Linux 4.19-rc7
[linux-2.6/btrfs-unstable.git] / drivers / pci / pci.c
blob51b6c81671c1e21baba57422cd90d8727992f8e1
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * PCI Bus Services, see include/linux/pci.h for further explanation.
5 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
6 * David Mosberger-Tang
8 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
9 */
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/of.h>
17 #include <linux/of_pci.h>
18 #include <linux/pci.h>
19 #include <linux/pm.h>
20 #include <linux/slab.h>
21 #include <linux/module.h>
22 #include <linux/spinlock.h>
23 #include <linux/string.h>
24 #include <linux/log2.h>
25 #include <linux/logic_pio.h>
26 #include <linux/pm_wakeup.h>
27 #include <linux/interrupt.h>
28 #include <linux/device.h>
29 #include <linux/pm_runtime.h>
30 #include <linux/pci_hotplug.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pci-ats.h>
33 #include <asm/setup.h>
34 #include <asm/dma.h>
35 #include <linux/aer.h>
36 #include "pci.h"
38 const char *pci_power_names[] = {
39 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
41 EXPORT_SYMBOL_GPL(pci_power_names);
43 int isa_dma_bridge_buggy;
44 EXPORT_SYMBOL(isa_dma_bridge_buggy);
46 int pci_pci_problems;
47 EXPORT_SYMBOL(pci_pci_problems);
49 unsigned int pci_pm_d3_delay;
51 static void pci_pme_list_scan(struct work_struct *work);
53 static LIST_HEAD(pci_pme_list);
54 static DEFINE_MUTEX(pci_pme_list_mutex);
55 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
57 struct pci_pme_device {
58 struct list_head list;
59 struct pci_dev *dev;
62 #define PME_TIMEOUT 1000 /* How long between PME checks */
64 static void pci_dev_d3_sleep(struct pci_dev *dev)
66 unsigned int delay = dev->d3_delay;
68 if (delay < pci_pm_d3_delay)
69 delay = pci_pm_d3_delay;
71 if (delay)
72 msleep(delay);
75 #ifdef CONFIG_PCI_DOMAINS
76 int pci_domains_supported = 1;
77 #endif
79 #define DEFAULT_CARDBUS_IO_SIZE (256)
80 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
81 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
82 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
83 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
85 #define DEFAULT_HOTPLUG_IO_SIZE (256)
86 #define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
87 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
88 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
89 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
91 #define DEFAULT_HOTPLUG_BUS_SIZE 1
92 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
94 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
97 * The default CLS is used if arch didn't set CLS explicitly and not
98 * all pci devices agree on the same value. Arch can override either
99 * the dfl or actual value as it sees fit. Don't forget this is
100 * measured in 32-bit words, not bytes.
102 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
103 u8 pci_cache_line_size;
106 * If we set up a device for bus mastering, we need to check the latency
107 * timer as certain BIOSes forget to set it properly.
109 unsigned int pcibios_max_latency = 255;
111 /* If set, the PCIe ARI capability will not be used. */
112 static bool pcie_ari_disabled;
114 /* If set, the PCIe ATS capability will not be used. */
115 static bool pcie_ats_disabled;
117 /* If set, the PCI config space of each device is printed during boot. */
118 bool pci_early_dump;
120 bool pci_ats_disabled(void)
122 return pcie_ats_disabled;
125 /* Disable bridge_d3 for all PCIe ports */
126 static bool pci_bridge_d3_disable;
127 /* Force bridge_d3 for all PCIe ports */
128 static bool pci_bridge_d3_force;
130 static int __init pcie_port_pm_setup(char *str)
132 if (!strcmp(str, "off"))
133 pci_bridge_d3_disable = true;
134 else if (!strcmp(str, "force"))
135 pci_bridge_d3_force = true;
136 return 1;
138 __setup("pcie_port_pm=", pcie_port_pm_setup);
140 /* Time to wait after a reset for device to become responsive */
141 #define PCIE_RESET_READY_POLL_MS 60000
144 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
145 * @bus: pointer to PCI bus structure to search
147 * Given a PCI bus, returns the highest PCI bus number present in the set
148 * including the given PCI bus and its list of child PCI buses.
150 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
152 struct pci_bus *tmp;
153 unsigned char max, n;
155 max = bus->busn_res.end;
156 list_for_each_entry(tmp, &bus->children, node) {
157 n = pci_bus_max_busnr(tmp);
158 if (n > max)
159 max = n;
161 return max;
163 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
165 #ifdef CONFIG_HAS_IOMEM
166 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
168 struct resource *res = &pdev->resource[bar];
171 * Make sure the BAR is actually a memory resource, not an IO resource
173 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
174 pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
175 return NULL;
177 return ioremap_nocache(res->start, resource_size(res));
179 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
181 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
184 * Make sure the BAR is actually a memory resource, not an IO resource
186 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
187 WARN_ON(1);
188 return NULL;
190 return ioremap_wc(pci_resource_start(pdev, bar),
191 pci_resource_len(pdev, bar));
193 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
194 #endif
197 * pci_dev_str_match_path - test if a path string matches a device
198 * @dev: the PCI device to test
199 * @p: string to match the device against
200 * @endptr: pointer to the string after the match
202 * Test if a string (typically from a kernel parameter) formatted as a
203 * path of device/function addresses matches a PCI device. The string must
204 * be of the form:
206 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
208 * A path for a device can be obtained using 'lspci -t'. Using a path
209 * is more robust against bus renumbering than using only a single bus,
210 * device and function address.
212 * Returns 1 if the string matches the device, 0 if it does not and
213 * a negative error code if it fails to parse the string.
215 static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
216 const char **endptr)
218 int ret;
219 int seg, bus, slot, func;
220 char *wpath, *p;
221 char end;
223 *endptr = strchrnul(path, ';');
225 wpath = kmemdup_nul(path, *endptr - path, GFP_KERNEL);
226 if (!wpath)
227 return -ENOMEM;
229 while (1) {
230 p = strrchr(wpath, '/');
231 if (!p)
232 break;
233 ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
234 if (ret != 2) {
235 ret = -EINVAL;
236 goto free_and_exit;
239 if (dev->devfn != PCI_DEVFN(slot, func)) {
240 ret = 0;
241 goto free_and_exit;
245 * Note: we don't need to get a reference to the upstream
246 * bridge because we hold a reference to the top level
247 * device which should hold a reference to the bridge,
248 * and so on.
250 dev = pci_upstream_bridge(dev);
251 if (!dev) {
252 ret = 0;
253 goto free_and_exit;
256 *p = 0;
259 ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
260 &func, &end);
261 if (ret != 4) {
262 seg = 0;
263 ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
264 if (ret != 3) {
265 ret = -EINVAL;
266 goto free_and_exit;
270 ret = (seg == pci_domain_nr(dev->bus) &&
271 bus == dev->bus->number &&
272 dev->devfn == PCI_DEVFN(slot, func));
274 free_and_exit:
275 kfree(wpath);
276 return ret;
280 * pci_dev_str_match - test if a string matches a device
281 * @dev: the PCI device to test
282 * @p: string to match the device against
283 * @endptr: pointer to the string after the match
285 * Test if a string (typically from a kernel parameter) matches a specified
286 * PCI device. The string may be of one of the following formats:
288 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
289 * pci:<vendor>:<device>[:<subvendor>:<subdevice>]
291 * The first format specifies a PCI bus/device/function address which
292 * may change if new hardware is inserted, if motherboard firmware changes,
293 * or due to changes caused in kernel parameters. If the domain is
294 * left unspecified, it is taken to be 0. In order to be robust against
295 * bus renumbering issues, a path of PCI device/function numbers may be used
296 * to address the specific device. The path for a device can be determined
297 * through the use of 'lspci -t'.
299 * The second format matches devices using IDs in the configuration
300 * space which may match multiple devices in the system. A value of 0
301 * for any field will match all devices. (Note: this differs from
302 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
303 * legacy reasons and convenience so users don't have to specify
304 * FFFFFFFFs on the command line.)
306 * Returns 1 if the string matches the device, 0 if it does not and
307 * a negative error code if the string cannot be parsed.
309 static int pci_dev_str_match(struct pci_dev *dev, const char *p,
310 const char **endptr)
312 int ret;
313 int count;
314 unsigned short vendor, device, subsystem_vendor, subsystem_device;
316 if (strncmp(p, "pci:", 4) == 0) {
317 /* PCI vendor/device (subvendor/subdevice) IDs are specified */
318 p += 4;
319 ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
320 &subsystem_vendor, &subsystem_device, &count);
321 if (ret != 4) {
322 ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
323 if (ret != 2)
324 return -EINVAL;
326 subsystem_vendor = 0;
327 subsystem_device = 0;
330 p += count;
332 if ((!vendor || vendor == dev->vendor) &&
333 (!device || device == dev->device) &&
334 (!subsystem_vendor ||
335 subsystem_vendor == dev->subsystem_vendor) &&
336 (!subsystem_device ||
337 subsystem_device == dev->subsystem_device))
338 goto found;
339 } else {
341 * PCI Bus, Device, Function IDs are specified
342 * (optionally, may include a path of devfns following it)
344 ret = pci_dev_str_match_path(dev, p, &p);
345 if (ret < 0)
346 return ret;
347 else if (ret)
348 goto found;
351 *endptr = p;
352 return 0;
354 found:
355 *endptr = p;
356 return 1;
359 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
360 u8 pos, int cap, int *ttl)
362 u8 id;
363 u16 ent;
365 pci_bus_read_config_byte(bus, devfn, pos, &pos);
367 while ((*ttl)--) {
368 if (pos < 0x40)
369 break;
370 pos &= ~3;
371 pci_bus_read_config_word(bus, devfn, pos, &ent);
373 id = ent & 0xff;
374 if (id == 0xff)
375 break;
376 if (id == cap)
377 return pos;
378 pos = (ent >> 8);
380 return 0;
383 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
384 u8 pos, int cap)
386 int ttl = PCI_FIND_CAP_TTL;
388 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
391 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
393 return __pci_find_next_cap(dev->bus, dev->devfn,
394 pos + PCI_CAP_LIST_NEXT, cap);
396 EXPORT_SYMBOL_GPL(pci_find_next_capability);
398 static int __pci_bus_find_cap_start(struct pci_bus *bus,
399 unsigned int devfn, u8 hdr_type)
401 u16 status;
403 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
404 if (!(status & PCI_STATUS_CAP_LIST))
405 return 0;
407 switch (hdr_type) {
408 case PCI_HEADER_TYPE_NORMAL:
409 case PCI_HEADER_TYPE_BRIDGE:
410 return PCI_CAPABILITY_LIST;
411 case PCI_HEADER_TYPE_CARDBUS:
412 return PCI_CB_CAPABILITY_LIST;
415 return 0;
419 * pci_find_capability - query for devices' capabilities
420 * @dev: PCI device to query
421 * @cap: capability code
423 * Tell if a device supports a given PCI capability.
424 * Returns the address of the requested capability structure within the
425 * device's PCI configuration space or 0 in case the device does not
426 * support it. Possible values for @cap:
428 * %PCI_CAP_ID_PM Power Management
429 * %PCI_CAP_ID_AGP Accelerated Graphics Port
430 * %PCI_CAP_ID_VPD Vital Product Data
431 * %PCI_CAP_ID_SLOTID Slot Identification
432 * %PCI_CAP_ID_MSI Message Signalled Interrupts
433 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
434 * %PCI_CAP_ID_PCIX PCI-X
435 * %PCI_CAP_ID_EXP PCI Express
437 int pci_find_capability(struct pci_dev *dev, int cap)
439 int pos;
441 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
442 if (pos)
443 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
445 return pos;
447 EXPORT_SYMBOL(pci_find_capability);
450 * pci_bus_find_capability - query for devices' capabilities
451 * @bus: the PCI bus to query
452 * @devfn: PCI device to query
453 * @cap: capability code
455 * Like pci_find_capability() but works for pci devices that do not have a
456 * pci_dev structure set up yet.
458 * Returns the address of the requested capability structure within the
459 * device's PCI configuration space or 0 in case the device does not
460 * support it.
462 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
464 int pos;
465 u8 hdr_type;
467 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
469 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
470 if (pos)
471 pos = __pci_find_next_cap(bus, devfn, pos, cap);
473 return pos;
475 EXPORT_SYMBOL(pci_bus_find_capability);
478 * pci_find_next_ext_capability - Find an extended capability
479 * @dev: PCI device to query
480 * @start: address at which to start looking (0 to start at beginning of list)
481 * @cap: capability code
483 * Returns the address of the next matching extended capability structure
484 * within the device's PCI configuration space or 0 if the device does
485 * not support it. Some capabilities can occur several times, e.g., the
486 * vendor-specific capability, and this provides a way to find them all.
488 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
490 u32 header;
491 int ttl;
492 int pos = PCI_CFG_SPACE_SIZE;
494 /* minimum 8 bytes per capability */
495 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
497 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
498 return 0;
500 if (start)
501 pos = start;
503 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
504 return 0;
507 * If we have no capabilities, this is indicated by cap ID,
508 * cap version and next pointer all being 0.
510 if (header == 0)
511 return 0;
513 while (ttl-- > 0) {
514 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
515 return pos;
517 pos = PCI_EXT_CAP_NEXT(header);
518 if (pos < PCI_CFG_SPACE_SIZE)
519 break;
521 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
522 break;
525 return 0;
527 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
530 * pci_find_ext_capability - Find an extended capability
531 * @dev: PCI device to query
532 * @cap: capability code
534 * Returns the address of the requested extended capability structure
535 * within the device's PCI configuration space or 0 if the device does
536 * not support it. Possible values for @cap:
538 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
539 * %PCI_EXT_CAP_ID_VC Virtual Channel
540 * %PCI_EXT_CAP_ID_DSN Device Serial Number
541 * %PCI_EXT_CAP_ID_PWR Power Budgeting
543 int pci_find_ext_capability(struct pci_dev *dev, int cap)
545 return pci_find_next_ext_capability(dev, 0, cap);
547 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
549 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
551 int rc, ttl = PCI_FIND_CAP_TTL;
552 u8 cap, mask;
554 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
555 mask = HT_3BIT_CAP_MASK;
556 else
557 mask = HT_5BIT_CAP_MASK;
559 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
560 PCI_CAP_ID_HT, &ttl);
561 while (pos) {
562 rc = pci_read_config_byte(dev, pos + 3, &cap);
563 if (rc != PCIBIOS_SUCCESSFUL)
564 return 0;
566 if ((cap & mask) == ht_cap)
567 return pos;
569 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
570 pos + PCI_CAP_LIST_NEXT,
571 PCI_CAP_ID_HT, &ttl);
574 return 0;
577 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
578 * @dev: PCI device to query
579 * @pos: Position from which to continue searching
580 * @ht_cap: Hypertransport capability code
582 * To be used in conjunction with pci_find_ht_capability() to search for
583 * all capabilities matching @ht_cap. @pos should always be a value returned
584 * from pci_find_ht_capability().
586 * NB. To be 100% safe against broken PCI devices, the caller should take
587 * steps to avoid an infinite loop.
589 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
591 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
593 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
596 * pci_find_ht_capability - query a device's Hypertransport capabilities
597 * @dev: PCI device to query
598 * @ht_cap: Hypertransport capability code
600 * Tell if a device supports a given Hypertransport capability.
601 * Returns an address within the device's PCI configuration space
602 * or 0 in case the device does not support the request capability.
603 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
604 * which has a Hypertransport capability matching @ht_cap.
606 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
608 int pos;
610 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
611 if (pos)
612 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
614 return pos;
616 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
619 * pci_find_parent_resource - return resource region of parent bus of given region
620 * @dev: PCI device structure contains resources to be searched
621 * @res: child resource record for which parent is sought
623 * For given resource region of given device, return the resource
624 * region of parent bus the given region is contained in.
626 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
627 struct resource *res)
629 const struct pci_bus *bus = dev->bus;
630 struct resource *r;
631 int i;
633 pci_bus_for_each_resource(bus, r, i) {
634 if (!r)
635 continue;
636 if (resource_contains(r, res)) {
639 * If the window is prefetchable but the BAR is
640 * not, the allocator made a mistake.
642 if (r->flags & IORESOURCE_PREFETCH &&
643 !(res->flags & IORESOURCE_PREFETCH))
644 return NULL;
647 * If we're below a transparent bridge, there may
648 * be both a positively-decoded aperture and a
649 * subtractively-decoded region that contain the BAR.
650 * We want the positively-decoded one, so this depends
651 * on pci_bus_for_each_resource() giving us those
652 * first.
654 return r;
657 return NULL;
659 EXPORT_SYMBOL(pci_find_parent_resource);
662 * pci_find_resource - Return matching PCI device resource
663 * @dev: PCI device to query
664 * @res: Resource to look for
666 * Goes over standard PCI resources (BARs) and checks if the given resource
667 * is partially or fully contained in any of them. In that case the
668 * matching resource is returned, %NULL otherwise.
670 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
672 int i;
674 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
675 struct resource *r = &dev->resource[i];
677 if (r->start && resource_contains(r, res))
678 return r;
681 return NULL;
683 EXPORT_SYMBOL(pci_find_resource);
686 * pci_find_pcie_root_port - return PCIe Root Port
687 * @dev: PCI device to query
689 * Traverse up the parent chain and return the PCIe Root Port PCI Device
690 * for a given PCI Device.
692 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
694 struct pci_dev *bridge, *highest_pcie_bridge = dev;
696 bridge = pci_upstream_bridge(dev);
697 while (bridge && pci_is_pcie(bridge)) {
698 highest_pcie_bridge = bridge;
699 bridge = pci_upstream_bridge(bridge);
702 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
703 return NULL;
705 return highest_pcie_bridge;
707 EXPORT_SYMBOL(pci_find_pcie_root_port);
710 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
711 * @dev: the PCI device to operate on
712 * @pos: config space offset of status word
713 * @mask: mask of bit(s) to care about in status word
715 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
717 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
719 int i;
721 /* Wait for Transaction Pending bit clean */
722 for (i = 0; i < 4; i++) {
723 u16 status;
724 if (i)
725 msleep((1 << (i - 1)) * 100);
727 pci_read_config_word(dev, pos, &status);
728 if (!(status & mask))
729 return 1;
732 return 0;
736 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
737 * @dev: PCI device to have its BARs restored
739 * Restore the BAR values for a given device, so as to make it
740 * accessible by its driver.
742 static void pci_restore_bars(struct pci_dev *dev)
744 int i;
746 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
747 pci_update_resource(dev, i);
750 static const struct pci_platform_pm_ops *pci_platform_pm;
752 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
754 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
755 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
756 return -EINVAL;
757 pci_platform_pm = ops;
758 return 0;
761 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
763 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
766 static inline int platform_pci_set_power_state(struct pci_dev *dev,
767 pci_power_t t)
769 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
772 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
774 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
777 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
779 return pci_platform_pm ?
780 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
783 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
785 return pci_platform_pm ?
786 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
789 static inline bool platform_pci_need_resume(struct pci_dev *dev)
791 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
795 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
796 * given PCI device
797 * @dev: PCI device to handle.
798 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
800 * RETURN VALUE:
801 * -EINVAL if the requested state is invalid.
802 * -EIO if device does not support PCI PM or its PM capabilities register has a
803 * wrong version, or device doesn't support the requested state.
804 * 0 if device already is in the requested state.
805 * 0 if device's power state has been successfully changed.
