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,
8 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
11 #include <linux/acpi.h>
12 #include <linux/kernel.h>
13 #include <linux/delay.h>
14 #include <linux/dmi.h>
15 #include <linux/init.h>
16 #include <linux/msi.h>
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/device.h>
28 #include <linux/pm_runtime.h>
29 #include <linux/pci_hotplug.h>
30 #include <linux/vmalloc.h>
32 #include <linux/aer.h>
33 #include <linux/bitfield.h>
36 DEFINE_MUTEX(pci_slot_mutex);
38 const char *pci_power_names[] = {
39 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
41 EXPORT_SYMBOL_GPL(pci_power_names);
44 int isa_dma_bridge_buggy;
45 EXPORT_SYMBOL(isa_dma_bridge_buggy);
49 EXPORT_SYMBOL(pci_pci_problems);
51 unsigned int pci_pm_d3hot_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 */
67 * Following exit from Conventional Reset, devices must be ready within 1 sec
68 * (PCIe r6.0 sec 6.6.1). A D3cold to D0 transition implies a Conventional
69 * Reset (PCIe r6.0 sec 5.8).
71 #define PCI_RESET_WAIT 1000 /* msec */
74 * Devices may extend the 1 sec period through Request Retry Status
75 * completions (PCIe r6.0 sec 2.3.1). The spec does not provide an upper
76 * limit, but 60 sec ought to be enough for any device to become
79 #define PCIE_RESET_READY_POLL_MS 60000 /* msec */
81 static void pci_dev_d3_sleep(struct pci_dev *dev)
83 unsigned int delay_ms = max(dev->d3hot_delay, pci_pm_d3hot_delay);
87 /* Use a 20% upper bound, 1ms minimum */
88 upper = max(DIV_ROUND_CLOSEST(delay_ms, 5), 1U);
89 usleep_range(delay_ms * USEC_PER_MSEC,
90 (delay_ms + upper) * USEC_PER_MSEC);
94 bool pci_reset_supported(struct pci_dev *dev)
96 return dev->reset_methods[0] != 0;
99 #ifdef CONFIG_PCI_DOMAINS
100 int pci_domains_supported = 1;
103 #define DEFAULT_CARDBUS_IO_SIZE (256)
104 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
105 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
106 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
107 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
109 #define DEFAULT_HOTPLUG_IO_SIZE (256)
110 #define DEFAULT_HOTPLUG_MMIO_SIZE (2*1024*1024)
111 #define DEFAULT_HOTPLUG_MMIO_PREF_SIZE (2*1024*1024)
112 /* hpiosize=nn can override this */
113 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
115 * pci=hpmmiosize=nnM overrides non-prefetchable MMIO size,
116 * pci=hpmmioprefsize=nnM overrides prefetchable MMIO size;
117 * pci=hpmemsize=nnM overrides both
119 unsigned long pci_hotplug_mmio_size = DEFAULT_HOTPLUG_MMIO_SIZE;
120 unsigned long pci_hotplug_mmio_pref_size = DEFAULT_HOTPLUG_MMIO_PREF_SIZE;
122 #define DEFAULT_HOTPLUG_BUS_SIZE 1
123 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
126 /* PCIe MPS/MRRS strategy; can be overridden by kernel command-line param */
127 #ifdef CONFIG_PCIE_BUS_TUNE_OFF
128 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_TUNE_OFF;
129 #elif defined CONFIG_PCIE_BUS_SAFE
130 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_SAFE;
131 #elif defined CONFIG_PCIE_BUS_PERFORMANCE
132 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_PERFORMANCE;
133 #elif defined CONFIG_PCIE_BUS_PEER2PEER
134 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_PEER2PEER;
136 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
140 * The default CLS is used if arch didn't set CLS explicitly and not
141 * all pci devices agree on the same value. Arch can override either
142 * the dfl or actual value as it sees fit. Don't forget this is
143 * measured in 32-bit words, not bytes.
145 u8 pci_dfl_cache_line_size __ro_after_init = L1_CACHE_BYTES >> 2;
146 u8 pci_cache_line_size __ro_after_init ;
149 * If we set up a device for bus mastering, we need to check the latency
150 * timer as certain BIOSes forget to set it properly.
152 unsigned int pcibios_max_latency = 255;
154 /* If set, the PCIe ARI capability will not be used. */
155 static bool pcie_ari_disabled;
157 /* If set, the PCIe ATS capability will not be used. */
158 static bool pcie_ats_disabled;
160 /* If set, the PCI config space of each device is printed during boot. */
163 bool pci_ats_disabled(void)
165 return pcie_ats_disabled;
167 EXPORT_SYMBOL_GPL(pci_ats_disabled);
169 /* Disable bridge_d3 for all PCIe ports */
170 static bool pci_bridge_d3_disable;
171 /* Force bridge_d3 for all PCIe ports */
172 static bool pci_bridge_d3_force;
174 static int __init pcie_port_pm_setup(char *str)
176 if (!strcmp(str, "off"))
177 pci_bridge_d3_disable = true;
178 else if (!strcmp(str, "force"))
179 pci_bridge_d3_force = true;
182 __setup("pcie_port_pm=", pcie_port_pm_setup);
185 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
186 * @bus: pointer to PCI bus structure to search
188 * Given a PCI bus, returns the highest PCI bus number present in the set
189 * including the given PCI bus and its list of child PCI buses.
191 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
194 unsigned char max, n;
196 max = bus->busn_res.end;
197 list_for_each_entry(tmp, &bus->children, node) {
198 n = pci_bus_max_busnr(tmp);
204 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
207 * pci_status_get_and_clear_errors - return and clear error bits in PCI_STATUS
208 * @pdev: the PCI device
210 * Returns error bits set in PCI_STATUS and clears them.
212 int pci_status_get_and_clear_errors(struct pci_dev *pdev)
217 ret = pci_read_config_word(pdev, PCI_STATUS, &status);
218 if (ret != PCIBIOS_SUCCESSFUL)
221 status &= PCI_STATUS_ERROR_BITS;
223 pci_write_config_word(pdev, PCI_STATUS, status);
227 EXPORT_SYMBOL_GPL(pci_status_get_and_clear_errors);
229 #ifdef CONFIG_HAS_IOMEM
230 static void __iomem *__pci_ioremap_resource(struct pci_dev *pdev, int bar,
233 struct resource *res = &pdev->resource[bar];
234 resource_size_t start = res->start;
235 resource_size_t size = resource_size(res);
238 * Make sure the BAR is actually a memory resource, not an IO resource
240 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
241 pci_err(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
246 return ioremap_wc(start, size);
248 return ioremap(start, size);
251 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
253 return __pci_ioremap_resource(pdev, bar, false);
255 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
257 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
259 return __pci_ioremap_resource(pdev, bar, true);
261 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
265 * pci_dev_str_match_path - test if a path string matches a device
266 * @dev: the PCI device to test
267 * @path: string to match the device against
268 * @endptr: pointer to the string after the match
270 * Test if a string (typically from a kernel parameter) formatted as a
271 * path of device/function addresses matches a PCI device. The string must
274 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
276 * A path for a device can be obtained using 'lspci -t'. Using a path
277 * is more robust against bus renumbering than using only a single bus,
278 * device and function address.
280 * Returns 1 if the string matches the device, 0 if it does not and
281 * a negative error code if it fails to parse the string.
283 static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
287 unsigned int seg, bus, slot, func;
291 *endptr = strchrnul(path, ';');
293 wpath = kmemdup_nul(path, *endptr - path, GFP_ATOMIC);
298 p = strrchr(wpath, '/');
301 ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
307 if (dev->devfn != PCI_DEVFN(slot, func)) {
313 * Note: we don't need to get a reference to the upstream
314 * bridge because we hold a reference to the top level
315 * device which should hold a reference to the bridge,
318 dev = pci_upstream_bridge(dev);
327 ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
331 ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
338 ret = (seg == pci_domain_nr(dev->bus) &&
339 bus == dev->bus->number &&
340 dev->devfn == PCI_DEVFN(slot, func));
348 * pci_dev_str_match - test if a string matches a device
349 * @dev: the PCI device to test
350 * @p: string to match the device against
351 * @endptr: pointer to the string after the match
353 * Test if a string (typically from a kernel parameter) matches a specified
354 * PCI device. The string may be of one of the following formats:
356 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
357 * pci:<vendor>:<device>[:<subvendor>:<subdevice>]
359 * The first format specifies a PCI bus/device/function address which
360 * may change if new hardware is inserted, if motherboard firmware changes,
361 * or due to changes caused in kernel parameters. If the domain is
362 * left unspecified, it is taken to be 0. In order to be robust against
363 * bus renumbering issues, a path of PCI device/function numbers may be used
364 * to address the specific device. The path for a device can be determined
365 * through the use of 'lspci -t'.
367 * The second format matches devices using IDs in the configuration
368 * space which may match multiple devices in the system. A value of 0
369 * for any field will match all devices. (Note: this differs from
370 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
371 * legacy reasons and convenience so users don't have to specify
372 * FFFFFFFFs on the command line.)
374 * Returns 1 if the string matches the device, 0 if it does not and
375 * a negative error code if the string cannot be parsed.
377 static int pci_dev_str_match(struct pci_dev *dev, const char *p,
382 unsigned short vendor, device, subsystem_vendor, subsystem_device;
384 if (strncmp(p, "pci:", 4) == 0) {
385 /* PCI vendor/device (subvendor/subdevice) IDs are specified */
387 ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
388 &subsystem_vendor, &subsystem_device, &count);
390 ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
394 subsystem_vendor = 0;
395 subsystem_device = 0;
400 if ((!vendor || vendor == dev->vendor) &&
401 (!device || device == dev->device) &&
402 (!subsystem_vendor ||
403 subsystem_vendor == dev->subsystem_vendor) &&
404 (!subsystem_device ||
405 subsystem_device == dev->subsystem_device))
409 * PCI Bus, Device, Function IDs are specified
410 * (optionally, may include a path of devfns following it)
412 ret = pci_dev_str_match_path(dev, p, &p);
427 static u8 __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
428 u8 pos, int cap, int *ttl)
433 pci_bus_read_config_byte(bus, devfn, pos, &pos);
439 pci_bus_read_config_word(bus, devfn, pos, &ent);
451 static u8 __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
454 int ttl = PCI_FIND_CAP_TTL;
456 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
459 u8 pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
461 return __pci_find_next_cap(dev->bus, dev->devfn,
462 pos + PCI_CAP_LIST_NEXT, cap);
464 EXPORT_SYMBOL_GPL(pci_find_next_capability);
466 static u8 __pci_bus_find_cap_start(struct pci_bus *bus,
467 unsigned int devfn, u8 hdr_type)
471 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
472 if (!(status & PCI_STATUS_CAP_LIST))
476 case PCI_HEADER_TYPE_NORMAL:
477 case PCI_HEADER_TYPE_BRIDGE:
478 return PCI_CAPABILITY_LIST;
479 case PCI_HEADER_TYPE_CARDBUS:
480 return PCI_CB_CAPABILITY_LIST;
487 * pci_find_capability - query for devices' capabilities
488 * @dev: PCI device to query
489 * @cap: capability code
491 * Tell if a device supports a given PCI capability.
492 * Returns the address of the requested capability structure within the
493 * device's PCI configuration space or 0 in case the device does not
494 * support it. Possible values for @cap include:
496 * %PCI_CAP_ID_PM Power Management
497 * %PCI_CAP_ID_AGP Accelerated Graphics Port
498 * %PCI_CAP_ID_VPD Vital Product Data
499 * %PCI_CAP_ID_SLOTID Slot Identification
500 * %PCI_CAP_ID_MSI Message Signalled Interrupts
501 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
502 * %PCI_CAP_ID_PCIX PCI-X
503 * %PCI_CAP_ID_EXP PCI Express
505 u8 pci_find_capability(struct pci_dev *dev, int cap)
509 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
511 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
515 EXPORT_SYMBOL(pci_find_capability);
518 * pci_bus_find_capability - query for devices' capabilities
519 * @bus: the PCI bus to query
520 * @devfn: PCI device to query
521 * @cap: capability code
523 * Like pci_find_capability() but works for PCI devices that do not have a
524 * pci_dev structure set up yet.
526 * Returns the address of the requested capability structure within the
527 * device's PCI configuration space or 0 in case the device does not
530 u8 pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
534 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
536 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & PCI_HEADER_TYPE_MASK);
538 pos = __pci_find_next_cap(bus, devfn, pos, cap);
542 EXPORT_SYMBOL(pci_bus_find_capability);
545 * pci_find_next_ext_capability - Find an extended capability
546 * @dev: PCI device to query
547 * @start: address at which to start looking (0 to start at beginning of list)
548 * @cap: capability code
550 * Returns the address of the next matching extended capability structure
551 * within the device's PCI configuration space or 0 if the device does
552 * not support it. Some capabilities can occur several times, e.g., the
553 * vendor-specific capability, and this provides a way to find them all.
555 u16 pci_find_next_ext_capability(struct pci_dev *dev, u16 start, int cap)
559 u16 pos = PCI_CFG_SPACE_SIZE;
561 /* minimum 8 bytes per capability */
562 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
564 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
570 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
574 * If we have no capabilities, this is indicated by cap ID,
575 * cap version and next pointer all being 0.
581 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
584 pos = PCI_EXT_CAP_NEXT(header);
585 if (pos < PCI_CFG_SPACE_SIZE)
588 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
594 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
597 * pci_find_ext_capability - Find an extended capability
598 * @dev: PCI device to query
599 * @cap: capability code
601 * Returns the address of the requested extended capability structure
602 * within the device's PCI configuration space or 0 if the device does
603 * not support it. Possible values for @cap include:
605 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
606 * %PCI_EXT_CAP_ID_VC Virtual Channel
607 * %PCI_EXT_CAP_ID_DSN Device Serial Number
608 * %PCI_EXT_CAP_ID_PWR Power Budgeting
610 u16 pci_find_ext_capability(struct pci_dev *dev, int cap)
612 return pci_find_next_ext_capability(dev, 0, cap);
614 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
617 * pci_get_dsn - Read and return the 8-byte Device Serial Number
618 * @dev: PCI device to query
620 * Looks up the PCI_EXT_CAP_ID_DSN and reads the 8 bytes of the Device Serial
623 * Returns the DSN, or zero if the capability does not exist.
625 u64 pci_get_dsn(struct pci_dev *dev)
631 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DSN);
636 * The Device Serial Number is two dwords offset 4 bytes from the
637 * capability position. The specification says that the first dword is
638 * the lower half, and the second dword is the upper half.
641 pci_read_config_dword(dev, pos, &dword);
643 pci_read_config_dword(dev, pos + 4, &dword);
644 dsn |= ((u64)dword) << 32;
648 EXPORT_SYMBOL_GPL(pci_get_dsn);
650 static u8 __pci_find_next_ht_cap(struct pci_dev *dev, u8 pos, int ht_cap)
652 int rc, ttl = PCI_FIND_CAP_TTL;
655 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
656 mask = HT_3BIT_CAP_MASK;
658 mask = HT_5BIT_CAP_MASK;
660 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
661 PCI_CAP_ID_HT, &ttl);
663 rc = pci_read_config_byte(dev, pos + 3, &cap);
664 if (rc != PCIBIOS_SUCCESSFUL)
667 if ((cap & mask) == ht_cap)
670 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
671 pos + PCI_CAP_LIST_NEXT,
672 PCI_CAP_ID_HT, &ttl);
679 * pci_find_next_ht_capability - query a device's HyperTransport capabilities
680 * @dev: PCI device to query
681 * @pos: Position from which to continue searching
682 * @ht_cap: HyperTransport capability code
684 * To be used in conjunction with pci_find_ht_capability() to search for
685 * all capabilities matching @ht_cap. @pos should always be a value returned
686 * from pci_find_ht_capability().
688 * NB. To be 100% safe against broken PCI devices, the caller should take
689 * steps to avoid an infinite loop.
691 u8 pci_find_next_ht_capability(struct pci_dev *dev, u8 pos, int ht_cap)
693 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
695 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
698 * pci_find_ht_capability - query a device's HyperTransport capabilities
699 * @dev: PCI device to query
700 * @ht_cap: HyperTransport capability code
702 * Tell if a device supports a given HyperTransport capability.
703 * Returns an address within the device's PCI configuration space
704 * or 0 in case the device does not support the request capability.
705 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
706 * which has a HyperTransport capability matching @ht_cap.
708 u8 pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
712 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
714 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
718 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
721 * pci_find_vsec_capability - Find a vendor-specific extended capability
722 * @dev: PCI device to query
723 * @vendor: Vendor ID for which capability is defined
724 * @cap: Vendor-specific capability ID
726 * If @dev has Vendor ID @vendor, search for a VSEC capability with
727 * VSEC ID @cap. If found, return the capability offset in
728 * config space; otherwise return 0.
730 u16 pci_find_vsec_capability(struct pci_dev *dev, u16 vendor, int cap)
736 if (vendor != dev->vendor)
739 while ((vsec = pci_find_next_ext_capability(dev, vsec,
740 PCI_EXT_CAP_ID_VNDR))) {
741 ret = pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER, &header);
742 if (ret != PCIBIOS_SUCCESSFUL)
745 if (PCI_VNDR_HEADER_ID(header) == cap)
751 EXPORT_SYMBOL_GPL(pci_find_vsec_capability);
754 * pci_find_dvsec_capability - Find DVSEC for vendor
755 * @dev: PCI device to query
756 * @vendor: Vendor ID to match for the DVSEC
757 * @dvsec: Designated Vendor-specific capability ID
759 * If DVSEC has Vendor ID @vendor and DVSEC ID @dvsec return the capability
760 * offset in config space; otherwise return 0.
762 u16 pci_find_dvsec_capability(struct pci_dev *dev, u16 vendor, u16 dvsec)
766 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_DVSEC);
773 pci_read_config_word(dev, pos + PCI_DVSEC_HEADER1, &v);
774 pci_read_config_word(dev, pos + PCI_DVSEC_HEADER2, &id);
775 if (vendor == v && dvsec == id)
778 pos = pci_find_next_ext_capability(dev, pos, PCI_EXT_CAP_ID_DVSEC);
783 EXPORT_SYMBOL_GPL(pci_find_dvsec_capability);
786 * pci_find_parent_resource - return resource region of parent bus of given
788 * @dev: PCI device structure contains resources to be searched
789 * @res: child resource record for which parent is sought
791 * For given resource region of given device, return the resource region of
792 * parent bus the given region is contained in.
794 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
795 struct resource *res)
797 const struct pci_bus *bus = dev->bus;
800 pci_bus_for_each_resource(bus, r) {
803 if (resource_contains(r, res)) {
806 * If the window is prefetchable but the BAR is
807 * not, the allocator made a mistake.
809 if (r->flags & IORESOURCE_PREFETCH &&
810 !(res->flags & IORESOURCE_PREFETCH))
814 * If we're below a transparent bridge, there may
815 * be both a positively-decoded aperture and a
816 * subtractively-decoded region that contain the BAR.
817 * We want the positively-decoded one, so this depends
818 * on pci_bus_for_each_resource() giving us those
826 EXPORT_SYMBOL(pci_find_parent_resource);
829 * pci_find_resource - Return matching PCI device resource
830 * @dev: PCI device to query
831 * @res: Resource to look for
833 * Goes over standard PCI resources (BARs) and checks if the given resource
834 * is partially or fully contained in any of them. In that case the
835 * matching resource is returned, %NULL otherwise.
837 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
841 for (i = 0; i < PCI_STD_NUM_BARS; i++) {
842 struct resource *r = &dev->resource[i];
844 if (r->start && resource_contains(r, res))
850 EXPORT_SYMBOL(pci_find_resource);
853 * pci_resource_name - Return the name of the PCI resource
854 * @dev: PCI device to query
855 * @i: index of the resource
857 * Return the standard PCI resource (BAR) name according to their index.
859 const char *pci_resource_name(struct pci_dev *dev, unsigned int i)
861 static const char * const bar_name[] = {
869 #ifdef CONFIG_PCI_IOV
877 "bridge window", /* "io" included in %pR */
878 "bridge window", /* "mem" included in %pR */
879 "bridge window", /* "mem pref" included in %pR */
881 static const char * const cardbus_name[] = {
888 #ifdef CONFIG_PCI_IOV
896 "CardBus bridge window 0", /* I/O */
897 "CardBus bridge window 1", /* I/O */
898 "CardBus bridge window 0", /* mem */
899 "CardBus bridge window 1", /* mem */
902 if (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS &&
903 i < ARRAY_SIZE(cardbus_name))
904 return cardbus_name[i];
906 if (i < ARRAY_SIZE(bar_name))
913 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
914 * @dev: the PCI device to operate on
915 * @pos: config space offset of status word
916 * @mask: mask of bit(s) to care about in status word
918 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
920 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
924 /* Wait for Transaction Pending bit clean */
925 for (i = 0; i < 4; i++) {
928 msleep((1 << (i - 1)) * 100);
930 pci_read_config_word(dev, pos, &status);
931 if (!(status & mask))
938 static int pci_acs_enable;
941 * pci_request_acs - ask for ACS to be enabled if supported
943 void pci_request_acs(void)
948 static const char *disable_acs_redir_param;
951 * pci_disable_acs_redir - disable ACS redirect capabilities
952 * @dev: the PCI device
954 * For only devices specified in the disable_acs_redir parameter.
