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