807 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
809 u16 pmcsr;
810 bool need_restore = false;
812 /* Check if we're already there */
813 if (dev->current_state == state)
814 return 0;
816 if (!dev->pm_cap)
817 return -EIO;
819 if (state < PCI_D0 || state > PCI_D3hot)
820 return -EINVAL;
822 /* Validate current state:
823 * Can enter D0 from any state, but if we can only go deeper
824 * to sleep if we're already in a low power state
826 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
827 && dev->current_state > state) {
828 pci_err(dev, "invalid power transition (from state %d to %d)\n",
829 dev->current_state, state);
830 return -EINVAL;
833 /* check if this device supports the desired state */
834 if ((state == PCI_D1 && !dev->d1_support)
835 || (state == PCI_D2 && !dev->d2_support))
836 return -EIO;
838 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
840 /* If we're (effectively) in D3, force entire word to 0.
841 * This doesn't affect PME_Status, disables PME_En, and
842 * sets PowerState to 0.
844 switch (dev->current_state) {
845 case PCI_D0:
846 case PCI_D1:
847 case PCI_D2:
848 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
849 pmcsr |= state;
850 break;
851 case PCI_D3hot:
852 case PCI_D3cold:
853 case PCI_UNKNOWN: /* Boot-up */
854 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
855 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
856 need_restore = true;
857 /* Fall-through: force to D0 */
858 default:
859 pmcsr = 0;
860 break;
863 /* enter specified state */
864 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
866 /* Mandatory power management transition delays */
867 /* see PCI PM 1.1 5.6.1 table 18 */
868 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
869 pci_dev_d3_sleep(dev);
870 else if (state == PCI_D2 || dev->current_state == PCI_D2)
871 udelay(PCI_PM_D2_DELAY);
873 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
874 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
875 if (dev->current_state != state && printk_ratelimit())
876 pci_info(dev, "Refused to change power state, currently in D%d\n",
877 dev->current_state);
880 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
881 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
882 * from D3hot to D0 _may_ perform an internal reset, thereby
883 * going to "D0 Uninitialized" rather than "D0 Initialized".
884 * For example, at least some versions of the 3c905B and the
885 * 3c556B exhibit this behaviour.
887 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
888 * devices in a D3hot state at boot. Consequently, we need to
889 * restore at least the BARs so that the device will be
890 * accessible to its driver.
892 if (need_restore)
893 pci_restore_bars(dev);
895 if (dev->bus->self)
896 pcie_aspm_pm_state_change(dev->bus->self);
898 return 0;
902 * pci_update_current_state - Read power state of given device and cache it
903 * @dev: PCI device to handle.
904 * @state: State to cache in case the device doesn't have the PM capability
906 * The power state is read from the PMCSR register, which however is
907 * inaccessible in D3cold. The platform firmware is therefore queried first
908 * to detect accessibility of the register. In case the platform firmware
909 * reports an incorrect state or the device isn't power manageable by the
910 * platform at all, we try to detect D3cold by testing accessibility of the
911 * vendor ID in config space.
913 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
915 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
916 !pci_device_is_present(dev)) {
917 dev->current_state = PCI_D3cold;
918 } else if (dev->pm_cap) {
919 u16 pmcsr;
921 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
922 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
923 } else {
924 dev->current_state = state;
929 * pci_power_up - Put the given device into D0 forcibly
930 * @dev: PCI device to power up
932 void pci_power_up(struct pci_dev *dev)
934 if (platform_pci_power_manageable(dev))
935 platform_pci_set_power_state(dev, PCI_D0);
937 pci_raw_set_power_state(dev, PCI_D0);
938 pci_update_current_state(dev, PCI_D0);
942 * pci_platform_power_transition - Use platform to change device power state
943 * @dev: PCI device to handle.
944 * @state: State to put the device into.
946 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
948 int error;
950 if (platform_pci_power_manageable(dev)) {
951 error = platform_pci_set_power_state(dev, state);
952 if (!error)
953 pci_update_current_state(dev, state);
954 } else
955 error = -ENODEV;
957 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
958 dev->current_state = PCI_D0;
960 return error;
964 * pci_wakeup - Wake up a PCI device
965 * @pci_dev: Device to handle.
966 * @ign: ignored parameter
968 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
970 pci_wakeup_event(pci_dev);
971 pm_request_resume(&pci_dev->dev);
972 return 0;
976 * pci_wakeup_bus - Walk given bus and wake up devices on it
977 * @bus: Top bus of the subtree to walk.
979 void pci_wakeup_bus(struct pci_bus *bus)
981 if (bus)
982 pci_walk_bus(bus, pci_wakeup, NULL);
986 * __pci_start_power_transition - Start power transition of a PCI device
987 * @dev: PCI device to handle.
988 * @state: State to put the device into.
990 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
992 if (state == PCI_D0) {
993 pci_platform_power_transition(dev, PCI_D0);
995 * Mandatory power management transition delays, see
996 * PCI Express Base Specification Revision 2.0 Section
997 * 6.6.1: Conventional Reset. Do not delay for
998 * devices powered on/off by corresponding bridge,
999 * because have already delayed for the bridge.
1001 if (dev->runtime_d3cold) {
1002 if (dev->d3cold_delay)
1003 msleep(dev->d3cold_delay);
1005 * When powering on a bridge from D3cold, the
1006 * whole hierarchy may be powered on into
1007 * D0uninitialized state, resume them to give
1008 * them a chance to suspend again
1010 pci_wakeup_bus(dev->subordinate);
1016 * __pci_dev_set_current_state - Set current state of a PCI device
1017 * @dev: Device to handle
1018 * @data: pointer to state to be set
1020 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1022 pci_power_t state = *(pci_power_t *)data;
1024 dev->current_state = state;
1025 return 0;
1029 * pci_bus_set_current_state - Walk given bus and set current state of devices
1030 * @bus: Top bus of the subtree to walk.
1031 * @state: state to be set
1033 void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1035 if (bus)
1036 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1040 * __pci_complete_power_transition - Complete power transition of a PCI device
1041 * @dev: PCI device to handle.
1042 * @state: State to put the device into.
1044 * This function should not be called directly by device drivers.
1046 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
1048 int ret;
1050 if (state <= PCI_D0)
1051 return -EINVAL;
1052 ret = pci_platform_power_transition(dev, state);
1053 /* Power off the bridge may power off the whole hierarchy */
1054 if (!ret && state == PCI_D3cold)
1055 pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1056 return ret;
1058 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
1061 * pci_set_power_state - Set the power state of a PCI device
1062 * @dev: PCI device to handle.
1063 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1065 * Transition a device to a new power state, using the platform firmware and/or
1066 * the device's PCI PM registers.
1068 * RETURN VALUE:
1069 * -EINVAL if the requested state is invalid.
1070 * -EIO if device does not support PCI PM or its PM capabilities register has a
1071 * wrong version, or device doesn't support the requested state.
1072 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1073 * 0 if device already is in the requested state.
1074 * 0 if the transition is to D3 but D3 is not supported.
1075 * 0 if device's power state has been successfully changed.
1077 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1079 int error;
1081 /* bound the state we're entering */
1082 if (state > PCI_D3cold)
1083 state = PCI_D3cold;
1084 else if (state < PCI_D0)
1085 state = PCI_D0;
1086 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1088 * If the device or the parent bridge do not support PCI PM,
1089 * ignore the request if we're doing anything other than putting
1090 * it into D0 (which would only happen on boot).
1092 return 0;
1094 /* Check if we're already there */
1095 if (dev->current_state == state)
1096 return 0;
1098 __pci_start_power_transition(dev, state);
1100 /* This device is quirked not to be put into D3, so
1101 don't put it in D3 */
1102 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1103 return 0;
1106 * To put device in D3cold, we put device into D3hot in native
1107 * way, then put device into D3cold with platform ops
1109 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1110 PCI_D3hot : state);
1112 if (!__pci_complete_power_transition(dev, state))
1113 error = 0;
1115 return error;
1117 EXPORT_SYMBOL(pci_set_power_state);
1120 * pci_choose_state - Choose the power state of a PCI device
1121 * @dev: PCI device to be suspended
1122 * @state: target sleep state for the whole system. This is the value
1123 * that is passed to suspend() function.
1125 * Returns PCI power state suitable for given device and given system
1126 * message.
1129 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
1131 pci_power_t ret;
1133 if (!dev->pm_cap)
1134 return PCI_D0;
1136 ret = platform_pci_choose_state(dev);
1137 if (ret != PCI_POWER_ERROR)
1138 return ret;
1140 switch (state.event) {
1141 case PM_EVENT_ON:
1142 return PCI_D0;
1143 case PM_EVENT_FREEZE:
1144 case PM_EVENT_PRETHAW:
1145 /* REVISIT both freeze and pre-thaw "should" use D0 */
1146 case PM_EVENT_SUSPEND:
1147 case PM_EVENT_HIBERNATE:
1148 return PCI_D3hot;
1149 default:
1150 pci_info(dev, "unrecognized suspend event %d\n",
1151 state.event);
1152 BUG();
1154 return PCI_D0;
1156 EXPORT_SYMBOL(pci_choose_state);
1158 #define PCI_EXP_SAVE_REGS 7
1160 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
1161 u16 cap, bool extended)
1163 struct pci_cap_saved_state *tmp;
1165 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
1166 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
1167 return tmp;
1169 return NULL;
1172 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1174 return _pci_find_saved_cap(dev, cap, false);
1177 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1179 return _pci_find_saved_cap(dev, cap, true);
1182 static int pci_save_pcie_state(struct pci_dev *dev)
1184 int i = 0;
1185 struct pci_cap_saved_state *save_state;
1186 u16 *cap;
1188 if (!pci_is_pcie(dev))
1189 return 0;
1191 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1192 if (!save_state) {
1193 pci_err(dev, "buffer not found in %s\n", __func__);
1194 return -ENOMEM;
1197 cap = (u16 *)&save_state->cap.data[0];
1198 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1199 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1200 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1201 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1202 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1203 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1204 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1206 return 0;
1209 static void pci_restore_pcie_state(struct pci_dev *dev)
1211 int i = 0;
1212 struct pci_cap_saved_state *save_state;
1213 u16 *cap;
1215 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1216 if (!save_state)
1217 return;
1219 cap = (u16 *)&save_state->cap.data[0];
1220 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1221 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1222 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1223 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1224 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1225 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1226 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1230 static int pci_save_pcix_state(struct pci_dev *dev)
1232 int pos;
1233 struct pci_cap_saved_state *save_state;
1235 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1236 if (!pos)
1237 return 0;
1239 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1240 if (!save_state) {
1241 pci_err(dev, "buffer not found in %s\n", __func__);
1242 return -ENOMEM;
1245 pci_read_config_word(dev, pos + PCI_X_CMD,
1246 (u16 *)save_state->cap.data);
1248 return 0;
1251 static void pci_restore_pcix_state(struct pci_dev *dev)
1253 int i = 0, pos;
1254 struct pci_cap_saved_state *save_state;
1255 u16 *cap;
1257 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1258 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1259 if (!save_state || !pos)
1260 return;
1261 cap = (u16 *)&save_state->cap.data[0];
1263 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1268 * pci_save_state - save the PCI configuration space of a device before suspending
1269 * @dev: - PCI device that we're dealing with
1271 int pci_save_state(struct pci_dev *dev)
1273 int i;
1274 /* XXX: 100% dword access ok here? */
1275 for (i = 0; i < 16; i++)
1276 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1277 dev->state_saved = true;
1279 i = pci_save_pcie_state(dev);
1280 if (i != 0)
1281 return i;
1283 i = pci_save_pcix_state(dev);
1284 if (i != 0)
1285 return i;
1287 return pci_save_vc_state(dev);
1289 EXPORT_SYMBOL(pci_save_state);
1291 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1292 u32 saved_val, int retry, bool force)
1294 u32 val;
1296 pci_read_config_dword(pdev, offset, &val);
1297 if (!force && val == saved_val)
1298 return;
1300 for (;;) {
1301 pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1302 offset, val, saved_val);
1303 pci_write_config_dword(pdev, offset, saved_val);
1304 if (retry-- <= 0)
1305 return;
1307 pci_read_config_dword(pdev, offset, &val);
1308 if (val == saved_val)
1309 return;
1311 mdelay(1);
1315 static void pci_restore_config_space_range(struct pci_dev *pdev,
1316 int start, int end, int retry,
1317 bool force)
1319 int index;
1321 for (index = end; index >= start; index--)
1322 pci_restore_config_dword(pdev, 4 * index,
1323 pdev->saved_config_space[index],
1324 retry, force);
1327 static void pci_restore_config_space(struct pci_dev *pdev)
1329 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1330 pci_restore_config_space_range(pdev, 10, 15, 0, false);
1331 /* Restore BARs before the command register. */
1332 pci_restore_config_space_range(pdev, 4, 9, 10, false);
1333 pci_restore_config_space_range(pdev, 0, 3, 0, false);
1334 } else if (pdev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
1335 pci_restore_config_space_range(pdev, 12, 15, 0, false);
1338 * Force rewriting of prefetch registers to avoid S3 resume
1339 * issues on Intel PCI bridges that occur when these
1340 * registers are not explicitly written.
1342 pci_restore_config_space_range(pdev, 9, 11, 0, true);
1343 pci_restore_config_space_range(pdev, 0, 8, 0, false);
1344 } else {
1345 pci_restore_config_space_range(pdev, 0, 15, 0, false);
1349 static void pci_restore_rebar_state(struct pci_dev *pdev)
1351 unsigned int pos, nbars, i;
1352 u32 ctrl;
1354 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
1355 if (!pos)
1356 return;
1358 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1359 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
1360 PCI_REBAR_CTRL_NBAR_SHIFT;
1362 for (i = 0; i < nbars; i++, pos += 8) {
1363 struct resource *res;
1364 int bar_idx, size;
1366 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
1367 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
1368 res = pdev->resource + bar_idx;
1369 size = order_base_2((resource_size(res) >> 20) | 1) - 1;
1370 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
1371 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
1372 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
1377 * pci_restore_state - Restore the saved state of a PCI device
1378 * @dev: - PCI device that we're dealing with
1380 void pci_restore_state(struct pci_dev *dev)
1382 if (!dev->state_saved)
1383 return;
1385 /* PCI Express register must be restored first */
1386 pci_restore_pcie_state(dev);
1387 pci_restore_pasid_state(dev);
1388 pci_restore_pri_state(dev);
1389 pci_restore_ats_state(dev);
1390 pci_restore_vc_state(dev);
1391 pci_restore_rebar_state(dev);
1393 pci_cleanup_aer_error_status_regs(dev);
1395 pci_restore_config_space(dev);
1397 pci_restore_pcix_state(dev);
1398 pci_restore_msi_state(dev);
1400 /* Restore ACS and IOV configuration state */
1401 pci_enable_acs(dev);
1402 pci_restore_iov_state(dev);
1404 dev->state_saved = false;
1406 EXPORT_SYMBOL(pci_restore_state);
1408 struct pci_saved_state {
1409 u32 config_space[16];
1410 struct pci_cap_saved_data cap[0];
1414 * pci_store_saved_state - Allocate and return an opaque struct containing
1415 * the device saved state.
1416 * @dev: PCI device that we're dealing with
1418 * Return NULL if no state or error.
1420 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1422 struct pci_saved_state *state;
1423 struct pci_cap_saved_state *tmp;
1424 struct pci_cap_saved_data *cap;
1425 size_t size;
1427 if (!dev->state_saved)
1428 return NULL;
1430 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1432 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1433 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1435 state = kzalloc(size, GFP_KERNEL);
1436 if (!state)
1437 return NULL;
1439 memcpy(state->config_space, dev->saved_config_space,
1440 sizeof(state->config_space));
1442 cap = state->cap;
1443 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1444 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1445 memcpy(cap, &tmp->cap, len);
1446 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1448 /* Empty cap_save terminates list */
1450 return state;
1452 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1455 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1456 * @dev: PCI device that we're dealing with
1457 * @state: Saved state returned from pci_store_saved_state()
1459 int pci_load_saved_state(struct pci_dev *dev,
1460 struct pci_saved_state *state)
1462 struct pci_cap_saved_data *cap;
1464 dev->state_saved = false;
1466 if (!state)
1467 return 0;
1469 memcpy(dev->saved_config_space, state->config_space,
1470 sizeof(state->config_space));
1472 cap = state->cap;
1473 while (cap->size) {
1474 struct pci_cap_saved_state *tmp;
1476 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1477 if (!tmp || tmp->cap.size != cap->size)
1478 return -EINVAL;
1480 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1481 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1482 sizeof(struct pci_cap_saved_data) + cap->size);
1485 dev->state_saved = true;
1486 return 0;
1488 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1491 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1492 * and free the memory allocated for it.
1493 * @dev: PCI device that we're dealing with
1494 * @state: Pointer to saved state returned from pci_store_saved_state()
1496 int pci_load_and_free_saved_state(struct pci_dev *dev,
1497 struct pci_saved_state **state)
1499 int ret = pci_load_saved_state(dev, *state);
1500 kfree(*state);
1501 *state = NULL;
1502 return ret;
1504 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1506 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1508 return pci_enable_resources(dev, bars);
1511 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1513 int err;
1514 struct pci_dev *bridge;
1515 u16 cmd;
1516 u8 pin;
1518 err = pci_set_power_state(dev, PCI_D0);
1519 if (err < 0 && err != -EIO)
1520 return err;
1522 bridge = pci_upstream_bridge(dev);
1523 if (bridge)
1524 pcie_aspm_powersave_config_link(bridge);
1526 err = pcibios_enable_device(dev, bars);
1527 if (err < 0)
1528 return err;
1529 pci_fixup_device(pci_fixup_enable, dev);
1531 if (dev->msi_enabled || dev->msix_enabled)
1532 return 0;
1534 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1535 if (pin) {
1536 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1537 if (cmd & PCI_COMMAND_INTX_DISABLE)
1538 pci_write_config_word(dev, PCI_COMMAND,
1539 cmd & ~PCI_COMMAND_INTX_DISABLE);
1542 return 0;
1546 * pci_reenable_device - Resume abandoned device
1547 * @dev: PCI device to be resumed
1549 * Note this function is a backend of pci_default_resume and is not supposed
1550 * to be called by normal code, write proper resume handler and use it instead.
1552 int pci_reenable_device(struct pci_dev *dev)
1554 if (pci_is_enabled(dev))
1555 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1556 return 0;
1558 EXPORT_SYMBOL(pci_reenable_device);
1560 static void pci_enable_bridge(struct pci_dev *dev)
1562 struct pci_dev *bridge;
1563 int retval;
1565 bridge = pci_upstream_bridge(dev);
1566 if (bridge)
1567 pci_enable_bridge(bridge);
1569 if (pci_is_enabled(dev)) {
1570 if (!dev->is_busmaster)
1571 pci_set_master(dev);
1572 return;
1575 retval = pci_enable_device(dev);
1576 if (retval)
1577 pci_err(dev, "Error enabling bridge (%d), continuing\n",
1578 retval);
1579 pci_set_master(dev);
1582 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1584 struct pci_dev *bridge;
1585 int err;
1586 int i, bars = 0;
1589 * Power state could be unknown at this point, either due to a fresh
1590 * boot or a device removal call. So get the current power state
1591 * so that things like MSI message writing will behave as expected
1592 * (e.g. if the device really is in D0 at enable time).
