linux/drivers/pci/pci.c
<<
>>
Prefs
   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#ifdef CONFIG_HAS_IOMEM
 177void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
 178{
 179        struct resource *res = &pdev->resource[bar];
 180
 181        /*
 182         * Make sure the BAR is actually a memory resource, not an IO resource
 183         */
 184        if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
 185                pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
 186                return NULL;
 187        }
 188        return ioremap(res->start, resource_size(res));
 189}
 190EXPORT_SYMBOL_GPL(pci_ioremap_bar);
 191
 192void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
 193{
 194        /*
 195         * Make sure the BAR is actually a memory resource, not an IO resource
 196         */
 197        if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
 198                WARN_ON(1);
 199                return NULL;
 200        }
 201        return ioremap_wc(pci_resource_start(pdev, bar),
 202                          pci_resource_len(pdev, bar));
 203}
 204EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
 205#endif
 206
 207/**
 208 * pci_dev_str_match_path - test if a path string matches a device
 209 * @dev: the PCI device to test
 210 * @path: string to match the device against
 211 * @endptr: pointer to the string after the match
 212 *
 213 * Test if a string (typically from a kernel parameter) formatted as a
 214 * path of device/function addresses matches a PCI device. The string must
 215 * be of the form:
 216 *
 217 *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
 218 *
 219 * A path for a device can be obtained using 'lspci -t'.  Using a path
 220 * is more robust against bus renumbering than using only a single bus,
 221 * device and function address.
 222 *
 223 * Returns 1 if the string matches the device, 0 if it does not and
 224 * a negative error code if it fails to parse the string.
 225 */
 226static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
 227                                  const char **endptr)
 228{
 229        int ret;
 230        int seg, bus, slot, func;
 231        char *wpath, *p;
 232        char end;
 233
 234        *endptr = strchrnul(path, ';');
 235
 236        wpath = kmemdup_nul(path, *endptr - path, GFP_KERNEL);
 237        if (!wpath)
 238                return -ENOMEM;
 239
 240        while (1) {
 241                p = strrchr(wpath, '/');
 242                if (!p)
 243                        break;
 244                ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
 245                if (ret != 2) {
 246                        ret = -EINVAL;
 247                        goto free_and_exit;
 248                }
 249
 250                if (dev->devfn != PCI_DEVFN(slot, func)) {
 251                        ret = 0;
 252                        goto free_and_exit;
 253                }
 254
 255                /*
 256                 * Note: we don't need to get a reference to the upstream
 257                 * bridge because we hold a reference to the top level
 258                 * device which should hold a reference to the bridge,
 259                 * and so on.
 260                 */
 261                dev = pci_upstream_bridge(dev);
 262                if (!dev) {
 263                        ret = 0;
 264                        goto free_and_exit;
 265                }
 266
 267                *p = 0;
 268        }
 269
 270        ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
 271                     &func, &end);
 272        if (ret != 4) {
 273                seg = 0;
 274                ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
 275                if (ret != 3) {
 276                        ret = -EINVAL;
 277                        goto free_and_exit;
 278                }
 279        }
 280
 281        ret = (seg == pci_domain_nr(dev->bus) &&
 282               bus == dev->bus->number &&
 283               dev->devfn == PCI_DEVFN(slot, func));
 284
 285free_and_exit:
 286        kfree(wpath);
 287        return ret;
 288}
 289
 290/**
 291 * pci_dev_str_match - test if a string matches a device
 292 * @dev: the PCI device to test
 293 * @p: string to match the device against
 294 * @endptr: pointer to the string after the match
 295 *
 296 * Test if a string (typically from a kernel parameter) matches a specified
 297 * PCI device. The string may be of one of the following formats:
 298 *
 299 *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
 300 *   pci:<vendor>:<device>[:<subvendor>:<subdevice>]
 301 *
 302 * The first format specifies a PCI bus/device/function address which
 303 * may change if new hardware is inserted, if motherboard firmware changes,
 304 * or due to changes caused in kernel parameters. If the domain is
 305 * left unspecified, it is taken to be 0.  In order to be robust against
 306 * bus renumbering issues, a path of PCI device/function numbers may be used
 307 * to address the specific device.  The path for a device can be determined
 308 * through the use of 'lspci -t'.
 309 *
 310 * The second format matches devices using IDs in the configuration
 311 * space which may match multiple devices in the system. A value of 0
 312 * for any field will match all devices. (Note: this differs from
 313 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
 314 * legacy reasons and convenience so users don't have to specify
 315 * FFFFFFFFs on the command line.)
 316 *
 317 * Returns 1 if the string matches the device, 0 if it does not and
 318 * a negative error code if the string cannot be parsed.
 319 */
 320static int pci_dev_str_match(struct pci_dev *dev, const char *p,
 321                             const char **endptr)
 322{
 323        int ret;
 324        int count;
 325        unsigned short vendor, device, subsystem_vendor, subsystem_device;
 326
 327        if (strncmp(p, "pci:", 4) == 0) {
 328                /* PCI vendor/device (subvendor/subdevice) IDs are specified */
 329                p += 4;
 330                ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
 331                             &subsystem_vendor, &subsystem_device, &count);
 332                if (ret != 4) {
 333                        ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
 334                        if (ret != 2)
 335                                return -EINVAL;
 336
 337                        subsystem_vendor = 0;
 338                        subsystem_device = 0;
 339                }
 340
 341                p += count;
 342
 343                if ((!vendor || vendor == dev->vendor) &&
 344                    (!device || device == dev->device) &&
 345                    (!subsystem_vendor ||
 346                            subsystem_vendor == dev->subsystem_vendor) &&
 347                    (!subsystem_device ||
 348                            subsystem_device == dev->subsystem_device))
 349                        goto found;
 350        } else {
 351                /*
 352                 * PCI Bus, Device, Function IDs are specified
 353                 * (optionally, may include a path of devfns following it)
 354                 */
 355                ret = pci_dev_str_match_path(dev, p, &p);
 356                if (ret < 0)
 357                        return ret;
 358                else if (ret)
 359                        goto found;
 360        }
 361
 362        *endptr = p;
 363        return 0;
 364
 365found:
 366        *endptr = p;
 367        return 1;
 368}
 369
 370static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
 371                                   u8 pos, int cap, int *ttl)
 372{
 373        u8 id;
 374        u16 ent;
 375
 376        pci_bus_read_config_byte(bus, devfn, pos, &pos);
 377
 378        while ((*ttl)--) {
 379                if (pos < 0x40)
 380                        break;
 381                pos &= ~3;
 382                pci_bus_read_config_word(bus, devfn, pos, &ent);
 383
 384                id = ent & 0xff;
 385                if (id == 0xff)
 386                        break;
 387                if (id == cap)
 388                        return pos;
 389                pos = (ent >> 8);
 390        }
 391        return 0;
 392}
 393
 394static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
 395                               u8 pos, int cap)
 396{
 397        int ttl = PCI_FIND_CAP_TTL;
 398
 399        return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
 400}
 401
 402int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
 403{
 404        return __pci_find_next_cap(dev->bus, dev->devfn,
 405                                   pos + PCI_CAP_LIST_NEXT, cap);
 406}
 407EXPORT_SYMBOL_GPL(pci_find_next_capability);
 408
 409static int __pci_bus_find_cap_start(struct pci_bus *bus,
 410                                    unsigned int devfn, u8 hdr_type)
 411{
 412        u16 status;
 413
 414        pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
 415        if (!(status & PCI_STATUS_CAP_LIST))
 416                return 0;
 417
 418        switch (hdr_type) {
 419        case PCI_HEADER_TYPE_NORMAL:
 420        case PCI_HEADER_TYPE_BRIDGE:
 421                return PCI_CAPABILITY_LIST;
 422        case PCI_HEADER_TYPE_CARDBUS:
 423                return PCI_CB_CAPABILITY_LIST;
 424        }
 425
 426        return 0;
 427}
 428
 429/**
 430 * pci_find_capability - query for devices' capabilities
 431 * @dev: PCI device to query
 432 * @cap: capability code
 433 *
 434 * Tell if a device supports a given PCI capability.
 435 * Returns the address of the requested capability structure within the
 436 * device's PCI configuration space or 0 in case the device does not
 437 * support it.  Possible values for @cap include:
 438 *
 439 *  %PCI_CAP_ID_PM           Power Management
 440 *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
 441 *  %PCI_CAP_ID_VPD          Vital Product Data
 442 *  %PCI_CAP_ID_SLOTID       Slot Identification
 443 *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
 444 *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
 445 *  %PCI_CAP_ID_PCIX         PCI-X
 446 *  %PCI_CAP_ID_EXP          PCI Express
 447 */
 448int pci_find_capability(struct pci_dev *dev, int cap)
 449{
 450        int pos;
 451
 452        pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 453        if (pos)
 454                pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
 455
 456        return pos;
 457}
 458EXPORT_SYMBOL(pci_find_capability);
 459
 460/**
 461 * pci_bus_find_capability - query for devices' capabilities
 462 * @bus: the PCI bus to query
 463 * @devfn: PCI device to query
 464 * @cap: capability code
 465 *
 466 * Like pci_find_capability() but works for PCI devices that do not have a
 467 * pci_dev structure set up yet.
 468 *
 469 * Returns the address of the requested capability structure within the
 470 * device's PCI configuration space or 0 in case the device does not
 471 * support it.
 472 */
 473int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
 474{
 475        int pos;
 476        u8 hdr_type;
 477
 478        pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
 479
 480        pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
 481        if (pos)
 482                pos = __pci_find_next_cap(bus, devfn, pos, cap);
 483
 484        return pos;
 485}
 486EXPORT_SYMBOL(pci_bus_find_capability);
 487
 488/**
 489 * pci_find_next_ext_capability - Find an extended capability
 490 * @dev: PCI device to query
 491 * @start: address at which to start looking (0 to start at beginning of list)
 492 * @cap: capability code
 493 *
 494 * Returns the address of the next matching extended capability structure
 495 * within the device's PCI configuration space or 0 if the device does
 496 * not support it.  Some capabilities can occur several times, e.g., the
 497 * vendor-specific capability, and this provides a way to find them all.
 498 */
 499int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
 500{
 501        u32 header;
 502        int ttl;
 503        int pos = PCI_CFG_SPACE_SIZE;
 504
 505        /* minimum 8 bytes per capability */
 506        ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
 507
 508        if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
 509                return 0;
 510
 511        if (start)
 512                pos = start;
 513
 514        if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 515                return 0;
 516
 517        /*
 518         * If we have no capabilities, this is indicated by cap ID,
 519         * cap version and next pointer all being 0.
 520         */
 521        if (header == 0)
 522                return 0;
 523
 524        while (ttl-- > 0) {
 525                if (PCI_EXT_CAP_ID(header) == cap && pos != start)
 526                        return pos;
 527
 528                pos = PCI_EXT_CAP_NEXT(header);
 529                if (pos < PCI_CFG_SPACE_SIZE)
 530                        break;
 531
 532                if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 533                        break;
 534        }
 535
 536        return 0;
 537}
 538EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
 539
 540/**
 541 * pci_find_ext_capability - Find an extended capability
 542 * @dev: PCI device to query
 543 * @cap: capability code
 544 *
 545 * Returns the address of the requested extended capability structure
 546 * within the device's PCI configuration space or 0 if the device does
 547 * not support it.  Possible values for @cap include:
 548 *
 549 *  %PCI_EXT_CAP_ID_ERR         Advanced Error Reporting
 550 *  %PCI_EXT_CAP_ID_VC          Virtual Channel
 551 *  %PCI_EXT_CAP_ID_DSN         Device Serial Number
 552 *  %PCI_EXT_CAP_ID_PWR         Power Budgeting
 553 */
 554int pci_find_ext_capability(struct pci_dev *dev, int cap)
 555{
 556        return pci_find_next_ext_capability(dev, 0, cap);
 557}
 558EXPORT_SYMBOL_GPL(pci_find_ext_capability);
 559
 560static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
 561{
 562        int rc, ttl = PCI_FIND_CAP_TTL;
 563        u8 cap, mask;
 564
 565        if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
 566                mask = HT_3BIT_CAP_MASK;
 567        else
 568                mask = HT_5BIT_CAP_MASK;
 569
 570        pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
 571                                      PCI_CAP_ID_HT, &ttl);
 572        while (pos) {
 573                rc = pci_read_config_byte(dev, pos + 3, &cap);
 574                if (rc != PCIBIOS_SUCCESSFUL)
 575                        return 0;
 576
 577                if ((cap & mask) == ht_cap)
 578                        return pos;
 579
 580                pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
 581                                              pos + PCI_CAP_LIST_NEXT,
 582                                              PCI_CAP_ID_HT, &ttl);
 583        }
 584
 585        return 0;
 586}
 587/**
 588 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
 589 * @dev: PCI device to query
 590 * @pos: Position from which to continue searching
 591 * @ht_cap: Hypertransport capability code
 592 *
 593 * To be used in conjunction with pci_find_ht_capability() to search for
 594 * all capabilities matching @ht_cap. @pos should always be a value returned
 595 * from pci_find_ht_capability().
 596 *
 597 * NB. To be 100% safe against broken PCI devices, the caller should take
 598 * steps to avoid an infinite loop.
 599 */
 600int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
 601{
 602        return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
 603}
 604EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
 605
 606/**
 607 * pci_find_ht_capability - query a device's Hypertransport capabilities
 608 * @dev: PCI device to query
 609 * @ht_cap: Hypertransport capability code
 610 *
 611 * Tell if a device supports a given Hypertransport capability.
 612 * Returns an address within the device's PCI configuration space
 613 * or 0 in case the device does not support the request capability.
 614 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
 615 * which has a Hypertransport capability matching @ht_cap.
 616 */
 617int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
 618{
 619        int pos;
 620
 621        pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 622        if (pos)
 623                pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
 624
 625        return pos;
 626}
 627EXPORT_SYMBOL_GPL(pci_find_ht_capability);
 628
 629/**
 630 * pci_find_parent_resource - return resource region of parent bus of given
 631 *                            region
 632 * @dev: PCI device structure contains resources to be searched
 633 * @res: child resource record for which parent is sought
 634 *
 635 * For given resource region of given device, return the resource region of
 636 * parent bus the given region is contained in.
 637 */
 638struct resource *pci_find_parent_resource(const struct pci_dev *dev,
 639                                          struct resource *res)
 640{
 641        const struct pci_bus *bus = dev->bus;
 642        struct resource *r;
 643        int i;
 644
 645        pci_bus_for_each_resource(bus, r, i) {
 646                if (!r)
 647                        continue;
 648                if (resource_contains(r, res)) {
 649
 650                        /*
 651                         * If the window is prefetchable but the BAR is
 652                         * not, the allocator made a mistake.
 653                         */
 654                        if (r->flags & IORESOURCE_PREFETCH &&
 655                            !(res->flags & IORESOURCE_PREFETCH))
 656                                return NULL;
 657
 658                        /*
 659                         * If we're below a transparent bridge, there may
 660                         * be both a positively-decoded aperture and a
 661                         * subtractively-decoded region that contain the BAR.
 662                         * We want the positively-decoded one, so this depends
 663                         * on pci_bus_for_each_resource() giving us those
 664                         * first.
 665                         */
 666                        return r;
 667                }
 668        }
 669        return NULL;
 670}
 671EXPORT_SYMBOL(pci_find_parent_resource);
 672
 673/**
 674 * pci_find_resource - Return matching PCI device resource
 675 * @dev: PCI device to query
 676 * @res: Resource to look for
 677 *
 678 * Goes over standard PCI resources (BARs) and checks if the given resource
 679 * is partially or fully contained in any of them. In that case the
 680 * matching resource is returned, %NULL otherwise.
 681 */
 682struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
 683{
 684        int i;
 685
 686        for (i = 0; i < PCI_STD_NUM_BARS; i++) {
 687                struct resource *r = &dev->resource[i];
 688
 689                if (r->start && resource_contains(r, res))
 690                        return r;
 691        }
 692
 693        return NULL;
 694}
 695EXPORT_SYMBOL(pci_find_resource);
 696
 697/**
 698 * pci_find_pcie_root_port - return PCIe Root Port
 699 * @dev: PCI device to query
 700 *
 701 * Traverse up the parent chain and return the PCIe Root Port PCI Device
 702 * for a given PCI Device.
 703 */
 704struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
 705{
 706        struct pci_dev *bridge, *highest_pcie_bridge = dev;
 707
 708        bridge = pci_upstream_bridge(dev);
 709        while (bridge && pci_is_pcie(bridge)) {
 710                highest_pcie_bridge = bridge;
 711                bridge = pci_upstream_bridge(bridge);
 712        }
 713
 714        if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
 715                return NULL;
 716
 717        return highest_pcie_bridge;
 718}
 719EXPORT_SYMBOL(pci_find_pcie_root_port);
 720
 721/**
 722 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
 723 * @dev: the PCI device to operate on
 724 * @pos: config space offset of status word
 725 * @mask: mask of bit(s) to care about in status word
 726 *
 727 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
 728 */
 729int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
 730{
 731        int i;
 732
 733        /* Wait for Transaction Pending bit clean */
 734        for (i = 0; i < 4; i++) {
 735                u16 status;
 736                if (i)
 737                        msleep((1 << (i - 1)) * 100);
 738
 739                pci_read_config_word(dev, pos, &status);
 740                if (!(status & mask))
 741                        return 1;
 742        }
 743
 744        return 0;
 745}
 746
 747/**
 748 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
 749 * @dev: PCI device to have its BARs restored
 750 *
 751 * Restore the BAR values for a given device, so as to make it
 752 * accessible by its driver.
 753 */
 754static void pci_restore_bars(struct pci_dev *dev)
 755{
 756        int i;
 757
 758        for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
 759                pci_update_resource(dev, i);
 760}
 761
 762static const struct pci_platform_pm_ops *pci_platform_pm;
 763
 764int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
 765{
 766        if (!ops->is_manageable || !ops->set_state  || !ops->get_state ||
 767            !ops->choose_state  || !ops->set_wakeup || !ops->need_resume)
 768                return -EINVAL;
 769        pci_platform_pm = ops;
 770        return 0;
 771}
 772
 773static inline bool platform_pci_power_manageable(struct pci_dev *dev)
 774{
 775        return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
 776}
 777
 778static inline int platform_pci_set_power_state(struct pci_dev *dev,
 779                                               pci_power_t t)
 780{
 781        return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
 782}
 783
 784static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
 785{
 786        return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
 787}
 788
 789static inline void platform_pci_refresh_power_state(struct pci_dev *dev)
 790{
 791        if (pci_platform_pm && pci_platform_pm->refresh_state)
 792                pci_platform_pm->refresh_state(dev);
 793}
 794
 795static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
 796{
 797        return pci_platform_pm ?
 798                        pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
 799}
 800
 801static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
 802{
 803        return pci_platform_pm ?
 804                        pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
 805}
 806
 807static inline bool platform_pci_need_resume(struct pci_dev *dev)
 808{
 809        return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
 810}
 811
 812static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
 813{
 814        return pci_platform_pm ? pci_platform_pm->bridge_d3(dev) : false;
 815}
 816
 817/**
 818 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
 819 *                           given PCI device
 820 * @dev: PCI device to handle.
 821 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
 822 *
 823 * RETURN VALUE:
 824 * -EINVAL if the requested state is invalid.
 825 * -EIO if device does not support PCI PM or its PM capabilities register has a
 826 * wrong version, or device doesn't support the requested state.
 827 * 0 if device already is in the requested state.
 828 * 0 if device's power state has been successfully changed.
 829 */
 830static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
 831{
 832        u16 pmcsr;
 833        bool need_restore = false;
 834
 835        /* Check if we're already there */
 836        if (dev->current_state == state)
 837                return 0;
 838
 839        if (!dev->pm_cap)
 840                return -EIO;
 841
 842        if (state < PCI_D0 || state > PCI_D3hot)
 843                return -EINVAL;
 844
 845        /*
 846         * Validate transition: We can enter D0 from any state, but if
 847         * we're already in a low-power state, we can only go deeper.  E.g.,
 848         * we can go from D1 to D3, but we can't go directly from D3 to D1;
 849         * we'd have to go from D3 to D0, then to D1.
 850         */
 851        if (state != PCI_D0 && dev->current_state <= PCI_D3cold
 852            && dev->current_state > state) {
 853                pci_err(dev, "invalid power transition (from %s to %s)\n",
 854                        pci_power_name(dev->current_state),
 855                        pci_power_name(state));
 856                return -EINVAL;
 857        }
 858
 859        /* Check if this device supports the desired state */
 860        if ((state == PCI_D1 && !dev->d1_support)
 861           || (state == PCI_D2 && !dev->d2_support))
 862                return -EIO;
 863
 864        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 865        if (pmcsr == (u16) ~0) {
 866                pci_err(dev, "can't change power state from %s to %s (config space inaccessible)\n",
 867                        pci_power_name(dev->current_state),
 868                        pci_power_name(state));
 869                return -EIO;
 870        }
 871
 872        /*
 873         * If we're (effectively) in D3, force entire word to 0.
 874         * This doesn't affect PME_Status, disables PME_En, and
 875         * sets PowerState to 0.
 876         */
 877        switch (dev->current_state) {
 878        case PCI_D0:
 879        case PCI_D1:
 880        case PCI_D2:
 881                pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
 882                pmcsr |= state;
 883                break;
 884        case PCI_D3hot:
 885        case PCI_D3cold:
 886        case PCI_UNKNOWN: /* Boot-up */
 887                if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
 888                 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
 889                        need_restore = true;
 890                /* Fall-through - force to D0 */
 891        default:
 892                pmcsr = 0;
 893                break;
 894        }
 895
 896        /* Enter specified state */
 897        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
 898
 899        /*
 900         * Mandatory power management transition delays; see PCI PM 1.1
 901         * 5.6.1 table 18
 902         */
 903        if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
 904                pci_dev_d3_sleep(dev);
 905        else if (state == PCI_D2 || dev->current_state == PCI_D2)
 906                msleep(PCI_PM_D2_DELAY);
 907
 908        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 909        dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
 910        if (dev->current_state != state)
 911                pci_info_ratelimited(dev, "refused to change power state from %s to %s\n",
 912                         pci_power_name(dev->current_state),
 913                         pci_power_name(state));
 914
 915        /*
 916         * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
 917         * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
 918         * from D3hot to D0 _may_ perform an internal reset, thereby
 919         * going to "D0 Uninitialized" rather than "D0 Initialized".
 920         * For example, at least some versions of the 3c905B and the
 921         * 3c556B exhibit this behaviour.
 922         *
 923         * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
 924         * devices in a D3hot state at boot.  Consequently, we need to
 925         * restore at least the BARs so that the device will be
 926         * accessible to its driver.
 927         */
 928        if (need_restore)
 929                pci_restore_bars(dev);
 930
 931        if (dev->bus->self)
 932                pcie_aspm_pm_state_change(dev->bus->self);
 933
 934        return 0;
 935}
 936
 937/**
 938 * pci_update_current_state - Read power state of given device and cache it
 939 * @dev: PCI device to handle.
 940 * @state: State to cache in case the device doesn't have the PM capability
 941 *
 942 * The power state is read from the PMCSR register, which however is
 943 * inaccessible in D3cold.  The platform firmware is therefore queried first
 944 * to detect accessibility of the register.  In case the platform firmware
 945 * reports an incorrect state or the device isn't power manageable by the
 946 * platform at all, we try to detect D3cold by testing accessibility of the
 947 * vendor ID in config space.
 948 */
 949void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
 950{
 951        if (platform_pci_get_power_state(dev) == PCI_D3cold ||
 952            !pci_device_is_present(dev)) {
 953                dev->current_state = PCI_D3cold;
 954        } else if (dev->pm_cap) {
 955                u16 pmcsr;
 956
 957                pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 958                dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
 959        } else {
 960                dev->current_state = state;
 961        }
 962}
 963
 964/**
 965 * pci_refresh_power_state - Refresh the given device's power state data
 966 * @dev: Target PCI device.
 967 *
 968 * Ask the platform to refresh the devices power state information and invoke
 969 * pci_update_current_state() to update its current PCI power state.
