xref: /linux/drivers/pci/pci.c (revision 6faadbbb7f9da70ce484f98f72223c20125a1009)
1 /*
2  *	PCI Bus Services, see include/linux/pci.h for further explanation.
3  *
4  *	Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
5  *	David Mosberger-Tang
6  *
7  *	Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
8  */
9 
10 #include <linux/acpi.h>
11 #include <linux/kernel.h>
12 #include <linux/delay.h>
13 #include <linux/dmi.h>
14 #include <linux/init.h>
15 #include <linux/of.h>
16 #include <linux/of_pci.h>
17 #include <linux/pci.h>
18 #include <linux/pm.h>
19 #include <linux/slab.h>
20 #include <linux/module.h>
21 #include <linux/spinlock.h>
22 #include <linux/string.h>
23 #include <linux/log2.h>
24 #include <linux/pci-aspm.h>
25 #include <linux/pm_wakeup.h>
26 #include <linux/interrupt.h>
27 #include <linux/device.h>
28 #include <linux/pm_runtime.h>
29 #include <linux/pci_hotplug.h>
30 #include <linux/vmalloc.h>
31 #include <linux/pci-ats.h>
32 #include <asm/setup.h>
33 #include <asm/dma.h>
34 #include <linux/aer.h>
35 #include "pci.h"
36 
37 const char *pci_power_names[] = {
38 	"error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
39 };
40 EXPORT_SYMBOL_GPL(pci_power_names);
41 
42 int isa_dma_bridge_buggy;
43 EXPORT_SYMBOL(isa_dma_bridge_buggy);
44 
45 int pci_pci_problems;
46 EXPORT_SYMBOL(pci_pci_problems);
47 
48 unsigned int pci_pm_d3_delay;
49 
50 static void pci_pme_list_scan(struct work_struct *work);
51 
52 static LIST_HEAD(pci_pme_list);
53 static DEFINE_MUTEX(pci_pme_list_mutex);
54 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
55 static DEFINE_MUTEX(pci_bridge_mutex);
56 
57 struct pci_pme_device {
58 	struct list_head list;
59 	struct pci_dev *dev;
60 };
61 
62 #define PME_TIMEOUT 1000 /* How long between PME checks */
63 
64 static void pci_dev_d3_sleep(struct pci_dev *dev)
65 {
66 	unsigned int delay = dev->d3_delay;
67 
68 	if (delay < pci_pm_d3_delay)
69 		delay = pci_pm_d3_delay;
70 
71 	if (delay)
72 		msleep(delay);
73 }
74 
75 #ifdef CONFIG_PCI_DOMAINS
76 int pci_domains_supported = 1;
77 #endif
78 
79 #define DEFAULT_CARDBUS_IO_SIZE		(256)
80 #define DEFAULT_CARDBUS_MEM_SIZE	(64*1024*1024)
81 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
82 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
83 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
84 
85 #define DEFAULT_HOTPLUG_IO_SIZE		(256)
86 #define DEFAULT_HOTPLUG_MEM_SIZE	(2*1024*1024)
87 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
88 unsigned long pci_hotplug_io_size  = DEFAULT_HOTPLUG_IO_SIZE;
89 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
90 
91 #define DEFAULT_HOTPLUG_BUS_SIZE	1
92 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
93 
94 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
95 
96 /*
97  * The default CLS is used if arch didn't set CLS explicitly and not
98  * all pci devices agree on the same value.  Arch can override either
99  * the dfl or actual value as it sees fit.  Don't forget this is
100  * measured in 32-bit words, not bytes.
101  */
102 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
103 u8 pci_cache_line_size;
104 
105 /*
106  * If we set up a device for bus mastering, we need to check the latency
107  * timer as certain BIOSes forget to set it properly.
108  */
109 unsigned int pcibios_max_latency = 255;
110 
111 /* If set, the PCIe ARI capability will not be used. */
112 static bool pcie_ari_disabled;
113 
114 /* Disable bridge_d3 for all PCIe ports */
115 static bool pci_bridge_d3_disable;
116 /* Force bridge_d3 for all PCIe ports */
117 static bool pci_bridge_d3_force;
118 
119 static int __init pcie_port_pm_setup(char *str)
120 {
121 	if (!strcmp(str, "off"))
122 		pci_bridge_d3_disable = true;
123 	else if (!strcmp(str, "force"))
124 		pci_bridge_d3_force = true;
125 	return 1;
126 }
127 __setup("pcie_port_pm=", pcie_port_pm_setup);
128 
129 /**
130  * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
131  * @bus: pointer to PCI bus structure to search
132  *
133  * Given a PCI bus, returns the highest PCI bus number present in the set
134  * including the given PCI bus and its list of child PCI buses.
135  */
136 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
137 {
138 	struct pci_bus *tmp;
139 	unsigned char max, n;
140 
141 	max = bus->busn_res.end;
142 	list_for_each_entry(tmp, &bus->children, node) {
143 		n = pci_bus_max_busnr(tmp);
144 		if (n > max)
145 			max = n;
146 	}
147 	return max;
148 }
149 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
150 
151 #ifdef CONFIG_HAS_IOMEM
152 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
153 {
154 	struct resource *res = &pdev->resource[bar];
155 
156 	/*
157 	 * Make sure the BAR is actually a memory resource, not an IO resource
158 	 */
159 	if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
160 		dev_warn(&pdev->dev, "can't ioremap BAR %d: %pR\n", bar, res);
161 		return NULL;
162 	}
163 	return ioremap_nocache(res->start, resource_size(res));
164 }
165 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
166 
167 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
168 {
169 	/*
170 	 * Make sure the BAR is actually a memory resource, not an IO resource
171 	 */
172 	if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
173 		WARN_ON(1);
174 		return NULL;
175 	}
176 	return ioremap_wc(pci_resource_start(pdev, bar),
177 			  pci_resource_len(pdev, bar));
178 }
179 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
180 #endif
181 
182 
183 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
184 				   u8 pos, int cap, int *ttl)
185 {
186 	u8 id;
187 	u16 ent;
188 
189 	pci_bus_read_config_byte(bus, devfn, pos, &pos);
190 
191 	while ((*ttl)--) {
192 		if (pos < 0x40)
193 			break;
194 		pos &= ~3;
195 		pci_bus_read_config_word(bus, devfn, pos, &ent);
196 
197 		id = ent & 0xff;
198 		if (id == 0xff)
199 			break;
200 		if (id == cap)
201 			return pos;
202 		pos = (ent >> 8);
203 	}
204 	return 0;
205 }
206 
207 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
208 			       u8 pos, int cap)
209 {
210 	int ttl = PCI_FIND_CAP_TTL;
211 
212 	return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
213 }
214 
215 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
216 {
217 	return __pci_find_next_cap(dev->bus, dev->devfn,
218 				   pos + PCI_CAP_LIST_NEXT, cap);
219 }
220 EXPORT_SYMBOL_GPL(pci_find_next_capability);
221 
222 static int __pci_bus_find_cap_start(struct pci_bus *bus,
223 				    unsigned int devfn, u8 hdr_type)
224 {
225 	u16 status;
226 
227 	pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
228 	if (!(status & PCI_STATUS_CAP_LIST))
229 		return 0;
230 
231 	switch (hdr_type) {
232 	case PCI_HEADER_TYPE_NORMAL:
233 	case PCI_HEADER_TYPE_BRIDGE:
234 		return PCI_CAPABILITY_LIST;
235 	case PCI_HEADER_TYPE_CARDBUS:
236 		return PCI_CB_CAPABILITY_LIST;
237 	}
238 
239 	return 0;
240 }
241 
242 /**
243  * pci_find_capability - query for devices' capabilities
244  * @dev: PCI device to query
245  * @cap: capability code
246  *
247  * Tell if a device supports a given PCI capability.
248  * Returns the address of the requested capability structure within the
249  * device's PCI configuration space or 0 in case the device does not
250  * support it.  Possible values for @cap:
251  *
252  *  %PCI_CAP_ID_PM           Power Management
253  *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
254  *  %PCI_CAP_ID_VPD          Vital Product Data
255  *  %PCI_CAP_ID_SLOTID       Slot Identification
256  *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
257  *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
258  *  %PCI_CAP_ID_PCIX         PCI-X
259  *  %PCI_CAP_ID_EXP          PCI Express
260  */
261 int pci_find_capability(struct pci_dev *dev, int cap)
262 {
263 	int pos;
264 
265 	pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
266 	if (pos)
267 		pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
268 
269 	return pos;
270 }
271 EXPORT_SYMBOL(pci_find_capability);
272 
273 /**
274  * pci_bus_find_capability - query for devices' capabilities
275  * @bus:   the PCI bus to query
276  * @devfn: PCI device to query
277  * @cap:   capability code
278  *
279  * Like pci_find_capability() but works for pci devices that do not have a
280  * pci_dev structure set up yet.
281  *
282  * Returns the address of the requested capability structure within the
283  * device's PCI configuration space or 0 in case the device does not
284  * support it.
285  */
286 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
287 {
288 	int pos;
289 	u8 hdr_type;
290 
291 	pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
292 
293 	pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
294 	if (pos)
295 		pos = __pci_find_next_cap(bus, devfn, pos, cap);
296 
297 	return pos;
298 }
299 EXPORT_SYMBOL(pci_bus_find_capability);
300 
301 /**
302  * pci_find_next_ext_capability - Find an extended capability
303  * @dev: PCI device to query
304  * @start: address at which to start looking (0 to start at beginning of list)
305  * @cap: capability code
306  *
307  * Returns the address of the next matching extended capability structure
308  * within the device's PCI configuration space or 0 if the device does
309  * not support it.  Some capabilities can occur several times, e.g., the
310  * vendor-specific capability, and this provides a way to find them all.
311  */
312 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
313 {
314 	u32 header;
315 	int ttl;
316 	int pos = PCI_CFG_SPACE_SIZE;
317 
318 	/* minimum 8 bytes per capability */
319 	ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
320 
321 	if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
322 		return 0;
323 
324 	if (start)
325 		pos = start;
326 
327 	if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
328 		return 0;
329 
330 	/*
331 	 * If we have no capabilities, this is indicated by cap ID,
332 	 * cap version and next pointer all being 0.
333 	 */
334 	if (header == 0)
335 		return 0;
336 
337 	while (ttl-- > 0) {
338 		if (PCI_EXT_CAP_ID(header) == cap && pos != start)
339 			return pos;
340 
341 		pos = PCI_EXT_CAP_NEXT(header);
342 		if (pos < PCI_CFG_SPACE_SIZE)
343 			break;
344 
345 		if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
346 			break;
347 	}
348 
349 	return 0;
350 }
351 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
352 
353 /**
354  * pci_find_ext_capability - Find an extended capability
355  * @dev: PCI device to query
356  * @cap: capability code
357  *
358  * Returns the address of the requested extended capability structure
359  * within the device's PCI configuration space or 0 if the device does
360  * not support it.  Possible values for @cap:
361  *
362  *  %PCI_EXT_CAP_ID_ERR		Advanced Error Reporting
363  *  %PCI_EXT_CAP_ID_VC		Virtual Channel
364  *  %PCI_EXT_CAP_ID_DSN		Device Serial Number
365  *  %PCI_EXT_CAP_ID_PWR		Power Budgeting
366  */
367 int pci_find_ext_capability(struct pci_dev *dev, int cap)
368 {
369 	return pci_find_next_ext_capability(dev, 0, cap);
370 }
371 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
372 
373 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
374 {
375 	int rc, ttl = PCI_FIND_CAP_TTL;
376 	u8 cap, mask;
377 
378 	if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
379 		mask = HT_3BIT_CAP_MASK;
380 	else
381 		mask = HT_5BIT_CAP_MASK;
382 
383 	pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
384 				      PCI_CAP_ID_HT, &ttl);
385 	while (pos) {
386 		rc = pci_read_config_byte(dev, pos + 3, &cap);
387 		if (rc != PCIBIOS_SUCCESSFUL)
388 			return 0;
389 
390 		if ((cap & mask) == ht_cap)
391 			return pos;
392 
393 		pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
394 					      pos + PCI_CAP_LIST_NEXT,
395 					      PCI_CAP_ID_HT, &ttl);
396 	}
397 
398 	return 0;
399 }
400 /**
401  * pci_find_next_ht_capability - query a device's Hypertransport capabilities
402  * @dev: PCI device to query
403  * @pos: Position from which to continue searching
404  * @ht_cap: Hypertransport capability code
405  *
406  * To be used in conjunction with pci_find_ht_capability() to search for
407  * all capabilities matching @ht_cap. @pos should always be a value returned
408  * from pci_find_ht_capability().
409  *
410  * NB. To be 100% safe against broken PCI devices, the caller should take
411  * steps to avoid an infinite loop.
412  */
413 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
414 {
415 	return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
416 }
417 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
418 
419 /**
420  * pci_find_ht_capability - query a device's Hypertransport capabilities
421  * @dev: PCI device to query
422  * @ht_cap: Hypertransport capability code
423  *
424  * Tell if a device supports a given Hypertransport capability.
425  * Returns an address within the device's PCI configuration space
426  * or 0 in case the device does not support the request capability.
427  * The address points to the PCI capability, of type PCI_CAP_ID_HT,
428  * which has a Hypertransport capability matching @ht_cap.
429  */
430 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
431 {
432 	int pos;
433 
434 	pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
435 	if (pos)
436 		pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
437 
438 	return pos;
439 }
440 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
441 
442 /**
443  * pci_find_parent_resource - return resource region of parent bus of given region
444  * @dev: PCI device structure contains resources to be searched
445  * @res: child resource record for which parent is sought
446  *
447  *  For given resource region of given device, return the resource
448  *  region of parent bus the given region is contained in.
449  */
450 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
451 					  struct resource *res)
452 {
453 	const struct pci_bus *bus = dev->bus;
454 	struct resource *r;
455 	int i;
456 
457 	pci_bus_for_each_resource(bus, r, i) {
458 		if (!r)
459 			continue;
460 		if (resource_contains(r, res)) {
461 
462 			/*
463 			 * If the window is prefetchable but the BAR is
464 			 * not, the allocator made a mistake.
465 			 */
466 			if (r->flags & IORESOURCE_PREFETCH &&
467 			    !(res->flags & IORESOURCE_PREFETCH))
468 				return NULL;
469 
470 			/*
471 			 * If we're below a transparent bridge, there may
472 			 * be both a positively-decoded aperture and a
473 			 * subtractively-decoded region that contain the BAR.
474 			 * We want the positively-decoded one, so this depends
475 			 * on pci_bus_for_each_resource() giving us those
476 			 * first.
477 			 */
478 			return r;
479 		}
480 	}
481 	return NULL;
482 }
483 EXPORT_SYMBOL(pci_find_parent_resource);
484 
485 /**
486  * pci_find_resource - Return matching PCI device resource
487  * @dev: PCI device to query
488  * @res: Resource to look for
489  *
490  * Goes over standard PCI resources (BARs) and checks if the given resource
491  * is partially or fully contained in any of them. In that case the
492  * matching resource is returned, %NULL otherwise.
493  */
494 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
495 {
496 	int i;
497 
498 	for (i = 0; i < PCI_ROM_RESOURCE; i++) {
499 		struct resource *r = &dev->resource[i];
500 
501 		if (r->start && resource_contains(r, res))
502 			return r;
503 	}
504 
505 	return NULL;
506 }
507 EXPORT_SYMBOL(pci_find_resource);
508 
509 /**
510  * pci_find_pcie_root_port - return PCIe Root Port
511  * @dev: PCI device to query
512  *
513  * Traverse up the parent chain and return the PCIe Root Port PCI Device
514  * for a given PCI Device.
515  */
516 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
517 {
518 	struct pci_dev *bridge, *highest_pcie_bridge = dev;
519 
520 	bridge = pci_upstream_bridge(dev);
521 	while (bridge && pci_is_pcie(bridge)) {
522 		highest_pcie_bridge = bridge;
523 		bridge = pci_upstream_bridge(bridge);
524 	}
525 
526 	if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
527 		return NULL;
528 
529 	return highest_pcie_bridge;
530 }
531 EXPORT_SYMBOL(pci_find_pcie_root_port);
532 
533 /**
534  * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
535  * @dev: the PCI device to operate on
536  * @pos: config space offset of status word
537  * @mask: mask of bit(s) to care about in status word
538  *
539  * Return 1 when mask bit(s) in status word clear, 0 otherwise.
540  */
541 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
542 {
543 	int i;
544 
545 	/* Wait for Transaction Pending bit clean */
546 	for (i = 0; i < 4; i++) {
547 		u16 status;
548 		if (i)
549 			msleep((1 << (i - 1)) * 100);
550 
551 		pci_read_config_word(dev, pos, &status);
552 		if (!(status & mask))
553 			return 1;
554 	}
555 
556 	return 0;
557 }
558 
559 /**
560  * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
561  * @dev: PCI device to have its BARs restored
562  *
563  * Restore the BAR values for a given device, so as to make it
564  * accessible by its driver.
565  */
566 static void pci_restore_bars(struct pci_dev *dev)
567 {
568 	int i;
569 
570 	for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
571 		pci_update_resource(dev, i);
572 }
573 
574 static const struct pci_platform_pm_ops *pci_platform_pm;
575 
576 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
577 {
578 	if (!ops->is_manageable || !ops->set_state  || !ops->get_state ||
579 	    !ops->choose_state  || !ops->set_wakeup || !ops->need_resume)
580 		return -EINVAL;
581 	pci_platform_pm = ops;
582 	return 0;
583 }
584 
585 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
586 {
587 	return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
588 }
589 
590 static inline int platform_pci_set_power_state(struct pci_dev *dev,
591 					       pci_power_t t)
592 {
593 	return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
594 }
595 
596 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
597 {
598 	return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
599 }
600 
601 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
602 {
603 	return pci_platform_pm ?
604 			pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
605 }
606 
607 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
608 {
609 	return pci_platform_pm ?
610 			pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
611 }
612 
613 static inline bool platform_pci_need_resume(struct pci_dev *dev)
614 {
615 	return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
616 }
617 
618 /**
619  * pci_raw_set_power_state - Use PCI PM registers to set the power state of
620  *                           given PCI device
621  * @dev: PCI device to handle.
622  * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
623  *
624  * RETURN VALUE:
625  * -EINVAL if the requested state is invalid.
626  * -EIO if device does not support PCI PM or its PM capabilities register has a
627  * wrong version, or device doesn't support the requested state.
628  * 0 if device already is in the requested state.
629  * 0 if device's power state has been successfully changed.
630  */
631 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
632 {
633 	u16 pmcsr;
634 	bool need_restore = false;
635 
636 	/* Check if we're already there */
637 	if (dev->current_state == state)
638 		return 0;
639 
640 	if (!dev->pm_cap)
641 		return -EIO;
642 
643 	if (state < PCI_D0 || state > PCI_D3hot)
644 		return -EINVAL;
645 
646 	/* Validate current state:
647 	 * Can enter D0 from any state, but if we can only go deeper
648 	 * to sleep if we're already in a low power state
649 	 */
650 	if (state != PCI_D0 && dev->current_state <= PCI_D3cold
651 	    && dev->current_state > state) {
652 		dev_err(&dev->dev, "invalid power transition (from state %d to %d)\n",
653 			dev->current_state, state);
654 		return -EINVAL;
655 	}
656 
657 	/* check if this device supports the desired state */
658 	if ((state == PCI_D1 && !dev->d1_support)
659 	   || (state == PCI_D2 && !dev->d2_support))
660 		return -EIO;
661 
662 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
663 
664 	/* If we're (effectively) in D3, force entire word to 0.
665 	 * This doesn't affect PME_Status, disables PME_En, and
666 	 * sets PowerState to 0.
667 	 */
668 	switch (dev->current_state) {
669 	case PCI_D0:
670 	case PCI_D1:
671 	case PCI_D2:
672 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
673 		pmcsr |= state;
674 		break;
675 	case PCI_D3hot:
676 	case PCI_D3cold:
677 	case PCI_UNKNOWN: /* Boot-up */
678 		if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
679 		 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
680 			need_restore = true;
681 		/* Fall-through: force to D0 */
682 	default:
683 		pmcsr = 0;
684 		break;
685 	}
686 
687 	/* enter specified state */
688 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
689 
690 	/* Mandatory power management transition delays */
691 	/* see PCI PM 1.1 5.6.1 table 18 */
692 	if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
693 		pci_dev_d3_sleep(dev);
694 	else if (state == PCI_D2 || dev->current_state == PCI_D2)
695 		udelay(PCI_PM_D2_DELAY);
696 
697 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
698 	dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
699 	if (dev->current_state != state && printk_ratelimit())
700 		dev_info(&dev->dev, "Refused to change power state, currently in D%d\n",
701 			 dev->current_state);
702 
703 	/*
704 	 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
705 	 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
706 	 * from D3hot to D0 _may_ perform an internal reset, thereby
707 	 * going to "D0 Uninitialized" rather than "D0 Initialized".
708 	 * For example, at least some versions of the 3c905B and the
709 	 * 3c556B exhibit this behaviour.
710 	 *
711 	 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
712 	 * devices in a D3hot state at boot.  Consequently, we need to
713 	 * restore at least the BARs so that the device will be
714 	 * accessible to its driver.
715 	 */
716 	if (need_restore)
717 		pci_restore_bars(dev);
718 
719 	if (dev->bus->self)
720 		pcie_aspm_pm_state_change(dev->bus->self);
721 
722 	return 0;
723 }
724 
725 /**
726  * pci_update_current_state - Read power state of given device and cache it
727  * @dev: PCI device to handle.
728  * @state: State to cache in case the device doesn't have the PM capability
729  *
730  * The power state is read from the PMCSR register, which however is
731  * inaccessible in D3cold.  The platform firmware is therefore queried first
732  * to detect accessibility of the register.  In case the platform firmware
733  * reports an incorrect state or the device isn't power manageable by the
734  * platform at all, we try to detect D3cold by testing accessibility of the
735  * vendor ID in config space.
