that it
5145 * requires the PCI device lock to be held.
5146 *
5147 * Returns 0 if the device function was successfully reset or negative if the
5148 * device doesn't support resetting a single function.
5149 */
5150int pci_reset_function_locked(struct pci_dev *dev)
5151{
5152	int rc;
5153
5154	if (!dev->reset_fn)
5155		return -ENOTTY;
5156
5157	pci_dev_save_and_disable(dev);
5158
5159	rc = __pci_reset_function_locked(dev);
5160
5161	pci_dev_restore(dev);
5162
5163	return rc;
5164}
5165EXPORT_SYMBOL_GPL(pci_reset_function_locked);
5166
5167/**
5168 * pci_try_reset_function - quiesce and reset a PCI device function
5169 * @dev: PCI device to reset
5170 *
5171 * Same as above, except return -EAGAIN if unable to lock device.
5172 */
5173int pci_try_reset_function(struct pci_dev *dev)
5174{
5175	int rc;
5176
5177	if (!dev->reset_fn)
5178		return -ENOTTY;
5179
5180	if (!pci_dev_trylock(dev))
5181		return -EAGAIN;
5182
5183	pci_dev_save_and_disable(dev);
5184	rc = __pci_reset_function_locked(dev);
5185	pci_dev_restore(dev);
5186	pci_dev_unlock(dev);
5187
5188	return rc;
5189}
5190EXPORT_SYMBOL_GPL(pci_try_reset_function);
5191
5192/* Do any devices on or below this bus prevent a bus reset? */
5193static bool pci_bus_resetable(struct pci_bus *bus)
5194{
5195	struct pci_dev *dev;
5196
5197
5198	if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5199		return false;
5200
5201	list_for_each_entry(dev, &bus->devices, bus_list) {
5202		if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5203		    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5204			return false;
5205	}
5206
5207	return true;
5208}
5209
5210/* Lock devices from the top of the tree down */
5211static void pci_bus_lock(struct pci_bus *bus)
5212{
5213	struct pci_dev *dev;
5214
5215	list_for_each_entry(dev, &bus->devices, bus_list) {
5216		pci_dev_lock(dev);
5217		if (dev->subordinate)
5218			pci_bus_lock(dev->subordinate);
5219	}
5220}
5221
5222/* Unlock devices from the bottom of the tree up */
5223static void pci_bus_unlock(struct pci_bus *bus)
5224{
5225	struct pci_dev *dev;
5226
5227	list_for_each_entry(dev, &bus->devices, bus_list) {
5228		if (dev->subordinate)
5229			pci_bus_unlock(dev->subordinate);
5230		pci_dev_unlock(dev);
5231	}
5232}
5233
5234/* Return 1 on successful lock, 0 on contention */
5235static int pci_bus_trylock(struct pci_bus *bus)
5236{
5237	struct pci_dev *dev;
5238
5239	list_for_each_entry(dev, &bus->devices, bus_list) {
5240		if (!pci_dev_trylock(dev))
5241			goto unlock;
5242		if (dev->subordinate) {
5243			if (!pci_bus_trylock(dev->subordinate)) {
5244				pci_dev_unlock(dev);
5245				goto unlock;
5246			}
5247		}
5248	}
5249	return 1;
5250
5251unlock:
5252	list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5253		if (dev->subordinate)
5254			pci_bus_unlock(dev->subordinate);
5255		pci_dev_unlock(dev);
5256	}
5257	return 0;
5258}
5259
5260/* Do any devices on or below this slot prevent a bus reset? */
5261static bool pci_slot_resetable(struct pci_slot *slot)
5262{
5263	struct pci_dev *dev;
5264
5265	if (slot->bus->self &&
5266	    (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5267		return false;
5268
5269	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5270		if (!dev->slot || dev->slot != slot)
5271			continue;
5272		if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5273		    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5274			return false;
5275	}
5276
5277	return true;
5278}
5279
5280/* Lock devices from the top of the tree down */
5281static void pci_slot_lock(struct pci_slot *slot)
5282{
5283	struct pci_dev *dev;
5284
5285	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5286		if (!dev->slot || dev->slot != slot)
5287			continue;
5288		pci_dev_lock(dev);
5289		if (dev->subordinate)
5290			pci_bus_lock(dev->subordinate);
5291	}
5292}
5293
5294/* Unlock devices from the bottom of the tree up */
5295static void pci_slot_unlock(struct pci_slot *slot)
5296{
5297	struct pci_dev *dev;
5298
5299	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5300		if (!dev->slot || dev->slot != slot)
5301			continue;
5302		if (dev->subordinate)
5303			pci_bus_unlock(dev->subordinate);
5304		pci_dev_unlock(dev);
5305	}
5306}
5307
5308/* Return 1 on successful lock, 0 on contention */
5309static int pci_slot_trylock(struct pci_slot *slot)
5310{
5311	struct pci_dev *dev;
5312
5313	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5314		if (!dev->slot || dev->slot != slot)
5315			continue;
5316		if (!pci_dev_trylock(dev))
5317			goto unlock;
5318		if (dev->subordinate) {
5319			if (!pci_bus_trylock(dev->subordinate)) {
5320				pci_dev_unlock(dev);
5321				goto unlock;
5322			}
5323		}
5324	}
5325	return 1;
5326
5327unlock:
5328	list_for_each_entry_continue_reverse(dev,
5329					     &slot->bus->devices, bus_list) {
5330		if (!dev->slot || dev->slot != slot)
5331			continue;
5332		if (dev->subordinate)
5333			pci_bus_unlock(dev->subordinate);
5334		pci_dev_unlock(dev);
5335	}
5336	return 0;
5337}
5338
5339/*
5340 * Save and disable devices from the top of the tree down while holding
5341 * the @dev mutex lock for the entire tree.