956 static void pci_disable_acs_redir(struct pci_dev *dev)
963 if (!disable_acs_redir_param)
966 p = disable_acs_redir_param;
968 ret = pci_dev_str_match(dev, p, &p);
970 pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
971 disable_acs_redir_param);
974 } else if (ret == 1) {
979 if (*p != ';' && *p != ',') {
980 /* End of param or invalid format */
989 if (!pci_dev_specific_disable_acs_redir(dev))
994 pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
998 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
1000 /* P2P Request & Completion Redirect */
1001 ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
1003 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
1005 pci_info(dev, "disabled ACS redirect\n");
1009 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilities
1010 * @dev: the PCI device
1012 static void pci_std_enable_acs(struct pci_dev *dev)
1022 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
1023 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
1025 /* Source Validation */
1026 ctrl |= (cap & PCI_ACS_SV);
1028 /* P2P Request Redirect */
1029 ctrl |= (cap & PCI_ACS_RR);
1031 /* P2P Completion Redirect */
1032 ctrl |= (cap & PCI_ACS_CR);
1034 /* Upstream Forwarding */
1035 ctrl |= (cap & PCI_ACS_UF);
1037 /* Enable Translation Blocking for external devices and noats */
1038 if (pci_ats_disabled() || dev->external_facing || dev->untrusted)
1039 ctrl |= (cap & PCI_ACS_TB);
1041 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
1045 * pci_enable_acs - enable ACS if hardware support it
1046 * @dev: the PCI device
1048 static void pci_enable_acs(struct pci_dev *dev)
1050 if (!pci_acs_enable)
1051 goto disable_acs_redir;
1053 if (!pci_dev_specific_enable_acs(dev))
1054 goto disable_acs_redir;
1056 pci_std_enable_acs(dev);
1060 * Note: pci_disable_acs_redir() must be called even if ACS was not
1061 * enabled by the kernel because it may have been enabled by
1062 * platform firmware. So if we are told to disable it, we should
1063 * always disable it after setting the kernel's default
1066 pci_disable_acs_redir(dev);
1070 * pcie_read_tlp_log - read TLP Header Log
1072 * @where: PCI Config offset of TLP Header Log
1073 * @tlp_log: TLP Log structure to fill
1075 * Fill @tlp_log from TLP Header Log registers, e.g., AER or DPC.
1077 * Return: 0 on success and filled TLP Log structure, <0 on error.
1079 int pcie_read_tlp_log(struct pci_dev *dev, int where,
1080 struct pcie_tlp_log *tlp_log)
1084 memset(tlp_log, 0, sizeof(*tlp_log));
1086 for (i = 0; i < 4; i++) {
1087 ret = pci_read_config_dword(dev, where + i * 4,
1090 return pcibios_err_to_errno(ret);
1095 EXPORT_SYMBOL_GPL(pcie_read_tlp_log);
1098 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
1099 * @dev: PCI device to have its BARs restored
1101 * Restore the BAR values for a given device, so as to make it
1102 * accessible by its driver.
1104 static void pci_restore_bars(struct pci_dev *dev)
1108 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
1109 pci_update_resource(dev, i);
1112 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
1114 if (pci_use_mid_pm())
1117 return acpi_pci_power_manageable(dev);
1120 static inline int platform_pci_set_power_state(struct pci_dev *dev,
1123 if (pci_use_mid_pm())
1124 return mid_pci_set_power_state(dev, t);
1126 return acpi_pci_set_power_state(dev, t);
1129 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
1131 if (pci_use_mid_pm())
1132 return mid_pci_get_power_state(dev);
1134 return acpi_pci_get_power_state(dev);
1137 static inline void platform_pci_refresh_power_state(struct pci_dev *dev)
1139 if (!pci_use_mid_pm())
1140 acpi_pci_refresh_power_state(dev);
1143 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
1145 if (pci_use_mid_pm())
1146 return PCI_POWER_ERROR;
1148 return acpi_pci_choose_state(dev);
1151 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
1153 if (pci_use_mid_pm())
1154 return PCI_POWER_ERROR;
1156 return acpi_pci_wakeup(dev, enable);
1159 static inline bool platform_pci_need_resume(struct pci_dev *dev)
1161 if (pci_use_mid_pm())
1164 return acpi_pci_need_resume(dev);
1167 static inline bool platform_pci_bridge_d3(struct pci_dev *dev)
1169 if (pci_use_mid_pm())
1172 return acpi_pci_bridge_d3(dev);
1176 * pci_update_current_state - Read power state of given device and cache it
1177 * @dev: PCI device to handle.
1178 * @state: State to cache in case the device doesn't have the PM capability
1180 * The power state is read from the PMCSR register, which however is
1181 * inaccessible in D3cold. The platform firmware is therefore queried first
1182 * to detect accessibility of the register. In case the platform firmware
1183 * reports an incorrect state or the device isn't power manageable by the
1184 * platform at all, we try to detect D3cold by testing accessibility of the
1185 * vendor ID in config space.
1187 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
1189 if (platform_pci_get_power_state(dev) == PCI_D3cold) {
1190 dev->current_state = PCI_D3cold;
1191 } else if (dev->pm_cap) {
1194 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1195 if (PCI_POSSIBLE_ERROR(pmcsr)) {
1196 dev->current_state = PCI_D3cold;
1199 dev->current_state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1201 dev->current_state = state;
1206 * pci_refresh_power_state - Refresh the given device's power state data
1207 * @dev: Target PCI device.
1209 * Ask the platform to refresh the devices power state information and invoke
1210 * pci_update_current_state() to update its current PCI power state.
1212 void pci_refresh_power_state(struct pci_dev *dev)
1214 platform_pci_refresh_power_state(dev);
1215 pci_update_current_state(dev, dev->current_state);
1219 * pci_platform_power_transition - Use platform to change device power state
1220 * @dev: PCI device to handle.
1221 * @state: State to put the device into.
1223 int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
1227 error = platform_pci_set_power_state(dev, state);
1229 pci_update_current_state(dev, state);
1230 else if (!dev->pm_cap) /* Fall back to PCI_D0 */
1231 dev->current_state = PCI_D0;
1235 EXPORT_SYMBOL_GPL(pci_platform_power_transition);
1237 static int pci_resume_one(struct pci_dev *pci_dev, void *ign)
1239 pm_request_resume(&pci_dev->dev);
1244 * pci_resume_bus - Walk given bus and runtime resume devices on it
1245 * @bus: Top bus of the subtree to walk.
1247 void pci_resume_bus(struct pci_bus *bus)
1250 pci_walk_bus(bus, pci_resume_one, NULL);
1253 static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
1256 bool retrain = false;
1257 struct pci_dev *bridge;
1259 if (pci_is_pcie(dev)) {
1260 bridge = pci_upstream_bridge(dev);
1266 * After reset, the device should not silently discard config
1267 * requests, but it may still indicate that it needs more time by
1268 * responding to them with CRS completions. The Root Port will
1269 * generally synthesize ~0 (PCI_ERROR_RESPONSE) data to complete
1270 * the read (except when CRS SV is enabled and the read was for the
1271 * Vendor ID; in that case it synthesizes 0x0001 data).
1273 * Wait for the device to return a non-CRS completion. Read the
1274 * Command register instead of Vendor ID so we don't have to
1275 * contend with the CRS SV value.
1280 if (pci_dev_is_disconnected(dev)) {
1281 pci_dbg(dev, "disconnected; not waiting\n");
1285 pci_read_config_dword(dev, PCI_COMMAND, &id);
1286 if (!PCI_POSSIBLE_ERROR(id))
1289 if (delay > timeout) {
1290 pci_warn(dev, "not ready %dms after %s; giving up\n",
1291 delay - 1, reset_type);
1295 if (delay > PCI_RESET_WAIT) {
1298 if (pcie_failed_link_retrain(bridge)) {
1303 pci_info(dev, "not ready %dms after %s; waiting\n",
1304 delay - 1, reset_type);
1311 if (delay > PCI_RESET_WAIT)
1312 pci_info(dev, "ready %dms after %s\n", delay - 1,
1315 pci_dbg(dev, "ready %dms after %s\n", delay - 1,
1322 * pci_power_up - Put the given device into D0
1323 * @dev: PCI device to power up
1325 * On success, return 0 or 1, depending on whether or not it is necessary to
1326 * restore the device's BARs subsequently (1 is returned in that case).
1328 * On failure, return a negative error code. Always return failure if @dev
1329 * lacks a Power Management Capability, even if the platform was able to
1330 * put the device in D0 via non-PCI means.
1332 int pci_power_up(struct pci_dev *dev)
1338 platform_pci_set_power_state(dev, PCI_D0);
1341 state = platform_pci_get_power_state(dev);
1342 if (state == PCI_UNKNOWN)
1343 dev->current_state = PCI_D0;
1345 dev->current_state = state;
1350 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1351 if (PCI_POSSIBLE_ERROR(pmcsr)) {
1352 pci_err(dev, "Unable to change power state from %s to D0, device inaccessible\n",
1353 pci_power_name(dev->current_state));
1354 dev->current_state = PCI_D3cold;
1358 state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1360 need_restore = (state == PCI_D3hot || dev->current_state >= PCI_D3hot) &&
1361 !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET);
1363 if (state == PCI_D0)
1367 * Force the entire word to 0. This doesn't affect PME_Status, disables
1368 * PME_En, and sets PowerState to 0.
1370 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, 0);
1372 /* Mandatory transition delays; see PCI PM 1.2. */
1373 if (state == PCI_D3hot)
1374 pci_dev_d3_sleep(dev);
1375 else if (state == PCI_D2)
1376 udelay(PCI_PM_D2_DELAY);
1379 dev->current_state = PCI_D0;
1387 * pci_set_full_power_state - Put a PCI device into D0 and update its state
1388 * @dev: PCI device to power up
1389 * @locked: whether pci_bus_sem is held
1391 * Call pci_power_up() to put @dev into D0, read from its PCI_PM_CTRL register
1392 * to confirm the state change, restore its BARs if they might be lost and
1393 * reconfigure ASPM in accordance with the new power state.
1395 * If pci_restore_state() is going to be called right after a power state change
1396 * to D0, it is more efficient to use pci_power_up() directly instead of this
1399 static int pci_set_full_power_state(struct pci_dev *dev, bool locked)
1404 ret = pci_power_up(dev);
1406 if (dev->current_state == PCI_D0)
1412 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1413 dev->current_state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1414 if (dev->current_state != PCI_D0) {
1415 pci_info_ratelimited(dev, "Refused to change power state from %s to D0\n",
1416 pci_power_name(dev->current_state));
1417 } else if (ret > 0) {
1419 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
1420 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
1421 * from D3hot to D0 _may_ perform an internal reset, thereby
1422 * going to "D0 Uninitialized" rather than "D0 Initialized".
1423 * For example, at least some versions of the 3c905B and the
1424 * 3c556B exhibit this behaviour.
1426 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
1427 * devices in a D3hot state at boot. Consequently, we need to
1428 * restore at least the BARs so that the device will be
1429 * accessible to its driver.
1431 pci_restore_bars(dev);
1435 pcie_aspm_pm_state_change(dev->bus->self, locked);
1441 * __pci_dev_set_current_state - Set current state of a PCI device
1442 * @dev: Device to handle
1443 * @data: pointer to state to be set
1445 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1447 pci_power_t state = *(pci_power_t *)data;
1449 dev->current_state = state;
1454 * pci_bus_set_current_state - Walk given bus and set current state of devices
1455 * @bus: Top bus of the subtree to walk.
1456 * @state: state to be set
1458 void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1461 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1464 static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state, bool locked)
1470 pci_walk_bus_locked(bus, __pci_dev_set_current_state, &state);
1472 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1476 * pci_set_low_power_state - Put a PCI device into a low-power state.
1477 * @dev: PCI device to handle.
1478 * @state: PCI power state (D1, D2, D3hot) to put the device into.
1479 * @locked: whether pci_bus_sem is held
1481 * Use the device's PCI_PM_CTRL register to put it into a low-power state.
1484 * -EINVAL if the requested state is invalid.
1485 * -EIO if device does not support PCI PM or its PM capabilities register has a
1486 * wrong version, or device doesn't support the requested state.
1487 * 0 if device already is in the requested state.
1488 * 0 if device's power state has been successfully changed.
1490 static int pci_set_low_power_state(struct pci_dev *dev, pci_power_t state, bool locked)
1498 * Validate transition: We can enter D0 from any state, but if
1499 * we're already in a low-power state, we can only go deeper. E.g.,
1500 * we can go from D1 to D3, but we can't go directly from D3 to D1;
1501 * we'd have to go from D3 to D0, then to D1.
1503 if (dev->current_state <= PCI_D3cold && dev->current_state > state) {
1504 pci_dbg(dev, "Invalid power transition (from %s to %s)\n",
1505 pci_power_name(dev->current_state),
1506 pci_power_name(state));
1510 /* Check if this device supports the desired state */
1511 if ((state == PCI_D1 && !dev->d1_support)
1512 || (state == PCI_D2 && !dev->d2_support))
1515 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1516 if (PCI_POSSIBLE_ERROR(pmcsr)) {
1517 pci_err(dev, "Unable to change power state from %s to %s, device inaccessible\n",
1518 pci_power_name(dev->current_state),
1519 pci_power_name(state));
1520 dev->current_state = PCI_D3cold;
1524 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
1527 /* Enter specified state */
1528 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1530 /* Mandatory power management transition delays; see PCI PM 1.2. */
1531 if (state == PCI_D3hot)
1532 pci_dev_d3_sleep(dev);
1533 else if (state == PCI_D2)
1534 udelay(PCI_PM_D2_DELAY);
1536 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1537 dev->current_state = pmcsr & PCI_PM_CTRL_STATE_MASK;
1538 if (dev->current_state != state)
1539 pci_info_ratelimited(dev, "Refused to change power state from %s to %s\n",
1540 pci_power_name(dev->current_state),
1541 pci_power_name(state));
1544 pcie_aspm_pm_state_change(dev->bus->self, locked);
1549 static int __pci_set_power_state(struct pci_dev *dev, pci_power_t state, bool locked)
1553 /* Bound the state we're entering */
1554 if (state > PCI_D3cold)
1556 else if (state < PCI_D0)
1558 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1561 * If the device or the parent bridge do not support PCI
1562 * PM, ignore the request if we're doing anything other
1563 * than putting it into D0 (which would only happen on
1568 /* Check if we're already there */
1569 if (dev->current_state == state)
1572 if (state == PCI_D0)
1573 return pci_set_full_power_state(dev, locked);
1576 * This device is quirked not to be put into D3, so don't put it in
1579 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1582 if (state == PCI_D3cold) {
1584 * To put the device in D3cold, put it into D3hot in the native
1585 * way, then put it into D3cold using platform ops.
1587 error = pci_set_low_power_state(dev, PCI_D3hot, locked);
1589 if (pci_platform_power_transition(dev, PCI_D3cold))
1592 /* Powering off a bridge may power off the whole hierarchy */
1593 if (dev->current_state == PCI_D3cold)
1594 __pci_bus_set_current_state(dev->subordinate, PCI_D3cold, locked);
1596 error = pci_set_low_power_state(dev, state, locked);
1598 if (pci_platform_power_transition(dev, state))
1606 * pci_set_power_state - Set the power state of a PCI device
1607 * @dev: PCI device to handle.
1608 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1610 * Transition a device to a new power state, using the platform firmware and/or
1611 * the device's PCI PM registers.
1614 * -EINVAL if the requested state is invalid.
1615 * -EIO if device does not support PCI PM or its PM capabilities register has a
1616 * wrong version, or device doesn't support the requested state.
1617 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1618 * 0 if device already is in the requested state.
1619 * 0 if the transition is to D3 but D3 is not supported.
1620 * 0 if device's power state has been successfully changed.
1622 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1624 return __pci_set_power_state(dev, state, false);
1626 EXPORT_SYMBOL(pci_set_power_state);
1628 int pci_set_power_state_locked(struct pci_dev *dev, pci_power_t state)
1630 lockdep_assert_held(&pci_bus_sem);
1632 return __pci_set_power_state(dev, state, true);
1634 EXPORT_SYMBOL(pci_set_power_state_locked);
1636 #define PCI_EXP_SAVE_REGS 7
1638 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1639 u16 cap, bool extended)
1641 struct pci_cap_saved_state *tmp;
1643 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1644 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1650 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1652 return _pci_find_saved_cap(dev, cap, false);
1655 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1657 return _pci_find_saved_cap(dev, cap, true);
1660 static int pci_save_pcie_state(struct pci_dev *dev)
1663 struct pci_cap_saved_state *save_state;
1666 if (!pci_is_pcie(dev))
1669 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1671 pci_err(dev, "buffer not found in %s\n", __func__);
1675 cap = (u16 *)&save_state->cap.data[0];
1676 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1677 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1678 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1679 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1680 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1681 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1682 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1684 pci_save_aspm_l1ss_state(dev);
1685 pci_save_ltr_state(dev);
1690 static void pci_restore_pcie_state(struct pci_dev *dev)
1693 struct pci_cap_saved_state *save_state;
1697 * Restore max latencies (in the LTR capability) before enabling
1698 * LTR itself in PCI_EXP_DEVCTL2.
1700 pci_restore_ltr_state(dev);
1701 pci_restore_aspm_l1ss_state(dev);
1703 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1708 * Downstream ports reset the LTR enable bit when link goes down.
1709 * Check and re-configure the bit here before restoring device.
1710 * PCIe r5.0, sec 7.5.3.16.
1712 pci_bridge_reconfigure_ltr(dev);
1714 cap = (u16 *)&save_state->cap.data[0];
1715 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1716 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1717 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1718 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1719 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1720 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1721 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1724 static int pci_save_pcix_state(struct pci_dev *dev)
1727 struct pci_cap_saved_state *save_state;
1729 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1733 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1735 pci_err(dev, "buffer not found in %s\n", __func__);
1739 pci_read_config_word(dev, pos + PCI_X_CMD,
1740 (u16 *)save_state->cap.data);
1745 static void pci_restore_pcix_state(struct pci_dev *dev)
1748 struct pci_cap_saved_state *save_state;
1751 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1752 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1753 if (!save_state || !pos)
1755 cap = (u16 *)&save_state->cap.data[0];
1757 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1761 * pci_save_state - save the PCI configuration space of a device before
1763 * @dev: PCI device that we're dealing with
1765 int pci_save_state(struct pci_dev *dev)
1768 /* XXX: 100% dword access ok here? */
1769 for (i = 0; i < 16; i++) {
1770 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1771 pci_dbg(dev, "save config %#04x: %#010x\n",
1772 i * 4, dev->saved_config_space[i]);
1774 dev->state_saved = true;
1776 i = pci_save_pcie_state(dev);
1780 i = pci_save_pcix_state(dev);
1784 pci_save_dpc_state(dev);
1785 pci_save_aer_state(dev);
1786 pci_save_ptm_state(dev);
1787 return pci_save_vc_state(dev);
1789 EXPORT_SYMBOL(pci_save_state);
1791 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1792 u32 saved_val, int retry, bool force)
1796 pci_read_config_dword(pdev, offset, &val);
1797 if (!force && val == saved_val)
1801 pci_dbg(pdev, "restore config %#04x: %#010x -> %#010x\n",
1802 offset, val, saved_val);
1803 pci_write_config_dword(pdev, offset, saved_val);
1807 pci_read_config_dword(pdev, offset, &val);
1808 if (val == saved_val)
1815 static void pci_restore_config_space_range(struct pci_dev *pdev,
1816 int start, int end, int retry,
1821 for (index = end; index >= start; index--)
1822 pci_restore_config_dword(pdev, 4 * index,
1823 pdev->saved_config_space[index],
1827 static void pci_restore_config_space(struct pci_dev *pdev)
1829 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1830 pci_restore_config_space_range(pdev, 10, 15, 0, false);
1831 /* Restore BARs before the command register. */
1832 pci_restore_config_space_range(pdev, 4, 9, 10, false);
1833 pci_restore_config_space_range(pdev, 0, 3, 0, false);
1834 } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1835 pci_restore_config_space_range(pdev, 12, 15, 0, false);
1838 * Force rewriting of prefetch registers to avoid S3 resume
1839 * issues on Intel PCI bridges that occur when these
1840 * registers are not explicitly written.
