Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * PCI Bus Services, see include/linux/pci.h for further explanation.
0004  *
0005  * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
0006  * David Mosberger-Tang
0007  *
0008  * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
0009  */
0010 
0011 #include <linux/acpi.h>
0012 #include <linux/kernel.h>
0013 #include <linux/delay.h>
0014 #include <linux/dmi.h>
0015 #include <linux/init.h>
0016 #include <linux/msi.h>
0017 #include <linux/of.h>
0018 #include <linux/pci.h>
0019 #include <linux/pm.h>
0020 #include <linux/slab.h>
0021 #include <linux/module.h>
0022 #include <linux/spinlock.h>
0023 #include <linux/string.h>
0024 #include <linux/log2.h>
0025 #include <linux/logic_pio.h>
0026 #include <linux/pm_wakeup.h>
0027 #include <linux/interrupt.h>
0028 #include <linux/device.h>
0029 #include <linux/pm_runtime.h>
0030 #include <linux/pci_hotplug.h>
0031 #include <linux/vmalloc.h>
0032 #include <asm/dma.h>
0033 #include <linux/aer.h>
0034 #include <linux/bitfield.h>
0035 #include "pci.h"
0036 
0037 DEFINE_MUTEX(pci_slot_mutex);
0038 
0039 const char *pci_power_names[] = {
0040     "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
0041 };
0042 EXPORT_SYMBOL_GPL(pci_power_names);
0043 
0044 #ifdef CONFIG_X86_32
0045 int isa_dma_bridge_buggy;
0046 EXPORT_SYMBOL(isa_dma_bridge_buggy);
0047 #endif
0048 
0049 int pci_pci_problems;
0050 EXPORT_SYMBOL(pci_pci_problems);
0051 
0052 unsigned int pci_pm_d3hot_delay;
0053 
0054 static void pci_pme_list_scan(struct work_struct *work);
0055 
0056 static LIST_HEAD(pci_pme_list);
0057 static DEFINE_MUTEX(pci_pme_list_mutex);
0058 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
0059 
0060 struct pci_pme_device {
0061     struct list_head list;
0062     struct pci_dev *dev;
0063 };
0064 
0065 #define PME_TIMEOUT 1000 /* How long between PME checks */
0066 
0067 static void pci_dev_d3_sleep(struct pci_dev *dev)
0068 {
0069     unsigned int delay = dev->d3hot_delay;
0070 
0071     if (delay < pci_pm_d3hot_delay)
0072         delay = pci_pm_d3hot_delay;
0073 
0074     if (delay)
0075         msleep(delay);
0076 }
0077 
0078 bool pci_reset_supported(struct pci_dev *dev)
0079 {
0080     return dev->reset_methods[0] != 0;
0081 }
0082 
0083 #ifdef CONFIG_PCI_DOMAINS
0084 int pci_domains_supported = 1;
0085 #endif
0086 
0087 #define DEFAULT_CARDBUS_IO_SIZE     (256)
0088 #define DEFAULT_CARDBUS_MEM_SIZE    (64*1024*1024)
0089 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
0090 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
0091 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
0092 
0093 #define DEFAULT_HOTPLUG_IO_SIZE     (256)
0094 #define DEFAULT_HOTPLUG_MMIO_SIZE   (2*1024*1024)
0095 #define DEFAULT_HOTPLUG_MMIO_PREF_SIZE  (2*1024*1024)
0096 /* hpiosize=nn can override this */
0097 unsigned long pci_hotplug_io_size  = DEFAULT_HOTPLUG_IO_SIZE;
0098 /*
0099  * pci=hpmmiosize=nnM overrides non-prefetchable MMIO size,
0100  * pci=hpmmioprefsize=nnM overrides prefetchable MMIO size;
0101  * pci=hpmemsize=nnM overrides both
0102  */
0103 unsigned long pci_hotplug_mmio_size = DEFAULT_HOTPLUG_MMIO_SIZE;
0104 unsigned long pci_hotplug_mmio_pref_size = DEFAULT_HOTPLUG_MMIO_PREF_SIZE;
0105 
0106 #define DEFAULT_HOTPLUG_BUS_SIZE    1
0107 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
0108 
0109 
0110 /* PCIe MPS/MRRS strategy; can be overridden by kernel command-line param */
0111 #ifdef CONFIG_PCIE_BUS_TUNE_OFF
0112 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_TUNE_OFF;
0113 #elif defined CONFIG_PCIE_BUS_SAFE
0114 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_SAFE;
0115 #elif defined CONFIG_PCIE_BUS_PERFORMANCE
0116 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_PERFORMANCE;
0117 #elif defined CONFIG_PCIE_BUS_PEER2PEER
0118 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_PEER2PEER;
0119 #else
0120 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
0121 #endif
0122 
0123 /*
0124  * The default CLS is used if arch didn't set CLS explicitly and not
0125  * all pci devices agree on the same value.  Arch can override either
0126  * the dfl or actual value as it sees fit.  Don't forget this is
0127  * measured in 32-bit words, not bytes.
0128  */
0129 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
0130 u8 pci_cache_line_size;
0131 
0132 /*
0133  * If we set up a device for bus mastering, we need to check the latency
0134  * timer as certain BIOSes forget to set it properly.
0135  */
0136 unsigned int pcibios_max_latency = 255;
0137 
0138 /* If set, the PCIe ARI capability will not be used. */
0139 static bool pcie_ari_disabled;
0140 
0141 /* If set, the PCIe ATS capability will not be used. */
0142 static bool pcie_ats_disabled;
0143 
0144 /* If set, the PCI config space of each device is printed during boot. */
0145 bool pci_early_dump;
0146 
0147 bool pci_ats_disabled(void)
0148 {
0149     return pcie_ats_disabled;
0150 }
0151 EXPORT_SYMBOL_GPL(pci_ats_disabled);
0152 
0153 /* Disable bridge_d3 for all PCIe ports */
0154 static bool pci_bridge_d3_disable;
0155 /* Force bridge_d3 for all PCIe ports */
0156 static bool pci_bridge_d3_force;
0157 
0158 static int __init pcie_port_pm_setup(char *str)
0159 {
0160     if (!strcmp(str, "off"))
0161         pci_bridge_d3_disable = true;
0162     else if (!strcmp(str, "force"))
0163         pci_bridge_d3_force = true;
0164     return 1;
0165 }
0166 __setup("pcie_port_pm=", pcie_port_pm_setup);
0167 
0168 /* Time to wait after a reset for device to become responsive */
0169 #define PCIE_RESET_READY_POLL_MS 60000
0170 
0171 /**
0172  * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
0173  * @bus: pointer to PCI bus structure to search
0174  *
0175  * Given a PCI bus, returns the highest PCI bus number present in the set
0176  * including the given PCI bus and its list of child PCI buses.
0177  */
0178 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
0179 {
0180     struct pci_bus *tmp;
0181     unsigned char max, n;
0182 
0183     max = bus->busn_res.end;
0184     list_for_each_entry(tmp, &bus->children, node) {
0185         n = pci_bus_max_busnr(tmp);
0186         if (n > max)
0187             max = n;
0188     }
0189     return max;
0190 }
0191 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
0192 
0193 /**
0194  * pci_status_get_and_clear_errors - return and clear error bits in PCI_STATUS
0195  * @pdev: the PCI device
0196  *
0197  * Returns error bits set in PCI_STATUS and clears them.
0198  */
0199 int pci_status_get_and_clear_errors(struct pci_dev *pdev)
0200 {
0201     u16 status;
0202     int ret;
0203 
0204     ret = pci_read_config_word(pdev, PCI_STATUS, &status);
0205     if (ret != PCIBIOS_SUCCESSFUL)
0206         return -EIO;
0207 
0208     status &= PCI_STATUS_ERROR_BITS;
0209     if (status)
0210         pci_write_config_word(pdev, PCI_STATUS, status);
0211 
0212     return status;
0213 }
0214 EXPORT_SYMBOL_GPL(pci_status_get_and_clear_errors);
0215 
0216 #ifdef CONFIG_HAS_IOMEM
0217 static void __iomem *__pci_ioremap_resource(struct pci_dev *pdev, int bar,
0218                         bool write_combine)
0219 {
0220     struct resource *res = &pdev->resource[bar];
0221     resource_size_t start = res->start;
0222     resource_size_t size = resource_size(res);
0223 
0224     /*
0225      * Make sure the BAR is actually a memory resource, not an IO resource
0226      */
0227     if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
0228         pci_err(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
0229         return NULL;
0230     }
0231 
0232     if (write_combine)
0233         return ioremap_wc(start, size);
0234 
0235     return ioremap(start, size);
0236 }
0237 
0238 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
0239 {
0240     return __pci_ioremap_resource(pdev, bar, false);
0241 }
0242 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
0243 
0244 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
0245 {
0246     return __pci_ioremap_resource(pdev, bar, true);
0247 }
0248 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
0249 #endif
0250 
0251 /**
0252  * pci_dev_str_match_path - test if a path string matches a device
0253  * @dev: the PCI device to test
0254  * @path: string to match the device against
0255  * @endptr: pointer to the string after the match
0256  *
0257  * Test if a string (typically from a kernel parameter) formatted as a
0258  * path of device/function addresses matches a PCI device. The string must
0259  * be of the form:
0260  *
0261  *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
0262  *
0263  * A path for a device can be obtained using 'lspci -t'.  Using a path
0264  * is more robust against bus renumbering than using only a single bus,
0265  * device and function address.
0266  *
0267  * Returns 1 if the string matches the device, 0 if it does not and
0268  * a negative error code if it fails to parse the string.
0269  */
0270 static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
0271                   const char **endptr)
0272 {
0273     int ret;
0274     unsigned int seg, bus, slot, func;
0275     char *wpath, *p;
0276     char end;
0277 
0278     *endptr = strchrnul(path, ';');
0279 
0280     wpath = kmemdup_nul(path, *endptr - path, GFP_ATOMIC);
0281     if (!wpath)
0282         return -ENOMEM;
0283 
0284     while (1) {
0285         p = strrchr(wpath, '/');
0286         if (!p)
0287             break;
0288         ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
0289         if (ret != 2) {
0290             ret = -EINVAL;
0291             goto free_and_exit;
0292         }
0293 
0294         if (dev->devfn != PCI_DEVFN(slot, func)) {
0295             ret = 0;
0296             goto free_and_exit;
0297         }
0298 
0299         /*
0300          * Note: we don't need to get a reference to the upstream
0301          * bridge because we hold a reference to the top level
0302          * device which should hold a reference to the bridge,
0303          * and so on.
0304          */
0305         dev = pci_upstream_bridge(dev);
0306         if (!dev) {
0307             ret = 0;
0308             goto free_and_exit;
0309         }
0310 
0311         *p = 0;
0312     }
0313 
0314     ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
0315              &func, &end);
0316     if (ret != 4) {
0317         seg = 0;
0318         ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
0319         if (ret != 3) {
0320             ret = -EINVAL;
0321             goto free_and_exit;
0322         }
0323     }
0324 
0325     ret = (seg == pci_domain_nr(dev->bus) &&
0326            bus == dev->bus->number &&
0327            dev->devfn == PCI_DEVFN(slot, func));
0328 
0329 free_and_exit:
0330     kfree(wpath);
0331     return ret;
0332 }
0333 
0334 /**
0335  * pci_dev_str_match - test if a string matches a device
0336  * @dev: the PCI device to test
0337  * @p: string to match the device against
0338  * @endptr: pointer to the string after the match
0339  *
0340  * Test if a string (typically from a kernel parameter) matches a specified
0341  * PCI device. The string may be of one of the following formats:
0342  *
0343  *   [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
0344  *   pci:<vendor>:<device>[:<subvendor>:<subdevice>]
0345  *
0346  * The first format specifies a PCI bus/device/function address which
0347  * may change if new hardware is inserted, if motherboard firmware changes,
0348  * or due to changes caused in kernel parameters. If the domain is
0349  * left unspecified, it is taken to be 0.  In order to be robust against
0350  * bus renumbering issues, a path of PCI device/function numbers may be used
0351  * to address the specific device.  The path for a device can be determined
0352  * through the use of 'lspci -t'.
0353  *
0354  * The second format matches devices using IDs in the configuration
0355  * space which may match multiple devices in the system. A value of 0
0356  * for any field will match all devices. (Note: this differs from
0357  * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
0358  * legacy reasons and convenience so users don't have to specify
0359  * FFFFFFFFs on the command line.)
0360  *
0361  * Returns 1 if the string matches the device, 0 if it does not and
0362  * a negative error code if the string cannot be parsed.
0363  */
0364 static int pci_dev_str_match(struct pci_dev *dev, const char *p,
0365                  const char **endptr)
0366 {
0367     int ret;
0368     int count;
0369     unsigned short vendor, device, subsystem_vendor, subsystem_device;
0370 
0371     if (strncmp(p, "pci:", 4) == 0) {
0372         /* PCI vendor/device (subvendor/subdevice) IDs are specified */
0373         p += 4;
0374         ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
0375                  &subsystem_vendor, &subsystem_device, &count);
0376         if (ret != 4) {
0377             ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
0378             if (ret != 2)
0379                 return -EINVAL;
0380 
0381             subsystem_vendor = 0;
0382             subsystem_device = 0;
0383         }
0384 
0385         p += count;
0386 
0387         if ((!vendor || vendor == dev->vendor) &&
0388             (!device || device == dev->device) &&
0389             (!subsystem_vendor ||
0390                 subsystem_vendor == dev->subsystem_vendor) &&
0391             (!subsystem_device ||
0392                 subsystem_device == dev->subsystem_device))
0393             goto found;
0394     } else {
0395         /*
0396          * PCI Bus, Device, Function IDs are specified
0397          * (optionally, may include a path of devfns following it)
0398          */
0399         ret = pci_dev_str_match_path(dev, p, &p);
0400         if (ret < 0)
0401             return ret;
0402         else if (ret)
0403             goto found;
0404     }
0405 
0406     *endptr = p;
0407     return 0;
0408 
0409 found:
0410     *endptr = p;
0411     return 1;
0412 }
0413 
0414 static u8 __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
0415                   u8 pos, int cap, int *ttl)
0416 {
0417     u8 id;
0418     u16 ent;
0419 
0420     pci_bus_read_config_byte(bus, devfn, pos, &pos);
0421 
0422     while ((*ttl)--) {
0423         if (pos < 0x40)
0424             break;
0425         pos &= ~3;
0426         pci_bus_read_config_word(bus, devfn, pos, &ent);
0427 
0428         id = ent & 0xff;
0429         if (id == 0xff)
0430             break;
0431         if (id == cap)
0432             return pos;
0433         pos = (ent >> 8);
0434     }
0435     return 0;
0436 }
0437 
0438 static u8 __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
0439                   u8 pos, int cap)
0440 {
0441     int ttl = PCI_FIND_CAP_TTL;
0442 
0443     return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
0444 }
0445 
0446 u8 pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
0447 {
0448     return __pci_find_next_cap(dev->bus, dev->devfn,
0449                    pos + PCI_CAP_LIST_NEXT, cap);
0450 }
0451 EXPORT_SYMBOL_GPL(pci_find_next_capability);
0452 
0453 static u8 __pci_bus_find_cap_start(struct pci_bus *bus,
0454                     unsigned int devfn, u8 hdr_type)
0455 {
0456     u16 status;
0457 
0458     pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
0459     if (!(status & PCI_STATUS_CAP_LIST))
0460         return 0;
0461 
0462     switch (hdr_type) {
0463     case PCI_HEADER_TYPE_NORMAL:
0464     case PCI_HEADER_TYPE_BRIDGE:
0465         return PCI_CAPABILITY_LIST;
0466     case PCI_HEADER_TYPE_CARDBUS:
0467         return PCI_CB_CAPABILITY_LIST;
0468     }
0469 
0470     return 0;
0471 }
0472 
0473 /**
0474  * pci_find_capability - query for devices' capabilities
0475  * @dev: PCI device to query
0476  * @cap: capability code
0477  *
0478  * Tell if a device supports a given PCI capability.
0479  * Returns the address of the requested capability structure within the
0480  * device's PCI configuration space or 0 in case the device does not
0481  * support it.  Possible values for @cap include:
0482  *
0483  *  %PCI_CAP_ID_PM           Power Management
0484  *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
0485  *  %PCI_CAP_ID_VPD          Vital Product Data
0486  *  %PCI_CAP_ID_SLOTID       Slot Identification
0487  *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
0488  *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
0489  *  %PCI_CAP_ID_PCIX         PCI-X
0490  *  %PCI_CAP_ID_EXP          PCI Express
0491  */
0492 u8 pci_find_capability(struct pci_dev *dev, int cap)
0493 {
0494     u8 pos;
0495 
0496     pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
0497     if (pos)
0498         pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
0499 
0500     return pos;
0501 }
0502 EXPORT_SYMBOL(pci_find_capability);
0503 
0504 /**
0505  * pci_bus_find_capability - query for devices' capabilities
0506  * @bus: the PCI bus to query
0507  * @devfn: PCI device to query
0508  * @cap: capability code
0509  *
0510  * Like pci_find_capability() but works for PCI devices that do not have a
0511  * pci_dev structure set up yet.
0512  *
0513  * Returns the address of the requested capability structure within the
0514  * device's PCI configuration space or 0 in case the device does not
0515  * support it.
0516  */
0517 u8 pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
0518 {
0519     u8 hdr_type, pos;
0520 
0521     pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
0522 
0523     pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
0524     if (pos)
0525         pos = __pci_find_next_cap(bus, devfn, pos, cap);
0526 
0527     return pos;
0528 }
0529 EXPORT_SYMBOL(pci_bus_find_capability);
0530 
0531 /**
0532  * pci_find_next_ext_capability - Find an extended capability
0533  * @dev: PCI device to query
0534  * @start: address at which to start looking (0 to start at beginning of list)
0535  * @cap: capability code
0536  *
0537  * Returns the address of the next matching extended capability structure
0538  * within the device's PCI configuration space or 0 if the device does
0539  * not support it.  Some capabilities can occur several times, e.g., the
0540  * vendor-specific capability, and this provides a way to find them all.
0541  */
0542 u16 pci_find_next_ext_capability(struct pci_dev *dev, u16 start, int cap)
0543 {
0544     u32 header;
0545     int ttl;
0546     u16 pos = PCI_CFG_SPACE_SIZE;
0547 
0548     /* minimum 8 bytes per capability */
0549     ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
0550 
0551     if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
0552         return 0;
0553 
0554     if (start)
0555         pos = start;
0556 
0557     if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
0558         return 0;
0559 
0560     /*
0561      * If we have no capabilities, this is indicated by cap ID,
0562      * cap version and next pointer all being 0.
0563      */
0564     if (header == 0)
0565         return 0;
0566 
0567     while (ttl-- > 0) {
0568         if (PCI_EXT_CAP_ID(header) == cap && pos != start)
0569             return pos;
0570 
0571         pos = PCI_EXT_CAP_NEXT(header);
0572         if (pos < PCI_CFG_SPACE_SIZE)
0573             break;
0574 
0575         if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
0576             break;
0577     }
0578 
0579     return 0;
0580 }
0581 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
0582 
0583 /**
0584  * pci_find_ext_capability - Find an extended capability
0585  * @dev: PCI device to query
0586  * @cap: capability code
0587  *
0588  * Returns the address of the requested extended capability structure
0589  * within the device's PCI configuration space or 0 if the device does
0590  * not support it.  Possible values for @cap include:
0591  *
0592  *  %PCI_EXT_CAP_ID_ERR     Advanced Error Reporting
0593  *  %PCI_EXT_CAP_ID_VC      Virtual Channel
0594  *  %PCI_EXT_CAP_ID_DSN     Device Serial Number
0595  *  %PCI_EXT_CAP_ID_PWR     Power Budgeting
0596  */
0597 u16 pci_find_ext_capability(struct pci_dev *dev, int cap)
0598 {
0599     return pci_find_next_ext_capability(dev, 0, cap);
0600 }
0601 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
0602 
0603 /**
0604  * pci_get_dsn - Read and return the 8-byte Device Serial Number
0605  * @dev: PCI device to query
0606  *
0607  * Looks up the PCI_EXT_CAP_ID_DSN and reads the 8 bytes of the Device Serial
0608  * Number.
0609  *
0610  * Returns the DSN, or zero if the capability does not exist.
0611  */
0612 u64 pci_get_dsn(struct pci_dev *dev)
0613 {
0614     u32 dword;
0615     u64 dsn;
0616     int pos;
0617 
0618     pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DSN);
0619     if (!pos)
0620         return 0;
0621 
0622     /*
0623      * The Device Serial Number is two dwords offset 4 bytes from the
0624      * capability position. The specification says that the first dword is
0625      * the lower half, and the second dword is the upper half.
0626      */
0627     pos += 4;
0628     pci_read_config_dword(dev, pos, &dword);
0629     dsn = (u64)dword;
0630     pci_read_config_dword(dev, pos + 4, &dword);
0631     dsn |= ((u64)dword) << 32;
0632 
0633     return dsn;
0634 }
0635 EXPORT_SYMBOL_GPL(pci_get_dsn);
0636 
0637 static u8 __pci_find_next_ht_cap(struct pci_dev *dev, u8 pos, int ht_cap)
0638 {
0639     int rc, ttl = PCI_FIND_CAP_TTL;
0640     u8 cap, mask;
0641 
0642     if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
0643         mask = HT_3BIT_CAP_MASK;
0644     else
0645         mask = HT_5BIT_CAP_MASK;
0646 
0647     pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
0648                       PCI_CAP_ID_HT, &ttl);
0649     while (pos) {
0650         rc = pci_read_config_byte(dev, pos + 3, &cap);
0651         if (rc != PCIBIOS_SUCCESSFUL)
0652             return 0;
0653 
0654         if ((cap & mask) == ht_cap)
0655             return pos;
0656 
0657         pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
0658                           pos + PCI_CAP_LIST_NEXT,
0659                           PCI_CAP_ID_HT, &ttl);
0660     }
0661 
0662     return 0;
0663 }
0664 
0665 /**
0666  * pci_find_next_ht_capability - query a device's HyperTransport capabilities
0667  * @dev: PCI device to query
0668  * @pos: Position from which to continue searching
0669  * @ht_cap: HyperTransport capability code
0670  *
0671  * To be used in conjunction with pci_find_ht_capability() to search for
0672  * all capabilities matching @ht_cap. @pos should always be a value returned
0673  * from pci_find_ht_capability().
0674  *
0675  * NB. To be 100% safe against broken PCI devices, the caller should take
0676  * steps to avoid an infinite loop.
0677  */
0678 u8 pci_find_next_ht_capability(struct pci_dev *dev, u8 pos, int ht_cap)
0679 {
0680     return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
0681 }
0682 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
0683 
0684 /**
0685  * pci_find_ht_capability - query a device's HyperTransport capabilities
0686  * @dev: PCI device to query
0687  * @ht_cap: HyperTransport capability code
0688  *
0689  * Tell if a device supports a given HyperTransport capability.
0690  * Returns an address within the device's PCI configuration space
0691  * or 0 in case the device does not support the request capability.
0692  * The address points to the PCI capability, of type PCI_CAP_ID_HT,
0693  * which has a HyperTransport capability matching @ht_cap.
0694  */
0695 u8 pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
0696 {
0697     u8 pos;
0698 
0699     pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
0700     if (pos)
0701         pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
0702 
0703     return pos;
0704 }
0705 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
0706 
0707 /**
0708  * pci_find_vsec_capability - Find a vendor-specific extended capability
0709  * @dev: PCI device to query
0710  * @vendor: Vendor ID for which capability is defined
0711  * @cap: Vendor-specific capability ID
0712  *
0713  * If @dev has Vendor ID @vendor, search for a VSEC capability with
0714  * VSEC ID @cap. If found, return the capability offset in
0715  * config space; otherwise return 0.
0716  */
0717 u16 pci_find_vsec_capability(struct pci_dev *dev, u16 vendor, int cap)
0718 {
0719     u16 vsec = 0;
0720     u32 header;
0721 
0722     if (vendor != dev->vendor)
0723         return 0;
0724 
0725     while ((vsec = pci_find_next_ext_capability(dev, vsec,
0726                              PCI_EXT_CAP_ID_VNDR))) {
0727         if (pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER,
0728                       &header) == PCIBIOS_SUCCESSFUL &&
0729             PCI_VNDR_HEADER_ID(header) == cap)
0730             return vsec;
0731     }
0732 
0733     return 0;
0734 }
0735 EXPORT_SYMBOL_GPL(pci_find_vsec_capability);
0736 
0737 /**
0738  * pci_find_dvsec_capability - Find DVSEC for vendor
0739  * @dev: PCI device to query
0740  * @vendor: Vendor ID to match for the DVSEC
0741  * @dvsec: Designated Vendor-specific capability ID
0742  *
0743  * If DVSEC has Vendor ID @vendor and DVSEC ID @dvsec return the capability
0744  * offset in config space; otherwise return 0.
0745  */
0746 u16 pci_find_dvsec_capability(struct pci_dev *dev, u16 vendor, u16 dvsec)
0747 {
0748     int pos;
0749 
0750     pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DVSEC);
0751     if (!pos)
0752         return 0;
0753 
0754     while (pos) {
0755         u16 v, id;
0756 
0757         pci_read_config_word(dev, pos + PCI_DVSEC_HEADER1, &v);
0758         pci_read_config_word(dev, pos + PCI_DVSEC_HEADER2, &id);
0759         if (vendor == v && dvsec == id)
0760             return pos;
0761 
0762         pos = pci_find_next_ext_capability(dev, pos, PCI_EXT_CAP_ID_DVSEC);
0763     }
0764 
0765     return 0;
0766 }
0767 EXPORT_SYMBOL_GPL(pci_find_dvsec_capability);
0768 
0769 /**
0770  * pci_find_parent_resource - return resource region of parent bus of given
0771  *                region
0772  * @dev: PCI device structure contains resources to be searched
0773  * @res: child resource record for which parent is sought
0774  *
0775  * For given resource region of given device, return the resource region of
0776  * parent bus the given region is contained in.
0777  */
0778 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
0779                       struct resource *res)
0780 {
0781     const struct pci_bus *bus = dev->bus;
0782     struct resource *r;
0783     int i;
0784 
0785     pci_bus_for_each_resource(bus, r, i) {
0786         if (!r)
0787             continue;
0788         if (resource_contains(r, res)) {
0789 
0790             /*
0791              * If the window is prefetchable but the BAR is
0792              * not, the allocator made a mistake.
0793              */
0794             if (r->flags & IORESOURCE_PREFETCH &&
0795                 !(res->flags & IORESOURCE_PREFETCH))
0796                 return NULL;
0797 
0798             /*
0799              * If we're below a transparent bridge, there may
0800              * be both a positively-decoded aperture and a
0801              * subtractively-decoded region that contain the BAR.
0802              * We want the positively-decoded one, so this depends
0803              * on pci_bus_for_each_resource() giving us those
0804              * first.
0805              */
0806             return r;
0807         }
0808     }
0809     return NULL;
0810 }
0811 EXPORT_SYMBOL(pci_find_parent_resource);
0812 
0813 /**
0814  * pci_find_resource - Return matching PCI device resource
0815  * @dev: PCI device to query
0816  * @res: Resource to look for
0817  *
0818  * Goes over standard PCI resources (BARs) and checks if the given resource
0819  * is partially or fully contained in any of them. In that case the
0820  * matching resource is returned, %NULL otherwise.