5342 */
5343static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5344{
5345	struct pci_dev *dev;
5346
5347	list_for_each_entry(dev, &bus->devices, bus_list) {
5348		pci_dev_save_and_disable(dev);
5349		if (dev->subordinate)
5350			pci_bus_save_and_disable_locked(dev->subordinate);
5351	}
5352}
5353
5354/*
5355 * Restore devices from top of the tree down while holding @dev mutex lock
5356 * for the entire tree.  Parent bridges need to be restored before we can
5357 * get to subordinate devices.
5358 */
5359static void pci_bus_restore_locked(struct pci_bus *bus)
5360{
5361	struct pci_dev *dev;
5362
5363	list_for_each_entry(dev, &bus->devices, bus_list) {
5364		pci_dev_restore(dev);
5365		if (dev->subordinate)
5366			pci_bus_restore_locked(dev->subordinate);
5367	}
5368}
5369
5370/*
5371 * Save and disable devices from the top of the tree down while holding
5372 * the @dev mutex lock for the entire tree.
5373 */
5374static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5375{
5376	struct pci_dev *dev;
5377
5378	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5379		if (!dev->slot || dev->slot != slot)
5380			continue;
5381		pci_dev_save_and_disable(dev);
5382		if (dev->subordinate)
5383			pci_bus_save_and_disable_locked(dev->subordinate);
5384	}
5385}
5386
5387/*
5388 * Restore devices from top of the tree down while holding @dev mutex lock
5389 * for the entire tree.  Parent bridges need to be restored before we can
5390 * get to subordinate devices.
5391 */
5392static void pci_slot_restore_locked(struct pci_slot *slot)
5393{
5394	struct pci_dev *dev;
5395
5396	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5397		if (!dev->slot || dev->slot != slot)
5398			continue;
5399		pci_dev_restore(dev);
5400		if (dev->subordinate)
5401			pci_bus_restore_locked(dev->subordinate);
5402	}
5403}
5404
5405static int pci_slot_reset(struct pci_slot *slot, int probe)
5406{
5407	int rc;
5408
5409	if (!slot || !pci_slot_resetable(slot))
5410		return -ENOTTY;
5411
5412	if (!probe)
5413		pci_slot_lock(slot);
5414
5415	might_sleep();
5416
5417	rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5418
5419	if (!probe)
5420		pci_slot_unlock(slot);
5421
5422	return rc;
5423}
5424
5425/**
5426 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5427 * @slot: PCI slot to probe
5428 *
5429 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5430 */
5431int pci_probe_reset_slot(struct pci_slot *slot)
5432{
5433	return pci_slot_reset(slot, 1);
5434}
5435EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5436
5437/**
5438 * __pci_reset_slot - Try to reset a PCI slot
5439 * @slot: PCI slot to reset
5440 *
5441 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5442 * independent of other slots.  For instance, some slots may support slot power
5443 * control.  In the case of a 1:1 bus to slot architecture, this function may
5444 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5445 * Generally a slot reset should be attempted before a bus reset.  All of the
5446 * function of the slot and any subordinate buses behind the slot are reset
5447 * through this function.  PCI config space of all devices in the slot and
5448 * behind the slot is saved before and restored after reset.
5449 *
5450 * Same as above except return -EAGAIN if the slot cannot be locked
5451 */
5452static int __pci_reset_slot(struct pci_slot *slot)
5453{
5454	int rc;
5455
5456	rc = pci_slot_reset(slot, 1);
5457	if (rc)
5458		return rc;
5459
5460	if (pci_slot_trylock(slot)) {
5461		pci_slot_save_and_disable_locked(slot);
5462		might_sleep();
5463		rc = pci_reset_hotplug_slot(slot->hotplug, 0);
5464		pci_slot_restore_locked(slot);
5465		pci_slot_unlock(slot);
5466	} else
5467		rc = -EAGAIN;
5468
5469	return rc;
5470}
5471
5472static int pci_bus_reset(struct pci_bus *bus, int probe)
5473{
5474	int ret;
5475
5476	if (!bus->self || !pci_bus_resetable(bus))
5477		return -ENOTTY;
5478
5479	if (probe)
5480		return 0;
5481
5482	pci_bus_lock(bus);
5483
5484	might_sleep();
5485
5486	ret = pci_bridge_secondary_bus_reset(bus->self);
5487
5488	pci_bus_unlock(bus);
5489
5490	return ret;
5491}
5492
5493/**
5494 * pci_bus_error_reset - reset the bridge's subordinate bus
5495 * @bridge: The parent device that connects to the bus to reset
5496 *
5497 * This function will first try to reset the slots on this bus if the method is
5498 * available. If slot reset fails or is not available, this will fall back to a
5499 * secondary bus reset.
