linux/drivers/pci/pci.c
<<
>>
Prefs
   1/*
   2 *      PCI Bus Services, see include/linux/pci.h for further explanation.
   3 *
   4 *      Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
   5 *      David Mosberger-Tang
   6 *
   7 *      Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
   8 */
   9
  10#include <linux/acpi.h>
  11#include <linux/kernel.h>
  12#include <linux/delay.h>
  13#include <linux/dmi.h>
  14#include <linux/init.h>
  15#include <linux/of.h>
  16#include <linux/of_pci.h>
  17#include <linux/pci.h>
  18#include <linux/pm.h>
  19#include <linux/slab.h>
  20#include <linux/module.h>
  21#include <linux/spinlock.h>
  22#include <linux/string.h>
  23#include <linux/log2.h>
  24#include <linux/pci-aspm.h>
  25#include <linux/pm_wakeup.h>
  26#include <linux/interrupt.h>
  27#include <linux/device.h>
  28#include <linux/pm_runtime.h>
  29#include <linux/pci_hotplug.h>
  30#include <linux/vmalloc.h>
  31#include <asm/setup.h>
  32#include <asm/dma.h>
  33#include <linux/aer.h>
  34#include "pci.h"
  35
  36const char *pci_power_names[] = {
  37        "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
  38};
  39EXPORT_SYMBOL_GPL(pci_power_names);
  40
  41int isa_dma_bridge_buggy;
  42EXPORT_SYMBOL(isa_dma_bridge_buggy);
  43
  44int pci_pci_problems;
  45EXPORT_SYMBOL(pci_pci_problems);
  46
  47unsigned int pci_pm_d3_delay;
  48
  49static void pci_pme_list_scan(struct work_struct *work);
  50
  51static LIST_HEAD(pci_pme_list);
  52static DEFINE_MUTEX(pci_pme_list_mutex);
  53static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
  54
  55struct pci_pme_device {
  56        struct list_head list;
  57        struct pci_dev *dev;
  58};
  59
  60#define PME_TIMEOUT 1000 /* How long between PME checks */
  61
  62static void pci_dev_d3_sleep(struct pci_dev *dev)
  63{
  64        unsigned int delay = dev->d3_delay;
  65
  66        if (delay < pci_pm_d3_delay)
  67                delay = pci_pm_d3_delay;
  68
  69        msleep(delay);
  70}
  71
  72#ifdef CONFIG_PCI_DOMAINS
  73int pci_domains_supported = 1;
  74#endif
  75
  76#define DEFAULT_CARDBUS_IO_SIZE         (256)
  77#define DEFAULT_CARDBUS_MEM_SIZE        (64*1024*1024)
  78/* pci=cbmemsize=nnM,cbiosize=nn can override this */
  79unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
  80unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
  81
  82#define DEFAULT_HOTPLUG_IO_SIZE         (256)
  83#define DEFAULT_HOTPLUG_MEM_SIZE        (2*1024*1024)
  84/* pci=hpmemsize=nnM,hpiosize=nn can override this */
  85unsigned long pci_hotplug_io_size  = DEFAULT_HOTPLUG_IO_SIZE;
  86unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
  87
  88#define DEFAULT_HOTPLUG_BUS_SIZE        1
  89unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
  90
  91enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
  92
  93/*
  94 * The default CLS is used if arch didn't set CLS explicitly and not
  95 * all pci devices agree on the same value.  Arch can override either
  96 * the dfl or actual value as it sees fit.  Don't forget this is
  97 * measured in 32-bit words, not bytes.
  98 */
  99u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
 100u8 pci_cache_line_size;
 101
 102/*
 103 * If we set up a device for bus mastering, we need to check the latency
 104 * timer as certain BIOSes forget to set it properly.
 105 */
 106unsigned int pcibios_max_latency = 255;
 107
 108/* If set, the PCIe ARI capability will not be used. */
 109static bool pcie_ari_disabled;
 110
 111/* Disable bridge_d3 for all PCIe ports */
 112static bool pci_bridge_d3_disable;
 113/* Force bridge_d3 for all PCIe ports */
 114static bool pci_bridge_d3_force;
 115
 116static int __init pcie_port_pm_setup(char *str)
 117{
 118        if (!strcmp(str, "off"))
 119                pci_bridge_d3_disable = true;
 120        else if (!strcmp(str, "force"))
 121                pci_bridge_d3_force = true;
 122        return 1;
 123}
 124__setup("pcie_port_pm=", pcie_port_pm_setup);
 125
 126/**
 127 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
 128 * @bus: pointer to PCI bus structure to search
 129 *
 130 * Given a PCI bus, returns the highest PCI bus number present in the set
 131 * including the given PCI bus and its list of child PCI buses.
 132 */
 133unsigned char pci_bus_max_busnr(struct pci_bus *bus)
 134{
 135        struct pci_bus *tmp;
 136        unsigned char max, n;
 137
 138        max = bus->busn_res.end;
 139        list_for_each_entry(tmp, &bus->children, node) {
 140                n = pci_bus_max_busnr(tmp);
 141                if (n > max)
 142                        max = n;
 143        }
 144        return max;
 145}
 146EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
 147
 148#ifdef CONFIG_HAS_IOMEM
 149void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
 150{
 151        struct resource *res = &pdev->resource[bar];
 152
 153        /*
 154         * Make sure the BAR is actually a memory resource, not an IO resource
 155         */
 156        if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
 157                dev_warn(&pdev->dev, "can't ioremap BAR %d: %pR\n", bar, res);
 158                return NULL;
 159        }
 160        return ioremap_nocache(res->start, resource_size(res));
 161}
 162EXPORT_SYMBOL_GPL(pci_ioremap_bar);
 163
 164void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
 165{
 166        /*
 167         * Make sure the BAR is actually a memory resource, not an IO resource
 168         */
 169        if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
 170                WARN_ON(1);
 171                return NULL;
 172        }
 173        return ioremap_wc(pci_resource_start(pdev, bar),
 174                          pci_resource_len(pdev, bar));
 175}
 176EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
 177#endif
 178
 179
 180static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
 181                                   u8 pos, int cap, int *ttl)
 182{
 183        u8 id;
 184        u16 ent;
 185
 186        pci_bus_read_config_byte(bus, devfn, pos, &pos);
 187
 188        while ((*ttl)--) {
 189                if (pos < 0x40)
 190                        break;
 191                pos &= ~3;
 192                pci_bus_read_config_word(bus, devfn, pos, &ent);
 193
 194                id = ent & 0xff;
 195                if (id == 0xff)
 196                        break;
 197                if (id == cap)
 198                        return pos;
 199                pos = (ent >> 8);
 200        }
 201        return 0;
 202}
 203
 204static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
 205                               u8 pos, int cap)
 206{
 207        int ttl = PCI_FIND_CAP_TTL;
 208
 209        return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
 210}
 211
 212int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
 213{
 214        return __pci_find_next_cap(dev->bus, dev->devfn,
 215                                   pos + PCI_CAP_LIST_NEXT, cap);
 216}
 217EXPORT_SYMBOL_GPL(pci_find_next_capability);
 218
 219static int __pci_bus_find_cap_start(struct pci_bus *bus,
 220                                    unsigned int devfn, u8 hdr_type)
 221{
 222        u16 status;
 223
 224        pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
 225        if (!(status & PCI_STATUS_CAP_LIST))
 226                return 0;
 227
 228        switch (hdr_type) {
 229        case PCI_HEADER_TYPE_NORMAL:
 230        case PCI_HEADER_TYPE_BRIDGE:
 231                return PCI_CAPABILITY_LIST;
 232        case PCI_HEADER_TYPE_CARDBUS:
 233                return PCI_CB_CAPABILITY_LIST;
 234        }
 235
 236        return 0;
 237}
 238
 239/**
 240 * pci_find_capability - query for devices' capabilities
 241 * @dev: PCI device to query
 242 * @cap: capability code
 243 *
 244 * Tell if a device supports a given PCI capability.
 245 * Returns the address of the requested capability structure within the
 246 * device's PCI configuration space or 0 in case the device does not
 247 * support it.  Possible values for @cap:
 248 *
 249 *  %PCI_CAP_ID_PM           Power Management
 250 *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
 251 *  %PCI_CAP_ID_VPD          Vital Product Data
 252 *  %PCI_CAP_ID_SLOTID       Slot Identification
 253 *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
 254 *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
 255 *  %PCI_CAP_ID_PCIX         PCI-X
 256 *  %PCI_CAP_ID_EXP          PCI Express
 257 */
 258int pci_find_capability(struct pci_dev *dev, int cap)
 259{
 260        int pos;
 261
 262        pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 263        if (pos)
 264                pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
 265
 266        return pos;
 267}
 268EXPORT_SYMBOL(pci_find_capability);
 269
 270/**
 271 * pci_bus_find_capability - query for devices' capabilities
 272 * @bus:   the PCI bus to query
 273 * @devfn: PCI device to query
 274 * @cap:   capability code
 275 *
 276 * Like pci_find_capability() but works for pci devices that do not have a
 277 * pci_dev structure set up yet.
 278 *
 279 * Returns the address of the requested capability structure within the
 280 * device's PCI configuration space or 0 in case the device does not
 281 * support it.
 282 */
 283int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
 284{
 285        int pos;
 286        u8 hdr_type;
 287
 288        pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
 289
 290        pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
 291        if (pos)
 292                pos = __pci_find_next_cap(bus, devfn, pos, cap);
 293
 294        return pos;
 295}
 296EXPORT_SYMBOL(pci_bus_find_capability);
 297
 298/**
 299 * pci_find_next_ext_capability - Find an extended capability
 300 * @dev: PCI device to query
 301 * @start: address at which to start looking (0 to start at beginning of list)
 302 * @cap: capability code
 303 *
 304 * Returns the address of the next matching extended capability structure
 305 * within the device's PCI configuration space or 0 if the device does
 306 * not support it.  Some capabilities can occur several times, e.g., the
 307 * vendor-specific capability, and this provides a way to find them all.
 308 */
 309int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
 310{
 311        u32 header;
 312        int ttl;
 313        int pos = PCI_CFG_SPACE_SIZE;
 314
 315        /* minimum 8 bytes per capability */
 316        ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
 317
 318        if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
 319                return 0;
 320
 321        if (start)
 322                pos = start;
 323
 324        if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 325                return 0;
 326
 327        /*
 328         * If we have no capabilities, this is indicated by cap ID,
 329         * cap version and next pointer all being 0.
 330         */
 331        if (header == 0)
 332                return 0;
 333
 334        while (ttl-- > 0) {
 335                if (PCI_EXT_CAP_ID(header) == cap && pos != start)
 336                        return pos;
 337
 338                pos = PCI_EXT_CAP_NEXT(header);
 339                if (pos < PCI_CFG_SPACE_SIZE)
 340                        break;
 341
 342                if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 343                        break;
 344        }
 345
 346        return 0;
 347}
 348EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
 349
 350/**
 351 * pci_find_ext_capability - Find an extended capability
 352 * @dev: PCI device to query
 353 * @cap: capability code
 354 *
 355 * Returns the address of the requested extended capability structure
 356 * within the device's PCI configuration space or 0 if the device does
 357 * not support it.  Possible values for @cap:
 358 *
 359 *  %PCI_EXT_CAP_ID_ERR         Advanced Error Reporting
 360 *  %PCI_EXT_CAP_ID_VC          Virtual Channel
 361 *  %PCI_EXT_CAP_ID_DSN         Device Serial Number
 362 *  %PCI_EXT_CAP_ID_PWR         Power Budgeting
 363 */
 364int pci_find_ext_capability(struct pci_dev *dev, int cap)
 365{
 366        return pci_find_next_ext_capability(dev, 0, cap);
 367}
 368EXPORT_SYMBOL_GPL(pci_find_ext_capability);
 369
 370static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
 371{
 372        int rc, ttl = PCI_FIND_CAP_TTL;
 373        u8 cap, mask;
 374
 375        if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
 376                mask = HT_3BIT_CAP_MASK;
 377        else
 378                mask = HT_5BIT_CAP_MASK;
 379
 380        pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
 381                                      PCI_CAP_ID_HT, &ttl);
 382        while (pos) {
 383                rc = pci_read_config_byte(dev, pos + 3, &cap);
 384                if (rc != PCIBIOS_SUCCESSFUL)
 385                        return 0;
 386
 387                if ((cap & mask) == ht_cap)
 388                        return pos;
 389
 390                pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
 391                                              pos + PCI_CAP_LIST_NEXT,
 392                                              PCI_CAP_ID_HT, &ttl);
 393        }
 394
 395        return 0;
 396}
 397/**
 398 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
 399 * @dev: PCI device to query
 400 * @pos: Position from which to continue searching
 401 * @ht_cap: Hypertransport capability code
 402 *
 403 * To be used in conjunction with pci_find_ht_capability() to search for
 404 * all capabilities matching @ht_cap. @pos should always be a value returned
 405 * from pci_find_ht_capability().
 406 *
 407 * NB. To be 100% safe against broken PCI devices, the caller should take
 408 * steps to avoid an infinite loop.
 409 */
 410int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
 411{
 412        return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
 413}
 414EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
 415
 416/**
 417 * pci_find_ht_capability - query a device's Hypertransport capabilities
 418 * @dev: PCI device to query
 419 * @ht_cap: Hypertransport capability code
 420 *
 421 * Tell if a device supports a given Hypertransport capability.
 422 * Returns an address within the device's PCI configuration space
 423 * or 0 in case the device does not support the request capability.
 424 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
 425 * which has a Hypertransport capability matching @ht_cap.
 426 */
 427int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
 428{
 429        int pos;
 430
 431        pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 432        if (pos)
 433                pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
 434
 435        return pos;
 436}
 437EXPORT_SYMBOL_GPL(pci_find_ht_capability);
 438
 439/**
 440 * pci_find_parent_resource - return resource region of parent bus of given region
 441 * @dev: PCI device structure contains resources to be searched
 442 * @res: child resource record for which parent is sought
 443 *
 444 *  For given resource region of given device, return the resource
 445 *  region of parent bus the given region is contained in.
 446 */
 447struct resource *pci_find_parent_resource(const struct pci_dev *dev,
 448                                          struct resource *res)
 449{
 450        const struct pci_bus *bus = dev->bus;
 451        struct resource *r;
 452        int i;
 453
 454        pci_bus_for_each_resource(bus, r, i) {
 455                if (!r)
 456                        continue;
 457                if (res->start && resource_contains(r, res)) {
 458
 459                        /*
 460                         * If the window is prefetchable but the BAR is
 461                         * not, the allocator made a mistake.
 462                         */
 463                        if (r->flags & IORESOURCE_PREFETCH &&
 464                            !(res->flags & IORESOURCE_PREFETCH))
 465                                return NULL;
 466
 467                        /*
 468                         * If we're below a transparent bridge, there may
 469                         * be both a positively-decoded aperture and a
 470                         * subtractively-decoded region that contain the BAR.
 471                         * We want the positively-decoded one, so this depends
 472                         * on pci_bus_for_each_resource() giving us those
 473                         * first.
 474                         */
 475                        return r;
 476                }
 477        }
 478        return NULL;
 479}
 480EXPORT_SYMBOL(pci_find_parent_resource);
 481
 482/**
 483 * pci_find_pcie_root_port - return PCIe Root Port
 484 * @dev: PCI device to query
 485 *
 486 * Traverse up the parent chain and return the PCIe Root Port PCI Device
 487 * for a given PCI Device.
 488 */
 489struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
 490{
 491        struct pci_dev *bridge, *highest_pcie_bridge = NULL;
 492
 493        bridge = pci_upstream_bridge(dev);
 494        while (bridge && pci_is_pcie(bridge)) {
 495                highest_pcie_bridge = bridge;
 496                bridge = pci_upstream_bridge(bridge);
 497        }
 498
 499        if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
 500                return NULL;
 501
 502        return highest_pcie_bridge;
 503}
 504EXPORT_SYMBOL(pci_find_pcie_root_port);
 505
 506/**
 507 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
 508 * @dev: the PCI device to operate on
 509 * @pos: config space offset of status word
 510 * @mask: mask of bit(s) to care about in status word
 511 *
 512 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
 513 */
 514int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
 515{
 516        int i;
 517
 518        /* Wait for Transaction Pending bit clean */
 519        for (i = 0; i < 4; i++) {
 520                u16 status;
 521                if (i)
 522                        msleep((1 << (i - 1)) * 100);
 523
 524                pci_read_config_word(dev, pos, &status);
 525                if (!(status & mask))
 526                        return 1;
 527        }
 528
 529        return 0;
 530}
 531
 532/**
 533 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
 534 * @dev: PCI device to have its BARs restored
 535 *
 536 * Restore the BAR values for a given device, so as to make it
 537 * accessible by its driver.
 538 */
 539static void pci_restore_bars(struct pci_dev *dev)
 540{
 541        int i;
 542
 543        /* Per SR-IOV spec 3.4.1.11, VF BARs are RO zero */
 544        if (dev->is_virtfn)
 545                return;
 546
 547        for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
 548                pci_update_resource(dev, i);
 549}
 550
 551static const struct pci_platform_pm_ops *pci_platform_pm;
 552
 553int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
 554{
 555        if (!ops->is_manageable || !ops->set_state || !ops->choose_state ||
 556            !ops->sleep_wake || !ops->run_wake || !ops->need_resume)
 557                return -EINVAL;
 558        pci_platform_pm = ops;
 559        return 0;
 560}
 561
 562static inline bool platform_pci_power_manageable(struct pci_dev *dev)
 563{
 564        return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
 565}
 566
 567static inline int platform_pci_set_power_state(struct pci_dev *dev,
 568                                               pci_power_t t)
 569{
 570        return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
 571}
 572
 573static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
 574{
 575        return pci_platform_pm ?
 576                        pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
 577}
 578
 579static inline int platform_pci_sleep_wake(struct pci_dev *dev, bool enable)
 580{
 581        return pci_platform_pm ?
 582                        pci_platform_pm->sleep_wake(dev, enable) : -ENODEV;
 583}
 584
 585static inline int platform_pci_run_wake(struct pci_dev *dev, bool enable)
 586{
 587        return pci_platform_pm ?
 588                        pci_platform_pm->run_wake(dev, enable) : -ENODEV;
 589}
 590
 591static inline bool platform_pci_need_resume(struct pci_dev *dev)
 592{
 593        return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
 594}
 595
 596/**
 597 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
 598 *                           given PCI device
 599 * @dev: PCI device to handle.
 600 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
 601 *
 602 * RETURN VALUE:
 603 * -EINVAL if the requested state is invalid.
 604 * -EIO if device does not support PCI PM or its PM capabilities register has a
 605 * wrong version, or device doesn't support the requested state.
 606 * 0 if device already is in the requested state.
 607 * 0 if device's power state has been successfully changed.
 608 */
 609static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
 610{
 611        u16 pmcsr;
 612        bool need_restore = false;
 613
 614        /* Check if we're already there */
 615        if (dev->current_state == state)
 616                return 0;
 617
 618        if (!dev->pm_cap)
 619                return -EIO;
 620
 621        if (state < PCI_D0 || state > PCI_D3hot)
 622                return -EINVAL;
 623
 624        /* Validate current state:
 625         * Can enter D0 from any state, but if we can only go deeper
 626         * to sleep if we're already in a low power state
 627         */
 628        if (state != PCI_D0 && dev->current_state <= PCI_D3cold
 629            && dev->current_state > state) {
 630                dev_err(&dev->dev, "invalid power transition (from state %d to %d)\n",
 631                        dev->current_state, state);
 632                return -EINVAL;
 633        }
 634
 635        /* check if this device supports the desired state */
 636        if ((state == PCI_D1 && !dev->d1_support)
 637           || (state == PCI_D2 && !dev->d2_support))
 638                return -EIO;
 639
 640        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 641
 642        /* If we're (effectively) in D3, force entire word to 0.
 643         * This doesn't affect PME_Status, disables PME_En, and
 644         * sets PowerState to 0.