736  */
737 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
738 {
739 	if (platform_pci_get_power_state(dev) == PCI_D3cold ||
740 	    !pci_device_is_present(dev)) {
741 		dev->current_state = PCI_D3cold;
742 	} else if (dev->pm_cap) {
743 		u16 pmcsr;
744 
745 		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
746 		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
747 	} else {
748 		dev->current_state = state;
749 	}
750 }
751 
752 /**
753  * pci_power_up - Put the given device into D0 forcibly
754  * @dev: PCI device to power up
755  */
756 void pci_power_up(struct pci_dev *dev)
757 {
758 	if (platform_pci_power_manageable(dev))
759 		platform_pci_set_power_state(dev, PCI_D0);
760 
761 	pci_raw_set_power_state(dev, PCI_D0);
762 	pci_update_current_state(dev, PCI_D0);
763 }
764 
765 /**
766  * pci_platform_power_transition - Use platform to change device power state
767  * @dev: PCI device to handle.
768  * @state: State to put the device into.
769  */
770 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
771 {
772 	int error;
773 
774 	if (platform_pci_power_manageable(dev)) {
775 		error = platform_pci_set_power_state(dev, state);
776 		if (!error)
777 			pci_update_current_state(dev, state);
778 	} else
779 		error = -ENODEV;
780 
781 	if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
782 		dev->current_state = PCI_D0;
783 
784 	return error;
785 }
786 
787 /**
788  * pci_wakeup - Wake up a PCI device
789  * @pci_dev: Device to handle.
790  * @ign: ignored parameter
791  */
792 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
793 {
794 	pci_wakeup_event(pci_dev);
795 	pm_request_resume(&pci_dev->dev);
796 	return 0;
797 }
798 
799 /**
800  * pci_wakeup_bus - Walk given bus and wake up devices on it
801  * @bus: Top bus of the subtree to walk.
802  */
803 static void pci_wakeup_bus(struct pci_bus *bus)
804 {
805 	if (bus)
806 		pci_walk_bus(bus, pci_wakeup, NULL);
807 }
808 
809 /**
810  * __pci_start_power_transition - Start power transition of a PCI device
811  * @dev: PCI device to handle.
812  * @state: State to put the device into.
813  */
814 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
815 {
816 	if (state == PCI_D0) {
817 		pci_platform_power_transition(dev, PCI_D0);
818 		/*
819 		 * Mandatory power management transition delays, see
820 		 * PCI Express Base Specification Revision 2.0 Section
821 		 * 6.6.1: Conventional Reset.  Do not delay for
822 		 * devices powered on/off by corresponding bridge,
823 		 * because have already delayed for the bridge.
824 		 */
825 		if (dev->runtime_d3cold) {
826 			if (dev->d3cold_delay)
827 				msleep(dev->d3cold_delay);
828 			/*
829 			 * When powering on a bridge from D3cold, the
830 			 * whole hierarchy may be powered on into
831 			 * D0uninitialized state, resume them to give
832 			 * them a chance to suspend again
833 			 */
834 			pci_wakeup_bus(dev->subordinate);
835 		}
836 	}
837 }
838 
839 /**
840  * __pci_dev_set_current_state - Set current state of a PCI device
841  * @dev: Device to handle
842  * @data: pointer to state to be set
843  */
844 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
845 {
846 	pci_power_t state = *(pci_power_t *)data;
847 
848 	dev->current_state = state;
849 	return 0;
850 }
851 
852 /**
853  * __pci_bus_set_current_state - Walk given bus and set current state of devices
854  * @bus: Top bus of the subtree to walk.
855  * @state: state to be set
856  */
857 static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
858 {
859 	if (bus)
860 		pci_walk_bus(bus, __pci_dev_set_current_state, &state);
861 }
862 
863 /**
864  * __pci_complete_power_transition - Complete power transition of a PCI device
865  * @dev: PCI device to handle.
866  * @state: State to put the device into.
867  *
868  * This function should not be called directly by device drivers.
869  */
870 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
871 {
872 	int ret;
873 
874 	if (state <= PCI_D0)
875 		return -EINVAL;
876 	ret = pci_platform_power_transition(dev, state);
877 	/* Power off the bridge may power off the whole hierarchy */
878 	if (!ret && state == PCI_D3cold)
879 		__pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
880 	return ret;
881 }
882 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
883 
884 /**
885  * pci_set_power_state - Set the power state of a PCI device
886  * @dev: PCI device to handle.
887  * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
888  *
889  * Transition a device to a new power state, using the platform firmware and/or
890  * the device's PCI PM registers.
891  *
892  * RETURN VALUE:
893  * -EINVAL if the requested state is invalid.
894  * -EIO if device does not support PCI PM or its PM capabilities register has a
895  * wrong version, or device doesn't support the requested state.
896  * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
897  * 0 if device already is in the requested state.
898  * 0 if the transition is to D3 but D3 is not supported.
899  * 0 if device's power state has been successfully changed.
900  */
901 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
902 {
903 	int error;
904 
905 	/* bound the state we're entering */
906 	if (state > PCI_D3cold)
907 		state = PCI_D3cold;
908 	else if (state < PCI_D0)
909 		state = PCI_D0;
910 	else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
911 		/*
912 		 * If the device or the parent bridge do not support PCI PM,
913 		 * ignore the request if we're doing anything other than putting
914 		 * it into D0 (which would only happen on boot).
915 		 */
916 		return 0;
917 
918 	/* Check if we're already there */
919 	if (dev->current_state == state)
920 		return 0;
921 
922 	__pci_start_power_transition(dev, state);
923 
924 	/* This device is quirked not to be put into D3, so
925 	   don't put it in D3 */
926 	if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
927 		return 0;
928 
929 	/*
930 	 * To put device in D3cold, we put device into D3hot in native
931 	 * way, then put device into D3cold with platform ops
932 	 */
933 	error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
934 					PCI_D3hot : state);
935 
936 	if (!__pci_complete_power_transition(dev, state))
937 		error = 0;
938 
939 	return error;
940 }
941 EXPORT_SYMBOL(pci_set_power_state);
942 
943 /**
944  * pci_choose_state - Choose the power state of a PCI device
945  * @dev: PCI device to be suspended
946  * @state: target sleep state for the whole system. This is the value
947  *	that is passed to suspend() function.
948  *
949  * Returns PCI power state suitable for given device and given system
950  * message.
951  */
952 
953 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
954 {
955 	pci_power_t ret;
956 
957 	if (!dev->pm_cap)
958 		return PCI_D0;
959 
960 	ret = platform_pci_choose_state(dev);
961 	if (ret != PCI_POWER_ERROR)
962 		return ret;
963 
964 	switch (state.event) {
965 	case PM_EVENT_ON:
966 		return PCI_D0;
967 	case PM_EVENT_FREEZE:
968 	case PM_EVENT_PRETHAW:
969 		/* REVISIT both freeze and pre-thaw "should" use D0 */
970 	case PM_EVENT_SUSPEND:
971 	case PM_EVENT_HIBERNATE:
972 		return PCI_D3hot;
973 	default:
974 		dev_info(&dev->dev, "unrecognized suspend event %d\n",
975 			 state.event);
976 		BUG();
977 	}
978 	return PCI_D0;
979 }
980 EXPORT_SYMBOL(pci_choose_state);
981 
982 #define PCI_EXP_SAVE_REGS	7
983 
984 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
985 						       u16 cap, bool extended)
986 {
987 	struct pci_cap_saved_state *tmp;
988 
989 	hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
990 		if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
991 			return tmp;
992 	}
993 	return NULL;
994 }
995 
996 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
997 {
998 	return _pci_find_saved_cap(dev, cap, false);
999 }
1000 
1001 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1002 {
1003 	return _pci_find_saved_cap(dev, cap, true);
1004 }
1005 
1006 static int pci_save_pcie_state(struct pci_dev *dev)
1007 {
1008 	int i = 0;
1009 	struct pci_cap_saved_state *save_state;
1010 	u16 *cap;
1011 
1012 	if (!pci_is_pcie(dev))
1013 		return 0;
1014 
1015 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1016 	if (!save_state) {
1017 		dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1018 		return -ENOMEM;
1019 	}
1020 
1021 	cap = (u16 *)&save_state->cap.data[0];
1022 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1023 	pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1024 	pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1025 	pcie_capability_read_word(dev, PCI_EXP_RTCTL,  &cap[i++]);
1026 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1027 	pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1028 	pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1029 
1030 	return 0;
1031 }
1032 
1033 static void pci_restore_pcie_state(struct pci_dev *dev)
1034 {
1035 	int i = 0;
1036 	struct pci_cap_saved_state *save_state;
1037 	u16 *cap;
1038 
1039 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1040 	if (!save_state)
1041 		return;
1042 
1043 	cap = (u16 *)&save_state->cap.data[0];
1044 	pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1045 	pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1046 	pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1047 	pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1048 	pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1049 	pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1050 	pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1051 }
1052 
1053 
1054 static int pci_save_pcix_state(struct pci_dev *dev)
1055 {
1056 	int pos;
1057 	struct pci_cap_saved_state *save_state;
1058 
1059 	pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1060 	if (!pos)
1061 		return 0;
1062 
1063 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1064 	if (!save_state) {
1065 		dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1066 		return -ENOMEM;
1067 	}
1068 
1069 	pci_read_config_word(dev, pos + PCI_X_CMD,
1070 			     (u16 *)save_state->cap.data);
1071 
1072 	return 0;
1073 }
1074 
1075 static void pci_restore_pcix_state(struct pci_dev *dev)
1076 {
1077 	int i = 0, pos;
1078 	struct pci_cap_saved_state *save_state;
1079 	u16 *cap;
1080 
1081 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1082 	pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1083 	if (!save_state || !pos)
1084 		return;
1085 	cap = (u16 *)&save_state->cap.data[0];
1086 
1087 	pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1088 }
1089 
1090 
1091 /**
1092  * pci_save_state - save the PCI configuration space of a device before suspending
1093  * @dev: - PCI device that we're dealing with
1094  */
1095 int pci_save_state(struct pci_dev *dev)
1096 {
1097 	int i;
1098 	/* XXX: 100% dword access ok here? */
1099 	for (i = 0; i < 16; i++)
1100 		pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1101 	dev->state_saved = true;
1102 
1103 	i = pci_save_pcie_state(dev);
1104 	if (i != 0)
1105 		return i;
1106 
1107 	i = pci_save_pcix_state(dev);
1108 	if (i != 0)
1109 		return i;
1110 
1111 	return pci_save_vc_state(dev);
1112 }
1113 EXPORT_SYMBOL(pci_save_state);
1114 
1115 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1116 				     u32 saved_val, int retry)
1117 {
1118 	u32 val;
1119 
1120 	pci_read_config_dword(pdev, offset, &val);
1121 	if (val == saved_val)
1122 		return;
1123 
1124 	for (;;) {
1125 		dev_dbg(&pdev->dev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1126 			offset, val, saved_val);
1127 		pci_write_config_dword(pdev, offset, saved_val);
1128 		if (retry-- <= 0)
1129 			return;
1130 
1131 		pci_read_config_dword(pdev, offset, &val);
1132 		if (val == saved_val)
1133 			return;
1134 
1135 		mdelay(1);
1136 	}
1137 }
1138 
1139 static void pci_restore_config_space_range(struct pci_dev *pdev,
1140 					   int start, int end, int retry)
1141 {
1142 	int index;
1143 
1144 	for (index = end; index >= start; index--)
1145 		pci_restore_config_dword(pdev, 4 * index,
1146 					 pdev->saved_config_space[index],
1147 					 retry);
1148 }
1149 
1150 static void pci_restore_config_space(struct pci_dev *pdev)
1151 {
1152 	if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1153 		pci_restore_config_space_range(pdev, 10, 15, 0);
1154 		/* Restore BARs before the command register. */
1155 		pci_restore_config_space_range(pdev, 4, 9, 10);
1156 		pci_restore_config_space_range(pdev, 0, 3, 0);
1157 	} else {
1158 		pci_restore_config_space_range(pdev, 0, 15, 0);
1159 	}
1160 }
1161 
1162 /**
1163  * pci_restore_state - Restore the saved state of a PCI device
1164  * @dev: - PCI device that we're dealing with
1165  */
1166 void pci_restore_state(struct pci_dev *dev)
1167 {
1168 	if (!dev->state_saved)
1169 		return;
1170 
1171 	/* PCI Express register must be restored first */
1172 	pci_restore_pcie_state(dev);
1173 	pci_restore_pasid_state(dev);
1174 	pci_restore_pri_state(dev);
1175 	pci_restore_ats_state(dev);
1176 	pci_restore_vc_state(dev);
1177 
1178 	pci_cleanup_aer_error_status_regs(dev);
1179 
1180 	pci_restore_config_space(dev);
1181 
1182 	pci_restore_pcix_state(dev);
1183 	pci_restore_msi_state(dev);
1184 
1185 	/* Restore ACS and IOV configuration state */
1186 	pci_enable_acs(dev);
1187 	pci_restore_iov_state(dev);
1188 
1189 	dev->state_saved = false;
1190 }
1191 EXPORT_SYMBOL(pci_restore_state);
1192 
1193 struct pci_saved_state {
1194 	u32 config_space[16];
1195 	struct pci_cap_saved_data cap[0];
1196 };
1197 
1198 /**
1199  * pci_store_saved_state - Allocate and return an opaque struct containing
1200  *			   the device saved state.
1201  * @dev: PCI device that we're dealing with
1202  *
1203  * Return NULL if no state or error.
1204  */
1205 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1206 {
1207 	struct pci_saved_state *state;
1208 	struct pci_cap_saved_state *tmp;
1209 	struct pci_cap_saved_data *cap;
1210 	size_t size;
1211 
1212 	if (!dev->state_saved)
1213 		return NULL;
1214 
1215 	size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1216 
1217 	hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1218 		size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1219 
1220 	state = kzalloc(size, GFP_KERNEL);
1221 	if (!state)
1222 		return NULL;
1223 
1224 	memcpy(state->config_space, dev->saved_config_space,
1225 	       sizeof(state->config_space));
1226 
1227 	cap = state->cap;
1228 	hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1229 		size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1230 		memcpy(cap, &tmp->cap, len);
1231 		cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1232 	}
1233 	/* Empty cap_save terminates list */
1234 
1235 	return state;
1236 }
1237 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1238 
1239 /**
1240  * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1241  * @dev: PCI device that we're dealing with
1242  * @state: Saved state returned from pci_store_saved_state()
1243  */
1244 int pci_load_saved_state(struct pci_dev *dev,
1245 			 struct pci_saved_state *state)
1246 {
1247 	struct pci_cap_saved_data *cap;
1248 
1249 	dev->state_saved = false;
1250 
1251 	if (!state)
1252 		return 0;
1253 
1254 	memcpy(dev->saved_config_space, state->config_space,
1255 	       sizeof(state->config_space));
1256 
1257 	cap = state->cap;
1258 	while (cap->size) {
1259 		struct pci_cap_saved_state *tmp;
1260 
1261 		tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1262 		if (!tmp || tmp->cap.size != cap->size)
1263 			return -EINVAL;
1264 
1265 		memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1266 		cap = (struct pci_cap_saved_data *)((u8 *)cap +
1267 		       sizeof(struct pci_cap_saved_data) + cap->size);
1268 	}
1269 
1270 	dev->state_saved = true;
1271 	return 0;
1272 }
1273 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1274 
1275 /**
1276  * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1277  *				   and free the memory allocated for it.
1278  * @dev: PCI device that we're dealing with
1279  * @state: Pointer to saved state returned from pci_store_saved_state()
1280  */
1281 int pci_load_and_free_saved_state(struct pci_dev *dev,
1282 				  struct pci_saved_state **state)
1283 {
1284 	int ret = pci_load_saved_state(dev, *state);
1285 	kfree(*state);
1286 	*state = NULL;
1287 	return ret;
1288 }
1289 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1290 
1291 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1292 {
1293 	return pci_enable_resources(dev, bars);
1294 }
1295 
1296 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1297 {
1298 	int err;
1299 	struct pci_dev *bridge;
1300 	u16 cmd;
1301 	u8 pin;
1302 
1303 	err = pci_set_power_state(dev, PCI_D0);
1304 	if (err < 0 && err != -EIO)
1305 		return err;
1306 
1307 	bridge = pci_upstream_bridge(dev);
1308 	if (bridge)
1309 		pcie_aspm_powersave_config_link(bridge);
1310 
1311 	err = pcibios_enable_device(dev, bars);
1312 	if (err < 0)
1313 		return err;
1314 	pci_fixup_device(pci_fixup_enable, dev);
1315 
1316 	if (dev->msi_enabled || dev->msix_enabled)
1317 		return 0;
1318 
1319 	pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1320 	if (pin) {
1321 		pci_read_config_word(dev, PCI_COMMAND, &cmd);
1322 		if (cmd & PCI_COMMAND_INTX_DISABLE)
1323 			pci_write_config_word(dev, PCI_COMMAND,
1324 					      cmd & ~PCI_COMMAND_INTX_DISABLE);
1325 	}
1326 
1327 	return 0;
1328 }
1329 
1330 /**
1331  * pci_reenable_device - Resume abandoned device
1332  * @dev: PCI device to be resumed
1333  *
1334  *  Note this function is a backend of pci_default_resume and is not supposed
1335  *  to be called by normal code, write proper resume handler and use it instead.
1336  */
1337 int pci_reenable_device(struct pci_dev *dev)
1338 {
1339 	if (pci_is_enabled(dev))
1340 		return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1341 	return 0;
1342 }
1343 EXPORT_SYMBOL(pci_reenable_device);
1344 
1345 static void pci_enable_bridge(struct pci_dev *dev)
1346 {
1347 	struct pci_dev *bridge;
1348 	int retval;
1349 
1350 	bridge = pci_upstream_bridge(dev);
1351 	if (bridge)
1352 		pci_enable_bridge(bridge);
1353 
1354 	/*
1355 	 * Hold pci_bridge_mutex to prevent a race when enabling two
1356 	 * devices below the bridge simultaneously.  The race may cause a
1357 	 * PCI_COMMAND_MEMORY update to be lost (see changelog).
1358 	 */
1359 	mutex_lock(&pci_bridge_mutex);
1360 	if (pci_is_enabled(dev)) {
1361 		if (!dev->is_busmaster)
1362 			pci_set_master(dev);
1363 		goto end;
1364 	}
1365 
1366 	retval = pci_enable_device(dev);
1367 	if (retval)
1368 		dev_err(&dev->dev, "Error enabling bridge (%d), continuing\n",
1369 			retval);
1370 	pci_set_master(dev);
1371 end:
1372 	mutex_unlock(&pci_bridge_mutex);
1373 }
1374 
1375 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1376 {
1377 	struct pci_dev *bridge;
1378 	int err;
1379 	int i, bars = 0;
1380 
1381 	/*
1382 	 * Power state could be unknown at this point, either due to a fresh
1383 	 * boot or a device removal call.  So get the current power state
1384 	 * so that things like MSI message writing will behave as expected
1385 	 * (e.g. if the device really is in D0 at enable time).
1386 	 */
1387 	if (dev->pm_cap) {
1388 		u16 pmcsr;
1389 		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1390 		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1391 	}
1392 
1393 	if (atomic_inc_return(&dev->enable_cnt) > 1)
1394 		return 0;		/* already enabled */
1395 
1396 	bridge = pci_upstream_bridge(dev);
1397 	if (bridge && !pci_is_enabled(bridge))
1398 		pci_enable_bridge(bridge);
1399 
1400 	/* only skip sriov related */
1401 	for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1402 		if (dev->resource[i].flags & flags)
1403 			bars |= (1 << i);
1404 	for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1405 		if (dev->resource[i].flags & flags)
1406 			bars |= (1 << i);
1407 
1408 	err = do_pci_enable_device(dev, bars);
1409 	if (err < 0)
1410 		atomic_dec(&dev->enable_cnt);
1411 	return err;
1412 }
1413 
1414 /**
1415  * pci_enable_device_io - Initialize a device for use with IO space
1416  * @dev: PCI device to be initialized
1417  *
1418  *  Initialize device before it's used by a driver. Ask low-level code
1419  *  to enable I/O resources. Wake up the device if it was suspended.
1420  *  Beware, this function can fail.
1421  */
1422 int pci_enable_device_io(struct pci_dev *dev)
1423 {
1424 	return pci_enable_device_flags(dev, IORESOURCE_IO);
1425 }
1426 EXPORT_SYMBOL(pci_enable_device_io);
1427 
1428 /**
1429  * pci_enable_device_mem - Initialize a device for use with Memory space
1430  * @dev: PCI device to be initialized
1431  *
1432  *  Initialize device before it's used by a driver. Ask low-level code
1433  *  to enable Memory resources. Wake up the device if it was suspended.
1434  *  Beware, this function can fail.
1435  */
1436 int pci_enable_device_mem(struct pci_dev *dev)
1437 {
1438 	return pci_enable_device_flags(dev, IORESOURCE_MEM);
1439 }
1440 EXPORT_SYMBOL(pci_enable_device_mem);
1441 
1442 /**
1443  * pci_enable_device - Initialize device before it's used by a driver.
1444  * @dev: PCI device to be initialized
1445  *
1446  *  Initialize device before it's used by a driver. Ask low-level code
1447  *  to enable I/O and memory. Wake up the device if it was suspended.
1448  *  Beware, this function can fail.
1449  *
1450  *  Note we don't actually enable the device many times if we call
1451  *  this function repeatedly (we just increment the count).
1452  */
1453 int pci_enable_device(struct pci_dev *dev)
1454 {
1455 	return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1456 }
1457 EXPORT_SYMBOL(pci_enable_device);
1458 
1459 /*
1460  * Managed PCI resources.  This manages device on/off, intx/msi/msix
1461  * on/off and BAR regions.  pci_dev itself records msi/msix status, so
1462  * there's no need to track it separately.  pci_devres is initialized
1463  * when a device is enabled using managed PCI device enable interface.
