Loading...
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
29const char *pci_power_names[] = {
30 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
31};
32EXPORT_SYMBOL_GPL(pci_power_names);
33
34int isa_dma_bridge_buggy;
35EXPORT_SYMBOL(isa_dma_bridge_buggy);
36
37int pci_pci_problems;
38EXPORT_SYMBOL(pci_pci_problems);
39
40unsigned int pci_pm_d3_delay;
41
42static void pci_pme_list_scan(struct work_struct *work);
43
44static LIST_HEAD(pci_pme_list);
45static DEFINE_MUTEX(pci_pme_list_mutex);
46static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
47
48struct 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
55static 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
66int 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 */
72unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
73unsigned 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 */
78unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
79unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
80
81enum 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 */
89u8 pci_dfl_cache_line_size __devinitdata = L1_CACHE_BYTES >> 2;
90u8 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 */
96unsigned int pcibios_max_latency = 255;
97
98/* If set, the PCIe ARI capability will not be used. */
99static 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 */
108unsigned char pci_bus_max_busnr(struct pci_bus* bus)
109{
110 struct list_head *tmp;
111 unsigned char max, n;
112
113 max = bus->subordinate;
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}
121EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
122
123#ifdef CONFIG_HAS_IOMEM
124void __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}
136EXPORT_SYMBOL_GPL(pci_ioremap_bar);
137#endif
138
139#if 0
140/**
141 * pci_max_busnr - returns maximum PCI bus number
142 *
143 * Returns the highest PCI bus number present in the system global list of
144 * PCI buses.
145 */
146unsigned char __devinit
147pci_max_busnr(void)
148{
149 struct pci_bus *bus = NULL;
150 unsigned char max, n;
151
152 max = 0;
153 while ((bus = pci_find_next_bus(bus)) != NULL) {
154 n = pci_bus_max_busnr(bus);
155 if(n > max)
156 max = n;
157 }
158 return max;
159}
160
161#endif /* 0 */
162
163#define PCI_FIND_CAP_TTL 48
164
165static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
166 u8 pos, int cap, int *ttl)
167{
168 u8 id;
169
170 while ((*ttl)--) {
171 pci_bus_read_config_byte(bus, devfn, pos, &pos);
172 if (pos < 0x40)
173 break;
174 pos &= ~3;
175 pci_bus_read_config_byte(bus, devfn, pos + PCI_CAP_LIST_ID,
176 &id);
177 if (id == 0xff)
178 break;
179 if (id == cap)
180 return pos;
181 pos += PCI_CAP_LIST_NEXT;
182 }
183 return 0;
184}
185
186static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
187 u8 pos, int cap)
188{
189 int ttl = PCI_FIND_CAP_TTL;
190
191 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
192}
193
194int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
195{
196 return __pci_find_next_cap(dev->bus, dev->devfn,
197 pos + PCI_CAP_LIST_NEXT, cap);
198}
199EXPORT_SYMBOL_GPL(pci_find_next_capability);
200
201static int __pci_bus_find_cap_start(struct pci_bus *bus,
202 unsigned int devfn, u8 hdr_type)
203{
204 u16 status;
205
206 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
207 if (!(status & PCI_STATUS_CAP_LIST))
208 return 0;
209
210 switch (hdr_type) {
211 case PCI_HEADER_TYPE_NORMAL:
212 case PCI_HEADER_TYPE_BRIDGE:
213 return PCI_CAPABILITY_LIST;
214 case PCI_HEADER_TYPE_CARDBUS:
215 return PCI_CB_CAPABILITY_LIST;
216 default:
217 return 0;
218 }
219
220 return 0;
221}
222
223/**
224 * pci_find_capability - query for devices' capabilities
225 * @dev: PCI device to query
226 * @cap: capability code
227 *
228 * Tell if a device supports a given PCI capability.
229 * Returns the address of the requested capability structure within the
230 * device's PCI configuration space or 0 in case the device does not
231 * support it. Possible values for @cap:
232 *
233 * %PCI_CAP_ID_PM Power Management
234 * %PCI_CAP_ID_AGP Accelerated Graphics Port
235 * %PCI_CAP_ID_VPD Vital Product Data
236 * %PCI_CAP_ID_SLOTID Slot Identification
237 * %PCI_CAP_ID_MSI Message Signalled Interrupts
238 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
239 * %PCI_CAP_ID_PCIX PCI-X
240 * %PCI_CAP_ID_EXP PCI Express
241 */
242int pci_find_capability(struct pci_dev *dev, int cap)
243{
244 int pos;
245
246 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
247 if (pos)
248 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
249
250 return pos;
251}
252
253/**
254 * pci_bus_find_capability - query for devices' capabilities
255 * @bus: the PCI bus to query
256 * @devfn: PCI device to query
257 * @cap: capability code
258 *
259 * Like pci_find_capability() but works for pci devices that do not have a
260 * pci_dev structure set up yet.
261 *
262 * Returns the address of the requested capability structure within the
263 * device's PCI configuration space or 0 in case the device does not
264 * support it.
265 */
266int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
267{
268 int pos;
269 u8 hdr_type;
270
271 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
272
273 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
274 if (pos)
275 pos = __pci_find_next_cap(bus, devfn, pos, cap);
276
277 return pos;
278}
279
280/**
281 * pci_find_ext_capability - Find an extended capability
282 * @dev: PCI device to query
283 * @cap: capability code
284 *
285 * Returns the address of the requested extended capability structure
286 * within the device's PCI configuration space or 0 if the device does
287 * not support it. Possible values for @cap:
288 *
289 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
290 * %PCI_EXT_CAP_ID_VC Virtual Channel
291 * %PCI_EXT_CAP_ID_DSN Device Serial Number
292 * %PCI_EXT_CAP_ID_PWR Power Budgeting
293 */
294int pci_find_ext_capability(struct pci_dev *dev, int cap)
295{
296 u32 header;
297 int ttl;
298 int pos = PCI_CFG_SPACE_SIZE;
299
300 /* minimum 8 bytes per capability */
301 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
302
303 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
304 return 0;
305
306 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
307 return 0;
308
309 /*
310 * If we have no capabilities, this is indicated by cap ID,
311 * cap version and next pointer all being 0.
312 */
313 if (header == 0)
314 return 0;
315
316 while (ttl-- > 0) {
317 if (PCI_EXT_CAP_ID(header) == cap)
318 return pos;
319
320 pos = PCI_EXT_CAP_NEXT(header);
321 if (pos < PCI_CFG_SPACE_SIZE)
322 break;
323
324 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
325 break;
326 }
327
328 return 0;
329}
330EXPORT_SYMBOL_GPL(pci_find_ext_capability);
331
332/**
333 * pci_bus_find_ext_capability - find an extended capability
334 * @bus: the PCI bus to query
335 * @devfn: PCI device to query
336 * @cap: capability code
337 *
338 * Like pci_find_ext_capability() but works for pci devices that do not have a
339 * pci_dev structure set up yet.
340 *
341 * Returns the address of the requested capability structure within the
342 * device's PCI configuration space or 0 in case the device does not
343 * support it.
344 */
345int pci_bus_find_ext_capability(struct pci_bus *bus, unsigned int devfn,
346 int cap)
347{
348 u32 header;
349 int ttl;
350 int pos = PCI_CFG_SPACE_SIZE;
351
352 /* minimum 8 bytes per capability */
353 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
354
355 if (!pci_bus_read_config_dword(bus, devfn, pos, &header))
356 return 0;
357 if (header == 0xffffffff || header == 0)
358 return 0;
359
360 while (ttl-- > 0) {
361 if (PCI_EXT_CAP_ID(header) == cap)
362 return pos;
363
364 pos = PCI_EXT_CAP_NEXT(header);
365 if (pos < PCI_CFG_SPACE_SIZE)
366 break;
367
368 if (!pci_bus_read_config_dword(bus, devfn, pos, &header))
369 break;
370 }
371
372 return 0;
373}
374
375static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
376{
377 int rc, ttl = PCI_FIND_CAP_TTL;
378 u8 cap, mask;
379
380 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
381 mask = HT_3BIT_CAP_MASK;
382 else
383 mask = HT_5BIT_CAP_MASK;
384
385 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
386 PCI_CAP_ID_HT, &ttl);
387 while (pos) {
388 rc = pci_read_config_byte(dev, pos + 3, &cap);
389 if (rc != PCIBIOS_SUCCESSFUL)
390 return 0;
391
392 if ((cap & mask) == ht_cap)
393 return pos;
394
395 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
396 pos + PCI_CAP_LIST_NEXT,
397 PCI_CAP_ID_HT, &ttl);
398 }
399
400 return 0;
401}
402/**
403 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
404 * @dev: PCI device to query
405 * @pos: Position from which to continue searching
406 * @ht_cap: Hypertransport capability code
407 *
408 * To be used in conjunction with pci_find_ht_capability() to search for
409 * all capabilities matching @ht_cap. @pos should always be a value returned
410 * from pci_find_ht_capability().
411 *
412 * NB. To be 100% safe against broken PCI devices, the caller should take
413 * steps to avoid an infinite loop.
414 */
415int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
416{
417 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
418}
419EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
420
421/**
422 * pci_find_ht_capability - query a device's Hypertransport capabilities
423 * @dev: PCI device to query
424 * @ht_cap: Hypertransport capability code
425 *
426 * Tell if a device supports a given Hypertransport capability.
427 * Returns an address within the device's PCI configuration space
428 * or 0 in case the device does not support the request capability.
429 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
430 * which has a Hypertransport capability matching @ht_cap.
431 */
432int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
433{
434 int pos;
435
436 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
437 if (pos)
438 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
439
440 return pos;
441}
442EXPORT_SYMBOL_GPL(pci_find_ht_capability);
443
444/**
445 * pci_find_parent_resource - return resource region of parent bus of given region
446 * @dev: PCI device structure contains resources to be searched
447 * @res: child resource record for which parent is sought
448 *
449 * For given resource region of given device, return the resource
450 * region of parent bus the given region is contained in or where
451 * it should be allocated from.
452 */
453struct resource *
454pci_find_parent_resource(const struct pci_dev *dev, struct resource *res)
455{
456 const struct pci_bus *bus = dev->bus;
457 int i;
458 struct resource *best = NULL, *r;
459
460 pci_bus_for_each_resource(bus, r, i) {
461 if (!r)
462 continue;
463 if (res->start && !(res->start >= r->start && res->end <= r->end))
464 continue; /* Not contained */
465 if ((res->flags ^ r->flags) & (IORESOURCE_IO | IORESOURCE_MEM))
466 continue; /* Wrong type */
467 if (!((res->flags ^ r->flags) & IORESOURCE_PREFETCH))
468 return r; /* Exact match */
469 /* We can't insert a non-prefetch resource inside a prefetchable parent .. */
470 if (r->flags & IORESOURCE_PREFETCH)
471 continue;
472 /* .. but we can put a prefetchable resource inside a non-prefetchable one */
473 if (!best)
474 best = r;
475 }
476 return best;
477}
478
479/**
480 * pci_restore_bars - restore a devices BAR values (e.g. after wake-up)
481 * @dev: PCI device to have its BARs restored
482 *
483 * Restore the BAR values for a given device, so as to make it
484 * accessible by its driver.
485 */
486static void
487pci_restore_bars(struct pci_dev *dev)
488{
489 int i;
490
491 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
492 pci_update_resource(dev, i);
493}
494
495static struct pci_platform_pm_ops *pci_platform_pm;
496
497int pci_set_platform_pm(struct pci_platform_pm_ops *ops)
498{
499 if (!ops->is_manageable || !ops->set_state || !ops->choose_state
500 || !ops->sleep_wake || !ops->can_wakeup)
501 return -EINVAL;
502 pci_platform_pm = ops;
503 return 0;
504}
505
506static inline bool platform_pci_power_manageable(struct pci_dev *dev)
507{
508 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
509}
510
511static inline int platform_pci_set_power_state(struct pci_dev *dev,
512 pci_power_t t)
513{
514 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
515}
516
517static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
518{
519 return pci_platform_pm ?
520 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
521}
522
523static inline bool platform_pci_can_wakeup(struct pci_dev *dev)
524{
525 return pci_platform_pm ? pci_platform_pm->can_wakeup(dev) : false;
526}
527
528static inline int platform_pci_sleep_wake(struct pci_dev *dev, bool enable)
529{
530 return pci_platform_pm ?
531 pci_platform_pm->sleep_wake(dev, enable) : -ENODEV;
532}
533
534static inline int platform_pci_run_wake(struct pci_dev *dev, bool enable)
535{
536 return pci_platform_pm ?
537 pci_platform_pm->run_wake(dev, enable) : -ENODEV;
538}
539
540/**
541 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
542 * given PCI device
543 * @dev: PCI device to handle.
544 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
545 *
546 * RETURN VALUE:
547 * -EINVAL if the requested state is invalid.
548 * -EIO if device does not support PCI PM or its PM capabilities register has a
549 * wrong version, or device doesn't support the requested state.
550 * 0 if device already is in the requested state.
551 * 0 if device's power state has been successfully changed.
552 */
553static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
554{
555 u16 pmcsr;
556 bool need_restore = false;
557
558 /* Check if we're already there */
559 if (dev->current_state == state)
560 return 0;
561
562 if (!dev->pm_cap)
563 return -EIO;
564
565 if (state < PCI_D0 || state > PCI_D3hot)
566 return -EINVAL;
567
568 /* Validate current state:
569 * Can enter D0 from any state, but if we can only go deeper
570 * to sleep if we're already in a low power state
571 */
572 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
573 && dev->current_state > state) {
574 dev_err(&dev->dev, "invalid power transition "
575 "(from state %d to %d)\n", dev->current_state, state);
576 return -EINVAL;
577 }
578
579 /* check if this device supports the desired state */
580 if ((state == PCI_D1 && !dev->d1_support)
581 || (state == PCI_D2 && !dev->d2_support))
582 return -EIO;
583
584 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
585
586 /* If we're (effectively) in D3, force entire word to 0.
587 * This doesn't affect PME_Status, disables PME_En, and
588 * sets PowerState to 0.
589 */
590 switch (dev->current_state) {
591 case PCI_D0:
592 case PCI_D1:
593 case PCI_D2:
594 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
595 pmcsr |= state;
596 break;
597 case PCI_D3hot:
598 case PCI_D3cold:
599 case PCI_UNKNOWN: /* Boot-up */
600 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
601 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
602 need_restore = true;
603 /* Fall-through: force to D0 */
604 default:
605 pmcsr = 0;
606 break;
607 }
608
609 /* enter specified state */
610 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
611
612 /* Mandatory power management transition delays */
613 /* see PCI PM 1.1 5.6.1 table 18 */
614 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
615 pci_dev_d3_sleep(dev);
616 else if (state == PCI_D2 || dev->current_state == PCI_D2)
617 udelay(PCI_PM_D2_DELAY);
618
619 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
620 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
621 if (dev->current_state != state && printk_ratelimit())
622 dev_info(&dev->dev, "Refused to change power state, "
623 "currently in D%d\n", dev->current_state);
624
625 /* According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
626 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
627 * from D3hot to D0 _may_ perform an internal reset, thereby
628 * going to "D0 Uninitialized" rather than "D0 Initialized".
629 * For example, at least some versions of the 3c905B and the
630 * 3c556B exhibit this behaviour.
631 *
632 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
633 * devices in a D3hot state at boot. Consequently, we need to
634 * restore at least the BARs so that the device will be
635 * accessible to its driver.
636 */
637 if (need_restore)
638 pci_restore_bars(dev);
639
640 if (dev->bus->self)
641 pcie_aspm_pm_state_change(dev->bus->self);
642
643 return 0;
644}
645
646/**
647 * pci_update_current_state - Read PCI power state of given device from its
648 * PCI PM registers and cache it
649 * @dev: PCI device to handle.
650 * @state: State to cache in case the device doesn't have the PM capability
651 */
652void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
653{
654 if (dev->pm_cap) {
655 u16 pmcsr;
656
657 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
658 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
659 } else {
660 dev->current_state = state;
661 }
662}
663
664/**
665 * pci_platform_power_transition - Use platform to change device power state
666 * @dev: PCI device to handle.
667 * @state: State to put the device into.
668 */
669static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
670{
671 int error;
672
673 if (platform_pci_power_manageable(dev)) {
674 error = platform_pci_set_power_state(dev, state);
675 if (!error)
676 pci_update_current_state(dev, state);
677 /* Fall back to PCI_D0 if native PM is not supported */
678 if (!dev->pm_cap)
679 dev->current_state = PCI_D0;
680 } else {
681 error = -ENODEV;
682 /* Fall back to PCI_D0 if native PM is not supported */
683 if (!dev->pm_cap)
684 dev->current_state = PCI_D0;
685 }
686
687 return error;
688}
689
690/**
691 * __pci_start_power_transition - Start power transition of a PCI device
692 * @dev: PCI device to handle.
693 * @state: State to put the device into.
694 */
695static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
696{
697 if (state == PCI_D0)
698 pci_platform_power_transition(dev, PCI_D0);
699}
700
701/**
702 * __pci_complete_power_transition - Complete power transition of a PCI device
703 * @dev: PCI device to handle.
704 * @state: State to put the device into.
705 *
706 * This function should not be called directly by device drivers.
707 */
708int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
709{
710 return state >= PCI_D0 ?
711 pci_platform_power_transition(dev, state) : -EINVAL;
712}
713EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
714
715/**
716 * pci_set_power_state - Set the power state of a PCI device
717 * @dev: PCI device to handle.
718 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
719 *
720 * Transition a device to a new power state, using the platform firmware and/or
721 * the device's PCI PM registers.
722 *
723 * RETURN VALUE:
724 * -EINVAL if the requested state is invalid.
725 * -EIO if device does not support PCI PM or its PM capabilities register has a
726 * wrong version, or device doesn't support the requested state.
727 * 0 if device already is in the requested state.
728 * 0 if device's power state has been successfully changed.
729 */
730int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
731{
732 int error;
733
734 /* bound the state we're entering */
735 if (state > PCI_D3hot)
736 state = PCI_D3hot;
737 else if (state < PCI_D0)
738 state = PCI_D0;
739 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
740 /*
741 * If the device or the parent bridge do not support PCI PM,
742 * ignore the request if we're doing anything other than putting
743 * it into D0 (which would only happen on boot).
744 */
745 return 0;
746
747 __pci_start_power_transition(dev, state);
748
749 /* This device is quirked not to be put into D3, so
750 don't put it in D3 */
751 if (state == PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
752 return 0;
753
754 error = pci_raw_set_power_state(dev, state);
755
756 if (!__pci_complete_power_transition(dev, state))
757 error = 0;
758 /*
759 * When aspm_policy is "powersave" this call ensures
760 * that ASPM is configured.
761 */
762 if (!error && dev->bus->self)
763 pcie_aspm_powersave_config_link(dev->bus->self);
764
765 return error;
766}
767
768/**
769 * pci_choose_state - Choose the power state of a PCI device
770 * @dev: PCI device to be suspended
771 * @state: target sleep state for the whole system. This is the value
772 * that is passed to suspend() function.
773 *
774 * Returns PCI power state suitable for given device and given system
775 * message.
776 */
777
778pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
779{
780 pci_power_t ret;
781
782 if (!pci_find_capability(dev, PCI_CAP_ID_PM))
783 return PCI_D0;
784
785 ret = platform_pci_choose_state(dev);
786 if (ret != PCI_POWER_ERROR)
787 return ret;
788
789 switch (state.event) {
790 case PM_EVENT_ON:
791 return PCI_D0;
792 case PM_EVENT_FREEZE:
793 case PM_EVENT_PRETHAW:
794 /* REVISIT both freeze and pre-thaw "should" use D0 */
795 case PM_EVENT_SUSPEND:
796 case PM_EVENT_HIBERNATE:
797 return PCI_D3hot;
798 default:
799 dev_info(&dev->dev, "unrecognized suspend event %d\n",
800 state.event);
801 BUG();
802 }
803 return PCI_D0;
804}
805
806EXPORT_SYMBOL(pci_choose_state);
807
808#define PCI_EXP_SAVE_REGS 7
809
810#define pcie_cap_has_devctl(type, flags) 1
811#define pcie_cap_has_lnkctl(type, flags) \
812 ((flags & PCI_EXP_FLAGS_VERS) > 1 || \
813 (type == PCI_EXP_TYPE_ROOT_PORT || \
814 type == PCI_EXP_TYPE_ENDPOINT || \
815 type == PCI_EXP_TYPE_LEG_END))
816#define pcie_cap_has_sltctl(type, flags) \
817 ((flags & PCI_EXP_FLAGS_VERS) > 1 || \
818 ((type == PCI_EXP_TYPE_ROOT_PORT) || \
819 (type == PCI_EXP_TYPE_DOWNSTREAM && \
820 (flags & PCI_EXP_FLAGS_SLOT))))
821#define pcie_cap_has_rtctl(type, flags) \
822 ((flags & PCI_EXP_FLAGS_VERS) > 1 || \
823 (type == PCI_EXP_TYPE_ROOT_PORT || \
824 type == PCI_EXP_TYPE_RC_EC))
825#define pcie_cap_has_devctl2(type, flags) \
826 ((flags & PCI_EXP_FLAGS_VERS) > 1)
827#define pcie_cap_has_lnkctl2(type, flags) \
828 ((flags & PCI_EXP_FLAGS_VERS) > 1)
829#define pcie_cap_has_sltctl2(type, flags) \
830 ((flags & PCI_EXP_FLAGS_VERS) > 1)
831
832static struct pci_cap_saved_state *pci_find_saved_cap(
833 struct pci_dev *pci_dev, char cap)
834{
835 struct pci_cap_saved_state *tmp;
836 struct hlist_node *pos;
837
838 hlist_for_each_entry(tmp, pos, &pci_dev->saved_cap_space, next) {
839 if (tmp->cap.cap_nr == cap)
840 return tmp;
841 }
842 return NULL;
843}
844
845static int pci_save_pcie_state(struct pci_dev *dev)
846{
847 int pos, i = 0;
848 struct pci_cap_saved_state *save_state;
849 u16 *cap;
850 u16 flags;
851
852 pos = pci_pcie_cap(dev);
853 if (!pos)
854 return 0;
855
856 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
857 if (!save_state) {
858 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
859 return -ENOMEM;
860 }
861 cap = (u16 *)&save_state->cap.data[0];
862
863 pci_read_config_word(dev, pos + PCI_EXP_FLAGS, &flags);
864
865 if (pcie_cap_has_devctl(dev->pcie_type, flags))
866 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &cap[i++]);
867 if (pcie_cap_has_lnkctl(dev->pcie_type, flags))
868 pci_read_config_word(dev, pos + PCI_EXP_LNKCTL, &cap[i++]);
869 if (pcie_cap_has_sltctl(dev->pcie_type, flags))
870 pci_read_config_word(dev, pos + PCI_EXP_SLTCTL, &cap[i++]);
871 if (pcie_cap_has_rtctl(dev->pcie_type, flags))
872 pci_read_config_word(dev, pos + PCI_EXP_RTCTL, &cap[i++]);
873 if (pcie_cap_has_devctl2(dev->pcie_type, flags))
874 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &cap[i++]);
875 if (pcie_cap_has_lnkctl2(dev->pcie_type, flags))
876 pci_read_config_word(dev, pos + PCI_EXP_LNKCTL2, &cap[i++]);
877 if (pcie_cap_has_sltctl2(dev->pcie_type, flags))
878 pci_read_config_word(dev, pos + PCI_EXP_SLTCTL2, &cap[i++]);
879
880 return 0;
881}
882
883static void pci_restore_pcie_state(struct pci_dev *dev)
884{
885 int i = 0, pos;
886 struct pci_cap_saved_state *save_state;
887 u16 *cap;
888 u16 flags;
889
890 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
891 pos = pci_find_capability(dev, PCI_CAP_ID_EXP);
892 if (!save_state || pos <= 0)
893 return;
894 cap = (u16 *)&save_state->cap.data[0];
895
896 pci_read_config_word(dev, pos + PCI_EXP_FLAGS, &flags);
897
898 if (pcie_cap_has_devctl(dev->pcie_type, flags))
899 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, cap[i++]);
900 if (pcie_cap_has_lnkctl(dev->pcie_type, flags))
901 pci_write_config_word(dev, pos + PCI_EXP_LNKCTL, cap[i++]);
902 if (pcie_cap_has_sltctl(dev->pcie_type, flags))
903 pci_write_config_word(dev, pos + PCI_EXP_SLTCTL, cap[i++]);
904 if (pcie_cap_has_rtctl(dev->pcie_type, flags))
905 pci_write_config_word(dev, pos + PCI_EXP_RTCTL, cap[i++]);
906 if (pcie_cap_has_devctl2(dev->pcie_type, flags))
907 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, cap[i++]);
908 if (pcie_cap_has_lnkctl2(dev->pcie_type, flags))
909 pci_write_config_word(dev, pos + PCI_EXP_LNKCTL2, cap[i++]);
910 if (pcie_cap_has_sltctl2(dev->pcie_type, flags))
911 pci_write_config_word(dev, pos + PCI_EXP_SLTCTL2, cap[i++]);
912}
913
914
915static int pci_save_pcix_state(struct pci_dev *dev)
916{
917 int pos;
918 struct pci_cap_saved_state *save_state;
919
920 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
921 if (pos <= 0)
922 return 0;
923
924 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
925 if (!save_state) {
926 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
927 return -ENOMEM;
928 }
929
930 pci_read_config_word(dev, pos + PCI_X_CMD,
931 (u16 *)save_state->cap.data);
932
933 return 0;
934}
935
936static void pci_restore_pcix_state(struct pci_dev *dev)
937{
938 int i = 0, pos;
939 struct pci_cap_saved_state *save_state;
940 u16 *cap;
941
942 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
943 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
944 if (!save_state || pos <= 0)
945 return;
946 cap = (u16 *)&save_state->cap.data[0];
947
948 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
949}
950
951
952/**
953 * pci_save_state - save the PCI configuration space of a device before suspending
954 * @dev: - PCI device that we're dealing with
955 */
956int
957pci_save_state(struct pci_dev *dev)
958{
959 int i;
960 /* XXX: 100% dword access ok here? */
961 for (i = 0; i < 16; i++)
962 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
963 dev->state_saved = true;
964 if ((i = pci_save_pcie_state(dev)) != 0)
965 return i;
966 if ((i = pci_save_pcix_state(dev)) != 0)
967 return i;
968 return 0;
969}
970
971static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
972 u32 saved_val, int retry)
973{
974 u32 val;
975
976 pci_read_config_dword(pdev, offset, &val);
977 if (val == saved_val)
978 return;
979
980 for (;;) {
981 dev_dbg(&pdev->dev, "restoring config space at offset "
982 "%#x (was %#x, writing %#x)\n", offset, val, saved_val);
983 pci_write_config_dword(pdev, offset, saved_val);
984 if (retry-- <= 0)
985 return;
986
987 pci_read_config_dword(pdev, offset, &val);
988 if (val == saved_val)
989 return;
990
991 mdelay(1);
992 }
993}
994
995static void pci_restore_config_space_range(struct pci_dev *pdev,
996 int start, int end, int retry)
997{
998 int index;
999
1000 for (index = end; index >= start; index--)
1001 pci_restore_config_dword(pdev, 4 * index,
1002 pdev->saved_config_space[index],
1003 retry);
1004}
1005
1006static void pci_restore_config_space(struct pci_dev *pdev)
1007{
1008 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1009 pci_restore_config_space_range(pdev, 10, 15, 0);
1010 /* Restore BARs before the command register. */
1011 pci_restore_config_space_range(pdev, 4, 9, 10);
1012 pci_restore_config_space_range(pdev, 0, 3, 0);
1013 } else {
1014 pci_restore_config_space_range(pdev, 0, 15, 0);
1015 }
1016}
1017
1018/**
1019 * pci_restore_state - Restore the saved state of a PCI device
1020 * @dev: - PCI device that we're dealing with
1021 */
1022void pci_restore_state(struct pci_dev *dev)
1023{
1024 if (!dev->state_saved)
1025 return;
1026
1027 /* PCI Express register must be restored first */
1028 pci_restore_pcie_state(dev);
1029 pci_restore_ats_state(dev);
1030
1031 pci_restore_config_space(dev);
1032
1033 pci_restore_pcix_state(dev);
1034 pci_restore_msi_state(dev);
1035 pci_restore_iov_state(dev);
1036
1037 dev->state_saved = false;
1038}
1039
1040struct pci_saved_state {
1041 u32 config_space[16];
1042 struct pci_cap_saved_data cap[0];
1043};
1044
1045/**
1046 * pci_store_saved_state - Allocate and return an opaque struct containing
1047 * the device saved state.
1048 * @dev: PCI device that we're dealing with
1049 *
1050 * Rerturn NULL if no state or error.
1051 */
1052struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1053{
1054 struct pci_saved_state *state;
1055 struct pci_cap_saved_state *tmp;
1056 struct pci_cap_saved_data *cap;
1057 struct hlist_node *pos;
1058 size_t size;
1059
1060 if (!dev->state_saved)
1061 return NULL;
1062
1063 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1064
1065 hlist_for_each_entry(tmp, pos, &dev->saved_cap_space, next)
1066 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1067
1068 state = kzalloc(size, GFP_KERNEL);
1069 if (!state)
1070 return NULL;
1071
1072 memcpy(state->config_space, dev->saved_config_space,
1073 sizeof(state->config_space));
1074
1075 cap = state->cap;
1076 hlist_for_each_entry(tmp, pos, &dev->saved_cap_space, next) {
1077 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1078 memcpy(cap, &tmp->cap, len);
1079 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1080 }
1081 /* Empty cap_save terminates list */
1082
1083 return state;
1084}
1085EXPORT_SYMBOL_GPL(pci_store_saved_state);
1086
1087/**
1088 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1089 * @dev: PCI device that we're dealing with
1090 * @state: Saved state returned from pci_store_saved_state()
1091 */
1092int pci_load_saved_state(struct pci_dev *dev, struct pci_saved_state *state)
1093{
1094 struct pci_cap_saved_data *cap;
1095
1096 dev->state_saved = false;
1097
1098 if (!state)
1099 return 0;
1100
1101 memcpy(dev->saved_config_space, state->config_space,
1102 sizeof(state->config_space));
1103
1104 cap = state->cap;
1105 while (cap->size) {
1106 struct pci_cap_saved_state *tmp;
1107
1108 tmp = pci_find_saved_cap(dev, cap->cap_nr);
1109 if (!tmp || tmp->cap.size != cap->size)
1110 return -EINVAL;
1111
1112 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1113 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1114 sizeof(struct pci_cap_saved_data) + cap->size);
1115 }
1116
1117 dev->state_saved = true;
1118 return 0;
1119}
1120EXPORT_SYMBOL_GPL(pci_load_saved_state);
1121
1122/**
1123 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1124 * and free the memory allocated for it.
1125 * @dev: PCI device that we're dealing with
1126 * @state: Pointer to saved state returned from pci_store_saved_state()
1127 */
1128int pci_load_and_free_saved_state(struct pci_dev *dev,
1129 struct pci_saved_state **state)
1130{
1131 int ret = pci_load_saved_state(dev, *state);
1132 kfree(*state);
1133 *state = NULL;
1134 return ret;
1135}
1136EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1137
1138static int do_pci_enable_device(struct pci_dev *dev, int bars)
1139{
1140 int err;
1141
1142 err = pci_set_power_state(dev, PCI_D0);
1143 if (err < 0 && err != -EIO)
1144 return err;
1145 err = pcibios_enable_device(dev, bars);
1146 if (err < 0)
1147 return err;
1148 pci_fixup_device(pci_fixup_enable, dev);
1149
1150 return 0;
1151}
1152
1153/**
1154 * pci_reenable_device - Resume abandoned device
1155 * @dev: PCI device to be resumed
1156 *
1157 * Note this function is a backend of pci_default_resume and is not supposed
1158 * to be called by normal code, write proper resume handler and use it instead.
1159 */
1160int pci_reenable_device(struct pci_dev *dev)
1161{
1162 if (pci_is_enabled(dev))
1163 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1164 return 0;
1165}
1166
1167static int __pci_enable_device_flags(struct pci_dev *dev,
1168 resource_size_t flags)
1169{
1170 int err;
1171 int i, bars = 0;
1172
1173 /*
1174 * Power state could be unknown at this point, either due to a fresh
1175 * boot or a device removal call. So get the current power state
1176 * so that things like MSI message writing will behave as expected
1177 * (e.g. if the device really is in D0 at enable time).
1178 */
1179 if (dev->pm_cap) {
1180 u16 pmcsr;
1181 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1182 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1183 }
1184
1185 if (atomic_add_return(1, &dev->enable_cnt) > 1)
1186 return 0; /* already enabled */
1187
1188 /* only skip sriov related */
1189 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1190 if (dev->resource[i].flags & flags)
1191 bars |= (1 << i);
1192 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1193 if (dev->resource[i].flags & flags)
1194 bars |= (1 << i);
1195
1196 err = do_pci_enable_device(dev, bars);
1197 if (err < 0)
1198 atomic_dec(&dev->enable_cnt);
1199 return err;
1200}
1201
1202/**
1203 * pci_enable_device_io - Initialize a device for use with IO space
1204 * @dev: PCI device to be initialized
1205 *
1206 * Initialize device before it's used by a driver. Ask low-level code
1207 * to enable I/O resources. Wake up the device if it was suspended.
1208 * Beware, this function can fail.
1209 */
1210int pci_enable_device_io(struct pci_dev *dev)
1211{
1212 return __pci_enable_device_flags(dev, IORESOURCE_IO);
1213}
1214
1215/**
1216 * pci_enable_device_mem - Initialize a device for use with Memory space
1217 * @dev: PCI device to be initialized
1218 *
1219 * Initialize device before it's used by a driver. Ask low-level code
1220 * to enable Memory resources. Wake up the device if it was suspended.
1221 * Beware, this function can fail.
1222 */
1223int pci_enable_device_mem(struct pci_dev *dev)
1224{
1225 return __pci_enable_device_flags(dev, IORESOURCE_MEM);
1226}
1227
1228/**
1229 * pci_enable_device - Initialize device before it's used by a driver.
1230 * @dev: PCI device to be initialized
1231 *
1232 * Initialize device before it's used by a driver. Ask low-level code
1233 * to enable I/O and memory. Wake up the device if it was suspended.
1234 * Beware, this function can fail.
1235 *
1236 * Note we don't actually enable the device many times if we call
1237 * this function repeatedly (we just increment the count).
1238 */
1239int pci_enable_device(struct pci_dev *dev)
1240{
1241 return __pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1242}
1243
1244/*
1245 * Managed PCI resources. This manages device on/off, intx/msi/msix
1246 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1247 * there's no need to track it separately. pci_devres is initialized
1248 * when a device is enabled using managed PCI device enable interface.
1249 */
1250struct pci_devres {
1251 unsigned int enabled:1;
1252 unsigned int pinned:1;
1253 unsigned int orig_intx:1;
1254 unsigned int restore_intx:1;
1255 u32 region_mask;
1256};
1257
1258static void pcim_release(struct device *gendev, void *res)
1259{
1260 struct pci_dev *dev = container_of(gendev, struct pci_dev, dev);
1261 struct pci_devres *this = res;
1262 int i;
1263
1264 if (dev->msi_enabled)
1265 pci_disable_msi(dev);
1266 if (dev->msix_enabled)
1267 pci_disable_msix(dev);
1268
1269 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1270 if (this->region_mask & (1 << i))
1271 pci_release_region(dev, i);
1272
1273 if (this->restore_intx)
1274 pci_intx(dev, this->orig_intx);
1275
1276 if (this->enabled && !this->pinned)
1277 pci_disable_device(dev);
1278}
1279
1280static struct pci_devres * get_pci_dr(struct pci_dev *pdev)
1281{
1282 struct pci_devres *dr, *new_dr;
1283
1284 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1285 if (dr)
1286 return dr;
1287
1288 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1289 if (!new_dr)
1290 return NULL;
1291 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1292}
1293
1294static struct pci_devres * find_pci_dr(struct pci_dev *pdev)
1295{
1296 if (pci_is_managed(pdev))
1297 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1298 return NULL;
1299}
1300
1301/**
1302 * pcim_enable_device - Managed pci_enable_device()
1303 * @pdev: PCI device to be initialized
1304 *
1305 * Managed pci_enable_device().