1842 pci_restore_config_space_range(pdev, 9, 11, 0, true);
1843 pci_restore_config_space_range(pdev, 0, 8, 0, false);
1845 pci_restore_config_space_range(pdev, 0, 15, 0, false);
1849 static void pci_restore_rebar_state(struct pci_dev *pdev)
1851 unsigned int pos, nbars, i;
1854 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1858 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1859 nbars = FIELD_GET(PCI_REBAR_CTRL_NBAR_MASK, ctrl);
1861 for (i = 0; i < nbars; i++, pos += 8) {
1862 struct resource *res;
1865 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1866 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1867 res = pdev->resource + bar_idx;
1868 size = pci_rebar_bytes_to_size(resource_size(res));
1869 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1870 ctrl |= FIELD_PREP(PCI_REBAR_CTRL_BAR_SIZE, size);
1871 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1876 * pci_restore_state - Restore the saved state of a PCI device
1877 * @dev: PCI device that we're dealing with
1879 void pci_restore_state(struct pci_dev *dev)
1881 if (!dev->state_saved)
1884 pci_restore_pcie_state(dev);
1885 pci_restore_pasid_state(dev);
1886 pci_restore_pri_state(dev);
1887 pci_restore_ats_state(dev);
1888 pci_restore_vc_state(dev);
1889 pci_restore_rebar_state(dev);
1890 pci_restore_dpc_state(dev);
1891 pci_restore_ptm_state(dev);
1893 pci_aer_clear_status(dev);
1894 pci_restore_aer_state(dev);
1896 pci_restore_config_space(dev);
1898 pci_restore_pcix_state(dev);
1899 pci_restore_msi_state(dev);
1901 /* Restore ACS and IOV configuration state */
1902 pci_enable_acs(dev);
1903 pci_restore_iov_state(dev);
1905 dev->state_saved = false;
1907 EXPORT_SYMBOL(pci_restore_state);
1909 struct pci_saved_state {
1910 u32 config_space[16];
1911 struct pci_cap_saved_data cap[];
1915 * pci_store_saved_state - Allocate and return an opaque struct containing
1916 * the device saved state.
1917 * @dev: PCI device that we're dealing with
1919 * Return NULL if no state or error.
1921 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1923 struct pci_saved_state *state;
1924 struct pci_cap_saved_state *tmp;
1925 struct pci_cap_saved_data *cap;
1928 if (!dev->state_saved)
1931 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1933 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1934 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1936 state = kzalloc(size, GFP_KERNEL);
1940 memcpy(state->config_space, dev->saved_config_space,
1941 sizeof(state->config_space));
1944 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1945 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1946 memcpy(cap, &tmp->cap, len);
1947 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1949 /* Empty cap_save terminates list */
1953 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1956 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1957 * @dev: PCI device that we're dealing with
1958 * @state: Saved state returned from pci_store_saved_state()
1960 int pci_load_saved_state(struct pci_dev *dev,
1961 struct pci_saved_state *state)
1963 struct pci_cap_saved_data *cap;
1965 dev->state_saved = false;
1970 memcpy(dev->saved_config_space, state->config_space,
1971 sizeof(state->config_space));
1975 struct pci_cap_saved_state *tmp;
1977 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1978 if (!tmp || tmp->cap.size != cap->size)
1981 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1982 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1983 sizeof(struct pci_cap_saved_data) + cap->size);
1986 dev->state_saved = true;
1989 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1992 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1993 * and free the memory allocated for it.
1994 * @dev: PCI device that we're dealing with
1995 * @state: Pointer to saved state returned from pci_store_saved_state()
1997 int pci_load_and_free_saved_state(struct pci_dev *dev,
1998 struct pci_saved_state **state)
2000 int ret = pci_load_saved_state(dev, *state);
2005 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
2007 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
2009 return pci_enable_resources(dev, bars);
2012 static int do_pci_enable_device(struct pci_dev *dev, int bars)
2015 struct pci_dev *bridge;
2019 err = pci_set_power_state(dev, PCI_D0);
2020 if (err < 0 && err != -EIO)
2023 bridge = pci_upstream_bridge(dev);
2025 pcie_aspm_powersave_config_link(bridge);
2027 err = pcibios_enable_device(dev, bars);
2030 pci_fixup_device(pci_fixup_enable, dev);
2032 if (dev->msi_enabled || dev->msix_enabled)
2035 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
2037 pci_read_config_word(dev, PCI_COMMAND, &cmd);
2038 if (cmd & PCI_COMMAND_INTX_DISABLE)
2039 pci_write_config_word(dev, PCI_COMMAND,
2040 cmd & ~PCI_COMMAND_INTX_DISABLE);
2047 * pci_reenable_device - Resume abandoned device
2048 * @dev: PCI device to be resumed
2050 * NOTE: This function is a backend of pci_default_resume() and is not supposed
2051 * to be called by normal code, write proper resume handler and use it instead.
2053 int pci_reenable_device(struct pci_dev *dev)
2055 if (pci_is_enabled(dev))
2056 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
2059 EXPORT_SYMBOL(pci_reenable_device);
2061 static void pci_enable_bridge(struct pci_dev *dev)
2063 struct pci_dev *bridge;
2066 bridge = pci_upstream_bridge(dev);
2068 pci_enable_bridge(bridge);
2070 if (pci_is_enabled(dev)) {
2071 if (!dev->is_busmaster)
2072 pci_set_master(dev);
2076 retval = pci_enable_device(dev);
2078 pci_err(dev, "Error enabling bridge (%d), continuing\n",
2080 pci_set_master(dev);
2083 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
2085 struct pci_dev *bridge;
2090 * Power state could be unknown at this point, either due to a fresh
2091 * boot or a device removal call. So get the current power state
2092 * so that things like MSI message writing will behave as expected
2093 * (e.g. if the device really is in D0 at enable time).
2095 pci_update_current_state(dev, dev->current_state);
2097 if (atomic_inc_return(&dev->enable_cnt) > 1)
2098 return 0; /* already enabled */
2100 bridge = pci_upstream_bridge(dev);
2102 pci_enable_bridge(bridge);
2104 /* only skip sriov related */
2105 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
2106 if (dev->resource[i].flags & flags)
2108 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
2109 if (dev->resource[i].flags & flags)
2112 err = do_pci_enable_device(dev, bars);
2114 atomic_dec(&dev->enable_cnt);
2119 * pci_enable_device_mem - Initialize a device for use with Memory space
2120 * @dev: PCI device to be initialized
2122 * Initialize device before it's used by a driver. Ask low-level code
2123 * to enable Memory resources. Wake up the device if it was suspended.
2124 * Beware, this function can fail.
2126 int pci_enable_device_mem(struct pci_dev *dev)
2128 return pci_enable_device_flags(dev, IORESOURCE_MEM);
2130 EXPORT_SYMBOL(pci_enable_device_mem);
2133 * pci_enable_device - Initialize device before it's used by a driver.
2134 * @dev: PCI device to be initialized
2136 * Initialize device before it's used by a driver. Ask low-level code
2137 * to enable I/O and memory. Wake up the device if it was suspended.
2138 * Beware, this function can fail.
2140 * Note we don't actually enable the device many times if we call
2141 * this function repeatedly (we just increment the count).
2143 int pci_enable_device(struct pci_dev *dev)
2145 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
2147 EXPORT_SYMBOL(pci_enable_device);
2150 * pcibios_device_add - provide arch specific hooks when adding device dev
2151 * @dev: the PCI device being added
2153 * Permits the platform to provide architecture specific functionality when
2154 * devices are added. This is the default implementation. Architecture
2155 * implementations can override this.
2157 int __weak pcibios_device_add(struct pci_dev *dev)
2163 * pcibios_release_device - provide arch specific hooks when releasing
2165 * @dev: the PCI device being released
2167 * Permits the platform to provide architecture specific functionality when
2168 * devices are released. This is the default implementation. Architecture
2169 * implementations can override this.
2171 void __weak pcibios_release_device(struct pci_dev *dev) {}
2174 * pcibios_disable_device - disable arch specific PCI resources for device dev
2175 * @dev: the PCI device to disable
2177 * Disables architecture specific PCI resources for the device. This
2178 * is the default implementation. Architecture implementations can
2181 void __weak pcibios_disable_device(struct pci_dev *dev) {}
2183 static void do_pci_disable_device(struct pci_dev *dev)
2187 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
2188 if (pci_command & PCI_COMMAND_MASTER) {
2189 pci_command &= ~PCI_COMMAND_MASTER;
2190 pci_write_config_word(dev, PCI_COMMAND, pci_command);
2193 pcibios_disable_device(dev);
2197 * pci_disable_enabled_device - Disable device without updating enable_cnt
2198 * @dev: PCI device to disable
2200 * NOTE: This function is a backend of PCI power management routines and is
2201 * not supposed to be called drivers.
2203 void pci_disable_enabled_device(struct pci_dev *dev)
2205 if (pci_is_enabled(dev))
2206 do_pci_disable_device(dev);
2210 * pci_disable_device - Disable PCI device after use
2211 * @dev: PCI device to be disabled
2213 * Signal to the system that the PCI device is not in use by the system
2214 * anymore. This only involves disabling PCI bus-mastering, if active.
2216 * Note we don't actually disable the device until all callers of
2217 * pci_enable_device() have called pci_disable_device().
2219 void pci_disable_device(struct pci_dev *dev)
2221 struct pci_devres *dr;
2223 dr = find_pci_dr(dev);
2227 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
2228 "disabling already-disabled device");
2230 if (atomic_dec_return(&dev->enable_cnt) != 0)
2233 do_pci_disable_device(dev);
2235 dev->is_busmaster = 0;
2237 EXPORT_SYMBOL(pci_disable_device);
2240 * pcibios_set_pcie_reset_state - set reset state for device dev
2241 * @dev: the PCIe device reset
2242 * @state: Reset state to enter into
2244 * Set the PCIe reset state for the device. This is the default
2245 * implementation. Architecture implementations can override this.
2247 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
2248 enum pcie_reset_state state)
2254 * pci_set_pcie_reset_state - set reset state for device dev
2255 * @dev: the PCIe device reset
2256 * @state: Reset state to enter into
2258 * Sets the PCI reset state for the device.
2260 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
2262 return pcibios_set_pcie_reset_state(dev, state);
2264 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
2266 #ifdef CONFIG_PCIEAER
2267 void pcie_clear_device_status(struct pci_dev *dev)
2271 pcie_capability_read_word(dev, PCI_EXP_DEVSTA, &sta);
2272 pcie_capability_write_word(dev, PCI_EXP_DEVSTA, sta);
2277 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
2278 * @dev: PCIe root port or event collector.
2280 void pcie_clear_root_pme_status(struct pci_dev *dev)
2282 pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
2286 * pci_check_pme_status - Check if given device has generated PME.
2287 * @dev: Device to check.
2289 * Check the PME status of the device and if set, clear it and clear PME enable
2290 * (if set). Return 'true' if PME status and PME enable were both set or
2291 * 'false' otherwise.
2293 bool pci_check_pme_status(struct pci_dev *dev)
2302 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
2303 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
2304 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
2307 /* Clear PME status. */
2308 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2309 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
2310 /* Disable PME to avoid interrupt flood. */
2311 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2315 pci_write_config_word(dev, pmcsr_pos, pmcsr);
2321 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
2322 * @dev: Device to handle.
2323 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
2325 * Check if @dev has generated PME and queue a resume request for it in that
2328 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
2330 if (pme_poll_reset && dev->pme_poll)
2331 dev->pme_poll = false;
2333 if (pci_check_pme_status(dev)) {
2334 pci_wakeup_event(dev);
2335 pm_request_resume(&dev->dev);
2341 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
2342 * @bus: Top bus of the subtree to walk.
2344 void pci_pme_wakeup_bus(struct pci_bus *bus)
2347 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
2352 * pci_pme_capable - check the capability of PCI device to generate PME#
2353 * @dev: PCI device to handle.
2354 * @state: PCI state from which device will issue PME#.
2356 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
2361 return !!(dev->pme_support & (1 << state));
2363 EXPORT_SYMBOL(pci_pme_capable);
2365 static void pci_pme_list_scan(struct work_struct *work)
2367 struct pci_pme_device *pme_dev, *n;
2369 mutex_lock(&pci_pme_list_mutex);
2370 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
2371 struct pci_dev *pdev = pme_dev->dev;
2373 if (pdev->pme_poll) {
2374 struct pci_dev *bridge = pdev->bus->self;
2375 struct device *dev = &pdev->dev;
2376 struct device *bdev = bridge ? &bridge->dev : NULL;
2380 * If we have a bridge, it should be in an active/D0
2381 * state or the configuration space of subordinate
2382 * devices may not be accessible or stable over the
2383 * course of the call.
2386 bref = pm_runtime_get_if_active(bdev);
2390 if (bridge->current_state != PCI_D0)
2395 * The device itself should be suspended but config
2396 * space must be accessible, therefore it cannot be in
2399 if (pm_runtime_suspended(dev) &&
2400 pdev->current_state != PCI_D3cold)
2401 pci_pme_wakeup(pdev, NULL);
2405 pm_runtime_put(bdev);
2407 list_del(&pme_dev->list);
2411 if (!list_empty(&pci_pme_list))
2412 queue_delayed_work(system_freezable_wq, &pci_pme_work,
2413 msecs_to_jiffies(PME_TIMEOUT));
2414 mutex_unlock(&pci_pme_list_mutex);
2417 static void __pci_pme_active(struct pci_dev *dev, bool enable)
2421 if (!dev->pme_support)
2424 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2425 /* Clear PME_Status by writing 1 to it and enable PME# */
2426 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2428 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2430 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2434 * pci_pme_restore - Restore PME configuration after config space restore.
2435 * @dev: PCI device to update.
2437 void pci_pme_restore(struct pci_dev *dev)
2441 if (!dev->pme_support)
2444 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2445 if (dev->wakeup_prepared) {
2446 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2447 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2449 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2450 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2452 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2456 * pci_pme_active - enable or disable PCI device's PME# function
2457 * @dev: PCI device to handle.
2458 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2460 * The caller must verify that the device is capable of generating PME# before
2461 * calling this function with @enable equal to 'true'.
2463 void pci_pme_active(struct pci_dev *dev, bool enable)
2465 __pci_pme_active(dev, enable);
2468 * PCI (as opposed to PCIe) PME requires that the device have
2469 * its PME# line hooked up correctly. Not all hardware vendors
2470 * do this, so the PME never gets delivered and the device
2471 * remains asleep. The easiest way around this is to
2472 * periodically walk the list of suspended devices and check
2473 * whether any have their PME flag set. The assumption is that
2474 * we'll wake up often enough anyway that this won't be a huge
2475 * hit, and the power savings from the devices will still be a
2478 * Although PCIe uses in-band PME message instead of PME# line
2479 * to report PME, PME does not work for some PCIe devices in
2480 * reality. For example, there are devices that set their PME
2481 * status bits, but don't really bother to send a PME message;
2482 * there are PCI Express Root Ports that don't bother to
2483 * trigger interrupts when they receive PME messages from the
2484 * devices below. So PME poll is used for PCIe devices too.
2487 if (dev->pme_poll) {
2488 struct pci_pme_device *pme_dev;
2490 pme_dev = kmalloc(sizeof(struct pci_pme_device),
2493 pci_warn(dev, "can't enable PME#\n");
2497 mutex_lock(&pci_pme_list_mutex);
2498 list_add(&pme_dev->list, &pci_pme_list);
2499 if (list_is_singular(&pci_pme_list))
2500 queue_delayed_work(system_freezable_wq,
2502 msecs_to_jiffies(PME_TIMEOUT));
2503 mutex_unlock(&pci_pme_list_mutex);
2505 mutex_lock(&pci_pme_list_mutex);
2506 list_for_each_entry(pme_dev, &pci_pme_list, list) {
2507 if (pme_dev->dev == dev) {
2508 list_del(&pme_dev->list);
2513 mutex_unlock(&pci_pme_list_mutex);
2517 pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2519 EXPORT_SYMBOL(pci_pme_active);
2522 * __pci_enable_wake - enable PCI device as wakeup event source
2523 * @dev: PCI device affected
2524 * @state: PCI state from which device will issue wakeup events
2525 * @enable: True to enable event generation; false to disable
2527 * This enables the device as a wakeup event source, or disables it.
2528 * When such events involves platform-specific hooks, those hooks are
2529 * called automatically by this routine.
2531 * Devices with legacy power management (no standard PCI PM capabilities)
2532 * always require such platform hooks.
2535 * 0 is returned on success
2536 * -EINVAL is returned if device is not supposed to wake up the system
2537 * Error code depending on the platform is returned if both the platform and
2538 * the native mechanism fail to enable the generation of wake-up events
2540 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2545 * Bridges that are not power-manageable directly only signal
2546 * wakeup on behalf of subordinate devices which is set up
2547 * elsewhere, so skip them. However, bridges that are
2548 * power-manageable may signal wakeup for themselves (for example,
2549 * on a hotplug event) and they need to be covered here.
2551 if (!pci_power_manageable(dev))
2554 /* Don't do the same thing twice in a row for one device. */
2555 if (!!enable == !!dev->wakeup_prepared)
2559 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2560 * Anderson we should be doing PME# wake enable followed by ACPI wake
2561 * enable. To disable wake-up we call the platform first, for symmetry.
2568 * Enable PME signaling if the device can signal PME from
2569 * D3cold regardless of whether or not it can signal PME from
2570 * the current target state, because that will allow it to
2571 * signal PME when the hierarchy above it goes into D3cold and
2572 * the device itself ends up in D3cold as a result of that.
2574 if (pci_pme_capable(dev, state) || pci_pme_capable(dev, PCI_D3cold))
2575 pci_pme_active(dev, true);
2578 error = platform_pci_set_wakeup(dev, true);
2582 dev->wakeup_prepared = true;
2584 platform_pci_set_wakeup(dev, false);
2585 pci_pme_active(dev, false);
2586 dev->wakeup_prepared = false;
2593 * pci_enable_wake - change wakeup settings for a PCI device
2594 * @pci_dev: Target device
2595 * @state: PCI state from which device will issue wakeup events
2596 * @enable: Whether or not to enable event generation
2598 * If @enable is set, check device_may_wakeup() for the device before calling
2599 * __pci_enable_wake() for it.
2601 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2603 if (enable && !device_may_wakeup(&pci_dev->dev))
2606 return __pci_enable_wake(pci_dev, state, enable);
2608 EXPORT_SYMBOL(pci_enable_wake);
2611 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2612 * @dev: PCI device to prepare
2613 * @enable: True to enable wake-up event generation; false to disable
2615 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2616 * and this function allows them to set that up cleanly - pci_enable_wake()
2617 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2618 * ordering constraints.
2620 * This function only returns error code if the device is not allowed to wake
2621 * up the system from sleep or it is not capable of generating PME# from both
2622 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2624 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2626 return pci_pme_capable(dev, PCI_D3cold) ?
2627 pci_enable_wake(dev, PCI_D3cold, enable) :
2628 pci_enable_wake(dev, PCI_D3hot, enable);
2630 EXPORT_SYMBOL(pci_wake_from_d3);
2633 * pci_target_state - find an appropriate low power state for a given PCI dev
2635 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2637 * Use underlying platform code to find a supported low power state for @dev.
2638 * If the platform can't manage @dev, return the deepest state from which it
2639 * can generate wake events, based on any available PME info.
2641 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2643 if (platform_pci_power_manageable(dev)) {
2645 * Call the platform to find the target state for the device.
2647 pci_power_t state = platform_pci_choose_state(dev);
2650 case PCI_POWER_ERROR:
2656 if (pci_no_d1d2(dev))
2664 * If the device is in D3cold even though it's not power-manageable by
2665 * the platform, it may have been powered down by non-standard means.
2666 * Best to let it slumber.
2668 if (dev->current_state == PCI_D3cold)
2670 else if (!dev->pm_cap)
2673 if (wakeup && dev->pme_support) {
2674 pci_power_t state = PCI_D3hot;
2677 * Find the deepest state from which the device can generate
2680 while (state && !(dev->pme_support & (1 << state)))
2685 else if (dev->pme_support & 1)
2693 * pci_prepare_to_sleep - prepare PCI device for system-wide transition
2694 * into a sleep state
2695 * @dev: Device to handle.