1464  */
1465 struct pci_devres {
1466 	unsigned int enabled:1;
1467 	unsigned int pinned:1;
1468 	unsigned int orig_intx:1;
1469 	unsigned int restore_intx:1;
1470 	u32 region_mask;
1471 };
1472 
1473 static void pcim_release(struct device *gendev, void *res)
1474 {
1475 	struct pci_dev *dev = to_pci_dev(gendev);
1476 	struct pci_devres *this = res;
1477 	int i;
1478 
1479 	if (dev->msi_enabled)
1480 		pci_disable_msi(dev);
1481 	if (dev->msix_enabled)
1482 		pci_disable_msix(dev);
1483 
1484 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1485 		if (this->region_mask & (1 << i))
1486 			pci_release_region(dev, i);
1487 
1488 	if (this->restore_intx)
1489 		pci_intx(dev, this->orig_intx);
1490 
1491 	if (this->enabled && !this->pinned)
1492 		pci_disable_device(dev);
1493 }
1494 
1495 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1496 {
1497 	struct pci_devres *dr, *new_dr;
1498 
1499 	dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1500 	if (dr)
1501 		return dr;
1502 
1503 	new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1504 	if (!new_dr)
1505 		return NULL;
1506 	return devres_get(&pdev->dev, new_dr, NULL, NULL);
1507 }
1508 
1509 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1510 {
1511 	if (pci_is_managed(pdev))
1512 		return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1513 	return NULL;
1514 }
1515 
1516 /**
1517  * pcim_enable_device - Managed pci_enable_device()
1518  * @pdev: PCI device to be initialized
1519  *
1520  * Managed pci_enable_device().
1521  */
1522 int pcim_enable_device(struct pci_dev *pdev)
1523 {
1524 	struct pci_devres *dr;
1525 	int rc;
1526 
1527 	dr = get_pci_dr(pdev);
1528 	if (unlikely(!dr))
1529 		return -ENOMEM;
1530 	if (dr->enabled)
1531 		return 0;
1532 
1533 	rc = pci_enable_device(pdev);
1534 	if (!rc) {
1535 		pdev->is_managed = 1;
1536 		dr->enabled = 1;
1537 	}
1538 	return rc;
1539 }
1540 EXPORT_SYMBOL(pcim_enable_device);
1541 
1542 /**
1543  * pcim_pin_device - Pin managed PCI device
1544  * @pdev: PCI device to pin
1545  *
1546  * Pin managed PCI device @pdev.  Pinned device won't be disabled on
1547  * driver detach.  @pdev must have been enabled with
1548  * pcim_enable_device().
1549  */
1550 void pcim_pin_device(struct pci_dev *pdev)
1551 {
1552 	struct pci_devres *dr;
1553 
1554 	dr = find_pci_dr(pdev);
1555 	WARN_ON(!dr || !dr->enabled);
1556 	if (dr)
1557 		dr->pinned = 1;
1558 }
1559 EXPORT_SYMBOL(pcim_pin_device);
1560 
1561 /*
1562  * pcibios_add_device - provide arch specific hooks when adding device dev
1563  * @dev: the PCI device being added
1564  *
1565  * Permits the platform to provide architecture specific functionality when
1566  * devices are added. This is the default implementation. Architecture
1567  * implementations can override this.
1568  */
1569 int __weak pcibios_add_device(struct pci_dev *dev)
1570 {
1571 	return 0;
1572 }
1573 
1574 /**
1575  * pcibios_release_device - provide arch specific hooks when releasing device dev
1576  * @dev: the PCI device being released
1577  *
1578  * Permits the platform to provide architecture specific functionality when
1579  * devices are released. This is the default implementation. Architecture
1580  * implementations can override this.
1581  */
1582 void __weak pcibios_release_device(struct pci_dev *dev) {}
1583 
1584 /**
1585  * pcibios_disable_device - disable arch specific PCI resources for device dev
1586  * @dev: the PCI device to disable
1587  *
1588  * Disables architecture specific PCI resources for the device. This
1589  * is the default implementation. Architecture implementations can
1590  * override this.
1591  */
1592 void __weak pcibios_disable_device(struct pci_dev *dev) {}
1593 
1594 /**
1595  * pcibios_penalize_isa_irq - penalize an ISA IRQ
1596  * @irq: ISA IRQ to penalize
1597  * @active: IRQ active or not
1598  *
1599  * Permits the platform to provide architecture-specific functionality when
1600  * penalizing ISA IRQs. This is the default implementation. Architecture
1601  * implementations can override this.
1602  */
1603 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1604 
1605 static void do_pci_disable_device(struct pci_dev *dev)
1606 {
1607 	u16 pci_command;
1608 
1609 	pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1610 	if (pci_command & PCI_COMMAND_MASTER) {
1611 		pci_command &= ~PCI_COMMAND_MASTER;
1612 		pci_write_config_word(dev, PCI_COMMAND, pci_command);
1613 	}
1614 
1615 	pcibios_disable_device(dev);
1616 }
1617 
1618 /**
1619  * pci_disable_enabled_device - Disable device without updating enable_cnt
1620  * @dev: PCI device to disable
1621  *
1622  * NOTE: This function is a backend of PCI power management routines and is
1623  * not supposed to be called drivers.
1624  */
1625 void pci_disable_enabled_device(struct pci_dev *dev)
1626 {
1627 	if (pci_is_enabled(dev))
1628 		do_pci_disable_device(dev);
1629 }
1630 
1631 /**
1632  * pci_disable_device - Disable PCI device after use
1633  * @dev: PCI device to be disabled
1634  *
1635  * Signal to the system that the PCI device is not in use by the system
1636  * anymore.  This only involves disabling PCI bus-mastering, if active.
1637  *
1638  * Note we don't actually disable the device until all callers of
1639  * pci_enable_device() have called pci_disable_device().
1640  */
1641 void pci_disable_device(struct pci_dev *dev)
1642 {
1643 	struct pci_devres *dr;
1644 
1645 	dr = find_pci_dr(dev);
1646 	if (dr)
1647 		dr->enabled = 0;
1648 
1649 	dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1650 		      "disabling already-disabled device");
1651 
1652 	if (atomic_dec_return(&dev->enable_cnt) != 0)
1653 		return;
1654 
1655 	do_pci_disable_device(dev);
1656 
1657 	dev->is_busmaster = 0;
1658 }
1659 EXPORT_SYMBOL(pci_disable_device);
1660 
1661 /**
1662  * pcibios_set_pcie_reset_state - set reset state for device dev
1663  * @dev: the PCIe device reset
1664  * @state: Reset state to enter into
1665  *
1666  *
1667  * Sets the PCIe reset state for the device. This is the default
1668  * implementation. Architecture implementations can override this.
1669  */
1670 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1671 					enum pcie_reset_state state)
1672 {
1673 	return -EINVAL;
1674 }
1675 
1676 /**
1677  * pci_set_pcie_reset_state - set reset state for device dev
1678  * @dev: the PCIe device reset
1679  * @state: Reset state to enter into
1680  *
1681  *
1682  * Sets the PCI reset state for the device.
1683  */
1684 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1685 {
1686 	return pcibios_set_pcie_reset_state(dev, state);
1687 }
1688 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1689 
1690 /**
1691  * pci_check_pme_status - Check if given device has generated PME.
1692  * @dev: Device to check.
1693  *
1694  * Check the PME status of the device and if set, clear it and clear PME enable
1695  * (if set).  Return 'true' if PME status and PME enable were both set or
1696  * 'false' otherwise.
1697  */
1698 bool pci_check_pme_status(struct pci_dev *dev)
1699 {
1700 	int pmcsr_pos;
1701 	u16 pmcsr;
1702 	bool ret = false;
1703 
1704 	if (!dev->pm_cap)
1705 		return false;
1706 
1707 	pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1708 	pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1709 	if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1710 		return false;
1711 
1712 	/* Clear PME status. */
1713 	pmcsr |= PCI_PM_CTRL_PME_STATUS;
1714 	if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1715 		/* Disable PME to avoid interrupt flood. */
1716 		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1717 		ret = true;
1718 	}
1719 
1720 	pci_write_config_word(dev, pmcsr_pos, pmcsr);
1721 
1722 	return ret;
1723 }
1724 
1725 /**
1726  * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1727  * @dev: Device to handle.
1728  * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1729  *
1730  * Check if @dev has generated PME and queue a resume request for it in that
1731  * case.
1732  */
1733 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1734 {
1735 	if (pme_poll_reset && dev->pme_poll)
1736 		dev->pme_poll = false;
1737 
1738 	if (pci_check_pme_status(dev)) {
1739 		pci_wakeup_event(dev);
1740 		pm_request_resume(&dev->dev);
1741 	}
1742 	return 0;
1743 }
1744 
1745 /**
1746  * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1747  * @bus: Top bus of the subtree to walk.
1748  */
1749 void pci_pme_wakeup_bus(struct pci_bus *bus)
1750 {
1751 	if (bus)
1752 		pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1753 }
1754 
1755 
1756 /**
1757  * pci_pme_capable - check the capability of PCI device to generate PME#
1758  * @dev: PCI device to handle.
1759  * @state: PCI state from which device will issue PME#.
1760  */
1761 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1762 {
1763 	if (!dev->pm_cap)
1764 		return false;
1765 
1766 	return !!(dev->pme_support & (1 << state));
1767 }
1768 EXPORT_SYMBOL(pci_pme_capable);
1769 
1770 static void pci_pme_list_scan(struct work_struct *work)
1771 {
1772 	struct pci_pme_device *pme_dev, *n;
1773 
1774 	mutex_lock(&pci_pme_list_mutex);
1775 	list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1776 		if (pme_dev->dev->pme_poll) {
1777 			struct pci_dev *bridge;
1778 
1779 			bridge = pme_dev->dev->bus->self;
1780 			/*
1781 			 * If bridge is in low power state, the
1782 			 * configuration space of subordinate devices
1783 			 * may be not accessible
1784 			 */
1785 			if (bridge && bridge->current_state != PCI_D0)
1786 				continue;
1787 			pci_pme_wakeup(pme_dev->dev, NULL);
1788 		} else {
1789 			list_del(&pme_dev->list);
1790 			kfree(pme_dev);
1791 		}
1792 	}
1793 	if (!list_empty(&pci_pme_list))
1794 		queue_delayed_work(system_freezable_wq, &pci_pme_work,
1795 				   msecs_to_jiffies(PME_TIMEOUT));
1796 	mutex_unlock(&pci_pme_list_mutex);
1797 }
1798 
1799 static void __pci_pme_active(struct pci_dev *dev, bool enable)
1800 {
1801 	u16 pmcsr;
1802 
1803 	if (!dev->pme_support)
1804 		return;
1805 
1806 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1807 	/* Clear PME_Status by writing 1 to it and enable PME# */
1808 	pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1809 	if (!enable)
1810 		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1811 
1812 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1813 }
1814 
1815 /**
1816  * pci_pme_restore - Restore PME configuration after config space restore.
1817  * @dev: PCI device to update.
1818  */
1819 void pci_pme_restore(struct pci_dev *dev)
1820 {
1821 	u16 pmcsr;
1822 
1823 	if (!dev->pme_support)
1824 		return;
1825 
1826 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1827 	if (dev->wakeup_prepared) {
1828 		pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1829 		pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
1830 	} else {
1831 		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1832 		pmcsr |= PCI_PM_CTRL_PME_STATUS;
1833 	}
1834 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1835 }
1836 
1837 /**
1838  * pci_pme_active - enable or disable PCI device's PME# function
1839  * @dev: PCI device to handle.
1840  * @enable: 'true' to enable PME# generation; 'false' to disable it.
1841  *
1842  * The caller must verify that the device is capable of generating PME# before
1843  * calling this function with @enable equal to 'true'.
1844  */
1845 void pci_pme_active(struct pci_dev *dev, bool enable)
1846 {
1847 	__pci_pme_active(dev, enable);
1848 
1849 	/*
1850 	 * PCI (as opposed to PCIe) PME requires that the device have
1851 	 * its PME# line hooked up correctly. Not all hardware vendors
1852 	 * do this, so the PME never gets delivered and the device
1853 	 * remains asleep. The easiest way around this is to
1854 	 * periodically walk the list of suspended devices and check
1855 	 * whether any have their PME flag set. The assumption is that
1856 	 * we'll wake up often enough anyway that this won't be a huge
1857 	 * hit, and the power savings from the devices will still be a
1858 	 * win.
1859 	 *
1860 	 * Although PCIe uses in-band PME message instead of PME# line
1861 	 * to report PME, PME does not work for some PCIe devices in
1862 	 * reality.  For example, there are devices that set their PME
1863 	 * status bits, but don't really bother to send a PME message;
1864 	 * there are PCI Express Root Ports that don't bother to
1865 	 * trigger interrupts when they receive PME messages from the
1866 	 * devices below.  So PME poll is used for PCIe devices too.
1867 	 */
1868 
1869 	if (dev->pme_poll) {
1870 		struct pci_pme_device *pme_dev;
1871 		if (enable) {
1872 			pme_dev = kmalloc(sizeof(struct pci_pme_device),
1873 					  GFP_KERNEL);
1874 			if (!pme_dev) {
1875 				dev_warn(&dev->dev, "can't enable PME#\n");
1876 				return;
1877 			}
1878 			pme_dev->dev = dev;
1879 			mutex_lock(&pci_pme_list_mutex);
1880 			list_add(&pme_dev->list, &pci_pme_list);
1881 			if (list_is_singular(&pci_pme_list))
1882 				queue_delayed_work(system_freezable_wq,
1883 						   &pci_pme_work,
1884 						   msecs_to_jiffies(PME_TIMEOUT));
1885 			mutex_unlock(&pci_pme_list_mutex);
1886 		} else {
1887 			mutex_lock(&pci_pme_list_mutex);
1888 			list_for_each_entry(pme_dev, &pci_pme_list, list) {
1889 				if (pme_dev->dev == dev) {
1890 					list_del(&pme_dev->list);
1891 					kfree(pme_dev);
1892 					break;
1893 				}
1894 			}
1895 			mutex_unlock(&pci_pme_list_mutex);
1896 		}
1897 	}
1898 
1899 	dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1900 }
1901 EXPORT_SYMBOL(pci_pme_active);
1902 
1903 /**
1904  * pci_enable_wake - enable PCI device as wakeup event source
1905  * @dev: PCI device affected
1906  * @state: PCI state from which device will issue wakeup events
1907  * @enable: True to enable event generation; false to disable
1908  *
1909  * This enables the device as a wakeup event source, or disables it.
1910  * When such events involves platform-specific hooks, those hooks are
1911  * called automatically by this routine.
1912  *
1913  * Devices with legacy power management (no standard PCI PM capabilities)
1914  * always require such platform hooks.
1915  *
1916  * RETURN VALUE:
1917  * 0 is returned on success
1918  * -EINVAL is returned if device is not supposed to wake up the system
1919  * Error code depending on the platform is returned if both the platform and
1920  * the native mechanism fail to enable the generation of wake-up events
1921  */
1922 int pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
1923 {
1924 	int ret = 0;
1925 
1926 	/*
1927 	 * Bridges can only signal wakeup on behalf of subordinate devices,
1928 	 * but that is set up elsewhere, so skip them.
1929 	 */
1930 	if (pci_has_subordinate(dev))
1931 		return 0;
1932 
1933 	/* Don't do the same thing twice in a row for one device. */
1934 	if (!!enable == !!dev->wakeup_prepared)
1935 		return 0;
1936 
1937 	/*
1938 	 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1939 	 * Anderson we should be doing PME# wake enable followed by ACPI wake
1940 	 * enable.  To disable wake-up we call the platform first, for symmetry.
1941 	 */
1942 
1943 	if (enable) {
1944 		int error;
1945 
1946 		if (pci_pme_capable(dev, state))
1947 			pci_pme_active(dev, true);
1948 		else
1949 			ret = 1;
1950 		error = platform_pci_set_wakeup(dev, true);
1951 		if (ret)
1952 			ret = error;
1953 		if (!ret)
1954 			dev->wakeup_prepared = true;
1955 	} else {
1956 		platform_pci_set_wakeup(dev, false);
1957 		pci_pme_active(dev, false);
1958 		dev->wakeup_prepared = false;
1959 	}
1960 
1961 	return ret;
1962 }
1963 EXPORT_SYMBOL(pci_enable_wake);
1964 
1965 /**
1966  * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1967  * @dev: PCI device to prepare
1968  * @enable: True to enable wake-up event generation; false to disable
1969  *
1970  * Many drivers want the device to wake up the system from D3_hot or D3_cold
1971  * and this function allows them to set that up cleanly - pci_enable_wake()
1972  * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1973  * ordering constraints.
1974  *
1975  * This function only returns error code if the device is not capable of
1976  * generating PME# from both D3_hot and D3_cold, and the platform is unable to
1977  * enable wake-up power for it.
1978  */
1979 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1980 {
1981 	return pci_pme_capable(dev, PCI_D3cold) ?
1982 			pci_enable_wake(dev, PCI_D3cold, enable) :
1983 			pci_enable_wake(dev, PCI_D3hot, enable);
1984 }
1985 EXPORT_SYMBOL(pci_wake_from_d3);
1986 
1987 /**
1988  * pci_target_state - find an appropriate low power state for a given PCI dev
1989  * @dev: PCI device
1990  * @wakeup: Whether or not wakeup functionality will be enabled for the device.
1991  *
1992  * Use underlying platform code to find a supported low power state for @dev.
1993  * If the platform can't manage @dev, return the deepest state from which it
1994  * can generate wake events, based on any available PME info.
1995  */
1996 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
1997 {
1998 	pci_power_t target_state = PCI_D3hot;
1999 
2000 	if (platform_pci_power_manageable(dev)) {
2001 		/*
2002 		 * Call the platform to choose the target state of the device
2003 		 * and enable wake-up from this state if supported.
2004 		 */
2005 		pci_power_t state = platform_pci_choose_state(dev);
2006 
2007 		switch (state) {
2008 		case PCI_POWER_ERROR:
2009 		case PCI_UNKNOWN:
2010 			break;
2011 		case PCI_D1:
2012 		case PCI_D2:
2013 			if (pci_no_d1d2(dev))
2014 				break;
2015 		default:
2016 			target_state = state;
2017 		}
2018 
2019 		return target_state;
2020 	}
2021 
2022 	if (!dev->pm_cap)
2023 		target_state = PCI_D0;
2024 
2025 	/*
2026 	 * If the device is in D3cold even though it's not power-manageable by
2027 	 * the platform, it may have been powered down by non-standard means.
2028 	 * Best to let it slumber.
2029 	 */
2030 	if (dev->current_state == PCI_D3cold)
2031 		target_state = PCI_D3cold;
2032 
2033 	if (wakeup) {
2034 		/*
2035 		 * Find the deepest state from which the device can generate
2036 		 * wake-up events, make it the target state and enable device
2037 		 * to generate PME#.
2038 		 */
2039 		if (dev->pme_support) {
2040 			while (target_state
2041 			      && !(dev->pme_support & (1 << target_state)))
2042 				target_state--;
2043 		}
2044 	}
2045 
2046 	return target_state;
2047 }
2048 
2049 /**
2050  * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
2051  * @dev: Device to handle.
2052  *
2053  * Choose the power state appropriate for the device depending on whether
2054  * it can wake up the system and/or is power manageable by the platform
2055  * (PCI_D3hot is the default) and put the device into that state.
2056  */
2057 int pci_prepare_to_sleep(struct pci_dev *dev)
2058 {
2059 	bool wakeup = device_may_wakeup(&dev->dev);
2060 	pci_power_t target_state = pci_target_state(dev, wakeup);
2061 	int error;
2062 
2063 	if (target_state == PCI_POWER_ERROR)
2064 		return -EIO;
2065 
2066 	pci_enable_wake(dev, target_state, wakeup);
2067 
2068 	error = pci_set_power_state(dev, target_state);
2069 
2070 	if (error)
2071 		pci_enable_wake(dev, target_state, false);
2072 
2073 	return error;
2074 }
2075 EXPORT_SYMBOL(pci_prepare_to_sleep);
2076 
2077 /**
2078  * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2079  * @dev: Device to handle.
2080  *
2081  * Disable device's system wake-up capability and put it into D0.
2082  */
2083 int pci_back_from_sleep(struct pci_dev *dev)
2084 {
2085 	pci_enable_wake(dev, PCI_D0, false);
2086 	return pci_set_power_state(dev, PCI_D0);
2087 }
2088 EXPORT_SYMBOL(pci_back_from_sleep);
2089 
2090 /**
2091  * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2092  * @dev: PCI device being suspended.
2093  *
2094  * Prepare @dev to generate wake-up events at run time and put it into a low
2095  * power state.
2096  */
2097 int pci_finish_runtime_suspend(struct pci_dev *dev)
2098 {
2099 	pci_power_t target_state;
2100 	int error;
2101 
2102 	target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2103 	if (target_state == PCI_POWER_ERROR)
2104 		return -EIO;
2105 
2106 	dev->runtime_d3cold = target_state == PCI_D3cold;
2107 
2108 	pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2109 
2110 	error = pci_set_power_state(dev, target_state);
2111 
2112 	if (error) {
2113 		pci_enable_wake(dev, target_state, false);
2114 		dev->runtime_d3cold = false;
2115 	}
2116 
2117 	return error;
2118 }
2119 
2120 /**
2121  * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2122  * @dev: Device to check.
2123  *
2124  * Return true if the device itself is capable of generating wake-up events
2125  * (through the platform or using the native PCIe PME) or if the device supports
2126  * PME and one of its upstream bridges can generate wake-up events.
2127  */
2128 bool pci_dev_run_wake(struct pci_dev *dev)
2129 {
2130 	struct pci_bus *bus = dev->bus;
2131 
2132 	if (device_can_wakeup(&dev->dev))
2133 		return true;
2134 
2135 	if (!dev->pme_support)
2136 		return false;
2137 
2138 	/* PME-capable in principle, but not from the target power state */
2139 	if (!pci_pme_capable(dev, pci_target_state(dev, false)))
2140 		return false;
2141 
2142 	while (bus->parent) {
2143 		struct pci_dev *bridge = bus->self;
2144 
2145 		if (device_can_wakeup(&bridge->dev))
2146 			return true;
2147 
2148 		bus = bus->parent;
2149 	}
2150 
2151 	/* We have reached the root bus. */
2152 	if (bus->bridge)
2153 		return device_can_wakeup(bus->bridge);
2154 
2155 	return false;
2156 }
2157 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2158 
2159 /**
2160  * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2161  * @pci_dev: Device to check.