1594 if (dev->pm_cap) {
1595 u16 pmcsr;
1596 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1597 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1600 if (atomic_inc_return(&dev->enable_cnt) > 1)
1601 return 0; /* already enabled */
1603 bridge = pci_upstream_bridge(dev);
1604 if (bridge)
1605 pci_enable_bridge(bridge);
1607 /* only skip sriov related */
1608 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1609 if (dev->resource[i].flags & flags)
1610 bars |= (1 << i);
1611 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1612 if (dev->resource[i].flags & flags)
1613 bars |= (1 << i);
1615 err = do_pci_enable_device(dev, bars);
1616 if (err < 0)
1617 atomic_dec(&dev->enable_cnt);
1618 return err;
1622 * pci_enable_device_io - Initialize a device for use with IO space
1623 * @dev: PCI device to be initialized
1625 * Initialize device before it's used by a driver. Ask low-level code
1626 * to enable I/O resources. Wake up the device if it was suspended.
1627 * Beware, this function can fail.
1629 int pci_enable_device_io(struct pci_dev *dev)
1631 return pci_enable_device_flags(dev, IORESOURCE_IO);
1633 EXPORT_SYMBOL(pci_enable_device_io);
1636 * pci_enable_device_mem - Initialize a device for use with Memory space
1637 * @dev: PCI device to be initialized
1639 * Initialize device before it's used by a driver. Ask low-level code
1640 * to enable Memory resources. Wake up the device if it was suspended.
1641 * Beware, this function can fail.
1643 int pci_enable_device_mem(struct pci_dev *dev)
1645 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1647 EXPORT_SYMBOL(pci_enable_device_mem);
1650 * pci_enable_device - Initialize device before it's used by a driver.
1651 * @dev: PCI device to be initialized
1653 * Initialize device before it's used by a driver. Ask low-level code
1654 * to enable I/O and memory. Wake up the device if it was suspended.
1655 * Beware, this function can fail.
1657 * Note we don't actually enable the device many times if we call
1658 * this function repeatedly (we just increment the count).
1660 int pci_enable_device(struct pci_dev *dev)
1662 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1664 EXPORT_SYMBOL(pci_enable_device);
1667 * Managed PCI resources. This manages device on/off, intx/msi/msix
1668 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1669 * there's no need to track it separately. pci_devres is initialized
1670 * when a device is enabled using managed PCI device enable interface.
1672 struct pci_devres {
1673 unsigned int enabled:1;
1674 unsigned int pinned:1;
1675 unsigned int orig_intx:1;
1676 unsigned int restore_intx:1;
1677 unsigned int mwi:1;
1678 u32 region_mask;
1681 static void pcim_release(struct device *gendev, void *res)
1683 struct pci_dev *dev = to_pci_dev(gendev);
1684 struct pci_devres *this = res;
1685 int i;
1687 if (dev->msi_enabled)
1688 pci_disable_msi(dev);
1689 if (dev->msix_enabled)
1690 pci_disable_msix(dev);
1692 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1693 if (this->region_mask & (1 << i))
1694 pci_release_region(dev, i);
1696 if (this->mwi)
1697 pci_clear_mwi(dev);
1699 if (this->restore_intx)
1700 pci_intx(dev, this->orig_intx);
1702 if (this->enabled && !this->pinned)
1703 pci_disable_device(dev);
1706 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1708 struct pci_devres *dr, *new_dr;
1710 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1711 if (dr)
1712 return dr;
1714 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1715 if (!new_dr)
1716 return NULL;
1717 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1720 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1722 if (pci_is_managed(pdev))
1723 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1724 return NULL;
1728 * pcim_enable_device - Managed pci_enable_device()
1729 * @pdev: PCI device to be initialized
1731 * Managed pci_enable_device().
1733 int pcim_enable_device(struct pci_dev *pdev)
1735 struct pci_devres *dr;
1736 int rc;
1738 dr = get_pci_dr(pdev);
1739 if (unlikely(!dr))
1740 return -ENOMEM;
1741 if (dr->enabled)
1742 return 0;
1744 rc = pci_enable_device(pdev);
1745 if (!rc) {
1746 pdev->is_managed = 1;
1747 dr->enabled = 1;
1749 return rc;
1751 EXPORT_SYMBOL(pcim_enable_device);
1754 * pcim_pin_device - Pin managed PCI device
1755 * @pdev: PCI device to pin
1757 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1758 * driver detach. @pdev must have been enabled with
1759 * pcim_enable_device().
1761 void pcim_pin_device(struct pci_dev *pdev)
1763 struct pci_devres *dr;
1765 dr = find_pci_dr(pdev);
1766 WARN_ON(!dr || !dr->enabled);
1767 if (dr)
1768 dr->pinned = 1;
1770 EXPORT_SYMBOL(pcim_pin_device);
1773 * pcibios_add_device - provide arch specific hooks when adding device dev
1774 * @dev: the PCI device being added
1776 * Permits the platform to provide architecture specific functionality when
1777 * devices are added. This is the default implementation. Architecture
1778 * implementations can override this.
1780 int __weak pcibios_add_device(struct pci_dev *dev)
1782 return 0;
1786 * pcibios_release_device - provide arch specific hooks when releasing device dev
1787 * @dev: the PCI device being released
1789 * Permits the platform to provide architecture specific functionality when
1790 * devices are released. This is the default implementation. Architecture
1791 * implementations can override this.
1793 void __weak pcibios_release_device(struct pci_dev *dev) {}
1796 * pcibios_disable_device - disable arch specific PCI resources for device dev
1797 * @dev: the PCI device to disable
1799 * Disables architecture specific PCI resources for the device. This
1800 * is the default implementation. Architecture implementations can
1801 * override this.
1803 void __weak pcibios_disable_device(struct pci_dev *dev) {}
1806 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1807 * @irq: ISA IRQ to penalize
1808 * @active: IRQ active or not
1810 * Permits the platform to provide architecture-specific functionality when
1811 * penalizing ISA IRQs. This is the default implementation. Architecture
1812 * implementations can override this.
1814 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1816 static void do_pci_disable_device(struct pci_dev *dev)
1818 u16 pci_command;
1820 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1821 if (pci_command & PCI_COMMAND_MASTER) {
1822 pci_command &= ~PCI_COMMAND_MASTER;
1823 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1826 pcibios_disable_device(dev);
1830 * pci_disable_enabled_device - Disable device without updating enable_cnt
1831 * @dev: PCI device to disable
1833 * NOTE: This function is a backend of PCI power management routines and is
1834 * not supposed to be called drivers.
1836 void pci_disable_enabled_device(struct pci_dev *dev)
1838 if (pci_is_enabled(dev))
1839 do_pci_disable_device(dev);
1843 * pci_disable_device - Disable PCI device after use
1844 * @dev: PCI device to be disabled
1846 * Signal to the system that the PCI device is not in use by the system
1847 * anymore. This only involves disabling PCI bus-mastering, if active.
1849 * Note we don't actually disable the device until all callers of
1850 * pci_enable_device() have called pci_disable_device().
1852 void pci_disable_device(struct pci_dev *dev)
1854 struct pci_devres *dr;
1856 dr = find_pci_dr(dev);
1857 if (dr)
1858 dr->enabled = 0;
1860 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1861 "disabling already-disabled device");
1863 if (atomic_dec_return(&dev->enable_cnt) != 0)
1864 return;
1866 do_pci_disable_device(dev);
1868 dev->is_busmaster = 0;
1870 EXPORT_SYMBOL(pci_disable_device);
1873 * pcibios_set_pcie_reset_state - set reset state for device dev
1874 * @dev: the PCIe device reset
1875 * @state: Reset state to enter into
1878 * Sets the PCIe reset state for the device. This is the default
1879 * implementation. Architecture implementations can override this.
1881 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1882 enum pcie_reset_state state)
1884 return -EINVAL;
1888 * pci_set_pcie_reset_state - set reset state for device dev
1889 * @dev: the PCIe device reset
1890 * @state: Reset state to enter into
1893 * Sets the PCI reset state for the device.
1895 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1897 return pcibios_set_pcie_reset_state(dev, state);
1899 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1902 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
1903 * @dev: PCIe root port or event collector.
1905 void pcie_clear_root_pme_status(struct pci_dev *dev)
1907 pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
1911 * pci_check_pme_status - Check if given device has generated PME.
1912 * @dev: Device to check.
1914 * Check the PME status of the device and if set, clear it and clear PME enable
1915 * (if set). Return 'true' if PME status and PME enable were both set or
1916 * 'false' otherwise.
1918 bool pci_check_pme_status(struct pci_dev *dev)
1920 int pmcsr_pos;
1921 u16 pmcsr;
1922 bool ret = false;
1924 if (!dev->pm_cap)
1925 return false;
1927 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1928 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1929 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1930 return false;
1932 /* Clear PME status. */
1933 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1934 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1935 /* Disable PME to avoid interrupt flood. */
1936 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1937 ret = true;
1940 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1942 return ret;
1946 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1947 * @dev: Device to handle.
1948 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1950 * Check if @dev has generated PME and queue a resume request for it in that
1951 * case.
1953 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1955 if (pme_poll_reset && dev->pme_poll)
1956 dev->pme_poll = false;
1958 if (pci_check_pme_status(dev)) {
1959 pci_wakeup_event(dev);
1960 pm_request_resume(&dev->dev);
1962 return 0;
1966 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1967 * @bus: Top bus of the subtree to walk.
1969 void pci_pme_wakeup_bus(struct pci_bus *bus)
1971 if (bus)
1972 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1977 * pci_pme_capable - check the capability of PCI device to generate PME#
1978 * @dev: PCI device to handle.
1979 * @state: PCI state from which device will issue PME#.
1981 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1983 if (!dev->pm_cap)
1984 return false;
1986 return !!(dev->pme_support & (1 << state));
1988 EXPORT_SYMBOL(pci_pme_capable);
1990 static void pci_pme_list_scan(struct work_struct *work)
1992 struct pci_pme_device *pme_dev, *n;
1994 mutex_lock(&pci_pme_list_mutex);
1995 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1996 if (pme_dev->dev->pme_poll) {
1997 struct pci_dev *bridge;
1999 bridge = pme_dev->dev->bus->self;
2001 * If bridge is in low power state, the
2002 * configuration space of subordinate devices
2003 * may be not accessible
2005 if (bridge && bridge->current_state != PCI_D0)
2006 continue;
2007 pci_pme_wakeup(pme_dev->dev, NULL);
2008 } else {
2009 list_del(&pme_dev->list);
2010 kfree(pme_dev);
2013 if (!list_empty(&pci_pme_list))
2014 queue_delayed_work(system_freezable_wq, &pci_pme_work,
2015 msecs_to_jiffies(PME_TIMEOUT));
2016 mutex_unlock(&pci_pme_list_mutex);
2019 static void __pci_pme_active(struct pci_dev *dev, bool enable)
2021 u16 pmcsr;
2023 if (!dev->pme_support)
2024 return;
2026 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2027 /* Clear PME_Status by writing 1 to it and enable PME# */
2028 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2029 if (!enable)
2030 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2032 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2036 * pci_pme_restore - Restore PME configuration after config space restore.
2037 * @dev: PCI device to update.
2039 void pci_pme_restore(struct pci_dev *dev)
2041 u16 pmcsr;
2043 if (!dev->pme_support)
2044 return;
2046 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2047 if (dev->wakeup_prepared) {
2048 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2049 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2050 } else {
2051 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2052 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2054 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2058 * pci_pme_active - enable or disable PCI device's PME# function
2059 * @dev: PCI device to handle.
2060 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2062 * The caller must verify that the device is capable of generating PME# before
2063 * calling this function with @enable equal to 'true'.
2065 void pci_pme_active(struct pci_dev *dev, bool enable)
2067 __pci_pme_active(dev, enable);
2070 * PCI (as opposed to PCIe) PME requires that the device have
2071 * its PME# line hooked up correctly. Not all hardware vendors
2072 * do this, so the PME never gets delivered and the device
2073 * remains asleep. The easiest way around this is to
2074 * periodically walk the list of suspended devices and check
2075 * whether any have their PME flag set. The assumption is that
2076 * we'll wake up often enough anyway that this won't be a huge
2077 * hit, and the power savings from the devices will still be a
2078 * win.
2080 * Although PCIe uses in-band PME message instead of PME# line
2081 * to report PME, PME does not work for some PCIe devices in
2082 * reality. For example, there are devices that set their PME
2083 * status bits, but don't really bother to send a PME message;
2084 * there are PCI Express Root Ports that don't bother to
2085 * trigger interrupts when they receive PME messages from the
2086 * devices below. So PME poll is used for PCIe devices too.
2089 if (dev->pme_poll) {
2090 struct pci_pme_device *pme_dev;
2091 if (enable) {
2092 pme_dev = kmalloc(sizeof(struct pci_pme_device),
2093 GFP_KERNEL);
2094 if (!pme_dev) {
2095 pci_warn(dev, "can't enable PME#\n");
2096 return;
2098 pme_dev->dev = dev;
2099 mutex_lock(&pci_pme_list_mutex);
2100 list_add(&pme_dev->list, &pci_pme_list);
2101 if (list_is_singular(&pci_pme_list))
2102 queue_delayed_work(system_freezable_wq,
2103 &pci_pme_work,
2104 msecs_to_jiffies(PME_TIMEOUT));
2105 mutex_unlock(&pci_pme_list_mutex);
2106 } else {
2107 mutex_lock(&pci_pme_list_mutex);
2108 list_for_each_entry(pme_dev, &pci_pme_list, list) {
2109 if (pme_dev->dev == dev) {
2110 list_del(&pme_dev->list);
2111 kfree(pme_dev);
2112 break;
2115 mutex_unlock(&pci_pme_list_mutex);
2119 pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2121 EXPORT_SYMBOL(pci_pme_active);
2124 * __pci_enable_wake - enable PCI device as wakeup event source
2125 * @dev: PCI device affected
2126 * @state: PCI state from which device will issue wakeup events
2127 * @enable: True to enable event generation; false to disable
2129 * This enables the device as a wakeup event source, or disables it.
2130 * When such events involves platform-specific hooks, those hooks are
2131 * called automatically by this routine.
2133 * Devices with legacy power management (no standard PCI PM capabilities)
2134 * always require such platform hooks.
2136 * RETURN VALUE:
2137 * 0 is returned on success
2138 * -EINVAL is returned if device is not supposed to wake up the system
2139 * Error code depending on the platform is returned if both the platform and
2140 * the native mechanism fail to enable the generation of wake-up events
2142 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2144 int ret = 0;
2147 * Bridges can only signal wakeup on behalf of subordinate devices,
2148 * but that is set up elsewhere, so skip them.
2150 if (pci_has_subordinate(dev))
2151 return 0;
2153 /* Don't do the same thing twice in a row for one device. */
2154 if (!!enable == !!dev->wakeup_prepared)
2155 return 0;
2158 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2159 * Anderson we should be doing PME# wake enable followed by ACPI wake
2160 * enable. To disable wake-up we call the platform first, for symmetry.
2163 if (enable) {
2164 int error;
2166 if (pci_pme_capable(dev, state))
2167 pci_pme_active(dev, true);
2168 else
2169 ret = 1;
2170 error = platform_pci_set_wakeup(dev, true);
2171 if (ret)
2172 ret = error;
2173 if (!ret)
2174 dev->wakeup_prepared = true;
2175 } else {
2176 platform_pci_set_wakeup(dev, false);
2177 pci_pme_active(dev, false);
2178 dev->wakeup_prepared = false;
2181 return ret;
2185 * pci_enable_wake - change wakeup settings for a PCI device
2186 * @pci_dev: Target device
2187 * @state: PCI state from which device will issue wakeup events
2188 * @enable: Whether or not to enable event generation
2190 * If @enable is set, check device_may_wakeup() for the device before calling
2191 * __pci_enable_wake() for it.
2193 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2195 if (enable && !device_may_wakeup(&pci_dev->dev))
2196 return -EINVAL;
2198 return __pci_enable_wake(pci_dev, state, enable);
2200 EXPORT_SYMBOL(pci_enable_wake);
2203 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2204 * @dev: PCI device to prepare
2205 * @enable: True to enable wake-up event generation; false to disable
2207 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2208 * and this function allows them to set that up cleanly - pci_enable_wake()
2209 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2210 * ordering constraints.
2212 * This function only returns error code if the device is not allowed to wake
2213 * up the system from sleep or it is not capable of generating PME# from both
2214 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2216 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2218 return pci_pme_capable(dev, PCI_D3cold) ?
2219 pci_enable_wake(dev, PCI_D3cold, enable) :
2220 pci_enable_wake(dev, PCI_D3hot, enable);
2222 EXPORT_SYMBOL(pci_wake_from_d3);
2225 * pci_target_state - find an appropriate low power state for a given PCI dev
2226 * @dev: PCI device
2227 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2229 * Use underlying platform code to find a supported low power state for @dev.
2230 * If the platform can't manage @dev, return the deepest state from which it
2231 * can generate wake events, based on any available PME info.
2233 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2235 pci_power_t target_state = PCI_D3hot;
2237 if (platform_pci_power_manageable(dev)) {
2239 * Call the platform to find the target state for the device.
2241 pci_power_t state = platform_pci_choose_state(dev);
2243 switch (state) {
2244 case PCI_POWER_ERROR:
2245 case PCI_UNKNOWN:
2246 break;
2247 case PCI_D1:
2248 case PCI_D2:
2249 if (pci_no_d1d2(dev))
2250 break;
2251 /* else: fall through */
2252 default:
2253 target_state = state;
2256 return target_state;
2259 if (!dev->pm_cap)
2260 target_state = PCI_D0;
2263 * If the device is in D3cold even though it's not power-manageable by
2264 * the platform, it may have been powered down by non-standard means.
2265 * Best to let it slumber.
2267 if (dev->current_state == PCI_D3cold)
2268 target_state = PCI_D3cold;
2270 if (wakeup) {
2272 * Find the deepest state from which the device can generate
2273 * PME#.
2275 if (dev->pme_support) {
2276 while (target_state
2277 && !(dev->pme_support & (1 << target_state)))
2278 target_state--;
2282 return target_state;
2286 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
2287 * @dev: Device to handle.
2289 * Choose the power state appropriate for the device depending on whether
2290 * it can wake up the system and/or is power manageable by the platform
2291 * (PCI_D3hot is the default) and put the device into that state.
2293 int pci_prepare_to_sleep(struct pci_dev *dev)
2295 bool wakeup = device_may_wakeup(&dev->dev);
2296 pci_power_t target_state = pci_target_state(dev, wakeup);
2297 int error;
2299 if (target_state == PCI_POWER_ERROR)
2300 return -EIO;
2302 pci_enable_wake(dev, target_state, wakeup);
2304 error = pci_set_power_state(dev, target_state);
2306 if (error)
2307 pci_enable_wake(dev, target_state, false);
2309 return error;
2311 EXPORT_SYMBOL(pci_prepare_to_sleep);
2314 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2315 * @dev: Device to handle.
2317 * Disable device's system wake-up capability and put it into D0.
2319 int pci_back_from_sleep(struct pci_dev *dev)
2321 pci_enable_wake(dev, PCI_D0, false);
2322 return pci_set_power_state(dev, PCI_D0);
2324 EXPORT_SYMBOL(pci_back_from_sleep);
2327 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2328 * @dev: PCI device being suspended.
2330 * Prepare @dev to generate wake-up events at run time and put it into a low
2331 * power state.
2333 int pci_finish_runtime_suspend(struct pci_dev *dev)
2335 pci_power_t target_state;
2336 int error;
2338 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2339 if (target_state == PCI_POWER_ERROR)
2340 return -EIO;
2342 dev->runtime_d3cold = target_state == PCI_D3cold;
2344 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2346 error = pci_set_power_state(dev, target_state);
2348 if (error) {
2349 pci_enable_wake(dev, target_state, false);
2350 dev->runtime_d3cold = false;
2353 return error;
2357 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2358 * @dev: Device to check.
