]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/pci/pci.c
02e844c1e63755e4e4a00e4b7d16d35942f35958
[mirror_ubuntu-artful-kernel.git] / drivers / pci / pci.c
1 /*
2 * PCI Bus Services, see include/linux/pci.h for further explanation.
3 *
4 * Copyright 1993 -- 1997 Drew Eckhardt, Frederic Potter,
5 * David Mosberger-Tang
6 *
7 * Copyright 1997 -- 2000 Martin Mares <mj@ucw.cz>
8 */
9
10 #include <linux/acpi.h>
11 #include <linux/kernel.h>
12 #include <linux/delay.h>
13 #include <linux/dmi.h>
14 #include <linux/init.h>
15 #include <linux/of.h>
16 #include <linux/of_pci.h>
17 #include <linux/pci.h>
18 #include <linux/pm.h>
19 #include <linux/slab.h>
20 #include <linux/module.h>
21 #include <linux/spinlock.h>
22 #include <linux/string.h>
23 #include <linux/log2.h>
24 #include <linux/pci-aspm.h>
25 #include <linux/pm_wakeup.h>
26 #include <linux/interrupt.h>
27 #include <linux/device.h>
28 #include <linux/pm_runtime.h>
29 #include <linux/pci_hotplug.h>
30 #include <linux/vmalloc.h>
31 #include <linux/pci-ats.h>
32 #include <asm/setup.h>
33 #include <asm/dma.h>
34 #include <linux/aer.h>
35 #include "pci.h"
36
37 const char *pci_power_names[] = {
38 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
39 };
40 EXPORT_SYMBOL_GPL(pci_power_names);
41
42 int isa_dma_bridge_buggy;
43 EXPORT_SYMBOL(isa_dma_bridge_buggy);
44
45 int pci_pci_problems;
46 EXPORT_SYMBOL(pci_pci_problems);
47
48 unsigned int pci_pm_d3_delay;
49
50 static void pci_pme_list_scan(struct work_struct *work);
51
52 static LIST_HEAD(pci_pme_list);
53 static DEFINE_MUTEX(pci_pme_list_mutex);
54 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
55
56 struct pci_pme_device {
57 struct list_head list;
58 struct pci_dev *dev;
59 };
60
61 #define PME_TIMEOUT 1000 /* How long between PME checks */
62
63 static void pci_dev_d3_sleep(struct pci_dev *dev)
64 {
65 unsigned int delay = dev->d3_delay;
66
67 if (delay < pci_pm_d3_delay)
68 delay = pci_pm_d3_delay;
69
70 if (delay)
71 msleep(delay);
72 }
73
74 #ifdef CONFIG_PCI_DOMAINS
75 int pci_domains_supported = 1;
76 #endif
77
78 #define DEFAULT_CARDBUS_IO_SIZE (256)
79 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
80 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
81 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
82 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
83
84 #define DEFAULT_HOTPLUG_IO_SIZE (256)
85 #define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
86 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
87 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
88 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
89
90 #define DEFAULT_HOTPLUG_BUS_SIZE 1
91 unsigned long pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
92
93 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
94
95 /*
96 * The default CLS is used if arch didn't set CLS explicitly and not
97 * all pci devices agree on the same value. Arch can override either
98 * the dfl or actual value as it sees fit. Don't forget this is
99 * measured in 32-bit words, not bytes.
100 */
101 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
102 u8 pci_cache_line_size;
103
104 /*
105 * If we set up a device for bus mastering, we need to check the latency
106 * timer as certain BIOSes forget to set it properly.
107 */
108 unsigned int pcibios_max_latency = 255;
109
110 /* If set, the PCIe ARI capability will not be used. */
111 static bool pcie_ari_disabled;
112
113 /* Disable bridge_d3 for all PCIe ports */
114 static bool pci_bridge_d3_disable;
115 /* Force bridge_d3 for all PCIe ports */
116 static bool pci_bridge_d3_force;
117
118 static int __init pcie_port_pm_setup(char *str)
119 {
120 if (!strcmp(str, "off"))
121 pci_bridge_d3_disable = true;
122 else if (!strcmp(str, "force"))
123 pci_bridge_d3_force = true;
124 return 1;
125 }
126 __setup("pcie_port_pm=", pcie_port_pm_setup);
127
128 /**
129 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
130 * @bus: pointer to PCI bus structure to search
131 *
132 * Given a PCI bus, returns the highest PCI bus number present in the set
133 * including the given PCI bus and its list of child PCI buses.
134 */
135 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
136 {
137 struct pci_bus *tmp;
138 unsigned char max, n;
139
140 max = bus->busn_res.end;
141 list_for_each_entry(tmp, &bus->children, node) {
142 n = pci_bus_max_busnr(tmp);
143 if (n > max)
144 max = n;
145 }
146 return max;
147 }
148 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
149
150 #ifdef CONFIG_HAS_IOMEM
151 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
152 {
153 struct resource *res = &pdev->resource[bar];
154
155 /*
156 * Make sure the BAR is actually a memory resource, not an IO resource
157 */
158 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
159 dev_warn(&pdev->dev, "can't ioremap BAR %d: %pR\n", bar, res);
160 return NULL;
161 }
162 return ioremap_nocache(res->start, resource_size(res));
163 }
164 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
165
166 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
167 {
168 /*
169 * Make sure the BAR is actually a memory resource, not an IO resource
170 */
171 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
172 WARN_ON(1);
173 return NULL;
174 }
175 return ioremap_wc(pci_resource_start(pdev, bar),
176 pci_resource_len(pdev, bar));
177 }
178 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
179 #endif
180
181
182 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
183 u8 pos, int cap, int *ttl)
184 {
185 u8 id;
186 u16 ent;
187
188 pci_bus_read_config_byte(bus, devfn, pos, &pos);
189
190 while ((*ttl)--) {
191 if (pos < 0x40)
192 break;
193 pos &= ~3;
194 pci_bus_read_config_word(bus, devfn, pos, &ent);
195
196 id = ent & 0xff;
197 if (id == 0xff)
198 break;
199 if (id == cap)
200 return pos;
201 pos = (ent >> 8);
202 }
203 return 0;
204 }
205
206 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
207 u8 pos, int cap)
208 {
209 int ttl = PCI_FIND_CAP_TTL;
210
211 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
212 }
213
214 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
215 {
216 return __pci_find_next_cap(dev->bus, dev->devfn,
217 pos + PCI_CAP_LIST_NEXT, cap);
218 }
219 EXPORT_SYMBOL_GPL(pci_find_next_capability);
220
221 static int __pci_bus_find_cap_start(struct pci_bus *bus,
222 unsigned int devfn, u8 hdr_type)
223 {
224 u16 status;
225
226 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
227 if (!(status & PCI_STATUS_CAP_LIST))
228 return 0;
229
230 switch (hdr_type) {
231 case PCI_HEADER_TYPE_NORMAL:
232 case PCI_HEADER_TYPE_BRIDGE:
233 return PCI_CAPABILITY_LIST;
234 case PCI_HEADER_TYPE_CARDBUS:
235 return PCI_CB_CAPABILITY_LIST;
236 }
237
238 return 0;
239 }
240
241 /**
242 * pci_find_capability - query for devices' capabilities
243 * @dev: PCI device to query
244 * @cap: capability code
245 *
246 * Tell if a device supports a given PCI capability.
247 * Returns the address of the requested capability structure within the
248 * device's PCI configuration space or 0 in case the device does not
249 * support it. Possible values for @cap:
250 *
251 * %PCI_CAP_ID_PM Power Management
252 * %PCI_CAP_ID_AGP Accelerated Graphics Port
253 * %PCI_CAP_ID_VPD Vital Product Data
254 * %PCI_CAP_ID_SLOTID Slot Identification
255 * %PCI_CAP_ID_MSI Message Signalled Interrupts
256 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
257 * %PCI_CAP_ID_PCIX PCI-X
258 * %PCI_CAP_ID_EXP PCI Express
259 */
260 int pci_find_capability(struct pci_dev *dev, int cap)
261 {
262 int pos;
263
264 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
265 if (pos)
266 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
267
268 return pos;
269 }
270 EXPORT_SYMBOL(pci_find_capability);
271
272 /**
273 * pci_bus_find_capability - query for devices' capabilities
274 * @bus: the PCI bus to query
275 * @devfn: PCI device to query
276 * @cap: capability code
277 *
278 * Like pci_find_capability() but works for pci devices that do not have a
279 * pci_dev structure set up yet.
280 *
281 * Returns the address of the requested capability structure within the
282 * device's PCI configuration space or 0 in case the device does not
283 * support it.
284 */
285 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
286 {
287 int pos;
288 u8 hdr_type;
289
290 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
291
292 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
293 if (pos)
294 pos = __pci_find_next_cap(bus, devfn, pos, cap);
295
296 return pos;
297 }
298 EXPORT_SYMBOL(pci_bus_find_capability);
299
300 /**
301 * pci_find_next_ext_capability - Find an extended capability
302 * @dev: PCI device to query
303 * @start: address at which to start looking (0 to start at beginning of list)
304 * @cap: capability code
305 *
306 * Returns the address of the next matching extended capability structure
307 * within the device's PCI configuration space or 0 if the device does
308 * not support it. Some capabilities can occur several times, e.g., the
309 * vendor-specific capability, and this provides a way to find them all.
310 */
311 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
312 {
313 u32 header;
314 int ttl;
315 int pos = PCI_CFG_SPACE_SIZE;
316
317 /* minimum 8 bytes per capability */
318 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
319
320 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
321 return 0;
322
323 if (start)
324 pos = start;
325
326 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
327 return 0;
328
329 /*
330 * If we have no capabilities, this is indicated by cap ID,
331 * cap version and next pointer all being 0.
332 */
333 if (header == 0)
334 return 0;
335
336 while (ttl-- > 0) {
337 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
338 return pos;
339
340 pos = PCI_EXT_CAP_NEXT(header);
341 if (pos < PCI_CFG_SPACE_SIZE)
342 break;
343
344 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
345 break;
346 }
347
348 return 0;
349 }
350 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
351
352 /**
353 * pci_find_ext_capability - Find an extended capability
354 * @dev: PCI device to query
355 * @cap: capability code
356 *
357 * Returns the address of the requested extended capability structure
358 * within the device's PCI configuration space or 0 if the device does
359 * not support it. Possible values for @cap:
360 *
361 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
362 * %PCI_EXT_CAP_ID_VC Virtual Channel
363 * %PCI_EXT_CAP_ID_DSN Device Serial Number
364 * %PCI_EXT_CAP_ID_PWR Power Budgeting
365 */
366 int pci_find_ext_capability(struct pci_dev *dev, int cap)
367 {
368 return pci_find_next_ext_capability(dev, 0, cap);
369 }
370 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
371
372 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
373 {
374 int rc, ttl = PCI_FIND_CAP_TTL;
375 u8 cap, mask;
376
377 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
378 mask = HT_3BIT_CAP_MASK;
379 else
380 mask = HT_5BIT_CAP_MASK;
381
382 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
383 PCI_CAP_ID_HT, &ttl);
384 while (pos) {
385 rc = pci_read_config_byte(dev, pos + 3, &cap);
386 if (rc != PCIBIOS_SUCCESSFUL)
387 return 0;
388
389 if ((cap & mask) == ht_cap)
390 return pos;
391
392 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
393 pos + PCI_CAP_LIST_NEXT,
394 PCI_CAP_ID_HT, &ttl);
395 }
396
397 return 0;
398 }
399 /**
400 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
401 * @dev: PCI device to query
402 * @pos: Position from which to continue searching
403 * @ht_cap: Hypertransport capability code
404 *
405 * To be used in conjunction with pci_find_ht_capability() to search for
406 * all capabilities matching @ht_cap. @pos should always be a value returned
407 * from pci_find_ht_capability().
408 *
409 * NB. To be 100% safe against broken PCI devices, the caller should take
410 * steps to avoid an infinite loop.
411 */
412 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
413 {
414 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
415 }
416 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
417
418 /**
419 * pci_find_ht_capability - query a device's Hypertransport capabilities
420 * @dev: PCI device to query
421 * @ht_cap: Hypertransport capability code
422 *
423 * Tell if a device supports a given Hypertransport capability.
424 * Returns an address within the device's PCI configuration space
425 * or 0 in case the device does not support the request capability.
426 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
427 * which has a Hypertransport capability matching @ht_cap.
428 */
429 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
430 {
431 int pos;
432
433 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
434 if (pos)
435 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
436
437 return pos;
438 }
439 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
440
441 /**
442 * pci_find_parent_resource - return resource region of parent bus of given region
443 * @dev: PCI device structure contains resources to be searched
444 * @res: child resource record for which parent is sought
445 *
446 * For given resource region of given device, return the resource
447 * region of parent bus the given region is contained in.
448 */
449 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
450 struct resource *res)
451 {
452 const struct pci_bus *bus = dev->bus;
453 struct resource *r;
454 int i;
455
456 pci_bus_for_each_resource(bus, r, i) {
457 if (!r)
458 continue;
459 if (resource_contains(r, res)) {
460
461 /*
462 * If the window is prefetchable but the BAR is
463 * not, the allocator made a mistake.
464 */
465 if (r->flags & IORESOURCE_PREFETCH &&
466 !(res->flags & IORESOURCE_PREFETCH))
467 return NULL;
468
469 /*
470 * If we're below a transparent bridge, there may
471 * be both a positively-decoded aperture and a
472 * subtractively-decoded region that contain the BAR.
473 * We want the positively-decoded one, so this depends
474 * on pci_bus_for_each_resource() giving us those
475 * first.
476 */
477 return r;
478 }
479 }
480 return NULL;
481 }
482 EXPORT_SYMBOL(pci_find_parent_resource);
483
484 /**
485 * pci_find_resource - Return matching PCI device resource
486 * @dev: PCI device to query
487 * @res: Resource to look for
488 *
489 * Goes over standard PCI resources (BARs) and checks if the given resource
490 * is partially or fully contained in any of them. In that case the
491 * matching resource is returned, %NULL otherwise.
492 */
493 struct resource *pci_find_resource(struct pci_dev *dev, struct resource *res)
494 {
495 int i;
496
497 for (i = 0; i < PCI_ROM_RESOURCE; i++) {
498 struct resource *r = &dev->resource[i];
499
500 if (r->start && resource_contains(r, res))
501 return r;
502 }
503
504 return NULL;
505 }
506 EXPORT_SYMBOL(pci_find_resource);
507
508 /**
509 * pci_find_pcie_root_port - return PCIe Root Port
510 * @dev: PCI device to query
511 *
512 * Traverse up the parent chain and return the PCIe Root Port PCI Device
513 * for a given PCI Device.
514 */
515 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
516 {
517 struct pci_dev *bridge, *highest_pcie_bridge = dev;
518
519 bridge = pci_upstream_bridge(dev);
520 while (bridge && pci_is_pcie(bridge)) {
521 highest_pcie_bridge = bridge;
522 bridge = pci_upstream_bridge(bridge);
523 }
524
525 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
526 return NULL;
527
528 return highest_pcie_bridge;
529 }
530 EXPORT_SYMBOL(pci_find_pcie_root_port);
531
532 /**
533 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
534 * @dev: the PCI device to operate on
535 * @pos: config space offset of status word
536 * @mask: mask of bit(s) to care about in status word
537 *
538 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
539 */
540 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
541 {
542 int i;
543
544 /* Wait for Transaction Pending bit clean */
545 for (i = 0; i < 4; i++) {
546 u16 status;
547 if (i)
548 msleep((1 << (i - 1)) * 100);
549
550 pci_read_config_word(dev, pos, &status);
551 if (!(status & mask))
552 return 1;
553 }
554
555 return 0;
556 }
557
558 /**
559 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
560 * @dev: PCI device to have its BARs restored
561 *
562 * Restore the BAR values for a given device, so as to make it
563 * accessible by its driver.
564 */
565 static void pci_restore_bars(struct pci_dev *dev)
566 {
567 int i;
568
569 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
570 pci_update_resource(dev, i);
571 }
572
573 static const struct pci_platform_pm_ops *pci_platform_pm;
574
575 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
576 {
577 if (!ops->is_manageable || !ops->set_state || !ops->get_state ||
578 !ops->choose_state || !ops->set_wakeup || !ops->need_resume)
579 return -EINVAL;
580 pci_platform_pm = ops;
581 return 0;
582 }
583
584 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
585 {
586 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
587 }
588
589 static inline int platform_pci_set_power_state(struct pci_dev *dev,
590 pci_power_t t)
591 {
592 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
593 }
594
595 static inline pci_power_t platform_pci_get_power_state(struct pci_dev *dev)
596 {
597 return pci_platform_pm ? pci_platform_pm->get_state(dev) : PCI_UNKNOWN;
598 }
599
600 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
601 {
602 return pci_platform_pm ?
603 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
604 }
605
606 static inline int platform_pci_set_wakeup(struct pci_dev *dev, bool enable)
607 {
608 return pci_platform_pm ?
609 pci_platform_pm->set_wakeup(dev, enable) : -ENODEV;
610 }
611
612 static inline bool platform_pci_need_resume(struct pci_dev *dev)
613 {
614 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
615 }
616
617 /**
618 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
619 * given PCI device
620 * @dev: PCI device to handle.
621 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
622 *
623 * RETURN VALUE:
624 * -EINVAL if the requested state is invalid.
625 * -EIO if device does not support PCI PM or its PM capabilities register has a
626 * wrong version, or device doesn't support the requested state.
627 * 0 if device already is in the requested state.
628 * 0 if device's power state has been successfully changed.
629 */
630 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
631 {
632 u16 pmcsr;
633 bool need_restore = false;
634
635 /* Check if we're already there */
636 if (dev->current_state == state)
637 return 0;
638
639 if (!dev->pm_cap)
640 return -EIO;
641
642 if (state < PCI_D0 || state > PCI_D3hot)
643 return -EINVAL;
644
645 /* Validate current state:
646 * Can enter D0 from any state, but if we can only go deeper
647 * to sleep if we're already in a low power state
648 */
649 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
650 && dev->current_state > state) {
651 dev_err(&dev->dev, "invalid power transition (from state %d to %d)\n",
652 dev->current_state, state);
653 return -EINVAL;
654 }
655
656 /* check if this device supports the desired state */
657 if ((state == PCI_D1 && !dev->d1_support)
658 || (state == PCI_D2 && !dev->d2_support))
659 return -EIO;
660
661 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
662
663 /* If we're (effectively) in D3, force entire word to 0.
664 * This doesn't affect PME_Status, disables PME_En, and
665 * sets PowerState to 0.
666 */
667 switch (dev->current_state) {
668 case PCI_D0:
669 case PCI_D1:
670 case PCI_D2:
671 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
672 pmcsr |= state;
673 break;
674 case PCI_D3hot:
675 case PCI_D3cold:
676 case PCI_UNKNOWN: /* Boot-up */
677 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
678 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
679 need_restore = true;
680 /* Fall-through: force to D0 */
681 default:
682 pmcsr = 0;
683 break;
684 }
685
686 /* enter specified state */
687 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
688
689 /* Mandatory power management transition delays */
690 /* see PCI PM 1.1 5.6.1 table 18 */
691 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
692 pci_dev_d3_sleep(dev);
693 else if (state == PCI_D2 || dev->current_state == PCI_D2)
694 udelay(PCI_PM_D2_DELAY);
695
696 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
697 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
698 if (dev->current_state != state && printk_ratelimit())
699 dev_info(&dev->dev, "Refused to change power state, currently in D%d\n",
700 dev->current_state);
701
702 /*
703 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
704 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
705 * from D3hot to D0 _may_ perform an internal reset, thereby
706 * going to "D0 Uninitialized" rather than "D0 Initialized".
707 * For example, at least some versions of the 3c905B and the
708 * 3c556B exhibit this behaviour.
709 *
710 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
711 * devices in a D3hot state at boot. Consequently, we need to
712 * restore at least the BARs so that the device will be
713 * accessible to its driver.
714 */
715 if (need_restore)
716 pci_restore_bars(dev);
717
718 if (dev->bus->self)
719 pcie_aspm_pm_state_change(dev->bus->self);
720
721 return 0;
722 }
723
724 /**
725 * pci_update_current_state - Read power state of given device and cache it
726 * @dev: PCI device to handle.
727 * @state: State to cache in case the device doesn't have the PM capability
728 *
729 * The power state is read from the PMCSR register, which however is
730 * inaccessible in D3cold. The platform firmware is therefore queried first
731 * to detect accessibility of the register. In case the platform firmware
732 * reports an incorrect state or the device isn't power manageable by the
733 * platform at all, we try to detect D3cold by testing accessibility of the
734 * vendor ID in config space.
