Linux 4.19.133
[linux/fpc-iii.git] / drivers / pci / pci.c
blob57a87a001b4f4f608c79c4f7940eac95a20335f8
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 DEFINE_MUTEX(pci_slot_mutex);
40 const char *pci_power_names[] = {
41 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
43 EXPORT_SYMBOL_GPL(pci_power_names);
45 int isa_dma_bridge_buggy;
46 EXPORT_SYMBOL(isa_dma_bridge_buggy);
48 int pci_pci_problems;
49 EXPORT_SYMBOL(pci_pci_problems);
51 unsigned int pci_pm_d3_delay;
53 static void pci_pme_list_scan(struct work_struct *work);
55 static LIST_HEAD(pci_pme_list);
56 static DEFINE_MUTEX(pci_pme_list_mutex);
57 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
59 struct pci_pme_device {
60 struct list_head list;
61 struct pci_dev *dev;
64 #define PME_TIMEOUT 1000 /* How long between PME checks */
66 static void pci_dev_d3_sleep(struct pci_dev *dev)
68 unsigned int delay = dev->d3_delay;
70 if (delay < pci_pm_d3_delay)
71 delay = pci_pm_d3_delay;
73 if (delay)
74 msleep(delay);
77 #ifdef CONFIG_PCI_DOMAINS
78 int pci_domains_supported = 1;
79 #endif
81 #define DEFAULT_CARDBUS_IO_SIZE (256)
82 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
83 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
84 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
85 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
87 #define DEFAULT_HOTPLUG_IO_SIZE (256)
88 #define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
89 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
90 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
91 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
93 #define DEFAULT_HOTPLUG_BUS_SIZE 1
94 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
96 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
99 * The default CLS is used if arch didn't set CLS explicitly and not
100 * all pci devices agree on the same value. Arch can override either
101 * the dfl or actual value as it sees fit. Don't forget this is
102 * measured in 32-bit words, not bytes.
104 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
105 u8 pci_cache_line_size;
108 * If we set up a device for bus mastering, we need to check the latency
109 * timer as certain BIOSes forget to set it properly.
111 unsigned int pcibios_max_latency = 255;
113 /* If set, the PCIe ARI capability will not be used. */
114 static bool pcie_ari_disabled;
116 /* If set, the PCIe ATS capability will not be used. */
117 static bool pcie_ats_disabled;
119 /* If set, the PCI config space of each device is printed during boot. */
120 bool pci_early_dump;
122 bool pci_ats_disabled(void)
124 return pcie_ats_disabled;
127 /* Disable bridge_d3 for all PCIe ports */
128 static bool pci_bridge_d3_disable;
129 /* Force bridge_d3 for all PCIe ports */
130 static bool pci_bridge_d3_force;
132 static int __init pcie_port_pm_setup(char *str)
134 if (!strcmp(str, "off"))
135 pci_bridge_d3_disable = true;
136 else if (!strcmp(str, "force"))
137 pci_bridge_d3_force = true;
138 return 1;
140 __setup("pcie_port_pm=", pcie_port_pm_setup);
142 /* Time to wait after a reset for device to become responsive */
143 #define PCIE_RESET_READY_POLL_MS 60000
146 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
147 * @bus: pointer to PCI bus structure to search
149 * Given a PCI bus, returns the highest PCI bus number present in the set
150 * including the given PCI bus and its list of child PCI buses.
152 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
154 struct pci_bus *tmp;
155 unsigned char max, n;
157 max = bus->busn_res.end;
158 list_for_each_entry(tmp, &bus->children, node) {
159 n = pci_bus_max_busnr(tmp);
160 if (n > max)
161 max = n;
163 return max;
165 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
167 #ifdef CONFIG_HAS_IOMEM
168 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
170 struct resource *res = &pdev->resource[bar];
173 * Make sure the BAR is actually a memory resource, not an IO resource
175 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
176 pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
177 return NULL;
179 return ioremap_nocache(res->start, resource_size(res));
181 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
183 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
186 * Make sure the BAR is actually a memory resource, not an IO resource
188 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
189 WARN_ON(1);
190 return NULL;
192 return ioremap_wc(pci_resource_start(pdev, bar),
193 pci_resource_len(pdev, bar));
195 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
196 #endif
199 * pci_dev_str_match_path - test if a path string matches a device
200 * @dev: the PCI device to test
201 * @p: string to match the device against
202 * @endptr: pointer to the string after the match
204 * Test if a string (typically from a kernel parameter) formatted as a
205 * path of device/function addresses matches a PCI device. The string must
206 * be of the form:
208 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
210 * A path for a device can be obtained using 'lspci -t'. Using a path
211 * is more robust against bus renumbering than using only a single bus,
212 * device and function address.
214 * Returns 1 if the string matches the device, 0 if it does not and
215 * a negative error code if it fails to parse the string.
217 static int pci_dev_str_match_path(struct pci_dev *dev, const char *path,
218 const char **endptr)
220 int ret;
221 int seg, bus, slot, func;
222 char *wpath, *p;
223 char end;
225 *endptr = strchrnul(path, ';');
227 wpath = kmemdup_nul(path, *endptr - path, GFP_KERNEL);
228 if (!wpath)
229 return -ENOMEM;
231 while (1) {
232 p = strrchr(wpath, '/');
233 if (!p)
234 break;
235 ret = sscanf(p, "/%x.%x%c", &slot, &func, &end);
236 if (ret != 2) {
237 ret = -EINVAL;
238 goto free_and_exit;
241 if (dev->devfn != PCI_DEVFN(slot, func)) {
242 ret = 0;
243 goto free_and_exit;
247 * Note: we don't need to get a reference to the upstream
248 * bridge because we hold a reference to the top level
249 * device which should hold a reference to the bridge,
250 * and so on.
252 dev = pci_upstream_bridge(dev);
253 if (!dev) {
254 ret = 0;
255 goto free_and_exit;
258 *p = 0;
261 ret = sscanf(wpath, "%x:%x:%x.%x%c", &seg, &bus, &slot,
262 &func, &end);
263 if (ret != 4) {
264 seg = 0;
265 ret = sscanf(wpath, "%x:%x.%x%c", &bus, &slot, &func, &end);
266 if (ret != 3) {
267 ret = -EINVAL;
268 goto free_and_exit;
272 ret = (seg == pci_domain_nr(dev->bus) &&
273 bus == dev->bus->number &&
274 dev->devfn == PCI_DEVFN(slot, func));
276 free_and_exit:
277 kfree(wpath);
278 return ret;
282 * pci_dev_str_match - test if a string matches a device
283 * @dev: the PCI device to test
284 * @p: string to match the device against
285 * @endptr: pointer to the string after the match
287 * Test if a string (typically from a kernel parameter) matches a specified
288 * PCI device. The string may be of one of the following formats:
290 * [<domain>:]<bus>:<device>.<func>[/<device>.<func>]*
291 * pci:<vendor>:<device>[:<subvendor>:<subdevice>]
293 * The first format specifies a PCI bus/device/function address which
294 * may change if new hardware is inserted, if motherboard firmware changes,
295 * or due to changes caused in kernel parameters. If the domain is
296 * left unspecified, it is taken to be 0. In order to be robust against
297 * bus renumbering issues, a path of PCI device/function numbers may be used
298 * to address the specific device. The path for a device can be determined
299 * through the use of 'lspci -t'.
301 * The second format matches devices using IDs in the configuration
302 * space which may match multiple devices in the system. A value of 0
303 * for any field will match all devices. (Note: this differs from
304 * in-kernel code that uses PCI_ANY_ID which is ~0; this is for
305 * legacy reasons and convenience so users don't have to specify
306 * FFFFFFFFs on the command line.)
308 * Returns 1 if the string matches the device, 0 if it does not and
309 * a negative error code if the string cannot be parsed.
311 static int pci_dev_str_match(struct pci_dev *dev, const char *p,
312 const char **endptr)
314 int ret;
315 int count;
316 unsigned short vendor, device, subsystem_vendor, subsystem_device;
318 if (strncmp(p, "pci:", 4) == 0) {
319 /* PCI vendor/device (subvendor/subdevice) IDs are specified */
320 p += 4;
321 ret = sscanf(p, "%hx:%hx:%hx:%hx%n", &vendor, &device,
322 &subsystem_vendor, &subsystem_device, &count);
323 if (ret != 4) {
324 ret = sscanf(p, "%hx:%hx%n", &vendor, &device, &count);
325 if (ret != 2)
326 return -EINVAL;
328 subsystem_vendor = 0;
329 subsystem_device = 0;
332 p += count;
334 if ((!vendor || vendor == dev->vendor) &&
335 (!device || device == dev->device) &&
336 (!subsystem_vendor ||
337 subsystem_vendor == dev->subsystem_vendor) &&
338 (!subsystem_device ||
339 subsystem_device == dev->subsystem_device))
340 goto found;
341 } else {
343 * PCI Bus, Device, Function IDs are specified
344 * (optionally, may include a path of devfns following it)
346 ret = pci_dev_str_match_path(dev, p, &p);
347 if (ret < 0)
348 return ret;
349 else if (ret)
350 goto found;
353 *endptr = p;
354 return 0;
356 found:
357 *endptr = p;
358 return 1;
361 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
362 u8 pos, int cap, int *ttl)
364 u8 id;
365 u16 ent;
367 pci_bus_read_config_byte(bus, devfn, pos, &pos);
369 while ((*ttl)--) {
370 if (pos < 0x40)
371 break;
372 pos &= ~3;
373 pci_bus_read_config_word(bus, devfn, pos, &ent);
375 id = ent & 0xff;
376 if (id == 0xff)
377 break;
378 if (id == cap)
379 return pos;
380 pos = (ent >> 8);
382 return 0;
385 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
386 u8 pos, int cap)
388 int ttl = PCI_FIND_CAP_TTL;
390 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
393 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
395 return __pci_find_next_cap(dev->bus, dev->devfn,
396 pos + PCI_CAP_LIST_NEXT, cap);
398 EXPORT_SYMBOL_GPL(pci_find_next_capability);
400 static int __pci_bus_find_cap_start(struct pci_bus *bus,
401 unsigned int devfn, u8 hdr_type)
403 u16 status;
405 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
406 if (!(status & PCI_STATUS_CAP_LIST))
407 return 0;
409 switch (hdr_type) {
410 case PCI_HEADER_TYPE_NORMAL:
411 case PCI_HEADER_TYPE_BRIDGE:
412 return PCI_CAPABILITY_LIST;
413 case PCI_HEADER_TYPE_CARDBUS:
414 return PCI_CB_CAPABILITY_LIST;
417 return 0;
421 * pci_find_capability - query for devices' capabilities
422 * @dev: PCI device to query
423 * @cap: capability code
425 * Tell if a device supports a given PCI capability.
426 * Returns the address of the requested capability structure within the
427 * device's PCI configuration space or 0 in case the device does not
428 * support it. Possible values for @cap:
430 * %PCI_CAP_ID_PM Power Management
431 * %PCI_CAP_ID_AGP Accelerated Graphics Port
432 * %PCI_CAP_ID_VPD Vital Product Data
433 * %PCI_CAP_ID_SLOTID Slot Identification
434 * %PCI_CAP_ID_MSI Message Signalled Interrupts
435 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
436 * %PCI_CAP_ID_PCIX PCI-X
437 * %PCI_CAP_ID_EXP PCI Express
439 int pci_find_capability(struct pci_dev *dev, int cap)
441 int pos;
443 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
444 if (pos)
445 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
447 return pos;
449 EXPORT_SYMBOL(pci_find_capability);
452 * pci_bus_find_capability - query for devices' capabilities
453 * @bus: the PCI bus to query
454 * @devfn: PCI device to query
455 * @cap: capability code
457 * Like pci_find_capability() but works for pci devices that do not have a
458 * pci_dev structure set up yet.
460 * Returns the address of the requested capability structure within the
461 * device's PCI configuration space or 0 in case the device does not
462 * support it.
464 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
466 int pos;
467 u8 hdr_type;
469 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
471 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
472 if (pos)
473 pos = __pci_find_next_cap(bus, devfn, pos, cap);
475 return pos;
477 EXPORT_SYMBOL(pci_bus_find_capability);
480 * pci_find_next_ext_capability - Find an extended capability
481 * @dev: PCI device to query
482 * @start: address at which to start looking (0 to start at beginning of list)
483 * @cap: capability code
485 * Returns the address of the next matching extended capability structure
486 * within the device's PCI configuration space or 0 if the device does
487 * not support it. Some capabilities can occur several times, e.g., the
488 * vendor-specific capability, and this provides a way to find them all.
490 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
492 u32 header;
493 int ttl;
494 int pos = PCI_CFG_SPACE_SIZE;
496 /* minimum 8 bytes per capability */
497 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
499 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
500 return 0;
502 if (start)
503 pos = start;
505 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
506 return 0;
509 * If we have no capabilities, this is indicated by cap ID,
510 * cap version and next pointer all being 0.
512 if (header == 0)
513 return 0;
515 while (ttl-- > 0) {
516 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
517 return pos;
519 pos = PCI_EXT_CAP_NEXT(header);
520 if (pos < PCI_CFG_SPACE_SIZE)
521 break;
523 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
524 break;
527 return 0;
529 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
532 * pci_find_ext_capability - Find an extended capability
533 * @dev: PCI device to query
534 * @cap: capability code
536 * Returns the address of the requested extended capability structure
537 * within the device's PCI configuration space or 0 if the device does
538 * not support it. Possible values for @cap:
540 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
541 * %PCI_EXT_CAP_ID_VC Virtual Channel
542 * %PCI_EXT_CAP_ID_DSN Device Serial Number
543 * %PCI_EXT_CAP_ID_PWR Power Budgeting
545 int pci_find_ext_capability(struct pci_dev *dev, int cap)
547 return pci_find_next_ext_capability(dev, 0, cap);
549 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
551 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
553 int rc, ttl = PCI_FIND_CAP_TTL;
554 u8 cap, mask;
556 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
557 mask = HT_3BIT_CAP_MASK;
558 else
559 mask = HT_5BIT_CAP_MASK;
561 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
562 PCI_CAP_ID_HT, &ttl);
563 while (pos) {
564 rc = pci_read_config_byte(dev, pos + 3, &cap);
565 if (rc != PCIBIOS_SUCCESSFUL)
566 return 0;
568 if ((cap & mask) == ht_cap)
569 return pos;
571 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
572 pos + PCI_CAP_LIST_NEXT,
573 PCI_CAP_ID_HT, &ttl);
576 return 0;
579 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
580 * @dev: PCI device to query
581 * @pos: Position from which to continue searching
582 * @ht_cap: Hypertransport capability code
584 * To be used in conjunction with pci_find_ht_capability() to search for
585 * all capabilities matching @ht_cap. @pos should always be a value returned
586 * from pci_find_ht_capability().
588 * NB. To be 100% safe against broken PCI devices, the caller should take
589 * steps to avoid an infinite loop.
591 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
593 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
595 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
598 * pci_find_ht_capability - query a device's Hypertransport capabilities
599 * @dev: PCI device to query
600 * @ht_cap: Hypertransport capability code
602 * Tell if a device supports a given Hypertransport capability.
603 * Returns an address within the device's PCI configuration space
604 * or 0 in case the device does not support the request capability.
605 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
606 * which has a Hypertransport capability matching @ht_cap.
608 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
610 int pos;
612 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
613 if (pos)
614 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
616 return pos;
618 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
621 * pci_find_parent_resource - return resource region of parent bus of given region
622 * @dev: PCI device structure contains resources to be searched
623 * @res: child resource record for which parent is sought
625 * For given resource region of given device, return the resource
626 * region of parent bus the given region is contained in.
628 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
629 struct resource *res)
631 const struct pci_bus *bus = dev->bus;
632 struct resource *r;
633 int i;
635 pci_bus_for_each_resource(bus, r, i) {
636 if (!r)
637 continue;
638 if (resource_contains(r, res)) {
641 * If the window is prefetchable but the BAR is
642 * not, the allocator made a mistake.
644 if (r->flags & IORESOURCE_PREFETCH &&
645 !(res->flags & IORESOURCE_PREFETCH))
646 return NULL;
649 * If we're below a transparent bridge, there may
650 * be both a positively-decoded aperture and a
651 * subtractively-decoded region that contain the BAR.
652 * We want the positively-decoded one, so this depends
653 * on pci_bus_for_each_resource() giving us those
654 * first.
656 return r;
659 return NULL;
661 EXPORT_SYMBOL(pci_find_parent_resource);
664 * pci_find_resource - Return matching PCI device resource
665 * @dev: PCI device to query
666 * @res: Resource to look for
668 * Goes over standard PCI resources (BARs) and checks if the given resource
669 * is partially or fully contained in any of them. In that case the
670 * matching resource is returned, %NULL otherwise.
672 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
674 int i;
676 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
677 struct resource *r = &dev->resource[i];
679 if (r->start && resource_contains(r, res))
680 return r;
683 return NULL;
685 EXPORT_SYMBOL(pci_find_resource);
688 * pci_find_pcie_root_port - return PCIe Root Port
689 * @dev: PCI device to query
691 * Traverse up the parent chain and return the PCIe Root Port PCI Device
692 * for a given PCI Device.