5500 */
5501int pci_bus_error_reset(struct pci_dev *bridge)
5502{
5503	struct pci_bus *bus = bridge->subordinate;
5504	struct pci_slot *slot;
5505
5506	if (!bus)
5507		return -ENOTTY;
5508
5509	mutex_lock(&pci_slot_mutex);
5510	if (list_empty(&bus->slots))
5511		goto bus_reset;
5512
5513	list_for_each_entry(slot, &bus->slots, list)
5514		if (pci_probe_reset_slot(slot))
5515			goto bus_reset;
5516
5517	list_for_each_entry(slot, &bus->slots, list)
5518		if (pci_slot_reset(slot, 0))
5519			goto bus_reset;
5520
5521	mutex_unlock(&pci_slot_mutex);
5522	return 0;
5523bus_reset:
5524	mutex_unlock(&pci_slot_mutex);
5525	return pci_bus_reset(bridge->subordinate, 0);
5526}
5527
5528/**
5529 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5530 * @bus: PCI bus to probe
5531 *
5532 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5533 */
5534int pci_probe_reset_bus(struct pci_bus *bus)
5535{
5536	return pci_bus_reset(bus, 1);
5537}
5538EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5539
5540/**
5541 * __pci_reset_bus - Try to reset a PCI bus
5542 * @bus: top level PCI bus to reset
5543 *
5544 * Same as above except return -EAGAIN if the bus cannot be locked
5545 */
5546static int __pci_reset_bus(struct pci_bus *bus)
5547{
5548	int rc;
5549
5550	rc = pci_bus_reset(bus, 1);
5551	if (rc)
5552		return rc;
5553
5554	if (pci_bus_trylock(bus)) {
5555		pci_bus_save_and_disable_locked(bus);
5556		might_sleep();
5557		rc = pci_bridge_secondary_bus_reset(bus->self);
5558		pci_bus_restore_locked(bus);
5559		pci_bus_unlock(bus);
5560	} else
5561		rc = -EAGAIN;
5562
5563	return rc;
5564}
5565
5566/**
5567 * pci_reset_bus - Try to reset a PCI bus
5568 * @pdev: top level PCI device to reset via slot/bus
5569 *
5570 * Same as above except return -EAGAIN if the bus cannot be locked
5571 */
5572int pci_reset_bus(struct pci_dev *pdev)
5573{
5574	return (!pci_probe_reset_slot(pdev->slot)) ?
5575	    __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5576}
5577EXPORT_SYMBOL_GPL(pci_reset_bus);
5578
5579/**
5580 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5581 * @dev: PCI device to query
5582 *
5583 * Returns mmrbc: maximum designed memory read count in bytes or
5584 * appropriate error value.
5585 */
5586int pcix_get_max_mmrbc(struct pci_dev *dev)
5587{
5588	int cap;
5589	u32 stat;
5590
5591	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5592	if (!cap)
5593		return -EINVAL;
5594
5595	if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5596		return -EINVAL;
5597
5598	return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5599}
5600EXPORT_SYMBOL(pcix_get_max_mmrbc);
5601
5602/**
5603 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5604 * @dev: PCI device to query
5605 *
5606 * Returns mmrbc: maximum memory read count in bytes or appropriate error
5607 * value.
5608 */
5609int pcix_get_mmrbc(struct pci_dev *dev)
5610{
5611	int cap;
5612	u16 cmd;
5613
5614	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5615	if (!cap)
5616		return -EINVAL;
5617
5618	if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5619		return -EINVAL;
5620
5621	return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5622}
5623EXPORT_SYMBOL(pcix_get_mmrbc);
5624
5625/**
5626 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5627 * @dev: PCI device to query
5628 * @mmrbc: maximum memory read count in bytes
5629 *    valid values are 512, 1024, 2048, 4096
5630 *
5631 * If possible sets maximum memory read byte count, some bridges have errata
5632 * that prevent this.
5633 */
5634int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5635{
5636	int cap;
5637	u32 stat, v, o;
5638	u16 cmd;
5639
5640	if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5641		return -EINVAL;
5642
5643	v = ffs(mmrbc) - 10;
5644
5645	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5646	if (!cap)
5647		return -EINVAL;
5648
5649	if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5650		return -EINVAL;
5651
5652	if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5653		return -E2BIG;
5654
5655	if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5656		return -EINVAL;
5657
5658	o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5659	if (o != v) {
5660		if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5661			return -EIO;
5662
5663		cmd &= ~PCI_X_CMD_MAX_READ;
5664		cmd |= v << 2;
5665		if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5666			return -EIO;
5667	}
5668	return 0;
5669}
5670EXPORT_SYMBOL(pcix_set_mmrbc);
5671
5672/**
5673 * pcie_get_readrq - get PCI Express read request size
5674 * @dev: PCI device to query
5675 *
5676 * Returns maximum memory read request in bytes or appropriate error value.
5677 */
5678int pcie_get_readrq(struct pci_dev *dev)
5679{
5680	u16 ctl;
5681
5682	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5683
5684	return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5685}
5686EXPORT_SYMBOL(pcie_get_readrq);
5687
5688/**
5689 * pcie_set_readrq - set PCI Express maximum memory read request
5690 * @dev: PCI device to query
5691 * @rq: maximum memory read count in bytes
5692 *    valid values are 128, 256, 512, 1024, 2048, 4096
5693 *
5694 * If possible sets maximum memory read request in bytes
5695 */
5696int pcie_set_readrq(struct pci_dev *dev, int rq)
5697{
5698	u16 v;
5699	int ret;
5700
5701	if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5702		return -EINVAL;
5703
5704	/*
5705	 * If using the "performance" PCIe config, we clamp the read rq
5706	 * size to the max packet size to keep the host bridge from
5707	 * generating requests larger than we can cope with.
