1 // SPDX-License-Identifier: GPL-2.0
3 * PCI Bus Services, see include/linux/pci.h for further explanation.
5 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
11 #include <linux/acpi.h>
12 #include <linux/kernel.h>
13 #include <linux/delay.h>
14 #include <linux/dmi.h>
15 #include <linux/init.h>
17 #include <linux/of_pci.h>
18 #include <linux/pci.h>
20 #include <linux/slab.h>
21 #include <linux/module.h>
22 #include <linux/spinlock.h>
23 #include <linux/string.h>
24 #include <linux/log2.h>
25 #include <linux/logic_pio.h>
26 #include <linux/pm_wakeup.h>
27 #include <linux/interrupt.h>
28 #include <linux/device.h>
29 #include <linux/pm_runtime.h>
30 #include <linux/pci_hotplug.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pci-ats.h>
33 #include <asm/setup.h>
35 #include <linux/aer.h>
38 DEFINE_MUTEX(pci_slot_mutex);
40 const char *pci_power_names[] = {
41 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
43 EXPORT_SYMBOL_GPL(pci_power_names);
45 int isa_dma_bridge_buggy;
46 EXPORT_SYMBOL(isa_dma_bridge_buggy);
49 EXPORT_SYMBOL(pci_pci_problems);
51 unsigned int pci_pm_d3_delay;
53 static void pci_pme_list_scan(struct work_struct *work);
55 static LIST_HEAD(pci_pme_list);
56 static DEFINE_MUTEX(pci_pme_list_mutex);
57 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
59 struct pci_pme_device {
60 struct list_head list;
64 #define PME_TIMEOUT 1000 /* How long between PME checks */
66 static void pci_dev_d3_sleep(struct pci_dev *dev)
68 unsigned int delay = dev->d3_delay;
70 if (delay < pci_pm_d3_delay)
71 delay = pci_pm_d3_delay;
77 #ifdef CONFIG_PCI_DOMAINS
78 int pci_domains_supported = 1;
81 #define DEFAULT_CARDBUS_IO_SIZE (256)
82 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
83 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
84 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
85 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
87 #define DEFAULT_HOTPLUG_IO_SIZE (256)
88 #define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
89 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
90 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
91 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
93 #define DEFAULT_HOTPLUG_BUS_SIZE 1
94 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
96 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
99 * The default CLS is used if arch didn't set CLS explicitly and not
100 * all pci devices agree on the same value. Arch can override either
101 * the dfl or actual value as it sees fit. Don't forget this is
102 * measured in 32-bit words, not bytes.
104 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
105 u8 pci_cache_line_size;
108 * If we set up a device for bus mastering, we need to check the latency
109 * timer as certain BIOSes forget to set it properly.
111 unsigned int pcibios_max_latency = 255;
113 /* If set, the PCIe ARI capability will not be used. */
114 static bool pcie_ari_disabled;
116 /* If set, the PCIe ATS capability will not be used. */
117 static bool pcie_ats_disabled;
119 /* If set, the PCI config space of each device is printed during boot. */
122 bool pci_ats_disabled(void)
124 return pcie_ats_disabled;
127 /* Disable bridge_d3 for all PCIe ports */
128 static bool pci_bridge_d3_disable;
129 /* Force bridge_d3 for all PCIe ports */
130 static bool pci_bridge_d3_force;
132 static int __init pcie_port_pm_setup(char *str)
134 if (!strcmp(str, "off"))
135 pci_bridge_d3_disable = true;
136 else if (!strcmp(str, "force"))
137 pci_bridge_d3_force = true;
140 __setup("pcie_port_pm=", pcie_port_pm_setup);
142 /* Time to wait after a reset for device to become responsive */
143 #define PCIE_RESET_READY_POLL_MS 60000
146 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
147 * @bus: pointer to PCI bus structure to search
149 * Given a PCI bus, returns the highest PCI bus number present in the set
150 * including the given PCI bus and its list of child PCI buses.
152 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
155 unsigned char max, n;
157 max = bus->busn_res.end;
158 list_for_each_entry(tmp, &bus->children, node) {
159 n = pci_bus_max_busnr(tmp);
165 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
167 #ifdef CONFIG_HAS_IOMEM
168 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
170 struct resource *res = &pdev->resource[bar];
173 * Make sure the BAR is actually a memory resource, not an IO resource
175 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
176 pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
179 return ioremap_nocache(res->start, resource_size(res));
181 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
183 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
186 * Make sure the BAR is actually a memory resource, not an IO resource
188 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
192 return ioremap_wc(pci_resource_start(pdev, bar),
193 pci_resource_len(pdev, bar));
195 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
199 * pci_dev_str_match_path - test if a path string matches a device
200 * @dev: the PCI device to test
201 * @path: string to match the device against
202 * @endptr: pointer to the string after the match
204 * Test if a string (typically from a kernel parameter) formatted as a
205 * path of device/function addresses matches a PCI device. The string must
208 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
210 * A path for a device can be obtained using 'lspci -t'. Using a path
211 * is more robust against bus renumbering than using only a single bus,
212 * device and function address.
214 * Returns 1 if the string matches the device, 0 if it does not and
215 * a negative error code if it fails to parse the string.
217 static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
221 int seg, bus, slot, func;
225 *endptr = strchrnul(path, ';');
227 wpath = kmemdup_nul(path, *endptr - path, GFP_KERNEL);
232 p = strrchr(wpath, '/');
235 ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
241 if (dev->devfn != PCI_DEVFN(slot, func)) {
247 * Note: we don't need to get a reference to the upstream
248 * bridge because we hold a reference to the top level
249 * device which should hold a reference to the bridge,
252 dev = pci_upstream_bridge(dev);
261 ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
265 ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
272 ret = (seg == pci_domain_nr(dev->bus) &&
273 bus == dev->bus->number &&
274 dev->devfn == PCI_DEVFN(slot, func));
282 * pci_dev_str_match - test if a string matches a device
283 * @dev: the PCI device to test
284 * @p: string to match the device against
285 * @endptr: pointer to the string after the match
287 * Test if a string (typically from a kernel parameter) matches a specified
288 * PCI device. The string may be of one of the following formats:
290 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
291 * pci:<vendor>:<device>[:<subvendor>:<subdevice>]
293 * The first format specifies a PCI bus/device/function address which
294 * may change if new hardware is inserted, if motherboard firmware changes,
295 * or due to changes caused in kernel parameters. If the domain is
296 * left unspecified, it is taken to be 0. In order to be robust against
297 * bus renumbering issues, a path of PCI device/function numbers may be used
298 * to address the specific device. The path for a device can be determined
299 * through the use of 'lspci -t'.
301 * The second format matches devices using IDs in the configuration
302 * space which may match multiple devices in the system. A value of 0
303 * for any field will match all devices. (Note: this differs from
304 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
305 * legacy reasons and convenience so users don't have to specify
306 * FFFFFFFFs on the command line.)
308 * Returns 1 if the string matches the device, 0 if it does not and
309 * a negative error code if the string cannot be parsed.
311 static int pci_dev_str_match(struct pci_dev *dev, const char *p,
316 unsigned short vendor, device, subsystem_vendor, subsystem_device;
318 if (strncmp(p, "pci:", 4) == 0) {
319 /* PCI vendor/device (subvendor/subdevice) IDs are specified */
321 ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
322 &subsystem_vendor, &subsystem_device, &count);
324 ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
328 subsystem_vendor = 0;
329 subsystem_device = 0;
334 if ((!vendor || vendor == dev->vendor) &&
335 (!device || device == dev->device) &&
336 (!subsystem_vendor ||
337 subsystem_vendor == dev->subsystem_vendor) &&
338 (!subsystem_device ||
339 subsystem_device == dev->subsystem_device))
343 * PCI Bus, Device, Function IDs are specified
344 * (optionally, may include a path of devfns following it)
346 ret = pci_dev_str_match_path(dev, p, &p);
361 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
362 u8 pos, int cap, int *ttl)
367 pci_bus_read_config_byte(bus, devfn, pos, &pos);
373 pci_bus_read_config_word(bus, devfn, pos, &ent);
385 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
388 int ttl = PCI_FIND_CAP_TTL;
390 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
393 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
395 return __pci_find_next_cap(dev->bus, dev->devfn,
396 pos + PCI_CAP_LIST_NEXT, cap);
398 EXPORT_SYMBOL_GPL(pci_find_next_capability);
400 static int __pci_bus_find_cap_start(struct pci_bus *bus,
401 unsigned int devfn, u8 hdr_type)
405 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
406 if (!(status & PCI_STATUS_CAP_LIST))
410 case PCI_HEADER_TYPE_NORMAL:
411 case PCI_HEADER_TYPE_BRIDGE:
412 return PCI_CAPABILITY_LIST;
413 case PCI_HEADER_TYPE_CARDBUS:
414 return PCI_CB_CAPABILITY_LIST;
421 * pci_find_capability - query for devices' capabilities
422 * @dev: PCI device to query
423 * @cap: capability code
425 * Tell if a device supports a given PCI capability.
426 * Returns the address of the requested capability structure within the
427 * device's PCI configuration space or 0 in case the device does not
428 * support it. Possible values for @cap include:
430 * %PCI_CAP_ID_PM Power Management
431 * %PCI_CAP_ID_AGP Accelerated Graphics Port
432 * %PCI_CAP_ID_VPD Vital Product Data
433 * %PCI_CAP_ID_SLOTID Slot Identification
434 * %PCI_CAP_ID_MSI Message Signalled Interrupts
435 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
436 * %PCI_CAP_ID_PCIX PCI-X
437 * %PCI_CAP_ID_EXP PCI Express
439 int pci_find_capability(struct pci_dev *dev, int cap)
443 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
445 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
449 EXPORT_SYMBOL(pci_find_capability);
452 * pci_bus_find_capability - query for devices' capabilities
453 * @bus: the PCI bus to query
454 * @devfn: PCI device to query
455 * @cap: capability code
457 * Like pci_find_capability() but works for PCI devices that do not have a
458 * pci_dev structure set up yet.
460 * Returns the address of the requested capability structure within the
461 * device's PCI configuration space or 0 in case the device does not
464 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
469 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
471 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
473 pos = __pci_find_next_cap(bus, devfn, pos, cap);
477 EXPORT_SYMBOL(pci_bus_find_capability);
480 * pci_find_next_ext_capability - Find an extended capability
481 * @dev: PCI device to query
482 * @start: address at which to start looking (0 to start at beginning of list)
483 * @cap: capability code
485 * Returns the address of the next matching extended capability structure
486 * within the device's PCI configuration space or 0 if the device does
487 * not support it. Some capabilities can occur several times, e.g., the
488 * vendor-specific capability, and this provides a way to find them all.
490 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
494 int pos = PCI_CFG_SPACE_SIZE;
496 /* minimum 8 bytes per capability */
497 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
499 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
505 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
509 * If we have no capabilities, this is indicated by cap ID,
510 * cap version and next pointer all being 0.
516 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
519 pos = PCI_EXT_CAP_NEXT(header);
520 if (pos < PCI_CFG_SPACE_SIZE)
523 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
529 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
532 * pci_find_ext_capability - Find an extended capability
533 * @dev: PCI device to query
534 * @cap: capability code
536 * Returns the address of the requested extended capability structure
537 * within the device's PCI configuration space or 0 if the device does
538 * not support it. Possible values for @cap include:
540 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
541 * %PCI_EXT_CAP_ID_VC Virtual Channel
542 * %PCI_EXT_CAP_ID_DSN Device Serial Number
543 * %PCI_EXT_CAP_ID_PWR Power Budgeting
545 int pci_find_ext_capability(struct pci_dev *dev, int cap)
547 return pci_find_next_ext_capability(dev, 0, cap);
549 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
551 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
553 int rc, ttl = PCI_FIND_CAP_TTL;
556 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
557 mask = HT_3BIT_CAP_MASK;
559 mask = HT_5BIT_CAP_MASK;
561 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
562 PCI_CAP_ID_HT, &ttl);
564 rc = pci_read_config_byte(dev, pos + 3, &cap);
565 if (rc != PCIBIOS_SUCCESSFUL)
568 if ((cap & mask) == ht_cap)
571 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
572 pos + PCI_CAP_LIST_NEXT,
573 PCI_CAP_ID_HT, &ttl);
579 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
580 * @dev: PCI device to query
581 * @pos: Position from which to continue searching
582 * @ht_cap: Hypertransport capability code
584 * To be used in conjunction with pci_find_ht_capability() to search for
585 * all capabilities matching @ht_cap. @pos should always be a value returned
586 * from pci_find_ht_capability().
588 * NB. To be 100% safe against broken PCI devices, the caller should take
589 * steps to avoid an infinite loop.
591 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
593 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
595 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
598 * pci_find_ht_capability - query a device's Hypertransport capabilities
599 * @dev: PCI device to query
600 * @ht_cap: Hypertransport capability code
602 * Tell if a device supports a given Hypertransport capability.
603 * Returns an address within the device's PCI configuration space
604 * or 0 in case the device does not support the request capability.
605 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
606 * which has a Hypertransport capability matching @ht_cap.
608 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
612 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
614 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
618 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
621 * pci_find_parent_resource - return resource region of parent bus of given
623 * @dev: PCI device structure contains resources to be searched
624 * @res: child resource record for which parent is sought
626 * For given resource region of given device, return the resource region of
627 * parent bus the given region is contained in.
629 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
630 struct resource *res)
632 const struct pci_bus *bus = dev->bus;
636 pci_bus_for_each_resource(bus, r, i) {
639 if (resource_contains(r, res)) {
642 * If the window is prefetchable but the BAR is
643 * not, the allocator made a mistake.
645 if (r->flags & IORESOURCE_PREFETCH &&
646 !(res->flags & IORESOURCE_PREFETCH))
650 * If we're below a transparent bridge, there may
651 * be both a positively-decoded aperture and a
652 * subtractively-decoded region that contain the BAR.
653 * We want the positively-decoded one, so this depends
654 * on pci_bus_for_each_resource() giving us those
662 EXPORT_SYMBOL(pci_find_parent_resource);
665 * pci_find_resource - Return matching PCI device resource
666 * @dev: PCI device to query
667 * @res: Resource to look for
669 * Goes over standard PCI resources (BARs) and checks if the given resource
670 * is partially or fully contained in any of them. In that case the
671 * matching resource is returned, %NULL otherwise.
673 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
677 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
678 struct resource *r = &dev->resource[i];
680 if (r->start && resource_contains(r, res))
686 EXPORT_SYMBOL(pci_find_resource);
689 * pci_find_pcie_root_port - return PCIe Root Port
690 * @dev: PCI device to query
692 * Traverse up the parent chain and return the PCIe Root Port PCI Device
693 * for a given PCI Device.
695 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
697 struct pci_dev *bridge, *highest_pcie_bridge = dev;
699 bridge = pci_upstream_bridge(dev);
700 while (bridge && pci_is_pcie(bridge)) {
701 highest_pcie_bridge = bridge;
702 bridge = pci_upstream_bridge(bridge);
705 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
708 return highest_pcie_bridge;
710 EXPORT_SYMBOL(pci_find_pcie_root_port);
713 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
714 * @dev: the PCI device to operate on
715 * @pos: config space offset of status word
716 * @mask: mask of bit(s) to care about in status word
718 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
720 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
724 /* Wait for Transaction Pending bit clean */
725 for (i = 0; i < 4; i++) {
728 msleep((1 << (i - 1)) * 100);
730 pci_read_config_word(dev, pos, &status);
731 if (!(status & mask))
739 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
740 * @dev: PCI device to have its BARs restored
742 * Restore the BAR values for a given device, so as to make it
743 * accessible by its driver.
745 static void pci_restore_bars(struct pci_dev *dev)
749 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
750 pci_update_resource(dev, i);
753 static const struct pci_platform_pm_ops *pci_platform_pm;
755 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
757 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
758 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
760 pci_platform_pm = ops;
764 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
766 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
769 static inline int platform_pci_set_power_state(struct pci_dev *dev,
772 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
775 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
777 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
780 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
782 return pci_platform_pm ?
783 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
786 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
788 return pci_platform_pm ?
789 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
792 static inline bool platform_pci_need_resume(struct pci_dev *dev)
794 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
797 static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
799 return pci_platform_pm ? pci_platform_pm->bridge_d3(dev) : false;
803 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
805 * @dev: PCI device to handle.
806 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
809 * -EINVAL if the requested state is invalid.
810 * -EIO if device does not support PCI PM or its PM capabilities register has a
811 * wrong version, or device doesn't support the requested state.
812 * 0 if device already is in the requested state.
813 * 0 if device's power state has been successfully changed.
815 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
818 bool need_restore = false;
820 /* Check if we're already there */
821 if (dev->current_state == state)
827 if (state < PCI_D0 || state > PCI_D3hot)
831 * Validate current state:
832 * Can enter D0 from any state, but if we can only go deeper
833 * to sleep if we're already in a low power state
835 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
836 && dev->current_state > state) {
837 pci_err(dev, "invalid power transition (from state %d to %d)\n",
838 dev->current_state, state);
842 /* Check if this device supports the desired state */
843 if ((state == PCI_D1 && !dev->d1_support)
844 || (state == PCI_D2 && !dev->d2_support))
847 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
850 * If we're (effectively) in D3, force entire word to 0.