2360 * Return true if the device itself is capable of generating wake-up events
2361 * (through the platform or using the native PCIe PME) or if the device supports
2362 * PME and one of its upstream bridges can generate wake-up events.
2364 bool pci_dev_run_wake(struct pci_dev *dev)
2366 struct pci_bus *bus = dev->bus;
2368 if (!dev->pme_support)
2369 return false;
2371 /* PME-capable in principle, but not from the target power state */
2372 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2373 return false;
2375 if (device_can_wakeup(&dev->dev))
2376 return true;
2378 while (bus->parent) {
2379 struct pci_dev *bridge = bus->self;
2381 if (device_can_wakeup(&bridge->dev))
2382 return true;
2384 bus = bus->parent;
2387 /* We have reached the root bus. */
2388 if (bus->bridge)
2389 return device_can_wakeup(bus->bridge);
2391 return false;
2393 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2396 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2397 * @pci_dev: Device to check.
2399 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2400 * reconfigured due to wakeup settings difference between system and runtime
2401 * suspend and the current power state of it is suitable for the upcoming
2402 * (system) transition.
2404 * If the device is not configured for system wakeup, disable PME for it before
2405 * returning 'true' to prevent it from waking up the system unnecessarily.
2407 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2409 struct device *dev = &pci_dev->dev;
2410 bool wakeup = device_may_wakeup(dev);
2412 if (!pm_runtime_suspended(dev)
2413 || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2414 || platform_pci_need_resume(pci_dev))
2415 return false;
2418 * At this point the device is good to go unless it's been configured
2419 * to generate PME at the runtime suspend time, but it is not supposed
2420 * to wake up the system. In that case, simply disable PME for it
2421 * (it will have to be re-enabled on exit from system resume).
2423 * If the device's power state is D3cold and the platform check above
2424 * hasn't triggered, the device's configuration is suitable and we don't
2425 * need to manipulate it at all.
2427 spin_lock_irq(&dev->power.lock);
2429 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2430 !wakeup)
2431 __pci_pme_active(pci_dev, false);
2433 spin_unlock_irq(&dev->power.lock);
2434 return true;
2438 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2439 * @pci_dev: Device to handle.
2441 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2442 * it might have been disabled during the prepare phase of system suspend if
2443 * the device was not configured for system wakeup.
2445 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2447 struct device *dev = &pci_dev->dev;
2449 if (!pci_dev_run_wake(pci_dev))
2450 return;
2452 spin_lock_irq(&dev->power.lock);
2454 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2455 __pci_pme_active(pci_dev, true);
2457 spin_unlock_irq(&dev->power.lock);
2460 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2462 struct device *dev = &pdev->dev;
2463 struct device *parent = dev->parent;
2465 if (parent)
2466 pm_runtime_get_sync(parent);
2467 pm_runtime_get_noresume(dev);
2469 * pdev->current_state is set to PCI_D3cold during suspending,
2470 * so wait until suspending completes
2472 pm_runtime_barrier(dev);
2474 * Only need to resume devices in D3cold, because config
2475 * registers are still accessible for devices suspended but
2476 * not in D3cold.
2478 if (pdev->current_state == PCI_D3cold)
2479 pm_runtime_resume(dev);
2482 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2484 struct device *dev = &pdev->dev;
2485 struct device *parent = dev->parent;
2487 pm_runtime_put(dev);
2488 if (parent)
2489 pm_runtime_put_sync(parent);
2493 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2494 * @bridge: Bridge to check
2496 * This function checks if it is possible to move the bridge to D3.
2497 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2499 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2501 if (!pci_is_pcie(bridge))
2502 return false;
2504 switch (pci_pcie_type(bridge)) {
2505 case PCI_EXP_TYPE_ROOT_PORT:
2506 case PCI_EXP_TYPE_UPSTREAM:
2507 case PCI_EXP_TYPE_DOWNSTREAM:
2508 if (pci_bridge_d3_disable)
2509 return false;
2512 * Hotplug ports handled by firmware in System Management Mode
2513 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2515 if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2516 return false;
2518 if (pci_bridge_d3_force)
2519 return true;
2521 /* Even the oldest 2010 Thunderbolt controller supports D3. */
2522 if (bridge->is_thunderbolt)
2523 return true;
2526 * Hotplug ports handled natively by the OS were not validated
2527 * by vendors for runtime D3 at least until 2018 because there
2528 * was no OS support.
2530 if (bridge->is_hotplug_bridge)
2531 return false;
2534 * It should be safe to put PCIe ports from 2015 or newer
2535 * to D3.
2537 if (dmi_get_bios_year() >= 2015)
2538 return true;
2539 break;
2542 return false;
2545 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2547 bool *d3cold_ok = data;
2549 if (/* The device needs to be allowed to go D3cold ... */
2550 dev->no_d3cold || !dev->d3cold_allowed ||
2552 /* ... and if it is wakeup capable to do so from D3cold. */
2553 (device_may_wakeup(&dev->dev) &&
2554 !pci_pme_capable(dev, PCI_D3cold)) ||
2556 /* If it is a bridge it must be allowed to go to D3. */
2557 !pci_power_manageable(dev))
2559 *d3cold_ok = false;
2561 return !*d3cold_ok;
2565 * pci_bridge_d3_update - Update bridge D3 capabilities
2566 * @dev: PCI device which is changed
2568 * Update upstream bridge PM capabilities accordingly depending on if the
2569 * device PM configuration was changed or the device is being removed. The
2570 * change is also propagated upstream.
2572 void pci_bridge_d3_update(struct pci_dev *dev)
2574 bool remove = !device_is_registered(&dev->dev);
2575 struct pci_dev *bridge;
2576 bool d3cold_ok = true;
2578 bridge = pci_upstream_bridge(dev);
2579 if (!bridge || !pci_bridge_d3_possible(bridge))
2580 return;
2583 * If D3 is currently allowed for the bridge, removing one of its
2584 * children won't change that.
2586 if (remove && bridge->bridge_d3)
2587 return;
2590 * If D3 is currently allowed for the bridge and a child is added or
2591 * changed, disallowance of D3 can only be caused by that child, so
2592 * we only need to check that single device, not any of its siblings.
2594 * If D3 is currently not allowed for the bridge, checking the device
2595 * first may allow us to skip checking its siblings.
2597 if (!remove)
2598 pci_dev_check_d3cold(dev, &d3cold_ok);
2601 * If D3 is currently not allowed for the bridge, this may be caused
2602 * either by the device being changed/removed or any of its siblings,
2603 * so we need to go through all children to find out if one of them
2604 * continues to block D3.
2606 if (d3cold_ok && !bridge->bridge_d3)
2607 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2608 &d3cold_ok);
2610 if (bridge->bridge_d3 != d3cold_ok) {
2611 bridge->bridge_d3 = d3cold_ok;
2612 /* Propagate change to upstream bridges */
2613 pci_bridge_d3_update(bridge);
2618 * pci_d3cold_enable - Enable D3cold for device
2619 * @dev: PCI device to handle
2621 * This function can be used in drivers to enable D3cold from the device
2622 * they handle. It also updates upstream PCI bridge PM capabilities
2623 * accordingly.
2625 void pci_d3cold_enable(struct pci_dev *dev)
2627 if (dev->no_d3cold) {
2628 dev->no_d3cold = false;
2629 pci_bridge_d3_update(dev);
2632 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2635 * pci_d3cold_disable - Disable D3cold for device
2636 * @dev: PCI device to handle
2638 * This function can be used in drivers to disable D3cold from the device
2639 * they handle. It also updates upstream PCI bridge PM capabilities
2640 * accordingly.
2642 void pci_d3cold_disable(struct pci_dev *dev)
2644 if (!dev->no_d3cold) {
2645 dev->no_d3cold = true;
2646 pci_bridge_d3_update(dev);
2649 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2652 * pci_pm_init - Initialize PM functions of given PCI device
2653 * @dev: PCI device to handle.
2655 void pci_pm_init(struct pci_dev *dev)
2657 int pm;
2658 u16 pmc;
2660 pm_runtime_forbid(&dev->dev);
2661 pm_runtime_set_active(&dev->dev);
2662 pm_runtime_enable(&dev->dev);
2663 device_enable_async_suspend(&dev->dev);
2664 dev->wakeup_prepared = false;
2666 dev->pm_cap = 0;
2667 dev->pme_support = 0;
2669 /* find PCI PM capability in list */
2670 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2671 if (!pm)
2672 return;
2673 /* Check device's ability to generate PME# */
2674 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2676 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2677 pci_err(dev, "unsupported PM cap regs version (%u)\n",
2678 pmc & PCI_PM_CAP_VER_MASK);
2679 return;
2682 dev->pm_cap = pm;
2683 dev->d3_delay = PCI_PM_D3_WAIT;
2684 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2685 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2686 dev->d3cold_allowed = true;
2688 dev->d1_support = false;
2689 dev->d2_support = false;
2690 if (!pci_no_d1d2(dev)) {
2691 if (pmc & PCI_PM_CAP_D1)
2692 dev->d1_support = true;
2693 if (pmc & PCI_PM_CAP_D2)
2694 dev->d2_support = true;
2696 if (dev->d1_support || dev->d2_support)
2697 pci_printk(KERN_DEBUG, dev, "supports%s%s\n",
2698 dev->d1_support ? " D1" : "",
2699 dev->d2_support ? " D2" : "");
2702 pmc &= PCI_PM_CAP_PME_MASK;
2703 if (pmc) {
2704 pci_printk(KERN_DEBUG, dev, "PME# supported from%s%s%s%s%s\n",
2705 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2706 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2707 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2708 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2709 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2710 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2711 dev->pme_poll = true;
2713 * Make device's PM flags reflect the wake-up capability, but
2714 * let the user space enable it to wake up the system as needed.
2716 device_set_wakeup_capable(&dev->dev, true);
2717 /* Disable the PME# generation functionality */
2718 pci_pme_active(dev, false);
2722 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2724 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2726 switch (prop) {
2727 case PCI_EA_P_MEM:
2728 case PCI_EA_P_VF_MEM:
2729 flags |= IORESOURCE_MEM;
2730 break;
2731 case PCI_EA_P_MEM_PREFETCH:
2732 case PCI_EA_P_VF_MEM_PREFETCH:
2733 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2734 break;
2735 case PCI_EA_P_IO:
2736 flags |= IORESOURCE_IO;
2737 break;
2738 default:
2739 return 0;
2742 return flags;
2745 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2746 u8 prop)
2748 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2749 return &dev->resource[bei];
2750 #ifdef CONFIG_PCI_IOV
2751 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2752 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2753 return &dev->resource[PCI_IOV_RESOURCES +
2754 bei - PCI_EA_BEI_VF_BAR0];
2755 #endif
2756 else if (bei == PCI_EA_BEI_ROM)
2757 return &dev->resource[PCI_ROM_RESOURCE];
2758 else
2759 return NULL;
2762 /* Read an Enhanced Allocation (EA) entry */
2763 static int pci_ea_read(struct pci_dev *dev, int offset)
2765 struct resource *res;
2766 int ent_size, ent_offset = offset;
2767 resource_size_t start, end;
2768 unsigned long flags;
2769 u32 dw0, bei, base, max_offset;
2770 u8 prop;
2771 bool support_64 = (sizeof(resource_size_t) >= 8);
2773 pci_read_config_dword(dev, ent_offset, &dw0);
2774 ent_offset += 4;
2776 /* Entry size field indicates DWORDs after 1st */
2777 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2779 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2780 goto out;
2782 bei = (dw0 & PCI_EA_BEI) >> 4;
2783 prop = (dw0 & PCI_EA_PP) >> 8;
2786 * If the Property is in the reserved range, try the Secondary
2787 * Property instead.
2789 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2790 prop = (dw0 & PCI_EA_SP) >> 16;
2791 if (prop > PCI_EA_P_BRIDGE_IO)
2792 goto out;
2794 res = pci_ea_get_resource(dev, bei, prop);
2795 if (!res) {
2796 pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
2797 goto out;
2800 flags = pci_ea_flags(dev, prop);
2801 if (!flags) {
2802 pci_err(dev, "Unsupported EA properties: %#x\n", prop);
2803 goto out;
2806 /* Read Base */
2807 pci_read_config_dword(dev, ent_offset, &base);
2808 start = (base & PCI_EA_FIELD_MASK);
2809 ent_offset += 4;
2811 /* Read MaxOffset */
2812 pci_read_config_dword(dev, ent_offset, &max_offset);
2813 ent_offset += 4;
2815 /* Read Base MSBs (if 64-bit entry) */
2816 if (base & PCI_EA_IS_64) {
2817 u32 base_upper;
2819 pci_read_config_dword(dev, ent_offset, &base_upper);
2820 ent_offset += 4;
2822 flags |= IORESOURCE_MEM_64;
2824 /* entry starts above 32-bit boundary, can't use */
2825 if (!support_64 && base_upper)
2826 goto out;
2828 if (support_64)
2829 start |= ((u64)base_upper << 32);
2832 end = start + (max_offset | 0x03);
2834 /* Read MaxOffset MSBs (if 64-bit entry) */
2835 if (max_offset & PCI_EA_IS_64) {
2836 u32 max_offset_upper;
2838 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2839 ent_offset += 4;
2841 flags |= IORESOURCE_MEM_64;
2843 /* entry too big, can't use */
2844 if (!support_64 && max_offset_upper)
2845 goto out;
2847 if (support_64)
2848 end += ((u64)max_offset_upper << 32);
2851 if (end < start) {
2852 pci_err(dev, "EA Entry crosses address boundary\n");
2853 goto out;
2856 if (ent_size != ent_offset - offset) {
2857 pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
2858 ent_size, ent_offset - offset);
2859 goto out;
2862 res->name = pci_name(dev);
2863 res->start = start;
2864 res->end = end;
2865 res->flags = flags;
2867 if (bei <= PCI_EA_BEI_BAR5)
2868 pci_printk(KERN_DEBUG, dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2869 bei, res, prop);
2870 else if (bei == PCI_EA_BEI_ROM)
2871 pci_printk(KERN_DEBUG, dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2872 res, prop);
2873 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2874 pci_printk(KERN_DEBUG, dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2875 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2876 else
2877 pci_printk(KERN_DEBUG, dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2878 bei, res, prop);
2880 out:
2881 return offset + ent_size;
2884 /* Enhanced Allocation Initialization */
2885 void pci_ea_init(struct pci_dev *dev)
2887 int ea;
2888 u8 num_ent;
2889 int offset;
2890 int i;
2892 /* find PCI EA capability in list */
2893 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2894 if (!ea)
2895 return;
2897 /* determine the number of entries */
2898 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2899 &num_ent);
2900 num_ent &= PCI_EA_NUM_ENT_MASK;
2902 offset = ea + PCI_EA_FIRST_ENT;
2904 /* Skip DWORD 2 for type 1 functions */
2905 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2906 offset += 4;
2908 /* parse each EA entry */
2909 for (i = 0; i < num_ent; ++i)
2910 offset = pci_ea_read(dev, offset);
2913 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2914 struct pci_cap_saved_state *new_cap)
2916 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2920 * _pci_add_cap_save_buffer - allocate buffer for saving given
2921 * capability registers
2922 * @dev: the PCI device
2923 * @cap: the capability to allocate the buffer for
2924 * @extended: Standard or Extended capability ID
2925 * @size: requested size of the buffer
2927 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2928 bool extended, unsigned int size)
2930 int pos;
2931 struct pci_cap_saved_state *save_state;
2933 if (extended)
2934 pos = pci_find_ext_capability(dev, cap);
2935 else
2936 pos = pci_find_capability(dev, cap);
2938 if (!pos)
2939 return 0;
2941 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2942 if (!save_state)
2943 return -ENOMEM;
2945 save_state->cap.cap_nr = cap;
2946 save_state->cap.cap_extended = extended;
2947 save_state->cap.size = size;
2948 pci_add_saved_cap(dev, save_state);
2950 return 0;
2953 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2955 return _pci_add_cap_save_buffer(dev, cap, false, size);
2958 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2960 return _pci_add_cap_save_buffer(dev, cap, true, size);
2964 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2965 * @dev: the PCI device
2967 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2969 int error;
2971 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2972 PCI_EXP_SAVE_REGS * sizeof(u16));
2973 if (error)
2974 pci_err(dev, "unable to preallocate PCI Express save buffer\n");
2976 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2977 if (error)
2978 pci_err(dev, "unable to preallocate PCI-X save buffer\n");
2980 pci_allocate_vc_save_buffers(dev);
2983 void pci_free_cap_save_buffers(struct pci_dev *dev)
2985 struct pci_cap_saved_state *tmp;
2986 struct hlist_node *n;
2988 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2989 kfree(tmp);
2993 * pci_configure_ari - enable or disable ARI forwarding
2994 * @dev: the PCI device
2996 * If @dev and its upstream bridge both support ARI, enable ARI in the
2997 * bridge. Otherwise, disable ARI in the bridge.
2999 void pci_configure_ari(struct pci_dev *dev)
3001 u32 cap;
3002 struct pci_dev *bridge;
3004 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3005 return;
3007 bridge = dev->bus->self;
3008 if (!bridge)
3009 return;
3011 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3012 if (!(cap & PCI_EXP_DEVCAP2_ARI))
3013 return;
3015 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3016 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3017 PCI_EXP_DEVCTL2_ARI);
3018 bridge->ari_enabled = 1;
3019 } else {
3020 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3021 PCI_EXP_DEVCTL2_ARI);
3022 bridge->ari_enabled = 0;
3026 static int pci_acs_enable;
3029 * pci_request_acs - ask for ACS to be enabled if supported
3031 void pci_request_acs(void)
3033 pci_acs_enable = 1;
3036 static const char *disable_acs_redir_param;
3039 * pci_disable_acs_redir - disable ACS redirect capabilities
3040 * @dev: the PCI device
3042 * For only devices specified in the disable_acs_redir parameter.
3044 static void pci_disable_acs_redir(struct pci_dev *dev)
3046 int ret = 0;
3047 const char *p;
3048 int pos;
3049 u16 ctrl;
3051 if (!disable_acs_redir_param)
3052 return;
3054 p = disable_acs_redir_param;
3055 while (*p) {
3056 ret = pci_dev_str_match(dev, p, &p);
3057 if (ret < 0) {
3058 pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
3059 disable_acs_redir_param);
3061 break;
3062 } else if (ret == 1) {
3063 /* Found a match */
3064 break;
3067 if (*p != ';' && *p != ',') {
3068 /* End of param or invalid format */
3069 break;
3071 p++;
3074 if (ret != 1)
3075 return;
3077 if (!pci_dev_specific_disable_acs_redir(dev))
3078 return;
3080 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3081 if (!pos) {
3082 pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
3083 return;
3086 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3088 /* P2P Request & Completion Redirect */
3089 ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
3091 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3093 pci_info(dev, "disabled ACS redirect\n");
3097 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
3098 * @dev: the PCI device
3100 static void pci_std_enable_acs(struct pci_dev *dev)
3102 int pos;
3103 u16 cap;
3104 u16 ctrl;
3106 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3107 if (!pos)
3108 return;
3110 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
3111 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3113 /* Source Validation */
3114 ctrl |= (cap & PCI_ACS_SV);
3116 /* P2P Request Redirect */
3117 ctrl |= (cap & PCI_ACS_RR);
3119 /* P2P Completion Redirect */
3120 ctrl |= (cap & PCI_ACS_CR);
3122 /* Upstream Forwarding */
3123 ctrl |= (cap & PCI_ACS_UF);
3125 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3129 * pci_enable_acs - enable ACS if hardware support it
3130 * @dev: the PCI device
3132 void pci_enable_acs(struct pci_dev *dev)
3134 if (!pci_acs_enable)
3135 goto disable_acs_redir;
3137 if (!pci_dev_specific_enable_acs(dev))
3138 goto disable_acs_redir;
3140 pci_std_enable_acs(dev);
3142 disable_acs_redir:
3144 * Note: pci_disable_acs_redir() must be called even if ACS was not
3145 * enabled by the kernel because it may have been enabled by
3146 * platform firmware. So if we are told to disable it, we should
3147 * always disable it after setting the kernel's default
3148 * preferences.
