xref: /linux/drivers/pci/pci.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
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/kernel.h>
11 #include <linux/delay.h>
12 #include <linux/init.h>
13 #include <linux/pci.h>
14 #include <linux/pm.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/spinlock.h>
18 #include <linux/string.h>
19 #include <linux/log2.h>
20 #include <linux/pci-aspm.h>
21 #include <linux/pm_wakeup.h>
22 #include <linux/interrupt.h>
23 #include <linux/device.h>
24 #include <linux/pm_runtime.h>
25 #include <asm-generic/pci-bridge.h>
26 #include <asm/setup.h>
27 #include "pci.h"
28 
29 const char *pci_power_names[] = {
30 	"error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
31 };
32 EXPORT_SYMBOL_GPL(pci_power_names);
33 
34 int isa_dma_bridge_buggy;
35 EXPORT_SYMBOL(isa_dma_bridge_buggy);
36 
37 int pci_pci_problems;
38 EXPORT_SYMBOL(pci_pci_problems);
39 
40 unsigned int pci_pm_d3_delay;
41 
42 static void pci_pme_list_scan(struct work_struct *work);
43 
44 static LIST_HEAD(pci_pme_list);
45 static DEFINE_MUTEX(pci_pme_list_mutex);
46 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
47 
48 struct pci_pme_device {
49 	struct list_head list;
50 	struct pci_dev *dev;
51 };
52 
53 #define PME_TIMEOUT 1000 /* How long between PME checks */
54 
55 static void pci_dev_d3_sleep(struct pci_dev *dev)
56 {
57 	unsigned int delay = dev->d3_delay;
58 
59 	if (delay < pci_pm_d3_delay)
60 		delay = pci_pm_d3_delay;
61 
62 	msleep(delay);
63 }
64 
65 #ifdef CONFIG_PCI_DOMAINS
66 int pci_domains_supported = 1;
67 #endif
68 
69 #define DEFAULT_CARDBUS_IO_SIZE		(256)
70 #define DEFAULT_CARDBUS_MEM_SIZE	(64*1024*1024)
71 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
72 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
73 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
74 
75 #define DEFAULT_HOTPLUG_IO_SIZE		(256)
76 #define DEFAULT_HOTPLUG_MEM_SIZE	(2*1024*1024)
77 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
78 unsigned long pci_hotplug_io_size  = DEFAULT_HOTPLUG_IO_SIZE;
79 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
80 
81 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_TUNE_OFF;
82 
83 /*
84  * The default CLS is used if arch didn't set CLS explicitly and not
85  * all pci devices agree on the same value.  Arch can override either
86  * the dfl or actual value as it sees fit.  Don't forget this is
87  * measured in 32-bit words, not bytes.
88  */
89 u8 pci_dfl_cache_line_size __devinitdata = L1_CACHE_BYTES >> 2;
90 u8 pci_cache_line_size;
91 
92 /*
93  * If we set up a device for bus mastering, we need to check the latency
94  * timer as certain BIOSes forget to set it properly.
95  */
96 unsigned int pcibios_max_latency = 255;
97 
98 /* If set, the PCIe ARI capability will not be used. */
99 static bool pcie_ari_disabled;
100 
101 /**
102  * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
103  * @bus: pointer to PCI bus structure to search
104  *
105  * Given a PCI bus, returns the highest PCI bus number present in the set
106  * including the given PCI bus and its list of child PCI buses.
107  */
108 unsigned char pci_bus_max_busnr(struct pci_bus* bus)
109 {
110 	struct list_head *tmp;
111 	unsigned char max, n;
112 
113 	max = bus->busn_res.end;
114 	list_for_each(tmp, &bus->children) {
115 		n = pci_bus_max_busnr(pci_bus_b(tmp));
116 		if(n > max)
117 			max = n;
118 	}
119 	return max;
120 }
121 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
122 
123 #ifdef CONFIG_HAS_IOMEM
124 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
125 {
126 	/*
127 	 * Make sure the BAR is actually a memory resource, not an IO resource
128 	 */
129 	if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
130 		WARN_ON(1);
131 		return NULL;
132 	}
133 	return ioremap_nocache(pci_resource_start(pdev, bar),
134 				     pci_resource_len(pdev, bar));
135 }
136 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
137 #endif
138 
139 #define PCI_FIND_CAP_TTL	48
140 
141 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
142 				   u8 pos, int cap, int *ttl)
143 {
144 	u8 id;
145 
146 	while ((*ttl)--) {
147 		pci_bus_read_config_byte(bus, devfn, pos, &pos);
148 		if (pos < 0x40)
149 			break;
150 		pos &= ~3;
151 		pci_bus_read_config_byte(bus, devfn, pos + PCI_CAP_LIST_ID,
152 					 &id);
153 		if (id == 0xff)
154 			break;
155 		if (id == cap)
156 			return pos;
157 		pos += PCI_CAP_LIST_NEXT;
158 	}
159 	return 0;
160 }
161 
162 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
163 			       u8 pos, int cap)
164 {
165 	int ttl = PCI_FIND_CAP_TTL;
166 
167 	return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
168 }
169 
170 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
171 {
172 	return __pci_find_next_cap(dev->bus, dev->devfn,
173 				   pos + PCI_CAP_LIST_NEXT, cap);
174 }
175 EXPORT_SYMBOL_GPL(pci_find_next_capability);
176 
177 static int __pci_bus_find_cap_start(struct pci_bus *bus,
178 				    unsigned int devfn, u8 hdr_type)
179 {
180 	u16 status;
181 
182 	pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
183 	if (!(status & PCI_STATUS_CAP_LIST))
184 		return 0;
185 
186 	switch (hdr_type) {
187 	case PCI_HEADER_TYPE_NORMAL:
188 	case PCI_HEADER_TYPE_BRIDGE:
189 		return PCI_CAPABILITY_LIST;
190 	case PCI_HEADER_TYPE_CARDBUS:
191 		return PCI_CB_CAPABILITY_LIST;
192 	default:
193 		return 0;
194 	}
195 
196 	return 0;
197 }
198 
199 /**
200  * pci_find_capability - query for devices' capabilities
201  * @dev: PCI device to query
202  * @cap: capability code
203  *
204  * Tell if a device supports a given PCI capability.
205  * Returns the address of the requested capability structure within the
206  * device's PCI configuration space or 0 in case the device does not
207  * support it.  Possible values for @cap:
208  *
209  *  %PCI_CAP_ID_PM           Power Management
210  *  %PCI_CAP_ID_AGP          Accelerated Graphics Port
211  *  %PCI_CAP_ID_VPD          Vital Product Data
212  *  %PCI_CAP_ID_SLOTID       Slot Identification
213  *  %PCI_CAP_ID_MSI          Message Signalled Interrupts
214  *  %PCI_CAP_ID_CHSWP        CompactPCI HotSwap
215  *  %PCI_CAP_ID_PCIX         PCI-X
216  *  %PCI_CAP_ID_EXP          PCI Express
217  */
218 int pci_find_capability(struct pci_dev *dev, int cap)
219 {
220 	int pos;
221 
222 	pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
223 	if (pos)
224 		pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
225 
226 	return pos;
227 }
228 
229 /**
230  * pci_bus_find_capability - query for devices' capabilities
231  * @bus:   the PCI bus to query
232  * @devfn: PCI device to query
233  * @cap:   capability code
234  *
235  * Like pci_find_capability() but works for pci devices that do not have a
236  * pci_dev structure set up yet.
237  *
238  * Returns the address of the requested capability structure within the
239  * device's PCI configuration space or 0 in case the device does not
240  * support it.
241  */
242 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
243 {
244 	int pos;
245 	u8 hdr_type;
246 
247 	pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
248 
249 	pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
250 	if (pos)
251 		pos = __pci_find_next_cap(bus, devfn, pos, cap);
252 
253 	return pos;
254 }
255 
256 /**
257  * pci_pcie_cap2 - query for devices' PCI_CAP_ID_EXP v2 capability structure
258  * @dev: PCI device to check
259  *
260  * Like pci_pcie_cap() but also checks that the PCIe capability version is
261  * >= 2.  Note that v1 capability structures could be sparse in that not
262  * all register fields were required.  v2 requires the entire structure to
263  * be present size wise, while still allowing for non-implemented registers
264  * to exist but they must be hardwired to 0.
265  *
266  * Due to the differences in the versions of capability structures, one
267  * must be careful not to try and access non-existant registers that may
268  * exist in early versions - v1 - of Express devices.
269  *
270  * Returns the offset of the PCIe capability structure as long as the
271  * capability version is >= 2; otherwise 0 is returned.
272  */
273 static int pci_pcie_cap2(struct pci_dev *dev)
274 {
275 	u16 flags;
276 	int pos;
277 
278 	pos = pci_pcie_cap(dev);
279 	if (pos) {
280 		pci_read_config_word(dev, pos + PCI_EXP_FLAGS, &flags);
281 		if ((flags & PCI_EXP_FLAGS_VERS) < 2)
282 			pos = 0;
283 	}
284 
285 	return pos;
286 }
287 
288 /**
289  * pci_find_ext_capability - Find an extended capability
290  * @dev: PCI device to query
291  * @cap: capability code
292  *
293  * Returns the address of the requested extended capability structure
294  * within the device's PCI configuration space or 0 if the device does
295  * not support it.  Possible values for @cap:
296  *
297  *  %PCI_EXT_CAP_ID_ERR		Advanced Error Reporting
298  *  %PCI_EXT_CAP_ID_VC		Virtual Channel
299  *  %PCI_EXT_CAP_ID_DSN		Device Serial Number
300  *  %PCI_EXT_CAP_ID_PWR		Power Budgeting
301  */
302 int pci_find_ext_capability(struct pci_dev *dev, int cap)
303 {
304 	u32 header;
305 	int ttl;
306 	int pos = PCI_CFG_SPACE_SIZE;
307 
308 	/* minimum 8 bytes per capability */
309 	ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
310 
311 	if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
312 		return 0;
313 
314 	if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
315 		return 0;
316 
317 	/*
318 	 * If we have no capabilities, this is indicated by cap ID,
319 	 * cap version and next pointer all being 0.
320 	 */
321 	if (header == 0)
322 		return 0;
323 
324 	while (ttl-- > 0) {
325 		if (PCI_EXT_CAP_ID(header) == cap)
326 			return pos;
327 
328 		pos = PCI_EXT_CAP_NEXT(header);
329 		if (pos < PCI_CFG_SPACE_SIZE)
330 			break;
331 
332 		if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
333 			break;
334 	}
335 
336 	return 0;
337 }
338 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
339 
340 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
341 {
342 	int rc, ttl = PCI_FIND_CAP_TTL;
343 	u8 cap, mask;
344 
345 	if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
346 		mask = HT_3BIT_CAP_MASK;
347 	else
348 		mask = HT_5BIT_CAP_MASK;
349 
350 	pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
351 				      PCI_CAP_ID_HT, &ttl);
352 	while (pos) {
353 		rc = pci_read_config_byte(dev, pos + 3, &cap);
354 		if (rc != PCIBIOS_SUCCESSFUL)
355 			return 0;
356 
357 		if ((cap & mask) == ht_cap)
358 			return pos;
359 
360 		pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
361 					      pos + PCI_CAP_LIST_NEXT,
362 					      PCI_CAP_ID_HT, &ttl);
363 	}
364 
365 	return 0;
366 }
367 /**
368  * pci_find_next_ht_capability - query a device's Hypertransport capabilities
369  * @dev: PCI device to query
370  * @pos: Position from which to continue searching
371  * @ht_cap: Hypertransport capability code
372  *
373  * To be used in conjunction with pci_find_ht_capability() to search for
374  * all capabilities matching @ht_cap. @pos should always be a value returned
375  * from pci_find_ht_capability().
376  *
377  * NB. To be 100% safe against broken PCI devices, the caller should take
378  * steps to avoid an infinite loop.
379  */
380 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
381 {
382 	return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
383 }
384 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
385 
386 /**
387  * pci_find_ht_capability - query a device's Hypertransport capabilities
388  * @dev: PCI device to query
389  * @ht_cap: Hypertransport capability code
390  *
391  * Tell if a device supports a given Hypertransport capability.
392  * Returns an address within the device's PCI configuration space
393  * or 0 in case the device does not support the request capability.
394  * The address points to the PCI capability, of type PCI_CAP_ID_HT,
395  * which has a Hypertransport capability matching @ht_cap.
396  */
397 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
398 {
399 	int pos;
400 
401 	pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
402 	if (pos)
403 		pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
404 
405 	return pos;
406 }
407 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
408 
409 /**
410  * pci_find_parent_resource - return resource region of parent bus of given region
411  * @dev: PCI device structure contains resources to be searched
412  * @res: child resource record for which parent is sought
413  *
414  *  For given resource region of given device, return the resource
415  *  region of parent bus the given region is contained in or where
416  *  it should be allocated from.
417  */
418 struct resource *
419 pci_find_parent_resource(const struct pci_dev *dev, struct resource *res)
420 {
421 	const struct pci_bus *bus = dev->bus;
422 	int i;
423 	struct resource *best = NULL, *r;
424 
425 	pci_bus_for_each_resource(bus, r, i) {
426 		if (!r)
427 			continue;
428 		if (res->start && !(res->start >= r->start && res->end <= r->end))
429 			continue;	/* Not contained */
430 		if ((res->flags ^ r->flags) & (IORESOURCE_IO | IORESOURCE_MEM))
431 			continue;	/* Wrong type */
432 		if (!((res->flags ^ r->flags) & IORESOURCE_PREFETCH))
433 			return r;	/* Exact match */
434 		/* We can't insert a non-prefetch resource inside a prefetchable parent .. */
435 		if (r->flags & IORESOURCE_PREFETCH)
436 			continue;
437 		/* .. but we can put a prefetchable resource inside a non-prefetchable one */
438 		if (!best)
439 			best = r;
440 	}
441 	return best;
442 }
443 
444 /**
445  * pci_restore_bars - restore a devices BAR values (e.g. after wake-up)
446  * @dev: PCI device to have its BARs restored
447  *
448  * Restore the BAR values for a given device, so as to make it
449  * accessible by its driver.
450  */
451 static void
452 pci_restore_bars(struct pci_dev *dev)
453 {
454 	int i;
455 
456 	for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
457 		pci_update_resource(dev, i);
458 }
459 
460 static struct pci_platform_pm_ops *pci_platform_pm;
461 
462 int pci_set_platform_pm(struct pci_platform_pm_ops *ops)
463 {
464 	if (!ops->is_manageable || !ops->set_state || !ops->choose_state
465 	    || !ops->sleep_wake || !ops->can_wakeup)
466 		return -EINVAL;
467 	pci_platform_pm = ops;
468 	return 0;
469 }
470 
471 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
472 {
473 	return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
474 }
475 
476 static inline int platform_pci_set_power_state(struct pci_dev *dev,
477                                                 pci_power_t t)
478 {
479 	return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
480 }
481 
482 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
483 {
484 	return pci_platform_pm ?
485 			pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
486 }
487 
488 static inline bool platform_pci_can_wakeup(struct pci_dev *dev)
489 {
490 	return pci_platform_pm ? pci_platform_pm->can_wakeup(dev) : false;
491 }
492 
493 static inline int platform_pci_sleep_wake(struct pci_dev *dev, bool enable)
494 {
495 	return pci_platform_pm ?
496 			pci_platform_pm->sleep_wake(dev, enable) : -ENODEV;
497 }
498 
499 static inline int platform_pci_run_wake(struct pci_dev *dev, bool enable)
500 {
501 	return pci_platform_pm ?
502 			pci_platform_pm->run_wake(dev, enable) : -ENODEV;
503 }
504 
505 /**
506  * pci_raw_set_power_state - Use PCI PM registers to set the power state of
507  *                           given PCI device
508  * @dev: PCI device to handle.
509  * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
510  *
511  * RETURN VALUE:
512  * -EINVAL if the requested state is invalid.
513  * -EIO if device does not support PCI PM or its PM capabilities register has a
514  * wrong version, or device doesn't support the requested state.
515  * 0 if device already is in the requested state.
516  * 0 if device's power state has been successfully changed.
517  */
518 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
519 {
520 	u16 pmcsr;
521 	bool need_restore = false;
522 
523 	/* Check if we're already there */
524 	if (dev->current_state == state)
525 		return 0;
526 
527 	if (!dev->pm_cap)
528 		return -EIO;
529 
530 	if (state < PCI_D0 || state > PCI_D3hot)
531 		return -EINVAL;
532 
533 	/* Validate current state:
534 	 * Can enter D0 from any state, but if we can only go deeper
535 	 * to sleep if we're already in a low power state
536 	 */
537 	if (state != PCI_D0 && dev->current_state <= PCI_D3cold
538 	    && dev->current_state > state) {
539 		dev_err(&dev->dev, "invalid power transition "
540 			"(from state %d to %d)\n", dev->current_state, state);
541 		return -EINVAL;
542 	}
543 
544 	/* check if this device supports the desired state */
545 	if ((state == PCI_D1 && !dev->d1_support)
546 	   || (state == PCI_D2 && !dev->d2_support))
547 		return -EIO;
548 
549 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
550 
551 	/* If we're (effectively) in D3, force entire word to 0.