851 * This doesn't affect PME_Status, disables PME_En, and
852 * sets PowerState to 0.
854 switch (dev->current_state) {
858 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
863 case PCI_UNKNOWN: /* Boot-up */
864 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
865 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
867 /* Fall-through - force to D0 */
873 /* Enter specified state */
874 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
877 * Mandatory power management transition delays; see PCI PM 1.1
880 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
881 pci_dev_d3_sleep(dev);
882 else if (state == PCI_D2 || dev->current_state == PCI_D2)
883 udelay(PCI_PM_D2_DELAY);
885 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
886 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
887 if (dev->current_state != state && printk_ratelimit())
888 pci_info(dev, "Refused to change power state, currently in D%d\n",
892 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
893 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
894 * from D3hot to D0 _may_ perform an internal reset, thereby
895 * going to "D0 Uninitialized" rather than "D0 Initialized".
896 * For example, at least some versions of the 3c905B and the
897 * 3c556B exhibit this behaviour.
899 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
900 * devices in a D3hot state at boot. Consequently, we need to
901 * restore at least the BARs so that the device will be
902 * accessible to its driver.
905 pci_restore_bars(dev);
908 pcie_aspm_pm_state_change(dev->bus->self);
914 * pci_update_current_state - Read power state of given device and cache it
915 * @dev: PCI device to handle.
916 * @state: State to cache in case the device doesn't have the PM capability
918 * The power state is read from the PMCSR register, which however is
919 * inaccessible in D3cold. The platform firmware is therefore queried first
920 * to detect accessibility of the register. In case the platform firmware
921 * reports an incorrect state or the device isn't power manageable by the
922 * platform at all, we try to detect D3cold by testing accessibility of the
923 * vendor ID in config space.
925 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
927 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
928 !pci_device_is_present(dev)) {
929 dev->current_state = PCI_D3cold;
930 } else if (dev->pm_cap) {
933 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
934 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
936 dev->current_state = state;
941 * pci_power_up - Put the given device into D0 forcibly
942 * @dev: PCI device to power up
944 void pci_power_up(struct pci_dev *dev)
946 if (platform_pci_power_manageable(dev))
947 platform_pci_set_power_state(dev, PCI_D0);
949 pci_raw_set_power_state(dev, PCI_D0);
950 pci_update_current_state(dev, PCI_D0);
954 * pci_platform_power_transition - Use platform to change device power state
955 * @dev: PCI device to handle.
956 * @state: State to put the device into.
958 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
962 if (platform_pci_power_manageable(dev)) {
963 error = platform_pci_set_power_state(dev, state);
965 pci_update_current_state(dev, state);
969 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
970 dev->current_state = PCI_D0;
976 * pci_wakeup - Wake up a PCI device
977 * @pci_dev: Device to handle.
978 * @ign: ignored parameter
980 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
982 pci_wakeup_event(pci_dev);
983 pm_request_resume(&pci_dev->dev);
988 * pci_wakeup_bus - Walk given bus and wake up devices on it
989 * @bus: Top bus of the subtree to walk.
991 void pci_wakeup_bus(struct pci_bus *bus)
994 pci_walk_bus(bus, pci_wakeup, NULL);
998 * __pci_start_power_transition - Start power transition of a PCI device
999 * @dev: PCI device to handle.
1000 * @state: State to put the device into.
1002 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
1004 if (state == PCI_D0) {
1005 pci_platform_power_transition(dev, PCI_D0);
1007 * Mandatory power management transition delays, see
1008 * PCI Express Base Specification Revision 2.0 Section
1009 * 6.6.1: Conventional Reset. Do not delay for
1010 * devices powered on/off by corresponding bridge,
1011 * because have already delayed for the bridge.
1013 if (dev->runtime_d3cold) {
1014 if (dev->d3cold_delay && !dev->imm_ready)
1015 msleep(dev->d3cold_delay);
1017 * When powering on a bridge from D3cold, the
1018 * whole hierarchy may be powered on into
1019 * D0uninitialized state, resume them to give
1020 * them a chance to suspend again
1022 pci_wakeup_bus(dev->subordinate);
1028 * __pci_dev_set_current_state - Set current state of a PCI device
1029 * @dev: Device to handle
1030 * @data: pointer to state to be set
1032 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1034 pci_power_t state = *(pci_power_t *)data;
1036 dev->current_state = state;
1041 * pci_bus_set_current_state - Walk given bus and set current state of devices
1042 * @bus: Top bus of the subtree to walk.
1043 * @state: state to be set
1045 void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1048 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1052 * __pci_complete_power_transition - Complete power transition of a PCI device
1053 * @dev: PCI device to handle.
1054 * @state: State to put the device into.
1056 * This function should not be called directly by device drivers.
1058 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
1062 if (state <= PCI_D0)
1064 ret = pci_platform_power_transition(dev, state);
1065 /* Power off the bridge may power off the whole hierarchy */
1066 if (!ret && state == PCI_D3cold)
1067 pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1070 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
1073 * pci_set_power_state - Set the power state of a PCI device
1074 * @dev: PCI device to handle.
1075 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1077 * Transition a device to a new power state, using the platform firmware and/or
1078 * the device's PCI PM registers.
1081 * -EINVAL if the requested state is invalid.
1082 * -EIO if device does not support PCI PM or its PM capabilities register has a
1083 * wrong version, or device doesn't support the requested state.
1084 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1085 * 0 if device already is in the requested state.
1086 * 0 if the transition is to D3 but D3 is not supported.
1087 * 0 if device's power state has been successfully changed.
1089 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1093 /* Bound the state we're entering */
1094 if (state > PCI_D3cold)
1096 else if (state < PCI_D0)
1098 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1101 * If the device or the parent bridge do not support PCI
1102 * PM, ignore the request if we're doing anything other
1103 * than putting it into D0 (which would only happen on
1108 /* Check if we're already there */
1109 if (dev->current_state == state)
1112 __pci_start_power_transition(dev, state);
1115 * This device is quirked not to be put into D3, so don't put it in
1118 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1122 * To put device in D3cold, we put device into D3hot in native
1123 * way, then put device into D3cold with platform ops
1125 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1128 if (!__pci_complete_power_transition(dev, state))
1133 EXPORT_SYMBOL(pci_set_power_state);
1136 * pci_choose_state - Choose the power state of a PCI device
1137 * @dev: PCI device to be suspended
1138 * @state: target sleep state for the whole system. This is the value
1139 * that is passed to suspend() function.
1141 * Returns PCI power state suitable for given device and given system
1144 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
1151 ret = platform_pci_choose_state(dev);
1152 if (ret != PCI_POWER_ERROR)
1155 switch (state.event) {
1158 case PM_EVENT_FREEZE:
1159 case PM_EVENT_PRETHAW:
1160 /* REVISIT both freeze and pre-thaw "should" use D0 */
1161 case PM_EVENT_SUSPEND:
1162 case PM_EVENT_HIBERNATE:
1165 pci_info(dev, "unrecognized suspend event %d\n",
1171 EXPORT_SYMBOL(pci_choose_state);
1173 #define PCI_EXP_SAVE_REGS 7
1175 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1176 u16 cap, bool extended)
1178 struct pci_cap_saved_state *tmp;
1180 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1181 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1187 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1189 return _pci_find_saved_cap(dev, cap, false);
1192 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1194 return _pci_find_saved_cap(dev, cap, true);
1197 static int pci_save_pcie_state(struct pci_dev *dev)
1200 struct pci_cap_saved_state *save_state;
1203 if (!pci_is_pcie(dev))
1206 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1208 pci_err(dev, "buffer not found in %s\n", __func__);
1212 cap = (u16 *)&save_state->cap.data[0];
1213 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1214 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1215 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1216 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1217 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1218 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1219 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1224 static void pci_restore_pcie_state(struct pci_dev *dev)
1227 struct pci_cap_saved_state *save_state;
1230 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1234 cap = (u16 *)&save_state->cap.data[0];
1235 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1236 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1237 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1238 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1239 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1240 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1241 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1244 static int pci_save_pcix_state(struct pci_dev *dev)
1247 struct pci_cap_saved_state *save_state;
1249 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1253 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1255 pci_err(dev, "buffer not found in %s\n", __func__);
1259 pci_read_config_word(dev, pos + PCI_X_CMD,
1260 (u16 *)save_state->cap.data);
1265 static void pci_restore_pcix_state(struct pci_dev *dev)
1268 struct pci_cap_saved_state *save_state;
1271 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1272 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1273 if (!save_state || !pos)
1275 cap = (u16 *)&save_state->cap.data[0];
1277 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1280 static void pci_save_ltr_state(struct pci_dev *dev)
1283 struct pci_cap_saved_state *save_state;
1286 if (!pci_is_pcie(dev))
1289 ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1293 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1295 pci_err(dev, "no suspend buffer for LTR; ASPM issues possible after resume\n");
1299 cap = (u16 *)&save_state->cap.data[0];
1300 pci_read_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, cap++);
1301 pci_read_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, cap++);
1304 static void pci_restore_ltr_state(struct pci_dev *dev)
1306 struct pci_cap_saved_state *save_state;
1310 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_LTR);
1311 ltr = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
1312 if (!save_state || !ltr)
1315 cap = (u16 *)&save_state->cap.data[0];
1316 pci_write_config_word(dev, ltr + PCI_LTR_MAX_SNOOP_LAT, *cap++);
1317 pci_write_config_word(dev, ltr + PCI_LTR_MAX_NOSNOOP_LAT, *cap++);
1321 * pci_save_state - save the PCI configuration space of a device before
1323 * @dev: PCI device that we're dealing with
1325 int pci_save_state(struct pci_dev *dev)
1328 /* XXX: 100% dword access ok here? */
1329 for (i = 0; i < 16; i++)
1330 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1331 dev->state_saved = true;
1333 i = pci_save_pcie_state(dev);
1337 i = pci_save_pcix_state(dev);
1341 pci_save_ltr_state(dev);
1342 pci_save_dpc_state(dev);
1343 return pci_save_vc_state(dev);
1345 EXPORT_SYMBOL(pci_save_state);
1347 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1348 u32 saved_val, int retry, bool force)
1352 pci_read_config_dword(pdev, offset, &val);
1353 if (!force && val == saved_val)
1357 pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1358 offset, val, saved_val);
1359 pci_write_config_dword(pdev, offset, saved_val);
1363 pci_read_config_dword(pdev, offset, &val);
1364 if (val == saved_val)
1371 static void pci_restore_config_space_range(struct pci_dev *pdev,
1372 int start, int end, int retry,
1377 for (index = end; index >= start; index--)
1378 pci_restore_config_dword(pdev, 4 * index,
1379 pdev->saved_config_space[index],
1383 static void pci_restore_config_space(struct pci_dev *pdev)
1385 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1386 pci_restore_config_space_range(pdev, 10, 15, 0, false);
1387 /* Restore BARs before the command register. */
1388 pci_restore_config_space_range(pdev, 4, 9, 10, false);
1389 pci_restore_config_space_range(pdev, 0, 3, 0, false);
1390 } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1391 pci_restore_config_space_range(pdev, 12, 15, 0, false);
1394 * Force rewriting of prefetch registers to avoid S3 resume
1395 * issues on Intel PCI bridges that occur when these
1396 * registers are not explicitly written.
1398 pci_restore_config_space_range(pdev, 9, 11, 0, true);
1399 pci_restore_config_space_range(pdev, 0, 8, 0, false);
1401 pci_restore_config_space_range(pdev, 0, 15, 0, false);
1405 static void pci_restore_rebar_state(struct pci_dev *pdev)
1407 unsigned int pos, nbars, i;
1410 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1414 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1415 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1416 PCI_REBAR_CTRL_NBAR_SHIFT;
1418 for (i = 0; i < nbars; i++, pos += 8) {
1419 struct resource *res;
1422 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1423 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1424 res = pdev->resource + bar_idx;
1425 size = order_base_2((resource_size(res) >> 20) | 1) - 1;
1426 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1427 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1428 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1433 * pci_restore_state - Restore the saved state of a PCI device
1434 * @dev: PCI device that we're dealing with
1436 void pci_restore_state(struct pci_dev *dev)
1438 if (!dev->state_saved)
1442 * Restore max latencies (in the LTR capability) before enabling
1443 * LTR itself (in the PCIe capability).
1445 pci_restore_ltr_state(dev);
1447 pci_restore_pcie_state(dev);
1448 pci_restore_pasid_state(dev);
1449 pci_restore_pri_state(dev);
1450 pci_restore_ats_state(dev);
1451 pci_restore_vc_state(dev);
1452 pci_restore_rebar_state(dev);
1453 pci_restore_dpc_state(dev);
1455 pci_cleanup_aer_error_status_regs(dev);
1457 pci_restore_config_space(dev);
1459 pci_restore_pcix_state(dev);
1460 pci_restore_msi_state(dev);
1462 /* Restore ACS and IOV configuration state */
1463 pci_enable_acs(dev);
1464 pci_restore_iov_state(dev);
1466 dev->state_saved = false;
1468 EXPORT_SYMBOL(pci_restore_state);
1470 struct pci_saved_state {
1471 u32 config_space[16];
1472 struct pci_cap_saved_data cap[0];
1476 * pci_store_saved_state - Allocate and return an opaque struct containing
1477 * the device saved state.
1478 * @dev: PCI device that we're dealing with
1480 * Return NULL if no state or error.
1482 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1484 struct pci_saved_state *state;
1485 struct pci_cap_saved_state *tmp;
1486 struct pci_cap_saved_data *cap;
1489 if (!dev->state_saved)
1492 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1494 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1495 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1497 state = kzalloc(size, GFP_KERNEL);
1501 memcpy(state->config_space, dev->saved_config_space,
1502 sizeof(state->config_space));
1505 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1506 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1507 memcpy(cap, &tmp->cap, len);
1508 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1510 /* Empty cap_save terminates list */
1514 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1517 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1518 * @dev: PCI device that we're dealing with
1519 * @state: Saved state returned from pci_store_saved_state()
1521 int pci_load_saved_state(struct pci_dev *dev,
1522 struct pci_saved_state *state)
1524 struct pci_cap_saved_data *cap;
1526 dev->state_saved = false;
1531 memcpy(dev->saved_config_space, state->config_space,
1532 sizeof(state->config_space));
1536 struct pci_cap_saved_state *tmp;
1538 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1539 if (!tmp || tmp->cap.size != cap->size)
1542 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1543 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1544 sizeof(struct pci_cap_saved_data) + cap->size);
1547 dev->state_saved = true;
1550 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1553 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1554 * and free the memory allocated for it.
1555 * @dev: PCI device that we're dealing with
1556 * @state: Pointer to saved state returned from pci_store_saved_state()
1558 int pci_load_and_free_saved_state(struct pci_dev *dev,
1559 struct pci_saved_state **state)
1561 int ret = pci_load_saved_state(dev, *state);
1566 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1568 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1570 return pci_enable_resources(dev, bars);
1573 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1576 struct pci_dev *bridge;
1580 err = pci_set_power_state(dev, PCI_D0);
1581 if (err < 0 && err != -EIO)
1584 bridge = pci_upstream_bridge(dev);
1586 pcie_aspm_powersave_config_link(bridge);
1588 err = pcibios_enable_device(dev, bars);
1591 pci_fixup_device(pci_fixup_enable, dev);
1593 if (dev->msi_enabled || dev->msix_enabled)
1596 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1598 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1599 if (cmd & PCI_COMMAND_INTX_DISABLE)
1600 pci_write_config_word(dev, PCI_COMMAND,
1601 cmd & ~PCI_COMMAND_INTX_DISABLE);
1608 * pci_reenable_device - Resume abandoned device
1609 * @dev: PCI device to be resumed
1611 * NOTE: This function is a backend of pci_default_resume() and is not supposed
1612 * to be called by normal code, write proper resume handler and use it instead.
1614 int pci_reenable_device(struct pci_dev *dev)
1616 if (pci_is_enabled(dev))
1617 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1620 EXPORT_SYMBOL(pci_reenable_device);
1622 static void pci_enable_bridge(struct pci_dev *dev)
1624 struct pci_dev *bridge;
1627 bridge = pci_upstream_bridge(dev);
1629 pci_enable_bridge(bridge);
1631 if (pci_is_enabled(dev)) {
1632 if (!dev->is_busmaster)
1633 pci_set_master(dev);
1637 retval = pci_enable_device(dev);
1639 pci_err(dev, "Error enabling bridge (%d), continuing\n",
1641 pci_set_master(dev);
1644 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1646 struct pci_dev *bridge;
1651 * Power state could be unknown at this point, either due to a fresh
1652 * boot or a device removal call. So get the current power state
1653 * so that things like MSI message writing will behave as expected
1654 * (e.g. if the device really is in D0 at enable time).