2697 * Choose the power state appropriate for the device depending on whether
2698 * it can wake up the system and/or is power manageable by the platform
2699 * (PCI_D3hot is the default) and put the device into that state.
2701 int pci_prepare_to_sleep(struct pci_dev *dev)
2703 bool wakeup = device_may_wakeup(&dev->dev);
2704 pci_power_t target_state = pci_target_state(dev, wakeup);
2707 if (target_state == PCI_POWER_ERROR)
2710 pci_enable_wake(dev, target_state, wakeup);
2712 error = pci_set_power_state(dev, target_state);
2715 pci_enable_wake(dev, target_state, false);
2719 EXPORT_SYMBOL(pci_prepare_to_sleep);
2722 * pci_back_from_sleep - turn PCI device on during system-wide transition
2723 * into working state
2724 * @dev: Device to handle.
2726 * Disable device's system wake-up capability and put it into D0.
2728 int pci_back_from_sleep(struct pci_dev *dev)
2730 int ret = pci_set_power_state(dev, PCI_D0);
2735 pci_enable_wake(dev, PCI_D0, false);
2738 EXPORT_SYMBOL(pci_back_from_sleep);
2741 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2742 * @dev: PCI device being suspended.
2744 * Prepare @dev to generate wake-up events at run time and put it into a low
2747 int pci_finish_runtime_suspend(struct pci_dev *dev)
2749 pci_power_t target_state;
2752 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2753 if (target_state == PCI_POWER_ERROR)
2756 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2758 error = pci_set_power_state(dev, target_state);
2761 pci_enable_wake(dev, target_state, false);
2767 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2768 * @dev: Device to check.
2770 * Return true if the device itself is capable of generating wake-up events
2771 * (through the platform or using the native PCIe PME) or if the device supports
2772 * PME and one of its upstream bridges can generate wake-up events.
2774 bool pci_dev_run_wake(struct pci_dev *dev)
2776 struct pci_bus *bus = dev->bus;
2778 if (!dev->pme_support)
2781 /* PME-capable in principle, but not from the target power state */
2782 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2785 if (device_can_wakeup(&dev->dev))
2788 while (bus->parent) {
2789 struct pci_dev *bridge = bus->self;
2791 if (device_can_wakeup(&bridge->dev))
2797 /* We have reached the root bus. */
2799 return device_can_wakeup(bus->bridge);
2803 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2806 * pci_dev_need_resume - Check if it is necessary to resume the device.
2807 * @pci_dev: Device to check.
2809 * Return 'true' if the device is not runtime-suspended or it has to be
2810 * reconfigured due to wakeup settings difference between system and runtime
2811 * suspend, or the current power state of it is not suitable for the upcoming
2812 * (system-wide) transition.
2814 bool pci_dev_need_resume(struct pci_dev *pci_dev)
2816 struct device *dev = &pci_dev->dev;
2817 pci_power_t target_state;
2819 if (!pm_runtime_suspended(dev) || platform_pci_need_resume(pci_dev))
2822 target_state = pci_target_state(pci_dev, device_may_wakeup(dev));
2825 * If the earlier platform check has not triggered, D3cold is just power
2826 * removal on top of D3hot, so no need to resume the device in that
2829 return target_state != pci_dev->current_state &&
2830 target_state != PCI_D3cold &&
2831 pci_dev->current_state != PCI_D3hot;
2835 * pci_dev_adjust_pme - Adjust PME setting for a suspended device.
2836 * @pci_dev: Device to check.
2838 * If the device is suspended and it is not configured for system wakeup,
2839 * disable PME for it to prevent it from waking up the system unnecessarily.
2841 * Note that if the device's power state is D3cold and the platform check in
2842 * pci_dev_need_resume() has not triggered, the device's configuration need not
2845 void pci_dev_adjust_pme(struct pci_dev *pci_dev)
2847 struct device *dev = &pci_dev->dev;
2849 spin_lock_irq(&dev->power.lock);
2851 if (pm_runtime_suspended(dev) && !device_may_wakeup(dev) &&
2852 pci_dev->current_state < PCI_D3cold)
2853 __pci_pme_active(pci_dev, false);
2855 spin_unlock_irq(&dev->power.lock);
2859 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2860 * @pci_dev: Device to handle.
2862 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2863 * it might have been disabled during the prepare phase of system suspend if
2864 * the device was not configured for system wakeup.
2866 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2868 struct device *dev = &pci_dev->dev;
2870 if (!pci_dev_run_wake(pci_dev))
2873 spin_lock_irq(&dev->power.lock);
2875 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2876 __pci_pme_active(pci_dev, true);
2878 spin_unlock_irq(&dev->power.lock);
2882 * pci_choose_state - Choose the power state of a PCI device.
2883 * @dev: Target PCI device.
2884 * @state: Target state for the whole system.
2886 * Returns PCI power state suitable for @dev and @state.
2888 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
2890 if (state.event == PM_EVENT_ON)
2893 return pci_target_state(dev, false);
2895 EXPORT_SYMBOL(pci_choose_state);
2897 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2899 struct device *dev = &pdev->dev;
2900 struct device *parent = dev->parent;
2903 pm_runtime_get_sync(parent);
2904 pm_runtime_get_noresume(dev);
2906 * pdev->current_state is set to PCI_D3cold during suspending,
2907 * so wait until suspending completes
2909 pm_runtime_barrier(dev);
2911 * Only need to resume devices in D3cold, because config
2912 * registers are still accessible for devices suspended but
2915 if (pdev->current_state == PCI_D3cold)
2916 pm_runtime_resume(dev);
2919 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2921 struct device *dev = &pdev->dev;
2922 struct device *parent = dev->parent;
2924 pm_runtime_put(dev);
2926 pm_runtime_put_sync(parent);
2929 static const struct dmi_system_id bridge_d3_blacklist[] = {
2933 * Gigabyte X299 root port is not marked as hotplug capable
2934 * which allows Linux to power manage it. However, this
2935 * confuses the BIOS SMI handler so don't power manage root
2936 * ports on that system.
2938 .ident = "X299 DESIGNARE EX-CF",
2940 DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2941 DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2946 * Downstream device is not accessible after putting a root port
2947 * into D3cold and back into D0 on Elo Continental Z2 board
2949 .ident = "Elo Continental Z2",
2951 DMI_MATCH(DMI_BOARD_VENDOR, "Elo Touch Solutions"),
2952 DMI_MATCH(DMI_BOARD_NAME, "Geminilake"),
2953 DMI_MATCH(DMI_BOARD_VERSION, "Continental Z2"),
2958 * Changing power state of root port dGPU is connected fails
2959 * https://gitlab.freedesktop.org/drm/amd/-/issues/3229
2961 .ident = "Hewlett-Packard HP Pavilion 17 Notebook PC/1972",
2963 DMI_MATCH(DMI_BOARD_VENDOR, "Hewlett-Packard"),
2964 DMI_MATCH(DMI_BOARD_NAME, "1972"),
2965 DMI_MATCH(DMI_BOARD_VERSION, "95.33"),
2973 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2974 * @bridge: Bridge to check
2976 * This function checks if it is possible to move the bridge to D3.
2977 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2979 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2981 if (!pci_is_pcie(bridge))
2984 switch (pci_pcie_type(bridge)) {
2985 case PCI_EXP_TYPE_ROOT_PORT:
2986 case PCI_EXP_TYPE_UPSTREAM:
2987 case PCI_EXP_TYPE_DOWNSTREAM:
2988 if (pci_bridge_d3_disable)
2992 * Hotplug ports handled by firmware in System Management Mode
2993 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2995 if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2998 if (pci_bridge_d3_force)
3001 /* Even the oldest 2010 Thunderbolt controller supports D3. */
3002 if (bridge->is_thunderbolt)
3005 /* Platform might know better if the bridge supports D3 */
3006 if (platform_pci_bridge_d3(bridge))
3010 * Hotplug ports handled natively by the OS were not validated
3011 * by vendors for runtime D3 at least until 2018 because there
3012 * was no OS support.
3014 if (bridge->is_hotplug_bridge)
3017 if (dmi_check_system(bridge_d3_blacklist))
3021 * It should be safe to put PCIe ports from 2015 or newer
3024 if (dmi_get_bios_year() >= 2015)
3032 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
3034 bool *d3cold_ok = data;
3036 if (/* The device needs to be allowed to go D3cold ... */
3037 dev->no_d3cold || !dev->d3cold_allowed ||
3039 /* ... and if it is wakeup capable to do so from D3cold. */
3040 (device_may_wakeup(&dev->dev) &&
3041 !pci_pme_capable(dev, PCI_D3cold)) ||
3043 /* If it is a bridge it must be allowed to go to D3. */
3044 !pci_power_manageable(dev))
3052 * pci_bridge_d3_update - Update bridge D3 capabilities
3053 * @dev: PCI device which is changed
3055 * Update upstream bridge PM capabilities accordingly depending on if the
3056 * device PM configuration was changed or the device is being removed. The
3057 * change is also propagated upstream.
3059 void pci_bridge_d3_update(struct pci_dev *dev)
3061 bool remove = !device_is_registered(&dev->dev);
3062 struct pci_dev *bridge;
3063 bool d3cold_ok = true;
3065 bridge = pci_upstream_bridge(dev);
3066 if (!bridge || !pci_bridge_d3_possible(bridge))
3070 * If D3 is currently allowed for the bridge, removing one of its
3071 * children won't change that.
3073 if (remove && bridge->bridge_d3)
3077 * If D3 is currently allowed for the bridge and a child is added or
3078 * changed, disallowance of D3 can only be caused by that child, so
3079 * we only need to check that single device, not any of its siblings.
3081 * If D3 is currently not allowed for the bridge, checking the device
3082 * first may allow us to skip checking its siblings.
3085 pci_dev_check_d3cold(dev, &d3cold_ok);
3088 * If D3 is currently not allowed for the bridge, this may be caused
3089 * either by the device being changed/removed or any of its siblings,
3090 * so we need to go through all children to find out if one of them
3091 * continues to block D3.
3093 if (d3cold_ok && !bridge->bridge_d3)
3094 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
3097 if (bridge->bridge_d3 != d3cold_ok) {
3098 bridge->bridge_d3 = d3cold_ok;
3099 /* Propagate change to upstream bridges */
3100 pci_bridge_d3_update(bridge);
3105 * pci_d3cold_enable - Enable D3cold for device
3106 * @dev: PCI device to handle
3108 * This function can be used in drivers to enable D3cold from the device
3109 * they handle. It also updates upstream PCI bridge PM capabilities
3112 void pci_d3cold_enable(struct pci_dev *dev)
3114 if (dev->no_d3cold) {
3115 dev->no_d3cold = false;
3116 pci_bridge_d3_update(dev);
3119 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
3122 * pci_d3cold_disable - Disable D3cold for device
3123 * @dev: PCI device to handle
3125 * This function can be used in drivers to disable D3cold from the device
3126 * they handle. It also updates upstream PCI bridge PM capabilities
3129 void pci_d3cold_disable(struct pci_dev *dev)
3131 if (!dev->no_d3cold) {
3132 dev->no_d3cold = true;
3133 pci_bridge_d3_update(dev);
3136 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
3139 * pci_pm_init - Initialize PM functions of given PCI device
3140 * @dev: PCI device to handle.
3142 void pci_pm_init(struct pci_dev *dev)
3148 pm_runtime_forbid(&dev->dev);
3149 pm_runtime_set_active(&dev->dev);
3150 pm_runtime_enable(&dev->dev);
3151 device_enable_async_suspend(&dev->dev);
3152 dev->wakeup_prepared = false;
3155 dev->pme_support = 0;
3157 /* find PCI PM capability in list */
3158 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
3161 /* Check device's ability to generate PME# */
3162 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
3164 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
3165 pci_err(dev, "unsupported PM cap regs version (%u)\n",
3166 pmc & PCI_PM_CAP_VER_MASK);
3171 dev->d3hot_delay = PCI_PM_D3HOT_WAIT;
3172 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
3173 dev->bridge_d3 = pci_bridge_d3_possible(dev);
3174 dev->d3cold_allowed = true;
3176 dev->d1_support = false;
3177 dev->d2_support = false;
3178 if (!pci_no_d1d2(dev)) {
3179 if (pmc & PCI_PM_CAP_D1)
3180 dev->d1_support = true;
3181 if (pmc & PCI_PM_CAP_D2)
3182 dev->d2_support = true;
3184 if (dev->d1_support || dev->d2_support)
3185 pci_info(dev, "supports%s%s\n",
3186 dev->d1_support ? " D1" : "",
3187 dev->d2_support ? " D2" : "");
3190 pmc &= PCI_PM_CAP_PME_MASK;
3192 pci_info(dev, "PME# supported from%s%s%s%s%s\n",
3193 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
3194 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
3195 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
3196 (pmc & PCI_PM_CAP_PME_D3hot) ? " D3hot" : "",
3197 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
3198 dev->pme_support = FIELD_GET(PCI_PM_CAP_PME_MASK, pmc);
3199 dev->pme_poll = true;
3201 * Make device's PM flags reflect the wake-up capability, but
3202 * let the user space enable it to wake up the system as needed.
3204 device_set_wakeup_capable(&dev->dev, true);
3205 /* Disable the PME# generation functionality */
3206 pci_pme_active(dev, false);
3209 pci_read_config_word(dev, PCI_STATUS, &status);
3210 if (status & PCI_STATUS_IMM_READY)
3214 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
3216 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
3220 case PCI_EA_P_VF_MEM:
3221 flags |= IORESOURCE_MEM;
3223 case PCI_EA_P_MEM_PREFETCH:
3224 case PCI_EA_P_VF_MEM_PREFETCH:
3225 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
3228 flags |= IORESOURCE_IO;
3237 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
3240 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
3241 return &dev->resource[bei];
3242 #ifdef CONFIG_PCI_IOV
3243 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
3244 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
3245 return &dev->resource[PCI_IOV_RESOURCES +
3246 bei - PCI_EA_BEI_VF_BAR0];
3248 else if (bei == PCI_EA_BEI_ROM)
3249 return &dev->resource[PCI_ROM_RESOURCE];
3254 /* Read an Enhanced Allocation (EA) entry */
3255 static int pci_ea_read(struct pci_dev *dev, int offset)
3257 struct resource *res;
3258 const char *res_name;
3259 int ent_size, ent_offset = offset;
3260 resource_size_t start, end;
3261 unsigned long flags;
3262 u32 dw0, bei, base, max_offset;
3264 bool support_64 = (sizeof(resource_size_t) >= 8);
3266 pci_read_config_dword(dev, ent_offset, &dw0);
3269 /* Entry size field indicates DWORDs after 1st */
3270 ent_size = (FIELD_GET(PCI_EA_ES, dw0) + 1) << 2;
3272 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
3275 bei = FIELD_GET(PCI_EA_BEI, dw0);
3276 prop = FIELD_GET(PCI_EA_PP, dw0);
3279 * If the Property is in the reserved range, try the Secondary
3282 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
3283 prop = FIELD_GET(PCI_EA_SP, dw0);
3284 if (prop > PCI_EA_P_BRIDGE_IO)
3287 res = pci_ea_get_resource(dev, bei, prop);
3288 res_name = pci_resource_name(dev, bei);
3290 pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
3294 flags = pci_ea_flags(dev, prop);
3296 pci_err(dev, "Unsupported EA properties: %#x\n", prop);
3301 pci_read_config_dword(dev, ent_offset, &base);
3302 start = (base & PCI_EA_FIELD_MASK);
3305 /* Read MaxOffset */
3306 pci_read_config_dword(dev, ent_offset, &max_offset);
3309 /* Read Base MSBs (if 64-bit entry) */
3310 if (base & PCI_EA_IS_64) {
3313 pci_read_config_dword(dev, ent_offset, &base_upper);
3316 flags |= IORESOURCE_MEM_64;
3318 /* entry starts above 32-bit boundary, can't use */
3319 if (!support_64 && base_upper)
3323 start |= ((u64)base_upper << 32);
3326 end = start + (max_offset | 0x03);
3328 /* Read MaxOffset MSBs (if 64-bit entry) */
3329 if (max_offset & PCI_EA_IS_64) {
3330 u32 max_offset_upper;
3332 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
3335 flags |= IORESOURCE_MEM_64;
3337 /* entry too big, can't use */
3338 if (!support_64 && max_offset_upper)
3342 end += ((u64)max_offset_upper << 32);
3346 pci_err(dev, "EA Entry crosses address boundary\n");
3350 if (ent_size != ent_offset - offset) {
3351 pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
3352 ent_size, ent_offset - offset);
3356 res->name = pci_name(dev);
3361 if (bei <= PCI_EA_BEI_BAR5)
3362 pci_info(dev, "%s %pR: from Enhanced Allocation, properties %#02x\n",
3363 res_name, res, prop);
3364 else if (bei == PCI_EA_BEI_ROM)
3365 pci_info(dev, "%s %pR: from Enhanced Allocation, properties %#02x\n",
3366 res_name, res, prop);
3367 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
3368 pci_info(dev, "%s %pR: from Enhanced Allocation, properties %#02x\n",
3369 res_name, res, prop);
3371 pci_info(dev, "BEI %d %pR: from Enhanced Allocation, properties %#02x\n",
3375 return offset + ent_size;
3378 /* Enhanced Allocation Initialization */
3379 void pci_ea_init(struct pci_dev *dev)
3386 /* find PCI EA capability in list */
3387 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
3391 /* determine the number of entries */
3392 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
3394 num_ent &= PCI_EA_NUM_ENT_MASK;
3396 offset = ea + PCI_EA_FIRST_ENT;
3398 /* Skip DWORD 2 for type 1 functions */
3399 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
3402 /* parse each EA entry */
3403 for (i = 0; i < num_ent; ++i)
3404 offset = pci_ea_read(dev, offset);
3407 static void pci_add_saved_cap(struct pci_dev *pci_dev,
3408 struct pci_cap_saved_state *new_cap)
3410 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
3414 * _pci_add_cap_save_buffer - allocate buffer for saving given
3415 * capability registers
3416 * @dev: the PCI device
3417 * @cap: the capability to allocate the buffer for
3418 * @extended: Standard or Extended capability ID
3419 * @size: requested size of the buffer
3421 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
3422 bool extended, unsigned int size)
3425 struct pci_cap_saved_state *save_state;
3428 pos = pci_find_ext_capability(dev, cap);
3430 pos = pci_find_capability(dev, cap);
3435 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
3439 save_state->cap.cap_nr = cap;
3440 save_state->cap.cap_extended = extended;
3441 save_state->cap.size = size;
3442 pci_add_saved_cap(dev, save_state);
3447 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
3449 return _pci_add_cap_save_buffer(dev, cap, false, size);
3452 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
3454 return _pci_add_cap_save_buffer(dev, cap, true, size);
3458 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
3459 * @dev: the PCI device
3461 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
3465 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3466 PCI_EXP_SAVE_REGS * sizeof(u16));
3468 pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3470 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3472 pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3474 error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_LTR,
3477 pci_err(dev, "unable to allocate suspend buffer for LTR\n");
3479 pci_allocate_vc_save_buffers(dev);
3482 void pci_free_cap_save_buffers(struct pci_dev *dev)
3484 struct pci_cap_saved_state *tmp;
3485 struct hlist_node *n;
3487 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3492 * pci_configure_ari - enable or disable ARI forwarding
3493 * @dev: the PCI device
3495 * If @dev and its upstream bridge both support ARI, enable ARI in the
3496 * bridge. Otherwise, disable ARI in the bridge.
3498 void pci_configure_ari(struct pci_dev *dev)
3501 struct pci_dev *bridge;
3503 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3506 bridge = dev->bus->self;
3510 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3511 if (!(cap & PCI_EXP_DEVCAP2_ARI))
3514 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3515 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3516 PCI_EXP_DEVCTL2_ARI);
3517 bridge->ari_enabled = 1;
3519 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3520 PCI_EXP_DEVCTL2_ARI);
3521 bridge->ari_enabled = 0;
3525 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3530 pos = pdev->acs_cap;
3535 * Except for egress control, capabilities are either required
3536 * or only required if controllable. Features missing from the
3537 * capability field can therefore be assumed as hard-wired enabled.
3539 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3540 acs_flags &= (cap | PCI_ACS_EC);
3542 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3543 return (ctrl & acs_flags) == acs_flags;
3547 * pci_acs_enabled - test ACS against required flags for a given device
3548 * @pdev: device to test
3549 * @acs_flags: required PCI ACS flags
3551 * Return true if the device supports the provided flags. Automatically
3552 * filters out flags that are not implemented on multifunction devices.