2162  *
2163  * Return 'true' if the device is runtime-suspended, it doesn't have to be
2164  * reconfigured due to wakeup settings difference between system and runtime
2165  * suspend and the current power state of it is suitable for the upcoming
2166  * (system) transition.
2167  *
2168  * If the device is not configured for system wakeup, disable PME for it before
2169  * returning 'true' to prevent it from waking up the system unnecessarily.
2170  */
2171 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2172 {
2173 	struct device *dev = &pci_dev->dev;
2174 	bool wakeup = device_may_wakeup(dev);
2175 
2176 	if (!pm_runtime_suspended(dev)
2177 	    || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2178 	    || platform_pci_need_resume(pci_dev)
2179 	    || (pci_dev->dev_flags & PCI_DEV_FLAGS_NEEDS_RESUME))
2180 		return false;
2181 
2182 	/*
2183 	 * At this point the device is good to go unless it's been configured
2184 	 * to generate PME at the runtime suspend time, but it is not supposed
2185 	 * to wake up the system.  In that case, simply disable PME for it
2186 	 * (it will have to be re-enabled on exit from system resume).
2187 	 *
2188 	 * If the device's power state is D3cold and the platform check above
2189 	 * hasn't triggered, the device's configuration is suitable and we don't
2190 	 * need to manipulate it at all.
2191 	 */
2192 	spin_lock_irq(&dev->power.lock);
2193 
2194 	if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2195 	    !wakeup)
2196 		__pci_pme_active(pci_dev, false);
2197 
2198 	spin_unlock_irq(&dev->power.lock);
2199 	return true;
2200 }
2201 
2202 /**
2203  * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2204  * @pci_dev: Device to handle.
2205  *
2206  * If the device is runtime suspended and wakeup-capable, enable PME for it as
2207  * it might have been disabled during the prepare phase of system suspend if
2208  * the device was not configured for system wakeup.
2209  */
2210 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2211 {
2212 	struct device *dev = &pci_dev->dev;
2213 
2214 	if (!pci_dev_run_wake(pci_dev))
2215 		return;
2216 
2217 	spin_lock_irq(&dev->power.lock);
2218 
2219 	if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2220 		__pci_pme_active(pci_dev, true);
2221 
2222 	spin_unlock_irq(&dev->power.lock);
2223 }
2224 
2225 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2226 {
2227 	struct device *dev = &pdev->dev;
2228 	struct device *parent = dev->parent;
2229 
2230 	if (parent)
2231 		pm_runtime_get_sync(parent);
2232 	pm_runtime_get_noresume(dev);
2233 	/*
2234 	 * pdev->current_state is set to PCI_D3cold during suspending,
2235 	 * so wait until suspending completes
2236 	 */
2237 	pm_runtime_barrier(dev);
2238 	/*
2239 	 * Only need to resume devices in D3cold, because config
2240 	 * registers are still accessible for devices suspended but
2241 	 * not in D3cold.
2242 	 */
2243 	if (pdev->current_state == PCI_D3cold)
2244 		pm_runtime_resume(dev);
2245 }
2246 
2247 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2248 {
2249 	struct device *dev = &pdev->dev;
2250 	struct device *parent = dev->parent;
2251 
2252 	pm_runtime_put(dev);
2253 	if (parent)
2254 		pm_runtime_put_sync(parent);
2255 }
2256 
2257 /**
2258  * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2259  * @bridge: Bridge to check
2260  *
2261  * This function checks if it is possible to move the bridge to D3.
2262  * Currently we only allow D3 for recent enough PCIe ports.
2263  */
2264 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2265 {
2266 	unsigned int year;
2267 
2268 	if (!pci_is_pcie(bridge))
2269 		return false;
2270 
2271 	switch (pci_pcie_type(bridge)) {
2272 	case PCI_EXP_TYPE_ROOT_PORT:
2273 	case PCI_EXP_TYPE_UPSTREAM:
2274 	case PCI_EXP_TYPE_DOWNSTREAM:
2275 		if (pci_bridge_d3_disable)
2276 			return false;
2277 
2278 		/*
2279 		 * Hotplug interrupts cannot be delivered if the link is down,
2280 		 * so parents of a hotplug port must stay awake. In addition,
2281 		 * hotplug ports handled by firmware in System Management Mode
2282 		 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2283 		 * For simplicity, disallow in general for now.
2284 		 */
2285 		if (bridge->is_hotplug_bridge)
2286 			return false;
2287 
2288 		if (pci_bridge_d3_force)
2289 			return true;
2290 
2291 		/*
2292 		 * It should be safe to put PCIe ports from 2015 or newer
2293 		 * to D3.
2294 		 */
2295 		if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
2296 		    year >= 2015) {
2297 			return true;
2298 		}
2299 		break;
2300 	}
2301 
2302 	return false;
2303 }
2304 
2305 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2306 {
2307 	bool *d3cold_ok = data;
2308 
2309 	if (/* The device needs to be allowed to go D3cold ... */
2310 	    dev->no_d3cold || !dev->d3cold_allowed ||
2311 
2312 	    /* ... and if it is wakeup capable to do so from D3cold. */
2313 	    (device_may_wakeup(&dev->dev) &&
2314 	     !pci_pme_capable(dev, PCI_D3cold)) ||
2315 
2316 	    /* If it is a bridge it must be allowed to go to D3. */
2317 	    !pci_power_manageable(dev))
2318 
2319 		*d3cold_ok = false;
2320 
2321 	return !*d3cold_ok;
2322 }
2323 
2324 /*
2325  * pci_bridge_d3_update - Update bridge D3 capabilities
2326  * @dev: PCI device which is changed
2327  *
2328  * Update upstream bridge PM capabilities accordingly depending on if the
2329  * device PM configuration was changed or the device is being removed.  The
2330  * change is also propagated upstream.
2331  */
2332 void pci_bridge_d3_update(struct pci_dev *dev)
2333 {
2334 	bool remove = !device_is_registered(&dev->dev);
2335 	struct pci_dev *bridge;
2336 	bool d3cold_ok = true;
2337 
2338 	bridge = pci_upstream_bridge(dev);
2339 	if (!bridge || !pci_bridge_d3_possible(bridge))
2340 		return;
2341 
2342 	/*
2343 	 * If D3 is currently allowed for the bridge, removing one of its
2344 	 * children won't change that.
2345 	 */
2346 	if (remove && bridge->bridge_d3)
2347 		return;
2348 
2349 	/*
2350 	 * If D3 is currently allowed for the bridge and a child is added or
2351 	 * changed, disallowance of D3 can only be caused by that child, so
2352 	 * we only need to check that single device, not any of its siblings.
2353 	 *
2354 	 * If D3 is currently not allowed for the bridge, checking the device
2355 	 * first may allow us to skip checking its siblings.
2356 	 */
2357 	if (!remove)
2358 		pci_dev_check_d3cold(dev, &d3cold_ok);
2359 
2360 	/*
2361 	 * If D3 is currently not allowed for the bridge, this may be caused
2362 	 * either by the device being changed/removed or any of its siblings,
2363 	 * so we need to go through all children to find out if one of them
2364 	 * continues to block D3.
2365 	 */
2366 	if (d3cold_ok && !bridge->bridge_d3)
2367 		pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2368 			     &d3cold_ok);
2369 
2370 	if (bridge->bridge_d3 != d3cold_ok) {
2371 		bridge->bridge_d3 = d3cold_ok;
2372 		/* Propagate change to upstream bridges */
2373 		pci_bridge_d3_update(bridge);
2374 	}
2375 }
2376 
2377 /**
2378  * pci_d3cold_enable - Enable D3cold for device
2379  * @dev: PCI device to handle
2380  *
2381  * This function can be used in drivers to enable D3cold from the device
2382  * they handle.  It also updates upstream PCI bridge PM capabilities
2383  * accordingly.
2384  */
2385 void pci_d3cold_enable(struct pci_dev *dev)
2386 {
2387 	if (dev->no_d3cold) {
2388 		dev->no_d3cold = false;
2389 		pci_bridge_d3_update(dev);
2390 	}
2391 }
2392 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2393 
2394 /**
2395  * pci_d3cold_disable - Disable D3cold for device
2396  * @dev: PCI device to handle
2397  *
2398  * This function can be used in drivers to disable D3cold from the device
2399  * they handle.  It also updates upstream PCI bridge PM capabilities
2400  * accordingly.
2401  */
2402 void pci_d3cold_disable(struct pci_dev *dev)
2403 {
2404 	if (!dev->no_d3cold) {
2405 		dev->no_d3cold = true;
2406 		pci_bridge_d3_update(dev);
2407 	}
2408 }
2409 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2410 
2411 /**
2412  * pci_pm_init - Initialize PM functions of given PCI device
2413  * @dev: PCI device to handle.
2414  */
2415 void pci_pm_init(struct pci_dev *dev)
2416 {
2417 	int pm;
2418 	u16 pmc;
2419 
2420 	pm_runtime_forbid(&dev->dev);
2421 	pm_runtime_set_active(&dev->dev);
2422 	pm_runtime_enable(&dev->dev);
2423 	device_enable_async_suspend(&dev->dev);
2424 	dev->wakeup_prepared = false;
2425 
2426 	dev->pm_cap = 0;
2427 	dev->pme_support = 0;
2428 
2429 	/* find PCI PM capability in list */
2430 	pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2431 	if (!pm)
2432 		return;
2433 	/* Check device's ability to generate PME# */
2434 	pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2435 
2436 	if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2437 		dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
2438 			pmc & PCI_PM_CAP_VER_MASK);
2439 		return;
2440 	}
2441 
2442 	dev->pm_cap = pm;
2443 	dev->d3_delay = PCI_PM_D3_WAIT;
2444 	dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2445 	dev->bridge_d3 = pci_bridge_d3_possible(dev);
2446 	dev->d3cold_allowed = true;
2447 
2448 	dev->d1_support = false;
2449 	dev->d2_support = false;
2450 	if (!pci_no_d1d2(dev)) {
2451 		if (pmc & PCI_PM_CAP_D1)
2452 			dev->d1_support = true;
2453 		if (pmc & PCI_PM_CAP_D2)
2454 			dev->d2_support = true;
2455 
2456 		if (dev->d1_support || dev->d2_support)
2457 			dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
2458 				   dev->d1_support ? " D1" : "",
2459 				   dev->d2_support ? " D2" : "");
2460 	}
2461 
2462 	pmc &= PCI_PM_CAP_PME_MASK;
2463 	if (pmc) {
2464 		dev_printk(KERN_DEBUG, &dev->dev,
2465 			 "PME# supported from%s%s%s%s%s\n",
2466 			 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2467 			 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2468 			 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2469 			 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2470 			 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2471 		dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2472 		dev->pme_poll = true;
2473 		/*
2474 		 * Make device's PM flags reflect the wake-up capability, but
2475 		 * let the user space enable it to wake up the system as needed.
2476 		 */
2477 		device_set_wakeup_capable(&dev->dev, true);
2478 		/* Disable the PME# generation functionality */
2479 		pci_pme_active(dev, false);
2480 	}
2481 }
2482 
2483 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2484 {
2485 	unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2486 
2487 	switch (prop) {
2488 	case PCI_EA_P_MEM:
2489 	case PCI_EA_P_VF_MEM:
2490 		flags |= IORESOURCE_MEM;
2491 		break;
2492 	case PCI_EA_P_MEM_PREFETCH:
2493 	case PCI_EA_P_VF_MEM_PREFETCH:
2494 		flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2495 		break;
2496 	case PCI_EA_P_IO:
2497 		flags |= IORESOURCE_IO;
2498 		break;
2499 	default:
2500 		return 0;
2501 	}
2502 
2503 	return flags;
2504 }
2505 
2506 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2507 					    u8 prop)
2508 {
2509 	if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2510 		return &dev->resource[bei];
2511 #ifdef CONFIG_PCI_IOV
2512 	else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2513 		 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2514 		return &dev->resource[PCI_IOV_RESOURCES +
2515 				      bei - PCI_EA_BEI_VF_BAR0];
2516 #endif
2517 	else if (bei == PCI_EA_BEI_ROM)
2518 		return &dev->resource[PCI_ROM_RESOURCE];
2519 	else
2520 		return NULL;
2521 }
2522 
2523 /* Read an Enhanced Allocation (EA) entry */
2524 static int pci_ea_read(struct pci_dev *dev, int offset)
2525 {
2526 	struct resource *res;
2527 	int ent_size, ent_offset = offset;
2528 	resource_size_t start, end;
2529 	unsigned long flags;
2530 	u32 dw0, bei, base, max_offset;
2531 	u8 prop;
2532 	bool support_64 = (sizeof(resource_size_t) >= 8);
2533 
2534 	pci_read_config_dword(dev, ent_offset, &dw0);
2535 	ent_offset += 4;
2536 
2537 	/* Entry size field indicates DWORDs after 1st */
2538 	ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2539 
2540 	if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2541 		goto out;
2542 
2543 	bei = (dw0 & PCI_EA_BEI) >> 4;
2544 	prop = (dw0 & PCI_EA_PP) >> 8;
2545 
2546 	/*
2547 	 * If the Property is in the reserved range, try the Secondary
2548 	 * Property instead.
2549 	 */
2550 	if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2551 		prop = (dw0 & PCI_EA_SP) >> 16;
2552 	if (prop > PCI_EA_P_BRIDGE_IO)
2553 		goto out;
2554 
2555 	res = pci_ea_get_resource(dev, bei, prop);
2556 	if (!res) {
2557 		dev_err(&dev->dev, "Unsupported EA entry BEI: %u\n", bei);
2558 		goto out;
2559 	}
2560 
2561 	flags = pci_ea_flags(dev, prop);
2562 	if (!flags) {
2563 		dev_err(&dev->dev, "Unsupported EA properties: %#x\n", prop);
2564 		goto out;
2565 	}
2566 
2567 	/* Read Base */
2568 	pci_read_config_dword(dev, ent_offset, &base);
2569 	start = (base & PCI_EA_FIELD_MASK);
2570 	ent_offset += 4;
2571 
2572 	/* Read MaxOffset */
2573 	pci_read_config_dword(dev, ent_offset, &max_offset);
2574 	ent_offset += 4;
2575 
2576 	/* Read Base MSBs (if 64-bit entry) */
2577 	if (base & PCI_EA_IS_64) {
2578 		u32 base_upper;
2579 
2580 		pci_read_config_dword(dev, ent_offset, &base_upper);
2581 		ent_offset += 4;
2582 
2583 		flags |= IORESOURCE_MEM_64;
2584 
2585 		/* entry starts above 32-bit boundary, can't use */
2586 		if (!support_64 && base_upper)
2587 			goto out;
2588 
2589 		if (support_64)
2590 			start |= ((u64)base_upper << 32);
2591 	}
2592 
2593 	end = start + (max_offset | 0x03);
2594 
2595 	/* Read MaxOffset MSBs (if 64-bit entry) */
2596 	if (max_offset & PCI_EA_IS_64) {
2597 		u32 max_offset_upper;
2598 
2599 		pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2600 		ent_offset += 4;
2601 
2602 		flags |= IORESOURCE_MEM_64;
2603 
2604 		/* entry too big, can't use */
2605 		if (!support_64 && max_offset_upper)
2606 			goto out;
2607 
2608 		if (support_64)
2609 			end += ((u64)max_offset_upper << 32);
2610 	}
2611 
2612 	if (end < start) {
2613 		dev_err(&dev->dev, "EA Entry crosses address boundary\n");
2614 		goto out;
2615 	}
2616 
2617 	if (ent_size != ent_offset - offset) {
2618 		dev_err(&dev->dev,
2619 			"EA Entry Size (%d) does not match length read (%d)\n",
2620 			ent_size, ent_offset - offset);
2621 		goto out;
2622 	}
2623 
2624 	res->name = pci_name(dev);
2625 	res->start = start;
2626 	res->end = end;
2627 	res->flags = flags;
2628 
2629 	if (bei <= PCI_EA_BEI_BAR5)
2630 		dev_printk(KERN_DEBUG, &dev->dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2631 			   bei, res, prop);
2632 	else if (bei == PCI_EA_BEI_ROM)
2633 		dev_printk(KERN_DEBUG, &dev->dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2634 			   res, prop);
2635 	else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2636 		dev_printk(KERN_DEBUG, &dev->dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2637 			   bei - PCI_EA_BEI_VF_BAR0, res, prop);
2638 	else
2639 		dev_printk(KERN_DEBUG, &dev->dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2640 			   bei, res, prop);
2641 
2642 out:
2643 	return offset + ent_size;
2644 }
2645 
2646 /* Enhanced Allocation Initialization */
2647 void pci_ea_init(struct pci_dev *dev)
2648 {
2649 	int ea;
2650 	u8 num_ent;
2651 	int offset;
2652 	int i;
2653 
2654 	/* find PCI EA capability in list */
2655 	ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2656 	if (!ea)
2657 		return;
2658 
2659 	/* determine the number of entries */
2660 	pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2661 					&num_ent);
2662 	num_ent &= PCI_EA_NUM_ENT_MASK;
2663 
2664 	offset = ea + PCI_EA_FIRST_ENT;
2665 
2666 	/* Skip DWORD 2 for type 1 functions */
2667 	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2668 		offset += 4;
2669 
2670 	/* parse each EA entry */
2671 	for (i = 0; i < num_ent; ++i)
2672 		offset = pci_ea_read(dev, offset);
2673 }
2674 
2675 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2676 	struct pci_cap_saved_state *new_cap)
2677 {
2678 	hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2679 }
2680 
2681 /**
2682  * _pci_add_cap_save_buffer - allocate buffer for saving given
2683  *                            capability registers
2684  * @dev: the PCI device
2685  * @cap: the capability to allocate the buffer for
2686  * @extended: Standard or Extended capability ID
2687  * @size: requested size of the buffer
2688  */
2689 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2690 				    bool extended, unsigned int size)
2691 {
2692 	int pos;
2693 	struct pci_cap_saved_state *save_state;
2694 
2695 	if (extended)
2696 		pos = pci_find_ext_capability(dev, cap);
2697 	else
2698 		pos = pci_find_capability(dev, cap);
2699 
2700 	if (!pos)
2701 		return 0;
2702 
2703 	save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2704 	if (!save_state)
2705 		return -ENOMEM;
2706 
2707 	save_state->cap.cap_nr = cap;
2708 	save_state->cap.cap_extended = extended;
2709 	save_state->cap.size = size;
2710 	pci_add_saved_cap(dev, save_state);
2711 
2712 	return 0;
2713 }
2714 
2715 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2716 {
2717 	return _pci_add_cap_save_buffer(dev, cap, false, size);
2718 }
2719 
2720 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2721 {
2722 	return _pci_add_cap_save_buffer(dev, cap, true, size);
2723 }
2724 
2725 /**
2726  * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2727  * @dev: the PCI device
2728  */
2729 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2730 {
2731 	int error;
2732 
2733 	error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2734 					PCI_EXP_SAVE_REGS * sizeof(u16));
2735 	if (error)
2736 		dev_err(&dev->dev,
2737 			"unable to preallocate PCI Express save buffer\n");
2738 
2739 	error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2740 	if (error)
2741 		dev_err(&dev->dev,
2742 			"unable to preallocate PCI-X save buffer\n");
2743 
2744 	pci_allocate_vc_save_buffers(dev);
2745 }
2746 
2747 void pci_free_cap_save_buffers(struct pci_dev *dev)
2748 {
2749 	struct pci_cap_saved_state *tmp;
2750 	struct hlist_node *n;
2751 
2752 	hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2753 		kfree(tmp);
2754 }
2755 
2756 /**
2757  * pci_configure_ari - enable or disable ARI forwarding
2758  * @dev: the PCI device
2759  *
2760  * If @dev and its upstream bridge both support ARI, enable ARI in the
2761  * bridge.  Otherwise, disable ARI in the bridge.
2762  */
2763 void pci_configure_ari(struct pci_dev *dev)
2764 {
2765 	u32 cap;
2766 	struct pci_dev *bridge;
2767 
2768 	if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2769 		return;
2770 
2771 	bridge = dev->bus->self;
2772 	if (!bridge)
2773 		return;
2774 
2775 	pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2776 	if (!(cap & PCI_EXP_DEVCAP2_ARI))
2777 		return;
2778 
2779 	if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2780 		pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2781 					 PCI_EXP_DEVCTL2_ARI);
2782 		bridge->ari_enabled = 1;
2783 	} else {
2784 		pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2785 					   PCI_EXP_DEVCTL2_ARI);
2786 		bridge->ari_enabled = 0;
2787 	}
2788 }
2789 
2790 static int pci_acs_enable;
2791 
2792 /**
2793  * pci_request_acs - ask for ACS to be enabled if supported
2794  */
2795 void pci_request_acs(void)
2796 {
2797 	pci_acs_enable = 1;
2798 }
2799 
2800 /**
2801  * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2802  * @dev: the PCI device
2803  */
2804 static void pci_std_enable_acs(struct pci_dev *dev)
2805 {
2806 	int pos;
2807 	u16 cap;
2808 	u16 ctrl;
2809 
2810 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2811 	if (!pos)
2812 		return;
2813 
2814 	pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2815 	pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2816 
2817 	/* Source Validation */
2818 	ctrl |= (cap & PCI_ACS_SV);
2819 
2820 	/* P2P Request Redirect */
2821 	ctrl |= (cap & PCI_ACS_RR);
2822 
2823 	/* P2P Completion Redirect */
2824 	ctrl |= (cap & PCI_ACS_CR);
2825 
2826 	/* Upstream Forwarding */
2827 	ctrl |= (cap & PCI_ACS_UF);
2828 
2829 	pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2830 }
2831 
2832 /**
2833  * pci_enable_acs - enable ACS if hardware support it
2834  * @dev: the PCI device
2835  */
2836 void pci_enable_acs(struct pci_dev *dev)
2837 {
2838 	if (!pci_acs_enable)
2839 		return;
2840 
2841 	if (!pci_dev_specific_enable_acs(dev))
2842 		return;
2843 
2844 	pci_std_enable_acs(dev);
2845 }
2846 
2847 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2848 {
2849 	int pos;
2850 	u16 cap, ctrl;
2851 
2852 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2853 	if (!pos)
2854 		return false;
2855 
2856 	/*
2857 	 * Except for egress control, capabilities are either required
2858 	 * or only required if controllable.  Features missing from the
2859 	 * capability field can therefore be assumed as hard-wired enabled.