1306 */
1307int pcim_enable_device(struct pci_dev *pdev)
1308{
1309 struct pci_devres *dr;
1310 int rc;
1311
1312 dr = get_pci_dr(pdev);
1313 if (unlikely(!dr))
1314 return -ENOMEM;
1315 if (dr->enabled)
1316 return 0;
1317
1318 rc = pci_enable_device(pdev);
1319 if (!rc) {
1320 pdev->is_managed = 1;
1321 dr->enabled = 1;
1322 }
1323 return rc;
1324}
1325
1326/**
1327 * pcim_pin_device - Pin managed PCI device
1328 * @pdev: PCI device to pin
1329 *
1330 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1331 * driver detach. @pdev must have been enabled with
1332 * pcim_enable_device().
1333 */
1334void pcim_pin_device(struct pci_dev *pdev)
1335{
1336 struct pci_devres *dr;
1337
1338 dr = find_pci_dr(pdev);
1339 WARN_ON(!dr || !dr->enabled);
1340 if (dr)
1341 dr->pinned = 1;
1342}
1343
1344/**
1345 * pcibios_disable_device - disable arch specific PCI resources for device dev
1346 * @dev: the PCI device to disable
1347 *
1348 * Disables architecture specific PCI resources for the device. This
1349 * is the default implementation. Architecture implementations can
1350 * override this.
1351 */
1352void __attribute__ ((weak)) pcibios_disable_device (struct pci_dev *dev) {}
1353
1354static void do_pci_disable_device(struct pci_dev *dev)
1355{
1356 u16 pci_command;
1357
1358 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1359 if (pci_command & PCI_COMMAND_MASTER) {
1360 pci_command &= ~PCI_COMMAND_MASTER;
1361 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1362 }
1363
1364 pcibios_disable_device(dev);
1365}
1366
1367/**
1368 * pci_disable_enabled_device - Disable device without updating enable_cnt
1369 * @dev: PCI device to disable
1370 *
1371 * NOTE: This function is a backend of PCI power management routines and is
1372 * not supposed to be called drivers.
1373 */
1374void pci_disable_enabled_device(struct pci_dev *dev)
1375{
1376 if (pci_is_enabled(dev))
1377 do_pci_disable_device(dev);
1378}
1379
1380/**
1381 * pci_disable_device - Disable PCI device after use
1382 * @dev: PCI device to be disabled
1383 *
1384 * Signal to the system that the PCI device is not in use by the system
1385 * anymore. This only involves disabling PCI bus-mastering, if active.
1386 *
1387 * Note we don't actually disable the device until all callers of
1388 * pci_enable_device() have called pci_disable_device().
1389 */
1390void
1391pci_disable_device(struct pci_dev *dev)
1392{
1393 struct pci_devres *dr;
1394
1395 dr = find_pci_dr(dev);
1396 if (dr)
1397 dr->enabled = 0;
1398
1399 if (atomic_sub_return(1, &dev->enable_cnt) != 0)
1400 return;
1401
1402 do_pci_disable_device(dev);
1403
1404 dev->is_busmaster = 0;
1405}
1406
1407/**
1408 * pcibios_set_pcie_reset_state - set reset state for device dev
1409 * @dev: the PCIe device reset
1410 * @state: Reset state to enter into
1411 *
1412 *
1413 * Sets the PCIe reset state for the device. This is the default
1414 * implementation. Architecture implementations can override this.
1415 */
1416int __attribute__ ((weak)) pcibios_set_pcie_reset_state(struct pci_dev *dev,
1417 enum pcie_reset_state state)
1418{
1419 return -EINVAL;
1420}
1421
1422/**
1423 * pci_set_pcie_reset_state - set reset state for device dev
1424 * @dev: the PCIe device reset
1425 * @state: Reset state to enter into
1426 *
1427 *
1428 * Sets the PCI reset state for the device.
1429 */
1430int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1431{
1432 return pcibios_set_pcie_reset_state(dev, state);
1433}
1434
1435/**
1436 * pci_check_pme_status - Check if given device has generated PME.
1437 * @dev: Device to check.
1438 *
1439 * Check the PME status of the device and if set, clear it and clear PME enable
1440 * (if set). Return 'true' if PME status and PME enable were both set or
1441 * 'false' otherwise.
1442 */
1443bool pci_check_pme_status(struct pci_dev *dev)
1444{
1445 int pmcsr_pos;
1446 u16 pmcsr;
1447 bool ret = false;
1448
1449 if (!dev->pm_cap)
1450 return false;
1451
1452 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1453 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1454 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1455 return false;
1456
1457 /* Clear PME status. */
1458 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1459 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1460 /* Disable PME to avoid interrupt flood. */
1461 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1462 ret = true;
1463 }
1464
1465 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1466
1467 return ret;
1468}
1469
1470/**
1471 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1472 * @dev: Device to handle.
1473 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1474 *
1475 * Check if @dev has generated PME and queue a resume request for it in that
1476 * case.
1477 */
1478static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1479{
1480 if (pme_poll_reset && dev->pme_poll)
1481 dev->pme_poll = false;
1482
1483 if (pci_check_pme_status(dev)) {
1484 pci_wakeup_event(dev);
1485 pm_request_resume(&dev->dev);
1486 }
1487 return 0;
1488}
1489
1490/**
1491 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1492 * @bus: Top bus of the subtree to walk.
1493 */
1494void pci_pme_wakeup_bus(struct pci_bus *bus)
1495{
1496 if (bus)
1497 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1498}
1499
1500/**
1501 * pci_pme_capable - check the capability of PCI device to generate PME#
1502 * @dev: PCI device to handle.
1503 * @state: PCI state from which device will issue PME#.
1504 */
1505bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1506{
1507 if (!dev->pm_cap)
1508 return false;
1509
1510 return !!(dev->pme_support & (1 << state));
1511}
1512
1513static void pci_pme_list_scan(struct work_struct *work)
1514{
1515 struct pci_pme_device *pme_dev, *n;
1516
1517 mutex_lock(&pci_pme_list_mutex);
1518 if (!list_empty(&pci_pme_list)) {
1519 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1520 if (pme_dev->dev->pme_poll) {
1521 pci_pme_wakeup(pme_dev->dev, NULL);
1522 } else {
1523 list_del(&pme_dev->list);
1524 kfree(pme_dev);
1525 }
1526 }
1527 if (!list_empty(&pci_pme_list))
1528 schedule_delayed_work(&pci_pme_work,
1529 msecs_to_jiffies(PME_TIMEOUT));
1530 }
1531 mutex_unlock(&pci_pme_list_mutex);
1532}
1533
1534/**
1535 * pci_pme_active - enable or disable PCI device's PME# function
1536 * @dev: PCI device to handle.
1537 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1538 *
1539 * The caller must verify that the device is capable of generating PME# before
1540 * calling this function with @enable equal to 'true'.
1541 */
1542void pci_pme_active(struct pci_dev *dev, bool enable)
1543{
1544 u16 pmcsr;
1545
1546 if (!dev->pm_cap)
1547 return;
1548
1549 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1550 /* Clear PME_Status by writing 1 to it and enable PME# */
1551 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1552 if (!enable)
1553 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1554
1555 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1556
1557 /* PCI (as opposed to PCIe) PME requires that the device have
1558 its PME# line hooked up correctly. Not all hardware vendors
1559 do this, so the PME never gets delivered and the device
1560 remains asleep. The easiest way around this is to
1561 periodically walk the list of suspended devices and check
1562 whether any have their PME flag set. The assumption is that
1563 we'll wake up often enough anyway that this won't be a huge
1564 hit, and the power savings from the devices will still be a
1565 win. */
1566
1567 if (dev->pme_poll) {
1568 struct pci_pme_device *pme_dev;
1569 if (enable) {
1570 pme_dev = kmalloc(sizeof(struct pci_pme_device),
1571 GFP_KERNEL);
1572 if (!pme_dev)
1573 goto out;
1574 pme_dev->dev = dev;
1575 mutex_lock(&pci_pme_list_mutex);
1576 list_add(&pme_dev->list, &pci_pme_list);
1577 if (list_is_singular(&pci_pme_list))
1578 schedule_delayed_work(&pci_pme_work,
1579 msecs_to_jiffies(PME_TIMEOUT));
1580 mutex_unlock(&pci_pme_list_mutex);
1581 } else {
1582 mutex_lock(&pci_pme_list_mutex);
1583 list_for_each_entry(pme_dev, &pci_pme_list, list) {
1584 if (pme_dev->dev == dev) {
1585 list_del(&pme_dev->list);
1586 kfree(pme_dev);
1587 break;
1588 }
1589 }
1590 mutex_unlock(&pci_pme_list_mutex);
1591 }
1592 }
1593
1594out:
1595 dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1596}
1597
1598/**
1599 * __pci_enable_wake - enable PCI device as wakeup event source
1600 * @dev: PCI device affected
1601 * @state: PCI state from which device will issue wakeup events
1602 * @runtime: True if the events are to be generated at run time
1603 * @enable: True to enable event generation; false to disable
1604 *
1605 * This enables the device as a wakeup event source, or disables it.
1606 * When such events involves platform-specific hooks, those hooks are
1607 * called automatically by this routine.
1608 *
1609 * Devices with legacy power management (no standard PCI PM capabilities)
1610 * always require such platform hooks.
1611 *
1612 * RETURN VALUE:
1613 * 0 is returned on success
1614 * -EINVAL is returned if device is not supposed to wake up the system
1615 * Error code depending on the platform is returned if both the platform and
1616 * the native mechanism fail to enable the generation of wake-up events
1617 */
1618int __pci_enable_wake(struct pci_dev *dev, pci_power_t state,
1619 bool runtime, bool enable)
1620{
1621 int ret = 0;
1622
1623 if (enable && !runtime && !device_may_wakeup(&dev->dev))
1624 return -EINVAL;
1625
1626 /* Don't do the same thing twice in a row for one device. */
1627 if (!!enable == !!dev->wakeup_prepared)
1628 return 0;
1629
1630 /*
1631 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1632 * Anderson we should be doing PME# wake enable followed by ACPI wake
1633 * enable. To disable wake-up we call the platform first, for symmetry.
1634 */
1635
1636 if (enable) {
1637 int error;
1638
1639 if (pci_pme_capable(dev, state))
1640 pci_pme_active(dev, true);
1641 else
1642 ret = 1;
1643 error = runtime ? platform_pci_run_wake(dev, true) :
1644 platform_pci_sleep_wake(dev, true);
1645 if (ret)
1646 ret = error;
1647 if (!ret)
1648 dev->wakeup_prepared = true;
1649 } else {
1650 if (runtime)
1651 platform_pci_run_wake(dev, false);
1652 else
1653 platform_pci_sleep_wake(dev, false);
1654 pci_pme_active(dev, false);
1655 dev->wakeup_prepared = false;
1656 }
1657
1658 return ret;
1659}
1660EXPORT_SYMBOL(__pci_enable_wake);
1661
1662/**
1663 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1664 * @dev: PCI device to prepare
1665 * @enable: True to enable wake-up event generation; false to disable
1666 *
1667 * Many drivers want the device to wake up the system from D3_hot or D3_cold
1668 * and this function allows them to set that up cleanly - pci_enable_wake()
1669 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1670 * ordering constraints.
1671 *
1672 * This function only returns error code if the device is not capable of
1673 * generating PME# from both D3_hot and D3_cold, and the platform is unable to
1674 * enable wake-up power for it.
1675 */
1676int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1677{
1678 return pci_pme_capable(dev, PCI_D3cold) ?
1679 pci_enable_wake(dev, PCI_D3cold, enable) :
1680 pci_enable_wake(dev, PCI_D3hot, enable);
1681}
1682
1683/**
1684 * pci_target_state - find an appropriate low power state for a given PCI dev
1685 * @dev: PCI device
1686 *
1687 * Use underlying platform code to find a supported low power state for @dev.
1688 * If the platform can't manage @dev, return the deepest state from which it
1689 * can generate wake events, based on any available PME info.
1690 */
1691pci_power_t pci_target_state(struct pci_dev *dev)
1692{
1693 pci_power_t target_state = PCI_D3hot;
1694
1695 if (platform_pci_power_manageable(dev)) {
1696 /*
1697 * Call the platform to choose the target state of the device
1698 * and enable wake-up from this state if supported.
1699 */
1700 pci_power_t state = platform_pci_choose_state(dev);
1701
1702 switch (state) {
1703 case PCI_POWER_ERROR:
1704 case PCI_UNKNOWN:
1705 break;
1706 case PCI_D1:
1707 case PCI_D2:
1708 if (pci_no_d1d2(dev))
1709 break;
1710 default:
1711 target_state = state;
1712 }
1713 } else if (!dev->pm_cap) {
1714 target_state = PCI_D0;
1715 } else if (device_may_wakeup(&dev->dev)) {
1716 /*
1717 * Find the deepest state from which the device can generate
1718 * wake-up events, make it the target state and enable device
1719 * to generate PME#.
1720 */
1721 if (dev->pme_support) {
1722 while (target_state
1723 && !(dev->pme_support & (1 << target_state)))
1724 target_state--;
1725 }
1726 }
1727
1728 return target_state;
1729}
1730
1731/**
1732 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
1733 * @dev: Device to handle.
1734 *
1735 * Choose the power state appropriate for the device depending on whether
1736 * it can wake up the system and/or is power manageable by the platform
1737 * (PCI_D3hot is the default) and put the device into that state.
1738 */
1739int pci_prepare_to_sleep(struct pci_dev *dev)
1740{
1741 pci_power_t target_state = pci_target_state(dev);
1742 int error;
1743
1744 if (target_state == PCI_POWER_ERROR)
1745 return -EIO;
1746
1747 pci_enable_wake(dev, target_state, device_may_wakeup(&dev->dev));
1748
1749 error = pci_set_power_state(dev, target_state);
1750
1751 if (error)
1752 pci_enable_wake(dev, target_state, false);
1753
1754 return error;
1755}
1756
1757/**
1758 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
1759 * @dev: Device to handle.
1760 *
1761 * Disable device's system wake-up capability and put it into D0.
1762 */
1763int pci_back_from_sleep(struct pci_dev *dev)
1764{
1765 pci_enable_wake(dev, PCI_D0, false);
1766 return pci_set_power_state(dev, PCI_D0);
1767}
1768
1769/**
1770 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
1771 * @dev: PCI device being suspended.
1772 *
1773 * Prepare @dev to generate wake-up events at run time and put it into a low
1774 * power state.
1775 */
1776int pci_finish_runtime_suspend(struct pci_dev *dev)
1777{
1778 pci_power_t target_state = pci_target_state(dev);
1779 int error;
1780
1781 if (target_state == PCI_POWER_ERROR)
1782 return -EIO;
1783
1784 __pci_enable_wake(dev, target_state, true, pci_dev_run_wake(dev));
1785
1786 error = pci_set_power_state(dev, target_state);
1787
1788 if (error)
1789 __pci_enable_wake(dev, target_state, true, false);
1790
1791 return error;
1792}
1793
1794/**
1795 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
1796 * @dev: Device to check.
1797 *
1798 * Return true if the device itself is cabable of generating wake-up events
1799 * (through the platform or using the native PCIe PME) or if the device supports
1800 * PME and one of its upstream bridges can generate wake-up events.
1801 */
1802bool pci_dev_run_wake(struct pci_dev *dev)
1803{
1804 struct pci_bus *bus = dev->bus;
1805
1806 if (device_run_wake(&dev->dev))
1807 return true;
1808
1809 if (!dev->pme_support)
1810 return false;
1811
1812 while (bus->parent) {
1813 struct pci_dev *bridge = bus->self;
1814
1815 if (device_run_wake(&bridge->dev))
1816 return true;
1817
1818 bus = bus->parent;
1819 }
1820
1821 /* We have reached the root bus. */
1822 if (bus->bridge)
1823 return device_run_wake(bus->bridge);
1824
1825 return false;
1826}
1827EXPORT_SYMBOL_GPL(pci_dev_run_wake);
1828
1829/**
1830 * pci_pm_init - Initialize PM functions of given PCI device
1831 * @dev: PCI device to handle.
1832 */
1833void pci_pm_init(struct pci_dev *dev)
1834{
1835 int pm;
1836 u16 pmc;
1837
1838 pm_runtime_forbid(&dev->dev);
1839 device_enable_async_suspend(&dev->dev);
1840 dev->wakeup_prepared = false;
1841
1842 dev->pm_cap = 0;
1843
1844 /* find PCI PM capability in list */
1845 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
1846 if (!pm)
1847 return;
1848 /* Check device's ability to generate PME# */
1849 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
1850
1851 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
1852 dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
1853 pmc & PCI_PM_CAP_VER_MASK);
1854 return;
1855 }
1856
1857 dev->pm_cap = pm;
1858 dev->d3_delay = PCI_PM_D3_WAIT;
1859
1860 dev->d1_support = false;
1861 dev->d2_support = false;
1862 if (!pci_no_d1d2(dev)) {
1863 if (pmc & PCI_PM_CAP_D1)
1864 dev->d1_support = true;
1865 if (pmc & PCI_PM_CAP_D2)
1866 dev->d2_support = true;
1867
1868 if (dev->d1_support || dev->d2_support)
1869 dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
1870 dev->d1_support ? " D1" : "",
1871 dev->d2_support ? " D2" : "");
1872 }
1873
1874 pmc &= PCI_PM_CAP_PME_MASK;
1875 if (pmc) {
1876 dev_printk(KERN_DEBUG, &dev->dev,
1877 "PME# supported from%s%s%s%s%s\n",
1878 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
1879 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
1880 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
1881 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
1882 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
1883 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
1884 dev->pme_poll = true;
1885 /*
1886 * Make device's PM flags reflect the wake-up capability, but
1887 * let the user space enable it to wake up the system as needed.
1888 */
1889 device_set_wakeup_capable(&dev->dev, true);
1890 /* Disable the PME# generation functionality */
1891 pci_pme_active(dev, false);
1892 } else {
1893 dev->pme_support = 0;
1894 }
1895}
1896
1897/**
1898 * platform_pci_wakeup_init - init platform wakeup if present
1899 * @dev: PCI device
1900 *
1901 * Some devices don't have PCI PM caps but can still generate wakeup
1902 * events through platform methods (like ACPI events). If @dev supports
1903 * platform wakeup events, set the device flag to indicate as much. This
1904 * may be redundant if the device also supports PCI PM caps, but double
1905 * initialization should be safe in that case.
1906 */
1907void platform_pci_wakeup_init(struct pci_dev *dev)
1908{
1909 if (!platform_pci_can_wakeup(dev))
1910 return;
1911
1912 device_set_wakeup_capable(&dev->dev, true);
1913 platform_pci_sleep_wake(dev, false);
1914}
1915
1916static void pci_add_saved_cap(struct pci_dev *pci_dev,
1917 struct pci_cap_saved_state *new_cap)
1918{
1919 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
1920}
1921
1922/**
1923 * pci_add_save_buffer - allocate buffer for saving given capability registers
1924 * @dev: the PCI device
1925 * @cap: the capability to allocate the buffer for
1926 * @size: requested size of the buffer
1927 */
1928static int pci_add_cap_save_buffer(
1929 struct pci_dev *dev, char cap, unsigned int size)
1930{
1931 int pos;
1932 struct pci_cap_saved_state *save_state;
1933
1934 pos = pci_find_capability(dev, cap);
1935 if (pos <= 0)
1936 return 0;
1937
1938 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
1939 if (!save_state)
1940 return -ENOMEM;
1941
1942 save_state->cap.cap_nr = cap;
1943 save_state->cap.size = size;
1944 pci_add_saved_cap(dev, save_state);
1945
1946 return 0;
1947}
1948
1949/**
1950 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
1951 * @dev: the PCI device
1952 */
1953void pci_allocate_cap_save_buffers(struct pci_dev *dev)
1954{
1955 int error;
1956
1957 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
1958 PCI_EXP_SAVE_REGS * sizeof(u16));
1959 if (error)
1960 dev_err(&dev->dev,
1961 "unable to preallocate PCI Express save buffer\n");
1962
1963 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
1964 if (error)
1965 dev_err(&dev->dev,
1966 "unable to preallocate PCI-X save buffer\n");
1967}
1968
1969void pci_free_cap_save_buffers(struct pci_dev *dev)
1970{
1971 struct pci_cap_saved_state *tmp;
1972 struct hlist_node *pos, *n;
1973
1974 hlist_for_each_entry_safe(tmp, pos, n, &dev->saved_cap_space, next)
1975 kfree(tmp);
1976}
1977
1978/**
1979 * pci_enable_ari - enable ARI forwarding if hardware support it
1980 * @dev: the PCI device
1981 */
1982void pci_enable_ari(struct pci_dev *dev)
1983{
1984 int pos;
1985 u32 cap;
1986 u16 flags, ctrl;
1987 struct pci_dev *bridge;
1988
1989 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
1990 return;
1991
1992 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
1993 if (!pos)
1994 return;
1995
1996 bridge = dev->bus->self;
1997 if (!bridge || !pci_is_pcie(bridge))
1998 return;
1999
2000 pos = pci_pcie_cap(bridge);
2001 if (!pos)
2002 return;
2003
2004 /* ARI is a PCIe v2 feature */
2005 pci_read_config_word(bridge, pos + PCI_EXP_FLAGS, &flags);
2006 if ((flags & PCI_EXP_FLAGS_VERS) < 2)
2007 return;
2008
2009 pci_read_config_dword(bridge, pos + PCI_EXP_DEVCAP2, &cap);
2010 if (!(cap & PCI_EXP_DEVCAP2_ARI))
2011 return;
2012
2013 pci_read_config_word(bridge, pos + PCI_EXP_DEVCTL2, &ctrl);
2014 ctrl |= PCI_EXP_DEVCTL2_ARI;
2015 pci_write_config_word(bridge, pos + PCI_EXP_DEVCTL2, ctrl);
2016
2017 bridge->ari_enabled = 1;
2018}
2019
2020/**
2021 * pci_enable_ido - enable ID-based ordering on a device
2022 * @dev: the PCI device
2023 * @type: which types of IDO to enable
2024 *
2025 * Enable ID-based ordering on @dev. @type can contain the bits
2026 * %PCI_EXP_IDO_REQUEST and/or %PCI_EXP_IDO_COMPLETION to indicate
2027 * which types of transactions are allowed to be re-ordered.
2028 */
2029void pci_enable_ido(struct pci_dev *dev, unsigned long type)
2030{
2031 int pos;
2032 u16 ctrl;
2033
2034 pos = pci_pcie_cap(dev);
2035 if (!pos)
2036 return;
2037
2038 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2039 if (type & PCI_EXP_IDO_REQUEST)
2040 ctrl |= PCI_EXP_IDO_REQ_EN;
2041 if (type & PCI_EXP_IDO_COMPLETION)
2042 ctrl |= PCI_EXP_IDO_CMP_EN;
2043 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2044}
2045EXPORT_SYMBOL(pci_enable_ido);
2046
2047/**
2048 * pci_disable_ido - disable ID-based ordering on a device
2049 * @dev: the PCI device
2050 * @type: which types of IDO to disable
2051 */
2052void pci_disable_ido(struct pci_dev *dev, unsigned long type)
2053{
2054 int pos;
2055 u16 ctrl;
2056
2057 if (!pci_is_pcie(dev))
2058 return;
2059
2060 pos = pci_pcie_cap(dev);
2061 if (!pos)
2062 return;
2063
2064 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2065 if (type & PCI_EXP_IDO_REQUEST)
2066 ctrl &= ~PCI_EXP_IDO_REQ_EN;
2067 if (type & PCI_EXP_IDO_COMPLETION)
2068 ctrl &= ~PCI_EXP_IDO_CMP_EN;
2069 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2070}
2071EXPORT_SYMBOL(pci_disable_ido);
2072
2073/**
2074 * pci_enable_obff - enable optimized buffer flush/fill
2075 * @dev: PCI device
2076 * @type: type of signaling to use
2077 *
2078 * Try to enable @type OBFF signaling on @dev. It will try using WAKE#
2079 * signaling if possible, falling back to message signaling only if
2080 * WAKE# isn't supported. @type should indicate whether the PCIe link
2081 * be brought out of L0s or L1 to send the message. It should be either
2082 * %PCI_EXP_OBFF_SIGNAL_ALWAYS or %PCI_OBFF_SIGNAL_L0.
2083 *
2084 * If your device can benefit from receiving all messages, even at the
2085 * power cost of bringing the link back up from a low power state, use
2086 * %PCI_EXP_OBFF_SIGNAL_ALWAYS. Otherwise, use %PCI_OBFF_SIGNAL_L0 (the
2087 * preferred type).
2088 *
2089 * RETURNS:
2090 * Zero on success, appropriate error number on failure.
2091 */
2092int pci_enable_obff(struct pci_dev *dev, enum pci_obff_signal_type type)
2093{
2094 int pos;
2095 u32 cap;
2096 u16 ctrl;
2097 int ret;
2098
2099 if (!pci_is_pcie(dev))
2100 return -ENOTSUPP;
2101
2102 pos = pci_pcie_cap(dev);
2103 if (!pos)
2104 return -ENOTSUPP;
2105
2106 pci_read_config_dword(dev, pos + PCI_EXP_DEVCAP2, &cap);
2107 if (!(cap & PCI_EXP_OBFF_MASK))
2108 return -ENOTSUPP; /* no OBFF support at all */
2109
2110 /* Make sure the topology supports OBFF as well */
2111 if (dev->bus) {
2112 ret = pci_enable_obff(dev->bus->self, type);
2113 if (ret)
2114 return ret;
2115 }
2116
2117 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2118 if (cap & PCI_EXP_OBFF_WAKE)
2119 ctrl |= PCI_EXP_OBFF_WAKE_EN;
2120 else {
2121 switch (type) {
2122 case PCI_EXP_OBFF_SIGNAL_L0:
2123 if (!(ctrl & PCI_EXP_OBFF_WAKE_EN))
2124 ctrl |= PCI_EXP_OBFF_MSGA_EN;
2125 break;
2126 case PCI_EXP_OBFF_SIGNAL_ALWAYS:
2127 ctrl &= ~PCI_EXP_OBFF_WAKE_EN;
2128 ctrl |= PCI_EXP_OBFF_MSGB_EN;
2129 break;
2130 default:
2131 WARN(1, "bad OBFF signal type\n");
2132 return -ENOTSUPP;
2133 }
2134 }
2135 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2136
2137 return 0;
2138}
2139EXPORT_SYMBOL(pci_enable_obff);
2140
2141/**
2142 * pci_disable_obff - disable optimized buffer flush/fill
2143 * @dev: PCI device
2144 *
2145 * Disable OBFF on @dev.
2146 */
2147void pci_disable_obff(struct pci_dev *dev)
2148{
2149 int pos;
2150 u16 ctrl;
2151
2152 if (!pci_is_pcie(dev))
2153 return;
2154
2155 pos = pci_pcie_cap(dev);
2156 if (!pos)
2157 return;
2158
2159 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2160 ctrl &= ~PCI_EXP_OBFF_WAKE_EN;
2161 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2162}
2163EXPORT_SYMBOL(pci_disable_obff);
2164
2165/**
2166 * pci_ltr_supported - check whether a device supports LTR
2167 * @dev: PCI device
2168 *
2169 * RETURNS:
2170 * True if @dev supports latency tolerance reporting, false otherwise.
2171 */
2172bool pci_ltr_supported(struct pci_dev *dev)
2173{
2174 int pos;
2175 u32 cap;
2176
2177 if (!pci_is_pcie(dev))
2178 return false;
2179
2180 pos = pci_pcie_cap(dev);
2181 if (!pos)
2182 return false;
2183
2184 pci_read_config_dword(dev, pos + PCI_EXP_DEVCAP2, &cap);
2185
2186 return cap & PCI_EXP_DEVCAP2_LTR;
2187}
2188EXPORT_SYMBOL(pci_ltr_supported);
2189
2190/**
2191 * pci_enable_ltr - enable latency tolerance reporting
2192 * @dev: PCI device
2193 *
2194 * Enable LTR on @dev if possible, which means enabling it first on
2195 * upstream ports.
2196 *
2197 * RETURNS:
2198 * Zero on success, errno on failure.
2199 */
2200int pci_enable_ltr(struct pci_dev *dev)
2201{
2202 int pos;
2203 u16 ctrl;
2204 int ret;
2205
2206 if (!pci_ltr_supported(dev))
2207 return -ENOTSUPP;
2208
2209 pos = pci_pcie_cap(dev);
2210 if (!pos)
2211 return -ENOTSUPP;
2212
2213 /* Only primary function can enable/disable LTR */
2214 if (PCI_FUNC(dev->devfn) != 0)
2215 return -EINVAL;
2216
2217 /* Enable upstream ports first */
2218 if (dev->bus) {
2219 ret = pci_enable_ltr(dev->bus->self);
2220 if (ret)
2221 return ret;
2222 }
2223
2224 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2225 ctrl |= PCI_EXP_LTR_EN;
2226 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2227
2228 return 0;
2229}
2230EXPORT_SYMBOL(pci_enable_ltr);
2231
2232/**
2233 * pci_disable_ltr - disable latency tolerance reporting
2234 * @dev: PCI device
2235 */
2236void pci_disable_ltr(struct pci_dev *dev)
2237{
2238 int pos;
2239 u16 ctrl;
2240
2241 if (!pci_ltr_supported(dev))
2242 return;
2243
2244 pos = pci_pcie_cap(dev);
2245 if (!pos)
2246 return;
2247
2248 /* Only primary function can enable/disable LTR */
2249 if (PCI_FUNC(dev->devfn) != 0)
2250 return;
2251
2252 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL2, &ctrl);
2253 ctrl &= ~PCI_EXP_LTR_EN;
2254 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL2, ctrl);
2255}
2256EXPORT_SYMBOL(pci_disable_ltr);
2257
2258static int __pci_ltr_scale(int *val)
2259{
2260 int scale = 0;
2261
2262 while (*val > 1023) {
2263 *val = (*val + 31) / 32;
2264 scale++;
2265 }
2266 return scale;
2267}
2268
2269/**
2270 * pci_set_ltr - set LTR latency values
2271 * @dev: PCI device
2272 * @snoop_lat_ns: snoop latency in nanoseconds
2273 * @nosnoop_lat_ns: nosnoop latency in nanoseconds
2274 *
2275 * Figure out the scale and set the LTR values accordingly.
2276 */
2277int pci_set_ltr(struct pci_dev *dev, int snoop_lat_ns, int nosnoop_lat_ns)
2278{
2279 int pos, ret, snoop_scale, nosnoop_scale;
2280 u16 val;
2281
2282 if (!pci_ltr_supported(dev))
2283 return -ENOTSUPP;
2284
2285 snoop_scale = __pci_ltr_scale(&snoop_lat_ns);
2286 nosnoop_scale = __pci_ltr_scale(&nosnoop_lat_ns);
2287
2288 if (snoop_lat_ns > PCI_LTR_VALUE_MASK ||
2289 nosnoop_lat_ns > PCI_LTR_VALUE_MASK)
2290 return -EINVAL;
2291
2292 if ((snoop_scale > (PCI_LTR_SCALE_MASK >> PCI_LTR_SCALE_SHIFT)) ||
2293 (nosnoop_scale > (PCI_LTR_SCALE_MASK >> PCI_LTR_SCALE_SHIFT)))
2294 return -EINVAL;
2295
2296 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_LTR);
2297 if (!pos)
2298 return -ENOTSUPP;
2299
2300 val = (snoop_scale << PCI_LTR_SCALE_SHIFT) | snoop_lat_ns;
2301 ret = pci_write_config_word(dev, pos + PCI_LTR_MAX_SNOOP_LAT, val);
2302 if (ret != 4)
2303 return -EIO;
2304
2305 val = (nosnoop_scale << PCI_LTR_SCALE_SHIFT) | nosnoop_lat_ns;
2306 ret = pci_write_config_word(dev, pos + PCI_LTR_MAX_NOSNOOP_LAT, val);
2307 if (ret != 4)
2308 return -EIO;
2309
2310 return 0;
2311}
2312EXPORT_SYMBOL(pci_set_ltr);
2313
2314static int pci_acs_enable;
2315
2316/**
2317 * pci_request_acs - ask for ACS to be enabled if supported
2318 */
2319void pci_request_acs(void)
2320{
2321 pci_acs_enable = 1;
2322}
2323
2324/**
2325 * pci_enable_acs - enable ACS if hardware support it
2326 * @dev: the PCI device
2327 */
2328void pci_enable_acs(struct pci_dev *dev)
2329{
2330 int pos;
2331 u16 cap;
2332 u16 ctrl;
2333
2334 if (!pci_acs_enable)
2335 return;
2336
2337 if (!pci_is_pcie(dev))
2338 return;
2339
2340 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2341 if (!pos)
2342 return;
2343
2344 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2345 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2346
2347 /* Source Validation */
2348 ctrl |= (cap & PCI_ACS_SV);
2349
2350 /* P2P Request Redirect */
2351 ctrl |= (cap & PCI_ACS_RR);
2352
2353 /* P2P Completion Redirect */
2354 ctrl |= (cap & PCI_ACS_CR);
2355
2356 /* Upstream Forwarding */
2357 ctrl |= (cap & PCI_ACS_UF);
2358
2359 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2360}
2361
2362/**
2363 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2364 * @dev: the PCI device
2365 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2366 *
2367 * Perform INTx swizzling for a device behind one level of bridge. This is
2368 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2369 * behind bridges on add-in cards. For devices with ARI enabled, the slot
2370 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2371 * the PCI Express Base Specification, Revision 2.1)
2372 */
2373u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2374{
2375 int slot;
2376
2377 if (pci_ari_enabled(dev->bus))
2378 slot = 0;
2379 else
2380 slot = PCI_SLOT(dev->devfn);
2381
2382 return (((pin - 1) + slot) % 4) + 1;
2383}
2384
2385int
2386pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
2387{
2388 u8 pin;
2389
2390 pin = dev->pin;
2391 if (!pin)
2392 return -1;
2393
2394 while (!pci_is_root_bus(dev->bus)) {
2395 pin = pci_swizzle_interrupt_pin(dev, pin);
2396 dev = dev->bus->self;
2397 }
2398 *bridge = dev;
2399 return pin;
2400}
2401
2402/**
2403 * pci_common_swizzle - swizzle INTx all the way to root bridge
2404 * @dev: the PCI device
2405 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2406 *
2407 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
2408 * bridges all the way up to a PCI root bus.
2409 */
2410u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
2411{
2412 u8 pin = *pinp;
2413
2414 while (!pci_is_root_bus(dev->bus)) {
2415 pin = pci_swizzle_interrupt_pin(dev, pin);
2416 dev = dev->bus->self;
2417 }
2418 *pinp = pin;
2419 return PCI_SLOT(dev->devfn);
2420}
2421
2422/**
2423 * pci_release_region - Release a PCI bar
2424 * @pdev: PCI device whose resources were previously reserved by pci_request_region
2425 * @bar: BAR to release
2426 *
2427 * Releases the PCI I/O and memory resources previously reserved by a
2428 * successful call to pci_request_region. Call this function only
2429 * after all use of the PCI regions has ceased.
2430 */
2431void pci_release_region(struct pci_dev *pdev, int bar)
2432{
2433 struct pci_devres *dr;
2434
2435 if (pci_resource_len(pdev, bar) == 0)
2436 return;
2437 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
2438 release_region(pci_resource_start(pdev, bar),
2439 pci_resource_len(pdev, bar));
2440 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
2441 release_mem_region(pci_resource_start(pdev, bar),
2442 pci_resource_len(pdev, bar));
2443
2444 dr = find_pci_dr(pdev);
2445 if (dr)
2446 dr->region_mask &= ~(1 << bar);
2447}
2448
2449/**
2450 * __pci_request_region - Reserved PCI I/O and memory resource
2451 * @pdev: PCI device whose resources are to be reserved
2452 * @bar: BAR to be reserved
2453 * @res_name: Name to be associated with resource.
2454 * @exclusive: whether the region access is exclusive or not
2455 *
2456 * Mark the PCI region associated with PCI device @pdev BR @bar as
2457 * being reserved by owner @res_name. Do not access any
2458 * address inside the PCI regions unless this call returns
2459 * successfully.
2460 *
2461 * If @exclusive is set, then the region is marked so that userspace
2462 * is explicitly not allowed to map the resource via /dev/mem or
2463 * sysfs MMIO access.
2464 *
2465 * Returns 0 on success, or %EBUSY on error. A warning
2466 * message is also printed on failure.
