Merge tag 'pstore-v4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux
[deliverable/linux.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/kernel.h>
11 #include <linux/delay.h>
12 #include <linux/init.h>
13 #include <linux/of.h>
14 #include <linux/of_pci.h>
15 #include <linux/pci.h>
16 #include <linux/pm.h>
17 #include <linux/slab.h>
18 #include <linux/module.h>
19 #include <linux/spinlock.h>
20 #include <linux/string.h>
21 #include <linux/log2.h>
22 #include <linux/pci-aspm.h>
23 #include <linux/pm_wakeup.h>
24 #include <linux/interrupt.h>
25 #include <linux/device.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/pci_hotplug.h>
28 #include <asm/setup.h>
29 #include <linux/aer.h>
30 #include "pci.h"
31
32 const char *pci_power_names[] = {
33 "error", "D0", "D1", "D2", "D3hot", "D3cold", "unknown",
34 };
35 EXPORT_SYMBOL_GPL(pci_power_names);
36
37 int isa_dma_bridge_buggy;
38 EXPORT_SYMBOL(isa_dma_bridge_buggy);
39
40 int pci_pci_problems;
41 EXPORT_SYMBOL(pci_pci_problems);
42
43 unsigned int pci_pm_d3_delay;
44
45 static void pci_pme_list_scan(struct work_struct *work);
46
47 static LIST_HEAD(pci_pme_list);
48 static DEFINE_MUTEX(pci_pme_list_mutex);
49 static DECLARE_DELAYED_WORK(pci_pme_work, pci_pme_list_scan);
50
51 struct pci_pme_device {
52 struct list_head list;
53 struct pci_dev *dev;
54 };
55
56 #define PME_TIMEOUT 1000 /* How long between PME checks */
57
58 static void pci_dev_d3_sleep(struct pci_dev *dev)
59 {
60 unsigned int delay = dev->d3_delay;
61
62 if (delay < pci_pm_d3_delay)
63 delay = pci_pm_d3_delay;
64
65 msleep(delay);
66 }
67
68 #ifdef CONFIG_PCI_DOMAINS
69 int pci_domains_supported = 1;
70 #endif
71
72 #define DEFAULT_CARDBUS_IO_SIZE (256)
73 #define DEFAULT_CARDBUS_MEM_SIZE (64*1024*1024)
74 /* pci=cbmemsize=nnM,cbiosize=nn can override this */
75 unsigned long pci_cardbus_io_size = DEFAULT_CARDBUS_IO_SIZE;
76 unsigned long pci_cardbus_mem_size = DEFAULT_CARDBUS_MEM_SIZE;
77
78 #define DEFAULT_HOTPLUG_IO_SIZE (256)
79 #define DEFAULT_HOTPLUG_MEM_SIZE (2*1024*1024)
80 /* pci=hpmemsize=nnM,hpiosize=nn can override this */
81 unsigned long pci_hotplug_io_size = DEFAULT_HOTPLUG_IO_SIZE;
82 unsigned long pci_hotplug_mem_size = DEFAULT_HOTPLUG_MEM_SIZE;
83
84 enum pcie_bus_config_types pcie_bus_config = PCIE_BUS_DEFAULT;
85
86 /*
87 * The default CLS is used if arch didn't set CLS explicitly and not
88 * all pci devices agree on the same value. Arch can override either
89 * the dfl or actual value as it sees fit. Don't forget this is
90 * measured in 32-bit words, not bytes.
91 */
92 u8 pci_dfl_cache_line_size = L1_CACHE_BYTES >> 2;
93 u8 pci_cache_line_size;
94
95 /*
96 * If we set up a device for bus mastering, we need to check the latency
97 * timer as certain BIOSes forget to set it properly.
98 */
99 unsigned int pcibios_max_latency = 255;
100
101 /* If set, the PCIe ARI capability will not be used. */
102 static bool pcie_ari_disabled;
103
104 /**
105 * pci_bus_max_busnr - returns maximum PCI bus number of given bus' children
106 * @bus: pointer to PCI bus structure to search
107 *
108 * Given a PCI bus, returns the highest PCI bus number present in the set
109 * including the given PCI bus and its list of child PCI buses.
110 */
111 unsigned char pci_bus_max_busnr(struct pci_bus *bus)
112 {
113 struct pci_bus *tmp;
114 unsigned char max, n;
115
116 max = bus->busn_res.end;
117 list_for_each_entry(tmp, &bus->children, node) {
118 n = pci_bus_max_busnr(tmp);
119 if (n > max)
120 max = n;
121 }
122 return max;
123 }
124 EXPORT_SYMBOL_GPL(pci_bus_max_busnr);
125
126 #ifdef CONFIG_HAS_IOMEM
127 void __iomem *pci_ioremap_bar(struct pci_dev *pdev, int bar)
128 {
129 struct resource *res = &pdev->resource[bar];
130
131 /*
132 * Make sure the BAR is actually a memory resource, not an IO resource
133 */
134 if (res->flags & IORESOURCE_UNSET || !(res->flags & IORESOURCE_MEM)) {
135 dev_warn(&pdev->dev, "can't ioremap BAR %d: %pR\n", bar, res);
136 return NULL;
137 }
138 return ioremap_nocache(res->start, resource_size(res));
139 }
140 EXPORT_SYMBOL_GPL(pci_ioremap_bar);
141
142 void __iomem *pci_ioremap_wc_bar(struct pci_dev *pdev, int bar)
143 {
144 /*
145 * Make sure the BAR is actually a memory resource, not an IO resource
146 */
147 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) {
148 WARN_ON(1);
149 return NULL;
150 }
151 return ioremap_wc(pci_resource_start(pdev, bar),
152 pci_resource_len(pdev, bar));
153 }
154 EXPORT_SYMBOL_GPL(pci_ioremap_wc_bar);
155 #endif
156
157
158 static int __pci_find_next_cap_ttl(struct pci_bus *bus, unsigned int devfn,
159 u8 pos, int cap, int *ttl)
160 {
161 u8 id;
162 u16 ent;
163
164 pci_bus_read_config_byte(bus, devfn, pos, &pos);
165
166 while ((*ttl)--) {
167 if (pos < 0x40)
168 break;
169 pos &= ~3;
170 pci_bus_read_config_word(bus, devfn, pos, &ent);
171
172 id = ent & 0xff;
173 if (id == 0xff)
174 break;
175 if (id == cap)
176 return pos;
177 pos = (ent >> 8);
178 }
179 return 0;
180 }
181
182 static int __pci_find_next_cap(struct pci_bus *bus, unsigned int devfn,
183 u8 pos, int cap)
184 {
185 int ttl = PCI_FIND_CAP_TTL;
186
187 return __pci_find_next_cap_ttl(bus, devfn, pos, cap, &ttl);
188 }
189
190 int pci_find_next_capability(struct pci_dev *dev, u8 pos, int cap)
191 {
192 return __pci_find_next_cap(dev->bus, dev->devfn,
193 pos + PCI_CAP_LIST_NEXT, cap);
194 }
195 EXPORT_SYMBOL_GPL(pci_find_next_capability);
196
197 static int __pci_bus_find_cap_start(struct pci_bus *bus,
198 unsigned int devfn, u8 hdr_type)
199 {
200 u16 status;
201
202 pci_bus_read_config_word(bus, devfn, PCI_STATUS, &status);
203 if (!(status & PCI_STATUS_CAP_LIST))
204 return 0;
205
206 switch (hdr_type) {
207 case PCI_HEADER_TYPE_NORMAL:
208 case PCI_HEADER_TYPE_BRIDGE:
209 return PCI_CAPABILITY_LIST;
210 case PCI_HEADER_TYPE_CARDBUS:
211 return PCI_CB_CAPABILITY_LIST;
212 }
213
214 return 0;
215 }
216
217 /**
218 * pci_find_capability - query for devices' capabilities
219 * @dev: PCI device to query
220 * @cap: capability code
221 *
222 * Tell if a device supports a given PCI capability.
223 * Returns the address of the requested capability structure within the
224 * device's PCI configuration space or 0 in case the device does not
225 * support it. Possible values for @cap:
226 *
227 * %PCI_CAP_ID_PM Power Management
228 * %PCI_CAP_ID_AGP Accelerated Graphics Port
229 * %PCI_CAP_ID_VPD Vital Product Data
230 * %PCI_CAP_ID_SLOTID Slot Identification
231 * %PCI_CAP_ID_MSI Message Signalled Interrupts
232 * %PCI_CAP_ID_CHSWP CompactPCI HotSwap
233 * %PCI_CAP_ID_PCIX PCI-X
234 * %PCI_CAP_ID_EXP PCI Express
235 */
236 int pci_find_capability(struct pci_dev *dev, int cap)
237 {
238 int pos;
239
240 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
241 if (pos)
242 pos = __pci_find_next_cap(dev->bus, dev->devfn, pos, cap);
243
244 return pos;
245 }
246 EXPORT_SYMBOL(pci_find_capability);
247
248 /**
249 * pci_bus_find_capability - query for devices' capabilities
250 * @bus: the PCI bus to query
251 * @devfn: PCI device to query
252 * @cap: capability code
253 *
254 * Like pci_find_capability() but works for pci devices that do not have a
255 * pci_dev structure set up yet.
256 *
257 * Returns the address of the requested capability structure within the
258 * device's PCI configuration space or 0 in case the device does not
259 * support it.
260 */
261 int pci_bus_find_capability(struct pci_bus *bus, unsigned int devfn, int cap)
262 {
263 int pos;
264 u8 hdr_type;
265
266 pci_bus_read_config_byte(bus, devfn, PCI_HEADER_TYPE, &hdr_type);
267
268 pos = __pci_bus_find_cap_start(bus, devfn, hdr_type & 0x7f);
269 if (pos)
270 pos = __pci_find_next_cap(bus, devfn, pos, cap);
271
272 return pos;
273 }
274 EXPORT_SYMBOL(pci_bus_find_capability);
275
276 /**
277 * pci_find_next_ext_capability - Find an extended capability
278 * @dev: PCI device to query
279 * @start: address at which to start looking (0 to start at beginning of list)
280 * @cap: capability code
281 *
282 * Returns the address of the next matching extended capability structure
283 * within the device's PCI configuration space or 0 if the device does
284 * not support it. Some capabilities can occur several times, e.g., the
285 * vendor-specific capability, and this provides a way to find them all.
286 */
287 int pci_find_next_ext_capability(struct pci_dev *dev, int start, int cap)
288 {
289 u32 header;
290 int ttl;
291 int pos = PCI_CFG_SPACE_SIZE;
292
293 /* minimum 8 bytes per capability */
294 ttl = (PCI_CFG_SPACE_EXP_SIZE - PCI_CFG_SPACE_SIZE) / 8;
295
296 if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
297 return 0;
298
299 if (start)
300 pos = start;
301
302 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
303 return 0;
304
305 /*
306 * If we have no capabilities, this is indicated by cap ID,
307 * cap version and next pointer all being 0.
308 */
309 if (header == 0)
310 return 0;
311
312 while (ttl-- > 0) {
313 if (PCI_EXT_CAP_ID(header) == cap && pos != start)
314 return pos;
315
316 pos = PCI_EXT_CAP_NEXT(header);
317 if (pos < PCI_CFG_SPACE_SIZE)
318 break;
319
320 if (pci_read_config_dword(dev, pos, &header) != PCIBIOS_SUCCESSFUL)
321 break;
322 }
323
324 return 0;
325 }
326 EXPORT_SYMBOL_GPL(pci_find_next_ext_capability);
327
328 /**
329 * pci_find_ext_capability - Find an extended capability
330 * @dev: PCI device to query
331 * @cap: capability code
332 *
333 * Returns the address of the requested extended capability structure
334 * within the device's PCI configuration space or 0 if the device does
335 * not support it. Possible values for @cap:
336 *
337 * %PCI_EXT_CAP_ID_ERR Advanced Error Reporting
338 * %PCI_EXT_CAP_ID_VC Virtual Channel
339 * %PCI_EXT_CAP_ID_DSN Device Serial Number
340 * %PCI_EXT_CAP_ID_PWR Power Budgeting
341 */
342 int pci_find_ext_capability(struct pci_dev *dev, int cap)
343 {
344 return pci_find_next_ext_capability(dev, 0, cap);
345 }
346 EXPORT_SYMBOL_GPL(pci_find_ext_capability);
347
348 static int __pci_find_next_ht_cap(struct pci_dev *dev, int pos, int ht_cap)
349 {
350 int rc, ttl = PCI_FIND_CAP_TTL;
351 u8 cap, mask;
352
353 if (ht_cap == HT_CAPTYPE_SLAVE || ht_cap == HT_CAPTYPE_HOST)
354 mask = HT_3BIT_CAP_MASK;
355 else
356 mask = HT_5BIT_CAP_MASK;
357
358 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn, pos,
359 PCI_CAP_ID_HT, &ttl);
360 while (pos) {
361 rc = pci_read_config_byte(dev, pos + 3, &cap);
362 if (rc != PCIBIOS_SUCCESSFUL)
363 return 0;
364
365 if ((cap & mask) == ht_cap)
366 return pos;
367
368 pos = __pci_find_next_cap_ttl(dev->bus, dev->devfn,
369 pos + PCI_CAP_LIST_NEXT,
370 PCI_CAP_ID_HT, &ttl);
371 }
372
373 return 0;
374 }
375 /**
376 * pci_find_next_ht_capability - query a device's Hypertransport capabilities
377 * @dev: PCI device to query
378 * @pos: Position from which to continue searching
379 * @ht_cap: Hypertransport capability code
380 *
381 * To be used in conjunction with pci_find_ht_capability() to search for
382 * all capabilities matching @ht_cap. @pos should always be a value returned
383 * from pci_find_ht_capability().
384 *
385 * NB. To be 100% safe against broken PCI devices, the caller should take
386 * steps to avoid an infinite loop.
387 */
388 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap)
389 {
390 return __pci_find_next_ht_cap(dev, pos + PCI_CAP_LIST_NEXT, ht_cap);
391 }
392 EXPORT_SYMBOL_GPL(pci_find_next_ht_capability);
393
394 /**
395 * pci_find_ht_capability - query a device's Hypertransport capabilities
396 * @dev: PCI device to query
397 * @ht_cap: Hypertransport capability code
398 *
399 * Tell if a device supports a given Hypertransport capability.
400 * Returns an address within the device's PCI configuration space
401 * or 0 in case the device does not support the request capability.
402 * The address points to the PCI capability, of type PCI_CAP_ID_HT,
403 * which has a Hypertransport capability matching @ht_cap.
404 */
405 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap)
406 {
407 int pos;
408
409 pos = __pci_bus_find_cap_start(dev->bus, dev->devfn, dev->hdr_type);
410 if (pos)
411 pos = __pci_find_next_ht_cap(dev, pos, ht_cap);
412
413 return pos;
414 }
415 EXPORT_SYMBOL_GPL(pci_find_ht_capability);
416
417 /**
418 * pci_find_parent_resource - return resource region of parent bus of given region
419 * @dev: PCI device structure contains resources to be searched
420 * @res: child resource record for which parent is sought
421 *
422 * For given resource region of given device, return the resource
423 * region of parent bus the given region is contained in.
424 */
425 struct resource *pci_find_parent_resource(const struct pci_dev *dev,
426 struct resource *res)
427 {
428 const struct pci_bus *bus = dev->bus;
429 struct resource *r;
430 int i;
431
432 pci_bus_for_each_resource(bus, r, i) {
433 if (!r)
434 continue;
435 if (res->start && resource_contains(r, res)) {
436
437 /*
438 * If the window is prefetchable but the BAR is
439 * not, the allocator made a mistake.
440 */
441 if (r->flags & IORESOURCE_PREFETCH &&
442 !(res->flags & IORESOURCE_PREFETCH))
443 return NULL;
444
445 /*
446 * If we're below a transparent bridge, there may
447 * be both a positively-decoded aperture and a
448 * subtractively-decoded region that contain the BAR.
449 * We want the positively-decoded one, so this depends
450 * on pci_bus_for_each_resource() giving us those
451 * first.
452 */
453 return r;
454 }
455 }
456 return NULL;
457 }
458 EXPORT_SYMBOL(pci_find_parent_resource);
459
460 /**
461 * pci_find_pcie_root_port - return PCIe Root Port
462 * @dev: PCI device to query
463 *
464 * Traverse up the parent chain and return the PCIe Root Port PCI Device
465 * for a given PCI Device.
466 */
467 struct pci_dev *pci_find_pcie_root_port(struct pci_dev *dev)
468 {
469 struct pci_dev *bridge, *highest_pcie_bridge = NULL;
470
471 bridge = pci_upstream_bridge(dev);
472 while (bridge && pci_is_pcie(bridge)) {
473 highest_pcie_bridge = bridge;
474 bridge = pci_upstream_bridge(bridge);
475 }
476
477 if (pci_pcie_type(highest_pcie_bridge) != PCI_EXP_TYPE_ROOT_PORT)
478 return NULL;
479
480 return highest_pcie_bridge;
481 }
482 EXPORT_SYMBOL(pci_find_pcie_root_port);
483
484 /**
485 * pci_wait_for_pending - wait for @mask bit(s) to clear in status word @pos
486 * @dev: the PCI device to operate on
487 * @pos: config space offset of status word
488 * @mask: mask of bit(s) to care about in status word
489 *
490 * Return 1 when mask bit(s) in status word clear, 0 otherwise.
491 */
492 int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask)
493 {
494 int i;
495
496 /* Wait for Transaction Pending bit clean */
497 for (i = 0; i < 4; i++) {
498 u16 status;
499 if (i)
500 msleep((1 << (i - 1)) * 100);
501
502 pci_read_config_word(dev, pos, &status);
503 if (!(status & mask))
504 return 1;
505 }
506
507 return 0;
508 }
509
510 /**
511 * pci_restore_bars - restore a device's BAR values (e.g. after wake-up)
512 * @dev: PCI device to have its BARs restored
513 *
514 * Restore the BAR values for a given device, so as to make it
515 * accessible by its driver.
516 */
517 static void pci_restore_bars(struct pci_dev *dev)
518 {
519 int i;
520
521 /* Per SR-IOV spec 3.4.1.11, VF BARs are RO zero */
522 if (dev->is_virtfn)
523 return;
524
525 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++)
526 pci_update_resource(dev, i);
527 }
528
529 static const struct pci_platform_pm_ops *pci_platform_pm;
530
531 int pci_set_platform_pm(const struct pci_platform_pm_ops *ops)
532 {
533 if (!ops->is_manageable || !ops->set_state || !ops->choose_state ||
534 !ops->sleep_wake || !ops->run_wake || !ops->need_resume)
535 return -EINVAL;
536 pci_platform_pm = ops;
537 return 0;
538 }
539
540 static inline bool platform_pci_power_manageable(struct pci_dev *dev)
541 {
542 return pci_platform_pm ? pci_platform_pm->is_manageable(dev) : false;
543 }
544
545 static inline int platform_pci_set_power_state(struct pci_dev *dev,
546 pci_power_t t)
547 {
548 return pci_platform_pm ? pci_platform_pm->set_state(dev, t) : -ENOSYS;
549 }
550
551 static inline pci_power_t platform_pci_choose_state(struct pci_dev *dev)
552 {
553 return pci_platform_pm ?
554 pci_platform_pm->choose_state(dev) : PCI_POWER_ERROR;
555 }
556
557 static inline int platform_pci_sleep_wake(struct pci_dev *dev, bool enable)
558 {
559 return pci_platform_pm ?
560 pci_platform_pm->sleep_wake(dev, enable) : -ENODEV;
561 }
562
563 static inline int platform_pci_run_wake(struct pci_dev *dev, bool enable)
564 {
565 return pci_platform_pm ?
566 pci_platform_pm->run_wake(dev, enable) : -ENODEV;
567 }
568
569 static inline bool platform_pci_need_resume(struct pci_dev *dev)
570 {
571 return pci_platform_pm ? pci_platform_pm->need_resume(dev) : false;
572 }
573
574 /**
575 * pci_raw_set_power_state - Use PCI PM registers to set the power state of
576 * given PCI device
577 * @dev: PCI device to handle.
578 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
579 *
580 * RETURN VALUE:
581 * -EINVAL if the requested state is invalid.
582 * -EIO if device does not support PCI PM or its PM capabilities register has a
583 * wrong version, or device doesn't support the requested state.
584 * 0 if device already is in the requested state.
585 * 0 if device's power state has been successfully changed.
586 */
587 static int pci_raw_set_power_state(struct pci_dev *dev, pci_power_t state)
588 {
589 u16 pmcsr;
590 bool need_restore = false;
591
592 /* Check if we're already there */
593 if (dev->current_state == state)
594 return 0;
595
596 if (!dev->pm_cap)
597 return -EIO;
598
599 if (state < PCI_D0 || state > PCI_D3hot)
600 return -EINVAL;
601
602 /* Validate current state:
603 * Can enter D0 from any state, but if we can only go deeper
604 * to sleep if we're already in a low power state
605 */
606 if (state != PCI_D0 && dev->current_state <= PCI_D3cold
607 && dev->current_state > state) {
608 dev_err(&dev->dev, "invalid power transition (from state %d to %d)\n",
609 dev->current_state, state);
610 return -EINVAL;
611 }
612
613 /* check if this device supports the desired state */
614 if ((state == PCI_D1 && !dev->d1_support)
615 || (state == PCI_D2 && !dev->d2_support))
616 return -EIO;
617
618 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
619
620 /* If we're (effectively) in D3, force entire word to 0.
621 * This doesn't affect PME_Status, disables PME_En, and
622 * sets PowerState to 0.