2860 	 */
2861 	pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2862 	acs_flags &= (cap | PCI_ACS_EC);
2863 
2864 	pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2865 	return (ctrl & acs_flags) == acs_flags;
2866 }
2867 
2868 /**
2869  * pci_acs_enabled - test ACS against required flags for a given device
2870  * @pdev: device to test
2871  * @acs_flags: required PCI ACS flags
2872  *
2873  * Return true if the device supports the provided flags.  Automatically
2874  * filters out flags that are not implemented on multifunction devices.
2875  *
2876  * Note that this interface checks the effective ACS capabilities of the
2877  * device rather than the actual capabilities.  For instance, most single
2878  * function endpoints are not required to support ACS because they have no
2879  * opportunity for peer-to-peer access.  We therefore return 'true'
2880  * regardless of whether the device exposes an ACS capability.  This makes
2881  * it much easier for callers of this function to ignore the actual type
2882  * or topology of the device when testing ACS support.
2883  */
2884 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2885 {
2886 	int ret;
2887 
2888 	ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2889 	if (ret >= 0)
2890 		return ret > 0;
2891 
2892 	/*
2893 	 * Conventional PCI and PCI-X devices never support ACS, either
2894 	 * effectively or actually.  The shared bus topology implies that
2895 	 * any device on the bus can receive or snoop DMA.
2896 	 */
2897 	if (!pci_is_pcie(pdev))
2898 		return false;
2899 
2900 	switch (pci_pcie_type(pdev)) {
2901 	/*
2902 	 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2903 	 * but since their primary interface is PCI/X, we conservatively
2904 	 * handle them as we would a non-PCIe device.
2905 	 */
2906 	case PCI_EXP_TYPE_PCIE_BRIDGE:
2907 	/*
2908 	 * PCIe 3.0, 6.12.1 excludes ACS on these devices.  "ACS is never
2909 	 * applicable... must never implement an ACS Extended Capability...".
2910 	 * This seems arbitrary, but we take a conservative interpretation
2911 	 * of this statement.
2912 	 */
2913 	case PCI_EXP_TYPE_PCI_BRIDGE:
2914 	case PCI_EXP_TYPE_RC_EC:
2915 		return false;
2916 	/*
2917 	 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2918 	 * implement ACS in order to indicate their peer-to-peer capabilities,
2919 	 * regardless of whether they are single- or multi-function devices.
2920 	 */
2921 	case PCI_EXP_TYPE_DOWNSTREAM:
2922 	case PCI_EXP_TYPE_ROOT_PORT:
2923 		return pci_acs_flags_enabled(pdev, acs_flags);
2924 	/*
2925 	 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2926 	 * implemented by the remaining PCIe types to indicate peer-to-peer
2927 	 * capabilities, but only when they are part of a multifunction
2928 	 * device.  The footnote for section 6.12 indicates the specific
2929 	 * PCIe types included here.
2930 	 */
2931 	case PCI_EXP_TYPE_ENDPOINT:
2932 	case PCI_EXP_TYPE_UPSTREAM:
2933 	case PCI_EXP_TYPE_LEG_END:
2934 	case PCI_EXP_TYPE_RC_END:
2935 		if (!pdev->multifunction)
2936 			break;
2937 
2938 		return pci_acs_flags_enabled(pdev, acs_flags);
2939 	}
2940 
2941 	/*
2942 	 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2943 	 * to single function devices with the exception of downstream ports.
2944 	 */
2945 	return true;
2946 }
2947 
2948 /**
2949  * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2950  * @start: starting downstream device
2951  * @end: ending upstream device or NULL to search to the root bus
2952  * @acs_flags: required flags
2953  *
2954  * Walk up a device tree from start to end testing PCI ACS support.  If
2955  * any step along the way does not support the required flags, return false.
2956  */
2957 bool pci_acs_path_enabled(struct pci_dev *start,
2958 			  struct pci_dev *end, u16 acs_flags)
2959 {
2960 	struct pci_dev *pdev, *parent = start;
2961 
2962 	do {
2963 		pdev = parent;
2964 
2965 		if (!pci_acs_enabled(pdev, acs_flags))
2966 			return false;
2967 
2968 		if (pci_is_root_bus(pdev->bus))
2969 			return (end == NULL);
2970 
2971 		parent = pdev->bus->self;
2972 	} while (pdev != end);
2973 
2974 	return true;
2975 }
2976 
2977 /**
2978  * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2979  * @dev: the PCI device
2980  * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
2981  *
2982  * Perform INTx swizzling for a device behind one level of bridge.  This is
2983  * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2984  * behind bridges on add-in cards.  For devices with ARI enabled, the slot
2985  * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2986  * the PCI Express Base Specification, Revision 2.1)
2987  */
2988 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2989 {
2990 	int slot;
2991 
2992 	if (pci_ari_enabled(dev->bus))
2993 		slot = 0;
2994 	else
2995 		slot = PCI_SLOT(dev->devfn);
2996 
2997 	return (((pin - 1) + slot) % 4) + 1;
2998 }
2999 
3000 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3001 {
3002 	u8 pin;
3003 
3004 	pin = dev->pin;
3005 	if (!pin)
3006 		return -1;
3007 
3008 	while (!pci_is_root_bus(dev->bus)) {
3009 		pin = pci_swizzle_interrupt_pin(dev, pin);
3010 		dev = dev->bus->self;
3011 	}
3012 	*bridge = dev;
3013 	return pin;
3014 }
3015 
3016 /**
3017  * pci_common_swizzle - swizzle INTx all the way to root bridge
3018  * @dev: the PCI device
3019  * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3020  *
3021  * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
3022  * bridges all the way up to a PCI root bus.
3023  */
3024 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3025 {
3026 	u8 pin = *pinp;
3027 
3028 	while (!pci_is_root_bus(dev->bus)) {
3029 		pin = pci_swizzle_interrupt_pin(dev, pin);
3030 		dev = dev->bus->self;
3031 	}
3032 	*pinp = pin;
3033 	return PCI_SLOT(dev->devfn);
3034 }
3035 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3036 
3037 /**
3038  *	pci_release_region - Release a PCI bar
3039  *	@pdev: PCI device whose resources were previously reserved by pci_request_region
3040  *	@bar: BAR to release
3041  *
3042  *	Releases the PCI I/O and memory resources previously reserved by a
3043  *	successful call to pci_request_region.  Call this function only
3044  *	after all use of the PCI regions has ceased.
3045  */
3046 void pci_release_region(struct pci_dev *pdev, int bar)
3047 {
3048 	struct pci_devres *dr;
3049 
3050 	if (pci_resource_len(pdev, bar) == 0)
3051 		return;
3052 	if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3053 		release_region(pci_resource_start(pdev, bar),
3054 				pci_resource_len(pdev, bar));
3055 	else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3056 		release_mem_region(pci_resource_start(pdev, bar),
3057 				pci_resource_len(pdev, bar));
3058 
3059 	dr = find_pci_dr(pdev);
3060 	if (dr)
3061 		dr->region_mask &= ~(1 << bar);
3062 }
3063 EXPORT_SYMBOL(pci_release_region);
3064 
3065 /**
3066  *	__pci_request_region - Reserved PCI I/O and memory resource
3067  *	@pdev: PCI device whose resources are to be reserved
3068  *	@bar: BAR to be reserved
3069  *	@res_name: Name to be associated with resource.
3070  *	@exclusive: whether the region access is exclusive or not
3071  *
3072  *	Mark the PCI region associated with PCI device @pdev BR @bar as
3073  *	being reserved by owner @res_name.  Do not access any
3074  *	address inside the PCI regions unless this call returns
3075  *	successfully.
3076  *
3077  *	If @exclusive is set, then the region is marked so that userspace
3078  *	is explicitly not allowed to map the resource via /dev/mem or
3079  *	sysfs MMIO access.
3080  *
3081  *	Returns 0 on success, or %EBUSY on error.  A warning
3082  *	message is also printed on failure.
3083  */
3084 static int __pci_request_region(struct pci_dev *pdev, int bar,
3085 				const char *res_name, int exclusive)
3086 {
3087 	struct pci_devres *dr;
3088 
3089 	if (pci_resource_len(pdev, bar) == 0)
3090 		return 0;
3091 
3092 	if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3093 		if (!request_region(pci_resource_start(pdev, bar),
3094 			    pci_resource_len(pdev, bar), res_name))
3095 			goto err_out;
3096 	} else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3097 		if (!__request_mem_region(pci_resource_start(pdev, bar),
3098 					pci_resource_len(pdev, bar), res_name,
3099 					exclusive))
3100 			goto err_out;
3101 	}
3102 
3103 	dr = find_pci_dr(pdev);
3104 	if (dr)
3105 		dr->region_mask |= 1 << bar;
3106 
3107 	return 0;
3108 
3109 err_out:
3110 	dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
3111 		 &pdev->resource[bar]);
3112 	return -EBUSY;
3113 }
3114 
3115 /**
3116  *	pci_request_region - Reserve PCI I/O and memory resource
3117  *	@pdev: PCI device whose resources are to be reserved
3118  *	@bar: BAR to be reserved
3119  *	@res_name: Name to be associated with resource
3120  *
3121  *	Mark the PCI region associated with PCI device @pdev BAR @bar as
3122  *	being reserved by owner @res_name.  Do not access any
3123  *	address inside the PCI regions unless this call returns
3124  *	successfully.
3125  *
3126  *	Returns 0 on success, or %EBUSY on error.  A warning
3127  *	message is also printed on failure.
3128  */
3129 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3130 {
3131 	return __pci_request_region(pdev, bar, res_name, 0);
3132 }
3133 EXPORT_SYMBOL(pci_request_region);
3134 
3135 /**
3136  *	pci_request_region_exclusive - Reserved PCI I/O and memory resource
3137  *	@pdev: PCI device whose resources are to be reserved
3138  *	@bar: BAR to be reserved
3139  *	@res_name: Name to be associated with resource.
3140  *
3141  *	Mark the PCI region associated with PCI device @pdev BR @bar as
3142  *	being reserved by owner @res_name.  Do not access any
3143  *	address inside the PCI regions unless this call returns
3144  *	successfully.
3145  *
3146  *	Returns 0 on success, or %EBUSY on error.  A warning
3147  *	message is also printed on failure.
3148  *
3149  *	The key difference that _exclusive makes it that userspace is
3150  *	explicitly not allowed to map the resource via /dev/mem or
3151  *	sysfs.
3152  */
3153 int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3154 				 const char *res_name)
3155 {
3156 	return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3157 }
3158 EXPORT_SYMBOL(pci_request_region_exclusive);
3159 
3160 /**
3161  * pci_release_selected_regions - Release selected PCI I/O and memory resources
3162  * @pdev: PCI device whose resources were previously reserved
3163  * @bars: Bitmask of BARs to be released
3164  *
3165  * Release selected PCI I/O and memory resources previously reserved.
3166  * Call this function only after all use of the PCI regions has ceased.
3167  */
3168 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3169 {
3170 	int i;
3171 
3172 	for (i = 0; i < 6; i++)
3173 		if (bars & (1 << i))
3174 			pci_release_region(pdev, i);
3175 }
3176 EXPORT_SYMBOL(pci_release_selected_regions);
3177 
3178 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3179 					  const char *res_name, int excl)
3180 {
3181 	int i;
3182 
3183 	for (i = 0; i < 6; i++)
3184 		if (bars & (1 << i))
3185 			if (__pci_request_region(pdev, i, res_name, excl))
3186 				goto err_out;
3187 	return 0;
3188 
3189 err_out:
3190 	while (--i >= 0)
3191 		if (bars & (1 << i))
3192 			pci_release_region(pdev, i);
3193 
3194 	return -EBUSY;
3195 }
3196 
3197 
3198 /**
3199  * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3200  * @pdev: PCI device whose resources are to be reserved
3201  * @bars: Bitmask of BARs to be requested
3202  * @res_name: Name to be associated with resource
3203  */
3204 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3205 				 const char *res_name)
3206 {
3207 	return __pci_request_selected_regions(pdev, bars, res_name, 0);
3208 }
3209 EXPORT_SYMBOL(pci_request_selected_regions);
3210 
3211 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3212 					   const char *res_name)
3213 {
3214 	return __pci_request_selected_regions(pdev, bars, res_name,
3215 			IORESOURCE_EXCLUSIVE);
3216 }
3217 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3218 
3219 /**
3220  *	pci_release_regions - Release reserved PCI I/O and memory resources
3221  *	@pdev: PCI device whose resources were previously reserved by pci_request_regions
3222  *
3223  *	Releases all PCI I/O and memory resources previously reserved by a
3224  *	successful call to pci_request_regions.  Call this function only
3225  *	after all use of the PCI regions has ceased.
3226  */
3227 
3228 void pci_release_regions(struct pci_dev *pdev)
3229 {
3230 	pci_release_selected_regions(pdev, (1 << 6) - 1);
3231 }
3232 EXPORT_SYMBOL(pci_release_regions);
3233 
3234 /**
3235  *	pci_request_regions - Reserved PCI I/O and memory resources
3236  *	@pdev: PCI device whose resources are to be reserved
3237  *	@res_name: Name to be associated with resource.
3238  *
3239  *	Mark all PCI regions associated with PCI device @pdev as
3240  *	being reserved by owner @res_name.  Do not access any
3241  *	address inside the PCI regions unless this call returns
3242  *	successfully.
3243  *
3244  *	Returns 0 on success, or %EBUSY on error.  A warning
3245  *	message is also printed on failure.
3246  */
3247 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3248 {
3249 	return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3250 }
3251 EXPORT_SYMBOL(pci_request_regions);
3252 
3253 /**
3254  *	pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3255  *	@pdev: PCI device whose resources are to be reserved
3256  *	@res_name: Name to be associated with resource.
3257  *
3258  *	Mark all PCI regions associated with PCI device @pdev as
3259  *	being reserved by owner @res_name.  Do not access any
3260  *	address inside the PCI regions unless this call returns
3261  *	successfully.
3262  *
3263  *	pci_request_regions_exclusive() will mark the region so that
3264  *	/dev/mem and the sysfs MMIO access will not be allowed.
3265  *
3266  *	Returns 0 on success, or %EBUSY on error.  A warning
3267  *	message is also printed on failure.
3268  */
3269 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3270 {
3271 	return pci_request_selected_regions_exclusive(pdev,
3272 					((1 << 6) - 1), res_name);
3273 }
3274 EXPORT_SYMBOL(pci_request_regions_exclusive);
3275 
3276 #ifdef PCI_IOBASE
3277 struct io_range {
3278 	struct list_head list;
3279 	phys_addr_t start;
3280 	resource_size_t size;
3281 };
3282 
3283 static LIST_HEAD(io_range_list);
3284 static DEFINE_SPINLOCK(io_range_lock);
3285 #endif
3286 
3287 /*
3288  * Record the PCI IO range (expressed as CPU physical address + size).
3289  * Return a negative value if an error has occured, zero otherwise
3290  */
3291 int __weak pci_register_io_range(phys_addr_t addr, resource_size_t size)
3292 {
3293 	int err = 0;
3294 
3295 #ifdef PCI_IOBASE
3296 	struct io_range *range;
3297 	resource_size_t allocated_size = 0;
3298 
3299 	/* check if the range hasn't been previously recorded */
3300 	spin_lock(&io_range_lock);
3301 	list_for_each_entry(range, &io_range_list, list) {
3302 		if (addr >= range->start && addr + size <= range->start + size) {
3303 			/* range already registered, bail out */
3304 			goto end_register;
3305 		}
3306 		allocated_size += range->size;
3307 	}
3308 
3309 	/* range not registed yet, check for available space */
3310 	if (allocated_size + size - 1 > IO_SPACE_LIMIT) {
3311 		/* if it's too big check if 64K space can be reserved */
3312 		if (allocated_size + SZ_64K - 1 > IO_SPACE_LIMIT) {
3313 			err = -E2BIG;
3314 			goto end_register;
3315 		}
3316 
3317 		size = SZ_64K;
3318 		pr_warn("Requested IO range too big, new size set to 64K\n");
3319 	}
3320 
3321 	/* add the range to the list */
3322 	range = kzalloc(sizeof(*range), GFP_ATOMIC);
3323 	if (!range) {
3324 		err = -ENOMEM;
3325 		goto end_register;
3326 	}
3327 
3328 	range->start = addr;
3329 	range->size = size;
3330 
3331 	list_add_tail(&range->list, &io_range_list);
3332 
3333 end_register:
3334 	spin_unlock(&io_range_lock);
3335 #endif
3336 
3337 	return err;
3338 }
3339 
3340 phys_addr_t pci_pio_to_address(unsigned long pio)
3341 {
3342 	phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3343 
3344 #ifdef PCI_IOBASE
3345 	struct io_range *range;
3346 	resource_size_t allocated_size = 0;
3347 
3348 	if (pio > IO_SPACE_LIMIT)
3349 		return address;
3350 
3351 	spin_lock(&io_range_lock);
3352 	list_for_each_entry(range, &io_range_list, list) {
3353 		if (pio >= allocated_size && pio < allocated_size + range->size) {
3354 			address = range->start + pio - allocated_size;
3355 			break;
3356 		}
3357 		allocated_size += range->size;
3358 	}
3359 	spin_unlock(&io_range_lock);
3360 #endif
3361 
3362 	return address;
3363 }
3364 
3365 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3366 {
3367 #ifdef PCI_IOBASE
3368 	struct io_range *res;
3369 	resource_size_t offset = 0;
3370 	unsigned long addr = -1;
3371 
3372 	spin_lock(&io_range_lock);
3373 	list_for_each_entry(res, &io_range_list, list) {
3374 		if (address >= res->start && address < res->start + res->size) {
3375 			addr = address - res->start + offset;
3376 			break;
3377 		}
3378 		offset += res->size;
3379 	}
3380 	spin_unlock(&io_range_lock);
3381 
3382 	return addr;
3383 #else
3384 	if (address > IO_SPACE_LIMIT)
3385 		return (unsigned long)-1;
3386 
3387 	return (unsigned long) address;
3388 #endif
3389 }
3390 
3391 /**
3392  *	pci_remap_iospace - Remap the memory mapped I/O space
3393  *	@res: Resource describing the I/O space
3394  *	@phys_addr: physical address of range to be mapped
3395  *
3396  *	Remap the memory mapped I/O space described by the @res
3397  *	and the CPU physical address @phys_addr into virtual address space.
3398  *	Only architectures that have memory mapped IO functions defined
3399  *	(and the PCI_IOBASE value defined) should call this function.
3400  */
3401 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3402 {
3403 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3404 	unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3405 
3406 	if (!(res->flags & IORESOURCE_IO))
3407 		return -EINVAL;
3408 
3409 	if (res->end > IO_SPACE_LIMIT)
3410 		return -EINVAL;
3411 
3412 	return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3413 				  pgprot_device(PAGE_KERNEL));
3414 #else
3415 	/* this architecture does not have memory mapped I/O space,
3416 	   so this function should never be called */
3417 	WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3418 	return -ENODEV;
3419 #endif
3420 }
3421 EXPORT_SYMBOL(pci_remap_iospace);
3422 
3423 /**
3424  *	pci_unmap_iospace - Unmap the memory mapped I/O space
3425  *	@res: resource to be unmapped
3426  *
3427  *	Unmap the CPU virtual address @res from virtual address space.
3428  *	Only architectures that have memory mapped IO functions defined
3429  *	(and the PCI_IOBASE value defined) should call this function.
3430  */
3431 void pci_unmap_iospace(struct resource *res)
3432 {
3433 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3434 	unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3435 
3436 	unmap_kernel_range(vaddr, resource_size(res));
3437 #endif
3438 }
3439 EXPORT_SYMBOL(pci_unmap_iospace);
3440 
3441 /**
3442  * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3443  * @dev: Generic device to remap IO address for
3444  * @offset: Resource address to map
3445  * @size: Size of map
3446  *
3447  * Managed pci_remap_cfgspace().  Map is automatically unmapped on driver
3448  * detach.
3449  */
3450 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3451 				      resource_size_t offset,
3452 				      resource_size_t size)
3453 {
3454 	void __iomem **ptr, *addr;
3455 
3456 	ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
3457 	if (!ptr)
3458 		return NULL;
3459 
3460 	addr = pci_remap_cfgspace(offset, size);
3461 	if (addr) {
3462 		*ptr = addr;
3463 		devres_add(dev, ptr);
3464 	} else
3465 		devres_free(ptr);
3466 
3467 	return addr;
3468 }
3469 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
3470 
3471 /**
3472  * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
3473  * @dev: generic device to handle the resource for
3474  * @res: configuration space resource to be handled
3475  *
3476  * Checks that a resource is a valid memory region, requests the memory
3477  * region and ioremaps with pci_remap_cfgspace() API that ensures the
3478  * proper PCI configuration space memory attributes are guaranteed.
3479  *
3480  * All operations are managed and will be undone on driver detach.