2467 */
2468static int __pci_request_region(struct pci_dev *pdev, int bar, const char *res_name,
2469 int exclusive)
2470{
2471 struct pci_devres *dr;
2472
2473 if (pci_resource_len(pdev, bar) == 0)
2474 return 0;
2475
2476 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
2477 if (!request_region(pci_resource_start(pdev, bar),
2478 pci_resource_len(pdev, bar), res_name))
2479 goto err_out;
2480 }
2481 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
2482 if (!__request_mem_region(pci_resource_start(pdev, bar),
2483 pci_resource_len(pdev, bar), res_name,
2484 exclusive))
2485 goto err_out;
2486 }
2487
2488 dr = find_pci_dr(pdev);
2489 if (dr)
2490 dr->region_mask |= 1 << bar;
2491
2492 return 0;
2493
2494err_out:
2495 dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
2496 &pdev->resource[bar]);
2497 return -EBUSY;
2498}
2499
2500/**
2501 * pci_request_region - Reserve PCI I/O and memory resource
2502 * @pdev: PCI device whose resources are to be reserved
2503 * @bar: BAR to be reserved
2504 * @res_name: Name to be associated with resource
2505 *
2506 * Mark the PCI region associated with PCI device @pdev BAR @bar as
2507 * being reserved by owner @res_name. Do not access any
2508 * address inside the PCI regions unless this call returns
2509 * successfully.
2510 *
2511 * Returns 0 on success, or %EBUSY on error. A warning
2512 * message is also printed on failure.
2513 */
2514int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
2515{
2516 return __pci_request_region(pdev, bar, res_name, 0);
2517}
2518
2519/**
2520 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
2521 * @pdev: PCI device whose resources are to be reserved
2522 * @bar: BAR to be reserved
2523 * @res_name: Name to be associated with resource.
2524 *
2525 * Mark the PCI region associated with PCI device @pdev BR @bar as
2526 * being reserved by owner @res_name. Do not access any
2527 * address inside the PCI regions unless this call returns
2528 * successfully.
2529 *
2530 * Returns 0 on success, or %EBUSY on error. A warning
2531 * message is also printed on failure.
2532 *
2533 * The key difference that _exclusive makes it that userspace is
2534 * explicitly not allowed to map the resource via /dev/mem or
2535 * sysfs.
2536 */
2537int pci_request_region_exclusive(struct pci_dev *pdev, int bar, const char *res_name)
2538{
2539 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
2540}
2541/**
2542 * pci_release_selected_regions - Release selected PCI I/O and memory resources
2543 * @pdev: PCI device whose resources were previously reserved
2544 * @bars: Bitmask of BARs to be released
2545 *
2546 * Release selected PCI I/O and memory resources previously reserved.
2547 * Call this function only after all use of the PCI regions has ceased.
2548 */
2549void pci_release_selected_regions(struct pci_dev *pdev, int bars)
2550{
2551 int i;
2552
2553 for (i = 0; i < 6; i++)
2554 if (bars & (1 << i))
2555 pci_release_region(pdev, i);
2556}
2557
2558int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
2559 const char *res_name, int excl)
2560{
2561 int i;
2562
2563 for (i = 0; i < 6; i++)
2564 if (bars & (1 << i))
2565 if (__pci_request_region(pdev, i, res_name, excl))
2566 goto err_out;
2567 return 0;
2568
2569err_out:
2570 while(--i >= 0)
2571 if (bars & (1 << i))
2572 pci_release_region(pdev, i);
2573
2574 return -EBUSY;
2575}
2576
2577
2578/**
2579 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
2580 * @pdev: PCI device whose resources are to be reserved
2581 * @bars: Bitmask of BARs to be requested
2582 * @res_name: Name to be associated with resource
2583 */
2584int pci_request_selected_regions(struct pci_dev *pdev, int bars,
2585 const char *res_name)
2586{
2587 return __pci_request_selected_regions(pdev, bars, res_name, 0);
2588}
2589
2590int pci_request_selected_regions_exclusive(struct pci_dev *pdev,
2591 int bars, const char *res_name)
2592{
2593 return __pci_request_selected_regions(pdev, bars, res_name,
2594 IORESOURCE_EXCLUSIVE);
2595}
2596
2597/**
2598 * pci_release_regions - Release reserved PCI I/O and memory resources
2599 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
2600 *
2601 * Releases all PCI I/O and memory resources previously reserved by a
2602 * successful call to pci_request_regions. Call this function only
2603 * after all use of the PCI regions has ceased.
2604 */
2605
2606void pci_release_regions(struct pci_dev *pdev)
2607{
2608 pci_release_selected_regions(pdev, (1 << 6) - 1);
2609}
2610
2611/**
2612 * pci_request_regions - Reserved PCI I/O and memory resources
2613 * @pdev: PCI device whose resources are to be reserved
2614 * @res_name: Name to be associated with resource.
2615 *
2616 * Mark all PCI regions associated with PCI device @pdev as
2617 * being reserved by owner @res_name. Do not access any
2618 * address inside the PCI regions unless this call returns
2619 * successfully.
2620 *
2621 * Returns 0 on success, or %EBUSY on error. A warning
2622 * message is also printed on failure.
2623 */
2624int pci_request_regions(struct pci_dev *pdev, const char *res_name)
2625{
2626 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
2627}
2628
2629/**
2630 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
2631 * @pdev: PCI device whose resources are to be reserved
2632 * @res_name: Name to be associated with resource.
2633 *
2634 * Mark all PCI regions associated with PCI device @pdev as
2635 * being reserved by owner @res_name. Do not access any
2636 * address inside the PCI regions unless this call returns
2637 * successfully.
2638 *
2639 * pci_request_regions_exclusive() will mark the region so that
2640 * /dev/mem and the sysfs MMIO access will not be allowed.
2641 *
2642 * Returns 0 on success, or %EBUSY on error. A warning
2643 * message is also printed on failure.
2644 */
2645int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
2646{
2647 return pci_request_selected_regions_exclusive(pdev,
2648 ((1 << 6) - 1), res_name);
2649}
2650
2651static void __pci_set_master(struct pci_dev *dev, bool enable)
2652{
2653 u16 old_cmd, cmd;
2654
2655 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
2656 if (enable)
2657 cmd = old_cmd | PCI_COMMAND_MASTER;
2658 else
2659 cmd = old_cmd & ~PCI_COMMAND_MASTER;
2660 if (cmd != old_cmd) {
2661 dev_dbg(&dev->dev, "%s bus mastering\n",
2662 enable ? "enabling" : "disabling");
2663 pci_write_config_word(dev, PCI_COMMAND, cmd);
2664 }
2665 dev->is_busmaster = enable;
2666}
2667
2668/**
2669 * pcibios_set_master - enable PCI bus-mastering for device dev
2670 * @dev: the PCI device to enable
2671 *
2672 * Enables PCI bus-mastering for the device. This is the default
2673 * implementation. Architecture specific implementations can override
2674 * this if necessary.
2675 */
2676void __weak pcibios_set_master(struct pci_dev *dev)
2677{
2678 u8 lat;
2679
2680 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
2681 if (pci_is_pcie(dev))
2682 return;
2683
2684 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
2685 if (lat < 16)
2686 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
2687 else if (lat > pcibios_max_latency)
2688 lat = pcibios_max_latency;
2689 else
2690 return;
2691 dev_printk(KERN_DEBUG, &dev->dev, "setting latency timer to %d\n", lat);
2692 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
2693}
2694
2695/**
2696 * pci_set_master - enables bus-mastering for device dev
2697 * @dev: the PCI device to enable
2698 *
2699 * Enables bus-mastering on the device and calls pcibios_set_master()
2700 * to do the needed arch specific settings.
2701 */
2702void pci_set_master(struct pci_dev *dev)
2703{
2704 __pci_set_master(dev, true);
2705 pcibios_set_master(dev);
2706}
2707
2708/**
2709 * pci_clear_master - disables bus-mastering for device dev
2710 * @dev: the PCI device to disable
2711 */
2712void pci_clear_master(struct pci_dev *dev)
2713{
2714 __pci_set_master(dev, false);
2715}
2716
2717/**
2718 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
2719 * @dev: the PCI device for which MWI is to be enabled
2720 *
2721 * Helper function for pci_set_mwi.
2722 * Originally copied from drivers/net/acenic.c.
2723 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
2724 *
2725 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
2726 */
2727int pci_set_cacheline_size(struct pci_dev *dev)
2728{
2729 u8 cacheline_size;
2730
2731 if (!pci_cache_line_size)
2732 return -EINVAL;
2733
2734 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
2735 equal to or multiple of the right value. */
2736 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
2737 if (cacheline_size >= pci_cache_line_size &&
2738 (cacheline_size % pci_cache_line_size) == 0)
2739 return 0;
2740
2741 /* Write the correct value. */
2742 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
2743 /* Read it back. */
2744 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
2745 if (cacheline_size == pci_cache_line_size)
2746 return 0;
2747
2748 dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not "
2749 "supported\n", pci_cache_line_size << 2);
2750
2751 return -EINVAL;
2752}
2753EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
2754
2755#ifdef PCI_DISABLE_MWI
2756int pci_set_mwi(struct pci_dev *dev)
2757{
2758 return 0;
2759}
2760
2761int pci_try_set_mwi(struct pci_dev *dev)
2762{
2763 return 0;
2764}
2765
2766void pci_clear_mwi(struct pci_dev *dev)
2767{
2768}
2769
2770#else
2771
2772/**
2773 * pci_set_mwi - enables memory-write-invalidate PCI transaction
2774 * @dev: the PCI device for which MWI is enabled
2775 *
2776 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
2777 *
2778 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
2779 */
2780int
2781pci_set_mwi(struct pci_dev *dev)
2782{
2783 int rc;
2784 u16 cmd;
2785
2786 rc = pci_set_cacheline_size(dev);
2787 if (rc)
2788 return rc;
2789
2790 pci_read_config_word(dev, PCI_COMMAND, &cmd);
2791 if (! (cmd & PCI_COMMAND_INVALIDATE)) {
2792 dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
2793 cmd |= PCI_COMMAND_INVALIDATE;
2794 pci_write_config_word(dev, PCI_COMMAND, cmd);
2795 }
2796
2797 return 0;
2798}
2799
2800/**
2801 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
2802 * @dev: the PCI device for which MWI is enabled
2803 *
2804 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
2805 * Callers are not required to check the return value.
2806 *
2807 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
2808 */
2809int pci_try_set_mwi(struct pci_dev *dev)
2810{
2811 int rc = pci_set_mwi(dev);
2812 return rc;
2813}
2814
2815/**
2816 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
2817 * @dev: the PCI device to disable
2818 *
2819 * Disables PCI Memory-Write-Invalidate transaction on the device
2820 */
2821void
2822pci_clear_mwi(struct pci_dev *dev)
2823{
2824 u16 cmd;
2825
2826 pci_read_config_word(dev, PCI_COMMAND, &cmd);
2827 if (cmd & PCI_COMMAND_INVALIDATE) {
2828 cmd &= ~PCI_COMMAND_INVALIDATE;
2829 pci_write_config_word(dev, PCI_COMMAND, cmd);
2830 }
2831}
2832#endif /* ! PCI_DISABLE_MWI */
2833
2834/**
2835 * pci_intx - enables/disables PCI INTx for device dev
2836 * @pdev: the PCI device to operate on
2837 * @enable: boolean: whether to enable or disable PCI INTx
2838 *
2839 * Enables/disables PCI INTx for device dev
2840 */
2841void
2842pci_intx(struct pci_dev *pdev, int enable)
2843{
2844 u16 pci_command, new;
2845
2846 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
2847
2848 if (enable) {
2849 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
2850 } else {
2851 new = pci_command | PCI_COMMAND_INTX_DISABLE;
2852 }
2853
2854 if (new != pci_command) {
2855 struct pci_devres *dr;
2856
2857 pci_write_config_word(pdev, PCI_COMMAND, new);
2858
2859 dr = find_pci_dr(pdev);
2860 if (dr && !dr->restore_intx) {
2861 dr->restore_intx = 1;
2862 dr->orig_intx = !enable;
2863 }
2864 }
2865}
2866
2867/**
2868 * pci_intx_mask_supported - probe for INTx masking support
2869 * @dev: the PCI device to operate on
2870 *
2871 * Check if the device dev support INTx masking via the config space
2872 * command word.
2873 */
2874bool pci_intx_mask_supported(struct pci_dev *dev)
2875{
2876 bool mask_supported = false;
2877 u16 orig, new;
2878
2879 pci_cfg_access_lock(dev);
2880
2881 pci_read_config_word(dev, PCI_COMMAND, &orig);
2882 pci_write_config_word(dev, PCI_COMMAND,
2883 orig ^ PCI_COMMAND_INTX_DISABLE);
2884 pci_read_config_word(dev, PCI_COMMAND, &new);
2885
2886 /*
2887 * There's no way to protect against hardware bugs or detect them
2888 * reliably, but as long as we know what the value should be, let's
2889 * go ahead and check it.
2890 */
2891 if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
2892 dev_err(&dev->dev, "Command register changed from "
2893 "0x%x to 0x%x: driver or hardware bug?\n", orig, new);
2894 } else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
2895 mask_supported = true;
2896 pci_write_config_word(dev, PCI_COMMAND, orig);
2897 }
2898
2899 pci_cfg_access_unlock(dev);
2900 return mask_supported;
2901}
2902EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
2903
2904static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
2905{
2906 struct pci_bus *bus = dev->bus;
2907 bool mask_updated = true;
2908 u32 cmd_status_dword;
2909 u16 origcmd, newcmd;
2910 unsigned long flags;
2911 bool irq_pending;
2912
2913 /*
2914 * We do a single dword read to retrieve both command and status.
2915 * Document assumptions that make this possible.
2916 */
2917 BUILD_BUG_ON(PCI_COMMAND % 4);
2918 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
2919
2920 raw_spin_lock_irqsave(&pci_lock, flags);
2921
2922 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
2923
2924 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
2925
2926 /*
2927 * Check interrupt status register to see whether our device
2928 * triggered the interrupt (when masking) or the next IRQ is
2929 * already pending (when unmasking).
2930 */
2931 if (mask != irq_pending) {
2932 mask_updated = false;
2933 goto done;
2934 }
2935
2936 origcmd = cmd_status_dword;
2937 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
2938 if (mask)
2939 newcmd |= PCI_COMMAND_INTX_DISABLE;
2940 if (newcmd != origcmd)
2941 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
2942
2943done:
2944 raw_spin_unlock_irqrestore(&pci_lock, flags);
2945
2946 return mask_updated;
2947}
2948
2949/**
2950 * pci_check_and_mask_intx - mask INTx on pending interrupt
2951 * @dev: the PCI device to operate on
2952 *
2953 * Check if the device dev has its INTx line asserted, mask it and
2954 * return true in that case. False is returned if not interrupt was
2955 * pending.
2956 */
2957bool pci_check_and_mask_intx(struct pci_dev *dev)
2958{
2959 return pci_check_and_set_intx_mask(dev, true);
2960}
2961EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
2962
2963/**
2964 * pci_check_and_mask_intx - unmask INTx of no interrupt is pending
2965 * @dev: the PCI device to operate on
2966 *
2967 * Check if the device dev has its INTx line asserted, unmask it if not
2968 * and return true. False is returned and the mask remains active if
2969 * there was still an interrupt pending.
2970 */
2971bool pci_check_and_unmask_intx(struct pci_dev *dev)
2972{
2973 return pci_check_and_set_intx_mask(dev, false);
2974}
2975EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
2976
2977/**
2978 * pci_msi_off - disables any msi or msix capabilities
2979 * @dev: the PCI device to operate on
2980 *
2981 * If you want to use msi see pci_enable_msi and friends.
2982 * This is a lower level primitive that allows us to disable
2983 * msi operation at the device level.
2984 */
2985void pci_msi_off(struct pci_dev *dev)
2986{
2987 int pos;
2988 u16 control;
2989
2990 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
2991 if (pos) {
2992 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
2993 control &= ~PCI_MSI_FLAGS_ENABLE;
2994 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
2995 }
2996 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
2997 if (pos) {
2998 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
2999 control &= ~PCI_MSIX_FLAGS_ENABLE;
3000 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
3001 }
3002}
3003EXPORT_SYMBOL_GPL(pci_msi_off);
3004
3005int pci_set_dma_max_seg_size(struct pci_dev *dev, unsigned int size)
3006{
3007 return dma_set_max_seg_size(&dev->dev, size);
3008}
3009EXPORT_SYMBOL(pci_set_dma_max_seg_size);
3010
3011int pci_set_dma_seg_boundary(struct pci_dev *dev, unsigned long mask)
3012{
3013 return dma_set_seg_boundary(&dev->dev, mask);
3014}
3015EXPORT_SYMBOL(pci_set_dma_seg_boundary);
3016
3017static int pcie_flr(struct pci_dev *dev, int probe)
3018{
3019 int i;
3020 int pos;
3021 u32 cap;
3022 u16 status, control;
3023
3024 pos = pci_pcie_cap(dev);
3025 if (!pos)
3026 return -ENOTTY;
3027
3028 pci_read_config_dword(dev, pos + PCI_EXP_DEVCAP, &cap);
3029 if (!(cap & PCI_EXP_DEVCAP_FLR))
3030 return -ENOTTY;
3031
3032 if (probe)
3033 return 0;
3034
3035 /* Wait for Transaction Pending bit clean */
3036 for (i = 0; i < 4; i++) {
3037 if (i)
3038 msleep((1 << (i - 1)) * 100);
3039
3040 pci_read_config_word(dev, pos + PCI_EXP_DEVSTA, &status);
3041 if (!(status & PCI_EXP_DEVSTA_TRPND))
3042 goto clear;
3043 }
3044
3045 dev_err(&dev->dev, "transaction is not cleared; "
3046 "proceeding with reset anyway\n");
3047
3048clear:
3049 pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &control);
3050 control |= PCI_EXP_DEVCTL_BCR_FLR;
3051 pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, control);
3052
3053 msleep(100);
3054
3055 return 0;
3056}
3057
3058static int pci_af_flr(struct pci_dev *dev, int probe)
3059{
3060 int i;
3061 int pos;
3062 u8 cap;
3063 u8 status;
3064
3065 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3066 if (!pos)
3067 return -ENOTTY;
3068
3069 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3070 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3071 return -ENOTTY;
3072
3073 if (probe)
3074 return 0;
3075
3076 /* Wait for Transaction Pending bit clean */
3077 for (i = 0; i < 4; i++) {
3078 if (i)
3079 msleep((1 << (i - 1)) * 100);
3080
3081 pci_read_config_byte(dev, pos + PCI_AF_STATUS, &status);
3082 if (!(status & PCI_AF_STATUS_TP))
3083 goto clear;
3084 }
3085
3086 dev_err(&dev->dev, "transaction is not cleared; "
3087 "proceeding with reset anyway\n");
3088
3089clear:
3090 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3091 msleep(100);
3092
3093 return 0;
3094}
3095
3096/**
3097 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3098 * @dev: Device to reset.
3099 * @probe: If set, only check if the device can be reset this way.
3100 *
3101 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3102 * unset, it will be reinitialized internally when going from PCI_D3hot to
3103 * PCI_D0. If that's the case and the device is not in a low-power state
3104 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3105 *
3106 * NOTE: This causes the caller to sleep for twice the device power transition
3107 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3108 * by devault (i.e. unless the @dev's d3_delay field has a different value).
3109 * Moreover, only devices in D0 can be reset by this function.
3110 */
3111static int pci_pm_reset(struct pci_dev *dev, int probe)
3112{
3113 u16 csr;
3114
3115 if (!dev->pm_cap)
3116 return -ENOTTY;
3117
3118 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3119 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3120 return -ENOTTY;
3121
3122 if (probe)
3123 return 0;
3124
3125 if (dev->current_state != PCI_D0)
3126 return -EINVAL;
3127
3128 csr &= ~PCI_PM_CTRL_STATE_MASK;
3129 csr |= PCI_D3hot;
3130 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3131 pci_dev_d3_sleep(dev);
3132
3133 csr &= ~PCI_PM_CTRL_STATE_MASK;
3134 csr |= PCI_D0;
3135 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3136 pci_dev_d3_sleep(dev);
3137
3138 return 0;
3139}
3140
3141static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
3142{
3143 u16 ctrl;
3144 struct pci_dev *pdev;
3145
3146 if (pci_is_root_bus(dev->bus) || dev->subordinate || !dev->bus->self)
3147 return -ENOTTY;
3148
3149 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3150 if (pdev != dev)
3151 return -ENOTTY;
3152
3153 if (probe)
3154 return 0;
3155
3156 pci_read_config_word(dev->bus->self, PCI_BRIDGE_CONTROL, &ctrl);
3157 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
3158 pci_write_config_word(dev->bus->self, PCI_BRIDGE_CONTROL, ctrl);
3159 msleep(100);
3160
3161 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
3162 pci_write_config_word(dev->bus->self, PCI_BRIDGE_CONTROL, ctrl);
3163 msleep(100);
3164
3165 return 0;
3166}
3167
3168static int __pci_dev_reset(struct pci_dev *dev, int probe)
3169{
3170 int rc;
3171
3172 might_sleep();
3173
3174 rc = pci_dev_specific_reset(dev, probe);
3175 if (rc != -ENOTTY)
3176 goto done;
3177
3178 rc = pcie_flr(dev, probe);
3179 if (rc != -ENOTTY)
3180 goto done;
3181
3182 rc = pci_af_flr(dev, probe);
3183 if (rc != -ENOTTY)
3184 goto done;
3185
3186 rc = pci_pm_reset(dev, probe);
3187 if (rc != -ENOTTY)
3188 goto done;
3189
3190 rc = pci_parent_bus_reset(dev, probe);
3191done:
3192 return rc;
3193}
3194
3195static int pci_dev_reset(struct pci_dev *dev, int probe)
3196{
3197 int rc;
3198
3199 if (!probe) {
3200 pci_cfg_access_lock(dev);
3201 /* block PM suspend, driver probe, etc. */
3202 device_lock(&dev->dev);
3203 }
3204
3205 rc = __pci_dev_reset(dev, probe);
3206
3207 if (!probe) {
3208 device_unlock(&dev->dev);
3209 pci_cfg_access_unlock(dev);
3210 }
3211 return rc;
3212}
3213/**
3214 * __pci_reset_function - reset a PCI device function
3215 * @dev: PCI device to reset
3216 *
3217 * Some devices allow an individual function to be reset without affecting
3218 * other functions in the same device. The PCI device must be responsive
3219 * to PCI config space in order to use this function.
3220 *
3221 * The device function is presumed to be unused when this function is called.
3222 * Resetting the device will make the contents of PCI configuration space
3223 * random, so any caller of this must be prepared to reinitialise the
3224 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3225 * etc.
3226 *
3227 * Returns 0 if the device function was successfully reset or negative if the
3228 * device doesn't support resetting a single function.
3229 */
3230int __pci_reset_function(struct pci_dev *dev)
3231{
3232 return pci_dev_reset(dev, 0);
3233}
3234EXPORT_SYMBOL_GPL(__pci_reset_function);
3235
3236/**
3237 * __pci_reset_function_locked - reset a PCI device function while holding
3238 * the @dev mutex lock.
3239 * @dev: PCI device to reset
3240 *
3241 * Some devices allow an individual function to be reset without affecting
3242 * other functions in the same device. The PCI device must be responsive
3243 * to PCI config space in order to use this function.
3244 *
3245 * The device function is presumed to be unused and the caller is holding
3246 * the device mutex lock when this function is called.
3247 * Resetting the device will make the contents of PCI configuration space
3248 * random, so any caller of this must be prepared to reinitialise the
3249 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3250 * etc.
3251 *
3252 * Returns 0 if the device function was successfully reset or negative if the
3253 * device doesn't support resetting a single function.
3254 */
3255int __pci_reset_function_locked(struct pci_dev *dev)
3256{
3257 return __pci_dev_reset(dev, 0);
3258}
3259EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
3260
3261/**
3262 * pci_probe_reset_function - check whether the device can be safely reset
3263 * @dev: PCI device to reset
3264 *
3265 * Some devices allow an individual function to be reset without affecting
3266 * other functions in the same device. The PCI device must be responsive
3267 * to PCI config space in order to use this function.
3268 *
3269 * Returns 0 if the device function can be reset or negative if the
3270 * device doesn't support resetting a single function.
3271 */
3272int pci_probe_reset_function(struct pci_dev *dev)
3273{
3274 return pci_dev_reset(dev, 1);
3275}
3276
3277/**
3278 * pci_reset_function - quiesce and reset a PCI device function
3279 * @dev: PCI device to reset
3280 *
3281 * Some devices allow an individual function to be reset without affecting
3282 * other functions in the same device. The PCI device must be responsive
3283 * to PCI config space in order to use this function.
3284 *
3285 * This function does not just reset the PCI portion of a device, but
3286 * clears all the state associated with the device. This function differs
3287 * from __pci_reset_function in that it saves and restores device state
3288 * over the reset.
3289 *
3290 * Returns 0 if the device function was successfully reset or negative if the
3291 * device doesn't support resetting a single function.
3292 */
3293int pci_reset_function(struct pci_dev *dev)
3294{
3295 int rc;
3296
3297 rc = pci_dev_reset(dev, 1);
3298 if (rc)
3299 return rc;
3300
3301 pci_save_state(dev);
3302
3303 /*
3304 * both INTx and MSI are disabled after the Interrupt Disable bit
3305 * is set and the Bus Master bit is cleared.
3306 */
3307 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
3308
3309 rc = pci_dev_reset(dev, 0);
3310
3311 pci_restore_state(dev);
3312
3313 return rc;
3314}
3315EXPORT_SYMBOL_GPL(pci_reset_function);
3316
3317/**
3318 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
3319 * @dev: PCI device to query
3320 *
3321 * Returns mmrbc: maximum designed memory read count in bytes
3322 * or appropriate error value.
3323 */
3324int pcix_get_max_mmrbc(struct pci_dev *dev)
3325{
3326 int cap;
3327 u32 stat;
3328
3329 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
3330 if (!cap)
3331 return -EINVAL;
3332
3333 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
3334 return -EINVAL;
3335
3336 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
3337}
3338EXPORT_SYMBOL(pcix_get_max_mmrbc);
3339
3340/**
3341 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
3342 * @dev: PCI device to query
3343 *
3344 * Returns mmrbc: maximum memory read count in bytes
3345 * or appropriate error value.
3346 */
3347int pcix_get_mmrbc(struct pci_dev *dev)
3348{
3349 int cap;
3350 u16 cmd;
3351
3352 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
3353 if (!cap)
3354 return -EINVAL;
3355
3356 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
3357 return -EINVAL;
3358
3359 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
3360}
3361EXPORT_SYMBOL(pcix_get_mmrbc);
3362
3363/**
3364 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
3365 * @dev: PCI device to query
3366 * @mmrbc: maximum memory read count in bytes
3367 * valid values are 512, 1024, 2048, 4096
3368 *
3369 * If possible sets maximum memory read byte count, some bridges have erratas
3370 * that prevent this.
3371 */
3372int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
3373{
3374 int cap;
3375 u32 stat, v, o;
3376 u16 cmd;
3377
3378 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
3379 return -EINVAL;
3380
3381 v = ffs(mmrbc) - 10;
3382
3383 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
3384 if (!cap)
3385 return -EINVAL;
3386
3387 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
3388 return -EINVAL;
3389
3390 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
3391 return -E2BIG;
3392
3393 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
3394 return -EINVAL;
3395
3396 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
3397 if (o != v) {
3398 if (v > o && dev->bus &&
3399 (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
3400 return -EIO;
3401
3402 cmd &= ~PCI_X_CMD_MAX_READ;
3403 cmd |= v << 2;
3404 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
3405 return -EIO;
3406 }
3407 return 0;
3408}
3409EXPORT_SYMBOL(pcix_set_mmrbc);
3410
3411/**
3412 * pcie_get_readrq - get PCI Express read request size
3413 * @dev: PCI device to query
3414 *
3415 * Returns maximum memory read request in bytes
3416 * or appropriate error value.
3417 */
3418int pcie_get_readrq(struct pci_dev *dev)
3419{
3420 int ret, cap;
3421 u16 ctl;
3422
3423 cap = pci_pcie_cap(dev);
3424 if (!cap)
3425 return -EINVAL;
3426
3427 ret = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3428 if (!ret)
3429 ret = 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
3430
3431 return ret;
3432}
3433EXPORT_SYMBOL(pcie_get_readrq);
3434
3435/**
3436 * pcie_set_readrq - set PCI Express maximum memory read request
3437 * @dev: PCI device to query
3438 * @rq: maximum memory read count in bytes
3439 * valid values are 128, 256, 512, 1024, 2048, 4096
3440 *
3441 * If possible sets maximum memory read request in bytes
3442 */
3443int pcie_set_readrq(struct pci_dev *dev, int rq)
3444{
3445 int cap, err = -EINVAL;
3446 u16 ctl, v;
3447
3448 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
3449 goto out;
3450
3451 cap = pci_pcie_cap(dev);
3452 if (!cap)
3453 goto out;
3454
3455 err = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3456 if (err)
3457 goto out;
3458 /*
3459 * If using the "performance" PCIe config, we clamp the
3460 * read rq size to the max packet size to prevent the
3461 * host bridge generating requests larger than we can
3462 * cope with
3463 */
3464 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
3465 int mps = pcie_get_mps(dev);
3466
3467 if (mps < 0)
3468 return mps;
3469 if (mps < rq)
3470 rq = mps;
3471 }
3472
3473 v = (ffs(rq) - 8) << 12;
3474
3475 if ((ctl & PCI_EXP_DEVCTL_READRQ) != v) {
3476 ctl &= ~PCI_EXP_DEVCTL_READRQ;
3477 ctl |= v;
3478 err = pci_write_config_word(dev, cap + PCI_EXP_DEVCTL, ctl);
3479 }
3480
3481out:
3482 return err;
3483}
3484EXPORT_SYMBOL(pcie_set_readrq);
3485
3486/**
3487 * pcie_get_mps - get PCI Express maximum payload size
3488 * @dev: PCI device to query
3489 *
3490 * Returns maximum payload size in bytes
3491 * or appropriate error value.
3492 */
3493int pcie_get_mps(struct pci_dev *dev)
3494{
3495 int ret, cap;
3496 u16 ctl;
3497
3498 cap = pci_pcie_cap(dev);
3499 if (!cap)
3500 return -EINVAL;
3501
3502 ret = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3503 if (!ret)
3504 ret = 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
3505
3506 return ret;
3507}
3508
3509/**
3510 * pcie_set_mps - set PCI Express maximum payload size
3511 * @dev: PCI device to query
3512 * @mps: maximum payload size in bytes
3513 * valid values are 128, 256, 512, 1024, 2048, 4096
3514 *
3515 * If possible sets maximum payload size
3516 */
3517int pcie_set_mps(struct pci_dev *dev, int mps)
3518{
3519 int cap, err = -EINVAL;
3520 u16 ctl, v;
3521
3522 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
3523 goto out;
3524
3525 v = ffs(mps) - 8;
3526 if (v > dev->pcie_mpss)
3527 goto out;
3528 v <<= 5;
3529
3530 cap = pci_pcie_cap(dev);
3531 if (!cap)
3532 goto out;
3533
3534 err = pci_read_config_word(dev, cap + PCI_EXP_DEVCTL, &ctl);
3535 if (err)
3536 goto out;
3537
3538 if ((ctl & PCI_EXP_DEVCTL_PAYLOAD) != v) {
3539 ctl &= ~PCI_EXP_DEVCTL_PAYLOAD;
3540 ctl |= v;
3541 err = pci_write_config_word(dev, cap + PCI_EXP_DEVCTL, ctl);
3542 }
3543out:
3544 return err;
3545}
3546
3547/**
3548 * pci_select_bars - Make BAR mask from the type of resource
3549 * @dev: the PCI device for which BAR mask is made
3550 * @flags: resource type mask to be selected
3551 *
3552 * This helper routine makes bar mask from the type of resource.
3553 */
3554int pci_select_bars(struct pci_dev *dev, unsigned long flags)
3555{
3556 int i, bars = 0;
3557 for (i = 0; i < PCI_NUM_RESOURCES; i++)
3558 if (pci_resource_flags(dev, i) & flags)
3559 bars |= (1 << i);
3560 return bars;
3561}
3562
3563/**
3564 * pci_resource_bar - get position of the BAR associated with a resource
3565 * @dev: the PCI device
3566 * @resno: the resource number
3567 * @type: the BAR type to be filled in
3568 *
3569 * Returns BAR position in config space, or 0 if the BAR is invalid.
3570 */
3571int pci_resource_bar(struct pci_dev *dev, int resno, enum pci_bar_type *type)
3572{
3573 int reg;
3574
3575 if (resno < PCI_ROM_RESOURCE) {
3576 *type = pci_bar_unknown;
3577 return PCI_BASE_ADDRESS_0 + 4 * resno;
3578 } else if (resno == PCI_ROM_RESOURCE) {
3579 *type = pci_bar_mem32;
3580 return dev->rom_base_reg;
3581 } else if (resno < PCI_BRIDGE_RESOURCES) {
3582 /* device specific resource */
3583 reg = pci_iov_resource_bar(dev, resno, type);
3584 if (reg)
3585 return reg;
3586 }
3587
3588 dev_err(&dev->dev, "BAR %d: invalid resource\n", resno);
3589 return 0;
3590}
3591
3592/* Some architectures require additional programming to enable VGA */
3593static arch_set_vga_state_t arch_set_vga_state;
3594
3595void __init pci_register_set_vga_state(arch_set_vga_state_t func)
3596{
3597 arch_set_vga_state = func; /* NULL disables */
3598}
3599
3600static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
3601 unsigned int command_bits, u32 flags)
3602{
3603 if (arch_set_vga_state)
3604 return arch_set_vga_state(dev, decode, command_bits,
3605 flags);
3606 return 0;
3607}
3608
3609/**
3610 * pci_set_vga_state - set VGA decode state on device and parents if requested
3611 * @dev: the PCI device
3612 * @decode: true = enable decoding, false = disable decoding
3613 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
3614 * @flags: traverse ancestors and change bridges
3615 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
3616 */
3617int pci_set_vga_state(struct pci_dev *dev, bool decode,
3618 unsigned int command_bits, u32 flags)
3619{
3620 struct pci_bus *bus;
3621 struct pci_dev *bridge;
3622 u16 cmd;
3623 int rc;
3624
3625 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) & (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
3626
3627 /* ARCH specific VGA enables */
3628 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
3629 if (rc)
3630 return rc;
3631
3632 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
3633 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3634 if (decode == true)
3635 cmd |= command_bits;
3636 else
3637 cmd &= ~command_bits;
3638 pci_write_config_word(dev, PCI_COMMAND, cmd);
3639 }
3640
3641 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
3642 return 0;
3643
3644 bus = dev->bus;
3645 while (bus) {
3646 bridge = bus->self;
3647 if (bridge) {
3648 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
3649 &cmd);
3650 if (decode == true)
3651 cmd |= PCI_BRIDGE_CTL_VGA;
3652 else
3653 cmd &= ~PCI_BRIDGE_CTL_VGA;
3654 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
3655 cmd);
3656 }
3657 bus = bus->parent;
3658 }
3659 return 0;
3660}
3661
3662#define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
3663static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
3664static DEFINE_SPINLOCK(resource_alignment_lock);
3665
3666/**
3667 * pci_specified_resource_alignment - get resource alignment specified by user.
3668 * @dev: the PCI device to get
3669 *
3670 * RETURNS: Resource alignment if it is specified.
3671 * Zero if it is not specified.
3672 */
3673resource_size_t pci_specified_resource_alignment(struct pci_dev *dev)
3674{
3675 int seg, bus, slot, func, align_order, count;
3676 resource_size_t align = 0;
3677 char *p;
3678
3679 spin_lock(&resource_alignment_lock);
3680 p = resource_alignment_param;
3681 while (*p) {
3682 count = 0;
3683 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
3684 p[count] == '@') {
3685 p += count + 1;
3686 } else {
3687 align_order = -1;
3688 }
3689 if (sscanf(p, "%x:%x:%x.%x%n",
3690 &seg, &bus, &slot, &func, &count) != 4) {
3691 seg = 0;
3692 if (sscanf(p, "%x:%x.%x%n",
3693 &bus, &slot, &func, &count) != 3) {
3694 /* Invalid format */
3695 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
3696 p);
3697 break;
3698 }
3699 }
3700 p += count;
3701 if (seg == pci_domain_nr(dev->bus) &&
3702 bus == dev->bus->number &&
3703 slot == PCI_SLOT(dev->devfn) &&
3704 func == PCI_FUNC(dev->devfn)) {
3705 if (align_order == -1) {
3706 align = PAGE_SIZE;
3707 } else {
3708 align = 1 << align_order;
3709 }
3710 /* Found */
3711 break;
3712 }
3713 if (*p != ';' && *p != ',') {
3714 /* End of param or invalid format */
3715 break;
3716 }
3717 p++;
3718 }
3719 spin_unlock(&resource_alignment_lock);
3720 return align;
3721}
3722
3723/**
3724 * pci_is_reassigndev - check if specified PCI is target device to reassign
3725 * @dev: the PCI device to check
3726 *
3727 * RETURNS: non-zero for PCI device is a target device to reassign,
3728 * or zero is not.