3554 * Note that this interface checks the effective ACS capabilities of the
3555 * device rather than the actual capabilities. For instance, most single
3556 * function endpoints are not required to support ACS because they have no
3557 * opportunity for peer-to-peer access. We therefore return 'true'
3558 * regardless of whether the device exposes an ACS capability. This makes
3559 * it much easier for callers of this function to ignore the actual type
3560 * or topology of the device when testing ACS support.
3562 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3566 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3571 * Conventional PCI and PCI-X devices never support ACS, either
3572 * effectively or actually. The shared bus topology implies that
3573 * any device on the bus can receive or snoop DMA.
3575 if (!pci_is_pcie(pdev))
3578 switch (pci_pcie_type(pdev)) {
3580 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3581 * but since their primary interface is PCI/X, we conservatively
3582 * handle them as we would a non-PCIe device.
3584 case PCI_EXP_TYPE_PCIE_BRIDGE:
3586 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
3587 * applicable... must never implement an ACS Extended Capability...".
3588 * This seems arbitrary, but we take a conservative interpretation
3589 * of this statement.
3591 case PCI_EXP_TYPE_PCI_BRIDGE:
3592 case PCI_EXP_TYPE_RC_EC:
3595 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3596 * implement ACS in order to indicate their peer-to-peer capabilities,
3597 * regardless of whether they are single- or multi-function devices.
3599 case PCI_EXP_TYPE_DOWNSTREAM:
3600 case PCI_EXP_TYPE_ROOT_PORT:
3601 return pci_acs_flags_enabled(pdev, acs_flags);
3603 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3604 * implemented by the remaining PCIe types to indicate peer-to-peer
3605 * capabilities, but only when they are part of a multifunction
3606 * device. The footnote for section 6.12 indicates the specific
3607 * PCIe types included here.
3609 case PCI_EXP_TYPE_ENDPOINT:
3610 case PCI_EXP_TYPE_UPSTREAM:
3611 case PCI_EXP_TYPE_LEG_END:
3612 case PCI_EXP_TYPE_RC_END:
3613 if (!pdev->multifunction)
3616 return pci_acs_flags_enabled(pdev, acs_flags);
3620 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3621 * to single function devices with the exception of downstream ports.
3627 * pci_acs_path_enabled - test ACS flags from start to end in a hierarchy
3628 * @start: starting downstream device
3629 * @end: ending upstream device or NULL to search to the root bus
3630 * @acs_flags: required flags
3632 * Walk up a device tree from start to end testing PCI ACS support. If
3633 * any step along the way does not support the required flags, return false.
3635 bool pci_acs_path_enabled(struct pci_dev *start,
3636 struct pci_dev *end, u16 acs_flags)
3638 struct pci_dev *pdev, *parent = start;
3643 if (!pci_acs_enabled(pdev, acs_flags))
3646 if (pci_is_root_bus(pdev->bus))
3647 return (end == NULL);
3649 parent = pdev->bus->self;
3650 } while (pdev != end);
3656 * pci_acs_init - Initialize ACS if hardware supports it
3657 * @dev: the PCI device
3659 void pci_acs_init(struct pci_dev *dev)
3661 dev->acs_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3664 * Attempt to enable ACS regardless of capability because some Root
3665 * Ports (e.g. those quirked with *_intel_pch_acs_*) do not have
3666 * the standard ACS capability but still support ACS via those
3669 pci_enable_acs(dev);
3673 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3677 * Helper to find the position of the ctrl register for a BAR.
3678 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3679 * Returns -ENOENT if no ctrl register for the BAR could be found.
3681 static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3683 unsigned int pos, nbars, i;
3686 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3690 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3691 nbars = FIELD_GET(PCI_REBAR_CTRL_NBAR_MASK, ctrl);
3693 for (i = 0; i < nbars; i++, pos += 8) {
3696 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3697 bar_idx = FIELD_GET(PCI_REBAR_CTRL_BAR_IDX, ctrl);
3706 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3708 * @bar: BAR to query
3710 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3711 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3713 u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3718 pos = pci_rebar_find_pos(pdev, bar);
3722 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3723 cap = FIELD_GET(PCI_REBAR_CAP_SIZES, cap);
3725 /* Sapphire RX 5600 XT Pulse has an invalid cap dword for BAR 0 */
3726 if (pdev->vendor == PCI_VENDOR_ID_ATI && pdev->device == 0x731f &&
3727 bar == 0 && cap == 0x700)
3732 EXPORT_SYMBOL(pci_rebar_get_possible_sizes);
3735 * pci_rebar_get_current_size - get the current size of a BAR
3737 * @bar: BAR to set size to
3739 * Read the size of a BAR from the resizable BAR config.
3740 * Returns size if found or negative error code.
3742 int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3747 pos = pci_rebar_find_pos(pdev, bar);
3751 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3752 return FIELD_GET(PCI_REBAR_CTRL_BAR_SIZE, ctrl);
3756 * pci_rebar_set_size - set a new size for a BAR
3758 * @bar: BAR to set size to
3759 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3761 * Set the new size of a BAR as defined in the spec.
3762 * Returns zero if resizing was successful, error code otherwise.
3764 int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3769 pos = pci_rebar_find_pos(pdev, bar);
3773 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3774 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3775 ctrl |= FIELD_PREP(PCI_REBAR_CTRL_BAR_SIZE, size);
3776 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3781 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3782 * @dev: the PCI device
3783 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3784 * PCI_EXP_DEVCAP2_ATOMIC_COMP32
3785 * PCI_EXP_DEVCAP2_ATOMIC_COMP64
3786 * PCI_EXP_DEVCAP2_ATOMIC_COMP128
3788 * Return 0 if all upstream bridges support AtomicOp routing, egress
3789 * blocking is disabled on all upstream ports, and the root port supports
3790 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3791 * AtomicOp completion), or negative otherwise.
3793 int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3795 struct pci_bus *bus = dev->bus;
3796 struct pci_dev *bridge;
3800 * Per PCIe r5.0, sec 9.3.5.10, the AtomicOp Requester Enable bit
3801 * in Device Control 2 is reserved in VFs and the PF value applies
3802 * to all associated VFs.
3807 if (!pci_is_pcie(dev))
3811 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3812 * AtomicOp requesters. For now, we only support endpoints as
3813 * requesters and root ports as completers. No endpoints as
3814 * completers, and no peer-to-peer.
3817 switch (pci_pcie_type(dev)) {
3818 case PCI_EXP_TYPE_ENDPOINT:
3819 case PCI_EXP_TYPE_LEG_END:
3820 case PCI_EXP_TYPE_RC_END:
3826 while (bus->parent) {
3829 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3831 switch (pci_pcie_type(bridge)) {
3832 /* Ensure switch ports support AtomicOp routing */
3833 case PCI_EXP_TYPE_UPSTREAM:
3834 case PCI_EXP_TYPE_DOWNSTREAM:
3835 if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3839 /* Ensure root port supports all the sizes we care about */
3840 case PCI_EXP_TYPE_ROOT_PORT:
3841 if ((cap & cap_mask) != cap_mask)
3846 /* Ensure upstream ports don't block AtomicOps on egress */
3847 if (pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM) {
3848 pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3850 if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3857 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3858 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3861 EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3864 * pci_release_region - Release a PCI bar
3865 * @pdev: PCI device whose resources were previously reserved by
3866 * pci_request_region()
3867 * @bar: BAR to release
3869 * Releases the PCI I/O and memory resources previously reserved by a
3870 * successful call to pci_request_region(). Call this function only
3871 * after all use of the PCI regions has ceased.
3873 void pci_release_region(struct pci_dev *pdev, int bar)
3875 struct pci_devres *dr;
3877 if (pci_resource_len(pdev, bar) == 0)
3879 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3880 release_region(pci_resource_start(pdev, bar),
3881 pci_resource_len(pdev, bar));
3882 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3883 release_mem_region(pci_resource_start(pdev, bar),
3884 pci_resource_len(pdev, bar));
3886 dr = find_pci_dr(pdev);
3888 dr->region_mask &= ~(1 << bar);
3890 EXPORT_SYMBOL(pci_release_region);
3893 * __pci_request_region - Reserved PCI I/O and memory resource
3894 * @pdev: PCI device whose resources are to be reserved
3895 * @bar: BAR to be reserved
3896 * @res_name: Name to be associated with resource.
3897 * @exclusive: whether the region access is exclusive or not
3899 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3900 * being reserved by owner @res_name. Do not access any
3901 * address inside the PCI regions unless this call returns
3904 * If @exclusive is set, then the region is marked so that userspace
3905 * is explicitly not allowed to map the resource via /dev/mem or
3906 * sysfs MMIO access.
3908 * Returns 0 on success, or %EBUSY on error. A warning
3909 * message is also printed on failure.
3911 static int __pci_request_region(struct pci_dev *pdev, int bar,
3912 const char *res_name, int exclusive)
3914 struct pci_devres *dr;
3916 if (pci_resource_len(pdev, bar) == 0)
3919 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3920 if (!request_region(pci_resource_start(pdev, bar),
3921 pci_resource_len(pdev, bar), res_name))
3923 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3924 if (!__request_mem_region(pci_resource_start(pdev, bar),
3925 pci_resource_len(pdev, bar), res_name,
3930 dr = find_pci_dr(pdev);
3932 dr->region_mask |= 1 << bar;
3937 pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3938 &pdev->resource[bar]);
3943 * pci_request_region - Reserve PCI I/O and memory resource
3944 * @pdev: PCI device whose resources are to be reserved
3945 * @bar: BAR to be reserved
3946 * @res_name: Name to be associated with resource
3948 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3949 * being reserved by owner @res_name. Do not access any
3950 * address inside the PCI regions unless this call returns
3953 * Returns 0 on success, or %EBUSY on error. A warning
3954 * message is also printed on failure.
3956 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3958 return __pci_request_region(pdev, bar, res_name, 0);
3960 EXPORT_SYMBOL(pci_request_region);
3963 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3964 * @pdev: PCI device whose resources were previously reserved
3965 * @bars: Bitmask of BARs to be released
3967 * Release selected PCI I/O and memory resources previously reserved.
3968 * Call this function only after all use of the PCI regions has ceased.
3970 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3974 for (i = 0; i < PCI_STD_NUM_BARS; i++)
3975 if (bars & (1 << i))
3976 pci_release_region(pdev, i);
3978 EXPORT_SYMBOL(pci_release_selected_regions);
3980 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3981 const char *res_name, int excl)
3985 for (i = 0; i < PCI_STD_NUM_BARS; i++)
3986 if (bars & (1 << i))
3987 if (__pci_request_region(pdev, i, res_name, excl))
3993 if (bars & (1 << i))
3994 pci_release_region(pdev, i);
4001 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
4002 * @pdev: PCI device whose resources are to be reserved
4003 * @bars: Bitmask of BARs to be requested
4004 * @res_name: Name to be associated with resource
4006 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
4007 const char *res_name)
4009 return __pci_request_selected_regions(pdev, bars, res_name, 0);
4011 EXPORT_SYMBOL(pci_request_selected_regions);
4013 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
4014 const char *res_name)
4016 return __pci_request_selected_regions(pdev, bars, res_name,
4017 IORESOURCE_EXCLUSIVE);
4019 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
4022 * pci_release_regions - Release reserved PCI I/O and memory resources
4023 * @pdev: PCI device whose resources were previously reserved by
4024 * pci_request_regions()
4026 * Releases all PCI I/O and memory resources previously reserved by a
4027 * successful call to pci_request_regions(). Call this function only
4028 * after all use of the PCI regions has ceased.
4031 void pci_release_regions(struct pci_dev *pdev)
4033 pci_release_selected_regions(pdev, (1 << PCI_STD_NUM_BARS) - 1);
4035 EXPORT_SYMBOL(pci_release_regions);
4038 * pci_request_regions - Reserve PCI I/O and memory resources
4039 * @pdev: PCI device whose resources are to be reserved
4040 * @res_name: Name to be associated with resource.
4042 * Mark all PCI regions associated with PCI device @pdev as
4043 * being reserved by owner @res_name. Do not access any
4044 * address inside the PCI regions unless this call returns
4047 * Returns 0 on success, or %EBUSY on error. A warning
4048 * message is also printed on failure.
4050 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
4052 return pci_request_selected_regions(pdev,
4053 ((1 << PCI_STD_NUM_BARS) - 1), res_name);
4055 EXPORT_SYMBOL(pci_request_regions);
4058 * pci_request_regions_exclusive - Reserve PCI I/O and memory resources
4059 * @pdev: PCI device whose resources are to be reserved
4060 * @res_name: Name to be associated with resource.
4062 * Mark all PCI regions associated with PCI device @pdev as being reserved
4063 * by owner @res_name. Do not access any address inside the PCI regions
4064 * unless this call returns successfully.
4066 * pci_request_regions_exclusive() will mark the region so that /dev/mem
4067 * and the sysfs MMIO access will not be allowed.
4069 * Returns 0 on success, or %EBUSY on error. A warning message is also
4070 * printed on failure.
4072 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
4074 return pci_request_selected_regions_exclusive(pdev,
4075 ((1 << PCI_STD_NUM_BARS) - 1), res_name);
4077 EXPORT_SYMBOL(pci_request_regions_exclusive);
4080 * Record the PCI IO range (expressed as CPU physical address + size).
4081 * Return a negative value if an error has occurred, zero otherwise
4083 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
4084 resource_size_t size)
4088 struct logic_pio_hwaddr *range;
4090 if (!size || addr + size < addr)
4093 range = kzalloc(sizeof(*range), GFP_ATOMIC);
4097 range->fwnode = fwnode;
4099 range->hw_start = addr;
4100 range->flags = LOGIC_PIO_CPU_MMIO;
4102 ret = logic_pio_register_range(range);
4106 /* Ignore duplicates due to deferred probing */
4114 phys_addr_t pci_pio_to_address(unsigned long pio)
4117 if (pio < MMIO_UPPER_LIMIT)
4118 return logic_pio_to_hwaddr(pio);
4121 return (phys_addr_t) OF_BAD_ADDR;
4123 EXPORT_SYMBOL_GPL(pci_pio_to_address);
4125 unsigned long __weak pci_address_to_pio(phys_addr_t address)
4128 return logic_pio_trans_cpuaddr(address);
4130 if (address > IO_SPACE_LIMIT)
4131 return (unsigned long)-1;
4133 return (unsigned long) address;
4138 * pci_remap_iospace - Remap the memory mapped I/O space
4139 * @res: Resource describing the I/O space
4140 * @phys_addr: physical address of range to be mapped
4142 * Remap the memory mapped I/O space described by the @res and the CPU
4143 * physical address @phys_addr into virtual address space. Only
4144 * architectures that have memory mapped IO functions defined (and the
4145 * PCI_IOBASE value defined) should call this function.
4147 #ifndef pci_remap_iospace
4148 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
4150 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4151 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4153 if (!(res->flags & IORESOURCE_IO))
4156 if (res->end > IO_SPACE_LIMIT)
4159 return vmap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
4160 pgprot_device(PAGE_KERNEL));
4163 * This architecture does not have memory mapped I/O space,
4164 * so this function should never be called
4166 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
4170 EXPORT_SYMBOL(pci_remap_iospace);
4174 * pci_unmap_iospace - Unmap the memory mapped I/O space
4175 * @res: resource to be unmapped
4177 * Unmap the CPU virtual address @res from virtual address space. Only
4178 * architectures that have memory mapped IO functions defined (and the
4179 * PCI_IOBASE value defined) should call this function.
4181 void pci_unmap_iospace(struct resource *res)
4183 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
4184 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
4186 vunmap_range(vaddr, vaddr + resource_size(res));
4189 EXPORT_SYMBOL(pci_unmap_iospace);
4191 static void __pci_set_master(struct pci_dev *dev, bool enable)
4195 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4197 cmd = old_cmd | PCI_COMMAND_MASTER;
4199 cmd = old_cmd & ~PCI_COMMAND_MASTER;
4200 if (cmd != old_cmd) {
4201 pci_dbg(dev, "%s bus mastering\n",
4202 enable ? "enabling" : "disabling");
4203 pci_write_config_word(dev, PCI_COMMAND, cmd);
4205 dev->is_busmaster = enable;
4209 * pcibios_setup - process "pci=" kernel boot arguments
4210 * @str: string used to pass in "pci=" kernel boot arguments
4212 * Process kernel boot arguments. This is the default implementation.
4213 * Architecture specific implementations can override this as necessary.
4215 char * __weak __init pcibios_setup(char *str)
4221 * pcibios_set_master - enable PCI bus-mastering for device dev
4222 * @dev: the PCI device to enable
4224 * Enables PCI bus-mastering for the device. This is the default
4225 * implementation. Architecture specific implementations can override
4226 * this if necessary.
4228 void __weak pcibios_set_master(struct pci_dev *dev)
4232 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4233 if (pci_is_pcie(dev))
4236 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4238 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4239 else if (lat > pcibios_max_latency)
4240 lat = pcibios_max_latency;
4244 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4248 * pci_set_master - enables bus-mastering for device dev
4249 * @dev: the PCI device to enable
4251 * Enables bus-mastering on the device and calls pcibios_set_master()
4252 * to do the needed arch specific settings.
4254 void pci_set_master(struct pci_dev *dev)
4256 __pci_set_master(dev, true);
4257 pcibios_set_master(dev);
4259 EXPORT_SYMBOL(pci_set_master);
4262 * pci_clear_master - disables bus-mastering for device dev
4263 * @dev: the PCI device to disable
4265 void pci_clear_master(struct pci_dev *dev)
4267 __pci_set_master(dev, false);
4269 EXPORT_SYMBOL(pci_clear_master);
4272 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4273 * @dev: the PCI device for which MWI is to be enabled
4275 * Helper function for pci_set_mwi.
4276 * Originally copied from drivers/net/acenic.c.
4277 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4279 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4281 int pci_set_cacheline_size(struct pci_dev *dev)
4285 if (!pci_cache_line_size)
4288 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4289 equal to or multiple of the right value. */
4290 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4291 if (cacheline_size >= pci_cache_line_size &&
4292 (cacheline_size % pci_cache_line_size) == 0)
4295 /* Write the correct value. */
4296 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4298 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4299 if (cacheline_size == pci_cache_line_size)
4302 pci_dbg(dev, "cache line size of %d is not supported\n",
4303 pci_cache_line_size << 2);
4307 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4310 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4311 * @dev: the PCI device for which MWI is enabled
4313 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4315 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4317 int pci_set_mwi(struct pci_dev *dev)
4319 #ifdef PCI_DISABLE_MWI
4325 rc = pci_set_cacheline_size(dev);
4329 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4330 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4331 pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4332 cmd |= PCI_COMMAND_INVALIDATE;
4333 pci_write_config_word(dev, PCI_COMMAND, cmd);
4338 EXPORT_SYMBOL(pci_set_mwi);
4341 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4342 * @dev: the PCI device for which MWI is enabled
4344 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4345 * Callers are not required to check the return value.
4347 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4349 int pci_try_set_mwi(struct pci_dev *dev)
4351 #ifdef PCI_DISABLE_MWI
4354 return pci_set_mwi(dev);
4357 EXPORT_SYMBOL(pci_try_set_mwi);
4360 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4361 * @dev: the PCI device to disable
4363 * Disables PCI Memory-Write-Invalidate transaction on the device
4365 void pci_clear_mwi(struct pci_dev *dev)
4367 #ifndef PCI_DISABLE_MWI
4370 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4371 if (cmd & PCI_COMMAND_INVALIDATE) {
4372 cmd &= ~PCI_COMMAND_INVALIDATE;
4373 pci_write_config_word(dev, PCI_COMMAND, cmd);
4377 EXPORT_SYMBOL(pci_clear_mwi);
4380 * pci_disable_parity - disable parity checking for device
4381 * @dev: the PCI device to operate on
4383 * Disable parity checking for device @dev
4385 void pci_disable_parity(struct pci_dev *dev)
4389 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4390 if (cmd & PCI_COMMAND_PARITY) {
4391 cmd &= ~PCI_COMMAND_PARITY;
4392 pci_write_config_word(dev, PCI_COMMAND, cmd);
4397 * pci_intx - enables/disables PCI INTx for device dev
4398 * @pdev: the PCI device to operate on
4399 * @enable: boolean: whether to enable or disable PCI INTx
4401 * Enables/disables PCI INTx for device @pdev
4403 void pci_intx(struct pci_dev *pdev, int enable)
4405 u16 pci_command, new;
4407 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4410 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4412 new = pci_command | PCI_COMMAND_INTX_DISABLE;
4414 if (new != pci_command) {
4415 struct pci_devres *dr;
4417 pci_write_config_word(pdev, PCI_COMMAND, new);
4419 dr = find_pci_dr(pdev);
4420 if (dr && !dr->restore_intx) {
4421 dr->restore_intx = 1;
4422 dr->orig_intx = !enable;
4426 EXPORT_SYMBOL_GPL(pci_intx);
4429 * pci_wait_for_pending_transaction - wait for pending transaction
4430 * @dev: the PCI device to operate on
4432 * Return 0 if transaction is pending 1 otherwise.