552 	 * This doesn't affect PME_Status, disables PME_En, and
553 	 * sets PowerState to 0.
554 	 */
555 	switch (dev->current_state) {
556 	case PCI_D0:
557 	case PCI_D1:
558 	case PCI_D2:
559 		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
560 		pmcsr |= state;
561 		break;
562 	case PCI_D3hot:
563 	case PCI_D3cold:
564 	case PCI_UNKNOWN: /* Boot-up */
565 		if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
566 		 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
567 			need_restore = true;
568 		/* Fall-through: force to D0 */
569 	default:
570 		pmcsr = 0;
571 		break;
572 	}
573 
574 	/* enter specified state */
575 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
576 
577 	/* Mandatory power management transition delays */
578 	/* see PCI PM 1.1 5.6.1 table 18 */
579 	if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
580 		pci_dev_d3_sleep(dev);
581 	else if (state == PCI_D2 || dev->current_state == PCI_D2)
582 		udelay(PCI_PM_D2_DELAY);
583 
584 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
585 	dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
586 	if (dev->current_state != state && printk_ratelimit())
587 		dev_info(&dev->dev, "Refused to change power state, "
588 			"currently in D%d\n", dev->current_state);
589 
590 	/*
591 	 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
592 	 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
593 	 * from D3hot to D0 _may_ perform an internal reset, thereby
594 	 * going to "D0 Uninitialized" rather than "D0 Initialized".
595 	 * For example, at least some versions of the 3c905B and the
596 	 * 3c556B exhibit this behaviour.
597 	 *
598 	 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
599 	 * devices in a D3hot state at boot.  Consequently, we need to
600 	 * restore at least the BARs so that the device will be
601 	 * accessible to its driver.
602 	 */
603 	if (need_restore)
604 		pci_restore_bars(dev);
605 
606 	if (dev->bus->self)
607 		pcie_aspm_pm_state_change(dev->bus->self);
608 
609 	return 0;
610 }
611 
612 /**
613  * pci_update_current_state - Read PCI power state of given device from its
614  *                            PCI PM registers and cache it
615  * @dev: PCI device to handle.
616  * @state: State to cache in case the device doesn't have the PM capability
617  */
618 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
619 {
620 	if (dev->pm_cap) {
621 		u16 pmcsr;
622 
623 		/*
624 		 * Configuration space is not accessible for device in
625 		 * D3cold, so just keep or set D3cold for safety
626 		 */
627 		if (dev->current_state == PCI_D3cold)
628 			return;
629 		if (state == PCI_D3cold) {
630 			dev->current_state = PCI_D3cold;
631 			return;
632 		}
633 		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
634 		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
635 	} else {
636 		dev->current_state = state;
637 	}
638 }
639 
640 /**
641  * pci_power_up - Put the given device into D0 forcibly
642  * @dev: PCI device to power up
643  */
644 void pci_power_up(struct pci_dev *dev)
645 {
646 	if (platform_pci_power_manageable(dev))
647 		platform_pci_set_power_state(dev, PCI_D0);
648 
649 	pci_raw_set_power_state(dev, PCI_D0);
650 	pci_update_current_state(dev, PCI_D0);
651 }
652 
653 /**
654  * pci_platform_power_transition - Use platform to change device power state
655  * @dev: PCI device to handle.
656  * @state: State to put the device into.
657  */
658 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
659 {
660 	int error;
661 
662 	if (platform_pci_power_manageable(dev)) {
663 		error = platform_pci_set_power_state(dev, state);
664 		if (!error)
665 			pci_update_current_state(dev, state);
666 		/* Fall back to PCI_D0 if native PM is not supported */
667 		if (!dev->pm_cap)
668 			dev->current_state = PCI_D0;
669 	} else {
670 		error = -ENODEV;
671 		/* Fall back to PCI_D0 if native PM is not supported */
672 		if (!dev->pm_cap)
673 			dev->current_state = PCI_D0;
674 	}
675 
676 	return error;
677 }
678 
679 /**
680  * __pci_start_power_transition - Start power transition of a PCI device
681  * @dev: PCI device to handle.
682  * @state: State to put the device into.
683  */
684 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
685 {
686 	if (state == PCI_D0) {
687 		pci_platform_power_transition(dev, PCI_D0);
688 		/*
689 		 * Mandatory power management transition delays, see
690 		 * PCI Express Base Specification Revision 2.0 Section
691 		 * 6.6.1: Conventional Reset.  Do not delay for
692 		 * devices powered on/off by corresponding bridge,
693 		 * because have already delayed for the bridge.
694 		 */
695 		if (dev->runtime_d3cold) {
696 			msleep(dev->d3cold_delay);
697 			/*
698 			 * When powering on a bridge from D3cold, the
699 			 * whole hierarchy may be powered on into
700 			 * D0uninitialized state, resume them to give
701 			 * them a chance to suspend again
702 			 */
703 			pci_wakeup_bus(dev->subordinate);
704 		}
705 	}
706 }
707 
708 /**
709  * __pci_dev_set_current_state - Set current state of a PCI device
710  * @dev: Device to handle
711  * @data: pointer to state to be set
712  */
713 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
714 {
715 	pci_power_t state = *(pci_power_t *)data;
716 
717 	dev->current_state = state;
718 	return 0;
719 }
720 
721 /**
722  * __pci_bus_set_current_state - Walk given bus and set current state of devices
723  * @bus: Top bus of the subtree to walk.
724  * @state: state to be set
725  */
726 static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
727 {
728 	if (bus)
729 		pci_walk_bus(bus, __pci_dev_set_current_state, &state);
730 }
731 
732 /**
733  * __pci_complete_power_transition - Complete power transition of a PCI device
734  * @dev: PCI device to handle.
735  * @state: State to put the device into.
736  *
737  * This function should not be called directly by device drivers.
738  */
739 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
740 {
741 	int ret;
742 
743 	if (state <= PCI_D0)
744 		return -EINVAL;
745 	ret = pci_platform_power_transition(dev, state);
746 	/* Power off the bridge may power off the whole hierarchy */
747 	if (!ret && state == PCI_D3cold)
748 		__pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
749 	return ret;
750 }
751 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
752 
753 /**
754  * pci_set_power_state - Set the power state of a PCI device
755  * @dev: PCI device to handle.
756  * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
757  *
758  * Transition a device to a new power state, using the platform firmware and/or
759  * the device's PCI PM registers.
760  *
761  * RETURN VALUE:
762  * -EINVAL if the requested state is invalid.
763  * -EIO if device does not support PCI PM or its PM capabilities register has a
764  * wrong version, or device doesn't support the requested state.
765  * 0 if device already is in the requested state.
766  * 0 if device's power state has been successfully changed.
767  */
768 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
769 {
770 	int error;
771 
772 	/* bound the state we're entering */
773 	if (state > PCI_D3cold)
774 		state = PCI_D3cold;
775 	else if (state < PCI_D0)
776 		state = PCI_D0;
777 	else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
778 		/*
779 		 * If the device or the parent bridge do not support PCI PM,
780 		 * ignore the request if we're doing anything other than putting
781 		 * it into D0 (which would only happen on boot).
782 		 */
783 		return 0;
784 
785 	/* Check if we're already there */
786 	if (dev->current_state == state)
787 		return 0;
788 
789 	__pci_start_power_transition(dev, state);
790 
791 	/* This device is quirked not to be put into D3, so
792 	   don't put it in D3 */
793 	if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
794 		return 0;
795 
796 	/*
797 	 * To put device in D3cold, we put device into D3hot in native
798 	 * way, then put device into D3cold with platform ops
799 	 */
800 	error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
801 					PCI_D3hot : state);
802 
803 	if (!__pci_complete_power_transition(dev, state))
804 		error = 0;
805 	/*
806 	 * When aspm_policy is "powersave" this call ensures
807 	 * that ASPM is configured.
808 	 */
809 	if (!error && dev->bus->self)
810 		pcie_aspm_powersave_config_link(dev->bus->self);
811 
812 	return error;
813 }
814 
815 /**
816  * pci_choose_state - Choose the power state of a PCI device
817  * @dev: PCI device to be suspended
818  * @state: target sleep state for the whole system. This is the value
819  *	that is passed to suspend() function.
820  *
821  * Returns PCI power state suitable for given device and given system
822  * message.
823  */
824 
825 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
826 {
827 	pci_power_t ret;
828 
829 	if (!pci_find_capability(dev, PCI_CAP_ID_PM))
830 		return PCI_D0;
831 
832 	ret = platform_pci_choose_state(dev);
833 	if (ret != PCI_POWER_ERROR)
834 		return ret;
835 
836 	switch (state.event) {
837 	case PM_EVENT_ON:
838 		return PCI_D0;
839 	case PM_EVENT_FREEZE:
840 	case PM_EVENT_PRETHAW:
841 		/* REVISIT both freeze and pre-thaw "should" use D0 */
842 	case PM_EVENT_SUSPEND:
843 	case PM_EVENT_HIBERNATE:
844 		return PCI_D3hot;
845 	default:
846 		dev_info(&dev->dev, "unrecognized suspend event %d\n",
847 			 state.event);
848 		BUG();
849 	}
850 	return PCI_D0;
851 }
852 
853 EXPORT_SYMBOL(pci_choose_state);
854 
855 #define PCI_EXP_SAVE_REGS	7
856 
857 #define pcie_cap_has_devctl(type, flags)	1
858 #define pcie_cap_has_lnkctl(type, flags)		\
859 		((flags & PCI_EXP_FLAGS_VERS) > 1 ||	\
860 		 (type == PCI_EXP_TYPE_ROOT_PORT ||	\
861 		  type == PCI_EXP_TYPE_ENDPOINT ||	\
862 		  type == PCI_EXP_TYPE_LEG_END))
863 #define pcie_cap_has_sltctl(type, flags)		\
864 		((flags & PCI_EXP_FLAGS_VERS) > 1 ||	\
865 		 ((type == PCI_EXP_TYPE_ROOT_PORT) ||	\
866 		  (type == PCI_EXP_TYPE_DOWNSTREAM &&	\
867 		   (flags & PCI_EXP_FLAGS_SLOT))))
868 #define pcie_cap_has_rtctl(type, flags)			\
869 		((flags & PCI_EXP_FLAGS_VERS) > 1 ||	\
870 		 (type == PCI_EXP_TYPE_ROOT_PORT ||	\
871 		  type == PCI_EXP_TYPE_RC_EC))
872 
873 static struct pci_cap_saved_state *pci_find_saved_cap(
874 	struct pci_dev *pci_dev, char cap)
875 {
876 	struct pci_cap_saved_state *tmp;
877 	struct hlist_node *pos;
878 
879 	hlist_for_each_entry(tmp, pos, &pci_dev->saved_cap_space, next) {
880 		if (tmp->cap.cap_nr == cap)
881 			return tmp;
882 	}
883 	return NULL;
884 }
885 
886 static int pci_save_pcie_state(struct pci_dev *dev)
887 {
888 	int pos, i = 0;
889 	struct pci_cap_saved_state *save_state;
890 	u16 *cap;
891 	u16 flags;
892 
893 	pos = pci_pcie_cap(dev);
894 	if (!pos)
895 		return 0;
896 
897 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
898 	if (!save_state) {
899 		dev_err(&dev->dev, "buffer not found in %s\n", __func__);
900 		return -ENOMEM;
901 	}
902 	cap = (u16 *)&save_state->cap.data[0];
903 
904 	pci_read_config_word(dev, pos + PCI_EXP_FLAGS, &flags);
905 
906 	if (pcie_cap_has_devctl(dev->pcie_type, flags))
907 		pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &cap[i++]);
908 	if (pcie_cap_has_lnkctl(dev->pcie_type, flags))
909 		pci_read_config_word(dev, pos + PCI_EXP_LNKCTL, &cap[i++]);
910 	if (pcie_cap_has_sltctl(dev->pcie_type, flags))
911 		pci_read_config_word(dev, pos + PCI_EXP_SLTCTL, &cap[i++]);
912 	if (pcie_cap_has_rtctl(dev->pcie_type, flags))
913 		pci_read_config_word(dev, pos + PCI_EXP_RTCTL, &cap[i++]);
914 
915 	pos = pci_pcie_cap2(dev);
916 	if (!pos)
917 		return 0;
918 
919 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &cap[i++]);
920 	pci_read_config_word(dev, pos + PCI_EXP_LNKCTL2, &cap[i++]);
921 	pci_read_config_word(dev, pos + PCI_EXP_SLTCTL2, &cap[i++]);
922 	return 0;
923 }
924 
925 static void pci_restore_pcie_state(struct pci_dev *dev)
926 {
927 	int i = 0, pos;
928 	struct pci_cap_saved_state *save_state;
929 	u16 *cap;
930 	u16 flags;
931 
932 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
933 	pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
934 	if (!save_state || pos <= 0)
935 		return;
936 	cap = (u16 *)&save_state->cap.data[0];
937 
938 	pci_read_config_word(dev, pos + PCI_EXP_FLAGS, &flags);
939 
940 	if (pcie_cap_has_devctl(dev->pcie_type, flags))
941 		pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, cap[i++]);
942 	if (pcie_cap_has_lnkctl(dev->pcie_type, flags))
943 		pci_write_config_word(dev, pos + PCI_EXP_LNKCTL, cap[i++]);
944 	if (pcie_cap_has_sltctl(dev->pcie_type, flags))
945 		pci_write_config_word(dev, pos + PCI_EXP_SLTCTL, cap[i++]);
946 	if (pcie_cap_has_rtctl(dev->pcie_type, flags))
947 		pci_write_config_word(dev, pos + PCI_EXP_RTCTL, cap[i++]);
948 
949 	pos = pci_pcie_cap2(dev);
950 	if (!pos)
951 		return;
952 
953 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, cap[i++]);
954 	pci_write_config_word(dev, pos + PCI_EXP_LNKCTL2, cap[i++]);
955 	pci_write_config_word(dev, pos + PCI_EXP_SLTCTL2, cap[i++]);
956 }
957 
958 
959 static int pci_save_pcix_state(struct pci_dev *dev)
960 {
961 	int pos;
962 	struct pci_cap_saved_state *save_state;
963 
964 	pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
965 	if (pos <= 0)
966 		return 0;
967 
968 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
969 	if (!save_state) {
970 		dev_err(&dev->dev, "buffer not found in %s\n", __func__);
971 		return -ENOMEM;
972 	}
973 
974 	pci_read_config_word(dev, pos + PCI_X_CMD,
975 			     (u16 *)save_state->cap.data);
976 
977 	return 0;
978 }
979 
980 static void pci_restore_pcix_state(struct pci_dev *dev)
981 {
982 	int i = 0, pos;
983 	struct pci_cap_saved_state *save_state;
984 	u16 *cap;
985 
986 	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
987 	pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
988 	if (!save_state || pos <= 0)
989 		return;
990 	cap = (u16 *)&save_state->cap.data[0];
991 
992 	pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
993 }
994 
995 
996 /**
997  * pci_save_state - save the PCI configuration space of a device before suspending
998  * @dev: - PCI device that we're dealing with
999  */
1000 int
1001 pci_save_state(struct pci_dev *dev)
1002 {
1003 	int i;
1004 	/* XXX: 100% dword access ok here? */
1005 	for (i = 0; i < 16; i++)
1006 		pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1007 	dev->state_saved = true;
1008 	if ((i = pci_save_pcie_state(dev)) != 0)
1009 		return i;
1010 	if ((i = pci_save_pcix_state(dev)) != 0)
1011 		return i;
1012 	return 0;
1013 }
1014 
1015 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1016 				     u32 saved_val, int retry)
1017 {
1018 	u32 val;
1019 
1020 	pci_read_config_dword(pdev, offset, &val);
1021 	if (val == saved_val)
1022 		return;
1023 
1024 	for (;;) {
1025 		dev_dbg(&pdev->dev, "restoring config space at offset "
1026 			"%#x (was %#x, writing %#x)\n", offset, val, saved_val);
1027 		pci_write_config_dword(pdev, offset, saved_val);
1028 		if (retry-- <= 0)
1029 			return;
1030 
1031 		pci_read_config_dword(pdev, offset, &val);
1032 		if (val == saved_val)
1033 			return;
1034 
1035 		mdelay(1);
1036 	}
1037 }
1038 
1039 static void pci_restore_config_space_range(struct pci_dev *pdev,
1040 					   int start, int end, int retry)
1041 {
1042 	int index;
1043 
1044 	for (index = end; index >= start; index--)
1045 		pci_restore_config_dword(pdev, 4 * index,
1046 					 pdev->saved_config_space[index],
1047 					 retry);
1048 }
1049 
1050 static void pci_restore_config_space(struct pci_dev *pdev)
1051 {
1052 	if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1053 		pci_restore_config_space_range(pdev, 10, 15, 0);
1054 		/* Restore BARs before the command register. */
1055 		pci_restore_config_space_range(pdev, 4, 9, 10);
1056 		pci_restore_config_space_range(pdev, 0, 3, 0);
1057 	} else {
1058 		pci_restore_config_space_range(pdev, 0, 15, 0);
1059 	}
1060 }
1061 
1062 /**
1063  * pci_restore_state - Restore the saved state of a PCI device
1064  * @dev: - PCI device that we're dealing with
1065  */
1066 void pci_restore_state(struct pci_dev *dev)
1067 {
1068 	if (!dev->state_saved)
1069 		return;
1070 
1071 	/* PCI Express register must be restored first */
1072 	pci_restore_pcie_state(dev);
1073 	pci_restore_ats_state(dev);
1074 
1075 	pci_restore_config_space(dev);
1076 
1077 	pci_restore_pcix_state(dev);
1078 	pci_restore_msi_state(dev);
1079 	pci_restore_iov_state(dev);
1080 
1081 	dev->state_saved = false;
1082 }
1083 
1084 struct pci_saved_state {
1085 	u32 config_space[16];
1086 	struct pci_cap_saved_data cap[0];
1087 };
1088 
1089 /**
1090  * pci_store_saved_state - Allocate and return an opaque struct containing
1091  *			   the device saved state.