735 */
736 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
737 {
738 if (platform_pci_get_power_state(dev) == PCI_D3cold ||
739 !pci_device_is_present(dev)) {
740 dev->current_state = PCI_D3cold;
741 } else if (dev->pm_cap) {
742 u16 pmcsr;
743
744 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
745 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
746 } else {
747 dev->current_state = state;
748 }
749 }
750
751 /**
752 * pci_power_up - Put the given device into D0 forcibly
753 * @dev: PCI device to power up
754 */
755 void pci_power_up(struct pci_dev *dev)
756 {
757 if (platform_pci_power_manageable(dev))
758 platform_pci_set_power_state(dev, PCI_D0);
759
760 pci_raw_set_power_state(dev, PCI_D0);
761 pci_update_current_state(dev, PCI_D0);
762 }
763
764 /**
765 * pci_platform_power_transition - Use platform to change device power state
766 * @dev: PCI device to handle.
767 * @state: State to put the device into.
768 */
769 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
770 {
771 int error;
772
773 if (platform_pci_power_manageable(dev)) {
774 error = platform_pci_set_power_state(dev, state);
775 if (!error)
776 pci_update_current_state(dev, state);
777 } else
778 error = -ENODEV;
779
780 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
781 dev->current_state = PCI_D0;
782
783 return error;
784 }
785
786 /**
787 * pci_wakeup - Wake up a PCI device
788 * @pci_dev: Device to handle.
789 * @ign: ignored parameter
790 */
791 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
792 {
793 pci_wakeup_event(pci_dev);
794 pm_request_resume(&pci_dev->dev);
795 return 0;
796 }
797
798 /**
799 * pci_wakeup_bus - Walk given bus and wake up devices on it
800 * @bus: Top bus of the subtree to walk.
801 */
802 static void pci_wakeup_bus(struct pci_bus *bus)
803 {
804 if (bus)
805 pci_walk_bus(bus, pci_wakeup, NULL);
806 }
807
808 /**
809 * __pci_start_power_transition - Start power transition of a PCI device
810 * @dev: PCI device to handle.
811 * @state: State to put the device into.
812 */
813 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
814 {
815 if (state == PCI_D0) {
816 pci_platform_power_transition(dev, PCI_D0);
817 /*
818 * Mandatory power management transition delays, see
819 * PCI Express Base Specification Revision 2.0 Section
820 * 6.6.1: Conventional Reset. Do not delay for
821 * devices powered on/off by corresponding bridge,
822 * because have already delayed for the bridge.
823 */
824 if (dev->runtime_d3cold) {
825 if (dev->d3cold_delay)
826 msleep(dev->d3cold_delay);
827 /*
828 * When powering on a bridge from D3cold, the
829 * whole hierarchy may be powered on into
830 * D0uninitialized state, resume them to give
831 * them a chance to suspend again
832 */
833 pci_wakeup_bus(dev->subordinate);
834 }
835 }
836 }
837
838 /**
839 * __pci_dev_set_current_state - Set current state of a PCI device
840 * @dev: Device to handle
841 * @data: pointer to state to be set
842 */
843 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
844 {
845 pci_power_t state = *(pci_power_t *)data;
846
847 dev->current_state = state;
848 return 0;
849 }
850
851 /**
852 * __pci_bus_set_current_state - Walk given bus and set current state of devices
853 * @bus: Top bus of the subtree to walk.
854 * @state: state to be set
855 */
856 static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
857 {
858 if (bus)
859 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
860 }
861
862 /**
863 * __pci_complete_power_transition - Complete power transition of a PCI device
864 * @dev: PCI device to handle.
865 * @state: State to put the device into.
866 *
867 * This function should not be called directly by device drivers.
868 */
869 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
870 {
871 int ret;
872
873 if (state <= PCI_D0)
874 return -EINVAL;
875 ret = pci_platform_power_transition(dev, state);
876 /* Power off the bridge may power off the whole hierarchy */
877 if (!ret && state == PCI_D3cold)
878 __pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
879 return ret;
880 }
881 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
882
883 /**
884 * pci_set_power_state - Set the power state of a PCI device
885 * @dev: PCI device to handle.
886 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
887 *
888 * Transition a device to a new power state, using the platform firmware and/or
889 * the device's PCI PM registers.
890 *
891 * RETURN VALUE:
892 * -EINVAL if the requested state is invalid.
893 * -EIO if device does not support PCI PM or its PM capabilities register has a
894 * wrong version, or device doesn't support the requested state.
895 * 0 if device already is in the requested state.
896 * 0 if device's power state has been successfully changed.
897 */
898 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
899 {
900 int error;
901
902 /* bound the state we're entering */
903 if (state > PCI_D3cold)
904 state = PCI_D3cold;
905 else if (state < PCI_D0)
906 state = PCI_D0;
907 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
908 /*
909 * If the device or the parent bridge do not support PCI PM,
910 * ignore the request if we're doing anything other than putting
911 * it into D0 (which would only happen on boot).
912 */
913 return 0;
914
915 /* Check if we're already there */
916 if (dev->current_state == state)
917 return 0;
918
919 __pci_start_power_transition(dev, state);
920
921 /* This device is quirked not to be put into D3, so
922 don't put it in D3 */
923 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
924 return 0;
925
926 /*
927 * To put device in D3cold, we put device into D3hot in native
928 * way, then put device into D3cold with platform ops
929 */
930 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
931 PCI_D3hot : state);
932
933 if (!__pci_complete_power_transition(dev, state))
934 error = 0;
935
936 return error;
937 }
938 EXPORT_SYMBOL(pci_set_power_state);
939
940 /**
941 * pci_choose_state - Choose the power state of a PCI device
942 * @dev: PCI device to be suspended
943 * @state: target sleep state for the whole system. This is the value
944 * that is passed to suspend() function.
945 *
946 * Returns PCI power state suitable for given device and given system
947 * message.
948 */
949
950 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
951 {
952 pci_power_t ret;
953
954 if (!dev->pm_cap)
955 return PCI_D0;
956
957 ret = platform_pci_choose_state(dev);
958 if (ret != PCI_POWER_ERROR)
959 return ret;
960
961 switch (state.event) {
962 case PM_EVENT_ON:
963 return PCI_D0;
964 case PM_EVENT_FREEZE:
965 case PM_EVENT_PRETHAW:
966 /* REVISIT both freeze and pre-thaw "should" use D0 */
967 case PM_EVENT_SUSPEND:
968 case PM_EVENT_HIBERNATE:
969 return PCI_D3hot;
970 default:
971 dev_info(&dev->dev, "unrecognized suspend event %d\n",
972 state.event);
973 BUG();
974 }
975 return PCI_D0;
976 }
977 EXPORT_SYMBOL(pci_choose_state);
978
979 #define PCI_EXP_SAVE_REGS 7
980
981 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
982 u16 cap, bool extended)
983 {
984 struct pci_cap_saved_state *tmp;
985
986 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
987 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
988 return tmp;
989 }
990 return NULL;
991 }
992
993 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
994 {
995 return _pci_find_saved_cap(dev, cap, false);
996 }
997
998 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
999 {
1000 return _pci_find_saved_cap(dev, cap, true);
1001 }
1002
1003 static int pci_save_pcie_state(struct pci_dev *dev)
1004 {
1005 int i = 0;
1006 struct pci_cap_saved_state *save_state;
1007 u16 *cap;
1008
1009 if (!pci_is_pcie(dev))
1010 return 0;
1011
1012 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1013 if (!save_state) {
1014 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1015 return -ENOMEM;
1016 }
1017
1018 cap = (u16 *)&save_state->cap.data[0];
1019 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
1020 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
1021 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
1022 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
1023 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
1024 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
1025 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
1026
1027 return 0;
1028 }
1029
1030 static void pci_restore_pcie_state(struct pci_dev *dev)
1031 {
1032 int i = 0;
1033 struct pci_cap_saved_state *save_state;
1034 u16 *cap;
1035
1036 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
1037 if (!save_state)
1038 return;
1039
1040 cap = (u16 *)&save_state->cap.data[0];
1041 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
1042 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1043 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1044 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1045 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1046 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1047 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1048 }
1049
1050
1051 static int pci_save_pcix_state(struct pci_dev *dev)
1052 {
1053 int pos;
1054 struct pci_cap_saved_state *save_state;
1055
1056 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1057 if (!pos)
1058 return 0;
1059
1060 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1061 if (!save_state) {
1062 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1063 return -ENOMEM;
1064 }
1065
1066 pci_read_config_word(dev, pos + PCI_X_CMD,
1067 (u16 *)save_state->cap.data);
1068
1069 return 0;
1070 }
1071
1072 static void pci_restore_pcix_state(struct pci_dev *dev)
1073 {
1074 int i = 0, pos;
1075 struct pci_cap_saved_state *save_state;
1076 u16 *cap;
1077
1078 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1079 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1080 if (!save_state || !pos)
1081 return;
1082 cap = (u16 *)&save_state->cap.data[0];
1083
1084 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1085 }
1086
1087
1088 /**
1089 * pci_save_state - save the PCI configuration space of a device before suspending
1090 * @dev: - PCI device that we're dealing with
1091 */
1092 int pci_save_state(struct pci_dev *dev)
1093 {
1094 int i;
1095 /* XXX: 100% dword access ok here? */
1096 for (i = 0; i < 16; i++)
1097 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1098 dev->state_saved = true;
1099
1100 i = pci_save_pcie_state(dev);
1101 if (i != 0)
1102 return i;
1103
1104 i = pci_save_pcix_state(dev);
1105 if (i != 0)
1106 return i;
1107
1108 return pci_save_vc_state(dev);
1109 }
1110 EXPORT_SYMBOL(pci_save_state);
1111
1112 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1113 u32 saved_val, int retry)
1114 {
1115 u32 val;
1116
1117 pci_read_config_dword(pdev, offset, &val);
1118 if (val == saved_val)
1119 return;
1120
1121 for (;;) {
1122 dev_dbg(&pdev->dev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1123 offset, val, saved_val);
1124 pci_write_config_dword(pdev, offset, saved_val);
1125 if (retry-- <= 0)
1126 return;
1127
1128 pci_read_config_dword(pdev, offset, &val);
1129 if (val == saved_val)
1130 return;
1131
1132 mdelay(1);
1133 }
1134 }
1135
1136 static void pci_restore_config_space_range(struct pci_dev *pdev,
1137 int start, int end, int retry)
1138 {
1139 int index;
1140
1141 for (index = end; index >= start; index--)
1142 pci_restore_config_dword(pdev, 4 * index,
1143 pdev->saved_config_space[index],
1144 retry);
1145 }
1146
1147 static void pci_restore_config_space(struct pci_dev *pdev)
1148 {
1149 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1150 pci_restore_config_space_range(pdev, 10, 15, 0);
1151 /* Restore BARs before the command register. */
1152 pci_restore_config_space_range(pdev, 4, 9, 10);
1153 pci_restore_config_space_range(pdev, 0, 3, 0);
1154 } else {
1155 pci_restore_config_space_range(pdev, 0, 15, 0);
1156 }
1157 }
1158
1159 /**
1160 * pci_restore_state - Restore the saved state of a PCI device
1161 * @dev: - PCI device that we're dealing with
1162 */
1163 void pci_restore_state(struct pci_dev *dev)
1164 {
1165 if (!dev->state_saved)
1166 return;
1167
1168 /* PCI Express register must be restored first */
1169 pci_restore_pcie_state(dev);
1170 pci_restore_pasid_state(dev);
1171 pci_restore_pri_state(dev);
1172 pci_restore_ats_state(dev);
1173 pci_restore_vc_state(dev);
1174
1175 pci_cleanup_aer_error_status_regs(dev);
1176
1177 pci_restore_config_space(dev);
1178
1179 pci_restore_pcix_state(dev);
1180 pci_restore_msi_state(dev);
1181
1182 /* Restore ACS and IOV configuration state */
1183 pci_enable_acs(dev);
1184 pci_restore_iov_state(dev);
1185
1186 dev->state_saved = false;
1187 }
1188 EXPORT_SYMBOL(pci_restore_state);
1189
1190 struct pci_saved_state {
1191 u32 config_space[16];
1192 struct pci_cap_saved_data cap[0];
1193 };
1194
1195 /**
1196 * pci_store_saved_state - Allocate and return an opaque struct containing
1197 * the device saved state.
1198 * @dev: PCI device that we're dealing with
1199 *
1200 * Return NULL if no state or error.
1201 */
1202 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1203 {
1204 struct pci_saved_state *state;
1205 struct pci_cap_saved_state *tmp;
1206 struct pci_cap_saved_data *cap;
1207 size_t size;
1208
1209 if (!dev->state_saved)
1210 return NULL;
1211
1212 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1213
1214 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1215 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1216
1217 state = kzalloc(size, GFP_KERNEL);
1218 if (!state)
1219 return NULL;
1220
1221 memcpy(state->config_space, dev->saved_config_space,
1222 sizeof(state->config_space));
1223
1224 cap = state->cap;
1225 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1226 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1227 memcpy(cap, &tmp->cap, len);
1228 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1229 }
1230 /* Empty cap_save terminates list */
1231
1232 return state;
1233 }
1234 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1235
1236 /**
1237 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1238 * @dev: PCI device that we're dealing with
1239 * @state: Saved state returned from pci_store_saved_state()
1240 */
1241 int pci_load_saved_state(struct pci_dev *dev,
1242 struct pci_saved_state *state)
1243 {
1244 struct pci_cap_saved_data *cap;
1245
1246 dev->state_saved = false;
1247
1248 if (!state)
1249 return 0;
1250
1251 memcpy(dev->saved_config_space, state->config_space,
1252 sizeof(state->config_space));
1253
1254 cap = state->cap;
1255 while (cap->size) {
1256 struct pci_cap_saved_state *tmp;
1257
1258 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1259 if (!tmp || tmp->cap.size != cap->size)
1260 return -EINVAL;
1261
1262 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1263 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1264 sizeof(struct pci_cap_saved_data) + cap->size);
1265 }
1266
1267 dev->state_saved = true;
1268 return 0;
1269 }
1270 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1271
1272 /**
1273 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1274 * and free the memory allocated for it.
1275 * @dev: PCI device that we're dealing with
1276 * @state: Pointer to saved state returned from pci_store_saved_state()
1277 */
1278 int pci_load_and_free_saved_state(struct pci_dev *dev,
1279 struct pci_saved_state **state)
1280 {
1281 int ret = pci_load_saved_state(dev, *state);
1282 kfree(*state);
1283 *state = NULL;
1284 return ret;
1285 }
1286 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1287
1288 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1289 {
1290 return pci_enable_resources(dev, bars);
1291 }
1292
1293 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1294 {
1295 int err;
1296 struct pci_dev *bridge;
1297 u16 cmd;
1298 u8 pin;
1299
1300 err = pci_set_power_state(dev, PCI_D0);
1301 if (err < 0 && err != -EIO)
1302 return err;
1303
1304 bridge = pci_upstream_bridge(dev);
1305 if (bridge)
1306 pcie_aspm_powersave_config_link(bridge);
1307
1308 err = pcibios_enable_device(dev, bars);
1309 if (err < 0)
1310 return err;
1311 pci_fixup_device(pci_fixup_enable, dev);
1312
1313 if (dev->msi_enabled || dev->msix_enabled)
1314 return 0;
1315
1316 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1317 if (pin) {
1318 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1319 if (cmd & PCI_COMMAND_INTX_DISABLE)
1320 pci_write_config_word(dev, PCI_COMMAND,
1321 cmd & ~PCI_COMMAND_INTX_DISABLE);
1322 }
1323
1324 return 0;
1325 }
1326
1327 /**
1328 * pci_reenable_device - Resume abandoned device
1329 * @dev: PCI device to be resumed
1330 *
1331 * Note this function is a backend of pci_default_resume and is not supposed
1332 * to be called by normal code, write proper resume handler and use it instead.
1333 */
1334 int pci_reenable_device(struct pci_dev *dev)
1335 {
1336 if (pci_is_enabled(dev))
1337 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1338 return 0;
1339 }
1340 EXPORT_SYMBOL(pci_reenable_device);
1341
1342 static void pci_enable_bridge(struct pci_dev *dev)
1343 {
1344 struct pci_dev *bridge;
1345 int retval;
1346
1347 bridge = pci_upstream_bridge(dev);
1348 if (bridge)
1349 pci_enable_bridge(bridge);
1350
1351 if (pci_is_enabled(dev)) {
1352 if (!dev->is_busmaster)
1353 pci_set_master(dev);
1354 return;
1355 }
1356
1357 retval = pci_enable_device(dev);
1358 if (retval)
1359 dev_err(&dev->dev, "Error enabling bridge (%d), continuing\n",
1360 retval);
1361 pci_set_master(dev);
1362 }
1363
1364 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1365 {
1366 struct pci_dev *bridge;
1367 int err;
1368 int i, bars = 0;
1369
1370 /*
1371 * Power state could be unknown at this point, either due to a fresh
1372 * boot or a device removal call. So get the current power state
1373 * so that things like MSI message writing will behave as expected
1374 * (e.g. if the device really is in D0 at enable time).
1375 */
1376 if (dev->pm_cap) {
1377 u16 pmcsr;
1378 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1379 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1380 }
1381
1382 if (atomic_inc_return(&dev->enable_cnt) > 1)
1383 return 0; /* already enabled */
1384
1385 bridge = pci_upstream_bridge(dev);
1386 if (bridge)
1387 pci_enable_bridge(bridge);
1388
1389 /* only skip sriov related */
1390 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1391 if (dev->resource[i].flags & flags)
1392 bars |= (1 << i);
1393 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1394 if (dev->resource[i].flags & flags)
1395 bars |= (1 << i);
1396
1397 err = do_pci_enable_device(dev, bars);
1398 if (err < 0)
1399 atomic_dec(&dev->enable_cnt);
1400 return err;
1401 }
1402
1403 /**
1404 * pci_enable_device_io - Initialize a device for use with IO space
1405 * @dev: PCI device to be initialized
1406 *
1407 * Initialize device before it's used by a driver. Ask low-level code
1408 * to enable I/O resources. Wake up the device if it was suspended.
1409 * Beware, this function can fail.
1410 */
1411 int pci_enable_device_io(struct pci_dev *dev)
1412 {
1413 return pci_enable_device_flags(dev, IORESOURCE_IO);
1414 }
1415 EXPORT_SYMBOL(pci_enable_device_io);
1416
1417 /**
1418 * pci_enable_device_mem - Initialize a device for use with Memory space
1419 * @dev: PCI device to be initialized
1420 *
1421 * Initialize device before it's used by a driver. Ask low-level code
1422 * to enable Memory resources. Wake up the device if it was suspended.
1423 * Beware, this function can fail.
1424 */
1425 int pci_enable_device_mem(struct pci_dev *dev)
1426 {
1427 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1428 }
1429 EXPORT_SYMBOL(pci_enable_device_mem);
1430
1431 /**
1432 * pci_enable_device - Initialize device before it's used by a driver.
1433 * @dev: PCI device to be initialized
1434 *
1435 * Initialize device before it's used by a driver. Ask low-level code
1436 * to enable I/O and memory. Wake up the device if it was suspended.
1437 * Beware, this function can fail.
1438 *
1439 * Note we don't actually enable the device many times if we call
1440 * this function repeatedly (we just increment the count).
1441 */
1442 int pci_enable_device(struct pci_dev *dev)
1443 {
1444 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1445 }
1446 EXPORT_SYMBOL(pci_enable_device);
1447
1448 /*
1449 * Managed PCI resources. This manages device on/off, intx/msi/msix
1450 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1451 * there's no need to track it separately. pci_devres is initialized
1452 * when a device is enabled using managed PCI device enable interface.
1453 */
1454 struct pci_devres {
1455 unsigned int enabled:1;
1456 unsigned int pinned:1;
1457 unsigned int orig_intx:1;
1458 unsigned int restore_intx:1;
1459 u32 region_mask;
1460 };
1461
1462 static void pcim_release(struct device *gendev, void *res)
1463 {
1464 struct pci_dev *dev = to_pci_dev(gendev);
1465 struct pci_devres *this = res;
1466 int i;
1467
1468 if (dev->msi_enabled)
1469 pci_disable_msi(dev);
1470 if (dev->msix_enabled)
1471 pci_disable_msix(dev);
1472
1473 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1474 if (this->region_mask & (1 << i))
1475 pci_release_region(dev, i);
1476
1477 if (this->restore_intx)
1478 pci_intx(dev, this->orig_intx);
1479
1480 if (this->enabled && !this->pinned)
1481 pci_disable_device(dev);
1482 }
1483
1484 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1485 {
1486 struct pci_devres *dr, *new_dr;
1487
1488 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1489 if (dr)
1490 return dr;
1491
1492 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1493 if (!new_dr)
1494 return NULL;
1495 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1496 }
1497
1498 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1499 {
1500 if (pci_is_managed(pdev))
1501 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1502 return NULL;
1503 }
1504
1505 /**
1506 * pcim_enable_device - Managed pci_enable_device()
1507 * @pdev: PCI device to be initialized
1508 *
1509 * Managed pci_enable_device().