5708	 */
5709	if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5710		int mps = pcie_get_mps(dev);
5711
5712		if (mps < rq)
5713			rq = mps;
5714	}
5715
5716	v = (ffs(rq) - 8) << 12;
5717
5718	ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5719						  PCI_EXP_DEVCTL_READRQ, v);
5720
5721	return pcibios_err_to_errno(ret);
5722}
5723EXPORT_SYMBOL(pcie_set_readrq);
5724
5725/**
5726 * pcie_get_mps - get PCI Express maximum payload size
5727 * @dev: PCI device to query
5728 *
5729 * Returns maximum payload size in bytes
5730 */
5731int pcie_get_mps(struct pci_dev *dev)
5732{
5733	u16 ctl;
5734
5735	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5736
5737	return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5738}
5739EXPORT_SYMBOL(pcie_get_mps);
5740
5741/**
5742 * pcie_set_mps - set PCI Express maximum payload size
5743 * @dev: PCI device to query
5744 * @mps: maximum payload size in bytes
5745 *    valid values are 128, 256, 512, 1024, 2048, 4096
5746 *
5747 * If possible sets maximum payload size
5748 */
5749int pcie_set_mps(struct pci_dev *dev, int mps)
5750{
5751	u16 v;
5752	int ret;
5753
5754	if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5755		return -EINVAL;
5756
5757	v = ffs(mps) - 8;
5758	if (v > dev->pcie_mpss)
5759		return -EINVAL;
5760	v <<= 5;
5761
5762	ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5763						  PCI_EXP_DEVCTL_PAYLOAD, v);
5764
5765	return pcibios_err_to_errno(ret);
5766}
5767EXPORT_SYMBOL(pcie_set_mps);
5768
5769/**
5770 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5771 *			      device and its bandwidth limitation
5772 * @dev: PCI device to query
5773 * @limiting_dev: storage for device causing the bandwidth limitation
5774 * @speed: storage for speed of limiting device
5775 * @width: storage for width of limiting device
5776 *
5777 * Walk up the PCI device chain and find the point where the minimum
5778 * bandwidth is available.  Return the bandwidth available there and (if
5779 * limiting_dev, speed, and width pointers are supplied) information about
5780 * that point.  The bandwidth returned is in Mb/s, i.e., megabits/second of
5781 * raw bandwidth.
5782 */
5783u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5784			     enum pci_bus_speed *speed,
5785			     enum pcie_link_width *width)
5786{
5787	u16 lnksta;
5788	enum pci_bus_speed next_speed;
5789	enum pcie_link_width next_width;
5790	u32 bw, next_bw;
5791
5792	if (speed)
5793		*speed = PCI_SPEED_UNKNOWN;
5794	if (width)
5795		*width = PCIE_LNK_WIDTH_UNKNOWN;
5796
5797	bw = 0;
5798
5799	while (dev) {
5800		pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5801
5802		next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5803		next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5804			PCI_EXP_LNKSTA_NLW_SHIFT;
5805
5806		next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5807
5808		/* Check if current device limits the total bandwidth */
5809		if (!bw || next_bw <= bw) {
5810			bw = next_bw;
5811
5812			if (limiting_dev)
5813				*limiting_dev = dev;
5814			if (speed)
5815				*speed = next_speed;
5816			if (width)
5817				*width = next_width;
5818		}
5819
5820		dev = pci_upstream_bridge(dev);
5821	}
5822
5823	return bw;
5824}
5825EXPORT_SYMBOL(pcie_bandwidth_available);
5826
5827/**
5828 * pcie_get_speed_cap - query for the PCI device's link speed capability
5829 * @dev: PCI device to query
5830 *
5831 * Query the PCI device speed capability.  Return the maximum link speed
5832 * supported by the device.
5833 */
5834enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5835{
5836	u32 lnkcap2, lnkcap;
5837
5838	/*
5839	 * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18.  The
5840	 * implementation note there recommends using the Supported Link
5841	 * Speeds Vector in Link Capabilities 2 when supported.
5842	 *
5843	 * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
5844	 * should use the Supported Link Speeds field in Link Capabilities,
5845	 * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
5846	 */
5847	pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5848
5849	/* PCIe r3.0-compliant */
5850	if (lnkcap2)
5851		return PCIE_LNKCAP2_SLS2SPEED(lnkcap2);
 
 
 
 
 
 
 
 
 
5852
5853	pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5854	if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
5855		return PCIE_SPEED_5_0GT;
5856	else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
5857		return PCIE_SPEED_2_5GT;
5858
5859	return PCI_SPEED_UNKNOWN;
5860}
5861EXPORT_SYMBOL(pcie_get_speed_cap);
5862
5863/**
5864 * pcie_get_width_cap - query for the PCI device's link width capability
5865 * @dev: PCI device to query
5866 *
5867 * Query the PCI device width capability.  Return the maximum link width
5868 * supported by the device.
5869 */
5870enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5871{
5872	u32 lnkcap;
5873
5874	pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5875	if (lnkcap)
5876		return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5877
5878	return PCIE_LNK_WIDTH_UNKNOWN;
5879}
5880EXPORT_SYMBOL(pcie_get_width_cap);
5881
5882/**
5883 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5884 * @dev: PCI device
5885 * @speed: storage for link speed
5886 * @width: storage for link width
5887 *
5888 * Calculate a PCI device's link bandwidth by querying for its link speed
5889 * and width, multiplying them, and applying encoding overhead.  The result
5890 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5891 */
5892u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5893			   enum pcie_link_width *width)
5894{
5895	*speed = pcie_get_speed_cap(dev);
5896	*width = pcie_get_width_cap(dev);
5897
5898	if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5899		return 0;
5900
5901	return *width * PCIE_SPEED2MBS_ENC(*speed);
5902}
5903
5904/**
5905 * __pcie_print_link_status - Report the PCI device's link speed and width
5906 * @dev: PCI device to query
5907 * @verbose: Print info even when enough bandwidth is available
5908 *
5909 * If the available bandwidth at the device is less than the device is
5910 * capable of, report the device's maximum possible bandwidth and the
5911 * upstream link that limits its performance.  If @verbose, always print
5912 * the available bandwidth, even if the device isn't constrained.