1092  * @dev: PCI device that we're dealing with
1093  *
1094  * Rerturn NULL if no state or error.
1095  */
1096 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1097 {
1098 	struct pci_saved_state *state;
1099 	struct pci_cap_saved_state *tmp;
1100 	struct pci_cap_saved_data *cap;
1101 	struct hlist_node *pos;
1102 	size_t size;
1103 
1104 	if (!dev->state_saved)
1105 		return NULL;
1106 
1107 	size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1108 
1109 	hlist_for_each_entry(tmp, pos, &dev->saved_cap_space, next)
1110 		size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1111 
1112 	state = kzalloc(size, GFP_KERNEL);
1113 	if (!state)
1114 		return NULL;
1115 
1116 	memcpy(state->config_space, dev->saved_config_space,
1117 	       sizeof(state->config_space));
1118 
1119 	cap = state->cap;
1120 	hlist_for_each_entry(tmp, pos, &dev->saved_cap_space, next) {
1121 		size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1122 		memcpy(cap, &tmp->cap, len);
1123 		cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1124 	}
1125 	/* Empty cap_save terminates list */
1126 
1127 	return state;
1128 }
1129 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1130 
1131 /**
1132  * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1133  * @dev: PCI device that we're dealing with
1134  * @state: Saved state returned from pci_store_saved_state()
1135  */
1136 int pci_load_saved_state(struct pci_dev *dev, struct pci_saved_state *state)
1137 {
1138 	struct pci_cap_saved_data *cap;
1139 
1140 	dev->state_saved = false;
1141 
1142 	if (!state)
1143 		return 0;
1144 
1145 	memcpy(dev->saved_config_space, state->config_space,
1146 	       sizeof(state->config_space));
1147 
1148 	cap = state->cap;
1149 	while (cap->size) {
1150 		struct pci_cap_saved_state *tmp;
1151 
1152 		tmp = pci_find_saved_cap(dev, cap->cap_nr);
1153 		if (!tmp || tmp->cap.size != cap->size)
1154 			return -EINVAL;
1155 
1156 		memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1157 		cap = (struct pci_cap_saved_data *)((u8 *)cap +
1158 		       sizeof(struct pci_cap_saved_data) + cap->size);
1159 	}
1160 
1161 	dev->state_saved = true;
1162 	return 0;
1163 }
1164 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1165 
1166 /**
1167  * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1168  *				   and free the memory allocated for it.
1169  * @dev: PCI device that we're dealing with
1170  * @state: Pointer to saved state returned from pci_store_saved_state()
1171  */
1172 int pci_load_and_free_saved_state(struct pci_dev *dev,
1173 				  struct pci_saved_state **state)
1174 {
1175 	int ret = pci_load_saved_state(dev, *state);
1176 	kfree(*state);
1177 	*state = NULL;
1178 	return ret;
1179 }
1180 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1181 
1182 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1183 {
1184 	int err;
1185 
1186 	err = pci_set_power_state(dev, PCI_D0);
1187 	if (err < 0 && err != -EIO)
1188 		return err;
1189 	err = pcibios_enable_device(dev, bars);
1190 	if (err < 0)
1191 		return err;
1192 	pci_fixup_device(pci_fixup_enable, dev);
1193 
1194 	return 0;
1195 }
1196 
1197 /**
1198  * pci_reenable_device - Resume abandoned device
1199  * @dev: PCI device to be resumed
1200  *
1201  *  Note this function is a backend of pci_default_resume and is not supposed
1202  *  to be called by normal code, write proper resume handler and use it instead.
1203  */
1204 int pci_reenable_device(struct pci_dev *dev)
1205 {
1206 	if (pci_is_enabled(dev))
1207 		return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1208 	return 0;
1209 }
1210 
1211 static int __pci_enable_device_flags(struct pci_dev *dev,
1212 				     resource_size_t flags)
1213 {
1214 	int err;
1215 	int i, bars = 0;
1216 
1217 	/*
1218 	 * Power state could be unknown at this point, either due to a fresh
1219 	 * boot or a device removal call.  So get the current power state
1220 	 * so that things like MSI message writing will behave as expected
1221 	 * (e.g. if the device really is in D0 at enable time).
1222 	 */
1223 	if (dev->pm_cap) {
1224 		u16 pmcsr;
1225 		pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1226 		dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1227 	}
1228 
1229 	if (atomic_add_return(1, &dev->enable_cnt) > 1)
1230 		return 0;		/* already enabled */
1231 
1232 	/* only skip sriov related */
1233 	for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1234 		if (dev->resource[i].flags & flags)
1235 			bars |= (1 << i);
1236 	for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1237 		if (dev->resource[i].flags & flags)
1238 			bars |= (1 << i);
1239 
1240 	err = do_pci_enable_device(dev, bars);
1241 	if (err < 0)
1242 		atomic_dec(&dev->enable_cnt);
1243 	return err;
1244 }
1245 
1246 /**
1247  * pci_enable_device_io - Initialize a device for use with IO space
1248  * @dev: PCI device to be initialized
1249  *
1250  *  Initialize device before it's used by a driver. Ask low-level code
1251  *  to enable I/O resources. Wake up the device if it was suspended.
1252  *  Beware, this function can fail.
1253  */
1254 int pci_enable_device_io(struct pci_dev *dev)
1255 {
1256 	return __pci_enable_device_flags(dev, IORESOURCE_IO);
1257 }
1258 
1259 /**
1260  * pci_enable_device_mem - Initialize a device for use with Memory space
1261  * @dev: PCI device to be initialized
1262  *
1263  *  Initialize device before it's used by a driver. Ask low-level code
1264  *  to enable Memory resources. Wake up the device if it was suspended.
1265  *  Beware, this function can fail.
1266  */
1267 int pci_enable_device_mem(struct pci_dev *dev)
1268 {
1269 	return __pci_enable_device_flags(dev, IORESOURCE_MEM);
1270 }
1271 
1272 /**
1273  * pci_enable_device - Initialize device before it's used by a driver.
1274  * @dev: PCI device to be initialized
1275  *
1276  *  Initialize device before it's used by a driver. Ask low-level code
1277  *  to enable I/O and memory. Wake up the device if it was suspended.
1278  *  Beware, this function can fail.
1279  *
1280  *  Note we don't actually enable the device many times if we call
1281  *  this function repeatedly (we just increment the count).
1282  */
1283 int pci_enable_device(struct pci_dev *dev)
1284 {
1285 	return __pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1286 }
1287 
1288 /*
1289  * Managed PCI resources.  This manages device on/off, intx/msi/msix
1290  * on/off and BAR regions.  pci_dev itself records msi/msix status, so
1291  * there's no need to track it separately.  pci_devres is initialized
1292  * when a device is enabled using managed PCI device enable interface.
1293  */
1294 struct pci_devres {
1295 	unsigned int enabled:1;
1296 	unsigned int pinned:1;
1297 	unsigned int orig_intx:1;
1298 	unsigned int restore_intx:1;
1299 	u32 region_mask;
1300 };
1301 
1302 static void pcim_release(struct device *gendev, void *res)
1303 {
1304 	struct pci_dev *dev = container_of(gendev, struct pci_dev, dev);
1305 	struct pci_devres *this = res;
1306 	int i;
1307 
1308 	if (dev->msi_enabled)
1309 		pci_disable_msi(dev);
1310 	if (dev->msix_enabled)
1311 		pci_disable_msix(dev);
1312 
1313 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1314 		if (this->region_mask & (1 << i))
1315 			pci_release_region(dev, i);
1316 
1317 	if (this->restore_intx)
1318 		pci_intx(dev, this->orig_intx);
1319 
1320 	if (this->enabled && !this->pinned)
1321 		pci_disable_device(dev);
1322 }
1323 
1324 static struct pci_devres * get_pci_dr(struct pci_dev *pdev)
1325 {
1326 	struct pci_devres *dr, *new_dr;
1327 
1328 	dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1329 	if (dr)
1330 		return dr;
1331 
1332 	new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1333 	if (!new_dr)
1334 		return NULL;
1335 	return devres_get(&pdev->dev, new_dr, NULL, NULL);
1336 }
1337 
1338 static struct pci_devres * find_pci_dr(struct pci_dev *pdev)
1339 {
1340 	if (pci_is_managed(pdev))
1341 		return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1342 	return NULL;
1343 }
1344 
1345 /**
1346  * pcim_enable_device - Managed pci_enable_device()
1347  * @pdev: PCI device to be initialized
1348  *
1349  * Managed pci_enable_device().
1350  */
1351 int pcim_enable_device(struct pci_dev *pdev)
1352 {
1353 	struct pci_devres *dr;
1354 	int rc;
1355 
1356 	dr = get_pci_dr(pdev);
1357 	if (unlikely(!dr))
1358 		return -ENOMEM;
1359 	if (dr->enabled)
1360 		return 0;
1361 
1362 	rc = pci_enable_device(pdev);
1363 	if (!rc) {
1364 		pdev->is_managed = 1;
1365 		dr->enabled = 1;
1366 	}
1367 	return rc;
1368 }
1369 
1370 /**
1371  * pcim_pin_device - Pin managed PCI device
1372  * @pdev: PCI device to pin
1373  *
1374  * Pin managed PCI device @pdev.  Pinned device won't be disabled on
1375  * driver detach.  @pdev must have been enabled with
1376  * pcim_enable_device().
1377  */
1378 void pcim_pin_device(struct pci_dev *pdev)
1379 {
1380 	struct pci_devres *dr;
1381 
1382 	dr = find_pci_dr(pdev);
1383 	WARN_ON(!dr || !dr->enabled);
1384 	if (dr)
1385 		dr->pinned = 1;
1386 }
1387 
1388 /**
1389  * pcibios_disable_device - disable arch specific PCI resources for device dev
1390  * @dev: the PCI device to disable
1391  *
1392  * Disables architecture specific PCI resources for the device. This
1393  * is the default implementation. Architecture implementations can
1394  * override this.
1395  */
1396 void __weak pcibios_disable_device (struct pci_dev *dev) {}
1397 
1398 static void do_pci_disable_device(struct pci_dev *dev)
1399 {
1400 	u16 pci_command;
1401 
1402 	pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1403 	if (pci_command & PCI_COMMAND_MASTER) {
1404 		pci_command &= ~PCI_COMMAND_MASTER;
1405 		pci_write_config_word(dev, PCI_COMMAND, pci_command);
1406 	}
1407 
1408 	pcibios_disable_device(dev);
1409 }
1410 
1411 /**
1412  * pci_disable_enabled_device - Disable device without updating enable_cnt
1413  * @dev: PCI device to disable
1414  *
1415  * NOTE: This function is a backend of PCI power management routines and is
1416  * not supposed to be called drivers.
1417  */
1418 void pci_disable_enabled_device(struct pci_dev *dev)
1419 {
1420 	if (pci_is_enabled(dev))
1421 		do_pci_disable_device(dev);
1422 }
1423 
1424 /**
1425  * pci_disable_device - Disable PCI device after use
1426  * @dev: PCI device to be disabled
1427  *
1428  * Signal to the system that the PCI device is not in use by the system
1429  * anymore.  This only involves disabling PCI bus-mastering, if active.
1430  *
1431  * Note we don't actually disable the device until all callers of
1432  * pci_enable_device() have called pci_disable_device().
1433  */
1434 void
1435 pci_disable_device(struct pci_dev *dev)
1436 {
1437 	struct pci_devres *dr;
1438 
1439 	dr = find_pci_dr(dev);
1440 	if (dr)
1441 		dr->enabled = 0;
1442 
1443 	if (atomic_sub_return(1, &dev->enable_cnt) != 0)
1444 		return;
1445 
1446 	do_pci_disable_device(dev);
1447 
1448 	dev->is_busmaster = 0;
1449 }
1450 
1451 /**
1452  * pcibios_set_pcie_reset_state - set reset state for device dev
1453  * @dev: the PCIe device reset
1454  * @state: Reset state to enter into
1455  *
1456  *
1457  * Sets the PCIe reset state for the device. This is the default
1458  * implementation. Architecture implementations can override this.
1459  */
1460 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1461 					enum pcie_reset_state state)
1462 {
1463 	return -EINVAL;
1464 }
1465 
1466 /**
1467  * pci_set_pcie_reset_state - set reset state for device dev
1468  * @dev: the PCIe device reset
1469  * @state: Reset state to enter into
1470  *
1471  *
1472  * Sets the PCI reset state for the device.
1473  */
1474 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1475 {
1476 	return pcibios_set_pcie_reset_state(dev, state);
1477 }
1478 
1479 /**
1480  * pci_check_pme_status - Check if given device has generated PME.
1481  * @dev: Device to check.
1482  *
1483  * Check the PME status of the device and if set, clear it and clear PME enable
1484  * (if set).  Return 'true' if PME status and PME enable were both set or
1485  * 'false' otherwise.
1486  */
1487 bool pci_check_pme_status(struct pci_dev *dev)
1488 {
1489 	int pmcsr_pos;
1490 	u16 pmcsr;
1491 	bool ret = false;
1492 
1493 	if (!dev->pm_cap)
1494 		return false;
1495 
1496 	pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1497 	pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1498 	if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1499 		return false;
1500 
1501 	/* Clear PME status. */
1502 	pmcsr |= PCI_PM_CTRL_PME_STATUS;
1503 	if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1504 		/* Disable PME to avoid interrupt flood. */
1505 		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1506 		ret = true;
1507 	}
1508 
1509 	pci_write_config_word(dev, pmcsr_pos, pmcsr);
1510 
1511 	return ret;
1512 }
1513 
1514 /**
1515  * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1516  * @dev: Device to handle.
1517  * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1518  *
1519  * Check if @dev has generated PME and queue a resume request for it in that
1520  * case.
1521  */
1522 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1523 {
1524 	if (pme_poll_reset && dev->pme_poll)
1525 		dev->pme_poll = false;
1526 
1527 	if (pci_check_pme_status(dev)) {
1528 		pci_wakeup_event(dev);
1529 		pm_request_resume(&dev->dev);
1530 	}
1531 	return 0;
1532 }
1533 
1534 /**
1535  * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1536  * @bus: Top bus of the subtree to walk.
1537  */
1538 void pci_pme_wakeup_bus(struct pci_bus *bus)
1539 {
1540 	if (bus)
1541 		pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1542 }
1543 
1544 /**
1545  * pci_wakeup - Wake up a PCI device
1546  * @dev: Device to handle.
1547  * @ign: ignored parameter
1548  */
1549 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
1550 {
1551 	pci_wakeup_event(pci_dev);
1552 	pm_request_resume(&pci_dev->dev);
1553 	return 0;
1554 }
1555 
1556 /**
1557  * pci_wakeup_bus - Walk given bus and wake up devices on it
1558  * @bus: Top bus of the subtree to walk.
1559  */
1560 void pci_wakeup_bus(struct pci_bus *bus)
1561 {
1562 	if (bus)
1563 		pci_walk_bus(bus, pci_wakeup, NULL);
1564 }
1565 
1566 /**
1567  * pci_pme_capable - check the capability of PCI device to generate PME#
1568  * @dev: PCI device to handle.
1569  * @state: PCI state from which device will issue PME#.
1570  */
1571 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1572 {
1573 	if (!dev->pm_cap)
1574 		return false;
1575 
1576 	return !!(dev->pme_support & (1 << state));
1577 }
1578 
1579 static void pci_pme_list_scan(struct work_struct *work)
1580 {
1581 	struct pci_pme_device *pme_dev, *n;
1582 
1583 	mutex_lock(&pci_pme_list_mutex);
1584 	if (!list_empty(&pci_pme_list)) {
1585 		list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1586 			if (pme_dev->dev->pme_poll) {
1587 				struct pci_dev *bridge;
1588 
1589 				bridge = pme_dev->dev->bus->self;
1590 				/*
1591 				 * If bridge is in low power state, the
1592 				 * configuration space of subordinate devices
1593 				 * may be not accessible
1594 				 */
1595 				if (bridge && bridge->current_state != PCI_D0)
1596 					continue;
1597 				pci_pme_wakeup(pme_dev->dev, NULL);
1598 			} else {
1599 				list_del(&pme_dev->list);
1600 				kfree(pme_dev);
1601 			}
1602 		}
1603 		if (!list_empty(&pci_pme_list))
1604 			schedule_delayed_work(&pci_pme_work,
1605 					      msecs_to_jiffies(PME_TIMEOUT));
1606 	}
1607 	mutex_unlock(&pci_pme_list_mutex);
1608 }
1609 
1610 /**
1611  * pci_pme_active - enable or disable PCI device's PME# function
1612  * @dev: PCI device to handle.