1658 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1659 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1662 if (atomic_inc_return(&dev->enable_cnt) > 1)
1663 return 0; /* already enabled */
1665 bridge = pci_upstream_bridge(dev);
1667 pci_enable_bridge(bridge);
1669 /* only skip sriov related */
1670 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1671 if (dev->resource[i].flags & flags)
1673 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1674 if (dev->resource[i].flags & flags)
1677 err = do_pci_enable_device(dev, bars);
1679 atomic_dec(&dev->enable_cnt);
1684 * pci_enable_device_io - Initialize a device for use with IO space
1685 * @dev: PCI device to be initialized
1687 * Initialize device before it's used by a driver. Ask low-level code
1688 * to enable I/O resources. Wake up the device if it was suspended.
1689 * Beware, this function can fail.
1691 int pci_enable_device_io(struct pci_dev *dev)
1693 return pci_enable_device_flags(dev, IORESOURCE_IO);
1695 EXPORT_SYMBOL(pci_enable_device_io);
1698 * pci_enable_device_mem - Initialize a device for use with Memory space
1699 * @dev: PCI device to be initialized
1701 * Initialize device before it's used by a driver. Ask low-level code
1702 * to enable Memory resources. Wake up the device if it was suspended.
1703 * Beware, this function can fail.
1705 int pci_enable_device_mem(struct pci_dev *dev)
1707 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1709 EXPORT_SYMBOL(pci_enable_device_mem);
1712 * pci_enable_device - Initialize device before it's used by a driver.
1713 * @dev: PCI device to be initialized
1715 * Initialize device before it's used by a driver. Ask low-level code
1716 * to enable I/O and memory. Wake up the device if it was suspended.
1717 * Beware, this function can fail.
1719 * Note we don't actually enable the device many times if we call
1720 * this function repeatedly (we just increment the count).
1722 int pci_enable_device(struct pci_dev *dev)
1724 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1726 EXPORT_SYMBOL(pci_enable_device);
1729 * Managed PCI resources. This manages device on/off, INTx/MSI/MSI-X
1730 * on/off and BAR regions. pci_dev itself records MSI/MSI-X status, so
1731 * there's no need to track it separately. pci_devres is initialized
1732 * when a device is enabled using managed PCI device enable interface.
1735 unsigned int enabled:1;
1736 unsigned int pinned:1;
1737 unsigned int orig_intx:1;
1738 unsigned int restore_intx:1;
1743 static void pcim_release(struct device *gendev, void *res)
1745 struct pci_dev *dev = to_pci_dev(gendev);
1746 struct pci_devres *this = res;
1749 if (dev->msi_enabled)
1750 pci_disable_msi(dev);
1751 if (dev->msix_enabled)
1752 pci_disable_msix(dev);
1754 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1755 if (this->region_mask & (1 << i))
1756 pci_release_region(dev, i);
1761 if (this->restore_intx)
1762 pci_intx(dev, this->orig_intx);
1764 if (this->enabled && !this->pinned)
1765 pci_disable_device(dev);
1768 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1770 struct pci_devres *dr, *new_dr;
1772 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1776 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1779 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1782 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1784 if (pci_is_managed(pdev))
1785 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1790 * pcim_enable_device - Managed pci_enable_device()
1791 * @pdev: PCI device to be initialized
1793 * Managed pci_enable_device().
1795 int pcim_enable_device(struct pci_dev *pdev)
1797 struct pci_devres *dr;
1800 dr = get_pci_dr(pdev);
1806 rc = pci_enable_device(pdev);
1808 pdev->is_managed = 1;
1813 EXPORT_SYMBOL(pcim_enable_device);
1816 * pcim_pin_device - Pin managed PCI device
1817 * @pdev: PCI device to pin
1819 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1820 * driver detach. @pdev must have been enabled with
1821 * pcim_enable_device().
1823 void pcim_pin_device(struct pci_dev *pdev)
1825 struct pci_devres *dr;
1827 dr = find_pci_dr(pdev);
1828 WARN_ON(!dr || !dr->enabled);
1832 EXPORT_SYMBOL(pcim_pin_device);
1835 * pcibios_add_device - provide arch specific hooks when adding device dev
1836 * @dev: the PCI device being added
1838 * Permits the platform to provide architecture specific functionality when
1839 * devices are added. This is the default implementation. Architecture
1840 * implementations can override this.
1842 int __weak pcibios_add_device(struct pci_dev *dev)
1848 * pcibios_release_device - provide arch specific hooks when releasing
1850 * @dev: the PCI device being released
1852 * Permits the platform to provide architecture specific functionality when
1853 * devices are released. This is the default implementation. Architecture
1854 * implementations can override this.
1856 void __weak pcibios_release_device(struct pci_dev *dev) {}
1859 * pcibios_disable_device - disable arch specific PCI resources for device dev
1860 * @dev: the PCI device to disable
1862 * Disables architecture specific PCI resources for the device. This
1863 * is the default implementation. Architecture implementations can
1866 void __weak pcibios_disable_device(struct pci_dev *dev) {}
1869 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1870 * @irq: ISA IRQ to penalize
1871 * @active: IRQ active or not
1873 * Permits the platform to provide architecture-specific functionality when
1874 * penalizing ISA IRQs. This is the default implementation. Architecture
1875 * implementations can override this.
1877 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1879 static void do_pci_disable_device(struct pci_dev *dev)
1883 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1884 if (pci_command & PCI_COMMAND_MASTER) {
1885 pci_command &= ~PCI_COMMAND_MASTER;
1886 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1889 pcibios_disable_device(dev);
1893 * pci_disable_enabled_device - Disable device without updating enable_cnt
1894 * @dev: PCI device to disable
1896 * NOTE: This function is a backend of PCI power management routines and is
1897 * not supposed to be called drivers.
1899 void pci_disable_enabled_device(struct pci_dev *dev)
1901 if (pci_is_enabled(dev))
1902 do_pci_disable_device(dev);
1906 * pci_disable_device - Disable PCI device after use
1907 * @dev: PCI device to be disabled
1909 * Signal to the system that the PCI device is not in use by the system
1910 * anymore. This only involves disabling PCI bus-mastering, if active.
1912 * Note we don't actually disable the device until all callers of
1913 * pci_enable_device() have called pci_disable_device().
1915 void pci_disable_device(struct pci_dev *dev)
1917 struct pci_devres *dr;
1919 dr = find_pci_dr(dev);
1923 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1924 "disabling already-disabled device");
1926 if (atomic_dec_return(&dev->enable_cnt) != 0)
1929 do_pci_disable_device(dev);
1931 dev->is_busmaster = 0;
1933 EXPORT_SYMBOL(pci_disable_device);
1936 * pcibios_set_pcie_reset_state - set reset state for device dev
1937 * @dev: the PCIe device reset
1938 * @state: Reset state to enter into
1940 * Set the PCIe reset state for the device. This is the default
1941 * implementation. Architecture implementations can override this.
1943 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1944 enum pcie_reset_state state)
1950 * pci_set_pcie_reset_state - set reset state for device dev
1951 * @dev: the PCIe device reset
1952 * @state: Reset state to enter into
1954 * Sets the PCI reset state for the device.
1956 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1958 return pcibios_set_pcie_reset_state(dev, state);
1960 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1963 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
1964 * @dev: PCIe root port or event collector.
1966 void pcie_clear_root_pme_status(struct pci_dev *dev)
1968 pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
1972 * pci_check_pme_status - Check if given device has generated PME.
1973 * @dev: Device to check.
1975 * Check the PME status of the device and if set, clear it and clear PME enable
1976 * (if set). Return 'true' if PME status and PME enable were both set or
1977 * 'false' otherwise.
1979 bool pci_check_pme_status(struct pci_dev *dev)
1988 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1989 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1990 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1993 /* Clear PME status. */
1994 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1995 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1996 /* Disable PME to avoid interrupt flood. */
1997 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2001 pci_write_config_word(dev, pmcsr_pos, pmcsr);
2007 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2008 * @dev: Device to handle.
2009 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2011 * Check if @dev has generated PME and queue a resume request for it in that
2014 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2016 if (pme_poll_reset && dev->pme_poll)
2017 dev->pme_poll = false;
2019 if (pci_check_pme_status(dev)) {
2020 pci_wakeup_event(dev);
2021 pm_request_resume(&dev->dev);
2027 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2028 * @bus: Top bus of the subtree to walk.
2030 void pci_pme_wakeup_bus(struct pci_bus *bus)
2033 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2038 * pci_pme_capable - check the capability of PCI device to generate PME#
2039 * @dev: PCI device to handle.
2040 * @state: PCI state from which device will issue PME#.
2042 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2047 return !!(dev->pme_support & (1 << state));
2049 EXPORT_SYMBOL(pci_pme_capable);
2051 static void pci_pme_list_scan(struct work_struct *work)
2053 struct pci_pme_device *pme_dev, *n;
2055 mutex_lock(&pci_pme_list_mutex);
2056 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2057 if (pme_dev->dev->pme_poll) {
2058 struct pci_dev *bridge;
2060 bridge = pme_dev->dev->bus->self;
2062 * If bridge is in low power state, the
2063 * configuration space of subordinate devices
2064 * may be not accessible
2066 if (bridge && bridge->current_state != PCI_D0)
2068 pci_pme_wakeup(pme_dev->dev, NULL);
2070 list_del(&pme_dev->list);
2074 if (!list_empty(&pci_pme_list))
2075 queue_delayed_work(system_freezable_wq, &pci_pme_work,
2076 msecs_to_jiffies(PME_TIMEOUT));
2077 mutex_unlock(&pci_pme_list_mutex);
2080 static void __pci_pme_active(struct pci_dev *dev, bool enable)
2084 if (!dev->pme_support)
2087 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2088 /* Clear PME_Status by writing 1 to it and enable PME# */
2089 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2091 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2093 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2097 * pci_pme_restore - Restore PME configuration after config space restore.
2098 * @dev: PCI device to update.
2100 void pci_pme_restore(struct pci_dev *dev)
2104 if (!dev->pme_support)
2107 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2108 if (dev->wakeup_prepared) {
2109 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2110 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2112 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2113 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2115 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2119 * pci_pme_active - enable or disable PCI device's PME# function
2120 * @dev: PCI device to handle.
2121 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2123 * The caller must verify that the device is capable of generating PME# before
2124 * calling this function with @enable equal to 'true'.
2126 void pci_pme_active(struct pci_dev *dev, bool enable)
2128 __pci_pme_active(dev, enable);
2131 * PCI (as opposed to PCIe) PME requires that the device have
2132 * its PME# line hooked up correctly. Not all hardware vendors
2133 * do this, so the PME never gets delivered and the device
2134 * remains asleep. The easiest way around this is to
2135 * periodically walk the list of suspended devices and check
2136 * whether any have their PME flag set. The assumption is that
2137 * we'll wake up often enough anyway that this won't be a huge
2138 * hit, and the power savings from the devices will still be a
2141 * Although PCIe uses in-band PME message instead of PME# line
2142 * to report PME, PME does not work for some PCIe devices in
2143 * reality. For example, there are devices that set their PME
2144 * status bits, but don't really bother to send a PME message;
2145 * there are PCI Express Root Ports that don't bother to
2146 * trigger interrupts when they receive PME messages from the
2147 * devices below. So PME poll is used for PCIe devices too.
2150 if (dev->pme_poll) {
2151 struct pci_pme_device *pme_dev;
2153 pme_dev = kmalloc(sizeof(struct pci_pme_device),
2156 pci_warn(dev, "can't enable PME#\n");
2160 mutex_lock(&pci_pme_list_mutex);
2161 list_add(&pme_dev->list, &pci_pme_list);
2162 if (list_is_singular(&pci_pme_list))
2163 queue_delayed_work(system_freezable_wq,
2165 msecs_to_jiffies(PME_TIMEOUT));
2166 mutex_unlock(&pci_pme_list_mutex);
2168 mutex_lock(&pci_pme_list_mutex);
2169 list_for_each_entry(pme_dev, &pci_pme_list, list) {
2170 if (pme_dev->dev == dev) {
2171 list_del(&pme_dev->list);
2176 mutex_unlock(&pci_pme_list_mutex);
2180 pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2182 EXPORT_SYMBOL(pci_pme_active);
2185 * __pci_enable_wake - enable PCI device as wakeup event source
2186 * @dev: PCI device affected
2187 * @state: PCI state from which device will issue wakeup events
2188 * @enable: True to enable event generation; false to disable
2190 * This enables the device as a wakeup event source, or disables it.
2191 * When such events involves platform-specific hooks, those hooks are
2192 * called automatically by this routine.
2194 * Devices with legacy power management (no standard PCI PM capabilities)
2195 * always require such platform hooks.
2198 * 0 is returned on success
2199 * -EINVAL is returned if device is not supposed to wake up the system
2200 * Error code depending on the platform is returned if both the platform and
2201 * the native mechanism fail to enable the generation of wake-up events
2203 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2208 * Bridges that are not power-manageable directly only signal
2209 * wakeup on behalf of subordinate devices which is set up
2210 * elsewhere, so skip them. However, bridges that are
2211 * power-manageable may signal wakeup for themselves (for example,
2212 * on a hotplug event) and they need to be covered here.
2214 if (!pci_power_manageable(dev))
2217 /* Don't do the same thing twice in a row for one device. */
2218 if (!!enable == !!dev->wakeup_prepared)
2222 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2223 * Anderson we should be doing PME# wake enable followed by ACPI wake
2224 * enable. To disable wake-up we call the platform first, for symmetry.
2230 if (pci_pme_capable(dev, state))
2231 pci_pme_active(dev, true);
2234 error = platform_pci_set_wakeup(dev, true);
2238 dev->wakeup_prepared = true;
2240 platform_pci_set_wakeup(dev, false);
2241 pci_pme_active(dev, false);
2242 dev->wakeup_prepared = false;
2249 * pci_enable_wake - change wakeup settings for a PCI device
2250 * @pci_dev: Target device
2251 * @state: PCI state from which device will issue wakeup events
2252 * @enable: Whether or not to enable event generation
2254 * If @enable is set, check device_may_wakeup() for the device before calling
2255 * __pci_enable_wake() for it.
2257 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2259 if (enable && !device_may_wakeup(&pci_dev->dev))
2262 return __pci_enable_wake(pci_dev, state, enable);
2264 EXPORT_SYMBOL(pci_enable_wake);
2267 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2268 * @dev: PCI device to prepare
2269 * @enable: True to enable wake-up event generation; false to disable
2271 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2272 * and this function allows them to set that up cleanly - pci_enable_wake()
2273 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2274 * ordering constraints.
2276 * This function only returns error code if the device is not allowed to wake
2277 * up the system from sleep or it is not capable of generating PME# from both
2278 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2280 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2282 return pci_pme_capable(dev, PCI_D3cold) ?
2283 pci_enable_wake(dev, PCI_D3cold, enable) :
2284 pci_enable_wake(dev, PCI_D3hot, enable);
2286 EXPORT_SYMBOL(pci_wake_from_d3);
2289 * pci_target_state - find an appropriate low power state for a given PCI dev
2291 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2293 * Use underlying platform code to find a supported low power state for @dev.
2294 * If the platform can't manage @dev, return the deepest state from which it
2295 * can generate wake events, based on any available PME info.
2297 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2299 pci_power_t target_state = PCI_D3hot;
2301 if (platform_pci_power_manageable(dev)) {
2303 * Call the platform to find the target state for the device.
2305 pci_power_t state = platform_pci_choose_state(dev);
2308 case PCI_POWER_ERROR:
2313 if (pci_no_d1d2(dev))
2315 /* else, fall through */
2317 target_state = state;
2320 return target_state;
2324 target_state = PCI_D0;
2327 * If the device is in D3cold even though it's not power-manageable by
2328 * the platform, it may have been powered down by non-standard means.
2329 * Best to let it slumber.
2331 if (dev->current_state == PCI_D3cold)
2332 target_state = PCI_D3cold;
2336 * Find the deepest state from which the device can generate
2339 if (dev->pme_support) {
2341 && !(dev->pme_support & (1 << target_state)))
2346 return target_state;
2350 * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2351 * into a sleep state
2352 * @dev: Device to handle.
2354 * Choose the power state appropriate for the device depending on whether
2355 * it can wake up the system and/or is power manageable by the platform
2356 * (PCI_D3hot is the default) and put the device into that state.
2358 int pci_prepare_to_sleep(struct pci_dev *dev)
2360 bool wakeup = device_may_wakeup(&dev->dev);
2361 pci_power_t target_state = pci_target_state(dev, wakeup);
2364 if (target_state == PCI_POWER_ERROR)
2367 pci_enable_wake(dev, target_state, wakeup);
2369 error = pci_set_power_state(dev, target_state);
2372 pci_enable_wake(dev, target_state, false);
2376 EXPORT_SYMBOL(pci_prepare_to_sleep);
2379 * pci_back_from_sleep - turn PCI device on during system-wide transition
2380 * into working state
2381 * @dev: Device to handle.
2383 * Disable device's system wake-up capability and put it into D0.