5913 */
5914void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
5915{
5916	enum pcie_link_width width, width_cap;
5917	enum pci_bus_speed speed, speed_cap;
5918	struct pci_dev *limiting_dev = NULL;
5919	u32 bw_avail, bw_cap;
5920
5921	bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5922	bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5923
5924	if (bw_avail >= bw_cap && verbose)
5925		pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5926			 bw_cap / 1000, bw_cap % 1000,
5927			 pci_speed_string(speed_cap), width_cap);
5928	else if (bw_avail < bw_cap)
5929		pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth, limited by %s x%d link at %s (capable of %u.%03u Gb/s with %s x%d link)\n",
5930			 bw_avail / 1000, bw_avail % 1000,
5931			 pci_speed_string(speed), width,
5932			 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5933			 bw_cap / 1000, bw_cap % 1000,
5934			 pci_speed_string(speed_cap), width_cap);
5935}
5936
5937/**
5938 * pcie_print_link_status - Report the PCI device's link speed and width
5939 * @dev: PCI device to query
5940 *
5941 * Report the available bandwidth at the device.
5942 */
5943void pcie_print_link_status(struct pci_dev *dev)
5944{
5945	__pcie_print_link_status(dev, true);
5946}
5947EXPORT_SYMBOL(pcie_print_link_status);
5948
5949/**
5950 * pci_select_bars - Make BAR mask from the type of resource
5951 * @dev: the PCI device for which BAR mask is made
5952 * @flags: resource type mask to be selected
5953 *
5954 * This helper routine makes bar mask from the type of resource.
5955 */
5956int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5957{
5958	int i, bars = 0;
5959	for (i = 0; i < PCI_NUM_RESOURCES; i++)
5960		if (pci_resource_flags(dev, i) & flags)
5961			bars |= (1 << i);
5962	return bars;
5963}
5964EXPORT_SYMBOL(pci_select_bars);
5965
5966/* Some architectures require additional programming to enable VGA */
5967static arch_set_vga_state_t arch_set_vga_state;
5968
5969void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5970{
5971	arch_set_vga_state = func;	/* NULL disables */
5972}
5973
5974static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5975				  unsigned int command_bits, u32 flags)
5976{
5977	if (arch_set_vga_state)
5978		return arch_set_vga_state(dev, decode, command_bits,
5979						flags);
5980	return 0;
5981}
5982
5983/**
5984 * pci_set_vga_state - set VGA decode state on device and parents if requested
5985 * @dev: the PCI device
5986 * @decode: true = enable decoding, false = disable decoding
5987 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5988 * @flags: traverse ancestors and change bridges
5989 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5990 */
5991int pci_set_vga_state(struct pci_dev *dev, bool decode,
5992		      unsigned int command_bits, u32 flags)
5993{
5994	struct pci_bus *bus;
5995	struct pci_dev *bridge;
5996	u16 cmd;
5997	int rc;
5998
5999	WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
6000
6001	/* ARCH specific VGA enables */
6002	rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
6003	if (rc)
6004		return rc;
6005
6006	if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
6007		pci_read_config_word(dev, PCI_COMMAND, &cmd);
6008		if (decode == true)
6009			cmd |= command_bits;
6010		else
6011			cmd &= ~command_bits;
6012		pci_write_config_word(dev, PCI_COMMAND, cmd);
6013	}
6014
6015	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
6016		return 0;
6017
6018	bus = dev->bus;
6019	while (bus) {
6020		bridge = bus->self;
6021		if (bridge) {
6022			pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
6023					     &cmd);
6024			if (decode == true)
6025				cmd |= PCI_BRIDGE_CTL_VGA;
6026			else
6027				cmd &= ~PCI_BRIDGE_CTL_VGA;
6028			pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
6029					      cmd);
6030		}
6031		bus = bus->parent;
6032	}
6033	return 0;
6034}
6035
6036#ifdef CONFIG_ACPI
6037bool pci_pr3_present(struct pci_dev *pdev)
6038{
6039	struct acpi_device *adev;
6040
6041	if (acpi_disabled)
6042		return false;
6043
6044	adev = ACPI_COMPANION(&pdev->dev);
6045	if (!adev)
6046		return false;
6047
6048	return adev->power.flags.power_resources &&
6049		acpi_has_method(adev->handle, "_PR3");
6050}
6051EXPORT_SYMBOL_GPL(pci_pr3_present);
6052#endif
6053
6054/**
6055 * pci_add_dma_alias - Add a DMA devfn alias for a device
6056 * @dev: the PCI device for which alias is added
6057 * @devfn_from: alias slot and function
6058 * @nr_devfns: number of subsequent devfns to alias
6059 *
6060 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
6061 * which is used to program permissible bus-devfn source addresses for DMA
6062 * requests in an IOMMU.  These aliases factor into IOMMU group creation
6063 * and are useful for devices generating DMA requests beyond or different
6064 * from their logical bus-devfn.  Examples include device quirks where the
6065 * device simply uses the wrong devfn, as well as non-transparent bridges
6066 * where the alias may be a proxy for devices in another domain.
6067 *
6068 * IOMMU group creation is performed during device discovery or addition,
6069 * prior to any potential DMA mapping and therefore prior to driver probing
6070 * (especially for userspace assigned devices where IOMMU group definition
6071 * cannot be left as a userspace activity).  DMA aliases should therefore
6072 * be configured via quirks, such as the PCI fixup header quirk.