0821  */
0822 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
0823 {
0824     int i;
0825 
0826     for (i = 0; i < PCI_STD_NUM_BARS; i++) {
0827         struct resource *r = &dev->resource[i];
0828 
0829         if (r->start && resource_contains(r, res))
0830             return r;
0831     }
0832 
0833     return NULL;
0834 }
0835 EXPORT_SYMBOL(pci_find_resource);
0836 
0837 /**
0838  * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
0839  * @dev: the PCI device to operate on
0840  * @pos: config space offset of status word
0841  * @mask: mask of bit(s) to care about in status word
0842  *
0843  * Return 1 when mask bit(s) in status word clear, 0 otherwise.
0844  */
0845 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
0846 {
0847     int i;
0848 
0849     /* Wait for Transaction Pending bit clean */
0850     for (i = 0; i < 4; i++) {
0851         u16 status;
0852         if (i)
0853             msleep((1 << (i - 1)) * 100);
0854 
0855         pci_read_config_word(dev, pos, &status);
0856         if (!(status & mask))
0857             return 1;
0858     }
0859 
0860     return 0;
0861 }
0862 
0863 static int pci_acs_enable;
0864 
0865 /**
0866  * pci_request_acs - ask for ACS to be enabled if supported
0867  */
0868 void pci_request_acs(void)
0869 {
0870     pci_acs_enable = 1;
0871 }
0872 
0873 static const char *disable_acs_redir_param;
0874 
0875 /**
0876  * pci_disable_acs_redir - disable ACS redirect capabilities
0877  * @dev: the PCI device
0878  *
0879  * For only devices specified in the disable_acs_redir parameter.
0880  */
0881 static void pci_disable_acs_redir(struct pci_dev *dev)
0882 {
0883     int ret = 0;
0884     const char *p;
0885     int pos;
0886     u16 ctrl;
0887 
0888     if (!disable_acs_redir_param)
0889         return;
0890 
0891     p = disable_acs_redir_param;
0892     while (*p) {
0893         ret = pci_dev_str_match(dev, p, &p);
0894         if (ret < 0) {
0895             pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
0896                      disable_acs_redir_param);
0897 
0898             break;
0899         } else if (ret == 1) {
0900             /* Found a match */
0901             break;
0902         }
0903 
0904         if (*p != ';' && *p != ',') {
0905             /* End of param or invalid format */
0906             break;
0907         }
0908         p++;
0909     }
0910 
0911     if (ret != 1)
0912         return;
0913 
0914     if (!pci_dev_specific_disable_acs_redir(dev))
0915         return;
0916 
0917     pos = dev->acs_cap;
0918     if (!pos) {
0919         pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
0920         return;
0921     }
0922 
0923     pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
0924 
0925     /* P2P Request & Completion Redirect */
0926     ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
0927 
0928     pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
0929 
0930     pci_info(dev, "disabled ACS redirect\n");
0931 }
0932 
0933 /**
0934  * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
0935  * @dev: the PCI device
0936  */
0937 static void pci_std_enable_acs(struct pci_dev *dev)
0938 {
0939     int pos;
0940     u16 cap;
0941     u16 ctrl;
0942 
0943     pos = dev->acs_cap;
0944     if (!pos)
0945         return;
0946 
0947     pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
0948     pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
0949 
0950     /* Source Validation */
0951     ctrl |= (cap & PCI_ACS_SV);
0952 
0953     /* P2P Request Redirect */
0954     ctrl |= (cap & PCI_ACS_RR);
0955 
0956     /* P2P Completion Redirect */
0957     ctrl |= (cap & PCI_ACS_CR);
0958 
0959     /* Upstream Forwarding */
0960     ctrl |= (cap & PCI_ACS_UF);
0961 
0962     /* Enable Translation Blocking for external devices and noats */
0963     if (pci_ats_disabled() || dev->external_facing || dev->untrusted)
0964         ctrl |= (cap & PCI_ACS_TB);
0965 
0966     pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
0967 }
0968 
0969 /**
0970  * pci_enable_acs - enable ACS if hardware support it
0971  * @dev: the PCI device
0972  */
0973 static void pci_enable_acs(struct pci_dev *dev)
0974 {
0975     if (!pci_acs_enable)
0976         goto disable_acs_redir;
0977 
0978     if (!pci_dev_specific_enable_acs(dev))
0979         goto disable_acs_redir;
0980 
0981     pci_std_enable_acs(dev);
0982 
0983 disable_acs_redir:
0984     /*
0985      * Note: pci_disable_acs_redir() must be called even if ACS was not
0986      * enabled by the kernel because it may have been enabled by
0987      * platform firmware.  So if we are told to disable it, we should
0988      * always disable it after setting the kernel's default
0989      * preferences.
0990      */
0991     pci_disable_acs_redir(dev);
0992 }
0993 
0994 /**
0995  * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
0996  * @dev: PCI device to have its BARs restored
0997  *
0998  * Restore the BAR values for a given device, so as to make it
0999  * accessible by its driver.
1000  */
1001 static void pci_restore_bars(struct pci_dev *dev)
1002 {
1003     int i;
1004 
1005     for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
1006         pci_update_resource(dev, i);
1007 }
1008 
1009 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
1010 {
1011     if (pci_use_mid_pm())
1012         return true;
1013 
1014     return acpi_pci_power_manageable(dev);
1015 }
1016 
1017 static inline int platform_pci_set_power_state(struct pci_dev *dev,
1018                            pci_power_t t)
1019 {
1020     if (pci_use_mid_pm())
1021         return mid_pci_set_power_state(dev, t);
1022 
1023     return acpi_pci_set_power_state(dev, t);
1024 }
1025 
1026 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
1027 {
1028     if (pci_use_mid_pm())
1029         return mid_pci_get_power_state(dev);
1030 
1031     return acpi_pci_get_power_state(dev);
1032 }
1033 
1034 static inline void platform_pci_refresh_power_state(struct pci_dev *dev)
1035 {
1036     if (!pci_use_mid_pm())
1037         acpi_pci_refresh_power_state(dev);
1038 }
1039 
1040 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
1041 {
1042     if (pci_use_mid_pm())
1043         return PCI_POWER_ERROR;
1044 
1045     return acpi_pci_choose_state(dev);
1046 }
1047 
1048 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
1049 {
1050     if (pci_use_mid_pm())
1051         return PCI_POWER_ERROR;
1052 
1053     return acpi_pci_wakeup(dev, enable);
1054 }
1055 
1056 static inline bool platform_pci_need_resume(struct pci_dev *dev)
1057 {
1058     if (pci_use_mid_pm())
1059         return false;
1060 
1061     return acpi_pci_need_resume(dev);
1062 }
1063 
1064 static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
1065 {
1066     if (pci_use_mid_pm())
1067         return false;
1068 
1069     return acpi_pci_bridge_d3(dev);
1070 }
1071 
1072 /**
1073  * pci_update_current_state - Read power state of given device and cache it
1074  * @dev: PCI device to handle.
1075  * @state: State to cache in case the device doesn't have the PM capability
1076  *
1077  * The power state is read from the PMCSR register, which however is
1078  * inaccessible in D3cold.  The platform firmware is therefore queried first
1079  * to detect accessibility of the register.  In case the platform firmware
1080  * reports an incorrect state or the device isn't power manageable by the
1081  * platform at all, we try to detect D3cold by testing accessibility of the
1082  * vendor ID in config space.
1083  */
1084 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
1085 {
1086     if (platform_pci_get_power_state(dev) == PCI_D3cold) {
1087         dev->current_state = PCI_D3cold;
1088     } else if (dev->pm_cap) {
1089         u16 pmcsr;
1090 
1091         pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1092         if (PCI_POSSIBLE_ERROR(pmcsr)) {
1093             dev->current_state = PCI_D3cold;
1094             return;
1095         }
1096         dev->current_state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1097     } else {
1098         dev->current_state = state;
1099     }
1100 }
1101 
1102 /**
1103  * pci_refresh_power_state - Refresh the given device's power state data
1104  * @dev: Target PCI device.
1105  *
1106  * Ask the platform to refresh the devices power state information and invoke
1107  * pci_update_current_state() to update its current PCI power state.
1108  */
1109 void pci_refresh_power_state(struct pci_dev *dev)
1110 {
1111     platform_pci_refresh_power_state(dev);
1112     pci_update_current_state(dev, dev->current_state);
1113 }
1114 
1115 /**
1116  * pci_platform_power_transition - Use platform to change device power state
1117  * @dev: PCI device to handle.
1118  * @state: State to put the device into.
1119  */
1120 int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
1121 {
1122     int error;
1123 
1124     error = platform_pci_set_power_state(dev, state);
1125     if (!error)
1126         pci_update_current_state(dev, state);
1127     else if (!dev->pm_cap) /* Fall back to PCI_D0 */
1128         dev->current_state = PCI_D0;
1129 
1130     return error;
1131 }
1132 EXPORT_SYMBOL_GPL(pci_platform_power_transition);
1133 
1134 static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
1135 {
1136     pm_request_resume(&pci_dev->dev);
1137     return 0;
1138 }
1139 
1140 /**
1141  * pci_resume_bus - Walk given bus and runtime resume devices on it
1142  * @bus: Top bus of the subtree to walk.
1143  */
1144 void pci_resume_bus(struct pci_bus *bus)
1145 {
1146     if (bus)
1147         pci_walk_bus(bus, pci_resume_one, NULL);
1148 }
1149 
1150 static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
1151 {
1152     int delay = 1;
1153     u32 id;
1154 
1155     /*
1156      * After reset, the device should not silently discard config
1157      * requests, but it may still indicate that it needs more time by
1158      * responding to them with CRS completions.  The Root Port will
1159      * generally synthesize ~0 (PCI_ERROR_RESPONSE) data to complete
1160      * the read (except when CRS SV is enabled and the read was for the
1161      * Vendor ID; in that case it synthesizes 0x0001 data).
1162      *
1163      * Wait for the device to return a non-CRS completion.  Read the
1164      * Command register instead of Vendor ID so we don't have to
1165      * contend with the CRS SV value.
1166      */
1167     pci_read_config_dword(dev, PCI_COMMAND, &id);
1168     while (PCI_POSSIBLE_ERROR(id)) {
1169         if (delay > timeout) {
1170             pci_warn(dev, "not ready %dms after %s; giving up\n",
1171                  delay - 1, reset_type);
1172             return -ENOTTY;
1173         }
1174 
1175         if (delay > 1000)
1176             pci_info(dev, "not ready %dms after %s; waiting\n",
1177                  delay - 1, reset_type);
1178 
1179         msleep(delay);
1180         delay *= 2;
1181         pci_read_config_dword(dev, PCI_COMMAND, &id);
1182     }
1183 
1184     if (delay > 1000)
1185         pci_info(dev, "ready %dms after %s\n", delay - 1,
1186              reset_type);
1187 
1188     return 0;
1189 }
1190 
1191 /**
1192  * pci_power_up - Put the given device into D0
1193  * @dev: PCI device to power up
1194  *
1195  * On success, return 0 or 1, depending on whether or not it is necessary to
1196  * restore the device's BARs subsequently (1 is returned in that case).
1197  */
1198 int pci_power_up(struct pci_dev *dev)
1199 {
1200     bool need_restore;
1201     pci_power_t state;
1202     u16 pmcsr;
1203 
1204     platform_pci_set_power_state(dev, PCI_D0);
1205 
1206     if (!dev->pm_cap) {
1207         state = platform_pci_get_power_state(dev);
1208         if (state == PCI_UNKNOWN)
1209             dev->current_state = PCI_D0;
1210         else
1211             dev->current_state = state;
1212 
1213         if (state == PCI_D0)
1214             return 0;
1215 
1216         return -EIO;
1217     }
1218 
1219     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1220     if (PCI_POSSIBLE_ERROR(pmcsr)) {
1221         pci_err(dev, "Unable to change power state from %s to D0, device inaccessible\n",
1222             pci_power_name(dev->current_state));
1223         dev->current_state = PCI_D3cold;
1224         return -EIO;
1225     }
1226 
1227     state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1228 
1229     need_restore = (state == PCI_D3hot || dev->current_state >= PCI_D3hot) &&
1230             !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET);
1231 
1232     if (state == PCI_D0)
1233         goto end;
1234 
1235     /*
1236      * Force the entire word to 0. This doesn't affect PME_Status, disables
1237      * PME_En, and sets PowerState to 0.
1238      */
1239     pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, 0);
1240 
1241     /* Mandatory transition delays; see PCI PM 1.2. */
1242     if (state == PCI_D3hot)
1243         pci_dev_d3_sleep(dev);
1244     else if (state == PCI_D2)
1245         udelay(PCI_PM_D2_DELAY);
1246 
1247 end:
1248     dev->current_state = PCI_D0;
1249     if (need_restore)
1250         return 1;
1251 
1252     return 0;
1253 }
1254 
1255 /**
1256  * pci_set_full_power_state - Put a PCI device into D0 and update its state
1257  * @dev: PCI device to power up
1258  *
1259  * Call pci_power_up() to put @dev into D0, read from its PCI_PM_CTRL register
1260  * to confirm the state change, restore its BARs if they might be lost and
1261  * reconfigure ASPM in acordance with the new power state.
1262  *
1263  * If pci_restore_state() is going to be called right after a power state change
1264  * to D0, it is more efficient to use pci_power_up() directly instead of this
1265  * function.
1266  */
1267 static int pci_set_full_power_state(struct pci_dev *dev)
1268 {
1269     u16 pmcsr;
1270     int ret;
1271 
1272     ret = pci_power_up(dev);
1273     if (ret < 0)
1274         return ret;
1275 
1276     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1277     dev->current_state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1278     if (dev->current_state != PCI_D0) {
1279         pci_info_ratelimited(dev, "Refused to change power state from %s to D0\n",
1280                      pci_power_name(dev->current_state));
1281     } else if (ret > 0) {
1282         /*
1283          * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
1284          * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
1285          * from D3hot to D0 _may_ perform an internal reset, thereby
1286          * going to "D0 Uninitialized" rather than "D0 Initialized".
1287          * For example, at least some versions of the 3c905B and the
1288          * 3c556B exhibit this behaviour.
1289          *
1290          * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
1291          * devices in a D3hot state at boot.  Consequently, we need to
1292          * restore at least the BARs so that the device will be
1293          * accessible to its driver.
1294          */
1295         pci_restore_bars(dev);
1296     }
1297 
1298     return 0;
1299 }
1300 
1301 /**
1302  * __pci_dev_set_current_state - Set current state of a PCI device
1303  * @dev: Device to handle
1304  * @data: pointer to state to be set
1305  */
1306 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1307 {
1308     pci_power_t state = *(pci_power_t *)data;
1309 
1310     dev->current_state = state;
1311     return 0;
1312 }
1313 
1314 /**
1315  * pci_bus_set_current_state - Walk given bus and set current state of devices
1316  * @bus: Top bus of the subtree to walk.
1317  * @state: state to be set
1318  */
1319 void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1320 {
1321     if (bus)
1322         pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1323 }
1324 
1325 /**
1326  * pci_set_low_power_state - Put a PCI device into a low-power state.
1327  * @dev: PCI device to handle.
1328  * @state: PCI power state (D1, D2, D3hot) to put the device into.
1329  *
1330  * Use the device's PCI_PM_CTRL register to put it into a low-power state.
1331  *
1332  * RETURN VALUE:
1333  * -EINVAL if the requested state is invalid.
1334  * -EIO if device does not support PCI PM or its PM capabilities register has a
1335  * wrong version, or device doesn't support the requested state.
1336  * 0 if device already is in the requested state.
1337  * 0 if device's power state has been successfully changed.
1338  */
1339 static int pci_set_low_power_state(struct pci_dev *dev, pci_power_t state)
1340 {
1341     u16 pmcsr;
1342 
1343     if (!dev->pm_cap)
1344         return -EIO;
1345 
1346     /*
1347      * Validate transition: We can enter D0 from any state, but if
1348      * we're already in a low-power state, we can only go deeper.  E.g.,
1349      * we can go from D1 to D3, but we can't go directly from D3 to D1;
1350      * we'd have to go from D3 to D0, then to D1.
1351      */
1352     if (dev->current_state <= PCI_D3cold && dev->current_state > state) {
1353         pci_dbg(dev, "Invalid power transition (from %s to %s)\n",
1354             pci_power_name(dev->current_state),
1355             pci_power_name(state));
1356         return -EINVAL;
1357     }
1358 
1359     /* Check if this device supports the desired state */
1360     if ((state == PCI_D1 && !dev->d1_support)
1361        || (state == PCI_D2 && !dev->d2_support))
1362         return -EIO;
1363 
1364     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1365     if (PCI_POSSIBLE_ERROR(pmcsr)) {
1366         pci_err(dev, "Unable to change power state from %s to %s, device inaccessible\n",
1367             pci_power_name(dev->current_state),
1368             pci_power_name(state));
1369         dev->current_state = PCI_D3cold;
1370         return -EIO;
1371     }
1372 
1373     pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1374     pmcsr |= state;
1375 
1376     /* Enter specified state */
1377     pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1378 
1379     /* Mandatory power management transition delays; see PCI PM 1.2. */
1380     if (state == PCI_D3hot)
1381         pci_dev_d3_sleep(dev);
1382     else if (state == PCI_D2)
1383         udelay(PCI_PM_D2_DELAY);
1384 
1385     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1386     dev->current_state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1387     if (dev->current_state != state)
1388         pci_info_ratelimited(dev, "Refused to change power state from %s to %s\n",
1389                      pci_power_name(dev->current_state),
1390                      pci_power_name(state));
1391 
1392     return 0;
1393 }
1394 
1395 /**
1396  * pci_set_power_state - Set the power state of a PCI device
1397  * @dev: PCI device to handle.
1398  * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1399  *
1400  * Transition a device to a new power state, using the platform firmware and/or
1401  * the device's PCI PM registers.
1402  *
1403  * RETURN VALUE:
1404  * -EINVAL if the requested state is invalid.
1405  * -EIO if device does not support PCI PM or its PM capabilities register has a
1406  * wrong version, or device doesn't support the requested state.
1407  * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1408  * 0 if device already is in the requested state.
1409  * 0 if the transition is to D3 but D3 is not supported.
1410  * 0 if device's power state has been successfully changed.
1411  */
1412 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1413 {
1414     int error;
1415 
1416     /* Bound the state we're entering */
1417     if (state > PCI_D3cold)
1418         state = PCI_D3cold;
1419     else if (state < PCI_D0)
1420         state = PCI_D0;
1421     else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1422 
1423         /*
1424          * If the device or the parent bridge do not support PCI
1425          * PM, ignore the request if we're doing anything other
1426          * than putting it into D0 (which would only happen on
1427          * boot).
1428          */
1429         return 0;
1430 
1431     /* Check if we're already there */
1432     if (dev->current_state == state)
1433         return 0;
1434 
1435     if (state == PCI_D0)
1436         return pci_set_full_power_state(dev);
1437 
1438     /*
1439      * This device is quirked not to be put into D3, so don't put it in
1440      * D3
1441      */
1442     if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1443         return 0;
1444 
1445     if (state == PCI_D3cold) {
1446         /*
1447          * To put the device in D3cold, put it into D3hot in the native
1448          * way, then put it into D3cold using platform ops.
1449          */
1450         error = pci_set_low_power_state(dev, PCI_D3hot);
1451 
1452         if (pci_platform_power_transition(dev, PCI_D3cold))
1453             return error;
1454 
1455         /* Powering off a bridge may power off the whole hierarchy */
1456         if (dev->current_state == PCI_D3cold)
1457             pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1458     } else {
1459         error = pci_set_low_power_state(dev, state);
1460 
1461         if (pci_platform_power_transition(dev, state))
1462             return error;
1463     }
1464 
1465     return 0;
1466 }
1467 EXPORT_SYMBOL(pci_set_power_state);
1468 
1469 #define PCI_EXP_SAVE_REGS   7
1470 
1471 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1472                                u16 cap, bool extended)
1473 {
1474     struct pci_cap_saved_state *tmp;
1475 
1476     hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1477         if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1478             return tmp;
1479     }
1480     return NULL;
1481 }
1482 
1483 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1484 {
1485     return _pci_find_saved_cap(dev, cap, false);
1486 }
1487 
1488 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1489 {
1490     return _pci_find_saved_cap(dev, cap, true);
1491 }
1492 
1493 static int pci_save_pcie_state(struct pci_dev *dev)
1494 {
1495     int i = 0;
1496     struct pci_cap_saved_state *save_state;
1497     u16 *cap;
1498 
1499     if (!pci_is_pcie(dev))
1500         return 0;
1501 
1502     save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1503     if (!save_state) {
1504         pci_err(dev, "buffer not found in %s\n", __func__);
1505         return -ENOMEM;
1506     }
1507 
1508     cap = (u16 *)&save_state->cap.data[0];
1509     pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1510     pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1511     pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1512     pcie_capability_read_word(dev, PCI_EXP_RTCTL,  &cap[i++]);
1513     pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1514     pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1515     pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1516 
1517     return 0;
1518 }
1519 
1520 void pci_bridge_reconfigure_ltr(struct pci_dev *dev)
1521 {
1522 #ifdef CONFIG_PCIEASPM
1523     struct pci_dev *bridge;
1524     u32 ctl;
1525 
1526     bridge = pci_upstream_bridge(dev);
1527     if (bridge && bridge->ltr_path) {
1528         pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2, &ctl);
1529         if (!(ctl & PCI_EXP_DEVCTL2_LTR_EN)) {
1530             pci_dbg(bridge, "re-enabling LTR\n");
1531             pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
1532                          PCI_EXP_DEVCTL2_LTR_EN);
1533         }
1534     }
1535 #endif
1536 }
1537 
1538 static void pci_restore_pcie_state(struct pci_dev *dev)
1539 {
1540     int i = 0;
1541     struct pci_cap_saved_state *save_state;
1542     u16 *cap;
1543 
1544     save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1545     if (!save_state)
1546         return;
1547 
1548     /*
1549      * Downstream ports reset the LTR enable bit when link goes down.
1550      * Check and re-configure the bit here before restoring device.
1551      * PCIe r5.0, sec 7.5.3.16.
1552      */
1553     pci_bridge_reconfigure_ltr(dev);
1554 
1555     cap = (u16 *)&save_state->cap.data[0];
1556     pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1557     pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1558     pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1559     pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1560     pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1561     pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1562     pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1563 }
1564 
1565 static int pci_save_pcix_state(struct pci_dev *dev)
1566 {
1567     int pos;
1568     struct pci_cap_saved_state *save_state;
1569 
1570     pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1571     if (!pos)
1572         return 0;
1573 
1574     save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1575     if (!save_state) {
1576         pci_err(dev, "buffer not found in %s\n", __func__);
1577         return -ENOMEM;
1578     }
1579 
1580     pci_read_config_word(dev, pos + PCI_X_CMD,
1581                  (u16 *)save_state->cap.data);
1582 
1583     return 0;
1584 }
1585 
1586 static void pci_restore_pcix_state(struct pci_dev *dev)
1587 {
1588     int i = 0, pos;
1589     struct pci_cap_saved_state *save_state;
1590     u16 *cap;
1591 
1592     save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1593     pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1594     if (!save_state || !pos)
1595         return;
1596     cap = (u16 *)&save_state->cap.data[0];
1597 
1598     pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1599 }
1600 
1601 static void pci_save_ltr_state(struct pci_dev *dev)
1602 {
1603     int ltr;
1604     struct pci_cap_saved_state *save_state;
1605     u32 *cap;
1606 
1607     if (!pci_is_pcie(dev))
1608         return;
1609 
1610     ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1611     if (!ltr)
1612         return;
1613 
1614     save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1615     if (!save_state) {
1616         pci_err(dev, "no suspend buffer for LTR; ASPM issues possible after resume\n");
1617         return;
1618     }
1619 
1620     /* Some broken devices only support dword access to LTR */
1621     cap = &save_state->cap.data[0];
1622     pci_read_config_dword(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, cap);
1623 }
1624 
1625 static void pci_restore_ltr_state(struct pci_dev *dev)
1626 {
1627     struct pci_cap_saved_state *save_state;
1628     int ltr;
1629     u32 *cap;
1630 
1631     save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1632     ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1633     if (!save_state || !ltr)
1634         return;
1635 
1636     /* Some broken devices only support dword access to LTR */
1637     cap = &save_state->cap.data[0];
1638     pci_write_config_dword(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, *cap);
1639 }
1640 
1641 /**
1642  * pci_save_state - save the PCI configuration space of a device before
1643  *          suspending
1644  * @dev: PCI device that we're dealing with
1645  */
1646 int pci_save_state(struct pci_dev *dev)
1647 {
1648     int i;
1649     /* XXX: 100% dword access ok here? */
1650     for (i = 0; i < 16; i++) {
1651         pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1652         pci_dbg(dev, "saving config space at offset %#x (reading %#x)\n",
1653             i * 4, dev->saved_config_space[i]);
1654     }
1655     dev->state_saved = true;
1656 
1657     i = pci_save_pcie_state(dev);
1658     if (i != 0)
1659         return i;
1660 
1661     i = pci_save_pcix_state(dev);
1662     if (i != 0)
1663         return i;
1664 
1665     pci_save_ltr_state(dev);
1666     pci_save_dpc_state(dev);
1667     pci_save_aer_state(dev);
1668     pci_save_ptm_state(dev);
1669     return pci_save_vc_state(dev);
1670 }
1671 EXPORT_SYMBOL(pci_save_state);
1672 
1673 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1674                      u32 saved_val, int retry, bool force)
1675 {
1676     u32 val;
1677 
1678     pci_read_config_dword(pdev, offset, &val);
1679     if (!force && val == saved_val)
1680         return;
1681 
1682     for (;;) {
1683         pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1684             offset, val, saved_val);
1685         pci_write_config_dword(pdev, offset, saved_val);
1686         if (retry-- <= 0)
1687             return;
1688 
1689         pci_read_config_dword(pdev, offset, &val);
1690         if (val == saved_val)
1691             return;
1692 
1693         mdelay(1);
1694     }
1695 }
1696 
1697 static void pci_restore_config_space_range(struct pci_dev *pdev,
1698                        int start, int end, int retry,
1699                        bool force)
1700 {
1701     int index;
1702 
1703     for (index = end; index >= start; index--)
1704         pci_restore_config_dword(pdev, 4 * index,
1705                      pdev->saved_config_space[index],
1706                      retry, force);
1707 }
1708 
1709 static void pci_restore_config_space(struct pci_dev *pdev)
1710 {
1711     if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1712         pci_restore_config_space_range(pdev, 10, 15, 0, false);
1713         /* Restore BARs before the command register. */
1714         pci_restore_config_space_range(pdev, 4, 9, 10, false);
1715         pci_restore_config_space_range(pdev, 0, 3, 0, false);
1716     } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1717         pci_restore_config_space_range(pdev, 12, 15, 0, false);
1718 
1719         /*
1720          * Force rewriting of prefetch registers to avoid S3 resume
1721          * issues on Intel PCI bridges that occur when these
1722          * registers are not explicitly written.