2385 int pci_back_from_sleep(struct pci_dev *dev)
2387 pci_enable_wake(dev, PCI_D0, false);
2388 return pci_set_power_state(dev, PCI_D0);
2390 EXPORT_SYMBOL(pci_back_from_sleep);
2393 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2394 * @dev: PCI device being suspended.
2396 * Prepare @dev to generate wake-up events at run time and put it into a low
2399 int pci_finish_runtime_suspend(struct pci_dev *dev)
2401 pci_power_t target_state;
2404 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2405 if (target_state == PCI_POWER_ERROR)
2408 dev->runtime_d3cold = target_state == PCI_D3cold;
2410 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2412 error = pci_set_power_state(dev, target_state);
2415 pci_enable_wake(dev, target_state, false);
2416 dev->runtime_d3cold = false;
2423 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2424 * @dev: Device to check.
2426 * Return true if the device itself is capable of generating wake-up events
2427 * (through the platform or using the native PCIe PME) or if the device supports
2428 * PME and one of its upstream bridges can generate wake-up events.
2430 bool pci_dev_run_wake(struct pci_dev *dev)
2432 struct pci_bus *bus = dev->bus;
2434 if (!dev->pme_support)
2437 /* PME-capable in principle, but not from the target power state */
2438 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2441 if (device_can_wakeup(&dev->dev))
2444 while (bus->parent) {
2445 struct pci_dev *bridge = bus->self;
2447 if (device_can_wakeup(&bridge->dev))
2453 /* We have reached the root bus. */
2455 return device_can_wakeup(bus->bridge);
2459 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2462 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2463 * @pci_dev: Device to check.
2465 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2466 * reconfigured due to wakeup settings difference between system and runtime
2467 * suspend and the current power state of it is suitable for the upcoming
2468 * (system) transition.
2470 * If the device is not configured for system wakeup, disable PME for it before
2471 * returning 'true' to prevent it from waking up the system unnecessarily.
2473 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2475 struct device *dev = &pci_dev->dev;
2476 bool wakeup = device_may_wakeup(dev);
2478 if (!pm_runtime_suspended(dev)
2479 || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2480 || platform_pci_need_resume(pci_dev))
2484 * At this point the device is good to go unless it's been configured
2485 * to generate PME at the runtime suspend time, but it is not supposed
2486 * to wake up the system. In that case, simply disable PME for it
2487 * (it will have to be re-enabled on exit from system resume).
2489 * If the device's power state is D3cold and the platform check above
2490 * hasn't triggered, the device's configuration is suitable and we don't
2491 * need to manipulate it at all.
2493 spin_lock_irq(&dev->power.lock);
2495 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2497 __pci_pme_active(pci_dev, false);
2499 spin_unlock_irq(&dev->power.lock);
2504 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2505 * @pci_dev: Device to handle.
2507 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2508 * it might have been disabled during the prepare phase of system suspend if
2509 * the device was not configured for system wakeup.
2511 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2513 struct device *dev = &pci_dev->dev;
2515 if (!pci_dev_run_wake(pci_dev))
2518 spin_lock_irq(&dev->power.lock);
2520 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2521 __pci_pme_active(pci_dev, true);
2523 spin_unlock_irq(&dev->power.lock);
2526 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2528 struct device *dev = &pdev->dev;
2529 struct device *parent = dev->parent;
2532 pm_runtime_get_sync(parent);
2533 pm_runtime_get_noresume(dev);
2535 * pdev->current_state is set to PCI_D3cold during suspending,
2536 * so wait until suspending completes
2538 pm_runtime_barrier(dev);
2540 * Only need to resume devices in D3cold, because config
2541 * registers are still accessible for devices suspended but
2544 if (pdev->current_state == PCI_D3cold)
2545 pm_runtime_resume(dev);
2548 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2550 struct device *dev = &pdev->dev;
2551 struct device *parent = dev->parent;
2553 pm_runtime_put(dev);
2555 pm_runtime_put_sync(parent);
2558 static const struct dmi_system_id bridge_d3_blacklist[] = {
2562 * Gigabyte X299 root port is not marked as hotplug capable
2563 * which allows Linux to power manage it. However, this
2564 * confuses the BIOS SMI handler so don't power manage root
2565 * ports on that system.
2567 .ident = "X299 DESIGNARE EX-CF",
2569 DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2570 DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2578 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2579 * @bridge: Bridge to check
2581 * This function checks if it is possible to move the bridge to D3.
2582 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2584 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2586 if (!pci_is_pcie(bridge))
2589 switch (pci_pcie_type(bridge)) {
2590 case PCI_EXP_TYPE_ROOT_PORT:
2591 case PCI_EXP_TYPE_UPSTREAM:
2592 case PCI_EXP_TYPE_DOWNSTREAM:
2593 if (pci_bridge_d3_disable)
2597 * Hotplug ports handled by firmware in System Management Mode
2598 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2600 if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2603 if (pci_bridge_d3_force)
2606 /* Even the oldest 2010 Thunderbolt controller supports D3. */
2607 if (bridge->is_thunderbolt)
2610 /* Platform might know better if the bridge supports D3 */
2611 if (platform_pci_bridge_d3(bridge))
2615 * Hotplug ports handled natively by the OS were not validated
2616 * by vendors for runtime D3 at least until 2018 because there
2617 * was no OS support.
2619 if (bridge->is_hotplug_bridge)
2622 if (dmi_check_system(bridge_d3_blacklist))
2626 * It should be safe to put PCIe ports from 2015 or newer
2629 if (dmi_get_bios_year() >= 2015)
2637 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2639 bool *d3cold_ok = data;
2641 if (/* The device needs to be allowed to go D3cold ... */
2642 dev->no_d3cold || !dev->d3cold_allowed ||
2644 /* ... and if it is wakeup capable to do so from D3cold. */
2645 (device_may_wakeup(&dev->dev) &&
2646 !pci_pme_capable(dev, PCI_D3cold)) ||
2648 /* If it is a bridge it must be allowed to go to D3. */
2649 !pci_power_manageable(dev))
2657 * pci_bridge_d3_update - Update bridge D3 capabilities
2658 * @dev: PCI device which is changed
2660 * Update upstream bridge PM capabilities accordingly depending on if the
2661 * device PM configuration was changed or the device is being removed. The
2662 * change is also propagated upstream.
2664 void pci_bridge_d3_update(struct pci_dev *dev)
2666 bool remove = !device_is_registered(&dev->dev);
2667 struct pci_dev *bridge;
2668 bool d3cold_ok = true;
2670 bridge = pci_upstream_bridge(dev);
2671 if (!bridge || !pci_bridge_d3_possible(bridge))
2675 * If D3 is currently allowed for the bridge, removing one of its
2676 * children won't change that.
2678 if (remove && bridge->bridge_d3)
2682 * If D3 is currently allowed for the bridge and a child is added or
2683 * changed, disallowance of D3 can only be caused by that child, so
2684 * we only need to check that single device, not any of its siblings.
2686 * If D3 is currently not allowed for the bridge, checking the device
2687 * first may allow us to skip checking its siblings.
2690 pci_dev_check_d3cold(dev, &d3cold_ok);
2693 * If D3 is currently not allowed for the bridge, this may be caused
2694 * either by the device being changed/removed or any of its siblings,
2695 * so we need to go through all children to find out if one of them
2696 * continues to block D3.
2698 if (d3cold_ok && !bridge->bridge_d3)
2699 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2702 if (bridge->bridge_d3 != d3cold_ok) {
2703 bridge->bridge_d3 = d3cold_ok;
2704 /* Propagate change to upstream bridges */
2705 pci_bridge_d3_update(bridge);
2710 * pci_d3cold_enable - Enable D3cold for device
2711 * @dev: PCI device to handle
2713 * This function can be used in drivers to enable D3cold from the device
2714 * they handle. It also updates upstream PCI bridge PM capabilities
2717 void pci_d3cold_enable(struct pci_dev *dev)
2719 if (dev->no_d3cold) {
2720 dev->no_d3cold = false;
2721 pci_bridge_d3_update(dev);
2724 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2727 * pci_d3cold_disable - Disable D3cold for device
2728 * @dev: PCI device to handle
2730 * This function can be used in drivers to disable D3cold from the device
2731 * they handle. It also updates upstream PCI bridge PM capabilities
2734 void pci_d3cold_disable(struct pci_dev *dev)
2736 if (!dev->no_d3cold) {
2737 dev->no_d3cold = true;
2738 pci_bridge_d3_update(dev);
2741 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2744 * pci_pm_init - Initialize PM functions of given PCI device
2745 * @dev: PCI device to handle.
2747 void pci_pm_init(struct pci_dev *dev)
2753 pm_runtime_forbid(&dev->dev);
2754 pm_runtime_set_active(&dev->dev);
2755 pm_runtime_enable(&dev->dev);
2756 device_enable_async_suspend(&dev->dev);
2757 dev->wakeup_prepared = false;
2760 dev->pme_support = 0;
2762 /* find PCI PM capability in list */
2763 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2766 /* Check device's ability to generate PME# */
2767 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2769 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2770 pci_err(dev, "unsupported PM cap regs version (%u)\n",
2771 pmc & PCI_PM_CAP_VER_MASK);
2776 dev->d3_delay = PCI_PM_D3_WAIT;
2777 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2778 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2779 dev->d3cold_allowed = true;
2781 dev->d1_support = false;
2782 dev->d2_support = false;
2783 if (!pci_no_d1d2(dev)) {
2784 if (pmc & PCI_PM_CAP_D1)
2785 dev->d1_support = true;
2786 if (pmc & PCI_PM_CAP_D2)
2787 dev->d2_support = true;
2789 if (dev->d1_support || dev->d2_support)
2790 pci_info(dev, "supports%s%s\n",
2791 dev->d1_support ? " D1" : "",
2792 dev->d2_support ? " D2" : "");
2795 pmc &= PCI_PM_CAP_PME_MASK;
2797 pci_info(dev, "PME# supported from%s%s%s%s%s\n",
2798 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2799 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2800 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2801 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2802 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2803 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2804 dev->pme_poll = true;
2806 * Make device's PM flags reflect the wake-up capability, but
2807 * let the user space enable it to wake up the system as needed.
2809 device_set_wakeup_capable(&dev->dev, true);
2810 /* Disable the PME# generation functionality */
2811 pci_pme_active(dev, false);
2814 pci_read_config_word(dev, PCI_STATUS, &status);
2815 if (status & PCI_STATUS_IMM_READY)
2819 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2821 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2825 case PCI_EA_P_VF_MEM:
2826 flags |= IORESOURCE_MEM;
2828 case PCI_EA_P_MEM_PREFETCH:
2829 case PCI_EA_P_VF_MEM_PREFETCH:
2830 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2833 flags |= IORESOURCE_IO;
2842 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2845 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2846 return &dev->resource[bei];
2847 #ifdef CONFIG_PCI_IOV
2848 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2849 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2850 return &dev->resource[PCI_IOV_RESOURCES +
2851 bei - PCI_EA_BEI_VF_BAR0];
2853 else if (bei == PCI_EA_BEI_ROM)
2854 return &dev->resource[PCI_ROM_RESOURCE];
2859 /* Read an Enhanced Allocation (EA) entry */
2860 static int pci_ea_read(struct pci_dev *dev, int offset)
2862 struct resource *res;
2863 int ent_size, ent_offset = offset;
2864 resource_size_t start, end;
2865 unsigned long flags;
2866 u32 dw0, bei, base, max_offset;
2868 bool support_64 = (sizeof(resource_size_t) >= 8);
2870 pci_read_config_dword(dev, ent_offset, &dw0);
2873 /* Entry size field indicates DWORDs after 1st */
2874 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2876 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2879 bei = (dw0 & PCI_EA_BEI) >> 4;
2880 prop = (dw0 & PCI_EA_PP) >> 8;
2883 * If the Property is in the reserved range, try the Secondary
2886 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2887 prop = (dw0 & PCI_EA_SP) >> 16;
2888 if (prop > PCI_EA_P_BRIDGE_IO)
2891 res = pci_ea_get_resource(dev, bei, prop);
2893 pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
2897 flags = pci_ea_flags(dev, prop);
2899 pci_err(dev, "Unsupported EA properties: %#x\n", prop);
2904 pci_read_config_dword(dev, ent_offset, &base);
2905 start = (base & PCI_EA_FIELD_MASK);
2908 /* Read MaxOffset */
2909 pci_read_config_dword(dev, ent_offset, &max_offset);
2912 /* Read Base MSBs (if 64-bit entry) */
2913 if (base & PCI_EA_IS_64) {
2916 pci_read_config_dword(dev, ent_offset, &base_upper);
2919 flags |= IORESOURCE_MEM_64;
2921 /* entry starts above 32-bit boundary, can't use */
2922 if (!support_64 && base_upper)
2926 start |= ((u64)base_upper << 32);
2929 end = start + (max_offset | 0x03);
2931 /* Read MaxOffset MSBs (if 64-bit entry) */
2932 if (max_offset & PCI_EA_IS_64) {
2933 u32 max_offset_upper;
2935 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2938 flags |= IORESOURCE_MEM_64;
2940 /* entry too big, can't use */
2941 if (!support_64 && max_offset_upper)
2945 end += ((u64)max_offset_upper << 32);
2949 pci_err(dev, "EA Entry crosses address boundary\n");
2953 if (ent_size != ent_offset - offset) {
2954 pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
2955 ent_size, ent_offset - offset);
2959 res->name = pci_name(dev);
2964 if (bei <= PCI_EA_BEI_BAR5)
2965 pci_info(dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2967 else if (bei == PCI_EA_BEI_ROM)
2968 pci_info(dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2970 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2971 pci_info(dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2972 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2974 pci_info(dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2978 return offset + ent_size;
2981 /* Enhanced Allocation Initialization */
2982 void pci_ea_init(struct pci_dev *dev)
2989 /* find PCI EA capability in list */
2990 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2994 /* determine the number of entries */
2995 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2997 num_ent &= PCI_EA_NUM_ENT_MASK;
2999 offset = ea + PCI_EA_FIRST_ENT;
3001 /* Skip DWORD 2 for type 1 functions */
3002 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3005 /* parse each EA entry */
3006 for (i = 0; i < num_ent; ++i)
3007 offset = pci_ea_read(dev, offset);
3010 static void pci_add_saved_cap(struct pci_dev *pci_dev,
3011 struct pci_cap_saved_state *new_cap)
3013 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3017 * _pci_add_cap_save_buffer - allocate buffer for saving given
3018 * capability registers
3019 * @dev: the PCI device
3020 * @cap: the capability to allocate the buffer for
3021 * @extended: Standard or Extended capability ID
3022 * @size: requested size of the buffer
3024 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3025 bool extended, unsigned int size)
3028 struct pci_cap_saved_state *save_state;
3031 pos = pci_find_ext_capability(dev, cap);
3033 pos = pci_find_capability(dev, cap);
3038 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3042 save_state->cap.cap_nr = cap;
3043 save_state->cap.cap_extended = extended;
3044 save_state->cap.size = size;
3045 pci_add_saved_cap(dev, save_state);
3050 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3052 return _pci_add_cap_save_buffer(dev, cap, false, size);
3055 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3057 return _pci_add_cap_save_buffer(dev, cap, true, size);
3061 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3062 * @dev: the PCI device
3064 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3068 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3069 PCI_EXP_SAVE_REGS * sizeof(u16));
3071 pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3073 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3075 pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3077 error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3080 pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3082 pci_allocate_vc_save_buffers(dev);
3085 void pci_free_cap_save_buffers(struct pci_dev *dev)
3087 struct pci_cap_saved_state *tmp;
3088 struct hlist_node *n;
3090 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3095 * pci_configure_ari - enable or disable ARI forwarding
3096 * @dev: the PCI device
3098 * If @dev and its upstream bridge both support ARI, enable ARI in the
3099 * bridge. Otherwise, disable ARI in the bridge.
3101 void pci_configure_ari(struct pci_dev *dev)
3104 struct pci_dev *bridge;
3106 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3109 bridge = dev->bus->self;
3113 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3114 if (!(cap & PCI_EXP_DEVCAP2_ARI))
3117 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3118 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3119 PCI_EXP_DEVCTL2_ARI);
3120 bridge->ari_enabled = 1;
3122 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3123 PCI_EXP_DEVCTL2_ARI);
3124 bridge->ari_enabled = 0;
3128 static int pci_acs_enable;
3131 * pci_request_acs - ask for ACS to be enabled if supported
3133 void pci_request_acs(void)
3138 static const char *disable_acs_redir_param;
3141 * pci_disable_acs_redir - disable ACS redirect capabilities
3142 * @dev: the PCI device
3144 * For only devices specified in the disable_acs_redir parameter.