 970 */
 971void pci_refresh_power_state(struct pci_dev *dev)
 972{
 973        if (platform_pci_power_manageable(dev))
 974                platform_pci_refresh_power_state(dev);
 975
 976        pci_update_current_state(dev, dev->current_state);
 977}
 978
 979/**
 980 * pci_platform_power_transition - Use platform to change device power state
 981 * @dev: PCI device to handle.
 982 * @state: State to put the device into.
 983 */
 984int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
 985{
 986        int error;
 987
 988        if (platform_pci_power_manageable(dev)) {
 989                error = platform_pci_set_power_state(dev, state);
 990                if (!error)
 991                        pci_update_current_state(dev, state);
 992        } else
 993                error = -ENODEV;
 994
 995        if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
 996                dev->current_state = PCI_D0;
 997
 998        return error;
 999}
1000EXPORT_SYMBOL_GPL(pci_platform_power_transition);
1001
1002/**
1003 * pci_wakeup - Wake up a PCI device
1004 * @pci_dev: Device to handle.
1005 * @ign: ignored parameter
1006 */
1007static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
1008{
1009        pci_wakeup_event(pci_dev);
1010        pm_request_resume(&pci_dev->dev);
1011        return 0;
1012}
1013
1014/**
1015 * pci_wakeup_bus - Walk given bus and wake up devices on it
1016 * @bus: Top bus of the subtree to walk.
1017 */
1018void pci_wakeup_bus(struct pci_bus *bus)
1019{
1020        if (bus)
1021                pci_walk_bus(bus, pci_wakeup, NULL);
1022}
1023
1024static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
1025{
1026        int delay = 1;
1027        u32 id;
1028
1029        /*
1030         * After reset, the device should not silently discard config
1031         * requests, but it may still indicate that it needs more time by
1032         * responding to them with CRS completions.  The Root Port will
1033         * generally synthesize ~0 data to complete the read (except when
1034         * CRS SV is enabled and the read was for the Vendor ID; in that
1035         * case it synthesizes 0x0001 data).
1036         *
1037         * Wait for the device to return a non-CRS completion.  Read the
1038         * Command register instead of Vendor ID so we don't have to
1039         * contend with the CRS SV value.
1040         */
1041        pci_read_config_dword(dev, PCI_COMMAND, &id);
1042        while (id == ~0) {
1043                if (delay > timeout) {
1044                        pci_warn(dev, "not ready %dms after %s; giving up\n",
1045                                 delay - 1, reset_type);
1046                        return -ENOTTY;
1047                }
1048
1049                if (delay > 1000)
1050                        pci_info(dev, "not ready %dms after %s; waiting\n",
1051                                 delay - 1, reset_type);
1052
1053                msleep(delay);
1054                delay *= 2;
1055                pci_read_config_dword(dev, PCI_COMMAND, &id);
1056        }
1057
1058        if (delay > 1000)
1059                pci_info(dev, "ready %dms after %s\n", delay - 1,
1060                         reset_type);
1061
1062        return 0;
1063}
1064
1065/**
1066 * pci_power_up - Put the given device into D0
1067 * @dev: PCI device to power up
1068 */
1069int pci_power_up(struct pci_dev *dev)
1070{
1071        pci_platform_power_transition(dev, PCI_D0);
1072
1073        /*
1074         * Mandatory power management transition delays are handled in
1075         * pci_pm_resume_noirq() and pci_pm_runtime_resume() of the
1076         * corresponding bridge.
1077         */
1078        if (dev->runtime_d3cold) {
1079                /*
1080                 * When powering on a bridge from D3cold, the whole hierarchy
1081                 * may be powered on into D0uninitialized state, resume them to
1082                 * give them a chance to suspend again
1083                 */
1084                pci_wakeup_bus(dev->subordinate);
1085        }
1086
1087        return pci_raw_set_power_state(dev, PCI_D0);
1088}
1089
1090/**
1091 * __pci_dev_set_current_state - Set current state of a PCI device
1092 * @dev: Device to handle
1093 * @data: pointer to state to be set
1094 */
1095static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1096{
1097        pci_power_t state = *(pci_power_t *)data;
1098
1099        dev->current_state = state;
1100        return 0;
1101}
1102
1103/**
1104 * pci_bus_set_current_state - Walk given bus and set current state of devices
1105 * @bus: Top bus of the subtree to walk.
1106 * @state: state to be set
1107 */
1108void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1109{
1110        if (bus)
1111                pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1112}
1113
1114/**
1115 * pci_set_power_state - Set the power state of a PCI device
1116 * @dev: PCI device to handle.
1117 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1118 *
1119 * Transition a device to a new power state, using the platform firmware and/or
1120 * the device's PCI PM registers.
1121 *
1122 * RETURN VALUE:
1123 * -EINVAL if the requested state is invalid.
1124 * -EIO if device does not support PCI PM or its PM capabilities register has a
1125 * wrong version, or device doesn't support the requested state.
1126 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1127 * 0 if device already is in the requested state.
1128 * 0 if the transition is to D3 but D3 is not supported.
1129 * 0 if device's power state has been successfully changed.
1130 */
1131int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1132{
1133        int error;
1134
1135        /* Bound the state we're entering */
1136        if (state > PCI_D3cold)
1137                state = PCI_D3cold;
1138        else if (state < PCI_D0)
1139                state = PCI_D0;
1140        else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1141
1142                /*
1143                 * If the device or the parent bridge do not support PCI
1144                 * PM, ignore the request if we're doing anything other
1145                 * than putting it into D0 (which would only happen on
1146                 * boot).
1147                 */
1148                return 0;
1149
1150        /* Check if we're already there */
1151        if (dev->current_state == state)
1152                return 0;
1153
1154        if (state == PCI_D0)
1155                return pci_power_up(dev);
1156
1157        /*
1158         * This device is quirked not to be put into D3, so don't put it in
1159         * D3
1160         */
1161        if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1162                return 0;
1163
1164        /*
1165         * To put device in D3cold, we put device into D3hot in native
1166         * way, then put device into D3cold with platform ops
1167         */
1168        error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1169                                        PCI_D3hot : state);
1170
1171        if (pci_platform_power_transition(dev, state))
1172                return error;
1173
1174        /* Powering off a bridge may power off the whole hierarchy */
1175        if (state == PCI_D3cold)
1176                pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1177
1178        return 0;
1179}
1180EXPORT_SYMBOL(pci_set_power_state);
1181
1182/**
1183 * pci_choose_state - Choose the power state of a PCI device
1184 * @dev: PCI device to be suspended
1185 * @state: target sleep state for the whole system. This is the value
1186 *         that is passed to suspend() function.
1187 *
1188 * Returns PCI power state suitable for given device and given system
1189 * message.
1190 */
1191pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
1192{
1193        pci_power_t ret;
1194
1195        if (!dev->pm_cap)
1196                return PCI_D0;
1197
1198        ret = platform_pci_choose_state(dev);
1199        if (ret != PCI_POWER_ERROR)
1200                return ret;
1201
1202        switch (state.event) {
1203        case PM_EVENT_ON:
1204                return PCI_D0;
1205        case PM_EVENT_FREEZE:
1206        case PM_EVENT_PRETHAW:
1207                /* REVISIT both freeze and pre-thaw "should" use D0 */
1208        case PM_EVENT_SUSPEND:
1209        case PM_EVENT_HIBERNATE:
1210                return PCI_D3hot;
1211        default:
1212                pci_info(dev, "unrecognized suspend event %d\n",
1213                         state.event);
1214                BUG();
1215        }
1216        return PCI_D0;
1217}
1218EXPORT_SYMBOL(pci_choose_state);
1219
1220#define PCI_EXP_SAVE_REGS       7
1221
1222static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1223                                                       u16 cap, bool extended)
1224{
1225        struct pci_cap_saved_state *tmp;
1226
1227        hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1228                if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1229                        return tmp;
1230        }
1231        return NULL;
1232}
1233
1234struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1235{
1236        return _pci_find_saved_cap(dev, cap, false);
1237}
1238
1239struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1240{
1241        return _pci_find_saved_cap(dev, cap, true);
1242}
1243
1244static int pci_save_pcie_state(struct pci_dev *dev)
1245{
1246        int i = 0;
1247        struct pci_cap_saved_state *save_state;
1248        u16 *cap;
1249
1250        if (!pci_is_pcie(dev))
1251                return 0;
1252
1253        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1254        if (!save_state) {
1255                pci_err(dev, "buffer not found in %s\n", __func__);
1256                return -ENOMEM;
1257        }
1258
1259        cap = (u16 *)&save_state->cap.data[0];
1260        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1261        pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1262        pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1263        pcie_capability_read_word(dev, PCI_EXP_RTCTL,  &cap[i++]);
1264        pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1265        pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1266        pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1267
1268        return 0;
1269}
1270
1271static void pci_restore_pcie_state(struct pci_dev *dev)
1272{
1273        int i = 0;
1274        struct pci_cap_saved_state *save_state;
1275        u16 *cap;
1276
1277        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1278        if (!save_state)
1279                return;
1280
1281        cap = (u16 *)&save_state->cap.data[0];
1282        pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1283        pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1284        pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1285        pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1286        pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1287        pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1288        pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1289}
1290
1291static int pci_save_pcix_state(struct pci_dev *dev)
1292{
1293        int pos;
1294        struct pci_cap_saved_state *save_state;
1295
1296        pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1297        if (!pos)
1298                return 0;
1299
1300        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1301        if (!save_state) {
1302                pci_err(dev, "buffer not found in %s\n", __func__);
1303                return -ENOMEM;
1304        }
1305
1306        pci_read_config_word(dev, pos + PCI_X_CMD,
1307                             (u16 *)save_state->cap.data);
1308
1309        return 0;
1310}
1311
1312static void pci_restore_pcix_state(struct pci_dev *dev)
1313{
1314        int i = 0, pos;
1315        struct pci_cap_saved_state *save_state;
1316        u16 *cap;
1317
1318        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1319        pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1320        if (!save_state || !pos)
1321                return;
1322        cap = (u16 *)&save_state->cap.data[0];
1323
1324        pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1325}
1326
1327static void pci_save_ltr_state(struct pci_dev *dev)
1328{
1329        int ltr;
1330        struct pci_cap_saved_state *save_state;
1331        u16 *cap;
1332
1333        if (!pci_is_pcie(dev))
1334                return;
1335
1336        ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1337        if (!ltr)
1338                return;
1339
1340        save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1341        if (!save_state) {
1342                pci_err(dev, "no suspend buffer for LTR; ASPM issues possible after resume\n");
1343                return;
1344        }
1345
1346        cap = (u16 *)&save_state->cap.data[0];
1347        pci_read_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, cap++);
1348        pci_read_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, cap++);
1349}
1350
1351static void pci_restore_ltr_state(struct pci_dev *dev)
1352{
1353        struct pci_cap_saved_state *save_state;
1354        int ltr;
1355        u16 *cap;
1356
1357        save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1358        ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1359        if (!save_state || !ltr)
1360                return;
1361
1362        cap = (u16 *)&save_state->cap.data[0];
1363        pci_write_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, *cap++);
1364        pci_write_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, *cap++);
1365}
1366
1367/**
1368 * pci_save_state - save the PCI configuration space of a device before
1369 *                  suspending
1370 * @dev: PCI device that we're dealing with
1371 */
1372int pci_save_state(struct pci_dev *dev)
1373{
1374        int i;
1375        /* XXX: 100% dword access ok here? */
1376        for (i = 0; i < 16; i++) {
1377                pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1378                pci_dbg(dev, "saving config space at offset %#x (reading %#x)\n",
1379                        i * 4, dev->saved_config_space[i]);
1380        }
1381        dev->state_saved = true;
1382
1383        i = pci_save_pcie_state(dev);
1384        if (i != 0)
1385                return i;
1386
1387        i = pci_save_pcix_state(dev);
1388        if (i != 0)
1389                return i;
1390
1391        pci_save_ltr_state(dev);
1392        pci_save_dpc_state(dev);
1393        pci_save_aer_state(dev);
1394        return pci_save_vc_state(dev);
1395}
1396EXPORT_SYMBOL(pci_save_state);
1397
1398static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1399                                     u32 saved_val, int retry, bool force)
1400{
1401        u32 val;
1402
1403        pci_read_config_dword(pdev, offset, &val);
1404        if (!force && val == saved_val)
1405                return;
1406
1407        for (;;) {
1408                pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1409                        offset, val, saved_val);
1410                pci_write_config_dword(pdev, offset, saved_val);
1411                if (retry-- <= 0)
1412                        return;
1413
1414                pci_read_config_dword(pdev, offset, &val);
1415                if (val == saved_val)
1416                        return;
1417
1418                mdelay(1);
1419        }
1420}
1421
1422static void pci_restore_config_space_range(struct pci_dev *pdev,
1423                                           int start, int end, int retry,
1424                                           bool force)
1425{
1426        int index;
1427
1428        for (index = end; index >= start; index--)
1429                pci_restore_config_dword(pdev, 4 * index,
1430                                         pdev->saved_config_space[index],
1431                                         retry, force);
1432}
1433
1434static void pci_restore_config_space(struct pci_dev *pdev)
1435{
1436        if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1437                pci_restore_config_space_range(pdev, 10, 15, 0, false);
1438                /* Restore BARs before the command register. */
1439                pci_restore_config_space_range(pdev, 4, 9, 10, false);
1440                pci_restore_config_space_range(pdev, 0, 3, 0, false);
1441        } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1442                pci_restore_config_space_range(pdev, 12, 15, 0, false);
1443
1444                /*
1445                 * Force rewriting of prefetch registers to avoid S3 resume
1446                 * issues on Intel PCI bridges that occur when these
1447                 * registers are not explicitly written.
1448                 */
1449                pci_restore_config_space_range(pdev, 9, 11, 0, true);
1450                pci_restore_config_space_range(pdev, 0, 8, 0, false);
1451        } else {
1452                pci_restore_config_space_range(pdev, 0, 15, 0, false);
1453        }
1454}
1455
1456static void pci_restore_rebar_state(struct pci_dev *pdev)
1457{
1458        unsigned int pos, nbars, i;
1459        u32 ctrl;
1460
1461        pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1462        if (!pos)
1463                return;
1464
1465        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1466        nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1467                    PCI_REBAR_CTRL_NBAR_SHIFT;
1468
1469        for (i = 0; i < nbars; i++, pos += 8) {
1470                struct resource *res;
1471                int bar_idx, size;
1472
1473                pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1474                bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1475                res = pdev->resource + bar_idx;
1476                size = ilog2(resource_size(res)) - 20;
1477                ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1478                ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1479                pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1480        }
1481}
1482
1483/**
1484 * pci_restore_state - Restore the saved state of a PCI device
1485 * @dev: PCI device that we're dealing with
1486 */
1487void pci_restore_state(struct pci_dev *dev)
1488{
1489        if (!dev->state_saved)
1490                return;
1491
1492        /*
1493         * Restore max latencies (in the LTR capability) before enabling
1494         * LTR itself (in the PCIe capability).
1495         */
1496        pci_restore_ltr_state(dev);
1497
1498        pci_restore_pcie_state(dev);
1499        pci_restore_pasid_state(dev);
1500        pci_restore_pri_state(dev);
1501        pci_restore_ats_state(dev);
1502        pci_restore_vc_state(dev);
1503        pci_restore_rebar_state(dev);
1504        pci_restore_dpc_state(dev);
1505
1506        pci_cleanup_aer_error_status_regs(dev);
1507        pci_restore_aer_state(dev);
1508
1509        pci_restore_config_space(dev);
1510
1511        pci_restore_pcix_state(dev);
1512        pci_restore_msi_state(dev);
1513
1514        /* Restore ACS and IOV configuration state */
1515        pci_enable_acs(dev);
1516        pci_restore_iov_state(dev);
1517
1518        dev->state_saved = false;
1519}
1520EXPORT_SYMBOL(pci_restore_state);
1521
1522struct pci_saved_state {
1523        u32 config_space[16];
1524        struct pci_cap_saved_data cap[0];
1525};
1526
1527/**
1528 * pci_store_saved_state - Allocate and return an opaque struct containing
1529 *                         the device saved state.
1530 * @dev: PCI device that we're dealing with
1531 *
1532 * Return NULL if no state or error.
1533 */
1534struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1535{
1536        struct pci_saved_state *state;
1537        struct pci_cap_saved_state *tmp;
1538        struct pci_cap_saved_data *cap;
1539        size_t size;
1540
1541        if (!dev->state_saved)
1542                return NULL;
1543
1544        size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1545
1546        hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1547                size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1548
1549        state = kzalloc(size, GFP_KERNEL);
1550        if (!state)
1551                return NULL;
1552
1553        memcpy(state->config_space, dev->saved_config_space,
1554               sizeof(state->config_space));
1555
1556        cap = state->cap;
1557        hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1558                size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1559                memcpy(cap, &tmp->cap, len);
1560                cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1561        }
1562        /* Empty cap_save terminates list */
1563
1564        return state;
1565}
1566EXPORT_SYMBOL_GPL(pci_store_saved_state);
1567
1568/**
1569 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1570 * @dev: PCI device that we're dealing with
1571 * @state: Saved state returned from pci_store_saved_state()
1572 */
1573int pci_load_saved_state(struct pci_dev *dev,
1574                         struct pci_saved_state *state)
1575{
1576        struct pci_cap_saved_data *cap;
1577
1578        dev->state_saved = false;
1579
1580        if (!state)
1581                return 0;
1582
1583        memcpy(dev->saved_config_space, state->config_space,
1584               sizeof(state->config_space));
1585
1586        cap = state->cap;
1587        while (cap->size) {
1588                struct pci_cap_saved_state *tmp;
1589
1590                tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1591                if (!tmp || tmp->cap.size != cap->size)
1592                        return -EINVAL;
1593
1594                memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1595                cap = (struct pci_cap_saved_data *)((u8 *)cap +
1596                       sizeof(struct pci_cap_saved_data) + cap->size);
1597        }
1598
1599        dev->state_saved = true;
1600        return 0;
1601}
1602EXPORT_SYMBOL_GPL(pci_load_saved_state);
1603
1604/**
1605 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1606 *                                 and free the memory allocated for it.
1607 * @dev: PCI device that we're dealing with
1608 * @state: Pointer to saved state returned from pci_store_saved_state()
1609 */
1610int pci_load_and_free_saved_state(struct pci_dev *dev,
1611                                  struct pci_saved_state **state)
1612{
1613        int ret = pci_load_saved_state(dev, *state);
1614        kfree(*state);
1615        *state = NULL;
1616        return ret;
1617}
1618EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1619
1620int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1621{
1622        return pci_enable_resources(dev, bars);
1623}
1624
1625static int do_pci_enable_device(struct pci_dev *dev, int bars)
1626{
1627        int err;
1628        struct pci_dev *bridge;
1629        u16 cmd;
1630        u8 pin;
1631
1632        err = pci_set_power_state(dev, PCI_D0);
1633        if (err < 0 && err != -EIO)
1634                return err;
1635
1636        bridge = pci_upstream_bridge(dev);
1637        if (bridge)
1638                pcie_aspm_powersave_config_link(bridge);
1639
1640        err = pcibios_enable_device(dev, bars);
1641        if (err < 0)
1642                return err;
1643        pci_fixup_device(pci_fixup_enable, dev);
1644
1645        if (dev->msi_enabled || dev->msix_enabled)
1646                return 0;
1647
1648        pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1649        if (pin) {
1650                pci_read_config_word(dev, PCI_COMMAND, &cmd);
1651                if (cmd & PCI_COMMAND_INTX_DISABLE)
1652                        pci_write_config_word(dev, PCI_COMMAND,
1653                                              cmd & ~PCI_COMMAND_INTX_DISABLE);
1654        }
1655
1656        return 0;
1657}
1658
1659/**
1660 * pci_reenable_device - Resume abandoned device
1661 * @dev: PCI device to be resumed
1662 *
1663 * NOTE: This function is a backend of pci_default_resume() and is not supposed
1664 * to be called by normal code, write proper resume handler and use it instead.
1665 */
1666int pci_reenable_device(struct pci_dev *dev)
1667{
1668        if (pci_is_enabled(dev))
1669                return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1670        return 0;
1671}
1672EXPORT_SYMBOL(pci_reenable_device);
1673
1674static void pci_enable_bridge(struct pci_dev *dev)
1675{
1676        struct pci_dev *bridge;
1677        int retval;
1678
1679        bridge = pci_upstream_bridge(dev);
1680        if (bridge)
1681                pci_enable_bridge(bridge);
1682
1683        if (pci_is_enabled(dev)) {
1684                if (!dev->is_busmaster)
1685                        pci_set_master(dev);
1686                return;
1687        }
1688
1689        retval = pci_enable_device(dev);
1690        if (retval)
1691                pci_err(dev, "Error enabling bridge (%d), continuing\n",
1692                        retval);
1693        pci_set_master(dev);
1694}
1695
1696static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1697{
1698        struct pci_dev *bridge;
1699        int err;
1700        int i, bars = 0;
1701
1702        /*
1703         * Power state could be unknown at this point, either due to a fresh
1704         * boot or a device removal call.  So get the current power state
1705         * so that things like MSI message writing will behave as expected
1706         * (e.g. if the device really is in D0 at enable time).
1707         */
1708        if (dev->pm_cap) {
1709                u16 pmcsr;
1710                pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1711                dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1712        }
1713
1714        if (atomic_inc_return(&dev->enable_cnt) > 1)
1715                return 0;               /* already enabled */
1716
1717        bridge = pci_upstream_bridge(dev);
1718        if (bridge)
1719                pci_enable_bridge(bridge);
1720
1721        /* only skip sriov related */
1722        for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1723                if (dev->resource[i].flags & flags)
1724                        bars |= (1 << i);
1725        for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1726                if (dev->resource[i].flags & flags)
1727                        bars |= (1 << i);
1728
1729        err = do_pci_enable_device(dev, bars);
1730        if (err < 0)
1731                atomic_dec(&dev->enable_cnt);
1732        return err;
1733}
1734
1735/**
1736 * pci_enable_device_io - Initialize a device for use with IO space
1737 * @dev: PCI device to be initialized
1738 *
1739 * Initialize device before it's used by a driver. Ask low-level code
1740 * to enable I/O resources. Wake up the device if it was suspended.
1741 * Beware, this function can fail.
1742 */
1743int pci_enable_device_io(struct pci_dev *dev)
1744{
1745        return pci_enable_device_flags(dev, IORESOURCE_IO);
1746}
1747EXPORT_SYMBOL(pci_enable_device_io);
1748
1749/**
1750 * pci_enable_device_mem - Initialize a device for use with Memory space
1751 * @dev: PCI device to be initialized
1752 *
1753 * Initialize device before it's used by a driver. Ask low-level code
1754 * to enable Memory resources. Wake up the device if it was suspended.
1755 * Beware, this function can fail.
1756 */
1757int pci_enable_device_mem(struct pci_dev *dev)
1758{
1759        return pci_enable_device_flags(dev, IORESOURCE_MEM);
1760}
1761EXPORT_SYMBOL(pci_enable_device_mem);
1762
1763/**
1764 * pci_enable_device - Initialize device before it's used by a driver.
1765 * @dev: PCI device to be initialized
1766 *
1767 * Initialize device before it's used by a driver. Ask low-level code
1768 * to enable I/O and memory. Wake up the device if it was suspended.
1769 * Beware, this function can fail.
1770 *
1771 * Note we don't actually enable the device many times if we call
1772 * this function repeatedly (we just increment the count).
1773 */
1774int pci_enable_device(struct pci_dev *dev)
1775{
1776        return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1777}
1778EXPORT_SYMBOL(pci_enable_device);
1779
1780/*
1781 * Managed PCI resources.  This manages device on/off, INTx/MSI/MSI-X
1782 * on/off and BAR regions.  pci_dev itself records MSI/MSI-X status, so
1783 * there's no need to track it separately.  pci_devres is initialized
1784 * when a device is enabled using managed PCI device enable interface.