1723          */
1724         pci_restore_config_space_range(pdev, 9, 11, 0, true);
1725         pci_restore_config_space_range(pdev, 0, 8, 0, false);
1726     } else {
1727         pci_restore_config_space_range(pdev, 0, 15, 0, false);
1728     }
1729 }
1730 
1731 static void pci_restore_rebar_state(struct pci_dev *pdev)
1732 {
1733     unsigned int pos, nbars, i;
1734     u32 ctrl;
1735 
1736     pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1737     if (!pos)
1738         return;
1739 
1740     pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1741     nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1742             PCI_REBAR_CTRL_NBAR_SHIFT;
1743 
1744     for (i = 0; i < nbars; i++, pos += 8) {
1745         struct resource *res;
1746         int bar_idx, size;
1747 
1748         pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1749         bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1750         res = pdev->resource + bar_idx;
1751         size = pci_rebar_bytes_to_size(resource_size(res));
1752         ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1753         ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1754         pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1755     }
1756 }
1757 
1758 /**
1759  * pci_restore_state - Restore the saved state of a PCI device
1760  * @dev: PCI device that we're dealing with
1761  */
1762 void pci_restore_state(struct pci_dev *dev)
1763 {
1764     if (!dev->state_saved)
1765         return;
1766 
1767     /*
1768      * Restore max latencies (in the LTR capability) before enabling
1769      * LTR itself (in the PCIe capability).
1770      */
1771     pci_restore_ltr_state(dev);
1772 
1773     pci_restore_pcie_state(dev);
1774     pci_restore_pasid_state(dev);
1775     pci_restore_pri_state(dev);
1776     pci_restore_ats_state(dev);
1777     pci_restore_vc_state(dev);
1778     pci_restore_rebar_state(dev);
1779     pci_restore_dpc_state(dev);
1780     pci_restore_ptm_state(dev);
1781 
1782     pci_aer_clear_status(dev);
1783     pci_restore_aer_state(dev);
1784 
1785     pci_restore_config_space(dev);
1786 
1787     pci_restore_pcix_state(dev);
1788     pci_restore_msi_state(dev);
1789 
1790     /* Restore ACS and IOV configuration state */
1791     pci_enable_acs(dev);
1792     pci_restore_iov_state(dev);
1793 
1794     dev->state_saved = false;
1795 }
1796 EXPORT_SYMBOL(pci_restore_state);
1797 
1798 struct pci_saved_state {
1799     u32 config_space[16];
1800     struct pci_cap_saved_data cap[];
1801 };
1802 
1803 /**
1804  * pci_store_saved_state - Allocate and return an opaque struct containing
1805  *             the device saved state.
1806  * @dev: PCI device that we're dealing with
1807  *
1808  * Return NULL if no state or error.
1809  */
1810 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1811 {
1812     struct pci_saved_state *state;
1813     struct pci_cap_saved_state *tmp;
1814     struct pci_cap_saved_data *cap;
1815     size_t size;
1816 
1817     if (!dev->state_saved)
1818         return NULL;
1819 
1820     size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1821 
1822     hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1823         size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1824 
1825     state = kzalloc(size, GFP_KERNEL);
1826     if (!state)
1827         return NULL;
1828 
1829     memcpy(state->config_space, dev->saved_config_space,
1830            sizeof(state->config_space));
1831 
1832     cap = state->cap;
1833     hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1834         size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1835         memcpy(cap, &tmp->cap, len);
1836         cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1837     }
1838     /* Empty cap_save terminates list */
1839 
1840     return state;
1841 }
1842 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1843 
1844 /**
1845  * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1846  * @dev: PCI device that we're dealing with
1847  * @state: Saved state returned from pci_store_saved_state()
1848  */
1849 int pci_load_saved_state(struct pci_dev *dev,
1850              struct pci_saved_state *state)
1851 {
1852     struct pci_cap_saved_data *cap;
1853 
1854     dev->state_saved = false;
1855 
1856     if (!state)
1857         return 0;
1858 
1859     memcpy(dev->saved_config_space, state->config_space,
1860            sizeof(state->config_space));
1861 
1862     cap = state->cap;
1863     while (cap->size) {
1864         struct pci_cap_saved_state *tmp;
1865 
1866         tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1867         if (!tmp || tmp->cap.size != cap->size)
1868             return -EINVAL;
1869 
1870         memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1871         cap = (struct pci_cap_saved_data *)((u8 *)cap +
1872                sizeof(struct pci_cap_saved_data) + cap->size);
1873     }
1874 
1875     dev->state_saved = true;
1876     return 0;
1877 }
1878 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1879 
1880 /**
1881  * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1882  *                 and free the memory allocated for it.
1883  * @dev: PCI device that we're dealing with
1884  * @state: Pointer to saved state returned from pci_store_saved_state()
1885  */
1886 int pci_load_and_free_saved_state(struct pci_dev *dev,
1887                   struct pci_saved_state **state)
1888 {
1889     int ret = pci_load_saved_state(dev, *state);
1890     kfree(*state);
1891     *state = NULL;
1892     return ret;
1893 }
1894 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1895 
1896 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1897 {
1898     return pci_enable_resources(dev, bars);
1899 }
1900 
1901 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1902 {
1903     int err;
1904     struct pci_dev *bridge;
1905     u16 cmd;
1906     u8 pin;
1907 
1908     err = pci_set_power_state(dev, PCI_D0);
1909     if (err < 0 && err != -EIO)
1910         return err;
1911 
1912     bridge = pci_upstream_bridge(dev);
1913     if (bridge)
1914         pcie_aspm_powersave_config_link(bridge);
1915 
1916     err = pcibios_enable_device(dev, bars);
1917     if (err < 0)
1918         return err;
1919     pci_fixup_device(pci_fixup_enable, dev);
1920 
1921     if (dev->msi_enabled || dev->msix_enabled)
1922         return 0;
1923 
1924     pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1925     if (pin) {
1926         pci_read_config_word(dev, PCI_COMMAND, &cmd);
1927         if (cmd & PCI_COMMAND_INTX_DISABLE)
1928             pci_write_config_word(dev, PCI_COMMAND,
1929                           cmd & ~PCI_COMMAND_INTX_DISABLE);
1930     }
1931 
1932     return 0;
1933 }
1934 
1935 /**
1936  * pci_reenable_device - Resume abandoned device
1937  * @dev: PCI device to be resumed
1938  *
1939  * NOTE: This function is a backend of pci_default_resume() and is not supposed
1940  * to be called by normal code, write proper resume handler and use it instead.
1941  */
1942 int pci_reenable_device(struct pci_dev *dev)
1943 {
1944     if (pci_is_enabled(dev))
1945         return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1946     return 0;
1947 }
1948 EXPORT_SYMBOL(pci_reenable_device);
1949 
1950 static void pci_enable_bridge(struct pci_dev *dev)
1951 {
1952     struct pci_dev *bridge;
1953     int retval;
1954 
1955     bridge = pci_upstream_bridge(dev);
1956     if (bridge)
1957         pci_enable_bridge(bridge);
1958 
1959     if (pci_is_enabled(dev)) {
1960         if (!dev->is_busmaster)
1961             pci_set_master(dev);
1962         return;
1963     }
1964 
1965     retval = pci_enable_device(dev);
1966     if (retval)
1967         pci_err(dev, "Error enabling bridge (%d), continuing\n",
1968             retval);
1969     pci_set_master(dev);
1970 }
1971 
1972 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1973 {
1974     struct pci_dev *bridge;
1975     int err;
1976     int i, bars = 0;
1977 
1978     /*
1979      * Power state could be unknown at this point, either due to a fresh
1980      * boot or a device removal call.  So get the current power state
1981      * so that things like MSI message writing will behave as expected
1982      * (e.g. if the device really is in D0 at enable time).
1983      */
1984     pci_update_current_state(dev, dev->current_state);
1985 
1986     if (atomic_inc_return(&dev->enable_cnt) > 1)
1987         return 0;       /* already enabled */
1988 
1989     bridge = pci_upstream_bridge(dev);
1990     if (bridge)
1991         pci_enable_bridge(bridge);
1992 
1993     /* only skip sriov related */
1994     for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1995         if (dev->resource[i].flags & flags)
1996             bars |= (1 << i);
1997     for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1998         if (dev->resource[i].flags & flags)
1999             bars |= (1 << i);
2000 
2001     err = do_pci_enable_device(dev, bars);
2002     if (err < 0)
2003         atomic_dec(&dev->enable_cnt);
2004     return err;
2005 }
2006 
2007 /**
2008  * pci_enable_device_io - Initialize a device for use with IO space
2009  * @dev: PCI device to be initialized
2010  *
2011  * Initialize device before it's used by a driver. Ask low-level code
2012  * to enable I/O resources. Wake up the device if it was suspended.
2013  * Beware, this function can fail.
2014  */
2015 int pci_enable_device_io(struct pci_dev *dev)
2016 {
2017     return pci_enable_device_flags(dev, IORESOURCE_IO);
2018 }
2019 EXPORT_SYMBOL(pci_enable_device_io);
2020 
2021 /**
2022  * pci_enable_device_mem - Initialize a device for use with Memory space
2023  * @dev: PCI device to be initialized
2024  *
2025  * Initialize device before it's used by a driver. Ask low-level code
2026  * to enable Memory resources. Wake up the device if it was suspended.
2027  * Beware, this function can fail.
2028  */
2029 int pci_enable_device_mem(struct pci_dev *dev)
2030 {
2031     return pci_enable_device_flags(dev, IORESOURCE_MEM);
2032 }
2033 EXPORT_SYMBOL(pci_enable_device_mem);
2034 
2035 /**
2036  * pci_enable_device - Initialize device before it's used by a driver.
2037  * @dev: PCI device to be initialized
2038  *
2039  * Initialize device before it's used by a driver. Ask low-level code
2040  * to enable I/O and memory. Wake up the device if it was suspended.
2041  * Beware, this function can fail.
2042  *
2043  * Note we don't actually enable the device many times if we call
2044  * this function repeatedly (we just increment the count).
2045  */
2046 int pci_enable_device(struct pci_dev *dev)
2047 {
2048     return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
2049 }
2050 EXPORT_SYMBOL(pci_enable_device);
2051 
2052 /*
2053  * Managed PCI resources.  This manages device on/off, INTx/MSI/MSI-X
2054  * on/off and BAR regions.  pci_dev itself records MSI/MSI-X status, so
2055  * there's no need to track it separately.  pci_devres is initialized
2056  * when a device is enabled using managed PCI device enable interface.
2057  */
2058 struct pci_devres {
2059     unsigned int enabled:1;
2060     unsigned int pinned:1;
2061     unsigned int orig_intx:1;
2062     unsigned int restore_intx:1;
2063     unsigned int mwi:1;
2064     u32 region_mask;
2065 };
2066 
2067 static void pcim_release(struct device *gendev, void *res)
2068 {
2069     struct pci_dev *dev = to_pci_dev(gendev);
2070     struct pci_devres *this = res;
2071     int i;
2072 
2073     for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
2074         if (this->region_mask & (1 << i))
2075             pci_release_region(dev, i);
2076 
2077     if (this->mwi)
2078         pci_clear_mwi(dev);
2079 
2080     if (this->restore_intx)
2081         pci_intx(dev, this->orig_intx);
2082 
2083     if (this->enabled && !this->pinned)
2084         pci_disable_device(dev);
2085 }
2086 
2087 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
2088 {
2089     struct pci_devres *dr, *new_dr;
2090 
2091     dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
2092     if (dr)
2093         return dr;
2094 
2095     new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
2096     if (!new_dr)
2097         return NULL;
2098     return devres_get(&pdev->dev, new_dr, NULL, NULL);
2099 }
2100 
2101 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
2102 {
2103     if (pci_is_managed(pdev))
2104         return devres_find(&pdev->dev, pcim_release, NULL, NULL);
2105     return NULL;
2106 }
2107 
2108 /**
2109  * pcim_enable_device - Managed pci_enable_device()
2110  * @pdev: PCI device to be initialized
2111  *
2112  * Managed pci_enable_device().
2113  */
2114 int pcim_enable_device(struct pci_dev *pdev)
2115 {
2116     struct pci_devres *dr;
2117     int rc;
2118 
2119     dr = get_pci_dr(pdev);
2120     if (unlikely(!dr))
2121         return -ENOMEM;
2122     if (dr->enabled)
2123         return 0;
2124 
2125     rc = pci_enable_device(pdev);
2126     if (!rc) {
2127         pdev->is_managed = 1;
2128         dr->enabled = 1;
2129     }
2130     return rc;
2131 }
2132 EXPORT_SYMBOL(pcim_enable_device);
2133 
2134 /**
2135  * pcim_pin_device - Pin managed PCI device
2136  * @pdev: PCI device to pin
2137  *
2138  * Pin managed PCI device @pdev.  Pinned device won't be disabled on
2139  * driver detach.  @pdev must have been enabled with
2140  * pcim_enable_device().
2141  */
2142 void pcim_pin_device(struct pci_dev *pdev)
2143 {
2144     struct pci_devres *dr;
2145 
2146     dr = find_pci_dr(pdev);
2147     WARN_ON(!dr || !dr->enabled);
2148     if (dr)
2149         dr->pinned = 1;
2150 }
2151 EXPORT_SYMBOL(pcim_pin_device);
2152 
2153 /*
2154  * pcibios_device_add - provide arch specific hooks when adding device dev
2155  * @dev: the PCI device being added
2156  *
2157  * Permits the platform to provide architecture specific functionality when
2158  * devices are added. This is the default implementation. Architecture
2159  * implementations can override this.
2160  */
2161 int __weak pcibios_device_add(struct pci_dev *dev)
2162 {
2163     return 0;
2164 }
2165 
2166 /**
2167  * pcibios_release_device - provide arch specific hooks when releasing
2168  *              device dev
2169  * @dev: the PCI device being released
2170  *
2171  * Permits the platform to provide architecture specific functionality when
2172  * devices are released. This is the default implementation. Architecture
2173  * implementations can override this.
2174  */
2175 void __weak pcibios_release_device(struct pci_dev *dev) {}
2176 
2177 /**
2178  * pcibios_disable_device - disable arch specific PCI resources for device dev
2179  * @dev: the PCI device to disable
2180  *
2181  * Disables architecture specific PCI resources for the device. This
2182  * is the default implementation. Architecture implementations can
2183  * override this.
2184  */
2185 void __weak pcibios_disable_device(struct pci_dev *dev) {}
2186 
2187 /**
2188  * pcibios_penalize_isa_irq - penalize an ISA IRQ
2189  * @irq: ISA IRQ to penalize
2190  * @active: IRQ active or not
2191  *
2192  * Permits the platform to provide architecture-specific functionality when
2193  * penalizing ISA IRQs. This is the default implementation. Architecture
2194  * implementations can override this.
2195  */
2196 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
2197 
2198 static void do_pci_disable_device(struct pci_dev *dev)
2199 {
2200     u16 pci_command;
2201 
2202     pci_read_config_word(dev, PCI_COMMAND, &pci_command);
2203     if (pci_command & PCI_COMMAND_MASTER) {
2204         pci_command &= ~PCI_COMMAND_MASTER;
2205         pci_write_config_word(dev, PCI_COMMAND, pci_command);
2206     }
2207 
2208     pcibios_disable_device(dev);
2209 }
2210 
2211 /**
2212  * pci_disable_enabled_device - Disable device without updating enable_cnt
2213  * @dev: PCI device to disable
2214  *
2215  * NOTE: This function is a backend of PCI power management routines and is
2216  * not supposed to be called drivers.
2217  */
2218 void pci_disable_enabled_device(struct pci_dev *dev)
2219 {
2220     if (pci_is_enabled(dev))
2221         do_pci_disable_device(dev);
2222 }
2223 
2224 /**
2225  * pci_disable_device - Disable PCI device after use
2226  * @dev: PCI device to be disabled
2227  *
2228  * Signal to the system that the PCI device is not in use by the system
2229  * anymore.  This only involves disabling PCI bus-mastering, if active.
2230  *
2231  * Note we don't actually disable the device until all callers of
2232  * pci_enable_device() have called pci_disable_device().
2233  */
2234 void pci_disable_device(struct pci_dev *dev)
2235 {
2236     struct pci_devres *dr;
2237 
2238     dr = find_pci_dr(dev);
2239     if (dr)
2240         dr->enabled = 0;
2241 
2242     dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
2243               "disabling already-disabled device");
2244 
2245     if (atomic_dec_return(&dev->enable_cnt) != 0)
2246         return;
2247 
2248     do_pci_disable_device(dev);
2249 
2250     dev->is_busmaster = 0;
2251 }
2252 EXPORT_SYMBOL(pci_disable_device);
2253 
2254 /**
2255  * pcibios_set_pcie_reset_state - set reset state for device dev
2256  * @dev: the PCIe device reset
2257  * @state: Reset state to enter into
2258  *
2259  * Set the PCIe reset state for the device. This is the default
2260  * implementation. Architecture implementations can override this.
2261  */
2262 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
2263                     enum pcie_reset_state state)
2264 {
2265     return -EINVAL;
2266 }
2267 
2268 /**
2269  * pci_set_pcie_reset_state - set reset state for device dev
2270  * @dev: the PCIe device reset
2271  * @state: Reset state to enter into
2272  *
2273  * Sets the PCI reset state for the device.
2274  */
2275 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
2276 {
2277     return pcibios_set_pcie_reset_state(dev, state);
2278 }
2279 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
2280 
2281 #ifdef CONFIG_PCIEAER
2282 void pcie_clear_device_status(struct pci_dev *dev)
2283 {
2284     u16 sta;
2285 
2286     pcie_capability_read_word(dev, PCI_EXP_DEVSTA, &sta);
2287     pcie_capability_write_word(dev, PCI_EXP_DEVSTA, sta);
2288 }
2289 #endif
2290 
2291 /**
2292  * pcie_clear_root_pme_status - Clear root port PME interrupt status.
2293  * @dev: PCIe root port or event collector.
2294  */
2295 void pcie_clear_root_pme_status(struct pci_dev *dev)
2296 {
2297     pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
2298 }
2299 
2300 /**
2301  * pci_check_pme_status - Check if given device has generated PME.
2302  * @dev: Device to check.
2303  *
2304  * Check the PME status of the device and if set, clear it and clear PME enable
2305  * (if set).  Return 'true' if PME status and PME enable were both set or
2306  * 'false' otherwise.
2307  */
2308 bool pci_check_pme_status(struct pci_dev *dev)
2309 {
2310     int pmcsr_pos;
2311     u16 pmcsr;
2312     bool ret = false;
2313 
2314     if (!dev->pm_cap)
2315         return false;
2316 
2317     pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
2318     pci_read_config_word(dev, pmcsr_pos, &pmcsr);
2319     if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
2320         return false;
2321 
2322     /* Clear PME status. */
2323     pmcsr |= PCI_PM_CTRL_PME_STATUS;
2324     if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
2325         /* Disable PME to avoid interrupt flood. */
2326         pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2327         ret = true;
2328     }
2329 
2330     pci_write_config_word(dev, pmcsr_pos, pmcsr);
2331 
2332     return ret;
2333 }
2334 
2335 /**
2336  * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2337  * @dev: Device to handle.
2338  * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2339  *
2340  * Check if @dev has generated PME and queue a resume request for it in that
2341  * case.
2342  */
2343 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2344 {
2345     if (pme_poll_reset && dev->pme_poll)
2346         dev->pme_poll = false;
2347 
2348     if (pci_check_pme_status(dev)) {
2349         pci_wakeup_event(dev);
2350         pm_request_resume(&dev->dev);
2351     }
2352     return 0;
2353 }
2354 
2355 /**
2356  * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2357  * @bus: Top bus of the subtree to walk.
2358  */
2359 void pci_pme_wakeup_bus(struct pci_bus *bus)
2360 {
2361     if (bus)
2362         pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2363 }
2364 
2365 
2366 /**
2367  * pci_pme_capable - check the capability of PCI device to generate PME#
2368  * @dev: PCI device to handle.
2369  * @state: PCI state from which device will issue PME#.
2370  */
2371 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2372 {
2373     if (!dev->pm_cap)
2374         return false;
2375 
2376     return !!(dev->pme_support & (1 << state));
2377 }
2378 EXPORT_SYMBOL(pci_pme_capable);
2379 
2380 static void pci_pme_list_scan(struct work_struct *work)
2381 {
2382     struct pci_pme_device *pme_dev, *n;
2383 
2384     mutex_lock(&pci_pme_list_mutex);
2385     list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2386         if (pme_dev->dev->pme_poll) {
2387             struct pci_dev *bridge;
2388 
2389             bridge = pme_dev->dev->bus->self;
2390             /*
2391              * If bridge is in low power state, the
2392              * configuration space of subordinate devices
2393              * may be not accessible
2394              */
2395             if (bridge && bridge->current_state != PCI_D0)
2396                 continue;
2397             /*
2398              * If the device is in D3cold it should not be
2399              * polled either.
2400              */
2401             if (pme_dev->dev->current_state == PCI_D3cold)
2402                 continue;
2403 
2404             pci_pme_wakeup(pme_dev->dev, NULL);
2405         } else {
2406             list_del(&pme_dev->list);
2407             kfree(pme_dev);
2408         }
2409     }
2410     if (!list_empty(&pci_pme_list))
2411         queue_delayed_work(system_freezable_wq, &pci_pme_work,
2412                    msecs_to_jiffies(PME_TIMEOUT));
2413     mutex_unlock(&pci_pme_list_mutex);
2414 }
2415 
2416 static void __pci_pme_active(struct pci_dev *dev, bool enable)
2417 {
2418     u16 pmcsr;
2419 
2420     if (!dev->pme_support)
2421         return;
2422 
2423     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2424     /* Clear PME_Status by writing 1 to it and enable PME# */
2425     pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2426     if (!enable)
2427         pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2428 
2429     pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2430 }
2431 
2432 /**
2433  * pci_pme_restore - Restore PME configuration after config space restore.
2434  * @dev: PCI device to update.
2435  */
2436 void pci_pme_restore(struct pci_dev *dev)
2437 {
2438     u16 pmcsr;
2439 
2440     if (!dev->pme_support)
2441         return;
2442 
2443     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2444     if (dev->wakeup_prepared) {
2445         pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2446         pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2447     } else {
2448         pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2449         pmcsr |= PCI_PM_CTRL_PME_STATUS;
2450     }
2451     pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2452 }
2453 
2454 /**
2455  * pci_pme_active - enable or disable PCI device's PME# function
2456  * @dev: PCI device to handle.
2457  * @enable: 'true' to enable PME# generation; 'false' to disable it.
2458  *
2459  * The caller must verify that the device is capable of generating PME# before
2460  * calling this function with @enable equal to 'true'.
2461  */
2462 void pci_pme_active(struct pci_dev *dev, bool enable)
2463 {
2464     __pci_pme_active(dev, enable);
2465 
2466     /*
2467      * PCI (as opposed to PCIe) PME requires that the device have
2468      * its PME# line hooked up correctly. Not all hardware vendors
2469      * do this, so the PME never gets delivered and the device
2470      * remains asleep. The easiest way around this is to
2471      * periodically walk the list of suspended devices and check
2472      * whether any have their PME flag set. The assumption is that
2473      * we'll wake up often enough anyway that this won't be a huge
2474      * hit, and the power savings from the devices will still be a
2475      * win.
2476      *
2477      * Although PCIe uses in-band PME message instead of PME# line
2478      * to report PME, PME does not work for some PCIe devices in
2479      * reality.  For example, there are devices that set their PME
2480      * status bits, but don't really bother to send a PME message;
2481      * there are PCI Express Root Ports that don't bother to
2482      * trigger interrupts when they receive PME messages from the
2483      * devices below.  So PME poll is used for PCIe devices too.
2484      */
2485 
2486     if (dev->pme_poll) {
2487         struct pci_pme_device *pme_dev;
2488         if (enable) {
2489             pme_dev = kmalloc(sizeof(struct pci_pme_device),
2490                       GFP_KERNEL);
2491             if (!pme_dev) {
2492                 pci_warn(dev, "can't enable PME#\n");
2493                 return;
2494             }
2495             pme_dev->dev = dev;
2496             mutex_lock(&pci_pme_list_mutex);
2497             list_add(&pme_dev->list, &pci_pme_list);
2498             if (list_is_singular(&pci_pme_list))
2499                 queue_delayed_work(system_freezable_wq,
2500                            &pci_pme_work,
2501                            msecs_to_jiffies(PME_TIMEOUT));
2502             mutex_unlock(&pci_pme_list_mutex);
2503         } else {
2504             mutex_lock(&pci_pme_list_mutex);
2505             list_for_each_entry(pme_dev, &pci_pme_list, list) {
2506                 if (pme_dev->dev == dev) {
2507                     list_del(&pme_dev->list);
2508                     kfree(pme_dev);
2509                     break;
2510                 }
2511             }
2512             mutex_unlock(&pci_pme_list_mutex);
2513         }
2514     }
2515 
2516     pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2517 }
2518 EXPORT_SYMBOL(pci_pme_active);
2519 
2520 /**
2521  * __pci_enable_wake - enable PCI device as wakeup event source
2522  * @dev: PCI device affected
2523  * @state: PCI state from which device will issue wakeup events
2524  * @enable: True to enable event generation; false to disable
2525  *
2526  * This enables the device as a wakeup event source, or disables it.
2527  * When such events involves platform-specific hooks, those hooks are
2528  * called automatically by this routine.
2529  *
2530  * Devices with legacy power management (no standard PCI PM capabilities)
2531  * always require such platform hooks.
2532  *
2533  * RETURN VALUE:
2534  * 0 is returned on success
2535  * -EINVAL is returned if device is not supposed to wake up the system
2536  * Error code depending on the platform is returned if both the platform and
2537  * the native mechanism fail to enable the generation of wake-up events
2538  */
2539 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2540 {
2541     int ret = 0;
2542 
2543     /*
2544      * Bridges that are not power-manageable directly only signal
2545      * wakeup on behalf of subordinate devices which is set up
2546      * elsewhere, so skip them. However, bridges that are
2547      * power-manageable may signal wakeup for themselves (for example,
2548      * on a hotplug event) and they need to be covered here.
2549      */
2550     if (!pci_power_manageable(dev))
2551         return 0;
2552 
2553     /* Don't do the same thing twice in a row for one device. */
2554     if (!!enable == !!dev->wakeup_prepared)
2555         return 0;
2556 
2557     /*
2558      * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2559      * Anderson we should be doing PME# wake enable followed by ACPI wake
2560      * enable.  To disable wake-up we call the platform first, for symmetry.
2561      */
2562 
2563     if (enable) {
2564         int error;
2565 
2566         /*
2567          * Enable PME signaling if the device can signal PME from
2568          * D3cold regardless of whether or not it can signal PME from
2569          * the current target state, because that will allow it to
2570          * signal PME when the hierarchy above it goes into D3cold and
2571          * the device itself ends up in D3cold as a result of that.
2572          */
2573         if (pci_pme_capable(dev, state) || pci_pme_capable(dev, PCI_D3cold))
2574             pci_pme_active(dev, true);
2575         else
2576             ret = 1;
2577         error = platform_pci_set_wakeup(dev, true);
2578         if (ret)
2579             ret = error;
2580         if (!ret)
2581             dev->wakeup_prepared = true;
2582     } else {
2583         platform_pci_set_wakeup(dev, false);
2584         pci_pme_active(dev, false);
2585         dev->wakeup_prepared = false;
2586     }
2587 
2588     return ret;
2589 }
2590 
2591 /**
2592  * pci_enable_wake - change wakeup settings for a PCI device
2593  * @pci_dev: Target device
2594  * @state: PCI state from which device will issue wakeup events
2595  * @enable: Whether or not to enable event generation
2596  *
2597  * If @enable is set, check device_may_wakeup() for the device before calling
2598  * __pci_enable_wake() for it.