3146 static void pci_disable_acs_redir(struct pci_dev *dev)
3153 if (!disable_acs_redir_param)
3156 p = disable_acs_redir_param;
3158 ret = pci_dev_str_match(dev, p, &p);
3160 pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
3161 disable_acs_redir_param);
3164 } else if (ret == 1) {
3169 if (*p != ';' && *p != ',') {
3170 /* End of param or invalid format */
3179 if (!pci_dev_specific_disable_acs_redir(dev))
3182 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3184 pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
3188 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3190 /* P2P Request & Completion Redirect */
3191 ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
3193 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3195 pci_info(dev, "disabled ACS redirect\n");
3199 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
3200 * @dev: the PCI device
3202 static void pci_std_enable_acs(struct pci_dev *dev)
3208 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3212 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
3213 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3215 /* Source Validation */
3216 ctrl |= (cap & PCI_ACS_SV);
3218 /* P2P Request Redirect */
3219 ctrl |= (cap & PCI_ACS_RR);
3221 /* P2P Completion Redirect */
3222 ctrl |= (cap & PCI_ACS_CR);
3224 /* Upstream Forwarding */
3225 ctrl |= (cap & PCI_ACS_UF);
3227 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3231 * pci_enable_acs - enable ACS if hardware support it
3232 * @dev: the PCI device
3234 void pci_enable_acs(struct pci_dev *dev)
3236 if (!pci_acs_enable)
3237 goto disable_acs_redir;
3239 if (!pci_dev_specific_enable_acs(dev))
3240 goto disable_acs_redir;
3242 pci_std_enable_acs(dev);
3246 * Note: pci_disable_acs_redir() must be called even if ACS was not
3247 * enabled by the kernel because it may have been enabled by
3248 * platform firmware. So if we are told to disable it, we should
3249 * always disable it after setting the kernel's default
3252 pci_disable_acs_redir(dev);
3255 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3260 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
3265 * Except for egress control, capabilities are either required
3266 * or only required if controllable. Features missing from the
3267 * capability field can therefore be assumed as hard-wired enabled.
3269 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3270 acs_flags &= (cap | PCI_ACS_EC);
3272 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3273 return (ctrl & acs_flags) == acs_flags;
3277 * pci_acs_enabled - test ACS against required flags for a given device
3278 * @pdev: device to test
3279 * @acs_flags: required PCI ACS flags
3281 * Return true if the device supports the provided flags. Automatically
3282 * filters out flags that are not implemented on multifunction devices.
3284 * Note that this interface checks the effective ACS capabilities of the
3285 * device rather than the actual capabilities. For instance, most single
3286 * function endpoints are not required to support ACS because they have no
3287 * opportunity for peer-to-peer access. We therefore return 'true'
3288 * regardless of whether the device exposes an ACS capability. This makes
3289 * it much easier for callers of this function to ignore the actual type
3290 * or topology of the device when testing ACS support.
3292 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3296 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3301 * Conventional PCI and PCI-X devices never support ACS, either
3302 * effectively or actually. The shared bus topology implies that
3303 * any device on the bus can receive or snoop DMA.
3305 if (!pci_is_pcie(pdev))
3308 switch (pci_pcie_type(pdev)) {
3310 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3311 * but since their primary interface is PCI/X, we conservatively
3312 * handle them as we would a non-PCIe device.
3314 case PCI_EXP_TYPE_PCIE_BRIDGE:
3316 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
3317 * applicable... must never implement an ACS Extended Capability...".
3318 * This seems arbitrary, but we take a conservative interpretation
3319 * of this statement.
3321 case PCI_EXP_TYPE_PCI_BRIDGE:
3322 case PCI_EXP_TYPE_RC_EC:
3325 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3326 * implement ACS in order to indicate their peer-to-peer capabilities,
3327 * regardless of whether they are single- or multi-function devices.
3329 case PCI_EXP_TYPE_DOWNSTREAM:
3330 case PCI_EXP_TYPE_ROOT_PORT:
3331 return pci_acs_flags_enabled(pdev, acs_flags);
3333 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3334 * implemented by the remaining PCIe types to indicate peer-to-peer
3335 * capabilities, but only when they are part of a multifunction
3336 * device. The footnote for section 6.12 indicates the specific
3337 * PCIe types included here.
3339 case PCI_EXP_TYPE_ENDPOINT:
3340 case PCI_EXP_TYPE_UPSTREAM:
3341 case PCI_EXP_TYPE_LEG_END:
3342 case PCI_EXP_TYPE_RC_END:
3343 if (!pdev->multifunction)
3346 return pci_acs_flags_enabled(pdev, acs_flags);
3350 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3351 * to single function devices with the exception of downstream ports.
3357 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3358 * @start: starting downstream device
3359 * @end: ending upstream device or NULL to search to the root bus
3360 * @acs_flags: required flags
3362 * Walk up a device tree from start to end testing PCI ACS support. If
3363 * any step along the way does not support the required flags, return false.
3365 bool pci_acs_path_enabled(struct pci_dev *start,
3366 struct pci_dev *end, u16 acs_flags)
3368 struct pci_dev *pdev, *parent = start;
3373 if (!pci_acs_enabled(pdev, acs_flags))
3376 if (pci_is_root_bus(pdev->bus))
3377 return (end == NULL);
3379 parent = pdev->bus->self;
3380 } while (pdev != end);
3386 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3390 * Helper to find the position of the ctrl register for a BAR.
3391 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3392 * Returns -ENOENT if no ctrl register for the BAR could be found.
3394 static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3396 unsigned int pos, nbars, i;
3399 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3403 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3404 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3405 PCI_REBAR_CTRL_NBAR_SHIFT;
3407 for (i = 0; i < nbars; i++, pos += 8) {
3410 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3411 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3420 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3422 * @bar: BAR to query
3424 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3425 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3427 u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3432 pos = pci_rebar_find_pos(pdev, bar);
3436 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3437 return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3441 * pci_rebar_get_current_size - get the current size of a BAR
3443 * @bar: BAR to set size to
3445 * Read the size of a BAR from the resizable BAR config.
3446 * Returns size if found or negative error code.
3448 int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3453 pos = pci_rebar_find_pos(pdev, bar);
3457 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3458 return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3462 * pci_rebar_set_size - set a new size for a BAR
3464 * @bar: BAR to set size to
3465 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3467 * Set the new size of a BAR as defined in the spec.
3468 * Returns zero if resizing was successful, error code otherwise.
3470 int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3475 pos = pci_rebar_find_pos(pdev, bar);
3479 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3480 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3481 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3482 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3487 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3488 * @dev: the PCI device
3489 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3490 * PCI_EXP_DEVCAP2_ATOMIC_COMP32
3491 * PCI_EXP_DEVCAP2_ATOMIC_COMP64
3492 * PCI_EXP_DEVCAP2_ATOMIC_COMP128
3494 * Return 0 if all upstream bridges support AtomicOp routing, egress
3495 * blocking is disabled on all upstream ports, and the root port supports
3496 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3497 * AtomicOp completion), or negative otherwise.
3499 int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3501 struct pci_bus *bus = dev->bus;
3502 struct pci_dev *bridge;
3505 if (!pci_is_pcie(dev))
3509 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3510 * AtomicOp requesters. For now, we only support endpoints as
3511 * requesters and root ports as completers. No endpoints as
3512 * completers, and no peer-to-peer.
3515 switch (pci_pcie_type(dev)) {
3516 case PCI_EXP_TYPE_ENDPOINT:
3517 case PCI_EXP_TYPE_LEG_END:
3518 case PCI_EXP_TYPE_RC_END:
3524 while (bus->parent) {
3527 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3529 switch (pci_pcie_type(bridge)) {
3530 /* Ensure switch ports support AtomicOp routing */
3531 case PCI_EXP_TYPE_UPSTREAM:
3532 case PCI_EXP_TYPE_DOWNSTREAM:
3533 if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3537 /* Ensure root port supports all the sizes we care about */
3538 case PCI_EXP_TYPE_ROOT_PORT:
3539 if ((cap & cap_mask) != cap_mask)
3544 /* Ensure upstream ports don't block AtomicOps on egress */
3545 if (!bridge->has_secondary_link) {
3546 pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3548 if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3555 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3556 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3559 EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3562 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3563 * @dev: the PCI device
3564 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3566 * Perform INTx swizzling for a device behind one level of bridge. This is
3567 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3568 * behind bridges on add-in cards. For devices with ARI enabled, the slot
3569 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3570 * the PCI Express Base Specification, Revision 2.1)
3572 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3576 if (pci_ari_enabled(dev->bus))
3579 slot = PCI_SLOT(dev->devfn);
3581 return (((pin - 1) + slot) % 4) + 1;
3584 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3592 while (!pci_is_root_bus(dev->bus)) {
3593 pin = pci_swizzle_interrupt_pin(dev, pin);
3594 dev = dev->bus->self;
3601 * pci_common_swizzle - swizzle INTx all the way to root bridge
3602 * @dev: the PCI device
3603 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3605 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3606 * bridges all the way up to a PCI root bus.
3608 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3612 while (!pci_is_root_bus(dev->bus)) {
3613 pin = pci_swizzle_interrupt_pin(dev, pin);
3614 dev = dev->bus->self;
3617 return PCI_SLOT(dev->devfn);
3619 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3622 * pci_release_region - Release a PCI bar
3623 * @pdev: PCI device whose resources were previously reserved by
3624 * pci_request_region()
3625 * @bar: BAR to release
3627 * Releases the PCI I/O and memory resources previously reserved by a
3628 * successful call to pci_request_region(). Call this function only
3629 * after all use of the PCI regions has ceased.
3631 void pci_release_region(struct pci_dev *pdev, int bar)
3633 struct pci_devres *dr;
3635 if (pci_resource_len(pdev, bar) == 0)
3637 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3638 release_region(pci_resource_start(pdev, bar),
3639 pci_resource_len(pdev, bar));
3640 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3641 release_mem_region(pci_resource_start(pdev, bar),
3642 pci_resource_len(pdev, bar));
3644 dr = find_pci_dr(pdev);
3646 dr->region_mask &= ~(1 << bar);
3648 EXPORT_SYMBOL(pci_release_region);
3651 * __pci_request_region - Reserved PCI I/O and memory resource
3652 * @pdev: PCI device whose resources are to be reserved
3653 * @bar: BAR to be reserved
3654 * @res_name: Name to be associated with resource.
3655 * @exclusive: whether the region access is exclusive or not
3657 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3658 * being reserved by owner @res_name. Do not access any
3659 * address inside the PCI regions unless this call returns
3662 * If @exclusive is set, then the region is marked so that userspace
3663 * is explicitly not allowed to map the resource via /dev/mem or
3664 * sysfs MMIO access.
3666 * Returns 0 on success, or %EBUSY on error. A warning
3667 * message is also printed on failure.
3669 static int __pci_request_region(struct pci_dev *pdev, int bar,
3670 const char *res_name, int exclusive)
3672 struct pci_devres *dr;
3674 if (pci_resource_len(pdev, bar) == 0)
3677 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3678 if (!request_region(pci_resource_start(pdev, bar),
3679 pci_resource_len(pdev, bar), res_name))
3681 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3682 if (!__request_mem_region(pci_resource_start(pdev, bar),
3683 pci_resource_len(pdev, bar), res_name,
3688 dr = find_pci_dr(pdev);
3690 dr->region_mask |= 1 << bar;
3695 pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3696 &pdev->resource[bar]);
3701 * pci_request_region - Reserve PCI I/O and memory resource
3702 * @pdev: PCI device whose resources are to be reserved
3703 * @bar: BAR to be reserved
3704 * @res_name: Name to be associated with resource
3706 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3707 * being reserved by owner @res_name. Do not access any
3708 * address inside the PCI regions unless this call returns
3711 * Returns 0 on success, or %EBUSY on error. A warning
3712 * message is also printed on failure.
3714 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3716 return __pci_request_region(pdev, bar, res_name, 0);
3718 EXPORT_SYMBOL(pci_request_region);
3721 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3722 * @pdev: PCI device whose resources were previously reserved
3723 * @bars: Bitmask of BARs to be released
3725 * Release selected PCI I/O and memory resources previously reserved.
3726 * Call this function only after all use of the PCI regions has ceased.
3728 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3732 for (i = 0; i < 6; i++)
3733 if (bars & (1 << i))
3734 pci_release_region(pdev, i);
3736 EXPORT_SYMBOL(pci_release_selected_regions);
3738 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3739 const char *res_name, int excl)
3743 for (i = 0; i < 6; i++)
3744 if (bars & (1 << i))
3745 if (__pci_request_region(pdev, i, res_name, excl))
3751 if (bars & (1 << i))
3752 pci_release_region(pdev, i);
3759 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3760 * @pdev: PCI device whose resources are to be reserved
3761 * @bars: Bitmask of BARs to be requested
3762 * @res_name: Name to be associated with resource
3764 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3765 const char *res_name)
3767 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3769 EXPORT_SYMBOL(pci_request_selected_regions);
3771 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3772 const char *res_name)
3774 return __pci_request_selected_regions(pdev, bars, res_name,
3775 IORESOURCE_EXCLUSIVE);
3777 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3780 * pci_release_regions - Release reserved PCI I/O and memory resources
3781 * @pdev: PCI device whose resources were previously reserved by
3782 * pci_request_regions()
3784 * Releases all PCI I/O and memory resources previously reserved by a
3785 * successful call to pci_request_regions(). Call this function only
3786 * after all use of the PCI regions has ceased.
3789 void pci_release_regions(struct pci_dev *pdev)
3791 pci_release_selected_regions(pdev, (1 << 6) - 1);
3793 EXPORT_SYMBOL(pci_release_regions);
3796 * pci_request_regions - Reserve PCI I/O and memory resources
3797 * @pdev: PCI device whose resources are to be reserved
3798 * @res_name: Name to be associated with resource.
3800 * Mark all PCI regions associated with PCI device @pdev as
3801 * being reserved by owner @res_name. Do not access any
3802 * address inside the PCI regions unless this call returns
3805 * Returns 0 on success, or %EBUSY on error. A warning
3806 * message is also printed on failure.
3808 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3810 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3812 EXPORT_SYMBOL(pci_request_regions);
3815 * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
3816 * @pdev: PCI device whose resources are to be reserved
3817 * @res_name: Name to be associated with resource.
3819 * Mark all PCI regions associated with PCI device @pdev as being reserved
3820 * by owner @res_name. Do not access any address inside the PCI regions
3821 * unless this call returns successfully.
3823 * pci_request_regions_exclusive() will mark the region so that /dev/mem
3824 * and the sysfs MMIO access will not be allowed.
3826 * Returns 0 on success, or %EBUSY on error. A warning message is also
3827 * printed on failure.
3829 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3831 return pci_request_selected_regions_exclusive(pdev,
3832 ((1 << 6) - 1), res_name);
3834 EXPORT_SYMBOL(pci_request_regions_exclusive);
3837 * Record the PCI IO range (expressed as CPU physical address + size).
3838 * Return a negative value if an error has occurred, zero otherwise
3840 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3841 resource_size_t size)
3845 struct logic_pio_hwaddr *range;
3847 if (!size || addr + size < addr)
3850 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3854 range->fwnode = fwnode;
3856 range->hw_start = addr;
3857 range->flags = LOGIC_PIO_CPU_MMIO;
3859 ret = logic_pio_register_range(range);
3867 phys_addr_t pci_pio_to_address(unsigned long pio)
3869 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3872 if (pio >= MMIO_UPPER_LIMIT)
3875 address = logic_pio_to_hwaddr(pio);
3881 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3884 return logic_pio_trans_cpuaddr(address);
3886 if (address > IO_SPACE_LIMIT)
3887 return (unsigned long)-1;
3889 return (unsigned long) address;
3894 * pci_remap_iospace - Remap the memory mapped I/O space
3895 * @res: Resource describing the I/O space
3896 * @phys_addr: physical address of range to be mapped
3898 * Remap the memory mapped I/O space described by the @res and the CPU
3899 * physical address @phys_addr into virtual address space. Only
3900 * architectures that have memory mapped IO functions defined (and the
3901 * PCI_IOBASE value defined) should call this function.
3903 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3905 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3906 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3908 if (!(res->flags & IORESOURCE_IO))
3911 if (res->end > IO_SPACE_LIMIT)
3914 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3915 pgprot_device(PAGE_KERNEL));
3918 * This architecture does not have memory mapped I/O space,
3919 * so this function should never be called
3921 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3925 EXPORT_SYMBOL(pci_remap_iospace);
3928 * pci_unmap_iospace - Unmap the memory mapped I/O space
3929 * @res: resource to be unmapped
3931 * Unmap the CPU virtual address @res from virtual address space. Only
3932 * architectures that have memory mapped IO functions defined (and the
3933 * PCI_IOBASE value defined) should call this function.