623 */
624 switch (dev->current_state) {
625 case PCI_D0:
626 case PCI_D1:
627 case PCI_D2:
628 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
629 pmcsr |= state;
630 break;
631 case PCI_D3hot:
632 case PCI_D3cold:
633 case PCI_UNKNOWN: /* Boot-up */
634 if ((pmcsr & PCI_PM_CTRL_STATE_MASK) == PCI_D3hot
635 && !(pmcsr & PCI_PM_CTRL_NO_SOFT_RESET))
636 need_restore = true;
637 /* Fall-through: force to D0 */
638 default:
639 pmcsr = 0;
640 break;
641 }
642
643 /* enter specified state */
644 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
645
646 /* Mandatory power management transition delays */
647 /* see PCI PM 1.1 5.6.1 table 18 */
648 if (state == PCI_D3hot || dev->current_state == PCI_D3hot)
649 pci_dev_d3_sleep(dev);
650 else if (state == PCI_D2 || dev->current_state == PCI_D2)
651 udelay(PCI_PM_D2_DELAY);
652
653 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
654 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
655 if (dev->current_state != state && printk_ratelimit())
656 dev_info(&dev->dev, "Refused to change power state, currently in D%d\n",
657 dev->current_state);
658
659 /*
660 * According to section 5.4.1 of the "PCI BUS POWER MANAGEMENT
661 * INTERFACE SPECIFICATION, REV. 1.2", a device transitioning
662 * from D3hot to D0 _may_ perform an internal reset, thereby
663 * going to "D0 Uninitialized" rather than "D0 Initialized".
664 * For example, at least some versions of the 3c905B and the
665 * 3c556B exhibit this behaviour.
666 *
667 * At least some laptop BIOSen (e.g. the Thinkpad T21) leave
668 * devices in a D3hot state at boot. Consequently, we need to
669 * restore at least the BARs so that the device will be
670 * accessible to its driver.
671 */
672 if (need_restore)
673 pci_restore_bars(dev);
674
675 if (dev->bus->self)
676 pcie_aspm_pm_state_change(dev->bus->self);
677
678 return 0;
679 }
680
681 /**
682 * pci_update_current_state - Read PCI power state of given device from its
683 * PCI PM registers and cache it
684 * @dev: PCI device to handle.
685 * @state: State to cache in case the device doesn't have the PM capability
686 */
687 void pci_update_current_state(struct pci_dev *dev, pci_power_t state)
688 {
689 if (dev->pm_cap) {
690 u16 pmcsr;
691
692 /*
693 * Configuration space is not accessible for device in
694 * D3cold, so just keep or set D3cold for safety
695 */
696 if (dev->current_state == PCI_D3cold)
697 return;
698 if (state == PCI_D3cold) {
699 dev->current_state = PCI_D3cold;
700 return;
701 }
702 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
703 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
704 } else {
705 dev->current_state = state;
706 }
707 }
708
709 /**
710 * pci_power_up - Put the given device into D0 forcibly
711 * @dev: PCI device to power up
712 */
713 void pci_power_up(struct pci_dev *dev)
714 {
715 if (platform_pci_power_manageable(dev))
716 platform_pci_set_power_state(dev, PCI_D0);
717
718 pci_raw_set_power_state(dev, PCI_D0);
719 pci_update_current_state(dev, PCI_D0);
720 }
721
722 /**
723 * pci_platform_power_transition - Use platform to change device power state
724 * @dev: PCI device to handle.
725 * @state: State to put the device into.
726 */
727 static int pci_platform_power_transition(struct pci_dev *dev, pci_power_t state)
728 {
729 int error;
730
731 if (platform_pci_power_manageable(dev)) {
732 error = platform_pci_set_power_state(dev, state);
733 if (!error)
734 pci_update_current_state(dev, state);
735 } else
736 error = -ENODEV;
737
738 if (error && !dev->pm_cap) /* Fall back to PCI_D0 */
739 dev->current_state = PCI_D0;
740
741 return error;
742 }
743
744 /**
745 * pci_wakeup - Wake up a PCI device
746 * @pci_dev: Device to handle.
747 * @ign: ignored parameter
748 */
749 static int pci_wakeup(struct pci_dev *pci_dev, void *ign)
750 {
751 pci_wakeup_event(pci_dev);
752 pm_request_resume(&pci_dev->dev);
753 return 0;
754 }
755
756 /**
757 * pci_wakeup_bus - Walk given bus and wake up devices on it
758 * @bus: Top bus of the subtree to walk.
759 */
760 static void pci_wakeup_bus(struct pci_bus *bus)
761 {
762 if (bus)
763 pci_walk_bus(bus, pci_wakeup, NULL);
764 }
765
766 /**
767 * __pci_start_power_transition - Start power transition of a PCI device
768 * @dev: PCI device to handle.
769 * @state: State to put the device into.
770 */
771 static void __pci_start_power_transition(struct pci_dev *dev, pci_power_t state)
772 {
773 if (state == PCI_D0) {
774 pci_platform_power_transition(dev, PCI_D0);
775 /*
776 * Mandatory power management transition delays, see
777 * PCI Express Base Specification Revision 2.0 Section
778 * 6.6.1: Conventional Reset. Do not delay for
779 * devices powered on/off by corresponding bridge,
780 * because have already delayed for the bridge.
781 */
782 if (dev->runtime_d3cold) {
783 msleep(dev->d3cold_delay);
784 /*
785 * When powering on a bridge from D3cold, the
786 * whole hierarchy may be powered on into
787 * D0uninitialized state, resume them to give
788 * them a chance to suspend again
789 */
790 pci_wakeup_bus(dev->subordinate);
791 }
792 }
793 }
794
795 /**
796 * __pci_dev_set_current_state - Set current state of a PCI device
797 * @dev: Device to handle
798 * @data: pointer to state to be set
799 */
800 static int __pci_dev_set_current_state(struct pci_dev *dev, void *data)
801 {
802 pci_power_t state = *(pci_power_t *)data;
803
804 dev->current_state = state;
805 return 0;
806 }
807
808 /**
809 * __pci_bus_set_current_state - Walk given bus and set current state of devices
810 * @bus: Top bus of the subtree to walk.
811 * @state: state to be set
812 */
813 static void __pci_bus_set_current_state(struct pci_bus *bus, pci_power_t state)
814 {
815 if (bus)
816 pci_walk_bus(bus, __pci_dev_set_current_state, &state);
817 }
818
819 /**
820 * __pci_complete_power_transition - Complete power transition of a PCI device
821 * @dev: PCI device to handle.
822 * @state: State to put the device into.
823 *
824 * This function should not be called directly by device drivers.
825 */
826 int __pci_complete_power_transition(struct pci_dev *dev, pci_power_t state)
827 {
828 int ret;
829
830 if (state <= PCI_D0)
831 return -EINVAL;
832 ret = pci_platform_power_transition(dev, state);
833 /* Power off the bridge may power off the whole hierarchy */
834 if (!ret && state == PCI_D3cold)
835 __pci_bus_set_current_state(dev->subordinate, PCI_D3cold);
836 return ret;
837 }
838 EXPORT_SYMBOL_GPL(__pci_complete_power_transition);
839
840 /**
841 * pci_set_power_state - Set the power state of a PCI device
842 * @dev: PCI device to handle.
843 * @state: PCI power state (D0, D1, D2, D3hot) to put the device into.
844 *
845 * Transition a device to a new power state, using the platform firmware and/or
846 * the device's PCI PM registers.
847 *
848 * RETURN VALUE:
849 * -EINVAL if the requested state is invalid.
850 * -EIO if device does not support PCI PM or its PM capabilities register has a
851 * wrong version, or device doesn't support the requested state.
852 * 0 if device already is in the requested state.
853 * 0 if device's power state has been successfully changed.
854 */
855 int pci_set_power_state(struct pci_dev *dev, pci_power_t state)
856 {
857 int error;
858
859 /* bound the state we're entering */
860 if (state > PCI_D3cold)
861 state = PCI_D3cold;
862 else if (state < PCI_D0)
863 state = PCI_D0;
864 else if ((state == PCI_D1 || state == PCI_D2) && pci_no_d1d2(dev))
865 /*
866 * If the device or the parent bridge do not support PCI PM,
867 * ignore the request if we're doing anything other than putting
868 * it into D0 (which would only happen on boot).
869 */
870 return 0;
871
872 /* Check if we're already there */
873 if (dev->current_state == state)
874 return 0;
875
876 __pci_start_power_transition(dev, state);
877
878 /* This device is quirked not to be put into D3, so
879 don't put it in D3 */
880 if (state >= PCI_D3hot && (dev->dev_flags & PCI_DEV_FLAGS_NO_D3))
881 return 0;
882
883 /*
884 * To put device in D3cold, we put device into D3hot in native
885 * way, then put device into D3cold with platform ops
886 */
887 error = pci_raw_set_power_state(dev, state > PCI_D3hot ?
888 PCI_D3hot : state);
889
890 if (!__pci_complete_power_transition(dev, state))
891 error = 0;
892
893 return error;
894 }
895 EXPORT_SYMBOL(pci_set_power_state);
896
897 /**
898 * pci_choose_state - Choose the power state of a PCI device
899 * @dev: PCI device to be suspended
900 * @state: target sleep state for the whole system. This is the value
901 * that is passed to suspend() function.
902 *
903 * Returns PCI power state suitable for given device and given system
904 * message.
905 */
906
907 pci_power_t pci_choose_state(struct pci_dev *dev, pm_message_t state)
908 {
909 pci_power_t ret;
910
911 if (!dev->pm_cap)
912 return PCI_D0;
913
914 ret = platform_pci_choose_state(dev);
915 if (ret != PCI_POWER_ERROR)
916 return ret;
917
918 switch (state.event) {
919 case PM_EVENT_ON:
920 return PCI_D0;
921 case PM_EVENT_FREEZE:
922 case PM_EVENT_PRETHAW:
923 /* REVISIT both freeze and pre-thaw "should" use D0 */
924 case PM_EVENT_SUSPEND:
925 case PM_EVENT_HIBERNATE:
926 return PCI_D3hot;
927 default:
928 dev_info(&dev->dev, "unrecognized suspend event %d\n",
929 state.event);
930 BUG();
931 }
932 return PCI_D0;
933 }
934 EXPORT_SYMBOL(pci_choose_state);
935
936 #define PCI_EXP_SAVE_REGS 7
937
938 static struct pci_cap_saved_state *_pci_find_saved_cap(struct pci_dev *pci_dev,
939 u16 cap, bool extended)
940 {
941 struct pci_cap_saved_state *tmp;
942
943 hlist_for_each_entry(tmp, &pci_dev->saved_cap_space, next) {
944 if (tmp->cap.cap_extended == extended && tmp->cap.cap_nr == cap)
945 return tmp;
946 }
947 return NULL;
948 }
949
950 struct pci_cap_saved_state *pci_find_saved_cap(struct pci_dev *dev, char cap)
951 {
952 return _pci_find_saved_cap(dev, cap, false);
953 }
954
955 struct pci_cap_saved_state *pci_find_saved_ext_cap(struct pci_dev *dev, u16 cap)
956 {
957 return _pci_find_saved_cap(dev, cap, true);
958 }
959
960 static int pci_save_pcie_state(struct pci_dev *dev)
961 {
962 int i = 0;
963 struct pci_cap_saved_state *save_state;
964 u16 *cap;
965
966 if (!pci_is_pcie(dev))
967 return 0;
968
969 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
970 if (!save_state) {
971 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
972 return -ENOMEM;
973 }
974
975 cap = (u16 *)&save_state->cap.data[0];
976 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &cap[i++]);
977 pcie_capability_read_word(dev, PCI_EXP_LNKCTL, &cap[i++]);
978 pcie_capability_read_word(dev, PCI_EXP_SLTCTL, &cap[i++]);
979 pcie_capability_read_word(dev, PCI_EXP_RTCTL, &cap[i++]);
980 pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &cap[i++]);
981 pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &cap[i++]);
982 pcie_capability_read_word(dev, PCI_EXP_SLTCTL2, &cap[i++]);
983
984 return 0;
985 }
986
987 static void pci_restore_pcie_state(struct pci_dev *dev)
988 {
989 int i = 0;
990 struct pci_cap_saved_state *save_state;
991 u16 *cap;
992
993 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_EXP);
994 if (!save_state)
995 return;
996
997 cap = (u16 *)&save_state->cap.data[0];
998 pcie_capability_write_word(dev, PCI_EXP_DEVCTL, cap[i++]);
999 pcie_capability_write_word(dev, PCI_EXP_LNKCTL, cap[i++]);
1000 pcie_capability_write_word(dev, PCI_EXP_SLTCTL, cap[i++]);
1001 pcie_capability_write_word(dev, PCI_EXP_RTCTL, cap[i++]);
1002 pcie_capability_write_word(dev, PCI_EXP_DEVCTL2, cap[i++]);
1003 pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, cap[i++]);
1004 pcie_capability_write_word(dev, PCI_EXP_SLTCTL2, cap[i++]);
1005 }
1006
1007
1008 static int pci_save_pcix_state(struct pci_dev *dev)
1009 {
1010 int pos;
1011 struct pci_cap_saved_state *save_state;
1012
1013 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1014 if (!pos)
1015 return 0;
1016
1017 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1018 if (!save_state) {
1019 dev_err(&dev->dev, "buffer not found in %s\n", __func__);
1020 return -ENOMEM;
1021 }
1022
1023 pci_read_config_word(dev, pos + PCI_X_CMD,
1024 (u16 *)save_state->cap.data);
1025
1026 return 0;
1027 }
1028
1029 static void pci_restore_pcix_state(struct pci_dev *dev)
1030 {
1031 int i = 0, pos;
1032 struct pci_cap_saved_state *save_state;
1033 u16 *cap;
1034
1035 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_PCIX);
1036 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1037 if (!save_state || !pos)
1038 return;
1039 cap = (u16 *)&save_state->cap.data[0];
1040
1041 pci_write_config_word(dev, pos + PCI_X_CMD, cap[i++]);
1042 }
1043
1044
1045 /**
1046 * pci_save_state - save the PCI configuration space of a device before suspending
1047 * @dev: - PCI device that we're dealing with
1048 */
1049 int pci_save_state(struct pci_dev *dev)
1050 {
1051 int i;
1052 /* XXX: 100% dword access ok here? */
1053 for (i = 0; i < 16; i++)
1054 pci_read_config_dword(dev, i * 4, &dev->saved_config_space[i]);
1055 dev->state_saved = true;
1056
1057 i = pci_save_pcie_state(dev);
1058 if (i != 0)
1059 return i;
1060
1061 i = pci_save_pcix_state(dev);
1062 if (i != 0)
1063 return i;
1064
1065 return pci_save_vc_state(dev);
1066 }
1067 EXPORT_SYMBOL(pci_save_state);
1068
1069 static void pci_restore_config_dword(struct pci_dev *pdev, int offset,
1070 u32 saved_val, int retry)
1071 {
1072 u32 val;
1073
1074 pci_read_config_dword(pdev, offset, &val);
1075 if (val == saved_val)
1076 return;
1077
1078 for (;;) {
1079 dev_dbg(&pdev->dev, "restoring config space at offset %#x (was %#x, writing %#x)\n",
1080 offset, val, saved_val);
1081 pci_write_config_dword(pdev, offset, saved_val);
1082 if (retry-- <= 0)
1083 return;
1084
1085 pci_read_config_dword(pdev, offset, &val);
1086 if (val == saved_val)
1087 return;
1088
1089 mdelay(1);
1090 }
1091 }
1092
1093 static void pci_restore_config_space_range(struct pci_dev *pdev,
1094 int start, int end, int retry)
1095 {
1096 int index;
1097
1098 for (index = end; index >= start; index--)
1099 pci_restore_config_dword(pdev, 4 * index,
1100 pdev->saved_config_space[index],
1101 retry);
1102 }
1103
1104 static void pci_restore_config_space(struct pci_dev *pdev)
1105 {
1106 if (pdev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
1107 pci_restore_config_space_range(pdev, 10, 15, 0);
1108 /* Restore BARs before the command register. */
1109 pci_restore_config_space_range(pdev, 4, 9, 10);
1110 pci_restore_config_space_range(pdev, 0, 3, 0);
1111 } else {
1112 pci_restore_config_space_range(pdev, 0, 15, 0);
1113 }
1114 }
1115
1116 /**
1117 * pci_restore_state - Restore the saved state of a PCI device
1118 * @dev: - PCI device that we're dealing with
1119 */
1120 void pci_restore_state(struct pci_dev *dev)
1121 {
1122 if (!dev->state_saved)
1123 return;
1124
1125 /* PCI Express register must be restored first */
1126 pci_restore_pcie_state(dev);
1127 pci_restore_ats_state(dev);
1128 pci_restore_vc_state(dev);
1129
1130 pci_cleanup_aer_error_status_regs(dev);
1131
1132 pci_restore_config_space(dev);
1133
1134 pci_restore_pcix_state(dev);
1135 pci_restore_msi_state(dev);
1136
1137 /* Restore ACS and IOV configuration state */
1138 pci_enable_acs(dev);
1139 pci_restore_iov_state(dev);
1140
1141 dev->state_saved = false;
1142 }
1143 EXPORT_SYMBOL(pci_restore_state);
1144
1145 struct pci_saved_state {
1146 u32 config_space[16];
1147 struct pci_cap_saved_data cap[0];
1148 };
1149
1150 /**
1151 * pci_store_saved_state - Allocate and return an opaque struct containing
1152 * the device saved state.
1153 * @dev: PCI device that we're dealing with
1154 *
1155 * Return NULL if no state or error.
1156 */
1157 struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
1158 {
1159 struct pci_saved_state *state;
1160 struct pci_cap_saved_state *tmp;
1161 struct pci_cap_saved_data *cap;
1162 size_t size;
1163
1164 if (!dev->state_saved)
1165 return NULL;
1166
1167 size = sizeof(*state) + sizeof(struct pci_cap_saved_data);
1168
1169 hlist_for_each_entry(tmp, &dev->saved_cap_space, next)
1170 size += sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1171
1172 state = kzalloc(size, GFP_KERNEL);
1173 if (!state)
1174 return NULL;
1175
1176 memcpy(state->config_space, dev->saved_config_space,
1177 sizeof(state->config_space));
1178
1179 cap = state->cap;
1180 hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
1181 size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
1182 memcpy(cap, &tmp->cap, len);
1183 cap = (struct pci_cap_saved_data *)((u8 *)cap + len);
1184 }
1185 /* Empty cap_save terminates list */
1186
1187 return state;
1188 }
1189 EXPORT_SYMBOL_GPL(pci_store_saved_state);
1190
1191 /**
1192 * pci_load_saved_state - Reload the provided save state into struct pci_dev.
1193 * @dev: PCI device that we're dealing with
1194 * @state: Saved state returned from pci_store_saved_state()
1195 */
1196 int pci_load_saved_state(struct pci_dev *dev,
1197 struct pci_saved_state *state)
1198 {
1199 struct pci_cap_saved_data *cap;
1200
1201 dev->state_saved = false;
1202
1203 if (!state)
1204 return 0;
1205
1206 memcpy(dev->saved_config_space, state->config_space,
1207 sizeof(state->config_space));
1208
1209 cap = state->cap;
1210 while (cap->size) {
1211 struct pci_cap_saved_state *tmp;
1212
1213 tmp = _pci_find_saved_cap(dev, cap->cap_nr, cap->cap_extended);
1214 if (!tmp || tmp->cap.size != cap->size)
1215 return -EINVAL;
1216
1217 memcpy(tmp->cap.data, cap->data, tmp->cap.size);
1218 cap = (struct pci_cap_saved_data *)((u8 *)cap +
1219 sizeof(struct pci_cap_saved_data) + cap->size);
1220 }
1221
1222 dev->state_saved = true;
1223 return 0;
1224 }
1225 EXPORT_SYMBOL_GPL(pci_load_saved_state);
1226
1227 /**
1228 * pci_load_and_free_saved_state - Reload the save state pointed to by state,
1229 * and free the memory allocated for it.
1230 * @dev: PCI device that we're dealing with
1231 * @state: Pointer to saved state returned from pci_store_saved_state()
1232 */
1233 int pci_load_and_free_saved_state(struct pci_dev *dev,
1234 struct pci_saved_state **state)
1235 {
1236 int ret = pci_load_saved_state(dev, *state);
1237 kfree(*state);
1238 *state = NULL;
1239 return ret;
1240 }
1241 EXPORT_SYMBOL_GPL(pci_load_and_free_saved_state);
1242
1243 int __weak pcibios_enable_device(struct pci_dev *dev, int bars)
1244 {
1245 return pci_enable_resources(dev, bars);
1246 }
1247
1248 static int do_pci_enable_device(struct pci_dev *dev, int bars)
1249 {
1250 int err;
1251 struct pci_dev *bridge;
1252 u16 cmd;
1253 u8 pin;
1254
1255 err = pci_set_power_state(dev, PCI_D0);
1256 if (err < 0 && err != -EIO)
1257 return err;
1258
1259 bridge = pci_upstream_bridge(dev);
1260 if (bridge)
1261 pcie_aspm_powersave_config_link(bridge);
1262
1263 err = pcibios_enable_device(dev, bars);
1264 if (err < 0)
1265 return err;
1266 pci_fixup_device(pci_fixup_enable, dev);
1267
1268 if (dev->msi_enabled || dev->msix_enabled)
1269 return 0;
1270
1271 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &pin);
1272 if (pin) {
1273 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1274 if (cmd & PCI_COMMAND_INTX_DISABLE)
1275 pci_write_config_word(dev, PCI_COMMAND,
1276 cmd & ~PCI_COMMAND_INTX_DISABLE);
1277 }
1278
1279 return 0;
1280 }
1281
1282 /**
1283 * pci_reenable_device - Resume abandoned device
1284 * @dev: PCI device to be resumed
1285 *
1286 * Note this function is a backend of pci_default_resume and is not supposed
1287 * to be called by normal code, write proper resume handler and use it instead.
1288 */
1289 int pci_reenable_device(struct pci_dev *dev)
1290 {
1291 if (pci_is_enabled(dev))
1292 return do_pci_enable_device(dev, (1 << PCI_NUM_RESOURCES) - 1);
1293 return 0;
1294 }
1295 EXPORT_SYMBOL(pci_reenable_device);
1296
1297 static void pci_enable_bridge(struct pci_dev *dev)
1298 {
1299 struct pci_dev *bridge;
1300 int retval;
1301
1302 bridge = pci_upstream_bridge(dev);
1303 if (bridge)
1304 pci_enable_bridge(bridge);
1305
1306 if (pci_is_enabled(dev)) {
1307 if (!dev->is_busmaster)
1308 pci_set_master(dev);
1309 return;
1310 }
1311
1312 retval = pci_enable_device(dev);
1313 if (retval)
1314 dev_err(&dev->dev, "Error enabling bridge (%d), continuing\n",
1315 retval);
1316 pci_set_master(dev);
1317 }
1318
1319 static int pci_enable_device_flags(struct pci_dev *dev, unsigned long flags)
1320 {
1321 struct pci_dev *bridge;
1322 int err;
1323 int i, bars = 0;
1324
1325 /*
1326 * Power state could be unknown at this point, either due to a fresh
1327 * boot or a device removal call. So get the current power state
1328 * so that things like MSI message writing will behave as expected
1329 * (e.g. if the device really is in D0 at enable time).