1785 */
1786struct pci_devres {
1787        unsigned int enabled:1;
1788        unsigned int pinned:1;
1789        unsigned int orig_intx:1;
1790        unsigned int restore_intx:1;
1791        unsigned int mwi:1;
1792        u32 region_mask;
1793};
1794
1795static void pcim_release(struct device *gendev, void *res)
1796{
1797        struct pci_dev *dev = to_pci_dev(gendev);
1798        struct pci_devres *this = res;
1799        int i;
1800
1801        if (dev->msi_enabled)
1802                pci_disable_msi(dev);
1803        if (dev->msix_enabled)
1804                pci_disable_msix(dev);
1805
1806        for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1807                if (this->region_mask & (1 << i))
1808                        pci_release_region(dev, i);
1809
1810        if (this->mwi)
1811                pci_clear_mwi(dev);
1812
1813        if (this->restore_intx)
1814                pci_intx(dev, this->orig_intx);
1815
1816        if (this->enabled && !this->pinned)
1817                pci_disable_device(dev);
1818}
1819
1820static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1821{
1822        struct pci_devres *dr, *new_dr;
1823
1824        dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1825        if (dr)
1826                return dr;
1827
1828        new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1829        if (!new_dr)
1830                return NULL;
1831        return devres_get(&pdev->dev, new_dr, NULL, NULL);
1832}
1833
1834static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1835{
1836        if (pci_is_managed(pdev))
1837                return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1838        return NULL;
1839}
1840
1841/**
1842 * pcim_enable_device - Managed pci_enable_device()
1843 * @pdev: PCI device to be initialized
1844 *
1845 * Managed pci_enable_device().
1846 */
1847int pcim_enable_device(struct pci_dev *pdev)
1848{
1849        struct pci_devres *dr;
1850        int rc;
1851
1852        dr = get_pci_dr(pdev);
1853        if (unlikely(!dr))
1854                return -ENOMEM;
1855        if (dr->enabled)
1856                return 0;
1857
1858        rc = pci_enable_device(pdev);
1859        if (!rc) {
1860                pdev->is_managed = 1;
1861                dr->enabled = 1;
1862        }
1863        return rc;
1864}
1865EXPORT_SYMBOL(pcim_enable_device);
1866
1867/**
1868 * pcim_pin_device - Pin managed PCI device
1869 * @pdev: PCI device to pin
1870 *
1871 * Pin managed PCI device @pdev.  Pinned device won't be disabled on
1872 * driver detach.  @pdev must have been enabled with
1873 * pcim_enable_device().
1874 */
1875void pcim_pin_device(struct pci_dev *pdev)
1876{
1877        struct pci_devres *dr;
1878
1879        dr = find_pci_dr(pdev);
1880        WARN_ON(!dr || !dr->enabled);
1881        if (dr)
1882                dr->pinned = 1;
1883}
1884EXPORT_SYMBOL(pcim_pin_device);
1885
1886/*
1887 * pcibios_add_device - provide arch specific hooks when adding device dev
1888 * @dev: the PCI device being added
1889 *
1890 * Permits the platform to provide architecture specific functionality when
1891 * devices are added. This is the default implementation. Architecture
1892 * implementations can override this.
1893 */
1894int __weak pcibios_add_device(struct pci_dev *dev)
1895{
1896        return 0;
1897}
1898
1899/**
1900 * pcibios_release_device - provide arch specific hooks when releasing
1901 *                          device dev
1902 * @dev: the PCI device being released
1903 *
1904 * Permits the platform to provide architecture specific functionality when
1905 * devices are released. This is the default implementation. Architecture
1906 * implementations can override this.
1907 */
1908void __weak pcibios_release_device(struct pci_dev *dev) {}
1909
1910/**
1911 * pcibios_disable_device - disable arch specific PCI resources for device dev
1912 * @dev: the PCI device to disable
1913 *
1914 * Disables architecture specific PCI resources for the device. This
1915 * is the default implementation. Architecture implementations can
1916 * override this.
1917 */
1918void __weak pcibios_disable_device(struct pci_dev *dev) {}
1919
1920/**
1921 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1922 * @irq: ISA IRQ to penalize
1923 * @active: IRQ active or not
1924 *
1925 * Permits the platform to provide architecture-specific functionality when
1926 * penalizing ISA IRQs. This is the default implementation. Architecture
1927 * implementations can override this.
1928 */
1929void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1930
1931static void do_pci_disable_device(struct pci_dev *dev)
1932{
1933        u16 pci_command;
1934
1935        pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1936        if (pci_command & PCI_COMMAND_MASTER) {
1937                pci_command &= ~PCI_COMMAND_MASTER;
1938                pci_write_config_word(dev, PCI_COMMAND, pci_command);
1939        }
1940
1941        pcibios_disable_device(dev);
1942}
1943
1944/**
1945 * pci_disable_enabled_device - Disable device without updating enable_cnt
1946 * @dev: PCI device to disable
1947 *
1948 * NOTE: This function is a backend of PCI power management routines and is
1949 * not supposed to be called drivers.
1950 */
1951void pci_disable_enabled_device(struct pci_dev *dev)
1952{
1953        if (pci_is_enabled(dev))
1954                do_pci_disable_device(dev);
1955}
1956
1957/**
1958 * pci_disable_device - Disable PCI device after use
1959 * @dev: PCI device to be disabled
1960 *
1961 * Signal to the system that the PCI device is not in use by the system
1962 * anymore.  This only involves disabling PCI bus-mastering, if active.
1963 *
1964 * Note we don't actually disable the device until all callers of
1965 * pci_enable_device() have called pci_disable_device().
1966 */
1967void pci_disable_device(struct pci_dev *dev)
1968{
1969        struct pci_devres *dr;
1970
1971        dr = find_pci_dr(dev);
1972        if (dr)
1973                dr->enabled = 0;
1974
1975        dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1976                      "disabling already-disabled device");
1977
1978        if (atomic_dec_return(&dev->enable_cnt) != 0)
1979                return;
1980
1981        do_pci_disable_device(dev);
1982
1983        dev->is_busmaster = 0;
1984}
1985EXPORT_SYMBOL(pci_disable_device);
1986
1987/**
1988 * pcibios_set_pcie_reset_state - set reset state for device dev
1989 * @dev: the PCIe device reset
1990 * @state: Reset state to enter into
1991 *
1992 * Set the PCIe reset state for the device. This is the default
1993 * implementation. Architecture implementations can override this.
1994 */
1995int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1996                                        enum pcie_reset_state state)
1997{
1998        return -EINVAL;
1999}
2000
2001/**
2002 * pci_set_pcie_reset_state - set reset state for device dev
2003 * @dev: the PCIe device reset
2004 * @state: Reset state to enter into
2005 *
2006 * Sets the PCI reset state for the device.
2007 */
2008int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
2009{
2010        return pcibios_set_pcie_reset_state(dev, state);
2011}
2012EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
2013
2014/**
2015 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
2016 * @dev: PCIe root port or event collector.
2017 */
2018void pcie_clear_root_pme_status(struct pci_dev *dev)
2019{
2020        pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
2021}
2022
2023/**
2024 * pci_check_pme_status - Check if given device has generated PME.
2025 * @dev: Device to check.
2026 *
2027 * Check the PME status of the device and if set, clear it and clear PME enable
2028 * (if set).  Return 'true' if PME status and PME enable were both set or
2029 * 'false' otherwise.
2030 */
2031bool pci_check_pme_status(struct pci_dev *dev)
2032{
2033        int pmcsr_pos;
2034        u16 pmcsr;
2035        bool ret = false;
2036
2037        if (!dev->pm_cap)
2038                return false;
2039
2040        pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
2041        pci_read_config_word(dev, pmcsr_pos, &pmcsr);
2042        if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
2043                return false;
2044
2045        /* Clear PME status. */
2046        pmcsr |= PCI_PM_CTRL_PME_STATUS;
2047        if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
2048                /* Disable PME to avoid interrupt flood. */
2049                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2050                ret = true;
2051        }
2052
2053        pci_write_config_word(dev, pmcsr_pos, pmcsr);
2054
2055        return ret;
2056}
2057
2058/**
2059 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2060 * @dev: Device to handle.
2061 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2062 *
2063 * Check if @dev has generated PME and queue a resume request for it in that
2064 * case.
2065 */
2066static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2067{
2068        if (pme_poll_reset && dev->pme_poll)
2069                dev->pme_poll = false;
2070
2071        if (pci_check_pme_status(dev)) {
2072                pci_wakeup_event(dev);
2073                pm_request_resume(&dev->dev);
2074        }
2075        return 0;
2076}
2077
2078/**
2079 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2080 * @bus: Top bus of the subtree to walk.
2081 */
2082void pci_pme_wakeup_bus(struct pci_bus *bus)
2083{
2084        if (bus)
2085                pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2086}
2087
2088
2089/**
2090 * pci_pme_capable - check the capability of PCI device to generate PME#
2091 * @dev: PCI device to handle.
2092 * @state: PCI state from which device will issue PME#.
2093 */
2094bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2095{
2096        if (!dev->pm_cap)
2097                return false;
2098
2099        return !!(dev->pme_support & (1 << state));
2100}
2101EXPORT_SYMBOL(pci_pme_capable);
2102
2103static void pci_pme_list_scan(struct work_struct *work)
2104{
2105        struct pci_pme_device *pme_dev, *n;
2106
2107        mutex_lock(&pci_pme_list_mutex);
2108        list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2109                if (pme_dev->dev->pme_poll) {
2110                        struct pci_dev *bridge;
2111
2112                        bridge = pme_dev->dev->bus->self;
2113                        /*
2114                         * If bridge is in low power state, the
2115                         * configuration space of subordinate devices
2116                         * may be not accessible
2117                         */
2118                        if (bridge && bridge->current_state != PCI_D0)
2119                                continue;
2120                        /*
2121                         * If the device is in D3cold it should not be
2122                         * polled either.
2123                         */
2124                        if (pme_dev->dev->current_state == PCI_D3cold)
2125                                continue;
2126
2127                        pci_pme_wakeup(pme_dev->dev, NULL);
2128                } else {
2129                        list_del(&pme_dev->list);
2130                        kfree(pme_dev);
2131                }
2132        }
2133        if (!list_empty(&pci_pme_list))
2134                queue_delayed_work(system_freezable_wq, &pci_pme_work,
2135                                   msecs_to_jiffies(PME_TIMEOUT));
2136        mutex_unlock(&pci_pme_list_mutex);
2137}
2138
2139static void __pci_pme_active(struct pci_dev *dev, bool enable)
2140{
2141        u16 pmcsr;
2142
2143        if (!dev->pme_support)
2144                return;
2145
2146        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2147        /* Clear PME_Status by writing 1 to it and enable PME# */
2148        pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2149        if (!enable)
2150                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2151
2152        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2153}
2154
2155/**
2156 * pci_pme_restore - Restore PME configuration after config space restore.
2157 * @dev: PCI device to update.
2158 */
2159void pci_pme_restore(struct pci_dev *dev)
2160{
2161        u16 pmcsr;
2162
2163        if (!dev->pme_support)
2164                return;
2165
2166        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2167        if (dev->wakeup_prepared) {
2168                pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2169                pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2170        } else {
2171                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2172                pmcsr |= PCI_PM_CTRL_PME_STATUS;
2173        }
2174        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2175}
2176
2177/**
2178 * pci_pme_active - enable or disable PCI device's PME# function
2179 * @dev: PCI device to handle.
2180 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2181 *
2182 * The caller must verify that the device is capable of generating PME# before
2183 * calling this function with @enable equal to 'true'.
2184 */
2185void pci_pme_active(struct pci_dev *dev, bool enable)
2186{
2187        __pci_pme_active(dev, enable);
2188
2189        /*
2190         * PCI (as opposed to PCIe) PME requires that the device have
2191         * its PME# line hooked up correctly. Not all hardware vendors
2192         * do this, so the PME never gets delivered and the device
2193         * remains asleep. The easiest way around this is to
2194         * periodically walk the list of suspended devices and check
2195         * whether any have their PME flag set. The assumption is that
2196         * we'll wake up often enough anyway that this won't be a huge
2197         * hit, and the power savings from the devices will still be a
2198         * win.
2199         *
2200         * Although PCIe uses in-band PME message instead of PME# line
2201         * to report PME, PME does not work for some PCIe devices in
2202         * reality.  For example, there are devices that set their PME
2203         * status bits, but don't really bother to send a PME message;
2204         * there are PCI Express Root Ports that don't bother to
2205         * trigger interrupts when they receive PME messages from the
2206         * devices below.  So PME poll is used for PCIe devices too.
2207         */
2208
2209        if (dev->pme_poll) {
2210                struct pci_pme_device *pme_dev;
2211                if (enable) {
2212                        pme_dev = kmalloc(sizeof(struct pci_pme_device),
2213                                          GFP_KERNEL);
2214                        if (!pme_dev) {
2215                                pci_warn(dev, "can't enable PME#\n");
2216                                return;
2217                        }
2218                        pme_dev->dev = dev;
2219                        mutex_lock(&pci_pme_list_mutex);
2220                        list_add(&pme_dev->list, &pci_pme_list);
2221                        if (list_is_singular(&pci_pme_list))
2222                                queue_delayed_work(system_freezable_wq,
2223                                                   &pci_pme_work,
2224                                                   msecs_to_jiffies(PME_TIMEOUT));
2225                        mutex_unlock(&pci_pme_list_mutex);
2226                } else {
2227                        mutex_lock(&pci_pme_list_mutex);
2228                        list_for_each_entry(pme_dev, &pci_pme_list, list) {
2229                                if (pme_dev->dev == dev) {
2230                                        list_del(&pme_dev->list);
2231                                        kfree(pme_dev);
2232                                        break;
2233                                }
2234                        }
2235                        mutex_unlock(&pci_pme_list_mutex);
2236                }
2237        }
2238
2239        pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2240}
2241EXPORT_SYMBOL(pci_pme_active);
2242
2243/**
2244 * __pci_enable_wake - enable PCI device as wakeup event source
2245 * @dev: PCI device affected
2246 * @state: PCI state from which device will issue wakeup events
2247 * @enable: True to enable event generation; false to disable
2248 *
2249 * This enables the device as a wakeup event source, or disables it.
2250 * When such events involves platform-specific hooks, those hooks are
2251 * called automatically by this routine.
2252 *
2253 * Devices with legacy power management (no standard PCI PM capabilities)
2254 * always require such platform hooks.
2255 *
2256 * RETURN VALUE:
2257 * 0 is returned on success
2258 * -EINVAL is returned if device is not supposed to wake up the system
2259 * Error code depending on the platform is returned if both the platform and
2260 * the native mechanism fail to enable the generation of wake-up events
2261 */
2262static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2263{
2264        int ret = 0;
2265
2266        /*
2267         * Bridges that are not power-manageable directly only signal
2268         * wakeup on behalf of subordinate devices which is set up
2269         * elsewhere, so skip them. However, bridges that are
2270         * power-manageable may signal wakeup for themselves (for example,
2271         * on a hotplug event) and they need to be covered here.
2272         */
2273        if (!pci_power_manageable(dev))
2274                return 0;
2275
2276        /* Don't do the same thing twice in a row for one device. */
2277        if (!!enable == !!dev->wakeup_prepared)
2278                return 0;
2279
2280        /*
2281         * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2282         * Anderson we should be doing PME# wake enable followed by ACPI wake
2283         * enable.  To disable wake-up we call the platform first, for symmetry.
2284         */
2285
2286        if (enable) {
2287                int error;
2288
2289                if (pci_pme_capable(dev, state))
2290                        pci_pme_active(dev, true);
2291                else
2292                        ret = 1;
2293                error = platform_pci_set_wakeup(dev, true);
2294                if (ret)
2295                        ret = error;
2296                if (!ret)
2297                        dev->wakeup_prepared = true;
2298        } else {
2299                platform_pci_set_wakeup(dev, false);
2300                pci_pme_active(dev, false);
2301                dev->wakeup_prepared = false;
2302        }
2303
2304        return ret;
2305}
2306
2307/**
2308 * pci_enable_wake - change wakeup settings for a PCI device
2309 * @pci_dev: Target device
2310 * @state: PCI state from which device will issue wakeup events
2311 * @enable: Whether or not to enable event generation
2312 *
2313 * If @enable is set, check device_may_wakeup() for the device before calling
2314 * __pci_enable_wake() for it.
2315 */
2316int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2317{
2318        if (enable && !device_may_wakeup(&pci_dev->dev))
2319                return -EINVAL;
2320
2321        return __pci_enable_wake(pci_dev, state, enable);
2322}
2323EXPORT_SYMBOL(pci_enable_wake);
2324
2325/**
2326 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2327 * @dev: PCI device to prepare
2328 * @enable: True to enable wake-up event generation; false to disable
2329 *
2330 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2331 * and this function allows them to set that up cleanly - pci_enable_wake()
2332 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2333 * ordering constraints.
2334 *
2335 * This function only returns error code if the device is not allowed to wake
2336 * up the system from sleep or it is not capable of generating PME# from both
2337 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2338 */
2339int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2340{
2341        return pci_pme_capable(dev, PCI_D3cold) ?
2342                        pci_enable_wake(dev, PCI_D3cold, enable) :
2343                        pci_enable_wake(dev, PCI_D3hot, enable);
2344}
2345EXPORT_SYMBOL(pci_wake_from_d3);
2346
2347/**
2348 * pci_target_state - find an appropriate low power state for a given PCI dev
2349 * @dev: PCI device
2350 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2351 *
2352 * Use underlying platform code to find a supported low power state for @dev.
2353 * If the platform can't manage @dev, return the deepest state from which it
2354 * can generate wake events, based on any available PME info.
2355 */
2356static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2357{
2358        pci_power_t target_state = PCI_D3hot;
2359
2360        if (platform_pci_power_manageable(dev)) {
2361                /*
2362                 * Call the platform to find the target state for the device.
2363                 */
2364                pci_power_t state = platform_pci_choose_state(dev);
2365
2366                switch (state) {
2367                case PCI_POWER_ERROR:
2368                case PCI_UNKNOWN:
2369                        break;
2370                case PCI_D1:
2371                case PCI_D2:
2372                        if (pci_no_d1d2(dev))
2373                                break;
2374                        /* else, fall through */
2375                default:
2376                        target_state = state;
2377                }
2378
2379                return target_state;
2380        }
2381
2382        if (!dev->pm_cap)
2383                target_state = PCI_D0;
2384
2385        /*
2386         * If the device is in D3cold even though it's not power-manageable by
2387         * the platform, it may have been powered down by non-standard means.
2388         * Best to let it slumber.
2389         */
2390        if (dev->current_state == PCI_D3cold)
2391                target_state = PCI_D3cold;
2392
2393        if (wakeup) {
2394                /*
2395                 * Find the deepest state from which the device can generate
2396                 * PME#.
2397                 */
2398                if (dev->pme_support) {
2399                        while (target_state
2400                              && !(dev->pme_support & (1 << target_state)))
2401                                target_state--;
2402                }
2403        }
2404
2405        return target_state;
2406}
2407
2408/**
2409 * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2410 *                        into a sleep state
2411 * @dev: Device to handle.
2412 *
2413 * Choose the power state appropriate for the device depending on whether
2414 * it can wake up the system and/or is power manageable by the platform
2415 * (PCI_D3hot is the default) and put the device into that state.
2416 */
2417int pci_prepare_to_sleep(struct pci_dev *dev)
2418{
2419        bool wakeup = device_may_wakeup(&dev->dev);
2420        pci_power_t target_state = pci_target_state(dev, wakeup);
2421        int error;
2422
2423        if (target_state == PCI_POWER_ERROR)
2424                return -EIO;
2425
2426        pci_enable_wake(dev, target_state, wakeup);
2427
2428        error = pci_set_power_state(dev, target_state);
2429
2430        if (error)
2431                pci_enable_wake(dev, target_state, false);
2432
2433        return error;
2434}
2435EXPORT_SYMBOL(pci_prepare_to_sleep);
2436
2437/**
2438 * pci_back_from_sleep - turn PCI device on during system-wide transition
2439 *                       into working state
2440 * @dev: Device to handle.
2441 *
2442 * Disable device's system wake-up capability and put it into D0.
2443 */
2444int pci_back_from_sleep(struct pci_dev *dev)
2445{
2446        pci_enable_wake(dev, PCI_D0, false);
2447        return pci_set_power_state(dev, PCI_D0);
2448}
2449EXPORT_SYMBOL(pci_back_from_sleep);
2450
2451/**
2452 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2453 * @dev: PCI device being suspended.
2454 *
2455 * Prepare @dev to generate wake-up events at run time and put it into a low
2456 * power state.
2457 */
2458int pci_finish_runtime_suspend(struct pci_dev *dev)
2459{
2460        pci_power_t target_state;
2461        int error;
2462
2463        target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2464        if (target_state == PCI_POWER_ERROR)
2465                return -EIO;
2466
2467        dev->runtime_d3cold = target_state == PCI_D3cold;
2468
2469        __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2470
2471        error = pci_set_power_state(dev, target_state);
2472
2473        if (error) {
2474                pci_enable_wake(dev, target_state, false);
2475                dev->runtime_d3cold = false;
2476        }
2477
2478        return error;
2479}
2480
2481/**
2482 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2483 * @dev: Device to check.
2484 *
2485 * Return true if the device itself is capable of generating wake-up events
2486 * (through the platform or using the native PCIe PME) or if the device supports
2487 * PME and one of its upstream bridges can generate wake-up events.
2488 */
2489bool pci_dev_run_wake(struct pci_dev *dev)
2490{
2491        struct pci_bus *bus = dev->bus;
2492
2493        if (!dev->pme_support)
2494                return false;
2495
2496        /* PME-capable in principle, but not from the target power state */
2497        if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2498                return false;
2499
2500        if (device_can_wakeup(&dev->dev))
2501                return true;
2502
2503        while (bus->parent) {
2504                struct pci_dev *bridge = bus->self;
2505
2506                if (device_can_wakeup(&bridge->dev))
2507                        return true;
2508
2509                bus = bus->parent;
2510        }
2511
2512        /* We have reached the root bus. */
2513        if (bus->bridge)
2514                return device_can_wakeup(bus->bridge);
2515
2516        return false;
2517}
2518EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2519
2520/**
2521 * pci_dev_need_resume - Check if it is necessary to resume the device.
2522 * @pci_dev: Device to check.
2523 *
2524 * Return 'true' if the device is not runtime-suspended or it has to be
2525 * reconfigured due to wakeup settings difference between system and runtime
2526 * suspend, or the current power state of it is not suitable for the upcoming
2527 * (system-wide) transition.
2528 */
2529bool pci_dev_need_resume(struct pci_dev *pci_dev)
2530{
2531        struct device *dev = &pci_dev->dev;
2532        pci_power_t target_state;
2533
2534        if (!pm_runtime_suspended(dev) || platform_pci_need_resume(pci_dev))
2535                return true;
2536
2537        target_state = pci_target_state(pci_dev, device_may_wakeup(dev));
2538
2539        /*
2540         * If the earlier platform check has not triggered, D3cold is just power
2541         * removal on top of D3hot, so no need to resume the device in that
2542         * case.
2543         */
2544        return target_state != pci_dev->current_state &&
2545                target_state != PCI_D3cold &&
2546                pci_dev->current_state != PCI_D3hot;
2547}
2548
2549/**
2550 * pci_dev_adjust_pme - Adjust PME setting for a suspended device.
2551 * @pci_dev: Device to check.
2552 *
2553 * If the device is suspended and it is not configured for system wakeup,
2554 * disable PME for it to prevent it from waking up the system unnecessarily.
2555 *
2556 * Note that if the device's power state is D3cold and the platform check in
2557 * pci_dev_need_resume() has not triggered, the device's configuration need not
2558 * be changed.