3935 void pci_unmap_iospace(struct resource *res)
3937 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3938 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3940 unmap_kernel_range(vaddr, resource_size(res));
3943 EXPORT_SYMBOL(pci_unmap_iospace);
3945 static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
3947 struct resource **res = ptr;
3949 pci_unmap_iospace(*res);
3953 * devm_pci_remap_iospace - Managed pci_remap_iospace()
3954 * @dev: Generic device to remap IO address for
3955 * @res: Resource describing the I/O space
3956 * @phys_addr: physical address of range to be mapped
3958 * Managed pci_remap_iospace(). Map is automatically unmapped on driver
3961 int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
3962 phys_addr_t phys_addr)
3964 const struct resource **ptr;
3967 ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
3971 error = pci_remap_iospace(res, phys_addr);
3976 devres_add(dev, ptr);
3981 EXPORT_SYMBOL(devm_pci_remap_iospace);
3984 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3985 * @dev: Generic device to remap IO address for
3986 * @offset: Resource address to map
3987 * @size: Size of map
3989 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
3992 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3993 resource_size_t offset,
3994 resource_size_t size)
3996 void __iomem **ptr, *addr;
3998 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
4002 addr = pci_remap_cfgspace(offset, size);
4005 devres_add(dev, ptr);
4011 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
4014 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
4015 * @dev: generic device to handle the resource for
4016 * @res: configuration space resource to be handled
4018 * Checks that a resource is a valid memory region, requests the memory
4019 * region and ioremaps with pci_remap_cfgspace() API that ensures the
4020 * proper PCI configuration space memory attributes are guaranteed.
4022 * All operations are managed and will be undone on driver detach.
4024 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
4025 * on failure. Usage example::
4027 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
4028 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
4030 * return PTR_ERR(base);
4032 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
4033 struct resource *res)
4035 resource_size_t size;
4037 void __iomem *dest_ptr;
4041 if (!res || resource_type(res) != IORESOURCE_MEM) {
4042 dev_err(dev, "invalid resource\n");
4043 return IOMEM_ERR_PTR(-EINVAL);
4046 size = resource_size(res);
4047 name = res->name ?: dev_name(dev);
4049 if (!devm_request_mem_region(dev, res->start, size, name)) {
4050 dev_err(dev, "can't request region for resource %pR\n", res);
4051 return IOMEM_ERR_PTR(-EBUSY);
4054 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4056 dev_err(dev, "ioremap failed for resource %pR\n", res);
4057 devm_release_mem_region(dev, res->start, size);
4058 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4063 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4065 static void __pci_set_master(struct pci_dev *dev, bool enable)
4069 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4071 cmd = old_cmd | PCI_COMMAND_MASTER;
4073 cmd = old_cmd & ~PCI_COMMAND_MASTER;
4074 if (cmd != old_cmd) {
4075 pci_dbg(dev, "%s bus mastering\n",
4076 enable ? "enabling" : "disabling");
4077 pci_write_config_word(dev, PCI_COMMAND, cmd);
4079 dev->is_busmaster = enable;
4083 * pcibios_setup - process "pci=" kernel boot arguments
4084 * @str: string used to pass in "pci=" kernel boot arguments
4086 * Process kernel boot arguments. This is the default implementation.
4087 * Architecture specific implementations can override this as necessary.
4089 char * __weak __init pcibios_setup(char *str)
4095 * pcibios_set_master - enable PCI bus-mastering for device dev
4096 * @dev: the PCI device to enable
4098 * Enables PCI bus-mastering for the device. This is the default
4099 * implementation. Architecture specific implementations can override
4100 * this if necessary.
4102 void __weak pcibios_set_master(struct pci_dev *dev)
4106 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4107 if (pci_is_pcie(dev))
4110 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4112 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4113 else if (lat > pcibios_max_latency)
4114 lat = pcibios_max_latency;
4118 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4122 * pci_set_master - enables bus-mastering for device dev
4123 * @dev: the PCI device to enable
4125 * Enables bus-mastering on the device and calls pcibios_set_master()
4126 * to do the needed arch specific settings.
4128 void pci_set_master(struct pci_dev *dev)
4130 __pci_set_master(dev, true);
4131 pcibios_set_master(dev);
4133 EXPORT_SYMBOL(pci_set_master);
4136 * pci_clear_master - disables bus-mastering for device dev
4137 * @dev: the PCI device to disable
4139 void pci_clear_master(struct pci_dev *dev)
4141 __pci_set_master(dev, false);
4143 EXPORT_SYMBOL(pci_clear_master);
4146 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4147 * @dev: the PCI device for which MWI is to be enabled
4149 * Helper function for pci_set_mwi.
4150 * Originally copied from drivers/net/acenic.c.
4153 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4155 int pci_set_cacheline_size(struct pci_dev *dev)
4159 if (!pci_cache_line_size)
4162 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4163 equal to or multiple of the right value. */
4164 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4165 if (cacheline_size >= pci_cache_line_size &&
4166 (cacheline_size % pci_cache_line_size) == 0)
4169 /* Write the correct value. */
4170 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4172 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4173 if (cacheline_size == pci_cache_line_size)
4176 pci_info(dev, "cache line size of %d is not supported\n",
4177 pci_cache_line_size << 2);
4181 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4184 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4185 * @dev: the PCI device for which MWI is enabled
4187 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4189 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4191 int pci_set_mwi(struct pci_dev *dev)
4193 #ifdef PCI_DISABLE_MWI
4199 rc = pci_set_cacheline_size(dev);
4203 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4204 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4205 pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4206 cmd |= PCI_COMMAND_INVALIDATE;
4207 pci_write_config_word(dev, PCI_COMMAND, cmd);
4212 EXPORT_SYMBOL(pci_set_mwi);
4215 * pcim_set_mwi - a device-managed pci_set_mwi()
4216 * @dev: the PCI device for which MWI is enabled
4218 * Managed pci_set_mwi().
4220 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4222 int pcim_set_mwi(struct pci_dev *dev)
4224 struct pci_devres *dr;
4226 dr = find_pci_dr(dev);
4231 return pci_set_mwi(dev);
4233 EXPORT_SYMBOL(pcim_set_mwi);
4236 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4237 * @dev: the PCI device for which MWI is enabled
4239 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4240 * Callers are not required to check the return value.
4242 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4244 int pci_try_set_mwi(struct pci_dev *dev)
4246 #ifdef PCI_DISABLE_MWI
4249 return pci_set_mwi(dev);
4252 EXPORT_SYMBOL(pci_try_set_mwi);
4255 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4256 * @dev: the PCI device to disable
4258 * Disables PCI Memory-Write-Invalidate transaction on the device
4260 void pci_clear_mwi(struct pci_dev *dev)
4262 #ifndef PCI_DISABLE_MWI
4265 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4266 if (cmd & PCI_COMMAND_INVALIDATE) {
4267 cmd &= ~PCI_COMMAND_INVALIDATE;
4268 pci_write_config_word(dev, PCI_COMMAND, cmd);
4272 EXPORT_SYMBOL(pci_clear_mwi);
4275 * pci_intx - enables/disables PCI INTx for device dev
4276 * @pdev: the PCI device to operate on
4277 * @enable: boolean: whether to enable or disable PCI INTx
4279 * Enables/disables PCI INTx for device @pdev
4281 void pci_intx(struct pci_dev *pdev, int enable)
4283 u16 pci_command, new;
4285 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4288 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4290 new = pci_command | PCI_COMMAND_INTX_DISABLE;
4292 if (new != pci_command) {
4293 struct pci_devres *dr;
4295 pci_write_config_word(pdev, PCI_COMMAND, new);
4297 dr = find_pci_dr(pdev);
4298 if (dr && !dr->restore_intx) {
4299 dr->restore_intx = 1;
4300 dr->orig_intx = !enable;
4304 EXPORT_SYMBOL_GPL(pci_intx);
4306 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4308 struct pci_bus *bus = dev->bus;
4309 bool mask_updated = true;
4310 u32 cmd_status_dword;
4311 u16 origcmd, newcmd;
4312 unsigned long flags;
4316 * We do a single dword read to retrieve both command and status.
4317 * Document assumptions that make this possible.
4319 BUILD_BUG_ON(PCI_COMMAND % 4);
4320 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4322 raw_spin_lock_irqsave(&pci_lock, flags);
4324 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4326 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4329 * Check interrupt status register to see whether our device
4330 * triggered the interrupt (when masking) or the next IRQ is
4331 * already pending (when unmasking).
4333 if (mask != irq_pending) {
4334 mask_updated = false;
4338 origcmd = cmd_status_dword;
4339 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4341 newcmd |= PCI_COMMAND_INTX_DISABLE;
4342 if (newcmd != origcmd)
4343 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4346 raw_spin_unlock_irqrestore(&pci_lock, flags);
4348 return mask_updated;
4352 * pci_check_and_mask_intx - mask INTx on pending interrupt
4353 * @dev: the PCI device to operate on
4355 * Check if the device dev has its INTx line asserted, mask it and return
4356 * true in that case. False is returned if no interrupt was pending.
4358 bool pci_check_and_mask_intx(struct pci_dev *dev)
4360 return pci_check_and_set_intx_mask(dev, true);
4362 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4365 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4366 * @dev: the PCI device to operate on
4368 * Check if the device dev has its INTx line asserted, unmask it if not and
4369 * return true. False is returned and the mask remains active if there was
4370 * still an interrupt pending.
4372 bool pci_check_and_unmask_intx(struct pci_dev *dev)
4374 return pci_check_and_set_intx_mask(dev, false);
4376 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4379 * pci_wait_for_pending_transaction - wait for pending transaction
4380 * @dev: the PCI device to operate on
4382 * Return 0 if transaction is pending 1 otherwise.
4384 int pci_wait_for_pending_transaction(struct pci_dev *dev)
4386 if (!pci_is_pcie(dev))
4389 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4390 PCI_EXP_DEVSTA_TRPND);
4392 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4394 static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
4400 * After reset, the device should not silently discard config
4401 * requests, but it may still indicate that it needs more time by
4402 * responding to them with CRS completions. The Root Port will
4403 * generally synthesize ~0 data to complete the read (except when
4404 * CRS SV is enabled and the read was for the Vendor ID; in that
4405 * case it synthesizes 0x0001 data).
4407 * Wait for the device to return a non-CRS completion. Read the
4408 * Command register instead of Vendor ID so we don't have to
4409 * contend with the CRS SV value.
4411 pci_read_config_dword(dev, PCI_COMMAND, &id);
4413 if (delay > timeout) {
4414 pci_warn(dev, "not ready %dms after %s; giving up\n",
4415 delay - 1, reset_type);
4420 pci_info(dev, "not ready %dms after %s; waiting\n",
4421 delay - 1, reset_type);
4425 pci_read_config_dword(dev, PCI_COMMAND, &id);
4429 pci_info(dev, "ready %dms after %s\n", delay - 1,
4436 * pcie_has_flr - check if a device supports function level resets
4437 * @dev: device to check
4439 * Returns true if the device advertises support for PCIe function level
4442 bool pcie_has_flr(struct pci_dev *dev)
4446 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4449 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4450 return cap & PCI_EXP_DEVCAP_FLR;
4452 EXPORT_SYMBOL_GPL(pcie_has_flr);
4455 * pcie_flr - initiate a PCIe function level reset
4456 * @dev: device to reset
4458 * Initiate a function level reset on @dev. The caller should ensure the
4459 * device supports FLR before calling this function, e.g. by using the
4460 * pcie_has_flr() helper.
4462 int pcie_flr(struct pci_dev *dev)
4464 if (!pci_wait_for_pending_transaction(dev))
4465 pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4467 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4473 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4474 * 100ms, but may silently discard requests while the FLR is in
4475 * progress. Wait 100ms before trying to access the device.
4479 return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4481 EXPORT_SYMBOL_GPL(pcie_flr);
4483 static int pci_af_flr(struct pci_dev *dev, int probe)
4488 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4492 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4495 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4496 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4503 * Wait for Transaction Pending bit to clear. A word-aligned test
4504 * is used, so we use the conrol offset rather than status and shift
4505 * the test bit to match.
4507 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4508 PCI_AF_STATUS_TP << 8))
4509 pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4511 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4517 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4518 * updated 27 July 2006; a device must complete an FLR within
4519 * 100ms, but may silently discard requests while the FLR is in
4520 * progress. Wait 100ms before trying to access the device.
4524 return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4528 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4529 * @dev: Device to reset.
4530 * @probe: If set, only check if the device can be reset this way.
4532 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4533 * unset, it will be reinitialized internally when going from PCI_D3hot to
4534 * PCI_D0. If that's the case and the device is not in a low-power state
4535 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4537 * NOTE: This causes the caller to sleep for twice the device power transition
4538 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4539 * by default (i.e. unless the @dev's d3_delay field has a different value).
4540 * Moreover, only devices in D0 can be reset by this function.
4542 static int pci_pm_reset(struct pci_dev *dev, int probe)
4546 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4549 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4550 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4556 if (dev->current_state != PCI_D0)
4559 csr &= ~PCI_PM_CTRL_STATE_MASK;
4561 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4562 pci_dev_d3_sleep(dev);
4564 csr &= ~PCI_PM_CTRL_STATE_MASK;
4566 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4567 pci_dev_d3_sleep(dev);
4569 return pci_dev_wait(dev, "PM D3->D0", PCIE_RESET_READY_POLL_MS);
4572 * pcie_wait_for_link - Wait until link is active or inactive
4573 * @pdev: Bridge device
4574 * @active: waiting for active or inactive?
4576 * Use this to wait till link becomes active or inactive.
4578 bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4585 * Some controllers might not implement link active reporting. In this
4586 * case, we wait for 1000 + 100 ms.
4588 if (!pdev->link_active_reporting) {
4594 * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4595 * after which we should expect an link active if the reset was
4596 * successful. If so, software must wait a minimum 100ms before sending
4597 * configuration requests to devices downstream this port.
4599 * If the link fails to activate, either the device was physically
4600 * removed or the link is permanently failed.
4605 pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4606 ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4616 else if (ret != active)
4617 pci_info(pdev, "Data Link Layer Link Active not %s in 1000 msec\n",
4618 active ? "set" : "cleared");
4619 return ret == active;
4622 void pci_reset_secondary_bus(struct pci_dev *dev)
4626 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4627 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4628 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4631 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4632 * this to 2ms to ensure that we meet the minimum requirement.
4636 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4637 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4640 * Trhfa for conventional PCI is 2^25 clock cycles.
4641 * Assuming a minimum 33MHz clock this results in a 1s
4642 * delay before we can consider subordinate devices to
4643 * be re-initialized. PCIe has some ways to shorten this,
4644 * but we don't make use of them yet.
4649 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4651 pci_reset_secondary_bus(dev);
4655 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4656 * @dev: Bridge device
4658 * Use the bridge control register to assert reset on the secondary bus.
4659 * Devices on the secondary bus are left in power-on state.
4661 int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4663 pcibios_reset_secondary_bus(dev);
4665 return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4667 EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4669 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4671 struct pci_dev *pdev;
4673 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4674 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4677 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4684 return pci_bridge_secondary_bus_reset(dev->bus->self);
4687 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4691 if (!hotplug || !try_module_get(hotplug->owner))
4694 if (hotplug->ops->reset_slot)
4695 rc = hotplug->ops->reset_slot(hotplug, probe);
4697 module_put(hotplug->owner);
4702 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4704 struct pci_dev *pdev;
4706 if (dev->subordinate || !dev->slot ||
4707 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4710 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4711 if (pdev != dev && pdev->slot == dev->slot)
4714 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4717 static void pci_dev_lock(struct pci_dev *dev)
4719 pci_cfg_access_lock(dev);
4720 /* block PM suspend, driver probe, etc. */
4721 device_lock(&dev->dev);
4724 /* Return 1 on successful lock, 0 on contention */
4725 static int pci_dev_trylock(struct pci_dev *dev)
4727 if (pci_cfg_access_trylock(dev)) {
4728 if (device_trylock(&dev->dev))
4730 pci_cfg_access_unlock(dev);
4736 static void pci_dev_unlock(struct pci_dev *dev)
4738 device_unlock(&dev->dev);
4739 pci_cfg_access_unlock(dev);
4742 static void pci_dev_save_and_disable(struct pci_dev *dev)
4744 const struct pci_error_handlers *err_handler =
4745 dev->driver ? dev->driver->err_handler : NULL;
4748 * dev->driver->err_handler->reset_prepare() is protected against
4749 * races with ->remove() by the device lock, which must be held by
4752 if (err_handler && err_handler->reset_prepare)
4753 err_handler->reset_prepare(dev);
4756 * Wake-up device prior to save. PM registers default to D0 after
4757 * reset and a simple register restore doesn't reliably return
4758 * to a non-D0 state anyway.
4760 pci_set_power_state(dev, PCI_D0);
4762 pci_save_state(dev);
4764 * Disable the device by clearing the Command register, except for
4765 * INTx-disable which is set. This not only disables MMIO and I/O port
4766 * BARs, but also prevents the device from being Bus Master, preventing
4767 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4768 * compliant devices, INTx-disable prevents legacy interrupts.
4770 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4773 static void pci_dev_restore(struct pci_dev *dev)
4775 const struct pci_error_handlers *err_handler =
4776 dev->driver ? dev->driver->err_handler : NULL;
4778 pci_restore_state(dev);
4781 * dev->driver->err_handler->reset_done() is protected against
4782 * races with ->remove() by the device lock, which must be held by
4785 if (err_handler && err_handler->reset_done)
4786 err_handler->reset_done(dev);
4790 * __pci_reset_function_locked - reset a PCI device function while holding
4791 * the @dev mutex lock.
4792 * @dev: PCI device to reset
4794 * Some devices allow an individual function to be reset without affecting
4795 * other functions in the same device. The PCI device must be responsive
4796 * to PCI config space in order to use this function.