1330 */
1331 if (dev->pm_cap) {
1332 u16 pmcsr;
1333 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1334 dev->current_state = (pmcsr & PCI_PM_CTRL_STATE_MASK);
1335 }
1336
1337 if (atomic_inc_return(&dev->enable_cnt) > 1)
1338 return 0; /* already enabled */
1339
1340 bridge = pci_upstream_bridge(dev);
1341 if (bridge)
1342 pci_enable_bridge(bridge);
1343
1344 /* only skip sriov related */
1345 for (i = 0; i <= PCI_ROM_RESOURCE; i++)
1346 if (dev->resource[i].flags & flags)
1347 bars |= (1 << i);
1348 for (i = PCI_BRIDGE_RESOURCES; i < DEVICE_COUNT_RESOURCE; i++)
1349 if (dev->resource[i].flags & flags)
1350 bars |= (1 << i);
1351
1352 err = do_pci_enable_device(dev, bars);
1353 if (err < 0)
1354 atomic_dec(&dev->enable_cnt);
1355 return err;
1356 }
1357
1358 /**
1359 * pci_enable_device_io - Initialize a device for use with IO space
1360 * @dev: PCI device to be initialized
1361 *
1362 * Initialize device before it's used by a driver. Ask low-level code
1363 * to enable I/O resources. Wake up the device if it was suspended.
1364 * Beware, this function can fail.
1365 */
1366 int pci_enable_device_io(struct pci_dev *dev)
1367 {
1368 return pci_enable_device_flags(dev, IORESOURCE_IO);
1369 }
1370 EXPORT_SYMBOL(pci_enable_device_io);
1371
1372 /**
1373 * pci_enable_device_mem - Initialize a device for use with Memory space
1374 * @dev: PCI device to be initialized
1375 *
1376 * Initialize device before it's used by a driver. Ask low-level code
1377 * to enable Memory resources. Wake up the device if it was suspended.
1378 * Beware, this function can fail.
1379 */
1380 int pci_enable_device_mem(struct pci_dev *dev)
1381 {
1382 return pci_enable_device_flags(dev, IORESOURCE_MEM);
1383 }
1384 EXPORT_SYMBOL(pci_enable_device_mem);
1385
1386 /**
1387 * pci_enable_device - Initialize device before it's used by a driver.
1388 * @dev: PCI device to be initialized
1389 *
1390 * Initialize device before it's used by a driver. Ask low-level code
1391 * to enable I/O and memory. Wake up the device if it was suspended.
1392 * Beware, this function can fail.
1393 *
1394 * Note we don't actually enable the device many times if we call
1395 * this function repeatedly (we just increment the count).
1396 */
1397 int pci_enable_device(struct pci_dev *dev)
1398 {
1399 return pci_enable_device_flags(dev, IORESOURCE_MEM | IORESOURCE_IO);
1400 }
1401 EXPORT_SYMBOL(pci_enable_device);
1402
1403 /*
1404 * Managed PCI resources. This manages device on/off, intx/msi/msix
1405 * on/off and BAR regions. pci_dev itself records msi/msix status, so
1406 * there's no need to track it separately. pci_devres is initialized
1407 * when a device is enabled using managed PCI device enable interface.
1408 */
1409 struct pci_devres {
1410 unsigned int enabled:1;
1411 unsigned int pinned:1;
1412 unsigned int orig_intx:1;
1413 unsigned int restore_intx:1;
1414 u32 region_mask;
1415 };
1416
1417 static void pcim_release(struct device *gendev, void *res)
1418 {
1419 struct pci_dev *dev = to_pci_dev(gendev);
1420 struct pci_devres *this = res;
1421 int i;
1422
1423 if (dev->msi_enabled)
1424 pci_disable_msi(dev);
1425 if (dev->msix_enabled)
1426 pci_disable_msix(dev);
1427
1428 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
1429 if (this->region_mask & (1 << i))
1430 pci_release_region(dev, i);
1431
1432 if (this->restore_intx)
1433 pci_intx(dev, this->orig_intx);
1434
1435 if (this->enabled && !this->pinned)
1436 pci_disable_device(dev);
1437 }
1438
1439 static struct pci_devres *get_pci_dr(struct pci_dev *pdev)
1440 {
1441 struct pci_devres *dr, *new_dr;
1442
1443 dr = devres_find(&pdev->dev, pcim_release, NULL, NULL);
1444 if (dr)
1445 return dr;
1446
1447 new_dr = devres_alloc(pcim_release, sizeof(*new_dr), GFP_KERNEL);
1448 if (!new_dr)
1449 return NULL;
1450 return devres_get(&pdev->dev, new_dr, NULL, NULL);
1451 }
1452
1453 static struct pci_devres *find_pci_dr(struct pci_dev *pdev)
1454 {
1455 if (pci_is_managed(pdev))
1456 return devres_find(&pdev->dev, pcim_release, NULL, NULL);
1457 return NULL;
1458 }
1459
1460 /**
1461 * pcim_enable_device - Managed pci_enable_device()
1462 * @pdev: PCI device to be initialized
1463 *
1464 * Managed pci_enable_device().
1465 */
1466 int pcim_enable_device(struct pci_dev *pdev)
1467 {
1468 struct pci_devres *dr;
1469 int rc;
1470
1471 dr = get_pci_dr(pdev);
1472 if (unlikely(!dr))
1473 return -ENOMEM;
1474 if (dr->enabled)
1475 return 0;
1476
1477 rc = pci_enable_device(pdev);
1478 if (!rc) {
1479 pdev->is_managed = 1;
1480 dr->enabled = 1;
1481 }
1482 return rc;
1483 }
1484 EXPORT_SYMBOL(pcim_enable_device);
1485
1486 /**
1487 * pcim_pin_device - Pin managed PCI device
1488 * @pdev: PCI device to pin
1489 *
1490 * Pin managed PCI device @pdev. Pinned device won't be disabled on
1491 * driver detach. @pdev must have been enabled with
1492 * pcim_enable_device().
1493 */
1494 void pcim_pin_device(struct pci_dev *pdev)
1495 {
1496 struct pci_devres *dr;
1497
1498 dr = find_pci_dr(pdev);
1499 WARN_ON(!dr || !dr->enabled);
1500 if (dr)
1501 dr->pinned = 1;
1502 }
1503 EXPORT_SYMBOL(pcim_pin_device);
1504
1505 /*
1506 * pcibios_add_device - provide arch specific hooks when adding device dev
1507 * @dev: the PCI device being added
1508 *
1509 * Permits the platform to provide architecture specific functionality when
1510 * devices are added. This is the default implementation. Architecture
1511 * implementations can override this.
1512 */
1513 int __weak pcibios_add_device(struct pci_dev *dev)
1514 {
1515 return 0;
1516 }
1517
1518 /**
1519 * pcibios_release_device - provide arch specific hooks when releasing device dev
1520 * @dev: the PCI device being released
1521 *
1522 * Permits the platform to provide architecture specific functionality when
1523 * devices are released. This is the default implementation. Architecture
1524 * implementations can override this.
1525 */
1526 void __weak pcibios_release_device(struct pci_dev *dev) {}
1527
1528 /**
1529 * pcibios_disable_device - disable arch specific PCI resources for device dev
1530 * @dev: the PCI device to disable
1531 *
1532 * Disables architecture specific PCI resources for the device. This
1533 * is the default implementation. Architecture implementations can
1534 * override this.
1535 */
1536 void __weak pcibios_disable_device(struct pci_dev *dev) {}
1537
1538 /**
1539 * pcibios_penalize_isa_irq - penalize an ISA IRQ
1540 * @irq: ISA IRQ to penalize
1541 * @active: IRQ active or not
1542 *
1543 * Permits the platform to provide architecture-specific functionality when
1544 * penalizing ISA IRQs. This is the default implementation. Architecture
1545 * implementations can override this.
1546 */
1547 void __weak pcibios_penalize_isa_irq(int irq, int active) {}
1548
1549 static void do_pci_disable_device(struct pci_dev *dev)
1550 {
1551 u16 pci_command;
1552
1553 pci_read_config_word(dev, PCI_COMMAND, &pci_command);
1554 if (pci_command & PCI_COMMAND_MASTER) {
1555 pci_command &= ~PCI_COMMAND_MASTER;
1556 pci_write_config_word(dev, PCI_COMMAND, pci_command);
1557 }
1558
1559 pcibios_disable_device(dev);
1560 }
1561
1562 /**
1563 * pci_disable_enabled_device - Disable device without updating enable_cnt
1564 * @dev: PCI device to disable
1565 *
1566 * NOTE: This function is a backend of PCI power management routines and is
1567 * not supposed to be called drivers.
1568 */
1569 void pci_disable_enabled_device(struct pci_dev *dev)
1570 {
1571 if (pci_is_enabled(dev))
1572 do_pci_disable_device(dev);
1573 }
1574
1575 /**
1576 * pci_disable_device - Disable PCI device after use
1577 * @dev: PCI device to be disabled
1578 *
1579 * Signal to the system that the PCI device is not in use by the system
1580 * anymore. This only involves disabling PCI bus-mastering, if active.
1581 *
1582 * Note we don't actually disable the device until all callers of
1583 * pci_enable_device() have called pci_disable_device().
1584 */
1585 void pci_disable_device(struct pci_dev *dev)
1586 {
1587 struct pci_devres *dr;
1588
1589 dr = find_pci_dr(dev);
1590 if (dr)
1591 dr->enabled = 0;
1592
1593 dev_WARN_ONCE(&dev->dev, atomic_read(&dev->enable_cnt) <= 0,
1594 "disabling already-disabled device");
1595
1596 if (atomic_dec_return(&dev->enable_cnt) != 0)
1597 return;
1598
1599 do_pci_disable_device(dev);
1600
1601 dev->is_busmaster = 0;
1602 }
1603 EXPORT_SYMBOL(pci_disable_device);
1604
1605 /**
1606 * pcibios_set_pcie_reset_state - set reset state for device dev
1607 * @dev: the PCIe device reset
1608 * @state: Reset state to enter into
1609 *
1610 *
1611 * Sets the PCIe reset state for the device. This is the default
1612 * implementation. Architecture implementations can override this.
1613 */
1614 int __weak pcibios_set_pcie_reset_state(struct pci_dev *dev,
1615 enum pcie_reset_state state)
1616 {
1617 return -EINVAL;
1618 }
1619
1620 /**
1621 * pci_set_pcie_reset_state - set reset state for device dev
1622 * @dev: the PCIe device reset
1623 * @state: Reset state to enter into
1624 *
1625 *
1626 * Sets the PCI reset state for the device.
1627 */
1628 int pci_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
1629 {
1630 return pcibios_set_pcie_reset_state(dev, state);
1631 }
1632 EXPORT_SYMBOL_GPL(pci_set_pcie_reset_state);
1633
1634 /**
1635 * pci_check_pme_status - Check if given device has generated PME.
1636 * @dev: Device to check.
1637 *
1638 * Check the PME status of the device and if set, clear it and clear PME enable
1639 * (if set). Return 'true' if PME status and PME enable were both set or
1640 * 'false' otherwise.
1641 */
1642 bool pci_check_pme_status(struct pci_dev *dev)
1643 {
1644 int pmcsr_pos;
1645 u16 pmcsr;
1646 bool ret = false;
1647
1648 if (!dev->pm_cap)
1649 return false;
1650
1651 pmcsr_pos = dev->pm_cap + PCI_PM_CTRL;
1652 pci_read_config_word(dev, pmcsr_pos, &pmcsr);
1653 if (!(pmcsr & PCI_PM_CTRL_PME_STATUS))
1654 return false;
1655
1656 /* Clear PME status. */
1657 pmcsr |= PCI_PM_CTRL_PME_STATUS;
1658 if (pmcsr & PCI_PM_CTRL_PME_ENABLE) {
1659 /* Disable PME to avoid interrupt flood. */
1660 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1661 ret = true;
1662 }
1663
1664 pci_write_config_word(dev, pmcsr_pos, pmcsr);
1665
1666 return ret;
1667 }
1668
1669 /**
1670 * pci_pme_wakeup - Wake up a PCI device if its PME Status bit is set.
1671 * @dev: Device to handle.
1672 * @pme_poll_reset: Whether or not to reset the device's pme_poll flag.
1673 *
1674 * Check if @dev has generated PME and queue a resume request for it in that
1675 * case.
1676 */
1677 static int pci_pme_wakeup(struct pci_dev *dev, void *pme_poll_reset)
1678 {
1679 if (pme_poll_reset && dev->pme_poll)
1680 dev->pme_poll = false;
1681
1682 if (pci_check_pme_status(dev)) {
1683 pci_wakeup_event(dev);
1684 pm_request_resume(&dev->dev);
1685 }
1686 return 0;
1687 }
1688
1689 /**
1690 * pci_pme_wakeup_bus - Walk given bus and wake up devices on it, if necessary.
1691 * @bus: Top bus of the subtree to walk.
1692 */
1693 void pci_pme_wakeup_bus(struct pci_bus *bus)
1694 {
1695 if (bus)
1696 pci_walk_bus(bus, pci_pme_wakeup, (void *)true);
1697 }
1698
1699
1700 /**
1701 * pci_pme_capable - check the capability of PCI device to generate PME#
1702 * @dev: PCI device to handle.
1703 * @state: PCI state from which device will issue PME#.
1704 */
1705 bool pci_pme_capable(struct pci_dev *dev, pci_power_t state)
1706 {
1707 if (!dev->pm_cap)
1708 return false;
1709
1710 return !!(dev->pme_support & (1 << state));
1711 }
1712 EXPORT_SYMBOL(pci_pme_capable);
1713
1714 static void pci_pme_list_scan(struct work_struct *work)
1715 {
1716 struct pci_pme_device *pme_dev, *n;
1717
1718 mutex_lock(&pci_pme_list_mutex);
1719 list_for_each_entry_safe(pme_dev, n, &pci_pme_list, list) {
1720 if (pme_dev->dev->pme_poll) {
1721 struct pci_dev *bridge;
1722
1723 bridge = pme_dev->dev->bus->self;
1724 /*
1725 * If bridge is in low power state, the
1726 * configuration space of subordinate devices
1727 * may be not accessible
1728 */
1729 if (bridge && bridge->current_state != PCI_D0)
1730 continue;
1731 pci_pme_wakeup(pme_dev->dev, NULL);
1732 } else {
1733 list_del(&pme_dev->list);
1734 kfree(pme_dev);
1735 }
1736 }
1737 if (!list_empty(&pci_pme_list))
1738 schedule_delayed_work(&pci_pme_work,
1739 msecs_to_jiffies(PME_TIMEOUT));
1740 mutex_unlock(&pci_pme_list_mutex);
1741 }
1742
1743 static void __pci_pme_active(struct pci_dev *dev, bool enable)
1744 {
1745 u16 pmcsr;
1746
1747 if (!dev->pme_support)
1748 return;
1749
1750 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &pmcsr);
1751 /* Clear PME_Status by writing 1 to it and enable PME# */
1752 pmcsr |= PCI_PM_CTRL_PME_STATUS | PCI_PM_CTRL_PME_ENABLE;
1753 if (!enable)
1754 pmcsr &= ~PCI_PM_CTRL_PME_ENABLE;
1755
1756 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, pmcsr);
1757 }
1758
1759 /**
1760 * pci_pme_active - enable or disable PCI device's PME# function
1761 * @dev: PCI device to handle.
1762 * @enable: 'true' to enable PME# generation; 'false' to disable it.
1763 *
1764 * The caller must verify that the device is capable of generating PME# before
1765 * calling this function with @enable equal to 'true'.
1766 */
1767 void pci_pme_active(struct pci_dev *dev, bool enable)
1768 {
1769 __pci_pme_active(dev, enable);
1770
1771 /*
1772 * PCI (as opposed to PCIe) PME requires that the device have
1773 * its PME# line hooked up correctly. Not all hardware vendors
1774 * do this, so the PME never gets delivered and the device
1775 * remains asleep. The easiest way around this is to
1776 * periodically walk the list of suspended devices and check
1777 * whether any have their PME flag set. The assumption is that
1778 * we'll wake up often enough anyway that this won't be a huge
1779 * hit, and the power savings from the devices will still be a
1780 * win.
1781 *
1782 * Although PCIe uses in-band PME message instead of PME# line
1783 * to report PME, PME does not work for some PCIe devices in
1784 * reality. For example, there are devices that set their PME
1785 * status bits, but don't really bother to send a PME message;
1786 * there are PCI Express Root Ports that don't bother to
1787 * trigger interrupts when they receive PME messages from the
1788 * devices below. So PME poll is used for PCIe devices too.
1789 */
1790
1791 if (dev->pme_poll) {
1792 struct pci_pme_device *pme_dev;
1793 if (enable) {
1794 pme_dev = kmalloc(sizeof(struct pci_pme_device),
1795 GFP_KERNEL);
1796 if (!pme_dev) {
1797 dev_warn(&dev->dev, "can't enable PME#\n");
1798 return;
1799 }
1800 pme_dev->dev = dev;
1801 mutex_lock(&pci_pme_list_mutex);
1802 list_add(&pme_dev->list, &pci_pme_list);
1803 if (list_is_singular(&pci_pme_list))
1804 schedule_delayed_work(&pci_pme_work,
1805 msecs_to_jiffies(PME_TIMEOUT));
1806 mutex_unlock(&pci_pme_list_mutex);
1807 } else {
1808 mutex_lock(&pci_pme_list_mutex);
1809 list_for_each_entry(pme_dev, &pci_pme_list, list) {
1810 if (pme_dev->dev == dev) {
1811 list_del(&pme_dev->list);
1812 kfree(pme_dev);
1813 break;
1814 }
1815 }
1816 mutex_unlock(&pci_pme_list_mutex);
1817 }
1818 }
1819
1820 dev_dbg(&dev->dev, "PME# %s\n", enable ? "enabled" : "disabled");
1821 }
1822 EXPORT_SYMBOL(pci_pme_active);
1823
1824 /**
1825 * __pci_enable_wake - enable PCI device as wakeup event source
1826 * @dev: PCI device affected
1827 * @state: PCI state from which device will issue wakeup events
1828 * @runtime: True if the events are to be generated at run time
1829 * @enable: True to enable event generation; false to disable
1830 *
1831 * This enables the device as a wakeup event source, or disables it.
1832 * When such events involves platform-specific hooks, those hooks are
1833 * called automatically by this routine.
1834 *
1835 * Devices with legacy power management (no standard PCI PM capabilities)
1836 * always require such platform hooks.
1837 *
1838 * RETURN VALUE:
1839 * 0 is returned on success
1840 * -EINVAL is returned if device is not supposed to wake up the system
1841 * Error code depending on the platform is returned if both the platform and
1842 * the native mechanism fail to enable the generation of wake-up events
1843 */
1844 int __pci_enable_wake(struct pci_dev *dev, pci_power_t state,
1845 bool runtime, bool enable)
1846 {
1847 int ret = 0;
1848
1849 if (enable && !runtime && !device_may_wakeup(&dev->dev))
1850 return -EINVAL;
1851
1852 /* Don't do the same thing twice in a row for one device. */
1853 if (!!enable == !!dev->wakeup_prepared)
1854 return 0;
1855
1856 /*
1857 * According to "PCI System Architecture" 4th ed. by Tom Shanley & Don
1858 * Anderson we should be doing PME# wake enable followed by ACPI wake
1859 * enable. To disable wake-up we call the platform first, for symmetry.
1860 */
1861
1862 if (enable) {
1863 int error;
1864
1865 if (pci_pme_capable(dev, state))
1866 pci_pme_active(dev, true);
1867 else
1868 ret = 1;
1869 error = runtime ? platform_pci_run_wake(dev, true) :
1870 platform_pci_sleep_wake(dev, true);
1871 if (ret)
1872 ret = error;
1873 if (!ret)
1874 dev->wakeup_prepared = true;
1875 } else {
1876 if (runtime)
1877 platform_pci_run_wake(dev, false);
1878 else
1879 platform_pci_sleep_wake(dev, false);
1880 pci_pme_active(dev, false);
1881 dev->wakeup_prepared = false;
1882 }
1883
1884 return ret;
1885 }
1886 EXPORT_SYMBOL(__pci_enable_wake);
1887
1888 /**
1889 * pci_wake_from_d3 - enable/disable device to wake up from D3_hot or D3_cold
1890 * @dev: PCI device to prepare
1891 * @enable: True to enable wake-up event generation; false to disable
1892 *
1893 * Many drivers want the device to wake up the system from D3_hot or D3_cold
1894 * and this function allows them to set that up cleanly - pci_enable_wake()
1895 * should not be called twice in a row to enable wake-up due to PCI PM vs ACPI
1896 * ordering constraints.
1897 *
1898 * This function only returns error code if the device is not capable of
1899 * generating PME# from both D3_hot and D3_cold, and the platform is unable to
1900 * enable wake-up power for it.
1901 */
1902 int pci_wake_from_d3(struct pci_dev *dev, bool enable)
1903 {
1904 return pci_pme_capable(dev, PCI_D3cold) ?
1905 pci_enable_wake(dev, PCI_D3cold, enable) :
1906 pci_enable_wake(dev, PCI_D3hot, enable);
1907 }
1908 EXPORT_SYMBOL(pci_wake_from_d3);
1909
1910 /**
1911 * pci_target_state - find an appropriate low power state for a given PCI dev
1912 * @dev: PCI device
1913 *
1914 * Use underlying platform code to find a supported low power state for @dev.
1915 * If the platform can't manage @dev, return the deepest state from which it
1916 * can generate wake events, based on any available PME info.