1613  * @enable: 'true' to enable PME# generation; 'false' to disable it.
1614  *
1615  * The caller must verify that the device is capable of generating PME# before
1616  * calling this function with @enable equal to 'true'.
1617  */
1618 void pci_pme_active(struct pci_dev *dev, bool enable)
1619 {
1620 	u16 pmcsr;
1621 
1622 	if (!dev->pm_cap)
1623 		return;
1624 
1625 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1626 	/* Clear PME_Status by writing 1 to it and enable PME# */
1627 	pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1628 	if (!enable)
1629 		pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1630 
1631 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1632 
1633 	/* PCI (as opposed to PCIe) PME requires that the device have
1634 	   its PME# line hooked up correctly. Not all hardware vendors
1635 	   do this, so the PME never gets delivered and the device
1636 	   remains asleep. The easiest way around this is to
1637 	   periodically walk the list of suspended devices and check
1638 	   whether any have their PME flag set. The assumption is that
1639 	   we'll wake up often enough anyway that this won't be a huge
1640 	   hit, and the power savings from the devices will still be a
1641 	   win. */
1642 
1643 	if (dev->pme_poll) {
1644 		struct pci_pme_device *pme_dev;
1645 		if (enable) {
1646 			pme_dev = kmalloc(sizeof(struct pci_pme_device),
1647 					  GFP_KERNEL);
1648 			if (!pme_dev)
1649 				goto out;
1650 			pme_dev->dev = dev;
1651 			mutex_lock(&pci_pme_list_mutex);
1652 			list_add(&pme_dev->list, &pci_pme_list);
1653 			if (list_is_singular(&pci_pme_list))
1654 				schedule_delayed_work(&pci_pme_work,
1655 						      msecs_to_jiffies(PME_TIMEOUT));
1656 			mutex_unlock(&pci_pme_list_mutex);
1657 		} else {
1658 			mutex_lock(&pci_pme_list_mutex);
1659 			list_for_each_entry(pme_dev, &pci_pme_list, list) {
1660 				if (pme_dev->dev == dev) {
1661 					list_del(&pme_dev->list);
1662 					kfree(pme_dev);
1663 					break;
1664 				}
1665 			}
1666 			mutex_unlock(&pci_pme_list_mutex);
1667 		}
1668 	}
1669 
1670 out:
1671 	dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1672 }
1673 
1674 /**
1675  * __pci_enable_wake - enable PCI device as wakeup event source
1676  * @dev: PCI device affected
1677  * @state: PCI state from which device will issue wakeup events
1678  * @runtime: True if the events are to be generated at run time
1679  * @enable: True to enable event generation; false to disable
1680  *
1681  * This enables the device as a wakeup event source, or disables it.
1682  * When such events involves platform-specific hooks, those hooks are
1683  * called automatically by this routine.
1684  *
1685  * Devices with legacy power management (no standard PCI PM capabilities)
1686  * always require such platform hooks.
1687  *
1688  * RETURN VALUE:
1689  * 0 is returned on success
1690  * -EINVAL is returned if device is not supposed to wake up the system
1691  * Error code depending on the platform is returned if both the platform and
1692  * the native mechanism fail to enable the generation of wake-up events
1693  */
1694 int __pci_enable_wake(struct pci_dev *dev, pci_power_t state,
1695 		      bool runtime, bool enable)
1696 {
1697 	int ret = 0;
1698 
1699 	if (enable && !runtime && !device_may_wakeup(&dev->dev))
1700 		return -EINVAL;
1701 
1702 	/* Don't do the same thing twice in a row for one device. */
1703 	if (!!enable == !!dev->wakeup_prepared)
1704 		return 0;
1705 
1706 	/*
1707 	 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1708 	 * Anderson we should be doing PME# wake enable followed by ACPI wake
1709 	 * enable.  To disable wake-up we call the platform first, for symmetry.
1710 	 */
1711 
1712 	if (enable) {
1713 		int error;
1714 
1715 		if (pci_pme_capable(dev, state))
1716 			pci_pme_active(dev, true);
1717 		else
1718 			ret = 1;
1719 		error = runtime ? platform_pci_run_wake(dev, true) :
1720 					platform_pci_sleep_wake(dev, true);
1721 		if (ret)
1722 			ret = error;
1723 		if (!ret)
1724 			dev->wakeup_prepared = true;
1725 	} else {
1726 		if (runtime)
1727 			platform_pci_run_wake(dev, false);
1728 		else
1729 			platform_pci_sleep_wake(dev, false);
1730 		pci_pme_active(dev, false);
1731 		dev->wakeup_prepared = false;
1732 	}
1733 
1734 	return ret;
1735 }
1736 EXPORT_SYMBOL(__pci_enable_wake);
1737 
1738 /**
1739  * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1740  * @dev: PCI device to prepare
1741  * @enable: True to enable wake-up event generation; false to disable
1742  *
1743  * Many drivers want the device to wake up the system from D3_hot or D3_cold
1744  * and this function allows them to set that up cleanly - pci_enable_wake()
1745  * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1746  * ordering constraints.
1747  *
1748  * This function only returns error code if the device is not capable of
1749  * generating PME# from both D3_hot and D3_cold, and the platform is unable to
1750  * enable wake-up power for it.
1751  */
1752 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1753 {
1754 	return pci_pme_capable(dev, PCI_D3cold) ?
1755 			pci_enable_wake(dev, PCI_D3cold, enable) :
1756 			pci_enable_wake(dev, PCI_D3hot, enable);
1757 }
1758 
1759 /**
1760  * pci_target_state - find an appropriate low power state for a given PCI dev
1761  * @dev: PCI device
1762  *
1763  * Use underlying platform code to find a supported low power state for @dev.
1764  * If the platform can't manage @dev, return the deepest state from which it
1765  * can generate wake events, based on any available PME info.
1766  */
1767 pci_power_t pci_target_state(struct pci_dev *dev)
1768 {
1769 	pci_power_t target_state = PCI_D3hot;
1770 
1771 	if (platform_pci_power_manageable(dev)) {
1772 		/*
1773 		 * Call the platform to choose the target state of the device
1774 		 * and enable wake-up from this state if supported.
1775 		 */
1776 		pci_power_t state = platform_pci_choose_state(dev);
1777 
1778 		switch (state) {
1779 		case PCI_POWER_ERROR:
1780 		case PCI_UNKNOWN:
1781 			break;
1782 		case PCI_D1:
1783 		case PCI_D2:
1784 			if (pci_no_d1d2(dev))
1785 				break;
1786 		default:
1787 			target_state = state;
1788 		}
1789 	} else if (!dev->pm_cap) {
1790 		target_state = PCI_D0;
1791 	} else if (device_may_wakeup(&dev->dev)) {
1792 		/*
1793 		 * Find the deepest state from which the device can generate
1794 		 * wake-up events, make it the target state and enable device
1795 		 * to generate PME#.
1796 		 */
1797 		if (dev->pme_support) {
1798 			while (target_state
1799 			      && !(dev->pme_support & (1 << target_state)))
1800 				target_state--;
1801 		}
1802 	}
1803 
1804 	return target_state;
1805 }
1806 
1807 /**
1808  * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
1809  * @dev: Device to handle.
1810  *
1811  * Choose the power state appropriate for the device depending on whether
1812  * it can wake up the system and/or is power manageable by the platform
1813  * (PCI_D3hot is the default) and put the device into that state.
1814  */
1815 int pci_prepare_to_sleep(struct pci_dev *dev)
1816 {
1817 	pci_power_t target_state = pci_target_state(dev);
1818 	int error;
1819 
1820 	if (target_state == PCI_POWER_ERROR)
1821 		return -EIO;
1822 
1823 	/* D3cold during system suspend/hibernate is not supported */
1824 	if (target_state > PCI_D3hot)
1825 		target_state = PCI_D3hot;
1826 
1827 	pci_enable_wake(dev, target_state, device_may_wakeup(&dev->dev));
1828 
1829 	error = pci_set_power_state(dev, target_state);
1830 
1831 	if (error)
1832 		pci_enable_wake(dev, target_state, false);
1833 
1834 	return error;
1835 }
1836 
1837 /**
1838  * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
1839  * @dev: Device to handle.
1840  *
1841  * Disable device's system wake-up capability and put it into D0.
1842  */
1843 int pci_back_from_sleep(struct pci_dev *dev)
1844 {
1845 	pci_enable_wake(dev, PCI_D0, false);
1846 	return pci_set_power_state(dev, PCI_D0);
1847 }
1848 
1849 /**
1850  * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
1851  * @dev: PCI device being suspended.
1852  *
1853  * Prepare @dev to generate wake-up events at run time and put it into a low
1854  * power state.
1855  */
1856 int pci_finish_runtime_suspend(struct pci_dev *dev)
1857 {
1858 	pci_power_t target_state = pci_target_state(dev);
1859 	int error;
1860 
1861 	if (target_state == PCI_POWER_ERROR)
1862 		return -EIO;
1863 
1864 	dev->runtime_d3cold = target_state == PCI_D3cold;
1865 
1866 	__pci_enable_wake(dev, target_state, true, pci_dev_run_wake(dev));
1867 
1868 	error = pci_set_power_state(dev, target_state);
1869 
1870 	if (error) {
1871 		__pci_enable_wake(dev, target_state, true, false);
1872 		dev->runtime_d3cold = false;
1873 	}
1874 
1875 	return error;
1876 }
1877 
1878 /**
1879  * pci_dev_run_wake - Check if device can generate run-time wake-up events.
1880  * @dev: Device to check.
1881  *
1882  * Return true if the device itself is cabable of generating wake-up events
1883  * (through the platform or using the native PCIe PME) or if the device supports
1884  * PME and one of its upstream bridges can generate wake-up events.
1885  */
1886 bool pci_dev_run_wake(struct pci_dev *dev)
1887 {
1888 	struct pci_bus *bus = dev->bus;
1889 
1890 	if (device_run_wake(&dev->dev))
1891 		return true;
1892 
1893 	if (!dev->pme_support)
1894 		return false;
1895 
1896 	while (bus->parent) {
1897 		struct pci_dev *bridge = bus->self;
1898 
1899 		if (device_run_wake(&bridge->dev))
1900 			return true;
1901 
1902 		bus = bus->parent;
1903 	}
1904 
1905 	/* We have reached the root bus. */
1906 	if (bus->bridge)
1907 		return device_run_wake(bus->bridge);
1908 
1909 	return false;
1910 }
1911 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
1912 
1913 /**
1914  * pci_pm_init - Initialize PM functions of given PCI device
1915  * @dev: PCI device to handle.
1916  */
1917 void pci_pm_init(struct pci_dev *dev)
1918 {
1919 	int pm;
1920 	u16 pmc;
1921 
1922 	pm_runtime_forbid(&dev->dev);
1923 	device_enable_async_suspend(&dev->dev);
1924 	dev->wakeup_prepared = false;
1925 
1926 	dev->pm_cap = 0;
1927 
1928 	/* find PCI PM capability in list */
1929 	pm = pci_find_capability(dev, PCI_CAP_ID_PM);
1930 	if (!pm)
1931 		return;
1932 	/* Check device's ability to generate PME# */
1933 	pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
1934 
1935 	if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
1936 		dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
1937 			pmc & PCI_PM_CAP_VER_MASK);
1938 		return;
1939 	}
1940 
1941 	dev->pm_cap = pm;
1942 	dev->d3_delay = PCI_PM_D3_WAIT;
1943 	dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
1944 
1945 	dev->d1_support = false;
1946 	dev->d2_support = false;
1947 	if (!pci_no_d1d2(dev)) {
1948 		if (pmc & PCI_PM_CAP_D1)
1949 			dev->d1_support = true;
1950 		if (pmc & PCI_PM_CAP_D2)
1951 			dev->d2_support = true;
1952 
1953 		if (dev->d1_support || dev->d2_support)
1954 			dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
1955 				   dev->d1_support ? " D1" : "",
1956 				   dev->d2_support ? " D2" : "");
1957 	}
1958 
1959 	pmc &= PCI_PM_CAP_PME_MASK;
1960 	if (pmc) {
1961 		dev_printk(KERN_DEBUG, &dev->dev,
1962 			 "PME# supported from%s%s%s%s%s\n",
1963 			 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
1964 			 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
1965 			 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
1966 			 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
1967 			 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
1968 		dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
1969 		dev->pme_poll = true;
1970 		/*
1971 		 * Make device's PM flags reflect the wake-up capability, but
1972 		 * let the user space enable it to wake up the system as needed.
1973 		 */
1974 		device_set_wakeup_capable(&dev->dev, true);
1975 		/* Disable the PME# generation functionality */
1976 		pci_pme_active(dev, false);
1977 	} else {
1978 		dev->pme_support = 0;
1979 	}
1980 }
1981 
1982 /**
1983  * platform_pci_wakeup_init - init platform wakeup if present
1984  * @dev: PCI device
1985  *
1986  * Some devices don't have PCI PM caps but can still generate wakeup
1987  * events through platform methods (like ACPI events).  If @dev supports
1988  * platform wakeup events, set the device flag to indicate as much.  This
1989  * may be redundant if the device also supports PCI PM caps, but double
1990  * initialization should be safe in that case.
1991  */
1992 void platform_pci_wakeup_init(struct pci_dev *dev)
1993 {
1994 	if (!platform_pci_can_wakeup(dev))
1995 		return;
1996 
1997 	device_set_wakeup_capable(&dev->dev, true);
1998 	platform_pci_sleep_wake(dev, false);
1999 }
2000 
2001 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2002 	struct pci_cap_saved_state *new_cap)
2003 {
2004 	hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2005 }
2006 
2007 /**
2008  * pci_add_save_buffer - allocate buffer for saving given capability registers
2009  * @dev: the PCI device
2010  * @cap: the capability to allocate the buffer for
2011  * @size: requested size of the buffer
2012  */
2013 static int pci_add_cap_save_buffer(
2014 	struct pci_dev *dev, char cap, unsigned int size)
2015 {
2016 	int pos;
2017 	struct pci_cap_saved_state *save_state;
2018 
2019 	pos = pci_find_capability(dev, cap);
2020 	if (pos <= 0)
2021 		return 0;
2022 
2023 	save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2024 	if (!save_state)
2025 		return -ENOMEM;
2026 
2027 	save_state->cap.cap_nr = cap;
2028 	save_state->cap.size = size;
2029 	pci_add_saved_cap(dev, save_state);
2030 
2031 	return 0;
2032 }
2033 
2034 /**
2035  * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2036  * @dev: the PCI device
2037  */
2038 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2039 {
2040 	int error;
2041 
2042 	error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2043 					PCI_EXP_SAVE_REGS * sizeof(u16));
2044 	if (error)
2045 		dev_err(&dev->dev,
2046 			"unable to preallocate PCI Express save buffer\n");
2047 
2048 	error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2049 	if (error)
2050 		dev_err(&dev->dev,
2051 			"unable to preallocate PCI-X save buffer\n");
2052 }
2053 
2054 void pci_free_cap_save_buffers(struct pci_dev *dev)
2055 {
2056 	struct pci_cap_saved_state *tmp;
2057 	struct hlist_node *pos, *n;
2058 
2059 	hlist_for_each_entry_safe(tmp, pos, n, &dev->saved_cap_space, next)
2060 		kfree(tmp);
2061 }
2062 
2063 /**
2064  * pci_enable_ari - enable ARI forwarding if hardware support it
2065  * @dev: the PCI device
2066  */
2067 void pci_enable_ari(struct pci_dev *dev)
2068 {
2069 	int pos;
2070 	u32 cap;
2071 	u16 ctrl;
2072 	struct pci_dev *bridge;
2073 
2074 	if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2075 		return;
2076 
2077 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
2078 	if (!pos)
2079 		return;
2080 
2081 	bridge = dev->bus->self;
2082 	if (!bridge)
2083 		return;
2084 
2085 	/* ARI is a PCIe cap v2 feature */
2086 	pos = pci_pcie_cap2(bridge);
2087 	if (!pos)
2088 		return;
2089 
2090 	pci_read_config_dword(bridge, pos + PCI_EXP_DEVCAP2, &cap);
2091 	if (!(cap & PCI_EXP_DEVCAP2_ARI))
2092 		return;
2093 
2094 	pci_read_config_word(bridge, pos + PCI_EXP_DEVCTL2, &ctrl);
2095 	ctrl |= PCI_EXP_DEVCTL2_ARI;
2096 	pci_write_config_word(bridge, pos + PCI_EXP_DEVCTL2, ctrl);
2097 
2098 	bridge->ari_enabled = 1;
2099 }
2100 
2101 /**
2102  * pci_enable_ido - enable ID-based Ordering on a device
2103  * @dev: the PCI device
2104  * @type: which types of IDO to enable
2105  *
2106  * Enable ID-based ordering on @dev.  @type can contain the bits
2107  * %PCI_EXP_IDO_REQUEST and/or %PCI_EXP_IDO_COMPLETION to indicate
2108  * which types of transactions are allowed to be re-ordered.