3150 pci_disable_acs_redir(dev);
3153 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3155 int pos;
3156 u16 cap, ctrl;
3158 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
3159 if (!pos)
3160 return false;
3163 * Except for egress control, capabilities are either required
3164 * or only required if controllable. Features missing from the
3165 * capability field can therefore be assumed as hard-wired enabled.
3167 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3168 acs_flags &= (cap | PCI_ACS_EC);
3170 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3171 return (ctrl & acs_flags) == acs_flags;
3175 * pci_acs_enabled - test ACS against required flags for a given device
3176 * @pdev: device to test
3177 * @acs_flags: required PCI ACS flags
3179 * Return true if the device supports the provided flags. Automatically
3180 * filters out flags that are not implemented on multifunction devices.
3182 * Note that this interface checks the effective ACS capabilities of the
3183 * device rather than the actual capabilities. For instance, most single
3184 * function endpoints are not required to support ACS because they have no
3185 * opportunity for peer-to-peer access. We therefore return 'true'
3186 * regardless of whether the device exposes an ACS capability. This makes
3187 * it much easier for callers of this function to ignore the actual type
3188 * or topology of the device when testing ACS support.
3190 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3192 int ret;
3194 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3195 if (ret >= 0)
3196 return ret > 0;
3199 * Conventional PCI and PCI-X devices never support ACS, either
3200 * effectively or actually. The shared bus topology implies that
3201 * any device on the bus can receive or snoop DMA.
3203 if (!pci_is_pcie(pdev))
3204 return false;
3206 switch (pci_pcie_type(pdev)) {
3208 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3209 * but since their primary interface is PCI/X, we conservatively
3210 * handle them as we would a non-PCIe device.
3212 case PCI_EXP_TYPE_PCIE_BRIDGE:
3214 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
3215 * applicable... must never implement an ACS Extended Capability...".
3216 * This seems arbitrary, but we take a conservative interpretation
3217 * of this statement.
3219 case PCI_EXP_TYPE_PCI_BRIDGE:
3220 case PCI_EXP_TYPE_RC_EC:
3221 return false;
3223 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3224 * implement ACS in order to indicate their peer-to-peer capabilities,
3225 * regardless of whether they are single- or multi-function devices.
3227 case PCI_EXP_TYPE_DOWNSTREAM:
3228 case PCI_EXP_TYPE_ROOT_PORT:
3229 return pci_acs_flags_enabled(pdev, acs_flags);
3231 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3232 * implemented by the remaining PCIe types to indicate peer-to-peer
3233 * capabilities, but only when they are part of a multifunction
3234 * device. The footnote for section 6.12 indicates the specific
3235 * PCIe types included here.
3237 case PCI_EXP_TYPE_ENDPOINT:
3238 case PCI_EXP_TYPE_UPSTREAM:
3239 case PCI_EXP_TYPE_LEG_END:
3240 case PCI_EXP_TYPE_RC_END:
3241 if (!pdev->multifunction)
3242 break;
3244 return pci_acs_flags_enabled(pdev, acs_flags);
3248 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3249 * to single function devices with the exception of downstream ports.
3251 return true;
3255 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3256 * @start: starting downstream device
3257 * @end: ending upstream device or NULL to search to the root bus
3258 * @acs_flags: required flags
3260 * Walk up a device tree from start to end testing PCI ACS support. If
3261 * any step along the way does not support the required flags, return false.
3263 bool pci_acs_path_enabled(struct pci_dev *start,
3264 struct pci_dev *end, u16 acs_flags)
3266 struct pci_dev *pdev, *parent = start;
3268 do {
3269 pdev = parent;
3271 if (!pci_acs_enabled(pdev, acs_flags))
3272 return false;
3274 if (pci_is_root_bus(pdev->bus))
3275 return (end == NULL);
3277 parent = pdev->bus->self;
3278 } while (pdev != end);
3280 return true;
3284 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3285 * @pdev: PCI device
3286 * @bar: BAR to find
3288 * Helper to find the position of the ctrl register for a BAR.
3289 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3290 * Returns -ENOENT if no ctrl register for the BAR could be found.
3292 static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3294 unsigned int pos, nbars, i;
3295 u32 ctrl;
3297 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3298 if (!pos)
3299 return -ENOTSUPP;
3301 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3302 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3303 PCI_REBAR_CTRL_NBAR_SHIFT;
3305 for (i = 0; i < nbars; i++, pos += 8) {
3306 int bar_idx;
3308 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3309 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3310 if (bar_idx == bar)
3311 return pos;
3314 return -ENOENT;
3318 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3319 * @pdev: PCI device
3320 * @bar: BAR to query
3322 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3323 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3325 u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3327 int pos;
3328 u32 cap;
3330 pos = pci_rebar_find_pos(pdev, bar);
3331 if (pos < 0)
3332 return 0;
3334 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3335 return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3339 * pci_rebar_get_current_size - get the current size of a BAR
3340 * @pdev: PCI device
3341 * @bar: BAR to set size to
3343 * Read the size of a BAR from the resizable BAR config.
3344 * Returns size if found or negative error code.
3346 int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3348 int pos;
3349 u32 ctrl;
3351 pos = pci_rebar_find_pos(pdev, bar);
3352 if (pos < 0)
3353 return pos;
3355 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3356 return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3360 * pci_rebar_set_size - set a new size for a BAR
3361 * @pdev: PCI device
3362 * @bar: BAR to set size to
3363 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3365 * Set the new size of a BAR as defined in the spec.
3366 * Returns zero if resizing was successful, error code otherwise.
3368 int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3370 int pos;
3371 u32 ctrl;
3373 pos = pci_rebar_find_pos(pdev, bar);
3374 if (pos < 0)
3375 return pos;
3377 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3378 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3379 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3380 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3381 return 0;
3385 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3386 * @dev: the PCI device
3387 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3388 * PCI_EXP_DEVCAP2_ATOMIC_COMP32
3389 * PCI_EXP_DEVCAP2_ATOMIC_COMP64
3390 * PCI_EXP_DEVCAP2_ATOMIC_COMP128
3392 * Return 0 if all upstream bridges support AtomicOp routing, egress
3393 * blocking is disabled on all upstream ports, and the root port supports
3394 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3395 * AtomicOp completion), or negative otherwise.
3397 int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3399 struct pci_bus *bus = dev->bus;
3400 struct pci_dev *bridge;
3401 u32 cap, ctl2;
3403 if (!pci_is_pcie(dev))
3404 return -EINVAL;
3407 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3408 * AtomicOp requesters. For now, we only support endpoints as
3409 * requesters and root ports as completers. No endpoints as
3410 * completers, and no peer-to-peer.
3413 switch (pci_pcie_type(dev)) {
3414 case PCI_EXP_TYPE_ENDPOINT:
3415 case PCI_EXP_TYPE_LEG_END:
3416 case PCI_EXP_TYPE_RC_END:
3417 break;
3418 default:
3419 return -EINVAL;
3422 while (bus->parent) {
3423 bridge = bus->self;
3425 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3427 switch (pci_pcie_type(bridge)) {
3428 /* Ensure switch ports support AtomicOp routing */
3429 case PCI_EXP_TYPE_UPSTREAM:
3430 case PCI_EXP_TYPE_DOWNSTREAM:
3431 if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3432 return -EINVAL;
3433 break;
3435 /* Ensure root port supports all the sizes we care about */
3436 case PCI_EXP_TYPE_ROOT_PORT:
3437 if ((cap & cap_mask) != cap_mask)
3438 return -EINVAL;
3439 break;
3442 /* Ensure upstream ports don't block AtomicOps on egress */
3443 if (!bridge->has_secondary_link) {
3444 pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3445 &ctl2);
3446 if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3447 return -EINVAL;
3450 bus = bus->parent;
3453 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3454 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3455 return 0;
3457 EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3460 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3461 * @dev: the PCI device
3462 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3464 * Perform INTx swizzling for a device behind one level of bridge. This is
3465 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3466 * behind bridges on add-in cards. For devices with ARI enabled, the slot
3467 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3468 * the PCI Express Base Specification, Revision 2.1)
3470 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3472 int slot;
3474 if (pci_ari_enabled(dev->bus))
3475 slot = 0;
3476 else
3477 slot = PCI_SLOT(dev->devfn);
3479 return (((pin - 1) + slot) % 4) + 1;
3482 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3484 u8 pin;
3486 pin = dev->pin;
3487 if (!pin)
3488 return -1;
3490 while (!pci_is_root_bus(dev->bus)) {
3491 pin = pci_swizzle_interrupt_pin(dev, pin);
3492 dev = dev->bus->self;
3494 *bridge = dev;
3495 return pin;
3499 * pci_common_swizzle - swizzle INTx all the way to root bridge
3500 * @dev: the PCI device
3501 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3503 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3504 * bridges all the way up to a PCI root bus.
3506 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3508 u8 pin = *pinp;
3510 while (!pci_is_root_bus(dev->bus)) {
3511 pin = pci_swizzle_interrupt_pin(dev, pin);
3512 dev = dev->bus->self;
3514 *pinp = pin;
3515 return PCI_SLOT(dev->devfn);
3517 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3520 * pci_release_region - Release a PCI bar
3521 * @pdev: PCI device whose resources were previously reserved by pci_request_region
3522 * @bar: BAR to release
3524 * Releases the PCI I/O and memory resources previously reserved by a
3525 * successful call to pci_request_region. Call this function only
3526 * after all use of the PCI regions has ceased.
3528 void pci_release_region(struct pci_dev *pdev, int bar)
3530 struct pci_devres *dr;
3532 if (pci_resource_len(pdev, bar) == 0)
3533 return;
3534 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3535 release_region(pci_resource_start(pdev, bar),
3536 pci_resource_len(pdev, bar));
3537 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3538 release_mem_region(pci_resource_start(pdev, bar),
3539 pci_resource_len(pdev, bar));
3541 dr = find_pci_dr(pdev);
3542 if (dr)
3543 dr->region_mask &= ~(1 << bar);
3545 EXPORT_SYMBOL(pci_release_region);
3548 * __pci_request_region - Reserved PCI I/O and memory resource
3549 * @pdev: PCI device whose resources are to be reserved
3550 * @bar: BAR to be reserved
3551 * @res_name: Name to be associated with resource.
3552 * @exclusive: whether the region access is exclusive or not
3554 * Mark the PCI region associated with PCI device @pdev BR @bar as
3555 * being reserved by owner @res_name. Do not access any
3556 * address inside the PCI regions unless this call returns
3557 * successfully.
3559 * If @exclusive is set, then the region is marked so that userspace
3560 * is explicitly not allowed to map the resource via /dev/mem or
3561 * sysfs MMIO access.
3563 * Returns 0 on success, or %EBUSY on error. A warning
3564 * message is also printed on failure.
3566 static int __pci_request_region(struct pci_dev *pdev, int bar,
3567 const char *res_name, int exclusive)
3569 struct pci_devres *dr;
3571 if (pci_resource_len(pdev, bar) == 0)
3572 return 0;
3574 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3575 if (!request_region(pci_resource_start(pdev, bar),
3576 pci_resource_len(pdev, bar), res_name))
3577 goto err_out;
3578 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3579 if (!__request_mem_region(pci_resource_start(pdev, bar),
3580 pci_resource_len(pdev, bar), res_name,
3581 exclusive))
3582 goto err_out;
3585 dr = find_pci_dr(pdev);
3586 if (dr)
3587 dr->region_mask |= 1 << bar;
3589 return 0;
3591 err_out:
3592 pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3593 &pdev->resource[bar]);
3594 return -EBUSY;
3598 * pci_request_region - Reserve PCI I/O and memory resource
3599 * @pdev: PCI device whose resources are to be reserved
3600 * @bar: BAR to be reserved
3601 * @res_name: Name to be associated with resource
3603 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3604 * being reserved by owner @res_name. Do not access any
3605 * address inside the PCI regions unless this call returns
3606 * successfully.
3608 * Returns 0 on success, or %EBUSY on error. A warning
3609 * message is also printed on failure.
3611 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3613 return __pci_request_region(pdev, bar, res_name, 0);
3615 EXPORT_SYMBOL(pci_request_region);
3618 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
3619 * @pdev: PCI device whose resources are to be reserved
3620 * @bar: BAR to be reserved
3621 * @res_name: Name to be associated with resource.
3623 * Mark the PCI region associated with PCI device @pdev BR @bar as
3624 * being reserved by owner @res_name. Do not access any
3625 * address inside the PCI regions unless this call returns
3626 * successfully.
3628 * Returns 0 on success, or %EBUSY on error. A warning
3629 * message is also printed on failure.
3631 * The key difference that _exclusive makes it that userspace is
3632 * explicitly not allowed to map the resource via /dev/mem or
3633 * sysfs.
3635 int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3636 const char *res_name)
3638 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3640 EXPORT_SYMBOL(pci_request_region_exclusive);
3643 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3644 * @pdev: PCI device whose resources were previously reserved
3645 * @bars: Bitmask of BARs to be released
3647 * Release selected PCI I/O and memory resources previously reserved.
3648 * Call this function only after all use of the PCI regions has ceased.
3650 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3652 int i;
3654 for (i = 0; i < 6; i++)
3655 if (bars & (1 << i))
3656 pci_release_region(pdev, i);
3658 EXPORT_SYMBOL(pci_release_selected_regions);
3660 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3661 const char *res_name, int excl)
3663 int i;
3665 for (i = 0; i < 6; i++)
3666 if (bars & (1 << i))
3667 if (__pci_request_region(pdev, i, res_name, excl))
3668 goto err_out;
3669 return 0;
3671 err_out:
3672 while (--i >= 0)
3673 if (bars & (1 << i))
3674 pci_release_region(pdev, i);
3676 return -EBUSY;
3681 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3682 * @pdev: PCI device whose resources are to be reserved
3683 * @bars: Bitmask of BARs to be requested
3684 * @res_name: Name to be associated with resource
3686 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3687 const char *res_name)
3689 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3691 EXPORT_SYMBOL(pci_request_selected_regions);
3693 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3694 const char *res_name)
3696 return __pci_request_selected_regions(pdev, bars, res_name,
3697 IORESOURCE_EXCLUSIVE);
3699 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3702 * pci_release_regions - Release reserved PCI I/O and memory resources
3703 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
3705 * Releases all PCI I/O and memory resources previously reserved by a
3706 * successful call to pci_request_regions. Call this function only
3707 * after all use of the PCI regions has ceased.
3710 void pci_release_regions(struct pci_dev *pdev)
3712 pci_release_selected_regions(pdev, (1 << 6) - 1);
3714 EXPORT_SYMBOL(pci_release_regions);
3717 * pci_request_regions - Reserved PCI I/O and memory resources
3718 * @pdev: PCI device whose resources are to be reserved
3719 * @res_name: Name to be associated with resource.
3721 * Mark all PCI regions associated with PCI device @pdev as
3722 * being reserved by owner @res_name. Do not access any
3723 * address inside the PCI regions unless this call returns
3724 * successfully.
3726 * Returns 0 on success, or %EBUSY on error. A warning
3727 * message is also printed on failure.
3729 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3731 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3733 EXPORT_SYMBOL(pci_request_regions);
3736 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3737 * @pdev: PCI device whose resources are to be reserved
3738 * @res_name: Name to be associated with resource.
3740 * Mark all PCI regions associated with PCI device @pdev as
3741 * being reserved by owner @res_name. Do not access any
3742 * address inside the PCI regions unless this call returns
3743 * successfully.
3745 * pci_request_regions_exclusive() will mark the region so that
3746 * /dev/mem and the sysfs MMIO access will not be allowed.
3748 * Returns 0 on success, or %EBUSY on error. A warning
3749 * message is also printed on failure.
3751 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3753 return pci_request_selected_regions_exclusive(pdev,
3754 ((1 << 6) - 1), res_name);
3756 EXPORT_SYMBOL(pci_request_regions_exclusive);
3759 * Record the PCI IO range (expressed as CPU physical address + size).
3760 * Return a negative value if an error has occured, zero otherwise
3762 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3763 resource_size_t size)
3765 int ret = 0;
3766 #ifdef PCI_IOBASE
3767 struct logic_pio_hwaddr *range;
3769 if (!size || addr + size < addr)
3770 return -EINVAL;
3772 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3773 if (!range)
3774 return -ENOMEM;
3776 range->fwnode = fwnode;
3777 range->size = size;
3778 range->hw_start = addr;
3779 range->flags = LOGIC_PIO_CPU_MMIO;
3781 ret = logic_pio_register_range(range);
3782 if (ret)
3783 kfree(range);
3784 #endif
3786 return ret;
3789 phys_addr_t pci_pio_to_address(unsigned long pio)
3791 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3793 #ifdef PCI_IOBASE
3794 if (pio >= MMIO_UPPER_LIMIT)
3795 return address;
3797 address = logic_pio_to_hwaddr(pio);
3798 #endif
3800 return address;
3803 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3805 #ifdef PCI_IOBASE
3806 return logic_pio_trans_cpuaddr(address);
3807 #else
3808 if (address > IO_SPACE_LIMIT)
3809 return (unsigned long)-1;
3811 return (unsigned long) address;
3812 #endif
3816 * pci_remap_iospace - Remap the memory mapped I/O space
3817 * @res: Resource describing the I/O space
3818 * @phys_addr: physical address of range to be mapped
3820 * Remap the memory mapped I/O space described by the @res
3821 * and the CPU physical address @phys_addr into virtual address space.
3822 * Only architectures that have memory mapped IO functions defined
3823 * (and the PCI_IOBASE value defined) should call this function.
3825 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3827 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3828 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3830 if (!(res->flags & IORESOURCE_IO))
3831 return -EINVAL;
3833 if (res->end > IO_SPACE_LIMIT)
3834 return -EINVAL;
3836 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3837 pgprot_device(PAGE_KERNEL));
3838 #else
3839 /* this architecture does not have memory mapped I/O space,
3840 so this function should never be called */
3841 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3842 return -ENODEV;
3843 #endif
3845 EXPORT_SYMBOL(pci_remap_iospace);
3848 * pci_unmap_iospace - Unmap the memory mapped I/O space
3849 * @res: resource to be unmapped
3851 * Unmap the CPU virtual address @res from virtual address space.
3852 * Only architectures that have memory mapped IO functions defined
3853 * (and the PCI_IOBASE value defined) should call this function.
3855 void pci_unmap_iospace(struct resource *res)
3857 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3858 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3860 unmap_kernel_range(vaddr, resource_size(res));
3861 #endif
3863 EXPORT_SYMBOL(pci_unmap_iospace);
3865 static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
3867 struct resource **res = ptr;
3869 pci_unmap_iospace(*res);
3873 * devm_pci_remap_iospace - Managed pci_remap_iospace()
3874 * @dev: Generic device to remap IO address for
3875 * @res: Resource describing the I/O space
3876 * @phys_addr: physical address of range to be mapped
3878 * Managed pci_remap_iospace(). Map is automatically unmapped on driver
3879 * detach.