694 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
696 struct pci_dev *bridge, *highest_pcie_bridge = dev;
698 bridge = pci_upstream_bridge(dev);
699 while (bridge && pci_is_pcie(bridge)) {
700 highest_pcie_bridge = bridge;
701 bridge = pci_upstream_bridge(bridge);
704 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
705 return NULL;
707 return highest_pcie_bridge;
709 EXPORT_SYMBOL(pci_find_pcie_root_port);
712 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
713 * @dev: the PCI device to operate on
714 * @pos: config space offset of status word
715 * @mask: mask of bit(s) to care about in status word
717 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
719 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
721 int i;
723 /* Wait for Transaction Pending bit clean */
724 for (i = 0; i < 4; i++) {
725 u16 status;
726 if (i)
727 msleep((1 << (i - 1)) * 100);
729 pci_read_config_word(dev, pos, &status);
730 if (!(status & mask))
731 return 1;
734 return 0;
738 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
739 * @dev: PCI device to have its BARs restored
741 * Restore the BAR values for a given device, so as to make it
742 * accessible by its driver.
744 static void pci_restore_bars(struct pci_dev *dev)
746 int i;
748 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
749 pci_update_resource(dev, i);
752 static const struct pci_platform_pm_ops *pci_platform_pm;
754 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
756 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
757 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
758 return -EINVAL;
759 pci_platform_pm = ops;
760 return 0;
763 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
765 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
768 static inline int platform_pci_set_power_state(struct pci_dev *dev,
769 pci_power_t t)
771 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
774 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
776 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
779 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
781 return pci_platform_pm ?
782 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
785 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
787 return pci_platform_pm ?
788 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
791 static inline bool platform_pci_need_resume(struct pci_dev *dev)
793 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
797 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
798 * given PCI device
799 * @dev: PCI device to handle.
800 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
802 * RETURN VALUE:
803 * -EINVAL if the requested state is invalid.
804 * -EIO if device does not support PCI PM or its PM capabilities register has a
805 * wrong version, or device doesn't support the requested state.
806 * 0 if device already is in the requested state.
807 * 0 if device's power state has been successfully changed.
809 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
811 u16 pmcsr;
812 bool need_restore = false;
814 /* Check if we're already there */
815 if (dev->current_state == state)
816 return 0;
818 if (!dev->pm_cap)
819 return -EIO;
821 if (state < PCI_D0 || state > PCI_D3hot)
822 return -EINVAL;
824 /* Validate current state:
825 * Can enter D0 from any state, but if we can only go deeper
826 * to sleep if we're already in a low power state
828 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
829 && dev->current_state > state) {
830 pci_err(dev, "invalid power transition (from state %d to %d)\n",
831 dev->current_state, state);
832 return -EINVAL;
835 /* check if this device supports the desired state */
836 if ((state == PCI_D1 && !dev->d1_support)
837 || (state == PCI_D2 && !dev->d2_support))
838 return -EIO;
840 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
842 /* If we're (effectively) in D3, force entire word to 0.
843 * This doesn't affect PME_Status, disables PME_En, and
844 * sets PowerState to 0.
846 switch (dev->current_state) {
847 case PCI_D0:
848 case PCI_D1:
849 case PCI_D2:
850 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
851 pmcsr |= state;
852 break;
853 case PCI_D3hot:
854 case PCI_D3cold:
855 case PCI_UNKNOWN: /* Boot-up */
856 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
857 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
858 need_restore = true;
859 /* Fall-through: force to D0 */
860 default:
861 pmcsr = 0;
862 break;
865 /* enter specified state */
866 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
868 /* Mandatory power management transition delays */
869 /* see PCI PM 1.1 5.6.1 table 18 */
870 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
871 pci_dev_d3_sleep(dev);
872 else if (state == PCI_D2 || dev->current_state == PCI_D2)
873 udelay(PCI_PM_D2_DELAY);
875 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
876 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
877 if (dev->current_state != state && printk_ratelimit())
878 pci_info(dev, "Refused to change power state, currently in D%d\n",
879 dev->current_state);
882 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
883 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
884 * from D3hot to D0 _may_ perform an internal reset, thereby
885 * going to "D0 Uninitialized" rather than "D0 Initialized".
886 * For example, at least some versions of the 3c905B and the
887 * 3c556B exhibit this behaviour.
889 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
890 * devices in a D3hot state at boot. Consequently, we need to
891 * restore at least the BARs so that the device will be
892 * accessible to its driver.
894 if (need_restore)
895 pci_restore_bars(dev);
897 if (dev->bus->self)
898 pcie_aspm_pm_state_change(dev->bus->self);
900 return 0;
904 * pci_update_current_state - Read power state of given device and cache it
905 * @dev: PCI device to handle.
906 * @state: State to cache in case the device doesn't have the PM capability
908 * The power state is read from the PMCSR register, which however is
909 * inaccessible in D3cold. The platform firmware is therefore queried first
910 * to detect accessibility of the register. In case the platform firmware
911 * reports an incorrect state or the device isn't power manageable by the
912 * platform at all, we try to detect D3cold by testing accessibility of the
913 * vendor ID in config space.
915 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
917 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
918 !pci_device_is_present(dev)) {
919 dev->current_state = PCI_D3cold;
920 } else if (dev->pm_cap) {
921 u16 pmcsr;
923 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
924 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
925 } else {
926 dev->current_state = state;
931 * pci_platform_power_transition - Use platform to change device power state
932 * @dev: PCI device to handle.
933 * @state: State to put the device into.
935 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
937 int error;
939 if (platform_pci_power_manageable(dev)) {
940 error = platform_pci_set_power_state(dev, state);
941 if (!error)
942 pci_update_current_state(dev, state);
943 } else
944 error = -ENODEV;
946 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
947 dev->current_state = PCI_D0;
949 return error;
953 * pci_wakeup - Wake up a PCI device
954 * @pci_dev: Device to handle.
955 * @ign: ignored parameter
957 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
959 pci_wakeup_event(pci_dev);
960 pm_request_resume(&pci_dev->dev);
961 return 0;
965 * pci_wakeup_bus - Walk given bus and wake up devices on it
966 * @bus: Top bus of the subtree to walk.
968 void pci_wakeup_bus(struct pci_bus *bus)
970 if (bus)
971 pci_walk_bus(bus, pci_wakeup, NULL);
975 * __pci_start_power_transition - Start power transition of a PCI device
976 * @dev: PCI device to handle.
977 * @state: State to put the device into.
979 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
981 if (state == PCI_D0) {
982 pci_platform_power_transition(dev, PCI_D0);
984 * Mandatory power management transition delays, see
985 * PCI Express Base Specification Revision 2.0 Section
986 * 6.6.1: Conventional Reset. Do not delay for
987 * devices powered on/off by corresponding bridge,
988 * because have already delayed for the bridge.
990 if (dev->runtime_d3cold) {
991 if (dev->d3cold_delay)
992 msleep(dev->d3cold_delay);
994 * When powering on a bridge from D3cold, the
995 * whole hierarchy may be powered on into
996 * D0uninitialized state, resume them to give
997 * them a chance to suspend again
999 pci_wakeup_bus(dev->subordinate);
1005 * __pci_dev_set_current_state - Set current state of a PCI device
1006 * @dev: Device to handle
1007 * @data: pointer to state to be set
1009 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
1011 pci_power_t state = *(pci_power_t *)data;
1013 dev->current_state = state;
1014 return 0;
1018 * pci_bus_set_current_state - Walk given bus and set current state of devices
1019 * @bus: Top bus of the subtree to walk.
1020 * @state: state to be set
1022 void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
1024 if (bus)
1025 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
1029 * __pci_complete_power_transition - Complete power transition of a PCI device
1030 * @dev: PCI device to handle.
1031 * @state: State to put the device into.
1033 * This function should not be called directly by device drivers.
1035 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
1037 int ret;
1039 if (state <= PCI_D0)
1040 return -EINVAL;
1041 ret = pci_platform_power_transition(dev, state);
1042 /* Power off the bridge may power off the whole hierarchy */
1043 if (!ret && state == PCI_D3cold)
1044 pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
1045 return ret;
1047 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
1050 * pci_set_power_state - Set the power state of a PCI device
1051 * @dev: PCI device to handle.
1052 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
1054 * Transition a device to a new power state, using the platform firmware and/or
1055 * the device's PCI PM registers.
1057 * RETURN VALUE:
1058 * -EINVAL if the requested state is invalid.
1059 * -EIO if device does not support PCI PM or its PM capabilities register has a
1060 * wrong version, or device doesn't support the requested state.
1061 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
1062 * 0 if device already is in the requested state.
1063 * 0 if the transition is to D3 but D3 is not supported.
1064 * 0 if device's power state has been successfully changed.
1066 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
1068 int error;
1070 /* bound the state we're entering */
1071 if (state > PCI_D3cold)
1072 state = PCI_D3cold;
1073 else if (state < PCI_D0)
1074 state = PCI_D0;
1075 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
1077 * If the device or the parent bridge do not support PCI PM,
1078 * ignore the request if we're doing anything other than putting
1079 * it into D0 (which would only happen on boot).
1081 return 0;
1083 /* Check if we're already there */
1084 if (dev->current_state == state)
1085 return 0;
1087 __pci_start_power_transition(dev, state);
1089 /* This device is quirked not to be put into D3, so
1090 don't put it in D3 */
1091 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
1092 return 0;
1095 * To put device in D3cold, we put device into D3hot in native
1096 * way, then put device into D3cold with platform ops
1098 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
1099 PCI_D3hot : state);
1101 if (!__pci_complete_power_transition(dev, state))
1102 error = 0;
1104 return error;
1106 EXPORT_SYMBOL(pci_set_power_state);
1109 * pci_power_up - Put the given device into D0 forcibly
1110 * @dev: PCI device to power up
1112 void pci_power_up(struct pci_dev *dev)
1114 __pci_start_power_transition(dev, PCI_D0);
1115 pci_raw_set_power_state(dev, PCI_D0);
1116 pci_update_current_state(dev, PCI_D0);
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 = ilog2(resource_size(res)) - 20;
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;
2008 * If the device is in D3cold it should not be
2009 * polled either.
2011 if (pme_dev->dev->current_state == PCI_D3cold)
2012 continue;
2014 pci_pme_wakeup(pme_dev->dev, NULL);
2015 } else {
2016 list_del(&pme_dev->list);
2017 kfree(pme_dev);
2020 if (!list_empty(&pci_pme_list))
2021 queue_delayed_work(system_freezable_wq, &pci_pme_work,
2022 msecs_to_jiffies(PME_TIMEOUT));
2023 mutex_unlock(&pci_pme_list_mutex);
2026 static void __pci_pme_active(struct pci_dev *dev, bool enable)
2028 u16 pmcsr;
2030 if (!dev->pme_support)
2031 return;
2033 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2034 /* Clear PME_Status by writing 1 to it and enable PME# */
2035 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
2036 if (!enable)
2037 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2039 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2043 * pci_pme_restore - Restore PME configuration after config space restore.
2044 * @dev: PCI device to update.
2046 void pci_pme_restore(struct pci_dev *dev)
2048 u16 pmcsr;
2050 if (!dev->pme_support)
2051 return;
2053 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
2054 if (dev->wakeup_prepared) {
2055 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
2056 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
2057 } else {
2058 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
2059 pmcsr |= PCI_PM_CTRL_PME_STATUS;
2061 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
2065 * pci_pme_active - enable or disable PCI device's PME# function
2066 * @dev: PCI device to handle.
2067 * @enable: 'true' to enable PME# generation; 'false' to disable it.
2069 * The caller must verify that the device is capable of generating PME# before
2070 * calling this function with @enable equal to 'true'.
2072 void pci_pme_active(struct pci_dev *dev, bool enable)
2074 __pci_pme_active(dev, enable);
2077 * PCI (as opposed to PCIe) PME requires that the device have
2078 * its PME# line hooked up correctly. Not all hardware vendors
2079 * do this, so the PME never gets delivered and the device
2080 * remains asleep. The easiest way around this is to
2081 * periodically walk the list of suspended devices and check
2082 * whether any have their PME flag set. The assumption is that
2083 * we'll wake up often enough anyway that this won't be a huge
2084 * hit, and the power savings from the devices will still be a
2085 * win.
2087 * Although PCIe uses in-band PME message instead of PME# line
2088 * to report PME, PME does not work for some PCIe devices in
2089 * reality. For example, there are devices that set their PME
2090 * status bits, but don't really bother to send a PME message;
2091 * there are PCI Express Root Ports that don't bother to
2092 * trigger interrupts when they receive PME messages from the
2093 * devices below. So PME poll is used for PCIe devices too.
2096 if (dev->pme_poll) {
2097 struct pci_pme_device *pme_dev;
2098 if (enable) {
2099 pme_dev = kmalloc(sizeof(struct pci_pme_device),
2100 GFP_KERNEL);
2101 if (!pme_dev) {
2102 pci_warn(dev, "can't enable PME#\n");
2103 return;
2105 pme_dev->dev = dev;
2106 mutex_lock(&pci_pme_list_mutex);
2107 list_add(&pme_dev->list, &pci_pme_list);
2108 if (list_is_singular(&pci_pme_list))
2109 queue_delayed_work(system_freezable_wq,
2110 &pci_pme_work,
2111 msecs_to_jiffies(PME_TIMEOUT));
2112 mutex_unlock(&pci_pme_list_mutex);
2113 } else {
2114 mutex_lock(&pci_pme_list_mutex);
2115 list_for_each_entry(pme_dev, &pci_pme_list, list) {
2116 if (pme_dev->dev == dev) {
2117 list_del(&pme_dev->list);
2118 kfree(pme_dev);
2119 break;
2122 mutex_unlock(&pci_pme_list_mutex);
2126 pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
2128 EXPORT_SYMBOL(pci_pme_active);
2131 * __pci_enable_wake - enable PCI device as wakeup event source
2132 * @dev: PCI device affected
2133 * @state: PCI state from which device will issue wakeup events
2134 * @enable: True to enable event generation; false to disable
2136 * This enables the device as a wakeup event source, or disables it.
2137 * When such events involves platform-specific hooks, those hooks are
2138 * called automatically by this routine.
2140 * Devices with legacy power management (no standard PCI PM capabilities)
2141 * always require such platform hooks.
2143 * RETURN VALUE:
2144 * 0 is returned on success
2145 * -EINVAL is returned if device is not supposed to wake up the system
2146 * Error code depending on the platform is returned if both the platform and
2147 * the native mechanism fail to enable the generation of wake-up events
2149 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
2151 int ret = 0;
2154 * Bridges can only signal wakeup on behalf of subordinate devices,
2155 * but that is set up elsewhere, so skip them.
2157 if (pci_has_subordinate(dev))
2158 return 0;
2160 /* Don't do the same thing twice in a row for one device. */
2161 if (!!enable == !!dev->wakeup_prepared)
2162 return 0;
2165 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
2166 * Anderson we should be doing PME# wake enable followed by ACPI wake
2167 * enable. To disable wake-up we call the platform first, for symmetry.
2170 if (enable) {
2171 int error;
2173 if (pci_pme_capable(dev, state))
2174 pci_pme_active(dev, true);
2175 else
2176 ret = 1;
2177 error = platform_pci_set_wakeup(dev, true);
2178 if (ret)
2179 ret = error;
2180 if (!ret)
2181 dev->wakeup_prepared = true;
2182 } else {
2183 platform_pci_set_wakeup(dev, false);
2184 pci_pme_active(dev, false);
2185 dev->wakeup_prepared = false;
2188 return ret;
2192 * pci_enable_wake - change wakeup settings for a PCI device
2193 * @pci_dev: Target device
2194 * @state: PCI state from which device will issue wakeup events
2195 * @enable: Whether or not to enable event generation
2197 * If @enable is set, check device_may_wakeup() for the device before calling
2198 * __pci_enable_wake() for it.
2200 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
2202 if (enable && !device_may_wakeup(&pci_dev->dev))
2203 return -EINVAL;
2205 return __pci_enable_wake(pci_dev, state, enable);
2207 EXPORT_SYMBOL(pci_enable_wake);
2210 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
2211 * @dev: PCI device to prepare
2212 * @enable: True to enable wake-up event generation; false to disable
2214 * Many drivers want the device to wake up the system from D3_hot or D3_cold
2215 * and this function allows them to set that up cleanly - pci_enable_wake()
2216 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
2217 * ordering constraints.
2219 * This function only returns error code if the device is not allowed to wake
2220 * up the system from sleep or it is not capable of generating PME# from both
2221 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2223 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2225 return pci_pme_capable(dev, PCI_D3cold) ?
2226 pci_enable_wake(dev, PCI_D3cold, enable) :
2227 pci_enable_wake(dev, PCI_D3hot, enable);
2229 EXPORT_SYMBOL(pci_wake_from_d3);
2232 * pci_target_state - find an appropriate low power state for a given PCI dev
2233 * @dev: PCI device
2234 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2236 * Use underlying platform code to find a supported low power state for @dev.
2237 * If the platform can't manage @dev, return the deepest state from which it
2238 * can generate wake events, based on any available PME info.
2240 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2242 pci_power_t target_state = PCI_D3hot;
2244 if (platform_pci_power_manageable(dev)) {
2246 * Call the platform to find the target state for the device.
2248 pci_power_t state = platform_pci_choose_state(dev);
2250 switch (state) {
2251 case PCI_POWER_ERROR:
2252 case PCI_UNKNOWN:
2253 break;
2254 case PCI_D1:
2255 case PCI_D2:
2256 if (pci_no_d1d2(dev))
2257 break;
2258 /* else: fall through */
2259 default:
2260 target_state = state;
2263 return target_state;
2266 if (!dev->pm_cap)
2267 target_state = PCI_D0;
2270 * If the device is in D3cold even though it's not power-manageable by
2271 * the platform, it may have been powered down by non-standard means.