 645         */
 646        switch (dev->current_state) {
 647        case PCI_D0:
 648        case PCI_D1:
 649        case PCI_D2:
 650                pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
 651                pmcsr |= state;
 652                break;
 653        case PCI_D3hot:
 654        case PCI_D3cold:
 655        case PCI_UNKNOWN: /* Boot-up */
 656                if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
 657                 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
 658                        need_restore = true;
 659                /* Fall-through: force to D0 */
 660        default:
 661                pmcsr = 0;
 662                break;
 663        }
 664
 665        /* enter specified state */
 666        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
 667
 668        /* Mandatory power management transition delays */
 669        /* see PCI PM 1.1 5.6.1 table 18 */
 670        if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
 671                pci_dev_d3_sleep(dev);
 672        else if (state == PCI_D2 || dev->current_state == PCI_D2)
 673                udelay(PCI_PM_D2_DELAY);
 674
 675        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 676        dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
 677        if (dev->current_state != state && printk_ratelimit())
 678                dev_info(&dev->dev, "Refused to change power state, currently in D%d\n",
 679                         dev->current_state);
 680
 681        /*
 682         * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
 683         * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
 684         * from D3hot to D0 _may_ perform an internal reset, thereby
 685         * going to "D0 Uninitialized" rather than "D0 Initialized".
 686         * For example, at least some versions of the 3c905B and the
 687         * 3c556B exhibit this behaviour.
 688         *
 689         * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
 690         * devices in a D3hot state at boot.  Consequently, we need to
 691         * restore at least the BARs so that the device will be
 692         * accessible to its driver.
 693         */
 694        if (need_restore)
 695                pci_restore_bars(dev);
 696
 697        if (dev->bus->self)
 698                pcie_aspm_pm_state_change(dev->bus->self);
 699
 700        return 0;
 701}
 702
 703/**
 704 * pci_update_current_state - Read PCI power state of given device from its
 705 *                            PCI PM registers and cache it
 706 * @dev: PCI device to handle.
 707 * @state: State to cache in case the device doesn't have the PM capability
 708 */
 709void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
 710{
 711        if (dev->pm_cap) {
 712                u16 pmcsr;
 713
 714                /*
 715                 * Configuration space is not accessible for device in
 716                 * D3cold, so just keep or set D3cold for safety
 717                 */
 718                if (dev->current_state == PCI_D3cold)
 719                        return;
 720                if (state == PCI_D3cold) {
 721                        dev->current_state = PCI_D3cold;
 722                        return;
 723                }
 724                pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
 725                dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
 726        } else {
 727                dev->current_state = state;
 728        }
 729}
 730
 731/**
 732 * pci_power_up - Put the given device into D0 forcibly
 733 * @dev: PCI device to power up
 734 */
 735void pci_power_up(struct pci_dev *dev)
 736{
 737        if (platform_pci_power_manageable(dev))
 738                platform_pci_set_power_state(dev, PCI_D0);
 739
 740        pci_raw_set_power_state(dev, PCI_D0);
 741        pci_update_current_state(dev, PCI_D0);
 742}
 743
 744/**
 745 * pci_platform_power_transition - Use platform to change device power state
 746 * @dev: PCI device to handle.
 747 * @state: State to put the device into.
 748 */
 749static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
 750{
 751        int error;
 752
 753        if (platform_pci_power_manageable(dev)) {
 754                error = platform_pci_set_power_state(dev, state);
 755                if (!error)
 756                        pci_update_current_state(dev, state);
 757        } else
 758                error = -ENODEV;
 759
 760        if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
 761                dev->current_state = PCI_D0;
 762
 763        return error;
 764}
 765
 766/**
 767 * pci_wakeup - Wake up a PCI device
 768 * @pci_dev: Device to handle.
 769 * @ign: ignored parameter
 770 */
 771static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
 772{
 773        pci_wakeup_event(pci_dev);
 774        pm_request_resume(&pci_dev->dev);
 775        return 0;
 776}
 777
 778/**
 779 * pci_wakeup_bus - Walk given bus and wake up devices on it
 780 * @bus: Top bus of the subtree to walk.
 781 */
 782static void pci_wakeup_bus(struct pci_bus *bus)
 783{
 784        if (bus)
 785                pci_walk_bus(bus, pci_wakeup, NULL);
 786}
 787
 788/**
 789 * __pci_start_power_transition - Start power transition of a PCI device
 790 * @dev: PCI device to handle.
 791 * @state: State to put the device into.
 792 */
 793static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
 794{
 795        if (state == PCI_D0) {
 796                pci_platform_power_transition(dev, PCI_D0);
 797                /*
 798                 * Mandatory power management transition delays, see
 799                 * PCI Express Base Specification Revision 2.0 Section
 800                 * 6.6.1: Conventional Reset.  Do not delay for
 801                 * devices powered on/off by corresponding bridge,
 802                 * because have already delayed for the bridge.
 803                 */
 804                if (dev->runtime_d3cold) {
 805                        msleep(dev->d3cold_delay);
 806                        /*
 807                         * When powering on a bridge from D3cold, the
 808                         * whole hierarchy may be powered on into
 809                         * D0uninitialized state, resume them to give
 810                         * them a chance to suspend again
 811                         */
 812                        pci_wakeup_bus(dev->subordinate);
 813                }
 814        }
 815}
 816
 817/**
 818 * __pci_dev_set_current_state - Set current state of a PCI device
 819 * @dev: Device to handle
 820 * @data: pointer to state to be set
 821 */
 822static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
 823{
 824        pci_power_t state = *(pci_power_t *)data;
 825
 826        dev->current_state = state;
 827        return 0;
 828}
 829
 830/**
 831 * __pci_bus_set_current_state - Walk given bus and set current state of devices
 832 * @bus: Top bus of the subtree to walk.
 833 * @state: state to be set
 834 */
 835static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
 836{
 837        if (bus)
 838                pci_walk_bus(bus, __pci_dev_set_current_state, &state);
 839}
 840
 841/**
 842 * __pci_complete_power_transition - Complete power transition of a PCI device
 843 * @dev: PCI device to handle.
 844 * @state: State to put the device into.
 845 *
 846 * This function should not be called directly by device drivers.
 847 */
 848int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
 849{
 850        int ret;
 851
 852        if (state <= PCI_D0)
 853                return -EINVAL;
 854        ret = pci_platform_power_transition(dev, state);
 855        /* Power off the bridge may power off the whole hierarchy */
 856        if (!ret && state == PCI_D3cold)
 857                __pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
 858        return ret;
 859}
 860EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
 861
 862/**
 863 * pci_set_power_state - Set the power state of a PCI device
 864 * @dev: PCI device to handle.
 865 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
 866 *
 867 * Transition a device to a new power state, using the platform firmware and/or
 868 * the device's PCI PM registers.
 869 *
 870 * RETURN VALUE:
 871 * -EINVAL if the requested state is invalid.
 872 * -EIO if device does not support PCI PM or its PM capabilities register has a
 873 * wrong version, or device doesn't support the requested state.
 874 * 0 if device already is in the requested state.
 875 * 0 if device's power state has been successfully changed.
 876 */
 877int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
 878{
 879        int error;
 880
 881        /* bound the state we're entering */
 882        if (state > PCI_D3cold)
 883                state = PCI_D3cold;
 884        else if (state < PCI_D0)
 885                state = PCI_D0;
 886        else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
 887                /*
 888                 * If the device or the parent bridge do not support PCI PM,
 889                 * ignore the request if we're doing anything other than putting
 890                 * it into D0 (which would only happen on boot).
 891                 */
 892                return 0;
 893
 894        /* Check if we're already there */
 895        if (dev->current_state == state)
 896                return 0;
 897
 898        __pci_start_power_transition(dev, state);
 899
 900        /* This device is quirked not to be put into D3, so
 901           don't put it in D3 */
 902        if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
 903                return 0;
 904
 905        /*
 906         * To put device in D3cold, we put device into D3hot in native
 907         * way, then put device into D3cold with platform ops
 908         */
 909        error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
 910                                        PCI_D3hot : state);
 911
 912        if (!__pci_complete_power_transition(dev, state))
 913                error = 0;
 914
 915        return error;
 916}
 917EXPORT_SYMBOL(pci_set_power_state);
 918
 919/**
 920 * pci_choose_state - Choose the power state of a PCI device
 921 * @dev: PCI device to be suspended
 922 * @state: target sleep state for the whole system. This is the value
 923 *      that is passed to suspend() function.
 924 *
 925 * Returns PCI power state suitable for given device and given system
 926 * message.
 927 */
 928
 929pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
 930{
 931        pci_power_t ret;
 932
 933        if (!dev->pm_cap)
 934                return PCI_D0;
 935
 936        ret = platform_pci_choose_state(dev);
 937        if (ret != PCI_POWER_ERROR)
 938                return ret;
 939
 940        switch (state.event) {
 941        case PM_EVENT_ON:
 942                return PCI_D0;
 943        case PM_EVENT_FREEZE:
 944        case PM_EVENT_PRETHAW:
 945                /* REVISIT both freeze and pre-thaw "should" use D0 */
 946        case PM_EVENT_SUSPEND:
 947        case PM_EVENT_HIBERNATE:
 948                return PCI_D3hot;
 949        default:
 950                dev_info(&dev->dev, "unrecognized suspend event %d\n",
 951                         state.event);
 952                BUG();
 953        }
 954        return PCI_D0;
 955}
 956EXPORT_SYMBOL(pci_choose_state);
 957
 958#define PCI_EXP_SAVE_REGS       7
 959
 960static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
 961                                                       u16 cap, bool extended)
 962{
 963        struct pci_cap_saved_state *tmp;
 964
 965        hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
 966                if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
 967                        return tmp;
 968        }
 969        return NULL;
 970}
 971
 972struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
 973{
 974        return _pci_find_saved_cap(dev, cap, false);
 975}
 976
 977struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
 978{
 979        return _pci_find_saved_cap(dev, cap, true);
 980}
 981
 982static int pci_save_pcie_state(struct pci_dev *dev)
 983{
 984        int i = 0;
 985        struct pci_cap_saved_state *save_state;
 986        u16 *cap;
 987
 988        if (!pci_is_pcie(dev))
 989                return 0;
 990
 991        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
 992        if (!save_state) {
 993                dev_err(&dev->dev, "buffer not found in %s\n", __func__);
 994                return -ENOMEM;
 995        }
 996
 997        cap = (u16 *)&save_state->cap.data[0];
 998        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
 999        pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1000        pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1001        pcie_capability_read_word(dev, PCI_EXP_RTCTL,  &cap[i++]);
1002        pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1003        pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1004        pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1005
1006        return 0;
1007}
1008
1009static void pci_restore_pcie_state(struct pci_dev *dev)
1010{
1011        int i = 0;
1012        struct pci_cap_saved_state *save_state;
1013        u16 *cap;
1014
1015        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1016        if (!save_state)
1017                return;
1018
1019        cap = (u16 *)&save_state->cap.data[0];
1020        pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1021        pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1022        pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1023        pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1024        pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1025        pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1026        pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1027}
1028
1029
1030static int pci_save_pcix_state(struct pci_dev *dev)
1031{
1032        int pos;
1033        struct pci_cap_saved_state *save_state;
1034
1035        pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1036        if (!pos)
1037                return 0;
1038
1039        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1040        if (!save_state) {
1041                dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1042                return -ENOMEM;
1043        }
1044
1045        pci_read_config_word(dev, pos + PCI_X_CMD,
1046                             (u16 *)save_state->cap.data);
1047
1048        return 0;
1049}
1050
1051static void pci_restore_pcix_state(struct pci_dev *dev)
1052{
1053        int i = 0, pos;
1054        struct pci_cap_saved_state *save_state;
1055        u16 *cap;
1056
1057        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1058        pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1059        if (!save_state || !pos)
1060                return;
1061        cap = (u16 *)&save_state->cap.data[0];
1062
1063        pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1064}
1065
1066
1067/**
1068 * pci_save_state - save the PCI configuration space of a device before suspending
1069 * @dev: - PCI device that we're dealing with
1070 */
1071int pci_save_state(struct pci_dev *dev)
1072{
1073        int i;
1074        /* XXX: 100% dword access ok here? */
1075        for (i = 0; i < 16; i++)
1076                pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1077        dev->state_saved = true;
1078
1079        i = pci_save_pcie_state(dev);
1080        if (i != 0)
1081                return i;
1082
1083        i = pci_save_pcix_state(dev);
1084        if (i != 0)
1085                return i;
1086
1087        return pci_save_vc_state(dev);
1088}
1089EXPORT_SYMBOL(pci_save_state);
1090
1091static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1092                                     u32 saved_val, int retry)
1093{
1094        u32 val;
1095
1096        pci_read_config_dword(pdev, offset, &val);
1097        if (val == saved_val)
1098                return;
1099
1100        for (;;) {
1101                dev_dbg(&pdev->dev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1102                        offset, val, saved_val);
1103                pci_write_config_dword(pdev, offset, saved_val);
1104                if (retry-- <= 0)
1105                        return;
1106
1107                pci_read_config_dword(pdev, offset, &val);
1108                if (val == saved_val)
1109                        return;
1110
1111                mdelay(1);
1112        }
1113}
1114
1115static void pci_restore_config_space_range(struct pci_dev *pdev,
1116                                           int start, int end, int retry)
1117{
1118        int index;
1119
1120        for (index = end; index >= start; index--)
1121                pci_restore_config_dword(pdev, 4 * index,
1122                                         pdev->saved_config_space[index],
1123                                         retry);
1124}
1125
1126static void pci_restore_config_space(struct pci_dev *pdev)
1127{
1128        if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1129                pci_restore_config_space_range(pdev, 10, 15, 0);
1130                /* Restore BARs before the command register. */
1131                pci_restore_config_space_range(pdev, 4, 9, 10);
1132                pci_restore_config_space_range(pdev, 0, 3, 0);
1133        } else {
1134                pci_restore_config_space_range(pdev, 0, 15, 0);
1135        }
1136}
1137
1138/**
1139 * pci_restore_state - Restore the saved state of a PCI device
1140 * @dev: - PCI device that we're dealing with
1141 */
1142void pci_restore_state(struct pci_dev *dev)
1143{
1144        if (!dev->state_saved)
1145                return;
1146
1147        /* PCI Express register must be restored first */
1148        pci_restore_pcie_state(dev);
1149        pci_restore_ats_state(dev);
1150        pci_restore_vc_state(dev);
1151
1152        pci_cleanup_aer_error_status_regs(dev);
1153
1154        pci_restore_config_space(dev);
1155
1156        pci_restore_pcix_state(dev);
1157        pci_restore_msi_state(dev);
1158
1159        /* Restore ACS and IOV configuration state */
1160        pci_enable_acs(dev);
1161        pci_restore_iov_state(dev);
1162
1163        dev->state_saved = false;
1164}
1165EXPORT_SYMBOL(pci_restore_state);
1166
1167struct pci_saved_state {
1168        u32 config_space[16];
1169        struct pci_cap_saved_data cap[0];
1170};
1171
1172/**
1173 * pci_store_saved_state - Allocate and return an opaque struct containing
1174 *                         the device saved state.
1175 * @dev: PCI device that we're dealing with
1176 *
1177 * Return NULL if no state or error.
1178 */
1179struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1180{
1181        struct pci_saved_state *state;
1182        struct pci_cap_saved_state *tmp;
1183        struct pci_cap_saved_data *cap;
1184        size_t size;
1185
1186        if (!dev->state_saved)
1187                return NULL;
1188
1189        size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1190
1191        hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1192                size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1193
1194        state = kzalloc(size, GFP_KERNEL);
1195        if (!state)
1196                return NULL;
1197
1198        memcpy(state->config_space, dev->saved_config_space,
1199               sizeof(state->config_space));
1200
1201        cap = state->cap;
1202        hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1203                size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1204                memcpy(cap, &tmp->cap, len);
1205                cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1206        }
1207        /* Empty cap_save terminates list */
1208
1209        return state;
1210}
1211EXPORT_SYMBOL_GPL(pci_store_saved_state);
1212
1213/**
1214 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1215 * @dev: PCI device that we're dealing with
1216 * @state: Saved state returned from pci_store_saved_state()
1217 */
1218int pci_load_saved_state(struct pci_dev *dev,
1219                         struct pci_saved_state *state)
1220{
1221        struct pci_cap_saved_data *cap;
1222
1223        dev->state_saved = false;
1224
1225        if (!state)
1226                return 0;
1227
1228        memcpy(dev->saved_config_space, state->config_space,
1229               sizeof(state->config_space));
1230
1231        cap = state->cap;
1232        while (cap->size) {
1233                struct pci_cap_saved_state *tmp;
1234
1235                tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1236                if (!tmp || tmp->cap.size != cap->size)
1237                        return -EINVAL;
1238
1239                memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1240                cap = (struct pci_cap_saved_data *)((u8 *)cap +
1241                       sizeof(struct pci_cap_saved_data) + cap->size);
1242        }
1243
1244        dev->state_saved = true;
1245        return 0;
1246}
1247EXPORT_SYMBOL_GPL(pci_load_saved_state);
1248
1249/**
1250 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1251 *                                 and free the memory allocated for it.
1252 * @dev: PCI device that we're dealing with
1253 * @state: Pointer to saved state returned from pci_store_saved_state()
1254 */
1255int pci_load_and_free_saved_state(struct pci_dev *dev,
1256                                  struct pci_saved_state **state)
1257{
1258        int ret = pci_load_saved_state(dev, *state);
1259        kfree(*state);
1260        *state = NULL;
1261        return ret;
1262}
1263EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1264
1265int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1266{
1267        return pci_enable_resources(dev, bars);
1268}
1269
1270static int do_pci_enable_device(struct pci_dev *dev, int bars)
1271{
1272        int err;
1273        struct pci_dev *bridge;
1274        u16 cmd;
1275        u8 pin;
1276
1277        err = pci_set_power_state(dev, PCI_D0);
1278        if (err < 0 && err != -EIO)
1279                return err;
1280
1281        bridge = pci_upstream_bridge(dev);
1282        if (bridge)
1283                pcie_aspm_powersave_config_link(bridge);
1284
1285        err = pcibios_enable_device(dev, bars);
1286        if (err < 0)
1287                return err;
1288        pci_fixup_device(pci_fixup_enable, dev);
1289
1290        if (dev->msi_enabled || dev->msix_enabled)
1291                return 0;
1292
1293        pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1294        if (pin) {
1295                pci_read_config_word(dev, PCI_COMMAND, &cmd);
1296                if (cmd & PCI_COMMAND_INTX_DISABLE)
1297                        pci_write_config_word(dev, PCI_COMMAND,
1298                                              cmd & ~PCI_COMMAND_INTX_DISABLE);
1299        }
1300
1301        return 0;
1302}
1303
1304/**
1305 * pci_reenable_device - Resume abandoned device
1306 * @dev: PCI device to be resumed
1307 *
1308 *  Note this function is a backend of pci_default_resume and is not supposed
1309 *  to be called by normal code, write proper resume handler and use it instead.
1310 */
1311int pci_reenable_device(struct pci_dev *dev)
1312{
1313        if (pci_is_enabled(dev))
1314                return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1315        return 0;
1316}
1317EXPORT_SYMBOL(pci_reenable_device);
1318
1319static void pci_enable_bridge(struct pci_dev *dev)
1320{
1321        struct pci_dev *bridge;
1322        int retval;
1323
1324        bridge = pci_upstream_bridge(dev);
1325        if (bridge)
1326                pci_enable_bridge(bridge);
1327
1328        if (pci_is_enabled(dev)) {
1329                if (!dev->is_busmaster)
1330                        pci_set_master(dev);
1331                return;
1332        }
1333
1334        retval = pci_enable_device(dev);
1335        if (retval)
1336                dev_err(&dev->dev, "Error enabling bridge (%d), continuing\n",
1337                        retval);
1338        pci_set_master(dev);
1339}
1340
1341static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1342{
1343        struct pci_dev *bridge;
1344        int err;
1345        int i, bars = 0;
1346
1347        /*
1348         * Power state could be unknown at this point, either due to a fresh
1349         * boot or a device removal call.  So get the current power state
1350         * so that things like MSI message writing will behave as expected
1351         * (e.g. if the device really is in D0 at enable time).