3481  *
3482  * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
3483  * on failure. Usage example:
3484  *
3485  *	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
3486  *	base = devm_pci_remap_cfg_resource(&pdev->dev, res);
3487  *	if (IS_ERR(base))
3488  *		return PTR_ERR(base);
3489  */
3490 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
3491 					  struct resource *res)
3492 {
3493 	resource_size_t size;
3494 	const char *name;
3495 	void __iomem *dest_ptr;
3496 
3497 	BUG_ON(!dev);
3498 
3499 	if (!res || resource_type(res) != IORESOURCE_MEM) {
3500 		dev_err(dev, "invalid resource\n");
3501 		return IOMEM_ERR_PTR(-EINVAL);
3502 	}
3503 
3504 	size = resource_size(res);
3505 	name = res->name ?: dev_name(dev);
3506 
3507 	if (!devm_request_mem_region(dev, res->start, size, name)) {
3508 		dev_err(dev, "can't request region for resource %pR\n", res);
3509 		return IOMEM_ERR_PTR(-EBUSY);
3510 	}
3511 
3512 	dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
3513 	if (!dest_ptr) {
3514 		dev_err(dev, "ioremap failed for resource %pR\n", res);
3515 		devm_release_mem_region(dev, res->start, size);
3516 		dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
3517 	}
3518 
3519 	return dest_ptr;
3520 }
3521 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
3522 
3523 static void __pci_set_master(struct pci_dev *dev, bool enable)
3524 {
3525 	u16 old_cmd, cmd;
3526 
3527 	pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3528 	if (enable)
3529 		cmd = old_cmd | PCI_COMMAND_MASTER;
3530 	else
3531 		cmd = old_cmd & ~PCI_COMMAND_MASTER;
3532 	if (cmd != old_cmd) {
3533 		dev_dbg(&dev->dev, "%s bus mastering\n",
3534 			enable ? "enabling" : "disabling");
3535 		pci_write_config_word(dev, PCI_COMMAND, cmd);
3536 	}
3537 	dev->is_busmaster = enable;
3538 }
3539 
3540 /**
3541  * pcibios_setup - process "pci=" kernel boot arguments
3542  * @str: string used to pass in "pci=" kernel boot arguments
3543  *
3544  * Process kernel boot arguments.  This is the default implementation.
3545  * Architecture specific implementations can override this as necessary.
3546  */
3547 char * __weak __init pcibios_setup(char *str)
3548 {
3549 	return str;
3550 }
3551 
3552 /**
3553  * pcibios_set_master - enable PCI bus-mastering for device dev
3554  * @dev: the PCI device to enable
3555  *
3556  * Enables PCI bus-mastering for the device.  This is the default
3557  * implementation.  Architecture specific implementations can override
3558  * this if necessary.
3559  */
3560 void __weak pcibios_set_master(struct pci_dev *dev)
3561 {
3562 	u8 lat;
3563 
3564 	/* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3565 	if (pci_is_pcie(dev))
3566 		return;
3567 
3568 	pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3569 	if (lat < 16)
3570 		lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3571 	else if (lat > pcibios_max_latency)
3572 		lat = pcibios_max_latency;
3573 	else
3574 		return;
3575 
3576 	pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3577 }
3578 
3579 /**
3580  * pci_set_master - enables bus-mastering for device dev
3581  * @dev: the PCI device to enable
3582  *
3583  * Enables bus-mastering on the device and calls pcibios_set_master()
3584  * to do the needed arch specific settings.
3585  */
3586 void pci_set_master(struct pci_dev *dev)
3587 {
3588 	__pci_set_master(dev, true);
3589 	pcibios_set_master(dev);
3590 }
3591 EXPORT_SYMBOL(pci_set_master);
3592 
3593 /**
3594  * pci_clear_master - disables bus-mastering for device dev
3595  * @dev: the PCI device to disable
3596  */
3597 void pci_clear_master(struct pci_dev *dev)
3598 {
3599 	__pci_set_master(dev, false);
3600 }
3601 EXPORT_SYMBOL(pci_clear_master);
3602 
3603 /**
3604  * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3605  * @dev: the PCI device for which MWI is to be enabled
3606  *
3607  * Helper function for pci_set_mwi.
3608  * Originally copied from drivers/net/acenic.c.
3609  * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3610  *
3611  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3612  */
3613 int pci_set_cacheline_size(struct pci_dev *dev)
3614 {
3615 	u8 cacheline_size;
3616 
3617 	if (!pci_cache_line_size)
3618 		return -EINVAL;
3619 
3620 	/* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3621 	   equal to or multiple of the right value. */
3622 	pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3623 	if (cacheline_size >= pci_cache_line_size &&
3624 	    (cacheline_size % pci_cache_line_size) == 0)
3625 		return 0;
3626 
3627 	/* Write the correct value. */
3628 	pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3629 	/* Read it back. */
3630 	pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3631 	if (cacheline_size == pci_cache_line_size)
3632 		return 0;
3633 
3634 	dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not supported\n",
3635 		   pci_cache_line_size << 2);
3636 
3637 	return -EINVAL;
3638 }
3639 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3640 
3641 /**
3642  * pci_set_mwi - enables memory-write-invalidate PCI transaction
3643  * @dev: the PCI device for which MWI is enabled
3644  *
3645  * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3646  *
3647  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3648  */
3649 int pci_set_mwi(struct pci_dev *dev)
3650 {
3651 #ifdef PCI_DISABLE_MWI
3652 	return 0;
3653 #else
3654 	int rc;
3655 	u16 cmd;
3656 
3657 	rc = pci_set_cacheline_size(dev);
3658 	if (rc)
3659 		return rc;
3660 
3661 	pci_read_config_word(dev, PCI_COMMAND, &cmd);
3662 	if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3663 		dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
3664 		cmd |= PCI_COMMAND_INVALIDATE;
3665 		pci_write_config_word(dev, PCI_COMMAND, cmd);
3666 	}
3667 	return 0;
3668 #endif
3669 }
3670 EXPORT_SYMBOL(pci_set_mwi);
3671 
3672 /**
3673  * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3674  * @dev: the PCI device for which MWI is enabled
3675  *
3676  * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3677  * Callers are not required to check the return value.
3678  *
3679  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3680  */
3681 int pci_try_set_mwi(struct pci_dev *dev)
3682 {
3683 #ifdef PCI_DISABLE_MWI
3684 	return 0;
3685 #else
3686 	return pci_set_mwi(dev);
3687 #endif
3688 }
3689 EXPORT_SYMBOL(pci_try_set_mwi);
3690 
3691 /**
3692  * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3693  * @dev: the PCI device to disable
3694  *
3695  * Disables PCI Memory-Write-Invalidate transaction on the device
3696  */
3697 void pci_clear_mwi(struct pci_dev *dev)
3698 {
3699 #ifndef PCI_DISABLE_MWI
3700 	u16 cmd;
3701 
3702 	pci_read_config_word(dev, PCI_COMMAND, &cmd);
3703 	if (cmd & PCI_COMMAND_INVALIDATE) {
3704 		cmd &= ~PCI_COMMAND_INVALIDATE;
3705 		pci_write_config_word(dev, PCI_COMMAND, cmd);
3706 	}
3707 #endif
3708 }
3709 EXPORT_SYMBOL(pci_clear_mwi);
3710 
3711 /**
3712  * pci_intx - enables/disables PCI INTx for device dev
3713  * @pdev: the PCI device to operate on
3714  * @enable: boolean: whether to enable or disable PCI INTx
3715  *
3716  * Enables/disables PCI INTx for device dev
3717  */
3718 void pci_intx(struct pci_dev *pdev, int enable)
3719 {
3720 	u16 pci_command, new;
3721 
3722 	pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3723 
3724 	if (enable)
3725 		new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3726 	else
3727 		new = pci_command | PCI_COMMAND_INTX_DISABLE;
3728 
3729 	if (new != pci_command) {
3730 		struct pci_devres *dr;
3731 
3732 		pci_write_config_word(pdev, PCI_COMMAND, new);
3733 
3734 		dr = find_pci_dr(pdev);
3735 		if (dr && !dr->restore_intx) {
3736 			dr->restore_intx = 1;
3737 			dr->orig_intx = !enable;
3738 		}
3739 	}
3740 }
3741 EXPORT_SYMBOL_GPL(pci_intx);
3742 
3743 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3744 {
3745 	struct pci_bus *bus = dev->bus;
3746 	bool mask_updated = true;
3747 	u32 cmd_status_dword;
3748 	u16 origcmd, newcmd;
3749 	unsigned long flags;
3750 	bool irq_pending;
3751 
3752 	/*
3753 	 * We do a single dword read to retrieve both command and status.
3754 	 * Document assumptions that make this possible.
3755 	 */
3756 	BUILD_BUG_ON(PCI_COMMAND % 4);
3757 	BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3758 
3759 	raw_spin_lock_irqsave(&pci_lock, flags);
3760 
3761 	bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3762 
3763 	irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3764 
3765 	/*
3766 	 * Check interrupt status register to see whether our device
3767 	 * triggered the interrupt (when masking) or the next IRQ is
3768 	 * already pending (when unmasking).
3769 	 */
3770 	if (mask != irq_pending) {
3771 		mask_updated = false;
3772 		goto done;
3773 	}
3774 
3775 	origcmd = cmd_status_dword;
3776 	newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3777 	if (mask)
3778 		newcmd |= PCI_COMMAND_INTX_DISABLE;
3779 	if (newcmd != origcmd)
3780 		bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3781 
3782 done:
3783 	raw_spin_unlock_irqrestore(&pci_lock, flags);
3784 
3785 	return mask_updated;
3786 }
3787 
3788 /**
3789  * pci_check_and_mask_intx - mask INTx on pending interrupt
3790  * @dev: the PCI device to operate on
3791  *
3792  * Check if the device dev has its INTx line asserted, mask it and
3793  * return true in that case. False is returned if no interrupt was
3794  * pending.
3795  */
3796 bool pci_check_and_mask_intx(struct pci_dev *dev)
3797 {
3798 	return pci_check_and_set_intx_mask(dev, true);
3799 }
3800 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3801 
3802 /**
3803  * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3804  * @dev: the PCI device to operate on
3805  *
3806  * Check if the device dev has its INTx line asserted, unmask it if not
3807  * and return true. False is returned and the mask remains active if
3808  * there was still an interrupt pending.
3809  */
3810 bool pci_check_and_unmask_intx(struct pci_dev *dev)
3811 {
3812 	return pci_check_and_set_intx_mask(dev, false);
3813 }
3814 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3815 
3816 /**
3817  * pci_wait_for_pending_transaction - waits for pending transaction
3818  * @dev: the PCI device to operate on
3819  *
3820  * Return 0 if transaction is pending 1 otherwise.
3821  */
3822 int pci_wait_for_pending_transaction(struct pci_dev *dev)
3823 {
3824 	if (!pci_is_pcie(dev))
3825 		return 1;
3826 
3827 	return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3828 				    PCI_EXP_DEVSTA_TRPND);
3829 }
3830 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3831 
3832 static void pci_flr_wait(struct pci_dev *dev)
3833 {
3834 	int delay = 1, timeout = 60000;
3835 	u32 id;
3836 
3837 	/*
3838 	 * Per PCIe r3.1, sec 6.6.2, a device must complete an FLR within
3839 	 * 100ms, but may silently discard requests while the FLR is in
3840 	 * progress.  Wait 100ms before trying to access the device.
3841 	 */
3842 	msleep(100);
3843 
3844 	/*
3845 	 * After 100ms, the device should not silently discard config
3846 	 * requests, but it may still indicate that it needs more time by
3847 	 * responding to them with CRS completions.  The Root Port will
3848 	 * generally synthesize ~0 data to complete the read (except when
3849 	 * CRS SV is enabled and the read was for the Vendor ID; in that
3850 	 * case it synthesizes 0x0001 data).
3851 	 *
3852 	 * Wait for the device to return a non-CRS completion.  Read the
3853 	 * Command register instead of Vendor ID so we don't have to
3854 	 * contend with the CRS SV value.
3855 	 */
3856 	pci_read_config_dword(dev, PCI_COMMAND, &id);
3857 	while (id == ~0) {
3858 		if (delay > timeout) {
3859 			dev_warn(&dev->dev, "not ready %dms after FLR; giving up\n",
3860 				 100 + delay - 1);
3861 			return;
3862 		}
3863 
3864 		if (delay > 1000)
3865 			dev_info(&dev->dev, "not ready %dms after FLR; waiting\n",
3866 				 100 + delay - 1);
3867 
3868 		msleep(delay);
3869 		delay *= 2;
3870 		pci_read_config_dword(dev, PCI_COMMAND, &id);
3871 	}
3872 
3873 	if (delay > 1000)
3874 		dev_info(&dev->dev, "ready %dms after FLR\n", 100 + delay - 1);
3875 }
3876 
3877 /**
3878  * pcie_has_flr - check if a device supports function level resets
3879  * @dev:	device to check
3880  *
3881  * Returns true if the device advertises support for PCIe function level
3882  * resets.
3883  */
3884 static bool pcie_has_flr(struct pci_dev *dev)
3885 {
3886 	u32 cap;
3887 
3888 	if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
3889 		return false;
3890 
3891 	pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
3892 	return cap & PCI_EXP_DEVCAP_FLR;
3893 }
3894 
3895 /**
3896  * pcie_flr - initiate a PCIe function level reset
3897  * @dev:	device to reset
3898  *
3899  * Initiate a function level reset on @dev.  The caller should ensure the
3900  * device supports FLR before calling this function, e.g. by using the
3901  * pcie_has_flr() helper.
3902  */
3903 void pcie_flr(struct pci_dev *dev)
3904 {
3905 	if (!pci_wait_for_pending_transaction(dev))
3906 		dev_err(&dev->dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
3907 
3908 	pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
3909 	pci_flr_wait(dev);
3910 }
3911 EXPORT_SYMBOL_GPL(pcie_flr);
3912 
3913 static int pci_af_flr(struct pci_dev *dev, int probe)
3914 {
3915 	int pos;
3916 	u8 cap;
3917 
3918 	pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3919 	if (!pos)
3920 		return -ENOTTY;
3921 
3922 	if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
3923 		return -ENOTTY;
3924 
3925 	pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3926 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3927 		return -ENOTTY;
3928 
3929 	if (probe)
3930 		return 0;
3931 
3932 	/*
3933 	 * Wait for Transaction Pending bit to clear.  A word-aligned test
3934 	 * is used, so we use the conrol offset rather than status and shift
3935 	 * the test bit to match.
3936 	 */
3937 	if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
3938 				 PCI_AF_STATUS_TP << 8))
3939 		dev_err(&dev->dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
3940 
3941 	pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3942 	pci_flr_wait(dev);
3943 	return 0;
3944 }
3945 
3946 /**
3947  * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3948  * @dev: Device to reset.
3949  * @probe: If set, only check if the device can be reset this way.
3950  *
3951  * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3952  * unset, it will be reinitialized internally when going from PCI_D3hot to
3953  * PCI_D0.  If that's the case and the device is not in a low-power state
3954  * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3955  *
3956  * NOTE: This causes the caller to sleep for twice the device power transition
3957  * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3958  * by default (i.e. unless the @dev's d3_delay field has a different value).
3959  * Moreover, only devices in D0 can be reset by this function.
3960  */
3961 static int pci_pm_reset(struct pci_dev *dev, int probe)
3962 {
3963 	u16 csr;
3964 
3965 	if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
3966 		return -ENOTTY;
3967 
3968 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3969 	if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3970 		return -ENOTTY;
3971 
3972 	if (probe)
3973 		return 0;
3974 
3975 	if (dev->current_state != PCI_D0)
3976 		return -EINVAL;
3977 
3978 	csr &= ~PCI_PM_CTRL_STATE_MASK;
3979 	csr |= PCI_D3hot;
3980 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3981 	pci_dev_d3_sleep(dev);
3982 
3983 	csr &= ~PCI_PM_CTRL_STATE_MASK;
3984 	csr |= PCI_D0;
3985 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3986 	pci_dev_d3_sleep(dev);
3987 
3988 	return 0;
3989 }
3990 
3991 void pci_reset_secondary_bus(struct pci_dev *dev)
3992 {
3993 	u16 ctrl;
3994 
3995 	pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
3996 	ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
3997 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3998 	/*
3999 	 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms.  Double
4000 	 * this to 2ms to ensure that we meet the minimum requirement.
4001 	 */
4002 	msleep(2);
4003 
4004 	ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4005 	pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4006 
4007 	/*
4008 	 * Trhfa for conventional PCI is 2^25 clock cycles.
4009 	 * Assuming a minimum 33MHz clock this results in a 1s
4010 	 * delay before we can consider subordinate devices to
4011 	 * be re-initialized.  PCIe has some ways to shorten this,
4012 	 * but we don't make use of them yet.
4013 	 */
4014 	ssleep(1);
4015 }
4016 
4017 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4018 {
4019 	pci_reset_secondary_bus(dev);
4020 }
4021 
4022 /**
4023  * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
4024  * @dev: Bridge device
4025  *
4026  * Use the bridge control register to assert reset on the secondary bus.
4027  * Devices on the secondary bus are left in power-on state.
4028  */
4029 void pci_reset_bridge_secondary_bus(struct pci_dev *dev)
4030 {
4031 	pcibios_reset_secondary_bus(dev);
4032 }
4033 EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
4034 
4035 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4036 {
4037 	struct pci_dev *pdev;
4038 
4039 	if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4040 	    !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4041 		return -ENOTTY;
4042 
4043 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4044 		if (pdev != dev)
4045 			return -ENOTTY;
4046 
4047 	if (probe)
4048 		return 0;
4049 
4050 	pci_reset_bridge_secondary_bus(dev->bus->self);
4051 
4052 	return 0;
4053 }
4054 
4055 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4056 {
4057 	int rc = -ENOTTY;
4058 
4059 	if (!hotplug || !try_module_get(hotplug->ops->owner))
4060 		return rc;
4061 
4062 	if (hotplug->ops->reset_slot)
4063 		rc = hotplug->ops->reset_slot(hotplug, probe);
4064 
4065 	module_put(hotplug->ops->owner);
4066 
4067 	return rc;
4068 }
4069 
4070 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4071 {
4072 	struct pci_dev *pdev;
4073 
4074 	if (dev->subordinate || !dev->slot ||
4075 	    dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4076 		return -ENOTTY;
4077 
4078 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4079 		if (pdev != dev && pdev->slot == dev->slot)
4080 			return -ENOTTY;
4081 
4082 	return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4083 }
4084 
4085 static void pci_dev_lock(struct pci_dev *dev)
4086 {
4087 	pci_cfg_access_lock(dev);
4088 	/* block PM suspend, driver probe, etc. */
4089 	device_lock(&dev->dev);
4090 }
4091 
4092 /* Return 1 on successful lock, 0 on contention */
4093 static int pci_dev_trylock(struct pci_dev *dev)
4094 {
4095 	if (pci_cfg_access_trylock(dev)) {
4096 		if (device_trylock(&dev->dev))
4097 			return 1;
4098 		pci_cfg_access_unlock(dev);
4099 	}
4100 
4101 	return 0;
4102 }
4103 
4104 static void pci_dev_unlock(struct pci_dev *dev)
4105 {
4106 	device_unlock(&dev->dev);
4107 	pci_cfg_access_unlock(dev);
4108 }
4109 
4110 static void pci_dev_save_and_disable(struct pci_dev *dev)
4111 {
4112 	const struct pci_error_handlers *err_handler =
4113 			dev->driver ? dev->driver->err_handler : NULL;
4114 
4115 	/*
4116 	 * dev->driver->err_handler->reset_prepare() is protected against
4117 	 * races with ->remove() by the device lock, which must be held by
4118 	 * the caller.
4119 	 */
4120 	if (err_handler && err_handler->reset_prepare)
4121 		err_handler->reset_prepare(dev);
4122 
4123 	/*
4124 	 * Wake-up device prior to save.  PM registers default to D0 after
4125 	 * reset and a simple register restore doesn't reliably return
4126 	 * to a non-D0 state anyway.
4127 	 */
4128 	pci_set_power_state(dev, PCI_D0);
4129 
4130 	pci_save_state(dev);
4131 	/*
4132 	 * Disable the device by clearing the Command register, except for
4133 	 * INTx-disable which is set.  This not only disables MMIO and I/O port
4134 	 * BARs, but also prevents the device from being Bus Master, preventing
4135 	 * DMA from the device including MSI/MSI-X interrupts.  For PCI 2.3
4136 	 * compliant devices, INTx-disable prevents legacy interrupts.
4137 	 */
4138 	pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4139 }
4140 
4141 static void pci_dev_restore(struct pci_dev *dev)
4142 {
4143 	const struct pci_error_handlers *err_handler =
4144 			dev->driver ? dev->driver->err_handler : NULL;
4145 
4146 	pci_restore_state(dev);
4147 
4148 	/*
4149 	 * dev->driver->err_handler->reset_done() is protected against
4150 	 * races with ->remove() by the device lock, which must be held by
4151 	 * the caller.
4152 	 */
4153 	if (err_handler && err_handler->reset_done)
4154 		err_handler->reset_done(dev);
4155 }
4156 
4157 /**
4158  * __pci_reset_function - reset a PCI device function
4159  * @dev: PCI device to reset
4160  *
4161  * Some devices allow an individual function to be reset without affecting
4162  * other functions in the same device.  The PCI device must be responsive
4163  * to PCI config space in order to use this function.
4164  *
4165  * The device function is presumed to be unused when this function is called.
4166  * Resetting the device will make the contents of PCI configuration space
4167  * random, so any caller of this must be prepared to reinitialise the
4168  * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4169  * etc.
4170  *
4171  * Returns 0 if the device function was successfully reset or negative if the
4172  * device doesn't support resetting a single function.
4173  */
4174 int __pci_reset_function(struct pci_dev *dev)
4175 {
4176 	int ret;
4177 
4178 	pci_dev_lock(dev);
4179 	ret = __pci_reset_function_locked(dev);
4180 	pci_dev_unlock(dev);
4181 
4182 	return ret;
4183 }
4184 EXPORT_SYMBOL_GPL(__pci_reset_function);
4185 
4186 /**
4187  * __pci_reset_function_locked - reset a PCI device function while holding
4188  * the @dev mutex lock.