2272 * Best to let it slumber.
2274 if (dev->current_state == PCI_D3cold)
2275 target_state = PCI_D3cold;
2277 if (wakeup) {
2279 * Find the deepest state from which the device can generate
2280 * PME#.
2282 if (dev->pme_support) {
2283 while (target_state
2284 && !(dev->pme_support & (1 << target_state)))
2285 target_state--;
2289 return target_state;
2293 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
2294 * @dev: Device to handle.
2296 * Choose the power state appropriate for the device depending on whether
2297 * it can wake up the system and/or is power manageable by the platform
2298 * (PCI_D3hot is the default) and put the device into that state.
2300 int pci_prepare_to_sleep(struct pci_dev *dev)
2302 bool wakeup = device_may_wakeup(&dev->dev);
2303 pci_power_t target_state = pci_target_state(dev, wakeup);
2304 int error;
2306 if (target_state == PCI_POWER_ERROR)
2307 return -EIO;
2309 pci_enable_wake(dev, target_state, wakeup);
2311 error = pci_set_power_state(dev, target_state);
2313 if (error)
2314 pci_enable_wake(dev, target_state, false);
2316 return error;
2318 EXPORT_SYMBOL(pci_prepare_to_sleep);
2321 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2322 * @dev: Device to handle.
2324 * Disable device's system wake-up capability and put it into D0.
2326 int pci_back_from_sleep(struct pci_dev *dev)
2328 pci_enable_wake(dev, PCI_D0, false);
2329 return pci_set_power_state(dev, PCI_D0);
2331 EXPORT_SYMBOL(pci_back_from_sleep);
2334 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2335 * @dev: PCI device being suspended.
2337 * Prepare @dev to generate wake-up events at run time and put it into a low
2338 * power state.
2340 int pci_finish_runtime_suspend(struct pci_dev *dev)
2342 pci_power_t target_state;
2343 int error;
2345 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2346 if (target_state == PCI_POWER_ERROR)
2347 return -EIO;
2349 dev->runtime_d3cold = target_state == PCI_D3cold;
2351 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2353 error = pci_set_power_state(dev, target_state);
2355 if (error) {
2356 pci_enable_wake(dev, target_state, false);
2357 dev->runtime_d3cold = false;
2360 return error;
2364 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2365 * @dev: Device to check.
2367 * Return true if the device itself is capable of generating wake-up events
2368 * (through the platform or using the native PCIe PME) or if the device supports
2369 * PME and one of its upstream bridges can generate wake-up events.
2371 bool pci_dev_run_wake(struct pci_dev *dev)
2373 struct pci_bus *bus = dev->bus;
2375 if (!dev->pme_support)
2376 return false;
2378 /* PME-capable in principle, but not from the target power state */
2379 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2380 return false;
2382 if (device_can_wakeup(&dev->dev))
2383 return true;
2385 while (bus->parent) {
2386 struct pci_dev *bridge = bus->self;
2388 if (device_can_wakeup(&bridge->dev))
2389 return true;
2391 bus = bus->parent;
2394 /* We have reached the root bus. */
2395 if (bus->bridge)
2396 return device_can_wakeup(bus->bridge);
2398 return false;
2400 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2403 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2404 * @pci_dev: Device to check.
2406 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2407 * reconfigured due to wakeup settings difference between system and runtime
2408 * suspend and the current power state of it is suitable for the upcoming
2409 * (system) transition.
2411 * If the device is not configured for system wakeup, disable PME for it before
2412 * returning 'true' to prevent it from waking up the system unnecessarily.
2414 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2416 struct device *dev = &pci_dev->dev;
2417 bool wakeup = device_may_wakeup(dev);
2419 if (!pm_runtime_suspended(dev)
2420 || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2421 || platform_pci_need_resume(pci_dev))
2422 return false;
2425 * At this point the device is good to go unless it's been configured
2426 * to generate PME at the runtime suspend time, but it is not supposed
2427 * to wake up the system. In that case, simply disable PME for it
2428 * (it will have to be re-enabled on exit from system resume).
2430 * If the device's power state is D3cold and the platform check above
2431 * hasn't triggered, the device's configuration is suitable and we don't
2432 * need to manipulate it at all.
2434 spin_lock_irq(&dev->power.lock);
2436 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2437 !wakeup)
2438 __pci_pme_active(pci_dev, false);
2440 spin_unlock_irq(&dev->power.lock);
2441 return true;
2445 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2446 * @pci_dev: Device to handle.
2448 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2449 * it might have been disabled during the prepare phase of system suspend if
2450 * the device was not configured for system wakeup.
2452 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2454 struct device *dev = &pci_dev->dev;
2456 if (!pci_dev_run_wake(pci_dev))
2457 return;
2459 spin_lock_irq(&dev->power.lock);
2461 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2462 __pci_pme_active(pci_dev, true);
2464 spin_unlock_irq(&dev->power.lock);
2467 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2469 struct device *dev = &pdev->dev;
2470 struct device *parent = dev->parent;
2472 if (parent)
2473 pm_runtime_get_sync(parent);
2474 pm_runtime_get_noresume(dev);
2476 * pdev->current_state is set to PCI_D3cold during suspending,
2477 * so wait until suspending completes
2479 pm_runtime_barrier(dev);
2481 * Only need to resume devices in D3cold, because config
2482 * registers are still accessible for devices suspended but
2483 * not in D3cold.
2485 if (pdev->current_state == PCI_D3cold)
2486 pm_runtime_resume(dev);
2489 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2491 struct device *dev = &pdev->dev;
2492 struct device *parent = dev->parent;
2494 pm_runtime_put(dev);
2495 if (parent)
2496 pm_runtime_put_sync(parent);
2499 static const struct dmi_system_id bridge_d3_blacklist[] = {
2500 #ifdef CONFIG_X86
2503 * Gigabyte X299 root port is not marked as hotplug capable
2504 * which allows Linux to power manage it. However, this
2505 * confuses the BIOS SMI handler so don't power manage root
2506 * ports on that system.
2508 .ident = "X299 DESIGNARE EX-CF",
2509 .matches = {
2510 DMI_MATCH(DMI_BOARD_VENDOR, "Gigabyte Technology Co., Ltd."),
2511 DMI_MATCH(DMI_BOARD_NAME, "X299 DESIGNARE EX-CF"),
2514 #endif
2519 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2520 * @bridge: Bridge to check
2522 * This function checks if it is possible to move the bridge to D3.
2523 * Currently we only allow D3 for recent enough PCIe ports and Thunderbolt.
2525 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2527 if (!pci_is_pcie(bridge))
2528 return false;
2530 switch (pci_pcie_type(bridge)) {
2531 case PCI_EXP_TYPE_ROOT_PORT:
2532 case PCI_EXP_TYPE_UPSTREAM:
2533 case PCI_EXP_TYPE_DOWNSTREAM:
2534 if (pci_bridge_d3_disable)
2535 return false;
2538 * Hotplug ports handled by firmware in System Management Mode
2539 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2541 if (bridge->is_hotplug_bridge && !pciehp_is_native(bridge))
2542 return false;
2544 if (pci_bridge_d3_force)
2545 return true;
2547 /* Even the oldest 2010 Thunderbolt controller supports D3. */
2548 if (bridge->is_thunderbolt)
2549 return true;
2552 * Hotplug ports handled natively by the OS were not validated
2553 * by vendors for runtime D3 at least until 2018 because there
2554 * was no OS support.
2556 if (bridge->is_hotplug_bridge)
2557 return false;
2559 if (dmi_check_system(bridge_d3_blacklist))
2560 return false;
2563 * It should be safe to put PCIe ports from 2015 or newer
2564 * to D3.
2566 if (dmi_get_bios_year() >= 2015)
2567 return true;
2568 break;
2571 return false;
2574 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2576 bool *d3cold_ok = data;
2578 if (/* The device needs to be allowed to go D3cold ... */
2579 dev->no_d3cold || !dev->d3cold_allowed ||
2581 /* ... and if it is wakeup capable to do so from D3cold. */
2582 (device_may_wakeup(&dev->dev) &&
2583 !pci_pme_capable(dev, PCI_D3cold)) ||
2585 /* If it is a bridge it must be allowed to go to D3. */
2586 !pci_power_manageable(dev))
2588 *d3cold_ok = false;
2590 return !*d3cold_ok;
2594 * pci_bridge_d3_update - Update bridge D3 capabilities
2595 * @dev: PCI device which is changed
2597 * Update upstream bridge PM capabilities accordingly depending on if the
2598 * device PM configuration was changed or the device is being removed. The
2599 * change is also propagated upstream.
2601 void pci_bridge_d3_update(struct pci_dev *dev)
2603 bool remove = !device_is_registered(&dev->dev);
2604 struct pci_dev *bridge;
2605 bool d3cold_ok = true;
2607 bridge = pci_upstream_bridge(dev);
2608 if (!bridge || !pci_bridge_d3_possible(bridge))
2609 return;
2612 * If D3 is currently allowed for the bridge, removing one of its
2613 * children won't change that.
2615 if (remove && bridge->bridge_d3)
2616 return;
2619 * If D3 is currently allowed for the bridge and a child is added or
2620 * changed, disallowance of D3 can only be caused by that child, so
2621 * we only need to check that single device, not any of its siblings.
2623 * If D3 is currently not allowed for the bridge, checking the device
2624 * first may allow us to skip checking its siblings.
2626 if (!remove)
2627 pci_dev_check_d3cold(dev, &d3cold_ok);
2630 * If D3 is currently not allowed for the bridge, this may be caused
2631 * either by the device being changed/removed or any of its siblings,
2632 * so we need to go through all children to find out if one of them
2633 * continues to block D3.
2635 if (d3cold_ok && !bridge->bridge_d3)
2636 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2637 &d3cold_ok);
2639 if (bridge->bridge_d3 != d3cold_ok) {
2640 bridge->bridge_d3 = d3cold_ok;
2641 /* Propagate change to upstream bridges */
2642 pci_bridge_d3_update(bridge);
2647 * pci_d3cold_enable - Enable D3cold for device
2648 * @dev: PCI device to handle
2650 * This function can be used in drivers to enable D3cold from the device
2651 * they handle. It also updates upstream PCI bridge PM capabilities
2652 * accordingly.
2654 void pci_d3cold_enable(struct pci_dev *dev)
2656 if (dev->no_d3cold) {
2657 dev->no_d3cold = false;
2658 pci_bridge_d3_update(dev);
2661 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2664 * pci_d3cold_disable - Disable D3cold for device
2665 * @dev: PCI device to handle
2667 * This function can be used in drivers to disable D3cold from the device
2668 * they handle. It also updates upstream PCI bridge PM capabilities
2669 * accordingly.
2671 void pci_d3cold_disable(struct pci_dev *dev)
2673 if (!dev->no_d3cold) {
2674 dev->no_d3cold = true;
2675 pci_bridge_d3_update(dev);
2678 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2681 * pci_pm_init - Initialize PM functions of given PCI device
2682 * @dev: PCI device to handle.
2684 void pci_pm_init(struct pci_dev *dev)
2686 int pm;
2687 u16 pmc;
2689 pm_runtime_forbid(&dev->dev);
2690 pm_runtime_set_active(&dev->dev);
2691 pm_runtime_enable(&dev->dev);
2692 device_enable_async_suspend(&dev->dev);
2693 dev->wakeup_prepared = false;
2695 dev->pm_cap = 0;
2696 dev->pme_support = 0;
2698 /* find PCI PM capability in list */
2699 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2700 if (!pm)
2701 return;
2702 /* Check device's ability to generate PME# */
2703 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2705 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2706 pci_err(dev, "unsupported PM cap regs version (%u)\n",
2707 pmc & PCI_PM_CAP_VER_MASK);
2708 return;
2711 dev->pm_cap = pm;
2712 dev->d3_delay = PCI_PM_D3_WAIT;
2713 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2714 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2715 dev->d3cold_allowed = true;
2717 dev->d1_support = false;
2718 dev->d2_support = false;
2719 if (!pci_no_d1d2(dev)) {
2720 if (pmc & PCI_PM_CAP_D1)
2721 dev->d1_support = true;
2722 if (pmc & PCI_PM_CAP_D2)
2723 dev->d2_support = true;
2725 if (dev->d1_support || dev->d2_support)
2726 pci_printk(KERN_DEBUG, dev, "supports%s%s\n",
2727 dev->d1_support ? " D1" : "",
2728 dev->d2_support ? " D2" : "");
2731 pmc &= PCI_PM_CAP_PME_MASK;
2732 if (pmc) {
2733 pci_printk(KERN_DEBUG, dev, "PME# supported from%s%s%s%s%s\n",
2734 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2735 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2736 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2737 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2738 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2739 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2740 dev->pme_poll = true;
2742 * Make device's PM flags reflect the wake-up capability, but
2743 * let the user space enable it to wake up the system as needed.
2745 device_set_wakeup_capable(&dev->dev, true);
2746 /* Disable the PME# generation functionality */
2747 pci_pme_active(dev, false);
2751 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2753 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2755 switch (prop) {
2756 case PCI_EA_P_MEM:
2757 case PCI_EA_P_VF_MEM:
2758 flags |= IORESOURCE_MEM;
2759 break;
2760 case PCI_EA_P_MEM_PREFETCH:
2761 case PCI_EA_P_VF_MEM_PREFETCH:
2762 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2763 break;
2764 case PCI_EA_P_IO:
2765 flags |= IORESOURCE_IO;
2766 break;
2767 default:
2768 return 0;
2771 return flags;
2774 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2775 u8 prop)
2777 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2778 return &dev->resource[bei];
2779 #ifdef CONFIG_PCI_IOV
2780 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2781 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2782 return &dev->resource[PCI_IOV_RESOURCES +
2783 bei - PCI_EA_BEI_VF_BAR0];
2784 #endif
2785 else if (bei == PCI_EA_BEI_ROM)
2786 return &dev->resource[PCI_ROM_RESOURCE];
2787 else
2788 return NULL;
2791 /* Read an Enhanced Allocation (EA) entry */
2792 static int pci_ea_read(struct pci_dev *dev, int offset)
2794 struct resource *res;
2795 int ent_size, ent_offset = offset;
2796 resource_size_t start, end;
2797 unsigned long flags;
2798 u32 dw0, bei, base, max_offset;
2799 u8 prop;
2800 bool support_64 = (sizeof(resource_size_t) >= 8);
2802 pci_read_config_dword(dev, ent_offset, &dw0);
2803 ent_offset += 4;
2805 /* Entry size field indicates DWORDs after 1st */
2806 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2808 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2809 goto out;
2811 bei = (dw0 & PCI_EA_BEI) >> 4;
2812 prop = (dw0 & PCI_EA_PP) >> 8;
2815 * If the Property is in the reserved range, try the Secondary
2816 * Property instead.