2599  */
2600 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2601 {
2602     if (enable && !device_may_wakeup(&pci_dev->dev))
2603         return -EINVAL;
2604 
2605     return __pci_enable_wake(pci_dev, state, enable);
2606 }
2607 EXPORT_SYMBOL(pci_enable_wake);
2608 
2609 /**
2610  * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2611  * @dev: PCI device to prepare
2612  * @enable: True to enable wake-up event generation; false to disable
2613  *
2614  * Many drivers want the device to wake up the system from D3_hot or D3_cold
2615  * and this function allows them to set that up cleanly - pci_enable_wake()
2616  * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2617  * ordering constraints.
2618  *
2619  * This function only returns error code if the device is not allowed to wake
2620  * up the system from sleep or it is not capable of generating PME# from both
2621  * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2622  */
2623 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2624 {
2625     return pci_pme_capable(dev, PCI_D3cold) ?
2626             pci_enable_wake(dev, PCI_D3cold, enable) :
2627             pci_enable_wake(dev, PCI_D3hot, enable);
2628 }
2629 EXPORT_SYMBOL(pci_wake_from_d3);
2630 
2631 /**
2632  * pci_target_state - find an appropriate low power state for a given PCI dev
2633  * @dev: PCI device
2634  * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2635  *
2636  * Use underlying platform code to find a supported low power state for @dev.
2637  * If the platform can't manage @dev, return the deepest state from which it
2638  * can generate wake events, based on any available PME info.
2639  */
2640 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2641 {
2642     if (platform_pci_power_manageable(dev)) {
2643         /*
2644          * Call the platform to find the target state for the device.
2645          */
2646         pci_power_t state = platform_pci_choose_state(dev);
2647 
2648         switch (state) {
2649         case PCI_POWER_ERROR:
2650         case PCI_UNKNOWN:
2651             return PCI_D3hot;
2652 
2653         case PCI_D1:
2654         case PCI_D2:
2655             if (pci_no_d1d2(dev))
2656                 return PCI_D3hot;
2657         }
2658 
2659         return state;
2660     }
2661 
2662     /*
2663      * If the device is in D3cold even though it's not power-manageable by
2664      * the platform, it may have been powered down by non-standard means.
2665      * Best to let it slumber.
2666      */
2667     if (dev->current_state == PCI_D3cold)
2668         return PCI_D3cold;
2669     else if (!dev->pm_cap)
2670         return PCI_D0;
2671 
2672     if (wakeup && dev->pme_support) {
2673         pci_power_t state = PCI_D3hot;
2674 
2675         /*
2676          * Find the deepest state from which the device can generate
2677          * PME#.
2678          */
2679         while (state && !(dev->pme_support & (1 << state)))
2680             state--;
2681 
2682         if (state)
2683             return state;
2684         else if (dev->pme_support & 1)
2685             return PCI_D0;
2686     }
2687 
2688     return PCI_D3hot;
2689 }
2690 
2691 /**
2692  * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2693  *            into a sleep state
2694  * @dev: Device to handle.
2695  *
2696  * Choose the power state appropriate for the device depending on whether
2697  * it can wake up the system and/or is power manageable by the platform
2698  * (PCI_D3hot is the default) and put the device into that state.
2699  */
2700 int pci_prepare_to_sleep(struct pci_dev *dev)
2701 {
2702     bool wakeup = device_may_wakeup(&dev->dev);
2703     pci_power_t target_state = pci_target_state(dev, wakeup);
2704     int error;
2705 
2706     if (target_state == PCI_POWER_ERROR)
2707         return -EIO;
2708 
2709     /*
2710      * There are systems (for example, Intel mobile chips since Coffee
2711      * Lake) where the power drawn while suspended can be significantly
2712      * reduced by disabling PTM on PCIe root ports as this allows the
2713      * port to enter a lower-power PM state and the SoC to reach a
2714      * lower-power idle state as a whole.
2715      */
2716     if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
2717         pci_disable_ptm(dev);
2718 
2719     pci_enable_wake(dev, target_state, wakeup);
2720 
2721     error = pci_set_power_state(dev, target_state);
2722 
2723     if (error) {
2724         pci_enable_wake(dev, target_state, false);
2725         pci_restore_ptm_state(dev);
2726     }
2727 
2728     return error;
2729 }
2730 EXPORT_SYMBOL(pci_prepare_to_sleep);
2731 
2732 /**
2733  * pci_back_from_sleep - turn PCI device on during system-wide transition
2734  *           into working state
2735  * @dev: Device to handle.
2736  *
2737  * Disable device's system wake-up capability and put it into D0.
2738  */
2739 int pci_back_from_sleep(struct pci_dev *dev)
2740 {
2741     int ret = pci_set_power_state(dev, PCI_D0);
2742 
2743     if (ret)
2744         return ret;
2745 
2746     pci_enable_wake(dev, PCI_D0, false);
2747     return 0;
2748 }
2749 EXPORT_SYMBOL(pci_back_from_sleep);
2750 
2751 /**
2752  * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2753  * @dev: PCI device being suspended.
2754  *
2755  * Prepare @dev to generate wake-up events at run time and put it into a low
2756  * power state.
2757  */
2758 int pci_finish_runtime_suspend(struct pci_dev *dev)
2759 {
2760     pci_power_t target_state;
2761     int error;
2762 
2763     target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2764     if (target_state == PCI_POWER_ERROR)
2765         return -EIO;
2766 
2767     /*
2768      * There are systems (for example, Intel mobile chips since Coffee
2769      * Lake) where the power drawn while suspended can be significantly
2770      * reduced by disabling PTM on PCIe root ports as this allows the
2771      * port to enter a lower-power PM state and the SoC to reach a
2772      * lower-power idle state as a whole.
2773      */
2774     if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
2775         pci_disable_ptm(dev);
2776 
2777     __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2778 
2779     error = pci_set_power_state(dev, target_state);
2780 
2781     if (error) {
2782         pci_enable_wake(dev, target_state, false);
2783         pci_restore_ptm_state(dev);
2784     }
2785 
2786     return error;
2787 }
2788 
2789 /**
2790  * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2791  * @dev: Device to check.
2792  *
2793  * Return true if the device itself is capable of generating wake-up events
2794  * (through the platform or using the native PCIe PME) or if the device supports
2795  * PME and one of its upstream bridges can generate wake-up events.
2796  */
2797 bool pci_dev_run_wake(struct pci_dev *dev)
2798 {
2799     struct pci_bus *bus = dev->bus;
2800 
2801     if (!dev->pme_support)
2802         return false;
2803 
2804     /* PME-capable in principle, but not from the target power state */
2805     if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2806         return false;
2807 
2808     if (device_can_wakeup(&dev->dev))
2809         return true;
2810 
2811     while (bus->parent) {
2812         struct pci_dev *bridge = bus->self;
2813 
2814         if (device_can_wakeup(&bridge->dev))
2815             return true;
2816 
2817         bus = bus->parent;
2818     }
2819 
2820     /* We have reached the root bus. */
2821     if (bus->bridge)
2822         return device_can_wakeup(bus->bridge);
2823 
2824     return false;
2825 }
2826 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2827 
2828 /**
2829  * pci_dev_need_resume - Check if it is necessary to resume the device.
2830  * @pci_dev: Device to check.
2831  *
2832  * Return 'true' if the device is not runtime-suspended or it has to be
2833  * reconfigured due to wakeup settings difference between system and runtime
2834  * suspend, or the current power state of it is not suitable for the upcoming
2835  * (system-wide) transition.
2836  */
2837 bool pci_dev_need_resume(struct pci_dev *pci_dev)
2838 {
2839     struct device *dev = &pci_dev->dev;
2840     pci_power_t target_state;
2841 
2842     if (!pm_runtime_suspended(dev) || platform_pci_need_resume(pci_dev))
2843         return true;
2844 
2845     target_state = pci_target_state(pci_dev, device_may_wakeup(dev));
2846 
2847     /*
2848      * If the earlier platform check has not triggered, D3cold is just power
2849      * removal on top of D3hot, so no need to resume the device in that
2850      * case.
2851      */
2852     return target_state != pci_dev->current_state &&
2853         target_state != PCI_D3cold &&
2854         pci_dev->current_state != PCI_D3hot;
2855 }
2856 
2857 /**
2858  * pci_dev_adjust_pme - Adjust PME setting for a suspended device.
2859  * @pci_dev: Device to check.
2860  *
2861  * If the device is suspended and it is not configured for system wakeup,
2862  * disable PME for it to prevent it from waking up the system unnecessarily.
2863  *
2864  * Note that if the device's power state is D3cold and the platform check in
2865  * pci_dev_need_resume() has not triggered, the device's configuration need not
2866  * be changed.
2867  */
2868 void pci_dev_adjust_pme(struct pci_dev *pci_dev)
2869 {
2870     struct device *dev = &pci_dev->dev;
2871 
2872     spin_lock_irq(&dev->power.lock);
2873 
2874     if (pm_runtime_suspended(dev) && !device_may_wakeup(dev) &&
2875         pci_dev->current_state < PCI_D3cold)
2876         __pci_pme_active(pci_dev, false);
2877 
2878     spin_unlock_irq(&dev->power.lock);
2879 }
2880 
2881 /**
2882  * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2883  * @pci_dev: Device to handle.
2884  *
2885  * If the device is runtime suspended and wakeup-capable, enable PME for it as
2886  * it might have been disabled during the prepare phase of system suspend if
2887  * the device was not configured for system wakeup.
2888  */
2889 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2890 {
2891     struct device *dev = &pci_dev->dev;
2892 
2893     if (!pci_dev_run_wake(pci_dev))
2894         return;
2895 
2896     spin_lock_irq(&dev->power.lock);
2897 
2898     if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2899         __pci_pme_active(pci_dev, true);
2900 
2901     spin_unlock_irq(&dev->power.lock);
2902 }
2903 
2904 /**
2905  * pci_choose_state - Choose the power state of a PCI device.
2906  * @dev: Target PCI device.
2907  * @state: Target state for the whole system.
2908  *
2909  * Returns PCI power state suitable for @dev and @state.
2910  */
2911 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
2912 {
2913     if (state.event == PM_EVENT_ON)
2914         return PCI_D0;
2915 
2916     return pci_target_state(dev, false);
2917 }
2918 EXPORT_SYMBOL(pci_choose_state);
2919 
2920 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2921 {
2922     struct device *dev = &pdev->dev;
2923     struct device *parent = dev->parent;
2924 
2925     if (parent)
2926         pm_runtime_get_sync(parent);
2927     pm_runtime_get_noresume(dev);
2928     /*
2929      * pdev->current_state is set to PCI_D3cold during suspending,
2930      * so wait until suspending completes
2931      */
2932     pm_runtime_barrier(dev);
2933     /*
2934      * Only need to resume devices in D3cold, because config
2935      * registers are still accessible for devices suspended but
2936      * not in D3cold.
2937      */
2938     if (pdev->current_state == PCI_D3cold)
2939         pm_runtime_resume(dev);
2940 }
2941 
2942 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2943 {
2944     struct device *dev = &pdev->dev;
2945     struct device *parent = dev->parent;
2946 
2947     pm_runtime_put(dev);
2948     if (parent)
2949         pm_runtime_put_sync(parent);
2950 }
2951 
2952 static const struct dmi_system_id bridge_d3_blacklist[] = {
2953 #ifdef CONFIG_X86
2954     {
2955         /*
2956          * Gigabyte X299 root port is not marked as hotplug capable
2957          * which allows Linux to power manage it.  However, this
2958          * confuses the BIOS SMI handler so don't power manage root
2959          * ports on that system.
2960          */
2961         .ident = "X299 DESIGNARE EX-CF",
2962         .matches = {
2963             DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2964             DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2965         },
2966     },
2967     {
2968         /*
2969          * Downstream device is not accessible after putting a root port
2970          * into D3cold and back into D0 on Elo i2.
2971          */
2972         .ident = "Elo i2",
2973         .matches = {
2974             DMI_MATCH(DMI_SYS_VENDOR, "Elo Touch Solutions"),
2975             DMI_MATCH(DMI_PRODUCT_NAME, "Elo i2"),
2976             DMI_MATCH(DMI_PRODUCT_VERSION, "RevB"),
2977         },
2978     },
2979 #endif
2980     { }
2981 };
2982 
2983 /**
2984  * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2985  * @bridge: Bridge to check
2986  *
2987  * This function checks if it is possible to move the bridge to D3.
2988  * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2989  */
2990 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2991 {
2992     if (!pci_is_pcie(bridge))
2993         return false;
2994 
2995     switch (pci_pcie_type(bridge)) {
2996     case PCI_EXP_TYPE_ROOT_PORT:
2997     case PCI_EXP_TYPE_UPSTREAM:
2998     case PCI_EXP_TYPE_DOWNSTREAM:
2999         if (pci_bridge_d3_disable)
3000             return false;
3001 
3002         /*
3003          * Hotplug ports handled by firmware in System Management Mode
3004          * may not be put into D3 by the OS (Thunderbolt on non-Macs).
3005          */
3006         if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
3007             return false;
3008 
3009         if (pci_bridge_d3_force)
3010             return true;
3011 
3012         /* Even the oldest 2010 Thunderbolt controller supports D3. */
3013         if (bridge->is_thunderbolt)
3014             return true;
3015 
3016         /* Platform might know better if the bridge supports D3 */
3017         if (platform_pci_bridge_d3(bridge))
3018             return true;
3019 
3020         /*
3021          * Hotplug ports handled natively by the OS were not validated
3022          * by vendors for runtime D3 at least until 2018 because there
3023          * was no OS support.
3024          */
3025         if (bridge->is_hotplug_bridge)
3026             return false;
3027 
3028         if (dmi_check_system(bridge_d3_blacklist))
3029             return false;
3030 
3031         /*
3032          * It should be safe to put PCIe ports from 2015 or newer
3033          * to D3.
3034          */
3035         if (dmi_get_bios_year() >= 2015)
3036             return true;
3037         break;
3038     }
3039 
3040     return false;
3041 }
3042 
3043 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
3044 {
3045     bool *d3cold_ok = data;
3046 
3047     if (/* The device needs to be allowed to go D3cold ... */
3048         dev->no_d3cold || !dev->d3cold_allowed ||
3049 
3050         /* ... and if it is wakeup capable to do so from D3cold. */
3051         (device_may_wakeup(&dev->dev) &&
3052          !pci_pme_capable(dev, PCI_D3cold)) ||
3053 
3054         /* If it is a bridge it must be allowed to go to D3. */
3055         !pci_power_manageable(dev))
3056 
3057         *d3cold_ok = false;
3058 
3059     return !*d3cold_ok;
3060 }
3061 
3062 /*
3063  * pci_bridge_d3_update - Update bridge D3 capabilities
3064  * @dev: PCI device which is changed
3065  *
3066  * Update upstream bridge PM capabilities accordingly depending on if the
3067  * device PM configuration was changed or the device is being removed.  The
3068  * change is also propagated upstream.
3069  */
3070 void pci_bridge_d3_update(struct pci_dev *dev)
3071 {
3072     bool remove = !device_is_registered(&dev->dev);
3073     struct pci_dev *bridge;
3074     bool d3cold_ok = true;
3075 
3076     bridge = pci_upstream_bridge(dev);
3077     if (!bridge || !pci_bridge_d3_possible(bridge))
3078         return;
3079 
3080     /*
3081      * If D3 is currently allowed for the bridge, removing one of its
3082      * children won't change that.
3083      */
3084     if (remove && bridge->bridge_d3)
3085         return;
3086 
3087     /*
3088      * If D3 is currently allowed for the bridge and a child is added or
3089      * changed, disallowance of D3 can only be caused by that child, so
3090      * we only need to check that single device, not any of its siblings.
3091      *
3092      * If D3 is currently not allowed for the bridge, checking the device
3093      * first may allow us to skip checking its siblings.
3094      */
3095     if (!remove)
3096         pci_dev_check_d3cold(dev, &d3cold_ok);
3097 
3098     /*
3099      * If D3 is currently not allowed for the bridge, this may be caused
3100      * either by the device being changed/removed or any of its siblings,
3101      * so we need to go through all children to find out if one of them
3102      * continues to block D3.
3103      */
3104     if (d3cold_ok && !bridge->bridge_d3)
3105         pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
3106                  &d3cold_ok);
3107 
3108     if (bridge->bridge_d3 != d3cold_ok) {
3109         bridge->bridge_d3 = d3cold_ok;
3110         /* Propagate change to upstream bridges */
3111         pci_bridge_d3_update(bridge);
3112     }
3113 }
3114 
3115 /**
3116  * pci_d3cold_enable - Enable D3cold for device
3117  * @dev: PCI device to handle
3118  *
3119  * This function can be used in drivers to enable D3cold from the device
3120  * they handle.  It also updates upstream PCI bridge PM capabilities
3121  * accordingly.
3122  */
3123 void pci_d3cold_enable(struct pci_dev *dev)
3124 {
3125     if (dev->no_d3cold) {
3126         dev->no_d3cold = false;
3127         pci_bridge_d3_update(dev);
3128     }
3129 }
3130 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
3131 
3132 /**
3133  * pci_d3cold_disable - Disable D3cold for device
3134  * @dev: PCI device to handle
3135  *
3136  * This function can be used in drivers to disable D3cold from the device
3137  * they handle.  It also updates upstream PCI bridge PM capabilities
3138  * accordingly.
3139  */
3140 void pci_d3cold_disable(struct pci_dev *dev)
3141 {
3142     if (!dev->no_d3cold) {
3143         dev->no_d3cold = true;
3144         pci_bridge_d3_update(dev);
3145     }
3146 }
3147 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
3148 
3149 /**
3150  * pci_pm_init - Initialize PM functions of given PCI device
3151  * @dev: PCI device to handle.
3152  */
3153 void pci_pm_init(struct pci_dev *dev)
3154 {
3155     int pm;
3156     u16 status;
3157     u16 pmc;
3158 
3159     pm_runtime_forbid(&dev->dev);
3160     pm_runtime_set_active(&dev->dev);
3161     pm_runtime_enable(&dev->dev);
3162     device_enable_async_suspend(&dev->dev);
3163     dev->wakeup_prepared = false;
3164 
3165     dev->pm_cap = 0;
3166     dev->pme_support = 0;
3167 
3168     /* find PCI PM capability in list */
3169     pm = pci_find_capability(dev, PCI_CAP_ID_PM);
3170     if (!pm)
3171         return;
3172     /* Check device's ability to generate PME# */
3173     pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
3174 
3175     if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
3176         pci_err(dev, "unsupported PM cap regs version (%u)\n",
3177             pmc & PCI_PM_CAP_VER_MASK);
3178         return;
3179     }
3180 
3181     dev->pm_cap = pm;
3182     dev->d3hot_delay = PCI_PM_D3HOT_WAIT;
3183     dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
3184     dev->bridge_d3 = pci_bridge_d3_possible(dev);
3185     dev->d3cold_allowed = true;
3186 
3187     dev->d1_support = false;
3188     dev->d2_support = false;
3189     if (!pci_no_d1d2(dev)) {
3190         if (pmc & PCI_PM_CAP_D1)
3191             dev->d1_support = true;
3192         if (pmc & PCI_PM_CAP_D2)
3193             dev->d2_support = true;
3194 
3195         if (dev->d1_support || dev->d2_support)
3196             pci_info(dev, "supports%s%s\n",
3197                    dev->d1_support ? " D1" : "",
3198                    dev->d2_support ? " D2" : "");
3199     }
3200 
3201     pmc &= PCI_PM_CAP_PME_MASK;
3202     if (pmc) {
3203         pci_info(dev, "PME# supported from%s%s%s%s%s\n",
3204              (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
3205              (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
3206              (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
3207              (pmc & PCI_PM_CAP_PME_D3hot) ? " D3hot" : "",
3208              (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
3209         dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
3210         dev->pme_poll = true;
3211         /*
3212          * Make device's PM flags reflect the wake-up capability, but
3213          * let the user space enable it to wake up the system as needed.
3214          */
3215         device_set_wakeup_capable(&dev->dev, true);
3216         /* Disable the PME# generation functionality */
3217         pci_pme_active(dev, false);
3218     }
3219 
3220     pci_read_config_word(dev, PCI_STATUS, &status);
3221     if (status & PCI_STATUS_IMM_READY)
3222         dev->imm_ready = 1;
3223 }
3224 
3225 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
3226 {
3227     unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
3228 
3229     switch (prop) {
3230     case PCI_EA_P_MEM:
3231     case PCI_EA_P_VF_MEM:
3232         flags |= IORESOURCE_MEM;
3233         break;
3234     case PCI_EA_P_MEM_PREFETCH:
3235     case PCI_EA_P_VF_MEM_PREFETCH:
3236         flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
3237         break;
3238     case PCI_EA_P_IO:
3239         flags |= IORESOURCE_IO;
3240         break;
3241     default:
3242         return 0;
3243     }
3244 
3245     return flags;
3246 }
3247 
3248 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
3249                         u8 prop)
3250 {
3251     if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
3252         return &dev->resource[bei];
3253 #ifdef CONFIG_PCI_IOV
3254     else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
3255          (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
3256         return &dev->resource[PCI_IOV_RESOURCES +
3257                       bei - PCI_EA_BEI_VF_BAR0];
3258 #endif
3259     else if (bei == PCI_EA_BEI_ROM)
3260         return &dev->resource[PCI_ROM_RESOURCE];
3261     else
3262         return NULL;
3263 }
3264 
3265 /* Read an Enhanced Allocation (EA) entry */
3266 static int pci_ea_read(struct pci_dev *dev, int offset)
3267 {
3268     struct resource *res;
3269     int ent_size, ent_offset = offset;
3270     resource_size_t start, end;
3271     unsigned long flags;
3272     u32 dw0, bei, base, max_offset;
3273     u8 prop;
3274     bool support_64 = (sizeof(resource_size_t) >= 8);
3275 
3276     pci_read_config_dword(dev, ent_offset, &dw0);
3277     ent_offset += 4;
3278 
3279     /* Entry size field indicates DWORDs after 1st */
3280     ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
3281 
3282     if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
3283         goto out;
3284 
3285     bei = (dw0 & PCI_EA_BEI) >> 4;
3286     prop = (dw0 & PCI_EA_PP) >> 8;
3287 
3288     /*
3289      * If the Property is in the reserved range, try the Secondary
3290      * Property instead.
3291      */
3292     if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
3293         prop = (dw0 & PCI_EA_SP) >> 16;
3294     if (prop > PCI_EA_P_BRIDGE_IO)
3295         goto out;
3296 
3297     res = pci_ea_get_resource(dev, bei, prop);
3298     if (!res) {
3299         pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
3300         goto out;
3301     }
3302 
3303     flags = pci_ea_flags(dev, prop);
3304     if (!flags) {
3305         pci_err(dev, "Unsupported EA properties: %#x\n", prop);
3306         goto out;
3307     }
3308 
3309     /* Read Base */
3310     pci_read_config_dword(dev, ent_offset, &base);
3311     start = (base & PCI_EA_FIELD_MASK);
3312     ent_offset += 4;
3313 
3314     /* Read MaxOffset */
3315     pci_read_config_dword(dev, ent_offset, &max_offset);
3316     ent_offset += 4;
3317 
3318     /* Read Base MSBs (if 64-bit entry) */
3319     if (base & PCI_EA_IS_64) {
3320         u32 base_upper;
3321 
3322         pci_read_config_dword(dev, ent_offset, &base_upper);
3323         ent_offset += 4;
3324 
3325         flags |= IORESOURCE_MEM_64;
3326 
3327         /* entry starts above 32-bit boundary, can't use */
3328         if (!support_64 && base_upper)
3329             goto out;
3330 
3331         if (support_64)
3332             start |= ((u64)base_upper << 32);
3333     }
3334 
3335     end = start + (max_offset | 0x03);
3336 
3337     /* Read MaxOffset MSBs (if 64-bit entry) */
3338     if (max_offset & PCI_EA_IS_64) {
3339         u32 max_offset_upper;
3340 
3341         pci_read_config_dword(dev, ent_offset, &max_offset_upper);
3342         ent_offset += 4;
3343 
3344         flags |= IORESOURCE_MEM_64;
3345 
3346         /* entry too big, can't use */
3347         if (!support_64 && max_offset_upper)
3348             goto out;
3349 
3350         if (support_64)
3351             end += ((u64)max_offset_upper << 32);
3352     }
3353 
3354     if (end < start) {
3355         pci_err(dev, "EA Entry crosses address boundary\n");
3356         goto out;
3357     }
3358 
3359     if (ent_size != ent_offset - offset) {
3360         pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
3361             ent_size, ent_offset - offset);
3362         goto out;
3363     }
3364 
3365     res->name = pci_name(dev);
3366     res->start = start;
3367     res->end = end;
3368     res->flags = flags;
3369 
3370     if (bei <= PCI_EA_BEI_BAR5)
3371         pci_info(dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3372                bei, res, prop);
3373     else if (bei == PCI_EA_BEI_ROM)
3374         pci_info(dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
3375                res, prop);
3376     else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
3377         pci_info(dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
3378                bei - PCI_EA_BEI_VF_BAR0, res, prop);
3379     else
3380         pci_info(dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
3381                bei, res, prop);
3382 
3383 out:
3384     return offset + ent_size;
3385 }
3386 
3387 /* Enhanced Allocation Initialization */
3388 void pci_ea_init(struct pci_dev *dev)
3389 {
3390     int ea;
3391     u8 num_ent;
3392     int offset;
3393     int i;
3394 
3395     /* find PCI EA capability in list */
3396     ea = pci_find_capability(dev, PCI_CAP_ID_EA);
3397     if (!ea)
3398         return;
3399 
3400     /* determine the number of entries */
3401     pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
3402                     &num_ent);
3403     num_ent &= PCI_EA_NUM_ENT_MASK;
3404 
3405     offset = ea + PCI_EA_FIRST_ENT;
3406 
3407     /* Skip DWORD 2 for type 1 functions */
3408     if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3409         offset += 4;
3410 
3411     /* parse each EA entry */
3412     for (i = 0; i < num_ent; ++i)
3413         offset = pci_ea_read(dev, offset);
3414 }
3415 
3416 static void pci_add_saved_cap(struct pci_dev *pci_dev,
3417     struct pci_cap_saved_state *new_cap)
3418 {
3419     hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3420 }
3421 
3422 /**
3423  * _pci_add_cap_save_buffer - allocate buffer for saving given
3424  *                capability registers
3425  * @dev: the PCI device
3426  * @cap: the capability to allocate the buffer for
3427  * @extended: Standard or Extended capability ID
3428  * @size: requested size of the buffer
3429  */
3430 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3431                     bool extended, unsigned int size)
3432 {
3433     int pos;
3434     struct pci_cap_saved_state *save_state;
3435 
3436     if (extended)
3437         pos = pci_find_ext_capability(dev, cap);
3438     else
3439         pos = pci_find_capability(dev, cap);
3440 
3441     if (!pos)
3442         return 0;
3443 
3444     save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3445     if (!save_state)
3446         return -ENOMEM;
3447 
3448     save_state->cap.cap_nr = cap;
3449     save_state->cap.cap_extended = extended;
3450     save_state->cap.size = size;
3451     pci_add_saved_cap(dev, save_state);
3452 
3453     return 0;
3454 }
3455 
3456 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3457 {
3458     return _pci_add_cap_save_buffer(dev, cap, false, size);
3459 }
3460 
3461 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3462 {
3463     return _pci_add_cap_save_buffer(dev, cap, true, size);
3464 }
3465 
3466 /**
3467  * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3468  * @dev: the PCI device
3469  */
3470 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3471 {
3472     int error;
3473 
3474     error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3475                     PCI_EXP_SAVE_REGS * sizeof(u16));
3476     if (error)
3477         pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3478 
3479     error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3480     if (error)
3481         pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3482 
3483     error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3484                         2 * sizeof(u16));
3485     if (error)
3486         pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3487 
3488     pci_allocate_vc_save_buffers(dev);
3489 }
3490 
3491 void pci_free_cap_save_buffers(struct pci_dev *dev)
3492 {
3493     struct pci_cap_saved_state *tmp;
3494     struct hlist_node *n;
3495 
3496     hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3497         kfree(tmp);
3498 }
3499 
3500 /**
3501  * pci_configure_ari - enable or disable ARI forwarding
3502  * @dev: the PCI device
3503  *
3504  * If @dev and its upstream bridge both support ARI, enable ARI in the
3505  * bridge.  Otherwise, disable ARI in the bridge.