3729 */
3730int pci_is_reassigndev(struct pci_dev *dev)
3731{
3732 return (pci_specified_resource_alignment(dev) != 0);
3733}
3734
3735/*
3736 * This function disables memory decoding and releases memory resources
3737 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
3738 * It also rounds up size to specified alignment.
3739 * Later on, the kernel will assign page-aligned memory resource back
3740 * to the device.
3741 */
3742void pci_reassigndev_resource_alignment(struct pci_dev *dev)
3743{
3744 int i;
3745 struct resource *r;
3746 resource_size_t align, size;
3747 u16 command;
3748
3749 if (!pci_is_reassigndev(dev))
3750 return;
3751
3752 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
3753 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
3754 dev_warn(&dev->dev,
3755 "Can't reassign resources to host bridge.\n");
3756 return;
3757 }
3758
3759 dev_info(&dev->dev,
3760 "Disabling memory decoding and releasing memory resources.\n");
3761 pci_read_config_word(dev, PCI_COMMAND, &command);
3762 command &= ~PCI_COMMAND_MEMORY;
3763 pci_write_config_word(dev, PCI_COMMAND, command);
3764
3765 align = pci_specified_resource_alignment(dev);
3766 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++) {
3767 r = &dev->resource[i];
3768 if (!(r->flags & IORESOURCE_MEM))
3769 continue;
3770 size = resource_size(r);
3771 if (size < align) {
3772 size = align;
3773 dev_info(&dev->dev,
3774 "Rounding up size of resource #%d to %#llx.\n",
3775 i, (unsigned long long)size);
3776 }
3777 r->end = size - 1;
3778 r->start = 0;
3779 }
3780 /* Need to disable bridge's resource window,
3781 * to enable the kernel to reassign new resource
3782 * window later on.
3783 */
3784 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
3785 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
3786 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
3787 r = &dev->resource[i];
3788 if (!(r->flags & IORESOURCE_MEM))
3789 continue;
3790 r->end = resource_size(r) - 1;
3791 r->start = 0;
3792 }
3793 pci_disable_bridge_window(dev);
3794 }
3795}
3796
3797ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
3798{
3799 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
3800 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
3801 spin_lock(&resource_alignment_lock);
3802 strncpy(resource_alignment_param, buf, count);
3803 resource_alignment_param[count] = '\0';
3804 spin_unlock(&resource_alignment_lock);
3805 return count;
3806}
3807
3808ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
3809{
3810 size_t count;
3811 spin_lock(&resource_alignment_lock);
3812 count = snprintf(buf, size, "%s", resource_alignment_param);
3813 spin_unlock(&resource_alignment_lock);
3814 return count;
3815}
3816
3817static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
3818{
3819 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
3820}
3821
3822static ssize_t pci_resource_alignment_store(struct bus_type *bus,
3823 const char *buf, size_t count)
3824{
3825 return pci_set_resource_alignment_param(buf, count);
3826}
3827
3828BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
3829 pci_resource_alignment_store);
3830
3831static int __init pci_resource_alignment_sysfs_init(void)
3832{
3833 return bus_create_file(&pci_bus_type,
3834 &bus_attr_resource_alignment);
3835}
3836
3837late_initcall(pci_resource_alignment_sysfs_init);
3838
3839static void __devinit pci_no_domains(void)
3840{
3841#ifdef CONFIG_PCI_DOMAINS
3842 pci_domains_supported = 0;
3843#endif
3844}
3845
3846/**
3847 * pci_ext_cfg_enabled - can we access extended PCI config space?
3848 * @dev: The PCI device of the root bridge.
3849 *
3850 * Returns 1 if we can access PCI extended config space (offsets
3851 * greater than 0xff). This is the default implementation. Architecture
3852 * implementations can override this.
3853 */
3854int __attribute__ ((weak)) pci_ext_cfg_avail(struct pci_dev *dev)
3855{
3856 return 1;
3857}
3858
3859void __weak pci_fixup_cardbus(struct pci_bus *bus)
3860{
3861}
3862EXPORT_SYMBOL(pci_fixup_cardbus);
3863
3864static int __init pci_setup(char *str)
3865{
3866 while (str) {
3867 char *k = strchr(str, ',');
3868 if (k)
3869 *k++ = 0;
3870 if (*str && (str = pcibios_setup(str)) && *str) {
3871 if (!strcmp(str, "nomsi")) {
3872 pci_no_msi();
3873 } else if (!strcmp(str, "noaer")) {
3874 pci_no_aer();
3875 } else if (!strncmp(str, "realloc=", 8)) {
3876 pci_realloc_get_opt(str + 8);
3877 } else if (!strncmp(str, "realloc", 7)) {
3878 pci_realloc_get_opt("on");
3879 } else if (!strcmp(str, "nodomains")) {
3880 pci_no_domains();
3881 } else if (!strncmp(str, "noari", 5)) {
3882 pcie_ari_disabled = true;
3883 } else if (!strncmp(str, "cbiosize=", 9)) {
3884 pci_cardbus_io_size = memparse(str + 9, &str);
3885 } else if (!strncmp(str, "cbmemsize=", 10)) {
3886 pci_cardbus_mem_size = memparse(str + 10, &str);
3887 } else if (!strncmp(str, "resource_alignment=", 19)) {
3888 pci_set_resource_alignment_param(str + 19,
3889 strlen(str + 19));
3890 } else if (!strncmp(str, "ecrc=", 5)) {
3891 pcie_ecrc_get_policy(str + 5);
3892 } else if (!strncmp(str, "hpiosize=", 9)) {
3893 pci_hotplug_io_size = memparse(str + 9, &str);
3894 } else if (!strncmp(str, "hpmemsize=", 10)) {
3895 pci_hotplug_mem_size = memparse(str + 10, &str);
3896 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
3897 pcie_bus_config = PCIE_BUS_TUNE_OFF;
3898 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
3899 pcie_bus_config = PCIE_BUS_SAFE;
3900 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
3901 pcie_bus_config = PCIE_BUS_PERFORMANCE;
3902 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
3903 pcie_bus_config = PCIE_BUS_PEER2PEER;
3904 } else if (!strncmp(str, "pcie_scan_all", 13)) {
3905 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
3906 } else {
3907 printk(KERN_ERR "PCI: Unknown option `%s'\n",
3908 str);
3909 }
3910 }
3911 str = k;
3912 }
3913 return 0;
3914}
3915early_param("pci", pci_setup);
3916
3917EXPORT_SYMBOL(pci_reenable_device);
3918EXPORT_SYMBOL(pci_enable_device_io);
3919EXPORT_SYMBOL(pci_enable_device_mem);
3920EXPORT_SYMBOL(pci_enable_device);
3921EXPORT_SYMBOL(pcim_enable_device);
3922EXPORT_SYMBOL(pcim_pin_device);
3923EXPORT_SYMBOL(pci_disable_device);
3924EXPORT_SYMBOL(pci_find_capability);
3925EXPORT_SYMBOL(pci_bus_find_capability);
3926EXPORT_SYMBOL(pci_release_regions);
3927EXPORT_SYMBOL(pci_request_regions);
3928EXPORT_SYMBOL(pci_request_regions_exclusive);
3929EXPORT_SYMBOL(pci_release_region);
3930EXPORT_SYMBOL(pci_request_region);
3931EXPORT_SYMBOL(pci_request_region_exclusive);
3932EXPORT_SYMBOL(pci_release_selected_regions);
3933EXPORT_SYMBOL(pci_request_selected_regions);
3934EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3935EXPORT_SYMBOL(pci_set_master);
3936EXPORT_SYMBOL(pci_clear_master);
3937EXPORT_SYMBOL(pci_set_mwi);
3938EXPORT_SYMBOL(pci_try_set_mwi);
3939EXPORT_SYMBOL(pci_clear_mwi);
3940EXPORT_SYMBOL_GPL(pci_intx);
3941EXPORT_SYMBOL(pci_assign_resource);
3942EXPORT_SYMBOL(pci_find_parent_resource);
3943EXPORT_SYMBOL(pci_select_bars);
3944
3945EXPORT_SYMBOL(pci_set_power_state);
3946EXPORT_SYMBOL(pci_save_state);
3947EXPORT_SYMBOL(pci_restore_state);
3948EXPORT_SYMBOL(pci_pme_capable);
3949EXPORT_SYMBOL(pci_pme_active);
3950EXPORT_SYMBOL(pci_wake_from_d3);
3951EXPORT_SYMBOL(pci_target_state);
3952EXPORT_SYMBOL(pci_prepare_to_sleep);
3953EXPORT_SYMBOL(pci_back_from_sleep);
3954EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * PCI Bus Services, see include/linux/pci.h for further explanation.
4 *
5 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
6 * David Mosberger-Tang
7 *
8 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
9 */
10
11#include <linux/acpi.h>
12#include <linux/kernel.h>
13#include <linux/delay.h>
14#include <linux/dmi.h>
15#include <linux/init.h>
16#include <linux/of.h>
17#include <linux/of_pci.h>
18#include <linux/pci.h>
19#include <linux/pm.h>
20#include <linux/slab.h>
21#include <linux/module.h>
22#include <linux/spinlock.h>
23#include <linux/string.h>
24#include <linux/log2.h>
25#include <linux/logic_pio.h>
26#include <linux/pci-aspm.h>
27#include <linux/pm_wakeup.h>
28#include <linux/interrupt.h>
29#include <linux/device.h>
30#include <linux/pm_runtime.h>
31#include <linux/pci_hotplug.h>
32#include <linux/vmalloc.h>
33#include <linux/pci-ats.h>
34#include <asm/setup.h>
35#include <asm/dma.h>
36#include <linux/aer.h>
37#include "pci.h"
38
39const char *pci_power_names[] = {
40 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
41};
42EXPORT_SYMBOL_GPL(pci_power_names);
43
44int isa_dma_bridge_buggy;
45EXPORT_SYMBOL(isa_dma_bridge_buggy);
46
47int pci_pci_problems;
48EXPORT_SYMBOL(pci_pci_problems);
49
50unsigned int pci_pm_d3_delay;
51
52static void pci_pme_list_scan(struct work_struct *work);
53
54static LIST_HEAD(pci_pme_list);
55static DEFINE_MUTEX(pci_pme_list_mutex);
56static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
57
58struct pci_pme_device {
59 struct list_head list;
60 struct pci_dev *dev;
61};
62
63#define PME_TIMEOUT 1000 /* How long between PME checks */
64
65static void pci_dev_d3_sleep(struct pci_dev *dev)
66{
67 unsigned int delay = dev->d3_delay;
68
69 if (delay < pci_pm_d3_delay)
70 delay = pci_pm_d3_delay;
71
72 if (delay)
73 msleep(delay);
74}
75
76#ifdef CONFIG_PCI_DOMAINS
77int pci_domains_supported = 1;
78#endif
79
80#define DEFAULT_CARDBUS_IO_SIZE (256)
81#define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
82/* pci=cbmemsize=nnM,cbiosize=nn can override this */
83unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
84unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
85
86#define DEFAULT_HOTPLUG_IO_SIZE (256)
87#define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
88/* pci=hpmemsize=nnM,hpiosize=nn can override this */
89unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
90unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
91
92#define DEFAULT_HOTPLUG_BUS_SIZE 1
93unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
94
95enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
96
97/*
98 * The default CLS is used if arch didn't set CLS explicitly and not
99 * all pci devices agree on the same value. Arch can override either
100 * the dfl or actual value as it sees fit. Don't forget this is
101 * measured in 32-bit words, not bytes.
102 */
103u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
104u8 pci_cache_line_size;
105
106/*
107 * If we set up a device for bus mastering, we need to check the latency
108 * timer as certain BIOSes forget to set it properly.
109 */
110unsigned int pcibios_max_latency = 255;
111
112/* If set, the PCIe ARI capability will not be used. */
113static bool pcie_ari_disabled;
114
115/* Disable bridge_d3 for all PCIe ports */
116static bool pci_bridge_d3_disable;
117/* Force bridge_d3 for all PCIe ports */
118static bool pci_bridge_d3_force;
119
120static int __init pcie_port_pm_setup(char *str)
121{
122 if (!strcmp(str, "off"))
123 pci_bridge_d3_disable = true;
124 else if (!strcmp(str, "force"))
125 pci_bridge_d3_force = true;
126 return 1;
127}
128__setup("pcie_port_pm=", pcie_port_pm_setup);
129
130/* Time to wait after a reset for device to become responsive */
131#define PCIE_RESET_READY_POLL_MS 60000
132
133/**
134 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
135 * @bus: pointer to PCI bus structure to search
136 *
137 * Given a PCI bus, returns the highest PCI bus number present in the set
138 * including the given PCI bus and its list of child PCI buses.
139 */
140unsigned char pci_bus_max_busnr(struct pci_bus *bus)
141{
142 struct pci_bus *tmp;
143 unsigned char max, n;
144
145 max = bus->busn_res.end;
146 list_for_each_entry(tmp, &bus->children, node) {
147 n = pci_bus_max_busnr(tmp);
148 if (n > max)
149 max = n;
150 }
151 return max;
152}
153EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
154
155#ifdef CONFIG_HAS_IOMEM
156void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
157{
158 struct resource *res = &pdev->resource[bar];
159
160 /*
161 * Make sure the BAR is actually a memory resource, not an IO resource
162 */
163 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
164 pci_warn(pdev, "can't ioremap BAR %d: %pR\n", bar, res);
165 return NULL;
166 }
167 return ioremap_nocache(res->start, resource_size(res));
168}
169EXPORT_SYMBOL_GPL(pci_ioremap_bar);
170
171void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
172{
173 /*
174 * Make sure the BAR is actually a memory resource, not an IO resource
175 */
176 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
177 WARN_ON(1);
178 return NULL;
179 }
180 return ioremap_wc(pci_resource_start(pdev, bar),
181 pci_resource_len(pdev, bar));
182}
183EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
184#endif
185
186
187static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
188 u8 pos, int cap, int *ttl)
189{
190 u8 id;
191 u16 ent;
192
193 pci_bus_read_config_byte(bus, devfn, pos, &pos);
194
195 while ((*ttl)--) {
196 if (pos < 0x40)
197 break;
198 pos &= ~3;
199 pci_bus_read_config_word(bus, devfn, pos, &ent);
200
201 id = ent & 0xff;
202 if (id == 0xff)
203 break;
204 if (id == cap)
205 return pos;
206 pos = (ent >> 8);
207 }
208 return 0;
209}
210
211static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
212 u8 pos, int cap)
213{
214 int ttl = PCI_FIND_CAP_TTL;
215
216 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
217}
218
219int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
220{
221 return __pci_find_next_cap(dev->bus, dev->devfn,
222 pos + PCI_CAP_LIST_NEXT, cap);
223}
224EXPORT_SYMBOL_GPL(pci_find_next_capability);
225
226static int __pci_bus_find_cap_start(struct pci_bus *bus,
227 unsigned int devfn, u8 hdr_type)
228{
229 u16 status;
230
231 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
232 if (!(status & PCI_STATUS_CAP_LIST))
233 return 0;
234
235 switch (hdr_type) {
236 case PCI_HEADER_TYPE_NORMAL:
237 case PCI_HEADER_TYPE_BRIDGE:
238 return PCI_CAPABILITY_LIST;
239 case PCI_HEADER_TYPE_CARDBUS:
240 return PCI_CB_CAPABILITY_LIST;
241 }
242
243 return 0;
244}
245
246/**
247 * pci_find_capability - query for devices' capabilities
248 * @dev: PCI device to query
249 * @cap: capability code
250 *
251 * Tell if a device supports a given PCI capability.
252 * Returns the address of the requested capability structure within the
253 * device's PCI configuration space or 0 in case the device does not
254 * support it. Possible values for @cap:
255 *
256 * %PCI_CAP_ID_PM Power Management
257 * %PCI_CAP_ID_AGP Accelerated Graphics Port
258 * %PCI_CAP_ID_VPD Vital Product Data
259 * %PCI_CAP_ID_SLOTID Slot Identification
260 * %PCI_CAP_ID_MSI Message Signalled Interrupts
261 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
262 * %PCI_CAP_ID_PCIX PCI-X
263 * %PCI_CAP_ID_EXP PCI Express
264 */
265int pci_find_capability(struct pci_dev *dev, int cap)
266{
267 int pos;
268
269 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
270 if (pos)
271 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
272
273 return pos;
274}
275EXPORT_SYMBOL(pci_find_capability);
276
277/**
278 * pci_bus_find_capability - query for devices' capabilities
279 * @bus: the PCI bus to query
280 * @devfn: PCI device to query
281 * @cap: capability code
282 *
283 * Like pci_find_capability() but works for pci devices that do not have a
284 * pci_dev structure set up yet.
285 *
286 * Returns the address of the requested capability structure within the
287 * device's PCI configuration space or 0 in case the device does not
288 * support it.
289 */
290int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
291{
292 int pos;
293 u8 hdr_type;
294
295 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
296
297 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
298 if (pos)
299 pos = __pci_find_next_cap(bus, devfn, pos, cap);
300
301 return pos;
302}
303EXPORT_SYMBOL(pci_bus_find_capability);
304
305/**
306 * pci_find_next_ext_capability - Find an extended capability
307 * @dev: PCI device to query
308 * @start: address at which to start looking (0 to start at beginning of list)
309 * @cap: capability code
310 *
311 * Returns the address of the next matching extended capability structure
312 * within the device's PCI configuration space or 0 if the device does
313 * not support it. Some capabilities can occur several times, e.g., the
314 * vendor-specific capability, and this provides a way to find them all.
315 */
316int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
317{
318 u32 header;
319 int ttl;
320 int pos = PCI_CFG_SPACE_SIZE;
321
322 /* minimum 8 bytes per capability */
323 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
324
325 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
326 return 0;
327
328 if (start)
329 pos = start;
330
331 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
332 return 0;
333
334 /*
335 * If we have no capabilities, this is indicated by cap ID,
336 * cap version and next pointer all being 0.
337 */
338 if (header == 0)
339 return 0;
340
341 while (ttl-- > 0) {
342 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
343 return pos;
344
345 pos = PCI_EXT_CAP_NEXT(header);
346 if (pos < PCI_CFG_SPACE_SIZE)
347 break;
348
349 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
350 break;
351 }
352
353 return 0;
354}
355EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
356
357/**
358 * pci_find_ext_capability - Find an extended capability
359 * @dev: PCI device to query
360 * @cap: capability code
361 *
362 * Returns the address of the requested extended capability structure
363 * within the device's PCI configuration space or 0 if the device does
364 * not support it. Possible values for @cap:
365 *
366 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
367 * %PCI_EXT_CAP_ID_VC Virtual Channel
368 * %PCI_EXT_CAP_ID_DSN Device Serial Number
369 * %PCI_EXT_CAP_ID_PWR Power Budgeting
370 */
371int pci_find_ext_capability(struct pci_dev *dev, int cap)
372{
373 return pci_find_next_ext_capability(dev, 0, cap);
374}
375EXPORT_SYMBOL_GPL(pci_find_ext_capability);
376
377static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
378{
379 int rc, ttl = PCI_FIND_CAP_TTL;
380 u8 cap, mask;
381
382 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
383 mask = HT_3BIT_CAP_MASK;
384 else
385 mask = HT_5BIT_CAP_MASK;
386
387 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
388 PCI_CAP_ID_HT, &ttl);
389 while (pos) {
390 rc = pci_read_config_byte(dev, pos + 3, &cap);
391 if (rc != PCIBIOS_SUCCESSFUL)
392 return 0;
393
394 if ((cap & mask) == ht_cap)
395 return pos;
396
397 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
398 pos + PCI_CAP_LIST_NEXT,
399 PCI_CAP_ID_HT, &ttl);
400 }
401
402 return 0;
403}
404/**
405 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
406 * @dev: PCI device to query
407 * @pos: Position from which to continue searching
408 * @ht_cap: Hypertransport capability code
409 *
410 * To be used in conjunction with pci_find_ht_capability() to search for
411 * all capabilities matching @ht_cap. @pos should always be a value returned
412 * from pci_find_ht_capability().
413 *
414 * NB. To be 100% safe against broken PCI devices, the caller should take
415 * steps to avoid an infinite loop.
416 */
417int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
418{
419 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
420}
421EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
422
423/**
424 * pci_find_ht_capability - query a device's Hypertransport capabilities
425 * @dev: PCI device to query
426 * @ht_cap: Hypertransport capability code
427 *
428 * Tell if a device supports a given Hypertransport capability.
429 * Returns an address within the device's PCI configuration space
430 * or 0 in case the device does not support the request capability.
431 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
432 * which has a Hypertransport capability matching @ht_cap.
433 */
434int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
435{
436 int pos;
437
438 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
439 if (pos)
440 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
441
442 return pos;
443}
444EXPORT_SYMBOL_GPL(pci_find_ht_capability);
445
446/**
447 * pci_find_parent_resource - return resource region of parent bus of given region
448 * @dev: PCI device structure contains resources to be searched
449 * @res: child resource record for which parent is sought
450 *
451 * For given resource region of given device, return the resource
452 * region of parent bus the given region is contained in.
453 */
454struct resource *pci_find_parent_resource(const struct pci_dev *dev,
455 struct resource *res)
456{
457 const struct pci_bus *bus = dev->bus;
458 struct resource *r;
459 int i;
460
461 pci_bus_for_each_resource(bus, r, i) {
462 if (!r)
463 continue;
464 if (resource_contains(r, res)) {
465
466 /*
467 * If the window is prefetchable but the BAR is
468 * not, the allocator made a mistake.
469 */
470 if (r->flags & IORESOURCE_PREFETCH &&
471 !(res->flags & IORESOURCE_PREFETCH))
472 return NULL;
473
474 /*
475 * If we're below a transparent bridge, there may
476 * be both a positively-decoded aperture and a
477 * subtractively-decoded region that contain the BAR.
478 * We want the positively-decoded one, so this depends
479 * on pci_bus_for_each_resource() giving us those
480 * first.
481 */
482 return r;
483 }
484 }
485 return NULL;
486}
487EXPORT_SYMBOL(pci_find_parent_resource);
488
489/**
490 * pci_find_resource - Return matching PCI device resource
491 * @dev: PCI device to query
492 * @res: Resource to look for
493 *
494 * Goes over standard PCI resources (BARs) and checks if the given resource
495 * is partially or fully contained in any of them. In that case the
496 * matching resource is returned, %NULL otherwise.
497 */
498struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
499{
500 int i;
501
502 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
503 struct resource *r = &dev->resource[i];
504
505 if (r->start && resource_contains(r, res))
506 return r;
507 }
508
509 return NULL;
510}
511EXPORT_SYMBOL(pci_find_resource);
512
513/**
514 * pci_find_pcie_root_port - return PCIe Root Port
515 * @dev: PCI device to query
516 *
517 * Traverse up the parent chain and return the PCIe Root Port PCI Device
518 * for a given PCI Device.
519 */
520struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
521{
522 struct pci_dev *bridge, *highest_pcie_bridge = dev;
523
524 bridge = pci_upstream_bridge(dev);
525 while (bridge && pci_is_pcie(bridge)) {
526 highest_pcie_bridge = bridge;
527 bridge = pci_upstream_bridge(bridge);
528 }
529
530 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
531 return NULL;
532
533 return highest_pcie_bridge;
534}
535EXPORT_SYMBOL(pci_find_pcie_root_port);
536
537/**
538 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
539 * @dev: the PCI device to operate on
540 * @pos: config space offset of status word
541 * @mask: mask of bit(s) to care about in status word
542 *
543 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
544 */
545int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
546{
547 int i;
548
549 /* Wait for Transaction Pending bit clean */
550 for (i = 0; i < 4; i++) {
551 u16 status;
552 if (i)
553 msleep((1 << (i - 1)) * 100);
554
555 pci_read_config_word(dev, pos, &status);
556 if (!(status & mask))
557 return 1;
558 }
559
560 return 0;
561}
562
563/**
564 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
565 * @dev: PCI device to have its BARs restored
566 *
567 * Restore the BAR values for a given device, so as to make it
568 * accessible by its driver.
569 */
570static void pci_restore_bars(struct pci_dev *dev)
571{
572 int i;
573
574 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
575 pci_update_resource(dev, i);
576}
577
578static const struct pci_platform_pm_ops *pci_platform_pm;
579
580int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
581{
582 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
583 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
584 return -EINVAL;
585 pci_platform_pm = ops;
586 return 0;
587}
588
589static inline bool platform_pci_power_manageable(struct pci_dev *dev)
590{
591 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
592}
593
594static inline int platform_pci_set_power_state(struct pci_dev *dev,
595 pci_power_t t)
596{
597 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
598}
599
600static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
601{
602 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
603}
604
605static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
606{
607 return pci_platform_pm ?
608 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
609}
610
611static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
612{
613 return pci_platform_pm ?
614 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
615}
616
617static inline bool platform_pci_need_resume(struct pci_dev *dev)
618{
619 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
620}
621
622/**
623 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
624 * given PCI device
625 * @dev: PCI device to handle.
626 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
627 *
628 * RETURN VALUE:
629 * -EINVAL if the requested state is invalid.
630 * -EIO if device does not support PCI PM or its PM capabilities register has a
631 * wrong version, or device doesn't support the requested state.
632 * 0 if device already is in the requested state.
633 * 0 if device's power state has been successfully changed.
634 */
635static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
636{
637 u16 pmcsr;
638 bool need_restore = false;
639
640 /* Check if we're already there */
641 if (dev->current_state == state)
642 return 0;
643
644 if (!dev->pm_cap)
645 return -EIO;
646
647 if (state < PCI_D0 || state > PCI_D3hot)
648 return -EINVAL;
649
650 /* Validate current state:
651 * Can enter D0 from any state, but if we can only go deeper
652 * to sleep if we're already in a low power state
653 */
654 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
655 && dev->current_state > state) {
656 pci_err(dev, "invalid power transition (from state %d to %d)\n",
657 dev->current_state, state);
658 return -EINVAL;
659 }
660
661 /* check if this device supports the desired state */
662 if ((state == PCI_D1 && !dev->d1_support)
663 || (state == PCI_D2 && !dev->d2_support))
664 return -EIO;
665
666 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
667
668 /* If we're (effectively) in D3, force entire word to 0.
669 * This doesn't affect PME_Status, disables PME_En, and
670 * sets PowerState to 0.
671 */
672 switch (dev->current_state) {
673 case PCI_D0:
674 case PCI_D1:
675 case PCI_D2:
676 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
677 pmcsr |= state;
678 break;
679 case PCI_D3hot:
680 case PCI_D3cold:
681 case PCI_UNKNOWN: /* Boot-up */
682 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
683 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
684 need_restore = true;
685 /* Fall-through: force to D0 */
686 default:
687 pmcsr = 0;
688 break;
689 }
690
691 /* enter specified state */
692 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
693
694 /* Mandatory power management transition delays */
695 /* see PCI PM 1.1 5.6.1 table 18 */
696 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
697 pci_dev_d3_sleep(dev);
698 else if (state == PCI_D2 || dev->current_state == PCI_D2)
699 udelay(PCI_PM_D2_DELAY);
700
701 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
702 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
703 if (dev->current_state != state && printk_ratelimit())
704 pci_info(dev, "Refused to change power state, currently in D%d\n",
705 dev->current_state);
706
707 /*
708 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
709 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
710 * from D3hot to D0 _may_ perform an internal reset, thereby
711 * going to "D0 Uninitialized" rather than "D0 Initialized".
712 * For example, at least some versions of the 3c905B and the
713 * 3c556B exhibit this behaviour.
714 *
715 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
716 * devices in a D3hot state at boot. Consequently, we need to
717 * restore at least the BARs so that the device will be
718 * accessible to its driver.
719 */
720 if (need_restore)
721 pci_restore_bars(dev);
722
723 if (dev->bus->self)
724 pcie_aspm_pm_state_change(dev->bus->self);
725
726 return 0;
727}
728
729/**
730 * pci_update_current_state - Read power state of given device and cache it
731 * @dev: PCI device to handle.
732 * @state: State to cache in case the device doesn't have the PM capability
733 *
734 * The power state is read from the PMCSR register, which however is
735 * inaccessible in D3cold. The platform firmware is therefore queried first
736 * to detect accessibility of the register. In case the platform firmware
737 * reports an incorrect state or the device isn't power manageable by the
738 * platform at all, we try to detect D3cold by testing accessibility of the
739 * vendor ID in config space.
740 */
741void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
742{
743 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
744 !pci_device_is_present(dev)) {
745 dev->current_state = PCI_D3cold;
746 } else if (dev->pm_cap) {
747 u16 pmcsr;
748
749 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
750 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
751 } else {
752 dev->current_state = state;
753 }
754}
755
756/**
757 * pci_power_up - Put the given device into D0 forcibly
758 * @dev: PCI device to power up
759 */
760void pci_power_up(struct pci_dev *dev)
761{
762 if (platform_pci_power_manageable(dev))
763 platform_pci_set_power_state(dev, PCI_D0);
764
765 pci_raw_set_power_state(dev, PCI_D0);
766 pci_update_current_state(dev, PCI_D0);
767}
768
769/**
770 * pci_platform_power_transition - Use platform to change device power state
771 * @dev: PCI device to handle.
772 * @state: State to put the device into.
773 */
774static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
775{
776 int error;
777
778 if (platform_pci_power_manageable(dev)) {
779 error = platform_pci_set_power_state(dev, state);
780 if (!error)
781 pci_update_current_state(dev, state);
782 } else
783 error = -ENODEV;
784
785 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
786 dev->current_state = PCI_D0;
787
788 return error;
789}
790
791/**
792 * pci_wakeup - Wake up a PCI device
793 * @pci_dev: Device to handle.
794 * @ign: ignored parameter
795 */
796static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
797{
798 pci_wakeup_event(pci_dev);
799 pm_request_resume(&pci_dev->dev);
800 return 0;
801}
802
803/**
804 * pci_wakeup_bus - Walk given bus and wake up devices on it
805 * @bus: Top bus of the subtree to walk.
806 */
807void pci_wakeup_bus(struct pci_bus *bus)
808{
809 if (bus)
810 pci_walk_bus(bus, pci_wakeup, NULL);
811}
812
813/**
814 * __pci_start_power_transition - Start power transition of a PCI device
815 * @dev: PCI device to handle.
816 * @state: State to put the device into.
817 */
818static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
819{
820 if (state == PCI_D0) {
821 pci_platform_power_transition(dev, PCI_D0);
822 /*
823 * Mandatory power management transition delays, see
824 * PCI Express Base Specification Revision 2.0 Section
825 * 6.6.1: Conventional Reset. Do not delay for
826 * devices powered on/off by corresponding bridge,
827 * because have already delayed for the bridge.
828 */
829 if (dev->runtime_d3cold) {
830 if (dev->d3cold_delay)
831 msleep(dev->d3cold_delay);
832 /*
833 * When powering on a bridge from D3cold, the
834 * whole hierarchy may be powered on into
835 * D0uninitialized state, resume them to give
836 * them a chance to suspend again
837 */
838 pci_wakeup_bus(dev->subordinate);
839 }
840 }
841}
842
843/**
844 * __pci_dev_set_current_state - Set current state of a PCI device
845 * @dev: Device to handle
846 * @data: pointer to state to be set
847 */
848static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
849{
850 pci_power_t state = *(pci_power_t *)data;
851
852 dev->current_state = state;
853 return 0;
854}
855
856/**
857 * pci_bus_set_current_state - Walk given bus and set current state of devices
858 * @bus: Top bus of the subtree to walk.
859 * @state: state to be set
860 */
861void pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
862{
863 if (bus)
864 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
865}
866
867/**
868 * __pci_complete_power_transition - Complete power transition of a PCI device
869 * @dev: PCI device to handle.
870 * @state: State to put the device into.
871 *
872 * This function should not be called directly by device drivers.
873 */
874int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
875{
876 int ret;
877
878 if (state <= PCI_D0)
879 return -EINVAL;
880 ret = pci_platform_power_transition(dev, state);
881 /* Power off the bridge may power off the whole hierarchy */
882 if (!ret && state == PCI_D3cold)
883 pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
884 return ret;
885}
886EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
887
888/**
889 * pci_set_power_state - Set the power state of a PCI device
890 * @dev: PCI device to handle.
891 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
892 *
893 * Transition a device to a new power state, using the platform firmware and/or
894 * the device's PCI PM registers.
895 *
896 * RETURN VALUE:
897 * -EINVAL if the requested state is invalid.
898 * -EIO if device does not support PCI PM or its PM capabilities register has a
899 * wrong version, or device doesn't support the requested state.
900 * 0 if the transition is to D1 or D2 but D1 and D2 are not supported.
901 * 0 if device already is in the requested state.
902 * 0 if the transition is to D3 but D3 is not supported.
903 * 0 if device's power state has been successfully changed.
904 */
905int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
906{
907 int error;
908
909 /* bound the state we're entering */
910 if (state > PCI_D3cold)
911 state = PCI_D3cold;
912 else if (state < PCI_D0)
913 state = PCI_D0;
914 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
915 /*
916 * If the device or the parent bridge do not support PCI PM,
917 * ignore the request if we're doing anything other than putting
918 * it into D0 (which would only happen on boot).
919 */
920 return 0;
921
922 /* Check if we're already there */
923 if (dev->current_state == state)
924 return 0;
925
926 __pci_start_power_transition(dev, state);
927
928 /* This device is quirked not to be put into D3, so
929 don't put it in D3 */
930 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
931 return 0;
932
933 /*
934 * To put device in D3cold, we put device into D3hot in native
935 * way, then put device into D3cold with platform ops
936 */
937 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
938 PCI_D3hot : state);
939
940 if (!__pci_complete_power_transition(dev, state))
941 error = 0;
942
943 return error;
944}
945EXPORT_SYMBOL(pci_set_power_state);
946
947/**
948 * pci_choose_state - Choose the power state of a PCI device
949 * @dev: PCI device to be suspended
950 * @state: target sleep state for the whole system. This is the value
951 * that is passed to suspend() function.
952 *
953 * Returns PCI power state suitable for given device and given system
954 * message.