1510 */
1511 int pcim_enable_device(struct pci_dev *pdev)
1512 {
1513 struct pci_devres *dr;
1514 int rc;
1515
1516 dr = get_pci_dr(pdev);
1517 if (unlikely(!dr))
1518 return -ENOMEM;
1519 if (dr->enabled)
1520 return 0;
1521
1522 rc = pci_enable_device(pdev);
1523 if (!rc) {
1524 pdev->is_managed = 1;
1525 dr->enabled = 1;
1526 }
1527 return rc;
1528 }
1529 EXPORT_SYMBOL(pcim_enable_device);
1530
1531 /**
1532 * pcim_pin_device - Pin managed PCI device
1533 * @pdev: PCI device to pin
1534 *
1535 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1536 * driver detach. @pdev must have been enabled with
1537 * pcim_enable_device().
1538 */
1539 void pcim_pin_device(struct pci_dev *pdev)
1540 {
1541 struct pci_devres *dr;
1542
1543 dr = find_pci_dr(pdev);
1544 WARN_ON(!dr || !dr->enabled);
1545 if (dr)
1546 dr->pinned = 1;
1547 }
1548 EXPORT_SYMBOL(pcim_pin_device);
1549
1550 /*
1551 * pcibios_add_device - provide arch specific hooks when adding device dev
1552 * @dev: the PCI device being added
1553 *
1554 * Permits the platform to provide architecture specific functionality when
1555 * devices are added. This is the default implementation. Architecture
1556 * implementations can override this.
1557 */
1558 int __weak pcibios_add_device(struct pci_dev *dev)
1559 {
1560 return 0;
1561 }
1562
1563 /**
1564 * pcibios_release_device - provide arch specific hooks when releasing device dev
1565 * @dev: the PCI device being released
1566 *
1567 * Permits the platform to provide architecture specific functionality when
1568 * devices are released. This is the default implementation. Architecture
1569 * implementations can override this.
1570 */
1571 void __weak pcibios_release_device(struct pci_dev *dev) {}
1572
1573 /**
1574 * pcibios_disable_device - disable arch specific PCI resources for device dev
1575 * @dev: the PCI device to disable
1576 *
1577 * Disables architecture specific PCI resources for the device. This
1578 * is the default implementation. Architecture implementations can
1579 * override this.
1580 */
1581 void __weak pcibios_disable_device(struct pci_dev *dev) {}
1582
1583 /**
1584 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1585 * @irq: ISA IRQ to penalize
1586 * @active: IRQ active or not
1587 *
1588 * Permits the platform to provide architecture-specific functionality when
1589 * penalizing ISA IRQs. This is the default implementation. Architecture
1590 * implementations can override this.
1591 */
1592 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1593
1594 static void do_pci_disable_device(struct pci_dev *dev)
1595 {
1596 u16 pci_command;
1597
1598 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1599 if (pci_command & PCI_COMMAND_MASTER) {
1600 pci_command &= ~PCI_COMMAND_MASTER;
1601 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1602 }
1603
1604 pcibios_disable_device(dev);
1605 }
1606
1607 /**
1608 * pci_disable_enabled_device - Disable device without updating enable_cnt
1609 * @dev: PCI device to disable
1610 *
1611 * NOTE: This function is a backend of PCI power management routines and is
1612 * not supposed to be called drivers.
1613 */
1614 void pci_disable_enabled_device(struct pci_dev *dev)
1615 {
1616 if (pci_is_enabled(dev))
1617 do_pci_disable_device(dev);
1618 }
1619
1620 /**
1621 * pci_disable_device - Disable PCI device after use
1622 * @dev: PCI device to be disabled
1623 *
1624 * Signal to the system that the PCI device is not in use by the system
1625 * anymore. This only involves disabling PCI bus-mastering, if active.
1626 *
1627 * Note we don't actually disable the device until all callers of
1628 * pci_enable_device() have called pci_disable_device().
1629 */
1630 void pci_disable_device(struct pci_dev *dev)
1631 {
1632 struct pci_devres *dr;
1633
1634 dr = find_pci_dr(dev);
1635 if (dr)
1636 dr->enabled = 0;
1637
1638 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1639 "disabling already-disabled device");
1640
1641 if (atomic_dec_return(&dev->enable_cnt) != 0)
1642 return;
1643
1644 do_pci_disable_device(dev);
1645
1646 dev->is_busmaster = 0;
1647 }
1648 EXPORT_SYMBOL(pci_disable_device);
1649
1650 /**
1651 * pcibios_set_pcie_reset_state - set reset state for device dev
1652 * @dev: the PCIe device reset
1653 * @state: Reset state to enter into
1654 *
1655 *
1656 * Sets the PCIe reset state for the device. This is the default
1657 * implementation. Architecture implementations can override this.
1658 */
1659 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1660 enum pcie_reset_state state)
1661 {
1662 return -EINVAL;
1663 }
1664
1665 /**
1666 * pci_set_pcie_reset_state - set reset state for device dev
1667 * @dev: the PCIe device reset
1668 * @state: Reset state to enter into
1669 *
1670 *
1671 * Sets the PCI reset state for the device.
1672 */
1673 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1674 {
1675 return pcibios_set_pcie_reset_state(dev, state);
1676 }
1677 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1678
1679 /**
1680 * pci_check_pme_status - Check if given device has generated PME.
1681 * @dev: Device to check.
1682 *
1683 * Check the PME status of the device and if set, clear it and clear PME enable
1684 * (if set). Return 'true' if PME status and PME enable were both set or
1685 * 'false' otherwise.
1686 */
1687 bool pci_check_pme_status(struct pci_dev *dev)
1688 {
1689 int pmcsr_pos;
1690 u16 pmcsr;
1691 bool ret = false;
1692
1693 if (!dev->pm_cap)
1694 return false;
1695
1696 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1697 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1698 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1699 return false;
1700
1701 /* Clear PME status. */
1702 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1703 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1704 /* Disable PME to avoid interrupt flood. */
1705 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1706 ret = true;
1707 }
1708
1709 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1710
1711 return ret;
1712 }
1713
1714 /**
1715 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1716 * @dev: Device to handle.
1717 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1718 *
1719 * Check if @dev has generated PME and queue a resume request for it in that
1720 * case.
1721 */
1722 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1723 {
1724 if (pme_poll_reset && dev->pme_poll)
1725 dev->pme_poll = false;
1726
1727 if (pci_check_pme_status(dev)) {
1728 pci_wakeup_event(dev);
1729 pm_request_resume(&dev->dev);
1730 }
1731 return 0;
1732 }
1733
1734 /**
1735 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1736 * @bus: Top bus of the subtree to walk.
1737 */
1738 void pci_pme_wakeup_bus(struct pci_bus *bus)
1739 {
1740 if (bus)
1741 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1742 }
1743
1744
1745 /**
1746 * pci_pme_capable - check the capability of PCI device to generate PME#
1747 * @dev: PCI device to handle.
1748 * @state: PCI state from which device will issue PME#.
1749 */
1750 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1751 {
1752 if (!dev->pm_cap)
1753 return false;
1754
1755 return !!(dev->pme_support & (1 << state));
1756 }
1757 EXPORT_SYMBOL(pci_pme_capable);
1758
1759 static void pci_pme_list_scan(struct work_struct *work)
1760 {
1761 struct pci_pme_device *pme_dev, *n;
1762
1763 mutex_lock(&pci_pme_list_mutex);
1764 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1765 if (pme_dev->dev->pme_poll) {
1766 struct pci_dev *bridge;
1767
1768 bridge = pme_dev->dev->bus->self;
1769 /*
1770 * If bridge is in low power state, the
1771 * configuration space of subordinate devices
1772 * may be not accessible
1773 */
1774 if (bridge && bridge->current_state != PCI_D0)
1775 continue;
1776 pci_pme_wakeup(pme_dev->dev, NULL);
1777 } else {
1778 list_del(&pme_dev->list);
1779 kfree(pme_dev);
1780 }
1781 }
1782 if (!list_empty(&pci_pme_list))
1783 queue_delayed_work(system_freezable_wq, &pci_pme_work,
1784 msecs_to_jiffies(PME_TIMEOUT));
1785 mutex_unlock(&pci_pme_list_mutex);
1786 }
1787
1788 static void __pci_pme_active(struct pci_dev *dev, bool enable)
1789 {
1790 u16 pmcsr;
1791
1792 if (!dev->pme_support)
1793 return;
1794
1795 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1796 /* Clear PME_Status by writing 1 to it and enable PME# */
1797 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1798 if (!enable)
1799 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1800
1801 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1802 }
1803
1804 /**
1805 * pci_pme_restore - Restore PME configuration after config space restore.
1806 * @dev: PCI device to update.
1807 */
1808 void pci_pme_restore(struct pci_dev *dev)
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 if (dev->wakeup_prepared) {
1817 pmcsr |= PCI_PM_CTRL_PME_ENABLE;
1818 pmcsr &= ~PCI_PM_CTRL_PME_STATUS;
1819 } else {
1820 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1821 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1822 }
1823 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1824 }
1825
1826 /**
1827 * pci_pme_active - enable or disable PCI device's PME# function
1828 * @dev: PCI device to handle.
1829 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1830 *
1831 * The caller must verify that the device is capable of generating PME# before
1832 * calling this function with @enable equal to 'true'.
1833 */
1834 void pci_pme_active(struct pci_dev *dev, bool enable)
1835 {
1836 __pci_pme_active(dev, enable);
1837
1838 /*
1839 * PCI (as opposed to PCIe) PME requires that the device have
1840 * its PME# line hooked up correctly. Not all hardware vendors
1841 * do this, so the PME never gets delivered and the device
1842 * remains asleep. The easiest way around this is to
1843 * periodically walk the list of suspended devices and check
1844 * whether any have their PME flag set. The assumption is that
1845 * we'll wake up often enough anyway that this won't be a huge
1846 * hit, and the power savings from the devices will still be a
1847 * win.
1848 *
1849 * Although PCIe uses in-band PME message instead of PME# line
1850 * to report PME, PME does not work for some PCIe devices in
1851 * reality. For example, there are devices that set their PME
1852 * status bits, but don't really bother to send a PME message;
1853 * there are PCI Express Root Ports that don't bother to
1854 * trigger interrupts when they receive PME messages from the
1855 * devices below. So PME poll is used for PCIe devices too.
1856 */
1857
1858 if (dev->pme_poll) {
1859 struct pci_pme_device *pme_dev;
1860 if (enable) {
1861 pme_dev = kmalloc(sizeof(struct pci_pme_device),
1862 GFP_KERNEL);
1863 if (!pme_dev) {
1864 dev_warn(&dev->dev, "can't enable PME#\n");
1865 return;
1866 }
1867 pme_dev->dev = dev;
1868 mutex_lock(&pci_pme_list_mutex);
1869 list_add(&pme_dev->list, &pci_pme_list);
1870 if (list_is_singular(&pci_pme_list))
1871 queue_delayed_work(system_freezable_wq,
1872 &pci_pme_work,
1873 msecs_to_jiffies(PME_TIMEOUT));
1874 mutex_unlock(&pci_pme_list_mutex);
1875 } else {
1876 mutex_lock(&pci_pme_list_mutex);
1877 list_for_each_entry(pme_dev, &pci_pme_list, list) {
1878 if (pme_dev->dev == dev) {
1879 list_del(&pme_dev->list);
1880 kfree(pme_dev);
1881 break;
1882 }
1883 }
1884 mutex_unlock(&pci_pme_list_mutex);
1885 }
1886 }
1887
1888 dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1889 }
1890 EXPORT_SYMBOL(pci_pme_active);
1891
1892 /**
1893 * __pci_enable_wake - enable PCI device as wakeup event source
1894 * @dev: PCI device affected
1895 * @state: PCI state from which device will issue wakeup events
1896 * @enable: True to enable event generation; false to disable
1897 *
1898 * This enables the device as a wakeup event source, or disables it.
1899 * When such events involves platform-specific hooks, those hooks are
1900 * called automatically by this routine.
1901 *
1902 * Devices with legacy power management (no standard PCI PM capabilities)
1903 * always require such platform hooks.
1904 *
1905 * RETURN VALUE:
1906 * 0 is returned on success
1907 * -EINVAL is returned if device is not supposed to wake up the system
1908 * Error code depending on the platform is returned if both the platform and
1909 * the native mechanism fail to enable the generation of wake-up events
1910 */
1911 static int __pci_enable_wake(struct pci_dev *dev, pci_power_t state, bool enable)
1912 {
1913 int ret = 0;
1914
1915 /* Don't do the same thing twice in a row for one device. */
1916 if (!!enable == !!dev->wakeup_prepared)
1917 return 0;
1918
1919 /*
1920 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1921 * Anderson we should be doing PME# wake enable followed by ACPI wake
1922 * enable. To disable wake-up we call the platform first, for symmetry.
1923 */
1924
1925 if (enable) {
1926 int error;
1927
1928 if (pci_pme_capable(dev, state))
1929 pci_pme_active(dev, true);
1930 else
1931 ret = 1;
1932 error = platform_pci_set_wakeup(dev, true);
1933 if (ret)
1934 ret = error;
1935 if (!ret)
1936 dev->wakeup_prepared = true;
1937 } else {
1938 platform_pci_set_wakeup(dev, false);
1939 pci_pme_active(dev, false);
1940 dev->wakeup_prepared = false;
1941 }
1942
1943 return ret;
1944 }
1945
1946 /**
1947 * pci_enable_wake - change wakeup settings for a PCI device
1948 * @pci_dev: Target device
1949 * @state: PCI state from which device will issue wakeup events
1950 * @enable: Whether or not to enable event generation
1951 *
1952 * If @enable is set, check device_may_wakeup() for the device before calling
1953 * __pci_enable_wake() for it.
1954 */
1955 int pci_enable_wake(struct pci_dev *pci_dev, pci_power_t state, bool enable)
1956 {
1957 if (enable && !device_may_wakeup(&pci_dev->dev))
1958 return -EINVAL;
1959
1960 return __pci_enable_wake(pci_dev, state, enable);
1961 }
1962 EXPORT_SYMBOL(pci_enable_wake);
1963
1964 /**
1965 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1966 * @dev: PCI device to prepare
1967 * @enable: True to enable wake-up event generation; false to disable
1968 *
1969 * Many drivers want the device to wake up the system from D3_hot or D3_cold
1970 * and this function allows them to set that up cleanly - pci_enable_wake()
1971 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1972 * ordering constraints.
1973 *
1974 * This function only returns error code if the device is not allowed to wake
1975 * up the system from sleep or it is not capable of generating PME# from both
1976 * D3_hot and D3_cold and the platform is unable to enable wake-up power for it.
1977 */
1978 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1979 {
1980 return pci_pme_capable(dev, PCI_D3cold) ?
1981 pci_enable_wake(dev, PCI_D3cold, enable) :
1982 pci_enable_wake(dev, PCI_D3hot, enable);
1983 }
1984 EXPORT_SYMBOL(pci_wake_from_d3);
1985
1986 /**
1987 * pci_target_state - find an appropriate low power state for a given PCI dev
1988 * @dev: PCI device
1989 * @wakeup: Whether or not wakeup functionality will be enabled for the device.
1990 *
1991 * Use underlying platform code to find a supported low power state for @dev.
1992 * If the platform can't manage @dev, return the deepest state from which it
1993 * can generate wake events, based on any available PME info.
1994 */
1995 static pci_power_t pci_target_state(struct pci_dev *dev, bool wakeup)
1996 {
1997 pci_power_t target_state = PCI_D3hot;
1998
1999 if (platform_pci_power_manageable(dev)) {
2000 /*
2001 * Call the platform to choose the target state of the device
2002 * and enable wake-up from this state if supported.
2003 */
2004 pci_power_t state = platform_pci_choose_state(dev);
2005
2006 switch (state) {
2007 case PCI_POWER_ERROR:
2008 case PCI_UNKNOWN:
2009 break;
2010 case PCI_D1:
2011 case PCI_D2:
2012 if (pci_no_d1d2(dev))
2013 break;
2014 default:
2015 target_state = state;
2016 }
2017
2018 return target_state;
2019 }
2020
2021 if (!dev->pm_cap)
2022 target_state = PCI_D0;
2023
2024 /*
2025 * If the device is in D3cold even though it's not power-manageable by
2026 * the platform, it may have been powered down by non-standard means.
2027 * Best to let it slumber.
2028 */
2029 if (dev->current_state == PCI_D3cold)
2030 target_state = PCI_D3cold;
2031
2032 if (wakeup) {
2033 /*
2034 * Find the deepest state from which the device can generate
2035 * wake-up events, make it the target state and enable device
2036 * to generate PME#.
2037 */
2038 if (dev->pme_support) {
2039 while (target_state
2040 && !(dev->pme_support & (1 << target_state)))
2041 target_state--;
2042 }
2043 }
2044
2045 return target_state;
2046 }
2047
2048 /**
2049 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
2050 * @dev: Device to handle.
2051 *
2052 * Choose the power state appropriate for the device depending on whether
2053 * it can wake up the system and/or is power manageable by the platform
2054 * (PCI_D3hot is the default) and put the device into that state.
2055 */
2056 int pci_prepare_to_sleep(struct pci_dev *dev)
2057 {
2058 bool wakeup = device_may_wakeup(&dev->dev);
2059 pci_power_t target_state = pci_target_state(dev, wakeup);
2060 int error;
2061
2062 if (target_state == PCI_POWER_ERROR)
2063 return -EIO;
2064
2065 pci_enable_wake(dev, target_state, wakeup);
2066
2067 error = pci_set_power_state(dev, target_state);
2068
2069 if (error)
2070 pci_enable_wake(dev, target_state, false);
2071
2072 return error;
2073 }
2074 EXPORT_SYMBOL(pci_prepare_to_sleep);
2075
2076 /**
2077 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
2078 * @dev: Device to handle.
2079 *
2080 * Disable device's system wake-up capability and put it into D0.
2081 */
2082 int pci_back_from_sleep(struct pci_dev *dev)
2083 {
2084 pci_enable_wake(dev, PCI_D0, false);
2085 return pci_set_power_state(dev, PCI_D0);
2086 }
2087 EXPORT_SYMBOL(pci_back_from_sleep);
2088
2089 /**
2090 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2091 * @dev: PCI device being suspended.
2092 *
2093 * Prepare @dev to generate wake-up events at run time and put it into a low
2094 * power state.
2095 */
2096 int pci_finish_runtime_suspend(struct pci_dev *dev)
2097 {
2098 pci_power_t target_state;
2099 int error;
2100
2101 target_state = pci_target_state(dev, device_can_wakeup(&dev->dev));
2102 if (target_state == PCI_POWER_ERROR)
2103 return -EIO;
2104
2105 dev->runtime_d3cold = target_state == PCI_D3cold;
2106
2107 __pci_enable_wake(dev, target_state, pci_dev_run_wake(dev));
2108
2109 error = pci_set_power_state(dev, target_state);
2110
2111 if (error) {
2112 pci_enable_wake(dev, target_state, false);
2113 dev->runtime_d3cold = false;
2114 }
2115
2116 return error;
2117 }
2118
2119 /**
2120 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2121 * @dev: Device to check.
2122 *
2123 * Return true if the device itself is capable of generating wake-up events
2124 * (through the platform or using the native PCIe PME) or if the device supports
2125 * PME and one of its upstream bridges can generate wake-up events.
2126 */
2127 bool pci_dev_run_wake(struct pci_dev *dev)
2128 {
2129 struct pci_bus *bus = dev->bus;
2130
2131 if (!dev->pme_support)
2132 return false;
2133
2134 /* PME-capable in principle, but not from the target power state */
2135 if (!pci_pme_capable(dev, pci_target_state(dev, true)))
2136 return false;
2137
2138 if (device_can_wakeup(&dev->dev))
2139 return true;
2140
2141 while (bus->parent) {
2142 struct pci_dev *bridge = bus->self;
2143
2144 if (device_can_wakeup(&bridge->dev))
2145 return true;
2146
2147 bus = bus->parent;
2148 }
2149
2150 /* We have reached the root bus. */
2151 if (bus->bridge)
2152 return device_can_wakeup(bus->bridge);
2153
2154 return false;
2155 }
2156 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2157
2158 /**
2159 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2160 * @pci_dev: Device to check.
2161 *
2162 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2163 * reconfigured due to wakeup settings difference between system and runtime
2164 * suspend and the current power state of it is suitable for the upcoming
2165 * (system) transition.
2166 *
2167 * If the device is not configured for system wakeup, disable PME for it before
2168 * returning 'true' to prevent it from waking up the system unnecessarily.