2818 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2819 prop = (dw0 & PCI_EA_SP) >> 16;
2820 if (prop > PCI_EA_P_BRIDGE_IO)
2821 goto out;
2823 res = pci_ea_get_resource(dev, bei, prop);
2824 if (!res) {
2825 pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
2826 goto out;
2829 flags = pci_ea_flags(dev, prop);
2830 if (!flags) {
2831 pci_err(dev, "Unsupported EA properties: %#x\n", prop);
2832 goto out;
2835 /* Read Base */
2836 pci_read_config_dword(dev, ent_offset, &base);
2837 start = (base & PCI_EA_FIELD_MASK);
2838 ent_offset += 4;
2840 /* Read MaxOffset */
2841 pci_read_config_dword(dev, ent_offset, &max_offset);
2842 ent_offset += 4;
2844 /* Read Base MSBs (if 64-bit entry) */
2845 if (base & PCI_EA_IS_64) {
2846 u32 base_upper;
2848 pci_read_config_dword(dev, ent_offset, &base_upper);
2849 ent_offset += 4;
2851 flags |= IORESOURCE_MEM_64;
2853 /* entry starts above 32-bit boundary, can't use */
2854 if (!support_64 && base_upper)
2855 goto out;
2857 if (support_64)
2858 start |= ((u64)base_upper << 32);
2861 end = start + (max_offset | 0x03);
2863 /* Read MaxOffset MSBs (if 64-bit entry) */
2864 if (max_offset & PCI_EA_IS_64) {
2865 u32 max_offset_upper;
2867 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2868 ent_offset += 4;
2870 flags |= IORESOURCE_MEM_64;
2872 /* entry too big, can't use */
2873 if (!support_64 && max_offset_upper)
2874 goto out;
2876 if (support_64)
2877 end += ((u64)max_offset_upper << 32);
2880 if (end < start) {
2881 pci_err(dev, "EA Entry crosses address boundary\n");
2882 goto out;
2885 if (ent_size != ent_offset - offset) {
2886 pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
2887 ent_size, ent_offset - offset);
2888 goto out;
2891 res->name = pci_name(dev);
2892 res->start = start;
2893 res->end = end;
2894 res->flags = flags;
2896 if (bei <= PCI_EA_BEI_BAR5)
2897 pci_printk(KERN_DEBUG, dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2898 bei, res, prop);
2899 else if (bei == PCI_EA_BEI_ROM)
2900 pci_printk(KERN_DEBUG, dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2901 res, prop);
2902 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2903 pci_printk(KERN_DEBUG, dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2904 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2905 else
2906 pci_printk(KERN_DEBUG, dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2907 bei, res, prop);
2909 out:
2910 return offset + ent_size;
2913 /* Enhanced Allocation Initialization */
2914 void pci_ea_init(struct pci_dev *dev)
2916 int ea;
2917 u8 num_ent;
2918 int offset;
2919 int i;
2921 /* find PCI EA capability in list */
2922 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2923 if (!ea)
2924 return;
2926 /* determine the number of entries */
2927 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2928 &num_ent);
2929 num_ent &= PCI_EA_NUM_ENT_MASK;
2931 offset = ea + PCI_EA_FIRST_ENT;
2933 /* Skip DWORD 2 for type 1 functions */
2934 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2935 offset += 4;
2937 /* parse each EA entry */
2938 for (i = 0; i < num_ent; ++i)
2939 offset = pci_ea_read(dev, offset);
2942 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2943 struct pci_cap_saved_state *new_cap)
2945 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2949 * _pci_add_cap_save_buffer - allocate buffer for saving given
2950 * capability registers
2951 * @dev: the PCI device
2952 * @cap: the capability to allocate the buffer for
2953 * @extended: Standard or Extended capability ID
2954 * @size: requested size of the buffer
2956 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2957 bool extended, unsigned int size)
2959 int pos;
2960 struct pci_cap_saved_state *save_state;
2962 if (extended)
2963 pos = pci_find_ext_capability(dev, cap);
2964 else
2965 pos = pci_find_capability(dev, cap);
2967 if (!pos)
2968 return 0;
2970 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2971 if (!save_state)
2972 return -ENOMEM;
2974 save_state->cap.cap_nr = cap;
2975 save_state->cap.cap_extended = extended;
2976 save_state->cap.size = size;
2977 pci_add_saved_cap(dev, save_state);
2979 return 0;
2982 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2984 return _pci_add_cap_save_buffer(dev, cap, false, size);
2987 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2989 return _pci_add_cap_save_buffer(dev, cap, true, size);
2993 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2994 * @dev: the PCI device
2996 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2998 int error;
3000 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
3001 PCI_EXP_SAVE_REGS * sizeof(u16));
3002 if (error)
3003 pci_err(dev, "unable to preallocate PCI Express save buffer\n");
3005 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
3006 if (error)
3007 pci_err(dev, "unable to preallocate PCI-X save buffer\n");
3009 pci_allocate_vc_save_buffers(dev);
3012 void pci_free_cap_save_buffers(struct pci_dev *dev)
3014 struct pci_cap_saved_state *tmp;
3015 struct hlist_node *n;
3017 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
3018 kfree(tmp);
3022 * pci_configure_ari - enable or disable ARI forwarding
3023 * @dev: the PCI device
3025 * If @dev and its upstream bridge both support ARI, enable ARI in the
3026 * bridge. Otherwise, disable ARI in the bridge.
3028 void pci_configure_ari(struct pci_dev *dev)
3030 u32 cap;
3031 struct pci_dev *bridge;
3033 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
3034 return;
3036 bridge = dev->bus->self;
3037 if (!bridge)
3038 return;
3040 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3041 if (!(cap & PCI_EXP_DEVCAP2_ARI))
3042 return;
3044 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
3045 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
3046 PCI_EXP_DEVCTL2_ARI);
3047 bridge->ari_enabled = 1;
3048 } else {
3049 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
3050 PCI_EXP_DEVCTL2_ARI);
3051 bridge->ari_enabled = 0;
3055 static int pci_acs_enable;
3058 * pci_request_acs - ask for ACS to be enabled if supported
3060 void pci_request_acs(void)
3062 pci_acs_enable = 1;
3065 static const char *disable_acs_redir_param;
3068 * pci_disable_acs_redir - disable ACS redirect capabilities
3069 * @dev: the PCI device
3071 * For only devices specified in the disable_acs_redir parameter.
3073 static void pci_disable_acs_redir(struct pci_dev *dev)
3075 int ret = 0;
3076 const char *p;
3077 int pos;
3078 u16 ctrl;
3080 if (!disable_acs_redir_param)
3081 return;
3083 p = disable_acs_redir_param;
3084 while (*p) {
3085 ret = pci_dev_str_match(dev, p, &p);
3086 if (ret < 0) {
3087 pr_info_once("PCI: Can't parse disable_acs_redir parameter: %s\n",
3088 disable_acs_redir_param);
3090 break;
3091 } else if (ret == 1) {
3092 /* Found a match */
3093 break;
3096 if (*p != ';' && *p != ',') {
3097 /* End of param or invalid format */
3098 break;
3100 p++;
3103 if (ret != 1)
3104 return;
3106 if (!pci_dev_specific_disable_acs_redir(dev))
3107 return;
3109 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3110 if (!pos) {
3111 pci_warn(dev, "cannot disable ACS redirect for this hardware as it does not have ACS capabilities\n");
3112 return;
3115 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3117 /* P2P Request & Completion Redirect */
3118 ctrl &= ~(PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC);
3120 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3122 pci_info(dev, "disabled ACS redirect\n");
3126 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
3127 * @dev: the PCI device
3129 static void pci_std_enable_acs(struct pci_dev *dev)
3131 int pos;
3132 u16 cap;
3133 u16 ctrl;
3135 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
3136 if (!pos)
3137 return;
3139 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
3140 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
3142 /* Source Validation */
3143 ctrl |= (cap & PCI_ACS_SV);
3145 /* P2P Request Redirect */
3146 ctrl |= (cap & PCI_ACS_RR);
3148 /* P2P Completion Redirect */
3149 ctrl |= (cap & PCI_ACS_CR);
3151 /* Upstream Forwarding */
3152 ctrl |= (cap & PCI_ACS_UF);
3154 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
3158 * pci_enable_acs - enable ACS if hardware support it
3159 * @dev: the PCI device
3161 void pci_enable_acs(struct pci_dev *dev)
3163 if (!pci_acs_enable)
3164 goto disable_acs_redir;
3166 if (!pci_dev_specific_enable_acs(dev))
3167 goto disable_acs_redir;
3169 pci_std_enable_acs(dev);
3171 disable_acs_redir:
3173 * Note: pci_disable_acs_redir() must be called even if ACS was not
3174 * enabled by the kernel because it may have been enabled by
3175 * platform firmware. So if we are told to disable it, we should
3176 * always disable it after setting the kernel's default
3177 * preferences.
3179 pci_disable_acs_redir(dev);
3182 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
3184 int pos;
3185 u16 cap, ctrl;
3187 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
3188 if (!pos)
3189 return false;
3192 * Except for egress control, capabilities are either required
3193 * or only required if controllable. Features missing from the
3194 * capability field can therefore be assumed as hard-wired enabled.
3196 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
3197 acs_flags &= (cap | PCI_ACS_EC);
3199 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
3200 return (ctrl & acs_flags) == acs_flags;
3204 * pci_acs_enabled - test ACS against required flags for a given device
3205 * @pdev: device to test
3206 * @acs_flags: required PCI ACS flags
3208 * Return true if the device supports the provided flags. Automatically
3209 * filters out flags that are not implemented on multifunction devices.
3211 * Note that this interface checks the effective ACS capabilities of the
3212 * device rather than the actual capabilities. For instance, most single
3213 * function endpoints are not required to support ACS because they have no
3214 * opportunity for peer-to-peer access. We therefore return 'true'
3215 * regardless of whether the device exposes an ACS capability. This makes
3216 * it much easier for callers of this function to ignore the actual type
3217 * or topology of the device when testing ACS support.
3219 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
3221 int ret;
3223 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
3224 if (ret >= 0)
3225 return ret > 0;
3228 * Conventional PCI and PCI-X devices never support ACS, either
3229 * effectively or actually. The shared bus topology implies that
3230 * any device on the bus can receive or snoop DMA.
3232 if (!pci_is_pcie(pdev))
3233 return false;
3235 switch (pci_pcie_type(pdev)) {
3237 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
3238 * but since their primary interface is PCI/X, we conservatively
3239 * handle them as we would a non-PCIe device.
3241 case PCI_EXP_TYPE_PCIE_BRIDGE:
3243 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
3244 * applicable... must never implement an ACS Extended Capability...".
3245 * This seems arbitrary, but we take a conservative interpretation
3246 * of this statement.
3248 case PCI_EXP_TYPE_PCI_BRIDGE:
3249 case PCI_EXP_TYPE_RC_EC:
3250 return false;
3252 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
3253 * implement ACS in order to indicate their peer-to-peer capabilities,
3254 * regardless of whether they are single- or multi-function devices.
3256 case PCI_EXP_TYPE_DOWNSTREAM:
3257 case PCI_EXP_TYPE_ROOT_PORT:
3258 return pci_acs_flags_enabled(pdev, acs_flags);
3260 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
3261 * implemented by the remaining PCIe types to indicate peer-to-peer
3262 * capabilities, but only when they are part of a multifunction
3263 * device. The footnote for section 6.12 indicates the specific
3264 * PCIe types included here.
3266 case PCI_EXP_TYPE_ENDPOINT:
3267 case PCI_EXP_TYPE_UPSTREAM:
3268 case PCI_EXP_TYPE_LEG_END:
3269 case PCI_EXP_TYPE_RC_END:
3270 if (!pdev->multifunction)
3271 break;
3273 return pci_acs_flags_enabled(pdev, acs_flags);
3277 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
3278 * to single function devices with the exception of downstream ports.
3280 return true;
3284 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
3285 * @start: starting downstream device
3286 * @end: ending upstream device or NULL to search to the root bus
3287 * @acs_flags: required flags
3289 * Walk up a device tree from start to end testing PCI ACS support. If
3290 * any step along the way does not support the required flags, return false.
3292 bool pci_acs_path_enabled(struct pci_dev *start,
3293 struct pci_dev *end, u16 acs_flags)
3295 struct pci_dev *pdev, *parent = start;
3297 do {
3298 pdev = parent;
3300 if (!pci_acs_enabled(pdev, acs_flags))
3301 return false;
3303 if (pci_is_root_bus(pdev->bus))
3304 return (end == NULL);
3306 parent = pdev->bus->self;
3307 } while (pdev != end);
3309 return true;
3313 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
3314 * @pdev: PCI device
3315 * @bar: BAR to find
3317 * Helper to find the position of the ctrl register for a BAR.
3318 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3319 * Returns -ENOENT if no ctrl register for the BAR could be found.
3321 static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3323 unsigned int pos, nbars, i;
3324 u32 ctrl;
3326 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3327 if (!pos)
3328 return -ENOTSUPP;
3330 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3331 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3332 PCI_REBAR_CTRL_NBAR_SHIFT;
3334 for (i = 0; i < nbars; i++, pos += 8) {
3335 int bar_idx;
3337 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3338 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3339 if (bar_idx == bar)
3340 return pos;
3343 return -ENOENT;
3347 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3348 * @pdev: PCI device
3349 * @bar: BAR to query
3351 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3352 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3354 u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3356 int pos;
3357 u32 cap;
3359 pos = pci_rebar_find_pos(pdev, bar);
3360 if (pos < 0)
3361 return 0;
3363 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3364 return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3368 * pci_rebar_get_current_size - get the current size of a BAR
3369 * @pdev: PCI device
3370 * @bar: BAR to set size to
3372 * Read the size of a BAR from the resizable BAR config.
3373 * Returns size if found or negative error code.
3375 int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3377 int pos;
3378 u32 ctrl;
3380 pos = pci_rebar_find_pos(pdev, bar);
3381 if (pos < 0)
3382 return pos;
3384 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3385 return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> PCI_REBAR_CTRL_BAR_SHIFT;
3389 * pci_rebar_set_size - set a new size for a BAR
3390 * @pdev: PCI device
3391 * @bar: BAR to set size to
3392 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3394 * Set the new size of a BAR as defined in the spec.
3395 * Returns zero if resizing was successful, error code otherwise.
3397 int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3399 int pos;
3400 u32 ctrl;
3402 pos = pci_rebar_find_pos(pdev, bar);
3403 if (pos < 0)
3404 return pos;
3406 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3407 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3408 ctrl |= size << PCI_REBAR_CTRL_BAR_SHIFT;
3409 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3410 return 0;
3414 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3415 * @dev: the PCI device
3416 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3417 * PCI_EXP_DEVCAP2_ATOMIC_COMP32
3418 * PCI_EXP_DEVCAP2_ATOMIC_COMP64
3419 * PCI_EXP_DEVCAP2_ATOMIC_COMP128
3421 * Return 0 if all upstream bridges support AtomicOp routing, egress
3422 * blocking is disabled on all upstream ports, and the root port supports
3423 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3424 * AtomicOp completion), or negative otherwise.
3426 int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3428 struct pci_bus *bus = dev->bus;
3429 struct pci_dev *bridge;
3430 u32 cap, ctl2;
3432 if (!pci_is_pcie(dev))
3433 return -EINVAL;
3436 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3437 * AtomicOp requesters. For now, we only support endpoints as
3438 * requesters and root ports as completers. No endpoints as
3439 * completers, and no peer-to-peer.
3442 switch (pci_pcie_type(dev)) {
3443 case PCI_EXP_TYPE_ENDPOINT:
3444 case PCI_EXP_TYPE_LEG_END:
3445 case PCI_EXP_TYPE_RC_END:
3446 break;
3447 default:
3448 return -EINVAL;
3451 while (bus->parent) {
3452 bridge = bus->self;
3454 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3456 switch (pci_pcie_type(bridge)) {
3457 /* Ensure switch ports support AtomicOp routing */
3458 case PCI_EXP_TYPE_UPSTREAM:
3459 case PCI_EXP_TYPE_DOWNSTREAM:
3460 if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3461 return -EINVAL;
3462 break;
3464 /* Ensure root port supports all the sizes we care about */
3465 case PCI_EXP_TYPE_ROOT_PORT:
3466 if ((cap & cap_mask) != cap_mask)
3467 return -EINVAL;
3468 break;
3471 /* Ensure upstream ports don't block AtomicOps on egress */
3472 if (!bridge->has_secondary_link) {
3473 pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3474 &ctl2);
3475 if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3476 return -EINVAL;
3479 bus = bus->parent;
3482 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3483 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3484 return 0;
3486 EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3489 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3490 * @dev: the PCI device
3491 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3493 * Perform INTx swizzling for a device behind one level of bridge. This is
3494 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3495 * behind bridges on add-in cards. For devices with ARI enabled, the slot
3496 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3497 * the PCI Express Base Specification, Revision 2.1)
3499 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3501 int slot;
3503 if (pci_ari_enabled(dev->bus))
3504 slot = 0;
3505 else
3506 slot = PCI_SLOT(dev->devfn);
3508 return (((pin - 1) + slot) % 4) + 1;
3511 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3513 u8 pin;
3515 pin = dev->pin;
3516 if (!pin)
3517 return -1;
3519 while (!pci_is_root_bus(dev->bus)) {
3520 pin = pci_swizzle_interrupt_pin(dev, pin);
3521 dev = dev->bus->self;
3523 *bridge = dev;
3524 return pin;
3528 * pci_common_swizzle - swizzle INTx all the way to root bridge
3529 * @dev: the PCI device
3530 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3532 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3533 * bridges all the way up to a PCI root bus.
3535 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3537 u8 pin = *pinp;
3539 while (!pci_is_root_bus(dev->bus)) {
3540 pin = pci_swizzle_interrupt_pin(dev, pin);
3541 dev = dev->bus->self;
3543 *pinp = pin;
3544 return PCI_SLOT(dev->devfn);
3546 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3549 * pci_release_region - Release a PCI bar
3550 * @pdev: PCI device whose resources were previously reserved by pci_request_region
3551 * @bar: BAR to release
3553 * Releases the PCI I/O and memory resources previously reserved by a
3554 * successful call to pci_request_region. Call this function only
3555 * after all use of the PCI regions has ceased.
3557 void pci_release_region(struct pci_dev *pdev, int bar)
3559 struct pci_devres *dr;
3561 if (pci_resource_len(pdev, bar) == 0)
3562 return;
3563 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3564 release_region(pci_resource_start(pdev, bar),
3565 pci_resource_len(pdev, bar));
3566 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3567 release_mem_region(pci_resource_start(pdev, bar),
3568 pci_resource_len(pdev, bar));
3570 dr = find_pci_dr(pdev);
3571 if (dr)
3572 dr->region_mask &= ~(1 << bar);
3574 EXPORT_SYMBOL(pci_release_region);
3577 * __pci_request_region - Reserved PCI I/O and memory resource
3578 * @pdev: PCI device whose resources are to be reserved
3579 * @bar: BAR to be reserved
3580 * @res_name: Name to be associated with resource.
3581 * @exclusive: whether the region access is exclusive or not
3583 * Mark the PCI region associated with PCI device @pdev BR @bar as
3584 * being reserved by owner @res_name. Do not access any
3585 * address inside the PCI regions unless this call returns
3586 * successfully.
3588 * If @exclusive is set, then the region is marked so that userspace
3589 * is explicitly not allowed to map the resource via /dev/mem or
3590 * sysfs MMIO access.
3592 * Returns 0 on success, or %EBUSY on error. A warning
3593 * message is also printed on failure.