1352         */
1353        if (dev->pm_cap) {
1354                u16 pmcsr;
1355                pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1356                dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1357        }
1358
1359        if (atomic_inc_return(&dev->enable_cnt) > 1)
1360                return 0;               /* already enabled */
1361
1362        bridge = pci_upstream_bridge(dev);
1363        if (bridge)
1364                pci_enable_bridge(bridge);
1365
1366        /* only skip sriov related */
1367        for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1368                if (dev->resource[i].flags & flags)
1369                        bars |= (1 << i);
1370        for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1371                if (dev->resource[i].flags & flags)
1372                        bars |= (1 << i);
1373
1374        err = do_pci_enable_device(dev, bars);
1375        if (err < 0)
1376                atomic_dec(&dev->enable_cnt);
1377        return err;
1378}
1379
1380/**
1381 * pci_enable_device_io - Initialize a device for use with IO space
1382 * @dev: PCI device to be initialized
1383 *
1384 *  Initialize device before it's used by a driver. Ask low-level code
1385 *  to enable I/O resources. Wake up the device if it was suspended.
1386 *  Beware, this function can fail.
1387 */
1388int pci_enable_device_io(struct pci_dev *dev)
1389{
1390        return pci_enable_device_flags(dev, IORESOURCE_IO);
1391}
1392EXPORT_SYMBOL(pci_enable_device_io);
1393
1394/**
1395 * pci_enable_device_mem - Initialize a device for use with Memory space
1396 * @dev: PCI device to be initialized
1397 *
1398 *  Initialize device before it's used by a driver. Ask low-level code
1399 *  to enable Memory resources. Wake up the device if it was suspended.
1400 *  Beware, this function can fail.
1401 */
1402int pci_enable_device_mem(struct pci_dev *dev)
1403{
1404        return pci_enable_device_flags(dev, IORESOURCE_MEM);
1405}
1406EXPORT_SYMBOL(pci_enable_device_mem);
1407
1408/**
1409 * pci_enable_device - Initialize device before it's used by a driver.
1410 * @dev: PCI device to be initialized
1411 *
1412 *  Initialize device before it's used by a driver. Ask low-level code
1413 *  to enable I/O and memory. Wake up the device if it was suspended.
1414 *  Beware, this function can fail.
1415 *
1416 *  Note we don't actually enable the device many times if we call
1417 *  this function repeatedly (we just increment the count).
1418 */
1419int pci_enable_device(struct pci_dev *dev)
1420{
1421        return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1422}
1423EXPORT_SYMBOL(pci_enable_device);
1424
1425/*
1426 * Managed PCI resources.  This manages device on/off, intx/msi/msix
1427 * on/off and BAR regions.  pci_dev itself records msi/msix status, so
1428 * there's no need to track it separately.  pci_devres is initialized
1429 * when a device is enabled using managed PCI device enable interface.
1430 */
1431struct pci_devres {
1432        unsigned int enabled:1;
1433        unsigned int pinned:1;
1434        unsigned int orig_intx:1;
1435        unsigned int restore_intx:1;
1436        u32 region_mask;
1437};
1438
1439static void pcim_release(struct device *gendev, void *res)
1440{
1441        struct pci_dev *dev = to_pci_dev(gendev);
1442        struct pci_devres *this = res;
1443        int i;
1444
1445        if (dev->msi_enabled)
1446                pci_disable_msi(dev);
1447        if (dev->msix_enabled)
1448                pci_disable_msix(dev);
1449
1450        for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1451                if (this->region_mask & (1 << i))
1452                        pci_release_region(dev, i);
1453
1454        if (this->restore_intx)
1455                pci_intx(dev, this->orig_intx);
1456
1457        if (this->enabled && !this->pinned)
1458                pci_disable_device(dev);
1459}
1460
1461static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1462{
1463        struct pci_devres *dr, *new_dr;
1464
1465        dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1466        if (dr)
1467                return dr;
1468
1469        new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1470        if (!new_dr)
1471                return NULL;
1472        return devres_get(&pdev->dev, new_dr, NULL, NULL);
1473}
1474
1475static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1476{
1477        if (pci_is_managed(pdev))
1478                return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1479        return NULL;
1480}
1481
1482/**
1483 * pcim_enable_device - Managed pci_enable_device()
1484 * @pdev: PCI device to be initialized
1485 *
1486 * Managed pci_enable_device().
1487 */
1488int pcim_enable_device(struct pci_dev *pdev)
1489{
1490        struct pci_devres *dr;
1491        int rc;
1492
1493        dr = get_pci_dr(pdev);
1494        if (unlikely(!dr))
1495                return -ENOMEM;
1496        if (dr->enabled)
1497                return 0;
1498
1499        rc = pci_enable_device(pdev);
1500        if (!rc) {
1501                pdev->is_managed = 1;
1502                dr->enabled = 1;
1503        }
1504        return rc;
1505}
1506EXPORT_SYMBOL(pcim_enable_device);
1507
1508/**
1509 * pcim_pin_device - Pin managed PCI device
1510 * @pdev: PCI device to pin
1511 *
1512 * Pin managed PCI device @pdev.  Pinned device won't be disabled on
1513 * driver detach.  @pdev must have been enabled with
1514 * pcim_enable_device().
1515 */
1516void pcim_pin_device(struct pci_dev *pdev)
1517{
1518        struct pci_devres *dr;
1519
1520        dr = find_pci_dr(pdev);
1521        WARN_ON(!dr || !dr->enabled);
1522        if (dr)
1523                dr->pinned = 1;
1524}
1525EXPORT_SYMBOL(pcim_pin_device);
1526
1527/*
1528 * pcibios_add_device - provide arch specific hooks when adding device dev
1529 * @dev: the PCI device being added
1530 *
1531 * Permits the platform to provide architecture specific functionality when
1532 * devices are added. This is the default implementation. Architecture
1533 * implementations can override this.
1534 */
1535int __weak pcibios_add_device(struct pci_dev *dev)
1536{
1537        return 0;
1538}
1539
1540/**
1541 * pcibios_release_device - provide arch specific hooks when releasing device dev
1542 * @dev: the PCI device being released
1543 *
1544 * Permits the platform to provide architecture specific functionality when
1545 * devices are released. This is the default implementation. Architecture
1546 * implementations can override this.
1547 */
1548void __weak pcibios_release_device(struct pci_dev *dev) {}
1549
1550/**
1551 * pcibios_disable_device - disable arch specific PCI resources for device dev
1552 * @dev: the PCI device to disable
1553 *
1554 * Disables architecture specific PCI resources for the device. This
1555 * is the default implementation. Architecture implementations can
1556 * override this.
1557 */
1558void __weak pcibios_disable_device(struct pci_dev *dev) {}
1559
1560/**
1561 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1562 * @irq: ISA IRQ to penalize
1563 * @active: IRQ active or not
1564 *
1565 * Permits the platform to provide architecture-specific functionality when
1566 * penalizing ISA IRQs. This is the default implementation. Architecture
1567 * implementations can override this.
1568 */
1569void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1570
1571static void do_pci_disable_device(struct pci_dev *dev)
1572{
1573        u16 pci_command;
1574
1575        pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1576        if (pci_command & PCI_COMMAND_MASTER) {
1577                pci_command &= ~PCI_COMMAND_MASTER;
1578                pci_write_config_word(dev, PCI_COMMAND, pci_command);
1579        }
1580
1581        pcibios_disable_device(dev);
1582}
1583
1584/**
1585 * pci_disable_enabled_device - Disable device without updating enable_cnt
1586 * @dev: PCI device to disable
1587 *
1588 * NOTE: This function is a backend of PCI power management routines and is
1589 * not supposed to be called drivers.
1590 */
1591void pci_disable_enabled_device(struct pci_dev *dev)
1592{
1593        if (pci_is_enabled(dev))
1594                do_pci_disable_device(dev);
1595}
1596
1597/**
1598 * pci_disable_device - Disable PCI device after use
1599 * @dev: PCI device to be disabled
1600 *
1601 * Signal to the system that the PCI device is not in use by the system
1602 * anymore.  This only involves disabling PCI bus-mastering, if active.
1603 *
1604 * Note we don't actually disable the device until all callers of
1605 * pci_enable_device() have called pci_disable_device().
1606 */
1607void pci_disable_device(struct pci_dev *dev)
1608{
1609        struct pci_devres *dr;
1610
1611        dr = find_pci_dr(dev);
1612        if (dr)
1613                dr->enabled = 0;
1614
1615        dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1616                      "disabling already-disabled device");
1617
1618        if (atomic_dec_return(&dev->enable_cnt) != 0)
1619                return;
1620
1621        do_pci_disable_device(dev);
1622
1623        dev->is_busmaster = 0;
1624}
1625EXPORT_SYMBOL(pci_disable_device);
1626
1627/**
1628 * pcibios_set_pcie_reset_state - set reset state for device dev
1629 * @dev: the PCIe device reset
1630 * @state: Reset state to enter into
1631 *
1632 *
1633 * Sets the PCIe reset state for the device. This is the default
1634 * implementation. Architecture implementations can override this.
1635 */
1636int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1637                                        enum pcie_reset_state state)
1638{
1639        return -EINVAL;
1640}
1641
1642/**
1643 * pci_set_pcie_reset_state - set reset state for device dev
1644 * @dev: the PCIe device reset
1645 * @state: Reset state to enter into
1646 *
1647 *
1648 * Sets the PCI reset state for the device.
1649 */
1650int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1651{
1652        return pcibios_set_pcie_reset_state(dev, state);
1653}
1654EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1655
1656/**
1657 * pci_check_pme_status - Check if given device has generated PME.
1658 * @dev: Device to check.
1659 *
1660 * Check the PME status of the device and if set, clear it and clear PME enable
1661 * (if set).  Return 'true' if PME status and PME enable were both set or
1662 * 'false' otherwise.
1663 */
1664bool pci_check_pme_status(struct pci_dev *dev)
1665{
1666        int pmcsr_pos;
1667        u16 pmcsr;
1668        bool ret = false;
1669
1670        if (!dev->pm_cap)
1671                return false;
1672
1673        pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1674        pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1675        if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1676                return false;
1677
1678        /* Clear PME status. */
1679        pmcsr |= PCI_PM_CTRL_PME_STATUS;
1680        if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1681                /* Disable PME to avoid interrupt flood. */
1682                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1683                ret = true;
1684        }
1685
1686        pci_write_config_word(dev, pmcsr_pos, pmcsr);
1687
1688        return ret;
1689}
1690
1691/**
1692 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1693 * @dev: Device to handle.
1694 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1695 *
1696 * Check if @dev has generated PME and queue a resume request for it in that
1697 * case.
1698 */
1699static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1700{
1701        if (pme_poll_reset && dev->pme_poll)
1702                dev->pme_poll = false;
1703
1704        if (pci_check_pme_status(dev)) {
1705                pci_wakeup_event(dev);
1706                pm_request_resume(&dev->dev);
1707        }
1708        return 0;
1709}
1710
1711/**
1712 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1713 * @bus: Top bus of the subtree to walk.
1714 */
1715void pci_pme_wakeup_bus(struct pci_bus *bus)
1716{
1717        if (bus)
1718                pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1719}
1720
1721
1722/**
1723 * pci_pme_capable - check the capability of PCI device to generate PME#
1724 * @dev: PCI device to handle.
1725 * @state: PCI state from which device will issue PME#.
1726 */
1727bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1728{
1729        if (!dev->pm_cap)
1730                return false;
1731
1732        return !!(dev->pme_support & (1 << state));
1733}
1734EXPORT_SYMBOL(pci_pme_capable);
1735
1736static void pci_pme_list_scan(struct work_struct *work)
1737{
1738        struct pci_pme_device *pme_dev, *n;
1739
1740        mutex_lock(&pci_pme_list_mutex);
1741        list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1742                if (pme_dev->dev->pme_poll) {
1743                        struct pci_dev *bridge;
1744
1745                        bridge = pme_dev->dev->bus->self;
1746                        /*
1747                         * If bridge is in low power state, the
1748                         * configuration space of subordinate devices
1749                         * may be not accessible
1750                         */
1751                        if (bridge && bridge->current_state != PCI_D0)
1752                                continue;
1753                        pci_pme_wakeup(pme_dev->dev, NULL);
1754                } else {
1755                        list_del(&pme_dev->list);
1756                        kfree(pme_dev);
1757                }
1758        }
1759        if (!list_empty(&pci_pme_list))
1760                schedule_delayed_work(&pci_pme_work,
1761                                      msecs_to_jiffies(PME_TIMEOUT));
1762        mutex_unlock(&pci_pme_list_mutex);
1763}
1764
1765static void __pci_pme_active(struct pci_dev *dev, bool enable)
1766{
1767        u16 pmcsr;
1768
1769        if (!dev->pme_support)
1770                return;
1771
1772        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1773        /* Clear PME_Status by writing 1 to it and enable PME# */
1774        pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1775        if (!enable)
1776                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1777
1778        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1779}
1780
1781/**
1782 * pci_pme_active - enable or disable PCI device's PME# function
1783 * @dev: PCI device to handle.
1784 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1785 *
1786 * The caller must verify that the device is capable of generating PME# before
1787 * calling this function with @enable equal to 'true'.
1788 */
1789void pci_pme_active(struct pci_dev *dev, bool enable)
1790{
1791        __pci_pme_active(dev, enable);
1792
1793        /*
1794         * PCI (as opposed to PCIe) PME requires that the device have
1795         * its PME# line hooked up correctly. Not all hardware vendors
1796         * do this, so the PME never gets delivered and the device
1797         * remains asleep. The easiest way around this is to
1798         * periodically walk the list of suspended devices and check
1799         * whether any have their PME flag set. The assumption is that
1800         * we'll wake up often enough anyway that this won't be a huge
1801         * hit, and the power savings from the devices will still be a
1802         * win.
1803         *
1804         * Although PCIe uses in-band PME message instead of PME# line
1805         * to report PME, PME does not work for some PCIe devices in
1806         * reality.  For example, there are devices that set their PME
1807         * status bits, but don't really bother to send a PME message;
1808         * there are PCI Express Root Ports that don't bother to
1809         * trigger interrupts when they receive PME messages from the
1810         * devices below.  So PME poll is used for PCIe devices too.
1811         */
1812
1813        if (dev->pme_poll) {
1814                struct pci_pme_device *pme_dev;
1815                if (enable) {
1816                        pme_dev = kmalloc(sizeof(struct pci_pme_device),
1817                                          GFP_KERNEL);
1818                        if (!pme_dev) {
1819                                dev_warn(&dev->dev, "can't enable PME#\n");
1820                                return;
1821                        }
1822                        pme_dev->dev = dev;
1823                        mutex_lock(&pci_pme_list_mutex);
1824                        list_add(&pme_dev->list, &pci_pme_list);
1825                        if (list_is_singular(&pci_pme_list))
1826                                schedule_delayed_work(&pci_pme_work,
1827                                                      msecs_to_jiffies(PME_TIMEOUT));
1828                        mutex_unlock(&pci_pme_list_mutex);
1829                } else {
1830                        mutex_lock(&pci_pme_list_mutex);
1831                        list_for_each_entry(pme_dev, &pci_pme_list, list) {
1832                                if (pme_dev->dev == dev) {
1833                                        list_del(&pme_dev->list);
1834                                        kfree(pme_dev);
1835                                        break;
1836                                }
1837                        }
1838                        mutex_unlock(&pci_pme_list_mutex);
1839                }
1840        }
1841
1842        dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1843}
1844EXPORT_SYMBOL(pci_pme_active);
1845
1846/**
1847 * __pci_enable_wake - enable PCI device as wakeup event source
1848 * @dev: PCI device affected
1849 * @state: PCI state from which device will issue wakeup events
1850 * @runtime: True if the events are to be generated at run time
1851 * @enable: True to enable event generation; false to disable
1852 *
1853 * This enables the device as a wakeup event source, or disables it.
1854 * When such events involves platform-specific hooks, those hooks are
1855 * called automatically by this routine.
1856 *
1857 * Devices with legacy power management (no standard PCI PM capabilities)
1858 * always require such platform hooks.
1859 *
1860 * RETURN VALUE:
1861 * 0 is returned on success
1862 * -EINVAL is returned if device is not supposed to wake up the system
1863 * Error code depending on the platform is returned if both the platform and
1864 * the native mechanism fail to enable the generation of wake-up events
1865 */
1866int __pci_enable_wake(struct pci_dev *dev, pci_power_t state,
1867                      bool runtime, bool enable)
1868{
1869        int ret = 0;
1870
1871        if (enable && !runtime && !device_may_wakeup(&dev->dev))
1872                return -EINVAL;
1873
1874        /* Don't do the same thing twice in a row for one device. */
1875        if (!!enable == !!dev->wakeup_prepared)
1876                return 0;
1877
1878        /*
1879         * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1880         * Anderson we should be doing PME# wake enable followed by ACPI wake
1881         * enable.  To disable wake-up we call the platform first, for symmetry.
1882         */
1883
1884        if (enable) {
1885                int error;
1886
1887                if (pci_pme_capable(dev, state))
1888                        pci_pme_active(dev, true);
1889                else
1890                        ret = 1;
1891                error = runtime ? platform_pci_run_wake(dev, true) :
1892                                        platform_pci_sleep_wake(dev, true);
1893                if (ret)
1894                        ret = error;
1895                if (!ret)
1896                        dev->wakeup_prepared = true;
1897        } else {
1898                if (runtime)
1899                        platform_pci_run_wake(dev, false);
1900                else
1901                        platform_pci_sleep_wake(dev, false);
1902                pci_pme_active(dev, false);
1903                dev->wakeup_prepared = false;
1904        }
1905
1906        return ret;
1907}
1908EXPORT_SYMBOL(__pci_enable_wake);
1909
1910/**
1911 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1912 * @dev: PCI device to prepare
1913 * @enable: True to enable wake-up event generation; false to disable
1914 *
1915 * Many drivers want the device to wake up the system from D3_hot or D3_cold
1916 * and this function allows them to set that up cleanly - pci_enable_wake()
1917 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1918 * ordering constraints.
1919 *
1920 * This function only returns error code if the device is not capable of
1921 * generating PME# from both D3_hot and D3_cold, and the platform is unable to
1922 * enable wake-up power for it.
1923 */
1924int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1925{
1926        return pci_pme_capable(dev, PCI_D3cold) ?
1927                        pci_enable_wake(dev, PCI_D3cold, enable) :
1928                        pci_enable_wake(dev, PCI_D3hot, enable);
1929}
1930EXPORT_SYMBOL(pci_wake_from_d3);
1931
1932/**
1933 * pci_target_state - find an appropriate low power state for a given PCI dev
1934 * @dev: PCI device
1935 *
1936 * Use underlying platform code to find a supported low power state for @dev.
1937 * If the platform can't manage @dev, return the deepest state from which it
1938 * can generate wake events, based on any available PME info.
1939 */
1940static pci_power_t pci_target_state(struct pci_dev *dev)
1941{
1942        pci_power_t target_state = PCI_D3hot;
1943
1944        if (platform_pci_power_manageable(dev)) {
1945                /*
1946                 * Call the platform to choose the target state of the device
1947                 * and enable wake-up from this state if supported.
1948                 */
1949                pci_power_t state = platform_pci_choose_state(dev);
1950
1951                switch (state) {
1952                case PCI_POWER_ERROR:
1953                case PCI_UNKNOWN:
1954                        break;
1955                case PCI_D1:
1956                case PCI_D2:
1957                        if (pci_no_d1d2(dev))
1958                                break;
1959                default:
1960                        target_state = state;
1961                }
1962        } else if (!dev->pm_cap) {
1963                target_state = PCI_D0;
1964        } else if (device_may_wakeup(&dev->dev)) {
1965                /*
1966                 * Find the deepest state from which the device can generate
1967                 * wake-up events, make it the target state and enable device
1968                 * to generate PME#.
1969                 */
1970                if (dev->pme_support) {
1971                        while (target_state
1972                              && !(dev->pme_support & (1 << target_state)))
1973                                target_state--;
1974                }
1975        }
1976
1977        return target_state;
1978}
1979
1980/**
1981 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
1982 * @dev: Device to handle.
1983 *
1984 * Choose the power state appropriate for the device depending on whether
1985 * it can wake up the system and/or is power manageable by the platform
1986 * (PCI_D3hot is the default) and put the device into that state.