2109  */
2110 void pci_enable_ido(struct pci_dev *dev, unsigned long type)
2111 {
2112 	int pos;
2113 	u16 ctrl;
2114 
2115 	/* ID-based Ordering is a PCIe cap v2 feature */
2116 	pos = pci_pcie_cap2(dev);
2117 	if (!pos)
2118 		return;
2119 
2120 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2121 	if (type & PCI_EXP_IDO_REQUEST)
2122 		ctrl |= PCI_EXP_IDO_REQ_EN;
2123 	if (type & PCI_EXP_IDO_COMPLETION)
2124 		ctrl |= PCI_EXP_IDO_CMP_EN;
2125 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2126 }
2127 EXPORT_SYMBOL(pci_enable_ido);
2128 
2129 /**
2130  * pci_disable_ido - disable ID-based ordering on a device
2131  * @dev: the PCI device
2132  * @type: which types of IDO to disable
2133  */
2134 void pci_disable_ido(struct pci_dev *dev, unsigned long type)
2135 {
2136 	int pos;
2137 	u16 ctrl;
2138 
2139 	/* ID-based Ordering is a PCIe cap v2 feature */
2140 	pos = pci_pcie_cap2(dev);
2141 	if (!pos)
2142 		return;
2143 
2144 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2145 	if (type & PCI_EXP_IDO_REQUEST)
2146 		ctrl &= ~PCI_EXP_IDO_REQ_EN;
2147 	if (type & PCI_EXP_IDO_COMPLETION)
2148 		ctrl &= ~PCI_EXP_IDO_CMP_EN;
2149 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2150 }
2151 EXPORT_SYMBOL(pci_disable_ido);
2152 
2153 /**
2154  * pci_enable_obff - enable optimized buffer flush/fill
2155  * @dev: PCI device
2156  * @type: type of signaling to use
2157  *
2158  * Try to enable @type OBFF signaling on @dev.  It will try using WAKE#
2159  * signaling if possible, falling back to message signaling only if
2160  * WAKE# isn't supported.  @type should indicate whether the PCIe link
2161  * be brought out of L0s or L1 to send the message.  It should be either
2162  * %PCI_EXP_OBFF_SIGNAL_ALWAYS or %PCI_OBFF_SIGNAL_L0.
2163  *
2164  * If your device can benefit from receiving all messages, even at the
2165  * power cost of bringing the link back up from a low power state, use
2166  * %PCI_EXP_OBFF_SIGNAL_ALWAYS.  Otherwise, use %PCI_OBFF_SIGNAL_L0 (the
2167  * preferred type).
2168  *
2169  * RETURNS:
2170  * Zero on success, appropriate error number on failure.
2171  */
2172 int pci_enable_obff(struct pci_dev *dev, enum pci_obff_signal_type type)
2173 {
2174 	int pos;
2175 	u32 cap;
2176 	u16 ctrl;
2177 	int ret;
2178 
2179 	/* OBFF is a PCIe cap v2 feature */
2180 	pos = pci_pcie_cap2(dev);
2181 	if (!pos)
2182 		return -ENOTSUPP;
2183 
2184 	pci_read_config_dword(dev, pos + PCI_EXP_DEVCAP2, &cap);
2185 	if (!(cap & PCI_EXP_OBFF_MASK))
2186 		return -ENOTSUPP; /* no OBFF support at all */
2187 
2188 	/* Make sure the topology supports OBFF as well */
2189 	if (dev->bus->self) {
2190 		ret = pci_enable_obff(dev->bus->self, type);
2191 		if (ret)
2192 			return ret;
2193 	}
2194 
2195 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2196 	if (cap & PCI_EXP_OBFF_WAKE)
2197 		ctrl |= PCI_EXP_OBFF_WAKE_EN;
2198 	else {
2199 		switch (type) {
2200 		case PCI_EXP_OBFF_SIGNAL_L0:
2201 			if (!(ctrl & PCI_EXP_OBFF_WAKE_EN))
2202 				ctrl |= PCI_EXP_OBFF_MSGA_EN;
2203 			break;
2204 		case PCI_EXP_OBFF_SIGNAL_ALWAYS:
2205 			ctrl &= ~PCI_EXP_OBFF_WAKE_EN;
2206 			ctrl |= PCI_EXP_OBFF_MSGB_EN;
2207 			break;
2208 		default:
2209 			WARN(1, "bad OBFF signal type\n");
2210 			return -ENOTSUPP;
2211 		}
2212 	}
2213 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2214 
2215 	return 0;
2216 }
2217 EXPORT_SYMBOL(pci_enable_obff);
2218 
2219 /**
2220  * pci_disable_obff - disable optimized buffer flush/fill
2221  * @dev: PCI device
2222  *
2223  * Disable OBFF on @dev.
2224  */
2225 void pci_disable_obff(struct pci_dev *dev)
2226 {
2227 	int pos;
2228 	u16 ctrl;
2229 
2230 	/* OBFF is a PCIe cap v2 feature */
2231 	pos = pci_pcie_cap2(dev);
2232 	if (!pos)
2233 		return;
2234 
2235 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2236 	ctrl &= ~PCI_EXP_OBFF_WAKE_EN;
2237 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2238 }
2239 EXPORT_SYMBOL(pci_disable_obff);
2240 
2241 /**
2242  * pci_ltr_supported - check whether a device supports LTR
2243  * @dev: PCI device
2244  *
2245  * RETURNS:
2246  * True if @dev supports latency tolerance reporting, false otherwise.
2247  */
2248 static bool pci_ltr_supported(struct pci_dev *dev)
2249 {
2250 	int pos;
2251 	u32 cap;
2252 
2253 	/* LTR is a PCIe cap v2 feature */
2254 	pos = pci_pcie_cap2(dev);
2255 	if (!pos)
2256 		return false;
2257 
2258 	pci_read_config_dword(dev, pos + PCI_EXP_DEVCAP2, &cap);
2259 
2260 	return cap & PCI_EXP_DEVCAP2_LTR;
2261 }
2262 
2263 /**
2264  * pci_enable_ltr - enable latency tolerance reporting
2265  * @dev: PCI device
2266  *
2267  * Enable LTR on @dev if possible, which means enabling it first on
2268  * upstream ports.
2269  *
2270  * RETURNS:
2271  * Zero on success, errno on failure.
2272  */
2273 int pci_enable_ltr(struct pci_dev *dev)
2274 {
2275 	int pos;
2276 	u16 ctrl;
2277 	int ret;
2278 
2279 	if (!pci_ltr_supported(dev))
2280 		return -ENOTSUPP;
2281 
2282 	/* LTR is a PCIe cap v2 feature */
2283 	pos = pci_pcie_cap2(dev);
2284 	if (!pos)
2285 		return -ENOTSUPP;
2286 
2287 	/* Only primary function can enable/disable LTR */
2288 	if (PCI_FUNC(dev->devfn) != 0)
2289 		return -EINVAL;
2290 
2291 	/* Enable upstream ports first */
2292 	if (dev->bus->self) {
2293 		ret = pci_enable_ltr(dev->bus->self);
2294 		if (ret)
2295 			return ret;
2296 	}
2297 
2298 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2299 	ctrl |= PCI_EXP_LTR_EN;
2300 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2301 
2302 	return 0;
2303 }
2304 EXPORT_SYMBOL(pci_enable_ltr);
2305 
2306 /**
2307  * pci_disable_ltr - disable latency tolerance reporting
2308  * @dev: PCI device
2309  */
2310 void pci_disable_ltr(struct pci_dev *dev)
2311 {
2312 	int pos;
2313 	u16 ctrl;
2314 
2315 	if (!pci_ltr_supported(dev))
2316 		return;
2317 
2318 	/* LTR is a PCIe cap v2 feature */
2319 	pos = pci_pcie_cap2(dev);
2320 	if (!pos)
2321 		return;
2322 
2323 	/* Only primary function can enable/disable LTR */
2324 	if (PCI_FUNC(dev->devfn) != 0)
2325 		return;
2326 
2327 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2328 	ctrl &= ~PCI_EXP_LTR_EN;
2329 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2330 }
2331 EXPORT_SYMBOL(pci_disable_ltr);
2332 
2333 static int __pci_ltr_scale(int *val)
2334 {
2335 	int scale = 0;
2336 
2337 	while (*val > 1023) {
2338 		*val = (*val + 31) / 32;
2339 		scale++;
2340 	}
2341 	return scale;
2342 }
2343 
2344 /**
2345  * pci_set_ltr - set LTR latency values
2346  * @dev: PCI device
2347  * @snoop_lat_ns: snoop latency in nanoseconds
2348  * @nosnoop_lat_ns: nosnoop latency in nanoseconds
2349  *
2350  * Figure out the scale and set the LTR values accordingly.
2351  */
2352 int pci_set_ltr(struct pci_dev *dev, int snoop_lat_ns, int nosnoop_lat_ns)
2353 {
2354 	int pos, ret, snoop_scale, nosnoop_scale;
2355 	u16 val;
2356 
2357 	if (!pci_ltr_supported(dev))
2358 		return -ENOTSUPP;
2359 
2360 	snoop_scale = __pci_ltr_scale(&snoop_lat_ns);
2361 	nosnoop_scale = __pci_ltr_scale(&nosnoop_lat_ns);
2362 
2363 	if (snoop_lat_ns > PCI_LTR_VALUE_MASK ||
2364 	    nosnoop_lat_ns > PCI_LTR_VALUE_MASK)
2365 		return -EINVAL;
2366 
2367 	if ((snoop_scale > (PCI_LTR_SCALE_MASK >> PCI_LTR_SCALE_SHIFT)) ||
2368 	    (nosnoop_scale > (PCI_LTR_SCALE_MASK >> PCI_LTR_SCALE_SHIFT)))
2369 		return -EINVAL;
2370 
2371 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
2372 	if (!pos)
2373 		return -ENOTSUPP;
2374 
2375 	val = (snoop_scale << PCI_LTR_SCALE_SHIFT) | snoop_lat_ns;
2376 	ret = pci_write_config_word(dev, pos + PCI_LTR_MAX_SNOOP_LAT, val);
2377 	if (ret != 4)
2378 		return -EIO;
2379 
2380 	val = (nosnoop_scale << PCI_LTR_SCALE_SHIFT) | nosnoop_lat_ns;
2381 	ret = pci_write_config_word(dev, pos + PCI_LTR_MAX_NOSNOOP_LAT, val);
2382 	if (ret != 4)
2383 		return -EIO;
2384 
2385 	return 0;
2386 }
2387 EXPORT_SYMBOL(pci_set_ltr);
2388 
2389 static int pci_acs_enable;
2390 
2391 /**
2392  * pci_request_acs - ask for ACS to be enabled if supported
2393  */
2394 void pci_request_acs(void)
2395 {
2396 	pci_acs_enable = 1;
2397 }
2398 
2399 /**
2400  * pci_enable_acs - enable ACS if hardware support it
2401  * @dev: the PCI device
2402  */
2403 void pci_enable_acs(struct pci_dev *dev)
2404 {
2405 	int pos;
2406 	u16 cap;
2407 	u16 ctrl;
2408 
2409 	if (!pci_acs_enable)
2410 		return;
2411 
2412 	if (!pci_is_pcie(dev))
2413 		return;
2414 
2415 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2416 	if (!pos)
2417 		return;
2418 
2419 	pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2420 	pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2421 
2422 	/* Source Validation */
2423 	ctrl |= (cap & PCI_ACS_SV);
2424 
2425 	/* P2P Request Redirect */
2426 	ctrl |= (cap & PCI_ACS_RR);
2427 
2428 	/* P2P Completion Redirect */
2429 	ctrl |= (cap & PCI_ACS_CR);
2430 
2431 	/* Upstream Forwarding */
2432 	ctrl |= (cap & PCI_ACS_UF);
2433 
2434 	pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2435 }
2436 
2437 /**
2438  * pci_acs_enabled - test ACS against required flags for a given device
2439  * @pdev: device to test
2440  * @acs_flags: required PCI ACS flags
2441  *
2442  * Return true if the device supports the provided flags.  Automatically
2443  * filters out flags that are not implemented on multifunction devices.
2444  */
2445 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2446 {
2447 	int pos, ret;
2448 	u16 ctrl;
2449 
2450 	ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2451 	if (ret >= 0)
2452 		return ret > 0;
2453 
2454 	if (!pci_is_pcie(pdev))
2455 		return false;
2456 
2457 	/* Filter out flags not applicable to multifunction */
2458 	if (pdev->multifunction)
2459 		acs_flags &= (PCI_ACS_RR | PCI_ACS_CR |
2460 			      PCI_ACS_EC | PCI_ACS_DT);
2461 
2462 	if (pdev->pcie_type == PCI_EXP_TYPE_DOWNSTREAM ||
2463 	    pdev->pcie_type == PCI_EXP_TYPE_ROOT_PORT ||
2464 	    pdev->multifunction) {
2465 		pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2466 		if (!pos)
2467 			return false;
2468 
2469 		pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2470 		if ((ctrl & acs_flags) != acs_flags)
2471 			return false;
2472 	}
2473 
2474 	return true;
2475 }
2476 
2477 /**
2478  * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2479  * @start: starting downstream device
2480  * @end: ending upstream device or NULL to search to the root bus
2481  * @acs_flags: required flags
2482  *
2483  * Walk up a device tree from start to end testing PCI ACS support.  If
2484  * any step along the way does not support the required flags, return false.
2485  */
2486 bool pci_acs_path_enabled(struct pci_dev *start,
2487 			  struct pci_dev *end, u16 acs_flags)
2488 {
2489 	struct pci_dev *pdev, *parent = start;
2490 
2491 	do {
2492 		pdev = parent;
2493 
2494 		if (!pci_acs_enabled(pdev, acs_flags))
2495 			return false;
2496 
2497 		if (pci_is_root_bus(pdev->bus))
2498 			return (end == NULL);
2499 
2500 		parent = pdev->bus->self;
2501 	} while (pdev != end);
2502 
2503 	return true;
2504 }
2505 
2506 /**
2507  * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2508  * @dev: the PCI device
2509  * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2510  *
2511  * Perform INTx swizzling for a device behind one level of bridge.  This is
2512  * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2513  * behind bridges on add-in cards.  For devices with ARI enabled, the slot
2514  * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2515  * the PCI Express Base Specification, Revision 2.1)
2516  */
2517 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2518 {
2519 	int slot;
2520 
2521 	if (pci_ari_enabled(dev->bus))
2522 		slot = 0;
2523 	else
2524 		slot = PCI_SLOT(dev->devfn);
2525 
2526 	return (((pin - 1) + slot) % 4) + 1;
2527 }
2528 
2529 int
2530 pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
2531 {
2532 	u8 pin;
2533 
2534 	pin = dev->pin;
2535 	if (!pin)
2536 		return -1;
2537 
2538 	while (!pci_is_root_bus(dev->bus)) {
2539 		pin = pci_swizzle_interrupt_pin(dev, pin);
2540 		dev = dev->bus->self;
2541 	}
2542 	*bridge = dev;
2543 	return pin;
2544 }
2545 
2546 /**
2547  * pci_common_swizzle - swizzle INTx all the way to root bridge
2548  * @dev: the PCI device
2549  * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2550  *
2551  * Perform INTx swizzling for a device.  This traverses through all PCI-to-PCI
2552  * bridges all the way up to a PCI root bus.
2553  */
2554 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
2555 {
2556 	u8 pin = *pinp;
2557 
2558 	while (!pci_is_root_bus(dev->bus)) {
2559 		pin = pci_swizzle_interrupt_pin(dev, pin);
2560 		dev = dev->bus->self;
2561 	}
2562 	*pinp = pin;
2563 	return PCI_SLOT(dev->devfn);
2564 }
2565 
2566 /**
2567  *	pci_release_region - Release a PCI bar
2568  *	@pdev: PCI device whose resources were previously reserved by pci_request_region
2569  *	@bar: BAR to release
2570  *
2571  *	Releases the PCI I/O and memory resources previously reserved by a
2572  *	successful call to pci_request_region.  Call this function only
2573  *	after all use of the PCI regions has ceased.