6073 */
6074void pci_add_dma_alias(struct pci_dev *dev, u8 devfn_from, unsigned nr_devfns)
6075{
6076	int devfn_to;
6077
6078	nr_devfns = min(nr_devfns, (unsigned) MAX_NR_DEVFNS - devfn_from);
6079	devfn_to = devfn_from + nr_devfns - 1;
6080
6081	if (!dev->dma_alias_mask)
6082		dev->dma_alias_mask = bitmap_zalloc(MAX_NR_DEVFNS, GFP_KERNEL);
6083	if (!dev->dma_alias_mask) {
6084		pci_warn(dev, "Unable to allocate DMA alias mask\n");
6085		return;
6086	}
6087
6088	bitmap_set(dev->dma_alias_mask, devfn_from, nr_devfns);
6089
6090	if (nr_devfns == 1)
6091		pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
6092				PCI_SLOT(devfn_from), PCI_FUNC(devfn_from));
6093	else if (nr_devfns > 1)
6094		pci_info(dev, "Enabling fixed DMA alias for devfn range from %02x.%d to %02x.%d\n",
6095				PCI_SLOT(devfn_from), PCI_FUNC(devfn_from),
6096				PCI_SLOT(devfn_to), PCI_FUNC(devfn_to));
6097}
6098
6099bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
6100{
6101	return (dev1->dma_alias_mask &&
6102		test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
6103	       (dev2->dma_alias_mask &&
6104		test_bit(dev1->devfn, dev2->dma_alias_mask)) ||
6105	       pci_real_dma_dev(dev1) == dev2 ||
6106	       pci_real_dma_dev(dev2) == dev1;
6107}
6108
6109bool pci_device_is_present(struct pci_dev *pdev)
6110{
6111	u32 v;
6112
6113	if (pci_dev_is_disconnected(pdev))
6114		return false;
6115	return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
6116}
6117EXPORT_SYMBOL_GPL(pci_device_is_present);
6118
6119void pci_ignore_hotplug(struct pci_dev *dev)
6120{
6121	struct pci_dev *bridge = dev->bus->self;
6122
6123	dev->ignore_hotplug = 1;
6124	/* Propagate the "ignore hotplug" setting to the parent bridge. */
6125	if (bridge)
6126		bridge->ignore_hotplug = 1;
6127}
6128EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
6129
6130/**
6131 * pci_real_dma_dev - Get PCI DMA device for PCI device
6132 * @dev: the PCI device that may have a PCI DMA alias
6133 *
6134 * Permits the platform to provide architecture-specific functionality to
6135 * devices needing to alias DMA to another PCI device on another PCI bus. If
6136 * the PCI device is on the same bus, it is recommended to use
6137 * pci_add_dma_alias(). This is the default implementation. Architecture
6138 * implementations can override this.
6139 */
6140struct pci_dev __weak *pci_real_dma_dev(struct pci_dev *dev)
6141{
6142	return dev;
6143}
6144
6145resource_size_t __weak pcibios_default_alignment(void)
6146{
6147	return 0;
6148}
6149
6150/*
6151 * Arches that don't want to expose struct resource to userland as-is in
6152 * sysfs and /proc can implement their own pci_resource_to_user().
6153 */
6154void __weak pci_resource_to_user(const struct pci_dev *dev, int bar,
6155				 const struct resource *rsrc,
6156				 resource_size_t *start, resource_size_t *end)
6157{
6158	*start = rsrc->start;
6159	*end = rsrc->end;
6160}
6161
6162static char *resource_alignment_param;
6163static DEFINE_SPINLOCK(resource_alignment_lock);
6164
6165/**
6166 * pci_specified_resource_alignment - get resource alignment specified by user.
6167 * @dev: the PCI device to get
6168 * @resize: whether or not to change resources' size when reassigning alignment
6169 *
6170 * RETURNS: Resource alignment if it is specified.
6171 *          Zero if it is not specified.
6172 */
6173static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
6174							bool *resize)
6175{
6176	int align_order, count;
6177	resource_size_t align = pcibios_default_alignment();
6178	const char *p;
6179	int ret;
6180
6181	spin_lock(&resource_alignment_lock);
6182	p = resource_alignment_param;
6183	if (!p || !*p)
6184		goto out;
6185	if (pci_has_flag(PCI_PROBE_ONLY)) {
6186		align = 0;
6187		pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
6188		goto out;
6189	}
6190
6191	while (*p) {
6192		count = 0;
6193		if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
6194							p[count] == '@') {
6195			p += count + 1;
6196		} else {
6197			align_order = -1;
6198		}
6199
6200		ret = pci_dev_str_match(dev, p, &p);
6201		if (ret == 1) {
6202			*resize = true;
6203			if (align_order == -1)
6204				align = PAGE_SIZE;
6205			else
6206				align = 1 << align_order;
6207			break;
6208		} else if (ret < 0) {
6209			pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
6210			       p);
6211			break;
6212		}
6213
6214		if (*p != ';' && *p != ',') {
6215			/* End of param or invalid format */
6216			break;
6217		}
6218		p++;
6219	}
6220out:
6221	spin_unlock(&resource_alignment_lock);
6222	return align;
6223}
6224
6225static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
6226					   resource_size_t align, bool resize)
6227{
6228	struct resource *r = &dev->resource[bar];
6229	resource_size_t size;
6230
6231	if (!(r->flags & IORESOURCE_MEM))
6232		return;
6233
6234	if (r->flags & IORESOURCE_PCI_FIXED) {
6235		pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
6236			 bar, r, (unsigned long long)align);
6237		return;
6238	}
6239
6240	size = resource_size(r);
6241	if (size >= align)
6242		return;
6243
6244	/*
6245	 * Increase the alignment of the resource.  There are two ways we
6246	 * can do this:
6247	 *
6248	 * 1) Increase the size of the resource.  BARs are aligned on their
6249	 *    size, so when we reallocate space for this resource, we'll
6250	 *    allocate it with the larger alignment.  This also prevents
6251	 *    assignment of any other BARs inside the alignment region, so
6252	 *    if we're requesting page alignment, this means no other BARs
6253	 *    will share the page.
6254	 *
6255	 *    The disadvantage is that this makes the resource larger than
6256	 *    the hardware BAR, which may break drivers that compute things
6257	 *    based on the resource size, e.g., to find registers at a
6258	 *    fixed offset before the end of the BAR.