1917 */
1918 static pci_power_t pci_target_state(struct pci_dev *dev)
1919 {
1920 pci_power_t target_state = PCI_D3hot;
1921
1922 if (platform_pci_power_manageable(dev)) {
1923 /*
1924 * Call the platform to choose the target state of the device
1925 * and enable wake-up from this state if supported.
1926 */
1927 pci_power_t state = platform_pci_choose_state(dev);
1928
1929 switch (state) {
1930 case PCI_POWER_ERROR:
1931 case PCI_UNKNOWN:
1932 break;
1933 case PCI_D1:
1934 case PCI_D2:
1935 if (pci_no_d1d2(dev))
1936 break;
1937 default:
1938 target_state = state;
1939 }
1940 } else if (!dev->pm_cap) {
1941 target_state = PCI_D0;
1942 } else if (device_may_wakeup(&dev->dev)) {
1943 /*
1944 * Find the deepest state from which the device can generate
1945 * wake-up events, make it the target state and enable device
1946 * to generate PME#.
1947 */
1948 if (dev->pme_support) {
1949 while (target_state
1950 && !(dev->pme_support & (1 << target_state)))
1951 target_state--;
1952 }
1953 }
1954
1955 return target_state;
1956 }
1957
1958 /**
1959 * pci_prepare_to_sleep - prepare PCI device for system-wide transition into a sleep state
1960 * @dev: Device to handle.
1961 *
1962 * Choose the power state appropriate for the device depending on whether
1963 * it can wake up the system and/or is power manageable by the platform
1964 * (PCI_D3hot is the default) and put the device into that state.
1965 */
1966 int pci_prepare_to_sleep(struct pci_dev *dev)
1967 {
1968 pci_power_t target_state = pci_target_state(dev);
1969 int error;
1970
1971 if (target_state == PCI_POWER_ERROR)
1972 return -EIO;
1973
1974 pci_enable_wake(dev, target_state, device_may_wakeup(&dev->dev));
1975
1976 error = pci_set_power_state(dev, target_state);
1977
1978 if (error)
1979 pci_enable_wake(dev, target_state, false);
1980
1981 return error;
1982 }
1983 EXPORT_SYMBOL(pci_prepare_to_sleep);
1984
1985 /**
1986 * pci_back_from_sleep - turn PCI device on during system-wide transition into working state
1987 * @dev: Device to handle.
1988 *
1989 * Disable device's system wake-up capability and put it into D0.
1990 */
1991 int pci_back_from_sleep(struct pci_dev *dev)
1992 {
1993 pci_enable_wake(dev, PCI_D0, false);
1994 return pci_set_power_state(dev, PCI_D0);
1995 }
1996 EXPORT_SYMBOL(pci_back_from_sleep);
1997
1998 /**
1999 * pci_finish_runtime_suspend - Carry out PCI-specific part of runtime suspend.
2000 * @dev: PCI device being suspended.
2001 *
2002 * Prepare @dev to generate wake-up events at run time and put it into a low
2003 * power state.
2004 */
2005 int pci_finish_runtime_suspend(struct pci_dev *dev)
2006 {
2007 pci_power_t target_state = pci_target_state(dev);
2008 int error;
2009
2010 if (target_state == PCI_POWER_ERROR)
2011 return -EIO;
2012
2013 dev->runtime_d3cold = target_state == PCI_D3cold;
2014
2015 __pci_enable_wake(dev, target_state, true, pci_dev_run_wake(dev));
2016
2017 error = pci_set_power_state(dev, target_state);
2018
2019 if (error) {
2020 __pci_enable_wake(dev, target_state, true, false);
2021 dev->runtime_d3cold = false;
2022 }
2023
2024 return error;
2025 }
2026
2027 /**
2028 * pci_dev_run_wake - Check if device can generate run-time wake-up events.
2029 * @dev: Device to check.
2030 *
2031 * Return true if the device itself is capable of generating wake-up events
2032 * (through the platform or using the native PCIe PME) or if the device supports
2033 * PME and one of its upstream bridges can generate wake-up events.
2034 */
2035 bool pci_dev_run_wake(struct pci_dev *dev)
2036 {
2037 struct pci_bus *bus = dev->bus;
2038
2039 if (device_run_wake(&dev->dev))
2040 return true;
2041
2042 if (!dev->pme_support)
2043 return false;
2044
2045 while (bus->parent) {
2046 struct pci_dev *bridge = bus->self;
2047
2048 if (device_run_wake(&bridge->dev))
2049 return true;
2050
2051 bus = bus->parent;
2052 }
2053
2054 /* We have reached the root bus. */
2055 if (bus->bridge)
2056 return device_run_wake(bus->bridge);
2057
2058 return false;
2059 }
2060 EXPORT_SYMBOL_GPL(pci_dev_run_wake);
2061
2062 /**
2063 * pci_dev_keep_suspended - Check if the device can stay in the suspended state.
2064 * @pci_dev: Device to check.
2065 *
2066 * Return 'true' if the device is runtime-suspended, it doesn't have to be
2067 * reconfigured due to wakeup settings difference between system and runtime
2068 * suspend and the current power state of it is suitable for the upcoming
2069 * (system) transition.
2070 *
2071 * If the device is not configured for system wakeup, disable PME for it before
2072 * returning 'true' to prevent it from waking up the system unnecessarily.
2073 */
2074 bool pci_dev_keep_suspended(struct pci_dev *pci_dev)
2075 {
2076 struct device *dev = &pci_dev->dev;
2077
2078 if (!pm_runtime_suspended(dev)
2079 || pci_target_state(pci_dev) != pci_dev->current_state
2080 || platform_pci_need_resume(pci_dev))
2081 return false;
2082
2083 /*
2084 * At this point the device is good to go unless it's been configured
2085 * to generate PME at the runtime suspend time, but it is not supposed
2086 * to wake up the system. In that case, simply disable PME for it
2087 * (it will have to be re-enabled on exit from system resume).
2088 *
2089 * If the device's power state is D3cold and the platform check above
2090 * hasn't triggered, the device's configuration is suitable and we don't
2091 * need to manipulate it at all.
2092 */
2093 spin_lock_irq(&dev->power.lock);
2094
2095 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold &&
2096 !device_may_wakeup(dev))
2097 __pci_pme_active(pci_dev, false);
2098
2099 spin_unlock_irq(&dev->power.lock);
2100 return true;
2101 }
2102
2103 /**
2104 * pci_dev_complete_resume - Finalize resume from system sleep for a device.
2105 * @pci_dev: Device to handle.
2106 *
2107 * If the device is runtime suspended and wakeup-capable, enable PME for it as
2108 * it might have been disabled during the prepare phase of system suspend if
2109 * the device was not configured for system wakeup.
2110 */
2111 void pci_dev_complete_resume(struct pci_dev *pci_dev)
2112 {
2113 struct device *dev = &pci_dev->dev;
2114
2115 if (!pci_dev_run_wake(pci_dev))
2116 return;
2117
2118 spin_lock_irq(&dev->power.lock);
2119
2120 if (pm_runtime_suspended(dev) && pci_dev->current_state < PCI_D3cold)
2121 __pci_pme_active(pci_dev, true);
2122
2123 spin_unlock_irq(&dev->power.lock);
2124 }
2125
2126 void pci_config_pm_runtime_get(struct pci_dev *pdev)
2127 {
2128 struct device *dev = &pdev->dev;
2129 struct device *parent = dev->parent;
2130
2131 if (parent)
2132 pm_runtime_get_sync(parent);
2133 pm_runtime_get_noresume(dev);
2134 /*
2135 * pdev->current_state is set to PCI_D3cold during suspending,
2136 * so wait until suspending completes
2137 */
2138 pm_runtime_barrier(dev);
2139 /*
2140 * Only need to resume devices in D3cold, because config
2141 * registers are still accessible for devices suspended but
2142 * not in D3cold.
2143 */
2144 if (pdev->current_state == PCI_D3cold)
2145 pm_runtime_resume(dev);
2146 }
2147
2148 void pci_config_pm_runtime_put(struct pci_dev *pdev)
2149 {
2150 struct device *dev = &pdev->dev;
2151 struct device *parent = dev->parent;
2152
2153 pm_runtime_put(dev);
2154 if (parent)
2155 pm_runtime_put_sync(parent);
2156 }
2157
2158 /**
2159 * pci_pm_init - Initialize PM functions of given PCI device
2160 * @dev: PCI device to handle.
2161 */
2162 void pci_pm_init(struct pci_dev *dev)
2163 {
2164 int pm;
2165 u16 pmc;
2166
2167 pm_runtime_forbid(&dev->dev);
2168 pm_runtime_set_active(&dev->dev);
2169 pm_runtime_enable(&dev->dev);
2170 device_enable_async_suspend(&dev->dev);
2171 dev->wakeup_prepared = false;
2172
2173 dev->pm_cap = 0;
2174 dev->pme_support = 0;
2175
2176 /* find PCI PM capability in list */
2177 pm = pci_find_capability(dev, PCI_CAP_ID_PM);
2178 if (!pm)
2179 return;
2180 /* Check device's ability to generate PME# */
2181 pci_read_config_word(dev, pm + PCI_PM_PMC, &pmc);
2182
2183 if ((pmc & PCI_PM_CAP_VER_MASK) > 3) {
2184 dev_err(&dev->dev, "unsupported PM cap regs version (%u)\n",
2185 pmc & PCI_PM_CAP_VER_MASK);
2186 return;
2187 }
2188
2189 dev->pm_cap = pm;
2190 dev->d3_delay = PCI_PM_D3_WAIT;
2191 dev->d3cold_delay = PCI_PM_D3COLD_WAIT;
2192 dev->d3cold_allowed = true;
2193
2194 dev->d1_support = false;
2195 dev->d2_support = false;
2196 if (!pci_no_d1d2(dev)) {
2197 if (pmc & PCI_PM_CAP_D1)
2198 dev->d1_support = true;
2199 if (pmc & PCI_PM_CAP_D2)
2200 dev->d2_support = true;
2201
2202 if (dev->d1_support || dev->d2_support)
2203 dev_printk(KERN_DEBUG, &dev->dev, "supports%s%s\n",
2204 dev->d1_support ? " D1" : "",
2205 dev->d2_support ? " D2" : "");
2206 }
2207
2208 pmc &= PCI_PM_CAP_PME_MASK;
2209 if (pmc) {
2210 dev_printk(KERN_DEBUG, &dev->dev,
2211 "PME# supported from%s%s%s%s%s\n",
2212 (pmc & PCI_PM_CAP_PME_D0) ? " D0" : "",
2213 (pmc & PCI_PM_CAP_PME_D1) ? " D1" : "",
2214 (pmc & PCI_PM_CAP_PME_D2) ? " D2" : "",
2215 (pmc & PCI_PM_CAP_PME_D3) ? " D3hot" : "",
2216 (pmc & PCI_PM_CAP_PME_D3cold) ? " D3cold" : "");
2217 dev->pme_support = pmc >> PCI_PM_CAP_PME_SHIFT;
2218 dev->pme_poll = true;
2219 /*
2220 * Make device's PM flags reflect the wake-up capability, but
2221 * let the user space enable it to wake up the system as needed.
2222 */
2223 device_set_wakeup_capable(&dev->dev, true);
2224 /* Disable the PME# generation functionality */
2225 pci_pme_active(dev, false);
2226 }
2227 }
2228
2229 static unsigned long pci_ea_flags(struct pci_dev *dev, u8 prop)
2230 {
2231 unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_PCI_EA_BEI;
2232
2233 switch (prop) {
2234 case PCI_EA_P_MEM:
2235 case PCI_EA_P_VF_MEM:
2236 flags |= IORESOURCE_MEM;
2237 break;
2238 case PCI_EA_P_MEM_PREFETCH:
2239 case PCI_EA_P_VF_MEM_PREFETCH:
2240 flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH;
2241 break;
2242 case PCI_EA_P_IO:
2243 flags |= IORESOURCE_IO;
2244 break;
2245 default:
2246 return 0;
2247 }
2248
2249 return flags;
2250 }
2251
2252 static struct resource *pci_ea_get_resource(struct pci_dev *dev, u8 bei,
2253 u8 prop)
2254 {
2255 if (bei <= PCI_EA_BEI_BAR5 && prop <= PCI_EA_P_IO)
2256 return &dev->resource[bei];
2257 #ifdef CONFIG_PCI_IOV
2258 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5 &&
2259 (prop == PCI_EA_P_VF_MEM || prop == PCI_EA_P_VF_MEM_PREFETCH))
2260 return &dev->resource[PCI_IOV_RESOURCES +
2261 bei - PCI_EA_BEI_VF_BAR0];
2262 #endif
2263 else if (bei == PCI_EA_BEI_ROM)
2264 return &dev->resource[PCI_ROM_RESOURCE];
2265 else
2266 return NULL;
2267 }
2268
2269 /* Read an Enhanced Allocation (EA) entry */
2270 static int pci_ea_read(struct pci_dev *dev, int offset)
2271 {
2272 struct resource *res;
2273 int ent_size, ent_offset = offset;
2274 resource_size_t start, end;
2275 unsigned long flags;
2276 u32 dw0, bei, base, max_offset;
2277 u8 prop;
2278 bool support_64 = (sizeof(resource_size_t) >= 8);
2279
2280 pci_read_config_dword(dev, ent_offset, &dw0);
2281 ent_offset += 4;
2282
2283 /* Entry size field indicates DWORDs after 1st */
2284 ent_size = ((dw0 & PCI_EA_ES) + 1) << 2;
2285
2286 if (!(dw0 & PCI_EA_ENABLE)) /* Entry not enabled */
2287 goto out;
2288
2289 bei = (dw0 & PCI_EA_BEI) >> 4;
2290 prop = (dw0 & PCI_EA_PP) >> 8;
2291
2292 /*
2293 * If the Property is in the reserved range, try the Secondary
2294 * Property instead.
2295 */
2296 if (prop > PCI_EA_P_BRIDGE_IO && prop < PCI_EA_P_MEM_RESERVED)
2297 prop = (dw0 & PCI_EA_SP) >> 16;
2298 if (prop > PCI_EA_P_BRIDGE_IO)
2299 goto out;
2300
2301 res = pci_ea_get_resource(dev, bei, prop);
2302 if (!res) {
2303 dev_err(&dev->dev, "Unsupported EA entry BEI: %u\n", bei);
2304 goto out;
2305 }
2306
2307 flags = pci_ea_flags(dev, prop);
2308 if (!flags) {
2309 dev_err(&dev->dev, "Unsupported EA properties: %#x\n", prop);
2310 goto out;
2311 }
2312
2313 /* Read Base */
2314 pci_read_config_dword(dev, ent_offset, &base);
2315 start = (base & PCI_EA_FIELD_MASK);
2316 ent_offset += 4;
2317
2318 /* Read MaxOffset */
2319 pci_read_config_dword(dev, ent_offset, &max_offset);
2320 ent_offset += 4;
2321
2322 /* Read Base MSBs (if 64-bit entry) */
2323 if (base & PCI_EA_IS_64) {
2324 u32 base_upper;
2325
2326 pci_read_config_dword(dev, ent_offset, &base_upper);
2327 ent_offset += 4;
2328
2329 flags |= IORESOURCE_MEM_64;
2330
2331 /* entry starts above 32-bit boundary, can't use */
2332 if (!support_64 && base_upper)
2333 goto out;
2334
2335 if (support_64)
2336 start |= ((u64)base_upper << 32);
2337 }
2338
2339 end = start + (max_offset | 0x03);
2340
2341 /* Read MaxOffset MSBs (if 64-bit entry) */
2342 if (max_offset & PCI_EA_IS_64) {
2343 u32 max_offset_upper;
2344
2345 pci_read_config_dword(dev, ent_offset, &max_offset_upper);
2346 ent_offset += 4;
2347
2348 flags |= IORESOURCE_MEM_64;
2349
2350 /* entry too big, can't use */
2351 if (!support_64 && max_offset_upper)
2352 goto out;
2353
2354 if (support_64)
2355 end += ((u64)max_offset_upper << 32);
2356 }
2357
2358 if (end < start) {
2359 dev_err(&dev->dev, "EA Entry crosses address boundary\n");
2360 goto out;
2361 }
2362
2363 if (ent_size != ent_offset - offset) {
2364 dev_err(&dev->dev,
2365 "EA Entry Size (%d) does not match length read (%d)\n",
2366 ent_size, ent_offset - offset);
2367 goto out;
2368 }
2369
2370 res->name = pci_name(dev);
2371 res->start = start;
2372 res->end = end;
2373 res->flags = flags;
2374
2375 if (bei <= PCI_EA_BEI_BAR5)
2376 dev_printk(KERN_DEBUG, &dev->dev, "BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2377 bei, res, prop);
2378 else if (bei == PCI_EA_BEI_ROM)
2379 dev_printk(KERN_DEBUG, &dev->dev, "ROM: %pR (from Enhanced Allocation, properties %#02x)\n",
2380 res, prop);
2381 else if (bei >= PCI_EA_BEI_VF_BAR0 && bei <= PCI_EA_BEI_VF_BAR5)
2382 dev_printk(KERN_DEBUG, &dev->dev, "VF BAR %d: %pR (from Enhanced Allocation, properties %#02x)\n",
2383 bei - PCI_EA_BEI_VF_BAR0, res, prop);
2384 else
2385 dev_printk(KERN_DEBUG, &dev->dev, "BEI %d res: %pR (from Enhanced Allocation, properties %#02x)\n",
2386 bei, res, prop);
2387
2388 out:
2389 return offset + ent_size;
2390 }
2391
2392 /* Enhanced Allocation Initialization */
2393 void pci_ea_init(struct pci_dev *dev)
2394 {
2395 int ea;
2396 u8 num_ent;
2397 int offset;
2398 int i;
2399
2400 /* find PCI EA capability in list */
2401 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
2402 if (!ea)
2403 return;
2404
2405 /* determine the number of entries */
2406 pci_bus_read_config_byte(dev->bus, dev->devfn, ea + PCI_EA_NUM_ENT,
2407 &num_ent);
2408 num_ent &= PCI_EA_NUM_ENT_MASK;
2409
2410 offset = ea + PCI_EA_FIRST_ENT;
2411
2412 /* Skip DWORD 2 for type 1 functions */
2413 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
2414 offset += 4;
2415
2416 /* parse each EA entry */
2417 for (i = 0; i < num_ent; ++i)
2418 offset = pci_ea_read(dev, offset);
2419 }
2420
2421 static void pci_add_saved_cap(struct pci_dev *pci_dev,
2422 struct pci_cap_saved_state *new_cap)
2423 {
2424 hlist_add_head(&new_cap->next, &pci_dev->saved_cap_space);
2425 }
2426
2427 /**
2428 * _pci_add_cap_save_buffer - allocate buffer for saving given
2429 * capability registers
2430 * @dev: the PCI device
2431 * @cap: the capability to allocate the buffer for
2432 * @extended: Standard or Extended capability ID
2433 * @size: requested size of the buffer
2434 */
2435 static int _pci_add_cap_save_buffer(struct pci_dev *dev, u16 cap,
2436 bool extended, unsigned int size)
2437 {
2438 int pos;
2439 struct pci_cap_saved_state *save_state;
2440
2441 if (extended)
2442 pos = pci_find_ext_capability(dev, cap);
2443 else
2444 pos = pci_find_capability(dev, cap);
2445
2446 if (!pos)
2447 return 0;
2448
2449 save_state = kzalloc(sizeof(*save_state) + size, GFP_KERNEL);
2450 if (!save_state)
2451 return -ENOMEM;
2452
2453 save_state->cap.cap_nr = cap;
2454 save_state->cap.cap_extended = extended;
2455 save_state->cap.size = size;
2456 pci_add_saved_cap(dev, save_state);
2457
2458 return 0;
2459 }
2460
2461 int pci_add_cap_save_buffer(struct pci_dev *dev, char cap, unsigned int size)
2462 {
2463 return _pci_add_cap_save_buffer(dev, cap, false, size);
2464 }
2465
2466 int pci_add_ext_cap_save_buffer(struct pci_dev *dev, u16 cap, unsigned int size)
2467 {
2468 return _pci_add_cap_save_buffer(dev, cap, true, size);
2469 }
2470
2471 /**
2472 * pci_allocate_cap_save_buffers - allocate buffers for saving capabilities
2473 * @dev: the PCI device
2474 */
2475 void pci_allocate_cap_save_buffers(struct pci_dev *dev)
2476 {
2477 int error;
2478
2479 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_EXP,
2480 PCI_EXP_SAVE_REGS * sizeof(u16));
2481 if (error)
2482 dev_err(&dev->dev,
2483 "unable to preallocate PCI Express save buffer\n");
2484
2485 error = pci_add_cap_save_buffer(dev, PCI_CAP_ID_PCIX, sizeof(u16));
2486 if (error)
2487 dev_err(&dev->dev,
2488 "unable to preallocate PCI-X save buffer\n");
2489
2490 pci_allocate_vc_save_buffers(dev);
2491 }
2492
2493 void pci_free_cap_save_buffers(struct pci_dev *dev)
2494 {
2495 struct pci_cap_saved_state *tmp;
2496 struct hlist_node *n;
2497
2498 hlist_for_each_entry_safe(tmp, n, &dev->saved_cap_space, next)
2499 kfree(tmp);
2500 }
2501
2502 /**
2503 * pci_configure_ari - enable or disable ARI forwarding
2504 * @dev: the PCI device
2505 *
2506 * If @dev and its upstream bridge both support ARI, enable ARI in the
2507 * bridge. Otherwise, disable ARI in the bridge.