955 */
956
957pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
958{
959 pci_power_t ret;
960
961 if (!dev->pm_cap)
962 return PCI_D0;
963
964 ret = platform_pci_choose_state(dev);
965 if (ret != PCI_POWER_ERROR)
966 return ret;
967
968 switch (state.event) {
969 case PM_EVENT_ON:
970 return PCI_D0;
971 case PM_EVENT_FREEZE:
972 case PM_EVENT_PRETHAW:
973 /* REVISIT both freeze and pre-thaw "should" use D0 */
974 case PM_EVENT_SUSPEND:
975 case PM_EVENT_HIBERNATE:
976 return PCI_D3hot;
977 default:
978 pci_info(dev, "unrecognized suspend event %d\n",
979 state.event);
980 BUG();
981 }
982 return PCI_D0;
983}
984EXPORT_SYMBOL(pci_choose_state);
985
986#define PCI_EXP_SAVE_REGS 7
987
988static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
989 u16 cap, bool extended)
990{
991 struct pci_cap_saved_state *tmp;
992
993 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
994 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
995 return tmp;
996 }
997 return NULL;
998}
999
1000struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
1001{
1002 return _pci_find_saved_cap(dev, cap, false);
1003}
1004
1005struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
1006{
1007 return _pci_find_saved_cap(dev, cap, true);
1008}
1009
1010static int pci_save_pcie_state(struct pci_dev *dev)
1011{
1012 int i = 0;
1013 struct pci_cap_saved_state *save_state;
1014 u16 *cap;
1015
1016 if (!pci_is_pcie(dev))
1017 return 0;
1018
1019 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1020 if (!save_state) {
1021 pci_err(dev, "buffer not found in %s\n", __func__);
1022 return -ENOMEM;
1023 }
1024
1025 cap = (u16 *)&save_state->cap.data[0];
1026 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1027 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1028 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1029 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1030 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1031 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1032 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1033
1034 return 0;
1035}
1036
1037static void pci_restore_pcie_state(struct pci_dev *dev)
1038{
1039 int i = 0;
1040 struct pci_cap_saved_state *save_state;
1041 u16 *cap;
1042
1043 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1044 if (!save_state)
1045 return;
1046
1047 cap = (u16 *)&save_state->cap.data[0];
1048 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1049 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1050 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1051 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1052 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1053 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1054 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1055}
1056
1057
1058static int pci_save_pcix_state(struct pci_dev *dev)
1059{
1060 int pos;
1061 struct pci_cap_saved_state *save_state;
1062
1063 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1064 if (!pos)
1065 return 0;
1066
1067 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1068 if (!save_state) {
1069 pci_err(dev, "buffer not found in %s\n", __func__);
1070 return -ENOMEM;
1071 }
1072
1073 pci_read_config_word(dev, pos + PCI_X_CMD,
1074 (u16 *)save_state->cap.data);
1075
1076 return 0;
1077}
1078
1079static void pci_restore_pcix_state(struct pci_dev *dev)
1080{
1081 int i = 0, pos;
1082 struct pci_cap_saved_state *save_state;
1083 u16 *cap;
1084
1085 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1086 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1087 if (!save_state || !pos)
1088 return;
1089 cap = (u16 *)&save_state->cap.data[0];
1090
1091 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1092}
1093
1094
1095/**
1096 * pci_save_state - save the PCI configuration space of a device before suspending
1097 * @dev: - PCI device that we're dealing with
1098 */
1099int pci_save_state(struct pci_dev *dev)
1100{
1101 int i;
1102 /* XXX: 100% dword access ok here? */
1103 for (i = 0; i < 16; i++)
1104 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1105 dev->state_saved = true;
1106
1107 i = pci_save_pcie_state(dev);
1108 if (i != 0)
1109 return i;
1110
1111 i = pci_save_pcix_state(dev);
1112 if (i != 0)
1113 return i;
1114
1115 return pci_save_vc_state(dev);
1116}
1117EXPORT_SYMBOL(pci_save_state);
1118
1119static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1120 u32 saved_val, int retry)
1121{
1122 u32 val;
1123
1124 pci_read_config_dword(pdev, offset, &val);
1125 if (val == saved_val)
1126 return;
1127
1128 for (;;) {
1129 pci_dbg(pdev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1130 offset, val, saved_val);
1131 pci_write_config_dword(pdev, offset, saved_val);
1132 if (retry-- <= 0)
1133 return;
1134
1135 pci_read_config_dword(pdev, offset, &val);
1136 if (val == saved_val)
1137 return;
1138
1139 mdelay(1);
1140 }
1141}
1142
1143static void pci_restore_config_space_range(struct pci_dev *pdev,
1144 int start, int end, int retry)
1145{
1146 int index;
1147
1148 for (index = end; index >= start; index--)
1149 pci_restore_config_dword(pdev, 4 * index,
1150 pdev->saved_config_space[index],
1151 retry);
1152}
1153
1154static void pci_restore_config_space(struct pci_dev *pdev)
1155{
1156 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1157 pci_restore_config_space_range(pdev, 10, 15, 0);
1158 /* Restore BARs before the command register. */
1159 pci_restore_config_space_range(pdev, 4, 9, 10);
1160 pci_restore_config_space_range(pdev, 0, 3, 0);
1161 } else {
1162 pci_restore_config_space_range(pdev, 0, 15, 0);
1163 }
1164}
1165
1166/**
1167 * pci_restore_state - Restore the saved state of a PCI device
1168 * @dev: - PCI device that we're dealing with
1169 */
1170void pci_restore_state(struct pci_dev *dev)
1171{
1172 if (!dev->state_saved)
1173 return;
1174
1175 /* PCI Express register must be restored first */
1176 pci_restore_pcie_state(dev);
1177 pci_restore_pasid_state(dev);
1178 pci_restore_pri_state(dev);
1179 pci_restore_ats_state(dev);
1180 pci_restore_vc_state(dev);
1181
1182 pci_cleanup_aer_error_status_regs(dev);
1183
1184 pci_restore_config_space(dev);
1185
1186 pci_restore_pcix_state(dev);
1187 pci_restore_msi_state(dev);
1188
1189 /* Restore ACS and IOV configuration state */
1190 pci_enable_acs(dev);
1191 pci_restore_iov_state(dev);
1192
1193 dev->state_saved = false;
1194}
1195EXPORT_SYMBOL(pci_restore_state);
1196
1197struct pci_saved_state {
1198 u32 config_space[16];
1199 struct pci_cap_saved_data cap[0];
1200};
1201
1202/**
1203 * pci_store_saved_state - Allocate and return an opaque struct containing
1204 * the device saved state.
1205 * @dev: PCI device that we're dealing with
1206 *
1207 * Return NULL if no state or error.
1208 */
1209struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1210{
1211 struct pci_saved_state *state;
1212 struct pci_cap_saved_state *tmp;
1213 struct pci_cap_saved_data *cap;
1214 size_t size;
1215
1216 if (!dev->state_saved)
1217 return NULL;
1218
1219 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1220
1221 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1222 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1223
1224 state = kzalloc(size, GFP_KERNEL);
1225 if (!state)
1226 return NULL;
1227
1228 memcpy(state->config_space, dev->saved_config_space,
1229 sizeof(state->config_space));
1230
1231 cap = state->cap;
1232 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1233 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1234 memcpy(cap, &tmp->cap, len);
1235 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1236 }
1237 /* Empty cap_save terminates list */
1238
1239 return state;
1240}
1241EXPORT_SYMBOL_GPL(pci_store_saved_state);
1242
1243/**
1244 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1245 * @dev: PCI device that we're dealing with
1246 * @state: Saved state returned from pci_store_saved_state()
1247 */
1248int pci_load_saved_state(struct pci_dev *dev,
1249 struct pci_saved_state *state)
1250{
1251 struct pci_cap_saved_data *cap;
1252
1253 dev->state_saved = false;
1254
1255 if (!state)
1256 return 0;
1257
1258 memcpy(dev->saved_config_space, state->config_space,
1259 sizeof(state->config_space));
1260
1261 cap = state->cap;
1262 while (cap->size) {
1263 struct pci_cap_saved_state *tmp;
1264
1265 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1266 if (!tmp || tmp->cap.size != cap->size)
1267 return -EINVAL;
1268
1269 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1270 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1271 sizeof(struct pci_cap_saved_data) + cap->size);
1272 }
1273
1274 dev->state_saved = true;
1275 return 0;
1276}
1277EXPORT_SYMBOL_GPL(pci_load_saved_state);
1278
1279/**
1280 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1281 * and free the memory allocated for it.
1282 * @dev: PCI device that we're dealing with
1283 * @state: Pointer to saved state returned from pci_store_saved_state()
1284 */
1285int pci_load_and_free_saved_state(struct pci_dev *dev,
1286 struct pci_saved_state **state)
1287{
1288 int ret = pci_load_saved_state(dev, *state);
1289 kfree(*state);
1290 *state = NULL;
1291 return ret;
1292}
1293EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1294
1295int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1296{
1297 return pci_enable_resources(dev, bars);
1298}
1299
1300static int do_pci_enable_device(struct pci_dev *dev, int bars)
1301{
1302 int err;
1303 struct pci_dev *bridge;
1304 u16 cmd;
1305 u8 pin;
1306
1307 err = pci_set_power_state(dev, PCI_D0);
1308 if (err < 0 && err != -EIO)
1309 return err;
1310
1311 bridge = pci_upstream_bridge(dev);
1312 if (bridge)
1313 pcie_aspm_powersave_config_link(bridge);
1314
1315 err = pcibios_enable_device(dev, bars);
1316 if (err < 0)
1317 return err;
1318 pci_fixup_device(pci_fixup_enable, dev);
1319
1320 if (dev->msi_enabled || dev->msix_enabled)
1321 return 0;
1322
1323 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1324 if (pin) {
1325 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1326 if (cmd & PCI_COMMAND_INTX_DISABLE)
1327 pci_write_config_word(dev, PCI_COMMAND,
1328 cmd & ~PCI_COMMAND_INTX_DISABLE);
1329 }
1330
1331 return 0;
1332}
1333
1334/**
1335 * pci_reenable_device - Resume abandoned device
1336 * @dev: PCI device to be resumed
1337 *
1338 * Note this function is a backend of pci_default_resume and is not supposed
1339 * to be called by normal code, write proper resume handler and use it instead.
1340 */
1341int pci_reenable_device(struct pci_dev *dev)
1342{
1343 if (pci_is_enabled(dev))
1344 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1345 return 0;
1346}
1347EXPORT_SYMBOL(pci_reenable_device);
1348
1349static void pci_enable_bridge(struct pci_dev *dev)
1350{
1351 struct pci_dev *bridge;
1352 int retval;
1353
1354 bridge = pci_upstream_bridge(dev);
1355 if (bridge)
1356 pci_enable_bridge(bridge);
1357
1358 if (pci_is_enabled(dev)) {
1359 if (!dev->is_busmaster)
1360 pci_set_master(dev);
1361 return;
1362 }
1363
1364 retval = pci_enable_device(dev);
1365 if (retval)
1366 pci_err(dev, "Error enabling bridge (%d), continuing\n",
1367 retval);
1368 pci_set_master(dev);
1369}
1370
1371static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1372{
1373 struct pci_dev *bridge;
1374 int err;
1375 int i, bars = 0;
1376
1377 /*
1378 * Power state could be unknown at this point, either due to a fresh
1379 * boot or a device removal call. So get the current power state
1380 * so that things like MSI message writing will behave as expected
1381 * (e.g. if the device really is in D0 at enable time).
1382 */
1383 if (dev->pm_cap) {
1384 u16 pmcsr;
1385 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1386 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1387 }
1388
1389 if (atomic_inc_return(&dev->enable_cnt) > 1)
1390 return 0; /* already enabled */
1391
1392 bridge = pci_upstream_bridge(dev);
1393 if (bridge)
1394 pci_enable_bridge(bridge);
1395
1396 /* only skip sriov related */
1397 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1398 if (dev->resource[i].flags & flags)
1399 bars |= (1 << i);
1400 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1401 if (dev->resource[i].flags & flags)
1402 bars |= (1 << i);
1403
1404 err = do_pci_enable_device(dev, bars);
1405 if (err < 0)
1406 atomic_dec(&dev->enable_cnt);
1407 return err;
1408}
1409
1410/**
1411 * pci_enable_device_io - Initialize a device for use with IO space
1412 * @dev: PCI device to be initialized
1413 *
1414 * Initialize device before it's used by a driver. Ask low-level code
1415 * to enable I/O resources. Wake up the device if it was suspended.
1416 * Beware, this function can fail.
1417 */
1418int pci_enable_device_io(struct pci_dev *dev)
1419{
1420 return pci_enable_device_flags(dev, IORESOURCE_IO);
1421}
1422EXPORT_SYMBOL(pci_enable_device_io);
1423
1424/**
1425 * pci_enable_device_mem - Initialize a device for use with Memory space
1426 * @dev: PCI device to be initialized
1427 *
1428 * Initialize device before it's used by a driver. Ask low-level code
1429 * to enable Memory resources. Wake up the device if it was suspended.
1430 * Beware, this function can fail.
1431 */
1432int pci_enable_device_mem(struct pci_dev *dev)
1433{
1434 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1435}
1436EXPORT_SYMBOL(pci_enable_device_mem);
1437
1438/**
1439 * pci_enable_device - Initialize device before it's used by a driver.
1440 * @dev: PCI device to be initialized
1441 *
1442 * Initialize device before it's used by a driver. Ask low-level code
1443 * to enable I/O and memory. Wake up the device if it was suspended.
1444 * Beware, this function can fail.
1445 *
1446 * Note we don't actually enable the device many times if we call
1447 * this function repeatedly (we just increment the count).
1448 */
1449int pci_enable_device(struct pci_dev *dev)
1450{
1451 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1452}
1453EXPORT_SYMBOL(pci_enable_device);
1454
1455/*
1456 * Managed PCI resources. This manages device on/off, intx/msi/msix
1457 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1458 * there's no need to track it separately. pci_devres is initialized
1459 * when a device is enabled using managed PCI device enable interface.
1460 */
1461struct pci_devres {
1462 unsigned int enabled:1;
1463 unsigned int pinned:1;
1464 unsigned int orig_intx:1;
1465 unsigned int restore_intx:1;
1466 unsigned int mwi:1;
1467 u32 region_mask;
1468};
1469
1470static void pcim_release(struct device *gendev, void *res)
1471{
1472 struct pci_dev *dev = to_pci_dev(gendev);
1473 struct pci_devres *this = res;
1474 int i;
1475
1476 if (dev->msi_enabled)
1477 pci_disable_msi(dev);
1478 if (dev->msix_enabled)
1479 pci_disable_msix(dev);
1480
1481 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1482 if (this->region_mask & (1 << i))
1483 pci_release_region(dev, i);
1484
1485 if (this->mwi)
1486 pci_clear_mwi(dev);
1487
1488 if (this->restore_intx)
1489 pci_intx(dev, this->orig_intx);
1490
1491 if (this->enabled && !this->pinned)
1492 pci_disable_device(dev);
1493}
1494
1495static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1496{
1497 struct pci_devres *dr, *new_dr;
1498
1499 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1500 if (dr)
1501 return dr;
1502
1503 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1504 if (!new_dr)
1505 return NULL;
1506 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1507}
1508
1509static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1510{
1511 if (pci_is_managed(pdev))
1512 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1513 return NULL;
1514}
1515
1516/**
1517 * pcim_enable_device - Managed pci_enable_device()
1518 * @pdev: PCI device to be initialized
1519 *
1520 * Managed pci_enable_device().
1521 */
1522int pcim_enable_device(struct pci_dev *pdev)
1523{
1524 struct pci_devres *dr;
1525 int rc;
1526
1527 dr = get_pci_dr(pdev);
1528 if (unlikely(!dr))
1529 return -ENOMEM;
1530 if (dr->enabled)
1531 return 0;
1532
1533 rc = pci_enable_device(pdev);
1534 if (!rc) {
1535 pdev->is_managed = 1;
1536 dr->enabled = 1;
1537 }
1538 return rc;
1539}
1540EXPORT_SYMBOL(pcim_enable_device);
1541
1542/**
1543 * pcim_pin_device - Pin managed PCI device
1544 * @pdev: PCI device to pin
1545 *
1546 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1547 * driver detach. @pdev must have been enabled with
1548 * pcim_enable_device().
1549 */
1550void pcim_pin_device(struct pci_dev *pdev)
1551{
1552 struct pci_devres *dr;
1553
1554 dr = find_pci_dr(pdev);
1555 WARN_ON(!dr || !dr->enabled);
1556 if (dr)
1557 dr->pinned = 1;
1558}
1559EXPORT_SYMBOL(pcim_pin_device);
1560
1561/*
1562 * pcibios_add_device - provide arch specific hooks when adding device dev
1563 * @dev: the PCI device being added
1564 *
1565 * Permits the platform to provide architecture specific functionality when
1566 * devices are added. This is the default implementation. Architecture
1567 * implementations can override this.
1568 */
1569int __weak pcibios_add_device(struct pci_dev *dev)
1570{
1571 return 0;
1572}
1573
1574/**
1575 * pcibios_release_device - provide arch specific hooks when releasing device dev
1576 * @dev: the PCI device being released
1577 *
1578 * Permits the platform to provide architecture specific functionality when
1579 * devices are released. This is the default implementation. Architecture
1580 * implementations can override this.
1581 */
1582void __weak pcibios_release_device(struct pci_dev *dev) {}
1583
1584/**
1585 * pcibios_disable_device - disable arch specific PCI resources for device dev
1586 * @dev: the PCI device to disable
1587 *
1588 * Disables architecture specific PCI resources for the device. This
1589 * is the default implementation. Architecture implementations can
1590 * override this.
1591 */
1592void __weak pcibios_disable_device(struct pci_dev *dev) {}
1593
1594/**
1595 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1596 * @irq: ISA IRQ to penalize
1597 * @active: IRQ active or not
1598 *
1599 * Permits the platform to provide architecture-specific functionality when
1600 * penalizing ISA IRQs. This is the default implementation. Architecture
1601 * implementations can override this.
1602 */
1603void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1604
1605static void do_pci_disable_device(struct pci_dev *dev)
1606{
1607 u16 pci_command;
1608
1609 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1610 if (pci_command & PCI_COMMAND_MASTER) {
1611 pci_command &= ~PCI_COMMAND_MASTER;
1612 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1613 }
1614
1615 pcibios_disable_device(dev);
1616}
1617
1618/**
1619 * pci_disable_enabled_device - Disable device without updating enable_cnt
1620 * @dev: PCI device to disable
1621 *
1622 * NOTE: This function is a backend of PCI power management routines and is
1623 * not supposed to be called drivers.
1624 */
1625void pci_disable_enabled_device(struct pci_dev *dev)
1626{
1627 if (pci_is_enabled(dev))
1628 do_pci_disable_device(dev);
1629}
1630
1631/**
1632 * pci_disable_device - Disable PCI device after use
1633 * @dev: PCI device to be disabled
1634 *
1635 * Signal to the system that the PCI device is not in use by the system
1636 * anymore. This only involves disabling PCI bus-mastering, if active.
1637 *
1638 * Note we don't actually disable the device until all callers of
1639 * pci_enable_device() have called pci_disable_device().
1640 */
1641void pci_disable_device(struct pci_dev *dev)
1642{
1643 struct pci_devres *dr;
1644
1645 dr = find_pci_dr(dev);
1646 if (dr)
1647 dr->enabled = 0;
1648
1649 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1650 "disabling already-disabled device");
1651
1652 if (atomic_dec_return(&dev->enable_cnt) != 0)
1653 return;
1654
1655 do_pci_disable_device(dev);
1656
1657 dev->is_busmaster = 0;
1658}
1659EXPORT_SYMBOL(pci_disable_device);
1660
1661/**
1662 * pcibios_set_pcie_reset_state - set reset state for device dev
1663 * @dev: the PCIe device reset
1664 * @state: Reset state to enter into
1665 *
1666 *
1667 * Sets the PCIe reset state for the device. This is the default
1668 * implementation. Architecture implementations can override this.
1669 */
1670int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1671 enum pcie_reset_state state)
1672{
1673 return -EINVAL;
1674}
1675
1676/**
1677 * pci_set_pcie_reset_state - set reset state for device dev
1678 * @dev: the PCIe device reset
1679 * @state: Reset state to enter into
1680 *
1681 *
1682 * Sets the PCI reset state for the device.
1683 */
1684int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1685{
1686 return pcibios_set_pcie_reset_state(dev, state);
1687}
1688EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1689
1690/**
1691 * pcie_clear_root_pme_status - Clear root port PME interrupt status.
1692 * @dev: PCIe root port or event collector.
1693 */
1694void pcie_clear_root_pme_status(struct pci_dev *dev)
1695{
1696 pcie_capability_set_dword(dev, PCI_EXP_RTSTA, PCI_EXP_RTSTA_PME);
1697}
1698
1699/**
1700 * pci_check_pme_status - Check if given device has generated PME.
1701 * @dev: Device to check.
1702 *
1703 * Check the PME status of the device and if set, clear it and clear PME enable
1704 * (if set). Return 'true' if PME status and PME enable were both set or
1705 * 'false' otherwise.
1706 */
1707bool pci_check_pme_status(struct pci_dev *dev)
1708{
1709 int pmcsr_pos;
1710 u16 pmcsr;
1711 bool ret = false;
1712
1713 if (!dev->pm_cap)
1714 return false;
1715
1716 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1717 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1718 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1719 return false;
1720
1721 /* Clear PME status. */
1722 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1723 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1724 /* Disable PME to avoid interrupt flood. */
1725 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1726 ret = true;
1727 }
1728
1729 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1730
1731 return ret;
1732}
1733
1734/**
1735 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1736 * @dev: Device to handle.
1737 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1738 *
1739 * Check if @dev has generated PME and queue a resume request for it in that
1740 * case.
1741 */
1742static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1743{
1744 if (pme_poll_reset && dev->pme_poll)
1745 dev->pme_poll = false;
1746
1747 if (pci_check_pme_status(dev)) {
1748 pci_wakeup_event(dev);
1749 pm_request_resume(&dev->dev);
1750 }
1751 return 0;
1752}
1753
1754/**
1755 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1756 * @bus: Top bus of the subtree to walk.
1757 */
1758void pci_pme_wakeup_bus(struct pci_bus *bus)
1759{
1760 if (bus)
1761 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1762}
1763
1764
1765/**
1766 * pci_pme_capable - check the capability of PCI device to generate PME#
1767 * @dev: PCI device to handle.
1768 * @state: PCI state from which device will issue PME#.
1769 */
1770bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1771{
1772 if (!dev->pm_cap)
1773 return false;
1774
1775 return !!(dev->pme_support & (1 << state));
1776}
1777EXPORT_SYMBOL(pci_pme_capable);
1778
1779static void pci_pme_list_scan(struct work_struct *work)
1780{
1781 struct pci_pme_device *pme_dev, *n;
1782
1783 mutex_lock(&pci_pme_list_mutex);
1784 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1785 if (pme_dev->dev->pme_poll) {
1786 struct pci_dev *bridge;
1787
1788 bridge = pme_dev->dev->bus->self;
1789 /*
1790 * If bridge is in low power state, the
1791 * configuration space of subordinate devices
1792 * may be not accessible
1793 */
1794 if (bridge && bridge->current_state != PCI_D0)
1795 continue;
1796 pci_pme_wakeup(pme_dev->dev, NULL);
1797 } else {
1798 list_del(&pme_dev->list);
1799 kfree(pme_dev);
1800 }
1801 }
1802 if (!list_empty(&pci_pme_list))
1803 queue_delayed_work(system_freezable_wq, &pci_pme_work,
1804 msecs_to_jiffies(PME_TIMEOUT));
1805 mutex_unlock(&pci_pme_list_mutex);
1806}
1807
1808static void __pci_pme_active(struct pci_dev *dev, bool enable)
1809{
1810 u16 pmcsr;
1811
1812 if (!dev->pme_support)
1813 return;
1814
1815 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1816 /* Clear PME_Status by writing 1 to it and enable PME# */
1817 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1818 if (!enable)
1819 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1820
1821 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1822}
1823
1824/**
1825 * pci_pme_restore - Restore PME configuration after config space restore.
1826 * @dev: PCI device to update.
1827 */
1828void pci_pme_restore(struct pci_dev *dev)
1829{
1830 u16 pmcsr;
1831
1832 if (!dev->pme_support)
1833 return;
1834
1835 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1836 if (dev->wakeup_prepared) {
1837 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1838 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
1839 } else {
1840 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1841 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1842 }
1843 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1844}
1845
1846/**
1847 * pci_pme_active - enable or disable PCI device's PME# function
1848 * @dev: PCI device to handle.
1849 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1850 *
1851 * The caller must verify that the device is capable of generating PME# before
1852 * calling this function with @enable equal to 'true'.
1853 */
1854void pci_pme_active(struct pci_dev *dev, bool enable)
1855{
1856 __pci_pme_active(dev, enable);
1857
1858 /*
1859 * PCI (as opposed to PCIe) PME requires that the device have
1860 * its PME# line hooked up correctly. Not all hardware vendors
1861 * do this, so the PME never gets delivered and the device
1862 * remains asleep. The easiest way around this is to
1863 * periodically walk the list of suspended devices and check
1864 * whether any have their PME flag set. The assumption is that
1865 * we'll wake up often enough anyway that this won't be a huge
1866 * hit, and the power savings from the devices will still be a
1867 * win.
1868 *
1869 * Although PCIe uses in-band PME message instead of PME# line
1870 * to report PME, PME does not work for some PCIe devices in
1871 * reality. For example, there are devices that set their PME
1872 * status bits, but don't really bother to send a PME message;
1873 * there are PCI Express Root Ports that don't bother to
1874 * trigger interrupts when they receive PME messages from the
1875 * devices below. So PME poll is used for PCIe devices too.
1876 */
1877
1878 if (dev->pme_poll) {
1879 struct pci_pme_device *pme_dev;
1880 if (enable) {
1881 pme_dev = kmalloc(sizeof(struct pci_pme_device),
1882 GFP_KERNEL);
1883 if (!pme_dev) {
1884 pci_warn(dev, "can't enable PME#\n");
1885 return;
1886 }
1887 pme_dev->dev = dev;
1888 mutex_lock(&pci_pme_list_mutex);
1889 list_add(&pme_dev->list, &pci_pme_list);
1890 if (list_is_singular(&pci_pme_list))
1891 queue_delayed_work(system_freezable_wq,
1892 &pci_pme_work,
1893 msecs_to_jiffies(PME_TIMEOUT));
1894 mutex_unlock(&pci_pme_list_mutex);
1895 } else {
1896 mutex_lock(&pci_pme_list_mutex);
1897 list_for_each_entry(pme_dev, &pci_pme_list, list) {
1898 if (pme_dev->dev == dev) {
1899 list_del(&pme_dev->list);
1900 kfree(pme_dev);
1901 break;
1902 }
1903 }
1904 mutex_unlock(&pci_pme_list_mutex);
1905 }
1906 }
1907
1908 pci_dbg(dev, "PME# %s\n", enable ? "enabled" : "disabled");
1909}
1910EXPORT_SYMBOL(pci_pme_active);
1911
1912/**
1913 * __pci_enable_wake - enable PCI device as wakeup event source
1914 * @dev: PCI device affected
1915 * @state: PCI state from which device will issue wakeup events
1916 * @enable: True to enable event generation; false to disable
1917 *
1918 * This enables the device as a wakeup event source, or disables it.
1919 * When such events involves platform-specific hooks, those hooks are
1920 * called automatically by this routine.
1921 *
1922 * Devices with legacy power management (no standard PCI PM capabilities)
1923 * always require such platform hooks.
1924 *
1925 * RETURN VALUE:
1926 * 0 is returned on success
1927 * -EINVAL is returned if device is not supposed to wake up the system
1928 * Error code depending on the platform is returned if both the platform and
1929 * the native mechanism fail to enable the generation of wake-up events
1930 */
1931static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
1932{
1933 int ret = 0;
1934
1935 /*
1936 * Bridges can only signal wakeup on behalf of subordinate devices,
1937 * but that is set up elsewhere, so skip them.
1938 */
1939 if (pci_has_subordinate(dev))
1940 return 0;
1941
1942 /* Don't do the same thing twice in a row for one device. */
1943 if (!!enable == !!dev->wakeup_prepared)
1944 return 0;
1945
1946 /*
1947 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1948 * Anderson we should be doing PME# wake enable followed by ACPI wake
1949 * enable. To disable wake-up we call the platform first, for symmetry.
1950 */
1951
1952 if (enable) {
1953 int error;
1954
1955 if (pci_pme_capable(dev, state))
1956 pci_pme_active(dev, true);
1957 else
1958 ret = 1;
1959 error = platform_pci_set_wakeup(dev, true);
1960 if (ret)
1961 ret = error;
1962 if (!ret)
1963 dev->wakeup_prepared = true;
1964 } else {
1965 platform_pci_set_wakeup(dev, false);
1966 pci_pme_active(dev, false);
1967 dev->wakeup_prepared = false;
1968 }
1969
1970 return ret;
1971}
1972
1973/**
1974 * pci_enable_wake - change wakeup settings for a PCI device
1975 * @pci_dev: Target device
1976 * @state: PCI state from which device will issue wakeup events
1977 * @enable: Whether or not to enable event generation
1978 *
1979 * If @enable is set, check device_may_wakeup() for the device before calling
1980 * __pci_enable_wake() for it.
1981 */
1982int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
1983{
1984 if (enable && !device_may_wakeup(&pci_dev->dev))
1985 return -EINVAL;
1986
1987 return __pci_enable_wake(pci_dev, state, enable);
1988}
1989EXPORT_SYMBOL(pci_enable_wake);
1990
1991/**
1992 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1993 * @dev: PCI device to prepare
1994 * @enable: True to enable wake-up event generation; false to disable
1995 *
1996 * Many drivers want the device to wake up the system from D3_hot or D3_cold
1997 * and this function allows them to set that up cleanly - pci_enable_wake()
1998 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1999 * ordering constraints.
2000 *
2001 * This function only returns error code if the device is not allowed to wake
2002 * up the system from sleep or it is not capable of generating PME# from both
2003 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
2004 */
2005int pci_wake_from_d3(struct pci_dev *dev, bool enable)
2006{
2007 return pci_pme_capable(dev, PCI_D3cold) ?
2008 pci_enable_wake(dev, PCI_D3cold, enable) :
2009 pci_enable_wake(dev, PCI_D3hot, enable);
2010}
2011EXPORT_SYMBOL(pci_wake_from_d3);
2012
2013/**
2014 * pci_target_state - find an appropriate low power state for a given PCI dev
2015 * @dev: PCI device
2016 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
2017 *
2018 * Use underlying platform code to find a supported low power state for @dev.
2019 * If the platform can't manage @dev, return the deepest state from which it
2020 * can generate wake events, based on any available PME info.
2021 */
2022static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
2023{
2024 pci_power_t target_state = PCI_D3hot;
2025
2026 if (platform_pci_power_manageable(dev)) {
2027 /*
2028 * Call the platform to choose the target state of the device
2029 * and enable wake-up from this state if supported.
2030 */
2031 pci_power_t state = platform_pci_choose_state(dev);
2032
2033 switch (state) {
2034 case PCI_POWER_ERROR:
2035 case PCI_UNKNOWN:
2036 break;
2037 case PCI_D1:
2038 case PCI_D2:
2039 if (pci_no_d1d2(dev))
2040 break;
2041 default:
2042 target_state = state;
2043 }
2044
2045 return target_state;
2046 }
2047
2048 if (!dev->pm_cap)
2049 target_state = PCI_D0;
2050
2051 /*
2052 * If the device is in D3cold even though it's not power-manageable by
2053 * the platform, it may have been powered down by non-standard means.
2054 * Best to let it slumber.
2055 */
2056 if (dev->current_state == PCI_D3cold)
2057 target_state = PCI_D3cold;
2058
2059 if (wakeup) {
2060 /*
2061 * Find the deepest state from which the device can generate
2062 * wake-up events, make it the target state and enable device
2063 * to generate PME#.
2064 */
2065 if (dev->pme_support) {
2066 while (target_state
2067 && !(dev->pme_support & (1 << target_state)))
2068 target_state--;
2069 }
2070 }
2071
2072 return target_state;
2073}
2074
2075/**
2076 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
2077 * @dev: Device to handle.
2078 *
2079 * Choose the power state appropriate for the device depending on whether
2080 * it can wake up the system and/or is power manageable by the platform
2081 * (PCI_D3hot is the default) and put the device into that state.
2082 */
2083int pci_prepare_to_sleep(struct pci_dev *dev)
2084{
2085 bool wakeup = device_may_wakeup(&dev->dev);
2086 pci_power_t target_state = pci_target_state(dev, wakeup);
2087 int error;
2088
2089 if (target_state == PCI_POWER_ERROR)
2090 return -EIO;
2091
2092 pci_enable_wake(dev, target_state, wakeup);
2093
2094 error = pci_set_power_state(dev, target_state);
2095
2096 if (error)
2097 pci_enable_wake(dev, target_state, false);
2098
2099 return error;
2100}
2101EXPORT_SYMBOL(pci_prepare_to_sleep);
2102
2103/**
2104 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2105 * @dev: Device to handle.
2106 *
2107 * Disable device's system wake-up capability and put it into D0.
2108 */
2109int pci_back_from_sleep(struct pci_dev *dev)
2110{
2111 pci_enable_wake(dev, PCI_D0, false);
2112 return pci_set_power_state(dev, PCI_D0);
2113}
2114EXPORT_SYMBOL(pci_back_from_sleep);
2115
2116/**
2117 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2118 * @dev: PCI device being suspended.
2119 *
2120 * Prepare @dev to generate wake-up events at run time and put it into a low
2121 * power state.
2122 */
2123int pci_finish_runtime_suspend(struct pci_dev *dev)
2124{
2125 pci_power_t target_state;
2126 int error;
2127
2128 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2129 if (target_state == PCI_POWER_ERROR)
2130 return -EIO;
2131
2132 dev->runtime_d3cold = target_state == PCI_D3cold;
2133
2134 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2135
2136 error = pci_set_power_state(dev, target_state);
2137
2138 if (error) {
2139 pci_enable_wake(dev, target_state, false);
2140 dev->runtime_d3cold = false;
2141 }
2142
2143 return error;
2144}
2145
2146/**
2147 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2148 * @dev: Device to check.
2149 *
2150 * Return true if the device itself is capable of generating wake-up events
2151 * (through the platform or using the native PCIe PME) or if the device supports
2152 * PME and one of its upstream bridges can generate wake-up events.
2153 */
2154bool pci_dev_run_wake(struct pci_dev *dev)
2155{
2156 struct pci_bus *bus = dev->bus;
2157
2158 if (!dev->pme_support)
2159 return false;
2160
2161 /* PME-capable in principle, but not from the target power state */
2162 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2163 return false;
2164
2165 if (device_can_wakeup(&dev->dev))
2166 return true;
2167
2168 while (bus->parent) {
2169 struct pci_dev *bridge = bus->self;
2170
2171 if (device_can_wakeup(&bridge->dev))
2172 return true;
2173
2174 bus = bus->parent;
2175 }
2176
2177 /* We have reached the root bus. */
2178 if (bus->bridge)
2179 return device_can_wakeup(bus->bridge);
2180
2181 return false;
2182}
2183EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2184
2185/**
2186 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2187 * @pci_dev: Device to check.
2188 *
2189 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2190 * reconfigured due to wakeup settings difference between system and runtime
2191 * suspend and the current power state of it is suitable for the upcoming
2192 * (system) transition.
2193 *
2194 * If the device is not configured for system wakeup, disable PME for it before
2195 * returning 'true' to prevent it from waking up the system unnecessarily.
2196 */
2197bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2198{
2199 struct device *dev = &pci_dev->dev;
2200 bool wakeup = device_may_wakeup(dev);
2201
2202 if (!pm_runtime_suspended(dev)
2203 || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2204 || platform_pci_need_resume(pci_dev))
2205 return false;
2206
2207 /*
2208 * At this point the device is good to go unless it's been configured
2209 * to generate PME at the runtime suspend time, but it is not supposed
2210 * to wake up the system. In that case, simply disable PME for it
2211 * (it will have to be re-enabled on exit from system resume).
2212 *
2213 * If the device's power state is D3cold and the platform check above
2214 * hasn't triggered, the device's configuration is suitable and we don't
2215 * need to manipulate it at all.
2216 */
2217 spin_lock_irq(&dev->power.lock);
2218
2219 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2220 !wakeup)
2221 __pci_pme_active(pci_dev, false);
2222
2223 spin_unlock_irq(&dev->power.lock);
2224 return true;
2225}
2226
2227/**
2228 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2229 * @pci_dev: Device to handle.
2230 *
2231 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2232 * it might have been disabled during the prepare phase of system suspend if
2233 * the device was not configured for system wakeup.
2234 */
2235void pci_dev_complete_resume(struct pci_dev *pci_dev)
2236{
2237 struct device *dev = &pci_dev->dev;
2238
2239 if (!pci_dev_run_wake(pci_dev))
2240 return;
2241
2242 spin_lock_irq(&dev->power.lock);
2243
2244 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2245 __pci_pme_active(pci_dev, true);
2246
2247 spin_unlock_irq(&dev->power.lock);
2248}
2249
2250void pci_config_pm_runtime_get(struct pci_dev *pdev)
2251{
2252 struct device *dev = &pdev->dev;
2253 struct device *parent = dev->parent;
2254
2255 if (parent)
2256 pm_runtime_get_sync(parent);
2257 pm_runtime_get_noresume(dev);
2258 /*
2259 * pdev->current_state is set to PCI_D3cold during suspending,
2260 * so wait until suspending completes
2261 */
2262 pm_runtime_barrier(dev);
2263 /*
2264 * Only need to resume devices in D3cold, because config
2265 * registers are still accessible for devices suspended but
2266 * not in D3cold.