3881 int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
3882 phys_addr_t phys_addr)
3884 const struct resource **ptr;
3885 int error;
3887 ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
3888 if (!ptr)
3889 return -ENOMEM;
3891 error = pci_remap_iospace(res, phys_addr);
3892 if (error) {
3893 devres_free(ptr);
3894 } else {
3895 *ptr = res;
3896 devres_add(dev, ptr);
3899 return error;
3901 EXPORT_SYMBOL(devm_pci_remap_iospace);
3904 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3905 * @dev: Generic device to remap IO address for
3906 * @offset: Resource address to map
3907 * @size: Size of map
3909 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
3910 * detach.
3912 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3913 resource_size_t offset,
3914 resource_size_t size)
3916 void __iomem **ptr, *addr;
3918 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
3919 if (!ptr)
3920 return NULL;
3922 addr = pci_remap_cfgspace(offset, size);
3923 if (addr) {
3924 *ptr = addr;
3925 devres_add(dev, ptr);
3926 } else
3927 devres_free(ptr);
3929 return addr;
3931 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
3934 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
3935 * @dev: generic device to handle the resource for
3936 * @res: configuration space resource to be handled
3938 * Checks that a resource is a valid memory region, requests the memory
3939 * region and ioremaps with pci_remap_cfgspace() API that ensures the
3940 * proper PCI configuration space memory attributes are guaranteed.
3942 * All operations are managed and will be undone on driver detach.
3944 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
3945 * on failure. Usage example::
3947 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
3948 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
3949 * if (IS_ERR(base))
3950 * return PTR_ERR(base);
3952 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
3953 struct resource *res)
3955 resource_size_t size;
3956 const char *name;
3957 void __iomem *dest_ptr;
3959 BUG_ON(!dev);
3961 if (!res || resource_type(res) != IORESOURCE_MEM) {
3962 dev_err(dev, "invalid resource\n");
3963 return IOMEM_ERR_PTR(-EINVAL);
3966 size = resource_size(res);
3967 name = res->name ?: dev_name(dev);
3969 if (!devm_request_mem_region(dev, res->start, size, name)) {
3970 dev_err(dev, "can't request region for resource %pR\n", res);
3971 return IOMEM_ERR_PTR(-EBUSY);
3974 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
3975 if (!dest_ptr) {
3976 dev_err(dev, "ioremap failed for resource %pR\n", res);
3977 devm_release_mem_region(dev, res->start, size);
3978 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
3981 return dest_ptr;
3983 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
3985 static void __pci_set_master(struct pci_dev *dev, bool enable)
3987 u16 old_cmd, cmd;
3989 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3990 if (enable)
3991 cmd = old_cmd | PCI_COMMAND_MASTER;
3992 else
3993 cmd = old_cmd & ~PCI_COMMAND_MASTER;
3994 if (cmd != old_cmd) {
3995 pci_dbg(dev, "%s bus mastering\n",
3996 enable ? "enabling" : "disabling");
3997 pci_write_config_word(dev, PCI_COMMAND, cmd);
3999 dev->is_busmaster = enable;
4003 * pcibios_setup - process "pci=" kernel boot arguments
4004 * @str: string used to pass in "pci=" kernel boot arguments
4006 * Process kernel boot arguments. This is the default implementation.
4007 * Architecture specific implementations can override this as necessary.
4009 char * __weak __init pcibios_setup(char *str)
4011 return str;
4015 * pcibios_set_master - enable PCI bus-mastering for device dev
4016 * @dev: the PCI device to enable
4018 * Enables PCI bus-mastering for the device. This is the default
4019 * implementation. Architecture specific implementations can override
4020 * this if necessary.
4022 void __weak pcibios_set_master(struct pci_dev *dev)
4024 u8 lat;
4026 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4027 if (pci_is_pcie(dev))
4028 return;
4030 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4031 if (lat < 16)
4032 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4033 else if (lat > pcibios_max_latency)
4034 lat = pcibios_max_latency;
4035 else
4036 return;
4038 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4042 * pci_set_master - enables bus-mastering for device dev
4043 * @dev: the PCI device to enable
4045 * Enables bus-mastering on the device and calls pcibios_set_master()
4046 * to do the needed arch specific settings.
4048 void pci_set_master(struct pci_dev *dev)
4050 __pci_set_master(dev, true);
4051 pcibios_set_master(dev);
4053 EXPORT_SYMBOL(pci_set_master);
4056 * pci_clear_master - disables bus-mastering for device dev
4057 * @dev: the PCI device to disable
4059 void pci_clear_master(struct pci_dev *dev)
4061 __pci_set_master(dev, false);
4063 EXPORT_SYMBOL(pci_clear_master);
4066 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4067 * @dev: the PCI device for which MWI is to be enabled
4069 * Helper function for pci_set_mwi.
4070 * Originally copied from drivers/net/acenic.c.
4071 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4073 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4075 int pci_set_cacheline_size(struct pci_dev *dev)
4077 u8 cacheline_size;
4079 if (!pci_cache_line_size)
4080 return -EINVAL;
4082 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4083 equal to or multiple of the right value. */
4084 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4085 if (cacheline_size >= pci_cache_line_size &&
4086 (cacheline_size % pci_cache_line_size) == 0)
4087 return 0;
4089 /* Write the correct value. */
4090 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4091 /* Read it back. */
4092 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4093 if (cacheline_size == pci_cache_line_size)
4094 return 0;
4096 pci_printk(KERN_DEBUG, dev, "cache line size of %d is not supported\n",
4097 pci_cache_line_size << 2);
4099 return -EINVAL;
4101 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4104 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4105 * @dev: the PCI device for which MWI is enabled
4107 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4109 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4111 int pci_set_mwi(struct pci_dev *dev)
4113 #ifdef PCI_DISABLE_MWI
4114 return 0;
4115 #else
4116 int rc;
4117 u16 cmd;
4119 rc = pci_set_cacheline_size(dev);
4120 if (rc)
4121 return rc;
4123 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4124 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4125 pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4126 cmd |= PCI_COMMAND_INVALIDATE;
4127 pci_write_config_word(dev, PCI_COMMAND, cmd);
4129 return 0;
4130 #endif
4132 EXPORT_SYMBOL(pci_set_mwi);
4135 * pcim_set_mwi - a device-managed pci_set_mwi()
4136 * @dev: the PCI device for which MWI is enabled
4138 * Managed pci_set_mwi().
4140 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4142 int pcim_set_mwi(struct pci_dev *dev)
4144 struct pci_devres *dr;
4146 dr = find_pci_dr(dev);
4147 if (!dr)
4148 return -ENOMEM;
4150 dr->mwi = 1;
4151 return pci_set_mwi(dev);
4153 EXPORT_SYMBOL(pcim_set_mwi);
4156 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4157 * @dev: the PCI device for which MWI is enabled
4159 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4160 * Callers are not required to check the return value.
4162 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4164 int pci_try_set_mwi(struct pci_dev *dev)
4166 #ifdef PCI_DISABLE_MWI
4167 return 0;
4168 #else
4169 return pci_set_mwi(dev);
4170 #endif
4172 EXPORT_SYMBOL(pci_try_set_mwi);
4175 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4176 * @dev: the PCI device to disable
4178 * Disables PCI Memory-Write-Invalidate transaction on the device
4180 void pci_clear_mwi(struct pci_dev *dev)
4182 #ifndef PCI_DISABLE_MWI
4183 u16 cmd;
4185 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4186 if (cmd & PCI_COMMAND_INVALIDATE) {
4187 cmd &= ~PCI_COMMAND_INVALIDATE;
4188 pci_write_config_word(dev, PCI_COMMAND, cmd);
4190 #endif
4192 EXPORT_SYMBOL(pci_clear_mwi);
4195 * pci_intx - enables/disables PCI INTx for device dev
4196 * @pdev: the PCI device to operate on
4197 * @enable: boolean: whether to enable or disable PCI INTx
4199 * Enables/disables PCI INTx for device dev
4201 void pci_intx(struct pci_dev *pdev, int enable)
4203 u16 pci_command, new;
4205 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4207 if (enable)
4208 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4209 else
4210 new = pci_command | PCI_COMMAND_INTX_DISABLE;
4212 if (new != pci_command) {
4213 struct pci_devres *dr;
4215 pci_write_config_word(pdev, PCI_COMMAND, new);
4217 dr = find_pci_dr(pdev);
4218 if (dr && !dr->restore_intx) {
4219 dr->restore_intx = 1;
4220 dr->orig_intx = !enable;
4224 EXPORT_SYMBOL_GPL(pci_intx);
4226 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4228 struct pci_bus *bus = dev->bus;
4229 bool mask_updated = true;
4230 u32 cmd_status_dword;
4231 u16 origcmd, newcmd;
4232 unsigned long flags;
4233 bool irq_pending;
4236 * We do a single dword read to retrieve both command and status.
4237 * Document assumptions that make this possible.
4239 BUILD_BUG_ON(PCI_COMMAND % 4);
4240 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4242 raw_spin_lock_irqsave(&pci_lock, flags);
4244 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4246 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4249 * Check interrupt status register to see whether our device
4250 * triggered the interrupt (when masking) or the next IRQ is
4251 * already pending (when unmasking).
4253 if (mask != irq_pending) {
4254 mask_updated = false;
4255 goto done;
4258 origcmd = cmd_status_dword;
4259 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4260 if (mask)
4261 newcmd |= PCI_COMMAND_INTX_DISABLE;
4262 if (newcmd != origcmd)
4263 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4265 done:
4266 raw_spin_unlock_irqrestore(&pci_lock, flags);
4268 return mask_updated;
4272 * pci_check_and_mask_intx - mask INTx on pending interrupt
4273 * @dev: the PCI device to operate on
4275 * Check if the device dev has its INTx line asserted, mask it and
4276 * return true in that case. False is returned if no interrupt was
4277 * pending.
4279 bool pci_check_and_mask_intx(struct pci_dev *dev)
4281 return pci_check_and_set_intx_mask(dev, true);
4283 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4286 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4287 * @dev: the PCI device to operate on
4289 * Check if the device dev has its INTx line asserted, unmask it if not
4290 * and return true. False is returned and the mask remains active if
4291 * there was still an interrupt pending.
4293 bool pci_check_and_unmask_intx(struct pci_dev *dev)
4295 return pci_check_and_set_intx_mask(dev, false);
4297 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4300 * pci_wait_for_pending_transaction - waits for pending transaction
4301 * @dev: the PCI device to operate on
4303 * Return 0 if transaction is pending 1 otherwise.
4305 int pci_wait_for_pending_transaction(struct pci_dev *dev)
4307 if (!pci_is_pcie(dev))
4308 return 1;
4310 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4311 PCI_EXP_DEVSTA_TRPND);
4313 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4315 static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
4317 int delay = 1;
4318 u32 id;
4321 * After reset, the device should not silently discard config
4322 * requests, but it may still indicate that it needs more time by
4323 * responding to them with CRS completions. The Root Port will
4324 * generally synthesize ~0 data to complete the read (except when
4325 * CRS SV is enabled and the read was for the Vendor ID; in that
4326 * case it synthesizes 0x0001 data).
4328 * Wait for the device to return a non-CRS completion. Read the
4329 * Command register instead of Vendor ID so we don't have to
4330 * contend with the CRS SV value.
4332 pci_read_config_dword(dev, PCI_COMMAND, &id);
4333 while (id == ~0) {
4334 if (delay > timeout) {
4335 pci_warn(dev, "not ready %dms after %s; giving up\n",
4336 delay - 1, reset_type);
4337 return -ENOTTY;
4340 if (delay > 1000)
4341 pci_info(dev, "not ready %dms after %s; waiting\n",
4342 delay - 1, reset_type);
4344 msleep(delay);
4345 delay *= 2;
4346 pci_read_config_dword(dev, PCI_COMMAND, &id);
4349 if (delay > 1000)
4350 pci_info(dev, "ready %dms after %s\n", delay - 1,
4351 reset_type);
4353 return 0;
4357 * pcie_has_flr - check if a device supports function level resets
4358 * @dev: device to check
4360 * Returns true if the device advertises support for PCIe function level
4361 * resets.
4363 bool pcie_has_flr(struct pci_dev *dev)
4365 u32 cap;
4367 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4368 return false;
4370 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4371 return cap & PCI_EXP_DEVCAP_FLR;
4373 EXPORT_SYMBOL_GPL(pcie_has_flr);
4376 * pcie_flr - initiate a PCIe function level reset
4377 * @dev: device to reset
4379 * Initiate a function level reset on @dev. The caller should ensure the
4380 * device supports FLR before calling this function, e.g. by using the
4381 * pcie_has_flr() helper.
4383 int pcie_flr(struct pci_dev *dev)
4385 if (!pci_wait_for_pending_transaction(dev))
4386 pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4388 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4391 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4392 * 100ms, but may silently discard requests while the FLR is in
4393 * progress. Wait 100ms before trying to access the device.
4395 msleep(100);
4397 return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4399 EXPORT_SYMBOL_GPL(pcie_flr);
4401 static int pci_af_flr(struct pci_dev *dev, int probe)
4403 int pos;
4404 u8 cap;
4406 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4407 if (!pos)
4408 return -ENOTTY;
4410 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4411 return -ENOTTY;
4413 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4414 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4415 return -ENOTTY;
4417 if (probe)
4418 return 0;
4421 * Wait for Transaction Pending bit to clear. A word-aligned test
4422 * is used, so we use the conrol offset rather than status and shift
4423 * the test bit to match.
4425 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4426 PCI_AF_STATUS_TP << 8))
4427 pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4429 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4432 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4433 * updated 27 July 2006; a device must complete an FLR within
4434 * 100ms, but may silently discard requests while the FLR is in
4435 * progress. Wait 100ms before trying to access the device.
4437 msleep(100);
4439 return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4443 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4444 * @dev: Device to reset.
4445 * @probe: If set, only check if the device can be reset this way.
4447 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4448 * unset, it will be reinitialized internally when going from PCI_D3hot to
4449 * PCI_D0. If that's the case and the device is not in a low-power state
4450 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4452 * NOTE: This causes the caller to sleep for twice the device power transition
4453 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4454 * by default (i.e. unless the @dev's d3_delay field has a different value).
4455 * Moreover, only devices in D0 can be reset by this function.
4457 static int pci_pm_reset(struct pci_dev *dev, int probe)
4459 u16 csr;
4461 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4462 return -ENOTTY;
4464 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4465 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4466 return -ENOTTY;
4468 if (probe)
4469 return 0;
4471 if (dev->current_state != PCI_D0)
4472 return -EINVAL;
4474 csr &= ~PCI_PM_CTRL_STATE_MASK;
4475 csr |= PCI_D3hot;
4476 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4477 pci_dev_d3_sleep(dev);
4479 csr &= ~PCI_PM_CTRL_STATE_MASK;
4480 csr |= PCI_D0;
4481 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4482 pci_dev_d3_sleep(dev);
4484 return pci_dev_wait(dev, "PM D3->D0", PCIE_RESET_READY_POLL_MS);
4487 * pcie_wait_for_link - Wait until link is active or inactive
4488 * @pdev: Bridge device
4489 * @active: waiting for active or inactive?
4491 * Use this to wait till link becomes active or inactive.
4493 bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4495 int timeout = 1000;
4496 bool ret;
4497 u16 lnk_status;
4499 for (;;) {
4500 pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4501 ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4502 if (ret == active)
4503 return true;
4504 if (timeout <= 0)
4505 break;
4506 msleep(10);
4507 timeout -= 10;
4510 pci_info(pdev, "Data Link Layer Link Active not %s in 1000 msec\n",
4511 active ? "set" : "cleared");
4513 return false;
4516 void pci_reset_secondary_bus(struct pci_dev *dev)
4518 u16 ctrl;
4520 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4521 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4522 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4525 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4526 * this to 2ms to ensure that we meet the minimum requirement.
4528 msleep(2);
4530 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4531 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4534 * Trhfa for conventional PCI is 2^25 clock cycles.
4535 * Assuming a minimum 33MHz clock this results in a 1s
4536 * delay before we can consider subordinate devices to
4537 * be re-initialized. PCIe has some ways to shorten this,
4538 * but we don't make use of them yet.
4540 ssleep(1);
4543 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4545 pci_reset_secondary_bus(dev);
4549 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4550 * @dev: Bridge device
4552 * Use the bridge control register to assert reset on the secondary bus.
4553 * Devices on the secondary bus are left in power-on state.
4555 int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4557 pcibios_reset_secondary_bus(dev);
4559 return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4561 EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4563 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4565 struct pci_dev *pdev;
4567 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4568 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4569 return -ENOTTY;
4571 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4572 if (pdev != dev)
4573 return -ENOTTY;
4575 if (probe)
4576 return 0;
4578 return pci_bridge_secondary_bus_reset(dev->bus->self);
4581 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4583 int rc = -ENOTTY;
4585 if (!hotplug || !try_module_get(hotplug->ops->owner))
4586 return rc;
4588 if (hotplug->ops->reset_slot)
4589 rc = hotplug->ops->reset_slot(hotplug, probe);
4591 module_put(hotplug->ops->owner);
4593 return rc;
4596 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4598 struct pci_dev *pdev;
4600 if (dev->subordinate || !dev->slot ||
4601 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4602 return -ENOTTY;
4604 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4605 if (pdev != dev && pdev->slot == dev->slot)
4606 return -ENOTTY;
4608 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4611 static void pci_dev_lock(struct pci_dev *dev)
4613 pci_cfg_access_lock(dev);
4614 /* block PM suspend, driver probe, etc. */
4615 device_lock(&dev->dev);
4618 /* Return 1 on successful lock, 0 on contention */
4619 static int pci_dev_trylock(struct pci_dev *dev)
4621 if (pci_cfg_access_trylock(dev)) {
4622 if (device_trylock(&dev->dev))
4623 return 1;
4624 pci_cfg_access_unlock(dev);
4627 return 0;
4630 static void pci_dev_unlock(struct pci_dev *dev)
4632 device_unlock(&dev->dev);
4633 pci_cfg_access_unlock(dev);
4636 static void pci_dev_save_and_disable(struct pci_dev *dev)
4638 const struct pci_error_handlers *err_handler =
4639 dev->driver ? dev->driver->err_handler : NULL;
4642 * dev->driver->err_handler->reset_prepare() is protected against
4643 * races with ->remove() by the device lock, which must be held by
4644 * the caller.
4646 if (err_handler && err_handler->reset_prepare)
4647 err_handler->reset_prepare(dev);
4650 * Wake-up device prior to save. PM registers default to D0 after
4651 * reset and a simple register restore doesn't reliably return
4652 * to a non-D0 state anyway.
4654 pci_set_power_state(dev, PCI_D0);
4656 pci_save_state(dev);
4658 * Disable the device by clearing the Command register, except for
4659 * INTx-disable which is set. This not only disables MMIO and I/O port
4660 * BARs, but also prevents the device from being Bus Master, preventing
4661 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4662 * compliant devices, INTx-disable prevents legacy interrupts.
4664 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4667 static void pci_dev_restore(struct pci_dev *dev)
4669 const struct pci_error_handlers *err_handler =
4670 dev->driver ? dev->driver->err_handler : NULL;
4672 pci_restore_state(dev);
4675 * dev->driver->err_handler->reset_done() is protected against
4676 * races with ->remove() by the device lock, which must be held by
4677 * the caller.
4679 if (err_handler && err_handler->reset_done)
4680 err_handler->reset_done(dev);
4684 * __pci_reset_function_locked - reset a PCI device function while holding
4685 * the @dev mutex lock.