4189  * @dev: PCI device to reset
4190  *
4191  * Some devices allow an individual function to be reset without affecting
4192  * other functions in the same device.  The PCI device must be responsive
4193  * to PCI config space in order to use this function.
4194  *
4195  * The device function is presumed to be unused and the caller is holding
4196  * the device mutex lock when this function is called.
4197  * Resetting the device will make the contents of PCI configuration space
4198  * random, so any caller of this must be prepared to reinitialise the
4199  * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4200  * etc.
4201  *
4202  * Returns 0 if the device function was successfully reset or negative if the
4203  * device doesn't support resetting a single function.
4204  */
4205 int __pci_reset_function_locked(struct pci_dev *dev)
4206 {
4207 	int rc;
4208 
4209 	might_sleep();
4210 
4211 	rc = pci_dev_specific_reset(dev, 0);
4212 	if (rc != -ENOTTY)
4213 		return rc;
4214 	if (pcie_has_flr(dev)) {
4215 		pcie_flr(dev);
4216 		return 0;
4217 	}
4218 	rc = pci_af_flr(dev, 0);
4219 	if (rc != -ENOTTY)
4220 		return rc;
4221 	rc = pci_pm_reset(dev, 0);
4222 	if (rc != -ENOTTY)
4223 		return rc;
4224 	rc = pci_dev_reset_slot_function(dev, 0);
4225 	if (rc != -ENOTTY)
4226 		return rc;
4227 	return pci_parent_bus_reset(dev, 0);
4228 }
4229 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4230 
4231 /**
4232  * pci_probe_reset_function - check whether the device can be safely reset
4233  * @dev: PCI device to reset
4234  *
4235  * Some devices allow an individual function to be reset without affecting
4236  * other functions in the same device.  The PCI device must be responsive
4237  * to PCI config space in order to use this function.
4238  *
4239  * Returns 0 if the device function can be reset or negative if the
4240  * device doesn't support resetting a single function.
4241  */
4242 int pci_probe_reset_function(struct pci_dev *dev)
4243 {
4244 	int rc;
4245 
4246 	might_sleep();
4247 
4248 	rc = pci_dev_specific_reset(dev, 1);
4249 	if (rc != -ENOTTY)
4250 		return rc;
4251 	if (pcie_has_flr(dev))
4252 		return 0;
4253 	rc = pci_af_flr(dev, 1);
4254 	if (rc != -ENOTTY)
4255 		return rc;
4256 	rc = pci_pm_reset(dev, 1);
4257 	if (rc != -ENOTTY)
4258 		return rc;
4259 	rc = pci_dev_reset_slot_function(dev, 1);
4260 	if (rc != -ENOTTY)
4261 		return rc;
4262 
4263 	return pci_parent_bus_reset(dev, 1);
4264 }
4265 
4266 /**
4267  * pci_reset_function - quiesce and reset a PCI device function
4268  * @dev: PCI device to reset
4269  *
4270  * Some devices allow an individual function to be reset without affecting
4271  * other functions in the same device.  The PCI device must be responsive
4272  * to PCI config space in order to use this function.
4273  *
4274  * This function does not just reset the PCI portion of a device, but
4275  * clears all the state associated with the device.  This function differs
4276  * from __pci_reset_function in that it saves and restores device state
4277  * over the reset.
4278  *
4279  * Returns 0 if the device function was successfully reset or negative if the
4280  * device doesn't support resetting a single function.
4281  */
4282 int pci_reset_function(struct pci_dev *dev)
4283 {
4284 	int rc;
4285 
4286 	rc = pci_probe_reset_function(dev);
4287 	if (rc)
4288 		return rc;
4289 
4290 	pci_dev_lock(dev);
4291 	pci_dev_save_and_disable(dev);
4292 
4293 	rc = __pci_reset_function_locked(dev);
4294 
4295 	pci_dev_restore(dev);
4296 	pci_dev_unlock(dev);
4297 
4298 	return rc;
4299 }
4300 EXPORT_SYMBOL_GPL(pci_reset_function);
4301 
4302 /**
4303  * pci_reset_function_locked - quiesce and reset a PCI device function
4304  * @dev: PCI device to reset
4305  *
4306  * Some devices allow an individual function to be reset without affecting
4307  * other functions in the same device.  The PCI device must be responsive
4308  * to PCI config space in order to use this function.
4309  *
4310  * This function does not just reset the PCI portion of a device, but
4311  * clears all the state associated with the device.  This function differs
4312  * from __pci_reset_function() in that it saves and restores device state
4313  * over the reset.  It also differs from pci_reset_function() in that it
4314  * requires the PCI device lock to be held.
4315  *
4316  * Returns 0 if the device function was successfully reset or negative if the
4317  * device doesn't support resetting a single function.
4318  */
4319 int pci_reset_function_locked(struct pci_dev *dev)
4320 {
4321 	int rc;
4322 
4323 	rc = pci_probe_reset_function(dev);
4324 	if (rc)
4325 		return rc;
4326 
4327 	pci_dev_save_and_disable(dev);
4328 
4329 	rc = __pci_reset_function_locked(dev);
4330 
4331 	pci_dev_restore(dev);
4332 
4333 	return rc;
4334 }
4335 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4336 
4337 /**
4338  * pci_try_reset_function - quiesce and reset a PCI device function
4339  * @dev: PCI device to reset
4340  *
4341  * Same as above, except return -EAGAIN if unable to lock device.
4342  */
4343 int pci_try_reset_function(struct pci_dev *dev)
4344 {
4345 	int rc;
4346 
4347 	rc = pci_probe_reset_function(dev);
4348 	if (rc)
4349 		return rc;
4350 
4351 	if (!pci_dev_trylock(dev))
4352 		return -EAGAIN;
4353 
4354 	pci_dev_save_and_disable(dev);
4355 	rc = __pci_reset_function_locked(dev);
4356 	pci_dev_unlock(dev);
4357 
4358 	pci_dev_restore(dev);
4359 	return rc;
4360 }
4361 EXPORT_SYMBOL_GPL(pci_try_reset_function);
4362 
4363 /* Do any devices on or below this bus prevent a bus reset? */
4364 static bool pci_bus_resetable(struct pci_bus *bus)
4365 {
4366 	struct pci_dev *dev;
4367 
4368 	list_for_each_entry(dev, &bus->devices, bus_list) {
4369 		if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4370 		    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4371 			return false;
4372 	}
4373 
4374 	return true;
4375 }
4376 
4377 /* Lock devices from the top of the tree down */
4378 static void pci_bus_lock(struct pci_bus *bus)
4379 {
4380 	struct pci_dev *dev;
4381 
4382 	list_for_each_entry(dev, &bus->devices, bus_list) {
4383 		pci_dev_lock(dev);
4384 		if (dev->subordinate)
4385 			pci_bus_lock(dev->subordinate);
4386 	}
4387 }
4388 
4389 /* Unlock devices from the bottom of the tree up */
4390 static void pci_bus_unlock(struct pci_bus *bus)
4391 {
4392 	struct pci_dev *dev;
4393 
4394 	list_for_each_entry(dev, &bus->devices, bus_list) {
4395 		if (dev->subordinate)
4396 			pci_bus_unlock(dev->subordinate);
4397 		pci_dev_unlock(dev);
4398 	}
4399 }
4400 
4401 /* Return 1 on successful lock, 0 on contention */
4402 static int pci_bus_trylock(struct pci_bus *bus)
4403 {
4404 	struct pci_dev *dev;
4405 
4406 	list_for_each_entry(dev, &bus->devices, bus_list) {
4407 		if (!pci_dev_trylock(dev))
4408 			goto unlock;
4409 		if (dev->subordinate) {
4410 			if (!pci_bus_trylock(dev->subordinate)) {
4411 				pci_dev_unlock(dev);
4412 				goto unlock;
4413 			}
4414 		}
4415 	}
4416 	return 1;
4417 
4418 unlock:
4419 	list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4420 		if (dev->subordinate)
4421 			pci_bus_unlock(dev->subordinate);
4422 		pci_dev_unlock(dev);
4423 	}
4424 	return 0;
4425 }
4426 
4427 /* Do any devices on or below this slot prevent a bus reset? */
4428 static bool pci_slot_resetable(struct pci_slot *slot)
4429 {
4430 	struct pci_dev *dev;
4431 
4432 	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4433 		if (!dev->slot || dev->slot != slot)
4434 			continue;
4435 		if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4436 		    (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4437 			return false;
4438 	}
4439 
4440 	return true;
4441 }
4442 
4443 /* Lock devices from the top of the tree down */
4444 static void pci_slot_lock(struct pci_slot *slot)
4445 {
4446 	struct pci_dev *dev;
4447 
4448 	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4449 		if (!dev->slot || dev->slot != slot)
4450 			continue;
4451 		pci_dev_lock(dev);
4452 		if (dev->subordinate)
4453 			pci_bus_lock(dev->subordinate);
4454 	}
4455 }
4456 
4457 /* Unlock devices from the bottom of the tree up */
4458 static void pci_slot_unlock(struct pci_slot *slot)
4459 {
4460 	struct pci_dev *dev;
4461 
4462 	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4463 		if (!dev->slot || dev->slot != slot)
4464 			continue;
4465 		if (dev->subordinate)
4466 			pci_bus_unlock(dev->subordinate);
4467 		pci_dev_unlock(dev);
4468 	}
4469 }
4470 
4471 /* Return 1 on successful lock, 0 on contention */
4472 static int pci_slot_trylock(struct pci_slot *slot)
4473 {
4474 	struct pci_dev *dev;
4475 
4476 	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4477 		if (!dev->slot || dev->slot != slot)
4478 			continue;
4479 		if (!pci_dev_trylock(dev))
4480 			goto unlock;
4481 		if (dev->subordinate) {
4482 			if (!pci_bus_trylock(dev->subordinate)) {
4483 				pci_dev_unlock(dev);
4484 				goto unlock;
4485 			}
4486 		}
4487 	}
4488 	return 1;
4489 
4490 unlock:
4491 	list_for_each_entry_continue_reverse(dev,
4492 					     &slot->bus->devices, bus_list) {
4493 		if (!dev->slot || dev->slot != slot)
4494 			continue;
4495 		if (dev->subordinate)
4496 			pci_bus_unlock(dev->subordinate);
4497 		pci_dev_unlock(dev);
4498 	}
4499 	return 0;
4500 }
4501 
4502 /* Save and disable devices from the top of the tree down */
4503 static void pci_bus_save_and_disable(struct pci_bus *bus)
4504 {
4505 	struct pci_dev *dev;
4506 
4507 	list_for_each_entry(dev, &bus->devices, bus_list) {
4508 		pci_dev_lock(dev);
4509 		pci_dev_save_and_disable(dev);
4510 		pci_dev_unlock(dev);
4511 		if (dev->subordinate)
4512 			pci_bus_save_and_disable(dev->subordinate);
4513 	}
4514 }
4515 
4516 /*
4517  * Restore devices from top of the tree down - parent bridges need to be
4518  * restored before we can get to subordinate devices.
4519  */
4520 static void pci_bus_restore(struct pci_bus *bus)
4521 {
4522 	struct pci_dev *dev;
4523 
4524 	list_for_each_entry(dev, &bus->devices, bus_list) {
4525 		pci_dev_lock(dev);
4526 		pci_dev_restore(dev);
4527 		pci_dev_unlock(dev);
4528 		if (dev->subordinate)
4529 			pci_bus_restore(dev->subordinate);
4530 	}
4531 }
4532 
4533 /* Save and disable devices from the top of the tree down */
4534 static void pci_slot_save_and_disable(struct pci_slot *slot)
4535 {
4536 	struct pci_dev *dev;
4537 
4538 	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4539 		if (!dev->slot || dev->slot != slot)
4540 			continue;
4541 		pci_dev_save_and_disable(dev);
4542 		if (dev->subordinate)
4543 			pci_bus_save_and_disable(dev->subordinate);
4544 	}
4545 }
4546 
4547 /*
4548  * Restore devices from top of the tree down - parent bridges need to be
4549  * restored before we can get to subordinate devices.
4550  */
4551 static void pci_slot_restore(struct pci_slot *slot)
4552 {
4553 	struct pci_dev *dev;
4554 
4555 	list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4556 		if (!dev->slot || dev->slot != slot)
4557 			continue;
4558 		pci_dev_restore(dev);
4559 		if (dev->subordinate)
4560 			pci_bus_restore(dev->subordinate);
4561 	}
4562 }
4563 
4564 static int pci_slot_reset(struct pci_slot *slot, int probe)
4565 {
4566 	int rc;
4567 
4568 	if (!slot || !pci_slot_resetable(slot))
4569 		return -ENOTTY;
4570 
4571 	if (!probe)
4572 		pci_slot_lock(slot);
4573 
4574 	might_sleep();
4575 
4576 	rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4577 
4578 	if (!probe)
4579 		pci_slot_unlock(slot);
4580 
4581 	return rc;
4582 }
4583 
4584 /**
4585  * pci_probe_reset_slot - probe whether a PCI slot can be reset
4586  * @slot: PCI slot to probe
4587  *
4588  * Return 0 if slot can be reset, negative if a slot reset is not supported.
4589  */
4590 int pci_probe_reset_slot(struct pci_slot *slot)
4591 {
4592 	return pci_slot_reset(slot, 1);
4593 }
4594 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4595 
4596 /**
4597  * pci_reset_slot - reset a PCI slot
4598  * @slot: PCI slot to reset
4599  *
4600  * A PCI bus may host multiple slots, each slot may support a reset mechanism
4601  * independent of other slots.  For instance, some slots may support slot power
4602  * control.  In the case of a 1:1 bus to slot architecture, this function may
4603  * wrap the bus reset to avoid spurious slot related events such as hotplug.
4604  * Generally a slot reset should be attempted before a bus reset.  All of the
4605  * function of the slot and any subordinate buses behind the slot are reset
4606  * through this function.  PCI config space of all devices in the slot and
4607  * behind the slot is saved before and restored after reset.
4608  *
4609  * Return 0 on success, non-zero on error.
4610  */
4611 int pci_reset_slot(struct pci_slot *slot)
4612 {
4613 	int rc;
4614 
4615 	rc = pci_slot_reset(slot, 1);
4616 	if (rc)
4617 		return rc;
4618 
4619 	pci_slot_save_and_disable(slot);
4620 
4621 	rc = pci_slot_reset(slot, 0);
4622 
4623 	pci_slot_restore(slot);
4624 
4625 	return rc;
4626 }
4627 EXPORT_SYMBOL_GPL(pci_reset_slot);
4628 
4629 /**
4630  * pci_try_reset_slot - Try to reset a PCI slot
4631  * @slot: PCI slot to reset
4632  *
4633  * Same as above except return -EAGAIN if the slot cannot be locked
4634  */
4635 int pci_try_reset_slot(struct pci_slot *slot)
4636 {
4637 	int rc;
4638 
4639 	rc = pci_slot_reset(slot, 1);
4640 	if (rc)
4641 		return rc;
4642 
4643 	pci_slot_save_and_disable(slot);
4644 
4645 	if (pci_slot_trylock(slot)) {
4646 		might_sleep();
4647 		rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4648 		pci_slot_unlock(slot);
4649 	} else
4650 		rc = -EAGAIN;
4651 
4652 	pci_slot_restore(slot);
4653 
4654 	return rc;
4655 }
4656 EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4657 
4658 static int pci_bus_reset(struct pci_bus *bus, int probe)
4659 {
4660 	if (!bus->self || !pci_bus_resetable(bus))
4661 		return -ENOTTY;
4662 
4663 	if (probe)
4664 		return 0;
4665 
4666 	pci_bus_lock(bus);
4667 
4668 	might_sleep();
4669 
4670 	pci_reset_bridge_secondary_bus(bus->self);
4671 
4672 	pci_bus_unlock(bus);
4673 
4674 	return 0;
4675 }
4676 
4677 /**
4678  * pci_probe_reset_bus - probe whether a PCI bus can be reset
4679  * @bus: PCI bus to probe
4680  *
4681  * Return 0 if bus can be reset, negative if a bus reset is not supported.
4682  */
4683 int pci_probe_reset_bus(struct pci_bus *bus)
4684 {
4685 	return pci_bus_reset(bus, 1);
4686 }
4687 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4688 
4689 /**
4690  * pci_reset_bus - reset a PCI bus
4691  * @bus: top level PCI bus to reset
4692  *
4693  * Do a bus reset on the given bus and any subordinate buses, saving
4694  * and restoring state of all devices.
4695  *
4696  * Return 0 on success, non-zero on error.
4697  */
4698 int pci_reset_bus(struct pci_bus *bus)
4699 {
4700 	int rc;
4701 
4702 	rc = pci_bus_reset(bus, 1);
4703 	if (rc)
4704 		return rc;
4705 
4706 	pci_bus_save_and_disable(bus);
4707 
4708 	rc = pci_bus_reset(bus, 0);
4709 
4710 	pci_bus_restore(bus);
4711 
4712 	return rc;
4713 }
4714 EXPORT_SYMBOL_GPL(pci_reset_bus);
4715 
4716 /**
4717  * pci_try_reset_bus - Try to reset a PCI bus
4718  * @bus: top level PCI bus to reset
4719  *
4720  * Same as above except return -EAGAIN if the bus cannot be locked
4721  */
4722 int pci_try_reset_bus(struct pci_bus *bus)
4723 {
4724 	int rc;
4725 
4726 	rc = pci_bus_reset(bus, 1);
4727 	if (rc)
4728 		return rc;
4729 
4730 	pci_bus_save_and_disable(bus);
4731 
4732 	if (pci_bus_trylock(bus)) {
4733 		might_sleep();
4734 		pci_reset_bridge_secondary_bus(bus->self);
4735 		pci_bus_unlock(bus);
4736 	} else
4737 		rc = -EAGAIN;
4738 
4739 	pci_bus_restore(bus);
4740 
4741 	return rc;
4742 }
4743 EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4744 
4745 /**
4746  * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4747  * @dev: PCI device to query
4748  *
4749  * Returns mmrbc: maximum designed memory read count in bytes
4750  *    or appropriate error value.
4751  */
4752 int pcix_get_max_mmrbc(struct pci_dev *dev)
4753 {
4754 	int cap;
4755 	u32 stat;
4756 
4757 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4758 	if (!cap)
4759 		return -EINVAL;
4760 
4761 	if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4762 		return -EINVAL;
4763 
4764 	return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4765 }
4766 EXPORT_SYMBOL(pcix_get_max_mmrbc);
4767 
4768 /**
4769  * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4770  * @dev: PCI device to query
4771  *
4772  * Returns mmrbc: maximum memory read count in bytes
4773  *    or appropriate error value.
4774  */
4775 int pcix_get_mmrbc(struct pci_dev *dev)
4776 {
4777 	int cap;
4778 	u16 cmd;
4779 
4780 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4781 	if (!cap)
4782 		return -EINVAL;
4783 
4784 	if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4785 		return -EINVAL;
4786 
4787 	return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4788 }
4789 EXPORT_SYMBOL(pcix_get_mmrbc);
4790 
4791 /**
4792  * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4793  * @dev: PCI device to query
4794  * @mmrbc: maximum memory read count in bytes
4795  *    valid values are 512, 1024, 2048, 4096
4796  *
4797  * If possible sets maximum memory read byte count, some bridges have erratas
4798  * that prevent this.
4799  */
4800 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4801 {
4802 	int cap;
4803 	u32 stat, v, o;
4804 	u16 cmd;
4805 
4806 	if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4807 		return -EINVAL;
4808 
4809 	v = ffs(mmrbc) - 10;
4810 
4811 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4812 	if (!cap)
4813 		return -EINVAL;
4814 
4815 	if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4816 		return -EINVAL;
4817 
4818 	if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4819 		return -E2BIG;
4820 
4821 	if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4822 		return -EINVAL;
4823 
4824 	o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4825 	if (o != v) {
4826 		if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4827 			return -EIO;
4828 
4829 		cmd &= ~PCI_X_CMD_MAX_READ;
4830 		cmd |= v << 2;
4831 		if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4832 			return -EIO;
4833 	}
4834 	return 0;
4835 }
4836 EXPORT_SYMBOL(pcix_set_mmrbc);
4837 
4838 /**
4839  * pcie_get_readrq - get PCI Express read request size
4840  * @dev: PCI device to query
4841  *
4842  * Returns maximum memory read request in bytes
4843  *    or appropriate error value.
4844  */
4845 int pcie_get_readrq(struct pci_dev *dev)
4846 {
4847 	u16 ctl;
4848 
4849 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4850 
4851 	return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
4852 }
4853 EXPORT_SYMBOL(pcie_get_readrq);
4854 
4855 /**
4856  * pcie_set_readrq - set PCI Express maximum memory read request
4857  * @dev: PCI device to query
4858  * @rq: maximum memory read count in bytes
4859  *    valid values are 128, 256, 512, 1024, 2048, 4096
4860  *
4861  * If possible sets maximum memory read request in bytes
4862  */
4863 int pcie_set_readrq(struct pci_dev *dev, int rq)
4864 {
4865 	u16 v;
4866 
4867 	if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
4868 		return -EINVAL;
4869 
4870 	/*
4871 	 * If using the "performance" PCIe config, we clamp the
4872 	 * read rq size to the max packet size to prevent the
4873 	 * host bridge generating requests larger than we can
4874 	 * cope with
4875 	 */
4876 	if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
4877 		int mps = pcie_get_mps(dev);
4878 
4879 		if (mps < rq)
4880 			rq = mps;
4881 	}
4882 
4883 	v = (ffs(rq) - 8) << 12;
4884 
4885 	return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4886 						  PCI_EXP_DEVCTL_READRQ, v);
4887 }
4888 EXPORT_SYMBOL(pcie_set_readrq);
4889 
4890 /**
4891  * pcie_get_mps - get PCI Express maximum payload size
4892  * @dev: PCI device to query
4893  *
4894  * Returns maximum payload size in bytes
4895  */
4896 int pcie_get_mps(struct pci_dev *dev)
4897 {
4898 	u16 ctl;
4899 
4900 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4901 
4902 	return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
4903 }
4904 EXPORT_SYMBOL(pcie_get_mps);
4905 
4906 /**
4907  * pcie_set_mps - set PCI Express maximum payload size
4908  * @dev: PCI device to query
4909  * @mps: maximum payload size in bytes
4910  *    valid values are 128, 256, 512, 1024, 2048, 4096
4911  *
4912  * If possible sets maximum payload size
4913  */
4914 int pcie_set_mps(struct pci_dev *dev, int mps)
4915 {
4916 	u16 v;
4917 
4918 	if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
4919 		return -EINVAL;
4920 
4921 	v = ffs(mps) - 8;
4922 	if (v > dev->pcie_mpss)
4923 		return -EINVAL;
4924 	v <<= 5;
4925 
4926 	return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4927 						  PCI_EXP_DEVCTL_PAYLOAD, v);
4928 }
4929 EXPORT_SYMBOL(pcie_set_mps);
4930 
4931 /**
4932  * pcie_get_minimum_link - determine minimum link settings of a PCI device
4933  * @dev: PCI device to query
4934  * @speed: storage for minimum speed
4935  * @width: storage for minimum width
4936  *
4937  * This function will walk up the PCI device chain and determine the minimum
4938  * link width and speed of the device.