3506  */
3507 void pci_configure_ari(struct pci_dev *dev)
3508 {
3509     u32 cap;
3510     struct pci_dev *bridge;
3511 
3512     if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3513         return;
3514 
3515     bridge = dev->bus->self;
3516     if (!bridge)
3517         return;
3518 
3519     pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3520     if (!(cap & PCI_EXP_DEVCAP2_ARI))
3521         return;
3522 
3523     if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3524         pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3525                      PCI_EXP_DEVCTL2_ARI);
3526         bridge->ari_enabled = 1;
3527     } else {
3528         pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3529                        PCI_EXP_DEVCTL2_ARI);
3530         bridge->ari_enabled = 0;
3531     }
3532 }
3533 
3534 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3535 {
3536     int pos;
3537     u16 cap, ctrl;
3538 
3539     pos = pdev->acs_cap;
3540     if (!pos)
3541         return false;
3542 
3543     /*
3544      * Except for egress control, capabilities are either required
3545      * or only required if controllable.  Features missing from the
3546      * capability field can therefore be assumed as hard-wired enabled.
3547      */
3548     pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3549     acs_flags &= (cap | PCI_ACS_EC);
3550 
3551     pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3552     return (ctrl & acs_flags) == acs_flags;
3553 }
3554 
3555 /**
3556  * pci_acs_enabled - test ACS against required flags for a given device
3557  * @pdev: device to test
3558  * @acs_flags: required PCI ACS flags
3559  *
3560  * Return true if the device supports the provided flags.  Automatically
3561  * filters out flags that are not implemented on multifunction devices.
3562  *
3563  * Note that this interface checks the effective ACS capabilities of the
3564  * device rather than the actual capabilities.  For instance, most single
3565  * function endpoints are not required to support ACS because they have no
3566  * opportunity for peer-to-peer access.  We therefore return 'true'
3567  * regardless of whether the device exposes an ACS capability.  This makes
3568  * it much easier for callers of this function to ignore the actual type
3569  * or topology of the device when testing ACS support.
3570  */
3571 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3572 {
3573     int ret;
3574 
3575     ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3576     if (ret >= 0)
3577         return ret > 0;
3578 
3579     /*
3580      * Conventional PCI and PCI-X devices never support ACS, either
3581      * effectively or actually.  The shared bus topology implies that
3582      * any device on the bus can receive or snoop DMA.
3583      */
3584     if (!pci_is_pcie(pdev))
3585         return false;
3586 
3587     switch (pci_pcie_type(pdev)) {
3588     /*
3589      * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3590      * but since their primary interface is PCI/X, we conservatively
3591      * handle them as we would a non-PCIe device.
3592      */
3593     case PCI_EXP_TYPE_PCIE_BRIDGE:
3594     /*
3595      * PCIe 3.0, 6.12.1 excludes ACS on these devices.  "ACS is never
3596      * applicable... must never implement an ACS Extended Capability...".
3597      * This seems arbitrary, but we take a conservative interpretation
3598      * of this statement.
3599      */
3600     case PCI_EXP_TYPE_PCI_BRIDGE:
3601     case PCI_EXP_TYPE_RC_EC:
3602         return false;
3603     /*
3604      * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3605      * implement ACS in order to indicate their peer-to-peer capabilities,
3606      * regardless of whether they are single- or multi-function devices.
3607      */
3608     case PCI_EXP_TYPE_DOWNSTREAM:
3609     case PCI_EXP_TYPE_ROOT_PORT:
3610         return pci_acs_flags_enabled(pdev, acs_flags);
3611     /*
3612      * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3613      * implemented by the remaining PCIe types to indicate peer-to-peer
3614      * capabilities, but only when they are part of a multifunction
3615      * device.  The footnote for section 6.12 indicates the specific
3616      * PCIe types included here.
3617      */
3618     case PCI_EXP_TYPE_ENDPOINT:
3619     case PCI_EXP_TYPE_UPSTREAM:
3620     case PCI_EXP_TYPE_LEG_END:
3621     case PCI_EXP_TYPE_RC_END:
3622         if (!pdev->multifunction)
3623             break;
3624 
3625         return pci_acs_flags_enabled(pdev, acs_flags);
3626     }
3627 
3628     /*
3629      * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3630      * to single function devices with the exception of downstream ports.
3631      */
3632     return true;
3633 }
3634 
3635 /**
3636  * pci_acs_path_enabled - test ACS flags from start to end in a hierarchy
3637  * @start: starting downstream device
3638  * @end: ending upstream device or NULL to search to the root bus
3639  * @acs_flags: required flags
3640  *
3641  * Walk up a device tree from start to end testing PCI ACS support.  If
3642  * any step along the way does not support the required flags, return false.
3643  */
3644 bool pci_acs_path_enabled(struct pci_dev *start,
3645               struct pci_dev *end, u16 acs_flags)
3646 {
3647     struct pci_dev *pdev, *parent = start;
3648 
3649     do {
3650         pdev = parent;
3651 
3652         if (!pci_acs_enabled(pdev, acs_flags))
3653             return false;
3654 
3655         if (pci_is_root_bus(pdev->bus))
3656             return (end == NULL);
3657 
3658         parent = pdev->bus->self;
3659     } while (pdev != end);
3660 
3661     return true;
3662 }
3663 
3664 /**
3665  * pci_acs_init - Initialize ACS if hardware supports it
3666  * @dev: the PCI device
3667  */
3668 void pci_acs_init(struct pci_dev *dev)
3669 {
3670     dev->acs_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3671 
3672     /*
3673      * Attempt to enable ACS regardless of capability because some Root
3674      * Ports (e.g. those quirked with *_intel_pch_acs_*) do not have
3675      * the standard ACS capability but still support ACS via those
3676      * quirks.
3677      */
3678     pci_enable_acs(dev);
3679 }
3680 
3681 /**
3682  * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3683  * @pdev: PCI device
3684  * @bar: BAR to find
3685  *
3686  * Helper to find the position of the ctrl register for a BAR.
3687  * Returns -ENOTSUPP if resizable BARs are not supported at all.
3688  * Returns -ENOENT if no ctrl register for the BAR could be found.
3689  */
3690 static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3691 {
3692     unsigned int pos, nbars, i;
3693     u32 ctrl;
3694 
3695     pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3696     if (!pos)
3697         return -ENOTSUPP;
3698 
3699     pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3700     nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3701             PCI_REBAR_CTRL_NBAR_SHIFT;
3702 
3703     for (i = 0; i < nbars; i++, pos += 8) {
3704         int bar_idx;
3705 
3706         pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3707         bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3708         if (bar_idx == bar)
3709             return pos;
3710     }
3711 
3712     return -ENOENT;
3713 }
3714 
3715 /**
3716  * pci_rebar_get_possible_sizes - get possible sizes for BAR
3717  * @pdev: PCI device
3718  * @bar: BAR to query
3719  *
3720  * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3721  * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3722  */
3723 u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3724 {
3725     int pos;
3726     u32 cap;
3727 
3728     pos = pci_rebar_find_pos(pdev, bar);
3729     if (pos < 0)
3730         return 0;
3731 
3732     pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3733     cap &= PCI_REBAR_CAP_SIZES;
3734 
3735     /* Sapphire RX 5600 XT Pulse has an invalid cap dword for BAR 0 */
3736     if (pdev->vendor == PCI_VENDOR_ID_ATI && pdev->device == 0x731f &&
3737         bar == 0 && cap == 0x7000)
3738         cap = 0x3f000;
3739 
3740     return cap >> 4;
3741 }
3742 EXPORT_SYMBOL(pci_rebar_get_possible_sizes);
3743 
3744 /**
3745  * pci_rebar_get_current_size - get the current size of a BAR
3746  * @pdev: PCI device
3747  * @bar: BAR to set size to
3748  *
3749  * Read the size of a BAR from the resizable BAR config.
3750  * Returns size if found or negative error code.
3751  */
3752 int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3753 {
3754     int pos;
3755     u32 ctrl;
3756 
3757     pos = pci_rebar_find_pos(pdev, bar);
3758     if (pos < 0)
3759         return pos;
3760 
3761     pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3762     return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3763 }
3764 
3765 /**
3766  * pci_rebar_set_size - set a new size for a BAR
3767  * @pdev: PCI device
3768  * @bar: BAR to set size to
3769  * @size: new size as defined in the spec (0=1MB, 19=512GB)
3770  *
3771  * Set the new size of a BAR as defined in the spec.
3772  * Returns zero if resizing was successful, error code otherwise.
3773  */
3774 int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3775 {
3776     int pos;
3777     u32 ctrl;
3778 
3779     pos = pci_rebar_find_pos(pdev, bar);
3780     if (pos < 0)
3781         return pos;
3782 
3783     pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3784     ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3785     ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3786     pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3787     return 0;
3788 }
3789 
3790 /**
3791  * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3792  * @dev: the PCI device
3793  * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3794  *  PCI_EXP_DEVCAP2_ATOMIC_COMP32
3795  *  PCI_EXP_DEVCAP2_ATOMIC_COMP64
3796  *  PCI_EXP_DEVCAP2_ATOMIC_COMP128
3797  *
3798  * Return 0 if all upstream bridges support AtomicOp routing, egress
3799  * blocking is disabled on all upstream ports, and the root port supports
3800  * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3801  * AtomicOp completion), or negative otherwise.
3802  */
3803 int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3804 {
3805     struct pci_bus *bus = dev->bus;
3806     struct pci_dev *bridge;
3807     u32 cap, ctl2;
3808 
3809     /*
3810      * Per PCIe r5.0, sec 9.3.5.10, the AtomicOp Requester Enable bit
3811      * in Device Control 2 is reserved in VFs and the PF value applies
3812      * to all associated VFs.
3813      */
3814     if (dev->is_virtfn)
3815         return -EINVAL;
3816 
3817     if (!pci_is_pcie(dev))
3818         return -EINVAL;
3819 
3820     /*
3821      * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3822      * AtomicOp requesters.  For now, we only support endpoints as
3823      * requesters and root ports as completers.  No endpoints as
3824      * completers, and no peer-to-peer.
3825      */
3826 
3827     switch (pci_pcie_type(dev)) {
3828     case PCI_EXP_TYPE_ENDPOINT:
3829     case PCI_EXP_TYPE_LEG_END:
3830     case PCI_EXP_TYPE_RC_END:
3831         break;
3832     default:
3833         return -EINVAL;
3834     }
3835 
3836     while (bus->parent) {
3837         bridge = bus->self;
3838 
3839         pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3840 
3841         switch (pci_pcie_type(bridge)) {
3842         /* Ensure switch ports support AtomicOp routing */
3843         case PCI_EXP_TYPE_UPSTREAM:
3844         case PCI_EXP_TYPE_DOWNSTREAM:
3845             if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3846                 return -EINVAL;
3847             break;
3848 
3849         /* Ensure root port supports all the sizes we care about */
3850         case PCI_EXP_TYPE_ROOT_PORT:
3851             if ((cap & cap_mask) != cap_mask)
3852                 return -EINVAL;
3853             break;
3854         }
3855 
3856         /* Ensure upstream ports don't block AtomicOps on egress */
3857         if (pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM) {
3858             pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3859                            &ctl2);
3860             if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3861                 return -EINVAL;
3862         }
3863 
3864         bus = bus->parent;
3865     }
3866 
3867     pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3868                  PCI_EXP_DEVCTL2_ATOMIC_REQ);
3869     return 0;
3870 }
3871 EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3872 
3873 /**
3874  * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3875  * @dev: the PCI device
3876  * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3877  *
3878  * Perform INTx swizzling for a device behind one level of bridge.  This is
3879  * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3880  * behind bridges on add-in cards.  For devices with ARI enabled, the slot
3881  * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3882  * the PCI Express Base Specification, Revision 2.1)
3883  */
3884 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3885 {
3886     int slot;
3887 
3888     if (pci_ari_enabled(dev->bus))
3889         slot = 0;
3890     else
3891         slot = PCI_SLOT(dev->devfn);
3892 
3893     return (((pin - 1) + slot) % 4) + 1;
3894 }
3895 
3896 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3897 {
3898     u8 pin;
3899 
3900     pin = dev->pin;
3901     if (!pin)
3902         return -1;
3903 
3904     while (!pci_is_root_bus(dev->bus)) {
3905         pin = pci_swizzle_interrupt_pin(dev, pin);
3906         dev = dev->bus->self;
3907     }
3908     *bridge = dev;
3909     return pin;
3910 }
3911 
3912 /**
3913  * pci_common_swizzle - swizzle INTx all the way to root bridge
3914  * @dev: the PCI device
3915  * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3916  *
3917  * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
3918  * bridges all the way up to a PCI root bus.
3919  */
3920 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3921 {
3922     u8 pin = *pinp;
3923 
3924     while (!pci_is_root_bus(dev->bus)) {
3925         pin = pci_swizzle_interrupt_pin(dev, pin);
3926         dev = dev->bus->self;
3927     }
3928     *pinp = pin;
3929     return PCI_SLOT(dev->devfn);
3930 }
3931 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3932 
3933 /**
3934  * pci_release_region - Release a PCI bar
3935  * @pdev: PCI device whose resources were previously reserved by
3936  *    pci_request_region()
3937  * @bar: BAR to release
3938  *
3939  * Releases the PCI I/O and memory resources previously reserved by a
3940  * successful call to pci_request_region().  Call this function only
3941  * after all use of the PCI regions has ceased.
3942  */
3943 void pci_release_region(struct pci_dev *pdev, int bar)
3944 {
3945     struct pci_devres *dr;
3946 
3947     if (pci_resource_len(pdev, bar) == 0)
3948         return;
3949     if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3950         release_region(pci_resource_start(pdev, bar),
3951                 pci_resource_len(pdev, bar));
3952     else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3953         release_mem_region(pci_resource_start(pdev, bar),
3954                 pci_resource_len(pdev, bar));
3955 
3956     dr = find_pci_dr(pdev);
3957     if (dr)
3958         dr->region_mask &= ~(1 << bar);
3959 }
3960 EXPORT_SYMBOL(pci_release_region);
3961 
3962 /**
3963  * __pci_request_region - Reserved PCI I/O and memory resource
3964  * @pdev: PCI device whose resources are to be reserved
3965  * @bar: BAR to be reserved
3966  * @res_name: Name to be associated with resource.
3967  * @exclusive: whether the region access is exclusive or not
3968  *
3969  * Mark the PCI region associated with PCI device @pdev BAR @bar as
3970  * being reserved by owner @res_name.  Do not access any
3971  * address inside the PCI regions unless this call returns
3972  * successfully.
3973  *
3974  * If @exclusive is set, then the region is marked so that userspace
3975  * is explicitly not allowed to map the resource via /dev/mem or
3976  * sysfs MMIO access.
3977  *
3978  * Returns 0 on success, or %EBUSY on error.  A warning
3979  * message is also printed on failure.
3980  */
3981 static int __pci_request_region(struct pci_dev *pdev, int bar,
3982                 const char *res_name, int exclusive)
3983 {
3984     struct pci_devres *dr;
3985 
3986     if (pci_resource_len(pdev, bar) == 0)
3987         return 0;
3988 
3989     if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3990         if (!request_region(pci_resource_start(pdev, bar),
3991                 pci_resource_len(pdev, bar), res_name))
3992             goto err_out;
3993     } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3994         if (!__request_mem_region(pci_resource_start(pdev, bar),
3995                     pci_resource_len(pdev, bar), res_name,
3996                     exclusive))
3997             goto err_out;
3998     }
3999 
4000     dr = find_pci_dr(pdev);
4001     if (dr)
4002         dr->region_mask |= 1 << bar;
4003 
4004     return 0;
4005 
4006 err_out:
4007     pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
4008          &pdev->resource[bar]);
4009     return -EBUSY;
4010 }
4011 
4012 /**
4013  * pci_request_region - Reserve PCI I/O and memory resource
4014  * @pdev: PCI device whose resources are to be reserved
4015  * @bar: BAR to be reserved
4016  * @res_name: Name to be associated with resource
4017  *
4018  * Mark the PCI region associated with PCI device @pdev BAR @bar as
4019  * being reserved by owner @res_name.  Do not access any
4020  * address inside the PCI regions unless this call returns
4021  * successfully.
4022  *
4023  * Returns 0 on success, or %EBUSY on error.  A warning
4024  * message is also printed on failure.
4025  */
4026 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
4027 {
4028     return __pci_request_region(pdev, bar, res_name, 0);
4029 }
4030 EXPORT_SYMBOL(pci_request_region);
4031 
4032 /**
4033  * pci_release_selected_regions - Release selected PCI I/O and memory resources
4034  * @pdev: PCI device whose resources were previously reserved
4035  * @bars: Bitmask of BARs to be released
4036  *
4037  * Release selected PCI I/O and memory resources previously reserved.
4038  * Call this function only after all use of the PCI regions has ceased.
4039  */
4040 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
4041 {
4042     int i;
4043 
4044     for (i = 0; i < PCI_STD_NUM_BARS; i++)
4045         if (bars & (1 << i))
4046             pci_release_region(pdev, i);
4047 }
4048 EXPORT_SYMBOL(pci_release_selected_regions);
4049 
4050 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
4051                       const char *res_name, int excl)
4052 {
4053     int i;
4054 
4055     for (i = 0; i < PCI_STD_NUM_BARS; i++)
4056         if (bars & (1 << i))
4057             if (__pci_request_region(pdev, i, res_name, excl))
4058                 goto err_out;
4059     return 0;
4060 
4061 err_out:
4062     while (--i >= 0)
4063         if (bars & (1 << i))
4064             pci_release_region(pdev, i);
4065 
4066     return -EBUSY;
4067 }
4068 
4069 
4070 /**
4071  * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
4072  * @pdev: PCI device whose resources are to be reserved
4073  * @bars: Bitmask of BARs to be requested
4074  * @res_name: Name to be associated with resource
4075  */
4076 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
4077                  const char *res_name)
4078 {
4079     return __pci_request_selected_regions(pdev, bars, res_name, 0);
4080 }
4081 EXPORT_SYMBOL(pci_request_selected_regions);
4082 
4083 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
4084                        const char *res_name)
4085 {
4086     return __pci_request_selected_regions(pdev, bars, res_name,
4087             IORESOURCE_EXCLUSIVE);
4088 }
4089 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
4090 
4091 /**
4092  * pci_release_regions - Release reserved PCI I/O and memory resources
4093  * @pdev: PCI device whose resources were previously reserved by
4094  *    pci_request_regions()
4095  *
4096  * Releases all PCI I/O and memory resources previously reserved by a
4097  * successful call to pci_request_regions().  Call this function only
4098  * after all use of the PCI regions has ceased.
4099  */
4100 
4101 void pci_release_regions(struct pci_dev *pdev)
4102 {
4103     pci_release_selected_regions(pdev, (1 << PCI_STD_NUM_BARS) - 1);
4104 }
4105 EXPORT_SYMBOL(pci_release_regions);
4106 
4107 /**
4108  * pci_request_regions - Reserve PCI I/O and memory resources
4109  * @pdev: PCI device whose resources are to be reserved
4110  * @res_name: Name to be associated with resource.
4111  *
4112  * Mark all PCI regions associated with PCI device @pdev as
4113  * being reserved by owner @res_name.  Do not access any
4114  * address inside the PCI regions unless this call returns
4115  * successfully.
4116  *
4117  * Returns 0 on success, or %EBUSY on error.  A warning
4118  * message is also printed on failure.
4119  */
4120 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
4121 {
4122     return pci_request_selected_regions(pdev,
4123             ((1 << PCI_STD_NUM_BARS) - 1), res_name);
4124 }
4125 EXPORT_SYMBOL(pci_request_regions);
4126 
4127 /**
4128  * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
4129  * @pdev: PCI device whose resources are to be reserved
4130  * @res_name: Name to be associated with resource.
4131  *
4132  * Mark all PCI regions associated with PCI device @pdev as being reserved
4133  * by owner @res_name.  Do not access any address inside the PCI regions
4134  * unless this call returns successfully.
4135  *
4136  * pci_request_regions_exclusive() will mark the region so that /dev/mem
4137  * and the sysfs MMIO access will not be allowed.
4138  *
4139  * Returns 0 on success, or %EBUSY on error.  A warning message is also
4140  * printed on failure.
4141  */
4142 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
4143 {
4144     return pci_request_selected_regions_exclusive(pdev,
4145                 ((1 << PCI_STD_NUM_BARS) - 1), res_name);
4146 }
4147 EXPORT_SYMBOL(pci_request_regions_exclusive);
4148 
4149 /*
4150  * Record the PCI IO range (expressed as CPU physical address + size).
4151  * Return a negative value if an error has occurred, zero otherwise
4152  */
4153 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
4154             resource_size_t size)
4155 {
4156     int ret = 0;
4157 #ifdef PCI_IOBASE
4158     struct logic_pio_hwaddr *range;
4159 
4160     if (!size || addr + size < addr)
4161         return -EINVAL;
4162 
4163     range = kzalloc(sizeof(*range), GFP_ATOMIC);
4164     if (!range)
4165         return -ENOMEM;
4166 
4167     range->fwnode = fwnode;
4168     range->size = size;
4169     range->hw_start = addr;
4170     range->flags = LOGIC_PIO_CPU_MMIO;
4171 
4172     ret = logic_pio_register_range(range);
4173     if (ret)
4174         kfree(range);
4175 
4176     /* Ignore duplicates due to deferred probing */
4177     if (ret == -EEXIST)
4178         ret = 0;
4179 #endif
4180 
4181     return ret;
4182 }
4183 
4184 phys_addr_t pci_pio_to_address(unsigned long pio)
4185 {
4186     phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
4187 
4188 #ifdef PCI_IOBASE
4189     if (pio >= MMIO_UPPER_LIMIT)
4190         return address;
4191 
4192     address = logic_pio_to_hwaddr(pio);
4193 #endif
4194 
4195     return address;
4196 }
4197 EXPORT_SYMBOL_GPL(pci_pio_to_address);
4198 
4199 unsigned long __weak pci_address_to_pio(phys_addr_t address)
4200 {
4201 #ifdef PCI_IOBASE
4202     return logic_pio_trans_cpuaddr(address);
4203 #else
4204     if (address > IO_SPACE_LIMIT)
4205         return (unsigned long)-1;
4206 
4207     return (unsigned long) address;
4208 #endif
4209 }
4210 
4211 /**
4212  * pci_remap_iospace - Remap the memory mapped I/O space
4213  * @res: Resource describing the I/O space
4214  * @phys_addr: physical address of range to be mapped
4215  *
4216  * Remap the memory mapped I/O space described by the @res and the CPU
4217  * physical address @phys_addr into virtual address space.  Only
4218  * architectures that have memory mapped IO functions defined (and the
4219  * PCI_IOBASE value defined) should call this function.
4220  */
4221 #ifndef pci_remap_iospace
4222 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
4223 {
4224 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4225     unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4226 
4227     if (!(res->flags & IORESOURCE_IO))
4228         return -EINVAL;
4229 
4230     if (res->end > IO_SPACE_LIMIT)
4231         return -EINVAL;
4232 
4233     return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
4234                   pgprot_device(PAGE_KERNEL));
4235 #else
4236     /*
4237      * This architecture does not have memory mapped I/O space,
4238      * so this function should never be called
4239      */
4240     WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
4241     return -ENODEV;
4242 #endif
4243 }
4244 EXPORT_SYMBOL(pci_remap_iospace);
4245 #endif
4246 
4247 /**
4248  * pci_unmap_iospace - Unmap the memory mapped I/O space
4249  * @res: resource to be unmapped
4250  *
4251  * Unmap the CPU virtual address @res from virtual address space.  Only
4252  * architectures that have memory mapped IO functions defined (and the
4253  * PCI_IOBASE value defined) should call this function.
4254  */
4255 void pci_unmap_iospace(struct resource *res)
4256 {
4257 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4258     unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4259 
4260     vunmap_range(vaddr, vaddr + resource_size(res));
4261 #endif
4262 }
4263 EXPORT_SYMBOL(pci_unmap_iospace);
4264 
4265 static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
4266 {
4267     struct resource **res = ptr;
4268 
4269     pci_unmap_iospace(*res);
4270 }
4271 
4272 /**
4273  * devm_pci_remap_iospace - Managed pci_remap_iospace()
4274  * @dev: Generic device to remap IO address for
4275  * @res: Resource describing the I/O space
4276  * @phys_addr: physical address of range to be mapped
4277  *
4278  * Managed pci_remap_iospace().  Map is automatically unmapped on driver
4279  * detach.
4280  */
4281 int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
4282                phys_addr_t phys_addr)
4283 {
4284     const struct resource **ptr;
4285     int error;
4286 
4287     ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
4288     if (!ptr)
4289         return -ENOMEM;
4290 
4291     error = pci_remap_iospace(res, phys_addr);
4292     if (error) {
4293         devres_free(ptr);
4294     } else  {
4295         *ptr = res;
4296         devres_add(dev, ptr);
4297     }
4298 
4299     return error;
4300 }
4301 EXPORT_SYMBOL(devm_pci_remap_iospace);
4302 
4303 /**
4304  * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
4305  * @dev: Generic device to remap IO address for
4306  * @offset: Resource address to map
4307  * @size: Size of map
4308  *
4309  * Managed pci_remap_cfgspace().  Map is automatically unmapped on driver
4310  * detach.
4311  */
4312 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
4313                       resource_size_t offset,
4314                       resource_size_t size)
4315 {
4316     void __iomem **ptr, *addr;
4317 
4318     ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
4319     if (!ptr)
4320         return NULL;
4321 
4322     addr = pci_remap_cfgspace(offset, size);
4323     if (addr) {
4324         *ptr = addr;
4325         devres_add(dev, ptr);
4326     } else
4327         devres_free(ptr);
4328 
4329     return addr;
4330 }
4331 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
4332 
4333 /**
4334  * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
4335  * @dev: generic device to handle the resource for
4336  * @res: configuration space resource to be handled
4337  *
4338  * Checks that a resource is a valid memory region, requests the memory
4339  * region and ioremaps with pci_remap_cfgspace() API that ensures the
4340  * proper PCI configuration space memory attributes are guaranteed.
4341  *
4342  * All operations are managed and will be undone on driver detach.