2559 */
2560void pci_dev_adjust_pme(struct pci_dev *pci_dev)
2561{
2562        struct device *dev = &pci_dev->dev;
2563
2564        spin_lock_irq(&dev->power.lock);
2565
2566        if (pm_runtime_suspended(dev) && !device_may_wakeup(dev) &&
2567            pci_dev->current_state < PCI_D3cold)
2568                __pci_pme_active(pci_dev, false);
2569
2570        spin_unlock_irq(&dev->power.lock);
2571}
2572
2573/**
2574 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2575 * @pci_dev: Device to handle.
2576 *
2577 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2578 * it might have been disabled during the prepare phase of system suspend if
2579 * the device was not configured for system wakeup.
2580 */
2581void pci_dev_complete_resume(struct pci_dev *pci_dev)
2582{
2583        struct device *dev = &pci_dev->dev;
2584
2585        if (!pci_dev_run_wake(pci_dev))
2586                return;
2587
2588        spin_lock_irq(&dev->power.lock);
2589
2590        if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2591                __pci_pme_active(pci_dev, true);
2592
2593        spin_unlock_irq(&dev->power.lock);
2594}
2595
2596void pci_config_pm_runtime_get(struct pci_dev *pdev)
2597{
2598        struct device *dev = &pdev->dev;
2599        struct device *parent = dev->parent;
2600
2601        if (parent)
2602                pm_runtime_get_sync(parent);
2603        pm_runtime_get_noresume(dev);
2604        /*
2605         * pdev->current_state is set to PCI_D3cold during suspending,
2606         * so wait until suspending completes
2607         */
2608        pm_runtime_barrier(dev);
2609        /*
2610         * Only need to resume devices in D3cold, because config
2611         * registers are still accessible for devices suspended but
2612         * not in D3cold.
2613         */
2614        if (pdev->current_state == PCI_D3cold)
2615                pm_runtime_resume(dev);
2616}
2617
2618void pci_config_pm_runtime_put(struct pci_dev *pdev)
2619{
2620        struct device *dev = &pdev->dev;
2621        struct device *parent = dev->parent;
2622
2623        pm_runtime_put(dev);
2624        if (parent)
2625                pm_runtime_put_sync(parent);
2626}
2627
2628static const struct dmi_system_id bridge_d3_blacklist[] = {
2629#ifdef CONFIG_X86
2630        {
2631                /*
2632                 * Gigabyte X299 root port is not marked as hotplug capable
2633                 * which allows Linux to power manage it.  However, this
2634                 * confuses the BIOS SMI handler so don't power manage root
2635                 * ports on that system.
2636                 */
2637                .ident = "X299 DESIGNARE EX-CF",
2638                .matches = {
2639                        DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2640                        DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2641                },
2642        },
2643#endif
2644        { }
2645};
2646
2647/**
2648 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2649 * @bridge: Bridge to check
2650 *
2651 * This function checks if it is possible to move the bridge to D3.
2652 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2653 */
2654bool pci_bridge_d3_possible(struct pci_dev *bridge)
2655{
2656        if (!pci_is_pcie(bridge))
2657                return false;
2658
2659        switch (pci_pcie_type(bridge)) {
2660        case PCI_EXP_TYPE_ROOT_PORT:
2661        case PCI_EXP_TYPE_UPSTREAM:
2662        case PCI_EXP_TYPE_DOWNSTREAM:
2663                if (pci_bridge_d3_disable)
2664                        return false;
2665
2666                /*
2667                 * Hotplug ports handled by firmware in System Management Mode
2668                 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2669                 */
2670                if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2671                        return false;
2672
2673                if (pci_bridge_d3_force)
2674                        return true;
2675
2676                /* Even the oldest 2010 Thunderbolt controller supports D3. */
2677                if (bridge->is_thunderbolt)
2678                        return true;
2679
2680                /* Platform might know better if the bridge supports D3 */
2681                if (platform_pci_bridge_d3(bridge))
2682                        return true;
2683
2684                /*
2685                 * Hotplug ports handled natively by the OS were not validated
2686                 * by vendors for runtime D3 at least until 2018 because there
2687                 * was no OS support.
2688                 */
2689                if (bridge->is_hotplug_bridge)
2690                        return false;
2691
2692                if (dmi_check_system(bridge_d3_blacklist))
2693                        return false;
2694
2695                /*
2696                 * It should be safe to put PCIe ports from 2015 or newer
2697                 * to D3.
2698                 */
2699                if (dmi_get_bios_year() >= 2015)
2700                        return true;
2701                break;
2702        }
2703
2704        return false;
2705}
2706
2707static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2708{
2709        bool *d3cold_ok = data;
2710
2711        if (/* The device needs to be allowed to go D3cold ... */
2712            dev->no_d3cold || !dev->d3cold_allowed ||
2713
2714            /* ... and if it is wakeup capable to do so from D3cold. */
2715            (device_may_wakeup(&dev->dev) &&
2716             !pci_pme_capable(dev, PCI_D3cold)) ||
2717
2718            /* If it is a bridge it must be allowed to go to D3. */
2719            !pci_power_manageable(dev))
2720
2721                *d3cold_ok = false;
2722
2723        return !*d3cold_ok;
2724}
2725
2726/*
2727 * pci_bridge_d3_update - Update bridge D3 capabilities
2728 * @dev: PCI device which is changed
2729 *
2730 * Update upstream bridge PM capabilities accordingly depending on if the
2731 * device PM configuration was changed or the device is being removed.  The
2732 * change is also propagated upstream.
2733 */
2734void pci_bridge_d3_update(struct pci_dev *dev)
2735{
2736        bool remove = !device_is_registered(&dev->dev);
2737        struct pci_dev *bridge;
2738        bool d3cold_ok = true;
2739
2740        bridge = pci_upstream_bridge(dev);
2741        if (!bridge || !pci_bridge_d3_possible(bridge))
2742                return;
2743
2744        /*
2745         * If D3 is currently allowed for the bridge, removing one of its
2746         * children won't change that.
2747         */
2748        if (remove && bridge->bridge_d3)
2749                return;
2750
2751        /*
2752         * If D3 is currently allowed for the bridge and a child is added or
2753         * changed, disallowance of D3 can only be caused by that child, so
2754         * we only need to check that single device, not any of its siblings.
2755         *
2756         * If D3 is currently not allowed for the bridge, checking the device
2757         * first may allow us to skip checking its siblings.
2758         */
2759        if (!remove)
2760                pci_dev_check_d3cold(dev, &d3cold_ok);
2761
2762        /*
2763         * If D3 is currently not allowed for the bridge, this may be caused
2764         * either by the device being changed/removed or any of its siblings,
2765         * so we need to go through all children to find out if one of them
2766         * continues to block D3.
2767         */
2768        if (d3cold_ok && !bridge->bridge_d3)
2769                pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2770                             &d3cold_ok);
2771
2772        if (bridge->bridge_d3 != d3cold_ok) {
2773                bridge->bridge_d3 = d3cold_ok;
2774                /* Propagate change to upstream bridges */
2775                pci_bridge_d3_update(bridge);
2776        }
2777}
2778
2779/**
2780 * pci_d3cold_enable - Enable D3cold for device
2781 * @dev: PCI device to handle
2782 *
2783 * This function can be used in drivers to enable D3cold from the device
2784 * they handle.  It also updates upstream PCI bridge PM capabilities
2785 * accordingly.
2786 */
2787void pci_d3cold_enable(struct pci_dev *dev)
2788{
2789        if (dev->no_d3cold) {
2790                dev->no_d3cold = false;
2791                pci_bridge_d3_update(dev);
2792        }
2793}
2794EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2795
2796/**
2797 * pci_d3cold_disable - Disable D3cold for device
2798 * @dev: PCI device to handle
2799 *
2800 * This function can be used in drivers to disable D3cold from the device
2801 * they handle.  It also updates upstream PCI bridge PM capabilities
2802 * accordingly.
2803 */
2804void pci_d3cold_disable(struct pci_dev *dev)
2805{
2806        if (!dev->no_d3cold) {
2807                dev->no_d3cold = true;
2808                pci_bridge_d3_update(dev);
2809        }
2810}
2811EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2812
2813/**
2814 * pci_pm_init - Initialize PM functions of given PCI device
2815 * @dev: PCI device to handle.
2816 */
2817void pci_pm_init(struct pci_dev *dev)
2818{
2819        int pm;
2820        u16 status;
2821        u16 pmc;
2822
2823        pm_runtime_forbid(&dev->dev);
2824        pm_runtime_set_active(&dev->dev);
2825        pm_runtime_enable(&dev->dev);
2826        device_enable_async_suspend(&dev->dev);
2827        dev->wakeup_prepared = false;
2828
2829        dev->pm_cap = 0;
2830        dev->pme_support = 0;
2831
2832        /* find PCI PM capability in list */
2833        pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2834        if (!pm)
2835                return;
2836        /* Check device's ability to generate PME# */
2837        pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2838
2839        if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2840                pci_err(dev, "unsupported PM cap regs version (%u)\n",
2841                        pmc & PCI_PM_CAP_VER_MASK);
2842                return;
2843        }
2844
2845        dev->pm_cap = pm;
2846        dev->d3_delay = PCI_PM_D3_WAIT;
2847        dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2848        dev->bridge_d3 = pci_bridge_d3_possible(dev);
2849        dev->d3cold_allowed = true;
2850
2851        dev->d1_support = false;
2852        dev->d2_support = false;
2853        if (!pci_no_d1d2(dev)) {
2854                if (pmc & PCI_PM_CAP_D1)
2855                        dev->d1_support = true;
2856                if (pmc & PCI_PM_CAP_D2)
2857                        dev->d2_support = true;
2858
2859                if (dev->d1_support || dev->d2_support)
2860                        pci_info(dev, "supports%s%s\n",
2861                                   dev->d1_support ? " D1" : "",
2862                                   dev->d2_support ? " D2" : "");
2863        }
2864
2865        pmc &= PCI_PM_CAP_PME_MASK;
2866        if (pmc) {
2867                pci_info(dev, "PME# supported from%s%s%s%s%s\n",
2868                         (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2869                         (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2870                         (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2871                         (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2872                         (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2873                dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2874                dev->pme_poll = true;
2875                /*
2876                 * Make device's PM flags reflect the wake-up capability, but
2877                 * let the user space enable it to wake up the system as needed.
2878                 */
2879                device_set_wakeup_capable(&dev->dev, true);
2880                /* Disable the PME# generation functionality */
2881                pci_pme_active(dev, false);
2882        }
2883
2884        pci_read_config_word(dev, PCI_STATUS, &status);
2885        if (status & PCI_STATUS_IMM_READY)
2886                dev->imm_ready = 1;
2887}
2888
2889static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2890{
2891        unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2892
2893        switch (prop) {
2894        case PCI_EA_P_MEM:
2895        case PCI_EA_P_VF_MEM:
2896                flags |= IORESOURCE_MEM;
2897                break;
2898        case PCI_EA_P_MEM_PREFETCH:
2899        case PCI_EA_P_VF_MEM_PREFETCH:
2900                flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2901                break;
2902        case PCI_EA_P_IO:
2903                flags |= IORESOURCE_IO;
2904                break;
2905        default:
2906                return 0;
2907        }
2908
2909        return flags;
2910}
2911
2912static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2913                                            u8 prop)
2914{
2915        if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2916                return &dev->resource[bei];
2917#ifdef CONFIG_PCI_IOV
2918        else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2919                 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2920                return &dev->resource[PCI_IOV_RESOURCES +
2921                                      bei - PCI_EA_BEI_VF_BAR0];
2922#endif
2923        else if (bei == PCI_EA_BEI_ROM)
2924                return &dev->resource[PCI_ROM_RESOURCE];
2925        else
2926                return NULL;
2927}
2928
2929/* Read an Enhanced Allocation (EA) entry */
2930static int pci_ea_read(struct pci_dev *dev, int offset)
2931{
2932        struct resource *res;
2933        int ent_size, ent_offset = offset;
2934        resource_size_t start, end;
2935        unsigned long flags;
2936        u32 dw0, bei, base, max_offset;
2937        u8 prop;
2938        bool support_64 = (sizeof(resource_size_t) >= 8);
2939
2940        pci_read_config_dword(dev, ent_offset, &dw0);
2941        ent_offset += 4;
2942
2943        /* Entry size field indicates DWORDs after 1st */
2944        ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2945
2946        if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2947                goto out;
2948
2949        bei = (dw0 & PCI_EA_BEI) >> 4;
2950        prop = (dw0 & PCI_EA_PP) >> 8;
2951
2952        /*
2953         * If the Property is in the reserved range, try the Secondary
2954         * Property instead.
2955         */
2956        if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2957                prop = (dw0 & PCI_EA_SP) >> 16;
2958        if (prop > PCI_EA_P_BRIDGE_IO)
2959                goto out;
2960
2961        res = pci_ea_get_resource(dev, bei, prop);
2962        if (!res) {
2963                pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
2964                goto out;
2965        }
2966
2967        flags = pci_ea_flags(dev, prop);
2968        if (!flags) {
2969                pci_err(dev, "Unsupported EA properties: %#x\n", prop);
2970                goto out;
2971        }
2972
2973        /* Read Base */
2974        pci_read_config_dword(dev, ent_offset, &base);
2975        start = (base & PCI_EA_FIELD_MASK);
2976        ent_offset += 4;
2977
2978        /* Read MaxOffset */
2979        pci_read_config_dword(dev, ent_offset, &max_offset);
2980        ent_offset += 4;
2981
2982        /* Read Base MSBs (if 64-bit entry) */
2983        if (base & PCI_EA_IS_64) {
2984                u32 base_upper;
2985
2986                pci_read_config_dword(dev, ent_offset, &base_upper);
2987                ent_offset += 4;
2988
2989                flags |= IORESOURCE_MEM_64;
2990
2991                /* entry starts above 32-bit boundary, can't use */
2992                if (!support_64 && base_upper)
2993                        goto out;
2994
2995                if (support_64)
2996                        start |= ((u64)base_upper << 32);
2997        }
2998
2999        end = start + (max_offset | 0x03);
3000
3001        /* Read MaxOffset MSBs (if 64-bit entry) */
3002        if (max_offset & PCI_EA_IS_64) {
3003                u32 max_offset_upper;
3004
3005                pci_read_config_dword(dev, ent_offset, &max_offset_upper);
3006                ent_offset += 4;
3007
3008                flags |= IORESOURCE_MEM_64;
3009
3010                /* entry too big, can't use */
3011                if (!support_64 && max_offset_upper)
3012                        goto out;
3013
3014                if (support_64)
3015                        end += ((u64)max_offset_upper << 32);
3016        }
3017
3018        if (end < start) {
3019                pci_err(dev, "EA Entry crosses address boundary\n");
3020                goto out;
3021        }
3022
3023        if (ent_size != ent_offset - offset) {
3024                pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
3025                        ent_size, ent_offset - offset);
3026                goto out;
3027        }
3028
3029        res->name = pci_name(dev);
3030        res->start = start;
3031        res->end = end;
3032        res->flags = flags;
3033
3034        if (bei <= PCI_EA_BEI_BAR5)
3035                pci_info(dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3036                           bei, res, prop);
3037        else if (bei == PCI_EA_BEI_ROM)
3038                pci_info(dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
3039                           res, prop);
3040        else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
3041                pci_info(dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3042                           bei - PCI_EA_BEI_VF_BAR0, res, prop);
3043        else
3044                pci_info(dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
3045                           bei, res, prop);
3046
3047out:
3048        return offset + ent_size;
3049}
3050
3051/* Enhanced Allocation Initialization */
3052void pci_ea_init(struct pci_dev *dev)
3053{
3054        int ea;
3055        u8 num_ent;
3056        int offset;
3057        int i;
3058
3059        /* find PCI EA capability in list */
3060        ea = pci_find_capability(dev, PCI_CAP_ID_EA);
3061        if (!ea)
3062                return;
3063
3064        /* determine the number of entries */
3065        pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
3066                                        &num_ent);
3067        num_ent &= PCI_EA_NUM_ENT_MASK;
3068
3069        offset = ea + PCI_EA_FIRST_ENT;
3070
3071        /* Skip DWORD 2 for type 1 functions */
3072        if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3073                offset += 4;
3074
3075        /* parse each EA entry */
3076        for (i = 0; i < num_ent; ++i)
3077                offset = pci_ea_read(dev, offset);
3078}
3079
3080static void pci_add_saved_cap(struct pci_dev *pci_dev,
3081        struct pci_cap_saved_state *new_cap)
3082{
3083        hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3084}
3085
3086/**
3087 * _pci_add_cap_save_buffer - allocate buffer for saving given
3088 *                            capability registers
3089 * @dev: the PCI device
3090 * @cap: the capability to allocate the buffer for
3091 * @extended: Standard or Extended capability ID
3092 * @size: requested size of the buffer
3093 */
3094static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3095                                    bool extended, unsigned int size)
3096{
3097        int pos;
3098        struct pci_cap_saved_state *save_state;
3099
3100        if (extended)
3101                pos = pci_find_ext_capability(dev, cap);
3102        else
3103                pos = pci_find_capability(dev, cap);
3104
3105        if (!pos)
3106                return 0;
3107
3108        save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3109        if (!save_state)
3110                return -ENOMEM;
3111
3112        save_state->cap.cap_nr = cap;
3113        save_state->cap.cap_extended = extended;
3114        save_state->cap.size = size;
3115        pci_add_saved_cap(dev, save_state);
3116
3117        return 0;
3118}
3119
3120int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3121{
3122        return _pci_add_cap_save_buffer(dev, cap, false, size);
3123}
3124
3125int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3126{
3127        return _pci_add_cap_save_buffer(dev, cap, true, size);
3128}
3129
3130/**
3131 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3132 * @dev: the PCI device
3133 */
3134void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3135{
3136        int error;
3137
3138        error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3139                                        PCI_EXP_SAVE_REGS * sizeof(u16));
3140        if (error)
3141                pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3142
3143        error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3144        if (error)
3145                pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3146
3147        error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3148                                            2 * sizeof(u16));
3149        if (error)
3150                pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3151
3152        pci_allocate_vc_save_buffers(dev);
3153}
3154
3155void pci_free_cap_save_buffers(struct pci_dev *dev)
3156{
3157        struct pci_cap_saved_state *tmp;
3158        struct hlist_node *n;
3159
3160        hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3161                kfree(tmp);
3162}
3163
3164/**
3165 * pci_configure_ari - enable or disable ARI forwarding
3166 * @dev: the PCI device
3167 *
3168 * If @dev and its upstream bridge both support ARI, enable ARI in the
3169 * bridge.  Otherwise, disable ARI in the bridge.
3170 */
3171void pci_configure_ari(struct pci_dev *dev)
3172{
3173        u32 cap;
3174        struct pci_dev *bridge;
3175
3176        if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3177                return;
3178
3179        bridge = dev->bus->self;
3180        if (!bridge)
3181                return;
3182
3183        pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3184        if (!(cap & PCI_EXP_DEVCAP2_ARI))
3185                return;
3186
3187        if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3188                pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3189                                         PCI_EXP_DEVCTL2_ARI);
3190                bridge->ari_enabled = 1;
3191        } else {
3192                pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3193                                           PCI_EXP_DEVCTL2_ARI);
3194                bridge->ari_enabled = 0;
3195        }
3196}
3197
3198static int pci_acs_enable;
3199
3200/**
3201 * pci_request_acs - ask for ACS to be enabled if supported
3202 */
3203void pci_request_acs(void)
3204{
3205        pci_acs_enable = 1;
3206}
3207
3208static const char *disable_acs_redir_param;
3209
3210/**
3211 * pci_disable_acs_redir - disable ACS redirect capabilities
3212 * @dev: the PCI device
3213 *
3214 * For only devices specified in the disable_acs_redir parameter.
3215 */
3216static void pci_disable_acs_redir(struct pci_dev *dev)
3217{
3218        int ret = 0;
3219        const char *p;
3220        int pos;
3221        u16 ctrl;
3222
3223        if (!disable_acs_redir_param)
3224                return;
3225
3226        p = disable_acs_redir_param;
3227        while (*p) {
3228                ret = pci_dev_str_match(dev, p, &p);
3229                if (ret < 0) {
3230                        pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
3231                                     disable_acs_redir_param);
3232
3233                        break;
3234                } else if (ret == 1) {
3235                        /* Found a match */
3236                        break;
3237                }
3238
3239                if (*p != ';' && *p != ',') {
3240                        /* End of param or invalid format */
3241                        break;
3242                }
3243                p++;
3244        }
3245
3246        if (ret != 1)
3247                return;
3248
3249        if (!pci_dev_specific_disable_acs_redir(dev))
3250                return;
3251
3252        pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3253        if (!pos) {
3254                pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
3255                return;
3256        }
3257
3258        pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3259
3260        /* P2P Request & Completion Redirect */
3261        ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
3262
3263        pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3264
3265        pci_info(dev, "disabled ACS redirect\n");
3266}
3267
3268/**
3269 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
3270 * @dev: the PCI device
3271 */
3272static void pci_std_enable_acs(struct pci_dev *dev)
3273{
3274        int pos;
3275        u16 cap;
3276        u16 ctrl;
3277
3278        pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3279        if (!pos)
3280                return;
3281
3282        pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
3283        pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3284
3285        /* Source Validation */
3286        ctrl |= (cap & PCI_ACS_SV);
3287
3288        /* P2P Request Redirect */
3289        ctrl |= (cap & PCI_ACS_RR);
3290
3291        /* P2P Completion Redirect */
3292        ctrl |= (cap & PCI_ACS_CR);
3293
3294        /* Upstream Forwarding */
3295        ctrl |= (cap & PCI_ACS_UF);
3296
3297        pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3298}
3299
3300/**
3301 * pci_enable_acs - enable ACS if hardware support it
3302 * @dev: the PCI device
3303 */
3304void pci_enable_acs(struct pci_dev *dev)
3305{
3306        if (!pci_acs_enable)
3307                goto disable_acs_redir;
3308
3309        if (!pci_dev_specific_enable_acs(dev))
3310                goto disable_acs_redir;
3311
3312        pci_std_enable_acs(dev);
3313
3314disable_acs_redir:
3315        /*
3316         * Note: pci_disable_acs_redir() must be called even if ACS was not
3317         * enabled by the kernel because it may have been enabled by
3318         * platform firmware.  So if we are told to disable it, we should
3319         * always disable it after setting the kernel's default
3320         * preferences.
3321         */
3322        pci_disable_acs_redir(dev);
3323}
3324
3325static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3326{
3327        int pos;
3328        u16 cap, ctrl;
3329
3330        pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
3331        if (!pos)
3332                return false;
3333
3334        /*
3335         * Except for egress control, capabilities are either required
3336         * or only required if controllable.  Features missing from the
3337         * capability field can therefore be assumed as hard-wired enabled.
3338         */
3339        pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3340        acs_flags &= (cap | PCI_ACS_EC);
3341
3342        pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3343        return (ctrl & acs_flags) == acs_flags;
3344}
3345
3346/**
3347 * pci_acs_enabled - test ACS against required flags for a given device
3348 * @pdev: device to test
3349 * @acs_flags: required PCI ACS flags
3350 *
3351 * Return true if the device supports the provided flags.  Automatically
3352 * filters out flags that are not implemented on multifunction devices.
3353 *
3354 * Note that this interface checks the effective ACS capabilities of the
3355 * device rather than the actual capabilities.  For instance, most single
3356 * function endpoints are not required to support ACS because they have no
3357 * opportunity for peer-to-peer access.  We therefore return 'true'
3358 * regardless of whether the device exposes an ACS capability.  This makes
3359 * it much easier for callers of this function to ignore the actual type
3360 * or topology of the device when testing ACS support.
3361 */
3362bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3363{
3364        int ret;
3365
3366        ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3367        if (ret >= 0)
3368                return ret > 0;
3369
3370        /*
3371         * Conventional PCI and PCI-X devices never support ACS, either
3372         * effectively or actually.  The shared bus topology implies that
3373         * any device on the bus can receive or snoop DMA.
3374         */
3375        if (!pci_is_pcie(pdev))
3376                return false;
3377
3378        switch (pci_pcie_type(pdev)) {
3379        /*
3380         * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3381         * but since their primary interface is PCI/X, we conservatively
3382         * handle them as we would a non-PCIe device.