4939  */
4940 int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
4941 			  enum pcie_link_width *width)
4942 {
4943 	int ret;
4944 
4945 	*speed = PCI_SPEED_UNKNOWN;
4946 	*width = PCIE_LNK_WIDTH_UNKNOWN;
4947 
4948 	while (dev) {
4949 		u16 lnksta;
4950 		enum pci_bus_speed next_speed;
4951 		enum pcie_link_width next_width;
4952 
4953 		ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
4954 		if (ret)
4955 			return ret;
4956 
4957 		next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
4958 		next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
4959 			PCI_EXP_LNKSTA_NLW_SHIFT;
4960 
4961 		if (next_speed < *speed)
4962 			*speed = next_speed;
4963 
4964 		if (next_width < *width)
4965 			*width = next_width;
4966 
4967 		dev = dev->bus->self;
4968 	}
4969 
4970 	return 0;
4971 }
4972 EXPORT_SYMBOL(pcie_get_minimum_link);
4973 
4974 /**
4975  * pci_select_bars - Make BAR mask from the type of resource
4976  * @dev: the PCI device for which BAR mask is made
4977  * @flags: resource type mask to be selected
4978  *
4979  * This helper routine makes bar mask from the type of resource.
4980  */
4981 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
4982 {
4983 	int i, bars = 0;
4984 	for (i = 0; i < PCI_NUM_RESOURCES; i++)
4985 		if (pci_resource_flags(dev, i) & flags)
4986 			bars |= (1 << i);
4987 	return bars;
4988 }
4989 EXPORT_SYMBOL(pci_select_bars);
4990 
4991 /* Some architectures require additional programming to enable VGA */
4992 static arch_set_vga_state_t arch_set_vga_state;
4993 
4994 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
4995 {
4996 	arch_set_vga_state = func;	/* NULL disables */
4997 }
4998 
4999 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5000 				  unsigned int command_bits, u32 flags)
5001 {
5002 	if (arch_set_vga_state)
5003 		return arch_set_vga_state(dev, decode, command_bits,
5004 						flags);
5005 	return 0;
5006 }
5007 
5008 /**
5009  * pci_set_vga_state - set VGA decode state on device and parents if requested
5010  * @dev: the PCI device
5011  * @decode: true = enable decoding, false = disable decoding
5012  * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5013  * @flags: traverse ancestors and change bridges
5014  * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5015  */
5016 int pci_set_vga_state(struct pci_dev *dev, bool decode,
5017 		      unsigned int command_bits, u32 flags)
5018 {
5019 	struct pci_bus *bus;
5020 	struct pci_dev *bridge;
5021 	u16 cmd;
5022 	int rc;
5023 
5024 	WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5025 
5026 	/* ARCH specific VGA enables */
5027 	rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5028 	if (rc)
5029 		return rc;
5030 
5031 	if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5032 		pci_read_config_word(dev, PCI_COMMAND, &cmd);
5033 		if (decode == true)
5034 			cmd |= command_bits;
5035 		else
5036 			cmd &= ~command_bits;
5037 		pci_write_config_word(dev, PCI_COMMAND, cmd);
5038 	}
5039 
5040 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5041 		return 0;
5042 
5043 	bus = dev->bus;
5044 	while (bus) {
5045 		bridge = bus->self;
5046 		if (bridge) {
5047 			pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5048 					     &cmd);
5049 			if (decode == true)
5050 				cmd |= PCI_BRIDGE_CTL_VGA;
5051 			else
5052 				cmd &= ~PCI_BRIDGE_CTL_VGA;
5053 			pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5054 					      cmd);
5055 		}
5056 		bus = bus->parent;
5057 	}
5058 	return 0;
5059 }
5060 
5061 /**
5062  * pci_add_dma_alias - Add a DMA devfn alias for a device
5063  * @dev: the PCI device for which alias is added
5064  * @devfn: alias slot and function
5065  *
5066  * This helper encodes 8-bit devfn as bit number in dma_alias_mask.
5067  * It should be called early, preferably as PCI fixup header quirk.
5068  */
5069 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5070 {
5071 	if (!dev->dma_alias_mask)
5072 		dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
5073 					      sizeof(long), GFP_KERNEL);
5074 	if (!dev->dma_alias_mask) {
5075 		dev_warn(&dev->dev, "Unable to allocate DMA alias mask\n");
5076 		return;
5077 	}
5078 
5079 	set_bit(devfn, dev->dma_alias_mask);
5080 	dev_info(&dev->dev, "Enabling fixed DMA alias to %02x.%d\n",
5081 		 PCI_SLOT(devfn), PCI_FUNC(devfn));
5082 }
5083 
5084 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5085 {
5086 	return (dev1->dma_alias_mask &&
5087 		test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5088 	       (dev2->dma_alias_mask &&
5089 		test_bit(dev1->devfn, dev2->dma_alias_mask));
5090 }
5091 
5092 bool pci_device_is_present(struct pci_dev *pdev)
5093 {
5094 	u32 v;
5095 
5096 	if (pci_dev_is_disconnected(pdev))
5097 		return false;
5098 	return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5099 }
5100 EXPORT_SYMBOL_GPL(pci_device_is_present);
5101 
5102 void pci_ignore_hotplug(struct pci_dev *dev)
5103 {
5104 	struct pci_dev *bridge = dev->bus->self;
5105 
5106 	dev->ignore_hotplug = 1;
5107 	/* Propagate the "ignore hotplug" setting to the parent bridge. */
5108 	if (bridge)
5109 		bridge->ignore_hotplug = 1;
5110 }
5111 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5112 
5113 resource_size_t __weak pcibios_default_alignment(void)
5114 {
5115 	return 0;
5116 }
5117 
5118 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5119 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5120 static DEFINE_SPINLOCK(resource_alignment_lock);
5121 
5122 /**
5123  * pci_specified_resource_alignment - get resource alignment specified by user.
5124  * @dev: the PCI device to get
5125  * @resize: whether or not to change resources' size when reassigning alignment
5126  *
5127  * RETURNS: Resource alignment if it is specified.
5128  *          Zero if it is not specified.
5129  */
5130 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5131 							bool *resize)
5132 {
5133 	int seg, bus, slot, func, align_order, count;
5134 	unsigned short vendor, device, subsystem_vendor, subsystem_device;
5135 	resource_size_t align = pcibios_default_alignment();
5136 	char *p;
5137 
5138 	spin_lock(&resource_alignment_lock);
5139 	p = resource_alignment_param;
5140 	if (!*p && !align)
5141 		goto out;
5142 	if (pci_has_flag(PCI_PROBE_ONLY)) {
5143 		align = 0;
5144 		pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5145 		goto out;
5146 	}
5147 
5148 	while (*p) {
5149 		count = 0;
5150 		if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5151 							p[count] == '@') {
5152 			p += count + 1;
5153 		} else {
5154 			align_order = -1;
5155 		}
5156 		if (strncmp(p, "pci:", 4) == 0) {
5157 			/* PCI vendor/device (subvendor/subdevice) ids are specified */
5158 			p += 4;
5159 			if (sscanf(p, "%hx:%hx:%hx:%hx%n",
5160 				&vendor, &device, &subsystem_vendor, &subsystem_device, &count) != 4) {
5161 				if (sscanf(p, "%hx:%hx%n", &vendor, &device, &count) != 2) {
5162 					printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: pci:%s\n",
5163 						p);
5164 					break;
5165 				}
5166 				subsystem_vendor = subsystem_device = 0;
5167 			}
5168 			p += count;
5169 			if ((!vendor || (vendor == dev->vendor)) &&
5170 				(!device || (device == dev->device)) &&
5171 				(!subsystem_vendor || (subsystem_vendor == dev->subsystem_vendor)) &&
5172 				(!subsystem_device || (subsystem_device == dev->subsystem_device))) {
5173 				*resize = true;
5174 				if (align_order == -1)
5175 					align = PAGE_SIZE;
5176 				else
5177 					align = 1 << align_order;
5178 				/* Found */
5179 				break;
5180 			}
5181 		}
5182 		else {
5183 			if (sscanf(p, "%x:%x:%x.%x%n",
5184 				&seg, &bus, &slot, &func, &count) != 4) {
5185 				seg = 0;
5186 				if (sscanf(p, "%x:%x.%x%n",
5187 						&bus, &slot, &func, &count) != 3) {
5188 					/* Invalid format */
5189 					printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
5190 						p);
5191 					break;
5192 				}
5193 			}
5194 			p += count;
5195 			if (seg == pci_domain_nr(dev->bus) &&
5196 				bus == dev->bus->number &&
5197 				slot == PCI_SLOT(dev->devfn) &&
5198 				func == PCI_FUNC(dev->devfn)) {
5199 				*resize = true;
5200 				if (align_order == -1)
5201 					align = PAGE_SIZE;
5202 				else
5203 					align = 1 << align_order;
5204 				/* Found */
5205 				break;
5206 			}
5207 		}
5208 		if (*p != ';' && *p != ',') {
5209 			/* End of param or invalid format */
5210 			break;
5211 		}
5212 		p++;
5213 	}
5214 out:
5215 	spin_unlock(&resource_alignment_lock);
5216 	return align;
5217 }
5218 
5219 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5220 					   resource_size_t align, bool resize)
5221 {
5222 	struct resource *r = &dev->resource[bar];
5223 	resource_size_t size;
5224 
5225 	if (!(r->flags & IORESOURCE_MEM))
5226 		return;
5227 
5228 	if (r->flags & IORESOURCE_PCI_FIXED) {
5229 		dev_info(&dev->dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5230 			 bar, r, (unsigned long long)align);
5231 		return;
5232 	}
5233 
5234 	size = resource_size(r);
5235 	if (size >= align)
5236 		return;
5237 
5238 	/*
5239 	 * Increase the alignment of the resource.  There are two ways we
5240 	 * can do this:
5241 	 *
5242 	 * 1) Increase the size of the resource.  BARs are aligned on their
5243 	 *    size, so when we reallocate space for this resource, we'll
5244 	 *    allocate it with the larger alignment.  This also prevents
5245 	 *    assignment of any other BARs inside the alignment region, so
5246 	 *    if we're requesting page alignment, this means no other BARs
5247 	 *    will share the page.
5248 	 *
5249 	 *    The disadvantage is that this makes the resource larger than
5250 	 *    the hardware BAR, which may break drivers that compute things
5251 	 *    based on the resource size, e.g., to find registers at a
5252 	 *    fixed offset before the end of the BAR.
5253 	 *
5254 	 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5255 	 *    set r->start to the desired alignment.  By itself this
5256 	 *    doesn't prevent other BARs being put inside the alignment
5257 	 *    region, but if we realign *every* resource of every device in
5258 	 *    the system, none of them will share an alignment region.
5259 	 *
5260 	 * When the user has requested alignment for only some devices via
5261 	 * the "pci=resource_alignment" argument, "resize" is true and we
5262 	 * use the first method.  Otherwise we assume we're aligning all
5263 	 * devices and we use the second.
5264 	 */
5265 
5266 	dev_info(&dev->dev, "BAR%d %pR: requesting alignment to %#llx\n",
5267 		 bar, r, (unsigned long long)align);
5268 
5269 	if (resize) {
5270 		r->start = 0;
5271 		r->end = align - 1;
5272 	} else {
5273 		r->flags &= ~IORESOURCE_SIZEALIGN;
5274 		r->flags |= IORESOURCE_STARTALIGN;
5275 		r->start = align;
5276 		r->end = r->start + size - 1;
5277 	}
5278 	r->flags |= IORESOURCE_UNSET;
5279 }
5280 
5281 /*
5282  * This function disables memory decoding and releases memory resources
5283  * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5284  * It also rounds up size to specified alignment.
5285  * Later on, the kernel will assign page-aligned memory resource back
5286  * to the device.
5287  */
5288 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5289 {
5290 	int i;
5291 	struct resource *r;
5292 	resource_size_t align;
5293 	u16 command;
5294 	bool resize = false;
5295 
5296 	/*
5297 	 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
5298 	 * 3.4.1.11.  Their resources are allocated from the space
5299 	 * described by the VF BARx register in the PF's SR-IOV capability.
5300 	 * We can't influence their alignment here.
5301 	 */
5302 	if (dev->is_virtfn)
5303 		return;
5304 
5305 	/* check if specified PCI is target device to reassign */
5306 	align = pci_specified_resource_alignment(dev, &resize);
5307 	if (!align)
5308 		return;
5309 
5310 	if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5311 	    (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5312 		dev_warn(&dev->dev,
5313 			"Can't reassign resources to host bridge.\n");
5314 		return;
5315 	}
5316 
5317 	dev_info(&dev->dev,
5318 		"Disabling memory decoding and releasing memory resources.\n");
5319 	pci_read_config_word(dev, PCI_COMMAND, &command);
5320 	command &= ~PCI_COMMAND_MEMORY;
5321 	pci_write_config_word(dev, PCI_COMMAND, command);
5322 
5323 	for (i = 0; i <= PCI_ROM_RESOURCE; i++)
5324 		pci_request_resource_alignment(dev, i, align, resize);
5325 
5326 	/*
5327 	 * Need to disable bridge's resource window,
5328 	 * to enable the kernel to reassign new resource
5329 	 * window later on.
5330 	 */
5331 	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5332 	    (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5333 		for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5334 			r = &dev->resource[i];
5335 			if (!(r->flags & IORESOURCE_MEM))
5336 				continue;
5337 			r->flags |= IORESOURCE_UNSET;
5338 			r->end = resource_size(r) - 1;
5339 			r->start = 0;
5340 		}
5341 		pci_disable_bridge_window(dev);
5342 	}
5343 }
5344 
5345 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5346 {
5347 	if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
5348 		count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
5349 	spin_lock(&resource_alignment_lock);
5350 	strncpy(resource_alignment_param, buf, count);
5351 	resource_alignment_param[count] = '\0';
5352 	spin_unlock(&resource_alignment_lock);
5353 	return count;
5354 }
5355 
5356 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
5357 {
5358 	size_t count;
5359 	spin_lock(&resource_alignment_lock);
5360 	count = snprintf(buf, size, "%s", resource_alignment_param);
5361 	spin_unlock(&resource_alignment_lock);
5362 	return count;
5363 }
5364 
5365 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
5366 {
5367 	return pci_get_resource_alignment_param(buf, PAGE_SIZE);
5368 }
5369 
5370 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
5371 					const char *buf, size_t count)
5372 {
5373 	return pci_set_resource_alignment_param(buf, count);
5374 }
5375 
5376 static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
5377 					pci_resource_alignment_store);
5378 
5379 static int __init pci_resource_alignment_sysfs_init(void)
5380 {
5381 	return bus_create_file(&pci_bus_type,
5382 					&bus_attr_resource_alignment);
5383 }
5384 late_initcall(pci_resource_alignment_sysfs_init);
5385 
5386 static void pci_no_domains(void)
5387 {
5388 #ifdef CONFIG_PCI_DOMAINS
5389 	pci_domains_supported = 0;
5390 #endif
5391 }
5392 
5393 #ifdef CONFIG_PCI_DOMAINS
5394 static atomic_t __domain_nr = ATOMIC_INIT(-1);
5395 
5396 int pci_get_new_domain_nr(void)
5397 {
5398 	return atomic_inc_return(&__domain_nr);
5399 }
5400 
5401 #ifdef CONFIG_PCI_DOMAINS_GENERIC
5402 static int of_pci_bus_find_domain_nr(struct device *parent)
5403 {
5404 	static int use_dt_domains = -1;
5405 	int domain = -1;
5406 
5407 	if (parent)
5408 		domain = of_get_pci_domain_nr(parent->of_node);
5409 	/*
5410 	 * Check DT domain and use_dt_domains values.
5411 	 *
5412 	 * If DT domain property is valid (domain >= 0) and
5413 	 * use_dt_domains != 0, the DT assignment is valid since this means
5414 	 * we have not previously allocated a domain number by using
5415 	 * pci_get_new_domain_nr(); we should also update use_dt_domains to
5416 	 * 1, to indicate that we have just assigned a domain number from
5417 	 * DT.
5418 	 *
5419 	 * If DT domain property value is not valid (ie domain < 0), and we
5420 	 * have not previously assigned a domain number from DT
5421 	 * (use_dt_domains != 1) we should assign a domain number by
5422 	 * using the:
5423 	 *
5424 	 * pci_get_new_domain_nr()
5425 	 *
5426 	 * API and update the use_dt_domains value to keep track of method we
5427 	 * are using to assign domain numbers (use_dt_domains = 0).
5428 	 *
5429 	 * All other combinations imply we have a platform that is trying
5430 	 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
5431 	 * which is a recipe for domain mishandling and it is prevented by
5432 	 * invalidating the domain value (domain = -1) and printing a
5433 	 * corresponding error.
5434 	 */
5435 	if (domain >= 0 && use_dt_domains) {
5436 		use_dt_domains = 1;
5437 	} else if (domain < 0 && use_dt_domains != 1) {
5438 		use_dt_domains = 0;
5439 		domain = pci_get_new_domain_nr();
5440 	} else {
5441 		dev_err(parent, "Node %pOF has inconsistent \"linux,pci-domain\" property in DT\n",
5442 			parent->of_node);
5443 		domain = -1;
5444 	}
5445 
5446 	return domain;
5447 }
5448 
5449 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
5450 {
5451 	return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
5452 			       acpi_pci_bus_find_domain_nr(bus);
5453 }
5454 #endif
5455 #endif
5456 
5457 /**
5458  * pci_ext_cfg_avail - can we access extended PCI config space?
5459  *
5460  * Returns 1 if we can access PCI extended config space (offsets
5461  * greater than 0xff). This is the default implementation. Architecture
5462  * implementations can override this.
5463  */
5464 int __weak pci_ext_cfg_avail(void)
5465 {
5466 	return 1;
5467 }
5468 
5469 void __weak pci_fixup_cardbus(struct pci_bus *bus)
5470 {
5471 }
5472 EXPORT_SYMBOL(pci_fixup_cardbus);
5473 
5474 static int __init pci_setup(char *str)
5475 {
5476 	while (str) {
5477 		char *k = strchr(str, ',');
5478 		if (k)
5479 			*k++ = 0;
5480 		if (*str && (str = pcibios_setup(str)) && *str) {
5481 			if (!strcmp(str, "nomsi")) {
5482 				pci_no_msi();
5483 			} else if (!strcmp(str, "noaer")) {
5484 				pci_no_aer();
5485 			} else if (!strncmp(str, "realloc=", 8)) {
5486 				pci_realloc_get_opt(str + 8);
5487 			} else if (!strncmp(str, "realloc", 7)) {
5488 				pci_realloc_get_opt("on");
5489 			} else if (!strcmp(str, "nodomains")) {
5490 				pci_no_domains();
5491 			} else if (!strncmp(str, "noari", 5)) {
5492 				pcie_ari_disabled = true;
5493 			} else if (!strncmp(str, "cbiosize=", 9)) {
5494 				pci_cardbus_io_size = memparse(str + 9, &str);
5495 			} else if (!strncmp(str, "cbmemsize=", 10)) {
5496 				pci_cardbus_mem_size = memparse(str + 10, &str);
5497 			} else if (!strncmp(str, "resource_alignment=", 19)) {
5498 				pci_set_resource_alignment_param(str + 19,
5499 							strlen(str + 19));
5500 			} else if (!strncmp(str, "ecrc=", 5)) {
5501 				pcie_ecrc_get_policy(str + 5);
5502 			} else if (!strncmp(str, "hpiosize=", 9)) {
5503 				pci_hotplug_io_size = memparse(str + 9, &str);
5504 			} else if (!strncmp(str, "hpmemsize=", 10)) {
5505 				pci_hotplug_mem_size = memparse(str + 10, &str);
5506 			} else if (!strncmp(str, "hpbussize=", 10)) {
5507 				pci_hotplug_bus_size =
5508 					simple_strtoul(str + 10, &str, 0);
5509 				if (pci_hotplug_bus_size > 0xff)
5510 					pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
5511 			} else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
5512 				pcie_bus_config = PCIE_BUS_TUNE_OFF;
5513 			} else if (!strncmp(str, "pcie_bus_safe", 13)) {
5514 				pcie_bus_config = PCIE_BUS_SAFE;
5515 			} else if (!strncmp(str, "pcie_bus_perf", 13)) {
5516 				pcie_bus_config = PCIE_BUS_PERFORMANCE;
5517 			} else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
5518 				pcie_bus_config = PCIE_BUS_PEER2PEER;
5519 			} else if (!strncmp(str, "pcie_scan_all", 13)) {
5520 				pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
5521 			} else {
5522 				printk(KERN_ERR "PCI: Unknown option `%s'\n",
5523 						str);
5524 			}
5525 		}
5526 		str = k;
5527 	}
5528 	return 0;
5529 }
5530 early_param("pci", pci_setup);
5531