6259	 *
6260	 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
6261	 *    set r->start to the desired alignment.  By itself this
6262	 *    doesn't prevent other BARs being put inside the alignment
6263	 *    region, but if we realign *every* resource of every device in
6264	 *    the system, none of them will share an alignment region.
6265	 *
6266	 * When the user has requested alignment for only some devices via
6267	 * the "pci=resource_alignment" argument, "resize" is true and we
6268	 * use the first method.  Otherwise we assume we're aligning all
6269	 * devices and we use the second.
6270	 */
6271
6272	pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
6273		 bar, r, (unsigned long long)align);
6274
6275	if (resize) {
6276		r->start = 0;
6277		r->end = align - 1;
6278	} else {
6279		r->flags &= ~IORESOURCE_SIZEALIGN;
6280		r->flags |= IORESOURCE_STARTALIGN;
6281		r->start = align;
6282		r->end = r->start + size - 1;
6283	}
6284	r->flags |= IORESOURCE_UNSET;
6285}
6286
6287/*
6288 * This function disables memory decoding and releases memory resources
6289 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6290 * It also rounds up size to specified alignment.
6291 * Later on, the kernel will assign page-aligned memory resource back
6292 * to the device.
6293 */
6294void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6295{
6296	int i;
6297	struct resource *r;
6298	resource_size_t align;
6299	u16 command;
6300	bool resize = false;
6301
6302	/*
6303	 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6304	 * 3.4.1.11.  Their resources are allocated from the space
6305	 * described by the VF BARx register in the PF's SR-IOV capability.
6306	 * We can't influence their alignment here.
6307	 */
6308	if (dev->is_virtfn)
6309		return;
6310
6311	/* check if specified PCI is target device to reassign */
6312	align = pci_specified_resource_alignment(dev, &resize);
6313	if (!align)
6314		return;
6315
6316	if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6317	    (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6318		pci_warn(dev, "Can't reassign resources to host bridge\n");
6319		return;
6320	}
6321
6322	pci_read_config_word(dev, PCI_COMMAND, &command);
6323	command &= ~PCI_COMMAND_MEMORY;
6324	pci_write_config_word(dev, PCI_COMMAND, command);
6325
6326	for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6327		pci_request_resource_alignment(dev, i, align, resize);
6328
6329	/*
6330	 * Need to disable bridge's resource window,
6331	 * to enable the kernel to reassign new resource
6332	 * window later on.
6333	 */
6334	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6335		for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6336			r = &dev->resource[i];
6337			if (!(r->flags & IORESOURCE_MEM))
6338				continue;
6339			r->flags |= IORESOURCE_UNSET;
6340			r->end = resource_size(r) - 1;
6341			r->start = 0;
6342		}
6343		pci_disable_bridge_window(dev);
6344	}
6345}
6346
6347static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
6348{
6349	size_t count = 0;
6350
6351	spin_lock(&resource_alignment_lock);
6352	if (resource_alignment_param)
6353		count = snprintf(buf, PAGE_SIZE, "%s", resource_alignment_param);
6354	spin_unlock(&resource_alignment_lock);
6355
6356	/*
6357	 * When set by the command line, resource_alignment_param will not
6358	 * have a trailing line feed, which is ugly. So conditionally add
6359	 * it here.
6360	 */
6361	if (count >= 2 && buf[count - 2] != '\n' && count < PAGE_SIZE - 1) {
6362		buf[count - 1] = '\n';
6363		buf[count++] = 0;
6364	}
6365
6366	return count;
6367}
6368
6369static ssize_t resource_alignment_store(struct bus_type *bus,
6370					const char *buf, size_t count)
6371{
6372	char *param = kstrndup(buf, count, GFP_KERNEL);
6373
6374	if (!param)
6375		return -ENOMEM;
6376
6377	spin_lock(&resource_alignment_lock);
6378	kfree(resource_alignment_param);
6379	resource_alignment_param = param;
6380	spin_unlock(&resource_alignment_lock);
6381	return count;
6382}
6383
6384static BUS_ATTR_RW(resource_alignment);
6385
6386static int __init pci_resource_alignment_sysfs_init(void)
6387{
6388	return bus_create_file(&pci_bus_type,
6389					&bus_attr_resource_alignment);
6390}
6391late_initcall(pci_resource_alignment_sysfs_init);
6392
6393static void pci_no_domains(void)
6394{
6395#ifdef CONFIG_PCI_DOMAINS
6396	pci_domains_supported = 0;
6397#endif
6398}
6399
6400#ifdef CONFIG_PCI_DOMAINS_GENERIC
6401static atomic_t __domain_nr = ATOMIC_INIT(-1);
6402
6403static int pci_get_new_domain_nr(void)
6404{
6405	return atomic_inc_return(&__domain_nr);
6406}
6407
6408static int of_pci_bus_find_domain_nr(struct device *parent)
6409{
6410	static int use_dt_domains = -1;
6411	int domain = -1;
6412
6413	if (parent)
6414		domain = of_get_pci_domain_nr(parent->of_node);
6415
6416	/*
6417	 * Check DT domain and use_dt_domains values.
6418	 *
6419	 * If DT domain property is valid (domain >= 0) and
6420	 * use_dt_domains != 0, the DT assignment is valid since this means
6421	 * we have not previously allocated a domain number by using
6422	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
6423	 * 1, to indicate that we have just assigned a domain number from
6424	 * DT.
6425	 *
6426	 * If DT domain property value is not valid (ie domain < 0), and we
6427	 * have not previously assigned a domain number from DT
6428	 * (use_dt_domains != 1) we should assign a domain number by
6429	 * using the:
6430	 *
6431	 * pci_get_new_domain_nr()
6432	 *
6433	 * API and update the use_dt_domains value to keep track of method we
6434	 * are using to assign domain numbers (use_dt_domains = 0).