2508 */
2509 void pci_configure_ari(struct pci_dev *dev)
2510 {
2511 u32 cap;
2512 struct pci_dev *bridge;
2513
2514 if (pcie_ari_disabled || !pci_is_pcie(dev) || dev->devfn)
2515 return;
2516
2517 bridge = dev->bus->self;
2518 if (!bridge)
2519 return;
2520
2521 pcie_capability_read_dword(bridge, PCI_EXP_DEVCAP2, &cap);
2522 if (!(cap & PCI_EXP_DEVCAP2_ARI))
2523 return;
2524
2525 if (pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI)) {
2526 pcie_capability_set_word(bridge, PCI_EXP_DEVCTL2,
2527 PCI_EXP_DEVCTL2_ARI);
2528 bridge->ari_enabled = 1;
2529 } else {
2530 pcie_capability_clear_word(bridge, PCI_EXP_DEVCTL2,
2531 PCI_EXP_DEVCTL2_ARI);
2532 bridge->ari_enabled = 0;
2533 }
2534 }
2535
2536 static int pci_acs_enable;
2537
2538 /**
2539 * pci_request_acs - ask for ACS to be enabled if supported
2540 */
2541 void pci_request_acs(void)
2542 {
2543 pci_acs_enable = 1;
2544 }
2545
2546 /**
2547 * pci_std_enable_acs - enable ACS on devices using standard ACS capabilites
2548 * @dev: the PCI device
2549 */
2550 static void pci_std_enable_acs(struct pci_dev *dev)
2551 {
2552 int pos;
2553 u16 cap;
2554 u16 ctrl;
2555
2556 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ACS);
2557 if (!pos)
2558 return;
2559
2560 pci_read_config_word(dev, pos + PCI_ACS_CAP, &cap);
2561 pci_read_config_word(dev, pos + PCI_ACS_CTRL, &ctrl);
2562
2563 /* Source Validation */
2564 ctrl |= (cap & PCI_ACS_SV);
2565
2566 /* P2P Request Redirect */
2567 ctrl |= (cap & PCI_ACS_RR);
2568
2569 /* P2P Completion Redirect */
2570 ctrl |= (cap & PCI_ACS_CR);
2571
2572 /* Upstream Forwarding */
2573 ctrl |= (cap & PCI_ACS_UF);
2574
2575 pci_write_config_word(dev, pos + PCI_ACS_CTRL, ctrl);
2576 }
2577
2578 /**
2579 * pci_enable_acs - enable ACS if hardware support it
2580 * @dev: the PCI device
2581 */
2582 void pci_enable_acs(struct pci_dev *dev)
2583 {
2584 if (!pci_acs_enable)
2585 return;
2586
2587 if (!pci_dev_specific_enable_acs(dev))
2588 return;
2589
2590 pci_std_enable_acs(dev);
2591 }
2592
2593 static bool pci_acs_flags_enabled(struct pci_dev *pdev, u16 acs_flags)
2594 {
2595 int pos;
2596 u16 cap, ctrl;
2597
2598 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS);
2599 if (!pos)
2600 return false;
2601
2602 /*
2603 * Except for egress control, capabilities are either required
2604 * or only required if controllable. Features missing from the
2605 * capability field can therefore be assumed as hard-wired enabled.
2606 */
2607 pci_read_config_word(pdev, pos + PCI_ACS_CAP, &cap);
2608 acs_flags &= (cap | PCI_ACS_EC);
2609
2610 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
2611 return (ctrl & acs_flags) == acs_flags;
2612 }
2613
2614 /**
2615 * pci_acs_enabled - test ACS against required flags for a given device
2616 * @pdev: device to test
2617 * @acs_flags: required PCI ACS flags
2618 *
2619 * Return true if the device supports the provided flags. Automatically
2620 * filters out flags that are not implemented on multifunction devices.
2621 *
2622 * Note that this interface checks the effective ACS capabilities of the
2623 * device rather than the actual capabilities. For instance, most single
2624 * function endpoints are not required to support ACS because they have no
2625 * opportunity for peer-to-peer access. We therefore return 'true'
2626 * regardless of whether the device exposes an ACS capability. This makes
2627 * it much easier for callers of this function to ignore the actual type
2628 * or topology of the device when testing ACS support.
2629 */
2630 bool pci_acs_enabled(struct pci_dev *pdev, u16 acs_flags)
2631 {
2632 int ret;
2633
2634 ret = pci_dev_specific_acs_enabled(pdev, acs_flags);
2635 if (ret >= 0)
2636 return ret > 0;
2637
2638 /*
2639 * Conventional PCI and PCI-X devices never support ACS, either
2640 * effectively or actually. The shared bus topology implies that
2641 * any device on the bus can receive or snoop DMA.
2642 */
2643 if (!pci_is_pcie(pdev))
2644 return false;
2645
2646 switch (pci_pcie_type(pdev)) {
2647 /*
2648 * PCI/X-to-PCIe bridges are not specifically mentioned by the spec,
2649 * but since their primary interface is PCI/X, we conservatively
2650 * handle them as we would a non-PCIe device.
2651 */
2652 case PCI_EXP_TYPE_PCIE_BRIDGE:
2653 /*
2654 * PCIe 3.0, 6.12.1 excludes ACS on these devices. "ACS is never
2655 * applicable... must never implement an ACS Extended Capability...".
2656 * This seems arbitrary, but we take a conservative interpretation
2657 * of this statement.
2658 */
2659 case PCI_EXP_TYPE_PCI_BRIDGE:
2660 case PCI_EXP_TYPE_RC_EC:
2661 return false;
2662 /*
2663 * PCIe 3.0, 6.12.1.1 specifies that downstream and root ports should
2664 * implement ACS in order to indicate their peer-to-peer capabilities,
2665 * regardless of whether they are single- or multi-function devices.
2666 */
2667 case PCI_EXP_TYPE_DOWNSTREAM:
2668 case PCI_EXP_TYPE_ROOT_PORT:
2669 return pci_acs_flags_enabled(pdev, acs_flags);
2670 /*
2671 * PCIe 3.0, 6.12.1.2 specifies ACS capabilities that should be
2672 * implemented by the remaining PCIe types to indicate peer-to-peer
2673 * capabilities, but only when they are part of a multifunction
2674 * device. The footnote for section 6.12 indicates the specific
2675 * PCIe types included here.
2676 */
2677 case PCI_EXP_TYPE_ENDPOINT:
2678 case PCI_EXP_TYPE_UPSTREAM:
2679 case PCI_EXP_TYPE_LEG_END:
2680 case PCI_EXP_TYPE_RC_END:
2681 if (!pdev->multifunction)
2682 break;
2683
2684 return pci_acs_flags_enabled(pdev, acs_flags);
2685 }
2686
2687 /*
2688 * PCIe 3.0, 6.12.1.3 specifies no ACS capabilities are applicable
2689 * to single function devices with the exception of downstream ports.
2690 */
2691 return true;
2692 }
2693
2694 /**
2695 * pci_acs_path_enable - test ACS flags from start to end in a hierarchy
2696 * @start: starting downstream device
2697 * @end: ending upstream device or NULL to search to the root bus
2698 * @acs_flags: required flags
2699 *
2700 * Walk up a device tree from start to end testing PCI ACS support. If
2701 * any step along the way does not support the required flags, return false.
2702 */
2703 bool pci_acs_path_enabled(struct pci_dev *start,
2704 struct pci_dev *end, u16 acs_flags)
2705 {
2706 struct pci_dev *pdev, *parent = start;
2707
2708 do {
2709 pdev = parent;
2710
2711 if (!pci_acs_enabled(pdev, acs_flags))
2712 return false;
2713
2714 if (pci_is_root_bus(pdev->bus))
2715 return (end == NULL);
2716
2717 parent = pdev->bus->self;
2718 } while (pdev != end);
2719
2720 return true;
2721 }
2722
2723 /**
2724 * pci_swizzle_interrupt_pin - swizzle INTx for device behind bridge
2725 * @dev: the PCI device
2726 * @pin: the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD)
2727 *
2728 * Perform INTx swizzling for a device behind one level of bridge. This is
2729 * required by section 9.1 of the PCI-to-PCI bridge specification for devices
2730 * behind bridges on add-in cards. For devices with ARI enabled, the slot
2731 * number is always 0 (see the Implementation Note in section 2.2.8.1 of
2732 * the PCI Express Base Specification, Revision 2.1)
2733 */
2734 u8 pci_swizzle_interrupt_pin(const struct pci_dev *dev, u8 pin)
2735 {
2736 int slot;
2737
2738 if (pci_ari_enabled(dev->bus))
2739 slot = 0;
2740 else
2741 slot = PCI_SLOT(dev->devfn);
2742
2743 return (((pin - 1) + slot) % 4) + 1;
2744 }
2745
2746 int pci_get_interrupt_pin(struct pci_dev *dev, struct pci_dev **bridge)
2747 {
2748 u8 pin;
2749
2750 pin = dev->pin;
2751 if (!pin)
2752 return -1;
2753
2754 while (!pci_is_root_bus(dev->bus)) {
2755 pin = pci_swizzle_interrupt_pin(dev, pin);
2756 dev = dev->bus->self;
2757 }
2758 *bridge = dev;
2759 return pin;
2760 }
2761
2762 /**
2763 * pci_common_swizzle - swizzle INTx all the way to root bridge
2764 * @dev: the PCI device
2765 * @pinp: pointer to the INTx pin value (1=INTA, 2=INTB, 3=INTD, 4=INTD)
2766 *
2767 * Perform INTx swizzling for a device. This traverses through all PCI-to-PCI
2768 * bridges all the way up to a PCI root bus.
2769 */
2770 u8 pci_common_swizzle(struct pci_dev *dev, u8 *pinp)
2771 {
2772 u8 pin = *pinp;
2773
2774 while (!pci_is_root_bus(dev->bus)) {
2775 pin = pci_swizzle_interrupt_pin(dev, pin);
2776 dev = dev->bus->self;
2777 }
2778 *pinp = pin;
2779 return PCI_SLOT(dev->devfn);
2780 }
2781 EXPORT_SYMBOL_GPL(pci_common_swizzle);
2782
2783 /**
2784 * pci_release_region - Release a PCI bar
2785 * @pdev: PCI device whose resources were previously reserved by pci_request_region
2786 * @bar: BAR to release
2787 *
2788 * Releases the PCI I/O and memory resources previously reserved by a
2789 * successful call to pci_request_region. Call this function only
2790 * after all use of the PCI regions has ceased.
2791 */
2792 void pci_release_region(struct pci_dev *pdev, int bar)
2793 {
2794 struct pci_devres *dr;
2795
2796 if (pci_resource_len(pdev, bar) == 0)
2797 return;
2798 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
2799 release_region(pci_resource_start(pdev, bar),
2800 pci_resource_len(pdev, bar));
2801 else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
2802 release_mem_region(pci_resource_start(pdev, bar),
2803 pci_resource_len(pdev, bar));
2804
2805 dr = find_pci_dr(pdev);
2806 if (dr)
2807 dr->region_mask &= ~(1 << bar);
2808 }
2809 EXPORT_SYMBOL(pci_release_region);
2810
2811 /**
2812 * __pci_request_region - Reserved PCI I/O and memory resource
2813 * @pdev: PCI device whose resources are to be reserved
2814 * @bar: BAR to be reserved
2815 * @res_name: Name to be associated with resource.
2816 * @exclusive: whether the region access is exclusive or not
2817 *
2818 * Mark the PCI region associated with PCI device @pdev BR @bar as
2819 * being reserved by owner @res_name. Do not access any
2820 * address inside the PCI regions unless this call returns
2821 * successfully.
2822 *
2823 * If @exclusive is set, then the region is marked so that userspace
2824 * is explicitly not allowed to map the resource via /dev/mem or
2825 * sysfs MMIO access.
2826 *
2827 * Returns 0 on success, or %EBUSY on error. A warning
2828 * message is also printed on failure.
2829 */
2830 static int __pci_request_region(struct pci_dev *pdev, int bar,
2831 const char *res_name, int exclusive)
2832 {
2833 struct pci_devres *dr;
2834
2835 if (pci_resource_len(pdev, bar) == 0)
2836 return 0;
2837
2838 if (pci_resource_flags(pdev, bar) & IORESOURCE_IO) {
2839 if (!request_region(pci_resource_start(pdev, bar),
2840 pci_resource_len(pdev, bar), res_name))
2841 goto err_out;
2842 } else if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM) {
2843 if (!__request_mem_region(pci_resource_start(pdev, bar),
2844 pci_resource_len(pdev, bar), res_name,
2845 exclusive))
2846 goto err_out;
2847 }
2848
2849 dr = find_pci_dr(pdev);
2850 if (dr)
2851 dr->region_mask |= 1 << bar;
2852
2853 return 0;
2854
2855 err_out:
2856 dev_warn(&pdev->dev, "BAR %d: can't reserve %pR\n", bar,
2857 &pdev->resource[bar]);
2858 return -EBUSY;
2859 }
2860
2861 /**
2862 * pci_request_region - Reserve PCI I/O and memory resource
2863 * @pdev: PCI device whose resources are to be reserved
2864 * @bar: BAR to be reserved
2865 * @res_name: Name to be associated with resource
2866 *
2867 * Mark the PCI region associated with PCI device @pdev BAR @bar as
2868 * being reserved by owner @res_name. Do not access any
2869 * address inside the PCI regions unless this call returns
2870 * successfully.
2871 *
2872 * Returns 0 on success, or %EBUSY on error. A warning
2873 * message is also printed on failure.
2874 */
2875 int pci_request_region(struct pci_dev *pdev, int bar, const char *res_name)
2876 {
2877 return __pci_request_region(pdev, bar, res_name, 0);
2878 }
2879 EXPORT_SYMBOL(pci_request_region);
2880
2881 /**
2882 * pci_request_region_exclusive - Reserved PCI I/O and memory resource
2883 * @pdev: PCI device whose resources are to be reserved
2884 * @bar: BAR to be reserved
2885 * @res_name: Name to be associated with resource.
2886 *
2887 * Mark the PCI region associated with PCI device @pdev BR @bar as
2888 * being reserved by owner @res_name. Do not access any
2889 * address inside the PCI regions unless this call returns
2890 * successfully.
2891 *
2892 * Returns 0 on success, or %EBUSY on error. A warning
2893 * message is also printed on failure.
2894 *
2895 * The key difference that _exclusive makes it that userspace is
2896 * explicitly not allowed to map the resource via /dev/mem or
2897 * sysfs.
2898 */
2899 int pci_request_region_exclusive(struct pci_dev *pdev, int bar,
2900 const char *res_name)
2901 {
2902 return __pci_request_region(pdev, bar, res_name, IORESOURCE_EXCLUSIVE);
2903 }
2904 EXPORT_SYMBOL(pci_request_region_exclusive);
2905
2906 /**
2907 * pci_release_selected_regions - Release selected PCI I/O and memory resources
2908 * @pdev: PCI device whose resources were previously reserved
2909 * @bars: Bitmask of BARs to be released
2910 *
2911 * Release selected PCI I/O and memory resources previously reserved.
2912 * Call this function only after all use of the PCI regions has ceased.
2913 */
2914 void pci_release_selected_regions(struct pci_dev *pdev, int bars)
2915 {
2916 int i;
2917
2918 for (i = 0; i < 6; i++)
2919 if (bars & (1 << i))
2920 pci_release_region(pdev, i);
2921 }
2922 EXPORT_SYMBOL(pci_release_selected_regions);
2923
2924 static int __pci_request_selected_regions(struct pci_dev *pdev, int bars,
2925 const char *res_name, int excl)
2926 {
2927 int i;
2928
2929 for (i = 0; i < 6; i++)
2930 if (bars & (1 << i))
2931 if (__pci_request_region(pdev, i, res_name, excl))
2932 goto err_out;
2933 return 0;
2934
2935 err_out:
2936 while (--i >= 0)
2937 if (bars & (1 << i))
2938 pci_release_region(pdev, i);
2939
2940 return -EBUSY;
2941 }
2942
2943
2944 /**
2945 * pci_request_selected_regions - Reserve selected PCI I/O and memory resources
2946 * @pdev: PCI device whose resources are to be reserved
2947 * @bars: Bitmask of BARs to be requested
2948 * @res_name: Name to be associated with resource
2949 */
2950 int pci_request_selected_regions(struct pci_dev *pdev, int bars,
2951 const char *res_name)
2952 {
2953 return __pci_request_selected_regions(pdev, bars, res_name, 0);
2954 }
2955 EXPORT_SYMBOL(pci_request_selected_regions);
2956
2957 int pci_request_selected_regions_exclusive(struct pci_dev *pdev, int bars,
2958 const char *res_name)
2959 {
2960 return __pci_request_selected_regions(pdev, bars, res_name,
2961 IORESOURCE_EXCLUSIVE);
2962 }
2963 EXPORT_SYMBOL(pci_request_selected_regions_exclusive);
2964
2965 /**
2966 * pci_release_regions - Release reserved PCI I/O and memory resources
2967 * @pdev: PCI device whose resources were previously reserved by pci_request_regions
2968 *
2969 * Releases all PCI I/O and memory resources previously reserved by a
2970 * successful call to pci_request_regions. Call this function only
2971 * after all use of the PCI regions has ceased.
2972 */
2973
2974 void pci_release_regions(struct pci_dev *pdev)
2975 {
2976 pci_release_selected_regions(pdev, (1 << 6) - 1);
2977 }
2978 EXPORT_SYMBOL(pci_release_regions);
2979
2980 /**
2981 * pci_request_regions - Reserved PCI I/O and memory resources
2982 * @pdev: PCI device whose resources are to be reserved
2983 * @res_name: Name to be associated with resource.
2984 *
2985 * Mark all PCI regions associated with PCI device @pdev as
2986 * being reserved by owner @res_name. Do not access any
2987 * address inside the PCI regions unless this call returns
2988 * successfully.
2989 *
2990 * Returns 0 on success, or %EBUSY on error. A warning
2991 * message is also printed on failure.
2992 */
2993 int pci_request_regions(struct pci_dev *pdev, const char *res_name)
2994 {
2995 return pci_request_selected_regions(pdev, ((1 << 6) - 1), res_name);
2996 }
2997 EXPORT_SYMBOL(pci_request_regions);
2998
2999 /**
3000 * pci_request_regions_exclusive - Reserved PCI I/O and memory resources
3001 * @pdev: PCI device whose resources are to be reserved
3002 * @res_name: Name to be associated with resource.
3003 *
3004 * Mark all PCI regions associated with PCI device @pdev as
3005 * being reserved by owner @res_name. Do not access any
3006 * address inside the PCI regions unless this call returns
3007 * successfully.
3008 *
3009 * pci_request_regions_exclusive() will mark the region so that
3010 * /dev/mem and the sysfs MMIO access will not be allowed.
3011 *
3012 * Returns 0 on success, or %EBUSY on error. A warning
3013 * message is also printed on failure.
3014 */
3015 int pci_request_regions_exclusive(struct pci_dev *pdev, const char *res_name)
3016 {
3017 return pci_request_selected_regions_exclusive(pdev,
3018 ((1 << 6) - 1), res_name);
3019 }
3020 EXPORT_SYMBOL(pci_request_regions_exclusive);
3021
3022 #ifdef PCI_IOBASE
3023 struct io_range {
3024 struct list_head list;
3025 phys_addr_t start;
3026 resource_size_t size;
3027 };
3028
3029 static LIST_HEAD(io_range_list);
3030 static DEFINE_SPINLOCK(io_range_lock);
3031 #endif
3032
3033 /*
3034 * Record the PCI IO range (expressed as CPU physical address + size).
3035 * Return a negative value if an error has occured, zero otherwise
3036 */
3037 int __weak pci_register_io_range(phys_addr_t addr, resource_size_t size)
3038 {
3039 int err = 0;
3040
3041 #ifdef PCI_IOBASE
3042 struct io_range *range;
3043 resource_size_t allocated_size = 0;
3044
3045 /* check if the range hasn't been previously recorded */
3046 spin_lock(&io_range_lock);
3047 list_for_each_entry(range, &io_range_list, list) {
3048 if (addr >= range->start && addr + size <= range->start + size) {
3049 /* range already registered, bail out */
3050 goto end_register;
3051 }
3052 allocated_size += range->size;
3053 }
3054
3055 /* range not registed yet, check for available space */
3056 if (allocated_size + size - 1 > IO_SPACE_LIMIT) {
3057 /* if it's too big check if 64K space can be reserved */
3058 if (allocated_size + SZ_64K - 1 > IO_SPACE_LIMIT) {
3059 err = -E2BIG;
3060 goto end_register;
3061 }
3062
3063 size = SZ_64K;
3064 pr_warn("Requested IO range too big, new size set to 64K\n");
3065 }
3066
3067 /* add the range to the list */
3068 range = kzalloc(sizeof(*range), GFP_ATOMIC);
3069 if (!range) {
3070 err = -ENOMEM;
3071 goto end_register;
3072 }
3073
3074 range->start = addr;
3075 range->size = size;
3076
3077 list_add_tail(&range->list, &io_range_list);
3078
3079 end_register:
3080 spin_unlock(&io_range_lock);
3081 #endif
3082
3083 return err;
3084 }
3085
3086 phys_addr_t pci_pio_to_address(unsigned long pio)
3087 {
3088 phys_addr_t address = (phys_addr_t)OF_BAD_ADDR;
3089
3090 #ifdef PCI_IOBASE
3091 struct io_range *range;
3092 resource_size_t allocated_size = 0;
3093
3094 if (pio > IO_SPACE_LIMIT)
3095 return address;
3096
3097 spin_lock(&io_range_lock);
3098 list_for_each_entry(range, &io_range_list, list) {
3099 if (pio >= allocated_size && pio < allocated_size + range->size) {
3100 address = range->start + pio - allocated_size;
3101 break;
3102 }
3103 allocated_size += range->size;
3104 }
3105 spin_unlock(&io_range_lock);
3106 #endif
3107
3108 return address;
3109 }
3110
3111 unsigned long __weak pci_address_to_pio(phys_addr_t address)
3112 {
3113 #ifdef PCI_IOBASE
3114 struct io_range *res;
3115 resource_size_t offset = 0;
3116 unsigned long addr = -1;
3117
3118 spin_lock(&io_range_lock);
3119 list_for_each_entry(res, &io_range_list, list) {
3120 if (address >= res->start && address < res->start + res->size) {
3121 addr = address - res->start + offset;
3122 break;
3123 }
3124 offset += res->size;
3125 }
3126 spin_unlock(&io_range_lock);
3127
3128 return addr;
3129 #else
3130 if (address > IO_SPACE_LIMIT)
3131 return (unsigned long)-1;
3132
3133 return (unsigned long) address;
3134 #endif
3135 }
3136
3137 /**
3138 * pci_remap_iospace - Remap the memory mapped I/O space
3139 * @res: Resource describing the I/O space
3140 * @phys_addr: physical address of range to be mapped
3141 *
3142 * Remap the memory mapped I/O space described by the @res
3143 * and the CPU physical address @phys_addr into virtual address space.