4798 * The device function is presumed to be unused and the caller is holding
4799 * the device mutex lock when this function is called.
4801 * Resetting the device will make the contents of PCI configuration space
4802 * random, so any caller of this must be prepared to reinitialise the
4803 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4806 * Returns 0 if the device function was successfully reset or negative if the
4807 * device doesn't support resetting a single function.
4809 int __pci_reset_function_locked(struct pci_dev *dev)
4816 * A reset method returns -ENOTTY if it doesn't support this device
4817 * and we should try the next method.
4819 * If it returns 0 (success), we're finished. If it returns any
4820 * other error, we're also finished: this indicates that further
4821 * reset mechanisms might be broken on the device.
4823 rc = pci_dev_specific_reset(dev, 0);
4826 if (pcie_has_flr(dev)) {
4831 rc = pci_af_flr(dev, 0);
4834 rc = pci_pm_reset(dev, 0);
4837 rc = pci_dev_reset_slot_function(dev, 0);
4840 return pci_parent_bus_reset(dev, 0);
4842 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4845 * pci_probe_reset_function - check whether the device can be safely reset
4846 * @dev: PCI device to reset
4848 * Some devices allow an individual function to be reset without affecting
4849 * other functions in the same device. The PCI device must be responsive
4850 * to PCI config space in order to use this function.
4852 * Returns 0 if the device function can be reset or negative if the
4853 * device doesn't support resetting a single function.
4855 int pci_probe_reset_function(struct pci_dev *dev)
4861 rc = pci_dev_specific_reset(dev, 1);
4864 if (pcie_has_flr(dev))
4866 rc = pci_af_flr(dev, 1);
4869 rc = pci_pm_reset(dev, 1);
4872 rc = pci_dev_reset_slot_function(dev, 1);
4876 return pci_parent_bus_reset(dev, 1);
4880 * pci_reset_function - quiesce and reset a PCI device function
4881 * @dev: PCI device to reset
4883 * Some devices allow an individual function to be reset without affecting
4884 * other functions in the same device. The PCI device must be responsive
4885 * to PCI config space in order to use this function.
4887 * This function does not just reset the PCI portion of a device, but
4888 * clears all the state associated with the device. This function differs
4889 * from __pci_reset_function_locked() in that it saves and restores device state
4890 * over the reset and takes the PCI device lock.
4892 * Returns 0 if the device function was successfully reset or negative if the
4893 * device doesn't support resetting a single function.
4895 int pci_reset_function(struct pci_dev *dev)
4903 pci_dev_save_and_disable(dev);
4905 rc = __pci_reset_function_locked(dev);
4907 pci_dev_restore(dev);
4908 pci_dev_unlock(dev);
4912 EXPORT_SYMBOL_GPL(pci_reset_function);
4915 * pci_reset_function_locked - quiesce and reset a PCI device function
4916 * @dev: PCI device to reset
4918 * Some devices allow an individual function to be reset without affecting
4919 * other functions in the same device. The PCI device must be responsive
4920 * to PCI config space in order to use this function.
4922 * This function does not just reset the PCI portion of a device, but
4923 * clears all the state associated with the device. This function differs
4924 * from __pci_reset_function_locked() in that it saves and restores device state
4925 * over the reset. It also differs from pci_reset_function() in that it
4926 * requires the PCI device lock to be held.
4928 * Returns 0 if the device function was successfully reset or negative if the
4929 * device doesn't support resetting a single function.
4931 int pci_reset_function_locked(struct pci_dev *dev)
4938 pci_dev_save_and_disable(dev);
4940 rc = __pci_reset_function_locked(dev);
4942 pci_dev_restore(dev);
4946 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4949 * pci_try_reset_function - quiesce and reset a PCI device function
4950 * @dev: PCI device to reset
4952 * Same as above, except return -EAGAIN if unable to lock device.
4954 int pci_try_reset_function(struct pci_dev *dev)
4961 if (!pci_dev_trylock(dev))
4964 pci_dev_save_and_disable(dev);
4965 rc = __pci_reset_function_locked(dev);
4966 pci_dev_restore(dev);
4967 pci_dev_unlock(dev);
4971 EXPORT_SYMBOL_GPL(pci_try_reset_function);
4973 /* Do any devices on or below this bus prevent a bus reset? */
4974 static bool pci_bus_resetable(struct pci_bus *bus)
4976 struct pci_dev *dev;
4979 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4982 list_for_each_entry(dev, &bus->devices, bus_list) {
4983 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4984 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4991 /* Lock devices from the top of the tree down */
4992 static void pci_bus_lock(struct pci_bus *bus)
4994 struct pci_dev *dev;
4996 list_for_each_entry(dev, &bus->devices, bus_list) {
4998 if (dev->subordinate)
4999 pci_bus_lock(dev->subordinate);
5003 /* Unlock devices from the bottom of the tree up */
5004 static void pci_bus_unlock(struct pci_bus *bus)
5006 struct pci_dev *dev;
5008 list_for_each_entry(dev, &bus->devices, bus_list) {
5009 if (dev->subordinate)
5010 pci_bus_unlock(dev->subordinate);
5011 pci_dev_unlock(dev);
5015 /* Return 1 on successful lock, 0 on contention */
5016 static int pci_bus_trylock(struct pci_bus *bus)
5018 struct pci_dev *dev;
5020 list_for_each_entry(dev, &bus->devices, bus_list) {
5021 if (!pci_dev_trylock(dev))
5023 if (dev->subordinate) {
5024 if (!pci_bus_trylock(dev->subordinate)) {
5025 pci_dev_unlock(dev);
5033 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5034 if (dev->subordinate)
5035 pci_bus_unlock(dev->subordinate);
5036 pci_dev_unlock(dev);
5041 /* Do any devices on or below this slot prevent a bus reset? */
5042 static bool pci_slot_resetable(struct pci_slot *slot)
5044 struct pci_dev *dev;
5046 if (slot->bus->self &&
5047 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5050 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5051 if (!dev->slot || dev->slot != slot)
5053 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5054 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
5061 /* Lock devices from the top of the tree down */
5062 static void pci_slot_lock(struct pci_slot *slot)
5064 struct pci_dev *dev;
5066 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5067 if (!dev->slot || dev->slot != slot)
5070 if (dev->subordinate)
5071 pci_bus_lock(dev->subordinate);
5075 /* Unlock devices from the bottom of the tree up */
5076 static void pci_slot_unlock(struct pci_slot *slot)
5078 struct pci_dev *dev;
5080 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5081 if (!dev->slot || dev->slot != slot)
5083 if (dev->subordinate)
5084 pci_bus_unlock(dev->subordinate);
5085 pci_dev_unlock(dev);
5089 /* Return 1 on successful lock, 0 on contention */
5090 static int pci_slot_trylock(struct pci_slot *slot)
5092 struct pci_dev *dev;
5094 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5095 if (!dev->slot || dev->slot != slot)
5097 if (!pci_dev_trylock(dev))
5099 if (dev->subordinate) {
5100 if (!pci_bus_trylock(dev->subordinate)) {
5101 pci_dev_unlock(dev);
5109 list_for_each_entry_continue_reverse(dev,
5110 &slot->bus->devices, bus_list) {
5111 if (!dev->slot || dev->slot != slot)
5113 if (dev->subordinate)
5114 pci_bus_unlock(dev->subordinate);
5115 pci_dev_unlock(dev);
5121 * Save and disable devices from the top of the tree down while holding
5122 * the @dev mutex lock for the entire tree.
5124 static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5126 struct pci_dev *dev;
5128 list_for_each_entry(dev, &bus->devices, bus_list) {
5129 pci_dev_save_and_disable(dev);
5130 if (dev->subordinate)
5131 pci_bus_save_and_disable_locked(dev->subordinate);
5136 * Restore devices from top of the tree down while holding @dev mutex lock
5137 * for the entire tree. Parent bridges need to be restored before we can
5138 * get to subordinate devices.
5140 static void pci_bus_restore_locked(struct pci_bus *bus)
5142 struct pci_dev *dev;
5144 list_for_each_entry(dev, &bus->devices, bus_list) {
5145 pci_dev_restore(dev);
5146 if (dev->subordinate)
5147 pci_bus_restore_locked(dev->subordinate);
5152 * Save and disable devices from the top of the tree down while holding
5153 * the @dev mutex lock for the entire tree.
5155 static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5157 struct pci_dev *dev;
5159 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5160 if (!dev->slot || dev->slot != slot)
5162 pci_dev_save_and_disable(dev);
5163 if (dev->subordinate)
5164 pci_bus_save_and_disable_locked(dev->subordinate);
5169 * Restore devices from top of the tree down while holding @dev mutex lock
5170 * for the entire tree. Parent bridges need to be restored before we can
5171 * get to subordinate devices.
5173 static void pci_slot_restore_locked(struct pci_slot *slot)
5175 struct pci_dev *dev;
5177 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5178 if (!dev->slot || dev->slot != slot)
5180 pci_dev_restore(dev);
5181 if (dev->subordinate)
5182 pci_bus_restore_locked(dev->subordinate);
5186 static int pci_slot_reset(struct pci_slot *slot, int probe)
5190 if (!slot || !pci_slot_resetable(slot))
5194 pci_slot_lock(slot);
5198 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5201 pci_slot_unlock(slot);
5207 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5208 * @slot: PCI slot to probe
5210 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5212 int pci_probe_reset_slot(struct pci_slot *slot)
5214 return pci_slot_reset(slot, 1);
5216 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5219 * __pci_reset_slot - Try to reset a PCI slot
5220 * @slot: PCI slot to reset
5222 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5223 * independent of other slots. For instance, some slots may support slot power
5224 * control. In the case of a 1:1 bus to slot architecture, this function may
5225 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5226 * Generally a slot reset should be attempted before a bus reset. All of the
5227 * function of the slot and any subordinate buses behind the slot are reset
5228 * through this function. PCI config space of all devices in the slot and
5229 * behind the slot is saved before and restored after reset.
5231 * Same as above except return -EAGAIN if the slot cannot be locked
5233 static int __pci_reset_slot(struct pci_slot *slot)
5237 rc = pci_slot_reset(slot, 1);
5241 if (pci_slot_trylock(slot)) {
5242 pci_slot_save_and_disable_locked(slot);
5244 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
5245 pci_slot_restore_locked(slot);
5246 pci_slot_unlock(slot);
5253 static int pci_bus_reset(struct pci_bus *bus, int probe)
5257 if (!bus->self || !pci_bus_resetable(bus))
5267 ret = pci_bridge_secondary_bus_reset(bus->self);
5269 pci_bus_unlock(bus);
5275 * pci_bus_error_reset - reset the bridge's subordinate bus
5276 * @bridge: The parent device that connects to the bus to reset
5278 * This function will first try to reset the slots on this bus if the method is
5279 * available. If slot reset fails or is not available, this will fall back to a
5280 * secondary bus reset.
5282 int pci_bus_error_reset(struct pci_dev *bridge)
5284 struct pci_bus *bus = bridge->subordinate;
5285 struct pci_slot *slot;
5290 mutex_lock(&pci_slot_mutex);
5291 if (list_empty(&bus->slots))
5294 list_for_each_entry(slot, &bus->slots, list)
5295 if (pci_probe_reset_slot(slot))
5298 list_for_each_entry(slot, &bus->slots, list)
5299 if (pci_slot_reset(slot, 0))
5302 mutex_unlock(&pci_slot_mutex);
5305 mutex_unlock(&pci_slot_mutex);
5306 return pci_bus_reset(bridge->subordinate, 0);
5310 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5311 * @bus: PCI bus to probe
5313 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5315 int pci_probe_reset_bus(struct pci_bus *bus)
5317 return pci_bus_reset(bus, 1);
5319 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5322 * __pci_reset_bus - Try to reset a PCI bus
5323 * @bus: top level PCI bus to reset
5325 * Same as above except return -EAGAIN if the bus cannot be locked
5327 static int __pci_reset_bus(struct pci_bus *bus)
5331 rc = pci_bus_reset(bus, 1);
5335 if (pci_bus_trylock(bus)) {
5336 pci_bus_save_and_disable_locked(bus);
5338 rc = pci_bridge_secondary_bus_reset(bus->self);
5339 pci_bus_restore_locked(bus);
5340 pci_bus_unlock(bus);
5348 * pci_reset_bus - Try to reset a PCI bus
5349 * @pdev: top level PCI device to reset via slot/bus
5351 * Same as above except return -EAGAIN if the bus cannot be locked
5353 int pci_reset_bus(struct pci_dev *pdev)
5355 return (!pci_probe_reset_slot(pdev->slot)) ?
5356 __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5358 EXPORT_SYMBOL_GPL(pci_reset_bus);
5361 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5362 * @dev: PCI device to query
5364 * Returns mmrbc: maximum designed memory read count in bytes or
5365 * appropriate error value.
5367 int pcix_get_max_mmrbc(struct pci_dev *dev)
5372 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5376 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5379 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5381 EXPORT_SYMBOL(pcix_get_max_mmrbc);
5384 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5385 * @dev: PCI device to query
5387 * Returns mmrbc: maximum memory read count in bytes or appropriate error
5390 int pcix_get_mmrbc(struct pci_dev *dev)
5395 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5399 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5402 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5404 EXPORT_SYMBOL(pcix_get_mmrbc);
5407 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5408 * @dev: PCI device to query
5409 * @mmrbc: maximum memory read count in bytes
5410 * valid values are 512, 1024, 2048, 4096
5412 * If possible sets maximum memory read byte count, some bridges have errata
5413 * that prevent this.
5415 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5421 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5424 v = ffs(mmrbc) - 10;
5426 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5430 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5433 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5436 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5439 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5441 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5444 cmd &= ~PCI_X_CMD_MAX_READ;
5446 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5451 EXPORT_SYMBOL(pcix_set_mmrbc);
5454 * pcie_get_readrq - get PCI Express read request size
5455 * @dev: PCI device to query
5457 * Returns maximum memory read request in bytes or appropriate error value.
5459 int pcie_get_readrq(struct pci_dev *dev)
5463 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5465 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5467 EXPORT_SYMBOL(pcie_get_readrq);
5470 * pcie_set_readrq - set PCI Express maximum memory read request
5471 * @dev: PCI device to query
5472 * @rq: maximum memory read count in bytes
5473 * valid values are 128, 256, 512, 1024, 2048, 4096
5475 * If possible sets maximum memory read request in bytes
5477 int pcie_set_readrq(struct pci_dev *dev, int rq)
5481 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5485 * If using the "performance" PCIe config, we clamp the read rq
5486 * size to the max packet size to keep the host bridge from
5487 * generating requests larger than we can cope with.
5489 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5490 int mps = pcie_get_mps(dev);
5496 v = (ffs(rq) - 8) << 12;
5498 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5499 PCI_EXP_DEVCTL_READRQ, v);
5501 EXPORT_SYMBOL(pcie_set_readrq);
5504 * pcie_get_mps - get PCI Express maximum payload size
5505 * @dev: PCI device to query
5507 * Returns maximum payload size in bytes
5509 int pcie_get_mps(struct pci_dev *dev)
5513 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5515 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5517 EXPORT_SYMBOL(pcie_get_mps);
5520 * pcie_set_mps - set PCI Express maximum payload size
5521 * @dev: PCI device to query
5522 * @mps: maximum payload size in bytes
5523 * valid values are 128, 256, 512, 1024, 2048, 4096
5525 * If possible sets maximum payload size
5527 int pcie_set_mps(struct pci_dev *dev, int mps)
5531 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5535 if (v > dev->pcie_mpss)
5539 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5540 PCI_EXP_DEVCTL_PAYLOAD, v);
5542 EXPORT_SYMBOL(pcie_set_mps);
5545 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5546 * device and its bandwidth limitation
5547 * @dev: PCI device to query
5548 * @limiting_dev: storage for device causing the bandwidth limitation
5549 * @speed: storage for speed of limiting device
5550 * @width: storage for width of limiting device
5552 * Walk up the PCI device chain and find the point where the minimum
5553 * bandwidth is available. Return the bandwidth available there and (if
5554 * limiting_dev, speed, and width pointers are supplied) information about
5555 * that point. The bandwidth returned is in Mb/s, i.e., megabits/second of
5558 u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5559 enum pci_bus_speed *speed,
5560 enum pcie_link_width *width)
5563 enum pci_bus_speed next_speed;
5564 enum pcie_link_width next_width;
5568 *speed = PCI_SPEED_UNKNOWN;
5570 *width = PCIE_LNK_WIDTH_UNKNOWN;
5575 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5577 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5578 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5579 PCI_EXP_LNKSTA_NLW_SHIFT;
5581 next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5583 /* Check if current device limits the total bandwidth */
5584 if (!bw || next_bw <= bw) {
5588 *limiting_dev = dev;
5590 *speed = next_speed;
5592 *width = next_width;
5595 dev = pci_upstream_bridge(dev);
5600 EXPORT_SYMBOL(pcie_bandwidth_available);
5603 * pcie_get_speed_cap - query for the PCI device's link speed capability
5604 * @dev: PCI device to query
5606 * Query the PCI device speed capability. Return the maximum link speed
5607 * supported by the device.