1987 */
1988int pci_prepare_to_sleep(struct pci_dev *dev)
1989{
1990        pci_power_t target_state = pci_target_state(dev);
1991        int error;
1992
1993        if (target_state == PCI_POWER_ERROR)
1994                return -EIO;
1995
1996        pci_enable_wake(dev, target_state, device_may_wakeup(&dev->dev));
1997
1998        error = pci_set_power_state(dev, target_state);
1999
2000        if (error)
2001                pci_enable_wake(dev, target_state, false);
2002
2003        return error;
2004}
2005EXPORT_SYMBOL(pci_prepare_to_sleep);
2006
2007/**
2008 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2009 * @dev: Device to handle.
2010 *
2011 * Disable device's system wake-up capability and put it into D0.
2012 */
2013int pci_back_from_sleep(struct pci_dev *dev)
2014{
2015        pci_enable_wake(dev, PCI_D0, false);
2016        return pci_set_power_state(dev, PCI_D0);
2017}
2018EXPORT_SYMBOL(pci_back_from_sleep);
2019
2020/**
2021 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2022 * @dev: PCI device being suspended.
2023 *
2024 * Prepare @dev to generate wake-up events at run time and put it into a low
2025 * power state.
2026 */
2027int pci_finish_runtime_suspend(struct pci_dev *dev)
2028{
2029        pci_power_t target_state = pci_target_state(dev);
2030        int error;
2031
2032        if (target_state == PCI_POWER_ERROR)
2033                return -EIO;
2034
2035        dev->runtime_d3cold = target_state == PCI_D3cold;
2036
2037        __pci_enable_wake(dev, target_state, true, pci_dev_run_wake(dev));
2038
2039        error = pci_set_power_state(dev, target_state);
2040
2041        if (error) {
2042                __pci_enable_wake(dev, target_state, true, false);
2043                dev->runtime_d3cold = false;
2044        }
2045
2046        return error;
2047}
2048
2049/**
2050 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2051 * @dev: Device to check.
2052 *
2053 * Return true if the device itself is capable of generating wake-up events
2054 * (through the platform or using the native PCIe PME) or if the device supports
2055 * PME and one of its upstream bridges can generate wake-up events.
2056 */
2057bool pci_dev_run_wake(struct pci_dev *dev)
2058{
2059        struct pci_bus *bus = dev->bus;
2060
2061        if (device_run_wake(&dev->dev))
2062                return true;
2063
2064        if (!dev->pme_support)
2065                return false;
2066
2067        while (bus->parent) {
2068                struct pci_dev *bridge = bus->self;
2069
2070                if (device_run_wake(&bridge->dev))
2071                        return true;
2072
2073                bus = bus->parent;
2074        }
2075
2076        /* We have reached the root bus. */
2077        if (bus->bridge)
2078                return device_run_wake(bus->bridge);
2079
2080        return false;
2081}
2082EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2083
2084/**
2085 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2086 * @pci_dev: Device to check.
2087 *
2088 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2089 * reconfigured due to wakeup settings difference between system and runtime
2090 * suspend and the current power state of it is suitable for the upcoming
2091 * (system) transition.
2092 *
2093 * If the device is not configured for system wakeup, disable PME for it before
2094 * returning 'true' to prevent it from waking up the system unnecessarily.
2095 */
2096bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2097{
2098        struct device *dev = &pci_dev->dev;
2099
2100        if (!pm_runtime_suspended(dev)
2101            || pci_target_state(pci_dev) != pci_dev->current_state
2102            || platform_pci_need_resume(pci_dev))
2103                return false;
2104
2105        /*
2106         * At this point the device is good to go unless it's been configured
2107         * to generate PME at the runtime suspend time, but it is not supposed
2108         * to wake up the system.  In that case, simply disable PME for it
2109         * (it will have to be re-enabled on exit from system resume).
2110         *
2111         * If the device's power state is D3cold and the platform check above
2112         * hasn't triggered, the device's configuration is suitable and we don't
2113         * need to manipulate it at all.
2114         */
2115        spin_lock_irq(&dev->power.lock);
2116
2117        if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2118            !device_may_wakeup(dev))
2119                __pci_pme_active(pci_dev, false);
2120
2121        spin_unlock_irq(&dev->power.lock);
2122        return true;
2123}
2124
2125/**
2126 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2127 * @pci_dev: Device to handle.
2128 *
2129 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2130 * it might have been disabled during the prepare phase of system suspend if
2131 * the device was not configured for system wakeup.
2132 */
2133void pci_dev_complete_resume(struct pci_dev *pci_dev)
2134{
2135        struct device *dev = &pci_dev->dev;
2136
2137        if (!pci_dev_run_wake(pci_dev))
2138                return;
2139
2140        spin_lock_irq(&dev->power.lock);
2141
2142        if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2143                __pci_pme_active(pci_dev, true);
2144
2145        spin_unlock_irq(&dev->power.lock);
2146}
2147
2148void pci_config_pm_runtime_get(struct pci_dev *pdev)
2149{
2150        struct device *dev = &pdev->dev;
2151        struct device *parent = dev->parent;
2152
2153        if (parent)
2154                pm_runtime_get_sync(parent);
2155        pm_runtime_get_noresume(dev);
2156        /*
2157         * pdev->current_state is set to PCI_D3cold during suspending,
2158         * so wait until suspending completes
2159         */
2160        pm_runtime_barrier(dev);
2161        /*
2162         * Only need to resume devices in D3cold, because config
2163         * registers are still accessible for devices suspended but
2164         * not in D3cold.
2165         */
2166        if (pdev->current_state == PCI_D3cold)
2167                pm_runtime_resume(dev);
2168}
2169
2170void pci_config_pm_runtime_put(struct pci_dev *pdev)
2171{
2172        struct device *dev = &pdev->dev;
2173        struct device *parent = dev->parent;
2174
2175        pm_runtime_put(dev);
2176        if (parent)
2177                pm_runtime_put_sync(parent);
2178}
2179
2180/**
2181 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2182 * @bridge: Bridge to check
2183 *
2184 * This function checks if it is possible to move the bridge to D3.
2185 * Currently we only allow D3 for recent enough PCIe ports.
2186 */
2187static bool pci_bridge_d3_possible(struct pci_dev *bridge)
2188{
2189        unsigned int year;
2190
2191        if (!pci_is_pcie(bridge))
2192                return false;
2193
2194        switch (pci_pcie_type(bridge)) {
2195        case PCI_EXP_TYPE_ROOT_PORT:
2196        case PCI_EXP_TYPE_UPSTREAM:
2197        case PCI_EXP_TYPE_DOWNSTREAM:
2198                if (pci_bridge_d3_disable)
2199                        return false;
2200                if (pci_bridge_d3_force)
2201                        return true;
2202
2203                /*
2204                 * It should be safe to put PCIe ports from 2015 or newer
2205                 * to D3.
2206                 */
2207                if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
2208                    year >= 2015) {
2209                        return true;
2210                }
2211                break;
2212        }
2213
2214        return false;
2215}
2216
2217static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2218{
2219        bool *d3cold_ok = data;
2220        bool no_d3cold;
2221
2222        /*
2223         * The device needs to be allowed to go D3cold and if it is wake
2224         * capable to do so from D3cold.
2225         */
2226        no_d3cold = dev->no_d3cold || !dev->d3cold_allowed ||
2227                (device_may_wakeup(&dev->dev) && !pci_pme_capable(dev, PCI_D3cold)) ||
2228                !pci_power_manageable(dev);
2229
2230        *d3cold_ok = !no_d3cold;
2231
2232        return no_d3cold;
2233}
2234
2235/*
2236 * pci_bridge_d3_update - Update bridge D3 capabilities
2237 * @dev: PCI device which is changed
2238 * @remove: Is the device being removed
2239 *
2240 * Update upstream bridge PM capabilities accordingly depending on if the
2241 * device PM configuration was changed or the device is being removed.  The
2242 * change is also propagated upstream.
2243 */
2244static void pci_bridge_d3_update(struct pci_dev *dev, bool remove)
2245{
2246        struct pci_dev *bridge;
2247        bool d3cold_ok = true;
2248
2249        bridge = pci_upstream_bridge(dev);
2250        if (!bridge || !pci_bridge_d3_possible(bridge))
2251                return;
2252
2253        pci_dev_get(bridge);
2254        /*
2255         * If the device is removed we do not care about its D3cold
2256         * capabilities.
2257         */
2258        if (!remove)
2259                pci_dev_check_d3cold(dev, &d3cold_ok);
2260
2261        if (d3cold_ok) {
2262                /*
2263                 * We need to go through all children to find out if all of
2264                 * them can still go to D3cold.
2265                 */
2266                pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2267                             &d3cold_ok);
2268        }
2269
2270        if (bridge->bridge_d3 != d3cold_ok) {
2271                bridge->bridge_d3 = d3cold_ok;
2272                /* Propagate change to upstream bridges */
2273                pci_bridge_d3_update(bridge, false);
2274        }
2275
2276        pci_dev_put(bridge);
2277}
2278
2279/**
2280 * pci_bridge_d3_device_changed - Update bridge D3 capabilities on change
2281 * @dev: PCI device that was changed
2282 *
2283 * If a device is added or its PM configuration, such as is it allowed to
2284 * enter D3cold, is changed this function updates upstream bridge PM
2285 * capabilities accordingly.
2286 */
2287void pci_bridge_d3_device_changed(struct pci_dev *dev)
2288{
2289        pci_bridge_d3_update(dev, false);
2290}
2291
2292/**
2293 * pci_bridge_d3_device_removed - Update bridge D3 capabilities on remove
2294 * @dev: PCI device being removed
2295 *
2296 * Function updates upstream bridge PM capabilities based on other devices
2297 * still left on the bus.
2298 */
2299void pci_bridge_d3_device_removed(struct pci_dev *dev)
2300{
2301        pci_bridge_d3_update(dev, true);
2302}
2303
2304/**
2305 * pci_d3cold_enable - Enable D3cold for device
2306 * @dev: PCI device to handle
2307 *
2308 * This function can be used in drivers to enable D3cold from the device
2309 * they handle.  It also updates upstream PCI bridge PM capabilities
2310 * accordingly.
2311 */
2312void pci_d3cold_enable(struct pci_dev *dev)
2313{
2314        if (dev->no_d3cold) {
2315                dev->no_d3cold = false;
2316                pci_bridge_d3_device_changed(dev);
2317        }
2318}
2319EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2320
2321/**
2322 * pci_d3cold_disable - Disable D3cold for device
2323 * @dev: PCI device to handle
2324 *
2325 * This function can be used in drivers to disable D3cold from the device
2326 * they handle.  It also updates upstream PCI bridge PM capabilities
2327 * accordingly.
2328 */
2329void pci_d3cold_disable(struct pci_dev *dev)
2330{
2331        if (!dev->no_d3cold) {
2332                dev->no_d3cold = true;
2333                pci_bridge_d3_device_changed(dev);
2334        }
2335}
2336EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2337
2338/**
2339 * pci_pm_init - Initialize PM functions of given PCI device
2340 * @dev: PCI device to handle.
2341 */
2342void pci_pm_init(struct pci_dev *dev)
2343{
2344        int pm;
2345        u16 pmc;
2346
2347        pm_runtime_forbid(&dev->dev);
2348        pm_runtime_set_active(&dev->dev);
2349        pm_runtime_enable(&dev->dev);
2350        device_enable_async_suspend(&dev->dev);
2351        dev->wakeup_prepared = false;
2352
2353        dev->pm_cap = 0;
2354        dev->pme_support = 0;
2355
2356        /* find PCI PM capability in list */
2357        pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2358        if (!pm)
2359                return;
2360        /* Check device's ability to generate PME# */
2361        pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2362
2363        if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2364                dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
2365                        pmc & PCI_PM_CAP_VER_MASK);
2366                return;
2367        }
2368
2369        dev->pm_cap = pm;
2370        dev->d3_delay = PCI_PM_D3_WAIT;
2371        dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2372        dev->bridge_d3 = pci_bridge_d3_possible(dev);
2373        dev->d3cold_allowed = true;
2374
2375        dev->d1_support = false;
2376        dev->d2_support = false;
2377        if (!pci_no_d1d2(dev)) {
2378                if (pmc & PCI_PM_CAP_D1)
2379                        dev->d1_support = true;
2380                if (pmc & PCI_PM_CAP_D2)
2381                        dev->d2_support = true;
2382
2383                if (dev->d1_support || dev->d2_support)
2384                        dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
2385                                   dev->d1_support ? " D1" : "",
2386                                   dev->d2_support ? " D2" : "");
2387        }
2388
2389        pmc &= PCI_PM_CAP_PME_MASK;
2390        if (pmc) {
2391                dev_printk(KERN_DEBUG, &dev->dev,
2392                         "PME# supported from%s%s%s%s%s\n",
2393                         (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2394                         (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2395                         (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2396                         (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2397                         (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2398                dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2399                dev->pme_poll = true;
2400                /*
2401                 * Make device's PM flags reflect the wake-up capability, but
2402                 * let the user space enable it to wake up the system as needed.
2403                 */
2404                device_set_wakeup_capable(&dev->dev, true);
2405                /* Disable the PME# generation functionality */
2406                pci_pme_active(dev, false);
2407        }
2408}
2409
2410static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2411{
2412        unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2413
2414        switch (prop) {
2415        case PCI_EA_P_MEM:
2416        case PCI_EA_P_VF_MEM:
2417                flags |= IORESOURCE_MEM;
2418                break;
2419        case PCI_EA_P_MEM_PREFETCH:
2420        case PCI_EA_P_VF_MEM_PREFETCH:
2421                flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2422                break;
2423        case PCI_EA_P_IO:
2424                flags |= IORESOURCE_IO;
2425                break;
2426        default:
2427                return 0;
2428        }
2429
2430        return flags;
2431}
2432
2433static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2434                                            u8 prop)
2435{
2436        if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2437                return &dev->resource[bei];
2438#ifdef CONFIG_PCI_IOV
2439        else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2440                 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2441                return &dev->resource[PCI_IOV_RESOURCES +
2442                                      bei - PCI_EA_BEI_VF_BAR0];
2443#endif
2444        else if (bei == PCI_EA_BEI_ROM)
2445                return &dev->resource[PCI_ROM_RESOURCE];
2446        else
2447                return NULL;
2448}
2449
2450/* Read an Enhanced Allocation (EA) entry */
2451static int pci_ea_read(struct pci_dev *dev, int offset)
2452{
2453        struct resource *res;
2454        int ent_size, ent_offset = offset;
2455        resource_size_t start, end;
2456        unsigned long flags;
2457        u32 dw0, bei, base, max_offset;
2458        u8 prop;
2459        bool support_64 = (sizeof(resource_size_t) >= 8);
2460
2461        pci_read_config_dword(dev, ent_offset, &dw0);
2462        ent_offset += 4;
2463
2464        /* Entry size field indicates DWORDs after 1st */
2465        ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2466
2467        if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2468                goto out;
2469
2470        bei = (dw0 & PCI_EA_BEI) >> 4;
2471        prop = (dw0 & PCI_EA_PP) >> 8;
2472
2473        /*
2474         * If the Property is in the reserved range, try the Secondary
2475         * Property instead.
2476         */
2477        if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2478                prop = (dw0 & PCI_EA_SP) >> 16;
2479        if (prop > PCI_EA_P_BRIDGE_IO)
2480                goto out;
2481
2482        res = pci_ea_get_resource(dev, bei, prop);
2483        if (!res) {
2484                dev_err(&dev->dev, "Unsupported EA entry BEI: %u\n", bei);
2485                goto out;
2486        }
2487
2488        flags = pci_ea_flags(dev, prop);
2489        if (!flags) {
2490                dev_err(&dev->dev, "Unsupported EA properties: %#x\n", prop);
2491                goto out;
2492        }
2493
2494        /* Read Base */
2495        pci_read_config_dword(dev, ent_offset, &base);
2496        start = (base & PCI_EA_FIELD_MASK);
2497        ent_offset += 4;
2498
2499        /* Read MaxOffset */
2500        pci_read_config_dword(dev, ent_offset, &max_offset);
2501        ent_offset += 4;
2502
2503        /* Read Base MSBs (if 64-bit entry) */
2504        if (base & PCI_EA_IS_64) {
2505                u32 base_upper;
2506
2507                pci_read_config_dword(dev, ent_offset, &base_upper);
2508                ent_offset += 4;
2509
2510                flags |= IORESOURCE_MEM_64;
2511
2512                /* entry starts above 32-bit boundary, can't use */
2513                if (!support_64 && base_upper)
2514                        goto out;
2515
2516                if (support_64)
2517                        start |= ((u64)base_upper << 32);
2518        }
2519
2520        end = start + (max_offset | 0x03);
2521
2522        /* Read MaxOffset MSBs (if 64-bit entry) */
2523        if (max_offset & PCI_EA_IS_64) {
2524                u32 max_offset_upper;
2525
2526                pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2527                ent_offset += 4;
2528
2529                flags |= IORESOURCE_MEM_64;
2530
2531                /* entry too big, can't use */
2532                if (!support_64 && max_offset_upper)
2533                        goto out;
2534
2535                if (support_64)
2536                        end += ((u64)max_offset_upper << 32);
2537        }
2538
2539        if (end < start) {
2540                dev_err(&dev->dev, "EA Entry crosses address boundary\n");
2541                goto out;
2542        }
2543
2544        if (ent_size != ent_offset - offset) {
2545                dev_err(&dev->dev,
2546                        "EA Entry Size (%d) does not match length read (%d)\n",
2547                        ent_size, ent_offset - offset);
2548                goto out;
2549        }
2550
2551        res->name = pci_name(dev);
2552        res->start = start;
2553        res->end = end;
2554        res->flags = flags;
2555
2556        if (bei <= PCI_EA_BEI_BAR5)
2557                dev_printk(KERN_DEBUG, &dev->dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2558                           bei, res, prop);
2559        else if (bei == PCI_EA_BEI_ROM)
2560                dev_printk(KERN_DEBUG, &dev->dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2561                           res, prop);
2562        else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2563                dev_printk(KERN_DEBUG, &dev->dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2564                           bei - PCI_EA_BEI_VF_BAR0, res, prop);
2565        else
2566                dev_printk(KERN_DEBUG, &dev->dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2567                           bei, res, prop);
2568
2569out:
2570        return offset + ent_size;
2571}
2572
2573/* Enhanced Allocation Initialization */
2574void pci_ea_init(struct pci_dev *dev)
2575{
2576        int ea;
2577        u8 num_ent;
2578        int offset;
2579        int i;
2580
2581        /* find PCI EA capability in list */
2582        ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2583        if (!ea)
2584                return;
2585
2586        /* determine the number of entries */
2587        pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2588                                        &num_ent);
2589        num_ent &= PCI_EA_NUM_ENT_MASK;
2590
2591        offset = ea + PCI_EA_FIRST_ENT;
2592
2593        /* Skip DWORD 2 for type 1 functions */
2594        if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2595                offset += 4;
2596
2597        /* parse each EA entry */
2598        for (i = 0; i < num_ent; ++i)
2599                offset = pci_ea_read(dev, offset);
2600}
2601
2602static void pci_add_saved_cap(struct pci_dev *pci_dev,
2603        struct pci_cap_saved_state *new_cap)
2604{
2605        hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2606}
2607
2608/**
2609 * _pci_add_cap_save_buffer - allocate buffer for saving given
2610 *                            capability registers
2611 * @dev: the PCI device
2612 * @cap: the capability to allocate the buffer for
2613 * @extended: Standard or Extended capability ID
2614 * @size: requested size of the buffer
2615 */
2616static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2617                                    bool extended, unsigned int size)
2618{
2619        int pos;
2620        struct pci_cap_saved_state *save_state;
2621
2622        if (extended)
2623                pos = pci_find_ext_capability(dev, cap);
2624        else
2625                pos = pci_find_capability(dev, cap);
2626
2627        if (!pos)
2628                return 0;
2629
2630        save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2631        if (!save_state)
2632                return -ENOMEM;
2633
2634        save_state->cap.cap_nr = cap;
2635        save_state->cap.cap_extended = extended;
2636        save_state->cap.size = size;
2637        pci_add_saved_cap(dev, save_state);
2638
2639        return 0;
2640}
2641
2642int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2643{
2644        return _pci_add_cap_save_buffer(dev, cap, false, size);
2645}
2646
2647int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2648{
2649        return _pci_add_cap_save_buffer(dev, cap, true, size);
2650}
2651
2652/**
2653 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2654 * @dev: the PCI device
2655 */
2656void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2657{
2658        int error;
2659
2660        error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2661                                        PCI_EXP_SAVE_REGS * sizeof(u16));
2662        if (error)
2663                dev_err(&dev->dev,
2664                        "unable to preallocate PCI Express save buffer\n");
2665
2666        error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2667        if (error)
2668                dev_err(&dev->dev,
2669                        "unable to preallocate PCI-X save buffer\n");
2670
2671        pci_allocate_vc_save_buffers(dev);
2672}
2673
2674void pci_free_cap_save_buffers(struct pci_dev *dev)
2675{
2676        struct pci_cap_saved_state *tmp;
2677        struct hlist_node *n;
2678
2679        hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2680                kfree(tmp);
2681}
2682
2683/**
2684 * pci_configure_ari - enable or disable ARI forwarding
2685 * @dev: the PCI device
2686 *
2687 * If @dev and its upstream bridge both support ARI, enable ARI in the
2688 * bridge.  Otherwise, disable ARI in the bridge.