3144 * Only architectures that have memory mapped IO functions defined
3145 * (and the PCI_IOBASE value defined) should call this function.
3146 */
3147 int __weak pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
3148 {
3149 #if defined(PCI_IOBASE) && defined(CONFIG_MMU)
3150 unsigned long vaddr = (unsigned long)PCI_IOBASE + res->start;
3151
3152 if (!(res->flags & IORESOURCE_IO))
3153 return -EINVAL;
3154
3155 if (res->end > IO_SPACE_LIMIT)
3156 return -EINVAL;
3157
3158 return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
3159 pgprot_device(PAGE_KERNEL));
3160 #else
3161 /* this architecture does not have memory mapped I/O space,
3162 so this function should never be called */
3163 WARN_ONCE(1, "This architecture does not support memory mapped I/O\n");
3164 return -ENODEV;
3165 #endif
3166 }
3167
3168 static void __pci_set_master(struct pci_dev *dev, bool enable)
3169 {
3170 u16 old_cmd, cmd;
3171
3172 pci_read_config_word(dev, PCI_COMMAND, &old_cmd);
3173 if (enable)
3174 cmd = old_cmd | PCI_COMMAND_MASTER;
3175 else
3176 cmd = old_cmd & ~PCI_COMMAND_MASTER;
3177 if (cmd != old_cmd) {
3178 dev_dbg(&dev->dev, "%s bus mastering\n",
3179 enable ? "enabling" : "disabling");
3180 pci_write_config_word(dev, PCI_COMMAND, cmd);
3181 }
3182 dev->is_busmaster = enable;
3183 }
3184
3185 /**
3186 * pcibios_setup - process "pci=" kernel boot arguments
3187 * @str: string used to pass in "pci=" kernel boot arguments
3188 *
3189 * Process kernel boot arguments. This is the default implementation.
3190 * Architecture specific implementations can override this as necessary.
3191 */
3192 char * __weak __init pcibios_setup(char *str)
3193 {
3194 return str;
3195 }
3196
3197 /**
3198 * pcibios_set_master - enable PCI bus-mastering for device dev
3199 * @dev: the PCI device to enable
3200 *
3201 * Enables PCI bus-mastering for the device. This is the default
3202 * implementation. Architecture specific implementations can override
3203 * this if necessary.
3204 */
3205 void __weak pcibios_set_master(struct pci_dev *dev)
3206 {
3207 u8 lat;
3208
3209 /* The latency timer doesn't apply to PCIe (either Type 0 or Type 1) */
3210 if (pci_is_pcie(dev))
3211 return;
3212
3213 pci_read_config_byte(dev, PCI_LATENCY_TIMER, &lat);
3214 if (lat < 16)
3215 lat = (64 <= pcibios_max_latency) ? 64 : pcibios_max_latency;
3216 else if (lat > pcibios_max_latency)
3217 lat = pcibios_max_latency;
3218 else
3219 return;
3220
3221 pci_write_config_byte(dev, PCI_LATENCY_TIMER, lat);
3222 }
3223
3224 /**
3225 * pci_set_master - enables bus-mastering for device dev
3226 * @dev: the PCI device to enable
3227 *
3228 * Enables bus-mastering on the device and calls pcibios_set_master()
3229 * to do the needed arch specific settings.
3230 */
3231 void pci_set_master(struct pci_dev *dev)
3232 {
3233 __pci_set_master(dev, true);
3234 pcibios_set_master(dev);
3235 }
3236 EXPORT_SYMBOL(pci_set_master);
3237
3238 /**
3239 * pci_clear_master - disables bus-mastering for device dev
3240 * @dev: the PCI device to disable
3241 */
3242 void pci_clear_master(struct pci_dev *dev)
3243 {
3244 __pci_set_master(dev, false);
3245 }
3246 EXPORT_SYMBOL(pci_clear_master);
3247
3248 /**
3249 * pci_set_cacheline_size - ensure the CACHE_LINE_SIZE register is programmed
3250 * @dev: the PCI device for which MWI is to be enabled
3251 *
3252 * Helper function for pci_set_mwi.
3253 * Originally copied from drivers/net/acenic.c.
3254 * Copyright 1998-2001 by Jes Sorensen, <jes@trained-monkey.org>.
3255 *
3256 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3257 */
3258 int pci_set_cacheline_size(struct pci_dev *dev)
3259 {
3260 u8 cacheline_size;
3261
3262 if (!pci_cache_line_size)
3263 return -EINVAL;
3264
3265 /* Validate current setting: the PCI_CACHE_LINE_SIZE must be
3266 equal to or multiple of the right value. */
3267 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3268 if (cacheline_size >= pci_cache_line_size &&
3269 (cacheline_size % pci_cache_line_size) == 0)
3270 return 0;
3271
3272 /* Write the correct value. */
3273 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
3274 /* Read it back. */
3275 pci_read_config_byte(dev, PCI_CACHE_LINE_SIZE, &cacheline_size);
3276 if (cacheline_size == pci_cache_line_size)
3277 return 0;
3278
3279 dev_printk(KERN_DEBUG, &dev->dev, "cache line size of %d is not supported\n",
3280 pci_cache_line_size << 2);
3281
3282 return -EINVAL;
3283 }
3284 EXPORT_SYMBOL_GPL(pci_set_cacheline_size);
3285
3286 /**
3287 * pci_set_mwi - enables memory-write-invalidate PCI transaction
3288 * @dev: the PCI device for which MWI is enabled
3289 *
3290 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3291 *
3292 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3293 */
3294 int pci_set_mwi(struct pci_dev *dev)
3295 {
3296 #ifdef PCI_DISABLE_MWI
3297 return 0;
3298 #else
3299 int rc;
3300 u16 cmd;
3301
3302 rc = pci_set_cacheline_size(dev);
3303 if (rc)
3304 return rc;
3305
3306 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3307 if (!(cmd & PCI_COMMAND_INVALIDATE)) {
3308 dev_dbg(&dev->dev, "enabling Mem-Wr-Inval\n");
3309 cmd |= PCI_COMMAND_INVALIDATE;
3310 pci_write_config_word(dev, PCI_COMMAND, cmd);
3311 }
3312 return 0;
3313 #endif
3314 }
3315 EXPORT_SYMBOL(pci_set_mwi);
3316
3317 /**
3318 * pci_try_set_mwi - enables memory-write-invalidate PCI transaction
3319 * @dev: the PCI device for which MWI is enabled
3320 *
3321 * Enables the Memory-Write-Invalidate transaction in %PCI_COMMAND.
3322 * Callers are not required to check the return value.
3323 *
3324 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
3325 */
3326 int pci_try_set_mwi(struct pci_dev *dev)
3327 {
3328 #ifdef PCI_DISABLE_MWI
3329 return 0;
3330 #else
3331 return pci_set_mwi(dev);
3332 #endif
3333 }
3334 EXPORT_SYMBOL(pci_try_set_mwi);
3335
3336 /**
3337 * pci_clear_mwi - disables Memory-Write-Invalidate for device dev
3338 * @dev: the PCI device to disable
3339 *
3340 * Disables PCI Memory-Write-Invalidate transaction on the device
3341 */
3342 void pci_clear_mwi(struct pci_dev *dev)
3343 {
3344 #ifndef PCI_DISABLE_MWI
3345 u16 cmd;
3346
3347 pci_read_config_word(dev, PCI_COMMAND, &cmd);
3348 if (cmd & PCI_COMMAND_INVALIDATE) {
3349 cmd &= ~PCI_COMMAND_INVALIDATE;
3350 pci_write_config_word(dev, PCI_COMMAND, cmd);
3351 }
3352 #endif
3353 }
3354 EXPORT_SYMBOL(pci_clear_mwi);
3355
3356 /**
3357 * pci_intx - enables/disables PCI INTx for device dev
3358 * @pdev: the PCI device to operate on
3359 * @enable: boolean: whether to enable or disable PCI INTx
3360 *
3361 * Enables/disables PCI INTx for device dev
3362 */
3363 void pci_intx(struct pci_dev *pdev, int enable)
3364 {
3365 u16 pci_command, new;
3366
3367 pci_read_config_word(pdev, PCI_COMMAND, &pci_command);
3368
3369 if (enable)
3370 new = pci_command & ~PCI_COMMAND_INTX_DISABLE;
3371 else
3372 new = pci_command | PCI_COMMAND_INTX_DISABLE;
3373
3374 if (new != pci_command) {
3375 struct pci_devres *dr;
3376
3377 pci_write_config_word(pdev, PCI_COMMAND, new);
3378
3379 dr = find_pci_dr(pdev);
3380 if (dr && !dr->restore_intx) {
3381 dr->restore_intx = 1;
3382 dr->orig_intx = !enable;
3383 }
3384 }
3385 }
3386 EXPORT_SYMBOL_GPL(pci_intx);
3387
3388 /**
3389 * pci_intx_mask_supported - probe for INTx masking support
3390 * @dev: the PCI device to operate on
3391 *
3392 * Check if the device dev support INTx masking via the config space
3393 * command word.
3394 */
3395 bool pci_intx_mask_supported(struct pci_dev *dev)
3396 {
3397 bool mask_supported = false;
3398 u16 orig, new;
3399
3400 if (dev->broken_intx_masking)
3401 return false;
3402
3403 pci_cfg_access_lock(dev);
3404
3405 pci_read_config_word(dev, PCI_COMMAND, &orig);
3406 pci_write_config_word(dev, PCI_COMMAND,
3407 orig ^ PCI_COMMAND_INTX_DISABLE);
3408 pci_read_config_word(dev, PCI_COMMAND, &new);
3409
3410 /*
3411 * There's no way to protect against hardware bugs or detect them
3412 * reliably, but as long as we know what the value should be, let's
3413 * go ahead and check it.
3414 */
3415 if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
3416 dev_err(&dev->dev, "Command register changed from 0x%x to 0x%x: driver or hardware bug?\n",
3417 orig, new);
3418 } else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
3419 mask_supported = true;
3420 pci_write_config_word(dev, PCI_COMMAND, orig);
3421 }
3422
3423 pci_cfg_access_unlock(dev);
3424 return mask_supported;
3425 }
3426 EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
3427
3428 static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
3429 {
3430 struct pci_bus *bus = dev->bus;
3431 bool mask_updated = true;
3432 u32 cmd_status_dword;
3433 u16 origcmd, newcmd;
3434 unsigned long flags;
3435 bool irq_pending;
3436
3437 /*
3438 * We do a single dword read to retrieve both command and status.
3439 * Document assumptions that make this possible.
3440 */
3441 BUILD_BUG_ON(PCI_COMMAND % 4);
3442 BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS);
3443
3444 raw_spin_lock_irqsave(&pci_lock, flags);
3445
3446 bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword);
3447
3448 irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT;
3449
3450 /*
3451 * Check interrupt status register to see whether our device
3452 * triggered the interrupt (when masking) or the next IRQ is
3453 * already pending (when unmasking).
3454 */
3455 if (mask != irq_pending) {
3456 mask_updated = false;
3457 goto done;
3458 }
3459
3460 origcmd = cmd_status_dword;
3461 newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
3462 if (mask)
3463 newcmd |= PCI_COMMAND_INTX_DISABLE;
3464 if (newcmd != origcmd)
3465 bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);
3466
3467 done:
3468 raw_spin_unlock_irqrestore(&pci_lock, flags);
3469
3470 return mask_updated;
3471 }
3472
3473 /**
3474 * pci_check_and_mask_intx - mask INTx on pending interrupt
3475 * @dev: the PCI device to operate on
3476 *
3477 * Check if the device dev has its INTx line asserted, mask it and
3478 * return true in that case. False is returned if not interrupt was
3479 * pending.
3480 */
3481 bool pci_check_and_mask_intx(struct pci_dev *dev)
3482 {
3483 return pci_check_and_set_intx_mask(dev, true);
3484 }
3485 EXPORT_SYMBOL_GPL(pci_check_and_mask_intx);
3486
3487 /**
3488 * pci_check_and_unmask_intx - unmask INTx if no interrupt is pending
3489 * @dev: the PCI device to operate on
3490 *
3491 * Check if the device dev has its INTx line asserted, unmask it if not
3492 * and return true. False is returned and the mask remains active if
3493 * there was still an interrupt pending.
3494 */
3495 bool pci_check_and_unmask_intx(struct pci_dev *dev)
3496 {
3497 return pci_check_and_set_intx_mask(dev, false);
3498 }
3499 EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx);
3500
3501 /**
3502 * pci_wait_for_pending_transaction - waits for pending transaction
3503 * @dev: the PCI device to operate on
3504 *
3505 * Return 0 if transaction is pending 1 otherwise.
3506 */
3507 int pci_wait_for_pending_transaction(struct pci_dev *dev)
3508 {
3509 if (!pci_is_pcie(dev))
3510 return 1;
3511
3512 return pci_wait_for_pending(dev, pci_pcie_cap(dev) + PCI_EXP_DEVSTA,
3513 PCI_EXP_DEVSTA_TRPND);
3514 }
3515 EXPORT_SYMBOL(pci_wait_for_pending_transaction);
3516
3517 /*
3518 * We should only need to wait 100ms after FLR, but some devices take longer.
3519 * Wait for up to 1000ms for config space to return something other than -1.
3520 * Intel IGD requires this when an LCD panel is attached. We read the 2nd
3521 * dword because VFs don't implement the 1st dword.
3522 */
3523 static void pci_flr_wait(struct pci_dev *dev)
3524 {
3525 int i = 0;
3526 u32 id;
3527
3528 do {
3529 msleep(100);
3530 pci_read_config_dword(dev, PCI_COMMAND, &id);
3531 } while (i++ < 10 && id == ~0);
3532
3533 if (id == ~0)
3534 dev_warn(&dev->dev, "Failed to return from FLR\n");
3535 else if (i > 1)
3536 dev_info(&dev->dev, "Required additional %dms to return from FLR\n",
3537 (i - 1) * 100);
3538 }
3539
3540 static int pcie_flr(struct pci_dev *dev, int probe)
3541 {
3542 u32 cap;
3543
3544 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
3545 if (!(cap & PCI_EXP_DEVCAP_FLR))
3546 return -ENOTTY;
3547
3548 if (probe)
3549 return 0;
3550
3551 if (!pci_wait_for_pending_transaction(dev))
3552 dev_err(&dev->dev, "timed out waiting for pending transaction; performing function level reset anyway\n");
3553
3554 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
3555 pci_flr_wait(dev);
3556 return 0;
3557 }
3558
3559 static int pci_af_flr(struct pci_dev *dev, int probe)
3560 {
3561 int pos;
3562 u8 cap;
3563
3564 pos = pci_find_capability(dev, PCI_CAP_ID_AF);
3565 if (!pos)
3566 return -ENOTTY;
3567
3568 pci_read_config_byte(dev, pos + PCI_AF_CAP, &cap);
3569 if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR))
3570 return -ENOTTY;
3571
3572 if (probe)
3573 return 0;
3574
3575 /*
3576 * Wait for Transaction Pending bit to clear. A word-aligned test
3577 * is used, so we use the conrol offset rather than status and shift
3578 * the test bit to match.
3579 */
3580 if (!pci_wait_for_pending(dev, pos + PCI_AF_CTRL,
3581 PCI_AF_STATUS_TP << 8))
3582 dev_err(&dev->dev, "timed out waiting for pending transaction; performing AF function level reset anyway\n");
3583
3584 pci_write_config_byte(dev, pos + PCI_AF_CTRL, PCI_AF_CTRL_FLR);
3585 pci_flr_wait(dev);
3586 return 0;
3587 }
3588
3589 /**
3590 * pci_pm_reset - Put device into PCI_D3 and back into PCI_D0.
3591 * @dev: Device to reset.
3592 * @probe: If set, only check if the device can be reset this way.
3593 *
3594 * If @dev supports native PCI PM and its PCI_PM_CTRL_NO_SOFT_RESET flag is
3595 * unset, it will be reinitialized internally when going from PCI_D3hot to
3596 * PCI_D0. If that's the case and the device is not in a low-power state
3597 * already, force it into PCI_D3hot and back to PCI_D0, causing it to be reset.
3598 *
3599 * NOTE: This causes the caller to sleep for twice the device power transition
3600 * cooldown period, which for the D0->D3hot and D3hot->D0 transitions is 10 ms
3601 * by default (i.e. unless the @dev's d3_delay field has a different value).
3602 * Moreover, only devices in D0 can be reset by this function.
3603 */
3604 static int pci_pm_reset(struct pci_dev *dev, int probe)
3605 {
3606 u16 csr;
3607
3608 if (!dev->pm_cap || dev->dev_flags & PCI_DEV_FLAGS_NO_PM_RESET)
3609 return -ENOTTY;
3610
3611 pci_read_config_word(dev, dev->pm_cap + PCI_PM_CTRL, &csr);
3612 if (csr & PCI_PM_CTRL_NO_SOFT_RESET)
3613 return -ENOTTY;
3614
3615 if (probe)
3616 return 0;
3617
3618 if (dev->current_state != PCI_D0)
3619 return -EINVAL;
3620
3621 csr &= ~PCI_PM_CTRL_STATE_MASK;
3622 csr |= PCI_D3hot;
3623 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3624 pci_dev_d3_sleep(dev);
3625
3626 csr &= ~PCI_PM_CTRL_STATE_MASK;
3627 csr |= PCI_D0;
3628 pci_write_config_word(dev, dev->pm_cap + PCI_PM_CTRL, csr);
3629 pci_dev_d3_sleep(dev);
3630
3631 return 0;
3632 }
3633
3634 void pci_reset_secondary_bus(struct pci_dev *dev)
3635 {
3636 u16 ctrl;
3637
3638 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &ctrl);
3639 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
3640 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3641 /*
3642 * PCI spec v3.0 7.6.4.2 requires minimum Trst of 1ms. Double
3643 * this to 2ms to ensure that we meet the minimum requirement.
3644 */
3645 msleep(2);
3646
3647 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
3648 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, ctrl);
3649
3650 /*
3651 * Trhfa for conventional PCI is 2^25 clock cycles.
3652 * Assuming a minimum 33MHz clock this results in a 1s
3653 * delay before we can consider subordinate devices to
3654 * be re-initialized. PCIe has some ways to shorten this,
3655 * but we don't make use of them yet.
3656 */
3657 ssleep(1);
3658 }
3659
3660 void __weak pcibios_reset_secondary_bus(struct pci_dev *dev)
3661 {
3662 pci_reset_secondary_bus(dev);
3663 }
3664
3665 /**
3666 * pci_reset_bridge_secondary_bus - Reset the secondary bus on a PCI bridge.
3667 * @dev: Bridge device
3668 *
3669 * Use the bridge control register to assert reset on the secondary bus.
3670 * Devices on the secondary bus are left in power-on state.
3671 */
3672 void pci_reset_bridge_secondary_bus(struct pci_dev *dev)
3673 {
3674 pcibios_reset_secondary_bus(dev);
3675 }
3676 EXPORT_SYMBOL_GPL(pci_reset_bridge_secondary_bus);
3677
3678 static int pci_parent_bus_reset(struct pci_dev *dev, int probe)
3679 {
3680 struct pci_dev *pdev;
3681
3682 if (pci_is_root_bus(dev->bus) || dev->subordinate ||
3683 !dev->bus->self || dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
3684 return -ENOTTY;
3685
3686 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3687 if (pdev != dev)
3688 return -ENOTTY;
3689
3690 if (probe)
3691 return 0;
3692
3693 pci_reset_bridge_secondary_bus(dev->bus->self);
3694
3695 return 0;
3696 }
3697
3698 static int pci_reset_hotplug_slot(struct hotplug_slot *hotplug, int probe)
3699 {
3700 int rc = -ENOTTY;
3701
3702 if (!hotplug || !try_module_get(hotplug->ops->owner))
3703 return rc;
3704
3705 if (hotplug->ops->reset_slot)
3706 rc = hotplug->ops->reset_slot(hotplug, probe);
3707
3708 module_put(hotplug->ops->owner);
3709
3710 return rc;
3711 }
3712
3713 static int pci_dev_reset_slot_function(struct pci_dev *dev, int probe)
3714 {
3715 struct pci_dev *pdev;
3716
3717 if (dev->subordinate || !dev->slot ||
3718 dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET)
3719 return -ENOTTY;
3720
3721 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
3722 if (pdev != dev && pdev->slot == dev->slot)
3723 return -ENOTTY;
3724
3725 return pci_reset_hotplug_slot(dev->slot->hotplug, probe);
3726 }
3727
3728 static int __pci_dev_reset(struct pci_dev *dev, int probe)
3729 {
3730 int rc;
3731
3732 might_sleep();
3733
3734 rc = pci_dev_specific_reset(dev, probe);
3735 if (rc != -ENOTTY)
3736 goto done;
3737
3738 rc = pcie_flr(dev, probe);
3739 if (rc != -ENOTTY)
3740 goto done;
3741
3742 rc = pci_af_flr(dev, probe);
3743 if (rc != -ENOTTY)
3744 goto done;
3745
3746 rc = pci_pm_reset(dev, probe);
3747 if (rc != -ENOTTY)
3748 goto done;
3749
3750 rc = pci_dev_reset_slot_function(dev, probe);
3751 if (rc != -ENOTTY)
3752 goto done;
3753
3754 rc = pci_parent_bus_reset(dev, probe);
3755 done:
3756 return rc;
3757 }
3758
3759 static void pci_dev_lock(struct pci_dev *dev)
3760 {
3761 pci_cfg_access_lock(dev);
3762 /* block PM suspend, driver probe, etc. */
3763 device_lock(&dev->dev);
3764 }
3765
3766 /* Return 1 on successful lock, 0 on contention */
3767 static int pci_dev_trylock(struct pci_dev *dev)
3768 {
3769 if (pci_cfg_access_trylock(dev)) {
3770 if (device_trylock(&dev->dev))
3771 return 1;
3772 pci_cfg_access_unlock(dev);
3773 }
3774
3775 return 0;
3776 }
3777
3778 static void pci_dev_unlock(struct pci_dev *dev)
3779 {
3780 device_unlock(&dev->dev);
3781 pci_cfg_access_unlock(dev);
3782 }
3783
3784 /**
3785 * pci_reset_notify - notify device driver of reset
3786 * @dev: device to be notified of reset
3787 * @prepare: 'true' if device is about to be reset; 'false' if reset attempt
3788 * completed
3789 *
3790 * Must be called prior to device access being disabled and after device
3791 * access is restored.