2169 */
2170 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2171 {
2172 struct device *dev = &pci_dev->dev;
2173 bool wakeup = device_may_wakeup(dev);
2174
2175 if (!pm_runtime_suspended(dev)
2176 || pci_target_state(pci_dev, wakeup) != pci_dev->current_state
2177 || platform_pci_need_resume(pci_dev)
2178 || (pci_dev->dev_flags & PCI_DEV_FLAGS_NEEDS_RESUME))
2179 return false;
2180
2181 /*
2182 * At this point the device is good to go unless it's been configured
2183 * to generate PME at the runtime suspend time, but it is not supposed
2184 * to wake up the system. In that case, simply disable PME for it
2185 * (it will have to be re-enabled on exit from system resume).
2186 *
2187 * If the device's power state is D3cold and the platform check above
2188 * hasn't triggered, the device's configuration is suitable and we don't
2189 * need to manipulate it at all.
2190 */
2191 spin_lock_irq(&dev->power.lock);
2192
2193 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2194 !wakeup)
2195 __pci_pme_active(pci_dev, false);
2196
2197 spin_unlock_irq(&dev->power.lock);
2198 return true;
2199 }
2200
2201 /**
2202 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2203 * @pci_dev: Device to handle.
2204 *
2205 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2206 * it might have been disabled during the prepare phase of system suspend if
2207 * the device was not configured for system wakeup.
2208 */
2209 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2210 {
2211 struct device *dev = &pci_dev->dev;
2212
2213 if (!pci_dev_run_wake(pci_dev))
2214 return;
2215
2216 spin_lock_irq(&dev->power.lock);
2217
2218 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2219 __pci_pme_active(pci_dev, true);
2220
2221 spin_unlock_irq(&dev->power.lock);
2222 }
2223
2224 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2225 {
2226 struct device *dev = &pdev->dev;
2227 struct device *parent = dev->parent;
2228
2229 if (parent)
2230 pm_runtime_get_sync(parent);
2231 pm_runtime_get_noresume(dev);
2232 /*
2233 * pdev->current_state is set to PCI_D3cold during suspending,
2234 * so wait until suspending completes
2235 */
2236 pm_runtime_barrier(dev);
2237 /*
2238 * Only need to resume devices in D3cold, because config
2239 * registers are still accessible for devices suspended but
2240 * not in D3cold.
2241 */
2242 if (pdev->current_state == PCI_D3cold)
2243 pm_runtime_resume(dev);
2244 }
2245
2246 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2247 {
2248 struct device *dev = &pdev->dev;
2249 struct device *parent = dev->parent;
2250
2251 pm_runtime_put(dev);
2252 if (parent)
2253 pm_runtime_put_sync(parent);
2254 }
2255
2256 /**
2257 * pci_bridge_d3_possible - Is it possible to put the bridge into D3
2258 * @bridge: Bridge to check
2259 *
2260 * This function checks if it is possible to move the bridge to D3.
2261 * Currently we only allow D3 for recent enough PCIe ports.
2262 */
2263 bool pci_bridge_d3_possible(struct pci_dev *bridge)
2264 {
2265 unsigned int year;
2266
2267 if (!pci_is_pcie(bridge))
2268 return false;
2269
2270 switch (pci_pcie_type(bridge)) {
2271 case PCI_EXP_TYPE_ROOT_PORT:
2272 case PCI_EXP_TYPE_UPSTREAM:
2273 case PCI_EXP_TYPE_DOWNSTREAM:
2274 if (pci_bridge_d3_disable)
2275 return false;
2276
2277 /*
2278 * Hotplug interrupts cannot be delivered if the link is down,
2279 * so parents of a hotplug port must stay awake. In addition,
2280 * hotplug ports handled by firmware in System Management Mode
2281 * may not be put into D3 by the OS (Thunderbolt on non-Macs).
2282 * For simplicity, disallow in general for now.
2283 */
2284 if (bridge->is_hotplug_bridge)
2285 return false;
2286
2287 if (pci_bridge_d3_force)
2288 return true;
2289
2290 /*
2291 * It should be safe to put PCIe ports from 2015 or newer
2292 * to D3.
2293 */
2294 if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
2295 year >= 2015) {
2296 return true;
2297 }
2298 break;
2299 }
2300
2301 return false;
2302 }
2303
2304 static int pci_dev_check_d3cold(struct pci_dev *dev, void *data)
2305 {
2306 bool *d3cold_ok = data;
2307
2308 if (/* The device needs to be allowed to go D3cold ... */
2309 dev->no_d3cold || !dev->d3cold_allowed ||
2310
2311 /* ... and if it is wakeup capable to do so from D3cold. */
2312 (device_may_wakeup(&dev->dev) &&
2313 !pci_pme_capable(dev, PCI_D3cold)) ||
2314
2315 /* If it is a bridge it must be allowed to go to D3. */
2316 !pci_power_manageable(dev))
2317
2318 *d3cold_ok = false;
2319
2320 return !*d3cold_ok;
2321 }
2322
2323 /*
2324 * pci_bridge_d3_update - Update bridge D3 capabilities
2325 * @dev: PCI device which is changed
2326 *
2327 * Update upstream bridge PM capabilities accordingly depending on if the
2328 * device PM configuration was changed or the device is being removed. The
2329 * change is also propagated upstream.
2330 */
2331 void pci_bridge_d3_update(struct pci_dev *dev)
2332 {
2333 bool remove = !device_is_registered(&dev->dev);
2334 struct pci_dev *bridge;
2335 bool d3cold_ok = true;
2336
2337 bridge = pci_upstream_bridge(dev);
2338 if (!bridge || !pci_bridge_d3_possible(bridge))
2339 return;
2340
2341 /*
2342 * If D3 is currently allowed for the bridge, removing one of its
2343 * children won't change that.
2344 */
2345 if (remove && bridge->bridge_d3)
2346 return;
2347
2348 /*
2349 * If D3 is currently allowed for the bridge and a child is added or
2350 * changed, disallowance of D3 can only be caused by that child, so
2351 * we only need to check that single device, not any of its siblings.
2352 *
2353 * If D3 is currently not allowed for the bridge, checking the device
2354 * first may allow us to skip checking its siblings.
2355 */
2356 if (!remove)
2357 pci_dev_check_d3cold(dev, &d3cold_ok);
2358
2359 /*
2360 * If D3 is currently not allowed for the bridge, this may be caused
2361 * either by the device being changed/removed or any of its siblings,
2362 * so we need to go through all children to find out if one of them
2363 * continues to block D3.
2364 */
2365 if (d3cold_ok && !bridge->bridge_d3)
2366 pci_walk_bus(bridge->subordinate, pci_dev_check_d3cold,
2367 &d3cold_ok);
2368
2369 if (bridge->bridge_d3 != d3cold_ok) {
2370 bridge->bridge_d3 = d3cold_ok;
2371 /* Propagate change to upstream bridges */
2372 pci_bridge_d3_update(bridge);
2373 }
2374 }
2375
2376 /**
2377 * pci_d3cold_enable - Enable D3cold for device
2378 * @dev: PCI device to handle
2379 *
2380 * This function can be used in drivers to enable D3cold from the device
2381 * they handle. It also updates upstream PCI bridge PM capabilities
2382 * accordingly.
2383 */
2384 void pci_d3cold_enable(struct pci_dev *dev)
2385 {
2386 if (dev->no_d3cold) {
2387 dev->no_d3cold = false;
2388 pci_bridge_d3_update(dev);
2389 }
2390 }
2391 EXPORT_SYMBOL_GPL(pci_d3cold_enable);
2392
2393 /**
2394 * pci_d3cold_disable - Disable D3cold for device
2395 * @dev: PCI device to handle
2396 *
2397 * This function can be used in drivers to disable D3cold from the device
2398 * they handle. It also updates upstream PCI bridge PM capabilities
2399 * accordingly.
2400 */
2401 void pci_d3cold_disable(struct pci_dev *dev)
2402 {
2403 if (!dev->no_d3cold) {
2404 dev->no_d3cold = true;
2405 pci_bridge_d3_update(dev);
2406 }
2407 }
2408 EXPORT_SYMBOL_GPL(pci_d3cold_disable);
2409
2410 /**
2411 * pci_pm_init - Initialize PM functions of given PCI device
2412 * @dev: PCI device to handle.
2413 */
2414 void pci_pm_init(struct pci_dev *dev)
2415 {
2416 int pm;
2417 u16 pmc;
2418
2419 pm_runtime_forbid(&dev->dev);
2420 pm_runtime_set_active(&dev->dev);
2421 pm_runtime_enable(&dev->dev);
2422 device_enable_async_suspend(&dev->dev);
2423 dev->wakeup_prepared = false;
2424
2425 dev->pm_cap = 0;
2426 dev->pme_support = 0;
2427
2428 /* find PCI PM capability in list */
2429 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2430 if (!pm)
2431 return;
2432 /* Check device's ability to generate PME# */
2433 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2434
2435 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2436 dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
2437 pmc & PCI_PM_CAP_VER_MASK);
2438 return;
2439 }
2440
2441 dev->pm_cap = pm;
2442 dev->d3_delay = PCI_PM_D3_WAIT;
2443 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2444 dev->bridge_d3 = pci_bridge_d3_possible(dev);
2445 dev->d3cold_allowed = true;
2446
2447 dev->d1_support = false;
2448 dev->d2_support = false;
2449 if (!pci_no_d1d2(dev)) {
2450 if (pmc & PCI_PM_CAP_D1)
2451 dev->d1_support = true;
2452 if (pmc & PCI_PM_CAP_D2)
2453 dev->d2_support = true;
2454
2455 if (dev->d1_support || dev->d2_support)
2456 dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
2457 dev->d1_support ? " D1" : "",
2458 dev->d2_support ? " D2" : "");
2459 }
2460
2461 pmc &= PCI_PM_CAP_PME_MASK;
2462 if (pmc) {
2463 dev_printk(KERN_DEBUG, &dev->dev,
2464 "PME# supported from%s%s%s%s%s\n",
2465 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2466 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2467 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2468 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2469 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2470 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2471 dev->pme_poll = true;
2472 /*
2473 * Make device's PM flags reflect the wake-up capability, but
2474 * let the user space enable it to wake up the system as needed.
2475 */
2476 device_set_wakeup_capable(&dev->dev, true);
2477 /* Disable the PME# generation functionality */
2478 pci_pme_active(dev, false);
2479 }
2480 }
2481
2482 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2483 {
2484 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2485
2486 switch (prop) {
2487 case PCI_EA_P_MEM:
2488 case PCI_EA_P_VF_MEM:
2489 flags |= IORESOURCE_MEM;
2490 break;
2491 case PCI_EA_P_MEM_PREFETCH:
2492 case PCI_EA_P_VF_MEM_PREFETCH:
2493 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2494 break;
2495 case PCI_EA_P_IO:
2496 flags |= IORESOURCE_IO;
2497 break;
2498 default:
2499 return 0;
2500 }
2501
2502 return flags;
2503 }
2504
2505 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2506 u8 prop)
2507 {
2508 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2509 return &dev->resource[bei];
2510 #ifdef CONFIG_PCI_IOV
2511 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2512 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2513 return &dev->resource[PCI_IOV_RESOURCES +
2514 bei - PCI_EA_BEI_VF_BAR0];
2515 #endif
2516 else if (bei == PCI_EA_BEI_ROM)
2517 return &dev->resource[PCI_ROM_RESOURCE];
2518 else
2519 return NULL;
2520 }
2521
2522 /* Read an Enhanced Allocation (EA) entry */
2523 static int pci_ea_read(struct pci_dev *dev, int offset)
2524 {
2525 struct resource *res;
2526 int ent_size, ent_offset = offset;
2527 resource_size_t start, end;
2528 unsigned long flags;
2529 u32 dw0, bei, base, max_offset;
2530 u8 prop;
2531 bool support_64 = (sizeof(resource_size_t) >= 8);
2532
2533 pci_read_config_dword(dev, ent_offset, &dw0);
2534 ent_offset += 4;
2535
2536 /* Entry size field indicates DWORDs after 1st */
2537 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2538
2539 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2540 goto out;
2541
2542 bei = (dw0 & PCI_EA_BEI) >> 4;
2543 prop = (dw0 & PCI_EA_PP) >> 8;
2544
2545 /*
2546 * If the Property is in the reserved range, try the Secondary
2547 * Property instead.
2548 */
2549 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2550 prop = (dw0 & PCI_EA_SP) >> 16;
2551 if (prop > PCI_EA_P_BRIDGE_IO)
2552 goto out;
2553
2554 res = pci_ea_get_resource(dev, bei, prop);
2555 if (!res) {
2556 dev_err(&dev->dev, "Unsupported EA entry BEI: %u\n", bei);
2557 goto out;
2558 }
2559
2560 flags = pci_ea_flags(dev, prop);
2561 if (!flags) {
2562 dev_err(&dev->dev, "Unsupported EA properties: %#x\n", prop);
2563 goto out;
2564 }
2565
2566 /* Read Base */
2567 pci_read_config_dword(dev, ent_offset, &base);
2568 start = (base & PCI_EA_FIELD_MASK);
2569 ent_offset += 4;
2570
2571 /* Read MaxOffset */
2572 pci_read_config_dword(dev, ent_offset, &max_offset);
2573 ent_offset += 4;
2574
2575 /* Read Base MSBs (if 64-bit entry) */
2576 if (base & PCI_EA_IS_64) {
2577 u32 base_upper;
2578
2579 pci_read_config_dword(dev, ent_offset, &base_upper);
2580 ent_offset += 4;
2581
2582 flags |= IORESOURCE_MEM_64;
2583
2584 /* entry starts above 32-bit boundary, can't use */
2585 if (!support_64 && base_upper)
2586 goto out;
2587
2588 if (support_64)
2589 start |= ((u64)base_upper << 32);
2590 }
2591
2592 end = start + (max_offset | 0x03);
2593
2594 /* Read MaxOffset MSBs (if 64-bit entry) */
2595 if (max_offset & PCI_EA_IS_64) {
2596 u32 max_offset_upper;
2597
2598 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2599 ent_offset += 4;
2600
2601 flags |= IORESOURCE_MEM_64;
2602
2603 /* entry too big, can't use */
2604 if (!support_64 && max_offset_upper)
2605 goto out;
2606
2607 if (support_64)
2608 end += ((u64)max_offset_upper << 32);
2609 }
2610
2611 if (end < start) {
2612 dev_err(&dev->dev, "EA Entry crosses address boundary\n");
2613 goto out;
2614 }
2615
2616 if (ent_size != ent_offset - offset) {
2617 dev_err(&dev->dev,
2618 "EA Entry Size (%d) does not match length read (%d)\n",
2619 ent_size, ent_offset - offset);
2620 goto out;
2621 }
2622
2623 res->name = pci_name(dev);
2624 res->start = start;
2625 res->end = end;
2626 res->flags = flags;
2627
2628 if (bei <= PCI_EA_BEI_BAR5)
2629 dev_printk(KERN_DEBUG, &dev->dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2630 bei, res, prop);
2631 else if (bei == PCI_EA_BEI_ROM)
2632 dev_printk(KERN_DEBUG, &dev->dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2633 res, prop);
2634 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2635 dev_printk(KERN_DEBUG, &dev->dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2636 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2637 else
2638 dev_printk(KERN_DEBUG, &dev->dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2639 bei, res, prop);
2640
2641 out:
2642 return offset + ent_size;
2643 }
2644
2645 /* Enhanced Allocation Initialization */
2646 void pci_ea_init(struct pci_dev *dev)
2647 {
2648 int ea;
2649 u8 num_ent;
2650 int offset;
2651 int i;
2652
2653 /* find PCI EA capability in list */
2654 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2655 if (!ea)
2656 return;
2657
2658 /* determine the number of entries */
2659 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2660 &num_ent);
2661 num_ent &= PCI_EA_NUM_ENT_MASK;
2662
2663 offset = ea + PCI_EA_FIRST_ENT;
2664
2665 /* Skip DWORD 2 for type 1 functions */
2666 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2667 offset += 4;
2668
2669 /* parse each EA entry */
2670 for (i = 0; i < num_ent; ++i)
2671 offset = pci_ea_read(dev, offset);
2672 }
2673
2674 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2675 struct pci_cap_saved_state *new_cap)
2676 {
2677 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2678 }
2679
2680 /**
2681 * _pci_add_cap_save_buffer - allocate buffer for saving given
2682 * capability registers
2683 * @dev: the PCI device
2684 * @cap: the capability to allocate the buffer for
2685 * @extended: Standard or Extended capability ID
2686 * @size: requested size of the buffer
2687 */
2688 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2689 bool extended, unsigned int size)
2690 {
2691 int pos;
2692 struct pci_cap_saved_state *save_state;
2693
2694 if (extended)
2695 pos = pci_find_ext_capability(dev, cap);
2696 else
2697 pos = pci_find_capability(dev, cap);
2698
2699 if (!pos)
2700 return 0;
2701
2702 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2703 if (!save_state)
2704 return -ENOMEM;
2705
2706 save_state->cap.cap_nr = cap;
2707 save_state->cap.cap_extended = extended;
2708 save_state->cap.size = size;
2709 pci_add_saved_cap(dev, save_state);
2710
2711 return 0;
2712 }
2713
2714 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2715 {
2716 return _pci_add_cap_save_buffer(dev, cap, false, size);
2717 }
2718
2719 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2720 {
2721 return _pci_add_cap_save_buffer(dev, cap, true, size);
2722 }
2723
2724 /**
2725 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2726 * @dev: the PCI device
2727 */
2728 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2729 {
2730 int error;
2731
2732 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2733 PCI_EXP_SAVE_REGS * sizeof(u16));
2734 if (error)
2735 dev_err(&dev->dev,
2736 "unable to preallocate PCI Express save buffer\n");
2737
2738 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2739 if (error)
2740 dev_err(&dev->dev,
2741 "unable to preallocate PCI-X save buffer\n");
2742
2743 pci_allocate_vc_save_buffers(dev);
2744 }
2745
2746 void pci_free_cap_save_buffers(struct pci_dev *dev)
2747 {
2748 struct pci_cap_saved_state *tmp;
2749 struct hlist_node *n;
2750
2751 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2752 kfree(tmp);
2753 }
2754
2755 /**
2756 * pci_configure_ari - enable or disable ARI forwarding
2757 * @dev: the PCI device
2758 *
2759 * If @dev and its upstream bridge both support ARI, enable ARI in the
2760 * bridge. Otherwise, disable ARI in the bridge.
2761 */
2762 void pci_configure_ari(struct pci_dev *dev)
2763 {
2764 u32 cap;
2765 struct pci_dev *bridge;
2766
2767 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2768 return;
2769
2770 bridge = dev->bus->self;
2771 if (!bridge)
2772 return;
2773
2774 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2775 if (!(cap & PCI_EXP_DEVCAP2_ARI))
2776 return;
2777
2778 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2779 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2780 PCI_EXP_DEVCTL2_ARI);
2781 bridge->ari_enabled = 1;
2782 } else {
2783 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2784 PCI_EXP_DEVCTL2_ARI);
2785 bridge->ari_enabled = 0;
2786 }
2787 }
2788
2789 static int pci_acs_enable;
2790
2791 /**
2792 * pci_request_acs - ask for ACS to be enabled if supported
2793 */
2794 void pci_request_acs(void)
2795 {
2796 pci_acs_enable = 1;
2797 }
2798
2799 /**
2800 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2801 * @dev: the PCI device
2802 */
2803 static void pci_std_enable_acs(struct pci_dev *dev)
2804 {
2805 int pos;
2806 u16 cap;
2807 u16 ctrl;
2808
2809 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2810 if (!pos)
2811 return;
2812
2813 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2814 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2815
2816 /* Source Validation */
2817 ctrl |= (cap & PCI_ACS_SV);
2818
2819 /* P2P Request Redirect */
2820 ctrl |= (cap & PCI_ACS_RR);
2821
2822 /* P2P Completion Redirect */
2823 ctrl |= (cap & PCI_ACS_CR);
2824
2825 /* Upstream Forwarding */
2826 ctrl |= (cap & PCI_ACS_UF);
2827
2828 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2829 }
2830
2831 /**
2832 * pci_enable_acs - enable ACS if hardware support it
2833 * @dev: the PCI device
2834 */
2835 void pci_enable_acs(struct pci_dev *dev)
2836 {
2837 if (!pci_acs_enable)
2838 return;
2839
2840 if (!pci_dev_specific_enable_acs(dev))
2841 return;
2842
2843 pci_std_enable_acs(dev);
2844 }
2845
2846 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2847 {
2848 int pos;
2849 u16 cap, ctrl;
2850
2851 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2852 if (!pos)
2853 return false;
2854
2855 /*
2856 * Except for egress control, capabilities are either required
2857 * or only required if controllable. Features missing from the
2858 * capability field can therefore be assumed as hard-wired enabled.