2689 */
2690void pci_configure_ari(struct pci_dev *dev)
2691{
2692        u32 cap;
2693        struct pci_dev *bridge;
2694
2695        if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2696                return;
2697
2698        bridge = dev->bus->self;
2699        if (!bridge)
2700                return;
2701
2702        pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2703        if (!(cap & PCI_EXP_DEVCAP2_ARI))
2704                return;
2705
2706        if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2707                pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2708                                         PCI_EXP_DEVCTL2_ARI);
2709                bridge->ari_enabled = 1;
2710        } else {
2711                pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2712                                           PCI_EXP_DEVCTL2_ARI);
2713                bridge->ari_enabled = 0;
2714        }
2715}
2716
2717static int pci_acs_enable;
2718
2719/**
2720 * pci_request_acs - ask for ACS to be enabled if supported
2721 */
2722void pci_request_acs(void)
2723{
2724        pci_acs_enable = 1;
2725}
2726
2727/**
2728 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2729 * @dev: the PCI device
2730 */
2731static void pci_std_enable_acs(struct pci_dev *dev)
2732{
2733        int pos;
2734        u16 cap;
2735        u16 ctrl;
2736
2737        pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2738        if (!pos)
2739                return;
2740
2741        pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2742        pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2743
2744        /* Source Validation */
2745        ctrl |= (cap & PCI_ACS_SV);
2746
2747        /* P2P Request Redirect */
2748        ctrl |= (cap & PCI_ACS_RR);
2749
2750        /* P2P Completion Redirect */
2751        ctrl |= (cap & PCI_ACS_CR);
2752
2753        /* Upstream Forwarding */
2754        ctrl |= (cap & PCI_ACS_UF);
2755
2756        pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2757}
2758
2759/**
2760 * pci_enable_acs - enable ACS if hardware support it
2761 * @dev: the PCI device
2762 */
2763void pci_enable_acs(struct pci_dev *dev)
2764{
2765        if (!pci_acs_enable)
2766                return;
2767
2768        if (!pci_dev_specific_enable_acs(dev))
2769                return;
2770
2771        pci_std_enable_acs(dev);
2772}
2773
2774static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2775{
2776        int pos;
2777        u16 cap, ctrl;
2778
2779        pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2780        if (!pos)
2781                return false;
2782
2783        /*
2784         * Except for egress control, capabilities are either required
2785         * or only required if controllable.  Features missing from the
2786         * capability field can therefore be assumed as hard-wired enabled.
2787         */
2788        pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2789        acs_flags &= (cap | PCI_ACS_EC);
2790
2791        pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2792        return (ctrl & acs_flags) == acs_flags;
2793}
2794
2795/**
2796 * pci_acs_enabled - test ACS against required flags for a given device
2797 * @pdev: device to test
2798 * @acs_flags: required PCI ACS flags
2799 *
2800 * Return true if the device supports the provided flags.  Automatically
2801 * filters out flags that are not implemented on multifunction devices.
2802 *
2803 * Note that this interface checks the effective ACS capabilities of the
2804 * device rather than the actual capabilities.  For instance, most single
2805 * function endpoints are not required to support ACS because they have no
2806 * opportunity for peer-to-peer access.  We therefore return 'true'
2807 * regardless of whether the device exposes an ACS capability.  This makes
2808 * it much easier for callers of this function to ignore the actual type
2809 * or topology of the device when testing ACS support.
2810 */
2811bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2812{
2813        int ret;
2814
2815        ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2816        if (ret >= 0)
2817                return ret > 0;
2818
2819        /*
2820         * Conventional PCI and PCI-X devices never support ACS, either
2821         * effectively or actually.  The shared bus topology implies that
2822         * any device on the bus can receive or snoop DMA.
2823         */
2824        if (!pci_is_pcie(pdev))
2825                return false;
2826
2827        switch (pci_pcie_type(pdev)) {
2828        /*
2829         * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2830         * but since their primary interface is PCI/X, we conservatively
2831         * handle them as we would a non-PCIe device.
2832         */
2833        case PCI_EXP_TYPE_PCIE_BRIDGE:
2834        /*
2835         * PCIe 3.0, 6.12.1 excludes ACS on these devices.  "ACS is never
2836         * applicable... must never implement an ACS Extended Capability...".
2837         * This seems arbitrary, but we take a conservative interpretation
2838         * of this statement.
2839         */
2840        case PCI_EXP_TYPE_PCI_BRIDGE:
2841        case PCI_EXP_TYPE_RC_EC:
2842                return false;
2843        /*
2844         * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2845         * implement ACS in order to indicate their peer-to-peer capabilities,
2846         * regardless of whether they are single- or multi-function devices.
2847         */
2848        case PCI_EXP_TYPE_DOWNSTREAM:
2849        case PCI_EXP_TYPE_ROOT_PORT:
2850                return pci_acs_flags_enabled(pdev, acs_flags);
2851        /*
2852         * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2853         * implemented by the remaining PCIe types to indicate peer-to-peer
2854         * capabilities, but only when they are part of a multifunction
2855         * device.  The footnote for section 6.12 indicates the specific
2856         * PCIe types included here.
2857         */
2858        case PCI_EXP_TYPE_ENDPOINT:
2859        case PCI_EXP_TYPE_UPSTREAM:
2860        case PCI_EXP_TYPE_LEG_END:
2861        case PCI_EXP_TYPE_RC_END:
2862                if (!pdev->multifunction)
2863                        break;
2864
2865                return pci_acs_flags_enabled(pdev, acs_flags);
2866        }
2867
2868        /*
2869         * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2870         * to single function devices with the exception of downstream ports.
2871         */
2872        return true;
2873}
2874
2875/**
2876 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2877 * @start: starting downstream device
2878 * @end: ending upstream device or NULL to search to the root bus
2879 * @acs_flags: required flags
2880 *
2881 * Walk up a device tree from start to end testing PCI ACS support.  If
2882 * any step along the way does not support the required flags, return false.
2883 */
2884bool pci_acs_path_enabled(struct pci_dev *start,
2885                          struct pci_dev *end, u16 acs_flags)
2886{
2887        struct pci_dev *pdev, *parent = start;
2888
2889        do {
2890                pdev = parent;
2891
2892                if (!pci_acs_enabled(pdev, acs_flags))
2893                        return false;
2894
2895                if (pci_is_root_bus(pdev->bus))
2896                        return (end == NULL);
2897
2898                parent = pdev->bus->self;
2899        } while (pdev != end);
2900
2901        return true;
2902}
2903
2904/**
2905 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2906 * @dev: the PCI device
2907 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
2908 *
2909 * Perform INTx swizzling for a device behind one level of bridge.  This is
2910 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2911 * behind bridges on add-in cards.  For devices with ARI enabled, the slot
2912 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2913 * the PCI Express Base Specification, Revision 2.1)
2914 */
2915u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2916{
2917        int slot;
2918
2919        if (pci_ari_enabled(dev->bus))
2920                slot = 0;
2921        else
2922                slot = PCI_SLOT(dev->devfn);
2923
2924        return (((pin - 1) + slot) % 4) + 1;
2925}
2926
2927int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
2928{
2929        u8 pin;
2930
2931        pin = dev->pin;
2932        if (!pin)
2933                return -1;
2934
2935        while (!pci_is_root_bus(dev->bus)) {
2936                pin = pci_swizzle_interrupt_pin(dev, pin);
2937                dev = dev->bus->self;
2938        }
2939        *bridge = dev;
2940        return pin;
2941}
2942
2943/**
2944 * pci_common_swizzle - swizzle INTx all the way to root bridge
2945 * @dev: the PCI device
2946 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2947 *
2948 * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
2949 * bridges all the way up to a PCI root bus.
2950 */
2951u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
2952{
2953        u8 pin = *pinp;
2954
2955        while (!pci_is_root_bus(dev->bus)) {
2956                pin = pci_swizzle_interrupt_pin(dev, pin);
2957                dev = dev->bus->self;
2958        }
2959        *pinp = pin;
2960        return PCI_SLOT(dev->devfn);
2961}
2962EXPORT_SYMBOL_GPL(pci_common_swizzle);
2963
2964/**
2965 *      pci_release_region - Release a PCI bar
2966 *      @pdev: PCI device whose resources were previously reserved by pci_request_region
2967 *      @bar: BAR to release
2968 *
2969 *      Releases the PCI I/O and memory resources previously reserved by a
2970 *      successful call to pci_request_region.  Call this function only
2971 *      after all use of the PCI regions has ceased.
2972 */
2973void pci_release_region(struct pci_dev *pdev, int bar)
2974{
2975        struct pci_devres *dr;
2976
2977        if (pci_resource_len(pdev, bar) == 0)
2978                return;
2979        if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
2980                release_region(pci_resource_start(pdev, bar),
2981                                pci_resource_len(pdev, bar));
2982        else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
2983                release_mem_region(pci_resource_start(pdev, bar),
2984                                pci_resource_len(pdev, bar));
2985
2986        dr = find_pci_dr(pdev);
2987        if (dr)
2988                dr->region_mask &= ~(1 << bar);
2989}
2990EXPORT_SYMBOL(pci_release_region);
2991
2992/**
2993 *      __pci_request_region - Reserved PCI I/O and memory resource
2994 *      @pdev: PCI device whose resources are to be reserved
2995 *      @bar: BAR to be reserved
2996 *      @res_name: Name to be associated with resource.
2997 *      @exclusive: whether the region access is exclusive or not
2998 *
2999 *      Mark the PCI region associated with PCI device @pdev BR @bar as
3000 *      being reserved by owner @res_name.  Do not access any
3001 *      address inside the PCI regions unless this call returns
3002 *      successfully.
3003 *
3004 *      If @exclusive is set, then the region is marked so that userspace
3005 *      is explicitly not allowed to map the resource via /dev/mem or
3006 *      sysfs MMIO access.
3007 *
3008 *      Returns 0 on success, or %EBUSY on error.  A warning
3009 *      message is also printed on failure.
3010 */
3011static int __pci_request_region(struct pci_dev *pdev, int bar,
3012                                const char *res_name, int exclusive)
3013{
3014        struct pci_devres *dr;
3015
3016        if (pci_resource_len(pdev, bar) == 0)
3017                return 0;
3018
3019        if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3020                if (!request_region(pci_resource_start(pdev, bar),
3021                            pci_resource_len(pdev, bar), res_name))
3022                        goto err_out;
3023        } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3024                if (!__request_mem_region(pci_resource_start(pdev, bar),
3025                                        pci_resource_len(pdev, bar), res_name,
3026                                        exclusive))
3027                        goto err_out;
3028        }
3029
3030        dr = find_pci_dr(pdev);
3031        if (dr)
3032                dr->region_mask |= 1 << bar;
3033
3034        return 0;
3035
3036err_out:
3037        dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
3038                 &pdev->resource[bar]);
3039        return -EBUSY;
3040}
3041
3042/**
3043 *      pci_request_region - Reserve PCI I/O and memory resource
3044 *      @pdev: PCI device whose resources are to be reserved
3045 *      @bar: BAR to be reserved
3046 *      @res_name: Name to be associated with resource
3047 *
3048 *      Mark the PCI region associated with PCI device @pdev BAR @bar as
3049 *      being reserved by owner @res_name.  Do not access any
3050 *      address inside the PCI regions unless this call returns
3051 *      successfully.
3052 *
3053 *      Returns 0 on success, or %EBUSY on error.  A warning
3054 *      message is also printed on failure.
3055 */
3056int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3057{
3058        return __pci_request_region(pdev, bar, res_name, 0);
3059}
3060EXPORT_SYMBOL(pci_request_region);
3061
3062/**
3063 *      pci_request_region_exclusive - Reserved PCI I/O and memory resource
3064 *      @pdev: PCI device whose resources are to be reserved
3065 *      @bar: BAR to be reserved
3066 *      @res_name: Name to be associated with resource.
3067 *
3068 *      Mark the PCI region associated with PCI device @pdev BR @bar as
3069 *      being reserved by owner @res_name.  Do not access any
3070 *      address inside the PCI regions unless this call returns
3071 *      successfully.
3072 *
3073 *      Returns 0 on success, or %EBUSY on error.  A warning
3074 *      message is also printed on failure.
3075 *
3076 *      The key difference that _exclusive makes it that userspace is
3077 *      explicitly not allowed to map the resource via /dev/mem or
3078 *      sysfs.
3079 */
3080int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3081                                 const char *res_name)
3082{
3083        return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3084}
3085EXPORT_SYMBOL(pci_request_region_exclusive);
3086
3087/**
3088 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3089 * @pdev: PCI device whose resources were previously reserved
3090 * @bars: Bitmask of BARs to be released
3091 *
3092 * Release selected PCI I/O and memory resources previously reserved.
3093 * Call this function only after all use of the PCI regions has ceased.
3094 */
3095void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3096{
3097        int i;
3098
3099        for (i = 0; i < 6; i++)
3100                if (bars & (1 << i))
3101                        pci_release_region(pdev, i);
3102}
3103EXPORT_SYMBOL(pci_release_selected_regions);
3104
3105static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3106                                          const char *res_name, int excl)
3107{
3108        int i;
3109
3110        for (i = 0; i < 6; i++)
3111                if (bars & (1 << i))
3112                        if (__pci_request_region(pdev, i, res_name, excl))
3113                                goto err_out;
3114        return 0;
3115
3116err_out:
3117        while (--i >= 0)
3118                if (bars & (1 << i))
3119                        pci_release_region(pdev, i);
3120
3121        return -EBUSY;
3122}
3123
3124
3125/**
3126 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3127 * @pdev: PCI device whose resources are to be reserved
3128 * @bars: Bitmask of BARs to be requested
3129 * @res_name: Name to be associated with resource
3130 */
3131int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3132                                 const char *res_name)
3133{
3134        return __pci_request_selected_regions(pdev, bars, res_name, 0);
3135}
3136EXPORT_SYMBOL(pci_request_selected_regions);
3137
3138int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3139                                           const char *res_name)
3140{
3141        return __pci_request_selected_regions(pdev, bars, res_name,
3142                        IORESOURCE_EXCLUSIVE);
3143}
3144EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3145
3146/**
3147 *      pci_release_regions - Release reserved PCI I/O and memory resources
3148 *      @pdev: PCI device whose resources were previously reserved by pci_request_regions
3149 *
3150 *      Releases all PCI I/O and memory resources previously reserved by a
3151 *      successful call to pci_request_regions.  Call this function only
3152 *      after all use of the PCI regions has ceased.
3153 */
3154
3155void pci_release_regions(struct pci_dev *pdev)
3156{
3157        pci_release_selected_regions(pdev, (1 << 6) - 1);
3158}
3159EXPORT_SYMBOL(pci_release_regions);
3160
3161/**
3162 *      pci_request_regions - Reserved PCI I/O and memory resources
3163 *      @pdev: PCI device whose resources are to be reserved
3164 *      @res_name: Name to be associated with resource.
3165 *
3166 *      Mark all PCI regions associated with PCI device @pdev as
3167 *      being reserved by owner @res_name.  Do not access any
3168 *      address inside the PCI regions unless this call returns
3169 *      successfully.
3170 *
3171 *      Returns 0 on success, or %EBUSY on error.  A warning
3172 *      message is also printed on failure.
3173 */
3174int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3175{
3176        return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3177}
3178EXPORT_SYMBOL(pci_request_regions);
3179
3180/**
3181 *      pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3182 *      @pdev: PCI device whose resources are to be reserved
3183 *      @res_name: Name to be associated with resource.
3184 *
3185 *      Mark all PCI regions associated with PCI device @pdev as
3186 *      being reserved by owner @res_name.  Do not access any
3187 *      address inside the PCI regions unless this call returns
3188 *      successfully.
3189 *
3190 *      pci_request_regions_exclusive() will mark the region so that
3191 *      /dev/mem and the sysfs MMIO access will not be allowed.
3192 *
3193 *      Returns 0 on success, or %EBUSY on error.  A warning
3194 *      message is also printed on failure.
3195 */
3196int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3197{
3198        return pci_request_selected_regions_exclusive(pdev,
3199                                        ((1 << 6) - 1), res_name);
3200}
3201EXPORT_SYMBOL(pci_request_regions_exclusive);
3202
3203#ifdef PCI_IOBASE
3204struct io_range {
3205        struct list_head list;
3206        phys_addr_t start;
3207        resource_size_t size;
3208};
3209
3210static LIST_HEAD(io_range_list);
3211static DEFINE_SPINLOCK(io_range_lock);
3212#endif
3213
3214/*
3215 * Record the PCI IO range (expressed as CPU physical address + size).
3216 * Return a negative value if an error has occured, zero otherwise
3217 */
3218int __weak pci_register_io_range(phys_addr_t addr, resource_size_t size)
3219{
3220        int err = 0;
3221
3222#ifdef PCI_IOBASE
3223        struct io_range *range;
3224        resource_size_t allocated_size = 0;
3225
3226        /* check if the range hasn't been previously recorded */
3227        spin_lock(&io_range_lock);
3228        list_for_each_entry(range, &io_range_list, list) {
3229                if (addr >= range->start && addr + size <= range->start + size) {
3230                        /* range already registered, bail out */
3231                        goto end_register;
3232                }
3233                allocated_size += range->size;
3234        }
3235
3236        /* range not registed yet, check for available space */
3237        if (allocated_size + size - 1 > IO_SPACE_LIMIT) {
3238                /* if it's too big check if 64K space can be reserved */
3239                if (allocated_size + SZ_64K - 1 > IO_SPACE_LIMIT) {
3240                        err = -E2BIG;
3241                        goto end_register;
3242                }
3243
3244                size = SZ_64K;
3245                pr_warn("Requested IO range too big, new size set to 64K\n");
3246        }
3247
3248        /* add the range to the list */
3249        range = kzalloc(sizeof(*range), GFP_ATOMIC);
3250        if (!range) {
3251                err = -ENOMEM;
3252                goto end_register;
3253        }
3254
3255        range->start = addr;
3256        range->size = size;
3257
3258        list_add_tail(&range->list, &io_range_list);
3259
3260end_register:
3261        spin_unlock(&io_range_lock);
3262#endif
3263
3264        return err;
3265}
3266
3267phys_addr_t pci_pio_to_address(unsigned long pio)
3268{
3269        phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3270
3271#ifdef PCI_IOBASE
3272        struct io_range *range;
3273        resource_size_t allocated_size = 0;
3274
3275        if (pio > IO_SPACE_LIMIT)
3276                return address;
3277
3278        spin_lock(&io_range_lock);
3279        list_for_each_entry(range, &io_range_list, list) {
3280                if (pio >= allocated_size && pio < allocated_size + range->size) {
3281                        address = range->start + pio - allocated_size;
3282                        break;
3283                }
3284                allocated_size += range->size;
3285        }
3286        spin_unlock(&io_range_lock);
3287#endif
3288
3289        return address;
3290}
3291
3292unsigned long __weak pci_address_to_pio(phys_addr_t address)
3293{
3294#ifdef PCI_IOBASE
3295        struct io_range *res;
3296        resource_size_t offset = 0;
3297        unsigned long addr = -1;
3298
3299        spin_lock(&io_range_lock);
3300        list_for_each_entry(res, &io_range_list, list) {
3301                if (address >= res->start && address < res->start + res->size) {
3302                        addr = address - res->start + offset;
3303                        break;
3304                }
3305                offset += res->size;
3306        }
3307        spin_unlock(&io_range_lock);
3308
3309        return addr;
3310#else
3311        if (address > IO_SPACE_LIMIT)
3312                return (unsigned long)-1;
3313
3314        return (unsigned long) address;
3315#endif
3316}
3317
3318/**
3319 *      pci_remap_iospace - Remap the memory mapped I/O space
3320 *      @res: Resource describing the I/O space
3321 *      @phys_addr: physical address of range to be mapped
3322 *
3323 *      Remap the memory mapped I/O space described by the @res
3324 *      and the CPU physical address @phys_addr into virtual address space.