4343  *
4344  * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
4345  * on failure. Usage example::
4346  *
4347  *  res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
4348  *  base = devm_pci_remap_cfg_resource(&pdev->dev, res);
4349  *  if (IS_ERR(base))
4350  *      return PTR_ERR(base);
4351  */
4352 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
4353                       struct resource *res)
4354 {
4355     resource_size_t size;
4356     const char *name;
4357     void __iomem *dest_ptr;
4358 
4359     BUG_ON(!dev);
4360 
4361     if (!res || resource_type(res) != IORESOURCE_MEM) {
4362         dev_err(dev, "invalid resource\n");
4363         return IOMEM_ERR_PTR(-EINVAL);
4364     }
4365 
4366     size = resource_size(res);
4367 
4368     if (res->name)
4369         name = devm_kasprintf(dev, GFP_KERNEL, "%s %s", dev_name(dev),
4370                       res->name);
4371     else
4372         name = devm_kstrdup(dev, dev_name(dev), GFP_KERNEL);
4373     if (!name)
4374         return IOMEM_ERR_PTR(-ENOMEM);
4375 
4376     if (!devm_request_mem_region(dev, res->start, size, name)) {
4377         dev_err(dev, "can't request region for resource %pR\n", res);
4378         return IOMEM_ERR_PTR(-EBUSY);
4379     }
4380 
4381     dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4382     if (!dest_ptr) {
4383         dev_err(dev, "ioremap failed for resource %pR\n", res);
4384         devm_release_mem_region(dev, res->start, size);
4385         dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4386     }
4387 
4388     return dest_ptr;
4389 }
4390 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4391 
4392 static void __pci_set_master(struct pci_dev *dev, bool enable)
4393 {
4394     u16 old_cmd, cmd;
4395 
4396     pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4397     if (enable)
4398         cmd = old_cmd | PCI_COMMAND_MASTER;
4399     else
4400         cmd = old_cmd & ~PCI_COMMAND_MASTER;
4401     if (cmd != old_cmd) {
4402         pci_dbg(dev, "%s bus mastering\n",
4403             enable ? "enabling" : "disabling");
4404         pci_write_config_word(dev, PCI_COMMAND, cmd);
4405     }
4406     dev->is_busmaster = enable;
4407 }
4408 
4409 /**
4410  * pcibios_setup - process "pci=" kernel boot arguments
4411  * @str: string used to pass in "pci=" kernel boot arguments
4412  *
4413  * Process kernel boot arguments.  This is the default implementation.
4414  * Architecture specific implementations can override this as necessary.
4415  */
4416 char * __weak __init pcibios_setup(char *str)
4417 {
4418     return str;
4419 }
4420 
4421 /**
4422  * pcibios_set_master - enable PCI bus-mastering for device dev
4423  * @dev: the PCI device to enable
4424  *
4425  * Enables PCI bus-mastering for the device.  This is the default
4426  * implementation.  Architecture specific implementations can override
4427  * this if necessary.
4428  */
4429 void __weak pcibios_set_master(struct pci_dev *dev)
4430 {
4431     u8 lat;
4432 
4433     /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4434     if (pci_is_pcie(dev))
4435         return;
4436 
4437     pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4438     if (lat < 16)
4439         lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4440     else if (lat > pcibios_max_latency)
4441         lat = pcibios_max_latency;
4442     else
4443         return;
4444 
4445     pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4446 }
4447 
4448 /**
4449  * pci_set_master - enables bus-mastering for device dev
4450  * @dev: the PCI device to enable
4451  *
4452  * Enables bus-mastering on the device and calls pcibios_set_master()
4453  * to do the needed arch specific settings.
4454  */
4455 void pci_set_master(struct pci_dev *dev)
4456 {
4457     __pci_set_master(dev, true);
4458     pcibios_set_master(dev);
4459 }
4460 EXPORT_SYMBOL(pci_set_master);
4461 
4462 /**
4463  * pci_clear_master - disables bus-mastering for device dev
4464  * @dev: the PCI device to disable
4465  */
4466 void pci_clear_master(struct pci_dev *dev)
4467 {
4468     __pci_set_master(dev, false);
4469 }
4470 EXPORT_SYMBOL(pci_clear_master);
4471 
4472 /**
4473  * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4474  * @dev: the PCI device for which MWI is to be enabled
4475  *
4476  * Helper function for pci_set_mwi.
4477  * Originally copied from drivers/net/acenic.c.
4478  * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4479  *
4480  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4481  */
4482 int pci_set_cacheline_size(struct pci_dev *dev)
4483 {
4484     u8 cacheline_size;
4485 
4486     if (!pci_cache_line_size)
4487         return -EINVAL;
4488 
4489     /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4490        equal to or multiple of the right value. */
4491     pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4492     if (cacheline_size >= pci_cache_line_size &&
4493         (cacheline_size % pci_cache_line_size) == 0)
4494         return 0;
4495 
4496     /* Write the correct value. */
4497     pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4498     /* Read it back. */
4499     pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4500     if (cacheline_size == pci_cache_line_size)
4501         return 0;
4502 
4503     pci_dbg(dev, "cache line size of %d is not supported\n",
4504            pci_cache_line_size << 2);
4505 
4506     return -EINVAL;
4507 }
4508 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4509 
4510 /**
4511  * pci_set_mwi - enables memory-write-invalidate PCI transaction
4512  * @dev: the PCI device for which MWI is enabled
4513  *
4514  * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4515  *
4516  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4517  */
4518 int pci_set_mwi(struct pci_dev *dev)
4519 {
4520 #ifdef PCI_DISABLE_MWI
4521     return 0;
4522 #else
4523     int rc;
4524     u16 cmd;
4525 
4526     rc = pci_set_cacheline_size(dev);
4527     if (rc)
4528         return rc;
4529 
4530     pci_read_config_word(dev, PCI_COMMAND, &cmd);
4531     if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4532         pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4533         cmd |= PCI_COMMAND_INVALIDATE;
4534         pci_write_config_word(dev, PCI_COMMAND, cmd);
4535     }
4536     return 0;
4537 #endif
4538 }
4539 EXPORT_SYMBOL(pci_set_mwi);
4540 
4541 /**
4542  * pcim_set_mwi - a device-managed pci_set_mwi()
4543  * @dev: the PCI device for which MWI is enabled
4544  *
4545  * Managed pci_set_mwi().
4546  *
4547  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4548  */
4549 int pcim_set_mwi(struct pci_dev *dev)
4550 {
4551     struct pci_devres *dr;
4552 
4553     dr = find_pci_dr(dev);
4554     if (!dr)
4555         return -ENOMEM;
4556 
4557     dr->mwi = 1;
4558     return pci_set_mwi(dev);
4559 }
4560 EXPORT_SYMBOL(pcim_set_mwi);
4561 
4562 /**
4563  * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4564  * @dev: the PCI device for which MWI is enabled
4565  *
4566  * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4567  * Callers are not required to check the return value.
4568  *
4569  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4570  */
4571 int pci_try_set_mwi(struct pci_dev *dev)
4572 {
4573 #ifdef PCI_DISABLE_MWI
4574     return 0;
4575 #else
4576     return pci_set_mwi(dev);
4577 #endif
4578 }
4579 EXPORT_SYMBOL(pci_try_set_mwi);
4580 
4581 /**
4582  * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4583  * @dev: the PCI device to disable
4584  *
4585  * Disables PCI Memory-Write-Invalidate transaction on the device
4586  */
4587 void pci_clear_mwi(struct pci_dev *dev)
4588 {
4589 #ifndef PCI_DISABLE_MWI
4590     u16 cmd;
4591 
4592     pci_read_config_word(dev, PCI_COMMAND, &cmd);
4593     if (cmd & PCI_COMMAND_INVALIDATE) {
4594         cmd &= ~PCI_COMMAND_INVALIDATE;
4595         pci_write_config_word(dev, PCI_COMMAND, cmd);
4596     }
4597 #endif
4598 }
4599 EXPORT_SYMBOL(pci_clear_mwi);
4600 
4601 /**
4602  * pci_disable_parity - disable parity checking for device
4603  * @dev: the PCI device to operate on
4604  *
4605  * Disable parity checking for device @dev
4606  */
4607 void pci_disable_parity(struct pci_dev *dev)
4608 {
4609     u16 cmd;
4610 
4611     pci_read_config_word(dev, PCI_COMMAND, &cmd);
4612     if (cmd & PCI_COMMAND_PARITY) {
4613         cmd &= ~PCI_COMMAND_PARITY;
4614         pci_write_config_word(dev, PCI_COMMAND, cmd);
4615     }
4616 }
4617 
4618 /**
4619  * pci_intx - enables/disables PCI INTx for device dev
4620  * @pdev: the PCI device to operate on
4621  * @enable: boolean: whether to enable or disable PCI INTx
4622  *
4623  * Enables/disables PCI INTx for device @pdev
4624  */
4625 void pci_intx(struct pci_dev *pdev, int enable)
4626 {
4627     u16 pci_command, new;
4628 
4629     pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4630 
4631     if (enable)
4632         new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4633     else
4634         new = pci_command | PCI_COMMAND_INTX_DISABLE;
4635 
4636     if (new != pci_command) {
4637         struct pci_devres *dr;
4638 
4639         pci_write_config_word(pdev, PCI_COMMAND, new);
4640 
4641         dr = find_pci_dr(pdev);
4642         if (dr && !dr->restore_intx) {
4643             dr->restore_intx = 1;
4644             dr->orig_intx = !enable;
4645         }
4646     }
4647 }
4648 EXPORT_SYMBOL_GPL(pci_intx);
4649 
4650 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4651 {
4652     struct pci_bus *bus = dev->bus;
4653     bool mask_updated = true;
4654     u32 cmd_status_dword;
4655     u16 origcmd, newcmd;
4656     unsigned long flags;
4657     bool irq_pending;
4658 
4659     /*
4660      * We do a single dword read to retrieve both command and status.
4661      * Document assumptions that make this possible.
4662      */
4663     BUILD_BUG_ON(PCI_COMMAND % 4);
4664     BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4665 
4666     raw_spin_lock_irqsave(&pci_lock, flags);
4667 
4668     bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4669 
4670     irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4671 
4672     /*
4673      * Check interrupt status register to see whether our device
4674      * triggered the interrupt (when masking) or the next IRQ is
4675      * already pending (when unmasking).
4676      */
4677     if (mask != irq_pending) {
4678         mask_updated = false;
4679         goto done;
4680     }
4681 
4682     origcmd = cmd_status_dword;
4683     newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4684     if (mask)
4685         newcmd |= PCI_COMMAND_INTX_DISABLE;
4686     if (newcmd != origcmd)
4687         bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4688 
4689 done:
4690     raw_spin_unlock_irqrestore(&pci_lock, flags);
4691 
4692     return mask_updated;
4693 }
4694 
4695 /**
4696  * pci_check_and_mask_intx - mask INTx on pending interrupt
4697  * @dev: the PCI device to operate on
4698  *
4699  * Check if the device dev has its INTx line asserted, mask it and return
4700  * true in that case. False is returned if no interrupt was pending.
4701  */
4702 bool pci_check_and_mask_intx(struct pci_dev *dev)
4703 {
4704     return pci_check_and_set_intx_mask(dev, true);
4705 }
4706 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4707 
4708 /**
4709  * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4710  * @dev: the PCI device to operate on
4711  *
4712  * Check if the device dev has its INTx line asserted, unmask it if not and
4713  * return true. False is returned and the mask remains active if there was
4714  * still an interrupt pending.
4715  */
4716 bool pci_check_and_unmask_intx(struct pci_dev *dev)
4717 {
4718     return pci_check_and_set_intx_mask(dev, false);
4719 }
4720 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4721 
4722 /**
4723  * pci_wait_for_pending_transaction - wait for pending transaction
4724  * @dev: the PCI device to operate on
4725  *
4726  * Return 0 if transaction is pending 1 otherwise.
4727  */
4728 int pci_wait_for_pending_transaction(struct pci_dev *dev)
4729 {
4730     if (!pci_is_pcie(dev))
4731         return 1;
4732 
4733     return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4734                     PCI_EXP_DEVSTA_TRPND);
4735 }
4736 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4737 
4738 /**
4739  * pcie_flr - initiate a PCIe function level reset
4740  * @dev: device to reset
4741  *
4742  * Initiate a function level reset unconditionally on @dev without
4743  * checking any flags and DEVCAP
4744  */
4745 int pcie_flr(struct pci_dev *dev)
4746 {
4747     if (!pci_wait_for_pending_transaction(dev))
4748         pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4749 
4750     pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4751 
4752     if (dev->imm_ready)
4753         return 0;
4754 
4755     /*
4756      * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4757      * 100ms, but may silently discard requests while the FLR is in
4758      * progress.  Wait 100ms before trying to access the device.
4759      */
4760     msleep(100);
4761 
4762     return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4763 }
4764 EXPORT_SYMBOL_GPL(pcie_flr);
4765 
4766 /**
4767  * pcie_reset_flr - initiate a PCIe function level reset
4768  * @dev: device to reset
4769  * @probe: if true, return 0 if device can be reset this way
4770  *
4771  * Initiate a function level reset on @dev.
4772  */
4773 int pcie_reset_flr(struct pci_dev *dev, bool probe)
4774 {
4775     if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4776         return -ENOTTY;
4777 
4778     if (!(dev->devcap & PCI_EXP_DEVCAP_FLR))
4779         return -ENOTTY;
4780 
4781     if (probe)
4782         return 0;
4783 
4784     return pcie_flr(dev);
4785 }
4786 EXPORT_SYMBOL_GPL(pcie_reset_flr);
4787 
4788 static int pci_af_flr(struct pci_dev *dev, bool probe)
4789 {
4790     int pos;
4791     u8 cap;
4792 
4793     pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4794     if (!pos)
4795         return -ENOTTY;
4796 
4797     if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4798         return -ENOTTY;
4799 
4800     pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4801     if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4802         return -ENOTTY;
4803 
4804     if (probe)
4805         return 0;
4806 
4807     /*
4808      * Wait for Transaction Pending bit to clear.  A word-aligned test
4809      * is used, so we use the control offset rather than status and shift
4810      * the test bit to match.
4811      */
4812     if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4813                  PCI_AF_STATUS_TP << 8))
4814         pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4815 
4816     pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4817 
4818     if (dev->imm_ready)
4819         return 0;
4820 
4821     /*
4822      * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4823      * updated 27 July 2006; a device must complete an FLR within
4824      * 100ms, but may silently discard requests while the FLR is in
4825      * progress.  Wait 100ms before trying to access the device.
4826      */
4827     msleep(100);
4828 
4829     return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4830 }
4831 
4832 /**
4833  * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4834  * @dev: Device to reset.
4835  * @probe: if true, return 0 if the device can be reset this way.
4836  *
4837  * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4838  * unset, it will be reinitialized internally when going from PCI_D3hot to
4839  * PCI_D0.  If that's the case and the device is not in a low-power state
4840  * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4841  *
4842  * NOTE: This causes the caller to sleep for twice the device power transition
4843  * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4844  * by default (i.e. unless the @dev's d3hot_delay field has a different value).
4845  * Moreover, only devices in D0 can be reset by this function.
4846  */
4847 static int pci_pm_reset(struct pci_dev *dev, bool probe)
4848 {
4849     u16 csr;
4850 
4851     if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4852         return -ENOTTY;
4853 
4854     pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4855     if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4856         return -ENOTTY;
4857 
4858     if (probe)
4859         return 0;
4860 
4861     if (dev->current_state != PCI_D0)
4862         return -EINVAL;
4863 
4864     csr &= ~PCI_PM_CTRL_STATE_MASK;
4865     csr |= PCI_D3hot;
4866     pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4867     pci_dev_d3_sleep(dev);
4868 
4869     csr &= ~PCI_PM_CTRL_STATE_MASK;
4870     csr |= PCI_D0;
4871     pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4872     pci_dev_d3_sleep(dev);
4873 
4874     return pci_dev_wait(dev, "PM D3hot->D0", PCIE_RESET_READY_POLL_MS);
4875 }
4876 
4877 /**
4878  * pcie_wait_for_link_delay - Wait until link is active or inactive
4879  * @pdev: Bridge device
4880  * @active: waiting for active or inactive?
4881  * @delay: Delay to wait after link has become active (in ms)
4882  *
4883  * Use this to wait till link becomes active or inactive.
4884  */
4885 static bool pcie_wait_for_link_delay(struct pci_dev *pdev, bool active,
4886                      int delay)
4887 {
4888     int timeout = 1000;
4889     bool ret;
4890     u16 lnk_status;
4891 
4892     /*
4893      * Some controllers might not implement link active reporting. In this
4894      * case, we wait for 1000 ms + any delay requested by the caller.
4895      */
4896     if (!pdev->link_active_reporting) {
4897         msleep(timeout + delay);
4898         return true;
4899     }
4900 
4901     /*
4902      * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4903      * after which we should expect an link active if the reset was
4904      * successful. If so, software must wait a minimum 100ms before sending
4905      * configuration requests to devices downstream this port.
4906      *
4907      * If the link fails to activate, either the device was physically
4908      * removed or the link is permanently failed.
4909      */
4910     if (active)
4911         msleep(20);
4912     for (;;) {
4913         pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4914         ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4915         if (ret == active)
4916             break;
4917         if (timeout <= 0)
4918             break;
4919         msleep(10);
4920         timeout -= 10;
4921     }
4922     if (active && ret)
4923         msleep(delay);
4924 
4925     return ret == active;
4926 }
4927 
4928 /**
4929  * pcie_wait_for_link - Wait until link is active or inactive
4930  * @pdev: Bridge device
4931  * @active: waiting for active or inactive?
4932  *
4933  * Use this to wait till link becomes active or inactive.
4934  */
4935 bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4936 {
4937     return pcie_wait_for_link_delay(pdev, active, 100);
4938 }
4939 
4940 /*
4941  * Find maximum D3cold delay required by all the devices on the bus.  The
4942  * spec says 100 ms, but firmware can lower it and we allow drivers to
4943  * increase it as well.
4944  *
4945  * Called with @pci_bus_sem locked for reading.
4946  */
4947 static int pci_bus_max_d3cold_delay(const struct pci_bus *bus)
4948 {
4949     const struct pci_dev *pdev;
4950     int min_delay = 100;
4951     int max_delay = 0;
4952 
4953     list_for_each_entry(pdev, &bus->devices, bus_list) {
4954         if (pdev->d3cold_delay < min_delay)
4955             min_delay = pdev->d3cold_delay;
4956         if (pdev->d3cold_delay > max_delay)
4957             max_delay = pdev->d3cold_delay;
4958     }
4959 
4960     return max(min_delay, max_delay);
4961 }
4962 
4963 /**
4964  * pci_bridge_wait_for_secondary_bus - Wait for secondary bus to be accessible
4965  * @dev: PCI bridge
4966  *
4967  * Handle necessary delays before access to the devices on the secondary
4968  * side of the bridge are permitted after D3cold to D0 transition.
4969  *
4970  * For PCIe this means the delays in PCIe 5.0 section 6.6.1. For
4971  * conventional PCI it means Tpvrh + Trhfa specified in PCI 3.0 section
4972  * 4.3.2.
4973  */
4974 void pci_bridge_wait_for_secondary_bus(struct pci_dev *dev)
4975 {
4976     struct pci_dev *child;
4977     int delay;
4978 
4979     if (pci_dev_is_disconnected(dev))
4980         return;
4981 
4982     if (!pci_is_bridge(dev) || !dev->bridge_d3)
4983         return;
4984 
4985     down_read(&pci_bus_sem);
4986 
4987     /*
4988      * We only deal with devices that are present currently on the bus.
4989      * For any hot-added devices the access delay is handled in pciehp
4990      * board_added(). In case of ACPI hotplug the firmware is expected
4991      * to configure the devices before OS is notified.
4992      */
4993     if (!dev->subordinate || list_empty(&dev->subordinate->devices)) {
4994         up_read(&pci_bus_sem);
4995         return;
4996     }
4997 
4998     /* Take d3cold_delay requirements into account */
4999     delay = pci_bus_max_d3cold_delay(dev->subordinate);
5000     if (!delay) {
5001         up_read(&pci_bus_sem);
5002         return;
5003     }
5004 
5005     child = list_first_entry(&dev->subordinate->devices, struct pci_dev,
5006                  bus_list);
5007     up_read(&pci_bus_sem);
5008 
5009     /*
5010      * Conventional PCI and PCI-X we need to wait Tpvrh + Trhfa before
5011      * accessing the device after reset (that is 1000 ms + 100 ms). In
5012      * practice this should not be needed because we don't do power
5013      * management for them (see pci_bridge_d3_possible()).
5014      */
5015     if (!pci_is_pcie(dev)) {
5016         pci_dbg(dev, "waiting %d ms for secondary bus\n", 1000 + delay);
5017         msleep(1000 + delay);
5018         return;
5019     }
5020 
5021     /*
5022      * For PCIe downstream and root ports that do not support speeds
5023      * greater than 5 GT/s need to wait minimum 100 ms. For higher
5024      * speeds (gen3) we need to wait first for the data link layer to
5025      * become active.
5026      *
5027      * However, 100 ms is the minimum and the PCIe spec says the
5028      * software must allow at least 1s before it can determine that the
5029      * device that did not respond is a broken device. There is
5030      * evidence that 100 ms is not always enough, for example certain
5031      * Titan Ridge xHCI controller does not always respond to
5032      * configuration requests if we only wait for 100 ms (see
5033      * https://bugzilla.kernel.org/show_bug.cgi?id=203885).
5034      *
5035      * Therefore we wait for 100 ms and check for the device presence.
5036      * If it is still not present give it an additional 100 ms.
5037      */
5038     if (!pcie_downstream_port(dev))
5039         return;
5040 
5041     if (pcie_get_speed_cap(dev) <= PCIE_SPEED_5_0GT) {
5042         pci_dbg(dev, "waiting %d ms for downstream link\n", delay);
5043         msleep(delay);
5044     } else {
5045         pci_dbg(dev, "waiting %d ms for downstream link, after activation\n",
5046             delay);
5047         if (!pcie_wait_for_link_delay(dev, true, delay)) {
5048             /* Did not train, no need to wait any further */
5049             pci_info(dev, "Data Link Layer Link Active not set in 1000 msec\n");
5050             return;
5051         }
5052     }
5053 
5054     if (!pci_device_is_present(child)) {
5055         pci_dbg(child, "waiting additional %d ms to become accessible\n", delay);
5056         msleep(delay);
5057     }
5058 }
5059 
5060 void pci_reset_secondary_bus(struct pci_dev *dev)
5061 {
5062     u16 ctrl;
5063 
5064     pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
5065     ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
5066     pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
5067 
5068     /*
5069      * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms.  Double
5070      * this to 2ms to ensure that we meet the minimum requirement.
5071      */
5072     msleep(2);
5073 
5074     ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
5075     pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
5076 
5077     /*
5078      * Trhfa for conventional PCI is 2^25 clock cycles.
5079      * Assuming a minimum 33MHz clock this results in a 1s
5080      * delay before we can consider subordinate devices to
5081      * be re-initialized.  PCIe has some ways to shorten this,
5082      * but we don't make use of them yet.
5083      */
5084     ssleep(1);
5085 }
5086 
5087 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
5088 {
5089     pci_reset_secondary_bus(dev);
5090 }
5091 
5092 /**
5093  * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
5094  * @dev: Bridge device
5095  *
5096  * Use the bridge control register to assert reset on the secondary bus.
5097  * Devices on the secondary bus are left in power-on state.
5098  */
5099 int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
5100 {
5101     pcibios_reset_secondary_bus(dev);
5102 
5103     return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
5104 }
5105 EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
5106 
5107 static int pci_parent_bus_reset(struct pci_dev *dev, bool probe)
5108 {
5109     struct pci_dev *pdev;
5110 
5111     if (pci_is_root_bus(dev->bus) || dev->subordinate ||
5112         !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
5113         return -ENOTTY;
5114 
5115     list_for_each_entry(pdev, &dev->bus->devices, bus_list)
5116         if (pdev != dev)
5117             return -ENOTTY;
5118 
5119     if (probe)
5120         return 0;
5121 
5122     return pci_bridge_secondary_bus_reset(dev->bus->self);
5123 }
5124 
5125 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, bool probe)
5126 {
5127     int rc = -ENOTTY;
5128 
5129     if (!hotplug || !try_module_get(hotplug->owner))
5130         return rc;
5131 
5132     if (hotplug->ops->reset_slot)
5133         rc = hotplug->ops->reset_slot(hotplug, probe);
5134 
5135     module_put(hotplug->owner);
5136 
5137     return rc;
5138 }
5139 
5140 static int pci_dev_reset_slot_function(struct pci_dev *dev, bool probe)
5141 {
5142     if (dev->multifunction || dev->subordinate || !dev->slot ||
5143         dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
5144         return -ENOTTY;
5145 
5146     return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
5147 }
5148 
5149 static int pci_reset_bus_function(struct pci_dev *dev, bool probe)
5150 {
5151     int rc;
5152 
5153     rc = pci_dev_reset_slot_function(dev, probe);
5154     if (rc != -ENOTTY)
5155         return rc;
5156     return pci_parent_bus_reset(dev, probe);
5157 }
5158 
5159 void pci_dev_lock(struct pci_dev *dev)
5160 {
5161     /* block PM suspend, driver probe, etc. */
5162     device_lock(&dev->dev);
5163     pci_cfg_access_lock(dev);
5164 }
5165 EXPORT_SYMBOL_GPL(pci_dev_lock);
5166 
5167 /* Return 1 on successful lock, 0 on contention */
5168 int pci_dev_trylock(struct pci_dev *dev)
5169 {
5170     if (device_trylock(&dev->dev)) {
5171         if (pci_cfg_access_trylock(dev))
5172             return 1;
5173         device_unlock(&dev->dev);
5174     }
5175 
5176     return 0;
5177 }
5178 EXPORT_SYMBOL_GPL(pci_dev_trylock);
5179 
5180 void pci_dev_unlock(struct pci_dev *dev)
5181 {
5182     pci_cfg_access_unlock(dev);
5183     device_unlock(&dev->dev);
5184 }
5185 EXPORT_SYMBOL_GPL(pci_dev_unlock);
5186 
5187 static void pci_dev_save_and_disable(struct pci_dev *dev)
5188 {
5189     const struct pci_error_handlers *err_handler =
5190             dev->driver ? dev->driver->err_handler : NULL;
5191 
5192     /*
5193      * dev->driver->err_handler->reset_prepare() is protected against
5194      * races with ->remove() by the device lock, which must be held by
5195      * the caller.
5196      */
5197     if (err_handler && err_handler->reset_prepare)
5198         err_handler->reset_prepare(dev);
5199 
5200     /*
5201      * Wake-up device prior to save.  PM registers default to D0 after
5202      * reset and a simple register restore doesn't reliably return
5203      * to a non-D0 state anyway.
5204      */
5205     pci_set_power_state(dev, PCI_D0);
5206 
5207     pci_save_state(dev);
5208     /*
5209      * Disable the device by clearing the Command register, except for
5210      * INTx-disable which is set.  This not only disables MMIO and I/O port
5211      * BARs, but also prevents the device from being Bus Master, preventing
5212      * DMA from the device including MSI/MSI-X interrupts.  For PCI 2.3
5213      * compliant devices, INTx-disable prevents legacy interrupts.
5214      */
5215     pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
5216 }
5217 
5218 static void pci_dev_restore(struct pci_dev *dev)
5219 {
5220     const struct pci_error_handlers *err_handler =
5221             dev->driver ? dev->driver->err_handler : NULL;
5222 
5223     pci_restore_state(dev);
5224 
5225     /*
5226      * dev->driver->err_handler->reset_done() is protected against
5227      * races with ->remove() by the device lock, which must be held by
5228      * the caller.