2859 */
2860 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2861 acs_flags &= (cap | PCI_ACS_EC);
2862
2863 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2864 return (ctrl & acs_flags) == acs_flags;
2865 }
2866
2867 /**
2868 * pci_acs_enabled - test ACS against required flags for a given device
2869 * @pdev: device to test
2870 * @acs_flags: required PCI ACS flags
2871 *
2872 * Return true if the device supports the provided flags. Automatically
2873 * filters out flags that are not implemented on multifunction devices.
2874 *
2875 * Note that this interface checks the effective ACS capabilities of the
2876 * device rather than the actual capabilities. For instance, most single
2877 * function endpoints are not required to support ACS because they have no
2878 * opportunity for peer-to-peer access. We therefore return 'true'
2879 * regardless of whether the device exposes an ACS capability. This makes
2880 * it much easier for callers of this function to ignore the actual type
2881 * or topology of the device when testing ACS support.
2882 */
2883 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2884 {
2885 int ret;
2886
2887 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2888 if (ret >= 0)
2889 return ret > 0;
2890
2891 /*
2892 * Conventional PCI and PCI-X devices never support ACS, either
2893 * effectively or actually. The shared bus topology implies that
2894 * any device on the bus can receive or snoop DMA.
2895 */
2896 if (!pci_is_pcie(pdev))
2897 return false;
2898
2899 switch (pci_pcie_type(pdev)) {
2900 /*
2901 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2902 * but since their primary interface is PCI/X, we conservatively
2903 * handle them as we would a non-PCIe device.
2904 */
2905 case PCI_EXP_TYPE_PCIE_BRIDGE:
2906 /*
2907 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
2908 * applicable... must never implement an ACS Extended Capability...".
2909 * This seems arbitrary, but we take a conservative interpretation
2910 * of this statement.
2911 */
2912 case PCI_EXP_TYPE_PCI_BRIDGE:
2913 case PCI_EXP_TYPE_RC_EC:
2914 return false;
2915 /*
2916 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2917 * implement ACS in order to indicate their peer-to-peer capabilities,
2918 * regardless of whether they are single- or multi-function devices.
2919 */
2920 case PCI_EXP_TYPE_DOWNSTREAM:
2921 case PCI_EXP_TYPE_ROOT_PORT:
2922 return pci_acs_flags_enabled(pdev, acs_flags);
2923 /*
2924 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2925 * implemented by the remaining PCIe types to indicate peer-to-peer
2926 * capabilities, but only when they are part of a multifunction
2927 * device. The footnote for section 6.12 indicates the specific
2928 * PCIe types included here.
2929 */
2930 case PCI_EXP_TYPE_ENDPOINT:
2931 case PCI_EXP_TYPE_UPSTREAM:
2932 case PCI_EXP_TYPE_LEG_END:
2933 case PCI_EXP_TYPE_RC_END:
2934 if (!pdev->multifunction)
2935 break;
2936
2937 return pci_acs_flags_enabled(pdev, acs_flags);
2938 }
2939
2940 /*
2941 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2942 * to single function devices with the exception of downstream ports.
2943 */
2944 return true;
2945 }
2946
2947 /**
2948 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2949 * @start: starting downstream device
2950 * @end: ending upstream device or NULL to search to the root bus
2951 * @acs_flags: required flags
2952 *
2953 * Walk up a device tree from start to end testing PCI ACS support. If
2954 * any step along the way does not support the required flags, return false.
2955 */
2956 bool pci_acs_path_enabled(struct pci_dev *start,
2957 struct pci_dev *end, u16 acs_flags)
2958 {
2959 struct pci_dev *pdev, *parent = start;
2960
2961 do {
2962 pdev = parent;
2963
2964 if (!pci_acs_enabled(pdev, acs_flags))
2965 return false;
2966
2967 if (pci_is_root_bus(pdev->bus))
2968 return (end == NULL);
2969
2970 parent = pdev->bus->self;
2971 } while (pdev != end);
2972
2973 return true;
2974 }
2975
2976 /**
2977 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2978 * @dev: the PCI device
2979 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
2980 *
2981 * Perform INTx swizzling for a device behind one level of bridge. This is
2982 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2983 * behind bridges on add-in cards. For devices with ARI enabled, the slot
2984 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2985 * the PCI Express Base Specification, Revision 2.1)
2986 */
2987 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2988 {
2989 int slot;
2990
2991 if (pci_ari_enabled(dev->bus))
2992 slot = 0;
2993 else
2994 slot = PCI_SLOT(dev->devfn);
2995
2996 return (((pin - 1) + slot) % 4) + 1;
2997 }
2998
2999 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
3000 {
3001 u8 pin;
3002
3003 pin = dev->pin;
3004 if (!pin)
3005 return -1;
3006
3007 while (!pci_is_root_bus(dev->bus)) {
3008 pin = pci_swizzle_interrupt_pin(dev, pin);
3009 dev = dev->bus->self;
3010 }
3011 *bridge = dev;
3012 return pin;
3013 }
3014
3015 /**
3016 * pci_common_swizzle - swizzle INTx all the way to root bridge
3017 * @dev: the PCI device
3018 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
3019 *
3020 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
3021 * bridges all the way up to a PCI root bus.
3022 */
3023 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
3024 {
3025 u8 pin = *pinp;
3026
3027 while (!pci_is_root_bus(dev->bus)) {
3028 pin = pci_swizzle_interrupt_pin(dev, pin);
3029 dev = dev->bus->self;
3030 }
3031 *pinp = pin;
3032 return PCI_SLOT(dev->devfn);
3033 }
3034 EXPORT_SYMBOL_GPL(pci_common_swizzle);
3035
3036 /**
3037 * pci_release_region - Release a PCI bar
3038 * @pdev: PCI device whose resources were previously reserved by pci_request_region
3039 * @bar: BAR to release
3040 *
3041 * Releases the PCI I/O and memory resources previously reserved by a
3042 * successful call to pci_request_region. Call this function only
3043 * after all use of the PCI regions has ceased.
3044 */
3045 void pci_release_region(struct pci_dev *pdev, int bar)
3046 {
3047 struct pci_devres *dr;
3048
3049 if (pci_resource_len(pdev, bar) == 0)
3050 return;
3051 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
3052 release_region(pci_resource_start(pdev, bar),
3053 pci_resource_len(pdev, bar));
3054 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
3055 release_mem_region(pci_resource_start(pdev, bar),
3056 pci_resource_len(pdev, bar));
3057
3058 dr = find_pci_dr(pdev);
3059 if (dr)
3060 dr->region_mask &= ~(1 << bar);
3061 }
3062 EXPORT_SYMBOL(pci_release_region);
3063
3064 /**
3065 * __pci_request_region - Reserved PCI I/O and memory resource
3066 * @pdev: PCI device whose resources are to be reserved
3067 * @bar: BAR to be reserved
3068 * @res_name: Name to be associated with resource.
3069 * @exclusive: whether the region access is exclusive or not
3070 *
3071 * Mark the PCI region associated with PCI device @pdev BR @bar as
3072 * being reserved by owner @res_name. Do not access any
3073 * address inside the PCI regions unless this call returns
3074 * successfully.
3075 *
3076 * If @exclusive is set, then the region is marked so that userspace
3077 * is explicitly not allowed to map the resource via /dev/mem or
3078 * sysfs MMIO access.
3079 *
3080 * Returns 0 on success, or %EBUSY on error. A warning
3081 * message is also printed on failure.
3082 */
3083 static int __pci_request_region(struct pci_dev *pdev, int bar,
3084 const char *res_name, int exclusive)
3085 {
3086 struct pci_devres *dr;
3087
3088 if (pci_resource_len(pdev, bar) == 0)
3089 return 0;
3090
3091 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
3092 if (!request_region(pci_resource_start(pdev, bar),
3093 pci_resource_len(pdev, bar), res_name))
3094 goto err_out;
3095 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
3096 if (!__request_mem_region(pci_resource_start(pdev, bar),
3097 pci_resource_len(pdev, bar), res_name,
3098 exclusive))
3099 goto err_out;
3100 }
3101
3102 dr = find_pci_dr(pdev);
3103 if (dr)
3104 dr->region_mask |= 1 << bar;
3105
3106 return 0;
3107
3108 err_out:
3109 dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
3110 &pdev->resource[bar]);
3111 return -EBUSY;
3112 }
3113
3114 /**
3115 * pci_request_region - Reserve PCI I/O and memory resource
3116 * @pdev: PCI device whose resources are to be reserved
3117 * @bar: BAR to be reserved
3118 * @res_name: Name to be associated with resource
3119 *
3120 * Mark the PCI region associated with PCI device @pdev BAR @bar as
3121 * being reserved by owner @res_name. Do not access any
3122 * address inside the PCI regions unless this call returns
3123 * successfully.
3124 *
3125 * Returns 0 on success, or %EBUSY on error. A warning
3126 * message is also printed on failure.
3127 */
3128 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
3129 {
3130 return __pci_request_region(pdev, bar, res_name, 0);
3131 }
3132 EXPORT_SYMBOL(pci_request_region);
3133
3134 /**
3135 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
3136 * @pdev: PCI device whose resources are to be reserved
3137 * @bar: BAR to be reserved
3138 * @res_name: Name to be associated with resource.
3139 *
3140 * Mark the PCI region associated with PCI device @pdev BR @bar as
3141 * being reserved by owner @res_name. Do not access any
3142 * address inside the PCI regions unless this call returns
3143 * successfully.
3144 *
3145 * Returns 0 on success, or %EBUSY on error. A warning
3146 * message is also printed on failure.
3147 *
3148 * The key difference that _exclusive makes it that userspace is
3149 * explicitly not allowed to map the resource via /dev/mem or
3150 * sysfs.
3151 */
3152 int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
3153 const char *res_name)
3154 {
3155 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
3156 }
3157 EXPORT_SYMBOL(pci_request_region_exclusive);
3158
3159 /**
3160 * pci_release_selected_regions - Release selected PCI I/O and memory resources
3161 * @pdev: PCI device whose resources were previously reserved
3162 * @bars: Bitmask of BARs to be released
3163 *
3164 * Release selected PCI I/O and memory resources previously reserved.
3165 * Call this function only after all use of the PCI regions has ceased.
3166 */
3167 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
3168 {
3169 int i;
3170
3171 for (i = 0; i < 6; i++)
3172 if (bars & (1 << i))
3173 pci_release_region(pdev, i);
3174 }
3175 EXPORT_SYMBOL(pci_release_selected_regions);
3176
3177 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
3178 const char *res_name, int excl)
3179 {
3180 int i;
3181
3182 for (i = 0; i < 6; i++)
3183 if (bars & (1 << i))
3184 if (__pci_request_region(pdev, i, res_name, excl))
3185 goto err_out;
3186 return 0;
3187
3188 err_out:
3189 while (--i >= 0)
3190 if (bars & (1 << i))
3191 pci_release_region(pdev, i);
3192
3193 return -EBUSY;
3194 }
3195
3196
3197 /**
3198 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
3199 * @pdev: PCI device whose resources are to be reserved
3200 * @bars: Bitmask of BARs to be requested
3201 * @res_name: Name to be associated with resource
3202 */
3203 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
3204 const char *res_name)
3205 {
3206 return __pci_request_selected_regions(pdev, bars, res_name, 0);
3207 }
3208 EXPORT_SYMBOL(pci_request_selected_regions);
3209
3210 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
3211 const char *res_name)
3212 {
3213 return __pci_request_selected_regions(pdev, bars, res_name,
3214 IORESOURCE_EXCLUSIVE);
3215 }
3216 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
3217
3218 /**
3219 * pci_release_regions - Release reserved PCI I/O and memory resources
3220 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
3221 *
3222 * Releases all PCI I/O and memory resources previously reserved by a
3223 * successful call to pci_request_regions. Call this function only
3224 * after all use of the PCI regions has ceased.
3225 */
3226
3227 void pci_release_regions(struct pci_dev *pdev)
3228 {
3229 pci_release_selected_regions(pdev, (1 << 6) - 1);
3230 }
3231 EXPORT_SYMBOL(pci_release_regions);
3232
3233 /**
3234 * pci_request_regions - Reserved PCI I/O and memory resources
3235 * @pdev: PCI device whose resources are to be reserved
3236 * @res_name: Name to be associated with resource.
3237 *
3238 * Mark all PCI regions associated with PCI device @pdev as
3239 * being reserved by owner @res_name. Do not access any
3240 * address inside the PCI regions unless this call returns
3241 * successfully.
3242 *
3243 * Returns 0 on success, or %EBUSY on error. A warning
3244 * message is also printed on failure.
3245 */
3246 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
3247 {
3248 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
3249 }
3250 EXPORT_SYMBOL(pci_request_regions);
3251
3252 /**
3253 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3254 * @pdev: PCI device whose resources are to be reserved
3255 * @res_name: Name to be associated with resource.
3256 *
3257 * Mark all PCI regions associated with PCI device @pdev as
3258 * being reserved by owner @res_name. Do not access any
3259 * address inside the PCI regions unless this call returns
3260 * successfully.
3261 *
3262 * pci_request_regions_exclusive() will mark the region so that
3263 * /dev/mem and the sysfs MMIO access will not be allowed.
3264 *
3265 * Returns 0 on success, or %EBUSY on error. A warning
3266 * message is also printed on failure.
3267 */
3268 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3269 {
3270 return pci_request_selected_regions_exclusive(pdev,
3271 ((1 << 6) - 1), res_name);
3272 }
3273 EXPORT_SYMBOL(pci_request_regions_exclusive);
3274
3275 #if defined(PCI_IOBASE) && !defined(CONFIG_LIBIO)
3276 struct io_range {
3277 struct list_head list;
3278 phys_addr_t start;
3279 resource_size_t size;
3280 };
3281
3282 static LIST_HEAD(io_range_list);
3283 static DEFINE_SPINLOCK(io_range_lock);
3284 #endif
3285
3286 /*
3287 * Record the PCI IO range (expressed as CPU physical address + size).
3288 * Return a negative value if an error has occured, zero otherwise
3289 */
3290 int pci_register_io_range(struct fwnode_handle *fwnode, phys_addr_t addr,
3291 resource_size_t size)
3292 {
3293 int err = 0;
3294
3295 #ifdef PCI_IOBASE
3296 #ifdef CONFIG_LIBIO
3297 struct libio_range *range, *tmprange;
3298
3299 if (!size || addr + size < addr)
3300 return -EINVAL;
3301
3302 WARN_ON(!PAGE_ALIGNED(addr) || !PAGE_ALIGNED(size));
3303
3304 range = kzalloc(sizeof(*range), GFP_KERNEL);
3305 if (!range)
3306 return -ENOMEM;
3307 range->node = fwnode;
3308 range->flags = IO_CPU_MMIO;
3309 range->size = size;
3310 range->hw_start = addr;
3311
3312 tmprange = register_libio_range(range);
3313 if (tmprange != range) {
3314 kfree(range);
3315 if (IS_ERR(tmprange))
3316 return -EFAULT;
3317 }
3318 #else
3319 struct io_range *range;
3320 resource_size_t allocated_size = 0;
3321
3322 /* check if the range hasn't been previously recorded */
3323 spin_lock(&io_range_lock);
3324 list_for_each_entry(range, &io_range_list, list) {
3325 if (addr >= range->start && addr + size <= range->start + size) {
3326 /* range already registered, bail out */
3327 goto end_register;
3328 }
3329 allocated_size += range->size;
3330 }
3331
3332 /* range not registed yet, check for available space */
3333 if (allocated_size + size - 1 > IO_SPACE_LIMIT) {
3334 /* if it's too big check if 64K space can be reserved */
3335 if (allocated_size + SZ_64K - 1 > IO_SPACE_LIMIT) {
3336 err = -E2BIG;
3337 goto end_register;
3338 }
3339
3340 size = SZ_64K;
3341 pr_warn("Requested IO range too big, new size set to 64K\n");
3342 }
3343
3344 /* add the range to the list */
3345 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3346 if (!range) {
3347 err = -ENOMEM;
3348 goto end_register;
3349 }
3350
3351 range->start = addr;
3352 range->size = size;
3353
3354 list_add_tail(&range->list, &io_range_list);
3355
3356 end_register:
3357 spin_unlock(&io_range_lock);
3358 #endif /* CONFIG_LIBIO */
3359 #endif /* PCI_IOBASE */
3360
3361 return err;
3362 }
3363
3364 phys_addr_t pci_pio_to_address(unsigned long pio)
3365 {
3366 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3367
3368 #ifdef PCI_IOBASE
3369 #ifdef CONFIG_LIBIO
3370 if (pio > IO_SPACE_LIMIT)
3371 return address;
3372
3373 address = libio_to_hwaddr(pio);
3374 #else
3375 struct io_range *range;
3376 resource_size_t allocated_size = 0;
3377
3378 if (pio > IO_SPACE_LIMIT)
3379 return address;
3380
3381 spin_lock(&io_range_lock);
3382 list_for_each_entry(range, &io_range_list, list) {
3383 if (pio >= allocated_size && pio < allocated_size + range->size) {
3384 address = range->start + pio - allocated_size;
3385 break;
3386 }
3387 allocated_size += range->size;
3388 }
3389 spin_unlock(&io_range_lock);
3390 #endif /* CONFIG_LIBIO */
3391 #endif /* PCI_IOBASE */
3392
3393 return address;
3394 }
3395
3396 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3397 {
3398 #ifdef PCI_IOBASE
3399 #ifdef CONFIG_LIBIO
3400 return libio_translate_cpuaddr(address);
3401 #else
3402 struct io_range *res;
3403 resource_size_t offset = 0;
3404 unsigned long addr = -1;
3405
3406 spin_lock(&io_range_lock);
3407 list_for_each_entry(res, &io_range_list, list) {
3408 if (address >= res->start && address < res->start + res->size) {
3409 addr = address - res->start + offset;
3410 break;
3411 }
3412 offset += res->size;
3413 }
3414 spin_unlock(&io_range_lock);
3415
3416 return addr;
3417 #endif
3418 #else
3419 #ifndef CONFIG_LIBIO
3420 if (address > IO_SPACE_LIMIT)
3421 return (unsigned long)-1;
3422 #endif
3423 return (unsigned long) address;
3424 #endif
3425 }
3426
3427 /**
3428 * pci_remap_iospace - Remap the memory mapped I/O space
3429 * @res: Resource describing the I/O space
3430 * @phys_addr: physical address of range to be mapped
3431 *
3432 * Remap the memory mapped I/O space described by the @res
3433 * and the CPU physical address @phys_addr into virtual address space.
3434 * Only architectures that have memory mapped IO functions defined
3435 * (and the PCI_IOBASE value defined) should call this function.
3436 */
3437 int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3438 {
3439 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3440 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3441
3442 if (!(res->flags & IORESOURCE_IO))
3443 return -EINVAL;
3444
3445 if (res->end > IO_SPACE_LIMIT)
3446 return -EINVAL;
3447
3448 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3449 pgprot_device(PAGE_KERNEL));
3450 #else
3451 /* this architecture does not have memory mapped I/O space,
3452 so this function should never be called */
3453 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3454 return -ENODEV;
3455 #endif
3456 }
3457 EXPORT_SYMBOL(pci_remap_iospace);
3458
3459 /**
3460 * pci_unmap_iospace - Unmap the memory mapped I/O space
3461 * @res: resource to be unmapped
3462 *
3463 * Unmap the CPU virtual address @res from virtual address space.
3464 * Only architectures that have memory mapped IO functions defined
3465 * (and the PCI_IOBASE value defined) should call this function.
3466 */
3467 void pci_unmap_iospace(struct resource *res)
3468 {
3469 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3470 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3471
3472 unmap_kernel_range(vaddr, resource_size(res));
3473 #endif
3474 }
3475 EXPORT_SYMBOL(pci_unmap_iospace);
3476
3477 /**
3478 * devm_pci_remap_cfgspace - Managed pci_remap_cfgspace()
3479 * @dev: Generic device to remap IO address for
3480 * @offset: Resource address to map
3481 * @size: Size of map
3482 *
3483 * Managed pci_remap_cfgspace(). Map is automatically unmapped on driver
3484 * detach.
3485 */
3486 void __iomem *devm_pci_remap_cfgspace(struct device *dev,
3487 resource_size_t offset,
3488 resource_size_t size)
3489 {
3490 void __iomem **ptr, *addr;
3491
3492 ptr = devres_alloc(devm_ioremap_release, sizeof(*ptr), GFP_KERNEL);
3493 if (!ptr)
3494 return NULL;
3495
3496 addr = pci_remap_cfgspace(offset, size);
3497 if (addr) {
3498 *ptr = addr;
3499 devres_add(dev, ptr);
3500 } else
3501 devres_free(ptr);
3502
3503 return addr;
3504 }
3505 EXPORT_SYMBOL(devm_pci_remap_cfgspace);
3506
3507 /**
3508 * devm_pci_remap_cfg_resource - check, request region and ioremap cfg resource
3509 * @dev: generic device to handle the resource for
3510 * @res: configuration space resource to be handled
3511 *
3512 * Checks that a resource is a valid memory region, requests the memory
3513 * region and ioremaps with pci_remap_cfgspace() API that ensures the
3514 * proper PCI configuration space memory attributes are guaranteed.
