linux/drivers/pci/pci.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * PCI Bus Services, see include/linux/pci.h for further explanation.
   4 *
   5 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
   6 * David Mosberger-Tang
   7 *
   8 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
   9 */
  10
  11#include <linux/acpi.h>
  12#include <linux/kernel.h>
  13#include <linux/delay.h>
  14#include <linux/dmi.h>
  15#include <linux/init.h>
  16#include <linux/msi.h>
  17#include <linux/of.h>
  18#include <linux/pci.h>
  19#include <linux/pm.h>
  20#include <linux/slab.h>
  21#include <linux/module.h>
  22#include <linux/spinlock.h>
  23#include <linux/string.h>
  24#include <linux/log2.h>
  25#include <linux/logic_pio.h>
  26#include <linux/pm_wakeup.h>
  27#include <linux/interrupt.h>
  28#include <linux/device.h>
  29#include <linux/pm_runtime.h>
  30#include <linux/pci_hotplug.h>
  31#include <linux/vmalloc.h>
  32#include <asm/dma.h>
  33#include <linux/aer.h>
  34#include <linux/bitfield.h>
  35#include "pci.h"
  36
  37DEFINE_MUTEX(pci_slot_mutex);
  38
  39const char *pci_power_names[] = {
  40        "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
  41};
  42EXPORT_SYMBOL_GPL(pci_power_names);
  43
  44int isa_dma_bridge_buggy;
  45EXPORT_SYMBOL(isa_dma_bridge_buggy);
  46
  47int pci_pci_problems;
  48EXPORT_SYMBOL(pci_pci_problems);
  49
  50unsigned int pci_pm_d3hot_delay;
  51
  52static void pci_pme_list_scan(struct work_struct *work);
  53
  54static LIST_HEAD(pci_pme_list);
  55static DEFINE_MUTEX(pci_pme_list_mutex);
  56static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
  57
  58struct pci_pme_device {
  59        struct list_head list;
  60        struct pci_dev *dev;
  61};
  62
  63#define PME_TIMEOUT 1000 /* How long between PME checks */
  64
  65static void pci_dev_d3_sleep(struct pci_dev *dev)
  66{
  67        unsigned int delay = dev->d3hot_delay;
  68
  69        if (delay < pci_pm_d3hot_delay)
  70                delay = pci_pm_d3hot_delay;
  71
  72        if (delay)
  73                msleep(delay);
  74}
  75
  76bool pci_reset_supported(struct pci_dev *dev)
  77{
  78        return dev->reset_methods[0] != 0;
  79}
  80
  81#ifdef CONFIG_PCI_DOMAINS
  82int pci_domains_supported = 1;
  83#endif
  84
  85#define DEFAULT_CARDBUS_IO_SIZE         (256)
  86#define DEFAULT_CARDBUS_MEM_SIZE        (64*1024*1024)
  87/* pci=cbmemsize=nnM,cbiosize=nn can override this */
  88unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
  89unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
  90
  91#define DEFAULT_HOTPLUG_IO_SIZE         (256)
  92#define DEFAULT_HOTPLUG_MMIO_SIZE       (2*1024*1024)
  93#define DEFAULT_HOTPLUG_MMIO_PREF_SIZE  (2*1024*1024)
  94/* hpiosize=nn can override this */
  95unsigned long pci_hotplug_io_size  = DEFAULT_HOTPLUG_IO_SIZE;
  96/*
  97 * pci=hpmmiosize=nnM overrides non-prefetchable MMIO size,
  98 * pci=hpmmioprefsize=nnM overrides prefetchable MMIO size;
  99 * pci=hpmemsize=nnM overrides both
 100 */
 101unsigned long pci_hotplug_mmio_size = DEFAULT_HOTPLUG_MMIO_SIZE;
 102unsigned long pci_hotplug_mmio_pref_size = DEFAULT_HOTPLUG_MMIO_PREF_SIZE;
 103
 104#define DEFAULT_HOTPLUG_BUS_SIZE        1
 105unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
 106
 107
 108/* PCIe MPS/MRRS strategy; can be overridden by kernel command-line param */
 109#ifdef CONFIG_PCIE_BUS_TUNE_OFF
 110enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_TUNE_OFF;
 111#elif defined CONFIG_PCIE_BUS_SAFE
 112enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_SAFE;
 113#elif defined CONFIG_PCIE_BUS_PERFORMANCE
 114enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_PERFORMANCE;
 115#elif defined CONFIG_PCIE_BUS_PEER2PEER
 116enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_PEER2PEER;
 117#else
 118enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
 119#endif
 120
 121/*
 122 * The default CLS is used if arch didn't set CLS explicitly and not
 123 * all pci devices agree on the same value.  Arch can override either
 124 * the dfl or actual value as it sees fit.  Don't forget this is
 125 * measured in 32-bit words, not bytes.
 126 */
 127u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
 128u8 pci_cache_line_size;
 129
 130/*
 131 * If we set up a device for bus mastering, we need to check the latency
 132 * timer as certain BIOSes forget to set it properly.
 133 */
 134unsigned int pcibios_max_latency = 255;
 135
 136/* If set, the PCIe ARI capability will not be used. */
 137static bool pcie_ari_disabled;
 138
 139/* If set, the PCIe ATS capability will not be used. */
 140static bool pcie_ats_disabled;
 141
 142/* If set, the PCI config space of each device is printed during boot. */
 143bool pci_early_dump;
 144
 145bool pci_ats_disabled(void)
 146{
 147        return pcie_ats_disabled;
 148}
 149EXPORT_SYMBOL_GPL(pci_ats_disabled);
 150
 151/* Disable bridge_d3 for all PCIe ports */
 152static bool pci_bridge_d3_disable;
 153/* Force bridge_d3 for all PCIe ports */
 154static bool pci_bridge_d3_force;
 155
 156static int __init pcie_port_pm_setup(char *str)
 157{
 158        if (!strcmp(str, "off"))
 159                pci_bridge_d3_disable = true;
 160        else if (!strcmp(str, "force"))
 161                pci_bridge_d3_force = true;
 162        return 1;
 163}
 164__setup("pcie_port_pm=", pcie_port_pm_setup);
 165
 166/* Time to wait after a reset for device to become responsive */
 167#define PCIE_RESET_READY_POLL_MS 60000
 168
 169/**
 170 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
 171 * @bus: pointer to PCI bus structure to search
 172 *
 173 * Given a PCI bus, returns the highest PCI bus number present in the set
 174 * including the given PCI bus and its list of child PCI buses.
 175 */
 176unsigned char pci_bus_max_busnr(struct pci_bus *bus)
 177{
 178        struct pci_bus *tmp;
 179        unsigned char max, n;
 180
 181        max = bus->busn_res.end;
 182        list_for_each_entry(tmp, &bus->children, node) {
 183                n = pci_bus_max_busnr(tmp);
 184                if (n > max)
 185                        max = n;
 186        }
 187        return max;
 188}
 189EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
 190
 191/**
 192 * pci_status_get_and_clear_errors - return and clear error bits in PCI_STATUS
 193 * @pdev: the PCI device
 194 *
 195 * Returns error bits set in PCI_STATUS and clears them.
 196 */
 197int pci_status_get_and_clear_errors(struct pci_dev *pdev)
 198{
 199        u16 status;
 200        int ret;
 201
 202        ret = pci_read_config_word(pdev, PCI_STATUS, &status);
 203        if (ret != PCIBIOS_SUCCESSFUL)
 204                return -EIO;
 205
 206        status &= PCI_STATUS_ERROR_BITS;
 207        if (status)
 208                pci_write_config_word(pdev, PCI_STATUS, status);
 209
 210        return status;
 211}
 212EXPORT_SYMBOL_GPL(pci_status_get_and_clear_errors);
 213
 214#ifdef CONFIG_HAS_IOMEM
 215static void __iomem *__pci_ioremap_resource(struct pci_dev *pdev, int bar,
 216                                            bool write_combine)
 217{
 218        struct resource *res = &pdev->resource[bar];
 219        resource_size_t start = res->start;
 220        resource_size_t size = resource_size(res);
 221
 222        /*
 223         * Make sure the BAR is actually a memory resource, not an IO resource
 224         */
 225        if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
 226                pci_err(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
 227                return NULL;
 228        }
 229
 230        if (write_combine)
 231                return ioremap_wc(start, size);
 232
 233        return ioremap(start, size);
 234}
 235
 236void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
 237{
 238        return __pci_ioremap_resource(pdev, bar, false);
 239}
 240EXPORT_SYMBOL_GPL(pci_ioremap_bar);
 241
 242void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
 243{
 244        return __pci_ioremap_resource(pdev, bar, true);
 245}
 246EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
 247#endif
 248
 249/**
 250 * pci_dev_str_match_path - test if a path string matches a device
 251 * @dev: the PCI device to test
 252 * @path: string to match the device against
 253 * @endptr: pointer to the string after the match
 254 *
 255 * Test if a string (typically from a kernel parameter) formatted as a
 256 * path of device/function addresses matches a PCI device. The string must
 257 * be of the form:
 258 *
 259 *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
 260 *
 261 * A path for a device can be obtained using 'lspci -t'.  Using a path
 262 * is more robust against bus renumbering than using only a single bus,
 263 * device and function address.
 264 *
 265 * Returns 1 if the string matches the device, 0 if it does not and
 266 * a negative error code if it fails to parse the string.
 267 */
 268static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
 269                                  const char **endptr)
 270{
 271        int ret;
 272        int seg, bus, slot, func;
 273        char *wpath, *p;
 274        char end;
 275
 276        *endptr = strchrnul(path, ';');
 277
 278        wpath = kmemdup_nul(path, *endptr - path, GFP_ATOMIC);
 279        if (!wpath)
 280                return -ENOMEM;
 281
 282        while (1) {
 283                p = strrchr(wpath, '/');
 284                if (!p)
 285                        break;
 286                ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
 287                if (ret != 2) {
 288                        ret = -EINVAL;
 289                        goto free_and_exit;
 290                }
 291
 292                if (dev->devfn != PCI_DEVFN(slot, func)) {
 293                        ret = 0;
 294                        goto free_and_exit;
 295                }
 296
 297                /*
 298                 * Note: we don't need to get a reference to the upstream
 299                 * bridge because we hold a reference to the top level
 300                 * device which should hold a reference to the bridge,
 301                 * and so on.
 302                 */
 303                dev = pci_upstream_bridge(dev);
 304                if (!dev) {
 305                        ret = 0;
 306                        goto free_and_exit;
 307                }
 308
 309                *p = 0;
 310        }
 311
 312        ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
 313                     &func, &end);
 314        if (ret != 4) {
 315                seg = 0;
 316                ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
 317                if (ret != 3) {
 318                        ret = -EINVAL;
 319                        goto free_and_exit;
 320                }
 321        }
 322
 323        ret = (seg == pci_domain_nr(dev->bus) &&
 324               bus == dev->bus->number &&
 325               dev->devfn == PCI_DEVFN(slot, func));
 326
 327free_and_exit:
 328        kfree(wpath);
 329        return ret;
 330}
 331
 332/**
 333 * pci_dev_str_match - test if a string matches a device
 334 * @dev: the PCI device to test
 335 * @p: string to match the device against
 336 * @endptr: pointer to the string after the match
 337 *
 338 * Test if a string (typically from a kernel parameter) matches a specified
 339 * PCI device. The string may be of one of the following formats:
 340 *
 341 *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
 342 *   pci:<vendor>:<device>[:<subvendor>:<subdevice>]
 343 *
 344 * The first format specifies a PCI bus/device/function address which
 345 * may change if new hardware is inserted, if motherboard firmware changes,
 346 * or due to changes caused in kernel parameters. If the domain is
 347 * left unspecified, it is taken to be 0.  In order to be robust against
 348 * bus renumbering issues, a path of PCI device/function numbers may be used
 349 * to address the specific device.  The path for a device can be determined
 350 * through the use of 'lspci -t'.
 351 *
 352 * The second format matches devices using IDs in the configuration
 353 * space which may match multiple devices in the system. A value of 0
 354 * for any field will match all devices. (Note: this differs from
 355 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
 356 * legacy reasons and convenience so users don't have to specify
 357 * FFFFFFFFs on the command line.)
 358 *
 359 * Returns 1 if the string matches the device, 0 if it does not and
 360 * a negative error code if the string cannot be parsed.
 361 */
 362static int pci_dev_str_match(struct pci_dev *dev, const char *p,
 363                             const char **endptr)
 364{
 365        int ret;
 366        int count;
 367        unsigned short vendor, device, subsystem_vendor, subsystem_device;
 368
 369        if (strncmp(p, "pci:", 4) == 0) {
 370                /* PCI vendor/device (subvendor/subdevice) IDs are specified */
 371                p += 4;
 372                ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
 373                             &subsystem_vendor, &subsystem_device, &count);
 374                if (ret != 4) {
 375                        ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
 376                        if (ret != 2)
 377                                return -EINVAL;
 378
 379                        subsystem_vendor = 0;
 380                        subsystem_device = 0;
 381                }
 382
 383                p += count;
 384
 385                if ((!vendor || vendor == dev->vendor) &&
 386                    (!device || device == dev->device) &&
 387                    (!subsystem_vendor ||
 388                            subsystem_vendor == dev->subsystem_vendor) &&
 389                    (!subsystem_device ||
 390                            subsystem_device == dev->subsystem_device))
 391                        goto found;
 392        } else {
 393                /*
 394                 * PCI Bus, Device, Function IDs are specified
 395                 * (optionally, may include a path of devfns following it)
 396                 */
 397                ret = pci_dev_str_match_path(dev, p, &p);
 398                if (ret < 0)
 399                        return ret;
 400                else if (ret)
 401                        goto found;
 402        }
 403
 404        *endptr = p;
 405        return 0;
 406
 407found:
 408        *endptr = p;
 409        return 1;
 410}
 411
 412static u8 __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
 413                                  u8 pos, int cap, int *ttl)
 414{
 415        u8 id;
 416        u16 ent;
 417
 418        pci_bus_read_config_byte(bus, devfn, pos, &pos);
 419
 420        while ((*ttl)--) {
 421                if (pos < 0x40)
 422                        break;
 423                pos &= ~3;
 424                pci_bus_read_config_word(bus, devfn, pos, &ent);
 425
 426                id = ent & 0xff;
 427                if (id == 0xff)
 428                        break;
 429                if (id == cap)
 430                        return pos;
 431                pos = (ent >> 8);
 432        }
 433        return 0;
 434}
 435
 436static u8 __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
 437                              u8 pos, int cap)
 438{
 439        int ttl = PCI_FIND_CAP_TTL;
 440
 441        return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
 442}
 443
 444u8 pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
 445{
 446        return __pci_find_next_cap(dev->bus, dev->devfn,
 447                                   pos + PCI_CAP_LIST_NEXT, cap);
 448}
 449EXPORT_SYMBOL_GPL(pci_find_next_capability);
 450
 451static u8 __pci_bus_find_cap_start(struct pci_bus *bus,
 452                                    unsigned int devfn, u8 hdr_type)
 453{
 454        u16 status;
 455
 456        pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
 457        if (!(status & PCI_STATUS_CAP_LIST))
 458                return 0;
 459
 460        switch (hdr_type) {
 461        case PCI_HEADER_TYPE_NORMAL:
 462        case PCI_HEADER_TYPE_BRIDGE:
 463                return PCI_CAPABILITY_LIST;
 464        case PCI_HEADER_TYPE_CARDBUS:
 465                return PCI_CB_CAPABILITY_LIST;
 466        }
 467
 468        return 0;
 469}
 470
 471/**
 472 * pci_find_capability - query for devices' capabilities
 473 * @dev: PCI device to query
 474 * @cap: capability code
 475 *
 476 * Tell if a device supports a given PCI capability.
 477 * Returns the address of the requested capability structure within the
 478 * device's PCI configuration space or 0 in case the device does not
 479 * support it.  Possible values for @cap include:
 480 *
 481 *  %PCI_CAP_ID_PM           Power Management
 482 *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
 483 *  %PCI_CAP_ID_VPD          Vital Product Data
 484 *  %PCI_CAP_ID_SLOTID       Slot Identification
 485 *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
 486 *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
 487 *  %PCI_CAP_ID_PCIX         PCI-X
 488 *  %PCI_CAP_ID_EXP          PCI Express
 489 */
 490u8 pci_find_capability(struct pci_dev *dev, int cap)
 491{
 492        u8 pos;
 493
 494        pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 495        if (pos)
 496                pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
 497
 498        return pos;
 499}
 500EXPORT_SYMBOL(pci_find_capability);
 501
 502/**
 503 * pci_bus_find_capability - query for devices' capabilities
 504 * @bus: the PCI bus to query
 505 * @devfn: PCI device to query
 506 * @cap: capability code
 507 *
 508 * Like pci_find_capability() but works for PCI devices that do not have a
 509 * pci_dev structure set up yet.
 510 *
 511 * Returns the address of the requested capability structure within the
 512 * device's PCI configuration space or 0 in case the device does not
 513 * support it.
 514 */
 515u8 pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
 516{
 517        u8 hdr_type, pos;
 518
 519        pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
 520
 521        pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
 522        if (pos)
 523                pos = __pci_find_next_cap(bus, devfn, pos, cap);
 524
 525        return pos;
 526}
 527EXPORT_SYMBOL(pci_bus_find_capability);
 528
 529/**
 530 * pci_find_next_ext_capability - Find an extended capability
 531 * @dev: PCI device to query
 532 * @start: address at which to start looking (0 to start at beginning of list)
 533 * @cap: capability code
 534 *
 535 * Returns the address of the next matching extended capability structure
 536 * within the device's PCI configuration space or 0 if the device does
 537 * not support it.  Some capabilities can occur several times, e.g., the
 538 * vendor-specific capability, and this provides a way to find them all.
 539 */
 540u16 pci_find_next_ext_capability(struct pci_dev *dev, u16 start, int cap)
 541{
 542        u32 header;
 543        int ttl;
 544        u16 pos = PCI_CFG_SPACE_SIZE;
 545
 546        /* minimum 8 bytes per capability */
 547        ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
 548
 549        if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
 550                return 0;
 551
 552        if (start)
 553                pos = start;
 554
 555        if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 556                return 0;
 557
 558        /*
 559         * If we have no capabilities, this is indicated by cap ID,
 560         * cap version and next pointer all being 0.
 561         */
 562        if (header == 0)
 563                return 0;
 564
 565        while (ttl-- > 0) {
 566                if (PCI_EXT_CAP_ID(header) == cap && pos != start)
 567                        return pos;
 568
 569                pos = PCI_EXT_CAP_NEXT(header);
 570                if (pos < PCI_CFG_SPACE_SIZE)
 571                        break;
 572
 573                if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
 574                        break;
 575        }
 576
 577        return 0;
 578}
 579EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
 580
 581/**
 582 * pci_find_ext_capability - Find an extended capability
 583 * @dev: PCI device to query
 584 * @cap: capability code
 585 *
 586 * Returns the address of the requested extended capability structure
 587 * within the device's PCI configuration space or 0 if the device does
 588 * not support it.  Possible values for @cap include:
 589 *
 590 *  %PCI_EXT_CAP_ID_ERR         Advanced Error Reporting
 591 *  %PCI_EXT_CAP_ID_VC          Virtual Channel
 592 *  %PCI_EXT_CAP_ID_DSN         Device Serial Number
 593 *  %PCI_EXT_CAP_ID_PWR         Power Budgeting
 594 */
 595u16 pci_find_ext_capability(struct pci_dev *dev, int cap)
 596{
 597        return pci_find_next_ext_capability(dev, 0, cap);
 598}
 599EXPORT_SYMBOL_GPL(pci_find_ext_capability);
 600
 601/**
 602 * pci_get_dsn - Read and return the 8-byte Device Serial Number
 603 * @dev: PCI device to query
 604 *
 605 * Looks up the PCI_EXT_CAP_ID_DSN and reads the 8 bytes of the Device Serial
 606 * Number.
 607 *
 608 * Returns the DSN, or zero if the capability does not exist.
 609 */
 610u64 pci_get_dsn(struct pci_dev *dev)
 611{
 612        u32 dword;
 613        u64 dsn;
 614        int pos;
 615
 616        pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DSN);
 617        if (!pos)
 618                return 0;
 619
 620        /*
 621         * The Device Serial Number is two dwords offset 4 bytes from the
 622         * capability position. The specification says that the first dword is
 623         * the lower half, and the second dword is the upper half.
 624         */
 625        pos += 4;
 626        pci_read_config_dword(dev, pos, &dword);
 627        dsn = (u64)dword;
 628        pci_read_config_dword(dev, pos + 4, &dword);
 629        dsn |= ((u64)dword) << 32;
 630
 631        return dsn;
 632}
 633EXPORT_SYMBOL_GPL(pci_get_dsn);
 634
 635static u8 __pci_find_next_ht_cap(struct pci_dev *dev, u8 pos, int ht_cap)
 636{
 637        int rc, ttl = PCI_FIND_CAP_TTL;
 638        u8 cap, mask;
 639
 640        if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
 641                mask = HT_3BIT_CAP_MASK;
 642        else
 643                mask = HT_5BIT_CAP_MASK;
 644
 645        pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
 646                                      PCI_CAP_ID_HT, &ttl);
 647        while (pos) {
 648                rc = pci_read_config_byte(dev, pos + 3, &cap);
 649                if (rc != PCIBIOS_SUCCESSFUL)
 650                        return 0;
 651
 652                if ((cap & mask) == ht_cap)
 653                        return pos;
 654
 655                pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
 656                                              pos + PCI_CAP_LIST_NEXT,
 657                                              PCI_CAP_ID_HT, &ttl);
 658        }
 659
 660        return 0;
 661}
 662
 663/**
 664 * pci_find_next_ht_capability - query a device's HyperTransport capabilities
 665 * @dev: PCI device to query
 666 * @pos: Position from which to continue searching
 667 * @ht_cap: HyperTransport capability code
 668 *
 669 * To be used in conjunction with pci_find_ht_capability() to search for
 670 * all capabilities matching @ht_cap. @pos should always be a value returned
 671 * from pci_find_ht_capability().
 672 *
 673 * NB. To be 100% safe against broken PCI devices, the caller should take
 674 * steps to avoid an infinite loop.
 675 */
 676u8 pci_find_next_ht_capability(struct pci_dev *dev, u8 pos, int ht_cap)
 677{
 678        return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
 679}
 680EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
 681
 682/**
 683 * pci_find_ht_capability - query a device's HyperTransport capabilities
 684 * @dev: PCI device to query
 685 * @ht_cap: HyperTransport capability code
 686 *
 687 * Tell if a device supports a given HyperTransport capability.
 688 * Returns an address within the device's PCI configuration space
 689 * or 0 in case the device does not support the request capability.
 690 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
 691 * which has a HyperTransport capability matching @ht_cap.
 692 */
 693u8 pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
 694{
 695        u8 pos;
 696
 697        pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
 698        if (pos)
 699                pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
 700
 701        return pos;
 702}
 703EXPORT_SYMBOL_GPL(pci_find_ht_capability);
 704
 705/**
 706 * pci_find_vsec_capability - Find a vendor-specific extended capability
 707 * @dev: PCI device to query
 708 * @vendor: Vendor ID for which capability is defined
 709 * @cap: Vendor-specific capability ID
 710 *
 711 * If @dev has Vendor ID @vendor, search for a VSEC capability with
 712 * VSEC ID @cap. If found, return the capability offset in
 713 * config space; otherwise return 0.
 714 */
 715u16 pci_find_vsec_capability(struct pci_dev *dev, u16 vendor, int cap)
 716{
 717        u16 vsec = 0;
 718        u32 header;
 719
 720        if (vendor != dev->vendor)
 721                return 0;
 722
 723        while ((vsec = pci_find_next_ext_capability(dev, vsec,
 724                                                     PCI_EXT_CAP_ID_VNDR))) {
 725                if (pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER,
 726                                          &header) == PCIBIOS_SUCCESSFUL &&
 727                    PCI_VNDR_HEADER_ID(header) == cap)
 728                        return vsec;
 729        }
 730
 731        return 0;
 732}
 733EXPORT_SYMBOL_GPL(pci_find_vsec_capability);
 734
 735/**
 736 * pci_find_parent_resource - return resource region of parent bus of given
 737 *                            region
 738 * @dev: PCI device structure contains resources to be searched
 739 * @res: child resource record for which parent is sought
 740 *
 741 * For given resource region of given device, return the resource region of
 742 * parent bus the given region is contained in.
 743 */
 744struct resource *pci_find_parent_resource(const struct pci_dev *dev,
 745                                          struct resource *res)
 746{
 747        const struct pci_bus *bus = dev->bus;
 748        struct resource *r;
 749        int i;
 750
 751        pci_bus_for_each_resource(bus, r, i) {
 752                if (!r)
 753                        continue;
 754                if (resource_contains(r, res)) {
 755
 756                        /*
 757                         * If the window is prefetchable but the BAR is
 758                         * not, the allocator made a mistake.
 759                         */
 760                        if (r->flags & IORESOURCE_PREFETCH &&
 761                            !(res->flags & IORESOURCE_PREFETCH))
 762                                return NULL;
 763
 764                        /*
 765                         * If we're below a transparent bridge, there may
 766                         * be both a positively-decoded aperture and a
 767                         * subtractively-decoded region that contain the BAR.
 768                         * We want the positively-decoded one, so this depends
 769                         * on pci_bus_for_each_resource() giving us those
 770                         * first.
 771                         */
 772                        return r;
 773                }
 774        }
 775        return NULL;
 776}
 777EXPORT_SYMBOL(pci_find_parent_resource);
 778
 779/**
 780 * pci_find_resource - Return matching PCI device resource
 781 * @dev: PCI device to query
 782 * @res: Resource to look for
 783 *
 784 * Goes over standard PCI resources (BARs) and checks if the given resource
 785 * is partially or fully contained in any of them. In that case the
 786 * matching resource is returned, %NULL otherwise.