4434 int pci_wait_for_pending_transaction(struct pci_dev *dev)
4436 if (!pci_is_pcie(dev))
4439 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4440 PCI_EXP_DEVSTA_TRPND);
4442 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4445 * pcie_flr - initiate a PCIe function level reset
4446 * @dev: device to reset
4448 * Initiate a function level reset unconditionally on @dev without
4449 * checking any flags and DEVCAP
4451 int pcie_flr(struct pci_dev *dev)
4453 if (!pci_wait_for_pending_transaction(dev))
4454 pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4456 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4462 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4463 * 100ms, but may silently discard requests while the FLR is in
4464 * progress. Wait 100ms before trying to access the device.
4468 return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4470 EXPORT_SYMBOL_GPL(pcie_flr);
4473 * pcie_reset_flr - initiate a PCIe function level reset
4474 * @dev: device to reset
4475 * @probe: if true, return 0 if device can be reset this way
4477 * Initiate a function level reset on @dev.
4479 int pcie_reset_flr(struct pci_dev *dev, bool probe)
4481 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4484 if (!(dev->devcap & PCI_EXP_DEVCAP_FLR))
4490 return pcie_flr(dev);
4492 EXPORT_SYMBOL_GPL(pcie_reset_flr);
4494 static int pci_af_flr(struct pci_dev *dev, bool probe)
4499 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4503 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4506 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4507 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4514 * Wait for Transaction Pending bit to clear. A word-aligned test
4515 * is used, so we use the control offset rather than status and shift
4516 * the test bit to match.
4518 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4519 PCI_AF_STATUS_TP << 8))
4520 pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4522 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4528 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4529 * updated 27 July 2006; a device must complete an FLR within
4530 * 100ms, but may silently discard requests while the FLR is in
4531 * progress. Wait 100ms before trying to access the device.
4535 return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4539 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4540 * @dev: Device to reset.
4541 * @probe: if true, return 0 if the device can be reset this way.
4543 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4544 * unset, it will be reinitialized internally when going from PCI_D3hot to
4545 * PCI_D0. If that's the case and the device is not in a low-power state
4546 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4548 * NOTE: This causes the caller to sleep for twice the device power transition
4549 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4550 * by default (i.e. unless the @dev's d3hot_delay field has a different value).
4551 * Moreover, only devices in D0 can be reset by this function.
4553 static int pci_pm_reset(struct pci_dev *dev, bool probe)
4557 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4560 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4561 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4567 if (dev->current_state != PCI_D0)
4570 csr &= ~PCI_PM_CTRL_STATE_MASK;
4572 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4573 pci_dev_d3_sleep(dev);
4575 csr &= ~PCI_PM_CTRL_STATE_MASK;
4577 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4578 pci_dev_d3_sleep(dev);
4580 return pci_dev_wait(dev, "PM D3hot->D0", PCIE_RESET_READY_POLL_MS);
4584 * pcie_wait_for_link_status - Wait for link status change
4585 * @pdev: Device whose link to wait for.
4586 * @use_lt: Use the LT bit if TRUE, or the DLLLA bit if FALSE.
4587 * @active: Waiting for active or inactive?
4589 * Return 0 if successful, or -ETIMEDOUT if status has not changed within
4590 * PCIE_LINK_RETRAIN_TIMEOUT_MS milliseconds.
4592 static int pcie_wait_for_link_status(struct pci_dev *pdev,
4593 bool use_lt, bool active)
4595 u16 lnksta_mask, lnksta_match;
4596 unsigned long end_jiffies;
4599 lnksta_mask = use_lt ? PCI_EXP_LNKSTA_LT : PCI_EXP_LNKSTA_DLLLA;
4600 lnksta_match = active ? lnksta_mask : 0;
4602 end_jiffies = jiffies + msecs_to_jiffies(PCIE_LINK_RETRAIN_TIMEOUT_MS);
4604 pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnksta);
4605 if ((lnksta & lnksta_mask) == lnksta_match)
4608 } while (time_before(jiffies, end_jiffies));
4614 * pcie_retrain_link - Request a link retrain and wait for it to complete
4615 * @pdev: Device whose link to retrain.
4616 * @use_lt: Use the LT bit if TRUE, or the DLLLA bit if FALSE, for status.
4618 * Retrain completion status is retrieved from the Link Status Register
4619 * according to @use_lt. It is not verified whether the use of the DLLLA
4622 * Return 0 if successful, or -ETIMEDOUT if training has not completed
4623 * within PCIE_LINK_RETRAIN_TIMEOUT_MS milliseconds.
4625 int pcie_retrain_link(struct pci_dev *pdev, bool use_lt)
4630 * Ensure the updated LNKCTL parameters are used during link
4631 * training by checking that there is no ongoing link training that
4632 * may have started before link parameters were changed, so as to
4633 * avoid LTSSM race as recommended in Implementation Note at the end
4634 * of PCIe r6.1 sec 7.5.3.7.
4636 rc = pcie_wait_for_link_status(pdev, true, false);
4640 pcie_capability_set_word(pdev, PCI_EXP_LNKCTL, PCI_EXP_LNKCTL_RL);
4641 if (pdev->clear_retrain_link) {
4643 * Due to an erratum in some devices the Retrain Link bit
4644 * needs to be cleared again manually to allow the link
4645 * training to succeed.
4647 pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, PCI_EXP_LNKCTL_RL);
4650 return pcie_wait_for_link_status(pdev, use_lt, !use_lt);
4654 * pcie_wait_for_link_delay - Wait until link is active or inactive
4655 * @pdev: Bridge device
4656 * @active: waiting for active or inactive?
4657 * @delay: Delay to wait after link has become active (in ms)
4659 * Use this to wait till link becomes active or inactive.
4661 static bool pcie_wait_for_link_delay(struct pci_dev *pdev, bool active,
4667 * Some controllers might not implement link active reporting. In this
4668 * case, we wait for 1000 ms + any delay requested by the caller.
4670 if (!pdev->link_active_reporting) {
4671 msleep(PCIE_LINK_RETRAIN_TIMEOUT_MS + delay);
4676 * PCIe r4.0 sec 6.6.1, a component must enter LTSSM Detect within 20ms,
4677 * after which we should expect an link active if the reset was
4678 * successful. If so, software must wait a minimum 100ms before sending
4679 * configuration requests to devices downstream this port.
4681 * If the link fails to activate, either the device was physically
4682 * removed or the link is permanently failed.
4686 rc = pcie_wait_for_link_status(pdev, false, active);
4689 rc = pcie_failed_link_retrain(pdev);
4704 * pcie_wait_for_link - Wait until link is active or inactive
4705 * @pdev: Bridge device
4706 * @active: waiting for active or inactive?
4708 * Use this to wait till link becomes active or inactive.
4710 bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4712 return pcie_wait_for_link_delay(pdev, active, 100);
4716 * Find maximum D3cold delay required by all the devices on the bus. The
4717 * spec says 100 ms, but firmware can lower it and we allow drivers to
4718 * increase it as well.
4720 * Called with @pci_bus_sem locked for reading.
4722 static int pci_bus_max_d3cold_delay(const struct pci_bus *bus)
4724 const struct pci_dev *pdev;
4725 int min_delay = 100;
4728 list_for_each_entry(pdev, &bus->devices, bus_list) {
4729 if (pdev->d3cold_delay < min_delay)
4730 min_delay = pdev->d3cold_delay;
4731 if (pdev->d3cold_delay > max_delay)
4732 max_delay = pdev->d3cold_delay;
4735 return max(min_delay, max_delay);
4739 * pci_bridge_wait_for_secondary_bus - Wait for secondary bus to be accessible
4741 * @reset_type: reset type in human-readable form
4743 * Handle necessary delays before access to the devices on the secondary
4744 * side of the bridge are permitted after D3cold to D0 transition
4745 * or Conventional Reset.
4747 * For PCIe this means the delays in PCIe 5.0 section 6.6.1. For
4748 * conventional PCI it means Tpvrh + Trhfa specified in PCI 3.0 section
4751 * Return 0 on success or -ENOTTY if the first device on the secondary bus
4752 * failed to become accessible.
4754 int pci_bridge_wait_for_secondary_bus(struct pci_dev *dev, char *reset_type)
4756 struct pci_dev *child;
4759 if (pci_dev_is_disconnected(dev))
4762 if (!pci_is_bridge(dev))
4765 down_read(&pci_bus_sem);
4768 * We only deal with devices that are present currently on the bus.
4769 * For any hot-added devices the access delay is handled in pciehp
4770 * board_added(). In case of ACPI hotplug the firmware is expected
4771 * to configure the devices before OS is notified.
4773 if (!dev->subordinate || list_empty(&dev->subordinate->devices)) {
4774 up_read(&pci_bus_sem);
4778 /* Take d3cold_delay requirements into account */
4779 delay = pci_bus_max_d3cold_delay(dev->subordinate);
4781 up_read(&pci_bus_sem);
4785 child = list_first_entry(&dev->subordinate->devices, struct pci_dev,
4787 up_read(&pci_bus_sem);
4790 * Conventional PCI and PCI-X we need to wait Tpvrh + Trhfa before
4791 * accessing the device after reset (that is 1000 ms + 100 ms).
4793 if (!pci_is_pcie(dev)) {
4794 pci_dbg(dev, "waiting %d ms for secondary bus\n", 1000 + delay);
4795 msleep(1000 + delay);
4800 * For PCIe downstream and root ports that do not support speeds
4801 * greater than 5 GT/s need to wait minimum 100 ms. For higher
4802 * speeds (gen3) we need to wait first for the data link layer to
4805 * However, 100 ms is the minimum and the PCIe spec says the
4806 * software must allow at least 1s before it can determine that the
4807 * device that did not respond is a broken device. Also device can
4808 * take longer than that to respond if it indicates so through Request
4809 * Retry Status completions.
4811 * Therefore we wait for 100 ms and check for the device presence
4812 * until the timeout expires.
4814 if (!pcie_downstream_port(dev))
4817 if (pcie_get_speed_cap(dev) <= PCIE_SPEED_5_0GT) {
4820 pci_dbg(dev, "waiting %d ms for downstream link\n", delay);
4823 if (!pci_dev_wait(child, reset_type, PCI_RESET_WAIT - delay))
4827 * If the port supports active link reporting we now check
4828 * whether the link is active and if not bail out early with
4829 * the assumption that the device is not present anymore.
4831 if (!dev->link_active_reporting)
4834 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &status);
4835 if (!(status & PCI_EXP_LNKSTA_DLLLA))
4838 return pci_dev_wait(child, reset_type,
4839 PCIE_RESET_READY_POLL_MS - PCI_RESET_WAIT);
4842 pci_dbg(dev, "waiting %d ms for downstream link, after activation\n",
4844 if (!pcie_wait_for_link_delay(dev, true, delay)) {
4845 /* Did not train, no need to wait any further */
4846 pci_info(dev, "Data Link Layer Link Active not set in 1000 msec\n");
4850 return pci_dev_wait(child, reset_type,
4851 PCIE_RESET_READY_POLL_MS - delay);
4854 void pci_reset_secondary_bus(struct pci_dev *dev)
4858 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4859 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4860 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4863 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4864 * this to 2ms to ensure that we meet the minimum requirement.
4868 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4869 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4872 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4874 pci_reset_secondary_bus(dev);
4878 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4879 * @dev: Bridge device
4881 * Use the bridge control register to assert reset on the secondary bus.
4882 * Devices on the secondary bus are left in power-on state.
4884 int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4886 pcibios_reset_secondary_bus(dev);
4888 return pci_bridge_wait_for_secondary_bus(dev, "bus reset");
4890 EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4892 static int pci_parent_bus_reset(struct pci_dev *dev, bool probe)
4894 struct pci_dev *pdev;
4896 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4897 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4900 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4907 return pci_bridge_secondary_bus_reset(dev->bus->self);
4910 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, bool probe)
4914 if (!hotplug || !try_module_get(hotplug->owner))
4917 if (hotplug->ops->reset_slot)
4918 rc = hotplug->ops->reset_slot(hotplug, probe);
4920 module_put(hotplug->owner);
4925 static int pci_dev_reset_slot_function(struct pci_dev *dev, bool probe)
4927 if (dev->multifunction || dev->subordinate || !dev->slot ||
4928 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4931 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4934 static u16 cxl_port_dvsec(struct pci_dev *dev)
4936 return pci_find_dvsec_capability(dev, PCI_VENDOR_ID_CXL,
4937 PCI_DVSEC_CXL_PORT);
4940 static bool cxl_sbr_masked(struct pci_dev *dev)
4945 dvsec = cxl_port_dvsec(dev);
4949 rc = pci_read_config_word(dev, dvsec + PCI_DVSEC_CXL_PORT_CTL, ®);
4950 if (rc || PCI_POSSIBLE_ERROR(reg))
4954 * Per CXL spec r3.1, sec 8.1.5.2, when "Unmask SBR" is 0, the SBR
4955 * bit in Bridge Control has no effect. When 1, the Port generates
4956 * hot reset when the SBR bit is set to 1.
4958 if (reg & PCI_DVSEC_CXL_PORT_CTL_UNMASK_SBR)
4964 static int pci_reset_bus_function(struct pci_dev *dev, bool probe)
4966 struct pci_dev *bridge = pci_upstream_bridge(dev);
4970 * If "dev" is below a CXL port that has SBR control masked, SBR
4971 * won't do anything, so return error.
4973 if (bridge && cxl_sbr_masked(bridge)) {
4980 rc = pci_dev_reset_slot_function(dev, probe);
4983 return pci_parent_bus_reset(dev, probe);
4986 static int cxl_reset_bus_function(struct pci_dev *dev, bool probe)
4988 struct pci_dev *bridge;
4989 u16 dvsec, reg, val;
4992 bridge = pci_upstream_bridge(dev);
4996 dvsec = cxl_port_dvsec(bridge);
5003 rc = pci_read_config_word(bridge, dvsec + PCI_DVSEC_CXL_PORT_CTL, ®);
5007 if (reg & PCI_DVSEC_CXL_PORT_CTL_UNMASK_SBR) {
5010 val = reg | PCI_DVSEC_CXL_PORT_CTL_UNMASK_SBR;
5011 pci_write_config_word(bridge, dvsec + PCI_DVSEC_CXL_PORT_CTL,
5015 rc = pci_reset_bus_function(dev, probe);
5018 pci_write_config_word(bridge, dvsec + PCI_DVSEC_CXL_PORT_CTL,
5024 void pci_dev_lock(struct pci_dev *dev)
5026 /* block PM suspend, driver probe, etc. */
5027 device_lock(&dev->dev);
5028 pci_cfg_access_lock(dev);
5030 EXPORT_SYMBOL_GPL(pci_dev_lock);
5032 /* Return 1 on successful lock, 0 on contention */
5033 int pci_dev_trylock(struct pci_dev *dev)
5035 if (device_trylock(&dev->dev)) {
5036 if (pci_cfg_access_trylock(dev))
5038 device_unlock(&dev->dev);
5043 EXPORT_SYMBOL_GPL(pci_dev_trylock);
5045 void pci_dev_unlock(struct pci_dev *dev)
5047 pci_cfg_access_unlock(dev);
5048 device_unlock(&dev->dev);
5050 EXPORT_SYMBOL_GPL(pci_dev_unlock);
5052 static void pci_dev_save_and_disable(struct pci_dev *dev)
5054 const struct pci_error_handlers *err_handler =
5055 dev->driver ? dev->driver->err_handler : NULL;
5058 * dev->driver->err_handler->reset_prepare() is protected against
5059 * races with ->remove() by the device lock, which must be held by
5062 if (err_handler && err_handler->reset_prepare)
5063 err_handler->reset_prepare(dev);
5066 * Wake-up device prior to save. PM registers default to D0 after
5067 * reset and a simple register restore doesn't reliably return
5068 * to a non-D0 state anyway.
5070 pci_set_power_state(dev, PCI_D0);
5072 pci_save_state(dev);
5074 * Disable the device by clearing the Command register, except for
5075 * INTx-disable which is set. This not only disables MMIO and I/O port
5076 * BARs, but also prevents the device from being Bus Master, preventing
5077 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
5078 * compliant devices, INTx-disable prevents legacy interrupts.
5080 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
5083 static void pci_dev_restore(struct pci_dev *dev)
5085 const struct pci_error_handlers *err_handler =
5086 dev->driver ? dev->driver->err_handler : NULL;
5088 pci_restore_state(dev);
5091 * dev->driver->err_handler->reset_done() is protected against
5092 * races with ->remove() by the device lock, which must be held by
5095 if (err_handler && err_handler->reset_done)
5096 err_handler->reset_done(dev);
5099 /* dev->reset_methods[] is a 0-terminated list of indices into this array */
5100 static const struct pci_reset_fn_method pci_reset_fn_methods[] = {
5102 { pci_dev_specific_reset, .name = "device_specific" },
5103 { pci_dev_acpi_reset, .name = "acpi" },
5104 { pcie_reset_flr, .name = "flr" },
5105 { pci_af_flr, .name = "af_flr" },
5106 { pci_pm_reset, .name = "pm" },
5107 { pci_reset_bus_function, .name = "bus" },
5108 { cxl_reset_bus_function, .name = "cxl_bus" },
5111 static ssize_t reset_method_show(struct device *dev,
5112 struct device_attribute *attr, char *buf)
5114 struct pci_dev *pdev = to_pci_dev(dev);
5118 for (i = 0; i < PCI_NUM_RESET_METHODS; i++) {
5119 m = pdev->reset_methods[i];
5123 len += sysfs_emit_at(buf, len, "%s%s", len ? " " : "",
5124 pci_reset_fn_methods[m].name);
5128 len += sysfs_emit_at(buf, len, "\n");
5133 static int reset_method_lookup(const char *name)
5137 for (m = 1; m < PCI_NUM_RESET_METHODS; m++) {
5138 if (sysfs_streq(name, pci_reset_fn_methods[m].name))
5142 return 0; /* not found */
5145 static ssize_t reset_method_store(struct device *dev,
5146 struct device_attribute *attr,
5147 const char *buf, size_t count)
5149 struct pci_dev *pdev = to_pci_dev(dev);
5150 char *options, *name;
5152 u8 reset_methods[PCI_NUM_RESET_METHODS] = { 0 };
5154 if (sysfs_streq(buf, "")) {
5155 pdev->reset_methods[0] = 0;
5156 pci_warn(pdev, "All device reset methods disabled by user");
5160 if (sysfs_streq(buf, "default")) {
5161 pci_init_reset_methods(pdev);
5165 options = kstrndup(buf, count, GFP_KERNEL);
5170 while ((name = strsep(&options, " ")) != NULL) {
5171 if (sysfs_streq(name, ""))
5176 m = reset_method_lookup(name);
5178 pci_err(pdev, "Invalid reset method '%s'", name);
5182 if (pci_reset_fn_methods[m].reset_fn(pdev, PCI_RESET_PROBE)) {
5183 pci_err(pdev, "Unsupported reset method '%s'", name);
5187 if (n == PCI_NUM_RESET_METHODS - 1) {
5188 pci_err(pdev, "Too many reset methods\n");
5192 reset_methods[n++] = m;
5195 reset_methods[n] = 0;
5197 /* Warn if dev-specific supported but not highest priority */
5198 if (pci_reset_fn_methods[1].reset_fn(pdev, PCI_RESET_PROBE) == 0 &&
5199 reset_methods[0] != 1)
5200 pci_warn(pdev, "Device-specific reset disabled/de-prioritized by user");
5201 memcpy(pdev->reset_methods, reset_methods, sizeof(pdev->reset_methods));
5206 /* Leave previous methods unchanged */
5210 static DEVICE_ATTR_RW(reset_method);
5212 static struct attribute *pci_dev_reset_method_attrs[] = {
5213 &dev_attr_reset_method.attr,
5217 static umode_t pci_dev_reset_method_attr_is_visible(struct kobject *kobj,
5218 struct attribute *a, int n)
5220 struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj));
5222 if (!pci_reset_supported(pdev))
5228 const struct attribute_group pci_dev_reset_method_attr_group = {
5229 .attrs = pci_dev_reset_method_attrs,
5230 .is_visible = pci_dev_reset_method_attr_is_visible,
5234 * __pci_reset_function_locked - reset a PCI device function while holding
5235 * the @dev mutex lock.