4686 * @dev: PCI device to reset
4688 * Some devices allow an individual function to be reset without affecting
4689 * other functions in the same device. The PCI device must be responsive
4690 * to PCI config space in order to use this function.
4692 * The device function is presumed to be unused and the caller is holding
4693 * the device mutex lock when this function is called.
4694 * Resetting the device will make the contents of PCI configuration space
4695 * random, so any caller of this must be prepared to reinitialise the
4696 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4697 * etc.
4699 * Returns 0 if the device function was successfully reset or negative if the
4700 * device doesn't support resetting a single function.
4702 int __pci_reset_function_locked(struct pci_dev *dev)
4704 int rc;
4706 might_sleep();
4709 * A reset method returns -ENOTTY if it doesn't support this device
4710 * and we should try the next method.
4712 * If it returns 0 (success), we're finished. If it returns any
4713 * other error, we're also finished: this indicates that further
4714 * reset mechanisms might be broken on the device.
4716 rc = pci_dev_specific_reset(dev, 0);
4717 if (rc != -ENOTTY)
4718 return rc;
4719 if (pcie_has_flr(dev)) {
4720 rc = pcie_flr(dev);
4721 if (rc != -ENOTTY)
4722 return rc;
4724 rc = pci_af_flr(dev, 0);
4725 if (rc != -ENOTTY)
4726 return rc;
4727 rc = pci_pm_reset(dev, 0);
4728 if (rc != -ENOTTY)
4729 return rc;
4730 rc = pci_dev_reset_slot_function(dev, 0);
4731 if (rc != -ENOTTY)
4732 return rc;
4733 return pci_parent_bus_reset(dev, 0);
4735 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4738 * pci_probe_reset_function - check whether the device can be safely reset
4739 * @dev: PCI device to reset
4741 * Some devices allow an individual function to be reset without affecting
4742 * other functions in the same device. The PCI device must be responsive
4743 * to PCI config space in order to use this function.
4745 * Returns 0 if the device function can be reset or negative if the
4746 * device doesn't support resetting a single function.
4748 int pci_probe_reset_function(struct pci_dev *dev)
4750 int rc;
4752 might_sleep();
4754 rc = pci_dev_specific_reset(dev, 1);
4755 if (rc != -ENOTTY)
4756 return rc;
4757 if (pcie_has_flr(dev))
4758 return 0;
4759 rc = pci_af_flr(dev, 1);
4760 if (rc != -ENOTTY)
4761 return rc;
4762 rc = pci_pm_reset(dev, 1);
4763 if (rc != -ENOTTY)
4764 return rc;
4765 rc = pci_dev_reset_slot_function(dev, 1);
4766 if (rc != -ENOTTY)
4767 return rc;
4769 return pci_parent_bus_reset(dev, 1);
4773 * pci_reset_function - quiesce and reset a PCI device function
4774 * @dev: PCI device to reset
4776 * Some devices allow an individual function to be reset without affecting
4777 * other functions in the same device. The PCI device must be responsive
4778 * to PCI config space in order to use this function.
4780 * This function does not just reset the PCI portion of a device, but
4781 * clears all the state associated with the device. This function differs
4782 * from __pci_reset_function_locked() in that it saves and restores device state
4783 * over the reset and takes the PCI device lock.
4785 * Returns 0 if the device function was successfully reset or negative if the
4786 * device doesn't support resetting a single function.
4788 int pci_reset_function(struct pci_dev *dev)
4790 int rc;
4792 if (!dev->reset_fn)
4793 return -ENOTTY;
4795 pci_dev_lock(dev);
4796 pci_dev_save_and_disable(dev);
4798 rc = __pci_reset_function_locked(dev);
4800 pci_dev_restore(dev);
4801 pci_dev_unlock(dev);
4803 return rc;
4805 EXPORT_SYMBOL_GPL(pci_reset_function);
4808 * pci_reset_function_locked - quiesce and reset a PCI device function
4809 * @dev: PCI device to reset
4811 * Some devices allow an individual function to be reset without affecting
4812 * other functions in the same device. The PCI device must be responsive
4813 * to PCI config space in order to use this function.
4815 * This function does not just reset the PCI portion of a device, but
4816 * clears all the state associated with the device. This function differs
4817 * from __pci_reset_function_locked() in that it saves and restores device state
4818 * over the reset. It also differs from pci_reset_function() in that it
4819 * requires the PCI device lock to be held.
4821 * Returns 0 if the device function was successfully reset or negative if the
4822 * device doesn't support resetting a single function.
4824 int pci_reset_function_locked(struct pci_dev *dev)
4826 int rc;
4828 if (!dev->reset_fn)
4829 return -ENOTTY;
4831 pci_dev_save_and_disable(dev);
4833 rc = __pci_reset_function_locked(dev);
4835 pci_dev_restore(dev);
4837 return rc;
4839 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4842 * pci_try_reset_function - quiesce and reset a PCI device function
4843 * @dev: PCI device to reset
4845 * Same as above, except return -EAGAIN if unable to lock device.
4847 int pci_try_reset_function(struct pci_dev *dev)
4849 int rc;
4851 if (!dev->reset_fn)
4852 return -ENOTTY;
4854 if (!pci_dev_trylock(dev))
4855 return -EAGAIN;
4857 pci_dev_save_and_disable(dev);
4858 rc = __pci_reset_function_locked(dev);
4859 pci_dev_restore(dev);
4860 pci_dev_unlock(dev);
4862 return rc;
4864 EXPORT_SYMBOL_GPL(pci_try_reset_function);
4866 /* Do any devices on or below this bus prevent a bus reset? */
4867 static bool pci_bus_resetable(struct pci_bus *bus)
4869 struct pci_dev *dev;
4872 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4873 return false;
4875 list_for_each_entry(dev, &bus->devices, bus_list) {
4876 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4877 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4878 return false;
4881 return true;
4884 /* Lock devices from the top of the tree down */
4885 static void pci_bus_lock(struct pci_bus *bus)
4887 struct pci_dev *dev;
4889 list_for_each_entry(dev, &bus->devices, bus_list) {
4890 pci_dev_lock(dev);
4891 if (dev->subordinate)
4892 pci_bus_lock(dev->subordinate);
4896 /* Unlock devices from the bottom of the tree up */
4897 static void pci_bus_unlock(struct pci_bus *bus)
4899 struct pci_dev *dev;
4901 list_for_each_entry(dev, &bus->devices, bus_list) {
4902 if (dev->subordinate)
4903 pci_bus_unlock(dev->subordinate);
4904 pci_dev_unlock(dev);
4908 /* Return 1 on successful lock, 0 on contention */
4909 static int pci_bus_trylock(struct pci_bus *bus)
4911 struct pci_dev *dev;
4913 list_for_each_entry(dev, &bus->devices, bus_list) {
4914 if (!pci_dev_trylock(dev))
4915 goto unlock;
4916 if (dev->subordinate) {
4917 if (!pci_bus_trylock(dev->subordinate)) {
4918 pci_dev_unlock(dev);
4919 goto unlock;
4923 return 1;
4925 unlock:
4926 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4927 if (dev->subordinate)
4928 pci_bus_unlock(dev->subordinate);
4929 pci_dev_unlock(dev);
4931 return 0;
4934 /* Do any devices on or below this slot prevent a bus reset? */
4935 static bool pci_slot_resetable(struct pci_slot *slot)
4937 struct pci_dev *dev;
4939 if (slot->bus->self &&
4940 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4941 return false;
4943 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4944 if (!dev->slot || dev->slot != slot)
4945 continue;
4946 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4947 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4948 return false;
4951 return true;
4954 /* Lock devices from the top of the tree down */
4955 static void pci_slot_lock(struct pci_slot *slot)
4957 struct pci_dev *dev;
4959 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4960 if (!dev->slot || dev->slot != slot)
4961 continue;
4962 pci_dev_lock(dev);
4963 if (dev->subordinate)
4964 pci_bus_lock(dev->subordinate);
4968 /* Unlock devices from the bottom of the tree up */
4969 static void pci_slot_unlock(struct pci_slot *slot)
4971 struct pci_dev *dev;
4973 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4974 if (!dev->slot || dev->slot != slot)
4975 continue;
4976 if (dev->subordinate)
4977 pci_bus_unlock(dev->subordinate);
4978 pci_dev_unlock(dev);
4982 /* Return 1 on successful lock, 0 on contention */
4983 static int pci_slot_trylock(struct pci_slot *slot)
4985 struct pci_dev *dev;
4987 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4988 if (!dev->slot || dev->slot != slot)
4989 continue;
4990 if (!pci_dev_trylock(dev))
4991 goto unlock;
4992 if (dev->subordinate) {
4993 if (!pci_bus_trylock(dev->subordinate)) {
4994 pci_dev_unlock(dev);
4995 goto unlock;
4999 return 1;
5001 unlock:
5002 list_for_each_entry_continue_reverse(dev,
5003 &slot->bus->devices, bus_list) {
5004 if (!dev->slot || dev->slot != slot)
5005 continue;
5006 if (dev->subordinate)
5007 pci_bus_unlock(dev->subordinate);
5008 pci_dev_unlock(dev);
5010 return 0;
5013 /* Save and disable devices from the top of the tree down */
5014 static void pci_bus_save_and_disable(struct pci_bus *bus)
5016 struct pci_dev *dev;
5018 list_for_each_entry(dev, &bus->devices, bus_list) {
5019 pci_dev_lock(dev);
5020 pci_dev_save_and_disable(dev);
5021 pci_dev_unlock(dev);
5022 if (dev->subordinate)
5023 pci_bus_save_and_disable(dev->subordinate);
5028 * Restore devices from top of the tree down - parent bridges need to be
5029 * restored before we can get to subordinate devices.
5031 static void pci_bus_restore(struct pci_bus *bus)
5033 struct pci_dev *dev;
5035 list_for_each_entry(dev, &bus->devices, bus_list) {
5036 pci_dev_lock(dev);
5037 pci_dev_restore(dev);
5038 pci_dev_unlock(dev);
5039 if (dev->subordinate)
5040 pci_bus_restore(dev->subordinate);
5044 /* Save and disable devices from the top of the tree down */
5045 static void pci_slot_save_and_disable(struct pci_slot *slot)
5047 struct pci_dev *dev;
5049 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5050 if (!dev->slot || dev->slot != slot)
5051 continue;
5052 pci_dev_save_and_disable(dev);
5053 if (dev->subordinate)
5054 pci_bus_save_and_disable(dev->subordinate);
5059 * Restore devices from top of the tree down - parent bridges need to be
5060 * restored before we can get to subordinate devices.
5062 static void pci_slot_restore(struct pci_slot *slot)
5064 struct pci_dev *dev;
5066 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5067 if (!dev->slot || dev->slot != slot)
5068 continue;
5069 pci_dev_lock(dev);
5070 pci_dev_restore(dev);
5071 pci_dev_unlock(dev);
5072 if (dev->subordinate)
5073 pci_bus_restore(dev->subordinate);
5077 static int pci_slot_reset(struct pci_slot *slot, int probe)
5079 int rc;
5081 if (!slot || !pci_slot_resetable(slot))
5082 return -ENOTTY;
5084 if (!probe)
5085 pci_slot_lock(slot);
5087 might_sleep();
5089 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5091 if (!probe)
5092 pci_slot_unlock(slot);
5094 return rc;
5098 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5099 * @slot: PCI slot to probe
5101 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5103 int pci_probe_reset_slot(struct pci_slot *slot)
5105 return pci_slot_reset(slot, 1);
5107 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5110 * __pci_reset_slot - Try to reset a PCI slot
5111 * @slot: PCI slot to reset
5113 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5114 * independent of other slots. For instance, some slots may support slot power
5115 * control. In the case of a 1:1 bus to slot architecture, this function may
5116 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5117 * Generally a slot reset should be attempted before a bus reset. All of the
5118 * function of the slot and any subordinate buses behind the slot are reset
5119 * through this function. PCI config space of all devices in the slot and
5120 * behind the slot is saved before and restored after reset.
5122 * Same as above except return -EAGAIN if the slot cannot be locked
5124 static int __pci_reset_slot(struct pci_slot *slot)
5126 int rc;
5128 rc = pci_slot_reset(slot, 1);
5129 if (rc)
5130 return rc;
5132 pci_slot_save_and_disable(slot);
5134 if (pci_slot_trylock(slot)) {
5135 might_sleep();
5136 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
5137 pci_slot_unlock(slot);
5138 } else
5139 rc = -EAGAIN;
5141 pci_slot_restore(slot);
5143 return rc;
5146 static int pci_bus_reset(struct pci_bus *bus, int probe)
5148 int ret;
5150 if (!bus->self || !pci_bus_resetable(bus))
5151 return -ENOTTY;
5153 if (probe)
5154 return 0;
5156 pci_bus_lock(bus);
5158 might_sleep();
5160 ret = pci_bridge_secondary_bus_reset(bus->self);
5162 pci_bus_unlock(bus);
5164 return ret;
5168 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5169 * @bus: PCI bus to probe
5171 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5173 int pci_probe_reset_bus(struct pci_bus *bus)
5175 return pci_bus_reset(bus, 1);
5177 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5180 * __pci_reset_bus - Try to reset a PCI bus
5181 * @bus: top level PCI bus to reset
5183 * Same as above except return -EAGAIN if the bus cannot be locked
5185 static int __pci_reset_bus(struct pci_bus *bus)
5187 int rc;
5189 rc = pci_bus_reset(bus, 1);
5190 if (rc)
5191 return rc;
5193 pci_bus_save_and_disable(bus);
5195 if (pci_bus_trylock(bus)) {
5196 might_sleep();
5197 rc = pci_bridge_secondary_bus_reset(bus->self);
5198 pci_bus_unlock(bus);
5199 } else
5200 rc = -EAGAIN;
5202 pci_bus_restore(bus);
5204 return rc;
5208 * pci_reset_bus - Try to reset a PCI bus
5209 * @pdev: top level PCI device to reset via slot/bus
5211 * Same as above except return -EAGAIN if the bus cannot be locked
5213 int pci_reset_bus(struct pci_dev *pdev)
5215 return (!pci_probe_reset_slot(pdev->slot)) ?
5216 __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5218 EXPORT_SYMBOL_GPL(pci_reset_bus);
5221 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5222 * @dev: PCI device to query
5224 * Returns mmrbc: maximum designed memory read count in bytes
5225 * or appropriate error value.
5227 int pcix_get_max_mmrbc(struct pci_dev *dev)
5229 int cap;
5230 u32 stat;
5232 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5233 if (!cap)
5234 return -EINVAL;
5236 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5237 return -EINVAL;
5239 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5241 EXPORT_SYMBOL(pcix_get_max_mmrbc);
5244 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5245 * @dev: PCI device to query
5247 * Returns mmrbc: maximum memory read count in bytes
5248 * or appropriate error value.
5250 int pcix_get_mmrbc(struct pci_dev *dev)
5252 int cap;
5253 u16 cmd;
5255 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5256 if (!cap)
5257 return -EINVAL;
5259 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5260 return -EINVAL;
5262 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5264 EXPORT_SYMBOL(pcix_get_mmrbc);
5267 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5268 * @dev: PCI device to query
5269 * @mmrbc: maximum memory read count in bytes
5270 * valid values are 512, 1024, 2048, 4096
5272 * If possible sets maximum memory read byte count, some bridges have erratas
5273 * that prevent this.
5275 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5277 int cap;
5278 u32 stat, v, o;
5279 u16 cmd;
5281 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5282 return -EINVAL;
5284 v = ffs(mmrbc) - 10;
5286 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5287 if (!cap)
5288 return -EINVAL;
5290 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5291 return -EINVAL;
5293 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5294 return -E2BIG;
5296 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5297 return -EINVAL;
5299 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5300 if (o != v) {
5301 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5302 return -EIO;
5304 cmd &= ~PCI_X_CMD_MAX_READ;
5305 cmd |= v << 2;
5306 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5307 return -EIO;
5309 return 0;
5311 EXPORT_SYMBOL(pcix_set_mmrbc);
5314 * pcie_get_readrq - get PCI Express read request size
5315 * @dev: PCI device to query
5317 * Returns maximum memory read request in bytes
5318 * or appropriate error value.
5320 int pcie_get_readrq(struct pci_dev *dev)
5322 u16 ctl;
5324 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5326 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5328 EXPORT_SYMBOL(pcie_get_readrq);
5331 * pcie_set_readrq - set PCI Express maximum memory read request
5332 * @dev: PCI device to query
5333 * @rq: maximum memory read count in bytes
5334 * valid values are 128, 256, 512, 1024, 2048, 4096
5336 * If possible sets maximum memory read request in bytes
5338 int pcie_set_readrq(struct pci_dev *dev, int rq)
5340 u16 v;
5342 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5343 return -EINVAL;
5346 * If using the "performance" PCIe config, we clamp the
5347 * read rq size to the max packet size to prevent the
5348 * host bridge generating requests larger than we can
5349 * cope with
5351 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5352 int mps = pcie_get_mps(dev);
5354 if (mps < rq)
5355 rq = mps;
5358 v = (ffs(rq) - 8) << 12;
5360 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5361 PCI_EXP_DEVCTL_READRQ, v);
5363 EXPORT_SYMBOL(pcie_set_readrq);
5366 * pcie_get_mps - get PCI Express maximum payload size
5367 * @dev: PCI device to query
5369 * Returns maximum payload size in bytes
5371 int pcie_get_mps(struct pci_dev *dev)
5373 u16 ctl;
5375 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5377 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5379 EXPORT_SYMBOL(pcie_get_mps);
5382 * pcie_set_mps - set PCI Express maximum payload size
5383 * @dev: PCI device to query
5384 * @mps: maximum payload size in bytes
5385 * valid values are 128, 256, 512, 1024, 2048, 4096
5387 * If possible sets maximum payload size
5389 int pcie_set_mps(struct pci_dev *dev, int mps)
5391 u16 v;
5393 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5394 return -EINVAL;
5396 v = ffs(mps) - 8;
5397 if (v > dev->pcie_mpss)
5398 return -EINVAL;
5399 v <<= 5;
5401 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5402 PCI_EXP_DEVCTL_PAYLOAD, v);
5404 EXPORT_SYMBOL(pcie_set_mps);
5407 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5408 * device and its bandwidth limitation
5409 * @dev: PCI device to query
5410 * @limiting_dev: storage for device causing the bandwidth limitation
5411 * @speed: storage for speed of limiting device
5412 * @width: storage for width of limiting device
5414 * Walk up the PCI device chain and find the point where the minimum
5415 * bandwidth is available. Return the bandwidth available there and (if
5416 * limiting_dev, speed, and width pointers are supplied) information about
5417 * that point. The bandwidth returned is in Mb/s, i.e., megabits/second of
5418 * raw bandwidth.
5420 u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5421 enum pci_bus_speed *speed,
5422 enum pcie_link_width *width)
5424 u16 lnksta;
5425 enum pci_bus_speed next_speed;
5426 enum pcie_link_width next_width;
5427 u32 bw, next_bw;
5429 if (speed)
5430 *speed = PCI_SPEED_UNKNOWN;
5431 if (width)
5432 *width = PCIE_LNK_WIDTH_UNKNOWN;
5434 bw = 0;
5436 while (dev) {
5437 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5439 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5440 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5441 PCI_EXP_LNKSTA_NLW_SHIFT;
5443 next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5445 /* Check if current device limits the total bandwidth */
5446 if (!bw || next_bw <= bw) {
5447 bw = next_bw;
5449 if (limiting_dev)
5450 *limiting_dev = dev;
5451 if (speed)
5452 *speed = next_speed;
5453 if (width)
5454 *width = next_width;
5457 dev = pci_upstream_bridge(dev);
5460 return bw;
5462 EXPORT_SYMBOL(pcie_bandwidth_available);
5465 * pcie_get_speed_cap - query for the PCI device's link speed capability
5466 * @dev: PCI device to query
5468 * Query the PCI device speed capability. Return the maximum link speed
5469 * supported by the device.