 787 */
 788struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
 789{
 790        int i;
 791
 792        for (i = 0; i < PCI_STD_NUM_BARS; i++) {
 793                struct resource *r = &dev->resource[i];
 794
 795                if (r->start && resource_contains(r, res))
 796                        return r;
 797        }
 798
 799        return NULL;
 800}
 801EXPORT_SYMBOL(pci_find_resource);
 802
 803/**
 804 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
 805 * @dev: the PCI device to operate on
 806 * @pos: config space offset of status word
 807 * @mask: mask of bit(s) to care about in status word
 808 *
 809 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
 810 */
 811int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
 812{
 813        int i;
 814
 815        /* Wait for Transaction Pending bit clean */
 816        for (i = 0; i < 4; i++) {
 817                u16 status;
 818                if (i)
 819                        msleep((1 << (i - 1)) * 100);
 820
 821                pci_read_config_word(dev, pos, &status);
 822                if (!(status & mask))
 823                        return 1;
 824        }
 825
 826        return 0;
 827}
 828
 829static int pci_acs_enable;
 830
 831/**
 832 * pci_request_acs - ask for ACS to be enabled if supported
 833 */
 834void pci_request_acs(void)
 835{
 836        pci_acs_enable = 1;
 837}
 838
 839static const char *disable_acs_redir_param;
 840
 841/**
 842 * pci_disable_acs_redir - disable ACS redirect capabilities
 843 * @dev: the PCI device
 844 *
 845 * For only devices specified in the disable_acs_redir parameter.
 846 */
 847static void pci_disable_acs_redir(struct pci_dev *dev)
 848{
 849        int ret = 0;
 850        const char *p;
 851        int pos;
 852        u16 ctrl;
 853
 854        if (!disable_acs_redir_param)
 855                return;
 856
 857        p = disable_acs_redir_param;
 858        while (*p) {
 859                ret = pci_dev_str_match(dev, p, &p);
 860                if (ret < 0) {
 861                        pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
 862                                     disable_acs_redir_param);
 863
 864                        break;
 865                } else if (ret == 1) {
 866                        /* Found a match */
 867                        break;
 868                }
 869
 870                if (*p != ';' && *p != ',') {
 871                        /* End of param or invalid format */
 872                        break;
 873                }
 874                p++;
 875        }
 876
 877        if (ret != 1)
 878                return;
 879
 880        if (!pci_dev_specific_disable_acs_redir(dev))
 881                return;
 882
 883        pos = dev->acs_cap;
 884        if (!pos) {
 885                pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
 886                return;
 887        }
 888
 889        pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
 890
 891        /* P2P Request & Completion Redirect */
 892        ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
 893
 894        pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
 895
 896        pci_info(dev, "disabled ACS redirect\n");
 897}
 898
 899/**
 900 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
 901 * @dev: the PCI device
 902 */
 903static void pci_std_enable_acs(struct pci_dev *dev)
 904{
 905        int pos;
 906        u16 cap;
 907        u16 ctrl;
 908
 909        pos = dev->acs_cap;
 910        if (!pos)
 911                return;
 912
 913        pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
 914        pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
 915
 916        /* Source Validation */
 917        ctrl |= (cap & PCI_ACS_SV);
 918
 919        /* P2P Request Redirect */
 920        ctrl |= (cap & PCI_ACS_RR);
 921
 922        /* P2P Completion Redirect */
 923        ctrl |= (cap & PCI_ACS_CR);
 924
 925        /* Upstream Forwarding */
 926        ctrl |= (cap & PCI_ACS_UF);
 927
 928        /* Enable Translation Blocking for external devices and noats */
 929        if (pci_ats_disabled() || dev->external_facing || dev->untrusted)
 930                ctrl |= (cap & PCI_ACS_TB);
 931
 932        pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
 933}
 934
 935/**
 936 * pci_enable_acs - enable ACS if hardware support it
 937 * @dev: the PCI device
 938 */
 939static void pci_enable_acs(struct pci_dev *dev)
 940{
 941        if (!pci_acs_enable)
 942                goto disable_acs_redir;
 943
 944        if (!pci_dev_specific_enable_acs(dev))
 945                goto disable_acs_redir;
 946
 947        pci_std_enable_acs(dev);
 948
 949disable_acs_redir:
 950        /*
 951         * Note: pci_disable_acs_redir() must be called even if ACS was not
 952         * enabled by the kernel because it may have been enabled by
 953         * platform firmware.  So if we are told to disable it, we should
 954         * always disable it after setting the kernel's default
 955         * preferences.
 956         */
 957        pci_disable_acs_redir(dev);
 958}
 959
 960/**
 961 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
 962 * @dev: PCI device to have its BARs restored
 963 *
 964 * Restore the BAR values for a given device, so as to make it
 965 * accessible by its driver.
 966 */
 967static void pci_restore_bars(struct pci_dev *dev)
 968{
 969        int i;
 970
 971        for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
 972                pci_update_resource(dev, i);
 973}
 974
 975static const struct pci_platform_pm_ops *pci_platform_pm;
 976
 977int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
 978{
 979        if (!ops->is_manageable || !ops->set_state  || !ops->get_state ||
 980            !ops->choose_state  || !ops->set_wakeup || !ops->need_resume)
 981                return -EINVAL;
 982        pci_platform_pm = ops;
 983        return 0;
 984}
 985
 986static inline bool platform_pci_power_manageable(struct pci_dev *dev)
 987{
 988        return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
 989}
 990
 991static inline int platform_pci_set_power_state(struct pci_dev *dev,
 992                                               pci_power_t t)
 993{
 994        return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
 995}
 996
 997static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
 998{
 999        return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
1000}
1001
1002static inline void platform_pci_refresh_power_state(struct pci_dev *dev)
1003{
1004        if (pci_platform_pm && pci_platform_pm->refresh_state)
1005                pci_platform_pm->refresh_state(dev);
1006}
1007
1008static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
1009{
1010        return pci_platform_pm ?
1011                        pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
1012}
1013
1014static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
1015{
1016        return pci_platform_pm ?
1017                        pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
1018}
1019
1020static inline bool platform_pci_need_resume(struct pci_dev *dev)
1021{
1022        return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
1023}
1024
1025static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
1026{
1027        if (pci_platform_pm && pci_platform_pm->bridge_d3)
1028                return pci_platform_pm->bridge_d3(dev);
1029        return false;
1030}
1031
1032/**
1033 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
1034 *                           given PCI device
1035 * @dev: PCI device to handle.
1036 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1037 *
1038 * RETURN VALUE:
1039 * -EINVAL if the requested state is invalid.
1040 * -EIO if device does not support PCI PM or its PM capabilities register has a
1041 * wrong version, or device doesn't support the requested state.
1042 * 0 if device already is in the requested state.
1043 * 0 if device's power state has been successfully changed.
1044 */
1045static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
1046{
1047        u16 pmcsr;
1048        bool need_restore = false;
1049
1050        /* Check if we're already there */
1051        if (dev->current_state == state)
1052                return 0;
1053
1054        if (!dev->pm_cap)
1055                return -EIO;
1056
1057        if (state < PCI_D0 || state > PCI_D3hot)
1058                return -EINVAL;
1059
1060        /*
1061         * Validate transition: We can enter D0 from any state, but if
1062         * we're already in a low-power state, we can only go deeper.  E.g.,
1063         * we can go from D1 to D3, but we can't go directly from D3 to D1;
1064         * we'd have to go from D3 to D0, then to D1.
1065         */
1066        if (state != PCI_D0 && dev->current_state <= PCI_D3cold
1067            && dev->current_state > state) {
1068                pci_err(dev, "invalid power transition (from %s to %s)\n",
1069                        pci_power_name(dev->current_state),
1070                        pci_power_name(state));
1071                return -EINVAL;
1072        }
1073
1074        /* Check if this device supports the desired state */
1075        if ((state == PCI_D1 && !dev->d1_support)
1076           || (state == PCI_D2 && !dev->d2_support))
1077                return -EIO;
1078
1079        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1080        if (pmcsr == (u16) ~0) {
1081                pci_err(dev, "can't change power state from %s to %s (config space inaccessible)\n",
1082                        pci_power_name(dev->current_state),
1083                        pci_power_name(state));
1084                return -EIO;
1085        }
1086
1087        /*
1088         * If we're (effectively) in D3, force entire word to 0.
1089         * This doesn't affect PME_Status, disables PME_En, and
1090         * sets PowerState to 0.
1091         */
1092        switch (dev->current_state) {
1093        case PCI_D0:
1094        case PCI_D1:
1095        case PCI_D2:
1096                pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1097                pmcsr |= state;
1098                break;
1099        case PCI_D3hot:
1100        case PCI_D3cold:
1101        case PCI_UNKNOWN: /* Boot-up */
1102                if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
1103                 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
1104                        need_restore = true;
1105                fallthrough;    /* force to D0 */
1106        default:
1107                pmcsr = 0;
1108                break;
1109        }
1110
1111        /* Enter specified state */
1112        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1113
1114        /*
1115         * Mandatory power management transition delays; see PCI PM 1.1
1116         * 5.6.1 table 18
1117         */
1118        if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
1119                pci_dev_d3_sleep(dev);
1120        else if (state == PCI_D2 || dev->current_state == PCI_D2)
1121                udelay(PCI_PM_D2_DELAY);
1122
1123        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1124        dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1125        if (dev->current_state != state)
1126                pci_info_ratelimited(dev, "refused to change power state from %s to %s\n",
1127                         pci_power_name(dev->current_state),
1128                         pci_power_name(state));
1129
1130        /*
1131         * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
1132         * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
1133         * from D3hot to D0 _may_ perform an internal reset, thereby
1134         * going to "D0 Uninitialized" rather than "D0 Initialized".
1135         * For example, at least some versions of the 3c905B and the
1136         * 3c556B exhibit this behaviour.
1137         *
1138         * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
1139         * devices in a D3hot state at boot.  Consequently, we need to
1140         * restore at least the BARs so that the device will be
1141         * accessible to its driver.
1142         */
1143        if (need_restore)
1144                pci_restore_bars(dev);
1145
1146        if (dev->bus->self)
1147                pcie_aspm_pm_state_change(dev->bus->self);
1148
1149        return 0;
1150}
1151
1152/**
1153 * pci_update_current_state - Read power state of given device and cache it
1154 * @dev: PCI device to handle.
1155 * @state: State to cache in case the device doesn't have the PM capability
1156 *
1157 * The power state is read from the PMCSR register, which however is
1158 * inaccessible in D3cold.  The platform firmware is therefore queried first
1159 * to detect accessibility of the register.  In case the platform firmware
1160 * reports an incorrect state or the device isn't power manageable by the
1161 * platform at all, we try to detect D3cold by testing accessibility of the
1162 * vendor ID in config space.
1163 */
1164void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
1165{
1166        if (platform_pci_get_power_state(dev) == PCI_D3cold ||
1167            !pci_device_is_present(dev)) {
1168                dev->current_state = PCI_D3cold;
1169        } else if (dev->pm_cap) {
1170                u16 pmcsr;
1171
1172                pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1173                dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1174        } else {
1175                dev->current_state = state;
1176        }
1177}
1178
1179/**
1180 * pci_refresh_power_state - Refresh the given device's power state data
1181 * @dev: Target PCI device.
1182 *
1183 * Ask the platform to refresh the devices power state information and invoke
1184 * pci_update_current_state() to update its current PCI power state.
1185 */
1186void pci_refresh_power_state(struct pci_dev *dev)
1187{
1188        if (platform_pci_power_manageable(dev))
1189                platform_pci_refresh_power_state(dev);
1190
1191        pci_update_current_state(dev, dev->current_state);
1192}
1193
1194/**
1195 * pci_platform_power_transition - Use platform to change device power state
1196 * @dev: PCI device to handle.
1197 * @state: State to put the device into.
1198 */
1199int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
1200{
1201        int error;
1202
1203        if (platform_pci_power_manageable(dev)) {
1204                error = platform_pci_set_power_state(dev, state);
1205                if (!error)
1206                        pci_update_current_state(dev, state);
1207        } else
1208                error = -ENODEV;
1209
1210        if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
1211                dev->current_state = PCI_D0;
1212
1213        return error;
1214}
1215EXPORT_SYMBOL_GPL(pci_platform_power_transition);
1216
1217static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
1218{
1219        pm_request_resume(&pci_dev->dev);
1220        return 0;
1221}
1222
1223/**
1224 * pci_resume_bus - Walk given bus and runtime resume devices on it
1225 * @bus: Top bus of the subtree to walk.
1226 */
1227void pci_resume_bus(struct pci_bus *bus)
1228{
1229        if (bus)
1230                pci_walk_bus(bus, pci_resume_one, NULL);
1231}
1232
1233static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
1234{
1235        int delay = 1;
1236        u32 id;
1237
1238        /*
1239         * After reset, the device should not silently discard config
1240         * requests, but it may still indicate that it needs more time by
1241         * responding to them with CRS completions.  The Root Port will
1242         * generally synthesize ~0 data to complete the read (except when
1243         * CRS SV is enabled and the read was for the Vendor ID; in that
1244         * case it synthesizes 0x0001 data).
1245         *
1246         * Wait for the device to return a non-CRS completion.  Read the
1247         * Command register instead of Vendor ID so we don't have to
1248         * contend with the CRS SV value.
1249         */
1250        pci_read_config_dword(dev, PCI_COMMAND, &id);
1251        while (id == ~0) {
1252                if (delay > timeout) {
1253                        pci_warn(dev, "not ready %dms after %s; giving up\n",
1254                                 delay - 1, reset_type);
1255                        return -ENOTTY;
1256                }
1257
1258                if (delay > 1000)
1259                        pci_info(dev, "not ready %dms after %s; waiting\n",
1260                                 delay - 1, reset_type);
1261
1262                msleep(delay);
1263                delay *= 2;
1264                pci_read_config_dword(dev, PCI_COMMAND, &id);
1265        }
1266
1267        if (delay > 1000)
1268                pci_info(dev, "ready %dms after %s\n", delay - 1,
1269                         reset_type);
1270
1271        return 0;
1272}
1273
1274/**
1275 * pci_power_up - Put the given device into D0
1276 * @dev: PCI device to power up
1277 */
1278int pci_power_up(struct pci_dev *dev)
1279{
1280        pci_platform_power_transition(dev, PCI_D0);
1281
1282        /*
1283         * Mandatory power management transition delays are handled in
1284         * pci_pm_resume_noirq() and pci_pm_runtime_resume() of the
1285         * corresponding bridge.
1286         */
1287        if (dev->runtime_d3cold) {
1288                /*
1289                 * When powering on a bridge from D3cold, the whole hierarchy
1290                 * may be powered on into D0uninitialized state, resume them to
1291                 * give them a chance to suspend again
1292                 */
1293                pci_resume_bus(dev->subordinate);
1294        }
1295
1296        return pci_raw_set_power_state(dev, PCI_D0);
1297}
1298
1299/**
1300 * __pci_dev_set_current_state - Set current state of a PCI device
1301 * @dev: Device to handle
1302 * @data: pointer to state to be set
1303 */
1304static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1305{
1306        pci_power_t state = *(pci_power_t *)data;
1307
1308        dev->current_state = state;
1309        return 0;
1310}
1311
1312/**
1313 * pci_bus_set_current_state - Walk given bus and set current state of devices
1314 * @bus: Top bus of the subtree to walk.
1315 * @state: state to be set
1316 */
1317void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1318{
1319        if (bus)
1320                pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1321}
1322
1323/**
1324 * pci_set_power_state - Set the power state of a PCI device
1325 * @dev: PCI device to handle.
1326 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1327 *
1328 * Transition a device to a new power state, using the platform firmware and/or
1329 * the device's PCI PM registers.
1330 *
1331 * RETURN VALUE:
1332 * -EINVAL if the requested state is invalid.
1333 * -EIO if device does not support PCI PM or its PM capabilities register has a
1334 * wrong version, or device doesn't support the requested state.
1335 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1336 * 0 if device already is in the requested state.
1337 * 0 if the transition is to D3 but D3 is not supported.
1338 * 0 if device's power state has been successfully changed.
1339 */
1340int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1341{
1342        int error;
1343
1344        /* Bound the state we're entering */
1345        if (state > PCI_D3cold)
1346                state = PCI_D3cold;
1347        else if (state < PCI_D0)
1348                state = PCI_D0;
1349        else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1350
1351                /*
1352                 * If the device or the parent bridge do not support PCI
1353                 * PM, ignore the request if we're doing anything other
1354                 * than putting it into D0 (which would only happen on
1355                 * boot).
1356                 */
1357                return 0;
1358
1359        /* Check if we're already there */
1360        if (dev->current_state == state)
1361                return 0;
1362
1363        if (state == PCI_D0)
1364                return pci_power_up(dev);
1365
1366        /*
1367         * This device is quirked not to be put into D3, so don't put it in
1368         * D3
1369         */
1370        if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1371                return 0;
1372
1373        /*
1374         * To put device in D3cold, we put device into D3hot in native
1375         * way, then put device into D3cold with platform ops
1376         */
1377        error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1378                                        PCI_D3hot : state);
1379
1380        if (pci_platform_power_transition(dev, state))
1381                return error;
1382
1383        /* Powering off a bridge may power off the whole hierarchy */
1384        if (state == PCI_D3cold)
1385                pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1386
1387        return 0;
1388}
1389EXPORT_SYMBOL(pci_set_power_state);
1390
1391/**
1392 * pci_choose_state - Choose the power state of a PCI device
1393 * @dev: PCI device to be suspended
1394 * @state: target sleep state for the whole system. This is the value
1395 *         that is passed to suspend() function.
1396 *
1397 * Returns PCI power state suitable for given device and given system
1398 * message.
1399 */
1400pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
1401{
1402        pci_power_t ret;
1403
1404        if (!dev->pm_cap)
1405                return PCI_D0;
1406
1407        ret = platform_pci_choose_state(dev);
1408        if (ret != PCI_POWER_ERROR)
1409                return ret;
1410
1411        switch (state.event) {
1412        case PM_EVENT_ON:
1413                return PCI_D0;
1414        case PM_EVENT_FREEZE:
1415        case PM_EVENT_PRETHAW:
1416                /* REVISIT both freeze and pre-thaw "should" use D0 */
1417        case PM_EVENT_SUSPEND:
1418        case PM_EVENT_HIBERNATE:
1419                return PCI_D3hot;
1420        default:
1421                pci_info(dev, "unrecognized suspend event %d\n",
1422                         state.event);
1423                BUG();
1424        }
1425        return PCI_D0;
1426}
1427EXPORT_SYMBOL(pci_choose_state);
1428
1429#define PCI_EXP_SAVE_REGS       7
1430
1431static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1432                                                       u16 cap, bool extended)
1433{
1434        struct pci_cap_saved_state *tmp;
1435
1436        hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1437                if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1438                        return tmp;
1439        }
1440        return NULL;
1441}
1442
1443struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1444{
1445        return _pci_find_saved_cap(dev, cap, false);
1446}
1447
1448struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1449{
1450        return _pci_find_saved_cap(dev, cap, true);
1451}
1452
1453static int pci_save_pcie_state(struct pci_dev *dev)
1454{
1455        int i = 0;
1456        struct pci_cap_saved_state *save_state;
1457        u16 *cap;
1458
1459        if (!pci_is_pcie(dev))
1460                return 0;
1461
1462        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1463        if (!save_state) {
1464                pci_err(dev, "buffer not found in %s\n", __func__);
1465                return -ENOMEM;
1466        }
1467
1468        cap = (u16 *)&save_state->cap.data[0];
1469        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1470        pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1471        pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1472        pcie_capability_read_word(dev, PCI_EXP_RTCTL,  &cap[i++]);
1473        pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1474        pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1475        pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1476
1477        return 0;
1478}
1479
1480static void pci_restore_pcie_state(struct pci_dev *dev)
1481{
1482        int i = 0;
1483        struct pci_cap_saved_state *save_state;
1484        u16 *cap;
1485
1486        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1487        if (!save_state)
1488                return;
1489
1490        cap = (u16 *)&save_state->cap.data[0];
1491        pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1492        pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1493        pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1494        pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1495        pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1496        pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1497        pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1498}
1499
1500static int pci_save_pcix_state(struct pci_dev *dev)
1501{
1502        int pos;
1503        struct pci_cap_saved_state *save_state;
1504
1505        pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1506        if (!pos)
1507                return 0;
1508
1509        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1510        if (!save_state) {
1511                pci_err(dev, "buffer not found in %s\n", __func__);
1512                return -ENOMEM;
1513        }
1514
1515        pci_read_config_word(dev, pos + PCI_X_CMD,
1516                             (u16 *)save_state->cap.data);
1517
1518        return 0;
1519}
1520
1521static void pci_restore_pcix_state(struct pci_dev *dev)
1522{
1523        int i = 0, pos;
1524        struct pci_cap_saved_state *save_state;
1525        u16 *cap;
1526
1527        save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1528        pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1529        if (!save_state || !pos)
1530                return;
1531        cap = (u16 *)&save_state->cap.data[0];
1532
1533        pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1534}
1535
1536static void pci_save_ltr_state(struct pci_dev *dev)
1537{
1538        int ltr;
1539        struct pci_cap_saved_state *save_state;
1540        u16 *cap;
1541
1542        if (!pci_is_pcie(dev))
1543                return;
1544
1545        ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1546        if (!ltr)
1547                return;
1548
1549        save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1550        if (!save_state) {
1551                pci_err(dev, "no suspend buffer for LTR; ASPM issues possible after resume\n");
1552                return;
1553        }
1554
1555        cap = (u16 *)&save_state->cap.data[0];
1556        pci_read_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, cap++);
1557        pci_read_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, cap++);
1558}
1559
1560static void pci_restore_ltr_state(struct pci_dev *dev)
1561{
1562        struct pci_cap_saved_state *save_state;
1563        int ltr;
1564        u16 *cap;
1565
1566        save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1567        ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1568        if (!save_state || !ltr)
1569                return;
1570
1571        cap = (u16 *)&save_state->cap.data[0];
1572        pci_write_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, *cap++);
1573        pci_write_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, *cap++);
1574}
1575
1576/**
1577 * pci_save_state - save the PCI configuration space of a device before
1578 *                  suspending
1579 * @dev: PCI device that we're dealing with
1580 */
1581int pci_save_state(struct pci_dev *dev)
1582{
1583        int i;
1584        /* XXX: 100% dword access ok here? */
1585        for (i = 0; i < 16; i++) {
1586                pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1587                pci_dbg(dev, "saving config space at offset %#x (reading %#x)\n",
1588                        i * 4, dev->saved_config_space[i]);
1589        }
1590        dev->state_saved = true;
1591
1592        i = pci_save_pcie_state(dev);
1593        if (i != 0)
1594                return i;
1595
1596        i = pci_save_pcix_state(dev);
1597        if (i != 0)
1598                return i;
1599
1600        pci_save_ltr_state(dev);
1601        pci_save_dpc_state(dev);
1602        pci_save_aer_state(dev);
1603        pci_save_ptm_state(dev);
1604        return pci_save_vc_state(dev);
1605}
1606EXPORT_SYMBOL(pci_save_state);
1607
1608static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1609                                     u32 saved_val, int retry, bool force)
1610{
1611        u32 val;
1612
1613        pci_read_config_dword(pdev, offset, &val);
1614        if (!force && val == saved_val)
1615                return;
1616
1617        for (;;) {
1618                pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1619                        offset, val, saved_val);
1620                pci_write_config_dword(pdev, offset, saved_val);
1621                if (retry-- <= 0)
1622                        return;
1623
1624                pci_read_config_dword(pdev, offset, &val);
1625                if (val == saved_val)
1626                        return;
1627
1628                mdelay(1);
1629        }
1630}
1631
1632static void pci_restore_config_space_range(struct pci_dev *pdev,
1633                                           int start, int end, int retry,
1634                                           bool force)
1635{
1636        int index;
1637
1638        for (index = end; index >= start; index--)
1639                pci_restore_config_dword(pdev, 4 * index,
1640                                         pdev->saved_config_space[index],
1641                                         retry, force);
1642}
1643
1644static void pci_restore_config_space(struct pci_dev *pdev)
1645{
1646        if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1647                pci_restore_config_space_range(pdev, 10, 15, 0, false);
1648                /* Restore BARs before the command register. */
1649                pci_restore_config_space_range(pdev, 4, 9, 10, false);
1650                pci_restore_config_space_range(pdev, 0, 3, 0, false);
1651        } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1652                pci_restore_config_space_range(pdev, 12, 15, 0, false);
1653
1654                /*
1655                 * Force rewriting of prefetch registers to avoid S3 resume
1656                 * issues on Intel PCI bridges that occur when these
1657                 * registers are not explicitly written.
1658                 */
1659                pci_restore_config_space_range(pdev, 9, 11, 0, true);
1660                pci_restore_config_space_range(pdev, 0, 8, 0, false);
1661        } else {
1662                pci_restore_config_space_range(pdev, 0, 15, 0, false);
1663        }
1664}
1665
1666static void pci_restore_rebar_state(struct pci_dev *pdev)
1667{
1668        unsigned int pos, nbars, i;
1669        u32 ctrl;
1670
1671        pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1672        if (!pos)
1673                return;
1674
1675        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1676        nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1677                    PCI_REBAR_CTRL_NBAR_SHIFT;
1678
1679        for (i = 0; i < nbars; i++, pos += 8) {
1680                struct resource *res;
1681                int bar_idx, size;
1682
1683                pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1684                bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1685                res = pdev->resource + bar_idx;
1686                size = pci_rebar_bytes_to_size(resource_size(res));
1687                ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1688                ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1689                pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1690        }
1691}
1692
1693/**
1694 * pci_restore_state - Restore the saved state of a PCI device
1695 * @dev: PCI device that we're dealing with
1696 */
1697void pci_restore_state(struct pci_dev *dev)
1698{
1699        if (!dev->state_saved)
1700                return;
1701
1702        /*
1703         * Restore max latencies (in the LTR capability) before enabling
1704         * LTR itself (in the PCIe capability).
1705         */
1706        pci_restore_ltr_state(dev);
1707
1708        pci_restore_pcie_state(dev);
1709        pci_restore_pasid_state(dev);
1710        pci_restore_pri_state(dev);
1711        pci_restore_ats_state(dev);
1712        pci_restore_vc_state(dev);
1713        pci_restore_rebar_state(dev);
1714        pci_restore_dpc_state(dev);
1715        pci_restore_ptm_state(dev);
1716
1717        pci_aer_clear_status(dev);
1718        pci_restore_aer_state(dev);
1719
1720        pci_restore_config_space(dev);
1721
1722        pci_restore_pcix_state(dev);
1723        pci_restore_msi_state(dev);
1724
1725        /* Restore ACS and IOV configuration state */
1726        pci_enable_acs(dev);
1727        pci_restore_iov_state(dev);
1728
1729        dev->state_saved = false;
1730}
1731EXPORT_SYMBOL(pci_restore_state);
1732
1733struct pci_saved_state {
1734        u32 config_space[16];
1735        struct pci_cap_saved_data cap[];
1736};
1737
1738/**
1739 * pci_store_saved_state - Allocate and return an opaque struct containing
1740 *                         the device saved state.