2574  */
2575 void pci_release_region(struct pci_dev *pdev, int bar)
2576 {
2577 	struct pci_devres *dr;
2578 
2579 	if (pci_resource_len(pdev, bar) == 0)
2580 		return;
2581 	if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
2582 		release_region(pci_resource_start(pdev, bar),
2583 				pci_resource_len(pdev, bar));
2584 	else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
2585 		release_mem_region(pci_resource_start(pdev, bar),
2586 				pci_resource_len(pdev, bar));
2587 
2588 	dr = find_pci_dr(pdev);
2589 	if (dr)
2590 		dr->region_mask &= ~(1 << bar);
2591 }
2592 
2593 /**
2594  *	__pci_request_region - Reserved PCI I/O and memory resource
2595  *	@pdev: PCI device whose resources are to be reserved
2596  *	@bar: BAR to be reserved
2597  *	@res_name: Name to be associated with resource.
2598  *	@exclusive: whether the region access is exclusive or not
2599  *
2600  *	Mark the PCI region associated with PCI device @pdev BR @bar as
2601  *	being reserved by owner @res_name.  Do not access any
2602  *	address inside the PCI regions unless this call returns
2603  *	successfully.
2604  *
2605  *	If @exclusive is set, then the region is marked so that userspace
2606  *	is explicitly not allowed to map the resource via /dev/mem or
2607  * 	sysfs MMIO access.
2608  *
2609  *	Returns 0 on success, or %EBUSY on error.  A warning
2610  *	message is also printed on failure.
2611  */
2612 static int __pci_request_region(struct pci_dev *pdev, int bar, const char *res_name,
2613 									int exclusive)
2614 {
2615 	struct pci_devres *dr;
2616 
2617 	if (pci_resource_len(pdev, bar) == 0)
2618 		return 0;
2619 
2620 	if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
2621 		if (!request_region(pci_resource_start(pdev, bar),
2622 			    pci_resource_len(pdev, bar), res_name))
2623 			goto err_out;
2624 	}
2625 	else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
2626 		if (!__request_mem_region(pci_resource_start(pdev, bar),
2627 					pci_resource_len(pdev, bar), res_name,
2628 					exclusive))
2629 			goto err_out;
2630 	}
2631 
2632 	dr = find_pci_dr(pdev);
2633 	if (dr)
2634 		dr->region_mask |= 1 << bar;
2635 
2636 	return 0;
2637 
2638 err_out:
2639 	dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
2640 		 &pdev->resource[bar]);
2641 	return -EBUSY;
2642 }
2643 
2644 /**
2645  *	pci_request_region - Reserve PCI I/O and memory resource
2646  *	@pdev: PCI device whose resources are to be reserved
2647  *	@bar: BAR to be reserved
2648  *	@res_name: Name to be associated with resource
2649  *
2650  *	Mark the PCI region associated with PCI device @pdev BAR @bar as
2651  *	being reserved by owner @res_name.  Do not access any
2652  *	address inside the PCI regions unless this call returns
2653  *	successfully.
2654  *
2655  *	Returns 0 on success, or %EBUSY on error.  A warning
2656  *	message is also printed on failure.
2657  */
2658 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
2659 {
2660 	return __pci_request_region(pdev, bar, res_name, 0);
2661 }
2662 
2663 /**
2664  *	pci_request_region_exclusive - Reserved PCI I/O and memory resource
2665  *	@pdev: PCI device whose resources are to be reserved
2666  *	@bar: BAR to be reserved
2667  *	@res_name: Name to be associated with resource.
2668  *
2669  *	Mark the PCI region associated with PCI device @pdev BR @bar as
2670  *	being reserved by owner @res_name.  Do not access any
2671  *	address inside the PCI regions unless this call returns
2672  *	successfully.
2673  *
2674  *	Returns 0 on success, or %EBUSY on error.  A warning
2675  *	message is also printed on failure.
2676  *
2677  *	The key difference that _exclusive makes it that userspace is
2678  *	explicitly not allowed to map the resource via /dev/mem or
2679  * 	sysfs.
2680  */
2681 int pci_request_region_exclusive(struct pci_dev *pdev, int bar, const char *res_name)
2682 {
2683 	return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
2684 }
2685 /**
2686  * pci_release_selected_regions - Release selected PCI I/O and memory resources
2687  * @pdev: PCI device whose resources were previously reserved
2688  * @bars: Bitmask of BARs to be released
2689  *
2690  * Release selected PCI I/O and memory resources previously reserved.
2691  * Call this function only after all use of the PCI regions has ceased.
2692  */
2693 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
2694 {
2695 	int i;
2696 
2697 	for (i = 0; i < 6; i++)
2698 		if (bars & (1 << i))
2699 			pci_release_region(pdev, i);
2700 }
2701 
2702 int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
2703 				 const char *res_name, int excl)
2704 {
2705 	int i;
2706 
2707 	for (i = 0; i < 6; i++)
2708 		if (bars & (1 << i))
2709 			if (__pci_request_region(pdev, i, res_name, excl))
2710 				goto err_out;
2711 	return 0;
2712 
2713 err_out:
2714 	while(--i >= 0)
2715 		if (bars & (1 << i))
2716 			pci_release_region(pdev, i);
2717 
2718 	return -EBUSY;
2719 }
2720 
2721 
2722 /**
2723  * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
2724  * @pdev: PCI device whose resources are to be reserved
2725  * @bars: Bitmask of BARs to be requested
2726  * @res_name: Name to be associated with resource
2727  */
2728 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
2729 				 const char *res_name)
2730 {
2731 	return __pci_request_selected_regions(pdev, bars, res_name, 0);
2732 }
2733 
2734 int pci_request_selected_regions_exclusive(struct pci_dev *pdev,
2735 				 int bars, const char *res_name)
2736 {
2737 	return __pci_request_selected_regions(pdev, bars, res_name,
2738 			IORESOURCE_EXCLUSIVE);
2739 }
2740 
2741 /**
2742  *	pci_release_regions - Release reserved PCI I/O and memory resources
2743  *	@pdev: PCI device whose resources were previously reserved by pci_request_regions
2744  *
2745  *	Releases all PCI I/O and memory resources previously reserved by a
2746  *	successful call to pci_request_regions.  Call this function only
2747  *	after all use of the PCI regions has ceased.
2748  */
2749 
2750 void pci_release_regions(struct pci_dev *pdev)
2751 {
2752 	pci_release_selected_regions(pdev, (1 << 6) - 1);
2753 }
2754 
2755 /**
2756  *	pci_request_regions - Reserved PCI I/O and memory resources
2757  *	@pdev: PCI device whose resources are to be reserved
2758  *	@res_name: Name to be associated with resource.
2759  *
2760  *	Mark all PCI regions associated with PCI device @pdev as
2761  *	being reserved by owner @res_name.  Do not access any
2762  *	address inside the PCI regions unless this call returns
2763  *	successfully.
2764  *
2765  *	Returns 0 on success, or %EBUSY on error.  A warning
2766  *	message is also printed on failure.
2767  */
2768 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
2769 {
2770 	return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
2771 }
2772 
2773 /**
2774  *	pci_request_regions_exclusive - Reserved PCI I/O and memory resources
2775  *	@pdev: PCI device whose resources are to be reserved
2776  *	@res_name: Name to be associated with resource.
2777  *
2778  *	Mark all PCI regions associated with PCI device @pdev as
2779  *	being reserved by owner @res_name.  Do not access any
2780  *	address inside the PCI regions unless this call returns
2781  *	successfully.
2782  *
2783  *	pci_request_regions_exclusive() will mark the region so that
2784  * 	/dev/mem and the sysfs MMIO access will not be allowed.
2785  *
2786  *	Returns 0 on success, or %EBUSY on error.  A warning
2787  *	message is also printed on failure.
2788  */
2789 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
2790 {
2791 	return pci_request_selected_regions_exclusive(pdev,
2792 					((1 << 6) - 1), res_name);
2793 }
2794 
2795 static void __pci_set_master(struct pci_dev *dev, bool enable)
2796 {
2797 	u16 old_cmd, cmd;
2798 
2799 	pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
2800 	if (enable)
2801 		cmd = old_cmd | PCI_COMMAND_MASTER;
2802 	else
2803 		cmd = old_cmd & ~PCI_COMMAND_MASTER;
2804 	if (cmd != old_cmd) {
2805 		dev_dbg(&dev->dev, "%s bus mastering\n",
2806 			enable ? "enabling" : "disabling");
2807 		pci_write_config_word(dev, PCI_COMMAND, cmd);
2808 	}
2809 	dev->is_busmaster = enable;
2810 }
2811 
2812 /**
2813  * pcibios_setup - process "pci=" kernel boot arguments
2814  * @str: string used to pass in "pci=" kernel boot arguments
2815  *
2816  * Process kernel boot arguments.  This is the default implementation.
2817  * Architecture specific implementations can override this as necessary.
2818  */
2819 char * __weak __init pcibios_setup(char *str)
2820 {
2821 	return str;
2822 }
2823 
2824 /**
2825  * pcibios_set_master - enable PCI bus-mastering for device dev
2826  * @dev: the PCI device to enable
2827  *
2828  * Enables PCI bus-mastering for the device.  This is the default
2829  * implementation.  Architecture specific implementations can override
2830  * this if necessary.
2831  */
2832 void __weak pcibios_set_master(struct pci_dev *dev)
2833 {
2834 	u8 lat;
2835 
2836 	/* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
2837 	if (pci_is_pcie(dev))
2838 		return;
2839 
2840 	pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
2841 	if (lat < 16)
2842 		lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
2843 	else if (lat > pcibios_max_latency)
2844 		lat = pcibios_max_latency;
2845 	else
2846 		return;
2847 	dev_printk(KERN_DEBUG, &dev->dev, "setting latency timer to %d\n", lat);
2848 	pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
2849 }
2850 
2851 /**
2852  * pci_set_master - enables bus-mastering for device dev
2853  * @dev: the PCI device to enable
2854  *
2855  * Enables bus-mastering on the device and calls pcibios_set_master()
2856  * to do the needed arch specific settings.
2857  */
2858 void pci_set_master(struct pci_dev *dev)
2859 {
2860 	__pci_set_master(dev, true);
2861 	pcibios_set_master(dev);
2862 }
2863 
2864 /**
2865  * pci_clear_master - disables bus-mastering for device dev
2866  * @dev: the PCI device to disable
2867  */
2868 void pci_clear_master(struct pci_dev *dev)
2869 {
2870 	__pci_set_master(dev, false);
2871 }
2872 
2873 /**
2874  * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
2875  * @dev: the PCI device for which MWI is to be enabled
2876  *
2877  * Helper function for pci_set_mwi.
2878  * Originally copied from drivers/net/acenic.c.
2879  * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
2880  *
2881  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
2882  */
2883 int pci_set_cacheline_size(struct pci_dev *dev)
2884 {
2885 	u8 cacheline_size;
2886 
2887 	if (!pci_cache_line_size)
2888 		return -EINVAL;
2889 
2890 	/* Validate current setting: the PCI_CACHE_LINE_SIZE must be
2891 	   equal to or multiple of the right value. */
2892 	pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
2893 	if (cacheline_size >= pci_cache_line_size &&
2894 	    (cacheline_size % pci_cache_line_size) == 0)
2895 		return 0;
2896 
2897 	/* Write the correct value. */
2898 	pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
2899 	/* Read it back. */
2900 	pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
2901 	if (cacheline_size == pci_cache_line_size)
2902 		return 0;
2903 
2904 	dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not "
2905 		   "supported\n", pci_cache_line_size << 2);
2906 
2907 	return -EINVAL;
2908 }
2909 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
2910 
2911 #ifdef PCI_DISABLE_MWI
2912 int pci_set_mwi(struct pci_dev *dev)
2913 {
2914 	return 0;
2915 }
2916 
2917 int pci_try_set_mwi(struct pci_dev *dev)
2918 {
2919 	return 0;
2920 }
2921 
2922 void pci_clear_mwi(struct pci_dev *dev)
2923 {
2924 }
2925 
2926 #else
2927 
2928 /**
2929  * pci_set_mwi - enables memory-write-invalidate PCI transaction
2930  * @dev: the PCI device for which MWI is enabled
2931  *
2932  * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
2933  *
2934  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
2935  */
2936 int
2937 pci_set_mwi(struct pci_dev *dev)
2938 {
2939 	int rc;
2940 	u16 cmd;
2941 
2942 	rc = pci_set_cacheline_size(dev);
2943 	if (rc)
2944 		return rc;
2945 
2946 	pci_read_config_word(dev, PCI_COMMAND, &cmd);
2947 	if (! (cmd & PCI_COMMAND_INVALIDATE)) {
2948 		dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
2949 		cmd |= PCI_COMMAND_INVALIDATE;
2950 		pci_write_config_word(dev, PCI_COMMAND, cmd);
2951 	}
2952 
2953 	return 0;
2954 }
2955 
2956 /**
2957  * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
2958  * @dev: the PCI device for which MWI is enabled
2959  *
2960  * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
2961  * Callers are not required to check the return value.
2962  *
2963  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
2964  */
2965 int pci_try_set_mwi(struct pci_dev *dev)
2966 {
2967 	int rc = pci_set_mwi(dev);
2968 	return rc;
2969 }
2970 
2971 /**
2972  * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
2973  * @dev: the PCI device to disable
2974  *
2975  * Disables PCI Memory-Write-Invalidate transaction on the device
2976  */
2977 void
2978 pci_clear_mwi(struct pci_dev *dev)
2979 {
2980 	u16 cmd;
2981 
2982 	pci_read_config_word(dev, PCI_COMMAND, &cmd);
2983 	if (cmd & PCI_COMMAND_INVALIDATE) {
2984 		cmd &= ~PCI_COMMAND_INVALIDATE;
2985 		pci_write_config_word(dev, PCI_COMMAND, cmd);
2986 	}
2987 }
2988 #endif /* ! PCI_DISABLE_MWI */
2989 
2990 /**
2991  * pci_intx - enables/disables PCI INTx for device dev
2992  * @pdev: the PCI device to operate on
2993  * @enable: boolean: whether to enable or disable PCI INTx
2994  *
2995  * Enables/disables PCI INTx for device dev
2996  */
2997 void
2998 pci_intx(struct pci_dev *pdev, int enable)
2999 {
3000 	u16 pci_command, new;
3001 
3002 	pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3003 
3004 	if (enable) {
3005 		new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3006 	} else {
3007 		new = pci_command | PCI_COMMAND_INTX_DISABLE;
3008 	}
3009 
3010 	if (new != pci_command) {
3011 		struct pci_devres *dr;
3012 
3013 		pci_write_config_word(pdev, PCI_COMMAND, new);
3014 
3015 		dr = find_pci_dr(pdev);
3016 		if (dr && !dr->restore_intx) {
3017 			dr->restore_intx = 1;
3018 			dr->orig_intx = !enable;
3019 		}
3020 	}
3021 }
3022 
3023 /**
3024  * pci_intx_mask_supported - probe for INTx masking support
3025  * @dev: the PCI device to operate on
3026  *
3027  * Check if the device dev support INTx masking via the config space
3028  * command word.
3029  */
3030 bool pci_intx_mask_supported(struct pci_dev *dev)
3031 {
3032 	bool mask_supported = false;
3033 	u16 orig, new;
3034 
3035 	if (dev->broken_intx_masking)
3036 		return false;
3037 
3038 	pci_cfg_access_lock(dev);
3039 
3040 	pci_read_config_word(dev, PCI_COMMAND, &orig);
3041 	pci_write_config_word(dev, PCI_COMMAND,
3042 			      orig ^ PCI_COMMAND_INTX_DISABLE);
3043 	pci_read_config_word(dev, PCI_COMMAND, &new);
3044 
3045 	/*
3046 	 * There's no way to protect against hardware bugs or detect them
3047 	 * reliably, but as long as we know what the value should be, let's
3048 	 * go ahead and check it.
3049 	 */
3050 	if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
3051 		dev_err(&dev->dev, "Command register changed from "
3052 			"0x%x to 0x%x: driver or hardware bug?\n", orig, new);
3053 	} else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
3054 		mask_supported = true;
3055 		pci_write_config_word(dev, PCI_COMMAND, orig);
3056 	}
3057 
3058 	pci_cfg_access_unlock(dev);
3059 	return mask_supported;
3060 }
3061 EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
3062 
3063 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3064 {
3065 	struct pci_bus *bus = dev->bus;
3066 	bool mask_updated = true;
3067 	u32 cmd_status_dword;
3068 	u16 origcmd, newcmd;
3069 	unsigned long flags;
3070 	bool irq_pending;
3071 
3072 	/*
3073 	 * We do a single dword read to retrieve both command and status.
3074 	 * Document assumptions that make this possible.
3075 	 */
3076 	BUILD_BUG_ON(PCI_COMMAND % 4);
3077 	BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3078 
3079 	raw_spin_lock_irqsave(&pci_lock, flags);
3080 
3081 	bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3082 
3083 	irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3084 
3085 	/*
3086 	 * Check interrupt status register to see whether our device
3087 	 * triggered the interrupt (when masking) or the next IRQ is
3088 	 * already pending (when unmasking).