3595 static int __pci_request_region(struct pci_dev *pdev, int bar,
3596 const char *res_name, int exclusive)
3598 struct pci_devres *dr;
3600 if (pci_resource_len(pdev, bar) == 0)
3601 return 0;
3603 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3604 if (!request_region(pci_resource_start(pdev, bar),
3605 pci_resource_len(pdev, bar), res_name))
3606 goto err_out;
3607 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3608 if (!__request_mem_region(pci_resource_start(pdev, bar),
3609 pci_resource_len(pdev, bar), res_name,
3610 exclusive))
3611 goto err_out;
3614 dr = find_pci_dr(pdev);
3615 if (dr)
3616 dr->region_mask |= 1 << bar;
3618 return 0;
3620 err_out:
3621 pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3622 &pdev->resource[bar]);
3623 return -EBUSY;
3627 * pci_request_region - Reserve PCI I/O and memory resource
3628 * @pdev: PCI device whose resources are to be reserved
3629 * @bar: BAR to be reserved
3630 * @res_name: Name to be associated with resource
3632 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3633 * being reserved by owner @res_name. Do not access any
3634 * address inside the PCI regions unless this call returns
3635 * successfully.
3637 * Returns 0 on success, or %EBUSY on error. A warning
3638 * message is also printed on failure.
3640 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3642 return __pci_request_region(pdev, bar, res_name, 0);
3644 EXPORT_SYMBOL(pci_request_region);
3647 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
3648 * @pdev: PCI device whose resources are to be reserved
3649 * @bar: BAR to be reserved
3650 * @res_name: Name to be associated with resource.
3652 * Mark the PCI region associated with PCI device @pdev BR @bar as
3653 * being reserved by owner @res_name. Do not access any
3654 * address inside the PCI regions unless this call returns
3655 * successfully.
3657 * Returns 0 on success, or %EBUSY on error. A warning
3658 * message is also printed on failure.
3660 * The key difference that _exclusive makes it that userspace is
3661 * explicitly not allowed to map the resource via /dev/mem or
3662 * sysfs.
3664 int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3665 const char *res_name)
3667 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3669 EXPORT_SYMBOL(pci_request_region_exclusive);
3672 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3673 * @pdev: PCI device whose resources were previously reserved
3674 * @bars: Bitmask of BARs to be released
3676 * Release selected PCI I/O and memory resources previously reserved.
3677 * Call this function only after all use of the PCI regions has ceased.
3679 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3681 int i;
3683 for (i = 0; i < 6; i++)
3684 if (bars & (1 << i))
3685 pci_release_region(pdev, i);
3687 EXPORT_SYMBOL(pci_release_selected_regions);
3689 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3690 const char *res_name, int excl)
3692 int i;
3694 for (i = 0; i < 6; i++)
3695 if (bars & (1 << i))
3696 if (__pci_request_region(pdev, i, res_name, excl))
3697 goto err_out;
3698 return 0;
3700 err_out:
3701 while (--i >= 0)
3702 if (bars & (1 << i))
3703 pci_release_region(pdev, i);
3705 return -EBUSY;
3710 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3711 * @pdev: PCI device whose resources are to be reserved
3712 * @bars: Bitmask of BARs to be requested
3713 * @res_name: Name to be associated with resource
3715 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3716 const char *res_name)
3718 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3720 EXPORT_SYMBOL(pci_request_selected_regions);
3722 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3723 const char *res_name)
3725 return __pci_request_selected_regions(pdev, bars, res_name,
3726 IORESOURCE_EXCLUSIVE);
3728 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3731 * pci_release_regions - Release reserved PCI I/O and memory resources
3732 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
3734 * Releases all PCI I/O and memory resources previously reserved by a
3735 * successful call to pci_request_regions. Call this function only
3736 * after all use of the PCI regions has ceased.
3739 void pci_release_regions(struct pci_dev *pdev)
3741 pci_release_selected_regions(pdev, (1 << 6) - 1);
3743 EXPORT_SYMBOL(pci_release_regions);
3746 * pci_request_regions - Reserved PCI I/O and memory resources
3747 * @pdev: PCI device whose resources are to be reserved
3748 * @res_name: Name to be associated with resource.
3750 * Mark all PCI regions associated with PCI device @pdev as
3751 * being reserved by owner @res_name. Do not access any
3752 * address inside the PCI regions unless this call returns
3753 * successfully.
3755 * Returns 0 on success, or %EBUSY on error. A warning
3756 * message is also printed on failure.
3758 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3760 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3762 EXPORT_SYMBOL(pci_request_regions);
3765 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3766 * @pdev: PCI device whose resources are to be reserved
3767 * @res_name: Name to be associated with resource.
3769 * Mark all PCI regions associated with PCI device @pdev as
3770 * being reserved by owner @res_name. Do not access any
3771 * address inside the PCI regions unless this call returns
3772 * successfully.
3774 * pci_request_regions_exclusive() will mark the region so that
3775 * /dev/mem and the sysfs MMIO access will not be allowed.
3777 * Returns 0 on success, or %EBUSY on error. A warning
3778 * message is also printed on failure.
3780 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3782 return pci_request_selected_regions_exclusive(pdev,
3783 ((1 << 6) - 1), res_name);
3785 EXPORT_SYMBOL(pci_request_regions_exclusive);
3788 * Record the PCI IO range (expressed as CPU physical address + size).
3789 * Return a negative value if an error has occured, zero otherwise
3791 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3792 resource_size_t size)
3794 int ret = 0;
3795 #ifdef PCI_IOBASE
3796 struct logic_pio_hwaddr *range;
3798 if (!size || addr + size < addr)
3799 return -EINVAL;
3801 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3802 if (!range)
3803 return -ENOMEM;
3805 range->fwnode = fwnode;
3806 range->size = size;
3807 range->hw_start = addr;
3808 range->flags = LOGIC_PIO_CPU_MMIO;
3810 ret = logic_pio_register_range(range);
3811 if (ret)
3812 kfree(range);
3813 #endif
3815 return ret;
3818 phys_addr_t pci_pio_to_address(unsigned long pio)
3820 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3822 #ifdef PCI_IOBASE
3823 if (pio >= MMIO_UPPER_LIMIT)
3824 return address;
3826 address = logic_pio_to_hwaddr(pio);
3827 #endif
3829 return address;
3832 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3834 #ifdef PCI_IOBASE
3835 return logic_pio_trans_cpuaddr(address);
3836 #else
3837 if (address > IO_SPACE_LIMIT)
3838 return (unsigned long)-1;
3840 return (unsigned long) address;
3841 #endif
3845 * pci_remap_iospace - Remap the memory mapped I/O space
3846 * @res: Resource describing the I/O space
3847 * @phys_addr: physical address of range to be mapped
3849 * Remap the memory mapped I/O space described by the @res
3850 * and the CPU physical address @phys_addr into virtual address space.
3851 * Only architectures that have memory mapped IO functions defined
3852 * (and the PCI_IOBASE value defined) should call this function.
3854 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3856 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3857 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3859 if (!(res->flags & IORESOURCE_IO))
3860 return -EINVAL;
3862 if (res->end > IO_SPACE_LIMIT)
3863 return -EINVAL;
3865 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3866 pgprot_device(PAGE_KERNEL));
3867 #else
3868 /* this architecture does not have memory mapped I/O space,
3869 so this function should never be called */
3870 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3871 return -ENODEV;
3872 #endif
3874 EXPORT_SYMBOL(pci_remap_iospace);
3877 * pci_unmap_iospace - Unmap the memory mapped I/O space
3878 * @res: resource to be unmapped
3880 * Unmap the CPU virtual address @res from virtual address space.
3881 * Only architectures that have memory mapped IO functions defined
3882 * (and the PCI_IOBASE value defined) should call this function.
3884 void pci_unmap_iospace(struct resource *res)
3886 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3887 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3889 unmap_kernel_range(vaddr, resource_size(res));
3890 #endif
3892 EXPORT_SYMBOL(pci_unmap_iospace);
3894 static void devm_pci_unmap_iospace(struct device *dev, void *ptr)
3896 struct resource **res = ptr;
3898 pci_unmap_iospace(*res);
3902 * devm_pci_remap_iospace - Managed pci_remap_iospace()
3903 * @dev: Generic device to remap IO address for
3904 * @res: Resource describing the I/O space
3905 * @phys_addr: physical address of range to be mapped
3907 * Managed pci_remap_iospace(). Map is automatically unmapped on driver
3908 * detach.
3910 int devm_pci_remap_iospace(struct device *dev, const struct resource *res,
3911 phys_addr_t phys_addr)
3913 const struct resource **ptr;
3914 int error;
3916 ptr = devres_alloc(devm_pci_unmap_iospace, sizeof(*ptr), GFP_KERNEL);
3917 if (!ptr)
3918 return -ENOMEM;
3920 error = pci_remap_iospace(res, phys_addr);
3921 if (error) {
3922 devres_free(ptr);
3923 } else {
3924 *ptr = res;
3925 devres_add(dev, ptr);
3928 return error;
3930 EXPORT_SYMBOL(devm_pci_remap_iospace);
3933 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3934 * @dev: Generic device to remap IO address for
3935 * @offset: Resource address to map
3936 * @size: Size of map
3938 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
3939 * detach.
3941 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3942 resource_size_t offset,
3943 resource_size_t size)
3945 void __iomem **ptr, *addr;
3947 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
3948 if (!ptr)
3949 return NULL;
3951 addr = pci_remap_cfgspace(offset, size);
3952 if (addr) {
3953 *ptr = addr;
3954 devres_add(dev, ptr);
3955 } else
3956 devres_free(ptr);
3958 return addr;
3960 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
3963 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
3964 * @dev: generic device to handle the resource for
3965 * @res: configuration space resource to be handled
3967 * Checks that a resource is a valid memory region, requests the memory
3968 * region and ioremaps with pci_remap_cfgspace() API that ensures the
3969 * proper PCI configuration space memory attributes are guaranteed.
3971 * All operations are managed and will be undone on driver detach.
3973 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
3974 * on failure. Usage example::
3976 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
3977 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
3978 * if (IS_ERR(base))
3979 * return PTR_ERR(base);
3981 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
3982 struct resource *res)
3984 resource_size_t size;
3985 const char *name;
3986 void __iomem *dest_ptr;
3988 BUG_ON(!dev);
3990 if (!res || resource_type(res) != IORESOURCE_MEM) {
3991 dev_err(dev, "invalid resource\n");
3992 return IOMEM_ERR_PTR(-EINVAL);
3995 size = resource_size(res);
3996 name = res->name ?: dev_name(dev);
3998 if (!devm_request_mem_region(dev, res->start, size, name)) {
3999 dev_err(dev, "can't request region for resource %pR\n", res);
4000 return IOMEM_ERR_PTR(-EBUSY);
4003 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
4004 if (!dest_ptr) {
4005 dev_err(dev, "ioremap failed for resource %pR\n", res);
4006 devm_release_mem_region(dev, res->start, size);
4007 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
4010 return dest_ptr;
4012 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
4014 static void __pci_set_master(struct pci_dev *dev, bool enable)
4016 u16 old_cmd, cmd;
4018 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
4019 if (enable)
4020 cmd = old_cmd | PCI_COMMAND_MASTER;
4021 else
4022 cmd = old_cmd & ~PCI_COMMAND_MASTER;
4023 if (cmd != old_cmd) {
4024 pci_dbg(dev, "%s bus mastering\n",
4025 enable ? "enabling" : "disabling");
4026 pci_write_config_word(dev, PCI_COMMAND, cmd);
4028 dev->is_busmaster = enable;
4032 * pcibios_setup - process "pci=" kernel boot arguments
4033 * @str: string used to pass in "pci=" kernel boot arguments
4035 * Process kernel boot arguments. This is the default implementation.
4036 * Architecture specific implementations can override this as necessary.
4038 char * __weak __init pcibios_setup(char *str)
4040 return str;
4044 * pcibios_set_master - enable PCI bus-mastering for device dev
4045 * @dev: the PCI device to enable
4047 * Enables PCI bus-mastering for the device. This is the default
4048 * implementation. Architecture specific implementations can override
4049 * this if necessary.
4051 void __weak pcibios_set_master(struct pci_dev *dev)
4053 u8 lat;
4055 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
4056 if (pci_is_pcie(dev))
4057 return;
4059 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
4060 if (lat < 16)
4061 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
4062 else if (lat > pcibios_max_latency)
4063 lat = pcibios_max_latency;
4064 else
4065 return;
4067 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
4071 * pci_set_master - enables bus-mastering for device dev
4072 * @dev: the PCI device to enable
4074 * Enables bus-mastering on the device and calls pcibios_set_master()
4075 * to do the needed arch specific settings.
4077 void pci_set_master(struct pci_dev *dev)
4079 __pci_set_master(dev, true);
4080 pcibios_set_master(dev);
4082 EXPORT_SYMBOL(pci_set_master);
4085 * pci_clear_master - disables bus-mastering for device dev
4086 * @dev: the PCI device to disable
4088 void pci_clear_master(struct pci_dev *dev)
4090 __pci_set_master(dev, false);
4092 EXPORT_SYMBOL(pci_clear_master);
4095 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
4096 * @dev: the PCI device for which MWI is to be enabled
4098 * Helper function for pci_set_mwi.
4099 * Originally copied from drivers/net/acenic.c.
4100 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
4102 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4104 int pci_set_cacheline_size(struct pci_dev *dev)
4106 u8 cacheline_size;
4108 if (!pci_cache_line_size)
4109 return -EINVAL;
4111 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
4112 equal to or multiple of the right value. */
4113 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4114 if (cacheline_size >= pci_cache_line_size &&
4115 (cacheline_size % pci_cache_line_size) == 0)
4116 return 0;
4118 /* Write the correct value. */
4119 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
4120 /* Read it back. */
4121 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
4122 if (cacheline_size == pci_cache_line_size)
4123 return 0;
4125 pci_printk(KERN_DEBUG, dev, "cache line size of %d is not supported\n",
4126 pci_cache_line_size << 2);
4128 return -EINVAL;
4130 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
4133 * pci_set_mwi - enables memory-write-invalidate PCI transaction
4134 * @dev: the PCI device for which MWI is enabled
4136 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4138 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4140 int pci_set_mwi(struct pci_dev *dev)
4142 #ifdef PCI_DISABLE_MWI
4143 return 0;
4144 #else
4145 int rc;
4146 u16 cmd;
4148 rc = pci_set_cacheline_size(dev);
4149 if (rc)
4150 return rc;
4152 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4153 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
4154 pci_dbg(dev, "enabling Mem-Wr-Inval\n");
4155 cmd |= PCI_COMMAND_INVALIDATE;
4156 pci_write_config_word(dev, PCI_COMMAND, cmd);
4158 return 0;
4159 #endif
4161 EXPORT_SYMBOL(pci_set_mwi);
4164 * pcim_set_mwi - a device-managed pci_set_mwi()
4165 * @dev: the PCI device for which MWI is enabled
4167 * Managed pci_set_mwi().
4169 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4171 int pcim_set_mwi(struct pci_dev *dev)
4173 struct pci_devres *dr;
4175 dr = find_pci_dr(dev);
4176 if (!dr)
4177 return -ENOMEM;
4179 dr->mwi = 1;
4180 return pci_set_mwi(dev);
4182 EXPORT_SYMBOL(pcim_set_mwi);
4185 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
4186 * @dev: the PCI device for which MWI is enabled
4188 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
4189 * Callers are not required to check the return value.
4191 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
4193 int pci_try_set_mwi(struct pci_dev *dev)
4195 #ifdef PCI_DISABLE_MWI
4196 return 0;
4197 #else
4198 return pci_set_mwi(dev);
4199 #endif
4201 EXPORT_SYMBOL(pci_try_set_mwi);
4204 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
4205 * @dev: the PCI device to disable
4207 * Disables PCI Memory-Write-Invalidate transaction on the device
4209 void pci_clear_mwi(struct pci_dev *dev)
4211 #ifndef PCI_DISABLE_MWI
4212 u16 cmd;
4214 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4215 if (cmd & PCI_COMMAND_INVALIDATE) {
4216 cmd &= ~PCI_COMMAND_INVALIDATE;
4217 pci_write_config_word(dev, PCI_COMMAND, cmd);
4219 #endif
4221 EXPORT_SYMBOL(pci_clear_mwi);
4224 * pci_intx - enables/disables PCI INTx for device dev
4225 * @pdev: the PCI device to operate on
4226 * @enable: boolean: whether to enable or disable PCI INTx
4228 * Enables/disables PCI INTx for device dev
4230 void pci_intx(struct pci_dev *pdev, int enable)
4232 u16 pci_command, new;
4234 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
4236 if (enable)
4237 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
4238 else
4239 new = pci_command | PCI_COMMAND_INTX_DISABLE;
4241 if (new != pci_command) {
4242 struct pci_devres *dr;
4244 pci_write_config_word(pdev, PCI_COMMAND, new);
4246 dr = find_pci_dr(pdev);
4247 if (dr && !dr->restore_intx) {
4248 dr->restore_intx = 1;
4249 dr->orig_intx = !enable;
4253 EXPORT_SYMBOL_GPL(pci_intx);
4255 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
4257 struct pci_bus *bus = dev->bus;
4258 bool mask_updated = true;
4259 u32 cmd_status_dword;
4260 u16 origcmd, newcmd;
4261 unsigned long flags;
4262 bool irq_pending;
4265 * We do a single dword read to retrieve both command and status.
4266 * Document assumptions that make this possible.