1741 * @dev: PCI device that we're dealing with
1742 *
1743 * Return NULL if no state or error.
1744 */
1745struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1746{
1747        struct pci_saved_state *state;
1748        struct pci_cap_saved_state *tmp;
1749        struct pci_cap_saved_data *cap;
1750        size_t size;
1751
1752        if (!dev->state_saved)
1753                return NULL;
1754
1755        size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1756
1757        hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1758                size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1759
1760        state = kzalloc(size, GFP_KERNEL);
1761        if (!state)
1762                return NULL;
1763
1764        memcpy(state->config_space, dev->saved_config_space,
1765               sizeof(state->config_space));
1766
1767        cap = state->cap;
1768        hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1769                size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1770                memcpy(cap, &tmp->cap, len);
1771                cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1772        }
1773        /* Empty cap_save terminates list */
1774
1775        return state;
1776}
1777EXPORT_SYMBOL_GPL(pci_store_saved_state);
1778
1779/**
1780 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1781 * @dev: PCI device that we're dealing with
1782 * @state: Saved state returned from pci_store_saved_state()
1783 */
1784int pci_load_saved_state(struct pci_dev *dev,
1785                         struct pci_saved_state *state)
1786{
1787        struct pci_cap_saved_data *cap;
1788
1789        dev->state_saved = false;
1790
1791        if (!state)
1792                return 0;
1793
1794        memcpy(dev->saved_config_space, state->config_space,
1795               sizeof(state->config_space));
1796
1797        cap = state->cap;
1798        while (cap->size) {
1799                struct pci_cap_saved_state *tmp;
1800
1801                tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1802                if (!tmp || tmp->cap.size != cap->size)
1803                        return -EINVAL;
1804
1805                memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1806                cap = (struct pci_cap_saved_data *)((u8 *)cap +
1807                       sizeof(struct pci_cap_saved_data) + cap->size);
1808        }
1809
1810        dev->state_saved = true;
1811        return 0;
1812}
1813EXPORT_SYMBOL_GPL(pci_load_saved_state);
1814
1815/**
1816 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1817 *                                 and free the memory allocated for it.
1818 * @dev: PCI device that we're dealing with
1819 * @state: Pointer to saved state returned from pci_store_saved_state()
1820 */
1821int pci_load_and_free_saved_state(struct pci_dev *dev,
1822                                  struct pci_saved_state **state)
1823{
1824        int ret = pci_load_saved_state(dev, *state);
1825        kfree(*state);
1826        *state = NULL;
1827        return ret;
1828}
1829EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1830
1831int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1832{
1833        return pci_enable_resources(dev, bars);
1834}
1835
1836static int do_pci_enable_device(struct pci_dev *dev, int bars)
1837{
1838        int err;
1839        struct pci_dev *bridge;
1840        u16 cmd;
1841        u8 pin;
1842
1843        err = pci_set_power_state(dev, PCI_D0);
1844        if (err < 0 && err != -EIO)
1845                return err;
1846
1847        bridge = pci_upstream_bridge(dev);
1848        if (bridge)
1849                pcie_aspm_powersave_config_link(bridge);
1850
1851        err = pcibios_enable_device(dev, bars);
1852        if (err < 0)
1853                return err;
1854        pci_fixup_device(pci_fixup_enable, dev);
1855
1856        if (dev->msi_enabled || dev->msix_enabled)
1857                return 0;
1858
1859        pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1860        if (pin) {
1861                pci_read_config_word(dev, PCI_COMMAND, &cmd);
1862                if (cmd & PCI_COMMAND_INTX_DISABLE)
1863                        pci_write_config_word(dev, PCI_COMMAND,
1864                                              cmd & ~PCI_COMMAND_INTX_DISABLE);
1865        }
1866
1867        return 0;
1868}
1869
1870/**
1871 * pci_reenable_device - Resume abandoned device
1872 * @dev: PCI device to be resumed
1873 *
1874 * NOTE: This function is a backend of pci_default_resume() and is not supposed
1875 * to be called by normal code, write proper resume handler and use it instead.
1876 */
1877int pci_reenable_device(struct pci_dev *dev)
1878{
1879        if (pci_is_enabled(dev))
1880                return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1881        return 0;
1882}
1883EXPORT_SYMBOL(pci_reenable_device);
1884
1885static void pci_enable_bridge(struct pci_dev *dev)
1886{
1887        struct pci_dev *bridge;
1888        int retval;
1889
1890        bridge = pci_upstream_bridge(dev);
1891        if (bridge)
1892                pci_enable_bridge(bridge);
1893
1894        if (pci_is_enabled(dev)) {
1895                if (!dev->is_busmaster)
1896                        pci_set_master(dev);
1897                return;
1898        }
1899
1900        retval = pci_enable_device(dev);
1901        if (retval)
1902                pci_err(dev, "Error enabling bridge (%d), continuing\n",
1903                        retval);
1904        pci_set_master(dev);
1905}
1906
1907static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1908{
1909        struct pci_dev *bridge;
1910        int err;
1911        int i, bars = 0;
1912
1913        /*
1914         * Power state could be unknown at this point, either due to a fresh
1915         * boot or a device removal call.  So get the current power state
1916         * so that things like MSI message writing will behave as expected
1917         * (e.g. if the device really is in D0 at enable time).
1918         */
1919        pci_update_current_state(dev, dev->current_state);
1920
1921        if (atomic_inc_return(&dev->enable_cnt) > 1)
1922                return 0;               /* already enabled */
1923
1924        bridge = pci_upstream_bridge(dev);
1925        if (bridge)
1926                pci_enable_bridge(bridge);
1927
1928        /* only skip sriov related */
1929        for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1930                if (dev->resource[i].flags & flags)
1931                        bars |= (1 << i);
1932        for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1933                if (dev->resource[i].flags & flags)
1934                        bars |= (1 << i);
1935
1936        err = do_pci_enable_device(dev, bars);
1937        if (err < 0)
1938                atomic_dec(&dev->enable_cnt);
1939        return err;
1940}
1941
1942/**
1943 * pci_enable_device_io - Initialize a device for use with IO space
1944 * @dev: PCI device to be initialized
1945 *
1946 * Initialize device before it's used by a driver. Ask low-level code
1947 * to enable I/O resources. Wake up the device if it was suspended.
1948 * Beware, this function can fail.
1949 */
1950int pci_enable_device_io(struct pci_dev *dev)
1951{
1952        return pci_enable_device_flags(dev, IORESOURCE_IO);
1953}
1954EXPORT_SYMBOL(pci_enable_device_io);
1955
1956/**
1957 * pci_enable_device_mem - Initialize a device for use with Memory space
1958 * @dev: PCI device to be initialized
1959 *
1960 * Initialize device before it's used by a driver. Ask low-level code
1961 * to enable Memory resources. Wake up the device if it was suspended.
1962 * Beware, this function can fail.
1963 */
1964int pci_enable_device_mem(struct pci_dev *dev)
1965{
1966        return pci_enable_device_flags(dev, IORESOURCE_MEM);
1967}
1968EXPORT_SYMBOL(pci_enable_device_mem);
1969
1970/**
1971 * pci_enable_device - Initialize device before it's used by a driver.
1972 * @dev: PCI device to be initialized
1973 *
1974 * Initialize device before it's used by a driver. Ask low-level code
1975 * to enable I/O and memory. Wake up the device if it was suspended.
1976 * Beware, this function can fail.
1977 *
1978 * Note we don't actually enable the device many times if we call
1979 * this function repeatedly (we just increment the count).
1980 */
1981int pci_enable_device(struct pci_dev *dev)
1982{
1983        return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1984}
1985EXPORT_SYMBOL(pci_enable_device);
1986
1987/*
1988 * Managed PCI resources.  This manages device on/off, INTx/MSI/MSI-X
1989 * on/off and BAR regions.  pci_dev itself records MSI/MSI-X status, so
1990 * there's no need to track it separately.  pci_devres is initialized
1991 * when a device is enabled using managed PCI device enable interface.
1992 */
1993struct pci_devres {
1994        unsigned int enabled:1;
1995        unsigned int pinned:1;
1996        unsigned int orig_intx:1;
1997        unsigned int restore_intx:1;
1998        unsigned int mwi:1;
1999        u32 region_mask;
2000};
2001
2002static void pcim_release(struct device *gendev, void *res)
2003{
2004        struct pci_dev *dev = to_pci_dev(gendev);
2005        struct pci_devres *this = res;
2006        int i;
2007
2008        if (dev->msi_enabled)
2009                pci_disable_msi(dev);
2010        if (dev->msix_enabled)
2011                pci_disable_msix(dev);
2012
2013        for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
2014                if (this->region_mask & (1 << i))
2015                        pci_release_region(dev, i);
2016
2017        if (this->mwi)
2018                pci_clear_mwi(dev);
2019
2020        if (this->restore_intx)
2021                pci_intx(dev, this->orig_intx);
2022
2023        if (this->enabled && !this->pinned)
2024                pci_disable_device(dev);
2025}
2026
2027static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
2028{
2029        struct pci_devres *dr, *new_dr;
2030
2031        dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
2032        if (dr)
2033                return dr;
2034
2035        new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
2036        if (!new_dr)
2037                return NULL;
2038        return devres_get(&pdev->dev, new_dr, NULL, NULL);
2039}
2040
2041static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
2042{
2043        if (pci_is_managed(pdev))
2044                return devres_find(&pdev->dev, pcim_release, NULL, NULL);
2045        return NULL;
2046}
2047
2048/**
2049 * pcim_enable_device - Managed pci_enable_device()
2050 * @pdev: PCI device to be initialized
2051 *
2052 * Managed pci_enable_device().
2053 */
2054int pcim_enable_device(struct pci_dev *pdev)
2055{
2056        struct pci_devres *dr;
2057        int rc;
2058
2059        dr = get_pci_dr(pdev);
2060        if (unlikely(!dr))
2061                return -ENOMEM;
2062        if (dr->enabled)
2063                return 0;
2064
2065        rc = pci_enable_device(pdev);
2066        if (!rc) {
2067                pdev->is_managed = 1;
2068                dr->enabled = 1;
2069        }
2070        return rc;
2071}
2072EXPORT_SYMBOL(pcim_enable_device);
2073
2074/**
2075 * pcim_pin_device - Pin managed PCI device
2076 * @pdev: PCI device to pin
2077 *
2078 * Pin managed PCI device @pdev.  Pinned device won't be disabled on
2079 * driver detach.  @pdev must have been enabled with
2080 * pcim_enable_device().
2081 */
2082void pcim_pin_device(struct pci_dev *pdev)
2083{
2084        struct pci_devres *dr;
2085
2086        dr = find_pci_dr(pdev);
2087        WARN_ON(!dr || !dr->enabled);
2088        if (dr)
2089                dr->pinned = 1;
2090}
2091EXPORT_SYMBOL(pcim_pin_device);
2092
2093/*
2094 * pcibios_add_device - provide arch specific hooks when adding device dev
2095 * @dev: the PCI device being added
2096 *
2097 * Permits the platform to provide architecture specific functionality when
2098 * devices are added. This is the default implementation. Architecture
2099 * implementations can override this.
2100 */
2101int __weak pcibios_add_device(struct pci_dev *dev)
2102{
2103        return 0;
2104}
2105
2106/**
2107 * pcibios_release_device - provide arch specific hooks when releasing
2108 *                          device dev
2109 * @dev: the PCI device being released
2110 *
2111 * Permits the platform to provide architecture specific functionality when
2112 * devices are released. This is the default implementation. Architecture
2113 * implementations can override this.
2114 */
2115void __weak pcibios_release_device(struct pci_dev *dev) {}
2116
2117/**
2118 * pcibios_disable_device - disable arch specific PCI resources for device dev
2119 * @dev: the PCI device to disable
2120 *
2121 * Disables architecture specific PCI resources for the device. This
2122 * is the default implementation. Architecture implementations can
2123 * override this.
2124 */
2125void __weak pcibios_disable_device(struct pci_dev *dev) {}
2126
2127/**
2128 * pcibios_penalize_isa_irq - penalize an ISA IRQ
2129 * @irq: ISA IRQ to penalize
2130 * @active: IRQ active or not
2131 *
2132 * Permits the platform to provide architecture-specific functionality when
2133 * penalizing ISA IRQs. This is the default implementation. Architecture
2134 * implementations can override this.
2135 */
2136void __weak pcibios_penalize_isa_irq(int irq, int active) {}
2137
2138static void do_pci_disable_device(struct pci_dev *dev)
2139{
2140        u16 pci_command;
2141
2142        pci_read_config_word(dev, PCI_COMMAND, &pci_command);
2143        if (pci_command & PCI_COMMAND_MASTER) {
2144                pci_command &= ~PCI_COMMAND_MASTER;
2145                pci_write_config_word(dev, PCI_COMMAND, pci_command);
2146        }
2147
2148        pcibios_disable_device(dev);
2149}
2150
2151/**
2152 * pci_disable_enabled_device - Disable device without updating enable_cnt
2153 * @dev: PCI device to disable
2154 *
2155 * NOTE: This function is a backend of PCI power management routines and is
2156 * not supposed to be called drivers.
2157 */
2158void pci_disable_enabled_device(struct pci_dev *dev)
2159{
2160        if (pci_is_enabled(dev))
2161                do_pci_disable_device(dev);
2162}
2163
2164/**
2165 * pci_disable_device - Disable PCI device after use
2166 * @dev: PCI device to be disabled
2167 *
2168 * Signal to the system that the PCI device is not in use by the system
2169 * anymore.  This only involves disabling PCI bus-mastering, if active.
2170 *
2171 * Note we don't actually disable the device until all callers of
2172 * pci_enable_device() have called pci_disable_device().
2173 */
2174void pci_disable_device(struct pci_dev *dev)
2175{
2176        struct pci_devres *dr;
2177
2178        dr = find_pci_dr(dev);
2179        if (dr)
2180                dr->enabled = 0;
2181
2182        dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
2183                      "disabling already-disabled device");
2184
2185        if (atomic_dec_return(&dev->enable_cnt) != 0)
2186                return;
2187
2188        do_pci_disable_device(dev);
2189
2190        dev->is_busmaster = 0;
2191}
2192EXPORT_SYMBOL(pci_disable_device);
2193
2194/**
2195 * pcibios_set_pcie_reset_state - set reset state for device dev
2196 * @dev: the PCIe device reset
2197 * @state: Reset state to enter into
2198 *
2199 * Set the PCIe reset state for the device. This is the default
2200 * implementation. Architecture implementations can override this.
2201 */
2202int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
2203                                        enum pcie_reset_state state)
2204{
2205        return -EINVAL;
2206}
2207
2208/**
2209 * pci_set_pcie_reset_state - set reset state for device dev
2210 * @dev: the PCIe device reset
2211 * @state: Reset state to enter into
2212 *
2213 * Sets the PCI reset state for the device.
2214 */
2215int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
2216{
2217        return pcibios_set_pcie_reset_state(dev, state);
2218}
2219EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
2220
2221void pcie_clear_device_status(struct pci_dev *dev)
2222{
2223        u16 sta;
2224
2225        pcie_capability_read_word(dev, PCI_EXP_DEVSTA, &sta);
2226        pcie_capability_write_word(dev, PCI_EXP_DEVSTA, sta);
2227}
2228
2229/**
2230 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
2231 * @dev: PCIe root port or event collector.
2232 */
2233void pcie_clear_root_pme_status(struct pci_dev *dev)
2234{
2235        pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
2236}
2237
2238/**
2239 * pci_check_pme_status - Check if given device has generated PME.
2240 * @dev: Device to check.
2241 *
2242 * Check the PME status of the device and if set, clear it and clear PME enable
2243 * (if set).  Return 'true' if PME status and PME enable were both set or
2244 * 'false' otherwise.
2245 */
2246bool pci_check_pme_status(struct pci_dev *dev)
2247{
2248        int pmcsr_pos;
2249        u16 pmcsr;
2250        bool ret = false;
2251
2252        if (!dev->pm_cap)
2253                return false;
2254
2255        pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
2256        pci_read_config_word(dev, pmcsr_pos, &pmcsr);
2257        if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
2258                return false;
2259
2260        /* Clear PME status. */
2261        pmcsr |= PCI_PM_CTRL_PME_STATUS;
2262        if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
2263                /* Disable PME to avoid interrupt flood. */
2264                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2265                ret = true;
2266        }
2267
2268        pci_write_config_word(dev, pmcsr_pos, pmcsr);
2269
2270        return ret;
2271}
2272
2273/**
2274 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2275 * @dev: Device to handle.
2276 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2277 *
2278 * Check if @dev has generated PME and queue a resume request for it in that
2279 * case.
2280 */
2281static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2282{
2283        if (pme_poll_reset && dev->pme_poll)
2284                dev->pme_poll = false;
2285
2286        if (pci_check_pme_status(dev)) {
2287                pci_wakeup_event(dev);
2288                pm_request_resume(&dev->dev);
2289        }
2290        return 0;
2291}
2292
2293/**
2294 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2295 * @bus: Top bus of the subtree to walk.
2296 */
2297void pci_pme_wakeup_bus(struct pci_bus *bus)
2298{
2299        if (bus)
2300                pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2301}
2302
2303
2304/**
2305 * pci_pme_capable - check the capability of PCI device to generate PME#
2306 * @dev: PCI device to handle.
2307 * @state: PCI state from which device will issue PME#.
2308 */
2309bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2310{
2311        if (!dev->pm_cap)
2312                return false;
2313
2314        return !!(dev->pme_support & (1 << state));
2315}
2316EXPORT_SYMBOL(pci_pme_capable);
2317
2318static void pci_pme_list_scan(struct work_struct *work)
2319{
2320        struct pci_pme_device *pme_dev, *n;
2321
2322        mutex_lock(&pci_pme_list_mutex);
2323        list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2324                if (pme_dev->dev->pme_poll) {
2325                        struct pci_dev *bridge;
2326
2327                        bridge = pme_dev->dev->bus->self;
2328                        /*
2329                         * If bridge is in low power state, the
2330                         * configuration space of subordinate devices
2331                         * may be not accessible
2332                         */
2333                        if (bridge && bridge->current_state != PCI_D0)
2334                                continue;
2335                        /*
2336                         * If the device is in D3cold it should not be
2337                         * polled either.
2338                         */
2339                        if (pme_dev->dev->current_state == PCI_D3cold)
2340                                continue;
2341
2342                        pci_pme_wakeup(pme_dev->dev, NULL);
2343                } else {
2344                        list_del(&pme_dev->list);
2345                        kfree(pme_dev);
2346                }
2347        }
2348        if (!list_empty(&pci_pme_list))
2349                queue_delayed_work(system_freezable_wq, &pci_pme_work,
2350                                   msecs_to_jiffies(PME_TIMEOUT));
2351        mutex_unlock(&pci_pme_list_mutex);
2352}
2353
2354static void __pci_pme_active(struct pci_dev *dev, bool enable)
2355{
2356        u16 pmcsr;
2357
2358        if (!dev->pme_support)
2359                return;
2360
2361        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2362        /* Clear PME_Status by writing 1 to it and enable PME# */
2363        pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2364        if (!enable)
2365                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2366
2367        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2368}
2369
2370/**
2371 * pci_pme_restore - Restore PME configuration after config space restore.
2372 * @dev: PCI device to update.
2373 */
2374void pci_pme_restore(struct pci_dev *dev)
2375{
2376        u16 pmcsr;
2377
2378        if (!dev->pme_support)
2379                return;
2380
2381        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2382        if (dev->wakeup_prepared) {
2383                pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2384                pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2385        } else {
2386                pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2387                pmcsr |= PCI_PM_CTRL_PME_STATUS;
2388        }
2389        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2390}
2391
2392/**
2393 * pci_pme_active - enable or disable PCI device's PME# function
2394 * @dev: PCI device to handle.
2395 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2396 *
2397 * The caller must verify that the device is capable of generating PME# before
2398 * calling this function with @enable equal to 'true'.
2399 */
2400void pci_pme_active(struct pci_dev *dev, bool enable)
2401{
2402        __pci_pme_active(dev, enable);
2403
2404        /*
2405         * PCI (as opposed to PCIe) PME requires that the device have
2406         * its PME# line hooked up correctly. Not all hardware vendors
2407         * do this, so the PME never gets delivered and the device
2408         * remains asleep. The easiest way around this is to
2409         * periodically walk the list of suspended devices and check
2410         * whether any have their PME flag set. The assumption is that
2411         * we'll wake up often enough anyway that this won't be a huge
2412         * hit, and the power savings from the devices will still be a
2413         * win.
2414         *
2415         * Although PCIe uses in-band PME message instead of PME# line
2416         * to report PME, PME does not work for some PCIe devices in
2417         * reality.  For example, there are devices that set their PME
2418         * status bits, but don't really bother to send a PME message;
2419         * there are PCI Express Root Ports that don't bother to
2420         * trigger interrupts when they receive PME messages from the
2421         * devices below.  So PME poll is used for PCIe devices too.
2422         */
2423
2424        if (dev->pme_poll) {
2425                struct pci_pme_device *pme_dev;
2426                if (enable) {
2427                        pme_dev = kmalloc(sizeof(struct pci_pme_device),
2428                                          GFP_KERNEL);
2429                        if (!pme_dev) {
2430                                pci_warn(dev, "can't enable PME#\n");
2431                                return;
2432                        }
2433                        pme_dev->dev = dev;
2434                        mutex_lock(&pci_pme_list_mutex);
2435                        list_add(&pme_dev->list, &pci_pme_list);
2436                        if (list_is_singular(&pci_pme_list))
2437                                queue_delayed_work(system_freezable_wq,
2438                                                   &pci_pme_work,
2439                                                   msecs_to_jiffies(PME_TIMEOUT));
2440                        mutex_unlock(&pci_pme_list_mutex);
2441                } else {
2442                        mutex_lock(&pci_pme_list_mutex);
2443                        list_for_each_entry(pme_dev, &pci_pme_list, list) {
2444                                if (pme_dev->dev == dev) {
2445                                        list_del(&pme_dev->list);
2446                                        kfree(pme_dev);
2447                                        break;
2448                                }
2449                        }
2450                        mutex_unlock(&pci_pme_list_mutex);
2451                }
2452        }
2453
2454        pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2455}
2456EXPORT_SYMBOL(pci_pme_active);
2457
2458/**
2459 * __pci_enable_wake - enable PCI device as wakeup event source
2460 * @dev: PCI device affected
2461 * @state: PCI state from which device will issue wakeup events
2462 * @enable: True to enable event generation; false to disable
2463 *
2464 * This enables the device as a wakeup event source, or disables it.
2465 * When such events involves platform-specific hooks, those hooks are
2466 * called automatically by this routine.
2467 *
2468 * Devices with legacy power management (no standard PCI PM capabilities)
2469 * always require such platform hooks.
2470 *
2471 * RETURN VALUE:
2472 * 0 is returned on success
2473 * -EINVAL is returned if device is not supposed to wake up the system
2474 * Error code depending on the platform is returned if both the platform and
2475 * the native mechanism fail to enable the generation of wake-up events
2476 */
2477static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2478{
2479        int ret = 0;
2480
2481        /*
2482         * Bridges that are not power-manageable directly only signal
2483         * wakeup on behalf of subordinate devices which is set up
2484         * elsewhere, so skip them. However, bridges that are
2485         * power-manageable may signal wakeup for themselves (for example,
2486         * on a hotplug event) and they need to be covered here.
2487         */
2488        if (!pci_power_manageable(dev))
2489                return 0;
2490
2491        /* Don't do the same thing twice in a row for one device. */
2492        if (!!enable == !!dev->wakeup_prepared)
2493                return 0;
2494
2495        /*
2496         * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2497         * Anderson we should be doing PME# wake enable followed by ACPI wake
2498         * enable.  To disable wake-up we call the platform first, for symmetry.
2499         */
2500
2501        if (enable) {
2502                int error;
2503
2504                /*
2505                 * Enable PME signaling if the device can signal PME from
2506                 * D3cold regardless of whether or not it can signal PME from
2507                 * the current target state, because that will allow it to
2508                 * signal PME when the hierarchy above it goes into D3cold and
2509                 * the device itself ends up in D3cold as a result of that.
2510                 */
2511                if (pci_pme_capable(dev, state) || pci_pme_capable(dev, PCI_D3cold))
2512                        pci_pme_active(dev, true);
2513                else
2514                        ret = 1;
2515                error = platform_pci_set_wakeup(dev, true);
2516                if (ret)
2517                        ret = error;
2518                if (!ret)
2519                        dev->wakeup_prepared = true;
2520        } else {
2521                platform_pci_set_wakeup(dev, false);
2522                pci_pme_active(dev, false);
2523                dev->wakeup_prepared = false;
2524        }
2525
2526        return ret;
2527}
2528
2529/**
2530 * pci_enable_wake - change wakeup settings for a PCI device
2531 * @pci_dev: Target device
2532 * @state: PCI state from which device will issue wakeup events
2533 * @enable: Whether or not to enable event generation
2534 *
2535 * If @enable is set, check device_may_wakeup() for the device before calling
2536 * __pci_enable_wake() for it.
2537 */
2538int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2539{
2540        if (enable && !device_may_wakeup(&pci_dev->dev))
2541                return -EINVAL;
2542
2543        return __pci_enable_wake(pci_dev, state, enable);
2544}
2545EXPORT_SYMBOL(pci_enable_wake);
2546
2547/**
2548 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2549 * @dev: PCI device to prepare
2550 * @enable: True to enable wake-up event generation; false to disable
2551 *
2552 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2553 * and this function allows them to set that up cleanly - pci_enable_wake()
2554 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2555 * ordering constraints.
2556 *
2557 * This function only returns error code if the device is not allowed to wake
2558 * up the system from sleep or it is not capable of generating PME# from both
2559 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2560 */
2561int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2562{
2563        return pci_pme_capable(dev, PCI_D3cold) ?
2564                        pci_enable_wake(dev, PCI_D3cold, enable) :
2565                        pci_enable_wake(dev, PCI_D3hot, enable);
2566}
2567EXPORT_SYMBOL(pci_wake_from_d3);
2568
2569/**
2570 * pci_target_state - find an appropriate low power state for a given PCI dev
2571 * @dev: PCI device
2572 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2573 *
2574 * Use underlying platform code to find a supported low power state for @dev.
2575 * If the platform can't manage @dev, return the deepest state from which it
2576 * can generate wake events, based on any available PME info.