5609 enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5611 u32 lnkcap2, lnkcap;
5614 * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18. The
5615 * implementation note there recommends using the Supported Link
5616 * Speeds Vector in Link Capabilities 2 when supported.
5618 * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
5619 * should use the Supported Link Speeds field in Link Capabilities,
5620 * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
5622 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5623 if (lnkcap2) { /* PCIe r3.0-compliant */
5624 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_16_0GB)
5625 return PCIE_SPEED_16_0GT;
5626 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
5627 return PCIE_SPEED_8_0GT;
5628 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
5629 return PCIE_SPEED_5_0GT;
5630 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
5631 return PCIE_SPEED_2_5GT;
5632 return PCI_SPEED_UNKNOWN;
5635 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5636 if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
5637 return PCIE_SPEED_5_0GT;
5638 else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
5639 return PCIE_SPEED_2_5GT;
5641 return PCI_SPEED_UNKNOWN;
5643 EXPORT_SYMBOL(pcie_get_speed_cap);
5646 * pcie_get_width_cap - query for the PCI device's link width capability
5647 * @dev: PCI device to query
5649 * Query the PCI device width capability. Return the maximum link width
5650 * supported by the device.
5652 enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5656 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5658 return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5660 return PCIE_LNK_WIDTH_UNKNOWN;
5662 EXPORT_SYMBOL(pcie_get_width_cap);
5665 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5667 * @speed: storage for link speed
5668 * @width: storage for link width
5670 * Calculate a PCI device's link bandwidth by querying for its link speed
5671 * and width, multiplying them, and applying encoding overhead. The result
5672 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5674 u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5675 enum pcie_link_width *width)
5677 *speed = pcie_get_speed_cap(dev);
5678 *width = pcie_get_width_cap(dev);
5680 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5683 return *width * PCIE_SPEED2MBS_ENC(*speed);
5687 * __pcie_print_link_status - Report the PCI device's link speed and width
5688 * @dev: PCI device to query
5689 * @verbose: Print info even when enough bandwidth is available
5691 * If the available bandwidth at the device is less than the device is
5692 * capable of, report the device's maximum possible bandwidth and the
5693 * upstream link that limits its performance. If @verbose, always print
5694 * the available bandwidth, even if the device isn't constrained.
5696 void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
5698 enum pcie_link_width width, width_cap;
5699 enum pci_bus_speed speed, speed_cap;
5700 struct pci_dev *limiting_dev = NULL;
5701 u32 bw_avail, bw_cap;
5703 bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5704 bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5706 if (bw_avail >= bw_cap && verbose)
5707 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5708 bw_cap / 1000, bw_cap % 1000,
5709 PCIE_SPEED2STR(speed_cap), width_cap);
5710 else if (bw_avail < bw_cap)
5711 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",
5712 bw_avail / 1000, bw_avail % 1000,
5713 PCIE_SPEED2STR(speed), width,
5714 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5715 bw_cap / 1000, bw_cap % 1000,
5716 PCIE_SPEED2STR(speed_cap), width_cap);
5720 * pcie_print_link_status - Report the PCI device's link speed and width
5721 * @dev: PCI device to query
5723 * Report the available bandwidth at the device.
5725 void pcie_print_link_status(struct pci_dev *dev)
5727 __pcie_print_link_status(dev, true);
5729 EXPORT_SYMBOL(pcie_print_link_status);
5732 * pci_select_bars - Make BAR mask from the type of resource
5733 * @dev: the PCI device for which BAR mask is made
5734 * @flags: resource type mask to be selected
5736 * This helper routine makes bar mask from the type of resource.
5738 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5741 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5742 if (pci_resource_flags(dev, i) & flags)
5746 EXPORT_SYMBOL(pci_select_bars);
5748 /* Some architectures require additional programming to enable VGA */
5749 static arch_set_vga_state_t arch_set_vga_state;
5751 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5753 arch_set_vga_state = func; /* NULL disables */
5756 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5757 unsigned int command_bits, u32 flags)
5759 if (arch_set_vga_state)
5760 return arch_set_vga_state(dev, decode, command_bits,
5766 * pci_set_vga_state - set VGA decode state on device and parents if requested
5767 * @dev: the PCI device
5768 * @decode: true = enable decoding, false = disable decoding
5769 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5770 * @flags: traverse ancestors and change bridges
5771 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5773 int pci_set_vga_state(struct pci_dev *dev, bool decode,
5774 unsigned int command_bits, u32 flags)
5776 struct pci_bus *bus;
5777 struct pci_dev *bridge;
5781 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5783 /* ARCH specific VGA enables */
5784 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5788 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5789 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5791 cmd |= command_bits;
5793 cmd &= ~command_bits;
5794 pci_write_config_word(dev, PCI_COMMAND, cmd);
5797 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5804 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5807 cmd |= PCI_BRIDGE_CTL_VGA;
5809 cmd &= ~PCI_BRIDGE_CTL_VGA;
5810 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5819 * pci_add_dma_alias - Add a DMA devfn alias for a device
5820 * @dev: the PCI device for which alias is added
5821 * @devfn: alias slot and function
5823 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
5824 * which is used to program permissible bus-devfn source addresses for DMA
5825 * requests in an IOMMU. These aliases factor into IOMMU group creation
5826 * and are useful for devices generating DMA requests beyond or different
5827 * from their logical bus-devfn. Examples include device quirks where the
5828 * device simply uses the wrong devfn, as well as non-transparent bridges
5829 * where the alias may be a proxy for devices in another domain.
5831 * IOMMU group creation is performed during device discovery or addition,
5832 * prior to any potential DMA mapping and therefore prior to driver probing
5833 * (especially for userspace assigned devices where IOMMU group definition
5834 * cannot be left as a userspace activity). DMA aliases should therefore
5835 * be configured via quirks, such as the PCI fixup header quirk.
5837 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5839 if (!dev->dma_alias_mask)
5840 dev->dma_alias_mask = bitmap_zalloc(U8_MAX, GFP_KERNEL);
5841 if (!dev->dma_alias_mask) {
5842 pci_warn(dev, "Unable to allocate DMA alias mask\n");
5846 set_bit(devfn, dev->dma_alias_mask);
5847 pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
5848 PCI_SLOT(devfn), PCI_FUNC(devfn));
5851 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5853 return (dev1->dma_alias_mask &&
5854 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5855 (dev2->dma_alias_mask &&
5856 test_bit(dev1->devfn, dev2->dma_alias_mask));
5859 bool pci_device_is_present(struct pci_dev *pdev)
5863 if (pci_dev_is_disconnected(pdev))
5865 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5867 EXPORT_SYMBOL_GPL(pci_device_is_present);
5869 void pci_ignore_hotplug(struct pci_dev *dev)
5871 struct pci_dev *bridge = dev->bus->self;
5873 dev->ignore_hotplug = 1;
5874 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5876 bridge->ignore_hotplug = 1;
5878 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5880 resource_size_t __weak pcibios_default_alignment(void)
5885 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5886 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5887 static DEFINE_SPINLOCK(resource_alignment_lock);
5890 * pci_specified_resource_alignment - get resource alignment specified by user.
5891 * @dev: the PCI device to get
5892 * @resize: whether or not to change resources' size when reassigning alignment
5894 * RETURNS: Resource alignment if it is specified.
5895 * Zero if it is not specified.
5897 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5900 int align_order, count;
5901 resource_size_t align = pcibios_default_alignment();
5905 spin_lock(&resource_alignment_lock);
5906 p = resource_alignment_param;
5909 if (pci_has_flag(PCI_PROBE_ONLY)) {
5911 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5917 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5924 ret = pci_dev_str_match(dev, p, &p);
5927 if (align_order == -1)
5930 align = 1 << align_order;
5932 } else if (ret < 0) {
5933 pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
5938 if (*p != ';' && *p != ',') {
5939 /* End of param or invalid format */
5945 spin_unlock(&resource_alignment_lock);
5949 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5950 resource_size_t align, bool resize)
5952 struct resource *r = &dev->resource[bar];
5953 resource_size_t size;
5955 if (!(r->flags & IORESOURCE_MEM))
5958 if (r->flags & IORESOURCE_PCI_FIXED) {
5959 pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5960 bar, r, (unsigned long long)align);
5964 size = resource_size(r);
5969 * Increase the alignment of the resource. There are two ways we
5972 * 1) Increase the size of the resource. BARs are aligned on their
5973 * size, so when we reallocate space for this resource, we'll
5974 * allocate it with the larger alignment. This also prevents
5975 * assignment of any other BARs inside the alignment region, so
5976 * if we're requesting page alignment, this means no other BARs
5977 * will share the page.
5979 * The disadvantage is that this makes the resource larger than
5980 * the hardware BAR, which may break drivers that compute things
5981 * based on the resource size, e.g., to find registers at a
5982 * fixed offset before the end of the BAR.
5984 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5985 * set r->start to the desired alignment. By itself this
5986 * doesn't prevent other BARs being put inside the alignment
5987 * region, but if we realign *every* resource of every device in
5988 * the system, none of them will share an alignment region.
5990 * When the user has requested alignment for only some devices via
5991 * the "pci=resource_alignment" argument, "resize" is true and we
5992 * use the first method. Otherwise we assume we're aligning all
5993 * devices and we use the second.
5996 pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
5997 bar, r, (unsigned long long)align);
6003 r->flags &= ~IORESOURCE_SIZEALIGN;
6004 r->flags |= IORESOURCE_STARTALIGN;
6006 r->end = r->start + size - 1;
6008 r->flags |= IORESOURCE_UNSET;
6012 * This function disables memory decoding and releases memory resources
6013 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6014 * It also rounds up size to specified alignment.
6015 * Later on, the kernel will assign page-aligned memory resource back
6018 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6022 resource_size_t align;
6024 bool resize = false;
6027 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6028 * 3.4.1.11. Their resources are allocated from the space
6029 * described by the VF BARx register in the PF's SR-IOV capability.
6030 * We can't influence their alignment here.
6035 /* check if specified PCI is target device to reassign */
6036 align = pci_specified_resource_alignment(dev, &resize);
6040 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6041 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6042 pci_warn(dev, "Can't reassign resources to host bridge\n");
6046 pci_read_config_word(dev, PCI_COMMAND, &command);
6047 command &= ~PCI_COMMAND_MEMORY;
6048 pci_write_config_word(dev, PCI_COMMAND, command);
6050 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6051 pci_request_resource_alignment(dev, i, align, resize);
6054 * Need to disable bridge's resource window,
6055 * to enable the kernel to reassign new resource
6058 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6059 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6060 r = &dev->resource[i];
6061 if (!(r->flags & IORESOURCE_MEM))
6063 r->flags |= IORESOURCE_UNSET;
6064 r->end = resource_size(r) - 1;
6067 pci_disable_bridge_window(dev);
6071 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
6073 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
6074 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
6075 spin_lock(&resource_alignment_lock);
6076 strncpy(resource_alignment_param, buf, count);
6077 resource_alignment_param[count] = '\0';
6078 spin_unlock(&resource_alignment_lock);
6082 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
6085 spin_lock(&resource_alignment_lock);
6086 count = snprintf(buf, size, "%s", resource_alignment_param);
6087 spin_unlock(&resource_alignment_lock);
6091 static ssize_t resource_alignment_show(struct bus_type *bus, char *buf)
6093 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
6096 static ssize_t resource_alignment_store(struct bus_type *bus,
6097 const char *buf, size_t count)
6099 return pci_set_resource_alignment_param(buf, count);
6102 static BUS_ATTR_RW(resource_alignment);
6104 static int __init pci_resource_alignment_sysfs_init(void)
6106 return bus_create_file(&pci_bus_type,
6107 &bus_attr_resource_alignment);
6109 late_initcall(pci_resource_alignment_sysfs_init);
6111 static void pci_no_domains(void)
6113 #ifdef CONFIG_PCI_DOMAINS
6114 pci_domains_supported = 0;
6118 #ifdef CONFIG_PCI_DOMAINS_GENERIC
6119 static atomic_t __domain_nr = ATOMIC_INIT(-1);
6121 static int pci_get_new_domain_nr(void)
6123 return atomic_inc_return(&__domain_nr);
6126 static int of_pci_bus_find_domain_nr(struct device *parent)
6128 static int use_dt_domains = -1;
6132 domain = of_get_pci_domain_nr(parent->of_node);
6135 * Check DT domain and use_dt_domains values.
6137 * If DT domain property is valid (domain >= 0) and
6138 * use_dt_domains != 0, the DT assignment is valid since this means
6139 * we have not previously allocated a domain number by using
6140 * pci_get_new_domain_nr(); we should also update use_dt_domains to
6141 * 1, to indicate that we have just assigned a domain number from
6144 * If DT domain property value is not valid (ie domain < 0), and we
6145 * have not previously assigned a domain number from DT
6146 * (use_dt_domains != 1) we should assign a domain number by
6149 * pci_get_new_domain_nr()
6151 * API and update the use_dt_domains value to keep track of method we
6152 * are using to assign domain numbers (use_dt_domains = 0).
6154 * All other combinations imply we have a platform that is trying
6155 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6156 * which is a recipe for domain mishandling and it is prevented by
6157 * invalidating the domain value (domain = -1) and printing a
6158 * corresponding error.
6160 if (domain >= 0 && use_dt_domains) {
6162 } else if (domain < 0 && use_dt_domains != 1) {
6164 domain = pci_get_new_domain_nr();
6167 pr_err("Node %pOF has ", parent->of_node);
6168 pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6175 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6177 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6178 acpi_pci_bus_find_domain_nr(bus);
6183 * pci_ext_cfg_avail - can we access extended PCI config space?
6185 * Returns 1 if we can access PCI extended config space (offsets
6186 * greater than 0xff). This is the default implementation. Architecture
6187 * implementations can override this.
6189 int __weak pci_ext_cfg_avail(void)
6194 void __weak pci_fixup_cardbus(struct pci_bus *bus)
6197 EXPORT_SYMBOL(pci_fixup_cardbus);
6199 static int __init pci_setup(char *str)
6202 char *k = strchr(str, ',');
6205 if (*str && (str = pcibios_setup(str)) && *str) {
6206 if (!strcmp(str, "nomsi")) {
6208 } else if (!strncmp(str, "noats", 5)) {
6209 pr_info("PCIe: ATS is disabled\n");
6210 pcie_ats_disabled = true;
6211 } else if (!strcmp(str, "noaer")) {
6213 } else if (!strcmp(str, "earlydump")) {
6214 pci_early_dump = true;
6215 } else if (!strncmp(str, "realloc=", 8)) {
6216 pci_realloc_get_opt(str + 8);
6217 } else if (!strncmp(str, "realloc", 7)) {
6218 pci_realloc_get_opt("on");
6219 } else if (!strcmp(str, "nodomains")) {
6221 } else if (!strncmp(str, "noari", 5)) {
6222 pcie_ari_disabled = true;
6223 } else if (!strncmp(str, "cbiosize=", 9)) {
6224 pci_cardbus_io_size = memparse(str + 9, &str);
6225 } else if (!strncmp(str, "cbmemsize=", 10)) {
6226 pci_cardbus_mem_size = memparse(str + 10, &str);
6227 } else if (!strncmp(str, "resource_alignment=", 19)) {
6228 pci_set_resource_alignment_param(str + 19,
6230 } else if (!strncmp(str, "ecrc=", 5)) {
6231 pcie_ecrc_get_policy(str + 5);
6232 } else if (!strncmp(str, "hpiosize=", 9)) {
6233 pci_hotplug_io_size = memparse(str + 9, &str);
6234 } else if (!strncmp(str, "hpmemsize=", 10)) {
6235 pci_hotplug_mem_size = memparse(str + 10, &str);
6236 } else if (!strncmp(str, "hpbussize=", 10)) {
6237 pci_hotplug_bus_size =
6238 simple_strtoul(str + 10, &str, 0);
6239 if (pci_hotplug_bus_size > 0xff)
6240 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6241 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6242 pcie_bus_config = PCIE_BUS_TUNE_OFF;
6243 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6244 pcie_bus_config = PCIE_BUS_SAFE;
6245 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6246 pcie_bus_config = PCIE_BUS_PERFORMANCE;
6247 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6248 pcie_bus_config = PCIE_BUS_PEER2PEER;
6249 } else if (!strncmp(str, "pcie_scan_all", 13)) {
6250 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6251 } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6252 disable_acs_redir_param = str + 18;
6254 pr_err("PCI: Unknown option `%s'\n", str);
6261 early_param("pci", pci_setup);
6264 * 'disable_acs_redir_param' is initialized in pci_setup(), above, to point
6265 * to data in the __initdata section which will be freed after the init
6266 * sequence is complete. We can't allocate memory in pci_setup() because some
6267 * architectures do not have any memory allocation service available during
6268 * an early_param() call. So we allocate memory and copy the variable here
6269 * before the init section is freed.
6271 static int __init pci_realloc_setup_params(void)
6273 disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6277 pure_initcall(pci_realloc_setup_params);