4268 BUILD_BUG_ON(PCI_COMMAND % 4);
4269 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
4271 raw_spin_lock_irqsave(&pci_lock, flags);
4273 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
4275 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
4278 * Check interrupt status register to see whether our device
4279 * triggered the interrupt (when masking) or the next IRQ is
4280 * already pending (when unmasking).
4282 if (mask != irq_pending) {
4283 mask_updated = false;
4284 goto done;
4287 origcmd = cmd_status_dword;
4288 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
4289 if (mask)
4290 newcmd |= PCI_COMMAND_INTX_DISABLE;
4291 if (newcmd != origcmd)
4292 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
4294 done:
4295 raw_spin_unlock_irqrestore(&pci_lock, flags);
4297 return mask_updated;
4301 * pci_check_and_mask_intx - mask INTx on pending interrupt
4302 * @dev: the PCI device to operate on
4304 * Check if the device dev has its INTx line asserted, mask it and
4305 * return true in that case. False is returned if no interrupt was
4306 * pending.
4308 bool pci_check_and_mask_intx(struct pci_dev *dev)
4310 return pci_check_and_set_intx_mask(dev, true);
4312 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
4315 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
4316 * @dev: the PCI device to operate on
4318 * Check if the device dev has its INTx line asserted, unmask it if not
4319 * and return true. False is returned and the mask remains active if
4320 * there was still an interrupt pending.
4322 bool pci_check_and_unmask_intx(struct pci_dev *dev)
4324 return pci_check_and_set_intx_mask(dev, false);
4326 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
4329 * pci_wait_for_pending_transaction - waits for pending transaction
4330 * @dev: the PCI device to operate on
4332 * Return 0 if transaction is pending 1 otherwise.
4334 int pci_wait_for_pending_transaction(struct pci_dev *dev)
4336 if (!pci_is_pcie(dev))
4337 return 1;
4339 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
4340 PCI_EXP_DEVSTA_TRPND);
4342 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
4344 static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
4346 int delay = 1;
4347 u32 id;
4350 * After reset, the device should not silently discard config
4351 * requests, but it may still indicate that it needs more time by
4352 * responding to them with CRS completions. The Root Port will
4353 * generally synthesize ~0 data to complete the read (except when
4354 * CRS SV is enabled and the read was for the Vendor ID; in that
4355 * case it synthesizes 0x0001 data).
4357 * Wait for the device to return a non-CRS completion. Read the
4358 * Command register instead of Vendor ID so we don't have to
4359 * contend with the CRS SV value.
4361 pci_read_config_dword(dev, PCI_COMMAND, &id);
4362 while (id == ~0) {
4363 if (delay > timeout) {
4364 pci_warn(dev, "not ready %dms after %s; giving up\n",
4365 delay - 1, reset_type);
4366 return -ENOTTY;
4369 if (delay > 1000)
4370 pci_info(dev, "not ready %dms after %s; waiting\n",
4371 delay - 1, reset_type);
4373 msleep(delay);
4374 delay *= 2;
4375 pci_read_config_dword(dev, PCI_COMMAND, &id);
4378 if (delay > 1000)
4379 pci_info(dev, "ready %dms after %s\n", delay - 1,
4380 reset_type);
4382 return 0;
4386 * pcie_has_flr - check if a device supports function level resets
4387 * @dev: device to check
4389 * Returns true if the device advertises support for PCIe function level
4390 * resets.
4392 bool pcie_has_flr(struct pci_dev *dev)
4394 u32 cap;
4396 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4397 return false;
4399 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4400 return cap & PCI_EXP_DEVCAP_FLR;
4402 EXPORT_SYMBOL_GPL(pcie_has_flr);
4405 * pcie_flr - initiate a PCIe function level reset
4406 * @dev: device to reset
4408 * Initiate a function level reset on @dev. The caller should ensure the
4409 * device supports FLR before calling this function, e.g. by using the
4410 * pcie_has_flr() helper.
4412 int pcie_flr(struct pci_dev *dev)
4414 if (!pci_wait_for_pending_transaction(dev))
4415 pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4417 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4420 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4421 * 100ms, but may silently discard requests while the FLR is in
4422 * progress. Wait 100ms before trying to access the device.
4424 msleep(100);
4426 return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4428 EXPORT_SYMBOL_GPL(pcie_flr);
4430 static int pci_af_flr(struct pci_dev *dev, int probe)
4432 int pos;
4433 u8 cap;
4435 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4436 if (!pos)
4437 return -ENOTTY;
4439 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4440 return -ENOTTY;
4442 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4443 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4444 return -ENOTTY;
4446 if (probe)
4447 return 0;
4450 * Wait for Transaction Pending bit to clear. A word-aligned test
4451 * is used, so we use the conrol offset rather than status and shift
4452 * the test bit to match.
4454 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4455 PCI_AF_STATUS_TP << 8))
4456 pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4458 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4461 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4462 * updated 27 July 2006; a device must complete an FLR within
4463 * 100ms, but may silently discard requests while the FLR is in
4464 * progress. Wait 100ms before trying to access the device.
4466 msleep(100);
4468 return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4472 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4473 * @dev: Device to reset.
4474 * @probe: If set, only check if the device can be reset this way.
4476 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4477 * unset, it will be reinitialized internally when going from PCI_D3hot to
4478 * PCI_D0. If that's the case and the device is not in a low-power state
4479 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4481 * NOTE: This causes the caller to sleep for twice the device power transition
4482 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4483 * by default (i.e. unless the @dev's d3_delay field has a different value).
4484 * Moreover, only devices in D0 can be reset by this function.
4486 static int pci_pm_reset(struct pci_dev *dev, int probe)
4488 u16 csr;
4490 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4491 return -ENOTTY;
4493 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4494 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4495 return -ENOTTY;
4497 if (probe)
4498 return 0;
4500 if (dev->current_state != PCI_D0)
4501 return -EINVAL;
4503 csr &= ~PCI_PM_CTRL_STATE_MASK;
4504 csr |= PCI_D3hot;
4505 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4506 pci_dev_d3_sleep(dev);
4508 csr &= ~PCI_PM_CTRL_STATE_MASK;
4509 csr |= PCI_D0;
4510 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4511 pci_dev_d3_sleep(dev);
4513 return pci_dev_wait(dev, "PM D3->D0", PCIE_RESET_READY_POLL_MS);
4516 * pcie_wait_for_link - Wait until link is active or inactive
4517 * @pdev: Bridge device
4518 * @active: waiting for active or inactive?
4520 * Use this to wait till link becomes active or inactive.
4522 bool pcie_wait_for_link(struct pci_dev *pdev, bool active)
4524 int timeout = 1000;
4525 bool ret;
4526 u16 lnk_status;
4528 for (;;) {
4529 pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
4530 ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
4531 if (ret == active)
4532 return true;
4533 if (timeout <= 0)
4534 break;
4535 msleep(10);
4536 timeout -= 10;
4539 pci_info(pdev, "Data Link Layer Link Active not %s in 1000 msec\n",
4540 active ? "set" : "cleared");
4542 return false;
4545 void pci_reset_secondary_bus(struct pci_dev *dev)
4547 u16 ctrl;
4549 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4550 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4551 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4554 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4555 * this to 2ms to ensure that we meet the minimum requirement.
4557 msleep(2);
4559 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4560 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4563 * Trhfa for conventional PCI is 2^25 clock cycles.
4564 * Assuming a minimum 33MHz clock this results in a 1s
4565 * delay before we can consider subordinate devices to
4566 * be re-initialized. PCIe has some ways to shorten this,
4567 * but we don't make use of them yet.
4569 ssleep(1);
4572 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4574 pci_reset_secondary_bus(dev);
4578 * pci_bridge_secondary_bus_reset - Reset the secondary bus on a PCI bridge.
4579 * @dev: Bridge device
4581 * Use the bridge control register to assert reset on the secondary bus.
4582 * Devices on the secondary bus are left in power-on state.
4584 int pci_bridge_secondary_bus_reset(struct pci_dev *dev)
4586 pcibios_reset_secondary_bus(dev);
4588 return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4590 EXPORT_SYMBOL_GPL(pci_bridge_secondary_bus_reset);
4592 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4594 struct pci_dev *pdev;
4596 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4597 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4598 return -ENOTTY;
4600 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4601 if (pdev != dev)
4602 return -ENOTTY;
4604 if (probe)
4605 return 0;
4607 return pci_bridge_secondary_bus_reset(dev->bus->self);
4610 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4612 int rc = -ENOTTY;
4614 if (!hotplug || !try_module_get(hotplug->ops->owner))
4615 return rc;
4617 if (hotplug->ops->reset_slot)
4618 rc = hotplug->ops->reset_slot(hotplug, probe);
4620 module_put(hotplug->ops->owner);
4622 return rc;
4625 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4627 struct pci_dev *pdev;
4629 if (dev->subordinate || !dev->slot ||
4630 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4631 return -ENOTTY;
4633 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4634 if (pdev != dev && pdev->slot == dev->slot)
4635 return -ENOTTY;
4637 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4640 static void pci_dev_lock(struct pci_dev *dev)
4642 pci_cfg_access_lock(dev);
4643 /* block PM suspend, driver probe, etc. */
4644 device_lock(&dev->dev);
4647 /* Return 1 on successful lock, 0 on contention */
4648 static int pci_dev_trylock(struct pci_dev *dev)
4650 if (pci_cfg_access_trylock(dev)) {
4651 if (device_trylock(&dev->dev))
4652 return 1;
4653 pci_cfg_access_unlock(dev);
4656 return 0;
4659 static void pci_dev_unlock(struct pci_dev *dev)
4661 device_unlock(&dev->dev);
4662 pci_cfg_access_unlock(dev);
4665 static void pci_dev_save_and_disable(struct pci_dev *dev)
4667 const struct pci_error_handlers *err_handler =
4668 dev->driver ? dev->driver->err_handler : NULL;
4671 * dev->driver->err_handler->reset_prepare() is protected against
4672 * races with ->remove() by the device lock, which must be held by
4673 * the caller.
4675 if (err_handler && err_handler->reset_prepare)
4676 err_handler->reset_prepare(dev);
4679 * Wake-up device prior to save. PM registers default to D0 after
4680 * reset and a simple register restore doesn't reliably return
4681 * to a non-D0 state anyway.
4683 pci_set_power_state(dev, PCI_D0);
4685 pci_save_state(dev);
4687 * Disable the device by clearing the Command register, except for
4688 * INTx-disable which is set. This not only disables MMIO and I/O port
4689 * BARs, but also prevents the device from being Bus Master, preventing
4690 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4691 * compliant devices, INTx-disable prevents legacy interrupts.
4693 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4696 static void pci_dev_restore(struct pci_dev *dev)
4698 const struct pci_error_handlers *err_handler =
4699 dev->driver ? dev->driver->err_handler : NULL;
4701 pci_restore_state(dev);
4704 * dev->driver->err_handler->reset_done() is protected against
4705 * races with ->remove() by the device lock, which must be held by
4706 * the caller.
4708 if (err_handler && err_handler->reset_done)
4709 err_handler->reset_done(dev);
4713 * __pci_reset_function_locked - reset a PCI device function while holding
4714 * the @dev mutex lock.
4715 * @dev: PCI device to reset
4717 * Some devices allow an individual function to be reset without affecting
4718 * other functions in the same device. The PCI device must be responsive
4719 * to PCI config space in order to use this function.
4721 * The device function is presumed to be unused and the caller is holding
4722 * the device mutex lock when this function is called.
4723 * Resetting the device will make the contents of PCI configuration space
4724 * random, so any caller of this must be prepared to reinitialise the
4725 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4726 * etc.
4728 * Returns 0 if the device function was successfully reset or negative if the
4729 * device doesn't support resetting a single function.
4731 int __pci_reset_function_locked(struct pci_dev *dev)
4733 int rc;
4735 might_sleep();
4738 * A reset method returns -ENOTTY if it doesn't support this device
4739 * and we should try the next method.
4741 * If it returns 0 (success), we're finished. If it returns any
4742 * other error, we're also finished: this indicates that further
4743 * reset mechanisms might be broken on the device.
4745 rc = pci_dev_specific_reset(dev, 0);
4746 if (rc != -ENOTTY)
4747 return rc;
4748 if (pcie_has_flr(dev)) {
4749 rc = pcie_flr(dev);
4750 if (rc != -ENOTTY)
4751 return rc;
4753 rc = pci_af_flr(dev, 0);
4754 if (rc != -ENOTTY)
4755 return rc;
4756 rc = pci_pm_reset(dev, 0);
4757 if (rc != -ENOTTY)
4758 return rc;
4759 rc = pci_dev_reset_slot_function(dev, 0);
4760 if (rc != -ENOTTY)
4761 return rc;
4762 return pci_parent_bus_reset(dev, 0);
4764 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4767 * pci_probe_reset_function - check whether the device can be safely reset
4768 * @dev: PCI device to reset
4770 * Some devices allow an individual function to be reset without affecting
4771 * other functions in the same device. The PCI device must be responsive
4772 * to PCI config space in order to use this function.
4774 * Returns 0 if the device function can be reset or negative if the
4775 * device doesn't support resetting a single function.
4777 int pci_probe_reset_function(struct pci_dev *dev)
4779 int rc;
4781 might_sleep();
4783 rc = pci_dev_specific_reset(dev, 1);
4784 if (rc != -ENOTTY)
4785 return rc;
4786 if (pcie_has_flr(dev))
4787 return 0;
4788 rc = pci_af_flr(dev, 1);
4789 if (rc != -ENOTTY)
4790 return rc;
4791 rc = pci_pm_reset(dev, 1);
4792 if (rc != -ENOTTY)
4793 return rc;
4794 rc = pci_dev_reset_slot_function(dev, 1);
4795 if (rc != -ENOTTY)
4796 return rc;
4798 return pci_parent_bus_reset(dev, 1);
4802 * pci_reset_function - quiesce and reset a PCI device function
4803 * @dev: PCI device to reset
4805 * Some devices allow an individual function to be reset without affecting
4806 * other functions in the same device. The PCI device must be responsive
4807 * to PCI config space in order to use this function.
4809 * This function does not just reset the PCI portion of a device, but
4810 * clears all the state associated with the device. This function differs
4811 * from __pci_reset_function_locked() in that it saves and restores device state
4812 * over the reset and takes the PCI device lock.
4814 * Returns 0 if the device function was successfully reset or negative if the
4815 * device doesn't support resetting a single function.
4817 int pci_reset_function(struct pci_dev *dev)
4819 int rc;
4821 if (!dev->reset_fn)
4822 return -ENOTTY;
4824 pci_dev_lock(dev);
4825 pci_dev_save_and_disable(dev);
4827 rc = __pci_reset_function_locked(dev);
4829 pci_dev_restore(dev);
4830 pci_dev_unlock(dev);
4832 return rc;
4834 EXPORT_SYMBOL_GPL(pci_reset_function);
4837 * pci_reset_function_locked - quiesce and reset a PCI device function
4838 * @dev: PCI device to reset
4840 * Some devices allow an individual function to be reset without affecting
4841 * other functions in the same device. The PCI device must be responsive
4842 * to PCI config space in order to use this function.
4844 * This function does not just reset the PCI portion of a device, but
4845 * clears all the state associated with the device. This function differs
4846 * from __pci_reset_function_locked() in that it saves and restores device state
4847 * over the reset. It also differs from pci_reset_function() in that it
4848 * requires the PCI device lock to be held.
4850 * Returns 0 if the device function was successfully reset or negative if the
4851 * device doesn't support resetting a single function.
4853 int pci_reset_function_locked(struct pci_dev *dev)
4855 int rc;
4857 if (!dev->reset_fn)
4858 return -ENOTTY;
4860 pci_dev_save_and_disable(dev);
4862 rc = __pci_reset_function_locked(dev);
4864 pci_dev_restore(dev);
4866 return rc;
4868 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4871 * pci_try_reset_function - quiesce and reset a PCI device function
4872 * @dev: PCI device to reset
4874 * Same as above, except return -EAGAIN if unable to lock device.