2577 */
2578static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2579{
2580        pci_power_t target_state = PCI_D3hot;
2581
2582        if (platform_pci_power_manageable(dev)) {
2583                /*
2584                 * Call the platform to find the target state for the device.
2585                 */
2586                pci_power_t state = platform_pci_choose_state(dev);
2587
2588                switch (state) {
2589                case PCI_POWER_ERROR:
2590                case PCI_UNKNOWN:
2591                        break;
2592                case PCI_D1:
2593                case PCI_D2:
2594                        if (pci_no_d1d2(dev))
2595                                break;
2596                        fallthrough;
2597                default:
2598                        target_state = state;
2599                }
2600
2601                return target_state;
2602        }
2603
2604        if (!dev->pm_cap)
2605                target_state = PCI_D0;
2606
2607        /*
2608         * If the device is in D3cold even though it's not power-manageable by
2609         * the platform, it may have been powered down by non-standard means.
2610         * Best to let it slumber.
2611         */
2612        if (dev->current_state == PCI_D3cold)
2613                target_state = PCI_D3cold;
2614
2615        if (wakeup && dev->pme_support) {
2616                pci_power_t state = target_state;
2617
2618                /*
2619                 * Find the deepest state from which the device can generate
2620                 * PME#.
2621                 */
2622                while (state && !(dev->pme_support & (1 << state)))
2623                        state--;
2624
2625                if (state)
2626                        return state;
2627                else if (dev->pme_support & 1)
2628                        return PCI_D0;
2629        }
2630
2631        return target_state;
2632}
2633
2634/**
2635 * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2636 *                        into a sleep state
2637 * @dev: Device to handle.
2638 *
2639 * Choose the power state appropriate for the device depending on whether
2640 * it can wake up the system and/or is power manageable by the platform
2641 * (PCI_D3hot is the default) and put the device into that state.
2642 */
2643int pci_prepare_to_sleep(struct pci_dev *dev)
2644{
2645        bool wakeup = device_may_wakeup(&dev->dev);
2646        pci_power_t target_state = pci_target_state(dev, wakeup);
2647        int error;
2648
2649        if (target_state == PCI_POWER_ERROR)
2650                return -EIO;
2651
2652        /*
2653         * There are systems (for example, Intel mobile chips since Coffee
2654         * Lake) where the power drawn while suspended can be significantly
2655         * reduced by disabling PTM on PCIe root ports as this allows the
2656         * port to enter a lower-power PM state and the SoC to reach a
2657         * lower-power idle state as a whole.
2658         */
2659        if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
2660                pci_disable_ptm(dev);
2661
2662        pci_enable_wake(dev, target_state, wakeup);
2663
2664        error = pci_set_power_state(dev, target_state);
2665
2666        if (error) {
2667                pci_enable_wake(dev, target_state, false);
2668                pci_restore_ptm_state(dev);
2669        }
2670
2671        return error;
2672}
2673EXPORT_SYMBOL(pci_prepare_to_sleep);
2674
2675/**
2676 * pci_back_from_sleep - turn PCI device on during system-wide transition
2677 *                       into working state
2678 * @dev: Device to handle.
2679 *
2680 * Disable device's system wake-up capability and put it into D0.
2681 */
2682int pci_back_from_sleep(struct pci_dev *dev)
2683{
2684        pci_enable_wake(dev, PCI_D0, false);
2685        return pci_set_power_state(dev, PCI_D0);
2686}
2687EXPORT_SYMBOL(pci_back_from_sleep);
2688
2689/**
2690 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2691 * @dev: PCI device being suspended.
2692 *
2693 * Prepare @dev to generate wake-up events at run time and put it into a low
2694 * power state.
2695 */
2696int pci_finish_runtime_suspend(struct pci_dev *dev)
2697{
2698        pci_power_t target_state;
2699        int error;
2700
2701        target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2702        if (target_state == PCI_POWER_ERROR)
2703                return -EIO;
2704
2705        dev->runtime_d3cold = target_state == PCI_D3cold;
2706
2707        /*
2708         * There are systems (for example, Intel mobile chips since Coffee
2709         * Lake) where the power drawn while suspended can be significantly
2710         * reduced by disabling PTM on PCIe root ports as this allows the
2711         * port to enter a lower-power PM state and the SoC to reach a
2712         * lower-power idle state as a whole.
2713         */
2714        if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
2715                pci_disable_ptm(dev);
2716
2717        __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2718
2719        error = pci_set_power_state(dev, target_state);
2720
2721        if (error) {
2722                pci_enable_wake(dev, target_state, false);
2723                pci_restore_ptm_state(dev);
2724                dev->runtime_d3cold = false;
2725        }
2726
2727        return error;
2728}
2729
2730/**
2731 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2732 * @dev: Device to check.
2733 *
2734 * Return true if the device itself is capable of generating wake-up events
2735 * (through the platform or using the native PCIe PME) or if the device supports
2736 * PME and one of its upstream bridges can generate wake-up events.
2737 */
2738bool pci_dev_run_wake(struct pci_dev *dev)
2739{
2740        struct pci_bus *bus = dev->bus;
2741
2742        if (!dev->pme_support)
2743                return false;
2744
2745        /* PME-capable in principle, but not from the target power state */
2746        if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2747                return false;
2748
2749        if (device_can_wakeup(&dev->dev))
2750                return true;
2751
2752        while (bus->parent) {
2753                struct pci_dev *bridge = bus->self;
2754
2755                if (device_can_wakeup(&bridge->dev))
2756                        return true;
2757
2758                bus = bus->parent;
2759        }
2760
2761        /* We have reached the root bus. */
2762        if (bus->bridge)
2763                return device_can_wakeup(bus->bridge);
2764
2765        return false;
2766}
2767EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2768
2769/**
2770 * pci_dev_need_resume - Check if it is necessary to resume the device.
2771 * @pci_dev: Device to check.
2772 *
2773 * Return 'true' if the device is not runtime-suspended or it has to be
2774 * reconfigured due to wakeup settings difference between system and runtime
2775 * suspend, or the current power state of it is not suitable for the upcoming
2776 * (system-wide) transition.
2777 */
2778bool pci_dev_need_resume(struct pci_dev *pci_dev)
2779{
2780        struct device *dev = &pci_dev->dev;
2781        pci_power_t target_state;
2782
2783        if (!pm_runtime_suspended(dev) || platform_pci_need_resume(pci_dev))
2784                return true;
2785
2786        target_state = pci_target_state(pci_dev, device_may_wakeup(dev));
2787
2788        /*
2789         * If the earlier platform check has not triggered, D3cold is just power
2790         * removal on top of D3hot, so no need to resume the device in that
2791         * case.
2792         */
2793        return target_state != pci_dev->current_state &&
2794                target_state != PCI_D3cold &&
2795                pci_dev->current_state != PCI_D3hot;
2796}
2797
2798/**
2799 * pci_dev_adjust_pme - Adjust PME setting for a suspended device.
2800 * @pci_dev: Device to check.
2801 *
2802 * If the device is suspended and it is not configured for system wakeup,
2803 * disable PME for it to prevent it from waking up the system unnecessarily.
2804 *
2805 * Note that if the device's power state is D3cold and the platform check in
2806 * pci_dev_need_resume() has not triggered, the device's configuration need not
2807 * be changed.
2808 */
2809void pci_dev_adjust_pme(struct pci_dev *pci_dev)
2810{
2811        struct device *dev = &pci_dev->dev;
2812
2813        spin_lock_irq(&dev->power.lock);
2814
2815        if (pm_runtime_suspended(dev) && !device_may_wakeup(dev) &&
2816            pci_dev->current_state < PCI_D3cold)
2817                __pci_pme_active(pci_dev, false);
2818
2819        spin_unlock_irq(&dev->power.lock);
2820}
2821
2822/**
2823 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2824 * @pci_dev: Device to handle.
2825 *
2826 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2827 * it might have been disabled during the prepare phase of system suspend if
2828 * the device was not configured for system wakeup.
2829 */
2830void pci_dev_complete_resume(struct pci_dev *pci_dev)
2831{
2832        struct device *dev = &pci_dev->dev;
2833
2834        if (!pci_dev_run_wake(pci_dev))
2835                return;
2836
2837        spin_lock_irq(&dev->power.lock);
2838
2839        if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2840                __pci_pme_active(pci_dev, true);
2841
2842        spin_unlock_irq(&dev->power.lock);
2843}
2844
2845void pci_config_pm_runtime_get(struct pci_dev *pdev)
2846{
2847        struct device *dev = &pdev->dev;
2848        struct device *parent = dev->parent;
2849
2850        if (parent)
2851                pm_runtime_get_sync(parent);
2852        pm_runtime_get_noresume(dev);
2853        /*
2854         * pdev->current_state is set to PCI_D3cold during suspending,
2855         * so wait until suspending completes
2856         */
2857        pm_runtime_barrier(dev);
2858        /*
2859         * Only need to resume devices in D3cold, because config
2860         * registers are still accessible for devices suspended but
2861         * not in D3cold.
2862         */
2863        if (pdev->current_state == PCI_D3cold)
2864                pm_runtime_resume(dev);
2865}
2866
2867void pci_config_pm_runtime_put(struct pci_dev *pdev)
2868{
2869        struct device *dev = &pdev->dev;
2870        struct device *parent = dev->parent;
2871
2872        pm_runtime_put(dev);
2873        if (parent)
2874                pm_runtime_put_sync(parent);
2875}
2876
2877static const struct dmi_system_id bridge_d3_blacklist[] = {
2878#ifdef CONFIG_X86
2879        {
2880                /*
2881                 * Gigabyte X299 root port is not marked as hotplug capable
2882                 * which allows Linux to power manage it.  However, this
2883                 * confuses the BIOS SMI handler so don't power manage root
2884                 * ports on that system.
2885                 */
2886                .ident = "X299 DESIGNARE EX-CF",
2887                .matches = {
2888                        DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2889                        DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2890                },
2891        },
2892#endif
2893        { }
2894};
2895
2896/**
2897 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2898 * @bridge: Bridge to check
2899 *
2900 * This function checks if it is possible to move the bridge to D3.
2901 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2902 */
2903bool pci_bridge_d3_possible(struct pci_dev *bridge)
2904{
2905        if (!pci_is_pcie(bridge))
2906                return false;
2907
2908        switch (pci_pcie_type(bridge)) {
2909        case PCI_EXP_TYPE_ROOT_PORT:
2910        case PCI_EXP_TYPE_UPSTREAM:
2911        case PCI_EXP_TYPE_DOWNSTREAM:
2912                if (pci_bridge_d3_disable)
2913                        return false;
2914
2915                /*
2916                 * Hotplug ports handled by firmware in System Management Mode
2917                 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2918                 */
2919                if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2920                        return false;
2921
2922                if (pci_bridge_d3_force)
2923                        return true;
2924
2925                /* Even the oldest 2010 Thunderbolt controller supports D3. */
2926                if (bridge->is_thunderbolt)
2927                        return true;
2928
2929                /* Platform might know better if the bridge supports D3 */
2930                if (platform_pci_bridge_d3(bridge))
2931                        return true;
2932
2933                /*
2934                 * Hotplug ports handled natively by the OS were not validated
2935                 * by vendors for runtime D3 at least until 2018 because there
2936                 * was no OS support.
2937                 */
2938                if (bridge->is_hotplug_bridge)
2939                        return false;
2940
2941                if (dmi_check_system(bridge_d3_blacklist))
2942                        return false;
2943
2944                /*
2945                 * It should be safe to put PCIe ports from 2015 or newer
2946                 * to D3.
2947                 */
2948                if (dmi_get_bios_year() >= 2015)
2949                        return true;
2950                break;
2951        }
2952
2953        return false;
2954}
2955
2956static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2957{
2958        bool *d3cold_ok = data;
2959
2960        if (/* The device needs to be allowed to go D3cold ... */
2961            dev->no_d3cold || !dev->d3cold_allowed ||
2962
2963            /* ... and if it is wakeup capable to do so from D3cold. */
2964            (device_may_wakeup(&dev->dev) &&
2965             !pci_pme_capable(dev, PCI_D3cold)) ||
2966
2967            /* If it is a bridge it must be allowed to go to D3. */
2968            !pci_power_manageable(dev))
2969
2970                *d3cold_ok = false;
2971
2972        return !*d3cold_ok;
2973}
2974
2975/*
2976 * pci_bridge_d3_update - Update bridge D3 capabilities
2977 * @dev: PCI device which is changed
2978 *
2979 * Update upstream bridge PM capabilities accordingly depending on if the
2980 * device PM configuration was changed or the device is being removed.  The
2981 * change is also propagated upstream.
2982 */
2983void pci_bridge_d3_update(struct pci_dev *dev)
2984{
2985        bool remove = !device_is_registered(&dev->dev);
2986        struct pci_dev *bridge;
2987        bool d3cold_ok = true;
2988
2989        bridge = pci_upstream_bridge(dev);
2990        if (!bridge || !pci_bridge_d3_possible(bridge))
2991                return;
2992
2993        /*
2994         * If D3 is currently allowed for the bridge, removing one of its
2995         * children won't change that.
2996         */
2997        if (remove && bridge->bridge_d3)
2998                return;
2999
3000        /*
3001         * If D3 is currently allowed for the bridge and a child is added or
3002         * changed, disallowance of D3 can only be caused by that child, so
3003         * we only need to check that single device, not any of its siblings.
3004         *
3005         * If D3 is currently not allowed for the bridge, checking the device
3006         * first may allow us to skip checking its siblings.
3007         */
3008        if (!remove)
3009                pci_dev_check_d3cold(dev, &d3cold_ok);
3010
3011        /*
3012         * If D3 is currently not allowed for the bridge, this may be caused
3013         * either by the device being changed/removed or any of its siblings,
3014         * so we need to go through all children to find out if one of them
3015         * continues to block D3.
3016         */
3017        if (d3cold_ok && !bridge->bridge_d3)
3018                pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
3019                             &d3cold_ok);
3020
3021        if (bridge->bridge_d3 != d3cold_ok) {
3022                bridge->bridge_d3 = d3cold_ok;
3023                /* Propagate change to upstream bridges */
3024                pci_bridge_d3_update(bridge);
3025        }
3026}
3027
3028/**
3029 * pci_d3cold_enable - Enable D3cold for device
3030 * @dev: PCI device to handle
3031 *
3032 * This function can be used in drivers to enable D3cold from the device
3033 * they handle.  It also updates upstream PCI bridge PM capabilities
3034 * accordingly.
3035 */
3036void pci_d3cold_enable(struct pci_dev *dev)
3037{
3038        if (dev->no_d3cold) {
3039                dev->no_d3cold = false;
3040                pci_bridge_d3_update(dev);
3041        }
3042}
3043EXPORT_SYMBOL_GPL(pci_d3cold_enable);
3044
3045/**
3046 * pci_d3cold_disable - Disable D3cold for device
3047 * @dev: PCI device to handle
3048 *
3049 * This function can be used in drivers to disable D3cold from the device
3050 * they handle.  It also updates upstream PCI bridge PM capabilities
3051 * accordingly.
3052 */
3053void pci_d3cold_disable(struct pci_dev *dev)
3054{
3055        if (!dev->no_d3cold) {
3056                dev->no_d3cold = true;
3057                pci_bridge_d3_update(dev);
3058        }
3059}
3060EXPORT_SYMBOL_GPL(pci_d3cold_disable);
3061
3062/**
3063 * pci_pm_init - Initialize PM functions of given PCI device
3064 * @dev: PCI device to handle.
3065 */
3066void pci_pm_init(struct pci_dev *dev)
3067{
3068        int pm;
3069        u16 status;
3070        u16 pmc;
3071
3072        pm_runtime_forbid(&dev->dev);
3073        pm_runtime_set_active(&dev->dev);
3074        pm_runtime_enable(&dev->dev);
3075        device_enable_async_suspend(&dev->dev);
3076        dev->wakeup_prepared = false;
3077
3078        dev->pm_cap = 0;
3079        dev->pme_support = 0;
3080
3081        /* find PCI PM capability in list */
3082        pm = pci_find_capability(dev, PCI_CAP_ID_PM);
3083        if (!pm)
3084                return;
3085        /* Check device's ability to generate PME# */
3086        pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
3087
3088        if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
3089                pci_err(dev, "unsupported PM cap regs version (%u)\n",
3090                        pmc & PCI_PM_CAP_VER_MASK);
3091                return;
3092        }
3093
3094        dev->pm_cap = pm;
3095        dev->d3hot_delay = PCI_PM_D3HOT_WAIT;
3096        dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
3097        dev->bridge_d3 = pci_bridge_d3_possible(dev);
3098        dev->d3cold_allowed = true;
3099
3100        dev->d1_support = false;
3101        dev->d2_support = false;
3102        if (!pci_no_d1d2(dev)) {
3103                if (pmc & PCI_PM_CAP_D1)
3104                        dev->d1_support = true;
3105                if (pmc & PCI_PM_CAP_D2)
3106                        dev->d2_support = true;
3107
3108                if (dev->d1_support || dev->d2_support)
3109                        pci_info(dev, "supports%s%s\n",
3110                                   dev->d1_support ? " D1" : "",
3111                                   dev->d2_support ? " D2" : "");
3112        }
3113
3114        pmc &= PCI_PM_CAP_PME_MASK;
3115        if (pmc) {
3116                pci_info(dev, "PME# supported from%s%s%s%s%s\n",
3117                         (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
3118                         (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
3119                         (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
3120                         (pmc & PCI_PM_CAP_PME_D3hot) ? " D3hot" : "",
3121                         (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
3122                dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
3123                dev->pme_poll = true;
3124                /*
3125                 * Make device's PM flags reflect the wake-up capability, but
3126                 * let the user space enable it to wake up the system as needed.
3127                 */
3128                device_set_wakeup_capable(&dev->dev, true);
3129                /* Disable the PME# generation functionality */
3130                pci_pme_active(dev, false);
3131        }
3132
3133        pci_read_config_word(dev, PCI_STATUS, &status);
3134        if (status & PCI_STATUS_IMM_READY)
3135                dev->imm_ready = 1;
3136}
3137
3138static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
3139{
3140        unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
3141
3142        switch (prop) {
3143        case PCI_EA_P_MEM:
3144        case PCI_EA_P_VF_MEM:
3145                flags |= IORESOURCE_MEM;
3146                break;
3147        case PCI_EA_P_MEM_PREFETCH:
3148        case PCI_EA_P_VF_MEM_PREFETCH:
3149                flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
3150                break;
3151        case PCI_EA_P_IO:
3152                flags |= IORESOURCE_IO;
3153                break;
3154        default:
3155                return 0;
3156        }
3157
3158        return flags;
3159}
3160
3161static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
3162                                            u8 prop)
3163{
3164        if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
3165                return &dev->resource[bei];
3166#ifdef CONFIG_PCI_IOV
3167        else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
3168                 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
3169                return &dev->resource[PCI_IOV_RESOURCES +
3170                                      bei - PCI_EA_BEI_VF_BAR0];
3171#endif
3172        else if (bei == PCI_EA_BEI_ROM)
3173                return &dev->resource[PCI_ROM_RESOURCE];
3174        else
3175                return NULL;
3176}
3177
3178/* Read an Enhanced Allocation (EA) entry */
3179static int pci_ea_read(struct pci_dev *dev, int offset)
3180{
3181        struct resource *res;
3182        int ent_size, ent_offset = offset;
3183        resource_size_t start, end;
3184        unsigned long flags;
3185        u32 dw0, bei, base, max_offset;
3186        u8 prop;
3187        bool support_64 = (sizeof(resource_size_t) >= 8);
3188
3189        pci_read_config_dword(dev, ent_offset, &dw0);
3190        ent_offset += 4;
3191
3192        /* Entry size field indicates DWORDs after 1st */
3193        ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
3194
3195        if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
3196                goto out;
3197
3198        bei = (dw0 & PCI_EA_BEI) >> 4;
3199        prop = (dw0 & PCI_EA_PP) >> 8;
3200
3201        /*
3202         * If the Property is in the reserved range, try the Secondary
3203         * Property instead.
3204         */
3205        if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
3206                prop = (dw0 & PCI_EA_SP) >> 16;
3207        if (prop > PCI_EA_P_BRIDGE_IO)
3208                goto out;
3209
3210        res = pci_ea_get_resource(dev, bei, prop);
3211        if (!res) {
3212                pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
3213                goto out;
3214        }
3215
3216        flags = pci_ea_flags(dev, prop);
3217        if (!flags) {
3218                pci_err(dev, "Unsupported EA properties: %#x\n", prop);
3219                goto out;
3220        }
3221
3222        /* Read Base */
3223        pci_read_config_dword(dev, ent_offset, &base);
3224        start = (base & PCI_EA_FIELD_MASK);
3225        ent_offset += 4;
3226
3227        /* Read MaxOffset */
3228        pci_read_config_dword(dev, ent_offset, &max_offset);
3229        ent_offset += 4;
3230
3231        /* Read Base MSBs (if 64-bit entry) */
3232        if (base & PCI_EA_IS_64) {
3233                u32 base_upper;
3234
3235                pci_read_config_dword(dev, ent_offset, &base_upper);
3236                ent_offset += 4;
3237
3238                flags |= IORESOURCE_MEM_64;
3239
3240                /* entry starts above 32-bit boundary, can't use */
3241                if (!support_64 && base_upper)
3242                        goto out;
3243
3244                if (support_64)
3245                        start |= ((u64)base_upper << 32);
3246        }
3247
3248        end = start + (max_offset | 0x03);
3249
3250        /* Read MaxOffset MSBs (if 64-bit entry) */
3251        if (max_offset & PCI_EA_IS_64) {
3252                u32 max_offset_upper;
3253
3254                pci_read_config_dword(dev, ent_offset, &max_offset_upper);
3255                ent_offset += 4;
3256
3257                flags |= IORESOURCE_MEM_64;
3258
3259                /* entry too big, can't use */
3260                if (!support_64 && max_offset_upper)
3261                        goto out;
3262
3263                if (support_64)
3264                        end += ((u64)max_offset_upper << 32);
3265        }
3266
3267        if (end < start) {
3268                pci_err(dev, "EA Entry crosses address boundary\n");
3269                goto out;
3270        }
3271
3272        if (ent_size != ent_offset - offset) {
3273                pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
3274                        ent_size, ent_offset - offset);
3275                goto out;
3276        }
3277
3278        res->name = pci_name(dev);
3279        res->start = start;
3280        res->end = end;
3281        res->flags = flags;
3282
3283        if (bei <= PCI_EA_BEI_BAR5)
3284                pci_info(dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3285                           bei, res, prop);
3286        else if (bei == PCI_EA_BEI_ROM)
3287                pci_info(dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
3288                           res, prop);
3289        else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
3290                pci_info(dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3291                           bei - PCI_EA_BEI_VF_BAR0, res, prop);
3292        else
3293                pci_info(dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
3294                           bei, res, prop);
3295
3296out:
3297        return offset + ent_size;
3298}
3299
3300/* Enhanced Allocation Initialization */
3301void pci_ea_init(struct pci_dev *dev)
3302{
3303        int ea;
3304        u8 num_ent;
3305        int offset;
3306        int i;
3307
3308        /* find PCI EA capability in list */
3309        ea = pci_find_capability(dev, PCI_CAP_ID_EA);
3310        if (!ea)
3311                return;
3312
3313        /* determine the number of entries */
3314        pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
3315                                        &num_ent);
3316        num_ent &= PCI_EA_NUM_ENT_MASK;
3317
3318        offset = ea + PCI_EA_FIRST_ENT;
3319
3320        /* Skip DWORD 2 for type 1 functions */
3321        if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3322                offset += 4;
3323
3324        /* parse each EA entry */
3325        for (i = 0; i < num_ent; ++i)
3326                offset = pci_ea_read(dev, offset);
3327}
3328
3329static void pci_add_saved_cap(struct pci_dev *pci_dev,
3330        struct pci_cap_saved_state *new_cap)
3331{
3332        hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3333}
3334
3335/**
3336 * _pci_add_cap_save_buffer - allocate buffer for saving given
3337 *                            capability registers
3338 * @dev: the PCI device
3339 * @cap: the capability to allocate the buffer for
3340 * @extended: Standard or Extended capability ID
3341 * @size: requested size of the buffer
3342 */
3343static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3344                                    bool extended, unsigned int size)
3345{
3346        int pos;
3347        struct pci_cap_saved_state *save_state;
3348
3349        if (extended)
3350                pos = pci_find_ext_capability(dev, cap);
3351        else
3352                pos = pci_find_capability(dev, cap);
3353
3354        if (!pos)
3355                return 0;
3356
3357        save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3358        if (!save_state)
3359                return -ENOMEM;
3360
3361        save_state->cap.cap_nr = cap;
3362        save_state->cap.cap_extended = extended;
3363        save_state->cap.size = size;
3364        pci_add_saved_cap(dev, save_state);
3365
3366        return 0;
3367}
3368
3369int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3370{
3371        return _pci_add_cap_save_buffer(dev, cap, false, size);
3372}
3373
3374int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3375{
3376        return _pci_add_cap_save_buffer(dev, cap, true, size);
3377}
3378
3379/**
3380 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3381 * @dev: the PCI device
3382 */
3383void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3384{
3385        int error;
3386
3387        error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3388                                        PCI_EXP_SAVE_REGS * sizeof(u16));
3389        if (error)
3390                pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3391
3392        error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3393        if (error)
3394                pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3395
3396        error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3397                                            2 * sizeof(u16));
3398        if (error)
3399                pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3400
3401        pci_allocate_vc_save_buffers(dev);
3402}
3403
3404void pci_free_cap_save_buffers(struct pci_dev *dev)
3405{
3406        struct pci_cap_saved_state *tmp;
3407        struct hlist_node *n;
3408
3409        hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3410                kfree(tmp);
3411}
3412
3413/**
3414 * pci_configure_ari - enable or disable ARI forwarding
3415 * @dev: the PCI device
3416 *
3417 * If @dev and its upstream bridge both support ARI, enable ARI in the
3418 * bridge.  Otherwise, disable ARI in the bridge.