3383         */
3384        case PCI_EXP_TYPE_PCIE_BRIDGE:
3385        /*
3386         * PCIe 3.0, 6.12.1 excludes ACS on these devices.  "ACS is never
3387         * applicable... must never implement an ACS Extended Capability...".
3388         * This seems arbitrary, but we take a conservative interpretation
3389         * of this statement.
3390         */
3391        case PCI_EXP_TYPE_PCI_BRIDGE:
3392        case PCI_EXP_TYPE_RC_EC:
3393                return false;
3394        /*
3395         * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3396         * implement ACS in order to indicate their peer-to-peer capabilities,
3397         * regardless of whether they are single- or multi-function devices.
3398         */
3399        case PCI_EXP_TYPE_DOWNSTREAM:
3400        case PCI_EXP_TYPE_ROOT_PORT:
3401                return pci_acs_flags_enabled(pdev, acs_flags);
3402        /*
3403         * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3404         * implemented by the remaining PCIe types to indicate peer-to-peer
3405         * capabilities, but only when they are part of a multifunction
3406         * device.  The footnote for section 6.12 indicates the specific
3407         * PCIe types included here.
3408         */
3409        case PCI_EXP_TYPE_ENDPOINT:
3410        case PCI_EXP_TYPE_UPSTREAM:
3411        case PCI_EXP_TYPE_LEG_END:
3412        case PCI_EXP_TYPE_RC_END:
3413                if (!pdev->multifunction)
3414                        break;
3415
3416                return pci_acs_flags_enabled(pdev, acs_flags);
3417        }
3418
3419        /*
3420         * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3421         * to single function devices with the exception of downstream ports.
3422         */
3423        return true;
3424}
3425
3426/**
3427 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3428 * @start: starting downstream device
3429 * @end: ending upstream device or NULL to search to the root bus
3430 * @acs_flags: required flags
3431 *
3432 * Walk up a device tree from start to end testing PCI ACS support.  If
3433 * any step along the way does not support the required flags, return false.
3434 */
3435bool pci_acs_path_enabled(struct pci_dev *start,
3436                          struct pci_dev *end, u16 acs_flags)
3437{
3438        struct pci_dev *pdev, *parent = start;
3439
3440        do {
3441                pdev = parent;
3442
3443                if (!pci_acs_enabled(pdev, acs_flags))
3444                        return false;
3445
3446                if (pci_is_root_bus(pdev->bus))
3447                        return (end == NULL);
3448
3449                parent = pdev->bus->self;
3450        } while (pdev != end);
3451
3452        return true;
3453}
3454
3455/**
3456 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3457 * @pdev: PCI device
3458 * @bar: BAR to find
3459 *
3460 * Helper to find the position of the ctrl register for a BAR.
3461 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3462 * Returns -ENOENT if no ctrl register for the BAR could be found.
3463 */
3464static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3465{
3466        unsigned int pos, nbars, i;
3467        u32 ctrl;
3468
3469        pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3470        if (!pos)
3471                return -ENOTSUPP;
3472
3473        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3474        nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3475                    PCI_REBAR_CTRL_NBAR_SHIFT;
3476
3477        for (i = 0; i < nbars; i++, pos += 8) {
3478                int bar_idx;
3479
3480                pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3481                bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3482                if (bar_idx == bar)
3483                        return pos;
3484        }
3485
3486        return -ENOENT;
3487}
3488
3489/**
3490 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3491 * @pdev: PCI device
3492 * @bar: BAR to query
3493 *
3494 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3495 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3496 */
3497u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3498{
3499        int pos;
3500        u32 cap;
3501
3502        pos = pci_rebar_find_pos(pdev, bar);
3503        if (pos < 0)
3504                return 0;
3505
3506        pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3507        return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3508}
3509
3510/**
3511 * pci_rebar_get_current_size - get the current size of a BAR
3512 * @pdev: PCI device
3513 * @bar: BAR to set size to
3514 *
3515 * Read the size of a BAR from the resizable BAR config.
3516 * Returns size if found or negative error code.
3517 */
3518int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3519{
3520        int pos;
3521        u32 ctrl;
3522
3523        pos = pci_rebar_find_pos(pdev, bar);
3524        if (pos < 0)
3525                return pos;
3526
3527        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3528        return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3529}
3530
3531/**
3532 * pci_rebar_set_size - set a new size for a BAR
3533 * @pdev: PCI device
3534 * @bar: BAR to set size to
3535 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3536 *
3537 * Set the new size of a BAR as defined in the spec.
3538 * Returns zero if resizing was successful, error code otherwise.
3539 */
3540int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3541{
3542        int pos;
3543        u32 ctrl;
3544
3545        pos = pci_rebar_find_pos(pdev, bar);
3546        if (pos < 0)
3547                return pos;
3548
3549        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3550        ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3551        ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3552        pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3553        return 0;
3554}
3555
3556/**
3557 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3558 * @dev: the PCI device
3559 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3560 *      PCI_EXP_DEVCAP2_ATOMIC_COMP32
3561 *      PCI_EXP_DEVCAP2_ATOMIC_COMP64
3562 *      PCI_EXP_DEVCAP2_ATOMIC_COMP128
3563 *
3564 * Return 0 if all upstream bridges support AtomicOp routing, egress
3565 * blocking is disabled on all upstream ports, and the root port supports
3566 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3567 * AtomicOp completion), or negative otherwise.
3568 */
3569int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3570{
3571        struct pci_bus *bus = dev->bus;
3572        struct pci_dev *bridge;
3573        u32 cap, ctl2;
3574
3575        if (!pci_is_pcie(dev))
3576                return -EINVAL;
3577
3578        /*
3579         * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3580         * AtomicOp requesters.  For now, we only support endpoints as
3581         * requesters and root ports as completers.  No endpoints as
3582         * completers, and no peer-to-peer.
3583         */
3584
3585        switch (pci_pcie_type(dev)) {
3586        case PCI_EXP_TYPE_ENDPOINT:
3587        case PCI_EXP_TYPE_LEG_END:
3588        case PCI_EXP_TYPE_RC_END:
3589                break;
3590        default:
3591                return -EINVAL;
3592        }
3593
3594        while (bus->parent) {
3595                bridge = bus->self;
3596
3597                pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3598
3599                switch (pci_pcie_type(bridge)) {
3600                /* Ensure switch ports support AtomicOp routing */
3601                case PCI_EXP_TYPE_UPSTREAM:
3602                case PCI_EXP_TYPE_DOWNSTREAM:
3603                        if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3604                                return -EINVAL;
3605                        break;
3606
3607                /* Ensure root port supports all the sizes we care about */
3608                case PCI_EXP_TYPE_ROOT_PORT:
3609                        if ((cap & cap_mask) != cap_mask)
3610                                return -EINVAL;
3611                        break;
3612                }
3613
3614                /* Ensure upstream ports don't block AtomicOps on egress */
3615                if (pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM) {
3616                        pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3617                                                   &ctl2);
3618                        if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3619                                return -EINVAL;
3620                }
3621
3622                bus = bus->parent;
3623        }
3624
3625        pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3626                                 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3627        return 0;
3628}
3629EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3630
3631/**
3632 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3633 * @dev: the PCI device
3634 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3635 *
3636 * Perform INTx swizzling for a device behind one level of bridge.  This is
3637 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3638 * behind bridges on add-in cards.  For devices with ARI enabled, the slot
3639 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3640 * the PCI Express Base Specification, Revision 2.1)
3641 */
3642u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3643{
3644        int slot;
3645
3646        if (pci_ari_enabled(dev->bus))
3647                slot = 0;
3648        else
3649                slot = PCI_SLOT(dev->devfn);
3650
3651        return (((pin - 1) + slot) % 4) + 1;
3652}
3653
3654int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3655{
3656        u8 pin;
3657
3658        pin = dev->pin;
3659        if (!pin)
3660                return -1;
3661
3662        while (!pci_is_root_bus(dev->bus)) {
3663                pin = pci_swizzle_interrupt_pin(dev, pin);
3664                dev = dev->bus->self;
3665        }
3666        *bridge = dev;
3667        return pin;
3668}
3669
3670/**
3671 * pci_common_swizzle - swizzle INTx all the way to root bridge
3672 * @dev: the PCI device
3673 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3674 *
3675 * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
3676 * bridges all the way up to a PCI root bus.
3677 */
3678u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3679{
3680        u8 pin = *pinp;
3681
3682        while (!pci_is_root_bus(dev->bus)) {
3683                pin = pci_swizzle_interrupt_pin(dev, pin);
3684                dev = dev->bus->self;
3685        }
3686        *pinp = pin;
3687        return PCI_SLOT(dev->devfn);
3688}
3689EXPORT_SYMBOL_GPL(pci_common_swizzle);
3690
3691/**
3692 * pci_release_region - Release a PCI bar
3693 * @pdev: PCI device whose resources were previously reserved by
3694 *        pci_request_region()
3695 * @bar: BAR to release
3696 *
3697 * Releases the PCI I/O and memory resources previously reserved by a
3698 * successful call to pci_request_region().  Call this function only
3699 * after all use of the PCI regions has ceased.
3700 */
3701void pci_release_region(struct pci_dev *pdev, int bar)
3702{
3703        struct pci_devres *dr;
3704
3705        if (pci_resource_len(pdev, bar) == 0)
3706                return;
3707        if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3708                release_region(pci_resource_start(pdev, bar),
3709                                pci_resource_len(pdev, bar));
3710        else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3711                release_mem_region(pci_resource_start(pdev, bar),
3712                                pci_resource_len(pdev, bar));
3713
3714        dr = find_pci_dr(pdev);
3715        if (dr)
3716                dr->region_mask &= ~(1 << bar);
3717}
3718EXPORT_SYMBOL(pci_release_region);
3719
3720/**
3721 * __pci_request_region - Reserved PCI I/O and memory resource
3722 * @pdev: PCI device whose resources are to be reserved
3723 * @bar: BAR to be reserved
3724 * @res_name: Name to be associated with resource.
3725 * @exclusive: whether the region access is exclusive or not
3726 *
3727 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3728 * being reserved by owner @res_name.  Do not access any
3729 * address inside the PCI regions unless this call returns
3730 * successfully.
3731 *
3732 * If @exclusive is set, then the region is marked so that userspace
3733 * is explicitly not allowed to map the resource via /dev/mem or
3734 * sysfs MMIO access.
3735 *
3736 * Returns 0 on success, or %EBUSY on error.  A warning
3737 * message is also printed on failure.
3738 */
3739static int __pci_request_region(struct pci_dev *pdev, int bar,
3740                                const char *res_name, int exclusive)
3741{
3742        struct pci_devres *dr;
3743
3744        if (pci_resource_len(pdev, bar) == 0)
3745                return 0;
3746
3747        if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3748                if (!request_region(pci_resource_start(pdev, bar),
3749                            pci_resource_len(pdev, bar), res_name))
3750                        goto err_out;
3751        } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3752                if (!__request_mem_region(pci_resource_start(pdev, bar),
3753                                        pci_resource_len(pdev, bar), res_name,
3754                                        exclusive))
3755                        goto err_out;
3756        }
3757
3758        dr = find_pci_dr(pdev);
3759        if (dr)
3760                dr->region_mask |= 1 << bar;
3761
3762        return 0;
3763
3764err_out:
3765        pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3766                 &pdev->resource[bar]);
3767        return -EBUSY;
3768}
3769
3770/**
3771 * pci_request_region - Reserve PCI I/O and memory resource
3772 * @pdev: PCI device whose resources are to be reserved
3773 * @bar: BAR to be reserved
3774 * @res_name: Name to be associated with resource
3775 *
3776 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3777 * being reserved by owner @res_name.  Do not access any
3778 * address inside the PCI regions unless this call returns
3779 * successfully.
3780 *
3781 * Returns 0 on success, or %EBUSY on error.  A warning
3782 * message is also printed on failure.
3783 */
3784int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3785{
3786        return __pci_request_region(pdev, bar, res_name, 0);
3787}
3788EXPORT_SYMBOL(pci_request_region);
3789
3790/**
3791 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3792 * @pdev: PCI device whose resources were previously reserved
3793 * @bars: Bitmask of BARs to be released
3794 *
3795 * Release selected PCI I/O and memory resources previously reserved.
3796 * Call this function only after all use of the PCI regions has ceased.
3797 */
3798void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3799{
3800        int i;
3801
3802        for (i = 0; i < PCI_STD_NUM_BARS; i++)
3803                if (bars & (1 << i))
3804                        pci_release_region(pdev, i);
3805}
3806EXPORT_SYMBOL(pci_release_selected_regions);
3807
3808static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3809                                          const char *res_name, int excl)
3810{
3811        int i;
3812
3813        for (i = 0; i < PCI_STD_NUM_BARS; i++)
3814                if (bars & (1 << i))
3815                        if (__pci_request_region(pdev, i, res_name, excl))
3816                                goto err_out;
3817        return 0;
3818
3819err_out:
3820        while (--i >= 0)
3821                if (bars & (1 << i))
3822                        pci_release_region(pdev, i);
3823
3824        return -EBUSY;
3825}
3826
3827
3828/**
3829 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3830 * @pdev: PCI device whose resources are to be reserved
3831 * @bars: Bitmask of BARs to be requested
3832 * @res_name: Name to be associated with resource
3833 */
3834int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3835                                 const char *res_name)
3836{
3837        return __pci_request_selected_regions(pdev, bars, res_name, 0);
3838}
3839EXPORT_SYMBOL(pci_request_selected_regions);
3840
3841int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3842                                           const char *res_name)
3843{
3844        return __pci_request_selected_regions(pdev, bars, res_name,
3845                        IORESOURCE_EXCLUSIVE);
3846}
3847EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3848
3849/**
3850 * pci_release_regions - Release reserved PCI I/O and memory resources
3851 * @pdev: PCI device whose resources were previously reserved by
3852 *        pci_request_regions()
3853 *
3854 * Releases all PCI I/O and memory resources previously reserved by a
3855 * successful call to pci_request_regions().  Call this function only
3856 * after all use of the PCI regions has ceased.
3857 */
3858
3859void pci_release_regions(struct pci_dev *pdev)
3860{
3861        pci_release_selected_regions(pdev, (1 << PCI_STD_NUM_BARS) - 1);
3862}
3863EXPORT_SYMBOL(pci_release_regions);
3864
3865/**
3866 * pci_request_regions - Reserve PCI I/O and memory resources
3867 * @pdev: PCI device whose resources are to be reserved
3868 * @res_name: Name to be associated with resource.
3869 *
3870 * Mark all PCI regions associated with PCI device @pdev as
3871 * being reserved by owner @res_name.  Do not access any
3872 * address inside the PCI regions unless this call returns
3873 * successfully.
3874 *
3875 * Returns 0 on success, or %EBUSY on error.  A warning
3876 * message is also printed on failure.
3877 */
3878int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3879{
3880        return pci_request_selected_regions(pdev,
3881                        ((1 << PCI_STD_NUM_BARS) - 1), res_name);
3882}
3883EXPORT_SYMBOL(pci_request_regions);
3884
3885/**
3886 * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
3887 * @pdev: PCI device whose resources are to be reserved
3888 * @res_name: Name to be associated with resource.
3889 *
3890 * Mark all PCI regions associated with PCI device @pdev as being reserved
3891 * by owner @res_name.  Do not access any address inside the PCI regions
3892 * unless this call returns successfully.
3893 *
3894 * pci_request_regions_exclusive() will mark the region so that /dev/mem
3895 * and the sysfs MMIO access will not be allowed.
3896 *
3897 * Returns 0 on success, or %EBUSY on error.  A warning message is also
3898 * printed on failure.
3899 */
3900int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3901{
3902        return pci_request_selected_regions_exclusive(pdev,
3903                                ((1 << PCI_STD_NUM_BARS) - 1), res_name);
3904}
3905EXPORT_SYMBOL(pci_request_regions_exclusive);
3906
3907/*
3908 * Record the PCI IO range (expressed as CPU physical address + size).
3909 * Return a negative value if an error has occurred, zero otherwise
3910 */
3911int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3912                        resource_size_t size)
3913{
3914        int ret = 0;
3915#ifdef PCI_IOBASE
3916        struct logic_pio_hwaddr *range;
3917
3918        if (!size || addr + size < addr)
3919                return -EINVAL;
3920
3921        range = kzalloc(sizeof(*range), GFP_ATOMIC);
3922        if (!range)
3923                return -ENOMEM;
3924
3925        range->fwnode = fwnode;
3926        range->size = size;
3927        range->hw_start = addr;
3928        range->flags = LOGIC_PIO_CPU_MMIO;
3929
3930        ret = logic_pio_register_range(range);
3931        if (ret)
3932                kfree(range);
3933#endif
3934
3935        return ret;
3936}
3937
3938phys_addr_t pci_pio_to_address(unsigned long pio)
3939{
3940        phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3941
3942#ifdef PCI_IOBASE
3943        if (pio >= MMIO_UPPER_LIMIT)
3944                return address;
3945
3946        address = logic_pio_to_hwaddr(pio);
3947#endif
3948
3949        return address;
3950}
3951
3952unsigned long __weak pci_address_to_pio(phys_addr_t address)
3953{
3954#ifdef PCI_IOBASE
3955        return logic_pio_trans_cpuaddr(address);
3956#else
3957        if (address > IO_SPACE_LIMIT)
3958                return (unsigned long)-1;
3959
3960        return (unsigned long) address;
3961#endif
3962}
3963
3964/**
3965 * pci_remap_iospace - Remap the memory mapped I/O space
3966 * @res: Resource describing the I/O space
3967 * @phys_addr: physical address of range to be mapped
3968 *
3969 * Remap the memory mapped I/O space described by the @res and the CPU
3970 * physical address @phys_addr into virtual address space.  Only
3971 * architectures that have memory mapped IO functions defined (and the
3972 * PCI_IOBASE value defined) should call this function.
3973 */
3974int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3975{
3976#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3977        unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3978
3979        if (!(res->flags & IORESOURCE_IO))
3980                return -EINVAL;
3981
3982        if (res->end > IO_SPACE_LIMIT)
3983                return -EINVAL;
3984
3985        return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3986                                  pgprot_device(PAGE_KERNEL));
3987#else
3988        /*
3989         * This architecture does not have memory mapped I/O space,
3990         * so this function should never be called
3991         */
3992        WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3993        return -ENODEV;
3994#endif
3995}
3996EXPORT_SYMBOL(pci_remap_iospace);
3997
3998/**
3999 * pci_unmap_iospace - Unmap the memory mapped I/O space
4000 * @res: resource to be unmapped
4001 *
4002 * Unmap the CPU virtual address @res from virtual address space.  Only
4003 * architectures that have memory mapped IO functions defined (and the
4004 * PCI_IOBASE value defined) should call this function.
4005 */
4006void pci_unmap_iospace(struct resource *res)
4007{
4008#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4009        unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4010
4011        unmap_kernel_range(vaddr, resource_size(res));
4012#endif
4013}
4014EXPORT_SYMBOL(pci_unmap_iospace);
4015
4016static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
4017{
4018        struct resource **res = ptr;
4019
4020        pci_unmap_iospace(*res);
4021}
4022
4023/**
4024 * devm_pci_remap_iospace - Managed pci_remap_iospace()
4025 * @dev: Generic device to remap IO address for
4026 * @res: Resource describing the I/O space
4027 * @phys_addr: physical address of range to be mapped
4028 *
4029 * Managed pci_remap_iospace().  Map is automatically unmapped on driver
4030 * detach.
4031 */
4032int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
4033                           phys_addr_t phys_addr)
4034{
4035        const struct resource **ptr;
4036        int error;
4037
4038        ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
4039        if (!ptr)
4040                return -ENOMEM;
4041
4042        error = pci_remap_iospace(res, phys_addr);
4043        if (error) {
4044                devres_free(ptr);
4045        } else  {
4046                *ptr = res;
4047                devres_add(dev, ptr);
4048        }
4049
4050        return error;
4051}
4052EXPORT_SYMBOL(devm_pci_remap_iospace);
4053
4054/**
4055 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
4056 * @dev: Generic device to remap IO address for
4057 * @offset: Resource address to map
4058 * @size: Size of map
4059 *
4060 * Managed pci_remap_cfgspace().  Map is automatically unmapped on driver
4061 * detach.
4062 */
4063void __iomem *devm_pci_remap_cfgspace(struct device *dev,
4064                                      resource_size_t offset,
4065                                      resource_size_t size)
4066{
4067        void __iomem **ptr, *addr;
4068
4069        ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
4070        if (!ptr)
4071                return NULL;
4072
4073        addr = pci_remap_cfgspace(offset, size);
4074        if (addr) {
4075                *ptr = addr;
4076                devres_add(dev, ptr);
4077        } else
4078                devres_free(ptr);
4079
4080        return addr;
4081}
4082EXPORT_SYMBOL(devm_pci_remap_cfgspace);
4083
4084/**
4085 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
4086 * @dev: generic device to handle the resource for
4087 * @res: configuration space resource to be handled
4088 *
4089 * Checks that a resource is a valid memory region, requests the memory
4090 * region and ioremaps with pci_remap_cfgspace() API that ensures the
4091 * proper PCI configuration space memory attributes are guaranteed.
4092 *
4093 * All operations are managed and will be undone on driver detach.
4094 *
4095 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
4096 * on failure. Usage example::
4097 *
4098 *      res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
4099 *      base = devm_pci_remap_cfg_resource(&pdev->dev, res);
4100 *      if (IS_ERR(base))
4101 *              return PTR_ERR(base);
4102 */
4103void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
4104                                          struct resource *res)
4105{
4106        resource_size_t size;
4107        const char *name;
4108        void __iomem *dest_ptr;
4109
4110        BUG_ON(!dev);
4111
4112        if (!res || resource_type(res) != IORESOURCE_MEM) {
4113                dev_err(dev, "invalid resource\n");
4114                return IOMEM_ERR_PTR(-EINVAL);
4115        }
4116
4117        size = resource_size(res);
4118        name = res->name ?: dev_name(dev);
4119
4120        if (!devm_request_mem_region(dev, res->start, size, name)) {
4121                dev_err(dev, "can't request region for resource %pR\n", res);
4122                return IOMEM_ERR_PTR(-EBUSY);
4123        }
4124
4125        dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4126        if (!dest_ptr) {
4127                dev_err(dev, "ioremap failed for resource %pR\n", res);
4128                devm_release_mem_region(dev, res->start, size);
4129                dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4130        }
4131
4132        return dest_ptr;
4133}
4134EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4135
4136static void __pci_set_master(struct pci_dev *dev, bool enable)
4137{
4138        u16 old_cmd, cmd;
4139
4140        pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4141        if (enable)
4142                cmd = old_cmd | PCI_COMMAND_MASTER;
4143        else
4144                cmd = old_cmd & ~PCI_COMMAND_MASTER;
4145        if (cmd != old_cmd) {
4146                pci_dbg(dev, "%s bus mastering\n",
4147                        enable ? "enabling" : "disabling");
4148                pci_write_config_word(dev, PCI_COMMAND, cmd);
4149        }
4150        dev->is_busmaster = enable;
4151}
4152
4153/**
4154 * pcibios_setup - process "pci=" kernel boot arguments
4155 * @str: string used to pass in "pci=" kernel boot arguments
4156 *
4157 * Process kernel boot arguments.  This is the default implementation.
4158 * Architecture specific implementations can override this as necessary.
4159 */
4160char * __weak __init pcibios_setup(char *str)
4161{
4162        return str;
4163}
4164
4165/**
4166 * pcibios_set_master - enable PCI bus-mastering for device dev
4167 * @dev: the PCI device to enable
4168 *
4169 * Enables PCI bus-mastering for the device.  This is the default
4170 * implementation.  Architecture specific implementations can override
4171 * this if necessary.