5229      */
5230     if (err_handler && err_handler->reset_done)
5231         err_handler->reset_done(dev);
5232 }
5233 
5234 /* dev->reset_methods[] is a 0-terminated list of indices into this array */
5235 static const struct pci_reset_fn_method pci_reset_fn_methods[] = {
5236     { },
5237     { pci_dev_specific_reset, .name = "device_specific" },
5238     { pci_dev_acpi_reset, .name = "acpi" },
5239     { pcie_reset_flr, .name = "flr" },
5240     { pci_af_flr, .name = "af_flr" },
5241     { pci_pm_reset, .name = "pm" },
5242     { pci_reset_bus_function, .name = "bus" },
5243 };
5244 
5245 static ssize_t reset_method_show(struct device *dev,
5246                  struct device_attribute *attr, char *buf)
5247 {
5248     struct pci_dev *pdev = to_pci_dev(dev);
5249     ssize_t len = 0;
5250     int i, m;
5251 
5252     for (i = 0; i < PCI_NUM_RESET_METHODS; i++) {
5253         m = pdev->reset_methods[i];
5254         if (!m)
5255             break;
5256 
5257         len += sysfs_emit_at(buf, len, "%s%s", len ? " " : "",
5258                      pci_reset_fn_methods[m].name);
5259     }
5260 
5261     if (len)
5262         len += sysfs_emit_at(buf, len, "\n");
5263 
5264     return len;
5265 }
5266 
5267 static int reset_method_lookup(const char *name)
5268 {
5269     int m;
5270 
5271     for (m = 1; m < PCI_NUM_RESET_METHODS; m++) {
5272         if (sysfs_streq(name, pci_reset_fn_methods[m].name))
5273             return m;
5274     }
5275 
5276     return 0;   /* not found */
5277 }
5278 
5279 static ssize_t reset_method_store(struct device *dev,
5280                   struct device_attribute *attr,
5281                   const char *buf, size_t count)
5282 {
5283     struct pci_dev *pdev = to_pci_dev(dev);
5284     char *options, *name;
5285     int m, n;
5286     u8 reset_methods[PCI_NUM_RESET_METHODS] = { 0 };
5287 
5288     if (sysfs_streq(buf, "")) {
5289         pdev->reset_methods[0] = 0;
5290         pci_warn(pdev, "All device reset methods disabled by user");
5291         return count;
5292     }
5293 
5294     if (sysfs_streq(buf, "default")) {
5295         pci_init_reset_methods(pdev);
5296         return count;
5297     }
5298 
5299     options = kstrndup(buf, count, GFP_KERNEL);
5300     if (!options)
5301         return -ENOMEM;
5302 
5303     n = 0;
5304     while ((name = strsep(&options, " ")) != NULL) {
5305         if (sysfs_streq(name, ""))
5306             continue;
5307 
5308         name = strim(name);
5309 
5310         m = reset_method_lookup(name);
5311         if (!m) {
5312             pci_err(pdev, "Invalid reset method '%s'", name);
5313             goto error;
5314         }
5315 
5316         if (pci_reset_fn_methods[m].reset_fn(pdev, PCI_RESET_PROBE)) {
5317             pci_err(pdev, "Unsupported reset method '%s'", name);
5318             goto error;
5319         }
5320 
5321         if (n == PCI_NUM_RESET_METHODS - 1) {
5322             pci_err(pdev, "Too many reset methods\n");
5323             goto error;
5324         }
5325 
5326         reset_methods[n++] = m;
5327     }
5328 
5329     reset_methods[n] = 0;
5330 
5331     /* Warn if dev-specific supported but not highest priority */
5332     if (pci_reset_fn_methods[1].reset_fn(pdev, PCI_RESET_PROBE) == 0 &&
5333         reset_methods[0] != 1)
5334         pci_warn(pdev, "Device-specific reset disabled/de-prioritized by user");
5335     memcpy(pdev->reset_methods, reset_methods, sizeof(pdev->reset_methods));
5336     kfree(options);
5337     return count;
5338 
5339 error:
5340     /* Leave previous methods unchanged */
5341     kfree(options);
5342     return -EINVAL;
5343 }
5344 static DEVICE_ATTR_RW(reset_method);
5345 
5346 static struct attribute *pci_dev_reset_method_attrs[] = {
5347     &dev_attr_reset_method.attr,
5348     NULL,
5349 };
5350 
5351 static umode_t pci_dev_reset_method_attr_is_visible(struct kobject *kobj,
5352                             struct attribute *a, int n)
5353 {
5354     struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj));
5355 
5356     if (!pci_reset_supported(pdev))
5357         return 0;
5358 
5359     return a->mode;
5360 }
5361 
5362 const struct attribute_group pci_dev_reset_method_attr_group = {
5363     .attrs = pci_dev_reset_method_attrs,
5364     .is_visible = pci_dev_reset_method_attr_is_visible,
5365 };
5366 
5367 /**
5368  * __pci_reset_function_locked - reset a PCI device function while holding
5369  * the @dev mutex lock.
5370  * @dev: PCI device to reset
5371  *
5372  * Some devices allow an individual function to be reset without affecting
5373  * other functions in the same device.  The PCI device must be responsive
5374  * to PCI config space in order to use this function.
5375  *
5376  * The device function is presumed to be unused and the caller is holding
5377  * the device mutex lock when this function is called.
5378  *
5379  * Resetting the device will make the contents of PCI configuration space
5380  * random, so any caller of this must be prepared to reinitialise the
5381  * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
5382  * etc.
5383  *
5384  * Returns 0 if the device function was successfully reset or negative if the
5385  * device doesn't support resetting a single function.
5386  */
5387 int __pci_reset_function_locked(struct pci_dev *dev)
5388 {
5389     int i, m, rc;
5390 
5391     might_sleep();
5392 
5393     /*
5394      * A reset method returns -ENOTTY if it doesn't support this device and
5395      * we should try the next method.
5396      *
5397      * If it returns 0 (success), we're finished.  If it returns any other
5398      * error, we're also finished: this indicates that further reset
5399      * mechanisms might be broken on the device.
5400      */
5401     for (i = 0; i < PCI_NUM_RESET_METHODS; i++) {
5402         m = dev->reset_methods[i];
5403         if (!m)
5404             return -ENOTTY;
5405 
5406         rc = pci_reset_fn_methods[m].reset_fn(dev, PCI_RESET_DO_RESET);
5407         if (!rc)
5408             return 0;
5409         if (rc != -ENOTTY)
5410             return rc;
5411     }
5412 
5413     return -ENOTTY;
5414 }
5415 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
5416 
5417 /**
5418  * pci_init_reset_methods - check whether device can be safely reset
5419  * and store supported reset mechanisms.
5420  * @dev: PCI device to check for reset mechanisms
5421  *
5422  * Some devices allow an individual function to be reset without affecting
5423  * other functions in the same device.  The PCI device must be in D0-D3hot
5424  * state.
5425  *
5426  * Stores reset mechanisms supported by device in reset_methods byte array
5427  * which is a member of struct pci_dev.
5428  */
5429 void pci_init_reset_methods(struct pci_dev *dev)
5430 {
5431     int m, i, rc;
5432 
5433     BUILD_BUG_ON(ARRAY_SIZE(pci_reset_fn_methods) != PCI_NUM_RESET_METHODS);
5434 
5435     might_sleep();
5436 
5437     i = 0;
5438     for (m = 1; m < PCI_NUM_RESET_METHODS; m++) {
5439         rc = pci_reset_fn_methods[m].reset_fn(dev, PCI_RESET_PROBE);
5440         if (!rc)
5441             dev->reset_methods[i++] = m;
5442         else if (rc != -ENOTTY)
5443             break;
5444     }
5445 
5446     dev->reset_methods[i] = 0;
5447 }
5448 
5449 /**
5450  * pci_reset_function - quiesce and reset a PCI device function
5451  * @dev: PCI device to reset
5452  *
5453  * Some devices allow an individual function to be reset without affecting
5454  * other functions in the same device.  The PCI device must be responsive
5455  * to PCI config space in order to use this function.
5456  *
5457  * This function does not just reset the PCI portion of a device, but
5458  * clears all the state associated with the device.  This function differs
5459  * from __pci_reset_function_locked() in that it saves and restores device state
5460  * over the reset and takes the PCI device lock.
5461  *
5462  * Returns 0 if the device function was successfully reset or negative if the
5463  * device doesn't support resetting a single function.
5464  */
5465 int pci_reset_function(struct pci_dev *dev)
5466 {
5467     int rc;
5468 
5469     if (!pci_reset_supported(dev))
5470         return -ENOTTY;
5471 
5472     pci_dev_lock(dev);
5473     pci_dev_save_and_disable(dev);
5474 
5475     rc = __pci_reset_function_locked(dev);
5476 
5477     pci_dev_restore(dev);
5478     pci_dev_unlock(dev);
5479 
5480     return rc;
5481 }
5482 EXPORT_SYMBOL_GPL(pci_reset_function);
5483 
5484 /**
5485  * pci_reset_function_locked - quiesce and reset a PCI device function
5486  * @dev: PCI device to reset
5487  *
5488  * Some devices allow an individual function to be reset without affecting
5489  * other functions in the same device.  The PCI device must be responsive
5490  * to PCI config space in order to use this function.
5491  *
5492  * This function does not just reset the PCI portion of a device, but
5493  * clears all the state associated with the device.  This function differs
5494  * from __pci_reset_function_locked() in that it saves and restores device state
5495  * over the reset.  It also differs from pci_reset_function() in that it
5496  * requires the PCI device lock to be held.
5497  *
5498  * Returns 0 if the device function was successfully reset or negative if the
5499  * device doesn't support resetting a single function.
5500  */
5501 int pci_reset_function_locked(struct pci_dev *dev)
5502 {
5503     int rc;
5504 
5505     if (!pci_reset_supported(dev))
5506         return -ENOTTY;
5507 
5508     pci_dev_save_and_disable(dev);
5509 
5510     rc = __pci_reset_function_locked(dev);
5511 
5512     pci_dev_restore(dev);
5513 
5514     return rc;
5515 }
5516 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
5517 
5518 /**
5519  * pci_try_reset_function - quiesce and reset a PCI device function
5520  * @dev: PCI device to reset
5521  *
5522  * Same as above, except return -EAGAIN if unable to lock device.
5523  */
5524 int pci_try_reset_function(struct pci_dev *dev)
5525 {
5526     int rc;
5527 
5528     if (!pci_reset_supported(dev))
5529         return -ENOTTY;
5530 
5531     if (!pci_dev_trylock(dev))
5532         return -EAGAIN;
5533 
5534     pci_dev_save_and_disable(dev);
5535     rc = __pci_reset_function_locked(dev);
5536     pci_dev_restore(dev);
5537     pci_dev_unlock(dev);
5538 
5539     return rc;
5540 }
5541 EXPORT_SYMBOL_GPL(pci_try_reset_function);
5542 
5543 /* Do any devices on or below this bus prevent a bus reset? */
5544 static bool pci_bus_resetable(struct pci_bus *bus)
5545 {
5546     struct pci_dev *dev;
5547 
5548 
5549     if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5550         return false;
5551 
5552     list_for_each_entry(dev, &bus->devices, bus_list) {
5553         if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5554             (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5555             return false;
5556     }
5557 
5558     return true;
5559 }
5560 
5561 /* Lock devices from the top of the tree down */
5562 static void pci_bus_lock(struct pci_bus *bus)
5563 {
5564     struct pci_dev *dev;
5565 
5566     list_for_each_entry(dev, &bus->devices, bus_list) {
5567         pci_dev_lock(dev);
5568         if (dev->subordinate)
5569             pci_bus_lock(dev->subordinate);
5570     }
5571 }
5572 
5573 /* Unlock devices from the bottom of the tree up */
5574 static void pci_bus_unlock(struct pci_bus *bus)
5575 {
5576     struct pci_dev *dev;
5577 
5578     list_for_each_entry(dev, &bus->devices, bus_list) {
5579         if (dev->subordinate)
5580             pci_bus_unlock(dev->subordinate);
5581         pci_dev_unlock(dev);
5582     }
5583 }
5584 
5585 /* Return 1 on successful lock, 0 on contention */
5586 static int pci_bus_trylock(struct pci_bus *bus)
5587 {
5588     struct pci_dev *dev;
5589 
5590     list_for_each_entry(dev, &bus->devices, bus_list) {
5591         if (!pci_dev_trylock(dev))
5592             goto unlock;
5593         if (dev->subordinate) {
5594             if (!pci_bus_trylock(dev->subordinate)) {
5595                 pci_dev_unlock(dev);
5596                 goto unlock;
5597             }
5598         }
5599     }
5600     return 1;
5601 
5602 unlock:
5603     list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5604         if (dev->subordinate)
5605             pci_bus_unlock(dev->subordinate);
5606         pci_dev_unlock(dev);
5607     }
5608     return 0;
5609 }
5610 
5611 /* Do any devices on or below this slot prevent a bus reset? */
5612 static bool pci_slot_resetable(struct pci_slot *slot)
5613 {
5614     struct pci_dev *dev;
5615 
5616     if (slot->bus->self &&
5617         (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5618         return false;
5619 
5620     list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5621         if (!dev->slot || dev->slot != slot)
5622             continue;
5623         if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5624             (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5625             return false;
5626     }
5627 
5628     return true;
5629 }
5630 
5631 /* Lock devices from the top of the tree down */
5632 static void pci_slot_lock(struct pci_slot *slot)
5633 {
5634     struct pci_dev *dev;
5635 
5636     list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5637         if (!dev->slot || dev->slot != slot)
5638             continue;
5639         pci_dev_lock(dev);
5640         if (dev->subordinate)
5641             pci_bus_lock(dev->subordinate);
5642     }
5643 }
5644 
5645 /* Unlock devices from the bottom of the tree up */
5646 static void pci_slot_unlock(struct pci_slot *slot)
5647 {
5648     struct pci_dev *dev;
5649 
5650     list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5651         if (!dev->slot || dev->slot != slot)
5652             continue;
5653         if (dev->subordinate)
5654             pci_bus_unlock(dev->subordinate);
5655         pci_dev_unlock(dev);
5656     }
5657 }
5658 
5659 /* Return 1 on successful lock, 0 on contention */
5660 static int pci_slot_trylock(struct pci_slot *slot)
5661 {
5662     struct pci_dev *dev;
5663 
5664     list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5665         if (!dev->slot || dev->slot != slot)
5666             continue;
5667         if (!pci_dev_trylock(dev))
5668             goto unlock;
5669         if (dev->subordinate) {
5670             if (!pci_bus_trylock(dev->subordinate)) {
5671                 pci_dev_unlock(dev);
5672                 goto unlock;
5673             }
5674         }
5675     }
5676     return 1;
5677 
5678 unlock:
5679     list_for_each_entry_continue_reverse(dev,
5680                          &slot->bus->devices, bus_list) {
5681         if (!dev->slot || dev->slot != slot)
5682             continue;
5683         if (dev->subordinate)
5684             pci_bus_unlock(dev->subordinate);
5685         pci_dev_unlock(dev);
5686     }
5687     return 0;
5688 }
5689 
5690 /*
5691  * Save and disable devices from the top of the tree down while holding
5692  * the @dev mutex lock for the entire tree.
5693  */
5694 static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5695 {
5696     struct pci_dev *dev;
5697 
5698     list_for_each_entry(dev, &bus->devices, bus_list) {
5699         pci_dev_save_and_disable(dev);
5700         if (dev->subordinate)
5701             pci_bus_save_and_disable_locked(dev->subordinate);
5702     }
5703 }
5704 
5705 /*
5706  * Restore devices from top of the tree down while holding @dev mutex lock
5707  * for the entire tree.  Parent bridges need to be restored before we can
5708  * get to subordinate devices.
5709  */
5710 static void pci_bus_restore_locked(struct pci_bus *bus)
5711 {
5712     struct pci_dev *dev;
5713 
5714     list_for_each_entry(dev, &bus->devices, bus_list) {
5715         pci_dev_restore(dev);
5716         if (dev->subordinate)
5717             pci_bus_restore_locked(dev->subordinate);
5718     }
5719 }
5720 
5721 /*
5722  * Save and disable devices from the top of the tree down while holding
5723  * the @dev mutex lock for the entire tree.
5724  */
5725 static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5726 {
5727     struct pci_dev *dev;
5728 
5729     list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5730         if (!dev->slot || dev->slot != slot)
5731             continue;
5732         pci_dev_save_and_disable(dev);
5733         if (dev->subordinate)
5734             pci_bus_save_and_disable_locked(dev->subordinate);
5735     }
5736 }
5737 
5738 /*
5739  * Restore devices from top of the tree down while holding @dev mutex lock
5740  * for the entire tree.  Parent bridges need to be restored before we can
5741  * get to subordinate devices.
5742  */
5743 static void pci_slot_restore_locked(struct pci_slot *slot)
5744 {
5745     struct pci_dev *dev;
5746 
5747     list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5748         if (!dev->slot || dev->slot != slot)
5749             continue;
5750         pci_dev_restore(dev);
5751         if (dev->subordinate)
5752             pci_bus_restore_locked(dev->subordinate);
5753     }
5754 }
5755 
5756 static int pci_slot_reset(struct pci_slot *slot, bool probe)
5757 {
5758     int rc;
5759 
5760     if (!slot || !pci_slot_resetable(slot))
5761         return -ENOTTY;
5762 
5763     if (!probe)
5764         pci_slot_lock(slot);
5765 
5766     might_sleep();
5767 
5768     rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5769 
5770     if (!probe)
5771         pci_slot_unlock(slot);
5772 
5773     return rc;
5774 }
5775 
5776 /**
5777  * pci_probe_reset_slot - probe whether a PCI slot can be reset
5778  * @slot: PCI slot to probe
5779  *
5780  * Return 0 if slot can be reset, negative if a slot reset is not supported.
5781  */
5782 int pci_probe_reset_slot(struct pci_slot *slot)
5783 {
5784     return pci_slot_reset(slot, PCI_RESET_PROBE);
5785 }
5786 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5787 
5788 /**
5789  * __pci_reset_slot - Try to reset a PCI slot
5790  * @slot: PCI slot to reset
5791  *
5792  * A PCI bus may host multiple slots, each slot may support a reset mechanism
5793  * independent of other slots.  For instance, some slots may support slot power
5794  * control.  In the case of a 1:1 bus to slot architecture, this function may
5795  * wrap the bus reset to avoid spurious slot related events such as hotplug.
5796  * Generally a slot reset should be attempted before a bus reset.  All of the
5797  * function of the slot and any subordinate buses behind the slot are reset
5798  * through this function.  PCI config space of all devices in the slot and
5799  * behind the slot is saved before and restored after reset.
5800  *
5801  * Same as above except return -EAGAIN if the slot cannot be locked
5802  */
5803 static int __pci_reset_slot(struct pci_slot *slot)
5804 {
5805     int rc;
5806 
5807     rc = pci_slot_reset(slot, PCI_RESET_PROBE);
5808     if (rc)
5809         return rc;
5810 
5811     if (pci_slot_trylock(slot)) {
5812         pci_slot_save_and_disable_locked(slot);
5813         might_sleep();
5814         rc = pci_reset_hotplug_slot(slot->hotplug, PCI_RESET_DO_RESET);
5815         pci_slot_restore_locked(slot);
5816         pci_slot_unlock(slot);
5817     } else
5818         rc = -EAGAIN;
5819 
5820     return rc;
5821 }
5822 
5823 static int pci_bus_reset(struct pci_bus *bus, bool probe)
5824 {
5825     int ret;
5826 
5827     if (!bus->self || !pci_bus_resetable(bus))
5828         return -ENOTTY;
5829 
5830     if (probe)
5831         return 0;
5832 
5833     pci_bus_lock(bus);
5834 
5835     might_sleep();
5836 
5837     ret = pci_bridge_secondary_bus_reset(bus->self);
5838 
5839     pci_bus_unlock(bus);
5840 
5841     return ret;
5842 }
5843 
5844 /**
5845  * pci_bus_error_reset - reset the bridge's subordinate bus
5846  * @bridge: The parent device that connects to the bus to reset
5847  *
5848  * This function will first try to reset the slots on this bus if the method is
5849  * available. If slot reset fails or is not available, this will fall back to a
5850  * secondary bus reset.
5851  */
5852 int pci_bus_error_reset(struct pci_dev *bridge)
5853 {
5854     struct pci_bus *bus = bridge->subordinate;
5855     struct pci_slot *slot;
5856 
5857     if (!bus)
5858         return -ENOTTY;
5859 
5860     mutex_lock(&pci_slot_mutex);
5861     if (list_empty(&bus->slots))
5862         goto bus_reset;
5863 
5864     list_for_each_entry(slot, &bus->slots, list)
5865         if (pci_probe_reset_slot(slot))
5866             goto bus_reset;
5867 
5868     list_for_each_entry(slot, &bus->slots, list)
5869         if (pci_slot_reset(slot, PCI_RESET_DO_RESET))
5870             goto bus_reset;
5871 
5872     mutex_unlock(&pci_slot_mutex);
5873     return 0;
5874 bus_reset:
5875     mutex_unlock(&pci_slot_mutex);
5876     return pci_bus_reset(bridge->subordinate, PCI_RESET_DO_RESET);
5877 }
5878 
5879 /**
5880  * pci_probe_reset_bus - probe whether a PCI bus can be reset
5881  * @bus: PCI bus to probe
5882  *
5883  * Return 0 if bus can be reset, negative if a bus reset is not supported.
5884  */
5885 int pci_probe_reset_bus(struct pci_bus *bus)
5886 {
5887     return pci_bus_reset(bus, PCI_RESET_PROBE);
5888 }
5889 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5890 
5891 /**
5892  * __pci_reset_bus - Try to reset a PCI bus
5893  * @bus: top level PCI bus to reset
5894  *
5895  * Same as above except return -EAGAIN if the bus cannot be locked
5896  */
5897 static int __pci_reset_bus(struct pci_bus *bus)
5898 {
5899     int rc;
5900 
5901     rc = pci_bus_reset(bus, PCI_RESET_PROBE);
5902     if (rc)
5903         return rc;
5904 
5905     if (pci_bus_trylock(bus)) {
5906         pci_bus_save_and_disable_locked(bus);
5907         might_sleep();
5908         rc = pci_bridge_secondary_bus_reset(bus->self);
5909         pci_bus_restore_locked(bus);
5910         pci_bus_unlock(bus);
5911     } else
5912         rc = -EAGAIN;
5913 
5914     return rc;
5915 }
5916 
5917 /**
5918  * pci_reset_bus - Try to reset a PCI bus
5919  * @pdev: top level PCI device to reset via slot/bus
5920  *
5921  * Same as above except return -EAGAIN if the bus cannot be locked
5922  */
5923 int pci_reset_bus(struct pci_dev *pdev)
5924 {
5925     return (!pci_probe_reset_slot(pdev->slot)) ?
5926         __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5927 }
5928 EXPORT_SYMBOL_GPL(pci_reset_bus);
5929 
5930 /**
5931  * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5932  * @dev: PCI device to query
5933  *
5934  * Returns mmrbc: maximum designed memory read count in bytes or
5935  * appropriate error value.
5936  */
5937 int pcix_get_max_mmrbc(struct pci_dev *dev)
5938 {
5939     int cap;
5940     u32 stat;
5941 
5942     cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5943     if (!cap)
5944         return -EINVAL;
5945 
5946     if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5947         return -EINVAL;
5948 
5949     return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5950 }
5951 EXPORT_SYMBOL(pcix_get_max_mmrbc);
5952 
5953 /**
5954  * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5955  * @dev: PCI device to query
5956  *
5957  * Returns mmrbc: maximum memory read count in bytes or appropriate error
5958  * value.
5959  */
5960 int pcix_get_mmrbc(struct pci_dev *dev)
5961 {
5962     int cap;
5963     u16 cmd;
5964 
5965     cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5966     if (!cap)
5967         return -EINVAL;
5968 
5969     if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5970         return -EINVAL;
5971 
5972     return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5973 }
5974 EXPORT_SYMBOL(pcix_get_mmrbc);
5975 
5976 /**
5977  * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5978  * @dev: PCI device to query
5979  * @mmrbc: maximum memory read count in bytes
5980  *    valid values are 512, 1024, 2048, 4096
5981  *
5982  * If possible sets maximum memory read byte count, some bridges have errata
5983  * that prevent this.
5984  */
5985 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5986 {
5987     int cap;
5988     u32 stat, v, o;
5989     u16 cmd;
5990 
5991     if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5992         return -EINVAL;
5993 
5994     v = ffs(mmrbc) - 10;
5995 
5996     cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5997     if (!cap)
5998         return -EINVAL;
5999 
6000     if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
6001         return -EINVAL;
6002 
6003     if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
6004         return -E2BIG;
6005 
6006     if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
6007         return -EINVAL;
6008 
6009     o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
6010     if (o != v) {
6011         if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
6012             return -EIO;
6013 
6014         cmd &= ~PCI_X_CMD_MAX_READ;
6015         cmd |= v << 2;
6016         if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
6017             return -EIO;
6018     }
6019     return 0;
6020 }
6021 EXPORT_SYMBOL(pcix_set_mmrbc);
6022 
6023 /**
6024  * pcie_get_readrq - get PCI Express read request size
6025  * @dev: PCI device to query
6026  *
6027  * Returns maximum memory read request in bytes or appropriate error value.
6028  */
6029 int pcie_get_readrq(struct pci_dev *dev)
6030 {
6031     u16 ctl;
6032 
6033     pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
6034 
6035     return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
6036 }
6037 EXPORT_SYMBOL(pcie_get_readrq);
6038 
6039 /**
6040  * pcie_set_readrq - set PCI Express maximum memory read request
6041  * @dev: PCI device to query
6042  * @rq: maximum memory read count in bytes
6043  *    valid values are 128, 256, 512, 1024, 2048, 4096
6044  *
6045  * If possible sets maximum memory read request in bytes
6046  */
6047 int pcie_set_readrq(struct pci_dev *dev, int rq)
6048 {
6049     u16 v;
6050     int ret;
6051 
6052     if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
6053         return -EINVAL;
6054 
6055     /*
6056      * If using the "performance" PCIe config, we clamp the read rq
6057      * size to the max packet size to keep the host bridge from
6058      * generating requests larger than we can cope with.
6059      */
6060     if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
6061         int mps = pcie_get_mps(dev);
6062 
6063         if (mps < rq)
6064             rq = mps;
6065     }
6066 
6067     v = (ffs(rq) - 8) << 12;
6068 
6069     ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
6070                           PCI_EXP_DEVCTL_READRQ, v);
6071 
6072     return pcibios_err_to_errno(ret);
6073 }
6074 EXPORT_SYMBOL(pcie_set_readrq);
6075 
6076 /**
6077  * pcie_get_mps - get PCI Express maximum payload size
6078  * @dev: PCI device to query
6079  *
6080  * Returns maximum payload size in bytes
6081  */
6082 int pcie_get_mps(struct pci_dev *dev)
6083 {
6084     u16 ctl;
6085 
6086     pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
6087 
6088     return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
6089 }
6090 EXPORT_SYMBOL(pcie_get_mps);
6091 
6092 /**
6093  * pcie_set_mps - set PCI Express maximum payload size
6094  * @dev: PCI device to query
6095  * @mps: maximum payload size in bytes
6096  *    valid values are 128, 256, 512, 1024, 2048, 4096
6097  *
6098  * If possible sets maximum payload size
6099  */
6100 int pcie_set_mps(struct pci_dev *dev, int mps)
6101 {
6102     u16 v;
6103     int ret;
6104 
6105     if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
6106         return -EINVAL;
6107 
6108     v = ffs(mps) - 8;
6109     if (v > dev->pcie_mpss)
6110         return -EINVAL;
6111     v <<= 5;
6112 
6113     ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
6114                           PCI_EXP_DEVCTL_PAYLOAD, v);
6115 
6116     return pcibios_err_to_errno(ret);
6117 }
6118 EXPORT_SYMBOL(pcie_set_mps);
6119 
6120 /**
6121  * pcie_bandwidth_available - determine minimum link settings of a PCIe
6122  *                device and its bandwidth limitation
6123  * @dev: PCI device to query
6124  * @limiting_dev: storage for device causing the bandwidth limitation
6125  * @speed: storage for speed of limiting device
6126  * @width: storage for width of limiting device
6127  *
6128  * Walk up the PCI device chain and find the point where the minimum
6129  * bandwidth is available.  Return the bandwidth available there and (if
6130  * limiting_dev, speed, and width pointers are supplied) information about
6131  * that point.  The bandwidth returned is in Mb/s, i.e., megabits/second of
6132  * raw bandwidth.