3419 */
3420void pci_configure_ari(struct pci_dev *dev)
3421{
3422        u32 cap;
3423        struct pci_dev *bridge;
3424
3425        if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3426                return;
3427
3428        bridge = dev->bus->self;
3429        if (!bridge)
3430                return;
3431
3432        pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3433        if (!(cap & PCI_EXP_DEVCAP2_ARI))
3434                return;
3435
3436        if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3437                pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3438                                         PCI_EXP_DEVCTL2_ARI);
3439                bridge->ari_enabled = 1;
3440        } else {
3441                pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3442                                           PCI_EXP_DEVCTL2_ARI);
3443                bridge->ari_enabled = 0;
3444        }
3445}
3446
3447static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3448{
3449        int pos;
3450        u16 cap, ctrl;
3451
3452        pos = pdev->acs_cap;
3453        if (!pos)
3454                return false;
3455
3456        /*
3457         * Except for egress control, capabilities are either required
3458         * or only required if controllable.  Features missing from the
3459         * capability field can therefore be assumed as hard-wired enabled.
3460         */
3461        pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3462        acs_flags &= (cap | PCI_ACS_EC);
3463
3464        pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3465        return (ctrl & acs_flags) == acs_flags;
3466}
3467
3468/**
3469 * pci_acs_enabled - test ACS against required flags for a given device
3470 * @pdev: device to test
3471 * @acs_flags: required PCI ACS flags
3472 *
3473 * Return true if the device supports the provided flags.  Automatically
3474 * filters out flags that are not implemented on multifunction devices.
3475 *
3476 * Note that this interface checks the effective ACS capabilities of the
3477 * device rather than the actual capabilities.  For instance, most single
3478 * function endpoints are not required to support ACS because they have no
3479 * opportunity for peer-to-peer access.  We therefore return 'true'
3480 * regardless of whether the device exposes an ACS capability.  This makes
3481 * it much easier for callers of this function to ignore the actual type
3482 * or topology of the device when testing ACS support.
3483 */
3484bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3485{
3486        int ret;
3487
3488        ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3489        if (ret >= 0)
3490                return ret > 0;
3491
3492        /*
3493         * Conventional PCI and PCI-X devices never support ACS, either
3494         * effectively or actually.  The shared bus topology implies that
3495         * any device on the bus can receive or snoop DMA.
3496         */
3497        if (!pci_is_pcie(pdev))
3498                return false;
3499
3500        switch (pci_pcie_type(pdev)) {
3501        /*
3502         * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3503         * but since their primary interface is PCI/X, we conservatively
3504         * handle them as we would a non-PCIe device.
3505         */
3506        case PCI_EXP_TYPE_PCIE_BRIDGE:
3507        /*
3508         * PCIe 3.0, 6.12.1 excludes ACS on these devices.  "ACS is never
3509         * applicable... must never implement an ACS Extended Capability...".
3510         * This seems arbitrary, but we take a conservative interpretation
3511         * of this statement.
3512         */
3513        case PCI_EXP_TYPE_PCI_BRIDGE:
3514        case PCI_EXP_TYPE_RC_EC:
3515                return false;
3516        /*
3517         * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3518         * implement ACS in order to indicate their peer-to-peer capabilities,
3519         * regardless of whether they are single- or multi-function devices.
3520         */
3521        case PCI_EXP_TYPE_DOWNSTREAM:
3522        case PCI_EXP_TYPE_ROOT_PORT:
3523                return pci_acs_flags_enabled(pdev, acs_flags);
3524        /*
3525         * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3526         * implemented by the remaining PCIe types to indicate peer-to-peer
3527         * capabilities, but only when they are part of a multifunction
3528         * device.  The footnote for section 6.12 indicates the specific
3529         * PCIe types included here.
3530         */
3531        case PCI_EXP_TYPE_ENDPOINT:
3532        case PCI_EXP_TYPE_UPSTREAM:
3533        case PCI_EXP_TYPE_LEG_END:
3534        case PCI_EXP_TYPE_RC_END:
3535                if (!pdev->multifunction)
3536                        break;
3537
3538                return pci_acs_flags_enabled(pdev, acs_flags);
3539        }
3540
3541        /*
3542         * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3543         * to single function devices with the exception of downstream ports.
3544         */
3545        return true;
3546}
3547
3548/**
3549 * pci_acs_path_enabled - test ACS flags from start to end in a hierarchy
3550 * @start: starting downstream device
3551 * @end: ending upstream device or NULL to search to the root bus
3552 * @acs_flags: required flags
3553 *
3554 * Walk up a device tree from start to end testing PCI ACS support.  If
3555 * any step along the way does not support the required flags, return false.
3556 */
3557bool pci_acs_path_enabled(struct pci_dev *start,
3558                          struct pci_dev *end, u16 acs_flags)
3559{
3560        struct pci_dev *pdev, *parent = start;
3561
3562        do {
3563                pdev = parent;
3564
3565                if (!pci_acs_enabled(pdev, acs_flags))
3566                        return false;
3567
3568                if (pci_is_root_bus(pdev->bus))
3569                        return (end == NULL);
3570
3571                parent = pdev->bus->self;
3572        } while (pdev != end);
3573
3574        return true;
3575}
3576
3577/**
3578 * pci_acs_init - Initialize ACS if hardware supports it
3579 * @dev: the PCI device
3580 */
3581void pci_acs_init(struct pci_dev *dev)
3582{
3583        dev->acs_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3584
3585        /*
3586         * Attempt to enable ACS regardless of capability because some Root
3587         * Ports (e.g. those quirked with *_intel_pch_acs_*) do not have
3588         * the standard ACS capability but still support ACS via those
3589         * quirks.
3590         */
3591        pci_enable_acs(dev);
3592}
3593
3594/**
3595 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3596 * @pdev: PCI device
3597 * @bar: BAR to find
3598 *
3599 * Helper to find the position of the ctrl register for a BAR.
3600 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3601 * Returns -ENOENT if no ctrl register for the BAR could be found.
3602 */
3603static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3604{
3605        unsigned int pos, nbars, i;
3606        u32 ctrl;
3607
3608        pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3609        if (!pos)
3610                return -ENOTSUPP;
3611
3612        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3613        nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3614                    PCI_REBAR_CTRL_NBAR_SHIFT;
3615
3616        for (i = 0; i < nbars; i++, pos += 8) {
3617                int bar_idx;
3618
3619                pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3620                bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3621                if (bar_idx == bar)
3622                        return pos;
3623        }
3624
3625        return -ENOENT;
3626}
3627
3628/**
3629 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3630 * @pdev: PCI device
3631 * @bar: BAR to query
3632 *
3633 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3634 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3635 */
3636u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3637{
3638        int pos;
3639        u32 cap;
3640
3641        pos = pci_rebar_find_pos(pdev, bar);
3642        if (pos < 0)
3643                return 0;
3644
3645        pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3646        cap &= PCI_REBAR_CAP_SIZES;
3647
3648        /* Sapphire RX 5600 XT Pulse has an invalid cap dword for BAR 0 */
3649        if (pdev->vendor == PCI_VENDOR_ID_ATI && pdev->device == 0x731f &&
3650            bar == 0 && cap == 0x7000)
3651                cap = 0x3f000;
3652
3653        return cap >> 4;
3654}
3655EXPORT_SYMBOL(pci_rebar_get_possible_sizes);
3656
3657/**
3658 * pci_rebar_get_current_size - get the current size of a BAR
3659 * @pdev: PCI device
3660 * @bar: BAR to set size to
3661 *
3662 * Read the size of a BAR from the resizable BAR config.
3663 * Returns size if found or negative error code.
3664 */
3665int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3666{
3667        int pos;
3668        u32 ctrl;
3669
3670        pos = pci_rebar_find_pos(pdev, bar);
3671        if (pos < 0)
3672                return pos;
3673
3674        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3675        return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3676}
3677
3678/**
3679 * pci_rebar_set_size - set a new size for a BAR
3680 * @pdev: PCI device
3681 * @bar: BAR to set size to
3682 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3683 *
3684 * Set the new size of a BAR as defined in the spec.
3685 * Returns zero if resizing was successful, error code otherwise.
3686 */
3687int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3688{
3689        int pos;
3690        u32 ctrl;
3691
3692        pos = pci_rebar_find_pos(pdev, bar);
3693        if (pos < 0)
3694                return pos;
3695
3696        pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3697        ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3698        ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3699        pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3700        return 0;
3701}
3702
3703/**
3704 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3705 * @dev: the PCI device
3706 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3707 *      PCI_EXP_DEVCAP2_ATOMIC_COMP32
3708 *      PCI_EXP_DEVCAP2_ATOMIC_COMP64
3709 *      PCI_EXP_DEVCAP2_ATOMIC_COMP128
3710 *
3711 * Return 0 if all upstream bridges support AtomicOp routing, egress
3712 * blocking is disabled on all upstream ports, and the root port supports
3713 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3714 * AtomicOp completion), or negative otherwise.
3715 */
3716int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3717{
3718        struct pci_bus *bus = dev->bus;
3719        struct pci_dev *bridge;
3720        u32 cap, ctl2;
3721
3722        if (!pci_is_pcie(dev))
3723                return -EINVAL;
3724
3725        /*
3726         * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3727         * AtomicOp requesters.  For now, we only support endpoints as
3728         * requesters and root ports as completers.  No endpoints as
3729         * completers, and no peer-to-peer.
3730         */
3731
3732        switch (pci_pcie_type(dev)) {
3733        case PCI_EXP_TYPE_ENDPOINT:
3734        case PCI_EXP_TYPE_LEG_END:
3735        case PCI_EXP_TYPE_RC_END:
3736                break;
3737        default:
3738                return -EINVAL;
3739        }
3740
3741        while (bus->parent) {
3742                bridge = bus->self;
3743
3744                pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3745
3746                switch (pci_pcie_type(bridge)) {
3747                /* Ensure switch ports support AtomicOp routing */
3748                case PCI_EXP_TYPE_UPSTREAM:
3749                case PCI_EXP_TYPE_DOWNSTREAM:
3750                        if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3751                                return -EINVAL;
3752                        break;
3753
3754                /* Ensure root port supports all the sizes we care about */
3755                case PCI_EXP_TYPE_ROOT_PORT:
3756                        if ((cap & cap_mask) != cap_mask)
3757                                return -EINVAL;
3758                        break;
3759                }
3760
3761                /* Ensure upstream ports don't block AtomicOps on egress */
3762                if (pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM) {
3763                        pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3764                                                   &ctl2);
3765                        if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3766                                return -EINVAL;
3767                }
3768
3769                bus = bus->parent;
3770        }
3771
3772        pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3773                                 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3774        return 0;
3775}
3776EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3777
3778/**
3779 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3780 * @dev: the PCI device
3781 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3782 *
3783 * Perform INTx swizzling for a device behind one level of bridge.  This is
3784 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3785 * behind bridges on add-in cards.  For devices with ARI enabled, the slot
3786 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3787 * the PCI Express Base Specification, Revision 2.1)
3788 */
3789u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3790{
3791        int slot;
3792
3793        if (pci_ari_enabled(dev->bus))
3794                slot = 0;
3795        else
3796                slot = PCI_SLOT(dev->devfn);
3797
3798        return (((pin - 1) + slot) % 4) + 1;
3799}
3800
3801int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3802{
3803        u8 pin;
3804
3805        pin = dev->pin;
3806        if (!pin)
3807                return -1;
3808
3809        while (!pci_is_root_bus(dev->bus)) {
3810                pin = pci_swizzle_interrupt_pin(dev, pin);
3811                dev = dev->bus->self;
3812        }
3813        *bridge = dev;
3814        return pin;
3815}
3816
3817/**
3818 * pci_common_swizzle - swizzle INTx all the way to root bridge
3819 * @dev: the PCI device
3820 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3821 *
3822 * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
3823 * bridges all the way up to a PCI root bus.
3824 */
3825u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3826{
3827        u8 pin = *pinp;
3828
3829        while (!pci_is_root_bus(dev->bus)) {
3830                pin = pci_swizzle_interrupt_pin(dev, pin);
3831                dev = dev->bus->self;
3832        }
3833        *pinp = pin;
3834        return PCI_SLOT(dev->devfn);
3835}
3836EXPORT_SYMBOL_GPL(pci_common_swizzle);
3837
3838/**
3839 * pci_release_region - Release a PCI bar
3840 * @pdev: PCI device whose resources were previously reserved by
3841 *        pci_request_region()
3842 * @bar: BAR to release
3843 *
3844 * Releases the PCI I/O and memory resources previously reserved by a
3845 * successful call to pci_request_region().  Call this function only
3846 * after all use of the PCI regions has ceased.
3847 */
3848void pci_release_region(struct pci_dev *pdev, int bar)
3849{
3850        struct pci_devres *dr;
3851
3852        if (pci_resource_len(pdev, bar) == 0)
3853                return;
3854        if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3855                release_region(pci_resource_start(pdev, bar),
3856                                pci_resource_len(pdev, bar));
3857        else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3858                release_mem_region(pci_resource_start(pdev, bar),
3859                                pci_resource_len(pdev, bar));
3860
3861        dr = find_pci_dr(pdev);
3862        if (dr)
3863                dr->region_mask &= ~(1 << bar);
3864}
3865EXPORT_SYMBOL(pci_release_region);
3866
3867/**
3868 * __pci_request_region - Reserved PCI I/O and memory resource
3869 * @pdev: PCI device whose resources are to be reserved
3870 * @bar: BAR to be reserved
3871 * @res_name: Name to be associated with resource.
3872 * @exclusive: whether the region access is exclusive or not
3873 *
3874 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3875 * being reserved by owner @res_name.  Do not access any
3876 * address inside the PCI regions unless this call returns
3877 * successfully.
3878 *
3879 * If @exclusive is set, then the region is marked so that userspace
3880 * is explicitly not allowed to map the resource via /dev/mem or
3881 * sysfs MMIO access.
3882 *
3883 * Returns 0 on success, or %EBUSY on error.  A warning
3884 * message is also printed on failure.
3885 */
3886static int __pci_request_region(struct pci_dev *pdev, int bar,
3887                                const char *res_name, int exclusive)
3888{
3889        struct pci_devres *dr;
3890
3891        if (pci_resource_len(pdev, bar) == 0)
3892                return 0;
3893
3894        if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3895                if (!request_region(pci_resource_start(pdev, bar),
3896                            pci_resource_len(pdev, bar), res_name))
3897                        goto err_out;
3898        } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3899                if (!__request_mem_region(pci_resource_start(pdev, bar),
3900                                        pci_resource_len(pdev, bar), res_name,
3901                                        exclusive))
3902                        goto err_out;
3903        }
3904
3905        dr = find_pci_dr(pdev);
3906        if (dr)
3907                dr->region_mask |= 1 << bar;
3908
3909        return 0;
3910
3911err_out:
3912        pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3913                 &pdev->resource[bar]);
3914        return -EBUSY;
3915}
3916
3917/**
3918 * pci_request_region - Reserve PCI I/O and memory resource
3919 * @pdev: PCI device whose resources are to be reserved
3920 * @bar: BAR to be reserved
3921 * @res_name: Name to be associated with resource
3922 *
3923 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3924 * being reserved by owner @res_name.  Do not access any
3925 * address inside the PCI regions unless this call returns
3926 * successfully.
3927 *
3928 * Returns 0 on success, or %EBUSY on error.  A warning
3929 * message is also printed on failure.
3930 */
3931int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3932{
3933        return __pci_request_region(pdev, bar, res_name, 0);
3934}
3935EXPORT_SYMBOL(pci_request_region);
3936
3937/**
3938 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3939 * @pdev: PCI device whose resources were previously reserved
3940 * @bars: Bitmask of BARs to be released
3941 *
3942 * Release selected PCI I/O and memory resources previously reserved.
3943 * Call this function only after all use of the PCI regions has ceased.
3944 */
3945void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3946{
3947        int i;
3948
3949        for (i = 0; i < PCI_STD_NUM_BARS; i++)
3950                if (bars & (1 << i))
3951                        pci_release_region(pdev, i);
3952}
3953EXPORT_SYMBOL(pci_release_selected_regions);
3954
3955static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3956                                          const char *res_name, int excl)
3957{
3958        int i;
3959
3960        for (i = 0; i < PCI_STD_NUM_BARS; i++)
3961                if (bars & (1 << i))
3962                        if (__pci_request_region(pdev, i, res_name, excl))
3963                                goto err_out;
3964        return 0;
3965
3966err_out:
3967        while (--i >= 0)
3968                if (bars & (1 << i))
3969                        pci_release_region(pdev, i);
3970
3971        return -EBUSY;
3972}
3973
3974
3975/**
3976 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3977 * @pdev: PCI device whose resources are to be reserved
3978 * @bars: Bitmask of BARs to be requested
3979 * @res_name: Name to be associated with resource
3980 */
3981int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3982                                 const char *res_name)
3983{
3984        return __pci_request_selected_regions(pdev, bars, res_name, 0);
3985}
3986EXPORT_SYMBOL(pci_request_selected_regions);
3987
3988int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3989                                           const char *res_name)
3990{
3991        return __pci_request_selected_regions(pdev, bars, res_name,
3992                        IORESOURCE_EXCLUSIVE);
3993}
3994EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3995
3996/**
3997 * pci_release_regions - Release reserved PCI I/O and memory resources
3998 * @pdev: PCI device whose resources were previously reserved by
3999 *        pci_request_regions()
4000 *
4001 * Releases all PCI I/O and memory resources previously reserved by a
4002 * successful call to pci_request_regions().  Call this function only
4003 * after all use of the PCI regions has ceased.
4004 */
4005
4006void pci_release_regions(struct pci_dev *pdev)
4007{
4008        pci_release_selected_regions(pdev, (1 << PCI_STD_NUM_BARS) - 1);
4009}
4010EXPORT_SYMBOL(pci_release_regions);
4011
4012/**
4013 * pci_request_regions - Reserve PCI I/O and memory resources
4014 * @pdev: PCI device whose resources are to be reserved
4015 * @res_name: Name to be associated with resource.
4016 *
4017 * Mark all PCI regions associated with PCI device @pdev as
4018 * being reserved by owner @res_name.  Do not access any
4019 * address inside the PCI regions unless this call returns
4020 * successfully.
4021 *
4022 * Returns 0 on success, or %EBUSY on error.  A warning
4023 * message is also printed on failure.
4024 */
4025int pci_request_regions(struct pci_dev *pdev, const char *res_name)
4026{
4027        return pci_request_selected_regions(pdev,
4028                        ((1 << PCI_STD_NUM_BARS) - 1), res_name);
4029}
4030EXPORT_SYMBOL(pci_request_regions);
4031
4032/**
4033 * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
4034 * @pdev: PCI device whose resources are to be reserved
4035 * @res_name: Name to be associated with resource.
4036 *
4037 * Mark all PCI regions associated with PCI device @pdev as being reserved
4038 * by owner @res_name.  Do not access any address inside the PCI regions
4039 * unless this call returns successfully.
4040 *
4041 * pci_request_regions_exclusive() will mark the region so that /dev/mem
4042 * and the sysfs MMIO access will not be allowed.
4043 *
4044 * Returns 0 on success, or %EBUSY on error.  A warning message is also
4045 * printed on failure.
4046 */
4047int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
4048{
4049        return pci_request_selected_regions_exclusive(pdev,
4050                                ((1 << PCI_STD_NUM_BARS) - 1), res_name);
4051}
4052EXPORT_SYMBOL(pci_request_regions_exclusive);
4053
4054/*
4055 * Record the PCI IO range (expressed as CPU physical address + size).
4056 * Return a negative value if an error has occurred, zero otherwise
4057 */
4058int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
4059                        resource_size_t size)
4060{
4061        int ret = 0;
4062#ifdef PCI_IOBASE
4063        struct logic_pio_hwaddr *range;
4064
4065        if (!size || addr + size < addr)
4066                return -EINVAL;
4067
4068        range = kzalloc(sizeof(*range), GFP_ATOMIC);
4069        if (!range)
4070                return -ENOMEM;
4071
4072        range->fwnode = fwnode;
4073        range->size = size;
4074        range->hw_start = addr;
4075        range->flags = LOGIC_PIO_CPU_MMIO;
4076
4077        ret = logic_pio_register_range(range);
4078        if (ret)
4079                kfree(range);
4080
4081        /* Ignore duplicates due to deferred probing */
4082        if (ret == -EEXIST)
4083                ret = 0;
4084#endif
4085
4086        return ret;
4087}
4088
4089phys_addr_t pci_pio_to_address(unsigned long pio)
4090{
4091        phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
4092
4093#ifdef PCI_IOBASE
4094        if (pio >= MMIO_UPPER_LIMIT)
4095                return address;
4096
4097        address = logic_pio_to_hwaddr(pio);
4098#endif
4099
4100        return address;
4101}
4102EXPORT_SYMBOL_GPL(pci_pio_to_address);
4103
4104unsigned long __weak pci_address_to_pio(phys_addr_t address)
4105{
4106#ifdef PCI_IOBASE
4107        return logic_pio_trans_cpuaddr(address);
4108#else
4109        if (address > IO_SPACE_LIMIT)
4110                return (unsigned long)-1;
4111
4112        return (unsigned long) address;
4113#endif
4114}
4115
4116/**
4117 * pci_remap_iospace - Remap the memory mapped I/O space
4118 * @res: Resource describing the I/O space
4119 * @phys_addr: physical address of range to be mapped
4120 *
4121 * Remap the memory mapped I/O space described by the @res and the CPU
4122 * physical address @phys_addr into virtual address space.  Only
4123 * architectures that have memory mapped IO functions defined (and the
4124 * PCI_IOBASE value defined) should call this function.
4125 */
4126int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
4127{
4128#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4129        unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4130
4131        if (!(res->flags & IORESOURCE_IO))
4132                return -EINVAL;
4133
4134        if (res->end > IO_SPACE_LIMIT)
4135                return -EINVAL;
4136
4137        return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
4138                                  pgprot_device(PAGE_KERNEL));
4139#else
4140        /*
4141         * This architecture does not have memory mapped I/O space,
4142         * so this function should never be called
4143         */
4144        WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
4145        return -ENODEV;
4146#endif
4147}
4148EXPORT_SYMBOL(pci_remap_iospace);
4149
4150/**
4151 * pci_unmap_iospace - Unmap the memory mapped I/O space
4152 * @res: resource to be unmapped
4153 *
4154 * Unmap the CPU virtual address @res from virtual address space.  Only
4155 * architectures that have memory mapped IO functions defined (and the
4156 * PCI_IOBASE value defined) should call this function.
4157 */
4158void pci_unmap_iospace(struct resource *res)
4159{
4160#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4161        unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4162
4163        vunmap_range(vaddr, vaddr + resource_size(res));
4164#endif
4165}
4166EXPORT_SYMBOL(pci_unmap_iospace);
4167
4168static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
4169{
4170        struct resource **res = ptr;
4171
4172        pci_unmap_iospace(*res);
4173}
4174
4175/**
4176 * devm_pci_remap_iospace - Managed pci_remap_iospace()
4177 * @dev: Generic device to remap IO address for
4178 * @res: Resource describing the I/O space
4179 * @phys_addr: physical address of range to be mapped
4180 *
4181 * Managed pci_remap_iospace().  Map is automatically unmapped on driver
4182 * detach.
4183 */
4184int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
4185                           phys_addr_t phys_addr)
4186{
4187        const struct resource **ptr;
4188        int error;
4189
4190        ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
4191        if (!ptr)
4192                return -ENOMEM;
4193
4194        error = pci_remap_iospace(res, phys_addr);
4195        if (error) {
4196                devres_free(ptr);
4197        } else  {
4198                *ptr = res;
4199                devres_add(dev, ptr);
4200        }
4201
4202        return error;
4203}
4204EXPORT_SYMBOL(devm_pci_remap_iospace);
4205
4206/**
4207 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
4208 * @dev: Generic device to remap IO address for
4209 * @offset: Resource address to map
4210 * @size: Size of map
4211 *
4212 * Managed pci_remap_cfgspace().  Map is automatically unmapped on driver
4213 * detach.
4214 */
4215void __iomem *devm_pci_remap_cfgspace(struct device *dev,
4216                                      resource_size_t offset,
4217                                      resource_size_t size)
4218{
4219        void __iomem **ptr, *addr;
4220
4221        ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
4222        if (!ptr)
4223                return NULL;
4224
4225        addr = pci_remap_cfgspace(offset, size);
4226        if (addr) {
4227                *ptr = addr;
4228                devres_add(dev, ptr);
4229        } else
4230                devres_free(ptr);
4231
4232        return addr;
4233}
4234EXPORT_SYMBOL(devm_pci_remap_cfgspace);
4235
4236/**
4237 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
4238 * @dev: generic device to handle the resource for
4239 * @res: configuration space resource to be handled
4240 *
4241 * Checks that a resource is a valid memory region, requests the memory
4242 * region and ioremaps with pci_remap_cfgspace() API that ensures the
4243 * proper PCI configuration space memory attributes are guaranteed.
4244 *
4245 * All operations are managed and will be undone on driver detach.
4246 *
4247 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
4248 * on failure. Usage example::
4249 *
4250 *      res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
4251 *      base = devm_pci_remap_cfg_resource(&pdev->dev, res);
4252 *      if (IS_ERR(base))
4253 *              return PTR_ERR(base);
4254 */
4255void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
4256                                          struct resource *res)
4257{
4258        resource_size_t size;
4259        const char *name;
4260        void __iomem *dest_ptr;
4261
4262        BUG_ON(!dev);
4263
4264        if (!res || resource_type(res) != IORESOURCE_MEM) {
4265                dev_err(dev, "invalid resource\n");
4266                return IOMEM_ERR_PTR(-EINVAL);
4267        }
4268
4269        size = resource_size(res);
4270
4271        if (res->name)
4272                name = devm_kasprintf(dev, GFP_KERNEL, "%s %s", dev_name(dev),
4273                                      res->name);
4274        else
4275                name = devm_kstrdup(dev, dev_name(dev), GFP_KERNEL);
4276        if (!name)
4277                return IOMEM_ERR_PTR(-ENOMEM);
4278
4279        if (!devm_request_mem_region(dev, res->start, size, name)) {
4280                dev_err(dev, "can't request region for resource %pR\n", res);
4281                return IOMEM_ERR_PTR(-EBUSY);
4282        }
4283
4284        dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4285        if (!dest_ptr) {
4286                dev_err(dev, "ioremap failed for resource %pR\n", res);
4287                devm_release_mem_region(dev, res->start, size);
4288                dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4289        }
4290
4291        return dest_ptr;
4292}
4293EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4294
4295static void __pci_set_master(struct pci_dev *dev, bool enable)
4296{
4297        u16 old_cmd, cmd;
4298
4299        pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4300        if (enable)
4301                cmd = old_cmd | PCI_COMMAND_MASTER;
4302        else
4303                cmd = old_cmd & ~PCI_COMMAND_MASTER;
4304        if (cmd != old_cmd) {
4305                pci_dbg(dev, "%s bus mastering\n",
4306                        enable ? "enabling" : "disabling");
4307                pci_write_config_word(dev, PCI_COMMAND, cmd);
4308        }
4309        dev->is_busmaster = enable;
4310}
4311
4312/**
4313 * pcibios_setup - process "pci=" kernel boot arguments
4314 * @str: string used to pass in "pci=" kernel boot arguments
4315 *
4316 * Process kernel boot arguments.  This is the default implementation.