4172 */
4173void __weak pcibios_set_master(struct pci_dev *dev)
4174{
4175        u8 lat;
4176
4177        /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4178        if (pci_is_pcie(dev))
4179                return;
4180
4181        pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4182        if (lat < 16)
4183                lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4184        else if (lat > pcibios_max_latency)
4185                lat = pcibios_max_latency;
4186        else
4187                return;
4188
4189        pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4190}
4191
4192/**
4193 * pci_set_master - enables bus-mastering for device dev
4194 * @dev: the PCI device to enable
4195 *
4196 * Enables bus-mastering on the device and calls pcibios_set_master()
4197 * to do the needed arch specific settings.
4198 */
4199void pci_set_master(struct pci_dev *dev)
4200{
4201        __pci_set_master(dev, true);
4202        pcibios_set_master(dev);
4203}
4204EXPORT_SYMBOL(pci_set_master);
4205
4206/**
4207 * pci_clear_master - disables bus-mastering for device dev
4208 * @dev: the PCI device to disable
4209 */
4210void pci_clear_master(struct pci_dev *dev)
4211{
4212        __pci_set_master(dev, false);
4213}
4214EXPORT_SYMBOL(pci_clear_master);
4215
4216/**
4217 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4218 * @dev: the PCI device for which MWI is to be enabled
4219 *
4220 * Helper function for pci_set_mwi.
4221 * Originally copied from drivers/net/acenic.c.
4222 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4223 *
4224 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4225 */
4226int pci_set_cacheline_size(struct pci_dev *dev)
4227{
4228        u8 cacheline_size;
4229
4230        if (!pci_cache_line_size)
4231                return -EINVAL;
4232
4233        /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4234           equal to or multiple of the right value. */
4235        pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4236        if (cacheline_size >= pci_cache_line_size &&
4237            (cacheline_size % pci_cache_line_size) == 0)
4238                return 0;
4239
4240        /* Write the correct value. */
4241        pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4242        /* Read it back. */
4243        pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4244        if (cacheline_size == pci_cache_line_size)
4245                return 0;
4246
4247        pci_info(dev, "cache line size of %d is not supported\n",
4248                   pci_cache_line_size << 2);
4249
4250        return -EINVAL;
4251}
4252EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4253
4254/**
4255 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4256 * @dev: the PCI device for which MWI is enabled
4257 *
4258 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4259 *
4260 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4261 */
4262int pci_set_mwi(struct pci_dev *dev)
4263{
4264#ifdef PCI_DISABLE_MWI
4265        return 0;
4266#else
4267        int rc;
4268        u16 cmd;
4269
4270        rc = pci_set_cacheline_size(dev);
4271        if (rc)
4272                return rc;
4273
4274        pci_read_config_word(dev, PCI_COMMAND, &cmd);
4275        if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4276                pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4277                cmd |= PCI_COMMAND_INVALIDATE;
4278                pci_write_config_word(dev, PCI_COMMAND, cmd);
4279        }
4280        return 0;
4281#endif
4282}
4283EXPORT_SYMBOL(pci_set_mwi);
4284
4285/**
4286 * pcim_set_mwi - a device-managed pci_set_mwi()
4287 * @dev: the PCI device for which MWI is enabled
4288 *
4289 * Managed pci_set_mwi().
4290 *
4291 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4292 */
4293int pcim_set_mwi(struct pci_dev *dev)
4294{
4295        struct pci_devres *dr;
4296
4297        dr = find_pci_dr(dev);
4298        if (!dr)
4299                return -ENOMEM;
4300
4301        dr->mwi = 1;
4302        return pci_set_mwi(dev);
4303}
4304EXPORT_SYMBOL(pcim_set_mwi);
4305
4306/**
4307 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4308 * @dev: the PCI device for which MWI is enabled
4309 *
4310 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4311 * Callers are not required to check the return value.
4312 *
4313 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4314 */
4315int pci_try_set_mwi(struct pci_dev *dev)
4316{
4317#ifdef PCI_DISABLE_MWI
4318        return 0;
4319#else
4320        return pci_set_mwi(dev);
4321#endif
4322}
4323EXPORT_SYMBOL(pci_try_set_mwi);
4324
4325/**
4326 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4327 * @dev: the PCI device to disable
4328 *
4329 * Disables PCI Memory-Write-Invalidate transaction on the device
4330 */
4331void pci_clear_mwi(struct pci_dev *dev)
4332{
4333#ifndef PCI_DISABLE_MWI
4334        u16 cmd;
4335
4336        pci_read_config_word(dev, PCI_COMMAND, &cmd);
4337        if (cmd & PCI_COMMAND_INVALIDATE) {
4338                cmd &= ~PCI_COMMAND_INVALIDATE;
4339                pci_write_config_word(dev, PCI_COMMAND, cmd);
4340        }
4341#endif
4342}
4343EXPORT_SYMBOL(pci_clear_mwi);
4344
4345/**
4346 * pci_intx - enables/disables PCI INTx for device dev
4347 * @pdev: the PCI device to operate on
4348 * @enable: boolean: whether to enable or disable PCI INTx
4349 *
4350 * Enables/disables PCI INTx for device @pdev
4351 */
4352void pci_intx(struct pci_dev *pdev, int enable)
4353{
4354        u16 pci_command, new;
4355
4356        pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4357
4358        if (enable)
4359                new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4360        else
4361                new = pci_command | PCI_COMMAND_INTX_DISABLE;
4362
4363        if (new != pci_command) {
4364                struct pci_devres *dr;
4365
4366                pci_write_config_word(pdev, PCI_COMMAND, new);
4367
4368                dr = find_pci_dr(pdev);
4369                if (dr && !dr->restore_intx) {
4370                        dr->restore_intx = 1;
4371                        dr->orig_intx = !enable;
4372                }
4373        }
4374}
4375EXPORT_SYMBOL_GPL(pci_intx);
4376
4377static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4378{
4379        struct pci_bus *bus = dev->bus;
4380        bool mask_updated = true;
4381        u32 cmd_status_dword;
4382        u16 origcmd, newcmd;
4383        unsigned long flags;
4384        bool irq_pending;
4385
4386        /*
4387         * We do a single dword read to retrieve both command and status.
4388         * Document assumptions that make this possible.
4389         */
4390        BUILD_BUG_ON(PCI_COMMAND % 4);
4391        BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4392
4393        raw_spin_lock_irqsave(&pci_lock, flags);
4394
4395        bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4396
4397        irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4398
4399        /*
4400         * Check interrupt status register to see whether our device
4401         * triggered the interrupt (when masking) or the next IRQ is
4402         * already pending (when unmasking).
4403         */
4404        if (mask != irq_pending) {
4405                mask_updated = false;
4406                goto done;
4407        }
4408
4409        origcmd = cmd_status_dword;
4410        newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4411        if (mask)
4412                newcmd |= PCI_COMMAND_INTX_DISABLE;
4413        if (newcmd != origcmd)
4414                bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4415
4416done:
4417        raw_spin_unlock_irqrestore(&pci_lock, flags);
4418
4419        return mask_updated;
4420}
4421
4422/**
4423 * pci_check_and_mask_intx - mask INTx on pending interrupt
4424 * @dev: the PCI device to operate on
4425 *
4426 * Check if the device dev has its INTx line asserted, mask it and return
4427 * true in that case. False is returned if no interrupt was pending.
4428 */
4429bool pci_check_and_mask_intx(struct pci_dev *dev)
4430{
4431        return pci_check_and_set_intx_mask(dev, true);
4432}
4433EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4434
4435/**
4436 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4437 * @dev: the PCI device to operate on
4438 *
4439 * Check if the device dev has its INTx line asserted, unmask it if not and
4440 * return true. False is returned and the mask remains active if there was
4441 * still an interrupt pending.
4442 */
4443bool pci_check_and_unmask_intx(struct pci_dev *dev)
4444{
4445        return pci_check_and_set_intx_mask(dev, false);
4446}
4447EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4448
4449/**
4450 * pci_wait_for_pending_transaction - wait for pending transaction
4451 * @dev: the PCI device to operate on
4452 *
4453 * Return 0 if transaction is pending 1 otherwise.
4454 */
4455int pci_wait_for_pending_transaction(struct pci_dev *dev)
4456{
4457        if (!pci_is_pcie(dev))
4458                return 1;
4459
4460        return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4461                                    PCI_EXP_DEVSTA_TRPND);
4462}
4463EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4464
4465/**
4466 * pcie_has_flr - check if a device supports function level resets
4467 * @dev: device to check
4468 *
4469 * Returns true if the device advertises support for PCIe function level
4470 * resets.
4471 */
4472bool pcie_has_flr(struct pci_dev *dev)
4473{
4474        u32 cap;
4475
4476        if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4477                return false;
4478
4479        pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4480        return cap & PCI_EXP_DEVCAP_FLR;
4481}
4482EXPORT_SYMBOL_GPL(pcie_has_flr);
4483
4484/**
4485 * pcie_flr - initiate a PCIe function level reset
4486 * @dev: device to reset
4487 *
4488 * Initiate a function level reset on @dev.  The caller should ensure the
4489 * device supports FLR before calling this function, e.g. by using the
4490 * pcie_has_flr() helper.
4491 */
4492int pcie_flr(struct pci_dev *dev)
4493{
4494        if (!pci_wait_for_pending_transaction(dev))
4495                pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4496
4497        pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4498
4499        if (dev->imm_ready)
4500                return 0;
4501
4502        /*
4503         * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4504         * 100ms, but may silently discard requests while the FLR is in
4505         * progress.  Wait 100ms before trying to access the device.
4506         */
4507        msleep(100);
4508
4509        return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4510}
4511EXPORT_SYMBOL_GPL(pcie_flr);
4512
4513static int pci_af_flr(struct pci_dev *dev, int probe)
4514{
4515        int pos;
4516        u8 cap;
4517
4518        pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4519        if (!pos)
4520                return -ENOTTY;
4521
4522        if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4523                return -ENOTTY;
4524
4525        pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4526        if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4527                return -ENOTTY;
4528
4529        if (probe)
4530                return 0;
4531
4532        /*
4533         * Wait for Transaction Pending bit to clear.  A word-aligned test
4534         * is used, so we use the control offset rather than status and shift
4535         * the test bit to match.
4536         */
4537        if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4538                                 PCI_AF_STATUS_TP << 8))
4539                pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4540
4541        pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4542
4543        if (dev->imm_ready)
4544                return 0;
4545
4546        /*
4547         * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4548         * updated 27 July 2006; a device must complete an FLR within
4549         * 100ms, but may silently discard requests while the FLR is in
4550         * progress.  Wait 100ms before trying to access the device.
4551         */
4552        msleep(100);
4553
4554        return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4555}
4556
4557/**
4558 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4559 * @dev: Device to reset.
4560 * @probe: If set, only check if the device can be reset this way.
4561 *
4562 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4563 * unset, it will be reinitialized internally when going from PCI_D3hot to
4564 * PCI_D0.  If that's the case and the device is not in a low-power state
4565 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4566 *
4567 * NOTE: This causes the caller to sleep for twice the device power transition
4568 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4569 * by default (i.e. unless the @dev's d3_delay field has a different value).
4570 * Moreover, only devices in D0 can be reset by this function.
4571 */
4572static int pci_pm_reset(struct pci_dev *dev, int probe)
4573{
4574        u16 csr;
4575
4576        if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4577                return -ENOTTY;
4578
4579        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4580        if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4581                return -ENOTTY;
4582
4583        if (probe)
4584                return 0;
4585
4586        if (dev->current_state != PCI_D0)
4587                return -EINVAL;
4588
4589        csr &= ~PCI_PM_CTRL_STATE_MASK;
4590        csr |= PCI_D3hot;
4591        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4592        pci_dev_d3_sleep(dev);
4593
4594        csr &= ~PCI_PM_CTRL_STATE_MASK;
4595        csr |= PCI_D0;
4596        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4597        pci_dev_d3_sleep(dev);
4598
4599        return pci_dev_wait(dev, "PM D3hot->D0", PCIE_RESET_READY_POLL_MS);
4600}
4601
4602/**
4603 * pcie_wait_for_link_delay - Wait until link is active or inactive
4604 * @pdev: Bridge device
4605 * @active: waiting for active or inactive?
4606 * @delay: Delay to wait after link has become active (in ms)
4607 *
4608 * Use this to wait till link becomes active or inactive.
4609 */
4610static bool pcie_wait_for_link_delay(struct pci_dev *pdev, bool active,
4611                                     int delay)
4612{
4613        int timeout = 1000;
4614        bool ret;
4615        u16 lnk_status;
4616
4617        /*
4618         * Some controllers might not implement link active reporting. In this
4619         * case, we wait for 1000 + 100 ms.
4620         */
4621        if (!pdev->link_active_reporting) {
4622                msleep(1100);
4623                return true;
4624        }
4625
4626        /*
4627         * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4628         * after which we should expect an link active if the reset was
4629         * successful. If so, software must wait a minimum 100ms before sending
4630         * configuration requests to devices downstream this port.
4631         *
4632         * If the link fails to activate, either the device was physically
4633         * removed or the link is permanently failed.
4634         */
4635        if (active)
4636                msleep(20);
4637        for (;;) {
4638                pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4639                ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4640                if (ret == active)
4641                        break;
4642                if (timeout <= 0)
4643                        break;
4644                msleep(10);
4645                timeout -= 10;
4646        }
4647        if (active && ret)
4648                msleep(delay);
4649        else if (ret != active)
4650                pci_info(pdev, "Data Link Layer Link Active not %s in 1000 msec\n",
4651                        active ? "set" : "cleared");
4652        return ret == active;
4653}
4654
4655/**
4656 * pcie_wait_for_link - Wait until link is active or inactive
4657 * @pdev: Bridge device
4658 * @active: waiting for active or inactive?
4659 *
4660 * Use this to wait till link becomes active or inactive.
4661 */
4662bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4663{
4664        return pcie_wait_for_link_delay(pdev, active, 100);
4665}
4666
4667/*
4668 * Find maximum D3cold delay required by all the devices on the bus.  The
4669 * spec says 100 ms, but firmware can lower it and we allow drivers to
4670 * increase it as well.
4671 *
4672 * Called with @pci_bus_sem locked for reading.
4673 */
4674static int pci_bus_max_d3cold_delay(const struct pci_bus *bus)
4675{
4676        const struct pci_dev *pdev;
4677        int min_delay = 100;
4678        int max_delay = 0;
4679
4680        list_for_each_entry(pdev, &bus->devices, bus_list) {
4681                if (pdev->d3cold_delay < min_delay)
4682                        min_delay = pdev->d3cold_delay;
4683                if (pdev->d3cold_delay > max_delay)
4684                        max_delay = pdev->d3cold_delay;
4685        }
4686
4687        return max(min_delay, max_delay);
4688}
4689
4690/**
4691 * pci_bridge_wait_for_secondary_bus - Wait for secondary bus to be accessible
4692 * @dev: PCI bridge
4693 *
4694 * Handle necessary delays before access to the devices on the secondary
4695 * side of the bridge are permitted after D3cold to D0 transition.
4696 *
4697 * For PCIe this means the delays in PCIe 5.0 section 6.6.1. For
4698 * conventional PCI it means Tpvrh + Trhfa specified in PCI 3.0 section
4699 * 4.3.2.
4700 */
4701void pci_bridge_wait_for_secondary_bus(struct pci_dev *dev)
4702{
4703        struct pci_dev *child;
4704        int delay;
4705
4706        if (pci_dev_is_disconnected(dev))
4707                return;
4708
4709        if (!pci_is_bridge(dev) || !dev->bridge_d3)
4710                return;
4711
4712        down_read(&pci_bus_sem);
4713
4714        /*
4715         * We only deal with devices that are present currently on the bus.
4716         * For any hot-added devices the access delay is handled in pciehp
4717         * board_added(). In case of ACPI hotplug the firmware is expected
4718         * to configure the devices before OS is notified.
4719         */
4720        if (!dev->subordinate || list_empty(&dev->subordinate->devices)) {
4721                up_read(&pci_bus_sem);
4722                return;
4723        }
4724
4725        /* Take d3cold_delay requirements into account */
4726        delay = pci_bus_max_d3cold_delay(dev->subordinate);
4727        if (!delay) {
4728                up_read(&pci_bus_sem);
4729                return;
4730        }
4731
4732        child = list_first_entry(&dev->subordinate->devices, struct pci_dev,
4733                                 bus_list);
4734        up_read(&pci_bus_sem);
4735
4736        /*
4737         * Conventional PCI and PCI-X we need to wait Tpvrh + Trhfa before
4738         * accessing the device after reset (that is 1000 ms + 100 ms). In
4739         * practice this should not be needed because we don't do power
4740         * management for them (see pci_bridge_d3_possible()).
4741         */
4742        if (!pci_is_pcie(dev)) {
4743                pci_dbg(dev, "waiting %d ms for secondary bus\n", 1000 + delay);
4744                msleep(1000 + delay);
4745                return;
4746        }
4747
4748        /*
4749         * For PCIe downstream and root ports that do not support speeds
4750         * greater than 5 GT/s need to wait minimum 100 ms. For higher
4751         * speeds (gen3) we need to wait first for the data link layer to
4752         * become active.
4753         *
4754         * However, 100 ms is the minimum and the PCIe spec says the
4755         * software must allow at least 1s before it can determine that the
4756         * device that did not respond is a broken device. There is
4757         * evidence that 100 ms is not always enough, for example certain
4758         * Titan Ridge xHCI controller does not always respond to
4759         * configuration requests if we only wait for 100 ms (see
4760         * https://bugzilla.kernel.org/show_bug.cgi?id=203885).
4761         *
4762         * Therefore we wait for 100 ms and check for the device presence.
4763         * If it is still not present give it an additional 100 ms.
4764         */
4765        if (!pcie_downstream_port(dev))
4766                return;
4767
4768        if (pcie_get_speed_cap(dev) <= PCIE_SPEED_5_0GT) {
4769                pci_dbg(dev, "waiting %d ms for downstream link\n", delay);
4770                msleep(delay);
4771        } else {
4772                pci_dbg(dev, "waiting %d ms for downstream link, after activation\n",
4773                        delay);
4774                if (!pcie_wait_for_link_delay(dev, true, delay)) {
4775                        /* Did not train, no need to wait any further */
4776                        return;
4777                }
4778        }
4779
4780        if (!pci_device_is_present(child)) {
4781                pci_dbg(child, "waiting additional %d ms to become accessible\n", delay);
4782                msleep(delay);
4783        }
4784}
4785
4786void pci_reset_secondary_bus(struct pci_dev *dev)
4787{
4788        u16 ctrl;
4789
4790        pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4791        ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4792        pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4793
4794        /*
4795         * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms.  Double
4796         * this to 2ms to ensure that we meet the minimum requirement.
4797         */
4798        msleep(2);
4799
4800        ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4801        pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4802
4803        /*
4804         * Trhfa for conventional PCI is 2^25 clock cycles.
4805         * Assuming a minimum 33MHz clock this results in a 1s
4806         * delay before we can consider subordinate devices to
4807         * be re-initialized.  PCIe has some ways to shorten this,
4808         * but we don't make use of them yet.
4809         */
4810        ssleep(1);
4811}
4812
4813void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4814{
4815        pci_reset_secondary_bus(dev);
4816}
4817
4818/**
4819 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4820 * @dev: Bridge device
4821 *
4822 * Use the bridge control register to assert reset on the secondary bus.
4823 * Devices on the secondary bus are left in power-on state.
4824 */
4825int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4826{
4827        pcibios_reset_secondary_bus(dev);
4828
4829        return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4830}
4831EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4832
4833static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4834{
4835        struct pci_dev *pdev;
4836
4837        if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4838            !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4839                return -ENOTTY;
4840
4841        list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4842                if (pdev != dev)
4843                        return -ENOTTY;
4844
4845        if (probe)
4846                return 0;
4847
4848        return pci_bridge_secondary_bus_reset(dev->bus->self);
4849}
4850
4851static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4852{
4853        int rc = -ENOTTY;
4854
4855        if (!hotplug || !try_module_get(hotplug->owner))
4856                return rc;
4857
4858        if (hotplug->ops->reset_slot)
4859                rc = hotplug->ops->reset_slot(hotplug, probe);
4860
4861        module_put(hotplug->owner);
4862
4863        return rc;
4864}
4865
4866static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4867{
4868        struct pci_dev *pdev;
4869
4870        if (dev->subordinate || !dev->slot ||
4871            dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4872                return -ENOTTY;
4873
4874        list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4875                if (pdev != dev && pdev->slot == dev->slot)
4876                        return -ENOTTY;
4877
4878        return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4879}
4880
4881static void pci_dev_lock(struct pci_dev *dev)
4882{
4883        pci_cfg_access_lock(dev);
4884        /* block PM suspend, driver probe, etc. */
4885        device_lock(&dev->dev);
4886}
4887
4888/* Return 1 on successful lock, 0 on contention */
4889static int pci_dev_trylock(struct pci_dev *dev)
4890{
4891        if (pci_cfg_access_trylock(dev)) {
4892                if (device_trylock(&dev->dev))
4893                        return 1;
4894                pci_cfg_access_unlock(dev);
4895        }
4896
4897        return 0;
4898}
4899
4900static void pci_dev_unlock(struct pci_dev *dev)
4901{
4902        device_unlock(&dev->dev);
4903        pci_cfg_access_unlock(dev);
4904}
4905
4906static void pci_dev_save_and_disable(struct pci_dev *dev)
4907{
4908        const struct pci_error_handlers *err_handler =
4909                        dev->driver ? dev->driver->err_handler : NULL;
4910
4911        /*
4912         * dev->driver->err_handler->reset_prepare() is protected against
4913         * races with ->remove() by the device lock, which must be held by
4914         * the caller.
4915         */
4916        if (err_handler && err_handler->reset_prepare)
4917                err_handler->reset_prepare(dev);
4918
4919        /*
4920         * Wake-up device prior to save.  PM registers default to D0 after
4921         * reset and a simple register restore doesn't reliably return
4922         * to a non-D0 state anyway.
4923         */
4924        pci_set_power_state(dev, PCI_D0);
4925
4926        pci_save_state(dev);
4927        /*
4928         * Disable the device by clearing the Command register, except for
4929         * INTx-disable which is set.  This not only disables MMIO and I/O port
4930         * BARs, but also prevents the device from being Bus Master, preventing
4931         * DMA from the device including MSI/MSI-X interrupts.  For PCI 2.3
4932         * compliant devices, INTx-disable prevents legacy interrupts.
4933         */
4934        pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4935}
4936
4937static void pci_dev_restore(struct pci_dev *dev)
4938{
4939        const struct pci_error_handlers *err_handler =
4940                        dev->driver ? dev->driver->err_handler : NULL;
4941
4942        pci_restore_state(dev);
4943
4944        /*
4945         * dev->driver->err_handler->reset_done() is protected against
4946         * races with ->remove() by the device lock, which must be held by
4947         * the caller.
4948         */
4949        if (err_handler && err_handler->reset_done)
4950                err_handler->reset_done(dev);
4951}
4952
4953/**
4954 * __pci_reset_function_locked - reset a PCI device function while holding
4955 * the @dev mutex lock.
4956 * @dev: PCI device to reset
4957 *
4958 * Some devices allow an individual function to be reset without affecting
4959 * other functions in the same device.  The PCI device must be responsive
4960 * to PCI config space in order to use this function.
4961 *
4962 * The device function is presumed to be unused and the caller is holding
4963 * the device mutex lock when this function is called.
4964 *
4965 * Resetting the device will make the contents of PCI configuration space
4966 * random, so any caller of this must be prepared to reinitialise the
4967 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4968 * etc.
4969 *
4970 * Returns 0 if the device function was successfully reset or negative if the
4971 * device doesn't support resetting a single function.
4972 */
4973int __pci_reset_function_locked(struct pci_dev *dev)
4974{
4975        int rc;
4976
4977        might_sleep();
4978
4979        /*
4980         * A reset method returns -ENOTTY if it doesn't support this device
4981         * and we should try the next method.
4982         *
4983         * If it returns 0 (success), we're finished.  If it returns any
4984         * other error, we're also finished: this indicates that further
4985         * reset mechanisms might be broken on the device.