4876 int pci_try_reset_function(struct pci_dev *dev)
4878 int rc;
4880 if (!dev->reset_fn)
4881 return -ENOTTY;
4883 if (!pci_dev_trylock(dev))
4884 return -EAGAIN;
4886 pci_dev_save_and_disable(dev);
4887 rc = __pci_reset_function_locked(dev);
4888 pci_dev_restore(dev);
4889 pci_dev_unlock(dev);
4891 return rc;
4893 EXPORT_SYMBOL_GPL(pci_try_reset_function);
4895 /* Do any devices on or below this bus prevent a bus reset? */
4896 static bool pci_bus_resetable(struct pci_bus *bus)
4898 struct pci_dev *dev;
4901 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4902 return false;
4904 list_for_each_entry(dev, &bus->devices, bus_list) {
4905 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4906 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4907 return false;
4910 return true;
4913 /* Lock devices from the top of the tree down */
4914 static void pci_bus_lock(struct pci_bus *bus)
4916 struct pci_dev *dev;
4918 list_for_each_entry(dev, &bus->devices, bus_list) {
4919 pci_dev_lock(dev);
4920 if (dev->subordinate)
4921 pci_bus_lock(dev->subordinate);
4925 /* Unlock devices from the bottom of the tree up */
4926 static void pci_bus_unlock(struct pci_bus *bus)
4928 struct pci_dev *dev;
4930 list_for_each_entry(dev, &bus->devices, bus_list) {
4931 if (dev->subordinate)
4932 pci_bus_unlock(dev->subordinate);
4933 pci_dev_unlock(dev);
4937 /* Return 1 on successful lock, 0 on contention */
4938 static int pci_bus_trylock(struct pci_bus *bus)
4940 struct pci_dev *dev;
4942 list_for_each_entry(dev, &bus->devices, bus_list) {
4943 if (!pci_dev_trylock(dev))
4944 goto unlock;
4945 if (dev->subordinate) {
4946 if (!pci_bus_trylock(dev->subordinate)) {
4947 pci_dev_unlock(dev);
4948 goto unlock;
4952 return 1;
4954 unlock:
4955 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4956 if (dev->subordinate)
4957 pci_bus_unlock(dev->subordinate);
4958 pci_dev_unlock(dev);
4960 return 0;
4963 /* Do any devices on or below this slot prevent a bus reset? */
4964 static bool pci_slot_resetable(struct pci_slot *slot)
4966 struct pci_dev *dev;
4968 if (slot->bus->self &&
4969 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4970 return false;
4972 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4973 if (!dev->slot || dev->slot != slot)
4974 continue;
4975 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4976 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4977 return false;
4980 return true;
4983 /* Lock devices from the top of the tree down */
4984 static void pci_slot_lock(struct pci_slot *slot)
4986 struct pci_dev *dev;
4988 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4989 if (!dev->slot || dev->slot != slot)
4990 continue;
4991 pci_dev_lock(dev);
4992 if (dev->subordinate)
4993 pci_bus_lock(dev->subordinate);
4997 /* Unlock devices from the bottom of the tree up */
4998 static void pci_slot_unlock(struct pci_slot *slot)
5000 struct pci_dev *dev;
5002 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5003 if (!dev->slot || dev->slot != slot)
5004 continue;
5005 if (dev->subordinate)
5006 pci_bus_unlock(dev->subordinate);
5007 pci_dev_unlock(dev);
5011 /* Return 1 on successful lock, 0 on contention */
5012 static int pci_slot_trylock(struct pci_slot *slot)
5014 struct pci_dev *dev;
5016 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5017 if (!dev->slot || dev->slot != slot)
5018 continue;
5019 if (!pci_dev_trylock(dev))
5020 goto unlock;
5021 if (dev->subordinate) {
5022 if (!pci_bus_trylock(dev->subordinate)) {
5023 pci_dev_unlock(dev);
5024 goto unlock;
5028 return 1;
5030 unlock:
5031 list_for_each_entry_continue_reverse(dev,
5032 &slot->bus->devices, bus_list) {
5033 if (!dev->slot || dev->slot != slot)
5034 continue;
5035 if (dev->subordinate)
5036 pci_bus_unlock(dev->subordinate);
5037 pci_dev_unlock(dev);
5039 return 0;
5043 * Save and disable devices from the top of the tree down while holding
5044 * the @dev mutex lock for the entire tree.
5046 static void pci_bus_save_and_disable_locked(struct pci_bus *bus)
5048 struct pci_dev *dev;
5050 list_for_each_entry(dev, &bus->devices, bus_list) {
5051 pci_dev_save_and_disable(dev);
5052 if (dev->subordinate)
5053 pci_bus_save_and_disable_locked(dev->subordinate);
5058 * Restore devices from top of the tree down while holding @dev mutex lock
5059 * for the entire tree. Parent bridges need to be restored before we can
5060 * get to subordinate devices.
5062 static void pci_bus_restore_locked(struct pci_bus *bus)
5064 struct pci_dev *dev;
5066 list_for_each_entry(dev, &bus->devices, bus_list) {
5067 pci_dev_restore(dev);
5068 if (dev->subordinate)
5069 pci_bus_restore_locked(dev->subordinate);
5074 * Save and disable devices from the top of the tree down while holding
5075 * the @dev mutex lock for the entire tree.
5077 static void pci_slot_save_and_disable_locked(struct pci_slot *slot)
5079 struct pci_dev *dev;
5081 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5082 if (!dev->slot || dev->slot != slot)
5083 continue;
5084 pci_dev_save_and_disable(dev);
5085 if (dev->subordinate)
5086 pci_bus_save_and_disable_locked(dev->subordinate);
5091 * Restore devices from top of the tree down while holding @dev mutex lock
5092 * for the entire tree. Parent bridges need to be restored before we can
5093 * get to subordinate devices.
5095 static void pci_slot_restore_locked(struct pci_slot *slot)
5097 struct pci_dev *dev;
5099 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
5100 if (!dev->slot || dev->slot != slot)
5101 continue;
5102 pci_dev_restore(dev);
5103 if (dev->subordinate)
5104 pci_bus_restore_locked(dev->subordinate);
5108 static int pci_slot_reset(struct pci_slot *slot, int probe)
5110 int rc;
5112 if (!slot || !pci_slot_resetable(slot))
5113 return -ENOTTY;
5115 if (!probe)
5116 pci_slot_lock(slot);
5118 might_sleep();
5120 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
5122 if (!probe)
5123 pci_slot_unlock(slot);
5125 return rc;
5129 * pci_probe_reset_slot - probe whether a PCI slot can be reset
5130 * @slot: PCI slot to probe
5132 * Return 0 if slot can be reset, negative if a slot reset is not supported.
5134 int pci_probe_reset_slot(struct pci_slot *slot)
5136 return pci_slot_reset(slot, 1);
5138 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
5141 * __pci_reset_slot - Try to reset a PCI slot
5142 * @slot: PCI slot to reset
5144 * A PCI bus may host multiple slots, each slot may support a reset mechanism
5145 * independent of other slots. For instance, some slots may support slot power
5146 * control. In the case of a 1:1 bus to slot architecture, this function may
5147 * wrap the bus reset to avoid spurious slot related events such as hotplug.
5148 * Generally a slot reset should be attempted before a bus reset. All of the
5149 * function of the slot and any subordinate buses behind the slot are reset
5150 * through this function. PCI config space of all devices in the slot and
5151 * behind the slot is saved before and restored after reset.
5153 * Same as above except return -EAGAIN if the slot cannot be locked
5155 static int __pci_reset_slot(struct pci_slot *slot)
5157 int rc;
5159 rc = pci_slot_reset(slot, 1);
5160 if (rc)
5161 return rc;
5163 if (pci_slot_trylock(slot)) {
5164 pci_slot_save_and_disable_locked(slot);
5165 might_sleep();
5166 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
5167 pci_slot_restore_locked(slot);
5168 pci_slot_unlock(slot);
5169 } else
5170 rc = -EAGAIN;
5172 return rc;
5175 static int pci_bus_reset(struct pci_bus *bus, int probe)
5177 int ret;
5179 if (!bus->self || !pci_bus_resetable(bus))
5180 return -ENOTTY;
5182 if (probe)
5183 return 0;
5185 pci_bus_lock(bus);
5187 might_sleep();
5189 ret = pci_bridge_secondary_bus_reset(bus->self);
5191 pci_bus_unlock(bus);
5193 return ret;
5197 * pci_bus_error_reset - reset the bridge's subordinate bus
5198 * @bridge: The parent device that connects to the bus to reset
5200 * This function will first try to reset the slots on this bus if the method is
5201 * available. If slot reset fails or is not available, this will fall back to a
5202 * secondary bus reset.
5204 int pci_bus_error_reset(struct pci_dev *bridge)
5206 struct pci_bus *bus = bridge->subordinate;
5207 struct pci_slot *slot;
5209 if (!bus)
5210 return -ENOTTY;
5212 mutex_lock(&pci_slot_mutex);
5213 if (list_empty(&bus->slots))
5214 goto bus_reset;
5216 list_for_each_entry(slot, &bus->slots, list)
5217 if (pci_probe_reset_slot(slot))
5218 goto bus_reset;
5220 list_for_each_entry(slot, &bus->slots, list)
5221 if (pci_slot_reset(slot, 0))
5222 goto bus_reset;
5224 mutex_unlock(&pci_slot_mutex);
5225 return 0;
5226 bus_reset:
5227 mutex_unlock(&pci_slot_mutex);
5228 return pci_bus_reset(bridge->subordinate, 0);
5232 * pci_probe_reset_bus - probe whether a PCI bus can be reset
5233 * @bus: PCI bus to probe
5235 * Return 0 if bus can be reset, negative if a bus reset is not supported.
5237 int pci_probe_reset_bus(struct pci_bus *bus)
5239 return pci_bus_reset(bus, 1);
5241 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
5244 * __pci_reset_bus - Try to reset a PCI bus
5245 * @bus: top level PCI bus to reset
5247 * Same as above except return -EAGAIN if the bus cannot be locked
5249 static int __pci_reset_bus(struct pci_bus *bus)
5251 int rc;
5253 rc = pci_bus_reset(bus, 1);
5254 if (rc)
5255 return rc;
5257 if (pci_bus_trylock(bus)) {
5258 pci_bus_save_and_disable_locked(bus);
5259 might_sleep();
5260 rc = pci_bridge_secondary_bus_reset(bus->self);
5261 pci_bus_restore_locked(bus);
5262 pci_bus_unlock(bus);
5263 } else
5264 rc = -EAGAIN;
5266 return rc;
5270 * pci_reset_bus - Try to reset a PCI bus
5271 * @pdev: top level PCI device to reset via slot/bus
5273 * Same as above except return -EAGAIN if the bus cannot be locked
5275 int pci_reset_bus(struct pci_dev *pdev)
5277 return (!pci_probe_reset_slot(pdev->slot)) ?
5278 __pci_reset_slot(pdev->slot) : __pci_reset_bus(pdev->bus);
5280 EXPORT_SYMBOL_GPL(pci_reset_bus);
5283 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
5284 * @dev: PCI device to query
5286 * Returns mmrbc: maximum designed memory read count in bytes
5287 * or appropriate error value.
5289 int pcix_get_max_mmrbc(struct pci_dev *dev)
5291 int cap;
5292 u32 stat;
5294 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5295 if (!cap)
5296 return -EINVAL;
5298 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5299 return -EINVAL;
5301 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
5303 EXPORT_SYMBOL(pcix_get_max_mmrbc);
5306 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
5307 * @dev: PCI device to query
5309 * Returns mmrbc: maximum memory read count in bytes
5310 * or appropriate error value.
5312 int pcix_get_mmrbc(struct pci_dev *dev)
5314 int cap;
5315 u16 cmd;
5317 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5318 if (!cap)
5319 return -EINVAL;
5321 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5322 return -EINVAL;
5324 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
5326 EXPORT_SYMBOL(pcix_get_mmrbc);
5329 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
5330 * @dev: PCI device to query
5331 * @mmrbc: maximum memory read count in bytes
5332 * valid values are 512, 1024, 2048, 4096
5334 * If possible sets maximum memory read byte count, some bridges have erratas
5335 * that prevent this.
5337 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
5339 int cap;
5340 u32 stat, v, o;
5341 u16 cmd;
5343 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
5344 return -EINVAL;
5346 v = ffs(mmrbc) - 10;
5348 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
5349 if (!cap)
5350 return -EINVAL;
5352 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
5353 return -EINVAL;
5355 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
5356 return -E2BIG;
5358 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
5359 return -EINVAL;
5361 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
5362 if (o != v) {
5363 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
5364 return -EIO;
5366 cmd &= ~PCI_X_CMD_MAX_READ;
5367 cmd |= v << 2;
5368 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
5369 return -EIO;
5371 return 0;
5373 EXPORT_SYMBOL(pcix_set_mmrbc);
5376 * pcie_get_readrq - get PCI Express read request size
5377 * @dev: PCI device to query
5379 * Returns maximum memory read request in bytes
5380 * or appropriate error value.
5382 int pcie_get_readrq(struct pci_dev *dev)
5384 u16 ctl;
5386 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5388 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5390 EXPORT_SYMBOL(pcie_get_readrq);
5393 * pcie_set_readrq - set PCI Express maximum memory read request
5394 * @dev: PCI device to query
5395 * @rq: maximum memory read count in bytes
5396 * valid values are 128, 256, 512, 1024, 2048, 4096
5398 * If possible sets maximum memory read request in bytes
5400 int pcie_set_readrq(struct pci_dev *dev, int rq)
5402 u16 v;
5404 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5405 return -EINVAL;
5408 * If using the "performance" PCIe config, we clamp the
5409 * read rq size to the max packet size to prevent the
5410 * host bridge generating requests larger than we can
5411 * cope with
5413 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5414 int mps = pcie_get_mps(dev);
5416 if (mps < rq)
5417 rq = mps;
5420 v = (ffs(rq) - 8) << 12;
5422 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5423 PCI_EXP_DEVCTL_READRQ, v);
5425 EXPORT_SYMBOL(pcie_set_readrq);
5428 * pcie_get_mps - get PCI Express maximum payload size
5429 * @dev: PCI device to query
5431 * Returns maximum payload size in bytes
5433 int pcie_get_mps(struct pci_dev *dev)
5435 u16 ctl;
5437 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5439 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5441 EXPORT_SYMBOL(pcie_get_mps);
5444 * pcie_set_mps - set PCI Express maximum payload size
5445 * @dev: PCI device to query
5446 * @mps: maximum payload size in bytes
5447 * valid values are 128, 256, 512, 1024, 2048, 4096
5449 * If possible sets maximum payload size
5451 int pcie_set_mps(struct pci_dev *dev, int mps)
5453 u16 v;
5455 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5456 return -EINVAL;
5458 v = ffs(mps) - 8;
5459 if (v > dev->pcie_mpss)
5460 return -EINVAL;
5461 v <<= 5;
5463 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5464 PCI_EXP_DEVCTL_PAYLOAD, v);
5466 EXPORT_SYMBOL(pcie_set_mps);
5469 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5470 * device and its bandwidth limitation
5471 * @dev: PCI device to query
5472 * @limiting_dev: storage for device causing the bandwidth limitation
5473 * @speed: storage for speed of limiting device
5474 * @width: storage for width of limiting device
5476 * Walk up the PCI device chain and find the point where the minimum
5477 * bandwidth is available. Return the bandwidth available there and (if
5478 * limiting_dev, speed, and width pointers are supplied) information about
5479 * that point. The bandwidth returned is in Mb/s, i.e., megabits/second of
5480 * raw bandwidth.
5482 u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5483 enum pci_bus_speed *speed,
5484 enum pcie_link_width *width)
5486 u16 lnksta;
5487 enum pci_bus_speed next_speed;
5488 enum pcie_link_width next_width;
5489 u32 bw, next_bw;
5491 if (speed)
5492 *speed = PCI_SPEED_UNKNOWN;
5493 if (width)
5494 *width = PCIE_LNK_WIDTH_UNKNOWN;
5496 bw = 0;
5498 while (dev) {
5499 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5501 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5502 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5503 PCI_EXP_LNKSTA_NLW_SHIFT;
5505 next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5507 /* Check if current device limits the total bandwidth */
5508 if (!bw || next_bw <= bw) {
5509 bw = next_bw;
5511 if (limiting_dev)
5512 *limiting_dev = dev;
5513 if (speed)
5514 *speed = next_speed;
5515 if (width)
5516 *width = next_width;
5519 dev = pci_upstream_bridge(dev);
5522 return bw;
5524 EXPORT_SYMBOL(pcie_bandwidth_available);
5527 * pcie_get_speed_cap - query for the PCI device's link speed capability
5528 * @dev: PCI device to query
5530 * Query the PCI device speed capability. Return the maximum link speed
5531 * supported by the device.
5533 enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5535 u32 lnkcap2, lnkcap;
5538 * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18. The
5539 * implementation note there recommends using the Supported Link
5540 * Speeds Vector in Link Capabilities 2 when supported.
5542 * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
5543 * should use the Supported Link Speeds field in Link Capabilities,
5544 * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
5546 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5547 if (lnkcap2) { /* PCIe r3.0-compliant */
5548 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_16_0GB)
5549 return PCIE_SPEED_16_0GT;
5550 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
5551 return PCIE_SPEED_8_0GT;
5552 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
5553 return PCIE_SPEED_5_0GT;
5554 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
5555 return PCIE_SPEED_2_5GT;
5556 return PCI_SPEED_UNKNOWN;
5559 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5560 if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
5561 return PCIE_SPEED_5_0GT;
5562 else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
5563 return PCIE_SPEED_2_5GT;
5565 return PCI_SPEED_UNKNOWN;
5567 EXPORT_SYMBOL(pcie_get_speed_cap);
5570 * pcie_get_width_cap - query for the PCI device's link width capability
5571 * @dev: PCI device to query
5573 * Query the PCI device width capability. Return the maximum link width
5574 * supported by the device.
5576 enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5578 u32 lnkcap;
5580 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5581 if (lnkcap)
5582 return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5584 return PCIE_LNK_WIDTH_UNKNOWN;
5586 EXPORT_SYMBOL(pcie_get_width_cap);
5589 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5590 * @dev: PCI device
5591 * @speed: storage for link speed
5592 * @width: storage for link width
5594 * Calculate a PCI device's link bandwidth by querying for its link speed
5595 * and width, multiplying them, and applying encoding overhead. The result
5596 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5598 u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5599 enum pcie_link_width *width)
5601 *speed = pcie_get_speed_cap(dev);
5602 *width = pcie_get_width_cap(dev);
5604 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5605 return 0;
5607 return *width * PCIE_SPEED2MBS_ENC(*speed);
5611 * __pcie_print_link_status - Report the PCI device's link speed and width
5612 * @dev: PCI device to query
5613 * @verbose: Print info even when enough bandwidth is available
5615 * If the available bandwidth at the device is less than the device is
5616 * capable of, report the device's maximum possible bandwidth and the
5617 * upstream link that limits its performance. If @verbose, always print
5618 * the available bandwidth, even if the device isn't constrained.