4317 * Architecture specific implementations can override this as necessary.
4318 */
4319char * __weak __init pcibios_setup(char *str)
4320{
4321        return str;
4322}
4323
4324/**
4325 * pcibios_set_master - enable PCI bus-mastering for device dev
4326 * @dev: the PCI device to enable
4327 *
4328 * Enables PCI bus-mastering for the device.  This is the default
4329 * implementation.  Architecture specific implementations can override
4330 * this if necessary.
4331 */
4332void __weak pcibios_set_master(struct pci_dev *dev)
4333{
4334        u8 lat;
4335
4336        /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4337        if (pci_is_pcie(dev))
4338                return;
4339
4340        pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4341        if (lat < 16)
4342                lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4343        else if (lat > pcibios_max_latency)
4344                lat = pcibios_max_latency;
4345        else
4346                return;
4347
4348        pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4349}
4350
4351/**
4352 * pci_set_master - enables bus-mastering for device dev
4353 * @dev: the PCI device to enable
4354 *
4355 * Enables bus-mastering on the device and calls pcibios_set_master()
4356 * to do the needed arch specific settings.
4357 */
4358void pci_set_master(struct pci_dev *dev)
4359{
4360        __pci_set_master(dev, true);
4361        pcibios_set_master(dev);
4362}
4363EXPORT_SYMBOL(pci_set_master);
4364
4365/**
4366 * pci_clear_master - disables bus-mastering for device dev
4367 * @dev: the PCI device to disable
4368 */
4369void pci_clear_master(struct pci_dev *dev)
4370{
4371        __pci_set_master(dev, false);
4372}
4373EXPORT_SYMBOL(pci_clear_master);
4374
4375/**
4376 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4377 * @dev: the PCI device for which MWI is to be enabled
4378 *
4379 * Helper function for pci_set_mwi.
4380 * Originally copied from drivers/net/acenic.c.
4381 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4382 *
4383 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4384 */
4385int pci_set_cacheline_size(struct pci_dev *dev)
4386{
4387        u8 cacheline_size;
4388
4389        if (!pci_cache_line_size)
4390                return -EINVAL;
4391
4392        /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4393           equal to or multiple of the right value. */
4394        pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4395        if (cacheline_size >= pci_cache_line_size &&
4396            (cacheline_size % pci_cache_line_size) == 0)
4397                return 0;
4398
4399        /* Write the correct value. */
4400        pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4401        /* Read it back. */
4402        pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4403        if (cacheline_size == pci_cache_line_size)
4404                return 0;
4405
4406        pci_dbg(dev, "cache line size of %d is not supported\n",
4407                   pci_cache_line_size << 2);
4408
4409        return -EINVAL;
4410}
4411EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4412
4413/**
4414 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4415 * @dev: the PCI device for which MWI is enabled
4416 *
4417 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4418 *
4419 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4420 */
4421int pci_set_mwi(struct pci_dev *dev)
4422{
4423#ifdef PCI_DISABLE_MWI
4424        return 0;
4425#else
4426        int rc;
4427        u16 cmd;
4428
4429        rc = pci_set_cacheline_size(dev);
4430        if (rc)
4431                return rc;
4432
4433        pci_read_config_word(dev, PCI_COMMAND, &cmd);
4434        if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4435                pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4436                cmd |= PCI_COMMAND_INVALIDATE;
4437                pci_write_config_word(dev, PCI_COMMAND, cmd);
4438        }
4439        return 0;
4440#endif
4441}
4442EXPORT_SYMBOL(pci_set_mwi);
4443
4444/**
4445 * pcim_set_mwi - a device-managed pci_set_mwi()
4446 * @dev: the PCI device for which MWI is enabled
4447 *
4448 * Managed pci_set_mwi().
4449 *
4450 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4451 */
4452int pcim_set_mwi(struct pci_dev *dev)
4453{
4454        struct pci_devres *dr;
4455
4456        dr = find_pci_dr(dev);
4457        if (!dr)
4458                return -ENOMEM;
4459
4460        dr->mwi = 1;
4461        return pci_set_mwi(dev);
4462}
4463EXPORT_SYMBOL(pcim_set_mwi);
4464
4465/**
4466 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4467 * @dev: the PCI device for which MWI is enabled
4468 *
4469 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4470 * Callers are not required to check the return value.
4471 *
4472 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4473 */
4474int pci_try_set_mwi(struct pci_dev *dev)
4475{
4476#ifdef PCI_DISABLE_MWI
4477        return 0;
4478#else
4479        return pci_set_mwi(dev);
4480#endif
4481}
4482EXPORT_SYMBOL(pci_try_set_mwi);
4483
4484/**
4485 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4486 * @dev: the PCI device to disable
4487 *
4488 * Disables PCI Memory-Write-Invalidate transaction on the device
4489 */
4490void pci_clear_mwi(struct pci_dev *dev)
4491{
4492#ifndef PCI_DISABLE_MWI
4493        u16 cmd;
4494
4495        pci_read_config_word(dev, PCI_COMMAND, &cmd);
4496        if (cmd & PCI_COMMAND_INVALIDATE) {
4497                cmd &= ~PCI_COMMAND_INVALIDATE;
4498                pci_write_config_word(dev, PCI_COMMAND, cmd);
4499        }
4500#endif
4501}
4502EXPORT_SYMBOL(pci_clear_mwi);
4503
4504/**
4505 * pci_disable_parity - disable parity checking for device
4506 * @dev: the PCI device to operate on
4507 *
4508 * Disable parity checking for device @dev
4509 */
4510void pci_disable_parity(struct pci_dev *dev)
4511{
4512        u16 cmd;
4513
4514        pci_read_config_word(dev, PCI_COMMAND, &cmd);
4515        if (cmd & PCI_COMMAND_PARITY) {
4516                cmd &= ~PCI_COMMAND_PARITY;
4517                pci_write_config_word(dev, PCI_COMMAND, cmd);
4518        }
4519}
4520
4521/**
4522 * pci_intx - enables/disables PCI INTx for device dev
4523 * @pdev: the PCI device to operate on
4524 * @enable: boolean: whether to enable or disable PCI INTx
4525 *
4526 * Enables/disables PCI INTx for device @pdev
4527 */
4528void pci_intx(struct pci_dev *pdev, int enable)
4529{
4530        u16 pci_command, new;
4531
4532        pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4533
4534        if (enable)
4535                new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4536        else
4537                new = pci_command | PCI_COMMAND_INTX_DISABLE;
4538
4539        if (new != pci_command) {
4540                struct pci_devres *dr;
4541
4542                pci_write_config_word(pdev, PCI_COMMAND, new);
4543
4544                dr = find_pci_dr(pdev);
4545                if (dr && !dr->restore_intx) {
4546                        dr->restore_intx = 1;
4547                        dr->orig_intx = !enable;
4548                }
4549        }
4550}
4551EXPORT_SYMBOL_GPL(pci_intx);
4552
4553static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4554{
4555        struct pci_bus *bus = dev->bus;
4556        bool mask_updated = true;
4557        u32 cmd_status_dword;
4558        u16 origcmd, newcmd;
4559        unsigned long flags;
4560        bool irq_pending;
4561
4562        /*
4563         * We do a single dword read to retrieve both command and status.
4564         * Document assumptions that make this possible.
4565         */
4566        BUILD_BUG_ON(PCI_COMMAND % 4);
4567        BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4568
4569        raw_spin_lock_irqsave(&pci_lock, flags);
4570
4571        bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4572
4573        irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4574
4575        /*
4576         * Check interrupt status register to see whether our device
4577         * triggered the interrupt (when masking) or the next IRQ is
4578         * already pending (when unmasking).
4579         */
4580        if (mask != irq_pending) {
4581                mask_updated = false;
4582                goto done;
4583        }
4584
4585        origcmd = cmd_status_dword;
4586        newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4587        if (mask)
4588                newcmd |= PCI_COMMAND_INTX_DISABLE;
4589        if (newcmd != origcmd)
4590                bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4591
4592done:
4593        raw_spin_unlock_irqrestore(&pci_lock, flags);
4594
4595        return mask_updated;
4596}
4597
4598/**
4599 * pci_check_and_mask_intx - mask INTx on pending interrupt
4600 * @dev: the PCI device to operate on
4601 *
4602 * Check if the device dev has its INTx line asserted, mask it and return
4603 * true in that case. False is returned if no interrupt was pending.
4604 */
4605bool pci_check_and_mask_intx(struct pci_dev *dev)
4606{
4607        return pci_check_and_set_intx_mask(dev, true);
4608}
4609EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4610
4611/**
4612 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4613 * @dev: the PCI device to operate on
4614 *
4615 * Check if the device dev has its INTx line asserted, unmask it if not and
4616 * return true. False is returned and the mask remains active if there was
4617 * still an interrupt pending.
4618 */
4619bool pci_check_and_unmask_intx(struct pci_dev *dev)
4620{
4621        return pci_check_and_set_intx_mask(dev, false);
4622}
4623EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4624
4625/**
4626 * pci_wait_for_pending_transaction - wait for pending transaction
4627 * @dev: the PCI device to operate on
4628 *
4629 * Return 0 if transaction is pending 1 otherwise.
4630 */
4631int pci_wait_for_pending_transaction(struct pci_dev *dev)
4632{
4633        if (!pci_is_pcie(dev))
4634                return 1;
4635
4636        return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4637                                    PCI_EXP_DEVSTA_TRPND);
4638}
4639EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4640
4641/**
4642 * pcie_flr - initiate a PCIe function level reset
4643 * @dev: device to reset
4644 *
4645 * Initiate a function level reset unconditionally on @dev without
4646 * checking any flags and DEVCAP
4647 */
4648int pcie_flr(struct pci_dev *dev)
4649{
4650        if (!pci_wait_for_pending_transaction(dev))
4651                pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4652
4653        pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4654
4655        if (dev->imm_ready)
4656                return 0;
4657
4658        /*
4659         * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4660         * 100ms, but may silently discard requests while the FLR is in
4661         * progress.  Wait 100ms before trying to access the device.
4662         */
4663        msleep(100);
4664
4665        return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4666}
4667EXPORT_SYMBOL_GPL(pcie_flr);
4668
4669/**
4670 * pcie_reset_flr - initiate a PCIe function level reset
4671 * @dev: device to reset
4672 * @probe: if true, return 0 if device can be reset this way
4673 *
4674 * Initiate a function level reset on @dev.
4675 */
4676int pcie_reset_flr(struct pci_dev *dev, bool probe)
4677{
4678        if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4679                return -ENOTTY;
4680
4681        if (!(dev->devcap & PCI_EXP_DEVCAP_FLR))
4682                return -ENOTTY;
4683
4684        if (probe)
4685                return 0;
4686
4687        return pcie_flr(dev);
4688}
4689EXPORT_SYMBOL_GPL(pcie_reset_flr);
4690
4691static int pci_af_flr(struct pci_dev *dev, bool probe)
4692{
4693        int pos;
4694        u8 cap;
4695
4696        pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4697        if (!pos)
4698                return -ENOTTY;
4699
4700        if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4701                return -ENOTTY;
4702
4703        pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4704        if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4705                return -ENOTTY;
4706
4707        if (probe)
4708                return 0;
4709
4710        /*
4711         * Wait for Transaction Pending bit to clear.  A word-aligned test
4712         * is used, so we use the control offset rather than status and shift
4713         * the test bit to match.
4714         */
4715        if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4716                                 PCI_AF_STATUS_TP << 8))
4717                pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4718
4719        pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4720
4721        if (dev->imm_ready)
4722                return 0;
4723
4724        /*
4725         * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4726         * updated 27 July 2006; a device must complete an FLR within
4727         * 100ms, but may silently discard requests while the FLR is in
4728         * progress.  Wait 100ms before trying to access the device.
4729         */
4730        msleep(100);
4731
4732        return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4733}
4734
4735/**
4736 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4737 * @dev: Device to reset.
4738 * @probe: if true, return 0 if the device can be reset this way.
4739 *
4740 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4741 * unset, it will be reinitialized internally when going from PCI_D3hot to
4742 * PCI_D0.  If that's the case and the device is not in a low-power state
4743 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4744 *
4745 * NOTE: This causes the caller to sleep for twice the device power transition
4746 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4747 * by default (i.e. unless the @dev's d3hot_delay field has a different value).
4748 * Moreover, only devices in D0 can be reset by this function.
4749 */
4750static int pci_pm_reset(struct pci_dev *dev, bool probe)
4751{
4752        u16 csr;
4753
4754        if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4755                return -ENOTTY;
4756
4757        pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4758        if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4759                return -ENOTTY;
4760
4761        if (probe)
4762                return 0;
4763
4764        if (dev->current_state != PCI_D0)
4765                return -EINVAL;
4766
4767        csr &= ~PCI_PM_CTRL_STATE_MASK;
4768        csr |= PCI_D3hot;
4769        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4770        pci_dev_d3_sleep(dev);
4771
4772        csr &= ~PCI_PM_CTRL_STATE_MASK;
4773        csr |= PCI_D0;
4774        pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4775        pci_dev_d3_sleep(dev);
4776
4777        return pci_dev_wait(dev, "PM D3hot->D0", PCIE_RESET_READY_POLL_MS);
4778}
4779
4780/**
4781 * pcie_wait_for_link_delay - Wait until link is active or inactive
4782 * @pdev: Bridge device
4783 * @active: waiting for active or inactive?
4784 * @delay: Delay to wait after link has become active (in ms)
4785 *
4786 * Use this to wait till link becomes active or inactive.
4787 */
4788static bool pcie_wait_for_link_delay(struct pci_dev *pdev, bool active,
4789                                     int delay)
4790{
4791        int timeout = 1000;
4792        bool ret;
4793        u16 lnk_status;
4794
4795        /*
4796         * Some controllers might not implement link active reporting. In this
4797         * case, we wait for 1000 ms + any delay requested by the caller.
4798         */
4799        if (!pdev->link_active_reporting) {
4800                msleep(timeout + delay);
4801                return true;
4802        }
4803
4804        /*
4805         * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4806         * after which we should expect an link active if the reset was
4807         * successful. If so, software must wait a minimum 100ms before sending
4808         * configuration requests to devices downstream this port.
4809         *
4810         * If the link fails to activate, either the device was physically
4811         * removed or the link is permanently failed.
4812         */
4813        if (active)
4814                msleep(20);
4815        for (;;) {
4816                pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4817                ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4818                if (ret == active)
4819                        break;
4820                if (timeout <= 0)
4821                        break;
4822                msleep(10);
4823                timeout -= 10;
4824        }
4825        if (active && ret)
4826                msleep(delay);
4827
4828        return ret == active;
4829}
4830
4831/**
4832 * pcie_wait_for_link - Wait until link is active or inactive
4833 * @pdev: Bridge device
4834 * @active: waiting for active or inactive?
4835 *
4836 * Use this to wait till link becomes active or inactive.
4837 */
4838bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4839{
4840        return pcie_wait_for_link_delay(pdev, active, 100);
4841}
4842
4843/*
4844 * Find maximum D3cold delay required by all the devices on the bus.  The
4845 * spec says 100 ms, but firmware can lower it and we allow drivers to
4846 * increase it as well.
4847 *
4848 * Called with @pci_bus_sem locked for reading.
4849 */
4850static int pci_bus_max_d3cold_delay(const struct pci_bus *bus)
4851{
4852        const struct pci_dev *pdev;
4853        int min_delay = 100;
4854        int max_delay = 0;
4855
4856        list_for_each_entry(pdev, &bus->devices, bus_list) {
4857                if (pdev->d3cold_delay < min_delay)
4858                        min_delay = pdev->d3cold_delay;
4859                if (pdev->d3cold_delay > max_delay)
4860                        max_delay = pdev->d3cold_delay;
4861        }
4862
4863        return max(min_delay, max_delay);
4864}
4865
4866/**
4867 * pci_bridge_wait_for_secondary_bus - Wait for secondary bus to be accessible
4868 * @dev: PCI bridge
4869 *
4870 * Handle necessary delays before access to the devices on the secondary
4871 * side of the bridge are permitted after D3cold to D0 transition.
4872 *
4873 * For PCIe this means the delays in PCIe 5.0 section 6.6.1. For
4874 * conventional PCI it means Tpvrh + Trhfa specified in PCI 3.0 section
4875 * 4.3.2.
4876 */
4877void pci_bridge_wait_for_secondary_bus(struct pci_dev *dev)
4878{
4879        struct pci_dev *child;
4880        int delay;
4881
4882        if (pci_dev_is_disconnected(dev))
4883                return;
4884
4885        if (!pci_is_bridge(dev) || !dev->bridge_d3)
4886                return;
4887
4888        down_read(&pci_bus_sem);
4889
4890        /*
4891         * We only deal with devices that are present currently on the bus.
4892         * For any hot-added devices the access delay is handled in pciehp
4893         * board_added(). In case of ACPI hotplug the firmware is expected
4894         * to configure the devices before OS is notified.
4895         */
4896        if (!dev->subordinate || list_empty(&dev->subordinate->devices)) {
4897                up_read(&pci_bus_sem);
4898                return;
4899        }
4900
4901        /* Take d3cold_delay requirements into account */
4902        delay = pci_bus_max_d3cold_delay(dev->subordinate);
4903        if (!delay) {
4904                up_read(&pci_bus_sem);
4905                return;
4906        }
4907
4908        child = list_first_entry(&dev->subordinate->devices, struct pci_dev,
4909                                 bus_list);
4910        up_read(&pci_bus_sem);
4911
4912        /*
4913         * Conventional PCI and PCI-X we need to wait Tpvrh + Trhfa before
4914         * accessing the device after reset (that is 1000 ms + 100 ms). In
4915         * practice this should not be needed because we don't do power
4916         * management for them (see pci_bridge_d3_possible()).
4917         */
4918        if (!pci_is_pcie(dev)) {
4919                pci_dbg(dev, "waiting %d ms for secondary bus\n", 1000 + delay);
4920                msleep(1000 + delay);
4921                return;
4922        }
4923
4924        /*
4925         * For PCIe downstream and root ports that do not support speeds
4926         * greater than 5 GT/s need to wait minimum 100 ms. For higher
4927         * speeds (gen3) we need to wait first for the data link layer to
4928         * become active.
4929         *
4930         * However, 100 ms is the minimum and the PCIe spec says the
4931         * software must allow at least 1s before it can determine that the
4932         * device that did not respond is a broken device. There is
4933         * evidence that 100 ms is not always enough, for example certain
4934         * Titan Ridge xHCI controller does not always respond to
4935         * configuration requests if we only wait for 100 ms (see
4936         * https://bugzilla.kernel.org/show_bug.cgi?id=203885).
4937         *
4938         * Therefore we wait for 100 ms and check for the device presence.
4939         * If it is still not present give it an additional 100 ms.
4940         */
4941        if (!pcie_downstream_port(dev))
4942                return;
4943
4944        if (pcie_get_speed_cap(dev) <= PCIE_SPEED_5_0GT) {
4945                pci_dbg(dev, "waiting %d ms for downstream link\n", delay);
4946                msleep(delay);
4947        } else {
4948                pci_dbg(dev, "waiting %d ms for downstream link, after activation\n",
4949                        delay);
4950                if (!pcie_wait_for_link_delay(dev, true, delay)) {
4951                        /* Did not train, no need to wait any further */
4952                        pci_info(dev, "Data Link Layer Link Active not set in 1000 msec\n");
4953                        return;
4954                }
4955        }
4956
4957        if (!pci_device_is_present(child)) {
4958                pci_dbg(child, "waiting additional %d ms to become accessible\n", delay);
4959                msleep(delay);
4960        }
4961}
4962
4963void pci_reset_secondary_bus(struct pci_dev *dev)
4964{
4965        u16 ctrl;
4966
4967        pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4968        ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4969        pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4970
4971        /*
4972         * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms.  Double
4973         * this to 2ms to ensure that we meet the minimum requirement.
4974         */
4975        msleep(2);
4976
4977        ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4978        pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4979
4980        /*
4981         * Trhfa for conventional PCI is 2^25 clock cycles.
4982         * Assuming a minimum 33MHz clock this results in a 1s
4983         * delay before we can consider subordinate devices to
4984         * be re-initialized.  PCIe has some ways to shorten this,
4985         * but we don't make use of them yet.
4986         */
4987        ssleep(1);
4988}
4989
4990void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4991{
4992        pci_reset_secondary_bus(dev);
4993}
4994
4995/**
4996 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4997 * @dev: Bridge device
4998 *
4999 * Use the bridge control register to assert reset on the secondary bus.
5000 * Devices on the secondary bus are left in power-on state.
5001 */
5002int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
5003{
5004        pcibios_reset_secondary_bus(dev);
5005
5006        return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
5007}
5008EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
5009
5010static int pci_parent_bus_reset(struct pci_dev *dev, bool probe)
5011{
5012        struct pci_dev *pdev;
5013
5014        if (pci_is_root_bus(dev->bus) || dev->subordinate ||
5015            !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
5016                return -ENOTTY;
5017
5018        list_for_each_entry(pdev, &dev->bus->devices, bus_list)
5019                if (pdev != dev)
5020                        return -ENOTTY;
5021
5022        if (probe)
5023                return 0;
5024
5025        return pci_bridge_secondary_bus_reset(dev->bus->self);
5026}
5027
5028static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, bool probe)
5029{
5030        int rc = -ENOTTY;
5031
5032        if (!hotplug || !try_module_get(hotplug->owner))
5033                return rc;
5034
5035        if (hotplug->ops->reset_slot)
5036                rc = hotplug->ops->reset_slot(hotplug, probe);
5037
5038        module_put(hotplug->owner);
5039
5040        return rc;
5041}
5042
5043static int pci_dev_reset_slot_function(struct pci_dev *dev, bool probe)
5044{
5045        if (dev->multifunction || dev->subordinate || !dev->slot ||
5046            dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
5047                return -ENOTTY;
5048
5049        return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
5050}
5051
5052static int pci_reset_bus_function(struct pci_dev *dev, bool probe)
5053{
5054        int rc;
5055
5056        rc = pci_dev_reset_slot_function(dev, probe);
5057        if (rc != -ENOTTY)
5058                return rc;
5059        return pci_parent_bus_reset(dev, probe);
5060}
5061
5062static void pci_dev_lock(struct pci_dev *dev)
5063{
5064        pci_cfg_access_lock(dev);
5065        /* block PM suspend, driver probe, etc. */
5066        device_lock(&dev->dev);
5067}
5068
5069/* Return 1 on successful lock, 0 on contention */
5070int pci_dev_trylock(struct pci_dev *dev)
5071{
5072        if (pci_cfg_access_trylock(dev)) {
5073                if (device_trylock(&dev->dev))
5074                        return 1;
5075                pci_cfg_access_unlock(dev);
5076        }
5077
5078        return 0;
5079}
5080EXPORT_SYMBOL_GPL(pci_dev_trylock);
5081
5082void pci_dev_unlock(struct pci_dev *dev)
5083{
5084        device_unlock(&dev->dev);
5085        pci_cfg_access_unlock(dev);
5086}
5087EXPORT_SYMBOL_GPL(pci_dev_unlock);
5088
5089static void pci_dev_save_and_disable(struct pci_dev *dev)
5090{
5091        const struct pci_error_handlers *err_handler =
5092                        dev->driver ? dev->driver->err_handler : NULL;
5093
5094        /*
5095         * dev->driver->err_handler->reset_prepare() is protected against
5096         * races with ->remove() by the device lock, which must be held by
5097         * the caller.
5098         */
5099        if (err_handler && err_handler->reset_prepare)
5100                err_handler->reset_prepare(dev);
5101
5102        /*
5103         * Wake-up device prior to save.  PM registers default to D0 after
5104         * reset and a simple register restore doesn't reliably return
5105         * to a non-D0 state anyway.
5106         */
5107        pci_set_power_state(dev, PCI_D0);
5108
5109        pci_save_state(dev);
5110        /*
5111         * Disable the device by clearing the Command register, except for
5112         * INTx-disable which is set.  This not only disables MMIO and I/O port
5113         * BARs, but also prevents the device from being Bus Master, preventing
5114         * DMA from the device including MSI/MSI-X interrupts.  For PCI 2.3
5115         * compliant devices, INTx-disable prevents legacy interrupts.
5116         */
5117        pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
5118}
5119
5120static void pci_dev_restore(struct pci_dev *dev)
5121{
5122        const struct pci_error_handlers *err_handler =
5123                        dev->driver ? dev->driver->err_handler : NULL;
5124
5125        pci_restore_state(dev);
5126
5127        /*
5128         * dev->driver->err_handler->reset_done() is protected against
5129         * races with ->remove() by the device lock, which must be held by
5130         * the caller.