4986         */
4987        rc = pci_dev_specific_reset(dev, 0);
4988        if (rc != -ENOTTY)
4989                return rc;
4990        if (pcie_has_flr(dev)) {
4991                rc = pcie_flr(dev);
4992                if (rc != -ENOTTY)
4993                        return rc;
4994        }
4995        rc = pci_af_flr(dev, 0);
4996        if (rc != -ENOTTY)
4997                return rc;
4998        rc = pci_pm_reset(dev, 0);
4999        if (rc != -ENOTTY)
5000                return rc;
5001        rc = pci_dev_reset_slot_function(dev, 0);
5002        if (rc != -ENOTTY)
5003                return rc;
5004        return pci_parent_bus_reset(dev, 0);
5005}
5006EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
5007
5008/**
5009 * pci_probe_reset_function - check whether the device can be safely reset
5010 * @dev: PCI device to reset
5011 *
5012 * Some devices allow an individual function to be reset without affecting
5013 * other functions in the same device.  The PCI device must be responsive
5014 * to PCI config space in order to use this function.
5015 *
5016 * Returns 0 if the device function can be reset or negative if the
5017 * device doesn't support resetting a single function.
5018 */
5019int pci_probe_reset_function(struct pci_dev *dev)
5020{
5021        int rc;
5022
5023        might_sleep();
5024
5025        rc = pci_dev_specific_reset(dev, 1);
5026        if (rc != -ENOTTY)
5027                return rc;
5028        if (pcie_has_flr(dev))
5029                return 0;
5030        rc = pci_af_flr(dev, 1);
5031        if (rc != -ENOTTY)
5032                return rc;
5033        rc = pci_pm_reset(dev, 1);
5034        if (rc != -ENOTTY)
5035                return rc;
5036        rc = pci_dev_reset_slot_function(dev, 1);
5037        if (rc != -ENOTTY)
5038                return rc;
5039
5040        return pci_parent_bus_reset(dev, 1);
5041}
5042
5043/**
5044 * pci_reset_function - quiesce and reset a PCI device function
5045 * @dev: PCI device to reset
5046 *
5047 * Some devices allow an individual function to be reset without affecting
5048 * other functions in the same device.  The PCI device must be responsive
5049 * to PCI config space in order to use this function.
5050 *
5051 * This function does not just reset the PCI portion of a device, but
5052 * clears all the state associated with the device.  This function differs
5053 * from __pci_reset_function_locked() in that it saves and restores device state
5054 * over the reset and takes the PCI device lock.
5055 *
5056 * Returns 0 if the device function was successfully reset or negative if the
5057 * device doesn't support resetting a single function.
5058 */
5059int pci_reset_function(struct pci_dev *dev)
5060{
5061        int rc;
5062
5063        if (!dev->reset_fn)
5064                return -ENOTTY;
5065
5066        pci_dev_lock(dev);
5067        pci_dev_save_and_disable(dev);
5068
5069        rc = __pci_reset_function_locked(dev);
5070
5071        pci_dev_restore(dev);
5072        pci_dev_unlock(dev);
5073
5074        return rc;
5075}
5076EXPORT_SYMBOL_GPL(pci_reset_function);
5077
5078/**
5079 * pci_reset_function_locked - quiesce and reset a PCI device function
5080 * @dev: PCI device to reset
5081 *
5082 * Some devices allow an individual function to be reset without affecting
5083 * other functions in the same device.  The PCI device must be responsive
5084 * to PCI config space in order to use this function.
5085 *
5086 * This function does not just reset the PCI portion of a device, but
5087 * clears all the state associated with the device.  This function differs
5088 * from __pci_reset_function_locked() in that it saves and restores device state
5089 * over the reset.  It also differs from pci_reset_function() in that it
5090 * requires the PCI device lock to be held.
5091 *
5092 * Returns 0 if the device function was successfully reset or negative if the
5093 * device doesn't support resetting a single function.
5094 */
5095int pci_reset_function_locked(struct pci_dev *dev)
5096{
5097        int rc;
5098
5099        if (!dev->reset_fn)
5100                return -ENOTTY;
5101
5102        pci_dev_save_and_disable(dev);
5103
5104        rc = __pci_reset_function_locked(dev);
5105
5106        pci_dev_restore(dev);
5107
5108        return rc;
5109}
5110EXPORT_SYMBOL_GPL(pci_reset_function_locked);
5111
5112/**
5113 * pci_try_reset_function - quiesce and reset a PCI device function
5114 * @dev: PCI device to reset
5115 *
5116 * Same as above, except return -EAGAIN if unable to lock device.
5117 */
5118int pci_try_reset_function(struct pci_dev *dev)
5119{
5120        int rc;
5121
5122        if (!dev->reset_fn)
5123                return -ENOTTY;
5124
5125        if (!pci_dev_trylock(dev))
5126                return -EAGAIN;
5127
5128        pci_dev_save_and_disable(dev);
5129        rc = __pci_reset_function_locked(dev);
5130        pci_dev_restore(dev);
5131        pci_dev_unlock(dev);
5132
5133        return rc;
5134}
5135EXPORT_SYMBOL_GPL(pci_try_reset_function);
5136
5137/* Do any devices on or below this bus prevent a bus reset? */
5138static bool pci_bus_resetable(struct pci_bus *bus)
5139{
5140        struct pci_dev *dev;
5141
5142
5143        if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5144                return false;
5145
5146        list_for_each_entry(dev, &bus->devices, bus_list) {
5147                if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5148                    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5149                        return false;
5150        }
5151
5152        return true;
5153}
5154
5155/* Lock devices from the top of the tree down */
5156static void pci_bus_lock(struct pci_bus *bus)
5157{
5158        struct pci_dev *dev;
5159
5160        list_for_each_entry(dev, &bus->devices, bus_list) {
5161                pci_dev_lock(dev);
5162                if (dev->subordinate)
5163                        pci_bus_lock(dev->subordinate);
5164        }
5165}
5166
5167/* Unlock devices from the bottom of the tree up */
5168static void pci_bus_unlock(struct pci_bus *bus)
5169{
5170        struct pci_dev *dev;
5171
5172        list_for_each_entry(dev, &bus->devices, bus_list) {
5173                if (dev->subordinate)
5174                        pci_bus_unlock(dev->subordinate);
5175                pci_dev_unlock(dev);
5176        }
5177}
5178
5179/* Return 1 on successful lock, 0 on contention */
5180static int pci_bus_trylock(struct pci_bus *bus)
5181{
5182        struct pci_dev *dev;
5183
5184        list_for_each_entry(dev, &bus->devices, bus_list) {
5185                if (!pci_dev_trylock(dev))
5186                        goto unlock;
5187                if (dev->subordinate) {
5188                        if (!pci_bus_trylock(dev->subordinate)) {
5189                                pci_dev_unlock(dev);
5190                                goto unlock;
5191                        }
5192                }
5193        }
5194        return 1;
5195
5196unlock:
5197        list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5198                if (dev->subordinate)
5199                        pci_bus_unlock(dev->subordinate);
5200                pci_dev_unlock(dev);
5201        }
5202        return 0;
5203}
5204
5205/* Do any devices on or below this slot prevent a bus reset? */
5206static bool pci_slot_resetable(struct pci_slot *slot)
5207{
5208        struct pci_dev *dev;
5209
5210        if (slot->bus->self &&
5211            (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5212                return false;
5213
5214        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5215                if (!dev->slot || dev->slot != slot)
5216                        continue;
5217                if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5218                    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5219                        return false;
5220        }
5221
5222        return true;
5223}
5224
5225/* Lock devices from the top of the tree down */
5226static void pci_slot_lock(struct pci_slot *slot)
5227{
5228        struct pci_dev *dev;
5229
5230        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5231                if (!dev->slot || dev->slot != slot)
5232                        continue;
5233                pci_dev_lock(dev);
5234                if (dev->subordinate)
5235                        pci_bus_lock(dev->subordinate);
5236        }
5237}
5238
5239/* Unlock devices from the bottom of the tree up */
5240static void pci_slot_unlock(struct pci_slot *slot)
5241{
5242        struct pci_dev *dev;
5243
5244        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5245                if (!dev->slot || dev->slot != slot)
5246                        continue;
5247                if (dev->subordinate)
5248                        pci_bus_unlock(dev->subordinate);
5249                pci_dev_unlock(dev);
5250        }
5251}
5252
5253/* Return 1 on successful lock, 0 on contention */
5254static int pci_slot_trylock(struct pci_slot *slot)
5255{
5256        struct pci_dev *dev;
5257
5258        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5259                if (!dev->slot || dev->slot != slot)
5260                        continue;
5261                if (!pci_dev_trylock(dev))
5262                        goto unlock;
5263                if (dev->subordinate) {
5264                        if (!pci_bus_trylock(dev->subordinate)) {
5265                                pci_dev_unlock(dev);
5266                                goto unlock;
5267                        }
5268                }
5269        }
5270        return 1;
5271
5272unlock:
5273        list_for_each_entry_continue_reverse(dev,
5274                                             &slot->bus->devices, bus_list) {
5275                if (!dev->slot || dev->slot != slot)
5276                        continue;
5277                if (dev->subordinate)
5278                        pci_bus_unlock(dev->subordinate);
5279                pci_dev_unlock(dev);
5280        }
5281        return 0;
5282}
5283
5284/*
5285 * Save and disable devices from the top of the tree down while holding
5286 * the @dev mutex lock for the entire tree.
5287 */
5288static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5289{
5290        struct pci_dev *dev;
5291
5292        list_for_each_entry(dev, &bus->devices, bus_list) {
5293                pci_dev_save_and_disable(dev);
5294                if (dev->subordinate)
5295                        pci_bus_save_and_disable_locked(dev->subordinate);
5296        }
5297}
5298
5299/*
5300 * Restore devices from top of the tree down while holding @dev mutex lock
5301 * for the entire tree.  Parent bridges need to be restored before we can
5302 * get to subordinate devices.
5303 */
5304static void pci_bus_restore_locked(struct pci_bus *bus)
5305{
5306        struct pci_dev *dev;
5307
5308        list_for_each_entry(dev, &bus->devices, bus_list) {
5309                pci_dev_restore(dev);
5310                if (dev->subordinate)
5311                        pci_bus_restore_locked(dev->subordinate);
5312        }
5313}
5314
5315/*
5316 * Save and disable devices from the top of the tree down while holding
5317 * the @dev mutex lock for the entire tree.
5318 */
5319static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5320{
5321        struct pci_dev *dev;
5322
5323        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5324                if (!dev->slot || dev->slot != slot)
5325                        continue;
5326                pci_dev_save_and_disable(dev);
5327                if (dev->subordinate)
5328                        pci_bus_save_and_disable_locked(dev->subordinate);
5329        }
5330}
5331
5332/*
5333 * Restore devices from top of the tree down while holding @dev mutex lock
5334 * for the entire tree.  Parent bridges need to be restored before we can
5335 * get to subordinate devices.
5336 */
5337static void pci_slot_restore_locked(struct pci_slot *slot)
5338{
5339        struct pci_dev *dev;
5340
5341        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5342                if (!dev->slot || dev->slot != slot)
5343                        continue;
5344                pci_dev_restore(dev);
5345                if (dev->subordinate)
5346                        pci_bus_restore_locked(dev->subordinate);
5347        }
5348}
5349
5350static int pci_slot_reset(struct pci_slot *slot, int probe)
5351{
5352        int rc;
5353
5354        if (!slot || !pci_slot_resetable(slot))
5355                return -ENOTTY;
5356
5357        if (!probe)
5358                pci_slot_lock(slot);
5359
5360        might_sleep();
5361
5362        rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5363
5364        if (!probe)
5365                pci_slot_unlock(slot);
5366
5367        return rc;
5368}
5369
5370/**
5371 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5372 * @slot: PCI slot to probe
5373 *
5374 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5375 */
5376int pci_probe_reset_slot(struct pci_slot *slot)
5377{
5378        return pci_slot_reset(slot, 1);
5379}
5380EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5381
5382/**
5383 * __pci_reset_slot - Try to reset a PCI slot
5384 * @slot: PCI slot to reset
5385 *
5386 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5387 * independent of other slots.  For instance, some slots may support slot power
5388 * control.  In the case of a 1:1 bus to slot architecture, this function may
5389 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5390 * Generally a slot reset should be attempted before a bus reset.  All of the
5391 * function of the slot and any subordinate buses behind the slot are reset
5392 * through this function.  PCI config space of all devices in the slot and
5393 * behind the slot is saved before and restored after reset.
5394 *
5395 * Same as above except return -EAGAIN if the slot cannot be locked
5396 */
5397static int __pci_reset_slot(struct pci_slot *slot)
5398{
5399        int rc;
5400
5401        rc = pci_slot_reset(slot, 1);
5402        if (rc)
5403                return rc;
5404
5405        if (pci_slot_trylock(slot)) {
5406                pci_slot_save_and_disable_locked(slot);
5407                might_sleep();
5408                rc = pci_reset_hotplug_slot(slot->hotplug, 0);
5409                pci_slot_restore_locked(slot);
5410                pci_slot_unlock(slot);
5411        } else
5412                rc = -EAGAIN;
5413
5414        return rc;
5415}
5416
5417static int pci_bus_reset(struct pci_bus *bus, int probe)
5418{
5419        int ret;
5420
5421        if (!bus->self || !pci_bus_resetable(bus))
5422                return -ENOTTY;
5423
5424        if (probe)
5425                return 0;
5426
5427        pci_bus_lock(bus);
5428
5429        might_sleep();
5430
5431        ret = pci_bridge_secondary_bus_reset(bus->self);
5432
5433        pci_bus_unlock(bus);
5434
5435        return ret;
5436}
5437
5438/**
5439 * pci_bus_error_reset - reset the bridge's subordinate bus
5440 * @bridge: The parent device that connects to the bus to reset
5441 *
5442 * This function will first try to reset the slots on this bus if the method is
5443 * available. If slot reset fails or is not available, this will fall back to a
5444 * secondary bus reset.
5445 */
5446int pci_bus_error_reset(struct pci_dev *bridge)
5447{
5448        struct pci_bus *bus = bridge->subordinate;
5449        struct pci_slot *slot;
5450
5451        if (!bus)
5452                return -ENOTTY;
5453
5454        mutex_lock(&pci_slot_mutex);
5455        if (list_empty(&bus->slots))
5456                goto bus_reset;
5457
5458        list_for_each_entry(slot, &bus->slots, list)
5459                if (pci_probe_reset_slot(slot))
5460                        goto bus_reset;
5461
5462        list_for_each_entry(slot, &bus->slots, list)
5463                if (pci_slot_reset(slot, 0))
5464                        goto bus_reset;
5465
5466        mutex_unlock(&pci_slot_mutex);
5467        return 0;
5468bus_reset:
5469        mutex_unlock(&pci_slot_mutex);
5470        return pci_bus_reset(bridge->subordinate, 0);
5471}
5472
5473/**
5474 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5475 * @bus: PCI bus to probe
5476 *
5477 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5478 */
5479int pci_probe_reset_bus(struct pci_bus *bus)
5480{
5481        return pci_bus_reset(bus, 1);
5482}
5483EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5484
5485/**
5486 * __pci_reset_bus - Try to reset a PCI bus
5487 * @bus: top level PCI bus to reset
5488 *
5489 * Same as above except return -EAGAIN if the bus cannot be locked
5490 */
5491static int __pci_reset_bus(struct pci_bus *bus)
5492{
5493        int rc;
5494
5495        rc = pci_bus_reset(bus, 1);
5496        if (rc)
5497                return rc;
5498
5499        if (pci_bus_trylock(bus)) {
5500                pci_bus_save_and_disable_locked(bus);
5501                might_sleep();
5502                rc = pci_bridge_secondary_bus_reset(bus->self);
5503                pci_bus_restore_locked(bus);
5504                pci_bus_unlock(bus);
5505        } else
5506                rc = -EAGAIN;
5507
5508        return rc;
5509}
5510
5511/**
5512 * pci_reset_bus - Try to reset a PCI bus
5513 * @pdev: top level PCI device to reset via slot/bus
5514 *
5515 * Same as above except return -EAGAIN if the bus cannot be locked
5516 */
5517int pci_reset_bus(struct pci_dev *pdev)
5518{
5519        return (!pci_probe_reset_slot(pdev->slot)) ?
5520            __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5521}
5522EXPORT_SYMBOL_GPL(pci_reset_bus);
5523
5524/**
5525 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5526 * @dev: PCI device to query
5527 *
5528 * Returns mmrbc: maximum designed memory read count in bytes or
5529 * appropriate error value.
5530 */
5531int pcix_get_max_mmrbc(struct pci_dev *dev)
5532{
5533        int cap;
5534        u32 stat;
5535
5536        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5537        if (!cap)
5538                return -EINVAL;
5539
5540        if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5541                return -EINVAL;
5542
5543        return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5544}
5545EXPORT_SYMBOL(pcix_get_max_mmrbc);
5546
5547/**
5548 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5549 * @dev: PCI device to query
5550 *
5551 * Returns mmrbc: maximum memory read count in bytes or appropriate error
5552 * value.
5553 */
5554int pcix_get_mmrbc(struct pci_dev *dev)
5555{
5556        int cap;
5557        u16 cmd;
5558
5559        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5560        if (!cap)
5561                return -EINVAL;
5562
5563        if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5564                return -EINVAL;
5565
5566        return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5567}
5568EXPORT_SYMBOL(pcix_get_mmrbc);
5569
5570/**
5571 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5572 * @dev: PCI device to query
5573 * @mmrbc: maximum memory read count in bytes
5574 *    valid values are 512, 1024, 2048, 4096
5575 *
5576 * If possible sets maximum memory read byte count, some bridges have errata
5577 * that prevent this.
5578 */
5579int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5580{
5581        int cap;
5582        u32 stat, v, o;
5583        u16 cmd;
5584
5585        if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5586                return -EINVAL;
5587
5588        v = ffs(mmrbc) - 10;
5589
5590        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5591        if (!cap)
5592                return -EINVAL;
5593
5594        if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5595                return -EINVAL;
5596
5597        if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5598                return -E2BIG;
5599
5600        if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5601                return -EINVAL;
5602
5603        o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5604        if (o != v) {
5605                if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5606                        return -EIO;
5607
5608                cmd &= ~PCI_X_CMD_MAX_READ;
5609                cmd |= v << 2;
5610                if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5611                        return -EIO;
5612        }
5613        return 0;
5614}
5615EXPORT_SYMBOL(pcix_set_mmrbc);
5616
5617/**
5618 * pcie_get_readrq - get PCI Express read request size
5619 * @dev: PCI device to query
5620 *
5621 * Returns maximum memory read request in bytes or appropriate error value.
5622 */
5623int pcie_get_readrq(struct pci_dev *dev)
5624{
5625        u16 ctl;
5626
5627        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5628
5629        return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5630}
5631EXPORT_SYMBOL(pcie_get_readrq);
5632
5633/**
5634 * pcie_set_readrq - set PCI Express maximum memory read request
5635 * @dev: PCI device to query
5636 * @rq: maximum memory read count in bytes
5637 *    valid values are 128, 256, 512, 1024, 2048, 4096
5638 *
5639 * If possible sets maximum memory read request in bytes
5640 */
5641int pcie_set_readrq(struct pci_dev *dev, int rq)
5642{
5643        u16 v;
5644
5645        if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5646                return -EINVAL;
5647
5648        /*
5649         * If using the "performance" PCIe config, we clamp the read rq
5650         * size to the max packet size to keep the host bridge from
5651         * generating requests larger than we can cope with.
5652         */
5653        if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5654                int mps = pcie_get_mps(dev);
5655
5656                if (mps < rq)
5657                        rq = mps;
5658        }
5659
5660        v = (ffs(rq) - 8) << 12;
5661
5662        return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5663                                                  PCI_EXP_DEVCTL_READRQ, v);
5664}
5665EXPORT_SYMBOL(pcie_set_readrq);
5666
5667/**
5668 * pcie_get_mps - get PCI Express maximum payload size
5669 * @dev: PCI device to query
5670 *
5671 * Returns maximum payload size in bytes
5672 */
5673int pcie_get_mps(struct pci_dev *dev)
5674{
5675        u16 ctl;
5676
5677        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5678
5679        return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5680}
5681EXPORT_SYMBOL(pcie_get_mps);
5682
5683/**
5684 * pcie_set_mps - set PCI Express maximum payload size
5685 * @dev: PCI device to query
5686 * @mps: maximum payload size in bytes
5687 *    valid values are 128, 256, 512, 1024, 2048, 4096
5688 *
5689 * If possible sets maximum payload size
5690 */
5691int pcie_set_mps(struct pci_dev *dev, int mps)
5692{
5693        u16 v;
5694
5695        if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5696                return -EINVAL;
5697
5698        v = ffs(mps) - 8;
5699        if (v > dev->pcie_mpss)
5700                return -EINVAL;
5701        v <<= 5;
5702
5703        return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5704                                                  PCI_EXP_DEVCTL_PAYLOAD, v);
5705}
5706EXPORT_SYMBOL(pcie_set_mps);
5707
5708/**
5709 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5710 *                            device and its bandwidth limitation
5711 * @dev: PCI device to query
5712 * @limiting_dev: storage for device causing the bandwidth limitation
5713 * @speed: storage for speed of limiting device
5714 * @width: storage for width of limiting device
5715 *
5716 * Walk up the PCI device chain and find the point where the minimum
5717 * bandwidth is available.  Return the bandwidth available there and (if
5718 * limiting_dev, speed, and width pointers are supplied) information about
5719 * that point.  The bandwidth returned is in Mb/s, i.e., megabits/second of
5720 * raw bandwidth.
5721 */
5722u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5723                             enum pci_bus_speed *speed,
5724                             enum pcie_link_width *width)
5725{
5726        u16 lnksta;
5727        enum pci_bus_speed next_speed;
5728        enum pcie_link_width next_width;
5729        u32 bw, next_bw;
5730
5731        if (speed)
5732                *speed = PCI_SPEED_UNKNOWN;
5733        if (width)
5734                *width = PCIE_LNK_WIDTH_UNKNOWN;
5735
5736        bw = 0;
5737
5738        while (dev) {
5739                pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5740
5741                next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5742                next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5743                        PCI_EXP_LNKSTA_NLW_SHIFT;
5744
5745                next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5746
5747                /* Check if current device limits the total bandwidth */
5748                if (!bw || next_bw <= bw) {
5749                        bw = next_bw;
5750
5751                        if (limiting_dev)
5752                                *limiting_dev = dev;
5753                        if (speed)
5754                                *speed = next_speed;
5755                        if (width)
5756                                *width = next_width;
5757                }
5758
5759                dev = pci_upstream_bridge(dev);
5760        }
5761
5762        return bw;
5763}
5764EXPORT_SYMBOL(pcie_bandwidth_available);
5765
5766/**
5767 * pcie_get_speed_cap - query for the PCI device's link speed capability
5768 * @dev: PCI device to query
5769 *
5770 * Query the PCI device speed capability.  Return the maximum link speed
5771 * supported by the device.
5772 */
5773enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5774{
5775        u32 lnkcap2, lnkcap;
5776
5777        /*
5778         * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18.  The
5779         * implementation note there recommends using the Supported Link
5780         * Speeds Vector in Link Capabilities 2 when supported.
5781         *
5782         * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
5783         * should use the Supported Link Speeds field in Link Capabilities,
5784         * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
5785         */
5786        pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5787        if (lnkcap2) { /* PCIe r3.0-compliant */
5788                if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_32_0GB)
5789                        return PCIE_SPEED_32_0GT;
5790                else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_16_0GB)
5791                        return PCIE_SPEED_16_0GT;
5792                else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
5793                        return PCIE_SPEED_8_0GT;
5794                else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
5795                        return PCIE_SPEED_5_0GT;
5796                else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
5797                        return PCIE_SPEED_2_5GT;
5798                return PCI_SPEED_UNKNOWN;
5799        }
5800
5801        pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5802        if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
5803                return PCIE_SPEED_5_0GT;
5804        else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
5805                return PCIE_SPEED_2_5GT;
5806
5807        return PCI_SPEED_UNKNOWN;
5808}
5809EXPORT_SYMBOL(pcie_get_speed_cap);
5810
5811/**
5812 * pcie_get_width_cap - query for the PCI device's link width capability
5813 * @dev: PCI device to query
5814 *
5815 * Query the PCI device width capability.  Return the maximum link width
5816 * supported by the device.