5471 enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5473 u32 lnkcap2, lnkcap;
5476 * PCIe r4.0 sec 7.5.3.18 recommends using the Supported Link
5477 * Speeds Vector in Link Capabilities 2 when supported, falling
5478 * back to Max Link Speed in Link Capabilities otherwise.
5480 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5481 if (lnkcap2) { /* PCIe r3.0-compliant */
5482 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_16_0GB)
5483 return PCIE_SPEED_16_0GT;
5484 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
5485 return PCIE_SPEED_8_0GT;
5486 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
5487 return PCIE_SPEED_5_0GT;
5488 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
5489 return PCIE_SPEED_2_5GT;
5490 return PCI_SPEED_UNKNOWN;
5493 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5494 if (lnkcap) {
5495 if (lnkcap & PCI_EXP_LNKCAP_SLS_16_0GB)
5496 return PCIE_SPEED_16_0GT;
5497 else if (lnkcap & PCI_EXP_LNKCAP_SLS_8_0GB)
5498 return PCIE_SPEED_8_0GT;
5499 else if (lnkcap & PCI_EXP_LNKCAP_SLS_5_0GB)
5500 return PCIE_SPEED_5_0GT;
5501 else if (lnkcap & PCI_EXP_LNKCAP_SLS_2_5GB)
5502 return PCIE_SPEED_2_5GT;
5505 return PCI_SPEED_UNKNOWN;
5507 EXPORT_SYMBOL(pcie_get_speed_cap);
5510 * pcie_get_width_cap - query for the PCI device's link width capability
5511 * @dev: PCI device to query
5513 * Query the PCI device width capability. Return the maximum link width
5514 * supported by the device.
5516 enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5518 u32 lnkcap;
5520 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5521 if (lnkcap)
5522 return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5524 return PCIE_LNK_WIDTH_UNKNOWN;
5526 EXPORT_SYMBOL(pcie_get_width_cap);
5529 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5530 * @dev: PCI device
5531 * @speed: storage for link speed
5532 * @width: storage for link width
5534 * Calculate a PCI device's link bandwidth by querying for its link speed
5535 * and width, multiplying them, and applying encoding overhead. The result
5536 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5538 u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5539 enum pcie_link_width *width)
5541 *speed = pcie_get_speed_cap(dev);
5542 *width = pcie_get_width_cap(dev);
5544 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5545 return 0;
5547 return *width * PCIE_SPEED2MBS_ENC(*speed);
5551 * __pcie_print_link_status - Report the PCI device's link speed and width
5552 * @dev: PCI device to query
5553 * @verbose: Print info even when enough bandwidth is available
5555 * If the available bandwidth at the device is less than the device is
5556 * capable of, report the device's maximum possible bandwidth and the
5557 * upstream link that limits its performance. If @verbose, always print
5558 * the available bandwidth, even if the device isn't constrained.
5560 void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
5562 enum pcie_link_width width, width_cap;
5563 enum pci_bus_speed speed, speed_cap;
5564 struct pci_dev *limiting_dev = NULL;
5565 u32 bw_avail, bw_cap;
5567 bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5568 bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5570 if (bw_avail >= bw_cap && verbose)
5571 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5572 bw_cap / 1000, bw_cap % 1000,
5573 PCIE_SPEED2STR(speed_cap), width_cap);
5574 else if (bw_avail < bw_cap)
5575 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",
5576 bw_avail / 1000, bw_avail % 1000,
5577 PCIE_SPEED2STR(speed), width,
5578 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5579 bw_cap / 1000, bw_cap % 1000,
5580 PCIE_SPEED2STR(speed_cap), width_cap);
5584 * pcie_print_link_status - Report the PCI device's link speed and width
5585 * @dev: PCI device to query
5587 * Report the available bandwidth at the device.
5589 void pcie_print_link_status(struct pci_dev *dev)
5591 __pcie_print_link_status(dev, true);
5593 EXPORT_SYMBOL(pcie_print_link_status);
5596 * pci_select_bars - Make BAR mask from the type of resource
5597 * @dev: the PCI device for which BAR mask is made
5598 * @flags: resource type mask to be selected
5600 * This helper routine makes bar mask from the type of resource.
5602 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5604 int i, bars = 0;
5605 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5606 if (pci_resource_flags(dev, i) & flags)
5607 bars |= (1 << i);
5608 return bars;
5610 EXPORT_SYMBOL(pci_select_bars);
5612 /* Some architectures require additional programming to enable VGA */
5613 static arch_set_vga_state_t arch_set_vga_state;
5615 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5617 arch_set_vga_state = func; /* NULL disables */
5620 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5621 unsigned int command_bits, u32 flags)
5623 if (arch_set_vga_state)
5624 return arch_set_vga_state(dev, decode, command_bits,
5625 flags);
5626 return 0;
5630 * pci_set_vga_state - set VGA decode state on device and parents if requested
5631 * @dev: the PCI device
5632 * @decode: true = enable decoding, false = disable decoding
5633 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5634 * @flags: traverse ancestors and change bridges
5635 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5637 int pci_set_vga_state(struct pci_dev *dev, bool decode,
5638 unsigned int command_bits, u32 flags)
5640 struct pci_bus *bus;
5641 struct pci_dev *bridge;
5642 u16 cmd;
5643 int rc;
5645 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5647 /* ARCH specific VGA enables */
5648 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5649 if (rc)
5650 return rc;
5652 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5653 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5654 if (decode == true)
5655 cmd |= command_bits;
5656 else
5657 cmd &= ~command_bits;
5658 pci_write_config_word(dev, PCI_COMMAND, cmd);
5661 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5662 return 0;
5664 bus = dev->bus;
5665 while (bus) {
5666 bridge = bus->self;
5667 if (bridge) {
5668 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5669 &cmd);
5670 if (decode == true)
5671 cmd |= PCI_BRIDGE_CTL_VGA;
5672 else
5673 cmd &= ~PCI_BRIDGE_CTL_VGA;
5674 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5675 cmd);
5677 bus = bus->parent;
5679 return 0;
5683 * pci_add_dma_alias - Add a DMA devfn alias for a device
5684 * @dev: the PCI device for which alias is added
5685 * @devfn: alias slot and function
5687 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
5688 * which is used to program permissible bus-devfn source addresses for DMA
5689 * requests in an IOMMU. These aliases factor into IOMMU group creation
5690 * and are useful for devices generating DMA requests beyond or different
5691 * from their logical bus-devfn. Examples include device quirks where the
5692 * device simply uses the wrong devfn, as well as non-transparent bridges
5693 * where the alias may be a proxy for devices in another domain.
5695 * IOMMU group creation is performed during device discovery or addition,
5696 * prior to any potential DMA mapping and therefore prior to driver probing
5697 * (especially for userspace assigned devices where IOMMU group definition
5698 * cannot be left as a userspace activity). DMA aliases should therefore
5699 * be configured via quirks, such as the PCI fixup header quirk.
5701 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5703 if (!dev->dma_alias_mask)
5704 dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
5705 sizeof(long), GFP_KERNEL);
5706 if (!dev->dma_alias_mask) {
5707 pci_warn(dev, "Unable to allocate DMA alias mask\n");
5708 return;
5711 set_bit(devfn, dev->dma_alias_mask);
5712 pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
5713 PCI_SLOT(devfn), PCI_FUNC(devfn));
5716 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5718 return (dev1->dma_alias_mask &&
5719 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5720 (dev2->dma_alias_mask &&
5721 test_bit(dev1->devfn, dev2->dma_alias_mask));
5724 bool pci_device_is_present(struct pci_dev *pdev)
5726 u32 v;
5728 if (pci_dev_is_disconnected(pdev))
5729 return false;
5730 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5732 EXPORT_SYMBOL_GPL(pci_device_is_present);
5734 void pci_ignore_hotplug(struct pci_dev *dev)
5736 struct pci_dev *bridge = dev->bus->self;
5738 dev->ignore_hotplug = 1;
5739 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5740 if (bridge)
5741 bridge->ignore_hotplug = 1;
5743 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5745 resource_size_t __weak pcibios_default_alignment(void)
5747 return 0;
5750 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5751 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5752 static DEFINE_SPINLOCK(resource_alignment_lock);
5755 * pci_specified_resource_alignment - get resource alignment specified by user.
5756 * @dev: the PCI device to get
5757 * @resize: whether or not to change resources' size when reassigning alignment
5759 * RETURNS: Resource alignment if it is specified.
5760 * Zero if it is not specified.
5762 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5763 bool *resize)
5765 int align_order, count;
5766 resource_size_t align = pcibios_default_alignment();
5767 const char *p;
5768 int ret;
5770 spin_lock(&resource_alignment_lock);
5771 p = resource_alignment_param;
5772 if (!*p && !align)
5773 goto out;
5774 if (pci_has_flag(PCI_PROBE_ONLY)) {
5775 align = 0;
5776 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5777 goto out;
5780 while (*p) {
5781 count = 0;
5782 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5783 p[count] == '@') {
5784 p += count + 1;
5785 } else {
5786 align_order = -1;
5789 ret = pci_dev_str_match(dev, p, &p);
5790 if (ret == 1) {
5791 *resize = true;
5792 if (align_order == -1)
5793 align = PAGE_SIZE;
5794 else
5795 align = 1 << align_order;
5796 break;
5797 } else if (ret < 0) {
5798 pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
5800 break;
5803 if (*p != ';' && *p != ',') {
5804 /* End of param or invalid format */
5805 break;
5807 p++;
5809 out:
5810 spin_unlock(&resource_alignment_lock);
5811 return align;
5814 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5815 resource_size_t align, bool resize)
5817 struct resource *r = &dev->resource[bar];
5818 resource_size_t size;
5820 if (!(r->flags & IORESOURCE_MEM))
5821 return;
5823 if (r->flags & IORESOURCE_PCI_FIXED) {
5824 pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5825 bar, r, (unsigned long long)align);
5826 return;
5829 size = resource_size(r);
5830 if (size >= align)
5831 return;
5834 * Increase the alignment of the resource. There are two ways we
5835 * can do this:
5837 * 1) Increase the size of the resource. BARs are aligned on their
5838 * size, so when we reallocate space for this resource, we'll
5839 * allocate it with the larger alignment. This also prevents
5840 * assignment of any other BARs inside the alignment region, so
5841 * if we're requesting page alignment, this means no other BARs
5842 * will share the page.
5844 * The disadvantage is that this makes the resource larger than
5845 * the hardware BAR, which may break drivers that compute things
5846 * based on the resource size, e.g., to find registers at a
5847 * fixed offset before the end of the BAR.
5849 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5850 * set r->start to the desired alignment. By itself this
5851 * doesn't prevent other BARs being put inside the alignment
5852 * region, but if we realign *every* resource of every device in
5853 * the system, none of them will share an alignment region.
5855 * When the user has requested alignment for only some devices via
5856 * the "pci=resource_alignment" argument, "resize" is true and we
5857 * use the first method. Otherwise we assume we're aligning all
5858 * devices and we use the second.
5861 pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
5862 bar, r, (unsigned long long)align);
5864 if (resize) {
5865 r->start = 0;
5866 r->end = align - 1;
5867 } else {
5868 r->flags &= ~IORESOURCE_SIZEALIGN;
5869 r->flags |= IORESOURCE_STARTALIGN;
5870 r->start = align;
5871 r->end = r->start + size - 1;
5873 r->flags |= IORESOURCE_UNSET;
5877 * This function disables memory decoding and releases memory resources
5878 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5879 * It also rounds up size to specified alignment.
5880 * Later on, the kernel will assign page-aligned memory resource back
5881 * to the device.
5883 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5885 int i;
5886 struct resource *r;
5887 resource_size_t align;
5888 u16 command;
5889 bool resize = false;
5892 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
5893 * 3.4.1.11. Their resources are allocated from the space
5894 * described by the VF BARx register in the PF's SR-IOV capability.
5895 * We can't influence their alignment here.
5897 if (dev->is_virtfn)
5898 return;
5900 /* check if specified PCI is target device to reassign */
5901 align = pci_specified_resource_alignment(dev, &resize);
5902 if (!align)
5903 return;
5905 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5906 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5907 pci_warn(dev, "Can't reassign resources to host bridge\n");
5908 return;
5911 pci_read_config_word(dev, PCI_COMMAND, &command);
5912 command &= ~PCI_COMMAND_MEMORY;
5913 pci_write_config_word(dev, PCI_COMMAND, command);
5915 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
5916 pci_request_resource_alignment(dev, i, align, resize);
5919 * Need to disable bridge's resource window,
5920 * to enable the kernel to reassign new resource
5921 * window later on.
5923 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5924 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5925 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5926 r = &dev->resource[i];
5927 if (!(r->flags & IORESOURCE_MEM))
5928 continue;
5929 r->flags |= IORESOURCE_UNSET;
5930 r->end = resource_size(r) - 1;
5931 r->start = 0;
5933 pci_disable_bridge_window(dev);
5937 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5939 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
5940 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
5941 spin_lock(&resource_alignment_lock);
5942 strncpy(resource_alignment_param, buf, count);
5943 resource_alignment_param[count] = '\0';
5944 spin_unlock(&resource_alignment_lock);
5945 return count;
5948 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
5950 size_t count;
5951 spin_lock(&resource_alignment_lock);
5952 count = snprintf(buf, size, "%s", resource_alignment_param);
5953 spin_unlock(&resource_alignment_lock);
5954 return count;
5957 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
5959 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
5962 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
5963 const char *buf, size_t count)
5965 return pci_set_resource_alignment_param(buf, count);
5968 static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
5969 pci_resource_alignment_store);
5971 static int __init pci_resource_alignment_sysfs_init(void)
5973 return bus_create_file(&pci_bus_type,
5974 &bus_attr_resource_alignment);
5976 late_initcall(pci_resource_alignment_sysfs_init);
5978 static void pci_no_domains(void)
5980 #ifdef CONFIG_PCI_DOMAINS
5981 pci_domains_supported = 0;
5982 #endif
5985 #ifdef CONFIG_PCI_DOMAINS_GENERIC
5986 static atomic_t __domain_nr = ATOMIC_INIT(-1);
5988 static int pci_get_new_domain_nr(void)
5990 return atomic_inc_return(&__domain_nr);
5993 static int of_pci_bus_find_domain_nr(struct device *parent)
5995 static int use_dt_domains = -1;
5996 int domain = -1;
5998 if (parent)
5999 domain = of_get_pci_domain_nr(parent->of_node);
6001 * Check DT domain and use_dt_domains values.
6003 * If DT domain property is valid (domain >= 0) and
6004 * use_dt_domains != 0, the DT assignment is valid since this means
6005 * we have not previously allocated a domain number by using
6006 * pci_get_new_domain_nr(); we should also update use_dt_domains to
6007 * 1, to indicate that we have just assigned a domain number from
6008 * DT.
6010 * If DT domain property value is not valid (ie domain < 0), and we
6011 * have not previously assigned a domain number from DT
6012 * (use_dt_domains != 1) we should assign a domain number by
6013 * using the:
6015 * pci_get_new_domain_nr()
6017 * API and update the use_dt_domains value to keep track of method we
6018 * are using to assign domain numbers (use_dt_domains = 0).
6020 * All other combinations imply we have a platform that is trying
6021 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6022 * which is a recipe for domain mishandling and it is prevented by
6023 * invalidating the domain value (domain = -1) and printing a
6024 * corresponding error.
6026 if (domain >= 0 && use_dt_domains) {
6027 use_dt_domains = 1;
6028 } else if (domain < 0 && use_dt_domains != 1) {
6029 use_dt_domains = 0;
6030 domain = pci_get_new_domain_nr();
6031 } else {
6032 if (parent)
6033 pr_err("Node %pOF has ", parent->of_node);
6034 pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6035 domain = -1;
6038 return domain;
6041 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6043 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6044 acpi_pci_bus_find_domain_nr(bus);
6046 #endif
6049 * pci_ext_cfg_avail - can we access extended PCI config space?
6051 * Returns 1 if we can access PCI extended config space (offsets
6052 * greater than 0xff). This is the default implementation. Architecture
6053 * implementations can override this.
6055 int __weak pci_ext_cfg_avail(void)
6057 return 1;
6060 void __weak pci_fixup_cardbus(struct pci_bus *bus)
6063 EXPORT_SYMBOL(pci_fixup_cardbus);
6065 static int __init pci_setup(char *str)
6067 while (str) {
6068 char *k = strchr(str, ',');
6069 if (k)
6070 *k++ = 0;
6071 if (*str && (str = pcibios_setup(str)) && *str) {
6072 if (!strcmp(str, "nomsi")) {
6073 pci_no_msi();
6074 } else if (!strncmp(str, "noats", 5)) {
6075 pr_info("PCIe: ATS is disabled\n");
6076 pcie_ats_disabled = true;
6077 } else if (!strcmp(str, "noaer")) {
6078 pci_no_aer();
6079 } else if (!strcmp(str, "earlydump")) {
6080 pci_early_dump = true;
6081 } else if (!strncmp(str, "realloc=", 8)) {
6082 pci_realloc_get_opt(str + 8);
6083 } else if (!strncmp(str, "realloc", 7)) {
6084 pci_realloc_get_opt("on");
6085 } else if (!strcmp(str, "nodomains")) {
6086 pci_no_domains();
6087 } else if (!strncmp(str, "noari", 5)) {
6088 pcie_ari_disabled = true;
6089 } else if (!strncmp(str, "cbiosize=", 9)) {
6090 pci_cardbus_io_size = memparse(str + 9, &str);
6091 } else if (!strncmp(str, "cbmemsize=", 10)) {
6092 pci_cardbus_mem_size = memparse(str + 10, &str);
6093 } else if (!strncmp(str, "resource_alignment=", 19)) {
6094 pci_set_resource_alignment_param(str + 19,
6095 strlen(str + 19));
6096 } else if (!strncmp(str, "ecrc=", 5)) {
6097 pcie_ecrc_get_policy(str + 5);
6098 } else if (!strncmp(str, "hpiosize=", 9)) {
6099 pci_hotplug_io_size = memparse(str + 9, &str);
6100 } else if (!strncmp(str, "hpmemsize=", 10)) {
6101 pci_hotplug_mem_size = memparse(str + 10, &str);
6102 } else if (!strncmp(str, "hpbussize=", 10)) {
6103 pci_hotplug_bus_size =
6104 simple_strtoul(str + 10, &str, 0);
6105 if (pci_hotplug_bus_size > 0xff)
6106 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6107 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6108 pcie_bus_config = PCIE_BUS_TUNE_OFF;
6109 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6110 pcie_bus_config = PCIE_BUS_SAFE;
6111 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6112 pcie_bus_config = PCIE_BUS_PERFORMANCE;
6113 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6114 pcie_bus_config = PCIE_BUS_PEER2PEER;
6115 } else if (!strncmp(str, "pcie_scan_all", 13)) {
6116 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6117 } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6118 disable_acs_redir_param = str + 18;
6119 } else {
6120 printk(KERN_ERR "PCI: Unknown option `%s'\n",
6121 str);
6124 str = k;
6126 return 0;
6128 early_param("pci", pci_setup);