3089 	 */
3090 	if (mask != irq_pending) {
3091 		mask_updated = false;
3092 		goto done;
3093 	}
3094 
3095 	origcmd = cmd_status_dword;
3096 	newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3097 	if (mask)
3098 		newcmd |= PCI_COMMAND_INTX_DISABLE;
3099 	if (newcmd != origcmd)
3100 		bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3101 
3102 done:
3103 	raw_spin_unlock_irqrestore(&pci_lock, flags);
3104 
3105 	return mask_updated;
3106 }
3107 
3108 /**
3109  * pci_check_and_mask_intx - mask INTx on pending interrupt
3110  * @dev: the PCI device to operate on
3111  *
3112  * Check if the device dev has its INTx line asserted, mask it and
3113  * return true in that case. False is returned if not interrupt was
3114  * pending.
3115  */
3116 bool pci_check_and_mask_intx(struct pci_dev *dev)
3117 {
3118 	return pci_check_and_set_intx_mask(dev, true);
3119 }
3120 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3121 
3122 /**
3123  * pci_check_and_mask_intx - unmask INTx of no interrupt is pending
3124  * @dev: the PCI device to operate on
3125  *
3126  * Check if the device dev has its INTx line asserted, unmask it if not
3127  * and return true. False is returned and the mask remains active if
3128  * there was still an interrupt pending.
3129  */
3130 bool pci_check_and_unmask_intx(struct pci_dev *dev)
3131 {
3132 	return pci_check_and_set_intx_mask(dev, false);
3133 }
3134 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3135 
3136 /**
3137  * pci_msi_off - disables any msi or msix capabilities
3138  * @dev: the PCI device to operate on
3139  *
3140  * If you want to use msi see pci_enable_msi and friends.
3141  * This is a lower level primitive that allows us to disable
3142  * msi operation at the device level.
3143  */
3144 void pci_msi_off(struct pci_dev *dev)
3145 {
3146 	int pos;
3147 	u16 control;
3148 
3149 	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
3150 	if (pos) {
3151 		pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
3152 		control &= ~PCI_MSI_FLAGS_ENABLE;
3153 		pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
3154 	}
3155 	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
3156 	if (pos) {
3157 		pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
3158 		control &= ~PCI_MSIX_FLAGS_ENABLE;
3159 		pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
3160 	}
3161 }
3162 EXPORT_SYMBOL_GPL(pci_msi_off);
3163 
3164 int pci_set_dma_max_seg_size(struct pci_dev *dev, unsigned int size)
3165 {
3166 	return dma_set_max_seg_size(&dev->dev, size);
3167 }
3168 EXPORT_SYMBOL(pci_set_dma_max_seg_size);
3169 
3170 int pci_set_dma_seg_boundary(struct pci_dev *dev, unsigned long mask)
3171 {
3172 	return dma_set_seg_boundary(&dev->dev, mask);
3173 }
3174 EXPORT_SYMBOL(pci_set_dma_seg_boundary);
3175 
3176 static int pcie_flr(struct pci_dev *dev, int probe)
3177 {
3178 	int i;
3179 	int pos;
3180 	u32 cap;
3181 	u16 status, control;
3182 
3183 	pos = pci_pcie_cap(dev);
3184 	if (!pos)
3185 		return -ENOTTY;
3186 
3187 	pci_read_config_dword(dev, pos + PCI_EXP_DEVCAP, &cap);
3188 	if (!(cap & PCI_EXP_DEVCAP_FLR))
3189 		return -ENOTTY;
3190 
3191 	if (probe)
3192 		return 0;
3193 
3194 	/* Wait for Transaction Pending bit clean */
3195 	for (i = 0; i < 4; i++) {
3196 		if (i)
3197 			msleep((1 << (i - 1)) * 100);
3198 
3199 		pci_read_config_word(dev, pos + PCI_EXP_DEVSTA, &status);
3200 		if (!(status & PCI_EXP_DEVSTA_TRPND))
3201 			goto clear;
3202 	}
3203 
3204 	dev_err(&dev->dev, "transaction is not cleared; "
3205 			"proceeding with reset anyway\n");
3206 
3207 clear:
3208 	pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &control);
3209 	control |= PCI_EXP_DEVCTL_BCR_FLR;
3210 	pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, control);
3211 
3212 	msleep(100);
3213 
3214 	return 0;
3215 }
3216 
3217 static int pci_af_flr(struct pci_dev *dev, int probe)
3218 {
3219 	int i;
3220 	int pos;
3221 	u8 cap;
3222 	u8 status;
3223 
3224 	pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3225 	if (!pos)
3226 		return -ENOTTY;
3227 
3228 	pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3229 	if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3230 		return -ENOTTY;
3231 
3232 	if (probe)
3233 		return 0;
3234 
3235 	/* Wait for Transaction Pending bit clean */
3236 	for (i = 0; i < 4; i++) {
3237 		if (i)
3238 			msleep((1 << (i - 1)) * 100);
3239 
3240 		pci_read_config_byte(dev, pos + PCI_AF_STATUS, &status);
3241 		if (!(status & PCI_AF_STATUS_TP))
3242 			goto clear;
3243 	}
3244 
3245 	dev_err(&dev->dev, "transaction is not cleared; "
3246 			"proceeding with reset anyway\n");
3247 
3248 clear:
3249 	pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3250 	msleep(100);
3251 
3252 	return 0;
3253 }
3254 
3255 /**
3256  * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3257  * @dev: Device to reset.
3258  * @probe: If set, only check if the device can be reset this way.
3259  *
3260  * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3261  * unset, it will be reinitialized internally when going from PCI_D3hot to
3262  * PCI_D0.  If that's the case and the device is not in a low-power state
3263  * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3264  *
3265  * NOTE: This causes the caller to sleep for twice the device power transition
3266  * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3267  * by devault (i.e. unless the @dev's d3_delay field has a different value).
3268  * Moreover, only devices in D0 can be reset by this function.
3269  */
3270 static int pci_pm_reset(struct pci_dev *dev, int probe)
3271 {
3272 	u16 csr;
3273 
3274 	if (!dev->pm_cap)
3275 		return -ENOTTY;
3276 
3277 	pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3278 	if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3279 		return -ENOTTY;
3280 
3281 	if (probe)
3282 		return 0;
3283 
3284 	if (dev->current_state != PCI_D0)
3285 		return -EINVAL;
3286 
3287 	csr &= ~PCI_PM_CTRL_STATE_MASK;
3288 	csr |= PCI_D3hot;
3289 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3290 	pci_dev_d3_sleep(dev);
3291 
3292 	csr &= ~PCI_PM_CTRL_STATE_MASK;
3293 	csr |= PCI_D0;
3294 	pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3295 	pci_dev_d3_sleep(dev);
3296 
3297 	return 0;
3298 }
3299 
3300 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
3301 {
3302 	u16 ctrl;
3303 	struct pci_dev *pdev;
3304 
3305 	if (pci_is_root_bus(dev->bus) || dev->subordinate || !dev->bus->self)
3306 		return -ENOTTY;
3307 
3308 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3309 		if (pdev != dev)
3310 			return -ENOTTY;
3311 
3312 	if (probe)
3313 		return 0;
3314 
3315 	pci_read_config_word(dev->bus->self, PCI_BRIDGE_CONTROL, &ctrl);
3316 	ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
3317 	pci_write_config_word(dev->bus->self, PCI_BRIDGE_CONTROL, ctrl);
3318 	msleep(100);
3319 
3320 	ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
3321 	pci_write_config_word(dev->bus->self, PCI_BRIDGE_CONTROL, ctrl);
3322 	msleep(100);
3323 
3324 	return 0;
3325 }
3326 
3327 static int __pci_dev_reset(struct pci_dev *dev, int probe)
3328 {
3329 	int rc;
3330 
3331 	might_sleep();
3332 
3333 	rc = pci_dev_specific_reset(dev, probe);
3334 	if (rc != -ENOTTY)
3335 		goto done;
3336 
3337 	rc = pcie_flr(dev, probe);
3338 	if (rc != -ENOTTY)
3339 		goto done;
3340 
3341 	rc = pci_af_flr(dev, probe);
3342 	if (rc != -ENOTTY)
3343 		goto done;
3344 
3345 	rc = pci_pm_reset(dev, probe);
3346 	if (rc != -ENOTTY)
3347 		goto done;
3348 
3349 	rc = pci_parent_bus_reset(dev, probe);
3350 done:
3351 	return rc;
3352 }
3353 
3354 static int pci_dev_reset(struct pci_dev *dev, int probe)
3355 {
3356 	int rc;
3357 
3358 	if (!probe) {
3359 		pci_cfg_access_lock(dev);
3360 		/* block PM suspend, driver probe, etc. */
3361 		device_lock(&dev->dev);
3362 	}
3363 
3364 	rc = __pci_dev_reset(dev, probe);
3365 
3366 	if (!probe) {
3367 		device_unlock(&dev->dev);
3368 		pci_cfg_access_unlock(dev);
3369 	}
3370 	return rc;
3371 }
3372 /**
3373  * __pci_reset_function - reset a PCI device function
3374  * @dev: PCI device to reset
3375  *
3376  * Some devices allow an individual function to be reset without affecting
3377  * other functions in the same device.  The PCI device must be responsive
3378  * to PCI config space in order to use this function.
3379  *
3380  * The device function is presumed to be unused when this function is called.
3381  * Resetting the device will make the contents of PCI configuration space
3382  * random, so any caller of this must be prepared to reinitialise the
3383  * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3384  * etc.
3385  *
3386  * Returns 0 if the device function was successfully reset or negative if the
3387  * device doesn't support resetting a single function.
3388  */
3389 int __pci_reset_function(struct pci_dev *dev)
3390 {
3391 	return pci_dev_reset(dev, 0);
3392 }
3393 EXPORT_SYMBOL_GPL(__pci_reset_function);
3394 
3395 /**
3396  * __pci_reset_function_locked - reset a PCI device function while holding
3397  * the @dev mutex lock.
3398  * @dev: PCI device to reset
3399  *
3400  * Some devices allow an individual function to be reset without affecting
3401  * other functions in the same device.  The PCI device must be responsive
3402  * to PCI config space in order to use this function.
3403  *
3404  * The device function is presumed to be unused and the caller is holding
3405  * the device mutex lock when this function is called.
3406  * Resetting the device will make the contents of PCI configuration space
3407  * random, so any caller of this must be prepared to reinitialise the
3408  * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3409  * etc.
3410  *
3411  * Returns 0 if the device function was successfully reset or negative if the
3412  * device doesn't support resetting a single function.
3413  */
3414 int __pci_reset_function_locked(struct pci_dev *dev)
3415 {
3416 	return __pci_dev_reset(dev, 0);
3417 }
3418 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
3419 
3420 /**
3421  * pci_probe_reset_function - check whether the device can be safely reset
3422  * @dev: PCI device to reset
3423  *
3424  * Some devices allow an individual function to be reset without affecting
3425  * other functions in the same device.  The PCI device must be responsive
3426  * to PCI config space in order to use this function.
3427  *
3428  * Returns 0 if the device function can be reset or negative if the
3429  * device doesn't support resetting a single function.
3430  */
3431 int pci_probe_reset_function(struct pci_dev *dev)
3432 {
3433 	return pci_dev_reset(dev, 1);
3434 }
3435 
3436 /**
3437  * pci_reset_function - quiesce and reset a PCI device function
3438  * @dev: PCI device to reset
3439  *
3440  * Some devices allow an individual function to be reset without affecting
3441  * other functions in the same device.  The PCI device must be responsive
3442  * to PCI config space in order to use this function.
3443  *
3444  * This function does not just reset the PCI portion of a device, but
3445  * clears all the state associated with the device.  This function differs
3446  * from __pci_reset_function in that it saves and restores device state
3447  * over the reset.
3448  *
3449  * Returns 0 if the device function was successfully reset or negative if the
3450  * device doesn't support resetting a single function.
3451  */
3452 int pci_reset_function(struct pci_dev *dev)
3453 {
3454 	int rc;
3455 
3456 	rc = pci_dev_reset(dev, 1);
3457 	if (rc)
3458 		return rc;
3459 
3460 	pci_save_state(dev);
3461 
3462 	/*
3463 	 * both INTx and MSI are disabled after the Interrupt Disable bit
3464 	 * is set and the Bus Master bit is cleared.
3465 	 */
3466 	pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
3467 
3468 	rc = pci_dev_reset(dev, 0);
3469 
3470 	pci_restore_state(dev);
3471 
3472 	return rc;
3473 }
3474 EXPORT_SYMBOL_GPL(pci_reset_function);
3475 
3476 /**
3477  * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
3478  * @dev: PCI device to query
3479  *
3480  * Returns mmrbc: maximum designed memory read count in bytes
3481  *    or appropriate error value.
3482  */
3483 int pcix_get_max_mmrbc(struct pci_dev *dev)
3484 {
3485 	int cap;
3486 	u32 stat;
3487 
3488 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
3489 	if (!cap)
3490 		return -EINVAL;
3491 
3492 	if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
3493 		return -EINVAL;
3494 
3495 	return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
3496 }
3497 EXPORT_SYMBOL(pcix_get_max_mmrbc);
3498 
3499 /**
3500  * pcix_get_mmrbc - get PCI-X maximum memory read byte count
3501  * @dev: PCI device to query
3502  *
3503  * Returns mmrbc: maximum memory read count in bytes
3504  *    or appropriate error value.
3505  */
3506 int pcix_get_mmrbc(struct pci_dev *dev)
3507 {
3508 	int cap;
3509 	u16 cmd;
3510 
3511 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
3512 	if (!cap)
3513 		return -EINVAL;
3514 
3515 	if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
3516 		return -EINVAL;
3517 
3518 	return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
3519 }
3520 EXPORT_SYMBOL(pcix_get_mmrbc);
3521 
3522 /**
3523  * pcix_set_mmrbc - set PCI-X maximum memory read byte count
3524  * @dev: PCI device to query
3525  * @mmrbc: maximum memory read count in bytes
3526  *    valid values are 512, 1024, 2048, 4096
3527  *
3528  * If possible sets maximum memory read byte count, some bridges have erratas
3529  * that prevent this.
3530  */
3531 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
3532 {
3533 	int cap;
3534 	u32 stat, v, o;
3535 	u16 cmd;
3536 
3537 	if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
3538 		return -EINVAL;
3539 
3540 	v = ffs(mmrbc) - 10;
3541 
3542 	cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
3543 	if (!cap)
3544 		return -EINVAL;
3545 
3546 	if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
3547 		return -EINVAL;
3548 
3549 	if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
3550 		return -E2BIG;
3551 
3552 	if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
3553 		return -EINVAL;
3554 
3555 	o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
3556 	if (o != v) {
3557 		if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
3558 			return -EIO;
3559 
3560 		cmd &= ~PCI_X_CMD_MAX_READ;
3561 		cmd |= v << 2;
3562 		if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
3563 			return -EIO;
3564 	}
3565 	return 0;
3566 }
3567 EXPORT_SYMBOL(pcix_set_mmrbc);
3568 
3569 /**
3570  * pcie_get_readrq - get PCI Express read request size
3571  * @dev: PCI device to query
3572  *
3573  * Returns maximum memory read request in bytes
3574  *    or appropriate error value.
3575  */
3576 int pcie_get_readrq(struct pci_dev *dev)
3577 {
3578 	int ret, cap;
3579 	u16 ctl;
3580 
3581 	cap = pci_pcie_cap(dev);
3582 	if (!cap)
3583 		return -EINVAL;
3584 
3585 	ret = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3586 	if (!ret)
3587 		ret = 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
3588 
3589 	return ret;
3590 }
3591 EXPORT_SYMBOL(pcie_get_readrq);
3592 
3593 /**
3594  * pcie_set_readrq - set PCI Express maximum memory read request
3595  * @dev: PCI device to query
3596  * @rq: maximum memory read count in bytes
3597  *    valid values are 128, 256, 512, 1024, 2048, 4096
3598  *
3599  * If possible sets maximum memory read request in bytes
3600  */
3601 int pcie_set_readrq(struct pci_dev *dev, int rq)
3602 {
3603 	int cap, err = -EINVAL;
3604 	u16 ctl, v;
3605 
3606 	if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
3607 		goto out;
3608 
3609 	cap = pci_pcie_cap(dev);
3610 	if (!cap)
3611 		goto out;
3612 
3613 	err = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3614 	if (err)
3615 		goto out;
3616 	/*
3617 	 * If using the "performance" PCIe config, we clamp the
3618 	 * read rq size to the max packet size to prevent the
3619 	 * host bridge generating requests larger than we can
3620 	 * cope with
3621 	 */
3622 	if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
3623 		int mps = pcie_get_mps(dev);
3624 
3625 		if (mps < 0)
3626 			return mps;
3627 		if (mps < rq)
3628 			rq = mps;
3629 	}
3630 
3631 	v = (ffs(rq) - 8) << 12;
3632 
3633 	if ((ctl & PCI_EXP_DEVCTL_READRQ) != v) {
3634 		ctl &= ~PCI_EXP_DEVCTL_READRQ;
3635 		ctl |= v;
3636 		err = pci_write_config_word(dev, cap + PCI_EXP_DEVCTL, ctl);
3637 	}
3638 
3639 out:
3640 	return err;
3641 }
3642 EXPORT_SYMBOL(pcie_set_readrq);
3643 
3644 /**
3645  * pcie_get_mps - get PCI Express maximum payload size
3646  * @dev: PCI device to query
3647  *
3648  * Returns maximum payload size in bytes
3649  *    or appropriate error value.