5620 void __pcie_print_link_status(struct pci_dev *dev, bool verbose)
5622 enum pcie_link_width width, width_cap;
5623 enum pci_bus_speed speed, speed_cap;
5624 struct pci_dev *limiting_dev = NULL;
5625 u32 bw_avail, bw_cap;
5627 bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5628 bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5630 if (bw_avail >= bw_cap && verbose)
5631 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5632 bw_cap / 1000, bw_cap % 1000,
5633 PCIE_SPEED2STR(speed_cap), width_cap);
5634 else if (bw_avail < bw_cap)
5635 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",
5636 bw_avail / 1000, bw_avail % 1000,
5637 PCIE_SPEED2STR(speed), width,
5638 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5639 bw_cap / 1000, bw_cap % 1000,
5640 PCIE_SPEED2STR(speed_cap), width_cap);
5644 * pcie_print_link_status - Report the PCI device's link speed and width
5645 * @dev: PCI device to query
5647 * Report the available bandwidth at the device.
5649 void pcie_print_link_status(struct pci_dev *dev)
5651 __pcie_print_link_status(dev, true);
5653 EXPORT_SYMBOL(pcie_print_link_status);
5656 * pci_select_bars - Make BAR mask from the type of resource
5657 * @dev: the PCI device for which BAR mask is made
5658 * @flags: resource type mask to be selected
5660 * This helper routine makes bar mask from the type of resource.
5662 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5664 int i, bars = 0;
5665 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5666 if (pci_resource_flags(dev, i) & flags)
5667 bars |= (1 << i);
5668 return bars;
5670 EXPORT_SYMBOL(pci_select_bars);
5672 /* Some architectures require additional programming to enable VGA */
5673 static arch_set_vga_state_t arch_set_vga_state;
5675 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5677 arch_set_vga_state = func; /* NULL disables */
5680 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5681 unsigned int command_bits, u32 flags)
5683 if (arch_set_vga_state)
5684 return arch_set_vga_state(dev, decode, command_bits,
5685 flags);
5686 return 0;
5690 * pci_set_vga_state - set VGA decode state on device and parents if requested
5691 * @dev: the PCI device
5692 * @decode: true = enable decoding, false = disable decoding
5693 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5694 * @flags: traverse ancestors and change bridges
5695 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5697 int pci_set_vga_state(struct pci_dev *dev, bool decode,
5698 unsigned int command_bits, u32 flags)
5700 struct pci_bus *bus;
5701 struct pci_dev *bridge;
5702 u16 cmd;
5703 int rc;
5705 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5707 /* ARCH specific VGA enables */
5708 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5709 if (rc)
5710 return rc;
5712 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5713 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5714 if (decode == true)
5715 cmd |= command_bits;
5716 else
5717 cmd &= ~command_bits;
5718 pci_write_config_word(dev, PCI_COMMAND, cmd);
5721 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5722 return 0;
5724 bus = dev->bus;
5725 while (bus) {
5726 bridge = bus->self;
5727 if (bridge) {
5728 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5729 &cmd);
5730 if (decode == true)
5731 cmd |= PCI_BRIDGE_CTL_VGA;
5732 else
5733 cmd &= ~PCI_BRIDGE_CTL_VGA;
5734 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5735 cmd);
5737 bus = bus->parent;
5739 return 0;
5743 * pci_add_dma_alias - Add a DMA devfn alias for a device
5744 * @dev: the PCI device for which alias is added
5745 * @devfn: alias slot and function
5747 * This helper encodes an 8-bit devfn as a bit number in dma_alias_mask
5748 * which is used to program permissible bus-devfn source addresses for DMA
5749 * requests in an IOMMU. These aliases factor into IOMMU group creation
5750 * and are useful for devices generating DMA requests beyond or different
5751 * from their logical bus-devfn. Examples include device quirks where the
5752 * device simply uses the wrong devfn, as well as non-transparent bridges
5753 * where the alias may be a proxy for devices in another domain.
5755 * IOMMU group creation is performed during device discovery or addition,
5756 * prior to any potential DMA mapping and therefore prior to driver probing
5757 * (especially for userspace assigned devices where IOMMU group definition
5758 * cannot be left as a userspace activity). DMA aliases should therefore
5759 * be configured via quirks, such as the PCI fixup header quirk.
5761 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5763 if (!dev->dma_alias_mask)
5764 dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
5765 sizeof(long), GFP_KERNEL);
5766 if (!dev->dma_alias_mask) {
5767 pci_warn(dev, "Unable to allocate DMA alias mask\n");
5768 return;
5771 set_bit(devfn, dev->dma_alias_mask);
5772 pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
5773 PCI_SLOT(devfn), PCI_FUNC(devfn));
5776 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5778 return (dev1->dma_alias_mask &&
5779 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5780 (dev2->dma_alias_mask &&
5781 test_bit(dev1->devfn, dev2->dma_alias_mask));
5784 bool pci_device_is_present(struct pci_dev *pdev)
5786 u32 v;
5788 if (pci_dev_is_disconnected(pdev))
5789 return false;
5790 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5792 EXPORT_SYMBOL_GPL(pci_device_is_present);
5794 void pci_ignore_hotplug(struct pci_dev *dev)
5796 struct pci_dev *bridge = dev->bus->self;
5798 dev->ignore_hotplug = 1;
5799 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5800 if (bridge)
5801 bridge->ignore_hotplug = 1;
5803 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5805 resource_size_t __weak pcibios_default_alignment(void)
5807 return 0;
5810 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5811 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5812 static DEFINE_SPINLOCK(resource_alignment_lock);
5815 * pci_specified_resource_alignment - get resource alignment specified by user.
5816 * @dev: the PCI device to get
5817 * @resize: whether or not to change resources' size when reassigning alignment
5819 * RETURNS: Resource alignment if it is specified.
5820 * Zero if it is not specified.
5822 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5823 bool *resize)
5825 int align_order, count;
5826 resource_size_t align = pcibios_default_alignment();
5827 const char *p;
5828 int ret;
5830 spin_lock(&resource_alignment_lock);
5831 p = resource_alignment_param;
5832 if (!*p && !align)
5833 goto out;
5834 if (pci_has_flag(PCI_PROBE_ONLY)) {
5835 align = 0;
5836 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5837 goto out;
5840 while (*p) {
5841 count = 0;
5842 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5843 p[count] == '@') {
5844 p += count + 1;
5845 } else {
5846 align_order = -1;
5849 ret = pci_dev_str_match(dev, p, &p);
5850 if (ret == 1) {
5851 *resize = true;
5852 if (align_order == -1)
5853 align = PAGE_SIZE;
5854 else
5855 align = 1 << align_order;
5856 break;
5857 } else if (ret < 0) {
5858 pr_err("PCI: Can't parse resource_alignment parameter: %s\n",
5860 break;
5863 if (*p != ';' && *p != ',') {
5864 /* End of param or invalid format */
5865 break;
5867 p++;
5869 out:
5870 spin_unlock(&resource_alignment_lock);
5871 return align;
5874 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5875 resource_size_t align, bool resize)
5877 struct resource *r = &dev->resource[bar];
5878 resource_size_t size;
5880 if (!(r->flags & IORESOURCE_MEM))
5881 return;
5883 if (r->flags & IORESOURCE_PCI_FIXED) {
5884 pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5885 bar, r, (unsigned long long)align);
5886 return;
5889 size = resource_size(r);
5890 if (size >= align)
5891 return;
5894 * Increase the alignment of the resource. There are two ways we
5895 * can do this:
5897 * 1) Increase the size of the resource. BARs are aligned on their
5898 * size, so when we reallocate space for this resource, we'll
5899 * allocate it with the larger alignment. This also prevents
5900 * assignment of any other BARs inside the alignment region, so
5901 * if we're requesting page alignment, this means no other BARs
5902 * will share the page.
5904 * The disadvantage is that this makes the resource larger than
5905 * the hardware BAR, which may break drivers that compute things
5906 * based on the resource size, e.g., to find registers at a
5907 * fixed offset before the end of the BAR.
5909 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5910 * set r->start to the desired alignment. By itself this
5911 * doesn't prevent other BARs being put inside the alignment
5912 * region, but if we realign *every* resource of every device in
5913 * the system, none of them will share an alignment region.
5915 * When the user has requested alignment for only some devices via
5916 * the "pci=resource_alignment" argument, "resize" is true and we
5917 * use the first method. Otherwise we assume we're aligning all
5918 * devices and we use the second.
5921 pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
5922 bar, r, (unsigned long long)align);
5924 if (resize) {
5925 r->start = 0;
5926 r->end = align - 1;
5927 } else {
5928 r->flags &= ~IORESOURCE_SIZEALIGN;
5929 r->flags |= IORESOURCE_STARTALIGN;
5930 r->start = align;
5931 r->end = r->start + size - 1;
5933 r->flags |= IORESOURCE_UNSET;
5937 * This function disables memory decoding and releases memory resources
5938 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5939 * It also rounds up size to specified alignment.
5940 * Later on, the kernel will assign page-aligned memory resource back
5941 * to the device.
5943 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5945 int i;
5946 struct resource *r;
5947 resource_size_t align;
5948 u16 command;
5949 bool resize = false;
5952 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
5953 * 3.4.1.11. Their resources are allocated from the space
5954 * described by the VF BARx register in the PF's SR-IOV capability.
5955 * We can't influence their alignment here.
5957 if (dev->is_virtfn)
5958 return;
5960 /* check if specified PCI is target device to reassign */
5961 align = pci_specified_resource_alignment(dev, &resize);
5962 if (!align)
5963 return;
5965 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5966 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5967 pci_warn(dev, "Can't reassign resources to host bridge\n");
5968 return;
5971 pci_read_config_word(dev, PCI_COMMAND, &command);
5972 command &= ~PCI_COMMAND_MEMORY;
5973 pci_write_config_word(dev, PCI_COMMAND, command);
5975 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
5976 pci_request_resource_alignment(dev, i, align, resize);
5979 * Need to disable bridge's resource window,
5980 * to enable the kernel to reassign new resource
5981 * window later on.
5983 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5984 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5985 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5986 r = &dev->resource[i];
5987 if (!(r->flags & IORESOURCE_MEM))
5988 continue;
5989 r->flags |= IORESOURCE_UNSET;
5990 r->end = resource_size(r) - 1;
5991 r->start = 0;
5993 pci_disable_bridge_window(dev);
5997 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5999 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
6000 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
6001 spin_lock(&resource_alignment_lock);
6002 strncpy(resource_alignment_param, buf, count);
6003 resource_alignment_param[count] = '\0';
6004 spin_unlock(&resource_alignment_lock);
6005 return count;
6008 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
6010 size_t count;
6011 spin_lock(&resource_alignment_lock);
6012 count = snprintf(buf, size, "%s", resource_alignment_param);
6013 spin_unlock(&resource_alignment_lock);
6014 return count;
6017 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
6019 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
6022 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
6023 const char *buf, size_t count)
6025 return pci_set_resource_alignment_param(buf, count);
6028 static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
6029 pci_resource_alignment_store);
6031 static int __init pci_resource_alignment_sysfs_init(void)
6033 return bus_create_file(&pci_bus_type,
6034 &bus_attr_resource_alignment);
6036 late_initcall(pci_resource_alignment_sysfs_init);
6038 static void pci_no_domains(void)
6040 #ifdef CONFIG_PCI_DOMAINS
6041 pci_domains_supported = 0;
6042 #endif
6045 #ifdef CONFIG_PCI_DOMAINS_GENERIC
6046 static atomic_t __domain_nr = ATOMIC_INIT(-1);
6048 static int pci_get_new_domain_nr(void)
6050 return atomic_inc_return(&__domain_nr);
6053 static int of_pci_bus_find_domain_nr(struct device *parent)
6055 static int use_dt_domains = -1;
6056 int domain = -1;
6058 if (parent)
6059 domain = of_get_pci_domain_nr(parent->of_node);
6061 * Check DT domain and use_dt_domains values.
6063 * If DT domain property is valid (domain >= 0) and
6064 * use_dt_domains != 0, the DT assignment is valid since this means
6065 * we have not previously allocated a domain number by using
6066 * pci_get_new_domain_nr(); we should also update use_dt_domains to
6067 * 1, to indicate that we have just assigned a domain number from
6068 * DT.
6070 * If DT domain property value is not valid (ie domain < 0), and we
6071 * have not previously assigned a domain number from DT
6072 * (use_dt_domains != 1) we should assign a domain number by
6073 * using the:
6075 * pci_get_new_domain_nr()
6077 * API and update the use_dt_domains value to keep track of method we
6078 * are using to assign domain numbers (use_dt_domains = 0).
6080 * All other combinations imply we have a platform that is trying
6081 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
6082 * which is a recipe for domain mishandling and it is prevented by
6083 * invalidating the domain value (domain = -1) and printing a
6084 * corresponding error.
6086 if (domain >= 0 && use_dt_domains) {
6087 use_dt_domains = 1;
6088 } else if (domain < 0 && use_dt_domains != 1) {
6089 use_dt_domains = 0;
6090 domain = pci_get_new_domain_nr();
6091 } else {
6092 if (parent)
6093 pr_err("Node %pOF has ", parent->of_node);
6094 pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
6095 domain = -1;
6098 return domain;
6101 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
6103 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
6104 acpi_pci_bus_find_domain_nr(bus);
6106 #endif
6109 * pci_ext_cfg_avail - can we access extended PCI config space?
6111 * Returns 1 if we can access PCI extended config space (offsets
6112 * greater than 0xff). This is the default implementation. Architecture
6113 * implementations can override this.
6115 int __weak pci_ext_cfg_avail(void)
6117 return 1;
6120 void __weak pci_fixup_cardbus(struct pci_bus *bus)
6123 EXPORT_SYMBOL(pci_fixup_cardbus);
6125 static int __init pci_setup(char *str)
6127 while (str) {
6128 char *k = strchr(str, ',');
6129 if (k)
6130 *k++ = 0;
6131 if (*str && (str = pcibios_setup(str)) && *str) {
6132 if (!strcmp(str, "nomsi")) {
6133 pci_no_msi();
6134 } else if (!strncmp(str, "noats", 5)) {
6135 pr_info("PCIe: ATS is disabled\n");
6136 pcie_ats_disabled = true;
6137 } else if (!strcmp(str, "noaer")) {
6138 pci_no_aer();
6139 } else if (!strcmp(str, "earlydump")) {
6140 pci_early_dump = true;
6141 } else if (!strncmp(str, "realloc=", 8)) {
6142 pci_realloc_get_opt(str + 8);
6143 } else if (!strncmp(str, "realloc", 7)) {
6144 pci_realloc_get_opt("on");
6145 } else if (!strcmp(str, "nodomains")) {
6146 pci_no_domains();
6147 } else if (!strncmp(str, "noari", 5)) {
6148 pcie_ari_disabled = true;
6149 } else if (!strncmp(str, "cbiosize=", 9)) {
6150 pci_cardbus_io_size = memparse(str + 9, &str);
6151 } else if (!strncmp(str, "cbmemsize=", 10)) {
6152 pci_cardbus_mem_size = memparse(str + 10, &str);
6153 } else if (!strncmp(str, "resource_alignment=", 19)) {
6154 pci_set_resource_alignment_param(str + 19,
6155 strlen(str + 19));
6156 } else if (!strncmp(str, "ecrc=", 5)) {
6157 pcie_ecrc_get_policy(str + 5);
6158 } else if (!strncmp(str, "hpiosize=", 9)) {
6159 pci_hotplug_io_size = memparse(str + 9, &str);
6160 } else if (!strncmp(str, "hpmemsize=", 10)) {
6161 pci_hotplug_mem_size = memparse(str + 10, &str);
6162 } else if (!strncmp(str, "hpbussize=", 10)) {
6163 pci_hotplug_bus_size =
6164 simple_strtoul(str + 10, &str, 0);
6165 if (pci_hotplug_bus_size > 0xff)
6166 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
6167 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
6168 pcie_bus_config = PCIE_BUS_TUNE_OFF;
6169 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
6170 pcie_bus_config = PCIE_BUS_SAFE;
6171 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
6172 pcie_bus_config = PCIE_BUS_PERFORMANCE;
6173 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
6174 pcie_bus_config = PCIE_BUS_PEER2PEER;
6175 } else if (!strncmp(str, "pcie_scan_all", 13)) {
6176 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
6177 } else if (!strncmp(str, "disable_acs_redir=", 18)) {
6178 disable_acs_redir_param = str + 18;
6179 } else {
6180 printk(KERN_ERR "PCI: Unknown option `%s'\n",
6181 str);
6184 str = k;
6186 return 0;
6188 early_param("pci", pci_setup);
6191 * 'disable_acs_redir_param' is initialized in pci_setup(), above, to point
6192 * to data in the __initdata section which will be freed after the init
6193 * sequence is complete. We can't allocate memory in pci_setup() because some
6194 * architectures do not have any memory allocation service available during
6195 * an early_param() call. So we allocate memory and copy the variable here
6196 * before the init section is freed.
6198 static int __init pci_realloc_setup_params(void)
6200 disable_acs_redir_param = kstrdup(disable_acs_redir_param, GFP_KERNEL);
6202 return 0;
6204 pure_initcall(pci_realloc_setup_params);