3792 */
3793 static void pci_reset_notify(struct pci_dev *dev, bool prepare)
3794 {
3795 const struct pci_error_handlers *err_handler =
3796 dev->driver ? dev->driver->err_handler : NULL;
3797 if (err_handler && err_handler->reset_notify)
3798 err_handler->reset_notify(dev, prepare);
3799 }
3800
3801 static void pci_dev_save_and_disable(struct pci_dev *dev)
3802 {
3803 pci_reset_notify(dev, true);
3804
3805 /*
3806 * Wake-up device prior to save. PM registers default to D0 after
3807 * reset and a simple register restore doesn't reliably return
3808 * to a non-D0 state anyway.
3809 */
3810 pci_set_power_state(dev, PCI_D0);
3811
3812 pci_save_state(dev);
3813 /*
3814 * Disable the device by clearing the Command register, except for
3815 * INTx-disable which is set. This not only disables MMIO and I/O port
3816 * BARs, but also prevents the device from being Bus Master, preventing
3817 * DMA from the device including MSI/MSI-X interrupts. For PCI 2.3
3818 * compliant devices, INTx-disable prevents legacy interrupts.
3819 */
3820 pci_write_config_word(dev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
3821 }
3822
3823 static void pci_dev_restore(struct pci_dev *dev)
3824 {
3825 pci_restore_state(dev);
3826 pci_reset_notify(dev, false);
3827 }
3828
3829 static int pci_dev_reset(struct pci_dev *dev, int probe)
3830 {
3831 int rc;
3832
3833 if (!probe)
3834 pci_dev_lock(dev);
3835
3836 rc = __pci_dev_reset(dev, probe);
3837
3838 if (!probe)
3839 pci_dev_unlock(dev);
3840
3841 return rc;
3842 }
3843
3844 /**
3845 * __pci_reset_function - reset a PCI device function
3846 * @dev: PCI device to reset
3847 *
3848 * Some devices allow an individual function to be reset without affecting
3849 * other functions in the same device. The PCI device must be responsive
3850 * to PCI config space in order to use this function.
3851 *
3852 * The device function is presumed to be unused when this function is called.
3853 * Resetting the device will make the contents of PCI configuration space
3854 * random, so any caller of this must be prepared to reinitialise the
3855 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3856 * etc.
3857 *
3858 * Returns 0 if the device function was successfully reset or negative if the
3859 * device doesn't support resetting a single function.
3860 */
3861 int __pci_reset_function(struct pci_dev *dev)
3862 {
3863 return pci_dev_reset(dev, 0);
3864 }
3865 EXPORT_SYMBOL_GPL(__pci_reset_function);
3866
3867 /**
3868 * __pci_reset_function_locked - reset a PCI device function while holding
3869 * the @dev mutex lock.
3870 * @dev: PCI device to reset
3871 *
3872 * Some devices allow an individual function to be reset without affecting
3873 * other functions in the same device. The PCI device must be responsive
3874 * to PCI config space in order to use this function.
3875 *
3876 * The device function is presumed to be unused and the caller is holding
3877 * the device mutex lock when this function is called.
3878 * Resetting the device will make the contents of PCI configuration space
3879 * random, so any caller of this must be prepared to reinitialise the
3880 * device including MSI, bus mastering, BARs, decoding IO and memory spaces,
3881 * etc.
3882 *
3883 * Returns 0 if the device function was successfully reset or negative if the
3884 * device doesn't support resetting a single function.
3885 */
3886 int __pci_reset_function_locked(struct pci_dev *dev)
3887 {
3888 return __pci_dev_reset(dev, 0);
3889 }
3890 EXPORT_SYMBOL_GPL(__pci_reset_function_locked);
3891
3892 /**
3893 * pci_probe_reset_function - check whether the device can be safely reset
3894 * @dev: PCI device to reset
3895 *
3896 * Some devices allow an individual function to be reset without affecting
3897 * other functions in the same device. The PCI device must be responsive
3898 * to PCI config space in order to use this function.
3899 *
3900 * Returns 0 if the device function can be reset or negative if the
3901 * device doesn't support resetting a single function.
3902 */
3903 int pci_probe_reset_function(struct pci_dev *dev)
3904 {
3905 return pci_dev_reset(dev, 1);
3906 }
3907
3908 /**
3909 * pci_reset_function - quiesce and reset a PCI device function
3910 * @dev: PCI device to reset
3911 *
3912 * Some devices allow an individual function to be reset without affecting
3913 * other functions in the same device. The PCI device must be responsive
3914 * to PCI config space in order to use this function.
3915 *
3916 * This function does not just reset the PCI portion of a device, but
3917 * clears all the state associated with the device. This function differs
3918 * from __pci_reset_function in that it saves and restores device state
3919 * over the reset.
3920 *
3921 * Returns 0 if the device function was successfully reset or negative if the
3922 * device doesn't support resetting a single function.
3923 */
3924 int pci_reset_function(struct pci_dev *dev)
3925 {
3926 int rc;
3927
3928 rc = pci_dev_reset(dev, 1);
3929 if (rc)
3930 return rc;
3931
3932 pci_dev_save_and_disable(dev);
3933
3934 rc = pci_dev_reset(dev, 0);
3935
3936 pci_dev_restore(dev);
3937
3938 return rc;
3939 }
3940 EXPORT_SYMBOL_GPL(pci_reset_function);
3941
3942 /**
3943 * pci_try_reset_function - quiesce and reset a PCI device function
3944 * @dev: PCI device to reset
3945 *
3946 * Same as above, except return -EAGAIN if unable to lock device.
3947 */
3948 int pci_try_reset_function(struct pci_dev *dev)
3949 {
3950 int rc;
3951
3952 rc = pci_dev_reset(dev, 1);
3953 if (rc)
3954 return rc;
3955
3956 pci_dev_save_and_disable(dev);
3957
3958 if (pci_dev_trylock(dev)) {
3959 rc = __pci_dev_reset(dev, 0);
3960 pci_dev_unlock(dev);
3961 } else
3962 rc = -EAGAIN;
3963
3964 pci_dev_restore(dev);
3965
3966 return rc;
3967 }
3968 EXPORT_SYMBOL_GPL(pci_try_reset_function);
3969
3970 /* Do any devices on or below this bus prevent a bus reset? */
3971 static bool pci_bus_resetable(struct pci_bus *bus)
3972 {
3973 struct pci_dev *dev;
3974
3975 list_for_each_entry(dev, &bus->devices, bus_list) {
3976 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
3977 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
3978 return false;
3979 }
3980
3981 return true;
3982 }
3983
3984 /* Lock devices from the top of the tree down */
3985 static void pci_bus_lock(struct pci_bus *bus)
3986 {
3987 struct pci_dev *dev;
3988
3989 list_for_each_entry(dev, &bus->devices, bus_list) {
3990 pci_dev_lock(dev);
3991 if (dev->subordinate)
3992 pci_bus_lock(dev->subordinate);
3993 }
3994 }
3995
3996 /* Unlock devices from the bottom of the tree up */
3997 static void pci_bus_unlock(struct pci_bus *bus)
3998 {
3999 struct pci_dev *dev;
4000
4001 list_for_each_entry(dev, &bus->devices, bus_list) {
4002 if (dev->subordinate)
4003 pci_bus_unlock(dev->subordinate);
4004 pci_dev_unlock(dev);
4005 }
4006 }
4007
4008 /* Return 1 on successful lock, 0 on contention */
4009 static int pci_bus_trylock(struct pci_bus *bus)
4010 {
4011 struct pci_dev *dev;
4012
4013 list_for_each_entry(dev, &bus->devices, bus_list) {
4014 if (!pci_dev_trylock(dev))
4015 goto unlock;
4016 if (dev->subordinate) {
4017 if (!pci_bus_trylock(dev->subordinate)) {
4018 pci_dev_unlock(dev);
4019 goto unlock;
4020 }
4021 }
4022 }
4023 return 1;
4024
4025 unlock:
4026 list_for_each_entry_continue_reverse(dev, &bus->devices, bus_list) {
4027 if (dev->subordinate)
4028 pci_bus_unlock(dev->subordinate);
4029 pci_dev_unlock(dev);
4030 }
4031 return 0;
4032 }
4033
4034 /* Do any devices on or below this slot prevent a bus reset? */
4035 static bool pci_slot_resetable(struct pci_slot *slot)
4036 {
4037 struct pci_dev *dev;
4038
4039 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4040 if (!dev->slot || dev->slot != slot)
4041 continue;
4042 if (dev->dev_flags & PCI_DEV_FLAGS_NO_BUS_RESET ||
4043 (dev->subordinate && !pci_bus_resetable(dev->subordinate)))
4044 return false;
4045 }
4046
4047 return true;
4048 }
4049
4050 /* Lock devices from the top of the tree down */
4051 static void pci_slot_lock(struct pci_slot *slot)
4052 {
4053 struct pci_dev *dev;
4054
4055 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4056 if (!dev->slot || dev->slot != slot)
4057 continue;
4058 pci_dev_lock(dev);
4059 if (dev->subordinate)
4060 pci_bus_lock(dev->subordinate);
4061 }
4062 }
4063
4064 /* Unlock devices from the bottom of the tree up */
4065 static void pci_slot_unlock(struct pci_slot *slot)
4066 {
4067 struct pci_dev *dev;
4068
4069 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4070 if (!dev->slot || dev->slot != slot)
4071 continue;
4072 if (dev->subordinate)
4073 pci_bus_unlock(dev->subordinate);
4074 pci_dev_unlock(dev);
4075 }
4076 }
4077
4078 /* Return 1 on successful lock, 0 on contention */
4079 static int pci_slot_trylock(struct pci_slot *slot)
4080 {
4081 struct pci_dev *dev;
4082
4083 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4084 if (!dev->slot || dev->slot != slot)
4085 continue;
4086 if (!pci_dev_trylock(dev))
4087 goto unlock;
4088 if (dev->subordinate) {
4089 if (!pci_bus_trylock(dev->subordinate)) {
4090 pci_dev_unlock(dev);
4091 goto unlock;
4092 }
4093 }
4094 }
4095 return 1;
4096
4097 unlock:
4098 list_for_each_entry_continue_reverse(dev,
4099 &slot->bus->devices, bus_list) {
4100 if (!dev->slot || dev->slot != slot)
4101 continue;
4102 if (dev->subordinate)
4103 pci_bus_unlock(dev->subordinate);
4104 pci_dev_unlock(dev);
4105 }
4106 return 0;
4107 }
4108
4109 /* Save and disable devices from the top of the tree down */
4110 static void pci_bus_save_and_disable(struct pci_bus *bus)
4111 {
4112 struct pci_dev *dev;
4113
4114 list_for_each_entry(dev, &bus->devices, bus_list) {
4115 pci_dev_save_and_disable(dev);
4116 if (dev->subordinate)
4117 pci_bus_save_and_disable(dev->subordinate);
4118 }
4119 }
4120
4121 /*
4122 * Restore devices from top of the tree down - parent bridges need to be
4123 * restored before we can get to subordinate devices.
4124 */
4125 static void pci_bus_restore(struct pci_bus *bus)
4126 {
4127 struct pci_dev *dev;
4128
4129 list_for_each_entry(dev, &bus->devices, bus_list) {
4130 pci_dev_restore(dev);
4131 if (dev->subordinate)
4132 pci_bus_restore(dev->subordinate);
4133 }
4134 }
4135
4136 /* Save and disable devices from the top of the tree down */
4137 static void pci_slot_save_and_disable(struct pci_slot *slot)
4138 {
4139 struct pci_dev *dev;
4140
4141 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4142 if (!dev->slot || dev->slot != slot)
4143 continue;
4144 pci_dev_save_and_disable(dev);
4145 if (dev->subordinate)
4146 pci_bus_save_and_disable(dev->subordinate);
4147 }
4148 }
4149
4150 /*
4151 * Restore devices from top of the tree down - parent bridges need to be
4152 * restored before we can get to subordinate devices.
4153 */
4154 static void pci_slot_restore(struct pci_slot *slot)
4155 {
4156 struct pci_dev *dev;
4157
4158 list_for_each_entry(dev, &slot->bus->devices, bus_list) {
4159 if (!dev->slot || dev->slot != slot)
4160 continue;
4161 pci_dev_restore(dev);
4162 if (dev->subordinate)
4163 pci_bus_restore(dev->subordinate);
4164 }
4165 }
4166
4167 static int pci_slot_reset(struct pci_slot *slot, int probe)
4168 {
4169 int rc;
4170
4171 if (!slot || !pci_slot_resetable(slot))
4172 return -ENOTTY;
4173
4174 if (!probe)
4175 pci_slot_lock(slot);
4176
4177 might_sleep();
4178
4179 rc = pci_reset_hotplug_slot(slot->hotplug, probe);
4180
4181 if (!probe)
4182 pci_slot_unlock(slot);
4183
4184 return rc;
4185 }
4186
4187 /**
4188 * pci_probe_reset_slot - probe whether a PCI slot can be reset
4189 * @slot: PCI slot to probe
4190 *
4191 * Return 0 if slot can be reset, negative if a slot reset is not supported.
4192 */
4193 int pci_probe_reset_slot(struct pci_slot *slot)
4194 {
4195 return pci_slot_reset(slot, 1);
4196 }
4197 EXPORT_SYMBOL_GPL(pci_probe_reset_slot);
4198
4199 /**
4200 * pci_reset_slot - reset a PCI slot
4201 * @slot: PCI slot to reset
4202 *
4203 * A PCI bus may host multiple slots, each slot may support a reset mechanism
4204 * independent of other slots. For instance, some slots may support slot power
4205 * control. In the case of a 1:1 bus to slot architecture, this function may
4206 * wrap the bus reset to avoid spurious slot related events such as hotplug.
4207 * Generally a slot reset should be attempted before a bus reset. All of the
4208 * function of the slot and any subordinate buses behind the slot are reset
4209 * through this function. PCI config space of all devices in the slot and
4210 * behind the slot is saved before and restored after reset.
4211 *
4212 * Return 0 on success, non-zero on error.
4213 */
4214 int pci_reset_slot(struct pci_slot *slot)
4215 {
4216 int rc;
4217
4218 rc = pci_slot_reset(slot, 1);
4219 if (rc)
4220 return rc;
4221
4222 pci_slot_save_and_disable(slot);
4223
4224 rc = pci_slot_reset(slot, 0);
4225
4226 pci_slot_restore(slot);
4227
4228 return rc;
4229 }
4230 EXPORT_SYMBOL_GPL(pci_reset_slot);
4231
4232 /**
4233 * pci_try_reset_slot - Try to reset a PCI slot
4234 * @slot: PCI slot to reset
4235 *
4236 * Same as above except return -EAGAIN if the slot cannot be locked
4237 */
4238 int pci_try_reset_slot(struct pci_slot *slot)
4239 {
4240 int rc;
4241
4242 rc = pci_slot_reset(slot, 1);
4243 if (rc)
4244 return rc;
4245
4246 pci_slot_save_and_disable(slot);
4247
4248 if (pci_slot_trylock(slot)) {
4249 might_sleep();
4250 rc = pci_reset_hotplug_slot(slot->hotplug, 0);
4251 pci_slot_unlock(slot);
4252 } else
4253 rc = -EAGAIN;
4254
4255 pci_slot_restore(slot);
4256
4257 return rc;
4258 }
4259 EXPORT_SYMBOL_GPL(pci_try_reset_slot);
4260
4261 static int pci_bus_reset(struct pci_bus *bus, int probe)
4262 {
4263 if (!bus->self || !pci_bus_resetable(bus))
4264 return -ENOTTY;
4265
4266 if (probe)
4267 return 0;
4268
4269 pci_bus_lock(bus);
4270
4271 might_sleep();
4272
4273 pci_reset_bridge_secondary_bus(bus->self);
4274
4275 pci_bus_unlock(bus);
4276
4277 return 0;
4278 }
4279
4280 /**
4281 * pci_probe_reset_bus - probe whether a PCI bus can be reset
4282 * @bus: PCI bus to probe
4283 *
4284 * Return 0 if bus can be reset, negative if a bus reset is not supported.
4285 */
4286 int pci_probe_reset_bus(struct pci_bus *bus)
4287 {
4288 return pci_bus_reset(bus, 1);
4289 }
4290 EXPORT_SYMBOL_GPL(pci_probe_reset_bus);
4291
4292 /**
4293 * pci_reset_bus - reset a PCI bus
4294 * @bus: top level PCI bus to reset
4295 *
4296 * Do a bus reset on the given bus and any subordinate buses, saving
4297 * and restoring state of all devices.
4298 *
4299 * Return 0 on success, non-zero on error.
4300 */
4301 int pci_reset_bus(struct pci_bus *bus)
4302 {
4303 int rc;
4304
4305 rc = pci_bus_reset(bus, 1);
4306 if (rc)
4307 return rc;
4308
4309 pci_bus_save_and_disable(bus);
4310
4311 rc = pci_bus_reset(bus, 0);
4312
4313 pci_bus_restore(bus);
4314
4315 return rc;
4316 }
4317 EXPORT_SYMBOL_GPL(pci_reset_bus);
4318
4319 /**
4320 * pci_try_reset_bus - Try to reset a PCI bus
4321 * @bus: top level PCI bus to reset
4322 *
4323 * Same as above except return -EAGAIN if the bus cannot be locked
4324 */
4325 int pci_try_reset_bus(struct pci_bus *bus)
4326 {
4327 int rc;
4328
4329 rc = pci_bus_reset(bus, 1);
4330 if (rc)
4331 return rc;
4332
4333 pci_bus_save_and_disable(bus);
4334
4335 if (pci_bus_trylock(bus)) {
4336 might_sleep();
4337 pci_reset_bridge_secondary_bus(bus->self);
4338 pci_bus_unlock(bus);
4339 } else
4340 rc = -EAGAIN;
4341
4342 pci_bus_restore(bus);
4343
4344 return rc;
4345 }
4346 EXPORT_SYMBOL_GPL(pci_try_reset_bus);
4347
4348 /**
4349 * pcix_get_max_mmrbc - get PCI-X maximum designed memory read byte count
4350 * @dev: PCI device to query
4351 *
4352 * Returns mmrbc: maximum designed memory read count in bytes
4353 * or appropriate error value.
4354 */
4355 int pcix_get_max_mmrbc(struct pci_dev *dev)
4356 {
4357 int cap;
4358 u32 stat;
4359
4360 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4361 if (!cap)
4362 return -EINVAL;
4363
4364 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4365 return -EINVAL;
4366
4367 return 512 << ((stat & PCI_X_STATUS_MAX_READ) >> 21);
4368 }
4369 EXPORT_SYMBOL(pcix_get_max_mmrbc);
4370
4371 /**
4372 * pcix_get_mmrbc - get PCI-X maximum memory read byte count
4373 * @dev: PCI device to query
4374 *
4375 * Returns mmrbc: maximum memory read count in bytes
4376 * or appropriate error value.
4377 */
4378 int pcix_get_mmrbc(struct pci_dev *dev)
4379 {
4380 int cap;
4381 u16 cmd;
4382
4383 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4384 if (!cap)
4385 return -EINVAL;
4386
4387 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4388 return -EINVAL;
4389
4390 return 512 << ((cmd & PCI_X_CMD_MAX_READ) >> 2);
4391 }
4392 EXPORT_SYMBOL(pcix_get_mmrbc);
4393
4394 /**
4395 * pcix_set_mmrbc - set PCI-X maximum memory read byte count
4396 * @dev: PCI device to query
4397 * @mmrbc: maximum memory read count in bytes
4398 * valid values are 512, 1024, 2048, 4096
4399 *
4400 * If possible sets maximum memory read byte count, some bridges have erratas
4401 * that prevent this.
4402 */
4403 int pcix_set_mmrbc(struct pci_dev *dev, int mmrbc)
4404 {
4405 int cap;
4406 u32 stat, v, o;
4407 u16 cmd;
4408
4409 if (mmrbc < 512 || mmrbc > 4096 || !is_power_of_2(mmrbc))
4410 return -EINVAL;
4411
4412 v = ffs(mmrbc) - 10;
4413
4414 cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
4415 if (!cap)
4416 return -EINVAL;
4417
4418 if (pci_read_config_dword(dev, cap + PCI_X_STATUS, &stat))
4419 return -EINVAL;
4420
4421 if (v > (stat & PCI_X_STATUS_MAX_READ) >> 21)
4422 return -E2BIG;
4423
4424 if (pci_read_config_word(dev, cap + PCI_X_CMD, &cmd))
4425 return -EINVAL;
4426
4427 o = (cmd & PCI_X_CMD_MAX_READ) >> 2;
4428 if (o != v) {
4429 if (v > o && (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MMRBC))
4430 return -EIO;
4431
4432 cmd &= ~PCI_X_CMD_MAX_READ;
4433 cmd |= v << 2;
4434 if (pci_write_config_word(dev, cap + PCI_X_CMD, cmd))
4435 return -EIO;
4436 }
4437 return 0;
4438 }
4439 EXPORT_SYMBOL(pcix_set_mmrbc);
4440
4441 /**
4442 * pcie_get_readrq - get PCI Express read request size
4443 * @dev: PCI device to query
4444 *
4445 * Returns maximum memory read request in bytes
4446 * or appropriate error value.