5236 * @dev: PCI device to reset
5238 * Some devices allow an individual function to be reset without affecting
5239 * other functions in the same device. The PCI device must be responsive
5240 * to PCI config space in order to use this function.
5242 * The device function is presumed to be unused and the caller is holding
5243 * the device mutex lock when this function is called.
5245 * Resetting the device will make the contents of PCI configuration space
5246 * random, so any caller of this must be prepared to reinitialise the
5247 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
5250 * Returns 0 if the device function was successfully reset or negative if the
5251 * device doesn't support resetting a single function.
5253 int __pci_reset_function_locked(struct pci_dev *dev)
5260 * A reset method returns -ENOTTY if it doesn't support this device and
5261 * we should try the next method.
5263 * If it returns 0 (success), we're finished. If it returns any other
5264 * error, we're also finished: this indicates that further reset
5265 * mechanisms might be broken on the device.
5267 for (i = 0; i < PCI_NUM_RESET_METHODS; i++) {
5268 m = dev->reset_methods[i];
5272 rc = pci_reset_fn_methods[m].reset_fn(dev, PCI_RESET_DO_RESET);
5281 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
5284 * pci_init_reset_methods - check whether device can be safely reset
5285 * and store supported reset mechanisms.
5286 * @dev: PCI device to check for reset mechanisms
5288 * Some devices allow an individual function to be reset without affecting
5289 * other functions in the same device. The PCI device must be in D0-D3hot
5292 * Stores reset mechanisms supported by device in reset_methods byte array
5293 * which is a member of struct pci_dev.
5295 void pci_init_reset_methods(struct pci_dev *dev)
5299 BUILD_BUG_ON(ARRAY_SIZE(pci_reset_fn_methods) != PCI_NUM_RESET_METHODS);
5304 for (m = 1; m < PCI_NUM_RESET_METHODS; m++) {
5305 rc = pci_reset_fn_methods[m].reset_fn(dev, PCI_RESET_PROBE);
5307 dev->reset_methods[i++] = m;
5308 else if (rc != -ENOTTY)
5312 dev->reset_methods[i] = 0;
5316 * pci_reset_function - quiesce and reset a PCI device function
5317 * @dev: PCI device to reset
5319 * Some devices allow an individual function to be reset without affecting
5320 * other functions in the same device. The PCI device must be responsive
5321 * to PCI config space in order to use this function.
5323 * This function does not just reset the PCI portion of a device, but
5324 * clears all the state associated with the device. This function differs
5325 * from __pci_reset_function_locked() in that it saves and restores device state
5326 * over the reset and takes the PCI device lock.
5328 * Returns 0 if the device function was successfully reset or negative if the
5329 * device doesn't support resetting a single function.
5331 int pci_reset_function(struct pci_dev *dev)
5333 struct pci_dev *bridge;
5336 if (!pci_reset_supported(dev))
5340 * If there's no upstream bridge, no locking is needed since there is
5341 * no upstream bridge configuration to hold consistent.
5343 bridge = pci_upstream_bridge(dev);
5345 pci_dev_lock(bridge);
5348 pci_dev_save_and_disable(dev);
5350 rc = __pci_reset_function_locked(dev);
5352 pci_dev_restore(dev);
5353 pci_dev_unlock(dev);
5356 pci_dev_unlock(bridge);
5360 EXPORT_SYMBOL_GPL(pci_reset_function);
5363 * pci_reset_function_locked - quiesce and reset a PCI device function
5364 * @dev: PCI device to reset
5366 * Some devices allow an individual function to be reset without affecting
5367 * other functions in the same device. The PCI device must be responsive
5368 * to PCI config space in order to use this function.
5370 * This function does not just reset the PCI portion of a device, but
5371 * clears all the state associated with the device. This function differs
5372 * from __pci_reset_function_locked() in that it saves and restores device state
5373 * over the reset. It also differs from pci_reset_function() in that it
5374 * requires the PCI device lock to be held.
5376 * Returns 0 if the device function was successfully reset or negative if the
5377 * device doesn't support resetting a single function.
5379 int pci_reset_function_locked(struct pci_dev *dev)
5383 if (!pci_reset_supported(dev))
5386 pci_dev_save_and_disable(dev);
5388 rc = __pci_reset_function_locked(dev);
5390 pci_dev_restore(dev);
5394 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
5397 * pci_try_reset_function - quiesce and reset a PCI device function
5398 * @dev: PCI device to reset
5400 * Same as above, except return -EAGAIN if unable to lock device.
5402 int pci_try_reset_function(struct pci_dev *dev)
5406 if (!pci_reset_supported(dev))
5409 if (!pci_dev_trylock(dev))
5412 pci_dev_save_and_disable(dev);
5413 rc = __pci_reset_function_locked(dev);
5414 pci_dev_restore(dev);
5415 pci_dev_unlock(dev);
5419 EXPORT_SYMBOL_GPL(pci_try_reset_function);
5421 /* Do any devices on or below this bus prevent a bus reset? */
5422 static bool pci_bus_resettable(struct pci_bus *bus)
5424 struct pci_dev *dev;
5427 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5430 list_for_each_entry(dev, &bus->devices, bus_list) {
5431 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5432 (dev->subordinate && !pci_bus_resettable(dev->subordinate)))
5439 /* Lock devices from the top of the tree down */
5440 static void pci_bus_lock(struct pci_bus *bus)
5442 struct pci_dev *dev;
5444 list_for_each_entry(dev, &bus->devices, bus_list) {
5446 if (dev->subordinate)
5447 pci_bus_lock(dev->subordinate);
5451 /* Unlock devices from the bottom of the tree up */
5452 static void pci_bus_unlock(struct pci_bus *bus)
5454 struct pci_dev *dev;
5456 list_for_each_entry(dev, &bus->devices, bus_list) {
5457 if (dev->subordinate)
5458 pci_bus_unlock(dev->subordinate);
5459 pci_dev_unlock(dev);
5463 /* Return 1 on successful lock, 0 on contention */
5464 static int pci_bus_trylock(struct pci_bus *bus)
5466 struct pci_dev *dev;
5468 list_for_each_entry(dev, &bus->devices, bus_list) {
5469 if (!pci_dev_trylock(dev))
5471 if (dev->subordinate) {
5472 if (!pci_bus_trylock(dev->subordinate)) {
5473 pci_dev_unlock(dev);
5481 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
5482 if (dev->subordinate)
5483 pci_bus_unlock(dev->subordinate);
5484 pci_dev_unlock(dev);
5489 /* Do any devices on or below this slot prevent a bus reset? */
5490 static bool pci_slot_resettable(struct pci_slot *slot)
5492 struct pci_dev *dev;
5494 if (slot->bus->self &&
5495 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
5498 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5499 if (!dev->slot || dev->slot != slot)
5501 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
5502 (dev->subordinate && !pci_bus_resettable(dev->subordinate)))
5509 /* Lock devices from the top of the tree down */
5510 static void pci_slot_lock(struct pci_slot *slot)
5512 struct pci_dev *dev;
5514 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5515 if (!dev->slot || dev->slot != slot)
5518 if (dev->subordinate)
5519 pci_bus_lock(dev->subordinate);
5523 /* Unlock devices from the bottom of the tree up */
5524 static void pci_slot_unlock(struct pci_slot *slot)
5526 struct pci_dev *dev;
5528 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5529 if (!dev->slot || dev->slot != slot)
5531 if (dev->subordinate)
5532 pci_bus_unlock(dev->subordinate);
5533 pci_dev_unlock(dev);
5537 /* Return 1 on successful lock, 0 on contention */
5538 static int pci_slot_trylock(struct pci_slot *slot)
5540 struct pci_dev *dev;
5542 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5543 if (!dev->slot || dev->slot != slot)
5545 if (!pci_dev_trylock(dev))
5547 if (dev->subordinate) {
5548 if (!pci_bus_trylock(dev->subordinate)) {
5549 pci_dev_unlock(dev);
5557 list_for_each_entry_continue_reverse(dev,
5558 &slot->bus->devices, bus_list) {
5559 if (!dev->slot || dev->slot != slot)
5561 if (dev->subordinate)
5562 pci_bus_unlock(dev->subordinate);
5563 pci_dev_unlock(dev);
5569 * Save and disable devices from the top of the tree down while holding
5570 * the @dev mutex lock for the entire tree.
5572 static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5574 struct pci_dev *dev;
5576 list_for_each_entry(dev, &bus->devices, bus_list) {
5577 pci_dev_save_and_disable(dev);
5578 if (dev->subordinate)
5579 pci_bus_save_and_disable_locked(dev->subordinate);
5584 * Restore devices from top of the tree down while holding @dev mutex lock
5585 * for the entire tree. Parent bridges need to be restored before we can
5586 * get to subordinate devices.
5588 static void pci_bus_restore_locked(struct pci_bus *bus)
5590 struct pci_dev *dev;
5592 list_for_each_entry(dev, &bus->devices, bus_list) {
5593 pci_dev_restore(dev);
5594 if (dev->subordinate)
5595 pci_bus_restore_locked(dev->subordinate);
5600 * Save and disable devices from the top of the tree down while holding
5601 * the @dev mutex lock for the entire tree.
5603 static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5605 struct pci_dev *dev;
5607 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5608 if (!dev->slot || dev->slot != slot)
5610 pci_dev_save_and_disable(dev);
5611 if (dev->subordinate)
5612 pci_bus_save_and_disable_locked(dev->subordinate);
5617 * Restore devices from top of the tree down while holding @dev mutex lock
5618 * for the entire tree. Parent bridges need to be restored before we can
5619 * get to subordinate devices.
5621 static void pci_slot_restore_locked(struct pci_slot *slot)
5623 struct pci_dev *dev;
5625 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5626 if (!dev->slot || dev->slot != slot)
5628 pci_dev_restore(dev);
5629 if (dev->subordinate)
5630 pci_bus_restore_locked(dev->subordinate);
5634 static int pci_slot_reset(struct pci_slot *slot, bool probe)
5638 if (!slot || !pci_slot_resettable(slot))
5642 pci_slot_lock(slot);
5646 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5649 pci_slot_unlock(slot);
5655 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5656 * @slot: PCI slot to probe
5658 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5660 int pci_probe_reset_slot(struct pci_slot *slot)
5662 return pci_slot_reset(slot, PCI_RESET_PROBE);
5664 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5667 * __pci_reset_slot - Try to reset a PCI slot
5668 * @slot: PCI slot to reset
5670 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5671 * independent of other slots. For instance, some slots may support slot power
5672 * control. In the case of a 1:1 bus to slot architecture, this function may
5673 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5674 * Generally a slot reset should be attempted before a bus reset. All of the
5675 * function of the slot and any subordinate buses behind the slot are reset
5676 * through this function. PCI config space of all devices in the slot and
5677 * behind the slot is saved before and restored after reset.
5679 * Same as above except return -EAGAIN if the slot cannot be locked
5681 static int __pci_reset_slot(struct pci_slot *slot)
5685 rc = pci_slot_reset(slot, PCI_RESET_PROBE);
5689 if (pci_slot_trylock(slot)) {
5690 pci_slot_save_and_disable_locked(slot);
5692 rc = pci_reset_hotplug_slot(slot->hotplug, PCI_RESET_DO_RESET);
5693 pci_slot_restore_locked(slot);
5694 pci_slot_unlock(slot);
5701 static int pci_bus_reset(struct pci_bus *bus, bool probe)
5705 if (!bus->self || !pci_bus_resettable(bus))
5715 ret = pci_bridge_secondary_bus_reset(bus->self);
5717 pci_bus_unlock(bus);
5723 * pci_bus_error_reset - reset the bridge's subordinate bus
5724 * @bridge: The parent device that connects to the bus to reset
5726 * This function will first try to reset the slots on this bus if the method is
5727 * available. If slot reset fails or is not available, this will fall back to a
5728 * secondary bus reset.
5730 int pci_bus_error_reset(struct pci_dev *bridge)
5732 struct pci_bus *bus = bridge->subordinate;
5733 struct pci_slot *slot;
5738 mutex_lock(&pci_slot_mutex);
5739 if (list_empty(&bus->slots))
5742 list_for_each_entry(slot, &bus->slots, list)
5743 if (pci_probe_reset_slot(slot))
5746 list_for_each_entry(slot, &bus->slots, list)
5747 if (pci_slot_reset(slot, PCI_RESET_DO_RESET))
5750 mutex_unlock(&pci_slot_mutex);
5753 mutex_unlock(&pci_slot_mutex);
5754 return pci_bus_reset(bridge->subordinate, PCI_RESET_DO_RESET);
5758 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5759 * @bus: PCI bus to probe
5761 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5763 int pci_probe_reset_bus(struct pci_bus *bus)
5765 return pci_bus_reset(bus, PCI_RESET_PROBE);
5767 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5770 * __pci_reset_bus - Try to reset a PCI bus
5771 * @bus: top level PCI bus to reset
5773 * Same as above except return -EAGAIN if the bus cannot be locked
5775 static int __pci_reset_bus(struct pci_bus *bus)
5779 rc = pci_bus_reset(bus, PCI_RESET_PROBE);
5783 if (pci_bus_trylock(bus)) {
5784 pci_bus_save_and_disable_locked(bus);
5786 rc = pci_bridge_secondary_bus_reset(bus->self);
5787 pci_bus_restore_locked(bus);
5788 pci_bus_unlock(bus);
5796 * pci_reset_bus - Try to reset a PCI bus
5797 * @pdev: top level PCI device to reset via slot/bus
5799 * Same as above except return -EAGAIN if the bus cannot be locked
5801 int pci_reset_bus(struct pci_dev *pdev)
5803 return (!pci_probe_reset_slot(pdev->slot)) ?
5804 __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5806 EXPORT_SYMBOL_GPL(pci_reset_bus);
5809 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5810 * @dev: PCI device to query
5812 * Returns mmrbc: maximum designed memory read count in bytes or
5813 * appropriate error value.
5815 int pcix_get_max_mmrbc(struct pci_dev *dev)
5820 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5824 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5827 return 512 << FIELD_GET(PCI_X_STATUS_MAX_READ, stat);
5829 EXPORT_SYMBOL(pcix_get_max_mmrbc);
5832 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5833 * @dev: PCI device to query
5835 * Returns mmrbc: maximum memory read count in bytes or appropriate error
5838 int pcix_get_mmrbc(struct pci_dev *dev)
5843 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5847 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5850 return 512 << FIELD_GET(PCI_X_CMD_MAX_READ, cmd);
5852 EXPORT_SYMBOL(pcix_get_mmrbc);
5855 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5856 * @dev: PCI device to query
5857 * @mmrbc: maximum memory read count in bytes
5858 * valid values are 512, 1024, 2048, 4096
5860 * If possible sets maximum memory read byte count, some bridges have errata
5861 * that prevent this.
5863 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5869 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5872 v = ffs(mmrbc) - 10;
5874 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5878 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5881 if (v > FIELD_GET(PCI_X_STATUS_MAX_READ, stat))
5884 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5887 o = FIELD_GET(PCI_X_CMD_MAX_READ, cmd);
5889 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5892 cmd &= ~PCI_X_CMD_MAX_READ;
5893 cmd |= FIELD_PREP(PCI_X_CMD_MAX_READ, v);
5894 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5899 EXPORT_SYMBOL(pcix_set_mmrbc);
5902 * pcie_get_readrq - get PCI Express read request size
5903 * @dev: PCI device to query
5905 * Returns maximum memory read request in bytes or appropriate error value.
5907 int pcie_get_readrq(struct pci_dev *dev)
5911 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5913 return 128 << FIELD_GET(PCI_EXP_DEVCTL_READRQ, ctl);
5915 EXPORT_SYMBOL(pcie_get_readrq);
5918 * pcie_set_readrq - set PCI Express maximum memory read request
5919 * @dev: PCI device to query
5920 * @rq: maximum memory read count in bytes
5921 * valid values are 128, 256, 512, 1024, 2048, 4096
5923 * If possible sets maximum memory read request in bytes
5925 int pcie_set_readrq(struct pci_dev *dev, int rq)
5929 struct pci_host_bridge *bridge = pci_find_host_bridge(dev->bus);
5931 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5935 * If using the "performance" PCIe config, we clamp the read rq
5936 * size to the max packet size to keep the host bridge from
5937 * generating requests larger than we can cope with.
5939 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5940 int mps = pcie_get_mps(dev);
5946 v = FIELD_PREP(PCI_EXP_DEVCTL_READRQ, ffs(rq) - 8);
5948 if (bridge->no_inc_mrrs) {
5949 int max_mrrs = pcie_get_readrq(dev);
5951 if (rq > max_mrrs) {
5952 pci_info(dev, "can't set Max_Read_Request_Size to %d; max is %d\n", rq, max_mrrs);
5957 ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5958 PCI_EXP_DEVCTL_READRQ, v);
5960 return pcibios_err_to_errno(ret);
5962 EXPORT_SYMBOL(pcie_set_readrq);
5965 * pcie_get_mps - get PCI Express maximum payload size
5966 * @dev: PCI device to query
5968 * Returns maximum payload size in bytes
5970 int pcie_get_mps(struct pci_dev *dev)
5974 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5976 return 128 << FIELD_GET(PCI_EXP_DEVCTL_PAYLOAD, ctl);
5978 EXPORT_SYMBOL(pcie_get_mps);
5981 * pcie_set_mps - set PCI Express maximum payload size
5982 * @dev: PCI device to query
5983 * @mps: maximum payload size in bytes
5984 * valid values are 128, 256, 512, 1024, 2048, 4096
5986 * If possible sets maximum payload size
5988 int pcie_set_mps(struct pci_dev *dev, int mps)
5993 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5997 if (v > dev->pcie_mpss)
5999 v = FIELD_PREP(PCI_EXP_DEVCTL_PAYLOAD, v);
6001 ret = pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
6002 PCI_EXP_DEVCTL_PAYLOAD, v);
6004 return pcibios_err_to_errno(ret);
6006 EXPORT_SYMBOL(pcie_set_mps);
6008 static enum pci_bus_speed to_pcie_link_speed(u16 lnksta)
6010 return pcie_link_speed[FIELD_GET(PCI_EXP_LNKSTA_CLS, lnksta)];
6013 int pcie_link_speed_mbps(struct pci_dev *pdev)
6018 err = pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnksta);
6022 switch (to_pcie_link_speed(lnksta)) {
6023 case PCIE_SPEED_2_5GT:
6025 case PCIE_SPEED_5_0GT:
6027 case PCIE_SPEED_8_0GT:
6029 case PCIE_SPEED_16_0GT:
6031 case PCIE_SPEED_32_0GT:
6033 case PCIE_SPEED_64_0GT:
6041 EXPORT_SYMBOL(pcie_link_speed_mbps);
6044 * pcie_bandwidth_available - determine minimum link settings of a PCIe
6045 * device and its bandwidth limitation
6046 * @dev: PCI device to query
6047 * @limiting_dev: storage for device causing the bandwidth limitation
6048 * @speed: storage for speed of limiting device
6049 * @width: storage for width of limiting device
6051 * Walk up the PCI device chain and find the point where the minimum
6052 * bandwidth is available. Return the bandwidth available there and (if
6053 * limiting_dev, speed, and width pointers are supplied) information about
6054 * that point. The bandwidth returned is in Mb/s, i.e., megabits/second of
6057 u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
6058 enum pci_bus_speed *speed,
6059 enum pcie_link_width *width)
6062 enum pci_bus_speed next_speed;
6063 enum pcie_link_width next_width;
6067 *speed = PCI_SPEED_UNKNOWN;
6069 *width = PCIE_LNK_WIDTH_UNKNOWN;
6074 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
6076 next_speed = to_pcie_link_speed(lnksta);
6077 next_width = FIELD_GET(PCI_EXP_LNKSTA_NLW, lnksta);
6079 next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
6081 /* Check if current device limits the total bandwidth */
6082 if (!bw || next_bw <= bw) {
6086 *limiting_dev = dev;
6088 *speed = next_speed;
6090 *width = next_width;
6093 dev = pci_upstream_bridge(dev);
6098 EXPORT_SYMBOL(pcie_bandwidth_available);
6101 * pcie_get_speed_cap - query for the PCI device's link speed capability
6102 * @dev: PCI device to query
6104 * Query the PCI device speed capability. Return the maximum link speed
6105 * supported by the device.
6107 enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
6109 u32 lnkcap2, lnkcap;
6112 * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18. The
6113 * implementation note there recommends using the Supported Link
6114 * Speeds Vector in Link Capabilities 2 when supported.