2267 */
2268 if (pdev->current_state == PCI_D3cold)
2269 pm_runtime_resume(dev);
2270}
2271
2272void pci_config_pm_runtime_put(struct pci_dev *pdev)
2273{
2274 struct device *dev = &pdev->dev;
2275 struct device *parent = dev->parent;
2276
2277 pm_runtime_put(dev);
2278 if (parent)
2279 pm_runtime_put_sync(parent);
2280}
2281
2282/**
2283 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2284 * @bridge: Bridge to check
2285 *
2286 * This function checks if it is possible to move the bridge to D3.
2287 * Currently we only allow D3 for recent enough PCIe ports.
2288 */
2289bool pci_bridge_d3_possible(struct pci_dev *bridge)
2290{
2291 if (!pci_is_pcie(bridge))
2292 return false;
2293
2294 switch (pci_pcie_type(bridge)) {
2295 case PCI_EXP_TYPE_ROOT_PORT:
2296 case PCI_EXP_TYPE_UPSTREAM:
2297 case PCI_EXP_TYPE_DOWNSTREAM:
2298 if (pci_bridge_d3_disable)
2299 return false;
2300
2301 /*
2302 * Hotplug interrupts cannot be delivered if the link is down,
2303 * so parents of a hotplug port must stay awake. In addition,
2304 * hotplug ports handled by firmware in System Management Mode
2305 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2306 * For simplicity, disallow in general for now.
2307 */
2308 if (bridge->is_hotplug_bridge)
2309 return false;
2310
2311 if (pci_bridge_d3_force)
2312 return true;
2313
2314 /*
2315 * It should be safe to put PCIe ports from 2015 or newer
2316 * to D3.
2317 */
2318 if (dmi_get_bios_year() >= 2015)
2319 return true;
2320 break;
2321 }
2322
2323 return false;
2324}
2325
2326static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2327{
2328 bool *d3cold_ok = data;
2329
2330 if (/* The device needs to be allowed to go D3cold ... */
2331 dev->no_d3cold || !dev->d3cold_allowed ||
2332
2333 /* ... and if it is wakeup capable to do so from D3cold. */
2334 (device_may_wakeup(&dev->dev) &&
2335 !pci_pme_capable(dev, PCI_D3cold)) ||
2336
2337 /* If it is a bridge it must be allowed to go to D3. */
2338 !pci_power_manageable(dev))
2339
2340 *d3cold_ok = false;
2341
2342 return !*d3cold_ok;
2343}
2344
2345/*
2346 * pci_bridge_d3_update - Update bridge D3 capabilities
2347 * @dev: PCI device which is changed
2348 *
2349 * Update upstream bridge PM capabilities accordingly depending on if the
2350 * device PM configuration was changed or the device is being removed. The
2351 * change is also propagated upstream.
2352 */
2353void pci_bridge_d3_update(struct pci_dev *dev)
2354{
2355 bool remove = !device_is_registered(&dev->dev);
2356 struct pci_dev *bridge;
2357 bool d3cold_ok = true;
2358
2359 bridge = pci_upstream_bridge(dev);
2360 if (!bridge || !pci_bridge_d3_possible(bridge))
2361 return;
2362
2363 /*
2364 * If D3 is currently allowed for the bridge, removing one of its
2365 * children won't change that.
2366 */
2367 if (remove && bridge->bridge_d3)
2368 return;
2369
2370 /*
2371 * If D3 is currently allowed for the bridge and a child is added or
2372 * changed, disallowance of D3 can only be caused by that child, so
2373 * we only need to check that single device, not any of its siblings.
2374 *
2375 * If D3 is currently not allowed for the bridge, checking the device
2376 * first may allow us to skip checking its siblings.
2377 */
2378 if (!remove)
2379 pci_dev_check_d3cold(dev, &d3cold_ok);
2380
2381 /*
2382 * If D3 is currently not allowed for the bridge, this may be caused
2383 * either by the device being changed/removed or any of its siblings,
2384 * so we need to go through all children to find out if one of them
2385 * continues to block D3.
2386 */
2387 if (d3cold_ok && !bridge->bridge_d3)
2388 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2389 &d3cold_ok);
2390
2391 if (bridge->bridge_d3 != d3cold_ok) {
2392 bridge->bridge_d3 = d3cold_ok;
2393 /* Propagate change to upstream bridges */
2394 pci_bridge_d3_update(bridge);
2395 }
2396}
2397
2398/**
2399 * pci_d3cold_enable - Enable D3cold for device
2400 * @dev: PCI device to handle
2401 *
2402 * This function can be used in drivers to enable D3cold from the device
2403 * they handle. It also updates upstream PCI bridge PM capabilities
2404 * accordingly.
2405 */
2406void pci_d3cold_enable(struct pci_dev *dev)
2407{
2408 if (dev->no_d3cold) {
2409 dev->no_d3cold = false;
2410 pci_bridge_d3_update(dev);
2411 }
2412}
2413EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2414
2415/**
2416 * pci_d3cold_disable - Disable D3cold for device
2417 * @dev: PCI device to handle
2418 *
2419 * This function can be used in drivers to disable D3cold from the device
2420 * they handle. It also updates upstream PCI bridge PM capabilities
2421 * accordingly.
2422 */
2423void pci_d3cold_disable(struct pci_dev *dev)
2424{
2425 if (!dev->no_d3cold) {
2426 dev->no_d3cold = true;
2427 pci_bridge_d3_update(dev);
2428 }
2429}
2430EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2431
2432/**
2433 * pci_pm_init - Initialize PM functions of given PCI device
2434 * @dev: PCI device to handle.
2435 */
2436void pci_pm_init(struct pci_dev *dev)
2437{
2438 int pm;
2439 u16 pmc;
2440
2441 pm_runtime_forbid(&dev->dev);
2442 pm_runtime_set_active(&dev->dev);
2443 pm_runtime_enable(&dev->dev);
2444 device_enable_async_suspend(&dev->dev);
2445 dev->wakeup_prepared = false;
2446
2447 dev->pm_cap = 0;
2448 dev->pme_support = 0;
2449
2450 /* find PCI PM capability in list */
2451 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2452 if (!pm)
2453 return;
2454 /* Check device's ability to generate PME# */
2455 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2456
2457 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2458 pci_err(dev, "unsupported PM cap regs version (%u)\n",
2459 pmc & PCI_PM_CAP_VER_MASK);
2460 return;
2461 }
2462
2463 dev->pm_cap = pm;
2464 dev->d3_delay = PCI_PM_D3_WAIT;
2465 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2466 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2467 dev->d3cold_allowed = true;
2468
2469 dev->d1_support = false;
2470 dev->d2_support = false;
2471 if (!pci_no_d1d2(dev)) {
2472 if (pmc & PCI_PM_CAP_D1)
2473 dev->d1_support = true;
2474 if (pmc & PCI_PM_CAP_D2)
2475 dev->d2_support = true;
2476
2477 if (dev->d1_support || dev->d2_support)
2478 pci_printk(KERN_DEBUG, dev, "supports%s%s\n",
2479 dev->d1_support ? " D1" : "",
2480 dev->d2_support ? " D2" : "");
2481 }
2482
2483 pmc &= PCI_PM_CAP_PME_MASK;
2484 if (pmc) {
2485 pci_printk(KERN_DEBUG, dev, "PME# supported from%s%s%s%s%s\n",
2486 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2487 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2488 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2489 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2490 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2491 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2492 dev->pme_poll = true;
2493 /*
2494 * Make device's PM flags reflect the wake-up capability, but
2495 * let the user space enable it to wake up the system as needed.
2496 */
2497 device_set_wakeup_capable(&dev->dev, true);
2498 /* Disable the PME# generation functionality */
2499 pci_pme_active(dev, false);
2500 }
2501}
2502
2503static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2504{
2505 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2506
2507 switch (prop) {
2508 case PCI_EA_P_MEM:
2509 case PCI_EA_P_VF_MEM:
2510 flags |= IORESOURCE_MEM;
2511 break;
2512 case PCI_EA_P_MEM_PREFETCH:
2513 case PCI_EA_P_VF_MEM_PREFETCH:
2514 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2515 break;
2516 case PCI_EA_P_IO:
2517 flags |= IORESOURCE_IO;
2518 break;
2519 default:
2520 return 0;
2521 }
2522
2523 return flags;
2524}
2525
2526static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2527 u8 prop)
2528{
2529 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2530 return &dev->resource[bei];
2531#ifdef CONFIG_PCI_IOV
2532 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2533 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2534 return &dev->resource[PCI_IOV_RESOURCES +
2535 bei - PCI_EA_BEI_VF_BAR0];
2536#endif
2537 else if (bei == PCI_EA_BEI_ROM)
2538 return &dev->resource[PCI_ROM_RESOURCE];
2539 else
2540 return NULL;
2541}
2542
2543/* Read an Enhanced Allocation (EA) entry */
2544static int pci_ea_read(struct pci_dev *dev, int offset)
2545{
2546 struct resource *res;
2547 int ent_size, ent_offset = offset;
2548 resource_size_t start, end;
2549 unsigned long flags;
2550 u32 dw0, bei, base, max_offset;
2551 u8 prop;
2552 bool support_64 = (sizeof(resource_size_t) >= 8);
2553
2554 pci_read_config_dword(dev, ent_offset, &dw0);
2555 ent_offset += 4;
2556
2557 /* Entry size field indicates DWORDs after 1st */
2558 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2559
2560 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2561 goto out;
2562
2563 bei = (dw0 & PCI_EA_BEI) >> 4;
2564 prop = (dw0 & PCI_EA_PP) >> 8;
2565
2566 /*
2567 * If the Property is in the reserved range, try the Secondary
2568 * Property instead.
2569 */
2570 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2571 prop = (dw0 & PCI_EA_SP) >> 16;
2572 if (prop > PCI_EA_P_BRIDGE_IO)
2573 goto out;
2574
2575 res = pci_ea_get_resource(dev, bei, prop);
2576 if (!res) {
2577 pci_err(dev, "Unsupported EA entry BEI: %u\n", bei);
2578 goto out;
2579 }
2580
2581 flags = pci_ea_flags(dev, prop);
2582 if (!flags) {
2583 pci_err(dev, "Unsupported EA properties: %#x\n", prop);
2584 goto out;
2585 }
2586
2587 /* Read Base */
2588 pci_read_config_dword(dev, ent_offset, &base);
2589 start = (base & PCI_EA_FIELD_MASK);
2590 ent_offset += 4;
2591
2592 /* Read MaxOffset */
2593 pci_read_config_dword(dev, ent_offset, &max_offset);
2594 ent_offset += 4;
2595
2596 /* Read Base MSBs (if 64-bit entry) */
2597 if (base & PCI_EA_IS_64) {
2598 u32 base_upper;
2599
2600 pci_read_config_dword(dev, ent_offset, &base_upper);
2601 ent_offset += 4;
2602
2603 flags |= IORESOURCE_MEM_64;
2604
2605 /* entry starts above 32-bit boundary, can't use */
2606 if (!support_64 && base_upper)
2607 goto out;
2608
2609 if (support_64)
2610 start |= ((u64)base_upper << 32);
2611 }
2612
2613 end = start + (max_offset | 0x03);
2614
2615 /* Read MaxOffset MSBs (if 64-bit entry) */
2616 if (max_offset & PCI_EA_IS_64) {
2617 u32 max_offset_upper;
2618
2619 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2620 ent_offset += 4;
2621
2622 flags |= IORESOURCE_MEM_64;
2623
2624 /* entry too big, can't use */
2625 if (!support_64 && max_offset_upper)
2626 goto out;
2627
2628 if (support_64)
2629 end += ((u64)max_offset_upper << 32);
2630 }
2631
2632 if (end < start) {
2633 pci_err(dev, "EA Entry crosses address boundary\n");
2634 goto out;
2635 }
2636
2637 if (ent_size != ent_offset - offset) {
2638 pci_err(dev, "EA Entry Size (%d) does not match length read (%d)\n",
2639 ent_size, ent_offset - offset);
2640 goto out;
2641 }
2642
2643 res->name = pci_name(dev);
2644 res->start = start;
2645 res->end = end;
2646 res->flags = flags;
2647
2648 if (bei <= PCI_EA_BEI_BAR5)
2649 pci_printk(KERN_DEBUG, dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2650 bei, res, prop);
2651 else if (bei == PCI_EA_BEI_ROM)
2652 pci_printk(KERN_DEBUG, dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2653 res, prop);
2654 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2655 pci_printk(KERN_DEBUG, dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2656 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2657 else
2658 pci_printk(KERN_DEBUG, dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2659 bei, res, prop);
2660
2661out:
2662 return offset + ent_size;
2663}
2664
2665/* Enhanced Allocation Initialization */
2666void pci_ea_init(struct pci_dev *dev)
2667{
2668 int ea;
2669 u8 num_ent;
2670 int offset;
2671 int i;
2672
2673 /* find PCI EA capability in list */
2674 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2675 if (!ea)
2676 return;
2677
2678 /* determine the number of entries */
2679 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2680 &num_ent);
2681 num_ent &= PCI_EA_NUM_ENT_MASK;
2682
2683 offset = ea + PCI_EA_FIRST_ENT;
2684
2685 /* Skip DWORD 2 for type 1 functions */
2686 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2687 offset += 4;
2688
2689 /* parse each EA entry */
2690 for (i = 0; i < num_ent; ++i)
2691 offset = pci_ea_read(dev, offset);
2692}
2693
2694static void pci_add_saved_cap(struct pci_dev *pci_dev,
2695 struct pci_cap_saved_state *new_cap)
2696{
2697 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2698}
2699
2700/**
2701 * _pci_add_cap_save_buffer - allocate buffer for saving given
2702 * capability registers
2703 * @dev: the PCI device
2704 * @cap: the capability to allocate the buffer for
2705 * @extended: Standard or Extended capability ID
2706 * @size: requested size of the buffer
2707 */
2708static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2709 bool extended, unsigned int size)
2710{
2711 int pos;
2712 struct pci_cap_saved_state *save_state;
2713
2714 if (extended)
2715 pos = pci_find_ext_capability(dev, cap);
2716 else
2717 pos = pci_find_capability(dev, cap);
2718
2719 if (!pos)
2720 return 0;
2721
2722 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2723 if (!save_state)
2724 return -ENOMEM;
2725
2726 save_state->cap.cap_nr = cap;
2727 save_state->cap.cap_extended = extended;
2728 save_state->cap.size = size;
2729 pci_add_saved_cap(dev, save_state);
2730
2731 return 0;
2732}
2733
2734int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2735{
2736 return _pci_add_cap_save_buffer(dev, cap, false, size);
2737}
2738
2739int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2740{
2741 return _pci_add_cap_save_buffer(dev, cap, true, size);
2742}
2743
2744/**
2745 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2746 * @dev: the PCI device
2747 */
2748void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2749{
2750 int error;
2751
2752 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2753 PCI_EXP_SAVE_REGS * sizeof(u16));
2754 if (error)
2755 pci_err(dev, "unable to preallocate PCI Express save buffer\n");
2756
2757 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2758 if (error)
2759 pci_err(dev, "unable to preallocate PCI-X save buffer\n");
2760
2761 pci_allocate_vc_save_buffers(dev);
2762}
2763
2764void pci_free_cap_save_buffers(struct pci_dev *dev)
2765{
2766 struct pci_cap_saved_state *tmp;
2767 struct hlist_node *n;
2768
2769 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2770 kfree(tmp);
2771}
2772
2773/**
2774 * pci_configure_ari - enable or disable ARI forwarding
2775 * @dev: the PCI device
2776 *
2777 * If @dev and its upstream bridge both support ARI, enable ARI in the
2778 * bridge. Otherwise, disable ARI in the bridge.
2779 */
2780void pci_configure_ari(struct pci_dev *dev)
2781{
2782 u32 cap;
2783 struct pci_dev *bridge;
2784
2785 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2786 return;
2787
2788 bridge = dev->bus->self;
2789 if (!bridge)
2790 return;
2791
2792 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2793 if (!(cap & PCI_EXP_DEVCAP2_ARI))
2794 return;
2795
2796 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2797 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2798 PCI_EXP_DEVCTL2_ARI);
2799 bridge->ari_enabled = 1;
2800 } else {
2801 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2802 PCI_EXP_DEVCTL2_ARI);
2803 bridge->ari_enabled = 0;
2804 }
2805}
2806
2807static int pci_acs_enable;
2808
2809/**
2810 * pci_request_acs - ask for ACS to be enabled if supported
2811 */
2812void pci_request_acs(void)
2813{
2814 pci_acs_enable = 1;
2815}
2816
2817/**
2818 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2819 * @dev: the PCI device
2820 */
2821static void pci_std_enable_acs(struct pci_dev *dev)
2822{
2823 int pos;
2824 u16 cap;
2825 u16 ctrl;
2826
2827 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2828 if (!pos)
2829 return;
2830
2831 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2832 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2833
2834 /* Source Validation */
2835 ctrl |= (cap & PCI_ACS_SV);
2836
2837 /* P2P Request Redirect */
2838 ctrl |= (cap & PCI_ACS_RR);
2839
2840 /* P2P Completion Redirect */
2841 ctrl |= (cap & PCI_ACS_CR);
2842
2843 /* Upstream Forwarding */
2844 ctrl |= (cap & PCI_ACS_UF);
2845
2846 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2847}
2848
2849/**
2850 * pci_enable_acs - enable ACS if hardware support it
2851 * @dev: the PCI device
2852 */
2853void pci_enable_acs(struct pci_dev *dev)
2854{
2855 if (!pci_acs_enable)
2856 return;
2857
2858 if (!pci_dev_specific_enable_acs(dev))
2859 return;
2860
2861 pci_std_enable_acs(dev);
2862}
2863
2864static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2865{
2866 int pos;
2867 u16 cap, ctrl;
2868
2869 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2870 if (!pos)
2871 return false;
2872
2873 /*
2874 * Except for egress control, capabilities are either required
2875 * or only required if controllable. Features missing from the
2876 * capability field can therefore be assumed as hard-wired enabled.
2877 */
2878 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2879 acs_flags &= (cap | PCI_ACS_EC);
2880
2881 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2882 return (ctrl & acs_flags) == acs_flags;
2883}
2884
2885/**
2886 * pci_acs_enabled - test ACS against required flags for a given device
2887 * @pdev: device to test
2888 * @acs_flags: required PCI ACS flags
2889 *
2890 * Return true if the device supports the provided flags. Automatically
2891 * filters out flags that are not implemented on multifunction devices.
2892 *
2893 * Note that this interface checks the effective ACS capabilities of the
2894 * device rather than the actual capabilities. For instance, most single
2895 * function endpoints are not required to support ACS because they have no
2896 * opportunity for peer-to-peer access. We therefore return 'true'
2897 * regardless of whether the device exposes an ACS capability. This makes
2898 * it much easier for callers of this function to ignore the actual type
2899 * or topology of the device when testing ACS support.
2900 */
2901bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2902{
2903 int ret;
2904
2905 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2906 if (ret >= 0)
2907 return ret > 0;
2908
2909 /*
2910 * Conventional PCI and PCI-X devices never support ACS, either
2911 * effectively or actually. The shared bus topology implies that
2912 * any device on the bus can receive or snoop DMA.
2913 */
2914 if (!pci_is_pcie(pdev))
2915 return false;
2916
2917 switch (pci_pcie_type(pdev)) {
2918 /*
2919 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2920 * but since their primary interface is PCI/X, we conservatively
2921 * handle them as we would a non-PCIe device.
2922 */
2923 case PCI_EXP_TYPE_PCIE_BRIDGE:
2924 /*
2925 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
2926 * applicable... must never implement an ACS Extended Capability...".
2927 * This seems arbitrary, but we take a conservative interpretation
2928 * of this statement.
2929 */
2930 case PCI_EXP_TYPE_PCI_BRIDGE:
2931 case PCI_EXP_TYPE_RC_EC:
2932 return false;
2933 /*
2934 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2935 * implement ACS in order to indicate their peer-to-peer capabilities,
2936 * regardless of whether they are single- or multi-function devices.
2937 */
2938 case PCI_EXP_TYPE_DOWNSTREAM:
2939 case PCI_EXP_TYPE_ROOT_PORT:
2940 return pci_acs_flags_enabled(pdev, acs_flags);
2941 /*
2942 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2943 * implemented by the remaining PCIe types to indicate peer-to-peer
2944 * capabilities, but only when they are part of a multifunction
2945 * device. The footnote for section 6.12 indicates the specific
2946 * PCIe types included here.
2947 */
2948 case PCI_EXP_TYPE_ENDPOINT:
2949 case PCI_EXP_TYPE_UPSTREAM:
2950 case PCI_EXP_TYPE_LEG_END:
2951 case PCI_EXP_TYPE_RC_END:
2952 if (!pdev->multifunction)
2953 break;
2954
2955 return pci_acs_flags_enabled(pdev, acs_flags);
2956 }
2957
2958 /*
2959 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2960 * to single function devices with the exception of downstream ports.
2961 */
2962 return true;
2963}
2964
2965/**
2966 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2967 * @start: starting downstream device
2968 * @end: ending upstream device or NULL to search to the root bus
2969 * @acs_flags: required flags
2970 *
2971 * Walk up a device tree from start to end testing PCI ACS support. If
2972 * any step along the way does not support the required flags, return false.
2973 */
2974bool pci_acs_path_enabled(struct pci_dev *start,
2975 struct pci_dev *end, u16 acs_flags)
2976{
2977 struct pci_dev *pdev, *parent = start;
2978
2979 do {
2980 pdev = parent;
2981
2982 if (!pci_acs_enabled(pdev, acs_flags))
2983 return false;
2984
2985 if (pci_is_root_bus(pdev->bus))
2986 return (end == NULL);
2987
2988 parent = pdev->bus->self;
2989 } while (pdev != end);
2990
2991 return true;
2992}
2993
2994/**
2995 * pci_rebar_find_pos - find position of resize ctrl reg for BAR
2996 * @pdev: PCI device
2997 * @bar: BAR to find
2998 *
2999 * Helper to find the position of the ctrl register for a BAR.
3000 * Returns -ENOTSUPP if resizable BARs are not supported at all.
3001 * Returns -ENOENT if no ctrl register for the BAR could be found.
3002 */
3003static int pci_rebar_find_pos(struct pci_dev *pdev, int bar)
3004{
3005 unsigned int pos, nbars, i;
3006 u32 ctrl;
3007
3008 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_REBAR);
3009 if (!pos)
3010 return -ENOTSUPP;
3011
3012 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3013 nbars = (ctrl & PCI_REBAR_CTRL_NBAR_MASK) >>
3014 PCI_REBAR_CTRL_NBAR_SHIFT;
3015
3016 for (i = 0; i < nbars; i++, pos += 8) {
3017 int bar_idx;
3018
3019 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3020 bar_idx = ctrl & PCI_REBAR_CTRL_BAR_IDX;
3021 if (bar_idx == bar)
3022 return pos;
3023 }
3024
3025 return -ENOENT;
3026}
3027
3028/**
3029 * pci_rebar_get_possible_sizes - get possible sizes for BAR
3030 * @pdev: PCI device
3031 * @bar: BAR to query
3032 *
3033 * Get the possible sizes of a resizable BAR as bitmask defined in the spec
3034 * (bit 0=1MB, bit 19=512GB). Returns 0 if BAR isn't resizable.
3035 */
3036u32 pci_rebar_get_possible_sizes(struct pci_dev *pdev, int bar)
3037{
3038 int pos;
3039 u32 cap;
3040
3041 pos = pci_rebar_find_pos(pdev, bar);
3042 if (pos < 0)
3043 return 0;
3044
3045 pci_read_config_dword(pdev, pos + PCI_REBAR_CAP, &cap);
3046 return (cap & PCI_REBAR_CAP_SIZES) >> 4;
3047}
3048
3049/**
3050 * pci_rebar_get_current_size - get the current size of a BAR
3051 * @pdev: PCI device
3052 * @bar: BAR to set size to
3053 *
3054 * Read the size of a BAR from the resizable BAR config.
3055 * Returns size if found or negative error code.
3056 */
3057int pci_rebar_get_current_size(struct pci_dev *pdev, int bar)
3058{
3059 int pos;
3060 u32 ctrl;
3061
3062 pos = pci_rebar_find_pos(pdev, bar);
3063 if (pos < 0)
3064 return pos;
3065
3066 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3067 return (ctrl & PCI_REBAR_CTRL_BAR_SIZE) >> 8;
3068}
3069
3070/**
3071 * pci_rebar_set_size - set a new size for a BAR
3072 * @pdev: PCI device
3073 * @bar: BAR to set size to
3074 * @size: new size as defined in the spec (0=1MB, 19=512GB)
3075 *
3076 * Set the new size of a BAR as defined in the spec.
3077 * Returns zero if resizing was successful, error code otherwise.
3078 */
3079int pci_rebar_set_size(struct pci_dev *pdev, int bar, int size)
3080{
3081 int pos;
3082 u32 ctrl;
3083
3084 pos = pci_rebar_find_pos(pdev, bar);
3085 if (pos < 0)
3086 return pos;
3087
3088 pci_read_config_dword(pdev, pos + PCI_REBAR_CTRL, &ctrl);
3089 ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
3090 ctrl |= size << 8;
3091 pci_write_config_dword(pdev, pos + PCI_REBAR_CTRL, ctrl);
3092 return 0;
3093}
3094
3095/**
3096 * pci_enable_atomic_ops_to_root - enable AtomicOp requests to root port
3097 * @dev: the PCI device
3098 * @cap_mask: mask of desired AtomicOp sizes, including one or more of:
3099 * PCI_EXP_DEVCAP2_ATOMIC_COMP32
3100 * PCI_EXP_DEVCAP2_ATOMIC_COMP64
3101 * PCI_EXP_DEVCAP2_ATOMIC_COMP128
3102 *
3103 * Return 0 if all upstream bridges support AtomicOp routing, egress
3104 * blocking is disabled on all upstream ports, and the root port supports
3105 * the requested completion capabilities (32-bit, 64-bit and/or 128-bit
3106 * AtomicOp completion), or negative otherwise.
3107 */
3108int pci_enable_atomic_ops_to_root(struct pci_dev *dev, u32 cap_mask)
3109{
3110 struct pci_bus *bus = dev->bus;
3111 struct pci_dev *bridge;
3112 u32 cap, ctl2;
3113
3114 if (!pci_is_pcie(dev))
3115 return -EINVAL;
3116
3117 /*
3118 * Per PCIe r4.0, sec 6.15, endpoints and root ports may be
3119 * AtomicOp requesters. For now, we only support endpoints as
3120 * requesters and root ports as completers. No endpoints as
3121 * completers, and no peer-to-peer.
3122 */
3123
3124 switch (pci_pcie_type(dev)) {
3125 case PCI_EXP_TYPE_ENDPOINT:
3126 case PCI_EXP_TYPE_LEG_END:
3127 case PCI_EXP_TYPE_RC_END:
3128 break;
3129 default:
3130 return -EINVAL;
3131 }
3132
3133 while (bus->parent) {
3134 bridge = bus->self;
3135
3136 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
3137
3138 switch (pci_pcie_type(bridge)) {
3139 /* Ensure switch ports support AtomicOp routing */
3140 case PCI_EXP_TYPE_UPSTREAM:
3141 case PCI_EXP_TYPE_DOWNSTREAM:
3142 if (!(cap & PCI_EXP_DEVCAP2_ATOMIC_ROUTE))
3143 return -EINVAL;
3144 break;
3145
3146 /* Ensure root port supports all the sizes we care about */
3147 case PCI_EXP_TYPE_ROOT_PORT:
3148 if ((cap & cap_mask) != cap_mask)
3149 return -EINVAL;
3150 break;
3151 }
3152
3153 /* Ensure upstream ports don't block AtomicOps on egress */
3154 if (!bridge->has_secondary_link) {
3155 pcie_capability_read_dword(bridge, PCI_EXP_DEVCTL2,
3156 &ctl2);
3157 if (ctl2 & PCI_EXP_DEVCTL2_ATOMIC_EGRESS_BLOCK)
3158 return -EINVAL;
3159 }
3160
3161 bus = bus->parent;
3162 }
3163
3164 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
3165 PCI_EXP_DEVCTL2_ATOMIC_REQ);
3166 return 0;
3167}
3168EXPORT_SYMBOL(pci_enable_atomic_ops_to_root);
3169
3170/**
3171 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
3172 * @dev: the PCI device
3173 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
3174 *
3175 * Perform INTx swizzling for a device behind one level of bridge. This is
3176 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
3177 * behind bridges on add-in cards. For devices with ARI enabled, the slot
3178 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
3179 * the PCI Express Base Specification, Revision 2.1)
3180 */
3181u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
3182{
3183 int slot;
3184
3185 if (pci_ari_enabled(dev->bus))
3186 slot = 0;
3187 else
3188 slot = PCI_SLOT(dev->devfn);
3189
3190 return (((pin - 1) + slot) % 4) + 1;
3191}
3192
3193int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3194{
3195 u8 pin;
3196
3197 pin = dev->pin;
3198 if (!pin)
3199 return -1;
3200
3201 while (!pci_is_root_bus(dev->bus)) {
3202 pin = pci_swizzle_interrupt_pin(dev, pin);
3203 dev = dev->bus->self;
3204 }
3205 *bridge = dev;
3206 return pin;
3207}
3208
3209/**
3210 * pci_common_swizzle - swizzle INTx all the way to root bridge
3211 * @dev: the PCI device
3212 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3213 *
3214 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3215 * bridges all the way up to a PCI root bus.
3216 */
3217u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3218{
3219 u8 pin = *pinp;
3220
3221 while (!pci_is_root_bus(dev->bus)) {
3222 pin = pci_swizzle_interrupt_pin(dev, pin);
3223 dev = dev->bus->self;
3224 }
3225 *pinp = pin;
3226 return PCI_SLOT(dev->devfn);
3227}
3228EXPORT_SYMBOL_GPL(pci_common_swizzle);
3229
3230/**
3231 * pci_release_region - Release a PCI bar
3232 * @pdev: PCI device whose resources were previously reserved by pci_request_region
3233 * @bar: BAR to release
3234 *
3235 * Releases the PCI I/O and memory resources previously reserved by a
3236 * successful call to pci_request_region. Call this function only
3237 * after all use of the PCI regions has ceased.
3238 */
3239void pci_release_region(struct pci_dev *pdev, int bar)
3240{
3241 struct pci_devres *dr;
3242
3243 if (pci_resource_len(pdev, bar) == 0)
3244 return;
3245 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3246 release_region(pci_resource_start(pdev, bar),
3247 pci_resource_len(pdev, bar));
3248 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3249 release_mem_region(pci_resource_start(pdev, bar),
3250 pci_resource_len(pdev, bar));
3251
3252 dr = find_pci_dr(pdev);
3253 if (dr)
3254 dr->region_mask &= ~(1 << bar);
3255}
3256EXPORT_SYMBOL(pci_release_region);
3257
3258/**
3259 * __pci_request_region - Reserved PCI I/O and memory resource
3260 * @pdev: PCI device whose resources are to be reserved
3261 * @bar: BAR to be reserved
3262 * @res_name: Name to be associated with resource.
3263 * @exclusive: whether the region access is exclusive or not
3264 *
3265 * Mark the PCI region associated with PCI device @pdev BR @bar as
3266 * being reserved by owner @res_name. Do not access any
3267 * address inside the PCI regions unless this call returns
3268 * successfully.
3269 *
3270 * If @exclusive is set, then the region is marked so that userspace
3271 * is explicitly not allowed to map the resource via /dev/mem or
3272 * sysfs MMIO access.
3273 *
3274 * Returns 0 on success, or %EBUSY on error. A warning
3275 * message is also printed on failure.
3276 */
3277static int __pci_request_region(struct pci_dev *pdev, int bar,
3278 const char *res_name, int exclusive)
3279{
3280 struct pci_devres *dr;
3281
3282 if (pci_resource_len(pdev, bar) == 0)
3283 return 0;
3284
3285 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3286 if (!request_region(pci_resource_start(pdev, bar),
3287 pci_resource_len(pdev, bar), res_name))
3288 goto err_out;
3289 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3290 if (!__request_mem_region(pci_resource_start(pdev, bar),
3291 pci_resource_len(pdev, bar), res_name,
3292 exclusive))
3293 goto err_out;
3294 }
3295
3296 dr = find_pci_dr(pdev);
3297 if (dr)
3298 dr->region_mask |= 1 << bar;
3299
3300 return 0;
3301
3302err_out:
3303 pci_warn(pdev, "BAR %d: can't reserve %pR\n", bar,
3304 &pdev->resource[bar]);
3305 return -EBUSY;
3306}
3307
3308/**
3309 * pci_request_region - Reserve PCI I/O and memory resource
3310 * @pdev: PCI device whose resources are to be reserved
3311 * @bar: BAR to be reserved
3312 * @res_name: Name to be associated with resource
3313 *
3314 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3315 * being reserved by owner @res_name. Do not access any
3316 * address inside the PCI regions unless this call returns
3317 * successfully.
3318 *
3319 * Returns 0 on success, or %EBUSY on error. A warning
3320 * message is also printed on failure.
3321 */
3322int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3323{
3324 return __pci_request_region(pdev, bar, res_name, 0);
3325}
3326EXPORT_SYMBOL(pci_request_region);
3327
3328/**
3329 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
3330 * @pdev: PCI device whose resources are to be reserved
3331 * @bar: BAR to be reserved
3332 * @res_name: Name to be associated with resource.
3333 *
3334 * Mark the PCI region associated with PCI device @pdev BR @bar as
3335 * being reserved by owner @res_name. Do not access any
3336 * address inside the PCI regions unless this call returns
3337 * successfully.
3338 *
3339 * Returns 0 on success, or %EBUSY on error. A warning
3340 * message is also printed on failure.
3341 *
3342 * The key difference that _exclusive makes it that userspace is
3343 * explicitly not allowed to map the resource via /dev/mem or
3344 * sysfs.
3345 */
3346int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3347 const char *res_name)
3348{
3349 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3350}
3351EXPORT_SYMBOL(pci_request_region_exclusive);
3352
3353/**
3354 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3355 * @pdev: PCI device whose resources were previously reserved
3356 * @bars: Bitmask of BARs to be released
3357 *
3358 * Release selected PCI I/O and memory resources previously reserved.
3359 * Call this function only after all use of the PCI regions has ceased.
3360 */
3361void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3362{
3363 int i;
3364
3365 for (i = 0; i < 6; i++)
3366 if (bars & (1 << i))
3367 pci_release_region(pdev, i);
3368}
3369EXPORT_SYMBOL(pci_release_selected_regions);
3370
3371static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3372 const char *res_name, int excl)
3373{
3374 int i;
3375
3376 for (i = 0; i < 6; i++)
3377 if (bars & (1 << i))
3378 if (__pci_request_region(pdev, i, res_name, excl))
3379 goto err_out;
3380 return 0;
3381
3382err_out:
3383 while (--i >= 0)
3384 if (bars & (1 << i))
3385 pci_release_region(pdev, i);
3386
3387 return -EBUSY;
3388}
3389
3390
3391/**
3392 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3393 * @pdev: PCI device whose resources are to be reserved
3394 * @bars: Bitmask of BARs to be requested
3395 * @res_name: Name to be associated with resource
3396 */
3397int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3398 const char *res_name)
3399{
3400 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3401}
3402EXPORT_SYMBOL(pci_request_selected_regions);
3403
3404int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3405 const char *res_name)
3406{
3407 return __pci_request_selected_regions(pdev, bars, res_name,
3408 IORESOURCE_EXCLUSIVE);
3409}
3410EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3411
3412/**
3413 * pci_release_regions - Release reserved PCI I/O and memory resources
3414 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
3415 *
3416 * Releases all PCI I/O and memory resources previously reserved by a
3417 * successful call to pci_request_regions. Call this function only
3418 * after all use of the PCI regions has ceased.
3419 */
3420
3421void pci_release_regions(struct pci_dev *pdev)
3422{
3423 pci_release_selected_regions(pdev, (1 << 6) - 1);
3424}
3425EXPORT_SYMBOL(pci_release_regions);
3426
3427/**
3428 * pci_request_regions - Reserved PCI I/O and memory resources
3429 * @pdev: PCI device whose resources are to be reserved
3430 * @res_name: Name to be associated with resource.
3431 *
3432 * Mark all PCI regions associated with PCI device @pdev as
3433 * being reserved by owner @res_name. Do not access any
3434 * address inside the PCI regions unless this call returns
3435 * successfully.
3436 *
3437 * Returns 0 on success, or %EBUSY on error. A warning
3438 * message is also printed on failure.