5817 */
5818enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5819{
5820        u32 lnkcap;
5821
5822        pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5823        if (lnkcap)
5824                return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5825
5826        return PCIE_LNK_WIDTH_UNKNOWN;
5827}
5828EXPORT_SYMBOL(pcie_get_width_cap);
5829
5830/**
5831 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5832 * @dev: PCI device
5833 * @speed: storage for link speed
5834 * @width: storage for link width
5835 *
5836 * Calculate a PCI device's link bandwidth by querying for its link speed
5837 * and width, multiplying them, and applying encoding overhead.  The result
5838 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5839 */
5840u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5841                           enum pcie_link_width *width)
5842{
5843        *speed = pcie_get_speed_cap(dev);
5844        *width = pcie_get_width_cap(dev);
5845
5846        if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5847                return 0;
5848
5849        return *width * PCIE_SPEED2MBS_ENC(*speed);
5850}
5851
5852/**
5853 * __pcie_print_link_status - Report the PCI device's link speed and width
5854 * @dev: PCI device to query
5855 * @verbose: Print info even when enough bandwidth is available
5856 *
5857 * If the available bandwidth at the device is less than the device is
5858 * capable of, report the device's maximum possible bandwidth and the
5859 * upstream link that limits its performance.  If @verbose, always print
5860 * the available bandwidth, even if the device isn't constrained.
5861 */
5862void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
5863{
5864        enum pcie_link_width width, width_cap;
5865        enum pci_bus_speed speed, speed_cap;
5866        struct pci_dev *limiting_dev = NULL;
5867        u32 bw_avail, bw_cap;
5868
5869        bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5870        bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5871
5872        if (bw_avail >= bw_cap && verbose)
5873                pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5874                         bw_cap / 1000, bw_cap % 1000,
5875                         PCIE_SPEED2STR(speed_cap), width_cap);
5876        else if (bw_avail < bw_cap)
5877                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",
5878                         bw_avail / 1000, bw_avail % 1000,
5879                         PCIE_SPEED2STR(speed), width,
5880                         limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5881                         bw_cap / 1000, bw_cap % 1000,
5882                         PCIE_SPEED2STR(speed_cap), width_cap);
5883}
5884
5885/**
5886 * pcie_print_link_status - Report the PCI device's link speed and width
5887 * @dev: PCI device to query
5888 *
5889 * Report the available bandwidth at the device.
5890 */
5891void pcie_print_link_status(struct pci_dev *dev)
5892{
5893        __pcie_print_link_status(dev, true);
5894}
5895EXPORT_SYMBOL(pcie_print_link_status);
5896
5897/**
5898 * pci_select_bars - Make BAR mask from the type of resource
5899 * @dev: the PCI device for which BAR mask is made
5900 * @flags: resource type mask to be selected
5901 *
5902 * This helper routine makes bar mask from the type of resource.
5903 */
5904int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5905{
5906        int i, bars = 0;
5907        for (i = 0; i < PCI_NUM_RESOURCES; i++)
5908                if (pci_resource_flags(dev, i) & flags)
5909                        bars |= (1 << i);
5910        return bars;
5911}
5912EXPORT_SYMBOL(pci_select_bars);
5913
5914/* Some architectures require additional programming to enable VGA */
5915static arch_set_vga_state_t arch_set_vga_state;
5916
5917void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5918{
5919        arch_set_vga_state = func;      /* NULL disables */
5920}
5921
5922static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5923                                  unsigned int command_bits, u32 flags)
5924{
5925        if (arch_set_vga_state)
5926                return arch_set_vga_state(dev, decode, command_bits,
5927                                                flags);
5928        return 0;
5929}
5930
5931/**
5932 * pci_set_vga_state - set VGA decode state on device and parents if requested
5933 * @dev: the PCI device
5934 * @decode: true = enable decoding, false = disable decoding
5935 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5936 * @flags: traverse ancestors and change bridges
5937 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5938 */
5939int pci_set_vga_state(struct pci_dev *dev, bool decode,
5940                      unsigned int command_bits, u32 flags)
5941{
5942        struct pci_bus *bus;
5943        struct pci_dev *bridge;
5944        u16 cmd;
5945        int rc;
5946
5947        WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5948
5949        /* ARCH specific VGA enables */
5950        rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5951        if (rc)
5952                return rc;
5953
5954        if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5955                pci_read_config_word(dev, PCI_COMMAND, &cmd);
5956                if (decode == true)
5957                        cmd |= command_bits;
5958                else
5959                        cmd &= ~command_bits;
5960                pci_write_config_word(dev, PCI_COMMAND, cmd);
5961        }
5962
5963        if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5964                return 0;
5965
5966        bus = dev->bus;
5967        while (bus) {
5968                bridge = bus->self;
5969                if (bridge) {
5970                        pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5971                                             &cmd);
5972                        if (decode == true)
5973                                cmd |= PCI_BRIDGE_CTL_VGA;
5974                        else
5975                                cmd &= ~PCI_BRIDGE_CTL_VGA;
5976                        pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5977                                              cmd);
5978                }
5979                bus = bus->parent;
5980        }
5981        return 0;
5982}
5983
5984#ifdef CONFIG_ACPI
5985bool pci_pr3_present(struct pci_dev *pdev)
5986{
5987        struct acpi_device *adev;
5988
5989        if (acpi_disabled)
5990                return false;
5991
5992        adev = ACPI_COMPANION(&pdev->dev);
5993        if (!adev)
5994                return false;
5995
5996        return adev->power.flags.power_resources &&
5997                acpi_has_method(adev->handle, "_PR3");
5998}
5999EXPORT_SYMBOL_GPL(pci_pr3_present);
6000#endif
6001
6002/**
6003 * pci_add_dma_alias - Add a DMA devfn alias for a device
6004 * @dev: the PCI device for which alias is added
6005 * @devfn_from: alias slot and function
6006 * @nr_devfns: number of subsequent devfns to alias
6007 *
6008 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
6009 * which is used to program permissible bus-devfn source addresses for DMA
6010 * requests in an IOMMU.  These aliases factor into IOMMU group creation
6011 * and are useful for devices generating DMA requests beyond or different
6012 * from their logical bus-devfn.  Examples include device quirks where the
6013 * device simply uses the wrong devfn, as well as non-transparent bridges
6014 * where the alias may be a proxy for devices in another domain.
6015 *
6016 * IOMMU group creation is performed during device discovery or addition,
6017 * prior to any potential DMA mapping and therefore prior to driver probing
6018 * (especially for userspace assigned devices where IOMMU group definition
6019 * cannot be left as a userspace activity).  DMA aliases should therefore
6020 * be configured via quirks, such as the PCI fixup header quirk.
6021 */
6022void pci_add_dma_alias(struct pci_dev *dev, u8 devfn_from, unsigned nr_devfns)
6023{
6024        int devfn_to;
6025
6026        nr_devfns = min(nr_devfns, (unsigned) MAX_NR_DEVFNS - devfn_from);
6027        devfn_to = devfn_from + nr_devfns - 1;
6028
6029        if (!dev->dma_alias_mask)
6030                dev->dma_alias_mask = bitmap_zalloc(MAX_NR_DEVFNS, GFP_KERNEL);
6031        if (!dev->dma_alias_mask) {
6032                pci_warn(dev, "Unable to allocate DMA alias mask\n");
6033                return;
6034        }
6035
6036        bitmap_set(dev->dma_alias_mask, devfn_from, nr_devfns);
6037
6038        if (nr_devfns == 1)
6039                pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
6040                                PCI_SLOT(devfn_from), PCI_FUNC(devfn_from));
6041        else if (nr_devfns > 1)
6042                pci_info(dev, "Enabling fixed DMA alias for devfn range from %02x.%d to %02x.%d\n",
6043                                PCI_SLOT(devfn_from), PCI_FUNC(devfn_from),
6044                                PCI_SLOT(devfn_to), PCI_FUNC(devfn_to));
6045}
6046
6047bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
6048{
6049        return (dev1->dma_alias_mask &&
6050                test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
6051               (dev2->dma_alias_mask &&
6052                test_bit(dev1->devfn, dev2->dma_alias_mask)) ||
6053               pci_real_dma_dev(dev1) == dev2 ||
6054               pci_real_dma_dev(dev2) == dev1;
6055}
6056
6057bool pci_device_is_present(struct pci_dev *pdev)
6058{
6059        u32 v;
6060
6061        if (pci_dev_is_disconnected(pdev))
6062                return false;
6063        return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
6064}
6065EXPORT_SYMBOL_GPL(pci_device_is_present);
6066
6067void pci_ignore_hotplug(struct pci_dev *dev)
6068{
6069        struct pci_dev *bridge = dev->bus->self;
6070
6071        dev->ignore_hotplug = 1;
6072        /* Propagate the "ignore hotplug" setting to the parent bridge. */
6073        if (bridge)
6074                bridge->ignore_hotplug = 1;
6075}
6076EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
6077
6078/**
6079 * pci_real_dma_dev - Get PCI DMA device for PCI device
6080 * @dev: the PCI device that may have a PCI DMA alias
6081 *
6082 * Permits the platform to provide architecture-specific functionality to
6083 * devices needing to alias DMA to another PCI device on another PCI bus. If
6084 * the PCI device is on the same bus, it is recommended to use
6085 * pci_add_dma_alias(). This is the default implementation. Architecture
6086 * implementations can override this.
6087 */
6088struct pci_dev __weak *pci_real_dma_dev(struct pci_dev *dev)
6089{
6090        return dev;
6091}
6092
6093resource_size_t __weak pcibios_default_alignment(void)
6094{
6095        return 0;
6096}
6097
6098/*
6099 * Arches that don't want to expose struct resource to userland as-is in
6100 * sysfs and /proc can implement their own pci_resource_to_user().
6101 */
6102void __weak pci_resource_to_user(const struct pci_dev *dev, int bar,
6103                                 const struct resource *rsrc,
6104                                 resource_size_t *start, resource_size_t *end)
6105{
6106        *start = rsrc->start;
6107        *end = rsrc->end;
6108}
6109
6110static char *resource_alignment_param;
6111static DEFINE_SPINLOCK(resource_alignment_lock);
6112
6113/**
6114 * pci_specified_resource_alignment - get resource alignment specified by user.
6115 * @dev: the PCI device to get
6116 * @resize: whether or not to change resources' size when reassigning alignment
6117 *
6118 * RETURNS: Resource alignment if it is specified.
6119 *          Zero if it is not specified.
6120 */
6121static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
6122                                                        bool *resize)
6123{
6124        int align_order, count;
6125        resource_size_t align = pcibios_default_alignment();
6126        const char *p;
6127        int ret;
6128
6129        spin_lock(&resource_alignment_lock);
6130        p = resource_alignment_param;
6131        if (!p || !*p)
6132                goto out;
6133        if (pci_has_flag(PCI_PROBE_ONLY)) {
6134                align = 0;
6135                pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
6136                goto out;
6137        }
6138
6139        while (*p) {
6140                count = 0;
6141                if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
6142                                                        p[count] == '@') {
6143                        p += count + 1;
6144                } else {
6145                        align_order = -1;
6146                }
6147
6148                ret = pci_dev_str_match(dev, p, &p);
6149                if (ret == 1) {
6150                        *resize = true;
6151                        if (align_order == -1)
6152                                align = PAGE_SIZE;
6153                        else
6154                                align = 1 << align_order;
6155                        break;
6156                } else if (ret < 0) {
6157                        pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
6158                               p);
6159                        break;
6160                }
6161
6162                if (*p != ';' && *p != ',') {
6163                        /* End of param or invalid format */
6164                        break;
6165                }
6166                p++;
6167        }
6168out:
6169        spin_unlock(&resource_alignment_lock);
6170        return align;
6171}
6172
6173static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
6174                                           resource_size_t align, bool resize)
6175{
6176        struct resource *r = &dev->resource[bar];
6177        resource_size_t size;
6178
6179        if (!(r->flags & IORESOURCE_MEM))
6180                return;
6181
6182        if (r->flags & IORESOURCE_PCI_FIXED) {
6183                pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
6184                         bar, r, (unsigned long long)align);
6185                return;
6186        }
6187
6188        size = resource_size(r);
6189        if (size >= align)
6190                return;
6191
6192        /*
6193         * Increase the alignment of the resource.  There are two ways we
6194         * can do this:
6195         *
6196         * 1) Increase the size of the resource.  BARs are aligned on their
6197         *    size, so when we reallocate space for this resource, we'll
6198         *    allocate it with the larger alignment.  This also prevents
6199         *    assignment of any other BARs inside the alignment region, so
6200         *    if we're requesting page alignment, this means no other BARs
6201         *    will share the page.
6202         *
6203         *    The disadvantage is that this makes the resource larger than
6204         *    the hardware BAR, which may break drivers that compute things
6205         *    based on the resource size, e.g., to find registers at a
6206         *    fixed offset before the end of the BAR.
6207         *
6208         * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
6209         *    set r->start to the desired alignment.  By itself this
6210         *    doesn't prevent other BARs being put inside the alignment
6211         *    region, but if we realign *every* resource of every device in
6212         *    the system, none of them will share an alignment region.
6213         *
6214         * When the user has requested alignment for only some devices via
6215         * the "pci=resource_alignment" argument, "resize" is true and we
6216         * use the first method.  Otherwise we assume we're aligning all
6217         * devices and we use the second.
6218         */
6219
6220        pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
6221                 bar, r, (unsigned long long)align);
6222
6223        if (resize) {
6224                r->start = 0;
6225                r->end = align - 1;
6226        } else {
6227                r->flags &= ~IORESOURCE_SIZEALIGN;
6228                r->flags |= IORESOURCE_STARTALIGN;
6229                r->start = align;
6230                r->end = r->start + size - 1;
6231        }
6232        r->flags |= IORESOURCE_UNSET;
6233}
6234
6235/*
6236 * This function disables memory decoding and releases memory resources
6237 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6238 * It also rounds up size to specified alignment.
6239 * Later on, the kernel will assign page-aligned memory resource back
6240 * to the device.
6241 */
6242void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6243{
6244        int i;
6245        struct resource *r;
6246        resource_size_t align;
6247        u16 command;
6248        bool resize = false;
6249
6250        /*
6251         * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6252         * 3.4.1.11.  Their resources are allocated from the space
6253         * described by the VF BARx register in the PF's SR-IOV capability.
6254         * We can't influence their alignment here.
6255         */
6256        if (dev->is_virtfn)
6257                return;
6258
6259        /* check if specified PCI is target device to reassign */
6260        align = pci_specified_resource_alignment(dev, &resize);
6261        if (!align)
6262                return;
6263
6264        if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6265            (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6266                pci_warn(dev, "Can't reassign resources to host bridge\n");
6267                return;
6268        }
6269
6270        pci_read_config_word(dev, PCI_COMMAND, &command);
6271        command &= ~PCI_COMMAND_MEMORY;
6272        pci_write_config_word(dev, PCI_COMMAND, command);
6273
6274        for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6275                pci_request_resource_alignment(dev, i, align, resize);
6276
6277        /*
6278         * Need to disable bridge's resource window,
6279         * to enable the kernel to reassign new resource
6280         * window later on.
6281         */
6282        if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6283                for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6284                        r = &dev->resource[i];
6285                        if (!(r->flags & IORESOURCE_MEM))
6286                                continue;
6287                        r->flags |= IORESOURCE_UNSET;
6288                        r->end = resource_size(r) - 1;
6289                        r->start = 0;
6290                }
6291                pci_disable_bridge_window(dev);
6292        }
6293}
6294
6295static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
6296{
6297        size_t count = 0;
6298
6299        spin_lock(&resource_alignment_lock);
6300        if (resource_alignment_param)
6301                count = snprintf(buf, PAGE_SIZE, "%s", resource_alignment_param);
6302        spin_unlock(&resource_alignment_lock);
6303
6304        /*
6305         * When set by the command line, resource_alignment_param will not
6306         * have a trailing line feed, which is ugly. So conditionally add
6307         * it here.
6308         */
6309        if (count >= 2 && buf[count - 2] != '\n' && count < PAGE_SIZE - 1) {
6310                buf[count - 1] = '\n';
6311                buf[count++] = 0;
6312        }
6313
6314        return count;
6315}
6316
6317static ssize_t resource_alignment_store(struct bus_type *bus,
6318                                        const char *buf, size_t count)
6319{
6320        char *param = kstrndup(buf, count, GFP_KERNEL);
6321
6322        if (!param)
6323                return -ENOMEM;
6324
6325        spin_lock(&resource_alignment_lock);
6326        kfree(resource_alignment_param);
6327        resource_alignment_param = param;
6328        spin_unlock(&resource_alignment_lock);
6329        return count;
6330}
6331
6332static BUS_ATTR_RW(resource_alignment);
6333
6334static int __init pci_resource_alignment_sysfs_init(void)
6335{
6336        return bus_create_file(&pci_bus_type,
6337                                        &bus_attr_resource_alignment);
6338}
6339late_initcall(pci_resource_alignment_sysfs_init);
6340
6341static void pci_no_domains(void)
6342{
6343#ifdef CONFIG_PCI_DOMAINS
6344        pci_domains_supported = 0;
6345#endif
6346}
6347
6348#ifdef CONFIG_PCI_DOMAINS_GENERIC
6349static atomic_t __domain_nr = ATOMIC_INIT(-1);
6350
6351static int pci_get_new_domain_nr(void)
6352{
6353        return atomic_inc_return(&__domain_nr);
6354}
6355
6356static int of_pci_bus_find_domain_nr(struct device *parent)
6357{
6358        static int use_dt_domains = -1;
6359        int domain = -1;
6360
6361        if (parent)
6362                domain = of_get_pci_domain_nr(parent->of_node);
6363
6364        /*
6365         * Check DT domain and use_dt_domains values.
6366         *
6367         * If DT domain property is valid (domain >= 0) and
6368         * use_dt_domains != 0, the DT assignment is valid since this means
6369         * we have not previously allocated a domain number by using
6370         * pci_get_new_domain_nr(); we should also update use_dt_domains to
6371         * 1, to indicate that we have just assigned a domain number from
6372         * DT.
6373         *
6374         * If DT domain property value is not valid (ie domain < 0), and we
6375         * have not previously assigned a domain number from DT
6376         * (use_dt_domains != 1) we should assign a domain number by
6377         * using the:
6378         *
6379         * pci_get_new_domain_nr()
6380         *
6381         * API and update the use_dt_domains value to keep track of method we
6382         * are using to assign domain numbers (use_dt_domains = 0).
6383         *
6384         * All other combinations imply we have a platform that is trying
6385         * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6386         * which is a recipe for domain mishandling and it is prevented by
6387         * invalidating the domain value (domain = -1) and printing a
6388         * corresponding error.
6389         */
6390        if (domain >= 0 && use_dt_domains) {
6391                use_dt_domains = 1;
6392        } else if (domain < 0 && use_dt_domains != 1) {
6393                use_dt_domains = 0;
6394                domain = pci_get_new_domain_nr();
6395        } else {
6396                if (parent)
6397                        pr_err("Node %pOF has ", parent->of_node);
6398                pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6399                domain = -1;
6400        }
6401
6402        return domain;
6403}
6404
6405int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6406{
6407        return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6408                               acpi_pci_bus_find_domain_nr(bus);
6409}
6410#endif
6411
6412/**
6413 * pci_ext_cfg_avail - can we access extended PCI config space?
6414 *
6415 * Returns 1 if we can access PCI extended config space (offsets
6416 * greater than 0xff). This is the default implementation. Architecture
6417 * implementations can override this.
6418 */
6419int __weak pci_ext_cfg_avail(void)
6420{
6421        return 1;
6422}
6423
6424void __weak pci_fixup_cardbus(struct pci_bus *bus)
6425{
6426}
6427EXPORT_SYMBOL(pci_fixup_cardbus);
6428
6429static int __init pci_setup(char *str)
6430{
6431        while (str) {
6432                char *k = strchr(str, ',');
6433                if (k)
6434                        *k++ = 0;
6435                if (*str && (str = pcibios_setup(str)) && *str) {
6436                        if (!strcmp(str, "nomsi")) {
6437                                pci_no_msi();
6438                        } else if (!strncmp(str, "noats", 5)) {
6439                                pr_info("PCIe: ATS is disabled\n");
6440                                pcie_ats_disabled = true;
6441                        } else if (!strcmp(str, "noaer")) {
6442                                pci_no_aer();
6443                        } else if (!strcmp(str, "earlydump")) {
6444                                pci_early_dump = true;
6445                        } else if (!strncmp(str, "realloc=", 8)) {
6446                                pci_realloc_get_opt(str + 8);
6447                        } else if (!strncmp(str, "realloc", 7)) {
6448                                pci_realloc_get_opt("on");
6449                        } else if (!strcmp(str, "nodomains")) {
6450                                pci_no_domains();
6451                        } else if (!strncmp(str, "noari", 5)) {
6452                                pcie_ari_disabled = true;
6453                        } else if (!strncmp(str, "cbiosize=", 9)) {
6454                                pci_cardbus_io_size = memparse(str + 9, &str);
6455                        } else if (!strncmp(str, "cbmemsize=", 10)) {
6456                                pci_cardbus_mem_size = memparse(str + 10, &str);
6457                        } else if (!strncmp(str, "resource_alignment=", 19)) {
6458                                resource_alignment_param = str + 19;
6459                        } else if (!strncmp(str, "ecrc=", 5)) {
6460                                pcie_ecrc_get_policy(str + 5);
6461                        } else if (!strncmp(str, "hpiosize=", 9)) {
6462                                pci_hotplug_io_size = memparse(str + 9, &str);
6463                        } else if (!strncmp(str, "hpmmiosize=", 11)) {
6464                                pci_hotplug_mmio_size = memparse(str + 11, &str);
6465                        } else if (!strncmp(str, "hpmmioprefsize=", 15)) {
6466                                pci_hotplug_mmio_pref_size = memparse(str + 15, &str);
6467                        } else if (!strncmp(str, "hpmemsize=", 10)) {
6468                                pci_hotplug_mmio_size = memparse(str + 10, &str);
6469                                pci_hotplug_mmio_pref_size = pci_hotplug_mmio_size;
6470                        } else if (!strncmp(str, "hpbussize=", 10)) {
6471                                pci_hotplug_bus_size =
6472                                        simple_strtoul(str + 10, &str, 0);
6473                                if (pci_hotplug_bus_size > 0xff)
6474                                        pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6475                        } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6476                                pcie_bus_config = PCIE_BUS_TUNE_OFF;
6477                        } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6478                                pcie_bus_config = PCIE_BUS_SAFE;
6479                        } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6480                                pcie_bus_config = PCIE_BUS_PERFORMANCE;
6481                        } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6482                                pcie_bus_config = PCIE_BUS_PEER2PEER;
6483                        } else if (!strncmp(str, "pcie_scan_all", 13)) {
6484                                pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6485                        } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6486                                disable_acs_redir_param = str + 18;
6487                        } else {
6488                                pr_err("PCI: Unknown option `%s'\n", str);
6489                        }
6490                }
6491                str = k;
6492        }
6493        return 0;
6494}
6495early_param("pci", pci_setup);
6496
6497/*
6498 * 'resource_alignment_param' and 'disable_acs_redir_param' are initialized
6499 * in pci_setup(), above, to point to data in the __initdata section which
6500 * will be freed after the init sequence is complete. We can't allocate memory
6501 * in pci_setup() because some architectures do not have any memory allocation
6502 * service available during an early_param() call. So we allocate memory and
6503 * copy the variable here before the init section is freed.
6504 *
6505 */
6506static int __init pci_realloc_setup_params(void)
6507{
6508        resource_alignment_param = kstrdup(resource_alignment_param,
6509                                           GFP_KERNEL);
6510        disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6511
6512        return 0;
6513}
6514pure_initcall(pci_realloc_setup_params);
6515