5131         */
5132        if (err_handler && err_handler->reset_done)
5133                err_handler->reset_done(dev);
5134}
5135
5136/* dev->reset_methods[] is a 0-terminated list of indices into this array */
5137static const struct pci_reset_fn_method pci_reset_fn_methods[] = {
5138        { },
5139        { pci_dev_specific_reset, .name = "device_specific" },
5140        { pci_dev_acpi_reset, .name = "acpi" },
5141        { pcie_reset_flr, .name = "flr" },
5142        { pci_af_flr, .name = "af_flr" },
5143        { pci_pm_reset, .name = "pm" },
5144        { pci_reset_bus_function, .name = "bus" },
5145};
5146
5147static ssize_t reset_method_show(struct device *dev,
5148                                 struct device_attribute *attr, char *buf)
5149{
5150        struct pci_dev *pdev = to_pci_dev(dev);
5151        ssize_t len = 0;
5152        int i, m;
5153
5154        for (i = 0; i < PCI_NUM_RESET_METHODS; i++) {
5155                m = pdev->reset_methods[i];
5156                if (!m)
5157                        break;
5158
5159                len += sysfs_emit_at(buf, len, "%s%s", len ? " " : "",
5160                                     pci_reset_fn_methods[m].name);
5161        }
5162
5163        if (len)
5164                len += sysfs_emit_at(buf, len, "\n");
5165
5166        return len;
5167}
5168
5169static int reset_method_lookup(const char *name)
5170{
5171        int m;
5172
5173        for (m = 1; m < PCI_NUM_RESET_METHODS; m++) {
5174                if (sysfs_streq(name, pci_reset_fn_methods[m].name))
5175                        return m;
5176        }
5177
5178        return 0;       /* not found */
5179}
5180
5181static ssize_t reset_method_store(struct device *dev,
5182                                  struct device_attribute *attr,
5183                                  const char *buf, size_t count)
5184{
5185        struct pci_dev *pdev = to_pci_dev(dev);
5186        char *options, *name;
5187        int m, n;
5188        u8 reset_methods[PCI_NUM_RESET_METHODS] = { 0 };
5189
5190        if (sysfs_streq(buf, "")) {
5191                pdev->reset_methods[0] = 0;
5192                pci_warn(pdev, "All device reset methods disabled by user");
5193                return count;
5194        }
5195
5196        if (sysfs_streq(buf, "default")) {
5197                pci_init_reset_methods(pdev);
5198                return count;
5199        }
5200
5201        options = kstrndup(buf, count, GFP_KERNEL);
5202        if (!options)
5203                return -ENOMEM;
5204
5205        n = 0;
5206        while ((name = strsep(&options, " ")) != NULL) {
5207                if (sysfs_streq(name, ""))
5208                        continue;
5209
5210                name = strim(name);
5211
5212                m = reset_method_lookup(name);
5213                if (!m) {
5214                        pci_err(pdev, "Invalid reset method '%s'", name);
5215                        goto error;
5216                }
5217
5218                if (pci_reset_fn_methods[m].reset_fn(pdev, PCI_RESET_PROBE)) {
5219                        pci_err(pdev, "Unsupported reset method '%s'", name);
5220                        goto error;
5221                }
5222
5223                if (n == PCI_NUM_RESET_METHODS - 1) {
5224                        pci_err(pdev, "Too many reset methods\n");
5225                        goto error;
5226                }
5227
5228                reset_methods[n++] = m;
5229        }
5230
5231        reset_methods[n] = 0;
5232
5233        /* Warn if dev-specific supported but not highest priority */
5234        if (pci_reset_fn_methods[1].reset_fn(pdev, PCI_RESET_PROBE) == 0 &&
5235            reset_methods[0] != 1)
5236                pci_warn(pdev, "Device-specific reset disabled/de-prioritized by user");
5237        memcpy(pdev->reset_methods, reset_methods, sizeof(pdev->reset_methods));
5238        kfree(options);
5239        return count;
5240
5241error:
5242        /* Leave previous methods unchanged */
5243        kfree(options);
5244        return -EINVAL;
5245}
5246static DEVICE_ATTR_RW(reset_method);
5247
5248static struct attribute *pci_dev_reset_method_attrs[] = {
5249        &dev_attr_reset_method.attr,
5250        NULL,
5251};
5252
5253static umode_t pci_dev_reset_method_attr_is_visible(struct kobject *kobj,
5254                                                    struct attribute *a, int n)
5255{
5256        struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj));
5257
5258        if (!pci_reset_supported(pdev))
5259                return 0;
5260
5261        return a->mode;
5262}
5263
5264const struct attribute_group pci_dev_reset_method_attr_group = {
5265        .attrs = pci_dev_reset_method_attrs,
5266        .is_visible = pci_dev_reset_method_attr_is_visible,
5267};
5268
5269/**
5270 * __pci_reset_function_locked - reset a PCI device function while holding
5271 * the @dev mutex lock.
5272 * @dev: PCI device to reset
5273 *
5274 * Some devices allow an individual function to be reset without affecting
5275 * other functions in the same device.  The PCI device must be responsive
5276 * to PCI config space in order to use this function.
5277 *
5278 * The device function is presumed to be unused and the caller is holding
5279 * the device mutex lock when this function is called.
5280 *
5281 * Resetting the device will make the contents of PCI configuration space
5282 * random, so any caller of this must be prepared to reinitialise the
5283 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
5284 * etc.
5285 *
5286 * Returns 0 if the device function was successfully reset or negative if the
5287 * device doesn't support resetting a single function.
5288 */
5289int __pci_reset_function_locked(struct pci_dev *dev)
5290{
5291        int i, m, rc = -ENOTTY;
5292
5293        might_sleep();
5294
5295        /*
5296         * A reset method returns -ENOTTY if it doesn't support this device and
5297         * we should try the next method.
5298         *
5299         * If it returns 0 (success), we're finished.  If it returns any other
5300         * error, we're also finished: this indicates that further reset
5301         * mechanisms might be broken on the device.
5302         */
5303        for (i = 0; i < PCI_NUM_RESET_METHODS; i++) {
5304                m = dev->reset_methods[i];
5305                if (!m)
5306                        return -ENOTTY;
5307
5308                rc = pci_reset_fn_methods[m].reset_fn(dev, PCI_RESET_DO_RESET);
5309                if (!rc)
5310                        return 0;
5311                if (rc != -ENOTTY)
5312                        return rc;
5313        }
5314
5315        return -ENOTTY;
5316}
5317EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
5318
5319/**
5320 * pci_init_reset_methods - check whether device can be safely reset
5321 * and store supported reset mechanisms.
5322 * @dev: PCI device to check for reset mechanisms
5323 *
5324 * Some devices allow an individual function to be reset without affecting
5325 * other functions in the same device.  The PCI device must be in D0-D3hot
5326 * state.
5327 *
5328 * Stores reset mechanisms supported by device in reset_methods byte array
5329 * which is a member of struct pci_dev.
5330 */
5331void pci_init_reset_methods(struct pci_dev *dev)
5332{
5333        int m, i, rc;
5334
5335        BUILD_BUG_ON(ARRAY_SIZE(pci_reset_fn_methods) != PCI_NUM_RESET_METHODS);
5336
5337        might_sleep();
5338
5339        i = 0;
5340        for (m = 1; m < PCI_NUM_RESET_METHODS; m++) {
5341                rc = pci_reset_fn_methods[m].reset_fn(dev, PCI_RESET_PROBE);
5342                if (!rc)
5343                        dev->reset_methods[i++] = m;
5344                else if (rc != -ENOTTY)
5345                        break;
5346        }
5347
5348        dev->reset_methods[i] = 0;
5349}
5350
5351/**
5352 * pci_reset_function - quiesce and reset a PCI device function
5353 * @dev: PCI device to reset
5354 *
5355 * Some devices allow an individual function to be reset without affecting
5356 * other functions in the same device.  The PCI device must be responsive
5357 * to PCI config space in order to use this function.
5358 *
5359 * This function does not just reset the PCI portion of a device, but
5360 * clears all the state associated with the device.  This function differs
5361 * from __pci_reset_function_locked() in that it saves and restores device state
5362 * over the reset and takes the PCI device lock.
5363 *
5364 * Returns 0 if the device function was successfully reset or negative if the
5365 * device doesn't support resetting a single function.
5366 */
5367int pci_reset_function(struct pci_dev *dev)
5368{
5369        int rc;
5370
5371        if (!pci_reset_supported(dev))
5372                return -ENOTTY;
5373
5374        pci_dev_lock(dev);
5375        pci_dev_save_and_disable(dev);
5376
5377        rc = __pci_reset_function_locked(dev);
5378
5379        pci_dev_restore(dev);
5380        pci_dev_unlock(dev);
5381
5382        return rc;
5383}
5384EXPORT_SYMBOL_GPL(pci_reset_function);
5385
5386/**
5387 * pci_reset_function_locked - quiesce and reset a PCI device function
5388 * @dev: PCI device to reset
5389 *
5390 * Some devices allow an individual function to be reset without affecting
5391 * other functions in the same device.  The PCI device must be responsive
5392 * to PCI config space in order to use this function.
5393 *
5394 * This function does not just reset the PCI portion of a device, but
5395 * clears all the state associated with the device.  This function differs
5396 * from __pci_reset_function_locked() in that it saves and restores device state
5397 * over the reset.  It also differs from pci_reset_function() in that it
5398 * requires the PCI device lock to be held.
5399 *
5400 * Returns 0 if the device function was successfully reset or negative if the
5401 * device doesn't support resetting a single function.
5402 */
5403int pci_reset_function_locked(struct pci_dev *dev)
5404{
5405        int rc;
5406
5407        if (!pci_reset_supported(dev))
5408                return -ENOTTY;
5409
5410        pci_dev_save_and_disable(dev);
5411
5412        rc = __pci_reset_function_locked(dev);
5413
5414        pci_dev_restore(dev);
5415
5416        return rc;
5417}
5418EXPORT_SYMBOL_GPL(pci_reset_function_locked);
5419
5420/**
5421 * pci_try_reset_function - quiesce and reset a PCI device function
5422 * @dev: PCI device to reset
5423 *
5424 * Same as above, except return -EAGAIN if unable to lock device.
5425 */
5426int pci_try_reset_function(struct pci_dev *dev)
5427{
5428        int rc;
5429
5430        if (!pci_reset_supported(dev))
5431                return -ENOTTY;
5432
5433        if (!pci_dev_trylock(dev))
5434                return -EAGAIN;
5435
5436        pci_dev_save_and_disable(dev);
5437        rc = __pci_reset_function_locked(dev);
5438        pci_dev_restore(dev);
5439        pci_dev_unlock(dev);
5440
5441        return rc;
5442}
5443EXPORT_SYMBOL_GPL(pci_try_reset_function);
5444
5445/* Do any devices on or below this bus prevent a bus reset? */
5446static bool pci_bus_resetable(struct pci_bus *bus)
5447{
5448        struct pci_dev *dev;
5449
5450
5451        if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5452                return false;
5453
5454        list_for_each_entry(dev, &bus->devices, bus_list) {
5455                if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5456                    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5457                        return false;
5458        }
5459
5460        return true;
5461}
5462
5463/* Lock devices from the top of the tree down */
5464static void pci_bus_lock(struct pci_bus *bus)
5465{
5466        struct pci_dev *dev;
5467
5468        list_for_each_entry(dev, &bus->devices, bus_list) {
5469                pci_dev_lock(dev);
5470                if (dev->subordinate)
5471                        pci_bus_lock(dev->subordinate);
5472        }
5473}
5474
5475/* Unlock devices from the bottom of the tree up */
5476static void pci_bus_unlock(struct pci_bus *bus)
5477{
5478        struct pci_dev *dev;
5479
5480        list_for_each_entry(dev, &bus->devices, bus_list) {
5481                if (dev->subordinate)
5482                        pci_bus_unlock(dev->subordinate);
5483                pci_dev_unlock(dev);
5484        }
5485}
5486
5487/* Return 1 on successful lock, 0 on contention */
5488static int pci_bus_trylock(struct pci_bus *bus)
5489{
5490        struct pci_dev *dev;
5491
5492        list_for_each_entry(dev, &bus->devices, bus_list) {
5493                if (!pci_dev_trylock(dev))
5494                        goto unlock;
5495                if (dev->subordinate) {
5496                        if (!pci_bus_trylock(dev->subordinate)) {
5497                                pci_dev_unlock(dev);
5498                                goto unlock;
5499                        }
5500                }
5501        }
5502        return 1;
5503
5504unlock:
5505        list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5506                if (dev->subordinate)
5507                        pci_bus_unlock(dev->subordinate);
5508                pci_dev_unlock(dev);
5509        }
5510        return 0;
5511}
5512
5513/* Do any devices on or below this slot prevent a bus reset? */
5514static bool pci_slot_resetable(struct pci_slot *slot)
5515{
5516        struct pci_dev *dev;
5517
5518        if (slot->bus->self &&
5519            (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5520                return false;
5521
5522        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5523                if (!dev->slot || dev->slot != slot)
5524                        continue;
5525                if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5526                    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5527                        return false;
5528        }
5529
5530        return true;
5531}
5532
5533/* Lock devices from the top of the tree down */
5534static void pci_slot_lock(struct pci_slot *slot)
5535{
5536        struct pci_dev *dev;
5537
5538        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5539                if (!dev->slot || dev->slot != slot)
5540                        continue;
5541                pci_dev_lock(dev);
5542                if (dev->subordinate)
5543                        pci_bus_lock(dev->subordinate);
5544        }
5545}
5546
5547/* Unlock devices from the bottom of the tree up */
5548static void pci_slot_unlock(struct pci_slot *slot)
5549{
5550        struct pci_dev *dev;
5551
5552        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5553                if (!dev->slot || dev->slot != slot)
5554                        continue;
5555                if (dev->subordinate)
5556                        pci_bus_unlock(dev->subordinate);
5557                pci_dev_unlock(dev);
5558        }
5559}
5560
5561/* Return 1 on successful lock, 0 on contention */
5562static int pci_slot_trylock(struct pci_slot *slot)
5563{
5564        struct pci_dev *dev;
5565
5566        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5567                if (!dev->slot || dev->slot != slot)
5568                        continue;
5569                if (!pci_dev_trylock(dev))
5570                        goto unlock;
5571                if (dev->subordinate) {
5572                        if (!pci_bus_trylock(dev->subordinate)) {
5573                                pci_dev_unlock(dev);
5574                                goto unlock;
5575                        }
5576                }
5577        }
5578        return 1;
5579
5580unlock:
5581        list_for_each_entry_continue_reverse(dev,
5582                                             &slot->bus->devices, bus_list) {
5583                if (!dev->slot || dev->slot != slot)
5584                        continue;
5585                if (dev->subordinate)
5586                        pci_bus_unlock(dev->subordinate);
5587                pci_dev_unlock(dev);
5588        }
5589        return 0;
5590}
5591
5592/*
5593 * Save and disable devices from the top of the tree down while holding
5594 * the @dev mutex lock for the entire tree.
5595 */
5596static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5597{
5598        struct pci_dev *dev;
5599
5600        list_for_each_entry(dev, &bus->devices, bus_list) {
5601                pci_dev_save_and_disable(dev);
5602                if (dev->subordinate)
5603                        pci_bus_save_and_disable_locked(dev->subordinate);
5604        }
5605}
5606
5607/*
5608 * Restore devices from top of the tree down while holding @dev mutex lock
5609 * for the entire tree.  Parent bridges need to be restored before we can
5610 * get to subordinate devices.
5611 */
5612static void pci_bus_restore_locked(struct pci_bus *bus)
5613{
5614        struct pci_dev *dev;
5615
5616        list_for_each_entry(dev, &bus->devices, bus_list) {
5617                pci_dev_restore(dev);
5618                if (dev->subordinate)
5619                        pci_bus_restore_locked(dev->subordinate);
5620        }
5621}
5622
5623/*
5624 * Save and disable devices from the top of the tree down while holding
5625 * the @dev mutex lock for the entire tree.
5626 */
5627static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5628{
5629        struct pci_dev *dev;
5630
5631        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5632                if (!dev->slot || dev->slot != slot)
5633                        continue;
5634                pci_dev_save_and_disable(dev);
5635                if (dev->subordinate)
5636                        pci_bus_save_and_disable_locked(dev->subordinate);
5637        }
5638}
5639
5640/*
5641 * Restore devices from top of the tree down while holding @dev mutex lock
5642 * for the entire tree.  Parent bridges need to be restored before we can
5643 * get to subordinate devices.
5644 */
5645static void pci_slot_restore_locked(struct pci_slot *slot)
5646{
5647        struct pci_dev *dev;
5648
5649        list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5650                if (!dev->slot || dev->slot != slot)
5651                        continue;
5652                pci_dev_restore(dev);
5653                if (dev->subordinate)
5654                        pci_bus_restore_locked(dev->subordinate);
5655        }
5656}
5657
5658static int pci_slot_reset(struct pci_slot *slot, bool probe)
5659{
5660        int rc;
5661
5662        if (!slot || !pci_slot_resetable(slot))
5663                return -ENOTTY;
5664
5665        if (!probe)
5666                pci_slot_lock(slot);
5667
5668        might_sleep();
5669
5670        rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5671
5672        if (!probe)
5673                pci_slot_unlock(slot);
5674
5675        return rc;
5676}
5677
5678/**
5679 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5680 * @slot: PCI slot to probe
5681 *
5682 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5683 */
5684int pci_probe_reset_slot(struct pci_slot *slot)
5685{
5686        return pci_slot_reset(slot, PCI_RESET_PROBE);
5687}
5688EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5689
5690/**
5691 * __pci_reset_slot - Try to reset a PCI slot
5692 * @slot: PCI slot to reset
5693 *
5694 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5695 * independent of other slots.  For instance, some slots may support slot power
5696 * control.  In the case of a 1:1 bus to slot architecture, this function may
5697 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5698 * Generally a slot reset should be attempted before a bus reset.  All of the
5699 * function of the slot and any subordinate buses behind the slot are reset
5700 * through this function.  PCI config space of all devices in the slot and
5701 * behind the slot is saved before and restored after reset.
5702 *
5703 * Same as above except return -EAGAIN if the slot cannot be locked
5704 */
5705static int __pci_reset_slot(struct pci_slot *slot)
5706{
5707        int rc;
5708
5709        rc = pci_slot_reset(slot, PCI_RESET_PROBE);
5710        if (rc)
5711                return rc;
5712
5713        if (pci_slot_trylock(slot)) {
5714                pci_slot_save_and_disable_locked(slot);
5715                might_sleep();
5716                rc = pci_reset_hotplug_slot(slot->hotplug, PCI_RESET_DO_RESET);
5717                pci_slot_restore_locked(slot);
5718                pci_slot_unlock(slot);
5719        } else
5720                rc = -EAGAIN;
5721
5722        return rc;
5723}
5724
5725static int pci_bus_reset(struct pci_bus *bus, bool probe)
5726{
5727        int ret;
5728
5729        if (!bus->self || !pci_bus_resetable(bus))
5730                return -ENOTTY;
5731
5732        if (probe)
5733                return 0;
5734
5735        pci_bus_lock(bus);
5736
5737        might_sleep();
5738
5739        ret = pci_bridge_secondary_bus_reset(bus->self);
5740
5741        pci_bus_unlock(bus);
5742
5743        return ret;
5744}
5745
5746/**
5747 * pci_bus_error_reset - reset the bridge's subordinate bus
5748 * @bridge: The parent device that connects to the bus to reset
5749 *
5750 * This function will first try to reset the slots on this bus if the method is
5751 * available. If slot reset fails or is not available, this will fall back to a
5752 * secondary bus reset.
5753 */
5754int pci_bus_error_reset(struct pci_dev *bridge)
5755{
5756        struct pci_bus *bus = bridge->subordinate;
5757        struct pci_slot *slot;
5758
5759        if (!bus)
5760                return -ENOTTY;
5761
5762        mutex_lock(&pci_slot_mutex);
5763        if (list_empty(&bus->slots))
5764                goto bus_reset;
5765
5766        list_for_each_entry(slot, &bus->slots, list)
5767                if (pci_probe_reset_slot(slot))
5768                        goto bus_reset;
5769
5770        list_for_each_entry(slot, &bus->slots, list)
5771                if (pci_slot_reset(slot, PCI_RESET_DO_RESET))
5772                        goto bus_reset;
5773
5774        mutex_unlock(&pci_slot_mutex);
5775        return 0;
5776bus_reset:
5777        mutex_unlock(&pci_slot_mutex);
5778        return pci_bus_reset(bridge->subordinate, PCI_RESET_DO_RESET);
5779}
5780
5781/**
5782 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5783 * @bus: PCI bus to probe
5784 *
5785 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5786 */
5787int pci_probe_reset_bus(struct pci_bus *bus)
5788{
5789        return pci_bus_reset(bus, PCI_RESET_PROBE);
5790}
5791EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5792
5793/**
5794 * __pci_reset_bus - Try to reset a PCI bus
5795 * @bus: top level PCI bus to reset
5796 *
5797 * Same as above except return -EAGAIN if the bus cannot be locked
5798 */
5799static int __pci_reset_bus(struct pci_bus *bus)
5800{
5801        int rc;
5802
5803        rc = pci_bus_reset(bus, PCI_RESET_PROBE);
5804        if (rc)
5805                return rc;
5806
5807        if (pci_bus_trylock(bus)) {
5808                pci_bus_save_and_disable_locked(bus);
5809                might_sleep();
5810                rc = pci_bridge_secondary_bus_reset(bus->self);
5811                pci_bus_restore_locked(bus);
5812                pci_bus_unlock(bus);
5813        } else
5814                rc = -EAGAIN;
5815
5816        return rc;
5817}
5818
5819/**
5820 * pci_reset_bus - Try to reset a PCI bus
5821 * @pdev: top level PCI device to reset via slot/bus
5822 *
5823 * Same as above except return -EAGAIN if the bus cannot be locked
5824 */
5825int pci_reset_bus(struct pci_dev *pdev)
5826{
5827        return (!pci_probe_reset_slot(pdev->slot)) ?
5828            __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5829}
5830EXPORT_SYMBOL_GPL(pci_reset_bus);
5831
5832/**
5833 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5834 * @dev: PCI device to query
5835 *
5836 * Returns mmrbc: maximum designed memory read count in bytes or
5837 * appropriate error value.
5838 */
5839int pcix_get_max_mmrbc(struct pci_dev *dev)
5840{
5841        int cap;
5842        u32 stat;
5843
5844        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5845        if (!cap)
5846                return -EINVAL;
5847
5848        if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5849                return -EINVAL;
5850
5851        return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5852}
5853EXPORT_SYMBOL(pcix_get_max_mmrbc);
5854
5855/**
5856 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5857 * @dev: PCI device to query
5858 *
5859 * Returns mmrbc: maximum memory read count in bytes or appropriate error
5860 * value.
5861 */
5862int pcix_get_mmrbc(struct pci_dev *dev)
5863{
5864        int cap;
5865        u16 cmd;
5866
5867        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5868        if (!cap)
5869                return -EINVAL;
5870
5871        if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5872                return -EINVAL;
5873
5874        return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5875}
5876EXPORT_SYMBOL(pcix_get_mmrbc);
5877
5878/**
5879 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5880 * @dev: PCI device to query
5881 * @mmrbc: maximum memory read count in bytes
5882 *    valid values are 512, 1024, 2048, 4096
5883 *
5884 * If possible sets maximum memory read byte count, some bridges have errata
5885 * that prevent this.
5886 */
5887int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5888{
5889        int cap;
5890        u32 stat, v, o;
5891        u16 cmd;
5892
5893        if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5894                return -EINVAL;
5895
5896        v = ffs(mmrbc) - 10;
5897
5898        cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5899        if (!cap)
5900                return -EINVAL;
5901
5902        if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5903                return -EINVAL;
5904
5905        if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5906                return -E2BIG;
5907
5908        if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5909                return -EINVAL;
5910
5911        o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5912        if (o != v) {
5913                if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5914                        return -EIO;
5915
5916                cmd &= ~PCI_X_CMD_MAX_READ;
5917                cmd |= v << 2;
5918                if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5919                        return -EIO;
5920        }
5921        return 0;
5922}
5923EXPORT_SYMBOL(pcix_set_mmrbc);
5924
5925/**
5926 * pcie_get_readrq - get PCI Express read request size
5927 * @dev: PCI device to query
5928 *
5929 * Returns maximum memory read request in bytes or appropriate error value.
5930 */
5931int pcie_get_readrq(struct pci_dev *dev)
5932{
5933        u16 ctl;
5934
5935        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5936
5937        return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5938}
5939EXPORT_SYMBOL(pcie_get_readrq);
5940
5941/**
5942 * pcie_set_readrq - set PCI Express maximum memory read request
5943 * @dev: PCI device to query
5944 * @rq: maximum memory read count in bytes
5945 *    valid values are 128, 256, 512, 1024, 2048, 4096
5946 *
5947 * If possible sets maximum memory read request in bytes
5948 */
5949int pcie_set_readrq(struct pci_dev *dev, int rq)
5950{
5951        u16 v;
5952        int ret;
5953
5954        if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5955                return -EINVAL;
5956
5957        /*
5958         * If using the "performance" PCIe config, we clamp the read rq
5959         * size to the max packet size to keep the host bridge from
5960         * generating requests larger than we can cope with.
5961         */
5962        if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5963                int mps = pcie_get_mps(dev);
5964
5965                if (mps < rq)
5966                        rq = mps;
5967        }
5968
5969        v = (ffs(rq) - 8) << 12;
5970
5971        ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5972                                                  PCI_EXP_DEVCTL_READRQ, v);
5973
5974        return pcibios_err_to_errno(ret);
5975}
5976EXPORT_SYMBOL(pcie_set_readrq);
5977
5978/**
5979 * pcie_get_mps - get PCI Express maximum payload size
5980 * @dev: PCI device to query
5981 *
5982 * Returns maximum payload size in bytes
5983 */
5984int pcie_get_mps(struct pci_dev *dev)
5985{
5986        u16 ctl;
5987
5988        pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5989
5990        return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5991}
5992EXPORT_SYMBOL(pcie_get_mps);
5993
5994/**
5995 * pcie_set_mps - set PCI Express maximum payload size
5996 * @dev: PCI device to query
5997 * @mps: maximum payload size in bytes
5998 *    valid values are 128, 256, 512, 1024, 2048, 4096
5999 *
6000 * If possible sets maximum payload size
6001 */
6002int pcie_set_mps(struct pci_dev *dev, int mps)
6003{
6004        u16 v;
6005        int ret;
6006
6007        if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
6008                return -EINVAL;
6009
6010        v = ffs(mps) - 8;
6011        if (v > dev->pcie_mpss)
6012                return -EINVAL;
6013        v <<= 5;
6014
6015        ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
6016                                                  PCI_EXP_DEVCTL_PAYLOAD, v);
6017
6018        return pcibios_err_to_errno(ret);
6019}
6020EXPORT_SYMBOL(pcie_set_mps);
6021
6022/**
6023 * pcie_bandwidth_available - determine minimum link settings of a PCIe
6024 *                            device and its bandwidth limitation
6025 * @dev: PCI device to query
6026 * @limiting_dev: storage for device causing the bandwidth limitation
6027 * @speed: storage for speed of limiting device
6028 * @width: storage for width of limiting device
6029 *
6030 * Walk up the PCI device chain and find the point where the minimum
6031 * bandwidth is available.  Return the bandwidth available there and (if
6032 * limiting_dev, speed, and width pointers are supplied) information about
6033 * that point.  The bandwidth returned is in Mb/s, i.e., megabits/second of
6034 * raw bandwidth.