6133  */
6134 u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
6135                  enum pci_bus_speed *speed,
6136                  enum pcie_link_width *width)
6137 {
6138     u16 lnksta;
6139     enum pci_bus_speed next_speed;
6140     enum pcie_link_width next_width;
6141     u32 bw, next_bw;
6142 
6143     if (speed)
6144         *speed = PCI_SPEED_UNKNOWN;
6145     if (width)
6146         *width = PCIE_LNK_WIDTH_UNKNOWN;
6147 
6148     bw = 0;
6149 
6150     while (dev) {
6151         pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
6152 
6153         next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
6154         next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
6155             PCI_EXP_LNKSTA_NLW_SHIFT;
6156 
6157         next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
6158 
6159         /* Check if current device limits the total bandwidth */
6160         if (!bw || next_bw <= bw) {
6161             bw = next_bw;
6162 
6163             if (limiting_dev)
6164                 *limiting_dev = dev;
6165             if (speed)
6166                 *speed = next_speed;
6167             if (width)
6168                 *width = next_width;
6169         }
6170 
6171         dev = pci_upstream_bridge(dev);
6172     }
6173 
6174     return bw;
6175 }
6176 EXPORT_SYMBOL(pcie_bandwidth_available);
6177 
6178 /**
6179  * pcie_get_speed_cap - query for the PCI device's link speed capability
6180  * @dev: PCI device to query
6181  *
6182  * Query the PCI device speed capability.  Return the maximum link speed
6183  * supported by the device.
6184  */
6185 enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
6186 {
6187     u32 lnkcap2, lnkcap;
6188 
6189     /*
6190      * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18.  The
6191      * implementation note there recommends using the Supported Link
6192      * Speeds Vector in Link Capabilities 2 when supported.
6193      *
6194      * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
6195      * should use the Supported Link Speeds field in Link Capabilities,
6196      * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
6197      */
6198     pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
6199 
6200     /* PCIe r3.0-compliant */
6201     if (lnkcap2)
6202         return PCIE_LNKCAP2_SLS2SPEED(lnkcap2);
6203 
6204     pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
6205     if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
6206         return PCIE_SPEED_5_0GT;
6207     else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
6208         return PCIE_SPEED_2_5GT;
6209 
6210     return PCI_SPEED_UNKNOWN;
6211 }
6212 EXPORT_SYMBOL(pcie_get_speed_cap);
6213 
6214 /**
6215  * pcie_get_width_cap - query for the PCI device's link width capability
6216  * @dev: PCI device to query
6217  *
6218  * Query the PCI device width capability.  Return the maximum link width
6219  * supported by the device.
6220  */
6221 enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
6222 {
6223     u32 lnkcap;
6224 
6225     pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
6226     if (lnkcap)
6227         return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
6228 
6229     return PCIE_LNK_WIDTH_UNKNOWN;
6230 }
6231 EXPORT_SYMBOL(pcie_get_width_cap);
6232 
6233 /**
6234  * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
6235  * @dev: PCI device
6236  * @speed: storage for link speed
6237  * @width: storage for link width
6238  *
6239  * Calculate a PCI device's link bandwidth by querying for its link speed
6240  * and width, multiplying them, and applying encoding overhead.  The result
6241  * is in Mb/s, i.e., megabits/second of raw bandwidth.
6242  */
6243 u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
6244                enum pcie_link_width *width)
6245 {
6246     *speed = pcie_get_speed_cap(dev);
6247     *width = pcie_get_width_cap(dev);
6248 
6249     if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
6250         return 0;
6251 
6252     return *width * PCIE_SPEED2MBS_ENC(*speed);
6253 }
6254 
6255 /**
6256  * __pcie_print_link_status - Report the PCI device's link speed and width
6257  * @dev: PCI device to query
6258  * @verbose: Print info even when enough bandwidth is available
6259  *
6260  * If the available bandwidth at the device is less than the device is
6261  * capable of, report the device's maximum possible bandwidth and the
6262  * upstream link that limits its performance.  If @verbose, always print
6263  * the available bandwidth, even if the device isn't constrained.
6264  */
6265 void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
6266 {
6267     enum pcie_link_width width, width_cap;
6268     enum pci_bus_speed speed, speed_cap;
6269     struct pci_dev *limiting_dev = NULL;
6270     u32 bw_avail, bw_cap;
6271 
6272     bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
6273     bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
6274 
6275     if (bw_avail >= bw_cap && verbose)
6276         pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
6277              bw_cap / 1000, bw_cap % 1000,
6278              pci_speed_string(speed_cap), width_cap);
6279     else if (bw_avail < bw_cap)
6280         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",
6281              bw_avail / 1000, bw_avail % 1000,
6282              pci_speed_string(speed), width,
6283              limiting_dev ? pci_name(limiting_dev) : "<unknown>",
6284              bw_cap / 1000, bw_cap % 1000,
6285              pci_speed_string(speed_cap), width_cap);
6286 }
6287 
6288 /**
6289  * pcie_print_link_status - Report the PCI device's link speed and width
6290  * @dev: PCI device to query
6291  *
6292  * Report the available bandwidth at the device.
6293  */
6294 void pcie_print_link_status(struct pci_dev *dev)
6295 {
6296     __pcie_print_link_status(dev, true);
6297 }
6298 EXPORT_SYMBOL(pcie_print_link_status);
6299 
6300 /**
6301  * pci_select_bars - Make BAR mask from the type of resource
6302  * @dev: the PCI device for which BAR mask is made
6303  * @flags: resource type mask to be selected
6304  *
6305  * This helper routine makes bar mask from the type of resource.
6306  */
6307 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
6308 {
6309     int i, bars = 0;
6310     for (i = 0; i < PCI_NUM_RESOURCES; i++)
6311         if (pci_resource_flags(dev, i) & flags)
6312             bars |= (1 << i);
6313     return bars;
6314 }
6315 EXPORT_SYMBOL(pci_select_bars);
6316 
6317 /* Some architectures require additional programming to enable VGA */
6318 static arch_set_vga_state_t arch_set_vga_state;
6319 
6320 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
6321 {
6322     arch_set_vga_state = func;  /* NULL disables */
6323 }
6324 
6325 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
6326                   unsigned int command_bits, u32 flags)
6327 {
6328     if (arch_set_vga_state)
6329         return arch_set_vga_state(dev, decode, command_bits,
6330                         flags);
6331     return 0;
6332 }
6333 
6334 /**
6335  * pci_set_vga_state - set VGA decode state on device and parents if requested
6336  * @dev: the PCI device
6337  * @decode: true = enable decoding, false = disable decoding
6338  * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
6339  * @flags: traverse ancestors and change bridges
6340  * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
6341  */
6342 int pci_set_vga_state(struct pci_dev *dev, bool decode,
6343               unsigned int command_bits, u32 flags)
6344 {
6345     struct pci_bus *bus;
6346     struct pci_dev *bridge;
6347     u16 cmd;
6348     int rc;
6349 
6350     WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
6351 
6352     /* ARCH specific VGA enables */
6353     rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
6354     if (rc)
6355         return rc;
6356 
6357     if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
6358         pci_read_config_word(dev, PCI_COMMAND, &cmd);
6359         if (decode)
6360             cmd |= command_bits;
6361         else
6362             cmd &= ~command_bits;
6363         pci_write_config_word(dev, PCI_COMMAND, cmd);
6364     }
6365 
6366     if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
6367         return 0;
6368 
6369     bus = dev->bus;
6370     while (bus) {
6371         bridge = bus->self;
6372         if (bridge) {
6373             pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
6374                          &cmd);
6375             if (decode)
6376                 cmd |= PCI_BRIDGE_CTL_VGA;
6377             else
6378                 cmd &= ~PCI_BRIDGE_CTL_VGA;
6379             pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
6380                           cmd);
6381         }
6382         bus = bus->parent;
6383     }
6384     return 0;
6385 }
6386 
6387 #ifdef CONFIG_ACPI
6388 bool pci_pr3_present(struct pci_dev *pdev)
6389 {
6390     struct acpi_device *adev;
6391 
6392     if (acpi_disabled)
6393         return false;
6394 
6395     adev = ACPI_COMPANION(&pdev->dev);
6396     if (!adev)
6397         return false;
6398 
6399     return adev->power.flags.power_resources &&
6400         acpi_has_method(adev->handle, "_PR3");
6401 }
6402 EXPORT_SYMBOL_GPL(pci_pr3_present);
6403 #endif
6404 
6405 /**
6406  * pci_add_dma_alias - Add a DMA devfn alias for a device
6407  * @dev: the PCI device for which alias is added
6408  * @devfn_from: alias slot and function
6409  * @nr_devfns: number of subsequent devfns to alias
6410  *
6411  * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
6412  * which is used to program permissible bus-devfn source addresses for DMA
6413  * requests in an IOMMU.  These aliases factor into IOMMU group creation
6414  * and are useful for devices generating DMA requests beyond or different
6415  * from their logical bus-devfn.  Examples include device quirks where the
6416  * device simply uses the wrong devfn, as well as non-transparent bridges
6417  * where the alias may be a proxy for devices in another domain.
6418  *
6419  * IOMMU group creation is performed during device discovery or addition,
6420  * prior to any potential DMA mapping and therefore prior to driver probing
6421  * (especially for userspace assigned devices where IOMMU group definition
6422  * cannot be left as a userspace activity).  DMA aliases should therefore
6423  * be configured via quirks, such as the PCI fixup header quirk.
6424  */
6425 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn_from,
6426                unsigned int nr_devfns)
6427 {
6428     int devfn_to;
6429 
6430     nr_devfns = min(nr_devfns, (unsigned int)MAX_NR_DEVFNS - devfn_from);
6431     devfn_to = devfn_from + nr_devfns - 1;
6432 
6433     if (!dev->dma_alias_mask)
6434         dev->dma_alias_mask = bitmap_zalloc(MAX_NR_DEVFNS, GFP_KERNEL);
6435     if (!dev->dma_alias_mask) {
6436         pci_warn(dev, "Unable to allocate DMA alias mask\n");
6437         return;
6438     }
6439 
6440     bitmap_set(dev->dma_alias_mask, devfn_from, nr_devfns);
6441 
6442     if (nr_devfns == 1)
6443         pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
6444                 PCI_SLOT(devfn_from), PCI_FUNC(devfn_from));
6445     else if (nr_devfns > 1)
6446         pci_info(dev, "Enabling fixed DMA alias for devfn range from %02x.%d to %02x.%d\n",
6447                 PCI_SLOT(devfn_from), PCI_FUNC(devfn_from),
6448                 PCI_SLOT(devfn_to), PCI_FUNC(devfn_to));
6449 }
6450 
6451 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
6452 {
6453     return (dev1->dma_alias_mask &&
6454         test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
6455            (dev2->dma_alias_mask &&
6456         test_bit(dev1->devfn, dev2->dma_alias_mask)) ||
6457            pci_real_dma_dev(dev1) == dev2 ||
6458            pci_real_dma_dev(dev2) == dev1;
6459 }
6460 
6461 bool pci_device_is_present(struct pci_dev *pdev)
6462 {
6463     u32 v;
6464 
6465     if (pci_dev_is_disconnected(pdev))
6466         return false;
6467     return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
6468 }
6469 EXPORT_SYMBOL_GPL(pci_device_is_present);
6470 
6471 void pci_ignore_hotplug(struct pci_dev *dev)
6472 {
6473     struct pci_dev *bridge = dev->bus->self;
6474 
6475     dev->ignore_hotplug = 1;
6476     /* Propagate the "ignore hotplug" setting to the parent bridge. */
6477     if (bridge)
6478         bridge->ignore_hotplug = 1;
6479 }
6480 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
6481 
6482 /**
6483  * pci_real_dma_dev - Get PCI DMA device for PCI device
6484  * @dev: the PCI device that may have a PCI DMA alias
6485  *
6486  * Permits the platform to provide architecture-specific functionality to
6487  * devices needing to alias DMA to another PCI device on another PCI bus. If
6488  * the PCI device is on the same bus, it is recommended to use
6489  * pci_add_dma_alias(). This is the default implementation. Architecture
6490  * implementations can override this.
6491  */
6492 struct pci_dev __weak *pci_real_dma_dev(struct pci_dev *dev)
6493 {
6494     return dev;
6495 }
6496 
6497 resource_size_t __weak pcibios_default_alignment(void)
6498 {
6499     return 0;
6500 }
6501 
6502 /*
6503  * Arches that don't want to expose struct resource to userland as-is in
6504  * sysfs and /proc can implement their own pci_resource_to_user().
6505  */
6506 void __weak pci_resource_to_user(const struct pci_dev *dev, int bar,
6507                  const struct resource *rsrc,
6508                  resource_size_t *start, resource_size_t *end)
6509 {
6510     *start = rsrc->start;
6511     *end = rsrc->end;
6512 }
6513 
6514 static char *resource_alignment_param;
6515 static DEFINE_SPINLOCK(resource_alignment_lock);
6516 
6517 /**
6518  * pci_specified_resource_alignment - get resource alignment specified by user.
6519  * @dev: the PCI device to get
6520  * @resize: whether or not to change resources' size when reassigning alignment
6521  *
6522  * RETURNS: Resource alignment if it is specified.
6523  *          Zero if it is not specified.
6524  */
6525 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
6526                             bool *resize)
6527 {
6528     int align_order, count;
6529     resource_size_t align = pcibios_default_alignment();
6530     const char *p;
6531     int ret;
6532 
6533     spin_lock(&resource_alignment_lock);
6534     p = resource_alignment_param;
6535     if (!p || !*p)
6536         goto out;
6537     if (pci_has_flag(PCI_PROBE_ONLY)) {
6538         align = 0;
6539         pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
6540         goto out;
6541     }
6542 
6543     while (*p) {
6544         count = 0;
6545         if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
6546             p[count] == '@') {
6547             p += count + 1;
6548             if (align_order > 63) {
6549                 pr_err("PCI: Invalid requested alignment (order %d)\n",
6550                        align_order);
6551                 align_order = PAGE_SHIFT;
6552             }
6553         } else {
6554             align_order = PAGE_SHIFT;
6555         }
6556 
6557         ret = pci_dev_str_match(dev, p, &p);
6558         if (ret == 1) {
6559             *resize = true;
6560             align = 1ULL << align_order;
6561             break;
6562         } else if (ret < 0) {
6563             pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
6564                    p);
6565             break;
6566         }
6567 
6568         if (*p != ';' && *p != ',') {
6569             /* End of param or invalid format */
6570             break;
6571         }
6572         p++;
6573     }
6574 out:
6575     spin_unlock(&resource_alignment_lock);
6576     return align;
6577 }
6578 
6579 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
6580                        resource_size_t align, bool resize)
6581 {
6582     struct resource *r = &dev->resource[bar];
6583     resource_size_t size;
6584 
6585     if (!(r->flags & IORESOURCE_MEM))
6586         return;
6587 
6588     if (r->flags & IORESOURCE_PCI_FIXED) {
6589         pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
6590              bar, r, (unsigned long long)align);
6591         return;
6592     }
6593 
6594     size = resource_size(r);
6595     if (size >= align)
6596         return;
6597 
6598     /*
6599      * Increase the alignment of the resource.  There are two ways we
6600      * can do this:
6601      *
6602      * 1) Increase the size of the resource.  BARs are aligned on their
6603      *    size, so when we reallocate space for this resource, we'll
6604      *    allocate it with the larger alignment.  This also prevents
6605      *    assignment of any other BARs inside the alignment region, so
6606      *    if we're requesting page alignment, this means no other BARs
6607      *    will share the page.
6608      *
6609      *    The disadvantage is that this makes the resource larger than
6610      *    the hardware BAR, which may break drivers that compute things
6611      *    based on the resource size, e.g., to find registers at a
6612      *    fixed offset before the end of the BAR.
6613      *
6614      * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
6615      *    set r->start to the desired alignment.  By itself this
6616      *    doesn't prevent other BARs being put inside the alignment
6617      *    region, but if we realign *every* resource of every device in
6618      *    the system, none of them will share an alignment region.
6619      *
6620      * When the user has requested alignment for only some devices via
6621      * the "pci=resource_alignment" argument, "resize" is true and we
6622      * use the first method.  Otherwise we assume we're aligning all
6623      * devices and we use the second.
6624      */
6625 
6626     pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
6627          bar, r, (unsigned long long)align);
6628 
6629     if (resize) {
6630         r->start = 0;
6631         r->end = align - 1;
6632     } else {
6633         r->flags &= ~IORESOURCE_SIZEALIGN;
6634         r->flags |= IORESOURCE_STARTALIGN;
6635         r->start = align;
6636         r->end = r->start + size - 1;
6637     }
6638     r->flags |= IORESOURCE_UNSET;
6639 }
6640 
6641 /*
6642  * This function disables memory decoding and releases memory resources
6643  * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6644  * It also rounds up size to specified alignment.
6645  * Later on, the kernel will assign page-aligned memory resource back
6646  * to the device.
6647  */
6648 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6649 {
6650     int i;
6651     struct resource *r;
6652     resource_size_t align;
6653     u16 command;
6654     bool resize = false;
6655 
6656     /*
6657      * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6658      * 3.4.1.11.  Their resources are allocated from the space
6659      * described by the VF BARx register in the PF's SR-IOV capability.
6660      * We can't influence their alignment here.
6661      */
6662     if (dev->is_virtfn)
6663         return;
6664 
6665     /* check if specified PCI is target device to reassign */
6666     align = pci_specified_resource_alignment(dev, &resize);
6667     if (!align)
6668         return;
6669 
6670     if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6671         (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6672         pci_warn(dev, "Can't reassign resources to host bridge\n");
6673         return;
6674     }
6675 
6676     pci_read_config_word(dev, PCI_COMMAND, &command);
6677     command &= ~PCI_COMMAND_MEMORY;
6678     pci_write_config_word(dev, PCI_COMMAND, command);
6679 
6680     for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6681         pci_request_resource_alignment(dev, i, align, resize);
6682 
6683     /*
6684      * Need to disable bridge's resource window,
6685      * to enable the kernel to reassign new resource
6686      * window later on.
6687      */
6688     if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6689         for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6690             r = &dev->resource[i];
6691             if (!(r->flags & IORESOURCE_MEM))
6692                 continue;
6693             r->flags |= IORESOURCE_UNSET;
6694             r->end = resource_size(r) - 1;
6695             r->start = 0;
6696         }
6697         pci_disable_bridge_window(dev);
6698     }
6699 }
6700 
6701 static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
6702 {
6703     size_t count = 0;
6704 
6705     spin_lock(&resource_alignment_lock);
6706     if (resource_alignment_param)
6707         count = sysfs_emit(buf, "%s\n", resource_alignment_param);
6708     spin_unlock(&resource_alignment_lock);
6709 
6710     return count;
6711 }
6712 
6713 static ssize_t resource_alignment_store(struct bus_type *bus,
6714                     const char *buf, size_t count)
6715 {
6716     char *param, *old, *end;
6717 
6718     if (count >= (PAGE_SIZE - 1))
6719         return -EINVAL;
6720 
6721     param = kstrndup(buf, count, GFP_KERNEL);
6722     if (!param)
6723         return -ENOMEM;
6724 
6725     end = strchr(param, '\n');
6726     if (end)
6727         *end = '\0';
6728 
6729     spin_lock(&resource_alignment_lock);
6730     old = resource_alignment_param;
6731     if (strlen(param)) {
6732         resource_alignment_param = param;
6733     } else {
6734         kfree(param);
6735         resource_alignment_param = NULL;
6736     }
6737     spin_unlock(&resource_alignment_lock);
6738 
6739     kfree(old);
6740 
6741     return count;
6742 }
6743 
6744 static BUS_ATTR_RW(resource_alignment);
6745 
6746 static int __init pci_resource_alignment_sysfs_init(void)
6747 {
6748     return bus_create_file(&pci_bus_type,
6749                     &bus_attr_resource_alignment);
6750 }
6751 late_initcall(pci_resource_alignment_sysfs_init);
6752 
6753 static void pci_no_domains(void)
6754 {
6755 #ifdef CONFIG_PCI_DOMAINS
6756     pci_domains_supported = 0;
6757 #endif
6758 }
6759 
6760 #ifdef CONFIG_PCI_DOMAINS_GENERIC
6761 static atomic_t __domain_nr = ATOMIC_INIT(-1);
6762 
6763 static int pci_get_new_domain_nr(void)
6764 {
6765     return atomic_inc_return(&__domain_nr);
6766 }
6767 
6768 static int of_pci_bus_find_domain_nr(struct device *parent)
6769 {
6770     static int use_dt_domains = -1;
6771     int domain = -1;
6772 
6773     if (parent)
6774         domain = of_get_pci_domain_nr(parent->of_node);
6775 
6776     /*
6777      * Check DT domain and use_dt_domains values.
6778      *
6779      * If DT domain property is valid (domain >= 0) and
6780      * use_dt_domains != 0, the DT assignment is valid since this means
6781      * we have not previously allocated a domain number by using
6782      * pci_get_new_domain_nr(); we should also update use_dt_domains to
6783      * 1, to indicate that we have just assigned a domain number from
6784      * DT.
6785      *
6786      * If DT domain property value is not valid (ie domain < 0), and we
6787      * have not previously assigned a domain number from DT
6788      * (use_dt_domains != 1) we should assign a domain number by
6789      * using the:
6790      *
6791      * pci_get_new_domain_nr()
6792      *
6793      * API and update the use_dt_domains value to keep track of method we
6794      * are using to assign domain numbers (use_dt_domains = 0).
6795      *
6796      * All other combinations imply we have a platform that is trying
6797      * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6798      * which is a recipe for domain mishandling and it is prevented by
6799      * invalidating the domain value (domain = -1) and printing a
6800      * corresponding error.
6801      */
6802     if (domain >= 0 && use_dt_domains) {
6803         use_dt_domains = 1;
6804     } else if (domain < 0 && use_dt_domains != 1) {
6805         use_dt_domains = 0;
6806         domain = pci_get_new_domain_nr();
6807     } else {
6808         if (parent)
6809             pr_err("Node %pOF has ", parent->of_node);
6810         pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6811         domain = -1;
6812     }
6813 
6814     return domain;
6815 }
6816 
6817 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6818 {
6819     return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6820                    acpi_pci_bus_find_domain_nr(bus);
6821 }
6822 #endif
6823 
6824 /**
6825  * pci_ext_cfg_avail - can we access extended PCI config space?
6826  *
6827  * Returns 1 if we can access PCI extended config space (offsets
6828  * greater than 0xff). This is the default implementation. Architecture
6829  * implementations can override this.
6830  */
6831 int __weak pci_ext_cfg_avail(void)
6832 {
6833     return 1;
6834 }
6835 
6836 void __weak pci_fixup_cardbus(struct pci_bus *bus)
6837 {
6838 }
6839 EXPORT_SYMBOL(pci_fixup_cardbus);
6840 
6841 static int __init pci_setup(char *str)
6842 {
6843     while (str) {
6844         char *k = strchr(str, ',');
6845         if (k)
6846             *k++ = 0;
6847         if (*str && (str = pcibios_setup(str)) && *str) {
6848             if (!strcmp(str, "nomsi")) {
6849                 pci_no_msi();
6850             } else if (!strncmp(str, "noats", 5)) {
6851                 pr_info("PCIe: ATS is disabled\n");
6852                 pcie_ats_disabled = true;
6853             } else if (!strcmp(str, "noaer")) {
6854                 pci_no_aer();
6855             } else if (!strcmp(str, "earlydump")) {
6856                 pci_early_dump = true;
6857             } else if (!strncmp(str, "realloc=", 8)) {
6858                 pci_realloc_get_opt(str + 8);
6859             } else if (!strncmp(str, "realloc", 7)) {
6860                 pci_realloc_get_opt("on");
6861             } else if (!strcmp(str, "nodomains")) {
6862                 pci_no_domains();
6863             } else if (!strncmp(str, "noari", 5)) {
6864                 pcie_ari_disabled = true;
6865             } else if (!strncmp(str, "cbiosize=", 9)) {
6866                 pci_cardbus_io_size = memparse(str + 9, &str);
6867             } else if (!strncmp(str, "cbmemsize=", 10)) {
6868                 pci_cardbus_mem_size = memparse(str + 10, &str);
6869             } else if (!strncmp(str, "resource_alignment=", 19)) {
6870                 resource_alignment_param = str + 19;
6871             } else if (!strncmp(str, "ecrc=", 5)) {
6872                 pcie_ecrc_get_policy(str + 5);
6873             } else if (!strncmp(str, "hpiosize=", 9)) {
6874                 pci_hotplug_io_size = memparse(str + 9, &str);
6875             } else if (!strncmp(str, "hpmmiosize=", 11)) {
6876                 pci_hotplug_mmio_size = memparse(str + 11, &str);
6877             } else if (!strncmp(str, "hpmmioprefsize=", 15)) {
6878                 pci_hotplug_mmio_pref_size = memparse(str + 15, &str);
6879             } else if (!strncmp(str, "hpmemsize=", 10)) {
6880                 pci_hotplug_mmio_size = memparse(str + 10, &str);
6881                 pci_hotplug_mmio_pref_size = pci_hotplug_mmio_size;
6882             } else if (!strncmp(str, "hpbussize=", 10)) {
6883                 pci_hotplug_bus_size =
6884                     simple_strtoul(str + 10, &str, 0);
6885                 if (pci_hotplug_bus_size > 0xff)
6886                     pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6887             } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6888                 pcie_bus_config = PCIE_BUS_TUNE_OFF;
6889             } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6890                 pcie_bus_config = PCIE_BUS_SAFE;
6891             } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6892                 pcie_bus_config = PCIE_BUS_PERFORMANCE;
6893             } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6894                 pcie_bus_config = PCIE_BUS_PEER2PEER;
6895             } else if (!strncmp(str, "pcie_scan_all", 13)) {
6896                 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6897             } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6898                 disable_acs_redir_param = str + 18;
6899             } else {
6900                 pr_err("PCI: Unknown option `%s'\n", str);
6901             }
6902         }
6903         str = k;
6904     }
6905     return 0;
6906 }
6907 early_param("pci", pci_setup);
6908 
6909 /*
6910  * 'resource_alignment_param' and 'disable_acs_redir_param' are initialized
6911  * in pci_setup(), above, to point to data in the __initdata section which
6912  * will be freed after the init sequence is complete. We can't allocate memory
6913  * in pci_setup() because some architectures do not have any memory allocation
6914  * service available during an early_param() call. So we allocate memory and
6915  * copy the variable here before the init section is freed.
6916  *
6917  */
6918 static int __init pci_realloc_setup_params(void)
6919 {
6920     resource_alignment_param = kstrdup(resource_alignment_param,
6921                        GFP_KERNEL);
6922     disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6923 
6924     return 0;
6925 }
6926 pure_initcall(pci_realloc_setup_params);