3325 *      Only architectures that have memory mapped IO functions defined
3326 *      (and the PCI_IOBASE value defined) should call this function.
3327 */
3328int __weak pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3329{
3330#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3331        unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3332
3333        if (!(res->flags & IORESOURCE_IO))
3334                return -EINVAL;
3335
3336        if (res->end > IO_SPACE_LIMIT)
3337                return -EINVAL;
3338
3339        return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3340                                  pgprot_device(PAGE_KERNEL));
3341#else
3342        /* this architecture does not have memory mapped I/O space,
3343           so this function should never be called */
3344        WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3345        return -ENODEV;
3346#endif
3347}
3348
3349/**
3350 *      pci_unmap_iospace - Unmap the memory mapped I/O space
3351 *      @res: resource to be unmapped
3352 *
3353 *      Unmap the CPU virtual address @res from virtual address space.
3354 *      Only architectures that have memory mapped IO functions defined
3355 *      (and the PCI_IOBASE value defined) should call this function.
3356 */
3357void pci_unmap_iospace(struct resource *res)
3358{
3359#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3360        unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3361
3362        unmap_kernel_range(vaddr, resource_size(res));
3363#endif
3364}
3365
3366static void __pci_set_master(struct pci_dev *dev, bool enable)
3367{
3368        u16 old_cmd, cmd;
3369
3370        pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3371        if (enable)
3372                cmd = old_cmd | PCI_COMMAND_MASTER;
3373        else
3374                cmd = old_cmd & ~PCI_COMMAND_MASTER;
3375        if (cmd != old_cmd) {
3376                dev_dbg(&dev->dev, "%s bus mastering\n",
3377                        enable ? "enabling" : "disabling");
3378                pci_write_config_word(dev, PCI_COMMAND, cmd);
3379        }
3380        dev->is_busmaster = enable;
3381}
3382
3383/**
3384 * pcibios_setup - process "pci=" kernel boot arguments
3385 * @str: string used to pass in "pci=" kernel boot arguments
3386 *
3387 * Process kernel boot arguments.  This is the default implementation.
3388 * Architecture specific implementations can override this as necessary.
3389 */
3390char * __weak __init pcibios_setup(char *str)
3391{
3392        return str;
3393}
3394
3395/**
3396 * pcibios_set_master - enable PCI bus-mastering for device dev
3397 * @dev: the PCI device to enable
3398 *
3399 * Enables PCI bus-mastering for the device.  This is the default
3400 * implementation.  Architecture specific implementations can override
3401 * this if necessary.
3402 */
3403void __weak pcibios_set_master(struct pci_dev *dev)
3404{
3405        u8 lat;
3406
3407        /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3408        if (pci_is_pcie(dev))
3409                return;
3410
3411        pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3412        if (lat < 16)
3413                lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3414        else if (lat > pcibios_max_latency)
3415                lat = pcibios_max_latency;
3416        else
3417                return;
3418
3419        pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3420}
3421
3422/**
3423 * pci_set_master - enables bus-mastering for device dev
3424 * @dev: the PCI device to enable
3425 *
3426 * Enables bus-mastering on the device and calls pcibios_set_master()
3427 * to do the needed arch specific settings.
3428 */
3429void pci_set_master(struct pci_dev *dev)
3430{
3431        __pci_set_master(dev, true);
3432        pcibios_set_master(dev);
3433}
3434EXPORT_SYMBOL(pci_set_master);
3435
3436/**
3437 * pci_clear_master - disables bus-mastering for device dev
3438 * @dev: the PCI device to disable
3439 */
3440void pci_clear_master(struct pci_dev *dev)
3441{
3442        __pci_set_master(dev, false);
3443}
3444EXPORT_SYMBOL(pci_clear_master);
3445
3446/**
3447 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3448 * @dev: the PCI device for which MWI is to be enabled
3449 *
3450 * Helper function for pci_set_mwi.
3451 * Originally copied from drivers/net/acenic.c.
3452 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3453 *
3454 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3455 */
3456int pci_set_cacheline_size(struct pci_dev *dev)
3457{
3458        u8 cacheline_size;
3459
3460        if (!pci_cache_line_size)
3461                return -EINVAL;
3462
3463        /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3464           equal to or multiple of the right value. */
3465        pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3466        if (cacheline_size >= pci_cache_line_size &&
3467            (cacheline_size % pci_cache_line_size) == 0)
3468                return 0;
3469
3470        /* Write the correct value. */
3471        pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3472        /* Read it back. */
3473        pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3474        if (cacheline_size == pci_cache_line_size)
3475                return 0;
3476
3477        dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not supported\n",
3478                   pci_cache_line_size << 2);
3479
3480        return -EINVAL;
3481}
3482EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3483
3484/**
3485 * pci_set_mwi - enables memory-write-invalidate PCI transaction
3486 * @dev: the PCI device for which MWI is enabled
3487 *
3488 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3489 *
3490 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3491 */
3492int pci_set_mwi(struct pci_dev *dev)
3493{
3494#ifdef PCI_DISABLE_MWI
3495        return 0;
3496#else
3497        int rc;
3498        u16 cmd;
3499
3500        rc = pci_set_cacheline_size(dev);
3501        if (rc)
3502                return rc;
3503
3504        pci_read_config_word(dev, PCI_COMMAND, &cmd);
3505        if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3506                dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
3507                cmd |= PCI_COMMAND_INVALIDATE;
3508                pci_write_config_word(dev, PCI_COMMAND, cmd);
3509        }
3510        return 0;
3511#endif
3512}
3513EXPORT_SYMBOL(pci_set_mwi);
3514
3515/**
3516 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3517 * @dev: the PCI device for which MWI is enabled
3518 *
3519 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3520 * Callers are not required to check the return value.
3521 *
3522 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3523 */
3524int pci_try_set_mwi(struct pci_dev *dev)
3525{
3526#ifdef PCI_DISABLE_MWI
3527        return 0;
3528#else
3529        return pci_set_mwi(dev);
3530#endif
3531}
3532EXPORT_SYMBOL(pci_try_set_mwi);
3533
3534/**
3535 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3536 * @dev: the PCI device to disable
3537 *
3538 * Disables PCI Memory-Write-Invalidate transaction on the device
3539 */
3540void pci_clear_mwi(struct pci_dev *dev)
3541{
3542#ifndef PCI_DISABLE_MWI
3543        u16 cmd;
3544
3545        pci_read_config_word(dev, PCI_COMMAND, &cmd);
3546        if (cmd & PCI_COMMAND_INVALIDATE) {
3547                cmd &= ~PCI_COMMAND_INVALIDATE;
3548                pci_write_config_word(dev, PCI_COMMAND, cmd);
3549        }
3550#endif
3551}
3552EXPORT_SYMBOL(pci_clear_mwi);
3553
3554/**
3555 * pci_intx - enables/disables PCI INTx for device dev
3556 * @pdev: the PCI device to operate on
3557 * @enable: boolean: whether to enable or disable PCI INTx
3558 *
3559 * Enables/disables PCI INTx for device dev
3560 */
3561void pci_intx(struct pci_dev *pdev, int enable)
3562{
3563        u16 pci_command, new;
3564
3565        pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3566
3567        if (enable)
3568                new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3569        else
3570                new = pci_command | PCI_COMMAND_INTX_DISABLE;
3571
3572        if (new != pci_command) {
3573                struct pci_devres *dr;
3574
3575                pci_write_config_word(pdev, PCI_COMMAND, new);
3576
3577                dr = find_pci_dr(pdev);
3578                if (dr && !dr->restore_intx) {
3579                        dr->restore_intx = 1;
3580                        dr->orig_intx = !enable;
3581                }
3582        }
3583}
3584EXPORT_SYMBOL_GPL(pci_intx);
3585
3586/**
3587 * pci_intx_mask_supported - probe for INTx masking support
3588 * @dev: the PCI device to operate on
3589 *
3590 * Check if the device dev support INTx masking via the config space
3591 * command word.
3592 */
3593bool pci_intx_mask_supported(struct pci_dev *dev)
3594{
3595        bool mask_supported = false;
3596        u16 orig, new;
3597
3598        if (dev->broken_intx_masking)
3599                return false;
3600
3601        pci_cfg_access_lock(dev);
3602
3603        pci_read_config_word(dev, PCI_COMMAND, &orig);
3604        pci_write_config_word(dev, PCI_COMMAND,
3605                              orig ^ PCI_COMMAND_INTX_DISABLE);
3606        pci_read_config_word(dev, PCI_COMMAND, &new);
3607
3608        /*
3609         * There's no way to protect against hardware bugs or detect them
3610         * reliably, but as long as we know what the value should be, let's
3611         * go ahead and check it.
3612         */
3613        if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
3614                dev_err(&dev->dev, "Command register changed from 0x%x to 0x%x: driver or hardware bug?\n",
3615                        orig, new);
3616        } else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
3617                mask_supported = true;
3618                pci_write_config_word(dev, PCI_COMMAND, orig);
3619        }
3620
3621        pci_cfg_access_unlock(dev);
3622        return mask_supported;
3623}
3624EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
3625
3626static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3627{
3628        struct pci_bus *bus = dev->bus;
3629        bool mask_updated = true;
3630        u32 cmd_status_dword;
3631        u16 origcmd, newcmd;
3632        unsigned long flags;
3633        bool irq_pending;
3634
3635        /*
3636         * We do a single dword read to retrieve both command and status.
3637         * Document assumptions that make this possible.
3638         */
3639        BUILD_BUG_ON(PCI_COMMAND % 4);
3640        BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3641
3642        raw_spin_lock_irqsave(&pci_lock, flags);
3643
3644        bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3645
3646        irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3647
3648        /*
3649         * Check interrupt status register to see whether our device
3650         * triggered the interrupt (when masking) or the next IRQ is
3651         * already pending (when unmasking).
3652         */
3653        if (mask != irq_pending) {
3654                mask_updated = false;
3655                goto done;
3656        }
3657
3658        origcmd = cmd_status_dword;
3659        newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3660        if (mask)
3661                newcmd |= PCI_COMMAND_INTX_DISABLE;
3662        if (newcmd != origcmd)
3663                bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3664
3665done:
3666        raw_spin_unlock_irqrestore(&pci_lock, flags);
3667
3668        return mask_updated;
3669}
3670
3671/**
3672 * pci_check_and_mask_intx - mask INTx on pending interrupt
3673 * @dev: the PCI device to operate on
3674 *
3675 * Check if the device dev has its INTx line asserted, mask it and
3676 * return true in that case. False is returned if not interrupt was
3677 * pending.
3678 */
3679bool pci_check_and_mask_intx(struct pci_dev *dev)
3680{
3681        return pci_check_and_set_intx_mask(dev, true);
3682}
3683EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3684
3685/**
3686 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3687 * @dev: the PCI device to operate on
3688 *
3689 * Check if the device dev has its INTx line asserted, unmask it if not
3690 * and return true. False is returned and the mask remains active if
3691 * there was still an interrupt pending.
3692 */
3693bool pci_check_and_unmask_intx(struct pci_dev *dev)
3694{
3695        return pci_check_and_set_intx_mask(dev, false);
3696}
3697EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3698
3699/**
3700 * pci_wait_for_pending_transaction - waits for pending transaction
3701 * @dev: the PCI device to operate on
3702 *
3703 * Return 0 if transaction is pending 1 otherwise.
3704 */
3705int pci_wait_for_pending_transaction(struct pci_dev *dev)
3706{
3707        if (!pci_is_pcie(dev))
3708                return 1;
3709
3710        return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3711                                    PCI_EXP_DEVSTA_TRPND);
3712}
3713EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3714
3715/*
3716 * We should only need to wait 100ms after FLR, but some devices take longer.
3717 * Wait for up to 1000ms for config space to return something other than -1.
3718 * Intel IGD requires this when an LCD panel is attached.  We read the 2nd
3719 * dword because VFs don't implement the 1st dword.
3720 */
3721static void pci_flr_wait(struct pci_dev *dev)
3722{
3723        int i = 0;
3724        u32 id;
3725
3726        do {
3727                msleep(100);
3728                pci_read_config_dword(dev, PCI_COMMAND, &id);
3729        } while (i++ < 10 && id == ~0);
3730
3731        if (id == ~0)
3732                dev_warn(&dev->dev, "Failed to return from FLR\n");
3733        else if (i > 1)
3734                dev_info(&dev->dev, "Required additional %dms to return from FLR\n",
3735                         (i - 1) * 100);
3736}
3737
3738static int pcie_flr(struct pci_dev *dev, int probe)
3739{
3740        u32 cap;
3741
3742        pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
3743        if (!(cap & PCI_EXP_DEVCAP_FLR))
3744                return -ENOTTY;
3745
3746        if (probe)
3747                return 0;
3748
3749        if (!pci_wait_for_pending_transaction(dev))
3750                dev_err(&dev->dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
3751
3752        pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
3753        pci_flr_wait(dev);
3754        return 0;
3755}
3756
3757static int pci_af_flr(struct pci_dev *dev, int probe)
3758{
3759        int pos;
3760        u8 cap;
3761
3762        pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3763        if (!pos)
3764                return -ENOTTY;
3765
3766        pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3767        if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3768                return -ENOTTY;
3769
3770        if (probe)
3771                return 0;
3772
3773        /*
3774         * Wait for Transaction Pending bit to clear.  A word-aligned test
3775         * is used, so we use the conrol offset rather than status and shift
3776         * the test bit to match.
3777         */
3778        if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
3779                                 PCI_AF_STATUS_TP << 8))
3780                dev_err(&dev->dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
3781
3782        pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3783        pci_flr_wait(dev);
3784        return 0;
3785}
3786
3787/**
3788 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3789 * @dev: Device to reset.
3790 * @probe: If set, only check if the device can be reset this way.
3791 *
3792 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3793 * unset, it will be reinitialized internally when going from PCI_D3hot to
3794 * PCI_D0.  If that's the case and the device is not in a low-power state
3795 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3796 *
3797 * NOTE: This causes the caller to sleep for twice the device power transition
3798 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3799 * by default (i.e. unless the @dev's d3_delay field has a different value).
3800 * Moreover, only devices in D0 can be reset by this function.
3801 */
3802static int pci_pm_reset(struct pci_dev *dev, int probe)
3803{
3804        u16 csr;
3805
3806        if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
3807                return -ENOTTY;
3808
3809        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3810        if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3811                return -ENOTTY;
3812
3813        if (probe)
3814                return 0;
3815
3816        if (dev->current_state != PCI_D0)
3817                return -EINVAL;
3818
3819        csr &= ~PCI_PM_CTRL_STATE_MASK;
3820        csr |= PCI_D3hot;
3821        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3822        pci_dev_d3_sleep(dev);
3823
3824        csr &= ~PCI_PM_CTRL_STATE_MASK;
3825        csr |= PCI_D0;
3826        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3827        pci_dev_d3_sleep(dev);
3828
3829        return 0;
3830}
3831
3832void pci_reset_secondary_bus(struct pci_dev *dev)
3833{
3834        u16 ctrl;
3835
3836        pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
3837        ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
3838        pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3839        /*
3840         * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms.  Double
3841         * this to 2ms to ensure that we meet the minimum requirement.
3842         */
3843        msleep(2);
3844
3845        ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
3846        pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3847
3848        /*
3849         * Trhfa for conventional PCI is 2^25 clock cycles.
3850         * Assuming a minimum 33MHz clock this results in a 1s
3851         * delay before we can consider subordinate devices to
3852         * be re-initialized.  PCIe has some ways to shorten this,
3853         * but we don't make use of them yet.
3854         */
3855        ssleep(1);
3856}
3857
3858void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
3859{
3860        pci_reset_secondary_bus(dev);
3861}
3862
3863/**
3864 * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
3865 * @dev: Bridge device
3866 *
3867 * Use the bridge control register to assert reset on the secondary bus.
3868 * Devices on the secondary bus are left in power-on state.
3869 */
3870void pci_reset_bridge_secondary_bus(struct pci_dev *dev)
3871{
3872        pcibios_reset_secondary_bus(dev);
3873}
3874EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
3875
3876static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
3877{
3878        struct pci_dev *pdev;
3879
3880        if (pci_is_root_bus(dev->bus) || dev->subordinate ||
3881            !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
3882                return -ENOTTY;
3883
3884        list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3885                if (pdev != dev)
3886                        return -ENOTTY;
3887
3888        if (probe)
3889                return 0;
3890
3891        pci_reset_bridge_secondary_bus(dev->bus->self);
3892
3893        return 0;
3894}
3895
3896static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
3897{
3898        int rc = -ENOTTY;
3899
3900        if (!hotplug || !try_module_get(hotplug->ops->owner))
3901                return rc;
3902
3903        if (hotplug->ops->reset_slot)
3904                rc = hotplug->ops->reset_slot(hotplug, probe);
3905
3906        module_put(hotplug->ops->owner);
3907
3908        return rc;
3909}
3910
3911static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
3912{
3913        struct pci_dev *pdev;
3914
3915        if (dev->subordinate || !dev->slot ||
3916            dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
3917                return -ENOTTY;
3918
3919        list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3920                if (pdev != dev && pdev->slot == dev->slot)
3921                        return -ENOTTY;
3922
3923        return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
3924}
3925
3926static int __pci_dev_reset(struct pci_dev *dev, int probe)
3927{
3928        int rc;
3929
3930        might_sleep();
3931
3932        rc = pci_dev_specific_reset(dev, probe);
3933        if (rc != -ENOTTY)
3934                goto done;
3935
3936        rc = pcie_flr(dev, probe);
3937        if (rc != -ENOTTY)
3938                goto done;
3939
3940        rc = pci_af_flr(dev, probe);
3941        if (rc != -ENOTTY)
3942                goto done;
3943
3944        rc = pci_pm_reset(dev, probe);
3945        if (rc != -ENOTTY)
3946                goto done;
3947
3948        rc = pci_dev_reset_slot_function(dev, probe);
3949        if (rc != -ENOTTY)
3950                goto done;
3951
3952        rc = pci_parent_bus_reset(dev, probe);
3953done:
3954        return rc;
3955}
3956
3957static void pci_dev_lock(struct pci_dev *dev)
3958{
3959        pci_cfg_access_lock(dev);
3960        /* block PM suspend, driver probe, etc. */
3961        device_lock(&dev->dev);
3962}
3963
3964/* Return 1 on successful lock, 0 on contention */
3965static int pci_dev_trylock(struct pci_dev *dev)
3966{
3967        if (pci_cfg_access_trylock(dev)) {
3968                if (device_trylock(&dev->dev))
3969                        return 1;
3970                pci_cfg_access_unlock(dev);
3971        }
3972
3973        return 0;
3974}
3975
3976static void pci_dev_unlock(struct pci_dev *dev)
3977{
3978        device_unlock(&dev->dev);
3979        pci_cfg_access_unlock(dev);
3980}
3981
3982/**
3983 * pci_reset_notify - notify device driver of reset
3984 * @dev: device to be notified of reset
3985 * @prepare: 'true' if device is about to be reset; 'false' if reset attempt
3986 *           completed
3987 *
3988 * Must be called prior to device access being disabled and after device
3989 * access is restored.