6035 */
6036u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
6037                             enum pci_bus_speed *speed,
6038                             enum pcie_link_width *width)
6039{
6040        u16 lnksta;
6041        enum pci_bus_speed next_speed;
6042        enum pcie_link_width next_width;
6043        u32 bw, next_bw;
6044
6045        if (speed)
6046                *speed = PCI_SPEED_UNKNOWN;
6047        if (width)
6048                *width = PCIE_LNK_WIDTH_UNKNOWN;
6049
6050        bw = 0;
6051
6052        while (dev) {
6053                pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
6054
6055                next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
6056                next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
6057                        PCI_EXP_LNKSTA_NLW_SHIFT;
6058
6059                next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
6060
6061                /* Check if current device limits the total bandwidth */
6062                if (!bw || next_bw <= bw) {
6063                        bw = next_bw;
6064
6065                        if (limiting_dev)
6066                                *limiting_dev = dev;
6067                        if (speed)
6068                                *speed = next_speed;
6069                        if (width)
6070                                *width = next_width;
6071                }
6072
6073                dev = pci_upstream_bridge(dev);
6074        }
6075
6076        return bw;
6077}
6078EXPORT_SYMBOL(pcie_bandwidth_available);
6079
6080/**
6081 * pcie_get_speed_cap - query for the PCI device's link speed capability
6082 * @dev: PCI device to query
6083 *
6084 * Query the PCI device speed capability.  Return the maximum link speed
6085 * supported by the device.
6086 */
6087enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
6088{
6089        u32 lnkcap2, lnkcap;
6090
6091        /*
6092         * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18.  The
6093         * implementation note there recommends using the Supported Link
6094         * Speeds Vector in Link Capabilities 2 when supported.
6095         *
6096         * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
6097         * should use the Supported Link Speeds field in Link Capabilities,
6098         * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
6099         */
6100        pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
6101
6102        /* PCIe r3.0-compliant */
6103        if (lnkcap2)
6104                return PCIE_LNKCAP2_SLS2SPEED(lnkcap2);
6105
6106        pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
6107        if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
6108                return PCIE_SPEED_5_0GT;
6109        else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
6110                return PCIE_SPEED_2_5GT;
6111
6112        return PCI_SPEED_UNKNOWN;
6113}
6114EXPORT_SYMBOL(pcie_get_speed_cap);
6115
6116/**
6117 * pcie_get_width_cap - query for the PCI device's link width capability
6118 * @dev: PCI device to query
6119 *
6120 * Query the PCI device width capability.  Return the maximum link width
6121 * supported by the device.
6122 */
6123enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
6124{
6125        u32 lnkcap;
6126
6127        pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
6128        if (lnkcap)
6129                return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
6130
6131        return PCIE_LNK_WIDTH_UNKNOWN;
6132}
6133EXPORT_SYMBOL(pcie_get_width_cap);
6134
6135/**
6136 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
6137 * @dev: PCI device
6138 * @speed: storage for link speed
6139 * @width: storage for link width
6140 *
6141 * Calculate a PCI device's link bandwidth by querying for its link speed
6142 * and width, multiplying them, and applying encoding overhead.  The result
6143 * is in Mb/s, i.e., megabits/second of raw bandwidth.
6144 */
6145u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
6146                           enum pcie_link_width *width)
6147{
6148        *speed = pcie_get_speed_cap(dev);
6149        *width = pcie_get_width_cap(dev);
6150
6151        if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
6152                return 0;
6153
6154        return *width * PCIE_SPEED2MBS_ENC(*speed);
6155}
6156
6157/**
6158 * __pcie_print_link_status - Report the PCI device's link speed and width
6159 * @dev: PCI device to query
6160 * @verbose: Print info even when enough bandwidth is available
6161 *
6162 * If the available bandwidth at the device is less than the device is
6163 * capable of, report the device's maximum possible bandwidth and the
6164 * upstream link that limits its performance.  If @verbose, always print
6165 * the available bandwidth, even if the device isn't constrained.
6166 */
6167void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
6168{
6169        enum pcie_link_width width, width_cap;
6170        enum pci_bus_speed speed, speed_cap;
6171        struct pci_dev *limiting_dev = NULL;
6172        u32 bw_avail, bw_cap;
6173
6174        bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
6175        bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
6176
6177        if (bw_avail >= bw_cap && verbose)
6178                pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
6179                         bw_cap / 1000, bw_cap % 1000,
6180                         pci_speed_string(speed_cap), width_cap);
6181        else if (bw_avail < bw_cap)
6182                pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth, limited by %s x%d link at %s (capable of %u.%03u Gb/s with %s x%d link)\n",
6183                         bw_avail / 1000, bw_avail % 1000,
6184                         pci_speed_string(speed), width,
6185                         limiting_dev ? pci_name(limiting_dev) : "<unknown>",
6186                         bw_cap / 1000, bw_cap % 1000,
6187                         pci_speed_string(speed_cap), width_cap);
6188}
6189
6190/**
6191 * pcie_print_link_status - Report the PCI device's link speed and width
6192 * @dev: PCI device to query
6193 *
6194 * Report the available bandwidth at the device.
6195 */
6196void pcie_print_link_status(struct pci_dev *dev)
6197{
6198        __pcie_print_link_status(dev, true);
6199}
6200EXPORT_SYMBOL(pcie_print_link_status);
6201
6202/**
6203 * pci_select_bars - Make BAR mask from the type of resource
6204 * @dev: the PCI device for which BAR mask is made
6205 * @flags: resource type mask to be selected
6206 *
6207 * This helper routine makes bar mask from the type of resource.
6208 */
6209int pci_select_bars(struct pci_dev *dev, unsigned long flags)
6210{
6211        int i, bars = 0;
6212        for (i = 0; i < PCI_NUM_RESOURCES; i++)
6213                if (pci_resource_flags(dev, i) & flags)
6214                        bars |= (1 << i);
6215        return bars;
6216}
6217EXPORT_SYMBOL(pci_select_bars);
6218
6219/* Some architectures require additional programming to enable VGA */
6220static arch_set_vga_state_t arch_set_vga_state;
6221
6222void __init pci_register_set_vga_state(arch_set_vga_state_t func)
6223{
6224        arch_set_vga_state = func;      /* NULL disables */
6225}
6226
6227static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
6228                                  unsigned int command_bits, u32 flags)
6229{
6230        if (arch_set_vga_state)
6231                return arch_set_vga_state(dev, decode, command_bits,
6232                                                flags);
6233        return 0;
6234}
6235
6236/**
6237 * pci_set_vga_state - set VGA decode state on device and parents if requested
6238 * @dev: the PCI device
6239 * @decode: true = enable decoding, false = disable decoding
6240 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
6241 * @flags: traverse ancestors and change bridges
6242 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
6243 */
6244int pci_set_vga_state(struct pci_dev *dev, bool decode,
6245                      unsigned int command_bits, u32 flags)
6246{
6247        struct pci_bus *bus;
6248        struct pci_dev *bridge;
6249        u16 cmd;
6250        int rc;
6251
6252        WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
6253
6254        /* ARCH specific VGA enables */
6255        rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
6256        if (rc)
6257                return rc;
6258
6259        if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
6260                pci_read_config_word(dev, PCI_COMMAND, &cmd);
6261                if (decode)
6262                        cmd |= command_bits;
6263                else
6264                        cmd &= ~command_bits;
6265                pci_write_config_word(dev, PCI_COMMAND, cmd);
6266        }
6267
6268        if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
6269                return 0;
6270
6271        bus = dev->bus;
6272        while (bus) {
6273                bridge = bus->self;
6274                if (bridge) {
6275                        pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
6276                                             &cmd);
6277                        if (decode)
6278                                cmd |= PCI_BRIDGE_CTL_VGA;
6279                        else
6280                                cmd &= ~PCI_BRIDGE_CTL_VGA;
6281                        pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
6282                                              cmd);
6283                }
6284                bus = bus->parent;
6285        }
6286        return 0;
6287}
6288
6289#ifdef CONFIG_ACPI
6290bool pci_pr3_present(struct pci_dev *pdev)
6291{
6292        struct acpi_device *adev;
6293
6294        if (acpi_disabled)
6295                return false;
6296
6297        adev = ACPI_COMPANION(&pdev->dev);
6298        if (!adev)
6299                return false;
6300
6301        return adev->power.flags.power_resources &&
6302                acpi_has_method(adev->handle, "_PR3");
6303}
6304EXPORT_SYMBOL_GPL(pci_pr3_present);
6305#endif
6306
6307/**
6308 * pci_add_dma_alias - Add a DMA devfn alias for a device
6309 * @dev: the PCI device for which alias is added
6310 * @devfn_from: alias slot and function
6311 * @nr_devfns: number of subsequent devfns to alias
6312 *
6313 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
6314 * which is used to program permissible bus-devfn source addresses for DMA
6315 * requests in an IOMMU.  These aliases factor into IOMMU group creation
6316 * and are useful for devices generating DMA requests beyond or different
6317 * from their logical bus-devfn.  Examples include device quirks where the
6318 * device simply uses the wrong devfn, as well as non-transparent bridges
6319 * where the alias may be a proxy for devices in another domain.
6320 *
6321 * IOMMU group creation is performed during device discovery or addition,
6322 * prior to any potential DMA mapping and therefore prior to driver probing
6323 * (especially for userspace assigned devices where IOMMU group definition
6324 * cannot be left as a userspace activity).  DMA aliases should therefore
6325 * be configured via quirks, such as the PCI fixup header quirk.
6326 */
6327void pci_add_dma_alias(struct pci_dev *dev, u8 devfn_from, unsigned nr_devfns)
6328{
6329        int devfn_to;
6330
6331        nr_devfns = min(nr_devfns, (unsigned) MAX_NR_DEVFNS - devfn_from);
6332        devfn_to = devfn_from + nr_devfns - 1;
6333
6334        if (!dev->dma_alias_mask)
6335                dev->dma_alias_mask = bitmap_zalloc(MAX_NR_DEVFNS, GFP_KERNEL);
6336        if (!dev->dma_alias_mask) {
6337                pci_warn(dev, "Unable to allocate DMA alias mask\n");
6338                return;
6339        }
6340
6341        bitmap_set(dev->dma_alias_mask, devfn_from, nr_devfns);
6342
6343        if (nr_devfns == 1)
6344                pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
6345                                PCI_SLOT(devfn_from), PCI_FUNC(devfn_from));
6346        else if (nr_devfns > 1)
6347                pci_info(dev, "Enabling fixed DMA alias for devfn range from %02x.%d to %02x.%d\n",
6348                                PCI_SLOT(devfn_from), PCI_FUNC(devfn_from),
6349                                PCI_SLOT(devfn_to), PCI_FUNC(devfn_to));
6350}
6351
6352bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
6353{
6354        return (dev1->dma_alias_mask &&
6355                test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
6356               (dev2->dma_alias_mask &&
6357                test_bit(dev1->devfn, dev2->dma_alias_mask)) ||
6358               pci_real_dma_dev(dev1) == dev2 ||
6359               pci_real_dma_dev(dev2) == dev1;
6360}
6361
6362bool pci_device_is_present(struct pci_dev *pdev)
6363{
6364        u32 v;
6365
6366        if (pci_dev_is_disconnected(pdev))
6367                return false;
6368        return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
6369}
6370EXPORT_SYMBOL_GPL(pci_device_is_present);
6371
6372void pci_ignore_hotplug(struct pci_dev *dev)
6373{
6374        struct pci_dev *bridge = dev->bus->self;
6375
6376        dev->ignore_hotplug = 1;
6377        /* Propagate the "ignore hotplug" setting to the parent bridge. */
6378        if (bridge)
6379                bridge->ignore_hotplug = 1;
6380}
6381EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
6382
6383/**
6384 * pci_real_dma_dev - Get PCI DMA device for PCI device
6385 * @dev: the PCI device that may have a PCI DMA alias
6386 *
6387 * Permits the platform to provide architecture-specific functionality to
6388 * devices needing to alias DMA to another PCI device on another PCI bus. If
6389 * the PCI device is on the same bus, it is recommended to use
6390 * pci_add_dma_alias(). This is the default implementation. Architecture
6391 * implementations can override this.
6392 */
6393struct pci_dev __weak *pci_real_dma_dev(struct pci_dev *dev)
6394{
6395        return dev;
6396}
6397
6398resource_size_t __weak pcibios_default_alignment(void)
6399{
6400        return 0;
6401}
6402
6403/*
6404 * Arches that don't want to expose struct resource to userland as-is in
6405 * sysfs and /proc can implement their own pci_resource_to_user().
6406 */
6407void __weak pci_resource_to_user(const struct pci_dev *dev, int bar,
6408                                 const struct resource *rsrc,
6409                                 resource_size_t *start, resource_size_t *end)
6410{
6411        *start = rsrc->start;
6412        *end = rsrc->end;
6413}
6414
6415static char *resource_alignment_param;
6416static DEFINE_SPINLOCK(resource_alignment_lock);
6417
6418/**
6419 * pci_specified_resource_alignment - get resource alignment specified by user.
6420 * @dev: the PCI device to get
6421 * @resize: whether or not to change resources' size when reassigning alignment
6422 *
6423 * RETURNS: Resource alignment if it is specified.
6424 *          Zero if it is not specified.
6425 */
6426static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
6427                                                        bool *resize)
6428{
6429        int align_order, count;
6430        resource_size_t align = pcibios_default_alignment();
6431        const char *p;
6432        int ret;
6433
6434        spin_lock(&resource_alignment_lock);
6435        p = resource_alignment_param;
6436        if (!p || !*p)
6437                goto out;
6438        if (pci_has_flag(PCI_PROBE_ONLY)) {
6439                align = 0;
6440                pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
6441                goto out;
6442        }
6443
6444        while (*p) {
6445                count = 0;
6446                if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
6447                    p[count] == '@') {
6448                        p += count + 1;
6449                        if (align_order > 63) {
6450                                pr_err("PCI: Invalid requested alignment (order %d)\n",
6451                                       align_order);
6452                                align_order = PAGE_SHIFT;
6453                        }
6454                } else {
6455                        align_order = PAGE_SHIFT;
6456                }
6457
6458                ret = pci_dev_str_match(dev, p, &p);
6459                if (ret == 1) {
6460                        *resize = true;
6461                        align = 1ULL << align_order;
6462                        break;
6463                } else if (ret < 0) {
6464                        pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
6465                               p);
6466                        break;
6467                }
6468
6469                if (*p != ';' && *p != ',') {
6470                        /* End of param or invalid format */
6471                        break;
6472                }
6473                p++;
6474        }
6475out:
6476        spin_unlock(&resource_alignment_lock);
6477        return align;
6478}
6479
6480static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
6481                                           resource_size_t align, bool resize)
6482{
6483        struct resource *r = &dev->resource[bar];
6484        resource_size_t size;
6485
6486        if (!(r->flags & IORESOURCE_MEM))
6487                return;
6488
6489        if (r->flags & IORESOURCE_PCI_FIXED) {
6490                pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
6491                         bar, r, (unsigned long long)align);
6492                return;
6493        }
6494
6495        size = resource_size(r);
6496        if (size >= align)
6497                return;
6498
6499        /*
6500         * Increase the alignment of the resource.  There are two ways we
6501         * can do this:
6502         *
6503         * 1) Increase the size of the resource.  BARs are aligned on their
6504         *    size, so when we reallocate space for this resource, we'll
6505         *    allocate it with the larger alignment.  This also prevents
6506         *    assignment of any other BARs inside the alignment region, so
6507         *    if we're requesting page alignment, this means no other BARs
6508         *    will share the page.
6509         *
6510         *    The disadvantage is that this makes the resource larger than
6511         *    the hardware BAR, which may break drivers that compute things
6512         *    based on the resource size, e.g., to find registers at a
6513         *    fixed offset before the end of the BAR.
6514         *
6515         * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
6516         *    set r->start to the desired alignment.  By itself this
6517         *    doesn't prevent other BARs being put inside the alignment
6518         *    region, but if we realign *every* resource of every device in
6519         *    the system, none of them will share an alignment region.
6520         *
6521         * When the user has requested alignment for only some devices via
6522         * the "pci=resource_alignment" argument, "resize" is true and we
6523         * use the first method.  Otherwise we assume we're aligning all
6524         * devices and we use the second.
6525         */
6526
6527        pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
6528                 bar, r, (unsigned long long)align);
6529
6530        if (resize) {
6531                r->start = 0;
6532                r->end = align - 1;
6533        } else {
6534                r->flags &= ~IORESOURCE_SIZEALIGN;
6535                r->flags |= IORESOURCE_STARTALIGN;
6536                r->start = align;
6537                r->end = r->start + size - 1;
6538        }
6539        r->flags |= IORESOURCE_UNSET;
6540}
6541
6542/*
6543 * This function disables memory decoding and releases memory resources
6544 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6545 * It also rounds up size to specified alignment.
6546 * Later on, the kernel will assign page-aligned memory resource back
6547 * to the device.
6548 */
6549void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6550{
6551        int i;
6552        struct resource *r;
6553        resource_size_t align;
6554        u16 command;
6555        bool resize = false;
6556
6557        /*
6558         * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6559         * 3.4.1.11.  Their resources are allocated from the space
6560         * described by the VF BARx register in the PF's SR-IOV capability.
6561         * We can't influence their alignment here.
6562         */
6563        if (dev->is_virtfn)
6564                return;
6565
6566        /* check if specified PCI is target device to reassign */
6567        align = pci_specified_resource_alignment(dev, &resize);
6568        if (!align)
6569                return;
6570
6571        if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6572            (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6573                pci_warn(dev, "Can't reassign resources to host bridge\n");
6574                return;
6575        }
6576
6577        pci_read_config_word(dev, PCI_COMMAND, &command);
6578        command &= ~PCI_COMMAND_MEMORY;
6579        pci_write_config_word(dev, PCI_COMMAND, command);
6580
6581        for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6582                pci_request_resource_alignment(dev, i, align, resize);
6583
6584        /*
6585         * Need to disable bridge's resource window,
6586         * to enable the kernel to reassign new resource
6587         * window later on.
6588         */
6589        if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6590                for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6591                        r = &dev->resource[i];
6592                        if (!(r->flags & IORESOURCE_MEM))
6593                                continue;
6594                        r->flags |= IORESOURCE_UNSET;
6595                        r->end = resource_size(r) - 1;
6596                        r->start = 0;
6597                }
6598                pci_disable_bridge_window(dev);
6599        }
6600}
6601
6602static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
6603{
6604        size_t count = 0;
6605
6606        spin_lock(&resource_alignment_lock);
6607        if (resource_alignment_param)
6608                count = sysfs_emit(buf, "%s\n", resource_alignment_param);
6609        spin_unlock(&resource_alignment_lock);
6610
6611        return count;
6612}
6613
6614static ssize_t resource_alignment_store(struct bus_type *bus,
6615                                        const char *buf, size_t count)
6616{
6617        char *param, *old, *end;
6618
6619        if (count >= (PAGE_SIZE - 1))
6620                return -EINVAL;
6621
6622        param = kstrndup(buf, count, GFP_KERNEL);
6623        if (!param)
6624                return -ENOMEM;
6625
6626        end = strchr(param, '\n');
6627        if (end)
6628                *end = '\0';
6629
6630        spin_lock(&resource_alignment_lock);
6631        old = resource_alignment_param;
6632        if (strlen(param)) {
6633                resource_alignment_param = param;
6634        } else {
6635                kfree(param);
6636                resource_alignment_param = NULL;
6637        }
6638        spin_unlock(&resource_alignment_lock);
6639
6640        kfree(old);
6641
6642        return count;
6643}
6644
6645static BUS_ATTR_RW(resource_alignment);
6646
6647static int __init pci_resource_alignment_sysfs_init(void)
6648{
6649        return bus_create_file(&pci_bus_type,
6650                                        &bus_attr_resource_alignment);
6651}
6652late_initcall(pci_resource_alignment_sysfs_init);
6653
6654static void pci_no_domains(void)
6655{
6656#ifdef CONFIG_PCI_DOMAINS
6657        pci_domains_supported = 0;
6658#endif
6659}
6660
6661#ifdef CONFIG_PCI_DOMAINS_GENERIC
6662static atomic_t __domain_nr = ATOMIC_INIT(-1);
6663
6664static int pci_get_new_domain_nr(void)
6665{
6666        return atomic_inc_return(&__domain_nr);
6667}
6668
6669static int of_pci_bus_find_domain_nr(struct device *parent)
6670{
6671        static int use_dt_domains = -1;
6672        int domain = -1;
6673
6674        if (parent)
6675                domain = of_get_pci_domain_nr(parent->of_node);
6676
6677        /*
6678         * Check DT domain and use_dt_domains values.
6679         *
6680         * If DT domain property is valid (domain >= 0) and
6681         * use_dt_domains != 0, the DT assignment is valid since this means
6682         * we have not previously allocated a domain number by using
6683         * pci_get_new_domain_nr(); we should also update use_dt_domains to
6684         * 1, to indicate that we have just assigned a domain number from
6685         * DT.
6686         *
6687         * If DT domain property value is not valid (ie domain < 0), and we
6688         * have not previously assigned a domain number from DT
6689         * (use_dt_domains != 1) we should assign a domain number by
6690         * using the:
6691         *
6692         * pci_get_new_domain_nr()
6693         *
6694         * API and update the use_dt_domains value to keep track of method we
6695         * are using to assign domain numbers (use_dt_domains = 0).
6696         *
6697         * All other combinations imply we have a platform that is trying
6698         * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6699         * which is a recipe for domain mishandling and it is prevented by
6700         * invalidating the domain value (domain = -1) and printing a
6701         * corresponding error.
6702         */
6703        if (domain >= 0 && use_dt_domains) {
6704                use_dt_domains = 1;
6705        } else if (domain < 0 && use_dt_domains != 1) {
6706                use_dt_domains = 0;
6707                domain = pci_get_new_domain_nr();
6708        } else {
6709                if (parent)
6710                        pr_err("Node %pOF has ", parent->of_node);
6711                pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6712                domain = -1;
6713        }
6714
6715        return domain;
6716}
6717
6718int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6719{
6720        return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6721                               acpi_pci_bus_find_domain_nr(bus);
6722}
6723#endif
6724
6725/**
6726 * pci_ext_cfg_avail - can we access extended PCI config space?
6727 *
6728 * Returns 1 if we can access PCI extended config space (offsets
6729 * greater than 0xff). This is the default implementation. Architecture
6730 * implementations can override this.
6731 */
6732int __weak pci_ext_cfg_avail(void)
6733{
6734        return 1;
6735}
6736
6737void __weak pci_fixup_cardbus(struct pci_bus *bus)
6738{
6739}
6740EXPORT_SYMBOL(pci_fixup_cardbus);
6741
6742static int __init pci_setup(char *str)
6743{
6744        while (str) {
6745                char *k = strchr(str, ',');
6746                if (k)
6747                        *k++ = 0;
6748                if (*str && (str = pcibios_setup(str)) && *str) {
6749                        if (!strcmp(str, "nomsi")) {
6750                                pci_no_msi();
6751                        } else if (!strncmp(str, "noats", 5)) {
6752                                pr_info("PCIe: ATS is disabled\n");
6753                                pcie_ats_disabled = true;
6754                        } else if (!strcmp(str, "noaer")) {
6755                                pci_no_aer();
6756                        } else if (!strcmp(str, "earlydump")) {
6757                                pci_early_dump = true;
6758                        } else if (!strncmp(str, "realloc=", 8)) {
6759                                pci_realloc_get_opt(str + 8);
6760                        } else if (!strncmp(str, "realloc", 7)) {
6761                                pci_realloc_get_opt("on");
6762                        } else if (!strcmp(str, "nodomains")) {
6763                                pci_no_domains();
6764                        } else if (!strncmp(str, "noari", 5)) {
6765                                pcie_ari_disabled = true;
6766                        } else if (!strncmp(str, "cbiosize=", 9)) {
6767                                pci_cardbus_io_size = memparse(str + 9, &str);
6768                        } else if (!strncmp(str, "cbmemsize=", 10)) {
6769                                pci_cardbus_mem_size = memparse(str + 10, &str);
6770                        } else if (!strncmp(str, "resource_alignment=", 19)) {
6771                                resource_alignment_param = str + 19;
6772                        } else if (!strncmp(str, "ecrc=", 5)) {
6773                                pcie_ecrc_get_policy(str + 5);
6774                        } else if (!strncmp(str, "hpiosize=", 9)) {
6775                                pci_hotplug_io_size = memparse(str + 9, &str);
6776                        } else if (!strncmp(str, "hpmmiosize=", 11)) {
6777                                pci_hotplug_mmio_size = memparse(str + 11, &str);
6778                        } else if (!strncmp(str, "hpmmioprefsize=", 15)) {
6779                                pci_hotplug_mmio_pref_size = memparse(str + 15, &str);
6780                        } else if (!strncmp(str, "hpmemsize=", 10)) {
6781                                pci_hotplug_mmio_size = memparse(str + 10, &str);
6782                                pci_hotplug_mmio_pref_size = pci_hotplug_mmio_size;
6783                        } else if (!strncmp(str, "hpbussize=", 10)) {
6784                                pci_hotplug_bus_size =
6785                                        simple_strtoul(str + 10, &str, 0);
6786                                if (pci_hotplug_bus_size > 0xff)
6787                                        pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6788                        } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6789                                pcie_bus_config = PCIE_BUS_TUNE_OFF;
6790                        } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6791                                pcie_bus_config = PCIE_BUS_SAFE;
6792                        } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6793                                pcie_bus_config = PCIE_BUS_PERFORMANCE;
6794                        } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6795                                pcie_bus_config = PCIE_BUS_PEER2PEER;
6796                        } else if (!strncmp(str, "pcie_scan_all", 13)) {
6797                                pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6798                        } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6799                                disable_acs_redir_param = str + 18;
6800                        } else {
6801                                pr_err("PCI: Unknown option `%s'\n", str);
6802                        }
6803                }
6804                str = k;
6805        }
6806        return 0;
6807}
6808early_param("pci", pci_setup);
6809
6810/*
6811 * 'resource_alignment_param' and 'disable_acs_redir_param' are initialized
6812 * in pci_setup(), above, to point to data in the __initdata section which
6813 * will be freed after the init sequence is complete. We can't allocate memory
6814 * in pci_setup() because some architectures do not have any memory allocation
6815 * service available during an early_param() call. So we allocate memory and
6816 * copy the variable here before the init section is freed.
6817 *
6818 */
6819static int __init pci_realloc_setup_params(void)
6820{
6821        resource_alignment_param = kstrdup(resource_alignment_param,
6822                                           GFP_KERNEL);
6823        disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6824
6825        return 0;
6826}
6827pure_initcall(pci_realloc_setup_params);
6828