6116 * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
6117 * should use the Supported Link Speeds field in Link Capabilities,
6118 * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
6120 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
6122 /* PCIe r3.0-compliant */
6124 return PCIE_LNKCAP2_SLS2SPEED(lnkcap2);
6126 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
6127 if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
6128 return PCIE_SPEED_5_0GT;
6129 else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
6130 return PCIE_SPEED_2_5GT;
6132 return PCI_SPEED_UNKNOWN;
6134 EXPORT_SYMBOL(pcie_get_speed_cap);
6137 * pcie_get_width_cap - query for the PCI device's link width capability
6138 * @dev: PCI device to query
6140 * Query the PCI device width capability. Return the maximum link width
6141 * supported by the device.
6143 enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
6147 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
6149 return FIELD_GET(PCI_EXP_LNKCAP_MLW, lnkcap);
6151 return PCIE_LNK_WIDTH_UNKNOWN;
6153 EXPORT_SYMBOL(pcie_get_width_cap);
6156 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
6158 * @speed: storage for link speed
6159 * @width: storage for link width
6161 * Calculate a PCI device's link bandwidth by querying for its link speed
6162 * and width, multiplying them, and applying encoding overhead. The result
6163 * is in Mb/s, i.e., megabits/second of raw bandwidth.
6165 static u32 pcie_bandwidth_capable(struct pci_dev *dev,
6166 enum pci_bus_speed *speed,
6167 enum pcie_link_width *width)
6169 *speed = pcie_get_speed_cap(dev);
6170 *width = pcie_get_width_cap(dev);
6172 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
6175 return *width * PCIE_SPEED2MBS_ENC(*speed);
6179 * __pcie_print_link_status - Report the PCI device's link speed and width
6180 * @dev: PCI device to query
6181 * @verbose: Print info even when enough bandwidth is available
6183 * If the available bandwidth at the device is less than the device is
6184 * capable of, report the device's maximum possible bandwidth and the
6185 * upstream link that limits its performance. If @verbose, always print
6186 * the available bandwidth, even if the device isn't constrained.
6188 void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
6190 enum pcie_link_width width, width_cap;
6191 enum pci_bus_speed speed, speed_cap;
6192 struct pci_dev *limiting_dev = NULL;
6193 u32 bw_avail, bw_cap;
6195 bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
6196 bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
6198 if (bw_avail >= bw_cap && verbose)
6199 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
6200 bw_cap / 1000, bw_cap % 1000,
6201 pci_speed_string(speed_cap), width_cap);
6202 else if (bw_avail < bw_cap)
6203 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",
6204 bw_avail / 1000, bw_avail % 1000,
6205 pci_speed_string(speed), width,
6206 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
6207 bw_cap / 1000, bw_cap % 1000,
6208 pci_speed_string(speed_cap), width_cap);
6212 * pcie_print_link_status - Report the PCI device's link speed and width
6213 * @dev: PCI device to query
6215 * Report the available bandwidth at the device.
6217 void pcie_print_link_status(struct pci_dev *dev)
6219 __pcie_print_link_status(dev, true);
6221 EXPORT_SYMBOL(pcie_print_link_status);
6224 * pci_select_bars - Make BAR mask from the type of resource
6225 * @dev: the PCI device for which BAR mask is made
6226 * @flags: resource type mask to be selected
6228 * This helper routine makes bar mask from the type of resource.
6230 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
6233 for (i = 0; i < PCI_NUM_RESOURCES; i++)
6234 if (pci_resource_flags(dev, i) & flags)
6238 EXPORT_SYMBOL(pci_select_bars);
6240 /* Some architectures require additional programming to enable VGA */
6241 static arch_set_vga_state_t arch_set_vga_state;
6243 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
6245 arch_set_vga_state = func; /* NULL disables */
6248 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
6249 unsigned int command_bits, u32 flags)
6251 if (arch_set_vga_state)
6252 return arch_set_vga_state(dev, decode, command_bits,
6258 * pci_set_vga_state - set VGA decode state on device and parents if requested
6259 * @dev: the PCI device
6260 * @decode: true = enable decoding, false = disable decoding
6261 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
6262 * @flags: traverse ancestors and change bridges
6263 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
6265 int pci_set_vga_state(struct pci_dev *dev, bool decode,
6266 unsigned int command_bits, u32 flags)
6268 struct pci_bus *bus;
6269 struct pci_dev *bridge;
6273 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
6275 /* ARCH specific VGA enables */
6276 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
6280 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
6281 pci_read_config_word(dev, PCI_COMMAND, &cmd);
6283 cmd |= command_bits;
6285 cmd &= ~command_bits;
6286 pci_write_config_word(dev, PCI_COMMAND, cmd);
6289 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
6296 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
6299 cmd |= PCI_BRIDGE_CTL_VGA;
6301 cmd &= ~PCI_BRIDGE_CTL_VGA;
6302 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
6311 bool pci_pr3_present(struct pci_dev *pdev)
6313 struct acpi_device *adev;
6318 adev = ACPI_COMPANION(&pdev->dev);
6322 return adev->power.flags.power_resources &&
6323 acpi_has_method(adev->handle, "_PR3");
6325 EXPORT_SYMBOL_GPL(pci_pr3_present);
6329 * pci_add_dma_alias - Add a DMA devfn alias for a device
6330 * @dev: the PCI device for which alias is added
6331 * @devfn_from: alias slot and function
6332 * @nr_devfns: number of subsequent devfns to alias
6334 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
6335 * which is used to program permissible bus-devfn source addresses for DMA
6336 * requests in an IOMMU. These aliases factor into IOMMU group creation
6337 * and are useful for devices generating DMA requests beyond or different
6338 * from their logical bus-devfn. Examples include device quirks where the
6339 * device simply uses the wrong devfn, as well as non-transparent bridges
6340 * where the alias may be a proxy for devices in another domain.
6342 * IOMMU group creation is performed during device discovery or addition,
6343 * prior to any potential DMA mapping and therefore prior to driver probing
6344 * (especially for userspace assigned devices where IOMMU group definition
6345 * cannot be left as a userspace activity). DMA aliases should therefore
6346 * be configured via quirks, such as the PCI fixup header quirk.
6348 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn_from,
6349 unsigned int nr_devfns)
6353 nr_devfns = min(nr_devfns, (unsigned int)MAX_NR_DEVFNS - devfn_from);
6354 devfn_to = devfn_from + nr_devfns - 1;
6356 if (!dev->dma_alias_mask)
6357 dev->dma_alias_mask = bitmap_zalloc(MAX_NR_DEVFNS, GFP_KERNEL);
6358 if (!dev->dma_alias_mask) {
6359 pci_warn(dev, "Unable to allocate DMA alias mask\n");
6363 bitmap_set(dev->dma_alias_mask, devfn_from, nr_devfns);
6366 pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
6367 PCI_SLOT(devfn_from), PCI_FUNC(devfn_from));
6368 else if (nr_devfns > 1)
6369 pci_info(dev, "Enabling fixed DMA alias for devfn range from %02x.%d to %02x.%d\n",
6370 PCI_SLOT(devfn_from), PCI_FUNC(devfn_from),
6371 PCI_SLOT(devfn_to), PCI_FUNC(devfn_to));
6374 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
6376 return (dev1->dma_alias_mask &&
6377 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
6378 (dev2->dma_alias_mask &&
6379 test_bit(dev1->devfn, dev2->dma_alias_mask)) ||
6380 pci_real_dma_dev(dev1) == dev2 ||
6381 pci_real_dma_dev(dev2) == dev1;
6384 bool pci_device_is_present(struct pci_dev *pdev)
6388 /* Check PF if pdev is a VF, since VF Vendor/Device IDs are 0xffff */
6389 pdev = pci_physfn(pdev);
6390 if (pci_dev_is_disconnected(pdev))
6392 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
6394 EXPORT_SYMBOL_GPL(pci_device_is_present);
6396 void pci_ignore_hotplug(struct pci_dev *dev)
6398 struct pci_dev *bridge = dev->bus->self;
6400 dev->ignore_hotplug = 1;
6401 /* Propagate the "ignore hotplug" setting to the parent bridge. */
6403 bridge->ignore_hotplug = 1;
6405 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
6408 * pci_real_dma_dev - Get PCI DMA device for PCI device
6409 * @dev: the PCI device that may have a PCI DMA alias
6411 * Permits the platform to provide architecture-specific functionality to
6412 * devices needing to alias DMA to another PCI device on another PCI bus. If
6413 * the PCI device is on the same bus, it is recommended to use
6414 * pci_add_dma_alias(). This is the default implementation. Architecture
6415 * implementations can override this.
6417 struct pci_dev __weak *pci_real_dma_dev(struct pci_dev *dev)
6422 resource_size_t __weak pcibios_default_alignment(void)
6428 * Arches that don't want to expose struct resource to userland as-is in
6429 * sysfs and /proc can implement their own pci_resource_to_user().
6431 void __weak pci_resource_to_user(const struct pci_dev *dev, int bar,
6432 const struct resource *rsrc,
6433 resource_size_t *start, resource_size_t *end)
6435 *start = rsrc->start;
6439 static char *resource_alignment_param;
6440 static DEFINE_SPINLOCK(resource_alignment_lock);
6443 * pci_specified_resource_alignment - get resource alignment specified by user.
6444 * @dev: the PCI device to get
6445 * @resize: whether or not to change resources' size when reassigning alignment
6447 * RETURNS: Resource alignment if it is specified.
6448 * Zero if it is not specified.
6450 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
6453 int align_order, count;
6454 resource_size_t align = pcibios_default_alignment();
6458 spin_lock(&resource_alignment_lock);
6459 p = resource_alignment_param;
6462 if (pci_has_flag(PCI_PROBE_ONLY)) {
6464 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
6470 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
6473 if (align_order > 63) {
6474 pr_err("PCI: Invalid requested alignment (order %d)\n",
6476 align_order = PAGE_SHIFT;
6479 align_order = PAGE_SHIFT;
6482 ret = pci_dev_str_match(dev, p, &p);
6485 align = 1ULL << align_order;
6487 } else if (ret < 0) {
6488 pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
6493 if (*p != ';' && *p != ',') {
6494 /* End of param or invalid format */
6500 spin_unlock(&resource_alignment_lock);
6504 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
6505 resource_size_t align, bool resize)
6507 struct resource *r = &dev->resource[bar];
6508 const char *r_name = pci_resource_name(dev, bar);
6509 resource_size_t size;
6511 if (!(r->flags & IORESOURCE_MEM))
6514 if (r->flags & IORESOURCE_PCI_FIXED) {
6515 pci_info(dev, "%s %pR: ignoring requested alignment %#llx\n",
6516 r_name, r, (unsigned long long)align);
6520 size = resource_size(r);
6525 * Increase the alignment of the resource. There are two ways we
6528 * 1) Increase the size of the resource. BARs are aligned on their
6529 * size, so when we reallocate space for this resource, we'll
6530 * allocate it with the larger alignment. This also prevents
6531 * assignment of any other BARs inside the alignment region, so
6532 * if we're requesting page alignment, this means no other BARs
6533 * will share the page.
6535 * The disadvantage is that this makes the resource larger than
6536 * the hardware BAR, which may break drivers that compute things
6537 * based on the resource size, e.g., to find registers at a
6538 * fixed offset before the end of the BAR.
6540 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
6541 * set r->start to the desired alignment. By itself this
6542 * doesn't prevent other BARs being put inside the alignment
6543 * region, but if we realign *every* resource of every device in
6544 * the system, none of them will share an alignment region.
6546 * When the user has requested alignment for only some devices via
6547 * the "pci=resource_alignment" argument, "resize" is true and we
6548 * use the first method. Otherwise we assume we're aligning all
6549 * devices and we use the second.
6552 pci_info(dev, "%s %pR: requesting alignment to %#llx\n",
6553 r_name, r, (unsigned long long)align);
6559 r->flags &= ~IORESOURCE_SIZEALIGN;
6560 r->flags |= IORESOURCE_STARTALIGN;
6562 r->end = r->start + size - 1;
6564 r->flags |= IORESOURCE_UNSET;
6568 * This function disables memory decoding and releases memory resources
6569 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
6570 * It also rounds up size to specified alignment.
6571 * Later on, the kernel will assign page-aligned memory resource back
6574 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
6578 resource_size_t align;
6580 bool resize = false;
6583 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
6584 * 3.4.1.11. Their resources are allocated from the space
6585 * described by the VF BARx register in the PF's SR-IOV capability.
6586 * We can't influence their alignment here.
6591 /* check if specified PCI is target device to reassign */
6592 align = pci_specified_resource_alignment(dev, &resize);
6596 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
6597 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
6598 pci_warn(dev, "Can't reassign resources to host bridge\n");
6602 pci_read_config_word(dev, PCI_COMMAND, &command);
6603 command &= ~PCI_COMMAND_MEMORY;
6604 pci_write_config_word(dev, PCI_COMMAND, command);
6606 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
6607 pci_request_resource_alignment(dev, i, align, resize);
6610 * Need to disable bridge's resource window,
6611 * to enable the kernel to reassign new resource
6614 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
6615 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
6616 r = &dev->resource[i];
6617 if (!(r->flags & IORESOURCE_MEM))
6619 r->flags |= IORESOURCE_UNSET;
6620 r->end = resource_size(r) - 1;
6623 pci_disable_bridge_window(dev);
6627 static ssize_t resource_alignment_show(const struct bus_type *bus, char *buf)
6631 spin_lock(&resource_alignment_lock);
6632 if (resource_alignment_param)
6633 count = sysfs_emit(buf, "%s\n", resource_alignment_param);
6634 spin_unlock(&resource_alignment_lock);
6639 static ssize_t resource_alignment_store(const struct bus_type *bus,
6640 const char *buf, size_t count)
6642 char *param, *old, *end;
6644 if (count >= (PAGE_SIZE - 1))
6647 param = kstrndup(buf, count, GFP_KERNEL);
6651 end = strchr(param, '\n');
6655 spin_lock(&resource_alignment_lock);
6656 old = resource_alignment_param;
6657 if (strlen(param)) {
6658 resource_alignment_param = param;
6661 resource_alignment_param = NULL;
6663 spin_unlock(&resource_alignment_lock);
6670 static BUS_ATTR_RW(resource_alignment);
6672 static int __init pci_resource_alignment_sysfs_init(void)
6674 return bus_create_file(&pci_bus_type,
6675 &bus_attr_resource_alignment);
6677 late_initcall(pci_resource_alignment_sysfs_init);
6679 static void pci_no_domains(void)
6681 #ifdef CONFIG_PCI_DOMAINS
6682 pci_domains_supported = 0;
6686 #ifdef CONFIG_PCI_DOMAINS_GENERIC
6687 static DEFINE_IDA(pci_domain_nr_static_ida);
6688 static DEFINE_IDA(pci_domain_nr_dynamic_ida);
6690 static void of_pci_reserve_static_domain_nr(void)
6692 struct device_node *np;
6695 for_each_node_by_type(np, "pci") {
6696 domain_nr = of_get_pci_domain_nr(np);
6700 * Permanently allocate domain_nr in dynamic_ida
6701 * to prevent it from dynamic allocation.
6703 ida_alloc_range(&pci_domain_nr_dynamic_ida,
6704 domain_nr, domain_nr, GFP_KERNEL);
6708 static int of_pci_bus_find_domain_nr(struct device *parent)
6710 static bool static_domains_reserved = false;
6713 /* On the first call scan device tree for static allocations. */
6714 if (!static_domains_reserved) {
6715 of_pci_reserve_static_domain_nr();
6716 static_domains_reserved = true;
6721 * If domain is in DT, allocate it in static IDA. This
6722 * prevents duplicate static allocations in case of errors
6725 domain_nr = of_get_pci_domain_nr(parent->of_node);
6727 return ida_alloc_range(&pci_domain_nr_static_ida,
6728 domain_nr, domain_nr,
6733 * If domain was not specified in DT, choose a free ID from dynamic
6734 * allocations. All domain numbers from DT are permanently in
6735 * dynamic allocations to prevent assigning them to other DT nodes
6736 * without static domain.
6738 return ida_alloc(&pci_domain_nr_dynamic_ida, GFP_KERNEL);
6741 static void of_pci_bus_release_domain_nr(struct pci_bus *bus, struct device *parent)
6743 if (bus->domain_nr < 0)
6746 /* Release domain from IDA where it was allocated. */
6747 if (of_get_pci_domain_nr(parent->of_node) == bus->domain_nr)
6748 ida_free(&pci_domain_nr_static_ida, bus->domain_nr);
6750 ida_free(&pci_domain_nr_dynamic_ida, bus->domain_nr);
6753 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6755 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6756 acpi_pci_bus_find_domain_nr(bus);
6759 void pci_bus_release_domain_nr(struct pci_bus *bus, struct device *parent)
6763 of_pci_bus_release_domain_nr(bus, parent);
6768 * pci_ext_cfg_avail - can we access extended PCI config space?
6770 * Returns 1 if we can access PCI extended config space (offsets
6771 * greater than 0xff). This is the default implementation. Architecture
6772 * implementations can override this.
6774 int __weak pci_ext_cfg_avail(void)
6779 void __weak pci_fixup_cardbus(struct pci_bus *bus)
6782 EXPORT_SYMBOL(pci_fixup_cardbus);
6784 static int __init pci_setup(char *str)
6787 char *k = strchr(str, ',');
6790 if (*str && (str = pcibios_setup(str)) && *str) {
6791 if (!strcmp(str, "nomsi")) {
6793 } else if (!strncmp(str, "noats", 5)) {
6794 pr_info("PCIe: ATS is disabled\n");
6795 pcie_ats_disabled = true;
6796 } else if (!strcmp(str, "noaer")) {
6798 } else if (!strcmp(str, "earlydump")) {
6799 pci_early_dump = true;
6800 } else if (!strncmp(str, "realloc=", 8)) {
6801 pci_realloc_get_opt(str + 8);
6802 } else if (!strncmp(str, "realloc", 7)) {
6803 pci_realloc_get_opt("on");
6804 } else if (!strcmp(str, "nodomains")) {
6806 } else if (!strncmp(str, "noari", 5)) {
6807 pcie_ari_disabled = true;
6808 } else if (!strncmp(str, "cbiosize=", 9)) {
6809 pci_cardbus_io_size = memparse(str + 9, &str);
6810 } else if (!strncmp(str, "cbmemsize=", 10)) {
6811 pci_cardbus_mem_size = memparse(str + 10, &str);
6812 } else if (!strncmp(str, "resource_alignment=", 19)) {
6813 resource_alignment_param = str + 19;
6814 } else if (!strncmp(str, "ecrc=", 5)) {
6815 pcie_ecrc_get_policy(str + 5);
6816 } else if (!strncmp(str, "hpiosize=", 9)) {
6817 pci_hotplug_io_size = memparse(str + 9, &str);
6818 } else if (!strncmp(str, "hpmmiosize=", 11)) {
6819 pci_hotplug_mmio_size = memparse(str + 11, &str);
6820 } else if (!strncmp(str, "hpmmioprefsize=", 15)) {
6821 pci_hotplug_mmio_pref_size = memparse(str + 15, &str);
6822 } else if (!strncmp(str, "hpmemsize=", 10)) {
6823 pci_hotplug_mmio_size = memparse(str + 10, &str);
6824 pci_hotplug_mmio_pref_size = pci_hotplug_mmio_size;
6825 } else if (!strncmp(str, "hpbussize=", 10)) {
6826 pci_hotplug_bus_size =
6827 simple_strtoul(str + 10, &str, 0);
6828 if (pci_hotplug_bus_size > 0xff)
6829 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6830 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6831 pcie_bus_config = PCIE_BUS_TUNE_OFF;
6832 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6833 pcie_bus_config = PCIE_BUS_SAFE;
6834 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6835 pcie_bus_config = PCIE_BUS_PERFORMANCE;
6836 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6837 pcie_bus_config = PCIE_BUS_PEER2PEER;
6838 } else if (!strncmp(str, "pcie_scan_all", 13)) {
6839 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6840 } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6841 disable_acs_redir_param = str + 18;
6843 pr_err("PCI: Unknown option `%s'\n", str);
6850 early_param("pci", pci_setup);
6853 * 'resource_alignment_param' and 'disable_acs_redir_param' are initialized
6854 * in pci_setup(), above, to point to data in the __initdata section which
6855 * will be freed after the init sequence is complete. We can't allocate memory
6856 * in pci_setup() because some architectures do not have any memory allocation
6857 * service available during an early_param() call. So we allocate memory and
6858 * copy the variable here before the init section is freed.
6861 static int __init pci_realloc_setup_params(void)
6863 resource_alignment_param = kstrdup(resource_alignment_param,
6865 disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6869 pure_initcall(pci_realloc_setup_params);