4447 */
4448 int pcie_get_readrq(struct pci_dev *dev)
4449 {
4450 u16 ctl;
4451
4452 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4453
4454 return 128 << ((ctl & PCI_EXP_DEVCTL_READRQ) >> 12);
4455 }
4456 EXPORT_SYMBOL(pcie_get_readrq);
4457
4458 /**
4459 * pcie_set_readrq - set PCI Express maximum memory read request
4460 * @dev: PCI device to query
4461 * @rq: maximum memory read count in bytes
4462 * valid values are 128, 256, 512, 1024, 2048, 4096
4463 *
4464 * If possible sets maximum memory read request in bytes
4465 */
4466 int pcie_set_readrq(struct pci_dev *dev, int rq)
4467 {
4468 u16 v;
4469
4470 if (rq < 128 || rq > 4096 || !is_power_of_2(rq))
4471 return -EINVAL;
4472
4473 /*
4474 * If using the "performance" PCIe config, we clamp the
4475 * read rq size to the max packet size to prevent the
4476 * host bridge generating requests larger than we can
4477 * cope with
4478 */
4479 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
4480 int mps = pcie_get_mps(dev);
4481
4482 if (mps < rq)
4483 rq = mps;
4484 }
4485
4486 v = (ffs(rq) - 8) << 12;
4487
4488 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4489 PCI_EXP_DEVCTL_READRQ, v);
4490 }
4491 EXPORT_SYMBOL(pcie_set_readrq);
4492
4493 /**
4494 * pcie_get_mps - get PCI Express maximum payload size
4495 * @dev: PCI device to query
4496 *
4497 * Returns maximum payload size in bytes
4498 */
4499 int pcie_get_mps(struct pci_dev *dev)
4500 {
4501 u16 ctl;
4502
4503 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
4504
4505 return 128 << ((ctl & PCI_EXP_DEVCTL_PAYLOAD) >> 5);
4506 }
4507 EXPORT_SYMBOL(pcie_get_mps);
4508
4509 /**
4510 * pcie_set_mps - set PCI Express maximum payload size
4511 * @dev: PCI device to query
4512 * @mps: maximum payload size in bytes
4513 * valid values are 128, 256, 512, 1024, 2048, 4096
4514 *
4515 * If possible sets maximum payload size
4516 */
4517 int pcie_set_mps(struct pci_dev *dev, int mps)
4518 {
4519 u16 v;
4520
4521 if (mps < 128 || mps > 4096 || !is_power_of_2(mps))
4522 return -EINVAL;
4523
4524 v = ffs(mps) - 8;
4525 if (v > dev->pcie_mpss)
4526 return -EINVAL;
4527 v <<= 5;
4528
4529 return pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
4530 PCI_EXP_DEVCTL_PAYLOAD, v);
4531 }
4532 EXPORT_SYMBOL(pcie_set_mps);
4533
4534 /**
4535 * pcie_get_minimum_link - determine minimum link settings of a PCI device
4536 * @dev: PCI device to query
4537 * @speed: storage for minimum speed
4538 * @width: storage for minimum width
4539 *
4540 * This function will walk up the PCI device chain and determine the minimum
4541 * link width and speed of the device.
4542 */
4543 int pcie_get_minimum_link(struct pci_dev *dev, enum pci_bus_speed *speed,
4544 enum pcie_link_width *width)
4545 {
4546 int ret;
4547
4548 *speed = PCI_SPEED_UNKNOWN;
4549 *width = PCIE_LNK_WIDTH_UNKNOWN;
4550
4551 while (dev) {
4552 u16 lnksta;
4553 enum pci_bus_speed next_speed;
4554 enum pcie_link_width next_width;
4555
4556 ret = pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
4557 if (ret)
4558 return ret;
4559
4560 next_speed = pcie_link_speed[lnksta & PCI_EXP_LNKSTA_CLS];
4561 next_width = (lnksta & PCI_EXP_LNKSTA_NLW) >>
4562 PCI_EXP_LNKSTA_NLW_SHIFT;
4563
4564 if (next_speed < *speed)
4565 *speed = next_speed;
4566
4567 if (next_width < *width)
4568 *width = next_width;
4569
4570 dev = dev->bus->self;
4571 }
4572
4573 return 0;
4574 }
4575 EXPORT_SYMBOL(pcie_get_minimum_link);
4576
4577 /**
4578 * pci_select_bars - Make BAR mask from the type of resource
4579 * @dev: the PCI device for which BAR mask is made
4580 * @flags: resource type mask to be selected
4581 *
4582 * This helper routine makes bar mask from the type of resource.
4583 */
4584 int pci_select_bars(struct pci_dev *dev, unsigned long flags)
4585 {
4586 int i, bars = 0;
4587 for (i = 0; i < PCI_NUM_RESOURCES; i++)
4588 if (pci_resource_flags(dev, i) & flags)
4589 bars |= (1 << i);
4590 return bars;
4591 }
4592 EXPORT_SYMBOL(pci_select_bars);
4593
4594 /**
4595 * pci_resource_bar - get position of the BAR associated with a resource
4596 * @dev: the PCI device
4597 * @resno: the resource number
4598 * @type: the BAR type to be filled in
4599 *
4600 * Returns BAR position in config space, or 0 if the BAR is invalid.
4601 */
4602 int pci_resource_bar(struct pci_dev *dev, int resno, enum pci_bar_type *type)
4603 {
4604 int reg;
4605
4606 if (resno < PCI_ROM_RESOURCE) {
4607 *type = pci_bar_unknown;
4608 return PCI_BASE_ADDRESS_0 + 4 * resno;
4609 } else if (resno == PCI_ROM_RESOURCE) {
4610 *type = pci_bar_mem32;
4611 return dev->rom_base_reg;
4612 } else if (resno < PCI_BRIDGE_RESOURCES) {
4613 /* device specific resource */
4614 *type = pci_bar_unknown;
4615 reg = pci_iov_resource_bar(dev, resno);
4616 if (reg)
4617 return reg;
4618 }
4619
4620 dev_err(&dev->dev, "BAR %d: invalid resource\n", resno);
4621 return 0;
4622 }
4623
4624 /* Some architectures require additional programming to enable VGA */
4625 static arch_set_vga_state_t arch_set_vga_state;
4626
4627 void __init pci_register_set_vga_state(arch_set_vga_state_t func)
4628 {
4629 arch_set_vga_state = func; /* NULL disables */
4630 }
4631
4632 static int pci_set_vga_state_arch(struct pci_dev *dev, bool decode,
4633 unsigned int command_bits, u32 flags)
4634 {
4635 if (arch_set_vga_state)
4636 return arch_set_vga_state(dev, decode, command_bits,
4637 flags);
4638 return 0;
4639 }
4640
4641 /**
4642 * pci_set_vga_state - set VGA decode state on device and parents if requested
4643 * @dev: the PCI device
4644 * @decode: true = enable decoding, false = disable decoding
4645 * @command_bits: PCI_COMMAND_IO and/or PCI_COMMAND_MEMORY
4646 * @flags: traverse ancestors and change bridges
4647 * CHANGE_BRIDGE_ONLY / CHANGE_BRIDGE
4648 */
4649 int pci_set_vga_state(struct pci_dev *dev, bool decode,
4650 unsigned int command_bits, u32 flags)
4651 {
4652 struct pci_bus *bus;
4653 struct pci_dev *bridge;
4654 u16 cmd;
4655 int rc;
4656
4657 WARN_ON((flags & PCI_VGA_STATE_CHANGE_DECODES) && (command_bits & ~(PCI_COMMAND_IO|PCI_COMMAND_MEMORY)));
4658
4659 /* ARCH specific VGA enables */
4660 rc = pci_set_vga_state_arch(dev, decode, command_bits, flags);
4661 if (rc)
4662 return rc;
4663
4664 if (flags & PCI_VGA_STATE_CHANGE_DECODES) {
4665 pci_read_config_word(dev, PCI_COMMAND, &cmd);
4666 if (decode == true)
4667 cmd |= command_bits;
4668 else
4669 cmd &= ~command_bits;
4670 pci_write_config_word(dev, PCI_COMMAND, cmd);
4671 }
4672
4673 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
4674 return 0;
4675
4676 bus = dev->bus;
4677 while (bus) {
4678 bridge = bus->self;
4679 if (bridge) {
4680 pci_read_config_word(bridge, PCI_BRIDGE_CONTROL,
4681 &cmd);
4682 if (decode == true)
4683 cmd |= PCI_BRIDGE_CTL_VGA;
4684 else
4685 cmd &= ~PCI_BRIDGE_CTL_VGA;
4686 pci_write_config_word(bridge, PCI_BRIDGE_CONTROL,
4687 cmd);
4688 }
4689 bus = bus->parent;
4690 }
4691 return 0;
4692 }
4693
4694 /**
4695 * pci_add_dma_alias - Add a DMA devfn alias for a device
4696 * @dev: the PCI device for which alias is added
4697 * @devfn: alias slot and function
4698 *
4699 * This helper encodes 8-bit devfn as bit number in dma_alias_mask.
4700 * It should be called early, preferably as PCI fixup header quirk.
4701 */
4702 void pci_add_dma_alias(struct pci_dev *dev, u8 devfn)
4703 {
4704 if (!dev->dma_alias_mask)
4705 dev->dma_alias_mask = kcalloc(BITS_TO_LONGS(U8_MAX),
4706 sizeof(long), GFP_KERNEL);
4707 if (!dev->dma_alias_mask) {
4708 dev_warn(&dev->dev, "Unable to allocate DMA alias mask\n");
4709 return;
4710 }
4711
4712 set_bit(devfn, dev->dma_alias_mask);
4713 dev_info(&dev->dev, "Enabling fixed DMA alias to %02x.%d\n",
4714 PCI_SLOT(devfn), PCI_FUNC(devfn));
4715 }
4716
4717 bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
4718 {
4719 return (dev1->dma_alias_mask &&
4720 test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
4721 (dev2->dma_alias_mask &&
4722 test_bit(dev1->devfn, dev2->dma_alias_mask));
4723 }
4724
4725 bool pci_device_is_present(struct pci_dev *pdev)
4726 {
4727 u32 v;
4728
4729 return pci_bus_read_dev_vendor_id(pdev->bus, pdev->devfn, &v, 0);
4730 }
4731 EXPORT_SYMBOL_GPL(pci_device_is_present);
4732
4733 void pci_ignore_hotplug(struct pci_dev *dev)
4734 {
4735 struct pci_dev *bridge = dev->bus->self;
4736
4737 dev->ignore_hotplug = 1;
4738 /* Propagate the "ignore hotplug" setting to the parent bridge. */
4739 if (bridge)
4740 bridge->ignore_hotplug = 1;
4741 }
4742 EXPORT_SYMBOL_GPL(pci_ignore_hotplug);
4743
4744 #define RESOURCE_ALIGNMENT_PARAM_SIZE COMMAND_LINE_SIZE
4745 static char resource_alignment_param[RESOURCE_ALIGNMENT_PARAM_SIZE] = {0};
4746 static DEFINE_SPINLOCK(resource_alignment_lock);
4747
4748 /**
4749 * pci_specified_resource_alignment - get resource alignment specified by user.
4750 * @dev: the PCI device to get
4751 *
4752 * RETURNS: Resource alignment if it is specified.
4753 * Zero if it is not specified.
4754 */
4755 static resource_size_t pci_specified_resource_alignment(struct pci_dev *dev)
4756 {
4757 int seg, bus, slot, func, align_order, count;
4758 resource_size_t align = 0;
4759 char *p;
4760
4761 spin_lock(&resource_alignment_lock);
4762 p = resource_alignment_param;
4763 while (*p) {
4764 count = 0;
4765 if (sscanf(p, "%d%n", &align_order, &count) == 1 &&
4766 p[count] == '@') {
4767 p += count + 1;
4768 } else {
4769 align_order = -1;
4770 }
4771 if (sscanf(p, "%x:%x:%x.%x%n",
4772 &seg, &bus, &slot, &func, &count) != 4) {
4773 seg = 0;
4774 if (sscanf(p, "%x:%x.%x%n",
4775 &bus, &slot, &func, &count) != 3) {
4776 /* Invalid format */
4777 printk(KERN_ERR "PCI: Can't parse resource_alignment parameter: %s\n",
4778 p);
4779 break;
4780 }
4781 }
4782 p += count;
4783 if (seg == pci_domain_nr(dev->bus) &&
4784 bus == dev->bus->number &&
4785 slot == PCI_SLOT(dev->devfn) &&
4786 func == PCI_FUNC(dev->devfn)) {
4787 if (align_order == -1)
4788 align = PAGE_SIZE;
4789 else
4790 align = 1 << align_order;
4791 /* Found */
4792 break;
4793 }
4794 if (*p != ';' && *p != ',') {
4795 /* End of param or invalid format */
4796 break;
4797 }
4798 p++;
4799 }
4800 spin_unlock(&resource_alignment_lock);
4801 return align;
4802 }
4803
4804 /*
4805 * This function disables memory decoding and releases memory resources
4806 * of the device specified by kernel's boot parameter 'pci=resource_alignment='.
4807 * It also rounds up size to specified alignment.
4808 * Later on, the kernel will assign page-aligned memory resource back
4809 * to the device.
4810 */
4811 void pci_reassigndev_resource_alignment(struct pci_dev *dev)
4812 {
4813 int i;
4814 struct resource *r;
4815 resource_size_t align, size;
4816 u16 command;
4817
4818 /* check if specified PCI is target device to reassign */
4819 align = pci_specified_resource_alignment(dev);
4820 if (!align)
4821 return;
4822
4823 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
4824 (dev->class >> 8) == PCI_CLASS_BRIDGE_HOST) {
4825 dev_warn(&dev->dev,
4826 "Can't reassign resources to host bridge.\n");
4827 return;
4828 }
4829
4830 dev_info(&dev->dev,
4831 "Disabling memory decoding and releasing memory resources.\n");
4832 pci_read_config_word(dev, PCI_COMMAND, &command);
4833 command &= ~PCI_COMMAND_MEMORY;
4834 pci_write_config_word(dev, PCI_COMMAND, command);
4835
4836 for (i = 0; i < PCI_BRIDGE_RESOURCES; i++) {
4837 r = &dev->resource[i];
4838 if (!(r->flags & IORESOURCE_MEM))
4839 continue;
4840 size = resource_size(r);
4841 if (size < align) {
4842 size = align;
4843 dev_info(&dev->dev,
4844 "Rounding up size of resource #%d to %#llx.\n",
4845 i, (unsigned long long)size);
4846 }
4847 r->flags |= IORESOURCE_UNSET;
4848 r->end = size - 1;
4849 r->start = 0;
4850 }
4851 /* Need to disable bridge's resource window,
4852 * to enable the kernel to reassign new resource
4853 * window later on.
4854 */
4855 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE &&
4856 (dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
4857 for (i = PCI_BRIDGE_RESOURCES; i < PCI_NUM_RESOURCES; i++) {
4858 r = &dev->resource[i];
4859 if (!(r->flags & IORESOURCE_MEM))
4860 continue;
4861 r->flags |= IORESOURCE_UNSET;
4862 r->end = resource_size(r) - 1;
4863 r->start = 0;
4864 }
4865 pci_disable_bridge_window(dev);
4866 }
4867 }
4868
4869 static ssize_t pci_set_resource_alignment_param(const char *buf, size_t count)
4870 {
4871 if (count > RESOURCE_ALIGNMENT_PARAM_SIZE - 1)
4872 count = RESOURCE_ALIGNMENT_PARAM_SIZE - 1;
4873 spin_lock(&resource_alignment_lock);
4874 strncpy(resource_alignment_param, buf, count);
4875 resource_alignment_param[count] = '\0';
4876 spin_unlock(&resource_alignment_lock);
4877 return count;
4878 }
4879
4880 static ssize_t pci_get_resource_alignment_param(char *buf, size_t size)
4881 {
4882 size_t count;
4883 spin_lock(&resource_alignment_lock);
4884 count = snprintf(buf, size, "%s", resource_alignment_param);
4885 spin_unlock(&resource_alignment_lock);
4886 return count;
4887 }
4888
4889 static ssize_t pci_resource_alignment_show(struct bus_type *bus, char *buf)
4890 {
4891 return pci_get_resource_alignment_param(buf, PAGE_SIZE);
4892 }
4893
4894 static ssize_t pci_resource_alignment_store(struct bus_type *bus,
4895 const char *buf, size_t count)
4896 {
4897 return pci_set_resource_alignment_param(buf, count);
4898 }
4899
4900 BUS_ATTR(resource_alignment, 0644, pci_resource_alignment_show,
4901 pci_resource_alignment_store);
4902
4903 static int __init pci_resource_alignment_sysfs_init(void)
4904 {
4905 return bus_create_file(&pci_bus_type,
4906 &bus_attr_resource_alignment);
4907 }
4908 late_initcall(pci_resource_alignment_sysfs_init);
4909
4910 static void pci_no_domains(void)
4911 {
4912 #ifdef CONFIG_PCI_DOMAINS
4913 pci_domains_supported = 0;
4914 #endif
4915 }
4916
4917 #ifdef CONFIG_PCI_DOMAINS
4918 static atomic_t __domain_nr = ATOMIC_INIT(-1);
4919
4920 int pci_get_new_domain_nr(void)
4921 {
4922 return atomic_inc_return(&__domain_nr);
4923 }
4924
4925 #ifdef CONFIG_PCI_DOMAINS_GENERIC
4926 void pci_bus_assign_domain_nr(struct pci_bus *bus, struct device *parent)
4927 {
4928 static int use_dt_domains = -1;
4929 int domain = -1;
4930
4931 if (parent)
4932 domain = of_get_pci_domain_nr(parent->of_node);
4933 /*
4934 * Check DT domain and use_dt_domains values.
4935 *
4936 * If DT domain property is valid (domain >= 0) and
4937 * use_dt_domains != 0, the DT assignment is valid since this means
4938 * we have not previously allocated a domain number by using
4939 * pci_get_new_domain_nr(); we should also update use_dt_domains to
4940 * 1, to indicate that we have just assigned a domain number from
4941 * DT.
4942 *
4943 * If DT domain property value is not valid (ie domain < 0), and we
4944 * have not previously assigned a domain number from DT
4945 * (use_dt_domains != 1) we should assign a domain number by
4946 * using the:
4947 *
4948 * pci_get_new_domain_nr()
4949 *
4950 * API and update the use_dt_domains value to keep track of method we
4951 * are using to assign domain numbers (use_dt_domains = 0).
4952 *
4953 * All other combinations imply we have a platform that is trying
4954 * to mix domain numbers obtained from DT and pci_get_new_domain_nr(),
4955 * which is a recipe for domain mishandling and it is prevented by
4956 * invalidating the domain value (domain = -1) and printing a
4957 * corresponding error.
4958 */
4959 if (domain >= 0 && use_dt_domains) {
4960 use_dt_domains = 1;
4961 } else if (domain < 0 && use_dt_domains != 1) {
4962 use_dt_domains = 0;
4963 domain = pci_get_new_domain_nr();
4964 } else {
4965 dev_err(parent, "Node %s has inconsistent \"linux,pci-domain\" property in DT\n",
4966 parent->of_node->full_name);
4967 domain = -1;
4968 }
4969
4970 bus->domain_nr = domain;
4971 }
4972 #endif
4973 #endif
4974
4975 /**
4976 * pci_ext_cfg_avail - can we access extended PCI config space?
4977 *
4978 * Returns 1 if we can access PCI extended config space (offsets
4979 * greater than 0xff). This is the default implementation. Architecture
4980 * implementations can override this.
4981 */
4982 int __weak pci_ext_cfg_avail(void)
4983 {
4984 return 1;
4985 }
4986
4987 void __weak pci_fixup_cardbus(struct pci_bus *bus)
4988 {
4989 }
4990 EXPORT_SYMBOL(pci_fixup_cardbus);
4991
4992 static int __init pci_setup(char *str)
4993 {
4994 while (str) {
4995 char *k = strchr(str, ',');
4996 if (k)
4997 *k++ = 0;
4998 if (*str && (str = pcibios_setup(str)) && *str) {
4999 if (!strcmp(str, "nomsi")) {
5000 pci_no_msi();
5001 } else if (!strcmp(str, "noaer")) {
5002 pci_no_aer();
5003 } else if (!strncmp(str, "realloc=", 8)) {
5004 pci_realloc_get_opt(str + 8);
5005 } else if (!strncmp(str, "realloc", 7)) {
5006 pci_realloc_get_opt("on");
5007 } else if (!strcmp(str, "nodomains")) {
5008 pci_no_domains();
5009 } else if (!strncmp(str, "noari", 5)) {
5010 pcie_ari_disabled = true;
5011 } else if (!strncmp(str, "cbiosize=", 9)) {
5012 pci_cardbus_io_size = memparse(str + 9, &str);
5013 } else if (!strncmp(str, "cbmemsize=", 10)) {
5014 pci_cardbus_mem_size = memparse(str + 10, &str);
5015 } else if (!strncmp(str, "resource_alignment=", 19)) {
5016 pci_set_resource_alignment_param(str + 19,
5017 strlen(str + 19));
5018 } else if (!strncmp(str, "ecrc=", 5)) {
5019 pcie_ecrc_get_policy(str + 5);
5020 } else if (!strncmp(str, "hpiosize=", 9)) {
5021 pci_hotplug_io_size = memparse(str + 9, &str);
5022 } else if (!strncmp(str, "hpmemsize=", 10)) {
5023 pci_hotplug_mem_size = memparse(str + 10, &str);
5024 } else if (!strncmp(str, "pcie_bus_tune_off", 17)) {
5025 pcie_bus_config = PCIE_BUS_TUNE_OFF;
5026 } else if (!strncmp(str, "pcie_bus_safe", 13)) {
5027 pcie_bus_config = PCIE_BUS_SAFE;
5028 } else if (!strncmp(str, "pcie_bus_perf", 13)) {
5029 pcie_bus_config = PCIE_BUS_PERFORMANCE;
5030 } else if (!strncmp(str, "pcie_bus_peer2peer", 18)) {
5031 pcie_bus_config = PCIE_BUS_PEER2PEER;
5032 } else if (!strncmp(str, "pcie_scan_all", 13)) {
5033 pci_add_flags(PCI_SCAN_ALL_PCIE_DEVS);
5034 } else {
5035 printk(KERN_ERR "PCI: Unknown option `%s'\n",
5036 str);
5037 }
5038 }
5039 str = k;
5040 }
5041 return 0;
5042 }
5043 early_param("pci", pci_setup);
This page took 0.210301 seconds and 6 git commands to generate.