6435	 *
6436	 * All other combinations imply we have a platform that is trying
6437	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6438	 * which is a recipe for domain mishandling and it is prevented by
6439	 * invalidating the domain value (domain = -1) and printing a
6440	 * corresponding error.
6441	 */
6442	if (domain >= 0 && use_dt_domains) {
6443		use_dt_domains = 1;
6444	} else if (domain < 0 && use_dt_domains != 1) {
6445		use_dt_domains = 0;
6446		domain = pci_get_new_domain_nr();
6447	} else {
6448		if (parent)
6449			pr_err("Node %pOF has ", parent->of_node);
6450		pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6451		domain = -1;
6452	}
6453
6454	return domain;
6455}
6456
6457int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6458{
6459	return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6460			       acpi_pci_bus_find_domain_nr(bus);
6461}
6462#endif
6463
6464/**
6465 * pci_ext_cfg_avail - can we access extended PCI config space?
6466 *
6467 * Returns 1 if we can access PCI extended config space (offsets
6468 * greater than 0xff). This is the default implementation. Architecture
6469 * implementations can override this.
6470 */
6471int __weak pci_ext_cfg_avail(void)
6472{
6473	return 1;
6474}
6475
6476void __weak pci_fixup_cardbus(struct pci_bus *bus)
6477{
6478}
6479EXPORT_SYMBOL(pci_fixup_cardbus);
6480
6481static int __init pci_setup(char *str)
6482{
6483	while (str) {
6484		char *k = strchr(str, ',');
6485		if (k)
6486			*k++ = 0;
6487		if (*str && (str = pcibios_setup(str)) && *str) {
6488			if (!strcmp(str, "nomsi")) {
6489				pci_no_msi();
6490			} else if (!strncmp(str, "noats", 5)) {
6491				pr_info("PCIe: ATS is disabled\n");
6492				pcie_ats_disabled = true;
6493			} else if (!strcmp(str, "noaer")) {
6494				pci_no_aer();
6495			} else if (!strcmp(str, "earlydump")) {
6496				pci_early_dump = true;
6497			} else if (!strncmp(str, "realloc=", 8)) {
6498				pci_realloc_get_opt(str + 8);
6499			} else if (!strncmp(str, "realloc", 7)) {
6500				pci_realloc_get_opt("on");
6501			} else if (!strcmp(str, "nodomains")) {
6502				pci_no_domains();
6503			} else if (!strncmp(str, "noari", 5)) {
6504				pcie_ari_disabled = true;
6505			} else if (!strncmp(str, "cbiosize=", 9)) {
6506				pci_cardbus_io_size = memparse(str + 9, &str);
6507			} else if (!strncmp(str, "cbmemsize=", 10)) {
6508				pci_cardbus_mem_size = memparse(str + 10, &str);
6509			} else if (!strncmp(str, "resource_alignment=", 19)) {
6510				resource_alignment_param = str + 19;
6511			} else if (!strncmp(str, "ecrc=", 5)) {
6512				pcie_ecrc_get_policy(str + 5);
6513			} else if (!strncmp(str, "hpiosize=", 9)) {
6514				pci_hotplug_io_size = memparse(str + 9, &str);
6515			} else if (!strncmp(str, "hpmmiosize=", 11)) {
6516				pci_hotplug_mmio_size = memparse(str + 11, &str);
6517			} else if (!strncmp(str, "hpmmioprefsize=", 15)) {
6518				pci_hotplug_mmio_pref_size = memparse(str + 15, &str);
6519			} else if (!strncmp(str, "hpmemsize=", 10)) {
6520				pci_hotplug_mmio_size = memparse(str + 10, &str);
6521				pci_hotplug_mmio_pref_size = pci_hotplug_mmio_size;
6522			} else if (!strncmp(str, "hpbussize=", 10)) {
6523				pci_hotplug_bus_size =
6524					simple_strtoul(str + 10, &str, 0);
6525				if (pci_hotplug_bus_size > 0xff)
6526					pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6527			} else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6528				pcie_bus_config = PCIE_BUS_TUNE_OFF;
6529			} else if (!strncmp(str, "pcie_bus_safe", 13)) {
6530				pcie_bus_config = PCIE_BUS_SAFE;
6531			} else if (!strncmp(str, "pcie_bus_perf", 13)) {
6532				pcie_bus_config = PCIE_BUS_PERFORMANCE;
6533			} else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6534				pcie_bus_config = PCIE_BUS_PEER2PEER;
6535			} else if (!strncmp(str, "pcie_scan_all", 13)) {
6536				pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6537			} else if (!strncmp(str, "disable_acs_redir=", 18)) {
6538				disable_acs_redir_param = str + 18;
6539			} else {
6540				pr_err("PCI: Unknown option `%s'\n", str);
6541			}
6542		}
6543		str = k;
6544	}
6545	return 0;
6546}
6547early_param("pci", pci_setup);
6548
6549/*
6550 * 'resource_alignment_param' and 'disable_acs_redir_param' are initialized
6551 * in pci_setup(), above, to point to data in the __initdata section which
6552 * will be freed after the init sequence is complete. We can't allocate memory
6553 * in pci_setup() because some architectures do not have any memory allocation
6554 * service available during an early_param() call. So we allocate memory and
6555 * copy the variable here before the init section is freed.
6556 *
6557 */
6558static int __init pci_realloc_setup_params(void)
6559{
6560	resource_alignment_param = kstrdup(resource_alignment_param,
6561					   GFP_KERNEL);
6562	disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6563
6564	return 0;
6565}
6566pure_initcall(pci_realloc_setup_params);