3515 *
3516 * All operations are managed and will be undone on driver detach.
3517 *
3518 * Returns a pointer to the remapped memory or an ERR_PTR() encoded error code
3519 * on failure. Usage example:
3520 *
3521 * res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
3522 * base = devm_pci_remap_cfg_resource(&pdev->dev, res);
3523 * if (IS_ERR(base))
3524 * return PTR_ERR(base);
3525 */
3526 void __iomem *devm_pci_remap_cfg_resource(struct device *dev,
3527 struct resource *res)
3528 {
3529 resource_size_t size;
3530 const char *name;
3531 void __iomem *dest_ptr;
3532
3533 BUG_ON(!dev);
3534
3535 if (!res || resource_type(res) != IORESOURCE_MEM) {
3536 dev_err(dev, "invalid resource\n");
3537 return IOMEM_ERR_PTR(-EINVAL);
3538 }
3539
3540 size = resource_size(res);
3541 name = res->name ?: dev_name(dev);
3542
3543 if (!devm_request_mem_region(dev, res->start, size, name)) {
3544 dev_err(dev, "can't request region for resource %pR\n", res);
3545 return IOMEM_ERR_PTR(-EBUSY);
3546 }
3547
3548 dest_ptr = devm_pci_remap_cfgspace(dev, res->start, size);
3549 if (!dest_ptr) {
3550 dev_err(dev, "ioremap failed for resource %pR\n", res);
3551 devm_release_mem_region(dev, res->start, size);
3552 dest_ptr = IOMEM_ERR_PTR(-ENOMEM);
3553 }
3554
3555 return dest_ptr;
3556 }
3557 EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
3558
3559 static void __pci_set_master(struct pci_dev *dev, bool enable)
3560 {
3561 u16 old_cmd, cmd;
3562
3563 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3564 if (enable)
3565 cmd = old_cmd | PCI_COMMAND_MASTER;
3566 else
3567 cmd = old_cmd & ~PCI_COMMAND_MASTER;
3568 if (cmd != old_cmd) {
3569 dev_dbg(&dev->dev, "%s bus mastering\n",
3570 enable ? "enabling" : "disabling");
3571 pci_write_config_word(dev, PCI_COMMAND, cmd);
3572 }
3573 dev->is_busmaster = enable;
3574 }
3575
3576 /**
3577 * pcibios_setup - process "pci=" kernel boot arguments
3578 * @str: string used to pass in "pci=" kernel boot arguments
3579 *
3580 * Process kernel boot arguments. This is the default implementation.
3581 * Architecture specific implementations can override this as necessary.
3582 */
3583 char * __weak __init pcibios_setup(char *str)
3584 {
3585 return str;
3586 }
3587
3588 /**
3589 * pcibios_set_master - enable PCI bus-mastering for device dev
3590 * @dev: the PCI device to enable
3591 *
3592 * Enables PCI bus-mastering for the device. This is the default
3593 * implementation. Architecture specific implementations can override
3594 * this if necessary.
3595 */
3596 void __weak pcibios_set_master(struct pci_dev *dev)
3597 {
3598 u8 lat;
3599
3600 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3601 if (pci_is_pcie(dev))
3602 return;
3603
3604 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3605 if (lat < 16)
3606 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3607 else if (lat > pcibios_max_latency)
3608 lat = pcibios_max_latency;
3609 else
3610 return;
3611
3612 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3613 }
3614
3615 /**
3616 * pci_set_master - enables bus-mastering for device dev
3617 * @dev: the PCI device to enable
3618 *
3619 * Enables bus-mastering on the device and calls pcibios_set_master()
3620 * to do the needed arch specific settings.
3621 */
3622 void pci_set_master(struct pci_dev *dev)
3623 {
3624 __pci_set_master(dev, true);
3625 pcibios_set_master(dev);
3626 }
3627 EXPORT_SYMBOL(pci_set_master);
3628
3629 /**
3630 * pci_clear_master - disables bus-mastering for device dev
3631 * @dev: the PCI device to disable
3632 */
3633 void pci_clear_master(struct pci_dev *dev)
3634 {
3635 __pci_set_master(dev, false);
3636 }
3637 EXPORT_SYMBOL(pci_clear_master);
3638
3639 /**
3640 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3641 * @dev: the PCI device for which MWI is to be enabled
3642 *
3643 * Helper function for pci_set_mwi.
3644 * Originally copied from drivers/net/acenic.c.
3645 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3646 *
3647 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3648 */
3649 int pci_set_cacheline_size(struct pci_dev *dev)
3650 {
3651 u8 cacheline_size;
3652
3653 if (!pci_cache_line_size)
3654 return -EINVAL;
3655
3656 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3657 equal to or multiple of the right value. */
3658 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3659 if (cacheline_size >= pci_cache_line_size &&
3660 (cacheline_size % pci_cache_line_size) == 0)
3661 return 0;
3662
3663 /* Write the correct value. */
3664 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3665 /* Read it back. */
3666 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3667 if (cacheline_size == pci_cache_line_size)
3668 return 0;
3669
3670 dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not supported\n",
3671 pci_cache_line_size << 2);
3672
3673 return -EINVAL;
3674 }
3675 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3676
3677 /**
3678 * pci_set_mwi - enables memory-write-invalidate PCI transaction
3679 * @dev: the PCI device for which MWI is enabled
3680 *
3681 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3682 *
3683 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3684 */
3685 int pci_set_mwi(struct pci_dev *dev)
3686 {
3687 #ifdef PCI_DISABLE_MWI
3688 return 0;
3689 #else
3690 int rc;
3691 u16 cmd;
3692
3693 rc = pci_set_cacheline_size(dev);
3694 if (rc)
3695 return rc;
3696
3697 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3698 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3699 dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
3700 cmd |= PCI_COMMAND_INVALIDATE;
3701 pci_write_config_word(dev, PCI_COMMAND, cmd);
3702 }
3703 return 0;
3704 #endif
3705 }
3706 EXPORT_SYMBOL(pci_set_mwi);
3707
3708 /**
3709 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3710 * @dev: the PCI device for which MWI is enabled
3711 *
3712 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3713 * Callers are not required to check the return value.
3714 *
3715 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3716 */
3717 int pci_try_set_mwi(struct pci_dev *dev)
3718 {
3719 #ifdef PCI_DISABLE_MWI
3720 return 0;
3721 #else
3722 return pci_set_mwi(dev);
3723 #endif
3724 }
3725 EXPORT_SYMBOL(pci_try_set_mwi);
3726
3727 /**
3728 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3729 * @dev: the PCI device to disable
3730 *
3731 * Disables PCI Memory-Write-Invalidate transaction on the device
3732 */
3733 void pci_clear_mwi(struct pci_dev *dev)
3734 {
3735 #ifndef PCI_DISABLE_MWI
3736 u16 cmd;
3737
3738 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3739 if (cmd & PCI_COMMAND_INVALIDATE) {
3740 cmd &= ~PCI_COMMAND_INVALIDATE;
3741 pci_write_config_word(dev, PCI_COMMAND, cmd);
3742 }
3743 #endif
3744 }
3745 EXPORT_SYMBOL(pci_clear_mwi);
3746
3747 /**
3748 * pci_intx - enables/disables PCI INTx for device dev
3749 * @pdev: the PCI device to operate on
3750 * @enable: boolean: whether to enable or disable PCI INTx
3751 *
3752 * Enables/disables PCI INTx for device dev
3753 */
3754 void pci_intx(struct pci_dev *pdev, int enable)
3755 {
3756 u16 pci_command, new;
3757
3758 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3759
3760 if (enable)
3761 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3762 else
3763 new = pci_command | PCI_COMMAND_INTX_DISABLE;
3764
3765 if (new != pci_command) {
3766 struct pci_devres *dr;
3767
3768 pci_write_config_word(pdev, PCI_COMMAND, new);
3769
3770 dr = find_pci_dr(pdev);
3771 if (dr && !dr->restore_intx) {
3772 dr->restore_intx = 1;
3773 dr->orig_intx = !enable;
3774 }
3775 }
3776 }
3777 EXPORT_SYMBOL_GPL(pci_intx);
3778
3779 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3780 {
3781 struct pci_bus *bus = dev->bus;
3782 bool mask_updated = true;
3783 u32 cmd_status_dword;
3784 u16 origcmd, newcmd;
3785 unsigned long flags;
3786 bool irq_pending;
3787
3788 /*
3789 * We do a single dword read to retrieve both command and status.
3790 * Document assumptions that make this possible.
3791 */
3792 BUILD_BUG_ON(PCI_COMMAND % 4);
3793 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3794
3795 raw_spin_lock_irqsave(&pci_lock, flags);
3796
3797 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3798
3799 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3800
3801 /*
3802 * Check interrupt status register to see whether our device
3803 * triggered the interrupt (when masking) or the next IRQ is
3804 * already pending (when unmasking).
3805 */
3806 if (mask != irq_pending) {
3807 mask_updated = false;
3808 goto done;
3809 }
3810
3811 origcmd = cmd_status_dword;
3812 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3813 if (mask)
3814 newcmd |= PCI_COMMAND_INTX_DISABLE;
3815 if (newcmd != origcmd)
3816 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3817
3818 done:
3819 raw_spin_unlock_irqrestore(&pci_lock, flags);
3820
3821 return mask_updated;
3822 }
3823
3824 /**
3825 * pci_check_and_mask_intx - mask INTx on pending interrupt
3826 * @dev: the PCI device to operate on
3827 *
3828 * Check if the device dev has its INTx line asserted, mask it and
3829 * return true in that case. False is returned if no interrupt was
3830 * pending.
3831 */
3832 bool pci_check_and_mask_intx(struct pci_dev *dev)
3833 {
3834 return pci_check_and_set_intx_mask(dev, true);
3835 }
3836 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3837
3838 /**
3839 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3840 * @dev: the PCI device to operate on
3841 *
3842 * Check if the device dev has its INTx line asserted, unmask it if not
3843 * and return true. False is returned and the mask remains active if
3844 * there was still an interrupt pending.
3845 */
3846 bool pci_check_and_unmask_intx(struct pci_dev *dev)
3847 {
3848 return pci_check_and_set_intx_mask(dev, false);
3849 }
3850 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3851
3852 /**
3853 * pci_wait_for_pending_transaction - waits for pending transaction
3854 * @dev: the PCI device to operate on
3855 *
3856 * Return 0 if transaction is pending 1 otherwise.
3857 */
3858 int pci_wait_for_pending_transaction(struct pci_dev *dev)
3859 {
3860 if (!pci_is_pcie(dev))
3861 return 1;
3862
3863 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3864 PCI_EXP_DEVSTA_TRPND);
3865 }
3866 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3867
3868 /*
3869 * We should only need to wait 100ms after FLR, but some devices take longer.
3870 * Wait for up to 1000ms for config space to return something other than -1.
3871 * Intel IGD requires this when an LCD panel is attached. We read the 2nd
3872 * dword because VFs don't implement the 1st dword.
3873 */
3874 static void pci_flr_wait(struct pci_dev *dev)
3875 {
3876 int i = 0;
3877 u32 id;
3878
3879 do {
3880 msleep(100);
3881 pci_read_config_dword(dev, PCI_COMMAND, &id);
3882 } while (i++ < 10 && id == ~0);
3883
3884 if (id == ~0)
3885 dev_warn(&dev->dev, "Failed to return from FLR\n");
3886 else if (i > 1)
3887 dev_info(&dev->dev, "Required additional %dms to return from FLR\n",
3888 (i - 1) * 100);
3889 }
3890
3891 /**
3892 * pcie_has_flr - check if a device supports function level resets
3893 * @dev: device to check
3894 *
3895 * Returns true if the device advertises support for PCIe function level
3896 * resets.
3897 */
3898 static bool pcie_has_flr(struct pci_dev *dev)
3899 {
3900 u32 cap;
3901
3902 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
3903 return false;
3904
3905 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
3906 return cap & PCI_EXP_DEVCAP_FLR;
3907 }
3908
3909 /**
3910 * pcie_flr - initiate a PCIe function level reset
3911 * @dev: device to reset
3912 *
3913 * Initiate a function level reset on @dev. The caller should ensure the
3914 * device supports FLR before calling this function, e.g. by using the
3915 * pcie_has_flr() helper.
3916 */
3917 void pcie_flr(struct pci_dev *dev)
3918 {
3919 if (!pci_wait_for_pending_transaction(dev))
3920 dev_err(&dev->dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
3921
3922 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
3923 pci_flr_wait(dev);
3924 }
3925 EXPORT_SYMBOL_GPL(pcie_flr);
3926
3927 static int pci_af_flr(struct pci_dev *dev, int probe)
3928 {
3929 int pos;
3930 u8 cap;
3931
3932 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3933 if (!pos)
3934 return -ENOTTY;
3935
3936 if (dev->dev_flags & PCI_DEV_FLAGS_NO_FLR_RESET)
3937 return -ENOTTY;
3938
3939 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3940 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3941 return -ENOTTY;
3942
3943 if (probe)
3944 return 0;
3945
3946 /*
3947 * Wait for Transaction Pending bit to clear. A word-aligned test
3948 * is used, so we use the conrol offset rather than status and shift
3949 * the test bit to match.
3950 */
3951 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
3952 PCI_AF_STATUS_TP << 8))
3953 dev_err(&dev->dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
3954
3955 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3956 pci_flr_wait(dev);
3957 return 0;
3958 }
3959
3960 /**
3961 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3962 * @dev: Device to reset.
3963 * @probe: If set, only check if the device can be reset this way.
3964 *
3965 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3966 * unset, it will be reinitialized internally when going from PCI_D3hot to
3967 * PCI_D0. If that's the case and the device is not in a low-power state
3968 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3969 *
3970 * NOTE: This causes the caller to sleep for twice the device power transition
3971 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3972 * by default (i.e. unless the @dev's d3_delay field has a different value).
3973 * Moreover, only devices in D0 can be reset by this function.
3974 */
3975 static int pci_pm_reset(struct pci_dev *dev, int probe)
3976 {
3977 u16 csr;
3978
3979 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
3980 return -ENOTTY;
3981
3982 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3983 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3984 return -ENOTTY;
3985
3986 if (probe)
3987 return 0;
3988
3989 if (dev->current_state != PCI_D0)
3990 return -EINVAL;
3991
3992 csr &= ~PCI_PM_CTRL_STATE_MASK;
3993 csr |= PCI_D3hot;
3994 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3995 pci_dev_d3_sleep(dev);
3996
3997 csr &= ~PCI_PM_CTRL_STATE_MASK;
3998 csr |= PCI_D0;
3999 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
4000 pci_dev_d3_sleep(dev);
4001
4002 return 0;
4003 }
4004
4005 void pci_reset_secondary_bus(struct pci_dev *dev)
4006 {
4007 u16 ctrl;
4008
4009 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
4010 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
4011 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4012 /*
4013 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
4014 * this to 2ms to ensure that we meet the minimum requirement.
4015 */
4016 msleep(2);
4017
4018 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
4019 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
4020
4021 /*
4022 * Trhfa for conventional PCI is 2^25 clock cycles.
4023 * Assuming a minimum 33MHz clock this results in a 1s
4024 * delay before we can consider subordinate devices to
4025 * be re-initialized. PCIe has some ways to shorten this,
4026 * but we don't make use of them yet.
4027 */
4028 ssleep(1);
4029 }
4030
4031 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
4032 {
4033 pci_reset_secondary_bus(dev);
4034 }
4035
4036 /**
4037 * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
4038 * @dev: Bridge device
4039 *
4040 * Use the bridge control register to assert reset on the secondary bus.
4041 * Devices on the secondary bus are left in power-on state.
4042 */
4043 void pci_reset_bridge_secondary_bus(struct pci_dev *dev)
4044 {
4045 pcibios_reset_secondary_bus(dev);
4046 }
4047 EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
4048
4049 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
4050 {
4051 struct pci_dev *pdev;
4052
4053 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
4054 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4055 return -ENOTTY;
4056
4057 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4058 if (pdev != dev)
4059 return -ENOTTY;
4060
4061 if (probe)
4062 return 0;
4063
4064 pci_reset_bridge_secondary_bus(dev->bus->self);
4065
4066 return 0;
4067 }
4068
4069 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
4070 {
4071 int rc = -ENOTTY;
4072
4073 if (!hotplug || !try_module_get(hotplug->ops->owner))
4074 return rc;
4075
4076 if (hotplug->ops->reset_slot)
4077 rc = hotplug->ops->reset_slot(hotplug, probe);
4078
4079 module_put(hotplug->ops->owner);
4080
4081 return rc;
4082 }
4083
4084 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
4085 {
4086 struct pci_dev *pdev;
4087
4088 if (dev->subordinate || !dev->slot ||
4089 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
4090 return -ENOTTY;
4091
4092 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
4093 if (pdev != dev && pdev->slot == dev->slot)
4094 return -ENOTTY;
4095
4096 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
4097 }
4098
4099 static void pci_dev_lock(struct pci_dev *dev)
4100 {
4101 pci_cfg_access_lock(dev);
4102 /* block PM suspend, driver probe, etc. */
4103 device_lock(&dev->dev);
4104 }
4105
4106 /* Return 1 on successful lock, 0 on contention */
4107 static int pci_dev_trylock(struct pci_dev *dev)
4108 {
4109 if (pci_cfg_access_trylock(dev)) {
4110 if (device_trylock(&dev->dev))
4111 return 1;
4112 pci_cfg_access_unlock(dev);
4113 }
4114
4115 return 0;
4116 }
4117
4118 static void pci_dev_unlock(struct pci_dev *dev)
4119 {
4120 device_unlock(&dev->dev);
4121 pci_cfg_access_unlock(dev);
4122 }
4123
4124 static void pci_dev_save_and_disable(struct pci_dev *dev)
4125 {
4126 const struct pci_error_handlers *err_handler =
4127 dev->driver ? dev->driver->err_handler : NULL;
4128
4129 /*
4130 * dev->driver->err_handler->reset_prepare() is protected against
4131 * races with ->remove() by the device lock, which must be held by
4132 * the caller.
4133 */
4134 if (err_handler && err_handler->reset_prepare)
4135 err_handler->reset_prepare(dev);
4136
4137 /*
4138 * Wake-up device prior to save. PM registers default to D0 after
4139 * reset and a simple register restore doesn't reliably return
4140 * to a non-D0 state anyway.
4141 */
4142 pci_set_power_state(dev, PCI_D0);
4143
4144 pci_save_state(dev);
4145 /*
4146 * Disable the device by clearing the Command register, except for
4147 * INTx-disable which is set. This not only disables MMIO and I/O port
4148 * BARs, but also prevents the device from being Bus Master, preventing
4149 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
4150 * compliant devices, INTx-disable prevents legacy interrupts.
4151 */
4152 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
4153 }
4154
4155 static void pci_dev_restore(struct pci_dev *dev)
4156 {
4157 const struct pci_error_handlers *err_handler =
4158 dev->driver ? dev->driver->err_handler : NULL;
4159
4160 pci_restore_state(dev);
4161
4162 /*
4163 * dev->driver->err_handler->reset_done() is protected against
4164 * races with ->remove() by the device lock, which must be held by
4165 * the caller.
4166 */
4167 if (err_handler && err_handler->reset_done)
4168 err_handler->reset_done(dev);
4169 }
4170
4171 /**
4172 * __pci_reset_function - reset a PCI device function
4173 * @dev: PCI device to reset
4174 *
4175 * Some devices allow an individual function to be reset without affecting
4176 * other functions in the same device. The PCI device must be responsive
4177 * to PCI config space in order to use this function.
4178 *
4179 * The device function is presumed to be unused when this function is called.
4180 * Resetting the device will make the contents of PCI configuration space
4181 * random, so any caller of this must be prepared to reinitialise the
4182 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4183 * etc.
4184 *
4185 * Returns 0 if the device function was successfully reset or negative if the
4186 * device doesn't support resetting a single function.
4187 */
4188 int __pci_reset_function(struct pci_dev *dev)
4189 {
4190 int ret;
4191
4192 pci_dev_lock(dev);
4193 ret = __pci_reset_function_locked(dev);
4194 pci_dev_unlock(dev);
4195
4196 return ret;
4197 }
4198 EXPORT_SYMBOL_GPL(__pci_reset_function);
4199
4200 /**
4201 * __pci_reset_function_locked - reset a PCI device function while holding
4202 * the @dev mutex lock.
4203 * @dev: PCI device to reset
4204 *
4205 * Some devices allow an individual function to be reset without affecting
4206 * other functions in the same device. The PCI device must be responsive
4207 * to PCI config space in order to use this function.
4208 *
4209 * The device function is presumed to be unused and the caller is holding
4210 * the device mutex lock when this function is called.