3990 */
3991static void pci_reset_notify(struct pci_dev *dev, bool prepare)
3992{
3993        const struct pci_error_handlers *err_handler =
3994                        dev->driver ? dev->driver->err_handler : NULL;
3995        if (err_handler && err_handler->reset_notify)
3996                err_handler->reset_notify(dev, prepare);
3997}
3998
3999static void pci_dev_save_and_disable(struct pci_dev *dev)
4000{
4001        pci_reset_notify(dev, true);
4002
4003        /*
4004         * Wake-up device prior to save.  PM registers default to D0 after
4005         * reset and a simple register restore doesn't reliably return
4006         * to a non-D0 state anyway.
4007         */
4008        pci_set_power_state(dev, PCI_D0);
4009
4010        pci_save_state(dev);
4011        /*
4012         * Disable the device by clearing the Command register, except for
4013         * INTx-disable which is set.  This not only disables MMIO and I/O port
4014         * BARs, but also prevents the device from being Bus Master, preventing
4015         * DMA from the device including MSI/MSI-X interrupts.  For PCI 2.3
4016         * compliant devices, INTx-disable prevents legacy interrupts.
4017         */
4018        pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4019}
4020
4021static void pci_dev_restore(struct pci_dev *dev)
4022{
4023        pci_restore_state(dev);
4024        pci_reset_notify(dev, false);
4025}
4026
4027static int pci_dev_reset(struct pci_dev *dev, int probe)
4028{
4029        int rc;
4030
4031        if (!probe)
4032                pci_dev_lock(dev);
4033
4034        rc = __pci_dev_reset(dev, probe);
4035
4036        if (!probe)
4037                pci_dev_unlock(dev);
4038
4039        return rc;
4040}
4041
4042/**
4043 * __pci_reset_function - reset a PCI device function
4044 * @dev: PCI device to reset
4045 *
4046 * Some devices allow an individual function to be reset without affecting
4047 * other functions in the same device.  The PCI device must be responsive
4048 * to PCI config space in order to use this function.
4049 *
4050 * The device function is presumed to be unused when this function is called.
4051 * Resetting the device will make the contents of PCI configuration space
4052 * random, so any caller of this must be prepared to reinitialise the
4053 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4054 * etc.
4055 *
4056 * Returns 0 if the device function was successfully reset or negative if the
4057 * device doesn't support resetting a single function.
4058 */
4059int __pci_reset_function(struct pci_dev *dev)
4060{
4061        return pci_dev_reset(dev, 0);
4062}
4063EXPORT_SYMBOL_GPL(__pci_reset_function);
4064
4065/**
4066 * __pci_reset_function_locked - reset a PCI device function while holding
4067 * the @dev mutex lock.
4068 * @dev: PCI device to reset
4069 *
4070 * Some devices allow an individual function to be reset without affecting
4071 * other functions in the same device.  The PCI device must be responsive
4072 * to PCI config space in order to use this function.
4073 *
4074 * The device function is presumed to be unused and the caller is holding
4075 * the device mutex lock when this function is called.
4076 * Resetting the device will make the contents of PCI configuration space
4077 * random, so any caller of this must be prepared to reinitialise the
4078 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4079 * etc.
4080 *
4081 * Returns 0 if the device function was successfully reset or negative if the
4082 * device doesn't support resetting a single function.
4083 */
4084int __pci_reset_function_locked(struct pci_dev *dev)
4085{
4086        return __pci_dev_reset(dev, 0);
4087}
4088EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4089
4090/**
4091 * pci_probe_reset_function - check whether the device can be safely reset
4092 * @dev: PCI device to reset
4093 *
4094 * Some devices allow an individual function to be reset without affecting
4095 * other functions in the same device.  The PCI device must be responsive
4096 * to PCI config space in order to use this function.
4097 *
4098 * Returns 0 if the device function can be reset or negative if the
4099 * device doesn't support resetting a single function.
4100 */
4101int pci_probe_reset_function(struct pci_dev *dev)
4102{
4103        return pci_dev_reset(dev, 1);
4104}
4105
4106/**
4107 * pci_reset_function - quiesce and reset a PCI device function
4108 * @dev: PCI device to reset
4109 *
4110 * Some devices allow an individual function to be reset without affecting
4111 * other functions in the same device.  The PCI device must be responsive
4112 * to PCI config space in order to use this function.
4113 *
4114 * This function does not just reset the PCI portion of a device, but
4115 * clears all the state associated with the device.  This function differs
4116 * from __pci_reset_function in that it saves and restores device state
4117 * over the reset.
4118 *
4119 * Returns 0 if the device function was successfully reset or negative if the
4120 * device doesn't support resetting a single function.
4121 */
4122int pci_reset_function(struct pci_dev *dev)
4123{
4124        int rc;
4125
4126        rc = pci_dev_reset(dev, 1);
4127        if (rc)
4128                return rc;
4129
4130        pci_dev_save_and_disable(dev);
4131
4132        rc = pci_dev_reset(dev, 0);
4133
4134        pci_dev_restore(dev);
4135
4136        return rc;
4137}
4138EXPORT_SYMBOL_GPL(pci_reset_function);
4139
4140/**
4141 * pci_try_reset_function - quiesce and reset a PCI device function
4142 * @dev: PCI device to reset
4143 *
4144 * Same as above, except return -EAGAIN if unable to lock device.
4145 */
4146int pci_try_reset_function(struct pci_dev *dev)
4147{
4148        int rc;
4149
4150        rc = pci_dev_reset(dev, 1);
4151        if (rc)
4152                return rc;
4153
4154        pci_dev_save_and_disable(dev);
4155
4156        if (pci_dev_trylock(dev)) {
4157                rc = __pci_dev_reset(dev, 0);
4158                pci_dev_unlock(dev);
4159        } else
4160                rc = -EAGAIN;
4161
4162        pci_dev_restore(dev);
4163
4164        return rc;
4165}
4166EXPORT_SYMBOL_GPL(pci_try_reset_function);
4167
4168/* Do any devices on or below this bus prevent a bus reset? */
4169static bool pci_bus_resetable(struct pci_bus *bus)
4170{
4171        struct pci_dev *dev;
4172
4173        list_for_each_entry(dev, &bus->devices, bus_list) {
4174                if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4175                    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4176                        return false;
4177        }
4178
4179        return true;
4180}
4181
4182/* Lock devices from the top of the tree down */
4183static void pci_bus_lock(struct pci_bus *bus)
4184{
4185        struct pci_dev *dev;
4186
4187        list_for_each_entry(dev, &bus->devices, bus_list) {
4188                pci_dev_lock(dev);
4189                if (dev->subordinate)
4190                        pci_bus_lock(dev->subordinate);
4191        }
4192}
4193
4194/* Unlock devices from the bottom of the tree up */
4195static void pci_bus_unlock(struct pci_bus *bus)
4196{
4197        struct pci_dev *dev;
4198
4199        list_for_each_entry(dev, &bus->devices, bus_list) {
4200                if (dev->subordinate)
4201                        pci_bus_unlock(dev->subordinate);
4202                pci_dev_unlock(dev);
4203        }
4204}
4205
4206/* Return 1 on successful lock, 0 on contention */
4207static int pci_bus_trylock(struct pci_bus *bus)
4208{
4209        struct pci_dev *dev;
4210
4211        list_for_each_entry(dev, &bus->devices, bus_list) {
4212                if (!pci_dev_trylock(dev))
4213                        goto unlock;
4214                if (dev->subordinate) {
4215                        if (!pci_bus_trylock(dev->subordinate)) {
4216                                pci_dev_unlock(dev);
4217                                goto unlock;
4218                        }
4219                }
4220        }
4221        return 1;
4222
4223unlock:
4224        list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4225                if (dev->subordinate)
4226                        pci_bus_unlock(dev->subordinate);
4227                pci_dev_unlock(dev);
4228        }
4229        return 0;
4230}
4231
4232/* Do any devices on or below this slot prevent a bus reset? */
4233static bool pci_slot_resetable(struct pci_slot *slot)
4234{
4235        struct pci_dev *dev;
4236
4237        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4238                if (!dev->slot || dev->slot != slot)
4239                        continue;
4240                if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4241                    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4242                        return false;
4243        }
4244
4245        return true;
4246}
4247
4248/* Lock devices from the top of the tree down */
4249static void pci_slot_lock(struct pci_slot *slot)
4250{
4251        struct pci_dev *dev;
4252
4253        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4254                if (!dev->slot || dev->slot != slot)
4255                        continue;
4256                pci_dev_lock(dev);
4257                if (dev->subordinate)
4258                        pci_bus_lock(dev->subordinate);
4259        }
4260}
4261
4262/* Unlock devices from the bottom of the tree up */
4263static void pci_slot_unlock(struct pci_slot *slot)
4264{
4265        struct pci_dev *dev;
4266
4267        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4268                if (!dev->slot || dev->slot != slot)
4269                        continue;
4270                if (dev->subordinate)
4271                        pci_bus_unlock(dev->subordinate);
4272                pci_dev_unlock(dev);
4273        }
4274}
4275
4276/* Return 1 on successful lock, 0 on contention */
4277static int pci_slot_trylock(struct pci_slot *slot)
4278{
4279        struct pci_dev *dev;
4280
4281        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4282                if (!dev->slot || dev->slot != slot)
4283                        continue;
4284                if (!pci_dev_trylock(dev))
4285                        goto unlock;
4286                if (dev->subordinate) {
4287                        if (!pci_bus_trylock(dev->subordinate)) {
4288                                pci_dev_unlock(dev);
4289                                goto unlock;
4290                        }
4291                }
4292        }
4293        return 1;
4294
4295unlock:
4296        list_for_each_entry_continue_reverse(dev,
4297                                             &slot->bus->devices, bus_list) {
4298                if (!dev->slot || dev->slot != slot)
4299                        continue;
4300                if (dev->subordinate)
4301                        pci_bus_unlock(dev->subordinate);
4302                pci_dev_unlock(dev);
4303        }
4304        return 0;
4305}
4306
4307/* Save and disable devices from the top of the tree down */
4308static void pci_bus_save_and_disable(struct pci_bus *bus)
4309{
4310        struct pci_dev *dev;
4311
4312        list_for_each_entry(dev, &bus->devices, bus_list) {
4313                pci_dev_save_and_disable(dev);
4314                if (dev->subordinate)
4315                        pci_bus_save_and_disable(dev->subordinate);
4316        }
4317}
4318
4319/*
4320 * Restore devices from top of the tree down - parent bridges need to be
4321 * restored before we can get to subordinate devices.
4322 */
4323static void pci_bus_restore(struct pci_bus *bus)
4324{
4325        struct pci_dev *dev;
4326
4327        list_for_each_entry(dev, &bus->devices, bus_list) {
4328                pci_dev_restore(dev);
4329                if (dev->subordinate)
4330                        pci_bus_restore(dev->subordinate);
4331        }
4332}
4333
4334/* Save and disable devices from the top of the tree down */
4335static void pci_slot_save_and_disable(struct pci_slot *slot)
4336{
4337        struct pci_dev *dev;
4338
4339        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4340                if (!dev->slot || dev->slot != slot)
4341                        continue;
4342                pci_dev_save_and_disable(dev);
4343                if (dev->subordinate)
4344                        pci_bus_save_and_disable(dev->subordinate);
4345        }
4346}
4347
4348/*
4349 * Restore devices from top of the tree down - parent bridges need to be
4350 * restored before we can get to subordinate devices.
4351 */
4352static void pci_slot_restore(struct pci_slot *slot)
4353{
4354        struct pci_dev *dev;
4355
4356        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4357                if (!dev->slot || dev->slot != slot)
4358                        continue;
4359                pci_dev_restore(dev);
4360                if (dev->subordinate)
4361                        pci_bus_restore(dev->subordinate);
4362        }
4363}
4364
4365static int pci_slot_reset(struct pci_slot *slot, int probe)
4366{
4367        int rc;
4368
4369        if (!slot || !pci_slot_resetable(slot))
4370                return -ENOTTY;
4371
4372        if (!probe)
4373                pci_slot_lock(slot);
4374
4375        might_sleep();
4376
4377        rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4378
4379        if (!probe)
4380                pci_slot_unlock(slot);
4381
4382        return rc;
4383}
4384
4385/**
4386 * pci_probe_reset_slot - probe whether a PCI slot can be reset
4387 * @slot: PCI slot to probe
4388 *
4389 * Return 0 if slot can be reset, negative if a slot reset is not supported.
4390 */
4391int pci_probe_reset_slot(struct pci_slot *slot)
4392{
4393        return pci_slot_reset(slot, 1);
4394}
4395EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4396
4397/**
4398 * pci_reset_slot - reset a PCI slot
4399 * @slot: PCI slot to reset
4400 *
4401 * A PCI bus may host multiple slots, each slot may support a reset mechanism
4402 * independent of other slots.  For instance, some slots may support slot power
4403 * control.  In the case of a 1:1 bus to slot architecture, this function may
4404 * wrap the bus reset to avoid spurious slot related events such as hotplug.
4405 * Generally a slot reset should be attempted before a bus reset.  All of the
4406 * function of the slot and any subordinate buses behind the slot are reset
4407 * through this function.  PCI config space of all devices in the slot and
4408 * behind the slot is saved before and restored after reset.
4409 *
4410 * Return 0 on success, non-zero on error.
4411 */
4412int pci_reset_slot(struct pci_slot *slot)
4413{
4414        int rc;
4415
4416        rc = pci_slot_reset(slot, 1);
4417        if (rc)
4418                return rc;
4419
4420        pci_slot_save_and_disable(slot);
4421
4422        rc = pci_slot_reset(slot, 0);
4423
4424        pci_slot_restore(slot);
4425
4426        return rc;
4427}
4428EXPORT_SYMBOL_GPL(pci_reset_slot);
4429
4430/**
4431 * pci_try_reset_slot - Try to reset a PCI slot
4432 * @slot: PCI slot to reset
4433 *
4434 * Same as above except return -EAGAIN if the slot cannot be locked
4435 */
4436int pci_try_reset_slot(struct pci_slot *slot)
4437{
4438        int rc;
4439
4440        rc = pci_slot_reset(slot, 1);
4441        if (rc)
4442                return rc;
4443
4444        pci_slot_save_and_disable(slot);
4445
4446        if (pci_slot_trylock(slot)) {
4447                might_sleep();
4448                rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4449                pci_slot_unlock(slot);
4450        } else
4451                rc = -EAGAIN;
4452
4453        pci_slot_restore(slot);
4454
4455        return rc;
4456}
4457EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4458
4459static int pci_bus_reset(struct pci_bus *bus, int probe)
4460{
4461        if (!bus->self || !pci_bus_resetable(bus))
4462                return -ENOTTY;
4463
4464        if (probe)
4465                return 0;
4466
4467        pci_bus_lock(bus);
4468
4469        might_sleep();
4470
4471        pci_reset_bridge_secondary_bus(bus->self);
4472
4473        pci_bus_unlock(bus);
4474
4475        return 0;
4476}
4477
4478/**
4479 * pci_probe_reset_bus - probe whether a PCI bus can be reset
4480 * @bus: PCI bus to probe
4481 *
4482 * Return 0 if bus can be reset, negative if a bus reset is not supported.
4483 */
4484int pci_probe_reset_bus(struct pci_bus *bus)
4485{
4486        return pci_bus_reset(bus, 1);
4487}
4488EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4489
4490/**
4491 * pci_reset_bus - reset a PCI bus
4492 * @bus: top level PCI bus to reset
4493 *
4494 * Do a bus reset on the given bus and any subordinate buses, saving
4495 * and restoring state of all devices.
4496 *
4497 * Return 0 on success, non-zero on error.
4498 */
4499int pci_reset_bus(struct pci_bus *bus)
4500{
4501        int rc;
4502
4503        rc = pci_bus_reset(bus, 1);
4504        if (rc)
4505                return rc;
4506
4507        pci_bus_save_and_disable(bus);
4508
4509        rc = pci_bus_reset(bus, 0);
4510
4511        pci_bus_restore(bus);
4512
4513        return rc;
4514}
4515EXPORT_SYMBOL_GPL(pci_reset_bus);
4516
4517/**
4518 * pci_try_reset_bus - Try to reset a PCI bus
4519 * @bus: top level PCI bus to reset
4520 *
4521 * Same as above except return -EAGAIN if the bus cannot be locked
4522 */
4523int pci_try_reset_bus(struct pci_bus *bus)
4524{
4525        int rc;
4526
4527        rc = pci_bus_reset(bus, 1);
4528        if (rc)
4529                return rc;
4530
4531        pci_bus_save_and_disable(bus);
4532
4533        if (pci_bus_trylock(bus)) {
4534                might_sleep();
4535                pci_reset_bridge_secondary_bus(bus->self);
4536                pci_bus_unlock(bus);
4537        } else
4538                rc = -EAGAIN;
4539
4540        pci_bus_restore(bus);
4541
4542        return rc;
4543}
4544EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4545
4546/**
4547 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4548 * @dev: PCI device to query
4549 *
4550 * Returns mmrbc: maximum designed memory read count in bytes
4551 *    or appropriate error value.
4552 */
4553int pcix_get_max_mmrbc(struct pci_dev *dev)
4554{
4555        int cap;
4556        u32 stat;
4557
4558        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4559        if (!cap)
4560                return -EINVAL;
4561
4562        if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4563                return -EINVAL;
4564
4565        return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4566}
4567EXPORT_SYMBOL(pcix_get_max_mmrbc);
4568
4569/**
4570 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4571 * @dev: PCI device to query
4572 *
4573 * Returns mmrbc: maximum memory read count in bytes
4574 *    or appropriate error value.
4575 */
4576int pcix_get_mmrbc(struct pci_dev *dev)
4577{
4578        int cap;
4579        u16 cmd;
4580
4581        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4582        if (!cap)
4583                return -EINVAL;
4584
4585        if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4586                return -EINVAL;
4587
4588        return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4589}
4590EXPORT_SYMBOL(pcix_get_mmrbc);
4591
4592/**
4593 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4594 * @dev: PCI device to query
4595 * @mmrbc: maximum memory read count in bytes
4596 *    valid values are 512, 1024, 2048, 4096
4597 *
4598 * If possible sets maximum memory read byte count, some bridges have erratas
4599 * that prevent this.
4600 */
4601int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4602{
4603        int cap;
4604        u32 stat, v, o;
4605        u16 cmd;
4606
4607        if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4608                return -EINVAL;
4609
4610        v = ffs(mmrbc) - 10;
4611
4612        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4613        if (!cap)
4614                return -EINVAL;
4615
4616        if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4617                return -EINVAL;
4618
4619        if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4620                return -E2BIG;
4621
4622        if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4623                return -EINVAL;
4624
4625        o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4626        if (o != v) {
4627                if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4628                        return -EIO;
4629
4630                cmd &= ~PCI_X_CMD_MAX_READ;
4631                cmd |= v << 2;
4632                if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4633                        return -EIO;
4634        }
4635        return 0;
4636}
4637EXPORT_SYMBOL(pcix_set_mmrbc);
4638
4639/**
4640 * pcie_get_readrq - get PCI Express read request size
4641 * @dev: PCI device to query
4642 *
4643 * Returns maximum memory read request in bytes
4644 *    or appropriate error value.