3439 */
3440int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3441{
3442 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3443}
3444EXPORT_SYMBOL(pci_request_regions);
3445
3446/**
3447 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3448 * @pdev: PCI device whose resources are to be reserved
3449 * @res_name: Name to be associated with resource.
3450 *
3451 * Mark all PCI regions associated with PCI device @pdev as
3452 * being reserved by owner @res_name. Do not access any
3453 * address inside the PCI regions unless this call returns
3454 * successfully.
3455 *
3456 * pci_request_regions_exclusive() will mark the region so that
3457 * /dev/mem and the sysfs MMIO access will not be allowed.
3458 *
3459 * Returns 0 on success, or %EBUSY on error. A warning
3460 * message is also printed on failure.
3461 */
3462int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3463{
3464 return pci_request_selected_regions_exclusive(pdev,
3465 ((1 << 6) - 1), res_name);
3466}
3467EXPORT_SYMBOL(pci_request_regions_exclusive);
3468
3469/*
3470 * Record the PCI IO range (expressed as CPU physical address + size).
3471 * Return a negative value if an error has occured, zero otherwise
3472 */
3473int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3474 resource_size_t size)
3475{
3476 int ret = 0;
3477#ifdef PCI_IOBASE
3478 struct logic_pio_hwaddr *range;
3479
3480 if (!size || addr + size < addr)
3481 return -EINVAL;
3482
3483 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3484 if (!range)
3485 return -ENOMEM;
3486
3487 range->fwnode = fwnode;
3488 range->size = size;
3489 range->hw_start = addr;
3490 range->flags = LOGIC_PIO_CPU_MMIO;
3491
3492 ret = logic_pio_register_range(range);
3493 if (ret)
3494 kfree(range);
3495#endif
3496
3497 return ret;
3498}
3499
3500phys_addr_t pci_pio_to_address(unsigned long pio)
3501{
3502 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3503
3504#ifdef PCI_IOBASE
3505 if (pio >= MMIO_UPPER_LIMIT)
3506 return address;
3507
3508 address = logic_pio_to_hwaddr(pio);
3509#endif
3510
3511 return address;
3512}
3513
3514unsigned long __weak pci_address_to_pio(phys_addr_t address)
3515{
3516#ifdef PCI_IOBASE
3517 return logic_pio_trans_cpuaddr(address);
3518#else
3519 if (address > IO_SPACE_LIMIT)
3520 return (unsigned long)-1;
3521
3522 return (unsigned long) address;
3523#endif
3524}
3525
3526/**
3527 * pci_remap_iospace - Remap the memory mapped I/O space
3528 * @res: Resource describing the I/O space
3529 * @phys_addr: physical address of range to be mapped
3530 *
3531 * Remap the memory mapped I/O space described by the @res
3532 * and the CPU physical address @phys_addr into virtual address space.
3533 * Only architectures that have memory mapped IO functions defined
3534 * (and the PCI_IOBASE value defined) should call this function.
3535 */
3536int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3537{
3538#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3539 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3540
3541 if (!(res->flags & IORESOURCE_IO))
3542 return -EINVAL;
3543
3544 if (res->end > IO_SPACE_LIMIT)
3545 return -EINVAL;
3546
3547 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3548 pgprot_device(PAGE_KERNEL));
3549#else
3550 /* this architecture does not have memory mapped I/O space,
3551 so this function should never be called */
3552 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3553 return -ENODEV;
3554#endif
3555}
3556EXPORT_SYMBOL(pci_remap_iospace);
3557
3558/**
3559 * pci_unmap_iospace - Unmap the memory mapped I/O space
3560 * @res: resource to be unmapped
3561 *
3562 * Unmap the CPU virtual address @res from virtual address space.
3563 * Only architectures that have memory mapped IO functions defined
3564 * (and the PCI_IOBASE value defined) should call this function.
3565 */
3566void pci_unmap_iospace(struct resource *res)
3567{
3568#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3569 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3570
3571 unmap_kernel_range(vaddr, resource_size(res));
3572#endif
3573}
3574EXPORT_SYMBOL(pci_unmap_iospace);
3575
3576/**
3577 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3578 * @dev: Generic device to remap IO address for
3579 * @offset: Resource address to map
3580 * @size: Size of map
3581 *
3582 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
3583 * detach.
3584 */
3585void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3586 resource_size_t offset,
3587 resource_size_t size)
3588{
3589 void __iomem **ptr, *addr;
3590
3591 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
3592 if (!ptr)
3593 return NULL;
3594
3595 addr = pci_remap_cfgspace(offset, size);
3596 if (addr) {
3597 *ptr = addr;
3598 devres_add(dev, ptr);
3599 } else
3600 devres_free(ptr);
3601
3602 return addr;
3603}
3604EXPORT_SYMBOL(devm_pci_remap_cfgspace);
3605
3606/**
3607 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
3608 * @dev: generic device to handle the resource for
3609 * @res: configuration space resource to be handled
3610 *
3611 * Checks that a resource is a valid memory region, requests the memory
3612 * region and ioremaps with pci_remap_cfgspace() API that ensures the
3613 * proper PCI configuration space memory attributes are guaranteed.
3614 *
3615 * All operations are managed and will be undone on driver detach.
3616 *
3617 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
3618 * on failure. Usage example::
3619 *
3620 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
3621 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
3622 * if (IS_ERR(base))
3623 * return PTR_ERR(base);
3624 */
3625void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
3626 struct resource *res)
3627{
3628 resource_size_t size;
3629 const char *name;
3630 void __iomem *dest_ptr;
3631
3632 BUG_ON(!dev);
3633
3634 if (!res || resource_type(res) != IORESOURCE_MEM) {
3635 dev_err(dev, "invalid resource\n");
3636 return IOMEM_ERR_PTR(-EINVAL);
3637 }
3638
3639 size = resource_size(res);
3640 name = res->name ?: dev_name(dev);
3641
3642 if (!devm_request_mem_region(dev, res->start, size, name)) {
3643 dev_err(dev, "can't request region for resource %pR\n", res);
3644 return IOMEM_ERR_PTR(-EBUSY);
3645 }
3646
3647 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
3648 if (!dest_ptr) {
3649 dev_err(dev, "ioremap failed for resource %pR\n", res);
3650 devm_release_mem_region(dev, res->start, size);
3651 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
3652 }
3653
3654 return dest_ptr;
3655}
3656EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
3657
3658static void __pci_set_master(struct pci_dev *dev, bool enable)
3659{
3660 u16 old_cmd, cmd;
3661
3662 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3663 if (enable)
3664 cmd = old_cmd | PCI_COMMAND_MASTER;
3665 else
3666 cmd = old_cmd & ~PCI_COMMAND_MASTER;
3667 if (cmd != old_cmd) {
3668 pci_dbg(dev, "%s bus mastering\n",
3669 enable ? "enabling" : "disabling");
3670 pci_write_config_word(dev, PCI_COMMAND, cmd);
3671 }
3672 dev->is_busmaster = enable;
3673}
3674
3675/**
3676 * pcibios_setup - process "pci=" kernel boot arguments
3677 * @str: string used to pass in "pci=" kernel boot arguments
3678 *
3679 * Process kernel boot arguments. This is the default implementation.
3680 * Architecture specific implementations can override this as necessary.
3681 */
3682char * __weak __init pcibios_setup(char *str)
3683{
3684 return str;
3685}
3686
3687/**
3688 * pcibios_set_master - enable PCI bus-mastering for device dev
3689 * @dev: the PCI device to enable
3690 *
3691 * Enables PCI bus-mastering for the device. This is the default
3692 * implementation. Architecture specific implementations can override
3693 * this if necessary.
3694 */
3695void __weak pcibios_set_master(struct pci_dev *dev)
3696{
3697 u8 lat;
3698
3699 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3700 if (pci_is_pcie(dev))
3701 return;
3702
3703 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3704 if (lat < 16)
3705 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3706 else if (lat > pcibios_max_latency)
3707 lat = pcibios_max_latency;
3708 else
3709 return;
3710
3711 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3712}
3713
3714/**
3715 * pci_set_master - enables bus-mastering for device dev
3716 * @dev: the PCI device to enable
3717 *
3718 * Enables bus-mastering on the device and calls pcibios_set_master()
3719 * to do the needed arch specific settings.
3720 */
3721void pci_set_master(struct pci_dev *dev)
3722{
3723 __pci_set_master(dev, true);
3724 pcibios_set_master(dev);
3725}
3726EXPORT_SYMBOL(pci_set_master);
3727
3728/**
3729 * pci_clear_master - disables bus-mastering for device dev
3730 * @dev: the PCI device to disable
3731 */
3732void pci_clear_master(struct pci_dev *dev)
3733{
3734 __pci_set_master(dev, false);
3735}
3736EXPORT_SYMBOL(pci_clear_master);
3737
3738/**
3739 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3740 * @dev: the PCI device for which MWI is to be enabled
3741 *
3742 * Helper function for pci_set_mwi.
3743 * Originally copied from drivers/net/acenic.c.
3744 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3745 *
3746 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3747 */
3748int pci_set_cacheline_size(struct pci_dev *dev)
3749{
3750 u8 cacheline_size;
3751
3752 if (!pci_cache_line_size)
3753 return -EINVAL;
3754
3755 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3756 equal to or multiple of the right value. */
3757 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3758 if (cacheline_size >= pci_cache_line_size &&
3759 (cacheline_size % pci_cache_line_size) == 0)
3760 return 0;
3761
3762 /* Write the correct value. */
3763 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3764 /* Read it back. */
3765 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3766 if (cacheline_size == pci_cache_line_size)
3767 return 0;
3768
3769 pci_printk(KERN_DEBUG, dev, "cache line size of %d is not supported\n",
3770 pci_cache_line_size << 2);
3771
3772 return -EINVAL;
3773}
3774EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3775
3776/**
3777 * pci_set_mwi - enables memory-write-invalidate PCI transaction
3778 * @dev: the PCI device for which MWI is enabled
3779 *
3780 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3781 *
3782 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3783 */
3784int pci_set_mwi(struct pci_dev *dev)
3785{
3786#ifdef PCI_DISABLE_MWI
3787 return 0;
3788#else
3789 int rc;
3790 u16 cmd;
3791
3792 rc = pci_set_cacheline_size(dev);
3793 if (rc)
3794 return rc;
3795
3796 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3797 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3798 pci_dbg(dev, "enabling Mem-Wr-Inval\n");
3799 cmd |= PCI_COMMAND_INVALIDATE;
3800 pci_write_config_word(dev, PCI_COMMAND, cmd);
3801 }
3802 return 0;
3803#endif
3804}
3805EXPORT_SYMBOL(pci_set_mwi);
3806
3807/**
3808 * pcim_set_mwi - a device-managed pci_set_mwi()
3809 * @dev: the PCI device for which MWI is enabled
3810 *
3811 * Managed pci_set_mwi().
3812 *
3813 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3814 */
3815int pcim_set_mwi(struct pci_dev *dev)
3816{
3817 struct pci_devres *dr;
3818
3819 dr = find_pci_dr(dev);
3820 if (!dr)
3821 return -ENOMEM;
3822
3823 dr->mwi = 1;
3824 return pci_set_mwi(dev);
3825}
3826EXPORT_SYMBOL(pcim_set_mwi);
3827
3828/**
3829 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3830 * @dev: the PCI device for which MWI is enabled
3831 *
3832 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3833 * Callers are not required to check the return value.
3834 *
3835 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3836 */
3837int pci_try_set_mwi(struct pci_dev *dev)
3838{
3839#ifdef PCI_DISABLE_MWI
3840 return 0;
3841#else
3842 return pci_set_mwi(dev);
3843#endif
3844}
3845EXPORT_SYMBOL(pci_try_set_mwi);
3846
3847/**
3848 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3849 * @dev: the PCI device to disable
3850 *
3851 * Disables PCI Memory-Write-Invalidate transaction on the device
3852 */
3853void pci_clear_mwi(struct pci_dev *dev)
3854{
3855#ifndef PCI_DISABLE_MWI
3856 u16 cmd;
3857
3858 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3859 if (cmd & PCI_COMMAND_INVALIDATE) {
3860 cmd &= ~PCI_COMMAND_INVALIDATE;
3861 pci_write_config_word(dev, PCI_COMMAND, cmd);
3862 }
3863#endif
3864}
3865EXPORT_SYMBOL(pci_clear_mwi);
3866
3867/**
3868 * pci_intx - enables/disables PCI INTx for device dev
3869 * @pdev: the PCI device to operate on
3870 * @enable: boolean: whether to enable or disable PCI INTx
3871 *
3872 * Enables/disables PCI INTx for device dev
3873 */
3874void pci_intx(struct pci_dev *pdev, int enable)
3875{
3876 u16 pci_command, new;
3877
3878 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3879
3880 if (enable)
3881 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3882 else
3883 new = pci_command | PCI_COMMAND_INTX_DISABLE;
3884
3885 if (new != pci_command) {
3886 struct pci_devres *dr;
3887
3888 pci_write_config_word(pdev, PCI_COMMAND, new);
3889
3890 dr = find_pci_dr(pdev);
3891 if (dr && !dr->restore_intx) {
3892 dr->restore_intx = 1;
3893 dr->orig_intx = !enable;
3894 }
3895 }
3896}
3897EXPORT_SYMBOL_GPL(pci_intx);
3898
3899static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3900{
3901 struct pci_bus *bus = dev->bus;
3902 bool mask_updated = true;
3903 u32 cmd_status_dword;
3904 u16 origcmd, newcmd;
3905 unsigned long flags;
3906 bool irq_pending;
3907
3908 /*
3909 * We do a single dword read to retrieve both command and status.
3910 * Document assumptions that make this possible.
3911 */
3912 BUILD_BUG_ON(PCI_COMMAND % 4);
3913 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3914
3915 raw_spin_lock_irqsave(&pci_lock, flags);
3916
3917 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3918
3919 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3920
3921 /*
3922 * Check interrupt status register to see whether our device
3923 * triggered the interrupt (when masking) or the next IRQ is
3924 * already pending (when unmasking).
3925 */
3926 if (mask != irq_pending) {
3927 mask_updated = false;
3928 goto done;
3929 }
3930
3931 origcmd = cmd_status_dword;
3932 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3933 if (mask)
3934 newcmd |= PCI_COMMAND_INTX_DISABLE;
3935 if (newcmd != origcmd)
3936 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3937
3938done:
3939 raw_spin_unlock_irqrestore(&pci_lock, flags);
3940
3941 return mask_updated;
3942}
3943
3944/**
3945 * pci_check_and_mask_intx - mask INTx on pending interrupt
3946 * @dev: the PCI device to operate on
3947 *
3948 * Check if the device dev has its INTx line asserted, mask it and
3949 * return true in that case. False is returned if no interrupt was
3950 * pending.
3951 */
3952bool pci_check_and_mask_intx(struct pci_dev *dev)
3953{
3954 return pci_check_and_set_intx_mask(dev, true);
3955}
3956EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3957
3958/**
3959 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3960 * @dev: the PCI device to operate on
3961 *
3962 * Check if the device dev has its INTx line asserted, unmask it if not
3963 * and return true. False is returned and the mask remains active if
3964 * there was still an interrupt pending.
3965 */
3966bool pci_check_and_unmask_intx(struct pci_dev *dev)
3967{
3968 return pci_check_and_set_intx_mask(dev, false);
3969}
3970EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3971
3972/**
3973 * pci_wait_for_pending_transaction - waits for pending transaction
3974 * @dev: the PCI device to operate on
3975 *
3976 * Return 0 if transaction is pending 1 otherwise.
3977 */
3978int pci_wait_for_pending_transaction(struct pci_dev *dev)
3979{
3980 if (!pci_is_pcie(dev))
3981 return 1;
3982
3983 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3984 PCI_EXP_DEVSTA_TRPND);
3985}
3986EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3987
3988static int pci_dev_wait(struct pci_dev *dev, char *reset_type, int timeout)
3989{
3990 int delay = 1;
3991 u32 id;
3992
3993 /*
3994 * After reset, the device should not silently discard config
3995 * requests, but it may still indicate that it needs more time by
3996 * responding to them with CRS completions. The Root Port will
3997 * generally synthesize ~0 data to complete the read (except when
3998 * CRS SV is enabled and the read was for the Vendor ID; in that
3999 * case it synthesizes 0x0001 data).
4000 *
4001 * Wait for the device to return a non-CRS completion. Read the
4002 * Command register instead of Vendor ID so we don't have to
4003 * contend with the CRS SV value.
4004 */
4005 pci_read_config_dword(dev, PCI_COMMAND, &id);
4006 while (id == ~0) {
4007 if (delay > timeout) {
4008 pci_warn(dev, "not ready %dms after %s; giving up\n",
4009 delay - 1, reset_type);
4010 return -ENOTTY;
4011 }
4012
4013 if (delay > 1000)
4014 pci_info(dev, "not ready %dms after %s; waiting\n",
4015 delay - 1, reset_type);
4016
4017 msleep(delay);
4018 delay *= 2;
4019 pci_read_config_dword(dev, PCI_COMMAND, &id);
4020 }
4021
4022 if (delay > 1000)
4023 pci_info(dev, "ready %dms after %s\n", delay - 1,
4024 reset_type);
4025
4026 return 0;
4027}
4028
4029/**
4030 * pcie_has_flr - check if a device supports function level resets
4031 * @dev: device to check
4032 *
4033 * Returns true if the device advertises support for PCIe function level
4034 * resets.
4035 */
4036static bool pcie_has_flr(struct pci_dev *dev)
4037{
4038 u32 cap;
4039
4040 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4041 return false;
4042
4043 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
4044 return cap & PCI_EXP_DEVCAP_FLR;
4045}
4046
4047/**
4048 * pcie_flr - initiate a PCIe function level reset
4049 * @dev: device to reset
4050 *
4051 * Initiate a function level reset on @dev. The caller should ensure the
4052 * device supports FLR before calling this function, e.g. by using the
4053 * pcie_has_flr() helper.
4054 */
4055int pcie_flr(struct pci_dev *dev)
4056{
4057 if (!pci_wait_for_pending_transaction(dev))
4058 pci_err(dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
4059
4060 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
4061
4062 /*
4063 * Per PCIe r4.0, sec 6.6.2, a device must complete an FLR within
4064 * 100ms, but may silently discard requests while the FLR is in
4065 * progress. Wait 100ms before trying to access the device.
4066 */
4067 msleep(100);
4068
4069 return pci_dev_wait(dev, "FLR", PCIE_RESET_READY_POLL_MS);
4070}
4071EXPORT_SYMBOL_GPL(pcie_flr);
4072
4073static int pci_af_flr(struct pci_dev *dev, int probe)
4074{
4075 int pos;
4076 u8 cap;
4077
4078 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
4079 if (!pos)
4080 return -ENOTTY;
4081
4082 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
4083 return -ENOTTY;
4084
4085 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
4086 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
4087 return -ENOTTY;
4088
4089 if (probe)
4090 return 0;
4091
4092 /*
4093 * Wait for Transaction Pending bit to clear. A word-aligned test
4094 * is used, so we use the conrol offset rather than status and shift
4095 * the test bit to match.
4096 */
4097 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
4098 PCI_AF_STATUS_TP << 8))
4099 pci_err(dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
4100
4101 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
4102
4103 /*
4104 * Per Advanced Capabilities for Conventional PCI ECN, 13 April 2006,
4105 * updated 27 July 2006; a device must complete an FLR within
4106 * 100ms, but may silently discard requests while the FLR is in
4107 * progress. Wait 100ms before trying to access the device.
4108 */
4109 msleep(100);
4110
4111 return pci_dev_wait(dev, "AF_FLR", PCIE_RESET_READY_POLL_MS);
4112}
4113
4114/**
4115 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
4116 * @dev: Device to reset.
4117 * @probe: If set, only check if the device can be reset this way.
4118 *
4119 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
4120 * unset, it will be reinitialized internally when going from PCI_D3hot to
4121 * PCI_D0. If that's the case and the device is not in a low-power state
4122 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
4123 *
4124 * NOTE: This causes the caller to sleep for twice the device power transition
4125 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
4126 * by default (i.e. unless the @dev's d3_delay field has a different value).
4127 * Moreover, only devices in D0 can be reset by this function.
4128 */
4129static int pci_pm_reset(struct pci_dev *dev, int probe)
4130{
4131 u16 csr;
4132
4133 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
4134 return -ENOTTY;
4135
4136 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
4137 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
4138 return -ENOTTY;
4139
4140 if (probe)
4141 return 0;
4142
4143 if (dev->current_state != PCI_D0)
4144 return -EINVAL;
4145
4146 csr &= ~PCI_PM_CTRL_STATE_MASK;
4147 csr |= PCI_D3hot;
4148 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4149 pci_dev_d3_sleep(dev);
4150
4151 csr &= ~PCI_PM_CTRL_STATE_MASK;
4152 csr |= PCI_D0;
4153 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4154 pci_dev_d3_sleep(dev);
4155
4156 return pci_dev_wait(dev, "PM D3->D0", PCIE_RESET_READY_POLL_MS);
4157}
4158
4159void pci_reset_secondary_bus(struct pci_dev *dev)
4160{
4161 u16 ctrl;
4162
4163 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4164 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4165 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4166
4167 /*
4168 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4169 * this to 2ms to ensure that we meet the minimum requirement.
4170 */
4171 msleep(2);
4172
4173 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4174 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4175
4176 /*
4177 * Trhfa for conventional PCI is 2^25 clock cycles.
4178 * Assuming a minimum 33MHz clock this results in a 1s
4179 * delay before we can consider subordinate devices to
4180 * be re-initialized. PCIe has some ways to shorten this,
4181 * but we don't make use of them yet.
4182 */
4183 ssleep(1);
4184}
4185
4186void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4187{
4188 pci_reset_secondary_bus(dev);
4189}
4190
4191/**
4192 * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
4193 * @dev: Bridge device
4194 *
4195 * Use the bridge control register to assert reset on the secondary bus.
4196 * Devices on the secondary bus are left in power-on state.
4197 */
4198int pci_reset_bridge_secondary_bus(struct pci_dev *dev)
4199{
4200 pcibios_reset_secondary_bus(dev);
4201
4202 return pci_dev_wait(dev, "bus reset", PCIE_RESET_READY_POLL_MS);
4203}
4204EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
4205
4206static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4207{
4208 struct pci_dev *pdev;
4209
4210 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4211 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4212 return -ENOTTY;
4213
4214 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4215 if (pdev != dev)
4216 return -ENOTTY;
4217
4218 if (probe)
4219 return 0;
4220
4221 pci_reset_bridge_secondary_bus(dev->bus->self);
4222
4223 return 0;
4224}
4225
4226static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4227{
4228 int rc = -ENOTTY;
4229
4230 if (!hotplug || !try_module_get(hotplug->ops->owner))
4231 return rc;
4232
4233 if (hotplug->ops->reset_slot)
4234 rc = hotplug->ops->reset_slot(hotplug, probe);
4235
4236 module_put(hotplug->ops->owner);
4237
4238 return rc;
4239}
4240
4241static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4242{
4243 struct pci_dev *pdev;
4244
4245 if (dev->subordinate || !dev->slot ||
4246 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4247 return -ENOTTY;
4248
4249 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4250 if (pdev != dev && pdev->slot == dev->slot)
4251 return -ENOTTY;
4252
4253 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4254}
4255
4256static void pci_dev_lock(struct pci_dev *dev)
4257{
4258 pci_cfg_access_lock(dev);
4259 /* block PM suspend, driver probe, etc. */
4260 device_lock(&dev->dev);
4261}
4262
4263/* Return 1 on successful lock, 0 on contention */
4264static int pci_dev_trylock(struct pci_dev *dev)
4265{
4266 if (pci_cfg_access_trylock(dev)) {
4267 if (device_trylock(&dev->dev))
4268 return 1;
4269 pci_cfg_access_unlock(dev);
4270 }
4271
4272 return 0;
4273}
4274
4275static void pci_dev_unlock(struct pci_dev *dev)
4276{
4277 device_unlock(&dev->dev);
4278 pci_cfg_access_unlock(dev);
4279}
4280
4281static void pci_dev_save_and_disable(struct pci_dev *dev)
4282{
4283 const struct pci_error_handlers *err_handler =
4284 dev->driver ? dev->driver->err_handler : NULL;
4285
4286 /*
4287 * dev->driver->err_handler->reset_prepare() is protected against
4288 * races with ->remove() by the device lock, which must be held by
4289 * the caller.
4290 */
4291 if (err_handler && err_handler->reset_prepare)
4292 err_handler->reset_prepare(dev);
4293
4294 /*
4295 * Wake-up device prior to save. PM registers default to D0 after
4296 * reset and a simple register restore doesn't reliably return
4297 * to a non-D0 state anyway.
4298 */
4299 pci_set_power_state(dev, PCI_D0);
4300
4301 pci_save_state(dev);
4302 /*
4303 * Disable the device by clearing the Command register, except for
4304 * INTx-disable which is set. This not only disables MMIO and I/O port
4305 * BARs, but also prevents the device from being Bus Master, preventing
4306 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4307 * compliant devices, INTx-disable prevents legacy interrupts.
4308 */
4309 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4310}
4311
4312static void pci_dev_restore(struct pci_dev *dev)
4313{
4314 const struct pci_error_handlers *err_handler =
4315 dev->driver ? dev->driver->err_handler : NULL;
4316
4317 pci_restore_state(dev);
4318
4319 /*
4320 * dev->driver->err_handler->reset_done() is protected against
4321 * races with ->remove() by the device lock, which must be held by
4322 * the caller.
4323 */
4324 if (err_handler && err_handler->reset_done)
4325 err_handler->reset_done(dev);
4326}
4327
4328/**
4329 * __pci_reset_function_locked - reset a PCI device function while holding
4330 * the @dev mutex lock.
4331 * @dev: PCI device to reset
4332 *
4333 * Some devices allow an individual function to be reset without affecting
4334 * other functions in the same device. The PCI device must be responsive
4335 * to PCI config space in order to use this function.
4336 *
4337 * The device function is presumed to be unused and the caller is holding
4338 * the device mutex lock when this function is called.
4339 * Resetting the device will make the contents of PCI configuration space
4340 * random, so any caller of this must be prepared to reinitialise the
4341 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4342 * etc.
4343 *
4344 * Returns 0 if the device function was successfully reset or negative if the
4345 * device doesn't support resetting a single function.
4346 */
4347int __pci_reset_function_locked(struct pci_dev *dev)
4348{
4349 int rc;
4350
4351 might_sleep();
4352
4353 /*
4354 * A reset method returns -ENOTTY if it doesn't support this device
4355 * and we should try the next method.
4356 *
4357 * If it returns 0 (success), we're finished. If it returns any
4358 * other error, we're also finished: this indicates that further
4359 * reset mechanisms might be broken on the device.
4360 */
4361 rc = pci_dev_specific_reset(dev, 0);
4362 if (rc != -ENOTTY)
4363 return rc;
4364 if (pcie_has_flr(dev)) {
4365 rc = pcie_flr(dev);
4366 if (rc != -ENOTTY)
4367 return rc;
4368 }
4369 rc = pci_af_flr(dev, 0);
4370 if (rc != -ENOTTY)
4371 return rc;
4372 rc = pci_pm_reset(dev, 0);
4373 if (rc != -ENOTTY)
4374 return rc;
4375 rc = pci_dev_reset_slot_function(dev, 0);
4376 if (rc != -ENOTTY)
4377 return rc;
4378 return pci_parent_bus_reset(dev, 0);
4379}
4380EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4381
4382/**
4383 * pci_probe_reset_function - check whether the device can be safely reset
4384 * @dev: PCI device to reset
4385 *
4386 * Some devices allow an individual function to be reset without affecting
4387 * other functions in the same device. The PCI device must be responsive
4388 * to PCI config space in order to use this function.
4389 *
4390 * Returns 0 if the device function can be reset or negative if the
4391 * device doesn't support resetting a single function.
4392 */
4393int pci_probe_reset_function(struct pci_dev *dev)
4394{
4395 int rc;
4396
4397 might_sleep();
4398
4399 rc = pci_dev_specific_reset(dev, 1);
4400 if (rc != -ENOTTY)
4401 return rc;
4402 if (pcie_has_flr(dev))
4403 return 0;
4404 rc = pci_af_flr(dev, 1);
4405 if (rc != -ENOTTY)
4406 return rc;
4407 rc = pci_pm_reset(dev, 1);
4408 if (rc != -ENOTTY)
4409 return rc;
4410 rc = pci_dev_reset_slot_function(dev, 1);
4411 if (rc != -ENOTTY)
4412 return rc;
4413
4414 return pci_parent_bus_reset(dev, 1);
4415}
4416
4417/**
4418 * pci_reset_function - quiesce and reset a PCI device function
4419 * @dev: PCI device to reset
4420 *
4421 * Some devices allow an individual function to be reset without affecting
4422 * other functions in the same device. The PCI device must be responsive
4423 * to PCI config space in order to use this function.
4424 *
4425 * This function does not just reset the PCI portion of a device, but
4426 * clears all the state associated with the device. This function differs
4427 * from __pci_reset_function_locked() in that it saves and restores device state
4428 * over the reset and takes the PCI device lock.
4429 *
4430 * Returns 0 if the device function was successfully reset or negative if the
4431 * device doesn't support resetting a single function.
4432 */
4433int pci_reset_function(struct pci_dev *dev)
4434{
4435 int rc;
4436
4437 if (!dev->reset_fn)
4438 return -ENOTTY;
4439
4440 pci_dev_lock(dev);
4441 pci_dev_save_and_disable(dev);
4442
4443 rc = __pci_reset_function_locked(dev);
4444
4445 pci_dev_restore(dev);
4446 pci_dev_unlock(dev);
4447
4448 return rc;
4449}
4450EXPORT_SYMBOL_GPL(pci_reset_function);
4451
4452/**
4453 * pci_reset_function_locked - quiesce and reset a PCI device function
4454 * @dev: PCI device to reset
4455 *
4456 * Some devices allow an individual function to be reset without affecting
4457 * other functions in the same device. The PCI device must be responsive
4458 * to PCI config space in order to use this function.
4459 *
4460 * This function does not just reset the PCI portion of a device, but
4461 * clears all the state associated with the device. This function differs
4462 * from __pci_reset_function_locked() in that it saves and restores device state
4463 * over the reset. It also differs from pci_reset_function() in that it
4464 * requires the PCI device lock to be held.
4465 *
4466 * Returns 0 if the device function was successfully reset or negative if the
4467 * device doesn't support resetting a single function.
4468 */
4469int pci_reset_function_locked(struct pci_dev *dev)
4470{
4471 int rc;
4472
4473 if (!dev->reset_fn)
4474 return -ENOTTY;
4475
4476 pci_dev_save_and_disable(dev);
4477
4478 rc = __pci_reset_function_locked(dev);
4479
4480 pci_dev_restore(dev);
4481
4482 return rc;
4483}
4484EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4485
4486/**
4487 * pci_try_reset_function - quiesce and reset a PCI device function
4488 * @dev: PCI device to reset
4489 *
4490 * Same as above, except return -EAGAIN if unable to lock device.
4491 */
4492int pci_try_reset_function(struct pci_dev *dev)
4493{
4494 int rc;
4495
4496 if (!dev->reset_fn)
4497 return -ENOTTY;
4498
4499 if (!pci_dev_trylock(dev))
4500 return -EAGAIN;
4501
4502 pci_dev_save_and_disable(dev);
4503 rc = __pci_reset_function_locked(dev);
4504 pci_dev_restore(dev);
4505 pci_dev_unlock(dev);
4506
4507 return rc;
4508}
4509EXPORT_SYMBOL_GPL(pci_try_reset_function);
4510
4511/* Do any devices on or below this bus prevent a bus reset? */
4512static bool pci_bus_resetable(struct pci_bus *bus)
4513{
4514 struct pci_dev *dev;
4515
4516
4517 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4518 return false;
4519
4520 list_for_each_entry(dev, &bus->devices, bus_list) {
4521 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4522 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4523 return false;
4524 }
4525
4526 return true;
4527}
4528
4529/* Lock devices from the top of the tree down */
4530static void pci_bus_lock(struct pci_bus *bus)
4531{
4532 struct pci_dev *dev;
4533
4534 list_for_each_entry(dev, &bus->devices, bus_list) {
4535 pci_dev_lock(dev);
4536 if (dev->subordinate)
4537 pci_bus_lock(dev->subordinate);
4538 }
4539}
4540
4541/* Unlock devices from the bottom of the tree up */
4542static void pci_bus_unlock(struct pci_bus *bus)
4543{
4544 struct pci_dev *dev;
4545
4546 list_for_each_entry(dev, &bus->devices, bus_list) {
4547 if (dev->subordinate)
4548 pci_bus_unlock(dev->subordinate);
4549 pci_dev_unlock(dev);
4550 }
4551}
4552
4553/* Return 1 on successful lock, 0 on contention */
4554static int pci_bus_trylock(struct pci_bus *bus)
4555{
4556 struct pci_dev *dev;
4557
4558 list_for_each_entry(dev, &bus->devices, bus_list) {
4559 if (!pci_dev_trylock(dev))
4560 goto unlock;
4561 if (dev->subordinate) {
4562 if (!pci_bus_trylock(dev->subordinate)) {
4563 pci_dev_unlock(dev);
4564 goto unlock;
4565 }
4566 }
4567 }
4568 return 1;
4569
4570unlock:
4571 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4572 if (dev->subordinate)
4573 pci_bus_unlock(dev->subordinate);
4574 pci_dev_unlock(dev);
4575 }
4576 return 0;
4577}
4578
4579/* Do any devices on or below this slot prevent a bus reset? */
4580static bool pci_slot_resetable(struct pci_slot *slot)
4581{
4582 struct pci_dev *dev;
4583
4584 if (slot->bus->self &&
4585 (slot->bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4586 return false;
4587
4588 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4589 if (!dev->slot || dev->slot != slot)
4590 continue;
4591 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4592 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4593 return false;
4594 }
4595
4596 return true;
4597}
4598
4599/* Lock devices from the top of the tree down */
4600static void pci_slot_lock(struct pci_slot *slot)
4601{
4602 struct pci_dev *dev;
4603
4604 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4605 if (!dev->slot || dev->slot != slot)
4606 continue;
4607 pci_dev_lock(dev);
4608 if (dev->subordinate)
4609 pci_bus_lock(dev->subordinate);
4610 }
4611}
4612
4613/* Unlock devices from the bottom of the tree up */
4614static void pci_slot_unlock(struct pci_slot *slot)
4615{
4616 struct pci_dev *dev;
4617
4618 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4619 if (!dev->slot || dev->slot != slot)
4620 continue;
4621 if (dev->subordinate)
4622 pci_bus_unlock(dev->subordinate);
4623 pci_dev_unlock(dev);
4624 }
4625}
4626
4627/* Return 1 on successful lock, 0 on contention */
4628static int pci_slot_trylock(struct pci_slot *slot)
4629{
4630 struct pci_dev *dev;
4631
4632 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4633 if (!dev->slot || dev->slot != slot)
4634 continue;
4635 if (!pci_dev_trylock(dev))
4636 goto unlock;
4637 if (dev->subordinate) {
4638 if (!pci_bus_trylock(dev->subordinate)) {
4639 pci_dev_unlock(dev);
4640 goto unlock;
4641 }
4642 }
4643 }
4644 return 1;
4645
4646unlock:
4647 list_for_each_entry_continue_reverse(dev,
4648 &slot->bus->devices, bus_list) {
4649 if (!dev->slot || dev->slot != slot)
4650 continue;
4651 if (dev->subordinate)
4652 pci_bus_unlock(dev->subordinate);
4653 pci_dev_unlock(dev);
4654 }
4655 return 0;
4656}
4657
4658/* Save and disable devices from the top of the tree down */
4659static void pci_bus_save_and_disable(struct pci_bus *bus)
4660{
4661 struct pci_dev *dev;
4662
4663 list_for_each_entry(dev, &bus->devices, bus_list) {
4664 pci_dev_lock(dev);
4665 pci_dev_save_and_disable(dev);
4666 pci_dev_unlock(dev);
4667 if (dev->subordinate)
4668 pci_bus_save_and_disable(dev->subordinate);
4669 }
4670}
4671
4672/*
4673 * Restore devices from top of the tree down - parent bridges need to be
4674 * restored before we can get to subordinate devices.