4211 * Resetting the device will make the contents of PCI configuration space
4212 * random, so any caller of this must be prepared to reinitialise the
4213 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
4214 * etc.
4215 *
4216 * Returns 0 if the device function was successfully reset or negative if the
4217 * device doesn't support resetting a single function.
4218 */
4219 int __pci_reset_function_locked(struct pci_dev *dev)
4220 {
4221 int rc;
4222
4223 might_sleep();
4224
4225 rc = pci_dev_specific_reset(dev, 0);
4226 if (rc != -ENOTTY)
4227 return rc;
4228 if (pcie_has_flr(dev)) {
4229 pcie_flr(dev);
4230 return 0;
4231 }
4232 rc = pci_af_flr(dev, 0);
4233 if (rc != -ENOTTY)
4234 return rc;
4235 rc = pci_pm_reset(dev, 0);
4236 if (rc != -ENOTTY)
4237 return rc;
4238 rc = pci_dev_reset_slot_function(dev, 0);
4239 if (rc != -ENOTTY)
4240 return rc;
4241 return pci_parent_bus_reset(dev, 0);
4242 }
4243 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
4244
4245 /**
4246 * pci_probe_reset_function - check whether the device can be safely reset
4247 * @dev: PCI device to reset
4248 *
4249 * Some devices allow an individual function to be reset without affecting
4250 * other functions in the same device. The PCI device must be responsive
4251 * to PCI config space in order to use this function.
4252 *
4253 * Returns 0 if the device function can be reset or negative if the
4254 * device doesn't support resetting a single function.
4255 */
4256 int pci_probe_reset_function(struct pci_dev *dev)
4257 {
4258 int rc;
4259
4260 might_sleep();
4261
4262 rc = pci_dev_specific_reset(dev, 1);
4263 if (rc != -ENOTTY)
4264 return rc;
4265 if (pcie_has_flr(dev))
4266 return 0;
4267 rc = pci_af_flr(dev, 1);
4268 if (rc != -ENOTTY)
4269 return rc;
4270 rc = pci_pm_reset(dev, 1);
4271 if (rc != -ENOTTY)
4272 return rc;
4273 rc = pci_dev_reset_slot_function(dev, 1);
4274 if (rc != -ENOTTY)
4275 return rc;
4276
4277 return pci_parent_bus_reset(dev, 1);
4278 }
4279
4280 /**
4281 * pci_reset_function - quiesce and reset a PCI device function
4282 * @dev: PCI device to reset
4283 *
4284 * Some devices allow an individual function to be reset without affecting
4285 * other functions in the same device. The PCI device must be responsive
4286 * to PCI config space in order to use this function.
4287 *
4288 * This function does not just reset the PCI portion of a device, but
4289 * clears all the state associated with the device. This function differs
4290 * from __pci_reset_function in that it saves and restores device state
4291 * over the reset.
4292 *
4293 * Returns 0 if the device function was successfully reset or negative if the
4294 * device doesn't support resetting a single function.
4295 */
4296 int pci_reset_function(struct pci_dev *dev)
4297 {
4298 int rc;
4299
4300 rc = pci_probe_reset_function(dev);
4301 if (rc)
4302 return rc;
4303
4304 pci_dev_lock(dev);
4305 pci_dev_save_and_disable(dev);
4306
4307 rc = __pci_reset_function_locked(dev);
4308
4309 pci_dev_restore(dev);
4310 pci_dev_unlock(dev);
4311
4312 return rc;
4313 }
4314 EXPORT_SYMBOL_GPL(pci_reset_function);
4315
4316 /**
4317 * pci_reset_function_locked - quiesce and reset a PCI device function
4318 * @dev: PCI device to reset
4319 *
4320 * Some devices allow an individual function to be reset without affecting
4321 * other functions in the same device. The PCI device must be responsive
4322 * to PCI config space in order to use this function.
4323 *
4324 * This function does not just reset the PCI portion of a device, but
4325 * clears all the state associated with the device. This function differs
4326 * from __pci_reset_function() in that it saves and restores device state
4327 * over the reset. It also differs from pci_reset_function() in that it
4328 * requires the PCI device lock to be held.
4329 *
4330 * Returns 0 if the device function was successfully reset or negative if the
4331 * device doesn't support resetting a single function.
4332 */
4333 int pci_reset_function_locked(struct pci_dev *dev)
4334 {
4335 int rc;
4336
4337 rc = pci_probe_reset_function(dev);
4338 if (rc)
4339 return rc;
4340
4341 pci_dev_save_and_disable(dev);
4342
4343 rc = __pci_reset_function_locked(dev);
4344
4345 pci_dev_restore(dev);
4346
4347 return rc;
4348 }
4349 EXPORT_SYMBOL_GPL(pci_reset_function_locked);
4350
4351 /**
4352 * pci_try_reset_function - quiesce and reset a PCI device function
4353 * @dev: PCI device to reset
4354 *
4355 * Same as above, except return -EAGAIN if unable to lock device.
4356 */
4357 int pci_try_reset_function(struct pci_dev *dev)
4358 {
4359 int rc;
4360
4361 rc = pci_probe_reset_function(dev);
4362 if (rc)
4363 return rc;
4364
4365 if (!pci_dev_trylock(dev))
4366 return -EAGAIN;
4367
4368 pci_dev_save_and_disable(dev);
4369 rc = __pci_reset_function_locked(dev);
4370 pci_dev_unlock(dev);
4371
4372 pci_dev_restore(dev);
4373 return rc;
4374 }
4375 EXPORT_SYMBOL_GPL(pci_try_reset_function);
4376
4377 /* Do any devices on or below this bus prevent a bus reset? */
4378 static bool pci_bus_resetable(struct pci_bus *bus)
4379 {
4380 struct pci_dev *dev;
4381
4382
4383 if (bus->self && (bus->self->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET))
4384 return false;
4385
4386 list_for_each_entry(dev, &bus->devices, bus_list) {
4387 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4388 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4389 return false;
4390 }
4391
4392 return true;
4393 }
4394
4395 /* Lock devices from the top of the tree down */
4396 static void pci_bus_lock(struct pci_bus *bus)
4397 {
4398 struct pci_dev *dev;
4399
4400 list_for_each_entry(dev, &bus->devices, bus_list) {
4401 pci_dev_lock(dev);
4402 if (dev->subordinate)
4403 pci_bus_lock(dev->subordinate);
4404 }
4405 }
4406
4407 /* Unlock devices from the bottom of the tree up */
4408 static void pci_bus_unlock(struct pci_bus *bus)
4409 {
4410 struct pci_dev *dev;
4411
4412 list_for_each_entry(dev, &bus->devices, bus_list) {
4413 if (dev->subordinate)
4414 pci_bus_unlock(dev->subordinate);
4415 pci_dev_unlock(dev);
4416 }
4417 }
4418
4419 /* Return 1 on successful lock, 0 on contention */
4420 static int pci_bus_trylock(struct pci_bus *bus)
4421 {
4422 struct pci_dev *dev;
4423
4424 list_for_each_entry(dev, &bus->devices, bus_list) {
4425 if (!pci_dev_trylock(dev))
4426 goto unlock;
4427 if (dev->subordinate) {
4428 if (!pci_bus_trylock(dev->subordinate)) {
4429 pci_dev_unlock(dev);
4430 goto unlock;
4431 }
4432 }
4433 }
4434 return 1;
4435
4436 unlock:
4437 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4438 if (dev->subordinate)
4439 pci_bus_unlock(dev->subordinate);
4440 pci_dev_unlock(dev);
4441 }
4442 return 0;
4443 }
4444
4445 /* Do any devices on or below this slot prevent a bus reset? */
4446 static bool pci_slot_resetable(struct pci_slot *slot)
4447 {
4448 struct pci_dev *dev;
4449
4450 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4451 if (!dev->slot || dev->slot != slot)
4452 continue;
4453 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4454 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4455 return false;
4456 }
4457
4458 return true;
4459 }
4460
4461 /* Lock devices from the top of the tree down */
4462 static void pci_slot_lock(struct pci_slot *slot)
4463 {
4464 struct pci_dev *dev;
4465
4466 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4467 if (!dev->slot || dev->slot != slot)
4468 continue;
4469 pci_dev_lock(dev);
4470 if (dev->subordinate)
4471 pci_bus_lock(dev->subordinate);
4472 }
4473 }
4474
4475 /* Unlock devices from the bottom of the tree up */
4476 static void pci_slot_unlock(struct pci_slot *slot)
4477 {
4478 struct pci_dev *dev;
4479
4480 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4481 if (!dev->slot || dev->slot != slot)
4482 continue;
4483 if (dev->subordinate)
4484 pci_bus_unlock(dev->subordinate);
4485 pci_dev_unlock(dev);
4486 }
4487 }
4488
4489 /* Return 1 on successful lock, 0 on contention */
4490 static int pci_slot_trylock(struct pci_slot *slot)
4491 {
4492 struct pci_dev *dev;
4493
4494 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4495 if (!dev->slot || dev->slot != slot)
4496 continue;
4497 if (!pci_dev_trylock(dev))
4498 goto unlock;
4499 if (dev->subordinate) {
4500 if (!pci_bus_trylock(dev->subordinate)) {
4501 pci_dev_unlock(dev);
4502 goto unlock;
4503 }
4504 }
4505 }
4506 return 1;
4507
4508 unlock:
4509 list_for_each_entry_continue_reverse(dev,
4510 &slot->bus->devices, bus_list) {
4511 if (!dev->slot || dev->slot != slot)
4512 continue;
4513 if (dev->subordinate)
4514 pci_bus_unlock(dev->subordinate);
4515 pci_dev_unlock(dev);
4516 }
4517 return 0;
4518 }
4519
4520 /* Save and disable devices from the top of the tree down */
4521 static void pci_bus_save_and_disable(struct pci_bus *bus)
4522 {
4523 struct pci_dev *dev;
4524
4525 list_for_each_entry(dev, &bus->devices, bus_list) {
4526 pci_dev_lock(dev);
4527 pci_dev_save_and_disable(dev);
4528 pci_dev_unlock(dev);
4529 if (dev->subordinate)
4530 pci_bus_save_and_disable(dev->subordinate);
4531 }
4532 }
4533
4534 /*
4535 * Restore devices from top of the tree down - parent bridges need to be
4536 * restored before we can get to subordinate devices.
4537 */
4538 static void pci_bus_restore(struct pci_bus *bus)
4539 {
4540 struct pci_dev *dev;
4541
4542 list_for_each_entry(dev, &bus->devices, bus_list) {
4543 pci_dev_lock(dev);
4544 pci_dev_restore(dev);
4545 pci_dev_unlock(dev);
4546 if (dev->subordinate)
4547 pci_bus_restore(dev->subordinate);
4548 }
4549 }
4550
4551 /* Save and disable devices from the top of the tree down */
4552 static void pci_slot_save_and_disable(struct pci_slot *slot)
4553 {
4554 struct pci_dev *dev;
4555
4556 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4557 if (!dev->slot || dev->slot != slot)
4558 continue;
4559 pci_dev_save_and_disable(dev);
4560 if (dev->subordinate)
4561 pci_bus_save_and_disable(dev->subordinate);
4562 }
4563 }
4564
4565 /*
4566 * Restore devices from top of the tree down - parent bridges need to be
4567 * restored before we can get to subordinate devices.
4568 */
4569 static void pci_slot_restore(struct pci_slot *slot)
4570 {
4571 struct pci_dev *dev;
4572
4573 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4574 if (!dev->slot || dev->slot != slot)
4575 continue;
4576 pci_dev_restore(dev);
4577 if (dev->subordinate)
4578 pci_bus_restore(dev->subordinate);
4579 }
4580 }
4581
4582 static int pci_slot_reset(struct pci_slot *slot, int probe)
4583 {
4584 int rc;
4585
4586 if (!slot || !pci_slot_resetable(slot))
4587 return -ENOTTY;
4588
4589 if (!probe)
4590 pci_slot_lock(slot);
4591
4592 might_sleep();
4593
4594 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4595
4596 if (!probe)
4597 pci_slot_unlock(slot);
4598
4599 return rc;
4600 }
4601
4602 /**
4603 * pci_probe_reset_slot - probe whether a PCI slot can be reset
4604 * @slot: PCI slot to probe
4605 *
4606 * Return 0 if slot can be reset, negative if a slot reset is not supported.
4607 */
4608 int pci_probe_reset_slot(struct pci_slot *slot)
4609 {
4610 return pci_slot_reset(slot, 1);
4611 }
4612 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4613
4614 /**
4615 * pci_reset_slot - reset a PCI slot
4616 * @slot: PCI slot to reset
4617 *
4618 * A PCI bus may host multiple slots, each slot may support a reset mechanism
4619 * independent of other slots. For instance, some slots may support slot power
4620 * control. In the case of a 1:1 bus to slot architecture, this function may
4621 * wrap the bus reset to avoid spurious slot related events such as hotplug.
4622 * Generally a slot reset should be attempted before a bus reset. All of the
4623 * function of the slot and any subordinate buses behind the slot are reset
4624 * through this function. PCI config space of all devices in the slot and
4625 * behind the slot is saved before and restored after reset.
4626 *
4627 * Return 0 on success, non-zero on error.
4628 */
4629 int pci_reset_slot(struct pci_slot *slot)
4630 {
4631 int rc;
4632
4633 rc = pci_slot_reset(slot, 1);
4634 if (rc)
4635 return rc;
4636
4637 pci_slot_save_and_disable(slot);
4638
4639 rc = pci_slot_reset(slot, 0);
4640
4641 pci_slot_restore(slot);
4642
4643 return rc;
4644 }
4645 EXPORT_SYMBOL_GPL(pci_reset_slot);
4646
4647 /**
4648 * pci_try_reset_slot - Try to reset a PCI slot
4649 * @slot: PCI slot to reset
4650 *
4651 * Same as above except return -EAGAIN if the slot cannot be locked
4652 */
4653 int pci_try_reset_slot(struct pci_slot *slot)
4654 {
4655 int rc;
4656
4657 rc = pci_slot_reset(slot, 1);
4658 if (rc)
4659 return rc;
4660
4661 pci_slot_save_and_disable(slot);
4662
4663 if (pci_slot_trylock(slot)) {
4664 might_sleep();
4665 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4666 pci_slot_unlock(slot);
4667 } else
4668 rc = -EAGAIN;
4669
4670 pci_slot_restore(slot);
4671
4672 return rc;
4673 }
4674 EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4675
4676 static int pci_bus_reset(struct pci_bus *bus, int probe)
4677 {
4678 if (!bus->self || !pci_bus_resetable(bus))
4679 return -ENOTTY;
4680
4681 if (probe)
4682 return 0;
4683
4684 pci_bus_lock(bus);
4685
4686 might_sleep();
4687
4688 pci_reset_bridge_secondary_bus(bus->self);
4689
4690 pci_bus_unlock(bus);
4691
4692 return 0;
4693 }
4694
4695 /**
4696 * pci_probe_reset_bus - probe whether a PCI bus can be reset
4697 * @bus: PCI bus to probe
4698 *
4699 * Return 0 if bus can be reset, negative if a bus reset is not supported.
4700 */
4701 int pci_probe_reset_bus(struct pci_bus *bus)
4702 {
4703 return pci_bus_reset(bus, 1);
4704 }
4705 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4706
4707 /**
4708 * pci_reset_bus - reset a PCI bus
4709 * @bus: top level PCI bus to reset
4710 *
4711 * Do a bus reset on the given bus and any subordinate buses, saving
4712 * and restoring state of all devices.
4713 *
4714 * Return 0 on success, non-zero on error.
4715 */
4716 int pci_reset_bus(struct pci_bus *bus)
4717 {
4718 int rc;
4719
4720 rc = pci_bus_reset(bus, 1);
4721 if (rc)
4722 return rc;
4723
4724 pci_bus_save_and_disable(bus);
4725
4726 rc = pci_bus_reset(bus, 0);
4727
4728 pci_bus_restore(bus);
4729
4730 return rc;
4731 }
4732 EXPORT_SYMBOL_GPL(pci_reset_bus);
4733
4734 /**
4735 * pci_try_reset_bus - Try to reset a PCI bus
4736 * @bus: top level PCI bus to reset
4737 *
4738 * Same as above except return -EAGAIN if the bus cannot be locked
4739 */
4740 int pci_try_reset_bus(struct pci_bus *bus)
4741 {
4742 int rc;
4743
4744 rc = pci_bus_reset(bus, 1);
4745 if (rc)
4746 return rc;
4747
4748 pci_bus_save_and_disable(bus);
4749
4750 if (pci_bus_trylock(bus)) {
4751 might_sleep();
4752 pci_reset_bridge_secondary_bus(bus->self);
4753 pci_bus_unlock(bus);
4754 } else
4755 rc = -EAGAIN;
4756
4757 pci_bus_restore(bus);
4758
4759 return rc;
4760 }
4761 EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4762
4763 /**
4764 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4765 * @dev: PCI device to query
4766 *
4767 * Returns mmrbc: maximum designed memory read count in bytes
4768 * or appropriate error value.
4769 */
4770 int pcix_get_max_mmrbc(struct pci_dev *dev)
4771 {
4772 int cap;
4773 u32 stat;
4774
4775 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4776 if (!cap)
4777 return -EINVAL;
4778
4779 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4780 return -EINVAL;
4781
4782 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4783 }
4784 EXPORT_SYMBOL(pcix_get_max_mmrbc);
4785
4786 /**
4787 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4788 * @dev: PCI device to query
4789 *
4790 * Returns mmrbc: maximum memory read count in bytes
4791 * or appropriate error value.
4792 */
4793 int pcix_get_mmrbc(struct pci_dev *dev)
4794 {
4795 int cap;
4796 u16 cmd;
4797
4798 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4799 if (!cap)
4800 return -EINVAL;
4801
4802 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4803 return -EINVAL;
4804
4805 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4806 }
4807 EXPORT_SYMBOL(pcix_get_mmrbc);
4808
4809 /**
4810 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4811 * @dev: PCI device to query
4812 * @mmrbc: maximum memory read count in bytes
4813 * valid values are 512, 1024, 2048, 4096
4814 *
4815 * If possible sets maximum memory read byte count, some bridges have erratas
4816 * that prevent this.
4817 */
4818 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4819 {
4820 int cap;
4821 u32 stat, v, o;
4822 u16 cmd;
4823
4824 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4825 return -EINVAL;
4826
4827 v = ffs(mmrbc) - 10;
4828
4829 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4830 if (!cap)
4831 return -EINVAL;
4832
4833 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4834 return -EINVAL;
4835
4836 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4837 return -E2BIG;
4838
4839 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4840 return -EINVAL;
4841
4842 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4843 if (o != v) {
4844 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4845 return -EIO;
4846
4847 cmd &= ~PCI_X_CMD_MAX_READ;
4848 cmd |= v << 2;
4849 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4850 return -EIO;
4851 }
4852 return 0;
4853 }
4854 EXPORT_SYMBOL(pcix_set_mmrbc);
4855
4856 /**
4857 * pcie_get_readrq - get PCI Express read request size
4858 * @dev: PCI device to query
4859 *
4860 * Returns maximum memory read request in bytes
4861 * or appropriate error value.
4862 */
4863 int pcie_get_readrq(struct pci_dev *dev)
4864 {
4865 u16 ctl;
4866
4867 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4868
4869 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
4870 }
4871 EXPORT_SYMBOL(pcie_get_readrq);
4872
4873 /**
4874 * pcie_set_readrq - set PCI Express maximum memory read request
4875 * @dev: PCI device to query
4876 * @rq: maximum memory read count in bytes
4877 * valid values are 128, 256, 512, 1024, 2048, 4096
4878 *
4879 * If possible sets maximum memory read request in bytes
4880 */
4881 int pcie_set_readrq(struct pci_dev *dev, int rq)
4882 {
4883 u16 v;
4884
4885 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
4886 return -EINVAL;
4887
4888 /*
4889 * If using the "performance" PCIe config, we clamp the
4890 * read rq size to the max packet size to prevent the
4891 * host bridge generating requests larger than we can
4892 * cope with
4893 */
4894 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
4895 int mps = pcie_get_mps(dev);
4896
4897 if (mps < rq)
4898 rq = mps;
4899 }
4900
4901 v = (ffs(rq) - 8) << 12;
4902
4903 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4904 PCI_EXP_DEVCTL_READRQ, v);
4905 }
4906 EXPORT_SYMBOL(pcie_set_readrq);
4907
4908 /**
4909 * pcie_get_mps - get PCI Express maximum payload size
4910 * @dev: PCI device to query
4911 *
4912 * Returns maximum payload size in bytes
4913 */
4914 int pcie_get_mps(struct pci_dev *dev)
4915 {
4916 u16 ctl;
4917
4918 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4919
4920 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
4921 }
4922 EXPORT_SYMBOL(pcie_get_mps);
4923
4924 /**
4925 * pcie_set_mps - set PCI Express maximum payload size
4926 * @dev: PCI device to query
4927 * @mps: maximum payload size in bytes
4928 * valid values are 128, 256, 512, 1024, 2048, 4096
4929 *
4930 * If possible sets maximum payload size
4931 */
4932 int pcie_set_mps(struct pci_dev *dev, int mps)
4933 {
4934 u16 v;
4935
4936 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
4937 return -EINVAL;
4938
4939 v = ffs(mps) - 8;
4940 if (v > dev->pcie_mpss)
4941 return -EINVAL;
4942 v <<= 5;
4943
4944 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4945 PCI_EXP_DEVCTL_PAYLOAD, v);
4946 }
4947 EXPORT_SYMBOL(pcie_set_mps);
4948
4949 /**
4950 * pcie_get_minimum_link - determine minimum link settings of a PCI device
4951 * @dev: PCI device to query
4952 * @speed: storage for minimum speed
4953 * @width: storage for minimum width
4954 *
4955 * This function will walk up the PCI device chain and determine the minimum
4956 * link width and speed of the device.