4645 */
4646int pcie_get_readrq(struct pci_dev *dev)
4647{
4648        u16 ctl;
4649
4650        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4651
4652        return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
4653}
4654EXPORT_SYMBOL(pcie_get_readrq);
4655
4656/**
4657 * pcie_set_readrq - set PCI Express maximum memory read request
4658 * @dev: PCI device to query
4659 * @rq: maximum memory read count in bytes
4660 *    valid values are 128, 256, 512, 1024, 2048, 4096
4661 *
4662 * If possible sets maximum memory read request in bytes
4663 */
4664int pcie_set_readrq(struct pci_dev *dev, int rq)
4665{
4666        u16 v;
4667
4668        if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
4669                return -EINVAL;
4670
4671        /*
4672         * If using the "performance" PCIe config, we clamp the
4673         * read rq size to the max packet size to prevent the
4674         * host bridge generating requests larger than we can
4675         * cope with
4676         */
4677        if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
4678                int mps = pcie_get_mps(dev);
4679
4680                if (mps < rq)
4681                        rq = mps;
4682        }
4683
4684        v = (ffs(rq) - 8) << 12;
4685
4686        return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4687                                                  PCI_EXP_DEVCTL_READRQ, v);
4688}
4689EXPORT_SYMBOL(pcie_set_readrq);
4690
4691/**
4692 * pcie_get_mps - get PCI Express maximum payload size
4693 * @dev: PCI device to query
4694 *
4695 * Returns maximum payload size in bytes
4696 */
4697int pcie_get_mps(struct pci_dev *dev)
4698{
4699        u16 ctl;
4700
4701        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4702
4703        return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
4704}
4705EXPORT_SYMBOL(pcie_get_mps);
4706
4707/**
4708 * pcie_set_mps - set PCI Express maximum payload size
4709 * @dev: PCI device to query
4710 * @mps: maximum payload size in bytes
4711 *    valid values are 128, 256, 512, 1024, 2048, 4096
4712 *
4713 * If possible sets maximum payload size
4714 */
4715int pcie_set_mps(struct pci_dev *dev, int mps)
4716{
4717        u16 v;
4718
4719        if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
4720                return -EINVAL;
4721
4722        v = ffs(mps) - 8;
4723        if (v > dev->pcie_mpss)
4724                return -EINVAL;
4725        v <<= 5;
4726
4727        return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4728                                                  PCI_EXP_DEVCTL_PAYLOAD, v);
4729}
4730EXPORT_SYMBOL(pcie_set_mps);
4731
4732/**
4733 * pcie_get_minimum_link - determine minimum link settings of a PCI device
4734 * @dev: PCI device to query
4735 * @speed: storage for minimum speed
4736 * @width: storage for minimum width
4737 *
4738 * This function will walk up the PCI device chain and determine the minimum
4739 * link width and speed of the device.
4740 */
4741int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
4742                          enum pcie_link_width *width)
4743{
4744        int ret;
4745
4746        *speed = PCI_SPEED_UNKNOWN;
4747        *width = PCIE_LNK_WIDTH_UNKNOWN;
4748
4749        while (dev) {
4750                u16 lnksta;
4751                enum pci_bus_speed next_speed;
4752                enum pcie_link_width next_width;
4753
4754                ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
4755                if (ret)
4756                        return ret;
4757
4758                next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
4759                next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
4760                        PCI_EXP_LNKSTA_NLW_SHIFT;
4761
4762                if (next_speed < *speed)
4763                        *speed = next_speed;
4764
4765                if (next_width < *width)
4766                        *width = next_width;
4767
4768                dev = dev->bus->self;
4769        }
4770
4771        return 0;
4772}
4773EXPORT_SYMBOL(pcie_get_minimum_link);
4774
4775/**
4776 * pci_select_bars - Make BAR mask from the type of resource
4777 * @dev: the PCI device for which BAR mask is made
4778 * @flags: resource type mask to be selected
4779 *
4780 * This helper routine makes bar mask from the type of resource.
4781 */
4782int pci_select_bars(struct pci_dev *dev, unsigned long flags)
4783{
4784        int i, bars = 0;
4785        for (i = 0; i < PCI_NUM_RESOURCES; i++)
4786                if (pci_resource_flags(dev, i) & flags)
4787                        bars |= (1 << i);
4788        return bars;
4789}
4790EXPORT_SYMBOL(pci_select_bars);
4791
4792/**
4793 * pci_resource_bar - get position of the BAR associated with a resource
4794 * @dev: the PCI device
4795 * @resno: the resource number
4796 * @type: the BAR type to be filled in
4797 *
4798 * Returns BAR position in config space, or 0 if the BAR is invalid.
4799 */
4800int pci_resource_bar(struct pci_dev *dev, int resno, enum pci_bar_type *type)
4801{
4802        int reg;
4803
4804        if (resno < PCI_ROM_RESOURCE) {
4805                *type = pci_bar_unknown;
4806                return PCI_BASE_ADDRESS_0 + 4 * resno;
4807        } else if (resno == PCI_ROM_RESOURCE) {
4808                *type = pci_bar_mem32;
4809                return dev->rom_base_reg;
4810        } else if (resno < PCI_BRIDGE_RESOURCES) {
4811                /* device specific resource */
4812                *type = pci_bar_unknown;
4813                reg = pci_iov_resource_bar(dev, resno);
4814                if (reg)
4815                        return reg;
4816        }
4817
4818        dev_err(&dev->dev, "BAR %d: invalid resource\n", resno);
4819        return 0;
4820}
4821
4822/* Some architectures require additional programming to enable VGA */
4823static arch_set_vga_state_t arch_set_vga_state;
4824
4825void __init pci_register_set_vga_state(arch_set_vga_state_t func)
4826{
4827        arch_set_vga_state = func;      /* NULL disables */
4828}
4829
4830static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
4831                                  unsigned int command_bits, u32 flags)
4832{
4833        if (arch_set_vga_state)
4834                return arch_set_vga_state(dev, decode, command_bits,
4835                                                flags);
4836        return 0;
4837}
4838
4839/**
4840 * pci_set_vga_state - set VGA decode state on device and parents if requested
4841 * @dev: the PCI device
4842 * @decode: true = enable decoding, false = disable decoding
4843 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
4844 * @flags: traverse ancestors and change bridges
4845 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
4846 */
4847int pci_set_vga_state(struct pci_dev *dev, bool decode,
4848                      unsigned int command_bits, u32 flags)
4849{
4850        struct pci_bus *bus;
4851        struct pci_dev *bridge;
4852        u16 cmd;
4853        int rc;
4854
4855        WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
4856
4857        /* ARCH specific VGA enables */
4858        rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
4859        if (rc)
4860                return rc;
4861
4862        if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
4863                pci_read_config_word(dev, PCI_COMMAND, &cmd);
4864                if (decode == true)
4865                        cmd |= command_bits;
4866                else
4867                        cmd &= ~command_bits;
4868                pci_write_config_word(dev, PCI_COMMAND, cmd);
4869        }
4870
4871        if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
4872                return 0;
4873
4874        bus = dev->bus;
4875        while (bus) {
4876                bridge = bus->self;
4877                if (bridge) {
4878                        pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
4879                                             &cmd);
4880                        if (decode == true)
4881                                cmd |= PCI_BRIDGE_CTL_VGA;
4882                        else
4883                                cmd &= ~PCI_BRIDGE_CTL_VGA;
4884                        pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
4885                                              cmd);
4886                }
4887                bus = bus->parent;
4888        }
4889        return 0;
4890}
4891
4892/**
4893 * pci_add_dma_alias - Add a DMA devfn alias for a device
4894 * @dev: the PCI device for which alias is added
4895 * @devfn: alias slot and function
4896 *
4897 * This helper encodes 8-bit devfn as bit number in dma_alias_mask.
4898 * It should be called early, preferably as PCI fixup header quirk.
4899 */
4900void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
4901{
4902        if (!dev->dma_alias_mask)
4903                dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
4904                                              sizeof(long), GFP_KERNEL);
4905        if (!dev->dma_alias_mask) {
4906                dev_warn(&dev->dev, "Unable to allocate DMA alias mask\n");
4907                return;
4908        }
4909
4910        set_bit(devfn, dev->dma_alias_mask);
4911        dev_info(&dev->dev, "Enabling fixed DMA alias to %02x.%d\n",
4912                 PCI_SLOT(devfn), PCI_FUNC(devfn));
4913}
4914
4915bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
4916{
4917        return (dev1->dma_alias_mask &&
4918                test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
4919               (dev2->dma_alias_mask &&
4920                test_bit(dev1->devfn, dev2->dma_alias_mask));
4921}
4922
4923bool pci_device_is_present(struct pci_dev *pdev)
4924{
4925        u32 v;
4926
4927        return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
4928}
4929EXPORT_SYMBOL_GPL(pci_device_is_present);
4930
4931void pci_ignore_hotplug(struct pci_dev *dev)
4932{
4933        struct pci_dev *bridge = dev->bus->self;
4934
4935        dev->ignore_hotplug = 1;
4936        /* Propagate the "ignore hotplug" setting to the parent bridge. */
4937        if (bridge)
4938                bridge->ignore_hotplug = 1;
4939}
4940EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
4941
4942#define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
4943static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
4944static DEFINE_SPINLOCK(resource_alignment_lock);
4945
4946/**
4947 * pci_specified_resource_alignment - get resource alignment specified by user.
4948 * @dev: the PCI device to get
4949 *
4950 * RETURNS: Resource alignment if it is specified.
4951 *          Zero if it is not specified.
4952 */
4953static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev)
4954{
4955        int seg, bus, slot, func, align_order, count;
4956        unsigned short vendor, device, subsystem_vendor, subsystem_device;
4957        resource_size_t align = 0;
4958        char *p;
4959
4960        spin_lock(&resource_alignment_lock);
4961        p = resource_alignment_param;
4962        while (*p) {
4963                count = 0;
4964                if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
4965                                                        p[count] == '@') {
4966                        p += count + 1;
4967                } else {
4968                        align_order = -1;
4969                }
4970                if (strncmp(p, "pci:", 4) == 0) {
4971                        /* PCI vendor/device (subvendor/subdevice) ids are specified */
4972                        p += 4;
4973                        if (sscanf(p, "%hx:%hx:%hx:%hx%n",
4974                                &vendor, &device, &subsystem_vendor, &subsystem_device, &count) != 4) {
4975                                if (sscanf(p, "%hx:%hx%n", &vendor, &device, &count) != 2) {
4976                                        printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: pci:%s\n",
4977                                                p);
4978                                        break;
4979                                }
4980                                subsystem_vendor = subsystem_device = 0;
4981                        }
4982                        p += count;
4983                        if ((!vendor || (vendor == dev->vendor)) &&
4984                                (!device || (device == dev->device)) &&
4985                                (!subsystem_vendor || (subsystem_vendor == dev->subsystem_vendor)) &&
4986                                (!subsystem_device || (subsystem_device == dev->subsystem_device))) {
4987                                if (align_order == -1)
4988                                        align = PAGE_SIZE;
4989                                else
4990                                        align = 1 << align_order;
4991                                /* Found */
4992                                break;
4993                        }
4994                }
4995                else {
4996                        if (sscanf(p, "%x:%x:%x.%x%n",
4997                                &seg, &bus, &slot, &func, &count) != 4) {
4998                                seg = 0;
4999                                if (sscanf(p, "%x:%x.%x%n",
5000                                                &bus, &slot, &func, &count) != 3) {
5001                                        /* Invalid format */
5002                                        printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
5003                                                p);
5004                                        break;
5005                                }
5006                        }
5007                        p += count;
5008                        if (seg == pci_domain_nr(dev->bus) &&
5009                                bus == dev->bus->number &&
5010                                slot == PCI_SLOT(dev->devfn) &&
5011                                func == PCI_FUNC(dev->devfn)) {
5012                                if (align_order == -1)
5013                                        align = PAGE_SIZE;
5014                                else
5015                                        align = 1 << align_order;
5016                                /* Found */
5017                                break;
5018                        }
5019                }
5020                if (*p != ';' && *p != ',') {
5021                        /* End of param or invalid format */
5022                        break;
5023                }
5024                p++;
5025        }
5026        spin_unlock(&resource_alignment_lock);
5027        return align;
5028}
5029
5030/*
5031 * This function disables memory decoding and releases memory resources
5032 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5033 * It also rounds up size to specified alignment.
5034 * Later on, the kernel will assign page-aligned memory resource back
5035 * to the device.
5036 */
5037void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5038{
5039        int i;
5040        struct resource *r;
5041        resource_size_t align, size;
5042        u16 command;
5043
5044        /* check if specified PCI is target device to reassign */
5045        align = pci_specified_resource_alignment(dev);
5046        if (!align)
5047                return;
5048
5049        if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5050            (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5051                dev_warn(&dev->dev,
5052                        "Can't reassign resources to host bridge.\n");
5053                return;
5054        }
5055
5056        dev_info(&dev->dev,
5057                "Disabling memory decoding and releasing memory resources.\n");
5058        pci_read_config_word(dev, PCI_COMMAND, &command);
5059        command &= ~PCI_COMMAND_MEMORY;
5060        pci_write_config_word(dev, PCI_COMMAND, command);
5061
5062        for (i = 0; i < PCI_BRIDGE_RESOURCES; i++) {
5063                r = &dev->resource[i];
5064                if (!(r->flags & IORESOURCE_MEM))
5065                        continue;
5066                size = resource_size(r);
5067                if (size < align) {
5068                        size = align;
5069                        dev_info(&dev->dev,
5070                                "Rounding up size of resource #%d to %#llx.\n",
5071                                i, (unsigned long long)size);
5072                }
5073                r->flags |= IORESOURCE_UNSET;
5074                r->end = size - 1;
5075                r->start = 0;
5076        }
5077        /* Need to disable bridge's resource window,
5078         * to enable the kernel to reassign new resource
5079         * window later on.
5080         */
5081        if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5082            (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5083                for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5084                        r = &dev->resource[i];
5085                        if (!(r->flags & IORESOURCE_MEM))
5086                                continue;
5087                        r->flags |= IORESOURCE_UNSET;
5088                        r->end = resource_size(r) - 1;
5089                        r->start = 0;
5090                }
5091                pci_disable_bridge_window(dev);
5092        }
5093}
5094
5095static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5096{
5097        if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
5098                count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
5099        spin_lock(&resource_alignment_lock);
5100        strncpy(resource_alignment_param, buf, count);
5101        resource_alignment_param[count] = '\0';
5102        spin_unlock(&resource_alignment_lock);
5103        return count;
5104}
5105
5106static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
5107{
5108        size_t count;
5109        spin_lock(&resource_alignment_lock);
5110        count = snprintf(buf, size, "%s", resource_alignment_param);
5111        spin_unlock(&resource_alignment_lock);
5112        return count;
5113}
5114
5115static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
5116{
5117        return pci_get_resource_alignment_param(buf, PAGE_SIZE);
5118}
5119
5120static ssize_t pci_resource_alignment_store(struct bus_type *bus,
5121                                        const char *buf, size_t count)
5122{
5123        return pci_set_resource_alignment_param(buf, count);
5124}
5125
5126static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
5127                                        pci_resource_alignment_store);
5128
5129static int __init pci_resource_alignment_sysfs_init(void)
5130{
5131        return bus_create_file(&pci_bus_type,
5132                                        &bus_attr_resource_alignment);
5133}
5134late_initcall(pci_resource_alignment_sysfs_init);
5135
5136static void pci_no_domains(void)
5137{
5138#ifdef CONFIG_PCI_DOMAINS
5139        pci_domains_supported = 0;
5140#endif
5141}
5142
5143#ifdef CONFIG_PCI_DOMAINS
5144static atomic_t __domain_nr = ATOMIC_INIT(-1);
5145
5146int pci_get_new_domain_nr(void)
5147{
5148        return atomic_inc_return(&__domain_nr);
5149}
5150
5151#ifdef CONFIG_PCI_DOMAINS_GENERIC
5152static int of_pci_bus_find_domain_nr(struct device *parent)
5153{
5154        static int use_dt_domains = -1;
5155        int domain = -1;
5156
5157        if (parent)
5158                domain = of_get_pci_domain_nr(parent->of_node);
5159        /*
5160         * Check DT domain and use_dt_domains values.
5161         *
5162         * If DT domain property is valid (domain >= 0) and
5163         * use_dt_domains != 0, the DT assignment is valid since this means
5164         * we have not previously allocated a domain number by using
5165         * pci_get_new_domain_nr(); we should also update use_dt_domains to
5166         * 1, to indicate that we have just assigned a domain number from
5167         * DT.
5168         *
5169         * If DT domain property value is not valid (ie domain < 0), and we
5170         * have not previously assigned a domain number from DT
5171         * (use_dt_domains != 1) we should assign a domain number by
5172         * using the:
5173         *
5174         * pci_get_new_domain_nr()
5175         *
5176         * API and update the use_dt_domains value to keep track of method we
5177         * are using to assign domain numbers (use_dt_domains = 0).
5178         *
5179         * All other combinations imply we have a platform that is trying
5180         * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
5181         * which is a recipe for domain mishandling and it is prevented by
5182         * invalidating the domain value (domain = -1) and printing a
5183         * corresponding error.
5184         */
5185        if (domain >= 0 && use_dt_domains) {
5186                use_dt_domains = 1;
5187        } else if (domain < 0 && use_dt_domains != 1) {
5188                use_dt_domains = 0;
5189                domain = pci_get_new_domain_nr();
5190        } else {
5191                dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
5192                        parent->of_node->full_name);
5193                domain = -1;
5194        }
5195
5196        return domain;
5197}
5198
5199int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
5200{
5201        return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
5202                               acpi_pci_bus_find_domain_nr(bus);
5203}
5204#endif
5205#endif
5206
5207/**
5208 * pci_ext_cfg_avail - can we access extended PCI config space?
5209 *
5210 * Returns 1 if we can access PCI extended config space (offsets
5211 * greater than 0xff). This is the default implementation. Architecture
5212 * implementations can override this.
5213 */
5214int __weak pci_ext_cfg_avail(void)
5215{
5216        return 1;
5217}
5218
5219void __weak pci_fixup_cardbus(struct pci_bus *bus)
5220{
5221}
5222EXPORT_SYMBOL(pci_fixup_cardbus);
5223
5224static int __init pci_setup(char *str)
5225{
5226        while (str) {
5227                char *k = strchr(str, ',');
5228                if (k)
5229                        *k++ = 0;
5230                if (*str && (str = pcibios_setup(str)) && *str) {
5231                        if (!strcmp(str, "nomsi")) {
5232                                pci_no_msi();
5233                        } else if (!strcmp(str, "noaer")) {
5234                                pci_no_aer();
5235                        } else if (!strncmp(str, "realloc=", 8)) {
5236                                pci_realloc_get_opt(str + 8);
5237                        } else if (!strncmp(str, "realloc", 7)) {
5238                                pci_realloc_get_opt("on");
5239                        } else if (!strcmp(str, "nodomains")) {
5240                                pci_no_domains();
5241                        } else if (!strncmp(str, "noari", 5)) {
5242                                pcie_ari_disabled = true;
5243                        } else if (!strncmp(str, "cbiosize=", 9)) {
5244                                pci_cardbus_io_size = memparse(str + 9, &str);
5245                        } else if (!strncmp(str, "cbmemsize=", 10)) {
5246                                pci_cardbus_mem_size = memparse(str + 10, &str);
5247                        } else if (!strncmp(str, "resource_alignment=", 19)) {
5248                                pci_set_resource_alignment_param(str + 19,
5249                                                        strlen(str + 19));
5250                        } else if (!strncmp(str, "ecrc=", 5)) {
5251                                pcie_ecrc_get_policy(str + 5);
5252                        } else if (!strncmp(str, "hpiosize=", 9)) {
5253                                pci_hotplug_io_size = memparse(str + 9, &str);
5254                        } else if (!strncmp(str, "hpmemsize=", 10)) {
5255                                pci_hotplug_mem_size = memparse(str + 10, &str);
5256                        } else if (!strncmp(str, "hpbussize=", 10)) {
5257                                pci_hotplug_bus_size =
5258                                        simple_strtoul(str + 10, &str, 0);
5259                                if (pci_hotplug_bus_size > 0xff)
5260                                        pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
5261                        } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
5262                                pcie_bus_config = PCIE_BUS_TUNE_OFF;
5263                        } else if (!strncmp(str, "pcie_bus_safe", 13)) {
5264                                pcie_bus_config = PCIE_BUS_SAFE;
5265                        } else if (!strncmp(str, "pcie_bus_perf", 13)) {
5266                                pcie_bus_config = PCIE_BUS_PERFORMANCE;
5267                        } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
5268                                pcie_bus_config = PCIE_BUS_PEER2PEER;
5269                        } else if (!strncmp(str, "pcie_scan_all", 13)) {
5270                                pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
5271                        } else {
5272                                printk(KERN_ERR "PCI: Unknown option `%s'\n",
5273                                                str);
5274                        }
5275                }
5276                str = k;
5277        }
5278        return 0;
5279}
5280early_param("pci", pci_setup);
5281