4675 */
4676static void pci_bus_restore(struct pci_bus *bus)
4677{
4678 struct pci_dev *dev;
4679
4680 list_for_each_entry(dev, &bus->devices, bus_list) {
4681 pci_dev_lock(dev);
4682 pci_dev_restore(dev);
4683 pci_dev_unlock(dev);
4684 if (dev->subordinate)
4685 pci_bus_restore(dev->subordinate);
4686 }
4687}
4688
4689/* Save and disable devices from the top of the tree down */
4690static void pci_slot_save_and_disable(struct pci_slot *slot)
4691{
4692 struct pci_dev *dev;
4693
4694 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4695 if (!dev->slot || dev->slot != slot)
4696 continue;
4697 pci_dev_save_and_disable(dev);
4698 if (dev->subordinate)
4699 pci_bus_save_and_disable(dev->subordinate);
4700 }
4701}
4702
4703/*
4704 * Restore devices from top of the tree down - parent bridges need to be
4705 * restored before we can get to subordinate devices.
4706 */
4707static void pci_slot_restore(struct pci_slot *slot)
4708{
4709 struct pci_dev *dev;
4710
4711 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4712 if (!dev->slot || dev->slot != slot)
4713 continue;
4714 pci_dev_lock(dev);
4715 pci_dev_restore(dev);
4716 pci_dev_unlock(dev);
4717 if (dev->subordinate)
4718 pci_bus_restore(dev->subordinate);
4719 }
4720}
4721
4722static int pci_slot_reset(struct pci_slot *slot, int probe)
4723{
4724 int rc;
4725
4726 if (!slot || !pci_slot_resetable(slot))
4727 return -ENOTTY;
4728
4729 if (!probe)
4730 pci_slot_lock(slot);
4731
4732 might_sleep();
4733
4734 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4735
4736 if (!probe)
4737 pci_slot_unlock(slot);
4738
4739 return rc;
4740}
4741
4742/**
4743 * pci_probe_reset_slot - probe whether a PCI slot can be reset
4744 * @slot: PCI slot to probe
4745 *
4746 * Return 0 if slot can be reset, negative if a slot reset is not supported.
4747 */
4748int pci_probe_reset_slot(struct pci_slot *slot)
4749{
4750 return pci_slot_reset(slot, 1);
4751}
4752EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4753
4754/**
4755 * pci_reset_slot - reset a PCI slot
4756 * @slot: PCI slot to reset
4757 *
4758 * A PCI bus may host multiple slots, each slot may support a reset mechanism
4759 * independent of other slots. For instance, some slots may support slot power
4760 * control. In the case of a 1:1 bus to slot architecture, this function may
4761 * wrap the bus reset to avoid spurious slot related events such as hotplug.
4762 * Generally a slot reset should be attempted before a bus reset. All of the
4763 * function of the slot and any subordinate buses behind the slot are reset
4764 * through this function. PCI config space of all devices in the slot and
4765 * behind the slot is saved before and restored after reset.
4766 *
4767 * Return 0 on success, non-zero on error.
4768 */
4769int pci_reset_slot(struct pci_slot *slot)
4770{
4771 int rc;
4772
4773 rc = pci_slot_reset(slot, 1);
4774 if (rc)
4775 return rc;
4776
4777 pci_slot_save_and_disable(slot);
4778
4779 rc = pci_slot_reset(slot, 0);
4780
4781 pci_slot_restore(slot);
4782
4783 return rc;
4784}
4785EXPORT_SYMBOL_GPL(pci_reset_slot);
4786
4787/**
4788 * pci_try_reset_slot - Try to reset a PCI slot
4789 * @slot: PCI slot to reset
4790 *
4791 * Same as above except return -EAGAIN if the slot cannot be locked
4792 */
4793int pci_try_reset_slot(struct pci_slot *slot)
4794{
4795 int rc;
4796
4797 rc = pci_slot_reset(slot, 1);
4798 if (rc)
4799 return rc;
4800
4801 pci_slot_save_and_disable(slot);
4802
4803 if (pci_slot_trylock(slot)) {
4804 might_sleep();
4805 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4806 pci_slot_unlock(slot);
4807 } else
4808 rc = -EAGAIN;
4809
4810 pci_slot_restore(slot);
4811
4812 return rc;
4813}
4814EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4815
4816static int pci_bus_reset(struct pci_bus *bus, int probe)
4817{
4818 if (!bus->self || !pci_bus_resetable(bus))
4819 return -ENOTTY;
4820
4821 if (probe)
4822 return 0;
4823
4824 pci_bus_lock(bus);
4825
4826 might_sleep();
4827
4828 pci_reset_bridge_secondary_bus(bus->self);
4829
4830 pci_bus_unlock(bus);
4831
4832 return 0;
4833}
4834
4835/**
4836 * pci_probe_reset_bus - probe whether a PCI bus can be reset
4837 * @bus: PCI bus to probe
4838 *
4839 * Return 0 if bus can be reset, negative if a bus reset is not supported.
4840 */
4841int pci_probe_reset_bus(struct pci_bus *bus)
4842{
4843 return pci_bus_reset(bus, 1);
4844}
4845EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4846
4847/**
4848 * pci_reset_bus - reset a PCI bus
4849 * @bus: top level PCI bus to reset
4850 *
4851 * Do a bus reset on the given bus and any subordinate buses, saving
4852 * and restoring state of all devices.
4853 *
4854 * Return 0 on success, non-zero on error.
4855 */
4856int pci_reset_bus(struct pci_bus *bus)
4857{
4858 int rc;
4859
4860 rc = pci_bus_reset(bus, 1);
4861 if (rc)
4862 return rc;
4863
4864 pci_bus_save_and_disable(bus);
4865
4866 rc = pci_bus_reset(bus, 0);
4867
4868 pci_bus_restore(bus);
4869
4870 return rc;
4871}
4872EXPORT_SYMBOL_GPL(pci_reset_bus);
4873
4874/**
4875 * pci_try_reset_bus - Try to reset a PCI bus
4876 * @bus: top level PCI bus to reset
4877 *
4878 * Same as above except return -EAGAIN if the bus cannot be locked
4879 */
4880int pci_try_reset_bus(struct pci_bus *bus)
4881{
4882 int rc;
4883
4884 rc = pci_bus_reset(bus, 1);
4885 if (rc)
4886 return rc;
4887
4888 pci_bus_save_and_disable(bus);
4889
4890 if (pci_bus_trylock(bus)) {
4891 might_sleep();
4892 pci_reset_bridge_secondary_bus(bus->self);
4893 pci_bus_unlock(bus);
4894 } else
4895 rc = -EAGAIN;
4896
4897 pci_bus_restore(bus);
4898
4899 return rc;
4900}
4901EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4902
4903/**
4904 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4905 * @dev: PCI device to query
4906 *
4907 * Returns mmrbc: maximum designed memory read count in bytes
4908 * or appropriate error value.
4909 */
4910int pcix_get_max_mmrbc(struct pci_dev *dev)
4911{
4912 int cap;
4913 u32 stat;
4914
4915 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4916 if (!cap)
4917 return -EINVAL;
4918
4919 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4920 return -EINVAL;
4921
4922 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4923}
4924EXPORT_SYMBOL(pcix_get_max_mmrbc);
4925
4926/**
4927 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4928 * @dev: PCI device to query
4929 *
4930 * Returns mmrbc: maximum memory read count in bytes
4931 * or appropriate error value.
4932 */
4933int pcix_get_mmrbc(struct pci_dev *dev)
4934{
4935 int cap;
4936 u16 cmd;
4937
4938 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4939 if (!cap)
4940 return -EINVAL;
4941
4942 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4943 return -EINVAL;
4944
4945 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4946}
4947EXPORT_SYMBOL(pcix_get_mmrbc);
4948
4949/**
4950 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4951 * @dev: PCI device to query
4952 * @mmrbc: maximum memory read count in bytes
4953 * valid values are 512, 1024, 2048, 4096
4954 *
4955 * If possible sets maximum memory read byte count, some bridges have erratas
4956 * that prevent this.
4957 */
4958int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4959{
4960 int cap;
4961 u32 stat, v, o;
4962 u16 cmd;
4963
4964 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4965 return -EINVAL;
4966
4967 v = ffs(mmrbc) - 10;
4968
4969 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4970 if (!cap)
4971 return -EINVAL;
4972
4973 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4974 return -EINVAL;
4975
4976 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4977 return -E2BIG;
4978
4979 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4980 return -EINVAL;
4981
4982 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4983 if (o != v) {
4984 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4985 return -EIO;
4986
4987 cmd &= ~PCI_X_CMD_MAX_READ;
4988 cmd |= v << 2;
4989 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4990 return -EIO;
4991 }
4992 return 0;
4993}
4994EXPORT_SYMBOL(pcix_set_mmrbc);
4995
4996/**
4997 * pcie_get_readrq - get PCI Express read request size
4998 * @dev: PCI device to query
4999 *
5000 * Returns maximum memory read request in bytes
5001 * or appropriate error value.
5002 */
5003int pcie_get_readrq(struct pci_dev *dev)
5004{
5005 u16 ctl;
5006
5007 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5008
5009 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
5010}
5011EXPORT_SYMBOL(pcie_get_readrq);
5012
5013/**
5014 * pcie_set_readrq - set PCI Express maximum memory read request
5015 * @dev: PCI device to query
5016 * @rq: maximum memory read count in bytes
5017 * valid values are 128, 256, 512, 1024, 2048, 4096
5018 *
5019 * If possible sets maximum memory read request in bytes
5020 */
5021int pcie_set_readrq(struct pci_dev *dev, int rq)
5022{
5023 u16 v;
5024
5025 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
5026 return -EINVAL;
5027
5028 /*
5029 * If using the "performance" PCIe config, we clamp the
5030 * read rq size to the max packet size to prevent the
5031 * host bridge generating requests larger than we can
5032 * cope with
5033 */
5034 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
5035 int mps = pcie_get_mps(dev);
5036
5037 if (mps < rq)
5038 rq = mps;
5039 }
5040
5041 v = (ffs(rq) - 8) << 12;
5042
5043 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5044 PCI_EXP_DEVCTL_READRQ, v);
5045}
5046EXPORT_SYMBOL(pcie_set_readrq);
5047
5048/**
5049 * pcie_get_mps - get PCI Express maximum payload size
5050 * @dev: PCI device to query
5051 *
5052 * Returns maximum payload size in bytes
5053 */
5054int pcie_get_mps(struct pci_dev *dev)
5055{
5056 u16 ctl;
5057
5058 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
5059
5060 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
5061}
5062EXPORT_SYMBOL(pcie_get_mps);
5063
5064/**
5065 * pcie_set_mps - set PCI Express maximum payload size
5066 * @dev: PCI device to query
5067 * @mps: maximum payload size in bytes
5068 * valid values are 128, 256, 512, 1024, 2048, 4096
5069 *
5070 * If possible sets maximum payload size
5071 */
5072int pcie_set_mps(struct pci_dev *dev, int mps)
5073{
5074 u16 v;
5075
5076 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
5077 return -EINVAL;
5078
5079 v = ffs(mps) - 8;
5080 if (v > dev->pcie_mpss)
5081 return -EINVAL;
5082 v <<= 5;
5083
5084 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
5085 PCI_EXP_DEVCTL_PAYLOAD, v);
5086}
5087EXPORT_SYMBOL(pcie_set_mps);
5088
5089/**
5090 * pcie_get_minimum_link - determine minimum link settings of a PCI device
5091 * @dev: PCI device to query
5092 * @speed: storage for minimum speed
5093 * @width: storage for minimum width
5094 *
5095 * This function will walk up the PCI device chain and determine the minimum
5096 * link width and speed of the device.
5097 */
5098int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
5099 enum pcie_link_width *width)
5100{
5101 int ret;
5102
5103 *speed = PCI_SPEED_UNKNOWN;
5104 *width = PCIE_LNK_WIDTH_UNKNOWN;
5105
5106 while (dev) {
5107 u16 lnksta;
5108 enum pci_bus_speed next_speed;
5109 enum pcie_link_width next_width;
5110
5111 ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5112 if (ret)
5113 return ret;
5114
5115 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5116 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5117 PCI_EXP_LNKSTA_NLW_SHIFT;
5118
5119 if (next_speed < *speed)
5120 *speed = next_speed;
5121
5122 if (next_width < *width)
5123 *width = next_width;
5124
5125 dev = dev->bus->self;
5126 }
5127
5128 return 0;
5129}
5130EXPORT_SYMBOL(pcie_get_minimum_link);
5131
5132/**
5133 * pcie_bandwidth_available - determine minimum link settings of a PCIe
5134 * device and its bandwidth limitation
5135 * @dev: PCI device to query
5136 * @limiting_dev: storage for device causing the bandwidth limitation
5137 * @speed: storage for speed of limiting device
5138 * @width: storage for width of limiting device
5139 *
5140 * Walk up the PCI device chain and find the point where the minimum
5141 * bandwidth is available. Return the bandwidth available there and (if
5142 * limiting_dev, speed, and width pointers are supplied) information about
5143 * that point. The bandwidth returned is in Mb/s, i.e., megabits/second of
5144 * raw bandwidth.
5145 */
5146u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
5147 enum pci_bus_speed *speed,
5148 enum pcie_link_width *width)
5149{
5150 u16 lnksta;
5151 enum pci_bus_speed next_speed;
5152 enum pcie_link_width next_width;
5153 u32 bw, next_bw;
5154
5155 if (speed)
5156 *speed = PCI_SPEED_UNKNOWN;
5157 if (width)
5158 *width = PCIE_LNK_WIDTH_UNKNOWN;
5159
5160 bw = 0;
5161
5162 while (dev) {
5163 pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
5164
5165 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
5166 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
5167 PCI_EXP_LNKSTA_NLW_SHIFT;
5168
5169 next_bw = next_width * PCIE_SPEED2MBS_ENC(next_speed);
5170
5171 /* Check if current device limits the total bandwidth */
5172 if (!bw || next_bw <= bw) {
5173 bw = next_bw;
5174
5175 if (limiting_dev)
5176 *limiting_dev = dev;
5177 if (speed)
5178 *speed = next_speed;
5179 if (width)
5180 *width = next_width;
5181 }
5182
5183 dev = pci_upstream_bridge(dev);
5184 }
5185
5186 return bw;
5187}
5188EXPORT_SYMBOL(pcie_bandwidth_available);
5189
5190/**
5191 * pcie_get_speed_cap - query for the PCI device's link speed capability
5192 * @dev: PCI device to query
5193 *
5194 * Query the PCI device speed capability. Return the maximum link speed
5195 * supported by the device.
5196 */
5197enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
5198{
5199 u32 lnkcap2, lnkcap;
5200
5201 /*
5202 * PCIe r4.0 sec 7.5.3.18 recommends using the Supported Link
5203 * Speeds Vector in Link Capabilities 2 when supported, falling
5204 * back to Max Link Speed in Link Capabilities otherwise.
5205 */
5206 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
5207 if (lnkcap2) { /* PCIe r3.0-compliant */
5208 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_16_0GB)
5209 return PCIE_SPEED_16_0GT;
5210 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
5211 return PCIE_SPEED_8_0GT;
5212 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
5213 return PCIE_SPEED_5_0GT;
5214 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
5215 return PCIE_SPEED_2_5GT;
5216 return PCI_SPEED_UNKNOWN;
5217 }
5218
5219 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5220 if (lnkcap) {
5221 if (lnkcap & PCI_EXP_LNKCAP_SLS_16_0GB)
5222 return PCIE_SPEED_16_0GT;
5223 else if (lnkcap & PCI_EXP_LNKCAP_SLS_8_0GB)
5224 return PCIE_SPEED_8_0GT;
5225 else if (lnkcap & PCI_EXP_LNKCAP_SLS_5_0GB)
5226 return PCIE_SPEED_5_0GT;
5227 else if (lnkcap & PCI_EXP_LNKCAP_SLS_2_5GB)
5228 return PCIE_SPEED_2_5GT;
5229 }
5230
5231 return PCI_SPEED_UNKNOWN;
5232}
5233
5234/**
5235 * pcie_get_width_cap - query for the PCI device's link width capability
5236 * @dev: PCI device to query
5237 *
5238 * Query the PCI device width capability. Return the maximum link width
5239 * supported by the device.
5240 */
5241enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev)
5242{
5243 u32 lnkcap;
5244
5245 pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
5246 if (lnkcap)
5247 return (lnkcap & PCI_EXP_LNKCAP_MLW) >> 4;
5248
5249 return PCIE_LNK_WIDTH_UNKNOWN;
5250}
5251
5252/**
5253 * pcie_bandwidth_capable - calculate a PCI device's link bandwidth capability
5254 * @dev: PCI device
5255 * @speed: storage for link speed
5256 * @width: storage for link width
5257 *
5258 * Calculate a PCI device's link bandwidth by querying for its link speed
5259 * and width, multiplying them, and applying encoding overhead. The result
5260 * is in Mb/s, i.e., megabits/second of raw bandwidth.
5261 */
5262u32 pcie_bandwidth_capable(struct pci_dev *dev, enum pci_bus_speed *speed,
5263 enum pcie_link_width *width)
5264{
5265 *speed = pcie_get_speed_cap(dev);
5266 *width = pcie_get_width_cap(dev);
5267
5268 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
5269 return 0;
5270
5271 return *width * PCIE_SPEED2MBS_ENC(*speed);
5272}
5273
5274/**
5275 * pcie_print_link_status - Report the PCI device's link speed and width
5276 * @dev: PCI device to query
5277 *
5278 * Report the available bandwidth at the device. If this is less than the
5279 * device is capable of, report the device's maximum possible bandwidth and
5280 * the upstream link that limits its performance to less than that.
5281 */
5282void pcie_print_link_status(struct pci_dev *dev)
5283{
5284 enum pcie_link_width width, width_cap;
5285 enum pci_bus_speed speed, speed_cap;
5286 struct pci_dev *limiting_dev = NULL;
5287 u32 bw_avail, bw_cap;
5288
5289 bw_cap = pcie_bandwidth_capable(dev, &speed_cap, &width_cap);
5290 bw_avail = pcie_bandwidth_available(dev, &limiting_dev, &speed, &width);
5291
5292 if (bw_avail >= bw_cap)
5293 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth (%s x%d link)\n",
5294 bw_cap / 1000, bw_cap % 1000,
5295 PCIE_SPEED2STR(speed_cap), width_cap);
5296 else
5297 pci_info(dev, "%u.%03u Gb/s available PCIe bandwidth, limited by %s x%d link at %s (capable of %u.%03u Gb/s with %s x%d link)\n",
5298 bw_avail / 1000, bw_avail % 1000,
5299 PCIE_SPEED2STR(speed), width,
5300 limiting_dev ? pci_name(limiting_dev) : "<unknown>",
5301 bw_cap / 1000, bw_cap % 1000,
5302 PCIE_SPEED2STR(speed_cap), width_cap);
5303}
5304EXPORT_SYMBOL(pcie_print_link_status);
5305
5306/**
5307 * pci_select_bars - Make BAR mask from the type of resource
5308 * @dev: the PCI device for which BAR mask is made
5309 * @flags: resource type mask to be selected
5310 *
5311 * This helper routine makes bar mask from the type of resource.
5312 */
5313int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5314{
5315 int i, bars = 0;
5316 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5317 if (pci_resource_flags(dev, i) & flags)
5318 bars |= (1 << i);
5319 return bars;
5320}
5321EXPORT_SYMBOL(pci_select_bars);
5322
5323/* Some architectures require additional programming to enable VGA */
5324static arch_set_vga_state_t arch_set_vga_state;
5325
5326void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5327{
5328 arch_set_vga_state = func; /* NULL disables */
5329}
5330
5331static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5332 unsigned int command_bits, u32 flags)
5333{
5334 if (arch_set_vga_state)
5335 return arch_set_vga_state(dev, decode, command_bits,
5336 flags);
5337 return 0;
5338}
5339
5340/**
5341 * pci_set_vga_state - set VGA decode state on device and parents if requested
5342 * @dev: the PCI device
5343 * @decode: true = enable decoding, false = disable decoding
5344 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5345 * @flags: traverse ancestors and change bridges
5346 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5347 */
5348int pci_set_vga_state(struct pci_dev *dev, bool decode,
5349 unsigned int command_bits, u32 flags)
5350{
5351 struct pci_bus *bus;
5352 struct pci_dev *bridge;
5353 u16 cmd;
5354 int rc;
5355
5356 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5357
5358 /* ARCH specific VGA enables */
5359 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5360 if (rc)
5361 return rc;
5362
5363 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5364 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5365 if (decode == true)
5366 cmd |= command_bits;
5367 else
5368 cmd &= ~command_bits;
5369 pci_write_config_word(dev, PCI_COMMAND, cmd);
5370 }
5371
5372 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5373 return 0;
5374
5375 bus = dev->bus;
5376 while (bus) {
5377 bridge = bus->self;
5378 if (bridge) {
5379 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5380 &cmd);
5381 if (decode == true)
5382 cmd |= PCI_BRIDGE_CTL_VGA;
5383 else
5384 cmd &= ~PCI_BRIDGE_CTL_VGA;
5385 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5386 cmd);
5387 }
5388 bus = bus->parent;
5389 }
5390 return 0;
5391}
5392
5393/**
5394 * pci_add_dma_alias - Add a DMA devfn alias for a device
5395 * @dev: the PCI device for which alias is added
5396 * @devfn: alias slot and function
5397 *
5398 * This helper encodes 8-bit devfn as bit number in dma_alias_mask.
5399 * It should be called early, preferably as PCI fixup header quirk.
5400 */
5401void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5402{
5403 if (!dev->dma_alias_mask)
5404 dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
5405 sizeof(long), GFP_KERNEL);
5406 if (!dev->dma_alias_mask) {
5407 pci_warn(dev, "Unable to allocate DMA alias mask\n");
5408 return;
5409 }
5410
5411 set_bit(devfn, dev->dma_alias_mask);
5412 pci_info(dev, "Enabling fixed DMA alias to %02x.%d\n",
5413 PCI_SLOT(devfn), PCI_FUNC(devfn));
5414}
5415
5416bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5417{
5418 return (dev1->dma_alias_mask &&
5419 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5420 (dev2->dma_alias_mask &&
5421 test_bit(dev1->devfn, dev2->dma_alias_mask));
5422}
5423
5424bool pci_device_is_present(struct pci_dev *pdev)
5425{
5426 u32 v;
5427
5428 if (pci_dev_is_disconnected(pdev))
5429 return false;
5430 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5431}
5432EXPORT_SYMBOL_GPL(pci_device_is_present);
5433
5434void pci_ignore_hotplug(struct pci_dev *dev)
5435{
5436 struct pci_dev *bridge = dev->bus->self;
5437
5438 dev->ignore_hotplug = 1;
5439 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5440 if (bridge)
5441 bridge->ignore_hotplug = 1;
5442}
5443EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5444
5445resource_size_t __weak pcibios_default_alignment(void)
5446{
5447 return 0;
5448}
5449
5450#define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5451static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5452static DEFINE_SPINLOCK(resource_alignment_lock);
5453
5454/**
5455 * pci_specified_resource_alignment - get resource alignment specified by user.
5456 * @dev: the PCI device to get
5457 * @resize: whether or not to change resources' size when reassigning alignment
5458 *
5459 * RETURNS: Resource alignment if it is specified.
5460 * Zero if it is not specified.
5461 */
5462static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5463 bool *resize)
5464{
5465 int seg, bus, slot, func, align_order, count;
5466 unsigned short vendor, device, subsystem_vendor, subsystem_device;
5467 resource_size_t align = pcibios_default_alignment();
5468 char *p;
5469
5470 spin_lock(&resource_alignment_lock);
5471 p = resource_alignment_param;
5472 if (!*p && !align)
5473 goto out;
5474 if (pci_has_flag(PCI_PROBE_ONLY)) {
5475 align = 0;
5476 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5477 goto out;
5478 }
5479
5480 while (*p) {
5481 count = 0;
5482 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5483 p[count] == '@') {
5484 p += count + 1;
5485 } else {
5486 align_order = -1;
5487 }
5488 if (strncmp(p, "pci:", 4) == 0) {
5489 /* PCI vendor/device (subvendor/subdevice) ids are specified */
5490 p += 4;
5491 if (sscanf(p, "%hx:%hx:%hx:%hx%n",
5492 &vendor, &device, &subsystem_vendor, &subsystem_device, &count) != 4) {
5493 if (sscanf(p, "%hx:%hx%n", &vendor, &device, &count) != 2) {
5494 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: pci:%s\n",
5495 p);
5496 break;
5497 }
5498 subsystem_vendor = subsystem_device = 0;
5499 }
5500 p += count;
5501 if ((!vendor || (vendor == dev->vendor)) &&
5502 (!device || (device == dev->device)) &&
5503 (!subsystem_vendor || (subsystem_vendor == dev->subsystem_vendor)) &&
5504 (!subsystem_device || (subsystem_device == dev->subsystem_device))) {
5505 *resize = true;
5506 if (align_order == -1)
5507 align = PAGE_SIZE;
5508 else
5509 align = 1 << align_order;
5510 /* Found */
5511 break;
5512 }
5513 }
5514 else {
5515 if (sscanf(p, "%x:%x:%x.%x%n",
5516 &seg, &bus, &slot, &func, &count) != 4) {
5517 seg = 0;
5518 if (sscanf(p, "%x:%x.%x%n",
5519 &bus, &slot, &func, &count) != 3) {
5520 /* Invalid format */
5521 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
5522 p);
5523 break;
5524 }
5525 }
5526 p += count;
5527 if (seg == pci_domain_nr(dev->bus) &&
5528 bus == dev->bus->number &&
5529 slot == PCI_SLOT(dev->devfn) &&
5530 func == PCI_FUNC(dev->devfn)) {
5531 *resize = true;
5532 if (align_order == -1)
5533 align = PAGE_SIZE;
5534 else
5535 align = 1 << align_order;
5536 /* Found */
5537 break;
5538 }
5539 }
5540 if (*p != ';' && *p != ',') {
5541 /* End of param or invalid format */
5542 break;
5543 }
5544 p++;
5545 }
5546out:
5547 spin_unlock(&resource_alignment_lock);
5548 return align;
5549}
5550
5551static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5552 resource_size_t align, bool resize)
5553{
5554 struct resource *r = &dev->resource[bar];
5555 resource_size_t size;
5556
5557 if (!(r->flags & IORESOURCE_MEM))
5558 return;
5559
5560 if (r->flags & IORESOURCE_PCI_FIXED) {
5561 pci_info(dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5562 bar, r, (unsigned long long)align);
5563 return;
5564 }
5565
5566 size = resource_size(r);
5567 if (size >= align)
5568 return;
5569
5570 /*
5571 * Increase the alignment of the resource. There are two ways we
5572 * can do this:
5573 *
5574 * 1) Increase the size of the resource. BARs are aligned on their
5575 * size, so when we reallocate space for this resource, we'll
5576 * allocate it with the larger alignment. This also prevents
5577 * assignment of any other BARs inside the alignment region, so
5578 * if we're requesting page alignment, this means no other BARs
5579 * will share the page.
5580 *
5581 * The disadvantage is that this makes the resource larger than
5582 * the hardware BAR, which may break drivers that compute things
5583 * based on the resource size, e.g., to find registers at a
5584 * fixed offset before the end of the BAR.
5585 *
5586 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5587 * set r->start to the desired alignment. By itself this
5588 * doesn't prevent other BARs being put inside the alignment
5589 * region, but if we realign *every* resource of every device in
5590 * the system, none of them will share an alignment region.
5591 *
5592 * When the user has requested alignment for only some devices via
5593 * the "pci=resource_alignment" argument, "resize" is true and we
5594 * use the first method. Otherwise we assume we're aligning all
5595 * devices and we use the second.
5596 */
5597
5598 pci_info(dev, "BAR%d %pR: requesting alignment to %#llx\n",
5599 bar, r, (unsigned long long)align);
5600
5601 if (resize) {
5602 r->start = 0;
5603 r->end = align - 1;
5604 } else {
5605 r->flags &= ~IORESOURCE_SIZEALIGN;
5606 r->flags |= IORESOURCE_STARTALIGN;
5607 r->start = align;
5608 r->end = r->start + size - 1;
5609 }
5610 r->flags |= IORESOURCE_UNSET;
5611}
5612
5613/*
5614 * This function disables memory decoding and releases memory resources
5615 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5616 * It also rounds up size to specified alignment.
5617 * Later on, the kernel will assign page-aligned memory resource back
5618 * to the device.
5619 */
5620void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5621{
5622 int i;
5623 struct resource *r;
5624 resource_size_t align;
5625 u16 command;
5626 bool resize = false;
5627
5628 /*
5629 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
5630 * 3.4.1.11. Their resources are allocated from the space
5631 * described by the VF BARx register in the PF's SR-IOV capability.
5632 * We can't influence their alignment here.
5633 */
5634 if (dev->is_virtfn)
5635 return;
5636
5637 /* check if specified PCI is target device to reassign */
5638 align = pci_specified_resource_alignment(dev, &resize);
5639 if (!align)
5640 return;
5641
5642 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5643 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5644 pci_warn(dev, "Can't reassign resources to host bridge\n");
5645 return;
5646 }
5647
5648 pci_read_config_word(dev, PCI_COMMAND, &command);
5649 command &= ~PCI_COMMAND_MEMORY;
5650 pci_write_config_word(dev, PCI_COMMAND, command);
5651
5652 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
5653 pci_request_resource_alignment(dev, i, align, resize);
5654
5655 /*
5656 * Need to disable bridge's resource window,
5657 * to enable the kernel to reassign new resource
5658 * window later on.
5659 */
5660 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5661 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5662 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5663 r = &dev->resource[i];
5664 if (!(r->flags & IORESOURCE_MEM))
5665 continue;
5666 r->flags |= IORESOURCE_UNSET;
5667 r->end = resource_size(r) - 1;
5668 r->start = 0;
5669 }
5670 pci_disable_bridge_window(dev);
5671 }
5672}
5673
5674static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5675{
5676 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
5677 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
5678 spin_lock(&resource_alignment_lock);
5679 strncpy(resource_alignment_param, buf, count);
5680 resource_alignment_param[count] = '\0';
5681 spin_unlock(&resource_alignment_lock);
5682 return count;
5683}
5684
5685static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
5686{
5687 size_t count;
5688 spin_lock(&resource_alignment_lock);
5689 count = snprintf(buf, size, "%s", resource_alignment_param);
5690 spin_unlock(&resource_alignment_lock);
5691 return count;
5692}
5693
5694static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
5695{
5696 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
5697}
5698
5699static ssize_t pci_resource_alignment_store(struct bus_type *bus,
5700 const char *buf, size_t count)
5701{
5702 return pci_set_resource_alignment_param(buf, count);
5703}
5704
5705static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
5706 pci_resource_alignment_store);
5707
5708static int __init pci_resource_alignment_sysfs_init(void)
5709{
5710 return bus_create_file(&pci_bus_type,
5711 &bus_attr_resource_alignment);
5712}
5713late_initcall(pci_resource_alignment_sysfs_init);
5714
5715static void pci_no_domains(void)
5716{
5717#ifdef CONFIG_PCI_DOMAINS
5718 pci_domains_supported = 0;
5719#endif
5720}
5721
5722#ifdef CONFIG_PCI_DOMAINS
5723static atomic_t __domain_nr = ATOMIC_INIT(-1);
5724
5725int pci_get_new_domain_nr(void)
5726{
5727 return atomic_inc_return(&__domain_nr);
5728}
5729
5730#ifdef CONFIG_PCI_DOMAINS_GENERIC
5731static int of_pci_bus_find_domain_nr(struct device *parent)
5732{
5733 static int use_dt_domains = -1;
5734 int domain = -1;
5735
5736 if (parent)
5737 domain = of_get_pci_domain_nr(parent->of_node);
5738 /*
5739 * Check DT domain and use_dt_domains values.
5740 *
5741 * If DT domain property is valid (domain >= 0) and
5742 * use_dt_domains != 0, the DT assignment is valid since this means
5743 * we have not previously allocated a domain number by using
5744 * pci_get_new_domain_nr(); we should also update use_dt_domains to
5745 * 1, to indicate that we have just assigned a domain number from
5746 * DT.
5747 *
5748 * If DT domain property value is not valid (ie domain < 0), and we
5749 * have not previously assigned a domain number from DT
5750 * (use_dt_domains != 1) we should assign a domain number by
5751 * using the:
5752 *
5753 * pci_get_new_domain_nr()
5754 *
5755 * API and update the use_dt_domains value to keep track of method we
5756 * are using to assign domain numbers (use_dt_domains = 0).
5757 *
5758 * All other combinations imply we have a platform that is trying
5759 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
5760 * which is a recipe for domain mishandling and it is prevented by
5761 * invalidating the domain value (domain = -1) and printing a
5762 * corresponding error.
5763 */
5764 if (domain >= 0 && use_dt_domains) {
5765 use_dt_domains = 1;
5766 } else if (domain < 0 && use_dt_domains != 1) {
5767 use_dt_domains = 0;
5768 domain = pci_get_new_domain_nr();
5769 } else {
5770 if (parent)
5771 pr_err("Node %pOF has ", parent->of_node);
5772 pr_err("Inconsistent \"linux,pci-domain\" property in DT\n");
5773 domain = -1;
5774 }
5775
5776 return domain;
5777}
5778
5779int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
5780{
5781 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
5782 acpi_pci_bus_find_domain_nr(bus);
5783}
5784#endif
5785#endif
5786
5787/**
5788 * pci_ext_cfg_avail - can we access extended PCI config space?
5789 *
5790 * Returns 1 if we can access PCI extended config space (offsets
5791 * greater than 0xff). This is the default implementation. Architecture
5792 * implementations can override this.
5793 */
5794int __weak pci_ext_cfg_avail(void)
5795{
5796 return 1;
5797}
5798
5799void __weak pci_fixup_cardbus(struct pci_bus *bus)
5800{
5801}
5802EXPORT_SYMBOL(pci_fixup_cardbus);
5803
5804static int __init pci_setup(char *str)
5805{
5806 while (str) {
5807 char *k = strchr(str, ',');
5808 if (k)
5809 *k++ = 0;
5810 if (*str && (str = pcibios_setup(str)) && *str) {
5811 if (!strcmp(str, "nomsi")) {
5812 pci_no_msi();
5813 } else if (!strcmp(str, "noaer")) {
5814 pci_no_aer();
5815 } else if (!strncmp(str, "realloc=", 8)) {
5816 pci_realloc_get_opt(str + 8);
5817 } else if (!strncmp(str, "realloc", 7)) {
5818 pci_realloc_get_opt("on");
5819 } else if (!strcmp(str, "nodomains")) {
5820 pci_no_domains();
5821 } else if (!strncmp(str, "noari", 5)) {
5822 pcie_ari_disabled = true;
5823 } else if (!strncmp(str, "cbiosize=", 9)) {
5824 pci_cardbus_io_size = memparse(str + 9, &str);
5825 } else if (!strncmp(str, "cbmemsize=", 10)) {
5826 pci_cardbus_mem_size = memparse(str + 10, &str);
5827 } else if (!strncmp(str, "resource_alignment=", 19)) {
5828 pci_set_resource_alignment_param(str + 19,
5829 strlen(str + 19));
5830 } else if (!strncmp(str, "ecrc=", 5)) {
5831 pcie_ecrc_get_policy(str + 5);
5832 } else if (!strncmp(str, "hpiosize=", 9)) {
5833 pci_hotplug_io_size = memparse(str + 9, &str);
5834 } else if (!strncmp(str, "hpmemsize=", 10)) {
5835 pci_hotplug_mem_size = memparse(str + 10, &str);
5836 } else if (!strncmp(str, "hpbussize=", 10)) {
5837 pci_hotplug_bus_size =
5838 simple_strtoul(str + 10, &str, 0);
5839 if (pci_hotplug_bus_size > 0xff)
5840 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
5841 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
5842 pcie_bus_config = PCIE_BUS_TUNE_OFF;
5843 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
5844 pcie_bus_config = PCIE_BUS_SAFE;
5845 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
5846 pcie_bus_config = PCIE_BUS_PERFORMANCE;
5847 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
5848 pcie_bus_config = PCIE_BUS_PEER2PEER;
5849 } else if (!strncmp(str, "pcie_scan_all", 13)) {
5850 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
5851 } else {
5852 printk(KERN_ERR "PCI: Unknown option `%s'\n",
5853 str);
5854 }
5855 }
5856 str = k;
5857 }
5858 return 0;
5859}
5860early_param("pci", pci_setup);