4957 */
4958 int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
4959 enum pcie_link_width *width)
4960 {
4961 int ret;
4962
4963 *speed = PCI_SPEED_UNKNOWN;
4964 *width = PCIE_LNK_WIDTH_UNKNOWN;
4965
4966 while (dev) {
4967 u16 lnksta;
4968 enum pci_bus_speed next_speed;
4969 enum pcie_link_width next_width;
4970
4971 ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
4972 if (ret)
4973 return ret;
4974
4975 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
4976 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
4977 PCI_EXP_LNKSTA_NLW_SHIFT;
4978
4979 if (next_speed < *speed)
4980 *speed = next_speed;
4981
4982 if (next_width < *width)
4983 *width = next_width;
4984
4985 dev = dev->bus->self;
4986 }
4987
4988 return 0;
4989 }
4990 EXPORT_SYMBOL(pcie_get_minimum_link);
4991
4992 /**
4993 * pci_select_bars - Make BAR mask from the type of resource
4994 * @dev: the PCI device for which BAR mask is made
4995 * @flags: resource type mask to be selected
4996 *
4997 * This helper routine makes bar mask from the type of resource.
4998 */
4999 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
5000 {
5001 int i, bars = 0;
5002 for (i = 0; i < PCI_NUM_RESOURCES; i++)
5003 if (pci_resource_flags(dev, i) & flags)
5004 bars |= (1 << i);
5005 return bars;
5006 }
5007 EXPORT_SYMBOL(pci_select_bars);
5008
5009 /* Some architectures require additional programming to enable VGA */
5010 static arch_set_vga_state_t arch_set_vga_state;
5011
5012 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
5013 {
5014 arch_set_vga_state = func; /* NULL disables */
5015 }
5016
5017 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
5018 unsigned int command_bits, u32 flags)
5019 {
5020 if (arch_set_vga_state)
5021 return arch_set_vga_state(dev, decode, command_bits,
5022 flags);
5023 return 0;
5024 }
5025
5026 /**
5027 * pci_set_vga_state - set VGA decode state on device and parents if requested
5028 * @dev: the PCI device
5029 * @decode: true = enable decoding, false = disable decoding
5030 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
5031 * @flags: traverse ancestors and change bridges
5032 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
5033 */
5034 int pci_set_vga_state(struct pci_dev *dev, bool decode,
5035 unsigned int command_bits, u32 flags)
5036 {
5037 struct pci_bus *bus;
5038 struct pci_dev *bridge;
5039 u16 cmd;
5040 int rc;
5041
5042 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
5043
5044 /* ARCH specific VGA enables */
5045 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
5046 if (rc)
5047 return rc;
5048
5049 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
5050 pci_read_config_word(dev, PCI_COMMAND, &cmd);
5051 if (decode == true)
5052 cmd |= command_bits;
5053 else
5054 cmd &= ~command_bits;
5055 pci_write_config_word(dev, PCI_COMMAND, cmd);
5056 }
5057
5058 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
5059 return 0;
5060
5061 bus = dev->bus;
5062 while (bus) {
5063 bridge = bus->self;
5064 if (bridge) {
5065 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
5066 &cmd);
5067 if (decode == true)
5068 cmd |= PCI_BRIDGE_CTL_VGA;
5069 else
5070 cmd &= ~PCI_BRIDGE_CTL_VGA;
5071 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
5072 cmd);
5073 }
5074 bus = bus->parent;
5075 }
5076 return 0;
5077 }
5078
5079 /**
5080 * pci_add_dma_alias - Add a DMA devfn alias for a device
5081 * @dev: the PCI device for which alias is added
5082 * @devfn: alias slot and function
5083 *
5084 * This helper encodes 8-bit devfn as bit number in dma_alias_mask.
5085 * It should be called early, preferably as PCI fixup header quirk.
5086 */
5087 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
5088 {
5089 if (!dev->dma_alias_mask)
5090 dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
5091 sizeof(long), GFP_KERNEL);
5092 if (!dev->dma_alias_mask) {
5093 dev_warn(&dev->dev, "Unable to allocate DMA alias mask\n");
5094 return;
5095 }
5096
5097 set_bit(devfn, dev->dma_alias_mask);
5098 dev_info(&dev->dev, "Enabling fixed DMA alias to %02x.%d\n",
5099 PCI_SLOT(devfn), PCI_FUNC(devfn));
5100 }
5101
5102 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
5103 {
5104 return (dev1->dma_alias_mask &&
5105 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
5106 (dev2->dma_alias_mask &&
5107 test_bit(dev1->devfn, dev2->dma_alias_mask));
5108 }
5109
5110 bool pci_device_is_present(struct pci_dev *pdev)
5111 {
5112 u32 v;
5113
5114 if (pci_dev_is_disconnected(pdev))
5115 return false;
5116 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
5117 }
5118 EXPORT_SYMBOL_GPL(pci_device_is_present);
5119
5120 void pci_ignore_hotplug(struct pci_dev *dev)
5121 {
5122 struct pci_dev *bridge = dev->bus->self;
5123
5124 dev->ignore_hotplug = 1;
5125 /* Propagate the "ignore hotplug" setting to the parent bridge. */
5126 if (bridge)
5127 bridge->ignore_hotplug = 1;
5128 }
5129 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
5130
5131 resource_size_t __weak pcibios_default_alignment(void)
5132 {
5133 return 0;
5134 }
5135
5136 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
5137 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
5138 static DEFINE_SPINLOCK(resource_alignment_lock);
5139
5140 /**
5141 * pci_specified_resource_alignment - get resource alignment specified by user.
5142 * @dev: the PCI device to get
5143 * @resize: whether or not to change resources' size when reassigning alignment
5144 *
5145 * RETURNS: Resource alignment if it is specified.
5146 * Zero if it is not specified.
5147 */
5148 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev,
5149 bool *resize)
5150 {
5151 int seg, bus, slot, func, align_order, count;
5152 unsigned short vendor, device, subsystem_vendor, subsystem_device;
5153 resource_size_t align = pcibios_default_alignment();
5154 char *p;
5155
5156 spin_lock(&resource_alignment_lock);
5157 p = resource_alignment_param;
5158 if (!*p && !align)
5159 goto out;
5160 if (pci_has_flag(PCI_PROBE_ONLY)) {
5161 align = 0;
5162 pr_info_once("PCI: Ignoring requested alignments (PCI_PROBE_ONLY)\n");
5163 goto out;
5164 }
5165
5166 while (*p) {
5167 count = 0;
5168 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
5169 p[count] == '@') {
5170 p += count + 1;
5171 } else {
5172 align_order = -1;
5173 }
5174 if (strncmp(p, "pci:", 4) == 0) {
5175 /* PCI vendor/device (subvendor/subdevice) ids are specified */
5176 p += 4;
5177 if (sscanf(p, "%hx:%hx:%hx:%hx%n",
5178 &vendor, &device, &subsystem_vendor, &subsystem_device, &count) != 4) {
5179 if (sscanf(p, "%hx:%hx%n", &vendor, &device, &count) != 2) {
5180 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: pci:%s\n",
5181 p);
5182 break;
5183 }
5184 subsystem_vendor = subsystem_device = 0;
5185 }
5186 p += count;
5187 if ((!vendor || (vendor == dev->vendor)) &&
5188 (!device || (device == dev->device)) &&
5189 (!subsystem_vendor || (subsystem_vendor == dev->subsystem_vendor)) &&
5190 (!subsystem_device || (subsystem_device == dev->subsystem_device))) {
5191 *resize = true;
5192 if (align_order == -1)
5193 align = PAGE_SIZE;
5194 else
5195 align = 1 << align_order;
5196 /* Found */
5197 break;
5198 }
5199 }
5200 else {
5201 if (sscanf(p, "%x:%x:%x.%x%n",
5202 &seg, &bus, &slot, &func, &count) != 4) {
5203 seg = 0;
5204 if (sscanf(p, "%x:%x.%x%n",
5205 &bus, &slot, &func, &count) != 3) {
5206 /* Invalid format */
5207 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
5208 p);
5209 break;
5210 }
5211 }
5212 p += count;
5213 if (seg == pci_domain_nr(dev->bus) &&
5214 bus == dev->bus->number &&
5215 slot == PCI_SLOT(dev->devfn) &&
5216 func == PCI_FUNC(dev->devfn)) {
5217 *resize = true;
5218 if (align_order == -1)
5219 align = PAGE_SIZE;
5220 else
5221 align = 1 << align_order;
5222 /* Found */
5223 break;
5224 }
5225 }
5226 if (*p != ';' && *p != ',') {
5227 /* End of param or invalid format */
5228 break;
5229 }
5230 p++;
5231 }
5232 out:
5233 spin_unlock(&resource_alignment_lock);
5234 return align;
5235 }
5236
5237 static void pci_request_resource_alignment(struct pci_dev *dev, int bar,
5238 resource_size_t align, bool resize)
5239 {
5240 struct resource *r = &dev->resource[bar];
5241 resource_size_t size;
5242
5243 if (!(r->flags & IORESOURCE_MEM))
5244 return;
5245
5246 if (r->flags & IORESOURCE_PCI_FIXED) {
5247 dev_info(&dev->dev, "BAR%d %pR: ignoring requested alignment %#llx\n",
5248 bar, r, (unsigned long long)align);
5249 return;
5250 }
5251
5252 size = resource_size(r);
5253 if (size >= align)
5254 return;
5255
5256 /*
5257 * Increase the alignment of the resource. There are two ways we
5258 * can do this:
5259 *
5260 * 1) Increase the size of the resource. BARs are aligned on their
5261 * size, so when we reallocate space for this resource, we'll
5262 * allocate it with the larger alignment. This also prevents
5263 * assignment of any other BARs inside the alignment region, so
5264 * if we're requesting page alignment, this means no other BARs
5265 * will share the page.
5266 *
5267 * The disadvantage is that this makes the resource larger than
5268 * the hardware BAR, which may break drivers that compute things
5269 * based on the resource size, e.g., to find registers at a
5270 * fixed offset before the end of the BAR.
5271 *
5272 * 2) Retain the resource size, but use IORESOURCE_STARTALIGN and
5273 * set r->start to the desired alignment. By itself this
5274 * doesn't prevent other BARs being put inside the alignment
5275 * region, but if we realign *every* resource of every device in
5276 * the system, none of them will share an alignment region.
5277 *
5278 * When the user has requested alignment for only some devices via
5279 * the "pci=resource_alignment" argument, "resize" is true and we
5280 * use the first method. Otherwise we assume we're aligning all
5281 * devices and we use the second.
5282 */
5283
5284 dev_info(&dev->dev, "BAR%d %pR: requesting alignment to %#llx\n",
5285 bar, r, (unsigned long long)align);
5286
5287 if (resize) {
5288 r->start = 0;
5289 r->end = align - 1;
5290 } else {
5291 r->flags &= ~IORESOURCE_SIZEALIGN;
5292 r->flags |= IORESOURCE_STARTALIGN;
5293 r->start = align;
5294 r->end = r->start + size - 1;
5295 }
5296 r->flags |= IORESOURCE_UNSET;
5297 }
5298
5299 /*
5300 * This function disables memory decoding and releases memory resources
5301 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
5302 * It also rounds up size to specified alignment.
5303 * Later on, the kernel will assign page-aligned memory resource back
5304 * to the device.
5305 */
5306 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
5307 {
5308 int i;
5309 struct resource *r;
5310 resource_size_t align;
5311 u16 command;
5312 bool resize = false;
5313
5314 /*
5315 * VF BARs are read-only zero according to SR-IOV spec r1.1, sec
5316 * 3.4.1.11. Their resources are allocated from the space
5317 * described by the VF BARx register in the PF's SR-IOV capability.
5318 * We can't influence their alignment here.
5319 */
5320 if (dev->is_virtfn)
5321 return;
5322
5323 /* check if specified PCI is target device to reassign */
5324 align = pci_specified_resource_alignment(dev, &resize);
5325 if (!align)
5326 return;
5327
5328 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
5329 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
5330 dev_warn(&dev->dev,
5331 "Can't reassign resources to host bridge.\n");
5332 return;
5333 }
5334
5335 dev_info(&dev->dev,
5336 "Disabling memory decoding and releasing memory resources.\n");
5337 pci_read_config_word(dev, PCI_COMMAND, &command);
5338 command &= ~PCI_COMMAND_MEMORY;
5339 pci_write_config_word(dev, PCI_COMMAND, command);
5340
5341 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
5342 pci_request_resource_alignment(dev, i, align, resize);
5343
5344 /*
5345 * Need to disable bridge's resource window,
5346 * to enable the kernel to reassign new resource
5347 * window later on.
5348 */
5349 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
5350 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
5351 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
5352 r = &dev->resource[i];
5353 if (!(r->flags & IORESOURCE_MEM))
5354 continue;
5355 r->flags |= IORESOURCE_UNSET;
5356 r->end = resource_size(r) - 1;
5357 r->start = 0;
5358 }
5359 pci_disable_bridge_window(dev);
5360 }
5361 }
5362
5363 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
5364 {
5365 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
5366 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
5367 spin_lock(&resource_alignment_lock);
5368 strncpy(resource_alignment_param, buf, count);
5369 resource_alignment_param[count] = '\0';
5370 spin_unlock(&resource_alignment_lock);
5371 return count;
5372 }
5373
5374 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
5375 {
5376 size_t count;
5377 spin_lock(&resource_alignment_lock);
5378 count = snprintf(buf, size, "%s", resource_alignment_param);
5379 spin_unlock(&resource_alignment_lock);
5380 return count;
5381 }
5382
5383 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
5384 {
5385 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
5386 }
5387
5388 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
5389 const char *buf, size_t count)
5390 {
5391 return pci_set_resource_alignment_param(buf, count);
5392 }
5393
5394 static BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
5395 pci_resource_alignment_store);
5396
5397 static int __init pci_resource_alignment_sysfs_init(void)
5398 {
5399 return bus_create_file(&pci_bus_type,
5400 &bus_attr_resource_alignment);
5401 }
5402 late_initcall(pci_resource_alignment_sysfs_init);
5403
5404 static void pci_no_domains(void)
5405 {
5406 #ifdef CONFIG_PCI_DOMAINS
5407 pci_domains_supported = 0;
5408 #endif
5409 }
5410
5411 #ifdef CONFIG_PCI_DOMAINS
5412 static atomic_t __domain_nr = ATOMIC_INIT(-1);
5413
5414 int pci_get_new_domain_nr(void)
5415 {
5416 return atomic_inc_return(&__domain_nr);
5417 }
5418
5419 #ifdef CONFIG_PCI_DOMAINS_GENERIC
5420 static int of_pci_bus_find_domain_nr(struct device *parent)
5421 {
5422 static int use_dt_domains = -1;
5423 int domain = -1;
5424
5425 if (parent)
5426 domain = of_get_pci_domain_nr(parent->of_node);
5427 /*
5428 * Check DT domain and use_dt_domains values.
5429 *
5430 * If DT domain property is valid (domain >= 0) and
5431 * use_dt_domains != 0, the DT assignment is valid since this means
5432 * we have not previously allocated a domain number by using
5433 * pci_get_new_domain_nr(); we should also update use_dt_domains to
5434 * 1, to indicate that we have just assigned a domain number from
5435 * DT.
5436 *
5437 * If DT domain property value is not valid (ie domain < 0), and we
5438 * have not previously assigned a domain number from DT
5439 * (use_dt_domains != 1) we should assign a domain number by
5440 * using the:
5441 *
5442 * pci_get_new_domain_nr()
5443 *
5444 * API and update the use_dt_domains value to keep track of method we
5445 * are using to assign domain numbers (use_dt_domains = 0).
5446 *
5447 * All other combinations imply we have a platform that is trying
5448 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
5449 * which is a recipe for domain mishandling and it is prevented by
5450 * invalidating the domain value (domain = -1) and printing a
5451 * corresponding error.
5452 */
5453 if (domain >= 0 && use_dt_domains) {
5454 use_dt_domains = 1;
5455 } else if (domain < 0 && use_dt_domains != 1) {
5456 use_dt_domains = 0;
5457 domain = pci_get_new_domain_nr();
5458 } else {
5459 dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
5460 parent->of_node->full_name);
5461 domain = -1;
5462 }
5463
5464 return domain;
5465 }
5466
5467 int pci_bus_find_domain_nr(struct pci_bus *bus, struct device *parent)
5468 {
5469 return acpi_disabled ? of_pci_bus_find_domain_nr(parent) :
5470 acpi_pci_bus_find_domain_nr(bus);
5471 }
5472 #endif
5473 #endif
5474
5475 /**
5476 * pci_ext_cfg_avail - can we access extended PCI config space?
5477 *
5478 * Returns 1 if we can access PCI extended config space (offsets
5479 * greater than 0xff). This is the default implementation. Architecture
5480 * implementations can override this.
5481 */
5482 int __weak pci_ext_cfg_avail(void)
5483 {
5484 return 1;
5485 }
5486
5487 void __weak pci_fixup_cardbus(struct pci_bus *bus)
5488 {
5489 }
5490 EXPORT_SYMBOL(pci_fixup_cardbus);
5491
5492 static int __init pci_setup(char *str)
5493 {
5494 while (str) {
5495 char *k = strchr(str, ',');
5496 if (k)
5497 *k++ = 0;
5498 if (*str && (str = pcibios_setup(str)) && *str) {
5499 if (!strcmp(str, "nomsi")) {
5500 pci_no_msi();
5501 } else if (!strcmp(str, "noaer")) {
5502 pci_no_aer();
5503 } else if (!strncmp(str, "realloc=", 8)) {
5504 pci_realloc_get_opt(str + 8);
5505 } else if (!strncmp(str, "realloc", 7)) {
5506 pci_realloc_get_opt("on");
5507 } else if (!strcmp(str, "nodomains")) {
5508 pci_no_domains();
5509 } else if (!strncmp(str, "noari", 5)) {
5510 pcie_ari_disabled = true;
5511 } else if (!strncmp(str, "cbiosize=", 9)) {
5512 pci_cardbus_io_size = memparse(str + 9, &str);
5513 } else if (!strncmp(str, "cbmemsize=", 10)) {
5514 pci_cardbus_mem_size = memparse(str + 10, &str);
5515 } else if (!strncmp(str, "resource_alignment=", 19)) {
5516 pci_set_resource_alignment_param(str + 19,
5517 strlen(str + 19));
5518 } else if (!strncmp(str, "ecrc=", 5)) {
5519 pcie_ecrc_get_policy(str + 5);
5520 } else if (!strncmp(str, "hpiosize=", 9)) {
5521 pci_hotplug_io_size = memparse(str + 9, &str);
5522 } else if (!strncmp(str, "hpmemsize=", 10)) {
5523 pci_hotplug_mem_size = memparse(str + 10, &str);
5524 } else if (!strncmp(str, "hpbussize=", 10)) {
5525 pci_hotplug_bus_size =
5526 simple_strtoul(str + 10, &str, 0);
5527 if (pci_hotplug_bus_size > 0xff)
5528 pci_hotplug_bus_size = DEFAULT_HOTPLUG_BUS_SIZE;
5529 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
5530 pcie_bus_config = PCIE_BUS_TUNE_OFF;
5531 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
5532 pcie_bus_config = PCIE_BUS_SAFE;
5533 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
5534 pcie_bus_config = PCIE_BUS_PERFORMANCE;
5535 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
5536 pcie_bus_config = PCIE_BUS_PEER2PEER;
5537 } else if (!strncmp(str, "pcie_scan_all", 13)) {
5538 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
5539 } else {
5540 printk(KERN_ERR "PCI: Unknown option `%s'\n",
5541 str);
5542 }
5543 }
5544 str = k;
5545 }
5546 return 0;
5547 }
5548 early_param("pci", pci_setup);