3650  */
3651 int pcie_get_mps(struct pci_dev *dev)
3652 {
3653 	int ret, cap;
3654 	u16 ctl;
3655 
3656 	cap = pci_pcie_cap(dev);
3657 	if (!cap)
3658 		return -EINVAL;
3659 
3660 	ret = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3661 	if (!ret)
3662 		ret = 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
3663 
3664 	return ret;
3665 }
3666 
3667 /**
3668  * pcie_set_mps - set PCI Express maximum payload size
3669  * @dev: PCI device to query
3670  * @mps: maximum payload size in bytes
3671  *    valid values are 128, 256, 512, 1024, 2048, 4096
3672  *
3673  * If possible sets maximum payload size
3674  */
3675 int pcie_set_mps(struct pci_dev *dev, int mps)
3676 {
3677 	int cap, err = -EINVAL;
3678 	u16 ctl, v;
3679 
3680 	if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
3681 		goto out;
3682 
3683 	v = ffs(mps) - 8;
3684 	if (v > dev->pcie_mpss)
3685 		goto out;
3686 	v <<= 5;
3687 
3688 	cap = pci_pcie_cap(dev);
3689 	if (!cap)
3690 		goto out;
3691 
3692 	err = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3693 	if (err)
3694 		goto out;
3695 
3696 	if ((ctl & PCI_EXP_DEVCTL_PAYLOAD) != v) {
3697 		ctl &= ~PCI_EXP_DEVCTL_PAYLOAD;
3698 		ctl |= v;
3699 		err = pci_write_config_word(dev, cap + PCI_EXP_DEVCTL, ctl);
3700 	}
3701 out:
3702 	return err;
3703 }
3704 
3705 /**
3706  * pci_select_bars - Make BAR mask from the type of resource
3707  * @dev: the PCI device for which BAR mask is made
3708  * @flags: resource type mask to be selected
3709  *
3710  * This helper routine makes bar mask from the type of resource.
3711  */
3712 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
3713 {
3714 	int i, bars = 0;
3715 	for (i = 0; i < PCI_NUM_RESOURCES; i++)
3716 		if (pci_resource_flags(dev, i) & flags)
3717 			bars |= (1 << i);
3718 	return bars;
3719 }
3720 
3721 /**
3722  * pci_resource_bar - get position of the BAR associated with a resource
3723  * @dev: the PCI device
3724  * @resno: the resource number
3725  * @type: the BAR type to be filled in
3726  *
3727  * Returns BAR position in config space, or 0 if the BAR is invalid.
3728  */
3729 int pci_resource_bar(struct pci_dev *dev, int resno, enum pci_bar_type *type)
3730 {
3731 	int reg;
3732 
3733 	if (resno < PCI_ROM_RESOURCE) {
3734 		*type = pci_bar_unknown;
3735 		return PCI_BASE_ADDRESS_0 + 4 * resno;
3736 	} else if (resno == PCI_ROM_RESOURCE) {
3737 		*type = pci_bar_mem32;
3738 		return dev->rom_base_reg;
3739 	} else if (resno < PCI_BRIDGE_RESOURCES) {
3740 		/* device specific resource */
3741 		reg = pci_iov_resource_bar(dev, resno, type);
3742 		if (reg)
3743 			return reg;
3744 	}
3745 
3746 	dev_err(&dev->dev, "BAR %d: invalid resource\n", resno);
3747 	return 0;
3748 }
3749 
3750 /* Some architectures require additional programming to enable VGA */
3751 static arch_set_vga_state_t arch_set_vga_state;
3752 
3753 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
3754 {
3755 	arch_set_vga_state = func;	/* NULL disables */
3756 }
3757 
3758 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
3759 		      unsigned int command_bits, u32 flags)
3760 {
3761 	if (arch_set_vga_state)
3762 		return arch_set_vga_state(dev, decode, command_bits,
3763 						flags);
3764 	return 0;
3765 }
3766 
3767 /**
3768  * pci_set_vga_state - set VGA decode state on device and parents if requested
3769  * @dev: the PCI device
3770  * @decode: true = enable decoding, false = disable decoding
3771  * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
3772  * @flags: traverse ancestors and change bridges
3773  * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
3774  */
3775 int pci_set_vga_state(struct pci_dev *dev, bool decode,
3776 		      unsigned int command_bits, u32 flags)
3777 {
3778 	struct pci_bus *bus;
3779 	struct pci_dev *bridge;
3780 	u16 cmd;
3781 	int rc;
3782 
3783 	WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) & (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
3784 
3785 	/* ARCH specific VGA enables */
3786 	rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
3787 	if (rc)
3788 		return rc;
3789 
3790 	if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
3791 		pci_read_config_word(dev, PCI_COMMAND, &cmd);
3792 		if (decode == true)
3793 			cmd |= command_bits;
3794 		else
3795 			cmd &= ~command_bits;
3796 		pci_write_config_word(dev, PCI_COMMAND, cmd);
3797 	}
3798 
3799 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
3800 		return 0;
3801 
3802 	bus = dev->bus;
3803 	while (bus) {
3804 		bridge = bus->self;
3805 		if (bridge) {
3806 			pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
3807 					     &cmd);
3808 			if (decode == true)
3809 				cmd |= PCI_BRIDGE_CTL_VGA;
3810 			else
3811 				cmd &= ~PCI_BRIDGE_CTL_VGA;
3812 			pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
3813 					      cmd);
3814 		}
3815 		bus = bus->parent;
3816 	}
3817 	return 0;
3818 }
3819 
3820 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
3821 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
3822 static DEFINE_SPINLOCK(resource_alignment_lock);
3823 
3824 /**
3825  * pci_specified_resource_alignment - get resource alignment specified by user.
3826  * @dev: the PCI device to get
3827  *
3828  * RETURNS: Resource alignment if it is specified.
3829  *          Zero if it is not specified.
3830  */
3831 resource_size_t pci_specified_resource_alignment(struct pci_dev *dev)
3832 {
3833 	int seg, bus, slot, func, align_order, count;
3834 	resource_size_t align = 0;
3835 	char *p;
3836 
3837 	spin_lock(&resource_alignment_lock);
3838 	p = resource_alignment_param;
3839 	while (*p) {
3840 		count = 0;
3841 		if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
3842 							p[count] == '@') {
3843 			p += count + 1;
3844 		} else {
3845 			align_order = -1;
3846 		}
3847 		if (sscanf(p, "%x:%x:%x.%x%n",
3848 			&seg, &bus, &slot, &func, &count) != 4) {
3849 			seg = 0;
3850 			if (sscanf(p, "%x:%x.%x%n",
3851 					&bus, &slot, &func, &count) != 3) {
3852 				/* Invalid format */
3853 				printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
3854 					p);
3855 				break;
3856 			}
3857 		}
3858 		p += count;
3859 		if (seg == pci_domain_nr(dev->bus) &&
3860 			bus == dev->bus->number &&
3861 			slot == PCI_SLOT(dev->devfn) &&
3862 			func == PCI_FUNC(dev->devfn)) {
3863 			if (align_order == -1) {
3864 				align = PAGE_SIZE;
3865 			} else {
3866 				align = 1 << align_order;
3867 			}
3868 			/* Found */
3869 			break;
3870 		}
3871 		if (*p != ';' && *p != ',') {
3872 			/* End of param or invalid format */
3873 			break;
3874 		}
3875 		p++;
3876 	}
3877 	spin_unlock(&resource_alignment_lock);
3878 	return align;
3879 }
3880 
3881 /**
3882  * pci_is_reassigndev - check if specified PCI is target device to reassign
3883  * @dev: the PCI device to check
3884  *
3885  * RETURNS: non-zero for PCI device is a target device to reassign,
3886  *          or zero is not.
3887  */
3888 int pci_is_reassigndev(struct pci_dev *dev)
3889 {
3890 	return (pci_specified_resource_alignment(dev) != 0);
3891 }
3892 
3893 /*
3894  * This function disables memory decoding and releases memory resources
3895  * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
3896  * It also rounds up size to specified alignment.
3897  * Later on, the kernel will assign page-aligned memory resource back
3898  * to the device.
3899  */
3900 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
3901 {
3902 	int i;
3903 	struct resource *r;
3904 	resource_size_t align, size;
3905 	u16 command;
3906 
3907 	if (!pci_is_reassigndev(dev))
3908 		return;
3909 
3910 	if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
3911 	    (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
3912 		dev_warn(&dev->dev,
3913 			"Can't reassign resources to host bridge.\n");
3914 		return;
3915 	}
3916 
3917 	dev_info(&dev->dev,
3918 		"Disabling memory decoding and releasing memory resources.\n");
3919 	pci_read_config_word(dev, PCI_COMMAND, &command);
3920 	command &= ~PCI_COMMAND_MEMORY;
3921 	pci_write_config_word(dev, PCI_COMMAND, command);
3922 
3923 	align = pci_specified_resource_alignment(dev);
3924 	for (i = 0; i < PCI_BRIDGE_RESOURCES; i++) {
3925 		r = &dev->resource[i];
3926 		if (!(r->flags & IORESOURCE_MEM))
3927 			continue;
3928 		size = resource_size(r);
3929 		if (size < align) {
3930 			size = align;
3931 			dev_info(&dev->dev,
3932 				"Rounding up size of resource #%d to %#llx.\n",
3933 				i, (unsigned long long)size);
3934 		}
3935 		r->end = size - 1;
3936 		r->start = 0;
3937 	}
3938 	/* Need to disable bridge's resource window,
3939 	 * to enable the kernel to reassign new resource
3940 	 * window later on.
3941 	 */
3942 	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
3943 	    (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
3944 		for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
3945 			r = &dev->resource[i];
3946 			if (!(r->flags & IORESOURCE_MEM))
3947 				continue;
3948 			r->end = resource_size(r) - 1;
3949 			r->start = 0;
3950 		}
3951 		pci_disable_bridge_window(dev);
3952 	}
3953 }
3954 
3955 ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
3956 {
3957 	if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
3958 		count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
3959 	spin_lock(&resource_alignment_lock);
3960 	strncpy(resource_alignment_param, buf, count);
3961 	resource_alignment_param[count] = '\0';
3962 	spin_unlock(&resource_alignment_lock);
3963 	return count;
3964 }
3965 
3966 ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
3967 {
3968 	size_t count;
3969 	spin_lock(&resource_alignment_lock);
3970 	count = snprintf(buf, size, "%s", resource_alignment_param);
3971 	spin_unlock(&resource_alignment_lock);
3972 	return count;
3973 }
3974 
3975 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
3976 {
3977 	return pci_get_resource_alignment_param(buf, PAGE_SIZE);
3978 }
3979 
3980 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
3981 					const char *buf, size_t count)
3982 {
3983 	return pci_set_resource_alignment_param(buf, count);
3984 }
3985 
3986 BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
3987 					pci_resource_alignment_store);
3988 
3989 static int __init pci_resource_alignment_sysfs_init(void)
3990 {
3991 	return bus_create_file(&pci_bus_type,
3992 					&bus_attr_resource_alignment);
3993 }
3994 
3995 late_initcall(pci_resource_alignment_sysfs_init);
3996 
3997 static void __devinit pci_no_domains(void)
3998 {
3999 #ifdef CONFIG_PCI_DOMAINS
4000 	pci_domains_supported = 0;
4001 #endif
4002 }
4003 
4004 /**
4005  * pci_ext_cfg_enabled - can we access extended PCI config space?
4006  * @dev: The PCI device of the root bridge.
4007  *
4008  * Returns 1 if we can access PCI extended config space (offsets
4009  * greater than 0xff). This is the default implementation. Architecture
4010  * implementations can override this.
4011  */
4012 int __weak pci_ext_cfg_avail(struct pci_dev *dev)
4013 {
4014 	return 1;
4015 }
4016 
4017 void __weak pci_fixup_cardbus(struct pci_bus *bus)
4018 {
4019 }
4020 EXPORT_SYMBOL(pci_fixup_cardbus);
4021 
4022 static int __init pci_setup(char *str)
4023 {
4024 	while (str) {
4025 		char *k = strchr(str, ',');
4026 		if (k)
4027 			*k++ = 0;
4028 		if (*str && (str = pcibios_setup(str)) && *str) {
4029 			if (!strcmp(str, "nomsi")) {
4030 				pci_no_msi();
4031 			} else if (!strcmp(str, "noaer")) {
4032 				pci_no_aer();
4033 			} else if (!strncmp(str, "realloc=", 8)) {
4034 				pci_realloc_get_opt(str + 8);
4035 			} else if (!strncmp(str, "realloc", 7)) {
4036 				pci_realloc_get_opt("on");
4037 			} else if (!strcmp(str, "nodomains")) {
4038 				pci_no_domains();
4039 			} else if (!strncmp(str, "noari", 5)) {
4040 				pcie_ari_disabled = true;
4041 			} else if (!strncmp(str, "cbiosize=", 9)) {
4042 				pci_cardbus_io_size = memparse(str + 9, &str);
4043 			} else if (!strncmp(str, "cbmemsize=", 10)) {
4044 				pci_cardbus_mem_size = memparse(str + 10, &str);
4045 			} else if (!strncmp(str, "resource_alignment=", 19)) {
4046 				pci_set_resource_alignment_param(str + 19,
4047 							strlen(str + 19));
4048 			} else if (!strncmp(str, "ecrc=", 5)) {
4049 				pcie_ecrc_get_policy(str + 5);
4050 			} else if (!strncmp(str, "hpiosize=", 9)) {
4051 				pci_hotplug_io_size = memparse(str + 9, &str);
4052 			} else if (!strncmp(str, "hpmemsize=", 10)) {
4053 				pci_hotplug_mem_size = memparse(str + 10, &str);
4054 			} else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
4055 				pcie_bus_config = PCIE_BUS_TUNE_OFF;
4056 			} else if (!strncmp(str, "pcie_bus_safe", 13)) {
4057 				pcie_bus_config = PCIE_BUS_SAFE;
4058 			} else if (!strncmp(str, "pcie_bus_perf", 13)) {
4059 				pcie_bus_config = PCIE_BUS_PERFORMANCE;
4060 			} else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
4061 				pcie_bus_config = PCIE_BUS_PEER2PEER;
4062 			} else if (!strncmp(str, "pcie_scan_all", 13)) {
4063 				pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
4064 			} else {
4065 				printk(KERN_ERR "PCI: Unknown option `%s'\n",
4066 						str);
4067 			}
4068 		}
4069 		str = k;
4070 	}
4071 	return 0;
4072 }
4073 early_param("pci", pci_setup);
4074 
4075 EXPORT_SYMBOL(pci_reenable_device);
4076 EXPORT_SYMBOL(pci_enable_device_io);
4077 EXPORT_SYMBOL(pci_enable_device_mem);
4078 EXPORT_SYMBOL(pci_enable_device);
4079 EXPORT_SYMBOL(pcim_enable_device);
4080 EXPORT_SYMBOL(pcim_pin_device);
4081 EXPORT_SYMBOL(pci_disable_device);
4082 EXPORT_SYMBOL(pci_find_capability);
4083 EXPORT_SYMBOL(pci_bus_find_capability);
4084 EXPORT_SYMBOL(pci_release_regions);
4085 EXPORT_SYMBOL(pci_request_regions);
4086 EXPORT_SYMBOL(pci_request_regions_exclusive);
4087 EXPORT_SYMBOL(pci_release_region);
4088 EXPORT_SYMBOL(pci_request_region);
4089 EXPORT_SYMBOL(pci_request_region_exclusive);
4090 EXPORT_SYMBOL(pci_release_selected_regions);
4091 EXPORT_SYMBOL(pci_request_selected_regions);
4092 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
4093 EXPORT_SYMBOL(pci_set_master);
4094 EXPORT_SYMBOL(pci_clear_master);
4095 EXPORT_SYMBOL(pci_set_mwi);
4096 EXPORT_SYMBOL(pci_try_set_mwi);
4097 EXPORT_SYMBOL(pci_clear_mwi);
4098 EXPORT_SYMBOL_GPL(pci_intx);
4099 EXPORT_SYMBOL(pci_assign_resource);
4100 EXPORT_SYMBOL(pci_find_parent_resource);
4101 EXPORT_SYMBOL(pci_select_bars);
4102 
4103 EXPORT_SYMBOL(pci_set_power_state);
4104 EXPORT_SYMBOL(pci_save_state);
4105 EXPORT_SYMBOL(pci_restore_state);
4106 EXPORT_SYMBOL(pci_pme_capable);
4107 EXPORT_SYMBOL(pci_pme_active);
4108 EXPORT_SYMBOL(pci_wake_from_d3);
4109 EXPORT_SYMBOL(pci_target_state);
4110 EXPORT_SYMBOL(pci_prepare_to_sleep);
4111 EXPORT_SYMBOL(pci_back_from_sleep);
4112 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
4113